xref: /titanic_54/usr/src/uts/common/inet/udp/udp.c (revision 6e04cb6cd0db0a60ffe2ed9f5db5c2ab91c702ff)
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  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright 2014, OmniTI Computer Consulting, Inc. All rights reserved.
25  */
26 /* Copyright (c) 1990 Mentat Inc. */
27 
28 #include <sys/sysmacros.h>
29 #include <sys/types.h>
30 #include <sys/stream.h>
31 #include <sys/stropts.h>
32 #include <sys/strlog.h>
33 #include <sys/strsun.h>
34 #define	_SUN_TPI_VERSION 2
35 #include <sys/tihdr.h>
36 #include <sys/timod.h>
37 #include <sys/ddi.h>
38 #include <sys/sunddi.h>
39 #include <sys/strsubr.h>
40 #include <sys/suntpi.h>
41 #include <sys/xti_inet.h>
42 #include <sys/kmem.h>
43 #include <sys/cred_impl.h>
44 #include <sys/policy.h>
45 #include <sys/priv.h>
46 #include <sys/ucred.h>
47 #include <sys/zone.h>
48 
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sockio.h>
52 #include <sys/vtrace.h>
53 #include <sys/sdt.h>
54 #include <sys/debug.h>
55 #include <sys/isa_defs.h>
56 #include <sys/random.h>
57 #include <netinet/in.h>
58 #include <netinet/ip6.h>
59 #include <netinet/icmp6.h>
60 #include <netinet/udp.h>
61 
62 #include <inet/common.h>
63 #include <inet/ip.h>
64 #include <inet/ip_impl.h>
65 #include <inet/ipsec_impl.h>
66 #include <inet/ip6.h>
67 #include <inet/ip_ire.h>
68 #include <inet/ip_if.h>
69 #include <inet/ip_multi.h>
70 #include <inet/ip_ndp.h>
71 #include <inet/proto_set.h>
72 #include <inet/mib2.h>
73 #include <inet/optcom.h>
74 #include <inet/snmpcom.h>
75 #include <inet/kstatcom.h>
76 #include <inet/ipclassifier.h>
77 #include <sys/squeue_impl.h>
78 #include <inet/ipnet.h>
79 #include <sys/ethernet.h>
80 
81 #include <sys/tsol/label.h>
82 #include <sys/tsol/tnet.h>
83 #include <rpc/pmap_prot.h>
84 
85 #include <inet/udp_impl.h>
86 
87 /*
88  * Synchronization notes:
89  *
90  * UDP is MT and uses the usual kernel synchronization primitives. There are 2
91  * locks, the fanout lock (uf_lock) and conn_lock. conn_lock
92  * protects the contents of the udp_t. uf_lock protects the address and the
93  * fanout information.
94  * The lock order is conn_lock -> uf_lock.
95  *
96  * The fanout lock uf_lock:
97  * When a UDP endpoint is bound to a local port, it is inserted into
98  * a bind hash list.  The list consists of an array of udp_fanout_t buckets.
99  * The size of the array is controlled by the udp_bind_fanout_size variable.
100  * This variable can be changed in /etc/system if the default value is
101  * not large enough.  Each bind hash bucket is protected by a per bucket
102  * lock.  It protects the udp_bind_hash and udp_ptpbhn fields in the udp_t
103  * structure and a few other fields in the udp_t. A UDP endpoint is removed
104  * from the bind hash list only when it is being unbound or being closed.
105  * The per bucket lock also protects a UDP endpoint's state changes.
106  *
107  * Plumbing notes:
108  * UDP is always a device driver. For compatibility with mibopen() code
109  * it is possible to I_PUSH "udp", but that results in pushing a passthrough
110  * dummy module.
111  *
112  * The above implies that we don't support any intermediate module to
113  * reside in between /dev/ip and udp -- in fact, we never supported such
114  * scenario in the past as the inter-layer communication semantics have
115  * always been private.
116  */
117 
118 /* For /etc/system control */
119 uint_t udp_bind_fanout_size = UDP_BIND_FANOUT_SIZE;
120 
121 static void	udp_addr_req(queue_t *q, mblk_t *mp);
122 static void	udp_tpi_bind(queue_t *q, mblk_t *mp);
123 static void	udp_bind_hash_insert(udp_fanout_t *uf, udp_t *udp);
124 static void	udp_bind_hash_remove(udp_t *udp, boolean_t caller_holds_lock);
125 static int	udp_build_hdr_template(conn_t *, const in6_addr_t *,
126     const in6_addr_t *, in_port_t, uint32_t);
127 static void	udp_capability_req(queue_t *q, mblk_t *mp);
128 static int	udp_tpi_close(queue_t *q, int flags);
129 static void	udp_close_free(conn_t *);
130 static void	udp_tpi_connect(queue_t *q, mblk_t *mp);
131 static void	udp_tpi_disconnect(queue_t *q, mblk_t *mp);
132 static void	udp_err_ack(queue_t *q, mblk_t *mp, t_scalar_t t_error,
133     int sys_error);
134 static void	udp_err_ack_prim(queue_t *q, mblk_t *mp, t_scalar_t primitive,
135     t_scalar_t tlierr, int sys_error);
136 static int	udp_extra_priv_ports_get(queue_t *q, mblk_t *mp, caddr_t cp,
137 		    cred_t *cr);
138 static int	udp_extra_priv_ports_add(queue_t *q, mblk_t *mp,
139 		    char *value, caddr_t cp, cred_t *cr);
140 static int	udp_extra_priv_ports_del(queue_t *q, mblk_t *mp,
141 		    char *value, caddr_t cp, cred_t *cr);
142 static void	udp_icmp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
143 static void	udp_icmp_error_ipv6(conn_t *connp, mblk_t *mp,
144     ip_recv_attr_t *ira);
145 static void	udp_info_req(queue_t *q, mblk_t *mp);
146 static void	udp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
147 static void	udp_lrput(queue_t *, mblk_t *);
148 static void	udp_lwput(queue_t *, mblk_t *);
149 static int	udp_open(queue_t *q, dev_t *devp, int flag, int sflag,
150 		    cred_t *credp, boolean_t isv6);
151 static int	udp_openv4(queue_t *q, dev_t *devp, int flag, int sflag,
152 		    cred_t *credp);
153 static int	udp_openv6(queue_t *q, dev_t *devp, int flag, int sflag,
154 		    cred_t *credp);
155 static boolean_t udp_opt_allow_udr_set(t_scalar_t level, t_scalar_t name);
156 int		udp_opt_set(conn_t *connp, uint_t optset_context,
157 		    int level, int name, uint_t inlen,
158 		    uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
159 		    void *thisdg_attrs, cred_t *cr);
160 int		udp_opt_get(conn_t *connp, int level, int name,
161 		    uchar_t *ptr);
162 static int	udp_output_connected(conn_t *connp, mblk_t *mp, cred_t *cr,
163 		    pid_t pid);
164 static int	udp_output_lastdst(conn_t *connp, mblk_t *mp, cred_t *cr,
165     pid_t pid, ip_xmit_attr_t *ixa);
166 static int	udp_output_newdst(conn_t *connp, mblk_t *data_mp, sin_t *sin,
167 		    sin6_t *sin6, ushort_t ipversion, cred_t *cr, pid_t,
168 		    ip_xmit_attr_t *ixa);
169 static mblk_t	*udp_prepend_hdr(conn_t *, ip_xmit_attr_t *, const ip_pkt_t *,
170     const in6_addr_t *, const in6_addr_t *, in_port_t, uint32_t, mblk_t *,
171     int *);
172 static mblk_t	*udp_prepend_header_template(conn_t *, ip_xmit_attr_t *,
173     mblk_t *, const in6_addr_t *, in_port_t, uint32_t, int *);
174 static void	udp_ud_err(queue_t *q, mblk_t *mp, t_scalar_t err);
175 static void	udp_ud_err_connected(conn_t *, t_scalar_t);
176 static void	udp_tpi_unbind(queue_t *q, mblk_t *mp);
177 static in_port_t udp_update_next_port(udp_t *udp, in_port_t port,
178     boolean_t random);
179 static void	udp_wput_other(queue_t *q, mblk_t *mp);
180 static void	udp_wput_iocdata(queue_t *q, mblk_t *mp);
181 static void	udp_wput_fallback(queue_t *q, mblk_t *mp);
182 static size_t	udp_set_rcv_hiwat(udp_t *udp, size_t size);
183 
184 static void	*udp_stack_init(netstackid_t stackid, netstack_t *ns);
185 static void	udp_stack_fini(netstackid_t stackid, void *arg);
186 
187 /* Common routines for TPI and socket module */
188 static void	udp_ulp_recv(conn_t *, mblk_t *, uint_t, ip_recv_attr_t *);
189 
190 /* Common routine for TPI and socket module */
191 static conn_t	*udp_do_open(cred_t *, boolean_t, int, int *);
192 static void	udp_do_close(conn_t *);
193 static int	udp_do_bind(conn_t *, struct sockaddr *, socklen_t, cred_t *,
194     boolean_t);
195 static int	udp_do_unbind(conn_t *);
196 
197 int		udp_getsockname(sock_lower_handle_t,
198     struct sockaddr *, socklen_t *, cred_t *);
199 int		udp_getpeername(sock_lower_handle_t,
200     struct sockaddr *, socklen_t *, cred_t *);
201 static int	udp_do_connect(conn_t *, const struct sockaddr *, socklen_t,
202     cred_t *, pid_t);
203 
204 #pragma inline(udp_output_connected, udp_output_newdst, udp_output_lastdst)
205 
206 /*
207  * Checks if the given destination addr/port is allowed out.
208  * If allowed, registers the (dest_addr/port, node_ID) mapping at Cluster.
209  * Called for each connect() and for sendto()/sendmsg() to a different
210  * destination.
211  * For connect(), called in udp_connect().
212  * For sendto()/sendmsg(), called in udp_output_newdst().
213  *
214  * This macro assumes that the cl_inet_connect2 hook is not NULL.
215  * Please check this before calling this macro.
216  *
217  * void
218  * CL_INET_UDP_CONNECT(conn_t cp, udp_t *udp, boolean_t is_outgoing,
219  *     in6_addr_t *faddrp, in_port_t (or uint16_t) fport, int err);
220  */
221 #define	CL_INET_UDP_CONNECT(cp, is_outgoing, faddrp, fport, err) {	\
222 	(err) = 0;							\
223 	/*								\
224 	 * Running in cluster mode - check and register active		\
225 	 * "connection" information					\
226 	 */								\
227 	if ((cp)->conn_ipversion == IPV4_VERSION)			\
228 		(err) = (*cl_inet_connect2)(				\
229 		    (cp)->conn_netstack->netstack_stackid,		\
230 		    IPPROTO_UDP, is_outgoing, AF_INET,			\
231 		    (uint8_t *)&((cp)->conn_laddr_v4),			\
232 		    (cp)->conn_lport,					\
233 		    (uint8_t *)&(V4_PART_OF_V6(*faddrp)),		\
234 		    (in_port_t)(fport), NULL);				\
235 	else								\
236 		(err) = (*cl_inet_connect2)(				\
237 		    (cp)->conn_netstack->netstack_stackid,		\
238 		    IPPROTO_UDP, is_outgoing, AF_INET6,			\
239 		    (uint8_t *)&((cp)->conn_laddr_v6),			\
240 		    (cp)->conn_lport,					\
241 		    (uint8_t *)(faddrp), (in_port_t)(fport), NULL);	\
242 }
243 
244 static struct module_info udp_mod_info =  {
245 	UDP_MOD_ID, UDP_MOD_NAME, 1, INFPSZ, UDP_RECV_HIWATER, UDP_RECV_LOWATER
246 };
247 
248 /*
249  * Entry points for UDP as a device.
250  * We have separate open functions for the /dev/udp and /dev/udp6 devices.
251  */
252 static struct qinit udp_rinitv4 = {
253 	NULL, NULL, udp_openv4, udp_tpi_close, NULL, &udp_mod_info, NULL
254 };
255 
256 static struct qinit udp_rinitv6 = {
257 	NULL, NULL, udp_openv6, udp_tpi_close, NULL, &udp_mod_info, NULL
258 };
259 
260 static struct qinit udp_winit = {
261 	(pfi_t)udp_wput, (pfi_t)ip_wsrv, NULL, NULL, NULL, &udp_mod_info
262 };
263 
264 /* UDP entry point during fallback */
265 struct qinit udp_fallback_sock_winit = {
266 	(pfi_t)udp_wput_fallback, NULL, NULL, NULL, NULL, &udp_mod_info
267 };
268 
269 /*
270  * UDP needs to handle I_LINK and I_PLINK since ifconfig
271  * likes to use it as a place to hang the various streams.
272  */
273 static struct qinit udp_lrinit = {
274 	(pfi_t)udp_lrput, NULL, udp_openv4, udp_tpi_close, NULL, &udp_mod_info
275 };
276 
277 static struct qinit udp_lwinit = {
278 	(pfi_t)udp_lwput, NULL, udp_openv4, udp_tpi_close, NULL, &udp_mod_info
279 };
280 
281 /* For AF_INET aka /dev/udp */
282 struct streamtab udpinfov4 = {
283 	&udp_rinitv4, &udp_winit, &udp_lrinit, &udp_lwinit
284 };
285 
286 /* For AF_INET6 aka /dev/udp6 */
287 struct streamtab udpinfov6 = {
288 	&udp_rinitv6, &udp_winit, &udp_lrinit, &udp_lwinit
289 };
290 
291 #define REUSELIST_MAX 64
292 struct reuselist {
293 	conn_t	*ru_conns[REUSELIST_MAX];
294 	int	ru_entries;			/* number of entries */
295 	int	ru_next;			/* round-robin pointer */
296 	kmutex_t ru_lock;
297 };
298 
299 #define	UDP_MAXPACKET_IPV4 (IP_MAXPACKET - UDPH_SIZE - IP_SIMPLE_HDR_LENGTH)
300 
301 /* Default structure copied into T_INFO_ACK messages */
302 static struct T_info_ack udp_g_t_info_ack_ipv4 = {
303 	T_INFO_ACK,
304 	UDP_MAXPACKET_IPV4,	/* TSDU_size. Excl. headers */
305 	T_INVALID,	/* ETSU_size.  udp does not support expedited data. */
306 	T_INVALID,	/* CDATA_size. udp does not support connect data. */
307 	T_INVALID,	/* DDATA_size. udp does not support disconnect data. */
308 	sizeof (sin_t),	/* ADDR_size. */
309 	0,		/* OPT_size - not initialized here */
310 	UDP_MAXPACKET_IPV4,	/* TIDU_size.  Excl. headers */
311 	T_CLTS,		/* SERV_type.  udp supports connection-less. */
312 	TS_UNBND,	/* CURRENT_state.  This is set from udp_state. */
313 	(XPG4_1|SENDZERO) /* PROVIDER_flag */
314 };
315 
316 #define	UDP_MAXPACKET_IPV6 (IP_MAXPACKET - UDPH_SIZE - IPV6_HDR_LEN)
317 
318 static	struct T_info_ack udp_g_t_info_ack_ipv6 = {
319 	T_INFO_ACK,
320 	UDP_MAXPACKET_IPV6,	/* TSDU_size.  Excl. headers */
321 	T_INVALID,	/* ETSU_size.  udp does not support expedited data. */
322 	T_INVALID,	/* CDATA_size. udp does not support connect data. */
323 	T_INVALID,	/* DDATA_size. udp does not support disconnect data. */
324 	sizeof (sin6_t), /* ADDR_size. */
325 	0,		/* OPT_size - not initialized here */
326 	UDP_MAXPACKET_IPV6,	/* TIDU_size. Excl. headers */
327 	T_CLTS,		/* SERV_type.  udp supports connection-less. */
328 	TS_UNBND,	/* CURRENT_state.  This is set from udp_state. */
329 	(XPG4_1|SENDZERO) /* PROVIDER_flag */
330 };
331 
332 /*
333  * UDP tunables related declarations. Definitions are in udp_tunables.c
334  */
335 extern mod_prop_info_t udp_propinfo_tbl[];
336 extern int udp_propinfo_count;
337 
338 /* Setable in /etc/system */
339 /* If set to 0, pick ephemeral port sequentially; otherwise randomly. */
340 uint32_t udp_random_anon_port = 1;
341 
342 /*
343  * Hook functions to enable cluster networking.
344  * On non-clustered systems these vectors must always be NULL
345  */
346 
347 void (*cl_inet_bind)(netstackid_t stack_id, uchar_t protocol,
348     sa_family_t addr_family, uint8_t *laddrp, in_port_t lport,
349     void *args) = NULL;
350 void (*cl_inet_unbind)(netstackid_t stack_id, uint8_t protocol,
351     sa_family_t addr_family, uint8_t *laddrp, in_port_t lport,
352     void *args) = NULL;
353 
354 typedef union T_primitives *t_primp_t;
355 
356 
357 /*
358  * udp_reuselist_add() and udp_reuselist_remove() are protected agains
359  * concurrent calls by uf_lock
360  * ru_lock protects add/remove against use of the list in udp_input()
361  */
362 static int
363 udp_reuselist_add(struct reuselist *reusep, conn_t *connp)
364 {
365 	ASSERT(MUTEX_HELD(&connp->conn_lock));
366 
367 	mutex_enter(&reusep->ru_lock);
368 
369 	if (reusep->ru_entries == REUSELIST_MAX) {
370 		mutex_exit(&reusep->ru_lock);
371 		return -1;
372 	}
373 
374 	reusep->ru_conns[reusep->ru_entries++] = connp;
375 	connp->conn_reuselist = reusep;
376 
377 	mutex_exit(&reusep->ru_lock);
378 	return 0;
379 }
380 
381 static void
382 udp_reuselist_remove(conn_t *connp)
383 {
384 	int i;
385 	struct reuselist *reusep = connp->conn_reuselist;
386 
387 	if (reusep == NULL)
388 		return;
389 
390 	ASSERT(MUTEX_HELD(&connp->conn_lock));
391 
392 	mutex_enter(&reusep->ru_lock);
393 
394 	for (i = 0; i < reusep->ru_entries; ++i) {
395 		if (reusep->ru_conns[i] == connp)
396 			break;
397 	}
398 	ASSERT(i < reusep->ru_entries);
399 
400 	/* move last entry into freed slot */
401 	if (--reusep->ru_entries == 0) {
402 		/* last entry, free list */
403 		mutex_exit(&reusep->ru_lock);
404 		mutex_destroy(&reusep->ru_lock);
405 		kmem_free(reusep, sizeof (*reusep));
406 		connp->conn_reuselist = NULL;
407 	} else {
408 		reusep->ru_conns[i] = reusep->ru_conns[reusep->ru_entries];
409 
410 		/*
411 		 * reset round-robin pointer, so it doesn't accidentally point
412 		 * to the last entry
413 		 */
414 		reusep->ru_next = 0;
415 		mutex_exit(&reusep->ru_lock);
416 	}
417 }
418 
419 /*
420  * Return the next anonymous port in the privileged port range for
421  * bind checking.
422  *
423  * Trusted Extension (TX) notes: TX allows administrator to mark or
424  * reserve ports as Multilevel ports (MLP). MLP has special function
425  * on TX systems. Once a port is made MLP, it's not available as
426  * ordinary port. This creates "holes" in the port name space. It
427  * may be necessary to skip the "holes" find a suitable anon port.
428  */
429 static in_port_t
430 udp_get_next_priv_port(udp_t *udp)
431 {
432 	static in_port_t next_priv_port = IPPORT_RESERVED - 1;
433 	in_port_t nextport;
434 	boolean_t restart = B_FALSE;
435 	udp_stack_t *us = udp->udp_us;
436 
437 retry:
438 	if (next_priv_port < us->us_min_anonpriv_port ||
439 	    next_priv_port >= IPPORT_RESERVED) {
440 		next_priv_port = IPPORT_RESERVED - 1;
441 		if (restart)
442 			return (0);
443 		restart = B_TRUE;
444 	}
445 
446 	if (is_system_labeled() &&
447 	    (nextport = tsol_next_port(crgetzone(udp->udp_connp->conn_cred),
448 	    next_priv_port, IPPROTO_UDP, B_FALSE)) != 0) {
449 		next_priv_port = nextport;
450 		goto retry;
451 	}
452 
453 	return (next_priv_port--);
454 }
455 
456 /*
457  * Hash list removal routine for udp_t structures.
458  */
459 static void
460 udp_bind_hash_remove(udp_t *udp, boolean_t caller_holds_lock)
461 {
462 	udp_t		*udpnext;
463 	kmutex_t	*lockp;
464 	udp_stack_t	*us = udp->udp_us;
465 	conn_t		*connp = udp->udp_connp;
466 
467 	if (udp->udp_ptpbhn == NULL)
468 		return;
469 
470 	/*
471 	 * Extract the lock pointer in case there are concurrent
472 	 * hash_remove's for this instance.
473 	 */
474 	ASSERT(connp->conn_lport != 0);
475 	if (!caller_holds_lock) {
476 		lockp = &us->us_bind_fanout[UDP_BIND_HASH(connp->conn_lport,
477 		    us->us_bind_fanout_size)].uf_lock;
478 		ASSERT(lockp != NULL);
479 		mutex_enter(lockp);
480 	}
481 	if (udp->udp_ptpbhn != NULL) {
482 		udpnext = udp->udp_bind_hash;
483 		if (udpnext != NULL) {
484 			udpnext->udp_ptpbhn = udp->udp_ptpbhn;
485 			udp->udp_bind_hash = NULL;
486 		}
487 		*udp->udp_ptpbhn = udpnext;
488 		udp->udp_ptpbhn = NULL;
489 	}
490 	udp_reuselist_remove(connp);
491 	if (!caller_holds_lock) {
492 		mutex_exit(lockp);
493 	}
494 }
495 
496 static void
497 udp_bind_hash_insert(udp_fanout_t *uf, udp_t *udp)
498 {
499 	conn_t	*connp = udp->udp_connp;
500 	udp_t	**udpp;
501 	udp_t	*udpnext;
502 	conn_t	*connext;
503 
504 	ASSERT(MUTEX_HELD(&uf->uf_lock));
505 	ASSERT(udp->udp_ptpbhn == NULL);
506 	udpp = &uf->uf_udp;
507 	udpnext = udpp[0];
508 	if (udpnext != NULL) {
509 		/*
510 		 * If the new udp bound to the INADDR_ANY address
511 		 * and the first one in the list is not bound to
512 		 * INADDR_ANY we skip all entries until we find the
513 		 * first one bound to INADDR_ANY.
514 		 * This makes sure that applications binding to a
515 		 * specific address get preference over those binding to
516 		 * INADDR_ANY.
517 		 */
518 		connext = udpnext->udp_connp;
519 		if (V6_OR_V4_INADDR_ANY(connp->conn_bound_addr_v6) &&
520 		    !V6_OR_V4_INADDR_ANY(connext->conn_bound_addr_v6)) {
521 			while ((udpnext = udpp[0]) != NULL &&
522 			    !V6_OR_V4_INADDR_ANY(connext->conn_bound_addr_v6)) {
523 				udpp = &(udpnext->udp_bind_hash);
524 			}
525 			if (udpnext != NULL)
526 				udpnext->udp_ptpbhn = &udp->udp_bind_hash;
527 		} else {
528 			udpnext->udp_ptpbhn = &udp->udp_bind_hash;
529 		}
530 	}
531 	udp->udp_bind_hash = udpnext;
532 	udp->udp_ptpbhn = udpp;
533 	udpp[0] = udp;
534 }
535 
536 /*
537  * This routine is called to handle each O_T_BIND_REQ/T_BIND_REQ message
538  * passed to udp_wput.
539  * It associates a port number and local address with the stream.
540  * It calls IP to verify the local IP address, and calls IP to insert
541  * the conn_t in the fanout table.
542  * If everything is ok it then sends the T_BIND_ACK back up.
543  *
544  * Note that UDP over IPv4 and IPv6 sockets can use the same port number
545  * without setting SO_REUSEADDR. This is needed so that they
546  * can be viewed as two independent transport protocols.
547  * However, anonymouns ports are allocated from the same range to avoid
548  * duplicating the us->us_next_port_to_try.
549  */
550 static void
551 udp_tpi_bind(queue_t *q, mblk_t *mp)
552 {
553 	sin_t		*sin;
554 	sin6_t		*sin6;
555 	mblk_t		*mp1;
556 	struct T_bind_req *tbr;
557 	conn_t		*connp;
558 	udp_t		*udp;
559 	int		error;
560 	struct sockaddr	*sa;
561 	cred_t		*cr;
562 
563 	/*
564 	 * All Solaris components should pass a db_credp
565 	 * for this TPI message, hence we ASSERT.
566 	 * But in case there is some other M_PROTO that looks
567 	 * like a TPI message sent by some other kernel
568 	 * component, we check and return an error.
569 	 */
570 	cr = msg_getcred(mp, NULL);
571 	ASSERT(cr != NULL);
572 	if (cr == NULL) {
573 		udp_err_ack(q, mp, TSYSERR, EINVAL);
574 		return;
575 	}
576 
577 	connp = Q_TO_CONN(q);
578 	udp = connp->conn_udp;
579 	if ((mp->b_wptr - mp->b_rptr) < sizeof (*tbr)) {
580 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
581 		    "udp_bind: bad req, len %u",
582 		    (uint_t)(mp->b_wptr - mp->b_rptr));
583 		udp_err_ack(q, mp, TPROTO, 0);
584 		return;
585 	}
586 	if (udp->udp_state != TS_UNBND) {
587 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
588 		    "udp_bind: bad state, %u", udp->udp_state);
589 		udp_err_ack(q, mp, TOUTSTATE, 0);
590 		return;
591 	}
592 	/*
593 	 * Reallocate the message to make sure we have enough room for an
594 	 * address.
595 	 */
596 	mp1 = reallocb(mp, sizeof (struct T_bind_ack) + sizeof (sin6_t), 1);
597 	if (mp1 == NULL) {
598 		udp_err_ack(q, mp, TSYSERR, ENOMEM);
599 		return;
600 	}
601 
602 	mp = mp1;
603 
604 	/* Reset the message type in preparation for shipping it back. */
605 	DB_TYPE(mp) = M_PCPROTO;
606 
607 	tbr = (struct T_bind_req *)mp->b_rptr;
608 	switch (tbr->ADDR_length) {
609 	case 0:			/* Request for a generic port */
610 		tbr->ADDR_offset = sizeof (struct T_bind_req);
611 		if (connp->conn_family == AF_INET) {
612 			tbr->ADDR_length = sizeof (sin_t);
613 			sin = (sin_t *)&tbr[1];
614 			*sin = sin_null;
615 			sin->sin_family = AF_INET;
616 			mp->b_wptr = (uchar_t *)&sin[1];
617 			sa = (struct sockaddr *)sin;
618 		} else {
619 			ASSERT(connp->conn_family == AF_INET6);
620 			tbr->ADDR_length = sizeof (sin6_t);
621 			sin6 = (sin6_t *)&tbr[1];
622 			*sin6 = sin6_null;
623 			sin6->sin6_family = AF_INET6;
624 			mp->b_wptr = (uchar_t *)&sin6[1];
625 			sa = (struct sockaddr *)sin6;
626 		}
627 		break;
628 
629 	case sizeof (sin_t):	/* Complete IPv4 address */
630 		sa = (struct sockaddr *)mi_offset_param(mp, tbr->ADDR_offset,
631 		    sizeof (sin_t));
632 		if (sa == NULL || !OK_32PTR((char *)sa)) {
633 			udp_err_ack(q, mp, TSYSERR, EINVAL);
634 			return;
635 		}
636 		if (connp->conn_family != AF_INET ||
637 		    sa->sa_family != AF_INET) {
638 			udp_err_ack(q, mp, TSYSERR, EAFNOSUPPORT);
639 			return;
640 		}
641 		break;
642 
643 	case sizeof (sin6_t):	/* complete IPv6 address */
644 		sa = (struct sockaddr *)mi_offset_param(mp, tbr->ADDR_offset,
645 		    sizeof (sin6_t));
646 		if (sa == NULL || !OK_32PTR((char *)sa)) {
647 			udp_err_ack(q, mp, TSYSERR, EINVAL);
648 			return;
649 		}
650 		if (connp->conn_family != AF_INET6 ||
651 		    sa->sa_family != AF_INET6) {
652 			udp_err_ack(q, mp, TSYSERR, EAFNOSUPPORT);
653 			return;
654 		}
655 		break;
656 
657 	default:		/* Invalid request */
658 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
659 		    "udp_bind: bad ADDR_length length %u", tbr->ADDR_length);
660 		udp_err_ack(q, mp, TBADADDR, 0);
661 		return;
662 	}
663 
664 	error = udp_do_bind(connp, sa, tbr->ADDR_length, cr,
665 	    tbr->PRIM_type != O_T_BIND_REQ);
666 
667 	if (error != 0) {
668 		if (error > 0) {
669 			udp_err_ack(q, mp, TSYSERR, error);
670 		} else {
671 			udp_err_ack(q, mp, -error, 0);
672 		}
673 	} else {
674 		tbr->PRIM_type = T_BIND_ACK;
675 		qreply(q, mp);
676 	}
677 }
678 
679 /*
680  * This routine handles each T_CONN_REQ message passed to udp.  It
681  * associates a default destination address with the stream.
682  *
683  * After various error checks are completed, udp_connect() lays
684  * the target address and port into the composite header template.
685  * Then we ask IP for information, including a source address if we didn't
686  * already have one. Finally we send up the T_OK_ACK reply message.
687  */
688 static void
689 udp_tpi_connect(queue_t *q, mblk_t *mp)
690 {
691 	conn_t	*connp = Q_TO_CONN(q);
692 	int	error;
693 	socklen_t	len;
694 	struct sockaddr		*sa;
695 	struct T_conn_req	*tcr;
696 	cred_t		*cr;
697 	pid_t		pid;
698 	/*
699 	 * All Solaris components should pass a db_credp
700 	 * for this TPI message, hence we ASSERT.
701 	 * But in case there is some other M_PROTO that looks
702 	 * like a TPI message sent by some other kernel
703 	 * component, we check and return an error.
704 	 */
705 	cr = msg_getcred(mp, &pid);
706 	ASSERT(cr != NULL);
707 	if (cr == NULL) {
708 		udp_err_ack(q, mp, TSYSERR, EINVAL);
709 		return;
710 	}
711 
712 	tcr = (struct T_conn_req *)mp->b_rptr;
713 
714 	/* A bit of sanity checking */
715 	if ((mp->b_wptr - mp->b_rptr) < sizeof (struct T_conn_req)) {
716 		udp_err_ack(q, mp, TPROTO, 0);
717 		return;
718 	}
719 
720 	if (tcr->OPT_length != 0) {
721 		udp_err_ack(q, mp, TBADOPT, 0);
722 		return;
723 	}
724 
725 	/*
726 	 * Determine packet type based on type of address passed in
727 	 * the request should contain an IPv4 or IPv6 address.
728 	 * Make sure that address family matches the type of
729 	 * family of the address passed down.
730 	 */
731 	len = tcr->DEST_length;
732 	switch (tcr->DEST_length) {
733 	default:
734 		udp_err_ack(q, mp, TBADADDR, 0);
735 		return;
736 
737 	case sizeof (sin_t):
738 		sa = (struct sockaddr *)mi_offset_param(mp, tcr->DEST_offset,
739 		    sizeof (sin_t));
740 		break;
741 
742 	case sizeof (sin6_t):
743 		sa = (struct sockaddr *)mi_offset_param(mp, tcr->DEST_offset,
744 		    sizeof (sin6_t));
745 		break;
746 	}
747 
748 	error = proto_verify_ip_addr(connp->conn_family, sa, len);
749 	if (error != 0) {
750 		udp_err_ack(q, mp, TSYSERR, error);
751 		return;
752 	}
753 
754 	error = udp_do_connect(connp, sa, len, cr, pid);
755 	if (error != 0) {
756 		if (error < 0)
757 			udp_err_ack(q, mp, -error, 0);
758 		else
759 			udp_err_ack(q, mp, TSYSERR, error);
760 	} else {
761 		mblk_t	*mp1;
762 		/*
763 		 * We have to send a connection confirmation to
764 		 * keep TLI happy.
765 		 */
766 		if (connp->conn_family == AF_INET) {
767 			mp1 = mi_tpi_conn_con(NULL, (char *)sa,
768 			    sizeof (sin_t), NULL, 0);
769 		} else {
770 			mp1 = mi_tpi_conn_con(NULL, (char *)sa,
771 			    sizeof (sin6_t), NULL, 0);
772 		}
773 		if (mp1 == NULL) {
774 			udp_err_ack(q, mp, TSYSERR, ENOMEM);
775 			return;
776 		}
777 
778 		/*
779 		 * Send ok_ack for T_CONN_REQ
780 		 */
781 		mp = mi_tpi_ok_ack_alloc(mp);
782 		if (mp == NULL) {
783 			/* Unable to reuse the T_CONN_REQ for the ack. */
784 			udp_err_ack_prim(q, mp1, T_CONN_REQ, TSYSERR, ENOMEM);
785 			return;
786 		}
787 
788 		putnext(connp->conn_rq, mp);
789 		putnext(connp->conn_rq, mp1);
790 	}
791 }
792 
793 static int
794 udp_tpi_close(queue_t *q, int flags)
795 {
796 	conn_t	*connp;
797 
798 	if (flags & SO_FALLBACK) {
799 		/*
800 		 * stream is being closed while in fallback
801 		 * simply free the resources that were allocated
802 		 */
803 		inet_minor_free(WR(q)->q_ptr, (dev_t)(RD(q)->q_ptr));
804 		qprocsoff(q);
805 		goto done;
806 	}
807 
808 	connp = Q_TO_CONN(q);
809 	udp_do_close(connp);
810 done:
811 	q->q_ptr = WR(q)->q_ptr = NULL;
812 	return (0);
813 }
814 
815 static void
816 udp_close_free(conn_t *connp)
817 {
818 	udp_t *udp = connp->conn_udp;
819 
820 	/* If there are any options associated with the stream, free them. */
821 	if (udp->udp_recv_ipp.ipp_fields != 0)
822 		ip_pkt_free(&udp->udp_recv_ipp);
823 
824 	/*
825 	 * Clear any fields which the kmem_cache constructor clears.
826 	 * Only udp_connp needs to be preserved.
827 	 * TBD: We should make this more efficient to avoid clearing
828 	 * everything.
829 	 */
830 	ASSERT(udp->udp_connp == connp);
831 	bzero(udp, sizeof (udp_t));
832 	udp->udp_connp = connp;
833 }
834 
835 static int
836 udp_do_disconnect(conn_t *connp)
837 {
838 	udp_t	*udp;
839 	udp_fanout_t *udpf;
840 	udp_stack_t *us;
841 	int	error;
842 
843 	udp = connp->conn_udp;
844 	us = udp->udp_us;
845 	mutex_enter(&connp->conn_lock);
846 	if (udp->udp_state != TS_DATA_XFER) {
847 		mutex_exit(&connp->conn_lock);
848 		return (-TOUTSTATE);
849 	}
850 	udpf = &us->us_bind_fanout[UDP_BIND_HASH(connp->conn_lport,
851 	    us->us_bind_fanout_size)];
852 	mutex_enter(&udpf->uf_lock);
853 	if (connp->conn_mcbc_bind)
854 		connp->conn_saddr_v6 = ipv6_all_zeros;
855 	else
856 		connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
857 	connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
858 	connp->conn_faddr_v6 = ipv6_all_zeros;
859 	connp->conn_fport = 0;
860 	udp->udp_state = TS_IDLE;
861 	mutex_exit(&udpf->uf_lock);
862 
863 	/* Remove any remnants of mapped address binding */
864 	if (connp->conn_family == AF_INET6)
865 		connp->conn_ipversion = IPV6_VERSION;
866 
867 	connp->conn_v6lastdst = ipv6_all_zeros;
868 	error = udp_build_hdr_template(connp, &connp->conn_saddr_v6,
869 	    &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
870 	mutex_exit(&connp->conn_lock);
871 	if (error != 0)
872 		return (error);
873 
874 	/*
875 	 * Tell IP to remove the full binding and revert
876 	 * to the local address binding.
877 	 */
878 	return (ip_laddr_fanout_insert(connp));
879 }
880 
881 static void
882 udp_tpi_disconnect(queue_t *q, mblk_t *mp)
883 {
884 	conn_t	*connp = Q_TO_CONN(q);
885 	int	error;
886 
887 	/*
888 	 * Allocate the largest primitive we need to send back
889 	 * T_error_ack is > than T_ok_ack
890 	 */
891 	mp = reallocb(mp, sizeof (struct T_error_ack), 1);
892 	if (mp == NULL) {
893 		/* Unable to reuse the T_DISCON_REQ for the ack. */
894 		udp_err_ack_prim(q, mp, T_DISCON_REQ, TSYSERR, ENOMEM);
895 		return;
896 	}
897 
898 	error = udp_do_disconnect(connp);
899 
900 	if (error != 0) {
901 		if (error < 0) {
902 			udp_err_ack(q, mp, -error, 0);
903 		} else {
904 			udp_err_ack(q, mp, TSYSERR, error);
905 		}
906 	} else {
907 		mp = mi_tpi_ok_ack_alloc(mp);
908 		ASSERT(mp != NULL);
909 		qreply(q, mp);
910 	}
911 }
912 
913 int
914 udp_disconnect(conn_t *connp)
915 {
916 	int error;
917 
918 	connp->conn_dgram_errind = B_FALSE;
919 	error = udp_do_disconnect(connp);
920 	if (error < 0)
921 		error = proto_tlitosyserr(-error);
922 
923 	return (error);
924 }
925 
926 /* This routine creates a T_ERROR_ACK message and passes it upstream. */
927 static void
928 udp_err_ack(queue_t *q, mblk_t *mp, t_scalar_t t_error, int sys_error)
929 {
930 	if ((mp = mi_tpi_err_ack_alloc(mp, t_error, sys_error)) != NULL)
931 		qreply(q, mp);
932 }
933 
934 /* Shorthand to generate and send TPI error acks to our client */
935 static void
936 udp_err_ack_prim(queue_t *q, mblk_t *mp, t_scalar_t primitive,
937     t_scalar_t t_error, int sys_error)
938 {
939 	struct T_error_ack	*teackp;
940 
941 	if ((mp = tpi_ack_alloc(mp, sizeof (struct T_error_ack),
942 	    M_PCPROTO, T_ERROR_ACK)) != NULL) {
943 		teackp = (struct T_error_ack *)mp->b_rptr;
944 		teackp->ERROR_prim = primitive;
945 		teackp->TLI_error = t_error;
946 		teackp->UNIX_error = sys_error;
947 		qreply(q, mp);
948 	}
949 }
950 
951 /* At minimum we need 4 bytes of UDP header */
952 #define	ICMP_MIN_UDP_HDR	4
953 
954 /*
955  * udp_icmp_input is called as conn_recvicmp to process ICMP messages.
956  * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
957  * Assumes that IP has pulled up everything up to and including the ICMP header.
958  */
959 /* ARGSUSED2 */
960 static void
961 udp_icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
962 {
963 	conn_t		*connp = (conn_t *)arg1;
964 	icmph_t		*icmph;
965 	ipha_t		*ipha;
966 	int		iph_hdr_length;
967 	udpha_t		*udpha;
968 	sin_t		sin;
969 	sin6_t		sin6;
970 	mblk_t		*mp1;
971 	int		error = 0;
972 	udp_t		*udp = connp->conn_udp;
973 
974 	ipha = (ipha_t *)mp->b_rptr;
975 
976 	ASSERT(OK_32PTR(mp->b_rptr));
977 
978 	if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
979 		ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
980 		udp_icmp_error_ipv6(connp, mp, ira);
981 		return;
982 	}
983 	ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
984 
985 	/* Skip past the outer IP and ICMP headers */
986 	ASSERT(IPH_HDR_LENGTH(ipha) == ira->ira_ip_hdr_length);
987 	iph_hdr_length = ira->ira_ip_hdr_length;
988 	icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
989 	ipha = (ipha_t *)&icmph[1];	/* Inner IP header */
990 
991 	/* Skip past the inner IP and find the ULP header */
992 	iph_hdr_length = IPH_HDR_LENGTH(ipha);
993 	udpha = (udpha_t *)((char *)ipha + iph_hdr_length);
994 
995 	switch (icmph->icmph_type) {
996 	case ICMP_DEST_UNREACHABLE:
997 		switch (icmph->icmph_code) {
998 		case ICMP_FRAGMENTATION_NEEDED: {
999 			ipha_t		*ipha;
1000 			ip_xmit_attr_t	*ixa;
1001 			/*
1002 			 * IP has already adjusted the path MTU.
1003 			 * But we need to adjust DF for IPv4.
1004 			 */
1005 			if (connp->conn_ipversion != IPV4_VERSION)
1006 				break;
1007 
1008 			ixa = conn_get_ixa(connp, B_FALSE);
1009 			if (ixa == NULL || ixa->ixa_ire == NULL) {
1010 				/*
1011 				 * Some other thread holds conn_ixa. We will
1012 				 * redo this on the next ICMP too big.
1013 				 */
1014 				if (ixa != NULL)
1015 					ixa_refrele(ixa);
1016 				break;
1017 			}
1018 			(void) ip_get_pmtu(ixa);
1019 
1020 			mutex_enter(&connp->conn_lock);
1021 			ipha = (ipha_t *)connp->conn_ht_iphc;
1022 			if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
1023 				ipha->ipha_fragment_offset_and_flags |=
1024 				    IPH_DF_HTONS;
1025 			} else {
1026 				ipha->ipha_fragment_offset_and_flags &=
1027 				    ~IPH_DF_HTONS;
1028 			}
1029 			mutex_exit(&connp->conn_lock);
1030 			ixa_refrele(ixa);
1031 			break;
1032 		}
1033 		case ICMP_PORT_UNREACHABLE:
1034 		case ICMP_PROTOCOL_UNREACHABLE:
1035 			error = ECONNREFUSED;
1036 			break;
1037 		default:
1038 			/* Transient errors */
1039 			break;
1040 		}
1041 		break;
1042 	default:
1043 		/* Transient errors */
1044 		break;
1045 	}
1046 	if (error == 0) {
1047 		freemsg(mp);
1048 		return;
1049 	}
1050 
1051 	/*
1052 	 * Deliver T_UDERROR_IND when the application has asked for it.
1053 	 * The socket layer enables this automatically when connected.
1054 	 */
1055 	if (!connp->conn_dgram_errind) {
1056 		freemsg(mp);
1057 		return;
1058 	}
1059 
1060 	switch (connp->conn_family) {
1061 	case AF_INET:
1062 		sin = sin_null;
1063 		sin.sin_family = AF_INET;
1064 		sin.sin_addr.s_addr = ipha->ipha_dst;
1065 		sin.sin_port = udpha->uha_dst_port;
1066 		if (IPCL_IS_NONSTR(connp)) {
1067 			mutex_enter(&connp->conn_lock);
1068 			if (udp->udp_state == TS_DATA_XFER) {
1069 				if (sin.sin_port == connp->conn_fport &&
1070 				    sin.sin_addr.s_addr ==
1071 				    connp->conn_faddr_v4) {
1072 					mutex_exit(&connp->conn_lock);
1073 					(*connp->conn_upcalls->su_set_error)
1074 					    (connp->conn_upper_handle, error);
1075 					goto done;
1076 				}
1077 			} else {
1078 				udp->udp_delayed_error = error;
1079 				*((sin_t *)&udp->udp_delayed_addr) = sin;
1080 			}
1081 			mutex_exit(&connp->conn_lock);
1082 		} else {
1083 			mp1 = mi_tpi_uderror_ind((char *)&sin, sizeof (sin_t),
1084 			    NULL, 0, error);
1085 			if (mp1 != NULL)
1086 				putnext(connp->conn_rq, mp1);
1087 		}
1088 		break;
1089 	case AF_INET6:
1090 		sin6 = sin6_null;
1091 		sin6.sin6_family = AF_INET6;
1092 		IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &sin6.sin6_addr);
1093 		sin6.sin6_port = udpha->uha_dst_port;
1094 		if (IPCL_IS_NONSTR(connp)) {
1095 			mutex_enter(&connp->conn_lock);
1096 			if (udp->udp_state == TS_DATA_XFER) {
1097 				if (sin6.sin6_port == connp->conn_fport &&
1098 				    IN6_ARE_ADDR_EQUAL(&sin6.sin6_addr,
1099 				    &connp->conn_faddr_v6)) {
1100 					mutex_exit(&connp->conn_lock);
1101 					(*connp->conn_upcalls->su_set_error)
1102 					    (connp->conn_upper_handle, error);
1103 					goto done;
1104 				}
1105 			} else {
1106 				udp->udp_delayed_error = error;
1107 				*((sin6_t *)&udp->udp_delayed_addr) = sin6;
1108 			}
1109 			mutex_exit(&connp->conn_lock);
1110 		} else {
1111 			mp1 = mi_tpi_uderror_ind((char *)&sin6, sizeof (sin6_t),
1112 			    NULL, 0, error);
1113 			if (mp1 != NULL)
1114 				putnext(connp->conn_rq, mp1);
1115 		}
1116 		break;
1117 	}
1118 done:
1119 	freemsg(mp);
1120 }
1121 
1122 /*
1123  * udp_icmp_error_ipv6 is called by udp_icmp_error to process ICMP for IPv6.
1124  * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
1125  * Assumes that IP has pulled up all the extension headers as well as the
1126  * ICMPv6 header.
1127  */
1128 static void
1129 udp_icmp_error_ipv6(conn_t *connp, mblk_t *mp, ip_recv_attr_t *ira)
1130 {
1131 	icmp6_t		*icmp6;
1132 	ip6_t		*ip6h, *outer_ip6h;
1133 	uint16_t	iph_hdr_length;
1134 	uint8_t		*nexthdrp;
1135 	udpha_t		*udpha;
1136 	sin6_t		sin6;
1137 	mblk_t		*mp1;
1138 	int		error = 0;
1139 	udp_t		*udp = connp->conn_udp;
1140 	udp_stack_t	*us = udp->udp_us;
1141 
1142 	outer_ip6h = (ip6_t *)mp->b_rptr;
1143 #ifdef DEBUG
1144 	if (outer_ip6h->ip6_nxt != IPPROTO_ICMPV6)
1145 		iph_hdr_length = ip_hdr_length_v6(mp, outer_ip6h);
1146 	else
1147 		iph_hdr_length = IPV6_HDR_LEN;
1148 	ASSERT(iph_hdr_length == ira->ira_ip_hdr_length);
1149 #endif
1150 	/* Skip past the outer IP and ICMP headers */
1151 	iph_hdr_length = ira->ira_ip_hdr_length;
1152 	icmp6 = (icmp6_t *)&mp->b_rptr[iph_hdr_length];
1153 
1154 	/* Skip past the inner IP and find the ULP header */
1155 	ip6h = (ip6_t *)&icmp6[1];	/* Inner IP header */
1156 	if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length, &nexthdrp)) {
1157 		freemsg(mp);
1158 		return;
1159 	}
1160 	udpha = (udpha_t *)((char *)ip6h + iph_hdr_length);
1161 
1162 	switch (icmp6->icmp6_type) {
1163 	case ICMP6_DST_UNREACH:
1164 		switch (icmp6->icmp6_code) {
1165 		case ICMP6_DST_UNREACH_NOPORT:
1166 			error = ECONNREFUSED;
1167 			break;
1168 		case ICMP6_DST_UNREACH_ADMIN:
1169 		case ICMP6_DST_UNREACH_NOROUTE:
1170 		case ICMP6_DST_UNREACH_BEYONDSCOPE:
1171 		case ICMP6_DST_UNREACH_ADDR:
1172 			/* Transient errors */
1173 			break;
1174 		default:
1175 			break;
1176 		}
1177 		break;
1178 	case ICMP6_PACKET_TOO_BIG: {
1179 		struct T_unitdata_ind	*tudi;
1180 		struct T_opthdr		*toh;
1181 		size_t			udi_size;
1182 		mblk_t			*newmp;
1183 		t_scalar_t		opt_length = sizeof (struct T_opthdr) +
1184 		    sizeof (struct ip6_mtuinfo);
1185 		sin6_t			*sin6;
1186 		struct ip6_mtuinfo	*mtuinfo;
1187 
1188 		/*
1189 		 * If the application has requested to receive path mtu
1190 		 * information, send up an empty message containing an
1191 		 * IPV6_PATHMTU ancillary data item.
1192 		 */
1193 		if (!connp->conn_ipv6_recvpathmtu)
1194 			break;
1195 
1196 		udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t) +
1197 		    opt_length;
1198 		if ((newmp = allocb(udi_size, BPRI_MED)) == NULL) {
1199 			UDPS_BUMP_MIB(us, udpInErrors);
1200 			break;
1201 		}
1202 
1203 		/*
1204 		 * newmp->b_cont is left to NULL on purpose.  This is an
1205 		 * empty message containing only ancillary data.
1206 		 */
1207 		newmp->b_datap->db_type = M_PROTO;
1208 		tudi = (struct T_unitdata_ind *)newmp->b_rptr;
1209 		newmp->b_wptr = (uchar_t *)tudi + udi_size;
1210 		tudi->PRIM_type = T_UNITDATA_IND;
1211 		tudi->SRC_length = sizeof (sin6_t);
1212 		tudi->SRC_offset = sizeof (struct T_unitdata_ind);
1213 		tudi->OPT_offset = tudi->SRC_offset + sizeof (sin6_t);
1214 		tudi->OPT_length = opt_length;
1215 
1216 		sin6 = (sin6_t *)&tudi[1];
1217 		bzero(sin6, sizeof (sin6_t));
1218 		sin6->sin6_family = AF_INET6;
1219 		sin6->sin6_addr = connp->conn_faddr_v6;
1220 
1221 		toh = (struct T_opthdr *)&sin6[1];
1222 		toh->level = IPPROTO_IPV6;
1223 		toh->name = IPV6_PATHMTU;
1224 		toh->len = opt_length;
1225 		toh->status = 0;
1226 
1227 		mtuinfo = (struct ip6_mtuinfo *)&toh[1];
1228 		bzero(mtuinfo, sizeof (struct ip6_mtuinfo));
1229 		mtuinfo->ip6m_addr.sin6_family = AF_INET6;
1230 		mtuinfo->ip6m_addr.sin6_addr = ip6h->ip6_dst;
1231 		mtuinfo->ip6m_mtu = icmp6->icmp6_mtu;
1232 		/*
1233 		 * We've consumed everything we need from the original
1234 		 * message.  Free it, then send our empty message.
1235 		 */
1236 		freemsg(mp);
1237 		udp_ulp_recv(connp, newmp, msgdsize(newmp), ira);
1238 		return;
1239 	}
1240 	case ICMP6_TIME_EXCEEDED:
1241 		/* Transient errors */
1242 		break;
1243 	case ICMP6_PARAM_PROB:
1244 		/* If this corresponds to an ICMP_PROTOCOL_UNREACHABLE */
1245 		if (icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER &&
1246 		    (uchar_t *)ip6h + icmp6->icmp6_pptr ==
1247 		    (uchar_t *)nexthdrp) {
1248 			error = ECONNREFUSED;
1249 			break;
1250 		}
1251 		break;
1252 	}
1253 	if (error == 0) {
1254 		freemsg(mp);
1255 		return;
1256 	}
1257 
1258 	/*
1259 	 * Deliver T_UDERROR_IND when the application has asked for it.
1260 	 * The socket layer enables this automatically when connected.
1261 	 */
1262 	if (!connp->conn_dgram_errind) {
1263 		freemsg(mp);
1264 		return;
1265 	}
1266 
1267 	sin6 = sin6_null;
1268 	sin6.sin6_family = AF_INET6;
1269 	sin6.sin6_addr = ip6h->ip6_dst;
1270 	sin6.sin6_port = udpha->uha_dst_port;
1271 	sin6.sin6_flowinfo = ip6h->ip6_vcf & ~IPV6_VERS_AND_FLOW_MASK;
1272 
1273 	if (IPCL_IS_NONSTR(connp)) {
1274 		mutex_enter(&connp->conn_lock);
1275 		if (udp->udp_state == TS_DATA_XFER) {
1276 			if (sin6.sin6_port == connp->conn_fport &&
1277 			    IN6_ARE_ADDR_EQUAL(&sin6.sin6_addr,
1278 			    &connp->conn_faddr_v6)) {
1279 				mutex_exit(&connp->conn_lock);
1280 				(*connp->conn_upcalls->su_set_error)
1281 				    (connp->conn_upper_handle, error);
1282 				goto done;
1283 			}
1284 		} else {
1285 			udp->udp_delayed_error = error;
1286 			*((sin6_t *)&udp->udp_delayed_addr) = sin6;
1287 		}
1288 		mutex_exit(&connp->conn_lock);
1289 	} else {
1290 		mp1 = mi_tpi_uderror_ind((char *)&sin6, sizeof (sin6_t),
1291 		    NULL, 0, error);
1292 		if (mp1 != NULL)
1293 			putnext(connp->conn_rq, mp1);
1294 	}
1295 done:
1296 	freemsg(mp);
1297 }
1298 
1299 /*
1300  * This routine responds to T_ADDR_REQ messages.  It is called by udp_wput.
1301  * The local address is filled in if endpoint is bound. The remote address
1302  * is filled in if remote address has been precified ("connected endpoint")
1303  * (The concept of connected CLTS sockets is alien to published TPI
1304  *  but we support it anyway).
1305  */
1306 static void
1307 udp_addr_req(queue_t *q, mblk_t *mp)
1308 {
1309 	struct sockaddr *sa;
1310 	mblk_t	*ackmp;
1311 	struct T_addr_ack *taa;
1312 	udp_t	*udp = Q_TO_UDP(q);
1313 	conn_t	*connp = udp->udp_connp;
1314 	uint_t	addrlen;
1315 
1316 	/* Make it large enough for worst case */
1317 	ackmp = reallocb(mp, sizeof (struct T_addr_ack) +
1318 	    2 * sizeof (sin6_t), 1);
1319 	if (ackmp == NULL) {
1320 		udp_err_ack(q, mp, TSYSERR, ENOMEM);
1321 		return;
1322 	}
1323 	taa = (struct T_addr_ack *)ackmp->b_rptr;
1324 
1325 	bzero(taa, sizeof (struct T_addr_ack));
1326 	ackmp->b_wptr = (uchar_t *)&taa[1];
1327 
1328 	taa->PRIM_type = T_ADDR_ACK;
1329 	ackmp->b_datap->db_type = M_PCPROTO;
1330 
1331 	if (connp->conn_family == AF_INET)
1332 		addrlen = sizeof (sin_t);
1333 	else
1334 		addrlen = sizeof (sin6_t);
1335 
1336 	mutex_enter(&connp->conn_lock);
1337 	/*
1338 	 * Note: Following code assumes 32 bit alignment of basic
1339 	 * data structures like sin_t and struct T_addr_ack.
1340 	 */
1341 	if (udp->udp_state != TS_UNBND) {
1342 		/*
1343 		 * Fill in local address first
1344 		 */
1345 		taa->LOCADDR_offset = sizeof (*taa);
1346 		taa->LOCADDR_length = addrlen;
1347 		sa = (struct sockaddr *)&taa[1];
1348 		(void) conn_getsockname(connp, sa, &addrlen);
1349 		ackmp->b_wptr += addrlen;
1350 	}
1351 	if (udp->udp_state == TS_DATA_XFER) {
1352 		/*
1353 		 * connected, fill remote address too
1354 		 */
1355 		taa->REMADDR_length = addrlen;
1356 		/* assumed 32-bit alignment */
1357 		taa->REMADDR_offset = taa->LOCADDR_offset + taa->LOCADDR_length;
1358 		sa = (struct sockaddr *)(ackmp->b_rptr + taa->REMADDR_offset);
1359 		(void) conn_getpeername(connp, sa, &addrlen);
1360 		ackmp->b_wptr += addrlen;
1361 	}
1362 	mutex_exit(&connp->conn_lock);
1363 	ASSERT(ackmp->b_wptr <= ackmp->b_datap->db_lim);
1364 	qreply(q, ackmp);
1365 }
1366 
1367 static void
1368 udp_copy_info(struct T_info_ack *tap, udp_t *udp)
1369 {
1370 	conn_t		*connp = udp->udp_connp;
1371 
1372 	if (connp->conn_family == AF_INET) {
1373 		*tap = udp_g_t_info_ack_ipv4;
1374 	} else {
1375 		*tap = udp_g_t_info_ack_ipv6;
1376 	}
1377 	tap->CURRENT_state = udp->udp_state;
1378 	tap->OPT_size = udp_max_optsize;
1379 }
1380 
1381 static void
1382 udp_do_capability_ack(udp_t *udp, struct T_capability_ack *tcap,
1383     t_uscalar_t cap_bits1)
1384 {
1385 	tcap->CAP_bits1 = 0;
1386 
1387 	if (cap_bits1 & TC1_INFO) {
1388 		udp_copy_info(&tcap->INFO_ack, udp);
1389 		tcap->CAP_bits1 |= TC1_INFO;
1390 	}
1391 }
1392 
1393 /*
1394  * This routine responds to T_CAPABILITY_REQ messages.  It is called by
1395  * udp_wput.  Much of the T_CAPABILITY_ACK information is copied from
1396  * udp_g_t_info_ack.  The current state of the stream is copied from
1397  * udp_state.
1398  */
1399 static void
1400 udp_capability_req(queue_t *q, mblk_t *mp)
1401 {
1402 	t_uscalar_t		cap_bits1;
1403 	struct T_capability_ack	*tcap;
1404 	udp_t	*udp = Q_TO_UDP(q);
1405 
1406 	cap_bits1 = ((struct T_capability_req *)mp->b_rptr)->CAP_bits1;
1407 
1408 	mp = tpi_ack_alloc(mp, sizeof (struct T_capability_ack),
1409 	    mp->b_datap->db_type, T_CAPABILITY_ACK);
1410 	if (!mp)
1411 		return;
1412 
1413 	tcap = (struct T_capability_ack *)mp->b_rptr;
1414 	udp_do_capability_ack(udp, tcap, cap_bits1);
1415 
1416 	qreply(q, mp);
1417 }
1418 
1419 /*
1420  * This routine responds to T_INFO_REQ messages.  It is called by udp_wput.
1421  * Most of the T_INFO_ACK information is copied from udp_g_t_info_ack.
1422  * The current state of the stream is copied from udp_state.
1423  */
1424 static void
1425 udp_info_req(queue_t *q, mblk_t *mp)
1426 {
1427 	udp_t *udp = Q_TO_UDP(q);
1428 
1429 	/* Create a T_INFO_ACK message. */
1430 	mp = tpi_ack_alloc(mp, sizeof (struct T_info_ack), M_PCPROTO,
1431 	    T_INFO_ACK);
1432 	if (!mp)
1433 		return;
1434 	udp_copy_info((struct T_info_ack *)mp->b_rptr, udp);
1435 	qreply(q, mp);
1436 }
1437 
1438 /* For /dev/udp aka AF_INET open */
1439 static int
1440 udp_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1441 {
1442 	return (udp_open(q, devp, flag, sflag, credp, B_FALSE));
1443 }
1444 
1445 /* For /dev/udp6 aka AF_INET6 open */
1446 static int
1447 udp_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1448 {
1449 	return (udp_open(q, devp, flag, sflag, credp, B_TRUE));
1450 }
1451 
1452 /*
1453  * This is the open routine for udp.  It allocates a udp_t structure for
1454  * the stream and, on the first open of the module, creates an ND table.
1455  */
1456 static int
1457 udp_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
1458     boolean_t isv6)
1459 {
1460 	udp_t		*udp;
1461 	conn_t		*connp;
1462 	dev_t		conn_dev;
1463 	vmem_t		*minor_arena;
1464 	int		err;
1465 
1466 	/* If the stream is already open, return immediately. */
1467 	if (q->q_ptr != NULL)
1468 		return (0);
1469 
1470 	if (sflag == MODOPEN)
1471 		return (EINVAL);
1472 
1473 	if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
1474 	    ((conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
1475 		minor_arena = ip_minor_arena_la;
1476 	} else {
1477 		/*
1478 		 * Either minor numbers in the large arena were exhausted
1479 		 * or a non socket application is doing the open.
1480 		 * Try to allocate from the small arena.
1481 		 */
1482 		if ((conn_dev = inet_minor_alloc(ip_minor_arena_sa)) == 0)
1483 			return (EBUSY);
1484 
1485 		minor_arena = ip_minor_arena_sa;
1486 	}
1487 
1488 	if (flag & SO_FALLBACK) {
1489 		/*
1490 		 * Non streams socket needs a stream to fallback to
1491 		 */
1492 		RD(q)->q_ptr = (void *)conn_dev;
1493 		WR(q)->q_qinfo = &udp_fallback_sock_winit;
1494 		WR(q)->q_ptr = (void *)minor_arena;
1495 		qprocson(q);
1496 		return (0);
1497 	}
1498 
1499 	connp = udp_do_open(credp, isv6, KM_SLEEP, &err);
1500 	if (connp == NULL) {
1501 		inet_minor_free(minor_arena, conn_dev);
1502 		return (err);
1503 	}
1504 	udp = connp->conn_udp;
1505 
1506 	*devp = makedevice(getemajor(*devp), (minor_t)conn_dev);
1507 	connp->conn_dev = conn_dev;
1508 	connp->conn_minor_arena = minor_arena;
1509 
1510 	/*
1511 	 * Initialize the udp_t structure for this stream.
1512 	 */
1513 	q->q_ptr = connp;
1514 	WR(q)->q_ptr = connp;
1515 	connp->conn_rq = q;
1516 	connp->conn_wq = WR(q);
1517 
1518 	/*
1519 	 * Since this conn_t/udp_t is not yet visible to anybody else we don't
1520 	 * need to lock anything.
1521 	 */
1522 	ASSERT(connp->conn_proto == IPPROTO_UDP);
1523 	ASSERT(connp->conn_udp == udp);
1524 	ASSERT(udp->udp_connp == connp);
1525 
1526 	if (flag & SO_SOCKSTR) {
1527 		udp->udp_issocket = B_TRUE;
1528 	}
1529 
1530 	WR(q)->q_hiwat = connp->conn_sndbuf;
1531 	WR(q)->q_lowat = connp->conn_sndlowat;
1532 
1533 	qprocson(q);
1534 
1535 	/* Set the Stream head write offset and high watermark. */
1536 	(void) proto_set_tx_wroff(q, connp, connp->conn_wroff);
1537 	(void) proto_set_rx_hiwat(q, connp,
1538 	    udp_set_rcv_hiwat(udp, connp->conn_rcvbuf));
1539 
1540 	mutex_enter(&connp->conn_lock);
1541 	connp->conn_state_flags &= ~CONN_INCIPIENT;
1542 	mutex_exit(&connp->conn_lock);
1543 	return (0);
1544 }
1545 
1546 /*
1547  * Which UDP options OK to set through T_UNITDATA_REQ...
1548  */
1549 /* ARGSUSED */
1550 static boolean_t
1551 udp_opt_allow_udr_set(t_scalar_t level, t_scalar_t name)
1552 {
1553 	return (B_TRUE);
1554 }
1555 
1556 /*
1557  * This routine gets default values of certain options whose default
1558  * values are maintained by protcol specific code
1559  */
1560 int
1561 udp_opt_default(queue_t *q, t_scalar_t level, t_scalar_t name, uchar_t *ptr)
1562 {
1563 	udp_t		*udp = Q_TO_UDP(q);
1564 	udp_stack_t *us = udp->udp_us;
1565 	int *i1 = (int *)ptr;
1566 
1567 	switch (level) {
1568 	case IPPROTO_IP:
1569 		switch (name) {
1570 		case IP_MULTICAST_TTL:
1571 			*ptr = (uchar_t)IP_DEFAULT_MULTICAST_TTL;
1572 			return (sizeof (uchar_t));
1573 		case IP_MULTICAST_LOOP:
1574 			*ptr = (uchar_t)IP_DEFAULT_MULTICAST_LOOP;
1575 			return (sizeof (uchar_t));
1576 		}
1577 		break;
1578 	case IPPROTO_IPV6:
1579 		switch (name) {
1580 		case IPV6_MULTICAST_HOPS:
1581 			*i1 = IP_DEFAULT_MULTICAST_TTL;
1582 			return (sizeof (int));
1583 		case IPV6_MULTICAST_LOOP:
1584 			*i1 = IP_DEFAULT_MULTICAST_LOOP;
1585 			return (sizeof (int));
1586 		case IPV6_UNICAST_HOPS:
1587 			*i1 = us->us_ipv6_hoplimit;
1588 			return (sizeof (int));
1589 		}
1590 		break;
1591 	}
1592 	return (-1);
1593 }
1594 
1595 /*
1596  * This routine retrieves the current status of socket options.
1597  * It returns the size of the option retrieved, or -1.
1598  */
1599 int
1600 udp_opt_get(conn_t *connp, t_scalar_t level, t_scalar_t name,
1601     uchar_t *ptr)
1602 {
1603 	int		*i1 = (int *)ptr;
1604 	udp_t		*udp = connp->conn_udp;
1605 	int		len;
1606 	conn_opt_arg_t	coas;
1607 	int		retval;
1608 
1609 	coas.coa_connp = connp;
1610 	coas.coa_ixa = connp->conn_ixa;
1611 	coas.coa_ipp = &connp->conn_xmit_ipp;
1612 	coas.coa_ancillary = B_FALSE;
1613 	coas.coa_changed = 0;
1614 
1615 	/*
1616 	 * We assume that the optcom framework has checked for the set
1617 	 * of levels and names that are supported, hence we don't worry
1618 	 * about rejecting based on that.
1619 	 * First check for UDP specific handling, then pass to common routine.
1620 	 */
1621 	switch (level) {
1622 	case IPPROTO_IP:
1623 		/*
1624 		 * Only allow IPv4 option processing on IPv4 sockets.
1625 		 */
1626 		if (connp->conn_family != AF_INET)
1627 			return (-1);
1628 
1629 		switch (name) {
1630 		case IP_OPTIONS:
1631 		case T_IP_OPTIONS:
1632 			mutex_enter(&connp->conn_lock);
1633 			if (!(udp->udp_recv_ipp.ipp_fields &
1634 			    IPPF_IPV4_OPTIONS)) {
1635 				mutex_exit(&connp->conn_lock);
1636 				return (0);
1637 			}
1638 
1639 			len = udp->udp_recv_ipp.ipp_ipv4_options_len;
1640 			ASSERT(len != 0);
1641 			bcopy(udp->udp_recv_ipp.ipp_ipv4_options, ptr, len);
1642 			mutex_exit(&connp->conn_lock);
1643 			return (len);
1644 		}
1645 		break;
1646 	case IPPROTO_UDP:
1647 		switch (name) {
1648 		case UDP_NAT_T_ENDPOINT:
1649 			mutex_enter(&connp->conn_lock);
1650 			*i1 = udp->udp_nat_t_endpoint;
1651 			mutex_exit(&connp->conn_lock);
1652 			return (sizeof (int));
1653 		case UDP_RCVHDR:
1654 			mutex_enter(&connp->conn_lock);
1655 			*i1 = udp->udp_rcvhdr ? 1 : 0;
1656 			mutex_exit(&connp->conn_lock);
1657 			return (sizeof (int));
1658 		}
1659 	}
1660 	mutex_enter(&connp->conn_lock);
1661 	retval = conn_opt_get(&coas, level, name, ptr);
1662 	mutex_exit(&connp->conn_lock);
1663 	return (retval);
1664 }
1665 
1666 /*
1667  * This routine retrieves the current status of socket options.
1668  * It returns the size of the option retrieved, or -1.
1669  */
1670 int
1671 udp_tpi_opt_get(queue_t *q, t_scalar_t level, t_scalar_t name, uchar_t *ptr)
1672 {
1673 	conn_t		*connp = Q_TO_CONN(q);
1674 	int		err;
1675 
1676 	err = udp_opt_get(connp, level, name, ptr);
1677 	return (err);
1678 }
1679 
1680 /*
1681  * This routine sets socket options.
1682  */
1683 int
1684 udp_do_opt_set(conn_opt_arg_t *coa, int level, int name,
1685     uint_t inlen, uchar_t *invalp, cred_t *cr, boolean_t checkonly)
1686 {
1687 	conn_t		*connp = coa->coa_connp;
1688 	ip_xmit_attr_t	*ixa = coa->coa_ixa;
1689 	udp_t		*udp = connp->conn_udp;
1690 	udp_stack_t	*us = udp->udp_us;
1691 	int		*i1 = (int *)invalp;
1692 	boolean_t 	onoff = (*i1 == 0) ? 0 : 1;
1693 	int		error;
1694 
1695 	ASSERT(MUTEX_NOT_HELD(&coa->coa_connp->conn_lock));
1696 	/*
1697 	 * First do UDP specific sanity checks and handle UDP specific
1698 	 * options. Note that some IPPROTO_UDP options are handled
1699 	 * by conn_opt_set.
1700 	 */
1701 	switch (level) {
1702 	case SOL_SOCKET:
1703 		switch (name) {
1704 		case SO_SNDBUF:
1705 			if (*i1 > us->us_max_buf) {
1706 				return (ENOBUFS);
1707 			}
1708 			break;
1709 		case SO_RCVBUF:
1710 			if (*i1 > us->us_max_buf) {
1711 				return (ENOBUFS);
1712 			}
1713 			break;
1714 
1715 		case SCM_UCRED: {
1716 			struct ucred_s *ucr;
1717 			cred_t *newcr;
1718 			ts_label_t *tsl;
1719 
1720 			/*
1721 			 * Only sockets that have proper privileges and are
1722 			 * bound to MLPs will have any other value here, so
1723 			 * this implicitly tests for privilege to set label.
1724 			 */
1725 			if (connp->conn_mlp_type == mlptSingle)
1726 				break;
1727 
1728 			ucr = (struct ucred_s *)invalp;
1729 			if (inlen < sizeof (*ucr) + sizeof (bslabel_t) ||
1730 			    ucr->uc_labeloff < sizeof (*ucr) ||
1731 			    ucr->uc_labeloff + sizeof (bslabel_t) > inlen)
1732 				return (EINVAL);
1733 			if (!checkonly) {
1734 				/*
1735 				 * Set ixa_tsl to the new label.
1736 				 * We assume that crgetzoneid doesn't change
1737 				 * as part of the SCM_UCRED.
1738 				 */
1739 				ASSERT(cr != NULL);
1740 				if ((tsl = crgetlabel(cr)) == NULL)
1741 					return (EINVAL);
1742 				newcr = copycred_from_bslabel(cr, UCLABEL(ucr),
1743 				    tsl->tsl_doi, KM_NOSLEEP);
1744 				if (newcr == NULL)
1745 					return (ENOSR);
1746 				ASSERT(newcr->cr_label != NULL);
1747 				/*
1748 				 * Move the hold on the cr_label to ixa_tsl by
1749 				 * setting cr_label to NULL. Then release newcr.
1750 				 */
1751 				ip_xmit_attr_replace_tsl(ixa, newcr->cr_label);
1752 				ixa->ixa_flags |= IXAF_UCRED_TSL;
1753 				newcr->cr_label = NULL;
1754 				crfree(newcr);
1755 				coa->coa_changed |= COA_HEADER_CHANGED;
1756 				coa->coa_changed |= COA_WROFF_CHANGED;
1757 			}
1758 			/* Fully handled this option. */
1759 			return (0);
1760 		}
1761 		}
1762 		break;
1763 	case IPPROTO_UDP:
1764 		switch (name) {
1765 		case UDP_NAT_T_ENDPOINT:
1766 			if ((error = secpolicy_ip_config(cr, B_FALSE)) != 0) {
1767 				return (error);
1768 			}
1769 
1770 			/*
1771 			 * Use conn_family instead so we can avoid ambiguitites
1772 			 * with AF_INET6 sockets that may switch from IPv4
1773 			 * to IPv6.
1774 			 */
1775 			if (connp->conn_family != AF_INET) {
1776 				return (EAFNOSUPPORT);
1777 			}
1778 
1779 			if (!checkonly) {
1780 				mutex_enter(&connp->conn_lock);
1781 				udp->udp_nat_t_endpoint = onoff;
1782 				mutex_exit(&connp->conn_lock);
1783 				coa->coa_changed |= COA_HEADER_CHANGED;
1784 				coa->coa_changed |= COA_WROFF_CHANGED;
1785 			}
1786 			/* Fully handled this option. */
1787 			return (0);
1788 		case UDP_RCVHDR:
1789 			mutex_enter(&connp->conn_lock);
1790 			udp->udp_rcvhdr = onoff;
1791 			mutex_exit(&connp->conn_lock);
1792 			return (0);
1793 		}
1794 		break;
1795 	}
1796 	error = conn_opt_set(coa, level, name, inlen, invalp,
1797 	    checkonly, cr);
1798 	return (error);
1799 }
1800 
1801 /*
1802  * This routine sets socket options.
1803  */
1804 int
1805 udp_opt_set(conn_t *connp, uint_t optset_context, int level,
1806     int name, uint_t inlen, uchar_t *invalp, uint_t *outlenp,
1807     uchar_t *outvalp, void *thisdg_attrs, cred_t *cr)
1808 {
1809 	udp_t		*udp = connp->conn_udp;
1810 	int		err;
1811 	conn_opt_arg_t	coas, *coa;
1812 	boolean_t	checkonly;
1813 	udp_stack_t	*us = udp->udp_us;
1814 
1815 	switch (optset_context) {
1816 	case SETFN_OPTCOM_CHECKONLY:
1817 		checkonly = B_TRUE;
1818 		/*
1819 		 * Note: Implies T_CHECK semantics for T_OPTCOM_REQ
1820 		 * inlen != 0 implies value supplied and
1821 		 * 	we have to "pretend" to set it.
1822 		 * inlen == 0 implies that there is no
1823 		 * 	value part in T_CHECK request and just validation
1824 		 * done elsewhere should be enough, we just return here.
1825 		 */
1826 		if (inlen == 0) {
1827 			*outlenp = 0;
1828 			return (0);
1829 		}
1830 		break;
1831 	case SETFN_OPTCOM_NEGOTIATE:
1832 		checkonly = B_FALSE;
1833 		break;
1834 	case SETFN_UD_NEGOTIATE:
1835 	case SETFN_CONN_NEGOTIATE:
1836 		checkonly = B_FALSE;
1837 		/*
1838 		 * Negotiating local and "association-related" options
1839 		 * through T_UNITDATA_REQ.
1840 		 *
1841 		 * Following routine can filter out ones we do not
1842 		 * want to be "set" this way.
1843 		 */
1844 		if (!udp_opt_allow_udr_set(level, name)) {
1845 			*outlenp = 0;
1846 			return (EINVAL);
1847 		}
1848 		break;
1849 	default:
1850 		/*
1851 		 * We should never get here
1852 		 */
1853 		*outlenp = 0;
1854 		return (EINVAL);
1855 	}
1856 
1857 	ASSERT((optset_context != SETFN_OPTCOM_CHECKONLY) ||
1858 	    (optset_context == SETFN_OPTCOM_CHECKONLY && inlen != 0));
1859 
1860 	if (thisdg_attrs != NULL) {
1861 		/* Options from T_UNITDATA_REQ */
1862 		coa = (conn_opt_arg_t *)thisdg_attrs;
1863 		ASSERT(coa->coa_connp == connp);
1864 		ASSERT(coa->coa_ixa != NULL);
1865 		ASSERT(coa->coa_ipp != NULL);
1866 		ASSERT(coa->coa_ancillary);
1867 	} else {
1868 		coa = &coas;
1869 		coas.coa_connp = connp;
1870 		/* Get a reference on conn_ixa to prevent concurrent mods */
1871 		coas.coa_ixa = conn_get_ixa(connp, B_TRUE);
1872 		if (coas.coa_ixa == NULL) {
1873 			*outlenp = 0;
1874 			return (ENOMEM);
1875 		}
1876 		coas.coa_ipp = &connp->conn_xmit_ipp;
1877 		coas.coa_ancillary = B_FALSE;
1878 		coas.coa_changed = 0;
1879 	}
1880 
1881 	err = udp_do_opt_set(coa, level, name, inlen, invalp,
1882 	    cr, checkonly);
1883 	if (err != 0) {
1884 errout:
1885 		if (!coa->coa_ancillary)
1886 			ixa_refrele(coa->coa_ixa);
1887 		*outlenp = 0;
1888 		return (err);
1889 	}
1890 	/* Handle DHCPINIT here outside of lock */
1891 	if (level == IPPROTO_IP && name == IP_DHCPINIT_IF) {
1892 		uint_t	ifindex;
1893 		ill_t	*ill;
1894 
1895 		ifindex = *(uint_t *)invalp;
1896 		if (ifindex == 0) {
1897 			ill = NULL;
1898 		} else {
1899 			ill = ill_lookup_on_ifindex(ifindex, B_FALSE,
1900 			    coa->coa_ixa->ixa_ipst);
1901 			if (ill == NULL) {
1902 				err = ENXIO;
1903 				goto errout;
1904 			}
1905 
1906 			mutex_enter(&ill->ill_lock);
1907 			if (ill->ill_state_flags & ILL_CONDEMNED) {
1908 				mutex_exit(&ill->ill_lock);
1909 				ill_refrele(ill);
1910 				err = ENXIO;
1911 				goto errout;
1912 			}
1913 			if (IS_VNI(ill)) {
1914 				mutex_exit(&ill->ill_lock);
1915 				ill_refrele(ill);
1916 				err = EINVAL;
1917 				goto errout;
1918 			}
1919 		}
1920 		mutex_enter(&connp->conn_lock);
1921 
1922 		if (connp->conn_dhcpinit_ill != NULL) {
1923 			/*
1924 			 * We've locked the conn so conn_cleanup_ill()
1925 			 * cannot clear conn_dhcpinit_ill -- so it's
1926 			 * safe to access the ill.
1927 			 */
1928 			ill_t *oill = connp->conn_dhcpinit_ill;
1929 
1930 			ASSERT(oill->ill_dhcpinit != 0);
1931 			atomic_dec_32(&oill->ill_dhcpinit);
1932 			ill_set_inputfn(connp->conn_dhcpinit_ill);
1933 			connp->conn_dhcpinit_ill = NULL;
1934 		}
1935 
1936 		if (ill != NULL) {
1937 			connp->conn_dhcpinit_ill = ill;
1938 			atomic_inc_32(&ill->ill_dhcpinit);
1939 			ill_set_inputfn(ill);
1940 			mutex_exit(&connp->conn_lock);
1941 			mutex_exit(&ill->ill_lock);
1942 			ill_refrele(ill);
1943 		} else {
1944 			mutex_exit(&connp->conn_lock);
1945 		}
1946 	}
1947 
1948 	/*
1949 	 * Common case of OK return with outval same as inval.
1950 	 */
1951 	if (invalp != outvalp) {
1952 		/* don't trust bcopy for identical src/dst */
1953 		(void) bcopy(invalp, outvalp, inlen);
1954 	}
1955 	*outlenp = inlen;
1956 
1957 	/*
1958 	 * If this was not ancillary data, then we rebuild the headers,
1959 	 * update the IRE/NCE, and IPsec as needed.
1960 	 * Since the label depends on the destination we go through
1961 	 * ip_set_destination first.
1962 	 */
1963 	if (coa->coa_ancillary) {
1964 		return (0);
1965 	}
1966 
1967 	if (coa->coa_changed & COA_ROUTE_CHANGED) {
1968 		in6_addr_t saddr, faddr, nexthop;
1969 		in_port_t fport;
1970 
1971 		/*
1972 		 * We clear lastdst to make sure we pick up the change
1973 		 * next time sending.
1974 		 * If we are connected we re-cache the information.
1975 		 * We ignore errors to preserve BSD behavior.
1976 		 * Note that we don't redo IPsec policy lookup here
1977 		 * since the final destination (or source) didn't change.
1978 		 */
1979 		mutex_enter(&connp->conn_lock);
1980 		connp->conn_v6lastdst = ipv6_all_zeros;
1981 
1982 		ip_attr_nexthop(coa->coa_ipp, coa->coa_ixa,
1983 		    &connp->conn_faddr_v6, &nexthop);
1984 		saddr = connp->conn_saddr_v6;
1985 		faddr = connp->conn_faddr_v6;
1986 		fport = connp->conn_fport;
1987 		mutex_exit(&connp->conn_lock);
1988 
1989 		if (!IN6_IS_ADDR_UNSPECIFIED(&faddr) &&
1990 		    !IN6_IS_ADDR_V4MAPPED_ANY(&faddr)) {
1991 			(void) ip_attr_connect(connp, coa->coa_ixa,
1992 			    &saddr, &faddr, &nexthop, fport, NULL, NULL,
1993 			    IPDF_ALLOW_MCBC | IPDF_VERIFY_DST);
1994 		}
1995 	}
1996 
1997 	ixa_refrele(coa->coa_ixa);
1998 
1999 	if (coa->coa_changed & COA_HEADER_CHANGED) {
2000 		/*
2001 		 * Rebuild the header template if we are connected.
2002 		 * Otherwise clear conn_v6lastdst so we rebuild the header
2003 		 * in the data path.
2004 		 */
2005 		mutex_enter(&connp->conn_lock);
2006 		if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) &&
2007 		    !IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_faddr_v6)) {
2008 			err = udp_build_hdr_template(connp,
2009 			    &connp->conn_saddr_v6, &connp->conn_faddr_v6,
2010 			    connp->conn_fport, connp->conn_flowinfo);
2011 			if (err != 0) {
2012 				mutex_exit(&connp->conn_lock);
2013 				return (err);
2014 			}
2015 		} else {
2016 			connp->conn_v6lastdst = ipv6_all_zeros;
2017 		}
2018 		mutex_exit(&connp->conn_lock);
2019 	}
2020 	if (coa->coa_changed & COA_RCVBUF_CHANGED) {
2021 		(void) proto_set_rx_hiwat(connp->conn_rq, connp,
2022 		    connp->conn_rcvbuf);
2023 	}
2024 	if ((coa->coa_changed & COA_SNDBUF_CHANGED) && !IPCL_IS_NONSTR(connp)) {
2025 		connp->conn_wq->q_hiwat = connp->conn_sndbuf;
2026 	}
2027 	if (coa->coa_changed & COA_WROFF_CHANGED) {
2028 		/* Increase wroff if needed */
2029 		uint_t wroff;
2030 
2031 		mutex_enter(&connp->conn_lock);
2032 		wroff = connp->conn_ht_iphc_allocated + us->us_wroff_extra;
2033 		if (udp->udp_nat_t_endpoint)
2034 			wroff += sizeof (uint32_t);
2035 		if (wroff > connp->conn_wroff) {
2036 			connp->conn_wroff = wroff;
2037 			mutex_exit(&connp->conn_lock);
2038 			(void) proto_set_tx_wroff(connp->conn_rq, connp, wroff);
2039 		} else {
2040 			mutex_exit(&connp->conn_lock);
2041 		}
2042 	}
2043 	return (err);
2044 }
2045 
2046 /* This routine sets socket options. */
2047 int
2048 udp_tpi_opt_set(queue_t *q, uint_t optset_context, int level, int name,
2049     uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
2050     void *thisdg_attrs, cred_t *cr)
2051 {
2052 	conn_t	*connp = Q_TO_CONN(q);
2053 	int error;
2054 
2055 	error = udp_opt_set(connp, optset_context, level, name, inlen, invalp,
2056 	    outlenp, outvalp, thisdg_attrs, cr);
2057 	return (error);
2058 }
2059 
2060 /*
2061  * Setup IP and UDP headers.
2062  * Returns NULL on allocation failure, in which case data_mp is freed.
2063  */
2064 mblk_t *
2065 udp_prepend_hdr(conn_t *connp, ip_xmit_attr_t *ixa, const ip_pkt_t *ipp,
2066     const in6_addr_t *v6src, const in6_addr_t *v6dst, in_port_t dstport,
2067     uint32_t flowinfo, mblk_t *data_mp, int *errorp)
2068 {
2069 	mblk_t		*mp;
2070 	udpha_t		*udpha;
2071 	udp_stack_t	*us = connp->conn_netstack->netstack_udp;
2072 	uint_t		data_len;
2073 	uint32_t	cksum;
2074 	udp_t		*udp = connp->conn_udp;
2075 	boolean_t	insert_spi = udp->udp_nat_t_endpoint;
2076 	uint_t		ulp_hdr_len;
2077 
2078 	data_len = msgdsize(data_mp);
2079 	ulp_hdr_len = UDPH_SIZE;
2080 	if (insert_spi)
2081 		ulp_hdr_len += sizeof (uint32_t);
2082 
2083 	mp = conn_prepend_hdr(ixa, ipp, v6src, v6dst, IPPROTO_UDP, flowinfo,
2084 	    ulp_hdr_len, data_mp, data_len, us->us_wroff_extra, &cksum, errorp);
2085 	if (mp == NULL) {
2086 		ASSERT(*errorp != 0);
2087 		return (NULL);
2088 	}
2089 
2090 	data_len += ulp_hdr_len;
2091 	ixa->ixa_pktlen = data_len + ixa->ixa_ip_hdr_length;
2092 
2093 	udpha = (udpha_t *)(mp->b_rptr + ixa->ixa_ip_hdr_length);
2094 	udpha->uha_src_port = connp->conn_lport;
2095 	udpha->uha_dst_port = dstport;
2096 	udpha->uha_checksum = 0;
2097 	udpha->uha_length = htons(data_len);
2098 
2099 	/*
2100 	 * If there was a routing option/header then conn_prepend_hdr
2101 	 * has massaged it and placed the pseudo-header checksum difference
2102 	 * in the cksum argument.
2103 	 *
2104 	 * Setup header length and prepare for ULP checksum done in IP.
2105 	 *
2106 	 * We make it easy for IP to include our pseudo header
2107 	 * by putting our length in uha_checksum.
2108 	 * The IP source, destination, and length have already been set by
2109 	 * conn_prepend_hdr.
2110 	 */
2111 	cksum += data_len;
2112 	cksum = (cksum >> 16) + (cksum & 0xFFFF);
2113 	ASSERT(cksum < 0x10000);
2114 
2115 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
2116 		ipha_t	*ipha = (ipha_t *)mp->b_rptr;
2117 
2118 		ASSERT(ntohs(ipha->ipha_length) == ixa->ixa_pktlen);
2119 
2120 		/* IP does the checksum if uha_checksum is non-zero */
2121 		if (us->us_do_checksum) {
2122 			if (cksum == 0)
2123 				udpha->uha_checksum = 0xffff;
2124 			else
2125 				udpha->uha_checksum = htons(cksum);
2126 		} else {
2127 			udpha->uha_checksum = 0;
2128 		}
2129 	} else {
2130 		ip6_t *ip6h = (ip6_t *)mp->b_rptr;
2131 
2132 		ASSERT(ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN == ixa->ixa_pktlen);
2133 		if (cksum == 0)
2134 			udpha->uha_checksum = 0xffff;
2135 		else
2136 			udpha->uha_checksum = htons(cksum);
2137 	}
2138 
2139 	/* Insert all-0s SPI now. */
2140 	if (insert_spi)
2141 		*((uint32_t *)(udpha + 1)) = 0;
2142 
2143 	return (mp);
2144 }
2145 
2146 static int
2147 udp_build_hdr_template(conn_t *connp, const in6_addr_t *v6src,
2148     const in6_addr_t *v6dst, in_port_t dstport, uint32_t flowinfo)
2149 {
2150 	udpha_t		*udpha;
2151 	int		error;
2152 
2153 	ASSERT(MUTEX_HELD(&connp->conn_lock));
2154 	/*
2155 	 * We clear lastdst to make sure we don't use the lastdst path
2156 	 * next time sending since we might not have set v6dst yet.
2157 	 */
2158 	connp->conn_v6lastdst = ipv6_all_zeros;
2159 
2160 	error = conn_build_hdr_template(connp, UDPH_SIZE, 0, v6src, v6dst,
2161 	    flowinfo);
2162 	if (error != 0)
2163 		return (error);
2164 
2165 	/*
2166 	 * Any routing header/option has been massaged. The checksum difference
2167 	 * is stored in conn_sum.
2168 	 */
2169 	udpha = (udpha_t *)connp->conn_ht_ulp;
2170 	udpha->uha_src_port = connp->conn_lport;
2171 	udpha->uha_dst_port = dstport;
2172 	udpha->uha_checksum = 0;
2173 	udpha->uha_length = htons(UDPH_SIZE);	/* Filled in later */
2174 	return (0);
2175 }
2176 
2177 static mblk_t *
2178 udp_queue_fallback(udp_t *udp, mblk_t *mp)
2179 {
2180 	ASSERT(MUTEX_HELD(&udp->udp_recv_lock));
2181 	if (IPCL_IS_NONSTR(udp->udp_connp)) {
2182 		/*
2183 		 * fallback has started but messages have not been moved yet
2184 		 */
2185 		if (udp->udp_fallback_queue_head == NULL) {
2186 			ASSERT(udp->udp_fallback_queue_tail == NULL);
2187 			udp->udp_fallback_queue_head = mp;
2188 			udp->udp_fallback_queue_tail = mp;
2189 		} else {
2190 			ASSERT(udp->udp_fallback_queue_tail != NULL);
2191 			udp->udp_fallback_queue_tail->b_next = mp;
2192 			udp->udp_fallback_queue_tail = mp;
2193 		}
2194 		return (NULL);
2195 	} else {
2196 		/*
2197 		 * Fallback completed, let the caller putnext() the mblk.
2198 		 */
2199 		return (mp);
2200 	}
2201 }
2202 
2203 /*
2204  * Deliver data to ULP. In case we have a socket, and it's falling back to
2205  * TPI, then we'll queue the mp for later processing.
2206  */
2207 static void
2208 udp_ulp_recv(conn_t *connp, mblk_t *mp, uint_t len, ip_recv_attr_t *ira)
2209 {
2210 	if (IPCL_IS_NONSTR(connp)) {
2211 		udp_t *udp = connp->conn_udp;
2212 		int error;
2213 
2214 		ASSERT(len == msgdsize(mp));
2215 		if ((*connp->conn_upcalls->su_recv)
2216 		    (connp->conn_upper_handle, mp, len, 0, &error, NULL) < 0) {
2217 			mutex_enter(&udp->udp_recv_lock);
2218 			if (error == ENOSPC) {
2219 				/*
2220 				 * let's confirm while holding the lock
2221 				 */
2222 				if ((*connp->conn_upcalls->su_recv)
2223 				    (connp->conn_upper_handle, NULL, 0, 0,
2224 				    &error, NULL) < 0) {
2225 					ASSERT(error == ENOSPC);
2226 					if (error == ENOSPC) {
2227 						connp->conn_flow_cntrld =
2228 						    B_TRUE;
2229 					}
2230 				}
2231 				mutex_exit(&udp->udp_recv_lock);
2232 			} else {
2233 				ASSERT(error == EOPNOTSUPP);
2234 				mp = udp_queue_fallback(udp, mp);
2235 				mutex_exit(&udp->udp_recv_lock);
2236 				if (mp != NULL)
2237 					putnext(connp->conn_rq, mp);
2238 			}
2239 		}
2240 		ASSERT(MUTEX_NOT_HELD(&udp->udp_recv_lock));
2241 	} else {
2242 		if (is_system_labeled()) {
2243 			ASSERT(ira->ira_cred != NULL);
2244 			/*
2245 			 * Provide for protocols above UDP such as RPC
2246 			 * NOPID leaves db_cpid unchanged.
2247 			 */
2248 			mblk_setcred(mp, ira->ira_cred, NOPID);
2249 		}
2250 
2251 		putnext(connp->conn_rq, mp);
2252 	}
2253 }
2254 
2255 /*
2256  * This is the inbound data path.
2257  * IP has already pulled up the IP plus UDP headers and verified alignment
2258  * etc.
2259  */
2260 /* ARGSUSED2 */
2261 static void
2262 udp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
2263 {
2264 	conn_t			*connp = (conn_t *)arg1;
2265 	struct T_unitdata_ind	*tudi;
2266 	uchar_t			*rptr;		/* Pointer to IP header */
2267 	int			hdr_length;	/* Length of IP+UDP headers */
2268 	int			udi_size;	/* Size of T_unitdata_ind */
2269 	int			pkt_len;
2270 	udp_t			*udp;
2271 	udpha_t			*udpha;
2272 	ip_pkt_t		ipps;
2273 	ip6_t			*ip6h;
2274 	mblk_t			*mp1;
2275 	uint32_t		udp_ipv4_options_len;
2276 	crb_t			recv_ancillary;
2277 	udp_stack_t		*us;
2278 	conn_t			*new = NULL;
2279 
2280 	ASSERT(connp->conn_flags & IPCL_UDPCONN);
2281 
2282 	mutex_enter(&connp->conn_lock);
2283 	if (connp->conn_reuselist != NULL) {
2284 		struct reuselist *reusep = connp->conn_reuselist;
2285 		int i;
2286 
2287 		/*
2288 		 * we have to balance the request between multiple sockets.
2289 		 * Currently we do this in a round-robin fashion. In the
2290 		 * reuselist we maintain a pointer to the last receiver.
2291 		 * TODO: we can add a check if the conn is full and skip to
2292 		 * the next.
2293 		 */
2294 		mutex_enter(&reusep->ru_lock);
2295 		i = reusep->ru_next;
2296 		new = reusep->ru_conns[i];
2297 		if (++i == reusep->ru_entries)
2298 			i = 0;
2299 		reusep->ru_next = i;
2300 		if (new == connp)
2301 			new = NULL;
2302 		else
2303 			CONN_INC_REF(new);
2304 		mutex_exit(&reusep->ru_lock);
2305 		mutex_exit(&connp->conn_lock);
2306 		if (new != NULL)
2307 			connp = new;
2308 	} else {
2309 		mutex_exit(&connp->conn_lock);
2310 	}
2311 
2312 	udp = connp->conn_udp;
2313 	us = udp->udp_us;
2314 	rptr = mp->b_rptr;
2315 
2316 	ASSERT(DB_TYPE(mp) == M_DATA);
2317 	ASSERT(OK_32PTR(rptr));
2318 	ASSERT(ira->ira_pktlen == msgdsize(mp));
2319 	pkt_len = ira->ira_pktlen;
2320 
2321 	/*
2322 	 * Get a snapshot of these and allow other threads to change
2323 	 * them after that. We need the same recv_ancillary when determining
2324 	 * the size as when adding the ancillary data items.
2325 	 */
2326 	mutex_enter(&connp->conn_lock);
2327 	udp_ipv4_options_len = udp->udp_recv_ipp.ipp_ipv4_options_len;
2328 	recv_ancillary = connp->conn_recv_ancillary;
2329 	mutex_exit(&connp->conn_lock);
2330 
2331 	hdr_length = ira->ira_ip_hdr_length;
2332 
2333 	/*
2334 	 * IP inspected the UDP header thus all of it must be in the mblk.
2335 	 * UDP length check is performed for IPv6 packets and IPv4 packets
2336 	 * to check if the size of the packet as specified
2337 	 * by the UDP header is the same as the length derived from the IP
2338 	 * header.
2339 	 */
2340 	udpha = (udpha_t *)(rptr + hdr_length);
2341 	if (pkt_len != ntohs(udpha->uha_length) + hdr_length)
2342 		goto tossit;
2343 
2344 	hdr_length += UDPH_SIZE;
2345 	ASSERT(MBLKL(mp) >= hdr_length);	/* IP did a pullup */
2346 
2347 	/* Initialize regardless of IP version */
2348 	ipps.ipp_fields = 0;
2349 
2350 	if (((ira->ira_flags & IRAF_IPV4_OPTIONS) ||
2351 	    udp_ipv4_options_len > 0) &&
2352 	    connp->conn_family == AF_INET) {
2353 		int	err;
2354 
2355 		/*
2356 		 * Record/update udp_recv_ipp with the lock
2357 		 * held. Not needed for AF_INET6 sockets
2358 		 * since they don't support a getsockopt of IP_OPTIONS.
2359 		 */
2360 		mutex_enter(&connp->conn_lock);
2361 		err = ip_find_hdr_v4((ipha_t *)rptr, &udp->udp_recv_ipp,
2362 		    B_TRUE);
2363 		if (err != 0) {
2364 			/* Allocation failed. Drop packet */
2365 			mutex_exit(&connp->conn_lock);
2366 			goto tossit;
2367 		}
2368 		mutex_exit(&connp->conn_lock);
2369 	}
2370 
2371 	if (recv_ancillary.crb_all != 0) {
2372 		/*
2373 		 * Record packet information in the ip_pkt_t
2374 		 */
2375 		if (ira->ira_flags & IRAF_IS_IPV4) {
2376 			ASSERT(IPH_HDR_VERSION(rptr) == IPV4_VERSION);
2377 			ASSERT(MBLKL(mp) >= sizeof (ipha_t));
2378 			ASSERT(((ipha_t *)rptr)->ipha_protocol == IPPROTO_UDP);
2379 			ASSERT(ira->ira_ip_hdr_length == IPH_HDR_LENGTH(rptr));
2380 
2381 			(void) ip_find_hdr_v4((ipha_t *)rptr, &ipps, B_FALSE);
2382 		} else {
2383 			uint8_t nexthdrp;
2384 
2385 			ASSERT(IPH_HDR_VERSION(rptr) == IPV6_VERSION);
2386 			/*
2387 			 * IPv6 packets can only be received by applications
2388 			 * that are prepared to receive IPv6 addresses.
2389 			 * The IP fanout must ensure this.
2390 			 */
2391 			ASSERT(connp->conn_family == AF_INET6);
2392 
2393 			ip6h = (ip6_t *)rptr;
2394 
2395 			/* We don't care about the length, but need the ipp */
2396 			hdr_length = ip_find_hdr_v6(mp, ip6h, B_TRUE, &ipps,
2397 			    &nexthdrp);
2398 			ASSERT(hdr_length == ira->ira_ip_hdr_length);
2399 			/* Restore */
2400 			hdr_length = ira->ira_ip_hdr_length + UDPH_SIZE;
2401 			ASSERT(nexthdrp == IPPROTO_UDP);
2402 		}
2403 	}
2404 
2405 	/*
2406 	 * This is the inbound data path.  Packets are passed upstream as
2407 	 * T_UNITDATA_IND messages.
2408 	 */
2409 	if (connp->conn_family == AF_INET) {
2410 		sin_t *sin;
2411 
2412 		ASSERT(IPH_HDR_VERSION((ipha_t *)rptr) == IPV4_VERSION);
2413 
2414 		/*
2415 		 * Normally only send up the source address.
2416 		 * If any ancillary data items are wanted we add those.
2417 		 */
2418 		udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin_t);
2419 		if (recv_ancillary.crb_all != 0) {
2420 			udi_size += conn_recvancillary_size(connp,
2421 			    recv_ancillary, ira, mp, &ipps);
2422 		}
2423 
2424 		/* Allocate a message block for the T_UNITDATA_IND structure. */
2425 		mp1 = allocb(udi_size, BPRI_MED);
2426 		if (mp1 == NULL)
2427 			goto tossit;
2428 		mp1->b_cont = mp;
2429 		mp1->b_datap->db_type = M_PROTO;
2430 		tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2431 		mp1->b_wptr = (uchar_t *)tudi + udi_size;
2432 		tudi->PRIM_type = T_UNITDATA_IND;
2433 		tudi->SRC_length = sizeof (sin_t);
2434 		tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2435 		tudi->OPT_offset = sizeof (struct T_unitdata_ind) +
2436 		    sizeof (sin_t);
2437 		udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin_t));
2438 		tudi->OPT_length = udi_size;
2439 		sin = (sin_t *)&tudi[1];
2440 		sin->sin_addr.s_addr = ((ipha_t *)rptr)->ipha_src;
2441 		sin->sin_port =	udpha->uha_src_port;
2442 		sin->sin_family = connp->conn_family;
2443 		*(uint32_t *)&sin->sin_zero[0] = 0;
2444 		*(uint32_t *)&sin->sin_zero[4] = 0;
2445 
2446 		/*
2447 		 * Add options if IP_RECVDSTADDR, IP_RECVIF, IP_RECVSLLA or
2448 		 * IP_RECVTTL has been set.
2449 		 */
2450 		if (udi_size != 0) {
2451 			conn_recvancillary_add(connp, recv_ancillary, ira,
2452 			    &ipps, (uchar_t *)&sin[1], udi_size);
2453 		}
2454 	} else {
2455 		sin6_t *sin6;
2456 
2457 		/*
2458 		 * Handle both IPv4 and IPv6 packets for IPv6 sockets.
2459 		 *
2460 		 * Normally we only send up the address. If receiving of any
2461 		 * optional receive side information is enabled, we also send
2462 		 * that up as options.
2463 		 */
2464 		udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t);
2465 
2466 		if (recv_ancillary.crb_all != 0) {
2467 			udi_size += conn_recvancillary_size(connp,
2468 			    recv_ancillary, ira, mp, &ipps);
2469 		}
2470 
2471 		mp1 = allocb(udi_size, BPRI_MED);
2472 		if (mp1 == NULL)
2473 			goto tossit;
2474 		mp1->b_cont = mp;
2475 		mp1->b_datap->db_type = M_PROTO;
2476 		tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2477 		mp1->b_wptr = (uchar_t *)tudi + udi_size;
2478 		tudi->PRIM_type = T_UNITDATA_IND;
2479 		tudi->SRC_length = sizeof (sin6_t);
2480 		tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2481 		tudi->OPT_offset = sizeof (struct T_unitdata_ind) +
2482 		    sizeof (sin6_t);
2483 		udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin6_t));
2484 		tudi->OPT_length = udi_size;
2485 		sin6 = (sin6_t *)&tudi[1];
2486 		if (ira->ira_flags & IRAF_IS_IPV4) {
2487 			in6_addr_t v6dst;
2488 
2489 			IN6_IPADDR_TO_V4MAPPED(((ipha_t *)rptr)->ipha_src,
2490 			    &sin6->sin6_addr);
2491 			IN6_IPADDR_TO_V4MAPPED(((ipha_t *)rptr)->ipha_dst,
2492 			    &v6dst);
2493 			sin6->sin6_flowinfo = 0;
2494 			sin6->sin6_scope_id = 0;
2495 			sin6->__sin6_src_id = ip_srcid_find_addr(&v6dst,
2496 			    IPCL_ZONEID(connp), us->us_netstack);
2497 		} else {
2498 			ip6h = (ip6_t *)rptr;
2499 
2500 			sin6->sin6_addr = ip6h->ip6_src;
2501 			/* No sin6_flowinfo per API */
2502 			sin6->sin6_flowinfo = 0;
2503 			/* For link-scope pass up scope id */
2504 			if (IN6_IS_ADDR_LINKSCOPE(&ip6h->ip6_src))
2505 				sin6->sin6_scope_id = ira->ira_ruifindex;
2506 			else
2507 				sin6->sin6_scope_id = 0;
2508 			sin6->__sin6_src_id = ip_srcid_find_addr(
2509 			    &ip6h->ip6_dst, IPCL_ZONEID(connp),
2510 			    us->us_netstack);
2511 		}
2512 		sin6->sin6_port = udpha->uha_src_port;
2513 		sin6->sin6_family = connp->conn_family;
2514 
2515 		if (udi_size != 0) {
2516 			conn_recvancillary_add(connp, recv_ancillary, ira,
2517 			    &ipps, (uchar_t *)&sin6[1], udi_size);
2518 		}
2519 	}
2520 
2521 	/*
2522 	 * DTrace this UDP input as udp:::receive (this is for IPv4, IPv6 and
2523 	 * loopback traffic).
2524 	 */
2525 	DTRACE_UDP5(receive, mblk_t *, NULL, ip_xmit_attr_t *, connp->conn_ixa,
2526 	    void_ip_t *, rptr, udp_t *, udp, udpha_t *, udpha);
2527 
2528 	/* Walk past the headers unless IP_RECVHDR was set. */
2529 	if (!udp->udp_rcvhdr) {
2530 		mp->b_rptr = rptr + hdr_length;
2531 		pkt_len -= hdr_length;
2532 	}
2533 
2534 	UDPS_BUMP_MIB(us, udpHCInDatagrams);
2535 	udp_ulp_recv(connp, mp1, pkt_len, ira);
2536 	if (new != NULL)
2537 		CONN_DEC_REF(new);
2538 	return;
2539 
2540 tossit:
2541 	freemsg(mp);
2542 	UDPS_BUMP_MIB(us, udpInErrors);
2543 	if (new != NULL)
2544 		CONN_DEC_REF(new);
2545 }
2546 
2547 /*
2548  * This routine creates a T_UDERROR_IND message and passes it upstream.
2549  * The address and options are copied from the T_UNITDATA_REQ message
2550  * passed in mp.  This message is freed.
2551  */
2552 static void
2553 udp_ud_err(queue_t *q, mblk_t *mp, t_scalar_t err)
2554 {
2555 	struct T_unitdata_req *tudr;
2556 	mblk_t	*mp1;
2557 	uchar_t *destaddr;
2558 	t_scalar_t destlen;
2559 	uchar_t	*optaddr;
2560 	t_scalar_t optlen;
2561 
2562 	if ((mp->b_wptr < mp->b_rptr) ||
2563 	    (MBLKL(mp)) < sizeof (struct T_unitdata_req)) {
2564 		goto done;
2565 	}
2566 	tudr = (struct T_unitdata_req *)mp->b_rptr;
2567 	destaddr = mp->b_rptr + tudr->DEST_offset;
2568 	if (destaddr < mp->b_rptr || destaddr >= mp->b_wptr ||
2569 	    destaddr + tudr->DEST_length < mp->b_rptr ||
2570 	    destaddr + tudr->DEST_length > mp->b_wptr) {
2571 		goto done;
2572 	}
2573 	optaddr = mp->b_rptr + tudr->OPT_offset;
2574 	if (optaddr < mp->b_rptr || optaddr >= mp->b_wptr ||
2575 	    optaddr + tudr->OPT_length < mp->b_rptr ||
2576 	    optaddr + tudr->OPT_length > mp->b_wptr) {
2577 		goto done;
2578 	}
2579 	destlen = tudr->DEST_length;
2580 	optlen = tudr->OPT_length;
2581 
2582 	mp1 = mi_tpi_uderror_ind((char *)destaddr, destlen,
2583 	    (char *)optaddr, optlen, err);
2584 	if (mp1 != NULL)
2585 		qreply(q, mp1);
2586 
2587 done:
2588 	freemsg(mp);
2589 }
2590 
2591 /*
2592  * This routine removes a port number association from a stream.  It
2593  * is called by udp_wput to handle T_UNBIND_REQ messages.
2594  */
2595 static void
2596 udp_tpi_unbind(queue_t *q, mblk_t *mp)
2597 {
2598 	conn_t	*connp = Q_TO_CONN(q);
2599 	int	error;
2600 
2601 	error = udp_do_unbind(connp);
2602 	if (error) {
2603 		if (error < 0)
2604 			udp_err_ack(q, mp, -error, 0);
2605 		else
2606 			udp_err_ack(q, mp, TSYSERR, error);
2607 		return;
2608 	}
2609 
2610 	mp = mi_tpi_ok_ack_alloc(mp);
2611 	ASSERT(mp != NULL);
2612 	ASSERT(((struct T_ok_ack *)mp->b_rptr)->PRIM_type == T_OK_ACK);
2613 	qreply(q, mp);
2614 }
2615 
2616 /*
2617  * Don't let port fall into the privileged range.
2618  * Since the extra privileged ports can be arbitrary we also
2619  * ensure that we exclude those from consideration.
2620  * us->us_epriv_ports is not sorted thus we loop over it until
2621  * there are no changes.
2622  */
2623 static in_port_t
2624 udp_update_next_port(udp_t *udp, in_port_t port, boolean_t random)
2625 {
2626 	int i, bump;
2627 	in_port_t nextport;
2628 	boolean_t restart = B_FALSE;
2629 	udp_stack_t *us = udp->udp_us;
2630 
2631 	if (random && udp_random_anon_port != 0) {
2632 		(void) random_get_pseudo_bytes((uint8_t *)&port,
2633 		    sizeof (in_port_t));
2634 		/*
2635 		 * Unless changed by a sys admin, the smallest anon port
2636 		 * is 32768 and the largest anon port is 65535.  It is
2637 		 * very likely (50%) for the random port to be smaller
2638 		 * than the smallest anon port.  When that happens,
2639 		 * add port % (anon port range) to the smallest anon
2640 		 * port to get the random port.  It should fall into the
2641 		 * valid anon port range.
2642 		 */
2643 		if ((port < us->us_smallest_anon_port) ||
2644 		    (port > us->us_largest_anon_port)) {
2645 			if (us->us_smallest_anon_port ==
2646 			    us->us_largest_anon_port) {
2647 				bump = 0;
2648 			} else {
2649 				bump = port % (us->us_largest_anon_port -
2650 				    us->us_smallest_anon_port);
2651 			}
2652 
2653 			port = us->us_smallest_anon_port + bump;
2654 		}
2655 	}
2656 
2657 retry:
2658 	if (port < us->us_smallest_anon_port)
2659 		port = us->us_smallest_anon_port;
2660 
2661 	if (port > us->us_largest_anon_port) {
2662 		port = us->us_smallest_anon_port;
2663 		if (restart)
2664 			return (0);
2665 		restart = B_TRUE;
2666 	}
2667 
2668 	if (port < us->us_smallest_nonpriv_port)
2669 		port = us->us_smallest_nonpriv_port;
2670 
2671 	for (i = 0; i < us->us_num_epriv_ports; i++) {
2672 		if (port == us->us_epriv_ports[i]) {
2673 			port++;
2674 			/*
2675 			 * Make sure that the port is in the
2676 			 * valid range.
2677 			 */
2678 			goto retry;
2679 		}
2680 	}
2681 
2682 	if (is_system_labeled() &&
2683 	    (nextport = tsol_next_port(crgetzone(udp->udp_connp->conn_cred),
2684 	    port, IPPROTO_UDP, B_TRUE)) != 0) {
2685 		port = nextport;
2686 		goto retry;
2687 	}
2688 
2689 	return (port);
2690 }
2691 
2692 /*
2693  * Handle T_UNITDATA_REQ with options. Both IPv4 and IPv6
2694  * Either tudr_mp or msg is set. If tudr_mp we take ancillary data from
2695  * the TPI options, otherwise we take them from msg_control.
2696  * If both sin and sin6 is set it is a connected socket and we use conn_faddr.
2697  * Always consumes mp; never consumes tudr_mp.
2698  */
2699 static int
2700 udp_output_ancillary(conn_t *connp, sin_t *sin, sin6_t *sin6, mblk_t *mp,
2701     mblk_t *tudr_mp, struct nmsghdr *msg, cred_t *cr, pid_t pid)
2702 {
2703 	udp_t		*udp = connp->conn_udp;
2704 	udp_stack_t	*us = udp->udp_us;
2705 	int		error;
2706 	ip_xmit_attr_t	*ixa;
2707 	ip_pkt_t	*ipp;
2708 	in6_addr_t	v6src;
2709 	in6_addr_t	v6dst;
2710 	in6_addr_t	v6nexthop;
2711 	in_port_t	dstport;
2712 	uint32_t	flowinfo;
2713 	uint_t		srcid;
2714 	int		is_absreq_failure = 0;
2715 	conn_opt_arg_t	coas, *coa;
2716 
2717 	ASSERT(tudr_mp != NULL || msg != NULL);
2718 
2719 	/*
2720 	 * Get ixa before checking state to handle a disconnect race.
2721 	 *
2722 	 * We need an exclusive copy of conn_ixa since the ancillary data
2723 	 * options might modify it. That copy has no pointers hence we
2724 	 * need to set them up once we've parsed the ancillary data.
2725 	 */
2726 	ixa = conn_get_ixa_exclusive(connp);
2727 	if (ixa == NULL) {
2728 		UDPS_BUMP_MIB(us, udpOutErrors);
2729 		freemsg(mp);
2730 		return (ENOMEM);
2731 	}
2732 	ASSERT(cr != NULL);
2733 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
2734 	ixa->ixa_cred = cr;
2735 	ixa->ixa_cpid = pid;
2736 	if (is_system_labeled()) {
2737 		/* We need to restart with a label based on the cred */
2738 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
2739 	}
2740 
2741 	/* In case previous destination was multicast or multirt */
2742 	ip_attr_newdst(ixa);
2743 
2744 	/* Get a copy of conn_xmit_ipp since the options might change it */
2745 	ipp = kmem_zalloc(sizeof (*ipp), KM_NOSLEEP);
2746 	if (ipp == NULL) {
2747 		ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
2748 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
2749 		ixa->ixa_cpid = connp->conn_cpid;
2750 		ixa_refrele(ixa);
2751 		UDPS_BUMP_MIB(us, udpOutErrors);
2752 		freemsg(mp);
2753 		return (ENOMEM);
2754 	}
2755 	mutex_enter(&connp->conn_lock);
2756 	error = ip_pkt_copy(&connp->conn_xmit_ipp, ipp, KM_NOSLEEP);
2757 	mutex_exit(&connp->conn_lock);
2758 	if (error != 0) {
2759 		UDPS_BUMP_MIB(us, udpOutErrors);
2760 		freemsg(mp);
2761 		goto done;
2762 	}
2763 
2764 	/*
2765 	 * Parse the options and update ixa and ipp as a result.
2766 	 * Note that ixa_tsl can be updated if SCM_UCRED.
2767 	 * ixa_refrele/ixa_inactivate will release any reference on ixa_tsl.
2768 	 */
2769 
2770 	coa = &coas;
2771 	coa->coa_connp = connp;
2772 	coa->coa_ixa = ixa;
2773 	coa->coa_ipp = ipp;
2774 	coa->coa_ancillary = B_TRUE;
2775 	coa->coa_changed = 0;
2776 
2777 	if (msg != NULL) {
2778 		error = process_auxiliary_options(connp, msg->msg_control,
2779 		    msg->msg_controllen, coa, &udp_opt_obj, udp_opt_set, cr);
2780 	} else {
2781 		struct T_unitdata_req *tudr;
2782 
2783 		tudr = (struct T_unitdata_req *)tudr_mp->b_rptr;
2784 		ASSERT(tudr->PRIM_type == T_UNITDATA_REQ);
2785 		error = tpi_optcom_buf(connp->conn_wq, tudr_mp,
2786 		    &tudr->OPT_length, tudr->OPT_offset, cr, &udp_opt_obj,
2787 		    coa, &is_absreq_failure);
2788 	}
2789 	if (error != 0) {
2790 		/*
2791 		 * Note: No special action needed in this
2792 		 * module for "is_absreq_failure"
2793 		 */
2794 		freemsg(mp);
2795 		UDPS_BUMP_MIB(us, udpOutErrors);
2796 		goto done;
2797 	}
2798 	ASSERT(is_absreq_failure == 0);
2799 
2800 	mutex_enter(&connp->conn_lock);
2801 	/*
2802 	 * If laddr is unspecified then we look at sin6_src_id.
2803 	 * We will give precedence to a source address set with IPV6_PKTINFO
2804 	 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
2805 	 * want ip_attr_connect to select a source (since it can fail) when
2806 	 * IPV6_PKTINFO is specified.
2807 	 * If this doesn't result in a source address then we get a source
2808 	 * from ip_attr_connect() below.
2809 	 */
2810 	v6src = connp->conn_saddr_v6;
2811 	if (sin != NULL) {
2812 		IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
2813 		dstport = sin->sin_port;
2814 		flowinfo = 0;
2815 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
2816 		ixa->ixa_flags |= IXAF_IS_IPV4;
2817 	} else if (sin6 != NULL) {
2818 		boolean_t v4mapped;
2819 
2820 		v6dst = sin6->sin6_addr;
2821 		dstport = sin6->sin6_port;
2822 		flowinfo = sin6->sin6_flowinfo;
2823 		srcid = sin6->__sin6_src_id;
2824 		if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
2825 			ixa->ixa_scopeid = sin6->sin6_scope_id;
2826 			ixa->ixa_flags |= IXAF_SCOPEID_SET;
2827 		} else {
2828 			ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
2829 		}
2830 		v4mapped = IN6_IS_ADDR_V4MAPPED(&v6dst);
2831 		if (v4mapped)
2832 			ixa->ixa_flags |= IXAF_IS_IPV4;
2833 		else
2834 			ixa->ixa_flags &= ~IXAF_IS_IPV4;
2835 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
2836 			if (!ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
2837 			    v4mapped, connp->conn_netstack)) {
2838 				/* Mismatch - v4mapped/v6 specified by srcid. */
2839 				mutex_exit(&connp->conn_lock);
2840 				error = EADDRNOTAVAIL;
2841 				goto failed;	/* Does freemsg() and mib. */
2842 			}
2843 		}
2844 	} else {
2845 		/* Connected case */
2846 		v6dst = connp->conn_faddr_v6;
2847 		dstport = connp->conn_fport;
2848 		flowinfo = connp->conn_flowinfo;
2849 	}
2850 	mutex_exit(&connp->conn_lock);
2851 
2852 	/* Handle IP_PKTINFO/IPV6_PKTINFO setting source address. */
2853 	if (ipp->ipp_fields & IPPF_ADDR) {
2854 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
2855 			if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
2856 				v6src = ipp->ipp_addr;
2857 		} else {
2858 			if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
2859 				v6src = ipp->ipp_addr;
2860 		}
2861 	}
2862 
2863 	ip_attr_nexthop(ipp, ixa, &v6dst, &v6nexthop);
2864 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
2865 	    &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST | IPDF_IPSEC);
2866 
2867 	switch (error) {
2868 	case 0:
2869 		break;
2870 	case EADDRNOTAVAIL:
2871 		/*
2872 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
2873 		 * Don't have the application see that errno
2874 		 */
2875 		error = ENETUNREACH;
2876 		goto failed;
2877 	case ENETDOWN:
2878 		/*
2879 		 * Have !ipif_addr_ready address; drop packet silently
2880 		 * until we can get applications to not send until we
2881 		 * are ready.
2882 		 */
2883 		error = 0;
2884 		goto failed;
2885 	case EHOSTUNREACH:
2886 	case ENETUNREACH:
2887 		if (ixa->ixa_ire != NULL) {
2888 			/*
2889 			 * Let conn_ip_output/ire_send_noroute return
2890 			 * the error and send any local ICMP error.
2891 			 */
2892 			error = 0;
2893 			break;
2894 		}
2895 		/* FALLTHRU */
2896 	default:
2897 	failed:
2898 		freemsg(mp);
2899 		UDPS_BUMP_MIB(us, udpOutErrors);
2900 		goto done;
2901 	}
2902 
2903 	/*
2904 	 * We might be going to a different destination than last time,
2905 	 * thus check that TX allows the communication and compute any
2906 	 * needed label.
2907 	 *
2908 	 * TSOL Note: We have an exclusive ipp and ixa for this thread so we
2909 	 * don't have to worry about concurrent threads.
2910 	 */
2911 	if (is_system_labeled()) {
2912 		/* Using UDP MLP requires SCM_UCRED from user */
2913 		if (connp->conn_mlp_type != mlptSingle &&
2914 		    !((ixa->ixa_flags & IXAF_UCRED_TSL))) {
2915 			UDPS_BUMP_MIB(us, udpOutErrors);
2916 			error = ECONNREFUSED;
2917 			freemsg(mp);
2918 			goto done;
2919 		}
2920 		/*
2921 		 * Check whether Trusted Solaris policy allows communication
2922 		 * with this host, and pretend that the destination is
2923 		 * unreachable if not.
2924 		 * Compute any needed label and place it in ipp_label_v4/v6.
2925 		 *
2926 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
2927 		 * ipp_label_v4/v6 to form the packet.
2928 		 *
2929 		 * Tsol note: We have ipp structure local to this thread so
2930 		 * no locking is needed.
2931 		 */
2932 		error = conn_update_label(connp, ixa, &v6dst, ipp);
2933 		if (error != 0) {
2934 			freemsg(mp);
2935 			UDPS_BUMP_MIB(us, udpOutErrors);
2936 			goto done;
2937 		}
2938 	}
2939 	mp = udp_prepend_hdr(connp, ixa, ipp, &v6src, &v6dst, dstport,
2940 	    flowinfo, mp, &error);
2941 	if (mp == NULL) {
2942 		ASSERT(error != 0);
2943 		UDPS_BUMP_MIB(us, udpOutErrors);
2944 		goto done;
2945 	}
2946 	if (ixa->ixa_pktlen > IP_MAXPACKET) {
2947 		error = EMSGSIZE;
2948 		UDPS_BUMP_MIB(us, udpOutErrors);
2949 		freemsg(mp);
2950 		goto done;
2951 	}
2952 	/* We're done.  Pass the packet to ip. */
2953 	UDPS_BUMP_MIB(us, udpHCOutDatagrams);
2954 
2955 	DTRACE_UDP5(send, mblk_t *, NULL, ip_xmit_attr_t *, ixa,
2956 	    void_ip_t *, mp->b_rptr, udp_t *, udp, udpha_t *,
2957 	    &mp->b_rptr[ixa->ixa_ip_hdr_length]);
2958 
2959 	error = conn_ip_output(mp, ixa);
2960 	/* No udpOutErrors if an error since IP increases its error counter */
2961 	switch (error) {
2962 	case 0:
2963 		break;
2964 	case EWOULDBLOCK:
2965 		(void) ixa_check_drain_insert(connp, ixa);
2966 		error = 0;
2967 		break;
2968 	case EADDRNOTAVAIL:
2969 		/*
2970 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
2971 		 * Don't have the application see that errno
2972 		 */
2973 		error = ENETUNREACH;
2974 		/* FALLTHRU */
2975 	default:
2976 		mutex_enter(&connp->conn_lock);
2977 		/*
2978 		 * Clear the source and v6lastdst so we call ip_attr_connect
2979 		 * for the next packet and try to pick a better source.
2980 		 */
2981 		if (connp->conn_mcbc_bind)
2982 			connp->conn_saddr_v6 = ipv6_all_zeros;
2983 		else
2984 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
2985 		connp->conn_v6lastdst = ipv6_all_zeros;
2986 		mutex_exit(&connp->conn_lock);
2987 		break;
2988 	}
2989 done:
2990 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
2991 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
2992 	ixa->ixa_cpid = connp->conn_cpid;
2993 	ixa_refrele(ixa);
2994 	ip_pkt_free(ipp);
2995 	kmem_free(ipp, sizeof (*ipp));
2996 	return (error);
2997 }
2998 
2999 /*
3000  * Handle sending an M_DATA for a connected socket.
3001  * Handles both IPv4 and IPv6.
3002  */
3003 static int
3004 udp_output_connected(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid)
3005 {
3006 	udp_t		*udp = connp->conn_udp;
3007 	udp_stack_t	*us = udp->udp_us;
3008 	int		error;
3009 	ip_xmit_attr_t	*ixa;
3010 
3011 	/*
3012 	 * If no other thread is using conn_ixa this just gets a reference to
3013 	 * conn_ixa. Otherwise we get a safe copy of conn_ixa.
3014 	 */
3015 	ixa = conn_get_ixa(connp, B_FALSE);
3016 	if (ixa == NULL) {
3017 		UDPS_BUMP_MIB(us, udpOutErrors);
3018 		freemsg(mp);
3019 		return (ENOMEM);
3020 	}
3021 
3022 	ASSERT(cr != NULL);
3023 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3024 	ixa->ixa_cred = cr;
3025 	ixa->ixa_cpid = pid;
3026 
3027 	mutex_enter(&connp->conn_lock);
3028 	mp = udp_prepend_header_template(connp, ixa, mp, &connp->conn_saddr_v6,
3029 	    connp->conn_fport, connp->conn_flowinfo, &error);
3030 
3031 	if (mp == NULL) {
3032 		ASSERT(error != 0);
3033 		mutex_exit(&connp->conn_lock);
3034 		ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3035 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3036 		ixa->ixa_cpid = connp->conn_cpid;
3037 		ixa_refrele(ixa);
3038 		UDPS_BUMP_MIB(us, udpOutErrors);
3039 		freemsg(mp);
3040 		return (error);
3041 	}
3042 
3043 	/*
3044 	 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
3045 	 * safe copy, then we need to fill in any pointers in it.
3046 	 */
3047 	if (ixa->ixa_ire == NULL) {
3048 		in6_addr_t	faddr, saddr;
3049 		in6_addr_t	nexthop;
3050 		in_port_t	fport;
3051 
3052 		saddr = connp->conn_saddr_v6;
3053 		faddr = connp->conn_faddr_v6;
3054 		fport = connp->conn_fport;
3055 		ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &faddr, &nexthop);
3056 		mutex_exit(&connp->conn_lock);
3057 
3058 		error = ip_attr_connect(connp, ixa, &saddr, &faddr, &nexthop,
3059 		    fport, NULL, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
3060 		    IPDF_IPSEC);
3061 		switch (error) {
3062 		case 0:
3063 			break;
3064 		case EADDRNOTAVAIL:
3065 			/*
3066 			 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3067 			 * Don't have the application see that errno
3068 			 */
3069 			error = ENETUNREACH;
3070 			goto failed;
3071 		case ENETDOWN:
3072 			/*
3073 			 * Have !ipif_addr_ready address; drop packet silently
3074 			 * until we can get applications to not send until we
3075 			 * are ready.
3076 			 */
3077 			error = 0;
3078 			goto failed;
3079 		case EHOSTUNREACH:
3080 		case ENETUNREACH:
3081 			if (ixa->ixa_ire != NULL) {
3082 				/*
3083 				 * Let conn_ip_output/ire_send_noroute return
3084 				 * the error and send any local ICMP error.
3085 				 */
3086 				error = 0;
3087 				break;
3088 			}
3089 			/* FALLTHRU */
3090 		default:
3091 		failed:
3092 			ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3093 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3094 			ixa->ixa_cpid = connp->conn_cpid;
3095 			ixa_refrele(ixa);
3096 			freemsg(mp);
3097 			UDPS_BUMP_MIB(us, udpOutErrors);
3098 			return (error);
3099 		}
3100 	} else {
3101 		/* Done with conn_t */
3102 		mutex_exit(&connp->conn_lock);
3103 	}
3104 	ASSERT(ixa->ixa_ire != NULL);
3105 
3106 	/* We're done.  Pass the packet to ip. */
3107 	UDPS_BUMP_MIB(us, udpHCOutDatagrams);
3108 
3109 	DTRACE_UDP5(send, mblk_t *, NULL, ip_xmit_attr_t *, ixa,
3110 	    void_ip_t *, mp->b_rptr, udp_t *, udp, udpha_t *,
3111 	    &mp->b_rptr[ixa->ixa_ip_hdr_length]);
3112 
3113 	error = conn_ip_output(mp, ixa);
3114 	/* No udpOutErrors if an error since IP increases its error counter */
3115 	switch (error) {
3116 	case 0:
3117 		break;
3118 	case EWOULDBLOCK:
3119 		(void) ixa_check_drain_insert(connp, ixa);
3120 		error = 0;
3121 		break;
3122 	case EADDRNOTAVAIL:
3123 		/*
3124 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3125 		 * Don't have the application see that errno
3126 		 */
3127 		error = ENETUNREACH;
3128 		break;
3129 	}
3130 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3131 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3132 	ixa->ixa_cpid = connp->conn_cpid;
3133 	ixa_refrele(ixa);
3134 	return (error);
3135 }
3136 
3137 /*
3138  * Handle sending an M_DATA to the last destination.
3139  * Handles both IPv4 and IPv6.
3140  *
3141  * NOTE: The caller must hold conn_lock and we drop it here.
3142  */
3143 static int
3144 udp_output_lastdst(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid,
3145     ip_xmit_attr_t *ixa)
3146 {
3147 	udp_t		*udp = connp->conn_udp;
3148 	udp_stack_t	*us = udp->udp_us;
3149 	int		error;
3150 
3151 	ASSERT(MUTEX_HELD(&connp->conn_lock));
3152 	ASSERT(ixa != NULL);
3153 
3154 	ASSERT(cr != NULL);
3155 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3156 	ixa->ixa_cred = cr;
3157 	ixa->ixa_cpid = pid;
3158 
3159 	mp = udp_prepend_header_template(connp, ixa, mp, &connp->conn_v6lastsrc,
3160 	    connp->conn_lastdstport, connp->conn_lastflowinfo, &error);
3161 
3162 	if (mp == NULL) {
3163 		ASSERT(error != 0);
3164 		mutex_exit(&connp->conn_lock);
3165 		ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3166 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3167 		ixa->ixa_cpid = connp->conn_cpid;
3168 		ixa_refrele(ixa);
3169 		UDPS_BUMP_MIB(us, udpOutErrors);
3170 		freemsg(mp);
3171 		return (error);
3172 	}
3173 
3174 	/*
3175 	 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
3176 	 * safe copy, then we need to fill in any pointers in it.
3177 	 */
3178 	if (ixa->ixa_ire == NULL) {
3179 		in6_addr_t	lastdst, lastsrc;
3180 		in6_addr_t	nexthop;
3181 		in_port_t	lastport;
3182 
3183 		lastsrc = connp->conn_v6lastsrc;
3184 		lastdst = connp->conn_v6lastdst;
3185 		lastport = connp->conn_lastdstport;
3186 		ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &lastdst, &nexthop);
3187 		mutex_exit(&connp->conn_lock);
3188 
3189 		error = ip_attr_connect(connp, ixa, &lastsrc, &lastdst,
3190 		    &nexthop, lastport, NULL, NULL, IPDF_ALLOW_MCBC |
3191 		    IPDF_VERIFY_DST | IPDF_IPSEC);
3192 		switch (error) {
3193 		case 0:
3194 			break;
3195 		case EADDRNOTAVAIL:
3196 			/*
3197 			 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3198 			 * Don't have the application see that errno
3199 			 */
3200 			error = ENETUNREACH;
3201 			goto failed;
3202 		case ENETDOWN:
3203 			/*
3204 			 * Have !ipif_addr_ready address; drop packet silently
3205 			 * until we can get applications to not send until we
3206 			 * are ready.
3207 			 */
3208 			error = 0;
3209 			goto failed;
3210 		case EHOSTUNREACH:
3211 		case ENETUNREACH:
3212 			if (ixa->ixa_ire != NULL) {
3213 				/*
3214 				 * Let conn_ip_output/ire_send_noroute return
3215 				 * the error and send any local ICMP error.
3216 				 */
3217 				error = 0;
3218 				break;
3219 			}
3220 			/* FALLTHRU */
3221 		default:
3222 		failed:
3223 			ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3224 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3225 			ixa->ixa_cpid = connp->conn_cpid;
3226 			ixa_refrele(ixa);
3227 			freemsg(mp);
3228 			UDPS_BUMP_MIB(us, udpOutErrors);
3229 			return (error);
3230 		}
3231 	} else {
3232 		/* Done with conn_t */
3233 		mutex_exit(&connp->conn_lock);
3234 	}
3235 
3236 	/* We're done.  Pass the packet to ip. */
3237 	UDPS_BUMP_MIB(us, udpHCOutDatagrams);
3238 
3239 	DTRACE_UDP5(send, mblk_t *, NULL, ip_xmit_attr_t *, ixa,
3240 	    void_ip_t *, mp->b_rptr, udp_t *, udp, udpha_t *,
3241 	    &mp->b_rptr[ixa->ixa_ip_hdr_length]);
3242 
3243 	error = conn_ip_output(mp, ixa);
3244 	/* No udpOutErrors if an error since IP increases its error counter */
3245 	switch (error) {
3246 	case 0:
3247 		break;
3248 	case EWOULDBLOCK:
3249 		(void) ixa_check_drain_insert(connp, ixa);
3250 		error = 0;
3251 		break;
3252 	case EADDRNOTAVAIL:
3253 		/*
3254 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3255 		 * Don't have the application see that errno
3256 		 */
3257 		error = ENETUNREACH;
3258 		/* FALLTHRU */
3259 	default:
3260 		mutex_enter(&connp->conn_lock);
3261 		/*
3262 		 * Clear the source and v6lastdst so we call ip_attr_connect
3263 		 * for the next packet and try to pick a better source.
3264 		 */
3265 		if (connp->conn_mcbc_bind)
3266 			connp->conn_saddr_v6 = ipv6_all_zeros;
3267 		else
3268 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
3269 		connp->conn_v6lastdst = ipv6_all_zeros;
3270 		mutex_exit(&connp->conn_lock);
3271 		break;
3272 	}
3273 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3274 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3275 	ixa->ixa_cpid = connp->conn_cpid;
3276 	ixa_refrele(ixa);
3277 	return (error);
3278 }
3279 
3280 
3281 /*
3282  * Prepend the header template and then fill in the source and
3283  * flowinfo. The caller needs to handle the destination address since
3284  * it's setting is different if rthdr or source route.
3285  *
3286  * Returns NULL is allocation failed or if the packet would exceed IP_MAXPACKET.
3287  * When it returns NULL it sets errorp.
3288  */
3289 static mblk_t *
3290 udp_prepend_header_template(conn_t *connp, ip_xmit_attr_t *ixa, mblk_t *mp,
3291     const in6_addr_t *v6src, in_port_t dstport, uint32_t flowinfo, int *errorp)
3292 {
3293 	udp_t		*udp = connp->conn_udp;
3294 	udp_stack_t	*us = udp->udp_us;
3295 	boolean_t	insert_spi = udp->udp_nat_t_endpoint;
3296 	uint_t		pktlen;
3297 	uint_t		alloclen;
3298 	uint_t		copylen;
3299 	uint8_t		*iph;
3300 	uint_t		ip_hdr_length;
3301 	udpha_t		*udpha;
3302 	uint32_t	cksum;
3303 	ip_pkt_t	*ipp;
3304 
3305 	ASSERT(MUTEX_HELD(&connp->conn_lock));
3306 
3307 	/*
3308 	 * Copy the header template and leave space for an SPI
3309 	 */
3310 	copylen = connp->conn_ht_iphc_len;
3311 	alloclen = copylen + (insert_spi ? sizeof (uint32_t) : 0);
3312 	pktlen = alloclen + msgdsize(mp);
3313 	if (pktlen > IP_MAXPACKET) {
3314 		freemsg(mp);
3315 		*errorp = EMSGSIZE;
3316 		return (NULL);
3317 	}
3318 	ixa->ixa_pktlen = pktlen;
3319 
3320 	/* check/fix buffer config, setup pointers into it */
3321 	iph = mp->b_rptr - alloclen;
3322 	if (DB_REF(mp) != 1 || iph < DB_BASE(mp) || !OK_32PTR(iph)) {
3323 		mblk_t *mp1;
3324 
3325 		mp1 = allocb(alloclen + us->us_wroff_extra, BPRI_MED);
3326 		if (mp1 == NULL) {
3327 			freemsg(mp);
3328 			*errorp = ENOMEM;
3329 			return (NULL);
3330 		}
3331 		mp1->b_wptr = DB_LIM(mp1);
3332 		mp1->b_cont = mp;
3333 		mp = mp1;
3334 		iph = (mp->b_wptr - alloclen);
3335 	}
3336 	mp->b_rptr = iph;
3337 	bcopy(connp->conn_ht_iphc, iph, copylen);
3338 	ip_hdr_length = (uint_t)(connp->conn_ht_ulp - connp->conn_ht_iphc);
3339 
3340 	ixa->ixa_ip_hdr_length = ip_hdr_length;
3341 	udpha = (udpha_t *)(iph + ip_hdr_length);
3342 
3343 	/*
3344 	 * Setup header length and prepare for ULP checksum done in IP.
3345 	 * udp_build_hdr_template has already massaged any routing header
3346 	 * and placed the result in conn_sum.
3347 	 *
3348 	 * We make it easy for IP to include our pseudo header
3349 	 * by putting our length in uha_checksum.
3350 	 */
3351 	cksum = pktlen - ip_hdr_length;
3352 	udpha->uha_length = htons(cksum);
3353 
3354 	cksum += connp->conn_sum;
3355 	cksum = (cksum >> 16) + (cksum & 0xFFFF);
3356 	ASSERT(cksum < 0x10000);
3357 
3358 	ipp = &connp->conn_xmit_ipp;
3359 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
3360 		ipha_t	*ipha = (ipha_t *)iph;
3361 
3362 		ipha->ipha_length = htons((uint16_t)pktlen);
3363 
3364 		/* IP does the checksum if uha_checksum is non-zero */
3365 		if (us->us_do_checksum)
3366 			udpha->uha_checksum = htons(cksum);
3367 
3368 		/* if IP_PKTINFO specified an addres it wins over bind() */
3369 		if ((ipp->ipp_fields & IPPF_ADDR) &&
3370 		    IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
3371 			ASSERT(ipp->ipp_addr_v4 != INADDR_ANY);
3372 			ipha->ipha_src = ipp->ipp_addr_v4;
3373 		} else {
3374 			IN6_V4MAPPED_TO_IPADDR(v6src, ipha->ipha_src);
3375 		}
3376 	} else {
3377 		ip6_t *ip6h = (ip6_t *)iph;
3378 
3379 		ip6h->ip6_plen =  htons((uint16_t)(pktlen - IPV6_HDR_LEN));
3380 		udpha->uha_checksum = htons(cksum);
3381 
3382 		/* if IP_PKTINFO specified an addres it wins over bind() */
3383 		if ((ipp->ipp_fields & IPPF_ADDR) &&
3384 		    !IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
3385 			ASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ipp->ipp_addr));
3386 			ip6h->ip6_src = ipp->ipp_addr;
3387 		} else {
3388 			ip6h->ip6_src = *v6src;
3389 		}
3390 		ip6h->ip6_vcf =
3391 		    (IPV6_DEFAULT_VERS_AND_FLOW & IPV6_VERS_AND_FLOW_MASK) |
3392 		    (flowinfo & ~IPV6_VERS_AND_FLOW_MASK);
3393 		if (ipp->ipp_fields & IPPF_TCLASS) {
3394 			/* Overrides the class part of flowinfo */
3395 			ip6h->ip6_vcf = IPV6_TCLASS_FLOW(ip6h->ip6_vcf,
3396 			    ipp->ipp_tclass);
3397 		}
3398 	}
3399 
3400 	/* Insert all-0s SPI now. */
3401 	if (insert_spi)
3402 		*((uint32_t *)(udpha + 1)) = 0;
3403 
3404 	udpha->uha_dst_port = dstport;
3405 	return (mp);
3406 }
3407 
3408 /*
3409  * Send a T_UDERR_IND in response to an M_DATA
3410  */
3411 static void
3412 udp_ud_err_connected(conn_t *connp, t_scalar_t error)
3413 {
3414 	struct sockaddr_storage ss;
3415 	sin_t		*sin;
3416 	sin6_t		*sin6;
3417 	struct sockaddr	*addr;
3418 	socklen_t	addrlen;
3419 	mblk_t		*mp1;
3420 
3421 	mutex_enter(&connp->conn_lock);
3422 	/* Initialize addr and addrlen as if they're passed in */
3423 	if (connp->conn_family == AF_INET) {
3424 		sin = (sin_t *)&ss;
3425 		*sin = sin_null;
3426 		sin->sin_family = AF_INET;
3427 		sin->sin_port = connp->conn_fport;
3428 		sin->sin_addr.s_addr = connp->conn_faddr_v4;
3429 		addr = (struct sockaddr *)sin;
3430 		addrlen = sizeof (*sin);
3431 	} else {
3432 		sin6 = (sin6_t *)&ss;
3433 		*sin6 = sin6_null;
3434 		sin6->sin6_family = AF_INET6;
3435 		sin6->sin6_port = connp->conn_fport;
3436 		sin6->sin6_flowinfo = connp->conn_flowinfo;
3437 		sin6->sin6_addr = connp->conn_faddr_v6;
3438 		if (IN6_IS_ADDR_LINKSCOPE(&connp->conn_faddr_v6) &&
3439 		    (connp->conn_ixa->ixa_flags & IXAF_SCOPEID_SET)) {
3440 			sin6->sin6_scope_id = connp->conn_ixa->ixa_scopeid;
3441 		} else {
3442 			sin6->sin6_scope_id = 0;
3443 		}
3444 		sin6->__sin6_src_id = 0;
3445 		addr = (struct sockaddr *)sin6;
3446 		addrlen = sizeof (*sin6);
3447 	}
3448 	mutex_exit(&connp->conn_lock);
3449 
3450 	mp1 = mi_tpi_uderror_ind((char *)addr, addrlen, NULL, 0, error);
3451 	if (mp1 != NULL)
3452 		putnext(connp->conn_rq, mp1);
3453 }
3454 
3455 /*
3456  * This routine handles all messages passed downstream.  It either
3457  * consumes the message or passes it downstream; it never queues a
3458  * a message.
3459  *
3460  * Also entry point for sockfs when udp is in "direct sockfs" mode.  This mode
3461  * is valid when we are directly beneath the stream head, and thus sockfs
3462  * is able to bypass STREAMS and directly call us, passing along the sockaddr
3463  * structure without the cumbersome T_UNITDATA_REQ interface for the case of
3464  * connected endpoints.
3465  */
3466 void
3467 udp_wput(queue_t *q, mblk_t *mp)
3468 {
3469 	sin6_t		*sin6;
3470 	sin_t		*sin = NULL;
3471 	uint_t		srcid;
3472 	conn_t		*connp = Q_TO_CONN(q);
3473 	udp_t		*udp = connp->conn_udp;
3474 	int		error = 0;
3475 	struct sockaddr	*addr = NULL;
3476 	socklen_t	addrlen;
3477 	udp_stack_t	*us = udp->udp_us;
3478 	struct T_unitdata_req *tudr;
3479 	mblk_t		*data_mp;
3480 	ushort_t	ipversion;
3481 	cred_t		*cr;
3482 	pid_t		pid;
3483 
3484 	/*
3485 	 * We directly handle several cases here: T_UNITDATA_REQ message
3486 	 * coming down as M_PROTO/M_PCPROTO and M_DATA messages for connected
3487 	 * socket.
3488 	 */
3489 	switch (DB_TYPE(mp)) {
3490 	case M_DATA:
3491 		if (!udp->udp_issocket || udp->udp_state != TS_DATA_XFER) {
3492 			/* Not connected; address is required */
3493 			UDPS_BUMP_MIB(us, udpOutErrors);
3494 			UDP_DBGSTAT(us, udp_data_notconn);
3495 			UDP_STAT(us, udp_out_err_notconn);
3496 			freemsg(mp);
3497 			return;
3498 		}
3499 		/*
3500 		 * All Solaris components should pass a db_credp
3501 		 * for this message, hence we ASSERT.
3502 		 * On production kernels we return an error to be robust against
3503 		 * random streams modules sitting on top of us.
3504 		 */
3505 		cr = msg_getcred(mp, &pid);
3506 		ASSERT(cr != NULL);
3507 		if (cr == NULL) {
3508 			UDPS_BUMP_MIB(us, udpOutErrors);
3509 			freemsg(mp);
3510 			return;
3511 		}
3512 		ASSERT(udp->udp_issocket);
3513 		UDP_DBGSTAT(us, udp_data_conn);
3514 		error = udp_output_connected(connp, mp, cr, pid);
3515 		if (error != 0) {
3516 			UDP_STAT(us, udp_out_err_output);
3517 			if (connp->conn_rq != NULL)
3518 				udp_ud_err_connected(connp, (t_scalar_t)error);
3519 #ifdef DEBUG
3520 			printf("udp_output_connected returned %d\n", error);
3521 #endif
3522 		}
3523 		return;
3524 
3525 	case M_PROTO:
3526 	case M_PCPROTO:
3527 		tudr = (struct T_unitdata_req *)mp->b_rptr;
3528 		if (MBLKL(mp) < sizeof (*tudr) ||
3529 		    ((t_primp_t)mp->b_rptr)->type != T_UNITDATA_REQ) {
3530 			udp_wput_other(q, mp);
3531 			return;
3532 		}
3533 		break;
3534 
3535 	default:
3536 		udp_wput_other(q, mp);
3537 		return;
3538 	}
3539 
3540 	/* Handle valid T_UNITDATA_REQ here */
3541 	data_mp = mp->b_cont;
3542 	if (data_mp == NULL) {
3543 		error = EPROTO;
3544 		goto ud_error2;
3545 	}
3546 	mp->b_cont = NULL;
3547 
3548 	if (!MBLKIN(mp, 0, tudr->DEST_offset + tudr->DEST_length)) {
3549 		error = EADDRNOTAVAIL;
3550 		goto ud_error2;
3551 	}
3552 
3553 	/*
3554 	 * All Solaris components should pass a db_credp
3555 	 * for this TPI message, hence we should ASSERT.
3556 	 * However, RPC (svc_clts_ksend) does this odd thing where it
3557 	 * passes the options from a T_UNITDATA_IND unchanged in a
3558 	 * T_UNITDATA_REQ. While that is the right thing to do for
3559 	 * some options, SCM_UCRED being the key one, this also makes it
3560 	 * pass down IP_RECVDSTADDR. Hence we can't ASSERT here.
3561 	 */
3562 	cr = msg_getcred(mp, &pid);
3563 	if (cr == NULL) {
3564 		cr = connp->conn_cred;
3565 		pid = connp->conn_cpid;
3566 	}
3567 
3568 	/*
3569 	 * If a port has not been bound to the stream, fail.
3570 	 * This is not a problem when sockfs is directly
3571 	 * above us, because it will ensure that the socket
3572 	 * is first bound before allowing data to be sent.
3573 	 */
3574 	if (udp->udp_state == TS_UNBND) {
3575 		error = EPROTO;
3576 		goto ud_error2;
3577 	}
3578 	addr = (struct sockaddr *)&mp->b_rptr[tudr->DEST_offset];
3579 	addrlen = tudr->DEST_length;
3580 
3581 	switch (connp->conn_family) {
3582 	case AF_INET6:
3583 		sin6 = (sin6_t *)addr;
3584 		if (!OK_32PTR((char *)sin6) || (addrlen != sizeof (sin6_t)) ||
3585 		    (sin6->sin6_family != AF_INET6)) {
3586 			error = EADDRNOTAVAIL;
3587 			goto ud_error2;
3588 		}
3589 
3590 		srcid = sin6->__sin6_src_id;
3591 		if (!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
3592 			/*
3593 			 * Destination is a non-IPv4-compatible IPv6 address.
3594 			 * Send out an IPv6 format packet.
3595 			 */
3596 
3597 			/*
3598 			 * If the local address is a mapped address return
3599 			 * an error.
3600 			 * It would be possible to send an IPv6 packet but the
3601 			 * response would never make it back to the application
3602 			 * since it is bound to a mapped address.
3603 			 */
3604 			if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
3605 				error = EADDRNOTAVAIL;
3606 				goto ud_error2;
3607 			}
3608 
3609 			UDP_DBGSTAT(us, udp_out_ipv6);
3610 
3611 			if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
3612 				sin6->sin6_addr = ipv6_loopback;
3613 			ipversion = IPV6_VERSION;
3614 		} else {
3615 			if (connp->conn_ipv6_v6only) {
3616 				error = EADDRNOTAVAIL;
3617 				goto ud_error2;
3618 			}
3619 
3620 			/*
3621 			 * If the local address is not zero or a mapped address
3622 			 * return an error.  It would be possible to send an
3623 			 * IPv4 packet but the response would never make it
3624 			 * back to the application since it is bound to a
3625 			 * non-mapped address.
3626 			 */
3627 			if (!IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6) &&
3628 			    !IN6_IS_ADDR_UNSPECIFIED(&connp->conn_saddr_v6)) {
3629 				error = EADDRNOTAVAIL;
3630 				goto ud_error2;
3631 			}
3632 			UDP_DBGSTAT(us, udp_out_mapped);
3633 
3634 			if (V4_PART_OF_V6(sin6->sin6_addr) == INADDR_ANY) {
3635 				V4_PART_OF_V6(sin6->sin6_addr) =
3636 				    htonl(INADDR_LOOPBACK);
3637 			}
3638 			ipversion = IPV4_VERSION;
3639 		}
3640 
3641 		if (tudr->OPT_length != 0) {
3642 			/*
3643 			 * If we are connected then the destination needs to be
3644 			 * the same as the connected one.
3645 			 */
3646 			if (udp->udp_state == TS_DATA_XFER &&
3647 			    !conn_same_as_last_v6(connp, sin6)) {
3648 				error = EISCONN;
3649 				goto ud_error2;
3650 			}
3651 			UDP_STAT(us, udp_out_opt);
3652 			error = udp_output_ancillary(connp, NULL, sin6,
3653 			    data_mp, mp, NULL, cr, pid);
3654 		} else {
3655 			ip_xmit_attr_t *ixa;
3656 
3657 			/*
3658 			 * We have to allocate an ip_xmit_attr_t before we grab
3659 			 * conn_lock and we need to hold conn_lock once we've
3660 			 * checked conn_same_as_last_v6 to handle concurrent
3661 			 * send* calls on a socket.
3662 			 */
3663 			ixa = conn_get_ixa(connp, B_FALSE);
3664 			if (ixa == NULL) {
3665 				error = ENOMEM;
3666 				goto ud_error2;
3667 			}
3668 			mutex_enter(&connp->conn_lock);
3669 
3670 			if (conn_same_as_last_v6(connp, sin6) &&
3671 			    connp->conn_lastsrcid == srcid &&
3672 			    ipsec_outbound_policy_current(ixa)) {
3673 				UDP_DBGSTAT(us, udp_out_lastdst);
3674 				/* udp_output_lastdst drops conn_lock */
3675 				error = udp_output_lastdst(connp, data_mp, cr,
3676 				    pid, ixa);
3677 			} else {
3678 				UDP_DBGSTAT(us, udp_out_diffdst);
3679 				/* udp_output_newdst drops conn_lock */
3680 				error = udp_output_newdst(connp, data_mp, NULL,
3681 				    sin6, ipversion, cr, pid, ixa);
3682 			}
3683 			ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
3684 		}
3685 		if (error == 0) {
3686 			freeb(mp);
3687 			return;
3688 		}
3689 		break;
3690 
3691 	case AF_INET:
3692 		sin = (sin_t *)addr;
3693 		if ((!OK_32PTR((char *)sin) || addrlen != sizeof (sin_t)) ||
3694 		    (sin->sin_family != AF_INET)) {
3695 			error = EADDRNOTAVAIL;
3696 			goto ud_error2;
3697 		}
3698 		UDP_DBGSTAT(us, udp_out_ipv4);
3699 		if (sin->sin_addr.s_addr == INADDR_ANY)
3700 			sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
3701 		ipversion = IPV4_VERSION;
3702 
3703 		srcid = 0;
3704 		if (tudr->OPT_length != 0) {
3705 			/*
3706 			 * If we are connected then the destination needs to be
3707 			 * the same as the connected one.
3708 			 */
3709 			if (udp->udp_state == TS_DATA_XFER &&
3710 			    !conn_same_as_last_v4(connp, sin)) {
3711 				error = EISCONN;
3712 				goto ud_error2;
3713 			}
3714 			UDP_STAT(us, udp_out_opt);
3715 			error = udp_output_ancillary(connp, sin, NULL,
3716 			    data_mp, mp, NULL, cr, pid);
3717 		} else {
3718 			ip_xmit_attr_t *ixa;
3719 
3720 			/*
3721 			 * We have to allocate an ip_xmit_attr_t before we grab
3722 			 * conn_lock and we need to hold conn_lock once we've
3723 			 * checked conn_same_as_last_v4 to handle concurrent
3724 			 * send* calls on a socket.
3725 			 */
3726 			ixa = conn_get_ixa(connp, B_FALSE);
3727 			if (ixa == NULL) {
3728 				error = ENOMEM;
3729 				goto ud_error2;
3730 			}
3731 			mutex_enter(&connp->conn_lock);
3732 
3733 			if (conn_same_as_last_v4(connp, sin) &&
3734 			    ipsec_outbound_policy_current(ixa)) {
3735 				UDP_DBGSTAT(us, udp_out_lastdst);
3736 				/* udp_output_lastdst drops conn_lock */
3737 				error = udp_output_lastdst(connp, data_mp, cr,
3738 				    pid, ixa);
3739 			} else {
3740 				UDP_DBGSTAT(us, udp_out_diffdst);
3741 				/* udp_output_newdst drops conn_lock */
3742 				error = udp_output_newdst(connp, data_mp, sin,
3743 				    NULL, ipversion, cr, pid, ixa);
3744 			}
3745 			ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
3746 		}
3747 		if (error == 0) {
3748 			freeb(mp);
3749 			return;
3750 		}
3751 		break;
3752 	}
3753 	UDP_STAT(us, udp_out_err_output);
3754 	ASSERT(mp != NULL);
3755 	/* mp is freed by the following routine */
3756 	udp_ud_err(q, mp, (t_scalar_t)error);
3757 	return;
3758 
3759 ud_error2:
3760 	UDPS_BUMP_MIB(us, udpOutErrors);
3761 	freemsg(data_mp);
3762 	UDP_STAT(us, udp_out_err_output);
3763 	ASSERT(mp != NULL);
3764 	/* mp is freed by the following routine */
3765 	udp_ud_err(q, mp, (t_scalar_t)error);
3766 }
3767 
3768 /*
3769  * Handle the case of the IP address, port, flow label being different
3770  * for both IPv4 and IPv6.
3771  *
3772  * NOTE: The caller must hold conn_lock and we drop it here.
3773  */
3774 static int
3775 udp_output_newdst(conn_t *connp, mblk_t *data_mp, sin_t *sin, sin6_t *sin6,
3776     ushort_t ipversion, cred_t *cr, pid_t pid, ip_xmit_attr_t *ixa)
3777 {
3778 	uint_t		srcid;
3779 	uint32_t	flowinfo;
3780 	udp_t		*udp = connp->conn_udp;
3781 	int		error = 0;
3782 	ip_xmit_attr_t	*oldixa;
3783 	udp_stack_t	*us = udp->udp_us;
3784 	in6_addr_t	v6src;
3785 	in6_addr_t	v6dst;
3786 	in6_addr_t	v6nexthop;
3787 	in_port_t	dstport;
3788 
3789 	ASSERT(MUTEX_HELD(&connp->conn_lock));
3790 	ASSERT(ixa != NULL);
3791 	/*
3792 	 * We hold conn_lock across all the use and modifications of
3793 	 * the conn_lastdst, conn_ixa, and conn_xmit_ipp to ensure that they
3794 	 * stay consistent.
3795 	 */
3796 
3797 	ASSERT(cr != NULL);
3798 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3799 	ixa->ixa_cred = cr;
3800 	ixa->ixa_cpid = pid;
3801 	if (is_system_labeled()) {
3802 		/* We need to restart with a label based on the cred */
3803 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
3804 	}
3805 
3806 	/*
3807 	 * If we are connected then the destination needs to be the
3808 	 * same as the connected one, which is not the case here since we
3809 	 * checked for that above.
3810 	 */
3811 	if (udp->udp_state == TS_DATA_XFER) {
3812 		mutex_exit(&connp->conn_lock);
3813 		error = EISCONN;
3814 		goto ud_error;
3815 	}
3816 
3817 	/* In case previous destination was multicast or multirt */
3818 	ip_attr_newdst(ixa);
3819 
3820 	/*
3821 	 * If laddr is unspecified then we look at sin6_src_id.
3822 	 * We will give precedence to a source address set with IPV6_PKTINFO
3823 	 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
3824 	 * want ip_attr_connect to select a source (since it can fail) when
3825 	 * IPV6_PKTINFO is specified.
3826 	 * If this doesn't result in a source address then we get a source
3827 	 * from ip_attr_connect() below.
3828 	 */
3829 	v6src = connp->conn_saddr_v6;
3830 	if (sin != NULL) {
3831 		IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
3832 		dstport = sin->sin_port;
3833 		flowinfo = 0;
3834 		/* Don't bother with ip_srcid_find_id(), but indicate anyway. */
3835 		srcid = 0;
3836 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3837 		ixa->ixa_flags |= IXAF_IS_IPV4;
3838 	} else {
3839 		boolean_t v4mapped;
3840 
3841 		v6dst = sin6->sin6_addr;
3842 		dstport = sin6->sin6_port;
3843 		flowinfo = sin6->sin6_flowinfo;
3844 		srcid = sin6->__sin6_src_id;
3845 		if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
3846 			ixa->ixa_scopeid = sin6->sin6_scope_id;
3847 			ixa->ixa_flags |= IXAF_SCOPEID_SET;
3848 		} else {
3849 			ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3850 		}
3851 		v4mapped = IN6_IS_ADDR_V4MAPPED(&v6dst);
3852 		if (v4mapped)
3853 			ixa->ixa_flags |= IXAF_IS_IPV4;
3854 		else
3855 			ixa->ixa_flags &= ~IXAF_IS_IPV4;
3856 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
3857 			if (!ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
3858 			    v4mapped, connp->conn_netstack)) {
3859 				/* Mismatched v4mapped/v6 specified by srcid. */
3860 				mutex_exit(&connp->conn_lock);
3861 				error = EADDRNOTAVAIL;
3862 				goto ud_error;
3863 			}
3864 		}
3865 	}
3866 	/* Handle IP_PKTINFO/IPV6_PKTINFO setting source address. */
3867 	if (connp->conn_xmit_ipp.ipp_fields & IPPF_ADDR) {
3868 		ip_pkt_t *ipp = &connp->conn_xmit_ipp;
3869 
3870 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
3871 			if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3872 				v6src = ipp->ipp_addr;
3873 		} else {
3874 			if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3875 				v6src = ipp->ipp_addr;
3876 		}
3877 	}
3878 
3879 	ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &v6dst, &v6nexthop);
3880 	mutex_exit(&connp->conn_lock);
3881 
3882 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
3883 	    &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST | IPDF_IPSEC);
3884 	switch (error) {
3885 	case 0:
3886 		break;
3887 	case EADDRNOTAVAIL:
3888 		/*
3889 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3890 		 * Don't have the application see that errno
3891 		 */
3892 		error = ENETUNREACH;
3893 		goto failed;
3894 	case ENETDOWN:
3895 		/*
3896 		 * Have !ipif_addr_ready address; drop packet silently
3897 		 * until we can get applications to not send until we
3898 		 * are ready.
3899 		 */
3900 		error = 0;
3901 		goto failed;
3902 	case EHOSTUNREACH:
3903 	case ENETUNREACH:
3904 		if (ixa->ixa_ire != NULL) {
3905 			/*
3906 			 * Let conn_ip_output/ire_send_noroute return
3907 			 * the error and send any local ICMP error.
3908 			 */
3909 			error = 0;
3910 			break;
3911 		}
3912 		/* FALLTHRU */
3913 	failed:
3914 	default:
3915 		goto ud_error;
3916 	}
3917 
3918 
3919 	/*
3920 	 * Cluster note: we let the cluster hook know that we are sending to a
3921 	 * new address and/or port.
3922 	 */
3923 	if (cl_inet_connect2 != NULL) {
3924 		CL_INET_UDP_CONNECT(connp, B_TRUE, &v6dst, dstport, error);
3925 		if (error != 0) {
3926 			error = EHOSTUNREACH;
3927 			goto ud_error;
3928 		}
3929 	}
3930 
3931 	mutex_enter(&connp->conn_lock);
3932 	/*
3933 	 * While we dropped the lock some other thread might have connected
3934 	 * this socket. If so we bail out with EISCONN to ensure that the
3935 	 * connecting thread is the one that updates conn_ixa, conn_ht_*
3936 	 * and conn_*last*.
3937 	 */
3938 	if (udp->udp_state == TS_DATA_XFER) {
3939 		mutex_exit(&connp->conn_lock);
3940 		error = EISCONN;
3941 		goto ud_error;
3942 	}
3943 
3944 	/*
3945 	 * We need to rebuild the headers if
3946 	 *  - we are labeling packets (could be different for different
3947 	 *    destinations)
3948 	 *  - we have a source route (or routing header) since we need to
3949 	 *    massage that to get the pseudo-header checksum
3950 	 *  - the IP version is different than the last time
3951 	 *  - a socket option with COA_HEADER_CHANGED has been set which
3952 	 *    set conn_v6lastdst to zero.
3953 	 *
3954 	 * Otherwise the prepend function will just update the src, dst,
3955 	 * dstport, and flow label.
3956 	 */
3957 	if (is_system_labeled()) {
3958 		/* TX MLP requires SCM_UCRED and don't have that here */
3959 		if (connp->conn_mlp_type != mlptSingle) {
3960 			mutex_exit(&connp->conn_lock);
3961 			error = ECONNREFUSED;
3962 			goto ud_error;
3963 		}
3964 		/*
3965 		 * Check whether Trusted Solaris policy allows communication
3966 		 * with this host, and pretend that the destination is
3967 		 * unreachable if not.
3968 		 * Compute any needed label and place it in ipp_label_v4/v6.
3969 		 *
3970 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
3971 		 * ipp_label_v4/v6 to form the packet.
3972 		 *
3973 		 * Tsol note: Since we hold conn_lock we know no other
3974 		 * thread manipulates conn_xmit_ipp.
3975 		 */
3976 		error = conn_update_label(connp, ixa, &v6dst,
3977 		    &connp->conn_xmit_ipp);
3978 		if (error != 0) {
3979 			mutex_exit(&connp->conn_lock);
3980 			goto ud_error;
3981 		}
3982 		/* Rebuild the header template */
3983 		error = udp_build_hdr_template(connp, &v6src, &v6dst, dstport,
3984 		    flowinfo);
3985 		if (error != 0) {
3986 			mutex_exit(&connp->conn_lock);
3987 			goto ud_error;
3988 		}
3989 	} else if ((connp->conn_xmit_ipp.ipp_fields &
3990 	    (IPPF_IPV4_OPTIONS|IPPF_RTHDR)) ||
3991 	    ipversion != connp->conn_lastipversion ||
3992 	    IN6_IS_ADDR_UNSPECIFIED(&connp->conn_v6lastdst)) {
3993 		/* Rebuild the header template */
3994 		error = udp_build_hdr_template(connp, &v6src, &v6dst, dstport,
3995 		    flowinfo);
3996 		if (error != 0) {
3997 			mutex_exit(&connp->conn_lock);
3998 			goto ud_error;
3999 		}
4000 	} else {
4001 		/* Simply update the destination address if no source route */
4002 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
4003 			ipha_t	*ipha = (ipha_t *)connp->conn_ht_iphc;
4004 
4005 			IN6_V4MAPPED_TO_IPADDR(&v6dst, ipha->ipha_dst);
4006 			if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
4007 				ipha->ipha_fragment_offset_and_flags |=
4008 				    IPH_DF_HTONS;
4009 			} else {
4010 				ipha->ipha_fragment_offset_and_flags &=
4011 				    ~IPH_DF_HTONS;
4012 			}
4013 		} else {
4014 			ip6_t *ip6h = (ip6_t *)connp->conn_ht_iphc;
4015 			ip6h->ip6_dst = v6dst;
4016 		}
4017 	}
4018 
4019 	/*
4020 	 * Remember the dst/dstport etc which corresponds to the built header
4021 	 * template and conn_ixa.
4022 	 */
4023 	oldixa = conn_replace_ixa(connp, ixa);
4024 	connp->conn_v6lastdst = v6dst;
4025 	connp->conn_lastipversion = ipversion;
4026 	connp->conn_lastdstport = dstport;
4027 	connp->conn_lastflowinfo = flowinfo;
4028 	connp->conn_lastscopeid = ixa->ixa_scopeid;
4029 	connp->conn_lastsrcid = srcid;
4030 	/* Also remember a source to use together with lastdst */
4031 	connp->conn_v6lastsrc = v6src;
4032 
4033 	data_mp = udp_prepend_header_template(connp, ixa, data_mp, &v6src,
4034 	    dstport, flowinfo, &error);
4035 
4036 	/* Done with conn_t */
4037 	mutex_exit(&connp->conn_lock);
4038 	ixa_refrele(oldixa);
4039 
4040 	if (data_mp == NULL) {
4041 		ASSERT(error != 0);
4042 		goto ud_error;
4043 	}
4044 
4045 	/* We're done.  Pass the packet to ip. */
4046 	UDPS_BUMP_MIB(us, udpHCOutDatagrams);
4047 
4048 	DTRACE_UDP5(send, mblk_t *, NULL, ip_xmit_attr_t *, ixa,
4049 	    void_ip_t *, data_mp->b_rptr, udp_t *, udp, udpha_t *,
4050 	    &data_mp->b_rptr[ixa->ixa_ip_hdr_length]);
4051 
4052 	error = conn_ip_output(data_mp, ixa);
4053 	/* No udpOutErrors if an error since IP increases its error counter */
4054 	switch (error) {
4055 	case 0:
4056 		break;
4057 	case EWOULDBLOCK:
4058 		(void) ixa_check_drain_insert(connp, ixa);
4059 		error = 0;
4060 		break;
4061 	case EADDRNOTAVAIL:
4062 		/*
4063 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
4064 		 * Don't have the application see that errno
4065 		 */
4066 		error = ENETUNREACH;
4067 		/* FALLTHRU */
4068 	default:
4069 		mutex_enter(&connp->conn_lock);
4070 		/*
4071 		 * Clear the source and v6lastdst so we call ip_attr_connect
4072 		 * for the next packet and try to pick a better source.
4073 		 */
4074 		if (connp->conn_mcbc_bind)
4075 			connp->conn_saddr_v6 = ipv6_all_zeros;
4076 		else
4077 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
4078 		connp->conn_v6lastdst = ipv6_all_zeros;
4079 		mutex_exit(&connp->conn_lock);
4080 		break;
4081 	}
4082 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
4083 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
4084 	ixa->ixa_cpid = connp->conn_cpid;
4085 	ixa_refrele(ixa);
4086 	return (error);
4087 
4088 ud_error:
4089 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
4090 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
4091 	ixa->ixa_cpid = connp->conn_cpid;
4092 	ixa_refrele(ixa);
4093 
4094 	freemsg(data_mp);
4095 	UDPS_BUMP_MIB(us, udpOutErrors);
4096 	UDP_STAT(us, udp_out_err_output);
4097 	return (error);
4098 }
4099 
4100 /* ARGSUSED */
4101 static void
4102 udp_wput_fallback(queue_t *wq, mblk_t *mp)
4103 {
4104 #ifdef DEBUG
4105 	cmn_err(CE_CONT, "udp_wput_fallback: Message in fallback \n");
4106 #endif
4107 	freemsg(mp);
4108 }
4109 
4110 
4111 /*
4112  * Handle special out-of-band ioctl requests (see PSARC/2008/265).
4113  */
4114 static void
4115 udp_wput_cmdblk(queue_t *q, mblk_t *mp)
4116 {
4117 	void	*data;
4118 	mblk_t	*datamp = mp->b_cont;
4119 	conn_t	*connp = Q_TO_CONN(q);
4120 	udp_t	*udp = connp->conn_udp;
4121 	cmdblk_t *cmdp = (cmdblk_t *)mp->b_rptr;
4122 
4123 	if (datamp == NULL || MBLKL(datamp) < cmdp->cb_len) {
4124 		cmdp->cb_error = EPROTO;
4125 		qreply(q, mp);
4126 		return;
4127 	}
4128 	data = datamp->b_rptr;
4129 
4130 	mutex_enter(&connp->conn_lock);
4131 	switch (cmdp->cb_cmd) {
4132 	case TI_GETPEERNAME:
4133 		if (udp->udp_state != TS_DATA_XFER)
4134 			cmdp->cb_error = ENOTCONN;
4135 		else
4136 			cmdp->cb_error = conn_getpeername(connp, data,
4137 			    &cmdp->cb_len);
4138 		break;
4139 	case TI_GETMYNAME:
4140 		cmdp->cb_error = conn_getsockname(connp, data, &cmdp->cb_len);
4141 		break;
4142 	default:
4143 		cmdp->cb_error = EINVAL;
4144 		break;
4145 	}
4146 	mutex_exit(&connp->conn_lock);
4147 
4148 	qreply(q, mp);
4149 }
4150 
4151 static void
4152 udp_use_pure_tpi(udp_t *udp)
4153 {
4154 	conn_t	*connp = udp->udp_connp;
4155 
4156 	mutex_enter(&connp->conn_lock);
4157 	udp->udp_issocket = B_FALSE;
4158 	mutex_exit(&connp->conn_lock);
4159 	UDP_STAT(udp->udp_us, udp_sock_fallback);
4160 }
4161 
4162 static void
4163 udp_wput_other(queue_t *q, mblk_t *mp)
4164 {
4165 	uchar_t	*rptr = mp->b_rptr;
4166 	struct iocblk *iocp;
4167 	conn_t	*connp = Q_TO_CONN(q);
4168 	udp_t	*udp = connp->conn_udp;
4169 	cred_t	*cr;
4170 
4171 	switch (mp->b_datap->db_type) {
4172 	case M_CMD:
4173 		udp_wput_cmdblk(q, mp);
4174 		return;
4175 
4176 	case M_PROTO:
4177 	case M_PCPROTO:
4178 		if (mp->b_wptr - rptr < sizeof (t_scalar_t)) {
4179 			/*
4180 			 * If the message does not contain a PRIM_type,
4181 			 * throw it away.
4182 			 */
4183 			freemsg(mp);
4184 			return;
4185 		}
4186 		switch (((t_primp_t)rptr)->type) {
4187 		case T_ADDR_REQ:
4188 			udp_addr_req(q, mp);
4189 			return;
4190 		case O_T_BIND_REQ:
4191 		case T_BIND_REQ:
4192 			udp_tpi_bind(q, mp);
4193 			return;
4194 		case T_CONN_REQ:
4195 			udp_tpi_connect(q, mp);
4196 			return;
4197 		case T_CAPABILITY_REQ:
4198 			udp_capability_req(q, mp);
4199 			return;
4200 		case T_INFO_REQ:
4201 			udp_info_req(q, mp);
4202 			return;
4203 		case T_UNITDATA_REQ:
4204 			/*
4205 			 * If a T_UNITDATA_REQ gets here, the address must
4206 			 * be bad.  Valid T_UNITDATA_REQs are handled
4207 			 * in udp_wput.
4208 			 */
4209 			udp_ud_err(q, mp, EADDRNOTAVAIL);
4210 			return;
4211 		case T_UNBIND_REQ:
4212 			udp_tpi_unbind(q, mp);
4213 			return;
4214 		case T_SVR4_OPTMGMT_REQ:
4215 			/*
4216 			 * All Solaris components should pass a db_credp
4217 			 * for this TPI message, hence we ASSERT.
4218 			 * But in case there is some other M_PROTO that looks
4219 			 * like a TPI message sent by some other kernel
4220 			 * component, we check and return an error.
4221 			 */
4222 			cr = msg_getcred(mp, NULL);
4223 			ASSERT(cr != NULL);
4224 			if (cr == NULL) {
4225 				udp_err_ack(q, mp, TSYSERR, EINVAL);
4226 				return;
4227 			}
4228 			if (!snmpcom_req(q, mp, udp_snmp_set, ip_snmp_get,
4229 			    cr)) {
4230 				svr4_optcom_req(q, mp, cr, &udp_opt_obj);
4231 			}
4232 			return;
4233 
4234 		case T_OPTMGMT_REQ:
4235 			/*
4236 			 * All Solaris components should pass a db_credp
4237 			 * for this TPI message, hence we ASSERT.
4238 			 * But in case there is some other M_PROTO that looks
4239 			 * like a TPI message sent by some other kernel
4240 			 * component, we check and return an error.
4241 			 */
4242 			cr = msg_getcred(mp, NULL);
4243 			ASSERT(cr != NULL);
4244 			if (cr == NULL) {
4245 				udp_err_ack(q, mp, TSYSERR, EINVAL);
4246 				return;
4247 			}
4248 			tpi_optcom_req(q, mp, cr, &udp_opt_obj);
4249 			return;
4250 
4251 		case T_DISCON_REQ:
4252 			udp_tpi_disconnect(q, mp);
4253 			return;
4254 
4255 		/* The following TPI message is not supported by udp. */
4256 		case O_T_CONN_RES:
4257 		case T_CONN_RES:
4258 			udp_err_ack(q, mp, TNOTSUPPORT, 0);
4259 			return;
4260 
4261 		/* The following 3 TPI requests are illegal for udp. */
4262 		case T_DATA_REQ:
4263 		case T_EXDATA_REQ:
4264 		case T_ORDREL_REQ:
4265 			udp_err_ack(q, mp, TNOTSUPPORT, 0);
4266 			return;
4267 		default:
4268 			break;
4269 		}
4270 		break;
4271 	case M_FLUSH:
4272 		if (*rptr & FLUSHW)
4273 			flushq(q, FLUSHDATA);
4274 		break;
4275 	case M_IOCTL:
4276 		iocp = (struct iocblk *)mp->b_rptr;
4277 		switch (iocp->ioc_cmd) {
4278 		case TI_GETPEERNAME:
4279 			if (udp->udp_state != TS_DATA_XFER) {
4280 				/*
4281 				 * If a default destination address has not
4282 				 * been associated with the stream, then we
4283 				 * don't know the peer's name.
4284 				 */
4285 				iocp->ioc_error = ENOTCONN;
4286 				iocp->ioc_count = 0;
4287 				mp->b_datap->db_type = M_IOCACK;
4288 				qreply(q, mp);
4289 				return;
4290 			}
4291 			/* FALLTHRU */
4292 		case TI_GETMYNAME:
4293 			/*
4294 			 * For TI_GETPEERNAME and TI_GETMYNAME, we first
4295 			 * need to copyin the user's strbuf structure.
4296 			 * Processing will continue in the M_IOCDATA case
4297 			 * below.
4298 			 */
4299 			mi_copyin(q, mp, NULL,
4300 			    SIZEOF_STRUCT(strbuf, iocp->ioc_flag));
4301 			return;
4302 		case _SIOCSOCKFALLBACK:
4303 			/*
4304 			 * Either sockmod is about to be popped and the
4305 			 * socket would now be treated as a plain stream,
4306 			 * or a module is about to be pushed so we have
4307 			 * to follow pure TPI semantics.
4308 			 */
4309 			if (!udp->udp_issocket) {
4310 				DB_TYPE(mp) = M_IOCNAK;
4311 				iocp->ioc_error = EINVAL;
4312 			} else {
4313 				udp_use_pure_tpi(udp);
4314 
4315 				DB_TYPE(mp) = M_IOCACK;
4316 				iocp->ioc_error = 0;
4317 			}
4318 			iocp->ioc_count = 0;
4319 			iocp->ioc_rval = 0;
4320 			qreply(q, mp);
4321 			return;
4322 		default:
4323 			break;
4324 		}
4325 		break;
4326 	case M_IOCDATA:
4327 		udp_wput_iocdata(q, mp);
4328 		return;
4329 	default:
4330 		/* Unrecognized messages are passed through without change. */
4331 		break;
4332 	}
4333 	ip_wput_nondata(q, mp);
4334 }
4335 
4336 /*
4337  * udp_wput_iocdata is called by udp_wput_other to handle all M_IOCDATA
4338  * messages.
4339  */
4340 static void
4341 udp_wput_iocdata(queue_t *q, mblk_t *mp)
4342 {
4343 	mblk_t		*mp1;
4344 	struct	iocblk *iocp = (struct iocblk *)mp->b_rptr;
4345 	STRUCT_HANDLE(strbuf, sb);
4346 	uint_t		addrlen;
4347 	conn_t		*connp = Q_TO_CONN(q);
4348 	udp_t		*udp = connp->conn_udp;
4349 
4350 	/* Make sure it is one of ours. */
4351 	switch (iocp->ioc_cmd) {
4352 	case TI_GETMYNAME:
4353 	case TI_GETPEERNAME:
4354 		break;
4355 	default:
4356 		ip_wput_nondata(q, mp);
4357 		return;
4358 	}
4359 
4360 	switch (mi_copy_state(q, mp, &mp1)) {
4361 	case -1:
4362 		return;
4363 	case MI_COPY_CASE(MI_COPY_IN, 1):
4364 		break;
4365 	case MI_COPY_CASE(MI_COPY_OUT, 1):
4366 		/*
4367 		 * The address has been copied out, so now
4368 		 * copyout the strbuf.
4369 		 */
4370 		mi_copyout(q, mp);
4371 		return;
4372 	case MI_COPY_CASE(MI_COPY_OUT, 2):
4373 		/*
4374 		 * The address and strbuf have been copied out.
4375 		 * We're done, so just acknowledge the original
4376 		 * M_IOCTL.
4377 		 */
4378 		mi_copy_done(q, mp, 0);
4379 		return;
4380 	default:
4381 		/*
4382 		 * Something strange has happened, so acknowledge
4383 		 * the original M_IOCTL with an EPROTO error.
4384 		 */
4385 		mi_copy_done(q, mp, EPROTO);
4386 		return;
4387 	}
4388 
4389 	/*
4390 	 * Now we have the strbuf structure for TI_GETMYNAME
4391 	 * and TI_GETPEERNAME.  Next we copyout the requested
4392 	 * address and then we'll copyout the strbuf.
4393 	 */
4394 	STRUCT_SET_HANDLE(sb, iocp->ioc_flag, (void *)mp1->b_rptr);
4395 
4396 	if (connp->conn_family == AF_INET)
4397 		addrlen = sizeof (sin_t);
4398 	else
4399 		addrlen = sizeof (sin6_t);
4400 
4401 	if (STRUCT_FGET(sb, maxlen) < addrlen) {
4402 		mi_copy_done(q, mp, EINVAL);
4403 		return;
4404 	}
4405 
4406 	switch (iocp->ioc_cmd) {
4407 	case TI_GETMYNAME:
4408 		break;
4409 	case TI_GETPEERNAME:
4410 		if (udp->udp_state != TS_DATA_XFER) {
4411 			mi_copy_done(q, mp, ENOTCONN);
4412 			return;
4413 		}
4414 		break;
4415 	}
4416 	mp1 = mi_copyout_alloc(q, mp, STRUCT_FGETP(sb, buf), addrlen, B_TRUE);
4417 	if (!mp1)
4418 		return;
4419 
4420 	STRUCT_FSET(sb, len, addrlen);
4421 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4422 	case TI_GETMYNAME:
4423 		(void) conn_getsockname(connp, (struct sockaddr *)mp1->b_wptr,
4424 		    &addrlen);
4425 		break;
4426 	case TI_GETPEERNAME:
4427 		(void) conn_getpeername(connp, (struct sockaddr *)mp1->b_wptr,
4428 		    &addrlen);
4429 		break;
4430 	}
4431 	mp1->b_wptr += addrlen;
4432 	/* Copy out the address */
4433 	mi_copyout(q, mp);
4434 }
4435 
4436 void
4437 udp_ddi_g_init(void)
4438 {
4439 	udp_max_optsize = optcom_max_optsize(udp_opt_obj.odb_opt_des_arr,
4440 	    udp_opt_obj.odb_opt_arr_cnt);
4441 
4442 	/*
4443 	 * We want to be informed each time a stack is created or
4444 	 * destroyed in the kernel, so we can maintain the
4445 	 * set of udp_stack_t's.
4446 	 */
4447 	netstack_register(NS_UDP, udp_stack_init, NULL, udp_stack_fini);
4448 }
4449 
4450 void
4451 udp_ddi_g_destroy(void)
4452 {
4453 	netstack_unregister(NS_UDP);
4454 }
4455 
4456 #define	INET_NAME	"ip"
4457 
4458 /*
4459  * Initialize the UDP stack instance.
4460  */
4461 static void *
4462 udp_stack_init(netstackid_t stackid, netstack_t *ns)
4463 {
4464 	udp_stack_t	*us;
4465 	int		i;
4466 	int		error = 0;
4467 	major_t		major;
4468 	size_t		arrsz;
4469 
4470 	us = (udp_stack_t *)kmem_zalloc(sizeof (*us), KM_SLEEP);
4471 	us->us_netstack = ns;
4472 
4473 	mutex_init(&us->us_epriv_port_lock, NULL, MUTEX_DEFAULT, NULL);
4474 	us->us_num_epriv_ports = UDP_NUM_EPRIV_PORTS;
4475 	us->us_epriv_ports[0] = ULP_DEF_EPRIV_PORT1;
4476 	us->us_epriv_ports[1] = ULP_DEF_EPRIV_PORT2;
4477 
4478 	/*
4479 	 * The smallest anonymous port in the priviledged port range which UDP
4480 	 * looks for free port.  Use in the option UDP_ANONPRIVBIND.
4481 	 */
4482 	us->us_min_anonpriv_port = 512;
4483 
4484 	us->us_bind_fanout_size = udp_bind_fanout_size;
4485 
4486 	/* Roundup variable that might have been modified in /etc/system */
4487 	if (!ISP2(us->us_bind_fanout_size)) {
4488 		/* Not a power of two. Round up to nearest power of two */
4489 		for (i = 0; i < 31; i++) {
4490 			if (us->us_bind_fanout_size < (1 << i))
4491 				break;
4492 		}
4493 		us->us_bind_fanout_size = 1 << i;
4494 	}
4495 	us->us_bind_fanout = kmem_zalloc(us->us_bind_fanout_size *
4496 	    sizeof (udp_fanout_t), KM_SLEEP);
4497 	for (i = 0; i < us->us_bind_fanout_size; i++) {
4498 		mutex_init(&us->us_bind_fanout[i].uf_lock, NULL, MUTEX_DEFAULT,
4499 		    NULL);
4500 	}
4501 
4502 	arrsz = udp_propinfo_count * sizeof (mod_prop_info_t);
4503 	us->us_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz,
4504 	    KM_SLEEP);
4505 	bcopy(udp_propinfo_tbl, us->us_propinfo_tbl, arrsz);
4506 
4507 	/* Allocate the per netstack stats */
4508 	mutex_enter(&cpu_lock);
4509 	us->us_sc_cnt = MAX(ncpus, boot_ncpus);
4510 	mutex_exit(&cpu_lock);
4511 	us->us_sc = kmem_zalloc(max_ncpus  * sizeof (udp_stats_cpu_t *),
4512 	    KM_SLEEP);
4513 	for (i = 0; i < us->us_sc_cnt; i++) {
4514 		us->us_sc[i] = kmem_zalloc(sizeof (udp_stats_cpu_t),
4515 		    KM_SLEEP);
4516 	}
4517 
4518 	us->us_kstat = udp_kstat2_init(stackid);
4519 	us->us_mibkp = udp_kstat_init(stackid);
4520 
4521 	major = mod_name_to_major(INET_NAME);
4522 	error = ldi_ident_from_major(major, &us->us_ldi_ident);
4523 	ASSERT(error == 0);
4524 	return (us);
4525 }
4526 
4527 /*
4528  * Free the UDP stack instance.
4529  */
4530 static void
4531 udp_stack_fini(netstackid_t stackid, void *arg)
4532 {
4533 	udp_stack_t *us = (udp_stack_t *)arg;
4534 	int i;
4535 
4536 	for (i = 0; i < us->us_bind_fanout_size; i++) {
4537 		mutex_destroy(&us->us_bind_fanout[i].uf_lock);
4538 	}
4539 
4540 	kmem_free(us->us_bind_fanout, us->us_bind_fanout_size *
4541 	    sizeof (udp_fanout_t));
4542 
4543 	us->us_bind_fanout = NULL;
4544 
4545 	for (i = 0; i < us->us_sc_cnt; i++)
4546 		kmem_free(us->us_sc[i], sizeof (udp_stats_cpu_t));
4547 	kmem_free(us->us_sc, max_ncpus * sizeof (udp_stats_cpu_t *));
4548 
4549 	kmem_free(us->us_propinfo_tbl,
4550 	    udp_propinfo_count * sizeof (mod_prop_info_t));
4551 	us->us_propinfo_tbl = NULL;
4552 
4553 	udp_kstat_fini(stackid, us->us_mibkp);
4554 	us->us_mibkp = NULL;
4555 
4556 	udp_kstat2_fini(stackid, us->us_kstat);
4557 	us->us_kstat = NULL;
4558 
4559 	mutex_destroy(&us->us_epriv_port_lock);
4560 	ldi_ident_release(us->us_ldi_ident);
4561 	kmem_free(us, sizeof (*us));
4562 }
4563 
4564 static size_t
4565 udp_set_rcv_hiwat(udp_t *udp, size_t size)
4566 {
4567 	udp_stack_t *us = udp->udp_us;
4568 
4569 	/* We add a bit of extra buffering */
4570 	size += size >> 1;
4571 	if (size > us->us_max_buf)
4572 		size = us->us_max_buf;
4573 
4574 	udp->udp_rcv_hiwat = size;
4575 	return (size);
4576 }
4577 
4578 /*
4579  * For the lower queue so that UDP can be a dummy mux.
4580  * Nobody should be sending
4581  * packets up this stream
4582  */
4583 static void
4584 udp_lrput(queue_t *q, mblk_t *mp)
4585 {
4586 	switch (mp->b_datap->db_type) {
4587 	case M_FLUSH:
4588 		/* Turn around */
4589 		if (*mp->b_rptr & FLUSHW) {
4590 			*mp->b_rptr &= ~FLUSHR;
4591 			qreply(q, mp);
4592 			return;
4593 		}
4594 		break;
4595 	}
4596 	freemsg(mp);
4597 }
4598 
4599 /*
4600  * For the lower queue so that UDP can be a dummy mux.
4601  * Nobody should be sending packets down this stream.
4602  */
4603 /* ARGSUSED */
4604 void
4605 udp_lwput(queue_t *q, mblk_t *mp)
4606 {
4607 	freemsg(mp);
4608 }
4609 
4610 /*
4611  * When a CPU is added, we need to allocate the per CPU stats struct.
4612  */
4613 void
4614 udp_stack_cpu_add(udp_stack_t *us, processorid_t cpu_seqid)
4615 {
4616 	int i;
4617 
4618 	if (cpu_seqid < us->us_sc_cnt)
4619 		return;
4620 	for (i = us->us_sc_cnt; i <= cpu_seqid; i++) {
4621 		ASSERT(us->us_sc[i] == NULL);
4622 		us->us_sc[i] = kmem_zalloc(sizeof (udp_stats_cpu_t),
4623 		    KM_SLEEP);
4624 	}
4625 	membar_producer();
4626 	us->us_sc_cnt = cpu_seqid + 1;
4627 }
4628 
4629 /*
4630  * Below routines for UDP socket module.
4631  */
4632 
4633 static conn_t *
4634 udp_do_open(cred_t *credp, boolean_t isv6, int flags, int *errorp)
4635 {
4636 	udp_t		*udp;
4637 	conn_t		*connp;
4638 	zoneid_t 	zoneid;
4639 	netstack_t 	*ns;
4640 	udp_stack_t 	*us;
4641 	int		len;
4642 
4643 	ASSERT(errorp != NULL);
4644 
4645 	if ((*errorp = secpolicy_basic_net_access(credp)) != 0)
4646 		return (NULL);
4647 
4648 	ns = netstack_find_by_cred(credp);
4649 	ASSERT(ns != NULL);
4650 	us = ns->netstack_udp;
4651 	ASSERT(us != NULL);
4652 
4653 	/*
4654 	 * For exclusive stacks we set the zoneid to zero
4655 	 * to make UDP operate as if in the global zone.
4656 	 */
4657 	if (ns->netstack_stackid != GLOBAL_NETSTACKID)
4658 		zoneid = GLOBAL_ZONEID;
4659 	else
4660 		zoneid = crgetzoneid(credp);
4661 
4662 	ASSERT(flags == KM_SLEEP || flags == KM_NOSLEEP);
4663 
4664 	connp = ipcl_conn_create(IPCL_UDPCONN, flags, ns);
4665 	if (connp == NULL) {
4666 		netstack_rele(ns);
4667 		*errorp = ENOMEM;
4668 		return (NULL);
4669 	}
4670 	udp = connp->conn_udp;
4671 
4672 	/*
4673 	 * ipcl_conn_create did a netstack_hold. Undo the hold that was
4674 	 * done by netstack_find_by_cred()
4675 	 */
4676 	netstack_rele(ns);
4677 
4678 	/*
4679 	 * Since this conn_t/udp_t is not yet visible to anybody else we don't
4680 	 * need to lock anything.
4681 	 */
4682 	ASSERT(connp->conn_proto == IPPROTO_UDP);
4683 	ASSERT(connp->conn_udp == udp);
4684 	ASSERT(udp->udp_connp == connp);
4685 
4686 	/* Set the initial state of the stream and the privilege status. */
4687 	udp->udp_state = TS_UNBND;
4688 	connp->conn_ixa->ixa_flags |= IXAF_VERIFY_SOURCE;
4689 	if (isv6) {
4690 		connp->conn_family = AF_INET6;
4691 		connp->conn_ipversion = IPV6_VERSION;
4692 		connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
4693 		connp->conn_default_ttl = us->us_ipv6_hoplimit;
4694 		len = sizeof (ip6_t) + UDPH_SIZE;
4695 	} else {
4696 		connp->conn_family = AF_INET;
4697 		connp->conn_ipversion = IPV4_VERSION;
4698 		connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
4699 		connp->conn_default_ttl = us->us_ipv4_ttl;
4700 		len = sizeof (ipha_t) + UDPH_SIZE;
4701 	}
4702 
4703 	ASSERT(connp->conn_ixa->ixa_protocol == connp->conn_proto);
4704 	connp->conn_xmit_ipp.ipp_unicast_hops = connp->conn_default_ttl;
4705 
4706 	connp->conn_ixa->ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
4707 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
4708 	/* conn_allzones can not be set this early, hence no IPCL_ZONEID */
4709 	connp->conn_ixa->ixa_zoneid = zoneid;
4710 
4711 	connp->conn_zoneid = zoneid;
4712 
4713 	/*
4714 	 * If the caller has the process-wide flag set, then default to MAC
4715 	 * exempt mode.  This allows read-down to unlabeled hosts.
4716 	 */
4717 	if (getpflags(NET_MAC_AWARE, credp) != 0)
4718 		connp->conn_mac_mode = CONN_MAC_AWARE;
4719 
4720 	connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
4721 
4722 	udp->udp_us = us;
4723 
4724 	connp->conn_rcvbuf = us->us_recv_hiwat;
4725 	connp->conn_sndbuf = us->us_xmit_hiwat;
4726 	connp->conn_sndlowat = us->us_xmit_lowat;
4727 	connp->conn_rcvlowat = udp_mod_info.mi_lowat;
4728 
4729 	connp->conn_wroff = len + us->us_wroff_extra;
4730 	connp->conn_so_type = SOCK_DGRAM;
4731 
4732 	connp->conn_recv = udp_input;
4733 	connp->conn_recvicmp = udp_icmp_input;
4734 	crhold(credp);
4735 	connp->conn_cred = credp;
4736 	connp->conn_cpid = curproc->p_pid;
4737 	connp->conn_open_time = ddi_get_lbolt64();
4738 	/* Cache things in ixa without an extra refhold */
4739 	ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
4740 	connp->conn_ixa->ixa_cred = connp->conn_cred;
4741 	connp->conn_ixa->ixa_cpid = connp->conn_cpid;
4742 	if (is_system_labeled())
4743 		connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
4744 
4745 	*((sin6_t *)&udp->udp_delayed_addr) = sin6_null;
4746 
4747 	if (us->us_pmtu_discovery)
4748 		connp->conn_ixa->ixa_flags |= IXAF_PMTU_DISCOVERY;
4749 
4750 	return (connp);
4751 }
4752 
4753 sock_lower_handle_t
4754 udp_create(int family, int type, int proto, sock_downcalls_t **sock_downcalls,
4755     uint_t *smodep, int *errorp, int flags, cred_t *credp)
4756 {
4757 	udp_t		*udp = NULL;
4758 	udp_stack_t	*us;
4759 	conn_t		*connp;
4760 	boolean_t	isv6;
4761 
4762 	if (type != SOCK_DGRAM || (family != AF_INET && family != AF_INET6) ||
4763 	    (proto != 0 && proto != IPPROTO_UDP)) {
4764 		*errorp = EPROTONOSUPPORT;
4765 		return (NULL);
4766 	}
4767 
4768 	if (family == AF_INET6)
4769 		isv6 = B_TRUE;
4770 	else
4771 		isv6 = B_FALSE;
4772 
4773 	connp = udp_do_open(credp, isv6, flags, errorp);
4774 	if (connp == NULL)
4775 		return (NULL);
4776 
4777 	udp = connp->conn_udp;
4778 	ASSERT(udp != NULL);
4779 	us = udp->udp_us;
4780 	ASSERT(us != NULL);
4781 
4782 	udp->udp_issocket = B_TRUE;
4783 	connp->conn_flags |= IPCL_NONSTR;
4784 
4785 	/*
4786 	 * Set flow control
4787 	 * Since this conn_t/udp_t is not yet visible to anybody else we don't
4788 	 * need to lock anything.
4789 	 */
4790 	(void) udp_set_rcv_hiwat(udp, connp->conn_rcvbuf);
4791 	udp->udp_rcv_disply_hiwat = connp->conn_rcvbuf;
4792 
4793 	connp->conn_flow_cntrld = B_FALSE;
4794 
4795 	mutex_enter(&connp->conn_lock);
4796 	connp->conn_state_flags &= ~CONN_INCIPIENT;
4797 	mutex_exit(&connp->conn_lock);
4798 
4799 	*errorp = 0;
4800 	*smodep = SM_ATOMIC;
4801 	*sock_downcalls = &sock_udp_downcalls;
4802 	return ((sock_lower_handle_t)connp);
4803 }
4804 
4805 /* ARGSUSED3 */
4806 void
4807 udp_activate(sock_lower_handle_t proto_handle, sock_upper_handle_t sock_handle,
4808     sock_upcalls_t *sock_upcalls, int flags, cred_t *cr)
4809 {
4810 	conn_t 		*connp = (conn_t *)proto_handle;
4811 	struct sock_proto_props sopp;
4812 
4813 	/* All Solaris components should pass a cred for this operation. */
4814 	ASSERT(cr != NULL);
4815 
4816 	connp->conn_upcalls = sock_upcalls;
4817 	connp->conn_upper_handle = sock_handle;
4818 
4819 	sopp.sopp_flags = SOCKOPT_WROFF | SOCKOPT_RCVHIWAT | SOCKOPT_RCVLOWAT |
4820 	    SOCKOPT_MAXBLK | SOCKOPT_MAXPSZ | SOCKOPT_MINPSZ;
4821 	sopp.sopp_wroff = connp->conn_wroff;
4822 	sopp.sopp_maxblk = INFPSZ;
4823 	sopp.sopp_rxhiwat = connp->conn_rcvbuf;
4824 	sopp.sopp_rxlowat = connp->conn_rcvlowat;
4825 	sopp.sopp_maxaddrlen = sizeof (sin6_t);
4826 	sopp.sopp_maxpsz =
4827 	    (connp->conn_family == AF_INET) ? UDP_MAXPACKET_IPV4 :
4828 	    UDP_MAXPACKET_IPV6;
4829 	sopp.sopp_minpsz = (udp_mod_info.mi_minpsz == 1) ? 0 :
4830 	    udp_mod_info.mi_minpsz;
4831 
4832 	(*connp->conn_upcalls->su_set_proto_props)(connp->conn_upper_handle,
4833 	    &sopp);
4834 }
4835 
4836 static void
4837 udp_do_close(conn_t *connp)
4838 {
4839 	udp_t	*udp;
4840 
4841 	ASSERT(connp != NULL && IPCL_IS_UDP(connp));
4842 	udp = connp->conn_udp;
4843 
4844 	if (cl_inet_unbind != NULL && udp->udp_state == TS_IDLE) {
4845 		/*
4846 		 * Running in cluster mode - register unbind information
4847 		 */
4848 		if (connp->conn_ipversion == IPV4_VERSION) {
4849 			(*cl_inet_unbind)(
4850 			    connp->conn_netstack->netstack_stackid,
4851 			    IPPROTO_UDP, AF_INET,
4852 			    (uint8_t *)(&V4_PART_OF_V6(connp->conn_laddr_v6)),
4853 			    (in_port_t)connp->conn_lport, NULL);
4854 		} else {
4855 			(*cl_inet_unbind)(
4856 			    connp->conn_netstack->netstack_stackid,
4857 			    IPPROTO_UDP, AF_INET6,
4858 			    (uint8_t *)&(connp->conn_laddr_v6),
4859 			    (in_port_t)connp->conn_lport, NULL);
4860 		}
4861 	}
4862 
4863 	mutex_enter(&connp->conn_lock);
4864 	udp_bind_hash_remove(udp, B_FALSE);
4865 	mutex_exit(&connp->conn_lock);
4866 
4867 	ip_quiesce_conn(connp);
4868 
4869 	if (!IPCL_IS_NONSTR(connp)) {
4870 		ASSERT(connp->conn_wq != NULL);
4871 		ASSERT(connp->conn_rq != NULL);
4872 		qprocsoff(connp->conn_rq);
4873 	}
4874 
4875 	udp_close_free(connp);
4876 
4877 	/*
4878 	 * Now we are truly single threaded on this stream, and can
4879 	 * delete the things hanging off the connp, and finally the connp.
4880 	 * We removed this connp from the fanout list, it cannot be
4881 	 * accessed thru the fanouts, and we already waited for the
4882 	 * conn_ref to drop to 0. We are already in close, so
4883 	 * there cannot be any other thread from the top. qprocsoff
4884 	 * has completed, and service has completed or won't run in
4885 	 * future.
4886 	 */
4887 	ASSERT(connp->conn_ref == 1);
4888 
4889 	if (!IPCL_IS_NONSTR(connp)) {
4890 		inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4891 	} else {
4892 		ip_free_helper_stream(connp);
4893 	}
4894 
4895 	connp->conn_ref--;
4896 	ipcl_conn_destroy(connp);
4897 }
4898 
4899 /* ARGSUSED1 */
4900 int
4901 udp_close(sock_lower_handle_t proto_handle, int flags, cred_t *cr)
4902 {
4903 	conn_t	*connp = (conn_t *)proto_handle;
4904 
4905 	/* All Solaris components should pass a cred for this operation. */
4906 	ASSERT(cr != NULL);
4907 
4908 	udp_do_close(connp);
4909 	return (0);
4910 }
4911 
4912 static int
4913 udp_do_bind(conn_t *connp, struct sockaddr *sa, socklen_t len, cred_t *cr,
4914     boolean_t bind_to_req_port_only)
4915 {
4916 	sin_t		*sin;
4917 	sin6_t		*sin6;
4918 	udp_t		*udp = connp->conn_udp;
4919 	int		error = 0;
4920 	ip_laddr_t	laddr_type = IPVL_UNICAST_UP;	/* INADDR_ANY */
4921 	in_port_t	port;		/* Host byte order */
4922 	in_port_t	requested_port;	/* Host byte order */
4923 	int		count;
4924 	ipaddr_t	v4src;		/* Set if AF_INET */
4925 	in6_addr_t	v6src;
4926 	int		loopmax;
4927 	udp_fanout_t	*udpf;
4928 	in_port_t	lport;		/* Network byte order */
4929 	uint_t		scopeid = 0;
4930 	zoneid_t	zoneid = IPCL_ZONEID(connp);
4931 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
4932 	boolean_t	is_inaddr_any;
4933 	mlp_type_t	addrtype, mlptype;
4934 	udp_stack_t	*us = udp->udp_us;
4935 	struct reuselist *reusep;
4936 
4937 	switch (len) {
4938 	case sizeof (sin_t):	/* Complete IPv4 address */
4939 		sin = (sin_t *)sa;
4940 
4941 		if (sin == NULL || !OK_32PTR((char *)sin))
4942 			return (EINVAL);
4943 
4944 		if (connp->conn_family != AF_INET ||
4945 		    sin->sin_family != AF_INET) {
4946 			return (EAFNOSUPPORT);
4947 		}
4948 		v4src = sin->sin_addr.s_addr;
4949 		IN6_IPADDR_TO_V4MAPPED(v4src, &v6src);
4950 		if (v4src != INADDR_ANY) {
4951 			laddr_type = ip_laddr_verify_v4(v4src, zoneid, ipst,
4952 			    B_TRUE);
4953 		}
4954 		port = ntohs(sin->sin_port);
4955 		break;
4956 
4957 	case sizeof (sin6_t):	/* complete IPv6 address */
4958 		sin6 = (sin6_t *)sa;
4959 
4960 		if (sin6 == NULL || !OK_32PTR((char *)sin6))
4961 			return (EINVAL);
4962 
4963 		if (connp->conn_family != AF_INET6 ||
4964 		    sin6->sin6_family != AF_INET6) {
4965 			return (EAFNOSUPPORT);
4966 		}
4967 		v6src = sin6->sin6_addr;
4968 		if (IN6_IS_ADDR_V4MAPPED(&v6src)) {
4969 			if (connp->conn_ipv6_v6only)
4970 				return (EADDRNOTAVAIL);
4971 
4972 			IN6_V4MAPPED_TO_IPADDR(&v6src, v4src);
4973 			if (v4src != INADDR_ANY) {
4974 				laddr_type = ip_laddr_verify_v4(v4src,
4975 				    zoneid, ipst, B_FALSE);
4976 			}
4977 		} else {
4978 			if (!IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
4979 				if (IN6_IS_ADDR_LINKSCOPE(&v6src))
4980 					scopeid = sin6->sin6_scope_id;
4981 				laddr_type = ip_laddr_verify_v6(&v6src,
4982 				    zoneid, ipst, B_TRUE, scopeid);
4983 			}
4984 		}
4985 		port = ntohs(sin6->sin6_port);
4986 		break;
4987 
4988 	default:		/* Invalid request */
4989 		(void) strlog(UDP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
4990 		    "udp_bind: bad ADDR_length length %u", len);
4991 		return (-TBADADDR);
4992 	}
4993 
4994 	/* Is the local address a valid unicast, multicast, or broadcast? */
4995 	if (laddr_type == IPVL_BAD)
4996 		return (EADDRNOTAVAIL);
4997 
4998 	requested_port = port;
4999 
5000 	if (requested_port == 0 || !bind_to_req_port_only)
5001 		bind_to_req_port_only = B_FALSE;
5002 	else		/* T_BIND_REQ and requested_port != 0 */
5003 		bind_to_req_port_only = B_TRUE;
5004 
5005 	if (requested_port == 0) {
5006 		/*
5007 		 * If the application passed in zero for the port number, it
5008 		 * doesn't care which port number we bind to. Get one in the
5009 		 * valid range.
5010 		 */
5011 		if (connp->conn_anon_priv_bind) {
5012 			port = udp_get_next_priv_port(udp);
5013 		} else {
5014 			port = udp_update_next_port(udp,
5015 			    us->us_next_port_to_try, B_TRUE);
5016 		}
5017 	} else {
5018 		/*
5019 		 * If the port is in the well-known privileged range,
5020 		 * make sure the caller was privileged.
5021 		 */
5022 		int i;
5023 		boolean_t priv = B_FALSE;
5024 
5025 		if (port < us->us_smallest_nonpriv_port) {
5026 			priv = B_TRUE;
5027 		} else {
5028 			for (i = 0; i < us->us_num_epriv_ports; i++) {
5029 				if (port == us->us_epriv_ports[i]) {
5030 					priv = B_TRUE;
5031 					break;
5032 				}
5033 			}
5034 		}
5035 
5036 		if (priv) {
5037 			if (secpolicy_net_privaddr(cr, port, IPPROTO_UDP) != 0)
5038 				return (-TACCES);
5039 		}
5040 	}
5041 
5042 	if (port == 0)
5043 		return (-TNOADDR);
5044 
5045 	/*
5046 	 * get some memory we might need later on for reuseport, avoid
5047 	 * KM_SLEEP under lock
5048 	 */
5049 	reusep = kmem_zalloc(sizeof (*reusep), KM_SLEEP);
5050 
5051 	mutex_enter(&connp->conn_lock);
5052 
5053 	if (!connp->conn_reuseport) {
5054 		kmem_free(reusep, sizeof (*reusep));
5055 		reusep = NULL;
5056 	}
5057 
5058 	/*
5059 	 * The state must be TS_UNBND. TPI mandates that users must send
5060 	 * TPI primitives only 1 at a time and wait for the response before
5061 	 * sending the next primitive.
5062 	 */
5063 	if (udp->udp_state != TS_UNBND) {
5064 		mutex_exit(&connp->conn_lock);
5065 		if (reusep != NULL)
5066 			kmem_free(reusep, sizeof (*reusep));
5067 		(void) strlog(UDP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
5068 		    "udp_bind: bad state, %u", udp->udp_state);
5069 		return (-TOUTSTATE);
5070 	}
5071 	/*
5072 	 * Copy the source address into our udp structure. This address
5073 	 * may still be zero; if so, IP will fill in the correct address
5074 	 * each time an outbound packet is passed to it. Since the udp is
5075 	 * not yet in the bind hash list, we don't grab the uf_lock to
5076 	 * change conn_ipversion
5077 	 */
5078 	if (connp->conn_family == AF_INET) {
5079 		ASSERT(sin != NULL);
5080 		ASSERT(connp->conn_ixa->ixa_flags & IXAF_IS_IPV4);
5081 	} else {
5082 		if (IN6_IS_ADDR_V4MAPPED(&v6src)) {
5083 			/*
5084 			 * no need to hold the uf_lock to set the conn_ipversion
5085 			 * since we are not yet in the fanout list
5086 			 */
5087 			connp->conn_ipversion = IPV4_VERSION;
5088 			connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
5089 		} else {
5090 			connp->conn_ipversion = IPV6_VERSION;
5091 			connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
5092 		}
5093 	}
5094 
5095 	/*
5096 	 * If conn_reuseaddr is not set, then we have to make sure that
5097 	 * the IP address and port number the application requested
5098 	 * (or we selected for the application) is not being used by
5099 	 * another stream.  If another stream is already using the
5100 	 * requested IP address and port, the behavior depends on
5101 	 * "bind_to_req_port_only". If set the bind fails; otherwise we
5102 	 * search for any an unused port to bind to the stream.
5103 	 *
5104 	 * As per the BSD semantics, as modified by the Deering multicast
5105 	 * changes, if udp_reuseaddr is set, then we allow multiple binds
5106 	 * to the same port independent of the local IP address.
5107 	 *
5108 	 * This is slightly different than in SunOS 4.X which did not
5109 	 * support IP multicast. Note that the change implemented by the
5110 	 * Deering multicast code effects all binds - not only binding
5111 	 * to IP multicast addresses.
5112 	 *
5113 	 * Note that when binding to port zero we ignore SO_REUSEADDR in
5114 	 * order to guarantee a unique port.
5115 	 */
5116 
5117 	count = 0;
5118 	if (connp->conn_anon_priv_bind) {
5119 		/*
5120 		 * loopmax = (IPPORT_RESERVED-1) -
5121 		 *    us->us_min_anonpriv_port + 1
5122 		 */
5123 		loopmax = IPPORT_RESERVED - us->us_min_anonpriv_port;
5124 	} else {
5125 		loopmax = us->us_largest_anon_port -
5126 		    us->us_smallest_anon_port + 1;
5127 	}
5128 
5129 	is_inaddr_any = V6_OR_V4_INADDR_ANY(v6src);
5130 
5131 	for (;;) {
5132 		udp_t		*udp1;
5133 		boolean_t	found_exclbind = B_FALSE;
5134 		conn_t		*connp1;
5135 
5136 		/*
5137 		 * Walk through the list of udp streams bound to
5138 		 * requested port with the same IP address.
5139 		 */
5140 		lport = htons(port);
5141 		udpf = &us->us_bind_fanout[UDP_BIND_HASH(lport,
5142 		    us->us_bind_fanout_size)];
5143 		mutex_enter(&udpf->uf_lock);
5144 		for (udp1 = udpf->uf_udp; udp1 != NULL;
5145 		    udp1 = udp1->udp_bind_hash) {
5146 			connp1 = udp1->udp_connp;
5147 
5148 			if (lport != connp1->conn_lport)
5149 				continue;
5150 
5151 			/*
5152 			 * On a labeled system, we must treat bindings to ports
5153 			 * on shared IP addresses by sockets with MAC exemption
5154 			 * privilege as being in all zones, as there's
5155 			 * otherwise no way to identify the right receiver.
5156 			 */
5157 			if (!IPCL_BIND_ZONE_MATCH(connp1, connp))
5158 				continue;
5159 
5160 			/*
5161 			 * If UDP_EXCLBIND is set for either the bound or
5162 			 * binding endpoint, the semantics of bind
5163 			 * is changed according to the following chart.
5164 			 *
5165 			 * spec = specified address (v4 or v6)
5166 			 * unspec = unspecified address (v4 or v6)
5167 			 * A = specified addresses are different for endpoints
5168 			 *
5169 			 * bound	bind to		allowed?
5170 			 * -------------------------------------
5171 			 * unspec	unspec		no
5172 			 * unspec	spec		no
5173 			 * spec		unspec		no
5174 			 * spec		spec		yes if A
5175 			 *
5176 			 * For labeled systems, SO_MAC_EXEMPT behaves the same
5177 			 * as UDP_EXCLBIND, except that zoneid is ignored.
5178 			 */
5179 			if (connp1->conn_exclbind || connp->conn_exclbind ||
5180 			    IPCL_CONNS_MAC(udp1->udp_connp, connp)) {
5181 				if (V6_OR_V4_INADDR_ANY(
5182 				    connp1->conn_bound_addr_v6) ||
5183 				    is_inaddr_any ||
5184 				    IN6_ARE_ADDR_EQUAL(
5185 				    &connp1->conn_bound_addr_v6,
5186 				    &v6src)) {
5187 					found_exclbind = B_TRUE;
5188 					break;
5189 				}
5190 				continue;
5191 			}
5192 
5193 			/*
5194 			 * Check ipversion to allow IPv4 and IPv6 sockets to
5195 			 * have disjoint port number spaces.
5196 			 */
5197 			if (connp->conn_ipversion != connp1->conn_ipversion) {
5198 
5199 				/*
5200 				 * On the first time through the loop, if the
5201 				 * the user intentionally specified a
5202 				 * particular port number, then ignore any
5203 				 * bindings of the other protocol that may
5204 				 * conflict. This allows the user to bind IPv6
5205 				 * alone and get both v4 and v6, or bind both
5206 				 * both and get each seperately. On subsequent
5207 				 * times through the loop, we're checking a
5208 				 * port that we chose (not the user) and thus
5209 				 * we do not allow casual duplicate bindings.
5210 				 */
5211 				if (count == 0 && requested_port != 0)
5212 					continue;
5213 			}
5214 
5215 			/*
5216 			 * No difference depending on SO_REUSEADDR.
5217 			 *
5218 			 * If existing port is bound to a
5219 			 * non-wildcard IP address and
5220 			 * the requesting stream is bound to
5221 			 * a distinct different IP addresses
5222 			 * (non-wildcard, also), keep going.
5223 			 */
5224 			if (!is_inaddr_any &&
5225 			    !V6_OR_V4_INADDR_ANY(connp1->conn_bound_addr_v6) &&
5226 			    !IN6_ARE_ADDR_EQUAL(&connp1->conn_laddr_v6,
5227 			    &v6src)) {
5228 				continue;
5229 			}
5230 
5231 			/*
5232 			 * if bound conn has reuseport set and conn requests
5233 			 * reuseport, check if cred matches. If they match,
5234 			 * allow conn to proceed.
5235 			 */
5236 			if (connp->conn_reuseport && connp1->conn_reuseport) {
5237 				cred_t *bcred = connp1->conn_cred;
5238 				cred_t *ncred = connp->conn_cred;
5239 				if (crgetuid(bcred) == crgetuid(ncred) &&
5240 				    crgetzoneid(bcred) == crgetzoneid(ncred)) {
5241 					/* just memorize one of the conns */
5242 					reusep->ru_conns[0] = connp1;
5243 					reusep->ru_entries = 1;
5244 					continue;
5245 				}
5246 			}
5247 
5248 			break;
5249 		}
5250 
5251 
5252 		if (!found_exclbind &&
5253 		    (connp->conn_reuseaddr && requested_port != 0)) {
5254 			if (reusep != NULL)
5255 				kmem_free(reusep, sizeof (*reusep));
5256 			break;
5257 		}
5258 
5259 		if (udp1 == NULL) {
5260 			/*
5261 			 * No other stream has this IP address and port number
5262 			 * or all have reuseport set. We can use it.
5263 			 */
5264 			if (connp->conn_reuseport) {
5265 				if (reusep->ru_entries > 0) {
5266 					/* add to a present reuselist */
5267 					struct reuselist *lp = reusep->
5268 						ru_conns[0]->conn_reuselist;
5269 
5270 					ASSERT(lp != NULL);
5271 					kmem_free(reusep, sizeof (*reusep));
5272 					if (udp_reuselist_add(lp, connp) < 0) {
5273 						/* table full */
5274 						mutex_exit(&udpf->uf_lock);
5275 						mutex_exit(&connp->conn_lock);
5276 						return (-TADDRBUSY);
5277 					}
5278 				} else {
5279 					/* use own new reuselist */
5280 					reusep->ru_conns[0] = connp;
5281 					reusep->ru_entries = 1;
5282 					mutex_init(&reusep->ru_lock, NULL,
5283 						MUTEX_DEFAULT, NULL);
5284 					connp->conn_reuselist = reusep;
5285 				}
5286 			}
5287 			break;
5288 		}
5289 		mutex_exit(&udpf->uf_lock);
5290 
5291 		if (connp->conn_reuseport) {
5292 			/* reject for all other cases */
5293 			mutex_exit(&connp->conn_lock);
5294 			kmem_free(reusep, sizeof (*reusep));
5295 			return (-TADDRBUSY);
5296 		}
5297 
5298 		if (bind_to_req_port_only) {
5299 			/*
5300 			 * We get here only when requested port
5301 			 * is bound (and only first  of the for()
5302 			 * loop iteration).
5303 			 *
5304 			 * The semantics of this bind request
5305 			 * require it to fail so we return from
5306 			 * the routine (and exit the loop).
5307 			 *
5308 			 */
5309 			mutex_exit(&connp->conn_lock);
5310 			return (-TADDRBUSY);
5311 		}
5312 
5313 		if (connp->conn_anon_priv_bind) {
5314 			port = udp_get_next_priv_port(udp);
5315 		} else {
5316 			if ((count == 0) && (requested_port != 0)) {
5317 				/*
5318 				 * If the application wants us to find
5319 				 * a port, get one to start with. Set
5320 				 * requested_port to 0, so that we will
5321 				 * update us->us_next_port_to_try below.
5322 				 */
5323 				port = udp_update_next_port(udp,
5324 				    us->us_next_port_to_try, B_TRUE);
5325 				requested_port = 0;
5326 			} else {
5327 				port = udp_update_next_port(udp, port + 1,
5328 				    B_FALSE);
5329 			}
5330 		}
5331 
5332 		if (port == 0 || ++count >= loopmax) {
5333 			/*
5334 			 * We've tried every possible port number and
5335 			 * there are none available, so send an error
5336 			 * to the user.
5337 			 */
5338 			mutex_exit(&connp->conn_lock);
5339 			return (-TNOADDR);
5340 		}
5341 	}
5342 
5343 	/*
5344 	 * Copy the source address into our udp structure.  This address
5345 	 * may still be zero; if so, ip_attr_connect will fill in the correct
5346 	 * address when a packet is about to be sent.
5347 	 * If we are binding to a broadcast or multicast address then
5348 	 * we just set the conn_bound_addr since we don't want to use
5349 	 * that as the source address when sending.
5350 	 */
5351 	connp->conn_bound_addr_v6 = v6src;
5352 	connp->conn_laddr_v6 = v6src;
5353 	if (scopeid != 0) {
5354 		connp->conn_ixa->ixa_flags |= IXAF_SCOPEID_SET;
5355 		connp->conn_ixa->ixa_scopeid = scopeid;
5356 		connp->conn_incoming_ifindex = scopeid;
5357 	} else {
5358 		connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
5359 		connp->conn_incoming_ifindex = connp->conn_bound_if;
5360 	}
5361 
5362 	switch (laddr_type) {
5363 	case IPVL_UNICAST_UP:
5364 	case IPVL_UNICAST_DOWN:
5365 		connp->conn_saddr_v6 = v6src;
5366 		connp->conn_mcbc_bind = B_FALSE;
5367 		break;
5368 	case IPVL_MCAST:
5369 	case IPVL_BCAST:
5370 		/* ip_set_destination will pick a source address later */
5371 		connp->conn_saddr_v6 = ipv6_all_zeros;
5372 		connp->conn_mcbc_bind = B_TRUE;
5373 		break;
5374 	}
5375 
5376 	/* Any errors after this point should use late_error */
5377 	connp->conn_lport = lport;
5378 
5379 	/*
5380 	 * Now reset the next anonymous port if the application requested
5381 	 * an anonymous port, or we handed out the next anonymous port.
5382 	 */
5383 	if ((requested_port == 0) && (!connp->conn_anon_priv_bind)) {
5384 		us->us_next_port_to_try = port + 1;
5385 	}
5386 
5387 	/* Initialize the T_BIND_ACK. */
5388 	if (connp->conn_family == AF_INET) {
5389 		sin->sin_port = connp->conn_lport;
5390 	} else {
5391 		sin6->sin6_port = connp->conn_lport;
5392 	}
5393 	udp->udp_state = TS_IDLE;
5394 	udp_bind_hash_insert(udpf, udp);
5395 	mutex_exit(&udpf->uf_lock);
5396 	mutex_exit(&connp->conn_lock);
5397 
5398 	if (cl_inet_bind) {
5399 		/*
5400 		 * Running in cluster mode - register bind information
5401 		 */
5402 		if (connp->conn_ipversion == IPV4_VERSION) {
5403 			(*cl_inet_bind)(connp->conn_netstack->netstack_stackid,
5404 			    IPPROTO_UDP, AF_INET, (uint8_t *)&v4src,
5405 			    (in_port_t)connp->conn_lport, NULL);
5406 		} else {
5407 			(*cl_inet_bind)(connp->conn_netstack->netstack_stackid,
5408 			    IPPROTO_UDP, AF_INET6, (uint8_t *)&v6src,
5409 			    (in_port_t)connp->conn_lport, NULL);
5410 		}
5411 	}
5412 
5413 	mutex_enter(&connp->conn_lock);
5414 	connp->conn_anon_port = (is_system_labeled() && requested_port == 0);
5415 	if (is_system_labeled() && (!connp->conn_anon_port ||
5416 	    connp->conn_anon_mlp)) {
5417 		uint16_t mlpport;
5418 		zone_t *zone;
5419 
5420 		zone = crgetzone(cr);
5421 		connp->conn_mlp_type =
5422 		    connp->conn_recv_ancillary.crb_recvucred ? mlptBoth :
5423 		    mlptSingle;
5424 		addrtype = tsol_mlp_addr_type(
5425 		    connp->conn_allzones ? ALL_ZONES : zone->zone_id,
5426 		    IPV6_VERSION, &v6src, us->us_netstack->netstack_ip);
5427 		if (addrtype == mlptSingle) {
5428 			error = -TNOADDR;
5429 			mutex_exit(&connp->conn_lock);
5430 			goto late_error;
5431 		}
5432 		mlpport = connp->conn_anon_port ? PMAPPORT : port;
5433 		mlptype = tsol_mlp_port_type(zone, IPPROTO_UDP, mlpport,
5434 		    addrtype);
5435 
5436 		/*
5437 		 * It is a coding error to attempt to bind an MLP port
5438 		 * without first setting SOL_SOCKET/SCM_UCRED.
5439 		 */
5440 		if (mlptype != mlptSingle &&
5441 		    connp->conn_mlp_type == mlptSingle) {
5442 			error = EINVAL;
5443 			mutex_exit(&connp->conn_lock);
5444 			goto late_error;
5445 		}
5446 
5447 		/*
5448 		 * It is an access violation to attempt to bind an MLP port
5449 		 * without NET_BINDMLP privilege.
5450 		 */
5451 		if (mlptype != mlptSingle &&
5452 		    secpolicy_net_bindmlp(cr) != 0) {
5453 			if (connp->conn_debug) {
5454 				(void) strlog(UDP_MOD_ID, 0, 1,
5455 				    SL_ERROR|SL_TRACE,
5456 				    "udp_bind: no priv for multilevel port %d",
5457 				    mlpport);
5458 			}
5459 			error = -TACCES;
5460 			mutex_exit(&connp->conn_lock);
5461 			goto late_error;
5462 		}
5463 
5464 		/*
5465 		 * If we're specifically binding a shared IP address and the
5466 		 * port is MLP on shared addresses, then check to see if this
5467 		 * zone actually owns the MLP.  Reject if not.
5468 		 */
5469 		if (mlptype == mlptShared && addrtype == mlptShared) {
5470 			/*
5471 			 * No need to handle exclusive-stack zones since
5472 			 * ALL_ZONES only applies to the shared stack.
5473 			 */
5474 			zoneid_t mlpzone;
5475 
5476 			mlpzone = tsol_mlp_findzone(IPPROTO_UDP,
5477 			    htons(mlpport));
5478 			if (connp->conn_zoneid != mlpzone) {
5479 				if (connp->conn_debug) {
5480 					(void) strlog(UDP_MOD_ID, 0, 1,
5481 					    SL_ERROR|SL_TRACE,
5482 					    "udp_bind: attempt to bind port "
5483 					    "%d on shared addr in zone %d "
5484 					    "(should be %d)",
5485 					    mlpport, connp->conn_zoneid,
5486 					    mlpzone);
5487 				}
5488 				error = -TACCES;
5489 				mutex_exit(&connp->conn_lock);
5490 				goto late_error;
5491 			}
5492 		}
5493 		if (connp->conn_anon_port) {
5494 			error = tsol_mlp_anon(zone, mlptype, connp->conn_proto,
5495 			    port, B_TRUE);
5496 			if (error != 0) {
5497 				if (connp->conn_debug) {
5498 					(void) strlog(UDP_MOD_ID, 0, 1,
5499 					    SL_ERROR|SL_TRACE,
5500 					    "udp_bind: cannot establish anon "
5501 					    "MLP for port %d", port);
5502 				}
5503 				error = -TACCES;
5504 				mutex_exit(&connp->conn_lock);
5505 				goto late_error;
5506 			}
5507 		}
5508 		connp->conn_mlp_type = mlptype;
5509 	}
5510 
5511 	/*
5512 	 * We create an initial header template here to make a subsequent
5513 	 * sendto have a starting point. Since conn_last_dst is zero the
5514 	 * first sendto will always follow the 'dst changed' code path.
5515 	 * Note that we defer massaging options and the related checksum
5516 	 * adjustment until we have a destination address.
5517 	 */
5518 	error = udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5519 	    &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5520 	if (error != 0) {
5521 		mutex_exit(&connp->conn_lock);
5522 		goto late_error;
5523 	}
5524 	/* Just in case */
5525 	connp->conn_faddr_v6 = ipv6_all_zeros;
5526 	connp->conn_fport = 0;
5527 	connp->conn_v6lastdst = ipv6_all_zeros;
5528 	mutex_exit(&connp->conn_lock);
5529 
5530 	error = ip_laddr_fanout_insert(connp);
5531 	if (error != 0)
5532 		goto late_error;
5533 
5534 	/* Bind succeeded */
5535 	return (0);
5536 
5537 late_error:
5538 	/* We had already picked the port number, and then the bind failed */
5539 	mutex_enter(&connp->conn_lock);
5540 	udpf = &us->us_bind_fanout[
5541 	    UDP_BIND_HASH(connp->conn_lport,
5542 	    us->us_bind_fanout_size)];
5543 	mutex_enter(&udpf->uf_lock);
5544 	connp->conn_saddr_v6 = ipv6_all_zeros;
5545 	connp->conn_bound_addr_v6 = ipv6_all_zeros;
5546 	connp->conn_laddr_v6 = ipv6_all_zeros;
5547 	if (scopeid != 0) {
5548 		connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
5549 		connp->conn_incoming_ifindex = connp->conn_bound_if;
5550 	}
5551 	udp->udp_state = TS_UNBND;
5552 	udp_bind_hash_remove(udp, B_TRUE);
5553 	connp->conn_lport = 0;
5554 	mutex_exit(&udpf->uf_lock);
5555 	connp->conn_anon_port = B_FALSE;
5556 	connp->conn_mlp_type = mlptSingle;
5557 
5558 	connp->conn_v6lastdst = ipv6_all_zeros;
5559 
5560 	/* Restore the header that was built above - different source address */
5561 	(void) udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5562 	    &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5563 	mutex_exit(&connp->conn_lock);
5564 	return (error);
5565 }
5566 
5567 int
5568 udp_bind(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5569     socklen_t len, cred_t *cr)
5570 {
5571 	int		error;
5572 	conn_t		*connp;
5573 
5574 	/* All Solaris components should pass a cred for this operation. */
5575 	ASSERT(cr != NULL);
5576 
5577 	connp = (conn_t *)proto_handle;
5578 
5579 	if (sa == NULL)
5580 		error = udp_do_unbind(connp);
5581 	else
5582 		error = udp_do_bind(connp, sa, len, cr, B_TRUE);
5583 
5584 	if (error < 0) {
5585 		if (error == -TOUTSTATE)
5586 			error = EINVAL;
5587 		else
5588 			error = proto_tlitosyserr(-error);
5589 	}
5590 
5591 	return (error);
5592 }
5593 
5594 static int
5595 udp_implicit_bind(conn_t *connp, cred_t *cr)
5596 {
5597 	sin6_t sin6addr;
5598 	sin_t *sin;
5599 	sin6_t *sin6;
5600 	socklen_t len;
5601 	int error;
5602 
5603 	/* All Solaris components should pass a cred for this operation. */
5604 	ASSERT(cr != NULL);
5605 
5606 	if (connp->conn_family == AF_INET) {
5607 		len = sizeof (struct sockaddr_in);
5608 		sin = (sin_t *)&sin6addr;
5609 		*sin = sin_null;
5610 		sin->sin_family = AF_INET;
5611 		sin->sin_addr.s_addr = INADDR_ANY;
5612 	} else {
5613 		ASSERT(connp->conn_family == AF_INET6);
5614 		len = sizeof (sin6_t);
5615 		sin6 = (sin6_t *)&sin6addr;
5616 		*sin6 = sin6_null;
5617 		sin6->sin6_family = AF_INET6;
5618 		V6_SET_ZERO(sin6->sin6_addr);
5619 	}
5620 
5621 	error = udp_do_bind(connp, (struct sockaddr *)&sin6addr, len,
5622 	    cr, B_FALSE);
5623 	return ((error < 0) ? proto_tlitosyserr(-error) : error);
5624 }
5625 
5626 /*
5627  * This routine removes a port number association from a stream. It
5628  * is called by udp_unbind and udp_tpi_unbind.
5629  */
5630 static int
5631 udp_do_unbind(conn_t *connp)
5632 {
5633 	udp_t 		*udp = connp->conn_udp;
5634 	udp_fanout_t	*udpf;
5635 	udp_stack_t	*us = udp->udp_us;
5636 
5637 	if (cl_inet_unbind != NULL) {
5638 		/*
5639 		 * Running in cluster mode - register unbind information
5640 		 */
5641 		if (connp->conn_ipversion == IPV4_VERSION) {
5642 			(*cl_inet_unbind)(
5643 			    connp->conn_netstack->netstack_stackid,
5644 			    IPPROTO_UDP, AF_INET,
5645 			    (uint8_t *)(&V4_PART_OF_V6(connp->conn_laddr_v6)),
5646 			    (in_port_t)connp->conn_lport, NULL);
5647 		} else {
5648 			(*cl_inet_unbind)(
5649 			    connp->conn_netstack->netstack_stackid,
5650 			    IPPROTO_UDP, AF_INET6,
5651 			    (uint8_t *)&(connp->conn_laddr_v6),
5652 			    (in_port_t)connp->conn_lport, NULL);
5653 		}
5654 	}
5655 
5656 	mutex_enter(&connp->conn_lock);
5657 	/* If a bind has not been done, we can't unbind. */
5658 	if (udp->udp_state == TS_UNBND) {
5659 		mutex_exit(&connp->conn_lock);
5660 		return (-TOUTSTATE);
5661 	}
5662 	udpf = &us->us_bind_fanout[UDP_BIND_HASH(connp->conn_lport,
5663 	    us->us_bind_fanout_size)];
5664 	mutex_enter(&udpf->uf_lock);
5665 	udp_bind_hash_remove(udp, B_TRUE);
5666 	connp->conn_saddr_v6 = ipv6_all_zeros;
5667 	connp->conn_bound_addr_v6 = ipv6_all_zeros;
5668 	connp->conn_laddr_v6 = ipv6_all_zeros;
5669 	connp->conn_mcbc_bind = B_FALSE;
5670 	connp->conn_lport = 0;
5671 	/* In case we were also connected */
5672 	connp->conn_faddr_v6 = ipv6_all_zeros;
5673 	connp->conn_fport = 0;
5674 	mutex_exit(&udpf->uf_lock);
5675 
5676 	connp->conn_v6lastdst = ipv6_all_zeros;
5677 	udp->udp_state = TS_UNBND;
5678 
5679 	(void) udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5680 	    &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5681 	mutex_exit(&connp->conn_lock);
5682 
5683 	ip_unbind(connp);
5684 
5685 	return (0);
5686 }
5687 
5688 /*
5689  * It associates a default destination address with the stream.
5690  */
5691 static int
5692 udp_do_connect(conn_t *connp, const struct sockaddr *sa, socklen_t len,
5693     cred_t *cr, pid_t pid)
5694 {
5695 	sin6_t		*sin6;
5696 	sin_t		*sin;
5697 	in6_addr_t 	v6dst;
5698 	ipaddr_t 	v4dst;
5699 	uint16_t 	dstport;
5700 	uint32_t 	flowinfo;
5701 	udp_fanout_t	*udpf;
5702 	udp_t		*udp, *udp1;
5703 	ushort_t	ipversion;
5704 	udp_stack_t	*us;
5705 	int		error;
5706 	conn_t		*connp1;
5707 	ip_xmit_attr_t	*ixa;
5708 	ip_xmit_attr_t	*oldixa;
5709 	uint_t		scopeid = 0;
5710 	uint_t		srcid = 0;
5711 	in6_addr_t	v6src = connp->conn_saddr_v6;
5712 	boolean_t	v4mapped;
5713 
5714 	udp = connp->conn_udp;
5715 	us = udp->udp_us;
5716 
5717 	/*
5718 	 * Address has been verified by the caller
5719 	 */
5720 	switch (len) {
5721 	default:
5722 		/*
5723 		 * Should never happen
5724 		 */
5725 		return (EINVAL);
5726 
5727 	case sizeof (sin_t):
5728 		sin = (sin_t *)sa;
5729 		v4dst = sin->sin_addr.s_addr;
5730 		dstport = sin->sin_port;
5731 		IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
5732 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
5733 		ipversion = IPV4_VERSION;
5734 		break;
5735 
5736 	case sizeof (sin6_t):
5737 		sin6 = (sin6_t *)sa;
5738 		v6dst = sin6->sin6_addr;
5739 		dstport = sin6->sin6_port;
5740 		srcid = sin6->__sin6_src_id;
5741 		v4mapped = IN6_IS_ADDR_V4MAPPED(&v6dst);
5742 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
5743 			if (!ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
5744 			    v4mapped, connp->conn_netstack)) {
5745 				/* Mismatch v4mapped/v6 specified by srcid. */
5746 				return (EADDRNOTAVAIL);
5747 			}
5748 		}
5749 		if (v4mapped) {
5750 			if (connp->conn_ipv6_v6only)
5751 				return (EADDRNOTAVAIL);
5752 
5753 			/*
5754 			 * Destination adress is mapped IPv6 address.
5755 			 * Source bound address should be unspecified or
5756 			 * IPv6 mapped address as well.
5757 			 */
5758 			if (!IN6_IS_ADDR_UNSPECIFIED(
5759 			    &connp->conn_bound_addr_v6) &&
5760 			    !IN6_IS_ADDR_V4MAPPED(&connp->conn_bound_addr_v6)) {
5761 				return (EADDRNOTAVAIL);
5762 			}
5763 			IN6_V4MAPPED_TO_IPADDR(&v6dst, v4dst);
5764 			ipversion = IPV4_VERSION;
5765 			flowinfo = 0;
5766 		} else {
5767 			ipversion = IPV6_VERSION;
5768 			flowinfo = sin6->sin6_flowinfo;
5769 			if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr))
5770 				scopeid = sin6->sin6_scope_id;
5771 		}
5772 		break;
5773 	}
5774 
5775 	if (dstport == 0)
5776 		return (-TBADADDR);
5777 
5778 	/*
5779 	 * If there is a different thread using conn_ixa then we get a new
5780 	 * copy and cut the old one loose from conn_ixa. Otherwise we use
5781 	 * conn_ixa and prevent any other thread from using/changing it.
5782 	 * Once connect() is done other threads can use conn_ixa since the
5783 	 * refcnt will be back at one.
5784 	 * We defer updating conn_ixa until later to handle any concurrent
5785 	 * conn_ixa_cleanup thread.
5786 	 */
5787 	ixa = conn_get_ixa(connp, B_FALSE);
5788 	if (ixa == NULL)
5789 		return (ENOMEM);
5790 
5791 	mutex_enter(&connp->conn_lock);
5792 	/*
5793 	 * This udp_t must have bound to a port already before doing a connect.
5794 	 * Reject if a connect is in progress (we drop conn_lock during
5795 	 * udp_do_connect).
5796 	 */
5797 	if (udp->udp_state == TS_UNBND || udp->udp_state == TS_WCON_CREQ) {
5798 		mutex_exit(&connp->conn_lock);
5799 		(void) strlog(UDP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
5800 		    "udp_connect: bad state, %u", udp->udp_state);
5801 		ixa_refrele(ixa);
5802 		return (-TOUTSTATE);
5803 	}
5804 	ASSERT(connp->conn_lport != 0 && udp->udp_ptpbhn != NULL);
5805 
5806 	udpf = &us->us_bind_fanout[UDP_BIND_HASH(connp->conn_lport,
5807 	    us->us_bind_fanout_size)];
5808 
5809 	mutex_enter(&udpf->uf_lock);
5810 	if (udp->udp_state == TS_DATA_XFER) {
5811 		/* Already connected - clear out state */
5812 		if (connp->conn_mcbc_bind)
5813 			connp->conn_saddr_v6 = ipv6_all_zeros;
5814 		else
5815 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
5816 		connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
5817 		connp->conn_faddr_v6 = ipv6_all_zeros;
5818 		connp->conn_fport = 0;
5819 		udp->udp_state = TS_IDLE;
5820 	}
5821 
5822 	connp->conn_fport = dstport;
5823 	connp->conn_ipversion = ipversion;
5824 	if (ipversion == IPV4_VERSION) {
5825 		/*
5826 		 * Interpret a zero destination to mean loopback.
5827 		 * Update the T_CONN_REQ (sin/sin6) since it is used to
5828 		 * generate the T_CONN_CON.
5829 		 */
5830 		if (v4dst == INADDR_ANY) {
5831 			v4dst = htonl(INADDR_LOOPBACK);
5832 			IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
5833 			if (connp->conn_family == AF_INET) {
5834 				sin->sin_addr.s_addr = v4dst;
5835 			} else {
5836 				sin6->sin6_addr = v6dst;
5837 			}
5838 		}
5839 		connp->conn_faddr_v6 = v6dst;
5840 		connp->conn_flowinfo = 0;
5841 	} else {
5842 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
5843 		/*
5844 		 * Interpret a zero destination to mean loopback.
5845 		 * Update the T_CONN_REQ (sin/sin6) since it is used to
5846 		 * generate the T_CONN_CON.
5847 		 */
5848 		if (IN6_IS_ADDR_UNSPECIFIED(&v6dst)) {
5849 			v6dst = ipv6_loopback;
5850 			sin6->sin6_addr = v6dst;
5851 		}
5852 		connp->conn_faddr_v6 = v6dst;
5853 		connp->conn_flowinfo = flowinfo;
5854 	}
5855 	mutex_exit(&udpf->uf_lock);
5856 
5857 	/*
5858 	 * We update our cred/cpid based on the caller of connect
5859 	 */
5860 	if (connp->conn_cred != cr) {
5861 		crhold(cr);
5862 		crfree(connp->conn_cred);
5863 		connp->conn_cred = cr;
5864 	}
5865 	connp->conn_cpid = pid;
5866 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
5867 	ixa->ixa_cred = cr;
5868 	ixa->ixa_cpid = pid;
5869 	if (is_system_labeled()) {
5870 		/* We need to restart with a label based on the cred */
5871 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
5872 	}
5873 
5874 	if (scopeid != 0) {
5875 		ixa->ixa_flags |= IXAF_SCOPEID_SET;
5876 		ixa->ixa_scopeid = scopeid;
5877 		connp->conn_incoming_ifindex = scopeid;
5878 	} else {
5879 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
5880 		connp->conn_incoming_ifindex = connp->conn_bound_if;
5881 	}
5882 	/*
5883 	 * conn_connect will drop conn_lock and reacquire it.
5884 	 * To prevent a send* from messing with this udp_t while the lock
5885 	 * is dropped we set udp_state and clear conn_v6lastdst.
5886 	 * That will make all send* fail with EISCONN.
5887 	 */
5888 	connp->conn_v6lastdst = ipv6_all_zeros;
5889 	udp->udp_state = TS_WCON_CREQ;
5890 
5891 	error = conn_connect(connp, NULL, IPDF_ALLOW_MCBC);
5892 	mutex_exit(&connp->conn_lock);
5893 	if (error != 0)
5894 		goto connect_failed;
5895 
5896 	/*
5897 	 * The addresses have been verified. Time to insert in
5898 	 * the correct fanout list.
5899 	 */
5900 	error = ipcl_conn_insert(connp);
5901 	if (error != 0)
5902 		goto connect_failed;
5903 
5904 	mutex_enter(&connp->conn_lock);
5905 	error = udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5906 	    &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5907 	if (error != 0) {
5908 		mutex_exit(&connp->conn_lock);
5909 		goto connect_failed;
5910 	}
5911 
5912 	udp->udp_state = TS_DATA_XFER;
5913 	/* Record this as the "last" send even though we haven't sent any */
5914 	connp->conn_v6lastdst = connp->conn_faddr_v6;
5915 	connp->conn_lastipversion = connp->conn_ipversion;
5916 	connp->conn_lastdstport = connp->conn_fport;
5917 	connp->conn_lastflowinfo = connp->conn_flowinfo;
5918 	connp->conn_lastscopeid = scopeid;
5919 	connp->conn_lastsrcid = srcid;
5920 	/* Also remember a source to use together with lastdst */
5921 	connp->conn_v6lastsrc = v6src;
5922 
5923 	oldixa = conn_replace_ixa(connp, ixa);
5924 	mutex_exit(&connp->conn_lock);
5925 	ixa_refrele(oldixa);
5926 
5927 	/*
5928 	 * We've picked a source address above. Now we can
5929 	 * verify that the src/port/dst/port is unique for all
5930 	 * connections in TS_DATA_XFER, skipping ourselves.
5931 	 */
5932 	mutex_enter(&udpf->uf_lock);
5933 	for (udp1 = udpf->uf_udp; udp1 != NULL; udp1 = udp1->udp_bind_hash) {
5934 		if (udp1->udp_state != TS_DATA_XFER)
5935 			continue;
5936 
5937 		if (udp1 == udp)
5938 			continue;
5939 
5940 		connp1 = udp1->udp_connp;
5941 		if (connp->conn_lport != connp1->conn_lport ||
5942 		    connp->conn_ipversion != connp1->conn_ipversion ||
5943 		    dstport != connp1->conn_fport ||
5944 		    !IN6_ARE_ADDR_EQUAL(&connp->conn_laddr_v6,
5945 		    &connp1->conn_laddr_v6) ||
5946 		    !IN6_ARE_ADDR_EQUAL(&v6dst, &connp1->conn_faddr_v6) ||
5947 		    !(IPCL_ZONE_MATCH(connp, connp1->conn_zoneid) ||
5948 		    IPCL_ZONE_MATCH(connp1, connp->conn_zoneid)))
5949 			continue;
5950 		mutex_exit(&udpf->uf_lock);
5951 		error = -TBADADDR;
5952 		goto connect_failed;
5953 	}
5954 	if (cl_inet_connect2 != NULL) {
5955 		CL_INET_UDP_CONNECT(connp, B_TRUE, &v6dst, dstport, error);
5956 		if (error != 0) {
5957 			mutex_exit(&udpf->uf_lock);
5958 			error = -TBADADDR;
5959 			goto connect_failed;
5960 		}
5961 	}
5962 	mutex_exit(&udpf->uf_lock);
5963 
5964 	ixa_refrele(ixa);
5965 	return (0);
5966 
5967 connect_failed:
5968 	if (ixa != NULL)
5969 		ixa_refrele(ixa);
5970 	mutex_enter(&connp->conn_lock);
5971 	mutex_enter(&udpf->uf_lock);
5972 	udp->udp_state = TS_IDLE;
5973 	connp->conn_faddr_v6 = ipv6_all_zeros;
5974 	connp->conn_fport = 0;
5975 	/* In case the source address was set above */
5976 	if (connp->conn_mcbc_bind)
5977 		connp->conn_saddr_v6 = ipv6_all_zeros;
5978 	else
5979 		connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
5980 	connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
5981 	mutex_exit(&udpf->uf_lock);
5982 
5983 	connp->conn_v6lastdst = ipv6_all_zeros;
5984 	connp->conn_flowinfo = 0;
5985 
5986 	(void) udp_build_hdr_template(connp, &connp->conn_saddr_v6,
5987 	    &connp->conn_faddr_v6, connp->conn_fport, connp->conn_flowinfo);
5988 	mutex_exit(&connp->conn_lock);
5989 	return (error);
5990 }
5991 
5992 static int
5993 udp_connect(sock_lower_handle_t proto_handle, const struct sockaddr *sa,
5994     socklen_t len, sock_connid_t *id, cred_t *cr)
5995 {
5996 	conn_t	*connp = (conn_t *)proto_handle;
5997 	udp_t	*udp = connp->conn_udp;
5998 	int	error;
5999 	boolean_t did_bind = B_FALSE;
6000 	pid_t	pid = curproc->p_pid;
6001 
6002 	/* All Solaris components should pass a cred for this operation. */
6003 	ASSERT(cr != NULL);
6004 
6005 	if (sa == NULL) {
6006 		/*
6007 		 * Disconnect
6008 		 * Make sure we are connected
6009 		 */
6010 		if (udp->udp_state != TS_DATA_XFER)
6011 			return (EINVAL);
6012 
6013 		error = udp_disconnect(connp);
6014 		return (error);
6015 	}
6016 
6017 	error = proto_verify_ip_addr(connp->conn_family, sa, len);
6018 	if (error != 0)
6019 		goto done;
6020 
6021 	/* do an implicit bind if necessary */
6022 	if (udp->udp_state == TS_UNBND) {
6023 		error = udp_implicit_bind(connp, cr);
6024 		/*
6025 		 * We could be racing with an actual bind, in which case
6026 		 * we would see EPROTO. We cross our fingers and try
6027 		 * to connect.
6028 		 */
6029 		if (!(error == 0 || error == EPROTO))
6030 			goto done;
6031 		did_bind = B_TRUE;
6032 	}
6033 	/*
6034 	 * set SO_DGRAM_ERRIND
6035 	 */
6036 	connp->conn_dgram_errind = B_TRUE;
6037 
6038 	error = udp_do_connect(connp, sa, len, cr, pid);
6039 
6040 	if (error != 0 && did_bind) {
6041 		int unbind_err;
6042 
6043 		unbind_err = udp_do_unbind(connp);
6044 		ASSERT(unbind_err == 0);
6045 	}
6046 
6047 	if (error == 0) {
6048 		*id = 0;
6049 		(*connp->conn_upcalls->su_connected)
6050 		    (connp->conn_upper_handle, 0, NULL, -1);
6051 	} else if (error < 0) {
6052 		error = proto_tlitosyserr(-error);
6053 	}
6054 
6055 done:
6056 	if (error != 0 && udp->udp_state == TS_DATA_XFER) {
6057 		/*
6058 		 * No need to hold locks to set state
6059 		 * after connect failure socket state is undefined
6060 		 * We set the state only to imitate old sockfs behavior
6061 		 */
6062 		udp->udp_state = TS_IDLE;
6063 	}
6064 	return (error);
6065 }
6066 
6067 int
6068 udp_send(sock_lower_handle_t proto_handle, mblk_t *mp, struct nmsghdr *msg,
6069     cred_t *cr)
6070 {
6071 	sin6_t		*sin6;
6072 	sin_t		*sin = NULL;
6073 	uint_t		srcid;
6074 	conn_t		*connp = (conn_t *)proto_handle;
6075 	udp_t		*udp = connp->conn_udp;
6076 	int		error = 0;
6077 	udp_stack_t	*us = udp->udp_us;
6078 	ushort_t	ipversion;
6079 	pid_t		pid = curproc->p_pid;
6080 	ip_xmit_attr_t	*ixa;
6081 
6082 	ASSERT(DB_TYPE(mp) == M_DATA);
6083 
6084 	/* All Solaris components should pass a cred for this operation. */
6085 	ASSERT(cr != NULL);
6086 
6087 	/* do an implicit bind if necessary */
6088 	if (udp->udp_state == TS_UNBND) {
6089 		error = udp_implicit_bind(connp, cr);
6090 		/*
6091 		 * We could be racing with an actual bind, in which case
6092 		 * we would see EPROTO. We cross our fingers and try
6093 		 * to connect.
6094 		 */
6095 		if (!(error == 0 || error == EPROTO)) {
6096 			freemsg(mp);
6097 			return (error);
6098 		}
6099 	}
6100 
6101 	/* Connected? */
6102 	if (msg->msg_name == NULL) {
6103 		if (udp->udp_state != TS_DATA_XFER) {
6104 			UDPS_BUMP_MIB(us, udpOutErrors);
6105 			return (EDESTADDRREQ);
6106 		}
6107 		if (msg->msg_controllen != 0) {
6108 			error = udp_output_ancillary(connp, NULL, NULL, mp,
6109 			    NULL, msg, cr, pid);
6110 		} else {
6111 			error = udp_output_connected(connp, mp, cr, pid);
6112 		}
6113 		if (us->us_sendto_ignerr)
6114 			return (0);
6115 		else
6116 			return (error);
6117 	}
6118 	if (udp->udp_state == TS_DATA_XFER) {
6119 		UDPS_BUMP_MIB(us, udpOutErrors);
6120 		return (EISCONN);
6121 	}
6122 	error = proto_verify_ip_addr(connp->conn_family,
6123 	    (struct sockaddr *)msg->msg_name, msg->msg_namelen);
6124 	if (error != 0) {
6125 		UDPS_BUMP_MIB(us, udpOutErrors);
6126 		return (error);
6127 	}
6128 	switch (connp->conn_family) {
6129 	case AF_INET6:
6130 		sin6 = (sin6_t *)msg->msg_name;
6131 
6132 		srcid = sin6->__sin6_src_id;
6133 
6134 		if (!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
6135 			/*
6136 			 * Destination is a non-IPv4-compatible IPv6 address.
6137 			 * Send out an IPv6 format packet.
6138 			 */
6139 
6140 			/*
6141 			 * If the local address is a mapped address return
6142 			 * an error.
6143 			 * It would be possible to send an IPv6 packet but the
6144 			 * response would never make it back to the application
6145 			 * since it is bound to a mapped address.
6146 			 */
6147 			if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
6148 				UDPS_BUMP_MIB(us, udpOutErrors);
6149 				return (EADDRNOTAVAIL);
6150 			}
6151 			if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
6152 				sin6->sin6_addr = ipv6_loopback;
6153 			ipversion = IPV6_VERSION;
6154 		} else {
6155 			if (connp->conn_ipv6_v6only) {
6156 				UDPS_BUMP_MIB(us, udpOutErrors);
6157 				return (EADDRNOTAVAIL);
6158 			}
6159 
6160 			/*
6161 			 * If the local address is not zero or a mapped address
6162 			 * return an error.  It would be possible to send an
6163 			 * IPv4 packet but the response would never make it
6164 			 * back to the application since it is bound to a
6165 			 * non-mapped address.
6166 			 */
6167 			if (!IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6) &&
6168 			    !IN6_IS_ADDR_UNSPECIFIED(&connp->conn_saddr_v6)) {
6169 				UDPS_BUMP_MIB(us, udpOutErrors);
6170 				return (EADDRNOTAVAIL);
6171 			}
6172 
6173 			if (V4_PART_OF_V6(sin6->sin6_addr) == INADDR_ANY) {
6174 				V4_PART_OF_V6(sin6->sin6_addr) =
6175 				    htonl(INADDR_LOOPBACK);
6176 			}
6177 			ipversion = IPV4_VERSION;
6178 		}
6179 
6180 		/*
6181 		 * We have to allocate an ip_xmit_attr_t before we grab
6182 		 * conn_lock and we need to hold conn_lock once we've check
6183 		 * conn_same_as_last_v6 to handle concurrent send* calls on a
6184 		 * socket.
6185 		 */
6186 		if (msg->msg_controllen == 0) {
6187 			ixa = conn_get_ixa(connp, B_FALSE);
6188 			if (ixa == NULL) {
6189 				UDPS_BUMP_MIB(us, udpOutErrors);
6190 				return (ENOMEM);
6191 			}
6192 		} else {
6193 			ixa = NULL;
6194 		}
6195 		mutex_enter(&connp->conn_lock);
6196 		if (udp->udp_delayed_error != 0) {
6197 			sin6_t  *sin2 = (sin6_t *)&udp->udp_delayed_addr;
6198 
6199 			error = udp->udp_delayed_error;
6200 			udp->udp_delayed_error = 0;
6201 
6202 			/* Compare IP address, port, and family */
6203 
6204 			if (sin6->sin6_port == sin2->sin6_port &&
6205 			    IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr,
6206 			    &sin2->sin6_addr) &&
6207 			    sin6->sin6_family == sin2->sin6_family) {
6208 				mutex_exit(&connp->conn_lock);
6209 				UDPS_BUMP_MIB(us, udpOutErrors);
6210 				if (ixa != NULL)
6211 					ixa_refrele(ixa);
6212 				return (error);
6213 			}
6214 		}
6215 
6216 		if (msg->msg_controllen != 0) {
6217 			mutex_exit(&connp->conn_lock);
6218 			ASSERT(ixa == NULL);
6219 			error = udp_output_ancillary(connp, NULL, sin6, mp,
6220 			    NULL, msg, cr, pid);
6221 		} else if (conn_same_as_last_v6(connp, sin6) &&
6222 		    connp->conn_lastsrcid == srcid &&
6223 		    ipsec_outbound_policy_current(ixa)) {
6224 			/* udp_output_lastdst drops conn_lock */
6225 			error = udp_output_lastdst(connp, mp, cr, pid, ixa);
6226 		} else {
6227 			/* udp_output_newdst drops conn_lock */
6228 			error = udp_output_newdst(connp, mp, NULL, sin6,
6229 			    ipversion, cr, pid, ixa);
6230 		}
6231 		ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
6232 		if (us->us_sendto_ignerr)
6233 			return (0);
6234 		else
6235 			return (error);
6236 	case AF_INET:
6237 		sin = (sin_t *)msg->msg_name;
6238 
6239 		ipversion = IPV4_VERSION;
6240 
6241 		if (sin->sin_addr.s_addr == INADDR_ANY)
6242 			sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
6243 
6244 		/*
6245 		 * We have to allocate an ip_xmit_attr_t before we grab
6246 		 * conn_lock and we need to hold conn_lock once we've check
6247 		 * conn_same_as_last_v6 to handle concurrent send* on a socket.
6248 		 */
6249 		if (msg->msg_controllen == 0) {
6250 			ixa = conn_get_ixa(connp, B_FALSE);
6251 			if (ixa == NULL) {
6252 				UDPS_BUMP_MIB(us, udpOutErrors);
6253 				return (ENOMEM);
6254 			}
6255 		} else {
6256 			ixa = NULL;
6257 		}
6258 		mutex_enter(&connp->conn_lock);
6259 		if (udp->udp_delayed_error != 0) {
6260 			sin_t  *sin2 = (sin_t *)&udp->udp_delayed_addr;
6261 
6262 			error = udp->udp_delayed_error;
6263 			udp->udp_delayed_error = 0;
6264 
6265 			/* Compare IP address and port */
6266 
6267 			if (sin->sin_port == sin2->sin_port &&
6268 			    sin->sin_addr.s_addr == sin2->sin_addr.s_addr) {
6269 				mutex_exit(&connp->conn_lock);
6270 				UDPS_BUMP_MIB(us, udpOutErrors);
6271 				if (ixa != NULL)
6272 					ixa_refrele(ixa);
6273 				return (error);
6274 			}
6275 		}
6276 		if (msg->msg_controllen != 0) {
6277 			mutex_exit(&connp->conn_lock);
6278 			ASSERT(ixa == NULL);
6279 			error = udp_output_ancillary(connp, sin, NULL, mp,
6280 			    NULL, msg, cr, pid);
6281 		} else if (conn_same_as_last_v4(connp, sin) &&
6282 		    ipsec_outbound_policy_current(ixa)) {
6283 			/* udp_output_lastdst drops conn_lock */
6284 			error = udp_output_lastdst(connp, mp, cr, pid, ixa);
6285 		} else {
6286 			/* udp_output_newdst drops conn_lock */
6287 			error = udp_output_newdst(connp, mp, sin, NULL,
6288 			    ipversion, cr, pid, ixa);
6289 		}
6290 		ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
6291 		if (us->us_sendto_ignerr)
6292 			return (0);
6293 		else
6294 			return (error);
6295 	default:
6296 		return (EINVAL);
6297 	}
6298 }
6299 
6300 int
6301 udp_fallback(sock_lower_handle_t proto_handle, queue_t *q,
6302     boolean_t issocket, so_proto_quiesced_cb_t quiesced_cb,
6303     sock_quiesce_arg_t *arg)
6304 {
6305 	conn_t 	*connp = (conn_t *)proto_handle;
6306 	udp_t	*udp;
6307 	struct T_capability_ack tca;
6308 	struct sockaddr_in6 laddr, faddr;
6309 	socklen_t laddrlen, faddrlen;
6310 	short opts;
6311 	struct stroptions *stropt;
6312 	mblk_t *mp, *stropt_mp;
6313 	int error;
6314 
6315 	udp = connp->conn_udp;
6316 
6317 	stropt_mp = allocb_wait(sizeof (*stropt), BPRI_HI, STR_NOSIG, NULL);
6318 
6319 	/*
6320 	 * setup the fallback stream that was allocated
6321 	 */
6322 	connp->conn_dev = (dev_t)RD(q)->q_ptr;
6323 	connp->conn_minor_arena = WR(q)->q_ptr;
6324 
6325 	RD(q)->q_ptr = WR(q)->q_ptr = connp;
6326 
6327 	WR(q)->q_qinfo = &udp_winit;
6328 
6329 	connp->conn_rq = RD(q);
6330 	connp->conn_wq = WR(q);
6331 
6332 	/* Notify stream head about options before sending up data */
6333 	stropt_mp->b_datap->db_type = M_SETOPTS;
6334 	stropt_mp->b_wptr += sizeof (*stropt);
6335 	stropt = (struct stroptions *)stropt_mp->b_rptr;
6336 	stropt->so_flags = SO_WROFF | SO_HIWAT;
6337 	stropt->so_wroff = connp->conn_wroff;
6338 	stropt->so_hiwat = udp->udp_rcv_disply_hiwat;
6339 	putnext(RD(q), stropt_mp);
6340 
6341 	/*
6342 	 * Free the helper stream
6343 	 */
6344 	ip_free_helper_stream(connp);
6345 
6346 	if (!issocket)
6347 		udp_use_pure_tpi(udp);
6348 
6349 	/*
6350 	 * Collect the information needed to sync with the sonode
6351 	 */
6352 	udp_do_capability_ack(udp, &tca, TC1_INFO);
6353 
6354 	laddrlen = faddrlen = sizeof (sin6_t);
6355 	(void) udp_getsockname((sock_lower_handle_t)connp,
6356 	    (struct sockaddr *)&laddr, &laddrlen, CRED());
6357 	error = udp_getpeername((sock_lower_handle_t)connp,
6358 	    (struct sockaddr *)&faddr, &faddrlen, CRED());
6359 	if (error != 0)
6360 		faddrlen = 0;
6361 
6362 	opts = 0;
6363 	if (connp->conn_dgram_errind)
6364 		opts |= SO_DGRAM_ERRIND;
6365 	if (connp->conn_ixa->ixa_flags & IXAF_DONTROUTE)
6366 		opts |= SO_DONTROUTE;
6367 
6368 	mp = (*quiesced_cb)(connp->conn_upper_handle, arg, &tca,
6369 	    (struct sockaddr *)&laddr, laddrlen,
6370 	    (struct sockaddr *)&faddr, faddrlen, opts);
6371 
6372 	mutex_enter(&udp->udp_recv_lock);
6373 	/*
6374 	 * Attempts to send data up during fallback will result in it being
6375 	 * queued in udp_t. First push up the datagrams obtained from the
6376 	 * socket, then any packets queued in udp_t.
6377 	 */
6378 	if (mp != NULL) {
6379 		mp->b_next = udp->udp_fallback_queue_head;
6380 		udp->udp_fallback_queue_head = mp;
6381 	}
6382 	while (udp->udp_fallback_queue_head != NULL) {
6383 		mp = udp->udp_fallback_queue_head;
6384 		udp->udp_fallback_queue_head = mp->b_next;
6385 		mutex_exit(&udp->udp_recv_lock);
6386 		mp->b_next = NULL;
6387 		putnext(RD(q), mp);
6388 		mutex_enter(&udp->udp_recv_lock);
6389 	}
6390 	udp->udp_fallback_queue_tail = udp->udp_fallback_queue_head;
6391 	/*
6392 	 * No longer a streams less socket
6393 	 */
6394 	mutex_enter(&connp->conn_lock);
6395 	connp->conn_flags &= ~IPCL_NONSTR;
6396 	mutex_exit(&connp->conn_lock);
6397 
6398 	mutex_exit(&udp->udp_recv_lock);
6399 
6400 	ASSERT(connp->conn_ref >= 1);
6401 
6402 	return (0);
6403 }
6404 
6405 /* ARGSUSED3 */
6406 int
6407 udp_getpeername(sock_lower_handle_t proto_handle, struct sockaddr *sa,
6408     socklen_t *salenp, cred_t *cr)
6409 {
6410 	conn_t	*connp = (conn_t *)proto_handle;
6411 	udp_t	*udp = connp->conn_udp;
6412 	int error;
6413 
6414 	/* All Solaris components should pass a cred for this operation. */
6415 	ASSERT(cr != NULL);
6416 
6417 	mutex_enter(&connp->conn_lock);
6418 	if (udp->udp_state != TS_DATA_XFER)
6419 		error = ENOTCONN;
6420 	else
6421 		error = conn_getpeername(connp, sa, salenp);
6422 	mutex_exit(&connp->conn_lock);
6423 	return (error);
6424 }
6425 
6426 /* ARGSUSED3 */
6427 int
6428 udp_getsockname(sock_lower_handle_t proto_handle, struct sockaddr *sa,
6429     socklen_t *salenp, cred_t *cr)
6430 {
6431 	conn_t	*connp = (conn_t *)proto_handle;
6432 	int error;
6433 
6434 	/* All Solaris components should pass a cred for this operation. */
6435 	ASSERT(cr != NULL);
6436 
6437 	mutex_enter(&connp->conn_lock);
6438 	error = conn_getsockname(connp, sa, salenp);
6439 	mutex_exit(&connp->conn_lock);
6440 	return (error);
6441 }
6442 
6443 int
6444 udp_getsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
6445     void *optvalp, socklen_t *optlen, cred_t *cr)
6446 {
6447 	conn_t		*connp = (conn_t *)proto_handle;
6448 	int		error;
6449 	t_uscalar_t	max_optbuf_len;
6450 	void		*optvalp_buf;
6451 	int		len;
6452 
6453 	/* All Solaris components should pass a cred for this operation. */
6454 	ASSERT(cr != NULL);
6455 
6456 	error = proto_opt_check(level, option_name, *optlen, &max_optbuf_len,
6457 	    udp_opt_obj.odb_opt_des_arr,
6458 	    udp_opt_obj.odb_opt_arr_cnt,
6459 	    B_FALSE, B_TRUE, cr);
6460 	if (error != 0) {
6461 		if (error < 0)
6462 			error = proto_tlitosyserr(-error);
6463 		return (error);
6464 	}
6465 
6466 	optvalp_buf = kmem_alloc(max_optbuf_len, KM_SLEEP);
6467 	len = udp_opt_get(connp, level, option_name, optvalp_buf);
6468 	if (len == -1) {
6469 		kmem_free(optvalp_buf, max_optbuf_len);
6470 		return (EINVAL);
6471 	}
6472 
6473 	/*
6474 	 * update optlen and copy option value
6475 	 */
6476 	t_uscalar_t size = MIN(len, *optlen);
6477 
6478 	bcopy(optvalp_buf, optvalp, size);
6479 	bcopy(&size, optlen, sizeof (size));
6480 
6481 	kmem_free(optvalp_buf, max_optbuf_len);
6482 	return (0);
6483 }
6484 
6485 int
6486 udp_setsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
6487     const void *optvalp, socklen_t optlen, cred_t *cr)
6488 {
6489 	conn_t		*connp = (conn_t *)proto_handle;
6490 	int		error;
6491 
6492 	/* All Solaris components should pass a cred for this operation. */
6493 	ASSERT(cr != NULL);
6494 
6495 	error = proto_opt_check(level, option_name, optlen, NULL,
6496 	    udp_opt_obj.odb_opt_des_arr,
6497 	    udp_opt_obj.odb_opt_arr_cnt,
6498 	    B_TRUE, B_FALSE, cr);
6499 
6500 	if (error != 0) {
6501 		if (error < 0)
6502 			error = proto_tlitosyserr(-error);
6503 		return (error);
6504 	}
6505 
6506 	error = udp_opt_set(connp, SETFN_OPTCOM_NEGOTIATE, level, option_name,
6507 	    optlen, (uchar_t *)optvalp, (uint_t *)&optlen, (uchar_t *)optvalp,
6508 	    NULL, cr);
6509 
6510 	ASSERT(error >= 0);
6511 
6512 	return (error);
6513 }
6514 
6515 void
6516 udp_clr_flowctrl(sock_lower_handle_t proto_handle)
6517 {
6518 	conn_t	*connp = (conn_t *)proto_handle;
6519 	udp_t	*udp = connp->conn_udp;
6520 
6521 	mutex_enter(&udp->udp_recv_lock);
6522 	connp->conn_flow_cntrld = B_FALSE;
6523 	mutex_exit(&udp->udp_recv_lock);
6524 }
6525 
6526 /* ARGSUSED2 */
6527 int
6528 udp_shutdown(sock_lower_handle_t proto_handle, int how, cred_t *cr)
6529 {
6530 	conn_t	*connp = (conn_t *)proto_handle;
6531 
6532 	/* All Solaris components should pass a cred for this operation. */
6533 	ASSERT(cr != NULL);
6534 
6535 	/* shut down the send side */
6536 	if (how != SHUT_RD)
6537 		(*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
6538 		    SOCK_OPCTL_SHUT_SEND, 0);
6539 	/* shut down the recv side */
6540 	if (how != SHUT_WR)
6541 		(*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
6542 		    SOCK_OPCTL_SHUT_RECV, 0);
6543 	return (0);
6544 }
6545 
6546 int
6547 udp_ioctl(sock_lower_handle_t proto_handle, int cmd, intptr_t arg,
6548     int mode, int32_t *rvalp, cred_t *cr)
6549 {
6550 	conn_t  	*connp = (conn_t *)proto_handle;
6551 	int		error;
6552 
6553 	/* All Solaris components should pass a cred for this operation. */
6554 	ASSERT(cr != NULL);
6555 
6556 	/*
6557 	 * If we don't have a helper stream then create one.
6558 	 * ip_create_helper_stream takes care of locking the conn_t,
6559 	 * so this check for NULL is just a performance optimization.
6560 	 */
6561 	if (connp->conn_helper_info == NULL) {
6562 		udp_stack_t *us = connp->conn_udp->udp_us;
6563 
6564 		ASSERT(us->us_ldi_ident != NULL);
6565 
6566 		/*
6567 		 * Create a helper stream for non-STREAMS socket.
6568 		 */
6569 		error = ip_create_helper_stream(connp, us->us_ldi_ident);
6570 		if (error != 0) {
6571 			ip0dbg(("tcp_ioctl: create of IP helper stream "
6572 			    "failed %d\n", error));
6573 			return (error);
6574 		}
6575 	}
6576 
6577 	switch (cmd) {
6578 		case _SIOCSOCKFALLBACK:
6579 		case TI_GETPEERNAME:
6580 		case TI_GETMYNAME:
6581 			ip1dbg(("udp_ioctl: cmd 0x%x on non streams socket",
6582 			    cmd));
6583 			error = EINVAL;
6584 			break;
6585 		default:
6586 			/*
6587 			 * Pass on to IP using helper stream
6588 			 */
6589 			error = ldi_ioctl(connp->conn_helper_info->iphs_handle,
6590 			    cmd, arg, mode, cr, rvalp);
6591 			break;
6592 	}
6593 	return (error);
6594 }
6595 
6596 /* ARGSUSED */
6597 int
6598 udp_accept(sock_lower_handle_t lproto_handle,
6599     sock_lower_handle_t eproto_handle, sock_upper_handle_t sock_handle,
6600     cred_t *cr)
6601 {
6602 	return (EOPNOTSUPP);
6603 }
6604 
6605 /* ARGSUSED */
6606 int
6607 udp_listen(sock_lower_handle_t proto_handle, int backlog, cred_t *cr)
6608 {
6609 	return (EOPNOTSUPP);
6610 }
6611 
6612 sock_downcalls_t sock_udp_downcalls = {
6613 	udp_activate,		/* sd_activate */
6614 	udp_accept,		/* sd_accept */
6615 	udp_bind,		/* sd_bind */
6616 	udp_listen,		/* sd_listen */
6617 	udp_connect,		/* sd_connect */
6618 	udp_getpeername,	/* sd_getpeername */
6619 	udp_getsockname,	/* sd_getsockname */
6620 	udp_getsockopt,		/* sd_getsockopt */
6621 	udp_setsockopt,		/* sd_setsockopt */
6622 	udp_send,		/* sd_send */
6623 	NULL,			/* sd_send_uio */
6624 	NULL,			/* sd_recv_uio */
6625 	NULL,			/* sd_poll */
6626 	udp_shutdown,		/* sd_shutdown */
6627 	udp_clr_flowctrl,	/* sd_setflowctrl */
6628 	udp_ioctl,		/* sd_ioctl */
6629 	udp_close		/* sd_close */
6630 };
6631