xref: /titanic_41/usr/src/uts/common/inet/ip/icmp.c (revision c39526b769298791ff5b0b6c5e761f49aabaeb4e)
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 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /* Copyright (c) 1990 Mentat Inc. */
26 
27 #include <sys/types.h>
28 #include <sys/stream.h>
29 #include <sys/stropts.h>
30 #include <sys/strlog.h>
31 #include <sys/strsun.h>
32 #define	_SUN_TPI_VERSION 2
33 #include <sys/tihdr.h>
34 #include <sys/timod.h>
35 #include <sys/ddi.h>
36 #include <sys/sunddi.h>
37 #include <sys/strsubr.h>
38 #include <sys/suntpi.h>
39 #include <sys/xti_inet.h>
40 #include <sys/cmn_err.h>
41 #include <sys/kmem.h>
42 #include <sys/cred_impl.h>
43 #include <sys/policy.h>
44 #include <sys/priv.h>
45 #include <sys/ucred.h>
46 #include <sys/zone.h>
47 
48 #include <sys/sockio.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/vtrace.h>
52 #include <sys/sdt.h>
53 #include <sys/debug.h>
54 #include <sys/isa_defs.h>
55 #include <sys/random.h>
56 #include <netinet/in.h>
57 #include <netinet/ip6.h>
58 #include <netinet/icmp6.h>
59 #include <netinet/udp.h>
60 
61 #include <inet/common.h>
62 #include <inet/ip.h>
63 #include <inet/ip_impl.h>
64 #include <inet/ipsec_impl.h>
65 #include <inet/ip6.h>
66 #include <inet/ip_ire.h>
67 #include <inet/ip_if.h>
68 #include <inet/ip_multi.h>
69 #include <inet/ip_ndp.h>
70 #include <inet/proto_set.h>
71 #include <inet/mib2.h>
72 #include <inet/nd.h>
73 #include <inet/optcom.h>
74 #include <inet/snmpcom.h>
75 #include <inet/kstatcom.h>
76 #include <inet/ipclassifier.h>
77 
78 #include <sys/tsol/label.h>
79 #include <sys/tsol/tnet.h>
80 
81 #include <inet/rawip_impl.h>
82 
83 #include <sys/disp.h>
84 
85 /*
86  * Synchronization notes:
87  *
88  * RAWIP is MT and uses the usual kernel synchronization primitives. We use
89  * conn_lock to protect the icmp_t.
90  *
91  * Plumbing notes:
92  * ICMP is always a device driver. For compatibility with mibopen() code
93  * it is possible to I_PUSH "icmp", but that results in pushing a passthrough
94  * dummy module.
95  */
96 
97 static void	icmp_addr_req(queue_t *q, mblk_t *mp);
98 static void	icmp_tpi_bind(queue_t *q, mblk_t *mp);
99 static void	icmp_bind_proto(icmp_t *icmp);
100 static int	icmp_build_hdr_template(conn_t *, const in6_addr_t *,
101     const in6_addr_t *, uint32_t);
102 static void	icmp_capability_req(queue_t *q, mblk_t *mp);
103 static int	icmp_close(queue_t *q, int flags);
104 static void	icmp_close_free(conn_t *);
105 static void	icmp_tpi_connect(queue_t *q, mblk_t *mp);
106 static void	icmp_tpi_disconnect(queue_t *q, mblk_t *mp);
107 static void	icmp_err_ack(queue_t *q, mblk_t *mp, t_scalar_t t_error,
108     int sys_error);
109 static void	icmp_err_ack_prim(queue_t *q, mblk_t *mp, t_scalar_t primitive,
110     t_scalar_t tlierr, int sys_error);
111 static void	icmp_icmp_input(void *arg1, mblk_t *mp, void *arg2,
112     ip_recv_attr_t *);
113 static void	icmp_icmp_error_ipv6(conn_t *connp, mblk_t *mp,
114     ip_recv_attr_t *);
115 static void	icmp_info_req(queue_t *q, mblk_t *mp);
116 static void	icmp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
117 static conn_t 	*icmp_open(int family, cred_t *credp, int *err, int flags);
118 static int	icmp_openv4(queue_t *q, dev_t *devp, int flag, int sflag,
119 		    cred_t *credp);
120 static int	icmp_openv6(queue_t *q, dev_t *devp, int flag, int sflag,
121 		    cred_t *credp);
122 static boolean_t icmp_opt_allow_udr_set(t_scalar_t level, t_scalar_t name);
123 int		icmp_opt_set(conn_t *connp, uint_t optset_context,
124 		    int level, int name, uint_t inlen,
125 		    uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
126 		    void *thisdg_attrs, cred_t *cr);
127 int		icmp_opt_get(conn_t *connp, int level, int name,
128 		    uchar_t *ptr);
129 static int	icmp_output_newdst(conn_t *connp, mblk_t *data_mp, sin_t *sin,
130 		    sin6_t *sin6, cred_t *cr, pid_t pid, ip_xmit_attr_t *ixa);
131 static mblk_t	*icmp_prepend_hdr(conn_t *, ip_xmit_attr_t *, const ip_pkt_t *,
132     const in6_addr_t *, const in6_addr_t *, uint32_t, mblk_t *, int *);
133 static mblk_t	*icmp_prepend_header_template(conn_t *, ip_xmit_attr_t *,
134     mblk_t *, const in6_addr_t *, uint32_t, int *);
135 static int	icmp_snmp_set(queue_t *q, t_scalar_t level, t_scalar_t name,
136 		    uchar_t *ptr, int len);
137 static void	icmp_ud_err(queue_t *q, mblk_t *mp, t_scalar_t err);
138 static void	icmp_tpi_unbind(queue_t *q, mblk_t *mp);
139 static void	icmp_wput(queue_t *q, mblk_t *mp);
140 static void	icmp_wput_fallback(queue_t *q, mblk_t *mp);
141 static void	icmp_wput_other(queue_t *q, mblk_t *mp);
142 static void	icmp_wput_iocdata(queue_t *q, mblk_t *mp);
143 static void	icmp_wput_restricted(queue_t *q, mblk_t *mp);
144 static void	icmp_ulp_recv(conn_t *, mblk_t *, uint_t);
145 
146 static void	*rawip_stack_init(netstackid_t stackid, netstack_t *ns);
147 static void	rawip_stack_fini(netstackid_t stackid, void *arg);
148 
149 static void	*rawip_kstat_init(netstackid_t stackid);
150 static void	rawip_kstat_fini(netstackid_t stackid, kstat_t *ksp);
151 static int	rawip_kstat_update(kstat_t *kp, int rw);
152 static void	rawip_stack_shutdown(netstackid_t stackid, void *arg);
153 
154 /* Common routines for TPI and socket module */
155 static conn_t	*rawip_do_open(int, cred_t *, int *, int);
156 static void	rawip_do_close(conn_t *);
157 static int	rawip_do_bind(conn_t *, struct sockaddr *, socklen_t);
158 static int	rawip_do_unbind(conn_t *);
159 static int	rawip_do_connect(conn_t *, const struct sockaddr *, socklen_t,
160     cred_t *, pid_t);
161 
162 int		rawip_getsockname(sock_lower_handle_t, struct sockaddr *,
163 		    socklen_t *, cred_t *);
164 int		rawip_getpeername(sock_lower_handle_t, struct sockaddr *,
165 		    socklen_t *, cred_t *);
166 
167 static struct module_info icmp_mod_info =  {
168 	5707, "icmp", 1, INFPSZ, 512, 128
169 };
170 
171 /*
172  * Entry points for ICMP as a device.
173  * We have separate open functions for the /dev/icmp and /dev/icmp6 devices.
174  */
175 static struct qinit icmprinitv4 = {
176 	NULL, NULL, icmp_openv4, icmp_close, NULL, &icmp_mod_info
177 };
178 
179 static struct qinit icmprinitv6 = {
180 	NULL, NULL, icmp_openv6, icmp_close, NULL, &icmp_mod_info
181 };
182 
183 static struct qinit icmpwinit = {
184 	(pfi_t)icmp_wput, (pfi_t)ip_wsrv, NULL, NULL, NULL, &icmp_mod_info
185 };
186 
187 /* ICMP entry point during fallback */
188 static struct qinit icmp_fallback_sock_winit = {
189 	(pfi_t)icmp_wput_fallback, NULL, NULL, NULL, NULL, &icmp_mod_info
190 };
191 
192 /* For AF_INET aka /dev/icmp */
193 struct streamtab icmpinfov4 = {
194 	&icmprinitv4, &icmpwinit
195 };
196 
197 /* For AF_INET6 aka /dev/icmp6 */
198 struct streamtab icmpinfov6 = {
199 	&icmprinitv6, &icmpwinit
200 };
201 
202 /* Default structure copied into T_INFO_ACK messages */
203 static struct T_info_ack icmp_g_t_info_ack = {
204 	T_INFO_ACK,
205 	IP_MAXPACKET,	 /* TSDU_size.  icmp allows maximum size messages. */
206 	T_INVALID,	/* ETSDU_size.  icmp does not support expedited data. */
207 	T_INVALID,	/* CDATA_size. icmp does not support connect data. */
208 	T_INVALID,	/* DDATA_size. icmp does not support disconnect data. */
209 	0,		/* ADDR_size - filled in later. */
210 	0,		/* OPT_size - not initialized here */
211 	IP_MAXPACKET,	/* TIDU_size.  icmp allows maximum size messages. */
212 	T_CLTS,		/* SERV_type.  icmp supports connection-less. */
213 	TS_UNBND,	/* CURRENT_state.  This is set from icmp_state. */
214 	(XPG4_1|SENDZERO) /* PROVIDER_flag */
215 };
216 
217 /*
218  * All of these are alterable, within the min/max values given, at run time.
219  *
220  * Note: All those tunables which do not start with "icmp_" are Committed and
221  * therefore are public. See PSARC 2009/306.
222  */
223 static mod_prop_info_t icmp_propinfo_tbl[] = {
224 	/* tunable - 0 */
225 	{ "icmp_wroff_extra", MOD_PROTO_RAWIP,
226 	    mod_set_uint32, mod_get_uint32,
227 	    {0, 128, 32}, {32} },
228 
229 	{ "icmp_ipv4_ttl", MOD_PROTO_RAWIP,
230 	    mod_set_uint32, mod_get_uint32,
231 	    {1, 255, 255}, {255} },
232 
233 	{ "icmp_ipv6_hoplimit", MOD_PROTO_RAWIP,
234 	    mod_set_uint32, mod_get_uint32,
235 	    {0, IPV6_MAX_HOPS, IPV6_DEFAULT_HOPS},
236 	    {IPV6_DEFAULT_HOPS} },
237 
238 	{ "icmp_bsd_compat", MOD_PROTO_RAWIP,
239 	    mod_set_boolean, mod_get_boolean,
240 	    {B_TRUE}, {B_TRUE} },
241 
242 	{ "send_maxbuf", MOD_PROTO_RAWIP,
243 	    mod_set_uint32, mod_get_uint32,
244 	    {4096, 65536, 8192}, {8192} },
245 
246 	{ "icmp_xmit_lowat", MOD_PROTO_RAWIP,
247 	    mod_set_uint32, mod_get_uint32,
248 	    {0, 65536, 1024}, {1024} },
249 
250 	{ "recv_maxbuf", MOD_PROTO_RAWIP,
251 	    mod_set_uint32, mod_get_uint32,
252 	    {4096, 65536, 8192}, {8192} },
253 
254 	{ "icmp_max_buf", MOD_PROTO_RAWIP,
255 	    mod_set_uint32, mod_get_uint32,
256 	    {65536, 1024*1024*1024, 256*1024}, {256 * 1024} },
257 
258 	{ "icmp_pmtu_discovery", MOD_PROTO_RAWIP,
259 	    mod_set_boolean, mod_get_boolean,
260 	    {B_FALSE}, {B_FALSE} },
261 
262 	{ "icmp_sendto_ignerr", MOD_PROTO_RAWIP,
263 	    mod_set_boolean, mod_get_boolean,
264 	    {B_FALSE}, {B_FALSE} },
265 
266 	{ "?", MOD_PROTO_RAWIP, NULL, mod_get_allprop, {0}, {0} },
267 
268 	{ NULL, 0, NULL, NULL, {0}, {0} }
269 };
270 
271 #define	is_wroff_extra			is_propinfo_tbl[0].prop_cur_uval
272 #define	is_ipv4_ttl			is_propinfo_tbl[1].prop_cur_uval
273 #define	is_ipv6_hoplimit		is_propinfo_tbl[2].prop_cur_uval
274 #define	is_bsd_compat			is_propinfo_tbl[3].prop_cur_bval
275 #define	is_xmit_hiwat			is_propinfo_tbl[4].prop_cur_uval
276 #define	is_xmit_lowat			is_propinfo_tbl[5].prop_cur_uval
277 #define	is_recv_hiwat			is_propinfo_tbl[6].prop_cur_uval
278 #define	is_max_buf			is_propinfo_tbl[7].prop_cur_uval
279 #define	is_pmtu_discovery		is_propinfo_tbl[8].prop_cur_bval
280 #define	is_sendto_ignerr		is_propinfo_tbl[9].prop_cur_bval
281 
282 typedef union T_primitives *t_primp_t;
283 
284 /*
285  * This routine is called to handle each O_T_BIND_REQ/T_BIND_REQ message
286  * passed to icmp_wput.
287  * It calls IP to verify the local IP address, and calls IP to insert
288  * the conn_t in the fanout table.
289  * If everything is ok it then sends the T_BIND_ACK back up.
290  */
291 static void
292 icmp_tpi_bind(queue_t *q, mblk_t *mp)
293 {
294 	int	error;
295 	struct sockaddr *sa;
296 	struct T_bind_req *tbr;
297 	socklen_t	len;
298 	sin_t	*sin;
299 	sin6_t	*sin6;
300 	icmp_t		*icmp;
301 	conn_t	*connp = Q_TO_CONN(q);
302 	mblk_t *mp1;
303 	cred_t *cr;
304 
305 	/*
306 	 * All Solaris components should pass a db_credp
307 	 * for this TPI message, hence we ASSERT.
308 	 * But in case there is some other M_PROTO that looks
309 	 * like a TPI message sent by some other kernel
310 	 * component, we check and return an error.
311 	 */
312 	cr = msg_getcred(mp, NULL);
313 	ASSERT(cr != NULL);
314 	if (cr == NULL) {
315 		icmp_err_ack(q, mp, TSYSERR, EINVAL);
316 		return;
317 	}
318 
319 	icmp = connp->conn_icmp;
320 	if ((mp->b_wptr - mp->b_rptr) < sizeof (*tbr)) {
321 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
322 		    "icmp_bind: bad req, len %u",
323 		    (uint_t)(mp->b_wptr - mp->b_rptr));
324 		icmp_err_ack(q, mp, TPROTO, 0);
325 		return;
326 	}
327 
328 	if (icmp->icmp_state != TS_UNBND) {
329 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
330 		    "icmp_bind: bad state, %u", icmp->icmp_state);
331 		icmp_err_ack(q, mp, TOUTSTATE, 0);
332 		return;
333 	}
334 
335 	/*
336 	 * Reallocate the message to make sure we have enough room for an
337 	 * address.
338 	 */
339 	mp1 = reallocb(mp, sizeof (struct T_bind_ack) + sizeof (sin6_t), 1);
340 	if (mp1 == NULL) {
341 		icmp_err_ack(q, mp, TSYSERR, ENOMEM);
342 		return;
343 	}
344 	mp = mp1;
345 
346 	/* Reset the message type in preparation for shipping it back. */
347 	DB_TYPE(mp) = M_PCPROTO;
348 	tbr = (struct T_bind_req *)mp->b_rptr;
349 	len = tbr->ADDR_length;
350 	switch (len) {
351 	case 0:	/* request for a generic port */
352 		tbr->ADDR_offset = sizeof (struct T_bind_req);
353 		if (connp->conn_family == AF_INET) {
354 			tbr->ADDR_length = sizeof (sin_t);
355 			sin = (sin_t *)&tbr[1];
356 			*sin = sin_null;
357 			sin->sin_family = AF_INET;
358 			mp->b_wptr = (uchar_t *)&sin[1];
359 			sa = (struct sockaddr *)sin;
360 			len = sizeof (sin_t);
361 		} else {
362 			ASSERT(connp->conn_family == AF_INET6);
363 			tbr->ADDR_length = sizeof (sin6_t);
364 			sin6 = (sin6_t *)&tbr[1];
365 			*sin6 = sin6_null;
366 			sin6->sin6_family = AF_INET6;
367 			mp->b_wptr = (uchar_t *)&sin6[1];
368 			sa = (struct sockaddr *)sin6;
369 			len = sizeof (sin6_t);
370 		}
371 		break;
372 
373 	case sizeof (sin_t):	/* Complete IPv4 address */
374 		sa = (struct sockaddr *)mi_offset_param(mp, tbr->ADDR_offset,
375 		    sizeof (sin_t));
376 		break;
377 
378 	case sizeof (sin6_t):	/* Complete IPv6 address */
379 		sa = (struct sockaddr *)mi_offset_param(mp,
380 		    tbr->ADDR_offset, sizeof (sin6_t));
381 		break;
382 
383 	default:
384 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
385 		    "icmp_bind: bad ADDR_length %u", tbr->ADDR_length);
386 		icmp_err_ack(q, mp, TBADADDR, 0);
387 		return;
388 	}
389 
390 	error = rawip_do_bind(connp, sa, len);
391 	if (error != 0) {
392 		if (error > 0) {
393 			icmp_err_ack(q, mp, TSYSERR, error);
394 		} else {
395 			icmp_err_ack(q, mp, -error, 0);
396 		}
397 	} else {
398 		tbr->PRIM_type = T_BIND_ACK;
399 		qreply(q, mp);
400 	}
401 }
402 
403 static int
404 rawip_do_bind(conn_t *connp, struct sockaddr *sa, socklen_t len)
405 {
406 	sin_t		*sin;
407 	sin6_t		*sin6;
408 	icmp_t		*icmp = connp->conn_icmp;
409 	int		error = 0;
410 	ip_laddr_t	laddr_type = IPVL_UNICAST_UP;	/* INADDR_ANY */
411 	in_port_t	lport;		/* Network byte order */
412 	ipaddr_t	v4src;		/* Set if AF_INET */
413 	in6_addr_t	v6src;
414 	uint_t		scopeid = 0;
415 	zoneid_t	zoneid = IPCL_ZONEID(connp);
416 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
417 
418 	if (sa == NULL || !OK_32PTR((char *)sa)) {
419 		return (EINVAL);
420 	}
421 
422 	switch (len) {
423 	case sizeof (sin_t):    /* Complete IPv4 address */
424 		sin = (sin_t *)sa;
425 		if (sin->sin_family != AF_INET ||
426 		    connp->conn_family != AF_INET) {
427 			/* TSYSERR, EAFNOSUPPORT */
428 			return (EAFNOSUPPORT);
429 		}
430 		v4src = sin->sin_addr.s_addr;
431 		IN6_IPADDR_TO_V4MAPPED(v4src, &v6src);
432 		if (v4src != INADDR_ANY) {
433 			laddr_type = ip_laddr_verify_v4(v4src, zoneid, ipst,
434 			    B_TRUE);
435 		}
436 		lport = sin->sin_port;
437 		break;
438 	case sizeof (sin6_t): /* Complete IPv6 address */
439 		sin6 = (sin6_t *)sa;
440 		if (sin6->sin6_family != AF_INET6 ||
441 		    connp->conn_family != AF_INET6) {
442 			/* TSYSERR, EAFNOSUPPORT */
443 			return (EAFNOSUPPORT);
444 		}
445 		/* No support for mapped addresses on raw sockets */
446 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
447 			/* TSYSERR, EADDRNOTAVAIL */
448 			return (EADDRNOTAVAIL);
449 		}
450 		v6src = sin6->sin6_addr;
451 		if (!IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
452 			if (IN6_IS_ADDR_LINKSCOPE(&v6src))
453 				scopeid = sin6->sin6_scope_id;
454 			laddr_type = ip_laddr_verify_v6(&v6src, zoneid, ipst,
455 			    B_TRUE, scopeid);
456 		}
457 		lport = sin6->sin6_port;
458 		break;
459 
460 	default:
461 		/* TBADADDR */
462 		return (EADDRNOTAVAIL);
463 	}
464 
465 	/* Is the local address a valid unicast, multicast, or broadcast? */
466 	if (laddr_type == IPVL_BAD)
467 		return (EADDRNOTAVAIL);
468 
469 	/*
470 	 * The state must be TS_UNBND.
471 	 */
472 	mutex_enter(&connp->conn_lock);
473 	if (icmp->icmp_state != TS_UNBND) {
474 		mutex_exit(&connp->conn_lock);
475 		return (-TOUTSTATE);
476 	}
477 
478 	/*
479 	 * Copy the source address into our icmp structure.  This address
480 	 * may still be zero; if so, ip will fill in the correct address
481 	 * each time an outbound packet is passed to it.
482 	 * If we are binding to a broadcast or multicast address then
483 	 * we just set the conn_bound_addr since we don't want to use
484 	 * that as the source address when sending.
485 	 */
486 	connp->conn_bound_addr_v6 = v6src;
487 	connp->conn_laddr_v6 = v6src;
488 	if (scopeid != 0) {
489 		connp->conn_ixa->ixa_flags |= IXAF_SCOPEID_SET;
490 		connp->conn_ixa->ixa_scopeid = scopeid;
491 		connp->conn_incoming_ifindex = scopeid;
492 	} else {
493 		connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
494 		connp->conn_incoming_ifindex = connp->conn_bound_if;
495 	}
496 
497 	switch (laddr_type) {
498 	case IPVL_UNICAST_UP:
499 	case IPVL_UNICAST_DOWN:
500 		connp->conn_saddr_v6 = v6src;
501 		connp->conn_mcbc_bind = B_FALSE;
502 		break;
503 	case IPVL_MCAST:
504 	case IPVL_BCAST:
505 		/* ip_set_destination will pick a source address later */
506 		connp->conn_saddr_v6 = ipv6_all_zeros;
507 		connp->conn_mcbc_bind = B_TRUE;
508 		break;
509 	}
510 
511 	/* Any errors after this point should use late_error */
512 
513 	/*
514 	 * Use sin_port/sin6_port since applications like psh use SOCK_RAW
515 	 * with IPPROTO_TCP.
516 	 */
517 	connp->conn_lport = lport;
518 	connp->conn_fport = 0;
519 
520 	if (connp->conn_family == AF_INET) {
521 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
522 	} else {
523 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
524 	}
525 
526 	icmp->icmp_state = TS_IDLE;
527 
528 	/*
529 	 * We create an initial header template here to make a subsequent
530 	 * sendto have a starting point. Since conn_last_dst is zero the
531 	 * first sendto will always follow the 'dst changed' code path.
532 	 * Note that we defer massaging options and the related checksum
533 	 * adjustment until we have a destination address.
534 	 */
535 	error = icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
536 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
537 	if (error != 0) {
538 		mutex_exit(&connp->conn_lock);
539 		goto late_error;
540 	}
541 	/* Just in case */
542 	connp->conn_faddr_v6 = ipv6_all_zeros;
543 	connp->conn_v6lastdst = ipv6_all_zeros;
544 	mutex_exit(&connp->conn_lock);
545 
546 	error = ip_laddr_fanout_insert(connp);
547 	if (error != 0)
548 		goto late_error;
549 
550 	/* Bind succeeded */
551 	return (0);
552 
553 late_error:
554 	mutex_enter(&connp->conn_lock);
555 	connp->conn_saddr_v6 = ipv6_all_zeros;
556 	connp->conn_bound_addr_v6 = ipv6_all_zeros;
557 	connp->conn_laddr_v6 = ipv6_all_zeros;
558 	if (scopeid != 0) {
559 		connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
560 		connp->conn_incoming_ifindex = connp->conn_bound_if;
561 	}
562 	icmp->icmp_state = TS_UNBND;
563 	connp->conn_v6lastdst = ipv6_all_zeros;
564 	connp->conn_lport = 0;
565 
566 	/* Restore the header that was built above - different source address */
567 	(void) icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
568 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
569 	mutex_exit(&connp->conn_lock);
570 	return (error);
571 }
572 
573 /*
574  * Tell IP to just bind to the protocol.
575  */
576 static void
577 icmp_bind_proto(icmp_t *icmp)
578 {
579 	conn_t	*connp = icmp->icmp_connp;
580 
581 	mutex_enter(&connp->conn_lock);
582 	connp->conn_saddr_v6 = ipv6_all_zeros;
583 	connp->conn_laddr_v6 = ipv6_all_zeros;
584 	connp->conn_faddr_v6 = ipv6_all_zeros;
585 	connp->conn_v6lastdst = ipv6_all_zeros;
586 	mutex_exit(&connp->conn_lock);
587 
588 	(void) ip_laddr_fanout_insert(connp);
589 }
590 
591 /*
592  * This routine handles each T_CONN_REQ message passed to icmp.  It
593  * associates a default destination address with the stream.
594  *
595  * After various error checks are completed, icmp_connect() lays
596  * the target address and port into the composite header template.
597  * Then we ask IP for information, including a source address if we didn't
598  * already have one. Finally we send up the T_OK_ACK reply message.
599  */
600 static void
601 icmp_tpi_connect(queue_t *q, mblk_t *mp)
602 {
603 	conn_t	*connp = Q_TO_CONN(q);
604 	struct T_conn_req	*tcr;
605 	struct sockaddr *sa;
606 	socklen_t len;
607 	int error;
608 	cred_t *cr;
609 	pid_t pid;
610 	/*
611 	 * All Solaris components should pass a db_credp
612 	 * for this TPI message, hence we ASSERT.
613 	 * But in case there is some other M_PROTO that looks
614 	 * like a TPI message sent by some other kernel
615 	 * component, we check and return an error.
616 	 */
617 	cr = msg_getcred(mp, &pid);
618 	ASSERT(cr != NULL);
619 	if (cr == NULL) {
620 		icmp_err_ack(q, mp, TSYSERR, EINVAL);
621 		return;
622 	}
623 
624 	tcr = (struct T_conn_req *)mp->b_rptr;
625 	/* Sanity checks */
626 	if ((mp->b_wptr - mp->b_rptr) < sizeof (struct T_conn_req)) {
627 		icmp_err_ack(q, mp, TPROTO, 0);
628 		return;
629 	}
630 
631 	if (tcr->OPT_length != 0) {
632 		icmp_err_ack(q, mp, TBADOPT, 0);
633 		return;
634 	}
635 
636 	len = tcr->DEST_length;
637 
638 	switch (len) {
639 	default:
640 		icmp_err_ack(q, mp, TBADADDR, 0);
641 		return;
642 	case sizeof (sin_t):
643 		sa = (struct sockaddr *)mi_offset_param(mp, tcr->DEST_offset,
644 		    sizeof (sin_t));
645 		break;
646 	case sizeof (sin6_t):
647 		sa = (struct sockaddr *)mi_offset_param(mp,
648 		    tcr->DEST_offset, sizeof (sin6_t));
649 		break;
650 	}
651 
652 	error = proto_verify_ip_addr(connp->conn_family, sa, len);
653 	if (error != 0) {
654 		icmp_err_ack(q, mp, TSYSERR, error);
655 		return;
656 	}
657 
658 	error = rawip_do_connect(connp, sa, len, cr, pid);
659 	if (error != 0) {
660 		if (error < 0) {
661 			icmp_err_ack(q, mp, -error, 0);
662 		} else {
663 			icmp_err_ack(q, mp, 0, error);
664 		}
665 	} else {
666 		mblk_t *mp1;
667 
668 		/*
669 		 * We have to send a connection confirmation to
670 		 * keep TLI happy.
671 		 */
672 		if (connp->conn_family == AF_INET) {
673 			mp1 = mi_tpi_conn_con(NULL, (char *)sa,
674 			    sizeof (sin_t), NULL, 0);
675 		} else {
676 			ASSERT(connp->conn_family == AF_INET6);
677 			mp1 = mi_tpi_conn_con(NULL, (char *)sa,
678 			    sizeof (sin6_t), NULL, 0);
679 		}
680 		if (mp1 == NULL) {
681 			icmp_err_ack(q, mp, TSYSERR, ENOMEM);
682 			return;
683 		}
684 
685 		/*
686 		 * Send ok_ack for T_CONN_REQ
687 		 */
688 		mp = mi_tpi_ok_ack_alloc(mp);
689 		if (mp == NULL) {
690 			/* Unable to reuse the T_CONN_REQ for the ack. */
691 			icmp_err_ack_prim(q, mp1, T_CONN_REQ, TSYSERR, ENOMEM);
692 			return;
693 		}
694 		putnext(connp->conn_rq, mp);
695 		putnext(connp->conn_rq, mp1);
696 	}
697 }
698 
699 static int
700 rawip_do_connect(conn_t *connp, const struct sockaddr *sa, socklen_t len,
701     cred_t *cr, pid_t pid)
702 {
703 	icmp_t		*icmp;
704 	sin_t		*sin;
705 	sin6_t		*sin6;
706 	int		error;
707 	uint16_t 	dstport;
708 	ipaddr_t	v4dst;
709 	in6_addr_t	v6dst;
710 	uint32_t	flowinfo;
711 	ip_xmit_attr_t	*ixa;
712 	uint_t		scopeid = 0;
713 	uint_t		srcid = 0;
714 	in6_addr_t	v6src = connp->conn_saddr_v6;
715 
716 	icmp = connp->conn_icmp;
717 
718 	if (sa == NULL || !OK_32PTR((char *)sa)) {
719 		return (EINVAL);
720 	}
721 
722 	ASSERT(sa != NULL && len != 0);
723 
724 	/*
725 	 * Determine packet type based on type of address passed in
726 	 * the request should contain an IPv4 or IPv6 address.
727 	 * Make sure that address family matches the type of
728 	 * family of the address passed down.
729 	 */
730 	switch (len) {
731 	case sizeof (sin_t):
732 		sin = (sin_t *)sa;
733 
734 		v4dst = sin->sin_addr.s_addr;
735 		dstport = sin->sin_port;
736 		IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
737 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
738 		break;
739 
740 	case sizeof (sin6_t):
741 		sin6 = (sin6_t *)sa;
742 
743 		/* No support for mapped addresses on raw sockets */
744 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
745 			return (EADDRNOTAVAIL);
746 		}
747 		v6dst = sin6->sin6_addr;
748 		dstport = sin6->sin6_port;
749 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
750 		flowinfo = sin6->sin6_flowinfo;
751 		if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr))
752 			scopeid = sin6->sin6_scope_id;
753 		srcid = sin6->__sin6_src_id;
754 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
755 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
756 			    connp->conn_netstack);
757 		}
758 		break;
759 	}
760 
761 	/*
762 	 * If there is a different thread using conn_ixa then we get a new
763 	 * copy and cut the old one loose from conn_ixa. Otherwise we use
764 	 * conn_ixa and prevent any other thread from using/changing it.
765 	 * Once connect() is done other threads can use conn_ixa since the
766 	 * refcnt will be back at one.
767 	 */
768 	ixa = conn_get_ixa(connp, B_TRUE);
769 	if (ixa == NULL)
770 		return (ENOMEM);
771 
772 	ASSERT(ixa->ixa_refcnt >= 2);
773 	ASSERT(ixa == connp->conn_ixa);
774 
775 	mutex_enter(&connp->conn_lock);
776 	/*
777 	 * This icmp_t must have bound already before doing a connect.
778 	 * Reject if a connect is in progress (we drop conn_lock during
779 	 * rawip_do_connect).
780 	 */
781 	if (icmp->icmp_state == TS_UNBND || icmp->icmp_state == TS_WCON_CREQ) {
782 		mutex_exit(&connp->conn_lock);
783 		ixa_refrele(ixa);
784 		return (-TOUTSTATE);
785 	}
786 
787 	if (icmp->icmp_state == TS_DATA_XFER) {
788 		/* Already connected - clear out state */
789 		if (connp->conn_mcbc_bind)
790 			connp->conn_saddr_v6 = ipv6_all_zeros;
791 		else
792 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
793 		connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
794 		connp->conn_faddr_v6 = ipv6_all_zeros;
795 		icmp->icmp_state = TS_IDLE;
796 	}
797 
798 	/*
799 	 * Use sin_port/sin6_port since applications like psh use SOCK_RAW
800 	 * with IPPROTO_TCP.
801 	 */
802 	connp->conn_fport = dstport;
803 	if (connp->conn_ipversion == IPV4_VERSION) {
804 		/*
805 		 * Interpret a zero destination to mean loopback.
806 		 * Update the T_CONN_REQ (sin/sin6) since it is used to
807 		 * generate the T_CONN_CON.
808 		 */
809 		if (v4dst == INADDR_ANY) {
810 			v4dst = htonl(INADDR_LOOPBACK);
811 			IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
812 			ASSERT(connp->conn_family == AF_INET);
813 			sin->sin_addr.s_addr = v4dst;
814 		}
815 		connp->conn_faddr_v6 = v6dst;
816 		connp->conn_flowinfo = 0;
817 	} else {
818 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
819 		/*
820 		 * Interpret a zero destination to mean loopback.
821 		 * Update the T_CONN_REQ (sin/sin6) since it is used to
822 		 * generate the T_CONN_CON.
823 		 */
824 		if (IN6_IS_ADDR_UNSPECIFIED(&v6dst)) {
825 			v6dst = ipv6_loopback;
826 			sin6->sin6_addr = v6dst;
827 		}
828 		connp->conn_faddr_v6 = v6dst;
829 		connp->conn_flowinfo = flowinfo;
830 	}
831 
832 	/*
833 	 * We update our cred/cpid based on the caller of connect
834 	 */
835 	if (connp->conn_cred != cr) {
836 		crhold(cr);
837 		crfree(connp->conn_cred);
838 		connp->conn_cred = cr;
839 	}
840 	connp->conn_cpid = pid;
841 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
842 	ixa->ixa_cred = cr;
843 	ixa->ixa_cpid = pid;
844 	if (is_system_labeled()) {
845 		/* We need to restart with a label based on the cred */
846 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
847 	}
848 
849 	if (scopeid != 0) {
850 		ixa->ixa_flags |= IXAF_SCOPEID_SET;
851 		ixa->ixa_scopeid = scopeid;
852 		connp->conn_incoming_ifindex = scopeid;
853 	} else {
854 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
855 		connp->conn_incoming_ifindex = connp->conn_bound_if;
856 	}
857 
858 	/*
859 	 * conn_connect will drop conn_lock and reacquire it.
860 	 * To prevent a send* from messing with this icmp_t while the lock
861 	 * is dropped we set icmp_state and clear conn_v6lastdst.
862 	 * That will make all send* fail with EISCONN.
863 	 */
864 	connp->conn_v6lastdst = ipv6_all_zeros;
865 	icmp->icmp_state = TS_WCON_CREQ;
866 
867 	error = conn_connect(connp, NULL, IPDF_ALLOW_MCBC);
868 	mutex_exit(&connp->conn_lock);
869 	if (error != 0)
870 		goto connect_failed;
871 
872 	/*
873 	 * The addresses have been verified. Time to insert in
874 	 * the correct fanout list.
875 	 */
876 	error = ipcl_conn_insert(connp);
877 	if (error != 0)
878 		goto connect_failed;
879 
880 	mutex_enter(&connp->conn_lock);
881 	error = icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
882 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
883 	if (error != 0) {
884 		mutex_exit(&connp->conn_lock);
885 		goto connect_failed;
886 	}
887 
888 	icmp->icmp_state = TS_DATA_XFER;
889 	/* Record this as the "last" send even though we haven't sent any */
890 	connp->conn_v6lastdst = connp->conn_faddr_v6;
891 	connp->conn_lastipversion = connp->conn_ipversion;
892 	connp->conn_lastdstport = connp->conn_fport;
893 	connp->conn_lastflowinfo = connp->conn_flowinfo;
894 	connp->conn_lastscopeid = scopeid;
895 	connp->conn_lastsrcid = srcid;
896 	/* Also remember a source to use together with lastdst */
897 	connp->conn_v6lastsrc = v6src;
898 	mutex_exit(&connp->conn_lock);
899 
900 	ixa_refrele(ixa);
901 	return (0);
902 
903 connect_failed:
904 	if (ixa != NULL)
905 		ixa_refrele(ixa);
906 	mutex_enter(&connp->conn_lock);
907 	icmp->icmp_state = TS_IDLE;
908 	/* In case the source address was set above */
909 	if (connp->conn_mcbc_bind)
910 		connp->conn_saddr_v6 = ipv6_all_zeros;
911 	else
912 		connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
913 	connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
914 	connp->conn_faddr_v6 = ipv6_all_zeros;
915 	connp->conn_v6lastdst = ipv6_all_zeros;
916 	connp->conn_flowinfo = 0;
917 
918 	(void) icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
919 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
920 	mutex_exit(&connp->conn_lock);
921 	return (error);
922 }
923 
924 static void
925 rawip_do_close(conn_t *connp)
926 {
927 	ASSERT(connp != NULL && IPCL_IS_RAWIP(connp));
928 
929 	ip_quiesce_conn(connp);
930 
931 	if (!IPCL_IS_NONSTR(connp)) {
932 		qprocsoff(connp->conn_rq);
933 	}
934 
935 	icmp_close_free(connp);
936 
937 	/*
938 	 * Now we are truly single threaded on this stream, and can
939 	 * delete the things hanging off the connp, and finally the connp.
940 	 * We removed this connp from the fanout list, it cannot be
941 	 * accessed thru the fanouts, and we already waited for the
942 	 * conn_ref to drop to 0. We are already in close, so
943 	 * there cannot be any other thread from the top. qprocsoff
944 	 * has completed, and service has completed or won't run in
945 	 * future.
946 	 */
947 	ASSERT(connp->conn_ref == 1);
948 
949 	if (!IPCL_IS_NONSTR(connp)) {
950 		inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
951 	} else {
952 		ip_free_helper_stream(connp);
953 	}
954 
955 	connp->conn_ref--;
956 	ipcl_conn_destroy(connp);
957 }
958 
959 static int
960 icmp_close(queue_t *q, int flags)
961 {
962 	conn_t  *connp;
963 
964 	if (flags & SO_FALLBACK) {
965 		/*
966 		 * stream is being closed while in fallback
967 		 * simply free the resources that were allocated
968 		 */
969 		inet_minor_free(WR(q)->q_ptr, (dev_t)(RD(q)->q_ptr));
970 		qprocsoff(q);
971 		goto done;
972 	}
973 
974 	connp = Q_TO_CONN(q);
975 	(void) rawip_do_close(connp);
976 done:
977 	q->q_ptr = WR(q)->q_ptr = NULL;
978 	return (0);
979 }
980 
981 static void
982 icmp_close_free(conn_t *connp)
983 {
984 	icmp_t *icmp = connp->conn_icmp;
985 
986 	if (icmp->icmp_filter != NULL) {
987 		kmem_free(icmp->icmp_filter, sizeof (icmp6_filter_t));
988 		icmp->icmp_filter = NULL;
989 	}
990 
991 	/*
992 	 * Clear any fields which the kmem_cache constructor clears.
993 	 * Only icmp_connp needs to be preserved.
994 	 * TBD: We should make this more efficient to avoid clearing
995 	 * everything.
996 	 */
997 	ASSERT(icmp->icmp_connp == connp);
998 	bzero(icmp, sizeof (icmp_t));
999 	icmp->icmp_connp = connp;
1000 }
1001 
1002 /*
1003  * This routine handles each T_DISCON_REQ message passed to icmp
1004  * as an indicating that ICMP is no longer connected. This results
1005  * in telling IP to restore the binding to just the local address.
1006  */
1007 static int
1008 icmp_do_disconnect(conn_t *connp)
1009 {
1010 	icmp_t	*icmp = connp->conn_icmp;
1011 	int	error;
1012 
1013 	mutex_enter(&connp->conn_lock);
1014 	if (icmp->icmp_state != TS_DATA_XFER) {
1015 		mutex_exit(&connp->conn_lock);
1016 		return (-TOUTSTATE);
1017 	}
1018 	if (connp->conn_mcbc_bind)
1019 		connp->conn_saddr_v6 = ipv6_all_zeros;
1020 	else
1021 		connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
1022 	connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
1023 	connp->conn_faddr_v6 = ipv6_all_zeros;
1024 	icmp->icmp_state = TS_IDLE;
1025 
1026 	connp->conn_v6lastdst = ipv6_all_zeros;
1027 	error = icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
1028 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
1029 	mutex_exit(&connp->conn_lock);
1030 	if (error != 0)
1031 		return (error);
1032 
1033 	/*
1034 	 * Tell IP to remove the full binding and revert
1035 	 * to the local address binding.
1036 	 */
1037 	return (ip_laddr_fanout_insert(connp));
1038 }
1039 
1040 static void
1041 icmp_tpi_disconnect(queue_t *q, mblk_t *mp)
1042 {
1043 	conn_t	*connp = Q_TO_CONN(q);
1044 	int	error;
1045 
1046 	/*
1047 	 * Allocate the largest primitive we need to send back
1048 	 * T_error_ack is > than T_ok_ack
1049 	 */
1050 	mp = reallocb(mp, sizeof (struct T_error_ack), 1);
1051 	if (mp == NULL) {
1052 		/* Unable to reuse the T_DISCON_REQ for the ack. */
1053 		icmp_err_ack_prim(q, mp, T_DISCON_REQ, TSYSERR, ENOMEM);
1054 		return;
1055 	}
1056 
1057 	error = icmp_do_disconnect(connp);
1058 
1059 	if (error != 0) {
1060 		if (error > 0) {
1061 			icmp_err_ack(q, mp, 0, error);
1062 		} else {
1063 			icmp_err_ack(q, mp, -error, 0);
1064 		}
1065 	} else {
1066 		mp = mi_tpi_ok_ack_alloc(mp);
1067 		ASSERT(mp != NULL);
1068 		qreply(q, mp);
1069 	}
1070 }
1071 
1072 static int
1073 icmp_disconnect(conn_t *connp)
1074 {
1075 	int	error;
1076 
1077 	connp->conn_dgram_errind = B_FALSE;
1078 
1079 	error = icmp_do_disconnect(connp);
1080 
1081 	if (error < 0)
1082 		error = proto_tlitosyserr(-error);
1083 	return (error);
1084 }
1085 
1086 /* This routine creates a T_ERROR_ACK message and passes it upstream. */
1087 static void
1088 icmp_err_ack(queue_t *q, mblk_t *mp, t_scalar_t t_error, int sys_error)
1089 {
1090 	if ((mp = mi_tpi_err_ack_alloc(mp, t_error, sys_error)) != NULL)
1091 		qreply(q, mp);
1092 }
1093 
1094 /* Shorthand to generate and send TPI error acks to our client */
1095 static void
1096 icmp_err_ack_prim(queue_t *q, mblk_t *mp, t_scalar_t primitive,
1097     t_scalar_t t_error, int sys_error)
1098 {
1099 	struct T_error_ack	*teackp;
1100 
1101 	if ((mp = tpi_ack_alloc(mp, sizeof (struct T_error_ack),
1102 	    M_PCPROTO, T_ERROR_ACK)) != NULL) {
1103 		teackp = (struct T_error_ack *)mp->b_rptr;
1104 		teackp->ERROR_prim = primitive;
1105 		teackp->TLI_error = t_error;
1106 		teackp->UNIX_error = sys_error;
1107 		qreply(q, mp);
1108 	}
1109 }
1110 
1111 /*
1112  * icmp_icmp_input is called as conn_recvicmp to process ICMP messages.
1113  * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
1114  * Assumes that IP has pulled up everything up to and including the ICMP header.
1115  */
1116 /* ARGSUSED2 */
1117 static void
1118 icmp_icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
1119 {
1120 	conn_t		*connp = (conn_t *)arg1;
1121 	icmp_t		*icmp = connp->conn_icmp;
1122 	icmph_t		*icmph;
1123 	ipha_t		*ipha;
1124 	int		iph_hdr_length;
1125 	sin_t		sin;
1126 	mblk_t		*mp1;
1127 	int		error = 0;
1128 
1129 	ipha = (ipha_t *)mp->b_rptr;
1130 
1131 	ASSERT(OK_32PTR(mp->b_rptr));
1132 
1133 	if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
1134 		ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
1135 		icmp_icmp_error_ipv6(connp, mp, ira);
1136 		return;
1137 	}
1138 	ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
1139 
1140 	/* Skip past the outer IP and ICMP headers */
1141 	ASSERT(IPH_HDR_LENGTH(ipha) == ira->ira_ip_hdr_length);
1142 	iph_hdr_length = ira->ira_ip_hdr_length;
1143 	icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1144 	ipha = (ipha_t *)&icmph[1];	/* Inner IP header */
1145 
1146 	iph_hdr_length = IPH_HDR_LENGTH(ipha);
1147 
1148 	switch (icmph->icmph_type) {
1149 	case ICMP_DEST_UNREACHABLE:
1150 		switch (icmph->icmph_code) {
1151 		case ICMP_FRAGMENTATION_NEEDED: {
1152 			ipha_t		*ipha;
1153 			ip_xmit_attr_t	*ixa;
1154 			/*
1155 			 * IP has already adjusted the path MTU.
1156 			 * But we need to adjust DF for IPv4.
1157 			 */
1158 			if (connp->conn_ipversion != IPV4_VERSION)
1159 				break;
1160 
1161 			ixa = conn_get_ixa(connp, B_FALSE);
1162 			if (ixa == NULL || ixa->ixa_ire == NULL) {
1163 				/*
1164 				 * Some other thread holds conn_ixa. We will
1165 				 * redo this on the next ICMP too big.
1166 				 */
1167 				if (ixa != NULL)
1168 					ixa_refrele(ixa);
1169 				break;
1170 			}
1171 			(void) ip_get_pmtu(ixa);
1172 
1173 			mutex_enter(&connp->conn_lock);
1174 			ipha = (ipha_t *)connp->conn_ht_iphc;
1175 			if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
1176 				ipha->ipha_fragment_offset_and_flags |=
1177 				    IPH_DF_HTONS;
1178 			} else {
1179 				ipha->ipha_fragment_offset_and_flags &=
1180 				    ~IPH_DF_HTONS;
1181 			}
1182 			mutex_exit(&connp->conn_lock);
1183 			ixa_refrele(ixa);
1184 			break;
1185 		}
1186 		case ICMP_PORT_UNREACHABLE:
1187 		case ICMP_PROTOCOL_UNREACHABLE:
1188 			error = ECONNREFUSED;
1189 			break;
1190 		default:
1191 			/* Transient errors */
1192 			break;
1193 		}
1194 		break;
1195 	default:
1196 		/* Transient errors */
1197 		break;
1198 	}
1199 	if (error == 0) {
1200 		freemsg(mp);
1201 		return;
1202 	}
1203 
1204 	/*
1205 	 * Deliver T_UDERROR_IND when the application has asked for it.
1206 	 * The socket layer enables this automatically when connected.
1207 	 */
1208 	if (!connp->conn_dgram_errind) {
1209 		freemsg(mp);
1210 		return;
1211 	}
1212 
1213 	sin = sin_null;
1214 	sin.sin_family = AF_INET;
1215 	sin.sin_addr.s_addr = ipha->ipha_dst;
1216 
1217 	if (IPCL_IS_NONSTR(connp)) {
1218 		mutex_enter(&connp->conn_lock);
1219 		if (icmp->icmp_state == TS_DATA_XFER) {
1220 			if (sin.sin_addr.s_addr == connp->conn_faddr_v4) {
1221 				mutex_exit(&connp->conn_lock);
1222 				(*connp->conn_upcalls->su_set_error)
1223 				    (connp->conn_upper_handle, error);
1224 				goto done;
1225 			}
1226 		} else {
1227 			icmp->icmp_delayed_error = error;
1228 			*((sin_t *)&icmp->icmp_delayed_addr) = sin;
1229 		}
1230 		mutex_exit(&connp->conn_lock);
1231 	} else {
1232 		mp1 = mi_tpi_uderror_ind((char *)&sin, sizeof (sin_t), NULL, 0,
1233 		    error);
1234 		if (mp1 != NULL)
1235 			putnext(connp->conn_rq, mp1);
1236 	}
1237 done:
1238 	freemsg(mp);
1239 }
1240 
1241 /*
1242  * icmp_icmp_error_ipv6 is called by icmp_icmp_error to process ICMP for IPv6.
1243  * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
1244  * Assumes that IP has pulled up all the extension headers as well as the
1245  * ICMPv6 header.
1246  */
1247 static void
1248 icmp_icmp_error_ipv6(conn_t *connp, mblk_t *mp, ip_recv_attr_t *ira)
1249 {
1250 	icmp6_t		*icmp6;
1251 	ip6_t		*ip6h, *outer_ip6h;
1252 	uint16_t	iph_hdr_length;
1253 	uint8_t		*nexthdrp;
1254 	sin6_t		sin6;
1255 	mblk_t		*mp1;
1256 	int		error = 0;
1257 	icmp_t		*icmp = connp->conn_icmp;
1258 
1259 	outer_ip6h = (ip6_t *)mp->b_rptr;
1260 #ifdef DEBUG
1261 	if (outer_ip6h->ip6_nxt != IPPROTO_ICMPV6)
1262 		iph_hdr_length = ip_hdr_length_v6(mp, outer_ip6h);
1263 	else
1264 		iph_hdr_length = IPV6_HDR_LEN;
1265 	ASSERT(iph_hdr_length == ira->ira_ip_hdr_length);
1266 #endif
1267 	/* Skip past the outer IP and ICMP headers */
1268 	iph_hdr_length = ira->ira_ip_hdr_length;
1269 	icmp6 = (icmp6_t *)&mp->b_rptr[iph_hdr_length];
1270 
1271 	ip6h = (ip6_t *)&icmp6[1];	/* Inner IP header */
1272 	if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length, &nexthdrp)) {
1273 		freemsg(mp);
1274 		return;
1275 	}
1276 
1277 	switch (icmp6->icmp6_type) {
1278 	case ICMP6_DST_UNREACH:
1279 		switch (icmp6->icmp6_code) {
1280 		case ICMP6_DST_UNREACH_NOPORT:
1281 			error = ECONNREFUSED;
1282 			break;
1283 		case ICMP6_DST_UNREACH_ADMIN:
1284 		case ICMP6_DST_UNREACH_NOROUTE:
1285 		case ICMP6_DST_UNREACH_BEYONDSCOPE:
1286 		case ICMP6_DST_UNREACH_ADDR:
1287 			/* Transient errors */
1288 			break;
1289 		default:
1290 			break;
1291 		}
1292 		break;
1293 	case ICMP6_PACKET_TOO_BIG: {
1294 		struct T_unitdata_ind	*tudi;
1295 		struct T_opthdr		*toh;
1296 		size_t			udi_size;
1297 		mblk_t			*newmp;
1298 		t_scalar_t		opt_length = sizeof (struct T_opthdr) +
1299 		    sizeof (struct ip6_mtuinfo);
1300 		sin6_t			*sin6;
1301 		struct ip6_mtuinfo	*mtuinfo;
1302 
1303 		/*
1304 		 * If the application has requested to receive path mtu
1305 		 * information, send up an empty message containing an
1306 		 * IPV6_PATHMTU ancillary data item.
1307 		 */
1308 		if (!connp->conn_ipv6_recvpathmtu)
1309 			break;
1310 
1311 		udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t) +
1312 		    opt_length;
1313 		if ((newmp = allocb(udi_size, BPRI_MED)) == NULL) {
1314 			BUMP_MIB(&icmp->icmp_is->is_rawip_mib, rawipInErrors);
1315 			break;
1316 		}
1317 
1318 		/*
1319 		 * newmp->b_cont is left to NULL on purpose.  This is an
1320 		 * empty message containing only ancillary data.
1321 		 */
1322 		newmp->b_datap->db_type = M_PROTO;
1323 		tudi = (struct T_unitdata_ind *)newmp->b_rptr;
1324 		newmp->b_wptr = (uchar_t *)tudi + udi_size;
1325 		tudi->PRIM_type = T_UNITDATA_IND;
1326 		tudi->SRC_length = sizeof (sin6_t);
1327 		tudi->SRC_offset = sizeof (struct T_unitdata_ind);
1328 		tudi->OPT_offset = tudi->SRC_offset + sizeof (sin6_t);
1329 		tudi->OPT_length = opt_length;
1330 
1331 		sin6 = (sin6_t *)&tudi[1];
1332 		bzero(sin6, sizeof (sin6_t));
1333 		sin6->sin6_family = AF_INET6;
1334 		sin6->sin6_addr = connp->conn_faddr_v6;
1335 
1336 		toh = (struct T_opthdr *)&sin6[1];
1337 		toh->level = IPPROTO_IPV6;
1338 		toh->name = IPV6_PATHMTU;
1339 		toh->len = opt_length;
1340 		toh->status = 0;
1341 
1342 		mtuinfo = (struct ip6_mtuinfo *)&toh[1];
1343 		bzero(mtuinfo, sizeof (struct ip6_mtuinfo));
1344 		mtuinfo->ip6m_addr.sin6_family = AF_INET6;
1345 		mtuinfo->ip6m_addr.sin6_addr = ip6h->ip6_dst;
1346 		mtuinfo->ip6m_mtu = icmp6->icmp6_mtu;
1347 		/*
1348 		 * We've consumed everything we need from the original
1349 		 * message.  Free it, then send our empty message.
1350 		 */
1351 		freemsg(mp);
1352 		icmp_ulp_recv(connp, newmp, msgdsize(newmp));
1353 		return;
1354 	}
1355 	case ICMP6_TIME_EXCEEDED:
1356 		/* Transient errors */
1357 		break;
1358 	case ICMP6_PARAM_PROB:
1359 		/* If this corresponds to an ICMP_PROTOCOL_UNREACHABLE */
1360 		if (icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER &&
1361 		    (uchar_t *)ip6h + icmp6->icmp6_pptr ==
1362 		    (uchar_t *)nexthdrp) {
1363 			error = ECONNREFUSED;
1364 			break;
1365 		}
1366 		break;
1367 	}
1368 	if (error == 0) {
1369 		freemsg(mp);
1370 		return;
1371 	}
1372 
1373 	/*
1374 	 * Deliver T_UDERROR_IND when the application has asked for it.
1375 	 * The socket layer enables this automatically when connected.
1376 	 */
1377 	if (!connp->conn_dgram_errind) {
1378 		freemsg(mp);
1379 		return;
1380 	}
1381 
1382 	sin6 = sin6_null;
1383 	sin6.sin6_family = AF_INET6;
1384 	sin6.sin6_addr = ip6h->ip6_dst;
1385 	sin6.sin6_flowinfo = ip6h->ip6_vcf & ~IPV6_VERS_AND_FLOW_MASK;
1386 	if (IPCL_IS_NONSTR(connp)) {
1387 		mutex_enter(&connp->conn_lock);
1388 		if (icmp->icmp_state == TS_DATA_XFER) {
1389 			if (IN6_ARE_ADDR_EQUAL(&sin6.sin6_addr,
1390 			    &connp->conn_faddr_v6)) {
1391 				mutex_exit(&connp->conn_lock);
1392 				(*connp->conn_upcalls->su_set_error)
1393 				    (connp->conn_upper_handle, error);
1394 				goto done;
1395 			}
1396 		} else {
1397 			icmp->icmp_delayed_error = error;
1398 			*((sin6_t *)&icmp->icmp_delayed_addr) = sin6;
1399 		}
1400 		mutex_exit(&connp->conn_lock);
1401 	} else {
1402 		mp1 = mi_tpi_uderror_ind((char *)&sin6, sizeof (sin6_t),
1403 		    NULL, 0, error);
1404 		if (mp1 != NULL)
1405 			putnext(connp->conn_rq, mp1);
1406 	}
1407 done:
1408 	freemsg(mp);
1409 }
1410 
1411 /*
1412  * This routine responds to T_ADDR_REQ messages.  It is called by icmp_wput.
1413  * The local address is filled in if endpoint is bound. The remote address
1414  * is filled in if remote address has been precified ("connected endpoint")
1415  * (The concept of connected CLTS sockets is alien to published TPI
1416  *  but we support it anyway).
1417  */
1418 static void
1419 icmp_addr_req(queue_t *q, mblk_t *mp)
1420 {
1421 	struct sockaddr *sa;
1422 	mblk_t	*ackmp;
1423 	struct T_addr_ack *taa;
1424 	icmp_t	*icmp = Q_TO_ICMP(q);
1425 	conn_t	*connp = icmp->icmp_connp;
1426 	uint_t	addrlen;
1427 
1428 	/* Make it large enough for worst case */
1429 	ackmp = reallocb(mp, sizeof (struct T_addr_ack) +
1430 	    2 * sizeof (sin6_t), 1);
1431 	if (ackmp == NULL) {
1432 		icmp_err_ack(q, mp, TSYSERR, ENOMEM);
1433 		return;
1434 	}
1435 	taa = (struct T_addr_ack *)ackmp->b_rptr;
1436 
1437 	bzero(taa, sizeof (struct T_addr_ack));
1438 	ackmp->b_wptr = (uchar_t *)&taa[1];
1439 
1440 	taa->PRIM_type = T_ADDR_ACK;
1441 	ackmp->b_datap->db_type = M_PCPROTO;
1442 
1443 	if (connp->conn_family == AF_INET)
1444 		addrlen = sizeof (sin_t);
1445 	else
1446 		addrlen = sizeof (sin6_t);
1447 
1448 	mutex_enter(&connp->conn_lock);
1449 	/*
1450 	 * Note: Following code assumes 32 bit alignment of basic
1451 	 * data structures like sin_t and struct T_addr_ack.
1452 	 */
1453 	if (icmp->icmp_state != TS_UNBND) {
1454 		/*
1455 		 * Fill in local address first
1456 		 */
1457 		taa->LOCADDR_offset = sizeof (*taa);
1458 		taa->LOCADDR_length = addrlen;
1459 		sa = (struct sockaddr *)&taa[1];
1460 		(void) conn_getsockname(connp, sa, &addrlen);
1461 		ackmp->b_wptr += addrlen;
1462 	}
1463 	if (icmp->icmp_state == TS_DATA_XFER) {
1464 		/*
1465 		 * connected, fill remote address too
1466 		 */
1467 		taa->REMADDR_length = addrlen;
1468 		/* assumed 32-bit alignment */
1469 		taa->REMADDR_offset = taa->LOCADDR_offset + taa->LOCADDR_length;
1470 		sa = (struct sockaddr *)(ackmp->b_rptr + taa->REMADDR_offset);
1471 		(void) conn_getpeername(connp, sa, &addrlen);
1472 		ackmp->b_wptr += addrlen;
1473 	}
1474 	mutex_exit(&connp->conn_lock);
1475 	ASSERT(ackmp->b_wptr <= ackmp->b_datap->db_lim);
1476 	qreply(q, ackmp);
1477 }
1478 
1479 static void
1480 icmp_copy_info(struct T_info_ack *tap, icmp_t *icmp)
1481 {
1482 	conn_t		*connp = icmp->icmp_connp;
1483 
1484 	*tap = icmp_g_t_info_ack;
1485 
1486 	if (connp->conn_family == AF_INET6)
1487 		tap->ADDR_size = sizeof (sin6_t);
1488 	else
1489 		tap->ADDR_size = sizeof (sin_t);
1490 	tap->CURRENT_state = icmp->icmp_state;
1491 	tap->OPT_size = icmp_max_optsize;
1492 }
1493 
1494 static void
1495 icmp_do_capability_ack(icmp_t *icmp, struct T_capability_ack *tcap,
1496     t_uscalar_t cap_bits1)
1497 {
1498 	tcap->CAP_bits1 = 0;
1499 
1500 	if (cap_bits1 & TC1_INFO) {
1501 		icmp_copy_info(&tcap->INFO_ack, icmp);
1502 		tcap->CAP_bits1 |= TC1_INFO;
1503 	}
1504 }
1505 
1506 /*
1507  * This routine responds to T_CAPABILITY_REQ messages.  It is called by
1508  * icmp_wput.  Much of the T_CAPABILITY_ACK information is copied from
1509  * icmp_g_t_info_ack.  The current state of the stream is copied from
1510  * icmp_state.
1511  */
1512 static void
1513 icmp_capability_req(queue_t *q, mblk_t *mp)
1514 {
1515 	icmp_t			*icmp = Q_TO_ICMP(q);
1516 	t_uscalar_t		cap_bits1;
1517 	struct T_capability_ack	*tcap;
1518 
1519 	cap_bits1 = ((struct T_capability_req *)mp->b_rptr)->CAP_bits1;
1520 
1521 	mp = tpi_ack_alloc(mp, sizeof (struct T_capability_ack),
1522 	    mp->b_datap->db_type, T_CAPABILITY_ACK);
1523 	if (!mp)
1524 		return;
1525 
1526 	tcap = (struct T_capability_ack *)mp->b_rptr;
1527 
1528 	icmp_do_capability_ack(icmp, tcap, cap_bits1);
1529 
1530 	qreply(q, mp);
1531 }
1532 
1533 /*
1534  * This routine responds to T_INFO_REQ messages.  It is called by icmp_wput.
1535  * Most of the T_INFO_ACK information is copied from icmp_g_t_info_ack.
1536  * The current state of the stream is copied from icmp_state.
1537  */
1538 static void
1539 icmp_info_req(queue_t *q, mblk_t *mp)
1540 {
1541 	icmp_t	*icmp = Q_TO_ICMP(q);
1542 
1543 	/* Create a T_INFO_ACK message. */
1544 	mp = tpi_ack_alloc(mp, sizeof (struct T_info_ack), M_PCPROTO,
1545 	    T_INFO_ACK);
1546 	if (!mp)
1547 		return;
1548 	icmp_copy_info((struct T_info_ack *)mp->b_rptr, icmp);
1549 	qreply(q, mp);
1550 }
1551 
1552 static int
1553 icmp_tpi_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
1554     int family)
1555 {
1556 	conn_t *connp;
1557 	dev_t	conn_dev;
1558 	int	error;
1559 
1560 	/* If the stream is already open, return immediately. */
1561 	if (q->q_ptr != NULL)
1562 		return (0);
1563 
1564 	if (sflag == MODOPEN)
1565 		return (EINVAL);
1566 
1567 	/*
1568 	 * Since ICMP is not used so heavily, allocating from the small
1569 	 * arena should be sufficient.
1570 	 */
1571 	if ((conn_dev = inet_minor_alloc(ip_minor_arena_sa)) == 0) {
1572 		return (EBUSY);
1573 	}
1574 
1575 	if (flag & SO_FALLBACK) {
1576 		/*
1577 		 * Non streams socket needs a stream to fallback to
1578 		 */
1579 		RD(q)->q_ptr = (void *)conn_dev;
1580 		WR(q)->q_qinfo = &icmp_fallback_sock_winit;
1581 		WR(q)->q_ptr = (void *)ip_minor_arena_sa;
1582 		qprocson(q);
1583 		return (0);
1584 	}
1585 
1586 	connp = rawip_do_open(family, credp, &error, KM_SLEEP);
1587 	if (connp == NULL) {
1588 		ASSERT(error != 0);
1589 		inet_minor_free(ip_minor_arena_sa, connp->conn_dev);
1590 		return (error);
1591 	}
1592 
1593 	*devp = makedevice(getemajor(*devp), (minor_t)conn_dev);
1594 	connp->conn_dev = conn_dev;
1595 	connp->conn_minor_arena = ip_minor_arena_sa;
1596 
1597 	/*
1598 	 * Initialize the icmp_t structure for this stream.
1599 	 */
1600 	q->q_ptr = connp;
1601 	WR(q)->q_ptr = connp;
1602 	connp->conn_rq = q;
1603 	connp->conn_wq = WR(q);
1604 
1605 	WR(q)->q_hiwat = connp->conn_sndbuf;
1606 	WR(q)->q_lowat = connp->conn_sndlowat;
1607 
1608 	qprocson(q);
1609 
1610 	/* Set the Stream head write offset. */
1611 	(void) proto_set_tx_wroff(q, connp, connp->conn_wroff);
1612 	(void) proto_set_rx_hiwat(connp->conn_rq, connp, connp->conn_rcvbuf);
1613 
1614 	mutex_enter(&connp->conn_lock);
1615 	connp->conn_state_flags &= ~CONN_INCIPIENT;
1616 	mutex_exit(&connp->conn_lock);
1617 
1618 	icmp_bind_proto(connp->conn_icmp);
1619 
1620 	return (0);
1621 }
1622 
1623 /* For /dev/icmp aka AF_INET open */
1624 static int
1625 icmp_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1626 {
1627 	return (icmp_tpi_open(q, devp, flag, sflag, credp, AF_INET));
1628 }
1629 
1630 /* For /dev/icmp6 aka AF_INET6 open */
1631 static int
1632 icmp_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1633 {
1634 	return (icmp_tpi_open(q, devp, flag, sflag, credp, AF_INET6));
1635 }
1636 
1637 /*
1638  * This is the open routine for icmp.  It allocates a icmp_t structure for
1639  * the stream and, on the first open of the module, creates an ND table.
1640  */
1641 static conn_t *
1642 rawip_do_open(int family, cred_t *credp, int *err, int flags)
1643 {
1644 	icmp_t	*icmp;
1645 	conn_t *connp;
1646 	zoneid_t zoneid;
1647 	netstack_t *ns;
1648 	icmp_stack_t *is;
1649 	int len;
1650 	boolean_t isv6 = B_FALSE;
1651 
1652 	*err = secpolicy_net_icmpaccess(credp);
1653 	if (*err != 0)
1654 		return (NULL);
1655 
1656 	if (family == AF_INET6)
1657 		isv6 = B_TRUE;
1658 
1659 	ns = netstack_find_by_cred(credp);
1660 	ASSERT(ns != NULL);
1661 	is = ns->netstack_icmp;
1662 	ASSERT(is != NULL);
1663 
1664 	/*
1665 	 * For exclusive stacks we set the zoneid to zero
1666 	 * to make ICMP operate as if in the global zone.
1667 	 */
1668 	if (ns->netstack_stackid != GLOBAL_NETSTACKID)
1669 		zoneid = GLOBAL_ZONEID;
1670 	else
1671 		zoneid = crgetzoneid(credp);
1672 
1673 	ASSERT(flags == KM_SLEEP || flags == KM_NOSLEEP);
1674 
1675 	connp = ipcl_conn_create(IPCL_RAWIPCONN, flags, ns);
1676 	icmp = connp->conn_icmp;
1677 
1678 	/*
1679 	 * ipcl_conn_create did a netstack_hold. Undo the hold that was
1680 	 * done by netstack_find_by_cred()
1681 	 */
1682 	netstack_rele(ns);
1683 
1684 	/*
1685 	 * Since this conn_t/icmp_t is not yet visible to anybody else we don't
1686 	 * need to lock anything.
1687 	 */
1688 	ASSERT(connp->conn_proto == IPPROTO_ICMP);
1689 	ASSERT(connp->conn_icmp == icmp);
1690 	ASSERT(icmp->icmp_connp == connp);
1691 
1692 	/* Set the initial state of the stream and the privilege status. */
1693 	icmp->icmp_state = TS_UNBND;
1694 	connp->conn_ixa->ixa_flags |= IXAF_VERIFY_SOURCE;
1695 	if (isv6) {
1696 		connp->conn_family = AF_INET6;
1697 		connp->conn_ipversion = IPV6_VERSION;
1698 		connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
1699 		connp->conn_proto = IPPROTO_ICMPV6;
1700 		/* May be changed by a SO_PROTOTYPE socket option. */
1701 		connp->conn_proto = IPPROTO_ICMPV6;
1702 		connp->conn_ixa->ixa_protocol = connp->conn_proto;
1703 		connp->conn_ixa->ixa_raw_cksum_offset = 2;
1704 		connp->conn_default_ttl = is->is_ipv6_hoplimit;
1705 		len = sizeof (ip6_t);
1706 	} else {
1707 		connp->conn_family = AF_INET;
1708 		connp->conn_ipversion = IPV4_VERSION;
1709 		connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
1710 		/* May be changed by a SO_PROTOTYPE socket option. */
1711 		connp->conn_proto = IPPROTO_ICMP;
1712 		connp->conn_ixa->ixa_protocol = connp->conn_proto;
1713 		connp->conn_default_ttl = is->is_ipv4_ttl;
1714 		len = sizeof (ipha_t);
1715 	}
1716 	connp->conn_xmit_ipp.ipp_unicast_hops = connp->conn_default_ttl;
1717 
1718 	connp->conn_ixa->ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1719 
1720 	/*
1721 	 * For the socket of protocol IPPROTO_RAW or when IP_HDRINCL is set,
1722 	 * the checksum is provided in the pre-built packet. We clear
1723 	 * IXAF_SET_ULP_CKSUM to tell IP that the application has sent a
1724 	 * complete IP header and not to compute the transport checksum.
1725 	 */
1726 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
1727 	/* conn_allzones can not be set this early, hence no IPCL_ZONEID */
1728 	connp->conn_ixa->ixa_zoneid = zoneid;
1729 
1730 	connp->conn_zoneid = zoneid;
1731 
1732 	/*
1733 	 * If the caller has the process-wide flag set, then default to MAC
1734 	 * exempt mode.  This allows read-down to unlabeled hosts.
1735 	 */
1736 	if (getpflags(NET_MAC_AWARE, credp) != 0)
1737 		connp->conn_mac_mode = CONN_MAC_AWARE;
1738 
1739 	connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
1740 
1741 	icmp->icmp_is = is;
1742 
1743 	connp->conn_rcvbuf = is->is_recv_hiwat;
1744 	connp->conn_sndbuf = is->is_xmit_hiwat;
1745 	connp->conn_sndlowat = is->is_xmit_lowat;
1746 	connp->conn_rcvlowat = icmp_mod_info.mi_lowat;
1747 
1748 	connp->conn_wroff = len + is->is_wroff_extra;
1749 	connp->conn_so_type = SOCK_RAW;
1750 
1751 	connp->conn_recv = icmp_input;
1752 	connp->conn_recvicmp = icmp_icmp_input;
1753 	crhold(credp);
1754 	connp->conn_cred = credp;
1755 	connp->conn_cpid = curproc->p_pid;
1756 	connp->conn_open_time = ddi_get_lbolt64();
1757 	/* Cache things in ixa without an extra refhold */
1758 	ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
1759 	connp->conn_ixa->ixa_cred = connp->conn_cred;
1760 	connp->conn_ixa->ixa_cpid = connp->conn_cpid;
1761 	if (is_system_labeled())
1762 		connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
1763 
1764 	connp->conn_flow_cntrld = B_FALSE;
1765 
1766 	if (is->is_pmtu_discovery)
1767 		connp->conn_ixa->ixa_flags |= IXAF_PMTU_DISCOVERY;
1768 
1769 	return (connp);
1770 }
1771 
1772 /*
1773  * Which ICMP options OK to set through T_UNITDATA_REQ...
1774  */
1775 /* ARGSUSED */
1776 static boolean_t
1777 icmp_opt_allow_udr_set(t_scalar_t level, t_scalar_t name)
1778 {
1779 	return (B_TRUE);
1780 }
1781 
1782 /*
1783  * This routine gets default values of certain options whose default
1784  * values are maintained by protcol specific code
1785  */
1786 int
1787 icmp_opt_default(queue_t *q, t_scalar_t level, t_scalar_t name, uchar_t *ptr)
1788 {
1789 	icmp_t *icmp = Q_TO_ICMP(q);
1790 	icmp_stack_t *is = icmp->icmp_is;
1791 	int *i1 = (int *)ptr;
1792 
1793 	switch (level) {
1794 	case IPPROTO_IP:
1795 		switch (name) {
1796 		case IP_MULTICAST_TTL:
1797 			*ptr = (uchar_t)IP_DEFAULT_MULTICAST_TTL;
1798 			return (sizeof (uchar_t));
1799 		case IP_MULTICAST_LOOP:
1800 			*ptr = (uchar_t)IP_DEFAULT_MULTICAST_LOOP;
1801 			return (sizeof (uchar_t));
1802 		}
1803 		break;
1804 	case IPPROTO_IPV6:
1805 		switch (name) {
1806 		case IPV6_MULTICAST_HOPS:
1807 			*i1 = IP_DEFAULT_MULTICAST_TTL;
1808 			return (sizeof (int));
1809 		case IPV6_MULTICAST_LOOP:
1810 			*i1 = IP_DEFAULT_MULTICAST_LOOP;
1811 			return (sizeof (int));
1812 		case IPV6_UNICAST_HOPS:
1813 			*i1 = is->is_ipv6_hoplimit;
1814 			return (sizeof (int));
1815 		}
1816 		break;
1817 	case IPPROTO_ICMPV6:
1818 		switch (name) {
1819 		case ICMP6_FILTER:
1820 			/* Make it look like "pass all" */
1821 			ICMP6_FILTER_SETPASSALL((icmp6_filter_t *)ptr);
1822 			return (sizeof (icmp6_filter_t));
1823 		}
1824 		break;
1825 	}
1826 	return (-1);
1827 }
1828 
1829 /*
1830  * This routine retrieves the current status of socket options.
1831  * It returns the size of the option retrieved, or -1.
1832  */
1833 int
1834 icmp_opt_get(conn_t *connp, int level, int name, uchar_t *ptr)
1835 {
1836 	icmp_t		*icmp = connp->conn_icmp;
1837 	int		*i1 = (int *)ptr;
1838 	conn_opt_arg_t	coas;
1839 	int		retval;
1840 
1841 	coas.coa_connp = connp;
1842 	coas.coa_ixa = connp->conn_ixa;
1843 	coas.coa_ipp = &connp->conn_xmit_ipp;
1844 	coas.coa_ancillary = B_FALSE;
1845 	coas.coa_changed = 0;
1846 
1847 	/*
1848 	 * We assume that the optcom framework has checked for the set
1849 	 * of levels and names that are supported, hence we don't worry
1850 	 * about rejecting based on that.
1851 	 * First check for ICMP specific handling, then pass to common routine.
1852 	 */
1853 	switch (level) {
1854 	case IPPROTO_IP:
1855 		/*
1856 		 * Only allow IPv4 option processing on IPv4 sockets.
1857 		 */
1858 		if (connp->conn_family != AF_INET)
1859 			return (-1);
1860 
1861 		switch (name) {
1862 		case IP_OPTIONS:
1863 		case T_IP_OPTIONS:
1864 			/* Options are passed up with each packet */
1865 			return (0);
1866 		case IP_HDRINCL:
1867 			mutex_enter(&connp->conn_lock);
1868 			*i1 = (int)icmp->icmp_hdrincl;
1869 			mutex_exit(&connp->conn_lock);
1870 			return (sizeof (int));
1871 		}
1872 		break;
1873 
1874 	case IPPROTO_IPV6:
1875 		/*
1876 		 * Only allow IPv6 option processing on native IPv6 sockets.
1877 		 */
1878 		if (connp->conn_family != AF_INET6)
1879 			return (-1);
1880 
1881 		switch (name) {
1882 		case IPV6_CHECKSUM:
1883 			/*
1884 			 * Return offset or -1 if no checksum offset.
1885 			 * Does not apply to IPPROTO_ICMPV6
1886 			 */
1887 			if (connp->conn_proto == IPPROTO_ICMPV6)
1888 				return (-1);
1889 
1890 			mutex_enter(&connp->conn_lock);
1891 			if (connp->conn_ixa->ixa_flags & IXAF_SET_RAW_CKSUM)
1892 				*i1 = connp->conn_ixa->ixa_raw_cksum_offset;
1893 			else
1894 				*i1 = -1;
1895 			mutex_exit(&connp->conn_lock);
1896 			return (sizeof (int));
1897 		}
1898 		break;
1899 
1900 	case IPPROTO_ICMPV6:
1901 		/*
1902 		 * Only allow IPv6 option processing on native IPv6 sockets.
1903 		 */
1904 		if (connp->conn_family != AF_INET6)
1905 			return (-1);
1906 
1907 		if (connp->conn_proto != IPPROTO_ICMPV6)
1908 			return (-1);
1909 
1910 		switch (name) {
1911 		case ICMP6_FILTER:
1912 			mutex_enter(&connp->conn_lock);
1913 			if (icmp->icmp_filter == NULL) {
1914 				/* Make it look like "pass all" */
1915 				ICMP6_FILTER_SETPASSALL((icmp6_filter_t *)ptr);
1916 			} else {
1917 				(void) bcopy(icmp->icmp_filter, ptr,
1918 				    sizeof (icmp6_filter_t));
1919 			}
1920 			mutex_exit(&connp->conn_lock);
1921 			return (sizeof (icmp6_filter_t));
1922 		}
1923 	}
1924 	mutex_enter(&connp->conn_lock);
1925 	retval = conn_opt_get(&coas, level, name, ptr);
1926 	mutex_exit(&connp->conn_lock);
1927 	return (retval);
1928 }
1929 
1930 /*
1931  * This routine retrieves the current status of socket options.
1932  * It returns the size of the option retrieved, or -1.
1933  */
1934 int
1935 icmp_tpi_opt_get(queue_t *q, int level, int name, uchar_t *ptr)
1936 {
1937 	conn_t		*connp = Q_TO_CONN(q);
1938 	int 		err;
1939 
1940 	err = icmp_opt_get(connp, level, name, ptr);
1941 	return (err);
1942 }
1943 
1944 /*
1945  * This routine sets socket options.
1946  */
1947 int
1948 icmp_do_opt_set(conn_opt_arg_t *coa, int level, int name,
1949     uint_t inlen, uchar_t *invalp, cred_t *cr, boolean_t checkonly)
1950 {
1951 	conn_t		*connp = coa->coa_connp;
1952 	ip_xmit_attr_t	*ixa = coa->coa_ixa;
1953 	icmp_t		*icmp = connp->conn_icmp;
1954 	icmp_stack_t	*is = icmp->icmp_is;
1955 	int		*i1 = (int *)invalp;
1956 	boolean_t	onoff = (*i1 == 0) ? 0 : 1;
1957 	int		error;
1958 
1959 	ASSERT(MUTEX_NOT_HELD(&coa->coa_connp->conn_lock));
1960 
1961 	/*
1962 	 * For fixed length options, no sanity check
1963 	 * of passed in length is done. It is assumed *_optcom_req()
1964 	 * routines do the right thing.
1965 	 */
1966 
1967 	switch (level) {
1968 	case SOL_SOCKET:
1969 		switch (name) {
1970 		case SO_PROTOTYPE:
1971 			if ((*i1 & 0xFF) != IPPROTO_ICMP &&
1972 			    (*i1 & 0xFF) != IPPROTO_ICMPV6 &&
1973 			    secpolicy_net_rawaccess(cr) != 0) {
1974 				return (EACCES);
1975 			}
1976 			if (checkonly)
1977 				break;
1978 
1979 			mutex_enter(&connp->conn_lock);
1980 			connp->conn_proto = *i1 & 0xFF;
1981 			ixa->ixa_protocol = connp->conn_proto;
1982 			if ((connp->conn_proto == IPPROTO_RAW ||
1983 			    connp->conn_proto == IPPROTO_IGMP) &&
1984 			    connp->conn_family == AF_INET) {
1985 				icmp->icmp_hdrincl = 1;
1986 				ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
1987 			} else if (connp->conn_proto == IPPROTO_UDP ||
1988 			    connp->conn_proto == IPPROTO_TCP ||
1989 			    connp->conn_proto == IPPROTO_SCTP) {
1990 				/* Used by test applications like psh */
1991 				icmp->icmp_hdrincl = 0;
1992 				ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
1993 			} else {
1994 				icmp->icmp_hdrincl = 0;
1995 				ixa->ixa_flags |= IXAF_SET_ULP_CKSUM;
1996 			}
1997 
1998 			if (connp->conn_family == AF_INET6 &&
1999 			    connp->conn_proto == IPPROTO_ICMPV6) {
2000 				/* Set offset for icmp6_cksum */
2001 				ixa->ixa_flags &= ~IXAF_SET_RAW_CKSUM;
2002 				ixa->ixa_raw_cksum_offset = 2;
2003 			}
2004 			if (icmp->icmp_filter != NULL &&
2005 			    connp->conn_proto != IPPROTO_ICMPV6) {
2006 				kmem_free(icmp->icmp_filter,
2007 				    sizeof (icmp6_filter_t));
2008 				icmp->icmp_filter = NULL;
2009 			}
2010 			mutex_exit(&connp->conn_lock);
2011 
2012 			coa->coa_changed |= COA_HEADER_CHANGED;
2013 			/*
2014 			 * For SCTP, we don't use icmp_bind_proto() for
2015 			 * raw socket binding.
2016 			 */
2017 			if (connp->conn_proto == IPPROTO_SCTP)
2018 				return (0);
2019 
2020 			coa->coa_changed |= COA_ICMP_BIND_NEEDED;
2021 			return (0);
2022 
2023 		case SO_SNDBUF:
2024 			if (*i1 > is->is_max_buf) {
2025 				return (ENOBUFS);
2026 			}
2027 			break;
2028 		case SO_RCVBUF:
2029 			if (*i1 > is->is_max_buf) {
2030 				return (ENOBUFS);
2031 			}
2032 			break;
2033 		}
2034 		break;
2035 
2036 	case IPPROTO_IP:
2037 		/*
2038 		 * Only allow IPv4 option processing on IPv4 sockets.
2039 		 */
2040 		if (connp->conn_family != AF_INET)
2041 			return (EINVAL);
2042 
2043 		switch (name) {
2044 		case IP_HDRINCL:
2045 			if (!checkonly) {
2046 				mutex_enter(&connp->conn_lock);
2047 				icmp->icmp_hdrincl = onoff;
2048 				if (onoff)
2049 					ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
2050 				else
2051 					ixa->ixa_flags |= IXAF_SET_ULP_CKSUM;
2052 				mutex_exit(&connp->conn_lock);
2053 			}
2054 			break;
2055 		}
2056 		break;
2057 
2058 	case IPPROTO_IPV6:
2059 		if (connp->conn_family != AF_INET6)
2060 			return (EINVAL);
2061 
2062 		switch (name) {
2063 		case IPV6_CHECKSUM:
2064 			/*
2065 			 * Integer offset into the user data of where the
2066 			 * checksum is located.
2067 			 * Offset of -1 disables option.
2068 			 * Does not apply to IPPROTO_ICMPV6.
2069 			 */
2070 			if (connp->conn_proto == IPPROTO_ICMPV6 ||
2071 			    coa->coa_ancillary) {
2072 				return (EINVAL);
2073 			}
2074 			if ((*i1 != -1) && ((*i1 < 0) || (*i1 & 0x1) != 0)) {
2075 				/* Negative or not 16 bit aligned offset */
2076 				return (EINVAL);
2077 			}
2078 			if (checkonly)
2079 				break;
2080 
2081 			mutex_enter(&connp->conn_lock);
2082 			if (*i1 == -1) {
2083 				ixa->ixa_flags &= ~IXAF_SET_RAW_CKSUM;
2084 				ixa->ixa_raw_cksum_offset = 0;
2085 				ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
2086 			} else {
2087 				ixa->ixa_flags |= IXAF_SET_RAW_CKSUM;
2088 				ixa->ixa_raw_cksum_offset = *i1;
2089 				ixa->ixa_flags |= IXAF_SET_ULP_CKSUM;
2090 			}
2091 			mutex_exit(&connp->conn_lock);
2092 			break;
2093 		}
2094 		break;
2095 
2096 	case IPPROTO_ICMPV6:
2097 		/*
2098 		 * Only allow IPv6 option processing on IPv6 sockets.
2099 		 */
2100 		if (connp->conn_family != AF_INET6)
2101 			return (EINVAL);
2102 		if (connp->conn_proto != IPPROTO_ICMPV6)
2103 			return (EINVAL);
2104 
2105 		switch (name) {
2106 		case ICMP6_FILTER:
2107 			if (checkonly)
2108 				break;
2109 
2110 			if ((inlen != 0) &&
2111 			    (inlen != sizeof (icmp6_filter_t)))
2112 				return (EINVAL);
2113 
2114 			mutex_enter(&connp->conn_lock);
2115 			if (inlen == 0) {
2116 				if (icmp->icmp_filter != NULL) {
2117 					kmem_free(icmp->icmp_filter,
2118 					    sizeof (icmp6_filter_t));
2119 					icmp->icmp_filter = NULL;
2120 				}
2121 			} else {
2122 				if (icmp->icmp_filter == NULL) {
2123 					icmp->icmp_filter = kmem_alloc(
2124 					    sizeof (icmp6_filter_t),
2125 					    KM_NOSLEEP);
2126 					if (icmp->icmp_filter == NULL) {
2127 						mutex_exit(&connp->conn_lock);
2128 						return (ENOBUFS);
2129 					}
2130 				}
2131 				(void) bcopy(invalp, icmp->icmp_filter, inlen);
2132 			}
2133 			mutex_exit(&connp->conn_lock);
2134 			break;
2135 		}
2136 		break;
2137 	}
2138 	error = conn_opt_set(coa, level, name, inlen, invalp,
2139 	    checkonly, cr);
2140 	return (error);
2141 }
2142 
2143 /*
2144  * This routine sets socket options.
2145  */
2146 int
2147 icmp_opt_set(conn_t *connp, uint_t optset_context, int level, int name,
2148     uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
2149     void *thisdg_attrs, cred_t *cr)
2150 {
2151 	icmp_t		*icmp = connp->conn_icmp;
2152 	int		err;
2153 	conn_opt_arg_t	coas, *coa;
2154 	boolean_t	checkonly;
2155 	icmp_stack_t	*is = icmp->icmp_is;
2156 
2157 	switch (optset_context) {
2158 	case SETFN_OPTCOM_CHECKONLY:
2159 		checkonly = B_TRUE;
2160 		/*
2161 		 * Note: Implies T_CHECK semantics for T_OPTCOM_REQ
2162 		 * inlen != 0 implies value supplied and
2163 		 * 	we have to "pretend" to set it.
2164 		 * inlen == 0 implies that there is no
2165 		 * 	value part in T_CHECK request and just validation
2166 		 * done elsewhere should be enough, we just return here.
2167 		 */
2168 		if (inlen == 0) {
2169 			*outlenp = 0;
2170 			return (0);
2171 		}
2172 		break;
2173 	case SETFN_OPTCOM_NEGOTIATE:
2174 		checkonly = B_FALSE;
2175 		break;
2176 	case SETFN_UD_NEGOTIATE:
2177 	case SETFN_CONN_NEGOTIATE:
2178 		checkonly = B_FALSE;
2179 		/*
2180 		 * Negotiating local and "association-related" options
2181 		 * through T_UNITDATA_REQ.
2182 		 *
2183 		 * Following routine can filter out ones we do not
2184 		 * want to be "set" this way.
2185 		 */
2186 		if (!icmp_opt_allow_udr_set(level, name)) {
2187 			*outlenp = 0;
2188 			return (EINVAL);
2189 		}
2190 		break;
2191 	default:
2192 		/*
2193 		 * We should never get here
2194 		 */
2195 		*outlenp = 0;
2196 		return (EINVAL);
2197 	}
2198 
2199 	ASSERT((optset_context != SETFN_OPTCOM_CHECKONLY) ||
2200 	    (optset_context == SETFN_OPTCOM_CHECKONLY && inlen != 0));
2201 
2202 	if (thisdg_attrs != NULL) {
2203 		/* Options from T_UNITDATA_REQ */
2204 		coa = (conn_opt_arg_t *)thisdg_attrs;
2205 		ASSERT(coa->coa_connp == connp);
2206 		ASSERT(coa->coa_ixa != NULL);
2207 		ASSERT(coa->coa_ipp != NULL);
2208 		ASSERT(coa->coa_ancillary);
2209 	} else {
2210 		coa = &coas;
2211 		coas.coa_connp = connp;
2212 		/* Get a reference on conn_ixa to prevent concurrent mods */
2213 		coas.coa_ixa = conn_get_ixa(connp, B_TRUE);
2214 		if (coas.coa_ixa == NULL) {
2215 			*outlenp = 0;
2216 			return (ENOMEM);
2217 		}
2218 		coas.coa_ipp = &connp->conn_xmit_ipp;
2219 		coas.coa_ancillary = B_FALSE;
2220 		coas.coa_changed = 0;
2221 	}
2222 
2223 	err = icmp_do_opt_set(coa, level, name, inlen, invalp,
2224 	    cr, checkonly);
2225 	if (err != 0) {
2226 errout:
2227 		if (!coa->coa_ancillary)
2228 			ixa_refrele(coa->coa_ixa);
2229 		*outlenp = 0;
2230 		return (err);
2231 	}
2232 
2233 	/*
2234 	 * Common case of OK return with outval same as inval.
2235 	 */
2236 	if (invalp != outvalp) {
2237 		/* don't trust bcopy for identical src/dst */
2238 		(void) bcopy(invalp, outvalp, inlen);
2239 	}
2240 	*outlenp = inlen;
2241 
2242 	/*
2243 	 * If this was not ancillary data, then we rebuild the headers,
2244 	 * update the IRE/NCE, and IPsec as needed.
2245 	 * Since the label depends on the destination we go through
2246 	 * ip_set_destination first.
2247 	 */
2248 	if (coa->coa_ancillary) {
2249 		return (0);
2250 	}
2251 
2252 	if (coa->coa_changed & COA_ROUTE_CHANGED) {
2253 		in6_addr_t saddr, faddr, nexthop;
2254 		in_port_t fport;
2255 
2256 		/*
2257 		 * We clear lastdst to make sure we pick up the change
2258 		 * next time sending.
2259 		 * If we are connected we re-cache the information.
2260 		 * We ignore errors to preserve BSD behavior.
2261 		 * Note that we don't redo IPsec policy lookup here
2262 		 * since the final destination (or source) didn't change.
2263 		 */
2264 		mutex_enter(&connp->conn_lock);
2265 		connp->conn_v6lastdst = ipv6_all_zeros;
2266 
2267 		ip_attr_nexthop(coa->coa_ipp, coa->coa_ixa,
2268 		    &connp->conn_faddr_v6, &nexthop);
2269 		saddr = connp->conn_saddr_v6;
2270 		faddr = connp->conn_faddr_v6;
2271 		fport = connp->conn_fport;
2272 		mutex_exit(&connp->conn_lock);
2273 
2274 		if (!IN6_IS_ADDR_UNSPECIFIED(&faddr) &&
2275 		    !IN6_IS_ADDR_V4MAPPED_ANY(&faddr)) {
2276 			(void) ip_attr_connect(connp, coa->coa_ixa,
2277 			    &saddr, &faddr, &nexthop, fport, NULL, NULL,
2278 			    IPDF_ALLOW_MCBC | IPDF_VERIFY_DST);
2279 		}
2280 	}
2281 
2282 	ixa_refrele(coa->coa_ixa);
2283 
2284 	if (coa->coa_changed & COA_HEADER_CHANGED) {
2285 		/*
2286 		 * Rebuild the header template if we are connected.
2287 		 * Otherwise clear conn_v6lastdst so we rebuild the header
2288 		 * in the data path.
2289 		 */
2290 		mutex_enter(&connp->conn_lock);
2291 		if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) &&
2292 		    !IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_faddr_v6)) {
2293 			err = icmp_build_hdr_template(connp,
2294 			    &connp->conn_saddr_v6, &connp->conn_faddr_v6,
2295 			    connp->conn_flowinfo);
2296 			if (err != 0) {
2297 				mutex_exit(&connp->conn_lock);
2298 				return (err);
2299 			}
2300 		} else {
2301 			connp->conn_v6lastdst = ipv6_all_zeros;
2302 		}
2303 		mutex_exit(&connp->conn_lock);
2304 	}
2305 	if (coa->coa_changed & COA_RCVBUF_CHANGED) {
2306 		(void) proto_set_rx_hiwat(connp->conn_rq, connp,
2307 		    connp->conn_rcvbuf);
2308 	}
2309 	if ((coa->coa_changed & COA_SNDBUF_CHANGED) && !IPCL_IS_NONSTR(connp)) {
2310 		connp->conn_wq->q_hiwat = connp->conn_sndbuf;
2311 	}
2312 	if (coa->coa_changed & COA_WROFF_CHANGED) {
2313 		/* Increase wroff if needed */
2314 		uint_t wroff;
2315 
2316 		mutex_enter(&connp->conn_lock);
2317 		wroff = connp->conn_ht_iphc_allocated + is->is_wroff_extra;
2318 		if (wroff > connp->conn_wroff) {
2319 			connp->conn_wroff = wroff;
2320 			mutex_exit(&connp->conn_lock);
2321 			(void) proto_set_tx_wroff(connp->conn_rq, connp, wroff);
2322 		} else {
2323 			mutex_exit(&connp->conn_lock);
2324 		}
2325 	}
2326 	if (coa->coa_changed & COA_ICMP_BIND_NEEDED) {
2327 		icmp_bind_proto(icmp);
2328 	}
2329 	return (err);
2330 }
2331 
2332 /* This routine sets socket options. */
2333 int
2334 icmp_tpi_opt_set(queue_t *q, uint_t optset_context, int level, int name,
2335     uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
2336     void *thisdg_attrs, cred_t *cr)
2337 {
2338 	conn_t	*connp = Q_TO_CONN(q);
2339 	int error;
2340 
2341 	error = icmp_opt_set(connp, optset_context, level, name, inlen, invalp,
2342 	    outlenp, outvalp, thisdg_attrs, cr);
2343 	return (error);
2344 }
2345 
2346 /*
2347  * Setup IP headers.
2348  *
2349  * Note that IP_HDRINCL has ipha_protocol that is different than conn_proto,
2350  * but icmp_output_hdrincl restores ipha_protocol once we return.
2351  */
2352 mblk_t *
2353 icmp_prepend_hdr(conn_t *connp, ip_xmit_attr_t *ixa, const ip_pkt_t *ipp,
2354     const in6_addr_t *v6src, const in6_addr_t *v6dst, uint32_t flowinfo,
2355     mblk_t *data_mp, int *errorp)
2356 {
2357 	mblk_t		*mp;
2358 	icmp_stack_t	*is = connp->conn_netstack->netstack_icmp;
2359 	uint_t		data_len;
2360 	uint32_t	cksum;
2361 
2362 	data_len = msgdsize(data_mp);
2363 	mp = conn_prepend_hdr(ixa, ipp, v6src, v6dst, connp->conn_proto,
2364 	    flowinfo, 0, data_mp, data_len, is->is_wroff_extra, &cksum, errorp);
2365 	if (mp == NULL) {
2366 		ASSERT(*errorp != 0);
2367 		return (NULL);
2368 	}
2369 
2370 	ixa->ixa_pktlen = data_len + ixa->ixa_ip_hdr_length;
2371 
2372 	/*
2373 	 * If there was a routing option/header then conn_prepend_hdr
2374 	 * has massaged it and placed the pseudo-header checksum difference
2375 	 * in the cksum argument.
2376 	 *
2377 	 * Prepare for ICMPv6 checksum done in IP.
2378 	 *
2379 	 * We make it easy for IP to include our pseudo header
2380 	 * by putting our length (and any routing header adjustment)
2381 	 * in the ICMPv6 checksum field.
2382 	 * The IP source, destination, and length have already been set by
2383 	 * conn_prepend_hdr.
2384 	 */
2385 	cksum += data_len;
2386 	cksum = (cksum >> 16) + (cksum & 0xFFFF);
2387 	ASSERT(cksum < 0x10000);
2388 
2389 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
2390 		ipha_t	*ipha = (ipha_t *)mp->b_rptr;
2391 
2392 		ASSERT(ntohs(ipha->ipha_length) == ixa->ixa_pktlen);
2393 	} else {
2394 		ip6_t	*ip6h = (ip6_t *)mp->b_rptr;
2395 		uint_t	cksum_offset = 0;
2396 
2397 		ASSERT(ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN == ixa->ixa_pktlen);
2398 
2399 		if (ixa->ixa_flags & IXAF_SET_ULP_CKSUM) {
2400 			if (connp->conn_proto == IPPROTO_ICMPV6) {
2401 				cksum_offset = ixa->ixa_ip_hdr_length +
2402 				    offsetof(icmp6_t, icmp6_cksum);
2403 			} else if (ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
2404 				cksum_offset = ixa->ixa_ip_hdr_length +
2405 				    ixa->ixa_raw_cksum_offset;
2406 			}
2407 		}
2408 		if (cksum_offset != 0) {
2409 			uint16_t *ptr;
2410 
2411 			/* Make sure the checksum fits in the first mblk */
2412 			if (cksum_offset + sizeof (short) > MBLKL(mp)) {
2413 				mblk_t *mp1;
2414 
2415 				mp1 = msgpullup(mp,
2416 				    cksum_offset + sizeof (short));
2417 				freemsg(mp);
2418 				if (mp1 == NULL) {
2419 					*errorp = ENOMEM;
2420 					return (NULL);
2421 				}
2422 				mp = mp1;
2423 				ip6h = (ip6_t *)mp->b_rptr;
2424 			}
2425 			ptr = (uint16_t *)(mp->b_rptr + cksum_offset);
2426 			*ptr = htons(cksum);
2427 		}
2428 	}
2429 
2430 	/* Note that we don't try to update wroff due to ancillary data */
2431 	return (mp);
2432 }
2433 
2434 static int
2435 icmp_build_hdr_template(conn_t *connp, const in6_addr_t *v6src,
2436     const in6_addr_t *v6dst, uint32_t flowinfo)
2437 {
2438 	int		error;
2439 
2440 	ASSERT(MUTEX_HELD(&connp->conn_lock));
2441 	/*
2442 	 * We clear lastdst to make sure we don't use the lastdst path
2443 	 * next time sending since we might not have set v6dst yet.
2444 	 */
2445 	connp->conn_v6lastdst = ipv6_all_zeros;
2446 
2447 	error = conn_build_hdr_template(connp, 0, 0, v6src, v6dst, flowinfo);
2448 	if (error != 0)
2449 		return (error);
2450 
2451 	/*
2452 	 * Any routing header/option has been massaged. The checksum difference
2453 	 * is stored in conn_sum.
2454 	 */
2455 	return (0);
2456 }
2457 
2458 static mblk_t *
2459 icmp_queue_fallback(icmp_t *icmp, mblk_t *mp)
2460 {
2461 	ASSERT(MUTEX_HELD(&icmp->icmp_recv_lock));
2462 	if (IPCL_IS_NONSTR(icmp->icmp_connp)) {
2463 		/*
2464 		 * fallback has started but messages have not been moved yet
2465 		 */
2466 		if (icmp->icmp_fallback_queue_head == NULL) {
2467 			ASSERT(icmp->icmp_fallback_queue_tail == NULL);
2468 			icmp->icmp_fallback_queue_head = mp;
2469 			icmp->icmp_fallback_queue_tail = mp;
2470 		} else {
2471 			ASSERT(icmp->icmp_fallback_queue_tail != NULL);
2472 			icmp->icmp_fallback_queue_tail->b_next = mp;
2473 			icmp->icmp_fallback_queue_tail = mp;
2474 		}
2475 		return (NULL);
2476 	} else {
2477 		/*
2478 		 * Fallback completed, let the caller putnext() the mblk.
2479 		 */
2480 		return (mp);
2481 	}
2482 }
2483 
2484 /*
2485  * Deliver data to ULP. In case we have a socket, and it's falling back to
2486  * TPI, then we'll queue the mp for later processing.
2487  */
2488 static void
2489 icmp_ulp_recv(conn_t *connp, mblk_t *mp, uint_t len)
2490 {
2491 	if (IPCL_IS_NONSTR(connp)) {
2492 		icmp_t *icmp = connp->conn_icmp;
2493 		int error;
2494 
2495 		ASSERT(len == msgdsize(mp));
2496 		if ((*connp->conn_upcalls->su_recv)
2497 		    (connp->conn_upper_handle, mp, len, 0, &error, NULL) < 0) {
2498 			mutex_enter(&icmp->icmp_recv_lock);
2499 			if (error == ENOSPC) {
2500 				/*
2501 				 * let's confirm while holding the lock
2502 				 */
2503 				if ((*connp->conn_upcalls->su_recv)
2504 				    (connp->conn_upper_handle, NULL, 0, 0,
2505 				    &error, NULL) < 0) {
2506 					ASSERT(error == ENOSPC);
2507 					if (error == ENOSPC) {
2508 						connp->conn_flow_cntrld =
2509 						    B_TRUE;
2510 					}
2511 				}
2512 				mutex_exit(&icmp->icmp_recv_lock);
2513 			} else {
2514 				ASSERT(error == EOPNOTSUPP);
2515 				mp = icmp_queue_fallback(icmp, mp);
2516 				mutex_exit(&icmp->icmp_recv_lock);
2517 				if (mp != NULL)
2518 					putnext(connp->conn_rq, mp);
2519 			}
2520 		}
2521 		ASSERT(MUTEX_NOT_HELD(&icmp->icmp_recv_lock));
2522 	} else {
2523 		putnext(connp->conn_rq, mp);
2524 	}
2525 }
2526 
2527 /*
2528  * This is the inbound data path.
2529  * IP has already pulled up the IP headers and verified alignment
2530  * etc.
2531  */
2532 /* ARGSUSED2 */
2533 static void
2534 icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
2535 {
2536 	conn_t			*connp = (conn_t *)arg1;
2537 	struct T_unitdata_ind	*tudi;
2538 	uchar_t			*rptr;		/* Pointer to IP header */
2539 	int			ip_hdr_length;
2540 	int			udi_size;	/* Size of T_unitdata_ind */
2541 	int			pkt_len;
2542 	icmp_t			*icmp;
2543 	ip_pkt_t		ipps;
2544 	ip6_t			*ip6h;
2545 	mblk_t			*mp1;
2546 	crb_t			recv_ancillary;
2547 	icmp_stack_t		*is;
2548 	sin_t			*sin;
2549 	sin6_t			*sin6;
2550 	ipha_t			*ipha;
2551 
2552 	ASSERT(connp->conn_flags & IPCL_RAWIPCONN);
2553 
2554 	icmp = connp->conn_icmp;
2555 	is = icmp->icmp_is;
2556 	rptr = mp->b_rptr;
2557 
2558 	ASSERT(DB_TYPE(mp) == M_DATA);
2559 	ASSERT(OK_32PTR(rptr));
2560 	ASSERT(ira->ira_pktlen == msgdsize(mp));
2561 	pkt_len = ira->ira_pktlen;
2562 
2563 	/*
2564 	 * Get a snapshot of these and allow other threads to change
2565 	 * them after that. We need the same recv_ancillary when determining
2566 	 * the size as when adding the ancillary data items.
2567 	 */
2568 	mutex_enter(&connp->conn_lock);
2569 	recv_ancillary = connp->conn_recv_ancillary;
2570 	mutex_exit(&connp->conn_lock);
2571 
2572 	ip_hdr_length = ira->ira_ip_hdr_length;
2573 	ASSERT(MBLKL(mp) >= ip_hdr_length);	/* IP did a pullup */
2574 
2575 	/* Initialize regardless of IP version */
2576 	ipps.ipp_fields = 0;
2577 
2578 	if (ira->ira_flags & IRAF_IS_IPV4) {
2579 		ASSERT(IPH_HDR_VERSION(rptr) == IPV4_VERSION);
2580 		ASSERT(MBLKL(mp) >= sizeof (ipha_t));
2581 		ASSERT(ira->ira_ip_hdr_length == IPH_HDR_LENGTH(rptr));
2582 
2583 		ipha = (ipha_t *)mp->b_rptr;
2584 		if (recv_ancillary.crb_all != 0)
2585 			(void) ip_find_hdr_v4(ipha, &ipps, B_FALSE);
2586 
2587 		/*
2588 		 * BSD for some reason adjusts ipha_length to exclude the
2589 		 * IP header length. We do the same.
2590 		 */
2591 		if (is->is_bsd_compat) {
2592 			ushort_t len;
2593 
2594 			len = ntohs(ipha->ipha_length);
2595 			if (mp->b_datap->db_ref > 1) {
2596 				/*
2597 				 * Allocate a new IP header so that we can
2598 				 * modify ipha_length.
2599 				 */
2600 				mblk_t	*mp1;
2601 
2602 				mp1 = allocb(ip_hdr_length, BPRI_MED);
2603 				if (mp1 == NULL) {
2604 					freemsg(mp);
2605 					BUMP_MIB(&is->is_rawip_mib,
2606 					    rawipInErrors);
2607 					return;
2608 				}
2609 				bcopy(rptr, mp1->b_rptr, ip_hdr_length);
2610 				mp->b_rptr = rptr + ip_hdr_length;
2611 				rptr = mp1->b_rptr;
2612 				ipha = (ipha_t *)rptr;
2613 				mp1->b_cont = mp;
2614 				mp1->b_wptr = rptr + ip_hdr_length;
2615 				mp = mp1;
2616 			}
2617 			len -= ip_hdr_length;
2618 			ipha->ipha_length = htons(len);
2619 		}
2620 
2621 		/*
2622 		 * For RAW sockets we not pass ICMP/IPv4 packets to AF_INET6
2623 		 * sockets. This is ensured by icmp_bind and the IP fanout code.
2624 		 */
2625 		ASSERT(connp->conn_family == AF_INET);
2626 
2627 		/*
2628 		 * This is the inbound data path.  Packets are passed upstream
2629 		 * as T_UNITDATA_IND messages with full IPv4 headers still
2630 		 * attached.
2631 		 */
2632 
2633 		/*
2634 		 * Normally only send up the source address.
2635 		 * If any ancillary data items are wanted we add those.
2636 		 */
2637 		udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin_t);
2638 		if (recv_ancillary.crb_all != 0) {
2639 			udi_size += conn_recvancillary_size(connp,
2640 			    recv_ancillary, ira, mp, &ipps);
2641 		}
2642 
2643 		/* Allocate a message block for the T_UNITDATA_IND structure. */
2644 		mp1 = allocb(udi_size, BPRI_MED);
2645 		if (mp1 == NULL) {
2646 			freemsg(mp);
2647 			BUMP_MIB(&is->is_rawip_mib, rawipInErrors);
2648 			return;
2649 		}
2650 		mp1->b_cont = mp;
2651 		tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2652 		mp1->b_datap->db_type = M_PROTO;
2653 		mp1->b_wptr = (uchar_t *)tudi + udi_size;
2654 		tudi->PRIM_type = T_UNITDATA_IND;
2655 		tudi->SRC_length = sizeof (sin_t);
2656 		tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2657 		sin = (sin_t *)&tudi[1];
2658 		*sin = sin_null;
2659 		sin->sin_family = AF_INET;
2660 		sin->sin_addr.s_addr = ipha->ipha_src;
2661 		*(uint32_t *)&sin->sin_zero[0] = 0;
2662 		*(uint32_t *)&sin->sin_zero[4] = 0;
2663 		tudi->OPT_offset =  sizeof (struct T_unitdata_ind) +
2664 		    sizeof (sin_t);
2665 		udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin_t));
2666 		tudi->OPT_length = udi_size;
2667 
2668 		/*
2669 		 * Add options if IP_RECVIF etc is set
2670 		 */
2671 		if (udi_size != 0) {
2672 			conn_recvancillary_add(connp, recv_ancillary, ira,
2673 			    &ipps, (uchar_t *)&sin[1], udi_size);
2674 		}
2675 		goto deliver;
2676 	}
2677 
2678 	ASSERT(IPH_HDR_VERSION(rptr) == IPV6_VERSION);
2679 	/*
2680 	 * IPv6 packets can only be received by applications
2681 	 * that are prepared to receive IPv6 addresses.
2682 	 * The IP fanout must ensure this.
2683 	 */
2684 	ASSERT(connp->conn_family == AF_INET6);
2685 
2686 	/*
2687 	 * Handle IPv6 packets. We don't pass up the IP headers with the
2688 	 * payload for IPv6.
2689 	 */
2690 
2691 	ip6h = (ip6_t *)rptr;
2692 	if (recv_ancillary.crb_all != 0) {
2693 		/*
2694 		 * Call on ip_find_hdr_v6 which gets individual lenghts of
2695 		 * extension headers (and pointers to them).
2696 		 */
2697 		uint8_t		nexthdr;
2698 
2699 		/* We don't care about the length or nextheader. */
2700 		(void) ip_find_hdr_v6(mp, ip6h, B_TRUE, &ipps, &nexthdr);
2701 
2702 		/*
2703 		 * We do not pass up hop-by-hop options or any other
2704 		 * extension header as part of the packet. Applications
2705 		 * that want to see them have to specify IPV6_RECV* socket
2706 		 * options. And conn_recvancillary_size/add explicitly
2707 		 * drops the TX option from IPV6_HOPOPTS as it does for UDP.
2708 		 *
2709 		 * If we had multilevel ICMP sockets, then we'd want to
2710 		 * modify conn_recvancillary_size/add to
2711 		 * allow the user to see the label.
2712 		 */
2713 	}
2714 
2715 	/*
2716 	 * Check a filter for ICMPv6 types if needed.
2717 	 * Verify raw checksums if needed.
2718 	 */
2719 	mutex_enter(&connp->conn_lock);
2720 	if (icmp->icmp_filter != NULL) {
2721 		int type;
2722 
2723 		/* Assumes that IP has done the pullupmsg */
2724 		type = mp->b_rptr[ip_hdr_length];
2725 
2726 		ASSERT(mp->b_rptr + ip_hdr_length <= mp->b_wptr);
2727 		if (ICMP6_FILTER_WILLBLOCK(type, icmp->icmp_filter)) {
2728 			mutex_exit(&connp->conn_lock);
2729 			freemsg(mp);
2730 			return;
2731 		}
2732 	}
2733 	if (connp->conn_ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
2734 		/* Checksum */
2735 		uint16_t	*up;
2736 		uint32_t	sum;
2737 		int		remlen;
2738 
2739 		up = (uint16_t *)&ip6h->ip6_src;
2740 
2741 		remlen = msgdsize(mp) - ip_hdr_length;
2742 		sum = htons(connp->conn_proto + remlen)
2743 		    + up[0] + up[1] + up[2] + up[3]
2744 		    + up[4] + up[5] + up[6] + up[7]
2745 		    + up[8] + up[9] + up[10] + up[11]
2746 		    + up[12] + up[13] + up[14] + up[15];
2747 		sum = (sum & 0xffff) + (sum >> 16);
2748 		sum = IP_CSUM(mp, ip_hdr_length, sum);
2749 		if (sum != 0) {
2750 			/* IPv6 RAW checksum failed */
2751 			ip0dbg(("icmp_rput: RAW checksum failed %x\n", sum));
2752 			mutex_exit(&connp->conn_lock);
2753 			freemsg(mp);
2754 			BUMP_MIB(&is->is_rawip_mib, rawipInCksumErrs);
2755 			return;
2756 		}
2757 	}
2758 	mutex_exit(&connp->conn_lock);
2759 
2760 	udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t);
2761 
2762 	if (recv_ancillary.crb_all != 0) {
2763 		udi_size += conn_recvancillary_size(connp,
2764 		    recv_ancillary, ira, mp, &ipps);
2765 	}
2766 
2767 	mp1 = allocb(udi_size, BPRI_MED);
2768 	if (mp1 == NULL) {
2769 		freemsg(mp);
2770 		BUMP_MIB(&is->is_rawip_mib, rawipInErrors);
2771 		return;
2772 	}
2773 	mp1->b_cont = mp;
2774 	mp1->b_datap->db_type = M_PROTO;
2775 	tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2776 	mp1->b_wptr = (uchar_t *)tudi + udi_size;
2777 	tudi->PRIM_type = T_UNITDATA_IND;
2778 	tudi->SRC_length = sizeof (sin6_t);
2779 	tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2780 	tudi->OPT_offset = sizeof (struct T_unitdata_ind) + sizeof (sin6_t);
2781 	udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin6_t));
2782 	tudi->OPT_length = udi_size;
2783 	sin6 = (sin6_t *)&tudi[1];
2784 	*sin6 = sin6_null;
2785 	sin6->sin6_port = 0;
2786 	sin6->sin6_family = AF_INET6;
2787 
2788 	sin6->sin6_addr = ip6h->ip6_src;
2789 	/* No sin6_flowinfo per API */
2790 	sin6->sin6_flowinfo = 0;
2791 	/* For link-scope pass up scope id */
2792 	if (IN6_IS_ADDR_LINKSCOPE(&ip6h->ip6_src))
2793 		sin6->sin6_scope_id = ira->ira_ruifindex;
2794 	else
2795 		sin6->sin6_scope_id = 0;
2796 	sin6->__sin6_src_id = ip_srcid_find_addr(&ip6h->ip6_dst,
2797 	    IPCL_ZONEID(connp), is->is_netstack);
2798 
2799 	if (udi_size != 0) {
2800 		conn_recvancillary_add(connp, recv_ancillary, ira,
2801 		    &ipps, (uchar_t *)&sin6[1], udi_size);
2802 	}
2803 
2804 	/* Skip all the IPv6 headers per API */
2805 	mp->b_rptr += ip_hdr_length;
2806 	pkt_len -= ip_hdr_length;
2807 
2808 deliver:
2809 	BUMP_MIB(&is->is_rawip_mib, rawipInDatagrams);
2810 	icmp_ulp_recv(connp, mp1, pkt_len);
2811 }
2812 
2813 /*
2814  * return SNMP stuff in buffer in mpdata. We don't hold any lock and report
2815  * information that can be changing beneath us.
2816  */
2817 mblk_t *
2818 icmp_snmp_get(queue_t *q, mblk_t *mpctl)
2819 {
2820 	mblk_t			*mpdata;
2821 	struct opthdr		*optp;
2822 	conn_t			*connp = Q_TO_CONN(q);
2823 	icmp_stack_t		*is = connp->conn_netstack->netstack_icmp;
2824 	mblk_t			*mp2ctl;
2825 
2826 	/*
2827 	 * make a copy of the original message
2828 	 */
2829 	mp2ctl = copymsg(mpctl);
2830 
2831 	if (mpctl == NULL ||
2832 	    (mpdata = mpctl->b_cont) == NULL) {
2833 		freemsg(mpctl);
2834 		freemsg(mp2ctl);
2835 		return (0);
2836 	}
2837 
2838 	/* fixed length structure for IPv4 and IPv6 counters */
2839 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
2840 	optp->level = EXPER_RAWIP;
2841 	optp->name = 0;
2842 	(void) snmp_append_data(mpdata, (char *)&is->is_rawip_mib,
2843 	    sizeof (is->is_rawip_mib));
2844 	optp->len = msgdsize(mpdata);
2845 	qreply(q, mpctl);
2846 
2847 	return (mp2ctl);
2848 }
2849 
2850 /*
2851  * Return 0 if invalid set request, 1 otherwise, including non-rawip requests.
2852  * TODO:  If this ever actually tries to set anything, it needs to be
2853  * to do the appropriate locking.
2854  */
2855 /* ARGSUSED */
2856 int
2857 icmp_snmp_set(queue_t *q, t_scalar_t level, t_scalar_t name,
2858     uchar_t *ptr, int len)
2859 {
2860 	switch (level) {
2861 	case EXPER_RAWIP:
2862 		return (0);
2863 	default:
2864 		return (1);
2865 	}
2866 }
2867 
2868 /*
2869  * This routine creates a T_UDERROR_IND message and passes it upstream.
2870  * The address and options are copied from the T_UNITDATA_REQ message
2871  * passed in mp.  This message is freed.
2872  */
2873 static void
2874 icmp_ud_err(queue_t *q, mblk_t *mp, t_scalar_t err)
2875 {
2876 	struct T_unitdata_req *tudr;
2877 	mblk_t	*mp1;
2878 	uchar_t *destaddr;
2879 	t_scalar_t destlen;
2880 	uchar_t	*optaddr;
2881 	t_scalar_t optlen;
2882 
2883 	if ((mp->b_wptr < mp->b_rptr) ||
2884 	    (MBLKL(mp)) < sizeof (struct T_unitdata_req)) {
2885 		goto done;
2886 	}
2887 	tudr = (struct T_unitdata_req *)mp->b_rptr;
2888 	destaddr = mp->b_rptr + tudr->DEST_offset;
2889 	if (destaddr < mp->b_rptr || destaddr >= mp->b_wptr ||
2890 	    destaddr + tudr->DEST_length < mp->b_rptr ||
2891 	    destaddr + tudr->DEST_length > mp->b_wptr) {
2892 		goto done;
2893 	}
2894 	optaddr = mp->b_rptr + tudr->OPT_offset;
2895 	if (optaddr < mp->b_rptr || optaddr >= mp->b_wptr ||
2896 	    optaddr + tudr->OPT_length < mp->b_rptr ||
2897 	    optaddr + tudr->OPT_length > mp->b_wptr) {
2898 		goto done;
2899 	}
2900 	destlen = tudr->DEST_length;
2901 	optlen = tudr->OPT_length;
2902 
2903 	mp1 = mi_tpi_uderror_ind((char *)destaddr, destlen,
2904 	    (char *)optaddr, optlen, err);
2905 	if (mp1 != NULL)
2906 		qreply(q, mp1);
2907 
2908 done:
2909 	freemsg(mp);
2910 }
2911 
2912 static int
2913 rawip_do_unbind(conn_t *connp)
2914 {
2915 	icmp_t	*icmp = connp->conn_icmp;
2916 
2917 	mutex_enter(&connp->conn_lock);
2918 	/* If a bind has not been done, we can't unbind. */
2919 	if (icmp->icmp_state == TS_UNBND) {
2920 		mutex_exit(&connp->conn_lock);
2921 		return (-TOUTSTATE);
2922 	}
2923 	connp->conn_saddr_v6 = ipv6_all_zeros;
2924 	connp->conn_bound_addr_v6 = ipv6_all_zeros;
2925 	connp->conn_laddr_v6 = ipv6_all_zeros;
2926 	connp->conn_mcbc_bind = B_FALSE;
2927 	connp->conn_lport = 0;
2928 	connp->conn_fport = 0;
2929 	/* In case we were also connected */
2930 	connp->conn_faddr_v6 = ipv6_all_zeros;
2931 	connp->conn_v6lastdst = ipv6_all_zeros;
2932 
2933 	icmp->icmp_state = TS_UNBND;
2934 
2935 	(void) icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
2936 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
2937 	mutex_exit(&connp->conn_lock);
2938 
2939 	ip_unbind(connp);
2940 	return (0);
2941 }
2942 
2943 /*
2944  * This routine is called by icmp_wput to handle T_UNBIND_REQ messages.
2945  * After some error checking, the message is passed downstream to ip.
2946  */
2947 static void
2948 icmp_tpi_unbind(queue_t *q, mblk_t *mp)
2949 {
2950 	conn_t	*connp = Q_TO_CONN(q);
2951 	int	error;
2952 
2953 	ASSERT(mp->b_cont == NULL);
2954 	error = rawip_do_unbind(connp);
2955 	if (error) {
2956 		if (error < 0) {
2957 			icmp_err_ack(q, mp, -error, 0);
2958 		} else {
2959 			icmp_err_ack(q, mp, 0, error);
2960 		}
2961 		return;
2962 	}
2963 
2964 	/*
2965 	 * Convert mp into a T_OK_ACK
2966 	 */
2967 
2968 	mp = mi_tpi_ok_ack_alloc(mp);
2969 
2970 	/*
2971 	 * should not happen in practice... T_OK_ACK is smaller than the
2972 	 * original message.
2973 	 */
2974 	ASSERT(mp != NULL);
2975 	ASSERT(((struct T_ok_ack *)mp->b_rptr)->PRIM_type == T_OK_ACK);
2976 	qreply(q, mp);
2977 }
2978 
2979 /*
2980  * Process IPv4 packets that already include an IP header.
2981  * Used when IP_HDRINCL has been set (implicit for IPPROTO_RAW and
2982  * IPPROTO_IGMP).
2983  * In this case we ignore the address and any options in the T_UNITDATA_REQ.
2984  *
2985  * The packet is assumed to have a base (20 byte) IP header followed
2986  * by the upper-layer protocol. We include any IP_OPTIONS including a
2987  * CIPSO label but otherwise preserve the base IP header.
2988  */
2989 static int
2990 icmp_output_hdrincl(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid)
2991 {
2992 	icmp_t		*icmp = connp->conn_icmp;
2993 	icmp_stack_t	*is = icmp->icmp_is;
2994 	ipha_t		iphas;
2995 	ipha_t		*ipha;
2996 	int		ip_hdr_length;
2997 	int		tp_hdr_len;
2998 	ip_xmit_attr_t	*ixa;
2999 	ip_pkt_t	*ipp;
3000 	in6_addr_t	v6src;
3001 	in6_addr_t	v6dst;
3002 	in6_addr_t	v6nexthop;
3003 	int		error;
3004 	boolean_t	do_ipsec;
3005 
3006 	/*
3007 	 * We need an exclusive copy of conn_ixa since the included IP
3008 	 * header could have any destination.
3009 	 * That copy has no pointers hence we
3010 	 * need to set them up once we've parsed the ancillary data.
3011 	 */
3012 	ixa = conn_get_ixa_exclusive(connp);
3013 	if (ixa == NULL) {
3014 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3015 		freemsg(mp);
3016 		return (ENOMEM);
3017 	}
3018 	ASSERT(cr != NULL);
3019 	/*
3020 	 * Caller has a reference on cr; from db_credp or because we
3021 	 * are running in process context.
3022 	 */
3023 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3024 	ixa->ixa_cred = cr;
3025 	ixa->ixa_cpid = pid;
3026 	if (is_system_labeled()) {
3027 		/* We need to restart with a label based on the cred */
3028 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
3029 	}
3030 
3031 	/* In case previous destination was multicast or multirt */
3032 	ip_attr_newdst(ixa);
3033 
3034 	/* Get a copy of conn_xmit_ipp since the TX label might change it */
3035 	ipp = kmem_zalloc(sizeof (*ipp), KM_NOSLEEP);
3036 	if (ipp == NULL) {
3037 		ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3038 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3039 		ixa->ixa_cpid = connp->conn_cpid;
3040 		ixa_refrele(ixa);
3041 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3042 		freemsg(mp);
3043 		return (ENOMEM);
3044 	}
3045 	mutex_enter(&connp->conn_lock);
3046 	error = ip_pkt_copy(&connp->conn_xmit_ipp, ipp, KM_NOSLEEP);
3047 	mutex_exit(&connp->conn_lock);
3048 	if (error != 0) {
3049 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3050 		freemsg(mp);
3051 		goto done;
3052 	}
3053 
3054 	/* Sanity check length of packet */
3055 	ipha = (ipha_t *)mp->b_rptr;
3056 
3057 	ip_hdr_length = IP_SIMPLE_HDR_LENGTH;
3058 	if ((mp->b_wptr - mp->b_rptr) < IP_SIMPLE_HDR_LENGTH) {
3059 		if (!pullupmsg(mp, IP_SIMPLE_HDR_LENGTH)) {
3060 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3061 			freemsg(mp);
3062 			goto done;
3063 		}
3064 		ipha = (ipha_t *)mp->b_rptr;
3065 	}
3066 	ipha->ipha_version_and_hdr_length =
3067 	    (IP_VERSION<<4) | (ip_hdr_length>>2);
3068 
3069 	/*
3070 	 * We set IXAF_DONTFRAG if the application set DF which makes
3071 	 * IP not fragment.
3072 	 */
3073 	ipha->ipha_fragment_offset_and_flags &= htons(IPH_DF);
3074 	if (ipha->ipha_fragment_offset_and_flags & htons(IPH_DF))
3075 		ixa->ixa_flags |= (IXAF_DONTFRAG | IXAF_PMTU_IPV4_DF);
3076 	else
3077 		ixa->ixa_flags &= ~(IXAF_DONTFRAG | IXAF_PMTU_IPV4_DF);
3078 
3079 	/* Even for multicast and broadcast we honor the apps ttl */
3080 	ixa->ixa_flags |= IXAF_NO_TTL_CHANGE;
3081 
3082 	/*
3083 	 * No source verification for non-local addresses
3084 	 */
3085 	if (ipha->ipha_src != INADDR_ANY &&
3086 	    ip_laddr_verify_v4(ipha->ipha_src, ixa->ixa_zoneid,
3087 	    is->is_netstack->netstack_ip, B_FALSE)
3088 	    != IPVL_UNICAST_UP) {
3089 		ixa->ixa_flags &= ~IXAF_VERIFY_SOURCE;
3090 	}
3091 
3092 	if (ipha->ipha_dst == INADDR_ANY)
3093 		ipha->ipha_dst = htonl(INADDR_LOOPBACK);
3094 
3095 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6src);
3096 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &v6dst);
3097 
3098 	/* Defer IPsec if it might need to look at ICMP type/code */
3099 	do_ipsec = ipha->ipha_protocol != IPPROTO_ICMP;
3100 	ixa->ixa_flags |= IXAF_IS_IPV4;
3101 
3102 	ip_attr_nexthop(ipp, ixa, &v6dst, &v6nexthop);
3103 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop,
3104 	    connp->conn_fport, &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
3105 	    (do_ipsec ? IPDF_IPSEC : 0));
3106 	switch (error) {
3107 	case 0:
3108 		break;
3109 	case EADDRNOTAVAIL:
3110 		/*
3111 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3112 		 * Don't have the application see that errno
3113 		 */
3114 		error = ENETUNREACH;
3115 		goto failed;
3116 	case ENETDOWN:
3117 		/*
3118 		 * Have !ipif_addr_ready address; drop packet silently
3119 		 * until we can get applications to not send until we
3120 		 * are ready.
3121 		 */
3122 		error = 0;
3123 		goto failed;
3124 	case EHOSTUNREACH:
3125 	case ENETUNREACH:
3126 		if (ixa->ixa_ire != NULL) {
3127 			/*
3128 			 * Let conn_ip_output/ire_send_noroute return
3129 			 * the error and send any local ICMP error.
3130 			 */
3131 			error = 0;
3132 			break;
3133 		}
3134 		/* FALLTHRU */
3135 	default:
3136 	failed:
3137 		freemsg(mp);
3138 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3139 		goto done;
3140 	}
3141 	if (ipha->ipha_src == INADDR_ANY)
3142 		IN6_V4MAPPED_TO_IPADDR(&v6src, ipha->ipha_src);
3143 
3144 	/*
3145 	 * We might be going to a different destination than last time,
3146 	 * thus check that TX allows the communication and compute any
3147 	 * needed label.
3148 	 *
3149 	 * TSOL Note: We have an exclusive ipp and ixa for this thread so we
3150 	 * don't have to worry about concurrent threads.
3151 	 */
3152 	if (is_system_labeled()) {
3153 		/*
3154 		 * Check whether Trusted Solaris policy allows communication
3155 		 * with this host, and pretend that the destination is
3156 		 * unreachable if not.
3157 		 * Compute any needed label and place it in ipp_label_v4/v6.
3158 		 *
3159 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
3160 		 * ipp_label_v4/v6 to form the packet.
3161 		 *
3162 		 * Tsol note: We have ipp structure local to this thread so
3163 		 * no locking is needed.
3164 		 */
3165 		error = conn_update_label(connp, ixa, &v6dst, ipp);
3166 		if (error != 0) {
3167 			freemsg(mp);
3168 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3169 			goto done;
3170 		}
3171 	}
3172 
3173 	/*
3174 	 * Save away a copy of the IPv4 header the application passed down
3175 	 * and then prepend an IPv4 header complete with any IP options
3176 	 * including label.
3177 	 * We need a struct copy since icmp_prepend_hdr will reuse the available
3178 	 * space in the mblk.
3179 	 */
3180 	iphas = *ipha;
3181 	mp->b_rptr += IP_SIMPLE_HDR_LENGTH;
3182 
3183 	mp = icmp_prepend_hdr(connp, ixa, ipp, &v6src, &v6dst, 0, mp, &error);
3184 	if (mp == NULL) {
3185 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3186 		ASSERT(error != 0);
3187 		goto done;
3188 	}
3189 	if (ixa->ixa_pktlen > IP_MAXPACKET) {
3190 		error = EMSGSIZE;
3191 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3192 		freemsg(mp);
3193 		goto done;
3194 	}
3195 	/* Restore key parts of the header that the application passed down */
3196 	ipha = (ipha_t *)mp->b_rptr;
3197 	ipha->ipha_type_of_service = iphas.ipha_type_of_service;
3198 	ipha->ipha_ident = iphas.ipha_ident;
3199 	ipha->ipha_fragment_offset_and_flags =
3200 	    iphas.ipha_fragment_offset_and_flags;
3201 	ipha->ipha_ttl = iphas.ipha_ttl;
3202 	ipha->ipha_protocol = iphas.ipha_protocol;
3203 	ipha->ipha_src = iphas.ipha_src;
3204 	ipha->ipha_dst = iphas.ipha_dst;
3205 
3206 	ixa->ixa_protocol = ipha->ipha_protocol;
3207 
3208 	/*
3209 	 * Make sure that the IP header plus any transport header that is
3210 	 * checksumed by ip_output is in the first mblk. (ip_output assumes
3211 	 * that at least the checksum field is in the first mblk.)
3212 	 */
3213 	switch (ipha->ipha_protocol) {
3214 	case IPPROTO_UDP:
3215 		tp_hdr_len = 8;
3216 		break;
3217 	case IPPROTO_TCP:
3218 		tp_hdr_len = 20;
3219 		break;
3220 	default:
3221 		tp_hdr_len = 0;
3222 		break;
3223 	}
3224 	ip_hdr_length = IPH_HDR_LENGTH(ipha);
3225 	if (mp->b_wptr - mp->b_rptr < ip_hdr_length + tp_hdr_len) {
3226 		if (!pullupmsg(mp, ip_hdr_length + tp_hdr_len)) {
3227 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3228 			if (mp->b_cont == NULL)
3229 				error = EINVAL;
3230 			else
3231 				error = ENOMEM;
3232 			freemsg(mp);
3233 			goto done;
3234 		}
3235 	}
3236 
3237 	if (!do_ipsec) {
3238 		/* Policy might differ for different ICMP type/code */
3239 		if (ixa->ixa_ipsec_policy != NULL) {
3240 			IPPOL_REFRELE(ixa->ixa_ipsec_policy);
3241 			ixa->ixa_ipsec_policy = NULL;
3242 			ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
3243 		}
3244 		mp = ip_output_attach_policy(mp, ipha, NULL, connp, ixa);
3245 		if (mp == NULL) {
3246 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3247 			error = EHOSTUNREACH;	/* IPsec policy failure */
3248 			goto done;
3249 		}
3250 	}
3251 
3252 	/* We're done.  Pass the packet to ip. */
3253 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3254 
3255 	error = conn_ip_output(mp, ixa);
3256 	/* No rawipOutErrors if an error since IP increases its error counter */
3257 	switch (error) {
3258 	case 0:
3259 		break;
3260 	case EWOULDBLOCK:
3261 		(void) ixa_check_drain_insert(connp, ixa);
3262 		error = 0;
3263 		break;
3264 	case EADDRNOTAVAIL:
3265 		/*
3266 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3267 		 * Don't have the application see that errno
3268 		 */
3269 		error = ENETUNREACH;
3270 		break;
3271 	}
3272 done:
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 	ip_pkt_free(ipp);
3278 	kmem_free(ipp, sizeof (*ipp));
3279 	return (error);
3280 }
3281 
3282 static mblk_t *
3283 icmp_output_attach_policy(mblk_t *mp, conn_t *connp, ip_xmit_attr_t *ixa)
3284 {
3285 	ipha_t	*ipha = NULL;
3286 	ip6_t	*ip6h = NULL;
3287 
3288 	if (ixa->ixa_flags & IXAF_IS_IPV4)
3289 		ipha = (ipha_t *)mp->b_rptr;
3290 	else
3291 		ip6h = (ip6_t *)mp->b_rptr;
3292 
3293 	if (ixa->ixa_ipsec_policy != NULL) {
3294 		IPPOL_REFRELE(ixa->ixa_ipsec_policy);
3295 		ixa->ixa_ipsec_policy = NULL;
3296 		ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
3297 	}
3298 	return (ip_output_attach_policy(mp, ipha, ip6h, connp, ixa));
3299 }
3300 
3301 /*
3302  * Handle T_UNITDATA_REQ with options. Both IPv4 and IPv6
3303  * Either tudr_mp or msg is set. If tudr_mp we take ancillary data from
3304  * the TPI options, otherwise we take them from msg_control.
3305  * If both sin and sin6 is set it is a connected socket and we use conn_faddr.
3306  * Always consumes mp; never consumes tudr_mp.
3307  */
3308 static int
3309 icmp_output_ancillary(conn_t *connp, sin_t *sin, sin6_t *sin6, mblk_t *mp,
3310     mblk_t *tudr_mp, struct nmsghdr *msg, cred_t *cr, pid_t pid)
3311 {
3312 	icmp_t		*icmp = connp->conn_icmp;
3313 	icmp_stack_t	*is = icmp->icmp_is;
3314 	int		error;
3315 	ip_xmit_attr_t	*ixa;
3316 	ip_pkt_t	*ipp;
3317 	in6_addr_t	v6src;
3318 	in6_addr_t	v6dst;
3319 	in6_addr_t	v6nexthop;
3320 	in_port_t	dstport;
3321 	uint32_t	flowinfo;
3322 	uint_t		srcid;
3323 	int		is_absreq_failure = 0;
3324 	conn_opt_arg_t	coas, *coa;
3325 
3326 	ASSERT(tudr_mp != NULL || msg != NULL);
3327 
3328 	/*
3329 	 * Get ixa before checking state to handle a disconnect race.
3330 	 *
3331 	 * We need an exclusive copy of conn_ixa since the ancillary data
3332 	 * options might modify it. That copy has no pointers hence we
3333 	 * need to set them up once we've parsed the ancillary data.
3334 	 */
3335 	ixa = conn_get_ixa_exclusive(connp);
3336 	if (ixa == NULL) {
3337 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3338 		freemsg(mp);
3339 		return (ENOMEM);
3340 	}
3341 	ASSERT(cr != NULL);
3342 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3343 	ixa->ixa_cred = cr;
3344 	ixa->ixa_cpid = pid;
3345 	if (is_system_labeled()) {
3346 		/* We need to restart with a label based on the cred */
3347 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
3348 	}
3349 
3350 	/* In case previous destination was multicast or multirt */
3351 	ip_attr_newdst(ixa);
3352 
3353 	/* Get a copy of conn_xmit_ipp since the options might change it */
3354 	ipp = kmem_zalloc(sizeof (*ipp), KM_NOSLEEP);
3355 	if (ipp == NULL) {
3356 		ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3357 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3358 		ixa->ixa_cpid = connp->conn_cpid;
3359 		ixa_refrele(ixa);
3360 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3361 		freemsg(mp);
3362 		return (ENOMEM);
3363 	}
3364 	mutex_enter(&connp->conn_lock);
3365 	error = ip_pkt_copy(&connp->conn_xmit_ipp, ipp, KM_NOSLEEP);
3366 	mutex_exit(&connp->conn_lock);
3367 	if (error != 0) {
3368 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3369 		freemsg(mp);
3370 		goto done;
3371 	}
3372 
3373 	/*
3374 	 * Parse the options and update ixa and ipp as a result.
3375 	 */
3376 
3377 	coa = &coas;
3378 	coa->coa_connp = connp;
3379 	coa->coa_ixa = ixa;
3380 	coa->coa_ipp = ipp;
3381 	coa->coa_ancillary = B_TRUE;
3382 	coa->coa_changed = 0;
3383 
3384 	if (msg != NULL) {
3385 		error = process_auxiliary_options(connp, msg->msg_control,
3386 		    msg->msg_controllen, coa, &icmp_opt_obj, icmp_opt_set, cr);
3387 	} else {
3388 		struct T_unitdata_req *tudr;
3389 
3390 		tudr = (struct T_unitdata_req *)tudr_mp->b_rptr;
3391 		ASSERT(tudr->PRIM_type == T_UNITDATA_REQ);
3392 		error = tpi_optcom_buf(connp->conn_wq, tudr_mp,
3393 		    &tudr->OPT_length, tudr->OPT_offset, cr, &icmp_opt_obj,
3394 		    coa, &is_absreq_failure);
3395 	}
3396 	if (error != 0) {
3397 		/*
3398 		 * Note: No special action needed in this
3399 		 * module for "is_absreq_failure"
3400 		 */
3401 		freemsg(mp);
3402 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3403 		goto done;
3404 	}
3405 	ASSERT(is_absreq_failure == 0);
3406 
3407 	mutex_enter(&connp->conn_lock);
3408 	/*
3409 	 * If laddr is unspecified then we look at sin6_src_id.
3410 	 * We will give precedence to a source address set with IPV6_PKTINFO
3411 	 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
3412 	 * want ip_attr_connect to select a source (since it can fail) when
3413 	 * IPV6_PKTINFO is specified.
3414 	 * If this doesn't result in a source address then we get a source
3415 	 * from ip_attr_connect() below.
3416 	 */
3417 	v6src = connp->conn_saddr_v6;
3418 	if (sin != NULL) {
3419 		IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
3420 		dstport = sin->sin_port;
3421 		flowinfo = 0;
3422 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3423 		ixa->ixa_flags |= IXAF_IS_IPV4;
3424 	} else if (sin6 != NULL) {
3425 		v6dst = sin6->sin6_addr;
3426 		dstport = sin6->sin6_port;
3427 		flowinfo = sin6->sin6_flowinfo;
3428 		srcid = sin6->__sin6_src_id;
3429 		if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
3430 			ixa->ixa_scopeid = sin6->sin6_scope_id;
3431 			ixa->ixa_flags |= IXAF_SCOPEID_SET;
3432 		} else {
3433 			ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3434 		}
3435 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
3436 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
3437 			    connp->conn_netstack);
3438 		}
3439 		if (IN6_IS_ADDR_V4MAPPED(&v6dst))
3440 			ixa->ixa_flags |= IXAF_IS_IPV4;
3441 		else
3442 			ixa->ixa_flags &= ~IXAF_IS_IPV4;
3443 	} else {
3444 		/* Connected case */
3445 		v6dst = connp->conn_faddr_v6;
3446 		flowinfo = connp->conn_flowinfo;
3447 	}
3448 	mutex_exit(&connp->conn_lock);
3449 	/* Handle IPV6_PKTINFO setting source address. */
3450 	if (IN6_IS_ADDR_UNSPECIFIED(&v6src) &&
3451 	    (ipp->ipp_fields & IPPF_ADDR)) {
3452 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
3453 			if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3454 				v6src = ipp->ipp_addr;
3455 		} else {
3456 			if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3457 				v6src = ipp->ipp_addr;
3458 		}
3459 	}
3460 	/*
3461 	 * Allow source not assigned to the system
3462 	 * only if it is not a local addresses
3463 	 */
3464 	if (!V6_OR_V4_INADDR_ANY(v6src)) {
3465 		ip_laddr_t laddr_type;
3466 
3467 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
3468 			ipaddr_t v4src;
3469 
3470 			IN6_V4MAPPED_TO_IPADDR(&v6src, v4src);
3471 			laddr_type = ip_laddr_verify_v4(v4src, ixa->ixa_zoneid,
3472 			    is->is_netstack->netstack_ip, B_FALSE);
3473 		} else {
3474 			laddr_type = ip_laddr_verify_v6(&v6src, ixa->ixa_zoneid,
3475 			    is->is_netstack->netstack_ip, B_FALSE, B_FALSE);
3476 		}
3477 		if (laddr_type != IPVL_UNICAST_UP)
3478 			ixa->ixa_flags &= ~IXAF_VERIFY_SOURCE;
3479 	}
3480 
3481 	ip_attr_nexthop(ipp, ixa, &v6dst, &v6nexthop);
3482 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
3483 	    &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST);
3484 
3485 	switch (error) {
3486 	case 0:
3487 		break;
3488 	case EADDRNOTAVAIL:
3489 		/*
3490 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3491 		 * Don't have the application see that errno
3492 		 */
3493 		error = ENETUNREACH;
3494 		goto failed;
3495 	case ENETDOWN:
3496 		/*
3497 		 * Have !ipif_addr_ready address; drop packet silently
3498 		 * until we can get applications to not send until we
3499 		 * are ready.
3500 		 */
3501 		error = 0;
3502 		goto failed;
3503 	case EHOSTUNREACH:
3504 	case ENETUNREACH:
3505 		if (ixa->ixa_ire != NULL) {
3506 			/*
3507 			 * Let conn_ip_output/ire_send_noroute return
3508 			 * the error and send any local ICMP error.
3509 			 */
3510 			error = 0;
3511 			break;
3512 		}
3513 		/* FALLTHRU */
3514 	default:
3515 	failed:
3516 		freemsg(mp);
3517 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3518 		goto done;
3519 	}
3520 
3521 	/*
3522 	 * We might be going to a different destination than last time,
3523 	 * thus check that TX allows the communication and compute any
3524 	 * needed label.
3525 	 *
3526 	 * TSOL Note: We have an exclusive ipp and ixa for this thread so we
3527 	 * don't have to worry about concurrent threads.
3528 	 */
3529 	if (is_system_labeled()) {
3530 		/*
3531 		 * Check whether Trusted Solaris policy allows communication
3532 		 * with this host, and pretend that the destination is
3533 		 * unreachable if not.
3534 		 * Compute any needed label and place it in ipp_label_v4/v6.
3535 		 *
3536 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
3537 		 * ipp_label_v4/v6 to form the packet.
3538 		 *
3539 		 * Tsol note: We have ipp structure local to this thread so
3540 		 * no locking is needed.
3541 		 */
3542 		error = conn_update_label(connp, ixa, &v6dst, ipp);
3543 		if (error != 0) {
3544 			freemsg(mp);
3545 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3546 			goto done;
3547 		}
3548 	}
3549 	mp = icmp_prepend_hdr(connp, ixa, ipp, &v6src, &v6dst, flowinfo, mp,
3550 	    &error);
3551 	if (mp == NULL) {
3552 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3553 		ASSERT(error != 0);
3554 		goto done;
3555 	}
3556 	if (ixa->ixa_pktlen > IP_MAXPACKET) {
3557 		error = EMSGSIZE;
3558 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3559 		freemsg(mp);
3560 		goto done;
3561 	}
3562 
3563 	/* Policy might differ for different ICMP type/code */
3564 	mp = icmp_output_attach_policy(mp, connp, ixa);
3565 	if (mp == NULL) {
3566 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3567 		error = EHOSTUNREACH;	/* IPsec policy failure */
3568 		goto done;
3569 	}
3570 
3571 	/* We're done.  Pass the packet to ip. */
3572 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3573 
3574 	error = conn_ip_output(mp, ixa);
3575 	if (!connp->conn_unspec_src)
3576 		ixa->ixa_flags |= IXAF_VERIFY_SOURCE;
3577 	/* No rawipOutErrors if an error since IP increases its error counter */
3578 	switch (error) {
3579 	case 0:
3580 		break;
3581 	case EWOULDBLOCK:
3582 		(void) ixa_check_drain_insert(connp, ixa);
3583 		error = 0;
3584 		break;
3585 	case EADDRNOTAVAIL:
3586 		/*
3587 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3588 		 * Don't have the application see that errno
3589 		 */
3590 		error = ENETUNREACH;
3591 		/* FALLTHRU */
3592 	default:
3593 		mutex_enter(&connp->conn_lock);
3594 		/*
3595 		 * Clear the source and v6lastdst so we call ip_attr_connect
3596 		 * for the next packet and try to pick a better source.
3597 		 */
3598 		if (connp->conn_mcbc_bind)
3599 			connp->conn_saddr_v6 = ipv6_all_zeros;
3600 		else
3601 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
3602 		connp->conn_v6lastdst = ipv6_all_zeros;
3603 		mutex_exit(&connp->conn_lock);
3604 		break;
3605 	}
3606 done:
3607 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3608 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3609 	ixa->ixa_cpid = connp->conn_cpid;
3610 	ixa_refrele(ixa);
3611 	ip_pkt_free(ipp);
3612 	kmem_free(ipp, sizeof (*ipp));
3613 	return (error);
3614 }
3615 
3616 /*
3617  * Handle sending an M_DATA for a connected socket.
3618  * Handles both IPv4 and IPv6.
3619  */
3620 int
3621 icmp_output_connected(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid)
3622 {
3623 	icmp_t		*icmp = connp->conn_icmp;
3624 	icmp_stack_t	*is = icmp->icmp_is;
3625 	int		error;
3626 	ip_xmit_attr_t	*ixa;
3627 	boolean_t	do_ipsec;
3628 
3629 	/*
3630 	 * If no other thread is using conn_ixa this just gets a reference to
3631 	 * conn_ixa. Otherwise we get a safe copy of conn_ixa.
3632 	 */
3633 	ixa = conn_get_ixa(connp, B_FALSE);
3634 	if (ixa == NULL) {
3635 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3636 		freemsg(mp);
3637 		return (ENOMEM);
3638 	}
3639 
3640 	ASSERT(cr != NULL);
3641 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3642 	ixa->ixa_cred = cr;
3643 	ixa->ixa_cpid = pid;
3644 
3645 	/* Defer IPsec if it might need to look at ICMP type/code */
3646 	switch (ixa->ixa_protocol) {
3647 	case IPPROTO_ICMP:
3648 	case IPPROTO_ICMPV6:
3649 		do_ipsec = B_FALSE;
3650 		break;
3651 	default:
3652 		do_ipsec = B_TRUE;
3653 	}
3654 
3655 	mutex_enter(&connp->conn_lock);
3656 	mp = icmp_prepend_header_template(connp, ixa, mp,
3657 	    &connp->conn_saddr_v6, connp->conn_flowinfo, &error);
3658 
3659 	if (mp == NULL) {
3660 		ASSERT(error != 0);
3661 		mutex_exit(&connp->conn_lock);
3662 		ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3663 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3664 		ixa->ixa_cpid = connp->conn_cpid;
3665 		ixa_refrele(ixa);
3666 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3667 		freemsg(mp);
3668 		return (error);
3669 	}
3670 
3671 	if (!do_ipsec) {
3672 		/* Policy might differ for different ICMP type/code */
3673 		mp = icmp_output_attach_policy(mp, connp, ixa);
3674 		if (mp == NULL) {
3675 			mutex_exit(&connp->conn_lock);
3676 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3677 			ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3678 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3679 			ixa->ixa_cpid = connp->conn_cpid;
3680 			ixa_refrele(ixa);
3681 			return (EHOSTUNREACH);	/* IPsec policy failure */
3682 		}
3683 	}
3684 
3685 	/*
3686 	 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
3687 	 * safe copy, then we need to fill in any pointers in it.
3688 	 */
3689 	if (ixa->ixa_ire == NULL) {
3690 		in6_addr_t	faddr, saddr;
3691 		in6_addr_t	nexthop;
3692 		in_port_t	fport;
3693 
3694 		saddr = connp->conn_saddr_v6;
3695 		faddr = connp->conn_faddr_v6;
3696 		fport = connp->conn_fport;
3697 		ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &faddr, &nexthop);
3698 		mutex_exit(&connp->conn_lock);
3699 
3700 		error = ip_attr_connect(connp, ixa, &saddr, &faddr, &nexthop,
3701 		    fport, NULL, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
3702 		    (do_ipsec ? IPDF_IPSEC : 0));
3703 		switch (error) {
3704 		case 0:
3705 			break;
3706 		case EADDRNOTAVAIL:
3707 			/*
3708 			 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3709 			 * Don't have the application see that errno
3710 			 */
3711 			error = ENETUNREACH;
3712 			goto failed;
3713 		case ENETDOWN:
3714 			/*
3715 			 * Have !ipif_addr_ready address; drop packet silently
3716 			 * until we can get applications to not send until we
3717 			 * are ready.
3718 			 */
3719 			error = 0;
3720 			goto failed;
3721 		case EHOSTUNREACH:
3722 		case ENETUNREACH:
3723 			if (ixa->ixa_ire != NULL) {
3724 				/*
3725 				 * Let conn_ip_output/ire_send_noroute return
3726 				 * the error and send any local ICMP error.
3727 				 */
3728 				error = 0;
3729 				break;
3730 			}
3731 			/* FALLTHRU */
3732 		default:
3733 		failed:
3734 			ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3735 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3736 			ixa->ixa_cpid = connp->conn_cpid;
3737 			ixa_refrele(ixa);
3738 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3739 			freemsg(mp);
3740 			return (error);
3741 		}
3742 	} else {
3743 		/* Done with conn_t */
3744 		mutex_exit(&connp->conn_lock);
3745 	}
3746 
3747 	/* We're done.  Pass the packet to ip. */
3748 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3749 
3750 	error = conn_ip_output(mp, ixa);
3751 	/* No rawipOutErrors if an error since IP increases its error counter */
3752 	switch (error) {
3753 	case 0:
3754 		break;
3755 	case EWOULDBLOCK:
3756 		(void) ixa_check_drain_insert(connp, ixa);
3757 		error = 0;
3758 		break;
3759 	case EADDRNOTAVAIL:
3760 		/*
3761 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3762 		 * Don't have the application see that errno
3763 		 */
3764 		error = ENETUNREACH;
3765 		break;
3766 	}
3767 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3768 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3769 	ixa->ixa_cpid = connp->conn_cpid;
3770 	ixa_refrele(ixa);
3771 	return (error);
3772 }
3773 
3774 /*
3775  * Handle sending an M_DATA to the last destination.
3776  * Handles both IPv4 and IPv6.
3777  *
3778  * NOTE: The caller must hold conn_lock and we drop it here.
3779  */
3780 int
3781 icmp_output_lastdst(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid,
3782     ip_xmit_attr_t *ixa)
3783 {
3784 	icmp_t		*icmp = connp->conn_icmp;
3785 	icmp_stack_t	*is = icmp->icmp_is;
3786 	int		error;
3787 	boolean_t	do_ipsec;
3788 
3789 	ASSERT(MUTEX_HELD(&connp->conn_lock));
3790 	ASSERT(ixa != NULL);
3791 
3792 	ASSERT(cr != NULL);
3793 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3794 	ixa->ixa_cred = cr;
3795 	ixa->ixa_cpid = pid;
3796 
3797 	/* Defer IPsec if it might need to look at ICMP type/code */
3798 	switch (ixa->ixa_protocol) {
3799 	case IPPROTO_ICMP:
3800 	case IPPROTO_ICMPV6:
3801 		do_ipsec = B_FALSE;
3802 		break;
3803 	default:
3804 		do_ipsec = B_TRUE;
3805 	}
3806 
3807 
3808 	mp = icmp_prepend_header_template(connp, ixa, mp,
3809 	    &connp->conn_v6lastsrc, connp->conn_lastflowinfo, &error);
3810 
3811 	if (mp == NULL) {
3812 		ASSERT(error != 0);
3813 		mutex_exit(&connp->conn_lock);
3814 		ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3815 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3816 		ixa->ixa_cpid = connp->conn_cpid;
3817 		ixa_refrele(ixa);
3818 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3819 		freemsg(mp);
3820 		return (error);
3821 	}
3822 
3823 	if (!do_ipsec) {
3824 		/* Policy might differ for different ICMP type/code */
3825 		mp = icmp_output_attach_policy(mp, connp, ixa);
3826 		if (mp == NULL) {
3827 			mutex_exit(&connp->conn_lock);
3828 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3829 			ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3830 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3831 			ixa->ixa_cpid = connp->conn_cpid;
3832 			ixa_refrele(ixa);
3833 			return (EHOSTUNREACH);	/* IPsec policy failure */
3834 		}
3835 	}
3836 
3837 	/*
3838 	 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
3839 	 * safe copy, then we need to fill in any pointers in it.
3840 	 */
3841 	if (ixa->ixa_ire == NULL) {
3842 		in6_addr_t	lastdst, lastsrc;
3843 		in6_addr_t	nexthop;
3844 		in_port_t	lastport;
3845 
3846 		lastsrc = connp->conn_v6lastsrc;
3847 		lastdst = connp->conn_v6lastdst;
3848 		lastport = connp->conn_lastdstport;
3849 		ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &lastdst, &nexthop);
3850 		mutex_exit(&connp->conn_lock);
3851 
3852 		error = ip_attr_connect(connp, ixa, &lastsrc, &lastdst,
3853 		    &nexthop, lastport, NULL, NULL, IPDF_ALLOW_MCBC |
3854 		    IPDF_VERIFY_DST | (do_ipsec ? IPDF_IPSEC : 0));
3855 		switch (error) {
3856 		case 0:
3857 			break;
3858 		case EADDRNOTAVAIL:
3859 			/*
3860 			 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3861 			 * Don't have the application see that errno
3862 			 */
3863 			error = ENETUNREACH;
3864 			goto failed;
3865 		case ENETDOWN:
3866 			/*
3867 			 * Have !ipif_addr_ready address; drop packet silently
3868 			 * until we can get applications to not send until we
3869 			 * are ready.
3870 			 */
3871 			error = 0;
3872 			goto failed;
3873 		case EHOSTUNREACH:
3874 		case ENETUNREACH:
3875 			if (ixa->ixa_ire != NULL) {
3876 				/*
3877 				 * Let conn_ip_output/ire_send_noroute return
3878 				 * the error and send any local ICMP error.
3879 				 */
3880 				error = 0;
3881 				break;
3882 			}
3883 			/* FALLTHRU */
3884 		default:
3885 		failed:
3886 			ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3887 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3888 			ixa->ixa_cpid = connp->conn_cpid;
3889 			ixa_refrele(ixa);
3890 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3891 			freemsg(mp);
3892 			return (error);
3893 		}
3894 	} else {
3895 		/* Done with conn_t */
3896 		mutex_exit(&connp->conn_lock);
3897 	}
3898 
3899 	/* We're done.  Pass the packet to ip. */
3900 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3901 	error = conn_ip_output(mp, ixa);
3902 	/* No rawipOutErrors if an error since IP increases its error counter */
3903 	switch (error) {
3904 	case 0:
3905 		break;
3906 	case EWOULDBLOCK:
3907 		(void) ixa_check_drain_insert(connp, ixa);
3908 		error = 0;
3909 		break;
3910 	case EADDRNOTAVAIL:
3911 		/*
3912 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3913 		 * Don't have the application see that errno
3914 		 */
3915 		error = ENETUNREACH;
3916 		/* FALLTHRU */
3917 	default:
3918 		mutex_enter(&connp->conn_lock);
3919 		/*
3920 		 * Clear the source and v6lastdst so we call ip_attr_connect
3921 		 * for the next packet and try to pick a better source.
3922 		 */
3923 		if (connp->conn_mcbc_bind)
3924 			connp->conn_saddr_v6 = ipv6_all_zeros;
3925 		else
3926 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
3927 		connp->conn_v6lastdst = ipv6_all_zeros;
3928 		mutex_exit(&connp->conn_lock);
3929 		break;
3930 	}
3931 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
3932 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3933 	ixa->ixa_cpid = connp->conn_cpid;
3934 	ixa_refrele(ixa);
3935 	return (error);
3936 }
3937 
3938 
3939 /*
3940  * Prepend the header template and then fill in the source and
3941  * flowinfo. The caller needs to handle the destination address since
3942  * it's setting is different if rthdr or source route.
3943  *
3944  * Returns NULL is allocation failed or if the packet would exceed IP_MAXPACKET.
3945  * When it returns NULL it sets errorp.
3946  */
3947 static mblk_t *
3948 icmp_prepend_header_template(conn_t *connp, ip_xmit_attr_t *ixa, mblk_t *mp,
3949     const in6_addr_t *v6src, uint32_t flowinfo, int *errorp)
3950 {
3951 	icmp_t		*icmp = connp->conn_icmp;
3952 	icmp_stack_t	*is = icmp->icmp_is;
3953 	uint_t		pktlen;
3954 	uint_t		copylen;
3955 	uint8_t		*iph;
3956 	uint_t		ip_hdr_length;
3957 	uint32_t	cksum;
3958 	ip_pkt_t	*ipp;
3959 
3960 	ASSERT(MUTEX_HELD(&connp->conn_lock));
3961 
3962 	/*
3963 	 * Copy the header template.
3964 	 */
3965 	copylen = connp->conn_ht_iphc_len;
3966 	pktlen = copylen + msgdsize(mp);
3967 	if (pktlen > IP_MAXPACKET) {
3968 		freemsg(mp);
3969 		*errorp = EMSGSIZE;
3970 		return (NULL);
3971 	}
3972 	ixa->ixa_pktlen = pktlen;
3973 
3974 	/* check/fix buffer config, setup pointers into it */
3975 	iph = mp->b_rptr - copylen;
3976 	if (DB_REF(mp) != 1 || iph < DB_BASE(mp) || !OK_32PTR(iph)) {
3977 		mblk_t *mp1;
3978 
3979 		mp1 = allocb(copylen + is->is_wroff_extra, BPRI_MED);
3980 		if (mp1 == NULL) {
3981 			freemsg(mp);
3982 			*errorp = ENOMEM;
3983 			return (NULL);
3984 		}
3985 		mp1->b_wptr = DB_LIM(mp1);
3986 		mp1->b_cont = mp;
3987 		mp = mp1;
3988 		iph = (mp->b_wptr - copylen);
3989 	}
3990 	mp->b_rptr = iph;
3991 	bcopy(connp->conn_ht_iphc, iph, copylen);
3992 	ip_hdr_length = (uint_t)(connp->conn_ht_ulp - connp->conn_ht_iphc);
3993 
3994 	ixa->ixa_ip_hdr_length = ip_hdr_length;
3995 
3996 	/*
3997 	 * Prepare for ICMPv6 checksum done in IP.
3998 	 *
3999 	 * icmp_build_hdr_template has already massaged any routing header
4000 	 * and placed the result in conn_sum.
4001 	 *
4002 	 * We make it easy for IP to include our pseudo header
4003 	 * by putting our length (and any routing header adjustment)
4004 	 * in the ICMPv6 checksum field.
4005 	 */
4006 	cksum = pktlen - ip_hdr_length;
4007 
4008 	cksum += connp->conn_sum;
4009 	cksum = (cksum >> 16) + (cksum & 0xFFFF);
4010 	ASSERT(cksum < 0x10000);
4011 
4012 	ipp = &connp->conn_xmit_ipp;
4013 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
4014 		ipha_t	*ipha = (ipha_t *)iph;
4015 
4016 		ipha->ipha_length = htons((uint16_t)pktlen);
4017 
4018 		/* if IP_PKTINFO specified an addres it wins over bind() */
4019 		if ((ipp->ipp_fields & IPPF_ADDR) &&
4020 		    IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
4021 			ASSERT(ipp->ipp_addr_v4 != INADDR_ANY);
4022 			ipha->ipha_src = ipp->ipp_addr_v4;
4023 		} else {
4024 			IN6_V4MAPPED_TO_IPADDR(v6src, ipha->ipha_src);
4025 		}
4026 	} else {
4027 		ip6_t *ip6h = (ip6_t *)iph;
4028 		uint_t	cksum_offset = 0;
4029 
4030 		ip6h->ip6_plen =  htons((uint16_t)(pktlen - IPV6_HDR_LEN));
4031 
4032 		/* if IP_PKTINFO specified an addres it wins over bind() */
4033 		if ((ipp->ipp_fields & IPPF_ADDR) &&
4034 		    !IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
4035 			ASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ipp->ipp_addr));
4036 			ip6h->ip6_src = ipp->ipp_addr;
4037 		} else {
4038 			ip6h->ip6_src = *v6src;
4039 		}
4040 		ip6h->ip6_vcf =
4041 		    (IPV6_DEFAULT_VERS_AND_FLOW & IPV6_VERS_AND_FLOW_MASK) |
4042 		    (flowinfo & ~IPV6_VERS_AND_FLOW_MASK);
4043 		if (ipp->ipp_fields & IPPF_TCLASS) {
4044 			/* Overrides the class part of flowinfo */
4045 			ip6h->ip6_vcf = IPV6_TCLASS_FLOW(ip6h->ip6_vcf,
4046 			    ipp->ipp_tclass);
4047 		}
4048 
4049 		if (ixa->ixa_flags & IXAF_SET_ULP_CKSUM) {
4050 			if (connp->conn_proto == IPPROTO_ICMPV6) {
4051 				cksum_offset = ixa->ixa_ip_hdr_length +
4052 				    offsetof(icmp6_t, icmp6_cksum);
4053 			} else if (ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
4054 				cksum_offset = ixa->ixa_ip_hdr_length +
4055 				    ixa->ixa_raw_cksum_offset;
4056 			}
4057 		}
4058 		if (cksum_offset != 0) {
4059 			uint16_t *ptr;
4060 
4061 			/* Make sure the checksum fits in the first mblk */
4062 			if (cksum_offset + sizeof (short) > MBLKL(mp)) {
4063 				mblk_t *mp1;
4064 
4065 				mp1 = msgpullup(mp,
4066 				    cksum_offset + sizeof (short));
4067 				freemsg(mp);
4068 				if (mp1 == NULL) {
4069 					*errorp = ENOMEM;
4070 					return (NULL);
4071 				}
4072 				mp = mp1;
4073 				iph = mp->b_rptr;
4074 				ip6h = (ip6_t *)iph;
4075 			}
4076 			ptr = (uint16_t *)(mp->b_rptr + cksum_offset);
4077 			*ptr = htons(cksum);
4078 		}
4079 	}
4080 
4081 	return (mp);
4082 }
4083 
4084 /*
4085  * This routine handles all messages passed downstream.  It either
4086  * consumes the message or passes it downstream; it never queues a
4087  * a message.
4088  */
4089 void
4090 icmp_wput(queue_t *q, mblk_t *mp)
4091 {
4092 	sin6_t		*sin6;
4093 	sin_t		*sin = NULL;
4094 	uint_t		srcid;
4095 	conn_t		*connp = Q_TO_CONN(q);
4096 	icmp_t		*icmp = connp->conn_icmp;
4097 	int		error = 0;
4098 	struct sockaddr	*addr = NULL;
4099 	socklen_t	addrlen;
4100 	icmp_stack_t	*is = icmp->icmp_is;
4101 	struct T_unitdata_req *tudr;
4102 	mblk_t		*data_mp;
4103 	cred_t		*cr;
4104 	pid_t		pid;
4105 
4106 	/*
4107 	 * We directly handle several cases here: T_UNITDATA_REQ message
4108 	 * coming down as M_PROTO/M_PCPROTO and M_DATA messages for connected
4109 	 * socket.
4110 	 */
4111 	switch (DB_TYPE(mp)) {
4112 	case M_DATA:
4113 		/* sockfs never sends down M_DATA */
4114 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4115 		freemsg(mp);
4116 		return;
4117 
4118 	case M_PROTO:
4119 	case M_PCPROTO:
4120 		tudr = (struct T_unitdata_req *)mp->b_rptr;
4121 		if (MBLKL(mp) < sizeof (*tudr) ||
4122 		    ((t_primp_t)mp->b_rptr)->type != T_UNITDATA_REQ) {
4123 			icmp_wput_other(q, mp);
4124 			return;
4125 		}
4126 		break;
4127 
4128 	default:
4129 		icmp_wput_other(q, mp);
4130 		return;
4131 	}
4132 
4133 	/* Handle valid T_UNITDATA_REQ here */
4134 	data_mp = mp->b_cont;
4135 	if (data_mp == NULL) {
4136 		error = EPROTO;
4137 		goto ud_error2;
4138 	}
4139 	mp->b_cont = NULL;
4140 
4141 	if (!MBLKIN(mp, 0, tudr->DEST_offset + tudr->DEST_length)) {
4142 		error = EADDRNOTAVAIL;
4143 		goto ud_error2;
4144 	}
4145 
4146 	/*
4147 	 * All Solaris components should pass a db_credp
4148 	 * for this message, hence we ASSERT.
4149 	 * On production kernels we return an error to be robust against
4150 	 * random streams modules sitting on top of us.
4151 	 */
4152 	cr = msg_getcred(mp, &pid);
4153 	ASSERT(cr != NULL);
4154 	if (cr == NULL) {
4155 		error = EINVAL;
4156 		goto ud_error2;
4157 	}
4158 
4159 	/*
4160 	 * If a port has not been bound to the stream, fail.
4161 	 * This is not a problem when sockfs is directly
4162 	 * above us, because it will ensure that the socket
4163 	 * is first bound before allowing data to be sent.
4164 	 */
4165 	if (icmp->icmp_state == TS_UNBND) {
4166 		error = EPROTO;
4167 		goto ud_error2;
4168 	}
4169 	addr = (struct sockaddr *)&mp->b_rptr[tudr->DEST_offset];
4170 	addrlen = tudr->DEST_length;
4171 
4172 	switch (connp->conn_family) {
4173 	case AF_INET6:
4174 		sin6 = (sin6_t *)addr;
4175 		if (!OK_32PTR((char *)sin6) || (addrlen != sizeof (sin6_t)) ||
4176 		    (sin6->sin6_family != AF_INET6)) {
4177 			error = EADDRNOTAVAIL;
4178 			goto ud_error2;
4179 		}
4180 
4181 		/* No support for mapped addresses on raw sockets */
4182 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
4183 			error = EADDRNOTAVAIL;
4184 			goto ud_error2;
4185 		}
4186 		srcid = sin6->__sin6_src_id;
4187 
4188 		/*
4189 		 * If the local address is a mapped address return
4190 		 * an error.
4191 		 * It would be possible to send an IPv6 packet but the
4192 		 * response would never make it back to the application
4193 		 * since it is bound to a mapped address.
4194 		 */
4195 		if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
4196 			error = EADDRNOTAVAIL;
4197 			goto ud_error2;
4198 		}
4199 
4200 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
4201 			sin6->sin6_addr = ipv6_loopback;
4202 
4203 		if (tudr->OPT_length != 0) {
4204 			/*
4205 			 * If we are connected then the destination needs to be
4206 			 * the same as the connected one.
4207 			 */
4208 			if (icmp->icmp_state == TS_DATA_XFER &&
4209 			    !conn_same_as_last_v6(connp, sin6)) {
4210 				error = EISCONN;
4211 				goto ud_error2;
4212 			}
4213 			error = icmp_output_ancillary(connp, NULL, sin6,
4214 			    data_mp, mp, NULL, cr, pid);
4215 		} else {
4216 			ip_xmit_attr_t *ixa;
4217 
4218 			/*
4219 			 * We have to allocate an ip_xmit_attr_t before we grab
4220 			 * conn_lock and we need to hold conn_lock once we've
4221 			 * checked conn_same_as_last_v6 to handle concurrent
4222 			 * send* calls on a socket.
4223 			 */
4224 			ixa = conn_get_ixa(connp, B_FALSE);
4225 			if (ixa == NULL) {
4226 				error = ENOMEM;
4227 				goto ud_error2;
4228 			}
4229 			mutex_enter(&connp->conn_lock);
4230 
4231 			if (conn_same_as_last_v6(connp, sin6) &&
4232 			    connp->conn_lastsrcid == srcid &&
4233 			    ipsec_outbound_policy_current(ixa)) {
4234 				/* icmp_output_lastdst drops conn_lock */
4235 				error = icmp_output_lastdst(connp, data_mp, cr,
4236 				    pid, ixa);
4237 			} else {
4238 				/* icmp_output_newdst drops conn_lock */
4239 				error = icmp_output_newdst(connp, data_mp, NULL,
4240 				    sin6, cr, pid, ixa);
4241 			}
4242 			ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
4243 		}
4244 		if (error == 0) {
4245 			freeb(mp);
4246 			return;
4247 		}
4248 		break;
4249 
4250 	case AF_INET:
4251 		sin = (sin_t *)addr;
4252 		if ((!OK_32PTR((char *)sin) || addrlen != sizeof (sin_t)) ||
4253 		    (sin->sin_family != AF_INET)) {
4254 			error = EADDRNOTAVAIL;
4255 			goto ud_error2;
4256 		}
4257 		if (sin->sin_addr.s_addr == INADDR_ANY)
4258 			sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
4259 
4260 		/* Protocol 255 contains full IP headers */
4261 		/* Read without holding lock */
4262 		if (icmp->icmp_hdrincl) {
4263 			if (MBLKL(data_mp) < IP_SIMPLE_HDR_LENGTH) {
4264 				if (!pullupmsg(data_mp, IP_SIMPLE_HDR_LENGTH)) {
4265 					error = EINVAL;
4266 					goto ud_error2;
4267 				}
4268 			}
4269 			error = icmp_output_hdrincl(connp, data_mp, cr, pid);
4270 			if (error == 0) {
4271 				freeb(mp);
4272 				return;
4273 			}
4274 			/* data_mp consumed above */
4275 			data_mp = NULL;
4276 			goto ud_error2;
4277 		}
4278 
4279 		if (tudr->OPT_length != 0) {
4280 			/*
4281 			 * If we are connected then the destination needs to be
4282 			 * the same as the connected one.
4283 			 */
4284 			if (icmp->icmp_state == TS_DATA_XFER &&
4285 			    !conn_same_as_last_v4(connp, sin)) {
4286 				error = EISCONN;
4287 				goto ud_error2;
4288 			}
4289 			error = icmp_output_ancillary(connp, sin, NULL,
4290 			    data_mp, mp, NULL, cr, pid);
4291 		} else {
4292 			ip_xmit_attr_t *ixa;
4293 
4294 			/*
4295 			 * We have to allocate an ip_xmit_attr_t before we grab
4296 			 * conn_lock and we need to hold conn_lock once we've
4297 			 * checked conn_same_as_last_v4 to handle concurrent
4298 			 * send* calls on a socket.
4299 			 */
4300 			ixa = conn_get_ixa(connp, B_FALSE);
4301 			if (ixa == NULL) {
4302 				error = ENOMEM;
4303 				goto ud_error2;
4304 			}
4305 			mutex_enter(&connp->conn_lock);
4306 
4307 			if (conn_same_as_last_v4(connp, sin) &&
4308 			    ipsec_outbound_policy_current(ixa)) {
4309 				/* icmp_output_lastdst drops conn_lock */
4310 				error = icmp_output_lastdst(connp, data_mp, cr,
4311 				    pid, ixa);
4312 			} else {
4313 				/* icmp_output_newdst drops conn_lock */
4314 				error = icmp_output_newdst(connp, data_mp, sin,
4315 				    NULL, cr, pid, ixa);
4316 			}
4317 			ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
4318 		}
4319 		if (error == 0) {
4320 			freeb(mp);
4321 			return;
4322 		}
4323 		break;
4324 	}
4325 	ASSERT(mp != NULL);
4326 	/* mp is freed by the following routine */
4327 	icmp_ud_err(q, mp, (t_scalar_t)error);
4328 	return;
4329 
4330 ud_error2:
4331 	BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4332 	freemsg(data_mp);
4333 	ASSERT(mp != NULL);
4334 	/* mp is freed by the following routine */
4335 	icmp_ud_err(q, mp, (t_scalar_t)error);
4336 }
4337 
4338 /*
4339  * Handle the case of the IP address or flow label being different
4340  * for both IPv4 and IPv6.
4341  *
4342  * NOTE: The caller must hold conn_lock and we drop it here.
4343  */
4344 static int
4345 icmp_output_newdst(conn_t *connp, mblk_t *data_mp, sin_t *sin, sin6_t *sin6,
4346     cred_t *cr, pid_t pid, ip_xmit_attr_t *ixa)
4347 {
4348 	icmp_t		*icmp = connp->conn_icmp;
4349 	icmp_stack_t	*is = icmp->icmp_is;
4350 	int		error;
4351 	ip_xmit_attr_t	*oldixa;
4352 	boolean_t	do_ipsec;
4353 	uint_t		srcid;
4354 	uint32_t	flowinfo;
4355 	in6_addr_t	v6src;
4356 	in6_addr_t	v6dst;
4357 	in6_addr_t	v6nexthop;
4358 	in_port_t	dstport;
4359 
4360 	ASSERT(MUTEX_HELD(&connp->conn_lock));
4361 	ASSERT(ixa != NULL);
4362 
4363 	/*
4364 	 * We hold conn_lock across all the use and modifications of
4365 	 * the conn_lastdst, conn_ixa, and conn_xmit_ipp to ensure that they
4366 	 * stay consistent.
4367 	 */
4368 
4369 	ASSERT(cr != NULL);
4370 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
4371 	ixa->ixa_cred = cr;
4372 	ixa->ixa_cpid = pid;
4373 	if (is_system_labeled()) {
4374 		/* We need to restart with a label based on the cred */
4375 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
4376 	}
4377 	/*
4378 	 * If we are connected then the destination needs to be the
4379 	 * same as the connected one, which is not the case here since we
4380 	 * checked for that above.
4381 	 */
4382 	if (icmp->icmp_state == TS_DATA_XFER) {
4383 		mutex_exit(&connp->conn_lock);
4384 		error = EISCONN;
4385 		goto ud_error;
4386 	}
4387 
4388 	/* In case previous destination was multicast or multirt */
4389 	ip_attr_newdst(ixa);
4390 
4391 	/*
4392 	 * If laddr is unspecified then we look at sin6_src_id.
4393 	 * We will give precedence to a source address set with IPV6_PKTINFO
4394 	 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
4395 	 * want ip_attr_connect to select a source (since it can fail) when
4396 	 * IPV6_PKTINFO is specified.
4397 	 * If this doesn't result in a source address then we get a source
4398 	 * from ip_attr_connect() below.
4399 	 */
4400 	v6src = connp->conn_saddr_v6;
4401 	if (sin != NULL) {
4402 		IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
4403 		dstport = sin->sin_port;
4404 		flowinfo = 0;
4405 		srcid = 0;
4406 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
4407 		if (srcid != 0 && V4_PART_OF_V6(&v6src) == INADDR_ANY) {
4408 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
4409 			    connp->conn_netstack);
4410 		}
4411 		ixa->ixa_flags |= IXAF_IS_IPV4;
4412 	} else {
4413 		v6dst = sin6->sin6_addr;
4414 		dstport = sin6->sin6_port;
4415 		flowinfo = sin6->sin6_flowinfo;
4416 		srcid = sin6->__sin6_src_id;
4417 		if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
4418 			ixa->ixa_scopeid = sin6->sin6_scope_id;
4419 			ixa->ixa_flags |= IXAF_SCOPEID_SET;
4420 		} else {
4421 			ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
4422 		}
4423 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
4424 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
4425 			    connp->conn_netstack);
4426 		}
4427 		if (IN6_IS_ADDR_V4MAPPED(&v6dst))
4428 			ixa->ixa_flags |= IXAF_IS_IPV4;
4429 		else
4430 			ixa->ixa_flags &= ~IXAF_IS_IPV4;
4431 	}
4432 	/* Handle IPV6_PKTINFO setting source address. */
4433 	if (IN6_IS_ADDR_UNSPECIFIED(&v6src) &&
4434 	    (connp->conn_xmit_ipp.ipp_fields & IPPF_ADDR)) {
4435 		ip_pkt_t *ipp = &connp->conn_xmit_ipp;
4436 
4437 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
4438 			if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
4439 				v6src = ipp->ipp_addr;
4440 		} else {
4441 			if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
4442 				v6src = ipp->ipp_addr;
4443 		}
4444 	}
4445 
4446 	/* Defer IPsec if it might need to look at ICMP type/code */
4447 	switch (ixa->ixa_protocol) {
4448 	case IPPROTO_ICMP:
4449 	case IPPROTO_ICMPV6:
4450 		do_ipsec = B_FALSE;
4451 		break;
4452 	default:
4453 		do_ipsec = B_TRUE;
4454 	}
4455 
4456 	ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &v6dst, &v6nexthop);
4457 	mutex_exit(&connp->conn_lock);
4458 
4459 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
4460 	    &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
4461 	    (do_ipsec ? IPDF_IPSEC : 0));
4462 	switch (error) {
4463 	case 0:
4464 		break;
4465 	case EADDRNOTAVAIL:
4466 		/*
4467 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
4468 		 * Don't have the application see that errno
4469 		 */
4470 		error = ENETUNREACH;
4471 		goto failed;
4472 	case ENETDOWN:
4473 		/*
4474 		 * Have !ipif_addr_ready address; drop packet silently
4475 		 * until we can get applications to not send until we
4476 		 * are ready.
4477 		 */
4478 		error = 0;
4479 		goto failed;
4480 	case EHOSTUNREACH:
4481 	case ENETUNREACH:
4482 		if (ixa->ixa_ire != NULL) {
4483 			/*
4484 			 * Let conn_ip_output/ire_send_noroute return
4485 			 * the error and send any local ICMP error.
4486 			 */
4487 			error = 0;
4488 			break;
4489 		}
4490 		/* FALLTHRU */
4491 	default:
4492 	failed:
4493 		goto ud_error;
4494 	}
4495 
4496 	mutex_enter(&connp->conn_lock);
4497 	/*
4498 	 * While we dropped the lock some other thread might have connected
4499 	 * this socket. If so we bail out with EISCONN to ensure that the
4500 	 * connecting thread is the one that updates conn_ixa, conn_ht_*
4501 	 * and conn_*last*.
4502 	 */
4503 	if (icmp->icmp_state == TS_DATA_XFER) {
4504 		mutex_exit(&connp->conn_lock);
4505 		error = EISCONN;
4506 		goto ud_error;
4507 	}
4508 
4509 	/*
4510 	 * We need to rebuild the headers if
4511 	 *  - we are labeling packets (could be different for different
4512 	 *    destinations)
4513 	 *  - we have a source route (or routing header) since we need to
4514 	 *    massage that to get the pseudo-header checksum
4515 	 *  - a socket option with COA_HEADER_CHANGED has been set which
4516 	 *    set conn_v6lastdst to zero.
4517 	 *
4518 	 * Otherwise the prepend function will just update the src, dst,
4519 	 * and flow label.
4520 	 */
4521 	if (is_system_labeled()) {
4522 		/* TX MLP requires SCM_UCRED and don't have that here */
4523 		if (connp->conn_mlp_type != mlptSingle) {
4524 			mutex_exit(&connp->conn_lock);
4525 			error = ECONNREFUSED;
4526 			goto ud_error;
4527 		}
4528 		/*
4529 		 * Check whether Trusted Solaris policy allows communication
4530 		 * with this host, and pretend that the destination is
4531 		 * unreachable if not.
4532 		 * Compute any needed label and place it in ipp_label_v4/v6.
4533 		 *
4534 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
4535 		 * ipp_label_v4/v6 to form the packet.
4536 		 *
4537 		 * Tsol note: Since we hold conn_lock we know no other
4538 		 * thread manipulates conn_xmit_ipp.
4539 		 */
4540 		error = conn_update_label(connp, ixa, &v6dst,
4541 		    &connp->conn_xmit_ipp);
4542 		if (error != 0) {
4543 			mutex_exit(&connp->conn_lock);
4544 			goto ud_error;
4545 		}
4546 		/* Rebuild the header template */
4547 		error = icmp_build_hdr_template(connp, &v6src, &v6dst,
4548 		    flowinfo);
4549 		if (error != 0) {
4550 			mutex_exit(&connp->conn_lock);
4551 			goto ud_error;
4552 		}
4553 	} else if (connp->conn_xmit_ipp.ipp_fields &
4554 	    (IPPF_IPV4_OPTIONS|IPPF_RTHDR) ||
4555 	    IN6_IS_ADDR_UNSPECIFIED(&connp->conn_v6lastdst)) {
4556 		/* Rebuild the header template */
4557 		error = icmp_build_hdr_template(connp, &v6src, &v6dst,
4558 		    flowinfo);
4559 		if (error != 0) {
4560 			mutex_exit(&connp->conn_lock);
4561 			goto ud_error;
4562 		}
4563 	} else {
4564 		/* Simply update the destination address if no source route */
4565 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
4566 			ipha_t	*ipha = (ipha_t *)connp->conn_ht_iphc;
4567 
4568 			IN6_V4MAPPED_TO_IPADDR(&v6dst, ipha->ipha_dst);
4569 			if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
4570 				ipha->ipha_fragment_offset_and_flags |=
4571 				    IPH_DF_HTONS;
4572 			} else {
4573 				ipha->ipha_fragment_offset_and_flags &=
4574 				    ~IPH_DF_HTONS;
4575 			}
4576 		} else {
4577 			ip6_t *ip6h = (ip6_t *)connp->conn_ht_iphc;
4578 			ip6h->ip6_dst = v6dst;
4579 		}
4580 	}
4581 
4582 	/*
4583 	 * Remember the dst etc which corresponds to the built header
4584 	 * template and conn_ixa.
4585 	 */
4586 	oldixa = conn_replace_ixa(connp, ixa);
4587 	connp->conn_v6lastdst = v6dst;
4588 	connp->conn_lastflowinfo = flowinfo;
4589 	connp->conn_lastscopeid = ixa->ixa_scopeid;
4590 	connp->conn_lastsrcid = srcid;
4591 	/* Also remember a source to use together with lastdst */
4592 	connp->conn_v6lastsrc = v6src;
4593 
4594 	data_mp = icmp_prepend_header_template(connp, ixa, data_mp, &v6src,
4595 	    flowinfo, &error);
4596 
4597 	/* Done with conn_t */
4598 	mutex_exit(&connp->conn_lock);
4599 	ixa_refrele(oldixa);
4600 
4601 	if (data_mp == NULL) {
4602 		ASSERT(error != 0);
4603 		goto ud_error;
4604 	}
4605 
4606 	if (!do_ipsec) {
4607 		/* Policy might differ for different ICMP type/code */
4608 		data_mp = icmp_output_attach_policy(data_mp, connp, ixa);
4609 		if (data_mp == NULL) {
4610 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4611 			error = EHOSTUNREACH;	/* IPsec policy failure */
4612 			goto done;
4613 		}
4614 	}
4615 
4616 	/* We're done.  Pass the packet to ip. */
4617 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
4618 
4619 	error = conn_ip_output(data_mp, ixa);
4620 	/* No rawipOutErrors if an error since IP increases its error counter */
4621 	switch (error) {
4622 	case 0:
4623 		break;
4624 	case EWOULDBLOCK:
4625 		(void) ixa_check_drain_insert(connp, ixa);
4626 		error = 0;
4627 		break;
4628 	case EADDRNOTAVAIL:
4629 		/*
4630 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
4631 		 * Don't have the application see that errno
4632 		 */
4633 		error = ENETUNREACH;
4634 		/* FALLTHRU */
4635 	default:
4636 		mutex_enter(&connp->conn_lock);
4637 		/*
4638 		 * Clear the source and v6lastdst so we call ip_attr_connect
4639 		 * for the next packet and try to pick a better source.
4640 		 */
4641 		if (connp->conn_mcbc_bind)
4642 			connp->conn_saddr_v6 = ipv6_all_zeros;
4643 		else
4644 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
4645 		connp->conn_v6lastdst = ipv6_all_zeros;
4646 		mutex_exit(&connp->conn_lock);
4647 		break;
4648 	}
4649 done:
4650 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
4651 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
4652 	ixa->ixa_cpid = connp->conn_cpid;
4653 	ixa_refrele(ixa);
4654 	return (error);
4655 
4656 ud_error:
4657 	ASSERT(!(ixa->ixa_free_flags & IXA_FREE_CRED));
4658 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
4659 	ixa->ixa_cpid = connp->conn_cpid;
4660 	ixa_refrele(ixa);
4661 
4662 	BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4663 	freemsg(data_mp);
4664 	return (error);
4665 }
4666 
4667 /* ARGSUSED */
4668 static void
4669 icmp_wput_fallback(queue_t *q, mblk_t *mp)
4670 {
4671 #ifdef DEBUG
4672 	cmn_err(CE_CONT, "icmp_wput_fallback: Message during fallback \n");
4673 #endif
4674 	freemsg(mp);
4675 }
4676 
4677 static void
4678 icmp_wput_other(queue_t *q, mblk_t *mp)
4679 {
4680 	uchar_t	*rptr = mp->b_rptr;
4681 	struct iocblk *iocp;
4682 	conn_t	*connp = Q_TO_CONN(q);
4683 	icmp_t	*icmp = connp->conn_icmp;
4684 	cred_t *cr;
4685 
4686 	switch (mp->b_datap->db_type) {
4687 	case M_PROTO:
4688 	case M_PCPROTO:
4689 		if (mp->b_wptr - rptr < sizeof (t_scalar_t)) {
4690 			/*
4691 			 * If the message does not contain a PRIM_type,
4692 			 * throw it away.
4693 			 */
4694 			freemsg(mp);
4695 			return;
4696 		}
4697 		switch (((t_primp_t)rptr)->type) {
4698 		case T_ADDR_REQ:
4699 			icmp_addr_req(q, mp);
4700 			return;
4701 		case O_T_BIND_REQ:
4702 		case T_BIND_REQ:
4703 			icmp_tpi_bind(q, mp);
4704 			return;
4705 		case T_CONN_REQ:
4706 			icmp_tpi_connect(q, mp);
4707 			return;
4708 		case T_CAPABILITY_REQ:
4709 			icmp_capability_req(q, mp);
4710 			return;
4711 		case T_INFO_REQ:
4712 			icmp_info_req(q, mp);
4713 			return;
4714 		case T_UNITDATA_REQ:
4715 			/*
4716 			 * If a T_UNITDATA_REQ gets here, the address must
4717 			 * be bad.  Valid T_UNITDATA_REQs are handled
4718 			 * in icmp_wput.
4719 			 */
4720 			icmp_ud_err(q, mp, EADDRNOTAVAIL);
4721 			return;
4722 		case T_UNBIND_REQ:
4723 			icmp_tpi_unbind(q, mp);
4724 			return;
4725 		case T_SVR4_OPTMGMT_REQ:
4726 			/*
4727 			 * All Solaris components should pass a db_credp
4728 			 * for this TPI message, hence we ASSERT.
4729 			 * But in case there is some other M_PROTO that looks
4730 			 * like a TPI message sent by some other kernel
4731 			 * component, we check and return an error.
4732 			 */
4733 			cr = msg_getcred(mp, NULL);
4734 			ASSERT(cr != NULL);
4735 			if (cr == NULL) {
4736 				icmp_err_ack(q, mp, TSYSERR, EINVAL);
4737 				return;
4738 			}
4739 
4740 			if (!snmpcom_req(q, mp, icmp_snmp_set, ip_snmp_get,
4741 			    cr)) {
4742 				svr4_optcom_req(q, mp, cr, &icmp_opt_obj);
4743 			}
4744 			return;
4745 
4746 		case T_OPTMGMT_REQ:
4747 			/*
4748 			 * All Solaris components should pass a db_credp
4749 			 * for this TPI message, hence we ASSERT.
4750 			 * But in case there is some other M_PROTO that looks
4751 			 * like a TPI message sent by some other kernel
4752 			 * component, we check and return an error.
4753 			 */
4754 			cr = msg_getcred(mp, NULL);
4755 			ASSERT(cr != NULL);
4756 			if (cr == NULL) {
4757 				icmp_err_ack(q, mp, TSYSERR, EINVAL);
4758 				return;
4759 			}
4760 			tpi_optcom_req(q, mp, cr, &icmp_opt_obj);
4761 			return;
4762 
4763 		case T_DISCON_REQ:
4764 			icmp_tpi_disconnect(q, mp);
4765 			return;
4766 
4767 		/* The following TPI message is not supported by icmp. */
4768 		case O_T_CONN_RES:
4769 		case T_CONN_RES:
4770 			icmp_err_ack(q, mp, TNOTSUPPORT, 0);
4771 			return;
4772 
4773 		/* The following 3 TPI requests are illegal for icmp. */
4774 		case T_DATA_REQ:
4775 		case T_EXDATA_REQ:
4776 		case T_ORDREL_REQ:
4777 			icmp_err_ack(q, mp, TNOTSUPPORT, 0);
4778 			return;
4779 		default:
4780 			break;
4781 		}
4782 		break;
4783 	case M_FLUSH:
4784 		if (*rptr & FLUSHW)
4785 			flushq(q, FLUSHDATA);
4786 		break;
4787 	case M_IOCTL:
4788 		iocp = (struct iocblk *)mp->b_rptr;
4789 		switch (iocp->ioc_cmd) {
4790 		case TI_GETPEERNAME:
4791 			if (icmp->icmp_state != TS_DATA_XFER) {
4792 				/*
4793 				 * If a default destination address has not
4794 				 * been associated with the stream, then we
4795 				 * don't know the peer's name.
4796 				 */
4797 				iocp->ioc_error = ENOTCONN;
4798 				iocp->ioc_count = 0;
4799 				mp->b_datap->db_type = M_IOCACK;
4800 				qreply(q, mp);
4801 				return;
4802 			}
4803 			/* FALLTHRU */
4804 		case TI_GETMYNAME:
4805 			/*
4806 			 * For TI_GETPEERNAME and TI_GETMYNAME, we first
4807 			 * need to copyin the user's strbuf structure.
4808 			 * Processing will continue in the M_IOCDATA case
4809 			 * below.
4810 			 */
4811 			mi_copyin(q, mp, NULL,
4812 			    SIZEOF_STRUCT(strbuf, iocp->ioc_flag));
4813 			return;
4814 		default:
4815 			break;
4816 		}
4817 		break;
4818 	case M_IOCDATA:
4819 		icmp_wput_iocdata(q, mp);
4820 		return;
4821 	default:
4822 		/* Unrecognized messages are passed through without change. */
4823 		break;
4824 	}
4825 	ip_wput_nondata(q, mp);
4826 }
4827 
4828 /*
4829  * icmp_wput_iocdata is called by icmp_wput_other to handle all M_IOCDATA
4830  * messages.
4831  */
4832 static void
4833 icmp_wput_iocdata(queue_t *q, mblk_t *mp)
4834 {
4835 	mblk_t		*mp1;
4836 	STRUCT_HANDLE(strbuf, sb);
4837 	uint_t		addrlen;
4838 	conn_t		*connp = Q_TO_CONN(q);
4839 	icmp_t		*icmp = connp->conn_icmp;
4840 
4841 	/* Make sure it is one of ours. */
4842 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4843 	case TI_GETMYNAME:
4844 	case TI_GETPEERNAME:
4845 		break;
4846 	default:
4847 		ip_wput_nondata(q, mp);
4848 		return;
4849 	}
4850 
4851 	switch (mi_copy_state(q, mp, &mp1)) {
4852 	case -1:
4853 		return;
4854 	case MI_COPY_CASE(MI_COPY_IN, 1):
4855 		break;
4856 	case MI_COPY_CASE(MI_COPY_OUT, 1):
4857 		/*
4858 		 * The address has been copied out, so now
4859 		 * copyout the strbuf.
4860 		 */
4861 		mi_copyout(q, mp);
4862 		return;
4863 	case MI_COPY_CASE(MI_COPY_OUT, 2):
4864 		/*
4865 		 * The address and strbuf have been copied out.
4866 		 * We're done, so just acknowledge the original
4867 		 * M_IOCTL.
4868 		 */
4869 		mi_copy_done(q, mp, 0);
4870 		return;
4871 	default:
4872 		/*
4873 		 * Something strange has happened, so acknowledge
4874 		 * the original M_IOCTL with an EPROTO error.
4875 		 */
4876 		mi_copy_done(q, mp, EPROTO);
4877 		return;
4878 	}
4879 
4880 	/*
4881 	 * Now we have the strbuf structure for TI_GETMYNAME
4882 	 * and TI_GETPEERNAME.  Next we copyout the requested
4883 	 * address and then we'll copyout the strbuf.
4884 	 */
4885 	STRUCT_SET_HANDLE(sb, ((struct iocblk *)mp->b_rptr)->ioc_flag,
4886 	    (void *)mp1->b_rptr);
4887 
4888 	if (connp->conn_family == AF_INET)
4889 		addrlen = sizeof (sin_t);
4890 	else
4891 		addrlen = sizeof (sin6_t);
4892 
4893 	if (STRUCT_FGET(sb, maxlen) < addrlen) {
4894 		mi_copy_done(q, mp, EINVAL);
4895 		return;
4896 	}
4897 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4898 	case TI_GETMYNAME:
4899 		break;
4900 	case TI_GETPEERNAME:
4901 		if (icmp->icmp_state != TS_DATA_XFER) {
4902 			mi_copy_done(q, mp, ENOTCONN);
4903 			return;
4904 		}
4905 		break;
4906 	default:
4907 		mi_copy_done(q, mp, EPROTO);
4908 		return;
4909 	}
4910 	mp1 = mi_copyout_alloc(q, mp, STRUCT_FGETP(sb, buf), addrlen, B_TRUE);
4911 	if (!mp1)
4912 		return;
4913 
4914 	STRUCT_FSET(sb, len, addrlen);
4915 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4916 	case TI_GETMYNAME:
4917 		(void) conn_getsockname(connp, (struct sockaddr *)mp1->b_wptr,
4918 		    &addrlen);
4919 		break;
4920 	case TI_GETPEERNAME:
4921 		(void) conn_getpeername(connp, (struct sockaddr *)mp1->b_wptr,
4922 		    &addrlen);
4923 		break;
4924 	}
4925 	mp1->b_wptr += addrlen;
4926 	/* Copy out the address */
4927 	mi_copyout(q, mp);
4928 }
4929 
4930 void
4931 icmp_ddi_g_init(void)
4932 {
4933 	icmp_max_optsize = optcom_max_optsize(icmp_opt_obj.odb_opt_des_arr,
4934 	    icmp_opt_obj.odb_opt_arr_cnt);
4935 
4936 	/*
4937 	 * We want to be informed each time a stack is created or
4938 	 * destroyed in the kernel, so we can maintain the
4939 	 * set of icmp_stack_t's.
4940 	 */
4941 	netstack_register(NS_ICMP, rawip_stack_init, NULL, rawip_stack_fini);
4942 }
4943 
4944 void
4945 icmp_ddi_g_destroy(void)
4946 {
4947 	netstack_unregister(NS_ICMP);
4948 }
4949 
4950 #define	INET_NAME	"ip"
4951 
4952 /*
4953  * Initialize the ICMP stack instance.
4954  */
4955 static void *
4956 rawip_stack_init(netstackid_t stackid, netstack_t *ns)
4957 {
4958 	icmp_stack_t	*is;
4959 	int		error = 0;
4960 	size_t		arrsz;
4961 	major_t		major;
4962 
4963 	is = (icmp_stack_t *)kmem_zalloc(sizeof (*is), KM_SLEEP);
4964 	is->is_netstack = ns;
4965 
4966 	arrsz = sizeof (icmp_propinfo_tbl);
4967 	is->is_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz, KM_SLEEP);
4968 	bcopy(icmp_propinfo_tbl, is->is_propinfo_tbl, arrsz);
4969 
4970 	is->is_ksp = rawip_kstat_init(stackid);
4971 
4972 	major = mod_name_to_major(INET_NAME);
4973 	error = ldi_ident_from_major(major, &is->is_ldi_ident);
4974 	ASSERT(error == 0);
4975 	return (is);
4976 }
4977 
4978 /*
4979  * Free the ICMP stack instance.
4980  */
4981 static void
4982 rawip_stack_fini(netstackid_t stackid, void *arg)
4983 {
4984 	icmp_stack_t *is = (icmp_stack_t *)arg;
4985 
4986 	kmem_free(is->is_propinfo_tbl, sizeof (icmp_propinfo_tbl));
4987 	is->is_propinfo_tbl = NULL;
4988 
4989 	rawip_kstat_fini(stackid, is->is_ksp);
4990 	is->is_ksp = NULL;
4991 	ldi_ident_release(is->is_ldi_ident);
4992 	kmem_free(is, sizeof (*is));
4993 }
4994 
4995 static void *
4996 rawip_kstat_init(netstackid_t stackid) {
4997 	kstat_t	*ksp;
4998 
4999 	rawip_named_kstat_t template = {
5000 		{ "inDatagrams",	KSTAT_DATA_UINT32, 0 },
5001 		{ "inCksumErrs",	KSTAT_DATA_UINT32, 0 },
5002 		{ "inErrors",		KSTAT_DATA_UINT32, 0 },
5003 		{ "outDatagrams",	KSTAT_DATA_UINT32, 0 },
5004 		{ "outErrors",		KSTAT_DATA_UINT32, 0 },
5005 	};
5006 
5007 	ksp = kstat_create_netstack("icmp", 0, "rawip", "mib2",
5008 					KSTAT_TYPE_NAMED,
5009 					NUM_OF_FIELDS(rawip_named_kstat_t),
5010 					0, stackid);
5011 	if (ksp == NULL || ksp->ks_data == NULL)
5012 		return (NULL);
5013 
5014 	bcopy(&template, ksp->ks_data, sizeof (template));
5015 	ksp->ks_update = rawip_kstat_update;
5016 	ksp->ks_private = (void *)(uintptr_t)stackid;
5017 
5018 	kstat_install(ksp);
5019 	return (ksp);
5020 }
5021 
5022 static void
5023 rawip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
5024 {
5025 	if (ksp != NULL) {
5026 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
5027 		kstat_delete_netstack(ksp, stackid);
5028 	}
5029 }
5030 
5031 static int
5032 rawip_kstat_update(kstat_t *ksp, int rw)
5033 {
5034 	rawip_named_kstat_t *rawipkp;
5035 	netstackid_t	stackid = (netstackid_t)(uintptr_t)ksp->ks_private;
5036 	netstack_t	*ns;
5037 	icmp_stack_t	*is;
5038 
5039 	if ((ksp == NULL) || (ksp->ks_data == NULL))
5040 		return (EIO);
5041 
5042 	if (rw == KSTAT_WRITE)
5043 		return (EACCES);
5044 
5045 	rawipkp = (rawip_named_kstat_t *)ksp->ks_data;
5046 
5047 	ns = netstack_find_by_stackid(stackid);
5048 	if (ns == NULL)
5049 		return (-1);
5050 	is = ns->netstack_icmp;
5051 	if (is == NULL) {
5052 		netstack_rele(ns);
5053 		return (-1);
5054 	}
5055 	rawipkp->inDatagrams.value.ui32 =  is->is_rawip_mib.rawipInDatagrams;
5056 	rawipkp->inCksumErrs.value.ui32 =  is->is_rawip_mib.rawipInCksumErrs;
5057 	rawipkp->inErrors.value.ui32 =	   is->is_rawip_mib.rawipInErrors;
5058 	rawipkp->outDatagrams.value.ui32 = is->is_rawip_mib.rawipOutDatagrams;
5059 	rawipkp->outErrors.value.ui32 =	   is->is_rawip_mib.rawipOutErrors;
5060 	netstack_rele(ns);
5061 	return (0);
5062 }
5063 
5064 /* ARGSUSED */
5065 int
5066 rawip_accept(sock_lower_handle_t lproto_handle,
5067     sock_lower_handle_t eproto_handle, sock_upper_handle_t sock_handle,
5068     cred_t *cr)
5069 {
5070 	return (EOPNOTSUPP);
5071 }
5072 
5073 /* ARGSUSED */
5074 int
5075 rawip_bind(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5076     socklen_t len, cred_t *cr)
5077 {
5078 	conn_t  *connp = (conn_t *)proto_handle;
5079 	int	error;
5080 
5081 	/* All Solaris components should pass a cred for this operation. */
5082 	ASSERT(cr != NULL);
5083 
5084 	/* Binding to a NULL address really means unbind */
5085 	if (sa == NULL)
5086 		error = rawip_do_unbind(connp);
5087 	else
5088 		error = rawip_do_bind(connp, sa, len);
5089 
5090 	if (error < 0) {
5091 		if (error == -TOUTSTATE)
5092 			error = EINVAL;
5093 		else
5094 			error = proto_tlitosyserr(-error);
5095 	}
5096 	return (error);
5097 }
5098 
5099 static int
5100 rawip_implicit_bind(conn_t *connp)
5101 {
5102 	sin6_t sin6addr;
5103 	sin_t *sin;
5104 	sin6_t *sin6;
5105 	socklen_t len;
5106 	int error;
5107 
5108 	if (connp->conn_family == AF_INET) {
5109 		len = sizeof (struct sockaddr_in);
5110 		sin = (sin_t *)&sin6addr;
5111 		*sin = sin_null;
5112 		sin->sin_family = AF_INET;
5113 		sin->sin_addr.s_addr = INADDR_ANY;
5114 	} else {
5115 		ASSERT(connp->conn_family == AF_INET6);
5116 		len = sizeof (sin6_t);
5117 		sin6 = (sin6_t *)&sin6addr;
5118 		*sin6 = sin6_null;
5119 		sin6->sin6_family = AF_INET6;
5120 		V6_SET_ZERO(sin6->sin6_addr);
5121 	}
5122 
5123 	error = rawip_do_bind(connp, (struct sockaddr *)&sin6addr, len);
5124 
5125 	return ((error < 0) ? proto_tlitosyserr(-error) : error);
5126 }
5127 
5128 static int
5129 rawip_unbind(conn_t *connp)
5130 {
5131 	int error;
5132 
5133 	error = rawip_do_unbind(connp);
5134 	if (error < 0) {
5135 		error = proto_tlitosyserr(-error);
5136 	}
5137 	return (error);
5138 }
5139 
5140 /* ARGSUSED */
5141 int
5142 rawip_listen(sock_lower_handle_t proto_handle, int backlog, cred_t *cr)
5143 {
5144 	return (EOPNOTSUPP);
5145 }
5146 
5147 int
5148 rawip_connect(sock_lower_handle_t proto_handle, const struct sockaddr *sa,
5149     socklen_t len, sock_connid_t *id, cred_t *cr)
5150 {
5151 	conn_t	*connp = (conn_t *)proto_handle;
5152 	icmp_t *icmp = connp->conn_icmp;
5153 	int	error;
5154 	boolean_t did_bind = B_FALSE;
5155 	pid_t	pid = curproc->p_pid;
5156 
5157 	/* All Solaris components should pass a cred for this operation. */
5158 	ASSERT(cr != NULL);
5159 
5160 	if (sa == NULL) {
5161 		/*
5162 		 * Disconnect
5163 		 * Make sure we are connected
5164 		 */
5165 		if (icmp->icmp_state != TS_DATA_XFER)
5166 			return (EINVAL);
5167 
5168 		error = icmp_disconnect(connp);
5169 		return (error);
5170 	}
5171 
5172 	error = proto_verify_ip_addr(connp->conn_family, sa, len);
5173 	if (error != 0)
5174 		return (error);
5175 
5176 	/* do an implicit bind if necessary */
5177 	if (icmp->icmp_state == TS_UNBND) {
5178 		error = rawip_implicit_bind(connp);
5179 		/*
5180 		 * We could be racing with an actual bind, in which case
5181 		 * we would see EPROTO. We cross our fingers and try
5182 		 * to connect.
5183 		 */
5184 		if (!(error == 0 || error == EPROTO))
5185 			return (error);
5186 		did_bind = B_TRUE;
5187 	}
5188 
5189 	/*
5190 	 * set SO_DGRAM_ERRIND
5191 	 */
5192 	connp->conn_dgram_errind = B_TRUE;
5193 
5194 	error = rawip_do_connect(connp, sa, len, cr, pid);
5195 	if (error != 0 && did_bind) {
5196 		int unbind_err;
5197 
5198 		unbind_err = rawip_unbind(connp);
5199 		ASSERT(unbind_err == 0);
5200 	}
5201 
5202 	if (error == 0) {
5203 		*id = 0;
5204 		(*connp->conn_upcalls->su_connected)(connp->conn_upper_handle,
5205 		    0, NULL, -1);
5206 	} else if (error < 0) {
5207 		error = proto_tlitosyserr(-error);
5208 	}
5209 	return (error);
5210 }
5211 
5212 /* ARGSUSED2 */
5213 int
5214 rawip_fallback(sock_lower_handle_t proto_handle, queue_t *q,
5215     boolean_t direct_sockfs, so_proto_quiesced_cb_t quiesced_cb)
5216 {
5217 	conn_t  *connp = (conn_t *)proto_handle;
5218 	icmp_t	*icmp;
5219 	struct T_capability_ack tca;
5220 	struct sockaddr_in6 laddr, faddr;
5221 	socklen_t laddrlen, faddrlen;
5222 	short opts;
5223 	struct stroptions *stropt;
5224 	mblk_t *stropt_mp;
5225 	int error;
5226 
5227 	icmp = connp->conn_icmp;
5228 
5229 	stropt_mp = allocb_wait(sizeof (*stropt), BPRI_HI, STR_NOSIG, NULL);
5230 
5231 	/*
5232 	 * setup the fallback stream that was allocated
5233 	 */
5234 	connp->conn_dev = (dev_t)RD(q)->q_ptr;
5235 	connp->conn_minor_arena = WR(q)->q_ptr;
5236 
5237 	RD(q)->q_ptr = WR(q)->q_ptr = connp;
5238 
5239 	WR(q)->q_qinfo = &icmpwinit;
5240 
5241 	connp->conn_rq = RD(q);
5242 	connp->conn_wq = WR(q);
5243 
5244 	/* Notify stream head about options before sending up data */
5245 	stropt_mp->b_datap->db_type = M_SETOPTS;
5246 	stropt_mp->b_wptr += sizeof (*stropt);
5247 	stropt = (struct stroptions *)stropt_mp->b_rptr;
5248 	stropt->so_flags = SO_WROFF | SO_HIWAT;
5249 	stropt->so_wroff = connp->conn_wroff;
5250 	stropt->so_hiwat = connp->conn_rcvbuf;
5251 	putnext(RD(q), stropt_mp);
5252 
5253 	/*
5254 	 * free helper stream
5255 	 */
5256 	ip_free_helper_stream(connp);
5257 
5258 	/*
5259 	 * Collect the information needed to sync with the sonode
5260 	 */
5261 	icmp_do_capability_ack(icmp, &tca, TC1_INFO);
5262 
5263 	laddrlen = faddrlen = sizeof (sin6_t);
5264 	(void) rawip_getsockname((sock_lower_handle_t)connp,
5265 	    (struct sockaddr *)&laddr, &laddrlen, CRED());
5266 	error = rawip_getpeername((sock_lower_handle_t)connp,
5267 	    (struct sockaddr *)&faddr, &faddrlen, CRED());
5268 	if (error != 0)
5269 		faddrlen = 0;
5270 	opts = 0;
5271 	if (connp->conn_dgram_errind)
5272 		opts |= SO_DGRAM_ERRIND;
5273 	if (connp->conn_ixa->ixa_flags & IXAF_DONTROUTE)
5274 		opts |= SO_DONTROUTE;
5275 
5276 	(*quiesced_cb)(connp->conn_upper_handle, q, &tca,
5277 	    (struct sockaddr *)&laddr, laddrlen,
5278 	    (struct sockaddr *)&faddr, faddrlen, opts);
5279 
5280 	/*
5281 	 * Attempts to send data up during fallback will result in it being
5282 	 * queued in icmp_t. Now we push up any queued packets.
5283 	 */
5284 	mutex_enter(&icmp->icmp_recv_lock);
5285 	while (icmp->icmp_fallback_queue_head != NULL) {
5286 		mblk_t	*mp;
5287 
5288 		mp = icmp->icmp_fallback_queue_head;
5289 		icmp->icmp_fallback_queue_head = mp->b_next;
5290 		mp->b_next = NULL;
5291 		mutex_exit(&icmp->icmp_recv_lock);
5292 		putnext(RD(q), mp);
5293 		mutex_enter(&icmp->icmp_recv_lock);
5294 	}
5295 	icmp->icmp_fallback_queue_tail = icmp->icmp_fallback_queue_head;
5296 
5297 	/*
5298 	 * No longer a streams less socket
5299 	 */
5300 	mutex_enter(&connp->conn_lock);
5301 	connp->conn_flags &= ~IPCL_NONSTR;
5302 	mutex_exit(&connp->conn_lock);
5303 
5304 	mutex_exit(&icmp->icmp_recv_lock);
5305 
5306 	ASSERT(icmp->icmp_fallback_queue_head == NULL &&
5307 	    icmp->icmp_fallback_queue_tail == NULL);
5308 
5309 	ASSERT(connp->conn_ref >= 1);
5310 
5311 	return (0);
5312 }
5313 
5314 /* ARGSUSED2 */
5315 sock_lower_handle_t
5316 rawip_create(int family, int type, int proto, sock_downcalls_t **sock_downcalls,
5317     uint_t *smodep, int *errorp, int flags, cred_t *credp)
5318 {
5319 	conn_t *connp;
5320 
5321 	if (type != SOCK_RAW || (family != AF_INET && family != AF_INET6)) {
5322 		*errorp = EPROTONOSUPPORT;
5323 		return (NULL);
5324 	}
5325 
5326 	connp = rawip_do_open(family, credp, errorp, flags);
5327 	if (connp != NULL) {
5328 		connp->conn_flags |= IPCL_NONSTR;
5329 
5330 		mutex_enter(&connp->conn_lock);
5331 		connp->conn_state_flags &= ~CONN_INCIPIENT;
5332 		mutex_exit(&connp->conn_lock);
5333 		*sock_downcalls = &sock_rawip_downcalls;
5334 		*smodep = SM_ATOMIC;
5335 	} else {
5336 		ASSERT(*errorp != 0);
5337 	}
5338 
5339 	return ((sock_lower_handle_t)connp);
5340 }
5341 
5342 /* ARGSUSED3 */
5343 void
5344 rawip_activate(sock_lower_handle_t proto_handle,
5345     sock_upper_handle_t sock_handle, sock_upcalls_t *sock_upcalls, int flags,
5346     cred_t *cr)
5347 {
5348 	conn_t 			*connp = (conn_t *)proto_handle;
5349 	struct sock_proto_props sopp;
5350 
5351 	/* All Solaris components should pass a cred for this operation. */
5352 	ASSERT(cr != NULL);
5353 
5354 	connp->conn_upcalls = sock_upcalls;
5355 	connp->conn_upper_handle = sock_handle;
5356 
5357 	sopp.sopp_flags = SOCKOPT_WROFF | SOCKOPT_RCVHIWAT | SOCKOPT_RCVLOWAT |
5358 	    SOCKOPT_MAXBLK | SOCKOPT_MAXPSZ | SOCKOPT_MINPSZ;
5359 	sopp.sopp_wroff = connp->conn_wroff;
5360 	sopp.sopp_rxhiwat = connp->conn_rcvbuf;
5361 	sopp.sopp_rxlowat = connp->conn_rcvlowat;
5362 	sopp.sopp_maxblk = INFPSZ;
5363 	sopp.sopp_maxpsz = IP_MAXPACKET;
5364 	sopp.sopp_minpsz = (icmp_mod_info.mi_minpsz == 1) ? 0 :
5365 	    icmp_mod_info.mi_minpsz;
5366 
5367 	(*connp->conn_upcalls->su_set_proto_props)
5368 	    (connp->conn_upper_handle, &sopp);
5369 
5370 	icmp_bind_proto(connp->conn_icmp);
5371 }
5372 
5373 /* ARGSUSED3 */
5374 int
5375 rawip_getpeername(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5376     socklen_t *salenp, cred_t *cr)
5377 {
5378 	conn_t  *connp = (conn_t *)proto_handle;
5379 	icmp_t  *icmp = connp->conn_icmp;
5380 	int	error;
5381 
5382 	/* All Solaris components should pass a cred for this operation. */
5383 	ASSERT(cr != NULL);
5384 
5385 	mutex_enter(&connp->conn_lock);
5386 	if (icmp->icmp_state != TS_DATA_XFER)
5387 		error = ENOTCONN;
5388 	else
5389 		error = conn_getpeername(connp, sa, salenp);
5390 	mutex_exit(&connp->conn_lock);
5391 	return (error);
5392 }
5393 
5394 /* ARGSUSED3 */
5395 int
5396 rawip_getsockname(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5397     socklen_t *salenp, cred_t *cr)
5398 {
5399 	conn_t  *connp = (conn_t *)proto_handle;
5400 	int	error;
5401 
5402 	/* All Solaris components should pass a cred for this operation. */
5403 	ASSERT(cr != NULL);
5404 
5405 	mutex_enter(&connp->conn_lock);
5406 	error = conn_getsockname(connp, sa, salenp);
5407 	mutex_exit(&connp->conn_lock);
5408 	return (error);
5409 }
5410 
5411 int
5412 rawip_setsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
5413     const void *optvalp, socklen_t optlen, cred_t *cr)
5414 {
5415 	conn_t	*connp = (conn_t *)proto_handle;
5416 	int error;
5417 
5418 	/* All Solaris components should pass a cred for this operation. */
5419 	ASSERT(cr != NULL);
5420 
5421 	error = proto_opt_check(level, option_name, optlen, NULL,
5422 	    icmp_opt_obj.odb_opt_des_arr,
5423 	    icmp_opt_obj.odb_opt_arr_cnt,
5424 	    B_TRUE, B_FALSE, cr);
5425 
5426 	if (error != 0) {
5427 		/*
5428 		 * option not recognized
5429 		 */
5430 		if (error < 0) {
5431 			error = proto_tlitosyserr(-error);
5432 		}
5433 		return (error);
5434 	}
5435 
5436 	error = icmp_opt_set(connp, SETFN_OPTCOM_NEGOTIATE, level,
5437 	    option_name, optlen, (uchar_t *)optvalp, (uint_t *)&optlen,
5438 	    (uchar_t *)optvalp, NULL, cr);
5439 
5440 	ASSERT(error >= 0);
5441 
5442 	return (error);
5443 }
5444 
5445 int
5446 rawip_getsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
5447     void *optvalp, socklen_t *optlen, cred_t *cr)
5448 {
5449 	int		error;
5450 	conn_t		*connp = (conn_t *)proto_handle;
5451 	t_uscalar_t	max_optbuf_len;
5452 	void		*optvalp_buf;
5453 	int		len;
5454 
5455 	/* All Solaris components should pass a cred for this operation. */
5456 	ASSERT(cr != NULL);
5457 
5458 	error = proto_opt_check(level, option_name, *optlen, &max_optbuf_len,
5459 	    icmp_opt_obj.odb_opt_des_arr,
5460 	    icmp_opt_obj.odb_opt_arr_cnt,
5461 	    B_FALSE, B_TRUE, cr);
5462 
5463 	if (error != 0) {
5464 		if (error < 0) {
5465 			error = proto_tlitosyserr(-error);
5466 		}
5467 		return (error);
5468 	}
5469 
5470 	optvalp_buf = kmem_alloc(max_optbuf_len, KM_SLEEP);
5471 	len = icmp_opt_get(connp, level, option_name, optvalp_buf);
5472 	if (len == -1) {
5473 		kmem_free(optvalp_buf, max_optbuf_len);
5474 		return (EINVAL);
5475 	}
5476 
5477 	/*
5478 	 * update optlen and copy option value
5479 	 */
5480 	t_uscalar_t size = MIN(len, *optlen);
5481 
5482 	bcopy(optvalp_buf, optvalp, size);
5483 	bcopy(&size, optlen, sizeof (size));
5484 
5485 	kmem_free(optvalp_buf, max_optbuf_len);
5486 	return (0);
5487 }
5488 
5489 /* ARGSUSED1 */
5490 int
5491 rawip_close(sock_lower_handle_t proto_handle, int flags, cred_t *cr)
5492 {
5493 	conn_t	*connp = (conn_t *)proto_handle;
5494 
5495 	/* All Solaris components should pass a cred for this operation. */
5496 	ASSERT(cr != NULL);
5497 
5498 	(void) rawip_do_close(connp);
5499 	return (0);
5500 }
5501 
5502 /* ARGSUSED2 */
5503 int
5504 rawip_shutdown(sock_lower_handle_t proto_handle, int how, cred_t *cr)
5505 {
5506 	conn_t  *connp = (conn_t *)proto_handle;
5507 
5508 	/* All Solaris components should pass a cred for this operation. */
5509 	ASSERT(cr != NULL);
5510 
5511 	/* shut down the send side */
5512 	if (how != SHUT_RD)
5513 		(*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
5514 		    SOCK_OPCTL_SHUT_SEND, 0);
5515 	/* shut down the recv side */
5516 	if (how != SHUT_WR)
5517 		(*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
5518 		    SOCK_OPCTL_SHUT_RECV, 0);
5519 	return (0);
5520 }
5521 
5522 void
5523 rawip_clr_flowctrl(sock_lower_handle_t proto_handle)
5524 {
5525 	conn_t  *connp = (conn_t *)proto_handle;
5526 	icmp_t	*icmp = connp->conn_icmp;
5527 
5528 	mutex_enter(&icmp->icmp_recv_lock);
5529 	connp->conn_flow_cntrld = B_FALSE;
5530 	mutex_exit(&icmp->icmp_recv_lock);
5531 }
5532 
5533 int
5534 rawip_ioctl(sock_lower_handle_t proto_handle, int cmd, intptr_t arg,
5535     int mode, int32_t *rvalp, cred_t *cr)
5536 {
5537 	conn_t  	*connp = (conn_t *)proto_handle;
5538 	int		error;
5539 
5540 	/* All Solaris components should pass a cred for this operation. */
5541 	ASSERT(cr != NULL);
5542 
5543 	/*
5544 	 * If we don't have a helper stream then create one.
5545 	 * ip_create_helper_stream takes care of locking the conn_t,
5546 	 * so this check for NULL is just a performance optimization.
5547 	 */
5548 	if (connp->conn_helper_info == NULL) {
5549 		icmp_stack_t *is = connp->conn_icmp->icmp_is;
5550 
5551 		ASSERT(is->is_ldi_ident != NULL);
5552 
5553 		/*
5554 		 * Create a helper stream for non-STREAMS socket.
5555 		 */
5556 		error = ip_create_helper_stream(connp, is->is_ldi_ident);
5557 		if (error != 0) {
5558 			ip0dbg(("rawip_ioctl: create of IP helper stream "
5559 			    "failed %d\n", error));
5560 			return (error);
5561 		}
5562 	}
5563 
5564 	switch (cmd) {
5565 	case _SIOCSOCKFALLBACK:
5566 	case TI_GETPEERNAME:
5567 	case TI_GETMYNAME:
5568 #ifdef DEBUG
5569 		cmn_err(CE_CONT, "icmp_ioctl cmd 0x%x on non streams"
5570 		    " socket", cmd);
5571 #endif
5572 		error = EINVAL;
5573 		break;
5574 	default:
5575 		/*
5576 		 * Pass on to IP using helper stream
5577 		 */
5578 		error = ldi_ioctl(connp->conn_helper_info->iphs_handle,
5579 		    cmd, arg, mode, cr, rvalp);
5580 		break;
5581 	}
5582 	return (error);
5583 }
5584 
5585 int
5586 rawip_send(sock_lower_handle_t proto_handle, mblk_t *mp, struct nmsghdr *msg,
5587     cred_t *cr)
5588 {
5589 	sin6_t		*sin6;
5590 	sin_t		*sin = NULL;
5591 	uint_t		srcid;
5592 	conn_t		*connp = (conn_t *)proto_handle;
5593 	icmp_t		*icmp = connp->conn_icmp;
5594 	int		error = 0;
5595 	icmp_stack_t	*is = icmp->icmp_is;
5596 	pid_t		pid = curproc->p_pid;
5597 	ip_xmit_attr_t	*ixa;
5598 
5599 	ASSERT(DB_TYPE(mp) == M_DATA);
5600 
5601 	/* All Solaris components should pass a cred for this operation. */
5602 	ASSERT(cr != NULL);
5603 
5604 	/* do an implicit bind if necessary */
5605 	if (icmp->icmp_state == TS_UNBND) {
5606 		error = rawip_implicit_bind(connp);
5607 		/*
5608 		 * We could be racing with an actual bind, in which case
5609 		 * we would see EPROTO. We cross our fingers and try
5610 		 * to connect.
5611 		 */
5612 		if (!(error == 0 || error == EPROTO)) {
5613 			freemsg(mp);
5614 			return (error);
5615 		}
5616 	}
5617 
5618 	/* Protocol 255 contains full IP headers */
5619 	/* Read without holding lock */
5620 	if (icmp->icmp_hdrincl) {
5621 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
5622 		if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH) {
5623 			if (!pullupmsg(mp, IP_SIMPLE_HDR_LENGTH)) {
5624 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5625 				freemsg(mp);
5626 				return (EINVAL);
5627 			}
5628 		}
5629 		error = icmp_output_hdrincl(connp, mp, cr, pid);
5630 		if (is->is_sendto_ignerr)
5631 			return (0);
5632 		else
5633 			return (error);
5634 	}
5635 
5636 	/* Connected? */
5637 	if (msg->msg_name == NULL) {
5638 		if (icmp->icmp_state != TS_DATA_XFER) {
5639 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5640 			return (EDESTADDRREQ);
5641 		}
5642 		if (msg->msg_controllen != 0) {
5643 			error = icmp_output_ancillary(connp, NULL, NULL, mp,
5644 			    NULL, msg, cr, pid);
5645 		} else {
5646 			error = icmp_output_connected(connp, mp, cr, pid);
5647 		}
5648 		if (is->is_sendto_ignerr)
5649 			return (0);
5650 		else
5651 			return (error);
5652 	}
5653 	if (icmp->icmp_state == TS_DATA_XFER) {
5654 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5655 		return (EISCONN);
5656 	}
5657 	error = proto_verify_ip_addr(connp->conn_family,
5658 	    (struct sockaddr *)msg->msg_name, msg->msg_namelen);
5659 	if (error != 0) {
5660 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5661 		return (error);
5662 	}
5663 	switch (connp->conn_family) {
5664 	case AF_INET6:
5665 		sin6 = (sin6_t *)msg->msg_name;
5666 
5667 		/* No support for mapped addresses on raw sockets */
5668 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
5669 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5670 			return (EADDRNOTAVAIL);
5671 		}
5672 		srcid = sin6->__sin6_src_id;
5673 
5674 		/*
5675 		 * If the local address is a mapped address return
5676 		 * an error.
5677 		 * It would be possible to send an IPv6 packet but the
5678 		 * response would never make it back to the application
5679 		 * since it is bound to a mapped address.
5680 		 */
5681 		if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
5682 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5683 			return (EADDRNOTAVAIL);
5684 		}
5685 
5686 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
5687 			sin6->sin6_addr = ipv6_loopback;
5688 
5689 		/*
5690 		 * We have to allocate an ip_xmit_attr_t before we grab
5691 		 * conn_lock and we need to hold conn_lock once we've check
5692 		 * conn_same_as_last_v6 to handle concurrent send* calls on a
5693 		 * socket.
5694 		 */
5695 		if (msg->msg_controllen == 0) {
5696 			ixa = conn_get_ixa(connp, B_FALSE);
5697 			if (ixa == NULL) {
5698 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5699 				return (ENOMEM);
5700 			}
5701 		} else {
5702 			ixa = NULL;
5703 		}
5704 		mutex_enter(&connp->conn_lock);
5705 		if (icmp->icmp_delayed_error != 0) {
5706 			sin6_t  *sin2 = (sin6_t *)&icmp->icmp_delayed_addr;
5707 
5708 			error = icmp->icmp_delayed_error;
5709 			icmp->icmp_delayed_error = 0;
5710 
5711 			/* Compare IP address and family */
5712 
5713 			if (IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr,
5714 			    &sin2->sin6_addr) &&
5715 			    sin6->sin6_family == sin2->sin6_family) {
5716 				mutex_exit(&connp->conn_lock);
5717 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5718 				if (ixa != NULL)
5719 					ixa_refrele(ixa);
5720 				return (error);
5721 			}
5722 		}
5723 		if (msg->msg_controllen != 0) {
5724 			mutex_exit(&connp->conn_lock);
5725 			ASSERT(ixa == NULL);
5726 			error = icmp_output_ancillary(connp, NULL, sin6, mp,
5727 			    NULL, msg, cr, pid);
5728 		} else if (conn_same_as_last_v6(connp, sin6) &&
5729 		    connp->conn_lastsrcid == srcid &&
5730 		    ipsec_outbound_policy_current(ixa)) {
5731 			/* icmp_output_lastdst drops conn_lock */
5732 			error = icmp_output_lastdst(connp, mp, cr, pid, ixa);
5733 		} else {
5734 			/* icmp_output_newdst drops conn_lock */
5735 			error = icmp_output_newdst(connp, mp, NULL, sin6, cr,
5736 			    pid, ixa);
5737 		}
5738 		ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
5739 		if (is->is_sendto_ignerr)
5740 			return (0);
5741 		else
5742 			return (error);
5743 	case AF_INET:
5744 		sin = (sin_t *)msg->msg_name;
5745 
5746 		if (sin->sin_addr.s_addr == INADDR_ANY)
5747 			sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
5748 
5749 		/*
5750 		 * We have to allocate an ip_xmit_attr_t before we grab
5751 		 * conn_lock and we need to hold conn_lock once we've check
5752 		 * conn_same_as_last_v6 to handle concurrent send* on a socket.
5753 		 */
5754 		if (msg->msg_controllen == 0) {
5755 			ixa = conn_get_ixa(connp, B_FALSE);
5756 			if (ixa == NULL) {
5757 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5758 				return (ENOMEM);
5759 			}
5760 		} else {
5761 			ixa = NULL;
5762 		}
5763 		mutex_enter(&connp->conn_lock);
5764 		if (icmp->icmp_delayed_error != 0) {
5765 			sin_t  *sin2 = (sin_t *)&icmp->icmp_delayed_addr;
5766 
5767 			error = icmp->icmp_delayed_error;
5768 			icmp->icmp_delayed_error = 0;
5769 
5770 			/* Compare IP address */
5771 
5772 			if (sin->sin_addr.s_addr == sin2->sin_addr.s_addr) {
5773 				mutex_exit(&connp->conn_lock);
5774 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5775 				if (ixa != NULL)
5776 					ixa_refrele(ixa);
5777 				return (error);
5778 			}
5779 		}
5780 
5781 		if (msg->msg_controllen != 0) {
5782 			mutex_exit(&connp->conn_lock);
5783 			ASSERT(ixa == NULL);
5784 			error = icmp_output_ancillary(connp, sin, NULL, mp,
5785 			    NULL, msg, cr, pid);
5786 		} else if (conn_same_as_last_v4(connp, sin) &&
5787 		    ipsec_outbound_policy_current(ixa)) {
5788 			/* icmp_output_lastdst drops conn_lock */
5789 			error = icmp_output_lastdst(connp, mp, cr, pid, ixa);
5790 		} else {
5791 			/* icmp_output_newdst drops conn_lock */
5792 			error = icmp_output_newdst(connp, mp, sin, NULL, cr,
5793 			    pid, ixa);
5794 		}
5795 		ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
5796 		if (is->is_sendto_ignerr)
5797 			return (0);
5798 		else
5799 			return (error);
5800 	default:
5801 		return (EINVAL);
5802 	}
5803 }
5804 
5805 sock_downcalls_t sock_rawip_downcalls = {
5806 	rawip_activate,
5807 	rawip_accept,
5808 	rawip_bind,
5809 	rawip_listen,
5810 	rawip_connect,
5811 	rawip_getpeername,
5812 	rawip_getsockname,
5813 	rawip_getsockopt,
5814 	rawip_setsockopt,
5815 	rawip_send,
5816 	NULL,
5817 	NULL,
5818 	NULL,
5819 	rawip_shutdown,
5820 	rawip_clr_flowctrl,
5821 	rawip_ioctl,
5822 	rawip_close
5823 };
5824