xref: /illumos-gate/usr/src/uts/common/inet/ip/ip.c (revision 1edba515a3484e0f74b638b203d462b3112ac84d)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 1990 Mentat Inc.
25  * Copyright (c) 2017 OmniTI Computer Consulting, Inc. All rights reserved.
26  * Copyright (c) 2016 by Delphix. All rights reserved.
27  * Copyright 2020 OmniOS Community Edition (OmniOSce) Association.
28  * Copyright 2021 Joyent, Inc.
29  * Copyright 2024 Oxide Computer Company
30  */
31 
32 #include <sys/types.h>
33 #include <sys/stream.h>
34 #include <sys/dlpi.h>
35 #include <sys/stropts.h>
36 #include <sys/sysmacros.h>
37 #include <sys/strsubr.h>
38 #include <sys/strlog.h>
39 #include <sys/strsun.h>
40 #include <sys/zone.h>
41 #define	_SUN_TPI_VERSION 2
42 #include <sys/tihdr.h>
43 #include <sys/xti_inet.h>
44 #include <sys/ddi.h>
45 #include <sys/suntpi.h>
46 #include <sys/cmn_err.h>
47 #include <sys/debug.h>
48 #include <sys/kobj.h>
49 #include <sys/modctl.h>
50 #include <sys/atomic.h>
51 #include <sys/policy.h>
52 #include <sys/priv.h>
53 #include <sys/taskq.h>
54 
55 #include <sys/systm.h>
56 #include <sys/param.h>
57 #include <sys/kmem.h>
58 #include <sys/sdt.h>
59 #include <sys/socket.h>
60 #include <sys/vtrace.h>
61 #include <sys/isa_defs.h>
62 #include <sys/mac.h>
63 #include <net/if.h>
64 #include <net/if_arp.h>
65 #include <net/route.h>
66 #include <sys/sockio.h>
67 #include <netinet/in.h>
68 #include <net/if_dl.h>
69 
70 #include <inet/common.h>
71 #include <inet/mi.h>
72 #include <inet/mib2.h>
73 #include <inet/nd.h>
74 #include <inet/arp.h>
75 #include <inet/snmpcom.h>
76 #include <inet/optcom.h>
77 #include <inet/kstatcom.h>
78 
79 #include <netinet/igmp_var.h>
80 #include <netinet/ip6.h>
81 #include <netinet/icmp6.h>
82 #include <netinet/sctp.h>
83 
84 #include <inet/ip.h>
85 #include <inet/ip_impl.h>
86 #include <inet/ip6.h>
87 #include <inet/ip6_asp.h>
88 #include <inet/tcp.h>
89 #include <inet/tcp_impl.h>
90 #include <inet/ip_multi.h>
91 #include <inet/ip_if.h>
92 #include <inet/ip_ire.h>
93 #include <inet/ip_ftable.h>
94 #include <inet/ip_rts.h>
95 #include <inet/ip_ndp.h>
96 #include <inet/ip_listutils.h>
97 #include <netinet/igmp.h>
98 #include <netinet/ip_mroute.h>
99 #include <inet/ipp_common.h>
100 #include <inet/cc.h>
101 
102 #include <net/pfkeyv2.h>
103 #include <inet/sadb.h>
104 #include <inet/ipsec_impl.h>
105 #include <inet/iptun/iptun_impl.h>
106 #include <inet/ipdrop.h>
107 #include <inet/ip_netinfo.h>
108 #include <inet/ilb_ip.h>
109 
110 #include <sys/ethernet.h>
111 #include <net/if_types.h>
112 #include <sys/cpuvar.h>
113 
114 #include <ipp/ipp.h>
115 #include <ipp/ipp_impl.h>
116 #include <ipp/ipgpc/ipgpc.h>
117 
118 #include <sys/pattr.h>
119 #include <inet/ipclassifier.h>
120 #include <inet/sctp_ip.h>
121 #include <inet/sctp/sctp_impl.h>
122 #include <inet/udp_impl.h>
123 #include <inet/rawip_impl.h>
124 #include <inet/rts_impl.h>
125 
126 #include <sys/tsol/label.h>
127 #include <sys/tsol/tnet.h>
128 
129 #include <sys/squeue_impl.h>
130 #include <inet/ip_arp.h>
131 
132 #include <sys/clock_impl.h>	/* For LBOLT_FASTPATH{,64} */
133 
134 /*
135  * Values for squeue switch:
136  * IP_SQUEUE_ENTER_NODRAIN: SQ_NODRAIN
137  * IP_SQUEUE_ENTER: SQ_PROCESS
138  * IP_SQUEUE_FILL: SQ_FILL
139  */
140 int ip_squeue_enter = IP_SQUEUE_ENTER;	/* Setable in /etc/system */
141 
142 int ip_squeue_flag;
143 
144 /*
145  * Setable in /etc/system
146  */
147 int ip_poll_normal_ms = 100;
148 int ip_poll_normal_ticks = 0;
149 int ip_modclose_ackwait_ms = 3000;
150 
151 /*
152  * It would be nice to have these present only in DEBUG systems, but the
153  * current design of the global symbol checking logic requires them to be
154  * unconditionally present.
155  */
156 uint_t ip_thread_data;			/* TSD key for debug support */
157 krwlock_t ip_thread_rwlock;
158 list_t	ip_thread_list;
159 
160 /*
161  * Structure to represent a linked list of msgblks. Used by ip_snmp_ functions.
162  */
163 
164 struct listptr_s {
165 	mblk_t	*lp_head;	/* pointer to the head of the list */
166 	mblk_t	*lp_tail;	/* pointer to the tail of the list */
167 };
168 
169 typedef struct listptr_s listptr_t;
170 
171 /*
172  * This is used by ip_snmp_get_mib2_ip_route_media and
173  * ip_snmp_get_mib2_ip6_route_media to carry the lists of return data.
174  */
175 typedef struct iproutedata_s {
176 	uint_t		ird_idx;
177 	uint_t		ird_flags;	/* see below */
178 	listptr_t	ird_route;	/* ipRouteEntryTable */
179 	listptr_t	ird_netmedia;	/* ipNetToMediaEntryTable */
180 	listptr_t	ird_attrs;	/* ipRouteAttributeTable */
181 } iproutedata_t;
182 
183 /* Include ire_testhidden and IRE_IF_CLONE routes */
184 #define	IRD_REPORT_ALL	0x01
185 
186 /*
187  * Cluster specific hooks. These should be NULL when booted as a non-cluster
188  */
189 
190 /*
191  * Hook functions to enable cluster networking
192  * On non-clustered systems these vectors must always be NULL.
193  *
194  * Hook function to Check ip specified ip address is a shared ip address
195  * in the cluster
196  *
197  */
198 int (*cl_inet_isclusterwide)(netstackid_t stack_id, uint8_t protocol,
199     sa_family_t addr_family, uint8_t *laddrp, void *args) = NULL;
200 
201 /*
202  * Hook function to generate cluster wide ip fragment identifier
203  */
204 uint32_t (*cl_inet_ipident)(netstackid_t stack_id, uint8_t protocol,
205     sa_family_t addr_family, uint8_t *laddrp, uint8_t *faddrp,
206     void *args) = NULL;
207 
208 /*
209  * Hook function to generate cluster wide SPI.
210  */
211 void (*cl_inet_getspi)(netstackid_t, uint8_t, uint8_t *, size_t,
212     void *) = NULL;
213 
214 /*
215  * Hook function to verify if the SPI is already utlized.
216  */
217 
218 int (*cl_inet_checkspi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
219 
220 /*
221  * Hook function to delete the SPI from the cluster wide repository.
222  */
223 
224 void (*cl_inet_deletespi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
225 
226 /*
227  * Hook function to inform the cluster when packet received on an IDLE SA
228  */
229 
230 void (*cl_inet_idlesa)(netstackid_t, uint8_t, uint32_t, sa_family_t,
231     in6_addr_t, in6_addr_t, void *) = NULL;
232 
233 /*
234  * Synchronization notes:
235  *
236  * IP is a fully D_MP STREAMS module/driver. Thus it does not depend on any
237  * MT level protection given by STREAMS. IP uses a combination of its own
238  * internal serialization mechanism and standard Solaris locking techniques.
239  * The internal serialization is per phyint.  This is used to serialize
240  * plumbing operations, IPMP operations, most set ioctls, etc.
241  *
242  * Plumbing is a long sequence of operations involving message
243  * exchanges between IP, ARP and device drivers. Many set ioctls are typically
244  * involved in plumbing operations. A natural model is to serialize these
245  * ioctls one per ill. For example plumbing of hme0 and qfe0 can go on in
246  * parallel without any interference. But various set ioctls on hme0 are best
247  * serialized, along with IPMP operations and processing of DLPI control
248  * messages received from drivers on a per phyint basis. This serialization is
249  * provided by the ipsq_t and primitives operating on this. Details can
250  * be found in ip_if.c above the core primitives operating on ipsq_t.
251  *
252  * Lookups of an ipif or ill by a thread return a refheld ipif / ill.
253  * Simiarly lookup of an ire by a thread also returns a refheld ire.
254  * In addition ipif's and ill's referenced by the ire are also indirectly
255  * refheld. Thus no ipif or ill can vanish as long as an ipif is refheld
256  * directly or indirectly. For example an SIOCSLIFADDR ioctl that changes the
257  * address of an ipif has to go through the ipsq_t. This ensures that only
258  * one such exclusive operation proceeds at any time on the ipif. It then
259  * waits for all refcnts
260  * associated with this ipif to come down to zero. The address is changed
261  * only after the ipif has been quiesced. Then the ipif is brought up again.
262  * More details are described above the comment in ip_sioctl_flags.
263  *
264  * Packet processing is based mostly on IREs and are fully multi-threaded
265  * using standard Solaris MT techniques.
266  *
267  * There are explicit locks in IP to handle:
268  * - The ip_g_head list maintained by mi_open_link() and friends.
269  *
270  * - The reassembly data structures (one lock per hash bucket)
271  *
272  * - conn_lock is meant to protect conn_t fields. The fields actually
273  *   protected by conn_lock are documented in the conn_t definition.
274  *
275  * - ire_lock to protect some of the fields of the ire, IRE tables
276  *   (one lock per hash bucket). Refer to ip_ire.c for details.
277  *
278  * - ndp_g_lock and ncec_lock for protecting NCEs.
279  *
280  * - ill_lock protects fields of the ill and ipif. Details in ip.h
281  *
282  * - ill_g_lock: This is a global reader/writer lock. Protects the following
283  *	* The AVL tree based global multi list of all ills.
284  *	* The linked list of all ipifs of an ill
285  *	* The <ipsq-xop> mapping
286  *	* <ill-phyint> association
287  *   Insertion/deletion of an ill in the system, insertion/deletion of an ipif
288  *   into an ill, changing the <ipsq-xop> mapping of an ill, changing the
289  *   <ill-phyint> assoc of an ill will all have to hold the ill_g_lock as
290  *   writer for the actual duration of the insertion/deletion/change.
291  *
292  * - ill_lock:  This is a per ill mutex.
293  *   It protects some members of the ill_t struct; see ip.h for details.
294  *   It also protects the <ill-phyint> assoc.
295  *   It also protects the list of ipifs hanging off the ill.
296  *
297  * - ipsq_lock: This is a per ipsq_t mutex lock.
298  *   This protects some members of the ipsq_t struct; see ip.h for details.
299  *   It also protects the <ipsq-ipxop> mapping
300  *
301  * - ipx_lock: This is a per ipxop_t mutex lock.
302  *   This protects some members of the ipxop_t struct; see ip.h for details.
303  *
304  * - phyint_lock: This is a per phyint mutex lock. Protects just the
305  *   phyint_flags
306  *
307  * - ip_addr_avail_lock: This is used to ensure the uniqueness of IP addresses.
308  *   This lock is held in ipif_up_done and the ipif is marked IPIF_UP and the
309  *   uniqueness check also done atomically.
310  *
311  * - ill_g_usesrc_lock: This readers/writer lock protects the usesrc
312  *   group list linked by ill_usesrc_grp_next. It also protects the
313  *   ill_usesrc_ifindex field. It is taken as a writer when a member of the
314  *   group is being added or deleted.  This lock is taken as a reader when
315  *   walking the list/group(eg: to get the number of members in a usesrc group).
316  *   Note, it is only necessary to take this lock if the ill_usesrc_grp_next
317  *   field is changing state i.e from NULL to non-NULL or vice-versa. For
318  *   example, it is not necessary to take this lock in the initial portion
319  *   of ip_sioctl_slifusesrc or at all in ip_sioctl_flags since these
320  *   operations are executed exclusively and that ensures that the "usesrc
321  *   group state" cannot change. The "usesrc group state" change can happen
322  *   only in the latter part of ip_sioctl_slifusesrc and in ill_delete.
323  *
324  * Changing <ill-phyint>, <ipsq-xop> assocications:
325  *
326  * To change the <ill-phyint> association, the ill_g_lock must be held
327  * as writer, and the ill_locks of both the v4 and v6 instance of the ill
328  * must be held.
329  *
330  * To change the <ipsq-xop> association, the ill_g_lock must be held as
331  * writer, the ipsq_lock must be held, and one must be writer on the ipsq.
332  * This is only done when ills are added or removed from IPMP groups.
333  *
334  * To add or delete an ipif from the list of ipifs hanging off the ill,
335  * ill_g_lock (writer) and ill_lock must be held and the thread must be
336  * a writer on the associated ipsq.
337  *
338  * To add or delete an ill to the system, the ill_g_lock must be held as
339  * writer and the thread must be a writer on the associated ipsq.
340  *
341  * To add or delete an ilm to an ill, the ill_lock must be held and the thread
342  * must be a writer on the associated ipsq.
343  *
344  * Lock hierarchy
345  *
346  * Some lock hierarchy scenarios are listed below.
347  *
348  * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock -> ipx_lock
349  * ill_g_lock -> ill_lock(s) -> phyint_lock
350  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock
351  * ill_g_lock -> ip_addr_avail_lock
352  * conn_lock -> irb_lock -> ill_lock -> ire_lock
353  * ill_g_lock -> ip_g_nd_lock
354  * ill_g_lock -> ips_ipmp_lock -> ill_lock -> nce_lock
355  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock -> nce_lock
356  * arl_lock -> ill_lock
357  * ips_ire_dep_lock -> irb_lock
358  *
359  * When more than 1 ill lock is needed to be held, all ill lock addresses
360  * are sorted on address and locked starting from highest addressed lock
361  * downward.
362  *
363  * Multicast scenarios
364  * ips_ill_g_lock -> ill_mcast_lock
365  * conn_ilg_lock -> ips_ill_g_lock -> ill_lock
366  * ill_mcast_serializer -> ill_mcast_lock -> ips_ipmp_lock -> ill_lock
367  * ill_mcast_serializer -> ill_mcast_lock -> connf_lock -> conn_lock
368  * ill_mcast_serializer -> ill_mcast_lock -> conn_ilg_lock
369  * ill_mcast_serializer -> ill_mcast_lock -> ips_igmp_timer_lock
370  *
371  * IPsec scenarios
372  *
373  * ipsa_lock -> ill_g_lock -> ill_lock
374  * ill_g_usesrc_lock -> ill_g_lock -> ill_lock
375  *
376  * Trusted Solaris scenarios
377  *
378  * igsa_lock -> gcgrp_rwlock -> gcgrp_lock
379  * igsa_lock -> gcdb_lock
380  * gcgrp_rwlock -> ire_lock
381  * gcgrp_rwlock -> gcdb_lock
382  *
383  * squeue(sq_lock), flow related (ft_lock, fe_lock) locking
384  *
385  * cpu_lock --> ill_lock --> sqset_lock --> sq_lock
386  * sq_lock -> conn_lock -> QLOCK(q)
387  * ill_lock -> ft_lock -> fe_lock
388  *
389  * Routing/forwarding table locking notes:
390  *
391  * Lock acquisition order: Radix tree lock, irb_lock.
392  * Requirements:
393  * i.  Walker must not hold any locks during the walker callback.
394  * ii  Walker must not see a truncated tree during the walk because of any node
395  *     deletion.
396  * iii Existing code assumes ire_bucket is valid if it is non-null and is used
397  *     in many places in the code to walk the irb list. Thus even if all the
398  *     ires in a bucket have been deleted, we still can't free the radix node
399  *     until the ires have actually been inactive'd (freed).
400  *
401  * Tree traversal - Need to hold the global tree lock in read mode.
402  * Before dropping the global tree lock, need to either increment the ire_refcnt
403  * to ensure that the radix node can't be deleted.
404  *
405  * Tree add - Need to hold the global tree lock in write mode to add a
406  * radix node. To prevent the node from being deleted, increment the
407  * irb_refcnt, after the node is added to the tree. The ire itself is
408  * added later while holding the irb_lock, but not the tree lock.
409  *
410  * Tree delete - Need to hold the global tree lock and irb_lock in write mode.
411  * All associated ires must be inactive (i.e. freed), and irb_refcnt
412  * must be zero.
413  *
414  * Walker - Increment irb_refcnt before calling the walker callback. Hold the
415  * global tree lock (read mode) for traversal.
416  *
417  * IRE dependencies - In some cases we hold ips_ire_dep_lock across ire_refrele
418  * hence we will acquire irb_lock while holding ips_ire_dep_lock.
419  *
420  * IPsec notes :
421  *
422  * IP interacts with the IPsec code (AH/ESP) by storing IPsec attributes
423  * in the ip_xmit_attr_t ip_recv_attr_t. For outbound datagrams, the
424  * ip_xmit_attr_t has the
425  * information used by the IPsec code for applying the right level of
426  * protection. The information initialized by IP in the ip_xmit_attr_t
427  * is determined by the per-socket policy or global policy in the system.
428  * For inbound datagrams, the ip_recv_attr_t
429  * starts out with nothing in it. It gets filled
430  * with the right information if it goes through the AH/ESP code, which
431  * happens if the incoming packet is secure. The information initialized
432  * by AH/ESP, is later used by IP (during fanouts to ULP) to see whether
433  * the policy requirements needed by per-socket policy or global policy
434  * is met or not.
435  *
436  * For fully connected sockets i.e dst, src [addr, port] is known,
437  * conn_policy_cached is set indicating that policy has been cached.
438  * conn_in_enforce_policy may or may not be set depending on whether
439  * there is a global policy match or per-socket policy match.
440  * Policy inheriting happpens in ip_policy_set once the destination is known.
441  * Once the right policy is set on the conn_t, policy cannot change for
442  * this socket. This makes life simpler for TCP (UDP ?) where
443  * re-transmissions go out with the same policy. For symmetry, policy
444  * is cached for fully connected UDP sockets also. Thus if policy is cached,
445  * it also implies that policy is latched i.e policy cannot change
446  * on these sockets. As we have the right policy on the conn, we don't
447  * have to lookup global policy for every outbound and inbound datagram
448  * and thus serving as an optimization. Note that a global policy change
449  * does not affect fully connected sockets if they have policy. If fully
450  * connected sockets did not have any policy associated with it, global
451  * policy change may affect them.
452  *
453  * IP Flow control notes:
454  * ---------------------
455  * Non-TCP streams are flow controlled by IP. The way this is accomplished
456  * differs when ILL_CAPAB_DLD_DIRECT is enabled for that IP instance. When
457  * ILL_DIRECT_CAPABLE(ill) is TRUE, IP can do direct function calls into
458  * GLDv3. Otherwise packets are sent down to lower layers using STREAMS
459  * functions.
460  *
461  * Per Tx ring udp flow control:
462  * This is applicable only when ILL_CAPAB_DLD_DIRECT capability is set in
463  * the ill (i.e. ILL_DIRECT_CAPABLE(ill) is true).
464  *
465  * The underlying link can expose multiple Tx rings to the GLDv3 mac layer.
466  * To achieve best performance, outgoing traffic need to be fanned out among
467  * these Tx ring. mac_tx() is called (via str_mdata_fastpath_put()) to send
468  * traffic out of the NIC and it takes a fanout hint. UDP connections pass
469  * the address of connp as fanout hint to mac_tx(). Under flow controlled
470  * condition, mac_tx() returns a non-NULL cookie (ip_mac_tx_cookie_t). This
471  * cookie points to a specific Tx ring that is blocked. The cookie is used to
472  * hash into an idl_tx_list[] entry in idl_tx_list[] array. Each idl_tx_list_t
473  * point to drain_lists (idl_t's). These drain list will store the blocked UDP
474  * connp's. The drain list is not a single list but a configurable number of
475  * lists.
476  *
477  * The diagram below shows idl_tx_list_t's and their drain_lists. ip_stack_t
478  * has an array of idl_tx_list_t. The size of the array is TX_FANOUT_SIZE
479  * which is equal to 128. This array in turn contains a pointer to idl_t[],
480  * the ip drain list. The idl_t[] array size is MIN(max_ncpus, 8). The drain
481  * list will point to the list of connp's that are flow controlled.
482  *
483  *                      ---------------   -------   -------   -------
484  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
485  *                   |  ---------------   -------   -------   -------
486  *                   |  ---------------   -------   -------   -------
487  *                   |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
488  * ----------------  |  ---------------   -------   -------   -------
489  * |idl_tx_list[0]|->|  ---------------   -------   -------   -------
490  * ----------------  |->|drain_list[2]|-->|connp|-->|connp|-->|connp|-->
491  *                   |  ---------------   -------   -------   -------
492  *                   .        .              .         .         .
493  *                   |  ---------------   -------   -------   -------
494  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
495  *                      ---------------   -------   -------   -------
496  *                      ---------------   -------   -------   -------
497  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
498  *                   |  ---------------   -------   -------   -------
499  *                   |  ---------------   -------   -------   -------
500  * ----------------  |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
501  * |idl_tx_list[1]|->|  ---------------   -------   -------   -------
502  * ----------------  |        .              .         .         .
503  *                   |  ---------------   -------   -------   -------
504  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
505  *                      ---------------   -------   -------   -------
506  *     .....
507  * ----------------
508  * |idl_tx_list[n]|-> ...
509  * ----------------
510  *
511  * When mac_tx() returns a cookie, the cookie is hashed into an index into
512  * ips_idl_tx_list[], and conn_drain_insert() is called with the idl_tx_list
513  * to insert the conn onto.  conn_drain_insert() asserts flow control for the
514  * sockets via su_txq_full() (non-STREAMS) or QFULL on conn_wq (STREAMS).
515  * Further, conn_blocked is set to indicate that the conn is blocked.
516  *
517  * GLDv3 calls ill_flow_enable() when flow control is relieved.  The cookie
518  * passed in the call to ill_flow_enable() identifies the blocked Tx ring and
519  * is again hashed to locate the appropriate idl_tx_list, which is then
520  * drained via conn_walk_drain().  conn_walk_drain() goes through each conn in
521  * the drain list and calls conn_drain_remove() to clear flow control (via
522  * calling su_txq_full() or clearing QFULL), and remove the conn from the
523  * drain list.
524  *
525  * Note that the drain list is not a single list but a (configurable) array of
526  * lists (8 elements by default).  Synchronization between drain insertion and
527  * flow control wakeup is handled by using idl_txl->txl_lock, and only
528  * conn_drain_insert() and conn_drain_remove() manipulate the drain list.
529  *
530  * Flow control via STREAMS is used when ILL_DIRECT_CAPABLE() returns FALSE.
531  * On the send side, if the packet cannot be sent down to the driver by IP
532  * (canput() fails), ip_xmit() drops the packet and returns EWOULDBLOCK to the
533  * caller, who may then invoke ixa_check_drain_insert() to insert the conn on
534  * the 0'th drain list.  When ip_wsrv() runs on the ill_wq because flow
535  * control has been relieved, the blocked conns in the 0'th drain list are
536  * drained as in the non-STREAMS case.
537  *
538  * In both the STREAMS and non-STREAMS cases, the sockfs upcall to set QFULL
539  * is done when the conn is inserted into the drain list (conn_drain_insert())
540  * and cleared when the conn is removed from the it (conn_drain_remove()).
541  *
542  * IPQOS notes:
543  *
544  * IPQoS Policies are applied to packets using IPPF (IP Policy framework)
545  * and IPQoS modules. IPPF includes hooks in IP at different control points
546  * (callout positions) which direct packets to IPQoS modules for policy
547  * processing. Policies, if present, are global.
548  *
549  * The callout positions are located in the following paths:
550  *		o local_in (packets destined for this host)
551  *		o local_out (packets orginating from this host )
552  *		o fwd_in  (packets forwarded by this m/c - inbound)
553  *		o fwd_out (packets forwarded by this m/c - outbound)
554  * Hooks at these callout points can be enabled/disabled using the ndd variable
555  * ip_policy_mask (a bit mask with the 4 LSB indicating the callout positions).
556  * By default all the callout positions are enabled.
557  *
558  * Outbound (local_out)
559  * Hooks are placed in ire_send_wire_v4 and ire_send_wire_v6.
560  *
561  * Inbound (local_in)
562  * Hooks are placed in ip_fanout_v4 and ip_fanout_v6.
563  *
564  * Forwarding (in and out)
565  * Hooks are placed in ire_recv_forward_v4/v6.
566  *
567  * IP Policy Framework processing (IPPF processing)
568  * Policy processing for a packet is initiated by ip_process, which ascertains
569  * that the classifier (ipgpc) is loaded and configured, failing which the
570  * packet resumes normal processing in IP. If the clasifier is present, the
571  * packet is acted upon by one or more IPQoS modules (action instances), per
572  * filters configured in ipgpc and resumes normal IP processing thereafter.
573  * An action instance can drop a packet in course of its processing.
574  *
575  * Zones notes:
576  *
577  * The partitioning rules for networking are as follows:
578  * 1) Packets coming from a zone must have a source address belonging to that
579  * zone.
580  * 2) Packets coming from a zone can only be sent on a physical interface on
581  * which the zone has an IP address.
582  * 3) Between two zones on the same machine, packet delivery is only allowed if
583  * there's a matching route for the destination and zone in the forwarding
584  * table.
585  * 4) The TCP and UDP port spaces are per-zone; that is, two processes in
586  * different zones can bind to the same port with the wildcard address
587  * (INADDR_ANY).
588  *
589  * The granularity of interface partitioning is at the logical interface level.
590  * Therefore, every zone has its own IP addresses, and incoming packets can be
591  * attributed to a zone unambiguously. A logical interface is placed into a zone
592  * using the SIOCSLIFZONE ioctl; this sets the ipif_zoneid field in the ipif_t
593  * structure. Rule (1) is implemented by modifying the source address selection
594  * algorithm so that the list of eligible addresses is filtered based on the
595  * sending process zone.
596  *
597  * The Internet Routing Entries (IREs) are either exclusive to a zone or shared
598  * across all zones, depending on their type. Here is the break-up:
599  *
600  * IRE type				Shared/exclusive
601  * --------				----------------
602  * IRE_BROADCAST			Exclusive
603  * IRE_DEFAULT (default routes)		Shared (*)
604  * IRE_LOCAL				Exclusive (x)
605  * IRE_LOOPBACK				Exclusive
606  * IRE_PREFIX (net routes)		Shared (*)
607  * IRE_IF_NORESOLVER (interface routes)	Exclusive
608  * IRE_IF_RESOLVER (interface routes)	Exclusive
609  * IRE_IF_CLONE (interface routes)	Exclusive
610  * IRE_HOST (host routes)		Shared (*)
611  *
612  * (*) A zone can only use a default or off-subnet route if the gateway is
613  * directly reachable from the zone, that is, if the gateway's address matches
614  * one of the zone's logical interfaces.
615  *
616  * (x) IRE_LOCAL are handled a bit differently.
617  * When ip_restrict_interzone_loopback is set (the default),
618  * ire_route_recursive restricts loopback using an IRE_LOCAL
619  * between zone to the case when L2 would have conceptually looped the packet
620  * back, i.e. the loopback which is required since neither Ethernet drivers
621  * nor Ethernet hardware loops them back. This is the case when the normal
622  * routes (ignoring IREs with different zoneids) would send out the packet on
623  * the same ill as the ill with which is IRE_LOCAL is associated.
624  *
625  * Multiple zones can share a common broadcast address; typically all zones
626  * share the 255.255.255.255 address. Incoming as well as locally originated
627  * broadcast packets must be dispatched to all the zones on the broadcast
628  * network. For directed broadcasts (e.g. 10.16.72.255) this is not trivial
629  * since some zones may not be on the 10.16.72/24 network. To handle this, each
630  * zone has its own set of IRE_BROADCAST entries; then, broadcast packets are
631  * sent to every zone that has an IRE_BROADCAST entry for the destination
632  * address on the input ill, see ip_input_broadcast().
633  *
634  * Applications in different zones can join the same multicast group address.
635  * The same logic applies for multicast as for broadcast. ip_input_multicast
636  * dispatches packets to all zones that have members on the physical interface.
637  */
638 
639 /*
640  * Squeue Fanout flags:
641  *	0: No fanout.
642  *	1: Fanout across all squeues
643  */
644 boolean_t	ip_squeue_fanout = 0;
645 
646 /*
647  * Maximum dups allowed per packet.
648  */
649 uint_t ip_max_frag_dups = 10;
650 
651 static int	ip_open(queue_t *q, dev_t *devp, int flag, int sflag,
652 		    cred_t *credp, boolean_t isv6);
653 static mblk_t	*ip_xmit_attach_llhdr(mblk_t *, nce_t *);
654 
655 static boolean_t icmp_inbound_verify_v4(mblk_t *, icmph_t *, ip_recv_attr_t *);
656 static void	icmp_inbound_too_big_v4(icmph_t *, ip_recv_attr_t *);
657 static void	icmp_inbound_error_fanout_v4(mblk_t *, icmph_t *,
658     ip_recv_attr_t *);
659 static void	icmp_options_update(ipha_t *);
660 static void	icmp_param_problem(mblk_t *, uint8_t,  ip_recv_attr_t *);
661 static void	icmp_pkt(mblk_t *, void *, size_t, ip_recv_attr_t *);
662 static mblk_t	*icmp_pkt_err_ok(mblk_t *, ip_recv_attr_t *);
663 static void	icmp_redirect_v4(mblk_t *mp, ipha_t *, icmph_t *,
664     ip_recv_attr_t *);
665 static void	icmp_send_redirect(mblk_t *, ipaddr_t, ip_recv_attr_t *);
666 static void	icmp_send_reply_v4(mblk_t *, ipha_t *, icmph_t *,
667     ip_recv_attr_t *);
668 
669 mblk_t		*ip_dlpi_alloc(size_t, t_uscalar_t);
670 char		*ip_dot_addr(ipaddr_t, char *);
671 mblk_t		*ip_carve_mp(mblk_t **, ssize_t);
672 static char	*ip_dot_saddr(uchar_t *, char *);
673 static int	ip_lrput(queue_t *, mblk_t *);
674 ipaddr_t	ip_net_mask(ipaddr_t);
675 char		*ip_nv_lookup(nv_t *, int);
676 int		ip_rput(queue_t *, mblk_t *);
677 static void	ip_rput_dlpi_writer(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
678 		    void *dummy_arg);
679 int		ip_snmp_get(queue_t *, mblk_t *, int, boolean_t);
680 static mblk_t	*ip_snmp_get_mib2_ip(queue_t *, mblk_t *,
681 		    mib2_ipIfStatsEntry_t *, ip_stack_t *, boolean_t);
682 static mblk_t	*ip_snmp_get_mib2_ip_traffic_stats(queue_t *, mblk_t *,
683 		    ip_stack_t *, boolean_t);
684 static mblk_t	*ip_snmp_get_mib2_ip6(queue_t *, mblk_t *, ip_stack_t *,
685 		    boolean_t);
686 static mblk_t	*ip_snmp_get_mib2_icmp(queue_t *, mblk_t *, ip_stack_t *ipst);
687 static mblk_t	*ip_snmp_get_mib2_icmp6(queue_t *, mblk_t *, ip_stack_t *ipst);
688 static mblk_t	*ip_snmp_get_mib2_igmp(queue_t *, mblk_t *, ip_stack_t *ipst);
689 static mblk_t	*ip_snmp_get_mib2_multi(queue_t *, mblk_t *, ip_stack_t *ipst);
690 static mblk_t	*ip_snmp_get_mib2_ip_addr(queue_t *, mblk_t *,
691 		    ip_stack_t *ipst, boolean_t);
692 static mblk_t	*ip_snmp_get_mib2_ip6_addr(queue_t *, mblk_t *,
693 		    ip_stack_t *ipst, boolean_t);
694 static mblk_t	*ip_snmp_get_mib2_ip_group_src(queue_t *, mblk_t *,
695 		    ip_stack_t *ipst);
696 static mblk_t	*ip_snmp_get_mib2_ip6_group_src(queue_t *, mblk_t *,
697 		    ip_stack_t *ipst);
698 static mblk_t	*ip_snmp_get_mib2_ip_group_mem(queue_t *, mblk_t *,
699 		    ip_stack_t *ipst);
700 static mblk_t	*ip_snmp_get_mib2_ip6_group_mem(queue_t *, mblk_t *,
701 		    ip_stack_t *ipst);
702 static mblk_t	*ip_snmp_get_mib2_virt_multi(queue_t *, mblk_t *,
703 		    ip_stack_t *ipst);
704 static mblk_t	*ip_snmp_get_mib2_multi_rtable(queue_t *, mblk_t *,
705 		    ip_stack_t *ipst);
706 static mblk_t	*ip_snmp_get_mib2_ip_route_media(queue_t *, mblk_t *, int,
707 		    ip_stack_t *ipst);
708 static mblk_t	*ip_snmp_get_mib2_ip6_route_media(queue_t *, mblk_t *, int,
709 		    ip_stack_t *ipst);
710 static void	ip_snmp_get2_v4(ire_t *, iproutedata_t *);
711 static void	ip_snmp_get2_v6_route(ire_t *, iproutedata_t *);
712 static void	ip_snmp_get2_v4_media(ncec_t *, void *);
713 static void	ip_snmp_get2_v6_media(ncec_t *, void *);
714 int		ip_snmp_set(queue_t *, int, int, uchar_t *, int);
715 
716 static mblk_t	*ip_fragment_copyhdr(uchar_t *, int, int, ip_stack_t *,
717 		    mblk_t *);
718 
719 static void	conn_drain_init(ip_stack_t *);
720 static void	conn_drain_fini(ip_stack_t *);
721 static void	conn_drain(conn_t *connp, boolean_t closing);
722 
723 static void	conn_walk_drain(ip_stack_t *, idl_tx_list_t *);
724 static void	conn_walk_sctp(pfv_t, void *, zoneid_t, netstack_t *);
725 
726 static void	*ip_stack_init(netstackid_t stackid, netstack_t *ns);
727 static void	ip_stack_shutdown(netstackid_t stackid, void *arg);
728 static void	ip_stack_fini(netstackid_t stackid, void *arg);
729 
730 static int	ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
731     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
732     ire_t *, conn_t *, boolean_t, const in6_addr_t *,  mcast_record_t,
733     const in6_addr_t *);
734 
735 static int	ip_squeue_switch(int);
736 
737 static void	*ip_kstat_init(netstackid_t, ip_stack_t *);
738 static void	ip_kstat_fini(netstackid_t, kstat_t *);
739 static int	ip_kstat_update(kstat_t *kp, int rw);
740 static void	*icmp_kstat_init(netstackid_t);
741 static void	icmp_kstat_fini(netstackid_t, kstat_t *);
742 static int	icmp_kstat_update(kstat_t *kp, int rw);
743 static void	*ip_kstat2_init(netstackid_t, ip_stat_t *);
744 static void	ip_kstat2_fini(netstackid_t, kstat_t *);
745 
746 static void	ipobs_init(ip_stack_t *);
747 static void	ipobs_fini(ip_stack_t *);
748 
749 static int	ip_tp_cpu_update(cpu_setup_t, int, void *);
750 
751 ipaddr_t	ip_g_all_ones = IP_HOST_MASK;
752 
753 static long ip_rput_pullups;
754 int	dohwcksum = 1;	/* use h/w cksum if supported by the hardware */
755 
756 vmem_t *ip_minor_arena_sa; /* for minor nos. from INET_MIN_DEV+2 thru 2^^18-1 */
757 vmem_t *ip_minor_arena_la; /* for minor nos. from 2^^18 thru 2^^32-1 */
758 
759 int	ip_debug;
760 
761 /*
762  * Multirouting/CGTP stuff
763  */
764 int	ip_cgtp_filter_rev = CGTP_FILTER_REV;	/* CGTP hooks version */
765 
766 /*
767  * IP tunables related declarations. Definitions are in ip_tunables.c
768  */
769 extern mod_prop_info_t ip_propinfo_tbl[];
770 extern int ip_propinfo_count;
771 
772 /*
773  * Table of IP ioctls encoding the various properties of the ioctl and
774  * indexed based on the last byte of the ioctl command. Occasionally there
775  * is a clash, and there is more than 1 ioctl with the same last byte.
776  * In such a case 1 ioctl is encoded in the ndx table and the remaining
777  * ioctls are encoded in the misc table. An entry in the ndx table is
778  * retrieved by indexing on the last byte of the ioctl command and comparing
779  * the ioctl command with the value in the ndx table. In the event of a
780  * mismatch the misc table is then searched sequentially for the desired
781  * ioctl command.
782  *
783  * Entry: <command> <copyin_size> <flags> <cmd_type> <function> <restart_func>
784  */
785 ip_ioctl_cmd_t ip_ndx_ioctl_table[] = {
786 	/* 000 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
787 	/* 001 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
788 	/* 002 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
789 	/* 003 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
790 	/* 004 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
791 	/* 005 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
792 	/* 006 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
793 	/* 007 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
794 	/* 008 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
795 	/* 009 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
796 
797 	/* 010 */ { SIOCADDRT,	sizeof (struct rtentry), IPI_PRIV,
798 			MISC_CMD, ip_siocaddrt, NULL },
799 	/* 011 */ { SIOCDELRT,	sizeof (struct rtentry), IPI_PRIV,
800 			MISC_CMD, ip_siocdelrt, NULL },
801 
802 	/* 012 */ { SIOCSIFADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
803 			IF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
804 	/* 013 */ { SIOCGIFADDR, sizeof (struct ifreq), IPI_GET_CMD,
805 			IF_CMD, ip_sioctl_get_addr, NULL },
806 
807 	/* 014 */ { SIOCSIFDSTADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
808 			IF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
809 	/* 015 */ { SIOCGIFDSTADDR, sizeof (struct ifreq),
810 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_dstaddr, NULL },
811 
812 	/* 016 */ { SIOCSIFFLAGS, sizeof (struct ifreq),
813 			IPI_PRIV | IPI_WR,
814 			IF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
815 	/* 017 */ { SIOCGIFFLAGS, sizeof (struct ifreq),
816 			IPI_MODOK | IPI_GET_CMD,
817 			IF_CMD, ip_sioctl_get_flags, NULL },
818 
819 	/* 018 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
820 	/* 019 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
821 
822 	/* copyin size cannot be coded for SIOCGIFCONF */
823 	/* 020 */ { O_SIOCGIFCONF, 0, IPI_GET_CMD,
824 			MISC_CMD, ip_sioctl_get_ifconf, NULL },
825 
826 	/* 021 */ { SIOCSIFMTU,	sizeof (struct ifreq), IPI_PRIV | IPI_WR,
827 			IF_CMD, ip_sioctl_mtu, NULL },
828 	/* 022 */ { SIOCGIFMTU,	sizeof (struct ifreq), IPI_GET_CMD,
829 			IF_CMD, ip_sioctl_get_mtu, NULL },
830 	/* 023 */ { SIOCGIFBRDADDR, sizeof (struct ifreq),
831 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_brdaddr, NULL },
832 	/* 024 */ { SIOCSIFBRDADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
833 			IF_CMD, ip_sioctl_brdaddr, NULL },
834 	/* 025 */ { SIOCGIFNETMASK, sizeof (struct ifreq),
835 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_netmask, NULL },
836 	/* 026 */ { SIOCSIFNETMASK, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
837 			IF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
838 	/* 027 */ { SIOCGIFMETRIC, sizeof (struct ifreq),
839 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_metric, NULL },
840 	/* 028 */ { SIOCSIFMETRIC, sizeof (struct ifreq), IPI_PRIV,
841 			IF_CMD, ip_sioctl_metric, NULL },
842 	/* 029 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
843 
844 	/* See 166-168 below for extended SIOC*XARP ioctls */
845 	/* 030 */ { SIOCSARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
846 			ARP_CMD, ip_sioctl_arp, NULL },
847 	/* 031 */ { SIOCGARP, sizeof (struct arpreq), IPI_GET_CMD,
848 			ARP_CMD, ip_sioctl_arp, NULL },
849 	/* 032 */ { SIOCDARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
850 			ARP_CMD, ip_sioctl_arp, NULL },
851 
852 	/* 033 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
853 	/* 034 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
854 	/* 035 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
855 	/* 036 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
856 	/* 037 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
857 	/* 038 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
858 	/* 039 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
859 	/* 040 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
860 	/* 041 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
861 	/* 042 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
862 	/* 043 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
863 	/* 044 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
864 	/* 045 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
865 	/* 046 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
866 	/* 047 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
867 	/* 048 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
868 	/* 049 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
869 	/* 050 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
870 	/* 051 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
871 	/* 052 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
872 	/* 053 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
873 
874 	/* 054 */ { IF_UNITSEL,	sizeof (int), IPI_PRIV | IPI_WR | IPI_MODOK,
875 			MISC_CMD, if_unitsel, if_unitsel_restart },
876 
877 	/* 055 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
878 	/* 056 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
879 	/* 057 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
880 	/* 058 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
881 	/* 059 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
882 	/* 060 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
883 	/* 061 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
884 	/* 062 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
885 	/* 063 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
886 	/* 064 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
887 	/* 065 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
888 	/* 066 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
889 	/* 067 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
890 	/* 068 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
891 	/* 069 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
892 	/* 070 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
893 	/* 071 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
894 	/* 072 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
895 
896 	/* 073 */ { SIOCSIFNAME, sizeof (struct ifreq),
897 			IPI_PRIV | IPI_WR | IPI_MODOK,
898 			IF_CMD, ip_sioctl_sifname, NULL },
899 
900 	/* 074 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
901 	/* 075 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
902 	/* 076 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
903 	/* 077 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
904 	/* 078 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
905 	/* 079 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
906 	/* 080 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
907 	/* 081 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
908 	/* 082 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
909 	/* 083 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
910 	/* 084 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
911 	/* 085 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
912 	/* 086 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
913 
914 	/* 087 */ { SIOCGIFNUM, sizeof (int), IPI_GET_CMD,
915 			MISC_CMD, ip_sioctl_get_ifnum, NULL },
916 	/* 088 */ { SIOCGIFMUXID, sizeof (struct ifreq), IPI_GET_CMD,
917 			IF_CMD, ip_sioctl_get_muxid, NULL },
918 	/* 089 */ { SIOCSIFMUXID, sizeof (struct ifreq),
919 			IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_muxid, NULL },
920 
921 	/* Both if and lif variants share same func */
922 	/* 090 */ { SIOCGIFINDEX, sizeof (struct ifreq), IPI_GET_CMD,
923 			IF_CMD, ip_sioctl_get_lifindex, NULL },
924 	/* Both if and lif variants share same func */
925 	/* 091 */ { SIOCSIFINDEX, sizeof (struct ifreq),
926 			IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_slifindex, NULL },
927 
928 	/* copyin size cannot be coded for SIOCGIFCONF */
929 	/* 092 */ { SIOCGIFCONF, 0, IPI_GET_CMD,
930 			MISC_CMD, ip_sioctl_get_ifconf, NULL },
931 	/* 093 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
932 	/* 094 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
933 	/* 095 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
934 	/* 096 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
935 	/* 097 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
936 	/* 098 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
937 	/* 099 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
938 	/* 100 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
939 	/* 101 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
940 	/* 102 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
941 	/* 103 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
942 	/* 104 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
943 	/* 105 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
944 	/* 106 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
945 	/* 107 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
946 	/* 108 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
947 	/* 109 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
948 
949 	/* 110 */ { SIOCLIFREMOVEIF, sizeof (struct lifreq),
950 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_removeif,
951 			ip_sioctl_removeif_restart },
952 	/* 111 */ { SIOCLIFADDIF, sizeof (struct lifreq),
953 			IPI_GET_CMD | IPI_PRIV | IPI_WR,
954 			LIF_CMD, ip_sioctl_addif, NULL },
955 #define	SIOCLIFADDR_NDX 112
956 	/* 112 */ { SIOCSLIFADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
957 			LIF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
958 	/* 113 */ { SIOCGLIFADDR, sizeof (struct lifreq),
959 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_addr, NULL },
960 	/* 114 */ { SIOCSLIFDSTADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
961 			LIF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
962 	/* 115 */ { SIOCGLIFDSTADDR, sizeof (struct lifreq),
963 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dstaddr, NULL },
964 	/* 116 */ { SIOCSLIFFLAGS, sizeof (struct lifreq),
965 			IPI_PRIV | IPI_WR,
966 			LIF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
967 	/* 117 */ { SIOCGLIFFLAGS, sizeof (struct lifreq),
968 			IPI_GET_CMD | IPI_MODOK,
969 			LIF_CMD, ip_sioctl_get_flags, NULL },
970 
971 	/* 118 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
972 	/* 119 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
973 
974 	/* 120 */ { O_SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
975 			ip_sioctl_get_lifconf, NULL },
976 	/* 121 */ { SIOCSLIFMTU, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
977 			LIF_CMD, ip_sioctl_mtu, NULL },
978 	/* 122 */ { SIOCGLIFMTU, sizeof (struct lifreq), IPI_GET_CMD,
979 			LIF_CMD, ip_sioctl_get_mtu, NULL },
980 	/* 123 */ { SIOCGLIFBRDADDR, sizeof (struct lifreq),
981 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_brdaddr, NULL },
982 	/* 124 */ { SIOCSLIFBRDADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
983 			LIF_CMD, ip_sioctl_brdaddr, NULL },
984 	/* 125 */ { SIOCGLIFNETMASK, sizeof (struct lifreq),
985 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_netmask, NULL },
986 	/* 126 */ { SIOCSLIFNETMASK, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
987 			LIF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
988 	/* 127 */ { SIOCGLIFMETRIC, sizeof (struct lifreq),
989 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_metric, NULL },
990 	/* 128 */ { SIOCSLIFMETRIC, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
991 			LIF_CMD, ip_sioctl_metric, NULL },
992 	/* 129 */ { SIOCSLIFNAME, sizeof (struct lifreq),
993 			IPI_PRIV | IPI_WR | IPI_MODOK,
994 			LIF_CMD, ip_sioctl_slifname,
995 			ip_sioctl_slifname_restart },
996 
997 	/* 130 */ { SIOCGLIFNUM, sizeof (struct lifnum), IPI_GET_CMD,
998 			MISC_CMD, ip_sioctl_get_lifnum, NULL },
999 	/* 131 */ { SIOCGLIFMUXID, sizeof (struct lifreq),
1000 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_muxid, NULL },
1001 	/* 132 */ { SIOCSLIFMUXID, sizeof (struct lifreq),
1002 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_muxid, NULL },
1003 	/* 133 */ { SIOCGLIFINDEX, sizeof (struct lifreq),
1004 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifindex, 0 },
1005 	/* 134 */ { SIOCSLIFINDEX, sizeof (struct lifreq),
1006 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifindex, 0 },
1007 	/* 135 */ { SIOCSLIFTOKEN, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1008 			LIF_CMD, ip_sioctl_token, NULL },
1009 	/* 136 */ { SIOCGLIFTOKEN, sizeof (struct lifreq),
1010 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_token, NULL },
1011 	/* 137 */ { SIOCSLIFSUBNET, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1012 			LIF_CMD, ip_sioctl_subnet, ip_sioctl_subnet_restart },
1013 	/* 138 */ { SIOCGLIFSUBNET, sizeof (struct lifreq),
1014 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_subnet, NULL },
1015 	/* 139 */ { SIOCSLIFLNKINFO, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1016 			LIF_CMD, ip_sioctl_lnkinfo, NULL },
1017 
1018 	/* 140 */ { SIOCGLIFLNKINFO, sizeof (struct lifreq),
1019 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lnkinfo, NULL },
1020 	/* 141 */ { SIOCLIFDELND, sizeof (struct lifreq), IPI_PRIV,
1021 			LIF_CMD, ip_siocdelndp_v6, NULL },
1022 	/* 142 */ { SIOCLIFGETND, sizeof (struct lifreq), IPI_GET_CMD,
1023 			LIF_CMD, ip_siocqueryndp_v6, NULL },
1024 	/* 143 */ { SIOCLIFSETND, sizeof (struct lifreq), IPI_PRIV,
1025 			LIF_CMD, ip_siocsetndp_v6, NULL },
1026 	/* 144 */ { SIOCTMYADDR, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1027 			MISC_CMD, ip_sioctl_tmyaddr, NULL },
1028 	/* 145 */ { SIOCTONLINK, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1029 			MISC_CMD, ip_sioctl_tonlink, NULL },
1030 	/* 146 */ { SIOCTMYSITE, sizeof (struct sioc_addrreq), 0,
1031 			MISC_CMD, ip_sioctl_tmysite, NULL },
1032 	/* 147 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1033 	/* 148 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1034 
1035 	/* Old *IPSECONFIG ioctls are now deprecated, now see spdsock.c */
1036 	/* 149 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1037 	/* 150 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1038 	/* 151 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1039 	/* 152 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1040 
1041 	/* 153 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1042 
1043 	/* 154 */ { SIOCGLIFBINDING, sizeof (struct lifreq), IPI_GET_CMD,
1044 			LIF_CMD, ip_sioctl_get_binding, NULL },
1045 	/* 155 */ { SIOCSLIFGROUPNAME, sizeof (struct lifreq),
1046 			IPI_PRIV | IPI_WR,
1047 			LIF_CMD, ip_sioctl_groupname, ip_sioctl_groupname },
1048 	/* 156 */ { SIOCGLIFGROUPNAME, sizeof (struct lifreq),
1049 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_groupname, NULL },
1050 	/* 157 */ { SIOCGLIFGROUPINFO, sizeof (lifgroupinfo_t),
1051 			IPI_GET_CMD, MISC_CMD, ip_sioctl_groupinfo, NULL },
1052 
1053 	/* Leave 158-160 unused; used to be SIOC*IFARP ioctls */
1054 	/* 158 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1055 	/* 159 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1056 	/* 160 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1057 
1058 	/* 161 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1059 
1060 	/* These are handled in ip_sioctl_copyin_setup itself */
1061 	/* 162 */ { SIOCGIP6ADDRPOLICY, 0, IPI_NULL_BCONT,
1062 			MISC_CMD, NULL, NULL },
1063 	/* 163 */ { SIOCSIP6ADDRPOLICY, 0, IPI_PRIV | IPI_NULL_BCONT,
1064 			MISC_CMD, NULL, NULL },
1065 	/* 164 */ { SIOCGDSTINFO, 0, IPI_GET_CMD, MISC_CMD, NULL, NULL },
1066 
1067 	/* 165 */ { SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
1068 			ip_sioctl_get_lifconf, NULL },
1069 
1070 	/* 166 */ { SIOCSXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1071 			XARP_CMD, ip_sioctl_arp, NULL },
1072 	/* 167 */ { SIOCGXARP, sizeof (struct xarpreq), IPI_GET_CMD,
1073 			XARP_CMD, ip_sioctl_arp, NULL },
1074 	/* 168 */ { SIOCDXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1075 			XARP_CMD, ip_sioctl_arp, NULL },
1076 
1077 	/* SIOCPOPSOCKFS is not handled by IP */
1078 	/* 169 */ { IPI_DONTCARE /* SIOCPOPSOCKFS */, 0, 0, 0, NULL, NULL },
1079 
1080 	/* 170 */ { SIOCGLIFZONE, sizeof (struct lifreq),
1081 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifzone, NULL },
1082 	/* 171 */ { SIOCSLIFZONE, sizeof (struct lifreq),
1083 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifzone,
1084 			ip_sioctl_slifzone_restart },
1085 	/* 172-174 are SCTP ioctls and not handled by IP */
1086 	/* 172 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1087 	/* 173 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1088 	/* 174 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1089 	/* 175 */ { SIOCGLIFUSESRC, sizeof (struct lifreq),
1090 			IPI_GET_CMD, LIF_CMD,
1091 			ip_sioctl_get_lifusesrc, 0 },
1092 	/* 176 */ { SIOCSLIFUSESRC, sizeof (struct lifreq),
1093 			IPI_PRIV | IPI_WR,
1094 			LIF_CMD, ip_sioctl_slifusesrc,
1095 			NULL },
1096 	/* 177 */ { SIOCGLIFSRCOF, 0, IPI_GET_CMD, MISC_CMD,
1097 			ip_sioctl_get_lifsrcof, NULL },
1098 	/* 178 */ { SIOCGMSFILTER, sizeof (struct group_filter), IPI_GET_CMD,
1099 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1100 	/* 179 */ { SIOCSMSFILTER, sizeof (struct group_filter), 0,
1101 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1102 	/* 180 */ { SIOCGIPMSFILTER, sizeof (struct ip_msfilter), IPI_GET_CMD,
1103 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1104 	/* 181 */ { SIOCSIPMSFILTER, sizeof (struct ip_msfilter), 0,
1105 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1106 	/* 182 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1107 	/* SIOCSENABLESDP is handled by SDP */
1108 	/* 183 */ { IPI_DONTCARE /* SIOCSENABLESDP */, 0, 0, 0, NULL, NULL },
1109 	/* 184 */ { IPI_DONTCARE /* SIOCSQPTR */, 0, 0, 0, NULL, NULL },
1110 	/* 185 */ { SIOCGIFHWADDR, sizeof (struct ifreq), IPI_GET_CMD,
1111 			IF_CMD, ip_sioctl_get_ifhwaddr, NULL },
1112 	/* 186 */ { IPI_DONTCARE /* SIOCGSTAMP */, 0, 0, 0, NULL, NULL },
1113 	/* 187 */ { SIOCILB, 0, IPI_PRIV | IPI_GET_CMD, MISC_CMD,
1114 			ip_sioctl_ilb_cmd, NULL },
1115 	/* 188 */ { SIOCGETPROP, 0, IPI_GET_CMD, 0, NULL, NULL },
1116 	/* 189 */ { SIOCSETPROP, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL},
1117 	/* 190 */ { SIOCGLIFDADSTATE, sizeof (struct lifreq),
1118 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dadstate, NULL },
1119 	/* 191 */ { SIOCSLIFPREFIX, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1120 			LIF_CMD, ip_sioctl_prefix, ip_sioctl_prefix_restart },
1121 	/* 192 */ { SIOCGLIFHWADDR, sizeof (struct lifreq), IPI_GET_CMD,
1122 			LIF_CMD, ip_sioctl_get_lifhwaddr, NULL }
1123 };
1124 
1125 int ip_ndx_ioctl_count = sizeof (ip_ndx_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1126 
1127 ip_ioctl_cmd_t ip_misc_ioctl_table[] = {
1128 	{ I_LINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1129 	{ I_UNLINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1130 	{ I_PLINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1131 	{ I_PUNLINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1132 	{ ND_GET,	0, 0, 0, NULL, NULL },
1133 	{ ND_SET,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1134 	{ IP_IOCTL,	0, 0, 0, NULL, NULL },
1135 	{ SIOCGETVIFCNT, sizeof (struct sioc_vif_req), IPI_GET_CMD,
1136 		MISC_CMD, mrt_ioctl},
1137 	{ SIOCGETSGCNT,	sizeof (struct sioc_sg_req), IPI_GET_CMD,
1138 		MISC_CMD, mrt_ioctl},
1139 	{ SIOCGETLSGCNT, sizeof (struct sioc_lsg_req), IPI_GET_CMD,
1140 		MISC_CMD, mrt_ioctl}
1141 };
1142 
1143 int ip_misc_ioctl_count =
1144     sizeof (ip_misc_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1145 
1146 int	conn_drain_nthreads;		/* Number of drainers reqd. */
1147 					/* Settable in /etc/system */
1148 /* Defined in ip_ire.c */
1149 extern uint32_t ip_ire_max_bucket_cnt, ip6_ire_max_bucket_cnt;
1150 extern uint32_t ip_ire_min_bucket_cnt, ip6_ire_min_bucket_cnt;
1151 extern uint32_t ip_ire_mem_ratio, ip_ire_cpu_ratio;
1152 
1153 static nv_t	ire_nv_arr[] = {
1154 	{ IRE_BROADCAST, "BROADCAST" },
1155 	{ IRE_LOCAL, "LOCAL" },
1156 	{ IRE_LOOPBACK, "LOOPBACK" },
1157 	{ IRE_DEFAULT, "DEFAULT" },
1158 	{ IRE_PREFIX, "PREFIX" },
1159 	{ IRE_IF_NORESOLVER, "IF_NORESOL" },
1160 	{ IRE_IF_RESOLVER, "IF_RESOLV" },
1161 	{ IRE_IF_CLONE, "IF_CLONE" },
1162 	{ IRE_HOST, "HOST" },
1163 	{ IRE_MULTICAST, "MULTICAST" },
1164 	{ IRE_NOROUTE, "NOROUTE" },
1165 	{ 0 }
1166 };
1167 
1168 nv_t	*ire_nv_tbl = ire_nv_arr;
1169 
1170 /* Simple ICMP IP Header Template */
1171 static ipha_t icmp_ipha = {
1172 	IP_SIMPLE_HDR_VERSION, 0, 0, 0, 0, 0, IPPROTO_ICMP
1173 };
1174 
1175 struct module_info ip_mod_info = {
1176 	IP_MOD_ID, IP_MOD_NAME, IP_MOD_MINPSZ, IP_MOD_MAXPSZ, IP_MOD_HIWAT,
1177 	IP_MOD_LOWAT
1178 };
1179 
1180 /*
1181  * Duplicate static symbols within a module confuses mdb; so we avoid the
1182  * problem by making the symbols here distinct from those in udp.c.
1183  */
1184 
1185 /*
1186  * Entry points for IP as a device and as a module.
1187  * We have separate open functions for the /dev/ip and /dev/ip6 devices.
1188  */
1189 static struct qinit iprinitv4 = {
1190 	ip_rput, NULL, ip_openv4, ip_close, NULL, &ip_mod_info
1191 };
1192 
1193 struct qinit iprinitv6 = {
1194 	ip_rput_v6, NULL, ip_openv6, ip_close, NULL, &ip_mod_info
1195 };
1196 
1197 static struct qinit ipwinit = {
1198 	ip_wput_nondata, ip_wsrv, NULL, NULL, NULL, &ip_mod_info
1199 };
1200 
1201 static struct qinit iplrinit = {
1202 	ip_lrput, NULL, ip_openv4, ip_close, NULL, &ip_mod_info
1203 };
1204 
1205 static struct qinit iplwinit = {
1206 	ip_lwput, NULL, NULL, NULL, NULL, &ip_mod_info
1207 };
1208 
1209 /* For AF_INET aka /dev/ip */
1210 struct streamtab ipinfov4 = {
1211 	&iprinitv4, &ipwinit, &iplrinit, &iplwinit
1212 };
1213 
1214 /* For AF_INET6 aka /dev/ip6 */
1215 struct streamtab ipinfov6 = {
1216 	&iprinitv6, &ipwinit, &iplrinit, &iplwinit
1217 };
1218 
1219 #ifdef	DEBUG
1220 boolean_t skip_sctp_cksum = B_FALSE;
1221 #endif
1222 
1223 /*
1224  * Generate an ICMP fragmentation needed message.
1225  * When called from ip_output side a minimal ip_recv_attr_t needs to be
1226  * constructed by the caller.
1227  */
1228 void
icmp_frag_needed(mblk_t * mp,int mtu,ip_recv_attr_t * ira)1229 icmp_frag_needed(mblk_t *mp, int mtu, ip_recv_attr_t *ira)
1230 {
1231 	icmph_t	icmph;
1232 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
1233 
1234 	mp = icmp_pkt_err_ok(mp, ira);
1235 	if (mp == NULL)
1236 		return;
1237 
1238 	bzero(&icmph, sizeof (icmph_t));
1239 	icmph.icmph_type = ICMP_DEST_UNREACHABLE;
1240 	icmph.icmph_code = ICMP_FRAGMENTATION_NEEDED;
1241 	icmph.icmph_du_mtu = htons((uint16_t)mtu);
1242 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutFragNeeded);
1243 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
1244 
1245 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
1246 }
1247 
1248 /*
1249  * icmp_inbound_v4 deals with ICMP messages that are handled by IP.
1250  * If the ICMP message is consumed by IP, i.e., it should not be delivered
1251  * to any IPPROTO_ICMP raw sockets, then it returns NULL.
1252  * Likewise, if the ICMP error is misformed (too short, etc), then it
1253  * returns NULL. The caller uses this to determine whether or not to send
1254  * to raw sockets.
1255  *
1256  * All error messages are passed to the matching transport stream.
1257  *
1258  * The following cases are handled by icmp_inbound:
1259  * 1) It needs to send a reply back and possibly delivering it
1260  *    to the "interested" upper clients.
1261  * 2) Return the mblk so that the caller can pass it to the RAW socket clients.
1262  * 3) It needs to change some values in IP only.
1263  * 4) It needs to change some values in IP and upper layers e.g TCP
1264  *    by delivering an error to the upper layers.
1265  *
1266  * We handle the above three cases in the context of IPsec in the
1267  * following way :
1268  *
1269  * 1) Send the reply back in the same way as the request came in.
1270  *    If it came in encrypted, it goes out encrypted. If it came in
1271  *    clear, it goes out in clear. Thus, this will prevent chosen
1272  *    plain text attack.
1273  * 2) The client may or may not expect things to come in secure.
1274  *    If it comes in secure, the policy constraints are checked
1275  *    before delivering it to the upper layers. If it comes in
1276  *    clear, ipsec_inbound_accept_clear will decide whether to
1277  *    accept this in clear or not. In both the cases, if the returned
1278  *    message (IP header + 8 bytes) that caused the icmp message has
1279  *    AH/ESP headers, it is sent up to AH/ESP for validation before
1280  *    sending up. If there are only 8 bytes of returned message, then
1281  *    upper client will not be notified.
1282  * 3) Check with global policy to see whether it matches the constaints.
1283  *    But this will be done only if icmp_accept_messages_in_clear is
1284  *    zero.
1285  * 4) If we need to change both in IP and ULP, then the decision taken
1286  *    while affecting the values in IP and while delivering up to TCP
1287  *    should be the same.
1288  *
1289  *	There are two cases.
1290  *
1291  *	a) If we reject data at the IP layer (ipsec_check_global_policy()
1292  *	   failed), we will not deliver it to the ULP, even though they
1293  *	   are *willing* to accept in *clear*. This is fine as our global
1294  *	   disposition to icmp messages asks us reject the datagram.
1295  *
1296  *	b) If we accept data at the IP layer (ipsec_check_global_policy()
1297  *	   succeeded or icmp_accept_messages_in_clear is 1), and not able
1298  *	   to deliver it to ULP (policy failed), it can lead to
1299  *	   consistency problems. The cases known at this time are
1300  *	   ICMP_DESTINATION_UNREACHABLE  messages with following code
1301  *	   values :
1302  *
1303  *	   - ICMP_FRAGMENTATION_NEEDED : IP adapts to the new value
1304  *	     and Upper layer rejects. Then the communication will
1305  *	     come to a stop. This is solved by making similar decisions
1306  *	     at both levels. Currently, when we are unable to deliver
1307  *	     to the Upper Layer (due to policy failures) while IP has
1308  *	     adjusted dce_pmtu, the next outbound datagram would
1309  *	     generate a local ICMP_FRAGMENTATION_NEEDED message - which
1310  *	     will be with the right level of protection. Thus the right
1311  *	     value will be communicated even if we are not able to
1312  *	     communicate when we get from the wire initially. But this
1313  *	     assumes there would be at least one outbound datagram after
1314  *	     IP has adjusted its dce_pmtu value. To make things
1315  *	     simpler, we accept in clear after the validation of
1316  *	     AH/ESP headers.
1317  *
1318  *	   - Other ICMP ERRORS : We may not be able to deliver it to the
1319  *	     upper layer depending on the level of protection the upper
1320  *	     layer expects and the disposition in ipsec_inbound_accept_clear().
1321  *	     ipsec_inbound_accept_clear() decides whether a given ICMP error
1322  *	     should be accepted in clear when the Upper layer expects secure.
1323  *	     Thus the communication may get aborted by some bad ICMP
1324  *	     packets.
1325  */
1326 mblk_t *
icmp_inbound_v4(mblk_t * mp,ip_recv_attr_t * ira)1327 icmp_inbound_v4(mblk_t *mp, ip_recv_attr_t *ira)
1328 {
1329 	icmph_t		*icmph;
1330 	ipha_t		*ipha;		/* Outer header */
1331 	int		ip_hdr_length;	/* Outer header length */
1332 	boolean_t	interested;
1333 	ipif_t		*ipif;
1334 	uint32_t	ts;
1335 	uint32_t	*tsp;
1336 	timestruc_t	now;
1337 	ill_t		*ill = ira->ira_ill;
1338 	ip_stack_t	*ipst = ill->ill_ipst;
1339 	zoneid_t	zoneid = ira->ira_zoneid;
1340 	int		len_needed;
1341 	mblk_t		*mp_ret = NULL;
1342 
1343 	ipha = (ipha_t *)mp->b_rptr;
1344 
1345 	BUMP_MIB(&ipst->ips_icmp_mib, icmpInMsgs);
1346 
1347 	ip_hdr_length = ira->ira_ip_hdr_length;
1348 	if ((mp->b_wptr - mp->b_rptr) < (ip_hdr_length + ICMPH_SIZE)) {
1349 		if (ira->ira_pktlen < (ip_hdr_length + ICMPH_SIZE)) {
1350 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1351 			ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1352 			freemsg(mp);
1353 			return (NULL);
1354 		}
1355 		/* Last chance to get real. */
1356 		ipha = ip_pullup(mp, ip_hdr_length + ICMPH_SIZE, ira);
1357 		if (ipha == NULL) {
1358 			BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1359 			freemsg(mp);
1360 			return (NULL);
1361 		}
1362 	}
1363 
1364 	/* The IP header will always be a multiple of four bytes */
1365 	icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1366 	ip2dbg(("icmp_inbound_v4: type %d code %d\n", icmph->icmph_type,
1367 	    icmph->icmph_code));
1368 
1369 	/*
1370 	 * We will set "interested" to "true" if we should pass a copy to
1371 	 * the transport or if we handle the packet locally.
1372 	 */
1373 	interested = B_FALSE;
1374 	switch (icmph->icmph_type) {
1375 	case ICMP_ECHO_REPLY:
1376 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchoReps);
1377 		break;
1378 	case ICMP_DEST_UNREACHABLE:
1379 		if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED)
1380 			BUMP_MIB(&ipst->ips_icmp_mib, icmpInFragNeeded);
1381 		interested = B_TRUE;	/* Pass up to transport */
1382 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInDestUnreachs);
1383 		break;
1384 	case ICMP_SOURCE_QUENCH:
1385 		interested = B_TRUE;	/* Pass up to transport */
1386 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInSrcQuenchs);
1387 		break;
1388 	case ICMP_REDIRECT:
1389 		if (!ipst->ips_ip_ignore_redirect)
1390 			interested = B_TRUE;
1391 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInRedirects);
1392 		break;
1393 	case ICMP_ECHO_REQUEST:
1394 		/*
1395 		 * Whether to respond to echo requests that come in as IP
1396 		 * broadcasts or as IP multicast is subject to debate
1397 		 * (what isn't?).  We aim to please, you pick it.
1398 		 * Default is do it.
1399 		 */
1400 		if (ira->ira_flags & IRAF_MULTICAST) {
1401 			/* multicast: respond based on tunable */
1402 			interested = ipst->ips_ip_g_resp_to_echo_mcast;
1403 		} else if (ira->ira_flags & IRAF_BROADCAST) {
1404 			/* broadcast: respond based on tunable */
1405 			interested = ipst->ips_ip_g_resp_to_echo_bcast;
1406 		} else {
1407 			/* unicast: always respond */
1408 			interested = B_TRUE;
1409 		}
1410 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchos);
1411 		if (!interested) {
1412 			/* We never pass these to RAW sockets */
1413 			freemsg(mp);
1414 			return (NULL);
1415 		}
1416 
1417 		/* Check db_ref to make sure we can modify the packet. */
1418 		if (mp->b_datap->db_ref > 1) {
1419 			mblk_t	*mp1;
1420 
1421 			mp1 = copymsg(mp);
1422 			freemsg(mp);
1423 			if (!mp1) {
1424 				BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1425 				return (NULL);
1426 			}
1427 			mp = mp1;
1428 			ipha = (ipha_t *)mp->b_rptr;
1429 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1430 		}
1431 		icmph->icmph_type = ICMP_ECHO_REPLY;
1432 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutEchoReps);
1433 		icmp_send_reply_v4(mp, ipha, icmph, ira);
1434 		return (NULL);
1435 
1436 	case ICMP_ROUTER_ADVERTISEMENT:
1437 	case ICMP_ROUTER_SOLICITATION:
1438 		break;
1439 	case ICMP_TIME_EXCEEDED:
1440 		interested = B_TRUE;	/* Pass up to transport */
1441 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimeExcds);
1442 		break;
1443 	case ICMP_PARAM_PROBLEM:
1444 		interested = B_TRUE;	/* Pass up to transport */
1445 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInParmProbs);
1446 		break;
1447 	case ICMP_TIME_STAMP_REQUEST:
1448 		/* Response to Time Stamp Requests is local policy. */
1449 		if (ipst->ips_ip_g_resp_to_timestamp) {
1450 			if (ira->ira_flags & IRAF_MULTIBROADCAST)
1451 				interested =
1452 				    ipst->ips_ip_g_resp_to_timestamp_bcast;
1453 			else
1454 				interested = B_TRUE;
1455 		}
1456 		if (!interested) {
1457 			/* We never pass these to RAW sockets */
1458 			freemsg(mp);
1459 			return (NULL);
1460 		}
1461 
1462 		/* Make sure we have enough of the packet */
1463 		len_needed = ip_hdr_length + ICMPH_SIZE +
1464 		    3 * sizeof (uint32_t);
1465 
1466 		if (mp->b_wptr - mp->b_rptr < len_needed) {
1467 			ipha = ip_pullup(mp, len_needed, ira);
1468 			if (ipha == NULL) {
1469 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1470 				ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1471 				    mp, ill);
1472 				freemsg(mp);
1473 				return (NULL);
1474 			}
1475 			/* Refresh following the pullup. */
1476 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1477 		}
1478 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestamps);
1479 		/* Check db_ref to make sure we can modify the packet. */
1480 		if (mp->b_datap->db_ref > 1) {
1481 			mblk_t	*mp1;
1482 
1483 			mp1 = copymsg(mp);
1484 			freemsg(mp);
1485 			if (!mp1) {
1486 				BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1487 				return (NULL);
1488 			}
1489 			mp = mp1;
1490 			ipha = (ipha_t *)mp->b_rptr;
1491 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1492 		}
1493 		icmph->icmph_type = ICMP_TIME_STAMP_REPLY;
1494 		tsp = (uint32_t *)&icmph[1];
1495 		tsp++;		/* Skip past 'originate time' */
1496 		/* Compute # of milliseconds since midnight */
1497 		gethrestime(&now);
1498 		ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
1499 		    NSEC2MSEC(now.tv_nsec);
1500 		*tsp++ = htonl(ts);	/* Lay in 'receive time' */
1501 		*tsp++ = htonl(ts);	/* Lay in 'send time' */
1502 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimestampReps);
1503 		icmp_send_reply_v4(mp, ipha, icmph, ira);
1504 		return (NULL);
1505 
1506 	case ICMP_TIME_STAMP_REPLY:
1507 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestampReps);
1508 		break;
1509 	case ICMP_INFO_REQUEST:
1510 		/* Per RFC 1122 3.2.2.7, ignore this. */
1511 	case ICMP_INFO_REPLY:
1512 		break;
1513 	case ICMP_ADDRESS_MASK_REQUEST:
1514 		if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1515 			interested =
1516 			    ipst->ips_ip_respond_to_address_mask_broadcast;
1517 		} else {
1518 			interested = B_TRUE;
1519 		}
1520 		if (!interested) {
1521 			/* We never pass these to RAW sockets */
1522 			freemsg(mp);
1523 			return (NULL);
1524 		}
1525 		len_needed = ip_hdr_length + ICMPH_SIZE + IP_ADDR_LEN;
1526 		if (mp->b_wptr - mp->b_rptr < len_needed) {
1527 			ipha = ip_pullup(mp, len_needed, ira);
1528 			if (ipha == NULL) {
1529 				BUMP_MIB(ill->ill_ip_mib,
1530 				    ipIfStatsInTruncatedPkts);
1531 				ip_drop_input("ipIfStatsInTruncatedPkts", mp,
1532 				    ill);
1533 				freemsg(mp);
1534 				return (NULL);
1535 			}
1536 			/* Refresh following the pullup. */
1537 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1538 		}
1539 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMasks);
1540 		/* Check db_ref to make sure we can modify the packet. */
1541 		if (mp->b_datap->db_ref > 1) {
1542 			mblk_t	*mp1;
1543 
1544 			mp1 = copymsg(mp);
1545 			freemsg(mp);
1546 			if (!mp1) {
1547 				BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1548 				return (NULL);
1549 			}
1550 			mp = mp1;
1551 			ipha = (ipha_t *)mp->b_rptr;
1552 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1553 		}
1554 		/*
1555 		 * Need the ipif with the mask be the same as the source
1556 		 * address of the mask reply. For unicast we have a specific
1557 		 * ipif. For multicast/broadcast we only handle onlink
1558 		 * senders, and use the source address to pick an ipif.
1559 		 */
1560 		ipif = ipif_lookup_addr(ipha->ipha_dst, ill, zoneid, ipst);
1561 		if (ipif == NULL) {
1562 			/* Broadcast or multicast */
1563 			ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1564 			if (ipif == NULL) {
1565 				freemsg(mp);
1566 				return (NULL);
1567 			}
1568 		}
1569 		icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
1570 		bcopy(&ipif->ipif_net_mask, &icmph[1], IP_ADDR_LEN);
1571 		ipif_refrele(ipif);
1572 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutAddrMaskReps);
1573 		icmp_send_reply_v4(mp, ipha, icmph, ira);
1574 		return (NULL);
1575 
1576 	case ICMP_ADDRESS_MASK_REPLY:
1577 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMaskReps);
1578 		break;
1579 	default:
1580 		interested = B_TRUE;	/* Pass up to transport */
1581 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInUnknowns);
1582 		break;
1583 	}
1584 	/*
1585 	 * See if there is an ICMP client to avoid an extra copymsg/freemsg
1586 	 * if there isn't one.
1587 	 */
1588 	if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_ICMP].connf_head != NULL) {
1589 		/* If there is an ICMP client and we want one too, copy it. */
1590 
1591 		if (!interested) {
1592 			/* Caller will deliver to RAW sockets */
1593 			return (mp);
1594 		}
1595 		mp_ret = copymsg(mp);
1596 		if (mp_ret == NULL) {
1597 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1598 			ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1599 		}
1600 	} else if (!interested) {
1601 		/* Neither we nor raw sockets are interested. Drop packet now */
1602 		freemsg(mp);
1603 		return (NULL);
1604 	}
1605 
1606 	/*
1607 	 * ICMP error or redirect packet. Make sure we have enough of
1608 	 * the header and that db_ref == 1 since we might end up modifying
1609 	 * the packet.
1610 	 */
1611 	if (mp->b_cont != NULL) {
1612 		if (ip_pullup(mp, -1, ira) == NULL) {
1613 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1614 			ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1615 			    mp, ill);
1616 			freemsg(mp);
1617 			return (mp_ret);
1618 		}
1619 	}
1620 
1621 	if (mp->b_datap->db_ref > 1) {
1622 		mblk_t	*mp1;
1623 
1624 		mp1 = copymsg(mp);
1625 		if (mp1 == NULL) {
1626 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1627 			ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1628 			freemsg(mp);
1629 			return (mp_ret);
1630 		}
1631 		freemsg(mp);
1632 		mp = mp1;
1633 	}
1634 
1635 	/*
1636 	 * In case mp has changed, verify the message before any further
1637 	 * processes.
1638 	 */
1639 	ipha = (ipha_t *)mp->b_rptr;
1640 	icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1641 	if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
1642 		freemsg(mp);
1643 		return (mp_ret);
1644 	}
1645 
1646 	switch (icmph->icmph_type) {
1647 	case ICMP_REDIRECT:
1648 		icmp_redirect_v4(mp, ipha, icmph, ira);
1649 		break;
1650 	case ICMP_DEST_UNREACHABLE:
1651 		if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED) {
1652 			/* Update DCE and adjust MTU is icmp header if needed */
1653 			icmp_inbound_too_big_v4(icmph, ira);
1654 		}
1655 		/* FALLTHROUGH */
1656 	default:
1657 		icmp_inbound_error_fanout_v4(mp, icmph, ira);
1658 		break;
1659 	}
1660 	return (mp_ret);
1661 }
1662 
1663 /*
1664  * Send an ICMP echo, timestamp or address mask reply.
1665  * The caller has already updated the payload part of the packet.
1666  * We handle the ICMP checksum, IP source address selection and feed
1667  * the packet into ip_output_simple.
1668  */
1669 static void
icmp_send_reply_v4(mblk_t * mp,ipha_t * ipha,icmph_t * icmph,ip_recv_attr_t * ira)1670 icmp_send_reply_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph,
1671     ip_recv_attr_t *ira)
1672 {
1673 	uint_t		ip_hdr_length = ira->ira_ip_hdr_length;
1674 	ill_t		*ill = ira->ira_ill;
1675 	ip_stack_t	*ipst = ill->ill_ipst;
1676 	ip_xmit_attr_t	ixas;
1677 
1678 	/* Send out an ICMP packet */
1679 	icmph->icmph_checksum = 0;
1680 	icmph->icmph_checksum = IP_CSUM(mp, ip_hdr_length, 0);
1681 	/* Reset time to live. */
1682 	ipha->ipha_ttl = ipst->ips_ip_def_ttl;
1683 	{
1684 		/* Swap source and destination addresses */
1685 		ipaddr_t tmp;
1686 
1687 		tmp = ipha->ipha_src;
1688 		ipha->ipha_src = ipha->ipha_dst;
1689 		ipha->ipha_dst = tmp;
1690 	}
1691 	ipha->ipha_ident = 0;
1692 	if (!IS_SIMPLE_IPH(ipha))
1693 		icmp_options_update(ipha);
1694 
1695 	bzero(&ixas, sizeof (ixas));
1696 	ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
1697 	ixas.ixa_zoneid = ira->ira_zoneid;
1698 	ixas.ixa_cred = kcred;
1699 	ixas.ixa_cpid = NOPID;
1700 	ixas.ixa_tsl = ira->ira_tsl;	/* Behave as a multi-level responder */
1701 	ixas.ixa_ifindex = 0;
1702 	ixas.ixa_ipst = ipst;
1703 	ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1704 
1705 	if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
1706 		/*
1707 		 * This packet should go out the same way as it
1708 		 * came in i.e in clear, independent of the IPsec policy
1709 		 * for transmitting packets.
1710 		 */
1711 		ixas.ixa_flags |= IXAF_NO_IPSEC;
1712 	} else {
1713 		if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
1714 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1715 			/* Note: mp already consumed and ip_drop_packet done */
1716 			return;
1717 		}
1718 	}
1719 	if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1720 		/*
1721 		 * Not one or our addresses (IRE_LOCALs), thus we let
1722 		 * ip_output_simple pick the source.
1723 		 */
1724 		ipha->ipha_src = INADDR_ANY;
1725 		ixas.ixa_flags |= IXAF_SET_SOURCE;
1726 	}
1727 	/* Should we send with DF and use dce_pmtu? */
1728 	if (ipst->ips_ipv4_icmp_return_pmtu) {
1729 		ixas.ixa_flags |= IXAF_PMTU_DISCOVERY;
1730 		ipha->ipha_fragment_offset_and_flags |= IPH_DF_HTONS;
1731 	}
1732 
1733 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
1734 
1735 	(void) ip_output_simple(mp, &ixas);
1736 	ixa_cleanup(&ixas);
1737 }
1738 
1739 /*
1740  * Verify the ICMP messages for either for ICMP error or redirect packet.
1741  * The caller should have fully pulled up the message. If it's a redirect
1742  * packet, only basic checks on IP header will be done; otherwise, verify
1743  * the packet by looking at the included ULP header.
1744  *
1745  * Called before icmp_inbound_error_fanout_v4 is called.
1746  */
1747 static boolean_t
icmp_inbound_verify_v4(mblk_t * mp,icmph_t * icmph,ip_recv_attr_t * ira)1748 icmp_inbound_verify_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
1749 {
1750 	ill_t		*ill = ira->ira_ill;
1751 	int		hdr_length;
1752 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
1753 	conn_t		*connp;
1754 	ipha_t		*ipha;	/* Inner IP header */
1755 
1756 	ipha = (ipha_t *)&icmph[1];
1757 	if ((uchar_t *)ipha + IP_SIMPLE_HDR_LENGTH > mp->b_wptr)
1758 		goto truncated;
1759 
1760 	hdr_length = IPH_HDR_LENGTH(ipha);
1761 
1762 	if ((IPH_HDR_VERSION(ipha) != IPV4_VERSION))
1763 		goto discard_pkt;
1764 
1765 	if (hdr_length < sizeof (ipha_t))
1766 		goto truncated;
1767 
1768 	if ((uchar_t *)ipha + hdr_length > mp->b_wptr)
1769 		goto truncated;
1770 
1771 	/*
1772 	 * Stop here for ICMP_REDIRECT.
1773 	 */
1774 	if (icmph->icmph_type == ICMP_REDIRECT)
1775 		return (B_TRUE);
1776 
1777 	/*
1778 	 * ICMP errors only.
1779 	 */
1780 	switch (ipha->ipha_protocol) {
1781 	case IPPROTO_UDP:
1782 		/*
1783 		 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1784 		 * transport header.
1785 		 */
1786 		if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1787 		    mp->b_wptr)
1788 			goto truncated;
1789 		break;
1790 	case IPPROTO_TCP: {
1791 		tcpha_t		*tcpha;
1792 
1793 		/*
1794 		 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1795 		 * transport header.
1796 		 */
1797 		if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1798 		    mp->b_wptr)
1799 			goto truncated;
1800 
1801 		tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
1802 		connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
1803 		    ipst);
1804 		if (connp == NULL)
1805 			goto discard_pkt;
1806 
1807 		if ((connp->conn_verifyicmp != NULL) &&
1808 		    !connp->conn_verifyicmp(connp, tcpha, icmph, NULL, ira)) {
1809 			CONN_DEC_REF(connp);
1810 			goto discard_pkt;
1811 		}
1812 		CONN_DEC_REF(connp);
1813 		break;
1814 	}
1815 	case IPPROTO_SCTP:
1816 		/*
1817 		 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1818 		 * transport header.
1819 		 */
1820 		if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1821 		    mp->b_wptr)
1822 			goto truncated;
1823 		break;
1824 	case IPPROTO_ESP:
1825 	case IPPROTO_AH:
1826 		break;
1827 	case IPPROTO_ENCAP:
1828 		if ((uchar_t *)ipha + hdr_length + sizeof (ipha_t) >
1829 		    mp->b_wptr)
1830 			goto truncated;
1831 		break;
1832 	default:
1833 		break;
1834 	}
1835 
1836 	return (B_TRUE);
1837 
1838 discard_pkt:
1839 	/* Bogus ICMP error. */
1840 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1841 	return (B_FALSE);
1842 
1843 truncated:
1844 	/* We pulled up everthing already. Must be truncated */
1845 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1846 	ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1847 	return (B_FALSE);
1848 }
1849 
1850 /* Table from RFC 1191 */
1851 static int icmp_frag_size_table[] =
1852 { 32000, 17914, 8166, 4352, 2002, 1496, 1006, 508, 296, 68 };
1853 
1854 /*
1855  * Process received ICMP Packet too big.
1856  * Just handles the DCE create/update, including using the above table of
1857  * PMTU guesses. The caller is responsible for validating the packet before
1858  * passing it in and also to fanout the ICMP error to any matching transport
1859  * conns. Assumes the message has been fully pulled up and verified.
1860  *
1861  * Before getting here, the caller has called icmp_inbound_verify_v4()
1862  * that should have verified with ULP to prevent undoing the changes we're
1863  * going to make to DCE. For example, TCP might have verified that the packet
1864  * which generated error is in the send window.
1865  *
1866  * In some cases modified this MTU in the ICMP header packet; the caller
1867  * should pass to the matching ULP after this returns.
1868  */
1869 static void
icmp_inbound_too_big_v4(icmph_t * icmph,ip_recv_attr_t * ira)1870 icmp_inbound_too_big_v4(icmph_t *icmph, ip_recv_attr_t *ira)
1871 {
1872 	dce_t		*dce;
1873 	int		old_mtu;
1874 	int		mtu, orig_mtu;
1875 	ipaddr_t	dst;
1876 	boolean_t	disable_pmtud;
1877 	ill_t		*ill = ira->ira_ill;
1878 	ip_stack_t	*ipst = ill->ill_ipst;
1879 	uint_t		hdr_length;
1880 	ipha_t		*ipha;
1881 
1882 	/* Caller already pulled up everything. */
1883 	ipha = (ipha_t *)&icmph[1];
1884 	ASSERT(icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
1885 	    icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED);
1886 	ASSERT(ill != NULL);
1887 
1888 	hdr_length = IPH_HDR_LENGTH(ipha);
1889 
1890 	/*
1891 	 * We handle path MTU for source routed packets since the DCE
1892 	 * is looked up using the final destination.
1893 	 */
1894 	dst = ip_get_dst(ipha);
1895 
1896 	dce = dce_lookup_and_add_v4(dst, ipst);
1897 	if (dce == NULL) {
1898 		/* Couldn't add a unique one - ENOMEM */
1899 		ip1dbg(("icmp_inbound_too_big_v4: no dce for 0x%x\n",
1900 		    ntohl(dst)));
1901 		return;
1902 	}
1903 
1904 	/* Check for MTU discovery advice as described in RFC 1191 */
1905 	mtu = ntohs(icmph->icmph_du_mtu);
1906 	orig_mtu = mtu;
1907 	disable_pmtud = B_FALSE;
1908 
1909 	mutex_enter(&dce->dce_lock);
1910 	if (dce->dce_flags & DCEF_PMTU)
1911 		old_mtu = dce->dce_pmtu;
1912 	else
1913 		old_mtu = ill->ill_mtu;
1914 
1915 	if (icmph->icmph_du_zero != 0 || mtu < ipst->ips_ip_pmtu_min) {
1916 		uint32_t length;
1917 		int	i;
1918 
1919 		/*
1920 		 * Use the table from RFC 1191 to figure out
1921 		 * the next "plateau" based on the length in
1922 		 * the original IP packet.
1923 		 */
1924 		length = ntohs(ipha->ipha_length);
1925 		DTRACE_PROBE2(ip4__pmtu__guess, dce_t *, dce,
1926 		    uint32_t, length);
1927 		if (old_mtu <= length &&
1928 		    old_mtu >= length - hdr_length) {
1929 			/*
1930 			 * Handle broken BSD 4.2 systems that
1931 			 * return the wrong ipha_length in ICMP
1932 			 * errors.
1933 			 */
1934 			ip1dbg(("Wrong mtu: sent %d, dce %d\n",
1935 			    length, old_mtu));
1936 			length -= hdr_length;
1937 		}
1938 		for (i = 0; i < A_CNT(icmp_frag_size_table); i++) {
1939 			if (length > icmp_frag_size_table[i])
1940 				break;
1941 		}
1942 		if (i == A_CNT(icmp_frag_size_table)) {
1943 			/* Smaller than IP_MIN_MTU! */
1944 			ip1dbg(("Too big for packet size %d\n",
1945 			    length));
1946 			disable_pmtud = B_TRUE;
1947 			mtu = ipst->ips_ip_pmtu_min;
1948 		} else {
1949 			mtu = icmp_frag_size_table[i];
1950 			ip1dbg(("Calculated mtu %d, packet size %d, "
1951 			    "before %d\n", mtu, length, old_mtu));
1952 			if (mtu < ipst->ips_ip_pmtu_min) {
1953 				mtu = ipst->ips_ip_pmtu_min;
1954 				disable_pmtud = B_TRUE;
1955 			}
1956 		}
1957 	}
1958 	if (disable_pmtud)
1959 		dce->dce_flags |= DCEF_TOO_SMALL_PMTU;
1960 	else
1961 		dce->dce_flags &= ~DCEF_TOO_SMALL_PMTU;
1962 
1963 	dce->dce_pmtu = MIN(old_mtu, mtu);
1964 	/* Prepare to send the new max frag size for the ULP. */
1965 	icmph->icmph_du_zero = 0;
1966 	icmph->icmph_du_mtu =  htons((uint16_t)dce->dce_pmtu);
1967 	DTRACE_PROBE4(ip4__pmtu__change, icmph_t *, icmph, dce_t *,
1968 	    dce, int, orig_mtu, int, mtu);
1969 
1970 	/* We now have a PMTU for sure */
1971 	dce->dce_flags |= DCEF_PMTU;
1972 	dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
1973 	mutex_exit(&dce->dce_lock);
1974 	/*
1975 	 * After dropping the lock the new value is visible to everyone.
1976 	 * Then we bump the generation number so any cached values reinspect
1977 	 * the dce_t.
1978 	 */
1979 	dce_increment_generation(dce);
1980 	dce_refrele(dce);
1981 }
1982 
1983 /*
1984  * If the packet in error is Self-Encapsulated, icmp_inbound_error_fanout_v4
1985  * calls this function.
1986  */
1987 static mblk_t *
icmp_inbound_self_encap_error_v4(mblk_t * mp,ipha_t * ipha,ipha_t * in_ipha)1988 icmp_inbound_self_encap_error_v4(mblk_t *mp, ipha_t *ipha, ipha_t *in_ipha)
1989 {
1990 	int length;
1991 
1992 	ASSERT(mp->b_datap->db_type == M_DATA);
1993 
1994 	/* icmp_inbound_v4 has already pulled up the whole error packet */
1995 	ASSERT(mp->b_cont == NULL);
1996 
1997 	/*
1998 	 * The length that we want to overlay is the inner header
1999 	 * and what follows it.
2000 	 */
2001 	length = msgdsize(mp) - ((uchar_t *)in_ipha - mp->b_rptr);
2002 
2003 	/*
2004 	 * Overlay the inner header and whatever follows it over the
2005 	 * outer header.
2006 	 */
2007 	bcopy((uchar_t *)in_ipha, (uchar_t *)ipha, length);
2008 
2009 	/* Adjust for what we removed */
2010 	mp->b_wptr -= (uchar_t *)in_ipha - (uchar_t *)ipha;
2011 	return (mp);
2012 }
2013 
2014 /*
2015  * Try to pass the ICMP message upstream in case the ULP cares.
2016  *
2017  * If the packet that caused the ICMP error is secure, we send
2018  * it to AH/ESP to make sure that the attached packet has a
2019  * valid association. ipha in the code below points to the
2020  * IP header of the packet that caused the error.
2021  *
2022  * For IPsec cases, we let the next-layer-up (which has access to
2023  * cached policy on the conn_t, or can query the SPD directly)
2024  * subtract out any IPsec overhead if they must.  We therefore make no
2025  * adjustments here for IPsec overhead.
2026  *
2027  * IFN could have been generated locally or by some router.
2028  *
2029  * LOCAL : ire_send_wire (before calling ipsec_out_process) can call
2030  * icmp_frag_needed/icmp_pkt2big_v6 to generated a local IFN.
2031  *	    This happens because IP adjusted its value of MTU on an
2032  *	    earlier IFN message and could not tell the upper layer,
2033  *	    the new adjusted value of MTU e.g. Packet was encrypted
2034  *	    or there was not enough information to fanout to upper
2035  *	    layers. Thus on the next outbound datagram, ire_send_wire
2036  *	    generates the IFN, where IPsec processing has *not* been
2037  *	    done.
2038  *
2039  *	    Note that we retain ixa_fragsize across IPsec thus once
2040  *	    we have picking ixa_fragsize and entered ipsec_out_process we do
2041  *	    no change the fragsize even if the path MTU changes before
2042  *	    we reach ip_output_post_ipsec.
2043  *
2044  *	    In the local case, IRAF_LOOPBACK will be set indicating
2045  *	    that IFN was generated locally.
2046  *
2047  * ROUTER : IFN could be secure or non-secure.
2048  *
2049  *	    * SECURE : We use the IPSEC_IN to fanout to AH/ESP if the
2050  *	      packet in error has AH/ESP headers to validate the AH/ESP
2051  *	      headers. AH/ESP will verify whether there is a valid SA or
2052  *	      not and send it back. We will fanout again if we have more
2053  *	      data in the packet.
2054  *
2055  *	      If the packet in error does not have AH/ESP, we handle it
2056  *	      like any other case.
2057  *
2058  *	    * NON_SECURE : If the packet in error has AH/ESP headers, we send it
2059  *	      up to AH/ESP for validation. AH/ESP will verify whether there is a
2060  *	      valid SA or not and send it back. We will fanout again if
2061  *	      we have more data in the packet.
2062  *
2063  *	      If the packet in error does not have AH/ESP, we handle it
2064  *	      like any other case.
2065  *
2066  * The caller must have called icmp_inbound_verify_v4.
2067  */
2068 static void
icmp_inbound_error_fanout_v4(mblk_t * mp,icmph_t * icmph,ip_recv_attr_t * ira)2069 icmp_inbound_error_fanout_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
2070 {
2071 	uint16_t	*up;	/* Pointer to ports in ULP header */
2072 	uint32_t	ports;	/* reversed ports for fanout */
2073 	ipha_t		ripha;	/* With reversed addresses */
2074 	ipha_t		*ipha;  /* Inner IP header */
2075 	uint_t		hdr_length; /* Inner IP header length */
2076 	tcpha_t		*tcpha;
2077 	conn_t		*connp;
2078 	ill_t		*ill = ira->ira_ill;
2079 	ip_stack_t	*ipst = ill->ill_ipst;
2080 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
2081 	ill_t		*rill = ira->ira_rill;
2082 
2083 	/* Caller already pulled up everything. */
2084 	ipha = (ipha_t *)&icmph[1];
2085 	ASSERT((uchar_t *)&ipha[1] <= mp->b_wptr);
2086 	ASSERT(mp->b_cont == NULL);
2087 
2088 	hdr_length = IPH_HDR_LENGTH(ipha);
2089 	ira->ira_protocol = ipha->ipha_protocol;
2090 
2091 	/*
2092 	 * We need a separate IP header with the source and destination
2093 	 * addresses reversed to do fanout/classification because the ipha in
2094 	 * the ICMP error is in the form we sent it out.
2095 	 */
2096 	ripha.ipha_src = ipha->ipha_dst;
2097 	ripha.ipha_dst = ipha->ipha_src;
2098 	ripha.ipha_protocol = ipha->ipha_protocol;
2099 	ripha.ipha_version_and_hdr_length = ipha->ipha_version_and_hdr_length;
2100 
2101 	ip2dbg(("icmp_inbound_error_v4: proto %d %x to %x: %d/%d\n",
2102 	    ripha.ipha_protocol, ntohl(ipha->ipha_src),
2103 	    ntohl(ipha->ipha_dst),
2104 	    icmph->icmph_type, icmph->icmph_code));
2105 
2106 	switch (ipha->ipha_protocol) {
2107 	case IPPROTO_UDP:
2108 		up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2109 
2110 		/* Attempt to find a client stream based on port. */
2111 		ip2dbg(("icmp_inbound_error_v4: UDP ports %d to %d\n",
2112 		    ntohs(up[0]), ntohs(up[1])));
2113 
2114 		/* Note that we send error to all matches. */
2115 		ira->ira_flags |= IRAF_ICMP_ERROR;
2116 		ip_fanout_udp_multi_v4(mp, &ripha, up[0], up[1], ira);
2117 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2118 		return;
2119 
2120 	case IPPROTO_TCP:
2121 		/*
2122 		 * Find a TCP client stream for this packet.
2123 		 * Note that we do a reverse lookup since the header is
2124 		 * in the form we sent it out.
2125 		 */
2126 		tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
2127 		connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
2128 		    ipst);
2129 		if (connp == NULL)
2130 			goto discard_pkt;
2131 
2132 		if (connp->conn_min_ttl != 0 &&
2133 		    connp->conn_min_ttl > ira->ira_ttl) {
2134 			CONN_DEC_REF(connp);
2135 			goto discard_pkt;
2136 		}
2137 		if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) ||
2138 		    (ira->ira_flags & IRAF_IPSEC_SECURE)) {
2139 			mp = ipsec_check_inbound_policy(mp, connp,
2140 			    ipha, NULL, ira);
2141 			if (mp == NULL) {
2142 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2143 				/* Note that mp is NULL */
2144 				ip_drop_input("ipIfStatsInDiscards", mp, ill);
2145 				CONN_DEC_REF(connp);
2146 				return;
2147 			}
2148 		}
2149 
2150 		ira->ira_flags |= IRAF_ICMP_ERROR;
2151 		ira->ira_ill = ira->ira_rill = NULL;
2152 		if (IPCL_IS_TCP(connp)) {
2153 			SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
2154 			    connp->conn_recvicmp, connp, ira, SQ_FILL,
2155 			    SQTAG_TCP_INPUT_ICMP_ERR);
2156 		} else {
2157 			/* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
2158 			(connp->conn_recv)(connp, mp, NULL, ira);
2159 			CONN_DEC_REF(connp);
2160 		}
2161 		ira->ira_ill = ill;
2162 		ira->ira_rill = rill;
2163 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2164 		return;
2165 
2166 	case IPPROTO_SCTP:
2167 		up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2168 		/* Find a SCTP client stream for this packet. */
2169 		((uint16_t *)&ports)[0] = up[1];
2170 		((uint16_t *)&ports)[1] = up[0];
2171 
2172 		ira->ira_flags |= IRAF_ICMP_ERROR;
2173 		ip_fanout_sctp(mp, &ripha, NULL, ports, ira);
2174 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2175 		return;
2176 
2177 	case IPPROTO_ESP:
2178 	case IPPROTO_AH:
2179 		if (!ipsec_loaded(ipss)) {
2180 			ip_proto_not_sup(mp, ira);
2181 			return;
2182 		}
2183 
2184 		if (ipha->ipha_protocol == IPPROTO_ESP)
2185 			mp = ipsecesp_icmp_error(mp, ira);
2186 		else
2187 			mp = ipsecah_icmp_error(mp, ira);
2188 		if (mp == NULL)
2189 			return;
2190 
2191 		/* Just in case ipsec didn't preserve the NULL b_cont */
2192 		if (mp->b_cont != NULL) {
2193 			if (!pullupmsg(mp, -1))
2194 				goto discard_pkt;
2195 		}
2196 
2197 		/*
2198 		 * Note that ira_pktlen and ira_ip_hdr_length are no longer
2199 		 * correct, but we don't use them any more here.
2200 		 *
2201 		 * If succesful, the mp has been modified to not include
2202 		 * the ESP/AH header so we can fanout to the ULP's icmp
2203 		 * error handler.
2204 		 */
2205 		if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2206 			goto truncated;
2207 
2208 		/* Verify the modified message before any further processes. */
2209 		ipha = (ipha_t *)mp->b_rptr;
2210 		hdr_length = IPH_HDR_LENGTH(ipha);
2211 		icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2212 		if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2213 			freemsg(mp);
2214 			return;
2215 		}
2216 
2217 		icmp_inbound_error_fanout_v4(mp, icmph, ira);
2218 		return;
2219 
2220 	case IPPROTO_ENCAP: {
2221 		/* Look for self-encapsulated packets that caused an error */
2222 		ipha_t *in_ipha;
2223 
2224 		/*
2225 		 * Caller has verified that length has to be
2226 		 * at least the size of IP header.
2227 		 */
2228 		ASSERT(hdr_length >= sizeof (ipha_t));
2229 		/*
2230 		 * Check the sanity of the inner IP header like
2231 		 * we did for the outer header.
2232 		 */
2233 		in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2234 		if ((IPH_HDR_VERSION(in_ipha) != IPV4_VERSION)) {
2235 			goto discard_pkt;
2236 		}
2237 		if (IPH_HDR_LENGTH(in_ipha) < sizeof (ipha_t)) {
2238 			goto discard_pkt;
2239 		}
2240 		/* Check for Self-encapsulated tunnels */
2241 		if (in_ipha->ipha_src == ipha->ipha_src &&
2242 		    in_ipha->ipha_dst == ipha->ipha_dst) {
2243 
2244 			mp = icmp_inbound_self_encap_error_v4(mp, ipha,
2245 			    in_ipha);
2246 			if (mp == NULL)
2247 				goto discard_pkt;
2248 
2249 			/*
2250 			 * Just in case self_encap didn't preserve the NULL
2251 			 * b_cont
2252 			 */
2253 			if (mp->b_cont != NULL) {
2254 				if (!pullupmsg(mp, -1))
2255 					goto discard_pkt;
2256 			}
2257 			/*
2258 			 * Note that ira_pktlen and ira_ip_hdr_length are no
2259 			 * longer correct, but we don't use them any more here.
2260 			 */
2261 			if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2262 				goto truncated;
2263 
2264 			/*
2265 			 * Verify the modified message before any further
2266 			 * processes.
2267 			 */
2268 			ipha = (ipha_t *)mp->b_rptr;
2269 			hdr_length = IPH_HDR_LENGTH(ipha);
2270 			icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2271 			if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2272 				freemsg(mp);
2273 				return;
2274 			}
2275 
2276 			/*
2277 			 * The packet in error is self-encapsualted.
2278 			 * And we are finding it further encapsulated
2279 			 * which we could not have possibly generated.
2280 			 */
2281 			if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2282 				goto discard_pkt;
2283 			}
2284 			icmp_inbound_error_fanout_v4(mp, icmph, ira);
2285 			return;
2286 		}
2287 		/* No self-encapsulated */
2288 	}
2289 	/* FALLTHROUGH */
2290 	case IPPROTO_IPV6:
2291 		if ((connp = ipcl_iptun_classify_v4(&ripha.ipha_src,
2292 		    &ripha.ipha_dst, ipst)) != NULL) {
2293 			ira->ira_flags |= IRAF_ICMP_ERROR;
2294 			connp->conn_recvicmp(connp, mp, NULL, ira);
2295 			CONN_DEC_REF(connp);
2296 			ira->ira_flags &= ~IRAF_ICMP_ERROR;
2297 			return;
2298 		}
2299 		/*
2300 		 * No IP tunnel is interested, fallthrough and see
2301 		 * if a raw socket will want it.
2302 		 */
2303 		/* FALLTHROUGH */
2304 	default:
2305 		ira->ira_flags |= IRAF_ICMP_ERROR;
2306 		ip_fanout_proto_v4(mp, &ripha, ira);
2307 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2308 		return;
2309 	}
2310 	/* NOTREACHED */
2311 discard_pkt:
2312 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2313 	ip1dbg(("icmp_inbound_error_fanout_v4: drop pkt\n"));
2314 	ip_drop_input("ipIfStatsInDiscards", mp, ill);
2315 	freemsg(mp);
2316 	return;
2317 
2318 truncated:
2319 	/* We pulled up everthing already. Must be truncated */
2320 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
2321 	ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
2322 	freemsg(mp);
2323 }
2324 
2325 /*
2326  * Common IP options parser.
2327  *
2328  * Setup routine: fill in *optp with options-parsing state, then
2329  * tail-call ipoptp_next to return the first option.
2330  */
2331 uint8_t
ipoptp_first(ipoptp_t * optp,ipha_t * ipha)2332 ipoptp_first(ipoptp_t *optp, ipha_t *ipha)
2333 {
2334 	uint32_t totallen; /* total length of all options */
2335 
2336 	totallen = ipha->ipha_version_and_hdr_length -
2337 	    (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
2338 	totallen <<= 2;
2339 	optp->ipoptp_next = (uint8_t *)(&ipha[1]);
2340 	optp->ipoptp_end = optp->ipoptp_next + totallen;
2341 	optp->ipoptp_flags = 0;
2342 	return (ipoptp_next(optp));
2343 }
2344 
2345 /* Like above but without an ipha_t */
2346 uint8_t
ipoptp_first2(ipoptp_t * optp,uint32_t totallen,uint8_t * opt)2347 ipoptp_first2(ipoptp_t *optp, uint32_t totallen, uint8_t *opt)
2348 {
2349 	optp->ipoptp_next = opt;
2350 	optp->ipoptp_end = optp->ipoptp_next + totallen;
2351 	optp->ipoptp_flags = 0;
2352 	return (ipoptp_next(optp));
2353 }
2354 
2355 /*
2356  * Common IP options parser: extract next option.
2357  */
2358 uint8_t
ipoptp_next(ipoptp_t * optp)2359 ipoptp_next(ipoptp_t *optp)
2360 {
2361 	uint8_t *end = optp->ipoptp_end;
2362 	uint8_t *cur = optp->ipoptp_next;
2363 	uint8_t opt, len, pointer;
2364 
2365 	/*
2366 	 * If cur > end already, then the ipoptp_end or ipoptp_next pointer
2367 	 * has been corrupted.
2368 	 */
2369 	ASSERT(cur <= end);
2370 
2371 	if (cur == end)
2372 		return (IPOPT_EOL);
2373 
2374 	opt = cur[IPOPT_OPTVAL];
2375 
2376 	/*
2377 	 * Skip any NOP options.
2378 	 */
2379 	while (opt == IPOPT_NOP) {
2380 		cur++;
2381 		if (cur == end)
2382 			return (IPOPT_EOL);
2383 		opt = cur[IPOPT_OPTVAL];
2384 	}
2385 
2386 	if (opt == IPOPT_EOL)
2387 		return (IPOPT_EOL);
2388 
2389 	/*
2390 	 * Option requiring a length.
2391 	 */
2392 	if ((cur + 1) >= end) {
2393 		optp->ipoptp_flags |= IPOPTP_ERROR;
2394 		return (IPOPT_EOL);
2395 	}
2396 	len = cur[IPOPT_OLEN];
2397 	if (len < 2) {
2398 		optp->ipoptp_flags |= IPOPTP_ERROR;
2399 		return (IPOPT_EOL);
2400 	}
2401 	optp->ipoptp_cur = cur;
2402 	optp->ipoptp_len = len;
2403 	optp->ipoptp_next = cur + len;
2404 	if (cur + len > end) {
2405 		optp->ipoptp_flags |= IPOPTP_ERROR;
2406 		return (IPOPT_EOL);
2407 	}
2408 
2409 	/*
2410 	 * For the options which require a pointer field, make sure
2411 	 * its there, and make sure it points to either something
2412 	 * inside this option, or the end of the option.
2413 	 */
2414 	pointer = IPOPT_EOL;
2415 	switch (opt) {
2416 	case IPOPT_RR:
2417 	case IPOPT_TS:
2418 	case IPOPT_LSRR:
2419 	case IPOPT_SSRR:
2420 		if (len <= IPOPT_OFFSET) {
2421 			optp->ipoptp_flags |= IPOPTP_ERROR;
2422 			return (opt);
2423 		}
2424 		pointer = cur[IPOPT_OFFSET];
2425 		if (pointer - 1 > len) {
2426 			optp->ipoptp_flags |= IPOPTP_ERROR;
2427 			return (opt);
2428 		}
2429 		break;
2430 	}
2431 
2432 	/*
2433 	 * Sanity check the pointer field based on the type of the
2434 	 * option.
2435 	 */
2436 	switch (opt) {
2437 	case IPOPT_RR:
2438 	case IPOPT_SSRR:
2439 	case IPOPT_LSRR:
2440 		if (pointer < IPOPT_MINOFF_SR)
2441 			optp->ipoptp_flags |= IPOPTP_ERROR;
2442 		break;
2443 	case IPOPT_TS:
2444 		if (pointer < IPOPT_MINOFF_IT)
2445 			optp->ipoptp_flags |= IPOPTP_ERROR;
2446 		/*
2447 		 * Note that the Internet Timestamp option also
2448 		 * contains two four bit fields (the Overflow field,
2449 		 * and the Flag field), which follow the pointer
2450 		 * field.  We don't need to check that these fields
2451 		 * fall within the length of the option because this
2452 		 * was implicitely done above.  We've checked that the
2453 		 * pointer value is at least IPOPT_MINOFF_IT, and that
2454 		 * it falls within the option.  Since IPOPT_MINOFF_IT >
2455 		 * IPOPT_POS_OV_FLG, we don't need the explicit check.
2456 		 */
2457 		ASSERT(len > IPOPT_POS_OV_FLG);
2458 		break;
2459 	}
2460 
2461 	return (opt);
2462 }
2463 
2464 /*
2465  * Use the outgoing IP header to create an IP_OPTIONS option the way
2466  * it was passed down from the application.
2467  *
2468  * This is compatible with BSD in that it returns
2469  * the reverse source route with the final destination
2470  * as the last entry. The first 4 bytes of the option
2471  * will contain the final destination.
2472  */
2473 int
ip_opt_get_user(conn_t * connp,uchar_t * buf)2474 ip_opt_get_user(conn_t *connp, uchar_t *buf)
2475 {
2476 	ipoptp_t	opts;
2477 	uchar_t		*opt;
2478 	uint8_t		optval;
2479 	uint8_t		optlen;
2480 	uint32_t	len = 0;
2481 	uchar_t		*buf1 = buf;
2482 	uint32_t	totallen;
2483 	ipaddr_t	dst;
2484 	ip_pkt_t	*ipp = &connp->conn_xmit_ipp;
2485 
2486 	if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
2487 		return (0);
2488 
2489 	totallen = ipp->ipp_ipv4_options_len;
2490 	if (totallen & 0x3)
2491 		return (0);
2492 
2493 	buf += IP_ADDR_LEN;	/* Leave room for final destination */
2494 	len += IP_ADDR_LEN;
2495 	bzero(buf1, IP_ADDR_LEN);
2496 
2497 	dst = connp->conn_faddr_v4;
2498 
2499 	for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
2500 	    optval != IPOPT_EOL;
2501 	    optval = ipoptp_next(&opts)) {
2502 		int	off;
2503 
2504 		opt = opts.ipoptp_cur;
2505 		if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
2506 			break;
2507 		}
2508 		optlen = opts.ipoptp_len;
2509 
2510 		switch (optval) {
2511 		case IPOPT_SSRR:
2512 		case IPOPT_LSRR:
2513 
2514 			/*
2515 			 * Insert destination as the first entry in the source
2516 			 * route and move down the entries on step.
2517 			 * The last entry gets placed at buf1.
2518 			 */
2519 			buf[IPOPT_OPTVAL] = optval;
2520 			buf[IPOPT_OLEN] = optlen;
2521 			buf[IPOPT_OFFSET] = optlen;
2522 
2523 			off = optlen - IP_ADDR_LEN;
2524 			if (off < 0) {
2525 				/* No entries in source route */
2526 				break;
2527 			}
2528 			/* Last entry in source route if not already set */
2529 			if (dst == INADDR_ANY)
2530 				bcopy(opt + off, buf1, IP_ADDR_LEN);
2531 			off -= IP_ADDR_LEN;
2532 
2533 			while (off > 0) {
2534 				bcopy(opt + off,
2535 				    buf + off + IP_ADDR_LEN,
2536 				    IP_ADDR_LEN);
2537 				off -= IP_ADDR_LEN;
2538 			}
2539 			/* ipha_dst into first slot */
2540 			bcopy(&dst, buf + off + IP_ADDR_LEN,
2541 			    IP_ADDR_LEN);
2542 			buf += optlen;
2543 			len += optlen;
2544 			break;
2545 
2546 		default:
2547 			bcopy(opt, buf, optlen);
2548 			buf += optlen;
2549 			len += optlen;
2550 			break;
2551 		}
2552 	}
2553 done:
2554 	/* Pad the resulting options */
2555 	while (len & 0x3) {
2556 		*buf++ = IPOPT_EOL;
2557 		len++;
2558 	}
2559 	return (len);
2560 }
2561 
2562 /*
2563  * Update any record route or timestamp options to include this host.
2564  * Reverse any source route option.
2565  * This routine assumes that the options are well formed i.e. that they
2566  * have already been checked.
2567  */
2568 static void
icmp_options_update(ipha_t * ipha)2569 icmp_options_update(ipha_t *ipha)
2570 {
2571 	ipoptp_t	opts;
2572 	uchar_t		*opt;
2573 	uint8_t		optval;
2574 	ipaddr_t	src;		/* Our local address */
2575 	ipaddr_t	dst;
2576 
2577 	ip2dbg(("icmp_options_update\n"));
2578 	src = ipha->ipha_src;
2579 	dst = ipha->ipha_dst;
2580 
2581 	for (optval = ipoptp_first(&opts, ipha);
2582 	    optval != IPOPT_EOL;
2583 	    optval = ipoptp_next(&opts)) {
2584 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
2585 		opt = opts.ipoptp_cur;
2586 		ip2dbg(("icmp_options_update: opt %d, len %d\n",
2587 		    optval, opts.ipoptp_len));
2588 		switch (optval) {
2589 			int off1, off2;
2590 		case IPOPT_SSRR:
2591 		case IPOPT_LSRR:
2592 			/*
2593 			 * Reverse the source route.  The first entry
2594 			 * should be the next to last one in the current
2595 			 * source route (the last entry is our address).
2596 			 * The last entry should be the final destination.
2597 			 */
2598 			off1 = IPOPT_MINOFF_SR - 1;
2599 			off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
2600 			if (off2 < 0) {
2601 				/* No entries in source route */
2602 				ip1dbg((
2603 				    "icmp_options_update: bad src route\n"));
2604 				break;
2605 			}
2606 			bcopy((char *)opt + off2, &dst, IP_ADDR_LEN);
2607 			bcopy(&ipha->ipha_dst, (char *)opt + off2, IP_ADDR_LEN);
2608 			bcopy(&dst, &ipha->ipha_dst, IP_ADDR_LEN);
2609 			off2 -= IP_ADDR_LEN;
2610 
2611 			while (off1 < off2) {
2612 				bcopy((char *)opt + off1, &src, IP_ADDR_LEN);
2613 				bcopy((char *)opt + off2, (char *)opt + off1,
2614 				    IP_ADDR_LEN);
2615 				bcopy(&src, (char *)opt + off2, IP_ADDR_LEN);
2616 				off1 += IP_ADDR_LEN;
2617 				off2 -= IP_ADDR_LEN;
2618 			}
2619 			opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
2620 			break;
2621 		}
2622 	}
2623 }
2624 
2625 /*
2626  * Process received ICMP Redirect messages.
2627  * Assumes the caller has verified that the headers are in the pulled up mblk.
2628  * Consumes mp.
2629  */
2630 static void
icmp_redirect_v4(mblk_t * mp,ipha_t * ipha,icmph_t * icmph,ip_recv_attr_t * ira)2631 icmp_redirect_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph, ip_recv_attr_t *ira)
2632 {
2633 	ire_t		*ire, *nire;
2634 	ire_t		*prev_ire;
2635 	ipaddr_t	src, dst, gateway;
2636 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
2637 	ipha_t		*inner_ipha;	/* Inner IP header */
2638 
2639 	/* Caller already pulled up everything. */
2640 	inner_ipha = (ipha_t *)&icmph[1];
2641 	src = ipha->ipha_src;
2642 	dst = inner_ipha->ipha_dst;
2643 	gateway = icmph->icmph_rd_gateway;
2644 	/* Make sure the new gateway is reachable somehow. */
2645 	ire = ire_ftable_lookup_v4(gateway, 0, 0, IRE_ONLINK, NULL,
2646 	    ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst, NULL);
2647 	/*
2648 	 * Make sure we had a route for the dest in question and that
2649 	 * that route was pointing to the old gateway (the source of the
2650 	 * redirect packet.)
2651 	 * We do longest match and then compare ire_gateway_addr below.
2652 	 */
2653 	prev_ire = ire_ftable_lookup_v4(dst, 0, 0, 0, NULL, ALL_ZONES,
2654 	    NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
2655 	/*
2656 	 * Check that
2657 	 *	the redirect was not from ourselves
2658 	 *	the new gateway and the old gateway are directly reachable
2659 	 */
2660 	if (prev_ire == NULL || ire == NULL ||
2661 	    (prev_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) ||
2662 	    (prev_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
2663 	    !(ire->ire_type & IRE_IF_ALL) ||
2664 	    prev_ire->ire_gateway_addr != src) {
2665 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2666 		ip_drop_input("icmpInBadRedirects - ire", mp, ira->ira_ill);
2667 		freemsg(mp);
2668 		if (ire != NULL)
2669 			ire_refrele(ire);
2670 		if (prev_ire != NULL)
2671 			ire_refrele(prev_ire);
2672 		return;
2673 	}
2674 
2675 	ire_refrele(prev_ire);
2676 	ire_refrele(ire);
2677 
2678 	/*
2679 	 * TODO: more precise handling for cases 0, 2, 3, the latter two
2680 	 * require TOS routing
2681 	 */
2682 	switch (icmph->icmph_code) {
2683 	case 0:
2684 	case 1:
2685 		/* TODO: TOS specificity for cases 2 and 3 */
2686 	case 2:
2687 	case 3:
2688 		break;
2689 	default:
2690 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2691 		ip_drop_input("icmpInBadRedirects - code", mp, ira->ira_ill);
2692 		freemsg(mp);
2693 		return;
2694 	}
2695 	/*
2696 	 * Create a Route Association.  This will allow us to remember that
2697 	 * someone we believe told us to use the particular gateway.
2698 	 */
2699 	ire = ire_create(
2700 	    (uchar_t *)&dst,			/* dest addr */
2701 	    (uchar_t *)&ip_g_all_ones,		/* mask */
2702 	    (uchar_t *)&gateway,		/* gateway addr */
2703 	    IRE_HOST,
2704 	    NULL,				/* ill */
2705 	    ALL_ZONES,
2706 	    (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST),
2707 	    NULL,				/* tsol_gc_t */
2708 	    ipst);
2709 
2710 	if (ire == NULL) {
2711 		freemsg(mp);
2712 		return;
2713 	}
2714 	nire = ire_add(ire);
2715 	/* Check if it was a duplicate entry */
2716 	if (nire != NULL && nire != ire) {
2717 		ASSERT(nire->ire_identical_ref > 1);
2718 		ire_delete(nire);
2719 		ire_refrele(nire);
2720 		nire = NULL;
2721 	}
2722 	ire = nire;
2723 	if (ire != NULL) {
2724 		ire_refrele(ire);		/* Held in ire_add */
2725 
2726 		/* tell routing sockets that we received a redirect */
2727 		ip_rts_change(RTM_REDIRECT, dst, gateway, IP_HOST_MASK, 0, src,
2728 		    (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST), 0,
2729 		    (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_AUTHOR), ipst);
2730 	}
2731 
2732 	/*
2733 	 * Delete any existing IRE_HOST type redirect ires for this destination.
2734 	 * This together with the added IRE has the effect of
2735 	 * modifying an existing redirect.
2736 	 */
2737 	prev_ire = ire_ftable_lookup_v4(dst, 0, src, IRE_HOST, NULL,
2738 	    ALL_ZONES, NULL, (MATCH_IRE_GW | MATCH_IRE_TYPE), 0, ipst, NULL);
2739 	if (prev_ire != NULL) {
2740 		if (prev_ire ->ire_flags & RTF_DYNAMIC)
2741 			ire_delete(prev_ire);
2742 		ire_refrele(prev_ire);
2743 	}
2744 
2745 	freemsg(mp);
2746 }
2747 
2748 /*
2749  * Generate an ICMP parameter problem message.
2750  * When called from ip_output side a minimal ip_recv_attr_t needs to be
2751  * constructed by the caller.
2752  */
2753 static void
icmp_param_problem(mblk_t * mp,uint8_t ptr,ip_recv_attr_t * ira)2754 icmp_param_problem(mblk_t *mp, uint8_t ptr, ip_recv_attr_t *ira)
2755 {
2756 	icmph_t	icmph;
2757 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
2758 
2759 	mp = icmp_pkt_err_ok(mp, ira);
2760 	if (mp == NULL)
2761 		return;
2762 
2763 	bzero(&icmph, sizeof (icmph_t));
2764 	icmph.icmph_type = ICMP_PARAM_PROBLEM;
2765 	icmph.icmph_pp_ptr = ptr;
2766 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutParmProbs);
2767 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
2768 }
2769 
2770 /*
2771  * Build and ship an IPv4 ICMP message using the packet data in mp, and
2772  * the ICMP header pointed to by "stuff".  (May be called as writer.)
2773  * Note: assumes that icmp_pkt_err_ok has been called to verify that
2774  * an icmp error packet can be sent.
2775  * Assigns an appropriate source address to the packet. If ipha_dst is
2776  * one of our addresses use it for source. Otherwise let ip_output_simple
2777  * pick the source address.
2778  */
2779 static void
icmp_pkt(mblk_t * mp,void * stuff,size_t len,ip_recv_attr_t * ira)2780 icmp_pkt(mblk_t *mp, void *stuff, size_t len, ip_recv_attr_t *ira)
2781 {
2782 	ipaddr_t dst;
2783 	icmph_t	*icmph;
2784 	ipha_t	*ipha;
2785 	uint_t	len_needed;
2786 	size_t	msg_len;
2787 	mblk_t	*mp1;
2788 	ipaddr_t src;
2789 	ire_t	*ire;
2790 	ip_xmit_attr_t ixas;
2791 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2792 
2793 	ipha = (ipha_t *)mp->b_rptr;
2794 
2795 	bzero(&ixas, sizeof (ixas));
2796 	ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
2797 	ixas.ixa_zoneid = ira->ira_zoneid;
2798 	ixas.ixa_ifindex = 0;
2799 	ixas.ixa_ipst = ipst;
2800 	ixas.ixa_cred = kcred;
2801 	ixas.ixa_cpid = NOPID;
2802 	ixas.ixa_tsl = ira->ira_tsl;	/* Behave as a multi-level responder */
2803 	ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
2804 
2805 	if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2806 		/*
2807 		 * Apply IPsec based on how IPsec was applied to
2808 		 * the packet that had the error.
2809 		 *
2810 		 * If it was an outbound packet that caused the ICMP
2811 		 * error, then the caller will have setup the IRA
2812 		 * appropriately.
2813 		 */
2814 		if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
2815 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2816 			/* Note: mp already consumed and ip_drop_packet done */
2817 			return;
2818 		}
2819 	} else {
2820 		/*
2821 		 * This is in clear. The icmp message we are building
2822 		 * here should go out in clear, independent of our policy.
2823 		 */
2824 		ixas.ixa_flags |= IXAF_NO_IPSEC;
2825 	}
2826 
2827 	/* Remember our eventual destination */
2828 	dst = ipha->ipha_src;
2829 
2830 	/*
2831 	 * If the packet was for one of our unicast addresses, make
2832 	 * sure we respond with that as the source. Otherwise
2833 	 * have ip_output_simple pick the source address.
2834 	 */
2835 	ire = ire_ftable_lookup_v4(ipha->ipha_dst, 0, 0,
2836 	    (IRE_LOCAL|IRE_LOOPBACK), NULL, ira->ira_zoneid, NULL,
2837 	    MATCH_IRE_TYPE|MATCH_IRE_ZONEONLY, 0, ipst, NULL);
2838 	if (ire != NULL) {
2839 		ire_refrele(ire);
2840 		src = ipha->ipha_dst;
2841 	} else {
2842 		src = INADDR_ANY;
2843 		ixas.ixa_flags |= IXAF_SET_SOURCE;
2844 	}
2845 
2846 	/*
2847 	 * Check if we can send back more then 8 bytes in addition to
2848 	 * the IP header.  We try to send 64 bytes of data and the internal
2849 	 * header in the special cases of ipv4 encapsulated ipv4 or ipv6.
2850 	 */
2851 	len_needed = IPH_HDR_LENGTH(ipha);
2852 	if (ipha->ipha_protocol == IPPROTO_ENCAP ||
2853 	    ipha->ipha_protocol == IPPROTO_IPV6) {
2854 		/*
2855 		 * NOTE: It is posssible that the inner packet is poorly
2856 		 * formed (e.g. IP version is corrupt, or v6 extension headers
2857 		 * got cut off).  The receiver of the ICMP message should see
2858 		 * what we saw.  In the absence of a sane inner-packet (which
2859 		 * protocol types IPPPROTO_ENCAP and IPPROTO_IPV6 indicate
2860 		 * would be an IP header), we should send the size of what is
2861 		 * normally expected to be there (either sizeof (ipha_t) or
2862 		 * sizeof (ip6_t).  It may be useful for diagnostic purposes.
2863 		 *
2864 		 * ALSO NOTE: "inner_ip6h" is the inner packet header, v4 or v6.
2865 		 */
2866 		ip6_t *inner_ip6h = (ip6_t *)((uchar_t *)ipha + len_needed);
2867 
2868 		if (!pullupmsg(mp, -1)) {
2869 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2870 			ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2871 			freemsg(mp);
2872 			return;
2873 		}
2874 		ipha = (ipha_t *)mp->b_rptr;
2875 
2876 		if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2877 			/*
2878 			 * Check the inner IP version here to guard against
2879 			 * bogons.
2880 			 */
2881 			if (IPH_HDR_VERSION(inner_ip6h) == IPV4_VERSION) {
2882 				len_needed +=
2883 				    IPH_HDR_LENGTH(((uchar_t *)inner_ip6h));
2884 			} else {
2885 				len_needed = sizeof (ipha_t);
2886 			}
2887 		} else {
2888 			ASSERT(ipha->ipha_protocol == IPPROTO_IPV6);
2889 			/* function called next-line checks inner IP version */
2890 			len_needed += ip_hdr_length_v6(mp, inner_ip6h);
2891 		}
2892 	}
2893 	len_needed += ipst->ips_ip_icmp_return;
2894 	msg_len = msgdsize(mp);
2895 	if (msg_len > len_needed) {
2896 		(void) adjmsg(mp, len_needed - msg_len);
2897 		msg_len = len_needed;
2898 	}
2899 	mp1 = allocb(sizeof (icmp_ipha) + len, BPRI_MED);
2900 	if (mp1 == NULL) {
2901 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutErrors);
2902 		freemsg(mp);
2903 		return;
2904 	}
2905 	mp1->b_cont = mp;
2906 	mp = mp1;
2907 
2908 	/*
2909 	 * Set IXAF_TRUSTED_ICMP so we can let the ICMP messages this
2910 	 * node generates be accepted in peace by all on-host destinations.
2911 	 * If we do NOT assume that all on-host destinations trust
2912 	 * self-generated ICMP messages, then rework here, ip6.c, and spd.c.
2913 	 * (Look for IXAF_TRUSTED_ICMP).
2914 	 */
2915 	ixas.ixa_flags |= IXAF_TRUSTED_ICMP;
2916 
2917 	ipha = (ipha_t *)mp->b_rptr;
2918 	mp1->b_wptr = (uchar_t *)ipha + (sizeof (icmp_ipha) + len);
2919 	*ipha = icmp_ipha;
2920 	ipha->ipha_src = src;
2921 	ipha->ipha_dst = dst;
2922 	ipha->ipha_ttl = ipst->ips_ip_def_ttl;
2923 	msg_len += sizeof (icmp_ipha) + len;
2924 	if (msg_len > IP_MAXPACKET) {
2925 		(void) adjmsg(mp, IP_MAXPACKET - msg_len);
2926 		msg_len = IP_MAXPACKET;
2927 	}
2928 	ipha->ipha_length = htons((uint16_t)msg_len);
2929 	icmph = (icmph_t *)&ipha[1];
2930 	bcopy(stuff, icmph, len);
2931 	icmph->icmph_checksum = 0;
2932 	icmph->icmph_checksum = IP_CSUM(mp, (int32_t)sizeof (ipha_t), 0);
2933 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
2934 
2935 	(void) ip_output_simple(mp, &ixas);
2936 	ixa_cleanup(&ixas);
2937 }
2938 
2939 /*
2940  * Determine if an ICMP error packet can be sent given the rate limit.
2941  * The limit consists of an average frequency (icmp_pkt_err_interval measured
2942  * in milliseconds) and a burst size. Burst size number of packets can
2943  * be sent arbitrarely closely spaced.
2944  * The state is tracked using two variables to implement an approximate
2945  * token bucket filter:
2946  *	icmp_pkt_err_last - lbolt value when the last burst started
2947  *	icmp_pkt_err_sent - number of packets sent in current burst
2948  */
2949 boolean_t
icmp_err_rate_limit(ip_stack_t * ipst)2950 icmp_err_rate_limit(ip_stack_t *ipst)
2951 {
2952 	clock_t now = TICK_TO_MSEC(ddi_get_lbolt());
2953 	uint_t refilled; /* Number of packets refilled in tbf since last */
2954 	/* Guard against changes by loading into local variable */
2955 	uint_t err_interval = ipst->ips_ip_icmp_err_interval;
2956 
2957 	if (err_interval == 0)
2958 		return (B_FALSE);
2959 
2960 	if (ipst->ips_icmp_pkt_err_last > now) {
2961 		/* 100HZ lbolt in ms for 32bit arch wraps every 49.7 days */
2962 		ipst->ips_icmp_pkt_err_last = 0;
2963 		ipst->ips_icmp_pkt_err_sent = 0;
2964 	}
2965 	/*
2966 	 * If we are in a burst update the token bucket filter.
2967 	 * Update the "last" time to be close to "now" but make sure
2968 	 * we don't loose precision.
2969 	 */
2970 	if (ipst->ips_icmp_pkt_err_sent != 0) {
2971 		refilled = (now - ipst->ips_icmp_pkt_err_last)/err_interval;
2972 		if (refilled > ipst->ips_icmp_pkt_err_sent) {
2973 			ipst->ips_icmp_pkt_err_sent = 0;
2974 		} else {
2975 			ipst->ips_icmp_pkt_err_sent -= refilled;
2976 			ipst->ips_icmp_pkt_err_last += refilled * err_interval;
2977 		}
2978 	}
2979 	if (ipst->ips_icmp_pkt_err_sent == 0) {
2980 		/* Start of new burst */
2981 		ipst->ips_icmp_pkt_err_last = now;
2982 	}
2983 	if (ipst->ips_icmp_pkt_err_sent < ipst->ips_ip_icmp_err_burst) {
2984 		ipst->ips_icmp_pkt_err_sent++;
2985 		ip1dbg(("icmp_err_rate_limit: %d sent in burst\n",
2986 		    ipst->ips_icmp_pkt_err_sent));
2987 		return (B_FALSE);
2988 	}
2989 	ip1dbg(("icmp_err_rate_limit: dropped\n"));
2990 	return (B_TRUE);
2991 }
2992 
2993 /*
2994  * Check if it is ok to send an IPv4 ICMP error packet in
2995  * response to the IPv4 packet in mp.
2996  * Free the message and return null if no
2997  * ICMP error packet should be sent.
2998  */
2999 static mblk_t *
icmp_pkt_err_ok(mblk_t * mp,ip_recv_attr_t * ira)3000 icmp_pkt_err_ok(mblk_t *mp, ip_recv_attr_t *ira)
3001 {
3002 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
3003 	icmph_t	*icmph;
3004 	ipha_t	*ipha;
3005 	uint_t	len_needed;
3006 
3007 	if (!mp)
3008 		return (NULL);
3009 	ipha = (ipha_t *)mp->b_rptr;
3010 	if (ip_csum_hdr(ipha)) {
3011 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInCksumErrs);
3012 		ip_drop_input("ipIfStatsInCksumErrs", mp, NULL);
3013 		freemsg(mp);
3014 		return (NULL);
3015 	}
3016 	if (ip_type_v4(ipha->ipha_dst, ipst) == IRE_BROADCAST ||
3017 	    ip_type_v4(ipha->ipha_src, ipst) == IRE_BROADCAST ||
3018 	    CLASSD(ipha->ipha_dst) ||
3019 	    CLASSD(ipha->ipha_src) ||
3020 	    (ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET)) {
3021 		/* Note: only errors to the fragment with offset 0 */
3022 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3023 		freemsg(mp);
3024 		return (NULL);
3025 	}
3026 	if (ipha->ipha_protocol == IPPROTO_ICMP) {
3027 		/*
3028 		 * Check the ICMP type.  RFC 1122 sez:  don't send ICMP
3029 		 * errors in response to any ICMP errors.
3030 		 */
3031 		len_needed = IPH_HDR_LENGTH(ipha) + ICMPH_SIZE;
3032 		if (mp->b_wptr - mp->b_rptr < len_needed) {
3033 			if (!pullupmsg(mp, len_needed)) {
3034 				BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3035 				freemsg(mp);
3036 				return (NULL);
3037 			}
3038 			ipha = (ipha_t *)mp->b_rptr;
3039 		}
3040 		icmph = (icmph_t *)
3041 		    (&((char *)ipha)[IPH_HDR_LENGTH(ipha)]);
3042 		switch (icmph->icmph_type) {
3043 		case ICMP_DEST_UNREACHABLE:
3044 		case ICMP_SOURCE_QUENCH:
3045 		case ICMP_TIME_EXCEEDED:
3046 		case ICMP_PARAM_PROBLEM:
3047 		case ICMP_REDIRECT:
3048 			BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3049 			freemsg(mp);
3050 			return (NULL);
3051 		default:
3052 			break;
3053 		}
3054 	}
3055 	/*
3056 	 * If this is a labeled system, then check to see if we're allowed to
3057 	 * send a response to this particular sender.  If not, then just drop.
3058 	 */
3059 	if (is_system_labeled() && !tsol_can_reply_error(mp, ira)) {
3060 		ip2dbg(("icmp_pkt_err_ok: can't respond to packet\n"));
3061 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3062 		freemsg(mp);
3063 		return (NULL);
3064 	}
3065 	if (icmp_err_rate_limit(ipst)) {
3066 		/*
3067 		 * Only send ICMP error packets every so often.
3068 		 * This should be done on a per port/source basis,
3069 		 * but for now this will suffice.
3070 		 */
3071 		freemsg(mp);
3072 		return (NULL);
3073 	}
3074 	return (mp);
3075 }
3076 
3077 /*
3078  * Called when a packet was sent out the same link that it arrived on.
3079  * Check if it is ok to send a redirect and then send it.
3080  */
3081 void
ip_send_potential_redirect_v4(mblk_t * mp,ipha_t * ipha,ire_t * ire,ip_recv_attr_t * ira)3082 ip_send_potential_redirect_v4(mblk_t *mp, ipha_t *ipha, ire_t *ire,
3083     ip_recv_attr_t *ira)
3084 {
3085 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
3086 	ipaddr_t	src, nhop;
3087 	mblk_t		*mp1;
3088 	ire_t		*nhop_ire;
3089 
3090 	/*
3091 	 * Check the source address to see if it originated
3092 	 * on the same logical subnet it is going back out on.
3093 	 * If so, we should be able to send it a redirect.
3094 	 * Avoid sending a redirect if the destination
3095 	 * is directly connected (i.e., we matched an IRE_ONLINK),
3096 	 * or if the packet was source routed out this interface.
3097 	 *
3098 	 * We avoid sending a redirect if the
3099 	 * destination is directly connected
3100 	 * because it is possible that multiple
3101 	 * IP subnets may have been configured on
3102 	 * the link, and the source may not
3103 	 * be on the same subnet as ip destination,
3104 	 * even though they are on the same
3105 	 * physical link.
3106 	 */
3107 	if ((ire->ire_type & IRE_ONLINK) ||
3108 	    ip_source_routed(ipha, ipst))
3109 		return;
3110 
3111 	nhop_ire = ire_nexthop(ire);
3112 	if (nhop_ire == NULL)
3113 		return;
3114 
3115 	nhop = nhop_ire->ire_addr;
3116 
3117 	if (nhop_ire->ire_type & IRE_IF_CLONE) {
3118 		ire_t	*ire2;
3119 
3120 		/* Follow ire_dep_parent to find non-clone IRE_INTERFACE */
3121 		mutex_enter(&nhop_ire->ire_lock);
3122 		ire2 = nhop_ire->ire_dep_parent;
3123 		if (ire2 != NULL)
3124 			ire_refhold(ire2);
3125 		mutex_exit(&nhop_ire->ire_lock);
3126 		ire_refrele(nhop_ire);
3127 		nhop_ire = ire2;
3128 	}
3129 	if (nhop_ire == NULL)
3130 		return;
3131 
3132 	ASSERT(!(nhop_ire->ire_type & IRE_IF_CLONE));
3133 
3134 	src = ipha->ipha_src;
3135 
3136 	/*
3137 	 * We look at the interface ire for the nexthop,
3138 	 * to see if ipha_src is in the same subnet
3139 	 * as the nexthop.
3140 	 */
3141 	if ((src & nhop_ire->ire_mask) == (nhop & nhop_ire->ire_mask)) {
3142 		/*
3143 		 * The source is directly connected.
3144 		 */
3145 		mp1 = copymsg(mp);
3146 		if (mp1 != NULL) {
3147 			icmp_send_redirect(mp1, nhop, ira);
3148 		}
3149 	}
3150 	ire_refrele(nhop_ire);
3151 }
3152 
3153 /*
3154  * Generate an ICMP redirect message.
3155  */
3156 static void
icmp_send_redirect(mblk_t * mp,ipaddr_t gateway,ip_recv_attr_t * ira)3157 icmp_send_redirect(mblk_t *mp, ipaddr_t gateway, ip_recv_attr_t *ira)
3158 {
3159 	icmph_t	icmph;
3160 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3161 
3162 	mp = icmp_pkt_err_ok(mp, ira);
3163 	if (mp == NULL)
3164 		return;
3165 
3166 	bzero(&icmph, sizeof (icmph_t));
3167 	icmph.icmph_type = ICMP_REDIRECT;
3168 	icmph.icmph_code = 1;
3169 	icmph.icmph_rd_gateway = gateway;
3170 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutRedirects);
3171 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3172 }
3173 
3174 /*
3175  * Generate an ICMP time exceeded message.
3176  */
3177 void
icmp_time_exceeded(mblk_t * mp,uint8_t code,ip_recv_attr_t * ira)3178 icmp_time_exceeded(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3179 {
3180 	icmph_t	icmph;
3181 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3182 
3183 	mp = icmp_pkt_err_ok(mp, ira);
3184 	if (mp == NULL)
3185 		return;
3186 
3187 	bzero(&icmph, sizeof (icmph_t));
3188 	icmph.icmph_type = ICMP_TIME_EXCEEDED;
3189 	icmph.icmph_code = code;
3190 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimeExcds);
3191 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3192 }
3193 
3194 /*
3195  * Generate an ICMP unreachable message.
3196  * When called from ip_output side a minimal ip_recv_attr_t needs to be
3197  * constructed by the caller.
3198  */
3199 void
icmp_unreachable(mblk_t * mp,uint8_t code,ip_recv_attr_t * ira)3200 icmp_unreachable(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3201 {
3202 	icmph_t	icmph;
3203 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3204 
3205 	mp = icmp_pkt_err_ok(mp, ira);
3206 	if (mp == NULL)
3207 		return;
3208 
3209 	bzero(&icmph, sizeof (icmph_t));
3210 	icmph.icmph_type = ICMP_DEST_UNREACHABLE;
3211 	icmph.icmph_code = code;
3212 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
3213 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3214 }
3215 
3216 /*
3217  * Latch in the IPsec state for a stream based the policy in the listener
3218  * and the actions in the ip_recv_attr_t.
3219  * Called directly from TCP and SCTP.
3220  */
3221 boolean_t
ip_ipsec_policy_inherit(conn_t * connp,conn_t * lconnp,ip_recv_attr_t * ira)3222 ip_ipsec_policy_inherit(conn_t *connp, conn_t *lconnp, ip_recv_attr_t *ira)
3223 {
3224 	ASSERT(lconnp->conn_policy != NULL);
3225 	ASSERT(connp->conn_policy == NULL);
3226 
3227 	IPPH_REFHOLD(lconnp->conn_policy);
3228 	connp->conn_policy = lconnp->conn_policy;
3229 
3230 	if (ira->ira_ipsec_action != NULL) {
3231 		if (connp->conn_latch == NULL) {
3232 			connp->conn_latch = iplatch_create();
3233 			if (connp->conn_latch == NULL)
3234 				return (B_FALSE);
3235 		}
3236 		ipsec_latch_inbound(connp, ira);
3237 	}
3238 	return (B_TRUE);
3239 }
3240 
3241 /*
3242  * Verify whether or not the IP address is a valid local address.
3243  * Could be a unicast, including one for a down interface.
3244  * If allow_mcbc then a multicast or broadcast address is also
3245  * acceptable.
3246  *
3247  * In the case of a broadcast/multicast address, however, the
3248  * upper protocol is expected to reset the src address
3249  * to zero when we return IPVL_MCAST/IPVL_BCAST so that
3250  * no packets are emitted with broadcast/multicast address as
3251  * source address (that violates hosts requirements RFC 1122)
3252  * The addresses valid for bind are:
3253  *	(1) - INADDR_ANY (0)
3254  *	(2) - IP address of an UP interface
3255  *	(3) - IP address of a DOWN interface
3256  *	(4) - valid local IP broadcast addresses. In this case
3257  *	the conn will only receive packets destined to
3258  *	the specified broadcast address.
3259  *	(5) - a multicast address. In this case
3260  *	the conn will only receive packets destined to
3261  *	the specified multicast address. Note: the
3262  *	application still has to issue an
3263  *	IP_ADD_MEMBERSHIP socket option.
3264  *
3265  * In all the above cases, the bound address must be valid in the current zone.
3266  * When the address is loopback, multicast or broadcast, there might be many
3267  * matching IREs so bind has to look up based on the zone.
3268  */
3269 ip_laddr_t
ip_laddr_verify_v4(ipaddr_t src_addr,zoneid_t zoneid,ip_stack_t * ipst,boolean_t allow_mcbc)3270 ip_laddr_verify_v4(ipaddr_t src_addr, zoneid_t zoneid,
3271     ip_stack_t *ipst, boolean_t allow_mcbc)
3272 {
3273 	ire_t *src_ire;
3274 
3275 	ASSERT(src_addr != INADDR_ANY);
3276 
3277 	src_ire = ire_ftable_lookup_v4(src_addr, 0, 0, 0,
3278 	    NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, 0, ipst, NULL);
3279 
3280 	/*
3281 	 * If an address other than in6addr_any is requested,
3282 	 * we verify that it is a valid address for bind
3283 	 * Note: Following code is in if-else-if form for
3284 	 * readability compared to a condition check.
3285 	 */
3286 	if (src_ire != NULL && (src_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK))) {
3287 		/*
3288 		 * (2) Bind to address of local UP interface
3289 		 */
3290 		ire_refrele(src_ire);
3291 		return (IPVL_UNICAST_UP);
3292 	} else if (src_ire != NULL && src_ire->ire_type & IRE_BROADCAST) {
3293 		/*
3294 		 * (4) Bind to broadcast address
3295 		 */
3296 		ire_refrele(src_ire);
3297 		if (allow_mcbc)
3298 			return (IPVL_BCAST);
3299 		else
3300 			return (IPVL_BAD);
3301 	} else if (CLASSD(src_addr)) {
3302 		/* (5) bind to multicast address. */
3303 		if (src_ire != NULL)
3304 			ire_refrele(src_ire);
3305 
3306 		if (allow_mcbc)
3307 			return (IPVL_MCAST);
3308 		else
3309 			return (IPVL_BAD);
3310 	} else {
3311 		ipif_t *ipif;
3312 
3313 		/*
3314 		 * (3) Bind to address of local DOWN interface?
3315 		 * (ipif_lookup_addr() looks up all interfaces
3316 		 * but we do not get here for UP interfaces
3317 		 * - case (2) above)
3318 		 */
3319 		if (src_ire != NULL)
3320 			ire_refrele(src_ire);
3321 
3322 		ipif = ipif_lookup_addr(src_addr, NULL, zoneid, ipst);
3323 		if (ipif == NULL)
3324 			return (IPVL_BAD);
3325 
3326 		/* Not a useful source? */
3327 		if (ipif->ipif_flags & (IPIF_NOLOCAL | IPIF_ANYCAST)) {
3328 			ipif_refrele(ipif);
3329 			return (IPVL_BAD);
3330 		}
3331 		ipif_refrele(ipif);
3332 		return (IPVL_UNICAST_DOWN);
3333 	}
3334 }
3335 
3336 /*
3337  * Insert in the bind fanout for IPv4 and IPv6.
3338  * The caller should already have used ip_laddr_verify_v*() before calling
3339  * this.
3340  */
3341 int
ip_laddr_fanout_insert(conn_t * connp)3342 ip_laddr_fanout_insert(conn_t *connp)
3343 {
3344 	int		error;
3345 
3346 	/*
3347 	 * Allow setting new policies. For example, disconnects result
3348 	 * in us being called. As we would have set conn_policy_cached
3349 	 * to B_TRUE before, we should set it to B_FALSE, so that policy
3350 	 * can change after the disconnect.
3351 	 */
3352 	connp->conn_policy_cached = B_FALSE;
3353 
3354 	error = ipcl_bind_insert(connp);
3355 	if (error != 0) {
3356 		if (connp->conn_anon_port) {
3357 			(void) tsol_mlp_anon(crgetzone(connp->conn_cred),
3358 			    connp->conn_mlp_type, connp->conn_proto,
3359 			    ntohs(connp->conn_lport), B_FALSE);
3360 		}
3361 		connp->conn_mlp_type = mlptSingle;
3362 	}
3363 	return (error);
3364 }
3365 
3366 /*
3367  * Verify that both the source and destination addresses are valid. If
3368  * IPDF_VERIFY_DST is not set, then the destination address may be unreachable,
3369  * i.e. have no route to it.  Protocols like TCP want to verify destination
3370  * reachability, while tunnels do not.
3371  *
3372  * Determine the route, the interface, and (optionally) the source address
3373  * to use to reach a given destination.
3374  * Note that we allow connect to broadcast and multicast addresses when
3375  * IPDF_ALLOW_MCBC is set.
3376  * first_hop and dst_addr are normally the same, but if source routing
3377  * they will differ; in that case the first_hop is what we'll use for the
3378  * routing lookup but the dce and label checks will be done on dst_addr,
3379  *
3380  * If uinfo is set, then we fill in the best available information
3381  * we have for the destination. This is based on (in priority order) any
3382  * metrics and path MTU stored in a dce_t, route metrics, and finally the
3383  * ill_mtu/ill_mc_mtu.
3384  *
3385  * Tsol note: If we have a source route then dst_addr != firsthop. But we
3386  * always do the label check on dst_addr.
3387  */
3388 int
ip_set_destination_v4(ipaddr_t * src_addrp,ipaddr_t dst_addr,ipaddr_t firsthop,ip_xmit_attr_t * ixa,iulp_t * uinfo,uint32_t flags,uint_t mac_mode)3389 ip_set_destination_v4(ipaddr_t *src_addrp, ipaddr_t dst_addr, ipaddr_t firsthop,
3390     ip_xmit_attr_t *ixa, iulp_t *uinfo, uint32_t flags, uint_t mac_mode)
3391 {
3392 	ire_t		*ire = NULL;
3393 	int		error = 0;
3394 	ipaddr_t	setsrc;				/* RTF_SETSRC */
3395 	zoneid_t	zoneid = ixa->ixa_zoneid;	/* Honors SO_ALLZONES */
3396 	ip_stack_t	*ipst = ixa->ixa_ipst;
3397 	dce_t		*dce;
3398 	uint_t		pmtu;
3399 	uint_t		generation;
3400 	nce_t		*nce;
3401 	ill_t		*ill = NULL;
3402 	boolean_t	multirt = B_FALSE;
3403 
3404 	ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
3405 
3406 	/*
3407 	 * We never send to zero; the ULPs map it to the loopback address.
3408 	 * We can't allow it since we use zero to mean unitialized in some
3409 	 * places.
3410 	 */
3411 	ASSERT(dst_addr != INADDR_ANY);
3412 
3413 	if (is_system_labeled()) {
3414 		ts_label_t *tsl = NULL;
3415 
3416 		error = tsol_check_dest(ixa->ixa_tsl, &dst_addr, IPV4_VERSION,
3417 		    mac_mode, (flags & IPDF_ZONE_IS_GLOBAL) != 0, &tsl);
3418 		if (error != 0)
3419 			return (error);
3420 		if (tsl != NULL) {
3421 			/* Update the label */
3422 			ip_xmit_attr_replace_tsl(ixa, tsl);
3423 		}
3424 	}
3425 
3426 	setsrc = INADDR_ANY;
3427 	/*
3428 	 * Select a route; For IPMP interfaces, we would only select
3429 	 * a "hidden" route (i.e., going through a specific under_ill)
3430 	 * if ixa_ifindex has been specified.
3431 	 */
3432 	ire = ip_select_route_v4(firsthop, *src_addrp, ixa,
3433 	    &generation, &setsrc, &error, &multirt);
3434 	ASSERT(ire != NULL);	/* IRE_NOROUTE if none found */
3435 	if (error != 0)
3436 		goto bad_addr;
3437 
3438 	/*
3439 	 * ire can't be a broadcast or multicast unless IPDF_ALLOW_MCBC is set.
3440 	 * If IPDF_VERIFY_DST is set, the destination must be reachable;
3441 	 * Otherwise the destination needn't be reachable.
3442 	 *
3443 	 * If we match on a reject or black hole, then we've got a
3444 	 * local failure.  May as well fail out the connect() attempt,
3445 	 * since it's never going to succeed.
3446 	 */
3447 	if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
3448 		/*
3449 		 * If we're verifying destination reachability, we always want
3450 		 * to complain here.
3451 		 *
3452 		 * If we're not verifying destination reachability but the
3453 		 * destination has a route, we still want to fail on the
3454 		 * temporary address and broadcast address tests.
3455 		 *
3456 		 * In both cases do we let the code continue so some reasonable
3457 		 * information is returned to the caller. That enables the
3458 		 * caller to use (and even cache) the IRE. conn_ip_ouput will
3459 		 * use the generation mismatch path to check for the unreachable
3460 		 * case thereby avoiding any specific check in the main path.
3461 		 */
3462 		ASSERT(generation == IRE_GENERATION_VERIFY);
3463 		if (flags & IPDF_VERIFY_DST) {
3464 			/*
3465 			 * Set errno but continue to set up ixa_ire to be
3466 			 * the RTF_REJECT|RTF_BLACKHOLE IRE.
3467 			 * That allows callers to use ip_output to get an
3468 			 * ICMP error back.
3469 			 */
3470 			if (!(ire->ire_type & IRE_HOST))
3471 				error = ENETUNREACH;
3472 			else
3473 				error = EHOSTUNREACH;
3474 		}
3475 	}
3476 
3477 	if ((ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST)) &&
3478 	    !(flags & IPDF_ALLOW_MCBC)) {
3479 		ire_refrele(ire);
3480 		ire = ire_reject(ipst, B_FALSE);
3481 		generation = IRE_GENERATION_VERIFY;
3482 		error = ENETUNREACH;
3483 	}
3484 
3485 	/* Cache things */
3486 	if (ixa->ixa_ire != NULL)
3487 		ire_refrele_notr(ixa->ixa_ire);
3488 #ifdef DEBUG
3489 	ire_refhold_notr(ire);
3490 	ire_refrele(ire);
3491 #endif
3492 	ixa->ixa_ire = ire;
3493 	ixa->ixa_ire_generation = generation;
3494 
3495 	/*
3496 	 * Ensure that ixa_dce is always set any time that ixa_ire is set,
3497 	 * since some callers will send a packet to conn_ip_output() even if
3498 	 * there's an error.
3499 	 */
3500 	if (flags & IPDF_UNIQUE_DCE) {
3501 		/* Fallback to the default dce if allocation fails */
3502 		dce = dce_lookup_and_add_v4(dst_addr, ipst);
3503 		if (dce != NULL)
3504 			generation = dce->dce_generation;
3505 		else
3506 			dce = dce_lookup_v4(dst_addr, ipst, &generation);
3507 	} else {
3508 		dce = dce_lookup_v4(dst_addr, ipst, &generation);
3509 	}
3510 	ASSERT(dce != NULL);
3511 	if (ixa->ixa_dce != NULL)
3512 		dce_refrele_notr(ixa->ixa_dce);
3513 #ifdef DEBUG
3514 	dce_refhold_notr(dce);
3515 	dce_refrele(dce);
3516 #endif
3517 	ixa->ixa_dce = dce;
3518 	ixa->ixa_dce_generation = generation;
3519 
3520 	/*
3521 	 * For multicast with multirt we have a flag passed back from
3522 	 * ire_lookup_multi_ill_v4 since we don't have an IRE for each
3523 	 * possible multicast address.
3524 	 * We also need a flag for multicast since we can't check
3525 	 * whether RTF_MULTIRT is set in ixa_ire for multicast.
3526 	 */
3527 	if (multirt) {
3528 		ixa->ixa_postfragfn = ip_postfrag_multirt_v4;
3529 		ixa->ixa_flags |= IXAF_MULTIRT_MULTICAST;
3530 	} else {
3531 		ixa->ixa_postfragfn = ire->ire_postfragfn;
3532 		ixa->ixa_flags &= ~IXAF_MULTIRT_MULTICAST;
3533 	}
3534 	if (!(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3535 		/* Get an nce to cache. */
3536 		nce = ire_to_nce(ire, firsthop, NULL);
3537 		if (nce == NULL) {
3538 			/* Allocation failure? */
3539 			ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3540 		} else {
3541 			if (ixa->ixa_nce != NULL)
3542 				nce_refrele(ixa->ixa_nce);
3543 			ixa->ixa_nce = nce;
3544 		}
3545 	}
3546 
3547 	/*
3548 	 * If the source address is a loopback address, the
3549 	 * destination had best be local or multicast.
3550 	 * If we are sending to an IRE_LOCAL using a loopback source then
3551 	 * it had better be the same zoneid.
3552 	 */
3553 	if (*src_addrp == htonl(INADDR_LOOPBACK)) {
3554 		if ((ire->ire_type & IRE_LOCAL) && ire->ire_zoneid != zoneid) {
3555 			ire = NULL;	/* Stored in ixa_ire */
3556 			error = EADDRNOTAVAIL;
3557 			goto bad_addr;
3558 		}
3559 		if (!(ire->ire_type & (IRE_LOOPBACK|IRE_LOCAL|IRE_MULTICAST))) {
3560 			ire = NULL;	/* Stored in ixa_ire */
3561 			error = EADDRNOTAVAIL;
3562 			goto bad_addr;
3563 		}
3564 	}
3565 	if (ire->ire_type & IRE_BROADCAST) {
3566 		/*
3567 		 * If the ULP didn't have a specified source, then we
3568 		 * make sure we reselect the source when sending
3569 		 * broadcasts out different interfaces.
3570 		 */
3571 		if (flags & IPDF_SELECT_SRC)
3572 			ixa->ixa_flags |= IXAF_SET_SOURCE;
3573 		else
3574 			ixa->ixa_flags &= ~IXAF_SET_SOURCE;
3575 	}
3576 
3577 	/*
3578 	 * Does the caller want us to pick a source address?
3579 	 */
3580 	if (flags & IPDF_SELECT_SRC) {
3581 		ipaddr_t	src_addr;
3582 
3583 		/*
3584 		 * We use use ire_nexthop_ill to avoid the under ipmp
3585 		 * interface for source address selection. Note that for ipmp
3586 		 * probe packets, ixa_ifindex would have been specified, and
3587 		 * the ip_select_route() invocation would have picked an ire
3588 		 * will ire_ill pointing at an under interface.
3589 		 */
3590 		ill = ire_nexthop_ill(ire);
3591 
3592 		/* If unreachable we have no ill but need some source */
3593 		if (ill == NULL) {
3594 			src_addr = htonl(INADDR_LOOPBACK);
3595 			/* Make sure we look for a better source address */
3596 			generation = SRC_GENERATION_VERIFY;
3597 		} else {
3598 			error = ip_select_source_v4(ill, setsrc, dst_addr,
3599 			    ixa->ixa_multicast_ifaddr, zoneid,
3600 			    ipst, &src_addr, &generation, NULL);
3601 			if (error != 0) {
3602 				ire = NULL;	/* Stored in ixa_ire */
3603 				goto bad_addr;
3604 			}
3605 		}
3606 
3607 		/*
3608 		 * We allow the source address to to down.
3609 		 * However, we check that we don't use the loopback address
3610 		 * as a source when sending out on the wire.
3611 		 */
3612 		if ((src_addr == htonl(INADDR_LOOPBACK)) &&
3613 		    !(ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK|IRE_MULTICAST)) &&
3614 		    !(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3615 			ire = NULL;	/* Stored in ixa_ire */
3616 			error = EADDRNOTAVAIL;
3617 			goto bad_addr;
3618 		}
3619 
3620 		*src_addrp = src_addr;
3621 		ixa->ixa_src_generation = generation;
3622 	}
3623 
3624 	/*
3625 	 * Make sure we don't leave an unreachable ixa_nce in place
3626 	 * since ip_select_route is used when we unplumb i.e., remove
3627 	 * references on ixa_ire, ixa_nce, and ixa_dce.
3628 	 */
3629 	nce = ixa->ixa_nce;
3630 	if (nce != NULL && nce->nce_is_condemned) {
3631 		nce_refrele(nce);
3632 		ixa->ixa_nce = NULL;
3633 		ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3634 	}
3635 
3636 	/*
3637 	 * The caller has set IXAF_PMTU_DISCOVERY if path MTU is desired.
3638 	 * However, we can't do it for IPv4 multicast or broadcast.
3639 	 */
3640 	if (ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST))
3641 		ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3642 
3643 	/*
3644 	 * Set initial value for fragmentation limit. Either conn_ip_output
3645 	 * or ULP might updates it when there are routing changes.
3646 	 * Handles a NULL ixa_ire->ire_ill or a NULL ixa_nce for RTF_REJECT.
3647 	 */
3648 	pmtu = ip_get_pmtu(ixa);
3649 	ixa->ixa_fragsize = pmtu;
3650 	/* Make sure ixa_fragsize and ixa_pmtu remain identical */
3651 	if (ixa->ixa_flags & IXAF_VERIFY_PMTU)
3652 		ixa->ixa_pmtu = pmtu;
3653 
3654 	/*
3655 	 * Extract information useful for some transports.
3656 	 * First we look for DCE metrics. Then we take what we have in
3657 	 * the metrics in the route, where the offlink is used if we have
3658 	 * one.
3659 	 */
3660 	if (uinfo != NULL) {
3661 		bzero(uinfo, sizeof (*uinfo));
3662 
3663 		if (dce->dce_flags & DCEF_UINFO)
3664 			*uinfo = dce->dce_uinfo;
3665 
3666 		rts_merge_metrics(uinfo, &ire->ire_metrics);
3667 
3668 		/* Allow ire_metrics to decrease the path MTU from above */
3669 		if (uinfo->iulp_mtu == 0 || uinfo->iulp_mtu > pmtu)
3670 			uinfo->iulp_mtu = pmtu;
3671 
3672 		uinfo->iulp_localnet = (ire->ire_type & IRE_ONLINK) != 0;
3673 		uinfo->iulp_loopback = (ire->ire_type & IRE_LOOPBACK) != 0;
3674 		uinfo->iulp_local = (ire->ire_type & IRE_LOCAL) != 0;
3675 	}
3676 
3677 	if (ill != NULL)
3678 		ill_refrele(ill);
3679 
3680 	return (error);
3681 
3682 bad_addr:
3683 	if (ire != NULL)
3684 		ire_refrele(ire);
3685 
3686 	if (ill != NULL)
3687 		ill_refrele(ill);
3688 
3689 	/*
3690 	 * Make sure we don't leave an unreachable ixa_nce in place
3691 	 * since ip_select_route is used when we unplumb i.e., remove
3692 	 * references on ixa_ire, ixa_nce, and ixa_dce.
3693 	 */
3694 	nce = ixa->ixa_nce;
3695 	if (nce != NULL && nce->nce_is_condemned) {
3696 		nce_refrele(nce);
3697 		ixa->ixa_nce = NULL;
3698 		ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3699 	}
3700 
3701 	return (error);
3702 }
3703 
3704 
3705 /*
3706  * Get the base MTU for the case when path MTU discovery is not used.
3707  * Takes the MTU of the IRE into account.
3708  */
3709 uint_t
ip_get_base_mtu(ill_t * ill,ire_t * ire)3710 ip_get_base_mtu(ill_t *ill, ire_t *ire)
3711 {
3712 	uint_t mtu;
3713 	uint_t iremtu = ire->ire_metrics.iulp_mtu;
3714 
3715 	if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST))
3716 		mtu = ill->ill_mc_mtu;
3717 	else
3718 		mtu = ill->ill_mtu;
3719 
3720 	if (iremtu != 0 && iremtu < mtu)
3721 		mtu = iremtu;
3722 
3723 	return (mtu);
3724 }
3725 
3726 /*
3727  * Get the PMTU for the attributes. Handles both IPv4 and IPv6.
3728  * Assumes that ixa_ire, dce, and nce have already been set up.
3729  *
3730  * The caller has set IXAF_PMTU_DISCOVERY if path MTU discovery is desired.
3731  * We avoid path MTU discovery if it is disabled with ndd.
3732  * Furtermore, if the path MTU is too small, then we don't set DF for IPv4.
3733  *
3734  * NOTE: We also used to turn it off for source routed packets. That
3735  * is no longer required since the dce is per final destination.
3736  */
3737 uint_t
ip_get_pmtu(ip_xmit_attr_t * ixa)3738 ip_get_pmtu(ip_xmit_attr_t *ixa)
3739 {
3740 	ip_stack_t	*ipst = ixa->ixa_ipst;
3741 	dce_t		*dce;
3742 	nce_t		*nce;
3743 	ire_t		*ire;
3744 	uint_t		pmtu;
3745 
3746 	ire = ixa->ixa_ire;
3747 	dce = ixa->ixa_dce;
3748 	nce = ixa->ixa_nce;
3749 
3750 	/*
3751 	 * If path MTU discovery has been turned off by ndd, then we ignore
3752 	 * any dce_pmtu and for IPv4 we will not set DF.
3753 	 */
3754 	if (!ipst->ips_ip_path_mtu_discovery)
3755 		ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3756 
3757 	pmtu = IP_MAXPACKET;
3758 	/*
3759 	 * We need to determine if it is acceptable to set DF for IPv4 or not
3760 	 * and for IPv6 if we need to use the minimum MTU. If a connection has
3761 	 * opted into path MTU discovery, then we can use 'DF' in IPv4 and do
3762 	 * not have to constrain ourselves to the IPv6 minimum MTU. There is a
3763 	 * second consideration here: IXAF_DONTFRAG. This is set as a result of
3764 	 * someone setting the IP_DONTFRAG or IPV6_DONTFRAG socket option. In
3765 	 * such a case, it is acceptable to set DF for IPv4 and to use a larger
3766 	 * MTU. Note, the actual MTU is constrained by the ill_t later on in
3767 	 * this function.
3768 	 */
3769 	if (ixa->ixa_flags & (IXAF_PMTU_DISCOVERY | IXAF_DONTFRAG)) {
3770 		ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3771 	} else {
3772 		ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3773 		if (!(ixa->ixa_flags & IXAF_IS_IPV4))
3774 			pmtu = IPV6_MIN_MTU;
3775 	}
3776 
3777 	/* Check if the PMTU is to old before we use it */
3778 	if ((dce->dce_flags & DCEF_PMTU) &&
3779 	    TICK_TO_SEC(ddi_get_lbolt64()) - dce->dce_last_change_time >
3780 	    ipst->ips_ip_pathmtu_interval) {
3781 		/*
3782 		 * Older than 20 minutes. Drop the path MTU information.
3783 		 */
3784 		mutex_enter(&dce->dce_lock);
3785 		dce->dce_flags &= ~(DCEF_PMTU|DCEF_TOO_SMALL_PMTU);
3786 		dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
3787 		mutex_exit(&dce->dce_lock);
3788 		dce_increment_generation(dce);
3789 	}
3790 
3791 	/* The metrics on the route can lower the path MTU */
3792 	if (ire->ire_metrics.iulp_mtu != 0 &&
3793 	    ire->ire_metrics.iulp_mtu < pmtu)
3794 		pmtu = ire->ire_metrics.iulp_mtu;
3795 
3796 	/*
3797 	 * If the path MTU is smaller than some minimum, we still use dce_pmtu
3798 	 * above (would be 576 for IPv4 and 1280 for IPv6), but we clear
3799 	 * IXAF_PMTU_IPV4_DF so that we avoid setting DF for IPv4.
3800 	 */
3801 	if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3802 		if (dce->dce_flags & DCEF_PMTU) {
3803 			if (dce->dce_pmtu < pmtu)
3804 				pmtu = dce->dce_pmtu;
3805 
3806 			if (dce->dce_flags & DCEF_TOO_SMALL_PMTU) {
3807 				ixa->ixa_flags |= IXAF_PMTU_TOO_SMALL;
3808 				ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3809 			} else {
3810 				ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3811 				ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3812 			}
3813 		} else {
3814 			ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3815 			ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3816 		}
3817 	}
3818 
3819 	/*
3820 	 * If we have an IRE_LOCAL we use the loopback mtu instead of
3821 	 * the ill for going out the wire i.e., IRE_LOCAL gets the same
3822 	 * mtu as IRE_LOOPBACK.
3823 	 */
3824 	if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) {
3825 		uint_t loopback_mtu;
3826 
3827 		loopback_mtu = (ire->ire_ipversion == IPV6_VERSION) ?
3828 		    ip_loopback_mtu_v6plus : ip_loopback_mtuplus;
3829 
3830 		if (loopback_mtu < pmtu)
3831 			pmtu = loopback_mtu;
3832 	} else if (nce != NULL) {
3833 		/*
3834 		 * Make sure we don't exceed the interface MTU.
3835 		 * In the case of RTF_REJECT or RTF_BLACKHOLE we might not have
3836 		 * an ill. We'd use the above IP_MAXPACKET in that case just
3837 		 * to tell the transport something larger than zero.
3838 		 */
3839 		if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST)) {
3840 			if (nce->nce_common->ncec_ill->ill_mc_mtu < pmtu)
3841 				pmtu = nce->nce_common->ncec_ill->ill_mc_mtu;
3842 			if (nce->nce_common->ncec_ill != nce->nce_ill &&
3843 			    nce->nce_ill->ill_mc_mtu < pmtu) {
3844 				/*
3845 				 * for interfaces in an IPMP group, the mtu of
3846 				 * the nce_ill (under_ill) could be different
3847 				 * from the mtu of the ncec_ill, so we take the
3848 				 * min of the two.
3849 				 */
3850 				pmtu = nce->nce_ill->ill_mc_mtu;
3851 			}
3852 		} else {
3853 			if (nce->nce_common->ncec_ill->ill_mtu < pmtu)
3854 				pmtu = nce->nce_common->ncec_ill->ill_mtu;
3855 			if (nce->nce_common->ncec_ill != nce->nce_ill &&
3856 			    nce->nce_ill->ill_mtu < pmtu) {
3857 				/*
3858 				 * for interfaces in an IPMP group, the mtu of
3859 				 * the nce_ill (under_ill) could be different
3860 				 * from the mtu of the ncec_ill, so we take the
3861 				 * min of the two.
3862 				 */
3863 				pmtu = nce->nce_ill->ill_mtu;
3864 			}
3865 		}
3866 	}
3867 
3868 	/*
3869 	 * Handle the IPV6_USE_MIN_MTU socket option or ancillary data.
3870 	 * Only applies to IPv6.
3871 	 */
3872 	if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3873 		if (ixa->ixa_flags & IXAF_USE_MIN_MTU) {
3874 			switch (ixa->ixa_use_min_mtu) {
3875 			case IPV6_USE_MIN_MTU_MULTICAST:
3876 				if (ire->ire_type & IRE_MULTICAST)
3877 					pmtu = IPV6_MIN_MTU;
3878 				break;
3879 			case IPV6_USE_MIN_MTU_ALWAYS:
3880 				pmtu = IPV6_MIN_MTU;
3881 				break;
3882 			case IPV6_USE_MIN_MTU_NEVER:
3883 				break;
3884 			}
3885 		} else {
3886 			/* Default is IPV6_USE_MIN_MTU_MULTICAST */
3887 			if (ire->ire_type & IRE_MULTICAST)
3888 				pmtu = IPV6_MIN_MTU;
3889 		}
3890 	}
3891 
3892 	/*
3893 	 * For multirouted IPv6 packets, the IP layer will insert a 8-byte
3894 	 * fragment header in every packet. We compensate for those cases by
3895 	 * returning a smaller path MTU to the ULP.
3896 	 *
3897 	 * In the case of CGTP then ip_output will add a fragment header.
3898 	 * Make sure there is room for it by telling a smaller number
3899 	 * to the transport.
3900 	 *
3901 	 * When IXAF_IPV6_ADDR_FRAGHDR we subtract the frag hdr here
3902 	 * so the ULPs consistently see a iulp_pmtu and ip_get_pmtu()
3903 	 * which is the size of the packets it can send.
3904 	 */
3905 	if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3906 		if ((ire->ire_flags & RTF_MULTIRT) ||
3907 		    (ixa->ixa_flags & IXAF_MULTIRT_MULTICAST)) {
3908 			pmtu -= sizeof (ip6_frag_t);
3909 			ixa->ixa_flags |= IXAF_IPV6_ADD_FRAGHDR;
3910 		}
3911 	}
3912 
3913 	return (pmtu);
3914 }
3915 
3916 /*
3917  * Carve "len" bytes out of an mblk chain, consuming any we empty, and duping
3918  * the final piece where we don't.  Return a pointer to the first mblk in the
3919  * result, and update the pointer to the next mblk to chew on.  If anything
3920  * goes wrong (i.e., dupb fails), we waste everything in sight and return a
3921  * NULL pointer.
3922  */
3923 mblk_t *
ip_carve_mp(mblk_t ** mpp,ssize_t len)3924 ip_carve_mp(mblk_t **mpp, ssize_t len)
3925 {
3926 	mblk_t	*mp0;
3927 	mblk_t	*mp1;
3928 	mblk_t	*mp2;
3929 
3930 	if (!len || !mpp || !(mp0 = *mpp))
3931 		return (NULL);
3932 	/* If we aren't going to consume the first mblk, we need a dup. */
3933 	if (mp0->b_wptr - mp0->b_rptr > len) {
3934 		mp1 = dupb(mp0);
3935 		if (mp1) {
3936 			/* Partition the data between the two mblks. */
3937 			mp1->b_wptr = mp1->b_rptr + len;
3938 			mp0->b_rptr = mp1->b_wptr;
3939 			/*
3940 			 * after adjustments if mblk not consumed is now
3941 			 * unaligned, try to align it. If this fails free
3942 			 * all messages and let upper layer recover.
3943 			 */
3944 			if (!OK_32PTR(mp0->b_rptr)) {
3945 				if (!pullupmsg(mp0, -1)) {
3946 					freemsg(mp0);
3947 					freemsg(mp1);
3948 					*mpp = NULL;
3949 					return (NULL);
3950 				}
3951 			}
3952 		}
3953 		return (mp1);
3954 	}
3955 	/* Eat through as many mblks as we need to get len bytes. */
3956 	len -= mp0->b_wptr - mp0->b_rptr;
3957 	for (mp2 = mp1 = mp0; (mp2 = mp2->b_cont) != 0 && len; mp1 = mp2) {
3958 		if (mp2->b_wptr - mp2->b_rptr > len) {
3959 			/*
3960 			 * We won't consume the entire last mblk.  Like
3961 			 * above, dup and partition it.
3962 			 */
3963 			mp1->b_cont = dupb(mp2);
3964 			mp1 = mp1->b_cont;
3965 			if (!mp1) {
3966 				/*
3967 				 * Trouble.  Rather than go to a lot of
3968 				 * trouble to clean up, we free the messages.
3969 				 * This won't be any worse than losing it on
3970 				 * the wire.
3971 				 */
3972 				freemsg(mp0);
3973 				freemsg(mp2);
3974 				*mpp = NULL;
3975 				return (NULL);
3976 			}
3977 			mp1->b_wptr = mp1->b_rptr + len;
3978 			mp2->b_rptr = mp1->b_wptr;
3979 			/*
3980 			 * after adjustments if mblk not consumed is now
3981 			 * unaligned, try to align it. If this fails free
3982 			 * all messages and let upper layer recover.
3983 			 */
3984 			if (!OK_32PTR(mp2->b_rptr)) {
3985 				if (!pullupmsg(mp2, -1)) {
3986 					freemsg(mp0);
3987 					freemsg(mp2);
3988 					*mpp = NULL;
3989 					return (NULL);
3990 				}
3991 			}
3992 			*mpp = mp2;
3993 			return (mp0);
3994 		}
3995 		/* Decrement len by the amount we just got. */
3996 		len -= mp2->b_wptr - mp2->b_rptr;
3997 	}
3998 	/*
3999 	 * len should be reduced to zero now.  If not our caller has
4000 	 * screwed up.
4001 	 */
4002 	if (len) {
4003 		/* Shouldn't happen! */
4004 		freemsg(mp0);
4005 		*mpp = NULL;
4006 		return (NULL);
4007 	}
4008 	/*
4009 	 * We consumed up to exactly the end of an mblk.  Detach the part
4010 	 * we are returning from the rest of the chain.
4011 	 */
4012 	mp1->b_cont = NULL;
4013 	*mpp = mp2;
4014 	return (mp0);
4015 }
4016 
4017 /* The ill stream is being unplumbed. Called from ip_close */
4018 int
ip_modclose(ill_t * ill)4019 ip_modclose(ill_t *ill)
4020 {
4021 	boolean_t success;
4022 	ipsq_t	*ipsq;
4023 	ipif_t	*ipif;
4024 	queue_t	*q = ill->ill_rq;
4025 	ip_stack_t	*ipst = ill->ill_ipst;
4026 	int	i;
4027 	arl_ill_common_t *ai = ill->ill_common;
4028 
4029 	/*
4030 	 * The punlink prior to this may have initiated a capability
4031 	 * negotiation. But ipsq_enter will block until that finishes or
4032 	 * times out.
4033 	 */
4034 	success = ipsq_enter(ill, B_FALSE, NEW_OP);
4035 
4036 	/*
4037 	 * Open/close/push/pop is guaranteed to be single threaded
4038 	 * per stream by STREAMS. FS guarantees that all references
4039 	 * from top are gone before close is called. So there can't
4040 	 * be another close thread that has set CONDEMNED on this ill.
4041 	 * and cause ipsq_enter to return failure.
4042 	 */
4043 	ASSERT(success);
4044 	ipsq = ill->ill_phyint->phyint_ipsq;
4045 
4046 	/*
4047 	 * Mark it condemned. No new reference will be made to this ill.
4048 	 * Lookup functions will return an error. Threads that try to
4049 	 * increment the refcnt must check for ILL_CAN_LOOKUP. This ensures
4050 	 * that the refcnt will drop down to zero.
4051 	 */
4052 	mutex_enter(&ill->ill_lock);
4053 	ill->ill_state_flags |= ILL_CONDEMNED;
4054 	for (ipif = ill->ill_ipif; ipif != NULL;
4055 	    ipif = ipif->ipif_next) {
4056 		ipif->ipif_state_flags |= IPIF_CONDEMNED;
4057 	}
4058 	/*
4059 	 * Wake up anybody waiting to enter the ipsq. ipsq_enter
4060 	 * returns  error if ILL_CONDEMNED is set
4061 	 */
4062 	cv_broadcast(&ill->ill_cv);
4063 	mutex_exit(&ill->ill_lock);
4064 
4065 	/*
4066 	 * Send all the deferred DLPI messages downstream which came in
4067 	 * during the small window right before ipsq_enter(). We do this
4068 	 * without waiting for the ACKs because all the ACKs for M_PROTO
4069 	 * messages are ignored in ip_rput() when ILL_CONDEMNED is set.
4070 	 */
4071 	ill_dlpi_send_deferred(ill);
4072 
4073 	/*
4074 	 * Shut down fragmentation reassembly.
4075 	 * ill_frag_timer won't start a timer again.
4076 	 * Now cancel any existing timer
4077 	 */
4078 	(void) untimeout(ill->ill_frag_timer_id);
4079 	(void) ill_frag_timeout(ill, 0);
4080 
4081 	/*
4082 	 * Call ill_delete to bring down the ipifs, ilms and ill on
4083 	 * this ill. Then wait for the refcnts to drop to zero.
4084 	 * ill_is_freeable checks whether the ill is really quiescent.
4085 	 * Then make sure that threads that are waiting to enter the
4086 	 * ipsq have seen the error returned by ipsq_enter and have
4087 	 * gone away. Then we call ill_delete_tail which does the
4088 	 * DL_UNBIND_REQ with the driver and then qprocsoff.
4089 	 */
4090 	ill_delete(ill);
4091 	mutex_enter(&ill->ill_lock);
4092 	while (!ill_is_freeable(ill))
4093 		cv_wait(&ill->ill_cv, &ill->ill_lock);
4094 
4095 	while (ill->ill_waiters)
4096 		cv_wait(&ill->ill_cv, &ill->ill_lock);
4097 
4098 	mutex_exit(&ill->ill_lock);
4099 
4100 	/*
4101 	 * ill_delete_tail drops reference on ill_ipst, but we need to keep
4102 	 * it held until the end of the function since the cleanup
4103 	 * below needs to be able to use the ip_stack_t.
4104 	 */
4105 	netstack_hold(ipst->ips_netstack);
4106 
4107 	/* qprocsoff is done via ill_delete_tail */
4108 	ill_delete_tail(ill);
4109 	/*
4110 	 * synchronously wait for arp stream to unbind. After this, we
4111 	 * cannot get any data packets up from the driver.
4112 	 */
4113 	arp_unbind_complete(ill);
4114 	ASSERT(ill->ill_ipst == NULL);
4115 
4116 	/*
4117 	 * Walk through all conns and qenable those that have queued data.
4118 	 * Close synchronization needs this to
4119 	 * be done to ensure that all upper layers blocked
4120 	 * due to flow control to the closing device
4121 	 * get unblocked.
4122 	 */
4123 	ip1dbg(("ip_wsrv: walking\n"));
4124 	for (i = 0; i < TX_FANOUT_SIZE; i++) {
4125 		conn_walk_drain(ipst, &ipst->ips_idl_tx_list[i]);
4126 	}
4127 
4128 	/*
4129 	 * ai can be null if this is an IPv6 ill, or if the IPv4
4130 	 * stream is being torn down before ARP was plumbed (e.g.,
4131 	 * /sbin/ifconfig plumbing a stream twice, and encountering
4132 	 * an error
4133 	 */
4134 	if (ai != NULL) {
4135 		ASSERT(!ill->ill_isv6);
4136 		mutex_enter(&ai->ai_lock);
4137 		ai->ai_ill = NULL;
4138 		if (ai->ai_arl == NULL) {
4139 			mutex_destroy(&ai->ai_lock);
4140 			kmem_free(ai, sizeof (*ai));
4141 		} else {
4142 			cv_signal(&ai->ai_ill_unplumb_done);
4143 			mutex_exit(&ai->ai_lock);
4144 		}
4145 	}
4146 
4147 	mutex_enter(&ipst->ips_ip_mi_lock);
4148 	mi_close_unlink(&ipst->ips_ip_g_head, (IDP)ill);
4149 	mutex_exit(&ipst->ips_ip_mi_lock);
4150 
4151 	/*
4152 	 * credp could be null if the open didn't succeed and ip_modopen
4153 	 * itself calls ip_close.
4154 	 */
4155 	if (ill->ill_credp != NULL)
4156 		crfree(ill->ill_credp);
4157 
4158 	mutex_destroy(&ill->ill_saved_ire_lock);
4159 	mutex_destroy(&ill->ill_lock);
4160 	rw_destroy(&ill->ill_mcast_lock);
4161 	mutex_destroy(&ill->ill_mcast_serializer);
4162 	list_destroy(&ill->ill_nce);
4163 
4164 	/*
4165 	 * Now we are done with the module close pieces that
4166 	 * need the netstack_t.
4167 	 */
4168 	netstack_rele(ipst->ips_netstack);
4169 
4170 	mi_close_free((IDP)ill);
4171 	q->q_ptr = WR(q)->q_ptr = NULL;
4172 
4173 	ipsq_exit(ipsq);
4174 
4175 	return (0);
4176 }
4177 
4178 /*
4179  * This is called as part of close() for IP, UDP, ICMP, and RTS
4180  * in order to quiesce the conn.
4181  */
4182 void
ip_quiesce_conn(conn_t * connp)4183 ip_quiesce_conn(conn_t *connp)
4184 {
4185 	boolean_t	drain_cleanup_reqd = B_FALSE;
4186 	boolean_t	conn_ioctl_cleanup_reqd = B_FALSE;
4187 	boolean_t	ilg_cleanup_reqd = B_FALSE;
4188 	ip_stack_t	*ipst;
4189 
4190 	ASSERT(!IPCL_IS_TCP(connp));
4191 	ipst = connp->conn_netstack->netstack_ip;
4192 
4193 	/*
4194 	 * Mark the conn as closing, and this conn must not be
4195 	 * inserted in future into any list. Eg. conn_drain_insert(),
4196 	 * won't insert this conn into the conn_drain_list.
4197 	 *
4198 	 * conn_idl, and conn_ilg cannot get set henceforth.
4199 	 */
4200 	mutex_enter(&connp->conn_lock);
4201 	ASSERT(!(connp->conn_state_flags & CONN_QUIESCED));
4202 	connp->conn_state_flags |= CONN_CLOSING;
4203 	if (connp->conn_idl != NULL)
4204 		drain_cleanup_reqd = B_TRUE;
4205 	if (connp->conn_oper_pending_ill != NULL)
4206 		conn_ioctl_cleanup_reqd = B_TRUE;
4207 	if (connp->conn_dhcpinit_ill != NULL) {
4208 		ASSERT(connp->conn_dhcpinit_ill->ill_dhcpinit != 0);
4209 		atomic_dec_32(&connp->conn_dhcpinit_ill->ill_dhcpinit);
4210 		ill_set_inputfn(connp->conn_dhcpinit_ill);
4211 		connp->conn_dhcpinit_ill = NULL;
4212 	}
4213 	if (connp->conn_ilg != NULL)
4214 		ilg_cleanup_reqd = B_TRUE;
4215 	mutex_exit(&connp->conn_lock);
4216 
4217 	if (conn_ioctl_cleanup_reqd)
4218 		conn_ioctl_cleanup(connp);
4219 
4220 	if (is_system_labeled() && connp->conn_anon_port) {
4221 		(void) tsol_mlp_anon(crgetzone(connp->conn_cred),
4222 		    connp->conn_mlp_type, connp->conn_proto,
4223 		    ntohs(connp->conn_lport), B_FALSE);
4224 		connp->conn_anon_port = 0;
4225 	}
4226 	connp->conn_mlp_type = mlptSingle;
4227 
4228 	/*
4229 	 * Remove this conn from any fanout list it is on.
4230 	 * and then wait for any threads currently operating
4231 	 * on this endpoint to finish
4232 	 */
4233 	ipcl_hash_remove(connp);
4234 
4235 	/*
4236 	 * Remove this conn from the drain list, and do any other cleanup that
4237 	 * may be required.  (TCP conns are never flow controlled, and
4238 	 * conn_idl will be NULL.)
4239 	 */
4240 	if (drain_cleanup_reqd && connp->conn_idl != NULL) {
4241 		idl_t *idl = connp->conn_idl;
4242 
4243 		mutex_enter(&idl->idl_lock);
4244 		conn_drain(connp, B_TRUE);
4245 		mutex_exit(&idl->idl_lock);
4246 	}
4247 
4248 	if (connp == ipst->ips_ip_g_mrouter)
4249 		(void) ip_mrouter_done(ipst);
4250 
4251 	if (ilg_cleanup_reqd)
4252 		ilg_delete_all(connp);
4253 
4254 	/*
4255 	 * Now conn refcnt can increase only thru CONN_INC_REF_LOCKED.
4256 	 * callers from write side can't be there now because close
4257 	 * is in progress. The only other caller is ipcl_walk
4258 	 * which checks for the condemned flag.
4259 	 */
4260 	mutex_enter(&connp->conn_lock);
4261 	connp->conn_state_flags |= CONN_CONDEMNED;
4262 	while (connp->conn_ref != 1)
4263 		cv_wait(&connp->conn_cv, &connp->conn_lock);
4264 	connp->conn_state_flags |= CONN_QUIESCED;
4265 	mutex_exit(&connp->conn_lock);
4266 }
4267 
4268 /* ARGSUSED */
4269 int
ip_close(queue_t * q,int flags,cred_t * credp __unused)4270 ip_close(queue_t *q, int flags, cred_t *credp __unused)
4271 {
4272 	conn_t		*connp;
4273 
4274 	/*
4275 	 * Call the appropriate delete routine depending on whether this is
4276 	 * a module or device.
4277 	 */
4278 	if (WR(q)->q_next != NULL) {
4279 		/* This is a module close */
4280 		return (ip_modclose((ill_t *)q->q_ptr));
4281 	}
4282 
4283 	connp = q->q_ptr;
4284 	ip_quiesce_conn(connp);
4285 
4286 	qprocsoff(q);
4287 
4288 	/*
4289 	 * Now we are truly single threaded on this stream, and can
4290 	 * delete the things hanging off the connp, and finally the connp.
4291 	 * We removed this connp from the fanout list, it cannot be
4292 	 * accessed thru the fanouts, and we already waited for the
4293 	 * conn_ref to drop to 0. We are already in close, so
4294 	 * there cannot be any other thread from the top. qprocsoff
4295 	 * has completed, and service has completed or won't run in
4296 	 * future.
4297 	 */
4298 	ASSERT(connp->conn_ref == 1);
4299 
4300 	inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4301 
4302 	connp->conn_ref--;
4303 	ipcl_conn_destroy(connp);
4304 
4305 	q->q_ptr = WR(q)->q_ptr = NULL;
4306 	return (0);
4307 }
4308 
4309 /*
4310  * Wapper around putnext() so that ip_rts_request can merely use
4311  * conn_recv.
4312  */
4313 /*ARGSUSED2*/
4314 static void
ip_conn_input(void * arg1,mblk_t * mp,void * arg2,ip_recv_attr_t * ira)4315 ip_conn_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4316 {
4317 	conn_t *connp = (conn_t *)arg1;
4318 
4319 	putnext(connp->conn_rq, mp);
4320 }
4321 
4322 /* Dummy in case ICMP error delivery is attempted to a /dev/ip instance */
4323 /* ARGSUSED */
4324 static void
ip_conn_input_icmp(void * arg1,mblk_t * mp,void * arg2,ip_recv_attr_t * ira)4325 ip_conn_input_icmp(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4326 {
4327 	freemsg(mp);
4328 }
4329 
4330 /*
4331  * Called when the module is about to be unloaded
4332  */
4333 void
ip_ddi_destroy(void)4334 ip_ddi_destroy(void)
4335 {
4336 	/* This needs to be called before destroying any transports. */
4337 	mutex_enter(&cpu_lock);
4338 	unregister_cpu_setup_func(ip_tp_cpu_update, NULL);
4339 	mutex_exit(&cpu_lock);
4340 
4341 	tnet_fini();
4342 
4343 	icmp_ddi_g_destroy();
4344 	rts_ddi_g_destroy();
4345 	udp_ddi_g_destroy();
4346 	sctp_ddi_g_destroy();
4347 	tcp_ddi_g_destroy();
4348 	ilb_ddi_g_destroy();
4349 	dce_g_destroy();
4350 	ipsec_policy_g_destroy();
4351 	ipcl_g_destroy();
4352 	ip_net_g_destroy();
4353 	ip_ire_g_fini();
4354 	inet_minor_destroy(ip_minor_arena_sa);
4355 #if defined(_LP64)
4356 	inet_minor_destroy(ip_minor_arena_la);
4357 #endif
4358 
4359 #ifdef DEBUG
4360 	list_destroy(&ip_thread_list);
4361 	rw_destroy(&ip_thread_rwlock);
4362 	tsd_destroy(&ip_thread_data);
4363 #endif
4364 
4365 	netstack_unregister(NS_IP);
4366 }
4367 
4368 /*
4369  * First step in cleanup.
4370  */
4371 /* ARGSUSED */
4372 static void
ip_stack_shutdown(netstackid_t stackid,void * arg)4373 ip_stack_shutdown(netstackid_t stackid, void *arg)
4374 {
4375 	ip_stack_t *ipst = (ip_stack_t *)arg;
4376 	kt_did_t ktid;
4377 
4378 #ifdef NS_DEBUG
4379 	printf("ip_stack_shutdown(%p, stack %d)\n", (void *)ipst, stackid);
4380 #endif
4381 
4382 	/*
4383 	 * Perform cleanup for special interfaces (loopback and IPMP).
4384 	 */
4385 	ip_interface_cleanup(ipst);
4386 
4387 	/*
4388 	 * The *_hook_shutdown()s start the process of notifying any
4389 	 * consumers that things are going away.... nothing is destroyed.
4390 	 */
4391 	ipv4_hook_shutdown(ipst);
4392 	ipv6_hook_shutdown(ipst);
4393 	arp_hook_shutdown(ipst);
4394 
4395 	mutex_enter(&ipst->ips_capab_taskq_lock);
4396 	ktid = ipst->ips_capab_taskq_thread->t_did;
4397 	ipst->ips_capab_taskq_quit = B_TRUE;
4398 	cv_signal(&ipst->ips_capab_taskq_cv);
4399 	mutex_exit(&ipst->ips_capab_taskq_lock);
4400 
4401 	/*
4402 	 * In rare occurrences, particularly on virtual hardware where CPUs can
4403 	 * be de-scheduled, the thread that we just signaled will not run until
4404 	 * after we have gotten through parts of ip_stack_fini. If that happens
4405 	 * then we'll try to grab the ips_capab_taskq_lock as part of returning
4406 	 * from cv_wait which no longer exists.
4407 	 */
4408 	thread_join(ktid);
4409 }
4410 
4411 /*
4412  * Free the IP stack instance.
4413  */
4414 static void
ip_stack_fini(netstackid_t stackid,void * arg)4415 ip_stack_fini(netstackid_t stackid, void *arg)
4416 {
4417 	ip_stack_t *ipst = (ip_stack_t *)arg;
4418 	int ret;
4419 
4420 #ifdef NS_DEBUG
4421 	printf("ip_stack_fini(%p, stack %d)\n", (void *)ipst, stackid);
4422 #endif
4423 	/*
4424 	 * At this point, all of the notifications that the events and
4425 	 * protocols are going away have been run, meaning that we can
4426 	 * now set about starting to clean things up.
4427 	 */
4428 	ipobs_fini(ipst);
4429 	ipv4_hook_destroy(ipst);
4430 	ipv6_hook_destroy(ipst);
4431 	arp_hook_destroy(ipst);
4432 	ip_net_destroy(ipst);
4433 
4434 	ipmp_destroy(ipst);
4435 
4436 	ip_kstat_fini(stackid, ipst->ips_ip_mibkp);
4437 	ipst->ips_ip_mibkp = NULL;
4438 	icmp_kstat_fini(stackid, ipst->ips_icmp_mibkp);
4439 	ipst->ips_icmp_mibkp = NULL;
4440 	ip_kstat2_fini(stackid, ipst->ips_ip_kstat);
4441 	ipst->ips_ip_kstat = NULL;
4442 	bzero(&ipst->ips_ip_statistics, sizeof (ipst->ips_ip_statistics));
4443 	ip6_kstat_fini(stackid, ipst->ips_ip6_kstat);
4444 	ipst->ips_ip6_kstat = NULL;
4445 	bzero(&ipst->ips_ip6_statistics, sizeof (ipst->ips_ip6_statistics));
4446 
4447 	kmem_free(ipst->ips_propinfo_tbl,
4448 	    ip_propinfo_count * sizeof (mod_prop_info_t));
4449 	ipst->ips_propinfo_tbl = NULL;
4450 
4451 	dce_stack_destroy(ipst);
4452 	ip_mrouter_stack_destroy(ipst);
4453 
4454 	/*
4455 	 * Quiesce all of our timers. Note we set the quiesce flags before we
4456 	 * call untimeout. The slowtimers may actually kick off another instance
4457 	 * of the non-slow timers.
4458 	 */
4459 	mutex_enter(&ipst->ips_igmp_timer_lock);
4460 	ipst->ips_igmp_timer_quiesce = B_TRUE;
4461 	mutex_exit(&ipst->ips_igmp_timer_lock);
4462 
4463 	mutex_enter(&ipst->ips_mld_timer_lock);
4464 	ipst->ips_mld_timer_quiesce = B_TRUE;
4465 	mutex_exit(&ipst->ips_mld_timer_lock);
4466 
4467 	mutex_enter(&ipst->ips_igmp_slowtimeout_lock);
4468 	ipst->ips_igmp_slowtimeout_quiesce = B_TRUE;
4469 	mutex_exit(&ipst->ips_igmp_slowtimeout_lock);
4470 
4471 	mutex_enter(&ipst->ips_mld_slowtimeout_lock);
4472 	ipst->ips_mld_slowtimeout_quiesce = B_TRUE;
4473 	mutex_exit(&ipst->ips_mld_slowtimeout_lock);
4474 
4475 	ret = untimeout(ipst->ips_igmp_timeout_id);
4476 	if (ret == -1) {
4477 		ASSERT(ipst->ips_igmp_timeout_id == 0);
4478 	} else {
4479 		ASSERT(ipst->ips_igmp_timeout_id != 0);
4480 		ipst->ips_igmp_timeout_id = 0;
4481 	}
4482 	ret = untimeout(ipst->ips_igmp_slowtimeout_id);
4483 	if (ret == -1) {
4484 		ASSERT(ipst->ips_igmp_slowtimeout_id == 0);
4485 	} else {
4486 		ASSERT(ipst->ips_igmp_slowtimeout_id != 0);
4487 		ipst->ips_igmp_slowtimeout_id = 0;
4488 	}
4489 	ret = untimeout(ipst->ips_mld_timeout_id);
4490 	if (ret == -1) {
4491 		ASSERT(ipst->ips_mld_timeout_id == 0);
4492 	} else {
4493 		ASSERT(ipst->ips_mld_timeout_id != 0);
4494 		ipst->ips_mld_timeout_id = 0;
4495 	}
4496 	ret = untimeout(ipst->ips_mld_slowtimeout_id);
4497 	if (ret == -1) {
4498 		ASSERT(ipst->ips_mld_slowtimeout_id == 0);
4499 	} else {
4500 		ASSERT(ipst->ips_mld_slowtimeout_id != 0);
4501 		ipst->ips_mld_slowtimeout_id = 0;
4502 	}
4503 
4504 	ip_ire_fini(ipst);
4505 	ip6_asp_free(ipst);
4506 	conn_drain_fini(ipst);
4507 	ipcl_destroy(ipst);
4508 
4509 	mutex_destroy(&ipst->ips_ndp4->ndp_g_lock);
4510 	mutex_destroy(&ipst->ips_ndp6->ndp_g_lock);
4511 	kmem_free(ipst->ips_ndp4, sizeof (ndp_g_t));
4512 	ipst->ips_ndp4 = NULL;
4513 	kmem_free(ipst->ips_ndp6, sizeof (ndp_g_t));
4514 	ipst->ips_ndp6 = NULL;
4515 
4516 	if (ipst->ips_loopback_ksp != NULL) {
4517 		kstat_delete_netstack(ipst->ips_loopback_ksp, stackid);
4518 		ipst->ips_loopback_ksp = NULL;
4519 	}
4520 
4521 	mutex_destroy(&ipst->ips_capab_taskq_lock);
4522 	cv_destroy(&ipst->ips_capab_taskq_cv);
4523 
4524 	rw_destroy(&ipst->ips_srcid_lock);
4525 
4526 	mutex_destroy(&ipst->ips_ip_mi_lock);
4527 	rw_destroy(&ipst->ips_ill_g_usesrc_lock);
4528 
4529 	mutex_destroy(&ipst->ips_igmp_timer_lock);
4530 	mutex_destroy(&ipst->ips_mld_timer_lock);
4531 	mutex_destroy(&ipst->ips_igmp_slowtimeout_lock);
4532 	mutex_destroy(&ipst->ips_mld_slowtimeout_lock);
4533 	mutex_destroy(&ipst->ips_ip_addr_avail_lock);
4534 	rw_destroy(&ipst->ips_ill_g_lock);
4535 
4536 	kmem_free(ipst->ips_phyint_g_list, sizeof (phyint_list_t));
4537 	ipst->ips_phyint_g_list = NULL;
4538 	kmem_free(ipst->ips_ill_g_heads, sizeof (ill_g_head_t) * MAX_G_HEADS);
4539 	ipst->ips_ill_g_heads = NULL;
4540 
4541 	ldi_ident_release(ipst->ips_ldi_ident);
4542 	kmem_free(ipst, sizeof (*ipst));
4543 }
4544 
4545 /*
4546  * This function is called from the TSD destructor, and is used to debug
4547  * reference count issues in IP. See block comment in <inet/ip_if.h> for
4548  * details.
4549  */
4550 static void
ip_thread_exit(void * phash)4551 ip_thread_exit(void *phash)
4552 {
4553 	th_hash_t *thh = phash;
4554 
4555 	rw_enter(&ip_thread_rwlock, RW_WRITER);
4556 	list_remove(&ip_thread_list, thh);
4557 	rw_exit(&ip_thread_rwlock);
4558 	mod_hash_destroy_hash(thh->thh_hash);
4559 	kmem_free(thh, sizeof (*thh));
4560 }
4561 
4562 /*
4563  * Called when the IP kernel module is loaded into the kernel
4564  */
4565 void
ip_ddi_init(void)4566 ip_ddi_init(void)
4567 {
4568 	ip_squeue_flag = ip_squeue_switch(ip_squeue_enter);
4569 
4570 	/*
4571 	 * For IP and TCP the minor numbers should start from 2 since we have 4
4572 	 * initial devices: ip, ip6, tcp, tcp6.
4573 	 */
4574 	/*
4575 	 * If this is a 64-bit kernel, then create two separate arenas -
4576 	 * one for TLIs in the range of INET_MIN_DEV+2 through 2^^18-1, and the
4577 	 * other for socket apps in the range 2^^18 through 2^^32-1.
4578 	 */
4579 	ip_minor_arena_la = NULL;
4580 	ip_minor_arena_sa = NULL;
4581 #if defined(_LP64)
4582 	if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4583 	    INET_MIN_DEV + 2, MAXMIN32, KM_SLEEP)) == NULL) {
4584 		cmn_err(CE_PANIC,
4585 		    "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4586 	}
4587 	if ((ip_minor_arena_la = inet_minor_create("ip_minor_arena_la",
4588 	    MAXMIN32 + 1, MAXMIN64, KM_SLEEP)) == NULL) {
4589 		cmn_err(CE_PANIC,
4590 		    "ip_ddi_init: ip_minor_arena_la creation failed\n");
4591 	}
4592 #else
4593 	if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4594 	    INET_MIN_DEV + 2, MAXMIN, KM_SLEEP)) == NULL) {
4595 		cmn_err(CE_PANIC,
4596 		    "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4597 	}
4598 #endif
4599 	ip_poll_normal_ticks = MSEC_TO_TICK_ROUNDUP(ip_poll_normal_ms);
4600 
4601 	ipcl_g_init();
4602 	ip_ire_g_init();
4603 	ip_net_g_init();
4604 
4605 #ifdef DEBUG
4606 	tsd_create(&ip_thread_data, ip_thread_exit);
4607 	rw_init(&ip_thread_rwlock, NULL, RW_DEFAULT, NULL);
4608 	list_create(&ip_thread_list, sizeof (th_hash_t),
4609 	    offsetof(th_hash_t, thh_link));
4610 #endif
4611 	ipsec_policy_g_init();
4612 	tcp_ddi_g_init();
4613 	sctp_ddi_g_init();
4614 	dce_g_init();
4615 
4616 	/*
4617 	 * We want to be informed each time a stack is created or
4618 	 * destroyed in the kernel, so we can maintain the
4619 	 * set of udp_stack_t's.
4620 	 */
4621 	netstack_register(NS_IP, ip_stack_init, ip_stack_shutdown,
4622 	    ip_stack_fini);
4623 
4624 	tnet_init();
4625 
4626 	udp_ddi_g_init();
4627 	rts_ddi_g_init();
4628 	icmp_ddi_g_init();
4629 	ilb_ddi_g_init();
4630 
4631 	/* This needs to be called after all transports are initialized. */
4632 	mutex_enter(&cpu_lock);
4633 	register_cpu_setup_func(ip_tp_cpu_update, NULL);
4634 	mutex_exit(&cpu_lock);
4635 }
4636 
4637 /*
4638  * Initialize the IP stack instance.
4639  */
4640 static void *
ip_stack_init(netstackid_t stackid,netstack_t * ns)4641 ip_stack_init(netstackid_t stackid, netstack_t *ns)
4642 {
4643 	ip_stack_t	*ipst;
4644 	size_t		arrsz;
4645 	major_t		major;
4646 
4647 #ifdef NS_DEBUG
4648 	printf("ip_stack_init(stack %d)\n", stackid);
4649 #endif
4650 
4651 	ipst = (ip_stack_t *)kmem_zalloc(sizeof (*ipst), KM_SLEEP);
4652 	ipst->ips_netstack = ns;
4653 
4654 	ipst->ips_ill_g_heads = kmem_zalloc(sizeof (ill_g_head_t) * MAX_G_HEADS,
4655 	    KM_SLEEP);
4656 	ipst->ips_phyint_g_list = kmem_zalloc(sizeof (phyint_list_t),
4657 	    KM_SLEEP);
4658 	ipst->ips_ndp4 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4659 	ipst->ips_ndp6 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4660 	mutex_init(&ipst->ips_ndp4->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4661 	mutex_init(&ipst->ips_ndp6->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4662 
4663 	mutex_init(&ipst->ips_igmp_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4664 	ipst->ips_igmp_deferred_next = INFINITY;
4665 	mutex_init(&ipst->ips_mld_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4666 	ipst->ips_mld_deferred_next = INFINITY;
4667 	mutex_init(&ipst->ips_igmp_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4668 	mutex_init(&ipst->ips_mld_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4669 	mutex_init(&ipst->ips_ip_mi_lock, NULL, MUTEX_DEFAULT, NULL);
4670 	mutex_init(&ipst->ips_ip_addr_avail_lock, NULL, MUTEX_DEFAULT, NULL);
4671 	rw_init(&ipst->ips_ill_g_lock, NULL, RW_DEFAULT, NULL);
4672 	rw_init(&ipst->ips_ill_g_usesrc_lock, NULL, RW_DEFAULT, NULL);
4673 
4674 	ipcl_init(ipst);
4675 	ip_ire_init(ipst);
4676 	ip6_asp_init(ipst);
4677 	ipif_init(ipst);
4678 	conn_drain_init(ipst);
4679 	ip_mrouter_stack_init(ipst);
4680 	dce_stack_init(ipst);
4681 
4682 	ipst->ips_ip_multirt_log_interval = 1000;
4683 
4684 	ipst->ips_ill_index = 1;
4685 
4686 	ipst->ips_saved_ip_forwarding = -1;
4687 	ipst->ips_reg_vif_num = ALL_VIFS;	/* Index to Register vif */
4688 
4689 	arrsz = ip_propinfo_count * sizeof (mod_prop_info_t);
4690 	ipst->ips_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz, KM_SLEEP);
4691 	bcopy(ip_propinfo_tbl, ipst->ips_propinfo_tbl, arrsz);
4692 
4693 	ipst->ips_ip_mibkp = ip_kstat_init(stackid, ipst);
4694 	ipst->ips_icmp_mibkp = icmp_kstat_init(stackid);
4695 	ipst->ips_ip_kstat = ip_kstat2_init(stackid, &ipst->ips_ip_statistics);
4696 	ipst->ips_ip6_kstat =
4697 	    ip6_kstat_init(stackid, &ipst->ips_ip6_statistics);
4698 
4699 	ipst->ips_ip_src_id = 1;
4700 	rw_init(&ipst->ips_srcid_lock, NULL, RW_DEFAULT, NULL);
4701 
4702 	ipst->ips_src_generation = SRC_GENERATION_INITIAL;
4703 
4704 	ip_net_init(ipst, ns);
4705 	ipv4_hook_init(ipst);
4706 	ipv6_hook_init(ipst);
4707 	arp_hook_init(ipst);
4708 	ipmp_init(ipst);
4709 	ipobs_init(ipst);
4710 
4711 	/*
4712 	 * Create the taskq dispatcher thread and initialize related stuff.
4713 	 */
4714 	mutex_init(&ipst->ips_capab_taskq_lock, NULL, MUTEX_DEFAULT, NULL);
4715 	cv_init(&ipst->ips_capab_taskq_cv, NULL, CV_DEFAULT, NULL);
4716 	ipst->ips_capab_taskq_thread = thread_create(NULL, 0,
4717 	    ill_taskq_dispatch, ipst, 0, &p0, TS_RUN, minclsyspri);
4718 
4719 	major = mod_name_to_major(INET_NAME);
4720 	(void) ldi_ident_from_major(major, &ipst->ips_ldi_ident);
4721 	return (ipst);
4722 }
4723 
4724 /*
4725  * Allocate and initialize a DLPI template of the specified length.  (May be
4726  * called as writer.)
4727  */
4728 mblk_t *
ip_dlpi_alloc(size_t len,t_uscalar_t prim)4729 ip_dlpi_alloc(size_t len, t_uscalar_t prim)
4730 {
4731 	mblk_t	*mp;
4732 
4733 	mp = allocb(len, BPRI_MED);
4734 	if (!mp)
4735 		return (NULL);
4736 
4737 	/*
4738 	 * DLPIv2 says that DL_INFO_REQ and DL_TOKEN_REQ (the latter
4739 	 * of which we don't seem to use) are sent with M_PCPROTO, and
4740 	 * that other DLPI are M_PROTO.
4741 	 */
4742 	if (prim == DL_INFO_REQ) {
4743 		mp->b_datap->db_type = M_PCPROTO;
4744 	} else {
4745 		mp->b_datap->db_type = M_PROTO;
4746 	}
4747 
4748 	mp->b_wptr = mp->b_rptr + len;
4749 	bzero(mp->b_rptr, len);
4750 	((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
4751 	return (mp);
4752 }
4753 
4754 /*
4755  * Allocate and initialize a DLPI notification.  (May be called as writer.)
4756  */
4757 mblk_t *
ip_dlnotify_alloc(uint_t notification,uint_t data)4758 ip_dlnotify_alloc(uint_t notification, uint_t data)
4759 {
4760 	dl_notify_ind_t	*notifyp;
4761 	mblk_t		*mp;
4762 
4763 	if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4764 		return (NULL);
4765 
4766 	notifyp = (dl_notify_ind_t *)mp->b_rptr;
4767 	notifyp->dl_notification = notification;
4768 	notifyp->dl_data = data;
4769 	return (mp);
4770 }
4771 
4772 mblk_t *
ip_dlnotify_alloc2(uint_t notification,uint_t data1,uint_t data2)4773 ip_dlnotify_alloc2(uint_t notification, uint_t data1, uint_t data2)
4774 {
4775 	dl_notify_ind_t	*notifyp;
4776 	mblk_t		*mp;
4777 
4778 	if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4779 		return (NULL);
4780 
4781 	notifyp = (dl_notify_ind_t *)mp->b_rptr;
4782 	notifyp->dl_notification = notification;
4783 	notifyp->dl_data1 = data1;
4784 	notifyp->dl_data2 = data2;
4785 	return (mp);
4786 }
4787 
4788 /*
4789  * Debug formatting routine.  Returns a character string representation of the
4790  * addr in buf, of the form xxx.xxx.xxx.xxx.  This routine takes the address
4791  * in the form of a ipaddr_t and calls ip_dot_saddr with a pointer.
4792  *
4793  * Once the ndd table-printing interfaces are removed, this can be changed to
4794  * standard dotted-decimal form.
4795  */
4796 char *
ip_dot_addr(ipaddr_t addr,char * buf)4797 ip_dot_addr(ipaddr_t addr, char *buf)
4798 {
4799 	uint8_t *ap = (uint8_t *)&addr;
4800 
4801 	(void) mi_sprintf(buf, "%03d.%03d.%03d.%03d",
4802 	    ap[0] & 0xFF, ap[1] & 0xFF, ap[2] & 0xFF, ap[3] & 0xFF);
4803 	return (buf);
4804 }
4805 
4806 /*
4807  * Write the given MAC address as a printable string in the usual colon-
4808  * separated format.
4809  */
4810 const char *
mac_colon_addr(const uint8_t * addr,size_t alen,char * buf,size_t buflen)4811 mac_colon_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
4812 {
4813 	char *bp;
4814 
4815 	if (alen == 0 || buflen < 4)
4816 		return ("?");
4817 	bp = buf;
4818 	for (;;) {
4819 		/*
4820 		 * If there are more MAC address bytes available, but we won't
4821 		 * have any room to print them, then add "..." to the string
4822 		 * instead.  See below for the 'magic number' explanation.
4823 		 */
4824 		if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
4825 			(void) strcpy(bp, "...");
4826 			break;
4827 		}
4828 		(void) sprintf(bp, "%02x", *addr++);
4829 		bp += 2;
4830 		if (--alen == 0)
4831 			break;
4832 		*bp++ = ':';
4833 		buflen -= 3;
4834 		/*
4835 		 * At this point, based on the first 'if' statement above,
4836 		 * either alen == 1 and buflen >= 3, or alen > 1 and
4837 		 * buflen >= 4.  The first case leaves room for the final "xx"
4838 		 * number and trailing NUL byte.  The second leaves room for at
4839 		 * least "...".  Thus the apparently 'magic' numbers chosen for
4840 		 * that statement.
4841 		 */
4842 	}
4843 	return (buf);
4844 }
4845 
4846 /*
4847  * Called when it is conceptually a ULP that would sent the packet
4848  * e.g., port unreachable and protocol unreachable. Check that the packet
4849  * would have passed the IPsec global policy before sending the error.
4850  *
4851  * Send an ICMP error after patching up the packet appropriately.
4852  * Uses ip_drop_input and bumps the appropriate MIB.
4853  */
4854 void
ip_fanout_send_icmp_v4(mblk_t * mp,uint_t icmp_type,uint_t icmp_code,ip_recv_attr_t * ira)4855 ip_fanout_send_icmp_v4(mblk_t *mp, uint_t icmp_type, uint_t icmp_code,
4856     ip_recv_attr_t *ira)
4857 {
4858 	ipha_t		*ipha;
4859 	boolean_t	secure;
4860 	ill_t		*ill = ira->ira_ill;
4861 	ip_stack_t	*ipst = ill->ill_ipst;
4862 	netstack_t	*ns = ipst->ips_netstack;
4863 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
4864 
4865 	secure = ira->ira_flags & IRAF_IPSEC_SECURE;
4866 
4867 	/*
4868 	 * We are generating an icmp error for some inbound packet.
4869 	 * Called from all ip_fanout_(udp, tcp, proto) functions.
4870 	 * Before we generate an error, check with global policy
4871 	 * to see whether this is allowed to enter the system. As
4872 	 * there is no "conn", we are checking with global policy.
4873 	 */
4874 	ipha = (ipha_t *)mp->b_rptr;
4875 	if (secure || ipss->ipsec_inbound_v4_policy_present) {
4876 		mp = ipsec_check_global_policy(mp, NULL, ipha, NULL, ira, ns);
4877 		if (mp == NULL)
4878 			return;
4879 	}
4880 
4881 	/* We never send errors for protocols that we do implement */
4882 	if (ira->ira_protocol == IPPROTO_ICMP ||
4883 	    ira->ira_protocol == IPPROTO_IGMP) {
4884 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4885 		ip_drop_input("ip_fanout_send_icmp_v4", mp, ill);
4886 		freemsg(mp);
4887 		return;
4888 	}
4889 	/*
4890 	 * Have to correct checksum since
4891 	 * the packet might have been
4892 	 * fragmented and the reassembly code in ip_rput
4893 	 * does not restore the IP checksum.
4894 	 */
4895 	ipha->ipha_hdr_checksum = 0;
4896 	ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
4897 
4898 	switch (icmp_type) {
4899 	case ICMP_DEST_UNREACHABLE:
4900 		switch (icmp_code) {
4901 		case ICMP_PROTOCOL_UNREACHABLE:
4902 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInUnknownProtos);
4903 			ip_drop_input("ipIfStatsInUnknownProtos", mp, ill);
4904 			break;
4905 		case ICMP_PORT_UNREACHABLE:
4906 			BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
4907 			ip_drop_input("ipIfStatsNoPorts", mp, ill);
4908 			break;
4909 		}
4910 
4911 		icmp_unreachable(mp, icmp_code, ira);
4912 		break;
4913 	default:
4914 #ifdef DEBUG
4915 		panic("ip_fanout_send_icmp_v4: wrong type");
4916 		/*NOTREACHED*/
4917 #else
4918 		freemsg(mp);
4919 		break;
4920 #endif
4921 	}
4922 }
4923 
4924 /*
4925  * Used to send an ICMP error message when a packet is received for
4926  * a protocol that is not supported. The mblk passed as argument
4927  * is consumed by this function.
4928  */
4929 void
ip_proto_not_sup(mblk_t * mp,ip_recv_attr_t * ira)4930 ip_proto_not_sup(mblk_t *mp, ip_recv_attr_t *ira)
4931 {
4932 	ipha_t		*ipha;
4933 
4934 	ipha = (ipha_t *)mp->b_rptr;
4935 	if (ira->ira_flags & IRAF_IS_IPV4) {
4936 		ASSERT(IPH_HDR_VERSION(ipha) == IP_VERSION);
4937 		ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
4938 		    ICMP_PROTOCOL_UNREACHABLE, ira);
4939 	} else {
4940 		ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
4941 		ip_fanout_send_icmp_v6(mp, ICMP6_PARAM_PROB,
4942 		    ICMP6_PARAMPROB_NEXTHEADER, ira);
4943 	}
4944 }
4945 
4946 /*
4947  * Deliver a rawip packet to the given conn, possibly applying ipsec policy.
4948  * Handles IPv4 and IPv6.
4949  * We are responsible for disposing of mp, such as by freemsg() or putnext()
4950  * Caller is responsible for dropping references to the conn.
4951  */
4952 void
ip_fanout_proto_conn(conn_t * connp,mblk_t * mp,ipha_t * ipha,ip6_t * ip6h,ip_recv_attr_t * ira)4953 ip_fanout_proto_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4954     ip_recv_attr_t *ira)
4955 {
4956 	ill_t		*ill = ira->ira_ill;
4957 	ip_stack_t	*ipst = ill->ill_ipst;
4958 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
4959 	boolean_t	secure;
4960 	uint_t		protocol = ira->ira_protocol;
4961 	iaflags_t	iraflags = ira->ira_flags;
4962 	queue_t		*rq;
4963 
4964 	secure = iraflags & IRAF_IPSEC_SECURE;
4965 
4966 	rq = connp->conn_rq;
4967 	if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
4968 		switch (protocol) {
4969 		case IPPROTO_ICMPV6:
4970 			BUMP_MIB(ill->ill_icmp6_mib, ipv6IfIcmpInOverflows);
4971 			break;
4972 		case IPPROTO_ICMP:
4973 			BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
4974 			break;
4975 		default:
4976 			BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
4977 			break;
4978 		}
4979 		freemsg(mp);
4980 		return;
4981 	}
4982 
4983 	ASSERT(!(IPCL_IS_IPTUN(connp)));
4984 
4985 	if (connp->conn_min_ttl != 0 && connp->conn_min_ttl > ira->ira_ttl) {
4986 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4987 		ip_drop_input("ipIfStatsInDiscards", mp, ill);
4988 		freemsg(mp);
4989 		return;
4990 	}
4991 
4992 	if (((iraflags & IRAF_IS_IPV4) ?
4993 	    CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
4994 	    CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
4995 	    secure) {
4996 		mp = ipsec_check_inbound_policy(mp, connp, ipha,
4997 		    ip6h, ira);
4998 		if (mp == NULL) {
4999 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5000 			/* Note that mp is NULL */
5001 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
5002 			return;
5003 		}
5004 	}
5005 
5006 	if (iraflags & IRAF_ICMP_ERROR) {
5007 		(connp->conn_recvicmp)(connp, mp, NULL, ira);
5008 	} else {
5009 		ill_t *rill = ira->ira_rill;
5010 
5011 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5012 		ira->ira_ill = ira->ira_rill = NULL;
5013 		/* Send it upstream */
5014 		(connp->conn_recv)(connp, mp, NULL, ira);
5015 		ira->ira_ill = ill;
5016 		ira->ira_rill = rill;
5017 	}
5018 }
5019 
5020 /*
5021  * Handle protocols with which IP is less intimate.  There
5022  * can be more than one stream bound to a particular
5023  * protocol.  When this is the case, normally each one gets a copy
5024  * of any incoming packets.
5025  *
5026  * IPsec NOTE :
5027  *
5028  * Don't allow a secure packet going up a non-secure connection.
5029  * We don't allow this because
5030  *
5031  * 1) Reply might go out in clear which will be dropped at
5032  *    the sending side.
5033  * 2) If the reply goes out in clear it will give the
5034  *    adversary enough information for getting the key in
5035  *    most of the cases.
5036  *
5037  * Moreover getting a secure packet when we expect clear
5038  * implies that SA's were added without checking for
5039  * policy on both ends. This should not happen once ISAKMP
5040  * is used to negotiate SAs as SAs will be added only after
5041  * verifying the policy.
5042  *
5043  * Zones notes:
5044  * Earlier in ip_input on a system with multiple shared-IP zones we
5045  * duplicate the multicast and broadcast packets and send them up
5046  * with each explicit zoneid that exists on that ill.
5047  * This means that here we can match the zoneid with SO_ALLZONES being special.
5048  */
5049 void
ip_fanout_proto_v4(mblk_t * mp,ipha_t * ipha,ip_recv_attr_t * ira)5050 ip_fanout_proto_v4(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
5051 {
5052 	mblk_t		*mp1;
5053 	ipaddr_t	laddr;
5054 	conn_t		*connp, *first_connp, *next_connp;
5055 	connf_t		*connfp;
5056 	ill_t		*ill = ira->ira_ill;
5057 	ip_stack_t	*ipst = ill->ill_ipst;
5058 
5059 	laddr = ipha->ipha_dst;
5060 
5061 	connfp = &ipst->ips_ipcl_proto_fanout_v4[ira->ira_protocol];
5062 	mutex_enter(&connfp->connf_lock);
5063 	connp = connfp->connf_head;
5064 	for (connp = connfp->connf_head; connp != NULL;
5065 	    connp = connp->conn_next) {
5066 		/* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5067 		if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5068 		    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5069 		    tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp))) {
5070 			break;
5071 		}
5072 	}
5073 
5074 	if (connp == NULL) {
5075 		/*
5076 		 * No one bound to these addresses.  Is
5077 		 * there a client that wants all
5078 		 * unclaimed datagrams?
5079 		 */
5080 		mutex_exit(&connfp->connf_lock);
5081 		ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
5082 		    ICMP_PROTOCOL_UNREACHABLE, ira);
5083 		return;
5084 	}
5085 
5086 	ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5087 
5088 	CONN_INC_REF(connp);
5089 	first_connp = connp;
5090 	connp = connp->conn_next;
5091 
5092 	for (;;) {
5093 		while (connp != NULL) {
5094 			/* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5095 			if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5096 			    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5097 			    tsol_receive_local(mp, &laddr, IPV4_VERSION,
5098 			    ira, connp)))
5099 				break;
5100 			connp = connp->conn_next;
5101 		}
5102 
5103 		if (connp == NULL) {
5104 			/* No more interested clients */
5105 			connp = first_connp;
5106 			break;
5107 		}
5108 		if (((mp1 = dupmsg(mp)) == NULL) &&
5109 		    ((mp1 = copymsg(mp)) == NULL)) {
5110 			/* Memory allocation failed */
5111 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5112 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
5113 			connp = first_connp;
5114 			break;
5115 		}
5116 
5117 		CONN_INC_REF(connp);
5118 		mutex_exit(&connfp->connf_lock);
5119 
5120 		ip_fanout_proto_conn(connp, mp1, (ipha_t *)mp1->b_rptr, NULL,
5121 		    ira);
5122 
5123 		mutex_enter(&connfp->connf_lock);
5124 		/* Follow the next pointer before releasing the conn. */
5125 		next_connp = connp->conn_next;
5126 		CONN_DEC_REF(connp);
5127 		connp = next_connp;
5128 	}
5129 
5130 	/* Last one.  Send it upstream. */
5131 	mutex_exit(&connfp->connf_lock);
5132 
5133 	ip_fanout_proto_conn(connp, mp, ipha, NULL, ira);
5134 
5135 	CONN_DEC_REF(connp);
5136 }
5137 
5138 /*
5139  * If we have a IPsec NAT-Traversal packet, strip the zero-SPI or
5140  * pass it along to ESP if the SPI is non-zero.  Returns the mblk if the mblk
5141  * is not consumed.
5142  *
5143  * One of three things can happen, all of which affect the passed-in mblk:
5144  *
5145  * 1.) The packet is stock UDP and gets its zero-SPI stripped.  Return mblk..
5146  *
5147  * 2.) The packet is ESP-in-UDP, gets transformed into an equivalent
5148  *     ESP packet, and is passed along to ESP for consumption.  Return NULL.
5149  *
5150  * 3.) The packet is an ESP-in-UDP Keepalive.  Drop it and return NULL.
5151  */
5152 mblk_t *
zero_spi_check(mblk_t * mp,ip_recv_attr_t * ira)5153 zero_spi_check(mblk_t *mp, ip_recv_attr_t *ira)
5154 {
5155 	int shift, plen, iph_len;
5156 	ipha_t *ipha;
5157 	udpha_t *udpha;
5158 	uint32_t *spi;
5159 	uint32_t esp_ports;
5160 	uint8_t *orptr;
5161 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
5162 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
5163 
5164 	ipha = (ipha_t *)mp->b_rptr;
5165 	iph_len = ira->ira_ip_hdr_length;
5166 	plen = ira->ira_pktlen;
5167 
5168 	if (plen - iph_len - sizeof (udpha_t) < sizeof (uint32_t)) {
5169 		/*
5170 		 * Most likely a keepalive for the benefit of an intervening
5171 		 * NAT.  These aren't for us, per se, so drop it.
5172 		 *
5173 		 * RFC 3947/8 doesn't say for sure what to do for 2-3
5174 		 * byte packets (keepalives are 1-byte), but we'll drop them
5175 		 * also.
5176 		 */
5177 		ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5178 		    DROPPER(ipss, ipds_esp_nat_t_ka), &ipss->ipsec_dropper);
5179 		return (NULL);
5180 	}
5181 
5182 	if (MBLKL(mp) < iph_len + sizeof (udpha_t) + sizeof (*spi)) {
5183 		/* might as well pull it all up - it might be ESP. */
5184 		if (!pullupmsg(mp, -1)) {
5185 			ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5186 			    DROPPER(ipss, ipds_esp_nomem),
5187 			    &ipss->ipsec_dropper);
5188 			return (NULL);
5189 		}
5190 
5191 		ipha = (ipha_t *)mp->b_rptr;
5192 	}
5193 	spi = (uint32_t *)(mp->b_rptr + iph_len + sizeof (udpha_t));
5194 	if (*spi == 0) {
5195 		/* UDP packet - remove 0-spi. */
5196 		shift = sizeof (uint32_t);
5197 	} else {
5198 		/* ESP-in-UDP packet - reduce to ESP. */
5199 		ipha->ipha_protocol = IPPROTO_ESP;
5200 		shift = sizeof (udpha_t);
5201 	}
5202 
5203 	/* Fix IP header */
5204 	ira->ira_pktlen = (plen - shift);
5205 	ipha->ipha_length = htons(ira->ira_pktlen);
5206 	ipha->ipha_hdr_checksum = 0;
5207 
5208 	orptr = mp->b_rptr;
5209 	mp->b_rptr += shift;
5210 
5211 	udpha = (udpha_t *)(orptr + iph_len);
5212 	if (*spi == 0) {
5213 		ASSERT((uint8_t *)ipha == orptr);
5214 		udpha->uha_length = htons(plen - shift - iph_len);
5215 		iph_len += sizeof (udpha_t);	/* For the call to ovbcopy(). */
5216 		esp_ports = 0;
5217 	} else {
5218 		esp_ports = *((uint32_t *)udpha);
5219 		ASSERT(esp_ports != 0);
5220 	}
5221 	ovbcopy(orptr, orptr + shift, iph_len);
5222 	if (esp_ports != 0) /* Punt up for ESP processing. */ {
5223 		ipha = (ipha_t *)(orptr + shift);
5224 
5225 		ira->ira_flags |= IRAF_ESP_UDP_PORTS;
5226 		ira->ira_esp_udp_ports = esp_ports;
5227 		ip_fanout_v4(mp, ipha, ira);
5228 		return (NULL);
5229 	}
5230 	return (mp);
5231 }
5232 
5233 /*
5234  * Deliver a udp packet to the given conn, possibly applying ipsec policy.
5235  * Handles IPv4 and IPv6.
5236  * We are responsible for disposing of mp, such as by freemsg() or putnext()
5237  * Caller is responsible for dropping references to the conn.
5238  */
5239 void
ip_fanout_udp_conn(conn_t * connp,mblk_t * mp,ipha_t * ipha,ip6_t * ip6h,ip_recv_attr_t * ira)5240 ip_fanout_udp_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
5241     ip_recv_attr_t *ira)
5242 {
5243 	ill_t		*ill = ira->ira_ill;
5244 	ip_stack_t	*ipst = ill->ill_ipst;
5245 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
5246 	boolean_t	secure;
5247 	iaflags_t	iraflags = ira->ira_flags;
5248 
5249 	secure = iraflags & IRAF_IPSEC_SECURE;
5250 
5251 	if (connp->conn_min_ttl != 0 && connp->conn_min_ttl > ira->ira_ttl) {
5252 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5253 		ip_drop_input("ipIfStatsInDiscards", mp, ill);
5254 		freemsg(mp);
5255 		return;
5256 	}
5257 
5258 	if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld :
5259 	    !canputnext(connp->conn_rq)) {
5260 		BUMP_MIB(ill->ill_ip_mib, udpIfStatsInOverflows);
5261 		freemsg(mp);
5262 		return;
5263 	}
5264 
5265 	if (((iraflags & IRAF_IS_IPV4) ?
5266 	    CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
5267 	    CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
5268 	    secure) {
5269 		mp = ipsec_check_inbound_policy(mp, connp, ipha,
5270 		    ip6h, ira);
5271 		if (mp == NULL) {
5272 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5273 			/* Note that mp is NULL */
5274 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
5275 			return;
5276 		}
5277 	}
5278 
5279 	/*
5280 	 * Since this code is not used for UDP unicast we don't need a NAT_T
5281 	 * check. Only ip_fanout_v4 has that check.
5282 	 */
5283 	if (ira->ira_flags & IRAF_ICMP_ERROR) {
5284 		(connp->conn_recvicmp)(connp, mp, NULL, ira);
5285 	} else {
5286 		ill_t *rill = ira->ira_rill;
5287 
5288 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5289 		ira->ira_ill = ira->ira_rill = NULL;
5290 		/* Send it upstream */
5291 		(connp->conn_recv)(connp, mp, NULL, ira);
5292 		ira->ira_ill = ill;
5293 		ira->ira_rill = rill;
5294 	}
5295 }
5296 
5297 /*
5298  * Fanout for UDP packets that are multicast or broadcast, and ICMP errors.
5299  * (Unicast fanout is handled in ip_input_v4.)
5300  *
5301  * If SO_REUSEADDR is set all multicast and broadcast packets
5302  * will be delivered to all conns bound to the same port.
5303  *
5304  * If there is at least one matching AF_INET receiver, then we will
5305  * ignore any AF_INET6 receivers.
5306  * In the special case where an AF_INET socket binds to 0.0.0.0/<port> and an
5307  * AF_INET6 socket binds to ::/<port>, only the AF_INET socket receives the IPv4
5308  * packets.
5309  *
5310  * Zones notes:
5311  * Earlier in ip_input on a system with multiple shared-IP zones we
5312  * duplicate the multicast and broadcast packets and send them up
5313  * with each explicit zoneid that exists on that ill.
5314  * This means that here we can match the zoneid with SO_ALLZONES being special.
5315  */
5316 void
ip_fanout_udp_multi_v4(mblk_t * mp,ipha_t * ipha,uint16_t lport,uint16_t fport,ip_recv_attr_t * ira)5317 ip_fanout_udp_multi_v4(mblk_t *mp, ipha_t *ipha, uint16_t lport, uint16_t fport,
5318     ip_recv_attr_t *ira)
5319 {
5320 	ipaddr_t	laddr;
5321 	in6_addr_t	v6faddr;
5322 	conn_t		*connp;
5323 	connf_t		*connfp;
5324 	ipaddr_t	faddr;
5325 	ill_t		*ill = ira->ira_ill;
5326 	ip_stack_t	*ipst = ill->ill_ipst;
5327 
5328 	ASSERT(ira->ira_flags & (IRAF_MULTIBROADCAST|IRAF_ICMP_ERROR));
5329 
5330 	laddr = ipha->ipha_dst;
5331 	faddr = ipha->ipha_src;
5332 
5333 	connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5334 	mutex_enter(&connfp->connf_lock);
5335 	connp = connfp->connf_head;
5336 
5337 	/*
5338 	 * If SO_REUSEADDR has been set on the first we send the
5339 	 * packet to all clients that have joined the group and
5340 	 * match the port.
5341 	 */
5342 	while (connp != NULL) {
5343 		if ((IPCL_UDP_MATCH(connp, lport, laddr, fport, faddr)) &&
5344 		    conn_wantpacket(connp, ira, ipha) &&
5345 		    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5346 		    tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5347 			break;
5348 		connp = connp->conn_next;
5349 	}
5350 
5351 	if (connp == NULL)
5352 		goto notfound;
5353 
5354 	CONN_INC_REF(connp);
5355 
5356 	if (connp->conn_reuseaddr) {
5357 		conn_t		*first_connp = connp;
5358 		conn_t		*next_connp;
5359 		mblk_t		*mp1;
5360 
5361 		connp = connp->conn_next;
5362 		for (;;) {
5363 			while (connp != NULL) {
5364 				if (IPCL_UDP_MATCH(connp, lport, laddr,
5365 				    fport, faddr) &&
5366 				    conn_wantpacket(connp, ira, ipha) &&
5367 				    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5368 				    tsol_receive_local(mp, &laddr, IPV4_VERSION,
5369 				    ira, connp)))
5370 					break;
5371 				connp = connp->conn_next;
5372 			}
5373 			if (connp == NULL) {
5374 				/* No more interested clients */
5375 				connp = first_connp;
5376 				break;
5377 			}
5378 			if (((mp1 = dupmsg(mp)) == NULL) &&
5379 			    ((mp1 = copymsg(mp)) == NULL)) {
5380 				/* Memory allocation failed */
5381 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5382 				ip_drop_input("ipIfStatsInDiscards", mp, ill);
5383 				connp = first_connp;
5384 				break;
5385 			}
5386 			CONN_INC_REF(connp);
5387 			mutex_exit(&connfp->connf_lock);
5388 
5389 			IP_STAT(ipst, ip_udp_fanmb);
5390 			ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5391 			    NULL, ira);
5392 			mutex_enter(&connfp->connf_lock);
5393 			/* Follow the next pointer before releasing the conn */
5394 			next_connp = connp->conn_next;
5395 			CONN_DEC_REF(connp);
5396 			connp = next_connp;
5397 		}
5398 	}
5399 
5400 	/* Last one.  Send it upstream. */
5401 	mutex_exit(&connfp->connf_lock);
5402 	IP_STAT(ipst, ip_udp_fanmb);
5403 	ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5404 	CONN_DEC_REF(connp);
5405 	return;
5406 
5407 notfound:
5408 	mutex_exit(&connfp->connf_lock);
5409 	/*
5410 	 * IPv6 endpoints bound to multicast IPv4-mapped addresses
5411 	 * have already been matched above, since they live in the IPv4
5412 	 * fanout tables. This implies we only need to
5413 	 * check for IPv6 in6addr_any endpoints here.
5414 	 * Thus we compare using ipv6_all_zeros instead of the destination
5415 	 * address, except for the multicast group membership lookup which
5416 	 * uses the IPv4 destination.
5417 	 */
5418 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6faddr);
5419 	connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5420 	mutex_enter(&connfp->connf_lock);
5421 	connp = connfp->connf_head;
5422 	/*
5423 	 * IPv4 multicast packet being delivered to an AF_INET6
5424 	 * in6addr_any endpoint.
5425 	 * Need to check conn_wantpacket(). Note that we use conn_wantpacket()
5426 	 * and not conn_wantpacket_v6() since any multicast membership is
5427 	 * for an IPv4-mapped multicast address.
5428 	 */
5429 	while (connp != NULL) {
5430 		if (IPCL_UDP_MATCH_V6(connp, lport, ipv6_all_zeros,
5431 		    fport, v6faddr) &&
5432 		    conn_wantpacket(connp, ira, ipha) &&
5433 		    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5434 		    tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5435 			break;
5436 		connp = connp->conn_next;
5437 	}
5438 
5439 	if (connp == NULL) {
5440 		/*
5441 		 * No one bound to this port.  Is
5442 		 * there a client that wants all
5443 		 * unclaimed datagrams?
5444 		 */
5445 		mutex_exit(&connfp->connf_lock);
5446 
5447 		if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_UDP].connf_head !=
5448 		    NULL) {
5449 			ASSERT(ira->ira_protocol == IPPROTO_UDP);
5450 			ip_fanout_proto_v4(mp, ipha, ira);
5451 		} else {
5452 			/*
5453 			 * We used to attempt to send an icmp error here, but
5454 			 * since this is known to be a multicast packet
5455 			 * and we don't send icmp errors in response to
5456 			 * multicast, just drop the packet and give up sooner.
5457 			 */
5458 			BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
5459 			freemsg(mp);
5460 		}
5461 		return;
5462 	}
5463 	CONN_INC_REF(connp);
5464 	ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5465 
5466 	/*
5467 	 * If SO_REUSEADDR has been set on the first we send the
5468 	 * packet to all clients that have joined the group and
5469 	 * match the port.
5470 	 */
5471 	if (connp->conn_reuseaddr) {
5472 		conn_t		*first_connp = connp;
5473 		conn_t		*next_connp;
5474 		mblk_t		*mp1;
5475 
5476 		connp = connp->conn_next;
5477 		for (;;) {
5478 			while (connp != NULL) {
5479 				if (IPCL_UDP_MATCH_V6(connp, lport,
5480 				    ipv6_all_zeros, fport, v6faddr) &&
5481 				    conn_wantpacket(connp, ira, ipha) &&
5482 				    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5483 				    tsol_receive_local(mp, &laddr, IPV4_VERSION,
5484 				    ira, connp)))
5485 					break;
5486 				connp = connp->conn_next;
5487 			}
5488 			if (connp == NULL) {
5489 				/* No more interested clients */
5490 				connp = first_connp;
5491 				break;
5492 			}
5493 			if (((mp1 = dupmsg(mp)) == NULL) &&
5494 			    ((mp1 = copymsg(mp)) == NULL)) {
5495 				/* Memory allocation failed */
5496 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5497 				ip_drop_input("ipIfStatsInDiscards", mp, ill);
5498 				connp = first_connp;
5499 				break;
5500 			}
5501 			CONN_INC_REF(connp);
5502 			mutex_exit(&connfp->connf_lock);
5503 
5504 			IP_STAT(ipst, ip_udp_fanmb);
5505 			ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5506 			    NULL, ira);
5507 			mutex_enter(&connfp->connf_lock);
5508 			/* Follow the next pointer before releasing the conn */
5509 			next_connp = connp->conn_next;
5510 			CONN_DEC_REF(connp);
5511 			connp = next_connp;
5512 		}
5513 	}
5514 
5515 	/* Last one.  Send it upstream. */
5516 	mutex_exit(&connfp->connf_lock);
5517 	IP_STAT(ipst, ip_udp_fanmb);
5518 	ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5519 	CONN_DEC_REF(connp);
5520 }
5521 
5522 /*
5523  * Split an incoming packet's IPv4 options into the label and the other options.
5524  * If 'allocate' is set it does memory allocation for the ip_pkt_t, including
5525  * clearing out any leftover label or options.
5526  * Otherwise it just makes ipp point into the packet.
5527  *
5528  * Returns zero if ok; ENOMEM if the buffer couldn't be allocated.
5529  */
5530 int
ip_find_hdr_v4(ipha_t * ipha,ip_pkt_t * ipp,boolean_t allocate)5531 ip_find_hdr_v4(ipha_t *ipha, ip_pkt_t *ipp, boolean_t allocate)
5532 {
5533 	uchar_t		*opt;
5534 	uint32_t	totallen;
5535 	uint32_t	optval;
5536 	uint32_t	optlen;
5537 
5538 	ipp->ipp_fields |= IPPF_HOPLIMIT | IPPF_TCLASS | IPPF_ADDR;
5539 	ipp->ipp_hoplimit = ipha->ipha_ttl;
5540 	ipp->ipp_type_of_service = ipha->ipha_type_of_service;
5541 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &ipp->ipp_addr);
5542 
5543 	/*
5544 	 * Get length (in 4 byte octets) of IP header options.
5545 	 */
5546 	totallen = ipha->ipha_version_and_hdr_length -
5547 	    (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5548 
5549 	if (totallen == 0) {
5550 		if (!allocate)
5551 			return (0);
5552 
5553 		/* Clear out anything from a previous packet */
5554 		if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5555 			kmem_free(ipp->ipp_ipv4_options,
5556 			    ipp->ipp_ipv4_options_len);
5557 			ipp->ipp_ipv4_options = NULL;
5558 			ipp->ipp_ipv4_options_len = 0;
5559 			ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5560 		}
5561 		if (ipp->ipp_fields & IPPF_LABEL_V4) {
5562 			kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5563 			ipp->ipp_label_v4 = NULL;
5564 			ipp->ipp_label_len_v4 = 0;
5565 			ipp->ipp_fields &= ~IPPF_LABEL_V4;
5566 		}
5567 		return (0);
5568 	}
5569 
5570 	totallen <<= 2;
5571 	opt = (uchar_t *)&ipha[1];
5572 	if (!is_system_labeled()) {
5573 
5574 	copyall:
5575 		if (!allocate) {
5576 			if (totallen != 0) {
5577 				ipp->ipp_ipv4_options = opt;
5578 				ipp->ipp_ipv4_options_len = totallen;
5579 				ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5580 			}
5581 			return (0);
5582 		}
5583 		/* Just copy all of options */
5584 		if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5585 			if (totallen == ipp->ipp_ipv4_options_len) {
5586 				bcopy(opt, ipp->ipp_ipv4_options, totallen);
5587 				return (0);
5588 			}
5589 			kmem_free(ipp->ipp_ipv4_options,
5590 			    ipp->ipp_ipv4_options_len);
5591 			ipp->ipp_ipv4_options = NULL;
5592 			ipp->ipp_ipv4_options_len = 0;
5593 			ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5594 		}
5595 		if (totallen == 0)
5596 			return (0);
5597 
5598 		ipp->ipp_ipv4_options = kmem_alloc(totallen, KM_NOSLEEP);
5599 		if (ipp->ipp_ipv4_options == NULL)
5600 			return (ENOMEM);
5601 		ipp->ipp_ipv4_options_len = totallen;
5602 		ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5603 		bcopy(opt, ipp->ipp_ipv4_options, totallen);
5604 		return (0);
5605 	}
5606 
5607 	if (allocate && (ipp->ipp_fields & IPPF_LABEL_V4)) {
5608 		kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5609 		ipp->ipp_label_v4 = NULL;
5610 		ipp->ipp_label_len_v4 = 0;
5611 		ipp->ipp_fields &= ~IPPF_LABEL_V4;
5612 	}
5613 
5614 	/*
5615 	 * Search for CIPSO option.
5616 	 * We assume CIPSO is first in options if it is present.
5617 	 * If it isn't, then ipp_opt_ipv4_options will not include the options
5618 	 * prior to the CIPSO option.
5619 	 */
5620 	while (totallen != 0) {
5621 		switch (optval = opt[IPOPT_OPTVAL]) {
5622 		case IPOPT_EOL:
5623 			return (0);
5624 		case IPOPT_NOP:
5625 			optlen = 1;
5626 			break;
5627 		default:
5628 			if (totallen <= IPOPT_OLEN)
5629 				return (EINVAL);
5630 			optlen = opt[IPOPT_OLEN];
5631 			if (optlen < 2)
5632 				return (EINVAL);
5633 		}
5634 		if (optlen > totallen)
5635 			return (EINVAL);
5636 
5637 		switch (optval) {
5638 		case IPOPT_COMSEC:
5639 			if (!allocate) {
5640 				ipp->ipp_label_v4 = opt;
5641 				ipp->ipp_label_len_v4 = optlen;
5642 				ipp->ipp_fields |= IPPF_LABEL_V4;
5643 			} else {
5644 				ipp->ipp_label_v4 = kmem_alloc(optlen,
5645 				    KM_NOSLEEP);
5646 				if (ipp->ipp_label_v4 == NULL)
5647 					return (ENOMEM);
5648 				ipp->ipp_label_len_v4 = optlen;
5649 				ipp->ipp_fields |= IPPF_LABEL_V4;
5650 				bcopy(opt, ipp->ipp_label_v4, optlen);
5651 			}
5652 			totallen -= optlen;
5653 			opt += optlen;
5654 
5655 			/* Skip padding bytes until we get to a multiple of 4 */
5656 			while ((totallen & 3) != 0 && opt[0] == IPOPT_NOP) {
5657 				totallen--;
5658 				opt++;
5659 			}
5660 			/* Remaining as ipp_ipv4_options */
5661 			goto copyall;
5662 		}
5663 		totallen -= optlen;
5664 		opt += optlen;
5665 	}
5666 	/* No CIPSO found; return everything as ipp_ipv4_options */
5667 	totallen = ipha->ipha_version_and_hdr_length -
5668 	    (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5669 	totallen <<= 2;
5670 	opt = (uchar_t *)&ipha[1];
5671 	goto copyall;
5672 }
5673 
5674 /*
5675  * Efficient versions of lookup for an IRE when we only
5676  * match the address.
5677  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5678  * Does not handle multicast addresses.
5679  */
5680 uint_t
ip_type_v4(ipaddr_t addr,ip_stack_t * ipst)5681 ip_type_v4(ipaddr_t addr, ip_stack_t *ipst)
5682 {
5683 	ire_t *ire;
5684 	uint_t result;
5685 
5686 	ire = ire_ftable_lookup_simple_v4(addr, 0, ipst, NULL);
5687 	ASSERT(ire != NULL);
5688 	if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5689 		result = IRE_NOROUTE;
5690 	else
5691 		result = ire->ire_type;
5692 	ire_refrele(ire);
5693 	return (result);
5694 }
5695 
5696 /*
5697  * Efficient versions of lookup for an IRE when we only
5698  * match the address.
5699  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5700  * Does not handle multicast addresses.
5701  */
5702 uint_t
ip_type_v6(const in6_addr_t * addr,ip_stack_t * ipst)5703 ip_type_v6(const in6_addr_t *addr, ip_stack_t *ipst)
5704 {
5705 	ire_t *ire;
5706 	uint_t result;
5707 
5708 	ire = ire_ftable_lookup_simple_v6(addr, 0, ipst, NULL);
5709 	ASSERT(ire != NULL);
5710 	if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5711 		result = IRE_NOROUTE;
5712 	else
5713 		result = ire->ire_type;
5714 	ire_refrele(ire);
5715 	return (result);
5716 }
5717 
5718 /*
5719  * Nobody should be sending
5720  * packets up this stream
5721  */
5722 static int
ip_lrput(queue_t * q,mblk_t * mp)5723 ip_lrput(queue_t *q, mblk_t *mp)
5724 {
5725 	switch (mp->b_datap->db_type) {
5726 	case M_FLUSH:
5727 		/* Turn around */
5728 		if (*mp->b_rptr & FLUSHW) {
5729 			*mp->b_rptr &= ~FLUSHR;
5730 			qreply(q, mp);
5731 			return (0);
5732 		}
5733 		break;
5734 	}
5735 	freemsg(mp);
5736 	return (0);
5737 }
5738 
5739 /* Nobody should be sending packets down this stream */
5740 /* ARGSUSED */
5741 int
ip_lwput(queue_t * q,mblk_t * mp)5742 ip_lwput(queue_t *q, mblk_t *mp)
5743 {
5744 	freemsg(mp);
5745 	return (0);
5746 }
5747 
5748 /*
5749  * Move the first hop in any source route to ipha_dst and remove that part of
5750  * the source route.  Called by other protocols.  Errors in option formatting
5751  * are ignored - will be handled by ip_output_options. Return the final
5752  * destination (either ipha_dst or the last entry in a source route.)
5753  */
5754 ipaddr_t
ip_massage_options(ipha_t * ipha,netstack_t * ns)5755 ip_massage_options(ipha_t *ipha, netstack_t *ns)
5756 {
5757 	ipoptp_t	opts;
5758 	uchar_t		*opt;
5759 	uint8_t		optval;
5760 	uint8_t		optlen;
5761 	ipaddr_t	dst;
5762 	int		i;
5763 	ip_stack_t	*ipst = ns->netstack_ip;
5764 
5765 	ip2dbg(("ip_massage_options\n"));
5766 	dst = ipha->ipha_dst;
5767 	for (optval = ipoptp_first(&opts, ipha);
5768 	    optval != IPOPT_EOL;
5769 	    optval = ipoptp_next(&opts)) {
5770 		opt = opts.ipoptp_cur;
5771 		switch (optval) {
5772 			uint8_t off;
5773 		case IPOPT_SSRR:
5774 		case IPOPT_LSRR:
5775 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
5776 				ip1dbg(("ip_massage_options: bad src route\n"));
5777 				break;
5778 			}
5779 			optlen = opts.ipoptp_len;
5780 			off = opt[IPOPT_OFFSET];
5781 			off--;
5782 		redo_srr:
5783 			if (optlen < IP_ADDR_LEN ||
5784 			    off > optlen - IP_ADDR_LEN) {
5785 				/* End of source route */
5786 				ip1dbg(("ip_massage_options: end of SR\n"));
5787 				break;
5788 			}
5789 			bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
5790 			ip1dbg(("ip_massage_options: next hop 0x%x\n",
5791 			    ntohl(dst)));
5792 			/*
5793 			 * Check if our address is present more than
5794 			 * once as consecutive hops in source route.
5795 			 * XXX verify per-interface ip_forwarding
5796 			 * for source route?
5797 			 */
5798 			if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
5799 				off += IP_ADDR_LEN;
5800 				goto redo_srr;
5801 			}
5802 			if (dst == htonl(INADDR_LOOPBACK)) {
5803 				ip1dbg(("ip_massage_options: loopback addr in "
5804 				    "source route!\n"));
5805 				break;
5806 			}
5807 			/*
5808 			 * Update ipha_dst to be the first hop and remove the
5809 			 * first hop from the source route (by overwriting
5810 			 * part of the option with NOP options).
5811 			 */
5812 			ipha->ipha_dst = dst;
5813 			/* Put the last entry in dst */
5814 			off = ((optlen - IP_ADDR_LEN - 3) & ~(IP_ADDR_LEN-1)) +
5815 			    3;
5816 			bcopy(&opt[off], &dst, IP_ADDR_LEN);
5817 
5818 			ip1dbg(("ip_massage_options: last hop 0x%x\n",
5819 			    ntohl(dst)));
5820 			/* Move down and overwrite */
5821 			opt[IP_ADDR_LEN] = opt[0];
5822 			opt[IP_ADDR_LEN+1] = opt[IPOPT_OLEN] - IP_ADDR_LEN;
5823 			opt[IP_ADDR_LEN+2] = opt[IPOPT_OFFSET];
5824 			for (i = 0; i < IP_ADDR_LEN; i++)
5825 				opt[i] = IPOPT_NOP;
5826 			break;
5827 		}
5828 	}
5829 	return (dst);
5830 }
5831 
5832 /*
5833  * Return the network mask
5834  * associated with the specified address.
5835  */
5836 ipaddr_t
ip_net_mask(ipaddr_t addr)5837 ip_net_mask(ipaddr_t addr)
5838 {
5839 	uchar_t	*up = (uchar_t *)&addr;
5840 	ipaddr_t mask = 0;
5841 	uchar_t	*maskp = (uchar_t *)&mask;
5842 
5843 #if defined(__x86)
5844 #define	TOTALLY_BRAIN_DAMAGED_C_COMPILER
5845 #endif
5846 #ifdef  TOTALLY_BRAIN_DAMAGED_C_COMPILER
5847 	maskp[0] = maskp[1] = maskp[2] = maskp[3] = 0;
5848 #endif
5849 	if (CLASSD(addr)) {
5850 		maskp[0] = 0xF0;
5851 		return (mask);
5852 	}
5853 
5854 	/* We assume Class E default netmask to be 32 */
5855 	if (CLASSE(addr))
5856 		return (0xffffffffU);
5857 
5858 	if (addr == 0)
5859 		return (0);
5860 	maskp[0] = 0xFF;
5861 	if ((up[0] & 0x80) == 0)
5862 		return (mask);
5863 
5864 	maskp[1] = 0xFF;
5865 	if ((up[0] & 0xC0) == 0x80)
5866 		return (mask);
5867 
5868 	maskp[2] = 0xFF;
5869 	if ((up[0] & 0xE0) == 0xC0)
5870 		return (mask);
5871 
5872 	/* Otherwise return no mask */
5873 	return ((ipaddr_t)0);
5874 }
5875 
5876 /* Name/Value Table Lookup Routine */
5877 char *
ip_nv_lookup(nv_t * nv,int value)5878 ip_nv_lookup(nv_t *nv, int value)
5879 {
5880 	if (!nv)
5881 		return (NULL);
5882 	for (; nv->nv_name; nv++) {
5883 		if (nv->nv_value == value)
5884 			return (nv->nv_name);
5885 	}
5886 	return ("unknown");
5887 }
5888 
5889 static int
ip_wait_for_info_ack(ill_t * ill)5890 ip_wait_for_info_ack(ill_t *ill)
5891 {
5892 	int err;
5893 
5894 	mutex_enter(&ill->ill_lock);
5895 	while (ill->ill_state_flags & ILL_LL_SUBNET_PENDING) {
5896 		/*
5897 		 * Return value of 0 indicates a pending signal.
5898 		 */
5899 		err = cv_wait_sig(&ill->ill_cv, &ill->ill_lock);
5900 		if (err == 0) {
5901 			mutex_exit(&ill->ill_lock);
5902 			return (EINTR);
5903 		}
5904 	}
5905 	mutex_exit(&ill->ill_lock);
5906 	/*
5907 	 * ip_rput_other could have set an error  in ill_error on
5908 	 * receipt of M_ERROR.
5909 	 */
5910 	return (ill->ill_error);
5911 }
5912 
5913 /*
5914  * This is a module open, i.e. this is a control stream for access
5915  * to a DLPI device.  We allocate an ill_t as the instance data in
5916  * this case.
5917  */
5918 static int
ip_modopen(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp)5919 ip_modopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5920 {
5921 	ill_t	*ill;
5922 	int	err;
5923 	zoneid_t zoneid;
5924 	netstack_t *ns;
5925 	ip_stack_t *ipst;
5926 
5927 	/*
5928 	 * Prevent unprivileged processes from pushing IP so that
5929 	 * they can't send raw IP.
5930 	 */
5931 	if (secpolicy_net_rawaccess(credp) != 0)
5932 		return (EPERM);
5933 
5934 	ns = netstack_find_by_cred(credp);
5935 	ASSERT(ns != NULL);
5936 	ipst = ns->netstack_ip;
5937 	ASSERT(ipst != NULL);
5938 
5939 	/*
5940 	 * For exclusive stacks we set the zoneid to zero
5941 	 * to make IP operate as if in the global zone.
5942 	 */
5943 	if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5944 		zoneid = GLOBAL_ZONEID;
5945 	else
5946 		zoneid = crgetzoneid(credp);
5947 
5948 	ill = (ill_t *)mi_open_alloc_sleep(sizeof (ill_t));
5949 	q->q_ptr = WR(q)->q_ptr = ill;
5950 	ill->ill_ipst = ipst;
5951 	ill->ill_zoneid = zoneid;
5952 
5953 	/*
5954 	 * ill_init initializes the ill fields and then sends down
5955 	 * down a DL_INFO_REQ after calling qprocson.
5956 	 */
5957 	err = ill_init(q, ill);
5958 
5959 	if (err != 0) {
5960 		mi_free(ill);
5961 		netstack_rele(ipst->ips_netstack);
5962 		q->q_ptr = NULL;
5963 		WR(q)->q_ptr = NULL;
5964 		return (err);
5965 	}
5966 
5967 	/*
5968 	 * Wait for the DL_INFO_ACK if a DL_INFO_REQ was sent.
5969 	 *
5970 	 * ill_init initializes the ipsq marking this thread as
5971 	 * writer
5972 	 */
5973 	ipsq_exit(ill->ill_phyint->phyint_ipsq);
5974 	err = ip_wait_for_info_ack(ill);
5975 	if (err == 0)
5976 		ill->ill_credp = credp;
5977 	else
5978 		goto fail;
5979 
5980 	crhold(credp);
5981 
5982 	mutex_enter(&ipst->ips_ip_mi_lock);
5983 	err = mi_open_link(&ipst->ips_ip_g_head, (IDP)q->q_ptr, devp, flag,
5984 	    sflag, credp);
5985 	mutex_exit(&ipst->ips_ip_mi_lock);
5986 fail:
5987 	if (err) {
5988 		(void) ip_close(q, 0, credp);
5989 		return (err);
5990 	}
5991 	return (0);
5992 }
5993 
5994 /* For /dev/ip aka AF_INET open */
5995 int
ip_openv4(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp)5996 ip_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5997 {
5998 	return (ip_open(q, devp, flag, sflag, credp, B_FALSE));
5999 }
6000 
6001 /* For /dev/ip6 aka AF_INET6 open */
6002 int
ip_openv6(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp)6003 ip_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
6004 {
6005 	return (ip_open(q, devp, flag, sflag, credp, B_TRUE));
6006 }
6007 
6008 /* IP open routine. */
6009 int
ip_open(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp,boolean_t isv6)6010 ip_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
6011     boolean_t isv6)
6012 {
6013 	conn_t		*connp;
6014 	major_t		maj;
6015 	zoneid_t	zoneid;
6016 	netstack_t	*ns;
6017 	ip_stack_t	*ipst;
6018 
6019 	/* Allow reopen. */
6020 	if (q->q_ptr != NULL)
6021 		return (0);
6022 
6023 	if (sflag & MODOPEN) {
6024 		/* This is a module open */
6025 		return (ip_modopen(q, devp, flag, sflag, credp));
6026 	}
6027 
6028 	if ((flag & ~(FKLYR)) == IP_HELPER_STR) {
6029 		/*
6030 		 * Non streams based socket looking for a stream
6031 		 * to access IP
6032 		 */
6033 		return (ip_helper_stream_setup(q, devp, flag, sflag,
6034 		    credp, isv6));
6035 	}
6036 
6037 	ns = netstack_find_by_cred(credp);
6038 	ASSERT(ns != NULL);
6039 	ipst = ns->netstack_ip;
6040 	ASSERT(ipst != NULL);
6041 
6042 	/*
6043 	 * For exclusive stacks we set the zoneid to zero
6044 	 * to make IP operate as if in the global zone.
6045 	 */
6046 	if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
6047 		zoneid = GLOBAL_ZONEID;
6048 	else
6049 		zoneid = crgetzoneid(credp);
6050 
6051 	/*
6052 	 * We are opening as a device. This is an IP client stream, and we
6053 	 * allocate an conn_t as the instance data.
6054 	 */
6055 	connp = ipcl_conn_create(IPCL_IPCCONN, KM_SLEEP, ipst->ips_netstack);
6056 
6057 	/*
6058 	 * ipcl_conn_create did a netstack_hold. Undo the hold that was
6059 	 * done by netstack_find_by_cred()
6060 	 */
6061 	netstack_rele(ipst->ips_netstack);
6062 
6063 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
6064 	/* conn_allzones can not be set this early, hence no IPCL_ZONEID */
6065 	connp->conn_ixa->ixa_zoneid = zoneid;
6066 	connp->conn_zoneid = zoneid;
6067 
6068 	connp->conn_rq = q;
6069 	q->q_ptr = WR(q)->q_ptr = connp;
6070 
6071 	/* Minor tells us which /dev entry was opened */
6072 	if (isv6) {
6073 		connp->conn_family = AF_INET6;
6074 		connp->conn_ipversion = IPV6_VERSION;
6075 		connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
6076 		connp->conn_ixa->ixa_src_preferences = IPV6_PREFER_SRC_DEFAULT;
6077 	} else {
6078 		connp->conn_family = AF_INET;
6079 		connp->conn_ipversion = IPV4_VERSION;
6080 		connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
6081 	}
6082 
6083 	if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
6084 	    ((connp->conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
6085 		connp->conn_minor_arena = ip_minor_arena_la;
6086 	} else {
6087 		/*
6088 		 * Either minor numbers in the large arena were exhausted
6089 		 * or a non socket application is doing the open.
6090 		 * Try to allocate from the small arena.
6091 		 */
6092 		if ((connp->conn_dev =
6093 		    inet_minor_alloc(ip_minor_arena_sa)) == 0) {
6094 			/* CONN_DEC_REF takes care of netstack_rele() */
6095 			q->q_ptr = WR(q)->q_ptr = NULL;
6096 			CONN_DEC_REF(connp);
6097 			return (EBUSY);
6098 		}
6099 		connp->conn_minor_arena = ip_minor_arena_sa;
6100 	}
6101 
6102 	maj = getemajor(*devp);
6103 	*devp = makedevice(maj, (minor_t)connp->conn_dev);
6104 
6105 	/*
6106 	 * connp->conn_cred is crfree()ed in ipcl_conn_destroy()
6107 	 */
6108 	connp->conn_cred = credp;
6109 	connp->conn_cpid = curproc->p_pid;
6110 	/* Cache things in ixa without an extra refhold */
6111 	ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
6112 	connp->conn_ixa->ixa_cred = connp->conn_cred;
6113 	connp->conn_ixa->ixa_cpid = connp->conn_cpid;
6114 	if (is_system_labeled())
6115 		connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
6116 
6117 	/*
6118 	 * Handle IP_IOC_RTS_REQUEST and other ioctls which use conn_recv
6119 	 */
6120 	connp->conn_recv = ip_conn_input;
6121 	connp->conn_recvicmp = ip_conn_input_icmp;
6122 
6123 	crhold(connp->conn_cred);
6124 
6125 	/*
6126 	 * If the caller has the process-wide flag set, then default to MAC
6127 	 * exempt mode.  This allows read-down to unlabeled hosts.
6128 	 */
6129 	if (getpflags(NET_MAC_AWARE, credp) != 0)
6130 		connp->conn_mac_mode = CONN_MAC_AWARE;
6131 
6132 	connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
6133 
6134 	connp->conn_rq = q;
6135 	connp->conn_wq = WR(q);
6136 
6137 	/* Non-zero default values */
6138 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP;
6139 
6140 	/*
6141 	 * Make the conn globally visible to walkers
6142 	 */
6143 	ASSERT(connp->conn_ref == 1);
6144 	mutex_enter(&connp->conn_lock);
6145 	connp->conn_state_flags &= ~CONN_INCIPIENT;
6146 	mutex_exit(&connp->conn_lock);
6147 
6148 	qprocson(q);
6149 
6150 	return (0);
6151 }
6152 
6153 /*
6154  * Set IPsec policy from an ipsec_req_t. If the req is not "zero" and valid,
6155  * all of them are copied to the conn_t. If the req is "zero", the policy is
6156  * zeroed out. A "zero" policy has zero ipsr_{ah,req,self_encap}_req
6157  * fields.
6158  * We keep only the latest setting of the policy and thus policy setting
6159  * is not incremental/cumulative.
6160  *
6161  * Requests to set policies with multiple alternative actions will
6162  * go through a different API.
6163  */
6164 int
ipsec_set_req(cred_t * cr,conn_t * connp,ipsec_req_t * req)6165 ipsec_set_req(cred_t *cr, conn_t *connp, ipsec_req_t *req)
6166 {
6167 	uint_t ah_req = 0;
6168 	uint_t esp_req = 0;
6169 	uint_t se_req = 0;
6170 	ipsec_act_t *actp = NULL;
6171 	uint_t nact;
6172 	ipsec_policy_head_t *ph;
6173 	boolean_t is_pol_reset, is_pol_inserted = B_FALSE;
6174 	int error = 0;
6175 	netstack_t	*ns = connp->conn_netstack;
6176 	ip_stack_t	*ipst = ns->netstack_ip;
6177 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
6178 
6179 #define	REQ_MASK (IPSEC_PREF_REQUIRED|IPSEC_PREF_NEVER)
6180 
6181 	/*
6182 	 * The IP_SEC_OPT option does not allow variable length parameters,
6183 	 * hence a request cannot be NULL.
6184 	 */
6185 	if (req == NULL)
6186 		return (EINVAL);
6187 
6188 	ah_req = req->ipsr_ah_req;
6189 	esp_req = req->ipsr_esp_req;
6190 	se_req = req->ipsr_self_encap_req;
6191 
6192 	/* Don't allow setting self-encap without one or more of AH/ESP. */
6193 	if (se_req != 0 && esp_req == 0 && ah_req == 0)
6194 		return (EINVAL);
6195 
6196 	/*
6197 	 * Are we dealing with a request to reset the policy (i.e.
6198 	 * zero requests).
6199 	 */
6200 	is_pol_reset = ((ah_req & REQ_MASK) == 0 &&
6201 	    (esp_req & REQ_MASK) == 0 &&
6202 	    (se_req & REQ_MASK) == 0);
6203 
6204 	if (!is_pol_reset) {
6205 		/*
6206 		 * If we couldn't load IPsec, fail with "protocol
6207 		 * not supported".
6208 		 * IPsec may not have been loaded for a request with zero
6209 		 * policies, so we don't fail in this case.
6210 		 */
6211 		mutex_enter(&ipss->ipsec_loader_lock);
6212 		if (ipss->ipsec_loader_state != IPSEC_LOADER_SUCCEEDED) {
6213 			mutex_exit(&ipss->ipsec_loader_lock);
6214 			return (EPROTONOSUPPORT);
6215 		}
6216 		mutex_exit(&ipss->ipsec_loader_lock);
6217 
6218 		/*
6219 		 * Test for valid requests. Invalid algorithms
6220 		 * need to be tested by IPsec code because new
6221 		 * algorithms can be added dynamically.
6222 		 */
6223 		if ((ah_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6224 		    (esp_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6225 		    (se_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0) {
6226 			return (EINVAL);
6227 		}
6228 
6229 		/*
6230 		 * Only privileged users can issue these
6231 		 * requests.
6232 		 */
6233 		if (((ah_req & IPSEC_PREF_NEVER) ||
6234 		    (esp_req & IPSEC_PREF_NEVER) ||
6235 		    (se_req & IPSEC_PREF_NEVER)) &&
6236 		    secpolicy_ip_config(cr, B_FALSE) != 0) {
6237 			return (EPERM);
6238 		}
6239 
6240 		/*
6241 		 * The IPSEC_PREF_REQUIRED and IPSEC_PREF_NEVER
6242 		 * are mutually exclusive.
6243 		 */
6244 		if (((ah_req & REQ_MASK) == REQ_MASK) ||
6245 		    ((esp_req & REQ_MASK) == REQ_MASK) ||
6246 		    ((se_req & REQ_MASK) == REQ_MASK)) {
6247 			/* Both of them are set */
6248 			return (EINVAL);
6249 		}
6250 	}
6251 
6252 	ASSERT(MUTEX_HELD(&connp->conn_lock));
6253 
6254 	/*
6255 	 * If we have already cached policies in conn_connect(), don't
6256 	 * let them change now. We cache policies for connections
6257 	 * whose src,dst [addr, port] is known.
6258 	 */
6259 	if (connp->conn_policy_cached) {
6260 		return (EINVAL);
6261 	}
6262 
6263 	/*
6264 	 * We have a zero policies, reset the connection policy if already
6265 	 * set. This will cause the connection to inherit the
6266 	 * global policy, if any.
6267 	 */
6268 	if (is_pol_reset) {
6269 		if (connp->conn_policy != NULL) {
6270 			IPPH_REFRELE(connp->conn_policy, ipst->ips_netstack);
6271 			connp->conn_policy = NULL;
6272 		}
6273 		connp->conn_in_enforce_policy = B_FALSE;
6274 		connp->conn_out_enforce_policy = B_FALSE;
6275 		return (0);
6276 	}
6277 
6278 	ph = connp->conn_policy = ipsec_polhead_split(connp->conn_policy,
6279 	    ipst->ips_netstack);
6280 	if (ph == NULL)
6281 		goto enomem;
6282 
6283 	ipsec_actvec_from_req(req, &actp, &nact, ipst->ips_netstack);
6284 	if (actp == NULL)
6285 		goto enomem;
6286 
6287 	/*
6288 	 * Always insert IPv4 policy entries, since they can also apply to
6289 	 * ipv6 sockets being used in ipv4-compat mode.
6290 	 */
6291 	if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6292 	    IPSEC_TYPE_INBOUND, ns))
6293 		goto enomem;
6294 	is_pol_inserted = B_TRUE;
6295 	if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6296 	    IPSEC_TYPE_OUTBOUND, ns))
6297 		goto enomem;
6298 
6299 	/*
6300 	 * We're looking at a v6 socket, also insert the v6-specific
6301 	 * entries.
6302 	 */
6303 	if (connp->conn_family == AF_INET6) {
6304 		if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6305 		    IPSEC_TYPE_INBOUND, ns))
6306 			goto enomem;
6307 		if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6308 		    IPSEC_TYPE_OUTBOUND, ns))
6309 			goto enomem;
6310 	}
6311 
6312 	ipsec_actvec_free(actp, nact);
6313 
6314 	/*
6315 	 * If the requests need security, set enforce_policy.
6316 	 * If the requests are IPSEC_PREF_NEVER, one should
6317 	 * still set conn_out_enforce_policy so that ip_set_destination
6318 	 * marks the ip_xmit_attr_t appropriatly. This is needed so that
6319 	 * for connections that we don't cache policy in at connect time,
6320 	 * if global policy matches in ip_output_attach_policy, we
6321 	 * don't wrongly inherit global policy. Similarly, we need
6322 	 * to set conn_in_enforce_policy also so that we don't verify
6323 	 * policy wrongly.
6324 	 */
6325 	if ((ah_req & REQ_MASK) != 0 ||
6326 	    (esp_req & REQ_MASK) != 0 ||
6327 	    (se_req & REQ_MASK) != 0) {
6328 		connp->conn_in_enforce_policy = B_TRUE;
6329 		connp->conn_out_enforce_policy = B_TRUE;
6330 	}
6331 
6332 	return (error);
6333 #undef REQ_MASK
6334 
6335 	/*
6336 	 * Common memory-allocation-failure exit path.
6337 	 */
6338 enomem:
6339 	if (actp != NULL)
6340 		ipsec_actvec_free(actp, nact);
6341 	if (is_pol_inserted)
6342 		ipsec_polhead_flush(ph, ns);
6343 	return (ENOMEM);
6344 }
6345 
6346 /*
6347  * Set socket options for joining and leaving multicast groups.
6348  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6349  * The caller has already check that the option name is consistent with
6350  * the address family of the socket.
6351  */
6352 int
ip_opt_set_multicast_group(conn_t * connp,t_scalar_t name,uchar_t * invalp,boolean_t inet6,boolean_t checkonly)6353 ip_opt_set_multicast_group(conn_t *connp, t_scalar_t name,
6354     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6355 {
6356 	int		*i1 = (int *)invalp;
6357 	int		error = 0;
6358 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
6359 	struct ip_mreq	*v4_mreqp;
6360 	struct ipv6_mreq *v6_mreqp;
6361 	struct group_req *greqp;
6362 	ire_t *ire;
6363 	boolean_t done = B_FALSE;
6364 	ipaddr_t ifaddr;
6365 	in6_addr_t v6group;
6366 	uint_t ifindex;
6367 	boolean_t mcast_opt = B_TRUE;
6368 	mcast_record_t fmode;
6369 	int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6370 	    ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6371 
6372 	switch (name) {
6373 	case IP_ADD_MEMBERSHIP:
6374 	case IPV6_JOIN_GROUP:
6375 		mcast_opt = B_FALSE;
6376 		/* FALLTHROUGH */
6377 	case MCAST_JOIN_GROUP:
6378 		fmode = MODE_IS_EXCLUDE;
6379 		optfn = ip_opt_add_group;
6380 		break;
6381 
6382 	case IP_DROP_MEMBERSHIP:
6383 	case IPV6_LEAVE_GROUP:
6384 		mcast_opt = B_FALSE;
6385 		/* FALLTHROUGH */
6386 	case MCAST_LEAVE_GROUP:
6387 		fmode = MODE_IS_INCLUDE;
6388 		optfn = ip_opt_delete_group;
6389 		break;
6390 	default:
6391 		/* Should not be reached. */
6392 		fmode = MODE_IS_INCLUDE;
6393 		optfn = NULL;
6394 		ASSERT(0);
6395 	}
6396 
6397 	if (mcast_opt) {
6398 		struct sockaddr_in *sin;
6399 		struct sockaddr_in6 *sin6;
6400 
6401 		greqp = (struct group_req *)i1;
6402 		if (greqp->gr_group.ss_family == AF_INET) {
6403 			sin = (struct sockaddr_in *)&(greqp->gr_group);
6404 			IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &v6group);
6405 		} else {
6406 			if (!inet6)
6407 				return (EINVAL);	/* Not on INET socket */
6408 
6409 			sin6 = (struct sockaddr_in6 *)&(greqp->gr_group);
6410 			v6group = sin6->sin6_addr;
6411 		}
6412 		ifaddr = INADDR_ANY;
6413 		ifindex = greqp->gr_interface;
6414 	} else if (inet6) {
6415 		v6_mreqp = (struct ipv6_mreq *)i1;
6416 		v6group = v6_mreqp->ipv6mr_multiaddr;
6417 		ifaddr = INADDR_ANY;
6418 		ifindex = v6_mreqp->ipv6mr_interface;
6419 	} else {
6420 		v4_mreqp = (struct ip_mreq *)i1;
6421 		IN6_INADDR_TO_V4MAPPED(&v4_mreqp->imr_multiaddr, &v6group);
6422 		ifaddr = (ipaddr_t)v4_mreqp->imr_interface.s_addr;
6423 		ifindex = 0;
6424 	}
6425 
6426 	/*
6427 	 * In the multirouting case, we need to replicate
6428 	 * the request on all interfaces that will take part
6429 	 * in replication.  We do so because multirouting is
6430 	 * reflective, thus we will probably receive multi-
6431 	 * casts on those interfaces.
6432 	 * The ip_multirt_apply_membership() succeeds if
6433 	 * the operation succeeds on at least one interface.
6434 	 */
6435 	if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6436 		ipaddr_t group;
6437 
6438 		IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6439 
6440 		ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6441 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6442 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6443 	} else {
6444 		ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6445 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6446 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6447 	}
6448 	if (ire != NULL) {
6449 		if (ire->ire_flags & RTF_MULTIRT) {
6450 			error = ip_multirt_apply_membership(optfn, ire, connp,
6451 			    checkonly, &v6group, fmode, &ipv6_all_zeros);
6452 			done = B_TRUE;
6453 		}
6454 		ire_refrele(ire);
6455 	}
6456 
6457 	if (!done) {
6458 		error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6459 		    fmode, &ipv6_all_zeros);
6460 	}
6461 	return (error);
6462 }
6463 
6464 /*
6465  * Set socket options for joining and leaving multicast groups
6466  * for specific sources.
6467  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6468  * The caller has already check that the option name is consistent with
6469  * the address family of the socket.
6470  */
6471 int
ip_opt_set_multicast_sources(conn_t * connp,t_scalar_t name,uchar_t * invalp,boolean_t inet6,boolean_t checkonly)6472 ip_opt_set_multicast_sources(conn_t *connp, t_scalar_t name,
6473     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6474 {
6475 	int		*i1 = (int *)invalp;
6476 	int		error = 0;
6477 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
6478 	struct ip_mreq_source *imreqp;
6479 	struct group_source_req *gsreqp;
6480 	in6_addr_t v6group, v6src;
6481 	uint32_t ifindex;
6482 	ipaddr_t ifaddr;
6483 	boolean_t mcast_opt = B_TRUE;
6484 	mcast_record_t fmode;
6485 	ire_t *ire;
6486 	boolean_t done = B_FALSE;
6487 	int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6488 	    ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6489 
6490 	switch (name) {
6491 	case IP_BLOCK_SOURCE:
6492 		mcast_opt = B_FALSE;
6493 		/* FALLTHROUGH */
6494 	case MCAST_BLOCK_SOURCE:
6495 		fmode = MODE_IS_EXCLUDE;
6496 		optfn = ip_opt_add_group;
6497 		break;
6498 
6499 	case IP_UNBLOCK_SOURCE:
6500 		mcast_opt = B_FALSE;
6501 		/* FALLTHROUGH */
6502 	case MCAST_UNBLOCK_SOURCE:
6503 		fmode = MODE_IS_EXCLUDE;
6504 		optfn = ip_opt_delete_group;
6505 		break;
6506 
6507 	case IP_ADD_SOURCE_MEMBERSHIP:
6508 		mcast_opt = B_FALSE;
6509 		/* FALLTHROUGH */
6510 	case MCAST_JOIN_SOURCE_GROUP:
6511 		fmode = MODE_IS_INCLUDE;
6512 		optfn = ip_opt_add_group;
6513 		break;
6514 
6515 	case IP_DROP_SOURCE_MEMBERSHIP:
6516 		mcast_opt = B_FALSE;
6517 		/* FALLTHROUGH */
6518 	case MCAST_LEAVE_SOURCE_GROUP:
6519 		fmode = MODE_IS_INCLUDE;
6520 		optfn = ip_opt_delete_group;
6521 		break;
6522 	default:
6523 		/* Should not be reached. */
6524 		optfn = NULL;
6525 		fmode = 0;
6526 		ASSERT(0);
6527 	}
6528 
6529 	if (mcast_opt) {
6530 		gsreqp = (struct group_source_req *)i1;
6531 		ifindex = gsreqp->gsr_interface;
6532 		if (gsreqp->gsr_group.ss_family == AF_INET) {
6533 			struct sockaddr_in *s;
6534 			s = (struct sockaddr_in *)&gsreqp->gsr_group;
6535 			IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6group);
6536 			s = (struct sockaddr_in *)&gsreqp->gsr_source;
6537 			IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6src);
6538 		} else {
6539 			struct sockaddr_in6 *s6;
6540 
6541 			if (!inet6)
6542 				return (EINVAL);	/* Not on INET socket */
6543 
6544 			s6 = (struct sockaddr_in6 *)&gsreqp->gsr_group;
6545 			v6group = s6->sin6_addr;
6546 			s6 = (struct sockaddr_in6 *)&gsreqp->gsr_source;
6547 			v6src = s6->sin6_addr;
6548 		}
6549 		ifaddr = INADDR_ANY;
6550 	} else {
6551 		imreqp = (struct ip_mreq_source *)i1;
6552 		IN6_INADDR_TO_V4MAPPED(&imreqp->imr_multiaddr, &v6group);
6553 		IN6_INADDR_TO_V4MAPPED(&imreqp->imr_sourceaddr, &v6src);
6554 		ifaddr = (ipaddr_t)imreqp->imr_interface.s_addr;
6555 		ifindex = 0;
6556 	}
6557 
6558 	/*
6559 	 * Handle src being mapped INADDR_ANY by changing it to unspecified.
6560 	 */
6561 	if (IN6_IS_ADDR_V4MAPPED_ANY(&v6src))
6562 		v6src = ipv6_all_zeros;
6563 
6564 	/*
6565 	 * In the multirouting case, we need to replicate
6566 	 * the request as noted in the mcast cases above.
6567 	 */
6568 	if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6569 		ipaddr_t group;
6570 
6571 		IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6572 
6573 		ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6574 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6575 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6576 	} else {
6577 		ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6578 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6579 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6580 	}
6581 	if (ire != NULL) {
6582 		if (ire->ire_flags & RTF_MULTIRT) {
6583 			error = ip_multirt_apply_membership(optfn, ire, connp,
6584 			    checkonly, &v6group, fmode, &v6src);
6585 			done = B_TRUE;
6586 		}
6587 		ire_refrele(ire);
6588 	}
6589 	if (!done) {
6590 		error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6591 		    fmode, &v6src);
6592 	}
6593 	return (error);
6594 }
6595 
6596 /*
6597  * Given a destination address and a pointer to where to put the information
6598  * this routine fills in the mtuinfo.
6599  * The socket must be connected.
6600  * For sctp conn_faddr is the primary address.
6601  */
6602 int
ip_fill_mtuinfo(conn_t * connp,ip_xmit_attr_t * ixa,struct ip6_mtuinfo * mtuinfo)6603 ip_fill_mtuinfo(conn_t *connp, ip_xmit_attr_t *ixa, struct ip6_mtuinfo *mtuinfo)
6604 {
6605 	uint32_t	pmtu = IP_MAXPACKET;
6606 	uint_t		scopeid;
6607 
6608 	if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6))
6609 		return (-1);
6610 
6611 	/* In case we never sent or called ip_set_destination_v4/v6 */
6612 	if (ixa->ixa_ire != NULL)
6613 		pmtu = ip_get_pmtu(ixa);
6614 
6615 	if (ixa->ixa_flags & IXAF_SCOPEID_SET)
6616 		scopeid = ixa->ixa_scopeid;
6617 	else
6618 		scopeid = 0;
6619 
6620 	bzero(mtuinfo, sizeof (*mtuinfo));
6621 	mtuinfo->ip6m_addr.sin6_family = AF_INET6;
6622 	mtuinfo->ip6m_addr.sin6_port = connp->conn_fport;
6623 	mtuinfo->ip6m_addr.sin6_addr = connp->conn_faddr_v6;
6624 	mtuinfo->ip6m_addr.sin6_scope_id = scopeid;
6625 	mtuinfo->ip6m_mtu = pmtu;
6626 
6627 	return (sizeof (struct ip6_mtuinfo));
6628 }
6629 
6630 /*
6631  * When the src multihoming is changed from weak to [strong, preferred]
6632  * ip_ire_rebind_walker is called to walk the list of all ire_t entries
6633  * and identify routes that were created by user-applications in the
6634  * unbound state (i.e., without RTA_IFP), and for which an ire_ill is not
6635  * currently defined. These routes are then 'rebound', i.e., their ire_ill
6636  * is selected by finding an interface route for the gateway.
6637  */
6638 /* ARGSUSED */
6639 void
ip_ire_rebind_walker(ire_t * ire,void * notused)6640 ip_ire_rebind_walker(ire_t *ire, void *notused)
6641 {
6642 	if (!ire->ire_unbound || ire->ire_ill != NULL)
6643 		return;
6644 	ire_rebind(ire);
6645 	ire_delete(ire);
6646 }
6647 
6648 /*
6649  * When the src multihoming is changed from  [strong, preferred] to weak,
6650  * ip_ire_unbind_walker is called to walk the list of all ire_t entries, and
6651  * set any entries that were created by user-applications in the unbound state
6652  * (i.e., without RTA_IFP) back to having a NULL ire_ill.
6653  */
6654 /* ARGSUSED */
6655 void
ip_ire_unbind_walker(ire_t * ire,void * notused)6656 ip_ire_unbind_walker(ire_t *ire, void *notused)
6657 {
6658 	ire_t *new_ire;
6659 
6660 	if (!ire->ire_unbound || ire->ire_ill == NULL)
6661 		return;
6662 	if (ire->ire_ipversion == IPV6_VERSION) {
6663 		new_ire = ire_create_v6(&ire->ire_addr_v6, &ire->ire_mask_v6,
6664 		    &ire->ire_gateway_addr_v6, ire->ire_type, NULL,
6665 		    ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6666 	} else {
6667 		new_ire = ire_create((uchar_t *)&ire->ire_addr,
6668 		    (uchar_t *)&ire->ire_mask,
6669 		    (uchar_t *)&ire->ire_gateway_addr, ire->ire_type, NULL,
6670 		    ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6671 	}
6672 	if (new_ire == NULL)
6673 		return;
6674 	new_ire->ire_unbound = B_TRUE;
6675 	/*
6676 	 * The bound ire must first be deleted so that we don't return
6677 	 * the existing one on the attempt to add the unbound new_ire.
6678 	 */
6679 	ire_delete(ire);
6680 	new_ire = ire_add(new_ire);
6681 	if (new_ire != NULL)
6682 		ire_refrele(new_ire);
6683 }
6684 
6685 /*
6686  * When the settings of ip*_strict_src_multihoming tunables are changed,
6687  * all cached routes need to be recomputed. This recomputation needs to be
6688  * done when going from weaker to stronger modes so that the cached ire
6689  * for the connection does not violate the current ip*_strict_src_multihoming
6690  * setting. It also needs to be done when going from stronger to weaker modes,
6691  * so that we fall back to matching on the longest-matching-route (as opposed
6692  * to a shorter match that may have been selected in the strong mode
6693  * to satisfy src_multihoming settings).
6694  *
6695  * The cached ixa_ire entires for all conn_t entries are marked as
6696  * "verify" so that they will be recomputed for the next packet.
6697  */
6698 void
conn_ire_revalidate(conn_t * connp,void * arg)6699 conn_ire_revalidate(conn_t *connp, void *arg)
6700 {
6701 	boolean_t isv6 = (boolean_t)arg;
6702 
6703 	if ((isv6 && connp->conn_ipversion != IPV6_VERSION) ||
6704 	    (!isv6 && connp->conn_ipversion != IPV4_VERSION))
6705 		return;
6706 	connp->conn_ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
6707 }
6708 
6709 /*
6710  * Handles both IPv4 and IPv6 reassembly - doing the out-of-order cases,
6711  * When an ipf is passed here for the first time, if
6712  * we already have in-order fragments on the queue, we convert from the fast-
6713  * path reassembly scheme to the hard-case scheme.  From then on, additional
6714  * fragments are reassembled here.  We keep track of the start and end offsets
6715  * of each piece, and the number of holes in the chain.  When the hole count
6716  * goes to zero, we are done!
6717  *
6718  * The ipf_count will be updated to account for any mblk(s) added (pointed to
6719  * by mp) or subtracted (freeb()ed dups), upon return the caller must update
6720  * ipfb_count and ill_frag_count by the difference of ipf_count before and
6721  * after the call to ip_reassemble().
6722  */
6723 int
ip_reassemble(mblk_t * mp,ipf_t * ipf,uint_t start,boolean_t more,ill_t * ill,size_t msg_len)6724 ip_reassemble(mblk_t *mp, ipf_t *ipf, uint_t start, boolean_t more, ill_t *ill,
6725     size_t msg_len)
6726 {
6727 	uint_t	end;
6728 	mblk_t	*next_mp;
6729 	mblk_t	*mp1;
6730 	uint_t	offset;
6731 	boolean_t incr_dups = B_TRUE;
6732 	boolean_t offset_zero_seen = B_FALSE;
6733 	boolean_t pkt_boundary_checked = B_FALSE;
6734 
6735 	/* If start == 0 then ipf_nf_hdr_len has to be set. */
6736 	ASSERT(start != 0 || ipf->ipf_nf_hdr_len != 0);
6737 
6738 	/* Add in byte count */
6739 	ipf->ipf_count += msg_len;
6740 	if (ipf->ipf_end) {
6741 		/*
6742 		 * We were part way through in-order reassembly, but now there
6743 		 * is a hole.  We walk through messages already queued, and
6744 		 * mark them for hard case reassembly.  We know that up till
6745 		 * now they were in order starting from offset zero.
6746 		 */
6747 		offset = 0;
6748 		for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6749 			IP_REASS_SET_START(mp1, offset);
6750 			if (offset == 0) {
6751 				ASSERT(ipf->ipf_nf_hdr_len != 0);
6752 				offset = -ipf->ipf_nf_hdr_len;
6753 			}
6754 			offset += mp1->b_wptr - mp1->b_rptr;
6755 			IP_REASS_SET_END(mp1, offset);
6756 		}
6757 		/* One hole at the end. */
6758 		ipf->ipf_hole_cnt = 1;
6759 		/* Brand it as a hard case, forever. */
6760 		ipf->ipf_end = 0;
6761 	}
6762 	/* Walk through all the new pieces. */
6763 	do {
6764 		end = start + (mp->b_wptr - mp->b_rptr);
6765 		/*
6766 		 * If start is 0, decrease 'end' only for the first mblk of
6767 		 * the fragment. Otherwise 'end' can get wrong value in the
6768 		 * second pass of the loop if first mblk is exactly the
6769 		 * size of ipf_nf_hdr_len.
6770 		 */
6771 		if (start == 0 && !offset_zero_seen) {
6772 			/* First segment */
6773 			ASSERT(ipf->ipf_nf_hdr_len != 0);
6774 			end -= ipf->ipf_nf_hdr_len;
6775 			offset_zero_seen = B_TRUE;
6776 		}
6777 		next_mp = mp->b_cont;
6778 		/*
6779 		 * We are checking to see if there is any interesing data
6780 		 * to process.  If there isn't and the mblk isn't the
6781 		 * one which carries the unfragmentable header then we
6782 		 * drop it.  It's possible to have just the unfragmentable
6783 		 * header come through without any data.  That needs to be
6784 		 * saved.
6785 		 *
6786 		 * If the assert at the top of this function holds then the
6787 		 * term "ipf->ipf_nf_hdr_len != 0" isn't needed.  This code
6788 		 * is infrequently traveled enough that the test is left in
6789 		 * to protect against future code changes which break that
6790 		 * invariant.
6791 		 */
6792 		if (start == end && start != 0 && ipf->ipf_nf_hdr_len != 0) {
6793 			/* Empty.  Blast it. */
6794 			IP_REASS_SET_START(mp, 0);
6795 			IP_REASS_SET_END(mp, 0);
6796 			/*
6797 			 * If the ipf points to the mblk we are about to free,
6798 			 * update ipf to point to the next mblk (or NULL
6799 			 * if none).
6800 			 */
6801 			if (ipf->ipf_mp->b_cont == mp)
6802 				ipf->ipf_mp->b_cont = next_mp;
6803 			freeb(mp);
6804 			continue;
6805 		}
6806 		mp->b_cont = NULL;
6807 		IP_REASS_SET_START(mp, start);
6808 		IP_REASS_SET_END(mp, end);
6809 		if (!ipf->ipf_tail_mp) {
6810 			ipf->ipf_tail_mp = mp;
6811 			ipf->ipf_mp->b_cont = mp;
6812 			if (start == 0 || !more) {
6813 				ipf->ipf_hole_cnt = 1;
6814 				/*
6815 				 * if the first fragment comes in more than one
6816 				 * mblk, this loop will be executed for each
6817 				 * mblk. Need to adjust hole count so exiting
6818 				 * this routine will leave hole count at 1.
6819 				 */
6820 				if (next_mp)
6821 					ipf->ipf_hole_cnt++;
6822 			} else
6823 				ipf->ipf_hole_cnt = 2;
6824 			continue;
6825 		} else if (ipf->ipf_last_frag_seen && !more &&
6826 		    !pkt_boundary_checked) {
6827 			/*
6828 			 * We check datagram boundary only if this fragment
6829 			 * claims to be the last fragment and we have seen a
6830 			 * last fragment in the past too. We do this only
6831 			 * once for a given fragment.
6832 			 *
6833 			 * start cannot be 0 here as fragments with start=0
6834 			 * and MF=0 gets handled as a complete packet. These
6835 			 * fragments should not reach here.
6836 			 */
6837 
6838 			if (start + msgdsize(mp) !=
6839 			    IP_REASS_END(ipf->ipf_tail_mp)) {
6840 				/*
6841 				 * We have two fragments both of which claim
6842 				 * to be the last fragment but gives conflicting
6843 				 * information about the whole datagram size.
6844 				 * Something fishy is going on. Drop the
6845 				 * fragment and free up the reassembly list.
6846 				 */
6847 				return (IP_REASS_FAILED);
6848 			}
6849 
6850 			/*
6851 			 * We shouldn't come to this code block again for this
6852 			 * particular fragment.
6853 			 */
6854 			pkt_boundary_checked = B_TRUE;
6855 		}
6856 
6857 		/* New stuff at or beyond tail? */
6858 		offset = IP_REASS_END(ipf->ipf_tail_mp);
6859 		if (start >= offset) {
6860 			if (ipf->ipf_last_frag_seen) {
6861 				/* current fragment is beyond last fragment */
6862 				return (IP_REASS_FAILED);
6863 			}
6864 			/* Link it on end. */
6865 			ipf->ipf_tail_mp->b_cont = mp;
6866 			ipf->ipf_tail_mp = mp;
6867 			if (more) {
6868 				if (start != offset)
6869 					ipf->ipf_hole_cnt++;
6870 			} else if (start == offset && next_mp == NULL)
6871 					ipf->ipf_hole_cnt--;
6872 			continue;
6873 		}
6874 		mp1 = ipf->ipf_mp->b_cont;
6875 		offset = IP_REASS_START(mp1);
6876 		/* New stuff at the front? */
6877 		if (start < offset) {
6878 			if (start == 0) {
6879 				if (end >= offset) {
6880 					/* Nailed the hole at the begining. */
6881 					ipf->ipf_hole_cnt--;
6882 				}
6883 			} else if (end < offset) {
6884 				/*
6885 				 * A hole, stuff, and a hole where there used
6886 				 * to be just a hole.
6887 				 */
6888 				ipf->ipf_hole_cnt++;
6889 			}
6890 			mp->b_cont = mp1;
6891 			/* Check for overlap. */
6892 			while (end > offset) {
6893 				if (end < IP_REASS_END(mp1)) {
6894 					mp->b_wptr -= end - offset;
6895 					IP_REASS_SET_END(mp, offset);
6896 					BUMP_MIB(ill->ill_ip_mib,
6897 					    ipIfStatsReasmPartDups);
6898 					break;
6899 				}
6900 				/* Did we cover another hole? */
6901 				if ((mp1->b_cont &&
6902 				    IP_REASS_END(mp1) !=
6903 				    IP_REASS_START(mp1->b_cont) &&
6904 				    end >= IP_REASS_START(mp1->b_cont)) ||
6905 				    (!ipf->ipf_last_frag_seen && !more)) {
6906 					ipf->ipf_hole_cnt--;
6907 				}
6908 				/* Clip out mp1. */
6909 				if ((mp->b_cont = mp1->b_cont) == NULL) {
6910 					/*
6911 					 * After clipping out mp1, this guy
6912 					 * is now hanging off the end.
6913 					 */
6914 					ipf->ipf_tail_mp = mp;
6915 				}
6916 				IP_REASS_SET_START(mp1, 0);
6917 				IP_REASS_SET_END(mp1, 0);
6918 				/* Subtract byte count */
6919 				ipf->ipf_count -= mp1->b_datap->db_lim -
6920 				    mp1->b_datap->db_base;
6921 				freeb(mp1);
6922 				BUMP_MIB(ill->ill_ip_mib,
6923 				    ipIfStatsReasmPartDups);
6924 				mp1 = mp->b_cont;
6925 				if (!mp1)
6926 					break;
6927 				offset = IP_REASS_START(mp1);
6928 			}
6929 			ipf->ipf_mp->b_cont = mp;
6930 			continue;
6931 		}
6932 		/*
6933 		 * The new piece starts somewhere between the start of the head
6934 		 * and before the end of the tail.
6935 		 */
6936 		for (; mp1; mp1 = mp1->b_cont) {
6937 			offset = IP_REASS_END(mp1);
6938 			if (start < offset) {
6939 				if (end <= offset) {
6940 					/* Nothing new. */
6941 					IP_REASS_SET_START(mp, 0);
6942 					IP_REASS_SET_END(mp, 0);
6943 					/* Subtract byte count */
6944 					ipf->ipf_count -= mp->b_datap->db_lim -
6945 					    mp->b_datap->db_base;
6946 					if (incr_dups) {
6947 						ipf->ipf_num_dups++;
6948 						incr_dups = B_FALSE;
6949 					}
6950 					freeb(mp);
6951 					BUMP_MIB(ill->ill_ip_mib,
6952 					    ipIfStatsReasmDuplicates);
6953 					break;
6954 				}
6955 				/*
6956 				 * Trim redundant stuff off beginning of new
6957 				 * piece.
6958 				 */
6959 				IP_REASS_SET_START(mp, offset);
6960 				mp->b_rptr += offset - start;
6961 				BUMP_MIB(ill->ill_ip_mib,
6962 				    ipIfStatsReasmPartDups);
6963 				start = offset;
6964 				if (!mp1->b_cont) {
6965 					/*
6966 					 * After trimming, this guy is now
6967 					 * hanging off the end.
6968 					 */
6969 					mp1->b_cont = mp;
6970 					ipf->ipf_tail_mp = mp;
6971 					if (!more) {
6972 						ipf->ipf_hole_cnt--;
6973 					}
6974 					break;
6975 				}
6976 			}
6977 			if (start >= IP_REASS_START(mp1->b_cont))
6978 				continue;
6979 			/* Fill a hole */
6980 			if (start > offset)
6981 				ipf->ipf_hole_cnt++;
6982 			mp->b_cont = mp1->b_cont;
6983 			mp1->b_cont = mp;
6984 			mp1 = mp->b_cont;
6985 			offset = IP_REASS_START(mp1);
6986 			if (end >= offset) {
6987 				ipf->ipf_hole_cnt--;
6988 				/* Check for overlap. */
6989 				while (end > offset) {
6990 					if (end < IP_REASS_END(mp1)) {
6991 						mp->b_wptr -= end - offset;
6992 						IP_REASS_SET_END(mp, offset);
6993 						/*
6994 						 * TODO we might bump
6995 						 * this up twice if there is
6996 						 * overlap at both ends.
6997 						 */
6998 						BUMP_MIB(ill->ill_ip_mib,
6999 						    ipIfStatsReasmPartDups);
7000 						break;
7001 					}
7002 					/* Did we cover another hole? */
7003 					if ((mp1->b_cont &&
7004 					    IP_REASS_END(mp1)
7005 					    != IP_REASS_START(mp1->b_cont) &&
7006 					    end >=
7007 					    IP_REASS_START(mp1->b_cont)) ||
7008 					    (!ipf->ipf_last_frag_seen &&
7009 					    !more)) {
7010 						ipf->ipf_hole_cnt--;
7011 					}
7012 					/* Clip out mp1. */
7013 					if ((mp->b_cont = mp1->b_cont) ==
7014 					    NULL) {
7015 						/*
7016 						 * After clipping out mp1,
7017 						 * this guy is now hanging
7018 						 * off the end.
7019 						 */
7020 						ipf->ipf_tail_mp = mp;
7021 					}
7022 					IP_REASS_SET_START(mp1, 0);
7023 					IP_REASS_SET_END(mp1, 0);
7024 					/* Subtract byte count */
7025 					ipf->ipf_count -=
7026 					    mp1->b_datap->db_lim -
7027 					    mp1->b_datap->db_base;
7028 					freeb(mp1);
7029 					BUMP_MIB(ill->ill_ip_mib,
7030 					    ipIfStatsReasmPartDups);
7031 					mp1 = mp->b_cont;
7032 					if (!mp1)
7033 						break;
7034 					offset = IP_REASS_START(mp1);
7035 				}
7036 			}
7037 			break;
7038 		}
7039 	} while (start = end, mp = next_mp);
7040 
7041 	/* Fragment just processed could be the last one. Remember this fact */
7042 	if (!more)
7043 		ipf->ipf_last_frag_seen = B_TRUE;
7044 
7045 	/* Still got holes? */
7046 	if (ipf->ipf_hole_cnt)
7047 		return (IP_REASS_PARTIAL);
7048 	/* Clean up overloaded fields to avoid upstream disasters. */
7049 	for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
7050 		IP_REASS_SET_START(mp1, 0);
7051 		IP_REASS_SET_END(mp1, 0);
7052 	}
7053 	return (IP_REASS_COMPLETE);
7054 }
7055 
7056 /*
7057  * Fragmentation reassembly.  Each ILL has a hash table for
7058  * queuing packets undergoing reassembly for all IPIFs
7059  * associated with the ILL.  The hash is based on the packet
7060  * IP ident field.  The ILL frag hash table was allocated
7061  * as a timer block at the time the ILL was created.  Whenever
7062  * there is anything on the reassembly queue, the timer will
7063  * be running.  Returns the reassembled packet if reassembly completes.
7064  */
7065 mblk_t *
ip_input_fragment(mblk_t * mp,ipha_t * ipha,ip_recv_attr_t * ira)7066 ip_input_fragment(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
7067 {
7068 	uint32_t	frag_offset_flags;
7069 	mblk_t		*t_mp;
7070 	ipaddr_t	dst;
7071 	uint8_t		proto = ipha->ipha_protocol;
7072 	uint32_t	sum_val;
7073 	uint16_t	sum_flags;
7074 	ipf_t		*ipf;
7075 	ipf_t		**ipfp;
7076 	ipfb_t		*ipfb;
7077 	uint16_t	ident;
7078 	uint32_t	offset;
7079 	ipaddr_t	src;
7080 	uint_t		hdr_length;
7081 	uint32_t	end;
7082 	mblk_t		*mp1;
7083 	mblk_t		*tail_mp;
7084 	size_t		count;
7085 	size_t		msg_len;
7086 	uint8_t		ecn_info = 0;
7087 	uint32_t	packet_size;
7088 	boolean_t	pruned = B_FALSE;
7089 	ill_t		*ill = ira->ira_ill;
7090 	ip_stack_t	*ipst = ill->ill_ipst;
7091 
7092 	/*
7093 	 * Drop the fragmented as early as possible, if
7094 	 * we don't have resource(s) to re-assemble.
7095 	 */
7096 	if (ipst->ips_ip_reass_queue_bytes == 0) {
7097 		freemsg(mp);
7098 		return (NULL);
7099 	}
7100 
7101 	/* Check for fragmentation offset; return if there's none */
7102 	if ((frag_offset_flags = ntohs(ipha->ipha_fragment_offset_and_flags) &
7103 	    (IPH_MF | IPH_OFFSET)) == 0)
7104 		return (mp);
7105 
7106 	/*
7107 	 * We utilize hardware computed checksum info only for UDP since
7108 	 * IP fragmentation is a normal occurrence for the protocol.  In
7109 	 * addition, checksum offload support for IP fragments carrying
7110 	 * UDP payload is commonly implemented across network adapters.
7111 	 */
7112 	ASSERT(ira->ira_rill != NULL);
7113 	if (proto == IPPROTO_UDP && dohwcksum &&
7114 	    ILL_HCKSUM_CAPABLE(ira->ira_rill) &&
7115 	    (DB_CKSUMFLAGS(mp) & (HCK_FULLCKSUM | HCK_PARTIALCKSUM))) {
7116 		mblk_t *mp1 = mp->b_cont;
7117 		int32_t len;
7118 
7119 		/* Record checksum information from the packet */
7120 		sum_val = (uint32_t)DB_CKSUM16(mp);
7121 		sum_flags = DB_CKSUMFLAGS(mp);
7122 
7123 		/* IP payload offset from beginning of mblk */
7124 		offset = ((uchar_t *)ipha + IPH_HDR_LENGTH(ipha)) - mp->b_rptr;
7125 
7126 		if ((sum_flags & HCK_PARTIALCKSUM) &&
7127 		    (mp1 == NULL || mp1->b_cont == NULL) &&
7128 		    offset >= DB_CKSUMSTART(mp) &&
7129 		    ((len = offset - DB_CKSUMSTART(mp)) & 1) == 0) {
7130 			uint32_t adj;
7131 			/*
7132 			 * Partial checksum has been calculated by hardware
7133 			 * and attached to the packet; in addition, any
7134 			 * prepended extraneous data is even byte aligned.
7135 			 * If any such data exists, we adjust the checksum;
7136 			 * this would also handle any postpended data.
7137 			 */
7138 			IP_ADJCKSUM_PARTIAL(mp->b_rptr + DB_CKSUMSTART(mp),
7139 			    mp, mp1, len, adj);
7140 
7141 			/* One's complement subtract extraneous checksum */
7142 			if (adj >= sum_val)
7143 				sum_val = ~(adj - sum_val) & 0xFFFF;
7144 			else
7145 				sum_val -= adj;
7146 		}
7147 	} else {
7148 		sum_val = 0;
7149 		sum_flags = 0;
7150 	}
7151 
7152 	/* Clear hardware checksumming flag */
7153 	DB_CKSUMFLAGS(mp) = 0;
7154 
7155 	ident = ipha->ipha_ident;
7156 	offset = (frag_offset_flags << 3) & 0xFFFF;
7157 	src = ipha->ipha_src;
7158 	dst = ipha->ipha_dst;
7159 	hdr_length = IPH_HDR_LENGTH(ipha);
7160 	end = ntohs(ipha->ipha_length) - hdr_length;
7161 
7162 	/* If end == 0 then we have a packet with no data, so just free it */
7163 	if (end == 0) {
7164 		freemsg(mp);
7165 		return (NULL);
7166 	}
7167 
7168 	/* Record the ECN field info. */
7169 	ecn_info = (ipha->ipha_type_of_service & 0x3);
7170 	if (offset != 0) {
7171 		/*
7172 		 * If this isn't the first piece, strip the header, and
7173 		 * add the offset to the end value.
7174 		 */
7175 		mp->b_rptr += hdr_length;
7176 		end += offset;
7177 	}
7178 
7179 	/* Handle vnic loopback of fragments */
7180 	if (mp->b_datap->db_ref > 2)
7181 		msg_len = 0;
7182 	else
7183 		msg_len = MBLKSIZE(mp);
7184 
7185 	tail_mp = mp;
7186 	while (tail_mp->b_cont != NULL) {
7187 		tail_mp = tail_mp->b_cont;
7188 		if (tail_mp->b_datap->db_ref <= 2)
7189 			msg_len += MBLKSIZE(tail_mp);
7190 	}
7191 
7192 	/* If the reassembly list for this ILL will get too big, prune it */
7193 	if ((msg_len + sizeof (*ipf) + ill->ill_frag_count) >=
7194 	    ipst->ips_ip_reass_queue_bytes) {
7195 		DTRACE_PROBE3(ip_reass_queue_bytes, uint_t, msg_len,
7196 		    uint_t, ill->ill_frag_count,
7197 		    uint_t, ipst->ips_ip_reass_queue_bytes);
7198 		ill_frag_prune(ill,
7199 		    (ipst->ips_ip_reass_queue_bytes < msg_len) ? 0 :
7200 		    (ipst->ips_ip_reass_queue_bytes - msg_len));
7201 		pruned = B_TRUE;
7202 	}
7203 
7204 	ipfb = &ill->ill_frag_hash_tbl[ILL_FRAG_HASH(src, ident)];
7205 	mutex_enter(&ipfb->ipfb_lock);
7206 
7207 	ipfp = &ipfb->ipfb_ipf;
7208 	/* Try to find an existing fragment queue for this packet. */
7209 	for (;;) {
7210 		ipf = ipfp[0];
7211 		if (ipf != NULL) {
7212 			/*
7213 			 * It has to match on ident and src/dst address.
7214 			 */
7215 			if (ipf->ipf_ident == ident &&
7216 			    ipf->ipf_src == src &&
7217 			    ipf->ipf_dst == dst &&
7218 			    ipf->ipf_protocol == proto) {
7219 				/*
7220 				 * If we have received too many
7221 				 * duplicate fragments for this packet
7222 				 * free it.
7223 				 */
7224 				if (ipf->ipf_num_dups > ip_max_frag_dups) {
7225 					ill_frag_free_pkts(ill, ipfb, ipf, 1);
7226 					freemsg(mp);
7227 					mutex_exit(&ipfb->ipfb_lock);
7228 					return (NULL);
7229 				}
7230 				/* Found it. */
7231 				break;
7232 			}
7233 			ipfp = &ipf->ipf_hash_next;
7234 			continue;
7235 		}
7236 
7237 		/*
7238 		 * If we pruned the list, do we want to store this new
7239 		 * fragment?. We apply an optimization here based on the
7240 		 * fact that most fragments will be received in order.
7241 		 * So if the offset of this incoming fragment is zero,
7242 		 * it is the first fragment of a new packet. We will
7243 		 * keep it.  Otherwise drop the fragment, as we have
7244 		 * probably pruned the packet already (since the
7245 		 * packet cannot be found).
7246 		 */
7247 		if (pruned && offset != 0) {
7248 			mutex_exit(&ipfb->ipfb_lock);
7249 			freemsg(mp);
7250 			return (NULL);
7251 		}
7252 
7253 		if (ipfb->ipfb_frag_pkts >= MAX_FRAG_PKTS(ipst))  {
7254 			/*
7255 			 * Too many fragmented packets in this hash
7256 			 * bucket. Free the oldest.
7257 			 */
7258 			ill_frag_free_pkts(ill, ipfb, ipfb->ipfb_ipf, 1);
7259 		}
7260 
7261 		/* New guy.  Allocate a frag message. */
7262 		mp1 = allocb(sizeof (*ipf), BPRI_MED);
7263 		if (mp1 == NULL) {
7264 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7265 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7266 			freemsg(mp);
7267 reass_done:
7268 			mutex_exit(&ipfb->ipfb_lock);
7269 			return (NULL);
7270 		}
7271 
7272 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmReqds);
7273 		mp1->b_cont = mp;
7274 
7275 		/* Initialize the fragment header. */
7276 		ipf = (ipf_t *)mp1->b_rptr;
7277 		ipf->ipf_mp = mp1;
7278 		ipf->ipf_ptphn = ipfp;
7279 		ipfp[0] = ipf;
7280 		ipf->ipf_hash_next = NULL;
7281 		ipf->ipf_ident = ident;
7282 		ipf->ipf_protocol = proto;
7283 		ipf->ipf_src = src;
7284 		ipf->ipf_dst = dst;
7285 		ipf->ipf_nf_hdr_len = 0;
7286 		/* Record reassembly start time. */
7287 		ipf->ipf_timestamp = gethrestime_sec();
7288 		/* Record ipf generation and account for frag header */
7289 		ipf->ipf_gen = ill->ill_ipf_gen++;
7290 		ipf->ipf_count = MBLKSIZE(mp1);
7291 		ipf->ipf_last_frag_seen = B_FALSE;
7292 		ipf->ipf_ecn = ecn_info;
7293 		ipf->ipf_num_dups = 0;
7294 		ipfb->ipfb_frag_pkts++;
7295 		ipf->ipf_checksum = 0;
7296 		ipf->ipf_checksum_flags = 0;
7297 
7298 		/* Store checksum value in fragment header */
7299 		if (sum_flags != 0) {
7300 			sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7301 			sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7302 			ipf->ipf_checksum = sum_val;
7303 			ipf->ipf_checksum_flags = sum_flags;
7304 		}
7305 
7306 		/*
7307 		 * We handle reassembly two ways.  In the easy case,
7308 		 * where all the fragments show up in order, we do
7309 		 * minimal bookkeeping, and just clip new pieces on
7310 		 * the end.  If we ever see a hole, then we go off
7311 		 * to ip_reassemble which has to mark the pieces and
7312 		 * keep track of the number of holes, etc.  Obviously,
7313 		 * the point of having both mechanisms is so we can
7314 		 * handle the easy case as efficiently as possible.
7315 		 */
7316 		if (offset == 0) {
7317 			/* Easy case, in-order reassembly so far. */
7318 			ipf->ipf_count += msg_len;
7319 			ipf->ipf_tail_mp = tail_mp;
7320 			/*
7321 			 * Keep track of next expected offset in
7322 			 * ipf_end.
7323 			 */
7324 			ipf->ipf_end = end;
7325 			ipf->ipf_nf_hdr_len = hdr_length;
7326 		} else {
7327 			/* Hard case, hole at the beginning. */
7328 			ipf->ipf_tail_mp = NULL;
7329 			/*
7330 			 * ipf_end == 0 means that we have given up
7331 			 * on easy reassembly.
7332 			 */
7333 			ipf->ipf_end = 0;
7334 
7335 			/* Forget checksum offload from now on */
7336 			ipf->ipf_checksum_flags = 0;
7337 
7338 			/*
7339 			 * ipf_hole_cnt is set by ip_reassemble.
7340 			 * ipf_count is updated by ip_reassemble.
7341 			 * No need to check for return value here
7342 			 * as we don't expect reassembly to complete
7343 			 * or fail for the first fragment itself.
7344 			 */
7345 			(void) ip_reassemble(mp, ipf,
7346 			    (frag_offset_flags & IPH_OFFSET) << 3,
7347 			    (frag_offset_flags & IPH_MF), ill, msg_len);
7348 		}
7349 		/* Update per ipfb and ill byte counts */
7350 		ipfb->ipfb_count += ipf->ipf_count;
7351 		ASSERT(ipfb->ipfb_count > 0);	/* Wraparound */
7352 		atomic_add_32(&ill->ill_frag_count, ipf->ipf_count);
7353 		/* If the frag timer wasn't already going, start it. */
7354 		mutex_enter(&ill->ill_lock);
7355 		ill_frag_timer_start(ill);
7356 		mutex_exit(&ill->ill_lock);
7357 		goto reass_done;
7358 	}
7359 
7360 	/*
7361 	 * If the packet's flag has changed (it could be coming up
7362 	 * from an interface different than the previous, therefore
7363 	 * possibly different checksum capability), then forget about
7364 	 * any stored checksum states.  Otherwise add the value to
7365 	 * the existing one stored in the fragment header.
7366 	 */
7367 	if (sum_flags != 0 && sum_flags == ipf->ipf_checksum_flags) {
7368 		sum_val += ipf->ipf_checksum;
7369 		sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7370 		sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7371 		ipf->ipf_checksum = sum_val;
7372 	} else if (ipf->ipf_checksum_flags != 0) {
7373 		/* Forget checksum offload from now on */
7374 		ipf->ipf_checksum_flags = 0;
7375 	}
7376 
7377 	/*
7378 	 * We have a new piece of a datagram which is already being
7379 	 * reassembled.  Update the ECN info if all IP fragments
7380 	 * are ECN capable.  If there is one which is not, clear
7381 	 * all the info.  If there is at least one which has CE
7382 	 * code point, IP needs to report that up to transport.
7383 	 */
7384 	if (ecn_info != IPH_ECN_NECT && ipf->ipf_ecn != IPH_ECN_NECT) {
7385 		if (ecn_info == IPH_ECN_CE)
7386 			ipf->ipf_ecn = IPH_ECN_CE;
7387 	} else {
7388 		ipf->ipf_ecn = IPH_ECN_NECT;
7389 	}
7390 	if (offset && ipf->ipf_end == offset) {
7391 		/* The new fragment fits at the end */
7392 		ipf->ipf_tail_mp->b_cont = mp;
7393 		/* Update the byte count */
7394 		ipf->ipf_count += msg_len;
7395 		/* Update per ipfb and ill byte counts */
7396 		ipfb->ipfb_count += msg_len;
7397 		ASSERT(ipfb->ipfb_count > 0);	/* Wraparound */
7398 		atomic_add_32(&ill->ill_frag_count, msg_len);
7399 		if (frag_offset_flags & IPH_MF) {
7400 			/* More to come. */
7401 			ipf->ipf_end = end;
7402 			ipf->ipf_tail_mp = tail_mp;
7403 			goto reass_done;
7404 		}
7405 	} else {
7406 		/* Go do the hard cases. */
7407 		int ret;
7408 
7409 		if (offset == 0)
7410 			ipf->ipf_nf_hdr_len = hdr_length;
7411 
7412 		/* Save current byte count */
7413 		count = ipf->ipf_count;
7414 		ret = ip_reassemble(mp, ipf,
7415 		    (frag_offset_flags & IPH_OFFSET) << 3,
7416 		    (frag_offset_flags & IPH_MF), ill, msg_len);
7417 		/* Count of bytes added and subtracted (freeb()ed) */
7418 		count = ipf->ipf_count - count;
7419 		if (count) {
7420 			/* Update per ipfb and ill byte counts */
7421 			ipfb->ipfb_count += count;
7422 			ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7423 			atomic_add_32(&ill->ill_frag_count, count);
7424 		}
7425 		if (ret == IP_REASS_PARTIAL) {
7426 			goto reass_done;
7427 		} else if (ret == IP_REASS_FAILED) {
7428 			/* Reassembly failed. Free up all resources */
7429 			ill_frag_free_pkts(ill, ipfb, ipf, 1);
7430 			for (t_mp = mp; t_mp != NULL; t_mp = t_mp->b_cont) {
7431 				IP_REASS_SET_START(t_mp, 0);
7432 				IP_REASS_SET_END(t_mp, 0);
7433 			}
7434 			freemsg(mp);
7435 			goto reass_done;
7436 		}
7437 		/* We will reach here iff 'ret' is IP_REASS_COMPLETE */
7438 	}
7439 	/*
7440 	 * We have completed reassembly.  Unhook the frag header from
7441 	 * the reassembly list.
7442 	 *
7443 	 * Before we free the frag header, record the ECN info
7444 	 * to report back to the transport.
7445 	 */
7446 	ecn_info = ipf->ipf_ecn;
7447 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmOKs);
7448 	ipfp = ipf->ipf_ptphn;
7449 
7450 	/* We need to supply these to caller */
7451 	if ((sum_flags = ipf->ipf_checksum_flags) != 0)
7452 		sum_val = ipf->ipf_checksum;
7453 	else
7454 		sum_val = 0;
7455 
7456 	mp1 = ipf->ipf_mp;
7457 	count = ipf->ipf_count;
7458 	ipf = ipf->ipf_hash_next;
7459 	if (ipf != NULL)
7460 		ipf->ipf_ptphn = ipfp;
7461 	ipfp[0] = ipf;
7462 	atomic_add_32(&ill->ill_frag_count, -count);
7463 	ASSERT(ipfb->ipfb_count >= count);
7464 	ipfb->ipfb_count -= count;
7465 	ipfb->ipfb_frag_pkts--;
7466 	mutex_exit(&ipfb->ipfb_lock);
7467 	/* Ditch the frag header. */
7468 	mp = mp1->b_cont;
7469 
7470 	freeb(mp1);
7471 
7472 	/* Restore original IP length in header. */
7473 	packet_size = (uint32_t)msgdsize(mp);
7474 	if (packet_size > IP_MAXPACKET) {
7475 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7476 		ip_drop_input("Reassembled packet too large", mp, ill);
7477 		freemsg(mp);
7478 		return (NULL);
7479 	}
7480 
7481 	if (DB_REF(mp) > 1) {
7482 		mblk_t *mp2 = copymsg(mp);
7483 
7484 		if (mp2 == NULL) {
7485 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7486 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7487 			freemsg(mp);
7488 			return (NULL);
7489 		}
7490 		freemsg(mp);
7491 		mp = mp2;
7492 	}
7493 	ipha = (ipha_t *)mp->b_rptr;
7494 
7495 	ipha->ipha_length = htons((uint16_t)packet_size);
7496 	/* We're now complete, zip the frag state */
7497 	ipha->ipha_fragment_offset_and_flags = 0;
7498 	/* Record the ECN info. */
7499 	ipha->ipha_type_of_service &= 0xFC;
7500 	ipha->ipha_type_of_service |= ecn_info;
7501 
7502 	/* Update the receive attributes */
7503 	ira->ira_pktlen = packet_size;
7504 	ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
7505 
7506 	/* Reassembly is successful; set checksum information in packet */
7507 	DB_CKSUM16(mp) = (uint16_t)sum_val;
7508 	DB_CKSUMFLAGS(mp) = sum_flags;
7509 	DB_CKSUMSTART(mp) = ira->ira_ip_hdr_length;
7510 
7511 	return (mp);
7512 }
7513 
7514 /*
7515  * Pullup function that should be used for IP input in order to
7516  * ensure we do not loose the L2 source address; we need the l2 source
7517  * address for IP_RECVSLLA and for ndp_input.
7518  *
7519  * We return either NULL or b_rptr.
7520  */
7521 void *
ip_pullup(mblk_t * mp,ssize_t len,ip_recv_attr_t * ira)7522 ip_pullup(mblk_t *mp, ssize_t len, ip_recv_attr_t *ira)
7523 {
7524 	ill_t		*ill = ira->ira_ill;
7525 
7526 	if (ip_rput_pullups++ == 0) {
7527 		(void) mi_strlog(ill->ill_rq, 1, SL_ERROR|SL_TRACE,
7528 		    "ip_pullup: %s forced us to "
7529 		    " pullup pkt, hdr len %ld, hdr addr %p",
7530 		    ill->ill_name, len, (void *)mp->b_rptr);
7531 	}
7532 	if (!(ira->ira_flags & IRAF_L2SRC_SET))
7533 		ip_setl2src(mp, ira, ira->ira_rill);
7534 	ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7535 	if (!pullupmsg(mp, len))
7536 		return (NULL);
7537 	else
7538 		return (mp->b_rptr);
7539 }
7540 
7541 /*
7542  * Make sure ira_l2src has an address. If we don't have one fill with zeros.
7543  * When called from the ULP ira_rill will be NULL hence the caller has to
7544  * pass in the ill.
7545  */
7546 /* ARGSUSED */
7547 void
ip_setl2src(mblk_t * mp,ip_recv_attr_t * ira,ill_t * ill)7548 ip_setl2src(mblk_t *mp, ip_recv_attr_t *ira, ill_t *ill)
7549 {
7550 	const uchar_t *addr;
7551 	int alen;
7552 
7553 	if (ira->ira_flags & IRAF_L2SRC_SET)
7554 		return;
7555 
7556 	ASSERT(ill != NULL);
7557 	alen = ill->ill_phys_addr_length;
7558 	ASSERT(alen <= sizeof (ira->ira_l2src));
7559 	if (ira->ira_mhip != NULL &&
7560 	    (addr = ira->ira_mhip->mhi_saddr) != NULL) {
7561 		bcopy(addr, ira->ira_l2src, alen);
7562 	} else if ((ira->ira_flags & IRAF_L2SRC_LOOPBACK) &&
7563 	    (addr = ill->ill_phys_addr) != NULL) {
7564 		bcopy(addr, ira->ira_l2src, alen);
7565 	} else {
7566 		bzero(ira->ira_l2src, alen);
7567 	}
7568 	ira->ira_flags |= IRAF_L2SRC_SET;
7569 }
7570 
7571 /*
7572  * check ip header length and align it.
7573  */
7574 mblk_t *
ip_check_and_align_header(mblk_t * mp,uint_t min_size,ip_recv_attr_t * ira)7575 ip_check_and_align_header(mblk_t *mp, uint_t min_size, ip_recv_attr_t *ira)
7576 {
7577 	ill_t	*ill = ira->ira_ill;
7578 	ssize_t len;
7579 
7580 	len = MBLKL(mp);
7581 
7582 	if (!OK_32PTR(mp->b_rptr))
7583 		IP_STAT(ill->ill_ipst, ip_notaligned);
7584 	else
7585 		IP_STAT(ill->ill_ipst, ip_recv_pullup);
7586 
7587 	/* Guard against bogus device drivers */
7588 	if (len < 0) {
7589 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7590 		ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7591 		freemsg(mp);
7592 		return (NULL);
7593 	}
7594 
7595 	if (len == 0) {
7596 		/* GLD sometimes sends up mblk with b_rptr == b_wptr! */
7597 		mblk_t *mp1 = mp->b_cont;
7598 
7599 		if (!(ira->ira_flags & IRAF_L2SRC_SET))
7600 			ip_setl2src(mp, ira, ira->ira_rill);
7601 		ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7602 
7603 		freeb(mp);
7604 		mp = mp1;
7605 		if (mp == NULL)
7606 			return (NULL);
7607 
7608 		if (OK_32PTR(mp->b_rptr) && MBLKL(mp) >= min_size)
7609 			return (mp);
7610 	}
7611 	if (ip_pullup(mp, min_size, ira) == NULL) {
7612 		if (msgdsize(mp) < min_size) {
7613 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7614 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7615 		} else {
7616 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7617 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7618 		}
7619 		freemsg(mp);
7620 		return (NULL);
7621 	}
7622 	return (mp);
7623 }
7624 
7625 /*
7626  * Common code for IPv4 and IPv6 to check and pullup multi-mblks
7627  */
7628 mblk_t *
ip_check_length(mblk_t * mp,uchar_t * rptr,ssize_t len,uint_t pkt_len,uint_t min_size,ip_recv_attr_t * ira)7629 ip_check_length(mblk_t *mp, uchar_t *rptr, ssize_t len,	uint_t pkt_len,
7630     uint_t min_size, ip_recv_attr_t *ira)
7631 {
7632 	ill_t	*ill = ira->ira_ill;
7633 
7634 	/*
7635 	 * Make sure we have data length consistent
7636 	 * with the IP header.
7637 	 */
7638 	if (mp->b_cont == NULL) {
7639 		/* pkt_len is based on ipha_len, not the mblk length */
7640 		if (pkt_len < min_size) {
7641 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7642 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7643 			freemsg(mp);
7644 			return (NULL);
7645 		}
7646 		if (len < 0) {
7647 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7648 			ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7649 			freemsg(mp);
7650 			return (NULL);
7651 		}
7652 		/* Drop any pad */
7653 		mp->b_wptr = rptr + pkt_len;
7654 	} else if ((len += msgdsize(mp->b_cont)) != 0) {
7655 		ASSERT(pkt_len >= min_size);
7656 		if (pkt_len < min_size) {
7657 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7658 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7659 			freemsg(mp);
7660 			return (NULL);
7661 		}
7662 		if (len < 0) {
7663 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7664 			ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7665 			freemsg(mp);
7666 			return (NULL);
7667 		}
7668 		/* Drop any pad */
7669 		(void) adjmsg(mp, -len);
7670 		/*
7671 		 * adjmsg may have freed an mblk from the chain, hence
7672 		 * invalidate any hw checksum here. This will force IP to
7673 		 * calculate the checksum in sw, but only for this packet.
7674 		 */
7675 		DB_CKSUMFLAGS(mp) = 0;
7676 		IP_STAT(ill->ill_ipst, ip_multimblk);
7677 	}
7678 	return (mp);
7679 }
7680 
7681 /*
7682  * Check that the IPv4 opt_len is consistent with the packet and pullup
7683  * the options.
7684  */
7685 mblk_t *
ip_check_optlen(mblk_t * mp,ipha_t * ipha,uint_t opt_len,uint_t pkt_len,ip_recv_attr_t * ira)7686 ip_check_optlen(mblk_t *mp, ipha_t *ipha, uint_t opt_len, uint_t pkt_len,
7687     ip_recv_attr_t *ira)
7688 {
7689 	ill_t	*ill = ira->ira_ill;
7690 	ssize_t len;
7691 
7692 	/* Assume no IPv6 packets arrive over the IPv4 queue */
7693 	if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
7694 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7695 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInWrongIPVersion);
7696 		ip_drop_input("IPvN packet on IPv4 ill", mp, ill);
7697 		freemsg(mp);
7698 		return (NULL);
7699 	}
7700 
7701 	if (opt_len > (15 - IP_SIMPLE_HDR_LENGTH_IN_WORDS)) {
7702 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7703 		ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7704 		freemsg(mp);
7705 		return (NULL);
7706 	}
7707 	/*
7708 	 * Recompute complete header length and make sure we
7709 	 * have access to all of it.
7710 	 */
7711 	len = ((size_t)opt_len + IP_SIMPLE_HDR_LENGTH_IN_WORDS) << 2;
7712 	if (len > (mp->b_wptr - mp->b_rptr)) {
7713 		if (len > pkt_len) {
7714 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7715 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7716 			freemsg(mp);
7717 			return (NULL);
7718 		}
7719 		if (ip_pullup(mp, len, ira) == NULL) {
7720 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7721 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7722 			freemsg(mp);
7723 			return (NULL);
7724 		}
7725 	}
7726 	return (mp);
7727 }
7728 
7729 /*
7730  * Returns a new ire, or the same ire, or NULL.
7731  * If a different IRE is returned, then it is held; the caller
7732  * needs to release it.
7733  * In no case is there any hold/release on the ire argument.
7734  */
7735 ire_t *
ip_check_multihome(void * addr,ire_t * ire,ill_t * ill)7736 ip_check_multihome(void *addr, ire_t *ire, ill_t *ill)
7737 {
7738 	ire_t		*new_ire;
7739 	ill_t		*ire_ill;
7740 	uint_t		ifindex;
7741 	ip_stack_t	*ipst = ill->ill_ipst;
7742 	boolean_t	strict_check = B_FALSE;
7743 
7744 	/*
7745 	 * IPMP common case: if IRE and ILL are in the same group, there's no
7746 	 * issue (e.g. packet received on an underlying interface matched an
7747 	 * IRE_LOCAL on its associated group interface).
7748 	 */
7749 	ASSERT(ire->ire_ill != NULL);
7750 	if (IS_IN_SAME_ILLGRP(ill, ire->ire_ill))
7751 		return (ire);
7752 
7753 	/*
7754 	 * Do another ire lookup here, using the ingress ill, to see if the
7755 	 * interface is in a usesrc group.
7756 	 * As long as the ills belong to the same group, we don't consider
7757 	 * them to be arriving on the wrong interface. Thus, if the switch
7758 	 * is doing inbound load spreading, we won't drop packets when the
7759 	 * ip*_strict_dst_multihoming switch is on.
7760 	 * We also need to check for IPIF_UNNUMBERED point2point interfaces
7761 	 * where the local address may not be unique. In this case we were
7762 	 * at the mercy of the initial ire lookup and the IRE_LOCAL it
7763 	 * actually returned. The new lookup, which is more specific, should
7764 	 * only find the IRE_LOCAL associated with the ingress ill if one
7765 	 * exists.
7766 	 */
7767 	if (ire->ire_ipversion == IPV4_VERSION) {
7768 		if (ipst->ips_ip_strict_dst_multihoming)
7769 			strict_check = B_TRUE;
7770 		new_ire = ire_ftable_lookup_v4(*((ipaddr_t *)addr), 0, 0,
7771 		    IRE_LOCAL, ill, ALL_ZONES, NULL,
7772 		    (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7773 	} else {
7774 		ASSERT(!IN6_IS_ADDR_MULTICAST((in6_addr_t *)addr));
7775 		if (ipst->ips_ipv6_strict_dst_multihoming)
7776 			strict_check = B_TRUE;
7777 		new_ire = ire_ftable_lookup_v6((in6_addr_t *)addr, NULL, NULL,
7778 		    IRE_LOCAL, ill, ALL_ZONES, NULL,
7779 		    (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7780 	}
7781 	/*
7782 	 * If the same ire that was returned in ip_input() is found then this
7783 	 * is an indication that usesrc groups are in use. The packet
7784 	 * arrived on a different ill in the group than the one associated with
7785 	 * the destination address.  If a different ire was found then the same
7786 	 * IP address must be hosted on multiple ills. This is possible with
7787 	 * unnumbered point2point interfaces. We switch to use this new ire in
7788 	 * order to have accurate interface statistics.
7789 	 */
7790 	if (new_ire != NULL) {
7791 		/* Note: held in one case but not the other? Caller handles */
7792 		if (new_ire != ire)
7793 			return (new_ire);
7794 		/* Unchanged */
7795 		ire_refrele(new_ire);
7796 		return (ire);
7797 	}
7798 
7799 	/*
7800 	 * Chase pointers once and store locally.
7801 	 */
7802 	ASSERT(ire->ire_ill != NULL);
7803 	ire_ill = ire->ire_ill;
7804 	ifindex = ill->ill_usesrc_ifindex;
7805 
7806 	/*
7807 	 * Check if it's a legal address on the 'usesrc' interface.
7808 	 * For IPMP data addresses the IRE_LOCAL is the upper, hence we
7809 	 * can just check phyint_ifindex.
7810 	 */
7811 	if (ifindex != 0 && ifindex == ire_ill->ill_phyint->phyint_ifindex) {
7812 		return (ire);
7813 	}
7814 
7815 	/*
7816 	 * If the ip*_strict_dst_multihoming switch is on then we can
7817 	 * only accept this packet if the interface is marked as routing.
7818 	 */
7819 	if (!(strict_check))
7820 		return (ire);
7821 
7822 	if ((ill->ill_flags & ire->ire_ill->ill_flags & ILLF_ROUTER) != 0) {
7823 		return (ire);
7824 	}
7825 	return (NULL);
7826 }
7827 
7828 /*
7829  * This function is used to construct a mac_header_info_s from a
7830  * DL_UNITDATA_IND message.
7831  * The address fields in the mhi structure points into the message,
7832  * thus the caller can't use those fields after freeing the message.
7833  *
7834  * We determine whether the packet received is a non-unicast packet
7835  * and in doing so, determine whether or not it is broadcast vs multicast.
7836  * For it to be a broadcast packet, we must have the appropriate mblk_t
7837  * hanging off the ill_t.  If this is either not present or doesn't match
7838  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7839  * to be multicast.  Thus NICs that have no broadcast address (or no
7840  * capability for one, such as point to point links) cannot return as
7841  * the packet being broadcast.
7842  */
7843 void
ip_dlur_to_mhi(ill_t * ill,mblk_t * mb,struct mac_header_info_s * mhip)7844 ip_dlur_to_mhi(ill_t *ill, mblk_t *mb, struct mac_header_info_s *mhip)
7845 {
7846 	dl_unitdata_ind_t *ind = (dl_unitdata_ind_t *)mb->b_rptr;
7847 	mblk_t *bmp;
7848 	uint_t extra_offset;
7849 
7850 	bzero(mhip, sizeof (struct mac_header_info_s));
7851 
7852 	mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7853 
7854 	if (ill->ill_sap_length < 0)
7855 		extra_offset = 0;
7856 	else
7857 		extra_offset = ill->ill_sap_length;
7858 
7859 	mhip->mhi_daddr = (uchar_t *)ind + ind->dl_dest_addr_offset +
7860 	    extra_offset;
7861 	mhip->mhi_saddr = (uchar_t *)ind + ind->dl_src_addr_offset +
7862 	    extra_offset;
7863 
7864 	if (!ind->dl_group_address)
7865 		return;
7866 
7867 	/* Multicast or broadcast */
7868 	mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7869 
7870 	if (ind->dl_dest_addr_offset > sizeof (*ind) &&
7871 	    ind->dl_dest_addr_offset + ind->dl_dest_addr_length < MBLKL(mb) &&
7872 	    (bmp = ill->ill_bcast_mp) != NULL) {
7873 		dl_unitdata_req_t *dlur;
7874 		uint8_t *bphys_addr;
7875 
7876 		dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7877 		bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
7878 		    extra_offset;
7879 
7880 		if (bcmp(mhip->mhi_daddr, bphys_addr,
7881 		    ind->dl_dest_addr_length) == 0)
7882 			mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7883 	}
7884 }
7885 
7886 /*
7887  * This function is used to construct a mac_header_info_s from a
7888  * M_DATA fastpath message from a DLPI driver.
7889  * The address fields in the mhi structure points into the message,
7890  * thus the caller can't use those fields after freeing the message.
7891  *
7892  * We determine whether the packet received is a non-unicast packet
7893  * and in doing so, determine whether or not it is broadcast vs multicast.
7894  * For it to be a broadcast packet, we must have the appropriate mblk_t
7895  * hanging off the ill_t.  If this is either not present or doesn't match
7896  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7897  * to be multicast.  Thus NICs that have no broadcast address (or no
7898  * capability for one, such as point to point links) cannot return as
7899  * the packet being broadcast.
7900  */
7901 void
ip_mdata_to_mhi(ill_t * ill,mblk_t * mp,struct mac_header_info_s * mhip)7902 ip_mdata_to_mhi(ill_t *ill, mblk_t *mp, struct mac_header_info_s *mhip)
7903 {
7904 	mblk_t *bmp;
7905 	struct ether_header *pether;
7906 
7907 	bzero(mhip, sizeof (struct mac_header_info_s));
7908 
7909 	mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7910 
7911 	pether = (struct ether_header *)((char *)mp->b_rptr
7912 	    - sizeof (struct ether_header));
7913 
7914 	/*
7915 	 * Make sure the interface is an ethernet type, since we don't
7916 	 * know the header format for anything but Ethernet. Also make
7917 	 * sure we are pointing correctly above db_base.
7918 	 */
7919 	if (ill->ill_type != IFT_ETHER)
7920 		return;
7921 
7922 retry:
7923 	if ((uchar_t *)pether < mp->b_datap->db_base)
7924 		return;
7925 
7926 	/* Is there a VLAN tag? */
7927 	if (ill->ill_isv6) {
7928 		if (pether->ether_type != htons(ETHERTYPE_IPV6)) {
7929 			pether = (struct ether_header *)((char *)pether - 4);
7930 			goto retry;
7931 		}
7932 	} else {
7933 		if (pether->ether_type != htons(ETHERTYPE_IP)) {
7934 			pether = (struct ether_header *)((char *)pether - 4);
7935 			goto retry;
7936 		}
7937 	}
7938 	mhip->mhi_daddr = (uchar_t *)&pether->ether_dhost;
7939 	mhip->mhi_saddr = (uchar_t *)&pether->ether_shost;
7940 
7941 	if (!(mhip->mhi_daddr[0] & 0x01))
7942 		return;
7943 
7944 	/* Multicast or broadcast */
7945 	mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7946 
7947 	if ((bmp = ill->ill_bcast_mp) != NULL) {
7948 		dl_unitdata_req_t *dlur;
7949 		uint8_t *bphys_addr;
7950 		uint_t	addrlen;
7951 
7952 		dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7953 		addrlen = dlur->dl_dest_addr_length;
7954 		if (ill->ill_sap_length < 0) {
7955 			bphys_addr = (uchar_t *)dlur +
7956 			    dlur->dl_dest_addr_offset;
7957 			addrlen += ill->ill_sap_length;
7958 		} else {
7959 			bphys_addr = (uchar_t *)dlur +
7960 			    dlur->dl_dest_addr_offset +
7961 			    ill->ill_sap_length;
7962 			addrlen -= ill->ill_sap_length;
7963 		}
7964 		if (bcmp(mhip->mhi_daddr, bphys_addr, addrlen) == 0)
7965 			mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7966 	}
7967 }
7968 
7969 /*
7970  * Handle anything but M_DATA messages
7971  * We see the DL_UNITDATA_IND which are part
7972  * of the data path, and also the other messages from the driver.
7973  */
7974 void
ip_rput_notdata(ill_t * ill,mblk_t * mp)7975 ip_rput_notdata(ill_t *ill, mblk_t *mp)
7976 {
7977 	mblk_t		*first_mp;
7978 	struct iocblk   *iocp;
7979 	struct mac_header_info_s mhi;
7980 
7981 	switch (DB_TYPE(mp)) {
7982 	case M_PROTO:
7983 	case M_PCPROTO: {
7984 		if (((dl_unitdata_ind_t *)mp->b_rptr)->dl_primitive !=
7985 		    DL_UNITDATA_IND) {
7986 			/* Go handle anything other than data elsewhere. */
7987 			ip_rput_dlpi(ill, mp);
7988 			return;
7989 		}
7990 
7991 		first_mp = mp;
7992 		mp = first_mp->b_cont;
7993 		first_mp->b_cont = NULL;
7994 
7995 		if (mp == NULL) {
7996 			freeb(first_mp);
7997 			return;
7998 		}
7999 		ip_dlur_to_mhi(ill, first_mp, &mhi);
8000 		if (ill->ill_isv6)
8001 			ip_input_v6(ill, NULL, mp, &mhi);
8002 		else
8003 			ip_input(ill, NULL, mp, &mhi);
8004 
8005 		/* Ditch the DLPI header. */
8006 		freeb(first_mp);
8007 		return;
8008 	}
8009 	case M_IOCACK:
8010 		iocp = (struct iocblk *)mp->b_rptr;
8011 		switch (iocp->ioc_cmd) {
8012 		case DL_IOC_HDR_INFO:
8013 			ill_fastpath_ack(ill, mp);
8014 			return;
8015 		default:
8016 			putnext(ill->ill_rq, mp);
8017 			return;
8018 		}
8019 		/* FALLTHROUGH */
8020 	case M_ERROR:
8021 	case M_HANGUP:
8022 		mutex_enter(&ill->ill_lock);
8023 		if (ill->ill_state_flags & ILL_CONDEMNED) {
8024 			mutex_exit(&ill->ill_lock);
8025 			freemsg(mp);
8026 			return;
8027 		}
8028 		ill_refhold_locked(ill);
8029 		mutex_exit(&ill->ill_lock);
8030 		qwriter_ip(ill, ill->ill_rq, mp, ip_rput_other, CUR_OP,
8031 		    B_FALSE);
8032 		return;
8033 	case M_CTL:
8034 		putnext(ill->ill_rq, mp);
8035 		return;
8036 	case M_IOCNAK:
8037 		ip1dbg(("got iocnak "));
8038 		iocp = (struct iocblk *)mp->b_rptr;
8039 		switch (iocp->ioc_cmd) {
8040 		case DL_IOC_HDR_INFO:
8041 			ip_rput_other(NULL, ill->ill_rq, mp, NULL);
8042 			return;
8043 		default:
8044 			break;
8045 		}
8046 		/* FALLTHROUGH */
8047 	default:
8048 		putnext(ill->ill_rq, mp);
8049 		return;
8050 	}
8051 }
8052 
8053 /* Read side put procedure.  Packets coming from the wire arrive here. */
8054 int
ip_rput(queue_t * q,mblk_t * mp)8055 ip_rput(queue_t *q, mblk_t *mp)
8056 {
8057 	ill_t	*ill;
8058 	union DL_primitives *dl;
8059 
8060 	ill = (ill_t *)q->q_ptr;
8061 
8062 	if (ill->ill_state_flags & (ILL_CONDEMNED | ILL_LL_SUBNET_PENDING)) {
8063 		/*
8064 		 * If things are opening or closing, only accept high-priority
8065 		 * DLPI messages.  (On open ill->ill_ipif has not yet been
8066 		 * created; on close, things hanging off the ill may have been
8067 		 * freed already.)
8068 		 */
8069 		dl = (union DL_primitives *)mp->b_rptr;
8070 		if (DB_TYPE(mp) != M_PCPROTO ||
8071 		    dl->dl_primitive == DL_UNITDATA_IND) {
8072 			inet_freemsg(mp);
8073 			return (0);
8074 		}
8075 	}
8076 	if (DB_TYPE(mp) == M_DATA) {
8077 		struct mac_header_info_s mhi;
8078 
8079 		ip_mdata_to_mhi(ill, mp, &mhi);
8080 		ip_input(ill, NULL, mp, &mhi);
8081 	} else {
8082 		ip_rput_notdata(ill, mp);
8083 	}
8084 	return (0);
8085 }
8086 
8087 /*
8088  * Move the information to a copy.
8089  */
8090 mblk_t *
ip_fix_dbref(mblk_t * mp,ip_recv_attr_t * ira)8091 ip_fix_dbref(mblk_t *mp, ip_recv_attr_t *ira)
8092 {
8093 	mblk_t		*mp1;
8094 	ill_t		*ill = ira->ira_ill;
8095 	ip_stack_t	*ipst = ill->ill_ipst;
8096 
8097 	IP_STAT(ipst, ip_db_ref);
8098 
8099 	/* Make sure we have ira_l2src before we loose the original mblk */
8100 	if (!(ira->ira_flags & IRAF_L2SRC_SET))
8101 		ip_setl2src(mp, ira, ira->ira_rill);
8102 
8103 	mp1 = copymsg(mp);
8104 	if (mp1 == NULL) {
8105 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
8106 		ip_drop_input("ipIfStatsInDiscards", mp, ill);
8107 		freemsg(mp);
8108 		return (NULL);
8109 	}
8110 	/* preserve the hardware checksum flags and data, if present */
8111 	if (DB_CKSUMFLAGS(mp) != 0) {
8112 		DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
8113 		DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
8114 		DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
8115 		DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
8116 		DB_CKSUM16(mp1) = DB_CKSUM16(mp);
8117 	}
8118 	freemsg(mp);
8119 	return (mp1);
8120 }
8121 
8122 static void
ip_dlpi_error(ill_t * ill,t_uscalar_t prim,t_uscalar_t dl_err,t_uscalar_t err)8123 ip_dlpi_error(ill_t *ill, t_uscalar_t prim, t_uscalar_t dl_err,
8124     t_uscalar_t err)
8125 {
8126 	if (dl_err == DL_SYSERR) {
8127 		(void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8128 		    "%s: %s failed: DL_SYSERR (errno %u)\n",
8129 		    ill->ill_name, dl_primstr(prim), err);
8130 		return;
8131 	}
8132 
8133 	(void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8134 	    "%s: %s failed: %s\n", ill->ill_name, dl_primstr(prim),
8135 	    dl_errstr(dl_err));
8136 }
8137 
8138 /*
8139  * ip_rput_dlpi is called by ip_rput to handle all DLPI messages other
8140  * than DL_UNITDATA_IND messages. If we need to process this message
8141  * exclusively, we call qwriter_ip, in which case we also need to call
8142  * ill_refhold before that, since qwriter_ip does an ill_refrele.
8143  */
8144 void
ip_rput_dlpi(ill_t * ill,mblk_t * mp)8145 ip_rput_dlpi(ill_t *ill, mblk_t *mp)
8146 {
8147 	dl_ok_ack_t	*dloa = (dl_ok_ack_t *)mp->b_rptr;
8148 	dl_error_ack_t	*dlea = (dl_error_ack_t *)dloa;
8149 	queue_t		*q = ill->ill_rq;
8150 	t_uscalar_t	prim = dloa->dl_primitive;
8151 	t_uscalar_t	reqprim = DL_PRIM_INVAL;
8152 
8153 	DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi",
8154 	    char *, dl_primstr(prim), ill_t *, ill);
8155 	ip1dbg(("ip_rput_dlpi"));
8156 
8157 	/*
8158 	 * If we received an ACK but didn't send a request for it, then it
8159 	 * can't be part of any pending operation; discard up-front.
8160 	 */
8161 	switch (prim) {
8162 	case DL_ERROR_ACK:
8163 		reqprim = dlea->dl_error_primitive;
8164 		ip2dbg(("ip_rput_dlpi(%s): DL_ERROR_ACK for %s (0x%x): %s "
8165 		    "(0x%x), unix %u\n", ill->ill_name, dl_primstr(reqprim),
8166 		    reqprim, dl_errstr(dlea->dl_errno), dlea->dl_errno,
8167 		    dlea->dl_unix_errno));
8168 		break;
8169 	case DL_OK_ACK:
8170 		reqprim = dloa->dl_correct_primitive;
8171 		break;
8172 	case DL_INFO_ACK:
8173 		reqprim = DL_INFO_REQ;
8174 		break;
8175 	case DL_BIND_ACK:
8176 		reqprim = DL_BIND_REQ;
8177 		break;
8178 	case DL_PHYS_ADDR_ACK:
8179 		reqprim = DL_PHYS_ADDR_REQ;
8180 		break;
8181 	case DL_NOTIFY_ACK:
8182 		reqprim = DL_NOTIFY_REQ;
8183 		break;
8184 	case DL_CAPABILITY_ACK:
8185 		reqprim = DL_CAPABILITY_REQ;
8186 		break;
8187 	}
8188 
8189 	if (prim != DL_NOTIFY_IND) {
8190 		if (reqprim == DL_PRIM_INVAL ||
8191 		    !ill_dlpi_pending(ill, reqprim)) {
8192 			/* Not a DLPI message we support or expected */
8193 			freemsg(mp);
8194 			return;
8195 		}
8196 		ip1dbg(("ip_rput: received %s for %s\n", dl_primstr(prim),
8197 		    dl_primstr(reqprim)));
8198 	}
8199 
8200 	switch (reqprim) {
8201 	case DL_UNBIND_REQ:
8202 		/*
8203 		 * NOTE: we mark the unbind as complete even if we got a
8204 		 * DL_ERROR_ACK, since there's not much else we can do.
8205 		 */
8206 		mutex_enter(&ill->ill_lock);
8207 		ill->ill_state_flags &= ~ILL_DL_UNBIND_IN_PROGRESS;
8208 		cv_signal(&ill->ill_cv);
8209 		mutex_exit(&ill->ill_lock);
8210 		break;
8211 
8212 	case DL_ENABMULTI_REQ:
8213 		if (prim == DL_OK_ACK) {
8214 			if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8215 				ill->ill_dlpi_multicast_state = IDS_OK;
8216 		}
8217 		break;
8218 	}
8219 
8220 	/*
8221 	 * The message is one we're waiting for (or DL_NOTIFY_IND), but we
8222 	 * need to become writer to continue to process it.  Because an
8223 	 * exclusive operation doesn't complete until replies to all queued
8224 	 * DLPI messages have been received, we know we're in the middle of an
8225 	 * exclusive operation and pass CUR_OP (except for DL_NOTIFY_IND).
8226 	 *
8227 	 * As required by qwriter_ip(), we refhold the ill; it will refrele.
8228 	 * Since this is on the ill stream we unconditionally bump up the
8229 	 * refcount without doing ILL_CAN_LOOKUP().
8230 	 */
8231 	ill_refhold(ill);
8232 	if (prim == DL_NOTIFY_IND)
8233 		qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, NEW_OP, B_FALSE);
8234 	else
8235 		qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, CUR_OP, B_FALSE);
8236 }
8237 
8238 /*
8239  * Handling of DLPI messages that require exclusive access to the ipsq.
8240  *
8241  * Need to do ipsq_pending_mp_get on ioctl completion, which could
8242  * happen here. (along with mi_copy_done)
8243  */
8244 /* ARGSUSED */
8245 static void
ip_rput_dlpi_writer(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * dummy_arg)8246 ip_rput_dlpi_writer(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8247 {
8248 	dl_ok_ack_t	*dloa = (dl_ok_ack_t *)mp->b_rptr;
8249 	dl_error_ack_t	*dlea = (dl_error_ack_t *)dloa;
8250 	int		err = 0;
8251 	ill_t		*ill = (ill_t *)q->q_ptr;
8252 	ipif_t		*ipif = NULL;
8253 	mblk_t		*mp1 = NULL;
8254 	conn_t		*connp = NULL;
8255 	t_uscalar_t	paddrreq;
8256 	mblk_t		*mp_hw;
8257 	boolean_t	success;
8258 	boolean_t	ioctl_aborted = B_FALSE;
8259 	boolean_t	log = B_TRUE;
8260 
8261 	DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer",
8262 	    char *, dl_primstr(dloa->dl_primitive), ill_t *, ill);
8263 
8264 	ip1dbg(("ip_rput_dlpi_writer .."));
8265 	ASSERT(ipsq->ipsq_xop == ill->ill_phyint->phyint_ipsq->ipsq_xop);
8266 	ASSERT(IAM_WRITER_ILL(ill));
8267 
8268 	ipif = ipsq->ipsq_xop->ipx_pending_ipif;
8269 	/*
8270 	 * The current ioctl could have been aborted by the user and a new
8271 	 * ioctl to bring up another ill could have started. We could still
8272 	 * get a response from the driver later.
8273 	 */
8274 	if (ipif != NULL && ipif->ipif_ill != ill)
8275 		ioctl_aborted = B_TRUE;
8276 
8277 	switch (dloa->dl_primitive) {
8278 	case DL_ERROR_ACK:
8279 		ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for %s\n",
8280 		    dl_primstr(dlea->dl_error_primitive)));
8281 
8282 		DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer error",
8283 		    char *, dl_primstr(dlea->dl_error_primitive),
8284 		    ill_t *, ill);
8285 
8286 		switch (dlea->dl_error_primitive) {
8287 		case DL_DISABMULTI_REQ:
8288 			ill_dlpi_done(ill, dlea->dl_error_primitive);
8289 			break;
8290 		case DL_PROMISCON_REQ:
8291 		case DL_PROMISCOFF_REQ:
8292 		case DL_UNBIND_REQ:
8293 		case DL_ATTACH_REQ:
8294 		case DL_INFO_REQ:
8295 			ill_dlpi_done(ill, dlea->dl_error_primitive);
8296 			break;
8297 		case DL_NOTIFY_REQ:
8298 			ill_dlpi_done(ill, DL_NOTIFY_REQ);
8299 			log = B_FALSE;
8300 			break;
8301 		case DL_PHYS_ADDR_REQ:
8302 			/*
8303 			 * For IPv6 only, there are two additional
8304 			 * phys_addr_req's sent to the driver to get the
8305 			 * IPv6 token and lla. This allows IP to acquire
8306 			 * the hardware address format for a given interface
8307 			 * without having built in knowledge of the hardware
8308 			 * address. ill_phys_addr_pend keeps track of the last
8309 			 * DL_PAR sent so we know which response we are
8310 			 * dealing with. ill_dlpi_done will update
8311 			 * ill_phys_addr_pend when it sends the next req.
8312 			 * We don't complete the IOCTL until all three DL_PARs
8313 			 * have been attempted, so set *_len to 0 and break.
8314 			 */
8315 			paddrreq = ill->ill_phys_addr_pend;
8316 			ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8317 			if (paddrreq == DL_IPV6_TOKEN) {
8318 				ill->ill_token_length = 0;
8319 				log = B_FALSE;
8320 				break;
8321 			} else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8322 				ill->ill_nd_lla_len = 0;
8323 				log = B_FALSE;
8324 				break;
8325 			}
8326 			/*
8327 			 * Something went wrong with the DL_PHYS_ADDR_REQ.
8328 			 * We presumably have an IOCTL hanging out waiting
8329 			 * for completion. Find it and complete the IOCTL
8330 			 * with the error noted.
8331 			 * However, ill_dl_phys was called on an ill queue
8332 			 * (from SIOCSLIFNAME), thus conn_pending_ill is not
8333 			 * set. But the ioctl is known to be pending on ill_wq.
8334 			 */
8335 			if (!ill->ill_ifname_pending)
8336 				break;
8337 			ill->ill_ifname_pending = 0;
8338 			if (!ioctl_aborted)
8339 				mp1 = ipsq_pending_mp_get(ipsq, &connp);
8340 			if (mp1 != NULL) {
8341 				/*
8342 				 * This operation (SIOCSLIFNAME) must have
8343 				 * happened on the ill. Assert there is no conn
8344 				 */
8345 				ASSERT(connp == NULL);
8346 				q = ill->ill_wq;
8347 			}
8348 			break;
8349 		case DL_BIND_REQ:
8350 			ill_dlpi_done(ill, DL_BIND_REQ);
8351 			if (ill->ill_ifname_pending)
8352 				break;
8353 			mutex_enter(&ill->ill_lock);
8354 			ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8355 			mutex_exit(&ill->ill_lock);
8356 			/*
8357 			 * Something went wrong with the bind.  We presumably
8358 			 * have an IOCTL hanging out waiting for completion.
8359 			 * Find it, take down the interface that was coming
8360 			 * up, and complete the IOCTL with the error noted.
8361 			 */
8362 			if (!ioctl_aborted)
8363 				mp1 = ipsq_pending_mp_get(ipsq, &connp);
8364 			if (mp1 != NULL) {
8365 				/*
8366 				 * This might be a result of a DL_NOTE_REPLUMB
8367 				 * notification. In that case, connp is NULL.
8368 				 */
8369 				if (connp != NULL)
8370 					q = CONNP_TO_WQ(connp);
8371 
8372 				(void) ipif_down(ipif, NULL, NULL);
8373 				/* error is set below the switch */
8374 			}
8375 			break;
8376 		case DL_ENABMULTI_REQ:
8377 			ill_dlpi_done(ill, DL_ENABMULTI_REQ);
8378 
8379 			if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8380 				ill->ill_dlpi_multicast_state = IDS_FAILED;
8381 			if (ill->ill_dlpi_multicast_state == IDS_FAILED) {
8382 
8383 				printf("ip: joining multicasts failed (%d)"
8384 				    " on %s - will use link layer "
8385 				    "broadcasts for multicast\n",
8386 				    dlea->dl_errno, ill->ill_name);
8387 
8388 				/*
8389 				 * Set up for multi_bcast; We are the
8390 				 * writer, so ok to access ill->ill_ipif
8391 				 * without any lock.
8392 				 */
8393 				mutex_enter(&ill->ill_phyint->phyint_lock);
8394 				ill->ill_phyint->phyint_flags |=
8395 				    PHYI_MULTI_BCAST;
8396 				mutex_exit(&ill->ill_phyint->phyint_lock);
8397 
8398 			}
8399 			freemsg(mp);	/* Don't want to pass this up */
8400 			return;
8401 		case DL_CAPABILITY_REQ:
8402 			ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for "
8403 			    "DL_CAPABILITY REQ\n"));
8404 			if (ill->ill_dlpi_capab_state == IDCS_PROBE_SENT)
8405 				ill->ill_dlpi_capab_state = IDCS_FAILED;
8406 			ill_capability_done(ill);
8407 			freemsg(mp);
8408 			return;
8409 		}
8410 		/*
8411 		 * Note the error for IOCTL completion (mp1 is set when
8412 		 * ready to complete ioctl). If ill_ifname_pending_err is
8413 		 * set, an error occured during plumbing (ill_ifname_pending),
8414 		 * so we want to report that error.
8415 		 *
8416 		 * NOTE: there are two addtional DL_PHYS_ADDR_REQ's
8417 		 * (DL_IPV6_TOKEN and DL_IPV6_LINK_LAYER_ADDR) that are
8418 		 * expected to get errack'd if the driver doesn't support
8419 		 * these flags (e.g. ethernet). log will be set to B_FALSE
8420 		 * if these error conditions are encountered.
8421 		 */
8422 		if (mp1 != NULL) {
8423 			if (ill->ill_ifname_pending_err != 0)  {
8424 				err = ill->ill_ifname_pending_err;
8425 				ill->ill_ifname_pending_err = 0;
8426 			} else {
8427 				err = dlea->dl_unix_errno ?
8428 				    dlea->dl_unix_errno : ENXIO;
8429 			}
8430 		/*
8431 		 * If we're plumbing an interface and an error hasn't already
8432 		 * been saved, set ill_ifname_pending_err to the error passed
8433 		 * up. Ignore the error if log is B_FALSE (see comment above).
8434 		 */
8435 		} else if (log && ill->ill_ifname_pending &&
8436 		    ill->ill_ifname_pending_err == 0) {
8437 			ill->ill_ifname_pending_err = dlea->dl_unix_errno ?
8438 			    dlea->dl_unix_errno : ENXIO;
8439 		}
8440 
8441 		if (log)
8442 			ip_dlpi_error(ill, dlea->dl_error_primitive,
8443 			    dlea->dl_errno, dlea->dl_unix_errno);
8444 		break;
8445 	case DL_CAPABILITY_ACK:
8446 		ill_capability_ack(ill, mp);
8447 		/*
8448 		 * The message has been handed off to ill_capability_ack
8449 		 * and must not be freed below
8450 		 */
8451 		mp = NULL;
8452 		break;
8453 
8454 	case DL_INFO_ACK:
8455 		/* Call a routine to handle this one. */
8456 		ill_dlpi_done(ill, DL_INFO_REQ);
8457 		ip_ll_subnet_defaults(ill, mp);
8458 		ASSERT(!MUTEX_HELD(&ill->ill_phyint->phyint_ipsq->ipsq_lock));
8459 		return;
8460 	case DL_BIND_ACK:
8461 		/*
8462 		 * We should have an IOCTL waiting on this unless
8463 		 * sent by ill_dl_phys, in which case just return
8464 		 */
8465 		ill_dlpi_done(ill, DL_BIND_REQ);
8466 
8467 		if (ill->ill_ifname_pending) {
8468 			DTRACE_PROBE2(ip__rput__dlpi__ifname__pending,
8469 			    ill_t *, ill, mblk_t *, mp);
8470 			break;
8471 		}
8472 		mutex_enter(&ill->ill_lock);
8473 		ill->ill_dl_up = 1;
8474 		ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8475 		mutex_exit(&ill->ill_lock);
8476 
8477 		if (!ioctl_aborted)
8478 			mp1 = ipsq_pending_mp_get(ipsq, &connp);
8479 		if (mp1 == NULL) {
8480 			DTRACE_PROBE1(ip__rput__dlpi__no__mblk, ill_t *, ill);
8481 			break;
8482 		}
8483 		/*
8484 		 * mp1 was added by ill_dl_up(). if that is a result of
8485 		 * a DL_NOTE_REPLUMB notification, connp could be NULL.
8486 		 */
8487 		if (connp != NULL)
8488 			q = CONNP_TO_WQ(connp);
8489 		/*
8490 		 * We are exclusive. So nothing can change even after
8491 		 * we get the pending mp.
8492 		 */
8493 		ip1dbg(("ip_rput_dlpi: bind_ack %s\n", ill->ill_name));
8494 		DTRACE_PROBE1(ip__rput__dlpi__bind__ack, ill_t *, ill);
8495 		ill_nic_event_dispatch(ill, 0, NE_UP, NULL, 0);
8496 
8497 		/*
8498 		 * Now bring up the resolver; when that is complete, we'll
8499 		 * create IREs.  Note that we intentionally mirror what
8500 		 * ipif_up() would have done, because we got here by way of
8501 		 * ill_dl_up(), which stopped ipif_up()'s processing.
8502 		 */
8503 		if (ill->ill_isv6) {
8504 			/*
8505 			 * v6 interfaces.
8506 			 * Unlike ARP which has to do another bind
8507 			 * and attach, once we get here we are
8508 			 * done with NDP
8509 			 */
8510 			(void) ipif_resolver_up(ipif, Res_act_initial);
8511 			if ((err = ipif_ndp_up(ipif, B_TRUE)) == 0)
8512 				err = ipif_up_done_v6(ipif);
8513 		} else if (ill->ill_net_type == IRE_IF_RESOLVER) {
8514 			/*
8515 			 * ARP and other v4 external resolvers.
8516 			 * Leave the pending mblk intact so that
8517 			 * the ioctl completes in ip_rput().
8518 			 */
8519 			if (connp != NULL)
8520 				mutex_enter(&connp->conn_lock);
8521 			mutex_enter(&ill->ill_lock);
8522 			success = ipsq_pending_mp_add(connp, ipif, q, mp1, 0);
8523 			mutex_exit(&ill->ill_lock);
8524 			if (connp != NULL)
8525 				mutex_exit(&connp->conn_lock);
8526 			if (success) {
8527 				err = ipif_resolver_up(ipif, Res_act_initial);
8528 				if (err == EINPROGRESS) {
8529 					freemsg(mp);
8530 					return;
8531 				}
8532 				mp1 = ipsq_pending_mp_get(ipsq, &connp);
8533 			} else {
8534 				/* The conn has started closing */
8535 				err = EINTR;
8536 			}
8537 		} else {
8538 			/*
8539 			 * This one is complete. Reply to pending ioctl.
8540 			 */
8541 			(void) ipif_resolver_up(ipif, Res_act_initial);
8542 			err = ipif_up_done(ipif);
8543 		}
8544 
8545 		if ((err == 0) && (ill->ill_up_ipifs)) {
8546 			err = ill_up_ipifs(ill, q, mp1);
8547 			if (err == EINPROGRESS) {
8548 				freemsg(mp);
8549 				return;
8550 			}
8551 		}
8552 
8553 		/*
8554 		 * If we have a moved ipif to bring up, and everything has
8555 		 * succeeded to this point, bring it up on the IPMP ill.
8556 		 * Otherwise, leave it down -- the admin can try to bring it
8557 		 * up by hand if need be.
8558 		 */
8559 		if (ill->ill_move_ipif != NULL) {
8560 			if (err != 0) {
8561 				ill->ill_move_ipif = NULL;
8562 			} else {
8563 				ipif = ill->ill_move_ipif;
8564 				ill->ill_move_ipif = NULL;
8565 				err = ipif_up(ipif, q, mp1);
8566 				if (err == EINPROGRESS) {
8567 					freemsg(mp);
8568 					return;
8569 				}
8570 			}
8571 		}
8572 		break;
8573 
8574 	case DL_NOTIFY_IND: {
8575 		dl_notify_ind_t *notify = (dl_notify_ind_t *)mp->b_rptr;
8576 		uint_t orig_mtu, orig_mc_mtu;
8577 
8578 		switch (notify->dl_notification) {
8579 		case DL_NOTE_PHYS_ADDR:
8580 			err = ill_set_phys_addr(ill, mp);
8581 			break;
8582 
8583 		case DL_NOTE_REPLUMB:
8584 			/*
8585 			 * Directly return after calling ill_replumb().
8586 			 * Note that we should not free mp as it is reused
8587 			 * in the ill_replumb() function.
8588 			 */
8589 			err = ill_replumb(ill, mp);
8590 			return;
8591 
8592 		case DL_NOTE_FASTPATH_FLUSH:
8593 			nce_flush(ill, B_FALSE);
8594 			break;
8595 
8596 		case DL_NOTE_SDU_SIZE:
8597 		case DL_NOTE_SDU_SIZE2:
8598 			/*
8599 			 * The dce and fragmentation code can cope with
8600 			 * this changing while packets are being sent.
8601 			 * When packets are sent ip_output will discover
8602 			 * a change.
8603 			 *
8604 			 * Change the MTU size of the interface.
8605 			 */
8606 			mutex_enter(&ill->ill_lock);
8607 			orig_mtu = ill->ill_mtu;
8608 			orig_mc_mtu = ill->ill_mc_mtu;
8609 			switch (notify->dl_notification) {
8610 			case DL_NOTE_SDU_SIZE:
8611 				ill->ill_current_frag =
8612 				    (uint_t)notify->dl_data;
8613 				ill->ill_mc_mtu = (uint_t)notify->dl_data;
8614 				break;
8615 			case DL_NOTE_SDU_SIZE2:
8616 				ill->ill_current_frag =
8617 				    (uint_t)notify->dl_data1;
8618 				ill->ill_mc_mtu = (uint_t)notify->dl_data2;
8619 				break;
8620 			}
8621 			if (ill->ill_current_frag > ill->ill_max_frag)
8622 				ill->ill_max_frag = ill->ill_current_frag;
8623 
8624 			if (!(ill->ill_flags & ILLF_FIXEDMTU)) {
8625 				ill->ill_mtu = ill->ill_current_frag;
8626 
8627 				/*
8628 				 * If ill_user_mtu was set (via
8629 				 * SIOCSLIFLNKINFO), clamp ill_mtu at it.
8630 				 */
8631 				if (ill->ill_user_mtu != 0 &&
8632 				    ill->ill_user_mtu < ill->ill_mtu)
8633 					ill->ill_mtu = ill->ill_user_mtu;
8634 
8635 				if (ill->ill_user_mtu != 0 &&
8636 				    ill->ill_user_mtu < ill->ill_mc_mtu)
8637 					ill->ill_mc_mtu = ill->ill_user_mtu;
8638 
8639 				if (ill->ill_isv6) {
8640 					if (ill->ill_mtu < IPV6_MIN_MTU)
8641 						ill->ill_mtu = IPV6_MIN_MTU;
8642 					if (ill->ill_mc_mtu < IPV6_MIN_MTU)
8643 						ill->ill_mc_mtu = IPV6_MIN_MTU;
8644 				} else {
8645 					if (ill->ill_mtu < IP_MIN_MTU)
8646 						ill->ill_mtu = IP_MIN_MTU;
8647 					if (ill->ill_mc_mtu < IP_MIN_MTU)
8648 						ill->ill_mc_mtu = IP_MIN_MTU;
8649 				}
8650 			} else if (ill->ill_mc_mtu > ill->ill_mtu) {
8651 				ill->ill_mc_mtu = ill->ill_mtu;
8652 			}
8653 
8654 			mutex_exit(&ill->ill_lock);
8655 			/*
8656 			 * Make sure all dce_generation checks find out
8657 			 * that ill_mtu/ill_mc_mtu has changed.
8658 			 */
8659 			if (orig_mtu != ill->ill_mtu ||
8660 			    orig_mc_mtu != ill->ill_mc_mtu) {
8661 				dce_increment_all_generations(ill->ill_isv6,
8662 				    ill->ill_ipst);
8663 			}
8664 
8665 			/*
8666 			 * Refresh IPMP meta-interface MTU if necessary.
8667 			 */
8668 			if (IS_UNDER_IPMP(ill))
8669 				ipmp_illgrp_refresh_mtu(ill->ill_grp);
8670 			break;
8671 
8672 		case DL_NOTE_LINK_UP:
8673 		case DL_NOTE_LINK_DOWN: {
8674 			/*
8675 			 * We are writer. ill / phyint / ipsq assocs stable.
8676 			 * The RUNNING flag reflects the state of the link.
8677 			 */
8678 			phyint_t *phyint = ill->ill_phyint;
8679 			uint64_t new_phyint_flags;
8680 			boolean_t changed = B_FALSE;
8681 			boolean_t went_up;
8682 
8683 			went_up = notify->dl_notification == DL_NOTE_LINK_UP;
8684 			mutex_enter(&phyint->phyint_lock);
8685 
8686 			new_phyint_flags = went_up ?
8687 			    phyint->phyint_flags | PHYI_RUNNING :
8688 			    phyint->phyint_flags & ~PHYI_RUNNING;
8689 
8690 			if (IS_IPMP(ill)) {
8691 				new_phyint_flags = went_up ?
8692 				    new_phyint_flags & ~PHYI_FAILED :
8693 				    new_phyint_flags | PHYI_FAILED;
8694 			}
8695 
8696 			if (new_phyint_flags != phyint->phyint_flags) {
8697 				phyint->phyint_flags = new_phyint_flags;
8698 				changed = B_TRUE;
8699 			}
8700 			mutex_exit(&phyint->phyint_lock);
8701 			/*
8702 			 * ill_restart_dad handles the DAD restart and routing
8703 			 * socket notification logic.
8704 			 */
8705 			if (changed) {
8706 				ill_restart_dad(phyint->phyint_illv4, went_up);
8707 				ill_restart_dad(phyint->phyint_illv6, went_up);
8708 			}
8709 			break;
8710 		}
8711 		case DL_NOTE_PROMISC_ON_PHYS: {
8712 			phyint_t *phyint = ill->ill_phyint;
8713 
8714 			mutex_enter(&phyint->phyint_lock);
8715 			phyint->phyint_flags |= PHYI_PROMISC;
8716 			mutex_exit(&phyint->phyint_lock);
8717 			break;
8718 		}
8719 		case DL_NOTE_PROMISC_OFF_PHYS: {
8720 			phyint_t *phyint = ill->ill_phyint;
8721 
8722 			mutex_enter(&phyint->phyint_lock);
8723 			phyint->phyint_flags &= ~PHYI_PROMISC;
8724 			mutex_exit(&phyint->phyint_lock);
8725 			break;
8726 		}
8727 		case DL_NOTE_CAPAB_RENEG:
8728 			/*
8729 			 * Something changed on the driver side.
8730 			 * It wants us to renegotiate the capabilities
8731 			 * on this ill. One possible cause is the aggregation
8732 			 * interface under us where a port got added or
8733 			 * went away.
8734 			 *
8735 			 * If the capability negotiation is already done
8736 			 * or is in progress, reset the capabilities and
8737 			 * mark the ill's ill_capab_reneg to be B_TRUE,
8738 			 * so that when the ack comes back, we can start
8739 			 * the renegotiation process.
8740 			 *
8741 			 * Note that if ill_capab_reneg is already B_TRUE
8742 			 * (ill_dlpi_capab_state is IDS_UNKNOWN in this case),
8743 			 * the capability resetting request has been sent
8744 			 * and the renegotiation has not been started yet;
8745 			 * nothing needs to be done in this case.
8746 			 */
8747 			ipsq_current_start(ipsq, ill->ill_ipif, 0);
8748 			ill_capability_reset(ill, B_TRUE);
8749 			ipsq_current_finish(ipsq);
8750 			break;
8751 
8752 		case DL_NOTE_ALLOWED_IPS:
8753 			ill_set_allowed_ips(ill, mp);
8754 			break;
8755 		default:
8756 			ip0dbg(("ip_rput_dlpi_writer: unknown notification "
8757 			    "type 0x%x for DL_NOTIFY_IND\n",
8758 			    notify->dl_notification));
8759 			break;
8760 		}
8761 
8762 		/*
8763 		 * As this is an asynchronous operation, we
8764 		 * should not call ill_dlpi_done
8765 		 */
8766 		break;
8767 	}
8768 	case DL_NOTIFY_ACK: {
8769 		dl_notify_ack_t *noteack = (dl_notify_ack_t *)mp->b_rptr;
8770 
8771 		if (noteack->dl_notifications & DL_NOTE_LINK_UP)
8772 			ill->ill_note_link = 1;
8773 		ill_dlpi_done(ill, DL_NOTIFY_REQ);
8774 		break;
8775 	}
8776 	case DL_PHYS_ADDR_ACK: {
8777 		/*
8778 		 * As part of plumbing the interface via SIOCSLIFNAME,
8779 		 * ill_dl_phys() will queue a series of DL_PHYS_ADDR_REQs,
8780 		 * whose answers we receive here.  As each answer is received,
8781 		 * we call ill_dlpi_done() to dispatch the next request as
8782 		 * we're processing the current one.  Once all answers have
8783 		 * been received, we use ipsq_pending_mp_get() to dequeue the
8784 		 * outstanding IOCTL and reply to it.  (Because ill_dl_phys()
8785 		 * is invoked from an ill queue, conn_oper_pending_ill is not
8786 		 * available, but we know the ioctl is pending on ill_wq.)
8787 		 */
8788 		uint_t	paddrlen, paddroff;
8789 		uint8_t	*addr;
8790 
8791 		paddrreq = ill->ill_phys_addr_pend;
8792 		paddrlen = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_length;
8793 		paddroff = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_offset;
8794 		addr = mp->b_rptr + paddroff;
8795 
8796 		ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8797 		if (paddrreq == DL_IPV6_TOKEN) {
8798 			/*
8799 			 * bcopy to low-order bits of ill_token
8800 			 *
8801 			 * XXX Temporary hack - currently, all known tokens
8802 			 * are 64 bits, so I'll cheat for the moment.
8803 			 */
8804 			bcopy(addr, &ill->ill_token.s6_addr32[2], paddrlen);
8805 			ill->ill_token_length = paddrlen;
8806 			break;
8807 		} else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8808 			ASSERT(ill->ill_nd_lla_mp == NULL);
8809 			ill_set_ndmp(ill, mp, paddroff, paddrlen);
8810 			mp = NULL;
8811 			break;
8812 		} else if (paddrreq == DL_CURR_DEST_ADDR) {
8813 			ASSERT(ill->ill_dest_addr_mp == NULL);
8814 			ill->ill_dest_addr_mp = mp;
8815 			ill->ill_dest_addr = addr;
8816 			mp = NULL;
8817 			if (ill->ill_isv6) {
8818 				ill_setdesttoken(ill);
8819 				ipif_setdestlinklocal(ill->ill_ipif);
8820 			}
8821 			break;
8822 		}
8823 
8824 		ASSERT(paddrreq == DL_CURR_PHYS_ADDR);
8825 		ASSERT(ill->ill_phys_addr_mp == NULL);
8826 		if (!ill->ill_ifname_pending)
8827 			break;
8828 		ill->ill_ifname_pending = 0;
8829 		if (!ioctl_aborted)
8830 			mp1 = ipsq_pending_mp_get(ipsq, &connp);
8831 		if (mp1 != NULL) {
8832 			ASSERT(connp == NULL);
8833 			q = ill->ill_wq;
8834 		}
8835 		/*
8836 		 * If any error acks received during the plumbing sequence,
8837 		 * ill_ifname_pending_err will be set. Break out and send up
8838 		 * the error to the pending ioctl.
8839 		 */
8840 		if (ill->ill_ifname_pending_err != 0) {
8841 			err = ill->ill_ifname_pending_err;
8842 			ill->ill_ifname_pending_err = 0;
8843 			break;
8844 		}
8845 
8846 		ill->ill_phys_addr_mp = mp;
8847 		ill->ill_phys_addr = (paddrlen == 0 ? NULL : addr);
8848 		mp = NULL;
8849 
8850 		/*
8851 		 * If paddrlen or ill_phys_addr_length is zero, the DLPI
8852 		 * provider doesn't support physical addresses.  We check both
8853 		 * paddrlen and ill_phys_addr_length because sppp (PPP) does
8854 		 * not have physical addresses, but historically adversises a
8855 		 * physical address length of 0 in its DL_INFO_ACK, but 6 in
8856 		 * its DL_PHYS_ADDR_ACK.
8857 		 */
8858 		if (paddrlen == 0 || ill->ill_phys_addr_length == 0) {
8859 			ill->ill_phys_addr = NULL;
8860 		} else if (paddrlen != ill->ill_phys_addr_length) {
8861 			ip0dbg(("DL_PHYS_ADDR_ACK: got addrlen %d, expected %d",
8862 			    paddrlen, ill->ill_phys_addr_length));
8863 			err = EINVAL;
8864 			break;
8865 		}
8866 
8867 		if (ill->ill_nd_lla_mp == NULL) {
8868 			if ((mp_hw = copyb(ill->ill_phys_addr_mp)) == NULL) {
8869 				err = ENOMEM;
8870 				break;
8871 			}
8872 			ill_set_ndmp(ill, mp_hw, paddroff, paddrlen);
8873 		}
8874 
8875 		if (ill->ill_isv6) {
8876 			ill_setdefaulttoken(ill);
8877 			ipif_setlinklocal(ill->ill_ipif);
8878 		}
8879 		break;
8880 	}
8881 	case DL_OK_ACK:
8882 		ip2dbg(("DL_OK_ACK %s (0x%x)\n",
8883 		    dl_primstr((int)dloa->dl_correct_primitive),
8884 		    dloa->dl_correct_primitive));
8885 		DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer ok",
8886 		    char *, dl_primstr(dloa->dl_correct_primitive),
8887 		    ill_t *, ill);
8888 
8889 		switch (dloa->dl_correct_primitive) {
8890 		case DL_ENABMULTI_REQ:
8891 		case DL_DISABMULTI_REQ:
8892 			ill_dlpi_done(ill, dloa->dl_correct_primitive);
8893 			break;
8894 		case DL_PROMISCON_REQ:
8895 		case DL_PROMISCOFF_REQ:
8896 		case DL_UNBIND_REQ:
8897 		case DL_ATTACH_REQ:
8898 			ill_dlpi_done(ill, dloa->dl_correct_primitive);
8899 			break;
8900 		}
8901 		break;
8902 	default:
8903 		break;
8904 	}
8905 
8906 	freemsg(mp);
8907 	if (mp1 == NULL)
8908 		return;
8909 
8910 	/*
8911 	 * The operation must complete without EINPROGRESS since
8912 	 * ipsq_pending_mp_get() has removed the mblk (mp1).  Otherwise,
8913 	 * the operation will be stuck forever inside the IPSQ.
8914 	 */
8915 	ASSERT(err != EINPROGRESS);
8916 
8917 	DTRACE_PROBE4(ipif__ioctl, char *, "ip_rput_dlpi_writer finish",
8918 	    int, ipsq->ipsq_xop->ipx_current_ioctl, ill_t *, ill,
8919 	    ipif_t *, NULL);
8920 
8921 	switch (ipsq->ipsq_xop->ipx_current_ioctl) {
8922 	case 0:
8923 		ipsq_current_finish(ipsq);
8924 		break;
8925 
8926 	case SIOCSLIFNAME:
8927 	case IF_UNITSEL: {
8928 		ill_t *ill_other = ILL_OTHER(ill);
8929 
8930 		/*
8931 		 * If SIOCSLIFNAME or IF_UNITSEL is about to succeed, and the
8932 		 * ill has a peer which is in an IPMP group, then place ill
8933 		 * into the same group.  One catch: although ifconfig plumbs
8934 		 * the appropriate IPMP meta-interface prior to plumbing this
8935 		 * ill, it is possible for multiple ifconfig applications to
8936 		 * race (or for another application to adjust plumbing), in
8937 		 * which case the IPMP meta-interface we need will be missing.
8938 		 * If so, kick the phyint out of the group.
8939 		 */
8940 		if (err == 0 && ill_other != NULL && IS_UNDER_IPMP(ill_other)) {
8941 			ipmp_grp_t	*grp = ill->ill_phyint->phyint_grp;
8942 			ipmp_illgrp_t	*illg;
8943 
8944 			illg = ill->ill_isv6 ? grp->gr_v6 : grp->gr_v4;
8945 			if (illg == NULL)
8946 				ipmp_phyint_leave_grp(ill->ill_phyint);
8947 			else
8948 				ipmp_ill_join_illgrp(ill, illg);
8949 		}
8950 
8951 		if (ipsq->ipsq_xop->ipx_current_ioctl == IF_UNITSEL)
8952 			ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8953 		else
8954 			ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8955 		break;
8956 	}
8957 	case SIOCLIFADDIF:
8958 		ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8959 		break;
8960 
8961 	default:
8962 		ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8963 		break;
8964 	}
8965 }
8966 
8967 /*
8968  * ip_rput_other is called by ip_rput to handle messages modifying the global
8969  * state in IP.  If 'ipsq' is non-NULL, caller is writer on it.
8970  */
8971 /* ARGSUSED */
8972 void
ip_rput_other(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * dummy_arg)8973 ip_rput_other(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8974 {
8975 	ill_t		*ill = q->q_ptr;
8976 	struct iocblk	*iocp;
8977 
8978 	ip1dbg(("ip_rput_other "));
8979 	if (ipsq != NULL) {
8980 		ASSERT(IAM_WRITER_IPSQ(ipsq));
8981 		ASSERT(ipsq->ipsq_xop ==
8982 		    ill->ill_phyint->phyint_ipsq->ipsq_xop);
8983 	}
8984 
8985 	switch (mp->b_datap->db_type) {
8986 	case M_ERROR:
8987 	case M_HANGUP:
8988 		/*
8989 		 * The device has a problem.  We force the ILL down.  It can
8990 		 * be brought up again manually using SIOCSIFFLAGS (via
8991 		 * ifconfig or equivalent).
8992 		 */
8993 		ASSERT(ipsq != NULL);
8994 		if (mp->b_rptr < mp->b_wptr)
8995 			ill->ill_error = (int)(*mp->b_rptr & 0xFF);
8996 		if (ill->ill_error == 0)
8997 			ill->ill_error = ENXIO;
8998 		if (!ill_down_start(q, mp))
8999 			return;
9000 		ipif_all_down_tail(ipsq, q, mp, NULL);
9001 		break;
9002 	case M_IOCNAK: {
9003 		iocp = (struct iocblk *)mp->b_rptr;
9004 
9005 		ASSERT(iocp->ioc_cmd == DL_IOC_HDR_INFO);
9006 		/*
9007 		 * If this was the first attempt, turn off the fastpath
9008 		 * probing.
9009 		 */
9010 		mutex_enter(&ill->ill_lock);
9011 		if (ill->ill_dlpi_fastpath_state == IDS_INPROGRESS) {
9012 			ill->ill_dlpi_fastpath_state = IDS_FAILED;
9013 			mutex_exit(&ill->ill_lock);
9014 			/*
9015 			 * don't flush the nce_t entries: we use them
9016 			 * as an index to the ncec itself.
9017 			 */
9018 			ip1dbg(("ip_rput: DLPI fastpath off on interface %s\n",
9019 			    ill->ill_name));
9020 		} else {
9021 			mutex_exit(&ill->ill_lock);
9022 		}
9023 		freemsg(mp);
9024 		break;
9025 	}
9026 	default:
9027 		ASSERT(0);
9028 		break;
9029 	}
9030 }
9031 
9032 /*
9033  * Update any source route, record route or timestamp options
9034  * When it fails it has consumed the message and BUMPed the MIB.
9035  */
9036 boolean_t
ip_forward_options(mblk_t * mp,ipha_t * ipha,ill_t * dst_ill,ip_recv_attr_t * ira)9037 ip_forward_options(mblk_t *mp, ipha_t *ipha, ill_t *dst_ill,
9038     ip_recv_attr_t *ira)
9039 {
9040 	ipoptp_t	opts;
9041 	uchar_t		*opt;
9042 	uint8_t		optval;
9043 	uint8_t		optlen;
9044 	ipaddr_t	dst;
9045 	ipaddr_t	ifaddr;
9046 	uint32_t	ts;
9047 	timestruc_t	now;
9048 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
9049 
9050 	ip2dbg(("ip_forward_options\n"));
9051 	dst = ipha->ipha_dst;
9052 	opt = NULL;
9053 
9054 	for (optval = ipoptp_first(&opts, ipha);
9055 	    optval != IPOPT_EOL;
9056 	    optval = ipoptp_next(&opts)) {
9057 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9058 		opt = opts.ipoptp_cur;
9059 		optlen = opts.ipoptp_len;
9060 		ip2dbg(("ip_forward_options: opt %d, len %d\n",
9061 		    optval, opts.ipoptp_len));
9062 		switch (optval) {
9063 			uint32_t off;
9064 		case IPOPT_SSRR:
9065 		case IPOPT_LSRR:
9066 			/* Check if adminstratively disabled */
9067 			if (!ipst->ips_ip_forward_src_routed) {
9068 				BUMP_MIB(dst_ill->ill_ip_mib,
9069 				    ipIfStatsForwProhibits);
9070 				ip_drop_input("ICMP_SOURCE_ROUTE_FAILED",
9071 				    mp, dst_ill);
9072 				icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED,
9073 				    ira);
9074 				return (B_FALSE);
9075 			}
9076 			if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9077 				/*
9078 				 * Must be partial since ip_input_options
9079 				 * checked for strict.
9080 				 */
9081 				break;
9082 			}
9083 			off = opt[IPOPT_OFFSET];
9084 			off--;
9085 		redo_srr:
9086 			if (optlen < IP_ADDR_LEN ||
9087 			    off > optlen - IP_ADDR_LEN) {
9088 				/* End of source route */
9089 				ip1dbg((
9090 				    "ip_forward_options: end of SR\n"));
9091 				break;
9092 			}
9093 			/* Pick a reasonable address on the outbound if */
9094 			ASSERT(dst_ill != NULL);
9095 			if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9096 			    INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9097 			    NULL) != 0) {
9098 				/* No source! Shouldn't happen */
9099 				ifaddr = INADDR_ANY;
9100 			}
9101 			bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9102 			bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9103 			ip1dbg(("ip_forward_options: next hop 0x%x\n",
9104 			    ntohl(dst)));
9105 
9106 			/*
9107 			 * Check if our address is present more than
9108 			 * once as consecutive hops in source route.
9109 			 */
9110 			if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9111 				off += IP_ADDR_LEN;
9112 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9113 				goto redo_srr;
9114 			}
9115 			ipha->ipha_dst = dst;
9116 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9117 			break;
9118 		case IPOPT_RR:
9119 			off = opt[IPOPT_OFFSET];
9120 			off--;
9121 			if (optlen < IP_ADDR_LEN ||
9122 			    off > optlen - IP_ADDR_LEN) {
9123 				/* No more room - ignore */
9124 				ip1dbg((
9125 				    "ip_forward_options: end of RR\n"));
9126 				break;
9127 			}
9128 			/* Pick a reasonable address on the outbound if */
9129 			ASSERT(dst_ill != NULL);
9130 			if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9131 			    INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9132 			    NULL) != 0) {
9133 				/* No source! Shouldn't happen */
9134 				ifaddr = INADDR_ANY;
9135 			}
9136 			bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9137 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9138 			break;
9139 		case IPOPT_TS:
9140 			off = 0;
9141 			/* Insert timestamp if there is room */
9142 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9143 			case IPOPT_TS_TSONLY:
9144 				off = IPOPT_TS_TIMELEN;
9145 				break;
9146 			case IPOPT_TS_PRESPEC:
9147 			case IPOPT_TS_PRESPEC_RFC791:
9148 				/* Verify that the address matched */
9149 				off = opt[IPOPT_OFFSET] - 1;
9150 				bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9151 				if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9152 					/* Not for us */
9153 					break;
9154 				}
9155 				/* FALLTHROUGH */
9156 			case IPOPT_TS_TSANDADDR:
9157 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9158 				break;
9159 			default:
9160 				/*
9161 				 * ip_*put_options should have already
9162 				 * dropped this packet.
9163 				 */
9164 				cmn_err(CE_PANIC, "ip_forward_options: "
9165 				    "unknown IT - bug in ip_input_options?\n");
9166 			}
9167 			if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9168 				/* Increase overflow counter */
9169 				off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9170 				opt[IPOPT_POS_OV_FLG] =
9171 				    (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9172 				    (off << 4));
9173 				break;
9174 			}
9175 			off = opt[IPOPT_OFFSET] - 1;
9176 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9177 			case IPOPT_TS_PRESPEC:
9178 			case IPOPT_TS_PRESPEC_RFC791:
9179 			case IPOPT_TS_TSANDADDR:
9180 				/* Pick a reasonable addr on the outbound if */
9181 				ASSERT(dst_ill != NULL);
9182 				if (ip_select_source_v4(dst_ill, INADDR_ANY,
9183 				    dst, INADDR_ANY, ALL_ZONES, ipst, &ifaddr,
9184 				    NULL, NULL) != 0) {
9185 					/* No source! Shouldn't happen */
9186 					ifaddr = INADDR_ANY;
9187 				}
9188 				bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9189 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9190 				/* FALLTHROUGH */
9191 			case IPOPT_TS_TSONLY:
9192 				off = opt[IPOPT_OFFSET] - 1;
9193 				/* Compute # of milliseconds since midnight */
9194 				gethrestime(&now);
9195 				ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9196 				    NSEC2MSEC(now.tv_nsec);
9197 				bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9198 				opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9199 				break;
9200 			}
9201 			break;
9202 		}
9203 	}
9204 	return (B_TRUE);
9205 }
9206 
9207 /*
9208  * Call ill_frag_timeout to do garbage collection. ill_frag_timeout
9209  * returns 'true' if there are still fragments left on the queue, in
9210  * which case we restart the timer.
9211  */
9212 void
ill_frag_timer(void * arg)9213 ill_frag_timer(void *arg)
9214 {
9215 	ill_t	*ill = (ill_t *)arg;
9216 	boolean_t frag_pending;
9217 	ip_stack_t *ipst = ill->ill_ipst;
9218 	time_t	timeout;
9219 
9220 	mutex_enter(&ill->ill_lock);
9221 	ASSERT(!ill->ill_fragtimer_executing);
9222 	if (ill->ill_state_flags & ILL_CONDEMNED) {
9223 		ill->ill_frag_timer_id = 0;
9224 		mutex_exit(&ill->ill_lock);
9225 		return;
9226 	}
9227 	ill->ill_fragtimer_executing = 1;
9228 	mutex_exit(&ill->ill_lock);
9229 
9230 	timeout = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9231 	    ipst->ips_ip_reassembly_timeout);
9232 
9233 	frag_pending = ill_frag_timeout(ill, timeout);
9234 
9235 	/*
9236 	 * Restart the timer, if we have fragments pending or if someone
9237 	 * wanted us to be scheduled again.
9238 	 */
9239 	mutex_enter(&ill->ill_lock);
9240 	ill->ill_fragtimer_executing = 0;
9241 	ill->ill_frag_timer_id = 0;
9242 	if (frag_pending || ill->ill_fragtimer_needrestart)
9243 		ill_frag_timer_start(ill);
9244 	mutex_exit(&ill->ill_lock);
9245 }
9246 
9247 void
ill_frag_timer_start(ill_t * ill)9248 ill_frag_timer_start(ill_t *ill)
9249 {
9250 	ip_stack_t *ipst = ill->ill_ipst;
9251 	clock_t	timeo_ms;
9252 
9253 	ASSERT(MUTEX_HELD(&ill->ill_lock));
9254 
9255 	/* If the ill is closing or opening don't proceed */
9256 	if (ill->ill_state_flags & ILL_CONDEMNED)
9257 		return;
9258 
9259 	if (ill->ill_fragtimer_executing) {
9260 		/*
9261 		 * ill_frag_timer is currently executing. Just record the
9262 		 * the fact that we want the timer to be restarted.
9263 		 * ill_frag_timer will post a timeout before it returns,
9264 		 * ensuring it will be called again.
9265 		 */
9266 		ill->ill_fragtimer_needrestart = 1;
9267 		return;
9268 	}
9269 
9270 	if (ill->ill_frag_timer_id == 0) {
9271 		timeo_ms = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9272 		    ipst->ips_ip_reassembly_timeout) * SECONDS;
9273 
9274 		/*
9275 		 * The timer is neither running nor is the timeout handler
9276 		 * executing. Post a timeout so that ill_frag_timer will be
9277 		 * called
9278 		 */
9279 		ill->ill_frag_timer_id = timeout(ill_frag_timer, ill,
9280 		    MSEC_TO_TICK(timeo_ms >> 1));
9281 		ill->ill_fragtimer_needrestart = 0;
9282 	}
9283 }
9284 
9285 /*
9286  * Update any source route, record route or timestamp options.
9287  * Check that we are at end of strict source route.
9288  * The options have already been checked for sanity in ip_input_options().
9289  */
9290 boolean_t
ip_input_local_options(mblk_t * mp,ipha_t * ipha,ip_recv_attr_t * ira)9291 ip_input_local_options(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
9292 {
9293 	ipoptp_t	opts;
9294 	uchar_t		*opt;
9295 	uint8_t		optval;
9296 	uint8_t		optlen;
9297 	ipaddr_t	dst;
9298 	ipaddr_t	ifaddr;
9299 	uint32_t	ts;
9300 	timestruc_t	now;
9301 	ill_t		*ill = ira->ira_ill;
9302 	ip_stack_t	*ipst = ill->ill_ipst;
9303 
9304 	ip2dbg(("ip_input_local_options\n"));
9305 	opt = NULL;
9306 
9307 	for (optval = ipoptp_first(&opts, ipha);
9308 	    optval != IPOPT_EOL;
9309 	    optval = ipoptp_next(&opts)) {
9310 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9311 		opt = opts.ipoptp_cur;
9312 		optlen = opts.ipoptp_len;
9313 		ip2dbg(("ip_input_local_options: opt %d, len %d\n",
9314 		    optval, optlen));
9315 		switch (optval) {
9316 			uint32_t off;
9317 		case IPOPT_SSRR:
9318 		case IPOPT_LSRR:
9319 			off = opt[IPOPT_OFFSET];
9320 			off--;
9321 			if (optlen < IP_ADDR_LEN ||
9322 			    off > optlen - IP_ADDR_LEN) {
9323 				/* End of source route */
9324 				ip1dbg(("ip_input_local_options: end of SR\n"));
9325 				break;
9326 			}
9327 			/*
9328 			 * This will only happen if two consecutive entries
9329 			 * in the source route contains our address or if
9330 			 * it is a packet with a loose source route which
9331 			 * reaches us before consuming the whole source route
9332 			 */
9333 			ip1dbg(("ip_input_local_options: not end of SR\n"));
9334 			if (optval == IPOPT_SSRR) {
9335 				goto bad_src_route;
9336 			}
9337 			/*
9338 			 * Hack: instead of dropping the packet truncate the
9339 			 * source route to what has been used by filling the
9340 			 * rest with IPOPT_NOP.
9341 			 */
9342 			opt[IPOPT_OLEN] = (uint8_t)off;
9343 			while (off < optlen) {
9344 				opt[off++] = IPOPT_NOP;
9345 			}
9346 			break;
9347 		case IPOPT_RR:
9348 			off = opt[IPOPT_OFFSET];
9349 			off--;
9350 			if (optlen < IP_ADDR_LEN ||
9351 			    off > optlen - IP_ADDR_LEN) {
9352 				/* No more room - ignore */
9353 				ip1dbg((
9354 				    "ip_input_local_options: end of RR\n"));
9355 				break;
9356 			}
9357 			/* Pick a reasonable address on the outbound if */
9358 			if (ip_select_source_v4(ill, INADDR_ANY, ipha->ipha_dst,
9359 			    INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9360 			    NULL) != 0) {
9361 				/* No source! Shouldn't happen */
9362 				ifaddr = INADDR_ANY;
9363 			}
9364 			bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9365 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9366 			break;
9367 		case IPOPT_TS:
9368 			off = 0;
9369 			/* Insert timestamp if there is romm */
9370 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9371 			case IPOPT_TS_TSONLY:
9372 				off = IPOPT_TS_TIMELEN;
9373 				break;
9374 			case IPOPT_TS_PRESPEC:
9375 			case IPOPT_TS_PRESPEC_RFC791:
9376 				/* Verify that the address matched */
9377 				off = opt[IPOPT_OFFSET] - 1;
9378 				bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9379 				if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9380 					/* Not for us */
9381 					break;
9382 				}
9383 				/* FALLTHROUGH */
9384 			case IPOPT_TS_TSANDADDR:
9385 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9386 				break;
9387 			default:
9388 				/*
9389 				 * ip_*put_options should have already
9390 				 * dropped this packet.
9391 				 */
9392 				cmn_err(CE_PANIC, "ip_input_local_options: "
9393 				    "unknown IT - bug in ip_input_options?\n");
9394 			}
9395 			if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9396 				/* Increase overflow counter */
9397 				off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9398 				opt[IPOPT_POS_OV_FLG] =
9399 				    (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9400 				    (off << 4));
9401 				break;
9402 			}
9403 			off = opt[IPOPT_OFFSET] - 1;
9404 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9405 			case IPOPT_TS_PRESPEC:
9406 			case IPOPT_TS_PRESPEC_RFC791:
9407 			case IPOPT_TS_TSANDADDR:
9408 				/* Pick a reasonable addr on the outbound if */
9409 				if (ip_select_source_v4(ill, INADDR_ANY,
9410 				    ipha->ipha_dst, INADDR_ANY, ALL_ZONES, ipst,
9411 				    &ifaddr, NULL, NULL) != 0) {
9412 					/* No source! Shouldn't happen */
9413 					ifaddr = INADDR_ANY;
9414 				}
9415 				bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9416 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9417 				/* FALLTHROUGH */
9418 			case IPOPT_TS_TSONLY:
9419 				off = opt[IPOPT_OFFSET] - 1;
9420 				/* Compute # of milliseconds since midnight */
9421 				gethrestime(&now);
9422 				ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9423 				    NSEC2MSEC(now.tv_nsec);
9424 				bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9425 				opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9426 				break;
9427 			}
9428 			break;
9429 		}
9430 	}
9431 	return (B_TRUE);
9432 
9433 bad_src_route:
9434 	/* make sure we clear any indication of a hardware checksum */
9435 	DB_CKSUMFLAGS(mp) = 0;
9436 	ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
9437 	icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9438 	return (B_FALSE);
9439 
9440 }
9441 
9442 /*
9443  * Process IP options in an inbound packet.  Always returns the nexthop.
9444  * Normally this is the passed in nexthop, but if there is an option
9445  * that effects the nexthop (such as a source route) that will be returned.
9446  * Sets *errorp if there is an error, in which case an ICMP error has been sent
9447  * and mp freed.
9448  */
9449 ipaddr_t
ip_input_options(ipha_t * ipha,ipaddr_t dst,mblk_t * mp,ip_recv_attr_t * ira,int * errorp)9450 ip_input_options(ipha_t *ipha, ipaddr_t dst, mblk_t *mp,
9451     ip_recv_attr_t *ira, int *errorp)
9452 {
9453 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
9454 	ipoptp_t	opts;
9455 	uchar_t		*opt;
9456 	uint8_t		optval;
9457 	uint8_t		optlen;
9458 	intptr_t	code = 0;
9459 	ire_t		*ire;
9460 
9461 	ip2dbg(("ip_input_options\n"));
9462 	opt = NULL;
9463 	*errorp = 0;
9464 	for (optval = ipoptp_first(&opts, ipha);
9465 	    optval != IPOPT_EOL;
9466 	    optval = ipoptp_next(&opts)) {
9467 		opt = opts.ipoptp_cur;
9468 		optlen = opts.ipoptp_len;
9469 		ip2dbg(("ip_input_options: opt %d, len %d\n",
9470 		    optval, optlen));
9471 		/*
9472 		 * Note: we need to verify the checksum before we
9473 		 * modify anything thus this routine only extracts the next
9474 		 * hop dst from any source route.
9475 		 */
9476 		switch (optval) {
9477 			uint32_t off;
9478 		case IPOPT_SSRR:
9479 		case IPOPT_LSRR:
9480 			if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9481 				if (optval == IPOPT_SSRR) {
9482 					ip1dbg(("ip_input_options: not next"
9483 					    " strict source route 0x%x\n",
9484 					    ntohl(dst)));
9485 					code = (char *)&ipha->ipha_dst -
9486 					    (char *)ipha;
9487 					goto param_prob; /* RouterReq's */
9488 				}
9489 				ip2dbg(("ip_input_options: "
9490 				    "not next source route 0x%x\n",
9491 				    ntohl(dst)));
9492 				break;
9493 			}
9494 
9495 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9496 				ip1dbg((
9497 				    "ip_input_options: bad option offset\n"));
9498 				code = (char *)&opt[IPOPT_OLEN] -
9499 				    (char *)ipha;
9500 				goto param_prob;
9501 			}
9502 			off = opt[IPOPT_OFFSET];
9503 			off--;
9504 		redo_srr:
9505 			if (optlen < IP_ADDR_LEN ||
9506 			    off > optlen - IP_ADDR_LEN) {
9507 				/* End of source route */
9508 				ip1dbg(("ip_input_options: end of SR\n"));
9509 				break;
9510 			}
9511 			bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9512 			ip1dbg(("ip_input_options: next hop 0x%x\n",
9513 			    ntohl(dst)));
9514 
9515 			/*
9516 			 * Check if our address is present more than
9517 			 * once as consecutive hops in source route.
9518 			 * XXX verify per-interface ip_forwarding
9519 			 * for source route?
9520 			 */
9521 			if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9522 				off += IP_ADDR_LEN;
9523 				goto redo_srr;
9524 			}
9525 
9526 			if (dst == htonl(INADDR_LOOPBACK)) {
9527 				ip1dbg(("ip_input_options: loopback addr in "
9528 				    "source route!\n"));
9529 				goto bad_src_route;
9530 			}
9531 			/*
9532 			 * For strict: verify that dst is directly
9533 			 * reachable.
9534 			 */
9535 			if (optval == IPOPT_SSRR) {
9536 				ire = ire_ftable_lookup_v4(dst, 0, 0,
9537 				    IRE_INTERFACE, NULL, ALL_ZONES,
9538 				    ira->ira_tsl,
9539 				    MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
9540 				    NULL);
9541 				if (ire == NULL) {
9542 					ip1dbg(("ip_input_options: SSRR not "
9543 					    "directly reachable: 0x%x\n",
9544 					    ntohl(dst)));
9545 					goto bad_src_route;
9546 				}
9547 				ire_refrele(ire);
9548 			}
9549 			/*
9550 			 * Defer update of the offset and the record route
9551 			 * until the packet is forwarded.
9552 			 */
9553 			break;
9554 		case IPOPT_RR:
9555 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9556 				ip1dbg((
9557 				    "ip_input_options: bad option offset\n"));
9558 				code = (char *)&opt[IPOPT_OLEN] -
9559 				    (char *)ipha;
9560 				goto param_prob;
9561 			}
9562 			break;
9563 		case IPOPT_TS:
9564 			/*
9565 			 * Verify that length >= 5 and that there is either
9566 			 * room for another timestamp or that the overflow
9567 			 * counter is not maxed out.
9568 			 */
9569 			code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
9570 			if (optlen < IPOPT_MINLEN_IT) {
9571 				goto param_prob;
9572 			}
9573 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9574 				ip1dbg((
9575 				    "ip_input_options: bad option offset\n"));
9576 				code = (char *)&opt[IPOPT_OFFSET] -
9577 				    (char *)ipha;
9578 				goto param_prob;
9579 			}
9580 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9581 			case IPOPT_TS_TSONLY:
9582 				off = IPOPT_TS_TIMELEN;
9583 				break;
9584 			case IPOPT_TS_TSANDADDR:
9585 			case IPOPT_TS_PRESPEC:
9586 			case IPOPT_TS_PRESPEC_RFC791:
9587 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9588 				break;
9589 			default:
9590 				code = (char *)&opt[IPOPT_POS_OV_FLG] -
9591 				    (char *)ipha;
9592 				goto param_prob;
9593 			}
9594 			if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
9595 			    (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
9596 				/*
9597 				 * No room and the overflow counter is 15
9598 				 * already.
9599 				 */
9600 				goto param_prob;
9601 			}
9602 			break;
9603 		}
9604 	}
9605 
9606 	if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0) {
9607 		return (dst);
9608 	}
9609 
9610 	ip1dbg(("ip_input_options: error processing IP options."));
9611 	code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
9612 
9613 param_prob:
9614 	/* make sure we clear any indication of a hardware checksum */
9615 	DB_CKSUMFLAGS(mp) = 0;
9616 	ip_drop_input("ICMP_PARAM_PROBLEM", mp, ira->ira_ill);
9617 	icmp_param_problem(mp, (uint8_t)code, ira);
9618 	*errorp = -1;
9619 	return (dst);
9620 
9621 bad_src_route:
9622 	/* make sure we clear any indication of a hardware checksum */
9623 	DB_CKSUMFLAGS(mp) = 0;
9624 	ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ira->ira_ill);
9625 	icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9626 	*errorp = -1;
9627 	return (dst);
9628 }
9629 
9630 /*
9631  * IP & ICMP info in >=14 msg's ...
9632  *  - ip fixed part (mib2_ip_t)
9633  *  - icmp fixed part (mib2_icmp_t)
9634  *  - ipAddrEntryTable (ip 20)		all IPv4 ipifs
9635  *  - ipRouteEntryTable (ip 21)		all IPv4 IREs
9636  *  - ipNetToMediaEntryTable (ip 22)	all IPv4 Neighbor Cache entries
9637  *  - ipRouteAttributeTable (ip 102)	labeled routes
9638  *  - ip multicast membership (ip_member_t)
9639  *  - ip multicast source filtering (ip_grpsrc_t)
9640  *  - igmp fixed part (struct igmpstat)
9641  *  - multicast routing stats (struct mrtstat)
9642  *  - multicast routing vifs (array of struct vifctl)
9643  *  - multicast routing routes (array of struct mfcctl)
9644  *  - ip6 fixed part (mib2_ipv6IfStatsEntry_t)
9645  *					One per ill plus one generic
9646  *  - icmp6 fixed part (mib2_ipv6IfIcmpEntry_t)
9647  *					One per ill plus one generic
9648  *  - ipv6RouteEntry			all IPv6 IREs
9649  *  - ipv6RouteAttributeTable (ip6 102)	labeled routes
9650  *  - ipv6NetToMediaEntry		all IPv6 Neighbor Cache entries
9651  *  - ipv6AddrEntry			all IPv6 ipifs
9652  *  - ipv6 multicast membership (ipv6_member_t)
9653  *  - ipv6 multicast source filtering (ipv6_grpsrc_t)
9654  *
9655  * NOTE: original mpctl is copied for msg's 2..N, since its ctl part is
9656  * already filled in by the caller.
9657  * If legacy_req is true then MIB structures needs to be truncated to their
9658  * legacy sizes before being returned.
9659  * Return value of 0 indicates that no messages were sent and caller
9660  * should free mpctl.
9661  */
9662 int
ip_snmp_get(queue_t * q,mblk_t * mpctl,int level,boolean_t legacy_req)9663 ip_snmp_get(queue_t *q, mblk_t *mpctl, int level, boolean_t legacy_req)
9664 {
9665 	ip_stack_t *ipst;
9666 	sctp_stack_t *sctps;
9667 
9668 	if (q->q_next != NULL) {
9669 		ipst = ILLQ_TO_IPST(q);
9670 	} else {
9671 		ipst = CONNQ_TO_IPST(q);
9672 	}
9673 	ASSERT(ipst != NULL);
9674 	sctps = ipst->ips_netstack->netstack_sctp;
9675 
9676 	if (mpctl == NULL || mpctl->b_cont == NULL) {
9677 		return (0);
9678 	}
9679 
9680 	/*
9681 	 * For the purposes of the (broken) packet shell use
9682 	 * of the level we make sure MIB2_TCP/MIB2_UDP can be used
9683 	 * to make TCP and UDP appear first in the list of mib items.
9684 	 * TBD: We could expand this and use it in netstat so that
9685 	 * the kernel doesn't have to produce large tables (connections,
9686 	 * routes, etc) when netstat only wants the statistics or a particular
9687 	 * table.
9688 	 */
9689 	if (!(level == MIB2_TCP || level == MIB2_UDP)) {
9690 		if ((mpctl = icmp_snmp_get(q, mpctl)) == NULL) {
9691 			return (1);
9692 		}
9693 	}
9694 
9695 	if (level != MIB2_TCP) {
9696 		if ((mpctl = udp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9697 			return (1);
9698 		}
9699 		if (level == MIB2_UDP) {
9700 			goto done;
9701 		}
9702 	}
9703 
9704 	if (level != MIB2_UDP) {
9705 		if ((mpctl = tcp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9706 			return (1);
9707 		}
9708 		if (level == MIB2_TCP) {
9709 			goto done;
9710 		}
9711 	}
9712 
9713 	if ((mpctl = ip_snmp_get_mib2_ip_traffic_stats(q, mpctl,
9714 	    ipst, legacy_req)) == NULL) {
9715 		return (1);
9716 	}
9717 
9718 	if ((mpctl = ip_snmp_get_mib2_ip6(q, mpctl, ipst,
9719 	    legacy_req)) == NULL) {
9720 		return (1);
9721 	}
9722 
9723 	if ((mpctl = ip_snmp_get_mib2_icmp(q, mpctl, ipst)) == NULL) {
9724 		return (1);
9725 	}
9726 
9727 	if ((mpctl = ip_snmp_get_mib2_icmp6(q, mpctl, ipst)) == NULL) {
9728 		return (1);
9729 	}
9730 
9731 	if ((mpctl = ip_snmp_get_mib2_igmp(q, mpctl, ipst)) == NULL) {
9732 		return (1);
9733 	}
9734 
9735 	if ((mpctl = ip_snmp_get_mib2_multi(q, mpctl, ipst)) == NULL) {
9736 		return (1);
9737 	}
9738 
9739 	if ((mpctl = ip_snmp_get_mib2_ip_addr(q, mpctl, ipst,
9740 	    legacy_req)) == NULL) {
9741 		return (1);
9742 	}
9743 
9744 	if ((mpctl = ip_snmp_get_mib2_ip6_addr(q, mpctl, ipst,
9745 	    legacy_req)) == NULL) {
9746 		return (1);
9747 	}
9748 
9749 	if ((mpctl = ip_snmp_get_mib2_ip_group_mem(q, mpctl, ipst)) == NULL) {
9750 		return (1);
9751 	}
9752 
9753 	if ((mpctl = ip_snmp_get_mib2_ip6_group_mem(q, mpctl, ipst)) == NULL) {
9754 		return (1);
9755 	}
9756 
9757 	if ((mpctl = ip_snmp_get_mib2_ip_group_src(q, mpctl, ipst)) == NULL) {
9758 		return (1);
9759 	}
9760 
9761 	if ((mpctl = ip_snmp_get_mib2_ip6_group_src(q, mpctl, ipst)) == NULL) {
9762 		return (1);
9763 	}
9764 
9765 	if ((mpctl = ip_snmp_get_mib2_virt_multi(q, mpctl, ipst)) == NULL) {
9766 		return (1);
9767 	}
9768 
9769 	if ((mpctl = ip_snmp_get_mib2_multi_rtable(q, mpctl, ipst)) == NULL) {
9770 		return (1);
9771 	}
9772 
9773 	mpctl = ip_snmp_get_mib2_ip_route_media(q, mpctl, level, ipst);
9774 	if (mpctl == NULL)
9775 		return (1);
9776 
9777 	mpctl = ip_snmp_get_mib2_ip6_route_media(q, mpctl, level, ipst);
9778 	if (mpctl == NULL)
9779 		return (1);
9780 
9781 	if ((mpctl = sctp_snmp_get_mib2(q, mpctl, sctps)) == NULL) {
9782 		return (1);
9783 	}
9784 	if ((mpctl = ip_snmp_get_mib2_ip_dce(q, mpctl, ipst)) == NULL) {
9785 		return (1);
9786 	}
9787 done:
9788 	freemsg(mpctl);
9789 	return (1);
9790 }
9791 
9792 /* Get global (legacy) IPv4 statistics */
9793 static mblk_t *
ip_snmp_get_mib2_ip(queue_t * q,mblk_t * mpctl,mib2_ipIfStatsEntry_t * ipmib,ip_stack_t * ipst,boolean_t legacy_req)9794 ip_snmp_get_mib2_ip(queue_t *q, mblk_t *mpctl, mib2_ipIfStatsEntry_t *ipmib,
9795     ip_stack_t *ipst, boolean_t legacy_req)
9796 {
9797 	mib2_ip_t		old_ip_mib;
9798 	struct opthdr		*optp;
9799 	mblk_t			*mp2ctl;
9800 	mib2_ipAddrEntry_t	mae;
9801 
9802 	/*
9803 	 * make a copy of the original message
9804 	 */
9805 	mp2ctl = copymsg(mpctl);
9806 
9807 	/* fixed length IP structure... */
9808 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9809 	optp->level = MIB2_IP;
9810 	optp->name = 0;
9811 	SET_MIB(old_ip_mib.ipForwarding,
9812 	    (WE_ARE_FORWARDING(ipst) ? 1 : 2));
9813 	SET_MIB(old_ip_mib.ipDefaultTTL,
9814 	    (uint32_t)ipst->ips_ip_def_ttl);
9815 	SET_MIB(old_ip_mib.ipReasmTimeout,
9816 	    ipst->ips_ip_reassembly_timeout);
9817 	SET_MIB(old_ip_mib.ipAddrEntrySize,
9818 	    (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9819 	    sizeof (mib2_ipAddrEntry_t));
9820 	SET_MIB(old_ip_mib.ipRouteEntrySize,
9821 	    sizeof (mib2_ipRouteEntry_t));
9822 	SET_MIB(old_ip_mib.ipNetToMediaEntrySize,
9823 	    sizeof (mib2_ipNetToMediaEntry_t));
9824 	SET_MIB(old_ip_mib.ipMemberEntrySize, sizeof (ip_member_t));
9825 	SET_MIB(old_ip_mib.ipGroupSourceEntrySize, sizeof (ip_grpsrc_t));
9826 	SET_MIB(old_ip_mib.ipRouteAttributeSize,
9827 	    sizeof (mib2_ipAttributeEntry_t));
9828 	SET_MIB(old_ip_mib.transportMLPSize, sizeof (mib2_transportMLPEntry_t));
9829 	SET_MIB(old_ip_mib.ipDestEntrySize, sizeof (dest_cache_entry_t));
9830 
9831 	/*
9832 	 * Grab the statistics from the new IP MIB
9833 	 */
9834 	SET_MIB(old_ip_mib.ipInReceives,
9835 	    (uint32_t)ipmib->ipIfStatsHCInReceives);
9836 	SET_MIB(old_ip_mib.ipInHdrErrors, ipmib->ipIfStatsInHdrErrors);
9837 	SET_MIB(old_ip_mib.ipInAddrErrors, ipmib->ipIfStatsInAddrErrors);
9838 	SET_MIB(old_ip_mib.ipForwDatagrams,
9839 	    (uint32_t)ipmib->ipIfStatsHCOutForwDatagrams);
9840 	SET_MIB(old_ip_mib.ipInUnknownProtos,
9841 	    ipmib->ipIfStatsInUnknownProtos);
9842 	SET_MIB(old_ip_mib.ipInDiscards, ipmib->ipIfStatsInDiscards);
9843 	SET_MIB(old_ip_mib.ipInDelivers,
9844 	    (uint32_t)ipmib->ipIfStatsHCInDelivers);
9845 	SET_MIB(old_ip_mib.ipOutRequests,
9846 	    (uint32_t)ipmib->ipIfStatsHCOutRequests);
9847 	SET_MIB(old_ip_mib.ipOutDiscards, ipmib->ipIfStatsOutDiscards);
9848 	SET_MIB(old_ip_mib.ipOutNoRoutes, ipmib->ipIfStatsOutNoRoutes);
9849 	SET_MIB(old_ip_mib.ipReasmReqds, ipmib->ipIfStatsReasmReqds);
9850 	SET_MIB(old_ip_mib.ipReasmOKs, ipmib->ipIfStatsReasmOKs);
9851 	SET_MIB(old_ip_mib.ipReasmFails, ipmib->ipIfStatsReasmFails);
9852 	SET_MIB(old_ip_mib.ipFragOKs, ipmib->ipIfStatsOutFragOKs);
9853 	SET_MIB(old_ip_mib.ipFragFails, ipmib->ipIfStatsOutFragFails);
9854 	SET_MIB(old_ip_mib.ipFragCreates, ipmib->ipIfStatsOutFragCreates);
9855 
9856 	/* ipRoutingDiscards is not being used */
9857 	SET_MIB(old_ip_mib.ipRoutingDiscards, 0);
9858 	SET_MIB(old_ip_mib.tcpInErrs, ipmib->tcpIfStatsInErrs);
9859 	SET_MIB(old_ip_mib.udpNoPorts, ipmib->udpIfStatsNoPorts);
9860 	SET_MIB(old_ip_mib.ipInCksumErrs, ipmib->ipIfStatsInCksumErrs);
9861 	SET_MIB(old_ip_mib.ipReasmDuplicates,
9862 	    ipmib->ipIfStatsReasmDuplicates);
9863 	SET_MIB(old_ip_mib.ipReasmPartDups, ipmib->ipIfStatsReasmPartDups);
9864 	SET_MIB(old_ip_mib.ipForwProhibits, ipmib->ipIfStatsForwProhibits);
9865 	SET_MIB(old_ip_mib.udpInCksumErrs, ipmib->udpIfStatsInCksumErrs);
9866 	SET_MIB(old_ip_mib.udpInOverflows, ipmib->udpIfStatsInOverflows);
9867 	SET_MIB(old_ip_mib.rawipInOverflows,
9868 	    ipmib->rawipIfStatsInOverflows);
9869 
9870 	SET_MIB(old_ip_mib.ipsecInSucceeded, ipmib->ipsecIfStatsInSucceeded);
9871 	SET_MIB(old_ip_mib.ipsecInFailed, ipmib->ipsecIfStatsInFailed);
9872 	SET_MIB(old_ip_mib.ipInIPv6, ipmib->ipIfStatsInWrongIPVersion);
9873 	SET_MIB(old_ip_mib.ipOutIPv6, ipmib->ipIfStatsOutWrongIPVersion);
9874 	SET_MIB(old_ip_mib.ipOutSwitchIPv6,
9875 	    ipmib->ipIfStatsOutSwitchIPVersion);
9876 
9877 	if (!snmp_append_data(mpctl->b_cont, (char *)&old_ip_mib,
9878 	    (int)sizeof (old_ip_mib))) {
9879 		ip1dbg(("ip_snmp_get_mib2_ip: failed to allocate %u bytes\n",
9880 		    (uint_t)sizeof (old_ip_mib)));
9881 	}
9882 
9883 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9884 	ip3dbg(("ip_snmp_get_mib2_ip: level %d, name %d, len %d\n",
9885 	    (int)optp->level, (int)optp->name, (int)optp->len));
9886 	qreply(q, mpctl);
9887 	return (mp2ctl);
9888 }
9889 
9890 /* Per interface IPv4 statistics */
9891 static mblk_t *
ip_snmp_get_mib2_ip_traffic_stats(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)9892 ip_snmp_get_mib2_ip_traffic_stats(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9893     boolean_t legacy_req)
9894 {
9895 	struct opthdr		*optp;
9896 	mblk_t			*mp2ctl;
9897 	ill_t			*ill;
9898 	ill_walk_context_t	ctx;
9899 	mblk_t			*mp_tail = NULL;
9900 	mib2_ipIfStatsEntry_t	global_ip_mib;
9901 	mib2_ipAddrEntry_t	mae;
9902 
9903 	/*
9904 	 * Make a copy of the original message
9905 	 */
9906 	mp2ctl = copymsg(mpctl);
9907 
9908 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9909 	optp->level = MIB2_IP;
9910 	optp->name = MIB2_IP_TRAFFIC_STATS;
9911 	/* Include "unknown interface" ip_mib */
9912 	ipst->ips_ip_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv4;
9913 	ipst->ips_ip_mib.ipIfStatsIfIndex =
9914 	    MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
9915 	SET_MIB(ipst->ips_ip_mib.ipIfStatsForwarding,
9916 	    (ipst->ips_ip_forwarding ? 1 : 2));
9917 	SET_MIB(ipst->ips_ip_mib.ipIfStatsDefaultTTL,
9918 	    (uint32_t)ipst->ips_ip_def_ttl);
9919 	SET_MIB(ipst->ips_ip_mib.ipIfStatsEntrySize,
9920 	    sizeof (mib2_ipIfStatsEntry_t));
9921 	SET_MIB(ipst->ips_ip_mib.ipIfStatsAddrEntrySize,
9922 	    sizeof (mib2_ipAddrEntry_t));
9923 	SET_MIB(ipst->ips_ip_mib.ipIfStatsRouteEntrySize,
9924 	    sizeof (mib2_ipRouteEntry_t));
9925 	SET_MIB(ipst->ips_ip_mib.ipIfStatsNetToMediaEntrySize,
9926 	    sizeof (mib2_ipNetToMediaEntry_t));
9927 	SET_MIB(ipst->ips_ip_mib.ipIfStatsMemberEntrySize,
9928 	    sizeof (ip_member_t));
9929 	SET_MIB(ipst->ips_ip_mib.ipIfStatsGroupSourceEntrySize,
9930 	    sizeof (ip_grpsrc_t));
9931 
9932 	bcopy(&ipst->ips_ip_mib, &global_ip_mib, sizeof (global_ip_mib));
9933 
9934 	if (legacy_req) {
9935 		SET_MIB(global_ip_mib.ipIfStatsAddrEntrySize,
9936 		    LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t));
9937 	}
9938 
9939 	if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9940 	    (char *)&global_ip_mib, (int)sizeof (global_ip_mib))) {
9941 		ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9942 		    "failed to allocate %u bytes\n",
9943 		    (uint_t)sizeof (global_ip_mib)));
9944 	}
9945 
9946 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9947 	ill = ILL_START_WALK_V4(&ctx, ipst);
9948 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9949 		ill->ill_ip_mib->ipIfStatsIfIndex =
9950 		    ill->ill_phyint->phyint_ifindex;
9951 		SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
9952 		    (ipst->ips_ip_forwarding ? 1 : 2));
9953 		SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultTTL,
9954 		    (uint32_t)ipst->ips_ip_def_ttl);
9955 
9956 		ip_mib2_add_ip_stats(&global_ip_mib, ill->ill_ip_mib);
9957 		if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9958 		    (char *)ill->ill_ip_mib,
9959 		    (int)sizeof (*ill->ill_ip_mib))) {
9960 			ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9961 			    "failed to allocate %u bytes\n",
9962 			    (uint_t)sizeof (*ill->ill_ip_mib)));
9963 		}
9964 	}
9965 	rw_exit(&ipst->ips_ill_g_lock);
9966 
9967 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9968 	ip3dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9969 	    "level %d, name %d, len %d\n",
9970 	    (int)optp->level, (int)optp->name, (int)optp->len));
9971 	qreply(q, mpctl);
9972 
9973 	if (mp2ctl == NULL)
9974 		return (NULL);
9975 
9976 	return (ip_snmp_get_mib2_ip(q, mp2ctl, &global_ip_mib, ipst,
9977 	    legacy_req));
9978 }
9979 
9980 /* Global IPv4 ICMP statistics */
9981 static mblk_t *
ip_snmp_get_mib2_icmp(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)9982 ip_snmp_get_mib2_icmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9983 {
9984 	struct opthdr		*optp;
9985 	mblk_t			*mp2ctl;
9986 
9987 	/*
9988 	 * Make a copy of the original message
9989 	 */
9990 	mp2ctl = copymsg(mpctl);
9991 
9992 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9993 	optp->level = MIB2_ICMP;
9994 	optp->name = 0;
9995 	if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_icmp_mib,
9996 	    (int)sizeof (ipst->ips_icmp_mib))) {
9997 		ip1dbg(("ip_snmp_get_mib2_icmp: failed to allocate %u bytes\n",
9998 		    (uint_t)sizeof (ipst->ips_icmp_mib)));
9999 	}
10000 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10001 	ip3dbg(("ip_snmp_get_mib2_icmp: level %d, name %d, len %d\n",
10002 	    (int)optp->level, (int)optp->name, (int)optp->len));
10003 	qreply(q, mpctl);
10004 	return (mp2ctl);
10005 }
10006 
10007 /* Global IPv4 IGMP statistics */
10008 static mblk_t *
ip_snmp_get_mib2_igmp(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10009 ip_snmp_get_mib2_igmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10010 {
10011 	struct opthdr		*optp;
10012 	mblk_t			*mp2ctl;
10013 
10014 	/*
10015 	 * make a copy of the original message
10016 	 */
10017 	mp2ctl = copymsg(mpctl);
10018 
10019 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10020 	optp->level = EXPER_IGMP;
10021 	optp->name = 0;
10022 	if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_igmpstat,
10023 	    (int)sizeof (ipst->ips_igmpstat))) {
10024 		ip1dbg(("ip_snmp_get_mib2_igmp: failed to allocate %u bytes\n",
10025 		    (uint_t)sizeof (ipst->ips_igmpstat)));
10026 	}
10027 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10028 	ip3dbg(("ip_snmp_get_mib2_igmp: level %d, name %d, len %d\n",
10029 	    (int)optp->level, (int)optp->name, (int)optp->len));
10030 	qreply(q, mpctl);
10031 	return (mp2ctl);
10032 }
10033 
10034 /* Global IPv4 Multicast Routing statistics */
10035 static mblk_t *
ip_snmp_get_mib2_multi(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10036 ip_snmp_get_mib2_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10037 {
10038 	struct opthdr		*optp;
10039 	mblk_t			*mp2ctl;
10040 
10041 	/*
10042 	 * make a copy of the original message
10043 	 */
10044 	mp2ctl = copymsg(mpctl);
10045 
10046 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10047 	optp->level = EXPER_DVMRP;
10048 	optp->name = 0;
10049 	if (!ip_mroute_stats(mpctl->b_cont, ipst)) {
10050 		ip0dbg(("ip_mroute_stats: failed\n"));
10051 	}
10052 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10053 	ip3dbg(("ip_snmp_get_mib2_multi: level %d, name %d, len %d\n",
10054 	    (int)optp->level, (int)optp->name, (int)optp->len));
10055 	qreply(q, mpctl);
10056 	return (mp2ctl);
10057 }
10058 
10059 /* IPv4 address information */
10060 static mblk_t *
ip_snmp_get_mib2_ip_addr(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)10061 ip_snmp_get_mib2_ip_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10062     boolean_t legacy_req)
10063 {
10064 	struct opthdr		*optp;
10065 	mblk_t			*mp2ctl;
10066 	mblk_t			*mp_tail = NULL;
10067 	ill_t			*ill;
10068 	ipif_t			*ipif;
10069 	uint_t			bitval;
10070 	mib2_ipAddrEntry_t	mae;
10071 	size_t			mae_size;
10072 	zoneid_t		zoneid;
10073 	ill_walk_context_t	ctx;
10074 
10075 	/*
10076 	 * make a copy of the original message
10077 	 */
10078 	mp2ctl = copymsg(mpctl);
10079 
10080 	mae_size = (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
10081 	    sizeof (mib2_ipAddrEntry_t);
10082 
10083 	/* ipAddrEntryTable */
10084 
10085 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10086 	optp->level = MIB2_IP;
10087 	optp->name = MIB2_IP_ADDR;
10088 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10089 
10090 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10091 	ill = ILL_START_WALK_V4(&ctx, ipst);
10092 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10093 		for (ipif = ill->ill_ipif; ipif != NULL;
10094 		    ipif = ipif->ipif_next) {
10095 			if (ipif->ipif_zoneid != zoneid &&
10096 			    ipif->ipif_zoneid != ALL_ZONES)
10097 				continue;
10098 			/* Sum of count from dead IRE_LO* and our current */
10099 			mae.ipAdEntInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10100 			if (ipif->ipif_ire_local != NULL) {
10101 				mae.ipAdEntInfo.ae_ibcnt +=
10102 				    ipif->ipif_ire_local->ire_ib_pkt_count;
10103 			}
10104 			mae.ipAdEntInfo.ae_obcnt = 0;
10105 			mae.ipAdEntInfo.ae_focnt = 0;
10106 
10107 			ipif_get_name(ipif, mae.ipAdEntIfIndex.o_bytes,
10108 			    OCTET_LENGTH);
10109 			mae.ipAdEntIfIndex.o_length =
10110 			    mi_strlen(mae.ipAdEntIfIndex.o_bytes);
10111 			mae.ipAdEntAddr = ipif->ipif_lcl_addr;
10112 			mae.ipAdEntNetMask = ipif->ipif_net_mask;
10113 			mae.ipAdEntInfo.ae_subnet = ipif->ipif_subnet;
10114 			mae.ipAdEntInfo.ae_subnet_len =
10115 			    ip_mask_to_plen(ipif->ipif_net_mask);
10116 			mae.ipAdEntInfo.ae_src_addr = ipif->ipif_lcl_addr;
10117 			for (bitval = 1;
10118 			    bitval &&
10119 			    !(bitval & ipif->ipif_brd_addr);
10120 			    bitval <<= 1)
10121 				noop;
10122 			mae.ipAdEntBcastAddr = bitval;
10123 			mae.ipAdEntReasmMaxSize = IP_MAXPACKET;
10124 			mae.ipAdEntInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10125 			mae.ipAdEntInfo.ae_metric  = ipif->ipif_ill->ill_metric;
10126 			mae.ipAdEntInfo.ae_broadcast_addr =
10127 			    ipif->ipif_brd_addr;
10128 			mae.ipAdEntInfo.ae_pp_dst_addr =
10129 			    ipif->ipif_pp_dst_addr;
10130 			mae.ipAdEntInfo.ae_flags = ipif->ipif_flags |
10131 			    ill->ill_flags | ill->ill_phyint->phyint_flags;
10132 			mae.ipAdEntRetransmitTime =
10133 			    ill->ill_reachable_retrans_time;
10134 
10135 			if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10136 			    (char *)&mae, (int)mae_size)) {
10137 				ip1dbg(("ip_snmp_get_mib2_ip_addr: failed to "
10138 				    "allocate %u bytes\n", (uint_t)mae_size));
10139 			}
10140 		}
10141 	}
10142 	rw_exit(&ipst->ips_ill_g_lock);
10143 
10144 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10145 	ip3dbg(("ip_snmp_get_mib2_ip_addr: level %d, name %d, len %d\n",
10146 	    (int)optp->level, (int)optp->name, (int)optp->len));
10147 	qreply(q, mpctl);
10148 	return (mp2ctl);
10149 }
10150 
10151 /* IPv6 address information */
10152 static mblk_t *
ip_snmp_get_mib2_ip6_addr(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)10153 ip_snmp_get_mib2_ip6_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10154     boolean_t legacy_req)
10155 {
10156 	struct opthdr		*optp;
10157 	mblk_t			*mp2ctl;
10158 	mblk_t			*mp_tail = NULL;
10159 	ill_t			*ill;
10160 	ipif_t			*ipif;
10161 	mib2_ipv6AddrEntry_t	mae6;
10162 	size_t			mae6_size;
10163 	zoneid_t		zoneid;
10164 	ill_walk_context_t	ctx;
10165 
10166 	/*
10167 	 * make a copy of the original message
10168 	 */
10169 	mp2ctl = copymsg(mpctl);
10170 
10171 	mae6_size = (legacy_req) ?
10172 	    LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t) :
10173 	    sizeof (mib2_ipv6AddrEntry_t);
10174 
10175 	/* ipv6AddrEntryTable */
10176 
10177 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10178 	optp->level = MIB2_IP6;
10179 	optp->name = MIB2_IP6_ADDR;
10180 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10181 
10182 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10183 	ill = ILL_START_WALK_V6(&ctx, ipst);
10184 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10185 		for (ipif = ill->ill_ipif; ipif != NULL;
10186 		    ipif = ipif->ipif_next) {
10187 			if (ipif->ipif_zoneid != zoneid &&
10188 			    ipif->ipif_zoneid != ALL_ZONES)
10189 				continue;
10190 			/* Sum of count from dead IRE_LO* and our current */
10191 			mae6.ipv6AddrInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10192 			if (ipif->ipif_ire_local != NULL) {
10193 				mae6.ipv6AddrInfo.ae_ibcnt +=
10194 				    ipif->ipif_ire_local->ire_ib_pkt_count;
10195 			}
10196 			mae6.ipv6AddrInfo.ae_obcnt = 0;
10197 			mae6.ipv6AddrInfo.ae_focnt = 0;
10198 
10199 			ipif_get_name(ipif, mae6.ipv6AddrIfIndex.o_bytes,
10200 			    OCTET_LENGTH);
10201 			mae6.ipv6AddrIfIndex.o_length =
10202 			    mi_strlen(mae6.ipv6AddrIfIndex.o_bytes);
10203 			mae6.ipv6AddrAddress = ipif->ipif_v6lcl_addr;
10204 			mae6.ipv6AddrPfxLength =
10205 			    ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
10206 			mae6.ipv6AddrInfo.ae_subnet = ipif->ipif_v6subnet;
10207 			mae6.ipv6AddrInfo.ae_subnet_len =
10208 			    mae6.ipv6AddrPfxLength;
10209 			mae6.ipv6AddrInfo.ae_src_addr = ipif->ipif_v6lcl_addr;
10210 
10211 			/* Type: stateless(1), stateful(2), unknown(3) */
10212 			if (ipif->ipif_flags & IPIF_ADDRCONF)
10213 				mae6.ipv6AddrType = 1;
10214 			else
10215 				mae6.ipv6AddrType = 2;
10216 			/* Anycast: true(1), false(2) */
10217 			if (ipif->ipif_flags & IPIF_ANYCAST)
10218 				mae6.ipv6AddrAnycastFlag = 1;
10219 			else
10220 				mae6.ipv6AddrAnycastFlag = 2;
10221 
10222 			/*
10223 			 * Address status: preferred(1), deprecated(2),
10224 			 * invalid(3), inaccessible(4), unknown(5)
10225 			 */
10226 			if (ipif->ipif_flags & IPIF_NOLOCAL)
10227 				mae6.ipv6AddrStatus = 3;
10228 			else if (ipif->ipif_flags & IPIF_DEPRECATED)
10229 				mae6.ipv6AddrStatus = 2;
10230 			else
10231 				mae6.ipv6AddrStatus = 1;
10232 			mae6.ipv6AddrInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10233 			mae6.ipv6AddrInfo.ae_metric  =
10234 			    ipif->ipif_ill->ill_metric;
10235 			mae6.ipv6AddrInfo.ae_pp_dst_addr =
10236 			    ipif->ipif_v6pp_dst_addr;
10237 			mae6.ipv6AddrInfo.ae_flags = ipif->ipif_flags |
10238 			    ill->ill_flags | ill->ill_phyint->phyint_flags;
10239 			mae6.ipv6AddrReasmMaxSize = IP_MAXPACKET;
10240 			mae6.ipv6AddrIdentifier = ill->ill_token;
10241 			mae6.ipv6AddrIdentifierLen = ill->ill_token_length;
10242 			mae6.ipv6AddrReachableTime = ill->ill_reachable_time;
10243 			mae6.ipv6AddrRetransmitTime =
10244 			    ill->ill_reachable_retrans_time;
10245 			if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10246 			    (char *)&mae6, (int)mae6_size)) {
10247 				ip1dbg(("ip_snmp_get_mib2_ip6_addr: failed to "
10248 				    "allocate %u bytes\n",
10249 				    (uint_t)mae6_size));
10250 			}
10251 		}
10252 	}
10253 	rw_exit(&ipst->ips_ill_g_lock);
10254 
10255 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10256 	ip3dbg(("ip_snmp_get_mib2_ip6_addr: level %d, name %d, len %d\n",
10257 	    (int)optp->level, (int)optp->name, (int)optp->len));
10258 	qreply(q, mpctl);
10259 	return (mp2ctl);
10260 }
10261 
10262 /* IPv4 multicast group membership. */
10263 static mblk_t *
ip_snmp_get_mib2_ip_group_mem(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10264 ip_snmp_get_mib2_ip_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10265 {
10266 	struct opthdr		*optp;
10267 	mblk_t			*mp2ctl;
10268 	ill_t			*ill;
10269 	ipif_t			*ipif;
10270 	ilm_t			*ilm;
10271 	ip_member_t		ipm;
10272 	mblk_t			*mp_tail = NULL;
10273 	ill_walk_context_t	ctx;
10274 	zoneid_t		zoneid;
10275 
10276 	/*
10277 	 * make a copy of the original message
10278 	 */
10279 	mp2ctl = copymsg(mpctl);
10280 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10281 
10282 	/* ipGroupMember table */
10283 	optp = (struct opthdr *)&mpctl->b_rptr[
10284 	    sizeof (struct T_optmgmt_ack)];
10285 	optp->level = MIB2_IP;
10286 	optp->name = EXPER_IP_GROUP_MEMBERSHIP;
10287 
10288 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10289 	ill = ILL_START_WALK_V4(&ctx, ipst);
10290 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10291 		/* Make sure the ill isn't going away. */
10292 		if (!ill_check_and_refhold(ill))
10293 			continue;
10294 		rw_exit(&ipst->ips_ill_g_lock);
10295 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10296 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10297 			if (ilm->ilm_zoneid != zoneid &&
10298 			    ilm->ilm_zoneid != ALL_ZONES)
10299 				continue;
10300 
10301 			/* Is there an ipif for ilm_ifaddr? */
10302 			for (ipif = ill->ill_ipif; ipif != NULL;
10303 			    ipif = ipif->ipif_next) {
10304 				if (!IPIF_IS_CONDEMNED(ipif) &&
10305 				    ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10306 				    ilm->ilm_ifaddr != INADDR_ANY)
10307 					break;
10308 			}
10309 			if (ipif != NULL) {
10310 				ipif_get_name(ipif,
10311 				    ipm.ipGroupMemberIfIndex.o_bytes,
10312 				    OCTET_LENGTH);
10313 			} else {
10314 				ill_get_name(ill,
10315 				    ipm.ipGroupMemberIfIndex.o_bytes,
10316 				    OCTET_LENGTH);
10317 			}
10318 			ipm.ipGroupMemberIfIndex.o_length =
10319 			    mi_strlen(ipm.ipGroupMemberIfIndex.o_bytes);
10320 
10321 			ipm.ipGroupMemberAddress = ilm->ilm_addr;
10322 			ipm.ipGroupMemberRefCnt = ilm->ilm_refcnt;
10323 			ipm.ipGroupMemberFilterMode = ilm->ilm_fmode;
10324 			if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10325 			    (char *)&ipm, (int)sizeof (ipm))) {
10326 				ip1dbg(("ip_snmp_get_mib2_ip_group: "
10327 				    "failed to allocate %u bytes\n",
10328 				    (uint_t)sizeof (ipm)));
10329 			}
10330 		}
10331 		rw_exit(&ill->ill_mcast_lock);
10332 		ill_refrele(ill);
10333 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10334 	}
10335 	rw_exit(&ipst->ips_ill_g_lock);
10336 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10337 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10338 	    (int)optp->level, (int)optp->name, (int)optp->len));
10339 	qreply(q, mpctl);
10340 	return (mp2ctl);
10341 }
10342 
10343 /* IPv6 multicast group membership. */
10344 static mblk_t *
ip_snmp_get_mib2_ip6_group_mem(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10345 ip_snmp_get_mib2_ip6_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10346 {
10347 	struct opthdr		*optp;
10348 	mblk_t			*mp2ctl;
10349 	ill_t			*ill;
10350 	ilm_t			*ilm;
10351 	ipv6_member_t		ipm6;
10352 	mblk_t			*mp_tail = NULL;
10353 	ill_walk_context_t	ctx;
10354 	zoneid_t		zoneid;
10355 
10356 	/*
10357 	 * make a copy of the original message
10358 	 */
10359 	mp2ctl = copymsg(mpctl);
10360 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10361 
10362 	/* ip6GroupMember table */
10363 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10364 	optp->level = MIB2_IP6;
10365 	optp->name = EXPER_IP6_GROUP_MEMBERSHIP;
10366 
10367 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10368 	ill = ILL_START_WALK_V6(&ctx, ipst);
10369 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10370 		/* Make sure the ill isn't going away. */
10371 		if (!ill_check_and_refhold(ill))
10372 			continue;
10373 		rw_exit(&ipst->ips_ill_g_lock);
10374 		/*
10375 		 * Normally we don't have any members on under IPMP interfaces.
10376 		 * We report them as a debugging aid.
10377 		 */
10378 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10379 		ipm6.ipv6GroupMemberIfIndex = ill->ill_phyint->phyint_ifindex;
10380 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10381 			if (ilm->ilm_zoneid != zoneid &&
10382 			    ilm->ilm_zoneid != ALL_ZONES)
10383 				continue;	/* not this zone */
10384 			ipm6.ipv6GroupMemberAddress = ilm->ilm_v6addr;
10385 			ipm6.ipv6GroupMemberRefCnt = ilm->ilm_refcnt;
10386 			ipm6.ipv6GroupMemberFilterMode = ilm->ilm_fmode;
10387 			if (!snmp_append_data2(mpctl->b_cont,
10388 			    &mp_tail,
10389 			    (char *)&ipm6, (int)sizeof (ipm6))) {
10390 				ip1dbg(("ip_snmp_get_mib2_ip6_group: "
10391 				    "failed to allocate %u bytes\n",
10392 				    (uint_t)sizeof (ipm6)));
10393 			}
10394 		}
10395 		rw_exit(&ill->ill_mcast_lock);
10396 		ill_refrele(ill);
10397 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10398 	}
10399 	rw_exit(&ipst->ips_ill_g_lock);
10400 
10401 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10402 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10403 	    (int)optp->level, (int)optp->name, (int)optp->len));
10404 	qreply(q, mpctl);
10405 	return (mp2ctl);
10406 }
10407 
10408 /* IP multicast filtered sources */
10409 static mblk_t *
ip_snmp_get_mib2_ip_group_src(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10410 ip_snmp_get_mib2_ip_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10411 {
10412 	struct opthdr		*optp;
10413 	mblk_t			*mp2ctl;
10414 	ill_t			*ill;
10415 	ipif_t			*ipif;
10416 	ilm_t			*ilm;
10417 	ip_grpsrc_t		ips;
10418 	mblk_t			*mp_tail = NULL;
10419 	ill_walk_context_t	ctx;
10420 	zoneid_t		zoneid;
10421 	int			i;
10422 	slist_t			*sl;
10423 
10424 	/*
10425 	 * make a copy of the original message
10426 	 */
10427 	mp2ctl = copymsg(mpctl);
10428 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10429 
10430 	/* ipGroupSource table */
10431 	optp = (struct opthdr *)&mpctl->b_rptr[
10432 	    sizeof (struct T_optmgmt_ack)];
10433 	optp->level = MIB2_IP;
10434 	optp->name = EXPER_IP_GROUP_SOURCES;
10435 
10436 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10437 	ill = ILL_START_WALK_V4(&ctx, ipst);
10438 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10439 		/* Make sure the ill isn't going away. */
10440 		if (!ill_check_and_refhold(ill))
10441 			continue;
10442 		rw_exit(&ipst->ips_ill_g_lock);
10443 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10444 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10445 			sl = ilm->ilm_filter;
10446 			if (ilm->ilm_zoneid != zoneid &&
10447 			    ilm->ilm_zoneid != ALL_ZONES)
10448 				continue;
10449 			if (SLIST_IS_EMPTY(sl))
10450 				continue;
10451 
10452 			/* Is there an ipif for ilm_ifaddr? */
10453 			for (ipif = ill->ill_ipif; ipif != NULL;
10454 			    ipif = ipif->ipif_next) {
10455 				if (!IPIF_IS_CONDEMNED(ipif) &&
10456 				    ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10457 				    ilm->ilm_ifaddr != INADDR_ANY)
10458 					break;
10459 			}
10460 			if (ipif != NULL) {
10461 				ipif_get_name(ipif,
10462 				    ips.ipGroupSourceIfIndex.o_bytes,
10463 				    OCTET_LENGTH);
10464 			} else {
10465 				ill_get_name(ill,
10466 				    ips.ipGroupSourceIfIndex.o_bytes,
10467 				    OCTET_LENGTH);
10468 			}
10469 			ips.ipGroupSourceIfIndex.o_length =
10470 			    mi_strlen(ips.ipGroupSourceIfIndex.o_bytes);
10471 
10472 			ips.ipGroupSourceGroup = ilm->ilm_addr;
10473 			for (i = 0; i < sl->sl_numsrc; i++) {
10474 				if (!IN6_IS_ADDR_V4MAPPED(&sl->sl_addr[i]))
10475 					continue;
10476 				IN6_V4MAPPED_TO_IPADDR(&sl->sl_addr[i],
10477 				    ips.ipGroupSourceAddress);
10478 				if (snmp_append_data2(mpctl->b_cont, &mp_tail,
10479 				    (char *)&ips, (int)sizeof (ips)) == 0) {
10480 					ip1dbg(("ip_snmp_get_mib2_ip_group_src:"
10481 					    " failed to allocate %u bytes\n",
10482 					    (uint_t)sizeof (ips)));
10483 				}
10484 			}
10485 		}
10486 		rw_exit(&ill->ill_mcast_lock);
10487 		ill_refrele(ill);
10488 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10489 	}
10490 	rw_exit(&ipst->ips_ill_g_lock);
10491 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10492 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10493 	    (int)optp->level, (int)optp->name, (int)optp->len));
10494 	qreply(q, mpctl);
10495 	return (mp2ctl);
10496 }
10497 
10498 /* IPv6 multicast filtered sources. */
10499 static mblk_t *
ip_snmp_get_mib2_ip6_group_src(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10500 ip_snmp_get_mib2_ip6_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10501 {
10502 	struct opthdr		*optp;
10503 	mblk_t			*mp2ctl;
10504 	ill_t			*ill;
10505 	ilm_t			*ilm;
10506 	ipv6_grpsrc_t		ips6;
10507 	mblk_t			*mp_tail = NULL;
10508 	ill_walk_context_t	ctx;
10509 	zoneid_t		zoneid;
10510 	int			i;
10511 	slist_t			*sl;
10512 
10513 	/*
10514 	 * make a copy of the original message
10515 	 */
10516 	mp2ctl = copymsg(mpctl);
10517 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10518 
10519 	/* ip6GroupMember table */
10520 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10521 	optp->level = MIB2_IP6;
10522 	optp->name = EXPER_IP6_GROUP_SOURCES;
10523 
10524 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10525 	ill = ILL_START_WALK_V6(&ctx, ipst);
10526 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10527 		/* Make sure the ill isn't going away. */
10528 		if (!ill_check_and_refhold(ill))
10529 			continue;
10530 		rw_exit(&ipst->ips_ill_g_lock);
10531 		/*
10532 		 * Normally we don't have any members on under IPMP interfaces.
10533 		 * We report them as a debugging aid.
10534 		 */
10535 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10536 		ips6.ipv6GroupSourceIfIndex = ill->ill_phyint->phyint_ifindex;
10537 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10538 			sl = ilm->ilm_filter;
10539 			if (ilm->ilm_zoneid != zoneid &&
10540 			    ilm->ilm_zoneid != ALL_ZONES)
10541 				continue;
10542 			if (SLIST_IS_EMPTY(sl))
10543 				continue;
10544 			ips6.ipv6GroupSourceGroup = ilm->ilm_v6addr;
10545 			for (i = 0; i < sl->sl_numsrc; i++) {
10546 				ips6.ipv6GroupSourceAddress = sl->sl_addr[i];
10547 				if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10548 				    (char *)&ips6, (int)sizeof (ips6))) {
10549 					ip1dbg(("ip_snmp_get_mib2_ip6_"
10550 					    "group_src: failed to allocate "
10551 					    "%u bytes\n",
10552 					    (uint_t)sizeof (ips6)));
10553 				}
10554 			}
10555 		}
10556 		rw_exit(&ill->ill_mcast_lock);
10557 		ill_refrele(ill);
10558 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10559 	}
10560 	rw_exit(&ipst->ips_ill_g_lock);
10561 
10562 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10563 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10564 	    (int)optp->level, (int)optp->name, (int)optp->len));
10565 	qreply(q, mpctl);
10566 	return (mp2ctl);
10567 }
10568 
10569 /* Multicast routing virtual interface table. */
10570 static mblk_t *
ip_snmp_get_mib2_virt_multi(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10571 ip_snmp_get_mib2_virt_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10572 {
10573 	struct opthdr		*optp;
10574 	mblk_t			*mp2ctl;
10575 
10576 	/*
10577 	 * make a copy of the original message
10578 	 */
10579 	mp2ctl = copymsg(mpctl);
10580 
10581 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10582 	optp->level = EXPER_DVMRP;
10583 	optp->name = EXPER_DVMRP_VIF;
10584 	if (!ip_mroute_vif(mpctl->b_cont, ipst)) {
10585 		ip0dbg(("ip_mroute_vif: failed\n"));
10586 	}
10587 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10588 	ip3dbg(("ip_snmp_get_mib2_virt_multi: level %d, name %d, len %d\n",
10589 	    (int)optp->level, (int)optp->name, (int)optp->len));
10590 	qreply(q, mpctl);
10591 	return (mp2ctl);
10592 }
10593 
10594 /* Multicast routing table. */
10595 static mblk_t *
ip_snmp_get_mib2_multi_rtable(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10596 ip_snmp_get_mib2_multi_rtable(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10597 {
10598 	struct opthdr		*optp;
10599 	mblk_t			*mp2ctl;
10600 
10601 	/*
10602 	 * make a copy of the original message
10603 	 */
10604 	mp2ctl = copymsg(mpctl);
10605 
10606 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10607 	optp->level = EXPER_DVMRP;
10608 	optp->name = EXPER_DVMRP_MRT;
10609 	if (!ip_mroute_mrt(mpctl->b_cont, ipst)) {
10610 		ip0dbg(("ip_mroute_mrt: failed\n"));
10611 	}
10612 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10613 	ip3dbg(("ip_snmp_get_mib2_multi_rtable: level %d, name %d, len %d\n",
10614 	    (int)optp->level, (int)optp->name, (int)optp->len));
10615 	qreply(q, mpctl);
10616 	return (mp2ctl);
10617 }
10618 
10619 /*
10620  * Return ipRouteEntryTable, ipNetToMediaEntryTable, and ipRouteAttributeTable
10621  * in one IRE walk.
10622  */
10623 static mblk_t *
ip_snmp_get_mib2_ip_route_media(queue_t * q,mblk_t * mpctl,int level,ip_stack_t * ipst)10624 ip_snmp_get_mib2_ip_route_media(queue_t *q, mblk_t *mpctl, int level,
10625     ip_stack_t *ipst)
10626 {
10627 	struct opthdr	*optp;
10628 	mblk_t		*mp2ctl;	/* Returned */
10629 	mblk_t		*mp3ctl;	/* nettomedia */
10630 	mblk_t		*mp4ctl;	/* routeattrs */
10631 	iproutedata_t	ird;
10632 	zoneid_t	zoneid;
10633 
10634 	/*
10635 	 * make copies of the original message
10636 	 *	- mp2ctl is returned unchanged to the caller for its use
10637 	 *	- mpctl is sent upstream as ipRouteEntryTable
10638 	 *	- mp3ctl is sent upstream as ipNetToMediaEntryTable
10639 	 *	- mp4ctl is sent upstream as ipRouteAttributeTable
10640 	 */
10641 	mp2ctl = copymsg(mpctl);
10642 	mp3ctl = copymsg(mpctl);
10643 	mp4ctl = copymsg(mpctl);
10644 	if (mp3ctl == NULL || mp4ctl == NULL) {
10645 		freemsg(mp4ctl);
10646 		freemsg(mp3ctl);
10647 		freemsg(mp2ctl);
10648 		freemsg(mpctl);
10649 		return (NULL);
10650 	}
10651 
10652 	bzero(&ird, sizeof (ird));
10653 
10654 	ird.ird_route.lp_head = mpctl->b_cont;
10655 	ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10656 	ird.ird_attrs.lp_head = mp4ctl->b_cont;
10657 	/*
10658 	 * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10659 	 * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10660 	 * intended a temporary solution until a proper MIB API is provided
10661 	 * that provides complete filtering/caller-opt-in.
10662 	 */
10663 	if (level == EXPER_IP_AND_ALL_IRES)
10664 		ird.ird_flags |= IRD_REPORT_ALL;
10665 
10666 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10667 	ire_walk_v4(ip_snmp_get2_v4, &ird, zoneid, ipst);
10668 
10669 	/* ipRouteEntryTable in mpctl */
10670 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10671 	optp->level = MIB2_IP;
10672 	optp->name = MIB2_IP_ROUTE;
10673 	optp->len = msgdsize(ird.ird_route.lp_head);
10674 	ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10675 	    (int)optp->level, (int)optp->name, (int)optp->len));
10676 	qreply(q, mpctl);
10677 
10678 	/* ipNetToMediaEntryTable in mp3ctl */
10679 	ncec_walk(NULL, ip_snmp_get2_v4_media, &ird, ipst);
10680 
10681 	optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10682 	optp->level = MIB2_IP;
10683 	optp->name = MIB2_IP_MEDIA;
10684 	optp->len = msgdsize(ird.ird_netmedia.lp_head);
10685 	ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10686 	    (int)optp->level, (int)optp->name, (int)optp->len));
10687 	qreply(q, mp3ctl);
10688 
10689 	/* ipRouteAttributeTable in mp4ctl */
10690 	optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10691 	optp->level = MIB2_IP;
10692 	optp->name = EXPER_IP_RTATTR;
10693 	optp->len = msgdsize(ird.ird_attrs.lp_head);
10694 	ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10695 	    (int)optp->level, (int)optp->name, (int)optp->len));
10696 	if (optp->len == 0)
10697 		freemsg(mp4ctl);
10698 	else
10699 		qreply(q, mp4ctl);
10700 
10701 	return (mp2ctl);
10702 }
10703 
10704 /*
10705  * Return ipv6RouteEntryTable and ipv6RouteAttributeTable in one IRE walk, and
10706  * ipv6NetToMediaEntryTable in an NDP walk.
10707  */
10708 static mblk_t *
ip_snmp_get_mib2_ip6_route_media(queue_t * q,mblk_t * mpctl,int level,ip_stack_t * ipst)10709 ip_snmp_get_mib2_ip6_route_media(queue_t *q, mblk_t *mpctl, int level,
10710     ip_stack_t *ipst)
10711 {
10712 	struct opthdr	*optp;
10713 	mblk_t		*mp2ctl;	/* Returned */
10714 	mblk_t		*mp3ctl;	/* nettomedia */
10715 	mblk_t		*mp4ctl;	/* routeattrs */
10716 	iproutedata_t	ird;
10717 	zoneid_t	zoneid;
10718 
10719 	/*
10720 	 * make copies of the original message
10721 	 *	- mp2ctl is returned unchanged to the caller for its use
10722 	 *	- mpctl is sent upstream as ipv6RouteEntryTable
10723 	 *	- mp3ctl is sent upstream as ipv6NetToMediaEntryTable
10724 	 *	- mp4ctl is sent upstream as ipv6RouteAttributeTable
10725 	 */
10726 	mp2ctl = copymsg(mpctl);
10727 	mp3ctl = copymsg(mpctl);
10728 	mp4ctl = copymsg(mpctl);
10729 	if (mp3ctl == NULL || mp4ctl == NULL) {
10730 		freemsg(mp4ctl);
10731 		freemsg(mp3ctl);
10732 		freemsg(mp2ctl);
10733 		freemsg(mpctl);
10734 		return (NULL);
10735 	}
10736 
10737 	bzero(&ird, sizeof (ird));
10738 
10739 	ird.ird_route.lp_head = mpctl->b_cont;
10740 	ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10741 	ird.ird_attrs.lp_head = mp4ctl->b_cont;
10742 	/*
10743 	 * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10744 	 * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10745 	 * intended a temporary solution until a proper MIB API is provided
10746 	 * that provides complete filtering/caller-opt-in.
10747 	 */
10748 	if (level == EXPER_IP_AND_ALL_IRES)
10749 		ird.ird_flags |= IRD_REPORT_ALL;
10750 
10751 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10752 	ire_walk_v6(ip_snmp_get2_v6_route, &ird, zoneid, ipst);
10753 
10754 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10755 	optp->level = MIB2_IP6;
10756 	optp->name = MIB2_IP6_ROUTE;
10757 	optp->len = msgdsize(ird.ird_route.lp_head);
10758 	ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10759 	    (int)optp->level, (int)optp->name, (int)optp->len));
10760 	qreply(q, mpctl);
10761 
10762 	/* ipv6NetToMediaEntryTable in mp3ctl */
10763 	ncec_walk(NULL, ip_snmp_get2_v6_media, &ird, ipst);
10764 
10765 	optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10766 	optp->level = MIB2_IP6;
10767 	optp->name = MIB2_IP6_MEDIA;
10768 	optp->len = msgdsize(ird.ird_netmedia.lp_head);
10769 	ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10770 	    (int)optp->level, (int)optp->name, (int)optp->len));
10771 	qreply(q, mp3ctl);
10772 
10773 	/* ipv6RouteAttributeTable in mp4ctl */
10774 	optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10775 	optp->level = MIB2_IP6;
10776 	optp->name = EXPER_IP_RTATTR;
10777 	optp->len = msgdsize(ird.ird_attrs.lp_head);
10778 	ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10779 	    (int)optp->level, (int)optp->name, (int)optp->len));
10780 	if (optp->len == 0)
10781 		freemsg(mp4ctl);
10782 	else
10783 		qreply(q, mp4ctl);
10784 
10785 	return (mp2ctl);
10786 }
10787 
10788 /*
10789  * IPv6 mib: One per ill
10790  */
10791 static mblk_t *
ip_snmp_get_mib2_ip6(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)10792 ip_snmp_get_mib2_ip6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10793     boolean_t legacy_req)
10794 {
10795 	struct opthdr		*optp;
10796 	mblk_t			*mp2ctl;
10797 	ill_t			*ill;
10798 	ill_walk_context_t	ctx;
10799 	mblk_t			*mp_tail = NULL;
10800 	mib2_ipv6AddrEntry_t	mae6;
10801 	mib2_ipIfStatsEntry_t	*ise;
10802 	size_t			ise_size, iae_size;
10803 
10804 	/*
10805 	 * Make a copy of the original message
10806 	 */
10807 	mp2ctl = copymsg(mpctl);
10808 
10809 	/* fixed length IPv6 structure ... */
10810 
10811 	if (legacy_req) {
10812 		ise_size = LEGACY_MIB_SIZE(&ipst->ips_ip6_mib,
10813 		    mib2_ipIfStatsEntry_t);
10814 		iae_size = LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t);
10815 	} else {
10816 		ise_size = sizeof (mib2_ipIfStatsEntry_t);
10817 		iae_size = sizeof (mib2_ipv6AddrEntry_t);
10818 	}
10819 
10820 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10821 	optp->level = MIB2_IP6;
10822 	optp->name = 0;
10823 	/* Include "unknown interface" ip6_mib */
10824 	ipst->ips_ip6_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv6;
10825 	ipst->ips_ip6_mib.ipIfStatsIfIndex =
10826 	    MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
10827 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsForwarding,
10828 	    ipst->ips_ipv6_forwarding ? 1 : 2);
10829 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsDefaultHopLimit,
10830 	    ipst->ips_ipv6_def_hops);
10831 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsEntrySize,
10832 	    sizeof (mib2_ipIfStatsEntry_t));
10833 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsAddrEntrySize,
10834 	    sizeof (mib2_ipv6AddrEntry_t));
10835 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsRouteEntrySize,
10836 	    sizeof (mib2_ipv6RouteEntry_t));
10837 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsNetToMediaEntrySize,
10838 	    sizeof (mib2_ipv6NetToMediaEntry_t));
10839 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsMemberEntrySize,
10840 	    sizeof (ipv6_member_t));
10841 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsGroupSourceEntrySize,
10842 	    sizeof (ipv6_grpsrc_t));
10843 
10844 	/*
10845 	 * Synchronize 64- and 32-bit counters
10846 	 */
10847 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInReceives,
10848 	    ipIfStatsHCInReceives);
10849 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInDelivers,
10850 	    ipIfStatsHCInDelivers);
10851 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutRequests,
10852 	    ipIfStatsHCOutRequests);
10853 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutForwDatagrams,
10854 	    ipIfStatsHCOutForwDatagrams);
10855 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutMcastPkts,
10856 	    ipIfStatsHCOutMcastPkts);
10857 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInMcastPkts,
10858 	    ipIfStatsHCInMcastPkts);
10859 
10860 	if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10861 	    (char *)&ipst->ips_ip6_mib, (int)ise_size)) {
10862 		ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate %u bytes\n",
10863 		    (uint_t)ise_size));
10864 	} else if (legacy_req) {
10865 		/* Adjust the EntrySize fields for legacy requests. */
10866 		ise =
10867 		    (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr - (int)ise_size);
10868 		SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10869 		SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10870 	}
10871 
10872 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10873 	ill = ILL_START_WALK_V6(&ctx, ipst);
10874 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10875 		ill->ill_ip_mib->ipIfStatsIfIndex =
10876 		    ill->ill_phyint->phyint_ifindex;
10877 		SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
10878 		    ipst->ips_ipv6_forwarding ? 1 : 2);
10879 		SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultHopLimit,
10880 		    ill->ill_max_hops);
10881 
10882 		/*
10883 		 * Synchronize 64- and 32-bit counters
10884 		 */
10885 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInReceives,
10886 		    ipIfStatsHCInReceives);
10887 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInDelivers,
10888 		    ipIfStatsHCInDelivers);
10889 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutRequests,
10890 		    ipIfStatsHCOutRequests);
10891 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutForwDatagrams,
10892 		    ipIfStatsHCOutForwDatagrams);
10893 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutMcastPkts,
10894 		    ipIfStatsHCOutMcastPkts);
10895 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInMcastPkts,
10896 		    ipIfStatsHCInMcastPkts);
10897 
10898 		if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10899 		    (char *)ill->ill_ip_mib, (int)ise_size)) {
10900 			ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate "
10901 			"%u bytes\n", (uint_t)ise_size));
10902 		} else if (legacy_req) {
10903 			/* Adjust the EntrySize fields for legacy requests. */
10904 			ise = (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr -
10905 			    (int)ise_size);
10906 			SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10907 			SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10908 		}
10909 	}
10910 	rw_exit(&ipst->ips_ill_g_lock);
10911 
10912 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10913 	ip3dbg(("ip_snmp_get_mib2_ip6: level %d, name %d, len %d\n",
10914 	    (int)optp->level, (int)optp->name, (int)optp->len));
10915 	qreply(q, mpctl);
10916 	return (mp2ctl);
10917 }
10918 
10919 /*
10920  * ICMPv6 mib: One per ill
10921  */
10922 static mblk_t *
ip_snmp_get_mib2_icmp6(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10923 ip_snmp_get_mib2_icmp6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10924 {
10925 	struct opthdr		*optp;
10926 	mblk_t			*mp2ctl;
10927 	ill_t			*ill;
10928 	ill_walk_context_t	ctx;
10929 	mblk_t			*mp_tail = NULL;
10930 	/*
10931 	 * Make a copy of the original message
10932 	 */
10933 	mp2ctl = copymsg(mpctl);
10934 
10935 	/* fixed length ICMPv6 structure ... */
10936 
10937 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10938 	optp->level = MIB2_ICMP6;
10939 	optp->name = 0;
10940 	/* Include "unknown interface" icmp6_mib */
10941 	ipst->ips_icmp6_mib.ipv6IfIcmpIfIndex =
10942 	    MIB2_UNKNOWN_INTERFACE; /* netstat flag */
10943 	ipst->ips_icmp6_mib.ipv6IfIcmpEntrySize =
10944 	    sizeof (mib2_ipv6IfIcmpEntry_t);
10945 	if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10946 	    (char *)&ipst->ips_icmp6_mib,
10947 	    (int)sizeof (ipst->ips_icmp6_mib))) {
10948 		ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate %u bytes\n",
10949 		    (uint_t)sizeof (ipst->ips_icmp6_mib)));
10950 	}
10951 
10952 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10953 	ill = ILL_START_WALK_V6(&ctx, ipst);
10954 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10955 		ill->ill_icmp6_mib->ipv6IfIcmpIfIndex =
10956 		    ill->ill_phyint->phyint_ifindex;
10957 		if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10958 		    (char *)ill->ill_icmp6_mib,
10959 		    (int)sizeof (*ill->ill_icmp6_mib))) {
10960 			ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate "
10961 			    "%u bytes\n",
10962 			    (uint_t)sizeof (*ill->ill_icmp6_mib)));
10963 		}
10964 	}
10965 	rw_exit(&ipst->ips_ill_g_lock);
10966 
10967 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10968 	ip3dbg(("ip_snmp_get_mib2_icmp6: level %d, name %d, len %d\n",
10969 	    (int)optp->level, (int)optp->name, (int)optp->len));
10970 	qreply(q, mpctl);
10971 	return (mp2ctl);
10972 }
10973 
10974 /*
10975  * ire_walk routine to create both ipRouteEntryTable and
10976  * ipRouteAttributeTable in one IRE walk
10977  */
10978 static void
ip_snmp_get2_v4(ire_t * ire,iproutedata_t * ird)10979 ip_snmp_get2_v4(ire_t *ire, iproutedata_t *ird)
10980 {
10981 	ill_t				*ill;
10982 	mib2_ipRouteEntry_t		*re;
10983 	mib2_ipAttributeEntry_t		iaes;
10984 	tsol_ire_gw_secattr_t		*attrp;
10985 	tsol_gc_t			*gc = NULL;
10986 	tsol_gcgrp_t			*gcgrp = NULL;
10987 	ip_stack_t			*ipst = ire->ire_ipst;
10988 
10989 	ASSERT(ire->ire_ipversion == IPV4_VERSION);
10990 
10991 	if (!(ird->ird_flags & IRD_REPORT_ALL)) {
10992 		if (ire->ire_testhidden)
10993 			return;
10994 		if (ire->ire_type & IRE_IF_CLONE)
10995 			return;
10996 	}
10997 
10998 	if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
10999 		return;
11000 
11001 	if ((attrp = ire->ire_gw_secattr) != NULL) {
11002 		mutex_enter(&attrp->igsa_lock);
11003 		if ((gc = attrp->igsa_gc) != NULL) {
11004 			gcgrp = gc->gc_grp;
11005 			ASSERT(gcgrp != NULL);
11006 			rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
11007 		}
11008 		mutex_exit(&attrp->igsa_lock);
11009 	}
11010 	/*
11011 	 * Return all IRE types for route table... let caller pick and choose
11012 	 */
11013 	re->ipRouteDest = ire->ire_addr;
11014 	ill = ire->ire_ill;
11015 	re->ipRouteIfIndex.o_length = 0;
11016 	if (ill != NULL) {
11017 		ill_get_name(ill, re->ipRouteIfIndex.o_bytes, OCTET_LENGTH);
11018 		re->ipRouteIfIndex.o_length =
11019 		    mi_strlen(re->ipRouteIfIndex.o_bytes);
11020 	}
11021 	re->ipRouteMetric1 = -1;
11022 	re->ipRouteMetric2 = -1;
11023 	re->ipRouteMetric3 = -1;
11024 	re->ipRouteMetric4 = -1;
11025 
11026 	re->ipRouteNextHop = ire->ire_gateway_addr;
11027 	/* indirect(4), direct(3), or invalid(2) */
11028 	if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11029 		re->ipRouteType = 2;
11030 	else if (ire->ire_type & IRE_ONLINK)
11031 		re->ipRouteType = 3;
11032 	else
11033 		re->ipRouteType = 4;
11034 
11035 	re->ipRouteProto = -1;
11036 	re->ipRouteAge = gethrestime_sec() - ire->ire_create_time;
11037 	re->ipRouteMask = ire->ire_mask;
11038 	re->ipRouteMetric5 = -1;
11039 	re->ipRouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11040 	if (ire->ire_ill != NULL && re->ipRouteInfo.re_max_frag == 0)
11041 		re->ipRouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11042 
11043 	re->ipRouteInfo.re_frag_flag	= 0;
11044 	re->ipRouteInfo.re_rtt		= 0;
11045 	re->ipRouteInfo.re_src_addr	= 0;
11046 	re->ipRouteInfo.re_ref		= ire->ire_refcnt;
11047 	re->ipRouteInfo.re_obpkt	= ire->ire_ob_pkt_count;
11048 	re->ipRouteInfo.re_ibpkt	= ire->ire_ib_pkt_count;
11049 	re->ipRouteInfo.re_flags	= ire->ire_flags;
11050 
11051 	/* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11052 	if (ire->ire_type & IRE_INTERFACE) {
11053 		ire_t *child;
11054 
11055 		rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11056 		child = ire->ire_dep_children;
11057 		while (child != NULL) {
11058 			re->ipRouteInfo.re_obpkt += child->ire_ob_pkt_count;
11059 			re->ipRouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11060 			child = child->ire_dep_sib_next;
11061 		}
11062 		rw_exit(&ipst->ips_ire_dep_lock);
11063 	}
11064 
11065 	if (ire->ire_flags & RTF_DYNAMIC) {
11066 		re->ipRouteInfo.re_ire_type	= IRE_HOST_REDIRECT;
11067 	} else {
11068 		re->ipRouteInfo.re_ire_type	= ire->ire_type;
11069 	}
11070 
11071 	if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11072 	    (char *)re, (int)sizeof (*re))) {
11073 		ip1dbg(("ip_snmp_get2_v4: failed to allocate %u bytes\n",
11074 		    (uint_t)sizeof (*re)));
11075 	}
11076 
11077 	if (gc != NULL) {
11078 		iaes.iae_routeidx = ird->ird_idx;
11079 		iaes.iae_doi = gc->gc_db->gcdb_doi;
11080 		iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11081 
11082 		if (!snmp_append_data2(ird->ird_attrs.lp_head,
11083 		    &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11084 			ip1dbg(("ip_snmp_get2_v4: failed to allocate %u "
11085 			    "bytes\n", (uint_t)sizeof (iaes)));
11086 		}
11087 	}
11088 
11089 	/* bump route index for next pass */
11090 	ird->ird_idx++;
11091 
11092 	kmem_free(re, sizeof (*re));
11093 	if (gcgrp != NULL)
11094 		rw_exit(&gcgrp->gcgrp_rwlock);
11095 }
11096 
11097 /*
11098  * ire_walk routine to create ipv6RouteEntryTable and ipRouteEntryTable.
11099  */
11100 static void
ip_snmp_get2_v6_route(ire_t * ire,iproutedata_t * ird)11101 ip_snmp_get2_v6_route(ire_t *ire, iproutedata_t *ird)
11102 {
11103 	ill_t				*ill;
11104 	mib2_ipv6RouteEntry_t		*re;
11105 	mib2_ipAttributeEntry_t		iaes;
11106 	tsol_ire_gw_secattr_t		*attrp;
11107 	tsol_gc_t			*gc = NULL;
11108 	tsol_gcgrp_t			*gcgrp = NULL;
11109 	ip_stack_t			*ipst = ire->ire_ipst;
11110 
11111 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
11112 
11113 	if (!(ird->ird_flags & IRD_REPORT_ALL)) {
11114 		if (ire->ire_testhidden)
11115 			return;
11116 		if (ire->ire_type & IRE_IF_CLONE)
11117 			return;
11118 	}
11119 
11120 	if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
11121 		return;
11122 
11123 	if ((attrp = ire->ire_gw_secattr) != NULL) {
11124 		mutex_enter(&attrp->igsa_lock);
11125 		if ((gc = attrp->igsa_gc) != NULL) {
11126 			gcgrp = gc->gc_grp;
11127 			ASSERT(gcgrp != NULL);
11128 			rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
11129 		}
11130 		mutex_exit(&attrp->igsa_lock);
11131 	}
11132 	/*
11133 	 * Return all IRE types for route table... let caller pick and choose
11134 	 */
11135 	re->ipv6RouteDest = ire->ire_addr_v6;
11136 	re->ipv6RoutePfxLength = ip_mask_to_plen_v6(&ire->ire_mask_v6);
11137 	re->ipv6RouteIndex = 0;	/* Unique when multiple with same dest/plen */
11138 	re->ipv6RouteIfIndex.o_length = 0;
11139 	ill = ire->ire_ill;
11140 	if (ill != NULL) {
11141 		ill_get_name(ill, re->ipv6RouteIfIndex.o_bytes, OCTET_LENGTH);
11142 		re->ipv6RouteIfIndex.o_length =
11143 		    mi_strlen(re->ipv6RouteIfIndex.o_bytes);
11144 	}
11145 
11146 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
11147 
11148 	mutex_enter(&ire->ire_lock);
11149 	re->ipv6RouteNextHop = ire->ire_gateway_addr_v6;
11150 	mutex_exit(&ire->ire_lock);
11151 
11152 	/* remote(4), local(3), or discard(2) */
11153 	if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11154 		re->ipv6RouteType = 2;
11155 	else if (ire->ire_type & IRE_ONLINK)
11156 		re->ipv6RouteType = 3;
11157 	else
11158 		re->ipv6RouteType = 4;
11159 
11160 	re->ipv6RouteProtocol	= -1;
11161 	re->ipv6RoutePolicy	= 0;
11162 	re->ipv6RouteAge	= gethrestime_sec() - ire->ire_create_time;
11163 	re->ipv6RouteNextHopRDI	= 0;
11164 	re->ipv6RouteWeight	= 0;
11165 	re->ipv6RouteMetric	= 0;
11166 	re->ipv6RouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11167 	if (ire->ire_ill != NULL && re->ipv6RouteInfo.re_max_frag == 0)
11168 		re->ipv6RouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11169 
11170 	re->ipv6RouteInfo.re_frag_flag	= 0;
11171 	re->ipv6RouteInfo.re_rtt	= 0;
11172 	re->ipv6RouteInfo.re_src_addr	= ipv6_all_zeros;
11173 	re->ipv6RouteInfo.re_obpkt	= ire->ire_ob_pkt_count;
11174 	re->ipv6RouteInfo.re_ibpkt	= ire->ire_ib_pkt_count;
11175 	re->ipv6RouteInfo.re_ref	= ire->ire_refcnt;
11176 	re->ipv6RouteInfo.re_flags	= ire->ire_flags;
11177 
11178 	/* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11179 	if (ire->ire_type & IRE_INTERFACE) {
11180 		ire_t *child;
11181 
11182 		rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11183 		child = ire->ire_dep_children;
11184 		while (child != NULL) {
11185 			re->ipv6RouteInfo.re_obpkt += child->ire_ob_pkt_count;
11186 			re->ipv6RouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11187 			child = child->ire_dep_sib_next;
11188 		}
11189 		rw_exit(&ipst->ips_ire_dep_lock);
11190 	}
11191 	if (ire->ire_flags & RTF_DYNAMIC) {
11192 		re->ipv6RouteInfo.re_ire_type	= IRE_HOST_REDIRECT;
11193 	} else {
11194 		re->ipv6RouteInfo.re_ire_type	= ire->ire_type;
11195 	}
11196 
11197 	if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11198 	    (char *)re, (int)sizeof (*re))) {
11199 		ip1dbg(("ip_snmp_get2_v6: failed to allocate %u bytes\n",
11200 		    (uint_t)sizeof (*re)));
11201 	}
11202 
11203 	if (gc != NULL) {
11204 		iaes.iae_routeidx = ird->ird_idx;
11205 		iaes.iae_doi = gc->gc_db->gcdb_doi;
11206 		iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11207 
11208 		if (!snmp_append_data2(ird->ird_attrs.lp_head,
11209 		    &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11210 			ip1dbg(("ip_snmp_get2_v6: failed to allocate %u "
11211 			    "bytes\n", (uint_t)sizeof (iaes)));
11212 		}
11213 	}
11214 
11215 	/* bump route index for next pass */
11216 	ird->ird_idx++;
11217 
11218 	kmem_free(re, sizeof (*re));
11219 	if (gcgrp != NULL)
11220 		rw_exit(&gcgrp->gcgrp_rwlock);
11221 }
11222 
11223 /*
11224  * ncec_walk routine to create ipv6NetToMediaEntryTable
11225  */
11226 static void
ip_snmp_get2_v6_media(ncec_t * ncec,void * ptr)11227 ip_snmp_get2_v6_media(ncec_t *ncec, void *ptr)
11228 {
11229 	iproutedata_t *ird		= ptr;
11230 	ill_t				*ill;
11231 	mib2_ipv6NetToMediaEntry_t	ntme;
11232 
11233 	ill = ncec->ncec_ill;
11234 	/* skip arpce entries, and loopback ncec entries */
11235 	if (ill->ill_isv6 == B_FALSE || ill->ill_net_type == IRE_LOOPBACK)
11236 		return;
11237 	/*
11238 	 * Neighbor cache entry attached to IRE with on-link
11239 	 * destination.
11240 	 * We report all IPMP groups on ncec_ill which is normally the upper.
11241 	 */
11242 	ntme.ipv6NetToMediaIfIndex = ill->ill_phyint->phyint_ifindex;
11243 	ntme.ipv6NetToMediaNetAddress = ncec->ncec_addr;
11244 	ntme.ipv6NetToMediaPhysAddress.o_length = ill->ill_phys_addr_length;
11245 	if (ncec->ncec_lladdr != NULL) {
11246 		bcopy(ncec->ncec_lladdr, ntme.ipv6NetToMediaPhysAddress.o_bytes,
11247 		    ntme.ipv6NetToMediaPhysAddress.o_length);
11248 	}
11249 	/*
11250 	 * Note: Returns ND_* states. Should be:
11251 	 * reachable(1), stale(2), delay(3), probe(4),
11252 	 * invalid(5), unknown(6)
11253 	 */
11254 	ntme.ipv6NetToMediaState = ncec->ncec_state;
11255 	ntme.ipv6NetToMediaLastUpdated = 0;
11256 
11257 	/* other(1), dynamic(2), static(3), local(4) */
11258 	if (NCE_MYADDR(ncec)) {
11259 		ntme.ipv6NetToMediaType = 4;
11260 	} else if (ncec->ncec_flags & NCE_F_PUBLISH) {
11261 		ntme.ipv6NetToMediaType = 1; /* proxy */
11262 	} else if (ncec->ncec_flags & NCE_F_STATIC) {
11263 		ntme.ipv6NetToMediaType = 3;
11264 	} else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST)) {
11265 		ntme.ipv6NetToMediaType = 1;
11266 	} else {
11267 		ntme.ipv6NetToMediaType = 2;
11268 	}
11269 
11270 	if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11271 	    &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11272 		ip1dbg(("ip_snmp_get2_v6_media: failed to allocate %u bytes\n",
11273 		    (uint_t)sizeof (ntme)));
11274 	}
11275 }
11276 
11277 int
nce2ace(ncec_t * ncec)11278 nce2ace(ncec_t *ncec)
11279 {
11280 	int flags = 0;
11281 
11282 	if (NCE_ISREACHABLE(ncec))
11283 		flags |= ACE_F_RESOLVED;
11284 	if (ncec->ncec_flags & NCE_F_AUTHORITY)
11285 		flags |= ACE_F_AUTHORITY;
11286 	if (ncec->ncec_flags & NCE_F_PUBLISH)
11287 		flags |= ACE_F_PUBLISH;
11288 	if ((ncec->ncec_flags & NCE_F_NONUD) != 0)
11289 		flags |= ACE_F_PERMANENT;
11290 	if (NCE_MYADDR(ncec))
11291 		flags |= (ACE_F_MYADDR | ACE_F_AUTHORITY);
11292 	if (ncec->ncec_flags & NCE_F_UNVERIFIED)
11293 		flags |= ACE_F_UNVERIFIED;
11294 	if (ncec->ncec_flags & NCE_F_AUTHORITY)
11295 		flags |= ACE_F_AUTHORITY;
11296 	if (ncec->ncec_flags & NCE_F_DELAYED)
11297 		flags |= ACE_F_DELAYED;
11298 	return (flags);
11299 }
11300 
11301 /*
11302  * ncec_walk routine to create ipNetToMediaEntryTable
11303  */
11304 static void
ip_snmp_get2_v4_media(ncec_t * ncec,void * ptr)11305 ip_snmp_get2_v4_media(ncec_t *ncec, void *ptr)
11306 {
11307 	iproutedata_t *ird		= ptr;
11308 	ill_t				*ill;
11309 	mib2_ipNetToMediaEntry_t	ntme;
11310 	const char			*name = "unknown";
11311 	ipaddr_t			ncec_addr;
11312 
11313 	ill = ncec->ncec_ill;
11314 	if (ill->ill_isv6 || (ncec->ncec_flags & NCE_F_BCAST) ||
11315 	    ill->ill_net_type == IRE_LOOPBACK)
11316 		return;
11317 
11318 	/* We report all IPMP groups on ncec_ill which is normally the upper. */
11319 	name = ill->ill_name;
11320 	/* Based on RFC 4293: other(1), inval(2), dyn(3), stat(4) */
11321 	if (NCE_MYADDR(ncec)) {
11322 		ntme.ipNetToMediaType = 4;
11323 	} else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST|NCE_F_PUBLISH)) {
11324 		ntme.ipNetToMediaType = 1;
11325 	} else {
11326 		ntme.ipNetToMediaType = 3;
11327 	}
11328 	ntme.ipNetToMediaIfIndex.o_length = MIN(OCTET_LENGTH, strlen(name));
11329 	bcopy(name, ntme.ipNetToMediaIfIndex.o_bytes,
11330 	    ntme.ipNetToMediaIfIndex.o_length);
11331 
11332 	IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ncec_addr);
11333 	bcopy(&ncec_addr, &ntme.ipNetToMediaNetAddress, sizeof (ncec_addr));
11334 
11335 	ntme.ipNetToMediaInfo.ntm_mask.o_length = sizeof (ipaddr_t);
11336 	ncec_addr = INADDR_BROADCAST;
11337 	bcopy(&ncec_addr, ntme.ipNetToMediaInfo.ntm_mask.o_bytes,
11338 	    sizeof (ncec_addr));
11339 	/*
11340 	 * map all the flags to the ACE counterpart.
11341 	 */
11342 	ntme.ipNetToMediaInfo.ntm_flags = nce2ace(ncec);
11343 
11344 	ntme.ipNetToMediaPhysAddress.o_length =
11345 	    MIN(OCTET_LENGTH, ill->ill_phys_addr_length);
11346 
11347 	if (!NCE_ISREACHABLE(ncec))
11348 		ntme.ipNetToMediaPhysAddress.o_length = 0;
11349 	else {
11350 		if (ncec->ncec_lladdr != NULL) {
11351 			bcopy(ncec->ncec_lladdr,
11352 			    ntme.ipNetToMediaPhysAddress.o_bytes,
11353 			    ntme.ipNetToMediaPhysAddress.o_length);
11354 		}
11355 	}
11356 
11357 	if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11358 	    &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11359 		ip1dbg(("ip_snmp_get2_v4_media: failed to allocate %u bytes\n",
11360 		    (uint_t)sizeof (ntme)));
11361 	}
11362 }
11363 
11364 /*
11365  * return (0) if invalid set request, 1 otherwise, including non-tcp requests
11366  */
11367 /* ARGSUSED */
11368 int
ip_snmp_set(queue_t * q,int level,int name,uchar_t * ptr,int len)11369 ip_snmp_set(queue_t *q, int level, int name, uchar_t *ptr, int len)
11370 {
11371 	switch (level) {
11372 	case MIB2_IP:
11373 	case MIB2_ICMP:
11374 		switch (name) {
11375 		default:
11376 			break;
11377 		}
11378 		return (1);
11379 	default:
11380 		return (1);
11381 	}
11382 }
11383 
11384 /*
11385  * When there exists both a 64- and 32-bit counter of a particular type
11386  * (i.e., InReceives), only the 64-bit counters are added.
11387  */
11388 void
ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t * o1,mib2_ipIfStatsEntry_t * o2)11389 ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t *o1, mib2_ipIfStatsEntry_t *o2)
11390 {
11391 	UPDATE_MIB(o1, ipIfStatsInHdrErrors, o2->ipIfStatsInHdrErrors);
11392 	UPDATE_MIB(o1, ipIfStatsInTooBigErrors, o2->ipIfStatsInTooBigErrors);
11393 	UPDATE_MIB(o1, ipIfStatsInNoRoutes, o2->ipIfStatsInNoRoutes);
11394 	UPDATE_MIB(o1, ipIfStatsInAddrErrors, o2->ipIfStatsInAddrErrors);
11395 	UPDATE_MIB(o1, ipIfStatsInUnknownProtos, o2->ipIfStatsInUnknownProtos);
11396 	UPDATE_MIB(o1, ipIfStatsInTruncatedPkts, o2->ipIfStatsInTruncatedPkts);
11397 	UPDATE_MIB(o1, ipIfStatsInDiscards, o2->ipIfStatsInDiscards);
11398 	UPDATE_MIB(o1, ipIfStatsOutDiscards, o2->ipIfStatsOutDiscards);
11399 	UPDATE_MIB(o1, ipIfStatsOutFragOKs, o2->ipIfStatsOutFragOKs);
11400 	UPDATE_MIB(o1, ipIfStatsOutFragFails, o2->ipIfStatsOutFragFails);
11401 	UPDATE_MIB(o1, ipIfStatsOutFragCreates, o2->ipIfStatsOutFragCreates);
11402 	UPDATE_MIB(o1, ipIfStatsReasmReqds, o2->ipIfStatsReasmReqds);
11403 	UPDATE_MIB(o1, ipIfStatsReasmOKs, o2->ipIfStatsReasmOKs);
11404 	UPDATE_MIB(o1, ipIfStatsReasmFails, o2->ipIfStatsReasmFails);
11405 	UPDATE_MIB(o1, ipIfStatsOutNoRoutes, o2->ipIfStatsOutNoRoutes);
11406 	UPDATE_MIB(o1, ipIfStatsReasmDuplicates, o2->ipIfStatsReasmDuplicates);
11407 	UPDATE_MIB(o1, ipIfStatsReasmPartDups, o2->ipIfStatsReasmPartDups);
11408 	UPDATE_MIB(o1, ipIfStatsForwProhibits, o2->ipIfStatsForwProhibits);
11409 	UPDATE_MIB(o1, udpInCksumErrs, o2->udpInCksumErrs);
11410 	UPDATE_MIB(o1, udpInOverflows, o2->udpInOverflows);
11411 	UPDATE_MIB(o1, rawipInOverflows, o2->rawipInOverflows);
11412 	UPDATE_MIB(o1, ipIfStatsInWrongIPVersion,
11413 	    o2->ipIfStatsInWrongIPVersion);
11414 	UPDATE_MIB(o1, ipIfStatsOutWrongIPVersion,
11415 	    o2->ipIfStatsInWrongIPVersion);
11416 	UPDATE_MIB(o1, ipIfStatsOutSwitchIPVersion,
11417 	    o2->ipIfStatsOutSwitchIPVersion);
11418 	UPDATE_MIB(o1, ipIfStatsHCInReceives, o2->ipIfStatsHCInReceives);
11419 	UPDATE_MIB(o1, ipIfStatsHCInOctets, o2->ipIfStatsHCInOctets);
11420 	UPDATE_MIB(o1, ipIfStatsHCInForwDatagrams,
11421 	    o2->ipIfStatsHCInForwDatagrams);
11422 	UPDATE_MIB(o1, ipIfStatsHCInDelivers, o2->ipIfStatsHCInDelivers);
11423 	UPDATE_MIB(o1, ipIfStatsHCOutRequests, o2->ipIfStatsHCOutRequests);
11424 	UPDATE_MIB(o1, ipIfStatsHCOutForwDatagrams,
11425 	    o2->ipIfStatsHCOutForwDatagrams);
11426 	UPDATE_MIB(o1, ipIfStatsOutFragReqds, o2->ipIfStatsOutFragReqds);
11427 	UPDATE_MIB(o1, ipIfStatsHCOutTransmits, o2->ipIfStatsHCOutTransmits);
11428 	UPDATE_MIB(o1, ipIfStatsHCOutOctets, o2->ipIfStatsHCOutOctets);
11429 	UPDATE_MIB(o1, ipIfStatsHCInMcastPkts, o2->ipIfStatsHCInMcastPkts);
11430 	UPDATE_MIB(o1, ipIfStatsHCInMcastOctets, o2->ipIfStatsHCInMcastOctets);
11431 	UPDATE_MIB(o1, ipIfStatsHCOutMcastPkts, o2->ipIfStatsHCOutMcastPkts);
11432 	UPDATE_MIB(o1, ipIfStatsHCOutMcastOctets,
11433 	    o2->ipIfStatsHCOutMcastOctets);
11434 	UPDATE_MIB(o1, ipIfStatsHCInBcastPkts, o2->ipIfStatsHCInBcastPkts);
11435 	UPDATE_MIB(o1, ipIfStatsHCOutBcastPkts, o2->ipIfStatsHCOutBcastPkts);
11436 	UPDATE_MIB(o1, ipsecInSucceeded, o2->ipsecInSucceeded);
11437 	UPDATE_MIB(o1, ipsecInFailed, o2->ipsecInFailed);
11438 	UPDATE_MIB(o1, ipInCksumErrs, o2->ipInCksumErrs);
11439 	UPDATE_MIB(o1, tcpInErrs, o2->tcpInErrs);
11440 	UPDATE_MIB(o1, udpNoPorts, o2->udpNoPorts);
11441 }
11442 
11443 void
ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t * o1,mib2_ipv6IfIcmpEntry_t * o2)11444 ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t *o1, mib2_ipv6IfIcmpEntry_t *o2)
11445 {
11446 	UPDATE_MIB(o1, ipv6IfIcmpInMsgs, o2->ipv6IfIcmpInMsgs);
11447 	UPDATE_MIB(o1, ipv6IfIcmpInErrors, o2->ipv6IfIcmpInErrors);
11448 	UPDATE_MIB(o1, ipv6IfIcmpInDestUnreachs, o2->ipv6IfIcmpInDestUnreachs);
11449 	UPDATE_MIB(o1, ipv6IfIcmpInAdminProhibs, o2->ipv6IfIcmpInAdminProhibs);
11450 	UPDATE_MIB(o1, ipv6IfIcmpInTimeExcds, o2->ipv6IfIcmpInTimeExcds);
11451 	UPDATE_MIB(o1, ipv6IfIcmpInParmProblems, o2->ipv6IfIcmpInParmProblems);
11452 	UPDATE_MIB(o1, ipv6IfIcmpInPktTooBigs, o2->ipv6IfIcmpInPktTooBigs);
11453 	UPDATE_MIB(o1, ipv6IfIcmpInEchos, o2->ipv6IfIcmpInEchos);
11454 	UPDATE_MIB(o1, ipv6IfIcmpInEchoReplies, o2->ipv6IfIcmpInEchoReplies);
11455 	UPDATE_MIB(o1, ipv6IfIcmpInRouterSolicits,
11456 	    o2->ipv6IfIcmpInRouterSolicits);
11457 	UPDATE_MIB(o1, ipv6IfIcmpInRouterAdvertisements,
11458 	    o2->ipv6IfIcmpInRouterAdvertisements);
11459 	UPDATE_MIB(o1, ipv6IfIcmpInNeighborSolicits,
11460 	    o2->ipv6IfIcmpInNeighborSolicits);
11461 	UPDATE_MIB(o1, ipv6IfIcmpInNeighborAdvertisements,
11462 	    o2->ipv6IfIcmpInNeighborAdvertisements);
11463 	UPDATE_MIB(o1, ipv6IfIcmpInRedirects, o2->ipv6IfIcmpInRedirects);
11464 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembQueries,
11465 	    o2->ipv6IfIcmpInGroupMembQueries);
11466 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembResponses,
11467 	    o2->ipv6IfIcmpInGroupMembResponses);
11468 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembReductions,
11469 	    o2->ipv6IfIcmpInGroupMembReductions);
11470 	UPDATE_MIB(o1, ipv6IfIcmpOutMsgs, o2->ipv6IfIcmpOutMsgs);
11471 	UPDATE_MIB(o1, ipv6IfIcmpOutErrors, o2->ipv6IfIcmpOutErrors);
11472 	UPDATE_MIB(o1, ipv6IfIcmpOutDestUnreachs,
11473 	    o2->ipv6IfIcmpOutDestUnreachs);
11474 	UPDATE_MIB(o1, ipv6IfIcmpOutAdminProhibs,
11475 	    o2->ipv6IfIcmpOutAdminProhibs);
11476 	UPDATE_MIB(o1, ipv6IfIcmpOutTimeExcds, o2->ipv6IfIcmpOutTimeExcds);
11477 	UPDATE_MIB(o1, ipv6IfIcmpOutParmProblems,
11478 	    o2->ipv6IfIcmpOutParmProblems);
11479 	UPDATE_MIB(o1, ipv6IfIcmpOutPktTooBigs, o2->ipv6IfIcmpOutPktTooBigs);
11480 	UPDATE_MIB(o1, ipv6IfIcmpOutEchos, o2->ipv6IfIcmpOutEchos);
11481 	UPDATE_MIB(o1, ipv6IfIcmpOutEchoReplies, o2->ipv6IfIcmpOutEchoReplies);
11482 	UPDATE_MIB(o1, ipv6IfIcmpOutRouterSolicits,
11483 	    o2->ipv6IfIcmpOutRouterSolicits);
11484 	UPDATE_MIB(o1, ipv6IfIcmpOutRouterAdvertisements,
11485 	    o2->ipv6IfIcmpOutRouterAdvertisements);
11486 	UPDATE_MIB(o1, ipv6IfIcmpOutNeighborSolicits,
11487 	    o2->ipv6IfIcmpOutNeighborSolicits);
11488 	UPDATE_MIB(o1, ipv6IfIcmpOutNeighborAdvertisements,
11489 	    o2->ipv6IfIcmpOutNeighborAdvertisements);
11490 	UPDATE_MIB(o1, ipv6IfIcmpOutRedirects, o2->ipv6IfIcmpOutRedirects);
11491 	UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembQueries,
11492 	    o2->ipv6IfIcmpOutGroupMembQueries);
11493 	UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembResponses,
11494 	    o2->ipv6IfIcmpOutGroupMembResponses);
11495 	UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembReductions,
11496 	    o2->ipv6IfIcmpOutGroupMembReductions);
11497 	UPDATE_MIB(o1, ipv6IfIcmpInOverflows, o2->ipv6IfIcmpInOverflows);
11498 	UPDATE_MIB(o1, ipv6IfIcmpBadHoplimit, o2->ipv6IfIcmpBadHoplimit);
11499 	UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborAdvertisements,
11500 	    o2->ipv6IfIcmpInBadNeighborAdvertisements);
11501 	UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborSolicitations,
11502 	    o2->ipv6IfIcmpInBadNeighborSolicitations);
11503 	UPDATE_MIB(o1, ipv6IfIcmpInBadRedirects, o2->ipv6IfIcmpInBadRedirects);
11504 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembTotal,
11505 	    o2->ipv6IfIcmpInGroupMembTotal);
11506 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadQueries,
11507 	    o2->ipv6IfIcmpInGroupMembBadQueries);
11508 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadReports,
11509 	    o2->ipv6IfIcmpInGroupMembBadReports);
11510 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembOurReports,
11511 	    o2->ipv6IfIcmpInGroupMembOurReports);
11512 }
11513 
11514 /*
11515  * Called before the options are updated to check if this packet will
11516  * be source routed from here.
11517  * This routine assumes that the options are well formed i.e. that they
11518  * have already been checked.
11519  */
11520 boolean_t
ip_source_routed(ipha_t * ipha,ip_stack_t * ipst)11521 ip_source_routed(ipha_t *ipha, ip_stack_t *ipst)
11522 {
11523 	ipoptp_t	opts;
11524 	uchar_t		*opt;
11525 	uint8_t		optval;
11526 	uint8_t		optlen;
11527 	ipaddr_t	dst;
11528 
11529 	if (IS_SIMPLE_IPH(ipha)) {
11530 		ip2dbg(("not source routed\n"));
11531 		return (B_FALSE);
11532 	}
11533 	dst = ipha->ipha_dst;
11534 	for (optval = ipoptp_first(&opts, ipha);
11535 	    optval != IPOPT_EOL;
11536 	    optval = ipoptp_next(&opts)) {
11537 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11538 		opt = opts.ipoptp_cur;
11539 		optlen = opts.ipoptp_len;
11540 		ip2dbg(("ip_source_routed: opt %d, len %d\n",
11541 		    optval, optlen));
11542 		switch (optval) {
11543 			uint32_t off;
11544 		case IPOPT_SSRR:
11545 		case IPOPT_LSRR:
11546 			/*
11547 			 * If dst is one of our addresses and there are some
11548 			 * entries left in the source route return (true).
11549 			 */
11550 			if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11551 				ip2dbg(("ip_source_routed: not next"
11552 				    " source route 0x%x\n",
11553 				    ntohl(dst)));
11554 				return (B_FALSE);
11555 			}
11556 			off = opt[IPOPT_OFFSET];
11557 			off--;
11558 			if (optlen < IP_ADDR_LEN ||
11559 			    off > optlen - IP_ADDR_LEN) {
11560 				/* End of source route */
11561 				ip1dbg(("ip_source_routed: end of SR\n"));
11562 				return (B_FALSE);
11563 			}
11564 			return (B_TRUE);
11565 		}
11566 	}
11567 	ip2dbg(("not source routed\n"));
11568 	return (B_FALSE);
11569 }
11570 
11571 /*
11572  * ip_unbind is called by the transports to remove a conn from
11573  * the fanout table.
11574  */
11575 void
ip_unbind(conn_t * connp)11576 ip_unbind(conn_t *connp)
11577 {
11578 
11579 	ASSERT(!MUTEX_HELD(&connp->conn_lock));
11580 
11581 	if (is_system_labeled() && connp->conn_anon_port) {
11582 		(void) tsol_mlp_anon(crgetzone(connp->conn_cred),
11583 		    connp->conn_mlp_type, connp->conn_proto,
11584 		    ntohs(connp->conn_lport), B_FALSE);
11585 		connp->conn_anon_port = 0;
11586 	}
11587 	connp->conn_mlp_type = mlptSingle;
11588 
11589 	ipcl_hash_remove(connp);
11590 }
11591 
11592 /*
11593  * Used for deciding the MSS size for the upper layer. Thus
11594  * we need to check the outbound policy values in the conn.
11595  */
11596 int
conn_ipsec_length(conn_t * connp)11597 conn_ipsec_length(conn_t *connp)
11598 {
11599 	ipsec_latch_t *ipl;
11600 
11601 	ipl = connp->conn_latch;
11602 	if (ipl == NULL)
11603 		return (0);
11604 
11605 	if (connp->conn_ixa->ixa_ipsec_policy == NULL)
11606 		return (0);
11607 
11608 	return (connp->conn_ixa->ixa_ipsec_policy->ipsp_act->ipa_ovhd);
11609 }
11610 
11611 /*
11612  * Returns an estimate of the IPsec headers size. This is used if
11613  * we don't want to call into IPsec to get the exact size.
11614  */
11615 int
ipsec_out_extra_length(ip_xmit_attr_t * ixa)11616 ipsec_out_extra_length(ip_xmit_attr_t *ixa)
11617 {
11618 	ipsec_action_t *a;
11619 
11620 	if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
11621 		return (0);
11622 
11623 	a = ixa->ixa_ipsec_action;
11624 	if (a == NULL) {
11625 		ASSERT(ixa->ixa_ipsec_policy != NULL);
11626 		a = ixa->ixa_ipsec_policy->ipsp_act;
11627 	}
11628 	ASSERT(a != NULL);
11629 
11630 	return (a->ipa_ovhd);
11631 }
11632 
11633 /*
11634  * If there are any source route options, return the true final
11635  * destination. Otherwise, return the destination.
11636  */
11637 ipaddr_t
ip_get_dst(ipha_t * ipha)11638 ip_get_dst(ipha_t *ipha)
11639 {
11640 	ipoptp_t	opts;
11641 	uchar_t		*opt;
11642 	uint8_t		optval;
11643 	uint8_t		optlen;
11644 	ipaddr_t	dst;
11645 	uint32_t off;
11646 
11647 	dst = ipha->ipha_dst;
11648 
11649 	if (IS_SIMPLE_IPH(ipha))
11650 		return (dst);
11651 
11652 	for (optval = ipoptp_first(&opts, ipha);
11653 	    optval != IPOPT_EOL;
11654 	    optval = ipoptp_next(&opts)) {
11655 		opt = opts.ipoptp_cur;
11656 		optlen = opts.ipoptp_len;
11657 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11658 		switch (optval) {
11659 		case IPOPT_SSRR:
11660 		case IPOPT_LSRR:
11661 			off = opt[IPOPT_OFFSET];
11662 			/*
11663 			 * If one of the conditions is true, it means
11664 			 * end of options and dst already has the right
11665 			 * value.
11666 			 */
11667 			if (!(optlen < IP_ADDR_LEN || off > optlen - 3)) {
11668 				off = optlen - IP_ADDR_LEN;
11669 				bcopy(&opt[off], &dst, IP_ADDR_LEN);
11670 			}
11671 			return (dst);
11672 		default:
11673 			break;
11674 		}
11675 	}
11676 
11677 	return (dst);
11678 }
11679 
11680 /*
11681  * Outbound IP fragmentation routine.
11682  * Assumes the caller has checked whether or not fragmentation should
11683  * be allowed. Here we copy the DF bit from the header to all the generated
11684  * fragments.
11685  */
11686 int
ip_fragment_v4(mblk_t * mp_orig,nce_t * nce,iaflags_t ixaflags,uint_t pkt_len,uint32_t max_frag,uint32_t xmit_hint,zoneid_t szone,zoneid_t nolzid,pfirepostfrag_t postfragfn,uintptr_t * ixa_cookie)11687 ip_fragment_v4(mblk_t *mp_orig, nce_t *nce, iaflags_t ixaflags,
11688     uint_t pkt_len, uint32_t max_frag, uint32_t xmit_hint, zoneid_t szone,
11689     zoneid_t nolzid, pfirepostfrag_t postfragfn, uintptr_t *ixa_cookie)
11690 {
11691 	int		i1;
11692 	int		hdr_len;
11693 	mblk_t		*hdr_mp;
11694 	ipha_t		*ipha;
11695 	int		ip_data_end;
11696 	int		len;
11697 	mblk_t		*mp = mp_orig;
11698 	int		offset;
11699 	ill_t		*ill = nce->nce_ill;
11700 	ip_stack_t	*ipst = ill->ill_ipst;
11701 	mblk_t		*carve_mp;
11702 	uint32_t	frag_flag;
11703 	uint_t		priority = mp->b_band;
11704 	int		error = 0;
11705 
11706 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragReqds);
11707 
11708 	if (pkt_len != msgdsize(mp)) {
11709 		ip0dbg(("Packet length mismatch: %d, %ld\n",
11710 		    pkt_len, msgdsize(mp)));
11711 		freemsg(mp);
11712 		return (EINVAL);
11713 	}
11714 
11715 	if (max_frag == 0) {
11716 		ip1dbg(("ip_fragment_v4: max_frag is zero. Dropping packet\n"));
11717 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11718 		ip_drop_output("FragFails: zero max_frag", mp, ill);
11719 		freemsg(mp);
11720 		return (EINVAL);
11721 	}
11722 
11723 	ASSERT(MBLKL(mp) >= sizeof (ipha_t));
11724 	ipha = (ipha_t *)mp->b_rptr;
11725 	ASSERT(ntohs(ipha->ipha_length) == pkt_len);
11726 	frag_flag = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_DF;
11727 
11728 	/*
11729 	 * Establish the starting offset.  May not be zero if we are fragging
11730 	 * a fragment that is being forwarded.
11731 	 */
11732 	offset = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET;
11733 
11734 	/* TODO why is this test needed? */
11735 	if (((max_frag - ntohs(ipha->ipha_length)) & ~7) < 8) {
11736 		/* TODO: notify ulp somehow */
11737 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11738 		ip_drop_output("FragFails: bad starting offset", mp, ill);
11739 		freemsg(mp);
11740 		return (EINVAL);
11741 	}
11742 
11743 	hdr_len = IPH_HDR_LENGTH(ipha);
11744 	ipha->ipha_hdr_checksum = 0;
11745 
11746 	/*
11747 	 * Establish the number of bytes maximum per frag, after putting
11748 	 * in the header.
11749 	 */
11750 	len = (max_frag - hdr_len) & ~7;
11751 
11752 	/* Get a copy of the header for the trailing frags */
11753 	hdr_mp = ip_fragment_copyhdr((uchar_t *)ipha, hdr_len, offset, ipst,
11754 	    mp);
11755 	if (hdr_mp == NULL) {
11756 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11757 		ip_drop_output("FragFails: no hdr_mp", mp, ill);
11758 		freemsg(mp);
11759 		return (ENOBUFS);
11760 	}
11761 
11762 	/* Store the starting offset, with the MoreFrags flag. */
11763 	i1 = offset | IPH_MF | frag_flag;
11764 	ipha->ipha_fragment_offset_and_flags = htons((uint16_t)i1);
11765 
11766 	/* Establish the ending byte offset, based on the starting offset. */
11767 	offset <<= 3;
11768 	ip_data_end = offset + ntohs(ipha->ipha_length) - hdr_len;
11769 
11770 	/* Store the length of the first fragment in the IP header. */
11771 	i1 = len + hdr_len;
11772 	ASSERT(i1 <= IP_MAXPACKET);
11773 	ipha->ipha_length = htons((uint16_t)i1);
11774 
11775 	/*
11776 	 * Compute the IP header checksum for the first frag.  We have to
11777 	 * watch out that we stop at the end of the header.
11778 	 */
11779 	ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11780 
11781 	/*
11782 	 * Now carve off the first frag.  Note that this will include the
11783 	 * original IP header.
11784 	 */
11785 	if (!(mp = ip_carve_mp(&mp_orig, i1))) {
11786 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11787 		ip_drop_output("FragFails: could not carve mp", mp_orig, ill);
11788 		freeb(hdr_mp);
11789 		freemsg(mp_orig);
11790 		return (ENOBUFS);
11791 	}
11792 
11793 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11794 
11795 	error = postfragfn(mp, nce, ixaflags, i1, xmit_hint, szone, nolzid,
11796 	    ixa_cookie);
11797 	if (error != 0 && error != EWOULDBLOCK) {
11798 		/* No point in sending the other fragments */
11799 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11800 		ip_drop_output("FragFails: postfragfn failed", mp_orig, ill);
11801 		freeb(hdr_mp);
11802 		freemsg(mp_orig);
11803 		return (error);
11804 	}
11805 
11806 	/* No need to redo state machine in loop */
11807 	ixaflags &= ~IXAF_REACH_CONF;
11808 
11809 	/* Advance the offset to the second frag starting point. */
11810 	offset += len;
11811 	/*
11812 	 * Update hdr_len from the copied header - there might be less options
11813 	 * in the later fragments.
11814 	 */
11815 	hdr_len = IPH_HDR_LENGTH(hdr_mp->b_rptr);
11816 	/* Loop until done. */
11817 	for (;;) {
11818 		uint16_t	offset_and_flags;
11819 		uint16_t	ip_len;
11820 
11821 		if (ip_data_end - offset > len) {
11822 			/*
11823 			 * Carve off the appropriate amount from the original
11824 			 * datagram.
11825 			 */
11826 			if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11827 				mp = NULL;
11828 				break;
11829 			}
11830 			/*
11831 			 * More frags after this one.  Get another copy
11832 			 * of the header.
11833 			 */
11834 			if (carve_mp->b_datap->db_ref == 1 &&
11835 			    hdr_mp->b_wptr - hdr_mp->b_rptr <
11836 			    carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11837 				/* Inline IP header */
11838 				carve_mp->b_rptr -= hdr_mp->b_wptr -
11839 				    hdr_mp->b_rptr;
11840 				bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11841 				    hdr_mp->b_wptr - hdr_mp->b_rptr);
11842 				mp = carve_mp;
11843 			} else {
11844 				if (!(mp = copyb(hdr_mp))) {
11845 					freemsg(carve_mp);
11846 					break;
11847 				}
11848 				/* Get priority marking, if any. */
11849 				mp->b_band = priority;
11850 				mp->b_cont = carve_mp;
11851 			}
11852 			ipha = (ipha_t *)mp->b_rptr;
11853 			offset_and_flags = IPH_MF;
11854 		} else {
11855 			/*
11856 			 * Last frag.  Consume the header. Set len to
11857 			 * the length of this last piece.
11858 			 */
11859 			len = ip_data_end - offset;
11860 
11861 			/*
11862 			 * Carve off the appropriate amount from the original
11863 			 * datagram.
11864 			 */
11865 			if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11866 				mp = NULL;
11867 				break;
11868 			}
11869 			if (carve_mp->b_datap->db_ref == 1 &&
11870 			    hdr_mp->b_wptr - hdr_mp->b_rptr <
11871 			    carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11872 				/* Inline IP header */
11873 				carve_mp->b_rptr -= hdr_mp->b_wptr -
11874 				    hdr_mp->b_rptr;
11875 				bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11876 				    hdr_mp->b_wptr - hdr_mp->b_rptr);
11877 				mp = carve_mp;
11878 				freeb(hdr_mp);
11879 				hdr_mp = mp;
11880 			} else {
11881 				mp = hdr_mp;
11882 				/* Get priority marking, if any. */
11883 				mp->b_band = priority;
11884 				mp->b_cont = carve_mp;
11885 			}
11886 			ipha = (ipha_t *)mp->b_rptr;
11887 			/* A frag of a frag might have IPH_MF non-zero */
11888 			offset_and_flags =
11889 			    ntohs(ipha->ipha_fragment_offset_and_flags) &
11890 			    IPH_MF;
11891 		}
11892 		offset_and_flags |= (uint16_t)(offset >> 3);
11893 		offset_and_flags |= (uint16_t)frag_flag;
11894 		/* Store the offset and flags in the IP header. */
11895 		ipha->ipha_fragment_offset_and_flags = htons(offset_and_flags);
11896 
11897 		/* Store the length in the IP header. */
11898 		ip_len = (uint16_t)(len + hdr_len);
11899 		ipha->ipha_length = htons(ip_len);
11900 
11901 		/*
11902 		 * Set the IP header checksum.	Note that mp is just
11903 		 * the header, so this is easy to pass to ip_csum.
11904 		 */
11905 		ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11906 
11907 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11908 
11909 		error = postfragfn(mp, nce, ixaflags, ip_len, xmit_hint, szone,
11910 		    nolzid, ixa_cookie);
11911 		/* All done if we just consumed the hdr_mp. */
11912 		if (mp == hdr_mp) {
11913 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragOKs);
11914 			return (error);
11915 		}
11916 		if (error != 0 && error != EWOULDBLOCK) {
11917 			DTRACE_PROBE2(ip__xmit__frag__fail, ill_t *, ill,
11918 			    mblk_t *, hdr_mp);
11919 			/* No point in sending the other fragments */
11920 			break;
11921 		}
11922 
11923 		/* Otherwise, advance and loop. */
11924 		offset += len;
11925 	}
11926 	/* Clean up following allocation failure. */
11927 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11928 	ip_drop_output("FragFails: loop ended", NULL, ill);
11929 	if (mp != hdr_mp)
11930 		freeb(hdr_mp);
11931 	if (mp != mp_orig)
11932 		freemsg(mp_orig);
11933 	return (error);
11934 }
11935 
11936 /*
11937  * Copy the header plus those options which have the copy bit set
11938  */
11939 static mblk_t *
ip_fragment_copyhdr(uchar_t * rptr,int hdr_len,int offset,ip_stack_t * ipst,mblk_t * src)11940 ip_fragment_copyhdr(uchar_t *rptr, int hdr_len, int offset, ip_stack_t *ipst,
11941     mblk_t *src)
11942 {
11943 	mblk_t	*mp;
11944 	uchar_t	*up;
11945 
11946 	/*
11947 	 * Quick check if we need to look for options without the copy bit
11948 	 * set
11949 	 */
11950 	mp = allocb_tmpl(ipst->ips_ip_wroff_extra + hdr_len, src);
11951 	if (!mp)
11952 		return (mp);
11953 	mp->b_rptr += ipst->ips_ip_wroff_extra;
11954 	if (hdr_len == IP_SIMPLE_HDR_LENGTH || offset != 0) {
11955 		bcopy(rptr, mp->b_rptr, hdr_len);
11956 		mp->b_wptr += hdr_len + ipst->ips_ip_wroff_extra;
11957 		return (mp);
11958 	}
11959 	up  = mp->b_rptr;
11960 	bcopy(rptr, up, IP_SIMPLE_HDR_LENGTH);
11961 	up += IP_SIMPLE_HDR_LENGTH;
11962 	rptr += IP_SIMPLE_HDR_LENGTH;
11963 	hdr_len -= IP_SIMPLE_HDR_LENGTH;
11964 	while (hdr_len > 0) {
11965 		uint32_t optval;
11966 		uint32_t optlen;
11967 
11968 		optval = *rptr;
11969 		if (optval == IPOPT_EOL)
11970 			break;
11971 		if (optval == IPOPT_NOP)
11972 			optlen = 1;
11973 		else
11974 			optlen = rptr[1];
11975 		if (optval & IPOPT_COPY) {
11976 			bcopy(rptr, up, optlen);
11977 			up += optlen;
11978 		}
11979 		rptr += optlen;
11980 		hdr_len -= optlen;
11981 	}
11982 	/*
11983 	 * Make sure that we drop an even number of words by filling
11984 	 * with EOL to the next word boundary.
11985 	 */
11986 	for (hdr_len = up - (mp->b_rptr + IP_SIMPLE_HDR_LENGTH);
11987 	    hdr_len & 0x3; hdr_len++)
11988 		*up++ = IPOPT_EOL;
11989 	mp->b_wptr = up;
11990 	/* Update header length */
11991 	mp->b_rptr[0] = (uint8_t)((IP_VERSION << 4) | ((up - mp->b_rptr) >> 2));
11992 	return (mp);
11993 }
11994 
11995 /*
11996  * Update any source route, record route, or timestamp options when
11997  * sending a packet back to ourselves.
11998  * Check that we are at end of strict source route.
11999  * The options have been sanity checked by ip_output_options().
12000  */
12001 void
ip_output_local_options(ipha_t * ipha,ip_stack_t * ipst)12002 ip_output_local_options(ipha_t *ipha, ip_stack_t *ipst)
12003 {
12004 	ipoptp_t	opts;
12005 	uchar_t		*opt;
12006 	uint8_t		optval;
12007 	uint8_t		optlen;
12008 	ipaddr_t	dst;
12009 	uint32_t	ts;
12010 	timestruc_t	now;
12011 	uint32_t	off = 0;
12012 
12013 	for (optval = ipoptp_first(&opts, ipha);
12014 	    optval != IPOPT_EOL;
12015 	    optval = ipoptp_next(&opts)) {
12016 		opt = opts.ipoptp_cur;
12017 		optlen = opts.ipoptp_len;
12018 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
12019 		switch (optval) {
12020 		case IPOPT_SSRR:
12021 		case IPOPT_LSRR:
12022 			off = opt[IPOPT_OFFSET];
12023 			off--;
12024 			if (optlen < IP_ADDR_LEN ||
12025 			    off > optlen - IP_ADDR_LEN) {
12026 				/* End of source route */
12027 				break;
12028 			}
12029 			/*
12030 			 * This will only happen if two consecutive entries
12031 			 * in the source route contains our address or if
12032 			 * it is a packet with a loose source route which
12033 			 * reaches us before consuming the whole source route
12034 			 */
12035 
12036 			if (optval == IPOPT_SSRR) {
12037 				return;
12038 			}
12039 			/*
12040 			 * Hack: instead of dropping the packet truncate the
12041 			 * source route to what has been used by filling the
12042 			 * rest with IPOPT_NOP.
12043 			 */
12044 			opt[IPOPT_OLEN] = (uint8_t)off;
12045 			while (off < optlen) {
12046 				opt[off++] = IPOPT_NOP;
12047 			}
12048 			break;
12049 		case IPOPT_RR:
12050 			off = opt[IPOPT_OFFSET];
12051 			off--;
12052 			if (optlen < IP_ADDR_LEN ||
12053 			    off > optlen - IP_ADDR_LEN) {
12054 				/* No more room - ignore */
12055 				ip1dbg((
12056 				    "ip_output_local_options: end of RR\n"));
12057 				break;
12058 			}
12059 			dst = htonl(INADDR_LOOPBACK);
12060 			bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
12061 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
12062 			break;
12063 		case IPOPT_TS:
12064 			/* Insert timestamp if there is romm */
12065 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12066 			case IPOPT_TS_TSONLY:
12067 				off = IPOPT_TS_TIMELEN;
12068 				break;
12069 			case IPOPT_TS_PRESPEC:
12070 			case IPOPT_TS_PRESPEC_RFC791:
12071 				/* Verify that the address matched */
12072 				off = opt[IPOPT_OFFSET] - 1;
12073 				bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
12074 				if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
12075 					/* Not for us */
12076 					break;
12077 				}
12078 				/* FALLTHROUGH */
12079 			case IPOPT_TS_TSANDADDR:
12080 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
12081 				break;
12082 			default:
12083 				/*
12084 				 * ip_*put_options should have already
12085 				 * dropped this packet.
12086 				 */
12087 				cmn_err(CE_PANIC, "ip_output_local_options: "
12088 				    "unknown IT - bug in ip_output_options?\n");
12089 			}
12090 			if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
12091 				/* Increase overflow counter */
12092 				off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
12093 				opt[IPOPT_POS_OV_FLG] = (uint8_t)
12094 				    (opt[IPOPT_POS_OV_FLG] & 0x0F) |
12095 				    (off << 4);
12096 				break;
12097 			}
12098 			off = opt[IPOPT_OFFSET] - 1;
12099 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12100 			case IPOPT_TS_PRESPEC:
12101 			case IPOPT_TS_PRESPEC_RFC791:
12102 			case IPOPT_TS_TSANDADDR:
12103 				dst = htonl(INADDR_LOOPBACK);
12104 				bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
12105 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
12106 				/* FALLTHROUGH */
12107 			case IPOPT_TS_TSONLY:
12108 				off = opt[IPOPT_OFFSET] - 1;
12109 				/* Compute # of milliseconds since midnight */
12110 				gethrestime(&now);
12111 				ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
12112 				    NSEC2MSEC(now.tv_nsec);
12113 				bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
12114 				opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
12115 				break;
12116 			}
12117 			break;
12118 		}
12119 	}
12120 }
12121 
12122 /*
12123  * Prepend an M_DATA fastpath header, and if none present prepend a
12124  * DL_UNITDATA_REQ. Frees the mblk on failure.
12125  *
12126  * nce_dlur_mp and nce_fp_mp can not disappear once they have been set.
12127  * If there is a change to them, the nce will be deleted (condemned) and
12128  * a new nce_t will be created when packets are sent. Thus we need no locks
12129  * to access those fields.
12130  *
12131  * We preserve b_band to support IPQoS. If a DL_UNITDATA_REQ is prepended
12132  * we place b_band in dl_priority.dl_max.
12133  */
12134 static mblk_t *
ip_xmit_attach_llhdr(mblk_t * mp,nce_t * nce)12135 ip_xmit_attach_llhdr(mblk_t *mp, nce_t *nce)
12136 {
12137 	uint_t	hlen;
12138 	mblk_t *mp1;
12139 	uint_t	priority;
12140 	uchar_t *rptr;
12141 
12142 	rptr = mp->b_rptr;
12143 
12144 	ASSERT(DB_TYPE(mp) == M_DATA);
12145 	priority = mp->b_band;
12146 
12147 	ASSERT(nce != NULL);
12148 	if ((mp1 = nce->nce_fp_mp) != NULL) {
12149 		hlen = MBLKL(mp1);
12150 		/*
12151 		 * Check if we have enough room to prepend fastpath
12152 		 * header
12153 		 */
12154 		if (hlen != 0 && (rptr - mp->b_datap->db_base) >= hlen) {
12155 			rptr -= hlen;
12156 			bcopy(mp1->b_rptr, rptr, hlen);
12157 			/*
12158 			 * Set the b_rptr to the start of the link layer
12159 			 * header
12160 			 */
12161 			mp->b_rptr = rptr;
12162 			return (mp);
12163 		}
12164 		mp1 = copyb(mp1);
12165 		if (mp1 == NULL) {
12166 			ill_t *ill = nce->nce_ill;
12167 
12168 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12169 			ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12170 			freemsg(mp);
12171 			return (NULL);
12172 		}
12173 		mp1->b_band = priority;
12174 		mp1->b_cont = mp;
12175 		DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
12176 		DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
12177 		DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
12178 		DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
12179 		DB_LSOMSS(mp1) = DB_LSOMSS(mp);
12180 		DTRACE_PROBE1(ip__xmit__copyb, (mblk_t *), mp1);
12181 		/*
12182 		 * XXX disable ICK_VALID and compute checksum
12183 		 * here; can happen if nce_fp_mp changes and
12184 		 * it can't be copied now due to insufficient
12185 		 * space. (unlikely, fp mp can change, but it
12186 		 * does not increase in length)
12187 		 */
12188 		return (mp1);
12189 	}
12190 	mp1 = copyb(nce->nce_dlur_mp);
12191 
12192 	if (mp1 == NULL) {
12193 		ill_t *ill = nce->nce_ill;
12194 
12195 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12196 		ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12197 		freemsg(mp);
12198 		return (NULL);
12199 	}
12200 	mp1->b_cont = mp;
12201 	if (priority != 0) {
12202 		mp1->b_band = priority;
12203 		((dl_unitdata_req_t *)(mp1->b_rptr))->dl_priority.dl_max =
12204 		    priority;
12205 	}
12206 	return (mp1);
12207 }
12208 
12209 /*
12210  * Finish the outbound IPsec processing. This function is called from
12211  * ipsec_out_process() if the IPsec packet was processed
12212  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12213  * asynchronously.
12214  *
12215  * This is common to IPv4 and IPv6.
12216  */
12217 int
ip_output_post_ipsec(mblk_t * mp,ip_xmit_attr_t * ixa)12218 ip_output_post_ipsec(mblk_t *mp, ip_xmit_attr_t *ixa)
12219 {
12220 	iaflags_t	ixaflags = ixa->ixa_flags;
12221 	uint_t		pktlen;
12222 
12223 
12224 	/* AH/ESP don't update ixa_pktlen when they modify the packet */
12225 	if (ixaflags & IXAF_IS_IPV4) {
12226 		ipha_t		*ipha = (ipha_t *)mp->b_rptr;
12227 
12228 		ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12229 		pktlen = ntohs(ipha->ipha_length);
12230 	} else {
12231 		ip6_t		*ip6h = (ip6_t *)mp->b_rptr;
12232 
12233 		ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12234 		pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12235 	}
12236 
12237 	/*
12238 	 * We release any hard reference on the SAs here to make
12239 	 * sure the SAs can be garbage collected. ipsr_sa has a soft reference
12240 	 * on the SAs.
12241 	 * If in the future we want the hard latching of the SAs in the
12242 	 * ip_xmit_attr_t then we should remove this.
12243 	 */
12244 	if (ixa->ixa_ipsec_esp_sa != NULL) {
12245 		IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12246 		ixa->ixa_ipsec_esp_sa = NULL;
12247 	}
12248 	if (ixa->ixa_ipsec_ah_sa != NULL) {
12249 		IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12250 		ixa->ixa_ipsec_ah_sa = NULL;
12251 	}
12252 
12253 	/* Do we need to fragment? */
12254 	if ((ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR) ||
12255 	    pktlen > ixa->ixa_fragsize) {
12256 		if (ixaflags & IXAF_IS_IPV4) {
12257 			ASSERT(!(ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR));
12258 			/*
12259 			 * We check for the DF case in ipsec_out_process
12260 			 * hence this only handles the non-DF case.
12261 			 */
12262 			return (ip_fragment_v4(mp, ixa->ixa_nce, ixa->ixa_flags,
12263 			    pktlen, ixa->ixa_fragsize,
12264 			    ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12265 			    ixa->ixa_no_loop_zoneid, ixa->ixa_postfragfn,
12266 			    &ixa->ixa_cookie));
12267 		} else {
12268 			mp = ip_fraghdr_add_v6(mp, ixa->ixa_ident, ixa);
12269 			if (mp == NULL) {
12270 				/* MIB and ip_drop_output already done */
12271 				return (ENOMEM);
12272 			}
12273 			pktlen += sizeof (ip6_frag_t);
12274 			if (pktlen > ixa->ixa_fragsize) {
12275 				return (ip_fragment_v6(mp, ixa->ixa_nce,
12276 				    ixa->ixa_flags, pktlen,
12277 				    ixa->ixa_fragsize, ixa->ixa_xmit_hint,
12278 				    ixa->ixa_zoneid, ixa->ixa_no_loop_zoneid,
12279 				    ixa->ixa_postfragfn, &ixa->ixa_cookie));
12280 			}
12281 		}
12282 	}
12283 	return ((ixa->ixa_postfragfn)(mp, ixa->ixa_nce, ixa->ixa_flags,
12284 	    pktlen, ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12285 	    ixa->ixa_no_loop_zoneid, NULL));
12286 }
12287 
12288 /*
12289  * Finish the inbound IPsec processing. This function is called from
12290  * ipsec_out_process() if the IPsec packet was processed
12291  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12292  * asynchronously.
12293  *
12294  * This is common to IPv4 and IPv6.
12295  */
12296 void
ip_input_post_ipsec(mblk_t * mp,ip_recv_attr_t * ira)12297 ip_input_post_ipsec(mblk_t *mp, ip_recv_attr_t *ira)
12298 {
12299 	iaflags_t	iraflags = ira->ira_flags;
12300 
12301 	/* Length might have changed */
12302 	if (iraflags & IRAF_IS_IPV4) {
12303 		ipha_t		*ipha = (ipha_t *)mp->b_rptr;
12304 
12305 		ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12306 		ira->ira_pktlen = ntohs(ipha->ipha_length);
12307 		ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
12308 		ira->ira_protocol = ipha->ipha_protocol;
12309 
12310 		ip_fanout_v4(mp, ipha, ira);
12311 	} else {
12312 		ip6_t		*ip6h = (ip6_t *)mp->b_rptr;
12313 		uint8_t		*nexthdrp;
12314 
12315 		ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12316 		ira->ira_pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12317 		if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ira->ira_ip_hdr_length,
12318 		    &nexthdrp)) {
12319 			/* Malformed packet */
12320 			BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
12321 			ip_drop_input("ipIfStatsInDiscards", mp, ira->ira_ill);
12322 			freemsg(mp);
12323 			return;
12324 		}
12325 		ira->ira_protocol = *nexthdrp;
12326 		ip_fanout_v6(mp, ip6h, ira);
12327 	}
12328 }
12329 
12330 /*
12331  * Select which AH & ESP SA's to use (if any) for the outbound packet.
12332  *
12333  * If this function returns B_TRUE, the requested SA's have been filled
12334  * into the ixa_ipsec_*_sa pointers.
12335  *
12336  * If the function returns B_FALSE, the packet has been "consumed", most
12337  * likely by an ACQUIRE sent up via PF_KEY to a key management daemon.
12338  *
12339  * The SA references created by the protocol-specific "select"
12340  * function will be released in ip_output_post_ipsec.
12341  */
12342 static boolean_t
ipsec_out_select_sa(mblk_t * mp,ip_xmit_attr_t * ixa)12343 ipsec_out_select_sa(mblk_t *mp, ip_xmit_attr_t *ixa)
12344 {
12345 	boolean_t need_ah_acquire = B_FALSE, need_esp_acquire = B_FALSE;
12346 	ipsec_policy_t *pp;
12347 	ipsec_action_t *ap;
12348 
12349 	ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12350 	ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12351 	    (ixa->ixa_ipsec_action != NULL));
12352 
12353 	ap = ixa->ixa_ipsec_action;
12354 	if (ap == NULL) {
12355 		pp = ixa->ixa_ipsec_policy;
12356 		ASSERT(pp != NULL);
12357 		ap = pp->ipsp_act;
12358 		ASSERT(ap != NULL);
12359 	}
12360 
12361 	/*
12362 	 * We have an action.  now, let's select SA's.
12363 	 * A side effect of setting ixa_ipsec_*_sa is that it will
12364 	 * be cached in the conn_t.
12365 	 */
12366 	if (ap->ipa_want_esp) {
12367 		if (ixa->ixa_ipsec_esp_sa == NULL) {
12368 			need_esp_acquire = !ipsec_outbound_sa(mp, ixa,
12369 			    IPPROTO_ESP);
12370 		}
12371 		ASSERT(need_esp_acquire || ixa->ixa_ipsec_esp_sa != NULL);
12372 	}
12373 
12374 	if (ap->ipa_want_ah) {
12375 		if (ixa->ixa_ipsec_ah_sa == NULL) {
12376 			need_ah_acquire = !ipsec_outbound_sa(mp, ixa,
12377 			    IPPROTO_AH);
12378 		}
12379 		ASSERT(need_ah_acquire || ixa->ixa_ipsec_ah_sa != NULL);
12380 		/*
12381 		 * The ESP and AH processing order needs to be preserved
12382 		 * when both protocols are required (ESP should be applied
12383 		 * before AH for an outbound packet). Force an ESP ACQUIRE
12384 		 * when both ESP and AH are required, and an AH ACQUIRE
12385 		 * is needed.
12386 		 */
12387 		if (ap->ipa_want_esp && need_ah_acquire)
12388 			need_esp_acquire = B_TRUE;
12389 	}
12390 
12391 	/*
12392 	 * Send an ACQUIRE (extended, regular, or both) if we need one.
12393 	 * Release SAs that got referenced, but will not be used until we
12394 	 * acquire _all_ of the SAs we need.
12395 	 */
12396 	if (need_ah_acquire || need_esp_acquire) {
12397 		if (ixa->ixa_ipsec_ah_sa != NULL) {
12398 			IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12399 			ixa->ixa_ipsec_ah_sa = NULL;
12400 		}
12401 		if (ixa->ixa_ipsec_esp_sa != NULL) {
12402 			IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12403 			ixa->ixa_ipsec_esp_sa = NULL;
12404 		}
12405 
12406 		sadb_acquire(mp, ixa, need_ah_acquire, need_esp_acquire);
12407 		return (B_FALSE);
12408 	}
12409 
12410 	return (B_TRUE);
12411 }
12412 
12413 /*
12414  * Handle IPsec output processing.
12415  * This function is only entered once for a given packet.
12416  * We try to do things synchronously, but if we need to have user-level
12417  * set up SAs, or ESP or AH uses asynchronous kEF, then the operation
12418  * will be completed
12419  *  - when the SAs are added in esp_add_sa_finish/ah_add_sa_finish
12420  *  - when asynchronous ESP is done it will do AH
12421  *
12422  * In all cases we come back in ip_output_post_ipsec() to fragment and
12423  * send out the packet.
12424  */
12425 int
ipsec_out_process(mblk_t * mp,ip_xmit_attr_t * ixa)12426 ipsec_out_process(mblk_t *mp, ip_xmit_attr_t *ixa)
12427 {
12428 	ill_t		*ill = ixa->ixa_nce->nce_ill;
12429 	ip_stack_t	*ipst = ixa->ixa_ipst;
12430 	ipsec_stack_t	*ipss;
12431 	ipsec_policy_t	*pp;
12432 	ipsec_action_t	*ap;
12433 
12434 	ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12435 
12436 	ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12437 	    (ixa->ixa_ipsec_action != NULL));
12438 
12439 	ipss = ipst->ips_netstack->netstack_ipsec;
12440 	if (!ipsec_loaded(ipss)) {
12441 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12442 		ip_drop_packet(mp, B_TRUE, ill,
12443 		    DROPPER(ipss, ipds_ip_ipsec_not_loaded),
12444 		    &ipss->ipsec_dropper);
12445 		return (ENOTSUP);
12446 	}
12447 
12448 	ap = ixa->ixa_ipsec_action;
12449 	if (ap == NULL) {
12450 		pp = ixa->ixa_ipsec_policy;
12451 		ASSERT(pp != NULL);
12452 		ap = pp->ipsp_act;
12453 		ASSERT(ap != NULL);
12454 	}
12455 
12456 	/* Handle explicit drop action and bypass. */
12457 	switch (ap->ipa_act.ipa_type) {
12458 	case IPSEC_ACT_DISCARD:
12459 	case IPSEC_ACT_REJECT:
12460 		ip_drop_packet(mp, B_FALSE, ill,
12461 		    DROPPER(ipss, ipds_spd_explicit), &ipss->ipsec_spd_dropper);
12462 		return (EHOSTUNREACH);	/* IPsec policy failure */
12463 	case IPSEC_ACT_BYPASS:
12464 		return (ip_output_post_ipsec(mp, ixa));
12465 	}
12466 
12467 	/*
12468 	 * The order of processing is first insert a IP header if needed.
12469 	 * Then insert the ESP header and then the AH header.
12470 	 */
12471 	if ((ixa->ixa_flags & IXAF_IS_IPV4) && ap->ipa_want_se) {
12472 		/*
12473 		 * First get the outer IP header before sending
12474 		 * it to ESP.
12475 		 */
12476 		ipha_t *oipha, *iipha;
12477 		mblk_t *outer_mp, *inner_mp;
12478 
12479 		if ((outer_mp = allocb(sizeof (ipha_t), BPRI_HI)) == NULL) {
12480 			(void) mi_strlog(ill->ill_rq, 0,
12481 			    SL_ERROR|SL_TRACE|SL_CONSOLE,
12482 			    "ipsec_out_process: "
12483 			    "Self-Encapsulation failed: Out of memory\n");
12484 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12485 			ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12486 			freemsg(mp);
12487 			return (ENOBUFS);
12488 		}
12489 		inner_mp = mp;
12490 		ASSERT(inner_mp->b_datap->db_type == M_DATA);
12491 		oipha = (ipha_t *)outer_mp->b_rptr;
12492 		iipha = (ipha_t *)inner_mp->b_rptr;
12493 		*oipha = *iipha;
12494 		outer_mp->b_wptr += sizeof (ipha_t);
12495 		oipha->ipha_length = htons(ntohs(iipha->ipha_length) +
12496 		    sizeof (ipha_t));
12497 		oipha->ipha_protocol = IPPROTO_ENCAP;
12498 		oipha->ipha_version_and_hdr_length =
12499 		    IP_SIMPLE_HDR_VERSION;
12500 		oipha->ipha_hdr_checksum = 0;
12501 		oipha->ipha_hdr_checksum = ip_csum_hdr(oipha);
12502 		outer_mp->b_cont = inner_mp;
12503 		mp = outer_mp;
12504 
12505 		ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
12506 	}
12507 
12508 	/* If we need to wait for a SA then we can't return any errno */
12509 	if (((ap->ipa_want_ah && (ixa->ixa_ipsec_ah_sa == NULL)) ||
12510 	    (ap->ipa_want_esp && (ixa->ixa_ipsec_esp_sa == NULL))) &&
12511 	    !ipsec_out_select_sa(mp, ixa))
12512 		return (0);
12513 
12514 	/*
12515 	 * By now, we know what SA's to use.  Toss over to ESP & AH
12516 	 * to do the heavy lifting.
12517 	 */
12518 	if (ap->ipa_want_esp) {
12519 		ASSERT(ixa->ixa_ipsec_esp_sa != NULL);
12520 
12521 		mp = ixa->ixa_ipsec_esp_sa->ipsa_output_func(mp, ixa);
12522 		if (mp == NULL) {
12523 			/*
12524 			 * Either it failed or is pending. In the former case
12525 			 * ipIfStatsInDiscards was increased.
12526 			 */
12527 			return (0);
12528 		}
12529 	}
12530 
12531 	if (ap->ipa_want_ah) {
12532 		ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
12533 
12534 		mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(mp, ixa);
12535 		if (mp == NULL) {
12536 			/*
12537 			 * Either it failed or is pending. In the former case
12538 			 * ipIfStatsInDiscards was increased.
12539 			 */
12540 			return (0);
12541 		}
12542 	}
12543 	/*
12544 	 * We are done with IPsec processing. Send it over
12545 	 * the wire.
12546 	 */
12547 	return (ip_output_post_ipsec(mp, ixa));
12548 }
12549 
12550 /*
12551  * ioctls that go through a down/up sequence may need to wait for the down
12552  * to complete. This involves waiting for the ire and ipif refcnts to go down
12553  * to zero. Subsequently the ioctl is restarted from ipif_ill_refrele_tail.
12554  */
12555 /* ARGSUSED */
12556 void
ip_reprocess_ioctl(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * dummy_arg)12557 ip_reprocess_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
12558 {
12559 	struct iocblk *iocp;
12560 	mblk_t *mp1;
12561 	ip_ioctl_cmd_t *ipip;
12562 	int err;
12563 	sin_t	*sin;
12564 	struct lifreq *lifr;
12565 	struct ifreq *ifr;
12566 
12567 	iocp = (struct iocblk *)mp->b_rptr;
12568 	ASSERT(ipsq != NULL);
12569 	/* Existence of mp1 verified in ip_wput_nondata */
12570 	mp1 = mp->b_cont->b_cont;
12571 	ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12572 	if (ipip->ipi_cmd == SIOCSLIFNAME || ipip->ipi_cmd == IF_UNITSEL) {
12573 		/*
12574 		 * Special case where ipx_current_ipif is not set:
12575 		 * ill_phyint_reinit merged the v4 and v6 into a single ipsq.
12576 		 * We are here as were not able to complete the operation in
12577 		 * ipif_set_values because we could not become exclusive on
12578 		 * the new ipsq.
12579 		 */
12580 		ill_t *ill = q->q_ptr;
12581 		ipsq_current_start(ipsq, ill->ill_ipif, ipip->ipi_cmd);
12582 	}
12583 	ASSERT(ipsq->ipsq_xop->ipx_current_ipif != NULL);
12584 
12585 	if (ipip->ipi_cmd_type == IF_CMD) {
12586 		/* This a old style SIOC[GS]IF* command */
12587 		ifr = (struct ifreq *)mp1->b_rptr;
12588 		sin = (sin_t *)&ifr->ifr_addr;
12589 	} else if (ipip->ipi_cmd_type == LIF_CMD) {
12590 		/* This a new style SIOC[GS]LIF* command */
12591 		lifr = (struct lifreq *)mp1->b_rptr;
12592 		sin = (sin_t *)&lifr->lifr_addr;
12593 	} else {
12594 		sin = NULL;
12595 	}
12596 
12597 	err = (*ipip->ipi_func_restart)(ipsq->ipsq_xop->ipx_current_ipif, sin,
12598 	    q, mp, ipip, mp1->b_rptr);
12599 
12600 	DTRACE_PROBE4(ipif__ioctl, char *, "ip_reprocess_ioctl finish",
12601 	    int, ipip->ipi_cmd,
12602 	    ill_t *, ipsq->ipsq_xop->ipx_current_ipif->ipif_ill,
12603 	    ipif_t *, ipsq->ipsq_xop->ipx_current_ipif);
12604 
12605 	ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12606 }
12607 
12608 /*
12609  * ioctl processing
12610  *
12611  * ioctl processing starts with ip_sioctl_copyin_setup(), which looks up
12612  * the ioctl command in the ioctl tables, determines the copyin data size
12613  * from the ipi_copyin_size field, and does an mi_copyin() of that size.
12614  *
12615  * ioctl processing then continues when the M_IOCDATA makes its way down to
12616  * ip_wput_nondata().  The ioctl is looked up again in the ioctl table, its
12617  * associated 'conn' is refheld till the end of the ioctl and the general
12618  * ioctl processing function ip_process_ioctl() is called to extract the
12619  * arguments and process the ioctl.  To simplify extraction, ioctl commands
12620  * are "typed" based on the arguments they take (e.g., LIF_CMD which takes a
12621  * `struct lifreq'), and a common extract function (e.g., ip_extract_lifreq())
12622  * is used to extract the ioctl's arguments.
12623  *
12624  * ip_process_ioctl determines if the ioctl needs to be serialized, and if
12625  * so goes thru the serialization primitive ipsq_try_enter. Then the
12626  * appropriate function to handle the ioctl is called based on the entry in
12627  * the ioctl table. ioctl completion is encapsulated in ip_ioctl_finish
12628  * which also refreleases the 'conn' that was refheld at the start of the
12629  * ioctl. Finally ipsq_exit is called if needed to exit the ipsq.
12630  *
12631  * Many exclusive ioctls go thru an internal down up sequence as part of
12632  * the operation. For example an attempt to change the IP address of an
12633  * ipif entails ipif_down, set address, ipif_up. Bringing down the interface
12634  * does all the cleanup such as deleting all ires that use this address.
12635  * Then we need to wait till all references to the interface go away.
12636  */
12637 void
ip_process_ioctl(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * arg)12638 ip_process_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *arg)
12639 {
12640 	struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
12641 	ip_ioctl_cmd_t *ipip = arg;
12642 	ip_extract_func_t *extract_funcp;
12643 	cmd_info_t ci;
12644 	int err;
12645 	boolean_t entered_ipsq = B_FALSE;
12646 
12647 	ip3dbg(("ip_process_ioctl: ioctl %X\n", iocp->ioc_cmd));
12648 
12649 	if (ipip == NULL)
12650 		ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12651 
12652 	/*
12653 	 * SIOCLIFADDIF needs to go thru a special path since the
12654 	 * ill may not exist yet. This happens in the case of lo0
12655 	 * which is created using this ioctl.
12656 	 */
12657 	if (ipip->ipi_cmd == SIOCLIFADDIF) {
12658 		err = ip_sioctl_addif(NULL, NULL, q, mp, NULL, NULL);
12659 		DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish",
12660 		    int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12661 		ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12662 		return;
12663 	}
12664 
12665 	ci.ci_ipif = NULL;
12666 	extract_funcp = NULL;
12667 	switch (ipip->ipi_cmd_type) {
12668 	case MISC_CMD:
12669 	case MSFILT_CMD:
12670 		/*
12671 		 * All MISC_CMD ioctls come in here -- e.g. SIOCGLIFCONF.
12672 		 */
12673 		if (ipip->ipi_cmd == IF_UNITSEL) {
12674 			/* ioctl comes down the ill */
12675 			ci.ci_ipif = ((ill_t *)q->q_ptr)->ill_ipif;
12676 			ipif_refhold(ci.ci_ipif);
12677 		}
12678 		err = 0;
12679 		ci.ci_sin = NULL;
12680 		ci.ci_sin6 = NULL;
12681 		ci.ci_lifr = NULL;
12682 		extract_funcp = NULL;
12683 		break;
12684 
12685 	case IF_CMD:
12686 	case LIF_CMD:
12687 		extract_funcp = ip_extract_lifreq;
12688 		break;
12689 
12690 	case ARP_CMD:
12691 	case XARP_CMD:
12692 		extract_funcp = ip_extract_arpreq;
12693 		break;
12694 
12695 	default:
12696 		ASSERT(0);
12697 	}
12698 
12699 	if (extract_funcp != NULL) {
12700 		err = (*extract_funcp)(q, mp, ipip, &ci);
12701 		if (err != 0) {
12702 			DTRACE_PROBE4(ipif__ioctl,
12703 			    char *, "ip_process_ioctl finish err",
12704 			    int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12705 			ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12706 			return;
12707 		}
12708 
12709 		/*
12710 		 * All of the extraction functions return a refheld ipif.
12711 		 */
12712 		ASSERT(ci.ci_ipif != NULL);
12713 	}
12714 
12715 	if (!(ipip->ipi_flags & IPI_WR)) {
12716 		/*
12717 		 * A return value of EINPROGRESS means the ioctl is
12718 		 * either queued and waiting for some reason or has
12719 		 * already completed.
12720 		 */
12721 		err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip,
12722 		    ci.ci_lifr);
12723 		if (ci.ci_ipif != NULL) {
12724 			DTRACE_PROBE4(ipif__ioctl,
12725 			    char *, "ip_process_ioctl finish RD",
12726 			    int, ipip->ipi_cmd, ill_t *, ci.ci_ipif->ipif_ill,
12727 			    ipif_t *, ci.ci_ipif);
12728 			ipif_refrele(ci.ci_ipif);
12729 		} else {
12730 			DTRACE_PROBE4(ipif__ioctl,
12731 			    char *, "ip_process_ioctl finish RD",
12732 			    int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12733 		}
12734 		ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12735 		return;
12736 	}
12737 
12738 	ASSERT(ci.ci_ipif != NULL);
12739 
12740 	/*
12741 	 * If ipsq is non-NULL, we are already being called exclusively
12742 	 */
12743 	ASSERT(ipsq == NULL || IAM_WRITER_IPSQ(ipsq));
12744 	if (ipsq == NULL) {
12745 		ipsq = ipsq_try_enter(ci.ci_ipif, NULL, q, mp, ip_process_ioctl,
12746 		    NEW_OP, B_TRUE);
12747 		if (ipsq == NULL) {
12748 			ipif_refrele(ci.ci_ipif);
12749 			return;
12750 		}
12751 		entered_ipsq = B_TRUE;
12752 	}
12753 	/*
12754 	 * Release the ipif so that ipif_down and friends that wait for
12755 	 * references to go away are not misled about the current ipif_refcnt
12756 	 * values. We are writer so we can access the ipif even after releasing
12757 	 * the ipif.
12758 	 */
12759 	ipif_refrele(ci.ci_ipif);
12760 
12761 	ipsq_current_start(ipsq, ci.ci_ipif, ipip->ipi_cmd);
12762 
12763 	/*
12764 	 * A return value of EINPROGRESS means the ioctl is
12765 	 * either queued and waiting for some reason or has
12766 	 * already completed.
12767 	 */
12768 	err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip, ci.ci_lifr);
12769 
12770 	DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish WR",
12771 	    int, ipip->ipi_cmd,
12772 	    ill_t *, ci.ci_ipif == NULL ? NULL : ci.ci_ipif->ipif_ill,
12773 	    ipif_t *, ci.ci_ipif);
12774 	ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12775 
12776 	if (entered_ipsq)
12777 		ipsq_exit(ipsq);
12778 }
12779 
12780 /*
12781  * Complete the ioctl. Typically ioctls use the mi package and need to
12782  * do mi_copyout/mi_copy_done.
12783  */
12784 void
ip_ioctl_finish(queue_t * q,mblk_t * mp,int err,int mode,ipsq_t * ipsq)12785 ip_ioctl_finish(queue_t *q, mblk_t *mp, int err, int mode, ipsq_t *ipsq)
12786 {
12787 	conn_t	*connp = NULL;
12788 
12789 	if (err == EINPROGRESS)
12790 		return;
12791 
12792 	if (CONN_Q(q)) {
12793 		connp = Q_TO_CONN(q);
12794 		ASSERT(connp->conn_ref >= 2);
12795 	}
12796 
12797 	switch (mode) {
12798 	case COPYOUT:
12799 		if (err == 0)
12800 			mi_copyout(q, mp);
12801 		else
12802 			mi_copy_done(q, mp, err);
12803 		break;
12804 
12805 	case NO_COPYOUT:
12806 		mi_copy_done(q, mp, err);
12807 		break;
12808 
12809 	default:
12810 		ASSERT(mode == CONN_CLOSE);	/* aborted through CONN_CLOSE */
12811 		break;
12812 	}
12813 
12814 	/*
12815 	 * The conn refhold and ioctlref placed on the conn at the start of the
12816 	 * ioctl are released here.
12817 	 */
12818 	if (connp != NULL) {
12819 		CONN_DEC_IOCTLREF(connp);
12820 		CONN_OPER_PENDING_DONE(connp);
12821 	}
12822 
12823 	if (ipsq != NULL)
12824 		ipsq_current_finish(ipsq);
12825 }
12826 
12827 /* Handles all non data messages */
12828 int
ip_wput_nondata(queue_t * q,mblk_t * mp)12829 ip_wput_nondata(queue_t *q, mblk_t *mp)
12830 {
12831 	mblk_t		*mp1;
12832 	struct iocblk	*iocp;
12833 	ip_ioctl_cmd_t	*ipip;
12834 	conn_t		*connp;
12835 	cred_t		*cr;
12836 	char		*proto_str;
12837 
12838 	if (CONN_Q(q))
12839 		connp = Q_TO_CONN(q);
12840 	else
12841 		connp = NULL;
12842 
12843 	iocp = NULL;
12844 	switch (DB_TYPE(mp)) {
12845 	case M_IOCTL:
12846 		/*
12847 		 * IOCTL processing begins in ip_sioctl_copyin_setup which
12848 		 * will arrange to copy in associated control structures.
12849 		 */
12850 		ip_sioctl_copyin_setup(q, mp);
12851 		return (0);
12852 	case M_IOCDATA:
12853 		/*
12854 		 * Ensure that this is associated with one of our trans-
12855 		 * parent ioctls.  If it's not ours, discard it if we're
12856 		 * running as a driver, or pass it on if we're a module.
12857 		 */
12858 		iocp = (struct iocblk *)mp->b_rptr;
12859 		ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12860 		if (ipip == NULL) {
12861 			if (q->q_next == NULL) {
12862 				goto nak;
12863 			} else {
12864 				putnext(q, mp);
12865 			}
12866 			return (0);
12867 		}
12868 		if ((q->q_next != NULL) && !(ipip->ipi_flags & IPI_MODOK)) {
12869 			/*
12870 			 * The ioctl is one we recognise, but is not consumed
12871 			 * by IP as a module and we are a module, so we drop
12872 			 */
12873 			goto nak;
12874 		}
12875 
12876 		/* IOCTL continuation following copyin or copyout. */
12877 		if (mi_copy_state(q, mp, NULL) == -1) {
12878 			/*
12879 			 * The copy operation failed.  mi_copy_state already
12880 			 * cleaned up, so we're out of here.
12881 			 */
12882 			return (0);
12883 		}
12884 		/*
12885 		 * If we just completed a copy in, we become writer and
12886 		 * continue processing in ip_sioctl_copyin_done.  If it
12887 		 * was a copy out, we call mi_copyout again.  If there is
12888 		 * nothing more to copy out, it will complete the IOCTL.
12889 		 */
12890 		if (MI_COPY_DIRECTION(mp) == MI_COPY_IN) {
12891 			if (!(mp1 = mp->b_cont) || !(mp1 = mp1->b_cont)) {
12892 				mi_copy_done(q, mp, EPROTO);
12893 				return (0);
12894 			}
12895 			/*
12896 			 * Check for cases that need more copying.  A return
12897 			 * value of 0 means a second copyin has been started,
12898 			 * so we return; a return value of 1 means no more
12899 			 * copying is needed, so we continue.
12900 			 */
12901 			if (ipip->ipi_cmd_type == MSFILT_CMD &&
12902 			    MI_COPY_COUNT(mp) == 1) {
12903 				if (ip_copyin_msfilter(q, mp) == 0)
12904 					return (0);
12905 			}
12906 			/*
12907 			 * Refhold the conn, till the ioctl completes. This is
12908 			 * needed in case the ioctl ends up in the pending mp
12909 			 * list. Every mp in the ipx_pending_mp list must have
12910 			 * a refhold on the conn to resume processing. The
12911 			 * refhold is released when the ioctl completes
12912 			 * (whether normally or abnormally). An ioctlref is also
12913 			 * placed on the conn to prevent TCP from removing the
12914 			 * queue needed to send the ioctl reply back.
12915 			 * In all cases ip_ioctl_finish is called to finish
12916 			 * the ioctl and release the refholds.
12917 			 */
12918 			if (connp != NULL) {
12919 				/* This is not a reentry */
12920 				CONN_INC_REF(connp);
12921 				CONN_INC_IOCTLREF(connp);
12922 			} else {
12923 				if (!(ipip->ipi_flags & IPI_MODOK)) {
12924 					mi_copy_done(q, mp, EINVAL);
12925 					return (0);
12926 				}
12927 			}
12928 
12929 			ip_process_ioctl(NULL, q, mp, ipip);
12930 
12931 		} else {
12932 			mi_copyout(q, mp);
12933 		}
12934 		return (0);
12935 
12936 	case M_IOCNAK:
12937 		/*
12938 		 * The only way we could get here is if a resolver didn't like
12939 		 * an IOCTL we sent it.	 This shouldn't happen.
12940 		 */
12941 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
12942 		    "ip_wput_nondata: unexpected M_IOCNAK, ioc_cmd 0x%x",
12943 		    ((struct iocblk *)mp->b_rptr)->ioc_cmd);
12944 		freemsg(mp);
12945 		return (0);
12946 	case M_IOCACK:
12947 		/* /dev/ip shouldn't see this */
12948 		goto nak;
12949 	case M_FLUSH:
12950 		if (*mp->b_rptr & FLUSHW)
12951 			flushq(q, FLUSHALL);
12952 		if (q->q_next) {
12953 			putnext(q, mp);
12954 			return (0);
12955 		}
12956 		if (*mp->b_rptr & FLUSHR) {
12957 			*mp->b_rptr &= ~FLUSHW;
12958 			qreply(q, mp);
12959 			return (0);
12960 		}
12961 		freemsg(mp);
12962 		return (0);
12963 	case M_CTL:
12964 		break;
12965 	case M_PROTO:
12966 	case M_PCPROTO:
12967 		/*
12968 		 * The only PROTO messages we expect are SNMP-related.
12969 		 */
12970 		switch (((union T_primitives *)mp->b_rptr)->type) {
12971 		case T_SVR4_OPTMGMT_REQ:
12972 			ip2dbg(("ip_wput_nondata: T_SVR4_OPTMGMT_REQ "
12973 			    "flags %x\n",
12974 			    ((struct T_optmgmt_req *)mp->b_rptr)->MGMT_flags));
12975 
12976 			if (connp == NULL) {
12977 				proto_str = "T_SVR4_OPTMGMT_REQ";
12978 				goto protonak;
12979 			}
12980 
12981 			/*
12982 			 * All Solaris components should pass a db_credp
12983 			 * for this TPI message, hence we ASSERT.
12984 			 * But in case there is some other M_PROTO that looks
12985 			 * like a TPI message sent by some other kernel
12986 			 * component, we check and return an error.
12987 			 */
12988 			cr = msg_getcred(mp, NULL);
12989 			ASSERT(cr != NULL);
12990 			if (cr == NULL) {
12991 				mp = mi_tpi_err_ack_alloc(mp, TSYSERR, EINVAL);
12992 				if (mp != NULL)
12993 					qreply(q, mp);
12994 				return (0);
12995 			}
12996 
12997 			if (!snmpcom_req(q, mp, ip_snmp_set, ip_snmp_get, cr)) {
12998 				proto_str = "Bad SNMPCOM request?";
12999 				goto protonak;
13000 			}
13001 			return (0);
13002 		default:
13003 			ip1dbg(("ip_wput_nondata: dropping M_PROTO prim %u\n",
13004 			    (int)*(uint_t *)mp->b_rptr));
13005 			freemsg(mp);
13006 			return (0);
13007 		}
13008 	default:
13009 		break;
13010 	}
13011 	if (q->q_next) {
13012 		putnext(q, mp);
13013 	} else
13014 		freemsg(mp);
13015 	return (0);
13016 
13017 nak:
13018 	iocp->ioc_error = EINVAL;
13019 	mp->b_datap->db_type = M_IOCNAK;
13020 	iocp->ioc_count = 0;
13021 	qreply(q, mp);
13022 	return (0);
13023 
13024 protonak:
13025 	cmn_err(CE_NOTE, "IP doesn't process %s as a module", proto_str);
13026 	if ((mp = mi_tpi_err_ack_alloc(mp, TPROTO, EINVAL)) != NULL)
13027 		qreply(q, mp);
13028 	return (0);
13029 }
13030 
13031 /*
13032  * Process IP options in an outbound packet.  Verify that the nexthop in a
13033  * strict source route is onlink.
13034  * Returns non-zero if something fails in which case an ICMP error has been
13035  * sent and mp freed.
13036  *
13037  * Assumes the ULP has called ip_massage_options to move nexthop into ipha_dst.
13038  */
13039 int
ip_output_options(mblk_t * mp,ipha_t * ipha,ip_xmit_attr_t * ixa,ill_t * ill)13040 ip_output_options(mblk_t *mp, ipha_t *ipha, ip_xmit_attr_t *ixa, ill_t *ill)
13041 {
13042 	ipoptp_t	opts;
13043 	uchar_t		*opt;
13044 	uint8_t		optval;
13045 	uint8_t		optlen;
13046 	ipaddr_t	dst;
13047 	intptr_t	code = 0;
13048 	ire_t		*ire;
13049 	ip_stack_t	*ipst = ixa->ixa_ipst;
13050 	ip_recv_attr_t	iras;
13051 
13052 	ip2dbg(("ip_output_options\n"));
13053 
13054 	opt = NULL;
13055 	dst = ipha->ipha_dst;
13056 	for (optval = ipoptp_first(&opts, ipha);
13057 	    optval != IPOPT_EOL;
13058 	    optval = ipoptp_next(&opts)) {
13059 		opt = opts.ipoptp_cur;
13060 		optlen = opts.ipoptp_len;
13061 		ip2dbg(("ip_output_options: opt %d, len %d\n",
13062 		    optval, optlen));
13063 		switch (optval) {
13064 			uint32_t off;
13065 		case IPOPT_SSRR:
13066 		case IPOPT_LSRR:
13067 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13068 				ip1dbg((
13069 				    "ip_output_options: bad option offset\n"));
13070 				code = (char *)&opt[IPOPT_OLEN] -
13071 				    (char *)ipha;
13072 				goto param_prob;
13073 			}
13074 			off = opt[IPOPT_OFFSET];
13075 			ip1dbg(("ip_output_options: next hop 0x%x\n",
13076 			    ntohl(dst)));
13077 			/*
13078 			 * For strict: verify that dst is directly
13079 			 * reachable.
13080 			 */
13081 			if (optval == IPOPT_SSRR) {
13082 				ire = ire_ftable_lookup_v4(dst, 0, 0,
13083 				    IRE_INTERFACE, NULL, ALL_ZONES,
13084 				    ixa->ixa_tsl,
13085 				    MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
13086 				    NULL);
13087 				if (ire == NULL) {
13088 					ip1dbg(("ip_output_options: SSRR not"
13089 					    " directly reachable: 0x%x\n",
13090 					    ntohl(dst)));
13091 					goto bad_src_route;
13092 				}
13093 				ire_refrele(ire);
13094 			}
13095 			break;
13096 		case IPOPT_RR:
13097 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13098 				ip1dbg((
13099 				    "ip_output_options: bad option offset\n"));
13100 				code = (char *)&opt[IPOPT_OLEN] -
13101 				    (char *)ipha;
13102 				goto param_prob;
13103 			}
13104 			break;
13105 		case IPOPT_TS:
13106 			/*
13107 			 * Verify that length >=5 and that there is either
13108 			 * room for another timestamp or that the overflow
13109 			 * counter is not maxed out.
13110 			 */
13111 			code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
13112 			if (optlen < IPOPT_MINLEN_IT) {
13113 				goto param_prob;
13114 			}
13115 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13116 				ip1dbg((
13117 				    "ip_output_options: bad option offset\n"));
13118 				code = (char *)&opt[IPOPT_OFFSET] -
13119 				    (char *)ipha;
13120 				goto param_prob;
13121 			}
13122 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
13123 			case IPOPT_TS_TSONLY:
13124 				off = IPOPT_TS_TIMELEN;
13125 				break;
13126 			case IPOPT_TS_TSANDADDR:
13127 			case IPOPT_TS_PRESPEC:
13128 			case IPOPT_TS_PRESPEC_RFC791:
13129 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
13130 				break;
13131 			default:
13132 				code = (char *)&opt[IPOPT_POS_OV_FLG] -
13133 				    (char *)ipha;
13134 				goto param_prob;
13135 			}
13136 			if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
13137 			    (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
13138 				/*
13139 				 * No room and the overflow counter is 15
13140 				 * already.
13141 				 */
13142 				goto param_prob;
13143 			}
13144 			break;
13145 		}
13146 	}
13147 
13148 	if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0)
13149 		return (0);
13150 
13151 	ip1dbg(("ip_output_options: error processing IP options."));
13152 	code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
13153 
13154 param_prob:
13155 	bzero(&iras, sizeof (iras));
13156 	iras.ira_ill = iras.ira_rill = ill;
13157 	iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13158 	iras.ira_rifindex = iras.ira_ruifindex;
13159 	iras.ira_flags = IRAF_IS_IPV4;
13160 
13161 	ip_drop_output("ip_output_options", mp, ill);
13162 	icmp_param_problem(mp, (uint8_t)code, &iras);
13163 	ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13164 	return (-1);
13165 
13166 bad_src_route:
13167 	bzero(&iras, sizeof (iras));
13168 	iras.ira_ill = iras.ira_rill = ill;
13169 	iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13170 	iras.ira_rifindex = iras.ira_ruifindex;
13171 	iras.ira_flags = IRAF_IS_IPV4;
13172 
13173 	ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
13174 	icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, &iras);
13175 	ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13176 	return (-1);
13177 }
13178 
13179 /*
13180  * The maximum value of conn_drain_list_cnt is CONN_MAXDRAINCNT.
13181  * conn_drain_list_cnt can be changed by setting conn_drain_nthreads
13182  * thru /etc/system.
13183  */
13184 #define	CONN_MAXDRAINCNT	64
13185 
13186 static void
conn_drain_init(ip_stack_t * ipst)13187 conn_drain_init(ip_stack_t *ipst)
13188 {
13189 	int i, j;
13190 	idl_tx_list_t *itl_tx;
13191 
13192 	ipst->ips_conn_drain_list_cnt = conn_drain_nthreads;
13193 
13194 	if ((ipst->ips_conn_drain_list_cnt == 0) ||
13195 	    (ipst->ips_conn_drain_list_cnt > CONN_MAXDRAINCNT)) {
13196 		/*
13197 		 * Default value of the number of drainers is the
13198 		 * number of cpus, subject to maximum of 8 drainers.
13199 		 */
13200 		if (boot_max_ncpus != -1)
13201 			ipst->ips_conn_drain_list_cnt = MIN(boot_max_ncpus, 8);
13202 		else
13203 			ipst->ips_conn_drain_list_cnt = MIN(max_ncpus, 8);
13204 	}
13205 
13206 	ipst->ips_idl_tx_list =
13207 	    kmem_zalloc(TX_FANOUT_SIZE * sizeof (idl_tx_list_t), KM_SLEEP);
13208 	for (i = 0; i < TX_FANOUT_SIZE; i++) {
13209 		itl_tx =  &ipst->ips_idl_tx_list[i];
13210 		itl_tx->txl_drain_list =
13211 		    kmem_zalloc(ipst->ips_conn_drain_list_cnt *
13212 		    sizeof (idl_t), KM_SLEEP);
13213 		mutex_init(&itl_tx->txl_lock, NULL, MUTEX_DEFAULT, NULL);
13214 		for (j = 0; j < ipst->ips_conn_drain_list_cnt; j++) {
13215 			mutex_init(&itl_tx->txl_drain_list[j].idl_lock, NULL,
13216 			    MUTEX_DEFAULT, NULL);
13217 			itl_tx->txl_drain_list[j].idl_itl = itl_tx;
13218 		}
13219 	}
13220 }
13221 
13222 static void
conn_drain_fini(ip_stack_t * ipst)13223 conn_drain_fini(ip_stack_t *ipst)
13224 {
13225 	int i;
13226 	idl_tx_list_t *itl_tx;
13227 
13228 	for (i = 0; i < TX_FANOUT_SIZE; i++) {
13229 		itl_tx =  &ipst->ips_idl_tx_list[i];
13230 		kmem_free(itl_tx->txl_drain_list,
13231 		    ipst->ips_conn_drain_list_cnt * sizeof (idl_t));
13232 	}
13233 	kmem_free(ipst->ips_idl_tx_list,
13234 	    TX_FANOUT_SIZE * sizeof (idl_tx_list_t));
13235 	ipst->ips_idl_tx_list = NULL;
13236 }
13237 
13238 /*
13239  * Flow control has blocked us from proceeding.  Insert the given conn in one
13240  * of the conn drain lists.  When flow control is unblocked, either ip_wsrv()
13241  * (STREAMS) or ill_flow_enable() (direct) will be called back, which in turn
13242  * will call conn_walk_drain().  See the flow control notes at the top of this
13243  * file for more details.
13244  */
13245 void
conn_drain_insert(conn_t * connp,idl_tx_list_t * tx_list)13246 conn_drain_insert(conn_t *connp, idl_tx_list_t *tx_list)
13247 {
13248 	idl_t	*idl = tx_list->txl_drain_list;
13249 	uint_t	index;
13250 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
13251 
13252 	mutex_enter(&connp->conn_lock);
13253 	if (connp->conn_state_flags & CONN_CLOSING) {
13254 		/*
13255 		 * The conn is closing as a result of which CONN_CLOSING
13256 		 * is set. Return.
13257 		 */
13258 		mutex_exit(&connp->conn_lock);
13259 		return;
13260 	} else if (connp->conn_idl == NULL) {
13261 		/*
13262 		 * Assign the next drain list round robin. We dont' use
13263 		 * a lock, and thus it may not be strictly round robin.
13264 		 * Atomicity of load/stores is enough to make sure that
13265 		 * conn_drain_list_index is always within bounds.
13266 		 */
13267 		index = tx_list->txl_drain_index;
13268 		ASSERT(index < ipst->ips_conn_drain_list_cnt);
13269 		connp->conn_idl = &tx_list->txl_drain_list[index];
13270 		index++;
13271 		if (index == ipst->ips_conn_drain_list_cnt)
13272 			index = 0;
13273 		tx_list->txl_drain_index = index;
13274 	} else {
13275 		ASSERT(connp->conn_idl->idl_itl == tx_list);
13276 	}
13277 	mutex_exit(&connp->conn_lock);
13278 
13279 	idl = connp->conn_idl;
13280 	mutex_enter(&idl->idl_lock);
13281 	if ((connp->conn_drain_prev != NULL) ||
13282 	    (connp->conn_state_flags & CONN_CLOSING)) {
13283 		/*
13284 		 * The conn is either already in the drain list or closing.
13285 		 * (We needed to check for CONN_CLOSING again since close can
13286 		 * sneak in between dropping conn_lock and acquiring idl_lock.)
13287 		 */
13288 		mutex_exit(&idl->idl_lock);
13289 		return;
13290 	}
13291 
13292 	/*
13293 	 * The conn is not in the drain list. Insert it at the
13294 	 * tail of the drain list. The drain list is circular
13295 	 * and doubly linked. idl_conn points to the 1st element
13296 	 * in the list.
13297 	 */
13298 	if (idl->idl_conn == NULL) {
13299 		idl->idl_conn = connp;
13300 		connp->conn_drain_next = connp;
13301 		connp->conn_drain_prev = connp;
13302 	} else {
13303 		conn_t *head = idl->idl_conn;
13304 
13305 		connp->conn_drain_next = head;
13306 		connp->conn_drain_prev = head->conn_drain_prev;
13307 		head->conn_drain_prev->conn_drain_next = connp;
13308 		head->conn_drain_prev = connp;
13309 	}
13310 	/*
13311 	 * For non streams based sockets assert flow control.
13312 	 */
13313 	conn_setqfull(connp, NULL);
13314 	mutex_exit(&idl->idl_lock);
13315 }
13316 
13317 static void
conn_drain_remove(conn_t * connp)13318 conn_drain_remove(conn_t *connp)
13319 {
13320 	idl_t *idl = connp->conn_idl;
13321 
13322 	if (idl != NULL) {
13323 		/*
13324 		 * Remove ourself from the drain list.
13325 		 */
13326 		if (connp->conn_drain_next == connp) {
13327 			/* Singleton in the list */
13328 			ASSERT(connp->conn_drain_prev == connp);
13329 			idl->idl_conn = NULL;
13330 		} else {
13331 			connp->conn_drain_prev->conn_drain_next =
13332 			    connp->conn_drain_next;
13333 			connp->conn_drain_next->conn_drain_prev =
13334 			    connp->conn_drain_prev;
13335 			if (idl->idl_conn == connp)
13336 				idl->idl_conn = connp->conn_drain_next;
13337 		}
13338 
13339 		/*
13340 		 * NOTE: because conn_idl is associated with a specific drain
13341 		 * list which in turn is tied to the index the TX ring
13342 		 * (txl_cookie) hashes to, and because the TX ring can change
13343 		 * over the lifetime of the conn_t, we must clear conn_idl so
13344 		 * a subsequent conn_drain_insert() will set conn_idl again
13345 		 * based on the latest txl_cookie.
13346 		 */
13347 		connp->conn_idl = NULL;
13348 	}
13349 	connp->conn_drain_next = NULL;
13350 	connp->conn_drain_prev = NULL;
13351 
13352 	conn_clrqfull(connp, NULL);
13353 	/*
13354 	 * For streams based sockets open up flow control.
13355 	 */
13356 	if (!IPCL_IS_NONSTR(connp))
13357 		enableok(connp->conn_wq);
13358 }
13359 
13360 /*
13361  * This conn is closing, and we are called from ip_close. OR
13362  * this conn is draining because flow-control on the ill has been relieved.
13363  *
13364  * We must also need to remove conn's on this idl from the list, and also
13365  * inform the sockfs upcalls about the change in flow-control.
13366  */
13367 static void
conn_drain(conn_t * connp,boolean_t closing)13368 conn_drain(conn_t *connp, boolean_t closing)
13369 {
13370 	idl_t *idl;
13371 	conn_t *next_connp;
13372 
13373 	/*
13374 	 * connp->conn_idl is stable at this point, and no lock is needed
13375 	 * to check it. If we are called from ip_close, close has already
13376 	 * set CONN_CLOSING, thus freezing the value of conn_idl, and
13377 	 * called us only because conn_idl is non-null. If we are called thru
13378 	 * service, conn_idl could be null, but it cannot change because
13379 	 * service is single-threaded per queue, and there cannot be another
13380 	 * instance of service trying to call conn_drain_insert on this conn
13381 	 * now.
13382 	 */
13383 	ASSERT(!closing || connp == NULL || connp->conn_idl != NULL);
13384 
13385 	/*
13386 	 * If the conn doesn't exist or is not on a drain list, bail.
13387 	 */
13388 	if (connp == NULL || connp->conn_idl == NULL ||
13389 	    connp->conn_drain_prev == NULL) {
13390 		return;
13391 	}
13392 
13393 	idl = connp->conn_idl;
13394 	ASSERT(MUTEX_HELD(&idl->idl_lock));
13395 
13396 	if (!closing) {
13397 		next_connp = connp->conn_drain_next;
13398 		while (next_connp != connp) {
13399 			conn_t *delconnp = next_connp;
13400 
13401 			next_connp = next_connp->conn_drain_next;
13402 			conn_drain_remove(delconnp);
13403 		}
13404 		ASSERT(connp->conn_drain_next == idl->idl_conn);
13405 	}
13406 	conn_drain_remove(connp);
13407 }
13408 
13409 /*
13410  * Write service routine. Shared perimeter entry point.
13411  * The device queue's messages has fallen below the low water mark and STREAMS
13412  * has backenabled the ill_wq. Send sockfs notification about flow-control on
13413  * each waiting conn.
13414  */
13415 int
ip_wsrv(queue_t * q)13416 ip_wsrv(queue_t *q)
13417 {
13418 	ill_t	*ill;
13419 
13420 	ill = (ill_t *)q->q_ptr;
13421 	if (ill->ill_state_flags == 0) {
13422 		ip_stack_t *ipst = ill->ill_ipst;
13423 
13424 		/*
13425 		 * The device flow control has opened up.
13426 		 * Walk through conn drain lists and qenable the
13427 		 * first conn in each list. This makes sense only
13428 		 * if the stream is fully plumbed and setup.
13429 		 * Hence the ill_state_flags check above.
13430 		 */
13431 		ip1dbg(("ip_wsrv: walking\n"));
13432 		conn_walk_drain(ipst, &ipst->ips_idl_tx_list[0]);
13433 		enableok(ill->ill_wq);
13434 	}
13435 	return (0);
13436 }
13437 
13438 /*
13439  * Callback to disable flow control in IP.
13440  *
13441  * This is a mac client callback added when the DLD_CAPAB_DIRECT capability
13442  * is enabled.
13443  *
13444  * When MAC_TX() is not able to send any more packets, dld sets its queue
13445  * to QFULL and enable the STREAMS flow control. Later, when the underlying
13446  * driver is able to continue to send packets, it calls mac_tx_(ring_)update()
13447  * function and wakes up corresponding mac worker threads, which in turn
13448  * calls this callback function, and disables flow control.
13449  */
13450 void
ill_flow_enable(void * arg,ip_mac_tx_cookie_t cookie)13451 ill_flow_enable(void *arg, ip_mac_tx_cookie_t cookie)
13452 {
13453 	ill_t *ill = (ill_t *)arg;
13454 	ip_stack_t *ipst = ill->ill_ipst;
13455 	idl_tx_list_t *idl_txl;
13456 
13457 	idl_txl = &ipst->ips_idl_tx_list[IDLHASHINDEX(cookie)];
13458 	mutex_enter(&idl_txl->txl_lock);
13459 	/* add code to to set a flag to indicate idl_txl is enabled */
13460 	conn_walk_drain(ipst, idl_txl);
13461 	mutex_exit(&idl_txl->txl_lock);
13462 }
13463 
13464 /*
13465  * Flow control has been relieved and STREAMS has backenabled us; drain
13466  * all the conn lists on `tx_list'.
13467  */
13468 static void
conn_walk_drain(ip_stack_t * ipst,idl_tx_list_t * tx_list)13469 conn_walk_drain(ip_stack_t *ipst, idl_tx_list_t *tx_list)
13470 {
13471 	int i;
13472 	idl_t *idl;
13473 
13474 	IP_STAT(ipst, ip_conn_walk_drain);
13475 
13476 	for (i = 0; i < ipst->ips_conn_drain_list_cnt; i++) {
13477 		idl = &tx_list->txl_drain_list[i];
13478 		mutex_enter(&idl->idl_lock);
13479 		conn_drain(idl->idl_conn, B_FALSE);
13480 		mutex_exit(&idl->idl_lock);
13481 	}
13482 }
13483 
13484 /*
13485  * Determine if the ill and multicast aspects of that packets
13486  * "matches" the conn.
13487  */
13488 boolean_t
conn_wantpacket(conn_t * connp,ip_recv_attr_t * ira,ipha_t * ipha)13489 conn_wantpacket(conn_t *connp, ip_recv_attr_t *ira, ipha_t *ipha)
13490 {
13491 	ill_t		*ill = ira->ira_rill;
13492 	zoneid_t	zoneid = ira->ira_zoneid;
13493 	uint_t		in_ifindex;
13494 	ipaddr_t	dst, src;
13495 
13496 	dst = ipha->ipha_dst;
13497 	src = ipha->ipha_src;
13498 
13499 	/*
13500 	 * conn_incoming_ifindex is set by IP_BOUND_IF which limits
13501 	 * unicast, broadcast and multicast reception to
13502 	 * conn_incoming_ifindex.
13503 	 * conn_wantpacket is called for unicast, broadcast and
13504 	 * multicast packets.
13505 	 */
13506 	in_ifindex = connp->conn_incoming_ifindex;
13507 
13508 	/* mpathd can bind to the under IPMP interface, which we allow */
13509 	if (in_ifindex != 0 && in_ifindex != ill->ill_phyint->phyint_ifindex) {
13510 		if (!IS_UNDER_IPMP(ill))
13511 			return (B_FALSE);
13512 
13513 		if (in_ifindex != ipmp_ill_get_ipmp_ifindex(ill))
13514 			return (B_FALSE);
13515 	}
13516 
13517 	if (!IPCL_ZONE_MATCH(connp, zoneid))
13518 		return (B_FALSE);
13519 
13520 	if (!(ira->ira_flags & IRAF_MULTICAST))
13521 		return (B_TRUE);
13522 
13523 	if (connp->conn_multi_router) {
13524 		/* multicast packet and multicast router socket: send up */
13525 		return (B_TRUE);
13526 	}
13527 
13528 	if (ipha->ipha_protocol == IPPROTO_PIM ||
13529 	    ipha->ipha_protocol == IPPROTO_RSVP)
13530 		return (B_TRUE);
13531 
13532 	return (conn_hasmembers_ill_withsrc_v4(connp, dst, src, ira->ira_ill));
13533 }
13534 
13535 void
conn_setqfull(conn_t * connp,boolean_t * flow_stopped)13536 conn_setqfull(conn_t *connp, boolean_t *flow_stopped)
13537 {
13538 	if (IPCL_IS_NONSTR(connp)) {
13539 		(*connp->conn_upcalls->su_txq_full)
13540 		    (connp->conn_upper_handle, B_TRUE);
13541 		if (flow_stopped != NULL)
13542 			*flow_stopped = B_TRUE;
13543 	} else {
13544 		queue_t *q = connp->conn_wq;
13545 
13546 		ASSERT(q != NULL);
13547 		if (!(q->q_flag & QFULL)) {
13548 			mutex_enter(QLOCK(q));
13549 			if (!(q->q_flag & QFULL)) {
13550 				/* still need to set QFULL */
13551 				q->q_flag |= QFULL;
13552 				/* set flow_stopped to true under QLOCK */
13553 				if (flow_stopped != NULL)
13554 					*flow_stopped = B_TRUE;
13555 				mutex_exit(QLOCK(q));
13556 			} else {
13557 				/* flow_stopped is left unchanged */
13558 				mutex_exit(QLOCK(q));
13559 			}
13560 		}
13561 	}
13562 }
13563 
13564 void
conn_clrqfull(conn_t * connp,boolean_t * flow_stopped)13565 conn_clrqfull(conn_t *connp, boolean_t *flow_stopped)
13566 {
13567 	if (IPCL_IS_NONSTR(connp)) {
13568 		(*connp->conn_upcalls->su_txq_full)
13569 		    (connp->conn_upper_handle, B_FALSE);
13570 		if (flow_stopped != NULL)
13571 			*flow_stopped = B_FALSE;
13572 	} else {
13573 		queue_t *q = connp->conn_wq;
13574 
13575 		ASSERT(q != NULL);
13576 		if (q->q_flag & QFULL) {
13577 			mutex_enter(QLOCK(q));
13578 			if (q->q_flag & QFULL) {
13579 				q->q_flag &= ~QFULL;
13580 				/* set flow_stopped to false under QLOCK */
13581 				if (flow_stopped != NULL)
13582 					*flow_stopped = B_FALSE;
13583 				mutex_exit(QLOCK(q));
13584 				if (q->q_flag & QWANTW)
13585 					qbackenable(q, 0);
13586 			} else {
13587 				/* flow_stopped is left unchanged */
13588 				mutex_exit(QLOCK(q));
13589 			}
13590 		}
13591 	}
13592 
13593 	mutex_enter(&connp->conn_lock);
13594 	connp->conn_blocked = B_FALSE;
13595 	mutex_exit(&connp->conn_lock);
13596 }
13597 
13598 /*
13599  * Return the length in bytes of the IPv4 headers (base header, label, and
13600  * other IP options) that will be needed based on the
13601  * ip_pkt_t structure passed by the caller.
13602  *
13603  * The returned length does not include the length of the upper level
13604  * protocol (ULP) header.
13605  * The caller needs to check that the length doesn't exceed the max for IPv4.
13606  */
13607 int
ip_total_hdrs_len_v4(const ip_pkt_t * ipp)13608 ip_total_hdrs_len_v4(const ip_pkt_t *ipp)
13609 {
13610 	int len;
13611 
13612 	len = IP_SIMPLE_HDR_LENGTH;
13613 	if (ipp->ipp_fields & IPPF_LABEL_V4) {
13614 		ASSERT(ipp->ipp_label_len_v4 != 0);
13615 		/* We need to round up here */
13616 		len += (ipp->ipp_label_len_v4 + 3) & ~3;
13617 	}
13618 
13619 	if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13620 		ASSERT(ipp->ipp_ipv4_options_len != 0);
13621 		ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13622 		len += ipp->ipp_ipv4_options_len;
13623 	}
13624 	return (len);
13625 }
13626 
13627 /*
13628  * All-purpose routine to build an IPv4 header with options based
13629  * on the abstract ip_pkt_t.
13630  *
13631  * The caller has to set the source and destination address as well as
13632  * ipha_length. The caller has to massage any source route and compensate
13633  * for the ULP pseudo-header checksum due to the source route.
13634  */
13635 void
ip_build_hdrs_v4(uchar_t * buf,uint_t buf_len,const ip_pkt_t * ipp,uint8_t protocol)13636 ip_build_hdrs_v4(uchar_t *buf, uint_t buf_len, const ip_pkt_t *ipp,
13637     uint8_t protocol)
13638 {
13639 	ipha_t	*ipha = (ipha_t *)buf;
13640 	uint8_t *cp;
13641 
13642 	/* Initialize IPv4 header */
13643 	ipha->ipha_type_of_service = ipp->ipp_type_of_service;
13644 	ipha->ipha_length = 0;	/* Caller will set later */
13645 	ipha->ipha_ident = 0;
13646 	ipha->ipha_fragment_offset_and_flags = 0;
13647 	ipha->ipha_ttl = ipp->ipp_unicast_hops;
13648 	ipha->ipha_protocol = protocol;
13649 	ipha->ipha_hdr_checksum = 0;
13650 
13651 	if ((ipp->ipp_fields & IPPF_ADDR) &&
13652 	    IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
13653 		ipha->ipha_src = ipp->ipp_addr_v4;
13654 
13655 	cp = (uint8_t *)&ipha[1];
13656 	if (ipp->ipp_fields & IPPF_LABEL_V4) {
13657 		ASSERT(ipp->ipp_label_len_v4 != 0);
13658 		bcopy(ipp->ipp_label_v4, cp, ipp->ipp_label_len_v4);
13659 		cp += ipp->ipp_label_len_v4;
13660 		/* We need to round up here */
13661 		while ((uintptr_t)cp & 0x3) {
13662 			*cp++ = IPOPT_NOP;
13663 		}
13664 	}
13665 
13666 	if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13667 		ASSERT(ipp->ipp_ipv4_options_len != 0);
13668 		ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13669 		bcopy(ipp->ipp_ipv4_options, cp, ipp->ipp_ipv4_options_len);
13670 		cp += ipp->ipp_ipv4_options_len;
13671 	}
13672 	ipha->ipha_version_and_hdr_length =
13673 	    (uint8_t)((IP_VERSION << 4) + buf_len / 4);
13674 
13675 	ASSERT((int)(cp - buf) == buf_len);
13676 }
13677 
13678 /* Allocate the private structure */
13679 static int
ip_priv_alloc(void ** bufp)13680 ip_priv_alloc(void **bufp)
13681 {
13682 	void	*buf;
13683 
13684 	if ((buf = kmem_alloc(sizeof (ip_priv_t), KM_NOSLEEP)) == NULL)
13685 		return (ENOMEM);
13686 
13687 	*bufp = buf;
13688 	return (0);
13689 }
13690 
13691 /* Function to delete the private structure */
13692 void
ip_priv_free(void * buf)13693 ip_priv_free(void *buf)
13694 {
13695 	ASSERT(buf != NULL);
13696 	kmem_free(buf, sizeof (ip_priv_t));
13697 }
13698 
13699 /*
13700  * The entry point for IPPF processing.
13701  * If the classifier (IPGPC_CLASSIFY) is not loaded and configured, the
13702  * routine just returns.
13703  *
13704  * When called, ip_process generates an ipp_packet_t structure
13705  * which holds the state information for this packet and invokes the
13706  * the classifier (via ipp_packet_process). The classification, depending on
13707  * configured filters, results in a list of actions for this packet. Invoking
13708  * an action may cause the packet to be dropped, in which case we return NULL.
13709  * proc indicates the callout position for
13710  * this packet and ill is the interface this packet arrived on or will leave
13711  * on (inbound and outbound resp.).
13712  *
13713  * We do the processing on the rill (mapped to the upper if ipmp), but MIB
13714  * on the ill corrsponding to the destination IP address.
13715  */
13716 mblk_t *
ip_process(ip_proc_t proc,mblk_t * mp,ill_t * rill,ill_t * ill)13717 ip_process(ip_proc_t proc, mblk_t *mp, ill_t *rill, ill_t *ill)
13718 {
13719 	ip_priv_t	*priv;
13720 	ipp_action_id_t	aid;
13721 	int		rc = 0;
13722 	ipp_packet_t	*pp;
13723 
13724 	/* If the classifier is not loaded, return  */
13725 	if ((aid = ipp_action_lookup(IPGPC_CLASSIFY)) == IPP_ACTION_INVAL) {
13726 		return (mp);
13727 	}
13728 
13729 	ASSERT(mp != NULL);
13730 
13731 	/* Allocate the packet structure */
13732 	rc = ipp_packet_alloc(&pp, "ip", aid);
13733 	if (rc != 0)
13734 		goto drop;
13735 
13736 	/* Allocate the private structure */
13737 	rc = ip_priv_alloc((void **)&priv);
13738 	if (rc != 0) {
13739 		ipp_packet_free(pp);
13740 		goto drop;
13741 	}
13742 	priv->proc = proc;
13743 	priv->ill_index = ill_get_upper_ifindex(rill);
13744 
13745 	ipp_packet_set_private(pp, priv, ip_priv_free);
13746 	ipp_packet_set_data(pp, mp);
13747 
13748 	/* Invoke the classifier */
13749 	rc = ipp_packet_process(&pp);
13750 	if (pp != NULL) {
13751 		mp = ipp_packet_get_data(pp);
13752 		ipp_packet_free(pp);
13753 		if (rc != 0)
13754 			goto drop;
13755 		return (mp);
13756 	} else {
13757 		/* No mp to trace in ip_drop_input/ip_drop_output  */
13758 		mp = NULL;
13759 	}
13760 drop:
13761 	if (proc == IPP_LOCAL_IN || proc == IPP_FWD_IN) {
13762 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
13763 		ip_drop_input("ip_process", mp, ill);
13764 	} else {
13765 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
13766 		ip_drop_output("ip_process", mp, ill);
13767 	}
13768 	freemsg(mp);
13769 	return (NULL);
13770 }
13771 
13772 /*
13773  * Propagate a multicast group membership operation (add/drop) on
13774  * all the interfaces crossed by the related multirt routes.
13775  * The call is considered successful if the operation succeeds
13776  * on at least one interface.
13777  *
13778  * This assumes that a set of IRE_HOST/RTF_MULTIRT has been created for the
13779  * multicast addresses with the ire argument being the first one.
13780  * We walk the bucket to find all the of those.
13781  *
13782  * Common to IPv4 and IPv6.
13783  */
13784 static int
ip_multirt_apply_membership(int (* fn)(conn_t *,boolean_t,const in6_addr_t *,ipaddr_t,uint_t,mcast_record_t,const in6_addr_t *),ire_t * ire,conn_t * connp,boolean_t checkonly,const in6_addr_t * v6group,mcast_record_t fmode,const in6_addr_t * v6src)13785 ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
13786     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
13787     ire_t *ire, conn_t *connp, boolean_t checkonly, const in6_addr_t *v6group,
13788     mcast_record_t fmode, const in6_addr_t *v6src)
13789 {
13790 	ire_t		*ire_gw;
13791 	irb_t		*irb;
13792 	int		ifindex;
13793 	int		error = 0;
13794 	int		result;
13795 	ip_stack_t	*ipst = ire->ire_ipst;
13796 	ipaddr_t	group;
13797 	boolean_t	isv6;
13798 	int		match_flags;
13799 
13800 	if (IN6_IS_ADDR_V4MAPPED(v6group)) {
13801 		IN6_V4MAPPED_TO_IPADDR(v6group, group);
13802 		isv6 = B_FALSE;
13803 	} else {
13804 		isv6 = B_TRUE;
13805 	}
13806 
13807 	irb = ire->ire_bucket;
13808 	ASSERT(irb != NULL);
13809 
13810 	result = 0;
13811 	irb_refhold(irb);
13812 	for (; ire != NULL; ire = ire->ire_next) {
13813 		if ((ire->ire_flags & RTF_MULTIRT) == 0)
13814 			continue;
13815 
13816 		/* We handle -ifp routes by matching on the ill if set */
13817 		match_flags = MATCH_IRE_TYPE;
13818 		if (ire->ire_ill != NULL)
13819 			match_flags |= MATCH_IRE_ILL;
13820 
13821 		if (isv6) {
13822 			if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6group))
13823 				continue;
13824 
13825 			ire_gw = ire_ftable_lookup_v6(&ire->ire_gateway_addr_v6,
13826 			    0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13827 			    match_flags, 0, ipst, NULL);
13828 		} else {
13829 			if (ire->ire_addr != group)
13830 				continue;
13831 
13832 			ire_gw = ire_ftable_lookup_v4(ire->ire_gateway_addr,
13833 			    0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13834 			    match_flags, 0, ipst, NULL);
13835 		}
13836 		/* No interface route exists for the gateway; skip this ire. */
13837 		if (ire_gw == NULL)
13838 			continue;
13839 		if (ire_gw->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
13840 			ire_refrele(ire_gw);
13841 			continue;
13842 		}
13843 		ASSERT(ire_gw->ire_ill != NULL);	/* IRE_INTERFACE */
13844 		ifindex = ire_gw->ire_ill->ill_phyint->phyint_ifindex;
13845 
13846 		/*
13847 		 * The operation is considered a success if
13848 		 * it succeeds at least once on any one interface.
13849 		 */
13850 		error = fn(connp, checkonly, v6group, INADDR_ANY, ifindex,
13851 		    fmode, v6src);
13852 		if (error == 0)
13853 			result = CGTP_MCAST_SUCCESS;
13854 
13855 		ire_refrele(ire_gw);
13856 	}
13857 	irb_refrele(irb);
13858 	/*
13859 	 * Consider the call as successful if we succeeded on at least
13860 	 * one interface. Otherwise, return the last encountered error.
13861 	 */
13862 	return (result == CGTP_MCAST_SUCCESS ? 0 : error);
13863 }
13864 
13865 /*
13866  * Return the expected CGTP hooks version number.
13867  */
13868 int
ip_cgtp_filter_supported(void)13869 ip_cgtp_filter_supported(void)
13870 {
13871 	return (ip_cgtp_filter_rev);
13872 }
13873 
13874 /*
13875  * CGTP hooks can be registered by invoking this function.
13876  * Checks that the version number matches.
13877  */
13878 int
ip_cgtp_filter_register(netstackid_t stackid,cgtp_filter_ops_t * ops)13879 ip_cgtp_filter_register(netstackid_t stackid, cgtp_filter_ops_t *ops)
13880 {
13881 	netstack_t *ns;
13882 	ip_stack_t *ipst;
13883 
13884 	if (ops->cfo_filter_rev != CGTP_FILTER_REV)
13885 		return (ENOTSUP);
13886 
13887 	ns = netstack_find_by_stackid(stackid);
13888 	if (ns == NULL)
13889 		return (EINVAL);
13890 	ipst = ns->netstack_ip;
13891 	ASSERT(ipst != NULL);
13892 
13893 	if (ipst->ips_ip_cgtp_filter_ops != NULL) {
13894 		netstack_rele(ns);
13895 		return (EALREADY);
13896 	}
13897 
13898 	ipst->ips_ip_cgtp_filter_ops = ops;
13899 
13900 	ill_set_inputfn_all(ipst);
13901 
13902 	netstack_rele(ns);
13903 	return (0);
13904 }
13905 
13906 /*
13907  * CGTP hooks can be unregistered by invoking this function.
13908  * Returns ENXIO if there was no registration.
13909  * Returns EBUSY if the ndd variable has not been turned off.
13910  */
13911 int
ip_cgtp_filter_unregister(netstackid_t stackid)13912 ip_cgtp_filter_unregister(netstackid_t stackid)
13913 {
13914 	netstack_t *ns;
13915 	ip_stack_t *ipst;
13916 
13917 	ns = netstack_find_by_stackid(stackid);
13918 	if (ns == NULL)
13919 		return (EINVAL);
13920 	ipst = ns->netstack_ip;
13921 	ASSERT(ipst != NULL);
13922 
13923 	if (ipst->ips_ip_cgtp_filter) {
13924 		netstack_rele(ns);
13925 		return (EBUSY);
13926 	}
13927 
13928 	if (ipst->ips_ip_cgtp_filter_ops == NULL) {
13929 		netstack_rele(ns);
13930 		return (ENXIO);
13931 	}
13932 	ipst->ips_ip_cgtp_filter_ops = NULL;
13933 
13934 	ill_set_inputfn_all(ipst);
13935 
13936 	netstack_rele(ns);
13937 	return (0);
13938 }
13939 
13940 /*
13941  * Check whether there is a CGTP filter registration.
13942  * Returns non-zero if there is a registration, otherwise returns zero.
13943  * Note: returns zero if bad stackid.
13944  */
13945 int
ip_cgtp_filter_is_registered(netstackid_t stackid)13946 ip_cgtp_filter_is_registered(netstackid_t stackid)
13947 {
13948 	netstack_t *ns;
13949 	ip_stack_t *ipst;
13950 	int ret;
13951 
13952 	ns = netstack_find_by_stackid(stackid);
13953 	if (ns == NULL)
13954 		return (0);
13955 	ipst = ns->netstack_ip;
13956 	ASSERT(ipst != NULL);
13957 
13958 	if (ipst->ips_ip_cgtp_filter_ops != NULL)
13959 		ret = 1;
13960 	else
13961 		ret = 0;
13962 
13963 	netstack_rele(ns);
13964 	return (ret);
13965 }
13966 
13967 static int
ip_squeue_switch(int val)13968 ip_squeue_switch(int val)
13969 {
13970 	int rval;
13971 
13972 	switch (val) {
13973 	case IP_SQUEUE_ENTER_NODRAIN:
13974 		rval = SQ_NODRAIN;
13975 		break;
13976 	case IP_SQUEUE_ENTER:
13977 		rval = SQ_PROCESS;
13978 		break;
13979 	case IP_SQUEUE_FILL:
13980 	default:
13981 		rval = SQ_FILL;
13982 		break;
13983 	}
13984 	return (rval);
13985 }
13986 
13987 static void *
ip_kstat2_init(netstackid_t stackid,ip_stat_t * ip_statisticsp)13988 ip_kstat2_init(netstackid_t stackid, ip_stat_t *ip_statisticsp)
13989 {
13990 	kstat_t *ksp;
13991 
13992 	ip_stat_t template = {
13993 		{ "ip_udp_fannorm",		KSTAT_DATA_UINT64 },
13994 		{ "ip_udp_fanmb",		KSTAT_DATA_UINT64 },
13995 		{ "ip_recv_pullup",		KSTAT_DATA_UINT64 },
13996 		{ "ip_db_ref",			KSTAT_DATA_UINT64 },
13997 		{ "ip_notaligned",		KSTAT_DATA_UINT64 },
13998 		{ "ip_multimblk",		KSTAT_DATA_UINT64 },
13999 		{ "ip_opt",			KSTAT_DATA_UINT64 },
14000 		{ "ipsec_proto_ahesp",		KSTAT_DATA_UINT64 },
14001 		{ "ip_conn_flputbq",		KSTAT_DATA_UINT64 },
14002 		{ "ip_conn_walk_drain",		KSTAT_DATA_UINT64 },
14003 		{ "ip_out_sw_cksum",		KSTAT_DATA_UINT64 },
14004 		{ "ip_out_sw_cksum_bytes",	KSTAT_DATA_UINT64 },
14005 		{ "ip_in_sw_cksum",		KSTAT_DATA_UINT64 },
14006 		{ "ip_ire_reclaim_calls",	KSTAT_DATA_UINT64 },
14007 		{ "ip_ire_reclaim_deleted",	KSTAT_DATA_UINT64 },
14008 		{ "ip_nce_reclaim_calls",	KSTAT_DATA_UINT64 },
14009 		{ "ip_nce_reclaim_deleted",	KSTAT_DATA_UINT64 },
14010 		{ "ip_nce_mcast_reclaim_calls",	KSTAT_DATA_UINT64 },
14011 		{ "ip_nce_mcast_reclaim_deleted",	KSTAT_DATA_UINT64 },
14012 		{ "ip_nce_mcast_reclaim_tqfail",	KSTAT_DATA_UINT64 },
14013 		{ "ip_dce_reclaim_calls",	KSTAT_DATA_UINT64 },
14014 		{ "ip_dce_reclaim_deleted",	KSTAT_DATA_UINT64 },
14015 		{ "ip_tcp_in_full_hw_cksum_err",	KSTAT_DATA_UINT64 },
14016 		{ "ip_tcp_in_part_hw_cksum_err",	KSTAT_DATA_UINT64 },
14017 		{ "ip_tcp_in_sw_cksum_err",		KSTAT_DATA_UINT64 },
14018 		{ "ip_udp_in_full_hw_cksum_err",	KSTAT_DATA_UINT64 },
14019 		{ "ip_udp_in_part_hw_cksum_err",	KSTAT_DATA_UINT64 },
14020 		{ "ip_udp_in_sw_cksum_err",	KSTAT_DATA_UINT64 },
14021 		{ "conn_in_recvdstaddr",	KSTAT_DATA_UINT64 },
14022 		{ "conn_in_recvopts",		KSTAT_DATA_UINT64 },
14023 		{ "conn_in_recvif",		KSTAT_DATA_UINT64 },
14024 		{ "conn_in_recvslla",		KSTAT_DATA_UINT64 },
14025 		{ "conn_in_recvucred",		KSTAT_DATA_UINT64 },
14026 		{ "conn_in_recvttl",		KSTAT_DATA_UINT64 },
14027 		{ "conn_in_recvtos",		KSTAT_DATA_UINT64 },
14028 		{ "conn_in_recvhopopts",	KSTAT_DATA_UINT64 },
14029 		{ "conn_in_recvhoplimit",	KSTAT_DATA_UINT64 },
14030 		{ "conn_in_recvdstopts",	KSTAT_DATA_UINT64 },
14031 		{ "conn_in_recvrthdrdstopts",	KSTAT_DATA_UINT64 },
14032 		{ "conn_in_recvrthdr",		KSTAT_DATA_UINT64 },
14033 		{ "conn_in_recvpktinfo",	KSTAT_DATA_UINT64 },
14034 		{ "conn_in_recvtclass",		KSTAT_DATA_UINT64 },
14035 		{ "conn_in_timestamp",		KSTAT_DATA_UINT64 },
14036 	};
14037 
14038 	ksp = kstat_create_netstack("ip", 0, "ipstat", "net",
14039 	    KSTAT_TYPE_NAMED, sizeof (template) / sizeof (kstat_named_t),
14040 	    KSTAT_FLAG_VIRTUAL, stackid);
14041 
14042 	if (ksp == NULL)
14043 		return (NULL);
14044 
14045 	bcopy(&template, ip_statisticsp, sizeof (template));
14046 	ksp->ks_data = (void *)ip_statisticsp;
14047 	ksp->ks_private = (void *)(uintptr_t)stackid;
14048 
14049 	kstat_install(ksp);
14050 	return (ksp);
14051 }
14052 
14053 static void
ip_kstat2_fini(netstackid_t stackid,kstat_t * ksp)14054 ip_kstat2_fini(netstackid_t stackid, kstat_t *ksp)
14055 {
14056 	if (ksp != NULL) {
14057 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14058 		kstat_delete_netstack(ksp, stackid);
14059 	}
14060 }
14061 
14062 static void *
ip_kstat_init(netstackid_t stackid,ip_stack_t * ipst)14063 ip_kstat_init(netstackid_t stackid, ip_stack_t *ipst)
14064 {
14065 	kstat_t	*ksp;
14066 
14067 	ip_named_kstat_t template = {
14068 		{ "forwarding",		KSTAT_DATA_UINT32, 0 },
14069 		{ "defaultTTL",		KSTAT_DATA_UINT32, 0 },
14070 		{ "inReceives",		KSTAT_DATA_UINT64, 0 },
14071 		{ "inHdrErrors",	KSTAT_DATA_UINT32, 0 },
14072 		{ "inAddrErrors",	KSTAT_DATA_UINT32, 0 },
14073 		{ "forwDatagrams",	KSTAT_DATA_UINT64, 0 },
14074 		{ "inUnknownProtos",	KSTAT_DATA_UINT32, 0 },
14075 		{ "inDiscards",		KSTAT_DATA_UINT32, 0 },
14076 		{ "inDelivers",		KSTAT_DATA_UINT64, 0 },
14077 		{ "outRequests",	KSTAT_DATA_UINT64, 0 },
14078 		{ "outDiscards",	KSTAT_DATA_UINT32, 0 },
14079 		{ "outNoRoutes",	KSTAT_DATA_UINT32, 0 },
14080 		{ "reasmTimeout",	KSTAT_DATA_UINT32, 0 },
14081 		{ "reasmReqds",		KSTAT_DATA_UINT32, 0 },
14082 		{ "reasmOKs",		KSTAT_DATA_UINT32, 0 },
14083 		{ "reasmFails",		KSTAT_DATA_UINT32, 0 },
14084 		{ "fragOKs",		KSTAT_DATA_UINT32, 0 },
14085 		{ "fragFails",		KSTAT_DATA_UINT32, 0 },
14086 		{ "fragCreates",	KSTAT_DATA_UINT32, 0 },
14087 		{ "addrEntrySize",	KSTAT_DATA_INT32, 0 },
14088 		{ "routeEntrySize",	KSTAT_DATA_INT32, 0 },
14089 		{ "netToMediaEntrySize",	KSTAT_DATA_INT32, 0 },
14090 		{ "routingDiscards",	KSTAT_DATA_UINT32, 0 },
14091 		{ "inErrs",		KSTAT_DATA_UINT32, 0 },
14092 		{ "noPorts",		KSTAT_DATA_UINT32, 0 },
14093 		{ "inCksumErrs",	KSTAT_DATA_UINT32, 0 },
14094 		{ "reasmDuplicates",	KSTAT_DATA_UINT32, 0 },
14095 		{ "reasmPartDups",	KSTAT_DATA_UINT32, 0 },
14096 		{ "forwProhibits",	KSTAT_DATA_UINT32, 0 },
14097 		{ "udpInCksumErrs",	KSTAT_DATA_UINT32, 0 },
14098 		{ "udpInOverflows",	KSTAT_DATA_UINT32, 0 },
14099 		{ "rawipInOverflows",	KSTAT_DATA_UINT32, 0 },
14100 		{ "ipsecInSucceeded",	KSTAT_DATA_UINT32, 0 },
14101 		{ "ipsecInFailed",	KSTAT_DATA_INT32, 0 },
14102 		{ "memberEntrySize",	KSTAT_DATA_INT32, 0 },
14103 		{ "inIPv6",		KSTAT_DATA_UINT32, 0 },
14104 		{ "outIPv6",		KSTAT_DATA_UINT32, 0 },
14105 		{ "outSwitchIPv6",	KSTAT_DATA_UINT32, 0 },
14106 	};
14107 
14108 	ksp = kstat_create_netstack("ip", 0, "ip", "mib2", KSTAT_TYPE_NAMED,
14109 	    NUM_OF_FIELDS(ip_named_kstat_t), 0, stackid);
14110 	if (ksp == NULL || ksp->ks_data == NULL)
14111 		return (NULL);
14112 
14113 	template.forwarding.value.ui32 = WE_ARE_FORWARDING(ipst) ? 1:2;
14114 	template.defaultTTL.value.ui32 = (uint32_t)ipst->ips_ip_def_ttl;
14115 	template.reasmTimeout.value.ui32 = ipst->ips_ip_reassembly_timeout;
14116 	template.addrEntrySize.value.i32 = sizeof (mib2_ipAddrEntry_t);
14117 	template.routeEntrySize.value.i32 = sizeof (mib2_ipRouteEntry_t);
14118 
14119 	template.netToMediaEntrySize.value.i32 =
14120 	    sizeof (mib2_ipNetToMediaEntry_t);
14121 
14122 	template.memberEntrySize.value.i32 = sizeof (ipv6_member_t);
14123 
14124 	bcopy(&template, ksp->ks_data, sizeof (template));
14125 	ksp->ks_update = ip_kstat_update;
14126 	ksp->ks_private = (void *)(uintptr_t)stackid;
14127 
14128 	kstat_install(ksp);
14129 	return (ksp);
14130 }
14131 
14132 static void
ip_kstat_fini(netstackid_t stackid,kstat_t * ksp)14133 ip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14134 {
14135 	if (ksp != NULL) {
14136 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14137 		kstat_delete_netstack(ksp, stackid);
14138 	}
14139 }
14140 
14141 static int
ip_kstat_update(kstat_t * kp,int rw)14142 ip_kstat_update(kstat_t *kp, int rw)
14143 {
14144 	ip_named_kstat_t *ipkp;
14145 	mib2_ipIfStatsEntry_t ipmib;
14146 	ill_walk_context_t ctx;
14147 	ill_t *ill;
14148 	netstackid_t	stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14149 	netstack_t	*ns;
14150 	ip_stack_t	*ipst;
14151 
14152 	if (kp->ks_data == NULL)
14153 		return (EIO);
14154 
14155 	if (rw == KSTAT_WRITE)
14156 		return (EACCES);
14157 
14158 	ns = netstack_find_by_stackid(stackid);
14159 	if (ns == NULL)
14160 		return (-1);
14161 	ipst = ns->netstack_ip;
14162 	if (ipst == NULL) {
14163 		netstack_rele(ns);
14164 		return (-1);
14165 	}
14166 	ipkp = (ip_named_kstat_t *)kp->ks_data;
14167 
14168 	bcopy(&ipst->ips_ip_mib, &ipmib, sizeof (ipmib));
14169 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
14170 	ill = ILL_START_WALK_V4(&ctx, ipst);
14171 	for (; ill != NULL; ill = ill_next(&ctx, ill))
14172 		ip_mib2_add_ip_stats(&ipmib, ill->ill_ip_mib);
14173 	rw_exit(&ipst->ips_ill_g_lock);
14174 
14175 	ipkp->forwarding.value.ui32 =		ipmib.ipIfStatsForwarding;
14176 	ipkp->defaultTTL.value.ui32 =		ipmib.ipIfStatsDefaultTTL;
14177 	ipkp->inReceives.value.ui64 =		ipmib.ipIfStatsHCInReceives;
14178 	ipkp->inHdrErrors.value.ui32 =		ipmib.ipIfStatsInHdrErrors;
14179 	ipkp->inAddrErrors.value.ui32 =		ipmib.ipIfStatsInAddrErrors;
14180 	ipkp->forwDatagrams.value.ui64 = ipmib.ipIfStatsHCOutForwDatagrams;
14181 	ipkp->inUnknownProtos.value.ui32 =	ipmib.ipIfStatsInUnknownProtos;
14182 	ipkp->inDiscards.value.ui32 =		ipmib.ipIfStatsInDiscards;
14183 	ipkp->inDelivers.value.ui64 =		ipmib.ipIfStatsHCInDelivers;
14184 	ipkp->outRequests.value.ui64 =		ipmib.ipIfStatsHCOutRequests;
14185 	ipkp->outDiscards.value.ui32 =		ipmib.ipIfStatsOutDiscards;
14186 	ipkp->outNoRoutes.value.ui32 =		ipmib.ipIfStatsOutNoRoutes;
14187 	ipkp->reasmTimeout.value.ui32 =		ipst->ips_ip_reassembly_timeout;
14188 	ipkp->reasmReqds.value.ui32 =		ipmib.ipIfStatsReasmReqds;
14189 	ipkp->reasmOKs.value.ui32 =		ipmib.ipIfStatsReasmOKs;
14190 	ipkp->reasmFails.value.ui32 =		ipmib.ipIfStatsReasmFails;
14191 	ipkp->fragOKs.value.ui32 =		ipmib.ipIfStatsOutFragOKs;
14192 	ipkp->fragFails.value.ui32 =		ipmib.ipIfStatsOutFragFails;
14193 	ipkp->fragCreates.value.ui32 =		ipmib.ipIfStatsOutFragCreates;
14194 
14195 	ipkp->routingDiscards.value.ui32 =	0;
14196 	ipkp->inErrs.value.ui32 =		ipmib.tcpIfStatsInErrs;
14197 	ipkp->noPorts.value.ui32 =		ipmib.udpIfStatsNoPorts;
14198 	ipkp->inCksumErrs.value.ui32 =		ipmib.ipIfStatsInCksumErrs;
14199 	ipkp->reasmDuplicates.value.ui32 =	ipmib.ipIfStatsReasmDuplicates;
14200 	ipkp->reasmPartDups.value.ui32 =	ipmib.ipIfStatsReasmPartDups;
14201 	ipkp->forwProhibits.value.ui32 =	ipmib.ipIfStatsForwProhibits;
14202 	ipkp->udpInCksumErrs.value.ui32 =	ipmib.udpIfStatsInCksumErrs;
14203 	ipkp->udpInOverflows.value.ui32 =	ipmib.udpIfStatsInOverflows;
14204 	ipkp->rawipInOverflows.value.ui32 =	ipmib.rawipIfStatsInOverflows;
14205 	ipkp->ipsecInSucceeded.value.ui32 =	ipmib.ipsecIfStatsInSucceeded;
14206 	ipkp->ipsecInFailed.value.i32 =		ipmib.ipsecIfStatsInFailed;
14207 
14208 	ipkp->inIPv6.value.ui32 =	ipmib.ipIfStatsInWrongIPVersion;
14209 	ipkp->outIPv6.value.ui32 =	ipmib.ipIfStatsOutWrongIPVersion;
14210 	ipkp->outSwitchIPv6.value.ui32 = ipmib.ipIfStatsOutSwitchIPVersion;
14211 
14212 	netstack_rele(ns);
14213 
14214 	return (0);
14215 }
14216 
14217 static void *
icmp_kstat_init(netstackid_t stackid)14218 icmp_kstat_init(netstackid_t stackid)
14219 {
14220 	kstat_t	*ksp;
14221 
14222 	icmp_named_kstat_t template = {
14223 		{ "inMsgs",		KSTAT_DATA_UINT32 },
14224 		{ "inErrors",		KSTAT_DATA_UINT32 },
14225 		{ "inDestUnreachs",	KSTAT_DATA_UINT32 },
14226 		{ "inTimeExcds",	KSTAT_DATA_UINT32 },
14227 		{ "inParmProbs",	KSTAT_DATA_UINT32 },
14228 		{ "inSrcQuenchs",	KSTAT_DATA_UINT32 },
14229 		{ "inRedirects",	KSTAT_DATA_UINT32 },
14230 		{ "inEchos",		KSTAT_DATA_UINT32 },
14231 		{ "inEchoReps",		KSTAT_DATA_UINT32 },
14232 		{ "inTimestamps",	KSTAT_DATA_UINT32 },
14233 		{ "inTimestampReps",	KSTAT_DATA_UINT32 },
14234 		{ "inAddrMasks",	KSTAT_DATA_UINT32 },
14235 		{ "inAddrMaskReps",	KSTAT_DATA_UINT32 },
14236 		{ "outMsgs",		KSTAT_DATA_UINT32 },
14237 		{ "outErrors",		KSTAT_DATA_UINT32 },
14238 		{ "outDestUnreachs",	KSTAT_DATA_UINT32 },
14239 		{ "outTimeExcds",	KSTAT_DATA_UINT32 },
14240 		{ "outParmProbs",	KSTAT_DATA_UINT32 },
14241 		{ "outSrcQuenchs",	KSTAT_DATA_UINT32 },
14242 		{ "outRedirects",	KSTAT_DATA_UINT32 },
14243 		{ "outEchos",		KSTAT_DATA_UINT32 },
14244 		{ "outEchoReps",	KSTAT_DATA_UINT32 },
14245 		{ "outTimestamps",	KSTAT_DATA_UINT32 },
14246 		{ "outTimestampReps",	KSTAT_DATA_UINT32 },
14247 		{ "outAddrMasks",	KSTAT_DATA_UINT32 },
14248 		{ "outAddrMaskReps",	KSTAT_DATA_UINT32 },
14249 		{ "inChksumErrs",	KSTAT_DATA_UINT32 },
14250 		{ "inUnknowns",		KSTAT_DATA_UINT32 },
14251 		{ "inFragNeeded",	KSTAT_DATA_UINT32 },
14252 		{ "outFragNeeded",	KSTAT_DATA_UINT32 },
14253 		{ "outDrops",		KSTAT_DATA_UINT32 },
14254 		{ "inOverFlows",	KSTAT_DATA_UINT32 },
14255 		{ "inBadRedirects",	KSTAT_DATA_UINT32 },
14256 	};
14257 
14258 	ksp = kstat_create_netstack("ip", 0, "icmp", "mib2", KSTAT_TYPE_NAMED,
14259 	    NUM_OF_FIELDS(icmp_named_kstat_t), 0, stackid);
14260 	if (ksp == NULL || ksp->ks_data == NULL)
14261 		return (NULL);
14262 
14263 	bcopy(&template, ksp->ks_data, sizeof (template));
14264 
14265 	ksp->ks_update = icmp_kstat_update;
14266 	ksp->ks_private = (void *)(uintptr_t)stackid;
14267 
14268 	kstat_install(ksp);
14269 	return (ksp);
14270 }
14271 
14272 static void
icmp_kstat_fini(netstackid_t stackid,kstat_t * ksp)14273 icmp_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14274 {
14275 	if (ksp != NULL) {
14276 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14277 		kstat_delete_netstack(ksp, stackid);
14278 	}
14279 }
14280 
14281 static int
icmp_kstat_update(kstat_t * kp,int rw)14282 icmp_kstat_update(kstat_t *kp, int rw)
14283 {
14284 	icmp_named_kstat_t *icmpkp;
14285 	netstackid_t	stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14286 	netstack_t	*ns;
14287 	ip_stack_t	*ipst;
14288 
14289 	if (kp->ks_data == NULL)
14290 		return (EIO);
14291 
14292 	if (rw == KSTAT_WRITE)
14293 		return (EACCES);
14294 
14295 	ns = netstack_find_by_stackid(stackid);
14296 	if (ns == NULL)
14297 		return (-1);
14298 	ipst = ns->netstack_ip;
14299 	if (ipst == NULL) {
14300 		netstack_rele(ns);
14301 		return (-1);
14302 	}
14303 	icmpkp = (icmp_named_kstat_t *)kp->ks_data;
14304 
14305 	icmpkp->inMsgs.value.ui32 =	    ipst->ips_icmp_mib.icmpInMsgs;
14306 	icmpkp->inErrors.value.ui32 =	    ipst->ips_icmp_mib.icmpInErrors;
14307 	icmpkp->inDestUnreachs.value.ui32 =
14308 	    ipst->ips_icmp_mib.icmpInDestUnreachs;
14309 	icmpkp->inTimeExcds.value.ui32 =    ipst->ips_icmp_mib.icmpInTimeExcds;
14310 	icmpkp->inParmProbs.value.ui32 =    ipst->ips_icmp_mib.icmpInParmProbs;
14311 	icmpkp->inSrcQuenchs.value.ui32 =   ipst->ips_icmp_mib.icmpInSrcQuenchs;
14312 	icmpkp->inRedirects.value.ui32 =    ipst->ips_icmp_mib.icmpInRedirects;
14313 	icmpkp->inEchos.value.ui32 =	    ipst->ips_icmp_mib.icmpInEchos;
14314 	icmpkp->inEchoReps.value.ui32 =	    ipst->ips_icmp_mib.icmpInEchoReps;
14315 	icmpkp->inTimestamps.value.ui32 =   ipst->ips_icmp_mib.icmpInTimestamps;
14316 	icmpkp->inTimestampReps.value.ui32 =
14317 	    ipst->ips_icmp_mib.icmpInTimestampReps;
14318 	icmpkp->inAddrMasks.value.ui32 =    ipst->ips_icmp_mib.icmpInAddrMasks;
14319 	icmpkp->inAddrMaskReps.value.ui32 =
14320 	    ipst->ips_icmp_mib.icmpInAddrMaskReps;
14321 	icmpkp->outMsgs.value.ui32 =	    ipst->ips_icmp_mib.icmpOutMsgs;
14322 	icmpkp->outErrors.value.ui32 =	    ipst->ips_icmp_mib.icmpOutErrors;
14323 	icmpkp->outDestUnreachs.value.ui32 =
14324 	    ipst->ips_icmp_mib.icmpOutDestUnreachs;
14325 	icmpkp->outTimeExcds.value.ui32 =   ipst->ips_icmp_mib.icmpOutTimeExcds;
14326 	icmpkp->outParmProbs.value.ui32 =   ipst->ips_icmp_mib.icmpOutParmProbs;
14327 	icmpkp->outSrcQuenchs.value.ui32 =
14328 	    ipst->ips_icmp_mib.icmpOutSrcQuenchs;
14329 	icmpkp->outRedirects.value.ui32 =   ipst->ips_icmp_mib.icmpOutRedirects;
14330 	icmpkp->outEchos.value.ui32 =	    ipst->ips_icmp_mib.icmpOutEchos;
14331 	icmpkp->outEchoReps.value.ui32 =    ipst->ips_icmp_mib.icmpOutEchoReps;
14332 	icmpkp->outTimestamps.value.ui32 =
14333 	    ipst->ips_icmp_mib.icmpOutTimestamps;
14334 	icmpkp->outTimestampReps.value.ui32 =
14335 	    ipst->ips_icmp_mib.icmpOutTimestampReps;
14336 	icmpkp->outAddrMasks.value.ui32 =
14337 	    ipst->ips_icmp_mib.icmpOutAddrMasks;
14338 	icmpkp->outAddrMaskReps.value.ui32 =
14339 	    ipst->ips_icmp_mib.icmpOutAddrMaskReps;
14340 	icmpkp->inCksumErrs.value.ui32 =    ipst->ips_icmp_mib.icmpInCksumErrs;
14341 	icmpkp->inUnknowns.value.ui32 =	    ipst->ips_icmp_mib.icmpInUnknowns;
14342 	icmpkp->inFragNeeded.value.ui32 =   ipst->ips_icmp_mib.icmpInFragNeeded;
14343 	icmpkp->outFragNeeded.value.ui32 =
14344 	    ipst->ips_icmp_mib.icmpOutFragNeeded;
14345 	icmpkp->outDrops.value.ui32 =	    ipst->ips_icmp_mib.icmpOutDrops;
14346 	icmpkp->inOverflows.value.ui32 =    ipst->ips_icmp_mib.icmpInOverflows;
14347 	icmpkp->inBadRedirects.value.ui32 =
14348 	    ipst->ips_icmp_mib.icmpInBadRedirects;
14349 
14350 	netstack_rele(ns);
14351 	return (0);
14352 }
14353 
14354 /*
14355  * This is the fanout function for raw socket opened for SCTP.  Note
14356  * that it is called after SCTP checks that there is no socket which
14357  * wants a packet.  Then before SCTP handles this out of the blue packet,
14358  * this function is called to see if there is any raw socket for SCTP.
14359  * If there is and it is bound to the correct address, the packet will
14360  * be sent to that socket.  Note that only one raw socket can be bound to
14361  * a port.  This is assured in ipcl_sctp_hash_insert();
14362  */
14363 void
ip_fanout_sctp_raw(mblk_t * mp,ipha_t * ipha,ip6_t * ip6h,uint32_t ports,ip_recv_attr_t * ira)14364 ip_fanout_sctp_raw(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h, uint32_t ports,
14365     ip_recv_attr_t *ira)
14366 {
14367 	conn_t		*connp;
14368 	queue_t		*rq;
14369 	boolean_t	secure;
14370 	ill_t		*ill = ira->ira_ill;
14371 	ip_stack_t	*ipst = ill->ill_ipst;
14372 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
14373 	sctp_stack_t	*sctps = ipst->ips_netstack->netstack_sctp;
14374 	iaflags_t	iraflags = ira->ira_flags;
14375 	ill_t		*rill = ira->ira_rill;
14376 
14377 	secure = iraflags & IRAF_IPSEC_SECURE;
14378 
14379 	connp = ipcl_classify_raw(mp, IPPROTO_SCTP, ports, ipha, ip6h,
14380 	    ira, ipst);
14381 	if (connp == NULL) {
14382 		/*
14383 		 * Although raw sctp is not summed, OOB chunks must be.
14384 		 * Drop the packet here if the sctp checksum failed.
14385 		 */
14386 		if (iraflags & IRAF_SCTP_CSUM_ERR) {
14387 			SCTPS_BUMP_MIB(sctps, sctpChecksumError);
14388 			freemsg(mp);
14389 			return;
14390 		}
14391 		ira->ira_ill = ira->ira_rill = NULL;
14392 		sctp_ootb_input(mp, ira, ipst);
14393 		ira->ira_ill = ill;
14394 		ira->ira_rill = rill;
14395 		return;
14396 	}
14397 
14398 	if (connp->conn_min_ttl != 0 && connp->conn_min_ttl > ira->ira_ttl) {
14399 		CONN_DEC_REF(connp);
14400 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
14401 		ip_drop_input("ipIfStatsInDiscards", mp, ill);
14402 		freemsg(mp);
14403 		return;
14404 	}
14405 
14406 	rq = connp->conn_rq;
14407 	if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
14408 		CONN_DEC_REF(connp);
14409 		BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
14410 		freemsg(mp);
14411 		return;
14412 	}
14413 	if (((iraflags & IRAF_IS_IPV4) ?
14414 	    CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
14415 	    CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
14416 	    secure) {
14417 		mp = ipsec_check_inbound_policy(mp, connp, ipha,
14418 		    ip6h, ira);
14419 		if (mp == NULL) {
14420 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
14421 			/* Note that mp is NULL */
14422 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
14423 			CONN_DEC_REF(connp);
14424 			return;
14425 		}
14426 	}
14427 
14428 	if (iraflags & IRAF_ICMP_ERROR) {
14429 		(connp->conn_recvicmp)(connp, mp, NULL, ira);
14430 	} else {
14431 		ill_t *rill = ira->ira_rill;
14432 
14433 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
14434 		/* This is the SOCK_RAW, IPPROTO_SCTP case. */
14435 		ira->ira_ill = ira->ira_rill = NULL;
14436 		(connp->conn_recv)(connp, mp, NULL, ira);
14437 		ira->ira_ill = ill;
14438 		ira->ira_rill = rill;
14439 	}
14440 	CONN_DEC_REF(connp);
14441 }
14442 
14443 /*
14444  * Free a packet that has the link-layer dl_unitdata_req_t or fast-path
14445  * header before the ip payload.
14446  */
14447 static void
ip_xmit_flowctl_drop(ill_t * ill,mblk_t * mp,boolean_t is_fp_mp,int fp_mp_len)14448 ip_xmit_flowctl_drop(ill_t *ill, mblk_t *mp, boolean_t is_fp_mp, int fp_mp_len)
14449 {
14450 	int len = (mp->b_wptr - mp->b_rptr);
14451 	mblk_t *ip_mp;
14452 
14453 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14454 	if (is_fp_mp || len != fp_mp_len) {
14455 		if (len > fp_mp_len) {
14456 			/*
14457 			 * fastpath header and ip header in the first mblk
14458 			 */
14459 			mp->b_rptr += fp_mp_len;
14460 		} else {
14461 			/*
14462 			 * ip_xmit_attach_llhdr had to prepend an mblk to
14463 			 * attach the fastpath header before ip header.
14464 			 */
14465 			ip_mp = mp->b_cont;
14466 			freeb(mp);
14467 			mp = ip_mp;
14468 			mp->b_rptr += (fp_mp_len - len);
14469 		}
14470 	} else {
14471 		ip_mp = mp->b_cont;
14472 		freeb(mp);
14473 		mp = ip_mp;
14474 	}
14475 	ip_drop_output("ipIfStatsOutDiscards - flow ctl", mp, ill);
14476 	freemsg(mp);
14477 }
14478 
14479 /*
14480  * Normal post fragmentation function.
14481  *
14482  * Send a packet using the passed in nce. This handles both IPv4 and IPv6
14483  * using the same state machine.
14484  *
14485  * We return an error on failure. In particular we return EWOULDBLOCK
14486  * when the driver flow controls. In that case this ensures that ip_wsrv runs
14487  * (currently by canputnext failure resulting in backenabling from GLD.)
14488  * This allows the callers of conn_ip_output() to use EWOULDBLOCK as an
14489  * indication that they can flow control until ip_wsrv() tells then to restart.
14490  *
14491  * If the nce passed by caller is incomplete, this function
14492  * queues the packet and if necessary, sends ARP request and bails.
14493  * If the Neighbor Cache passed is fully resolved, we simply prepend
14494  * the link-layer header to the packet, do ipsec hw acceleration
14495  * work if necessary, and send the packet out on the wire.
14496  */
14497 /* ARGSUSED6 */
14498 int
ip_xmit(mblk_t * mp,nce_t * nce,iaflags_t ixaflags,uint_t pkt_len,uint32_t xmit_hint,zoneid_t szone,zoneid_t nolzid,uintptr_t * ixacookie)14499 ip_xmit(mblk_t *mp, nce_t *nce, iaflags_t ixaflags, uint_t pkt_len,
14500     uint32_t xmit_hint, zoneid_t szone, zoneid_t nolzid, uintptr_t *ixacookie)
14501 {
14502 	queue_t		*wq;
14503 	ill_t		*ill = nce->nce_ill;
14504 	ip_stack_t	*ipst = ill->ill_ipst;
14505 	uint64_t	delta;
14506 	boolean_t	isv6 = ill->ill_isv6;
14507 	boolean_t	fp_mp;
14508 	ncec_t		*ncec = nce->nce_common;
14509 	int64_t		now = LBOLT_FASTPATH64;
14510 	boolean_t	is_probe;
14511 
14512 	DTRACE_PROBE1(ip__xmit, nce_t *, nce);
14513 
14514 	ASSERT(mp != NULL);
14515 	ASSERT(mp->b_datap->db_type == M_DATA);
14516 	ASSERT3U(pkt_len, ==, msgdsize(mp));
14517 
14518 	/*
14519 	 * If we have already been here and are coming back after ARP/ND.
14520 	 * the IXAF_NO_TRACE flag is set. We skip FW_HOOKS, DTRACE and ipobs
14521 	 * in that case since they have seen the packet when it came here
14522 	 * the first time.
14523 	 */
14524 	if (ixaflags & IXAF_NO_TRACE)
14525 		goto sendit;
14526 
14527 	if (ixaflags & IXAF_IS_IPV4) {
14528 		ipha_t *ipha = (ipha_t *)mp->b_rptr;
14529 
14530 		ASSERT(!isv6);
14531 		ASSERT3U(pkt_len, ==,
14532 		    ntohs(((ipha_t *)mp->b_rptr)->ipha_length));
14533 		if (HOOKS4_INTERESTED_PHYSICAL_OUT(ipst) &&
14534 		    !(ixaflags & IXAF_NO_PFHOOK)) {
14535 			int	error;
14536 
14537 			FW_HOOKS(ipst->ips_ip4_physical_out_event,
14538 			    ipst->ips_ipv4firewall_physical_out,
14539 			    NULL, ill, ipha, mp, mp, 0, ipst, error);
14540 			DTRACE_PROBE1(ip4__physical__out__end,
14541 			    mblk_t *, mp);
14542 			if (mp == NULL)
14543 				return (error);
14544 
14545 			/* The length could have changed */
14546 			pkt_len = msgdsize(mp);
14547 		}
14548 		if (ipst->ips_ip4_observe.he_interested) {
14549 			/*
14550 			 * Note that for TX the zoneid is the sending
14551 			 * zone, whether or not MLP is in play.
14552 			 * Since the szone argument is the IP zoneid (i.e.,
14553 			 * zero for exclusive-IP zones) and ipobs wants
14554 			 * the system zoneid, we map it here.
14555 			 */
14556 			szone = IP_REAL_ZONEID(szone, ipst);
14557 
14558 			/*
14559 			 * On the outbound path the destination zone will be
14560 			 * unknown as we're sending this packet out on the
14561 			 * wire.
14562 			 */
14563 			ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14564 			    ill, ipst);
14565 		}
14566 		DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14567 		    void_ip_t *, ipha,  __dtrace_ipsr_ill_t *, ill,
14568 		    ipha_t *, ipha, ip6_t *, NULL, int, 0);
14569 	} else {
14570 		ip6_t *ip6h = (ip6_t *)mp->b_rptr;
14571 
14572 		ASSERT(isv6);
14573 		ASSERT(pkt_len ==
14574 		    ntohs(((ip6_t *)mp->b_rptr)->ip6_plen) + IPV6_HDR_LEN);
14575 		if (HOOKS6_INTERESTED_PHYSICAL_OUT(ipst) &&
14576 		    !(ixaflags & IXAF_NO_PFHOOK)) {
14577 			int	error;
14578 
14579 			FW_HOOKS6(ipst->ips_ip6_physical_out_event,
14580 			    ipst->ips_ipv6firewall_physical_out,
14581 			    NULL, ill, ip6h, mp, mp, 0, ipst, error);
14582 			DTRACE_PROBE1(ip6__physical__out__end,
14583 			    mblk_t *, mp);
14584 			if (mp == NULL)
14585 				return (error);
14586 
14587 			/* The length could have changed */
14588 			pkt_len = msgdsize(mp);
14589 		}
14590 		if (ipst->ips_ip6_observe.he_interested) {
14591 			/* See above */
14592 			szone = IP_REAL_ZONEID(szone, ipst);
14593 
14594 			ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14595 			    ill, ipst);
14596 		}
14597 		DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14598 		    void_ip_t *, ip6h,  __dtrace_ipsr_ill_t *, ill,
14599 		    ipha_t *, NULL, ip6_t *, ip6h, int, 0);
14600 	}
14601 
14602 sendit:
14603 	/*
14604 	 * We check the state without a lock because the state can never
14605 	 * move "backwards" to initial or incomplete.
14606 	 */
14607 	switch (ncec->ncec_state) {
14608 	case ND_REACHABLE:
14609 	case ND_STALE:
14610 	case ND_DELAY:
14611 	case ND_PROBE:
14612 		mp = ip_xmit_attach_llhdr(mp, nce);
14613 		if (mp == NULL) {
14614 			/*
14615 			 * ip_xmit_attach_llhdr has increased
14616 			 * ipIfStatsOutDiscards and called ip_drop_output()
14617 			 */
14618 			return (ENOBUFS);
14619 		}
14620 		/*
14621 		 * check if nce_fastpath completed and we tagged on a
14622 		 * copy of nce_fp_mp in ip_xmit_attach_llhdr().
14623 		 */
14624 		fp_mp = (mp->b_datap->db_type == M_DATA);
14625 
14626 		if (fp_mp &&
14627 		    (ill->ill_capabilities & ILL_CAPAB_DLD_DIRECT)) {
14628 			ill_dld_direct_t *idd;
14629 
14630 			idd = &ill->ill_dld_capab->idc_direct;
14631 			/*
14632 			 * Send the packet directly to DLD, where it
14633 			 * may be queued depending on the availability
14634 			 * of transmit resources at the media layer.
14635 			 * Return value should be taken into
14636 			 * account and flow control the TCP.
14637 			 */
14638 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14639 			UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14640 			    pkt_len);
14641 
14642 			if (ixaflags & IXAF_NO_DEV_FLOW_CTL) {
14643 				(void) idd->idd_tx_df(idd->idd_tx_dh, mp,
14644 				    (uintptr_t)xmit_hint, IP_DROP_ON_NO_DESC);
14645 			} else {
14646 				uintptr_t cookie;
14647 
14648 				if ((cookie = idd->idd_tx_df(idd->idd_tx_dh,
14649 				    mp, (uintptr_t)xmit_hint, 0)) != 0) {
14650 					if (ixacookie != NULL)
14651 						*ixacookie = cookie;
14652 					return (EWOULDBLOCK);
14653 				}
14654 			}
14655 		} else {
14656 			wq = ill->ill_wq;
14657 
14658 			if (!(ixaflags & IXAF_NO_DEV_FLOW_CTL) &&
14659 			    !canputnext(wq)) {
14660 				if (ixacookie != NULL)
14661 					*ixacookie = 0;
14662 				ip_xmit_flowctl_drop(ill, mp, fp_mp,
14663 				    nce->nce_fp_mp != NULL ?
14664 				    MBLKL(nce->nce_fp_mp) : 0);
14665 				return (EWOULDBLOCK);
14666 			}
14667 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14668 			UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14669 			    pkt_len);
14670 			putnext(wq, mp);
14671 		}
14672 
14673 		/*
14674 		 * The rest of this function implements Neighbor Unreachability
14675 		 * detection. Determine if the ncec is eligible for NUD.
14676 		 */
14677 		if (ncec->ncec_flags & NCE_F_NONUD)
14678 			return (0);
14679 
14680 		ASSERT(ncec->ncec_state != ND_INCOMPLETE);
14681 
14682 		/*
14683 		 * Check for upper layer advice
14684 		 */
14685 		if (ixaflags & IXAF_REACH_CONF) {
14686 			timeout_id_t tid;
14687 
14688 			/*
14689 			 * It should be o.k. to check the state without
14690 			 * a lock here, at most we lose an advice.
14691 			 */
14692 			ncec->ncec_last = TICK_TO_MSEC(now);
14693 			if (ncec->ncec_state != ND_REACHABLE) {
14694 				mutex_enter(&ncec->ncec_lock);
14695 				ncec->ncec_state = ND_REACHABLE;
14696 				tid = ncec->ncec_timeout_id;
14697 				ncec->ncec_timeout_id = 0;
14698 				mutex_exit(&ncec->ncec_lock);
14699 				(void) untimeout(tid);
14700 				if (ip_debug > 2) {
14701 					/* ip1dbg */
14702 					pr_addr_dbg("ip_xmit: state"
14703 					    " for %s changed to"
14704 					    " REACHABLE\n", AF_INET6,
14705 					    &ncec->ncec_addr);
14706 				}
14707 			}
14708 			return (0);
14709 		}
14710 
14711 		delta =  TICK_TO_MSEC(now) - ncec->ncec_last;
14712 		ip1dbg(("ip_xmit: delta = %" PRId64
14713 		    " ill_reachable_time = %d \n", delta,
14714 		    ill->ill_reachable_time));
14715 		if (delta > (uint64_t)ill->ill_reachable_time) {
14716 			mutex_enter(&ncec->ncec_lock);
14717 			switch (ncec->ncec_state) {
14718 			case ND_REACHABLE:
14719 				ASSERT((ncec->ncec_flags & NCE_F_NONUD) == 0);
14720 				/* FALLTHROUGH */
14721 			case ND_STALE:
14722 				/*
14723 				 * ND_REACHABLE is identical to
14724 				 * ND_STALE in this specific case. If
14725 				 * reachable time has expired for this
14726 				 * neighbor (delta is greater than
14727 				 * reachable time), conceptually, the
14728 				 * neighbor cache is no longer in
14729 				 * REACHABLE state, but already in
14730 				 * STALE state.  So the correct
14731 				 * transition here is to ND_DELAY.
14732 				 */
14733 				ncec->ncec_state = ND_DELAY;
14734 				mutex_exit(&ncec->ncec_lock);
14735 				nce_restart_timer(ncec,
14736 				    ipst->ips_delay_first_probe_time);
14737 				if (ip_debug > 3) {
14738 					/* ip2dbg */
14739 					pr_addr_dbg("ip_xmit: state"
14740 					    " for %s changed to"
14741 					    " DELAY\n", AF_INET6,
14742 					    &ncec->ncec_addr);
14743 				}
14744 				break;
14745 			case ND_DELAY:
14746 			case ND_PROBE:
14747 				mutex_exit(&ncec->ncec_lock);
14748 				/* Timers have already started */
14749 				break;
14750 			case ND_UNREACHABLE:
14751 				/*
14752 				 * nce_timer has detected that this ncec
14753 				 * is unreachable and initiated deleting
14754 				 * this ncec.
14755 				 * This is a harmless race where we found the
14756 				 * ncec before it was deleted and have
14757 				 * just sent out a packet using this
14758 				 * unreachable ncec.
14759 				 */
14760 				mutex_exit(&ncec->ncec_lock);
14761 				break;
14762 			default:
14763 				ASSERT(0);
14764 				mutex_exit(&ncec->ncec_lock);
14765 			}
14766 		}
14767 		return (0);
14768 
14769 	case ND_INCOMPLETE:
14770 		/*
14771 		 * the state could have changed since we didn't hold the lock.
14772 		 * Re-verify state under lock.
14773 		 */
14774 		is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14775 		mutex_enter(&ncec->ncec_lock);
14776 		if (NCE_ISREACHABLE(ncec)) {
14777 			mutex_exit(&ncec->ncec_lock);
14778 			goto sendit;
14779 		}
14780 		/* queue the packet */
14781 		nce_queue_mp(ncec, mp, is_probe);
14782 		mutex_exit(&ncec->ncec_lock);
14783 		DTRACE_PROBE2(ip__xmit__incomplete,
14784 		    (ncec_t *), ncec, (mblk_t *), mp);
14785 		return (0);
14786 
14787 	case ND_INITIAL:
14788 		/*
14789 		 * State could have changed since we didn't hold the lock, so
14790 		 * re-verify state.
14791 		 */
14792 		is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14793 		mutex_enter(&ncec->ncec_lock);
14794 		if (NCE_ISREACHABLE(ncec))  {
14795 			mutex_exit(&ncec->ncec_lock);
14796 			goto sendit;
14797 		}
14798 		nce_queue_mp(ncec, mp, is_probe);
14799 		if (ncec->ncec_state == ND_INITIAL) {
14800 			ncec->ncec_state = ND_INCOMPLETE;
14801 			mutex_exit(&ncec->ncec_lock);
14802 			/*
14803 			 * figure out the source we want to use
14804 			 * and resolve it.
14805 			 */
14806 			ip_ndp_resolve(ncec);
14807 		} else  {
14808 			mutex_exit(&ncec->ncec_lock);
14809 		}
14810 		return (0);
14811 
14812 	case ND_UNREACHABLE:
14813 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14814 		ip_drop_output("ipIfStatsOutDiscards - ND_UNREACHABLE",
14815 		    mp, ill);
14816 		freemsg(mp);
14817 		return (0);
14818 
14819 	default:
14820 		ASSERT(0);
14821 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14822 		ip_drop_output("ipIfStatsOutDiscards - ND_other",
14823 		    mp, ill);
14824 		freemsg(mp);
14825 		return (ENETUNREACH);
14826 	}
14827 }
14828 
14829 /*
14830  * Return B_TRUE if the buffers differ in length or content.
14831  * This is used for comparing extension header buffers.
14832  * Note that an extension header would be declared different
14833  * even if all that changed was the next header value in that header i.e.
14834  * what really changed is the next extension header.
14835  */
14836 boolean_t
ip_cmpbuf(const void * abuf,uint_t alen,boolean_t b_valid,const void * bbuf,uint_t blen)14837 ip_cmpbuf(const void *abuf, uint_t alen, boolean_t b_valid, const void *bbuf,
14838     uint_t blen)
14839 {
14840 	if (!b_valid)
14841 		blen = 0;
14842 
14843 	if (alen != blen)
14844 		return (B_TRUE);
14845 	if (alen == 0)
14846 		return (B_FALSE);	/* Both zero length */
14847 	return (bcmp(abuf, bbuf, alen));
14848 }
14849 
14850 /*
14851  * Preallocate memory for ip_savebuf(). Returns B_TRUE if ok.
14852  * Return B_FALSE if memory allocation fails - don't change any state!
14853  */
14854 boolean_t
ip_allocbuf(void ** dstp,uint_t * dstlenp,boolean_t src_valid,const void * src,uint_t srclen)14855 ip_allocbuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14856     const void *src, uint_t srclen)
14857 {
14858 	void *dst;
14859 
14860 	if (!src_valid)
14861 		srclen = 0;
14862 
14863 	ASSERT(*dstlenp == 0);
14864 	if (src != NULL && srclen != 0) {
14865 		dst = mi_alloc(srclen, BPRI_MED);
14866 		if (dst == NULL)
14867 			return (B_FALSE);
14868 	} else {
14869 		dst = NULL;
14870 	}
14871 	if (*dstp != NULL)
14872 		mi_free(*dstp);
14873 	*dstp = dst;
14874 	*dstlenp = dst == NULL ? 0 : srclen;
14875 	return (B_TRUE);
14876 }
14877 
14878 /*
14879  * Replace what is in *dst, *dstlen with the source.
14880  * Assumes ip_allocbuf has already been called.
14881  */
14882 void
ip_savebuf(void ** dstp,uint_t * dstlenp,boolean_t src_valid,const void * src,uint_t srclen)14883 ip_savebuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14884     const void *src, uint_t srclen)
14885 {
14886 	if (!src_valid)
14887 		srclen = 0;
14888 
14889 	ASSERT(*dstlenp == srclen);
14890 	if (src != NULL && srclen != 0)
14891 		bcopy(src, *dstp, srclen);
14892 }
14893 
14894 /*
14895  * Free the storage pointed to by the members of an ip_pkt_t.
14896  */
14897 void
ip_pkt_free(ip_pkt_t * ipp)14898 ip_pkt_free(ip_pkt_t *ipp)
14899 {
14900 	uint_t	fields = ipp->ipp_fields;
14901 
14902 	if (fields & IPPF_HOPOPTS) {
14903 		kmem_free(ipp->ipp_hopopts, ipp->ipp_hopoptslen);
14904 		ipp->ipp_hopopts = NULL;
14905 		ipp->ipp_hopoptslen = 0;
14906 	}
14907 	if (fields & IPPF_RTHDRDSTOPTS) {
14908 		kmem_free(ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen);
14909 		ipp->ipp_rthdrdstopts = NULL;
14910 		ipp->ipp_rthdrdstoptslen = 0;
14911 	}
14912 	if (fields & IPPF_DSTOPTS) {
14913 		kmem_free(ipp->ipp_dstopts, ipp->ipp_dstoptslen);
14914 		ipp->ipp_dstopts = NULL;
14915 		ipp->ipp_dstoptslen = 0;
14916 	}
14917 	if (fields & IPPF_RTHDR) {
14918 		kmem_free(ipp->ipp_rthdr, ipp->ipp_rthdrlen);
14919 		ipp->ipp_rthdr = NULL;
14920 		ipp->ipp_rthdrlen = 0;
14921 	}
14922 	if (fields & IPPF_IPV4_OPTIONS) {
14923 		kmem_free(ipp->ipp_ipv4_options, ipp->ipp_ipv4_options_len);
14924 		ipp->ipp_ipv4_options = NULL;
14925 		ipp->ipp_ipv4_options_len = 0;
14926 	}
14927 	if (fields & IPPF_LABEL_V4) {
14928 		kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
14929 		ipp->ipp_label_v4 = NULL;
14930 		ipp->ipp_label_len_v4 = 0;
14931 	}
14932 	if (fields & IPPF_LABEL_V6) {
14933 		kmem_free(ipp->ipp_label_v6, ipp->ipp_label_len_v6);
14934 		ipp->ipp_label_v6 = NULL;
14935 		ipp->ipp_label_len_v6 = 0;
14936 	}
14937 	ipp->ipp_fields &= ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14938 	    IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14939 }
14940 
14941 /*
14942  * Copy from src to dst and allocate as needed.
14943  * Returns zero or ENOMEM.
14944  *
14945  * The caller must initialize dst to zero.
14946  */
14947 int
ip_pkt_copy(ip_pkt_t * src,ip_pkt_t * dst,int kmflag)14948 ip_pkt_copy(ip_pkt_t *src, ip_pkt_t *dst, int kmflag)
14949 {
14950 	uint_t	fields = src->ipp_fields;
14951 
14952 	/* Start with fields that don't require memory allocation */
14953 	dst->ipp_fields = fields &
14954 	    ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14955 	    IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14956 
14957 	dst->ipp_addr = src->ipp_addr;
14958 	dst->ipp_unicast_hops = src->ipp_unicast_hops;
14959 	dst->ipp_hoplimit = src->ipp_hoplimit;
14960 	dst->ipp_tclass = src->ipp_tclass;
14961 	dst->ipp_type_of_service = src->ipp_type_of_service;
14962 
14963 	if (!(fields & (IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14964 	    IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6)))
14965 		return (0);
14966 
14967 	if (fields & IPPF_HOPOPTS) {
14968 		dst->ipp_hopopts = kmem_alloc(src->ipp_hopoptslen, kmflag);
14969 		if (dst->ipp_hopopts == NULL) {
14970 			ip_pkt_free(dst);
14971 			return (ENOMEM);
14972 		}
14973 		dst->ipp_fields |= IPPF_HOPOPTS;
14974 		bcopy(src->ipp_hopopts, dst->ipp_hopopts,
14975 		    src->ipp_hopoptslen);
14976 		dst->ipp_hopoptslen = src->ipp_hopoptslen;
14977 	}
14978 	if (fields & IPPF_RTHDRDSTOPTS) {
14979 		dst->ipp_rthdrdstopts = kmem_alloc(src->ipp_rthdrdstoptslen,
14980 		    kmflag);
14981 		if (dst->ipp_rthdrdstopts == NULL) {
14982 			ip_pkt_free(dst);
14983 			return (ENOMEM);
14984 		}
14985 		dst->ipp_fields |= IPPF_RTHDRDSTOPTS;
14986 		bcopy(src->ipp_rthdrdstopts, dst->ipp_rthdrdstopts,
14987 		    src->ipp_rthdrdstoptslen);
14988 		dst->ipp_rthdrdstoptslen = src->ipp_rthdrdstoptslen;
14989 	}
14990 	if (fields & IPPF_DSTOPTS) {
14991 		dst->ipp_dstopts = kmem_alloc(src->ipp_dstoptslen, kmflag);
14992 		if (dst->ipp_dstopts == NULL) {
14993 			ip_pkt_free(dst);
14994 			return (ENOMEM);
14995 		}
14996 		dst->ipp_fields |= IPPF_DSTOPTS;
14997 		bcopy(src->ipp_dstopts, dst->ipp_dstopts,
14998 		    src->ipp_dstoptslen);
14999 		dst->ipp_dstoptslen = src->ipp_dstoptslen;
15000 	}
15001 	if (fields & IPPF_RTHDR) {
15002 		dst->ipp_rthdr = kmem_alloc(src->ipp_rthdrlen, kmflag);
15003 		if (dst->ipp_rthdr == NULL) {
15004 			ip_pkt_free(dst);
15005 			return (ENOMEM);
15006 		}
15007 		dst->ipp_fields |= IPPF_RTHDR;
15008 		bcopy(src->ipp_rthdr, dst->ipp_rthdr,
15009 		    src->ipp_rthdrlen);
15010 		dst->ipp_rthdrlen = src->ipp_rthdrlen;
15011 	}
15012 	if (fields & IPPF_IPV4_OPTIONS) {
15013 		dst->ipp_ipv4_options = kmem_alloc(src->ipp_ipv4_options_len,
15014 		    kmflag);
15015 		if (dst->ipp_ipv4_options == NULL) {
15016 			ip_pkt_free(dst);
15017 			return (ENOMEM);
15018 		}
15019 		dst->ipp_fields |= IPPF_IPV4_OPTIONS;
15020 		bcopy(src->ipp_ipv4_options, dst->ipp_ipv4_options,
15021 		    src->ipp_ipv4_options_len);
15022 		dst->ipp_ipv4_options_len = src->ipp_ipv4_options_len;
15023 	}
15024 	if (fields & IPPF_LABEL_V4) {
15025 		dst->ipp_label_v4 = kmem_alloc(src->ipp_label_len_v4, kmflag);
15026 		if (dst->ipp_label_v4 == NULL) {
15027 			ip_pkt_free(dst);
15028 			return (ENOMEM);
15029 		}
15030 		dst->ipp_fields |= IPPF_LABEL_V4;
15031 		bcopy(src->ipp_label_v4, dst->ipp_label_v4,
15032 		    src->ipp_label_len_v4);
15033 		dst->ipp_label_len_v4 = src->ipp_label_len_v4;
15034 	}
15035 	if (fields & IPPF_LABEL_V6) {
15036 		dst->ipp_label_v6 = kmem_alloc(src->ipp_label_len_v6, kmflag);
15037 		if (dst->ipp_label_v6 == NULL) {
15038 			ip_pkt_free(dst);
15039 			return (ENOMEM);
15040 		}
15041 		dst->ipp_fields |= IPPF_LABEL_V6;
15042 		bcopy(src->ipp_label_v6, dst->ipp_label_v6,
15043 		    src->ipp_label_len_v6);
15044 		dst->ipp_label_len_v6 = src->ipp_label_len_v6;
15045 	}
15046 	if (fields & IPPF_FRAGHDR) {
15047 		dst->ipp_fraghdr = kmem_alloc(src->ipp_fraghdrlen, kmflag);
15048 		if (dst->ipp_fraghdr == NULL) {
15049 			ip_pkt_free(dst);
15050 			return (ENOMEM);
15051 		}
15052 		dst->ipp_fields |= IPPF_FRAGHDR;
15053 		bcopy(src->ipp_fraghdr, dst->ipp_fraghdr,
15054 		    src->ipp_fraghdrlen);
15055 		dst->ipp_fraghdrlen = src->ipp_fraghdrlen;
15056 	}
15057 	return (0);
15058 }
15059 
15060 /*
15061  * Returns INADDR_ANY if no source route
15062  */
15063 ipaddr_t
ip_pkt_source_route_v4(const ip_pkt_t * ipp)15064 ip_pkt_source_route_v4(const ip_pkt_t *ipp)
15065 {
15066 	ipaddr_t	nexthop = INADDR_ANY;
15067 	ipoptp_t	opts;
15068 	uchar_t		*opt;
15069 	uint8_t		optval;
15070 	uint8_t		optlen;
15071 	uint32_t	totallen;
15072 
15073 	if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
15074 		return (INADDR_ANY);
15075 
15076 	totallen = ipp->ipp_ipv4_options_len;
15077 	if (totallen & 0x3)
15078 		return (INADDR_ANY);
15079 
15080 	for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15081 	    optval != IPOPT_EOL;
15082 	    optval = ipoptp_next(&opts)) {
15083 		opt = opts.ipoptp_cur;
15084 		switch (optval) {
15085 			uint8_t off;
15086 		case IPOPT_SSRR:
15087 		case IPOPT_LSRR:
15088 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15089 				break;
15090 			}
15091 			optlen = opts.ipoptp_len;
15092 			off = opt[IPOPT_OFFSET];
15093 			off--;
15094 			if (optlen < IP_ADDR_LEN ||
15095 			    off > optlen - IP_ADDR_LEN) {
15096 				/* End of source route */
15097 				break;
15098 			}
15099 			bcopy((char *)opt + off, &nexthop, IP_ADDR_LEN);
15100 			if (nexthop == htonl(INADDR_LOOPBACK)) {
15101 				/* Ignore */
15102 				nexthop = INADDR_ANY;
15103 				break;
15104 			}
15105 			break;
15106 		}
15107 	}
15108 	return (nexthop);
15109 }
15110 
15111 /*
15112  * Reverse a source route.
15113  */
15114 void
ip_pkt_source_route_reverse_v4(ip_pkt_t * ipp)15115 ip_pkt_source_route_reverse_v4(ip_pkt_t *ipp)
15116 {
15117 	ipaddr_t	tmp;
15118 	ipoptp_t	opts;
15119 	uchar_t		*opt;
15120 	uint8_t		optval;
15121 	uint32_t	totallen;
15122 
15123 	if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
15124 		return;
15125 
15126 	totallen = ipp->ipp_ipv4_options_len;
15127 	if (totallen & 0x3)
15128 		return;
15129 
15130 	for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15131 	    optval != IPOPT_EOL;
15132 	    optval = ipoptp_next(&opts)) {
15133 		uint8_t off1, off2;
15134 
15135 		opt = opts.ipoptp_cur;
15136 		switch (optval) {
15137 		case IPOPT_SSRR:
15138 		case IPOPT_LSRR:
15139 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15140 				break;
15141 			}
15142 			off1 = IPOPT_MINOFF_SR - 1;
15143 			off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
15144 			while (off2 > off1) {
15145 				bcopy(opt + off2, &tmp, IP_ADDR_LEN);
15146 				bcopy(opt + off1, opt + off2, IP_ADDR_LEN);
15147 				bcopy(&tmp, opt + off2, IP_ADDR_LEN);
15148 				off2 -= IP_ADDR_LEN;
15149 				off1 += IP_ADDR_LEN;
15150 			}
15151 			opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
15152 			break;
15153 		}
15154 	}
15155 }
15156 
15157 /*
15158  * Returns NULL if no routing header
15159  */
15160 in6_addr_t *
ip_pkt_source_route_v6(const ip_pkt_t * ipp)15161 ip_pkt_source_route_v6(const ip_pkt_t *ipp)
15162 {
15163 	in6_addr_t	*nexthop = NULL;
15164 	ip6_rthdr0_t	*rthdr;
15165 
15166 	if (!(ipp->ipp_fields & IPPF_RTHDR))
15167 		return (NULL);
15168 
15169 	rthdr = (ip6_rthdr0_t *)ipp->ipp_rthdr;
15170 	if (rthdr->ip6r0_segleft == 0)
15171 		return (NULL);
15172 
15173 	nexthop = (in6_addr_t *)((char *)rthdr + sizeof (*rthdr));
15174 	return (nexthop);
15175 }
15176 
15177 zoneid_t
ip_get_zoneid_v4(ipaddr_t addr,mblk_t * mp,ip_recv_attr_t * ira,zoneid_t lookup_zoneid)15178 ip_get_zoneid_v4(ipaddr_t addr, mblk_t *mp, ip_recv_attr_t *ira,
15179     zoneid_t lookup_zoneid)
15180 {
15181 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
15182 	ire_t		*ire;
15183 	int		ire_flags = MATCH_IRE_TYPE;
15184 	zoneid_t	zoneid = ALL_ZONES;
15185 
15186 	if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15187 		return (ALL_ZONES);
15188 
15189 	if (lookup_zoneid != ALL_ZONES)
15190 		ire_flags |= MATCH_IRE_ZONEONLY;
15191 	ire = ire_ftable_lookup_v4(addr, 0, 0, IRE_LOCAL | IRE_LOOPBACK,
15192 	    NULL, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15193 	if (ire != NULL) {
15194 		zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15195 		ire_refrele(ire);
15196 	}
15197 	return (zoneid);
15198 }
15199 
15200 zoneid_t
ip_get_zoneid_v6(in6_addr_t * addr,mblk_t * mp,const ill_t * ill,ip_recv_attr_t * ira,zoneid_t lookup_zoneid)15201 ip_get_zoneid_v6(in6_addr_t *addr, mblk_t *mp, const ill_t *ill,
15202     ip_recv_attr_t *ira, zoneid_t lookup_zoneid)
15203 {
15204 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
15205 	ire_t		*ire;
15206 	int		ire_flags = MATCH_IRE_TYPE;
15207 	zoneid_t	zoneid = ALL_ZONES;
15208 
15209 	if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15210 		return (ALL_ZONES);
15211 
15212 	if (IN6_IS_ADDR_LINKLOCAL(addr))
15213 		ire_flags |= MATCH_IRE_ILL;
15214 
15215 	if (lookup_zoneid != ALL_ZONES)
15216 		ire_flags |= MATCH_IRE_ZONEONLY;
15217 	ire = ire_ftable_lookup_v6(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15218 	    ill, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15219 	if (ire != NULL) {
15220 		zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15221 		ire_refrele(ire);
15222 	}
15223 	return (zoneid);
15224 }
15225 
15226 /*
15227  * IP obserability hook support functions.
15228  */
15229 static void
ipobs_init(ip_stack_t * ipst)15230 ipobs_init(ip_stack_t *ipst)
15231 {
15232 	netid_t id;
15233 
15234 	id = net_getnetidbynetstackid(ipst->ips_netstack->netstack_stackid);
15235 
15236 	ipst->ips_ip4_observe_pr = net_protocol_lookup(id, NHF_INET);
15237 	VERIFY(ipst->ips_ip4_observe_pr != NULL);
15238 
15239 	ipst->ips_ip6_observe_pr = net_protocol_lookup(id, NHF_INET6);
15240 	VERIFY(ipst->ips_ip6_observe_pr != NULL);
15241 }
15242 
15243 static void
ipobs_fini(ip_stack_t * ipst)15244 ipobs_fini(ip_stack_t *ipst)
15245 {
15246 
15247 	VERIFY(net_protocol_release(ipst->ips_ip4_observe_pr) == 0);
15248 	VERIFY(net_protocol_release(ipst->ips_ip6_observe_pr) == 0);
15249 }
15250 
15251 /*
15252  * hook_pkt_observe_t is composed in network byte order so that the
15253  * entire mblk_t chain handed into hook_run can be used as-is.
15254  * The caveat is that use of the fields, such as the zone fields,
15255  * requires conversion into host byte order first.
15256  */
15257 void
ipobs_hook(mblk_t * mp,int htype,zoneid_t zsrc,zoneid_t zdst,const ill_t * ill,ip_stack_t * ipst)15258 ipobs_hook(mblk_t *mp, int htype, zoneid_t zsrc, zoneid_t zdst,
15259     const ill_t *ill, ip_stack_t *ipst)
15260 {
15261 	hook_pkt_observe_t *hdr;
15262 	uint64_t grifindex;
15263 	mblk_t *imp;
15264 
15265 	imp = allocb(sizeof (*hdr), BPRI_HI);
15266 	if (imp == NULL)
15267 		return;
15268 
15269 	hdr = (hook_pkt_observe_t *)imp->b_rptr;
15270 	/*
15271 	 * b_wptr is set to make the apparent size of the data in the mblk_t
15272 	 * to exclude the pointers at the end of hook_pkt_observer_t.
15273 	 */
15274 	imp->b_wptr = imp->b_rptr + sizeof (dl_ipnetinfo_t);
15275 	imp->b_cont = mp;
15276 
15277 	ASSERT(DB_TYPE(mp) == M_DATA);
15278 
15279 	if (IS_UNDER_IPMP(ill))
15280 		grifindex = ipmp_ill_get_ipmp_ifindex(ill);
15281 	else
15282 		grifindex = 0;
15283 
15284 	hdr->hpo_version = 1;
15285 	hdr->hpo_htype = htons(htype);
15286 	hdr->hpo_pktlen = htonl((ulong_t)msgdsize(mp));
15287 	hdr->hpo_ifindex = htonl(ill->ill_phyint->phyint_ifindex);
15288 	hdr->hpo_grifindex = htonl(grifindex);
15289 	hdr->hpo_zsrc = htonl(zsrc);
15290 	hdr->hpo_zdst = htonl(zdst);
15291 	hdr->hpo_pkt = imp;
15292 	hdr->hpo_ctx = ipst->ips_netstack;
15293 
15294 	if (ill->ill_isv6) {
15295 		hdr->hpo_family = AF_INET6;
15296 		(void) hook_run(ipst->ips_ipv6_net_data->netd_hooks,
15297 		    ipst->ips_ipv6observing, (hook_data_t)hdr);
15298 	} else {
15299 		hdr->hpo_family = AF_INET;
15300 		(void) hook_run(ipst->ips_ipv4_net_data->netd_hooks,
15301 		    ipst->ips_ipv4observing, (hook_data_t)hdr);
15302 	}
15303 
15304 	imp->b_cont = NULL;
15305 	freemsg(imp);
15306 }
15307 
15308 /*
15309  * Utility routine that checks if `v4srcp' is a valid address on underlying
15310  * interface `ill'.  If `ipifp' is non-NULL, it's set to a held ipif
15311  * associated with `v4srcp' on success.  NOTE: if this is not called from
15312  * inside the IPSQ (ill_g_lock is not held), `ill' may be removed from the
15313  * group during or after this lookup.
15314  */
15315 boolean_t
ipif_lookup_testaddr_v4(ill_t * ill,const in_addr_t * v4srcp,ipif_t ** ipifp)15316 ipif_lookup_testaddr_v4(ill_t *ill, const in_addr_t *v4srcp, ipif_t **ipifp)
15317 {
15318 	ipif_t *ipif;
15319 
15320 	ipif = ipif_lookup_addr_exact(*v4srcp, ill, ill->ill_ipst);
15321 	if (ipif != NULL) {
15322 		if (ipifp != NULL)
15323 			*ipifp = ipif;
15324 		else
15325 			ipif_refrele(ipif);
15326 		return (B_TRUE);
15327 	}
15328 
15329 	ip1dbg(("ipif_lookup_testaddr_v4: cannot find ipif for src %x\n",
15330 	    *v4srcp));
15331 	return (B_FALSE);
15332 }
15333 
15334 /*
15335  * Transport protocol call back function for CPU state change.
15336  */
15337 /* ARGSUSED */
15338 static int
ip_tp_cpu_update(cpu_setup_t what,int id,void * arg)15339 ip_tp_cpu_update(cpu_setup_t what, int id, void *arg)
15340 {
15341 	processorid_t cpu_seqid;
15342 	netstack_handle_t nh;
15343 	netstack_t *ns;
15344 
15345 	ASSERT(MUTEX_HELD(&cpu_lock));
15346 
15347 	switch (what) {
15348 	case CPU_CONFIG:
15349 	case CPU_ON:
15350 	case CPU_INIT:
15351 	case CPU_CPUPART_IN:
15352 		cpu_seqid = cpu[id]->cpu_seqid;
15353 		netstack_next_init(&nh);
15354 		while ((ns = netstack_next(&nh)) != NULL) {
15355 			tcp_stack_cpu_add(ns->netstack_tcp, cpu_seqid);
15356 			sctp_stack_cpu_add(ns->netstack_sctp, cpu_seqid);
15357 			udp_stack_cpu_add(ns->netstack_udp, cpu_seqid);
15358 			netstack_rele(ns);
15359 		}
15360 		netstack_next_fini(&nh);
15361 		break;
15362 	case CPU_UNCONFIG:
15363 	case CPU_OFF:
15364 	case CPU_CPUPART_OUT:
15365 		/*
15366 		 * Nothing to do.  We don't remove the per CPU stats from
15367 		 * the IP stack even when the CPU goes offline.
15368 		 */
15369 		break;
15370 	default:
15371 		break;
15372 	}
15373 	return (0);
15374 }
15375