xref: /titanic_41/usr/src/uts/common/inet/ip/ip.c (revision 6df9eb5dad3c727a3f2973d0b83af404cc4f5423)
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) 2012 Joyent, Inc. All rights reserved.
26  * Copyright (c) 2014, OmniTI Computer Consulting, Inc. All rights reserved.
27  */
28 
29 #include <sys/types.h>
30 #include <sys/stream.h>
31 #include <sys/dlpi.h>
32 #include <sys/stropts.h>
33 #include <sys/sysmacros.h>
34 #include <sys/strsubr.h>
35 #include <sys/strlog.h>
36 #include <sys/strsun.h>
37 #include <sys/zone.h>
38 #define	_SUN_TPI_VERSION 2
39 #include <sys/tihdr.h>
40 #include <sys/xti_inet.h>
41 #include <sys/ddi.h>
42 #include <sys/suntpi.h>
43 #include <sys/cmn_err.h>
44 #include <sys/debug.h>
45 #include <sys/kobj.h>
46 #include <sys/modctl.h>
47 #include <sys/atomic.h>
48 #include <sys/policy.h>
49 #include <sys/priv.h>
50 #include <sys/taskq.h>
51 
52 #include <sys/systm.h>
53 #include <sys/param.h>
54 #include <sys/kmem.h>
55 #include <sys/sdt.h>
56 #include <sys/socket.h>
57 #include <sys/vtrace.h>
58 #include <sys/isa_defs.h>
59 #include <sys/mac.h>
60 #include <net/if.h>
61 #include <net/if_arp.h>
62 #include <net/route.h>
63 #include <sys/sockio.h>
64 #include <netinet/in.h>
65 #include <net/if_dl.h>
66 
67 #include <inet/common.h>
68 #include <inet/mi.h>
69 #include <inet/mib2.h>
70 #include <inet/nd.h>
71 #include <inet/arp.h>
72 #include <inet/snmpcom.h>
73 #include <inet/optcom.h>
74 #include <inet/kstatcom.h>
75 
76 #include <netinet/igmp_var.h>
77 #include <netinet/ip6.h>
78 #include <netinet/icmp6.h>
79 #include <netinet/sctp.h>
80 
81 #include <inet/ip.h>
82 #include <inet/ip_impl.h>
83 #include <inet/ip6.h>
84 #include <inet/ip6_asp.h>
85 #include <inet/tcp.h>
86 #include <inet/tcp_impl.h>
87 #include <inet/ip_multi.h>
88 #include <inet/ip_if.h>
89 #include <inet/ip_ire.h>
90 #include <inet/ip_ftable.h>
91 #include <inet/ip_rts.h>
92 #include <inet/ip_ndp.h>
93 #include <inet/ip_listutils.h>
94 #include <netinet/igmp.h>
95 #include <netinet/ip_mroute.h>
96 #include <inet/ipp_common.h>
97 
98 #include <net/pfkeyv2.h>
99 #include <inet/sadb.h>
100 #include <inet/ipsec_impl.h>
101 #include <inet/iptun/iptun_impl.h>
102 #include <inet/ipdrop.h>
103 #include <inet/ip_netinfo.h>
104 #include <inet/ilb_ip.h>
105 
106 #include <sys/ethernet.h>
107 #include <net/if_types.h>
108 #include <sys/cpuvar.h>
109 
110 #include <ipp/ipp.h>
111 #include <ipp/ipp_impl.h>
112 #include <ipp/ipgpc/ipgpc.h>
113 
114 #include <sys/pattr.h>
115 #include <inet/ipclassifier.h>
116 #include <inet/sctp_ip.h>
117 #include <inet/sctp/sctp_impl.h>
118 #include <inet/udp_impl.h>
119 #include <inet/rawip_impl.h>
120 #include <inet/rts_impl.h>
121 
122 #include <sys/tsol/label.h>
123 #include <sys/tsol/tnet.h>
124 
125 #include <sys/squeue_impl.h>
126 #include <inet/ip_arp.h>
127 
128 #include <sys/clock_impl.h>	/* For LBOLT_FASTPATH{,64} */
129 
130 /*
131  * Values for squeue switch:
132  * IP_SQUEUE_ENTER_NODRAIN: SQ_NODRAIN
133  * IP_SQUEUE_ENTER: SQ_PROCESS
134  * IP_SQUEUE_FILL: SQ_FILL
135  */
136 int ip_squeue_enter = IP_SQUEUE_ENTER;	/* Setable in /etc/system */
137 
138 int ip_squeue_flag;
139 
140 /*
141  * Setable in /etc/system
142  */
143 int ip_poll_normal_ms = 100;
144 int ip_poll_normal_ticks = 0;
145 int ip_modclose_ackwait_ms = 3000;
146 
147 /*
148  * It would be nice to have these present only in DEBUG systems, but the
149  * current design of the global symbol checking logic requires them to be
150  * unconditionally present.
151  */
152 uint_t ip_thread_data;			/* TSD key for debug support */
153 krwlock_t ip_thread_rwlock;
154 list_t	ip_thread_list;
155 
156 /*
157  * Structure to represent a linked list of msgblks. Used by ip_snmp_ functions.
158  */
159 
160 struct listptr_s {
161 	mblk_t	*lp_head;	/* pointer to the head of the list */
162 	mblk_t	*lp_tail;	/* pointer to the tail of the list */
163 };
164 
165 typedef struct listptr_s listptr_t;
166 
167 /*
168  * This is used by ip_snmp_get_mib2_ip_route_media and
169  * ip_snmp_get_mib2_ip6_route_media to carry the lists of return data.
170  */
171 typedef struct iproutedata_s {
172 	uint_t		ird_idx;
173 	uint_t		ird_flags;	/* see below */
174 	listptr_t	ird_route;	/* ipRouteEntryTable */
175 	listptr_t	ird_netmedia;	/* ipNetToMediaEntryTable */
176 	listptr_t	ird_attrs;	/* ipRouteAttributeTable */
177 } iproutedata_t;
178 
179 /* Include ire_testhidden and IRE_IF_CLONE routes */
180 #define	IRD_REPORT_ALL	0x01
181 
182 /*
183  * Cluster specific hooks. These should be NULL when booted as a non-cluster
184  */
185 
186 /*
187  * Hook functions to enable cluster networking
188  * On non-clustered systems these vectors must always be NULL.
189  *
190  * Hook function to Check ip specified ip address is a shared ip address
191  * in the cluster
192  *
193  */
194 int (*cl_inet_isclusterwide)(netstackid_t stack_id, uint8_t protocol,
195     sa_family_t addr_family, uint8_t *laddrp, void *args) = NULL;
196 
197 /*
198  * Hook function to generate cluster wide ip fragment identifier
199  */
200 uint32_t (*cl_inet_ipident)(netstackid_t stack_id, uint8_t protocol,
201     sa_family_t addr_family, uint8_t *laddrp, uint8_t *faddrp,
202     void *args) = NULL;
203 
204 /*
205  * Hook function to generate cluster wide SPI.
206  */
207 void (*cl_inet_getspi)(netstackid_t, uint8_t, uint8_t *, size_t,
208     void *) = NULL;
209 
210 /*
211  * Hook function to verify if the SPI is already utlized.
212  */
213 
214 int (*cl_inet_checkspi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
215 
216 /*
217  * Hook function to delete the SPI from the cluster wide repository.
218  */
219 
220 void (*cl_inet_deletespi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
221 
222 /*
223  * Hook function to inform the cluster when packet received on an IDLE SA
224  */
225 
226 void (*cl_inet_idlesa)(netstackid_t, uint8_t, uint32_t, sa_family_t,
227     in6_addr_t, in6_addr_t, void *) = NULL;
228 
229 /*
230  * Synchronization notes:
231  *
232  * IP is a fully D_MP STREAMS module/driver. Thus it does not depend on any
233  * MT level protection given by STREAMS. IP uses a combination of its own
234  * internal serialization mechanism and standard Solaris locking techniques.
235  * The internal serialization is per phyint.  This is used to serialize
236  * plumbing operations, IPMP operations, most set ioctls, etc.
237  *
238  * Plumbing is a long sequence of operations involving message
239  * exchanges between IP, ARP and device drivers. Many set ioctls are typically
240  * involved in plumbing operations. A natural model is to serialize these
241  * ioctls one per ill. For example plumbing of hme0 and qfe0 can go on in
242  * parallel without any interference. But various set ioctls on hme0 are best
243  * serialized, along with IPMP operations and processing of DLPI control
244  * messages received from drivers on a per phyint basis. This serialization is
245  * provided by the ipsq_t and primitives operating on this. Details can
246  * be found in ip_if.c above the core primitives operating on ipsq_t.
247  *
248  * Lookups of an ipif or ill by a thread return a refheld ipif / ill.
249  * Simiarly lookup of an ire by a thread also returns a refheld ire.
250  * In addition ipif's and ill's referenced by the ire are also indirectly
251  * refheld. Thus no ipif or ill can vanish as long as an ipif is refheld
252  * directly or indirectly. For example an SIOCSLIFADDR ioctl that changes the
253  * address of an ipif has to go through the ipsq_t. This ensures that only
254  * one such exclusive operation proceeds at any time on the ipif. It then
255  * waits for all refcnts
256  * associated with this ipif to come down to zero. The address is changed
257  * only after the ipif has been quiesced. Then the ipif is brought up again.
258  * More details are described above the comment in ip_sioctl_flags.
259  *
260  * Packet processing is based mostly on IREs and are fully multi-threaded
261  * using standard Solaris MT techniques.
262  *
263  * There are explicit locks in IP to handle:
264  * - The ip_g_head list maintained by mi_open_link() and friends.
265  *
266  * - The reassembly data structures (one lock per hash bucket)
267  *
268  * - conn_lock is meant to protect conn_t fields. The fields actually
269  *   protected by conn_lock are documented in the conn_t definition.
270  *
271  * - ire_lock to protect some of the fields of the ire, IRE tables
272  *   (one lock per hash bucket). Refer to ip_ire.c for details.
273  *
274  * - ndp_g_lock and ncec_lock for protecting NCEs.
275  *
276  * - ill_lock protects fields of the ill and ipif. Details in ip.h
277  *
278  * - ill_g_lock: This is a global reader/writer lock. Protects the following
279  *	* The AVL tree based global multi list of all ills.
280  *	* The linked list of all ipifs of an ill
281  *	* The <ipsq-xop> mapping
282  *	* <ill-phyint> association
283  *   Insertion/deletion of an ill in the system, insertion/deletion of an ipif
284  *   into an ill, changing the <ipsq-xop> mapping of an ill, changing the
285  *   <ill-phyint> assoc of an ill will all have to hold the ill_g_lock as
286  *   writer for the actual duration of the insertion/deletion/change.
287  *
288  * - ill_lock:  This is a per ill mutex.
289  *   It protects some members of the ill_t struct; see ip.h for details.
290  *   It also protects the <ill-phyint> assoc.
291  *   It also protects the list of ipifs hanging off the ill.
292  *
293  * - ipsq_lock: This is a per ipsq_t mutex lock.
294  *   This protects some members of the ipsq_t struct; see ip.h for details.
295  *   It also protects the <ipsq-ipxop> mapping
296  *
297  * - ipx_lock: This is a per ipxop_t mutex lock.
298  *   This protects some members of the ipxop_t struct; see ip.h for details.
299  *
300  * - phyint_lock: This is a per phyint mutex lock. Protects just the
301  *   phyint_flags
302  *
303  * - ip_addr_avail_lock: This is used to ensure the uniqueness of IP addresses.
304  *   This lock is held in ipif_up_done and the ipif is marked IPIF_UP and the
305  *   uniqueness check also done atomically.
306  *
307  * - ill_g_usesrc_lock: This readers/writer lock protects the usesrc
308  *   group list linked by ill_usesrc_grp_next. It also protects the
309  *   ill_usesrc_ifindex field. It is taken as a writer when a member of the
310  *   group is being added or deleted.  This lock is taken as a reader when
311  *   walking the list/group(eg: to get the number of members in a usesrc group).
312  *   Note, it is only necessary to take this lock if the ill_usesrc_grp_next
313  *   field is changing state i.e from NULL to non-NULL or vice-versa. For
314  *   example, it is not necessary to take this lock in the initial portion
315  *   of ip_sioctl_slifusesrc or at all in ip_sioctl_flags since these
316  *   operations are executed exclusively and that ensures that the "usesrc
317  *   group state" cannot change. The "usesrc group state" change can happen
318  *   only in the latter part of ip_sioctl_slifusesrc and in ill_delete.
319  *
320  * Changing <ill-phyint>, <ipsq-xop> assocications:
321  *
322  * To change the <ill-phyint> association, the ill_g_lock must be held
323  * as writer, and the ill_locks of both the v4 and v6 instance of the ill
324  * must be held.
325  *
326  * To change the <ipsq-xop> association, the ill_g_lock must be held as
327  * writer, the ipsq_lock must be held, and one must be writer on the ipsq.
328  * This is only done when ills are added or removed from IPMP groups.
329  *
330  * To add or delete an ipif from the list of ipifs hanging off the ill,
331  * ill_g_lock (writer) and ill_lock must be held and the thread must be
332  * a writer on the associated ipsq.
333  *
334  * To add or delete an ill to the system, the ill_g_lock must be held as
335  * writer and the thread must be a writer on the associated ipsq.
336  *
337  * To add or delete an ilm to an ill, the ill_lock must be held and the thread
338  * must be a writer on the associated ipsq.
339  *
340  * Lock hierarchy
341  *
342  * Some lock hierarchy scenarios are listed below.
343  *
344  * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock -> ipx_lock
345  * ill_g_lock -> ill_lock(s) -> phyint_lock
346  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock
347  * ill_g_lock -> ip_addr_avail_lock
348  * conn_lock -> irb_lock -> ill_lock -> ire_lock
349  * ill_g_lock -> ip_g_nd_lock
350  * ill_g_lock -> ips_ipmp_lock -> ill_lock -> nce_lock
351  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock -> nce_lock
352  * arl_lock -> ill_lock
353  * ips_ire_dep_lock -> irb_lock
354  *
355  * When more than 1 ill lock is needed to be held, all ill lock addresses
356  * are sorted on address and locked starting from highest addressed lock
357  * downward.
358  *
359  * Multicast scenarios
360  * ips_ill_g_lock -> ill_mcast_lock
361  * conn_ilg_lock -> ips_ill_g_lock -> ill_lock
362  * ill_mcast_serializer -> ill_mcast_lock -> ips_ipmp_lock -> ill_lock
363  * ill_mcast_serializer -> ill_mcast_lock -> connf_lock -> conn_lock
364  * ill_mcast_serializer -> ill_mcast_lock -> conn_ilg_lock
365  * ill_mcast_serializer -> ill_mcast_lock -> ips_igmp_timer_lock
366  *
367  * IPsec scenarios
368  *
369  * ipsa_lock -> ill_g_lock -> ill_lock
370  * ill_g_usesrc_lock -> ill_g_lock -> ill_lock
371  *
372  * Trusted Solaris scenarios
373  *
374  * igsa_lock -> gcgrp_rwlock -> gcgrp_lock
375  * igsa_lock -> gcdb_lock
376  * gcgrp_rwlock -> ire_lock
377  * gcgrp_rwlock -> gcdb_lock
378  *
379  * squeue(sq_lock), flow related (ft_lock, fe_lock) locking
380  *
381  * cpu_lock --> ill_lock --> sqset_lock --> sq_lock
382  * sq_lock -> conn_lock -> QLOCK(q)
383  * ill_lock -> ft_lock -> fe_lock
384  *
385  * Routing/forwarding table locking notes:
386  *
387  * Lock acquisition order: Radix tree lock, irb_lock.
388  * Requirements:
389  * i.  Walker must not hold any locks during the walker callback.
390  * ii  Walker must not see a truncated tree during the walk because of any node
391  *     deletion.
392  * iii Existing code assumes ire_bucket is valid if it is non-null and is used
393  *     in many places in the code to walk the irb list. Thus even if all the
394  *     ires in a bucket have been deleted, we still can't free the radix node
395  *     until the ires have actually been inactive'd (freed).
396  *
397  * Tree traversal - Need to hold the global tree lock in read mode.
398  * Before dropping the global tree lock, need to either increment the ire_refcnt
399  * to ensure that the radix node can't be deleted.
400  *
401  * Tree add - Need to hold the global tree lock in write mode to add a
402  * radix node. To prevent the node from being deleted, increment the
403  * irb_refcnt, after the node is added to the tree. The ire itself is
404  * added later while holding the irb_lock, but not the tree lock.
405  *
406  * Tree delete - Need to hold the global tree lock and irb_lock in write mode.
407  * All associated ires must be inactive (i.e. freed), and irb_refcnt
408  * must be zero.
409  *
410  * Walker - Increment irb_refcnt before calling the walker callback. Hold the
411  * global tree lock (read mode) for traversal.
412  *
413  * IRE dependencies - In some cases we hold ips_ire_dep_lock across ire_refrele
414  * hence we will acquire irb_lock while holding ips_ire_dep_lock.
415  *
416  * IPsec notes :
417  *
418  * IP interacts with the IPsec code (AH/ESP) by storing IPsec attributes
419  * in the ip_xmit_attr_t ip_recv_attr_t. For outbound datagrams, the
420  * ip_xmit_attr_t has the
421  * information used by the IPsec code for applying the right level of
422  * protection. The information initialized by IP in the ip_xmit_attr_t
423  * is determined by the per-socket policy or global policy in the system.
424  * For inbound datagrams, the ip_recv_attr_t
425  * starts out with nothing in it. It gets filled
426  * with the right information if it goes through the AH/ESP code, which
427  * happens if the incoming packet is secure. The information initialized
428  * by AH/ESP, is later used by IP (during fanouts to ULP) to see whether
429  * the policy requirements needed by per-socket policy or global policy
430  * is met or not.
431  *
432  * For fully connected sockets i.e dst, src [addr, port] is known,
433  * conn_policy_cached is set indicating that policy has been cached.
434  * conn_in_enforce_policy may or may not be set depending on whether
435  * there is a global policy match or per-socket policy match.
436  * Policy inheriting happpens in ip_policy_set once the destination is known.
437  * Once the right policy is set on the conn_t, policy cannot change for
438  * this socket. This makes life simpler for TCP (UDP ?) where
439  * re-transmissions go out with the same policy. For symmetry, policy
440  * is cached for fully connected UDP sockets also. Thus if policy is cached,
441  * it also implies that policy is latched i.e policy cannot change
442  * on these sockets. As we have the right policy on the conn, we don't
443  * have to lookup global policy for every outbound and inbound datagram
444  * and thus serving as an optimization. Note that a global policy change
445  * does not affect fully connected sockets if they have policy. If fully
446  * connected sockets did not have any policy associated with it, global
447  * policy change may affect them.
448  *
449  * IP Flow control notes:
450  * ---------------------
451  * Non-TCP streams are flow controlled by IP. The way this is accomplished
452  * differs when ILL_CAPAB_DLD_DIRECT is enabled for that IP instance. When
453  * ILL_DIRECT_CAPABLE(ill) is TRUE, IP can do direct function calls into
454  * GLDv3. Otherwise packets are sent down to lower layers using STREAMS
455  * functions.
456  *
457  * Per Tx ring udp flow control:
458  * This is applicable only when ILL_CAPAB_DLD_DIRECT capability is set in
459  * the ill (i.e. ILL_DIRECT_CAPABLE(ill) is true).
460  *
461  * The underlying link can expose multiple Tx rings to the GLDv3 mac layer.
462  * To achieve best performance, outgoing traffic need to be fanned out among
463  * these Tx ring. mac_tx() is called (via str_mdata_fastpath_put()) to send
464  * traffic out of the NIC and it takes a fanout hint. UDP connections pass
465  * the address of connp as fanout hint to mac_tx(). Under flow controlled
466  * condition, mac_tx() returns a non-NULL cookie (ip_mac_tx_cookie_t). This
467  * cookie points to a specific Tx ring that is blocked. The cookie is used to
468  * hash into an idl_tx_list[] entry in idl_tx_list[] array. Each idl_tx_list_t
469  * point to drain_lists (idl_t's). These drain list will store the blocked UDP
470  * connp's. The drain list is not a single list but a configurable number of
471  * lists.
472  *
473  * The diagram below shows idl_tx_list_t's and their drain_lists. ip_stack_t
474  * has an array of idl_tx_list_t. The size of the array is TX_FANOUT_SIZE
475  * which is equal to 128. This array in turn contains a pointer to idl_t[],
476  * the ip drain list. The idl_t[] array size is MIN(max_ncpus, 8). The drain
477  * list will point to the list of connp's that are flow controlled.
478  *
479  *                      ---------------   -------   -------   -------
480  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
481  *                   |  ---------------   -------   -------   -------
482  *                   |  ---------------   -------   -------   -------
483  *                   |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
484  * ----------------  |  ---------------   -------   -------   -------
485  * |idl_tx_list[0]|->|  ---------------   -------   -------   -------
486  * ----------------  |->|drain_list[2]|-->|connp|-->|connp|-->|connp|-->
487  *                   |  ---------------   -------   -------   -------
488  *                   .        .              .         .         .
489  *                   |  ---------------   -------   -------   -------
490  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
491  *                      ---------------   -------   -------   -------
492  *                      ---------------   -------   -------   -------
493  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
494  *                   |  ---------------   -------   -------   -------
495  *                   |  ---------------   -------   -------   -------
496  * ----------------  |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
497  * |idl_tx_list[1]|->|  ---------------   -------   -------   -------
498  * ----------------  |        .              .         .         .
499  *                   |  ---------------   -------   -------   -------
500  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
501  *                      ---------------   -------   -------   -------
502  *     .....
503  * ----------------
504  * |idl_tx_list[n]|-> ...
505  * ----------------
506  *
507  * When mac_tx() returns a cookie, the cookie is hashed into an index into
508  * ips_idl_tx_list[], and conn_drain_insert() is called with the idl_tx_list
509  * to insert the conn onto.  conn_drain_insert() asserts flow control for the
510  * sockets via su_txq_full() (non-STREAMS) or QFULL on conn_wq (STREAMS).
511  * Further, conn_blocked is set to indicate that the conn is blocked.
512  *
513  * GLDv3 calls ill_flow_enable() when flow control is relieved.  The cookie
514  * passed in the call to ill_flow_enable() identifies the blocked Tx ring and
515  * is again hashed to locate the appropriate idl_tx_list, which is then
516  * drained via conn_walk_drain().  conn_walk_drain() goes through each conn in
517  * the drain list and calls conn_drain_remove() to clear flow control (via
518  * calling su_txq_full() or clearing QFULL), and remove the conn from the
519  * drain list.
520  *
521  * Note that the drain list is not a single list but a (configurable) array of
522  * lists (8 elements by default).  Synchronization between drain insertion and
523  * flow control wakeup is handled by using idl_txl->txl_lock, and only
524  * conn_drain_insert() and conn_drain_remove() manipulate the drain list.
525  *
526  * Flow control via STREAMS is used when ILL_DIRECT_CAPABLE() returns FALSE.
527  * On the send side, if the packet cannot be sent down to the driver by IP
528  * (canput() fails), ip_xmit() drops the packet and returns EWOULDBLOCK to the
529  * caller, who may then invoke ixa_check_drain_insert() to insert the conn on
530  * the 0'th drain list.  When ip_wsrv() runs on the ill_wq because flow
531  * control has been relieved, the blocked conns in the 0'th drain list are
532  * drained as in the non-STREAMS case.
533  *
534  * In both the STREAMS and non-STREAMS cases, the sockfs upcall to set QFULL
535  * is done when the conn is inserted into the drain list (conn_drain_insert())
536  * and cleared when the conn is removed from the it (conn_drain_remove()).
537  *
538  * IPQOS notes:
539  *
540  * IPQoS Policies are applied to packets using IPPF (IP Policy framework)
541  * and IPQoS modules. IPPF includes hooks in IP at different control points
542  * (callout positions) which direct packets to IPQoS modules for policy
543  * processing. Policies, if present, are global.
544  *
545  * The callout positions are located in the following paths:
546  *		o local_in (packets destined for this host)
547  *		o local_out (packets orginating from this host )
548  *		o fwd_in  (packets forwarded by this m/c - inbound)
549  *		o fwd_out (packets forwarded by this m/c - outbound)
550  * Hooks at these callout points can be enabled/disabled using the ndd variable
551  * ip_policy_mask (a bit mask with the 4 LSB indicating the callout positions).
552  * By default all the callout positions are enabled.
553  *
554  * Outbound (local_out)
555  * Hooks are placed in ire_send_wire_v4 and ire_send_wire_v6.
556  *
557  * Inbound (local_in)
558  * Hooks are placed in ip_fanout_v4 and ip_fanout_v6.
559  *
560  * Forwarding (in and out)
561  * Hooks are placed in ire_recv_forward_v4/v6.
562  *
563  * IP Policy Framework processing (IPPF processing)
564  * Policy processing for a packet is initiated by ip_process, which ascertains
565  * that the classifier (ipgpc) is loaded and configured, failing which the
566  * packet resumes normal processing in IP. If the clasifier is present, the
567  * packet is acted upon by one or more IPQoS modules (action instances), per
568  * filters configured in ipgpc and resumes normal IP processing thereafter.
569  * An action instance can drop a packet in course of its processing.
570  *
571  * Zones notes:
572  *
573  * The partitioning rules for networking are as follows:
574  * 1) Packets coming from a zone must have a source address belonging to that
575  * zone.
576  * 2) Packets coming from a zone can only be sent on a physical interface on
577  * which the zone has an IP address.
578  * 3) Between two zones on the same machine, packet delivery is only allowed if
579  * there's a matching route for the destination and zone in the forwarding
580  * table.
581  * 4) The TCP and UDP port spaces are per-zone; that is, two processes in
582  * different zones can bind to the same port with the wildcard address
583  * (INADDR_ANY).
584  *
585  * The granularity of interface partitioning is at the logical interface level.
586  * Therefore, every zone has its own IP addresses, and incoming packets can be
587  * attributed to a zone unambiguously. A logical interface is placed into a zone
588  * using the SIOCSLIFZONE ioctl; this sets the ipif_zoneid field in the ipif_t
589  * structure. Rule (1) is implemented by modifying the source address selection
590  * algorithm so that the list of eligible addresses is filtered based on the
591  * sending process zone.
592  *
593  * The Internet Routing Entries (IREs) are either exclusive to a zone or shared
594  * across all zones, depending on their type. Here is the break-up:
595  *
596  * IRE type				Shared/exclusive
597  * --------				----------------
598  * IRE_BROADCAST			Exclusive
599  * IRE_DEFAULT (default routes)		Shared (*)
600  * IRE_LOCAL				Exclusive (x)
601  * IRE_LOOPBACK				Exclusive
602  * IRE_PREFIX (net routes)		Shared (*)
603  * IRE_IF_NORESOLVER (interface routes)	Exclusive
604  * IRE_IF_RESOLVER (interface routes)	Exclusive
605  * IRE_IF_CLONE (interface routes)	Exclusive
606  * IRE_HOST (host routes)		Shared (*)
607  *
608  * (*) A zone can only use a default or off-subnet route if the gateway is
609  * directly reachable from the zone, that is, if the gateway's address matches
610  * one of the zone's logical interfaces.
611  *
612  * (x) IRE_LOCAL are handled a bit differently.
613  * When ip_restrict_interzone_loopback is set (the default),
614  * ire_route_recursive restricts loopback using an IRE_LOCAL
615  * between zone to the case when L2 would have conceptually looped the packet
616  * back, i.e. the loopback which is required since neither Ethernet drivers
617  * nor Ethernet hardware loops them back. This is the case when the normal
618  * routes (ignoring IREs with different zoneids) would send out the packet on
619  * the same ill as the ill with which is IRE_LOCAL is associated.
620  *
621  * Multiple zones can share a common broadcast address; typically all zones
622  * share the 255.255.255.255 address. Incoming as well as locally originated
623  * broadcast packets must be dispatched to all the zones on the broadcast
624  * network. For directed broadcasts (e.g. 10.16.72.255) this is not trivial
625  * since some zones may not be on the 10.16.72/24 network. To handle this, each
626  * zone has its own set of IRE_BROADCAST entries; then, broadcast packets are
627  * sent to every zone that has an IRE_BROADCAST entry for the destination
628  * address on the input ill, see ip_input_broadcast().
629  *
630  * Applications in different zones can join the same multicast group address.
631  * The same logic applies for multicast as for broadcast. ip_input_multicast
632  * dispatches packets to all zones that have members on the physical interface.
633  */
634 
635 /*
636  * Squeue Fanout flags:
637  *	0: No fanout.
638  *	1: Fanout across all squeues
639  */
640 boolean_t	ip_squeue_fanout = 0;
641 
642 /*
643  * Maximum dups allowed per packet.
644  */
645 uint_t ip_max_frag_dups = 10;
646 
647 static int	ip_open(queue_t *q, dev_t *devp, int flag, int sflag,
648 		    cred_t *credp, boolean_t isv6);
649 static mblk_t	*ip_xmit_attach_llhdr(mblk_t *, nce_t *);
650 
651 static boolean_t icmp_inbound_verify_v4(mblk_t *, icmph_t *, ip_recv_attr_t *);
652 static void	icmp_inbound_too_big_v4(icmph_t *, ip_recv_attr_t *);
653 static void	icmp_inbound_error_fanout_v4(mblk_t *, icmph_t *,
654     ip_recv_attr_t *);
655 static void	icmp_options_update(ipha_t *);
656 static void	icmp_param_problem(mblk_t *, uint8_t,  ip_recv_attr_t *);
657 static void	icmp_pkt(mblk_t *, void *, size_t, ip_recv_attr_t *);
658 static mblk_t	*icmp_pkt_err_ok(mblk_t *, ip_recv_attr_t *);
659 static void	icmp_redirect_v4(mblk_t *mp, ipha_t *, icmph_t *,
660     ip_recv_attr_t *);
661 static void	icmp_send_redirect(mblk_t *, ipaddr_t, ip_recv_attr_t *);
662 static void	icmp_send_reply_v4(mblk_t *, ipha_t *, icmph_t *,
663     ip_recv_attr_t *);
664 
665 mblk_t		*ip_dlpi_alloc(size_t, t_uscalar_t);
666 char		*ip_dot_addr(ipaddr_t, char *);
667 mblk_t		*ip_carve_mp(mblk_t **, ssize_t);
668 int		ip_close(queue_t *, int);
669 static char	*ip_dot_saddr(uchar_t *, char *);
670 static void	ip_lrput(queue_t *, mblk_t *);
671 ipaddr_t	ip_net_mask(ipaddr_t);
672 char		*ip_nv_lookup(nv_t *, int);
673 void	ip_rput(queue_t *, mblk_t *);
674 static void	ip_rput_dlpi_writer(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
675 		    void *dummy_arg);
676 int		ip_snmp_get(queue_t *, mblk_t *, int, boolean_t);
677 static mblk_t	*ip_snmp_get_mib2_ip(queue_t *, mblk_t *,
678 		    mib2_ipIfStatsEntry_t *, ip_stack_t *, boolean_t);
679 static mblk_t	*ip_snmp_get_mib2_ip_traffic_stats(queue_t *, mblk_t *,
680 		    ip_stack_t *, boolean_t);
681 static mblk_t	*ip_snmp_get_mib2_ip6(queue_t *, mblk_t *, ip_stack_t *,
682 		    boolean_t);
683 static mblk_t	*ip_snmp_get_mib2_icmp(queue_t *, mblk_t *, ip_stack_t *ipst);
684 static mblk_t	*ip_snmp_get_mib2_icmp6(queue_t *, mblk_t *, ip_stack_t *ipst);
685 static mblk_t	*ip_snmp_get_mib2_igmp(queue_t *, mblk_t *, ip_stack_t *ipst);
686 static mblk_t	*ip_snmp_get_mib2_multi(queue_t *, mblk_t *, ip_stack_t *ipst);
687 static mblk_t	*ip_snmp_get_mib2_ip_addr(queue_t *, mblk_t *,
688 		    ip_stack_t *ipst, boolean_t);
689 static mblk_t	*ip_snmp_get_mib2_ip6_addr(queue_t *, mblk_t *,
690 		    ip_stack_t *ipst, boolean_t);
691 static mblk_t	*ip_snmp_get_mib2_ip_group_src(queue_t *, mblk_t *,
692 		    ip_stack_t *ipst);
693 static mblk_t	*ip_snmp_get_mib2_ip6_group_src(queue_t *, mblk_t *,
694 		    ip_stack_t *ipst);
695 static mblk_t	*ip_snmp_get_mib2_ip_group_mem(queue_t *, mblk_t *,
696 		    ip_stack_t *ipst);
697 static mblk_t	*ip_snmp_get_mib2_ip6_group_mem(queue_t *, mblk_t *,
698 		    ip_stack_t *ipst);
699 static mblk_t	*ip_snmp_get_mib2_virt_multi(queue_t *, mblk_t *,
700 		    ip_stack_t *ipst);
701 static mblk_t	*ip_snmp_get_mib2_multi_rtable(queue_t *, mblk_t *,
702 		    ip_stack_t *ipst);
703 static mblk_t	*ip_snmp_get_mib2_ip_route_media(queue_t *, mblk_t *, int,
704 		    ip_stack_t *ipst);
705 static mblk_t	*ip_snmp_get_mib2_ip6_route_media(queue_t *, mblk_t *, int,
706 		    ip_stack_t *ipst);
707 static void	ip_snmp_get2_v4(ire_t *, iproutedata_t *);
708 static void	ip_snmp_get2_v6_route(ire_t *, iproutedata_t *);
709 static int	ip_snmp_get2_v4_media(ncec_t *, iproutedata_t *);
710 static int	ip_snmp_get2_v6_media(ncec_t *, iproutedata_t *);
711 int		ip_snmp_set(queue_t *, int, int, uchar_t *, int);
712 
713 static mblk_t	*ip_fragment_copyhdr(uchar_t *, int, int, ip_stack_t *,
714 		    mblk_t *);
715 
716 static void	conn_drain_init(ip_stack_t *);
717 static void	conn_drain_fini(ip_stack_t *);
718 static void	conn_drain(conn_t *connp, boolean_t closing);
719 
720 static void	conn_walk_drain(ip_stack_t *, idl_tx_list_t *);
721 static void	conn_walk_sctp(pfv_t, void *, zoneid_t, netstack_t *);
722 
723 static void	*ip_stack_init(netstackid_t stackid, netstack_t *ns);
724 static void	ip_stack_shutdown(netstackid_t stackid, void *arg);
725 static void	ip_stack_fini(netstackid_t stackid, void *arg);
726 
727 static int	ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
728     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
729     ire_t *, conn_t *, boolean_t, const in6_addr_t *,  mcast_record_t,
730     const in6_addr_t *);
731 
732 static int	ip_squeue_switch(int);
733 
734 static void	*ip_kstat_init(netstackid_t, ip_stack_t *);
735 static void	ip_kstat_fini(netstackid_t, kstat_t *);
736 static int	ip_kstat_update(kstat_t *kp, int rw);
737 static void	*icmp_kstat_init(netstackid_t);
738 static void	icmp_kstat_fini(netstackid_t, kstat_t *);
739 static int	icmp_kstat_update(kstat_t *kp, int rw);
740 static void	*ip_kstat2_init(netstackid_t, ip_stat_t *);
741 static void	ip_kstat2_fini(netstackid_t, kstat_t *);
742 
743 static void	ipobs_init(ip_stack_t *);
744 static void	ipobs_fini(ip_stack_t *);
745 
746 static int	ip_tp_cpu_update(cpu_setup_t, int, void *);
747 
748 ipaddr_t	ip_g_all_ones = IP_HOST_MASK;
749 
750 static long ip_rput_pullups;
751 int	dohwcksum = 1;	/* use h/w cksum if supported by the hardware */
752 
753 vmem_t *ip_minor_arena_sa; /* for minor nos. from INET_MIN_DEV+2 thru 2^^18-1 */
754 vmem_t *ip_minor_arena_la; /* for minor nos. from 2^^18 thru 2^^32-1 */
755 
756 int	ip_debug;
757 
758 /*
759  * Multirouting/CGTP stuff
760  */
761 int	ip_cgtp_filter_rev = CGTP_FILTER_REV;	/* CGTP hooks version */
762 
763 /*
764  * IP tunables related declarations. Definitions are in ip_tunables.c
765  */
766 extern mod_prop_info_t ip_propinfo_tbl[];
767 extern int ip_propinfo_count;
768 
769 /*
770  * Table of IP ioctls encoding the various properties of the ioctl and
771  * indexed based on the last byte of the ioctl command. Occasionally there
772  * is a clash, and there is more than 1 ioctl with the same last byte.
773  * In such a case 1 ioctl is encoded in the ndx table and the remaining
774  * ioctls are encoded in the misc table. An entry in the ndx table is
775  * retrieved by indexing on the last byte of the ioctl command and comparing
776  * the ioctl command with the value in the ndx table. In the event of a
777  * mismatch the misc table is then searched sequentially for the desired
778  * ioctl command.
779  *
780  * Entry: <command> <copyin_size> <flags> <cmd_type> <function> <restart_func>
781  */
782 ip_ioctl_cmd_t ip_ndx_ioctl_table[] = {
783 	/* 000 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
784 	/* 001 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
785 	/* 002 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
786 	/* 003 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
787 	/* 004 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
788 	/* 005 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
789 	/* 006 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
790 	/* 007 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
791 	/* 008 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
792 	/* 009 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
793 
794 	/* 010 */ { SIOCADDRT,	sizeof (struct rtentry), IPI_PRIV,
795 			MISC_CMD, ip_siocaddrt, NULL },
796 	/* 011 */ { SIOCDELRT,	sizeof (struct rtentry), IPI_PRIV,
797 			MISC_CMD, ip_siocdelrt, NULL },
798 
799 	/* 012 */ { SIOCSIFADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
800 			IF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
801 	/* 013 */ { SIOCGIFADDR, sizeof (struct ifreq), IPI_GET_CMD,
802 			IF_CMD, ip_sioctl_get_addr, NULL },
803 
804 	/* 014 */ { SIOCSIFDSTADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
805 			IF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
806 	/* 015 */ { SIOCGIFDSTADDR, sizeof (struct ifreq),
807 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_dstaddr, NULL },
808 
809 	/* 016 */ { SIOCSIFFLAGS, sizeof (struct ifreq),
810 			IPI_PRIV | IPI_WR,
811 			IF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
812 	/* 017 */ { SIOCGIFFLAGS, sizeof (struct ifreq),
813 			IPI_MODOK | IPI_GET_CMD,
814 			IF_CMD, ip_sioctl_get_flags, NULL },
815 
816 	/* 018 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
817 	/* 019 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
818 
819 	/* copyin size cannot be coded for SIOCGIFCONF */
820 	/* 020 */ { O_SIOCGIFCONF, 0, IPI_GET_CMD,
821 			MISC_CMD, ip_sioctl_get_ifconf, NULL },
822 
823 	/* 021 */ { SIOCSIFMTU,	sizeof (struct ifreq), IPI_PRIV | IPI_WR,
824 			IF_CMD, ip_sioctl_mtu, NULL },
825 	/* 022 */ { SIOCGIFMTU,	sizeof (struct ifreq), IPI_GET_CMD,
826 			IF_CMD, ip_sioctl_get_mtu, NULL },
827 	/* 023 */ { SIOCGIFBRDADDR, sizeof (struct ifreq),
828 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_brdaddr, NULL },
829 	/* 024 */ { SIOCSIFBRDADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
830 			IF_CMD, ip_sioctl_brdaddr, NULL },
831 	/* 025 */ { SIOCGIFNETMASK, sizeof (struct ifreq),
832 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_netmask, NULL },
833 	/* 026 */ { SIOCSIFNETMASK, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
834 			IF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
835 	/* 027 */ { SIOCGIFMETRIC, sizeof (struct ifreq),
836 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_metric, NULL },
837 	/* 028 */ { SIOCSIFMETRIC, sizeof (struct ifreq), IPI_PRIV,
838 			IF_CMD, ip_sioctl_metric, NULL },
839 	/* 029 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
840 
841 	/* See 166-168 below for extended SIOC*XARP ioctls */
842 	/* 030 */ { SIOCSARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
843 			ARP_CMD, ip_sioctl_arp, NULL },
844 	/* 031 */ { SIOCGARP, sizeof (struct arpreq), IPI_GET_CMD,
845 			ARP_CMD, ip_sioctl_arp, NULL },
846 	/* 032 */ { SIOCDARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
847 			ARP_CMD, ip_sioctl_arp, NULL },
848 
849 	/* 033 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
850 	/* 034 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
851 	/* 035 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
852 	/* 036 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
853 	/* 037 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
854 	/* 038 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
855 	/* 039 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
856 	/* 040 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
857 	/* 041 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
858 	/* 042 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
859 	/* 043 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
860 	/* 044 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
861 	/* 045 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
862 	/* 046 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
863 	/* 047 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
864 	/* 048 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
865 	/* 049 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
866 	/* 050 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
867 	/* 051 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
868 	/* 052 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
869 	/* 053 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
870 
871 	/* 054 */ { IF_UNITSEL,	sizeof (int), IPI_PRIV | IPI_WR | IPI_MODOK,
872 			MISC_CMD, if_unitsel, if_unitsel_restart },
873 
874 	/* 055 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
875 	/* 056 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
876 	/* 057 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
877 	/* 058 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
878 	/* 059 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
879 	/* 060 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
880 	/* 061 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
881 	/* 062 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
882 	/* 063 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
883 	/* 064 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
884 	/* 065 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
885 	/* 066 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
886 	/* 067 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
887 	/* 068 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
888 	/* 069 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
889 	/* 070 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
890 	/* 071 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
891 	/* 072 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
892 
893 	/* 073 */ { SIOCSIFNAME, sizeof (struct ifreq),
894 			IPI_PRIV | IPI_WR | IPI_MODOK,
895 			IF_CMD, ip_sioctl_sifname, NULL },
896 
897 	/* 074 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
898 	/* 075 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
899 	/* 076 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
900 	/* 077 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
901 	/* 078 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
902 	/* 079 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
903 	/* 080 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
904 	/* 081 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
905 	/* 082 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
906 	/* 083 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
907 	/* 084 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
908 	/* 085 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
909 	/* 086 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
910 
911 	/* 087 */ { SIOCGIFNUM, sizeof (int), IPI_GET_CMD,
912 			MISC_CMD, ip_sioctl_get_ifnum, NULL },
913 	/* 088 */ { SIOCGIFMUXID, sizeof (struct ifreq), IPI_GET_CMD,
914 			IF_CMD, ip_sioctl_get_muxid, NULL },
915 	/* 089 */ { SIOCSIFMUXID, sizeof (struct ifreq),
916 			IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_muxid, NULL },
917 
918 	/* Both if and lif variants share same func */
919 	/* 090 */ { SIOCGIFINDEX, sizeof (struct ifreq), IPI_GET_CMD,
920 			IF_CMD, ip_sioctl_get_lifindex, NULL },
921 	/* Both if and lif variants share same func */
922 	/* 091 */ { SIOCSIFINDEX, sizeof (struct ifreq),
923 			IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_slifindex, NULL },
924 
925 	/* copyin size cannot be coded for SIOCGIFCONF */
926 	/* 092 */ { SIOCGIFCONF, 0, IPI_GET_CMD,
927 			MISC_CMD, ip_sioctl_get_ifconf, NULL },
928 	/* 093 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
929 	/* 094 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
930 	/* 095 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
931 	/* 096 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
932 	/* 097 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
933 	/* 098 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
934 	/* 099 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
935 	/* 100 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
936 	/* 101 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
937 	/* 102 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
938 	/* 103 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
939 	/* 104 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
940 	/* 105 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
941 	/* 106 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
942 	/* 107 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
943 	/* 108 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
944 	/* 109 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
945 
946 	/* 110 */ { SIOCLIFREMOVEIF, sizeof (struct lifreq),
947 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_removeif,
948 			ip_sioctl_removeif_restart },
949 	/* 111 */ { SIOCLIFADDIF, sizeof (struct lifreq),
950 			IPI_GET_CMD | IPI_PRIV | IPI_WR,
951 			LIF_CMD, ip_sioctl_addif, NULL },
952 #define	SIOCLIFADDR_NDX 112
953 	/* 112 */ { SIOCSLIFADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
954 			LIF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
955 	/* 113 */ { SIOCGLIFADDR, sizeof (struct lifreq),
956 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_addr, NULL },
957 	/* 114 */ { SIOCSLIFDSTADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
958 			LIF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
959 	/* 115 */ { SIOCGLIFDSTADDR, sizeof (struct lifreq),
960 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dstaddr, NULL },
961 	/* 116 */ { SIOCSLIFFLAGS, sizeof (struct lifreq),
962 			IPI_PRIV | IPI_WR,
963 			LIF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
964 	/* 117 */ { SIOCGLIFFLAGS, sizeof (struct lifreq),
965 			IPI_GET_CMD | IPI_MODOK,
966 			LIF_CMD, ip_sioctl_get_flags, NULL },
967 
968 	/* 118 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
969 	/* 119 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
970 
971 	/* 120 */ { O_SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
972 			ip_sioctl_get_lifconf, NULL },
973 	/* 121 */ { SIOCSLIFMTU, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
974 			LIF_CMD, ip_sioctl_mtu, NULL },
975 	/* 122 */ { SIOCGLIFMTU, sizeof (struct lifreq), IPI_GET_CMD,
976 			LIF_CMD, ip_sioctl_get_mtu, NULL },
977 	/* 123 */ { SIOCGLIFBRDADDR, sizeof (struct lifreq),
978 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_brdaddr, NULL },
979 	/* 124 */ { SIOCSLIFBRDADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
980 			LIF_CMD, ip_sioctl_brdaddr, NULL },
981 	/* 125 */ { SIOCGLIFNETMASK, sizeof (struct lifreq),
982 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_netmask, NULL },
983 	/* 126 */ { SIOCSLIFNETMASK, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
984 			LIF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
985 	/* 127 */ { SIOCGLIFMETRIC, sizeof (struct lifreq),
986 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_metric, NULL },
987 	/* 128 */ { SIOCSLIFMETRIC, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
988 			LIF_CMD, ip_sioctl_metric, NULL },
989 	/* 129 */ { SIOCSLIFNAME, sizeof (struct lifreq),
990 			IPI_PRIV | IPI_WR | IPI_MODOK,
991 			LIF_CMD, ip_sioctl_slifname,
992 			ip_sioctl_slifname_restart },
993 
994 	/* 130 */ { SIOCGLIFNUM, sizeof (struct lifnum), IPI_GET_CMD,
995 			MISC_CMD, ip_sioctl_get_lifnum, NULL },
996 	/* 131 */ { SIOCGLIFMUXID, sizeof (struct lifreq),
997 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_muxid, NULL },
998 	/* 132 */ { SIOCSLIFMUXID, sizeof (struct lifreq),
999 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_muxid, NULL },
1000 	/* 133 */ { SIOCGLIFINDEX, sizeof (struct lifreq),
1001 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifindex, 0 },
1002 	/* 134 */ { SIOCSLIFINDEX, sizeof (struct lifreq),
1003 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifindex, 0 },
1004 	/* 135 */ { SIOCSLIFTOKEN, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1005 			LIF_CMD, ip_sioctl_token, NULL },
1006 	/* 136 */ { SIOCGLIFTOKEN, sizeof (struct lifreq),
1007 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_token, NULL },
1008 	/* 137 */ { SIOCSLIFSUBNET, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1009 			LIF_CMD, ip_sioctl_subnet, ip_sioctl_subnet_restart },
1010 	/* 138 */ { SIOCGLIFSUBNET, sizeof (struct lifreq),
1011 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_subnet, NULL },
1012 	/* 139 */ { SIOCSLIFLNKINFO, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1013 			LIF_CMD, ip_sioctl_lnkinfo, NULL },
1014 
1015 	/* 140 */ { SIOCGLIFLNKINFO, sizeof (struct lifreq),
1016 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lnkinfo, NULL },
1017 	/* 141 */ { SIOCLIFDELND, sizeof (struct lifreq), IPI_PRIV,
1018 			LIF_CMD, ip_siocdelndp_v6, NULL },
1019 	/* 142 */ { SIOCLIFGETND, sizeof (struct lifreq), IPI_GET_CMD,
1020 			LIF_CMD, ip_siocqueryndp_v6, NULL },
1021 	/* 143 */ { SIOCLIFSETND, sizeof (struct lifreq), IPI_PRIV,
1022 			LIF_CMD, ip_siocsetndp_v6, NULL },
1023 	/* 144 */ { SIOCTMYADDR, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1024 			MISC_CMD, ip_sioctl_tmyaddr, NULL },
1025 	/* 145 */ { SIOCTONLINK, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1026 			MISC_CMD, ip_sioctl_tonlink, NULL },
1027 	/* 146 */ { SIOCTMYSITE, sizeof (struct sioc_addrreq), 0,
1028 			MISC_CMD, ip_sioctl_tmysite, NULL },
1029 	/* 147 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1030 	/* 148 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1031 
1032 	/* Old *IPSECONFIG ioctls are now deprecated, now see spdsock.c */
1033 	/* 149 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1034 	/* 150 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1035 	/* 151 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1036 	/* 152 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1037 
1038 	/* 153 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1039 
1040 	/* 154 */ { SIOCGLIFBINDING, sizeof (struct lifreq), IPI_GET_CMD,
1041 			LIF_CMD, ip_sioctl_get_binding, NULL },
1042 	/* 155 */ { SIOCSLIFGROUPNAME, sizeof (struct lifreq),
1043 			IPI_PRIV | IPI_WR,
1044 			LIF_CMD, ip_sioctl_groupname, ip_sioctl_groupname },
1045 	/* 156 */ { SIOCGLIFGROUPNAME, sizeof (struct lifreq),
1046 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_groupname, NULL },
1047 	/* 157 */ { SIOCGLIFGROUPINFO, sizeof (lifgroupinfo_t),
1048 			IPI_GET_CMD, MISC_CMD, ip_sioctl_groupinfo, NULL },
1049 
1050 	/* Leave 158-160 unused; used to be SIOC*IFARP ioctls */
1051 	/* 158 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1052 	/* 159 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1053 	/* 160 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1054 
1055 	/* 161 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1056 
1057 	/* These are handled in ip_sioctl_copyin_setup itself */
1058 	/* 162 */ { SIOCGIP6ADDRPOLICY, 0, IPI_NULL_BCONT,
1059 			MISC_CMD, NULL, NULL },
1060 	/* 163 */ { SIOCSIP6ADDRPOLICY, 0, IPI_PRIV | IPI_NULL_BCONT,
1061 			MISC_CMD, NULL, NULL },
1062 	/* 164 */ { SIOCGDSTINFO, 0, IPI_GET_CMD, MISC_CMD, NULL, NULL },
1063 
1064 	/* 165 */ { SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
1065 			ip_sioctl_get_lifconf, NULL },
1066 
1067 	/* 166 */ { SIOCSXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1068 			XARP_CMD, ip_sioctl_arp, NULL },
1069 	/* 167 */ { SIOCGXARP, sizeof (struct xarpreq), IPI_GET_CMD,
1070 			XARP_CMD, ip_sioctl_arp, NULL },
1071 	/* 168 */ { SIOCDXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1072 			XARP_CMD, ip_sioctl_arp, NULL },
1073 
1074 	/* SIOCPOPSOCKFS is not handled by IP */
1075 	/* 169 */ { IPI_DONTCARE /* SIOCPOPSOCKFS */, 0, 0, 0, NULL, NULL },
1076 
1077 	/* 170 */ { SIOCGLIFZONE, sizeof (struct lifreq),
1078 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifzone, NULL },
1079 	/* 171 */ { SIOCSLIFZONE, sizeof (struct lifreq),
1080 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifzone,
1081 			ip_sioctl_slifzone_restart },
1082 	/* 172-174 are SCTP ioctls and not handled by IP */
1083 	/* 172 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1084 	/* 173 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1085 	/* 174 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1086 	/* 175 */ { SIOCGLIFUSESRC, sizeof (struct lifreq),
1087 			IPI_GET_CMD, LIF_CMD,
1088 			ip_sioctl_get_lifusesrc, 0 },
1089 	/* 176 */ { SIOCSLIFUSESRC, sizeof (struct lifreq),
1090 			IPI_PRIV | IPI_WR,
1091 			LIF_CMD, ip_sioctl_slifusesrc,
1092 			NULL },
1093 	/* 177 */ { SIOCGLIFSRCOF, 0, IPI_GET_CMD, MISC_CMD,
1094 			ip_sioctl_get_lifsrcof, NULL },
1095 	/* 178 */ { SIOCGMSFILTER, sizeof (struct group_filter), IPI_GET_CMD,
1096 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1097 	/* 179 */ { SIOCSMSFILTER, sizeof (struct group_filter), 0,
1098 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1099 	/* 180 */ { SIOCGIPMSFILTER, sizeof (struct ip_msfilter), IPI_GET_CMD,
1100 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1101 	/* 181 */ { SIOCSIPMSFILTER, sizeof (struct ip_msfilter), 0,
1102 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1103 	/* 182 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1104 	/* SIOCSENABLESDP is handled by SDP */
1105 	/* 183 */ { IPI_DONTCARE /* SIOCSENABLESDP */, 0, 0, 0, NULL, NULL },
1106 	/* 184 */ { IPI_DONTCARE /* SIOCSQPTR */, 0, 0, 0, NULL, NULL },
1107 	/* 185 */ { SIOCGIFHWADDR, sizeof (struct ifreq), IPI_GET_CMD,
1108 			IF_CMD, ip_sioctl_get_ifhwaddr, NULL },
1109 	/* 186 */ { IPI_DONTCARE /* SIOCGSTAMP */, 0, 0, 0, NULL, NULL },
1110 	/* 187 */ { SIOCILB, 0, IPI_PRIV | IPI_GET_CMD, MISC_CMD,
1111 			ip_sioctl_ilb_cmd, NULL },
1112 	/* 188 */ { SIOCGETPROP, 0, IPI_GET_CMD, 0, NULL, NULL },
1113 	/* 189 */ { SIOCSETPROP, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL},
1114 	/* 190 */ { SIOCGLIFDADSTATE, sizeof (struct lifreq),
1115 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dadstate, NULL },
1116 	/* 191 */ { SIOCSLIFPREFIX, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1117 			LIF_CMD, ip_sioctl_prefix, ip_sioctl_prefix_restart },
1118 	/* 192 */ { SIOCGLIFHWADDR, sizeof (struct lifreq), IPI_GET_CMD,
1119 			LIF_CMD, ip_sioctl_get_lifhwaddr, NULL }
1120 };
1121 
1122 int ip_ndx_ioctl_count = sizeof (ip_ndx_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1123 
1124 ip_ioctl_cmd_t ip_misc_ioctl_table[] = {
1125 	{ I_LINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1126 	{ I_UNLINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1127 	{ I_PLINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1128 	{ I_PUNLINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1129 	{ ND_GET,	0, 0, 0, NULL, NULL },
1130 	{ ND_SET,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1131 	{ IP_IOCTL,	0, 0, 0, NULL, NULL },
1132 	{ SIOCGETVIFCNT, sizeof (struct sioc_vif_req), IPI_GET_CMD,
1133 		MISC_CMD, mrt_ioctl},
1134 	{ SIOCGETSGCNT,	sizeof (struct sioc_sg_req), IPI_GET_CMD,
1135 		MISC_CMD, mrt_ioctl},
1136 	{ SIOCGETLSGCNT, sizeof (struct sioc_lsg_req), IPI_GET_CMD,
1137 		MISC_CMD, mrt_ioctl}
1138 };
1139 
1140 int ip_misc_ioctl_count =
1141     sizeof (ip_misc_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1142 
1143 int	conn_drain_nthreads;		/* Number of drainers reqd. */
1144 					/* Settable in /etc/system */
1145 /* Defined in ip_ire.c */
1146 extern uint32_t ip_ire_max_bucket_cnt, ip6_ire_max_bucket_cnt;
1147 extern uint32_t ip_ire_min_bucket_cnt, ip6_ire_min_bucket_cnt;
1148 extern uint32_t ip_ire_mem_ratio, ip_ire_cpu_ratio;
1149 
1150 static nv_t	ire_nv_arr[] = {
1151 	{ IRE_BROADCAST, "BROADCAST" },
1152 	{ IRE_LOCAL, "LOCAL" },
1153 	{ IRE_LOOPBACK, "LOOPBACK" },
1154 	{ IRE_DEFAULT, "DEFAULT" },
1155 	{ IRE_PREFIX, "PREFIX" },
1156 	{ IRE_IF_NORESOLVER, "IF_NORESOL" },
1157 	{ IRE_IF_RESOLVER, "IF_RESOLV" },
1158 	{ IRE_IF_CLONE, "IF_CLONE" },
1159 	{ IRE_HOST, "HOST" },
1160 	{ IRE_MULTICAST, "MULTICAST" },
1161 	{ IRE_NOROUTE, "NOROUTE" },
1162 	{ 0 }
1163 };
1164 
1165 nv_t	*ire_nv_tbl = ire_nv_arr;
1166 
1167 /* Simple ICMP IP Header Template */
1168 static ipha_t icmp_ipha = {
1169 	IP_SIMPLE_HDR_VERSION, 0, 0, 0, 0, 0, IPPROTO_ICMP
1170 };
1171 
1172 struct module_info ip_mod_info = {
1173 	IP_MOD_ID, IP_MOD_NAME, IP_MOD_MINPSZ, IP_MOD_MAXPSZ, IP_MOD_HIWAT,
1174 	IP_MOD_LOWAT
1175 };
1176 
1177 /*
1178  * Duplicate static symbols within a module confuses mdb; so we avoid the
1179  * problem by making the symbols here distinct from those in udp.c.
1180  */
1181 
1182 /*
1183  * Entry points for IP as a device and as a module.
1184  * We have separate open functions for the /dev/ip and /dev/ip6 devices.
1185  */
1186 static struct qinit iprinitv4 = {
1187 	(pfi_t)ip_rput, NULL, ip_openv4, ip_close, NULL,
1188 	&ip_mod_info
1189 };
1190 
1191 struct qinit iprinitv6 = {
1192 	(pfi_t)ip_rput_v6, NULL, ip_openv6, ip_close, NULL,
1193 	&ip_mod_info
1194 };
1195 
1196 static struct qinit ipwinit = {
1197 	(pfi_t)ip_wput_nondata, (pfi_t)ip_wsrv, NULL, NULL, NULL,
1198 	&ip_mod_info
1199 };
1200 
1201 static struct qinit iplrinit = {
1202 	(pfi_t)ip_lrput, NULL, ip_openv4, ip_close, NULL,
1203 	&ip_mod_info
1204 };
1205 
1206 static struct qinit iplwinit = {
1207 	(pfi_t)ip_lwput, NULL, NULL, NULL, NULL,
1208 	&ip_mod_info
1209 };
1210 
1211 /* For AF_INET aka /dev/ip */
1212 struct streamtab ipinfov4 = {
1213 	&iprinitv4, &ipwinit, &iplrinit, &iplwinit
1214 };
1215 
1216 /* For AF_INET6 aka /dev/ip6 */
1217 struct streamtab ipinfov6 = {
1218 	&iprinitv6, &ipwinit, &iplrinit, &iplwinit
1219 };
1220 
1221 #ifdef	DEBUG
1222 boolean_t skip_sctp_cksum = B_FALSE;
1223 #endif
1224 
1225 /*
1226  * Generate an ICMP fragmentation needed message.
1227  * When called from ip_output side a minimal ip_recv_attr_t needs to be
1228  * constructed by the caller.
1229  */
1230 void
icmp_frag_needed(mblk_t * mp,int mtu,ip_recv_attr_t * ira)1231 icmp_frag_needed(mblk_t *mp, int mtu, ip_recv_attr_t *ira)
1232 {
1233 	icmph_t	icmph;
1234 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
1235 
1236 	mp = icmp_pkt_err_ok(mp, ira);
1237 	if (mp == NULL)
1238 		return;
1239 
1240 	bzero(&icmph, sizeof (icmph_t));
1241 	icmph.icmph_type = ICMP_DEST_UNREACHABLE;
1242 	icmph.icmph_code = ICMP_FRAGMENTATION_NEEDED;
1243 	icmph.icmph_du_mtu = htons((uint16_t)mtu);
1244 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutFragNeeded);
1245 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
1246 
1247 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
1248 }
1249 
1250 /*
1251  * icmp_inbound_v4 deals with ICMP messages that are handled by IP.
1252  * If the ICMP message is consumed by IP, i.e., it should not be delivered
1253  * to any IPPROTO_ICMP raw sockets, then it returns NULL.
1254  * Likewise, if the ICMP error is misformed (too short, etc), then it
1255  * returns NULL. The caller uses this to determine whether or not to send
1256  * to raw sockets.
1257  *
1258  * All error messages are passed to the matching transport stream.
1259  *
1260  * The following cases are handled by icmp_inbound:
1261  * 1) It needs to send a reply back and possibly delivering it
1262  *    to the "interested" upper clients.
1263  * 2) Return the mblk so that the caller can pass it to the RAW socket clients.
1264  * 3) It needs to change some values in IP only.
1265  * 4) It needs to change some values in IP and upper layers e.g TCP
1266  *    by delivering an error to the upper layers.
1267  *
1268  * We handle the above three cases in the context of IPsec in the
1269  * following way :
1270  *
1271  * 1) Send the reply back in the same way as the request came in.
1272  *    If it came in encrypted, it goes out encrypted. If it came in
1273  *    clear, it goes out in clear. Thus, this will prevent chosen
1274  *    plain text attack.
1275  * 2) The client may or may not expect things to come in secure.
1276  *    If it comes in secure, the policy constraints are checked
1277  *    before delivering it to the upper layers. If it comes in
1278  *    clear, ipsec_inbound_accept_clear will decide whether to
1279  *    accept this in clear or not. In both the cases, if the returned
1280  *    message (IP header + 8 bytes) that caused the icmp message has
1281  *    AH/ESP headers, it is sent up to AH/ESP for validation before
1282  *    sending up. If there are only 8 bytes of returned message, then
1283  *    upper client will not be notified.
1284  * 3) Check with global policy to see whether it matches the constaints.
1285  *    But this will be done only if icmp_accept_messages_in_clear is
1286  *    zero.
1287  * 4) If we need to change both in IP and ULP, then the decision taken
1288  *    while affecting the values in IP and while delivering up to TCP
1289  *    should be the same.
1290  *
1291  * 	There are two cases.
1292  *
1293  * 	a) If we reject data at the IP layer (ipsec_check_global_policy()
1294  *	   failed), we will not deliver it to the ULP, even though they
1295  *	   are *willing* to accept in *clear*. This is fine as our global
1296  *	   disposition to icmp messages asks us reject the datagram.
1297  *
1298  *	b) If we accept data at the IP layer (ipsec_check_global_policy()
1299  *	   succeeded or icmp_accept_messages_in_clear is 1), and not able
1300  *	   to deliver it to ULP (policy failed), it can lead to
1301  *	   consistency problems. The cases known at this time are
1302  *	   ICMP_DESTINATION_UNREACHABLE  messages with following code
1303  *	   values :
1304  *
1305  *	   - ICMP_FRAGMENTATION_NEEDED : IP adapts to the new value
1306  *	     and Upper layer rejects. Then the communication will
1307  *	     come to a stop. This is solved by making similar decisions
1308  *	     at both levels. Currently, when we are unable to deliver
1309  *	     to the Upper Layer (due to policy failures) while IP has
1310  *	     adjusted dce_pmtu, the next outbound datagram would
1311  *	     generate a local ICMP_FRAGMENTATION_NEEDED message - which
1312  *	     will be with the right level of protection. Thus the right
1313  *	     value will be communicated even if we are not able to
1314  *	     communicate when we get from the wire initially. But this
1315  *	     assumes there would be at least one outbound datagram after
1316  *	     IP has adjusted its dce_pmtu value. To make things
1317  *	     simpler, we accept in clear after the validation of
1318  *	     AH/ESP headers.
1319  *
1320  *	   - Other ICMP ERRORS : We may not be able to deliver it to the
1321  *	     upper layer depending on the level of protection the upper
1322  *	     layer expects and the disposition in ipsec_inbound_accept_clear().
1323  *	     ipsec_inbound_accept_clear() decides whether a given ICMP error
1324  *	     should be accepted in clear when the Upper layer expects secure.
1325  *	     Thus the communication may get aborted by some bad ICMP
1326  *	     packets.
1327  */
1328 mblk_t *
icmp_inbound_v4(mblk_t * mp,ip_recv_attr_t * ira)1329 icmp_inbound_v4(mblk_t *mp, ip_recv_attr_t *ira)
1330 {
1331 	icmph_t		*icmph;
1332 	ipha_t		*ipha;		/* Outer header */
1333 	int		ip_hdr_length;	/* Outer header length */
1334 	boolean_t	interested;
1335 	ipif_t		*ipif;
1336 	uint32_t	ts;
1337 	uint32_t	*tsp;
1338 	timestruc_t	now;
1339 	ill_t		*ill = ira->ira_ill;
1340 	ip_stack_t	*ipst = ill->ill_ipst;
1341 	zoneid_t	zoneid = ira->ira_zoneid;
1342 	int		len_needed;
1343 	mblk_t		*mp_ret = NULL;
1344 
1345 	ipha = (ipha_t *)mp->b_rptr;
1346 
1347 	BUMP_MIB(&ipst->ips_icmp_mib, icmpInMsgs);
1348 
1349 	ip_hdr_length = ira->ira_ip_hdr_length;
1350 	if ((mp->b_wptr - mp->b_rptr) < (ip_hdr_length + ICMPH_SIZE)) {
1351 		if (ira->ira_pktlen < (ip_hdr_length + ICMPH_SIZE)) {
1352 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1353 			ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1354 			freemsg(mp);
1355 			return (NULL);
1356 		}
1357 		/* Last chance to get real. */
1358 		ipha = ip_pullup(mp, ip_hdr_length + ICMPH_SIZE, ira);
1359 		if (ipha == NULL) {
1360 			BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1361 			freemsg(mp);
1362 			return (NULL);
1363 		}
1364 	}
1365 
1366 	/* The IP header will always be a multiple of four bytes */
1367 	icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1368 	ip2dbg(("icmp_inbound_v4: type %d code %d\n", icmph->icmph_type,
1369 	    icmph->icmph_code));
1370 
1371 	/*
1372 	 * We will set "interested" to "true" if we should pass a copy to
1373 	 * the transport or if we handle the packet locally.
1374 	 */
1375 	interested = B_FALSE;
1376 	switch (icmph->icmph_type) {
1377 	case ICMP_ECHO_REPLY:
1378 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchoReps);
1379 		break;
1380 	case ICMP_DEST_UNREACHABLE:
1381 		if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED)
1382 			BUMP_MIB(&ipst->ips_icmp_mib, icmpInFragNeeded);
1383 		interested = B_TRUE;	/* Pass up to transport */
1384 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInDestUnreachs);
1385 		break;
1386 	case ICMP_SOURCE_QUENCH:
1387 		interested = B_TRUE;	/* Pass up to transport */
1388 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInSrcQuenchs);
1389 		break;
1390 	case ICMP_REDIRECT:
1391 		if (!ipst->ips_ip_ignore_redirect)
1392 			interested = B_TRUE;
1393 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInRedirects);
1394 		break;
1395 	case ICMP_ECHO_REQUEST:
1396 		/*
1397 		 * Whether to respond to echo requests that come in as IP
1398 		 * broadcasts or as IP multicast is subject to debate
1399 		 * (what isn't?).  We aim to please, you pick it.
1400 		 * Default is do it.
1401 		 */
1402 		if (ira->ira_flags & IRAF_MULTICAST) {
1403 			/* multicast: respond based on tunable */
1404 			interested = ipst->ips_ip_g_resp_to_echo_mcast;
1405 		} else if (ira->ira_flags & IRAF_BROADCAST) {
1406 			/* broadcast: respond based on tunable */
1407 			interested = ipst->ips_ip_g_resp_to_echo_bcast;
1408 		} else {
1409 			/* unicast: always respond */
1410 			interested = B_TRUE;
1411 		}
1412 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchos);
1413 		if (!interested) {
1414 			/* We never pass these to RAW sockets */
1415 			freemsg(mp);
1416 			return (NULL);
1417 		}
1418 
1419 		/* Check db_ref to make sure we can modify the packet. */
1420 		if (mp->b_datap->db_ref > 1) {
1421 			mblk_t	*mp1;
1422 
1423 			mp1 = copymsg(mp);
1424 			freemsg(mp);
1425 			if (!mp1) {
1426 				BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1427 				return (NULL);
1428 			}
1429 			mp = mp1;
1430 			ipha = (ipha_t *)mp->b_rptr;
1431 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1432 		}
1433 		icmph->icmph_type = ICMP_ECHO_REPLY;
1434 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutEchoReps);
1435 		icmp_send_reply_v4(mp, ipha, icmph, ira);
1436 		return (NULL);
1437 
1438 	case ICMP_ROUTER_ADVERTISEMENT:
1439 	case ICMP_ROUTER_SOLICITATION:
1440 		break;
1441 	case ICMP_TIME_EXCEEDED:
1442 		interested = B_TRUE;	/* Pass up to transport */
1443 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimeExcds);
1444 		break;
1445 	case ICMP_PARAM_PROBLEM:
1446 		interested = B_TRUE;	/* Pass up to transport */
1447 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInParmProbs);
1448 		break;
1449 	case ICMP_TIME_STAMP_REQUEST:
1450 		/* Response to Time Stamp Requests is local policy. */
1451 		if (ipst->ips_ip_g_resp_to_timestamp) {
1452 			if (ira->ira_flags & IRAF_MULTIBROADCAST)
1453 				interested =
1454 				    ipst->ips_ip_g_resp_to_timestamp_bcast;
1455 			else
1456 				interested = B_TRUE;
1457 		}
1458 		if (!interested) {
1459 			/* We never pass these to RAW sockets */
1460 			freemsg(mp);
1461 			return (NULL);
1462 		}
1463 
1464 		/* Make sure we have enough of the packet */
1465 		len_needed = ip_hdr_length + ICMPH_SIZE +
1466 		    3 * sizeof (uint32_t);
1467 
1468 		if (mp->b_wptr - mp->b_rptr < len_needed) {
1469 			ipha = ip_pullup(mp, len_needed, ira);
1470 			if (ipha == NULL) {
1471 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1472 				ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1473 				    mp, ill);
1474 				freemsg(mp);
1475 				return (NULL);
1476 			}
1477 			/* Refresh following the pullup. */
1478 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1479 		}
1480 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestamps);
1481 		/* Check db_ref to make sure we can modify the packet. */
1482 		if (mp->b_datap->db_ref > 1) {
1483 			mblk_t	*mp1;
1484 
1485 			mp1 = copymsg(mp);
1486 			freemsg(mp);
1487 			if (!mp1) {
1488 				BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1489 				return (NULL);
1490 			}
1491 			mp = mp1;
1492 			ipha = (ipha_t *)mp->b_rptr;
1493 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1494 		}
1495 		icmph->icmph_type = ICMP_TIME_STAMP_REPLY;
1496 		tsp = (uint32_t *)&icmph[1];
1497 		tsp++;		/* Skip past 'originate time' */
1498 		/* Compute # of milliseconds since midnight */
1499 		gethrestime(&now);
1500 		ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
1501 		    NSEC2MSEC(now.tv_nsec);
1502 		*tsp++ = htonl(ts);	/* Lay in 'receive time' */
1503 		*tsp++ = htonl(ts);	/* Lay in 'send time' */
1504 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimestampReps);
1505 		icmp_send_reply_v4(mp, ipha, icmph, ira);
1506 		return (NULL);
1507 
1508 	case ICMP_TIME_STAMP_REPLY:
1509 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestampReps);
1510 		break;
1511 	case ICMP_INFO_REQUEST:
1512 		/* Per RFC 1122 3.2.2.7, ignore this. */
1513 	case ICMP_INFO_REPLY:
1514 		break;
1515 	case ICMP_ADDRESS_MASK_REQUEST:
1516 		if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1517 			interested =
1518 			    ipst->ips_ip_respond_to_address_mask_broadcast;
1519 		} else {
1520 			interested = B_TRUE;
1521 		}
1522 		if (!interested) {
1523 			/* We never pass these to RAW sockets */
1524 			freemsg(mp);
1525 			return (NULL);
1526 		}
1527 		len_needed = ip_hdr_length + ICMPH_SIZE + IP_ADDR_LEN;
1528 		if (mp->b_wptr - mp->b_rptr < len_needed) {
1529 			ipha = ip_pullup(mp, len_needed, ira);
1530 			if (ipha == NULL) {
1531 				BUMP_MIB(ill->ill_ip_mib,
1532 				    ipIfStatsInTruncatedPkts);
1533 				ip_drop_input("ipIfStatsInTruncatedPkts", mp,
1534 				    ill);
1535 				freemsg(mp);
1536 				return (NULL);
1537 			}
1538 			/* Refresh following the pullup. */
1539 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1540 		}
1541 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMasks);
1542 		/* Check db_ref to make sure we can modify the packet. */
1543 		if (mp->b_datap->db_ref > 1) {
1544 			mblk_t	*mp1;
1545 
1546 			mp1 = copymsg(mp);
1547 			freemsg(mp);
1548 			if (!mp1) {
1549 				BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1550 				return (NULL);
1551 			}
1552 			mp = mp1;
1553 			ipha = (ipha_t *)mp->b_rptr;
1554 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1555 		}
1556 		/*
1557 		 * Need the ipif with the mask be the same as the source
1558 		 * address of the mask reply. For unicast we have a specific
1559 		 * ipif. For multicast/broadcast we only handle onlink
1560 		 * senders, and use the source address to pick an ipif.
1561 		 */
1562 		ipif = ipif_lookup_addr(ipha->ipha_dst, ill, zoneid, ipst);
1563 		if (ipif == NULL) {
1564 			/* Broadcast or multicast */
1565 			ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1566 			if (ipif == NULL) {
1567 				freemsg(mp);
1568 				return (NULL);
1569 			}
1570 		}
1571 		icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
1572 		bcopy(&ipif->ipif_net_mask, &icmph[1], IP_ADDR_LEN);
1573 		ipif_refrele(ipif);
1574 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutAddrMaskReps);
1575 		icmp_send_reply_v4(mp, ipha, icmph, ira);
1576 		return (NULL);
1577 
1578 	case ICMP_ADDRESS_MASK_REPLY:
1579 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMaskReps);
1580 		break;
1581 	default:
1582 		interested = B_TRUE;	/* Pass up to transport */
1583 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInUnknowns);
1584 		break;
1585 	}
1586 	/*
1587 	 * See if there is an ICMP client to avoid an extra copymsg/freemsg
1588 	 * if there isn't one.
1589 	 */
1590 	if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_ICMP].connf_head != NULL) {
1591 		/* If there is an ICMP client and we want one too, copy it. */
1592 
1593 		if (!interested) {
1594 			/* Caller will deliver to RAW sockets */
1595 			return (mp);
1596 		}
1597 		mp_ret = copymsg(mp);
1598 		if (mp_ret == NULL) {
1599 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1600 			ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1601 		}
1602 	} else if (!interested) {
1603 		/* Neither we nor raw sockets are interested. Drop packet now */
1604 		freemsg(mp);
1605 		return (NULL);
1606 	}
1607 
1608 	/*
1609 	 * ICMP error or redirect packet. Make sure we have enough of
1610 	 * the header and that db_ref == 1 since we might end up modifying
1611 	 * the packet.
1612 	 */
1613 	if (mp->b_cont != NULL) {
1614 		if (ip_pullup(mp, -1, ira) == NULL) {
1615 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1616 			ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1617 			    mp, ill);
1618 			freemsg(mp);
1619 			return (mp_ret);
1620 		}
1621 	}
1622 
1623 	if (mp->b_datap->db_ref > 1) {
1624 		mblk_t	*mp1;
1625 
1626 		mp1 = copymsg(mp);
1627 		if (mp1 == NULL) {
1628 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1629 			ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1630 			freemsg(mp);
1631 			return (mp_ret);
1632 		}
1633 		freemsg(mp);
1634 		mp = mp1;
1635 	}
1636 
1637 	/*
1638 	 * In case mp has changed, verify the message before any further
1639 	 * processes.
1640 	 */
1641 	ipha = (ipha_t *)mp->b_rptr;
1642 	icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1643 	if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
1644 		freemsg(mp);
1645 		return (mp_ret);
1646 	}
1647 
1648 	switch (icmph->icmph_type) {
1649 	case ICMP_REDIRECT:
1650 		icmp_redirect_v4(mp, ipha, icmph, ira);
1651 		break;
1652 	case ICMP_DEST_UNREACHABLE:
1653 		if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED) {
1654 			/* Update DCE and adjust MTU is icmp header if needed */
1655 			icmp_inbound_too_big_v4(icmph, ira);
1656 		}
1657 		/* FALLTHRU */
1658 	default:
1659 		icmp_inbound_error_fanout_v4(mp, icmph, ira);
1660 		break;
1661 	}
1662 	return (mp_ret);
1663 }
1664 
1665 /*
1666  * Send an ICMP echo, timestamp or address mask reply.
1667  * The caller has already updated the payload part of the packet.
1668  * We handle the ICMP checksum, IP source address selection and feed
1669  * the packet into ip_output_simple.
1670  */
1671 static void
icmp_send_reply_v4(mblk_t * mp,ipha_t * ipha,icmph_t * icmph,ip_recv_attr_t * ira)1672 icmp_send_reply_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph,
1673     ip_recv_attr_t *ira)
1674 {
1675 	uint_t		ip_hdr_length = ira->ira_ip_hdr_length;
1676 	ill_t		*ill = ira->ira_ill;
1677 	ip_stack_t	*ipst = ill->ill_ipst;
1678 	ip_xmit_attr_t	ixas;
1679 
1680 	/* Send out an ICMP packet */
1681 	icmph->icmph_checksum = 0;
1682 	icmph->icmph_checksum = IP_CSUM(mp, ip_hdr_length, 0);
1683 	/* Reset time to live. */
1684 	ipha->ipha_ttl = ipst->ips_ip_def_ttl;
1685 	{
1686 		/* Swap source and destination addresses */
1687 		ipaddr_t tmp;
1688 
1689 		tmp = ipha->ipha_src;
1690 		ipha->ipha_src = ipha->ipha_dst;
1691 		ipha->ipha_dst = tmp;
1692 	}
1693 	ipha->ipha_ident = 0;
1694 	if (!IS_SIMPLE_IPH(ipha))
1695 		icmp_options_update(ipha);
1696 
1697 	bzero(&ixas, sizeof (ixas));
1698 	ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
1699 	ixas.ixa_zoneid = ira->ira_zoneid;
1700 	ixas.ixa_cred = kcred;
1701 	ixas.ixa_cpid = NOPID;
1702 	ixas.ixa_tsl = ira->ira_tsl;	/* Behave as a multi-level responder */
1703 	ixas.ixa_ifindex = 0;
1704 	ixas.ixa_ipst = ipst;
1705 	ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1706 
1707 	if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
1708 		/*
1709 		 * This packet should go out the same way as it
1710 		 * came in i.e in clear, independent of the IPsec policy
1711 		 * for transmitting packets.
1712 		 */
1713 		ixas.ixa_flags |= IXAF_NO_IPSEC;
1714 	} else {
1715 		if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
1716 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1717 			/* Note: mp already consumed and ip_drop_packet done */
1718 			return;
1719 		}
1720 	}
1721 	if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1722 		/*
1723 		 * Not one or our addresses (IRE_LOCALs), thus we let
1724 		 * ip_output_simple pick the source.
1725 		 */
1726 		ipha->ipha_src = INADDR_ANY;
1727 		ixas.ixa_flags |= IXAF_SET_SOURCE;
1728 	}
1729 	/* Should we send with DF and use dce_pmtu? */
1730 	if (ipst->ips_ipv4_icmp_return_pmtu) {
1731 		ixas.ixa_flags |= IXAF_PMTU_DISCOVERY;
1732 		ipha->ipha_fragment_offset_and_flags |= IPH_DF_HTONS;
1733 	}
1734 
1735 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
1736 
1737 	(void) ip_output_simple(mp, &ixas);
1738 	ixa_cleanup(&ixas);
1739 }
1740 
1741 /*
1742  * Verify the ICMP messages for either for ICMP error or redirect packet.
1743  * The caller should have fully pulled up the message. If it's a redirect
1744  * packet, only basic checks on IP header will be done; otherwise, verify
1745  * the packet by looking at the included ULP header.
1746  *
1747  * Called before icmp_inbound_error_fanout_v4 is called.
1748  */
1749 static boolean_t
icmp_inbound_verify_v4(mblk_t * mp,icmph_t * icmph,ip_recv_attr_t * ira)1750 icmp_inbound_verify_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
1751 {
1752 	ill_t		*ill = ira->ira_ill;
1753 	int		hdr_length;
1754 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
1755 	conn_t		*connp;
1756 	ipha_t		*ipha;	/* Inner IP header */
1757 
1758 	ipha = (ipha_t *)&icmph[1];
1759 	if ((uchar_t *)ipha + IP_SIMPLE_HDR_LENGTH > mp->b_wptr)
1760 		goto truncated;
1761 
1762 	hdr_length = IPH_HDR_LENGTH(ipha);
1763 
1764 	if ((IPH_HDR_VERSION(ipha) != IPV4_VERSION))
1765 		goto discard_pkt;
1766 
1767 	if (hdr_length < sizeof (ipha_t))
1768 		goto truncated;
1769 
1770 	if ((uchar_t *)ipha + hdr_length > mp->b_wptr)
1771 		goto truncated;
1772 
1773 	/*
1774 	 * Stop here for ICMP_REDIRECT.
1775 	 */
1776 	if (icmph->icmph_type == ICMP_REDIRECT)
1777 		return (B_TRUE);
1778 
1779 	/*
1780 	 * ICMP errors only.
1781 	 */
1782 	switch (ipha->ipha_protocol) {
1783 	case IPPROTO_UDP:
1784 		/*
1785 		 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1786 		 * transport header.
1787 		 */
1788 		if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1789 		    mp->b_wptr)
1790 			goto truncated;
1791 		break;
1792 	case IPPROTO_TCP: {
1793 		tcpha_t		*tcpha;
1794 
1795 		/*
1796 		 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1797 		 * transport header.
1798 		 */
1799 		if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1800 		    mp->b_wptr)
1801 			goto truncated;
1802 
1803 		tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
1804 		connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
1805 		    ipst);
1806 		if (connp == NULL)
1807 			goto discard_pkt;
1808 
1809 		if ((connp->conn_verifyicmp != NULL) &&
1810 		    !connp->conn_verifyicmp(connp, tcpha, icmph, NULL, ira)) {
1811 			CONN_DEC_REF(connp);
1812 			goto discard_pkt;
1813 		}
1814 		CONN_DEC_REF(connp);
1815 		break;
1816 	}
1817 	case IPPROTO_SCTP:
1818 		/*
1819 		 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1820 		 * transport header.
1821 		 */
1822 		if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1823 		    mp->b_wptr)
1824 			goto truncated;
1825 		break;
1826 	case IPPROTO_ESP:
1827 	case IPPROTO_AH:
1828 		break;
1829 	case IPPROTO_ENCAP:
1830 		if ((uchar_t *)ipha + hdr_length + sizeof (ipha_t) >
1831 		    mp->b_wptr)
1832 			goto truncated;
1833 		break;
1834 	default:
1835 		break;
1836 	}
1837 
1838 	return (B_TRUE);
1839 
1840 discard_pkt:
1841 	/* Bogus ICMP error. */
1842 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1843 	return (B_FALSE);
1844 
1845 truncated:
1846 	/* We pulled up everthing already. Must be truncated */
1847 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1848 	ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1849 	return (B_FALSE);
1850 }
1851 
1852 /* Table from RFC 1191 */
1853 static int icmp_frag_size_table[] =
1854 { 32000, 17914, 8166, 4352, 2002, 1496, 1006, 508, 296, 68 };
1855 
1856 /*
1857  * Process received ICMP Packet too big.
1858  * Just handles the DCE create/update, including using the above table of
1859  * PMTU guesses. The caller is responsible for validating the packet before
1860  * passing it in and also to fanout the ICMP error to any matching transport
1861  * conns. Assumes the message has been fully pulled up and verified.
1862  *
1863  * Before getting here, the caller has called icmp_inbound_verify_v4()
1864  * that should have verified with ULP to prevent undoing the changes we're
1865  * going to make to DCE. For example, TCP might have verified that the packet
1866  * which generated error is in the send window.
1867  *
1868  * In some cases modified this MTU in the ICMP header packet; the caller
1869  * should pass to the matching ULP after this returns.
1870  */
1871 static void
icmp_inbound_too_big_v4(icmph_t * icmph,ip_recv_attr_t * ira)1872 icmp_inbound_too_big_v4(icmph_t *icmph, ip_recv_attr_t *ira)
1873 {
1874 	dce_t		*dce;
1875 	int		old_mtu;
1876 	int		mtu, orig_mtu;
1877 	ipaddr_t	dst;
1878 	boolean_t	disable_pmtud;
1879 	ill_t		*ill = ira->ira_ill;
1880 	ip_stack_t	*ipst = ill->ill_ipst;
1881 	uint_t		hdr_length;
1882 	ipha_t		*ipha;
1883 
1884 	/* Caller already pulled up everything. */
1885 	ipha = (ipha_t *)&icmph[1];
1886 	ASSERT(icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
1887 	    icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED);
1888 	ASSERT(ill != NULL);
1889 
1890 	hdr_length = IPH_HDR_LENGTH(ipha);
1891 
1892 	/*
1893 	 * We handle path MTU for source routed packets since the DCE
1894 	 * is looked up using the final destination.
1895 	 */
1896 	dst = ip_get_dst(ipha);
1897 
1898 	dce = dce_lookup_and_add_v4(dst, ipst);
1899 	if (dce == NULL) {
1900 		/* Couldn't add a unique one - ENOMEM */
1901 		ip1dbg(("icmp_inbound_too_big_v4: no dce for 0x%x\n",
1902 		    ntohl(dst)));
1903 		return;
1904 	}
1905 
1906 	/* Check for MTU discovery advice as described in RFC 1191 */
1907 	mtu = ntohs(icmph->icmph_du_mtu);
1908 	orig_mtu = mtu;
1909 	disable_pmtud = B_FALSE;
1910 
1911 	mutex_enter(&dce->dce_lock);
1912 	if (dce->dce_flags & DCEF_PMTU)
1913 		old_mtu = dce->dce_pmtu;
1914 	else
1915 		old_mtu = ill->ill_mtu;
1916 
1917 	if (icmph->icmph_du_zero != 0 || mtu < ipst->ips_ip_pmtu_min) {
1918 		uint32_t length;
1919 		int	i;
1920 
1921 		/*
1922 		 * Use the table from RFC 1191 to figure out
1923 		 * the next "plateau" based on the length in
1924 		 * the original IP packet.
1925 		 */
1926 		length = ntohs(ipha->ipha_length);
1927 		DTRACE_PROBE2(ip4__pmtu__guess, dce_t *, dce,
1928 		    uint32_t, length);
1929 		if (old_mtu <= length &&
1930 		    old_mtu >= length - hdr_length) {
1931 			/*
1932 			 * Handle broken BSD 4.2 systems that
1933 			 * return the wrong ipha_length in ICMP
1934 			 * errors.
1935 			 */
1936 			ip1dbg(("Wrong mtu: sent %d, dce %d\n",
1937 			    length, old_mtu));
1938 			length -= hdr_length;
1939 		}
1940 		for (i = 0; i < A_CNT(icmp_frag_size_table); i++) {
1941 			if (length > icmp_frag_size_table[i])
1942 				break;
1943 		}
1944 		if (i == A_CNT(icmp_frag_size_table)) {
1945 			/* Smaller than IP_MIN_MTU! */
1946 			ip1dbg(("Too big for packet size %d\n",
1947 			    length));
1948 			disable_pmtud = B_TRUE;
1949 			mtu = ipst->ips_ip_pmtu_min;
1950 		} else {
1951 			mtu = icmp_frag_size_table[i];
1952 			ip1dbg(("Calculated mtu %d, packet size %d, "
1953 			    "before %d\n", mtu, length, old_mtu));
1954 			if (mtu < ipst->ips_ip_pmtu_min) {
1955 				mtu = ipst->ips_ip_pmtu_min;
1956 				disable_pmtud = B_TRUE;
1957 			}
1958 		}
1959 	}
1960 	if (disable_pmtud)
1961 		dce->dce_flags |= DCEF_TOO_SMALL_PMTU;
1962 	else
1963 		dce->dce_flags &= ~DCEF_TOO_SMALL_PMTU;
1964 
1965 	dce->dce_pmtu = MIN(old_mtu, mtu);
1966 	/* Prepare to send the new max frag size for the ULP. */
1967 	icmph->icmph_du_zero = 0;
1968 	icmph->icmph_du_mtu =  htons((uint16_t)dce->dce_pmtu);
1969 	DTRACE_PROBE4(ip4__pmtu__change, icmph_t *, icmph, dce_t *,
1970 	    dce, int, orig_mtu, int, mtu);
1971 
1972 	/* We now have a PMTU for sure */
1973 	dce->dce_flags |= DCEF_PMTU;
1974 	dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
1975 	mutex_exit(&dce->dce_lock);
1976 	/*
1977 	 * After dropping the lock the new value is visible to everyone.
1978 	 * Then we bump the generation number so any cached values reinspect
1979 	 * the dce_t.
1980 	 */
1981 	dce_increment_generation(dce);
1982 	dce_refrele(dce);
1983 }
1984 
1985 /*
1986  * If the packet in error is Self-Encapsulated, icmp_inbound_error_fanout_v4
1987  * calls this function.
1988  */
1989 static mblk_t *
icmp_inbound_self_encap_error_v4(mblk_t * mp,ipha_t * ipha,ipha_t * in_ipha)1990 icmp_inbound_self_encap_error_v4(mblk_t *mp, ipha_t *ipha, ipha_t *in_ipha)
1991 {
1992 	int length;
1993 
1994 	ASSERT(mp->b_datap->db_type == M_DATA);
1995 
1996 	/* icmp_inbound_v4 has already pulled up the whole error packet */
1997 	ASSERT(mp->b_cont == NULL);
1998 
1999 	/*
2000 	 * The length that we want to overlay is the inner header
2001 	 * and what follows it.
2002 	 */
2003 	length = msgdsize(mp) - ((uchar_t *)in_ipha - mp->b_rptr);
2004 
2005 	/*
2006 	 * Overlay the inner header and whatever follows it over the
2007 	 * outer header.
2008 	 */
2009 	bcopy((uchar_t *)in_ipha, (uchar_t *)ipha, length);
2010 
2011 	/* Adjust for what we removed */
2012 	mp->b_wptr -= (uchar_t *)in_ipha - (uchar_t *)ipha;
2013 	return (mp);
2014 }
2015 
2016 /*
2017  * Try to pass the ICMP message upstream in case the ULP cares.
2018  *
2019  * If the packet that caused the ICMP error is secure, we send
2020  * it to AH/ESP to make sure that the attached packet has a
2021  * valid association. ipha in the code below points to the
2022  * IP header of the packet that caused the error.
2023  *
2024  * For IPsec cases, we let the next-layer-up (which has access to
2025  * cached policy on the conn_t, or can query the SPD directly)
2026  * subtract out any IPsec overhead if they must.  We therefore make no
2027  * adjustments here for IPsec overhead.
2028  *
2029  * IFN could have been generated locally or by some router.
2030  *
2031  * LOCAL : ire_send_wire (before calling ipsec_out_process) can call
2032  * icmp_frag_needed/icmp_pkt2big_v6 to generated a local IFN.
2033  *	    This happens because IP adjusted its value of MTU on an
2034  *	    earlier IFN message and could not tell the upper layer,
2035  *	    the new adjusted value of MTU e.g. Packet was encrypted
2036  *	    or there was not enough information to fanout to upper
2037  *	    layers. Thus on the next outbound datagram, ire_send_wire
2038  *	    generates the IFN, where IPsec processing has *not* been
2039  *	    done.
2040  *
2041  *	    Note that we retain ixa_fragsize across IPsec thus once
2042  *	    we have picking ixa_fragsize and entered ipsec_out_process we do
2043  *	    no change the fragsize even if the path MTU changes before
2044  *	    we reach ip_output_post_ipsec.
2045  *
2046  *	    In the local case, IRAF_LOOPBACK will be set indicating
2047  *	    that IFN was generated locally.
2048  *
2049  * ROUTER : IFN could be secure or non-secure.
2050  *
2051  *	    * SECURE : We use the IPSEC_IN to fanout to AH/ESP if the
2052  *	      packet in error has AH/ESP headers to validate the AH/ESP
2053  *	      headers. AH/ESP will verify whether there is a valid SA or
2054  *	      not and send it back. We will fanout again if we have more
2055  *	      data in the packet.
2056  *
2057  *	      If the packet in error does not have AH/ESP, we handle it
2058  *	      like any other case.
2059  *
2060  *	    * NON_SECURE : If the packet in error has AH/ESP headers, we send it
2061  *	      up to AH/ESP for validation. AH/ESP will verify whether there is a
2062  *	      valid SA or not and send it back. We will fanout again if
2063  *	      we have more data in the packet.
2064  *
2065  *	      If the packet in error does not have AH/ESP, we handle it
2066  *	      like any other case.
2067  *
2068  * The caller must have called icmp_inbound_verify_v4.
2069  */
2070 static void
icmp_inbound_error_fanout_v4(mblk_t * mp,icmph_t * icmph,ip_recv_attr_t * ira)2071 icmp_inbound_error_fanout_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
2072 {
2073 	uint16_t	*up;	/* Pointer to ports in ULP header */
2074 	uint32_t	ports;	/* reversed ports for fanout */
2075 	ipha_t		ripha;	/* With reversed addresses */
2076 	ipha_t		*ipha;  /* Inner IP header */
2077 	uint_t		hdr_length; /* Inner IP header length */
2078 	tcpha_t		*tcpha;
2079 	conn_t		*connp;
2080 	ill_t		*ill = ira->ira_ill;
2081 	ip_stack_t	*ipst = ill->ill_ipst;
2082 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
2083 	ill_t		*rill = ira->ira_rill;
2084 
2085 	/* Caller already pulled up everything. */
2086 	ipha = (ipha_t *)&icmph[1];
2087 	ASSERT((uchar_t *)&ipha[1] <= mp->b_wptr);
2088 	ASSERT(mp->b_cont == NULL);
2089 
2090 	hdr_length = IPH_HDR_LENGTH(ipha);
2091 	ira->ira_protocol = ipha->ipha_protocol;
2092 
2093 	/*
2094 	 * We need a separate IP header with the source and destination
2095 	 * addresses reversed to do fanout/classification because the ipha in
2096 	 * the ICMP error is in the form we sent it out.
2097 	 */
2098 	ripha.ipha_src = ipha->ipha_dst;
2099 	ripha.ipha_dst = ipha->ipha_src;
2100 	ripha.ipha_protocol = ipha->ipha_protocol;
2101 	ripha.ipha_version_and_hdr_length = ipha->ipha_version_and_hdr_length;
2102 
2103 	ip2dbg(("icmp_inbound_error_v4: proto %d %x to %x: %d/%d\n",
2104 	    ripha.ipha_protocol, ntohl(ipha->ipha_src),
2105 	    ntohl(ipha->ipha_dst),
2106 	    icmph->icmph_type, icmph->icmph_code));
2107 
2108 	switch (ipha->ipha_protocol) {
2109 	case IPPROTO_UDP:
2110 		up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2111 
2112 		/* Attempt to find a client stream based on port. */
2113 		ip2dbg(("icmp_inbound_error_v4: UDP ports %d to %d\n",
2114 		    ntohs(up[0]), ntohs(up[1])));
2115 
2116 		/* Note that we send error to all matches. */
2117 		ira->ira_flags |= IRAF_ICMP_ERROR;
2118 		ip_fanout_udp_multi_v4(mp, &ripha, up[0], up[1], ira);
2119 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2120 		return;
2121 
2122 	case IPPROTO_TCP:
2123 		/*
2124 		 * Find a TCP client stream for this packet.
2125 		 * Note that we do a reverse lookup since the header is
2126 		 * in the form we sent it out.
2127 		 */
2128 		tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
2129 		connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
2130 		    ipst);
2131 		if (connp == NULL)
2132 			goto discard_pkt;
2133 
2134 		if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) ||
2135 		    (ira->ira_flags & IRAF_IPSEC_SECURE)) {
2136 			mp = ipsec_check_inbound_policy(mp, connp,
2137 			    ipha, NULL, ira);
2138 			if (mp == NULL) {
2139 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2140 				/* Note that mp is NULL */
2141 				ip_drop_input("ipIfStatsInDiscards", mp, ill);
2142 				CONN_DEC_REF(connp);
2143 				return;
2144 			}
2145 		}
2146 
2147 		ira->ira_flags |= IRAF_ICMP_ERROR;
2148 		ira->ira_ill = ira->ira_rill = NULL;
2149 		if (IPCL_IS_TCP(connp)) {
2150 			SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
2151 			    connp->conn_recvicmp, connp, ira, SQ_FILL,
2152 			    SQTAG_TCP_INPUT_ICMP_ERR);
2153 		} else {
2154 			/* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
2155 			(connp->conn_recv)(connp, mp, NULL, ira);
2156 			CONN_DEC_REF(connp);
2157 		}
2158 		ira->ira_ill = ill;
2159 		ira->ira_rill = rill;
2160 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2161 		return;
2162 
2163 	case IPPROTO_SCTP:
2164 		up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2165 		/* Find a SCTP client stream for this packet. */
2166 		((uint16_t *)&ports)[0] = up[1];
2167 		((uint16_t *)&ports)[1] = up[0];
2168 
2169 		ira->ira_flags |= IRAF_ICMP_ERROR;
2170 		ip_fanout_sctp(mp, &ripha, NULL, ports, ira);
2171 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2172 		return;
2173 
2174 	case IPPROTO_ESP:
2175 	case IPPROTO_AH:
2176 		if (!ipsec_loaded(ipss)) {
2177 			ip_proto_not_sup(mp, ira);
2178 			return;
2179 		}
2180 
2181 		if (ipha->ipha_protocol == IPPROTO_ESP)
2182 			mp = ipsecesp_icmp_error(mp, ira);
2183 		else
2184 			mp = ipsecah_icmp_error(mp, ira);
2185 		if (mp == NULL)
2186 			return;
2187 
2188 		/* Just in case ipsec didn't preserve the NULL b_cont */
2189 		if (mp->b_cont != NULL) {
2190 			if (!pullupmsg(mp, -1))
2191 				goto discard_pkt;
2192 		}
2193 
2194 		/*
2195 		 * Note that ira_pktlen and ira_ip_hdr_length are no longer
2196 		 * correct, but we don't use them any more here.
2197 		 *
2198 		 * If succesful, the mp has been modified to not include
2199 		 * the ESP/AH header so we can fanout to the ULP's icmp
2200 		 * error handler.
2201 		 */
2202 		if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2203 			goto truncated;
2204 
2205 		/* Verify the modified message before any further processes. */
2206 		ipha = (ipha_t *)mp->b_rptr;
2207 		hdr_length = IPH_HDR_LENGTH(ipha);
2208 		icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2209 		if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2210 			freemsg(mp);
2211 			return;
2212 		}
2213 
2214 		icmp_inbound_error_fanout_v4(mp, icmph, ira);
2215 		return;
2216 
2217 	case IPPROTO_ENCAP: {
2218 		/* Look for self-encapsulated packets that caused an error */
2219 		ipha_t *in_ipha;
2220 
2221 		/*
2222 		 * Caller has verified that length has to be
2223 		 * at least the size of IP header.
2224 		 */
2225 		ASSERT(hdr_length >= sizeof (ipha_t));
2226 		/*
2227 		 * Check the sanity of the inner IP header like
2228 		 * we did for the outer header.
2229 		 */
2230 		in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2231 		if ((IPH_HDR_VERSION(in_ipha) != IPV4_VERSION)) {
2232 			goto discard_pkt;
2233 		}
2234 		if (IPH_HDR_LENGTH(in_ipha) < sizeof (ipha_t)) {
2235 			goto discard_pkt;
2236 		}
2237 		/* Check for Self-encapsulated tunnels */
2238 		if (in_ipha->ipha_src == ipha->ipha_src &&
2239 		    in_ipha->ipha_dst == ipha->ipha_dst) {
2240 
2241 			mp = icmp_inbound_self_encap_error_v4(mp, ipha,
2242 			    in_ipha);
2243 			if (mp == NULL)
2244 				goto discard_pkt;
2245 
2246 			/*
2247 			 * Just in case self_encap didn't preserve the NULL
2248 			 * b_cont
2249 			 */
2250 			if (mp->b_cont != NULL) {
2251 				if (!pullupmsg(mp, -1))
2252 					goto discard_pkt;
2253 			}
2254 			/*
2255 			 * Note that ira_pktlen and ira_ip_hdr_length are no
2256 			 * longer correct, but we don't use them any more here.
2257 			 */
2258 			if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2259 				goto truncated;
2260 
2261 			/*
2262 			 * Verify the modified message before any further
2263 			 * processes.
2264 			 */
2265 			ipha = (ipha_t *)mp->b_rptr;
2266 			hdr_length = IPH_HDR_LENGTH(ipha);
2267 			icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2268 			if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2269 				freemsg(mp);
2270 				return;
2271 			}
2272 
2273 			/*
2274 			 * The packet in error is self-encapsualted.
2275 			 * And we are finding it further encapsulated
2276 			 * which we could not have possibly generated.
2277 			 */
2278 			if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2279 				goto discard_pkt;
2280 			}
2281 			icmp_inbound_error_fanout_v4(mp, icmph, ira);
2282 			return;
2283 		}
2284 		/* No self-encapsulated */
2285 		/* FALLTHRU */
2286 	}
2287 	case IPPROTO_IPV6:
2288 		if ((connp = ipcl_iptun_classify_v4(&ripha.ipha_src,
2289 		    &ripha.ipha_dst, ipst)) != NULL) {
2290 			ira->ira_flags |= IRAF_ICMP_ERROR;
2291 			connp->conn_recvicmp(connp, mp, NULL, ira);
2292 			CONN_DEC_REF(connp);
2293 			ira->ira_flags &= ~IRAF_ICMP_ERROR;
2294 			return;
2295 		}
2296 		/*
2297 		 * No IP tunnel is interested, fallthrough and see
2298 		 * if a raw socket will want it.
2299 		 */
2300 		/* FALLTHRU */
2301 	default:
2302 		ira->ira_flags |= IRAF_ICMP_ERROR;
2303 		ip_fanout_proto_v4(mp, &ripha, ira);
2304 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2305 		return;
2306 	}
2307 	/* NOTREACHED */
2308 discard_pkt:
2309 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2310 	ip1dbg(("icmp_inbound_error_fanout_v4: drop pkt\n"));
2311 	ip_drop_input("ipIfStatsInDiscards", mp, ill);
2312 	freemsg(mp);
2313 	return;
2314 
2315 truncated:
2316 	/* We pulled up everthing already. Must be truncated */
2317 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
2318 	ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
2319 	freemsg(mp);
2320 }
2321 
2322 /*
2323  * Common IP options parser.
2324  *
2325  * Setup routine: fill in *optp with options-parsing state, then
2326  * tail-call ipoptp_next to return the first option.
2327  */
2328 uint8_t
ipoptp_first(ipoptp_t * optp,ipha_t * ipha)2329 ipoptp_first(ipoptp_t *optp, ipha_t *ipha)
2330 {
2331 	uint32_t totallen; /* total length of all options */
2332 
2333 	totallen = ipha->ipha_version_and_hdr_length -
2334 	    (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
2335 	totallen <<= 2;
2336 	optp->ipoptp_next = (uint8_t *)(&ipha[1]);
2337 	optp->ipoptp_end = optp->ipoptp_next + totallen;
2338 	optp->ipoptp_flags = 0;
2339 	return (ipoptp_next(optp));
2340 }
2341 
2342 /* Like above but without an ipha_t */
2343 uint8_t
ipoptp_first2(ipoptp_t * optp,uint32_t totallen,uint8_t * opt)2344 ipoptp_first2(ipoptp_t *optp, uint32_t totallen, uint8_t *opt)
2345 {
2346 	optp->ipoptp_next = opt;
2347 	optp->ipoptp_end = optp->ipoptp_next + totallen;
2348 	optp->ipoptp_flags = 0;
2349 	return (ipoptp_next(optp));
2350 }
2351 
2352 /*
2353  * Common IP options parser: extract next option.
2354  */
2355 uint8_t
ipoptp_next(ipoptp_t * optp)2356 ipoptp_next(ipoptp_t *optp)
2357 {
2358 	uint8_t *end = optp->ipoptp_end;
2359 	uint8_t *cur = optp->ipoptp_next;
2360 	uint8_t opt, len, pointer;
2361 
2362 	/*
2363 	 * If cur > end already, then the ipoptp_end or ipoptp_next pointer
2364 	 * has been corrupted.
2365 	 */
2366 	ASSERT(cur <= end);
2367 
2368 	if (cur == end)
2369 		return (IPOPT_EOL);
2370 
2371 	opt = cur[IPOPT_OPTVAL];
2372 
2373 	/*
2374 	 * Skip any NOP options.
2375 	 */
2376 	while (opt == IPOPT_NOP) {
2377 		cur++;
2378 		if (cur == end)
2379 			return (IPOPT_EOL);
2380 		opt = cur[IPOPT_OPTVAL];
2381 	}
2382 
2383 	if (opt == IPOPT_EOL)
2384 		return (IPOPT_EOL);
2385 
2386 	/*
2387 	 * Option requiring a length.
2388 	 */
2389 	if ((cur + 1) >= end) {
2390 		optp->ipoptp_flags |= IPOPTP_ERROR;
2391 		return (IPOPT_EOL);
2392 	}
2393 	len = cur[IPOPT_OLEN];
2394 	if (len < 2) {
2395 		optp->ipoptp_flags |= IPOPTP_ERROR;
2396 		return (IPOPT_EOL);
2397 	}
2398 	optp->ipoptp_cur = cur;
2399 	optp->ipoptp_len = len;
2400 	optp->ipoptp_next = cur + len;
2401 	if (cur + len > end) {
2402 		optp->ipoptp_flags |= IPOPTP_ERROR;
2403 		return (IPOPT_EOL);
2404 	}
2405 
2406 	/*
2407 	 * For the options which require a pointer field, make sure
2408 	 * its there, and make sure it points to either something
2409 	 * inside this option, or the end of the option.
2410 	 */
2411 	switch (opt) {
2412 	case IPOPT_RR:
2413 	case IPOPT_TS:
2414 	case IPOPT_LSRR:
2415 	case IPOPT_SSRR:
2416 		if (len <= IPOPT_OFFSET) {
2417 			optp->ipoptp_flags |= IPOPTP_ERROR;
2418 			return (opt);
2419 		}
2420 		pointer = cur[IPOPT_OFFSET];
2421 		if (pointer - 1 > len) {
2422 			optp->ipoptp_flags |= IPOPTP_ERROR;
2423 			return (opt);
2424 		}
2425 		break;
2426 	}
2427 
2428 	/*
2429 	 * Sanity check the pointer field based on the type of the
2430 	 * option.
2431 	 */
2432 	switch (opt) {
2433 	case IPOPT_RR:
2434 	case IPOPT_SSRR:
2435 	case IPOPT_LSRR:
2436 		if (pointer < IPOPT_MINOFF_SR)
2437 			optp->ipoptp_flags |= IPOPTP_ERROR;
2438 		break;
2439 	case IPOPT_TS:
2440 		if (pointer < IPOPT_MINOFF_IT)
2441 			optp->ipoptp_flags |= IPOPTP_ERROR;
2442 		/*
2443 		 * Note that the Internet Timestamp option also
2444 		 * contains two four bit fields (the Overflow field,
2445 		 * and the Flag field), which follow the pointer
2446 		 * field.  We don't need to check that these fields
2447 		 * fall within the length of the option because this
2448 		 * was implicitely done above.  We've checked that the
2449 		 * pointer value is at least IPOPT_MINOFF_IT, and that
2450 		 * it falls within the option.  Since IPOPT_MINOFF_IT >
2451 		 * IPOPT_POS_OV_FLG, we don't need the explicit check.
2452 		 */
2453 		ASSERT(len > IPOPT_POS_OV_FLG);
2454 		break;
2455 	}
2456 
2457 	return (opt);
2458 }
2459 
2460 /*
2461  * Use the outgoing IP header to create an IP_OPTIONS option the way
2462  * it was passed down from the application.
2463  *
2464  * This is compatible with BSD in that it returns
2465  * the reverse source route with the final destination
2466  * as the last entry. The first 4 bytes of the option
2467  * will contain the final destination.
2468  */
2469 int
ip_opt_get_user(conn_t * connp,uchar_t * buf)2470 ip_opt_get_user(conn_t *connp, uchar_t *buf)
2471 {
2472 	ipoptp_t	opts;
2473 	uchar_t		*opt;
2474 	uint8_t		optval;
2475 	uint8_t		optlen;
2476 	uint32_t	len = 0;
2477 	uchar_t		*buf1 = buf;
2478 	uint32_t	totallen;
2479 	ipaddr_t	dst;
2480 	ip_pkt_t	*ipp = &connp->conn_xmit_ipp;
2481 
2482 	if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
2483 		return (0);
2484 
2485 	totallen = ipp->ipp_ipv4_options_len;
2486 	if (totallen & 0x3)
2487 		return (0);
2488 
2489 	buf += IP_ADDR_LEN;	/* Leave room for final destination */
2490 	len += IP_ADDR_LEN;
2491 	bzero(buf1, IP_ADDR_LEN);
2492 
2493 	dst = connp->conn_faddr_v4;
2494 
2495 	for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
2496 	    optval != IPOPT_EOL;
2497 	    optval = ipoptp_next(&opts)) {
2498 		int	off;
2499 
2500 		opt = opts.ipoptp_cur;
2501 		if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
2502 			break;
2503 		}
2504 		optlen = opts.ipoptp_len;
2505 
2506 		switch (optval) {
2507 		case IPOPT_SSRR:
2508 		case IPOPT_LSRR:
2509 
2510 			/*
2511 			 * Insert destination as the first entry in the source
2512 			 * route and move down the entries on step.
2513 			 * The last entry gets placed at buf1.
2514 			 */
2515 			buf[IPOPT_OPTVAL] = optval;
2516 			buf[IPOPT_OLEN] = optlen;
2517 			buf[IPOPT_OFFSET] = optlen;
2518 
2519 			off = optlen - IP_ADDR_LEN;
2520 			if (off < 0) {
2521 				/* No entries in source route */
2522 				break;
2523 			}
2524 			/* Last entry in source route if not already set */
2525 			if (dst == INADDR_ANY)
2526 				bcopy(opt + off, buf1, IP_ADDR_LEN);
2527 			off -= IP_ADDR_LEN;
2528 
2529 			while (off > 0) {
2530 				bcopy(opt + off,
2531 				    buf + off + IP_ADDR_LEN,
2532 				    IP_ADDR_LEN);
2533 				off -= IP_ADDR_LEN;
2534 			}
2535 			/* ipha_dst into first slot */
2536 			bcopy(&dst, buf + off + IP_ADDR_LEN,
2537 			    IP_ADDR_LEN);
2538 			buf += optlen;
2539 			len += optlen;
2540 			break;
2541 
2542 		default:
2543 			bcopy(opt, buf, optlen);
2544 			buf += optlen;
2545 			len += optlen;
2546 			break;
2547 		}
2548 	}
2549 done:
2550 	/* Pad the resulting options */
2551 	while (len & 0x3) {
2552 		*buf++ = IPOPT_EOL;
2553 		len++;
2554 	}
2555 	return (len);
2556 }
2557 
2558 /*
2559  * Update any record route or timestamp options to include this host.
2560  * Reverse any source route option.
2561  * This routine assumes that the options are well formed i.e. that they
2562  * have already been checked.
2563  */
2564 static void
icmp_options_update(ipha_t * ipha)2565 icmp_options_update(ipha_t *ipha)
2566 {
2567 	ipoptp_t	opts;
2568 	uchar_t		*opt;
2569 	uint8_t		optval;
2570 	ipaddr_t	src;		/* Our local address */
2571 	ipaddr_t	dst;
2572 
2573 	ip2dbg(("icmp_options_update\n"));
2574 	src = ipha->ipha_src;
2575 	dst = ipha->ipha_dst;
2576 
2577 	for (optval = ipoptp_first(&opts, ipha);
2578 	    optval != IPOPT_EOL;
2579 	    optval = ipoptp_next(&opts)) {
2580 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
2581 		opt = opts.ipoptp_cur;
2582 		ip2dbg(("icmp_options_update: opt %d, len %d\n",
2583 		    optval, opts.ipoptp_len));
2584 		switch (optval) {
2585 			int off1, off2;
2586 		case IPOPT_SSRR:
2587 		case IPOPT_LSRR:
2588 			/*
2589 			 * Reverse the source route.  The first entry
2590 			 * should be the next to last one in the current
2591 			 * source route (the last entry is our address).
2592 			 * The last entry should be the final destination.
2593 			 */
2594 			off1 = IPOPT_MINOFF_SR - 1;
2595 			off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
2596 			if (off2 < 0) {
2597 				/* No entries in source route */
2598 				ip1dbg((
2599 				    "icmp_options_update: bad src route\n"));
2600 				break;
2601 			}
2602 			bcopy((char *)opt + off2, &dst, IP_ADDR_LEN);
2603 			bcopy(&ipha->ipha_dst, (char *)opt + off2, IP_ADDR_LEN);
2604 			bcopy(&dst, &ipha->ipha_dst, IP_ADDR_LEN);
2605 			off2 -= IP_ADDR_LEN;
2606 
2607 			while (off1 < off2) {
2608 				bcopy((char *)opt + off1, &src, IP_ADDR_LEN);
2609 				bcopy((char *)opt + off2, (char *)opt + off1,
2610 				    IP_ADDR_LEN);
2611 				bcopy(&src, (char *)opt + off2, IP_ADDR_LEN);
2612 				off1 += IP_ADDR_LEN;
2613 				off2 -= IP_ADDR_LEN;
2614 			}
2615 			opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
2616 			break;
2617 		}
2618 	}
2619 }
2620 
2621 /*
2622  * Process received ICMP Redirect messages.
2623  * Assumes the caller has verified that the headers are in the pulled up mblk.
2624  * Consumes mp.
2625  */
2626 static void
icmp_redirect_v4(mblk_t * mp,ipha_t * ipha,icmph_t * icmph,ip_recv_attr_t * ira)2627 icmp_redirect_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph, ip_recv_attr_t *ira)
2628 {
2629 	ire_t		*ire, *nire;
2630 	ire_t		*prev_ire;
2631 	ipaddr_t  	src, dst, gateway;
2632 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
2633 	ipha_t		*inner_ipha;	/* Inner IP header */
2634 
2635 	/* Caller already pulled up everything. */
2636 	inner_ipha = (ipha_t *)&icmph[1];
2637 	src = ipha->ipha_src;
2638 	dst = inner_ipha->ipha_dst;
2639 	gateway = icmph->icmph_rd_gateway;
2640 	/* Make sure the new gateway is reachable somehow. */
2641 	ire = ire_ftable_lookup_v4(gateway, 0, 0, IRE_ONLINK, NULL,
2642 	    ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst, NULL);
2643 	/*
2644 	 * Make sure we had a route for the dest in question and that
2645 	 * that route was pointing to the old gateway (the source of the
2646 	 * redirect packet.)
2647 	 * We do longest match and then compare ire_gateway_addr below.
2648 	 */
2649 	prev_ire = ire_ftable_lookup_v4(dst, 0, 0, 0, NULL, ALL_ZONES,
2650 	    NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
2651 	/*
2652 	 * Check that
2653 	 *	the redirect was not from ourselves
2654 	 *	the new gateway and the old gateway are directly reachable
2655 	 */
2656 	if (prev_ire == NULL || ire == NULL ||
2657 	    (prev_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) ||
2658 	    (prev_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
2659 	    !(ire->ire_type & IRE_IF_ALL) ||
2660 	    prev_ire->ire_gateway_addr != src) {
2661 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2662 		ip_drop_input("icmpInBadRedirects - ire", mp, ira->ira_ill);
2663 		freemsg(mp);
2664 		if (ire != NULL)
2665 			ire_refrele(ire);
2666 		if (prev_ire != NULL)
2667 			ire_refrele(prev_ire);
2668 		return;
2669 	}
2670 
2671 	ire_refrele(prev_ire);
2672 	ire_refrele(ire);
2673 
2674 	/*
2675 	 * TODO: more precise handling for cases 0, 2, 3, the latter two
2676 	 * require TOS routing
2677 	 */
2678 	switch (icmph->icmph_code) {
2679 	case 0:
2680 	case 1:
2681 		/* TODO: TOS specificity for cases 2 and 3 */
2682 	case 2:
2683 	case 3:
2684 		break;
2685 	default:
2686 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2687 		ip_drop_input("icmpInBadRedirects - code", mp, ira->ira_ill);
2688 		freemsg(mp);
2689 		return;
2690 	}
2691 	/*
2692 	 * Create a Route Association.  This will allow us to remember that
2693 	 * someone we believe told us to use the particular gateway.
2694 	 */
2695 	ire = ire_create(
2696 	    (uchar_t *)&dst,			/* dest addr */
2697 	    (uchar_t *)&ip_g_all_ones,		/* mask */
2698 	    (uchar_t *)&gateway,		/* gateway addr */
2699 	    IRE_HOST,
2700 	    NULL,				/* ill */
2701 	    ALL_ZONES,
2702 	    (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST),
2703 	    NULL,				/* tsol_gc_t */
2704 	    ipst);
2705 
2706 	if (ire == NULL) {
2707 		freemsg(mp);
2708 		return;
2709 	}
2710 	nire = ire_add(ire);
2711 	/* Check if it was a duplicate entry */
2712 	if (nire != NULL && nire != ire) {
2713 		ASSERT(nire->ire_identical_ref > 1);
2714 		ire_delete(nire);
2715 		ire_refrele(nire);
2716 		nire = NULL;
2717 	}
2718 	ire = nire;
2719 	if (ire != NULL) {
2720 		ire_refrele(ire);		/* Held in ire_add */
2721 
2722 		/* tell routing sockets that we received a redirect */
2723 		ip_rts_change(RTM_REDIRECT, dst, gateway, IP_HOST_MASK, 0, src,
2724 		    (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST), 0,
2725 		    (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_AUTHOR), ipst);
2726 	}
2727 
2728 	/*
2729 	 * Delete any existing IRE_HOST type redirect ires for this destination.
2730 	 * This together with the added IRE has the effect of
2731 	 * modifying an existing redirect.
2732 	 */
2733 	prev_ire = ire_ftable_lookup_v4(dst, 0, src, IRE_HOST, NULL,
2734 	    ALL_ZONES, NULL, (MATCH_IRE_GW | MATCH_IRE_TYPE), 0, ipst, NULL);
2735 	if (prev_ire != NULL) {
2736 		if (prev_ire ->ire_flags & RTF_DYNAMIC)
2737 			ire_delete(prev_ire);
2738 		ire_refrele(prev_ire);
2739 	}
2740 
2741 	freemsg(mp);
2742 }
2743 
2744 /*
2745  * Generate an ICMP parameter problem message.
2746  * When called from ip_output side a minimal ip_recv_attr_t needs to be
2747  * constructed by the caller.
2748  */
2749 static void
icmp_param_problem(mblk_t * mp,uint8_t ptr,ip_recv_attr_t * ira)2750 icmp_param_problem(mblk_t *mp, uint8_t ptr, ip_recv_attr_t *ira)
2751 {
2752 	icmph_t	icmph;
2753 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
2754 
2755 	mp = icmp_pkt_err_ok(mp, ira);
2756 	if (mp == NULL)
2757 		return;
2758 
2759 	bzero(&icmph, sizeof (icmph_t));
2760 	icmph.icmph_type = ICMP_PARAM_PROBLEM;
2761 	icmph.icmph_pp_ptr = ptr;
2762 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutParmProbs);
2763 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
2764 }
2765 
2766 /*
2767  * Build and ship an IPv4 ICMP message using the packet data in mp, and
2768  * the ICMP header pointed to by "stuff".  (May be called as writer.)
2769  * Note: assumes that icmp_pkt_err_ok has been called to verify that
2770  * an icmp error packet can be sent.
2771  * Assigns an appropriate source address to the packet. If ipha_dst is
2772  * one of our addresses use it for source. Otherwise let ip_output_simple
2773  * pick the source address.
2774  */
2775 static void
icmp_pkt(mblk_t * mp,void * stuff,size_t len,ip_recv_attr_t * ira)2776 icmp_pkt(mblk_t *mp, void *stuff, size_t len, ip_recv_attr_t *ira)
2777 {
2778 	ipaddr_t dst;
2779 	icmph_t	*icmph;
2780 	ipha_t	*ipha;
2781 	uint_t	len_needed;
2782 	size_t	msg_len;
2783 	mblk_t	*mp1;
2784 	ipaddr_t src;
2785 	ire_t	*ire;
2786 	ip_xmit_attr_t ixas;
2787 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2788 
2789 	ipha = (ipha_t *)mp->b_rptr;
2790 
2791 	bzero(&ixas, sizeof (ixas));
2792 	ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
2793 	ixas.ixa_zoneid = ira->ira_zoneid;
2794 	ixas.ixa_ifindex = 0;
2795 	ixas.ixa_ipst = ipst;
2796 	ixas.ixa_cred = kcred;
2797 	ixas.ixa_cpid = NOPID;
2798 	ixas.ixa_tsl = ira->ira_tsl;	/* Behave as a multi-level responder */
2799 	ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
2800 
2801 	if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2802 		/*
2803 		 * Apply IPsec based on how IPsec was applied to
2804 		 * the packet that had the error.
2805 		 *
2806 		 * If it was an outbound packet that caused the ICMP
2807 		 * error, then the caller will have setup the IRA
2808 		 * appropriately.
2809 		 */
2810 		if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
2811 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2812 			/* Note: mp already consumed and ip_drop_packet done */
2813 			return;
2814 		}
2815 	} else {
2816 		/*
2817 		 * This is in clear. The icmp message we are building
2818 		 * here should go out in clear, independent of our policy.
2819 		 */
2820 		ixas.ixa_flags |= IXAF_NO_IPSEC;
2821 	}
2822 
2823 	/* Remember our eventual destination */
2824 	dst = ipha->ipha_src;
2825 
2826 	/*
2827 	 * If the packet was for one of our unicast addresses, make
2828 	 * sure we respond with that as the source. Otherwise
2829 	 * have ip_output_simple pick the source address.
2830 	 */
2831 	ire = ire_ftable_lookup_v4(ipha->ipha_dst, 0, 0,
2832 	    (IRE_LOCAL|IRE_LOOPBACK), NULL, ira->ira_zoneid, NULL,
2833 	    MATCH_IRE_TYPE|MATCH_IRE_ZONEONLY, 0, ipst, NULL);
2834 	if (ire != NULL) {
2835 		ire_refrele(ire);
2836 		src = ipha->ipha_dst;
2837 	} else {
2838 		src = INADDR_ANY;
2839 		ixas.ixa_flags |= IXAF_SET_SOURCE;
2840 	}
2841 
2842 	/*
2843 	 * Check if we can send back more then 8 bytes in addition to
2844 	 * the IP header.  We try to send 64 bytes of data and the internal
2845 	 * header in the special cases of ipv4 encapsulated ipv4 or ipv6.
2846 	 */
2847 	len_needed = IPH_HDR_LENGTH(ipha);
2848 	if (ipha->ipha_protocol == IPPROTO_ENCAP ||
2849 	    ipha->ipha_protocol == IPPROTO_IPV6) {
2850 		if (!pullupmsg(mp, -1)) {
2851 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2852 			ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2853 			freemsg(mp);
2854 			return;
2855 		}
2856 		ipha = (ipha_t *)mp->b_rptr;
2857 
2858 		if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2859 			len_needed += IPH_HDR_LENGTH(((uchar_t *)ipha +
2860 			    len_needed));
2861 		} else {
2862 			ip6_t *ip6h = (ip6_t *)((uchar_t *)ipha + len_needed);
2863 
2864 			ASSERT(ipha->ipha_protocol == IPPROTO_IPV6);
2865 			len_needed += ip_hdr_length_v6(mp, ip6h);
2866 		}
2867 	}
2868 	len_needed += ipst->ips_ip_icmp_return;
2869 	msg_len = msgdsize(mp);
2870 	if (msg_len > len_needed) {
2871 		(void) adjmsg(mp, len_needed - msg_len);
2872 		msg_len = len_needed;
2873 	}
2874 	mp1 = allocb(sizeof (icmp_ipha) + len, BPRI_MED);
2875 	if (mp1 == NULL) {
2876 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutErrors);
2877 		freemsg(mp);
2878 		return;
2879 	}
2880 	mp1->b_cont = mp;
2881 	mp = mp1;
2882 
2883 	/*
2884 	 * Set IXAF_TRUSTED_ICMP so we can let the ICMP messages this
2885 	 * node generates be accepted in peace by all on-host destinations.
2886 	 * If we do NOT assume that all on-host destinations trust
2887 	 * self-generated ICMP messages, then rework here, ip6.c, and spd.c.
2888 	 * (Look for IXAF_TRUSTED_ICMP).
2889 	 */
2890 	ixas.ixa_flags |= IXAF_TRUSTED_ICMP;
2891 
2892 	ipha = (ipha_t *)mp->b_rptr;
2893 	mp1->b_wptr = (uchar_t *)ipha + (sizeof (icmp_ipha) + len);
2894 	*ipha = icmp_ipha;
2895 	ipha->ipha_src = src;
2896 	ipha->ipha_dst = dst;
2897 	ipha->ipha_ttl = ipst->ips_ip_def_ttl;
2898 	msg_len += sizeof (icmp_ipha) + len;
2899 	if (msg_len > IP_MAXPACKET) {
2900 		(void) adjmsg(mp, IP_MAXPACKET - msg_len);
2901 		msg_len = IP_MAXPACKET;
2902 	}
2903 	ipha->ipha_length = htons((uint16_t)msg_len);
2904 	icmph = (icmph_t *)&ipha[1];
2905 	bcopy(stuff, icmph, len);
2906 	icmph->icmph_checksum = 0;
2907 	icmph->icmph_checksum = IP_CSUM(mp, (int32_t)sizeof (ipha_t), 0);
2908 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
2909 
2910 	(void) ip_output_simple(mp, &ixas);
2911 	ixa_cleanup(&ixas);
2912 }
2913 
2914 /*
2915  * Determine if an ICMP error packet can be sent given the rate limit.
2916  * The limit consists of an average frequency (icmp_pkt_err_interval measured
2917  * in milliseconds) and a burst size. Burst size number of packets can
2918  * be sent arbitrarely closely spaced.
2919  * The state is tracked using two variables to implement an approximate
2920  * token bucket filter:
2921  *	icmp_pkt_err_last - lbolt value when the last burst started
2922  *	icmp_pkt_err_sent - number of packets sent in current burst
2923  */
2924 boolean_t
icmp_err_rate_limit(ip_stack_t * ipst)2925 icmp_err_rate_limit(ip_stack_t *ipst)
2926 {
2927 	clock_t now = TICK_TO_MSEC(ddi_get_lbolt());
2928 	uint_t refilled; /* Number of packets refilled in tbf since last */
2929 	/* Guard against changes by loading into local variable */
2930 	uint_t err_interval = ipst->ips_ip_icmp_err_interval;
2931 
2932 	if (err_interval == 0)
2933 		return (B_FALSE);
2934 
2935 	if (ipst->ips_icmp_pkt_err_last > now) {
2936 		/* 100HZ lbolt in ms for 32bit arch wraps every 49.7 days */
2937 		ipst->ips_icmp_pkt_err_last = 0;
2938 		ipst->ips_icmp_pkt_err_sent = 0;
2939 	}
2940 	/*
2941 	 * If we are in a burst update the token bucket filter.
2942 	 * Update the "last" time to be close to "now" but make sure
2943 	 * we don't loose precision.
2944 	 */
2945 	if (ipst->ips_icmp_pkt_err_sent != 0) {
2946 		refilled = (now - ipst->ips_icmp_pkt_err_last)/err_interval;
2947 		if (refilled > ipst->ips_icmp_pkt_err_sent) {
2948 			ipst->ips_icmp_pkt_err_sent = 0;
2949 		} else {
2950 			ipst->ips_icmp_pkt_err_sent -= refilled;
2951 			ipst->ips_icmp_pkt_err_last += refilled * err_interval;
2952 		}
2953 	}
2954 	if (ipst->ips_icmp_pkt_err_sent == 0) {
2955 		/* Start of new burst */
2956 		ipst->ips_icmp_pkt_err_last = now;
2957 	}
2958 	if (ipst->ips_icmp_pkt_err_sent < ipst->ips_ip_icmp_err_burst) {
2959 		ipst->ips_icmp_pkt_err_sent++;
2960 		ip1dbg(("icmp_err_rate_limit: %d sent in burst\n",
2961 		    ipst->ips_icmp_pkt_err_sent));
2962 		return (B_FALSE);
2963 	}
2964 	ip1dbg(("icmp_err_rate_limit: dropped\n"));
2965 	return (B_TRUE);
2966 }
2967 
2968 /*
2969  * Check if it is ok to send an IPv4 ICMP error packet in
2970  * response to the IPv4 packet in mp.
2971  * Free the message and return null if no
2972  * ICMP error packet should be sent.
2973  */
2974 static mblk_t *
icmp_pkt_err_ok(mblk_t * mp,ip_recv_attr_t * ira)2975 icmp_pkt_err_ok(mblk_t *mp, ip_recv_attr_t *ira)
2976 {
2977 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
2978 	icmph_t	*icmph;
2979 	ipha_t	*ipha;
2980 	uint_t	len_needed;
2981 
2982 	if (!mp)
2983 		return (NULL);
2984 	ipha = (ipha_t *)mp->b_rptr;
2985 	if (ip_csum_hdr(ipha)) {
2986 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInCksumErrs);
2987 		ip_drop_input("ipIfStatsInCksumErrs", mp, NULL);
2988 		freemsg(mp);
2989 		return (NULL);
2990 	}
2991 	if (ip_type_v4(ipha->ipha_dst, ipst) == IRE_BROADCAST ||
2992 	    ip_type_v4(ipha->ipha_src, ipst) == IRE_BROADCAST ||
2993 	    CLASSD(ipha->ipha_dst) ||
2994 	    CLASSD(ipha->ipha_src) ||
2995 	    (ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET)) {
2996 		/* Note: only errors to the fragment with offset 0 */
2997 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
2998 		freemsg(mp);
2999 		return (NULL);
3000 	}
3001 	if (ipha->ipha_protocol == IPPROTO_ICMP) {
3002 		/*
3003 		 * Check the ICMP type.  RFC 1122 sez:  don't send ICMP
3004 		 * errors in response to any ICMP errors.
3005 		 */
3006 		len_needed = IPH_HDR_LENGTH(ipha) + ICMPH_SIZE;
3007 		if (mp->b_wptr - mp->b_rptr < len_needed) {
3008 			if (!pullupmsg(mp, len_needed)) {
3009 				BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3010 				freemsg(mp);
3011 				return (NULL);
3012 			}
3013 			ipha = (ipha_t *)mp->b_rptr;
3014 		}
3015 		icmph = (icmph_t *)
3016 		    (&((char *)ipha)[IPH_HDR_LENGTH(ipha)]);
3017 		switch (icmph->icmph_type) {
3018 		case ICMP_DEST_UNREACHABLE:
3019 		case ICMP_SOURCE_QUENCH:
3020 		case ICMP_TIME_EXCEEDED:
3021 		case ICMP_PARAM_PROBLEM:
3022 		case ICMP_REDIRECT:
3023 			BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3024 			freemsg(mp);
3025 			return (NULL);
3026 		default:
3027 			break;
3028 		}
3029 	}
3030 	/*
3031 	 * If this is a labeled system, then check to see if we're allowed to
3032 	 * send a response to this particular sender.  If not, then just drop.
3033 	 */
3034 	if (is_system_labeled() && !tsol_can_reply_error(mp, ira)) {
3035 		ip2dbg(("icmp_pkt_err_ok: can't respond to packet\n"));
3036 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3037 		freemsg(mp);
3038 		return (NULL);
3039 	}
3040 	if (icmp_err_rate_limit(ipst)) {
3041 		/*
3042 		 * Only send ICMP error packets every so often.
3043 		 * This should be done on a per port/source basis,
3044 		 * but for now this will suffice.
3045 		 */
3046 		freemsg(mp);
3047 		return (NULL);
3048 	}
3049 	return (mp);
3050 }
3051 
3052 /*
3053  * Called when a packet was sent out the same link that it arrived on.
3054  * Check if it is ok to send a redirect and then send it.
3055  */
3056 void
ip_send_potential_redirect_v4(mblk_t * mp,ipha_t * ipha,ire_t * ire,ip_recv_attr_t * ira)3057 ip_send_potential_redirect_v4(mblk_t *mp, ipha_t *ipha, ire_t *ire,
3058     ip_recv_attr_t *ira)
3059 {
3060 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
3061 	ipaddr_t	src, nhop;
3062 	mblk_t		*mp1;
3063 	ire_t		*nhop_ire;
3064 
3065 	/*
3066 	 * Check the source address to see if it originated
3067 	 * on the same logical subnet it is going back out on.
3068 	 * If so, we should be able to send it a redirect.
3069 	 * Avoid sending a redirect if the destination
3070 	 * is directly connected (i.e., we matched an IRE_ONLINK),
3071 	 * or if the packet was source routed out this interface.
3072 	 *
3073 	 * We avoid sending a redirect if the
3074 	 * destination is directly connected
3075 	 * because it is possible that multiple
3076 	 * IP subnets may have been configured on
3077 	 * the link, and the source may not
3078 	 * be on the same subnet as ip destination,
3079 	 * even though they are on the same
3080 	 * physical link.
3081 	 */
3082 	if ((ire->ire_type & IRE_ONLINK) ||
3083 	    ip_source_routed(ipha, ipst))
3084 		return;
3085 
3086 	nhop_ire = ire_nexthop(ire);
3087 	if (nhop_ire == NULL)
3088 		return;
3089 
3090 	nhop = nhop_ire->ire_addr;
3091 
3092 	if (nhop_ire->ire_type & IRE_IF_CLONE) {
3093 		ire_t	*ire2;
3094 
3095 		/* Follow ire_dep_parent to find non-clone IRE_INTERFACE */
3096 		mutex_enter(&nhop_ire->ire_lock);
3097 		ire2 = nhop_ire->ire_dep_parent;
3098 		if (ire2 != NULL)
3099 			ire_refhold(ire2);
3100 		mutex_exit(&nhop_ire->ire_lock);
3101 		ire_refrele(nhop_ire);
3102 		nhop_ire = ire2;
3103 	}
3104 	if (nhop_ire == NULL)
3105 		return;
3106 
3107 	ASSERT(!(nhop_ire->ire_type & IRE_IF_CLONE));
3108 
3109 	src = ipha->ipha_src;
3110 
3111 	/*
3112 	 * We look at the interface ire for the nexthop,
3113 	 * to see if ipha_src is in the same subnet
3114 	 * as the nexthop.
3115 	 */
3116 	if ((src & nhop_ire->ire_mask) == (nhop & nhop_ire->ire_mask)) {
3117 		/*
3118 		 * The source is directly connected.
3119 		 */
3120 		mp1 = copymsg(mp);
3121 		if (mp1 != NULL) {
3122 			icmp_send_redirect(mp1, nhop, ira);
3123 		}
3124 	}
3125 	ire_refrele(nhop_ire);
3126 }
3127 
3128 /*
3129  * Generate an ICMP redirect message.
3130  */
3131 static void
icmp_send_redirect(mblk_t * mp,ipaddr_t gateway,ip_recv_attr_t * ira)3132 icmp_send_redirect(mblk_t *mp, ipaddr_t gateway, ip_recv_attr_t *ira)
3133 {
3134 	icmph_t	icmph;
3135 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3136 
3137 	mp = icmp_pkt_err_ok(mp, ira);
3138 	if (mp == NULL)
3139 		return;
3140 
3141 	bzero(&icmph, sizeof (icmph_t));
3142 	icmph.icmph_type = ICMP_REDIRECT;
3143 	icmph.icmph_code = 1;
3144 	icmph.icmph_rd_gateway = gateway;
3145 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutRedirects);
3146 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3147 }
3148 
3149 /*
3150  * Generate an ICMP time exceeded message.
3151  */
3152 void
icmp_time_exceeded(mblk_t * mp,uint8_t code,ip_recv_attr_t * ira)3153 icmp_time_exceeded(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3154 {
3155 	icmph_t	icmph;
3156 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3157 
3158 	mp = icmp_pkt_err_ok(mp, ira);
3159 	if (mp == NULL)
3160 		return;
3161 
3162 	bzero(&icmph, sizeof (icmph_t));
3163 	icmph.icmph_type = ICMP_TIME_EXCEEDED;
3164 	icmph.icmph_code = code;
3165 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimeExcds);
3166 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3167 }
3168 
3169 /*
3170  * Generate an ICMP unreachable message.
3171  * When called from ip_output side a minimal ip_recv_attr_t needs to be
3172  * constructed by the caller.
3173  */
3174 void
icmp_unreachable(mblk_t * mp,uint8_t code,ip_recv_attr_t * ira)3175 icmp_unreachable(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3176 {
3177 	icmph_t	icmph;
3178 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3179 
3180 	mp = icmp_pkt_err_ok(mp, ira);
3181 	if (mp == NULL)
3182 		return;
3183 
3184 	bzero(&icmph, sizeof (icmph_t));
3185 	icmph.icmph_type = ICMP_DEST_UNREACHABLE;
3186 	icmph.icmph_code = code;
3187 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
3188 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3189 }
3190 
3191 /*
3192  * Latch in the IPsec state for a stream based the policy in the listener
3193  * and the actions in the ip_recv_attr_t.
3194  * Called directly from TCP and SCTP.
3195  */
3196 boolean_t
ip_ipsec_policy_inherit(conn_t * connp,conn_t * lconnp,ip_recv_attr_t * ira)3197 ip_ipsec_policy_inherit(conn_t *connp, conn_t *lconnp, ip_recv_attr_t *ira)
3198 {
3199 	ASSERT(lconnp->conn_policy != NULL);
3200 	ASSERT(connp->conn_policy == NULL);
3201 
3202 	IPPH_REFHOLD(lconnp->conn_policy);
3203 	connp->conn_policy = lconnp->conn_policy;
3204 
3205 	if (ira->ira_ipsec_action != NULL) {
3206 		if (connp->conn_latch == NULL) {
3207 			connp->conn_latch = iplatch_create();
3208 			if (connp->conn_latch == NULL)
3209 				return (B_FALSE);
3210 		}
3211 		ipsec_latch_inbound(connp, ira);
3212 	}
3213 	return (B_TRUE);
3214 }
3215 
3216 /*
3217  * Verify whether or not the IP address is a valid local address.
3218  * Could be a unicast, including one for a down interface.
3219  * If allow_mcbc then a multicast or broadcast address is also
3220  * acceptable.
3221  *
3222  * In the case of a broadcast/multicast address, however, the
3223  * upper protocol is expected to reset the src address
3224  * to zero when we return IPVL_MCAST/IPVL_BCAST so that
3225  * no packets are emitted with broadcast/multicast address as
3226  * source address (that violates hosts requirements RFC 1122)
3227  * The addresses valid for bind are:
3228  *	(1) - INADDR_ANY (0)
3229  *	(2) - IP address of an UP interface
3230  *	(3) - IP address of a DOWN interface
3231  *	(4) - valid local IP broadcast addresses. In this case
3232  *	the conn will only receive packets destined to
3233  *	the specified broadcast address.
3234  *	(5) - a multicast address. In this case
3235  *	the conn will only receive packets destined to
3236  *	the specified multicast address. Note: the
3237  *	application still has to issue an
3238  *	IP_ADD_MEMBERSHIP socket option.
3239  *
3240  * In all the above cases, the bound address must be valid in the current zone.
3241  * When the address is loopback, multicast or broadcast, there might be many
3242  * matching IREs so bind has to look up based on the zone.
3243  */
3244 ip_laddr_t
ip_laddr_verify_v4(ipaddr_t src_addr,zoneid_t zoneid,ip_stack_t * ipst,boolean_t allow_mcbc)3245 ip_laddr_verify_v4(ipaddr_t src_addr, zoneid_t zoneid,
3246     ip_stack_t *ipst, boolean_t allow_mcbc)
3247 {
3248 	ire_t *src_ire;
3249 
3250 	ASSERT(src_addr != INADDR_ANY);
3251 
3252 	src_ire = ire_ftable_lookup_v4(src_addr, 0, 0, 0,
3253 	    NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, 0, ipst, NULL);
3254 
3255 	/*
3256 	 * If an address other than in6addr_any is requested,
3257 	 * we verify that it is a valid address for bind
3258 	 * Note: Following code is in if-else-if form for
3259 	 * readability compared to a condition check.
3260 	 */
3261 	if (src_ire != NULL && (src_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK))) {
3262 		/*
3263 		 * (2) Bind to address of local UP interface
3264 		 */
3265 		ire_refrele(src_ire);
3266 		return (IPVL_UNICAST_UP);
3267 	} else if (src_ire != NULL && src_ire->ire_type & IRE_BROADCAST) {
3268 		/*
3269 		 * (4) Bind to broadcast address
3270 		 */
3271 		ire_refrele(src_ire);
3272 		if (allow_mcbc)
3273 			return (IPVL_BCAST);
3274 		else
3275 			return (IPVL_BAD);
3276 	} else if (CLASSD(src_addr)) {
3277 		/* (5) bind to multicast address. */
3278 		if (src_ire != NULL)
3279 			ire_refrele(src_ire);
3280 
3281 		if (allow_mcbc)
3282 			return (IPVL_MCAST);
3283 		else
3284 			return (IPVL_BAD);
3285 	} else {
3286 		ipif_t *ipif;
3287 
3288 		/*
3289 		 * (3) Bind to address of local DOWN interface?
3290 		 * (ipif_lookup_addr() looks up all interfaces
3291 		 * but we do not get here for UP interfaces
3292 		 * - case (2) above)
3293 		 */
3294 		if (src_ire != NULL)
3295 			ire_refrele(src_ire);
3296 
3297 		ipif = ipif_lookup_addr(src_addr, NULL, zoneid, ipst);
3298 		if (ipif == NULL)
3299 			return (IPVL_BAD);
3300 
3301 		/* Not a useful source? */
3302 		if (ipif->ipif_flags & (IPIF_NOLOCAL | IPIF_ANYCAST)) {
3303 			ipif_refrele(ipif);
3304 			return (IPVL_BAD);
3305 		}
3306 		ipif_refrele(ipif);
3307 		return (IPVL_UNICAST_DOWN);
3308 	}
3309 }
3310 
3311 /*
3312  * Insert in the bind fanout for IPv4 and IPv6.
3313  * The caller should already have used ip_laddr_verify_v*() before calling
3314  * this.
3315  */
3316 int
ip_laddr_fanout_insert(conn_t * connp)3317 ip_laddr_fanout_insert(conn_t *connp)
3318 {
3319 	int		error;
3320 
3321 	/*
3322 	 * Allow setting new policies. For example, disconnects result
3323 	 * in us being called. As we would have set conn_policy_cached
3324 	 * to B_TRUE before, we should set it to B_FALSE, so that policy
3325 	 * can change after the disconnect.
3326 	 */
3327 	connp->conn_policy_cached = B_FALSE;
3328 
3329 	error = ipcl_bind_insert(connp);
3330 	if (error != 0) {
3331 		if (connp->conn_anon_port) {
3332 			(void) tsol_mlp_anon(crgetzone(connp->conn_cred),
3333 			    connp->conn_mlp_type, connp->conn_proto,
3334 			    ntohs(connp->conn_lport), B_FALSE);
3335 		}
3336 		connp->conn_mlp_type = mlptSingle;
3337 	}
3338 	return (error);
3339 }
3340 
3341 /*
3342  * Verify that both the source and destination addresses are valid. If
3343  * IPDF_VERIFY_DST is not set, then the destination address may be unreachable,
3344  * i.e. have no route to it.  Protocols like TCP want to verify destination
3345  * reachability, while tunnels do not.
3346  *
3347  * Determine the route, the interface, and (optionally) the source address
3348  * to use to reach a given destination.
3349  * Note that we allow connect to broadcast and multicast addresses when
3350  * IPDF_ALLOW_MCBC is set.
3351  * first_hop and dst_addr are normally the same, but if source routing
3352  * they will differ; in that case the first_hop is what we'll use for the
3353  * routing lookup but the dce and label checks will be done on dst_addr,
3354  *
3355  * If uinfo is set, then we fill in the best available information
3356  * we have for the destination. This is based on (in priority order) any
3357  * metrics and path MTU stored in a dce_t, route metrics, and finally the
3358  * ill_mtu/ill_mc_mtu.
3359  *
3360  * Tsol note: If we have a source route then dst_addr != firsthop. But we
3361  * always do the label check on dst_addr.
3362  */
3363 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)3364 ip_set_destination_v4(ipaddr_t *src_addrp, ipaddr_t dst_addr, ipaddr_t firsthop,
3365     ip_xmit_attr_t *ixa, iulp_t *uinfo, uint32_t flags, uint_t mac_mode)
3366 {
3367 	ire_t		*ire = NULL;
3368 	int		error = 0;
3369 	ipaddr_t	setsrc;				/* RTF_SETSRC */
3370 	zoneid_t	zoneid = ixa->ixa_zoneid;	/* Honors SO_ALLZONES */
3371 	ip_stack_t	*ipst = ixa->ixa_ipst;
3372 	dce_t		*dce;
3373 	uint_t		pmtu;
3374 	uint_t		generation;
3375 	nce_t		*nce;
3376 	ill_t		*ill = NULL;
3377 	boolean_t	multirt = B_FALSE;
3378 
3379 	ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
3380 
3381 	/*
3382 	 * We never send to zero; the ULPs map it to the loopback address.
3383 	 * We can't allow it since we use zero to mean unitialized in some
3384 	 * places.
3385 	 */
3386 	ASSERT(dst_addr != INADDR_ANY);
3387 
3388 	if (is_system_labeled()) {
3389 		ts_label_t *tsl = NULL;
3390 
3391 		error = tsol_check_dest(ixa->ixa_tsl, &dst_addr, IPV4_VERSION,
3392 		    mac_mode, (flags & IPDF_ZONE_IS_GLOBAL) != 0, &tsl);
3393 		if (error != 0)
3394 			return (error);
3395 		if (tsl != NULL) {
3396 			/* Update the label */
3397 			ip_xmit_attr_replace_tsl(ixa, tsl);
3398 		}
3399 	}
3400 
3401 	setsrc = INADDR_ANY;
3402 	/*
3403 	 * Select a route; For IPMP interfaces, we would only select
3404 	 * a "hidden" route (i.e., going through a specific under_ill)
3405 	 * if ixa_ifindex has been specified.
3406 	 */
3407 	ire = ip_select_route_v4(firsthop, *src_addrp, ixa,
3408 	    &generation, &setsrc, &error, &multirt);
3409 	ASSERT(ire != NULL);	/* IRE_NOROUTE if none found */
3410 	if (error != 0)
3411 		goto bad_addr;
3412 
3413 	/*
3414 	 * ire can't be a broadcast or multicast unless IPDF_ALLOW_MCBC is set.
3415 	 * If IPDF_VERIFY_DST is set, the destination must be reachable;
3416 	 * Otherwise the destination needn't be reachable.
3417 	 *
3418 	 * If we match on a reject or black hole, then we've got a
3419 	 * local failure.  May as well fail out the connect() attempt,
3420 	 * since it's never going to succeed.
3421 	 */
3422 	if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
3423 		/*
3424 		 * If we're verifying destination reachability, we always want
3425 		 * to complain here.
3426 		 *
3427 		 * If we're not verifying destination reachability but the
3428 		 * destination has a route, we still want to fail on the
3429 		 * temporary address and broadcast address tests.
3430 		 *
3431 		 * In both cases do we let the code continue so some reasonable
3432 		 * information is returned to the caller. That enables the
3433 		 * caller to use (and even cache) the IRE. conn_ip_ouput will
3434 		 * use the generation mismatch path to check for the unreachable
3435 		 * case thereby avoiding any specific check in the main path.
3436 		 */
3437 		ASSERT(generation == IRE_GENERATION_VERIFY);
3438 		if (flags & IPDF_VERIFY_DST) {
3439 			/*
3440 			 * Set errno but continue to set up ixa_ire to be
3441 			 * the RTF_REJECT|RTF_BLACKHOLE IRE.
3442 			 * That allows callers to use ip_output to get an
3443 			 * ICMP error back.
3444 			 */
3445 			if (!(ire->ire_type & IRE_HOST))
3446 				error = ENETUNREACH;
3447 			else
3448 				error = EHOSTUNREACH;
3449 		}
3450 	}
3451 
3452 	if ((ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST)) &&
3453 	    !(flags & IPDF_ALLOW_MCBC)) {
3454 		ire_refrele(ire);
3455 		ire = ire_reject(ipst, B_FALSE);
3456 		generation = IRE_GENERATION_VERIFY;
3457 		error = ENETUNREACH;
3458 	}
3459 
3460 	/* Cache things */
3461 	if (ixa->ixa_ire != NULL)
3462 		ire_refrele_notr(ixa->ixa_ire);
3463 #ifdef DEBUG
3464 	ire_refhold_notr(ire);
3465 	ire_refrele(ire);
3466 #endif
3467 	ixa->ixa_ire = ire;
3468 	ixa->ixa_ire_generation = generation;
3469 
3470 	/*
3471 	 * Ensure that ixa_dce is always set any time that ixa_ire is set,
3472 	 * since some callers will send a packet to conn_ip_output() even if
3473 	 * there's an error.
3474 	 */
3475 	if (flags & IPDF_UNIQUE_DCE) {
3476 		/* Fallback to the default dce if allocation fails */
3477 		dce = dce_lookup_and_add_v4(dst_addr, ipst);
3478 		if (dce != NULL)
3479 			generation = dce->dce_generation;
3480 		else
3481 			dce = dce_lookup_v4(dst_addr, ipst, &generation);
3482 	} else {
3483 		dce = dce_lookup_v4(dst_addr, ipst, &generation);
3484 	}
3485 	ASSERT(dce != NULL);
3486 	if (ixa->ixa_dce != NULL)
3487 		dce_refrele_notr(ixa->ixa_dce);
3488 #ifdef DEBUG
3489 	dce_refhold_notr(dce);
3490 	dce_refrele(dce);
3491 #endif
3492 	ixa->ixa_dce = dce;
3493 	ixa->ixa_dce_generation = generation;
3494 
3495 	/*
3496 	 * For multicast with multirt we have a flag passed back from
3497 	 * ire_lookup_multi_ill_v4 since we don't have an IRE for each
3498 	 * possible multicast address.
3499 	 * We also need a flag for multicast since we can't check
3500 	 * whether RTF_MULTIRT is set in ixa_ire for multicast.
3501 	 */
3502 	if (multirt) {
3503 		ixa->ixa_postfragfn = ip_postfrag_multirt_v4;
3504 		ixa->ixa_flags |= IXAF_MULTIRT_MULTICAST;
3505 	} else {
3506 		ixa->ixa_postfragfn = ire->ire_postfragfn;
3507 		ixa->ixa_flags &= ~IXAF_MULTIRT_MULTICAST;
3508 	}
3509 	if (!(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3510 		/* Get an nce to cache. */
3511 		nce = ire_to_nce(ire, firsthop, NULL);
3512 		if (nce == NULL) {
3513 			/* Allocation failure? */
3514 			ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3515 		} else {
3516 			if (ixa->ixa_nce != NULL)
3517 				nce_refrele(ixa->ixa_nce);
3518 			ixa->ixa_nce = nce;
3519 		}
3520 	}
3521 
3522 	/*
3523 	 * If the source address is a loopback address, the
3524 	 * destination had best be local or multicast.
3525 	 * If we are sending to an IRE_LOCAL using a loopback source then
3526 	 * it had better be the same zoneid.
3527 	 */
3528 	if (*src_addrp == htonl(INADDR_LOOPBACK)) {
3529 		if ((ire->ire_type & IRE_LOCAL) && ire->ire_zoneid != zoneid) {
3530 			ire = NULL;	/* Stored in ixa_ire */
3531 			error = EADDRNOTAVAIL;
3532 			goto bad_addr;
3533 		}
3534 		if (!(ire->ire_type & (IRE_LOOPBACK|IRE_LOCAL|IRE_MULTICAST))) {
3535 			ire = NULL;	/* Stored in ixa_ire */
3536 			error = EADDRNOTAVAIL;
3537 			goto bad_addr;
3538 		}
3539 	}
3540 	if (ire->ire_type & IRE_BROADCAST) {
3541 		/*
3542 		 * If the ULP didn't have a specified source, then we
3543 		 * make sure we reselect the source when sending
3544 		 * broadcasts out different interfaces.
3545 		 */
3546 		if (flags & IPDF_SELECT_SRC)
3547 			ixa->ixa_flags |= IXAF_SET_SOURCE;
3548 		else
3549 			ixa->ixa_flags &= ~IXAF_SET_SOURCE;
3550 	}
3551 
3552 	/*
3553 	 * Does the caller want us to pick a source address?
3554 	 */
3555 	if (flags & IPDF_SELECT_SRC) {
3556 		ipaddr_t	src_addr;
3557 
3558 		/*
3559 		 * We use use ire_nexthop_ill to avoid the under ipmp
3560 		 * interface for source address selection. Note that for ipmp
3561 		 * probe packets, ixa_ifindex would have been specified, and
3562 		 * the ip_select_route() invocation would have picked an ire
3563 		 * will ire_ill pointing at an under interface.
3564 		 */
3565 		ill = ire_nexthop_ill(ire);
3566 
3567 		/* If unreachable we have no ill but need some source */
3568 		if (ill == NULL) {
3569 			src_addr = htonl(INADDR_LOOPBACK);
3570 			/* Make sure we look for a better source address */
3571 			generation = SRC_GENERATION_VERIFY;
3572 		} else {
3573 			error = ip_select_source_v4(ill, setsrc, dst_addr,
3574 			    ixa->ixa_multicast_ifaddr, zoneid,
3575 			    ipst, &src_addr, &generation, NULL);
3576 			if (error != 0) {
3577 				ire = NULL;	/* Stored in ixa_ire */
3578 				goto bad_addr;
3579 			}
3580 		}
3581 
3582 		/*
3583 		 * We allow the source address to to down.
3584 		 * However, we check that we don't use the loopback address
3585 		 * as a source when sending out on the wire.
3586 		 */
3587 		if ((src_addr == htonl(INADDR_LOOPBACK)) &&
3588 		    !(ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK|IRE_MULTICAST)) &&
3589 		    !(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3590 			ire = NULL;	/* Stored in ixa_ire */
3591 			error = EADDRNOTAVAIL;
3592 			goto bad_addr;
3593 		}
3594 
3595 		*src_addrp = src_addr;
3596 		ixa->ixa_src_generation = generation;
3597 	}
3598 
3599 	/*
3600 	 * Make sure we don't leave an unreachable ixa_nce in place
3601 	 * since ip_select_route is used when we unplumb i.e., remove
3602 	 * references on ixa_ire, ixa_nce, and ixa_dce.
3603 	 */
3604 	nce = ixa->ixa_nce;
3605 	if (nce != NULL && nce->nce_is_condemned) {
3606 		nce_refrele(nce);
3607 		ixa->ixa_nce = NULL;
3608 		ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3609 	}
3610 
3611 	/*
3612 	 * The caller has set IXAF_PMTU_DISCOVERY if path MTU is desired.
3613 	 * However, we can't do it for IPv4 multicast or broadcast.
3614 	 */
3615 	if (ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST))
3616 		ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3617 
3618 	/*
3619 	 * Set initial value for fragmentation limit. Either conn_ip_output
3620 	 * or ULP might updates it when there are routing changes.
3621 	 * Handles a NULL ixa_ire->ire_ill or a NULL ixa_nce for RTF_REJECT.
3622 	 */
3623 	pmtu = ip_get_pmtu(ixa);
3624 	ixa->ixa_fragsize = pmtu;
3625 	/* Make sure ixa_fragsize and ixa_pmtu remain identical */
3626 	if (ixa->ixa_flags & IXAF_VERIFY_PMTU)
3627 		ixa->ixa_pmtu = pmtu;
3628 
3629 	/*
3630 	 * Extract information useful for some transports.
3631 	 * First we look for DCE metrics. Then we take what we have in
3632 	 * the metrics in the route, where the offlink is used if we have
3633 	 * one.
3634 	 */
3635 	if (uinfo != NULL) {
3636 		bzero(uinfo, sizeof (*uinfo));
3637 
3638 		if (dce->dce_flags & DCEF_UINFO)
3639 			*uinfo = dce->dce_uinfo;
3640 
3641 		rts_merge_metrics(uinfo, &ire->ire_metrics);
3642 
3643 		/* Allow ire_metrics to decrease the path MTU from above */
3644 		if (uinfo->iulp_mtu == 0 || uinfo->iulp_mtu > pmtu)
3645 			uinfo->iulp_mtu = pmtu;
3646 
3647 		uinfo->iulp_localnet = (ire->ire_type & IRE_ONLINK) != 0;
3648 		uinfo->iulp_loopback = (ire->ire_type & IRE_LOOPBACK) != 0;
3649 		uinfo->iulp_local = (ire->ire_type & IRE_LOCAL) != 0;
3650 	}
3651 
3652 	if (ill != NULL)
3653 		ill_refrele(ill);
3654 
3655 	return (error);
3656 
3657 bad_addr:
3658 	if (ire != NULL)
3659 		ire_refrele(ire);
3660 
3661 	if (ill != NULL)
3662 		ill_refrele(ill);
3663 
3664 	/*
3665 	 * Make sure we don't leave an unreachable ixa_nce in place
3666 	 * since ip_select_route is used when we unplumb i.e., remove
3667 	 * references on ixa_ire, ixa_nce, and ixa_dce.
3668 	 */
3669 	nce = ixa->ixa_nce;
3670 	if (nce != NULL && nce->nce_is_condemned) {
3671 		nce_refrele(nce);
3672 		ixa->ixa_nce = NULL;
3673 		ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3674 	}
3675 
3676 	return (error);
3677 }
3678 
3679 
3680 /*
3681  * Get the base MTU for the case when path MTU discovery is not used.
3682  * Takes the MTU of the IRE into account.
3683  */
3684 uint_t
ip_get_base_mtu(ill_t * ill,ire_t * ire)3685 ip_get_base_mtu(ill_t *ill, ire_t *ire)
3686 {
3687 	uint_t mtu;
3688 	uint_t iremtu = ire->ire_metrics.iulp_mtu;
3689 
3690 	if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST))
3691 		mtu = ill->ill_mc_mtu;
3692 	else
3693 		mtu = ill->ill_mtu;
3694 
3695 	if (iremtu != 0 && iremtu < mtu)
3696 		mtu = iremtu;
3697 
3698 	return (mtu);
3699 }
3700 
3701 /*
3702  * Get the PMTU for the attributes. Handles both IPv4 and IPv6.
3703  * Assumes that ixa_ire, dce, and nce have already been set up.
3704  *
3705  * The caller has set IXAF_PMTU_DISCOVERY if path MTU discovery is desired.
3706  * We avoid path MTU discovery if it is disabled with ndd.
3707  * Furtermore, if the path MTU is too small, then we don't set DF for IPv4.
3708  *
3709  * NOTE: We also used to turn it off for source routed packets. That
3710  * is no longer required since the dce is per final destination.
3711  */
3712 uint_t
ip_get_pmtu(ip_xmit_attr_t * ixa)3713 ip_get_pmtu(ip_xmit_attr_t *ixa)
3714 {
3715 	ip_stack_t	*ipst = ixa->ixa_ipst;
3716 	dce_t		*dce;
3717 	nce_t		*nce;
3718 	ire_t		*ire;
3719 	uint_t		pmtu;
3720 
3721 	ire = ixa->ixa_ire;
3722 	dce = ixa->ixa_dce;
3723 	nce = ixa->ixa_nce;
3724 
3725 	/*
3726 	 * If path MTU discovery has been turned off by ndd, then we ignore
3727 	 * any dce_pmtu and for IPv4 we will not set DF.
3728 	 */
3729 	if (!ipst->ips_ip_path_mtu_discovery)
3730 		ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3731 
3732 	pmtu = IP_MAXPACKET;
3733 	/*
3734 	 * Decide whether whether IPv4 sets DF
3735 	 * For IPv6 "no DF" means to use the 1280 mtu
3736 	 */
3737 	if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3738 		ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3739 	} else {
3740 		ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3741 		if (!(ixa->ixa_flags & IXAF_IS_IPV4))
3742 			pmtu = IPV6_MIN_MTU;
3743 	}
3744 
3745 	/* Check if the PMTU is to old before we use it */
3746 	if ((dce->dce_flags & DCEF_PMTU) &&
3747 	    TICK_TO_SEC(ddi_get_lbolt64()) - dce->dce_last_change_time >
3748 	    ipst->ips_ip_pathmtu_interval) {
3749 		/*
3750 		 * Older than 20 minutes. Drop the path MTU information.
3751 		 */
3752 		mutex_enter(&dce->dce_lock);
3753 		dce->dce_flags &= ~(DCEF_PMTU|DCEF_TOO_SMALL_PMTU);
3754 		dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
3755 		mutex_exit(&dce->dce_lock);
3756 		dce_increment_generation(dce);
3757 	}
3758 
3759 	/* The metrics on the route can lower the path MTU */
3760 	if (ire->ire_metrics.iulp_mtu != 0 &&
3761 	    ire->ire_metrics.iulp_mtu < pmtu)
3762 		pmtu = ire->ire_metrics.iulp_mtu;
3763 
3764 	/*
3765 	 * If the path MTU is smaller than some minimum, we still use dce_pmtu
3766 	 * above (would be 576 for IPv4 and 1280 for IPv6), but we clear
3767 	 * IXAF_PMTU_IPV4_DF so that we avoid setting DF for IPv4.
3768 	 */
3769 	if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3770 		if (dce->dce_flags & DCEF_PMTU) {
3771 			if (dce->dce_pmtu < pmtu)
3772 				pmtu = dce->dce_pmtu;
3773 
3774 			if (dce->dce_flags & DCEF_TOO_SMALL_PMTU) {
3775 				ixa->ixa_flags |= IXAF_PMTU_TOO_SMALL;
3776 				ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3777 			} else {
3778 				ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3779 				ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3780 			}
3781 		} else {
3782 			ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3783 			ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3784 		}
3785 	}
3786 
3787 	/*
3788 	 * If we have an IRE_LOCAL we use the loopback mtu instead of
3789 	 * the ill for going out the wire i.e., IRE_LOCAL gets the same
3790 	 * mtu as IRE_LOOPBACK.
3791 	 */
3792 	if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) {
3793 		uint_t loopback_mtu;
3794 
3795 		loopback_mtu = (ire->ire_ipversion == IPV6_VERSION) ?
3796 		    ip_loopback_mtu_v6plus : ip_loopback_mtuplus;
3797 
3798 		if (loopback_mtu < pmtu)
3799 			pmtu = loopback_mtu;
3800 	} else if (nce != NULL) {
3801 		/*
3802 		 * Make sure we don't exceed the interface MTU.
3803 		 * In the case of RTF_REJECT or RTF_BLACKHOLE we might not have
3804 		 * an ill. We'd use the above IP_MAXPACKET in that case just
3805 		 * to tell the transport something larger than zero.
3806 		 */
3807 		if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST)) {
3808 			if (nce->nce_common->ncec_ill->ill_mc_mtu < pmtu)
3809 				pmtu = nce->nce_common->ncec_ill->ill_mc_mtu;
3810 			if (nce->nce_common->ncec_ill != nce->nce_ill &&
3811 			    nce->nce_ill->ill_mc_mtu < pmtu) {
3812 				/*
3813 				 * for interfaces in an IPMP group, the mtu of
3814 				 * the nce_ill (under_ill) could be different
3815 				 * from the mtu of the ncec_ill, so we take the
3816 				 * min of the two.
3817 				 */
3818 				pmtu = nce->nce_ill->ill_mc_mtu;
3819 			}
3820 		} else {
3821 			if (nce->nce_common->ncec_ill->ill_mtu < pmtu)
3822 				pmtu = nce->nce_common->ncec_ill->ill_mtu;
3823 			if (nce->nce_common->ncec_ill != nce->nce_ill &&
3824 			    nce->nce_ill->ill_mtu < pmtu) {
3825 				/*
3826 				 * for interfaces in an IPMP group, the mtu of
3827 				 * the nce_ill (under_ill) could be different
3828 				 * from the mtu of the ncec_ill, so we take the
3829 				 * min of the two.
3830 				 */
3831 				pmtu = nce->nce_ill->ill_mtu;
3832 			}
3833 		}
3834 	}
3835 
3836 	/*
3837 	 * Handle the IPV6_USE_MIN_MTU socket option or ancillary data.
3838 	 * Only applies to IPv6.
3839 	 */
3840 	if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3841 		if (ixa->ixa_flags & IXAF_USE_MIN_MTU) {
3842 			switch (ixa->ixa_use_min_mtu) {
3843 			case IPV6_USE_MIN_MTU_MULTICAST:
3844 				if (ire->ire_type & IRE_MULTICAST)
3845 					pmtu = IPV6_MIN_MTU;
3846 				break;
3847 			case IPV6_USE_MIN_MTU_ALWAYS:
3848 				pmtu = IPV6_MIN_MTU;
3849 				break;
3850 			case IPV6_USE_MIN_MTU_NEVER:
3851 				break;
3852 			}
3853 		} else {
3854 			/* Default is IPV6_USE_MIN_MTU_MULTICAST */
3855 			if (ire->ire_type & IRE_MULTICAST)
3856 				pmtu = IPV6_MIN_MTU;
3857 		}
3858 	}
3859 
3860 	/*
3861 	 * After receiving an ICMPv6 "packet too big" message with a
3862 	 * MTU < 1280, and for multirouted IPv6 packets, the IP layer
3863 	 * will insert a 8-byte fragment header in every packet. We compensate
3864 	 * for those cases by returning a smaller path MTU to the ULP.
3865 	 *
3866 	 * In the case of CGTP then ip_output will add a fragment header.
3867 	 * Make sure there is room for it by telling a smaller number
3868 	 * to the transport.
3869 	 *
3870 	 * When IXAF_IPV6_ADDR_FRAGHDR we subtract the frag hdr here
3871 	 * so the ULPs consistently see a iulp_pmtu and ip_get_pmtu()
3872 	 * which is the size of the packets it can send.
3873 	 */
3874 	if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3875 		if ((dce->dce_flags & DCEF_TOO_SMALL_PMTU) ||
3876 		    (ire->ire_flags & RTF_MULTIRT) ||
3877 		    (ixa->ixa_flags & IXAF_MULTIRT_MULTICAST)) {
3878 			pmtu -= sizeof (ip6_frag_t);
3879 			ixa->ixa_flags |= IXAF_IPV6_ADD_FRAGHDR;
3880 		}
3881 	}
3882 
3883 	return (pmtu);
3884 }
3885 
3886 /*
3887  * Carve "len" bytes out of an mblk chain, consuming any we empty, and duping
3888  * the final piece where we don't.  Return a pointer to the first mblk in the
3889  * result, and update the pointer to the next mblk to chew on.  If anything
3890  * goes wrong (i.e., dupb fails), we waste everything in sight and return a
3891  * NULL pointer.
3892  */
3893 mblk_t *
ip_carve_mp(mblk_t ** mpp,ssize_t len)3894 ip_carve_mp(mblk_t **mpp, ssize_t len)
3895 {
3896 	mblk_t	*mp0;
3897 	mblk_t	*mp1;
3898 	mblk_t	*mp2;
3899 
3900 	if (!len || !mpp || !(mp0 = *mpp))
3901 		return (NULL);
3902 	/* If we aren't going to consume the first mblk, we need a dup. */
3903 	if (mp0->b_wptr - mp0->b_rptr > len) {
3904 		mp1 = dupb(mp0);
3905 		if (mp1) {
3906 			/* Partition the data between the two mblks. */
3907 			mp1->b_wptr = mp1->b_rptr + len;
3908 			mp0->b_rptr = mp1->b_wptr;
3909 			/*
3910 			 * after adjustments if mblk not consumed is now
3911 			 * unaligned, try to align it. If this fails free
3912 			 * all messages and let upper layer recover.
3913 			 */
3914 			if (!OK_32PTR(mp0->b_rptr)) {
3915 				if (!pullupmsg(mp0, -1)) {
3916 					freemsg(mp0);
3917 					freemsg(mp1);
3918 					*mpp = NULL;
3919 					return (NULL);
3920 				}
3921 			}
3922 		}
3923 		return (mp1);
3924 	}
3925 	/* Eat through as many mblks as we need to get len bytes. */
3926 	len -= mp0->b_wptr - mp0->b_rptr;
3927 	for (mp2 = mp1 = mp0; (mp2 = mp2->b_cont) != 0 && len; mp1 = mp2) {
3928 		if (mp2->b_wptr - mp2->b_rptr > len) {
3929 			/*
3930 			 * We won't consume the entire last mblk.  Like
3931 			 * above, dup and partition it.
3932 			 */
3933 			mp1->b_cont = dupb(mp2);
3934 			mp1 = mp1->b_cont;
3935 			if (!mp1) {
3936 				/*
3937 				 * Trouble.  Rather than go to a lot of
3938 				 * trouble to clean up, we free the messages.
3939 				 * This won't be any worse than losing it on
3940 				 * the wire.
3941 				 */
3942 				freemsg(mp0);
3943 				freemsg(mp2);
3944 				*mpp = NULL;
3945 				return (NULL);
3946 			}
3947 			mp1->b_wptr = mp1->b_rptr + len;
3948 			mp2->b_rptr = mp1->b_wptr;
3949 			/*
3950 			 * after adjustments if mblk not consumed is now
3951 			 * unaligned, try to align it. If this fails free
3952 			 * all messages and let upper layer recover.
3953 			 */
3954 			if (!OK_32PTR(mp2->b_rptr)) {
3955 				if (!pullupmsg(mp2, -1)) {
3956 					freemsg(mp0);
3957 					freemsg(mp2);
3958 					*mpp = NULL;
3959 					return (NULL);
3960 				}
3961 			}
3962 			*mpp = mp2;
3963 			return (mp0);
3964 		}
3965 		/* Decrement len by the amount we just got. */
3966 		len -= mp2->b_wptr - mp2->b_rptr;
3967 	}
3968 	/*
3969 	 * len should be reduced to zero now.  If not our caller has
3970 	 * screwed up.
3971 	 */
3972 	if (len) {
3973 		/* Shouldn't happen! */
3974 		freemsg(mp0);
3975 		*mpp = NULL;
3976 		return (NULL);
3977 	}
3978 	/*
3979 	 * We consumed up to exactly the end of an mblk.  Detach the part
3980 	 * we are returning from the rest of the chain.
3981 	 */
3982 	mp1->b_cont = NULL;
3983 	*mpp = mp2;
3984 	return (mp0);
3985 }
3986 
3987 /* The ill stream is being unplumbed. Called from ip_close */
3988 int
ip_modclose(ill_t * ill)3989 ip_modclose(ill_t *ill)
3990 {
3991 	boolean_t success;
3992 	ipsq_t	*ipsq;
3993 	ipif_t	*ipif;
3994 	queue_t	*q = ill->ill_rq;
3995 	ip_stack_t	*ipst = ill->ill_ipst;
3996 	int	i;
3997 	arl_ill_common_t *ai = ill->ill_common;
3998 
3999 	/*
4000 	 * The punlink prior to this may have initiated a capability
4001 	 * negotiation. But ipsq_enter will block until that finishes or
4002 	 * times out.
4003 	 */
4004 	success = ipsq_enter(ill, B_FALSE, NEW_OP);
4005 
4006 	/*
4007 	 * Open/close/push/pop is guaranteed to be single threaded
4008 	 * per stream by STREAMS. FS guarantees that all references
4009 	 * from top are gone before close is called. So there can't
4010 	 * be another close thread that has set CONDEMNED on this ill.
4011 	 * and cause ipsq_enter to return failure.
4012 	 */
4013 	ASSERT(success);
4014 	ipsq = ill->ill_phyint->phyint_ipsq;
4015 
4016 	/*
4017 	 * Mark it condemned. No new reference will be made to this ill.
4018 	 * Lookup functions will return an error. Threads that try to
4019 	 * increment the refcnt must check for ILL_CAN_LOOKUP. This ensures
4020 	 * that the refcnt will drop down to zero.
4021 	 */
4022 	mutex_enter(&ill->ill_lock);
4023 	ill->ill_state_flags |= ILL_CONDEMNED;
4024 	for (ipif = ill->ill_ipif; ipif != NULL;
4025 	    ipif = ipif->ipif_next) {
4026 		ipif->ipif_state_flags |= IPIF_CONDEMNED;
4027 	}
4028 	/*
4029 	 * Wake up anybody waiting to enter the ipsq. ipsq_enter
4030 	 * returns  error if ILL_CONDEMNED is set
4031 	 */
4032 	cv_broadcast(&ill->ill_cv);
4033 	mutex_exit(&ill->ill_lock);
4034 
4035 	/*
4036 	 * Send all the deferred DLPI messages downstream which came in
4037 	 * during the small window right before ipsq_enter(). We do this
4038 	 * without waiting for the ACKs because all the ACKs for M_PROTO
4039 	 * messages are ignored in ip_rput() when ILL_CONDEMNED is set.
4040 	 */
4041 	ill_dlpi_send_deferred(ill);
4042 
4043 	/*
4044 	 * Shut down fragmentation reassembly.
4045 	 * ill_frag_timer won't start a timer again.
4046 	 * Now cancel any existing timer
4047 	 */
4048 	(void) untimeout(ill->ill_frag_timer_id);
4049 	(void) ill_frag_timeout(ill, 0);
4050 
4051 	/*
4052 	 * Call ill_delete to bring down the ipifs, ilms and ill on
4053 	 * this ill. Then wait for the refcnts to drop to zero.
4054 	 * ill_is_freeable checks whether the ill is really quiescent.
4055 	 * Then make sure that threads that are waiting to enter the
4056 	 * ipsq have seen the error returned by ipsq_enter and have
4057 	 * gone away. Then we call ill_delete_tail which does the
4058 	 * DL_UNBIND_REQ with the driver and then qprocsoff.
4059 	 */
4060 	ill_delete(ill);
4061 	mutex_enter(&ill->ill_lock);
4062 	while (!ill_is_freeable(ill))
4063 		cv_wait(&ill->ill_cv, &ill->ill_lock);
4064 
4065 	while (ill->ill_waiters)
4066 		cv_wait(&ill->ill_cv, &ill->ill_lock);
4067 
4068 	mutex_exit(&ill->ill_lock);
4069 
4070 	/*
4071 	 * ill_delete_tail drops reference on ill_ipst, but we need to keep
4072 	 * it held until the end of the function since the cleanup
4073 	 * below needs to be able to use the ip_stack_t.
4074 	 */
4075 	netstack_hold(ipst->ips_netstack);
4076 
4077 	/* qprocsoff is done via ill_delete_tail */
4078 	ill_delete_tail(ill);
4079 	/*
4080 	 * synchronously wait for arp stream to unbind. After this, we
4081 	 * cannot get any data packets up from the driver.
4082 	 */
4083 	arp_unbind_complete(ill);
4084 	ASSERT(ill->ill_ipst == NULL);
4085 
4086 	/*
4087 	 * Walk through all conns and qenable those that have queued data.
4088 	 * Close synchronization needs this to
4089 	 * be done to ensure that all upper layers blocked
4090 	 * due to flow control to the closing device
4091 	 * get unblocked.
4092 	 */
4093 	ip1dbg(("ip_wsrv: walking\n"));
4094 	for (i = 0; i < TX_FANOUT_SIZE; i++) {
4095 		conn_walk_drain(ipst, &ipst->ips_idl_tx_list[i]);
4096 	}
4097 
4098 	/*
4099 	 * ai can be null if this is an IPv6 ill, or if the IPv4
4100 	 * stream is being torn down before ARP was plumbed (e.g.,
4101 	 * /sbin/ifconfig plumbing a stream twice, and encountering
4102 	 * an error
4103 	 */
4104 	if (ai != NULL) {
4105 		ASSERT(!ill->ill_isv6);
4106 		mutex_enter(&ai->ai_lock);
4107 		ai->ai_ill = NULL;
4108 		if (ai->ai_arl == NULL) {
4109 			mutex_destroy(&ai->ai_lock);
4110 			kmem_free(ai, sizeof (*ai));
4111 		} else {
4112 			cv_signal(&ai->ai_ill_unplumb_done);
4113 			mutex_exit(&ai->ai_lock);
4114 		}
4115 	}
4116 
4117 	mutex_enter(&ipst->ips_ip_mi_lock);
4118 	mi_close_unlink(&ipst->ips_ip_g_head, (IDP)ill);
4119 	mutex_exit(&ipst->ips_ip_mi_lock);
4120 
4121 	/*
4122 	 * credp could be null if the open didn't succeed and ip_modopen
4123 	 * itself calls ip_close.
4124 	 */
4125 	if (ill->ill_credp != NULL)
4126 		crfree(ill->ill_credp);
4127 
4128 	mutex_destroy(&ill->ill_saved_ire_lock);
4129 	mutex_destroy(&ill->ill_lock);
4130 	rw_destroy(&ill->ill_mcast_lock);
4131 	mutex_destroy(&ill->ill_mcast_serializer);
4132 	list_destroy(&ill->ill_nce);
4133 
4134 	/*
4135 	 * Now we are done with the module close pieces that
4136 	 * need the netstack_t.
4137 	 */
4138 	netstack_rele(ipst->ips_netstack);
4139 
4140 	mi_close_free((IDP)ill);
4141 	q->q_ptr = WR(q)->q_ptr = NULL;
4142 
4143 	ipsq_exit(ipsq);
4144 
4145 	return (0);
4146 }
4147 
4148 /*
4149  * This is called as part of close() for IP, UDP, ICMP, and RTS
4150  * in order to quiesce the conn.
4151  */
4152 void
ip_quiesce_conn(conn_t * connp)4153 ip_quiesce_conn(conn_t *connp)
4154 {
4155 	boolean_t	drain_cleanup_reqd = B_FALSE;
4156 	boolean_t	conn_ioctl_cleanup_reqd = B_FALSE;
4157 	boolean_t	ilg_cleanup_reqd = B_FALSE;
4158 	ip_stack_t	*ipst;
4159 
4160 	ASSERT(!IPCL_IS_TCP(connp));
4161 	ipst = connp->conn_netstack->netstack_ip;
4162 
4163 	/*
4164 	 * Mark the conn as closing, and this conn must not be
4165 	 * inserted in future into any list. Eg. conn_drain_insert(),
4166 	 * won't insert this conn into the conn_drain_list.
4167 	 *
4168 	 * conn_idl, and conn_ilg cannot get set henceforth.
4169 	 */
4170 	mutex_enter(&connp->conn_lock);
4171 	ASSERT(!(connp->conn_state_flags & CONN_QUIESCED));
4172 	connp->conn_state_flags |= CONN_CLOSING;
4173 	if (connp->conn_idl != NULL)
4174 		drain_cleanup_reqd = B_TRUE;
4175 	if (connp->conn_oper_pending_ill != NULL)
4176 		conn_ioctl_cleanup_reqd = B_TRUE;
4177 	if (connp->conn_dhcpinit_ill != NULL) {
4178 		ASSERT(connp->conn_dhcpinit_ill->ill_dhcpinit != 0);
4179 		atomic_dec_32(&connp->conn_dhcpinit_ill->ill_dhcpinit);
4180 		ill_set_inputfn(connp->conn_dhcpinit_ill);
4181 		connp->conn_dhcpinit_ill = NULL;
4182 	}
4183 	if (connp->conn_ilg != NULL)
4184 		ilg_cleanup_reqd = B_TRUE;
4185 	mutex_exit(&connp->conn_lock);
4186 
4187 	if (conn_ioctl_cleanup_reqd)
4188 		conn_ioctl_cleanup(connp);
4189 
4190 	if (is_system_labeled() && connp->conn_anon_port) {
4191 		(void) tsol_mlp_anon(crgetzone(connp->conn_cred),
4192 		    connp->conn_mlp_type, connp->conn_proto,
4193 		    ntohs(connp->conn_lport), B_FALSE);
4194 		connp->conn_anon_port = 0;
4195 	}
4196 	connp->conn_mlp_type = mlptSingle;
4197 
4198 	/*
4199 	 * Remove this conn from any fanout list it is on.
4200 	 * and then wait for any threads currently operating
4201 	 * on this endpoint to finish
4202 	 */
4203 	ipcl_hash_remove(connp);
4204 
4205 	/*
4206 	 * Remove this conn from the drain list, and do any other cleanup that
4207 	 * may be required.  (TCP conns are never flow controlled, and
4208 	 * conn_idl will be NULL.)
4209 	 */
4210 	if (drain_cleanup_reqd && connp->conn_idl != NULL) {
4211 		idl_t *idl = connp->conn_idl;
4212 
4213 		mutex_enter(&idl->idl_lock);
4214 		conn_drain(connp, B_TRUE);
4215 		mutex_exit(&idl->idl_lock);
4216 	}
4217 
4218 	if (connp == ipst->ips_ip_g_mrouter)
4219 		(void) ip_mrouter_done(ipst);
4220 
4221 	if (ilg_cleanup_reqd)
4222 		ilg_delete_all(connp);
4223 
4224 	/*
4225 	 * Now conn refcnt can increase only thru CONN_INC_REF_LOCKED.
4226 	 * callers from write side can't be there now because close
4227 	 * is in progress. The only other caller is ipcl_walk
4228 	 * which checks for the condemned flag.
4229 	 */
4230 	mutex_enter(&connp->conn_lock);
4231 	connp->conn_state_flags |= CONN_CONDEMNED;
4232 	while (connp->conn_ref != 1)
4233 		cv_wait(&connp->conn_cv, &connp->conn_lock);
4234 	connp->conn_state_flags |= CONN_QUIESCED;
4235 	mutex_exit(&connp->conn_lock);
4236 }
4237 
4238 /* ARGSUSED */
4239 int
ip_close(queue_t * q,int flags)4240 ip_close(queue_t *q, int flags)
4241 {
4242 	conn_t		*connp;
4243 
4244 	/*
4245 	 * Call the appropriate delete routine depending on whether this is
4246 	 * a module or device.
4247 	 */
4248 	if (WR(q)->q_next != NULL) {
4249 		/* This is a module close */
4250 		return (ip_modclose((ill_t *)q->q_ptr));
4251 	}
4252 
4253 	connp = q->q_ptr;
4254 	ip_quiesce_conn(connp);
4255 
4256 	qprocsoff(q);
4257 
4258 	/*
4259 	 * Now we are truly single threaded on this stream, and can
4260 	 * delete the things hanging off the connp, and finally the connp.
4261 	 * We removed this connp from the fanout list, it cannot be
4262 	 * accessed thru the fanouts, and we already waited for the
4263 	 * conn_ref to drop to 0. We are already in close, so
4264 	 * there cannot be any other thread from the top. qprocsoff
4265 	 * has completed, and service has completed or won't run in
4266 	 * future.
4267 	 */
4268 	ASSERT(connp->conn_ref == 1);
4269 
4270 	inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4271 
4272 	connp->conn_ref--;
4273 	ipcl_conn_destroy(connp);
4274 
4275 	q->q_ptr = WR(q)->q_ptr = NULL;
4276 	return (0);
4277 }
4278 
4279 /*
4280  * Wapper around putnext() so that ip_rts_request can merely use
4281  * conn_recv.
4282  */
4283 /*ARGSUSED2*/
4284 static void
ip_conn_input(void * arg1,mblk_t * mp,void * arg2,ip_recv_attr_t * ira)4285 ip_conn_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4286 {
4287 	conn_t *connp = (conn_t *)arg1;
4288 
4289 	putnext(connp->conn_rq, mp);
4290 }
4291 
4292 /* Dummy in case ICMP error delivery is attempted to a /dev/ip instance */
4293 /* ARGSUSED */
4294 static void
ip_conn_input_icmp(void * arg1,mblk_t * mp,void * arg2,ip_recv_attr_t * ira)4295 ip_conn_input_icmp(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4296 {
4297 	freemsg(mp);
4298 }
4299 
4300 /*
4301  * Called when the module is about to be unloaded
4302  */
4303 void
ip_ddi_destroy(void)4304 ip_ddi_destroy(void)
4305 {
4306 	/* This needs to be called before destroying any transports. */
4307 	mutex_enter(&cpu_lock);
4308 	unregister_cpu_setup_func(ip_tp_cpu_update, NULL);
4309 	mutex_exit(&cpu_lock);
4310 
4311 	tnet_fini();
4312 
4313 	icmp_ddi_g_destroy();
4314 	rts_ddi_g_destroy();
4315 	udp_ddi_g_destroy();
4316 	sctp_ddi_g_destroy();
4317 	tcp_ddi_g_destroy();
4318 	ilb_ddi_g_destroy();
4319 	dce_g_destroy();
4320 	ipsec_policy_g_destroy();
4321 	ipcl_g_destroy();
4322 	ip_net_g_destroy();
4323 	ip_ire_g_fini();
4324 	inet_minor_destroy(ip_minor_arena_sa);
4325 #if defined(_LP64)
4326 	inet_minor_destroy(ip_minor_arena_la);
4327 #endif
4328 
4329 #ifdef DEBUG
4330 	list_destroy(&ip_thread_list);
4331 	rw_destroy(&ip_thread_rwlock);
4332 	tsd_destroy(&ip_thread_data);
4333 #endif
4334 
4335 	netstack_unregister(NS_IP);
4336 }
4337 
4338 /*
4339  * First step in cleanup.
4340  */
4341 /* ARGSUSED */
4342 static void
ip_stack_shutdown(netstackid_t stackid,void * arg)4343 ip_stack_shutdown(netstackid_t stackid, void *arg)
4344 {
4345 	ip_stack_t *ipst = (ip_stack_t *)arg;
4346 	kt_did_t ktid;
4347 
4348 #ifdef NS_DEBUG
4349 	printf("ip_stack_shutdown(%p, stack %d)\n", (void *)ipst, stackid);
4350 #endif
4351 
4352 	/*
4353 	 * Perform cleanup for special interfaces (loopback and IPMP).
4354 	 */
4355 	ip_interface_cleanup(ipst);
4356 
4357 	/*
4358 	 * The *_hook_shutdown()s start the process of notifying any
4359 	 * consumers that things are going away.... nothing is destroyed.
4360 	 */
4361 	ipv4_hook_shutdown(ipst);
4362 	ipv6_hook_shutdown(ipst);
4363 	arp_hook_shutdown(ipst);
4364 
4365 	mutex_enter(&ipst->ips_capab_taskq_lock);
4366 	ktid = ipst->ips_capab_taskq_thread->t_did;
4367 	ipst->ips_capab_taskq_quit = B_TRUE;
4368 	cv_signal(&ipst->ips_capab_taskq_cv);
4369 	mutex_exit(&ipst->ips_capab_taskq_lock);
4370 
4371 	/*
4372 	 * In rare occurrences, particularly on virtual hardware where CPUs can
4373 	 * be de-scheduled, the thread that we just signaled will not run until
4374 	 * after we have gotten through parts of ip_stack_fini. If that happens
4375 	 * then we'll try to grab the ips_capab_taskq_lock as part of returning
4376 	 * from cv_wait which no longer exists.
4377 	 */
4378 	thread_join(ktid);
4379 }
4380 
4381 /*
4382  * Free the IP stack instance.
4383  */
4384 static void
ip_stack_fini(netstackid_t stackid,void * arg)4385 ip_stack_fini(netstackid_t stackid, void *arg)
4386 {
4387 	ip_stack_t *ipst = (ip_stack_t *)arg;
4388 	int ret;
4389 
4390 #ifdef NS_DEBUG
4391 	printf("ip_stack_fini(%p, stack %d)\n", (void *)ipst, stackid);
4392 #endif
4393 	/*
4394 	 * At this point, all of the notifications that the events and
4395 	 * protocols are going away have been run, meaning that we can
4396 	 * now set about starting to clean things up.
4397 	 */
4398 	ipobs_fini(ipst);
4399 	ipv4_hook_destroy(ipst);
4400 	ipv6_hook_destroy(ipst);
4401 	arp_hook_destroy(ipst);
4402 	ip_net_destroy(ipst);
4403 
4404 	ipmp_destroy(ipst);
4405 
4406 	ip_kstat_fini(stackid, ipst->ips_ip_mibkp);
4407 	ipst->ips_ip_mibkp = NULL;
4408 	icmp_kstat_fini(stackid, ipst->ips_icmp_mibkp);
4409 	ipst->ips_icmp_mibkp = NULL;
4410 	ip_kstat2_fini(stackid, ipst->ips_ip_kstat);
4411 	ipst->ips_ip_kstat = NULL;
4412 	bzero(&ipst->ips_ip_statistics, sizeof (ipst->ips_ip_statistics));
4413 	ip6_kstat_fini(stackid, ipst->ips_ip6_kstat);
4414 	ipst->ips_ip6_kstat = NULL;
4415 	bzero(&ipst->ips_ip6_statistics, sizeof (ipst->ips_ip6_statistics));
4416 
4417 	kmem_free(ipst->ips_propinfo_tbl,
4418 	    ip_propinfo_count * sizeof (mod_prop_info_t));
4419 	ipst->ips_propinfo_tbl = NULL;
4420 
4421 	dce_stack_destroy(ipst);
4422 	ip_mrouter_stack_destroy(ipst);
4423 
4424 	/*
4425 	 * Quiesce all of our timers. Note we set the quiesce flags before we
4426 	 * call untimeout. The slowtimers may actually kick off another instance
4427 	 * of the non-slow timers.
4428 	 */
4429 	mutex_enter(&ipst->ips_igmp_timer_lock);
4430 	ipst->ips_igmp_timer_quiesce = B_TRUE;
4431 	mutex_exit(&ipst->ips_igmp_timer_lock);
4432 
4433 	mutex_enter(&ipst->ips_mld_timer_lock);
4434 	ipst->ips_mld_timer_quiesce = B_TRUE;
4435 	mutex_exit(&ipst->ips_mld_timer_lock);
4436 
4437 	mutex_enter(&ipst->ips_igmp_slowtimeout_lock);
4438 	ipst->ips_igmp_slowtimeout_quiesce = B_TRUE;
4439 	mutex_exit(&ipst->ips_igmp_slowtimeout_lock);
4440 
4441 	mutex_enter(&ipst->ips_mld_slowtimeout_lock);
4442 	ipst->ips_mld_slowtimeout_quiesce = B_TRUE;
4443 	mutex_exit(&ipst->ips_mld_slowtimeout_lock);
4444 
4445 	ret = untimeout(ipst->ips_igmp_timeout_id);
4446 	if (ret == -1) {
4447 		ASSERT(ipst->ips_igmp_timeout_id == 0);
4448 	} else {
4449 		ASSERT(ipst->ips_igmp_timeout_id != 0);
4450 		ipst->ips_igmp_timeout_id = 0;
4451 	}
4452 	ret = untimeout(ipst->ips_igmp_slowtimeout_id);
4453 	if (ret == -1) {
4454 		ASSERT(ipst->ips_igmp_slowtimeout_id == 0);
4455 	} else {
4456 		ASSERT(ipst->ips_igmp_slowtimeout_id != 0);
4457 		ipst->ips_igmp_slowtimeout_id = 0;
4458 	}
4459 	ret = untimeout(ipst->ips_mld_timeout_id);
4460 	if (ret == -1) {
4461 		ASSERT(ipst->ips_mld_timeout_id == 0);
4462 	} else {
4463 		ASSERT(ipst->ips_mld_timeout_id != 0);
4464 		ipst->ips_mld_timeout_id = 0;
4465 	}
4466 	ret = untimeout(ipst->ips_mld_slowtimeout_id);
4467 	if (ret == -1) {
4468 		ASSERT(ipst->ips_mld_slowtimeout_id == 0);
4469 	} else {
4470 		ASSERT(ipst->ips_mld_slowtimeout_id != 0);
4471 		ipst->ips_mld_slowtimeout_id = 0;
4472 	}
4473 
4474 	ip_ire_fini(ipst);
4475 	ip6_asp_free(ipst);
4476 	conn_drain_fini(ipst);
4477 	ipcl_destroy(ipst);
4478 
4479 	mutex_destroy(&ipst->ips_ndp4->ndp_g_lock);
4480 	mutex_destroy(&ipst->ips_ndp6->ndp_g_lock);
4481 	kmem_free(ipst->ips_ndp4, sizeof (ndp_g_t));
4482 	ipst->ips_ndp4 = NULL;
4483 	kmem_free(ipst->ips_ndp6, sizeof (ndp_g_t));
4484 	ipst->ips_ndp6 = NULL;
4485 
4486 	if (ipst->ips_loopback_ksp != NULL) {
4487 		kstat_delete_netstack(ipst->ips_loopback_ksp, stackid);
4488 		ipst->ips_loopback_ksp = NULL;
4489 	}
4490 
4491 	mutex_destroy(&ipst->ips_capab_taskq_lock);
4492 	cv_destroy(&ipst->ips_capab_taskq_cv);
4493 
4494 	rw_destroy(&ipst->ips_srcid_lock);
4495 
4496 	mutex_destroy(&ipst->ips_ip_mi_lock);
4497 	rw_destroy(&ipst->ips_ill_g_usesrc_lock);
4498 
4499 	mutex_destroy(&ipst->ips_igmp_timer_lock);
4500 	mutex_destroy(&ipst->ips_mld_timer_lock);
4501 	mutex_destroy(&ipst->ips_igmp_slowtimeout_lock);
4502 	mutex_destroy(&ipst->ips_mld_slowtimeout_lock);
4503 	mutex_destroy(&ipst->ips_ip_addr_avail_lock);
4504 	rw_destroy(&ipst->ips_ill_g_lock);
4505 
4506 	kmem_free(ipst->ips_phyint_g_list, sizeof (phyint_list_t));
4507 	ipst->ips_phyint_g_list = NULL;
4508 	kmem_free(ipst->ips_ill_g_heads, sizeof (ill_g_head_t) * MAX_G_HEADS);
4509 	ipst->ips_ill_g_heads = NULL;
4510 
4511 	ldi_ident_release(ipst->ips_ldi_ident);
4512 	kmem_free(ipst, sizeof (*ipst));
4513 }
4514 
4515 /*
4516  * This function is called from the TSD destructor, and is used to debug
4517  * reference count issues in IP. See block comment in <inet/ip_if.h> for
4518  * details.
4519  */
4520 static void
ip_thread_exit(void * phash)4521 ip_thread_exit(void *phash)
4522 {
4523 	th_hash_t *thh = phash;
4524 
4525 	rw_enter(&ip_thread_rwlock, RW_WRITER);
4526 	list_remove(&ip_thread_list, thh);
4527 	rw_exit(&ip_thread_rwlock);
4528 	mod_hash_destroy_hash(thh->thh_hash);
4529 	kmem_free(thh, sizeof (*thh));
4530 }
4531 
4532 /*
4533  * Called when the IP kernel module is loaded into the kernel
4534  */
4535 void
ip_ddi_init(void)4536 ip_ddi_init(void)
4537 {
4538 	ip_squeue_flag = ip_squeue_switch(ip_squeue_enter);
4539 
4540 	/*
4541 	 * For IP and TCP the minor numbers should start from 2 since we have 4
4542 	 * initial devices: ip, ip6, tcp, tcp6.
4543 	 */
4544 	/*
4545 	 * If this is a 64-bit kernel, then create two separate arenas -
4546 	 * one for TLIs in the range of INET_MIN_DEV+2 through 2^^18-1, and the
4547 	 * other for socket apps in the range 2^^18 through 2^^32-1.
4548 	 */
4549 	ip_minor_arena_la = NULL;
4550 	ip_minor_arena_sa = NULL;
4551 #if defined(_LP64)
4552 	if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4553 	    INET_MIN_DEV + 2, MAXMIN32, KM_SLEEP)) == NULL) {
4554 		cmn_err(CE_PANIC,
4555 		    "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4556 	}
4557 	if ((ip_minor_arena_la = inet_minor_create("ip_minor_arena_la",
4558 	    MAXMIN32 + 1, MAXMIN64, KM_SLEEP)) == NULL) {
4559 		cmn_err(CE_PANIC,
4560 		    "ip_ddi_init: ip_minor_arena_la creation failed\n");
4561 	}
4562 #else
4563 	if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4564 	    INET_MIN_DEV + 2, MAXMIN, KM_SLEEP)) == NULL) {
4565 		cmn_err(CE_PANIC,
4566 		    "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4567 	}
4568 #endif
4569 	ip_poll_normal_ticks = MSEC_TO_TICK_ROUNDUP(ip_poll_normal_ms);
4570 
4571 	ipcl_g_init();
4572 	ip_ire_g_init();
4573 	ip_net_g_init();
4574 
4575 #ifdef DEBUG
4576 	tsd_create(&ip_thread_data, ip_thread_exit);
4577 	rw_init(&ip_thread_rwlock, NULL, RW_DEFAULT, NULL);
4578 	list_create(&ip_thread_list, sizeof (th_hash_t),
4579 	    offsetof(th_hash_t, thh_link));
4580 #endif
4581 	ipsec_policy_g_init();
4582 	tcp_ddi_g_init();
4583 	sctp_ddi_g_init();
4584 	dce_g_init();
4585 
4586 	/*
4587 	 * We want to be informed each time a stack is created or
4588 	 * destroyed in the kernel, so we can maintain the
4589 	 * set of udp_stack_t's.
4590 	 */
4591 	netstack_register(NS_IP, ip_stack_init, ip_stack_shutdown,
4592 	    ip_stack_fini);
4593 
4594 	tnet_init();
4595 
4596 	udp_ddi_g_init();
4597 	rts_ddi_g_init();
4598 	icmp_ddi_g_init();
4599 	ilb_ddi_g_init();
4600 
4601 	/* This needs to be called after all transports are initialized. */
4602 	mutex_enter(&cpu_lock);
4603 	register_cpu_setup_func(ip_tp_cpu_update, NULL);
4604 	mutex_exit(&cpu_lock);
4605 }
4606 
4607 /*
4608  * Initialize the IP stack instance.
4609  */
4610 static void *
ip_stack_init(netstackid_t stackid,netstack_t * ns)4611 ip_stack_init(netstackid_t stackid, netstack_t *ns)
4612 {
4613 	ip_stack_t	*ipst;
4614 	size_t		arrsz;
4615 	major_t		major;
4616 
4617 #ifdef NS_DEBUG
4618 	printf("ip_stack_init(stack %d)\n", stackid);
4619 #endif
4620 
4621 	ipst = (ip_stack_t *)kmem_zalloc(sizeof (*ipst), KM_SLEEP);
4622 	ipst->ips_netstack = ns;
4623 
4624 	ipst->ips_ill_g_heads = kmem_zalloc(sizeof (ill_g_head_t) * MAX_G_HEADS,
4625 	    KM_SLEEP);
4626 	ipst->ips_phyint_g_list = kmem_zalloc(sizeof (phyint_list_t),
4627 	    KM_SLEEP);
4628 	ipst->ips_ndp4 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4629 	ipst->ips_ndp6 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4630 	mutex_init(&ipst->ips_ndp4->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4631 	mutex_init(&ipst->ips_ndp6->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4632 
4633 	mutex_init(&ipst->ips_igmp_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4634 	ipst->ips_igmp_deferred_next = INFINITY;
4635 	mutex_init(&ipst->ips_mld_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4636 	ipst->ips_mld_deferred_next = INFINITY;
4637 	mutex_init(&ipst->ips_igmp_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4638 	mutex_init(&ipst->ips_mld_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4639 	mutex_init(&ipst->ips_ip_mi_lock, NULL, MUTEX_DEFAULT, NULL);
4640 	mutex_init(&ipst->ips_ip_addr_avail_lock, NULL, MUTEX_DEFAULT, NULL);
4641 	rw_init(&ipst->ips_ill_g_lock, NULL, RW_DEFAULT, NULL);
4642 	rw_init(&ipst->ips_ill_g_usesrc_lock, NULL, RW_DEFAULT, NULL);
4643 
4644 	ipcl_init(ipst);
4645 	ip_ire_init(ipst);
4646 	ip6_asp_init(ipst);
4647 	ipif_init(ipst);
4648 	conn_drain_init(ipst);
4649 	ip_mrouter_stack_init(ipst);
4650 	dce_stack_init(ipst);
4651 
4652 	ipst->ips_ip_multirt_log_interval = 1000;
4653 
4654 	ipst->ips_ill_index = 1;
4655 
4656 	ipst->ips_saved_ip_forwarding = -1;
4657 	ipst->ips_reg_vif_num = ALL_VIFS; 	/* Index to Register vif */
4658 
4659 	arrsz = ip_propinfo_count * sizeof (mod_prop_info_t);
4660 	ipst->ips_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz, KM_SLEEP);
4661 	bcopy(ip_propinfo_tbl, ipst->ips_propinfo_tbl, arrsz);
4662 
4663 	ipst->ips_ip_mibkp = ip_kstat_init(stackid, ipst);
4664 	ipst->ips_icmp_mibkp = icmp_kstat_init(stackid);
4665 	ipst->ips_ip_kstat = ip_kstat2_init(stackid, &ipst->ips_ip_statistics);
4666 	ipst->ips_ip6_kstat =
4667 	    ip6_kstat_init(stackid, &ipst->ips_ip6_statistics);
4668 
4669 	ipst->ips_ip_src_id = 1;
4670 	rw_init(&ipst->ips_srcid_lock, NULL, RW_DEFAULT, NULL);
4671 
4672 	ipst->ips_src_generation = SRC_GENERATION_INITIAL;
4673 
4674 	ip_net_init(ipst, ns);
4675 	ipv4_hook_init(ipst);
4676 	ipv6_hook_init(ipst);
4677 	arp_hook_init(ipst);
4678 	ipmp_init(ipst);
4679 	ipobs_init(ipst);
4680 
4681 	/*
4682 	 * Create the taskq dispatcher thread and initialize related stuff.
4683 	 */
4684 	mutex_init(&ipst->ips_capab_taskq_lock, NULL, MUTEX_DEFAULT, NULL);
4685 	cv_init(&ipst->ips_capab_taskq_cv, NULL, CV_DEFAULT, NULL);
4686 	ipst->ips_capab_taskq_thread = thread_create(NULL, 0,
4687 	    ill_taskq_dispatch, ipst, 0, &p0, TS_RUN, minclsyspri);
4688 
4689 	major = mod_name_to_major(INET_NAME);
4690 	(void) ldi_ident_from_major(major, &ipst->ips_ldi_ident);
4691 	return (ipst);
4692 }
4693 
4694 /*
4695  * Allocate and initialize a DLPI template of the specified length.  (May be
4696  * called as writer.)
4697  */
4698 mblk_t *
ip_dlpi_alloc(size_t len,t_uscalar_t prim)4699 ip_dlpi_alloc(size_t len, t_uscalar_t prim)
4700 {
4701 	mblk_t	*mp;
4702 
4703 	mp = allocb(len, BPRI_MED);
4704 	if (!mp)
4705 		return (NULL);
4706 
4707 	/*
4708 	 * DLPIv2 says that DL_INFO_REQ and DL_TOKEN_REQ (the latter
4709 	 * of which we don't seem to use) are sent with M_PCPROTO, and
4710 	 * that other DLPI are M_PROTO.
4711 	 */
4712 	if (prim == DL_INFO_REQ) {
4713 		mp->b_datap->db_type = M_PCPROTO;
4714 	} else {
4715 		mp->b_datap->db_type = M_PROTO;
4716 	}
4717 
4718 	mp->b_wptr = mp->b_rptr + len;
4719 	bzero(mp->b_rptr, len);
4720 	((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
4721 	return (mp);
4722 }
4723 
4724 /*
4725  * Allocate and initialize a DLPI notification.  (May be called as writer.)
4726  */
4727 mblk_t *
ip_dlnotify_alloc(uint_t notification,uint_t data)4728 ip_dlnotify_alloc(uint_t notification, uint_t data)
4729 {
4730 	dl_notify_ind_t	*notifyp;
4731 	mblk_t		*mp;
4732 
4733 	if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4734 		return (NULL);
4735 
4736 	notifyp = (dl_notify_ind_t *)mp->b_rptr;
4737 	notifyp->dl_notification = notification;
4738 	notifyp->dl_data = data;
4739 	return (mp);
4740 }
4741 
4742 mblk_t *
ip_dlnotify_alloc2(uint_t notification,uint_t data1,uint_t data2)4743 ip_dlnotify_alloc2(uint_t notification, uint_t data1, uint_t data2)
4744 {
4745 	dl_notify_ind_t	*notifyp;
4746 	mblk_t		*mp;
4747 
4748 	if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4749 		return (NULL);
4750 
4751 	notifyp = (dl_notify_ind_t *)mp->b_rptr;
4752 	notifyp->dl_notification = notification;
4753 	notifyp->dl_data1 = data1;
4754 	notifyp->dl_data2 = data2;
4755 	return (mp);
4756 }
4757 
4758 /*
4759  * Debug formatting routine.  Returns a character string representation of the
4760  * addr in buf, of the form xxx.xxx.xxx.xxx.  This routine takes the address
4761  * in the form of a ipaddr_t and calls ip_dot_saddr with a pointer.
4762  *
4763  * Once the ndd table-printing interfaces are removed, this can be changed to
4764  * standard dotted-decimal form.
4765  */
4766 char *
ip_dot_addr(ipaddr_t addr,char * buf)4767 ip_dot_addr(ipaddr_t addr, char *buf)
4768 {
4769 	uint8_t *ap = (uint8_t *)&addr;
4770 
4771 	(void) mi_sprintf(buf, "%03d.%03d.%03d.%03d",
4772 	    ap[0] & 0xFF, ap[1] & 0xFF, ap[2] & 0xFF, ap[3] & 0xFF);
4773 	return (buf);
4774 }
4775 
4776 /*
4777  * Write the given MAC address as a printable string in the usual colon-
4778  * separated format.
4779  */
4780 const char *
mac_colon_addr(const uint8_t * addr,size_t alen,char * buf,size_t buflen)4781 mac_colon_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
4782 {
4783 	char *bp;
4784 
4785 	if (alen == 0 || buflen < 4)
4786 		return ("?");
4787 	bp = buf;
4788 	for (;;) {
4789 		/*
4790 		 * If there are more MAC address bytes available, but we won't
4791 		 * have any room to print them, then add "..." to the string
4792 		 * instead.  See below for the 'magic number' explanation.
4793 		 */
4794 		if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
4795 			(void) strcpy(bp, "...");
4796 			break;
4797 		}
4798 		(void) sprintf(bp, "%02x", *addr++);
4799 		bp += 2;
4800 		if (--alen == 0)
4801 			break;
4802 		*bp++ = ':';
4803 		buflen -= 3;
4804 		/*
4805 		 * At this point, based on the first 'if' statement above,
4806 		 * either alen == 1 and buflen >= 3, or alen > 1 and
4807 		 * buflen >= 4.  The first case leaves room for the final "xx"
4808 		 * number and trailing NUL byte.  The second leaves room for at
4809 		 * least "...".  Thus the apparently 'magic' numbers chosen for
4810 		 * that statement.
4811 		 */
4812 	}
4813 	return (buf);
4814 }
4815 
4816 /*
4817  * Called when it is conceptually a ULP that would sent the packet
4818  * e.g., port unreachable and protocol unreachable. Check that the packet
4819  * would have passed the IPsec global policy before sending the error.
4820  *
4821  * Send an ICMP error after patching up the packet appropriately.
4822  * Uses ip_drop_input and bumps the appropriate MIB.
4823  */
4824 void
ip_fanout_send_icmp_v4(mblk_t * mp,uint_t icmp_type,uint_t icmp_code,ip_recv_attr_t * ira)4825 ip_fanout_send_icmp_v4(mblk_t *mp, uint_t icmp_type, uint_t icmp_code,
4826     ip_recv_attr_t *ira)
4827 {
4828 	ipha_t		*ipha;
4829 	boolean_t	secure;
4830 	ill_t		*ill = ira->ira_ill;
4831 	ip_stack_t	*ipst = ill->ill_ipst;
4832 	netstack_t	*ns = ipst->ips_netstack;
4833 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
4834 
4835 	secure = ira->ira_flags & IRAF_IPSEC_SECURE;
4836 
4837 	/*
4838 	 * We are generating an icmp error for some inbound packet.
4839 	 * Called from all ip_fanout_(udp, tcp, proto) functions.
4840 	 * Before we generate an error, check with global policy
4841 	 * to see whether this is allowed to enter the system. As
4842 	 * there is no "conn", we are checking with global policy.
4843 	 */
4844 	ipha = (ipha_t *)mp->b_rptr;
4845 	if (secure || ipss->ipsec_inbound_v4_policy_present) {
4846 		mp = ipsec_check_global_policy(mp, NULL, ipha, NULL, ira, ns);
4847 		if (mp == NULL)
4848 			return;
4849 	}
4850 
4851 	/* We never send errors for protocols that we do implement */
4852 	if (ira->ira_protocol == IPPROTO_ICMP ||
4853 	    ira->ira_protocol == IPPROTO_IGMP) {
4854 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4855 		ip_drop_input("ip_fanout_send_icmp_v4", mp, ill);
4856 		freemsg(mp);
4857 		return;
4858 	}
4859 	/*
4860 	 * Have to correct checksum since
4861 	 * the packet might have been
4862 	 * fragmented and the reassembly code in ip_rput
4863 	 * does not restore the IP checksum.
4864 	 */
4865 	ipha->ipha_hdr_checksum = 0;
4866 	ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
4867 
4868 	switch (icmp_type) {
4869 	case ICMP_DEST_UNREACHABLE:
4870 		switch (icmp_code) {
4871 		case ICMP_PROTOCOL_UNREACHABLE:
4872 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInUnknownProtos);
4873 			ip_drop_input("ipIfStatsInUnknownProtos", mp, ill);
4874 			break;
4875 		case ICMP_PORT_UNREACHABLE:
4876 			BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
4877 			ip_drop_input("ipIfStatsNoPorts", mp, ill);
4878 			break;
4879 		}
4880 
4881 		icmp_unreachable(mp, icmp_code, ira);
4882 		break;
4883 	default:
4884 #ifdef DEBUG
4885 		panic("ip_fanout_send_icmp_v4: wrong type");
4886 		/*NOTREACHED*/
4887 #else
4888 		freemsg(mp);
4889 		break;
4890 #endif
4891 	}
4892 }
4893 
4894 /*
4895  * Used to send an ICMP error message when a packet is received for
4896  * a protocol that is not supported. The mblk passed as argument
4897  * is consumed by this function.
4898  */
4899 void
ip_proto_not_sup(mblk_t * mp,ip_recv_attr_t * ira)4900 ip_proto_not_sup(mblk_t *mp, ip_recv_attr_t *ira)
4901 {
4902 	ipha_t		*ipha;
4903 
4904 	ipha = (ipha_t *)mp->b_rptr;
4905 	if (ira->ira_flags & IRAF_IS_IPV4) {
4906 		ASSERT(IPH_HDR_VERSION(ipha) == IP_VERSION);
4907 		ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
4908 		    ICMP_PROTOCOL_UNREACHABLE, ira);
4909 	} else {
4910 		ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
4911 		ip_fanout_send_icmp_v6(mp, ICMP6_PARAM_PROB,
4912 		    ICMP6_PARAMPROB_NEXTHEADER, ira);
4913 	}
4914 }
4915 
4916 /*
4917  * Deliver a rawip packet to the given conn, possibly applying ipsec policy.
4918  * Handles IPv4 and IPv6.
4919  * We are responsible for disposing of mp, such as by freemsg() or putnext()
4920  * Caller is responsible for dropping references to the conn.
4921  */
4922 void
ip_fanout_proto_conn(conn_t * connp,mblk_t * mp,ipha_t * ipha,ip6_t * ip6h,ip_recv_attr_t * ira)4923 ip_fanout_proto_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4924     ip_recv_attr_t *ira)
4925 {
4926 	ill_t		*ill = ira->ira_ill;
4927 	ip_stack_t	*ipst = ill->ill_ipst;
4928 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
4929 	boolean_t	secure;
4930 	uint_t		protocol = ira->ira_protocol;
4931 	iaflags_t	iraflags = ira->ira_flags;
4932 	queue_t		*rq;
4933 
4934 	secure = iraflags & IRAF_IPSEC_SECURE;
4935 
4936 	rq = connp->conn_rq;
4937 	if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
4938 		switch (protocol) {
4939 		case IPPROTO_ICMPV6:
4940 			BUMP_MIB(ill->ill_icmp6_mib, ipv6IfIcmpInOverflows);
4941 			break;
4942 		case IPPROTO_ICMP:
4943 			BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
4944 			break;
4945 		default:
4946 			BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
4947 			break;
4948 		}
4949 		freemsg(mp);
4950 		return;
4951 	}
4952 
4953 	ASSERT(!(IPCL_IS_IPTUN(connp)));
4954 
4955 	if (((iraflags & IRAF_IS_IPV4) ?
4956 	    CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
4957 	    CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
4958 	    secure) {
4959 		mp = ipsec_check_inbound_policy(mp, connp, ipha,
4960 		    ip6h, ira);
4961 		if (mp == NULL) {
4962 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4963 			/* Note that mp is NULL */
4964 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
4965 			return;
4966 		}
4967 	}
4968 
4969 	if (iraflags & IRAF_ICMP_ERROR) {
4970 		(connp->conn_recvicmp)(connp, mp, NULL, ira);
4971 	} else {
4972 		ill_t *rill = ira->ira_rill;
4973 
4974 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
4975 		ira->ira_ill = ira->ira_rill = NULL;
4976 		/* Send it upstream */
4977 		(connp->conn_recv)(connp, mp, NULL, ira);
4978 		ira->ira_ill = ill;
4979 		ira->ira_rill = rill;
4980 	}
4981 }
4982 
4983 /*
4984  * Handle protocols with which IP is less intimate.  There
4985  * can be more than one stream bound to a particular
4986  * protocol.  When this is the case, normally each one gets a copy
4987  * of any incoming packets.
4988  *
4989  * IPsec NOTE :
4990  *
4991  * Don't allow a secure packet going up a non-secure connection.
4992  * We don't allow this because
4993  *
4994  * 1) Reply might go out in clear which will be dropped at
4995  *    the sending side.
4996  * 2) If the reply goes out in clear it will give the
4997  *    adversary enough information for getting the key in
4998  *    most of the cases.
4999  *
5000  * Moreover getting a secure packet when we expect clear
5001  * implies that SA's were added without checking for
5002  * policy on both ends. This should not happen once ISAKMP
5003  * is used to negotiate SAs as SAs will be added only after
5004  * verifying the policy.
5005  *
5006  * Zones notes:
5007  * Earlier in ip_input on a system with multiple shared-IP zones we
5008  * duplicate the multicast and broadcast packets and send them up
5009  * with each explicit zoneid that exists on that ill.
5010  * This means that here we can match the zoneid with SO_ALLZONES being special.
5011  */
5012 void
ip_fanout_proto_v4(mblk_t * mp,ipha_t * ipha,ip_recv_attr_t * ira)5013 ip_fanout_proto_v4(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
5014 {
5015 	mblk_t		*mp1;
5016 	ipaddr_t	laddr;
5017 	conn_t		*connp, *first_connp, *next_connp;
5018 	connf_t		*connfp;
5019 	ill_t		*ill = ira->ira_ill;
5020 	ip_stack_t	*ipst = ill->ill_ipst;
5021 
5022 	laddr = ipha->ipha_dst;
5023 
5024 	connfp = &ipst->ips_ipcl_proto_fanout_v4[ira->ira_protocol];
5025 	mutex_enter(&connfp->connf_lock);
5026 	connp = connfp->connf_head;
5027 	for (connp = connfp->connf_head; connp != NULL;
5028 	    connp = connp->conn_next) {
5029 		/* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5030 		if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5031 		    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5032 		    tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp))) {
5033 			break;
5034 		}
5035 	}
5036 
5037 	if (connp == NULL) {
5038 		/*
5039 		 * No one bound to these addresses.  Is
5040 		 * there a client that wants all
5041 		 * unclaimed datagrams?
5042 		 */
5043 		mutex_exit(&connfp->connf_lock);
5044 		ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
5045 		    ICMP_PROTOCOL_UNREACHABLE, ira);
5046 		return;
5047 	}
5048 
5049 	ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5050 
5051 	CONN_INC_REF(connp);
5052 	first_connp = connp;
5053 	connp = connp->conn_next;
5054 
5055 	for (;;) {
5056 		while (connp != NULL) {
5057 			/* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5058 			if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5059 			    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5060 			    tsol_receive_local(mp, &laddr, IPV4_VERSION,
5061 			    ira, connp)))
5062 				break;
5063 			connp = connp->conn_next;
5064 		}
5065 
5066 		if (connp == NULL) {
5067 			/* No more interested clients */
5068 			connp = first_connp;
5069 			break;
5070 		}
5071 		if (((mp1 = dupmsg(mp)) == NULL) &&
5072 		    ((mp1 = copymsg(mp)) == NULL)) {
5073 			/* Memory allocation failed */
5074 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5075 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
5076 			connp = first_connp;
5077 			break;
5078 		}
5079 
5080 		CONN_INC_REF(connp);
5081 		mutex_exit(&connfp->connf_lock);
5082 
5083 		ip_fanout_proto_conn(connp, mp1, (ipha_t *)mp1->b_rptr, NULL,
5084 		    ira);
5085 
5086 		mutex_enter(&connfp->connf_lock);
5087 		/* Follow the next pointer before releasing the conn. */
5088 		next_connp = connp->conn_next;
5089 		CONN_DEC_REF(connp);
5090 		connp = next_connp;
5091 	}
5092 
5093 	/* Last one.  Send it upstream. */
5094 	mutex_exit(&connfp->connf_lock);
5095 
5096 	ip_fanout_proto_conn(connp, mp, ipha, NULL, ira);
5097 
5098 	CONN_DEC_REF(connp);
5099 }
5100 
5101 /*
5102  * If we have a IPsec NAT-Traversal packet, strip the zero-SPI or
5103  * pass it along to ESP if the SPI is non-zero.  Returns the mblk if the mblk
5104  * is not consumed.
5105  *
5106  * One of three things can happen, all of which affect the passed-in mblk:
5107  *
5108  * 1.) The packet is stock UDP and gets its zero-SPI stripped.  Return mblk..
5109  *
5110  * 2.) The packet is ESP-in-UDP, gets transformed into an equivalent
5111  *     ESP packet, and is passed along to ESP for consumption.  Return NULL.
5112  *
5113  * 3.) The packet is an ESP-in-UDP Keepalive.  Drop it and return NULL.
5114  */
5115 mblk_t *
zero_spi_check(mblk_t * mp,ip_recv_attr_t * ira)5116 zero_spi_check(mblk_t *mp, ip_recv_attr_t *ira)
5117 {
5118 	int shift, plen, iph_len;
5119 	ipha_t *ipha;
5120 	udpha_t *udpha;
5121 	uint32_t *spi;
5122 	uint32_t esp_ports;
5123 	uint8_t *orptr;
5124 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
5125 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
5126 
5127 	ipha = (ipha_t *)mp->b_rptr;
5128 	iph_len = ira->ira_ip_hdr_length;
5129 	plen = ira->ira_pktlen;
5130 
5131 	if (plen - iph_len - sizeof (udpha_t) < sizeof (uint32_t)) {
5132 		/*
5133 		 * Most likely a keepalive for the benefit of an intervening
5134 		 * NAT.  These aren't for us, per se, so drop it.
5135 		 *
5136 		 * RFC 3947/8 doesn't say for sure what to do for 2-3
5137 		 * byte packets (keepalives are 1-byte), but we'll drop them
5138 		 * also.
5139 		 */
5140 		ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5141 		    DROPPER(ipss, ipds_esp_nat_t_ka), &ipss->ipsec_dropper);
5142 		return (NULL);
5143 	}
5144 
5145 	if (MBLKL(mp) < iph_len + sizeof (udpha_t) + sizeof (*spi)) {
5146 		/* might as well pull it all up - it might be ESP. */
5147 		if (!pullupmsg(mp, -1)) {
5148 			ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5149 			    DROPPER(ipss, ipds_esp_nomem),
5150 			    &ipss->ipsec_dropper);
5151 			return (NULL);
5152 		}
5153 
5154 		ipha = (ipha_t *)mp->b_rptr;
5155 	}
5156 	spi = (uint32_t *)(mp->b_rptr + iph_len + sizeof (udpha_t));
5157 	if (*spi == 0) {
5158 		/* UDP packet - remove 0-spi. */
5159 		shift = sizeof (uint32_t);
5160 	} else {
5161 		/* ESP-in-UDP packet - reduce to ESP. */
5162 		ipha->ipha_protocol = IPPROTO_ESP;
5163 		shift = sizeof (udpha_t);
5164 	}
5165 
5166 	/* Fix IP header */
5167 	ira->ira_pktlen = (plen - shift);
5168 	ipha->ipha_length = htons(ira->ira_pktlen);
5169 	ipha->ipha_hdr_checksum = 0;
5170 
5171 	orptr = mp->b_rptr;
5172 	mp->b_rptr += shift;
5173 
5174 	udpha = (udpha_t *)(orptr + iph_len);
5175 	if (*spi == 0) {
5176 		ASSERT((uint8_t *)ipha == orptr);
5177 		udpha->uha_length = htons(plen - shift - iph_len);
5178 		iph_len += sizeof (udpha_t);	/* For the call to ovbcopy(). */
5179 		esp_ports = 0;
5180 	} else {
5181 		esp_ports = *((uint32_t *)udpha);
5182 		ASSERT(esp_ports != 0);
5183 	}
5184 	ovbcopy(orptr, orptr + shift, iph_len);
5185 	if (esp_ports != 0) /* Punt up for ESP processing. */ {
5186 		ipha = (ipha_t *)(orptr + shift);
5187 
5188 		ira->ira_flags |= IRAF_ESP_UDP_PORTS;
5189 		ira->ira_esp_udp_ports = esp_ports;
5190 		ip_fanout_v4(mp, ipha, ira);
5191 		return (NULL);
5192 	}
5193 	return (mp);
5194 }
5195 
5196 /*
5197  * Deliver a udp packet to the given conn, possibly applying ipsec policy.
5198  * Handles IPv4 and IPv6.
5199  * We are responsible for disposing of mp, such as by freemsg() or putnext()
5200  * Caller is responsible for dropping references to the conn.
5201  */
5202 void
ip_fanout_udp_conn(conn_t * connp,mblk_t * mp,ipha_t * ipha,ip6_t * ip6h,ip_recv_attr_t * ira)5203 ip_fanout_udp_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
5204     ip_recv_attr_t *ira)
5205 {
5206 	ill_t		*ill = ira->ira_ill;
5207 	ip_stack_t	*ipst = ill->ill_ipst;
5208 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
5209 	boolean_t	secure;
5210 	iaflags_t	iraflags = ira->ira_flags;
5211 
5212 	secure = iraflags & IRAF_IPSEC_SECURE;
5213 
5214 	if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld :
5215 	    !canputnext(connp->conn_rq)) {
5216 		BUMP_MIB(ill->ill_ip_mib, udpIfStatsInOverflows);
5217 		freemsg(mp);
5218 		return;
5219 	}
5220 
5221 	if (((iraflags & IRAF_IS_IPV4) ?
5222 	    CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
5223 	    CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
5224 	    secure) {
5225 		mp = ipsec_check_inbound_policy(mp, connp, ipha,
5226 		    ip6h, ira);
5227 		if (mp == NULL) {
5228 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5229 			/* Note that mp is NULL */
5230 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
5231 			return;
5232 		}
5233 	}
5234 
5235 	/*
5236 	 * Since this code is not used for UDP unicast we don't need a NAT_T
5237 	 * check. Only ip_fanout_v4 has that check.
5238 	 */
5239 	if (ira->ira_flags & IRAF_ICMP_ERROR) {
5240 		(connp->conn_recvicmp)(connp, mp, NULL, ira);
5241 	} else {
5242 		ill_t *rill = ira->ira_rill;
5243 
5244 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5245 		ira->ira_ill = ira->ira_rill = NULL;
5246 		/* Send it upstream */
5247 		(connp->conn_recv)(connp, mp, NULL, ira);
5248 		ira->ira_ill = ill;
5249 		ira->ira_rill = rill;
5250 	}
5251 }
5252 
5253 /*
5254  * Fanout for UDP packets that are multicast or broadcast, and ICMP errors.
5255  * (Unicast fanout is handled in ip_input_v4.)
5256  *
5257  * If SO_REUSEADDR is set all multicast and broadcast packets
5258  * will be delivered to all conns bound to the same port.
5259  *
5260  * If there is at least one matching AF_INET receiver, then we will
5261  * ignore any AF_INET6 receivers.
5262  * In the special case where an AF_INET socket binds to 0.0.0.0/<port> and an
5263  * AF_INET6 socket binds to ::/<port>, only the AF_INET socket receives the IPv4
5264  * packets.
5265  *
5266  * Zones notes:
5267  * Earlier in ip_input on a system with multiple shared-IP zones we
5268  * duplicate the multicast and broadcast packets and send them up
5269  * with each explicit zoneid that exists on that ill.
5270  * This means that here we can match the zoneid with SO_ALLZONES being special.
5271  */
5272 void
ip_fanout_udp_multi_v4(mblk_t * mp,ipha_t * ipha,uint16_t lport,uint16_t fport,ip_recv_attr_t * ira)5273 ip_fanout_udp_multi_v4(mblk_t *mp, ipha_t *ipha, uint16_t lport, uint16_t fport,
5274     ip_recv_attr_t *ira)
5275 {
5276 	ipaddr_t	laddr;
5277 	in6_addr_t	v6faddr;
5278 	conn_t		*connp;
5279 	connf_t		*connfp;
5280 	ipaddr_t	faddr;
5281 	ill_t		*ill = ira->ira_ill;
5282 	ip_stack_t	*ipst = ill->ill_ipst;
5283 
5284 	ASSERT(ira->ira_flags & (IRAF_MULTIBROADCAST|IRAF_ICMP_ERROR));
5285 
5286 	laddr = ipha->ipha_dst;
5287 	faddr = ipha->ipha_src;
5288 
5289 	connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5290 	mutex_enter(&connfp->connf_lock);
5291 	connp = connfp->connf_head;
5292 
5293 	/*
5294 	 * If SO_REUSEADDR has been set on the first we send the
5295 	 * packet to all clients that have joined the group and
5296 	 * match the port.
5297 	 */
5298 	while (connp != NULL) {
5299 		if ((IPCL_UDP_MATCH(connp, lport, laddr, fport, faddr)) &&
5300 		    conn_wantpacket(connp, ira, ipha) &&
5301 		    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5302 		    tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5303 			break;
5304 		connp = connp->conn_next;
5305 	}
5306 
5307 	if (connp == NULL)
5308 		goto notfound;
5309 
5310 	CONN_INC_REF(connp);
5311 
5312 	if (connp->conn_reuseaddr) {
5313 		conn_t		*first_connp = connp;
5314 		conn_t		*next_connp;
5315 		mblk_t		*mp1;
5316 
5317 		connp = connp->conn_next;
5318 		for (;;) {
5319 			while (connp != NULL) {
5320 				if (IPCL_UDP_MATCH(connp, lport, laddr,
5321 				    fport, faddr) &&
5322 				    conn_wantpacket(connp, ira, ipha) &&
5323 				    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5324 				    tsol_receive_local(mp, &laddr, IPV4_VERSION,
5325 				    ira, connp)))
5326 					break;
5327 				connp = connp->conn_next;
5328 			}
5329 			if (connp == NULL) {
5330 				/* No more interested clients */
5331 				connp = first_connp;
5332 				break;
5333 			}
5334 			if (((mp1 = dupmsg(mp)) == NULL) &&
5335 			    ((mp1 = copymsg(mp)) == NULL)) {
5336 				/* Memory allocation failed */
5337 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5338 				ip_drop_input("ipIfStatsInDiscards", mp, ill);
5339 				connp = first_connp;
5340 				break;
5341 			}
5342 			CONN_INC_REF(connp);
5343 			mutex_exit(&connfp->connf_lock);
5344 
5345 			IP_STAT(ipst, ip_udp_fanmb);
5346 			ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5347 			    NULL, ira);
5348 			mutex_enter(&connfp->connf_lock);
5349 			/* Follow the next pointer before releasing the conn */
5350 			next_connp = connp->conn_next;
5351 			CONN_DEC_REF(connp);
5352 			connp = next_connp;
5353 		}
5354 	}
5355 
5356 	/* Last one.  Send it upstream. */
5357 	mutex_exit(&connfp->connf_lock);
5358 	IP_STAT(ipst, ip_udp_fanmb);
5359 	ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5360 	CONN_DEC_REF(connp);
5361 	return;
5362 
5363 notfound:
5364 	mutex_exit(&connfp->connf_lock);
5365 	/*
5366 	 * IPv6 endpoints bound to multicast IPv4-mapped addresses
5367 	 * have already been matched above, since they live in the IPv4
5368 	 * fanout tables. This implies we only need to
5369 	 * check for IPv6 in6addr_any endpoints here.
5370 	 * Thus we compare using ipv6_all_zeros instead of the destination
5371 	 * address, except for the multicast group membership lookup which
5372 	 * uses the IPv4 destination.
5373 	 */
5374 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6faddr);
5375 	connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5376 	mutex_enter(&connfp->connf_lock);
5377 	connp = connfp->connf_head;
5378 	/*
5379 	 * IPv4 multicast packet being delivered to an AF_INET6
5380 	 * in6addr_any endpoint.
5381 	 * Need to check conn_wantpacket(). Note that we use conn_wantpacket()
5382 	 * and not conn_wantpacket_v6() since any multicast membership is
5383 	 * for an IPv4-mapped multicast address.
5384 	 */
5385 	while (connp != NULL) {
5386 		if (IPCL_UDP_MATCH_V6(connp, lport, ipv6_all_zeros,
5387 		    fport, v6faddr) &&
5388 		    conn_wantpacket(connp, ira, ipha) &&
5389 		    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5390 		    tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5391 			break;
5392 		connp = connp->conn_next;
5393 	}
5394 
5395 	if (connp == NULL) {
5396 		/*
5397 		 * No one bound to this port.  Is
5398 		 * there a client that wants all
5399 		 * unclaimed datagrams?
5400 		 */
5401 		mutex_exit(&connfp->connf_lock);
5402 
5403 		if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_UDP].connf_head !=
5404 		    NULL) {
5405 			ASSERT(ira->ira_protocol == IPPROTO_UDP);
5406 			ip_fanout_proto_v4(mp, ipha, ira);
5407 		} else {
5408 			/*
5409 			 * We used to attempt to send an icmp error here, but
5410 			 * since this is known to be a multicast packet
5411 			 * and we don't send icmp errors in response to
5412 			 * multicast, just drop the packet and give up sooner.
5413 			 */
5414 			BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
5415 			freemsg(mp);
5416 		}
5417 		return;
5418 	}
5419 	ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5420 
5421 	/*
5422 	 * If SO_REUSEADDR has been set on the first we send the
5423 	 * packet to all clients that have joined the group and
5424 	 * match the port.
5425 	 */
5426 	if (connp->conn_reuseaddr) {
5427 		conn_t		*first_connp = connp;
5428 		conn_t		*next_connp;
5429 		mblk_t		*mp1;
5430 
5431 		CONN_INC_REF(connp);
5432 		connp = connp->conn_next;
5433 		for (;;) {
5434 			while (connp != NULL) {
5435 				if (IPCL_UDP_MATCH_V6(connp, lport,
5436 				    ipv6_all_zeros, fport, v6faddr) &&
5437 				    conn_wantpacket(connp, ira, ipha) &&
5438 				    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5439 				    tsol_receive_local(mp, &laddr, IPV4_VERSION,
5440 				    ira, connp)))
5441 					break;
5442 				connp = connp->conn_next;
5443 			}
5444 			if (connp == NULL) {
5445 				/* No more interested clients */
5446 				connp = first_connp;
5447 				break;
5448 			}
5449 			if (((mp1 = dupmsg(mp)) == NULL) &&
5450 			    ((mp1 = copymsg(mp)) == NULL)) {
5451 				/* Memory allocation failed */
5452 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5453 				ip_drop_input("ipIfStatsInDiscards", mp, ill);
5454 				connp = first_connp;
5455 				break;
5456 			}
5457 			CONN_INC_REF(connp);
5458 			mutex_exit(&connfp->connf_lock);
5459 
5460 			IP_STAT(ipst, ip_udp_fanmb);
5461 			ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5462 			    NULL, ira);
5463 			mutex_enter(&connfp->connf_lock);
5464 			/* Follow the next pointer before releasing the conn */
5465 			next_connp = connp->conn_next;
5466 			CONN_DEC_REF(connp);
5467 			connp = next_connp;
5468 		}
5469 	}
5470 
5471 	/* Last one.  Send it upstream. */
5472 	mutex_exit(&connfp->connf_lock);
5473 	IP_STAT(ipst, ip_udp_fanmb);
5474 	ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5475 	CONN_DEC_REF(connp);
5476 }
5477 
5478 /*
5479  * Split an incoming packet's IPv4 options into the label and the other options.
5480  * If 'allocate' is set it does memory allocation for the ip_pkt_t, including
5481  * clearing out any leftover label or options.
5482  * Otherwise it just makes ipp point into the packet.
5483  *
5484  * Returns zero if ok; ENOMEM if the buffer couldn't be allocated.
5485  */
5486 int
ip_find_hdr_v4(ipha_t * ipha,ip_pkt_t * ipp,boolean_t allocate)5487 ip_find_hdr_v4(ipha_t *ipha, ip_pkt_t *ipp, boolean_t allocate)
5488 {
5489 	uchar_t		*opt;
5490 	uint32_t	totallen;
5491 	uint32_t	optval;
5492 	uint32_t	optlen;
5493 
5494 	ipp->ipp_fields |= IPPF_HOPLIMIT | IPPF_TCLASS | IPPF_ADDR;
5495 	ipp->ipp_hoplimit = ipha->ipha_ttl;
5496 	ipp->ipp_type_of_service = ipha->ipha_type_of_service;
5497 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &ipp->ipp_addr);
5498 
5499 	/*
5500 	 * Get length (in 4 byte octets) of IP header options.
5501 	 */
5502 	totallen = ipha->ipha_version_and_hdr_length -
5503 	    (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5504 
5505 	if (totallen == 0) {
5506 		if (!allocate)
5507 			return (0);
5508 
5509 		/* Clear out anything from a previous packet */
5510 		if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5511 			kmem_free(ipp->ipp_ipv4_options,
5512 			    ipp->ipp_ipv4_options_len);
5513 			ipp->ipp_ipv4_options = NULL;
5514 			ipp->ipp_ipv4_options_len = 0;
5515 			ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5516 		}
5517 		if (ipp->ipp_fields & IPPF_LABEL_V4) {
5518 			kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5519 			ipp->ipp_label_v4 = NULL;
5520 			ipp->ipp_label_len_v4 = 0;
5521 			ipp->ipp_fields &= ~IPPF_LABEL_V4;
5522 		}
5523 		return (0);
5524 	}
5525 
5526 	totallen <<= 2;
5527 	opt = (uchar_t *)&ipha[1];
5528 	if (!is_system_labeled()) {
5529 
5530 	copyall:
5531 		if (!allocate) {
5532 			if (totallen != 0) {
5533 				ipp->ipp_ipv4_options = opt;
5534 				ipp->ipp_ipv4_options_len = totallen;
5535 				ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5536 			}
5537 			return (0);
5538 		}
5539 		/* Just copy all of options */
5540 		if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5541 			if (totallen == ipp->ipp_ipv4_options_len) {
5542 				bcopy(opt, ipp->ipp_ipv4_options, totallen);
5543 				return (0);
5544 			}
5545 			kmem_free(ipp->ipp_ipv4_options,
5546 			    ipp->ipp_ipv4_options_len);
5547 			ipp->ipp_ipv4_options = NULL;
5548 			ipp->ipp_ipv4_options_len = 0;
5549 			ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5550 		}
5551 		if (totallen == 0)
5552 			return (0);
5553 
5554 		ipp->ipp_ipv4_options = kmem_alloc(totallen, KM_NOSLEEP);
5555 		if (ipp->ipp_ipv4_options == NULL)
5556 			return (ENOMEM);
5557 		ipp->ipp_ipv4_options_len = totallen;
5558 		ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5559 		bcopy(opt, ipp->ipp_ipv4_options, totallen);
5560 		return (0);
5561 	}
5562 
5563 	if (allocate && (ipp->ipp_fields & IPPF_LABEL_V4)) {
5564 		kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5565 		ipp->ipp_label_v4 = NULL;
5566 		ipp->ipp_label_len_v4 = 0;
5567 		ipp->ipp_fields &= ~IPPF_LABEL_V4;
5568 	}
5569 
5570 	/*
5571 	 * Search for CIPSO option.
5572 	 * We assume CIPSO is first in options if it is present.
5573 	 * If it isn't, then ipp_opt_ipv4_options will not include the options
5574 	 * prior to the CIPSO option.
5575 	 */
5576 	while (totallen != 0) {
5577 		switch (optval = opt[IPOPT_OPTVAL]) {
5578 		case IPOPT_EOL:
5579 			return (0);
5580 		case IPOPT_NOP:
5581 			optlen = 1;
5582 			break;
5583 		default:
5584 			if (totallen <= IPOPT_OLEN)
5585 				return (EINVAL);
5586 			optlen = opt[IPOPT_OLEN];
5587 			if (optlen < 2)
5588 				return (EINVAL);
5589 		}
5590 		if (optlen > totallen)
5591 			return (EINVAL);
5592 
5593 		switch (optval) {
5594 		case IPOPT_COMSEC:
5595 			if (!allocate) {
5596 				ipp->ipp_label_v4 = opt;
5597 				ipp->ipp_label_len_v4 = optlen;
5598 				ipp->ipp_fields |= IPPF_LABEL_V4;
5599 			} else {
5600 				ipp->ipp_label_v4 = kmem_alloc(optlen,
5601 				    KM_NOSLEEP);
5602 				if (ipp->ipp_label_v4 == NULL)
5603 					return (ENOMEM);
5604 				ipp->ipp_label_len_v4 = optlen;
5605 				ipp->ipp_fields |= IPPF_LABEL_V4;
5606 				bcopy(opt, ipp->ipp_label_v4, optlen);
5607 			}
5608 			totallen -= optlen;
5609 			opt += optlen;
5610 
5611 			/* Skip padding bytes until we get to a multiple of 4 */
5612 			while ((totallen & 3) != 0 && opt[0] == IPOPT_NOP) {
5613 				totallen--;
5614 				opt++;
5615 			}
5616 			/* Remaining as ipp_ipv4_options */
5617 			goto copyall;
5618 		}
5619 		totallen -= optlen;
5620 		opt += optlen;
5621 	}
5622 	/* No CIPSO found; return everything as ipp_ipv4_options */
5623 	totallen = ipha->ipha_version_and_hdr_length -
5624 	    (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5625 	totallen <<= 2;
5626 	opt = (uchar_t *)&ipha[1];
5627 	goto copyall;
5628 }
5629 
5630 /*
5631  * Efficient versions of lookup for an IRE when we only
5632  * match the address.
5633  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5634  * Does not handle multicast addresses.
5635  */
5636 uint_t
ip_type_v4(ipaddr_t addr,ip_stack_t * ipst)5637 ip_type_v4(ipaddr_t addr, ip_stack_t *ipst)
5638 {
5639 	ire_t *ire;
5640 	uint_t result;
5641 
5642 	ire = ire_ftable_lookup_simple_v4(addr, 0, ipst, NULL);
5643 	ASSERT(ire != NULL);
5644 	if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5645 		result = IRE_NOROUTE;
5646 	else
5647 		result = ire->ire_type;
5648 	ire_refrele(ire);
5649 	return (result);
5650 }
5651 
5652 /*
5653  * Efficient versions of lookup for an IRE when we only
5654  * match the address.
5655  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5656  * Does not handle multicast addresses.
5657  */
5658 uint_t
ip_type_v6(const in6_addr_t * addr,ip_stack_t * ipst)5659 ip_type_v6(const in6_addr_t *addr, ip_stack_t *ipst)
5660 {
5661 	ire_t *ire;
5662 	uint_t result;
5663 
5664 	ire = ire_ftable_lookup_simple_v6(addr, 0, ipst, NULL);
5665 	ASSERT(ire != NULL);
5666 	if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5667 		result = IRE_NOROUTE;
5668 	else
5669 		result = ire->ire_type;
5670 	ire_refrele(ire);
5671 	return (result);
5672 }
5673 
5674 /*
5675  * Nobody should be sending
5676  * packets up this stream
5677  */
5678 static void
ip_lrput(queue_t * q,mblk_t * mp)5679 ip_lrput(queue_t *q, mblk_t *mp)
5680 {
5681 	switch (mp->b_datap->db_type) {
5682 	case M_FLUSH:
5683 		/* Turn around */
5684 		if (*mp->b_rptr & FLUSHW) {
5685 			*mp->b_rptr &= ~FLUSHR;
5686 			qreply(q, mp);
5687 			return;
5688 		}
5689 		break;
5690 	}
5691 	freemsg(mp);
5692 }
5693 
5694 /* Nobody should be sending packets down this stream */
5695 /* ARGSUSED */
5696 void
ip_lwput(queue_t * q,mblk_t * mp)5697 ip_lwput(queue_t *q, mblk_t *mp)
5698 {
5699 	freemsg(mp);
5700 }
5701 
5702 /*
5703  * Move the first hop in any source route to ipha_dst and remove that part of
5704  * the source route.  Called by other protocols.  Errors in option formatting
5705  * are ignored - will be handled by ip_output_options. Return the final
5706  * destination (either ipha_dst or the last entry in a source route.)
5707  */
5708 ipaddr_t
ip_massage_options(ipha_t * ipha,netstack_t * ns)5709 ip_massage_options(ipha_t *ipha, netstack_t *ns)
5710 {
5711 	ipoptp_t	opts;
5712 	uchar_t		*opt;
5713 	uint8_t		optval;
5714 	uint8_t		optlen;
5715 	ipaddr_t	dst;
5716 	int		i;
5717 	ip_stack_t	*ipst = ns->netstack_ip;
5718 
5719 	ip2dbg(("ip_massage_options\n"));
5720 	dst = ipha->ipha_dst;
5721 	for (optval = ipoptp_first(&opts, ipha);
5722 	    optval != IPOPT_EOL;
5723 	    optval = ipoptp_next(&opts)) {
5724 		opt = opts.ipoptp_cur;
5725 		switch (optval) {
5726 			uint8_t off;
5727 		case IPOPT_SSRR:
5728 		case IPOPT_LSRR:
5729 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
5730 				ip1dbg(("ip_massage_options: bad src route\n"));
5731 				break;
5732 			}
5733 			optlen = opts.ipoptp_len;
5734 			off = opt[IPOPT_OFFSET];
5735 			off--;
5736 		redo_srr:
5737 			if (optlen < IP_ADDR_LEN ||
5738 			    off > optlen - IP_ADDR_LEN) {
5739 				/* End of source route */
5740 				ip1dbg(("ip_massage_options: end of SR\n"));
5741 				break;
5742 			}
5743 			bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
5744 			ip1dbg(("ip_massage_options: next hop 0x%x\n",
5745 			    ntohl(dst)));
5746 			/*
5747 			 * Check if our address is present more than
5748 			 * once as consecutive hops in source route.
5749 			 * XXX verify per-interface ip_forwarding
5750 			 * for source route?
5751 			 */
5752 			if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
5753 				off += IP_ADDR_LEN;
5754 				goto redo_srr;
5755 			}
5756 			if (dst == htonl(INADDR_LOOPBACK)) {
5757 				ip1dbg(("ip_massage_options: loopback addr in "
5758 				    "source route!\n"));
5759 				break;
5760 			}
5761 			/*
5762 			 * Update ipha_dst to be the first hop and remove the
5763 			 * first hop from the source route (by overwriting
5764 			 * part of the option with NOP options).
5765 			 */
5766 			ipha->ipha_dst = dst;
5767 			/* Put the last entry in dst */
5768 			off = ((optlen - IP_ADDR_LEN - 3) & ~(IP_ADDR_LEN-1)) +
5769 			    3;
5770 			bcopy(&opt[off], &dst, IP_ADDR_LEN);
5771 
5772 			ip1dbg(("ip_massage_options: last hop 0x%x\n",
5773 			    ntohl(dst)));
5774 			/* Move down and overwrite */
5775 			opt[IP_ADDR_LEN] = opt[0];
5776 			opt[IP_ADDR_LEN+1] = opt[IPOPT_OLEN] - IP_ADDR_LEN;
5777 			opt[IP_ADDR_LEN+2] = opt[IPOPT_OFFSET];
5778 			for (i = 0; i < IP_ADDR_LEN; i++)
5779 				opt[i] = IPOPT_NOP;
5780 			break;
5781 		}
5782 	}
5783 	return (dst);
5784 }
5785 
5786 /*
5787  * Return the network mask
5788  * associated with the specified address.
5789  */
5790 ipaddr_t
ip_net_mask(ipaddr_t addr)5791 ip_net_mask(ipaddr_t addr)
5792 {
5793 	uchar_t	*up = (uchar_t *)&addr;
5794 	ipaddr_t mask = 0;
5795 	uchar_t	*maskp = (uchar_t *)&mask;
5796 
5797 #if defined(__i386) || defined(__amd64)
5798 #define	TOTALLY_BRAIN_DAMAGED_C_COMPILER
5799 #endif
5800 #ifdef  TOTALLY_BRAIN_DAMAGED_C_COMPILER
5801 	maskp[0] = maskp[1] = maskp[2] = maskp[3] = 0;
5802 #endif
5803 	if (CLASSD(addr)) {
5804 		maskp[0] = 0xF0;
5805 		return (mask);
5806 	}
5807 
5808 	/* We assume Class E default netmask to be 32 */
5809 	if (CLASSE(addr))
5810 		return (0xffffffffU);
5811 
5812 	if (addr == 0)
5813 		return (0);
5814 	maskp[0] = 0xFF;
5815 	if ((up[0] & 0x80) == 0)
5816 		return (mask);
5817 
5818 	maskp[1] = 0xFF;
5819 	if ((up[0] & 0xC0) == 0x80)
5820 		return (mask);
5821 
5822 	maskp[2] = 0xFF;
5823 	if ((up[0] & 0xE0) == 0xC0)
5824 		return (mask);
5825 
5826 	/* Otherwise return no mask */
5827 	return ((ipaddr_t)0);
5828 }
5829 
5830 /* Name/Value Table Lookup Routine */
5831 char *
ip_nv_lookup(nv_t * nv,int value)5832 ip_nv_lookup(nv_t *nv, int value)
5833 {
5834 	if (!nv)
5835 		return (NULL);
5836 	for (; nv->nv_name; nv++) {
5837 		if (nv->nv_value == value)
5838 			return (nv->nv_name);
5839 	}
5840 	return ("unknown");
5841 }
5842 
5843 static int
ip_wait_for_info_ack(ill_t * ill)5844 ip_wait_for_info_ack(ill_t *ill)
5845 {
5846 	int err;
5847 
5848 	mutex_enter(&ill->ill_lock);
5849 	while (ill->ill_state_flags & ILL_LL_SUBNET_PENDING) {
5850 		/*
5851 		 * Return value of 0 indicates a pending signal.
5852 		 */
5853 		err = cv_wait_sig(&ill->ill_cv, &ill->ill_lock);
5854 		if (err == 0) {
5855 			mutex_exit(&ill->ill_lock);
5856 			return (EINTR);
5857 		}
5858 	}
5859 	mutex_exit(&ill->ill_lock);
5860 	/*
5861 	 * ip_rput_other could have set an error  in ill_error on
5862 	 * receipt of M_ERROR.
5863 	 */
5864 	return (ill->ill_error);
5865 }
5866 
5867 /*
5868  * This is a module open, i.e. this is a control stream for access
5869  * to a DLPI device.  We allocate an ill_t as the instance data in
5870  * this case.
5871  */
5872 static int
ip_modopen(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp)5873 ip_modopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5874 {
5875 	ill_t	*ill;
5876 	int	err;
5877 	zoneid_t zoneid;
5878 	netstack_t *ns;
5879 	ip_stack_t *ipst;
5880 
5881 	/*
5882 	 * Prevent unprivileged processes from pushing IP so that
5883 	 * they can't send raw IP.
5884 	 */
5885 	if (secpolicy_net_rawaccess(credp) != 0)
5886 		return (EPERM);
5887 
5888 	ns = netstack_find_by_cred(credp);
5889 	ASSERT(ns != NULL);
5890 	ipst = ns->netstack_ip;
5891 	ASSERT(ipst != NULL);
5892 
5893 	/*
5894 	 * For exclusive stacks we set the zoneid to zero
5895 	 * to make IP operate as if in the global zone.
5896 	 */
5897 	if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5898 		zoneid = GLOBAL_ZONEID;
5899 	else
5900 		zoneid = crgetzoneid(credp);
5901 
5902 	ill = (ill_t *)mi_open_alloc_sleep(sizeof (ill_t));
5903 	q->q_ptr = WR(q)->q_ptr = ill;
5904 	ill->ill_ipst = ipst;
5905 	ill->ill_zoneid = zoneid;
5906 
5907 	/*
5908 	 * ill_init initializes the ill fields and then sends down
5909 	 * down a DL_INFO_REQ after calling qprocson.
5910 	 */
5911 	err = ill_init(q, ill);
5912 
5913 	if (err != 0) {
5914 		mi_free(ill);
5915 		netstack_rele(ipst->ips_netstack);
5916 		q->q_ptr = NULL;
5917 		WR(q)->q_ptr = NULL;
5918 		return (err);
5919 	}
5920 
5921 	/*
5922 	 * Wait for the DL_INFO_ACK if a DL_INFO_REQ was sent.
5923 	 *
5924 	 * ill_init initializes the ipsq marking this thread as
5925 	 * writer
5926 	 */
5927 	ipsq_exit(ill->ill_phyint->phyint_ipsq);
5928 	err = ip_wait_for_info_ack(ill);
5929 	if (err == 0)
5930 		ill->ill_credp = credp;
5931 	else
5932 		goto fail;
5933 
5934 	crhold(credp);
5935 
5936 	mutex_enter(&ipst->ips_ip_mi_lock);
5937 	err = mi_open_link(&ipst->ips_ip_g_head, (IDP)q->q_ptr, devp, flag,
5938 	    sflag, credp);
5939 	mutex_exit(&ipst->ips_ip_mi_lock);
5940 fail:
5941 	if (err) {
5942 		(void) ip_close(q, 0);
5943 		return (err);
5944 	}
5945 	return (0);
5946 }
5947 
5948 /* For /dev/ip aka AF_INET open */
5949 int
ip_openv4(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp)5950 ip_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5951 {
5952 	return (ip_open(q, devp, flag, sflag, credp, B_FALSE));
5953 }
5954 
5955 /* For /dev/ip6 aka AF_INET6 open */
5956 int
ip_openv6(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp)5957 ip_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5958 {
5959 	return (ip_open(q, devp, flag, sflag, credp, B_TRUE));
5960 }
5961 
5962 /* IP open routine. */
5963 int
ip_open(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp,boolean_t isv6)5964 ip_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
5965     boolean_t isv6)
5966 {
5967 	conn_t 		*connp;
5968 	major_t		maj;
5969 	zoneid_t	zoneid;
5970 	netstack_t	*ns;
5971 	ip_stack_t	*ipst;
5972 
5973 	/* Allow reopen. */
5974 	if (q->q_ptr != NULL)
5975 		return (0);
5976 
5977 	if (sflag & MODOPEN) {
5978 		/* This is a module open */
5979 		return (ip_modopen(q, devp, flag, sflag, credp));
5980 	}
5981 
5982 	if ((flag & ~(FKLYR)) == IP_HELPER_STR) {
5983 		/*
5984 		 * Non streams based socket looking for a stream
5985 		 * to access IP
5986 		 */
5987 		return (ip_helper_stream_setup(q, devp, flag, sflag,
5988 		    credp, isv6));
5989 	}
5990 
5991 	ns = netstack_find_by_cred(credp);
5992 	ASSERT(ns != NULL);
5993 	ipst = ns->netstack_ip;
5994 	ASSERT(ipst != NULL);
5995 
5996 	/*
5997 	 * For exclusive stacks we set the zoneid to zero
5998 	 * to make IP operate as if in the global zone.
5999 	 */
6000 	if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
6001 		zoneid = GLOBAL_ZONEID;
6002 	else
6003 		zoneid = crgetzoneid(credp);
6004 
6005 	/*
6006 	 * We are opening as a device. This is an IP client stream, and we
6007 	 * allocate an conn_t as the instance data.
6008 	 */
6009 	connp = ipcl_conn_create(IPCL_IPCCONN, KM_SLEEP, ipst->ips_netstack);
6010 
6011 	/*
6012 	 * ipcl_conn_create did a netstack_hold. Undo the hold that was
6013 	 * done by netstack_find_by_cred()
6014 	 */
6015 	netstack_rele(ipst->ips_netstack);
6016 
6017 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
6018 	/* conn_allzones can not be set this early, hence no IPCL_ZONEID */
6019 	connp->conn_ixa->ixa_zoneid = zoneid;
6020 	connp->conn_zoneid = zoneid;
6021 
6022 	connp->conn_rq = q;
6023 	q->q_ptr = WR(q)->q_ptr = connp;
6024 
6025 	/* Minor tells us which /dev entry was opened */
6026 	if (isv6) {
6027 		connp->conn_family = AF_INET6;
6028 		connp->conn_ipversion = IPV6_VERSION;
6029 		connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
6030 		connp->conn_ixa->ixa_src_preferences = IPV6_PREFER_SRC_DEFAULT;
6031 	} else {
6032 		connp->conn_family = AF_INET;
6033 		connp->conn_ipversion = IPV4_VERSION;
6034 		connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
6035 	}
6036 
6037 	if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
6038 	    ((connp->conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
6039 		connp->conn_minor_arena = ip_minor_arena_la;
6040 	} else {
6041 		/*
6042 		 * Either minor numbers in the large arena were exhausted
6043 		 * or a non socket application is doing the open.
6044 		 * Try to allocate from the small arena.
6045 		 */
6046 		if ((connp->conn_dev =
6047 		    inet_minor_alloc(ip_minor_arena_sa)) == 0) {
6048 			/* CONN_DEC_REF takes care of netstack_rele() */
6049 			q->q_ptr = WR(q)->q_ptr = NULL;
6050 			CONN_DEC_REF(connp);
6051 			return (EBUSY);
6052 		}
6053 		connp->conn_minor_arena = ip_minor_arena_sa;
6054 	}
6055 
6056 	maj = getemajor(*devp);
6057 	*devp = makedevice(maj, (minor_t)connp->conn_dev);
6058 
6059 	/*
6060 	 * connp->conn_cred is crfree()ed in ipcl_conn_destroy()
6061 	 */
6062 	connp->conn_cred = credp;
6063 	connp->conn_cpid = curproc->p_pid;
6064 	/* Cache things in ixa without an extra refhold */
6065 	ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
6066 	connp->conn_ixa->ixa_cred = connp->conn_cred;
6067 	connp->conn_ixa->ixa_cpid = connp->conn_cpid;
6068 	if (is_system_labeled())
6069 		connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
6070 
6071 	/*
6072 	 * Handle IP_IOC_RTS_REQUEST and other ioctls which use conn_recv
6073 	 */
6074 	connp->conn_recv = ip_conn_input;
6075 	connp->conn_recvicmp = ip_conn_input_icmp;
6076 
6077 	crhold(connp->conn_cred);
6078 
6079 	/*
6080 	 * If the caller has the process-wide flag set, then default to MAC
6081 	 * exempt mode.  This allows read-down to unlabeled hosts.
6082 	 */
6083 	if (getpflags(NET_MAC_AWARE, credp) != 0)
6084 		connp->conn_mac_mode = CONN_MAC_AWARE;
6085 
6086 	connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
6087 
6088 	connp->conn_rq = q;
6089 	connp->conn_wq = WR(q);
6090 
6091 	/* Non-zero default values */
6092 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP;
6093 
6094 	/*
6095 	 * Make the conn globally visible to walkers
6096 	 */
6097 	ASSERT(connp->conn_ref == 1);
6098 	mutex_enter(&connp->conn_lock);
6099 	connp->conn_state_flags &= ~CONN_INCIPIENT;
6100 	mutex_exit(&connp->conn_lock);
6101 
6102 	qprocson(q);
6103 
6104 	return (0);
6105 }
6106 
6107 /*
6108  * Set IPsec policy from an ipsec_req_t. If the req is not "zero" and valid,
6109  * all of them are copied to the conn_t. If the req is "zero", the policy is
6110  * zeroed out. A "zero" policy has zero ipsr_{ah,req,self_encap}_req
6111  * fields.
6112  * We keep only the latest setting of the policy and thus policy setting
6113  * is not incremental/cumulative.
6114  *
6115  * Requests to set policies with multiple alternative actions will
6116  * go through a different API.
6117  */
6118 int
ipsec_set_req(cred_t * cr,conn_t * connp,ipsec_req_t * req)6119 ipsec_set_req(cred_t *cr, conn_t *connp, ipsec_req_t *req)
6120 {
6121 	uint_t ah_req = 0;
6122 	uint_t esp_req = 0;
6123 	uint_t se_req = 0;
6124 	ipsec_act_t *actp = NULL;
6125 	uint_t nact;
6126 	ipsec_policy_head_t *ph;
6127 	boolean_t is_pol_reset, is_pol_inserted = B_FALSE;
6128 	int error = 0;
6129 	netstack_t	*ns = connp->conn_netstack;
6130 	ip_stack_t	*ipst = ns->netstack_ip;
6131 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
6132 
6133 #define	REQ_MASK (IPSEC_PREF_REQUIRED|IPSEC_PREF_NEVER)
6134 
6135 	/*
6136 	 * The IP_SEC_OPT option does not allow variable length parameters,
6137 	 * hence a request cannot be NULL.
6138 	 */
6139 	if (req == NULL)
6140 		return (EINVAL);
6141 
6142 	ah_req = req->ipsr_ah_req;
6143 	esp_req = req->ipsr_esp_req;
6144 	se_req = req->ipsr_self_encap_req;
6145 
6146 	/* Don't allow setting self-encap without one or more of AH/ESP. */
6147 	if (se_req != 0 && esp_req == 0 && ah_req == 0)
6148 		return (EINVAL);
6149 
6150 	/*
6151 	 * Are we dealing with a request to reset the policy (i.e.
6152 	 * zero requests).
6153 	 */
6154 	is_pol_reset = ((ah_req & REQ_MASK) == 0 &&
6155 	    (esp_req & REQ_MASK) == 0 &&
6156 	    (se_req & REQ_MASK) == 0);
6157 
6158 	if (!is_pol_reset) {
6159 		/*
6160 		 * If we couldn't load IPsec, fail with "protocol
6161 		 * not supported".
6162 		 * IPsec may not have been loaded for a request with zero
6163 		 * policies, so we don't fail in this case.
6164 		 */
6165 		mutex_enter(&ipss->ipsec_loader_lock);
6166 		if (ipss->ipsec_loader_state != IPSEC_LOADER_SUCCEEDED) {
6167 			mutex_exit(&ipss->ipsec_loader_lock);
6168 			return (EPROTONOSUPPORT);
6169 		}
6170 		mutex_exit(&ipss->ipsec_loader_lock);
6171 
6172 		/*
6173 		 * Test for valid requests. Invalid algorithms
6174 		 * need to be tested by IPsec code because new
6175 		 * algorithms can be added dynamically.
6176 		 */
6177 		if ((ah_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6178 		    (esp_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6179 		    (se_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0) {
6180 			return (EINVAL);
6181 		}
6182 
6183 		/*
6184 		 * Only privileged users can issue these
6185 		 * requests.
6186 		 */
6187 		if (((ah_req & IPSEC_PREF_NEVER) ||
6188 		    (esp_req & IPSEC_PREF_NEVER) ||
6189 		    (se_req & IPSEC_PREF_NEVER)) &&
6190 		    secpolicy_ip_config(cr, B_FALSE) != 0) {
6191 			return (EPERM);
6192 		}
6193 
6194 		/*
6195 		 * The IPSEC_PREF_REQUIRED and IPSEC_PREF_NEVER
6196 		 * are mutually exclusive.
6197 		 */
6198 		if (((ah_req & REQ_MASK) == REQ_MASK) ||
6199 		    ((esp_req & REQ_MASK) == REQ_MASK) ||
6200 		    ((se_req & REQ_MASK) == REQ_MASK)) {
6201 			/* Both of them are set */
6202 			return (EINVAL);
6203 		}
6204 	}
6205 
6206 	ASSERT(MUTEX_HELD(&connp->conn_lock));
6207 
6208 	/*
6209 	 * If we have already cached policies in conn_connect(), don't
6210 	 * let them change now. We cache policies for connections
6211 	 * whose src,dst [addr, port] is known.
6212 	 */
6213 	if (connp->conn_policy_cached) {
6214 		return (EINVAL);
6215 	}
6216 
6217 	/*
6218 	 * We have a zero policies, reset the connection policy if already
6219 	 * set. This will cause the connection to inherit the
6220 	 * global policy, if any.
6221 	 */
6222 	if (is_pol_reset) {
6223 		if (connp->conn_policy != NULL) {
6224 			IPPH_REFRELE(connp->conn_policy, ipst->ips_netstack);
6225 			connp->conn_policy = NULL;
6226 		}
6227 		connp->conn_in_enforce_policy = B_FALSE;
6228 		connp->conn_out_enforce_policy = B_FALSE;
6229 		return (0);
6230 	}
6231 
6232 	ph = connp->conn_policy = ipsec_polhead_split(connp->conn_policy,
6233 	    ipst->ips_netstack);
6234 	if (ph == NULL)
6235 		goto enomem;
6236 
6237 	ipsec_actvec_from_req(req, &actp, &nact, ipst->ips_netstack);
6238 	if (actp == NULL)
6239 		goto enomem;
6240 
6241 	/*
6242 	 * Always insert IPv4 policy entries, since they can also apply to
6243 	 * ipv6 sockets being used in ipv4-compat mode.
6244 	 */
6245 	if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6246 	    IPSEC_TYPE_INBOUND, ns))
6247 		goto enomem;
6248 	is_pol_inserted = B_TRUE;
6249 	if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6250 	    IPSEC_TYPE_OUTBOUND, ns))
6251 		goto enomem;
6252 
6253 	/*
6254 	 * We're looking at a v6 socket, also insert the v6-specific
6255 	 * entries.
6256 	 */
6257 	if (connp->conn_family == AF_INET6) {
6258 		if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6259 		    IPSEC_TYPE_INBOUND, ns))
6260 			goto enomem;
6261 		if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6262 		    IPSEC_TYPE_OUTBOUND, ns))
6263 			goto enomem;
6264 	}
6265 
6266 	ipsec_actvec_free(actp, nact);
6267 
6268 	/*
6269 	 * If the requests need security, set enforce_policy.
6270 	 * If the requests are IPSEC_PREF_NEVER, one should
6271 	 * still set conn_out_enforce_policy so that ip_set_destination
6272 	 * marks the ip_xmit_attr_t appropriatly. This is needed so that
6273 	 * for connections that we don't cache policy in at connect time,
6274 	 * if global policy matches in ip_output_attach_policy, we
6275 	 * don't wrongly inherit global policy. Similarly, we need
6276 	 * to set conn_in_enforce_policy also so that we don't verify
6277 	 * policy wrongly.
6278 	 */
6279 	if ((ah_req & REQ_MASK) != 0 ||
6280 	    (esp_req & REQ_MASK) != 0 ||
6281 	    (se_req & REQ_MASK) != 0) {
6282 		connp->conn_in_enforce_policy = B_TRUE;
6283 		connp->conn_out_enforce_policy = B_TRUE;
6284 	}
6285 
6286 	return (error);
6287 #undef REQ_MASK
6288 
6289 	/*
6290 	 * Common memory-allocation-failure exit path.
6291 	 */
6292 enomem:
6293 	if (actp != NULL)
6294 		ipsec_actvec_free(actp, nact);
6295 	if (is_pol_inserted)
6296 		ipsec_polhead_flush(ph, ns);
6297 	return (ENOMEM);
6298 }
6299 
6300 /*
6301  * Set socket options for joining and leaving multicast groups.
6302  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6303  * The caller has already check that the option name is consistent with
6304  * the address family of the socket.
6305  */
6306 int
ip_opt_set_multicast_group(conn_t * connp,t_scalar_t name,uchar_t * invalp,boolean_t inet6,boolean_t checkonly)6307 ip_opt_set_multicast_group(conn_t *connp, t_scalar_t name,
6308     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6309 {
6310 	int		*i1 = (int *)invalp;
6311 	int		error = 0;
6312 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
6313 	struct ip_mreq	*v4_mreqp;
6314 	struct ipv6_mreq *v6_mreqp;
6315 	struct group_req *greqp;
6316 	ire_t *ire;
6317 	boolean_t done = B_FALSE;
6318 	ipaddr_t ifaddr;
6319 	in6_addr_t v6group;
6320 	uint_t ifindex;
6321 	boolean_t mcast_opt = B_TRUE;
6322 	mcast_record_t fmode;
6323 	int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6324 	    ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6325 
6326 	switch (name) {
6327 	case IP_ADD_MEMBERSHIP:
6328 	case IPV6_JOIN_GROUP:
6329 		mcast_opt = B_FALSE;
6330 		/* FALLTHRU */
6331 	case MCAST_JOIN_GROUP:
6332 		fmode = MODE_IS_EXCLUDE;
6333 		optfn = ip_opt_add_group;
6334 		break;
6335 
6336 	case IP_DROP_MEMBERSHIP:
6337 	case IPV6_LEAVE_GROUP:
6338 		mcast_opt = B_FALSE;
6339 		/* FALLTHRU */
6340 	case MCAST_LEAVE_GROUP:
6341 		fmode = MODE_IS_INCLUDE;
6342 		optfn = ip_opt_delete_group;
6343 		break;
6344 	default:
6345 		ASSERT(0);
6346 	}
6347 
6348 	if (mcast_opt) {
6349 		struct sockaddr_in *sin;
6350 		struct sockaddr_in6 *sin6;
6351 
6352 		greqp = (struct group_req *)i1;
6353 		if (greqp->gr_group.ss_family == AF_INET) {
6354 			sin = (struct sockaddr_in *)&(greqp->gr_group);
6355 			IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &v6group);
6356 		} else {
6357 			if (!inet6)
6358 				return (EINVAL);	/* Not on INET socket */
6359 
6360 			sin6 = (struct sockaddr_in6 *)&(greqp->gr_group);
6361 			v6group = sin6->sin6_addr;
6362 		}
6363 		ifaddr = INADDR_ANY;
6364 		ifindex = greqp->gr_interface;
6365 	} else if (inet6) {
6366 		v6_mreqp = (struct ipv6_mreq *)i1;
6367 		v6group = v6_mreqp->ipv6mr_multiaddr;
6368 		ifaddr = INADDR_ANY;
6369 		ifindex = v6_mreqp->ipv6mr_interface;
6370 	} else {
6371 		v4_mreqp = (struct ip_mreq *)i1;
6372 		IN6_INADDR_TO_V4MAPPED(&v4_mreqp->imr_multiaddr, &v6group);
6373 		ifaddr = (ipaddr_t)v4_mreqp->imr_interface.s_addr;
6374 		ifindex = 0;
6375 	}
6376 
6377 	/*
6378 	 * In the multirouting case, we need to replicate
6379 	 * the request on all interfaces that will take part
6380 	 * in replication.  We do so because multirouting is
6381 	 * reflective, thus we will probably receive multi-
6382 	 * casts on those interfaces.
6383 	 * The ip_multirt_apply_membership() succeeds if
6384 	 * the operation succeeds on at least one interface.
6385 	 */
6386 	if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6387 		ipaddr_t group;
6388 
6389 		IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6390 
6391 		ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6392 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6393 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6394 	} else {
6395 		ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6396 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6397 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6398 	}
6399 	if (ire != NULL) {
6400 		if (ire->ire_flags & RTF_MULTIRT) {
6401 			error = ip_multirt_apply_membership(optfn, ire, connp,
6402 			    checkonly, &v6group, fmode, &ipv6_all_zeros);
6403 			done = B_TRUE;
6404 		}
6405 		ire_refrele(ire);
6406 	}
6407 
6408 	if (!done) {
6409 		error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6410 		    fmode, &ipv6_all_zeros);
6411 	}
6412 	return (error);
6413 }
6414 
6415 /*
6416  * Set socket options for joining and leaving multicast groups
6417  * for specific sources.
6418  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6419  * The caller has already check that the option name is consistent with
6420  * the address family of the socket.
6421  */
6422 int
ip_opt_set_multicast_sources(conn_t * connp,t_scalar_t name,uchar_t * invalp,boolean_t inet6,boolean_t checkonly)6423 ip_opt_set_multicast_sources(conn_t *connp, t_scalar_t name,
6424     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6425 {
6426 	int		*i1 = (int *)invalp;
6427 	int		error = 0;
6428 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
6429 	struct ip_mreq_source *imreqp;
6430 	struct group_source_req *gsreqp;
6431 	in6_addr_t v6group, v6src;
6432 	uint32_t ifindex;
6433 	ipaddr_t ifaddr;
6434 	boolean_t mcast_opt = B_TRUE;
6435 	mcast_record_t fmode;
6436 	ire_t *ire;
6437 	boolean_t done = B_FALSE;
6438 	int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6439 	    ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6440 
6441 	switch (name) {
6442 	case IP_BLOCK_SOURCE:
6443 		mcast_opt = B_FALSE;
6444 		/* FALLTHRU */
6445 	case MCAST_BLOCK_SOURCE:
6446 		fmode = MODE_IS_EXCLUDE;
6447 		optfn = ip_opt_add_group;
6448 		break;
6449 
6450 	case IP_UNBLOCK_SOURCE:
6451 		mcast_opt = B_FALSE;
6452 		/* FALLTHRU */
6453 	case MCAST_UNBLOCK_SOURCE:
6454 		fmode = MODE_IS_EXCLUDE;
6455 		optfn = ip_opt_delete_group;
6456 		break;
6457 
6458 	case IP_ADD_SOURCE_MEMBERSHIP:
6459 		mcast_opt = B_FALSE;
6460 		/* FALLTHRU */
6461 	case MCAST_JOIN_SOURCE_GROUP:
6462 		fmode = MODE_IS_INCLUDE;
6463 		optfn = ip_opt_add_group;
6464 		break;
6465 
6466 	case IP_DROP_SOURCE_MEMBERSHIP:
6467 		mcast_opt = B_FALSE;
6468 		/* FALLTHRU */
6469 	case MCAST_LEAVE_SOURCE_GROUP:
6470 		fmode = MODE_IS_INCLUDE;
6471 		optfn = ip_opt_delete_group;
6472 		break;
6473 	default:
6474 		ASSERT(0);
6475 	}
6476 
6477 	if (mcast_opt) {
6478 		gsreqp = (struct group_source_req *)i1;
6479 		ifindex = gsreqp->gsr_interface;
6480 		if (gsreqp->gsr_group.ss_family == AF_INET) {
6481 			struct sockaddr_in *s;
6482 			s = (struct sockaddr_in *)&gsreqp->gsr_group;
6483 			IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6group);
6484 			s = (struct sockaddr_in *)&gsreqp->gsr_source;
6485 			IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6src);
6486 		} else {
6487 			struct sockaddr_in6 *s6;
6488 
6489 			if (!inet6)
6490 				return (EINVAL);	/* Not on INET socket */
6491 
6492 			s6 = (struct sockaddr_in6 *)&gsreqp->gsr_group;
6493 			v6group = s6->sin6_addr;
6494 			s6 = (struct sockaddr_in6 *)&gsreqp->gsr_source;
6495 			v6src = s6->sin6_addr;
6496 		}
6497 		ifaddr = INADDR_ANY;
6498 	} else {
6499 		imreqp = (struct ip_mreq_source *)i1;
6500 		IN6_INADDR_TO_V4MAPPED(&imreqp->imr_multiaddr, &v6group);
6501 		IN6_INADDR_TO_V4MAPPED(&imreqp->imr_sourceaddr, &v6src);
6502 		ifaddr = (ipaddr_t)imreqp->imr_interface.s_addr;
6503 		ifindex = 0;
6504 	}
6505 
6506 	/*
6507 	 * Handle src being mapped INADDR_ANY by changing it to unspecified.
6508 	 */
6509 	if (IN6_IS_ADDR_V4MAPPED_ANY(&v6src))
6510 		v6src = ipv6_all_zeros;
6511 
6512 	/*
6513 	 * In the multirouting case, we need to replicate
6514 	 * the request as noted in the mcast cases above.
6515 	 */
6516 	if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6517 		ipaddr_t group;
6518 
6519 		IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6520 
6521 		ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6522 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6523 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6524 	} else {
6525 		ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6526 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6527 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6528 	}
6529 	if (ire != NULL) {
6530 		if (ire->ire_flags & RTF_MULTIRT) {
6531 			error = ip_multirt_apply_membership(optfn, ire, connp,
6532 			    checkonly, &v6group, fmode, &v6src);
6533 			done = B_TRUE;
6534 		}
6535 		ire_refrele(ire);
6536 	}
6537 	if (!done) {
6538 		error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6539 		    fmode, &v6src);
6540 	}
6541 	return (error);
6542 }
6543 
6544 /*
6545  * Given a destination address and a pointer to where to put the information
6546  * this routine fills in the mtuinfo.
6547  * The socket must be connected.
6548  * For sctp conn_faddr is the primary address.
6549  */
6550 int
ip_fill_mtuinfo(conn_t * connp,ip_xmit_attr_t * ixa,struct ip6_mtuinfo * mtuinfo)6551 ip_fill_mtuinfo(conn_t *connp, ip_xmit_attr_t *ixa, struct ip6_mtuinfo *mtuinfo)
6552 {
6553 	uint32_t	pmtu = IP_MAXPACKET;
6554 	uint_t		scopeid;
6555 
6556 	if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6))
6557 		return (-1);
6558 
6559 	/* In case we never sent or called ip_set_destination_v4/v6 */
6560 	if (ixa->ixa_ire != NULL)
6561 		pmtu = ip_get_pmtu(ixa);
6562 
6563 	if (ixa->ixa_flags & IXAF_SCOPEID_SET)
6564 		scopeid = ixa->ixa_scopeid;
6565 	else
6566 		scopeid = 0;
6567 
6568 	bzero(mtuinfo, sizeof (*mtuinfo));
6569 	mtuinfo->ip6m_addr.sin6_family = AF_INET6;
6570 	mtuinfo->ip6m_addr.sin6_port = connp->conn_fport;
6571 	mtuinfo->ip6m_addr.sin6_addr = connp->conn_faddr_v6;
6572 	mtuinfo->ip6m_addr.sin6_scope_id = scopeid;
6573 	mtuinfo->ip6m_mtu = pmtu;
6574 
6575 	return (sizeof (struct ip6_mtuinfo));
6576 }
6577 
6578 /*
6579  * When the src multihoming is changed from weak to [strong, preferred]
6580  * ip_ire_rebind_walker is called to walk the list of all ire_t entries
6581  * and identify routes that were created by user-applications in the
6582  * unbound state (i.e., without RTA_IFP), and for which an ire_ill is not
6583  * currently defined. These routes are then 'rebound', i.e., their ire_ill
6584  * is selected by finding an interface route for the gateway.
6585  */
6586 /* ARGSUSED */
6587 void
ip_ire_rebind_walker(ire_t * ire,void * notused)6588 ip_ire_rebind_walker(ire_t *ire, void *notused)
6589 {
6590 	if (!ire->ire_unbound || ire->ire_ill != NULL)
6591 		return;
6592 	ire_rebind(ire);
6593 	ire_delete(ire);
6594 }
6595 
6596 /*
6597  * When the src multihoming is changed from  [strong, preferred] to weak,
6598  * ip_ire_unbind_walker is called to walk the list of all ire_t entries, and
6599  * set any entries that were created by user-applications in the unbound state
6600  * (i.e., without RTA_IFP) back to having a NULL ire_ill.
6601  */
6602 /* ARGSUSED */
6603 void
ip_ire_unbind_walker(ire_t * ire,void * notused)6604 ip_ire_unbind_walker(ire_t *ire, void *notused)
6605 {
6606 	ire_t *new_ire;
6607 
6608 	if (!ire->ire_unbound || ire->ire_ill == NULL)
6609 		return;
6610 	if (ire->ire_ipversion == IPV6_VERSION) {
6611 		new_ire = ire_create_v6(&ire->ire_addr_v6, &ire->ire_mask_v6,
6612 		    &ire->ire_gateway_addr_v6, ire->ire_type, NULL,
6613 		    ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6614 	} else {
6615 		new_ire = ire_create((uchar_t *)&ire->ire_addr,
6616 		    (uchar_t *)&ire->ire_mask,
6617 		    (uchar_t *)&ire->ire_gateway_addr, ire->ire_type, NULL,
6618 		    ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6619 	}
6620 	if (new_ire == NULL)
6621 		return;
6622 	new_ire->ire_unbound = B_TRUE;
6623 	/*
6624 	 * The bound ire must first be deleted so that we don't return
6625 	 * the existing one on the attempt to add the unbound new_ire.
6626 	 */
6627 	ire_delete(ire);
6628 	new_ire = ire_add(new_ire);
6629 	if (new_ire != NULL)
6630 		ire_refrele(new_ire);
6631 }
6632 
6633 /*
6634  * When the settings of ip*_strict_src_multihoming tunables are changed,
6635  * all cached routes need to be recomputed. This recomputation needs to be
6636  * done when going from weaker to stronger modes so that the cached ire
6637  * for the connection does not violate the current ip*_strict_src_multihoming
6638  * setting. It also needs to be done when going from stronger to weaker modes,
6639  * so that we fall back to matching on the longest-matching-route (as opposed
6640  * to a shorter match that may have been selected in the strong mode
6641  * to satisfy src_multihoming settings).
6642  *
6643  * The cached ixa_ire entires for all conn_t entries are marked as
6644  * "verify" so that they will be recomputed for the next packet.
6645  */
6646 void
conn_ire_revalidate(conn_t * connp,void * arg)6647 conn_ire_revalidate(conn_t *connp, void *arg)
6648 {
6649 	boolean_t isv6 = (boolean_t)arg;
6650 
6651 	if ((isv6 && connp->conn_ipversion != IPV6_VERSION) ||
6652 	    (!isv6 && connp->conn_ipversion != IPV4_VERSION))
6653 		return;
6654 	connp->conn_ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
6655 }
6656 
6657 /*
6658  * Handles both IPv4 and IPv6 reassembly - doing the out-of-order cases,
6659  * When an ipf is passed here for the first time, if
6660  * we already have in-order fragments on the queue, we convert from the fast-
6661  * path reassembly scheme to the hard-case scheme.  From then on, additional
6662  * fragments are reassembled here.  We keep track of the start and end offsets
6663  * of each piece, and the number of holes in the chain.  When the hole count
6664  * goes to zero, we are done!
6665  *
6666  * The ipf_count will be updated to account for any mblk(s) added (pointed to
6667  * by mp) or subtracted (freeb()ed dups), upon return the caller must update
6668  * ipfb_count and ill_frag_count by the difference of ipf_count before and
6669  * after the call to ip_reassemble().
6670  */
6671 int
ip_reassemble(mblk_t * mp,ipf_t * ipf,uint_t start,boolean_t more,ill_t * ill,size_t msg_len)6672 ip_reassemble(mblk_t *mp, ipf_t *ipf, uint_t start, boolean_t more, ill_t *ill,
6673     size_t msg_len)
6674 {
6675 	uint_t	end;
6676 	mblk_t	*next_mp;
6677 	mblk_t	*mp1;
6678 	uint_t	offset;
6679 	boolean_t incr_dups = B_TRUE;
6680 	boolean_t offset_zero_seen = B_FALSE;
6681 	boolean_t pkt_boundary_checked = B_FALSE;
6682 
6683 	/* If start == 0 then ipf_nf_hdr_len has to be set. */
6684 	ASSERT(start != 0 || ipf->ipf_nf_hdr_len != 0);
6685 
6686 	/* Add in byte count */
6687 	ipf->ipf_count += msg_len;
6688 	if (ipf->ipf_end) {
6689 		/*
6690 		 * We were part way through in-order reassembly, but now there
6691 		 * is a hole.  We walk through messages already queued, and
6692 		 * mark them for hard case reassembly.  We know that up till
6693 		 * now they were in order starting from offset zero.
6694 		 */
6695 		offset = 0;
6696 		for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6697 			IP_REASS_SET_START(mp1, offset);
6698 			if (offset == 0) {
6699 				ASSERT(ipf->ipf_nf_hdr_len != 0);
6700 				offset = -ipf->ipf_nf_hdr_len;
6701 			}
6702 			offset += mp1->b_wptr - mp1->b_rptr;
6703 			IP_REASS_SET_END(mp1, offset);
6704 		}
6705 		/* One hole at the end. */
6706 		ipf->ipf_hole_cnt = 1;
6707 		/* Brand it as a hard case, forever. */
6708 		ipf->ipf_end = 0;
6709 	}
6710 	/* Walk through all the new pieces. */
6711 	do {
6712 		end = start + (mp->b_wptr - mp->b_rptr);
6713 		/*
6714 		 * If start is 0, decrease 'end' only for the first mblk of
6715 		 * the fragment. Otherwise 'end' can get wrong value in the
6716 		 * second pass of the loop if first mblk is exactly the
6717 		 * size of ipf_nf_hdr_len.
6718 		 */
6719 		if (start == 0 && !offset_zero_seen) {
6720 			/* First segment */
6721 			ASSERT(ipf->ipf_nf_hdr_len != 0);
6722 			end -= ipf->ipf_nf_hdr_len;
6723 			offset_zero_seen = B_TRUE;
6724 		}
6725 		next_mp = mp->b_cont;
6726 		/*
6727 		 * We are checking to see if there is any interesing data
6728 		 * to process.  If there isn't and the mblk isn't the
6729 		 * one which carries the unfragmentable header then we
6730 		 * drop it.  It's possible to have just the unfragmentable
6731 		 * header come through without any data.  That needs to be
6732 		 * saved.
6733 		 *
6734 		 * If the assert at the top of this function holds then the
6735 		 * term "ipf->ipf_nf_hdr_len != 0" isn't needed.  This code
6736 		 * is infrequently traveled enough that the test is left in
6737 		 * to protect against future code changes which break that
6738 		 * invariant.
6739 		 */
6740 		if (start == end && start != 0 && ipf->ipf_nf_hdr_len != 0) {
6741 			/* Empty.  Blast it. */
6742 			IP_REASS_SET_START(mp, 0);
6743 			IP_REASS_SET_END(mp, 0);
6744 			/*
6745 			 * If the ipf points to the mblk we are about to free,
6746 			 * update ipf to point to the next mblk (or NULL
6747 			 * if none).
6748 			 */
6749 			if (ipf->ipf_mp->b_cont == mp)
6750 				ipf->ipf_mp->b_cont = next_mp;
6751 			freeb(mp);
6752 			continue;
6753 		}
6754 		mp->b_cont = NULL;
6755 		IP_REASS_SET_START(mp, start);
6756 		IP_REASS_SET_END(mp, end);
6757 		if (!ipf->ipf_tail_mp) {
6758 			ipf->ipf_tail_mp = mp;
6759 			ipf->ipf_mp->b_cont = mp;
6760 			if (start == 0 || !more) {
6761 				ipf->ipf_hole_cnt = 1;
6762 				/*
6763 				 * if the first fragment comes in more than one
6764 				 * mblk, this loop will be executed for each
6765 				 * mblk. Need to adjust hole count so exiting
6766 				 * this routine will leave hole count at 1.
6767 				 */
6768 				if (next_mp)
6769 					ipf->ipf_hole_cnt++;
6770 			} else
6771 				ipf->ipf_hole_cnt = 2;
6772 			continue;
6773 		} else if (ipf->ipf_last_frag_seen && !more &&
6774 		    !pkt_boundary_checked) {
6775 			/*
6776 			 * We check datagram boundary only if this fragment
6777 			 * claims to be the last fragment and we have seen a
6778 			 * last fragment in the past too. We do this only
6779 			 * once for a given fragment.
6780 			 *
6781 			 * start cannot be 0 here as fragments with start=0
6782 			 * and MF=0 gets handled as a complete packet. These
6783 			 * fragments should not reach here.
6784 			 */
6785 
6786 			if (start + msgdsize(mp) !=
6787 			    IP_REASS_END(ipf->ipf_tail_mp)) {
6788 				/*
6789 				 * We have two fragments both of which claim
6790 				 * to be the last fragment but gives conflicting
6791 				 * information about the whole datagram size.
6792 				 * Something fishy is going on. Drop the
6793 				 * fragment and free up the reassembly list.
6794 				 */
6795 				return (IP_REASS_FAILED);
6796 			}
6797 
6798 			/*
6799 			 * We shouldn't come to this code block again for this
6800 			 * particular fragment.
6801 			 */
6802 			pkt_boundary_checked = B_TRUE;
6803 		}
6804 
6805 		/* New stuff at or beyond tail? */
6806 		offset = IP_REASS_END(ipf->ipf_tail_mp);
6807 		if (start >= offset) {
6808 			if (ipf->ipf_last_frag_seen) {
6809 				/* current fragment is beyond last fragment */
6810 				return (IP_REASS_FAILED);
6811 			}
6812 			/* Link it on end. */
6813 			ipf->ipf_tail_mp->b_cont = mp;
6814 			ipf->ipf_tail_mp = mp;
6815 			if (more) {
6816 				if (start != offset)
6817 					ipf->ipf_hole_cnt++;
6818 			} else if (start == offset && next_mp == NULL)
6819 					ipf->ipf_hole_cnt--;
6820 			continue;
6821 		}
6822 		mp1 = ipf->ipf_mp->b_cont;
6823 		offset = IP_REASS_START(mp1);
6824 		/* New stuff at the front? */
6825 		if (start < offset) {
6826 			if (start == 0) {
6827 				if (end >= offset) {
6828 					/* Nailed the hole at the begining. */
6829 					ipf->ipf_hole_cnt--;
6830 				}
6831 			} else if (end < offset) {
6832 				/*
6833 				 * A hole, stuff, and a hole where there used
6834 				 * to be just a hole.
6835 				 */
6836 				ipf->ipf_hole_cnt++;
6837 			}
6838 			mp->b_cont = mp1;
6839 			/* Check for overlap. */
6840 			while (end > offset) {
6841 				if (end < IP_REASS_END(mp1)) {
6842 					mp->b_wptr -= end - offset;
6843 					IP_REASS_SET_END(mp, offset);
6844 					BUMP_MIB(ill->ill_ip_mib,
6845 					    ipIfStatsReasmPartDups);
6846 					break;
6847 				}
6848 				/* Did we cover another hole? */
6849 				if ((mp1->b_cont &&
6850 				    IP_REASS_END(mp1) !=
6851 				    IP_REASS_START(mp1->b_cont) &&
6852 				    end >= IP_REASS_START(mp1->b_cont)) ||
6853 				    (!ipf->ipf_last_frag_seen && !more)) {
6854 					ipf->ipf_hole_cnt--;
6855 				}
6856 				/* Clip out mp1. */
6857 				if ((mp->b_cont = mp1->b_cont) == NULL) {
6858 					/*
6859 					 * After clipping out mp1, this guy
6860 					 * is now hanging off the end.
6861 					 */
6862 					ipf->ipf_tail_mp = mp;
6863 				}
6864 				IP_REASS_SET_START(mp1, 0);
6865 				IP_REASS_SET_END(mp1, 0);
6866 				/* Subtract byte count */
6867 				ipf->ipf_count -= mp1->b_datap->db_lim -
6868 				    mp1->b_datap->db_base;
6869 				freeb(mp1);
6870 				BUMP_MIB(ill->ill_ip_mib,
6871 				    ipIfStatsReasmPartDups);
6872 				mp1 = mp->b_cont;
6873 				if (!mp1)
6874 					break;
6875 				offset = IP_REASS_START(mp1);
6876 			}
6877 			ipf->ipf_mp->b_cont = mp;
6878 			continue;
6879 		}
6880 		/*
6881 		 * The new piece starts somewhere between the start of the head
6882 		 * and before the end of the tail.
6883 		 */
6884 		for (; mp1; mp1 = mp1->b_cont) {
6885 			offset = IP_REASS_END(mp1);
6886 			if (start < offset) {
6887 				if (end <= offset) {
6888 					/* Nothing new. */
6889 					IP_REASS_SET_START(mp, 0);
6890 					IP_REASS_SET_END(mp, 0);
6891 					/* Subtract byte count */
6892 					ipf->ipf_count -= mp->b_datap->db_lim -
6893 					    mp->b_datap->db_base;
6894 					if (incr_dups) {
6895 						ipf->ipf_num_dups++;
6896 						incr_dups = B_FALSE;
6897 					}
6898 					freeb(mp);
6899 					BUMP_MIB(ill->ill_ip_mib,
6900 					    ipIfStatsReasmDuplicates);
6901 					break;
6902 				}
6903 				/*
6904 				 * Trim redundant stuff off beginning of new
6905 				 * piece.
6906 				 */
6907 				IP_REASS_SET_START(mp, offset);
6908 				mp->b_rptr += offset - start;
6909 				BUMP_MIB(ill->ill_ip_mib,
6910 				    ipIfStatsReasmPartDups);
6911 				start = offset;
6912 				if (!mp1->b_cont) {
6913 					/*
6914 					 * After trimming, this guy is now
6915 					 * hanging off the end.
6916 					 */
6917 					mp1->b_cont = mp;
6918 					ipf->ipf_tail_mp = mp;
6919 					if (!more) {
6920 						ipf->ipf_hole_cnt--;
6921 					}
6922 					break;
6923 				}
6924 			}
6925 			if (start >= IP_REASS_START(mp1->b_cont))
6926 				continue;
6927 			/* Fill a hole */
6928 			if (start > offset)
6929 				ipf->ipf_hole_cnt++;
6930 			mp->b_cont = mp1->b_cont;
6931 			mp1->b_cont = mp;
6932 			mp1 = mp->b_cont;
6933 			offset = IP_REASS_START(mp1);
6934 			if (end >= offset) {
6935 				ipf->ipf_hole_cnt--;
6936 				/* Check for overlap. */
6937 				while (end > offset) {
6938 					if (end < IP_REASS_END(mp1)) {
6939 						mp->b_wptr -= end - offset;
6940 						IP_REASS_SET_END(mp, offset);
6941 						/*
6942 						 * TODO we might bump
6943 						 * this up twice if there is
6944 						 * overlap at both ends.
6945 						 */
6946 						BUMP_MIB(ill->ill_ip_mib,
6947 						    ipIfStatsReasmPartDups);
6948 						break;
6949 					}
6950 					/* Did we cover another hole? */
6951 					if ((mp1->b_cont &&
6952 					    IP_REASS_END(mp1)
6953 					    != IP_REASS_START(mp1->b_cont) &&
6954 					    end >=
6955 					    IP_REASS_START(mp1->b_cont)) ||
6956 					    (!ipf->ipf_last_frag_seen &&
6957 					    !more)) {
6958 						ipf->ipf_hole_cnt--;
6959 					}
6960 					/* Clip out mp1. */
6961 					if ((mp->b_cont = mp1->b_cont) ==
6962 					    NULL) {
6963 						/*
6964 						 * After clipping out mp1,
6965 						 * this guy is now hanging
6966 						 * off the end.
6967 						 */
6968 						ipf->ipf_tail_mp = mp;
6969 					}
6970 					IP_REASS_SET_START(mp1, 0);
6971 					IP_REASS_SET_END(mp1, 0);
6972 					/* Subtract byte count */
6973 					ipf->ipf_count -=
6974 					    mp1->b_datap->db_lim -
6975 					    mp1->b_datap->db_base;
6976 					freeb(mp1);
6977 					BUMP_MIB(ill->ill_ip_mib,
6978 					    ipIfStatsReasmPartDups);
6979 					mp1 = mp->b_cont;
6980 					if (!mp1)
6981 						break;
6982 					offset = IP_REASS_START(mp1);
6983 				}
6984 			}
6985 			break;
6986 		}
6987 	} while (start = end, mp = next_mp);
6988 
6989 	/* Fragment just processed could be the last one. Remember this fact */
6990 	if (!more)
6991 		ipf->ipf_last_frag_seen = B_TRUE;
6992 
6993 	/* Still got holes? */
6994 	if (ipf->ipf_hole_cnt)
6995 		return (IP_REASS_PARTIAL);
6996 	/* Clean up overloaded fields to avoid upstream disasters. */
6997 	for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6998 		IP_REASS_SET_START(mp1, 0);
6999 		IP_REASS_SET_END(mp1, 0);
7000 	}
7001 	return (IP_REASS_COMPLETE);
7002 }
7003 
7004 /*
7005  * Fragmentation reassembly.  Each ILL has a hash table for
7006  * queuing packets undergoing reassembly for all IPIFs
7007  * associated with the ILL.  The hash is based on the packet
7008  * IP ident field.  The ILL frag hash table was allocated
7009  * as a timer block at the time the ILL was created.  Whenever
7010  * there is anything on the reassembly queue, the timer will
7011  * be running.  Returns the reassembled packet if reassembly completes.
7012  */
7013 mblk_t *
ip_input_fragment(mblk_t * mp,ipha_t * ipha,ip_recv_attr_t * ira)7014 ip_input_fragment(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
7015 {
7016 	uint32_t	frag_offset_flags;
7017 	mblk_t		*t_mp;
7018 	ipaddr_t	dst;
7019 	uint8_t		proto = ipha->ipha_protocol;
7020 	uint32_t	sum_val;
7021 	uint16_t	sum_flags;
7022 	ipf_t		*ipf;
7023 	ipf_t		**ipfp;
7024 	ipfb_t		*ipfb;
7025 	uint16_t	ident;
7026 	uint32_t	offset;
7027 	ipaddr_t	src;
7028 	uint_t		hdr_length;
7029 	uint32_t	end;
7030 	mblk_t		*mp1;
7031 	mblk_t		*tail_mp;
7032 	size_t		count;
7033 	size_t		msg_len;
7034 	uint8_t		ecn_info = 0;
7035 	uint32_t	packet_size;
7036 	boolean_t	pruned = B_FALSE;
7037 	ill_t		*ill = ira->ira_ill;
7038 	ip_stack_t	*ipst = ill->ill_ipst;
7039 
7040 	/*
7041 	 * Drop the fragmented as early as possible, if
7042 	 * we don't have resource(s) to re-assemble.
7043 	 */
7044 	if (ipst->ips_ip_reass_queue_bytes == 0) {
7045 		freemsg(mp);
7046 		return (NULL);
7047 	}
7048 
7049 	/* Check for fragmentation offset; return if there's none */
7050 	if ((frag_offset_flags = ntohs(ipha->ipha_fragment_offset_and_flags) &
7051 	    (IPH_MF | IPH_OFFSET)) == 0)
7052 		return (mp);
7053 
7054 	/*
7055 	 * We utilize hardware computed checksum info only for UDP since
7056 	 * IP fragmentation is a normal occurrence for the protocol.  In
7057 	 * addition, checksum offload support for IP fragments carrying
7058 	 * UDP payload is commonly implemented across network adapters.
7059 	 */
7060 	ASSERT(ira->ira_rill != NULL);
7061 	if (proto == IPPROTO_UDP && dohwcksum &&
7062 	    ILL_HCKSUM_CAPABLE(ira->ira_rill) &&
7063 	    (DB_CKSUMFLAGS(mp) & (HCK_FULLCKSUM | HCK_PARTIALCKSUM))) {
7064 		mblk_t *mp1 = mp->b_cont;
7065 		int32_t len;
7066 
7067 		/* Record checksum information from the packet */
7068 		sum_val = (uint32_t)DB_CKSUM16(mp);
7069 		sum_flags = DB_CKSUMFLAGS(mp);
7070 
7071 		/* IP payload offset from beginning of mblk */
7072 		offset = ((uchar_t *)ipha + IPH_HDR_LENGTH(ipha)) - mp->b_rptr;
7073 
7074 		if ((sum_flags & HCK_PARTIALCKSUM) &&
7075 		    (mp1 == NULL || mp1->b_cont == NULL) &&
7076 		    offset >= DB_CKSUMSTART(mp) &&
7077 		    ((len = offset - DB_CKSUMSTART(mp)) & 1) == 0) {
7078 			uint32_t adj;
7079 			/*
7080 			 * Partial checksum has been calculated by hardware
7081 			 * and attached to the packet; in addition, any
7082 			 * prepended extraneous data is even byte aligned.
7083 			 * If any such data exists, we adjust the checksum;
7084 			 * this would also handle any postpended data.
7085 			 */
7086 			IP_ADJCKSUM_PARTIAL(mp->b_rptr + DB_CKSUMSTART(mp),
7087 			    mp, mp1, len, adj);
7088 
7089 			/* One's complement subtract extraneous checksum */
7090 			if (adj >= sum_val)
7091 				sum_val = ~(adj - sum_val) & 0xFFFF;
7092 			else
7093 				sum_val -= adj;
7094 		}
7095 	} else {
7096 		sum_val = 0;
7097 		sum_flags = 0;
7098 	}
7099 
7100 	/* Clear hardware checksumming flag */
7101 	DB_CKSUMFLAGS(mp) = 0;
7102 
7103 	ident = ipha->ipha_ident;
7104 	offset = (frag_offset_flags << 3) & 0xFFFF;
7105 	src = ipha->ipha_src;
7106 	dst = ipha->ipha_dst;
7107 	hdr_length = IPH_HDR_LENGTH(ipha);
7108 	end = ntohs(ipha->ipha_length) - hdr_length;
7109 
7110 	/* If end == 0 then we have a packet with no data, so just free it */
7111 	if (end == 0) {
7112 		freemsg(mp);
7113 		return (NULL);
7114 	}
7115 
7116 	/* Record the ECN field info. */
7117 	ecn_info = (ipha->ipha_type_of_service & 0x3);
7118 	if (offset != 0) {
7119 		/*
7120 		 * If this isn't the first piece, strip the header, and
7121 		 * add the offset to the end value.
7122 		 */
7123 		mp->b_rptr += hdr_length;
7124 		end += offset;
7125 	}
7126 
7127 	/* Handle vnic loopback of fragments */
7128 	if (mp->b_datap->db_ref > 2)
7129 		msg_len = 0;
7130 	else
7131 		msg_len = MBLKSIZE(mp);
7132 
7133 	tail_mp = mp;
7134 	while (tail_mp->b_cont != NULL) {
7135 		tail_mp = tail_mp->b_cont;
7136 		if (tail_mp->b_datap->db_ref <= 2)
7137 			msg_len += MBLKSIZE(tail_mp);
7138 	}
7139 
7140 	/* If the reassembly list for this ILL will get too big, prune it */
7141 	if ((msg_len + sizeof (*ipf) + ill->ill_frag_count) >=
7142 	    ipst->ips_ip_reass_queue_bytes) {
7143 		DTRACE_PROBE3(ip_reass_queue_bytes, uint_t, msg_len,
7144 		    uint_t, ill->ill_frag_count,
7145 		    uint_t, ipst->ips_ip_reass_queue_bytes);
7146 		ill_frag_prune(ill,
7147 		    (ipst->ips_ip_reass_queue_bytes < msg_len) ? 0 :
7148 		    (ipst->ips_ip_reass_queue_bytes - msg_len));
7149 		pruned = B_TRUE;
7150 	}
7151 
7152 	ipfb = &ill->ill_frag_hash_tbl[ILL_FRAG_HASH(src, ident)];
7153 	mutex_enter(&ipfb->ipfb_lock);
7154 
7155 	ipfp = &ipfb->ipfb_ipf;
7156 	/* Try to find an existing fragment queue for this packet. */
7157 	for (;;) {
7158 		ipf = ipfp[0];
7159 		if (ipf != NULL) {
7160 			/*
7161 			 * It has to match on ident and src/dst address.
7162 			 */
7163 			if (ipf->ipf_ident == ident &&
7164 			    ipf->ipf_src == src &&
7165 			    ipf->ipf_dst == dst &&
7166 			    ipf->ipf_protocol == proto) {
7167 				/*
7168 				 * If we have received too many
7169 				 * duplicate fragments for this packet
7170 				 * free it.
7171 				 */
7172 				if (ipf->ipf_num_dups > ip_max_frag_dups) {
7173 					ill_frag_free_pkts(ill, ipfb, ipf, 1);
7174 					freemsg(mp);
7175 					mutex_exit(&ipfb->ipfb_lock);
7176 					return (NULL);
7177 				}
7178 				/* Found it. */
7179 				break;
7180 			}
7181 			ipfp = &ipf->ipf_hash_next;
7182 			continue;
7183 		}
7184 
7185 		/*
7186 		 * If we pruned the list, do we want to store this new
7187 		 * fragment?. We apply an optimization here based on the
7188 		 * fact that most fragments will be received in order.
7189 		 * So if the offset of this incoming fragment is zero,
7190 		 * it is the first fragment of a new packet. We will
7191 		 * keep it.  Otherwise drop the fragment, as we have
7192 		 * probably pruned the packet already (since the
7193 		 * packet cannot be found).
7194 		 */
7195 		if (pruned && offset != 0) {
7196 			mutex_exit(&ipfb->ipfb_lock);
7197 			freemsg(mp);
7198 			return (NULL);
7199 		}
7200 
7201 		if (ipfb->ipfb_frag_pkts >= MAX_FRAG_PKTS(ipst))  {
7202 			/*
7203 			 * Too many fragmented packets in this hash
7204 			 * bucket. Free the oldest.
7205 			 */
7206 			ill_frag_free_pkts(ill, ipfb, ipfb->ipfb_ipf, 1);
7207 		}
7208 
7209 		/* New guy.  Allocate a frag message. */
7210 		mp1 = allocb(sizeof (*ipf), BPRI_MED);
7211 		if (mp1 == NULL) {
7212 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7213 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7214 			freemsg(mp);
7215 reass_done:
7216 			mutex_exit(&ipfb->ipfb_lock);
7217 			return (NULL);
7218 		}
7219 
7220 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmReqds);
7221 		mp1->b_cont = mp;
7222 
7223 		/* Initialize the fragment header. */
7224 		ipf = (ipf_t *)mp1->b_rptr;
7225 		ipf->ipf_mp = mp1;
7226 		ipf->ipf_ptphn = ipfp;
7227 		ipfp[0] = ipf;
7228 		ipf->ipf_hash_next = NULL;
7229 		ipf->ipf_ident = ident;
7230 		ipf->ipf_protocol = proto;
7231 		ipf->ipf_src = src;
7232 		ipf->ipf_dst = dst;
7233 		ipf->ipf_nf_hdr_len = 0;
7234 		/* Record reassembly start time. */
7235 		ipf->ipf_timestamp = gethrestime_sec();
7236 		/* Record ipf generation and account for frag header */
7237 		ipf->ipf_gen = ill->ill_ipf_gen++;
7238 		ipf->ipf_count = MBLKSIZE(mp1);
7239 		ipf->ipf_last_frag_seen = B_FALSE;
7240 		ipf->ipf_ecn = ecn_info;
7241 		ipf->ipf_num_dups = 0;
7242 		ipfb->ipfb_frag_pkts++;
7243 		ipf->ipf_checksum = 0;
7244 		ipf->ipf_checksum_flags = 0;
7245 
7246 		/* Store checksum value in fragment header */
7247 		if (sum_flags != 0) {
7248 			sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7249 			sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7250 			ipf->ipf_checksum = sum_val;
7251 			ipf->ipf_checksum_flags = sum_flags;
7252 		}
7253 
7254 		/*
7255 		 * We handle reassembly two ways.  In the easy case,
7256 		 * where all the fragments show up in order, we do
7257 		 * minimal bookkeeping, and just clip new pieces on
7258 		 * the end.  If we ever see a hole, then we go off
7259 		 * to ip_reassemble which has to mark the pieces and
7260 		 * keep track of the number of holes, etc.  Obviously,
7261 		 * the point of having both mechanisms is so we can
7262 		 * handle the easy case as efficiently as possible.
7263 		 */
7264 		if (offset == 0) {
7265 			/* Easy case, in-order reassembly so far. */
7266 			ipf->ipf_count += msg_len;
7267 			ipf->ipf_tail_mp = tail_mp;
7268 			/*
7269 			 * Keep track of next expected offset in
7270 			 * ipf_end.
7271 			 */
7272 			ipf->ipf_end = end;
7273 			ipf->ipf_nf_hdr_len = hdr_length;
7274 		} else {
7275 			/* Hard case, hole at the beginning. */
7276 			ipf->ipf_tail_mp = NULL;
7277 			/*
7278 			 * ipf_end == 0 means that we have given up
7279 			 * on easy reassembly.
7280 			 */
7281 			ipf->ipf_end = 0;
7282 
7283 			/* Forget checksum offload from now on */
7284 			ipf->ipf_checksum_flags = 0;
7285 
7286 			/*
7287 			 * ipf_hole_cnt is set by ip_reassemble.
7288 			 * ipf_count is updated by ip_reassemble.
7289 			 * No need to check for return value here
7290 			 * as we don't expect reassembly to complete
7291 			 * or fail for the first fragment itself.
7292 			 */
7293 			(void) ip_reassemble(mp, ipf,
7294 			    (frag_offset_flags & IPH_OFFSET) << 3,
7295 			    (frag_offset_flags & IPH_MF), ill, msg_len);
7296 		}
7297 		/* Update per ipfb and ill byte counts */
7298 		ipfb->ipfb_count += ipf->ipf_count;
7299 		ASSERT(ipfb->ipfb_count > 0);	/* Wraparound */
7300 		atomic_add_32(&ill->ill_frag_count, ipf->ipf_count);
7301 		/* If the frag timer wasn't already going, start it. */
7302 		mutex_enter(&ill->ill_lock);
7303 		ill_frag_timer_start(ill);
7304 		mutex_exit(&ill->ill_lock);
7305 		goto reass_done;
7306 	}
7307 
7308 	/*
7309 	 * If the packet's flag has changed (it could be coming up
7310 	 * from an interface different than the previous, therefore
7311 	 * possibly different checksum capability), then forget about
7312 	 * any stored checksum states.  Otherwise add the value to
7313 	 * the existing one stored in the fragment header.
7314 	 */
7315 	if (sum_flags != 0 && sum_flags == ipf->ipf_checksum_flags) {
7316 		sum_val += ipf->ipf_checksum;
7317 		sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7318 		sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7319 		ipf->ipf_checksum = sum_val;
7320 	} else if (ipf->ipf_checksum_flags != 0) {
7321 		/* Forget checksum offload from now on */
7322 		ipf->ipf_checksum_flags = 0;
7323 	}
7324 
7325 	/*
7326 	 * We have a new piece of a datagram which is already being
7327 	 * reassembled.  Update the ECN info if all IP fragments
7328 	 * are ECN capable.  If there is one which is not, clear
7329 	 * all the info.  If there is at least one which has CE
7330 	 * code point, IP needs to report that up to transport.
7331 	 */
7332 	if (ecn_info != IPH_ECN_NECT && ipf->ipf_ecn != IPH_ECN_NECT) {
7333 		if (ecn_info == IPH_ECN_CE)
7334 			ipf->ipf_ecn = IPH_ECN_CE;
7335 	} else {
7336 		ipf->ipf_ecn = IPH_ECN_NECT;
7337 	}
7338 	if (offset && ipf->ipf_end == offset) {
7339 		/* The new fragment fits at the end */
7340 		ipf->ipf_tail_mp->b_cont = mp;
7341 		/* Update the byte count */
7342 		ipf->ipf_count += msg_len;
7343 		/* Update per ipfb and ill byte counts */
7344 		ipfb->ipfb_count += msg_len;
7345 		ASSERT(ipfb->ipfb_count > 0);	/* Wraparound */
7346 		atomic_add_32(&ill->ill_frag_count, msg_len);
7347 		if (frag_offset_flags & IPH_MF) {
7348 			/* More to come. */
7349 			ipf->ipf_end = end;
7350 			ipf->ipf_tail_mp = tail_mp;
7351 			goto reass_done;
7352 		}
7353 	} else {
7354 		/* Go do the hard cases. */
7355 		int ret;
7356 
7357 		if (offset == 0)
7358 			ipf->ipf_nf_hdr_len = hdr_length;
7359 
7360 		/* Save current byte count */
7361 		count = ipf->ipf_count;
7362 		ret = ip_reassemble(mp, ipf,
7363 		    (frag_offset_flags & IPH_OFFSET) << 3,
7364 		    (frag_offset_flags & IPH_MF), ill, msg_len);
7365 		/* Count of bytes added and subtracted (freeb()ed) */
7366 		count = ipf->ipf_count - count;
7367 		if (count) {
7368 			/* Update per ipfb and ill byte counts */
7369 			ipfb->ipfb_count += count;
7370 			ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7371 			atomic_add_32(&ill->ill_frag_count, count);
7372 		}
7373 		if (ret == IP_REASS_PARTIAL) {
7374 			goto reass_done;
7375 		} else if (ret == IP_REASS_FAILED) {
7376 			/* Reassembly failed. Free up all resources */
7377 			ill_frag_free_pkts(ill, ipfb, ipf, 1);
7378 			for (t_mp = mp; t_mp != NULL; t_mp = t_mp->b_cont) {
7379 				IP_REASS_SET_START(t_mp, 0);
7380 				IP_REASS_SET_END(t_mp, 0);
7381 			}
7382 			freemsg(mp);
7383 			goto reass_done;
7384 		}
7385 		/* We will reach here iff 'ret' is IP_REASS_COMPLETE */
7386 	}
7387 	/*
7388 	 * We have completed reassembly.  Unhook the frag header from
7389 	 * the reassembly list.
7390 	 *
7391 	 * Before we free the frag header, record the ECN info
7392 	 * to report back to the transport.
7393 	 */
7394 	ecn_info = ipf->ipf_ecn;
7395 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmOKs);
7396 	ipfp = ipf->ipf_ptphn;
7397 
7398 	/* We need to supply these to caller */
7399 	if ((sum_flags = ipf->ipf_checksum_flags) != 0)
7400 		sum_val = ipf->ipf_checksum;
7401 	else
7402 		sum_val = 0;
7403 
7404 	mp1 = ipf->ipf_mp;
7405 	count = ipf->ipf_count;
7406 	ipf = ipf->ipf_hash_next;
7407 	if (ipf != NULL)
7408 		ipf->ipf_ptphn = ipfp;
7409 	ipfp[0] = ipf;
7410 	atomic_add_32(&ill->ill_frag_count, -count);
7411 	ASSERT(ipfb->ipfb_count >= count);
7412 	ipfb->ipfb_count -= count;
7413 	ipfb->ipfb_frag_pkts--;
7414 	mutex_exit(&ipfb->ipfb_lock);
7415 	/* Ditch the frag header. */
7416 	mp = mp1->b_cont;
7417 
7418 	freeb(mp1);
7419 
7420 	/* Restore original IP length in header. */
7421 	packet_size = (uint32_t)msgdsize(mp);
7422 	if (packet_size > IP_MAXPACKET) {
7423 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7424 		ip_drop_input("Reassembled packet too large", mp, ill);
7425 		freemsg(mp);
7426 		return (NULL);
7427 	}
7428 
7429 	if (DB_REF(mp) > 1) {
7430 		mblk_t *mp2 = copymsg(mp);
7431 
7432 		if (mp2 == NULL) {
7433 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7434 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7435 			freemsg(mp);
7436 			return (NULL);
7437 		}
7438 		freemsg(mp);
7439 		mp = mp2;
7440 	}
7441 	ipha = (ipha_t *)mp->b_rptr;
7442 
7443 	ipha->ipha_length = htons((uint16_t)packet_size);
7444 	/* We're now complete, zip the frag state */
7445 	ipha->ipha_fragment_offset_and_flags = 0;
7446 	/* Record the ECN info. */
7447 	ipha->ipha_type_of_service &= 0xFC;
7448 	ipha->ipha_type_of_service |= ecn_info;
7449 
7450 	/* Update the receive attributes */
7451 	ira->ira_pktlen = packet_size;
7452 	ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
7453 
7454 	/* Reassembly is successful; set checksum information in packet */
7455 	DB_CKSUM16(mp) = (uint16_t)sum_val;
7456 	DB_CKSUMFLAGS(mp) = sum_flags;
7457 	DB_CKSUMSTART(mp) = ira->ira_ip_hdr_length;
7458 
7459 	return (mp);
7460 }
7461 
7462 /*
7463  * Pullup function that should be used for IP input in order to
7464  * ensure we do not loose the L2 source address; we need the l2 source
7465  * address for IP_RECVSLLA and for ndp_input.
7466  *
7467  * We return either NULL or b_rptr.
7468  */
7469 void *
ip_pullup(mblk_t * mp,ssize_t len,ip_recv_attr_t * ira)7470 ip_pullup(mblk_t *mp, ssize_t len, ip_recv_attr_t *ira)
7471 {
7472 	ill_t		*ill = ira->ira_ill;
7473 
7474 	if (ip_rput_pullups++ == 0) {
7475 		(void) mi_strlog(ill->ill_rq, 1, SL_ERROR|SL_TRACE,
7476 		    "ip_pullup: %s forced us to "
7477 		    " pullup pkt, hdr len %ld, hdr addr %p",
7478 		    ill->ill_name, len, (void *)mp->b_rptr);
7479 	}
7480 	if (!(ira->ira_flags & IRAF_L2SRC_SET))
7481 		ip_setl2src(mp, ira, ira->ira_rill);
7482 	ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7483 	if (!pullupmsg(mp, len))
7484 		return (NULL);
7485 	else
7486 		return (mp->b_rptr);
7487 }
7488 
7489 /*
7490  * Make sure ira_l2src has an address. If we don't have one fill with zeros.
7491  * When called from the ULP ira_rill will be NULL hence the caller has to
7492  * pass in the ill.
7493  */
7494 /* ARGSUSED */
7495 void
ip_setl2src(mblk_t * mp,ip_recv_attr_t * ira,ill_t * ill)7496 ip_setl2src(mblk_t *mp, ip_recv_attr_t *ira, ill_t *ill)
7497 {
7498 	const uchar_t *addr;
7499 	int alen;
7500 
7501 	if (ira->ira_flags & IRAF_L2SRC_SET)
7502 		return;
7503 
7504 	ASSERT(ill != NULL);
7505 	alen = ill->ill_phys_addr_length;
7506 	ASSERT(alen <= sizeof (ira->ira_l2src));
7507 	if (ira->ira_mhip != NULL &&
7508 	    (addr = ira->ira_mhip->mhi_saddr) != NULL) {
7509 		bcopy(addr, ira->ira_l2src, alen);
7510 	} else if ((ira->ira_flags & IRAF_L2SRC_LOOPBACK) &&
7511 	    (addr = ill->ill_phys_addr) != NULL) {
7512 		bcopy(addr, ira->ira_l2src, alen);
7513 	} else {
7514 		bzero(ira->ira_l2src, alen);
7515 	}
7516 	ira->ira_flags |= IRAF_L2SRC_SET;
7517 }
7518 
7519 /*
7520  * check ip header length and align it.
7521  */
7522 mblk_t *
ip_check_and_align_header(mblk_t * mp,uint_t min_size,ip_recv_attr_t * ira)7523 ip_check_and_align_header(mblk_t *mp, uint_t min_size, ip_recv_attr_t *ira)
7524 {
7525 	ill_t	*ill = ira->ira_ill;
7526 	ssize_t len;
7527 
7528 	len = MBLKL(mp);
7529 
7530 	if (!OK_32PTR(mp->b_rptr))
7531 		IP_STAT(ill->ill_ipst, ip_notaligned);
7532 	else
7533 		IP_STAT(ill->ill_ipst, ip_recv_pullup);
7534 
7535 	/* Guard against bogus device drivers */
7536 	if (len < 0) {
7537 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7538 		ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7539 		freemsg(mp);
7540 		return (NULL);
7541 	}
7542 
7543 	if (len == 0) {
7544 		/* GLD sometimes sends up mblk with b_rptr == b_wptr! */
7545 		mblk_t *mp1 = mp->b_cont;
7546 
7547 		if (!(ira->ira_flags & IRAF_L2SRC_SET))
7548 			ip_setl2src(mp, ira, ira->ira_rill);
7549 		ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7550 
7551 		freeb(mp);
7552 		mp = mp1;
7553 		if (mp == NULL)
7554 			return (NULL);
7555 
7556 		if (OK_32PTR(mp->b_rptr) && MBLKL(mp) >= min_size)
7557 			return (mp);
7558 	}
7559 	if (ip_pullup(mp, min_size, ira) == NULL) {
7560 		if (msgdsize(mp) < min_size) {
7561 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7562 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7563 		} else {
7564 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7565 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7566 		}
7567 		freemsg(mp);
7568 		return (NULL);
7569 	}
7570 	return (mp);
7571 }
7572 
7573 /*
7574  * Common code for IPv4 and IPv6 to check and pullup multi-mblks
7575  */
7576 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)7577 ip_check_length(mblk_t *mp, uchar_t *rptr, ssize_t len,	uint_t pkt_len,
7578     uint_t min_size, ip_recv_attr_t *ira)
7579 {
7580 	ill_t	*ill = ira->ira_ill;
7581 
7582 	/*
7583 	 * Make sure we have data length consistent
7584 	 * with the IP header.
7585 	 */
7586 	if (mp->b_cont == NULL) {
7587 		/* pkt_len is based on ipha_len, not the mblk length */
7588 		if (pkt_len < min_size) {
7589 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7590 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7591 			freemsg(mp);
7592 			return (NULL);
7593 		}
7594 		if (len < 0) {
7595 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7596 			ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7597 			freemsg(mp);
7598 			return (NULL);
7599 		}
7600 		/* Drop any pad */
7601 		mp->b_wptr = rptr + pkt_len;
7602 	} else if ((len += msgdsize(mp->b_cont)) != 0) {
7603 		ASSERT(pkt_len >= min_size);
7604 		if (pkt_len < min_size) {
7605 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7606 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7607 			freemsg(mp);
7608 			return (NULL);
7609 		}
7610 		if (len < 0) {
7611 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7612 			ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7613 			freemsg(mp);
7614 			return (NULL);
7615 		}
7616 		/* Drop any pad */
7617 		(void) adjmsg(mp, -len);
7618 		/*
7619 		 * adjmsg may have freed an mblk from the chain, hence
7620 		 * invalidate any hw checksum here. This will force IP to
7621 		 * calculate the checksum in sw, but only for this packet.
7622 		 */
7623 		DB_CKSUMFLAGS(mp) = 0;
7624 		IP_STAT(ill->ill_ipst, ip_multimblk);
7625 	}
7626 	return (mp);
7627 }
7628 
7629 /*
7630  * Check that the IPv4 opt_len is consistent with the packet and pullup
7631  * the options.
7632  */
7633 mblk_t *
ip_check_optlen(mblk_t * mp,ipha_t * ipha,uint_t opt_len,uint_t pkt_len,ip_recv_attr_t * ira)7634 ip_check_optlen(mblk_t *mp, ipha_t *ipha, uint_t opt_len, uint_t pkt_len,
7635     ip_recv_attr_t *ira)
7636 {
7637 	ill_t	*ill = ira->ira_ill;
7638 	ssize_t len;
7639 
7640 	/* Assume no IPv6 packets arrive over the IPv4 queue */
7641 	if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
7642 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7643 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInWrongIPVersion);
7644 		ip_drop_input("IPvN packet on IPv4 ill", mp, ill);
7645 		freemsg(mp);
7646 		return (NULL);
7647 	}
7648 
7649 	if (opt_len > (15 - IP_SIMPLE_HDR_LENGTH_IN_WORDS)) {
7650 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7651 		ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7652 		freemsg(mp);
7653 		return (NULL);
7654 	}
7655 	/*
7656 	 * Recompute complete header length and make sure we
7657 	 * have access to all of it.
7658 	 */
7659 	len = ((size_t)opt_len + IP_SIMPLE_HDR_LENGTH_IN_WORDS) << 2;
7660 	if (len > (mp->b_wptr - mp->b_rptr)) {
7661 		if (len > pkt_len) {
7662 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7663 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7664 			freemsg(mp);
7665 			return (NULL);
7666 		}
7667 		if (ip_pullup(mp, len, ira) == NULL) {
7668 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7669 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7670 			freemsg(mp);
7671 			return (NULL);
7672 		}
7673 	}
7674 	return (mp);
7675 }
7676 
7677 /*
7678  * Returns a new ire, or the same ire, or NULL.
7679  * If a different IRE is returned, then it is held; the caller
7680  * needs to release it.
7681  * In no case is there any hold/release on the ire argument.
7682  */
7683 ire_t *
ip_check_multihome(void * addr,ire_t * ire,ill_t * ill)7684 ip_check_multihome(void *addr, ire_t *ire, ill_t *ill)
7685 {
7686 	ire_t		*new_ire;
7687 	ill_t		*ire_ill;
7688 	uint_t		ifindex;
7689 	ip_stack_t	*ipst = ill->ill_ipst;
7690 	boolean_t	strict_check = B_FALSE;
7691 
7692 	/*
7693 	 * IPMP common case: if IRE and ILL are in the same group, there's no
7694 	 * issue (e.g. packet received on an underlying interface matched an
7695 	 * IRE_LOCAL on its associated group interface).
7696 	 */
7697 	ASSERT(ire->ire_ill != NULL);
7698 	if (IS_IN_SAME_ILLGRP(ill, ire->ire_ill))
7699 		return (ire);
7700 
7701 	/*
7702 	 * Do another ire lookup here, using the ingress ill, to see if the
7703 	 * interface is in a usesrc group.
7704 	 * As long as the ills belong to the same group, we don't consider
7705 	 * them to be arriving on the wrong interface. Thus, if the switch
7706 	 * is doing inbound load spreading, we won't drop packets when the
7707 	 * ip*_strict_dst_multihoming switch is on.
7708 	 * We also need to check for IPIF_UNNUMBERED point2point interfaces
7709 	 * where the local address may not be unique. In this case we were
7710 	 * at the mercy of the initial ire lookup and the IRE_LOCAL it
7711 	 * actually returned. The new lookup, which is more specific, should
7712 	 * only find the IRE_LOCAL associated with the ingress ill if one
7713 	 * exists.
7714 	 */
7715 	if (ire->ire_ipversion == IPV4_VERSION) {
7716 		if (ipst->ips_ip_strict_dst_multihoming)
7717 			strict_check = B_TRUE;
7718 		new_ire = ire_ftable_lookup_v4(*((ipaddr_t *)addr), 0, 0,
7719 		    IRE_LOCAL, ill, ALL_ZONES, NULL,
7720 		    (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7721 	} else {
7722 		ASSERT(!IN6_IS_ADDR_MULTICAST((in6_addr_t *)addr));
7723 		if (ipst->ips_ipv6_strict_dst_multihoming)
7724 			strict_check = B_TRUE;
7725 		new_ire = ire_ftable_lookup_v6((in6_addr_t *)addr, NULL, NULL,
7726 		    IRE_LOCAL, ill, ALL_ZONES, NULL,
7727 		    (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7728 	}
7729 	/*
7730 	 * If the same ire that was returned in ip_input() is found then this
7731 	 * is an indication that usesrc groups are in use. The packet
7732 	 * arrived on a different ill in the group than the one associated with
7733 	 * the destination address.  If a different ire was found then the same
7734 	 * IP address must be hosted on multiple ills. This is possible with
7735 	 * unnumbered point2point interfaces. We switch to use this new ire in
7736 	 * order to have accurate interface statistics.
7737 	 */
7738 	if (new_ire != NULL) {
7739 		/* Note: held in one case but not the other? Caller handles */
7740 		if (new_ire != ire)
7741 			return (new_ire);
7742 		/* Unchanged */
7743 		ire_refrele(new_ire);
7744 		return (ire);
7745 	}
7746 
7747 	/*
7748 	 * Chase pointers once and store locally.
7749 	 */
7750 	ASSERT(ire->ire_ill != NULL);
7751 	ire_ill = ire->ire_ill;
7752 	ifindex = ill->ill_usesrc_ifindex;
7753 
7754 	/*
7755 	 * Check if it's a legal address on the 'usesrc' interface.
7756 	 * For IPMP data addresses the IRE_LOCAL is the upper, hence we
7757 	 * can just check phyint_ifindex.
7758 	 */
7759 	if (ifindex != 0 && ifindex == ire_ill->ill_phyint->phyint_ifindex) {
7760 		return (ire);
7761 	}
7762 
7763 	/*
7764 	 * If the ip*_strict_dst_multihoming switch is on then we can
7765 	 * only accept this packet if the interface is marked as routing.
7766 	 */
7767 	if (!(strict_check))
7768 		return (ire);
7769 
7770 	if ((ill->ill_flags & ire->ire_ill->ill_flags & ILLF_ROUTER) != 0) {
7771 		return (ire);
7772 	}
7773 	return (NULL);
7774 }
7775 
7776 /*
7777  * This function is used to construct a mac_header_info_s from a
7778  * DL_UNITDATA_IND message.
7779  * The address fields in the mhi structure points into the message,
7780  * thus the caller can't use those fields after freeing the message.
7781  *
7782  * We determine whether the packet received is a non-unicast packet
7783  * and in doing so, determine whether or not it is broadcast vs multicast.
7784  * For it to be a broadcast packet, we must have the appropriate mblk_t
7785  * hanging off the ill_t.  If this is either not present or doesn't match
7786  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7787  * to be multicast.  Thus NICs that have no broadcast address (or no
7788  * capability for one, such as point to point links) cannot return as
7789  * the packet being broadcast.
7790  */
7791 void
ip_dlur_to_mhi(ill_t * ill,mblk_t * mb,struct mac_header_info_s * mhip)7792 ip_dlur_to_mhi(ill_t *ill, mblk_t *mb, struct mac_header_info_s *mhip)
7793 {
7794 	dl_unitdata_ind_t *ind = (dl_unitdata_ind_t *)mb->b_rptr;
7795 	mblk_t *bmp;
7796 	uint_t extra_offset;
7797 
7798 	bzero(mhip, sizeof (struct mac_header_info_s));
7799 
7800 	mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7801 
7802 	if (ill->ill_sap_length < 0)
7803 		extra_offset = 0;
7804 	else
7805 		extra_offset = ill->ill_sap_length;
7806 
7807 	mhip->mhi_daddr = (uchar_t *)ind + ind->dl_dest_addr_offset +
7808 	    extra_offset;
7809 	mhip->mhi_saddr = (uchar_t *)ind + ind->dl_src_addr_offset +
7810 	    extra_offset;
7811 
7812 	if (!ind->dl_group_address)
7813 		return;
7814 
7815 	/* Multicast or broadcast */
7816 	mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7817 
7818 	if (ind->dl_dest_addr_offset > sizeof (*ind) &&
7819 	    ind->dl_dest_addr_offset + ind->dl_dest_addr_length < MBLKL(mb) &&
7820 	    (bmp = ill->ill_bcast_mp) != NULL) {
7821 		dl_unitdata_req_t *dlur;
7822 		uint8_t *bphys_addr;
7823 
7824 		dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7825 		bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
7826 		    extra_offset;
7827 
7828 		if (bcmp(mhip->mhi_daddr, bphys_addr,
7829 		    ind->dl_dest_addr_length) == 0)
7830 			mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7831 	}
7832 }
7833 
7834 /*
7835  * This function is used to construct a mac_header_info_s from a
7836  * M_DATA fastpath message from a DLPI driver.
7837  * The address fields in the mhi structure points into the message,
7838  * thus the caller can't use those fields after freeing the message.
7839  *
7840  * We determine whether the packet received is a non-unicast packet
7841  * and in doing so, determine whether or not it is broadcast vs multicast.
7842  * For it to be a broadcast packet, we must have the appropriate mblk_t
7843  * hanging off the ill_t.  If this is either not present or doesn't match
7844  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7845  * to be multicast.  Thus NICs that have no broadcast address (or no
7846  * capability for one, such as point to point links) cannot return as
7847  * the packet being broadcast.
7848  */
7849 void
ip_mdata_to_mhi(ill_t * ill,mblk_t * mp,struct mac_header_info_s * mhip)7850 ip_mdata_to_mhi(ill_t *ill, mblk_t *mp, struct mac_header_info_s *mhip)
7851 {
7852 	mblk_t *bmp;
7853 	struct ether_header *pether;
7854 
7855 	bzero(mhip, sizeof (struct mac_header_info_s));
7856 
7857 	mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7858 
7859 	pether = (struct ether_header *)((char *)mp->b_rptr
7860 	    - sizeof (struct ether_header));
7861 
7862 	/*
7863 	 * Make sure the interface is an ethernet type, since we don't
7864 	 * know the header format for anything but Ethernet. Also make
7865 	 * sure we are pointing correctly above db_base.
7866 	 */
7867 	if (ill->ill_type != IFT_ETHER)
7868 		return;
7869 
7870 retry:
7871 	if ((uchar_t *)pether < mp->b_datap->db_base)
7872 		return;
7873 
7874 	/* Is there a VLAN tag? */
7875 	if (ill->ill_isv6) {
7876 		if (pether->ether_type != htons(ETHERTYPE_IPV6)) {
7877 			pether = (struct ether_header *)((char *)pether - 4);
7878 			goto retry;
7879 		}
7880 	} else {
7881 		if (pether->ether_type != htons(ETHERTYPE_IP)) {
7882 			pether = (struct ether_header *)((char *)pether - 4);
7883 			goto retry;
7884 		}
7885 	}
7886 	mhip->mhi_daddr = (uchar_t *)&pether->ether_dhost;
7887 	mhip->mhi_saddr = (uchar_t *)&pether->ether_shost;
7888 
7889 	if (!(mhip->mhi_daddr[0] & 0x01))
7890 		return;
7891 
7892 	/* Multicast or broadcast */
7893 	mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7894 
7895 	if ((bmp = ill->ill_bcast_mp) != NULL) {
7896 		dl_unitdata_req_t *dlur;
7897 		uint8_t *bphys_addr;
7898 		uint_t	addrlen;
7899 
7900 		dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7901 		addrlen = dlur->dl_dest_addr_length;
7902 		if (ill->ill_sap_length < 0) {
7903 			bphys_addr = (uchar_t *)dlur +
7904 			    dlur->dl_dest_addr_offset;
7905 			addrlen += ill->ill_sap_length;
7906 		} else {
7907 			bphys_addr = (uchar_t *)dlur +
7908 			    dlur->dl_dest_addr_offset +
7909 			    ill->ill_sap_length;
7910 			addrlen -= ill->ill_sap_length;
7911 		}
7912 		if (bcmp(mhip->mhi_daddr, bphys_addr, addrlen) == 0)
7913 			mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7914 	}
7915 }
7916 
7917 /*
7918  * Handle anything but M_DATA messages
7919  * We see the DL_UNITDATA_IND which are part
7920  * of the data path, and also the other messages from the driver.
7921  */
7922 void
ip_rput_notdata(ill_t * ill,mblk_t * mp)7923 ip_rput_notdata(ill_t *ill, mblk_t *mp)
7924 {
7925 	mblk_t		*first_mp;
7926 	struct iocblk   *iocp;
7927 	struct mac_header_info_s mhi;
7928 
7929 	switch (DB_TYPE(mp)) {
7930 	case M_PROTO:
7931 	case M_PCPROTO: {
7932 		if (((dl_unitdata_ind_t *)mp->b_rptr)->dl_primitive !=
7933 		    DL_UNITDATA_IND) {
7934 			/* Go handle anything other than data elsewhere. */
7935 			ip_rput_dlpi(ill, mp);
7936 			return;
7937 		}
7938 
7939 		first_mp = mp;
7940 		mp = first_mp->b_cont;
7941 		first_mp->b_cont = NULL;
7942 
7943 		if (mp == NULL) {
7944 			freeb(first_mp);
7945 			return;
7946 		}
7947 		ip_dlur_to_mhi(ill, first_mp, &mhi);
7948 		if (ill->ill_isv6)
7949 			ip_input_v6(ill, NULL, mp, &mhi);
7950 		else
7951 			ip_input(ill, NULL, mp, &mhi);
7952 
7953 		/* Ditch the DLPI header. */
7954 		freeb(first_mp);
7955 		return;
7956 	}
7957 	case M_IOCACK:
7958 		iocp = (struct iocblk *)mp->b_rptr;
7959 		switch (iocp->ioc_cmd) {
7960 		case DL_IOC_HDR_INFO:
7961 			ill_fastpath_ack(ill, mp);
7962 			return;
7963 		default:
7964 			putnext(ill->ill_rq, mp);
7965 			return;
7966 		}
7967 		/* FALLTHRU */
7968 	case M_ERROR:
7969 	case M_HANGUP:
7970 		mutex_enter(&ill->ill_lock);
7971 		if (ill->ill_state_flags & ILL_CONDEMNED) {
7972 			mutex_exit(&ill->ill_lock);
7973 			freemsg(mp);
7974 			return;
7975 		}
7976 		ill_refhold_locked(ill);
7977 		mutex_exit(&ill->ill_lock);
7978 		qwriter_ip(ill, ill->ill_rq, mp, ip_rput_other, CUR_OP,
7979 		    B_FALSE);
7980 		return;
7981 	case M_CTL:
7982 		putnext(ill->ill_rq, mp);
7983 		return;
7984 	case M_IOCNAK:
7985 		ip1dbg(("got iocnak "));
7986 		iocp = (struct iocblk *)mp->b_rptr;
7987 		switch (iocp->ioc_cmd) {
7988 		case DL_IOC_HDR_INFO:
7989 			ip_rput_other(NULL, ill->ill_rq, mp, NULL);
7990 			return;
7991 		default:
7992 			break;
7993 		}
7994 		/* FALLTHRU */
7995 	default:
7996 		putnext(ill->ill_rq, mp);
7997 		return;
7998 	}
7999 }
8000 
8001 /* Read side put procedure.  Packets coming from the wire arrive here. */
8002 void
ip_rput(queue_t * q,mblk_t * mp)8003 ip_rput(queue_t *q, mblk_t *mp)
8004 {
8005 	ill_t	*ill;
8006 	union DL_primitives *dl;
8007 
8008 	ill = (ill_t *)q->q_ptr;
8009 
8010 	if (ill->ill_state_flags & (ILL_CONDEMNED | ILL_LL_SUBNET_PENDING)) {
8011 		/*
8012 		 * If things are opening or closing, only accept high-priority
8013 		 * DLPI messages.  (On open ill->ill_ipif has not yet been
8014 		 * created; on close, things hanging off the ill may have been
8015 		 * freed already.)
8016 		 */
8017 		dl = (union DL_primitives *)mp->b_rptr;
8018 		if (DB_TYPE(mp) != M_PCPROTO ||
8019 		    dl->dl_primitive == DL_UNITDATA_IND) {
8020 			inet_freemsg(mp);
8021 			return;
8022 		}
8023 	}
8024 	if (DB_TYPE(mp) == M_DATA) {
8025 		struct mac_header_info_s mhi;
8026 
8027 		ip_mdata_to_mhi(ill, mp, &mhi);
8028 		ip_input(ill, NULL, mp, &mhi);
8029 	} else {
8030 		ip_rput_notdata(ill, mp);
8031 	}
8032 }
8033 
8034 /*
8035  * Move the information to a copy.
8036  */
8037 mblk_t *
ip_fix_dbref(mblk_t * mp,ip_recv_attr_t * ira)8038 ip_fix_dbref(mblk_t *mp, ip_recv_attr_t *ira)
8039 {
8040 	mblk_t		*mp1;
8041 	ill_t		*ill = ira->ira_ill;
8042 	ip_stack_t	*ipst = ill->ill_ipst;
8043 
8044 	IP_STAT(ipst, ip_db_ref);
8045 
8046 	/* Make sure we have ira_l2src before we loose the original mblk */
8047 	if (!(ira->ira_flags & IRAF_L2SRC_SET))
8048 		ip_setl2src(mp, ira, ira->ira_rill);
8049 
8050 	mp1 = copymsg(mp);
8051 	if (mp1 == NULL) {
8052 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
8053 		ip_drop_input("ipIfStatsInDiscards", mp, ill);
8054 		freemsg(mp);
8055 		return (NULL);
8056 	}
8057 	/* preserve the hardware checksum flags and data, if present */
8058 	if (DB_CKSUMFLAGS(mp) != 0) {
8059 		DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
8060 		DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
8061 		DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
8062 		DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
8063 		DB_CKSUM16(mp1) = DB_CKSUM16(mp);
8064 	}
8065 	freemsg(mp);
8066 	return (mp1);
8067 }
8068 
8069 static void
ip_dlpi_error(ill_t * ill,t_uscalar_t prim,t_uscalar_t dl_err,t_uscalar_t err)8070 ip_dlpi_error(ill_t *ill, t_uscalar_t prim, t_uscalar_t dl_err,
8071     t_uscalar_t err)
8072 {
8073 	if (dl_err == DL_SYSERR) {
8074 		(void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8075 		    "%s: %s failed: DL_SYSERR (errno %u)\n",
8076 		    ill->ill_name, dl_primstr(prim), err);
8077 		return;
8078 	}
8079 
8080 	(void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8081 	    "%s: %s failed: %s\n", ill->ill_name, dl_primstr(prim),
8082 	    dl_errstr(dl_err));
8083 }
8084 
8085 /*
8086  * ip_rput_dlpi is called by ip_rput to handle all DLPI messages other
8087  * than DL_UNITDATA_IND messages. If we need to process this message
8088  * exclusively, we call qwriter_ip, in which case we also need to call
8089  * ill_refhold before that, since qwriter_ip does an ill_refrele.
8090  */
8091 void
ip_rput_dlpi(ill_t * ill,mblk_t * mp)8092 ip_rput_dlpi(ill_t *ill, mblk_t *mp)
8093 {
8094 	dl_ok_ack_t	*dloa = (dl_ok_ack_t *)mp->b_rptr;
8095 	dl_error_ack_t	*dlea = (dl_error_ack_t *)dloa;
8096 	queue_t		*q = ill->ill_rq;
8097 	t_uscalar_t	prim = dloa->dl_primitive;
8098 	t_uscalar_t	reqprim = DL_PRIM_INVAL;
8099 
8100 	DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi",
8101 	    char *, dl_primstr(prim), ill_t *, ill);
8102 	ip1dbg(("ip_rput_dlpi"));
8103 
8104 	/*
8105 	 * If we received an ACK but didn't send a request for it, then it
8106 	 * can't be part of any pending operation; discard up-front.
8107 	 */
8108 	switch (prim) {
8109 	case DL_ERROR_ACK:
8110 		reqprim = dlea->dl_error_primitive;
8111 		ip2dbg(("ip_rput_dlpi(%s): DL_ERROR_ACK for %s (0x%x): %s "
8112 		    "(0x%x), unix %u\n", ill->ill_name, dl_primstr(reqprim),
8113 		    reqprim, dl_errstr(dlea->dl_errno), dlea->dl_errno,
8114 		    dlea->dl_unix_errno));
8115 		break;
8116 	case DL_OK_ACK:
8117 		reqprim = dloa->dl_correct_primitive;
8118 		break;
8119 	case DL_INFO_ACK:
8120 		reqprim = DL_INFO_REQ;
8121 		break;
8122 	case DL_BIND_ACK:
8123 		reqprim = DL_BIND_REQ;
8124 		break;
8125 	case DL_PHYS_ADDR_ACK:
8126 		reqprim = DL_PHYS_ADDR_REQ;
8127 		break;
8128 	case DL_NOTIFY_ACK:
8129 		reqprim = DL_NOTIFY_REQ;
8130 		break;
8131 	case DL_CAPABILITY_ACK:
8132 		reqprim = DL_CAPABILITY_REQ;
8133 		break;
8134 	}
8135 
8136 	if (prim != DL_NOTIFY_IND) {
8137 		if (reqprim == DL_PRIM_INVAL ||
8138 		    !ill_dlpi_pending(ill, reqprim)) {
8139 			/* Not a DLPI message we support or expected */
8140 			freemsg(mp);
8141 			return;
8142 		}
8143 		ip1dbg(("ip_rput: received %s for %s\n", dl_primstr(prim),
8144 		    dl_primstr(reqprim)));
8145 	}
8146 
8147 	switch (reqprim) {
8148 	case DL_UNBIND_REQ:
8149 		/*
8150 		 * NOTE: we mark the unbind as complete even if we got a
8151 		 * DL_ERROR_ACK, since there's not much else we can do.
8152 		 */
8153 		mutex_enter(&ill->ill_lock);
8154 		ill->ill_state_flags &= ~ILL_DL_UNBIND_IN_PROGRESS;
8155 		cv_signal(&ill->ill_cv);
8156 		mutex_exit(&ill->ill_lock);
8157 		break;
8158 
8159 	case DL_ENABMULTI_REQ:
8160 		if (prim == DL_OK_ACK) {
8161 			if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8162 				ill->ill_dlpi_multicast_state = IDS_OK;
8163 		}
8164 		break;
8165 	}
8166 
8167 	/*
8168 	 * The message is one we're waiting for (or DL_NOTIFY_IND), but we
8169 	 * need to become writer to continue to process it.  Because an
8170 	 * exclusive operation doesn't complete until replies to all queued
8171 	 * DLPI messages have been received, we know we're in the middle of an
8172 	 * exclusive operation and pass CUR_OP (except for DL_NOTIFY_IND).
8173 	 *
8174 	 * As required by qwriter_ip(), we refhold the ill; it will refrele.
8175 	 * Since this is on the ill stream we unconditionally bump up the
8176 	 * refcount without doing ILL_CAN_LOOKUP().
8177 	 */
8178 	ill_refhold(ill);
8179 	if (prim == DL_NOTIFY_IND)
8180 		qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, NEW_OP, B_FALSE);
8181 	else
8182 		qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, CUR_OP, B_FALSE);
8183 }
8184 
8185 /*
8186  * Handling of DLPI messages that require exclusive access to the ipsq.
8187  *
8188  * Need to do ipsq_pending_mp_get on ioctl completion, which could
8189  * happen here. (along with mi_copy_done)
8190  */
8191 /* ARGSUSED */
8192 static void
ip_rput_dlpi_writer(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * dummy_arg)8193 ip_rput_dlpi_writer(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8194 {
8195 	dl_ok_ack_t	*dloa = (dl_ok_ack_t *)mp->b_rptr;
8196 	dl_error_ack_t	*dlea = (dl_error_ack_t *)dloa;
8197 	int		err = 0;
8198 	ill_t		*ill = (ill_t *)q->q_ptr;
8199 	ipif_t		*ipif = NULL;
8200 	mblk_t		*mp1 = NULL;
8201 	conn_t		*connp = NULL;
8202 	t_uscalar_t	paddrreq;
8203 	mblk_t		*mp_hw;
8204 	boolean_t	success;
8205 	boolean_t	ioctl_aborted = B_FALSE;
8206 	boolean_t	log = B_TRUE;
8207 
8208 	DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer",
8209 	    char *, dl_primstr(dloa->dl_primitive), ill_t *, ill);
8210 
8211 	ip1dbg(("ip_rput_dlpi_writer .."));
8212 	ASSERT(ipsq->ipsq_xop == ill->ill_phyint->phyint_ipsq->ipsq_xop);
8213 	ASSERT(IAM_WRITER_ILL(ill));
8214 
8215 	ipif = ipsq->ipsq_xop->ipx_pending_ipif;
8216 	/*
8217 	 * The current ioctl could have been aborted by the user and a new
8218 	 * ioctl to bring up another ill could have started. We could still
8219 	 * get a response from the driver later.
8220 	 */
8221 	if (ipif != NULL && ipif->ipif_ill != ill)
8222 		ioctl_aborted = B_TRUE;
8223 
8224 	switch (dloa->dl_primitive) {
8225 	case DL_ERROR_ACK:
8226 		ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for %s\n",
8227 		    dl_primstr(dlea->dl_error_primitive)));
8228 
8229 		DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer error",
8230 		    char *, dl_primstr(dlea->dl_error_primitive),
8231 		    ill_t *, ill);
8232 
8233 		switch (dlea->dl_error_primitive) {
8234 		case DL_DISABMULTI_REQ:
8235 			ill_dlpi_done(ill, dlea->dl_error_primitive);
8236 			break;
8237 		case DL_PROMISCON_REQ:
8238 		case DL_PROMISCOFF_REQ:
8239 		case DL_UNBIND_REQ:
8240 		case DL_ATTACH_REQ:
8241 		case DL_INFO_REQ:
8242 			ill_dlpi_done(ill, dlea->dl_error_primitive);
8243 			break;
8244 		case DL_NOTIFY_REQ:
8245 			ill_dlpi_done(ill, DL_NOTIFY_REQ);
8246 			log = B_FALSE;
8247 			break;
8248 		case DL_PHYS_ADDR_REQ:
8249 			/*
8250 			 * For IPv6 only, there are two additional
8251 			 * phys_addr_req's sent to the driver to get the
8252 			 * IPv6 token and lla. This allows IP to acquire
8253 			 * the hardware address format for a given interface
8254 			 * without having built in knowledge of the hardware
8255 			 * address. ill_phys_addr_pend keeps track of the last
8256 			 * DL_PAR sent so we know which response we are
8257 			 * dealing with. ill_dlpi_done will update
8258 			 * ill_phys_addr_pend when it sends the next req.
8259 			 * We don't complete the IOCTL until all three DL_PARs
8260 			 * have been attempted, so set *_len to 0 and break.
8261 			 */
8262 			paddrreq = ill->ill_phys_addr_pend;
8263 			ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8264 			if (paddrreq == DL_IPV6_TOKEN) {
8265 				ill->ill_token_length = 0;
8266 				log = B_FALSE;
8267 				break;
8268 			} else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8269 				ill->ill_nd_lla_len = 0;
8270 				log = B_FALSE;
8271 				break;
8272 			}
8273 			/*
8274 			 * Something went wrong with the DL_PHYS_ADDR_REQ.
8275 			 * We presumably have an IOCTL hanging out waiting
8276 			 * for completion. Find it and complete the IOCTL
8277 			 * with the error noted.
8278 			 * However, ill_dl_phys was called on an ill queue
8279 			 * (from SIOCSLIFNAME), thus conn_pending_ill is not
8280 			 * set. But the ioctl is known to be pending on ill_wq.
8281 			 */
8282 			if (!ill->ill_ifname_pending)
8283 				break;
8284 			ill->ill_ifname_pending = 0;
8285 			if (!ioctl_aborted)
8286 				mp1 = ipsq_pending_mp_get(ipsq, &connp);
8287 			if (mp1 != NULL) {
8288 				/*
8289 				 * This operation (SIOCSLIFNAME) must have
8290 				 * happened on the ill. Assert there is no conn
8291 				 */
8292 				ASSERT(connp == NULL);
8293 				q = ill->ill_wq;
8294 			}
8295 			break;
8296 		case DL_BIND_REQ:
8297 			ill_dlpi_done(ill, DL_BIND_REQ);
8298 			if (ill->ill_ifname_pending)
8299 				break;
8300 			mutex_enter(&ill->ill_lock);
8301 			ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8302 			mutex_exit(&ill->ill_lock);
8303 			/*
8304 			 * Something went wrong with the bind.  We presumably
8305 			 * have an IOCTL hanging out waiting for completion.
8306 			 * Find it, take down the interface that was coming
8307 			 * up, and complete the IOCTL with the error noted.
8308 			 */
8309 			if (!ioctl_aborted)
8310 				mp1 = ipsq_pending_mp_get(ipsq, &connp);
8311 			if (mp1 != NULL) {
8312 				/*
8313 				 * This might be a result of a DL_NOTE_REPLUMB
8314 				 * notification. In that case, connp is NULL.
8315 				 */
8316 				if (connp != NULL)
8317 					q = CONNP_TO_WQ(connp);
8318 
8319 				(void) ipif_down(ipif, NULL, NULL);
8320 				/* error is set below the switch */
8321 			}
8322 			break;
8323 		case DL_ENABMULTI_REQ:
8324 			ill_dlpi_done(ill, DL_ENABMULTI_REQ);
8325 
8326 			if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8327 				ill->ill_dlpi_multicast_state = IDS_FAILED;
8328 			if (ill->ill_dlpi_multicast_state == IDS_FAILED) {
8329 
8330 				printf("ip: joining multicasts failed (%d)"
8331 				    " on %s - will use link layer "
8332 				    "broadcasts for multicast\n",
8333 				    dlea->dl_errno, ill->ill_name);
8334 
8335 				/*
8336 				 * Set up for multi_bcast; We are the
8337 				 * writer, so ok to access ill->ill_ipif
8338 				 * without any lock.
8339 				 */
8340 				mutex_enter(&ill->ill_phyint->phyint_lock);
8341 				ill->ill_phyint->phyint_flags |=
8342 				    PHYI_MULTI_BCAST;
8343 				mutex_exit(&ill->ill_phyint->phyint_lock);
8344 
8345 			}
8346 			freemsg(mp);	/* Don't want to pass this up */
8347 			return;
8348 		case DL_CAPABILITY_REQ:
8349 			ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for "
8350 			    "DL_CAPABILITY REQ\n"));
8351 			if (ill->ill_dlpi_capab_state == IDCS_PROBE_SENT)
8352 				ill->ill_dlpi_capab_state = IDCS_FAILED;
8353 			ill_capability_done(ill);
8354 			freemsg(mp);
8355 			return;
8356 		}
8357 		/*
8358 		 * Note the error for IOCTL completion (mp1 is set when
8359 		 * ready to complete ioctl). If ill_ifname_pending_err is
8360 		 * set, an error occured during plumbing (ill_ifname_pending),
8361 		 * so we want to report that error.
8362 		 *
8363 		 * NOTE: there are two addtional DL_PHYS_ADDR_REQ's
8364 		 * (DL_IPV6_TOKEN and DL_IPV6_LINK_LAYER_ADDR) that are
8365 		 * expected to get errack'd if the driver doesn't support
8366 		 * these flags (e.g. ethernet). log will be set to B_FALSE
8367 		 * if these error conditions are encountered.
8368 		 */
8369 		if (mp1 != NULL) {
8370 			if (ill->ill_ifname_pending_err != 0)  {
8371 				err = ill->ill_ifname_pending_err;
8372 				ill->ill_ifname_pending_err = 0;
8373 			} else {
8374 				err = dlea->dl_unix_errno ?
8375 				    dlea->dl_unix_errno : ENXIO;
8376 			}
8377 		/*
8378 		 * If we're plumbing an interface and an error hasn't already
8379 		 * been saved, set ill_ifname_pending_err to the error passed
8380 		 * up. Ignore the error if log is B_FALSE (see comment above).
8381 		 */
8382 		} else if (log && ill->ill_ifname_pending &&
8383 		    ill->ill_ifname_pending_err == 0) {
8384 			ill->ill_ifname_pending_err = dlea->dl_unix_errno ?
8385 			    dlea->dl_unix_errno : ENXIO;
8386 		}
8387 
8388 		if (log)
8389 			ip_dlpi_error(ill, dlea->dl_error_primitive,
8390 			    dlea->dl_errno, dlea->dl_unix_errno);
8391 		break;
8392 	case DL_CAPABILITY_ACK:
8393 		ill_capability_ack(ill, mp);
8394 		/*
8395 		 * The message has been handed off to ill_capability_ack
8396 		 * and must not be freed below
8397 		 */
8398 		mp = NULL;
8399 		break;
8400 
8401 	case DL_INFO_ACK:
8402 		/* Call a routine to handle this one. */
8403 		ill_dlpi_done(ill, DL_INFO_REQ);
8404 		ip_ll_subnet_defaults(ill, mp);
8405 		ASSERT(!MUTEX_HELD(&ill->ill_phyint->phyint_ipsq->ipsq_lock));
8406 		return;
8407 	case DL_BIND_ACK:
8408 		/*
8409 		 * We should have an IOCTL waiting on this unless
8410 		 * sent by ill_dl_phys, in which case just return
8411 		 */
8412 		ill_dlpi_done(ill, DL_BIND_REQ);
8413 
8414 		if (ill->ill_ifname_pending) {
8415 			DTRACE_PROBE2(ip__rput__dlpi__ifname__pending,
8416 			    ill_t *, ill, mblk_t *, mp);
8417 			break;
8418 		}
8419 		mutex_enter(&ill->ill_lock);
8420 		ill->ill_dl_up = 1;
8421 		ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8422 		mutex_exit(&ill->ill_lock);
8423 
8424 		if (!ioctl_aborted)
8425 			mp1 = ipsq_pending_mp_get(ipsq, &connp);
8426 		if (mp1 == NULL) {
8427 			DTRACE_PROBE1(ip__rput__dlpi__no__mblk, ill_t *, ill);
8428 			break;
8429 		}
8430 		/*
8431 		 * mp1 was added by ill_dl_up(). if that is a result of
8432 		 * a DL_NOTE_REPLUMB notification, connp could be NULL.
8433 		 */
8434 		if (connp != NULL)
8435 			q = CONNP_TO_WQ(connp);
8436 		/*
8437 		 * We are exclusive. So nothing can change even after
8438 		 * we get the pending mp.
8439 		 */
8440 		ip1dbg(("ip_rput_dlpi: bind_ack %s\n", ill->ill_name));
8441 		DTRACE_PROBE1(ip__rput__dlpi__bind__ack, ill_t *, ill);
8442 		ill_nic_event_dispatch(ill, 0, NE_UP, NULL, 0);
8443 
8444 		/*
8445 		 * Now bring up the resolver; when that is complete, we'll
8446 		 * create IREs.  Note that we intentionally mirror what
8447 		 * ipif_up() would have done, because we got here by way of
8448 		 * ill_dl_up(), which stopped ipif_up()'s processing.
8449 		 */
8450 		if (ill->ill_isv6) {
8451 			/*
8452 			 * v6 interfaces.
8453 			 * Unlike ARP which has to do another bind
8454 			 * and attach, once we get here we are
8455 			 * done with NDP
8456 			 */
8457 			(void) ipif_resolver_up(ipif, Res_act_initial);
8458 			if ((err = ipif_ndp_up(ipif, B_TRUE)) == 0)
8459 				err = ipif_up_done_v6(ipif);
8460 		} else if (ill->ill_net_type == IRE_IF_RESOLVER) {
8461 			/*
8462 			 * ARP and other v4 external resolvers.
8463 			 * Leave the pending mblk intact so that
8464 			 * the ioctl completes in ip_rput().
8465 			 */
8466 			if (connp != NULL)
8467 				mutex_enter(&connp->conn_lock);
8468 			mutex_enter(&ill->ill_lock);
8469 			success = ipsq_pending_mp_add(connp, ipif, q, mp1, 0);
8470 			mutex_exit(&ill->ill_lock);
8471 			if (connp != NULL)
8472 				mutex_exit(&connp->conn_lock);
8473 			if (success) {
8474 				err = ipif_resolver_up(ipif, Res_act_initial);
8475 				if (err == EINPROGRESS) {
8476 					freemsg(mp);
8477 					return;
8478 				}
8479 				mp1 = ipsq_pending_mp_get(ipsq, &connp);
8480 			} else {
8481 				/* The conn has started closing */
8482 				err = EINTR;
8483 			}
8484 		} else {
8485 			/*
8486 			 * This one is complete. Reply to pending ioctl.
8487 			 */
8488 			(void) ipif_resolver_up(ipif, Res_act_initial);
8489 			err = ipif_up_done(ipif);
8490 		}
8491 
8492 		if ((err == 0) && (ill->ill_up_ipifs)) {
8493 			err = ill_up_ipifs(ill, q, mp1);
8494 			if (err == EINPROGRESS) {
8495 				freemsg(mp);
8496 				return;
8497 			}
8498 		}
8499 
8500 		/*
8501 		 * If we have a moved ipif to bring up, and everything has
8502 		 * succeeded to this point, bring it up on the IPMP ill.
8503 		 * Otherwise, leave it down -- the admin can try to bring it
8504 		 * up by hand if need be.
8505 		 */
8506 		if (ill->ill_move_ipif != NULL) {
8507 			if (err != 0) {
8508 				ill->ill_move_ipif = NULL;
8509 			} else {
8510 				ipif = ill->ill_move_ipif;
8511 				ill->ill_move_ipif = NULL;
8512 				err = ipif_up(ipif, q, mp1);
8513 				if (err == EINPROGRESS) {
8514 					freemsg(mp);
8515 					return;
8516 				}
8517 			}
8518 		}
8519 		break;
8520 
8521 	case DL_NOTIFY_IND: {
8522 		dl_notify_ind_t *notify = (dl_notify_ind_t *)mp->b_rptr;
8523 		uint_t orig_mtu, orig_mc_mtu;
8524 
8525 		switch (notify->dl_notification) {
8526 		case DL_NOTE_PHYS_ADDR:
8527 			err = ill_set_phys_addr(ill, mp);
8528 			break;
8529 
8530 		case DL_NOTE_REPLUMB:
8531 			/*
8532 			 * Directly return after calling ill_replumb().
8533 			 * Note that we should not free mp as it is reused
8534 			 * in the ill_replumb() function.
8535 			 */
8536 			err = ill_replumb(ill, mp);
8537 			return;
8538 
8539 		case DL_NOTE_FASTPATH_FLUSH:
8540 			nce_flush(ill, B_FALSE);
8541 			break;
8542 
8543 		case DL_NOTE_SDU_SIZE:
8544 		case DL_NOTE_SDU_SIZE2:
8545 			/*
8546 			 * The dce and fragmentation code can cope with
8547 			 * this changing while packets are being sent.
8548 			 * When packets are sent ip_output will discover
8549 			 * a change.
8550 			 *
8551 			 * Change the MTU size of the interface.
8552 			 */
8553 			mutex_enter(&ill->ill_lock);
8554 			orig_mtu = ill->ill_mtu;
8555 			orig_mc_mtu = ill->ill_mc_mtu;
8556 			switch (notify->dl_notification) {
8557 			case DL_NOTE_SDU_SIZE:
8558 				ill->ill_current_frag =
8559 				    (uint_t)notify->dl_data;
8560 				ill->ill_mc_mtu = (uint_t)notify->dl_data;
8561 				break;
8562 			case DL_NOTE_SDU_SIZE2:
8563 				ill->ill_current_frag =
8564 				    (uint_t)notify->dl_data1;
8565 				ill->ill_mc_mtu = (uint_t)notify->dl_data2;
8566 				break;
8567 			}
8568 			if (ill->ill_current_frag > ill->ill_max_frag)
8569 				ill->ill_max_frag = ill->ill_current_frag;
8570 
8571 			if (!(ill->ill_flags & ILLF_FIXEDMTU)) {
8572 				ill->ill_mtu = ill->ill_current_frag;
8573 
8574 				/*
8575 				 * If ill_user_mtu was set (via
8576 				 * SIOCSLIFLNKINFO), clamp ill_mtu at it.
8577 				 */
8578 				if (ill->ill_user_mtu != 0 &&
8579 				    ill->ill_user_mtu < ill->ill_mtu)
8580 					ill->ill_mtu = ill->ill_user_mtu;
8581 
8582 				if (ill->ill_user_mtu != 0 &&
8583 				    ill->ill_user_mtu < ill->ill_mc_mtu)
8584 					ill->ill_mc_mtu = ill->ill_user_mtu;
8585 
8586 				if (ill->ill_isv6) {
8587 					if (ill->ill_mtu < IPV6_MIN_MTU)
8588 						ill->ill_mtu = IPV6_MIN_MTU;
8589 					if (ill->ill_mc_mtu < IPV6_MIN_MTU)
8590 						ill->ill_mc_mtu = IPV6_MIN_MTU;
8591 				} else {
8592 					if (ill->ill_mtu < IP_MIN_MTU)
8593 						ill->ill_mtu = IP_MIN_MTU;
8594 					if (ill->ill_mc_mtu < IP_MIN_MTU)
8595 						ill->ill_mc_mtu = IP_MIN_MTU;
8596 				}
8597 			} else if (ill->ill_mc_mtu > ill->ill_mtu) {
8598 				ill->ill_mc_mtu = ill->ill_mtu;
8599 			}
8600 
8601 			mutex_exit(&ill->ill_lock);
8602 			/*
8603 			 * Make sure all dce_generation checks find out
8604 			 * that ill_mtu/ill_mc_mtu has changed.
8605 			 */
8606 			if (orig_mtu != ill->ill_mtu ||
8607 			    orig_mc_mtu != ill->ill_mc_mtu) {
8608 				dce_increment_all_generations(ill->ill_isv6,
8609 				    ill->ill_ipst);
8610 			}
8611 
8612 			/*
8613 			 * Refresh IPMP meta-interface MTU if necessary.
8614 			 */
8615 			if (IS_UNDER_IPMP(ill))
8616 				ipmp_illgrp_refresh_mtu(ill->ill_grp);
8617 			break;
8618 
8619 		case DL_NOTE_LINK_UP:
8620 		case DL_NOTE_LINK_DOWN: {
8621 			/*
8622 			 * We are writer. ill / phyint / ipsq assocs stable.
8623 			 * The RUNNING flag reflects the state of the link.
8624 			 */
8625 			phyint_t *phyint = ill->ill_phyint;
8626 			uint64_t new_phyint_flags;
8627 			boolean_t changed = B_FALSE;
8628 			boolean_t went_up;
8629 
8630 			went_up = notify->dl_notification == DL_NOTE_LINK_UP;
8631 			mutex_enter(&phyint->phyint_lock);
8632 
8633 			new_phyint_flags = went_up ?
8634 			    phyint->phyint_flags | PHYI_RUNNING :
8635 			    phyint->phyint_flags & ~PHYI_RUNNING;
8636 
8637 			if (IS_IPMP(ill)) {
8638 				new_phyint_flags = went_up ?
8639 				    new_phyint_flags & ~PHYI_FAILED :
8640 				    new_phyint_flags | PHYI_FAILED;
8641 			}
8642 
8643 			if (new_phyint_flags != phyint->phyint_flags) {
8644 				phyint->phyint_flags = new_phyint_flags;
8645 				changed = B_TRUE;
8646 			}
8647 			mutex_exit(&phyint->phyint_lock);
8648 			/*
8649 			 * ill_restart_dad handles the DAD restart and routing
8650 			 * socket notification logic.
8651 			 */
8652 			if (changed) {
8653 				ill_restart_dad(phyint->phyint_illv4, went_up);
8654 				ill_restart_dad(phyint->phyint_illv6, went_up);
8655 			}
8656 			break;
8657 		}
8658 		case DL_NOTE_PROMISC_ON_PHYS: {
8659 			phyint_t *phyint = ill->ill_phyint;
8660 
8661 			mutex_enter(&phyint->phyint_lock);
8662 			phyint->phyint_flags |= PHYI_PROMISC;
8663 			mutex_exit(&phyint->phyint_lock);
8664 			break;
8665 		}
8666 		case DL_NOTE_PROMISC_OFF_PHYS: {
8667 			phyint_t *phyint = ill->ill_phyint;
8668 
8669 			mutex_enter(&phyint->phyint_lock);
8670 			phyint->phyint_flags &= ~PHYI_PROMISC;
8671 			mutex_exit(&phyint->phyint_lock);
8672 			break;
8673 		}
8674 		case DL_NOTE_CAPAB_RENEG:
8675 			/*
8676 			 * Something changed on the driver side.
8677 			 * It wants us to renegotiate the capabilities
8678 			 * on this ill. One possible cause is the aggregation
8679 			 * interface under us where a port got added or
8680 			 * went away.
8681 			 *
8682 			 * If the capability negotiation is already done
8683 			 * or is in progress, reset the capabilities and
8684 			 * mark the ill's ill_capab_reneg to be B_TRUE,
8685 			 * so that when the ack comes back, we can start
8686 			 * the renegotiation process.
8687 			 *
8688 			 * Note that if ill_capab_reneg is already B_TRUE
8689 			 * (ill_dlpi_capab_state is IDS_UNKNOWN in this case),
8690 			 * the capability resetting request has been sent
8691 			 * and the renegotiation has not been started yet;
8692 			 * nothing needs to be done in this case.
8693 			 */
8694 			ipsq_current_start(ipsq, ill->ill_ipif, 0);
8695 			ill_capability_reset(ill, B_TRUE);
8696 			ipsq_current_finish(ipsq);
8697 			break;
8698 
8699 		case DL_NOTE_ALLOWED_IPS:
8700 			ill_set_allowed_ips(ill, mp);
8701 			break;
8702 		default:
8703 			ip0dbg(("ip_rput_dlpi_writer: unknown notification "
8704 			    "type 0x%x for DL_NOTIFY_IND\n",
8705 			    notify->dl_notification));
8706 			break;
8707 		}
8708 
8709 		/*
8710 		 * As this is an asynchronous operation, we
8711 		 * should not call ill_dlpi_done
8712 		 */
8713 		break;
8714 	}
8715 	case DL_NOTIFY_ACK: {
8716 		dl_notify_ack_t *noteack = (dl_notify_ack_t *)mp->b_rptr;
8717 
8718 		if (noteack->dl_notifications & DL_NOTE_LINK_UP)
8719 			ill->ill_note_link = 1;
8720 		ill_dlpi_done(ill, DL_NOTIFY_REQ);
8721 		break;
8722 	}
8723 	case DL_PHYS_ADDR_ACK: {
8724 		/*
8725 		 * As part of plumbing the interface via SIOCSLIFNAME,
8726 		 * ill_dl_phys() will queue a series of DL_PHYS_ADDR_REQs,
8727 		 * whose answers we receive here.  As each answer is received,
8728 		 * we call ill_dlpi_done() to dispatch the next request as
8729 		 * we're processing the current one.  Once all answers have
8730 		 * been received, we use ipsq_pending_mp_get() to dequeue the
8731 		 * outstanding IOCTL and reply to it.  (Because ill_dl_phys()
8732 		 * is invoked from an ill queue, conn_oper_pending_ill is not
8733 		 * available, but we know the ioctl is pending on ill_wq.)
8734 		 */
8735 		uint_t	paddrlen, paddroff;
8736 		uint8_t	*addr;
8737 
8738 		paddrreq = ill->ill_phys_addr_pend;
8739 		paddrlen = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_length;
8740 		paddroff = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_offset;
8741 		addr = mp->b_rptr + paddroff;
8742 
8743 		ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8744 		if (paddrreq == DL_IPV6_TOKEN) {
8745 			/*
8746 			 * bcopy to low-order bits of ill_token
8747 			 *
8748 			 * XXX Temporary hack - currently, all known tokens
8749 			 * are 64 bits, so I'll cheat for the moment.
8750 			 */
8751 			bcopy(addr, &ill->ill_token.s6_addr32[2], paddrlen);
8752 			ill->ill_token_length = paddrlen;
8753 			break;
8754 		} else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8755 			ASSERT(ill->ill_nd_lla_mp == NULL);
8756 			ill_set_ndmp(ill, mp, paddroff, paddrlen);
8757 			mp = NULL;
8758 			break;
8759 		} else if (paddrreq == DL_CURR_DEST_ADDR) {
8760 			ASSERT(ill->ill_dest_addr_mp == NULL);
8761 			ill->ill_dest_addr_mp = mp;
8762 			ill->ill_dest_addr = addr;
8763 			mp = NULL;
8764 			if (ill->ill_isv6) {
8765 				ill_setdesttoken(ill);
8766 				ipif_setdestlinklocal(ill->ill_ipif);
8767 			}
8768 			break;
8769 		}
8770 
8771 		ASSERT(paddrreq == DL_CURR_PHYS_ADDR);
8772 		ASSERT(ill->ill_phys_addr_mp == NULL);
8773 		if (!ill->ill_ifname_pending)
8774 			break;
8775 		ill->ill_ifname_pending = 0;
8776 		if (!ioctl_aborted)
8777 			mp1 = ipsq_pending_mp_get(ipsq, &connp);
8778 		if (mp1 != NULL) {
8779 			ASSERT(connp == NULL);
8780 			q = ill->ill_wq;
8781 		}
8782 		/*
8783 		 * If any error acks received during the plumbing sequence,
8784 		 * ill_ifname_pending_err will be set. Break out and send up
8785 		 * the error to the pending ioctl.
8786 		 */
8787 		if (ill->ill_ifname_pending_err != 0) {
8788 			err = ill->ill_ifname_pending_err;
8789 			ill->ill_ifname_pending_err = 0;
8790 			break;
8791 		}
8792 
8793 		ill->ill_phys_addr_mp = mp;
8794 		ill->ill_phys_addr = (paddrlen == 0 ? NULL : addr);
8795 		mp = NULL;
8796 
8797 		/*
8798 		 * If paddrlen or ill_phys_addr_length is zero, the DLPI
8799 		 * provider doesn't support physical addresses.  We check both
8800 		 * paddrlen and ill_phys_addr_length because sppp (PPP) does
8801 		 * not have physical addresses, but historically adversises a
8802 		 * physical address length of 0 in its DL_INFO_ACK, but 6 in
8803 		 * its DL_PHYS_ADDR_ACK.
8804 		 */
8805 		if (paddrlen == 0 || ill->ill_phys_addr_length == 0) {
8806 			ill->ill_phys_addr = NULL;
8807 		} else if (paddrlen != ill->ill_phys_addr_length) {
8808 			ip0dbg(("DL_PHYS_ADDR_ACK: got addrlen %d, expected %d",
8809 			    paddrlen, ill->ill_phys_addr_length));
8810 			err = EINVAL;
8811 			break;
8812 		}
8813 
8814 		if (ill->ill_nd_lla_mp == NULL) {
8815 			if ((mp_hw = copyb(ill->ill_phys_addr_mp)) == NULL) {
8816 				err = ENOMEM;
8817 				break;
8818 			}
8819 			ill_set_ndmp(ill, mp_hw, paddroff, paddrlen);
8820 		}
8821 
8822 		if (ill->ill_isv6) {
8823 			ill_setdefaulttoken(ill);
8824 			ipif_setlinklocal(ill->ill_ipif);
8825 		}
8826 		break;
8827 	}
8828 	case DL_OK_ACK:
8829 		ip2dbg(("DL_OK_ACK %s (0x%x)\n",
8830 		    dl_primstr((int)dloa->dl_correct_primitive),
8831 		    dloa->dl_correct_primitive));
8832 		DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer ok",
8833 		    char *, dl_primstr(dloa->dl_correct_primitive),
8834 		    ill_t *, ill);
8835 
8836 		switch (dloa->dl_correct_primitive) {
8837 		case DL_ENABMULTI_REQ:
8838 		case DL_DISABMULTI_REQ:
8839 			ill_dlpi_done(ill, dloa->dl_correct_primitive);
8840 			break;
8841 		case DL_PROMISCON_REQ:
8842 		case DL_PROMISCOFF_REQ:
8843 		case DL_UNBIND_REQ:
8844 		case DL_ATTACH_REQ:
8845 			ill_dlpi_done(ill, dloa->dl_correct_primitive);
8846 			break;
8847 		}
8848 		break;
8849 	default:
8850 		break;
8851 	}
8852 
8853 	freemsg(mp);
8854 	if (mp1 == NULL)
8855 		return;
8856 
8857 	/*
8858 	 * The operation must complete without EINPROGRESS since
8859 	 * ipsq_pending_mp_get() has removed the mblk (mp1).  Otherwise,
8860 	 * the operation will be stuck forever inside the IPSQ.
8861 	 */
8862 	ASSERT(err != EINPROGRESS);
8863 
8864 	DTRACE_PROBE4(ipif__ioctl, char *, "ip_rput_dlpi_writer finish",
8865 	    int, ipsq->ipsq_xop->ipx_current_ioctl, ill_t *, ill,
8866 	    ipif_t *, NULL);
8867 
8868 	switch (ipsq->ipsq_xop->ipx_current_ioctl) {
8869 	case 0:
8870 		ipsq_current_finish(ipsq);
8871 		break;
8872 
8873 	case SIOCSLIFNAME:
8874 	case IF_UNITSEL: {
8875 		ill_t *ill_other = ILL_OTHER(ill);
8876 
8877 		/*
8878 		 * If SIOCSLIFNAME or IF_UNITSEL is about to succeed, and the
8879 		 * ill has a peer which is in an IPMP group, then place ill
8880 		 * into the same group.  One catch: although ifconfig plumbs
8881 		 * the appropriate IPMP meta-interface prior to plumbing this
8882 		 * ill, it is possible for multiple ifconfig applications to
8883 		 * race (or for another application to adjust plumbing), in
8884 		 * which case the IPMP meta-interface we need will be missing.
8885 		 * If so, kick the phyint out of the group.
8886 		 */
8887 		if (err == 0 && ill_other != NULL && IS_UNDER_IPMP(ill_other)) {
8888 			ipmp_grp_t	*grp = ill->ill_phyint->phyint_grp;
8889 			ipmp_illgrp_t	*illg;
8890 
8891 			illg = ill->ill_isv6 ? grp->gr_v6 : grp->gr_v4;
8892 			if (illg == NULL)
8893 				ipmp_phyint_leave_grp(ill->ill_phyint);
8894 			else
8895 				ipmp_ill_join_illgrp(ill, illg);
8896 		}
8897 
8898 		if (ipsq->ipsq_xop->ipx_current_ioctl == IF_UNITSEL)
8899 			ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8900 		else
8901 			ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8902 		break;
8903 	}
8904 	case SIOCLIFADDIF:
8905 		ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8906 		break;
8907 
8908 	default:
8909 		ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8910 		break;
8911 	}
8912 }
8913 
8914 /*
8915  * ip_rput_other is called by ip_rput to handle messages modifying the global
8916  * state in IP.  If 'ipsq' is non-NULL, caller is writer on it.
8917  */
8918 /* ARGSUSED */
8919 void
ip_rput_other(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * dummy_arg)8920 ip_rput_other(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8921 {
8922 	ill_t		*ill = q->q_ptr;
8923 	struct iocblk	*iocp;
8924 
8925 	ip1dbg(("ip_rput_other "));
8926 	if (ipsq != NULL) {
8927 		ASSERT(IAM_WRITER_IPSQ(ipsq));
8928 		ASSERT(ipsq->ipsq_xop ==
8929 		    ill->ill_phyint->phyint_ipsq->ipsq_xop);
8930 	}
8931 
8932 	switch (mp->b_datap->db_type) {
8933 	case M_ERROR:
8934 	case M_HANGUP:
8935 		/*
8936 		 * The device has a problem.  We force the ILL down.  It can
8937 		 * be brought up again manually using SIOCSIFFLAGS (via
8938 		 * ifconfig or equivalent).
8939 		 */
8940 		ASSERT(ipsq != NULL);
8941 		if (mp->b_rptr < mp->b_wptr)
8942 			ill->ill_error = (int)(*mp->b_rptr & 0xFF);
8943 		if (ill->ill_error == 0)
8944 			ill->ill_error = ENXIO;
8945 		if (!ill_down_start(q, mp))
8946 			return;
8947 		ipif_all_down_tail(ipsq, q, mp, NULL);
8948 		break;
8949 	case M_IOCNAK: {
8950 		iocp = (struct iocblk *)mp->b_rptr;
8951 
8952 		ASSERT(iocp->ioc_cmd == DL_IOC_HDR_INFO);
8953 		/*
8954 		 * If this was the first attempt, turn off the fastpath
8955 		 * probing.
8956 		 */
8957 		mutex_enter(&ill->ill_lock);
8958 		if (ill->ill_dlpi_fastpath_state == IDS_INPROGRESS) {
8959 			ill->ill_dlpi_fastpath_state = IDS_FAILED;
8960 			mutex_exit(&ill->ill_lock);
8961 			/*
8962 			 * don't flush the nce_t entries: we use them
8963 			 * as an index to the ncec itself.
8964 			 */
8965 			ip1dbg(("ip_rput: DLPI fastpath off on interface %s\n",
8966 			    ill->ill_name));
8967 		} else {
8968 			mutex_exit(&ill->ill_lock);
8969 		}
8970 		freemsg(mp);
8971 		break;
8972 	}
8973 	default:
8974 		ASSERT(0);
8975 		break;
8976 	}
8977 }
8978 
8979 /*
8980  * Update any source route, record route or timestamp options
8981  * When it fails it has consumed the message and BUMPed the MIB.
8982  */
8983 boolean_t
ip_forward_options(mblk_t * mp,ipha_t * ipha,ill_t * dst_ill,ip_recv_attr_t * ira)8984 ip_forward_options(mblk_t *mp, ipha_t *ipha, ill_t *dst_ill,
8985     ip_recv_attr_t *ira)
8986 {
8987 	ipoptp_t	opts;
8988 	uchar_t		*opt;
8989 	uint8_t		optval;
8990 	uint8_t		optlen;
8991 	ipaddr_t	dst;
8992 	ipaddr_t	ifaddr;
8993 	uint32_t	ts;
8994 	timestruc_t	now;
8995 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
8996 
8997 	ip2dbg(("ip_forward_options\n"));
8998 	dst = ipha->ipha_dst;
8999 	for (optval = ipoptp_first(&opts, ipha);
9000 	    optval != IPOPT_EOL;
9001 	    optval = ipoptp_next(&opts)) {
9002 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9003 		opt = opts.ipoptp_cur;
9004 		optlen = opts.ipoptp_len;
9005 		ip2dbg(("ip_forward_options: opt %d, len %d\n",
9006 		    optval, opts.ipoptp_len));
9007 		switch (optval) {
9008 			uint32_t off;
9009 		case IPOPT_SSRR:
9010 		case IPOPT_LSRR:
9011 			/* Check if adminstratively disabled */
9012 			if (!ipst->ips_ip_forward_src_routed) {
9013 				BUMP_MIB(dst_ill->ill_ip_mib,
9014 				    ipIfStatsForwProhibits);
9015 				ip_drop_input("ICMP_SOURCE_ROUTE_FAILED",
9016 				    mp, dst_ill);
9017 				icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED,
9018 				    ira);
9019 				return (B_FALSE);
9020 			}
9021 			if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9022 				/*
9023 				 * Must be partial since ip_input_options
9024 				 * checked for strict.
9025 				 */
9026 				break;
9027 			}
9028 			off = opt[IPOPT_OFFSET];
9029 			off--;
9030 		redo_srr:
9031 			if (optlen < IP_ADDR_LEN ||
9032 			    off > optlen - IP_ADDR_LEN) {
9033 				/* End of source route */
9034 				ip1dbg((
9035 				    "ip_forward_options: end of SR\n"));
9036 				break;
9037 			}
9038 			/* Pick a reasonable address on the outbound if */
9039 			ASSERT(dst_ill != NULL);
9040 			if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9041 			    INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9042 			    NULL) != 0) {
9043 				/* No source! Shouldn't happen */
9044 				ifaddr = INADDR_ANY;
9045 			}
9046 			bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9047 			bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9048 			ip1dbg(("ip_forward_options: next hop 0x%x\n",
9049 			    ntohl(dst)));
9050 
9051 			/*
9052 			 * Check if our address is present more than
9053 			 * once as consecutive hops in source route.
9054 			 */
9055 			if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9056 				off += IP_ADDR_LEN;
9057 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9058 				goto redo_srr;
9059 			}
9060 			ipha->ipha_dst = dst;
9061 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9062 			break;
9063 		case IPOPT_RR:
9064 			off = opt[IPOPT_OFFSET];
9065 			off--;
9066 			if (optlen < IP_ADDR_LEN ||
9067 			    off > optlen - IP_ADDR_LEN) {
9068 				/* No more room - ignore */
9069 				ip1dbg((
9070 				    "ip_forward_options: end of RR\n"));
9071 				break;
9072 			}
9073 			/* Pick a reasonable address on the outbound if */
9074 			ASSERT(dst_ill != NULL);
9075 			if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9076 			    INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9077 			    NULL) != 0) {
9078 				/* No source! Shouldn't happen */
9079 				ifaddr = INADDR_ANY;
9080 			}
9081 			bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9082 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9083 			break;
9084 		case IPOPT_TS:
9085 			/* Insert timestamp if there is room */
9086 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9087 			case IPOPT_TS_TSONLY:
9088 				off = IPOPT_TS_TIMELEN;
9089 				break;
9090 			case IPOPT_TS_PRESPEC:
9091 			case IPOPT_TS_PRESPEC_RFC791:
9092 				/* Verify that the address matched */
9093 				off = opt[IPOPT_OFFSET] - 1;
9094 				bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9095 				if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9096 					/* Not for us */
9097 					break;
9098 				}
9099 				/* FALLTHRU */
9100 			case IPOPT_TS_TSANDADDR:
9101 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9102 				break;
9103 			default:
9104 				/*
9105 				 * ip_*put_options should have already
9106 				 * dropped this packet.
9107 				 */
9108 				cmn_err(CE_PANIC, "ip_forward_options: "
9109 				    "unknown IT - bug in ip_input_options?\n");
9110 				return (B_TRUE);	/* Keep "lint" happy */
9111 			}
9112 			if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9113 				/* Increase overflow counter */
9114 				off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9115 				opt[IPOPT_POS_OV_FLG] =
9116 				    (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9117 				    (off << 4));
9118 				break;
9119 			}
9120 			off = opt[IPOPT_OFFSET] - 1;
9121 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9122 			case IPOPT_TS_PRESPEC:
9123 			case IPOPT_TS_PRESPEC_RFC791:
9124 			case IPOPT_TS_TSANDADDR:
9125 				/* Pick a reasonable addr on the outbound if */
9126 				ASSERT(dst_ill != NULL);
9127 				if (ip_select_source_v4(dst_ill, INADDR_ANY,
9128 				    dst, INADDR_ANY, ALL_ZONES, ipst, &ifaddr,
9129 				    NULL, NULL) != 0) {
9130 					/* No source! Shouldn't happen */
9131 					ifaddr = INADDR_ANY;
9132 				}
9133 				bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9134 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9135 				/* FALLTHRU */
9136 			case IPOPT_TS_TSONLY:
9137 				off = opt[IPOPT_OFFSET] - 1;
9138 				/* Compute # of milliseconds since midnight */
9139 				gethrestime(&now);
9140 				ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9141 				    NSEC2MSEC(now.tv_nsec);
9142 				bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9143 				opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9144 				break;
9145 			}
9146 			break;
9147 		}
9148 	}
9149 	return (B_TRUE);
9150 }
9151 
9152 /*
9153  * Call ill_frag_timeout to do garbage collection. ill_frag_timeout
9154  * returns 'true' if there are still fragments left on the queue, in
9155  * which case we restart the timer.
9156  */
9157 void
ill_frag_timer(void * arg)9158 ill_frag_timer(void *arg)
9159 {
9160 	ill_t	*ill = (ill_t *)arg;
9161 	boolean_t frag_pending;
9162 	ip_stack_t *ipst = ill->ill_ipst;
9163 	time_t	timeout;
9164 
9165 	mutex_enter(&ill->ill_lock);
9166 	ASSERT(!ill->ill_fragtimer_executing);
9167 	if (ill->ill_state_flags & ILL_CONDEMNED) {
9168 		ill->ill_frag_timer_id = 0;
9169 		mutex_exit(&ill->ill_lock);
9170 		return;
9171 	}
9172 	ill->ill_fragtimer_executing = 1;
9173 	mutex_exit(&ill->ill_lock);
9174 
9175 	timeout = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9176 	    ipst->ips_ip_reassembly_timeout);
9177 
9178 	frag_pending = ill_frag_timeout(ill, timeout);
9179 
9180 	/*
9181 	 * Restart the timer, if we have fragments pending or if someone
9182 	 * wanted us to be scheduled again.
9183 	 */
9184 	mutex_enter(&ill->ill_lock);
9185 	ill->ill_fragtimer_executing = 0;
9186 	ill->ill_frag_timer_id = 0;
9187 	if (frag_pending || ill->ill_fragtimer_needrestart)
9188 		ill_frag_timer_start(ill);
9189 	mutex_exit(&ill->ill_lock);
9190 }
9191 
9192 void
ill_frag_timer_start(ill_t * ill)9193 ill_frag_timer_start(ill_t *ill)
9194 {
9195 	ip_stack_t *ipst = ill->ill_ipst;
9196 	clock_t	timeo_ms;
9197 
9198 	ASSERT(MUTEX_HELD(&ill->ill_lock));
9199 
9200 	/* If the ill is closing or opening don't proceed */
9201 	if (ill->ill_state_flags & ILL_CONDEMNED)
9202 		return;
9203 
9204 	if (ill->ill_fragtimer_executing) {
9205 		/*
9206 		 * ill_frag_timer is currently executing. Just record the
9207 		 * the fact that we want the timer to be restarted.
9208 		 * ill_frag_timer will post a timeout before it returns,
9209 		 * ensuring it will be called again.
9210 		 */
9211 		ill->ill_fragtimer_needrestart = 1;
9212 		return;
9213 	}
9214 
9215 	if (ill->ill_frag_timer_id == 0) {
9216 		timeo_ms = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9217 		    ipst->ips_ip_reassembly_timeout) * SECONDS;
9218 
9219 		/*
9220 		 * The timer is neither running nor is the timeout handler
9221 		 * executing. Post a timeout so that ill_frag_timer will be
9222 		 * called
9223 		 */
9224 		ill->ill_frag_timer_id = timeout(ill_frag_timer, ill,
9225 		    MSEC_TO_TICK(timeo_ms >> 1));
9226 		ill->ill_fragtimer_needrestart = 0;
9227 	}
9228 }
9229 
9230 /*
9231  * Update any source route, record route or timestamp options.
9232  * Check that we are at end of strict source route.
9233  * The options have already been checked for sanity in ip_input_options().
9234  */
9235 boolean_t
ip_input_local_options(mblk_t * mp,ipha_t * ipha,ip_recv_attr_t * ira)9236 ip_input_local_options(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
9237 {
9238 	ipoptp_t	opts;
9239 	uchar_t		*opt;
9240 	uint8_t		optval;
9241 	uint8_t		optlen;
9242 	ipaddr_t	dst;
9243 	ipaddr_t	ifaddr;
9244 	uint32_t	ts;
9245 	timestruc_t	now;
9246 	ill_t		*ill = ira->ira_ill;
9247 	ip_stack_t	*ipst = ill->ill_ipst;
9248 
9249 	ip2dbg(("ip_input_local_options\n"));
9250 
9251 	for (optval = ipoptp_first(&opts, ipha);
9252 	    optval != IPOPT_EOL;
9253 	    optval = ipoptp_next(&opts)) {
9254 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9255 		opt = opts.ipoptp_cur;
9256 		optlen = opts.ipoptp_len;
9257 		ip2dbg(("ip_input_local_options: opt %d, len %d\n",
9258 		    optval, optlen));
9259 		switch (optval) {
9260 			uint32_t off;
9261 		case IPOPT_SSRR:
9262 		case IPOPT_LSRR:
9263 			off = opt[IPOPT_OFFSET];
9264 			off--;
9265 			if (optlen < IP_ADDR_LEN ||
9266 			    off > optlen - IP_ADDR_LEN) {
9267 				/* End of source route */
9268 				ip1dbg(("ip_input_local_options: end of SR\n"));
9269 				break;
9270 			}
9271 			/*
9272 			 * This will only happen if two consecutive entries
9273 			 * in the source route contains our address or if
9274 			 * it is a packet with a loose source route which
9275 			 * reaches us before consuming the whole source route
9276 			 */
9277 			ip1dbg(("ip_input_local_options: not end of SR\n"));
9278 			if (optval == IPOPT_SSRR) {
9279 				goto bad_src_route;
9280 			}
9281 			/*
9282 			 * Hack: instead of dropping the packet truncate the
9283 			 * source route to what has been used by filling the
9284 			 * rest with IPOPT_NOP.
9285 			 */
9286 			opt[IPOPT_OLEN] = (uint8_t)off;
9287 			while (off < optlen) {
9288 				opt[off++] = IPOPT_NOP;
9289 			}
9290 			break;
9291 		case IPOPT_RR:
9292 			off = opt[IPOPT_OFFSET];
9293 			off--;
9294 			if (optlen < IP_ADDR_LEN ||
9295 			    off > optlen - IP_ADDR_LEN) {
9296 				/* No more room - ignore */
9297 				ip1dbg((
9298 				    "ip_input_local_options: end of RR\n"));
9299 				break;
9300 			}
9301 			/* Pick a reasonable address on the outbound if */
9302 			if (ip_select_source_v4(ill, INADDR_ANY, ipha->ipha_dst,
9303 			    INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9304 			    NULL) != 0) {
9305 				/* No source! Shouldn't happen */
9306 				ifaddr = INADDR_ANY;
9307 			}
9308 			bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9309 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9310 			break;
9311 		case IPOPT_TS:
9312 			/* Insert timestamp if there is romm */
9313 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9314 			case IPOPT_TS_TSONLY:
9315 				off = IPOPT_TS_TIMELEN;
9316 				break;
9317 			case IPOPT_TS_PRESPEC:
9318 			case IPOPT_TS_PRESPEC_RFC791:
9319 				/* Verify that the address matched */
9320 				off = opt[IPOPT_OFFSET] - 1;
9321 				bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9322 				if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9323 					/* Not for us */
9324 					break;
9325 				}
9326 				/* FALLTHRU */
9327 			case IPOPT_TS_TSANDADDR:
9328 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9329 				break;
9330 			default:
9331 				/*
9332 				 * ip_*put_options should have already
9333 				 * dropped this packet.
9334 				 */
9335 				cmn_err(CE_PANIC, "ip_input_local_options: "
9336 				    "unknown IT - bug in ip_input_options?\n");
9337 				return (B_TRUE);	/* Keep "lint" happy */
9338 			}
9339 			if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9340 				/* Increase overflow counter */
9341 				off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9342 				opt[IPOPT_POS_OV_FLG] =
9343 				    (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9344 				    (off << 4));
9345 				break;
9346 			}
9347 			off = opt[IPOPT_OFFSET] - 1;
9348 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9349 			case IPOPT_TS_PRESPEC:
9350 			case IPOPT_TS_PRESPEC_RFC791:
9351 			case IPOPT_TS_TSANDADDR:
9352 				/* Pick a reasonable addr on the outbound if */
9353 				if (ip_select_source_v4(ill, INADDR_ANY,
9354 				    ipha->ipha_dst, INADDR_ANY, ALL_ZONES, ipst,
9355 				    &ifaddr, NULL, NULL) != 0) {
9356 					/* No source! Shouldn't happen */
9357 					ifaddr = INADDR_ANY;
9358 				}
9359 				bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9360 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9361 				/* FALLTHRU */
9362 			case IPOPT_TS_TSONLY:
9363 				off = opt[IPOPT_OFFSET] - 1;
9364 				/* Compute # of milliseconds since midnight */
9365 				gethrestime(&now);
9366 				ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9367 				    NSEC2MSEC(now.tv_nsec);
9368 				bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9369 				opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9370 				break;
9371 			}
9372 			break;
9373 		}
9374 	}
9375 	return (B_TRUE);
9376 
9377 bad_src_route:
9378 	/* make sure we clear any indication of a hardware checksum */
9379 	DB_CKSUMFLAGS(mp) = 0;
9380 	ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
9381 	icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9382 	return (B_FALSE);
9383 
9384 }
9385 
9386 /*
9387  * Process IP options in an inbound packet.  Always returns the nexthop.
9388  * Normally this is the passed in nexthop, but if there is an option
9389  * that effects the nexthop (such as a source route) that will be returned.
9390  * Sets *errorp if there is an error, in which case an ICMP error has been sent
9391  * and mp freed.
9392  */
9393 ipaddr_t
ip_input_options(ipha_t * ipha,ipaddr_t dst,mblk_t * mp,ip_recv_attr_t * ira,int * errorp)9394 ip_input_options(ipha_t *ipha, ipaddr_t dst, mblk_t *mp,
9395     ip_recv_attr_t *ira, int *errorp)
9396 {
9397 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
9398 	ipoptp_t	opts;
9399 	uchar_t		*opt;
9400 	uint8_t		optval;
9401 	uint8_t		optlen;
9402 	intptr_t	code = 0;
9403 	ire_t		*ire;
9404 
9405 	ip2dbg(("ip_input_options\n"));
9406 	*errorp = 0;
9407 	for (optval = ipoptp_first(&opts, ipha);
9408 	    optval != IPOPT_EOL;
9409 	    optval = ipoptp_next(&opts)) {
9410 		opt = opts.ipoptp_cur;
9411 		optlen = opts.ipoptp_len;
9412 		ip2dbg(("ip_input_options: opt %d, len %d\n",
9413 		    optval, optlen));
9414 		/*
9415 		 * Note: we need to verify the checksum before we
9416 		 * modify anything thus this routine only extracts the next
9417 		 * hop dst from any source route.
9418 		 */
9419 		switch (optval) {
9420 			uint32_t off;
9421 		case IPOPT_SSRR:
9422 		case IPOPT_LSRR:
9423 			if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9424 				if (optval == IPOPT_SSRR) {
9425 					ip1dbg(("ip_input_options: not next"
9426 					    " strict source route 0x%x\n",
9427 					    ntohl(dst)));
9428 					code = (char *)&ipha->ipha_dst -
9429 					    (char *)ipha;
9430 					goto param_prob; /* RouterReq's */
9431 				}
9432 				ip2dbg(("ip_input_options: "
9433 				    "not next source route 0x%x\n",
9434 				    ntohl(dst)));
9435 				break;
9436 			}
9437 
9438 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9439 				ip1dbg((
9440 				    "ip_input_options: bad option offset\n"));
9441 				code = (char *)&opt[IPOPT_OLEN] -
9442 				    (char *)ipha;
9443 				goto param_prob;
9444 			}
9445 			off = opt[IPOPT_OFFSET];
9446 			off--;
9447 		redo_srr:
9448 			if (optlen < IP_ADDR_LEN ||
9449 			    off > optlen - IP_ADDR_LEN) {
9450 				/* End of source route */
9451 				ip1dbg(("ip_input_options: end of SR\n"));
9452 				break;
9453 			}
9454 			bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9455 			ip1dbg(("ip_input_options: next hop 0x%x\n",
9456 			    ntohl(dst)));
9457 
9458 			/*
9459 			 * Check if our address is present more than
9460 			 * once as consecutive hops in source route.
9461 			 * XXX verify per-interface ip_forwarding
9462 			 * for source route?
9463 			 */
9464 			if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9465 				off += IP_ADDR_LEN;
9466 				goto redo_srr;
9467 			}
9468 
9469 			if (dst == htonl(INADDR_LOOPBACK)) {
9470 				ip1dbg(("ip_input_options: loopback addr in "
9471 				    "source route!\n"));
9472 				goto bad_src_route;
9473 			}
9474 			/*
9475 			 * For strict: verify that dst is directly
9476 			 * reachable.
9477 			 */
9478 			if (optval == IPOPT_SSRR) {
9479 				ire = ire_ftable_lookup_v4(dst, 0, 0,
9480 				    IRE_INTERFACE, NULL, ALL_ZONES,
9481 				    ira->ira_tsl,
9482 				    MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
9483 				    NULL);
9484 				if (ire == NULL) {
9485 					ip1dbg(("ip_input_options: SSRR not "
9486 					    "directly reachable: 0x%x\n",
9487 					    ntohl(dst)));
9488 					goto bad_src_route;
9489 				}
9490 				ire_refrele(ire);
9491 			}
9492 			/*
9493 			 * Defer update of the offset and the record route
9494 			 * until the packet is forwarded.
9495 			 */
9496 			break;
9497 		case IPOPT_RR:
9498 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9499 				ip1dbg((
9500 				    "ip_input_options: bad option offset\n"));
9501 				code = (char *)&opt[IPOPT_OLEN] -
9502 				    (char *)ipha;
9503 				goto param_prob;
9504 			}
9505 			break;
9506 		case IPOPT_TS:
9507 			/*
9508 			 * Verify that length >= 5 and that there is either
9509 			 * room for another timestamp or that the overflow
9510 			 * counter is not maxed out.
9511 			 */
9512 			code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
9513 			if (optlen < IPOPT_MINLEN_IT) {
9514 				goto param_prob;
9515 			}
9516 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9517 				ip1dbg((
9518 				    "ip_input_options: bad option offset\n"));
9519 				code = (char *)&opt[IPOPT_OFFSET] -
9520 				    (char *)ipha;
9521 				goto param_prob;
9522 			}
9523 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9524 			case IPOPT_TS_TSONLY:
9525 				off = IPOPT_TS_TIMELEN;
9526 				break;
9527 			case IPOPT_TS_TSANDADDR:
9528 			case IPOPT_TS_PRESPEC:
9529 			case IPOPT_TS_PRESPEC_RFC791:
9530 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9531 				break;
9532 			default:
9533 				code = (char *)&opt[IPOPT_POS_OV_FLG] -
9534 				    (char *)ipha;
9535 				goto param_prob;
9536 			}
9537 			if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
9538 			    (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
9539 				/*
9540 				 * No room and the overflow counter is 15
9541 				 * already.
9542 				 */
9543 				goto param_prob;
9544 			}
9545 			break;
9546 		}
9547 	}
9548 
9549 	if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0) {
9550 		return (dst);
9551 	}
9552 
9553 	ip1dbg(("ip_input_options: error processing IP options."));
9554 	code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
9555 
9556 param_prob:
9557 	/* make sure we clear any indication of a hardware checksum */
9558 	DB_CKSUMFLAGS(mp) = 0;
9559 	ip_drop_input("ICMP_PARAM_PROBLEM", mp, ira->ira_ill);
9560 	icmp_param_problem(mp, (uint8_t)code, ira);
9561 	*errorp = -1;
9562 	return (dst);
9563 
9564 bad_src_route:
9565 	/* make sure we clear any indication of a hardware checksum */
9566 	DB_CKSUMFLAGS(mp) = 0;
9567 	ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ira->ira_ill);
9568 	icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9569 	*errorp = -1;
9570 	return (dst);
9571 }
9572 
9573 /*
9574  * IP & ICMP info in >=14 msg's ...
9575  *  - ip fixed part (mib2_ip_t)
9576  *  - icmp fixed part (mib2_icmp_t)
9577  *  - ipAddrEntryTable (ip 20)		all IPv4 ipifs
9578  *  - ipRouteEntryTable (ip 21)		all IPv4 IREs
9579  *  - ipNetToMediaEntryTable (ip 22)	all IPv4 Neighbor Cache entries
9580  *  - ipRouteAttributeTable (ip 102)	labeled routes
9581  *  - ip multicast membership (ip_member_t)
9582  *  - ip multicast source filtering (ip_grpsrc_t)
9583  *  - igmp fixed part (struct igmpstat)
9584  *  - multicast routing stats (struct mrtstat)
9585  *  - multicast routing vifs (array of struct vifctl)
9586  *  - multicast routing routes (array of struct mfcctl)
9587  *  - ip6 fixed part (mib2_ipv6IfStatsEntry_t)
9588  *					One per ill plus one generic
9589  *  - icmp6 fixed part (mib2_ipv6IfIcmpEntry_t)
9590  *					One per ill plus one generic
9591  *  - ipv6RouteEntry			all IPv6 IREs
9592  *  - ipv6RouteAttributeTable (ip6 102)	labeled routes
9593  *  - ipv6NetToMediaEntry		all IPv6 Neighbor Cache entries
9594  *  - ipv6AddrEntry			all IPv6 ipifs
9595  *  - ipv6 multicast membership (ipv6_member_t)
9596  *  - ipv6 multicast source filtering (ipv6_grpsrc_t)
9597  *
9598  * NOTE: original mpctl is copied for msg's 2..N, since its ctl part is
9599  * already filled in by the caller.
9600  * If legacy_req is true then MIB structures needs to be truncated to their
9601  * legacy sizes before being returned.
9602  * Return value of 0 indicates that no messages were sent and caller
9603  * should free mpctl.
9604  */
9605 int
ip_snmp_get(queue_t * q,mblk_t * mpctl,int level,boolean_t legacy_req)9606 ip_snmp_get(queue_t *q, mblk_t *mpctl, int level, boolean_t legacy_req)
9607 {
9608 	ip_stack_t *ipst;
9609 	sctp_stack_t *sctps;
9610 
9611 	if (q->q_next != NULL) {
9612 		ipst = ILLQ_TO_IPST(q);
9613 	} else {
9614 		ipst = CONNQ_TO_IPST(q);
9615 	}
9616 	ASSERT(ipst != NULL);
9617 	sctps = ipst->ips_netstack->netstack_sctp;
9618 
9619 	if (mpctl == NULL || mpctl->b_cont == NULL) {
9620 		return (0);
9621 	}
9622 
9623 	/*
9624 	 * For the purposes of the (broken) packet shell use
9625 	 * of the level we make sure MIB2_TCP/MIB2_UDP can be used
9626 	 * to make TCP and UDP appear first in the list of mib items.
9627 	 * TBD: We could expand this and use it in netstat so that
9628 	 * the kernel doesn't have to produce large tables (connections,
9629 	 * routes, etc) when netstat only wants the statistics or a particular
9630 	 * table.
9631 	 */
9632 	if (!(level == MIB2_TCP || level == MIB2_UDP)) {
9633 		if ((mpctl = icmp_snmp_get(q, mpctl)) == NULL) {
9634 			return (1);
9635 		}
9636 	}
9637 
9638 	if (level != MIB2_TCP) {
9639 		if ((mpctl = udp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9640 			return (1);
9641 		}
9642 	}
9643 
9644 	if (level != MIB2_UDP) {
9645 		if ((mpctl = tcp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9646 			return (1);
9647 		}
9648 	}
9649 
9650 	if ((mpctl = ip_snmp_get_mib2_ip_traffic_stats(q, mpctl,
9651 	    ipst, legacy_req)) == NULL) {
9652 		return (1);
9653 	}
9654 
9655 	if ((mpctl = ip_snmp_get_mib2_ip6(q, mpctl, ipst,
9656 	    legacy_req)) == NULL) {
9657 		return (1);
9658 	}
9659 
9660 	if ((mpctl = ip_snmp_get_mib2_icmp(q, mpctl, ipst)) == NULL) {
9661 		return (1);
9662 	}
9663 
9664 	if ((mpctl = ip_snmp_get_mib2_icmp6(q, mpctl, ipst)) == NULL) {
9665 		return (1);
9666 	}
9667 
9668 	if ((mpctl = ip_snmp_get_mib2_igmp(q, mpctl, ipst)) == NULL) {
9669 		return (1);
9670 	}
9671 
9672 	if ((mpctl = ip_snmp_get_mib2_multi(q, mpctl, ipst)) == NULL) {
9673 		return (1);
9674 	}
9675 
9676 	if ((mpctl = ip_snmp_get_mib2_ip_addr(q, mpctl, ipst,
9677 	    legacy_req)) == NULL) {
9678 		return (1);
9679 	}
9680 
9681 	if ((mpctl = ip_snmp_get_mib2_ip6_addr(q, mpctl, ipst,
9682 	    legacy_req)) == NULL) {
9683 		return (1);
9684 	}
9685 
9686 	if ((mpctl = ip_snmp_get_mib2_ip_group_mem(q, mpctl, ipst)) == NULL) {
9687 		return (1);
9688 	}
9689 
9690 	if ((mpctl = ip_snmp_get_mib2_ip6_group_mem(q, mpctl, ipst)) == NULL) {
9691 		return (1);
9692 	}
9693 
9694 	if ((mpctl = ip_snmp_get_mib2_ip_group_src(q, mpctl, ipst)) == NULL) {
9695 		return (1);
9696 	}
9697 
9698 	if ((mpctl = ip_snmp_get_mib2_ip6_group_src(q, mpctl, ipst)) == NULL) {
9699 		return (1);
9700 	}
9701 
9702 	if ((mpctl = ip_snmp_get_mib2_virt_multi(q, mpctl, ipst)) == NULL) {
9703 		return (1);
9704 	}
9705 
9706 	if ((mpctl = ip_snmp_get_mib2_multi_rtable(q, mpctl, ipst)) == NULL) {
9707 		return (1);
9708 	}
9709 
9710 	mpctl = ip_snmp_get_mib2_ip_route_media(q, mpctl, level, ipst);
9711 	if (mpctl == NULL)
9712 		return (1);
9713 
9714 	mpctl = ip_snmp_get_mib2_ip6_route_media(q, mpctl, level, ipst);
9715 	if (mpctl == NULL)
9716 		return (1);
9717 
9718 	if ((mpctl = sctp_snmp_get_mib2(q, mpctl, sctps)) == NULL) {
9719 		return (1);
9720 	}
9721 	if ((mpctl = ip_snmp_get_mib2_ip_dce(q, mpctl, ipst)) == NULL) {
9722 		return (1);
9723 	}
9724 	freemsg(mpctl);
9725 	return (1);
9726 }
9727 
9728 /* Get global (legacy) IPv4 statistics */
9729 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)9730 ip_snmp_get_mib2_ip(queue_t *q, mblk_t *mpctl, mib2_ipIfStatsEntry_t *ipmib,
9731     ip_stack_t *ipst, boolean_t legacy_req)
9732 {
9733 	mib2_ip_t		old_ip_mib;
9734 	struct opthdr		*optp;
9735 	mblk_t			*mp2ctl;
9736 	mib2_ipAddrEntry_t	mae;
9737 
9738 	/*
9739 	 * make a copy of the original message
9740 	 */
9741 	mp2ctl = copymsg(mpctl);
9742 
9743 	/* fixed length IP structure... */
9744 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9745 	optp->level = MIB2_IP;
9746 	optp->name = 0;
9747 	SET_MIB(old_ip_mib.ipForwarding,
9748 	    (WE_ARE_FORWARDING(ipst) ? 1 : 2));
9749 	SET_MIB(old_ip_mib.ipDefaultTTL,
9750 	    (uint32_t)ipst->ips_ip_def_ttl);
9751 	SET_MIB(old_ip_mib.ipReasmTimeout,
9752 	    ipst->ips_ip_reassembly_timeout);
9753 	SET_MIB(old_ip_mib.ipAddrEntrySize,
9754 	    (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9755 	    sizeof (mib2_ipAddrEntry_t));
9756 	SET_MIB(old_ip_mib.ipRouteEntrySize,
9757 	    sizeof (mib2_ipRouteEntry_t));
9758 	SET_MIB(old_ip_mib.ipNetToMediaEntrySize,
9759 	    sizeof (mib2_ipNetToMediaEntry_t));
9760 	SET_MIB(old_ip_mib.ipMemberEntrySize, sizeof (ip_member_t));
9761 	SET_MIB(old_ip_mib.ipGroupSourceEntrySize, sizeof (ip_grpsrc_t));
9762 	SET_MIB(old_ip_mib.ipRouteAttributeSize,
9763 	    sizeof (mib2_ipAttributeEntry_t));
9764 	SET_MIB(old_ip_mib.transportMLPSize, sizeof (mib2_transportMLPEntry_t));
9765 	SET_MIB(old_ip_mib.ipDestEntrySize, sizeof (dest_cache_entry_t));
9766 
9767 	/*
9768 	 * Grab the statistics from the new IP MIB
9769 	 */
9770 	SET_MIB(old_ip_mib.ipInReceives,
9771 	    (uint32_t)ipmib->ipIfStatsHCInReceives);
9772 	SET_MIB(old_ip_mib.ipInHdrErrors, ipmib->ipIfStatsInHdrErrors);
9773 	SET_MIB(old_ip_mib.ipInAddrErrors, ipmib->ipIfStatsInAddrErrors);
9774 	SET_MIB(old_ip_mib.ipForwDatagrams,
9775 	    (uint32_t)ipmib->ipIfStatsHCOutForwDatagrams);
9776 	SET_MIB(old_ip_mib.ipInUnknownProtos,
9777 	    ipmib->ipIfStatsInUnknownProtos);
9778 	SET_MIB(old_ip_mib.ipInDiscards, ipmib->ipIfStatsInDiscards);
9779 	SET_MIB(old_ip_mib.ipInDelivers,
9780 	    (uint32_t)ipmib->ipIfStatsHCInDelivers);
9781 	SET_MIB(old_ip_mib.ipOutRequests,
9782 	    (uint32_t)ipmib->ipIfStatsHCOutRequests);
9783 	SET_MIB(old_ip_mib.ipOutDiscards, ipmib->ipIfStatsOutDiscards);
9784 	SET_MIB(old_ip_mib.ipOutNoRoutes, ipmib->ipIfStatsOutNoRoutes);
9785 	SET_MIB(old_ip_mib.ipReasmReqds, ipmib->ipIfStatsReasmReqds);
9786 	SET_MIB(old_ip_mib.ipReasmOKs, ipmib->ipIfStatsReasmOKs);
9787 	SET_MIB(old_ip_mib.ipReasmFails, ipmib->ipIfStatsReasmFails);
9788 	SET_MIB(old_ip_mib.ipFragOKs, ipmib->ipIfStatsOutFragOKs);
9789 	SET_MIB(old_ip_mib.ipFragFails, ipmib->ipIfStatsOutFragFails);
9790 	SET_MIB(old_ip_mib.ipFragCreates, ipmib->ipIfStatsOutFragCreates);
9791 
9792 	/* ipRoutingDiscards is not being used */
9793 	SET_MIB(old_ip_mib.ipRoutingDiscards, 0);
9794 	SET_MIB(old_ip_mib.tcpInErrs, ipmib->tcpIfStatsInErrs);
9795 	SET_MIB(old_ip_mib.udpNoPorts, ipmib->udpIfStatsNoPorts);
9796 	SET_MIB(old_ip_mib.ipInCksumErrs, ipmib->ipIfStatsInCksumErrs);
9797 	SET_MIB(old_ip_mib.ipReasmDuplicates,
9798 	    ipmib->ipIfStatsReasmDuplicates);
9799 	SET_MIB(old_ip_mib.ipReasmPartDups, ipmib->ipIfStatsReasmPartDups);
9800 	SET_MIB(old_ip_mib.ipForwProhibits, ipmib->ipIfStatsForwProhibits);
9801 	SET_MIB(old_ip_mib.udpInCksumErrs, ipmib->udpIfStatsInCksumErrs);
9802 	SET_MIB(old_ip_mib.udpInOverflows, ipmib->udpIfStatsInOverflows);
9803 	SET_MIB(old_ip_mib.rawipInOverflows,
9804 	    ipmib->rawipIfStatsInOverflows);
9805 
9806 	SET_MIB(old_ip_mib.ipsecInSucceeded, ipmib->ipsecIfStatsInSucceeded);
9807 	SET_MIB(old_ip_mib.ipsecInFailed, ipmib->ipsecIfStatsInFailed);
9808 	SET_MIB(old_ip_mib.ipInIPv6, ipmib->ipIfStatsInWrongIPVersion);
9809 	SET_MIB(old_ip_mib.ipOutIPv6, ipmib->ipIfStatsOutWrongIPVersion);
9810 	SET_MIB(old_ip_mib.ipOutSwitchIPv6,
9811 	    ipmib->ipIfStatsOutSwitchIPVersion);
9812 
9813 	if (!snmp_append_data(mpctl->b_cont, (char *)&old_ip_mib,
9814 	    (int)sizeof (old_ip_mib))) {
9815 		ip1dbg(("ip_snmp_get_mib2_ip: failed to allocate %u bytes\n",
9816 		    (uint_t)sizeof (old_ip_mib)));
9817 	}
9818 
9819 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9820 	ip3dbg(("ip_snmp_get_mib2_ip: level %d, name %d, len %d\n",
9821 	    (int)optp->level, (int)optp->name, (int)optp->len));
9822 	qreply(q, mpctl);
9823 	return (mp2ctl);
9824 }
9825 
9826 /* Per interface IPv4 statistics */
9827 static mblk_t *
ip_snmp_get_mib2_ip_traffic_stats(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)9828 ip_snmp_get_mib2_ip_traffic_stats(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9829     boolean_t legacy_req)
9830 {
9831 	struct opthdr		*optp;
9832 	mblk_t			*mp2ctl;
9833 	ill_t			*ill;
9834 	ill_walk_context_t	ctx;
9835 	mblk_t			*mp_tail = NULL;
9836 	mib2_ipIfStatsEntry_t	global_ip_mib;
9837 	mib2_ipAddrEntry_t	mae;
9838 
9839 	/*
9840 	 * Make a copy of the original message
9841 	 */
9842 	mp2ctl = copymsg(mpctl);
9843 
9844 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9845 	optp->level = MIB2_IP;
9846 	optp->name = MIB2_IP_TRAFFIC_STATS;
9847 	/* Include "unknown interface" ip_mib */
9848 	ipst->ips_ip_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv4;
9849 	ipst->ips_ip_mib.ipIfStatsIfIndex =
9850 	    MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
9851 	SET_MIB(ipst->ips_ip_mib.ipIfStatsForwarding,
9852 	    (ipst->ips_ip_forwarding ? 1 : 2));
9853 	SET_MIB(ipst->ips_ip_mib.ipIfStatsDefaultTTL,
9854 	    (uint32_t)ipst->ips_ip_def_ttl);
9855 	SET_MIB(ipst->ips_ip_mib.ipIfStatsEntrySize,
9856 	    sizeof (mib2_ipIfStatsEntry_t));
9857 	SET_MIB(ipst->ips_ip_mib.ipIfStatsAddrEntrySize,
9858 	    sizeof (mib2_ipAddrEntry_t));
9859 	SET_MIB(ipst->ips_ip_mib.ipIfStatsRouteEntrySize,
9860 	    sizeof (mib2_ipRouteEntry_t));
9861 	SET_MIB(ipst->ips_ip_mib.ipIfStatsNetToMediaEntrySize,
9862 	    sizeof (mib2_ipNetToMediaEntry_t));
9863 	SET_MIB(ipst->ips_ip_mib.ipIfStatsMemberEntrySize,
9864 	    sizeof (ip_member_t));
9865 	SET_MIB(ipst->ips_ip_mib.ipIfStatsGroupSourceEntrySize,
9866 	    sizeof (ip_grpsrc_t));
9867 
9868 	bcopy(&ipst->ips_ip_mib, &global_ip_mib, sizeof (global_ip_mib));
9869 
9870 	if (legacy_req) {
9871 		SET_MIB(global_ip_mib.ipIfStatsAddrEntrySize,
9872 		    LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t));
9873 	}
9874 
9875 	if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9876 	    (char *)&global_ip_mib, (int)sizeof (global_ip_mib))) {
9877 		ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9878 		    "failed to allocate %u bytes\n",
9879 		    (uint_t)sizeof (global_ip_mib)));
9880 	}
9881 
9882 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9883 	ill = ILL_START_WALK_V4(&ctx, ipst);
9884 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9885 		ill->ill_ip_mib->ipIfStatsIfIndex =
9886 		    ill->ill_phyint->phyint_ifindex;
9887 		SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
9888 		    (ipst->ips_ip_forwarding ? 1 : 2));
9889 		SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultTTL,
9890 		    (uint32_t)ipst->ips_ip_def_ttl);
9891 
9892 		ip_mib2_add_ip_stats(&global_ip_mib, ill->ill_ip_mib);
9893 		if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9894 		    (char *)ill->ill_ip_mib,
9895 		    (int)sizeof (*ill->ill_ip_mib))) {
9896 			ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9897 			    "failed to allocate %u bytes\n",
9898 			    (uint_t)sizeof (*ill->ill_ip_mib)));
9899 		}
9900 	}
9901 	rw_exit(&ipst->ips_ill_g_lock);
9902 
9903 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9904 	ip3dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9905 	    "level %d, name %d, len %d\n",
9906 	    (int)optp->level, (int)optp->name, (int)optp->len));
9907 	qreply(q, mpctl);
9908 
9909 	if (mp2ctl == NULL)
9910 		return (NULL);
9911 
9912 	return (ip_snmp_get_mib2_ip(q, mp2ctl, &global_ip_mib, ipst,
9913 	    legacy_req));
9914 }
9915 
9916 /* Global IPv4 ICMP statistics */
9917 static mblk_t *
ip_snmp_get_mib2_icmp(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)9918 ip_snmp_get_mib2_icmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9919 {
9920 	struct opthdr		*optp;
9921 	mblk_t			*mp2ctl;
9922 
9923 	/*
9924 	 * Make a copy of the original message
9925 	 */
9926 	mp2ctl = copymsg(mpctl);
9927 
9928 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9929 	optp->level = MIB2_ICMP;
9930 	optp->name = 0;
9931 	if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_icmp_mib,
9932 	    (int)sizeof (ipst->ips_icmp_mib))) {
9933 		ip1dbg(("ip_snmp_get_mib2_icmp: failed to allocate %u bytes\n",
9934 		    (uint_t)sizeof (ipst->ips_icmp_mib)));
9935 	}
9936 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9937 	ip3dbg(("ip_snmp_get_mib2_icmp: level %d, name %d, len %d\n",
9938 	    (int)optp->level, (int)optp->name, (int)optp->len));
9939 	qreply(q, mpctl);
9940 	return (mp2ctl);
9941 }
9942 
9943 /* Global IPv4 IGMP statistics */
9944 static mblk_t *
ip_snmp_get_mib2_igmp(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)9945 ip_snmp_get_mib2_igmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9946 {
9947 	struct opthdr		*optp;
9948 	mblk_t			*mp2ctl;
9949 
9950 	/*
9951 	 * make a copy of the original message
9952 	 */
9953 	mp2ctl = copymsg(mpctl);
9954 
9955 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9956 	optp->level = EXPER_IGMP;
9957 	optp->name = 0;
9958 	if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_igmpstat,
9959 	    (int)sizeof (ipst->ips_igmpstat))) {
9960 		ip1dbg(("ip_snmp_get_mib2_igmp: failed to allocate %u bytes\n",
9961 		    (uint_t)sizeof (ipst->ips_igmpstat)));
9962 	}
9963 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9964 	ip3dbg(("ip_snmp_get_mib2_igmp: level %d, name %d, len %d\n",
9965 	    (int)optp->level, (int)optp->name, (int)optp->len));
9966 	qreply(q, mpctl);
9967 	return (mp2ctl);
9968 }
9969 
9970 /* Global IPv4 Multicast Routing statistics */
9971 static mblk_t *
ip_snmp_get_mib2_multi(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)9972 ip_snmp_get_mib2_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9973 {
9974 	struct opthdr		*optp;
9975 	mblk_t			*mp2ctl;
9976 
9977 	/*
9978 	 * make a copy of the original message
9979 	 */
9980 	mp2ctl = copymsg(mpctl);
9981 
9982 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9983 	optp->level = EXPER_DVMRP;
9984 	optp->name = 0;
9985 	if (!ip_mroute_stats(mpctl->b_cont, ipst)) {
9986 		ip0dbg(("ip_mroute_stats: failed\n"));
9987 	}
9988 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9989 	ip3dbg(("ip_snmp_get_mib2_multi: level %d, name %d, len %d\n",
9990 	    (int)optp->level, (int)optp->name, (int)optp->len));
9991 	qreply(q, mpctl);
9992 	return (mp2ctl);
9993 }
9994 
9995 /* IPv4 address information */
9996 static mblk_t *
ip_snmp_get_mib2_ip_addr(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)9997 ip_snmp_get_mib2_ip_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9998     boolean_t legacy_req)
9999 {
10000 	struct opthdr		*optp;
10001 	mblk_t			*mp2ctl;
10002 	mblk_t			*mp_tail = NULL;
10003 	ill_t			*ill;
10004 	ipif_t			*ipif;
10005 	uint_t			bitval;
10006 	mib2_ipAddrEntry_t	mae;
10007 	size_t			mae_size;
10008 	zoneid_t		zoneid;
10009 	ill_walk_context_t	ctx;
10010 
10011 	/*
10012 	 * make a copy of the original message
10013 	 */
10014 	mp2ctl = copymsg(mpctl);
10015 
10016 	mae_size = (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
10017 	    sizeof (mib2_ipAddrEntry_t);
10018 
10019 	/* ipAddrEntryTable */
10020 
10021 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10022 	optp->level = MIB2_IP;
10023 	optp->name = MIB2_IP_ADDR;
10024 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10025 
10026 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10027 	ill = ILL_START_WALK_V4(&ctx, ipst);
10028 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10029 		for (ipif = ill->ill_ipif; ipif != NULL;
10030 		    ipif = ipif->ipif_next) {
10031 			if (ipif->ipif_zoneid != zoneid &&
10032 			    ipif->ipif_zoneid != ALL_ZONES)
10033 				continue;
10034 			/* Sum of count from dead IRE_LO* and our current */
10035 			mae.ipAdEntInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10036 			if (ipif->ipif_ire_local != NULL) {
10037 				mae.ipAdEntInfo.ae_ibcnt +=
10038 				    ipif->ipif_ire_local->ire_ib_pkt_count;
10039 			}
10040 			mae.ipAdEntInfo.ae_obcnt = 0;
10041 			mae.ipAdEntInfo.ae_focnt = 0;
10042 
10043 			ipif_get_name(ipif, mae.ipAdEntIfIndex.o_bytes,
10044 			    OCTET_LENGTH);
10045 			mae.ipAdEntIfIndex.o_length =
10046 			    mi_strlen(mae.ipAdEntIfIndex.o_bytes);
10047 			mae.ipAdEntAddr = ipif->ipif_lcl_addr;
10048 			mae.ipAdEntNetMask = ipif->ipif_net_mask;
10049 			mae.ipAdEntInfo.ae_subnet = ipif->ipif_subnet;
10050 			mae.ipAdEntInfo.ae_subnet_len =
10051 			    ip_mask_to_plen(ipif->ipif_net_mask);
10052 			mae.ipAdEntInfo.ae_src_addr = ipif->ipif_lcl_addr;
10053 			for (bitval = 1;
10054 			    bitval &&
10055 			    !(bitval & ipif->ipif_brd_addr);
10056 			    bitval <<= 1)
10057 				noop;
10058 			mae.ipAdEntBcastAddr = bitval;
10059 			mae.ipAdEntReasmMaxSize = IP_MAXPACKET;
10060 			mae.ipAdEntInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10061 			mae.ipAdEntInfo.ae_metric  = ipif->ipif_ill->ill_metric;
10062 			mae.ipAdEntInfo.ae_broadcast_addr =
10063 			    ipif->ipif_brd_addr;
10064 			mae.ipAdEntInfo.ae_pp_dst_addr =
10065 			    ipif->ipif_pp_dst_addr;
10066 			mae.ipAdEntInfo.ae_flags = ipif->ipif_flags |
10067 			    ill->ill_flags | ill->ill_phyint->phyint_flags;
10068 			mae.ipAdEntRetransmitTime =
10069 			    ill->ill_reachable_retrans_time;
10070 
10071 			if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10072 			    (char *)&mae, (int)mae_size)) {
10073 				ip1dbg(("ip_snmp_get_mib2_ip_addr: failed to "
10074 				    "allocate %u bytes\n", (uint_t)mae_size));
10075 			}
10076 		}
10077 	}
10078 	rw_exit(&ipst->ips_ill_g_lock);
10079 
10080 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10081 	ip3dbg(("ip_snmp_get_mib2_ip_addr: level %d, name %d, len %d\n",
10082 	    (int)optp->level, (int)optp->name, (int)optp->len));
10083 	qreply(q, mpctl);
10084 	return (mp2ctl);
10085 }
10086 
10087 /* IPv6 address information */
10088 static mblk_t *
ip_snmp_get_mib2_ip6_addr(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)10089 ip_snmp_get_mib2_ip6_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10090     boolean_t legacy_req)
10091 {
10092 	struct opthdr		*optp;
10093 	mblk_t			*mp2ctl;
10094 	mblk_t			*mp_tail = NULL;
10095 	ill_t			*ill;
10096 	ipif_t			*ipif;
10097 	mib2_ipv6AddrEntry_t	mae6;
10098 	size_t			mae6_size;
10099 	zoneid_t		zoneid;
10100 	ill_walk_context_t	ctx;
10101 
10102 	/*
10103 	 * make a copy of the original message
10104 	 */
10105 	mp2ctl = copymsg(mpctl);
10106 
10107 	mae6_size = (legacy_req) ?
10108 	    LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t) :
10109 	    sizeof (mib2_ipv6AddrEntry_t);
10110 
10111 	/* ipv6AddrEntryTable */
10112 
10113 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10114 	optp->level = MIB2_IP6;
10115 	optp->name = MIB2_IP6_ADDR;
10116 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10117 
10118 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10119 	ill = ILL_START_WALK_V6(&ctx, ipst);
10120 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10121 		for (ipif = ill->ill_ipif; ipif != NULL;
10122 		    ipif = ipif->ipif_next) {
10123 			if (ipif->ipif_zoneid != zoneid &&
10124 			    ipif->ipif_zoneid != ALL_ZONES)
10125 				continue;
10126 			/* Sum of count from dead IRE_LO* and our current */
10127 			mae6.ipv6AddrInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10128 			if (ipif->ipif_ire_local != NULL) {
10129 				mae6.ipv6AddrInfo.ae_ibcnt +=
10130 				    ipif->ipif_ire_local->ire_ib_pkt_count;
10131 			}
10132 			mae6.ipv6AddrInfo.ae_obcnt = 0;
10133 			mae6.ipv6AddrInfo.ae_focnt = 0;
10134 
10135 			ipif_get_name(ipif, mae6.ipv6AddrIfIndex.o_bytes,
10136 			    OCTET_LENGTH);
10137 			mae6.ipv6AddrIfIndex.o_length =
10138 			    mi_strlen(mae6.ipv6AddrIfIndex.o_bytes);
10139 			mae6.ipv6AddrAddress = ipif->ipif_v6lcl_addr;
10140 			mae6.ipv6AddrPfxLength =
10141 			    ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
10142 			mae6.ipv6AddrInfo.ae_subnet = ipif->ipif_v6subnet;
10143 			mae6.ipv6AddrInfo.ae_subnet_len =
10144 			    mae6.ipv6AddrPfxLength;
10145 			mae6.ipv6AddrInfo.ae_src_addr = ipif->ipif_v6lcl_addr;
10146 
10147 			/* Type: stateless(1), stateful(2), unknown(3) */
10148 			if (ipif->ipif_flags & IPIF_ADDRCONF)
10149 				mae6.ipv6AddrType = 1;
10150 			else
10151 				mae6.ipv6AddrType = 2;
10152 			/* Anycast: true(1), false(2) */
10153 			if (ipif->ipif_flags & IPIF_ANYCAST)
10154 				mae6.ipv6AddrAnycastFlag = 1;
10155 			else
10156 				mae6.ipv6AddrAnycastFlag = 2;
10157 
10158 			/*
10159 			 * Address status: preferred(1), deprecated(2),
10160 			 * invalid(3), inaccessible(4), unknown(5)
10161 			 */
10162 			if (ipif->ipif_flags & IPIF_NOLOCAL)
10163 				mae6.ipv6AddrStatus = 3;
10164 			else if (ipif->ipif_flags & IPIF_DEPRECATED)
10165 				mae6.ipv6AddrStatus = 2;
10166 			else
10167 				mae6.ipv6AddrStatus = 1;
10168 			mae6.ipv6AddrInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10169 			mae6.ipv6AddrInfo.ae_metric  =
10170 			    ipif->ipif_ill->ill_metric;
10171 			mae6.ipv6AddrInfo.ae_pp_dst_addr =
10172 			    ipif->ipif_v6pp_dst_addr;
10173 			mae6.ipv6AddrInfo.ae_flags = ipif->ipif_flags |
10174 			    ill->ill_flags | ill->ill_phyint->phyint_flags;
10175 			mae6.ipv6AddrReasmMaxSize = IP_MAXPACKET;
10176 			mae6.ipv6AddrIdentifier = ill->ill_token;
10177 			mae6.ipv6AddrIdentifierLen = ill->ill_token_length;
10178 			mae6.ipv6AddrReachableTime = ill->ill_reachable_time;
10179 			mae6.ipv6AddrRetransmitTime =
10180 			    ill->ill_reachable_retrans_time;
10181 			if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10182 			    (char *)&mae6, (int)mae6_size)) {
10183 				ip1dbg(("ip_snmp_get_mib2_ip6_addr: failed to "
10184 				    "allocate %u bytes\n",
10185 				    (uint_t)mae6_size));
10186 			}
10187 		}
10188 	}
10189 	rw_exit(&ipst->ips_ill_g_lock);
10190 
10191 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10192 	ip3dbg(("ip_snmp_get_mib2_ip6_addr: level %d, name %d, len %d\n",
10193 	    (int)optp->level, (int)optp->name, (int)optp->len));
10194 	qreply(q, mpctl);
10195 	return (mp2ctl);
10196 }
10197 
10198 /* IPv4 multicast group membership. */
10199 static mblk_t *
ip_snmp_get_mib2_ip_group_mem(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10200 ip_snmp_get_mib2_ip_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10201 {
10202 	struct opthdr		*optp;
10203 	mblk_t			*mp2ctl;
10204 	ill_t			*ill;
10205 	ipif_t			*ipif;
10206 	ilm_t			*ilm;
10207 	ip_member_t		ipm;
10208 	mblk_t			*mp_tail = NULL;
10209 	ill_walk_context_t	ctx;
10210 	zoneid_t		zoneid;
10211 
10212 	/*
10213 	 * make a copy of the original message
10214 	 */
10215 	mp2ctl = copymsg(mpctl);
10216 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10217 
10218 	/* ipGroupMember table */
10219 	optp = (struct opthdr *)&mpctl->b_rptr[
10220 	    sizeof (struct T_optmgmt_ack)];
10221 	optp->level = MIB2_IP;
10222 	optp->name = EXPER_IP_GROUP_MEMBERSHIP;
10223 
10224 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10225 	ill = ILL_START_WALK_V4(&ctx, ipst);
10226 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10227 		/* Make sure the ill isn't going away. */
10228 		if (!ill_check_and_refhold(ill))
10229 			continue;
10230 		rw_exit(&ipst->ips_ill_g_lock);
10231 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10232 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10233 			if (ilm->ilm_zoneid != zoneid &&
10234 			    ilm->ilm_zoneid != ALL_ZONES)
10235 				continue;
10236 
10237 			/* Is there an ipif for ilm_ifaddr? */
10238 			for (ipif = ill->ill_ipif; ipif != NULL;
10239 			    ipif = ipif->ipif_next) {
10240 				if (!IPIF_IS_CONDEMNED(ipif) &&
10241 				    ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10242 				    ilm->ilm_ifaddr != INADDR_ANY)
10243 					break;
10244 			}
10245 			if (ipif != NULL) {
10246 				ipif_get_name(ipif,
10247 				    ipm.ipGroupMemberIfIndex.o_bytes,
10248 				    OCTET_LENGTH);
10249 			} else {
10250 				ill_get_name(ill,
10251 				    ipm.ipGroupMemberIfIndex.o_bytes,
10252 				    OCTET_LENGTH);
10253 			}
10254 			ipm.ipGroupMemberIfIndex.o_length =
10255 			    mi_strlen(ipm.ipGroupMemberIfIndex.o_bytes);
10256 
10257 			ipm.ipGroupMemberAddress = ilm->ilm_addr;
10258 			ipm.ipGroupMemberRefCnt = ilm->ilm_refcnt;
10259 			ipm.ipGroupMemberFilterMode = ilm->ilm_fmode;
10260 			if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10261 			    (char *)&ipm, (int)sizeof (ipm))) {
10262 				ip1dbg(("ip_snmp_get_mib2_ip_group: "
10263 				    "failed to allocate %u bytes\n",
10264 				    (uint_t)sizeof (ipm)));
10265 			}
10266 		}
10267 		rw_exit(&ill->ill_mcast_lock);
10268 		ill_refrele(ill);
10269 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10270 	}
10271 	rw_exit(&ipst->ips_ill_g_lock);
10272 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10273 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10274 	    (int)optp->level, (int)optp->name, (int)optp->len));
10275 	qreply(q, mpctl);
10276 	return (mp2ctl);
10277 }
10278 
10279 /* IPv6 multicast group membership. */
10280 static mblk_t *
ip_snmp_get_mib2_ip6_group_mem(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10281 ip_snmp_get_mib2_ip6_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10282 {
10283 	struct opthdr		*optp;
10284 	mblk_t			*mp2ctl;
10285 	ill_t			*ill;
10286 	ilm_t			*ilm;
10287 	ipv6_member_t		ipm6;
10288 	mblk_t			*mp_tail = NULL;
10289 	ill_walk_context_t	ctx;
10290 	zoneid_t		zoneid;
10291 
10292 	/*
10293 	 * make a copy of the original message
10294 	 */
10295 	mp2ctl = copymsg(mpctl);
10296 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10297 
10298 	/* ip6GroupMember table */
10299 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10300 	optp->level = MIB2_IP6;
10301 	optp->name = EXPER_IP6_GROUP_MEMBERSHIP;
10302 
10303 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10304 	ill = ILL_START_WALK_V6(&ctx, ipst);
10305 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10306 		/* Make sure the ill isn't going away. */
10307 		if (!ill_check_and_refhold(ill))
10308 			continue;
10309 		rw_exit(&ipst->ips_ill_g_lock);
10310 		/*
10311 		 * Normally we don't have any members on under IPMP interfaces.
10312 		 * We report them as a debugging aid.
10313 		 */
10314 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10315 		ipm6.ipv6GroupMemberIfIndex = ill->ill_phyint->phyint_ifindex;
10316 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10317 			if (ilm->ilm_zoneid != zoneid &&
10318 			    ilm->ilm_zoneid != ALL_ZONES)
10319 				continue;	/* not this zone */
10320 			ipm6.ipv6GroupMemberAddress = ilm->ilm_v6addr;
10321 			ipm6.ipv6GroupMemberRefCnt = ilm->ilm_refcnt;
10322 			ipm6.ipv6GroupMemberFilterMode = ilm->ilm_fmode;
10323 			if (!snmp_append_data2(mpctl->b_cont,
10324 			    &mp_tail,
10325 			    (char *)&ipm6, (int)sizeof (ipm6))) {
10326 				ip1dbg(("ip_snmp_get_mib2_ip6_group: "
10327 				    "failed to allocate %u bytes\n",
10328 				    (uint_t)sizeof (ipm6)));
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 
10337 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10338 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10339 	    (int)optp->level, (int)optp->name, (int)optp->len));
10340 	qreply(q, mpctl);
10341 	return (mp2ctl);
10342 }
10343 
10344 /* IP multicast filtered sources */
10345 static mblk_t *
ip_snmp_get_mib2_ip_group_src(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10346 ip_snmp_get_mib2_ip_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10347 {
10348 	struct opthdr		*optp;
10349 	mblk_t			*mp2ctl;
10350 	ill_t			*ill;
10351 	ipif_t			*ipif;
10352 	ilm_t			*ilm;
10353 	ip_grpsrc_t		ips;
10354 	mblk_t			*mp_tail = NULL;
10355 	ill_walk_context_t	ctx;
10356 	zoneid_t		zoneid;
10357 	int			i;
10358 	slist_t			*sl;
10359 
10360 	/*
10361 	 * make a copy of the original message
10362 	 */
10363 	mp2ctl = copymsg(mpctl);
10364 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10365 
10366 	/* ipGroupSource table */
10367 	optp = (struct opthdr *)&mpctl->b_rptr[
10368 	    sizeof (struct T_optmgmt_ack)];
10369 	optp->level = MIB2_IP;
10370 	optp->name = EXPER_IP_GROUP_SOURCES;
10371 
10372 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10373 	ill = ILL_START_WALK_V4(&ctx, ipst);
10374 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10375 		/* Make sure the ill isn't going away. */
10376 		if (!ill_check_and_refhold(ill))
10377 			continue;
10378 		rw_exit(&ipst->ips_ill_g_lock);
10379 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10380 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10381 			sl = ilm->ilm_filter;
10382 			if (ilm->ilm_zoneid != zoneid &&
10383 			    ilm->ilm_zoneid != ALL_ZONES)
10384 				continue;
10385 			if (SLIST_IS_EMPTY(sl))
10386 				continue;
10387 
10388 			/* Is there an ipif for ilm_ifaddr? */
10389 			for (ipif = ill->ill_ipif; ipif != NULL;
10390 			    ipif = ipif->ipif_next) {
10391 				if (!IPIF_IS_CONDEMNED(ipif) &&
10392 				    ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10393 				    ilm->ilm_ifaddr != INADDR_ANY)
10394 					break;
10395 			}
10396 			if (ipif != NULL) {
10397 				ipif_get_name(ipif,
10398 				    ips.ipGroupSourceIfIndex.o_bytes,
10399 				    OCTET_LENGTH);
10400 			} else {
10401 				ill_get_name(ill,
10402 				    ips.ipGroupSourceIfIndex.o_bytes,
10403 				    OCTET_LENGTH);
10404 			}
10405 			ips.ipGroupSourceIfIndex.o_length =
10406 			    mi_strlen(ips.ipGroupSourceIfIndex.o_bytes);
10407 
10408 			ips.ipGroupSourceGroup = ilm->ilm_addr;
10409 			for (i = 0; i < sl->sl_numsrc; i++) {
10410 				if (!IN6_IS_ADDR_V4MAPPED(&sl->sl_addr[i]))
10411 					continue;
10412 				IN6_V4MAPPED_TO_IPADDR(&sl->sl_addr[i],
10413 				    ips.ipGroupSourceAddress);
10414 				if (snmp_append_data2(mpctl->b_cont, &mp_tail,
10415 				    (char *)&ips, (int)sizeof (ips)) == 0) {
10416 					ip1dbg(("ip_snmp_get_mib2_ip_group_src:"
10417 					    " failed to allocate %u bytes\n",
10418 					    (uint_t)sizeof (ips)));
10419 				}
10420 			}
10421 		}
10422 		rw_exit(&ill->ill_mcast_lock);
10423 		ill_refrele(ill);
10424 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10425 	}
10426 	rw_exit(&ipst->ips_ill_g_lock);
10427 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10428 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10429 	    (int)optp->level, (int)optp->name, (int)optp->len));
10430 	qreply(q, mpctl);
10431 	return (mp2ctl);
10432 }
10433 
10434 /* IPv6 multicast filtered sources. */
10435 static mblk_t *
ip_snmp_get_mib2_ip6_group_src(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10436 ip_snmp_get_mib2_ip6_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10437 {
10438 	struct opthdr		*optp;
10439 	mblk_t			*mp2ctl;
10440 	ill_t			*ill;
10441 	ilm_t			*ilm;
10442 	ipv6_grpsrc_t		ips6;
10443 	mblk_t			*mp_tail = NULL;
10444 	ill_walk_context_t	ctx;
10445 	zoneid_t		zoneid;
10446 	int			i;
10447 	slist_t			*sl;
10448 
10449 	/*
10450 	 * make a copy of the original message
10451 	 */
10452 	mp2ctl = copymsg(mpctl);
10453 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10454 
10455 	/* ip6GroupMember table */
10456 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10457 	optp->level = MIB2_IP6;
10458 	optp->name = EXPER_IP6_GROUP_SOURCES;
10459 
10460 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10461 	ill = ILL_START_WALK_V6(&ctx, ipst);
10462 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10463 		/* Make sure the ill isn't going away. */
10464 		if (!ill_check_and_refhold(ill))
10465 			continue;
10466 		rw_exit(&ipst->ips_ill_g_lock);
10467 		/*
10468 		 * Normally we don't have any members on under IPMP interfaces.
10469 		 * We report them as a debugging aid.
10470 		 */
10471 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10472 		ips6.ipv6GroupSourceIfIndex = ill->ill_phyint->phyint_ifindex;
10473 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10474 			sl = ilm->ilm_filter;
10475 			if (ilm->ilm_zoneid != zoneid &&
10476 			    ilm->ilm_zoneid != ALL_ZONES)
10477 				continue;
10478 			if (SLIST_IS_EMPTY(sl))
10479 				continue;
10480 			ips6.ipv6GroupSourceGroup = ilm->ilm_v6addr;
10481 			for (i = 0; i < sl->sl_numsrc; i++) {
10482 				ips6.ipv6GroupSourceAddress = sl->sl_addr[i];
10483 				if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10484 				    (char *)&ips6, (int)sizeof (ips6))) {
10485 					ip1dbg(("ip_snmp_get_mib2_ip6_"
10486 					    "group_src: failed to allocate "
10487 					    "%u bytes\n",
10488 					    (uint_t)sizeof (ips6)));
10489 				}
10490 			}
10491 		}
10492 		rw_exit(&ill->ill_mcast_lock);
10493 		ill_refrele(ill);
10494 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10495 	}
10496 	rw_exit(&ipst->ips_ill_g_lock);
10497 
10498 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10499 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10500 	    (int)optp->level, (int)optp->name, (int)optp->len));
10501 	qreply(q, mpctl);
10502 	return (mp2ctl);
10503 }
10504 
10505 /* Multicast routing virtual interface table. */
10506 static mblk_t *
ip_snmp_get_mib2_virt_multi(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10507 ip_snmp_get_mib2_virt_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10508 {
10509 	struct opthdr		*optp;
10510 	mblk_t			*mp2ctl;
10511 
10512 	/*
10513 	 * make a copy of the original message
10514 	 */
10515 	mp2ctl = copymsg(mpctl);
10516 
10517 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10518 	optp->level = EXPER_DVMRP;
10519 	optp->name = EXPER_DVMRP_VIF;
10520 	if (!ip_mroute_vif(mpctl->b_cont, ipst)) {
10521 		ip0dbg(("ip_mroute_vif: failed\n"));
10522 	}
10523 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10524 	ip3dbg(("ip_snmp_get_mib2_virt_multi: level %d, name %d, len %d\n",
10525 	    (int)optp->level, (int)optp->name, (int)optp->len));
10526 	qreply(q, mpctl);
10527 	return (mp2ctl);
10528 }
10529 
10530 /* Multicast routing table. */
10531 static mblk_t *
ip_snmp_get_mib2_multi_rtable(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10532 ip_snmp_get_mib2_multi_rtable(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10533 {
10534 	struct opthdr		*optp;
10535 	mblk_t			*mp2ctl;
10536 
10537 	/*
10538 	 * make a copy of the original message
10539 	 */
10540 	mp2ctl = copymsg(mpctl);
10541 
10542 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10543 	optp->level = EXPER_DVMRP;
10544 	optp->name = EXPER_DVMRP_MRT;
10545 	if (!ip_mroute_mrt(mpctl->b_cont, ipst)) {
10546 		ip0dbg(("ip_mroute_mrt: failed\n"));
10547 	}
10548 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10549 	ip3dbg(("ip_snmp_get_mib2_multi_rtable: level %d, name %d, len %d\n",
10550 	    (int)optp->level, (int)optp->name, (int)optp->len));
10551 	qreply(q, mpctl);
10552 	return (mp2ctl);
10553 }
10554 
10555 /*
10556  * Return ipRouteEntryTable, ipNetToMediaEntryTable, and ipRouteAttributeTable
10557  * in one IRE walk.
10558  */
10559 static mblk_t *
ip_snmp_get_mib2_ip_route_media(queue_t * q,mblk_t * mpctl,int level,ip_stack_t * ipst)10560 ip_snmp_get_mib2_ip_route_media(queue_t *q, mblk_t *mpctl, int level,
10561     ip_stack_t *ipst)
10562 {
10563 	struct opthdr	*optp;
10564 	mblk_t		*mp2ctl;	/* Returned */
10565 	mblk_t		*mp3ctl;	/* nettomedia */
10566 	mblk_t		*mp4ctl;	/* routeattrs */
10567 	iproutedata_t	ird;
10568 	zoneid_t	zoneid;
10569 
10570 	/*
10571 	 * make copies of the original message
10572 	 *	- mp2ctl is returned unchanged to the caller for his use
10573 	 *	- mpctl is sent upstream as ipRouteEntryTable
10574 	 *	- mp3ctl is sent upstream as ipNetToMediaEntryTable
10575 	 *	- mp4ctl is sent upstream as ipRouteAttributeTable
10576 	 */
10577 	mp2ctl = copymsg(mpctl);
10578 	mp3ctl = copymsg(mpctl);
10579 	mp4ctl = copymsg(mpctl);
10580 	if (mp3ctl == NULL || mp4ctl == NULL) {
10581 		freemsg(mp4ctl);
10582 		freemsg(mp3ctl);
10583 		freemsg(mp2ctl);
10584 		freemsg(mpctl);
10585 		return (NULL);
10586 	}
10587 
10588 	bzero(&ird, sizeof (ird));
10589 
10590 	ird.ird_route.lp_head = mpctl->b_cont;
10591 	ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10592 	ird.ird_attrs.lp_head = mp4ctl->b_cont;
10593 	/*
10594 	 * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10595 	 * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10596 	 * intended a temporary solution until a proper MIB API is provided
10597 	 * that provides complete filtering/caller-opt-in.
10598 	 */
10599 	if (level == EXPER_IP_AND_ALL_IRES)
10600 		ird.ird_flags |= IRD_REPORT_ALL;
10601 
10602 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10603 	ire_walk_v4(ip_snmp_get2_v4, &ird, zoneid, ipst);
10604 
10605 	/* ipRouteEntryTable in mpctl */
10606 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10607 	optp->level = MIB2_IP;
10608 	optp->name = MIB2_IP_ROUTE;
10609 	optp->len = msgdsize(ird.ird_route.lp_head);
10610 	ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10611 	    (int)optp->level, (int)optp->name, (int)optp->len));
10612 	qreply(q, mpctl);
10613 
10614 	/* ipNetToMediaEntryTable in mp3ctl */
10615 	ncec_walk(NULL, ip_snmp_get2_v4_media, &ird, ipst);
10616 
10617 	optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10618 	optp->level = MIB2_IP;
10619 	optp->name = MIB2_IP_MEDIA;
10620 	optp->len = msgdsize(ird.ird_netmedia.lp_head);
10621 	ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10622 	    (int)optp->level, (int)optp->name, (int)optp->len));
10623 	qreply(q, mp3ctl);
10624 
10625 	/* ipRouteAttributeTable in mp4ctl */
10626 	optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10627 	optp->level = MIB2_IP;
10628 	optp->name = EXPER_IP_RTATTR;
10629 	optp->len = msgdsize(ird.ird_attrs.lp_head);
10630 	ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10631 	    (int)optp->level, (int)optp->name, (int)optp->len));
10632 	if (optp->len == 0)
10633 		freemsg(mp4ctl);
10634 	else
10635 		qreply(q, mp4ctl);
10636 
10637 	return (mp2ctl);
10638 }
10639 
10640 /*
10641  * Return ipv6RouteEntryTable and ipv6RouteAttributeTable in one IRE walk, and
10642  * ipv6NetToMediaEntryTable in an NDP walk.
10643  */
10644 static mblk_t *
ip_snmp_get_mib2_ip6_route_media(queue_t * q,mblk_t * mpctl,int level,ip_stack_t * ipst)10645 ip_snmp_get_mib2_ip6_route_media(queue_t *q, mblk_t *mpctl, int level,
10646     ip_stack_t *ipst)
10647 {
10648 	struct opthdr	*optp;
10649 	mblk_t		*mp2ctl;	/* Returned */
10650 	mblk_t		*mp3ctl;	/* nettomedia */
10651 	mblk_t		*mp4ctl;	/* routeattrs */
10652 	iproutedata_t	ird;
10653 	zoneid_t	zoneid;
10654 
10655 	/*
10656 	 * make copies of the original message
10657 	 *	- mp2ctl is returned unchanged to the caller for his use
10658 	 *	- mpctl is sent upstream as ipv6RouteEntryTable
10659 	 *	- mp3ctl is sent upstream as ipv6NetToMediaEntryTable
10660 	 *	- mp4ctl is sent upstream as ipv6RouteAttributeTable
10661 	 */
10662 	mp2ctl = copymsg(mpctl);
10663 	mp3ctl = copymsg(mpctl);
10664 	mp4ctl = copymsg(mpctl);
10665 	if (mp3ctl == NULL || mp4ctl == NULL) {
10666 		freemsg(mp4ctl);
10667 		freemsg(mp3ctl);
10668 		freemsg(mp2ctl);
10669 		freemsg(mpctl);
10670 		return (NULL);
10671 	}
10672 
10673 	bzero(&ird, sizeof (ird));
10674 
10675 	ird.ird_route.lp_head = mpctl->b_cont;
10676 	ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10677 	ird.ird_attrs.lp_head = mp4ctl->b_cont;
10678 	/*
10679 	 * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10680 	 * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10681 	 * intended a temporary solution until a proper MIB API is provided
10682 	 * that provides complete filtering/caller-opt-in.
10683 	 */
10684 	if (level == EXPER_IP_AND_ALL_IRES)
10685 		ird.ird_flags |= IRD_REPORT_ALL;
10686 
10687 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10688 	ire_walk_v6(ip_snmp_get2_v6_route, &ird, zoneid, ipst);
10689 
10690 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10691 	optp->level = MIB2_IP6;
10692 	optp->name = MIB2_IP6_ROUTE;
10693 	optp->len = msgdsize(ird.ird_route.lp_head);
10694 	ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10695 	    (int)optp->level, (int)optp->name, (int)optp->len));
10696 	qreply(q, mpctl);
10697 
10698 	/* ipv6NetToMediaEntryTable in mp3ctl */
10699 	ncec_walk(NULL, ip_snmp_get2_v6_media, &ird, ipst);
10700 
10701 	optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10702 	optp->level = MIB2_IP6;
10703 	optp->name = MIB2_IP6_MEDIA;
10704 	optp->len = msgdsize(ird.ird_netmedia.lp_head);
10705 	ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10706 	    (int)optp->level, (int)optp->name, (int)optp->len));
10707 	qreply(q, mp3ctl);
10708 
10709 	/* ipv6RouteAttributeTable in mp4ctl */
10710 	optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10711 	optp->level = MIB2_IP6;
10712 	optp->name = EXPER_IP_RTATTR;
10713 	optp->len = msgdsize(ird.ird_attrs.lp_head);
10714 	ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10715 	    (int)optp->level, (int)optp->name, (int)optp->len));
10716 	if (optp->len == 0)
10717 		freemsg(mp4ctl);
10718 	else
10719 		qreply(q, mp4ctl);
10720 
10721 	return (mp2ctl);
10722 }
10723 
10724 /*
10725  * IPv6 mib: One per ill
10726  */
10727 static mblk_t *
ip_snmp_get_mib2_ip6(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)10728 ip_snmp_get_mib2_ip6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10729     boolean_t legacy_req)
10730 {
10731 	struct opthdr		*optp;
10732 	mblk_t			*mp2ctl;
10733 	ill_t			*ill;
10734 	ill_walk_context_t	ctx;
10735 	mblk_t			*mp_tail = NULL;
10736 	mib2_ipv6AddrEntry_t	mae6;
10737 	mib2_ipIfStatsEntry_t	*ise;
10738 	size_t			ise_size, iae_size;
10739 
10740 	/*
10741 	 * Make a copy of the original message
10742 	 */
10743 	mp2ctl = copymsg(mpctl);
10744 
10745 	/* fixed length IPv6 structure ... */
10746 
10747 	if (legacy_req) {
10748 		ise_size = LEGACY_MIB_SIZE(&ipst->ips_ip6_mib,
10749 		    mib2_ipIfStatsEntry_t);
10750 		iae_size = LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t);
10751 	} else {
10752 		ise_size = sizeof (mib2_ipIfStatsEntry_t);
10753 		iae_size = sizeof (mib2_ipv6AddrEntry_t);
10754 	}
10755 
10756 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10757 	optp->level = MIB2_IP6;
10758 	optp->name = 0;
10759 	/* Include "unknown interface" ip6_mib */
10760 	ipst->ips_ip6_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv6;
10761 	ipst->ips_ip6_mib.ipIfStatsIfIndex =
10762 	    MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
10763 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsForwarding,
10764 	    ipst->ips_ipv6_forwarding ? 1 : 2);
10765 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsDefaultHopLimit,
10766 	    ipst->ips_ipv6_def_hops);
10767 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsEntrySize,
10768 	    sizeof (mib2_ipIfStatsEntry_t));
10769 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsAddrEntrySize,
10770 	    sizeof (mib2_ipv6AddrEntry_t));
10771 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsRouteEntrySize,
10772 	    sizeof (mib2_ipv6RouteEntry_t));
10773 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsNetToMediaEntrySize,
10774 	    sizeof (mib2_ipv6NetToMediaEntry_t));
10775 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsMemberEntrySize,
10776 	    sizeof (ipv6_member_t));
10777 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsGroupSourceEntrySize,
10778 	    sizeof (ipv6_grpsrc_t));
10779 
10780 	/*
10781 	 * Synchronize 64- and 32-bit counters
10782 	 */
10783 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInReceives,
10784 	    ipIfStatsHCInReceives);
10785 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInDelivers,
10786 	    ipIfStatsHCInDelivers);
10787 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutRequests,
10788 	    ipIfStatsHCOutRequests);
10789 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutForwDatagrams,
10790 	    ipIfStatsHCOutForwDatagrams);
10791 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutMcastPkts,
10792 	    ipIfStatsHCOutMcastPkts);
10793 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInMcastPkts,
10794 	    ipIfStatsHCInMcastPkts);
10795 
10796 	if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10797 	    (char *)&ipst->ips_ip6_mib, (int)ise_size)) {
10798 		ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate %u bytes\n",
10799 		    (uint_t)ise_size));
10800 	} else if (legacy_req) {
10801 		/* Adjust the EntrySize fields for legacy requests. */
10802 		ise =
10803 		    (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr - (int)ise_size);
10804 		SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10805 		SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10806 	}
10807 
10808 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10809 	ill = ILL_START_WALK_V6(&ctx, ipst);
10810 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10811 		ill->ill_ip_mib->ipIfStatsIfIndex =
10812 		    ill->ill_phyint->phyint_ifindex;
10813 		SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
10814 		    ipst->ips_ipv6_forwarding ? 1 : 2);
10815 		SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultHopLimit,
10816 		    ill->ill_max_hops);
10817 
10818 		/*
10819 		 * Synchronize 64- and 32-bit counters
10820 		 */
10821 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInReceives,
10822 		    ipIfStatsHCInReceives);
10823 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInDelivers,
10824 		    ipIfStatsHCInDelivers);
10825 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutRequests,
10826 		    ipIfStatsHCOutRequests);
10827 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutForwDatagrams,
10828 		    ipIfStatsHCOutForwDatagrams);
10829 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutMcastPkts,
10830 		    ipIfStatsHCOutMcastPkts);
10831 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInMcastPkts,
10832 		    ipIfStatsHCInMcastPkts);
10833 
10834 		if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10835 		    (char *)ill->ill_ip_mib, (int)ise_size)) {
10836 			ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate "
10837 			"%u bytes\n", (uint_t)ise_size));
10838 		} else if (legacy_req) {
10839 			/* Adjust the EntrySize fields for legacy requests. */
10840 			ise = (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr -
10841 			    (int)ise_size);
10842 			SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10843 			SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10844 		}
10845 	}
10846 	rw_exit(&ipst->ips_ill_g_lock);
10847 
10848 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10849 	ip3dbg(("ip_snmp_get_mib2_ip6: level %d, name %d, len %d\n",
10850 	    (int)optp->level, (int)optp->name, (int)optp->len));
10851 	qreply(q, mpctl);
10852 	return (mp2ctl);
10853 }
10854 
10855 /*
10856  * ICMPv6 mib: One per ill
10857  */
10858 static mblk_t *
ip_snmp_get_mib2_icmp6(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10859 ip_snmp_get_mib2_icmp6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10860 {
10861 	struct opthdr		*optp;
10862 	mblk_t			*mp2ctl;
10863 	ill_t			*ill;
10864 	ill_walk_context_t	ctx;
10865 	mblk_t			*mp_tail = NULL;
10866 	/*
10867 	 * Make a copy of the original message
10868 	 */
10869 	mp2ctl = copymsg(mpctl);
10870 
10871 	/* fixed length ICMPv6 structure ... */
10872 
10873 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10874 	optp->level = MIB2_ICMP6;
10875 	optp->name = 0;
10876 	/* Include "unknown interface" icmp6_mib */
10877 	ipst->ips_icmp6_mib.ipv6IfIcmpIfIndex =
10878 	    MIB2_UNKNOWN_INTERFACE; /* netstat flag */
10879 	ipst->ips_icmp6_mib.ipv6IfIcmpEntrySize =
10880 	    sizeof (mib2_ipv6IfIcmpEntry_t);
10881 	if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10882 	    (char *)&ipst->ips_icmp6_mib,
10883 	    (int)sizeof (ipst->ips_icmp6_mib))) {
10884 		ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate %u bytes\n",
10885 		    (uint_t)sizeof (ipst->ips_icmp6_mib)));
10886 	}
10887 
10888 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10889 	ill = ILL_START_WALK_V6(&ctx, ipst);
10890 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10891 		ill->ill_icmp6_mib->ipv6IfIcmpIfIndex =
10892 		    ill->ill_phyint->phyint_ifindex;
10893 		if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10894 		    (char *)ill->ill_icmp6_mib,
10895 		    (int)sizeof (*ill->ill_icmp6_mib))) {
10896 			ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate "
10897 			    "%u bytes\n",
10898 			    (uint_t)sizeof (*ill->ill_icmp6_mib)));
10899 		}
10900 	}
10901 	rw_exit(&ipst->ips_ill_g_lock);
10902 
10903 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10904 	ip3dbg(("ip_snmp_get_mib2_icmp6: level %d, name %d, len %d\n",
10905 	    (int)optp->level, (int)optp->name, (int)optp->len));
10906 	qreply(q, mpctl);
10907 	return (mp2ctl);
10908 }
10909 
10910 /*
10911  * ire_walk routine to create both ipRouteEntryTable and
10912  * ipRouteAttributeTable in one IRE walk
10913  */
10914 static void
ip_snmp_get2_v4(ire_t * ire,iproutedata_t * ird)10915 ip_snmp_get2_v4(ire_t *ire, iproutedata_t *ird)
10916 {
10917 	ill_t				*ill;
10918 	mib2_ipRouteEntry_t		*re;
10919 	mib2_ipAttributeEntry_t		iaes;
10920 	tsol_ire_gw_secattr_t		*attrp;
10921 	tsol_gc_t			*gc = NULL;
10922 	tsol_gcgrp_t			*gcgrp = NULL;
10923 	ip_stack_t			*ipst = ire->ire_ipst;
10924 
10925 	ASSERT(ire->ire_ipversion == IPV4_VERSION);
10926 
10927 	if (!(ird->ird_flags & IRD_REPORT_ALL)) {
10928 		if (ire->ire_testhidden)
10929 			return;
10930 		if (ire->ire_type & IRE_IF_CLONE)
10931 			return;
10932 	}
10933 
10934 	if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
10935 		return;
10936 
10937 	if ((attrp = ire->ire_gw_secattr) != NULL) {
10938 		mutex_enter(&attrp->igsa_lock);
10939 		if ((gc = attrp->igsa_gc) != NULL) {
10940 			gcgrp = gc->gc_grp;
10941 			ASSERT(gcgrp != NULL);
10942 			rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
10943 		}
10944 		mutex_exit(&attrp->igsa_lock);
10945 	}
10946 	/*
10947 	 * Return all IRE types for route table... let caller pick and choose
10948 	 */
10949 	re->ipRouteDest = ire->ire_addr;
10950 	ill = ire->ire_ill;
10951 	re->ipRouteIfIndex.o_length = 0;
10952 	if (ill != NULL) {
10953 		ill_get_name(ill, re->ipRouteIfIndex.o_bytes, OCTET_LENGTH);
10954 		re->ipRouteIfIndex.o_length =
10955 		    mi_strlen(re->ipRouteIfIndex.o_bytes);
10956 	}
10957 	re->ipRouteMetric1 = -1;
10958 	re->ipRouteMetric2 = -1;
10959 	re->ipRouteMetric3 = -1;
10960 	re->ipRouteMetric4 = -1;
10961 
10962 	re->ipRouteNextHop = ire->ire_gateway_addr;
10963 	/* indirect(4), direct(3), or invalid(2) */
10964 	if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
10965 		re->ipRouteType = 2;
10966 	else if (ire->ire_type & IRE_ONLINK)
10967 		re->ipRouteType = 3;
10968 	else
10969 		re->ipRouteType = 4;
10970 
10971 	re->ipRouteProto = -1;
10972 	re->ipRouteAge = gethrestime_sec() - ire->ire_create_time;
10973 	re->ipRouteMask = ire->ire_mask;
10974 	re->ipRouteMetric5 = -1;
10975 	re->ipRouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
10976 	if (ire->ire_ill != NULL && re->ipRouteInfo.re_max_frag == 0)
10977 		re->ipRouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
10978 
10979 	re->ipRouteInfo.re_frag_flag	= 0;
10980 	re->ipRouteInfo.re_rtt		= 0;
10981 	re->ipRouteInfo.re_src_addr	= 0;
10982 	re->ipRouteInfo.re_ref		= ire->ire_refcnt;
10983 	re->ipRouteInfo.re_obpkt	= ire->ire_ob_pkt_count;
10984 	re->ipRouteInfo.re_ibpkt	= ire->ire_ib_pkt_count;
10985 	re->ipRouteInfo.re_flags	= ire->ire_flags;
10986 
10987 	/* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
10988 	if (ire->ire_type & IRE_INTERFACE) {
10989 		ire_t *child;
10990 
10991 		rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
10992 		child = ire->ire_dep_children;
10993 		while (child != NULL) {
10994 			re->ipRouteInfo.re_obpkt += child->ire_ob_pkt_count;
10995 			re->ipRouteInfo.re_ibpkt += child->ire_ib_pkt_count;
10996 			child = child->ire_dep_sib_next;
10997 		}
10998 		rw_exit(&ipst->ips_ire_dep_lock);
10999 	}
11000 
11001 	if (ire->ire_flags & RTF_DYNAMIC) {
11002 		re->ipRouteInfo.re_ire_type	= IRE_HOST_REDIRECT;
11003 	} else {
11004 		re->ipRouteInfo.re_ire_type	= ire->ire_type;
11005 	}
11006 
11007 	if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11008 	    (char *)re, (int)sizeof (*re))) {
11009 		ip1dbg(("ip_snmp_get2_v4: failed to allocate %u bytes\n",
11010 		    (uint_t)sizeof (*re)));
11011 	}
11012 
11013 	if (gc != NULL) {
11014 		iaes.iae_routeidx = ird->ird_idx;
11015 		iaes.iae_doi = gc->gc_db->gcdb_doi;
11016 		iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11017 
11018 		if (!snmp_append_data2(ird->ird_attrs.lp_head,
11019 		    &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11020 			ip1dbg(("ip_snmp_get2_v4: failed to allocate %u "
11021 			    "bytes\n", (uint_t)sizeof (iaes)));
11022 		}
11023 	}
11024 
11025 	/* bump route index for next pass */
11026 	ird->ird_idx++;
11027 
11028 	kmem_free(re, sizeof (*re));
11029 	if (gcgrp != NULL)
11030 		rw_exit(&gcgrp->gcgrp_rwlock);
11031 }
11032 
11033 /*
11034  * ire_walk routine to create ipv6RouteEntryTable and ipRouteEntryTable.
11035  */
11036 static void
ip_snmp_get2_v6_route(ire_t * ire,iproutedata_t * ird)11037 ip_snmp_get2_v6_route(ire_t *ire, iproutedata_t *ird)
11038 {
11039 	ill_t				*ill;
11040 	mib2_ipv6RouteEntry_t		*re;
11041 	mib2_ipAttributeEntry_t		iaes;
11042 	tsol_ire_gw_secattr_t		*attrp;
11043 	tsol_gc_t			*gc = NULL;
11044 	tsol_gcgrp_t			*gcgrp = NULL;
11045 	ip_stack_t			*ipst = ire->ire_ipst;
11046 
11047 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
11048 
11049 	if (!(ird->ird_flags & IRD_REPORT_ALL)) {
11050 		if (ire->ire_testhidden)
11051 			return;
11052 		if (ire->ire_type & IRE_IF_CLONE)
11053 			return;
11054 	}
11055 
11056 	if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
11057 		return;
11058 
11059 	if ((attrp = ire->ire_gw_secattr) != NULL) {
11060 		mutex_enter(&attrp->igsa_lock);
11061 		if ((gc = attrp->igsa_gc) != NULL) {
11062 			gcgrp = gc->gc_grp;
11063 			ASSERT(gcgrp != NULL);
11064 			rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
11065 		}
11066 		mutex_exit(&attrp->igsa_lock);
11067 	}
11068 	/*
11069 	 * Return all IRE types for route table... let caller pick and choose
11070 	 */
11071 	re->ipv6RouteDest = ire->ire_addr_v6;
11072 	re->ipv6RoutePfxLength = ip_mask_to_plen_v6(&ire->ire_mask_v6);
11073 	re->ipv6RouteIndex = 0;	/* Unique when multiple with same dest/plen */
11074 	re->ipv6RouteIfIndex.o_length = 0;
11075 	ill = ire->ire_ill;
11076 	if (ill != NULL) {
11077 		ill_get_name(ill, re->ipv6RouteIfIndex.o_bytes, OCTET_LENGTH);
11078 		re->ipv6RouteIfIndex.o_length =
11079 		    mi_strlen(re->ipv6RouteIfIndex.o_bytes);
11080 	}
11081 
11082 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
11083 
11084 	mutex_enter(&ire->ire_lock);
11085 	re->ipv6RouteNextHop = ire->ire_gateway_addr_v6;
11086 	mutex_exit(&ire->ire_lock);
11087 
11088 	/* remote(4), local(3), or discard(2) */
11089 	if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11090 		re->ipv6RouteType = 2;
11091 	else if (ire->ire_type & IRE_ONLINK)
11092 		re->ipv6RouteType = 3;
11093 	else
11094 		re->ipv6RouteType = 4;
11095 
11096 	re->ipv6RouteProtocol	= -1;
11097 	re->ipv6RoutePolicy	= 0;
11098 	re->ipv6RouteAge	= gethrestime_sec() - ire->ire_create_time;
11099 	re->ipv6RouteNextHopRDI	= 0;
11100 	re->ipv6RouteWeight	= 0;
11101 	re->ipv6RouteMetric	= 0;
11102 	re->ipv6RouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11103 	if (ire->ire_ill != NULL && re->ipv6RouteInfo.re_max_frag == 0)
11104 		re->ipv6RouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11105 
11106 	re->ipv6RouteInfo.re_frag_flag	= 0;
11107 	re->ipv6RouteInfo.re_rtt	= 0;
11108 	re->ipv6RouteInfo.re_src_addr	= ipv6_all_zeros;
11109 	re->ipv6RouteInfo.re_obpkt	= ire->ire_ob_pkt_count;
11110 	re->ipv6RouteInfo.re_ibpkt	= ire->ire_ib_pkt_count;
11111 	re->ipv6RouteInfo.re_ref	= ire->ire_refcnt;
11112 	re->ipv6RouteInfo.re_flags	= ire->ire_flags;
11113 
11114 	/* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11115 	if (ire->ire_type & IRE_INTERFACE) {
11116 		ire_t *child;
11117 
11118 		rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11119 		child = ire->ire_dep_children;
11120 		while (child != NULL) {
11121 			re->ipv6RouteInfo.re_obpkt += child->ire_ob_pkt_count;
11122 			re->ipv6RouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11123 			child = child->ire_dep_sib_next;
11124 		}
11125 		rw_exit(&ipst->ips_ire_dep_lock);
11126 	}
11127 	if (ire->ire_flags & RTF_DYNAMIC) {
11128 		re->ipv6RouteInfo.re_ire_type	= IRE_HOST_REDIRECT;
11129 	} else {
11130 		re->ipv6RouteInfo.re_ire_type	= ire->ire_type;
11131 	}
11132 
11133 	if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11134 	    (char *)re, (int)sizeof (*re))) {
11135 		ip1dbg(("ip_snmp_get2_v6: failed to allocate %u bytes\n",
11136 		    (uint_t)sizeof (*re)));
11137 	}
11138 
11139 	if (gc != NULL) {
11140 		iaes.iae_routeidx = ird->ird_idx;
11141 		iaes.iae_doi = gc->gc_db->gcdb_doi;
11142 		iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11143 
11144 		if (!snmp_append_data2(ird->ird_attrs.lp_head,
11145 		    &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11146 			ip1dbg(("ip_snmp_get2_v6: failed to allocate %u "
11147 			    "bytes\n", (uint_t)sizeof (iaes)));
11148 		}
11149 	}
11150 
11151 	/* bump route index for next pass */
11152 	ird->ird_idx++;
11153 
11154 	kmem_free(re, sizeof (*re));
11155 	if (gcgrp != NULL)
11156 		rw_exit(&gcgrp->gcgrp_rwlock);
11157 }
11158 
11159 /*
11160  * ncec_walk routine to create ipv6NetToMediaEntryTable
11161  */
11162 static int
ip_snmp_get2_v6_media(ncec_t * ncec,iproutedata_t * ird)11163 ip_snmp_get2_v6_media(ncec_t *ncec, iproutedata_t *ird)
11164 {
11165 	ill_t				*ill;
11166 	mib2_ipv6NetToMediaEntry_t	ntme;
11167 
11168 	ill = ncec->ncec_ill;
11169 	/* skip arpce entries, and loopback ncec entries */
11170 	if (ill->ill_isv6 == B_FALSE || ill->ill_net_type == IRE_LOOPBACK)
11171 		return (0);
11172 	/*
11173 	 * Neighbor cache entry attached to IRE with on-link
11174 	 * destination.
11175 	 * We report all IPMP groups on ncec_ill which is normally the upper.
11176 	 */
11177 	ntme.ipv6NetToMediaIfIndex = ill->ill_phyint->phyint_ifindex;
11178 	ntme.ipv6NetToMediaNetAddress = ncec->ncec_addr;
11179 	ntme.ipv6NetToMediaPhysAddress.o_length = ill->ill_phys_addr_length;
11180 	if (ncec->ncec_lladdr != NULL) {
11181 		bcopy(ncec->ncec_lladdr, ntme.ipv6NetToMediaPhysAddress.o_bytes,
11182 		    ntme.ipv6NetToMediaPhysAddress.o_length);
11183 	}
11184 	/*
11185 	 * Note: Returns ND_* states. Should be:
11186 	 * reachable(1), stale(2), delay(3), probe(4),
11187 	 * invalid(5), unknown(6)
11188 	 */
11189 	ntme.ipv6NetToMediaState = ncec->ncec_state;
11190 	ntme.ipv6NetToMediaLastUpdated = 0;
11191 
11192 	/* other(1), dynamic(2), static(3), local(4) */
11193 	if (NCE_MYADDR(ncec)) {
11194 		ntme.ipv6NetToMediaType = 4;
11195 	} else if (ncec->ncec_flags & NCE_F_PUBLISH) {
11196 		ntme.ipv6NetToMediaType = 1; /* proxy */
11197 	} else if (ncec->ncec_flags & NCE_F_STATIC) {
11198 		ntme.ipv6NetToMediaType = 3;
11199 	} else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST)) {
11200 		ntme.ipv6NetToMediaType = 1;
11201 	} else {
11202 		ntme.ipv6NetToMediaType = 2;
11203 	}
11204 
11205 	if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11206 	    &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11207 		ip1dbg(("ip_snmp_get2_v6_media: failed to allocate %u bytes\n",
11208 		    (uint_t)sizeof (ntme)));
11209 	}
11210 	return (0);
11211 }
11212 
11213 int
nce2ace(ncec_t * ncec)11214 nce2ace(ncec_t *ncec)
11215 {
11216 	int flags = 0;
11217 
11218 	if (NCE_ISREACHABLE(ncec))
11219 		flags |= ACE_F_RESOLVED;
11220 	if (ncec->ncec_flags & NCE_F_AUTHORITY)
11221 		flags |= ACE_F_AUTHORITY;
11222 	if (ncec->ncec_flags & NCE_F_PUBLISH)
11223 		flags |= ACE_F_PUBLISH;
11224 	if ((ncec->ncec_flags & NCE_F_NONUD) != 0)
11225 		flags |= ACE_F_PERMANENT;
11226 	if (NCE_MYADDR(ncec))
11227 		flags |= (ACE_F_MYADDR | ACE_F_AUTHORITY);
11228 	if (ncec->ncec_flags & NCE_F_UNVERIFIED)
11229 		flags |= ACE_F_UNVERIFIED;
11230 	if (ncec->ncec_flags & NCE_F_AUTHORITY)
11231 		flags |= ACE_F_AUTHORITY;
11232 	if (ncec->ncec_flags & NCE_F_DELAYED)
11233 		flags |= ACE_F_DELAYED;
11234 	return (flags);
11235 }
11236 
11237 /*
11238  * ncec_walk routine to create ipNetToMediaEntryTable
11239  */
11240 static int
ip_snmp_get2_v4_media(ncec_t * ncec,iproutedata_t * ird)11241 ip_snmp_get2_v4_media(ncec_t *ncec, iproutedata_t *ird)
11242 {
11243 	ill_t				*ill;
11244 	mib2_ipNetToMediaEntry_t	ntme;
11245 	const char			*name = "unknown";
11246 	ipaddr_t			ncec_addr;
11247 
11248 	ill = ncec->ncec_ill;
11249 	if (ill->ill_isv6 || (ncec->ncec_flags & NCE_F_BCAST) ||
11250 	    ill->ill_net_type == IRE_LOOPBACK)
11251 		return (0);
11252 
11253 	/* We report all IPMP groups on ncec_ill which is normally the upper. */
11254 	name = ill->ill_name;
11255 	/* Based on RFC 4293: other(1), inval(2), dyn(3), stat(4) */
11256 	if (NCE_MYADDR(ncec)) {
11257 		ntme.ipNetToMediaType = 4;
11258 	} else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST|NCE_F_PUBLISH)) {
11259 		ntme.ipNetToMediaType = 1;
11260 	} else {
11261 		ntme.ipNetToMediaType = 3;
11262 	}
11263 	ntme.ipNetToMediaIfIndex.o_length = MIN(OCTET_LENGTH, strlen(name));
11264 	bcopy(name, ntme.ipNetToMediaIfIndex.o_bytes,
11265 	    ntme.ipNetToMediaIfIndex.o_length);
11266 
11267 	IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ncec_addr);
11268 	bcopy(&ncec_addr, &ntme.ipNetToMediaNetAddress, sizeof (ncec_addr));
11269 
11270 	ntme.ipNetToMediaInfo.ntm_mask.o_length = sizeof (ipaddr_t);
11271 	ncec_addr = INADDR_BROADCAST;
11272 	bcopy(&ncec_addr, ntme.ipNetToMediaInfo.ntm_mask.o_bytes,
11273 	    sizeof (ncec_addr));
11274 	/*
11275 	 * map all the flags to the ACE counterpart.
11276 	 */
11277 	ntme.ipNetToMediaInfo.ntm_flags = nce2ace(ncec);
11278 
11279 	ntme.ipNetToMediaPhysAddress.o_length =
11280 	    MIN(OCTET_LENGTH, ill->ill_phys_addr_length);
11281 
11282 	if (!NCE_ISREACHABLE(ncec))
11283 		ntme.ipNetToMediaPhysAddress.o_length = 0;
11284 	else {
11285 		if (ncec->ncec_lladdr != NULL) {
11286 			bcopy(ncec->ncec_lladdr,
11287 			    ntme.ipNetToMediaPhysAddress.o_bytes,
11288 			    ntme.ipNetToMediaPhysAddress.o_length);
11289 		}
11290 	}
11291 
11292 	if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11293 	    &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11294 		ip1dbg(("ip_snmp_get2_v4_media: failed to allocate %u bytes\n",
11295 		    (uint_t)sizeof (ntme)));
11296 	}
11297 	return (0);
11298 }
11299 
11300 /*
11301  * return (0) if invalid set request, 1 otherwise, including non-tcp requests
11302  */
11303 /* ARGSUSED */
11304 int
ip_snmp_set(queue_t * q,int level,int name,uchar_t * ptr,int len)11305 ip_snmp_set(queue_t *q, int level, int name, uchar_t *ptr, int len)
11306 {
11307 	switch (level) {
11308 	case MIB2_IP:
11309 	case MIB2_ICMP:
11310 		switch (name) {
11311 		default:
11312 			break;
11313 		}
11314 		return (1);
11315 	default:
11316 		return (1);
11317 	}
11318 }
11319 
11320 /*
11321  * When there exists both a 64- and 32-bit counter of a particular type
11322  * (i.e., InReceives), only the 64-bit counters are added.
11323  */
11324 void
ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t * o1,mib2_ipIfStatsEntry_t * o2)11325 ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t *o1, mib2_ipIfStatsEntry_t *o2)
11326 {
11327 	UPDATE_MIB(o1, ipIfStatsInHdrErrors, o2->ipIfStatsInHdrErrors);
11328 	UPDATE_MIB(o1, ipIfStatsInTooBigErrors, o2->ipIfStatsInTooBigErrors);
11329 	UPDATE_MIB(o1, ipIfStatsInNoRoutes, o2->ipIfStatsInNoRoutes);
11330 	UPDATE_MIB(o1, ipIfStatsInAddrErrors, o2->ipIfStatsInAddrErrors);
11331 	UPDATE_MIB(o1, ipIfStatsInUnknownProtos, o2->ipIfStatsInUnknownProtos);
11332 	UPDATE_MIB(o1, ipIfStatsInTruncatedPkts, o2->ipIfStatsInTruncatedPkts);
11333 	UPDATE_MIB(o1, ipIfStatsInDiscards, o2->ipIfStatsInDiscards);
11334 	UPDATE_MIB(o1, ipIfStatsOutDiscards, o2->ipIfStatsOutDiscards);
11335 	UPDATE_MIB(o1, ipIfStatsOutFragOKs, o2->ipIfStatsOutFragOKs);
11336 	UPDATE_MIB(o1, ipIfStatsOutFragFails, o2->ipIfStatsOutFragFails);
11337 	UPDATE_MIB(o1, ipIfStatsOutFragCreates, o2->ipIfStatsOutFragCreates);
11338 	UPDATE_MIB(o1, ipIfStatsReasmReqds, o2->ipIfStatsReasmReqds);
11339 	UPDATE_MIB(o1, ipIfStatsReasmOKs, o2->ipIfStatsReasmOKs);
11340 	UPDATE_MIB(o1, ipIfStatsReasmFails, o2->ipIfStatsReasmFails);
11341 	UPDATE_MIB(o1, ipIfStatsOutNoRoutes, o2->ipIfStatsOutNoRoutes);
11342 	UPDATE_MIB(o1, ipIfStatsReasmDuplicates, o2->ipIfStatsReasmDuplicates);
11343 	UPDATE_MIB(o1, ipIfStatsReasmPartDups, o2->ipIfStatsReasmPartDups);
11344 	UPDATE_MIB(o1, ipIfStatsForwProhibits, o2->ipIfStatsForwProhibits);
11345 	UPDATE_MIB(o1, udpInCksumErrs, o2->udpInCksumErrs);
11346 	UPDATE_MIB(o1, udpInOverflows, o2->udpInOverflows);
11347 	UPDATE_MIB(o1, rawipInOverflows, o2->rawipInOverflows);
11348 	UPDATE_MIB(o1, ipIfStatsInWrongIPVersion,
11349 	    o2->ipIfStatsInWrongIPVersion);
11350 	UPDATE_MIB(o1, ipIfStatsOutWrongIPVersion,
11351 	    o2->ipIfStatsInWrongIPVersion);
11352 	UPDATE_MIB(o1, ipIfStatsOutSwitchIPVersion,
11353 	    o2->ipIfStatsOutSwitchIPVersion);
11354 	UPDATE_MIB(o1, ipIfStatsHCInReceives, o2->ipIfStatsHCInReceives);
11355 	UPDATE_MIB(o1, ipIfStatsHCInOctets, o2->ipIfStatsHCInOctets);
11356 	UPDATE_MIB(o1, ipIfStatsHCInForwDatagrams,
11357 	    o2->ipIfStatsHCInForwDatagrams);
11358 	UPDATE_MIB(o1, ipIfStatsHCInDelivers, o2->ipIfStatsHCInDelivers);
11359 	UPDATE_MIB(o1, ipIfStatsHCOutRequests, o2->ipIfStatsHCOutRequests);
11360 	UPDATE_MIB(o1, ipIfStatsHCOutForwDatagrams,
11361 	    o2->ipIfStatsHCOutForwDatagrams);
11362 	UPDATE_MIB(o1, ipIfStatsOutFragReqds, o2->ipIfStatsOutFragReqds);
11363 	UPDATE_MIB(o1, ipIfStatsHCOutTransmits, o2->ipIfStatsHCOutTransmits);
11364 	UPDATE_MIB(o1, ipIfStatsHCOutOctets, o2->ipIfStatsHCOutOctets);
11365 	UPDATE_MIB(o1, ipIfStatsHCInMcastPkts, o2->ipIfStatsHCInMcastPkts);
11366 	UPDATE_MIB(o1, ipIfStatsHCInMcastOctets, o2->ipIfStatsHCInMcastOctets);
11367 	UPDATE_MIB(o1, ipIfStatsHCOutMcastPkts, o2->ipIfStatsHCOutMcastPkts);
11368 	UPDATE_MIB(o1, ipIfStatsHCOutMcastOctets,
11369 	    o2->ipIfStatsHCOutMcastOctets);
11370 	UPDATE_MIB(o1, ipIfStatsHCInBcastPkts, o2->ipIfStatsHCInBcastPkts);
11371 	UPDATE_MIB(o1, ipIfStatsHCOutBcastPkts, o2->ipIfStatsHCOutBcastPkts);
11372 	UPDATE_MIB(o1, ipsecInSucceeded, o2->ipsecInSucceeded);
11373 	UPDATE_MIB(o1, ipsecInFailed, o2->ipsecInFailed);
11374 	UPDATE_MIB(o1, ipInCksumErrs, o2->ipInCksumErrs);
11375 	UPDATE_MIB(o1, tcpInErrs, o2->tcpInErrs);
11376 	UPDATE_MIB(o1, udpNoPorts, o2->udpNoPorts);
11377 }
11378 
11379 void
ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t * o1,mib2_ipv6IfIcmpEntry_t * o2)11380 ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t *o1, mib2_ipv6IfIcmpEntry_t *o2)
11381 {
11382 	UPDATE_MIB(o1, ipv6IfIcmpInMsgs, o2->ipv6IfIcmpInMsgs);
11383 	UPDATE_MIB(o1, ipv6IfIcmpInErrors, o2->ipv6IfIcmpInErrors);
11384 	UPDATE_MIB(o1, ipv6IfIcmpInDestUnreachs, o2->ipv6IfIcmpInDestUnreachs);
11385 	UPDATE_MIB(o1, ipv6IfIcmpInAdminProhibs, o2->ipv6IfIcmpInAdminProhibs);
11386 	UPDATE_MIB(o1, ipv6IfIcmpInTimeExcds, o2->ipv6IfIcmpInTimeExcds);
11387 	UPDATE_MIB(o1, ipv6IfIcmpInParmProblems, o2->ipv6IfIcmpInParmProblems);
11388 	UPDATE_MIB(o1, ipv6IfIcmpInPktTooBigs, o2->ipv6IfIcmpInPktTooBigs);
11389 	UPDATE_MIB(o1, ipv6IfIcmpInEchos, o2->ipv6IfIcmpInEchos);
11390 	UPDATE_MIB(o1, ipv6IfIcmpInEchoReplies, o2->ipv6IfIcmpInEchoReplies);
11391 	UPDATE_MIB(o1, ipv6IfIcmpInRouterSolicits,
11392 	    o2->ipv6IfIcmpInRouterSolicits);
11393 	UPDATE_MIB(o1, ipv6IfIcmpInRouterAdvertisements,
11394 	    o2->ipv6IfIcmpInRouterAdvertisements);
11395 	UPDATE_MIB(o1, ipv6IfIcmpInNeighborSolicits,
11396 	    o2->ipv6IfIcmpInNeighborSolicits);
11397 	UPDATE_MIB(o1, ipv6IfIcmpInNeighborAdvertisements,
11398 	    o2->ipv6IfIcmpInNeighborAdvertisements);
11399 	UPDATE_MIB(o1, ipv6IfIcmpInRedirects, o2->ipv6IfIcmpInRedirects);
11400 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembQueries,
11401 	    o2->ipv6IfIcmpInGroupMembQueries);
11402 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembResponses,
11403 	    o2->ipv6IfIcmpInGroupMembResponses);
11404 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembReductions,
11405 	    o2->ipv6IfIcmpInGroupMembReductions);
11406 	UPDATE_MIB(o1, ipv6IfIcmpOutMsgs, o2->ipv6IfIcmpOutMsgs);
11407 	UPDATE_MIB(o1, ipv6IfIcmpOutErrors, o2->ipv6IfIcmpOutErrors);
11408 	UPDATE_MIB(o1, ipv6IfIcmpOutDestUnreachs,
11409 	    o2->ipv6IfIcmpOutDestUnreachs);
11410 	UPDATE_MIB(o1, ipv6IfIcmpOutAdminProhibs,
11411 	    o2->ipv6IfIcmpOutAdminProhibs);
11412 	UPDATE_MIB(o1, ipv6IfIcmpOutTimeExcds, o2->ipv6IfIcmpOutTimeExcds);
11413 	UPDATE_MIB(o1, ipv6IfIcmpOutParmProblems,
11414 	    o2->ipv6IfIcmpOutParmProblems);
11415 	UPDATE_MIB(o1, ipv6IfIcmpOutPktTooBigs, o2->ipv6IfIcmpOutPktTooBigs);
11416 	UPDATE_MIB(o1, ipv6IfIcmpOutEchos, o2->ipv6IfIcmpOutEchos);
11417 	UPDATE_MIB(o1, ipv6IfIcmpOutEchoReplies, o2->ipv6IfIcmpOutEchoReplies);
11418 	UPDATE_MIB(o1, ipv6IfIcmpOutRouterSolicits,
11419 	    o2->ipv6IfIcmpOutRouterSolicits);
11420 	UPDATE_MIB(o1, ipv6IfIcmpOutRouterAdvertisements,
11421 	    o2->ipv6IfIcmpOutRouterAdvertisements);
11422 	UPDATE_MIB(o1, ipv6IfIcmpOutNeighborSolicits,
11423 	    o2->ipv6IfIcmpOutNeighborSolicits);
11424 	UPDATE_MIB(o1, ipv6IfIcmpOutNeighborAdvertisements,
11425 	    o2->ipv6IfIcmpOutNeighborAdvertisements);
11426 	UPDATE_MIB(o1, ipv6IfIcmpOutRedirects, o2->ipv6IfIcmpOutRedirects);
11427 	UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembQueries,
11428 	    o2->ipv6IfIcmpOutGroupMembQueries);
11429 	UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembResponses,
11430 	    o2->ipv6IfIcmpOutGroupMembResponses);
11431 	UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembReductions,
11432 	    o2->ipv6IfIcmpOutGroupMembReductions);
11433 	UPDATE_MIB(o1, ipv6IfIcmpInOverflows, o2->ipv6IfIcmpInOverflows);
11434 	UPDATE_MIB(o1, ipv6IfIcmpBadHoplimit, o2->ipv6IfIcmpBadHoplimit);
11435 	UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborAdvertisements,
11436 	    o2->ipv6IfIcmpInBadNeighborAdvertisements);
11437 	UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborSolicitations,
11438 	    o2->ipv6IfIcmpInBadNeighborSolicitations);
11439 	UPDATE_MIB(o1, ipv6IfIcmpInBadRedirects, o2->ipv6IfIcmpInBadRedirects);
11440 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembTotal,
11441 	    o2->ipv6IfIcmpInGroupMembTotal);
11442 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadQueries,
11443 	    o2->ipv6IfIcmpInGroupMembBadQueries);
11444 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadReports,
11445 	    o2->ipv6IfIcmpInGroupMembBadReports);
11446 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembOurReports,
11447 	    o2->ipv6IfIcmpInGroupMembOurReports);
11448 }
11449 
11450 /*
11451  * Called before the options are updated to check if this packet will
11452  * be source routed from here.
11453  * This routine assumes that the options are well formed i.e. that they
11454  * have already been checked.
11455  */
11456 boolean_t
ip_source_routed(ipha_t * ipha,ip_stack_t * ipst)11457 ip_source_routed(ipha_t *ipha, ip_stack_t *ipst)
11458 {
11459 	ipoptp_t	opts;
11460 	uchar_t		*opt;
11461 	uint8_t		optval;
11462 	uint8_t		optlen;
11463 	ipaddr_t	dst;
11464 
11465 	if (IS_SIMPLE_IPH(ipha)) {
11466 		ip2dbg(("not source routed\n"));
11467 		return (B_FALSE);
11468 	}
11469 	dst = ipha->ipha_dst;
11470 	for (optval = ipoptp_first(&opts, ipha);
11471 	    optval != IPOPT_EOL;
11472 	    optval = ipoptp_next(&opts)) {
11473 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11474 		opt = opts.ipoptp_cur;
11475 		optlen = opts.ipoptp_len;
11476 		ip2dbg(("ip_source_routed: opt %d, len %d\n",
11477 		    optval, optlen));
11478 		switch (optval) {
11479 			uint32_t off;
11480 		case IPOPT_SSRR:
11481 		case IPOPT_LSRR:
11482 			/*
11483 			 * If dst is one of our addresses and there are some
11484 			 * entries left in the source route return (true).
11485 			 */
11486 			if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11487 				ip2dbg(("ip_source_routed: not next"
11488 				    " source route 0x%x\n",
11489 				    ntohl(dst)));
11490 				return (B_FALSE);
11491 			}
11492 			off = opt[IPOPT_OFFSET];
11493 			off--;
11494 			if (optlen < IP_ADDR_LEN ||
11495 			    off > optlen - IP_ADDR_LEN) {
11496 				/* End of source route */
11497 				ip1dbg(("ip_source_routed: end of SR\n"));
11498 				return (B_FALSE);
11499 			}
11500 			return (B_TRUE);
11501 		}
11502 	}
11503 	ip2dbg(("not source routed\n"));
11504 	return (B_FALSE);
11505 }
11506 
11507 /*
11508  * ip_unbind is called by the transports to remove a conn from
11509  * the fanout table.
11510  */
11511 void
ip_unbind(conn_t * connp)11512 ip_unbind(conn_t *connp)
11513 {
11514 
11515 	ASSERT(!MUTEX_HELD(&connp->conn_lock));
11516 
11517 	if (is_system_labeled() && connp->conn_anon_port) {
11518 		(void) tsol_mlp_anon(crgetzone(connp->conn_cred),
11519 		    connp->conn_mlp_type, connp->conn_proto,
11520 		    ntohs(connp->conn_lport), B_FALSE);
11521 		connp->conn_anon_port = 0;
11522 	}
11523 	connp->conn_mlp_type = mlptSingle;
11524 
11525 	ipcl_hash_remove(connp);
11526 }
11527 
11528 /*
11529  * Used for deciding the MSS size for the upper layer. Thus
11530  * we need to check the outbound policy values in the conn.
11531  */
11532 int
conn_ipsec_length(conn_t * connp)11533 conn_ipsec_length(conn_t *connp)
11534 {
11535 	ipsec_latch_t *ipl;
11536 
11537 	ipl = connp->conn_latch;
11538 	if (ipl == NULL)
11539 		return (0);
11540 
11541 	if (connp->conn_ixa->ixa_ipsec_policy == NULL)
11542 		return (0);
11543 
11544 	return (connp->conn_ixa->ixa_ipsec_policy->ipsp_act->ipa_ovhd);
11545 }
11546 
11547 /*
11548  * Returns an estimate of the IPsec headers size. This is used if
11549  * we don't want to call into IPsec to get the exact size.
11550  */
11551 int
ipsec_out_extra_length(ip_xmit_attr_t * ixa)11552 ipsec_out_extra_length(ip_xmit_attr_t *ixa)
11553 {
11554 	ipsec_action_t *a;
11555 
11556 	if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
11557 		return (0);
11558 
11559 	a = ixa->ixa_ipsec_action;
11560 	if (a == NULL) {
11561 		ASSERT(ixa->ixa_ipsec_policy != NULL);
11562 		a = ixa->ixa_ipsec_policy->ipsp_act;
11563 	}
11564 	ASSERT(a != NULL);
11565 
11566 	return (a->ipa_ovhd);
11567 }
11568 
11569 /*
11570  * If there are any source route options, return the true final
11571  * destination. Otherwise, return the destination.
11572  */
11573 ipaddr_t
ip_get_dst(ipha_t * ipha)11574 ip_get_dst(ipha_t *ipha)
11575 {
11576 	ipoptp_t	opts;
11577 	uchar_t		*opt;
11578 	uint8_t		optval;
11579 	uint8_t		optlen;
11580 	ipaddr_t	dst;
11581 	uint32_t off;
11582 
11583 	dst = ipha->ipha_dst;
11584 
11585 	if (IS_SIMPLE_IPH(ipha))
11586 		return (dst);
11587 
11588 	for (optval = ipoptp_first(&opts, ipha);
11589 	    optval != IPOPT_EOL;
11590 	    optval = ipoptp_next(&opts)) {
11591 		opt = opts.ipoptp_cur;
11592 		optlen = opts.ipoptp_len;
11593 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11594 		switch (optval) {
11595 		case IPOPT_SSRR:
11596 		case IPOPT_LSRR:
11597 			off = opt[IPOPT_OFFSET];
11598 			/*
11599 			 * If one of the conditions is true, it means
11600 			 * end of options and dst already has the right
11601 			 * value.
11602 			 */
11603 			if (!(optlen < IP_ADDR_LEN || off > optlen - 3)) {
11604 				off = optlen - IP_ADDR_LEN;
11605 				bcopy(&opt[off], &dst, IP_ADDR_LEN);
11606 			}
11607 			return (dst);
11608 		default:
11609 			break;
11610 		}
11611 	}
11612 
11613 	return (dst);
11614 }
11615 
11616 /*
11617  * Outbound IP fragmentation routine.
11618  * Assumes the caller has checked whether or not fragmentation should
11619  * be allowed. Here we copy the DF bit from the header to all the generated
11620  * fragments.
11621  */
11622 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)11623 ip_fragment_v4(mblk_t *mp_orig, nce_t *nce, iaflags_t ixaflags,
11624     uint_t pkt_len, uint32_t max_frag, uint32_t xmit_hint, zoneid_t szone,
11625     zoneid_t nolzid, pfirepostfrag_t postfragfn, uintptr_t *ixa_cookie)
11626 {
11627 	int		i1;
11628 	int		hdr_len;
11629 	mblk_t		*hdr_mp;
11630 	ipha_t		*ipha;
11631 	int		ip_data_end;
11632 	int		len;
11633 	mblk_t		*mp = mp_orig;
11634 	int		offset;
11635 	ill_t		*ill = nce->nce_ill;
11636 	ip_stack_t	*ipst = ill->ill_ipst;
11637 	mblk_t		*carve_mp;
11638 	uint32_t	frag_flag;
11639 	uint_t		priority = mp->b_band;
11640 	int		error = 0;
11641 
11642 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragReqds);
11643 
11644 	if (pkt_len != msgdsize(mp)) {
11645 		ip0dbg(("Packet length mismatch: %d, %ld\n",
11646 		    pkt_len, msgdsize(mp)));
11647 		freemsg(mp);
11648 		return (EINVAL);
11649 	}
11650 
11651 	if (max_frag == 0) {
11652 		ip1dbg(("ip_fragment_v4: max_frag is zero. Dropping packet\n"));
11653 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11654 		ip_drop_output("FragFails: zero max_frag", mp, ill);
11655 		freemsg(mp);
11656 		return (EINVAL);
11657 	}
11658 
11659 	ASSERT(MBLKL(mp) >= sizeof (ipha_t));
11660 	ipha = (ipha_t *)mp->b_rptr;
11661 	ASSERT(ntohs(ipha->ipha_length) == pkt_len);
11662 	frag_flag = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_DF;
11663 
11664 	/*
11665 	 * Establish the starting offset.  May not be zero if we are fragging
11666 	 * a fragment that is being forwarded.
11667 	 */
11668 	offset = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET;
11669 
11670 	/* TODO why is this test needed? */
11671 	if (((max_frag - ntohs(ipha->ipha_length)) & ~7) < 8) {
11672 		/* TODO: notify ulp somehow */
11673 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11674 		ip_drop_output("FragFails: bad starting offset", mp, ill);
11675 		freemsg(mp);
11676 		return (EINVAL);
11677 	}
11678 
11679 	hdr_len = IPH_HDR_LENGTH(ipha);
11680 	ipha->ipha_hdr_checksum = 0;
11681 
11682 	/*
11683 	 * Establish the number of bytes maximum per frag, after putting
11684 	 * in the header.
11685 	 */
11686 	len = (max_frag - hdr_len) & ~7;
11687 
11688 	/* Get a copy of the header for the trailing frags */
11689 	hdr_mp = ip_fragment_copyhdr((uchar_t *)ipha, hdr_len, offset, ipst,
11690 	    mp);
11691 	if (hdr_mp == NULL) {
11692 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11693 		ip_drop_output("FragFails: no hdr_mp", mp, ill);
11694 		freemsg(mp);
11695 		return (ENOBUFS);
11696 	}
11697 
11698 	/* Store the starting offset, with the MoreFrags flag. */
11699 	i1 = offset | IPH_MF | frag_flag;
11700 	ipha->ipha_fragment_offset_and_flags = htons((uint16_t)i1);
11701 
11702 	/* Establish the ending byte offset, based on the starting offset. */
11703 	offset <<= 3;
11704 	ip_data_end = offset + ntohs(ipha->ipha_length) - hdr_len;
11705 
11706 	/* Store the length of the first fragment in the IP header. */
11707 	i1 = len + hdr_len;
11708 	ASSERT(i1 <= IP_MAXPACKET);
11709 	ipha->ipha_length = htons((uint16_t)i1);
11710 
11711 	/*
11712 	 * Compute the IP header checksum for the first frag.  We have to
11713 	 * watch out that we stop at the end of the header.
11714 	 */
11715 	ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11716 
11717 	/*
11718 	 * Now carve off the first frag.  Note that this will include the
11719 	 * original IP header.
11720 	 */
11721 	if (!(mp = ip_carve_mp(&mp_orig, i1))) {
11722 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11723 		ip_drop_output("FragFails: could not carve mp", mp_orig, ill);
11724 		freeb(hdr_mp);
11725 		freemsg(mp_orig);
11726 		return (ENOBUFS);
11727 	}
11728 
11729 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11730 
11731 	error = postfragfn(mp, nce, ixaflags, i1, xmit_hint, szone, nolzid,
11732 	    ixa_cookie);
11733 	if (error != 0 && error != EWOULDBLOCK) {
11734 		/* No point in sending the other fragments */
11735 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11736 		ip_drop_output("FragFails: postfragfn failed", mp_orig, ill);
11737 		freeb(hdr_mp);
11738 		freemsg(mp_orig);
11739 		return (error);
11740 	}
11741 
11742 	/* No need to redo state machine in loop */
11743 	ixaflags &= ~IXAF_REACH_CONF;
11744 
11745 	/* Advance the offset to the second frag starting point. */
11746 	offset += len;
11747 	/*
11748 	 * Update hdr_len from the copied header - there might be less options
11749 	 * in the later fragments.
11750 	 */
11751 	hdr_len = IPH_HDR_LENGTH(hdr_mp->b_rptr);
11752 	/* Loop until done. */
11753 	for (;;) {
11754 		uint16_t	offset_and_flags;
11755 		uint16_t	ip_len;
11756 
11757 		if (ip_data_end - offset > len) {
11758 			/*
11759 			 * Carve off the appropriate amount from the original
11760 			 * datagram.
11761 			 */
11762 			if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11763 				mp = NULL;
11764 				break;
11765 			}
11766 			/*
11767 			 * More frags after this one.  Get another copy
11768 			 * of the header.
11769 			 */
11770 			if (carve_mp->b_datap->db_ref == 1 &&
11771 			    hdr_mp->b_wptr - hdr_mp->b_rptr <
11772 			    carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11773 				/* Inline IP header */
11774 				carve_mp->b_rptr -= hdr_mp->b_wptr -
11775 				    hdr_mp->b_rptr;
11776 				bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11777 				    hdr_mp->b_wptr - hdr_mp->b_rptr);
11778 				mp = carve_mp;
11779 			} else {
11780 				if (!(mp = copyb(hdr_mp))) {
11781 					freemsg(carve_mp);
11782 					break;
11783 				}
11784 				/* Get priority marking, if any. */
11785 				mp->b_band = priority;
11786 				mp->b_cont = carve_mp;
11787 			}
11788 			ipha = (ipha_t *)mp->b_rptr;
11789 			offset_and_flags = IPH_MF;
11790 		} else {
11791 			/*
11792 			 * Last frag.  Consume the header. Set len to
11793 			 * the length of this last piece.
11794 			 */
11795 			len = ip_data_end - offset;
11796 
11797 			/*
11798 			 * Carve off the appropriate amount from the original
11799 			 * datagram.
11800 			 */
11801 			if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11802 				mp = NULL;
11803 				break;
11804 			}
11805 			if (carve_mp->b_datap->db_ref == 1 &&
11806 			    hdr_mp->b_wptr - hdr_mp->b_rptr <
11807 			    carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11808 				/* Inline IP header */
11809 				carve_mp->b_rptr -= hdr_mp->b_wptr -
11810 				    hdr_mp->b_rptr;
11811 				bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11812 				    hdr_mp->b_wptr - hdr_mp->b_rptr);
11813 				mp = carve_mp;
11814 				freeb(hdr_mp);
11815 				hdr_mp = mp;
11816 			} else {
11817 				mp = hdr_mp;
11818 				/* Get priority marking, if any. */
11819 				mp->b_band = priority;
11820 				mp->b_cont = carve_mp;
11821 			}
11822 			ipha = (ipha_t *)mp->b_rptr;
11823 			/* A frag of a frag might have IPH_MF non-zero */
11824 			offset_and_flags =
11825 			    ntohs(ipha->ipha_fragment_offset_and_flags) &
11826 			    IPH_MF;
11827 		}
11828 		offset_and_flags |= (uint16_t)(offset >> 3);
11829 		offset_and_flags |= (uint16_t)frag_flag;
11830 		/* Store the offset and flags in the IP header. */
11831 		ipha->ipha_fragment_offset_and_flags = htons(offset_and_flags);
11832 
11833 		/* Store the length in the IP header. */
11834 		ip_len = (uint16_t)(len + hdr_len);
11835 		ipha->ipha_length = htons(ip_len);
11836 
11837 		/*
11838 		 * Set the IP header checksum.	Note that mp is just
11839 		 * the header, so this is easy to pass to ip_csum.
11840 		 */
11841 		ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11842 
11843 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11844 
11845 		error = postfragfn(mp, nce, ixaflags, ip_len, xmit_hint, szone,
11846 		    nolzid, ixa_cookie);
11847 		/* All done if we just consumed the hdr_mp. */
11848 		if (mp == hdr_mp) {
11849 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragOKs);
11850 			return (error);
11851 		}
11852 		if (error != 0 && error != EWOULDBLOCK) {
11853 			DTRACE_PROBE2(ip__xmit__frag__fail, ill_t *, ill,
11854 			    mblk_t *, hdr_mp);
11855 			/* No point in sending the other fragments */
11856 			break;
11857 		}
11858 
11859 		/* Otherwise, advance and loop. */
11860 		offset += len;
11861 	}
11862 	/* Clean up following allocation failure. */
11863 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11864 	ip_drop_output("FragFails: loop ended", NULL, ill);
11865 	if (mp != hdr_mp)
11866 		freeb(hdr_mp);
11867 	if (mp != mp_orig)
11868 		freemsg(mp_orig);
11869 	return (error);
11870 }
11871 
11872 /*
11873  * Copy the header plus those options which have the copy bit set
11874  */
11875 static mblk_t *
ip_fragment_copyhdr(uchar_t * rptr,int hdr_len,int offset,ip_stack_t * ipst,mblk_t * src)11876 ip_fragment_copyhdr(uchar_t *rptr, int hdr_len, int offset, ip_stack_t *ipst,
11877     mblk_t *src)
11878 {
11879 	mblk_t	*mp;
11880 	uchar_t	*up;
11881 
11882 	/*
11883 	 * Quick check if we need to look for options without the copy bit
11884 	 * set
11885 	 */
11886 	mp = allocb_tmpl(ipst->ips_ip_wroff_extra + hdr_len, src);
11887 	if (!mp)
11888 		return (mp);
11889 	mp->b_rptr += ipst->ips_ip_wroff_extra;
11890 	if (hdr_len == IP_SIMPLE_HDR_LENGTH || offset != 0) {
11891 		bcopy(rptr, mp->b_rptr, hdr_len);
11892 		mp->b_wptr += hdr_len + ipst->ips_ip_wroff_extra;
11893 		return (mp);
11894 	}
11895 	up  = mp->b_rptr;
11896 	bcopy(rptr, up, IP_SIMPLE_HDR_LENGTH);
11897 	up += IP_SIMPLE_HDR_LENGTH;
11898 	rptr += IP_SIMPLE_HDR_LENGTH;
11899 	hdr_len -= IP_SIMPLE_HDR_LENGTH;
11900 	while (hdr_len > 0) {
11901 		uint32_t optval;
11902 		uint32_t optlen;
11903 
11904 		optval = *rptr;
11905 		if (optval == IPOPT_EOL)
11906 			break;
11907 		if (optval == IPOPT_NOP)
11908 			optlen = 1;
11909 		else
11910 			optlen = rptr[1];
11911 		if (optval & IPOPT_COPY) {
11912 			bcopy(rptr, up, optlen);
11913 			up += optlen;
11914 		}
11915 		rptr += optlen;
11916 		hdr_len -= optlen;
11917 	}
11918 	/*
11919 	 * Make sure that we drop an even number of words by filling
11920 	 * with EOL to the next word boundary.
11921 	 */
11922 	for (hdr_len = up - (mp->b_rptr + IP_SIMPLE_HDR_LENGTH);
11923 	    hdr_len & 0x3; hdr_len++)
11924 		*up++ = IPOPT_EOL;
11925 	mp->b_wptr = up;
11926 	/* Update header length */
11927 	mp->b_rptr[0] = (uint8_t)((IP_VERSION << 4) | ((up - mp->b_rptr) >> 2));
11928 	return (mp);
11929 }
11930 
11931 /*
11932  * Update any source route, record route, or timestamp options when
11933  * sending a packet back to ourselves.
11934  * Check that we are at end of strict source route.
11935  * The options have been sanity checked by ip_output_options().
11936  */
11937 void
ip_output_local_options(ipha_t * ipha,ip_stack_t * ipst)11938 ip_output_local_options(ipha_t *ipha, ip_stack_t *ipst)
11939 {
11940 	ipoptp_t	opts;
11941 	uchar_t		*opt;
11942 	uint8_t		optval;
11943 	uint8_t		optlen;
11944 	ipaddr_t	dst;
11945 	uint32_t	ts;
11946 	timestruc_t	now;
11947 
11948 	for (optval = ipoptp_first(&opts, ipha);
11949 	    optval != IPOPT_EOL;
11950 	    optval = ipoptp_next(&opts)) {
11951 		opt = opts.ipoptp_cur;
11952 		optlen = opts.ipoptp_len;
11953 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11954 		switch (optval) {
11955 			uint32_t off;
11956 		case IPOPT_SSRR:
11957 		case IPOPT_LSRR:
11958 			off = opt[IPOPT_OFFSET];
11959 			off--;
11960 			if (optlen < IP_ADDR_LEN ||
11961 			    off > optlen - IP_ADDR_LEN) {
11962 				/* End of source route */
11963 				break;
11964 			}
11965 			/*
11966 			 * This will only happen if two consecutive entries
11967 			 * in the source route contains our address or if
11968 			 * it is a packet with a loose source route which
11969 			 * reaches us before consuming the whole source route
11970 			 */
11971 
11972 			if (optval == IPOPT_SSRR) {
11973 				return;
11974 			}
11975 			/*
11976 			 * Hack: instead of dropping the packet truncate the
11977 			 * source route to what has been used by filling the
11978 			 * rest with IPOPT_NOP.
11979 			 */
11980 			opt[IPOPT_OLEN] = (uint8_t)off;
11981 			while (off < optlen) {
11982 				opt[off++] = IPOPT_NOP;
11983 			}
11984 			break;
11985 		case IPOPT_RR:
11986 			off = opt[IPOPT_OFFSET];
11987 			off--;
11988 			if (optlen < IP_ADDR_LEN ||
11989 			    off > optlen - IP_ADDR_LEN) {
11990 				/* No more room - ignore */
11991 				ip1dbg((
11992 				    "ip_output_local_options: end of RR\n"));
11993 				break;
11994 			}
11995 			dst = htonl(INADDR_LOOPBACK);
11996 			bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
11997 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
11998 			break;
11999 		case IPOPT_TS:
12000 			/* Insert timestamp if there is romm */
12001 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12002 			case IPOPT_TS_TSONLY:
12003 				off = IPOPT_TS_TIMELEN;
12004 				break;
12005 			case IPOPT_TS_PRESPEC:
12006 			case IPOPT_TS_PRESPEC_RFC791:
12007 				/* Verify that the address matched */
12008 				off = opt[IPOPT_OFFSET] - 1;
12009 				bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
12010 				if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
12011 					/* Not for us */
12012 					break;
12013 				}
12014 				/* FALLTHRU */
12015 			case IPOPT_TS_TSANDADDR:
12016 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
12017 				break;
12018 			default:
12019 				/*
12020 				 * ip_*put_options should have already
12021 				 * dropped this packet.
12022 				 */
12023 				cmn_err(CE_PANIC, "ip_output_local_options: "
12024 				    "unknown IT - bug in ip_output_options?\n");
12025 				return;	/* Keep "lint" happy */
12026 			}
12027 			if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
12028 				/* Increase overflow counter */
12029 				off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
12030 				opt[IPOPT_POS_OV_FLG] = (uint8_t)
12031 				    (opt[IPOPT_POS_OV_FLG] & 0x0F) |
12032 				    (off << 4);
12033 				break;
12034 			}
12035 			off = opt[IPOPT_OFFSET] - 1;
12036 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12037 			case IPOPT_TS_PRESPEC:
12038 			case IPOPT_TS_PRESPEC_RFC791:
12039 			case IPOPT_TS_TSANDADDR:
12040 				dst = htonl(INADDR_LOOPBACK);
12041 				bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
12042 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
12043 				/* FALLTHRU */
12044 			case IPOPT_TS_TSONLY:
12045 				off = opt[IPOPT_OFFSET] - 1;
12046 				/* Compute # of milliseconds since midnight */
12047 				gethrestime(&now);
12048 				ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
12049 				    NSEC2MSEC(now.tv_nsec);
12050 				bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
12051 				opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
12052 				break;
12053 			}
12054 			break;
12055 		}
12056 	}
12057 }
12058 
12059 /*
12060  * Prepend an M_DATA fastpath header, and if none present prepend a
12061  * DL_UNITDATA_REQ. Frees the mblk on failure.
12062  *
12063  * nce_dlur_mp and nce_fp_mp can not disappear once they have been set.
12064  * If there is a change to them, the nce will be deleted (condemned) and
12065  * a new nce_t will be created when packets are sent. Thus we need no locks
12066  * to access those fields.
12067  *
12068  * We preserve b_band to support IPQoS. If a DL_UNITDATA_REQ is prepended
12069  * we place b_band in dl_priority.dl_max.
12070  */
12071 static mblk_t *
ip_xmit_attach_llhdr(mblk_t * mp,nce_t * nce)12072 ip_xmit_attach_llhdr(mblk_t *mp, nce_t *nce)
12073 {
12074 	uint_t	hlen;
12075 	mblk_t *mp1;
12076 	uint_t	priority;
12077 	uchar_t *rptr;
12078 
12079 	rptr = mp->b_rptr;
12080 
12081 	ASSERT(DB_TYPE(mp) == M_DATA);
12082 	priority = mp->b_band;
12083 
12084 	ASSERT(nce != NULL);
12085 	if ((mp1 = nce->nce_fp_mp) != NULL) {
12086 		hlen = MBLKL(mp1);
12087 		/*
12088 		 * Check if we have enough room to prepend fastpath
12089 		 * header
12090 		 */
12091 		if (hlen != 0 && (rptr - mp->b_datap->db_base) >= hlen) {
12092 			rptr -= hlen;
12093 			bcopy(mp1->b_rptr, rptr, hlen);
12094 			/*
12095 			 * Set the b_rptr to the start of the link layer
12096 			 * header
12097 			 */
12098 			mp->b_rptr = rptr;
12099 			return (mp);
12100 		}
12101 		mp1 = copyb(mp1);
12102 		if (mp1 == NULL) {
12103 			ill_t *ill = nce->nce_ill;
12104 
12105 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12106 			ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12107 			freemsg(mp);
12108 			return (NULL);
12109 		}
12110 		mp1->b_band = priority;
12111 		mp1->b_cont = mp;
12112 		DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
12113 		DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
12114 		DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
12115 		DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
12116 		DB_LSOMSS(mp1) = DB_LSOMSS(mp);
12117 		DTRACE_PROBE1(ip__xmit__copyb, (mblk_t *), mp1);
12118 		/*
12119 		 * XXX disable ICK_VALID and compute checksum
12120 		 * here; can happen if nce_fp_mp changes and
12121 		 * it can't be copied now due to insufficient
12122 		 * space. (unlikely, fp mp can change, but it
12123 		 * does not increase in length)
12124 		 */
12125 		return (mp1);
12126 	}
12127 	mp1 = copyb(nce->nce_dlur_mp);
12128 
12129 	if (mp1 == NULL) {
12130 		ill_t *ill = nce->nce_ill;
12131 
12132 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12133 		ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12134 		freemsg(mp);
12135 		return (NULL);
12136 	}
12137 	mp1->b_cont = mp;
12138 	if (priority != 0) {
12139 		mp1->b_band = priority;
12140 		((dl_unitdata_req_t *)(mp1->b_rptr))->dl_priority.dl_max =
12141 		    priority;
12142 	}
12143 	return (mp1);
12144 }
12145 
12146 /*
12147  * Finish the outbound IPsec processing. This function is called from
12148  * ipsec_out_process() if the IPsec packet was processed
12149  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12150  * asynchronously.
12151  *
12152  * This is common to IPv4 and IPv6.
12153  */
12154 int
ip_output_post_ipsec(mblk_t * mp,ip_xmit_attr_t * ixa)12155 ip_output_post_ipsec(mblk_t *mp, ip_xmit_attr_t *ixa)
12156 {
12157 	iaflags_t	ixaflags = ixa->ixa_flags;
12158 	uint_t		pktlen;
12159 
12160 
12161 	/* AH/ESP don't update ixa_pktlen when they modify the packet */
12162 	if (ixaflags & IXAF_IS_IPV4) {
12163 		ipha_t		*ipha = (ipha_t *)mp->b_rptr;
12164 
12165 		ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12166 		pktlen = ntohs(ipha->ipha_length);
12167 	} else {
12168 		ip6_t		*ip6h = (ip6_t *)mp->b_rptr;
12169 
12170 		ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12171 		pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12172 	}
12173 
12174 	/*
12175 	 * We release any hard reference on the SAs here to make
12176 	 * sure the SAs can be garbage collected. ipsr_sa has a soft reference
12177 	 * on the SAs.
12178 	 * If in the future we want the hard latching of the SAs in the
12179 	 * ip_xmit_attr_t then we should remove this.
12180 	 */
12181 	if (ixa->ixa_ipsec_esp_sa != NULL) {
12182 		IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12183 		ixa->ixa_ipsec_esp_sa = NULL;
12184 	}
12185 	if (ixa->ixa_ipsec_ah_sa != NULL) {
12186 		IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12187 		ixa->ixa_ipsec_ah_sa = NULL;
12188 	}
12189 
12190 	/* Do we need to fragment? */
12191 	if ((ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR) ||
12192 	    pktlen > ixa->ixa_fragsize) {
12193 		if (ixaflags & IXAF_IS_IPV4) {
12194 			ASSERT(!(ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR));
12195 			/*
12196 			 * We check for the DF case in ipsec_out_process
12197 			 * hence this only handles the non-DF case.
12198 			 */
12199 			return (ip_fragment_v4(mp, ixa->ixa_nce, ixa->ixa_flags,
12200 			    pktlen, ixa->ixa_fragsize,
12201 			    ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12202 			    ixa->ixa_no_loop_zoneid, ixa->ixa_postfragfn,
12203 			    &ixa->ixa_cookie));
12204 		} else {
12205 			mp = ip_fraghdr_add_v6(mp, ixa->ixa_ident, ixa);
12206 			if (mp == NULL) {
12207 				/* MIB and ip_drop_output already done */
12208 				return (ENOMEM);
12209 			}
12210 			pktlen += sizeof (ip6_frag_t);
12211 			if (pktlen > ixa->ixa_fragsize) {
12212 				return (ip_fragment_v6(mp, ixa->ixa_nce,
12213 				    ixa->ixa_flags, pktlen,
12214 				    ixa->ixa_fragsize, ixa->ixa_xmit_hint,
12215 				    ixa->ixa_zoneid, ixa->ixa_no_loop_zoneid,
12216 				    ixa->ixa_postfragfn, &ixa->ixa_cookie));
12217 			}
12218 		}
12219 	}
12220 	return ((ixa->ixa_postfragfn)(mp, ixa->ixa_nce, ixa->ixa_flags,
12221 	    pktlen, ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12222 	    ixa->ixa_no_loop_zoneid, NULL));
12223 }
12224 
12225 /*
12226  * Finish the inbound IPsec processing. This function is called from
12227  * ipsec_out_process() if the IPsec packet was processed
12228  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12229  * asynchronously.
12230  *
12231  * This is common to IPv4 and IPv6.
12232  */
12233 void
ip_input_post_ipsec(mblk_t * mp,ip_recv_attr_t * ira)12234 ip_input_post_ipsec(mblk_t *mp, ip_recv_attr_t *ira)
12235 {
12236 	iaflags_t	iraflags = ira->ira_flags;
12237 
12238 	/* Length might have changed */
12239 	if (iraflags & IRAF_IS_IPV4) {
12240 		ipha_t		*ipha = (ipha_t *)mp->b_rptr;
12241 
12242 		ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12243 		ira->ira_pktlen = ntohs(ipha->ipha_length);
12244 		ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
12245 		ira->ira_protocol = ipha->ipha_protocol;
12246 
12247 		ip_fanout_v4(mp, ipha, ira);
12248 	} else {
12249 		ip6_t		*ip6h = (ip6_t *)mp->b_rptr;
12250 		uint8_t		*nexthdrp;
12251 
12252 		ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12253 		ira->ira_pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12254 		if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ira->ira_ip_hdr_length,
12255 		    &nexthdrp)) {
12256 			/* Malformed packet */
12257 			BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
12258 			ip_drop_input("ipIfStatsInDiscards", mp, ira->ira_ill);
12259 			freemsg(mp);
12260 			return;
12261 		}
12262 		ira->ira_protocol = *nexthdrp;
12263 		ip_fanout_v6(mp, ip6h, ira);
12264 	}
12265 }
12266 
12267 /*
12268  * Select which AH & ESP SA's to use (if any) for the outbound packet.
12269  *
12270  * If this function returns B_TRUE, the requested SA's have been filled
12271  * into the ixa_ipsec_*_sa pointers.
12272  *
12273  * If the function returns B_FALSE, the packet has been "consumed", most
12274  * likely by an ACQUIRE sent up via PF_KEY to a key management daemon.
12275  *
12276  * The SA references created by the protocol-specific "select"
12277  * function will be released in ip_output_post_ipsec.
12278  */
12279 static boolean_t
ipsec_out_select_sa(mblk_t * mp,ip_xmit_attr_t * ixa)12280 ipsec_out_select_sa(mblk_t *mp, ip_xmit_attr_t *ixa)
12281 {
12282 	boolean_t need_ah_acquire = B_FALSE, need_esp_acquire = B_FALSE;
12283 	ipsec_policy_t *pp;
12284 	ipsec_action_t *ap;
12285 
12286 	ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12287 	ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12288 	    (ixa->ixa_ipsec_action != NULL));
12289 
12290 	ap = ixa->ixa_ipsec_action;
12291 	if (ap == NULL) {
12292 		pp = ixa->ixa_ipsec_policy;
12293 		ASSERT(pp != NULL);
12294 		ap = pp->ipsp_act;
12295 		ASSERT(ap != NULL);
12296 	}
12297 
12298 	/*
12299 	 * We have an action.  now, let's select SA's.
12300 	 * A side effect of setting ixa_ipsec_*_sa is that it will
12301 	 * be cached in the conn_t.
12302 	 */
12303 	if (ap->ipa_want_esp) {
12304 		if (ixa->ixa_ipsec_esp_sa == NULL) {
12305 			need_esp_acquire = !ipsec_outbound_sa(mp, ixa,
12306 			    IPPROTO_ESP);
12307 		}
12308 		ASSERT(need_esp_acquire || ixa->ixa_ipsec_esp_sa != NULL);
12309 	}
12310 
12311 	if (ap->ipa_want_ah) {
12312 		if (ixa->ixa_ipsec_ah_sa == NULL) {
12313 			need_ah_acquire = !ipsec_outbound_sa(mp, ixa,
12314 			    IPPROTO_AH);
12315 		}
12316 		ASSERT(need_ah_acquire || ixa->ixa_ipsec_ah_sa != NULL);
12317 		/*
12318 		 * The ESP and AH processing order needs to be preserved
12319 		 * when both protocols are required (ESP should be applied
12320 		 * before AH for an outbound packet). Force an ESP ACQUIRE
12321 		 * when both ESP and AH are required, and an AH ACQUIRE
12322 		 * is needed.
12323 		 */
12324 		if (ap->ipa_want_esp && need_ah_acquire)
12325 			need_esp_acquire = B_TRUE;
12326 	}
12327 
12328 	/*
12329 	 * Send an ACQUIRE (extended, regular, or both) if we need one.
12330 	 * Release SAs that got referenced, but will not be used until we
12331 	 * acquire _all_ of the SAs we need.
12332 	 */
12333 	if (need_ah_acquire || need_esp_acquire) {
12334 		if (ixa->ixa_ipsec_ah_sa != NULL) {
12335 			IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12336 			ixa->ixa_ipsec_ah_sa = NULL;
12337 		}
12338 		if (ixa->ixa_ipsec_esp_sa != NULL) {
12339 			IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12340 			ixa->ixa_ipsec_esp_sa = NULL;
12341 		}
12342 
12343 		sadb_acquire(mp, ixa, need_ah_acquire, need_esp_acquire);
12344 		return (B_FALSE);
12345 	}
12346 
12347 	return (B_TRUE);
12348 }
12349 
12350 /*
12351  * Handle IPsec output processing.
12352  * This function is only entered once for a given packet.
12353  * We try to do things synchronously, but if we need to have user-level
12354  * set up SAs, or ESP or AH uses asynchronous kEF, then the operation
12355  * will be completed
12356  *  - when the SAs are added in esp_add_sa_finish/ah_add_sa_finish
12357  *  - when asynchronous ESP is done it will do AH
12358  *
12359  * In all cases we come back in ip_output_post_ipsec() to fragment and
12360  * send out the packet.
12361  */
12362 int
ipsec_out_process(mblk_t * mp,ip_xmit_attr_t * ixa)12363 ipsec_out_process(mblk_t *mp, ip_xmit_attr_t *ixa)
12364 {
12365 	ill_t		*ill = ixa->ixa_nce->nce_ill;
12366 	ip_stack_t	*ipst = ixa->ixa_ipst;
12367 	ipsec_stack_t	*ipss;
12368 	ipsec_policy_t	*pp;
12369 	ipsec_action_t	*ap;
12370 
12371 	ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12372 
12373 	ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12374 	    (ixa->ixa_ipsec_action != NULL));
12375 
12376 	ipss = ipst->ips_netstack->netstack_ipsec;
12377 	if (!ipsec_loaded(ipss)) {
12378 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12379 		ip_drop_packet(mp, B_TRUE, ill,
12380 		    DROPPER(ipss, ipds_ip_ipsec_not_loaded),
12381 		    &ipss->ipsec_dropper);
12382 		return (ENOTSUP);
12383 	}
12384 
12385 	ap = ixa->ixa_ipsec_action;
12386 	if (ap == NULL) {
12387 		pp = ixa->ixa_ipsec_policy;
12388 		ASSERT(pp != NULL);
12389 		ap = pp->ipsp_act;
12390 		ASSERT(ap != NULL);
12391 	}
12392 
12393 	/* Handle explicit drop action and bypass. */
12394 	switch (ap->ipa_act.ipa_type) {
12395 	case IPSEC_ACT_DISCARD:
12396 	case IPSEC_ACT_REJECT:
12397 		ip_drop_packet(mp, B_FALSE, ill,
12398 		    DROPPER(ipss, ipds_spd_explicit), &ipss->ipsec_spd_dropper);
12399 		return (EHOSTUNREACH);	/* IPsec policy failure */
12400 	case IPSEC_ACT_BYPASS:
12401 		return (ip_output_post_ipsec(mp, ixa));
12402 	}
12403 
12404 	/*
12405 	 * The order of processing is first insert a IP header if needed.
12406 	 * Then insert the ESP header and then the AH header.
12407 	 */
12408 	if ((ixa->ixa_flags & IXAF_IS_IPV4) && ap->ipa_want_se) {
12409 		/*
12410 		 * First get the outer IP header before sending
12411 		 * it to ESP.
12412 		 */
12413 		ipha_t *oipha, *iipha;
12414 		mblk_t *outer_mp, *inner_mp;
12415 
12416 		if ((outer_mp = allocb(sizeof (ipha_t), BPRI_HI)) == NULL) {
12417 			(void) mi_strlog(ill->ill_rq, 0,
12418 			    SL_ERROR|SL_TRACE|SL_CONSOLE,
12419 			    "ipsec_out_process: "
12420 			    "Self-Encapsulation failed: Out of memory\n");
12421 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12422 			ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12423 			freemsg(mp);
12424 			return (ENOBUFS);
12425 		}
12426 		inner_mp = mp;
12427 		ASSERT(inner_mp->b_datap->db_type == M_DATA);
12428 		oipha = (ipha_t *)outer_mp->b_rptr;
12429 		iipha = (ipha_t *)inner_mp->b_rptr;
12430 		*oipha = *iipha;
12431 		outer_mp->b_wptr += sizeof (ipha_t);
12432 		oipha->ipha_length = htons(ntohs(iipha->ipha_length) +
12433 		    sizeof (ipha_t));
12434 		oipha->ipha_protocol = IPPROTO_ENCAP;
12435 		oipha->ipha_version_and_hdr_length =
12436 		    IP_SIMPLE_HDR_VERSION;
12437 		oipha->ipha_hdr_checksum = 0;
12438 		oipha->ipha_hdr_checksum = ip_csum_hdr(oipha);
12439 		outer_mp->b_cont = inner_mp;
12440 		mp = outer_mp;
12441 
12442 		ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
12443 	}
12444 
12445 	/* If we need to wait for a SA then we can't return any errno */
12446 	if (((ap->ipa_want_ah && (ixa->ixa_ipsec_ah_sa == NULL)) ||
12447 	    (ap->ipa_want_esp && (ixa->ixa_ipsec_esp_sa == NULL))) &&
12448 	    !ipsec_out_select_sa(mp, ixa))
12449 		return (0);
12450 
12451 	/*
12452 	 * By now, we know what SA's to use.  Toss over to ESP & AH
12453 	 * to do the heavy lifting.
12454 	 */
12455 	if (ap->ipa_want_esp) {
12456 		ASSERT(ixa->ixa_ipsec_esp_sa != NULL);
12457 
12458 		mp = ixa->ixa_ipsec_esp_sa->ipsa_output_func(mp, ixa);
12459 		if (mp == NULL) {
12460 			/*
12461 			 * Either it failed or is pending. In the former case
12462 			 * ipIfStatsInDiscards was increased.
12463 			 */
12464 			return (0);
12465 		}
12466 	}
12467 
12468 	if (ap->ipa_want_ah) {
12469 		ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
12470 
12471 		mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(mp, ixa);
12472 		if (mp == NULL) {
12473 			/*
12474 			 * Either it failed or is pending. In the former case
12475 			 * ipIfStatsInDiscards was increased.
12476 			 */
12477 			return (0);
12478 		}
12479 	}
12480 	/*
12481 	 * We are done with IPsec processing. Send it over
12482 	 * the wire.
12483 	 */
12484 	return (ip_output_post_ipsec(mp, ixa));
12485 }
12486 
12487 /*
12488  * ioctls that go through a down/up sequence may need to wait for the down
12489  * to complete. This involves waiting for the ire and ipif refcnts to go down
12490  * to zero. Subsequently the ioctl is restarted from ipif_ill_refrele_tail.
12491  */
12492 /* ARGSUSED */
12493 void
ip_reprocess_ioctl(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * dummy_arg)12494 ip_reprocess_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
12495 {
12496 	struct iocblk *iocp;
12497 	mblk_t *mp1;
12498 	ip_ioctl_cmd_t *ipip;
12499 	int err;
12500 	sin_t	*sin;
12501 	struct lifreq *lifr;
12502 	struct ifreq *ifr;
12503 
12504 	iocp = (struct iocblk *)mp->b_rptr;
12505 	ASSERT(ipsq != NULL);
12506 	/* Existence of mp1 verified in ip_wput_nondata */
12507 	mp1 = mp->b_cont->b_cont;
12508 	ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12509 	if (ipip->ipi_cmd == SIOCSLIFNAME || ipip->ipi_cmd == IF_UNITSEL) {
12510 		/*
12511 		 * Special case where ipx_current_ipif is not set:
12512 		 * ill_phyint_reinit merged the v4 and v6 into a single ipsq.
12513 		 * We are here as were not able to complete the operation in
12514 		 * ipif_set_values because we could not become exclusive on
12515 		 * the new ipsq.
12516 		 */
12517 		ill_t *ill = q->q_ptr;
12518 		ipsq_current_start(ipsq, ill->ill_ipif, ipip->ipi_cmd);
12519 	}
12520 	ASSERT(ipsq->ipsq_xop->ipx_current_ipif != NULL);
12521 
12522 	if (ipip->ipi_cmd_type == IF_CMD) {
12523 		/* This a old style SIOC[GS]IF* command */
12524 		ifr = (struct ifreq *)mp1->b_rptr;
12525 		sin = (sin_t *)&ifr->ifr_addr;
12526 	} else if (ipip->ipi_cmd_type == LIF_CMD) {
12527 		/* This a new style SIOC[GS]LIF* command */
12528 		lifr = (struct lifreq *)mp1->b_rptr;
12529 		sin = (sin_t *)&lifr->lifr_addr;
12530 	} else {
12531 		sin = NULL;
12532 	}
12533 
12534 	err = (*ipip->ipi_func_restart)(ipsq->ipsq_xop->ipx_current_ipif, sin,
12535 	    q, mp, ipip, mp1->b_rptr);
12536 
12537 	DTRACE_PROBE4(ipif__ioctl, char *, "ip_reprocess_ioctl finish",
12538 	    int, ipip->ipi_cmd,
12539 	    ill_t *, ipsq->ipsq_xop->ipx_current_ipif->ipif_ill,
12540 	    ipif_t *, ipsq->ipsq_xop->ipx_current_ipif);
12541 
12542 	ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12543 }
12544 
12545 /*
12546  * ioctl processing
12547  *
12548  * ioctl processing starts with ip_sioctl_copyin_setup(), which looks up
12549  * the ioctl command in the ioctl tables, determines the copyin data size
12550  * from the ipi_copyin_size field, and does an mi_copyin() of that size.
12551  *
12552  * ioctl processing then continues when the M_IOCDATA makes its way down to
12553  * ip_wput_nondata().  The ioctl is looked up again in the ioctl table, its
12554  * associated 'conn' is refheld till the end of the ioctl and the general
12555  * ioctl processing function ip_process_ioctl() is called to extract the
12556  * arguments and process the ioctl.  To simplify extraction, ioctl commands
12557  * are "typed" based on the arguments they take (e.g., LIF_CMD which takes a
12558  * `struct lifreq'), and a common extract function (e.g., ip_extract_lifreq())
12559  * is used to extract the ioctl's arguments.
12560  *
12561  * ip_process_ioctl determines if the ioctl needs to be serialized, and if
12562  * so goes thru the serialization primitive ipsq_try_enter. Then the
12563  * appropriate function to handle the ioctl is called based on the entry in
12564  * the ioctl table. ioctl completion is encapsulated in ip_ioctl_finish
12565  * which also refreleases the 'conn' that was refheld at the start of the
12566  * ioctl. Finally ipsq_exit is called if needed to exit the ipsq.
12567  *
12568  * Many exclusive ioctls go thru an internal down up sequence as part of
12569  * the operation. For example an attempt to change the IP address of an
12570  * ipif entails ipif_down, set address, ipif_up. Bringing down the interface
12571  * does all the cleanup such as deleting all ires that use this address.
12572  * Then we need to wait till all references to the interface go away.
12573  */
12574 void
ip_process_ioctl(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * arg)12575 ip_process_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *arg)
12576 {
12577 	struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
12578 	ip_ioctl_cmd_t *ipip = arg;
12579 	ip_extract_func_t *extract_funcp;
12580 	cmd_info_t ci;
12581 	int err;
12582 	boolean_t entered_ipsq = B_FALSE;
12583 
12584 	ip3dbg(("ip_process_ioctl: ioctl %X\n", iocp->ioc_cmd));
12585 
12586 	if (ipip == NULL)
12587 		ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12588 
12589 	/*
12590 	 * SIOCLIFADDIF needs to go thru a special path since the
12591 	 * ill may not exist yet. This happens in the case of lo0
12592 	 * which is created using this ioctl.
12593 	 */
12594 	if (ipip->ipi_cmd == SIOCLIFADDIF) {
12595 		err = ip_sioctl_addif(NULL, NULL, q, mp, NULL, NULL);
12596 		DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish",
12597 		    int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12598 		ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12599 		return;
12600 	}
12601 
12602 	ci.ci_ipif = NULL;
12603 	switch (ipip->ipi_cmd_type) {
12604 	case MISC_CMD:
12605 	case MSFILT_CMD:
12606 		/*
12607 		 * All MISC_CMD ioctls come in here -- e.g. SIOCGLIFCONF.
12608 		 */
12609 		if (ipip->ipi_cmd == IF_UNITSEL) {
12610 			/* ioctl comes down the ill */
12611 			ci.ci_ipif = ((ill_t *)q->q_ptr)->ill_ipif;
12612 			ipif_refhold(ci.ci_ipif);
12613 		}
12614 		err = 0;
12615 		ci.ci_sin = NULL;
12616 		ci.ci_sin6 = NULL;
12617 		ci.ci_lifr = NULL;
12618 		extract_funcp = NULL;
12619 		break;
12620 
12621 	case IF_CMD:
12622 	case LIF_CMD:
12623 		extract_funcp = ip_extract_lifreq;
12624 		break;
12625 
12626 	case ARP_CMD:
12627 	case XARP_CMD:
12628 		extract_funcp = ip_extract_arpreq;
12629 		break;
12630 
12631 	default:
12632 		ASSERT(0);
12633 	}
12634 
12635 	if (extract_funcp != NULL) {
12636 		err = (*extract_funcp)(q, mp, ipip, &ci);
12637 		if (err != 0) {
12638 			DTRACE_PROBE4(ipif__ioctl,
12639 			    char *, "ip_process_ioctl finish err",
12640 			    int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12641 			ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12642 			return;
12643 		}
12644 
12645 		/*
12646 		 * All of the extraction functions return a refheld ipif.
12647 		 */
12648 		ASSERT(ci.ci_ipif != NULL);
12649 	}
12650 
12651 	if (!(ipip->ipi_flags & IPI_WR)) {
12652 		/*
12653 		 * A return value of EINPROGRESS means the ioctl is
12654 		 * either queued and waiting for some reason or has
12655 		 * already completed.
12656 		 */
12657 		err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip,
12658 		    ci.ci_lifr);
12659 		if (ci.ci_ipif != NULL) {
12660 			DTRACE_PROBE4(ipif__ioctl,
12661 			    char *, "ip_process_ioctl finish RD",
12662 			    int, ipip->ipi_cmd, ill_t *, ci.ci_ipif->ipif_ill,
12663 			    ipif_t *, ci.ci_ipif);
12664 			ipif_refrele(ci.ci_ipif);
12665 		} else {
12666 			DTRACE_PROBE4(ipif__ioctl,
12667 			    char *, "ip_process_ioctl finish RD",
12668 			    int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12669 		}
12670 		ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12671 		return;
12672 	}
12673 
12674 	ASSERT(ci.ci_ipif != NULL);
12675 
12676 	/*
12677 	 * If ipsq is non-NULL, we are already being called exclusively
12678 	 */
12679 	ASSERT(ipsq == NULL || IAM_WRITER_IPSQ(ipsq));
12680 	if (ipsq == NULL) {
12681 		ipsq = ipsq_try_enter(ci.ci_ipif, NULL, q, mp, ip_process_ioctl,
12682 		    NEW_OP, B_TRUE);
12683 		if (ipsq == NULL) {
12684 			ipif_refrele(ci.ci_ipif);
12685 			return;
12686 		}
12687 		entered_ipsq = B_TRUE;
12688 	}
12689 	/*
12690 	 * Release the ipif so that ipif_down and friends that wait for
12691 	 * references to go away are not misled about the current ipif_refcnt
12692 	 * values. We are writer so we can access the ipif even after releasing
12693 	 * the ipif.
12694 	 */
12695 	ipif_refrele(ci.ci_ipif);
12696 
12697 	ipsq_current_start(ipsq, ci.ci_ipif, ipip->ipi_cmd);
12698 
12699 	/*
12700 	 * A return value of EINPROGRESS means the ioctl is
12701 	 * either queued and waiting for some reason or has
12702 	 * already completed.
12703 	 */
12704 	err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip, ci.ci_lifr);
12705 
12706 	DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish WR",
12707 	    int, ipip->ipi_cmd,
12708 	    ill_t *, ci.ci_ipif == NULL ? NULL : ci.ci_ipif->ipif_ill,
12709 	    ipif_t *, ci.ci_ipif);
12710 	ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12711 
12712 	if (entered_ipsq)
12713 		ipsq_exit(ipsq);
12714 }
12715 
12716 /*
12717  * Complete the ioctl. Typically ioctls use the mi package and need to
12718  * do mi_copyout/mi_copy_done.
12719  */
12720 void
ip_ioctl_finish(queue_t * q,mblk_t * mp,int err,int mode,ipsq_t * ipsq)12721 ip_ioctl_finish(queue_t *q, mblk_t *mp, int err, int mode, ipsq_t *ipsq)
12722 {
12723 	conn_t	*connp = NULL;
12724 
12725 	if (err == EINPROGRESS)
12726 		return;
12727 
12728 	if (CONN_Q(q)) {
12729 		connp = Q_TO_CONN(q);
12730 		ASSERT(connp->conn_ref >= 2);
12731 	}
12732 
12733 	switch (mode) {
12734 	case COPYOUT:
12735 		if (err == 0)
12736 			mi_copyout(q, mp);
12737 		else
12738 			mi_copy_done(q, mp, err);
12739 		break;
12740 
12741 	case NO_COPYOUT:
12742 		mi_copy_done(q, mp, err);
12743 		break;
12744 
12745 	default:
12746 		ASSERT(mode == CONN_CLOSE);	/* aborted through CONN_CLOSE */
12747 		break;
12748 	}
12749 
12750 	/*
12751 	 * The conn refhold and ioctlref placed on the conn at the start of the
12752 	 * ioctl are released here.
12753 	 */
12754 	if (connp != NULL) {
12755 		CONN_DEC_IOCTLREF(connp);
12756 		CONN_OPER_PENDING_DONE(connp);
12757 	}
12758 
12759 	if (ipsq != NULL)
12760 		ipsq_current_finish(ipsq);
12761 }
12762 
12763 /* Handles all non data messages */
12764 void
ip_wput_nondata(queue_t * q,mblk_t * mp)12765 ip_wput_nondata(queue_t *q, mblk_t *mp)
12766 {
12767 	mblk_t		*mp1;
12768 	struct iocblk	*iocp;
12769 	ip_ioctl_cmd_t	*ipip;
12770 	conn_t		*connp;
12771 	cred_t		*cr;
12772 	char		*proto_str;
12773 
12774 	if (CONN_Q(q))
12775 		connp = Q_TO_CONN(q);
12776 	else
12777 		connp = NULL;
12778 
12779 	switch (DB_TYPE(mp)) {
12780 	case M_IOCTL:
12781 		/*
12782 		 * IOCTL processing begins in ip_sioctl_copyin_setup which
12783 		 * will arrange to copy in associated control structures.
12784 		 */
12785 		ip_sioctl_copyin_setup(q, mp);
12786 		return;
12787 	case M_IOCDATA:
12788 		/*
12789 		 * Ensure that this is associated with one of our trans-
12790 		 * parent ioctls.  If it's not ours, discard it if we're
12791 		 * running as a driver, or pass it on if we're a module.
12792 		 */
12793 		iocp = (struct iocblk *)mp->b_rptr;
12794 		ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12795 		if (ipip == NULL) {
12796 			if (q->q_next == NULL) {
12797 				goto nak;
12798 			} else {
12799 				putnext(q, mp);
12800 			}
12801 			return;
12802 		}
12803 		if ((q->q_next != NULL) && !(ipip->ipi_flags & IPI_MODOK)) {
12804 			/*
12805 			 * The ioctl is one we recognise, but is not consumed
12806 			 * by IP as a module and we are a module, so we drop
12807 			 */
12808 			goto nak;
12809 		}
12810 
12811 		/* IOCTL continuation following copyin or copyout. */
12812 		if (mi_copy_state(q, mp, NULL) == -1) {
12813 			/*
12814 			 * The copy operation failed.  mi_copy_state already
12815 			 * cleaned up, so we're out of here.
12816 			 */
12817 			return;
12818 		}
12819 		/*
12820 		 * If we just completed a copy in, we become writer and
12821 		 * continue processing in ip_sioctl_copyin_done.  If it
12822 		 * was a copy out, we call mi_copyout again.  If there is
12823 		 * nothing more to copy out, it will complete the IOCTL.
12824 		 */
12825 		if (MI_COPY_DIRECTION(mp) == MI_COPY_IN) {
12826 			if (!(mp1 = mp->b_cont) || !(mp1 = mp1->b_cont)) {
12827 				mi_copy_done(q, mp, EPROTO);
12828 				return;
12829 			}
12830 			/*
12831 			 * Check for cases that need more copying.  A return
12832 			 * value of 0 means a second copyin has been started,
12833 			 * so we return; a return value of 1 means no more
12834 			 * copying is needed, so we continue.
12835 			 */
12836 			if (ipip->ipi_cmd_type == MSFILT_CMD &&
12837 			    MI_COPY_COUNT(mp) == 1) {
12838 				if (ip_copyin_msfilter(q, mp) == 0)
12839 					return;
12840 			}
12841 			/*
12842 			 * Refhold the conn, till the ioctl completes. This is
12843 			 * needed in case the ioctl ends up in the pending mp
12844 			 * list. Every mp in the ipx_pending_mp list must have
12845 			 * a refhold on the conn to resume processing. The
12846 			 * refhold is released when the ioctl completes
12847 			 * (whether normally or abnormally). An ioctlref is also
12848 			 * placed on the conn to prevent TCP from removing the
12849 			 * queue needed to send the ioctl reply back.
12850 			 * In all cases ip_ioctl_finish is called to finish
12851 			 * the ioctl and release the refholds.
12852 			 */
12853 			if (connp != NULL) {
12854 				/* This is not a reentry */
12855 				CONN_INC_REF(connp);
12856 				CONN_INC_IOCTLREF(connp);
12857 			} else {
12858 				if (!(ipip->ipi_flags & IPI_MODOK)) {
12859 					mi_copy_done(q, mp, EINVAL);
12860 					return;
12861 				}
12862 			}
12863 
12864 			ip_process_ioctl(NULL, q, mp, ipip);
12865 
12866 		} else {
12867 			mi_copyout(q, mp);
12868 		}
12869 		return;
12870 
12871 	case M_IOCNAK:
12872 		/*
12873 		 * The only way we could get here is if a resolver didn't like
12874 		 * an IOCTL we sent it.	 This shouldn't happen.
12875 		 */
12876 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
12877 		    "ip_wput_nondata: unexpected M_IOCNAK, ioc_cmd 0x%x",
12878 		    ((struct iocblk *)mp->b_rptr)->ioc_cmd);
12879 		freemsg(mp);
12880 		return;
12881 	case M_IOCACK:
12882 		/* /dev/ip shouldn't see this */
12883 		goto nak;
12884 	case M_FLUSH:
12885 		if (*mp->b_rptr & FLUSHW)
12886 			flushq(q, FLUSHALL);
12887 		if (q->q_next) {
12888 			putnext(q, mp);
12889 			return;
12890 		}
12891 		if (*mp->b_rptr & FLUSHR) {
12892 			*mp->b_rptr &= ~FLUSHW;
12893 			qreply(q, mp);
12894 			return;
12895 		}
12896 		freemsg(mp);
12897 		return;
12898 	case M_CTL:
12899 		break;
12900 	case M_PROTO:
12901 	case M_PCPROTO:
12902 		/*
12903 		 * The only PROTO messages we expect are SNMP-related.
12904 		 */
12905 		switch (((union T_primitives *)mp->b_rptr)->type) {
12906 		case T_SVR4_OPTMGMT_REQ:
12907 			ip2dbg(("ip_wput_nondata: T_SVR4_OPTMGMT_REQ "
12908 			    "flags %x\n",
12909 			    ((struct T_optmgmt_req *)mp->b_rptr)->MGMT_flags));
12910 
12911 			if (connp == NULL) {
12912 				proto_str = "T_SVR4_OPTMGMT_REQ";
12913 				goto protonak;
12914 			}
12915 
12916 			/*
12917 			 * All Solaris components should pass a db_credp
12918 			 * for this TPI message, hence we ASSERT.
12919 			 * But in case there is some other M_PROTO that looks
12920 			 * like a TPI message sent by some other kernel
12921 			 * component, we check and return an error.
12922 			 */
12923 			cr = msg_getcred(mp, NULL);
12924 			ASSERT(cr != NULL);
12925 			if (cr == NULL) {
12926 				mp = mi_tpi_err_ack_alloc(mp, TSYSERR, EINVAL);
12927 				if (mp != NULL)
12928 					qreply(q, mp);
12929 				return;
12930 			}
12931 
12932 			if (!snmpcom_req(q, mp, ip_snmp_set, ip_snmp_get, cr)) {
12933 				proto_str = "Bad SNMPCOM request?";
12934 				goto protonak;
12935 			}
12936 			return;
12937 		default:
12938 			ip1dbg(("ip_wput_nondata: dropping M_PROTO prim %u\n",
12939 			    (int)*(uint_t *)mp->b_rptr));
12940 			freemsg(mp);
12941 			return;
12942 		}
12943 	default:
12944 		break;
12945 	}
12946 	if (q->q_next) {
12947 		putnext(q, mp);
12948 	} else
12949 		freemsg(mp);
12950 	return;
12951 
12952 nak:
12953 	iocp->ioc_error = EINVAL;
12954 	mp->b_datap->db_type = M_IOCNAK;
12955 	iocp->ioc_count = 0;
12956 	qreply(q, mp);
12957 	return;
12958 
12959 protonak:
12960 	cmn_err(CE_NOTE, "IP doesn't process %s as a module", proto_str);
12961 	if ((mp = mi_tpi_err_ack_alloc(mp, TPROTO, EINVAL)) != NULL)
12962 		qreply(q, mp);
12963 }
12964 
12965 /*
12966  * Process IP options in an outbound packet.  Verify that the nexthop in a
12967  * strict source route is onlink.
12968  * Returns non-zero if something fails in which case an ICMP error has been
12969  * sent and mp freed.
12970  *
12971  * Assumes the ULP has called ip_massage_options to move nexthop into ipha_dst.
12972  */
12973 int
ip_output_options(mblk_t * mp,ipha_t * ipha,ip_xmit_attr_t * ixa,ill_t * ill)12974 ip_output_options(mblk_t *mp, ipha_t *ipha, ip_xmit_attr_t *ixa, ill_t *ill)
12975 {
12976 	ipoptp_t	opts;
12977 	uchar_t		*opt;
12978 	uint8_t		optval;
12979 	uint8_t		optlen;
12980 	ipaddr_t	dst;
12981 	intptr_t	code = 0;
12982 	ire_t		*ire;
12983 	ip_stack_t	*ipst = ixa->ixa_ipst;
12984 	ip_recv_attr_t	iras;
12985 
12986 	ip2dbg(("ip_output_options\n"));
12987 
12988 	dst = ipha->ipha_dst;
12989 	for (optval = ipoptp_first(&opts, ipha);
12990 	    optval != IPOPT_EOL;
12991 	    optval = ipoptp_next(&opts)) {
12992 		opt = opts.ipoptp_cur;
12993 		optlen = opts.ipoptp_len;
12994 		ip2dbg(("ip_output_options: opt %d, len %d\n",
12995 		    optval, optlen));
12996 		switch (optval) {
12997 			uint32_t off;
12998 		case IPOPT_SSRR:
12999 		case IPOPT_LSRR:
13000 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13001 				ip1dbg((
13002 				    "ip_output_options: bad option offset\n"));
13003 				code = (char *)&opt[IPOPT_OLEN] -
13004 				    (char *)ipha;
13005 				goto param_prob;
13006 			}
13007 			off = opt[IPOPT_OFFSET];
13008 			ip1dbg(("ip_output_options: next hop 0x%x\n",
13009 			    ntohl(dst)));
13010 			/*
13011 			 * For strict: verify that dst is directly
13012 			 * reachable.
13013 			 */
13014 			if (optval == IPOPT_SSRR) {
13015 				ire = ire_ftable_lookup_v4(dst, 0, 0,
13016 				    IRE_INTERFACE, NULL, ALL_ZONES,
13017 				    ixa->ixa_tsl,
13018 				    MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
13019 				    NULL);
13020 				if (ire == NULL) {
13021 					ip1dbg(("ip_output_options: SSRR not"
13022 					    " directly reachable: 0x%x\n",
13023 					    ntohl(dst)));
13024 					goto bad_src_route;
13025 				}
13026 				ire_refrele(ire);
13027 			}
13028 			break;
13029 		case IPOPT_RR:
13030 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13031 				ip1dbg((
13032 				    "ip_output_options: bad option offset\n"));
13033 				code = (char *)&opt[IPOPT_OLEN] -
13034 				    (char *)ipha;
13035 				goto param_prob;
13036 			}
13037 			break;
13038 		case IPOPT_TS:
13039 			/*
13040 			 * Verify that length >=5 and that there is either
13041 			 * room for another timestamp or that the overflow
13042 			 * counter is not maxed out.
13043 			 */
13044 			code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
13045 			if (optlen < IPOPT_MINLEN_IT) {
13046 				goto param_prob;
13047 			}
13048 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13049 				ip1dbg((
13050 				    "ip_output_options: bad option offset\n"));
13051 				code = (char *)&opt[IPOPT_OFFSET] -
13052 				    (char *)ipha;
13053 				goto param_prob;
13054 			}
13055 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
13056 			case IPOPT_TS_TSONLY:
13057 				off = IPOPT_TS_TIMELEN;
13058 				break;
13059 			case IPOPT_TS_TSANDADDR:
13060 			case IPOPT_TS_PRESPEC:
13061 			case IPOPT_TS_PRESPEC_RFC791:
13062 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
13063 				break;
13064 			default:
13065 				code = (char *)&opt[IPOPT_POS_OV_FLG] -
13066 				    (char *)ipha;
13067 				goto param_prob;
13068 			}
13069 			if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
13070 			    (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
13071 				/*
13072 				 * No room and the overflow counter is 15
13073 				 * already.
13074 				 */
13075 				goto param_prob;
13076 			}
13077 			break;
13078 		}
13079 	}
13080 
13081 	if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0)
13082 		return (0);
13083 
13084 	ip1dbg(("ip_output_options: error processing IP options."));
13085 	code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
13086 
13087 param_prob:
13088 	bzero(&iras, sizeof (iras));
13089 	iras.ira_ill = iras.ira_rill = ill;
13090 	iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13091 	iras.ira_rifindex = iras.ira_ruifindex;
13092 	iras.ira_flags = IRAF_IS_IPV4;
13093 
13094 	ip_drop_output("ip_output_options", mp, ill);
13095 	icmp_param_problem(mp, (uint8_t)code, &iras);
13096 	ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13097 	return (-1);
13098 
13099 bad_src_route:
13100 	bzero(&iras, sizeof (iras));
13101 	iras.ira_ill = iras.ira_rill = ill;
13102 	iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13103 	iras.ira_rifindex = iras.ira_ruifindex;
13104 	iras.ira_flags = IRAF_IS_IPV4;
13105 
13106 	ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
13107 	icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, &iras);
13108 	ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13109 	return (-1);
13110 }
13111 
13112 /*
13113  * The maximum value of conn_drain_list_cnt is CONN_MAXDRAINCNT.
13114  * conn_drain_list_cnt can be changed by setting conn_drain_nthreads
13115  * thru /etc/system.
13116  */
13117 #define	CONN_MAXDRAINCNT	64
13118 
13119 static void
conn_drain_init(ip_stack_t * ipst)13120 conn_drain_init(ip_stack_t *ipst)
13121 {
13122 	int i, j;
13123 	idl_tx_list_t *itl_tx;
13124 
13125 	ipst->ips_conn_drain_list_cnt = conn_drain_nthreads;
13126 
13127 	if ((ipst->ips_conn_drain_list_cnt == 0) ||
13128 	    (ipst->ips_conn_drain_list_cnt > CONN_MAXDRAINCNT)) {
13129 		/*
13130 		 * Default value of the number of drainers is the
13131 		 * number of cpus, subject to maximum of 8 drainers.
13132 		 */
13133 		if (boot_max_ncpus != -1)
13134 			ipst->ips_conn_drain_list_cnt = MIN(boot_max_ncpus, 8);
13135 		else
13136 			ipst->ips_conn_drain_list_cnt = MIN(max_ncpus, 8);
13137 	}
13138 
13139 	ipst->ips_idl_tx_list =
13140 	    kmem_zalloc(TX_FANOUT_SIZE * sizeof (idl_tx_list_t), KM_SLEEP);
13141 	for (i = 0; i < TX_FANOUT_SIZE; i++) {
13142 		itl_tx =  &ipst->ips_idl_tx_list[i];
13143 		itl_tx->txl_drain_list =
13144 		    kmem_zalloc(ipst->ips_conn_drain_list_cnt *
13145 		    sizeof (idl_t), KM_SLEEP);
13146 		mutex_init(&itl_tx->txl_lock, NULL, MUTEX_DEFAULT, NULL);
13147 		for (j = 0; j < ipst->ips_conn_drain_list_cnt; j++) {
13148 			mutex_init(&itl_tx->txl_drain_list[j].idl_lock, NULL,
13149 			    MUTEX_DEFAULT, NULL);
13150 			itl_tx->txl_drain_list[j].idl_itl = itl_tx;
13151 		}
13152 	}
13153 }
13154 
13155 static void
conn_drain_fini(ip_stack_t * ipst)13156 conn_drain_fini(ip_stack_t *ipst)
13157 {
13158 	int i;
13159 	idl_tx_list_t *itl_tx;
13160 
13161 	for (i = 0; i < TX_FANOUT_SIZE; i++) {
13162 		itl_tx =  &ipst->ips_idl_tx_list[i];
13163 		kmem_free(itl_tx->txl_drain_list,
13164 		    ipst->ips_conn_drain_list_cnt * sizeof (idl_t));
13165 	}
13166 	kmem_free(ipst->ips_idl_tx_list,
13167 	    TX_FANOUT_SIZE * sizeof (idl_tx_list_t));
13168 	ipst->ips_idl_tx_list = NULL;
13169 }
13170 
13171 /*
13172  * Flow control has blocked us from proceeding.  Insert the given conn in one
13173  * of the conn drain lists.  When flow control is unblocked, either ip_wsrv()
13174  * (STREAMS) or ill_flow_enable() (direct) will be called back, which in turn
13175  * will call conn_walk_drain().  See the flow control notes at the top of this
13176  * file for more details.
13177  */
13178 void
conn_drain_insert(conn_t * connp,idl_tx_list_t * tx_list)13179 conn_drain_insert(conn_t *connp, idl_tx_list_t *tx_list)
13180 {
13181 	idl_t	*idl = tx_list->txl_drain_list;
13182 	uint_t	index;
13183 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
13184 
13185 	mutex_enter(&connp->conn_lock);
13186 	if (connp->conn_state_flags & CONN_CLOSING) {
13187 		/*
13188 		 * The conn is closing as a result of which CONN_CLOSING
13189 		 * is set. Return.
13190 		 */
13191 		mutex_exit(&connp->conn_lock);
13192 		return;
13193 	} else if (connp->conn_idl == NULL) {
13194 		/*
13195 		 * Assign the next drain list round robin. We dont' use
13196 		 * a lock, and thus it may not be strictly round robin.
13197 		 * Atomicity of load/stores is enough to make sure that
13198 		 * conn_drain_list_index is always within bounds.
13199 		 */
13200 		index = tx_list->txl_drain_index;
13201 		ASSERT(index < ipst->ips_conn_drain_list_cnt);
13202 		connp->conn_idl = &tx_list->txl_drain_list[index];
13203 		index++;
13204 		if (index == ipst->ips_conn_drain_list_cnt)
13205 			index = 0;
13206 		tx_list->txl_drain_index = index;
13207 	} else {
13208 		ASSERT(connp->conn_idl->idl_itl == tx_list);
13209 	}
13210 	mutex_exit(&connp->conn_lock);
13211 
13212 	idl = connp->conn_idl;
13213 	mutex_enter(&idl->idl_lock);
13214 	if ((connp->conn_drain_prev != NULL) ||
13215 	    (connp->conn_state_flags & CONN_CLOSING)) {
13216 		/*
13217 		 * The conn is either already in the drain list or closing.
13218 		 * (We needed to check for CONN_CLOSING again since close can
13219 		 * sneak in between dropping conn_lock and acquiring idl_lock.)
13220 		 */
13221 		mutex_exit(&idl->idl_lock);
13222 		return;
13223 	}
13224 
13225 	/*
13226 	 * The conn is not in the drain list. Insert it at the
13227 	 * tail of the drain list. The drain list is circular
13228 	 * and doubly linked. idl_conn points to the 1st element
13229 	 * in the list.
13230 	 */
13231 	if (idl->idl_conn == NULL) {
13232 		idl->idl_conn = connp;
13233 		connp->conn_drain_next = connp;
13234 		connp->conn_drain_prev = connp;
13235 	} else {
13236 		conn_t *head = idl->idl_conn;
13237 
13238 		connp->conn_drain_next = head;
13239 		connp->conn_drain_prev = head->conn_drain_prev;
13240 		head->conn_drain_prev->conn_drain_next = connp;
13241 		head->conn_drain_prev = connp;
13242 	}
13243 	/*
13244 	 * For non streams based sockets assert flow control.
13245 	 */
13246 	conn_setqfull(connp, NULL);
13247 	mutex_exit(&idl->idl_lock);
13248 }
13249 
13250 static void
conn_drain_remove(conn_t * connp)13251 conn_drain_remove(conn_t *connp)
13252 {
13253 	idl_t *idl = connp->conn_idl;
13254 
13255 	if (idl != NULL) {
13256 		/*
13257 		 * Remove ourself from the drain list.
13258 		 */
13259 		if (connp->conn_drain_next == connp) {
13260 			/* Singleton in the list */
13261 			ASSERT(connp->conn_drain_prev == connp);
13262 			idl->idl_conn = NULL;
13263 		} else {
13264 			connp->conn_drain_prev->conn_drain_next =
13265 			    connp->conn_drain_next;
13266 			connp->conn_drain_next->conn_drain_prev =
13267 			    connp->conn_drain_prev;
13268 			if (idl->idl_conn == connp)
13269 				idl->idl_conn = connp->conn_drain_next;
13270 		}
13271 
13272 		/*
13273 		 * NOTE: because conn_idl is associated with a specific drain
13274 		 * list which in turn is tied to the index the TX ring
13275 		 * (txl_cookie) hashes to, and because the TX ring can change
13276 		 * over the lifetime of the conn_t, we must clear conn_idl so
13277 		 * a subsequent conn_drain_insert() will set conn_idl again
13278 		 * based on the latest txl_cookie.
13279 		 */
13280 		connp->conn_idl = NULL;
13281 	}
13282 	connp->conn_drain_next = NULL;
13283 	connp->conn_drain_prev = NULL;
13284 
13285 	conn_clrqfull(connp, NULL);
13286 	/*
13287 	 * For streams based sockets open up flow control.
13288 	 */
13289 	if (!IPCL_IS_NONSTR(connp))
13290 		enableok(connp->conn_wq);
13291 }
13292 
13293 /*
13294  * This conn is closing, and we are called from ip_close. OR
13295  * this conn is draining because flow-control on the ill has been relieved.
13296  *
13297  * We must also need to remove conn's on this idl from the list, and also
13298  * inform the sockfs upcalls about the change in flow-control.
13299  */
13300 static void
conn_drain(conn_t * connp,boolean_t closing)13301 conn_drain(conn_t *connp, boolean_t closing)
13302 {
13303 	idl_t *idl;
13304 	conn_t *next_connp;
13305 
13306 	/*
13307 	 * connp->conn_idl is stable at this point, and no lock is needed
13308 	 * to check it. If we are called from ip_close, close has already
13309 	 * set CONN_CLOSING, thus freezing the value of conn_idl, and
13310 	 * called us only because conn_idl is non-null. If we are called thru
13311 	 * service, conn_idl could be null, but it cannot change because
13312 	 * service is single-threaded per queue, and there cannot be another
13313 	 * instance of service trying to call conn_drain_insert on this conn
13314 	 * now.
13315 	 */
13316 	ASSERT(!closing || connp == NULL || connp->conn_idl != NULL);
13317 
13318 	/*
13319 	 * If the conn doesn't exist or is not on a drain list, bail.
13320 	 */
13321 	if (connp == NULL || connp->conn_idl == NULL ||
13322 	    connp->conn_drain_prev == NULL) {
13323 		return;
13324 	}
13325 
13326 	idl = connp->conn_idl;
13327 	ASSERT(MUTEX_HELD(&idl->idl_lock));
13328 
13329 	if (!closing) {
13330 		next_connp = connp->conn_drain_next;
13331 		while (next_connp != connp) {
13332 			conn_t *delconnp = next_connp;
13333 
13334 			next_connp = next_connp->conn_drain_next;
13335 			conn_drain_remove(delconnp);
13336 		}
13337 		ASSERT(connp->conn_drain_next == idl->idl_conn);
13338 	}
13339 	conn_drain_remove(connp);
13340 }
13341 
13342 /*
13343  * Write service routine. Shared perimeter entry point.
13344  * The device queue's messages has fallen below the low water mark and STREAMS
13345  * has backenabled the ill_wq. Send sockfs notification about flow-control on
13346  * each waiting conn.
13347  */
13348 void
ip_wsrv(queue_t * q)13349 ip_wsrv(queue_t *q)
13350 {
13351 	ill_t	*ill;
13352 
13353 	ill = (ill_t *)q->q_ptr;
13354 	if (ill->ill_state_flags == 0) {
13355 		ip_stack_t *ipst = ill->ill_ipst;
13356 
13357 		/*
13358 		 * The device flow control has opened up.
13359 		 * Walk through conn drain lists and qenable the
13360 		 * first conn in each list. This makes sense only
13361 		 * if the stream is fully plumbed and setup.
13362 		 * Hence the ill_state_flags check above.
13363 		 */
13364 		ip1dbg(("ip_wsrv: walking\n"));
13365 		conn_walk_drain(ipst, &ipst->ips_idl_tx_list[0]);
13366 		enableok(ill->ill_wq);
13367 	}
13368 }
13369 
13370 /*
13371  * Callback to disable flow control in IP.
13372  *
13373  * This is a mac client callback added when the DLD_CAPAB_DIRECT capability
13374  * is enabled.
13375  *
13376  * When MAC_TX() is not able to send any more packets, dld sets its queue
13377  * to QFULL and enable the STREAMS flow control. Later, when the underlying
13378  * driver is able to continue to send packets, it calls mac_tx_(ring_)update()
13379  * function and wakes up corresponding mac worker threads, which in turn
13380  * calls this callback function, and disables flow control.
13381  */
13382 void
ill_flow_enable(void * arg,ip_mac_tx_cookie_t cookie)13383 ill_flow_enable(void *arg, ip_mac_tx_cookie_t cookie)
13384 {
13385 	ill_t *ill = (ill_t *)arg;
13386 	ip_stack_t *ipst = ill->ill_ipst;
13387 	idl_tx_list_t *idl_txl;
13388 
13389 	idl_txl = &ipst->ips_idl_tx_list[IDLHASHINDEX(cookie)];
13390 	mutex_enter(&idl_txl->txl_lock);
13391 	/* add code to to set a flag to indicate idl_txl is enabled */
13392 	conn_walk_drain(ipst, idl_txl);
13393 	mutex_exit(&idl_txl->txl_lock);
13394 }
13395 
13396 /*
13397  * Flow control has been relieved and STREAMS has backenabled us; drain
13398  * all the conn lists on `tx_list'.
13399  */
13400 static void
conn_walk_drain(ip_stack_t * ipst,idl_tx_list_t * tx_list)13401 conn_walk_drain(ip_stack_t *ipst, idl_tx_list_t *tx_list)
13402 {
13403 	int i;
13404 	idl_t *idl;
13405 
13406 	IP_STAT(ipst, ip_conn_walk_drain);
13407 
13408 	for (i = 0; i < ipst->ips_conn_drain_list_cnt; i++) {
13409 		idl = &tx_list->txl_drain_list[i];
13410 		mutex_enter(&idl->idl_lock);
13411 		conn_drain(idl->idl_conn, B_FALSE);
13412 		mutex_exit(&idl->idl_lock);
13413 	}
13414 }
13415 
13416 /*
13417  * Determine if the ill and multicast aspects of that packets
13418  * "matches" the conn.
13419  */
13420 boolean_t
conn_wantpacket(conn_t * connp,ip_recv_attr_t * ira,ipha_t * ipha)13421 conn_wantpacket(conn_t *connp, ip_recv_attr_t *ira, ipha_t *ipha)
13422 {
13423 	ill_t		*ill = ira->ira_rill;
13424 	zoneid_t	zoneid = ira->ira_zoneid;
13425 	uint_t		in_ifindex;
13426 	ipaddr_t	dst, src;
13427 
13428 	dst = ipha->ipha_dst;
13429 	src = ipha->ipha_src;
13430 
13431 	/*
13432 	 * conn_incoming_ifindex is set by IP_BOUND_IF which limits
13433 	 * unicast, broadcast and multicast reception to
13434 	 * conn_incoming_ifindex.
13435 	 * conn_wantpacket is called for unicast, broadcast and
13436 	 * multicast packets.
13437 	 */
13438 	in_ifindex = connp->conn_incoming_ifindex;
13439 
13440 	/* mpathd can bind to the under IPMP interface, which we allow */
13441 	if (in_ifindex != 0 && in_ifindex != ill->ill_phyint->phyint_ifindex) {
13442 		if (!IS_UNDER_IPMP(ill))
13443 			return (B_FALSE);
13444 
13445 		if (in_ifindex != ipmp_ill_get_ipmp_ifindex(ill))
13446 			return (B_FALSE);
13447 	}
13448 
13449 	if (!IPCL_ZONE_MATCH(connp, zoneid))
13450 		return (B_FALSE);
13451 
13452 	if (!(ira->ira_flags & IRAF_MULTICAST))
13453 		return (B_TRUE);
13454 
13455 	if (connp->conn_multi_router) {
13456 		/* multicast packet and multicast router socket: send up */
13457 		return (B_TRUE);
13458 	}
13459 
13460 	if (ipha->ipha_protocol == IPPROTO_PIM ||
13461 	    ipha->ipha_protocol == IPPROTO_RSVP)
13462 		return (B_TRUE);
13463 
13464 	return (conn_hasmembers_ill_withsrc_v4(connp, dst, src, ira->ira_ill));
13465 }
13466 
13467 void
conn_setqfull(conn_t * connp,boolean_t * flow_stopped)13468 conn_setqfull(conn_t *connp, boolean_t *flow_stopped)
13469 {
13470 	if (IPCL_IS_NONSTR(connp)) {
13471 		(*connp->conn_upcalls->su_txq_full)
13472 		    (connp->conn_upper_handle, B_TRUE);
13473 		if (flow_stopped != NULL)
13474 			*flow_stopped = B_TRUE;
13475 	} else {
13476 		queue_t *q = connp->conn_wq;
13477 
13478 		ASSERT(q != NULL);
13479 		if (!(q->q_flag & QFULL)) {
13480 			mutex_enter(QLOCK(q));
13481 			if (!(q->q_flag & QFULL)) {
13482 				/* still need to set QFULL */
13483 				q->q_flag |= QFULL;
13484 				/* set flow_stopped to true under QLOCK */
13485 				if (flow_stopped != NULL)
13486 					*flow_stopped = B_TRUE;
13487 				mutex_exit(QLOCK(q));
13488 			} else {
13489 				/* flow_stopped is left unchanged */
13490 				mutex_exit(QLOCK(q));
13491 			}
13492 		}
13493 	}
13494 }
13495 
13496 void
conn_clrqfull(conn_t * connp,boolean_t * flow_stopped)13497 conn_clrqfull(conn_t *connp, boolean_t *flow_stopped)
13498 {
13499 	if (IPCL_IS_NONSTR(connp)) {
13500 		(*connp->conn_upcalls->su_txq_full)
13501 		    (connp->conn_upper_handle, B_FALSE);
13502 		if (flow_stopped != NULL)
13503 			*flow_stopped = B_FALSE;
13504 	} else {
13505 		queue_t *q = connp->conn_wq;
13506 
13507 		ASSERT(q != NULL);
13508 		if (q->q_flag & QFULL) {
13509 			mutex_enter(QLOCK(q));
13510 			if (q->q_flag & QFULL) {
13511 				q->q_flag &= ~QFULL;
13512 				/* set flow_stopped to false under QLOCK */
13513 				if (flow_stopped != NULL)
13514 					*flow_stopped = B_FALSE;
13515 				mutex_exit(QLOCK(q));
13516 				if (q->q_flag & QWANTW)
13517 					qbackenable(q, 0);
13518 			} else {
13519 				/* flow_stopped is left unchanged */
13520 				mutex_exit(QLOCK(q));
13521 			}
13522 		}
13523 	}
13524 
13525 	mutex_enter(&connp->conn_lock);
13526 	connp->conn_blocked = B_FALSE;
13527 	mutex_exit(&connp->conn_lock);
13528 }
13529 
13530 /*
13531  * Return the length in bytes of the IPv4 headers (base header, label, and
13532  * other IP options) that will be needed based on the
13533  * ip_pkt_t structure passed by the caller.
13534  *
13535  * The returned length does not include the length of the upper level
13536  * protocol (ULP) header.
13537  * The caller needs to check that the length doesn't exceed the max for IPv4.
13538  */
13539 int
ip_total_hdrs_len_v4(const ip_pkt_t * ipp)13540 ip_total_hdrs_len_v4(const ip_pkt_t *ipp)
13541 {
13542 	int len;
13543 
13544 	len = IP_SIMPLE_HDR_LENGTH;
13545 	if (ipp->ipp_fields & IPPF_LABEL_V4) {
13546 		ASSERT(ipp->ipp_label_len_v4 != 0);
13547 		/* We need to round up here */
13548 		len += (ipp->ipp_label_len_v4 + 3) & ~3;
13549 	}
13550 
13551 	if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13552 		ASSERT(ipp->ipp_ipv4_options_len != 0);
13553 		ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13554 		len += ipp->ipp_ipv4_options_len;
13555 	}
13556 	return (len);
13557 }
13558 
13559 /*
13560  * All-purpose routine to build an IPv4 header with options based
13561  * on the abstract ip_pkt_t.
13562  *
13563  * The caller has to set the source and destination address as well as
13564  * ipha_length. The caller has to massage any source route and compensate
13565  * for the ULP pseudo-header checksum due to the source route.
13566  */
13567 void
ip_build_hdrs_v4(uchar_t * buf,uint_t buf_len,const ip_pkt_t * ipp,uint8_t protocol)13568 ip_build_hdrs_v4(uchar_t *buf, uint_t buf_len, const ip_pkt_t *ipp,
13569     uint8_t protocol)
13570 {
13571 	ipha_t	*ipha = (ipha_t *)buf;
13572 	uint8_t *cp;
13573 
13574 	/* Initialize IPv4 header */
13575 	ipha->ipha_type_of_service = ipp->ipp_type_of_service;
13576 	ipha->ipha_length = 0;	/* Caller will set later */
13577 	ipha->ipha_ident = 0;
13578 	ipha->ipha_fragment_offset_and_flags = 0;
13579 	ipha->ipha_ttl = ipp->ipp_unicast_hops;
13580 	ipha->ipha_protocol = protocol;
13581 	ipha->ipha_hdr_checksum = 0;
13582 
13583 	if ((ipp->ipp_fields & IPPF_ADDR) &&
13584 	    IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
13585 		ipha->ipha_src = ipp->ipp_addr_v4;
13586 
13587 	cp = (uint8_t *)&ipha[1];
13588 	if (ipp->ipp_fields & IPPF_LABEL_V4) {
13589 		ASSERT(ipp->ipp_label_len_v4 != 0);
13590 		bcopy(ipp->ipp_label_v4, cp, ipp->ipp_label_len_v4);
13591 		cp += ipp->ipp_label_len_v4;
13592 		/* We need to round up here */
13593 		while ((uintptr_t)cp & 0x3) {
13594 			*cp++ = IPOPT_NOP;
13595 		}
13596 	}
13597 
13598 	if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13599 		ASSERT(ipp->ipp_ipv4_options_len != 0);
13600 		ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13601 		bcopy(ipp->ipp_ipv4_options, cp, ipp->ipp_ipv4_options_len);
13602 		cp += ipp->ipp_ipv4_options_len;
13603 	}
13604 	ipha->ipha_version_and_hdr_length =
13605 	    (uint8_t)((IP_VERSION << 4) + buf_len / 4);
13606 
13607 	ASSERT((int)(cp - buf) == buf_len);
13608 }
13609 
13610 /* Allocate the private structure */
13611 static int
ip_priv_alloc(void ** bufp)13612 ip_priv_alloc(void **bufp)
13613 {
13614 	void	*buf;
13615 
13616 	if ((buf = kmem_alloc(sizeof (ip_priv_t), KM_NOSLEEP)) == NULL)
13617 		return (ENOMEM);
13618 
13619 	*bufp = buf;
13620 	return (0);
13621 }
13622 
13623 /* Function to delete the private structure */
13624 void
ip_priv_free(void * buf)13625 ip_priv_free(void *buf)
13626 {
13627 	ASSERT(buf != NULL);
13628 	kmem_free(buf, sizeof (ip_priv_t));
13629 }
13630 
13631 /*
13632  * The entry point for IPPF processing.
13633  * If the classifier (IPGPC_CLASSIFY) is not loaded and configured, the
13634  * routine just returns.
13635  *
13636  * When called, ip_process generates an ipp_packet_t structure
13637  * which holds the state information for this packet and invokes the
13638  * the classifier (via ipp_packet_process). The classification, depending on
13639  * configured filters, results in a list of actions for this packet. Invoking
13640  * an action may cause the packet to be dropped, in which case we return NULL.
13641  * proc indicates the callout position for
13642  * this packet and ill is the interface this packet arrived on or will leave
13643  * on (inbound and outbound resp.).
13644  *
13645  * We do the processing on the rill (mapped to the upper if ipmp), but MIB
13646  * on the ill corrsponding to the destination IP address.
13647  */
13648 mblk_t *
ip_process(ip_proc_t proc,mblk_t * mp,ill_t * rill,ill_t * ill)13649 ip_process(ip_proc_t proc, mblk_t *mp, ill_t *rill, ill_t *ill)
13650 {
13651 	ip_priv_t	*priv;
13652 	ipp_action_id_t	aid;
13653 	int		rc = 0;
13654 	ipp_packet_t	*pp;
13655 
13656 	/* If the classifier is not loaded, return  */
13657 	if ((aid = ipp_action_lookup(IPGPC_CLASSIFY)) == IPP_ACTION_INVAL) {
13658 		return (mp);
13659 	}
13660 
13661 	ASSERT(mp != NULL);
13662 
13663 	/* Allocate the packet structure */
13664 	rc = ipp_packet_alloc(&pp, "ip", aid);
13665 	if (rc != 0)
13666 		goto drop;
13667 
13668 	/* Allocate the private structure */
13669 	rc = ip_priv_alloc((void **)&priv);
13670 	if (rc != 0) {
13671 		ipp_packet_free(pp);
13672 		goto drop;
13673 	}
13674 	priv->proc = proc;
13675 	priv->ill_index = ill_get_upper_ifindex(rill);
13676 
13677 	ipp_packet_set_private(pp, priv, ip_priv_free);
13678 	ipp_packet_set_data(pp, mp);
13679 
13680 	/* Invoke the classifier */
13681 	rc = ipp_packet_process(&pp);
13682 	if (pp != NULL) {
13683 		mp = ipp_packet_get_data(pp);
13684 		ipp_packet_free(pp);
13685 		if (rc != 0)
13686 			goto drop;
13687 		return (mp);
13688 	} else {
13689 		/* No mp to trace in ip_drop_input/ip_drop_output  */
13690 		mp = NULL;
13691 	}
13692 drop:
13693 	if (proc == IPP_LOCAL_IN || proc == IPP_FWD_IN) {
13694 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
13695 		ip_drop_input("ip_process", mp, ill);
13696 	} else {
13697 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
13698 		ip_drop_output("ip_process", mp, ill);
13699 	}
13700 	freemsg(mp);
13701 	return (NULL);
13702 }
13703 
13704 /*
13705  * Propagate a multicast group membership operation (add/drop) on
13706  * all the interfaces crossed by the related multirt routes.
13707  * The call is considered successful if the operation succeeds
13708  * on at least one interface.
13709  *
13710  * This assumes that a set of IRE_HOST/RTF_MULTIRT has been created for the
13711  * multicast addresses with the ire argument being the first one.
13712  * We walk the bucket to find all the of those.
13713  *
13714  * Common to IPv4 and IPv6.
13715  */
13716 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)13717 ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
13718     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
13719     ire_t *ire, conn_t *connp, boolean_t checkonly, const in6_addr_t *v6group,
13720     mcast_record_t fmode, const in6_addr_t *v6src)
13721 {
13722 	ire_t		*ire_gw;
13723 	irb_t		*irb;
13724 	int		ifindex;
13725 	int		error = 0;
13726 	int		result;
13727 	ip_stack_t	*ipst = ire->ire_ipst;
13728 	ipaddr_t	group;
13729 	boolean_t	isv6;
13730 	int		match_flags;
13731 
13732 	if (IN6_IS_ADDR_V4MAPPED(v6group)) {
13733 		IN6_V4MAPPED_TO_IPADDR(v6group, group);
13734 		isv6 = B_FALSE;
13735 	} else {
13736 		isv6 = B_TRUE;
13737 	}
13738 
13739 	irb = ire->ire_bucket;
13740 	ASSERT(irb != NULL);
13741 
13742 	result = 0;
13743 	irb_refhold(irb);
13744 	for (; ire != NULL; ire = ire->ire_next) {
13745 		if ((ire->ire_flags & RTF_MULTIRT) == 0)
13746 			continue;
13747 
13748 		/* We handle -ifp routes by matching on the ill if set */
13749 		match_flags = MATCH_IRE_TYPE;
13750 		if (ire->ire_ill != NULL)
13751 			match_flags |= MATCH_IRE_ILL;
13752 
13753 		if (isv6) {
13754 			if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6group))
13755 				continue;
13756 
13757 			ire_gw = ire_ftable_lookup_v6(&ire->ire_gateway_addr_v6,
13758 			    0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13759 			    match_flags, 0, ipst, NULL);
13760 		} else {
13761 			if (ire->ire_addr != group)
13762 				continue;
13763 
13764 			ire_gw = ire_ftable_lookup_v4(ire->ire_gateway_addr,
13765 			    0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13766 			    match_flags, 0, ipst, NULL);
13767 		}
13768 		/* No interface route exists for the gateway; skip this ire. */
13769 		if (ire_gw == NULL)
13770 			continue;
13771 		if (ire_gw->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
13772 			ire_refrele(ire_gw);
13773 			continue;
13774 		}
13775 		ASSERT(ire_gw->ire_ill != NULL);	/* IRE_INTERFACE */
13776 		ifindex = ire_gw->ire_ill->ill_phyint->phyint_ifindex;
13777 
13778 		/*
13779 		 * The operation is considered a success if
13780 		 * it succeeds at least once on any one interface.
13781 		 */
13782 		error = fn(connp, checkonly, v6group, INADDR_ANY, ifindex,
13783 		    fmode, v6src);
13784 		if (error == 0)
13785 			result = CGTP_MCAST_SUCCESS;
13786 
13787 		ire_refrele(ire_gw);
13788 	}
13789 	irb_refrele(irb);
13790 	/*
13791 	 * Consider the call as successful if we succeeded on at least
13792 	 * one interface. Otherwise, return the last encountered error.
13793 	 */
13794 	return (result == CGTP_MCAST_SUCCESS ? 0 : error);
13795 }
13796 
13797 /*
13798  * Return the expected CGTP hooks version number.
13799  */
13800 int
ip_cgtp_filter_supported(void)13801 ip_cgtp_filter_supported(void)
13802 {
13803 	return (ip_cgtp_filter_rev);
13804 }
13805 
13806 /*
13807  * CGTP hooks can be registered by invoking this function.
13808  * Checks that the version number matches.
13809  */
13810 int
ip_cgtp_filter_register(netstackid_t stackid,cgtp_filter_ops_t * ops)13811 ip_cgtp_filter_register(netstackid_t stackid, cgtp_filter_ops_t *ops)
13812 {
13813 	netstack_t *ns;
13814 	ip_stack_t *ipst;
13815 
13816 	if (ops->cfo_filter_rev != CGTP_FILTER_REV)
13817 		return (ENOTSUP);
13818 
13819 	ns = netstack_find_by_stackid(stackid);
13820 	if (ns == NULL)
13821 		return (EINVAL);
13822 	ipst = ns->netstack_ip;
13823 	ASSERT(ipst != NULL);
13824 
13825 	if (ipst->ips_ip_cgtp_filter_ops != NULL) {
13826 		netstack_rele(ns);
13827 		return (EALREADY);
13828 	}
13829 
13830 	ipst->ips_ip_cgtp_filter_ops = ops;
13831 
13832 	ill_set_inputfn_all(ipst);
13833 
13834 	netstack_rele(ns);
13835 	return (0);
13836 }
13837 
13838 /*
13839  * CGTP hooks can be unregistered by invoking this function.
13840  * Returns ENXIO if there was no registration.
13841  * Returns EBUSY if the ndd variable has not been turned off.
13842  */
13843 int
ip_cgtp_filter_unregister(netstackid_t stackid)13844 ip_cgtp_filter_unregister(netstackid_t stackid)
13845 {
13846 	netstack_t *ns;
13847 	ip_stack_t *ipst;
13848 
13849 	ns = netstack_find_by_stackid(stackid);
13850 	if (ns == NULL)
13851 		return (EINVAL);
13852 	ipst = ns->netstack_ip;
13853 	ASSERT(ipst != NULL);
13854 
13855 	if (ipst->ips_ip_cgtp_filter) {
13856 		netstack_rele(ns);
13857 		return (EBUSY);
13858 	}
13859 
13860 	if (ipst->ips_ip_cgtp_filter_ops == NULL) {
13861 		netstack_rele(ns);
13862 		return (ENXIO);
13863 	}
13864 	ipst->ips_ip_cgtp_filter_ops = NULL;
13865 
13866 	ill_set_inputfn_all(ipst);
13867 
13868 	netstack_rele(ns);
13869 	return (0);
13870 }
13871 
13872 /*
13873  * Check whether there is a CGTP filter registration.
13874  * Returns non-zero if there is a registration, otherwise returns zero.
13875  * Note: returns zero if bad stackid.
13876  */
13877 int
ip_cgtp_filter_is_registered(netstackid_t stackid)13878 ip_cgtp_filter_is_registered(netstackid_t stackid)
13879 {
13880 	netstack_t *ns;
13881 	ip_stack_t *ipst;
13882 	int ret;
13883 
13884 	ns = netstack_find_by_stackid(stackid);
13885 	if (ns == NULL)
13886 		return (0);
13887 	ipst = ns->netstack_ip;
13888 	ASSERT(ipst != NULL);
13889 
13890 	if (ipst->ips_ip_cgtp_filter_ops != NULL)
13891 		ret = 1;
13892 	else
13893 		ret = 0;
13894 
13895 	netstack_rele(ns);
13896 	return (ret);
13897 }
13898 
13899 static int
ip_squeue_switch(int val)13900 ip_squeue_switch(int val)
13901 {
13902 	int rval;
13903 
13904 	switch (val) {
13905 	case IP_SQUEUE_ENTER_NODRAIN:
13906 		rval = SQ_NODRAIN;
13907 		break;
13908 	case IP_SQUEUE_ENTER:
13909 		rval = SQ_PROCESS;
13910 		break;
13911 	case IP_SQUEUE_FILL:
13912 	default:
13913 		rval = SQ_FILL;
13914 		break;
13915 	}
13916 	return (rval);
13917 }
13918 
13919 static void *
ip_kstat2_init(netstackid_t stackid,ip_stat_t * ip_statisticsp)13920 ip_kstat2_init(netstackid_t stackid, ip_stat_t *ip_statisticsp)
13921 {
13922 	kstat_t *ksp;
13923 
13924 	ip_stat_t template = {
13925 		{ "ip_udp_fannorm", 		KSTAT_DATA_UINT64 },
13926 		{ "ip_udp_fanmb", 		KSTAT_DATA_UINT64 },
13927 		{ "ip_recv_pullup", 		KSTAT_DATA_UINT64 },
13928 		{ "ip_db_ref",			KSTAT_DATA_UINT64 },
13929 		{ "ip_notaligned",		KSTAT_DATA_UINT64 },
13930 		{ "ip_multimblk",		KSTAT_DATA_UINT64 },
13931 		{ "ip_opt",			KSTAT_DATA_UINT64 },
13932 		{ "ipsec_proto_ahesp",		KSTAT_DATA_UINT64 },
13933 		{ "ip_conn_flputbq",		KSTAT_DATA_UINT64 },
13934 		{ "ip_conn_walk_drain",		KSTAT_DATA_UINT64 },
13935 		{ "ip_out_sw_cksum",		KSTAT_DATA_UINT64 },
13936 		{ "ip_out_sw_cksum_bytes",	KSTAT_DATA_UINT64 },
13937 		{ "ip_in_sw_cksum",		KSTAT_DATA_UINT64 },
13938 		{ "ip_ire_reclaim_calls",	KSTAT_DATA_UINT64 },
13939 		{ "ip_ire_reclaim_deleted",	KSTAT_DATA_UINT64 },
13940 		{ "ip_nce_reclaim_calls",	KSTAT_DATA_UINT64 },
13941 		{ "ip_nce_reclaim_deleted",	KSTAT_DATA_UINT64 },
13942 		{ "ip_dce_reclaim_calls",	KSTAT_DATA_UINT64 },
13943 		{ "ip_dce_reclaim_deleted",	KSTAT_DATA_UINT64 },
13944 		{ "ip_tcp_in_full_hw_cksum_err",	KSTAT_DATA_UINT64 },
13945 		{ "ip_tcp_in_part_hw_cksum_err",	KSTAT_DATA_UINT64 },
13946 		{ "ip_tcp_in_sw_cksum_err",		KSTAT_DATA_UINT64 },
13947 		{ "ip_udp_in_full_hw_cksum_err",	KSTAT_DATA_UINT64 },
13948 		{ "ip_udp_in_part_hw_cksum_err",	KSTAT_DATA_UINT64 },
13949 		{ "ip_udp_in_sw_cksum_err",	KSTAT_DATA_UINT64 },
13950 		{ "conn_in_recvdstaddr",	KSTAT_DATA_UINT64 },
13951 		{ "conn_in_recvopts",		KSTAT_DATA_UINT64 },
13952 		{ "conn_in_recvif",		KSTAT_DATA_UINT64 },
13953 		{ "conn_in_recvslla",		KSTAT_DATA_UINT64 },
13954 		{ "conn_in_recvucred",		KSTAT_DATA_UINT64 },
13955 		{ "conn_in_recvttl",		KSTAT_DATA_UINT64 },
13956 		{ "conn_in_recvhopopts",	KSTAT_DATA_UINT64 },
13957 		{ "conn_in_recvhoplimit",	KSTAT_DATA_UINT64 },
13958 		{ "conn_in_recvdstopts",	KSTAT_DATA_UINT64 },
13959 		{ "conn_in_recvrthdrdstopts",	KSTAT_DATA_UINT64 },
13960 		{ "conn_in_recvrthdr",		KSTAT_DATA_UINT64 },
13961 		{ "conn_in_recvpktinfo",	KSTAT_DATA_UINT64 },
13962 		{ "conn_in_recvtclass",		KSTAT_DATA_UINT64 },
13963 		{ "conn_in_timestamp",		KSTAT_DATA_UINT64 },
13964 	};
13965 
13966 	ksp = kstat_create_netstack("ip", 0, "ipstat", "net",
13967 	    KSTAT_TYPE_NAMED, sizeof (template) / sizeof (kstat_named_t),
13968 	    KSTAT_FLAG_VIRTUAL, stackid);
13969 
13970 	if (ksp == NULL)
13971 		return (NULL);
13972 
13973 	bcopy(&template, ip_statisticsp, sizeof (template));
13974 	ksp->ks_data = (void *)ip_statisticsp;
13975 	ksp->ks_private = (void *)(uintptr_t)stackid;
13976 
13977 	kstat_install(ksp);
13978 	return (ksp);
13979 }
13980 
13981 static void
ip_kstat2_fini(netstackid_t stackid,kstat_t * ksp)13982 ip_kstat2_fini(netstackid_t stackid, kstat_t *ksp)
13983 {
13984 	if (ksp != NULL) {
13985 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
13986 		kstat_delete_netstack(ksp, stackid);
13987 	}
13988 }
13989 
13990 static void *
ip_kstat_init(netstackid_t stackid,ip_stack_t * ipst)13991 ip_kstat_init(netstackid_t stackid, ip_stack_t *ipst)
13992 {
13993 	kstat_t	*ksp;
13994 
13995 	ip_named_kstat_t template = {
13996 		{ "forwarding",		KSTAT_DATA_UINT32, 0 },
13997 		{ "defaultTTL",		KSTAT_DATA_UINT32, 0 },
13998 		{ "inReceives",		KSTAT_DATA_UINT64, 0 },
13999 		{ "inHdrErrors",	KSTAT_DATA_UINT32, 0 },
14000 		{ "inAddrErrors",	KSTAT_DATA_UINT32, 0 },
14001 		{ "forwDatagrams",	KSTAT_DATA_UINT64, 0 },
14002 		{ "inUnknownProtos",	KSTAT_DATA_UINT32, 0 },
14003 		{ "inDiscards",		KSTAT_DATA_UINT32, 0 },
14004 		{ "inDelivers",		KSTAT_DATA_UINT64, 0 },
14005 		{ "outRequests",	KSTAT_DATA_UINT64, 0 },
14006 		{ "outDiscards",	KSTAT_DATA_UINT32, 0 },
14007 		{ "outNoRoutes",	KSTAT_DATA_UINT32, 0 },
14008 		{ "reasmTimeout",	KSTAT_DATA_UINT32, 0 },
14009 		{ "reasmReqds",		KSTAT_DATA_UINT32, 0 },
14010 		{ "reasmOKs",		KSTAT_DATA_UINT32, 0 },
14011 		{ "reasmFails",		KSTAT_DATA_UINT32, 0 },
14012 		{ "fragOKs",		KSTAT_DATA_UINT32, 0 },
14013 		{ "fragFails",		KSTAT_DATA_UINT32, 0 },
14014 		{ "fragCreates",	KSTAT_DATA_UINT32, 0 },
14015 		{ "addrEntrySize",	KSTAT_DATA_INT32, 0 },
14016 		{ "routeEntrySize",	KSTAT_DATA_INT32, 0 },
14017 		{ "netToMediaEntrySize",	KSTAT_DATA_INT32, 0 },
14018 		{ "routingDiscards",	KSTAT_DATA_UINT32, 0 },
14019 		{ "inErrs",		KSTAT_DATA_UINT32, 0 },
14020 		{ "noPorts",		KSTAT_DATA_UINT32, 0 },
14021 		{ "inCksumErrs",	KSTAT_DATA_UINT32, 0 },
14022 		{ "reasmDuplicates",	KSTAT_DATA_UINT32, 0 },
14023 		{ "reasmPartDups",	KSTAT_DATA_UINT32, 0 },
14024 		{ "forwProhibits",	KSTAT_DATA_UINT32, 0 },
14025 		{ "udpInCksumErrs",	KSTAT_DATA_UINT32, 0 },
14026 		{ "udpInOverflows",	KSTAT_DATA_UINT32, 0 },
14027 		{ "rawipInOverflows",	KSTAT_DATA_UINT32, 0 },
14028 		{ "ipsecInSucceeded",	KSTAT_DATA_UINT32, 0 },
14029 		{ "ipsecInFailed",	KSTAT_DATA_INT32, 0 },
14030 		{ "memberEntrySize",	KSTAT_DATA_INT32, 0 },
14031 		{ "inIPv6",		KSTAT_DATA_UINT32, 0 },
14032 		{ "outIPv6",		KSTAT_DATA_UINT32, 0 },
14033 		{ "outSwitchIPv6",	KSTAT_DATA_UINT32, 0 },
14034 	};
14035 
14036 	ksp = kstat_create_netstack("ip", 0, "ip", "mib2", KSTAT_TYPE_NAMED,
14037 	    NUM_OF_FIELDS(ip_named_kstat_t), 0, stackid);
14038 	if (ksp == NULL || ksp->ks_data == NULL)
14039 		return (NULL);
14040 
14041 	template.forwarding.value.ui32 = WE_ARE_FORWARDING(ipst) ? 1:2;
14042 	template.defaultTTL.value.ui32 = (uint32_t)ipst->ips_ip_def_ttl;
14043 	template.reasmTimeout.value.ui32 = ipst->ips_ip_reassembly_timeout;
14044 	template.addrEntrySize.value.i32 = sizeof (mib2_ipAddrEntry_t);
14045 	template.routeEntrySize.value.i32 = sizeof (mib2_ipRouteEntry_t);
14046 
14047 	template.netToMediaEntrySize.value.i32 =
14048 	    sizeof (mib2_ipNetToMediaEntry_t);
14049 
14050 	template.memberEntrySize.value.i32 = sizeof (ipv6_member_t);
14051 
14052 	bcopy(&template, ksp->ks_data, sizeof (template));
14053 	ksp->ks_update = ip_kstat_update;
14054 	ksp->ks_private = (void *)(uintptr_t)stackid;
14055 
14056 	kstat_install(ksp);
14057 	return (ksp);
14058 }
14059 
14060 static void
ip_kstat_fini(netstackid_t stackid,kstat_t * ksp)14061 ip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14062 {
14063 	if (ksp != NULL) {
14064 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14065 		kstat_delete_netstack(ksp, stackid);
14066 	}
14067 }
14068 
14069 static int
ip_kstat_update(kstat_t * kp,int rw)14070 ip_kstat_update(kstat_t *kp, int rw)
14071 {
14072 	ip_named_kstat_t *ipkp;
14073 	mib2_ipIfStatsEntry_t ipmib;
14074 	ill_walk_context_t ctx;
14075 	ill_t *ill;
14076 	netstackid_t	stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14077 	netstack_t	*ns;
14078 	ip_stack_t	*ipst;
14079 
14080 	if (kp == NULL || kp->ks_data == NULL)
14081 		return (EIO);
14082 
14083 	if (rw == KSTAT_WRITE)
14084 		return (EACCES);
14085 
14086 	ns = netstack_find_by_stackid(stackid);
14087 	if (ns == NULL)
14088 		return (-1);
14089 	ipst = ns->netstack_ip;
14090 	if (ipst == NULL) {
14091 		netstack_rele(ns);
14092 		return (-1);
14093 	}
14094 	ipkp = (ip_named_kstat_t *)kp->ks_data;
14095 
14096 	bcopy(&ipst->ips_ip_mib, &ipmib, sizeof (ipmib));
14097 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
14098 	ill = ILL_START_WALK_V4(&ctx, ipst);
14099 	for (; ill != NULL; ill = ill_next(&ctx, ill))
14100 		ip_mib2_add_ip_stats(&ipmib, ill->ill_ip_mib);
14101 	rw_exit(&ipst->ips_ill_g_lock);
14102 
14103 	ipkp->forwarding.value.ui32 =		ipmib.ipIfStatsForwarding;
14104 	ipkp->defaultTTL.value.ui32 =		ipmib.ipIfStatsDefaultTTL;
14105 	ipkp->inReceives.value.ui64 =		ipmib.ipIfStatsHCInReceives;
14106 	ipkp->inHdrErrors.value.ui32 =		ipmib.ipIfStatsInHdrErrors;
14107 	ipkp->inAddrErrors.value.ui32 =		ipmib.ipIfStatsInAddrErrors;
14108 	ipkp->forwDatagrams.value.ui64 = ipmib.ipIfStatsHCOutForwDatagrams;
14109 	ipkp->inUnknownProtos.value.ui32 =	ipmib.ipIfStatsInUnknownProtos;
14110 	ipkp->inDiscards.value.ui32 =		ipmib.ipIfStatsInDiscards;
14111 	ipkp->inDelivers.value.ui64 =		ipmib.ipIfStatsHCInDelivers;
14112 	ipkp->outRequests.value.ui64 =		ipmib.ipIfStatsHCOutRequests;
14113 	ipkp->outDiscards.value.ui32 =		ipmib.ipIfStatsOutDiscards;
14114 	ipkp->outNoRoutes.value.ui32 =		ipmib.ipIfStatsOutNoRoutes;
14115 	ipkp->reasmTimeout.value.ui32 =		ipst->ips_ip_reassembly_timeout;
14116 	ipkp->reasmReqds.value.ui32 =		ipmib.ipIfStatsReasmReqds;
14117 	ipkp->reasmOKs.value.ui32 =		ipmib.ipIfStatsReasmOKs;
14118 	ipkp->reasmFails.value.ui32 =		ipmib.ipIfStatsReasmFails;
14119 	ipkp->fragOKs.value.ui32 =		ipmib.ipIfStatsOutFragOKs;
14120 	ipkp->fragFails.value.ui32 =		ipmib.ipIfStatsOutFragFails;
14121 	ipkp->fragCreates.value.ui32 =		ipmib.ipIfStatsOutFragCreates;
14122 
14123 	ipkp->routingDiscards.value.ui32 =	0;
14124 	ipkp->inErrs.value.ui32 =		ipmib.tcpIfStatsInErrs;
14125 	ipkp->noPorts.value.ui32 =		ipmib.udpIfStatsNoPorts;
14126 	ipkp->inCksumErrs.value.ui32 =		ipmib.ipIfStatsInCksumErrs;
14127 	ipkp->reasmDuplicates.value.ui32 =	ipmib.ipIfStatsReasmDuplicates;
14128 	ipkp->reasmPartDups.value.ui32 =	ipmib.ipIfStatsReasmPartDups;
14129 	ipkp->forwProhibits.value.ui32 =	ipmib.ipIfStatsForwProhibits;
14130 	ipkp->udpInCksumErrs.value.ui32 =	ipmib.udpIfStatsInCksumErrs;
14131 	ipkp->udpInOverflows.value.ui32 =	ipmib.udpIfStatsInOverflows;
14132 	ipkp->rawipInOverflows.value.ui32 =	ipmib.rawipIfStatsInOverflows;
14133 	ipkp->ipsecInSucceeded.value.ui32 =	ipmib.ipsecIfStatsInSucceeded;
14134 	ipkp->ipsecInFailed.value.i32 =		ipmib.ipsecIfStatsInFailed;
14135 
14136 	ipkp->inIPv6.value.ui32 =	ipmib.ipIfStatsInWrongIPVersion;
14137 	ipkp->outIPv6.value.ui32 =	ipmib.ipIfStatsOutWrongIPVersion;
14138 	ipkp->outSwitchIPv6.value.ui32 = ipmib.ipIfStatsOutSwitchIPVersion;
14139 
14140 	netstack_rele(ns);
14141 
14142 	return (0);
14143 }
14144 
14145 static void *
icmp_kstat_init(netstackid_t stackid)14146 icmp_kstat_init(netstackid_t stackid)
14147 {
14148 	kstat_t	*ksp;
14149 
14150 	icmp_named_kstat_t template = {
14151 		{ "inMsgs",		KSTAT_DATA_UINT32 },
14152 		{ "inErrors",		KSTAT_DATA_UINT32 },
14153 		{ "inDestUnreachs",	KSTAT_DATA_UINT32 },
14154 		{ "inTimeExcds",	KSTAT_DATA_UINT32 },
14155 		{ "inParmProbs",	KSTAT_DATA_UINT32 },
14156 		{ "inSrcQuenchs",	KSTAT_DATA_UINT32 },
14157 		{ "inRedirects",	KSTAT_DATA_UINT32 },
14158 		{ "inEchos",		KSTAT_DATA_UINT32 },
14159 		{ "inEchoReps",		KSTAT_DATA_UINT32 },
14160 		{ "inTimestamps",	KSTAT_DATA_UINT32 },
14161 		{ "inTimestampReps",	KSTAT_DATA_UINT32 },
14162 		{ "inAddrMasks",	KSTAT_DATA_UINT32 },
14163 		{ "inAddrMaskReps",	KSTAT_DATA_UINT32 },
14164 		{ "outMsgs",		KSTAT_DATA_UINT32 },
14165 		{ "outErrors",		KSTAT_DATA_UINT32 },
14166 		{ "outDestUnreachs",	KSTAT_DATA_UINT32 },
14167 		{ "outTimeExcds",	KSTAT_DATA_UINT32 },
14168 		{ "outParmProbs",	KSTAT_DATA_UINT32 },
14169 		{ "outSrcQuenchs",	KSTAT_DATA_UINT32 },
14170 		{ "outRedirects",	KSTAT_DATA_UINT32 },
14171 		{ "outEchos",		KSTAT_DATA_UINT32 },
14172 		{ "outEchoReps",	KSTAT_DATA_UINT32 },
14173 		{ "outTimestamps",	KSTAT_DATA_UINT32 },
14174 		{ "outTimestampReps",	KSTAT_DATA_UINT32 },
14175 		{ "outAddrMasks",	KSTAT_DATA_UINT32 },
14176 		{ "outAddrMaskReps",	KSTAT_DATA_UINT32 },
14177 		{ "inChksumErrs",	KSTAT_DATA_UINT32 },
14178 		{ "inUnknowns",		KSTAT_DATA_UINT32 },
14179 		{ "inFragNeeded",	KSTAT_DATA_UINT32 },
14180 		{ "outFragNeeded",	KSTAT_DATA_UINT32 },
14181 		{ "outDrops",		KSTAT_DATA_UINT32 },
14182 		{ "inOverFlows",	KSTAT_DATA_UINT32 },
14183 		{ "inBadRedirects",	KSTAT_DATA_UINT32 },
14184 	};
14185 
14186 	ksp = kstat_create_netstack("ip", 0, "icmp", "mib2", KSTAT_TYPE_NAMED,
14187 	    NUM_OF_FIELDS(icmp_named_kstat_t), 0, stackid);
14188 	if (ksp == NULL || ksp->ks_data == NULL)
14189 		return (NULL);
14190 
14191 	bcopy(&template, ksp->ks_data, sizeof (template));
14192 
14193 	ksp->ks_update = icmp_kstat_update;
14194 	ksp->ks_private = (void *)(uintptr_t)stackid;
14195 
14196 	kstat_install(ksp);
14197 	return (ksp);
14198 }
14199 
14200 static void
icmp_kstat_fini(netstackid_t stackid,kstat_t * ksp)14201 icmp_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14202 {
14203 	if (ksp != NULL) {
14204 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14205 		kstat_delete_netstack(ksp, stackid);
14206 	}
14207 }
14208 
14209 static int
icmp_kstat_update(kstat_t * kp,int rw)14210 icmp_kstat_update(kstat_t *kp, int rw)
14211 {
14212 	icmp_named_kstat_t *icmpkp;
14213 	netstackid_t	stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14214 	netstack_t	*ns;
14215 	ip_stack_t	*ipst;
14216 
14217 	if ((kp == NULL) || (kp->ks_data == NULL))
14218 		return (EIO);
14219 
14220 	if (rw == KSTAT_WRITE)
14221 		return (EACCES);
14222 
14223 	ns = netstack_find_by_stackid(stackid);
14224 	if (ns == NULL)
14225 		return (-1);
14226 	ipst = ns->netstack_ip;
14227 	if (ipst == NULL) {
14228 		netstack_rele(ns);
14229 		return (-1);
14230 	}
14231 	icmpkp = (icmp_named_kstat_t *)kp->ks_data;
14232 
14233 	icmpkp->inMsgs.value.ui32 =	    ipst->ips_icmp_mib.icmpInMsgs;
14234 	icmpkp->inErrors.value.ui32 =	    ipst->ips_icmp_mib.icmpInErrors;
14235 	icmpkp->inDestUnreachs.value.ui32 =
14236 	    ipst->ips_icmp_mib.icmpInDestUnreachs;
14237 	icmpkp->inTimeExcds.value.ui32 =    ipst->ips_icmp_mib.icmpInTimeExcds;
14238 	icmpkp->inParmProbs.value.ui32 =    ipst->ips_icmp_mib.icmpInParmProbs;
14239 	icmpkp->inSrcQuenchs.value.ui32 =   ipst->ips_icmp_mib.icmpInSrcQuenchs;
14240 	icmpkp->inRedirects.value.ui32 =    ipst->ips_icmp_mib.icmpInRedirects;
14241 	icmpkp->inEchos.value.ui32 =	    ipst->ips_icmp_mib.icmpInEchos;
14242 	icmpkp->inEchoReps.value.ui32 =	    ipst->ips_icmp_mib.icmpInEchoReps;
14243 	icmpkp->inTimestamps.value.ui32 =   ipst->ips_icmp_mib.icmpInTimestamps;
14244 	icmpkp->inTimestampReps.value.ui32 =
14245 	    ipst->ips_icmp_mib.icmpInTimestampReps;
14246 	icmpkp->inAddrMasks.value.ui32 =    ipst->ips_icmp_mib.icmpInAddrMasks;
14247 	icmpkp->inAddrMaskReps.value.ui32 =
14248 	    ipst->ips_icmp_mib.icmpInAddrMaskReps;
14249 	icmpkp->outMsgs.value.ui32 =	    ipst->ips_icmp_mib.icmpOutMsgs;
14250 	icmpkp->outErrors.value.ui32 =	    ipst->ips_icmp_mib.icmpOutErrors;
14251 	icmpkp->outDestUnreachs.value.ui32 =
14252 	    ipst->ips_icmp_mib.icmpOutDestUnreachs;
14253 	icmpkp->outTimeExcds.value.ui32 =   ipst->ips_icmp_mib.icmpOutTimeExcds;
14254 	icmpkp->outParmProbs.value.ui32 =   ipst->ips_icmp_mib.icmpOutParmProbs;
14255 	icmpkp->outSrcQuenchs.value.ui32 =
14256 	    ipst->ips_icmp_mib.icmpOutSrcQuenchs;
14257 	icmpkp->outRedirects.value.ui32 =   ipst->ips_icmp_mib.icmpOutRedirects;
14258 	icmpkp->outEchos.value.ui32 =	    ipst->ips_icmp_mib.icmpOutEchos;
14259 	icmpkp->outEchoReps.value.ui32 =    ipst->ips_icmp_mib.icmpOutEchoReps;
14260 	icmpkp->outTimestamps.value.ui32 =
14261 	    ipst->ips_icmp_mib.icmpOutTimestamps;
14262 	icmpkp->outTimestampReps.value.ui32 =
14263 	    ipst->ips_icmp_mib.icmpOutTimestampReps;
14264 	icmpkp->outAddrMasks.value.ui32 =
14265 	    ipst->ips_icmp_mib.icmpOutAddrMasks;
14266 	icmpkp->outAddrMaskReps.value.ui32 =
14267 	    ipst->ips_icmp_mib.icmpOutAddrMaskReps;
14268 	icmpkp->inCksumErrs.value.ui32 =    ipst->ips_icmp_mib.icmpInCksumErrs;
14269 	icmpkp->inUnknowns.value.ui32 =	    ipst->ips_icmp_mib.icmpInUnknowns;
14270 	icmpkp->inFragNeeded.value.ui32 =   ipst->ips_icmp_mib.icmpInFragNeeded;
14271 	icmpkp->outFragNeeded.value.ui32 =
14272 	    ipst->ips_icmp_mib.icmpOutFragNeeded;
14273 	icmpkp->outDrops.value.ui32 =	    ipst->ips_icmp_mib.icmpOutDrops;
14274 	icmpkp->inOverflows.value.ui32 =    ipst->ips_icmp_mib.icmpInOverflows;
14275 	icmpkp->inBadRedirects.value.ui32 =
14276 	    ipst->ips_icmp_mib.icmpInBadRedirects;
14277 
14278 	netstack_rele(ns);
14279 	return (0);
14280 }
14281 
14282 /*
14283  * This is the fanout function for raw socket opened for SCTP.  Note
14284  * that it is called after SCTP checks that there is no socket which
14285  * wants a packet.  Then before SCTP handles this out of the blue packet,
14286  * this function is called to see if there is any raw socket for SCTP.
14287  * If there is and it is bound to the correct address, the packet will
14288  * be sent to that socket.  Note that only one raw socket can be bound to
14289  * a port.  This is assured in ipcl_sctp_hash_insert();
14290  */
14291 void
ip_fanout_sctp_raw(mblk_t * mp,ipha_t * ipha,ip6_t * ip6h,uint32_t ports,ip_recv_attr_t * ira)14292 ip_fanout_sctp_raw(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h, uint32_t ports,
14293     ip_recv_attr_t *ira)
14294 {
14295 	conn_t		*connp;
14296 	queue_t		*rq;
14297 	boolean_t	secure;
14298 	ill_t		*ill = ira->ira_ill;
14299 	ip_stack_t	*ipst = ill->ill_ipst;
14300 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
14301 	sctp_stack_t	*sctps = ipst->ips_netstack->netstack_sctp;
14302 	iaflags_t	iraflags = ira->ira_flags;
14303 	ill_t		*rill = ira->ira_rill;
14304 
14305 	secure = iraflags & IRAF_IPSEC_SECURE;
14306 
14307 	connp = ipcl_classify_raw(mp, IPPROTO_SCTP, ports, ipha, ip6h,
14308 	    ira, ipst);
14309 	if (connp == NULL) {
14310 		/*
14311 		 * Although raw sctp is not summed, OOB chunks must be.
14312 		 * Drop the packet here if the sctp checksum failed.
14313 		 */
14314 		if (iraflags & IRAF_SCTP_CSUM_ERR) {
14315 			SCTPS_BUMP_MIB(sctps, sctpChecksumError);
14316 			freemsg(mp);
14317 			return;
14318 		}
14319 		ira->ira_ill = ira->ira_rill = NULL;
14320 		sctp_ootb_input(mp, ira, ipst);
14321 		ira->ira_ill = ill;
14322 		ira->ira_rill = rill;
14323 		return;
14324 	}
14325 	rq = connp->conn_rq;
14326 	if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
14327 		CONN_DEC_REF(connp);
14328 		BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
14329 		freemsg(mp);
14330 		return;
14331 	}
14332 	if (((iraflags & IRAF_IS_IPV4) ?
14333 	    CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
14334 	    CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
14335 	    secure) {
14336 		mp = ipsec_check_inbound_policy(mp, connp, ipha,
14337 		    ip6h, ira);
14338 		if (mp == NULL) {
14339 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
14340 			/* Note that mp is NULL */
14341 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
14342 			CONN_DEC_REF(connp);
14343 			return;
14344 		}
14345 	}
14346 
14347 	if (iraflags & IRAF_ICMP_ERROR) {
14348 		(connp->conn_recvicmp)(connp, mp, NULL, ira);
14349 	} else {
14350 		ill_t *rill = ira->ira_rill;
14351 
14352 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
14353 		/* This is the SOCK_RAW, IPPROTO_SCTP case. */
14354 		ira->ira_ill = ira->ira_rill = NULL;
14355 		(connp->conn_recv)(connp, mp, NULL, ira);
14356 		ira->ira_ill = ill;
14357 		ira->ira_rill = rill;
14358 	}
14359 	CONN_DEC_REF(connp);
14360 }
14361 
14362 /*
14363  * Free a packet that has the link-layer dl_unitdata_req_t or fast-path
14364  * header before the ip payload.
14365  */
14366 static void
ip_xmit_flowctl_drop(ill_t * ill,mblk_t * mp,boolean_t is_fp_mp,int fp_mp_len)14367 ip_xmit_flowctl_drop(ill_t *ill, mblk_t *mp, boolean_t is_fp_mp, int fp_mp_len)
14368 {
14369 	int len = (mp->b_wptr - mp->b_rptr);
14370 	mblk_t *ip_mp;
14371 
14372 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14373 	if (is_fp_mp || len != fp_mp_len) {
14374 		if (len > fp_mp_len) {
14375 			/*
14376 			 * fastpath header and ip header in the first mblk
14377 			 */
14378 			mp->b_rptr += fp_mp_len;
14379 		} else {
14380 			/*
14381 			 * ip_xmit_attach_llhdr had to prepend an mblk to
14382 			 * attach the fastpath header before ip header.
14383 			 */
14384 			ip_mp = mp->b_cont;
14385 			freeb(mp);
14386 			mp = ip_mp;
14387 			mp->b_rptr += (fp_mp_len - len);
14388 		}
14389 	} else {
14390 		ip_mp = mp->b_cont;
14391 		freeb(mp);
14392 		mp = ip_mp;
14393 	}
14394 	ip_drop_output("ipIfStatsOutDiscards - flow ctl", mp, ill);
14395 	freemsg(mp);
14396 }
14397 
14398 /*
14399  * Normal post fragmentation function.
14400  *
14401  * Send a packet using the passed in nce. This handles both IPv4 and IPv6
14402  * using the same state machine.
14403  *
14404  * We return an error on failure. In particular we return EWOULDBLOCK
14405  * when the driver flow controls. In that case this ensures that ip_wsrv runs
14406  * (currently by canputnext failure resulting in backenabling from GLD.)
14407  * This allows the callers of conn_ip_output() to use EWOULDBLOCK as an
14408  * indication that they can flow control until ip_wsrv() tells then to restart.
14409  *
14410  * If the nce passed by caller is incomplete, this function
14411  * queues the packet and if necessary, sends ARP request and bails.
14412  * If the Neighbor Cache passed is fully resolved, we simply prepend
14413  * the link-layer header to the packet, do ipsec hw acceleration
14414  * work if necessary, and send the packet out on the wire.
14415  */
14416 /* ARGSUSED6 */
14417 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)14418 ip_xmit(mblk_t *mp, nce_t *nce, iaflags_t ixaflags, uint_t pkt_len,
14419     uint32_t xmit_hint, zoneid_t szone, zoneid_t nolzid, uintptr_t *ixacookie)
14420 {
14421 	queue_t		*wq;
14422 	ill_t		*ill = nce->nce_ill;
14423 	ip_stack_t	*ipst = ill->ill_ipst;
14424 	uint64_t	delta;
14425 	boolean_t	isv6 = ill->ill_isv6;
14426 	boolean_t	fp_mp;
14427 	ncec_t		*ncec = nce->nce_common;
14428 	int64_t		now = LBOLT_FASTPATH64;
14429 	boolean_t	is_probe;
14430 
14431 	DTRACE_PROBE1(ip__xmit, nce_t *, nce);
14432 
14433 	ASSERT(mp != NULL);
14434 	ASSERT(mp->b_datap->db_type == M_DATA);
14435 	ASSERT(pkt_len == msgdsize(mp));
14436 
14437 	/*
14438 	 * If we have already been here and are coming back after ARP/ND.
14439 	 * the IXAF_NO_TRACE flag is set. We skip FW_HOOKS, DTRACE and ipobs
14440 	 * in that case since they have seen the packet when it came here
14441 	 * the first time.
14442 	 */
14443 	if (ixaflags & IXAF_NO_TRACE)
14444 		goto sendit;
14445 
14446 	if (ixaflags & IXAF_IS_IPV4) {
14447 		ipha_t *ipha = (ipha_t *)mp->b_rptr;
14448 
14449 		ASSERT(!isv6);
14450 		ASSERT(pkt_len == ntohs(((ipha_t *)mp->b_rptr)->ipha_length));
14451 		if (HOOKS4_INTERESTED_PHYSICAL_OUT(ipst) &&
14452 		    !(ixaflags & IXAF_NO_PFHOOK)) {
14453 			int	error;
14454 
14455 			FW_HOOKS(ipst->ips_ip4_physical_out_event,
14456 			    ipst->ips_ipv4firewall_physical_out,
14457 			    NULL, ill, ipha, mp, mp, 0, ipst, error);
14458 			DTRACE_PROBE1(ip4__physical__out__end,
14459 			    mblk_t *, mp);
14460 			if (mp == NULL)
14461 				return (error);
14462 
14463 			/* The length could have changed */
14464 			pkt_len = msgdsize(mp);
14465 		}
14466 		if (ipst->ips_ip4_observe.he_interested) {
14467 			/*
14468 			 * Note that for TX the zoneid is the sending
14469 			 * zone, whether or not MLP is in play.
14470 			 * Since the szone argument is the IP zoneid (i.e.,
14471 			 * zero for exclusive-IP zones) and ipobs wants
14472 			 * the system zoneid, we map it here.
14473 			 */
14474 			szone = IP_REAL_ZONEID(szone, ipst);
14475 
14476 			/*
14477 			 * On the outbound path the destination zone will be
14478 			 * unknown as we're sending this packet out on the
14479 			 * wire.
14480 			 */
14481 			ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14482 			    ill, ipst);
14483 		}
14484 		DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14485 		    void_ip_t *, ipha,  __dtrace_ipsr_ill_t *, ill,
14486 		    ipha_t *, ipha, ip6_t *, NULL, int, 0);
14487 	} else {
14488 		ip6_t *ip6h = (ip6_t *)mp->b_rptr;
14489 
14490 		ASSERT(isv6);
14491 		ASSERT(pkt_len ==
14492 		    ntohs(((ip6_t *)mp->b_rptr)->ip6_plen) + IPV6_HDR_LEN);
14493 		if (HOOKS6_INTERESTED_PHYSICAL_OUT(ipst) &&
14494 		    !(ixaflags & IXAF_NO_PFHOOK)) {
14495 			int	error;
14496 
14497 			FW_HOOKS6(ipst->ips_ip6_physical_out_event,
14498 			    ipst->ips_ipv6firewall_physical_out,
14499 			    NULL, ill, ip6h, mp, mp, 0, ipst, error);
14500 			DTRACE_PROBE1(ip6__physical__out__end,
14501 			    mblk_t *, mp);
14502 			if (mp == NULL)
14503 				return (error);
14504 
14505 			/* The length could have changed */
14506 			pkt_len = msgdsize(mp);
14507 		}
14508 		if (ipst->ips_ip6_observe.he_interested) {
14509 			/* See above */
14510 			szone = IP_REAL_ZONEID(szone, ipst);
14511 
14512 			ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14513 			    ill, ipst);
14514 		}
14515 		DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14516 		    void_ip_t *, ip6h,  __dtrace_ipsr_ill_t *, ill,
14517 		    ipha_t *, NULL, ip6_t *, ip6h, int, 0);
14518 	}
14519 
14520 sendit:
14521 	/*
14522 	 * We check the state without a lock because the state can never
14523 	 * move "backwards" to initial or incomplete.
14524 	 */
14525 	switch (ncec->ncec_state) {
14526 	case ND_REACHABLE:
14527 	case ND_STALE:
14528 	case ND_DELAY:
14529 	case ND_PROBE:
14530 		mp = ip_xmit_attach_llhdr(mp, nce);
14531 		if (mp == NULL) {
14532 			/*
14533 			 * ip_xmit_attach_llhdr has increased
14534 			 * ipIfStatsOutDiscards and called ip_drop_output()
14535 			 */
14536 			return (ENOBUFS);
14537 		}
14538 		/*
14539 		 * check if nce_fastpath completed and we tagged on a
14540 		 * copy of nce_fp_mp in ip_xmit_attach_llhdr().
14541 		 */
14542 		fp_mp = (mp->b_datap->db_type == M_DATA);
14543 
14544 		if (fp_mp &&
14545 		    (ill->ill_capabilities & ILL_CAPAB_DLD_DIRECT)) {
14546 			ill_dld_direct_t *idd;
14547 
14548 			idd = &ill->ill_dld_capab->idc_direct;
14549 			/*
14550 			 * Send the packet directly to DLD, where it
14551 			 * may be queued depending on the availability
14552 			 * of transmit resources at the media layer.
14553 			 * Return value should be taken into
14554 			 * account and flow control the TCP.
14555 			 */
14556 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14557 			UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14558 			    pkt_len);
14559 
14560 			if (ixaflags & IXAF_NO_DEV_FLOW_CTL) {
14561 				(void) idd->idd_tx_df(idd->idd_tx_dh, mp,
14562 				    (uintptr_t)xmit_hint, IP_DROP_ON_NO_DESC);
14563 			} else {
14564 				uintptr_t cookie;
14565 
14566 				if ((cookie = idd->idd_tx_df(idd->idd_tx_dh,
14567 				    mp, (uintptr_t)xmit_hint, 0)) != 0) {
14568 					if (ixacookie != NULL)
14569 						*ixacookie = cookie;
14570 					return (EWOULDBLOCK);
14571 				}
14572 			}
14573 		} else {
14574 			wq = ill->ill_wq;
14575 
14576 			if (!(ixaflags & IXAF_NO_DEV_FLOW_CTL) &&
14577 			    !canputnext(wq)) {
14578 				if (ixacookie != NULL)
14579 					*ixacookie = 0;
14580 				ip_xmit_flowctl_drop(ill, mp, fp_mp,
14581 				    nce->nce_fp_mp != NULL ?
14582 				    MBLKL(nce->nce_fp_mp) : 0);
14583 				return (EWOULDBLOCK);
14584 			}
14585 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14586 			UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14587 			    pkt_len);
14588 			putnext(wq, mp);
14589 		}
14590 
14591 		/*
14592 		 * The rest of this function implements Neighbor Unreachability
14593 		 * detection. Determine if the ncec is eligible for NUD.
14594 		 */
14595 		if (ncec->ncec_flags & NCE_F_NONUD)
14596 			return (0);
14597 
14598 		ASSERT(ncec->ncec_state != ND_INCOMPLETE);
14599 
14600 		/*
14601 		 * Check for upper layer advice
14602 		 */
14603 		if (ixaflags & IXAF_REACH_CONF) {
14604 			timeout_id_t tid;
14605 
14606 			/*
14607 			 * It should be o.k. to check the state without
14608 			 * a lock here, at most we lose an advice.
14609 			 */
14610 			ncec->ncec_last = TICK_TO_MSEC(now);
14611 			if (ncec->ncec_state != ND_REACHABLE) {
14612 				mutex_enter(&ncec->ncec_lock);
14613 				ncec->ncec_state = ND_REACHABLE;
14614 				tid = ncec->ncec_timeout_id;
14615 				ncec->ncec_timeout_id = 0;
14616 				mutex_exit(&ncec->ncec_lock);
14617 				(void) untimeout(tid);
14618 				if (ip_debug > 2) {
14619 					/* ip1dbg */
14620 					pr_addr_dbg("ip_xmit: state"
14621 					    " for %s changed to"
14622 					    " REACHABLE\n", AF_INET6,
14623 					    &ncec->ncec_addr);
14624 				}
14625 			}
14626 			return (0);
14627 		}
14628 
14629 		delta =  TICK_TO_MSEC(now) - ncec->ncec_last;
14630 		ip1dbg(("ip_xmit: delta = %" PRId64
14631 		    " ill_reachable_time = %d \n", delta,
14632 		    ill->ill_reachable_time));
14633 		if (delta > (uint64_t)ill->ill_reachable_time) {
14634 			mutex_enter(&ncec->ncec_lock);
14635 			switch (ncec->ncec_state) {
14636 			case ND_REACHABLE:
14637 				ASSERT((ncec->ncec_flags & NCE_F_NONUD) == 0);
14638 				/* FALLTHROUGH */
14639 			case ND_STALE:
14640 				/*
14641 				 * ND_REACHABLE is identical to
14642 				 * ND_STALE in this specific case. If
14643 				 * reachable time has expired for this
14644 				 * neighbor (delta is greater than
14645 				 * reachable time), conceptually, the
14646 				 * neighbor cache is no longer in
14647 				 * REACHABLE state, but already in
14648 				 * STALE state.  So the correct
14649 				 * transition here is to ND_DELAY.
14650 				 */
14651 				ncec->ncec_state = ND_DELAY;
14652 				mutex_exit(&ncec->ncec_lock);
14653 				nce_restart_timer(ncec,
14654 				    ipst->ips_delay_first_probe_time);
14655 				if (ip_debug > 3) {
14656 					/* ip2dbg */
14657 					pr_addr_dbg("ip_xmit: state"
14658 					    " for %s changed to"
14659 					    " DELAY\n", AF_INET6,
14660 					    &ncec->ncec_addr);
14661 				}
14662 				break;
14663 			case ND_DELAY:
14664 			case ND_PROBE:
14665 				mutex_exit(&ncec->ncec_lock);
14666 				/* Timers have already started */
14667 				break;
14668 			case ND_UNREACHABLE:
14669 				/*
14670 				 * nce_timer has detected that this ncec
14671 				 * is unreachable and initiated deleting
14672 				 * this ncec.
14673 				 * This is a harmless race where we found the
14674 				 * ncec before it was deleted and have
14675 				 * just sent out a packet using this
14676 				 * unreachable ncec.
14677 				 */
14678 				mutex_exit(&ncec->ncec_lock);
14679 				break;
14680 			default:
14681 				ASSERT(0);
14682 				mutex_exit(&ncec->ncec_lock);
14683 			}
14684 		}
14685 		return (0);
14686 
14687 	case ND_INCOMPLETE:
14688 		/*
14689 		 * the state could have changed since we didn't hold the lock.
14690 		 * Re-verify state under lock.
14691 		 */
14692 		is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14693 		mutex_enter(&ncec->ncec_lock);
14694 		if (NCE_ISREACHABLE(ncec)) {
14695 			mutex_exit(&ncec->ncec_lock);
14696 			goto sendit;
14697 		}
14698 		/* queue the packet */
14699 		nce_queue_mp(ncec, mp, is_probe);
14700 		mutex_exit(&ncec->ncec_lock);
14701 		DTRACE_PROBE2(ip__xmit__incomplete,
14702 		    (ncec_t *), ncec, (mblk_t *), mp);
14703 		return (0);
14704 
14705 	case ND_INITIAL:
14706 		/*
14707 		 * State could have changed since we didn't hold the lock, so
14708 		 * re-verify state.
14709 		 */
14710 		is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14711 		mutex_enter(&ncec->ncec_lock);
14712 		if (NCE_ISREACHABLE(ncec))  {
14713 			mutex_exit(&ncec->ncec_lock);
14714 			goto sendit;
14715 		}
14716 		nce_queue_mp(ncec, mp, is_probe);
14717 		if (ncec->ncec_state == ND_INITIAL) {
14718 			ncec->ncec_state = ND_INCOMPLETE;
14719 			mutex_exit(&ncec->ncec_lock);
14720 			/*
14721 			 * figure out the source we want to use
14722 			 * and resolve it.
14723 			 */
14724 			ip_ndp_resolve(ncec);
14725 		} else  {
14726 			mutex_exit(&ncec->ncec_lock);
14727 		}
14728 		return (0);
14729 
14730 	case ND_UNREACHABLE:
14731 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14732 		ip_drop_output("ipIfStatsOutDiscards - ND_UNREACHABLE",
14733 		    mp, ill);
14734 		freemsg(mp);
14735 		return (0);
14736 
14737 	default:
14738 		ASSERT(0);
14739 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14740 		ip_drop_output("ipIfStatsOutDiscards - ND_other",
14741 		    mp, ill);
14742 		freemsg(mp);
14743 		return (ENETUNREACH);
14744 	}
14745 }
14746 
14747 /*
14748  * Return B_TRUE if the buffers differ in length or content.
14749  * This is used for comparing extension header buffers.
14750  * Note that an extension header would be declared different
14751  * even if all that changed was the next header value in that header i.e.
14752  * what really changed is the next extension header.
14753  */
14754 boolean_t
ip_cmpbuf(const void * abuf,uint_t alen,boolean_t b_valid,const void * bbuf,uint_t blen)14755 ip_cmpbuf(const void *abuf, uint_t alen, boolean_t b_valid, const void *bbuf,
14756     uint_t blen)
14757 {
14758 	if (!b_valid)
14759 		blen = 0;
14760 
14761 	if (alen != blen)
14762 		return (B_TRUE);
14763 	if (alen == 0)
14764 		return (B_FALSE);	/* Both zero length */
14765 	return (bcmp(abuf, bbuf, alen));
14766 }
14767 
14768 /*
14769  * Preallocate memory for ip_savebuf(). Returns B_TRUE if ok.
14770  * Return B_FALSE if memory allocation fails - don't change any state!
14771  */
14772 boolean_t
ip_allocbuf(void ** dstp,uint_t * dstlenp,boolean_t src_valid,const void * src,uint_t srclen)14773 ip_allocbuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14774     const void *src, uint_t srclen)
14775 {
14776 	void *dst;
14777 
14778 	if (!src_valid)
14779 		srclen = 0;
14780 
14781 	ASSERT(*dstlenp == 0);
14782 	if (src != NULL && srclen != 0) {
14783 		dst = mi_alloc(srclen, BPRI_MED);
14784 		if (dst == NULL)
14785 			return (B_FALSE);
14786 	} else {
14787 		dst = NULL;
14788 	}
14789 	if (*dstp != NULL)
14790 		mi_free(*dstp);
14791 	*dstp = dst;
14792 	*dstlenp = dst == NULL ? 0 : srclen;
14793 	return (B_TRUE);
14794 }
14795 
14796 /*
14797  * Replace what is in *dst, *dstlen with the source.
14798  * Assumes ip_allocbuf has already been called.
14799  */
14800 void
ip_savebuf(void ** dstp,uint_t * dstlenp,boolean_t src_valid,const void * src,uint_t srclen)14801 ip_savebuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14802     const void *src, uint_t srclen)
14803 {
14804 	if (!src_valid)
14805 		srclen = 0;
14806 
14807 	ASSERT(*dstlenp == srclen);
14808 	if (src != NULL && srclen != 0)
14809 		bcopy(src, *dstp, srclen);
14810 }
14811 
14812 /*
14813  * Free the storage pointed to by the members of an ip_pkt_t.
14814  */
14815 void
ip_pkt_free(ip_pkt_t * ipp)14816 ip_pkt_free(ip_pkt_t *ipp)
14817 {
14818 	uint_t	fields = ipp->ipp_fields;
14819 
14820 	if (fields & IPPF_HOPOPTS) {
14821 		kmem_free(ipp->ipp_hopopts, ipp->ipp_hopoptslen);
14822 		ipp->ipp_hopopts = NULL;
14823 		ipp->ipp_hopoptslen = 0;
14824 	}
14825 	if (fields & IPPF_RTHDRDSTOPTS) {
14826 		kmem_free(ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen);
14827 		ipp->ipp_rthdrdstopts = NULL;
14828 		ipp->ipp_rthdrdstoptslen = 0;
14829 	}
14830 	if (fields & IPPF_DSTOPTS) {
14831 		kmem_free(ipp->ipp_dstopts, ipp->ipp_dstoptslen);
14832 		ipp->ipp_dstopts = NULL;
14833 		ipp->ipp_dstoptslen = 0;
14834 	}
14835 	if (fields & IPPF_RTHDR) {
14836 		kmem_free(ipp->ipp_rthdr, ipp->ipp_rthdrlen);
14837 		ipp->ipp_rthdr = NULL;
14838 		ipp->ipp_rthdrlen = 0;
14839 	}
14840 	if (fields & IPPF_IPV4_OPTIONS) {
14841 		kmem_free(ipp->ipp_ipv4_options, ipp->ipp_ipv4_options_len);
14842 		ipp->ipp_ipv4_options = NULL;
14843 		ipp->ipp_ipv4_options_len = 0;
14844 	}
14845 	if (fields & IPPF_LABEL_V4) {
14846 		kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
14847 		ipp->ipp_label_v4 = NULL;
14848 		ipp->ipp_label_len_v4 = 0;
14849 	}
14850 	if (fields & IPPF_LABEL_V6) {
14851 		kmem_free(ipp->ipp_label_v6, ipp->ipp_label_len_v6);
14852 		ipp->ipp_label_v6 = NULL;
14853 		ipp->ipp_label_len_v6 = 0;
14854 	}
14855 	ipp->ipp_fields &= ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14856 	    IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14857 }
14858 
14859 /*
14860  * Copy from src to dst and allocate as needed.
14861  * Returns zero or ENOMEM.
14862  *
14863  * The caller must initialize dst to zero.
14864  */
14865 int
ip_pkt_copy(ip_pkt_t * src,ip_pkt_t * dst,int kmflag)14866 ip_pkt_copy(ip_pkt_t *src, ip_pkt_t *dst, int kmflag)
14867 {
14868 	uint_t	fields = src->ipp_fields;
14869 
14870 	/* Start with fields that don't require memory allocation */
14871 	dst->ipp_fields = fields &
14872 	    ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14873 	    IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14874 
14875 	dst->ipp_addr = src->ipp_addr;
14876 	dst->ipp_unicast_hops = src->ipp_unicast_hops;
14877 	dst->ipp_hoplimit = src->ipp_hoplimit;
14878 	dst->ipp_tclass = src->ipp_tclass;
14879 	dst->ipp_type_of_service = src->ipp_type_of_service;
14880 
14881 	if (!(fields & (IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14882 	    IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6)))
14883 		return (0);
14884 
14885 	if (fields & IPPF_HOPOPTS) {
14886 		dst->ipp_hopopts = kmem_alloc(src->ipp_hopoptslen, kmflag);
14887 		if (dst->ipp_hopopts == NULL) {
14888 			ip_pkt_free(dst);
14889 			return (ENOMEM);
14890 		}
14891 		dst->ipp_fields |= IPPF_HOPOPTS;
14892 		bcopy(src->ipp_hopopts, dst->ipp_hopopts,
14893 		    src->ipp_hopoptslen);
14894 		dst->ipp_hopoptslen = src->ipp_hopoptslen;
14895 	}
14896 	if (fields & IPPF_RTHDRDSTOPTS) {
14897 		dst->ipp_rthdrdstopts = kmem_alloc(src->ipp_rthdrdstoptslen,
14898 		    kmflag);
14899 		if (dst->ipp_rthdrdstopts == NULL) {
14900 			ip_pkt_free(dst);
14901 			return (ENOMEM);
14902 		}
14903 		dst->ipp_fields |= IPPF_RTHDRDSTOPTS;
14904 		bcopy(src->ipp_rthdrdstopts, dst->ipp_rthdrdstopts,
14905 		    src->ipp_rthdrdstoptslen);
14906 		dst->ipp_rthdrdstoptslen = src->ipp_rthdrdstoptslen;
14907 	}
14908 	if (fields & IPPF_DSTOPTS) {
14909 		dst->ipp_dstopts = kmem_alloc(src->ipp_dstoptslen, kmflag);
14910 		if (dst->ipp_dstopts == NULL) {
14911 			ip_pkt_free(dst);
14912 			return (ENOMEM);
14913 		}
14914 		dst->ipp_fields |= IPPF_DSTOPTS;
14915 		bcopy(src->ipp_dstopts, dst->ipp_dstopts,
14916 		    src->ipp_dstoptslen);
14917 		dst->ipp_dstoptslen = src->ipp_dstoptslen;
14918 	}
14919 	if (fields & IPPF_RTHDR) {
14920 		dst->ipp_rthdr = kmem_alloc(src->ipp_rthdrlen, kmflag);
14921 		if (dst->ipp_rthdr == NULL) {
14922 			ip_pkt_free(dst);
14923 			return (ENOMEM);
14924 		}
14925 		dst->ipp_fields |= IPPF_RTHDR;
14926 		bcopy(src->ipp_rthdr, dst->ipp_rthdr,
14927 		    src->ipp_rthdrlen);
14928 		dst->ipp_rthdrlen = src->ipp_rthdrlen;
14929 	}
14930 	if (fields & IPPF_IPV4_OPTIONS) {
14931 		dst->ipp_ipv4_options = kmem_alloc(src->ipp_ipv4_options_len,
14932 		    kmflag);
14933 		if (dst->ipp_ipv4_options == NULL) {
14934 			ip_pkt_free(dst);
14935 			return (ENOMEM);
14936 		}
14937 		dst->ipp_fields |= IPPF_IPV4_OPTIONS;
14938 		bcopy(src->ipp_ipv4_options, dst->ipp_ipv4_options,
14939 		    src->ipp_ipv4_options_len);
14940 		dst->ipp_ipv4_options_len = src->ipp_ipv4_options_len;
14941 	}
14942 	if (fields & IPPF_LABEL_V4) {
14943 		dst->ipp_label_v4 = kmem_alloc(src->ipp_label_len_v4, kmflag);
14944 		if (dst->ipp_label_v4 == NULL) {
14945 			ip_pkt_free(dst);
14946 			return (ENOMEM);
14947 		}
14948 		dst->ipp_fields |= IPPF_LABEL_V4;
14949 		bcopy(src->ipp_label_v4, dst->ipp_label_v4,
14950 		    src->ipp_label_len_v4);
14951 		dst->ipp_label_len_v4 = src->ipp_label_len_v4;
14952 	}
14953 	if (fields & IPPF_LABEL_V6) {
14954 		dst->ipp_label_v6 = kmem_alloc(src->ipp_label_len_v6, kmflag);
14955 		if (dst->ipp_label_v6 == NULL) {
14956 			ip_pkt_free(dst);
14957 			return (ENOMEM);
14958 		}
14959 		dst->ipp_fields |= IPPF_LABEL_V6;
14960 		bcopy(src->ipp_label_v6, dst->ipp_label_v6,
14961 		    src->ipp_label_len_v6);
14962 		dst->ipp_label_len_v6 = src->ipp_label_len_v6;
14963 	}
14964 	if (fields & IPPF_FRAGHDR) {
14965 		dst->ipp_fraghdr = kmem_alloc(src->ipp_fraghdrlen, kmflag);
14966 		if (dst->ipp_fraghdr == NULL) {
14967 			ip_pkt_free(dst);
14968 			return (ENOMEM);
14969 		}
14970 		dst->ipp_fields |= IPPF_FRAGHDR;
14971 		bcopy(src->ipp_fraghdr, dst->ipp_fraghdr,
14972 		    src->ipp_fraghdrlen);
14973 		dst->ipp_fraghdrlen = src->ipp_fraghdrlen;
14974 	}
14975 	return (0);
14976 }
14977 
14978 /*
14979  * Returns INADDR_ANY if no source route
14980  */
14981 ipaddr_t
ip_pkt_source_route_v4(const ip_pkt_t * ipp)14982 ip_pkt_source_route_v4(const ip_pkt_t *ipp)
14983 {
14984 	ipaddr_t	nexthop = INADDR_ANY;
14985 	ipoptp_t	opts;
14986 	uchar_t		*opt;
14987 	uint8_t		optval;
14988 	uint8_t		optlen;
14989 	uint32_t	totallen;
14990 
14991 	if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
14992 		return (INADDR_ANY);
14993 
14994 	totallen = ipp->ipp_ipv4_options_len;
14995 	if (totallen & 0x3)
14996 		return (INADDR_ANY);
14997 
14998 	for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
14999 	    optval != IPOPT_EOL;
15000 	    optval = ipoptp_next(&opts)) {
15001 		opt = opts.ipoptp_cur;
15002 		switch (optval) {
15003 			uint8_t off;
15004 		case IPOPT_SSRR:
15005 		case IPOPT_LSRR:
15006 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15007 				break;
15008 			}
15009 			optlen = opts.ipoptp_len;
15010 			off = opt[IPOPT_OFFSET];
15011 			off--;
15012 			if (optlen < IP_ADDR_LEN ||
15013 			    off > optlen - IP_ADDR_LEN) {
15014 				/* End of source route */
15015 				break;
15016 			}
15017 			bcopy((char *)opt + off, &nexthop, IP_ADDR_LEN);
15018 			if (nexthop == htonl(INADDR_LOOPBACK)) {
15019 				/* Ignore */
15020 				nexthop = INADDR_ANY;
15021 				break;
15022 			}
15023 			break;
15024 		}
15025 	}
15026 	return (nexthop);
15027 }
15028 
15029 /*
15030  * Reverse a source route.
15031  */
15032 void
ip_pkt_source_route_reverse_v4(ip_pkt_t * ipp)15033 ip_pkt_source_route_reverse_v4(ip_pkt_t *ipp)
15034 {
15035 	ipaddr_t	tmp;
15036 	ipoptp_t	opts;
15037 	uchar_t		*opt;
15038 	uint8_t		optval;
15039 	uint32_t	totallen;
15040 
15041 	if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
15042 		return;
15043 
15044 	totallen = ipp->ipp_ipv4_options_len;
15045 	if (totallen & 0x3)
15046 		return;
15047 
15048 	for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15049 	    optval != IPOPT_EOL;
15050 	    optval = ipoptp_next(&opts)) {
15051 		uint8_t off1, off2;
15052 
15053 		opt = opts.ipoptp_cur;
15054 		switch (optval) {
15055 		case IPOPT_SSRR:
15056 		case IPOPT_LSRR:
15057 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15058 				break;
15059 			}
15060 			off1 = IPOPT_MINOFF_SR - 1;
15061 			off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
15062 			while (off2 > off1) {
15063 				bcopy(opt + off2, &tmp, IP_ADDR_LEN);
15064 				bcopy(opt + off1, opt + off2, IP_ADDR_LEN);
15065 				bcopy(&tmp, opt + off2, IP_ADDR_LEN);
15066 				off2 -= IP_ADDR_LEN;
15067 				off1 += IP_ADDR_LEN;
15068 			}
15069 			opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
15070 			break;
15071 		}
15072 	}
15073 }
15074 
15075 /*
15076  * Returns NULL if no routing header
15077  */
15078 in6_addr_t *
ip_pkt_source_route_v6(const ip_pkt_t * ipp)15079 ip_pkt_source_route_v6(const ip_pkt_t *ipp)
15080 {
15081 	in6_addr_t	*nexthop = NULL;
15082 	ip6_rthdr0_t	*rthdr;
15083 
15084 	if (!(ipp->ipp_fields & IPPF_RTHDR))
15085 		return (NULL);
15086 
15087 	rthdr = (ip6_rthdr0_t *)ipp->ipp_rthdr;
15088 	if (rthdr->ip6r0_segleft == 0)
15089 		return (NULL);
15090 
15091 	nexthop = (in6_addr_t *)((char *)rthdr + sizeof (*rthdr));
15092 	return (nexthop);
15093 }
15094 
15095 zoneid_t
ip_get_zoneid_v4(ipaddr_t addr,mblk_t * mp,ip_recv_attr_t * ira,zoneid_t lookup_zoneid)15096 ip_get_zoneid_v4(ipaddr_t addr, mblk_t *mp, ip_recv_attr_t *ira,
15097     zoneid_t lookup_zoneid)
15098 {
15099 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
15100 	ire_t		*ire;
15101 	int		ire_flags = MATCH_IRE_TYPE;
15102 	zoneid_t	zoneid = ALL_ZONES;
15103 
15104 	if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15105 		return (ALL_ZONES);
15106 
15107 	if (lookup_zoneid != ALL_ZONES)
15108 		ire_flags |= MATCH_IRE_ZONEONLY;
15109 	ire = ire_ftable_lookup_v4(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15110 	    NULL, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15111 	if (ire != NULL) {
15112 		zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15113 		ire_refrele(ire);
15114 	}
15115 	return (zoneid);
15116 }
15117 
15118 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)15119 ip_get_zoneid_v6(in6_addr_t *addr, mblk_t *mp, const ill_t *ill,
15120     ip_recv_attr_t *ira, zoneid_t lookup_zoneid)
15121 {
15122 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
15123 	ire_t		*ire;
15124 	int		ire_flags = MATCH_IRE_TYPE;
15125 	zoneid_t	zoneid = ALL_ZONES;
15126 
15127 	if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15128 		return (ALL_ZONES);
15129 
15130 	if (IN6_IS_ADDR_LINKLOCAL(addr))
15131 		ire_flags |= MATCH_IRE_ILL;
15132 
15133 	if (lookup_zoneid != ALL_ZONES)
15134 		ire_flags |= MATCH_IRE_ZONEONLY;
15135 	ire = ire_ftable_lookup_v6(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15136 	    ill, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15137 	if (ire != NULL) {
15138 		zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15139 		ire_refrele(ire);
15140 	}
15141 	return (zoneid);
15142 }
15143 
15144 /*
15145  * IP obserability hook support functions.
15146  */
15147 static void
ipobs_init(ip_stack_t * ipst)15148 ipobs_init(ip_stack_t *ipst)
15149 {
15150 	netid_t id;
15151 
15152 	id = net_getnetidbynetstackid(ipst->ips_netstack->netstack_stackid);
15153 
15154 	ipst->ips_ip4_observe_pr = net_protocol_lookup(id, NHF_INET);
15155 	VERIFY(ipst->ips_ip4_observe_pr != NULL);
15156 
15157 	ipst->ips_ip6_observe_pr = net_protocol_lookup(id, NHF_INET6);
15158 	VERIFY(ipst->ips_ip6_observe_pr != NULL);
15159 }
15160 
15161 static void
ipobs_fini(ip_stack_t * ipst)15162 ipobs_fini(ip_stack_t *ipst)
15163 {
15164 
15165 	VERIFY(net_protocol_release(ipst->ips_ip4_observe_pr) == 0);
15166 	VERIFY(net_protocol_release(ipst->ips_ip6_observe_pr) == 0);
15167 }
15168 
15169 /*
15170  * hook_pkt_observe_t is composed in network byte order so that the
15171  * entire mblk_t chain handed into hook_run can be used as-is.
15172  * The caveat is that use of the fields, such as the zone fields,
15173  * requires conversion into host byte order first.
15174  */
15175 void
ipobs_hook(mblk_t * mp,int htype,zoneid_t zsrc,zoneid_t zdst,const ill_t * ill,ip_stack_t * ipst)15176 ipobs_hook(mblk_t *mp, int htype, zoneid_t zsrc, zoneid_t zdst,
15177     const ill_t *ill, ip_stack_t *ipst)
15178 {
15179 	hook_pkt_observe_t *hdr;
15180 	uint64_t grifindex;
15181 	mblk_t *imp;
15182 
15183 	imp = allocb(sizeof (*hdr), BPRI_HI);
15184 	if (imp == NULL)
15185 		return;
15186 
15187 	hdr = (hook_pkt_observe_t *)imp->b_rptr;
15188 	/*
15189 	 * b_wptr is set to make the apparent size of the data in the mblk_t
15190 	 * to exclude the pointers at the end of hook_pkt_observer_t.
15191 	 */
15192 	imp->b_wptr = imp->b_rptr + sizeof (dl_ipnetinfo_t);
15193 	imp->b_cont = mp;
15194 
15195 	ASSERT(DB_TYPE(mp) == M_DATA);
15196 
15197 	if (IS_UNDER_IPMP(ill))
15198 		grifindex = ipmp_ill_get_ipmp_ifindex(ill);
15199 	else
15200 		grifindex = 0;
15201 
15202 	hdr->hpo_version = 1;
15203 	hdr->hpo_htype = htons(htype);
15204 	hdr->hpo_pktlen = htonl((ulong_t)msgdsize(mp));
15205 	hdr->hpo_ifindex = htonl(ill->ill_phyint->phyint_ifindex);
15206 	hdr->hpo_grifindex = htonl(grifindex);
15207 	hdr->hpo_zsrc = htonl(zsrc);
15208 	hdr->hpo_zdst = htonl(zdst);
15209 	hdr->hpo_pkt = imp;
15210 	hdr->hpo_ctx = ipst->ips_netstack;
15211 
15212 	if (ill->ill_isv6) {
15213 		hdr->hpo_family = AF_INET6;
15214 		(void) hook_run(ipst->ips_ipv6_net_data->netd_hooks,
15215 		    ipst->ips_ipv6observing, (hook_data_t)hdr);
15216 	} else {
15217 		hdr->hpo_family = AF_INET;
15218 		(void) hook_run(ipst->ips_ipv4_net_data->netd_hooks,
15219 		    ipst->ips_ipv4observing, (hook_data_t)hdr);
15220 	}
15221 
15222 	imp->b_cont = NULL;
15223 	freemsg(imp);
15224 }
15225 
15226 /*
15227  * Utility routine that checks if `v4srcp' is a valid address on underlying
15228  * interface `ill'.  If `ipifp' is non-NULL, it's set to a held ipif
15229  * associated with `v4srcp' on success.  NOTE: if this is not called from
15230  * inside the IPSQ (ill_g_lock is not held), `ill' may be removed from the
15231  * group during or after this lookup.
15232  */
15233 boolean_t
ipif_lookup_testaddr_v4(ill_t * ill,const in_addr_t * v4srcp,ipif_t ** ipifp)15234 ipif_lookup_testaddr_v4(ill_t *ill, const in_addr_t *v4srcp, ipif_t **ipifp)
15235 {
15236 	ipif_t *ipif;
15237 
15238 	ipif = ipif_lookup_addr_exact(*v4srcp, ill, ill->ill_ipst);
15239 	if (ipif != NULL) {
15240 		if (ipifp != NULL)
15241 			*ipifp = ipif;
15242 		else
15243 			ipif_refrele(ipif);
15244 		return (B_TRUE);
15245 	}
15246 
15247 	ip1dbg(("ipif_lookup_testaddr_v4: cannot find ipif for src %x\n",
15248 	    *v4srcp));
15249 	return (B_FALSE);
15250 }
15251 
15252 /*
15253  * Transport protocol call back function for CPU state change.
15254  */
15255 /* ARGSUSED */
15256 static int
ip_tp_cpu_update(cpu_setup_t what,int id,void * arg)15257 ip_tp_cpu_update(cpu_setup_t what, int id, void *arg)
15258 {
15259 	processorid_t cpu_seqid;
15260 	netstack_handle_t nh;
15261 	netstack_t *ns;
15262 
15263 	ASSERT(MUTEX_HELD(&cpu_lock));
15264 
15265 	switch (what) {
15266 	case CPU_CONFIG:
15267 	case CPU_ON:
15268 	case CPU_INIT:
15269 	case CPU_CPUPART_IN:
15270 		cpu_seqid = cpu[id]->cpu_seqid;
15271 		netstack_next_init(&nh);
15272 		while ((ns = netstack_next(&nh)) != NULL) {
15273 			tcp_stack_cpu_add(ns->netstack_tcp, cpu_seqid);
15274 			sctp_stack_cpu_add(ns->netstack_sctp, cpu_seqid);
15275 			udp_stack_cpu_add(ns->netstack_udp, cpu_seqid);
15276 			netstack_rele(ns);
15277 		}
15278 		netstack_next_fini(&nh);
15279 		break;
15280 	case CPU_UNCONFIG:
15281 	case CPU_OFF:
15282 	case CPU_CPUPART_OUT:
15283 		/*
15284 		 * Nothing to do.  We don't remove the per CPU stats from
15285 		 * the IP stack even when the CPU goes offline.
15286 		 */
15287 		break;
15288 	default:
15289 		break;
15290 	}
15291 	return (0);
15292 }
15293