xref: /titanic_51/usr/src/uts/common/inet/ip/icmp.c (revision 8fc99e42676a23421c75e76660640f9765d693b1)
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 int	icmp_param_get(queue_t *q, mblk_t *mp, caddr_t cp, cred_t *cr);
132 static boolean_t icmp_param_register(IDP *ndp, icmpparam_t *icmppa, int cnt);
133 static int	icmp_param_set(queue_t *q, mblk_t *mp, char *value,
134 		    caddr_t cp, cred_t *cr);
135 static mblk_t	*icmp_prepend_hdr(conn_t *, ip_xmit_attr_t *, const ip_pkt_t *,
136     const in6_addr_t *, const in6_addr_t *, uint32_t, mblk_t *, int *);
137 static mblk_t	*icmp_prepend_header_template(conn_t *, ip_xmit_attr_t *,
138     mblk_t *, const in6_addr_t *, uint32_t, int *);
139 static int	icmp_snmp_set(queue_t *q, t_scalar_t level, t_scalar_t name,
140 		    uchar_t *ptr, int len);
141 static void	icmp_ud_err(queue_t *q, mblk_t *mp, t_scalar_t err);
142 static void	icmp_tpi_unbind(queue_t *q, mblk_t *mp);
143 static void	icmp_wput(queue_t *q, mblk_t *mp);
144 static void	icmp_wput_fallback(queue_t *q, mblk_t *mp);
145 static void	icmp_wput_other(queue_t *q, mblk_t *mp);
146 static void	icmp_wput_iocdata(queue_t *q, mblk_t *mp);
147 static void	icmp_wput_restricted(queue_t *q, mblk_t *mp);
148 static void	icmp_ulp_recv(conn_t *, mblk_t *, uint_t);
149 
150 static void	*rawip_stack_init(netstackid_t stackid, netstack_t *ns);
151 static void	rawip_stack_fini(netstackid_t stackid, void *arg);
152 
153 static void	*rawip_kstat_init(netstackid_t stackid);
154 static void	rawip_kstat_fini(netstackid_t stackid, kstat_t *ksp);
155 static int	rawip_kstat_update(kstat_t *kp, int rw);
156 static void	rawip_stack_shutdown(netstackid_t stackid, void *arg);
157 
158 /* Common routines for TPI and socket module */
159 static conn_t	*rawip_do_open(int, cred_t *, int *, int);
160 static void	rawip_do_close(conn_t *);
161 static int	rawip_do_bind(conn_t *, struct sockaddr *, socklen_t);
162 static int	rawip_do_unbind(conn_t *);
163 static int	rawip_do_connect(conn_t *, const struct sockaddr *, socklen_t,
164     cred_t *, pid_t);
165 
166 int		rawip_getsockname(sock_lower_handle_t, struct sockaddr *,
167 		    socklen_t *, cred_t *);
168 int		rawip_getpeername(sock_lower_handle_t, struct sockaddr *,
169 		    socklen_t *, cred_t *);
170 
171 static struct module_info icmp_mod_info =  {
172 	5707, "icmp", 1, INFPSZ, 512, 128
173 };
174 
175 /*
176  * Entry points for ICMP as a device.
177  * We have separate open functions for the /dev/icmp and /dev/icmp6 devices.
178  */
179 static struct qinit icmprinitv4 = {
180 	NULL, NULL, icmp_openv4, icmp_close, NULL, &icmp_mod_info
181 };
182 
183 static struct qinit icmprinitv6 = {
184 	NULL, NULL, icmp_openv6, icmp_close, NULL, &icmp_mod_info
185 };
186 
187 static struct qinit icmpwinit = {
188 	(pfi_t)icmp_wput, (pfi_t)ip_wsrv, NULL, NULL, NULL, &icmp_mod_info
189 };
190 
191 /* ICMP entry point during fallback */
192 static struct qinit icmp_fallback_sock_winit = {
193 	(pfi_t)icmp_wput_fallback, NULL, NULL, NULL, NULL, &icmp_mod_info
194 };
195 
196 /* For AF_INET aka /dev/icmp */
197 struct streamtab icmpinfov4 = {
198 	&icmprinitv4, &icmpwinit
199 };
200 
201 /* For AF_INET6 aka /dev/icmp6 */
202 struct streamtab icmpinfov6 = {
203 	&icmprinitv6, &icmpwinit
204 };
205 
206 static sin_t	sin_null;	/* Zero address for quick clears */
207 static sin6_t	sin6_null;	/* Zero address for quick clears */
208 
209 /* Default structure copied into T_INFO_ACK messages */
210 static struct T_info_ack icmp_g_t_info_ack = {
211 	T_INFO_ACK,
212 	IP_MAXPACKET,	 /* TSDU_size.  icmp allows maximum size messages. */
213 	T_INVALID,	/* ETSDU_size.  icmp does not support expedited data. */
214 	T_INVALID,	/* CDATA_size. icmp does not support connect data. */
215 	T_INVALID,	/* DDATA_size. icmp does not support disconnect data. */
216 	0,		/* ADDR_size - filled in later. */
217 	0,		/* OPT_size - not initialized here */
218 	IP_MAXPACKET,	/* TIDU_size.  icmp allows maximum size messages. */
219 	T_CLTS,		/* SERV_type.  icmp supports connection-less. */
220 	TS_UNBND,	/* CURRENT_state.  This is set from icmp_state. */
221 	(XPG4_1|SENDZERO) /* PROVIDER_flag */
222 };
223 
224 /*
225  * Table of ND variables supported by icmp.  These are loaded into is_nd
226  * when the stack instance is created.
227  * All of these are alterable, within the min/max values given, at run time.
228  */
229 static icmpparam_t	icmp_param_arr[] = {
230 	/* min	max	value	name */
231 	{ 0,	128,	32,	"icmp_wroff_extra" },
232 	{ 1,	255,	255,	"icmp_ipv4_ttl" },
233 	{ 0, IPV6_MAX_HOPS, IPV6_DEFAULT_HOPS,	"icmp_ipv6_hoplimit"},
234 	{ 0,	1,	1,	"icmp_bsd_compat" },
235 	{ 4096,	65536,	8192,	"icmp_xmit_hiwat"},
236 	{ 0,	65536,	1024,	"icmp_xmit_lowat"},
237 	{ 4096,	65536,	8192,	"icmp_recv_hiwat"},
238 	{ 65536, 1024*1024*1024, 256*1024,	"icmp_max_buf"},
239 	{ 0,	1,	0,	"icmp_pmtu_discovery" },
240 	{ 0,	1,	0,	"icmp_sendto_ignerr" },
241 };
242 #define	is_wroff_extra			is_param_arr[0].icmp_param_value
243 #define	is_ipv4_ttl			is_param_arr[1].icmp_param_value
244 #define	is_ipv6_hoplimit		is_param_arr[2].icmp_param_value
245 #define	is_bsd_compat			is_param_arr[3].icmp_param_value
246 #define	is_xmit_hiwat			is_param_arr[4].icmp_param_value
247 #define	is_xmit_lowat			is_param_arr[5].icmp_param_value
248 #define	is_recv_hiwat			is_param_arr[6].icmp_param_value
249 #define	is_max_buf			is_param_arr[7].icmp_param_value
250 #define	is_pmtu_discovery		is_param_arr[8].icmp_param_value
251 #define	is_sendto_ignerr		is_param_arr[9].icmp_param_value
252 
253 typedef union T_primitives *t_primp_t;
254 
255 /*
256  * This routine is called to handle each O_T_BIND_REQ/T_BIND_REQ message
257  * passed to icmp_wput.
258  * It calls IP to verify the local IP address, and calls IP to insert
259  * the conn_t in the fanout table.
260  * If everything is ok it then sends the T_BIND_ACK back up.
261  */
262 static void
263 icmp_tpi_bind(queue_t *q, mblk_t *mp)
264 {
265 	int	error;
266 	struct sockaddr *sa;
267 	struct T_bind_req *tbr;
268 	socklen_t	len;
269 	sin_t	*sin;
270 	sin6_t	*sin6;
271 	icmp_t		*icmp;
272 	conn_t	*connp = Q_TO_CONN(q);
273 	mblk_t *mp1;
274 	cred_t *cr;
275 
276 	/*
277 	 * All Solaris components should pass a db_credp
278 	 * for this TPI message, hence we ASSERT.
279 	 * But in case there is some other M_PROTO that looks
280 	 * like a TPI message sent by some other kernel
281 	 * component, we check and return an error.
282 	 */
283 	cr = msg_getcred(mp, NULL);
284 	ASSERT(cr != NULL);
285 	if (cr == NULL) {
286 		icmp_err_ack(q, mp, TSYSERR, EINVAL);
287 		return;
288 	}
289 
290 	icmp = connp->conn_icmp;
291 	if ((mp->b_wptr - mp->b_rptr) < sizeof (*tbr)) {
292 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
293 		    "icmp_bind: bad req, len %u",
294 		    (uint_t)(mp->b_wptr - mp->b_rptr));
295 		icmp_err_ack(q, mp, TPROTO, 0);
296 		return;
297 	}
298 
299 	if (icmp->icmp_state != TS_UNBND) {
300 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
301 		    "icmp_bind: bad state, %u", icmp->icmp_state);
302 		icmp_err_ack(q, mp, TOUTSTATE, 0);
303 		return;
304 	}
305 
306 	/*
307 	 * Reallocate the message to make sure we have enough room for an
308 	 * address.
309 	 */
310 	mp1 = reallocb(mp, sizeof (struct T_bind_ack) + sizeof (sin6_t), 1);
311 	if (mp1 == NULL) {
312 		icmp_err_ack(q, mp, TSYSERR, ENOMEM);
313 		return;
314 	}
315 	mp = mp1;
316 
317 	/* Reset the message type in preparation for shipping it back. */
318 	DB_TYPE(mp) = M_PCPROTO;
319 	tbr = (struct T_bind_req *)mp->b_rptr;
320 	len = tbr->ADDR_length;
321 	switch (len) {
322 	case 0:	/* request for a generic port */
323 		tbr->ADDR_offset = sizeof (struct T_bind_req);
324 		if (connp->conn_family == AF_INET) {
325 			tbr->ADDR_length = sizeof (sin_t);
326 			sin = (sin_t *)&tbr[1];
327 			*sin = sin_null;
328 			sin->sin_family = AF_INET;
329 			mp->b_wptr = (uchar_t *)&sin[1];
330 			sa = (struct sockaddr *)sin;
331 			len = sizeof (sin_t);
332 		} else {
333 			ASSERT(connp->conn_family == AF_INET6);
334 			tbr->ADDR_length = sizeof (sin6_t);
335 			sin6 = (sin6_t *)&tbr[1];
336 			*sin6 = sin6_null;
337 			sin6->sin6_family = AF_INET6;
338 			mp->b_wptr = (uchar_t *)&sin6[1];
339 			sa = (struct sockaddr *)sin6;
340 			len = sizeof (sin6_t);
341 		}
342 		break;
343 
344 	case sizeof (sin_t):	/* Complete IPv4 address */
345 		sa = (struct sockaddr *)mi_offset_param(mp, tbr->ADDR_offset,
346 		    sizeof (sin_t));
347 		break;
348 
349 	case sizeof (sin6_t):	/* Complete IPv6 address */
350 		sa = (struct sockaddr *)mi_offset_param(mp,
351 		    tbr->ADDR_offset, sizeof (sin6_t));
352 		break;
353 
354 	default:
355 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
356 		    "icmp_bind: bad ADDR_length %u", tbr->ADDR_length);
357 		icmp_err_ack(q, mp, TBADADDR, 0);
358 		return;
359 	}
360 
361 	error = rawip_do_bind(connp, sa, len);
362 	if (error != 0) {
363 		if (error > 0) {
364 			icmp_err_ack(q, mp, TSYSERR, error);
365 		} else {
366 			icmp_err_ack(q, mp, -error, 0);
367 		}
368 	} else {
369 		tbr->PRIM_type = T_BIND_ACK;
370 		qreply(q, mp);
371 	}
372 }
373 
374 static int
375 rawip_do_bind(conn_t *connp, struct sockaddr *sa, socklen_t len)
376 {
377 	sin_t		*sin;
378 	sin6_t		*sin6;
379 	icmp_t		*icmp = connp->conn_icmp;
380 	int		error = 0;
381 	ip_laddr_t	laddr_type = IPVL_UNICAST_UP;	/* INADDR_ANY */
382 	in_port_t	lport;		/* Network byte order */
383 	ipaddr_t	v4src;		/* Set if AF_INET */
384 	in6_addr_t	v6src;
385 	uint_t		scopeid = 0;
386 	zoneid_t	zoneid = IPCL_ZONEID(connp);
387 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
388 
389 	if (sa == NULL || !OK_32PTR((char *)sa)) {
390 		return (EINVAL);
391 	}
392 
393 	switch (len) {
394 	case sizeof (sin_t):    /* Complete IPv4 address */
395 		sin = (sin_t *)sa;
396 		if (sin->sin_family != AF_INET ||
397 		    connp->conn_family != AF_INET) {
398 			/* TSYSERR, EAFNOSUPPORT */
399 			return (EAFNOSUPPORT);
400 		}
401 		v4src = sin->sin_addr.s_addr;
402 		IN6_IPADDR_TO_V4MAPPED(v4src, &v6src);
403 		if (v4src != INADDR_ANY) {
404 			laddr_type = ip_laddr_verify_v4(v4src, zoneid, ipst,
405 			    B_TRUE);
406 		}
407 		lport = sin->sin_port;
408 		break;
409 	case sizeof (sin6_t): /* Complete IPv6 address */
410 		sin6 = (sin6_t *)sa;
411 		if (sin6->sin6_family != AF_INET6 ||
412 		    connp->conn_family != AF_INET6) {
413 			/* TSYSERR, EAFNOSUPPORT */
414 			return (EAFNOSUPPORT);
415 		}
416 		/* No support for mapped addresses on raw sockets */
417 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
418 			/* TSYSERR, EADDRNOTAVAIL */
419 			return (EADDRNOTAVAIL);
420 		}
421 		v6src = sin6->sin6_addr;
422 		if (!IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
423 			if (IN6_IS_ADDR_LINKSCOPE(&v6src))
424 				scopeid = sin6->sin6_scope_id;
425 			laddr_type = ip_laddr_verify_v6(&v6src, zoneid, ipst,
426 			    B_TRUE, scopeid);
427 		}
428 		lport = sin6->sin6_port;
429 		break;
430 
431 	default:
432 		/* TBADADDR */
433 		return (EADDRNOTAVAIL);
434 	}
435 
436 	/* Is the local address a valid unicast, multicast, or broadcast? */
437 	if (laddr_type == IPVL_BAD)
438 		return (EADDRNOTAVAIL);
439 
440 	/*
441 	 * The state must be TS_UNBND.
442 	 */
443 	mutex_enter(&connp->conn_lock);
444 	if (icmp->icmp_state != TS_UNBND) {
445 		mutex_exit(&connp->conn_lock);
446 		return (-TOUTSTATE);
447 	}
448 
449 	/*
450 	 * Copy the source address into our icmp structure.  This address
451 	 * may still be zero; if so, ip will fill in the correct address
452 	 * each time an outbound packet is passed to it.
453 	 * If we are binding to a broadcast or multicast address then
454 	 * we just set the conn_bound_addr since we don't want to use
455 	 * that as the source address when sending.
456 	 */
457 	connp->conn_bound_addr_v6 = v6src;
458 	connp->conn_laddr_v6 = v6src;
459 	if (scopeid != 0) {
460 		connp->conn_ixa->ixa_flags |= IXAF_SCOPEID_SET;
461 		connp->conn_ixa->ixa_scopeid = scopeid;
462 		connp->conn_incoming_ifindex = scopeid;
463 	} else {
464 		connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
465 		connp->conn_incoming_ifindex = connp->conn_bound_if;
466 	}
467 
468 	switch (laddr_type) {
469 	case IPVL_UNICAST_UP:
470 	case IPVL_UNICAST_DOWN:
471 		connp->conn_saddr_v6 = v6src;
472 		connp->conn_mcbc_bind = B_FALSE;
473 		break;
474 	case IPVL_MCAST:
475 	case IPVL_BCAST:
476 		/* ip_set_destination will pick a source address later */
477 		connp->conn_saddr_v6 = ipv6_all_zeros;
478 		connp->conn_mcbc_bind = B_TRUE;
479 		break;
480 	}
481 
482 	/* Any errors after this point should use late_error */
483 
484 	/*
485 	 * Use sin_port/sin6_port since applications like psh use SOCK_RAW
486 	 * with IPPROTO_TCP.
487 	 */
488 	connp->conn_lport = lport;
489 	connp->conn_fport = 0;
490 
491 	if (connp->conn_family == AF_INET) {
492 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
493 	} else {
494 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
495 	}
496 
497 	icmp->icmp_state = TS_IDLE;
498 
499 	/*
500 	 * We create an initial header template here to make a subsequent
501 	 * sendto have a starting point. Since conn_last_dst is zero the
502 	 * first sendto will always follow the 'dst changed' code path.
503 	 * Note that we defer massaging options and the related checksum
504 	 * adjustment until we have a destination address.
505 	 */
506 	error = icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
507 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
508 	if (error != 0) {
509 		mutex_exit(&connp->conn_lock);
510 		goto late_error;
511 	}
512 	/* Just in case */
513 	connp->conn_faddr_v6 = ipv6_all_zeros;
514 	connp->conn_v6lastdst = ipv6_all_zeros;
515 	mutex_exit(&connp->conn_lock);
516 
517 	error = ip_laddr_fanout_insert(connp);
518 	if (error != 0)
519 		goto late_error;
520 
521 	/* Bind succeeded */
522 	return (0);
523 
524 late_error:
525 	mutex_enter(&connp->conn_lock);
526 	connp->conn_saddr_v6 = ipv6_all_zeros;
527 	connp->conn_bound_addr_v6 = ipv6_all_zeros;
528 	connp->conn_laddr_v6 = ipv6_all_zeros;
529 	if (scopeid != 0) {
530 		connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
531 		connp->conn_incoming_ifindex = connp->conn_bound_if;
532 	}
533 	icmp->icmp_state = TS_UNBND;
534 	connp->conn_v6lastdst = ipv6_all_zeros;
535 	connp->conn_lport = 0;
536 
537 	/* Restore the header that was built above - different source address */
538 	(void) icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
539 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
540 	mutex_exit(&connp->conn_lock);
541 	return (error);
542 }
543 
544 /*
545  * Tell IP to just bind to the protocol.
546  */
547 static void
548 icmp_bind_proto(icmp_t *icmp)
549 {
550 	conn_t	*connp = icmp->icmp_connp;
551 
552 	mutex_enter(&connp->conn_lock);
553 	connp->conn_saddr_v6 = ipv6_all_zeros;
554 	connp->conn_laddr_v6 = ipv6_all_zeros;
555 	connp->conn_faddr_v6 = ipv6_all_zeros;
556 	connp->conn_v6lastdst = ipv6_all_zeros;
557 	mutex_exit(&connp->conn_lock);
558 
559 	(void) ip_laddr_fanout_insert(connp);
560 }
561 
562 /*
563  * This routine handles each T_CONN_REQ message passed to icmp.  It
564  * associates a default destination address with the stream.
565  *
566  * After various error checks are completed, icmp_connect() lays
567  * the target address and port into the composite header template.
568  * Then we ask IP for information, including a source address if we didn't
569  * already have one. Finally we send up the T_OK_ACK reply message.
570  */
571 static void
572 icmp_tpi_connect(queue_t *q, mblk_t *mp)
573 {
574 	conn_t	*connp = Q_TO_CONN(q);
575 	struct T_conn_req	*tcr;
576 	struct sockaddr *sa;
577 	socklen_t len;
578 	int error;
579 	cred_t *cr;
580 	pid_t pid;
581 	/*
582 	 * All Solaris components should pass a db_credp
583 	 * for this TPI message, hence we ASSERT.
584 	 * But in case there is some other M_PROTO that looks
585 	 * like a TPI message sent by some other kernel
586 	 * component, we check and return an error.
587 	 */
588 	cr = msg_getcred(mp, &pid);
589 	ASSERT(cr != NULL);
590 	if (cr == NULL) {
591 		icmp_err_ack(q, mp, TSYSERR, EINVAL);
592 		return;
593 	}
594 
595 	tcr = (struct T_conn_req *)mp->b_rptr;
596 	/* Sanity checks */
597 	if ((mp->b_wptr - mp->b_rptr) < sizeof (struct T_conn_req)) {
598 		icmp_err_ack(q, mp, TPROTO, 0);
599 		return;
600 	}
601 
602 	if (tcr->OPT_length != 0) {
603 		icmp_err_ack(q, mp, TBADOPT, 0);
604 		return;
605 	}
606 
607 	len = tcr->DEST_length;
608 
609 	switch (len) {
610 	default:
611 		icmp_err_ack(q, mp, TBADADDR, 0);
612 		return;
613 	case sizeof (sin_t):
614 		sa = (struct sockaddr *)mi_offset_param(mp, tcr->DEST_offset,
615 		    sizeof (sin_t));
616 		break;
617 	case sizeof (sin6_t):
618 		sa = (struct sockaddr *)mi_offset_param(mp,
619 		    tcr->DEST_offset, sizeof (sin6_t));
620 		break;
621 	}
622 
623 	error = proto_verify_ip_addr(connp->conn_family, sa, len);
624 	if (error != 0) {
625 		icmp_err_ack(q, mp, TSYSERR, error);
626 		return;
627 	}
628 
629 	error = rawip_do_connect(connp, sa, len, cr, pid);
630 	if (error != 0) {
631 		if (error < 0) {
632 			icmp_err_ack(q, mp, -error, 0);
633 		} else {
634 			icmp_err_ack(q, mp, 0, error);
635 		}
636 	} else {
637 		mblk_t *mp1;
638 
639 		/*
640 		 * We have to send a connection confirmation to
641 		 * keep TLI happy.
642 		 */
643 		if (connp->conn_family == AF_INET) {
644 			mp1 = mi_tpi_conn_con(NULL, (char *)sa,
645 			    sizeof (sin_t), NULL, 0);
646 		} else {
647 			ASSERT(connp->conn_family == AF_INET6);
648 			mp1 = mi_tpi_conn_con(NULL, (char *)sa,
649 			    sizeof (sin6_t), NULL, 0);
650 		}
651 		if (mp1 == NULL) {
652 			icmp_err_ack(q, mp, TSYSERR, ENOMEM);
653 			return;
654 		}
655 
656 		/*
657 		 * Send ok_ack for T_CONN_REQ
658 		 */
659 		mp = mi_tpi_ok_ack_alloc(mp);
660 		if (mp == NULL) {
661 			/* Unable to reuse the T_CONN_REQ for the ack. */
662 			icmp_err_ack_prim(q, mp1, T_CONN_REQ, TSYSERR, ENOMEM);
663 			return;
664 		}
665 		putnext(connp->conn_rq, mp);
666 		putnext(connp->conn_rq, mp1);
667 	}
668 }
669 
670 static int
671 rawip_do_connect(conn_t *connp, const struct sockaddr *sa, socklen_t len,
672     cred_t *cr, pid_t pid)
673 {
674 	icmp_t		*icmp;
675 	sin_t		*sin;
676 	sin6_t		*sin6;
677 	int		error;
678 	uint16_t 	dstport;
679 	ipaddr_t	v4dst;
680 	in6_addr_t	v6dst;
681 	uint32_t	flowinfo;
682 	ip_xmit_attr_t	*ixa;
683 	uint_t		scopeid = 0;
684 	uint_t		srcid = 0;
685 	in6_addr_t	v6src = connp->conn_saddr_v6;
686 
687 	icmp = connp->conn_icmp;
688 
689 	if (sa == NULL || !OK_32PTR((char *)sa)) {
690 		return (EINVAL);
691 	}
692 
693 	ASSERT(sa != NULL && len != 0);
694 
695 	/*
696 	 * Determine packet type based on type of address passed in
697 	 * the request should contain an IPv4 or IPv6 address.
698 	 * Make sure that address family matches the type of
699 	 * family of the address passed down.
700 	 */
701 	switch (len) {
702 	case sizeof (sin_t):
703 		sin = (sin_t *)sa;
704 
705 		v4dst = sin->sin_addr.s_addr;
706 		dstport = sin->sin_port;
707 		IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
708 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
709 		break;
710 
711 	case sizeof (sin6_t):
712 		sin6 = (sin6_t *)sa;
713 
714 		/* No support for mapped addresses on raw sockets */
715 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
716 			return (EADDRNOTAVAIL);
717 		}
718 		v6dst = sin6->sin6_addr;
719 		dstport = sin6->sin6_port;
720 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
721 		flowinfo = sin6->sin6_flowinfo;
722 		if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr))
723 			scopeid = sin6->sin6_scope_id;
724 		srcid = sin6->__sin6_src_id;
725 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
726 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
727 			    connp->conn_netstack);
728 		}
729 		break;
730 	}
731 
732 	/*
733 	 * If there is a different thread using conn_ixa then we get a new
734 	 * copy and cut the old one loose from conn_ixa. Otherwise we use
735 	 * conn_ixa and prevent any other thread from using/changing it.
736 	 * Once connect() is done other threads can use conn_ixa since the
737 	 * refcnt will be back at one.
738 	 */
739 	ixa = conn_get_ixa(connp, B_TRUE);
740 	if (ixa == NULL)
741 		return (ENOMEM);
742 
743 	ASSERT(ixa->ixa_refcnt >= 2);
744 	ASSERT(ixa == connp->conn_ixa);
745 
746 	mutex_enter(&connp->conn_lock);
747 	/*
748 	 * This icmp_t must have bound already before doing a connect.
749 	 * Reject if a connect is in progress (we drop conn_lock during
750 	 * rawip_do_connect).
751 	 */
752 	if (icmp->icmp_state == TS_UNBND || icmp->icmp_state == TS_WCON_CREQ) {
753 		mutex_exit(&connp->conn_lock);
754 		ixa_refrele(ixa);
755 		return (-TOUTSTATE);
756 	}
757 
758 	if (icmp->icmp_state == TS_DATA_XFER) {
759 		/* Already connected - clear out state */
760 		if (connp->conn_mcbc_bind)
761 			connp->conn_saddr_v6 = ipv6_all_zeros;
762 		else
763 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
764 		connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
765 		connp->conn_faddr_v6 = ipv6_all_zeros;
766 		icmp->icmp_state = TS_IDLE;
767 	}
768 
769 	/*
770 	 * Use sin_port/sin6_port since applications like psh use SOCK_RAW
771 	 * with IPPROTO_TCP.
772 	 */
773 	connp->conn_fport = dstport;
774 	if (connp->conn_ipversion == IPV4_VERSION) {
775 		/*
776 		 * Interpret a zero destination to mean loopback.
777 		 * Update the T_CONN_REQ (sin/sin6) since it is used to
778 		 * generate the T_CONN_CON.
779 		 */
780 		if (v4dst == INADDR_ANY) {
781 			v4dst = htonl(INADDR_LOOPBACK);
782 			IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
783 			ASSERT(connp->conn_family == AF_INET);
784 			sin->sin_addr.s_addr = v4dst;
785 		}
786 		connp->conn_faddr_v6 = v6dst;
787 		connp->conn_flowinfo = 0;
788 	} else {
789 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
790 		/*
791 		 * Interpret a zero destination to mean loopback.
792 		 * Update the T_CONN_REQ (sin/sin6) since it is used to
793 		 * generate the T_CONN_CON.
794 		 */
795 		if (IN6_IS_ADDR_UNSPECIFIED(&v6dst)) {
796 			v6dst = ipv6_loopback;
797 			sin6->sin6_addr = v6dst;
798 		}
799 		connp->conn_faddr_v6 = v6dst;
800 		connp->conn_flowinfo = flowinfo;
801 	}
802 
803 	/*
804 	 * We update our cred/cpid based on the caller of connect
805 	 */
806 	if (connp->conn_cred != cr) {
807 		crhold(cr);
808 		crfree(connp->conn_cred);
809 		connp->conn_cred = cr;
810 	}
811 	connp->conn_cpid = pid;
812 	ixa->ixa_cred = cr;
813 	ixa->ixa_cpid = pid;
814 	if (is_system_labeled()) {
815 		/* We need to restart with a label based on the cred */
816 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
817 	}
818 
819 	if (scopeid != 0) {
820 		ixa->ixa_flags |= IXAF_SCOPEID_SET;
821 		ixa->ixa_scopeid = scopeid;
822 		connp->conn_incoming_ifindex = scopeid;
823 	} else {
824 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
825 		connp->conn_incoming_ifindex = connp->conn_bound_if;
826 	}
827 
828 	/*
829 	 * conn_connect will drop conn_lock and reacquire it.
830 	 * To prevent a send* from messing with this icmp_t while the lock
831 	 * is dropped we set icmp_state and clear conn_v6lastdst.
832 	 * That will make all send* fail with EISCONN.
833 	 */
834 	connp->conn_v6lastdst = ipv6_all_zeros;
835 	icmp->icmp_state = TS_WCON_CREQ;
836 
837 	error = conn_connect(connp, NULL, IPDF_ALLOW_MCBC);
838 	mutex_exit(&connp->conn_lock);
839 	if (error != 0)
840 		goto connect_failed;
841 
842 	/*
843 	 * The addresses have been verified. Time to insert in
844 	 * the correct fanout list.
845 	 */
846 	error = ipcl_conn_insert(connp);
847 	if (error != 0)
848 		goto connect_failed;
849 
850 	mutex_enter(&connp->conn_lock);
851 	error = icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
852 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
853 	if (error != 0) {
854 		mutex_exit(&connp->conn_lock);
855 		goto connect_failed;
856 	}
857 
858 	icmp->icmp_state = TS_DATA_XFER;
859 	/* Record this as the "last" send even though we haven't sent any */
860 	connp->conn_v6lastdst = connp->conn_faddr_v6;
861 	connp->conn_lastipversion = connp->conn_ipversion;
862 	connp->conn_lastdstport = connp->conn_fport;
863 	connp->conn_lastflowinfo = connp->conn_flowinfo;
864 	connp->conn_lastscopeid = scopeid;
865 	connp->conn_lastsrcid = srcid;
866 	/* Also remember a source to use together with lastdst */
867 	connp->conn_v6lastsrc = v6src;
868 	mutex_exit(&connp->conn_lock);
869 
870 	ixa_refrele(ixa);
871 	return (0);
872 
873 connect_failed:
874 	if (ixa != NULL)
875 		ixa_refrele(ixa);
876 	mutex_enter(&connp->conn_lock);
877 	icmp->icmp_state = TS_IDLE;
878 	/* In case the source address was set above */
879 	if (connp->conn_mcbc_bind)
880 		connp->conn_saddr_v6 = ipv6_all_zeros;
881 	else
882 		connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
883 	connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
884 	connp->conn_faddr_v6 = ipv6_all_zeros;
885 	connp->conn_v6lastdst = ipv6_all_zeros;
886 	connp->conn_flowinfo = 0;
887 
888 	(void) icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
889 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
890 	mutex_exit(&connp->conn_lock);
891 	return (error);
892 }
893 
894 static void
895 rawip_do_close(conn_t *connp)
896 {
897 	ASSERT(connp != NULL && IPCL_IS_RAWIP(connp));
898 
899 	ip_quiesce_conn(connp);
900 
901 	if (!IPCL_IS_NONSTR(connp)) {
902 		qprocsoff(connp->conn_rq);
903 	}
904 
905 	icmp_close_free(connp);
906 
907 	/*
908 	 * Now we are truly single threaded on this stream, and can
909 	 * delete the things hanging off the connp, and finally the connp.
910 	 * We removed this connp from the fanout list, it cannot be
911 	 * accessed thru the fanouts, and we already waited for the
912 	 * conn_ref to drop to 0. We are already in close, so
913 	 * there cannot be any other thread from the top. qprocsoff
914 	 * has completed, and service has completed or won't run in
915 	 * future.
916 	 */
917 	ASSERT(connp->conn_ref == 1);
918 
919 	if (!IPCL_IS_NONSTR(connp)) {
920 		inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
921 	} else {
922 		ip_free_helper_stream(connp);
923 	}
924 
925 	connp->conn_ref--;
926 	ipcl_conn_destroy(connp);
927 }
928 
929 static int
930 icmp_close(queue_t *q, int flags)
931 {
932 	conn_t  *connp;
933 
934 	if (flags & SO_FALLBACK) {
935 		/*
936 		 * stream is being closed while in fallback
937 		 * simply free the resources that were allocated
938 		 */
939 		inet_minor_free(WR(q)->q_ptr, (dev_t)(RD(q)->q_ptr));
940 		qprocsoff(q);
941 		goto done;
942 	}
943 
944 	connp = Q_TO_CONN(q);
945 	(void) rawip_do_close(connp);
946 done:
947 	q->q_ptr = WR(q)->q_ptr = NULL;
948 	return (0);
949 }
950 
951 static void
952 icmp_close_free(conn_t *connp)
953 {
954 	icmp_t *icmp = connp->conn_icmp;
955 
956 	if (icmp->icmp_filter != NULL) {
957 		kmem_free(icmp->icmp_filter, sizeof (icmp6_filter_t));
958 		icmp->icmp_filter = NULL;
959 	}
960 
961 	/*
962 	 * Clear any fields which the kmem_cache constructor clears.
963 	 * Only icmp_connp needs to be preserved.
964 	 * TBD: We should make this more efficient to avoid clearing
965 	 * everything.
966 	 */
967 	ASSERT(icmp->icmp_connp == connp);
968 	bzero(icmp, sizeof (icmp_t));
969 	icmp->icmp_connp = connp;
970 }
971 
972 /*
973  * This routine handles each T_DISCON_REQ message passed to icmp
974  * as an indicating that ICMP is no longer connected. This results
975  * in telling IP to restore the binding to just the local address.
976  */
977 static int
978 icmp_do_disconnect(conn_t *connp)
979 {
980 	icmp_t	*icmp = connp->conn_icmp;
981 	int	error;
982 
983 	mutex_enter(&connp->conn_lock);
984 	if (icmp->icmp_state != TS_DATA_XFER) {
985 		mutex_exit(&connp->conn_lock);
986 		return (-TOUTSTATE);
987 	}
988 	if (connp->conn_mcbc_bind)
989 		connp->conn_saddr_v6 = ipv6_all_zeros;
990 	else
991 		connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
992 	connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
993 	connp->conn_faddr_v6 = ipv6_all_zeros;
994 	icmp->icmp_state = TS_IDLE;
995 
996 	connp->conn_v6lastdst = ipv6_all_zeros;
997 	error = icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
998 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
999 	mutex_exit(&connp->conn_lock);
1000 	if (error != 0)
1001 		return (error);
1002 
1003 	/*
1004 	 * Tell IP to remove the full binding and revert
1005 	 * to the local address binding.
1006 	 */
1007 	return (ip_laddr_fanout_insert(connp));
1008 }
1009 
1010 static void
1011 icmp_tpi_disconnect(queue_t *q, mblk_t *mp)
1012 {
1013 	conn_t	*connp = Q_TO_CONN(q);
1014 	int	error;
1015 
1016 	/*
1017 	 * Allocate the largest primitive we need to send back
1018 	 * T_error_ack is > than T_ok_ack
1019 	 */
1020 	mp = reallocb(mp, sizeof (struct T_error_ack), 1);
1021 	if (mp == NULL) {
1022 		/* Unable to reuse the T_DISCON_REQ for the ack. */
1023 		icmp_err_ack_prim(q, mp, T_DISCON_REQ, TSYSERR, ENOMEM);
1024 		return;
1025 	}
1026 
1027 	error = icmp_do_disconnect(connp);
1028 
1029 	if (error != 0) {
1030 		if (error > 0) {
1031 			icmp_err_ack(q, mp, 0, error);
1032 		} else {
1033 			icmp_err_ack(q, mp, -error, 0);
1034 		}
1035 	} else {
1036 		mp = mi_tpi_ok_ack_alloc(mp);
1037 		ASSERT(mp != NULL);
1038 		qreply(q, mp);
1039 	}
1040 }
1041 
1042 static int
1043 icmp_disconnect(conn_t *connp)
1044 {
1045 	int	error;
1046 
1047 	connp->conn_dgram_errind = B_FALSE;
1048 
1049 	error = icmp_do_disconnect(connp);
1050 
1051 	if (error < 0)
1052 		error = proto_tlitosyserr(-error);
1053 	return (error);
1054 }
1055 
1056 /* This routine creates a T_ERROR_ACK message and passes it upstream. */
1057 static void
1058 icmp_err_ack(queue_t *q, mblk_t *mp, t_scalar_t t_error, int sys_error)
1059 {
1060 	if ((mp = mi_tpi_err_ack_alloc(mp, t_error, sys_error)) != NULL)
1061 		qreply(q, mp);
1062 }
1063 
1064 /* Shorthand to generate and send TPI error acks to our client */
1065 static void
1066 icmp_err_ack_prim(queue_t *q, mblk_t *mp, t_scalar_t primitive,
1067     t_scalar_t t_error, int sys_error)
1068 {
1069 	struct T_error_ack	*teackp;
1070 
1071 	if ((mp = tpi_ack_alloc(mp, sizeof (struct T_error_ack),
1072 	    M_PCPROTO, T_ERROR_ACK)) != NULL) {
1073 		teackp = (struct T_error_ack *)mp->b_rptr;
1074 		teackp->ERROR_prim = primitive;
1075 		teackp->TLI_error = t_error;
1076 		teackp->UNIX_error = sys_error;
1077 		qreply(q, mp);
1078 	}
1079 }
1080 
1081 /*
1082  * icmp_icmp_input is called as conn_recvicmp to process ICMP messages.
1083  * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
1084  * Assumes that IP has pulled up everything up to and including the ICMP header.
1085  */
1086 /* ARGSUSED2 */
1087 static void
1088 icmp_icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
1089 {
1090 	conn_t		*connp = (conn_t *)arg1;
1091 	icmp_t		*icmp = connp->conn_icmp;
1092 	icmph_t		*icmph;
1093 	ipha_t		*ipha;
1094 	int		iph_hdr_length;
1095 	sin_t		sin;
1096 	mblk_t		*mp1;
1097 	int		error = 0;
1098 
1099 	ipha = (ipha_t *)mp->b_rptr;
1100 
1101 	ASSERT(OK_32PTR(mp->b_rptr));
1102 
1103 	if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
1104 		ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
1105 		icmp_icmp_error_ipv6(connp, mp, ira);
1106 		return;
1107 	}
1108 	ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
1109 
1110 	/* Skip past the outer IP and ICMP headers */
1111 	ASSERT(IPH_HDR_LENGTH(ipha) == ira->ira_ip_hdr_length);
1112 	iph_hdr_length = ira->ira_ip_hdr_length;
1113 	icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1114 	ipha = (ipha_t *)&icmph[1];	/* Inner IP header */
1115 
1116 	iph_hdr_length = IPH_HDR_LENGTH(ipha);
1117 
1118 	switch (icmph->icmph_type) {
1119 	case ICMP_DEST_UNREACHABLE:
1120 		switch (icmph->icmph_code) {
1121 		case ICMP_FRAGMENTATION_NEEDED: {
1122 			ipha_t		*ipha;
1123 			ip_xmit_attr_t	*ixa;
1124 			/*
1125 			 * IP has already adjusted the path MTU.
1126 			 * But we need to adjust DF for IPv4.
1127 			 */
1128 			if (connp->conn_ipversion != IPV4_VERSION)
1129 				break;
1130 
1131 			ixa = conn_get_ixa(connp, B_FALSE);
1132 			if (ixa == NULL || ixa->ixa_ire == NULL) {
1133 				/*
1134 				 * Some other thread holds conn_ixa. We will
1135 				 * redo this on the next ICMP too big.
1136 				 */
1137 				if (ixa != NULL)
1138 					ixa_refrele(ixa);
1139 				break;
1140 			}
1141 			(void) ip_get_pmtu(ixa);
1142 
1143 			mutex_enter(&connp->conn_lock);
1144 			ipha = (ipha_t *)connp->conn_ht_iphc;
1145 			if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
1146 				ipha->ipha_fragment_offset_and_flags |=
1147 				    IPH_DF_HTONS;
1148 			} else {
1149 				ipha->ipha_fragment_offset_and_flags &=
1150 				    ~IPH_DF_HTONS;
1151 			}
1152 			mutex_exit(&connp->conn_lock);
1153 			ixa_refrele(ixa);
1154 			break;
1155 		}
1156 		case ICMP_PORT_UNREACHABLE:
1157 		case ICMP_PROTOCOL_UNREACHABLE:
1158 			error = ECONNREFUSED;
1159 			break;
1160 		default:
1161 			/* Transient errors */
1162 			break;
1163 		}
1164 		break;
1165 	default:
1166 		/* Transient errors */
1167 		break;
1168 	}
1169 	if (error == 0) {
1170 		freemsg(mp);
1171 		return;
1172 	}
1173 
1174 	/*
1175 	 * Deliver T_UDERROR_IND when the application has asked for it.
1176 	 * The socket layer enables this automatically when connected.
1177 	 */
1178 	if (!connp->conn_dgram_errind) {
1179 		freemsg(mp);
1180 		return;
1181 	}
1182 
1183 	sin = sin_null;
1184 	sin.sin_family = AF_INET;
1185 	sin.sin_addr.s_addr = ipha->ipha_dst;
1186 
1187 	if (IPCL_IS_NONSTR(connp)) {
1188 		mutex_enter(&connp->conn_lock);
1189 		if (icmp->icmp_state == TS_DATA_XFER) {
1190 			if (sin.sin_addr.s_addr == connp->conn_faddr_v4) {
1191 				mutex_exit(&connp->conn_lock);
1192 				(*connp->conn_upcalls->su_set_error)
1193 				    (connp->conn_upper_handle, error);
1194 				goto done;
1195 			}
1196 		} else {
1197 			icmp->icmp_delayed_error = error;
1198 			*((sin_t *)&icmp->icmp_delayed_addr) = sin;
1199 		}
1200 		mutex_exit(&connp->conn_lock);
1201 	} else {
1202 		mp1 = mi_tpi_uderror_ind((char *)&sin, sizeof (sin_t), NULL, 0,
1203 		    error);
1204 		if (mp1 != NULL)
1205 			putnext(connp->conn_rq, mp1);
1206 	}
1207 done:
1208 	freemsg(mp);
1209 }
1210 
1211 /*
1212  * icmp_icmp_error_ipv6 is called by icmp_icmp_error to process ICMP for IPv6.
1213  * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
1214  * Assumes that IP has pulled up all the extension headers as well as the
1215  * ICMPv6 header.
1216  */
1217 static void
1218 icmp_icmp_error_ipv6(conn_t *connp, mblk_t *mp, ip_recv_attr_t *ira)
1219 {
1220 	icmp6_t		*icmp6;
1221 	ip6_t		*ip6h, *outer_ip6h;
1222 	uint16_t	iph_hdr_length;
1223 	uint8_t		*nexthdrp;
1224 	sin6_t		sin6;
1225 	mblk_t		*mp1;
1226 	int		error = 0;
1227 	icmp_t		*icmp = connp->conn_icmp;
1228 
1229 	outer_ip6h = (ip6_t *)mp->b_rptr;
1230 #ifdef DEBUG
1231 	if (outer_ip6h->ip6_nxt != IPPROTO_ICMPV6)
1232 		iph_hdr_length = ip_hdr_length_v6(mp, outer_ip6h);
1233 	else
1234 		iph_hdr_length = IPV6_HDR_LEN;
1235 	ASSERT(iph_hdr_length == ira->ira_ip_hdr_length);
1236 #endif
1237 	/* Skip past the outer IP and ICMP headers */
1238 	iph_hdr_length = ira->ira_ip_hdr_length;
1239 	icmp6 = (icmp6_t *)&mp->b_rptr[iph_hdr_length];
1240 
1241 	ip6h = (ip6_t *)&icmp6[1];	/* Inner IP header */
1242 	if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length, &nexthdrp)) {
1243 		freemsg(mp);
1244 		return;
1245 	}
1246 
1247 	switch (icmp6->icmp6_type) {
1248 	case ICMP6_DST_UNREACH:
1249 		switch (icmp6->icmp6_code) {
1250 		case ICMP6_DST_UNREACH_NOPORT:
1251 			error = ECONNREFUSED;
1252 			break;
1253 		case ICMP6_DST_UNREACH_ADMIN:
1254 		case ICMP6_DST_UNREACH_NOROUTE:
1255 		case ICMP6_DST_UNREACH_BEYONDSCOPE:
1256 		case ICMP6_DST_UNREACH_ADDR:
1257 			/* Transient errors */
1258 			break;
1259 		default:
1260 			break;
1261 		}
1262 		break;
1263 	case ICMP6_PACKET_TOO_BIG: {
1264 		struct T_unitdata_ind	*tudi;
1265 		struct T_opthdr		*toh;
1266 		size_t			udi_size;
1267 		mblk_t			*newmp;
1268 		t_scalar_t		opt_length = sizeof (struct T_opthdr) +
1269 		    sizeof (struct ip6_mtuinfo);
1270 		sin6_t			*sin6;
1271 		struct ip6_mtuinfo	*mtuinfo;
1272 
1273 		/*
1274 		 * If the application has requested to receive path mtu
1275 		 * information, send up an empty message containing an
1276 		 * IPV6_PATHMTU ancillary data item.
1277 		 */
1278 		if (!connp->conn_ipv6_recvpathmtu)
1279 			break;
1280 
1281 		udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t) +
1282 		    opt_length;
1283 		if ((newmp = allocb(udi_size, BPRI_MED)) == NULL) {
1284 			BUMP_MIB(&icmp->icmp_is->is_rawip_mib, rawipInErrors);
1285 			break;
1286 		}
1287 
1288 		/*
1289 		 * newmp->b_cont is left to NULL on purpose.  This is an
1290 		 * empty message containing only ancillary data.
1291 		 */
1292 		newmp->b_datap->db_type = M_PROTO;
1293 		tudi = (struct T_unitdata_ind *)newmp->b_rptr;
1294 		newmp->b_wptr = (uchar_t *)tudi + udi_size;
1295 		tudi->PRIM_type = T_UNITDATA_IND;
1296 		tudi->SRC_length = sizeof (sin6_t);
1297 		tudi->SRC_offset = sizeof (struct T_unitdata_ind);
1298 		tudi->OPT_offset = tudi->SRC_offset + sizeof (sin6_t);
1299 		tudi->OPT_length = opt_length;
1300 
1301 		sin6 = (sin6_t *)&tudi[1];
1302 		bzero(sin6, sizeof (sin6_t));
1303 		sin6->sin6_family = AF_INET6;
1304 		sin6->sin6_addr = connp->conn_faddr_v6;
1305 
1306 		toh = (struct T_opthdr *)&sin6[1];
1307 		toh->level = IPPROTO_IPV6;
1308 		toh->name = IPV6_PATHMTU;
1309 		toh->len = opt_length;
1310 		toh->status = 0;
1311 
1312 		mtuinfo = (struct ip6_mtuinfo *)&toh[1];
1313 		bzero(mtuinfo, sizeof (struct ip6_mtuinfo));
1314 		mtuinfo->ip6m_addr.sin6_family = AF_INET6;
1315 		mtuinfo->ip6m_addr.sin6_addr = ip6h->ip6_dst;
1316 		mtuinfo->ip6m_mtu = icmp6->icmp6_mtu;
1317 		/*
1318 		 * We've consumed everything we need from the original
1319 		 * message.  Free it, then send our empty message.
1320 		 */
1321 		freemsg(mp);
1322 		icmp_ulp_recv(connp, newmp, msgdsize(newmp));
1323 		return;
1324 	}
1325 	case ICMP6_TIME_EXCEEDED:
1326 		/* Transient errors */
1327 		break;
1328 	case ICMP6_PARAM_PROB:
1329 		/* If this corresponds to an ICMP_PROTOCOL_UNREACHABLE */
1330 		if (icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER &&
1331 		    (uchar_t *)ip6h + icmp6->icmp6_pptr ==
1332 		    (uchar_t *)nexthdrp) {
1333 			error = ECONNREFUSED;
1334 			break;
1335 		}
1336 		break;
1337 	}
1338 	if (error == 0) {
1339 		freemsg(mp);
1340 		return;
1341 	}
1342 
1343 	/*
1344 	 * Deliver T_UDERROR_IND when the application has asked for it.
1345 	 * The socket layer enables this automatically when connected.
1346 	 */
1347 	if (!connp->conn_dgram_errind) {
1348 		freemsg(mp);
1349 		return;
1350 	}
1351 
1352 	sin6 = sin6_null;
1353 	sin6.sin6_family = AF_INET6;
1354 	sin6.sin6_addr = ip6h->ip6_dst;
1355 	sin6.sin6_flowinfo = ip6h->ip6_vcf & ~IPV6_VERS_AND_FLOW_MASK;
1356 	if (IPCL_IS_NONSTR(connp)) {
1357 		mutex_enter(&connp->conn_lock);
1358 		if (icmp->icmp_state == TS_DATA_XFER) {
1359 			if (IN6_ARE_ADDR_EQUAL(&sin6.sin6_addr,
1360 			    &connp->conn_faddr_v6)) {
1361 				mutex_exit(&connp->conn_lock);
1362 				(*connp->conn_upcalls->su_set_error)
1363 				    (connp->conn_upper_handle, error);
1364 				goto done;
1365 			}
1366 		} else {
1367 			icmp->icmp_delayed_error = error;
1368 			*((sin6_t *)&icmp->icmp_delayed_addr) = sin6;
1369 		}
1370 		mutex_exit(&connp->conn_lock);
1371 	} else {
1372 		mp1 = mi_tpi_uderror_ind((char *)&sin6, sizeof (sin6_t),
1373 		    NULL, 0, error);
1374 		if (mp1 != NULL)
1375 			putnext(connp->conn_rq, mp1);
1376 	}
1377 done:
1378 	freemsg(mp);
1379 }
1380 
1381 /*
1382  * This routine responds to T_ADDR_REQ messages.  It is called by icmp_wput.
1383  * The local address is filled in if endpoint is bound. The remote address
1384  * is filled in if remote address has been precified ("connected endpoint")
1385  * (The concept of connected CLTS sockets is alien to published TPI
1386  *  but we support it anyway).
1387  */
1388 static void
1389 icmp_addr_req(queue_t *q, mblk_t *mp)
1390 {
1391 	struct sockaddr *sa;
1392 	mblk_t	*ackmp;
1393 	struct T_addr_ack *taa;
1394 	icmp_t	*icmp = Q_TO_ICMP(q);
1395 	conn_t	*connp = icmp->icmp_connp;
1396 	uint_t	addrlen;
1397 
1398 	/* Make it large enough for worst case */
1399 	ackmp = reallocb(mp, sizeof (struct T_addr_ack) +
1400 	    2 * sizeof (sin6_t), 1);
1401 	if (ackmp == NULL) {
1402 		icmp_err_ack(q, mp, TSYSERR, ENOMEM);
1403 		return;
1404 	}
1405 	taa = (struct T_addr_ack *)ackmp->b_rptr;
1406 
1407 	bzero(taa, sizeof (struct T_addr_ack));
1408 	ackmp->b_wptr = (uchar_t *)&taa[1];
1409 
1410 	taa->PRIM_type = T_ADDR_ACK;
1411 	ackmp->b_datap->db_type = M_PCPROTO;
1412 
1413 	if (connp->conn_family == AF_INET)
1414 		addrlen = sizeof (sin_t);
1415 	else
1416 		addrlen = sizeof (sin6_t);
1417 
1418 	mutex_enter(&connp->conn_lock);
1419 	/*
1420 	 * Note: Following code assumes 32 bit alignment of basic
1421 	 * data structures like sin_t and struct T_addr_ack.
1422 	 */
1423 	if (icmp->icmp_state != TS_UNBND) {
1424 		/*
1425 		 * Fill in local address first
1426 		 */
1427 		taa->LOCADDR_offset = sizeof (*taa);
1428 		taa->LOCADDR_length = addrlen;
1429 		sa = (struct sockaddr *)&taa[1];
1430 		(void) conn_getsockname(connp, sa, &addrlen);
1431 		ackmp->b_wptr += addrlen;
1432 	}
1433 	if (icmp->icmp_state == TS_DATA_XFER) {
1434 		/*
1435 		 * connected, fill remote address too
1436 		 */
1437 		taa->REMADDR_length = addrlen;
1438 		/* assumed 32-bit alignment */
1439 		taa->REMADDR_offset = taa->LOCADDR_offset + taa->LOCADDR_length;
1440 		sa = (struct sockaddr *)(ackmp->b_rptr + taa->REMADDR_offset);
1441 		(void) conn_getpeername(connp, sa, &addrlen);
1442 		ackmp->b_wptr += addrlen;
1443 	}
1444 	mutex_exit(&connp->conn_lock);
1445 	ASSERT(ackmp->b_wptr <= ackmp->b_datap->db_lim);
1446 	qreply(q, ackmp);
1447 }
1448 
1449 static void
1450 icmp_copy_info(struct T_info_ack *tap, icmp_t *icmp)
1451 {
1452 	conn_t		*connp = icmp->icmp_connp;
1453 
1454 	*tap = icmp_g_t_info_ack;
1455 
1456 	if (connp->conn_family == AF_INET6)
1457 		tap->ADDR_size = sizeof (sin6_t);
1458 	else
1459 		tap->ADDR_size = sizeof (sin_t);
1460 	tap->CURRENT_state = icmp->icmp_state;
1461 	tap->OPT_size = icmp_max_optsize;
1462 }
1463 
1464 static void
1465 icmp_do_capability_ack(icmp_t *icmp, struct T_capability_ack *tcap,
1466     t_uscalar_t cap_bits1)
1467 {
1468 	tcap->CAP_bits1 = 0;
1469 
1470 	if (cap_bits1 & TC1_INFO) {
1471 		icmp_copy_info(&tcap->INFO_ack, icmp);
1472 		tcap->CAP_bits1 |= TC1_INFO;
1473 	}
1474 }
1475 
1476 /*
1477  * This routine responds to T_CAPABILITY_REQ messages.  It is called by
1478  * icmp_wput.  Much of the T_CAPABILITY_ACK information is copied from
1479  * icmp_g_t_info_ack.  The current state of the stream is copied from
1480  * icmp_state.
1481  */
1482 static void
1483 icmp_capability_req(queue_t *q, mblk_t *mp)
1484 {
1485 	icmp_t			*icmp = Q_TO_ICMP(q);
1486 	t_uscalar_t		cap_bits1;
1487 	struct T_capability_ack	*tcap;
1488 
1489 	cap_bits1 = ((struct T_capability_req *)mp->b_rptr)->CAP_bits1;
1490 
1491 	mp = tpi_ack_alloc(mp, sizeof (struct T_capability_ack),
1492 	    mp->b_datap->db_type, T_CAPABILITY_ACK);
1493 	if (!mp)
1494 		return;
1495 
1496 	tcap = (struct T_capability_ack *)mp->b_rptr;
1497 
1498 	icmp_do_capability_ack(icmp, tcap, cap_bits1);
1499 
1500 	qreply(q, mp);
1501 }
1502 
1503 /*
1504  * This routine responds to T_INFO_REQ messages.  It is called by icmp_wput.
1505  * Most of the T_INFO_ACK information is copied from icmp_g_t_info_ack.
1506  * The current state of the stream is copied from icmp_state.
1507  */
1508 static void
1509 icmp_info_req(queue_t *q, mblk_t *mp)
1510 {
1511 	icmp_t	*icmp = Q_TO_ICMP(q);
1512 
1513 	/* Create a T_INFO_ACK message. */
1514 	mp = tpi_ack_alloc(mp, sizeof (struct T_info_ack), M_PCPROTO,
1515 	    T_INFO_ACK);
1516 	if (!mp)
1517 		return;
1518 	icmp_copy_info((struct T_info_ack *)mp->b_rptr, icmp);
1519 	qreply(q, mp);
1520 }
1521 
1522 static int
1523 icmp_tpi_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
1524     int family)
1525 {
1526 	conn_t *connp;
1527 	dev_t	conn_dev;
1528 	int	error;
1529 
1530 	/* If the stream is already open, return immediately. */
1531 	if (q->q_ptr != NULL)
1532 		return (0);
1533 
1534 	if (sflag == MODOPEN)
1535 		return (EINVAL);
1536 
1537 	/*
1538 	 * Since ICMP is not used so heavily, allocating from the small
1539 	 * arena should be sufficient.
1540 	 */
1541 	if ((conn_dev = inet_minor_alloc(ip_minor_arena_sa)) == 0) {
1542 		return (EBUSY);
1543 	}
1544 
1545 	if (flag & SO_FALLBACK) {
1546 		/*
1547 		 * Non streams socket needs a stream to fallback to
1548 		 */
1549 		RD(q)->q_ptr = (void *)conn_dev;
1550 		WR(q)->q_qinfo = &icmp_fallback_sock_winit;
1551 		WR(q)->q_ptr = (void *)ip_minor_arena_sa;
1552 		qprocson(q);
1553 		return (0);
1554 	}
1555 
1556 	connp = rawip_do_open(family, credp, &error, KM_SLEEP);
1557 	if (connp == NULL) {
1558 		ASSERT(error != 0);
1559 		inet_minor_free(ip_minor_arena_sa, connp->conn_dev);
1560 		return (error);
1561 	}
1562 
1563 	*devp = makedevice(getemajor(*devp), (minor_t)conn_dev);
1564 	connp->conn_dev = conn_dev;
1565 	connp->conn_minor_arena = ip_minor_arena_sa;
1566 
1567 	/*
1568 	 * Initialize the icmp_t structure for this stream.
1569 	 */
1570 	q->q_ptr = connp;
1571 	WR(q)->q_ptr = connp;
1572 	connp->conn_rq = q;
1573 	connp->conn_wq = WR(q);
1574 
1575 	WR(q)->q_hiwat = connp->conn_sndbuf;
1576 	WR(q)->q_lowat = connp->conn_sndlowat;
1577 
1578 	qprocson(q);
1579 
1580 	/* Set the Stream head write offset. */
1581 	(void) proto_set_tx_wroff(q, connp, connp->conn_wroff);
1582 	(void) proto_set_rx_hiwat(connp->conn_rq, connp, connp->conn_rcvbuf);
1583 
1584 	mutex_enter(&connp->conn_lock);
1585 	connp->conn_state_flags &= ~CONN_INCIPIENT;
1586 	mutex_exit(&connp->conn_lock);
1587 
1588 	icmp_bind_proto(connp->conn_icmp);
1589 
1590 	return (0);
1591 }
1592 
1593 /* For /dev/icmp aka AF_INET open */
1594 static int
1595 icmp_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1596 {
1597 	return (icmp_tpi_open(q, devp, flag, sflag, credp, AF_INET));
1598 }
1599 
1600 /* For /dev/icmp6 aka AF_INET6 open */
1601 static int
1602 icmp_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1603 {
1604 	return (icmp_tpi_open(q, devp, flag, sflag, credp, AF_INET6));
1605 }
1606 
1607 /*
1608  * This is the open routine for icmp.  It allocates a icmp_t structure for
1609  * the stream and, on the first open of the module, creates an ND table.
1610  */
1611 static conn_t *
1612 rawip_do_open(int family, cred_t *credp, int *err, int flags)
1613 {
1614 	icmp_t	*icmp;
1615 	conn_t *connp;
1616 	zoneid_t zoneid;
1617 	netstack_t *ns;
1618 	icmp_stack_t *is;
1619 	int len;
1620 	boolean_t isv6 = B_FALSE;
1621 
1622 	*err = secpolicy_net_icmpaccess(credp);
1623 	if (*err != 0)
1624 		return (NULL);
1625 
1626 	if (family == AF_INET6)
1627 		isv6 = B_TRUE;
1628 
1629 	ns = netstack_find_by_cred(credp);
1630 	ASSERT(ns != NULL);
1631 	is = ns->netstack_icmp;
1632 	ASSERT(is != NULL);
1633 
1634 	/*
1635 	 * For exclusive stacks we set the zoneid to zero
1636 	 * to make ICMP operate as if in the global zone.
1637 	 */
1638 	if (ns->netstack_stackid != GLOBAL_NETSTACKID)
1639 		zoneid = GLOBAL_ZONEID;
1640 	else
1641 		zoneid = crgetzoneid(credp);
1642 
1643 	ASSERT(flags == KM_SLEEP || flags == KM_NOSLEEP);
1644 
1645 	connp = ipcl_conn_create(IPCL_RAWIPCONN, flags, ns);
1646 	icmp = connp->conn_icmp;
1647 
1648 	/*
1649 	 * ipcl_conn_create did a netstack_hold. Undo the hold that was
1650 	 * done by netstack_find_by_cred()
1651 	 */
1652 	netstack_rele(ns);
1653 
1654 	/*
1655 	 * Since this conn_t/icmp_t is not yet visible to anybody else we don't
1656 	 * need to lock anything.
1657 	 */
1658 	ASSERT(connp->conn_proto == IPPROTO_ICMP);
1659 	ASSERT(connp->conn_icmp == icmp);
1660 	ASSERT(icmp->icmp_connp == connp);
1661 
1662 	/* Set the initial state of the stream and the privilege status. */
1663 	icmp->icmp_state = TS_UNBND;
1664 	connp->conn_ixa->ixa_flags |= IXAF_VERIFY_SOURCE;
1665 	if (isv6) {
1666 		connp->conn_family = AF_INET6;
1667 		connp->conn_ipversion = IPV6_VERSION;
1668 		connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
1669 		connp->conn_proto = IPPROTO_ICMPV6;
1670 		/* May be changed by a SO_PROTOTYPE socket option. */
1671 		connp->conn_proto = IPPROTO_ICMPV6;
1672 		connp->conn_ixa->ixa_protocol = connp->conn_proto;
1673 		connp->conn_ixa->ixa_raw_cksum_offset = 2;
1674 		connp->conn_default_ttl = is->is_ipv6_hoplimit;
1675 		len = sizeof (ip6_t);
1676 	} else {
1677 		connp->conn_family = AF_INET;
1678 		connp->conn_ipversion = IPV4_VERSION;
1679 		connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
1680 		/* May be changed by a SO_PROTOTYPE socket option. */
1681 		connp->conn_proto = IPPROTO_ICMP;
1682 		connp->conn_ixa->ixa_protocol = connp->conn_proto;
1683 		connp->conn_default_ttl = is->is_ipv4_ttl;
1684 		len = sizeof (ipha_t);
1685 	}
1686 	connp->conn_xmit_ipp.ipp_unicast_hops = connp->conn_default_ttl;
1687 
1688 	connp->conn_ixa->ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1689 
1690 	/*
1691 	 * For the socket of protocol IPPROTO_RAW or when IP_HDRINCL is set,
1692 	 * the checksum is provided in the pre-built packet. We clear
1693 	 * IXAF_SET_ULP_CKSUM to tell IP that the application has sent a
1694 	 * complete IP header and not to compute the transport checksum.
1695 	 */
1696 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
1697 	/* conn_allzones can not be set this early, hence no IPCL_ZONEID */
1698 	connp->conn_ixa->ixa_zoneid = zoneid;
1699 
1700 	connp->conn_zoneid = zoneid;
1701 
1702 	/*
1703 	 * If the caller has the process-wide flag set, then default to MAC
1704 	 * exempt mode.  This allows read-down to unlabeled hosts.
1705 	 */
1706 	if (getpflags(NET_MAC_AWARE, credp) != 0)
1707 		connp->conn_mac_mode = CONN_MAC_AWARE;
1708 
1709 	connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
1710 
1711 	icmp->icmp_is = is;
1712 
1713 	connp->conn_rcvbuf = is->is_recv_hiwat;
1714 	connp->conn_sndbuf = is->is_xmit_hiwat;
1715 	connp->conn_sndlowat = is->is_xmit_lowat;
1716 	connp->conn_rcvlowat = icmp_mod_info.mi_lowat;
1717 
1718 	connp->conn_wroff = len + is->is_wroff_extra;
1719 	connp->conn_so_type = SOCK_RAW;
1720 
1721 	connp->conn_recv = icmp_input;
1722 	connp->conn_recvicmp = icmp_icmp_input;
1723 	crhold(credp);
1724 	connp->conn_cred = credp;
1725 	connp->conn_cpid = curproc->p_pid;
1726 	connp->conn_open_time = ddi_get_lbolt64();
1727 	/* Cache things in ixa without an extra refhold */
1728 	connp->conn_ixa->ixa_cred = connp->conn_cred;
1729 	connp->conn_ixa->ixa_cpid = connp->conn_cpid;
1730 	if (is_system_labeled())
1731 		connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
1732 
1733 	connp->conn_flow_cntrld = B_FALSE;
1734 
1735 	if (is->is_pmtu_discovery)
1736 		connp->conn_ixa->ixa_flags |= IXAF_PMTU_DISCOVERY;
1737 
1738 	return (connp);
1739 }
1740 
1741 /*
1742  * Which ICMP options OK to set through T_UNITDATA_REQ...
1743  */
1744 /* ARGSUSED */
1745 static boolean_t
1746 icmp_opt_allow_udr_set(t_scalar_t level, t_scalar_t name)
1747 {
1748 	return (B_TRUE);
1749 }
1750 
1751 /*
1752  * This routine gets default values of certain options whose default
1753  * values are maintained by protcol specific code
1754  */
1755 int
1756 icmp_opt_default(queue_t *q, t_scalar_t level, t_scalar_t name, uchar_t *ptr)
1757 {
1758 	icmp_t *icmp = Q_TO_ICMP(q);
1759 	icmp_stack_t *is = icmp->icmp_is;
1760 	int *i1 = (int *)ptr;
1761 
1762 	switch (level) {
1763 	case IPPROTO_IP:
1764 		switch (name) {
1765 		case IP_MULTICAST_TTL:
1766 			*ptr = (uchar_t)IP_DEFAULT_MULTICAST_TTL;
1767 			return (sizeof (uchar_t));
1768 		case IP_MULTICAST_LOOP:
1769 			*ptr = (uchar_t)IP_DEFAULT_MULTICAST_LOOP;
1770 			return (sizeof (uchar_t));
1771 		}
1772 		break;
1773 	case IPPROTO_IPV6:
1774 		switch (name) {
1775 		case IPV6_MULTICAST_HOPS:
1776 			*i1 = IP_DEFAULT_MULTICAST_TTL;
1777 			return (sizeof (int));
1778 		case IPV6_MULTICAST_LOOP:
1779 			*i1 = IP_DEFAULT_MULTICAST_LOOP;
1780 			return (sizeof (int));
1781 		case IPV6_UNICAST_HOPS:
1782 			*i1 = is->is_ipv6_hoplimit;
1783 			return (sizeof (int));
1784 		}
1785 		break;
1786 	case IPPROTO_ICMPV6:
1787 		switch (name) {
1788 		case ICMP6_FILTER:
1789 			/* Make it look like "pass all" */
1790 			ICMP6_FILTER_SETPASSALL((icmp6_filter_t *)ptr);
1791 			return (sizeof (icmp6_filter_t));
1792 		}
1793 		break;
1794 	}
1795 	return (-1);
1796 }
1797 
1798 /*
1799  * This routine retrieves the current status of socket options.
1800  * It returns the size of the option retrieved, or -1.
1801  */
1802 int
1803 icmp_opt_get(conn_t *connp, int level, int name, uchar_t *ptr)
1804 {
1805 	icmp_t		*icmp = connp->conn_icmp;
1806 	int		*i1 = (int *)ptr;
1807 	conn_opt_arg_t	coas;
1808 	int		retval;
1809 
1810 	coas.coa_connp = connp;
1811 	coas.coa_ixa = connp->conn_ixa;
1812 	coas.coa_ipp = &connp->conn_xmit_ipp;
1813 	coas.coa_ancillary = B_FALSE;
1814 	coas.coa_changed = 0;
1815 
1816 	/*
1817 	 * We assume that the optcom framework has checked for the set
1818 	 * of levels and names that are supported, hence we don't worry
1819 	 * about rejecting based on that.
1820 	 * First check for ICMP specific handling, then pass to common routine.
1821 	 */
1822 	switch (level) {
1823 	case IPPROTO_IP:
1824 		/*
1825 		 * Only allow IPv4 option processing on IPv4 sockets.
1826 		 */
1827 		if (connp->conn_family != AF_INET)
1828 			return (-1);
1829 
1830 		switch (name) {
1831 		case IP_OPTIONS:
1832 		case T_IP_OPTIONS:
1833 			/* Options are passed up with each packet */
1834 			return (0);
1835 		case IP_HDRINCL:
1836 			mutex_enter(&connp->conn_lock);
1837 			*i1 = (int)icmp->icmp_hdrincl;
1838 			mutex_exit(&connp->conn_lock);
1839 			return (sizeof (int));
1840 		}
1841 		break;
1842 
1843 	case IPPROTO_IPV6:
1844 		/*
1845 		 * Only allow IPv6 option processing on native IPv6 sockets.
1846 		 */
1847 		if (connp->conn_family != AF_INET6)
1848 			return (-1);
1849 
1850 		switch (name) {
1851 		case IPV6_CHECKSUM:
1852 			/*
1853 			 * Return offset or -1 if no checksum offset.
1854 			 * Does not apply to IPPROTO_ICMPV6
1855 			 */
1856 			if (connp->conn_proto == IPPROTO_ICMPV6)
1857 				return (-1);
1858 
1859 			mutex_enter(&connp->conn_lock);
1860 			if (connp->conn_ixa->ixa_flags & IXAF_SET_RAW_CKSUM)
1861 				*i1 = connp->conn_ixa->ixa_raw_cksum_offset;
1862 			else
1863 				*i1 = -1;
1864 			mutex_exit(&connp->conn_lock);
1865 			return (sizeof (int));
1866 		}
1867 		break;
1868 
1869 	case IPPROTO_ICMPV6:
1870 		/*
1871 		 * Only allow IPv6 option processing on native IPv6 sockets.
1872 		 */
1873 		if (connp->conn_family != AF_INET6)
1874 			return (-1);
1875 
1876 		if (connp->conn_proto != IPPROTO_ICMPV6)
1877 			return (-1);
1878 
1879 		switch (name) {
1880 		case ICMP6_FILTER:
1881 			mutex_enter(&connp->conn_lock);
1882 			if (icmp->icmp_filter == NULL) {
1883 				/* Make it look like "pass all" */
1884 				ICMP6_FILTER_SETPASSALL((icmp6_filter_t *)ptr);
1885 			} else {
1886 				(void) bcopy(icmp->icmp_filter, ptr,
1887 				    sizeof (icmp6_filter_t));
1888 			}
1889 			mutex_exit(&connp->conn_lock);
1890 			return (sizeof (icmp6_filter_t));
1891 		}
1892 	}
1893 	mutex_enter(&connp->conn_lock);
1894 	retval = conn_opt_get(&coas, level, name, ptr);
1895 	mutex_exit(&connp->conn_lock);
1896 	return (retval);
1897 }
1898 
1899 /*
1900  * This routine retrieves the current status of socket options.
1901  * It returns the size of the option retrieved, or -1.
1902  */
1903 int
1904 icmp_tpi_opt_get(queue_t *q, int level, int name, uchar_t *ptr)
1905 {
1906 	conn_t		*connp = Q_TO_CONN(q);
1907 	int 		err;
1908 
1909 	err = icmp_opt_get(connp, level, name, ptr);
1910 	return (err);
1911 }
1912 
1913 /*
1914  * This routine sets socket options.
1915  */
1916 int
1917 icmp_do_opt_set(conn_opt_arg_t *coa, int level, int name,
1918     uint_t inlen, uchar_t *invalp, cred_t *cr, boolean_t checkonly)
1919 {
1920 	conn_t		*connp = coa->coa_connp;
1921 	ip_xmit_attr_t	*ixa = coa->coa_ixa;
1922 	icmp_t		*icmp = connp->conn_icmp;
1923 	icmp_stack_t	*is = icmp->icmp_is;
1924 	int		*i1 = (int *)invalp;
1925 	boolean_t	onoff = (*i1 == 0) ? 0 : 1;
1926 	int		error;
1927 
1928 	ASSERT(MUTEX_NOT_HELD(&coa->coa_connp->conn_lock));
1929 
1930 	/*
1931 	 * For fixed length options, no sanity check
1932 	 * of passed in length is done. It is assumed *_optcom_req()
1933 	 * routines do the right thing.
1934 	 */
1935 
1936 	switch (level) {
1937 	case SOL_SOCKET:
1938 		switch (name) {
1939 		case SO_PROTOTYPE:
1940 			if ((*i1 & 0xFF) != IPPROTO_ICMP &&
1941 			    (*i1 & 0xFF) != IPPROTO_ICMPV6 &&
1942 			    secpolicy_net_rawaccess(cr) != 0) {
1943 				return (EACCES);
1944 			}
1945 			if (checkonly)
1946 				break;
1947 
1948 			mutex_enter(&connp->conn_lock);
1949 			connp->conn_proto = *i1 & 0xFF;
1950 			ixa->ixa_protocol = connp->conn_proto;
1951 			if ((connp->conn_proto == IPPROTO_RAW ||
1952 			    connp->conn_proto == IPPROTO_IGMP) &&
1953 			    connp->conn_family == AF_INET) {
1954 				icmp->icmp_hdrincl = 1;
1955 				ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
1956 			} else if (connp->conn_proto == IPPROTO_UDP ||
1957 			    connp->conn_proto == IPPROTO_TCP ||
1958 			    connp->conn_proto == IPPROTO_SCTP) {
1959 				/* Used by test applications like psh */
1960 				icmp->icmp_hdrincl = 0;
1961 				ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
1962 			} else {
1963 				icmp->icmp_hdrincl = 0;
1964 				ixa->ixa_flags |= IXAF_SET_ULP_CKSUM;
1965 			}
1966 
1967 			if (connp->conn_family == AF_INET6 &&
1968 			    connp->conn_proto == IPPROTO_ICMPV6) {
1969 				/* Set offset for icmp6_cksum */
1970 				ixa->ixa_flags &= ~IXAF_SET_RAW_CKSUM;
1971 				ixa->ixa_raw_cksum_offset = 2;
1972 			}
1973 			if (icmp->icmp_filter != NULL &&
1974 			    connp->conn_proto != IPPROTO_ICMPV6) {
1975 				kmem_free(icmp->icmp_filter,
1976 				    sizeof (icmp6_filter_t));
1977 				icmp->icmp_filter = NULL;
1978 			}
1979 			mutex_exit(&connp->conn_lock);
1980 
1981 			coa->coa_changed |= COA_HEADER_CHANGED;
1982 			/*
1983 			 * For SCTP, we don't use icmp_bind_proto() for
1984 			 * raw socket binding.
1985 			 */
1986 			if (connp->conn_proto == IPPROTO_SCTP)
1987 				return (0);
1988 
1989 			coa->coa_changed |= COA_ICMP_BIND_NEEDED;
1990 			return (0);
1991 
1992 		case SO_SNDBUF:
1993 			if (*i1 > is->is_max_buf) {
1994 				return (ENOBUFS);
1995 			}
1996 			break;
1997 		case SO_RCVBUF:
1998 			if (*i1 > is->is_max_buf) {
1999 				return (ENOBUFS);
2000 			}
2001 			break;
2002 		}
2003 		break;
2004 
2005 	case IPPROTO_IP:
2006 		/*
2007 		 * Only allow IPv4 option processing on IPv4 sockets.
2008 		 */
2009 		if (connp->conn_family != AF_INET)
2010 			return (EINVAL);
2011 
2012 		switch (name) {
2013 		case IP_HDRINCL:
2014 			if (!checkonly) {
2015 				mutex_enter(&connp->conn_lock);
2016 				icmp->icmp_hdrincl = onoff;
2017 				if (onoff)
2018 					ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
2019 				else
2020 					ixa->ixa_flags |= IXAF_SET_ULP_CKSUM;
2021 				mutex_exit(&connp->conn_lock);
2022 			}
2023 			break;
2024 		}
2025 		break;
2026 
2027 	case IPPROTO_IPV6:
2028 		if (connp->conn_family != AF_INET6)
2029 			return (EINVAL);
2030 
2031 		switch (name) {
2032 		case IPV6_CHECKSUM:
2033 			/*
2034 			 * Integer offset into the user data of where the
2035 			 * checksum is located.
2036 			 * Offset of -1 disables option.
2037 			 * Does not apply to IPPROTO_ICMPV6.
2038 			 */
2039 			if (connp->conn_proto == IPPROTO_ICMPV6 ||
2040 			    coa->coa_ancillary) {
2041 				return (EINVAL);
2042 			}
2043 			if ((*i1 != -1) && ((*i1 < 0) || (*i1 & 0x1) != 0)) {
2044 				/* Negative or not 16 bit aligned offset */
2045 				return (EINVAL);
2046 			}
2047 			if (checkonly)
2048 				break;
2049 
2050 			mutex_enter(&connp->conn_lock);
2051 			if (*i1 == -1) {
2052 				ixa->ixa_flags &= ~IXAF_SET_RAW_CKSUM;
2053 				ixa->ixa_raw_cksum_offset = 0;
2054 				ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
2055 			} else {
2056 				ixa->ixa_flags |= IXAF_SET_RAW_CKSUM;
2057 				ixa->ixa_raw_cksum_offset = *i1;
2058 				ixa->ixa_flags |= IXAF_SET_ULP_CKSUM;
2059 			}
2060 			mutex_exit(&connp->conn_lock);
2061 			break;
2062 		}
2063 		break;
2064 
2065 	case IPPROTO_ICMPV6:
2066 		/*
2067 		 * Only allow IPv6 option processing on IPv6 sockets.
2068 		 */
2069 		if (connp->conn_family != AF_INET6)
2070 			return (EINVAL);
2071 		if (connp->conn_proto != IPPROTO_ICMPV6)
2072 			return (EINVAL);
2073 
2074 		switch (name) {
2075 		case ICMP6_FILTER:
2076 			if (checkonly)
2077 				break;
2078 
2079 			if ((inlen != 0) &&
2080 			    (inlen != sizeof (icmp6_filter_t)))
2081 				return (EINVAL);
2082 
2083 			mutex_enter(&connp->conn_lock);
2084 			if (inlen == 0) {
2085 				if (icmp->icmp_filter != NULL) {
2086 					kmem_free(icmp->icmp_filter,
2087 					    sizeof (icmp6_filter_t));
2088 					icmp->icmp_filter = NULL;
2089 				}
2090 			} else {
2091 				if (icmp->icmp_filter == NULL) {
2092 					icmp->icmp_filter = kmem_alloc(
2093 					    sizeof (icmp6_filter_t),
2094 					    KM_NOSLEEP);
2095 					if (icmp->icmp_filter == NULL) {
2096 						mutex_exit(&connp->conn_lock);
2097 						return (ENOBUFS);
2098 					}
2099 				}
2100 				(void) bcopy(invalp, icmp->icmp_filter, inlen);
2101 			}
2102 			mutex_exit(&connp->conn_lock);
2103 			break;
2104 		}
2105 		break;
2106 	}
2107 	error = conn_opt_set(coa, level, name, inlen, invalp,
2108 	    checkonly, cr);
2109 	return (error);
2110 }
2111 
2112 /*
2113  * This routine sets socket options.
2114  */
2115 int
2116 icmp_opt_set(conn_t *connp, uint_t optset_context, int level, int name,
2117     uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
2118     void *thisdg_attrs, cred_t *cr)
2119 {
2120 	icmp_t		*icmp = connp->conn_icmp;
2121 	int		err;
2122 	conn_opt_arg_t	coas, *coa;
2123 	boolean_t	checkonly;
2124 	icmp_stack_t	*is = icmp->icmp_is;
2125 
2126 	switch (optset_context) {
2127 	case SETFN_OPTCOM_CHECKONLY:
2128 		checkonly = B_TRUE;
2129 		/*
2130 		 * Note: Implies T_CHECK semantics for T_OPTCOM_REQ
2131 		 * inlen != 0 implies value supplied and
2132 		 * 	we have to "pretend" to set it.
2133 		 * inlen == 0 implies that there is no
2134 		 * 	value part in T_CHECK request and just validation
2135 		 * done elsewhere should be enough, we just return here.
2136 		 */
2137 		if (inlen == 0) {
2138 			*outlenp = 0;
2139 			return (0);
2140 		}
2141 		break;
2142 	case SETFN_OPTCOM_NEGOTIATE:
2143 		checkonly = B_FALSE;
2144 		break;
2145 	case SETFN_UD_NEGOTIATE:
2146 	case SETFN_CONN_NEGOTIATE:
2147 		checkonly = B_FALSE;
2148 		/*
2149 		 * Negotiating local and "association-related" options
2150 		 * through T_UNITDATA_REQ.
2151 		 *
2152 		 * Following routine can filter out ones we do not
2153 		 * want to be "set" this way.
2154 		 */
2155 		if (!icmp_opt_allow_udr_set(level, name)) {
2156 			*outlenp = 0;
2157 			return (EINVAL);
2158 		}
2159 		break;
2160 	default:
2161 		/*
2162 		 * We should never get here
2163 		 */
2164 		*outlenp = 0;
2165 		return (EINVAL);
2166 	}
2167 
2168 	ASSERT((optset_context != SETFN_OPTCOM_CHECKONLY) ||
2169 	    (optset_context == SETFN_OPTCOM_CHECKONLY && inlen != 0));
2170 
2171 	if (thisdg_attrs != NULL) {
2172 		/* Options from T_UNITDATA_REQ */
2173 		coa = (conn_opt_arg_t *)thisdg_attrs;
2174 		ASSERT(coa->coa_connp == connp);
2175 		ASSERT(coa->coa_ixa != NULL);
2176 		ASSERT(coa->coa_ipp != NULL);
2177 		ASSERT(coa->coa_ancillary);
2178 	} else {
2179 		coa = &coas;
2180 		coas.coa_connp = connp;
2181 		/* Get a reference on conn_ixa to prevent concurrent mods */
2182 		coas.coa_ixa = conn_get_ixa(connp, B_TRUE);
2183 		if (coas.coa_ixa == NULL) {
2184 			*outlenp = 0;
2185 			return (ENOMEM);
2186 		}
2187 		coas.coa_ipp = &connp->conn_xmit_ipp;
2188 		coas.coa_ancillary = B_FALSE;
2189 		coas.coa_changed = 0;
2190 	}
2191 
2192 	err = icmp_do_opt_set(coa, level, name, inlen, invalp,
2193 	    cr, checkonly);
2194 	if (err != 0) {
2195 errout:
2196 		if (!coa->coa_ancillary)
2197 			ixa_refrele(coa->coa_ixa);
2198 		*outlenp = 0;
2199 		return (err);
2200 	}
2201 
2202 	/*
2203 	 * Common case of OK return with outval same as inval.
2204 	 */
2205 	if (invalp != outvalp) {
2206 		/* don't trust bcopy for identical src/dst */
2207 		(void) bcopy(invalp, outvalp, inlen);
2208 	}
2209 	*outlenp = inlen;
2210 
2211 	/*
2212 	 * If this was not ancillary data, then we rebuild the headers,
2213 	 * update the IRE/NCE, and IPsec as needed.
2214 	 * Since the label depends on the destination we go through
2215 	 * ip_set_destination first.
2216 	 */
2217 	if (coa->coa_ancillary) {
2218 		return (0);
2219 	}
2220 
2221 	if (coa->coa_changed & COA_ROUTE_CHANGED) {
2222 		in6_addr_t saddr, faddr, nexthop;
2223 		in_port_t fport;
2224 
2225 		/*
2226 		 * We clear lastdst to make sure we pick up the change
2227 		 * next time sending.
2228 		 * If we are connected we re-cache the information.
2229 		 * We ignore errors to preserve BSD behavior.
2230 		 * Note that we don't redo IPsec policy lookup here
2231 		 * since the final destination (or source) didn't change.
2232 		 */
2233 		mutex_enter(&connp->conn_lock);
2234 		connp->conn_v6lastdst = ipv6_all_zeros;
2235 
2236 		ip_attr_nexthop(coa->coa_ipp, coa->coa_ixa,
2237 		    &connp->conn_faddr_v6, &nexthop);
2238 		saddr = connp->conn_saddr_v6;
2239 		faddr = connp->conn_faddr_v6;
2240 		fport = connp->conn_fport;
2241 		mutex_exit(&connp->conn_lock);
2242 
2243 		if (!IN6_IS_ADDR_UNSPECIFIED(&faddr) &&
2244 		    !IN6_IS_ADDR_V4MAPPED_ANY(&faddr)) {
2245 			(void) ip_attr_connect(connp, coa->coa_ixa,
2246 			    &saddr, &faddr, &nexthop, fport, NULL, NULL,
2247 			    IPDF_ALLOW_MCBC | IPDF_VERIFY_DST);
2248 		}
2249 	}
2250 
2251 	ixa_refrele(coa->coa_ixa);
2252 
2253 	if (coa->coa_changed & COA_HEADER_CHANGED) {
2254 		/*
2255 		 * Rebuild the header template if we are connected.
2256 		 * Otherwise clear conn_v6lastdst so we rebuild the header
2257 		 * in the data path.
2258 		 */
2259 		mutex_enter(&connp->conn_lock);
2260 		if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) &&
2261 		    !IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_faddr_v6)) {
2262 			err = icmp_build_hdr_template(connp,
2263 			    &connp->conn_saddr_v6, &connp->conn_faddr_v6,
2264 			    connp->conn_flowinfo);
2265 			if (err != 0) {
2266 				mutex_exit(&connp->conn_lock);
2267 				return (err);
2268 			}
2269 		} else {
2270 			connp->conn_v6lastdst = ipv6_all_zeros;
2271 		}
2272 		mutex_exit(&connp->conn_lock);
2273 	}
2274 	if (coa->coa_changed & COA_RCVBUF_CHANGED) {
2275 		(void) proto_set_rx_hiwat(connp->conn_rq, connp,
2276 		    connp->conn_rcvbuf);
2277 	}
2278 	if ((coa->coa_changed & COA_SNDBUF_CHANGED) && !IPCL_IS_NONSTR(connp)) {
2279 		connp->conn_wq->q_hiwat = connp->conn_sndbuf;
2280 	}
2281 	if (coa->coa_changed & COA_WROFF_CHANGED) {
2282 		/* Increase wroff if needed */
2283 		uint_t wroff;
2284 
2285 		mutex_enter(&connp->conn_lock);
2286 		wroff = connp->conn_ht_iphc_allocated + is->is_wroff_extra;
2287 		if (wroff > connp->conn_wroff) {
2288 			connp->conn_wroff = wroff;
2289 			mutex_exit(&connp->conn_lock);
2290 			(void) proto_set_tx_wroff(connp->conn_rq, connp, wroff);
2291 		} else {
2292 			mutex_exit(&connp->conn_lock);
2293 		}
2294 	}
2295 	if (coa->coa_changed & COA_ICMP_BIND_NEEDED) {
2296 		icmp_bind_proto(icmp);
2297 	}
2298 	return (err);
2299 }
2300 
2301 /* This routine sets socket options. */
2302 int
2303 icmp_tpi_opt_set(queue_t *q, uint_t optset_context, int level, int name,
2304     uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
2305     void *thisdg_attrs, cred_t *cr)
2306 {
2307 	conn_t	*connp = Q_TO_CONN(q);
2308 	int error;
2309 
2310 	error = icmp_opt_set(connp, optset_context, level, name, inlen, invalp,
2311 	    outlenp, outvalp, thisdg_attrs, cr);
2312 	return (error);
2313 }
2314 
2315 /*
2316  * Setup IP headers.
2317  *
2318  * Note that IP_HDRINCL has ipha_protocol that is different than conn_proto,
2319  * but icmp_output_hdrincl restores ipha_protocol once we return.
2320  */
2321 mblk_t *
2322 icmp_prepend_hdr(conn_t *connp, ip_xmit_attr_t *ixa, const ip_pkt_t *ipp,
2323     const in6_addr_t *v6src, const in6_addr_t *v6dst, uint32_t flowinfo,
2324     mblk_t *data_mp, int *errorp)
2325 {
2326 	mblk_t		*mp;
2327 	icmp_stack_t	*is = connp->conn_netstack->netstack_icmp;
2328 	uint_t		data_len;
2329 	uint32_t	cksum;
2330 
2331 	data_len = msgdsize(data_mp);
2332 	mp = conn_prepend_hdr(ixa, ipp, v6src, v6dst, connp->conn_proto,
2333 	    flowinfo, 0, data_mp, data_len, is->is_wroff_extra, &cksum, errorp);
2334 	if (mp == NULL) {
2335 		ASSERT(*errorp != 0);
2336 		return (NULL);
2337 	}
2338 
2339 	ixa->ixa_pktlen = data_len + ixa->ixa_ip_hdr_length;
2340 
2341 	/*
2342 	 * If there was a routing option/header then conn_prepend_hdr
2343 	 * has massaged it and placed the pseudo-header checksum difference
2344 	 * in the cksum argument.
2345 	 *
2346 	 * Prepare for ICMPv6 checksum done in IP.
2347 	 *
2348 	 * We make it easy for IP to include our pseudo header
2349 	 * by putting our length (and any routing header adjustment)
2350 	 * in the ICMPv6 checksum field.
2351 	 * The IP source, destination, and length have already been set by
2352 	 * conn_prepend_hdr.
2353 	 */
2354 	cksum += data_len;
2355 	cksum = (cksum >> 16) + (cksum & 0xFFFF);
2356 	ASSERT(cksum < 0x10000);
2357 
2358 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
2359 		ipha_t	*ipha = (ipha_t *)mp->b_rptr;
2360 
2361 		ASSERT(ntohs(ipha->ipha_length) == ixa->ixa_pktlen);
2362 	} else {
2363 		ip6_t	*ip6h = (ip6_t *)mp->b_rptr;
2364 		uint_t	cksum_offset = 0;
2365 
2366 		ASSERT(ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN == ixa->ixa_pktlen);
2367 
2368 		if (ixa->ixa_flags & IXAF_SET_ULP_CKSUM) {
2369 			if (connp->conn_proto == IPPROTO_ICMPV6) {
2370 				cksum_offset = ixa->ixa_ip_hdr_length +
2371 				    offsetof(icmp6_t, icmp6_cksum);
2372 			} else if (ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
2373 				cksum_offset = ixa->ixa_ip_hdr_length +
2374 				    ixa->ixa_raw_cksum_offset;
2375 			}
2376 		}
2377 		if (cksum_offset != 0) {
2378 			uint16_t *ptr;
2379 
2380 			/* Make sure the checksum fits in the first mblk */
2381 			if (cksum_offset + sizeof (short) > MBLKL(mp)) {
2382 				mblk_t *mp1;
2383 
2384 				mp1 = msgpullup(mp,
2385 				    cksum_offset + sizeof (short));
2386 				freemsg(mp);
2387 				if (mp1 == NULL) {
2388 					*errorp = ENOMEM;
2389 					return (NULL);
2390 				}
2391 				mp = mp1;
2392 				ip6h = (ip6_t *)mp->b_rptr;
2393 			}
2394 			ptr = (uint16_t *)(mp->b_rptr + cksum_offset);
2395 			*ptr = htons(cksum);
2396 		}
2397 	}
2398 
2399 	/* Note that we don't try to update wroff due to ancillary data */
2400 	return (mp);
2401 }
2402 
2403 static int
2404 icmp_build_hdr_template(conn_t *connp, const in6_addr_t *v6src,
2405     const in6_addr_t *v6dst, uint32_t flowinfo)
2406 {
2407 	int		error;
2408 
2409 	ASSERT(MUTEX_HELD(&connp->conn_lock));
2410 	/*
2411 	 * We clear lastdst to make sure we don't use the lastdst path
2412 	 * next time sending since we might not have set v6dst yet.
2413 	 */
2414 	connp->conn_v6lastdst = ipv6_all_zeros;
2415 
2416 	error = conn_build_hdr_template(connp, 0, 0, v6src, v6dst, flowinfo);
2417 	if (error != 0)
2418 		return (error);
2419 
2420 	/*
2421 	 * Any routing header/option has been massaged. The checksum difference
2422 	 * is stored in conn_sum.
2423 	 */
2424 	return (0);
2425 }
2426 
2427 /*
2428  * This routine retrieves the value of an ND variable in a icmpparam_t
2429  * structure.  It is called through nd_getset when a user reads the
2430  * variable.
2431  */
2432 /* ARGSUSED */
2433 static int
2434 icmp_param_get(queue_t *q, mblk_t *mp, caddr_t cp, cred_t *cr)
2435 {
2436 	icmpparam_t	*icmppa = (icmpparam_t *)cp;
2437 
2438 	(void) mi_mpprintf(mp, "%d", icmppa->icmp_param_value);
2439 	return (0);
2440 }
2441 
2442 /*
2443  * Walk through the param array specified registering each element with the
2444  * named dispatch (ND) handler.
2445  */
2446 static boolean_t
2447 icmp_param_register(IDP *ndp, icmpparam_t *icmppa, int cnt)
2448 {
2449 	for (; cnt-- > 0; icmppa++) {
2450 		if (icmppa->icmp_param_name && icmppa->icmp_param_name[0]) {
2451 			if (!nd_load(ndp, icmppa->icmp_param_name,
2452 			    icmp_param_get, icmp_param_set,
2453 			    (caddr_t)icmppa)) {
2454 				nd_free(ndp);
2455 				return (B_FALSE);
2456 			}
2457 		}
2458 	}
2459 	return (B_TRUE);
2460 }
2461 
2462 /* This routine sets an ND variable in a icmpparam_t structure. */
2463 /* ARGSUSED */
2464 static int
2465 icmp_param_set(queue_t *q, mblk_t *mp, char *value, caddr_t cp, cred_t *cr)
2466 {
2467 	long		new_value;
2468 	icmpparam_t	*icmppa = (icmpparam_t *)cp;
2469 
2470 	/*
2471 	 * Fail the request if the new value does not lie within the
2472 	 * required bounds.
2473 	 */
2474 	if (ddi_strtol(value, NULL, 10, &new_value) != 0 ||
2475 	    new_value < icmppa->icmp_param_min ||
2476 	    new_value > icmppa->icmp_param_max) {
2477 		return (EINVAL);
2478 	}
2479 	/* Set the new value */
2480 	icmppa->icmp_param_value = new_value;
2481 	return (0);
2482 }
2483 
2484 static mblk_t *
2485 icmp_queue_fallback(icmp_t *icmp, mblk_t *mp)
2486 {
2487 	ASSERT(MUTEX_HELD(&icmp->icmp_recv_lock));
2488 	if (IPCL_IS_NONSTR(icmp->icmp_connp)) {
2489 		/*
2490 		 * fallback has started but messages have not been moved yet
2491 		 */
2492 		if (icmp->icmp_fallback_queue_head == NULL) {
2493 			ASSERT(icmp->icmp_fallback_queue_tail == NULL);
2494 			icmp->icmp_fallback_queue_head = mp;
2495 			icmp->icmp_fallback_queue_tail = mp;
2496 		} else {
2497 			ASSERT(icmp->icmp_fallback_queue_tail != NULL);
2498 			icmp->icmp_fallback_queue_tail->b_next = mp;
2499 			icmp->icmp_fallback_queue_tail = mp;
2500 		}
2501 		return (NULL);
2502 	} else {
2503 		/*
2504 		 * Fallback completed, let the caller putnext() the mblk.
2505 		 */
2506 		return (mp);
2507 	}
2508 }
2509 
2510 /*
2511  * Deliver data to ULP. In case we have a socket, and it's falling back to
2512  * TPI, then we'll queue the mp for later processing.
2513  */
2514 static void
2515 icmp_ulp_recv(conn_t *connp, mblk_t *mp, uint_t len)
2516 {
2517 	if (IPCL_IS_NONSTR(connp)) {
2518 		icmp_t *icmp = connp->conn_icmp;
2519 		int error;
2520 
2521 		ASSERT(len == msgdsize(mp));
2522 		if ((*connp->conn_upcalls->su_recv)
2523 		    (connp->conn_upper_handle, mp, len, 0, &error, NULL) < 0) {
2524 			mutex_enter(&icmp->icmp_recv_lock);
2525 			if (error == ENOSPC) {
2526 				/*
2527 				 * let's confirm while holding the lock
2528 				 */
2529 				if ((*connp->conn_upcalls->su_recv)
2530 				    (connp->conn_upper_handle, NULL, 0, 0,
2531 				    &error, NULL) < 0) {
2532 					ASSERT(error == ENOSPC);
2533 					if (error == ENOSPC) {
2534 						connp->conn_flow_cntrld =
2535 						    B_TRUE;
2536 					}
2537 				}
2538 				mutex_exit(&icmp->icmp_recv_lock);
2539 			} else {
2540 				ASSERT(error == EOPNOTSUPP);
2541 				mp = icmp_queue_fallback(icmp, mp);
2542 				mutex_exit(&icmp->icmp_recv_lock);
2543 				if (mp != NULL)
2544 					putnext(connp->conn_rq, mp);
2545 			}
2546 		}
2547 		ASSERT(MUTEX_NOT_HELD(&icmp->icmp_recv_lock));
2548 	} else {
2549 		putnext(connp->conn_rq, mp);
2550 	}
2551 }
2552 
2553 /*
2554  * This is the inbound data path.
2555  * IP has already pulled up the IP headers and verified alignment
2556  * etc.
2557  */
2558 /* ARGSUSED2 */
2559 static void
2560 icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
2561 {
2562 	conn_t			*connp = (conn_t *)arg1;
2563 	struct T_unitdata_ind	*tudi;
2564 	uchar_t			*rptr;		/* Pointer to IP header */
2565 	int			ip_hdr_length;
2566 	int			udi_size;	/* Size of T_unitdata_ind */
2567 	int			pkt_len;
2568 	icmp_t			*icmp;
2569 	ip_pkt_t		ipps;
2570 	ip6_t			*ip6h;
2571 	mblk_t			*mp1;
2572 	crb_t			recv_ancillary;
2573 	icmp_stack_t		*is;
2574 	sin_t			*sin;
2575 	sin6_t			*sin6;
2576 	ipha_t			*ipha;
2577 
2578 	ASSERT(connp->conn_flags & IPCL_RAWIPCONN);
2579 
2580 	icmp = connp->conn_icmp;
2581 	is = icmp->icmp_is;
2582 	rptr = mp->b_rptr;
2583 
2584 	ASSERT(DB_TYPE(mp) == M_DATA);
2585 	ASSERT(OK_32PTR(rptr));
2586 	ASSERT(ira->ira_pktlen == msgdsize(mp));
2587 	pkt_len = ira->ira_pktlen;
2588 
2589 	/*
2590 	 * Get a snapshot of these and allow other threads to change
2591 	 * them after that. We need the same recv_ancillary when determining
2592 	 * the size as when adding the ancillary data items.
2593 	 */
2594 	mutex_enter(&connp->conn_lock);
2595 	recv_ancillary = connp->conn_recv_ancillary;
2596 	mutex_exit(&connp->conn_lock);
2597 
2598 	ip_hdr_length = ira->ira_ip_hdr_length;
2599 	ASSERT(MBLKL(mp) >= ip_hdr_length);	/* IP did a pullup */
2600 
2601 	/* Initialize regardless of IP version */
2602 	ipps.ipp_fields = 0;
2603 
2604 	if (ira->ira_flags & IRAF_IS_IPV4) {
2605 		ASSERT(IPH_HDR_VERSION(rptr) == IPV4_VERSION);
2606 		ASSERT(MBLKL(mp) >= sizeof (ipha_t));
2607 		ASSERT(ira->ira_ip_hdr_length == IPH_HDR_LENGTH(rptr));
2608 
2609 		ipha = (ipha_t *)mp->b_rptr;
2610 		if (recv_ancillary.crb_all != 0)
2611 			(void) ip_find_hdr_v4(ipha, &ipps, B_FALSE);
2612 
2613 		/*
2614 		 * BSD for some reason adjusts ipha_length to exclude the
2615 		 * IP header length. We do the same.
2616 		 */
2617 		if (is->is_bsd_compat) {
2618 			ushort_t len;
2619 
2620 			len = ntohs(ipha->ipha_length);
2621 			if (mp->b_datap->db_ref > 1) {
2622 				/*
2623 				 * Allocate a new IP header so that we can
2624 				 * modify ipha_length.
2625 				 */
2626 				mblk_t	*mp1;
2627 
2628 				mp1 = allocb(ip_hdr_length, BPRI_MED);
2629 				if (mp1 == NULL) {
2630 					freemsg(mp);
2631 					BUMP_MIB(&is->is_rawip_mib,
2632 					    rawipInErrors);
2633 					return;
2634 				}
2635 				bcopy(rptr, mp1->b_rptr, ip_hdr_length);
2636 				mp->b_rptr = rptr + ip_hdr_length;
2637 				rptr = mp1->b_rptr;
2638 				ipha = (ipha_t *)rptr;
2639 				mp1->b_cont = mp;
2640 				mp1->b_wptr = rptr + ip_hdr_length;
2641 				mp = mp1;
2642 			}
2643 			len -= ip_hdr_length;
2644 			ipha->ipha_length = htons(len);
2645 		}
2646 
2647 		/*
2648 		 * For RAW sockets we not pass ICMP/IPv4 packets to AF_INET6
2649 		 * sockets. This is ensured by icmp_bind and the IP fanout code.
2650 		 */
2651 		ASSERT(connp->conn_family == AF_INET);
2652 
2653 		/*
2654 		 * This is the inbound data path.  Packets are passed upstream
2655 		 * as T_UNITDATA_IND messages with full IPv4 headers still
2656 		 * attached.
2657 		 */
2658 
2659 		/*
2660 		 * Normally only send up the source address.
2661 		 * If any ancillary data items are wanted we add those.
2662 		 */
2663 		udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin_t);
2664 		if (recv_ancillary.crb_all != 0) {
2665 			udi_size += conn_recvancillary_size(connp,
2666 			    recv_ancillary, ira, mp, &ipps);
2667 		}
2668 
2669 		/* Allocate a message block for the T_UNITDATA_IND structure. */
2670 		mp1 = allocb(udi_size, BPRI_MED);
2671 		if (mp1 == NULL) {
2672 			freemsg(mp);
2673 			BUMP_MIB(&is->is_rawip_mib, rawipInErrors);
2674 			return;
2675 		}
2676 		mp1->b_cont = mp;
2677 		tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2678 		mp1->b_datap->db_type = M_PROTO;
2679 		mp1->b_wptr = (uchar_t *)tudi + udi_size;
2680 		tudi->PRIM_type = T_UNITDATA_IND;
2681 		tudi->SRC_length = sizeof (sin_t);
2682 		tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2683 		sin = (sin_t *)&tudi[1];
2684 		*sin = sin_null;
2685 		sin->sin_family = AF_INET;
2686 		sin->sin_addr.s_addr = ipha->ipha_src;
2687 		*(uint32_t *)&sin->sin_zero[0] = 0;
2688 		*(uint32_t *)&sin->sin_zero[4] = 0;
2689 		tudi->OPT_offset =  sizeof (struct T_unitdata_ind) +
2690 		    sizeof (sin_t);
2691 		udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin_t));
2692 		tudi->OPT_length = udi_size;
2693 
2694 		/*
2695 		 * Add options if IP_RECVIF etc is set
2696 		 */
2697 		if (udi_size != 0) {
2698 			conn_recvancillary_add(connp, recv_ancillary, ira,
2699 			    &ipps, (uchar_t *)&sin[1], udi_size);
2700 		}
2701 		goto deliver;
2702 	}
2703 
2704 	ASSERT(IPH_HDR_VERSION(rptr) == IPV6_VERSION);
2705 	/*
2706 	 * IPv6 packets can only be received by applications
2707 	 * that are prepared to receive IPv6 addresses.
2708 	 * The IP fanout must ensure this.
2709 	 */
2710 	ASSERT(connp->conn_family == AF_INET6);
2711 
2712 	/*
2713 	 * Handle IPv6 packets. We don't pass up the IP headers with the
2714 	 * payload for IPv6.
2715 	 */
2716 
2717 	ip6h = (ip6_t *)rptr;
2718 	if (recv_ancillary.crb_all != 0) {
2719 		/*
2720 		 * Call on ip_find_hdr_v6 which gets individual lenghts of
2721 		 * extension headers (and pointers to them).
2722 		 */
2723 		uint8_t		nexthdr;
2724 
2725 		/* We don't care about the length or nextheader. */
2726 		(void) ip_find_hdr_v6(mp, ip6h, B_TRUE, &ipps, &nexthdr);
2727 
2728 		/*
2729 		 * We do not pass up hop-by-hop options or any other
2730 		 * extension header as part of the packet. Applications
2731 		 * that want to see them have to specify IPV6_RECV* socket
2732 		 * options. And conn_recvancillary_size/add explicitly
2733 		 * drops the TX option from IPV6_HOPOPTS as it does for UDP.
2734 		 *
2735 		 * If we had multilevel ICMP sockets, then we'd want to
2736 		 * modify conn_recvancillary_size/add to
2737 		 * allow the user to see the label.
2738 		 */
2739 	}
2740 
2741 	/*
2742 	 * Check a filter for ICMPv6 types if needed.
2743 	 * Verify raw checksums if needed.
2744 	 */
2745 	mutex_enter(&connp->conn_lock);
2746 	if (icmp->icmp_filter != NULL) {
2747 		int type;
2748 
2749 		/* Assumes that IP has done the pullupmsg */
2750 		type = mp->b_rptr[ip_hdr_length];
2751 
2752 		ASSERT(mp->b_rptr + ip_hdr_length <= mp->b_wptr);
2753 		if (ICMP6_FILTER_WILLBLOCK(type, icmp->icmp_filter)) {
2754 			mutex_exit(&connp->conn_lock);
2755 			freemsg(mp);
2756 			return;
2757 		}
2758 	}
2759 	if (connp->conn_ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
2760 		/* Checksum */
2761 		uint16_t	*up;
2762 		uint32_t	sum;
2763 		int		remlen;
2764 
2765 		up = (uint16_t *)&ip6h->ip6_src;
2766 
2767 		remlen = msgdsize(mp) - ip_hdr_length;
2768 		sum = htons(connp->conn_proto + remlen)
2769 		    + up[0] + up[1] + up[2] + up[3]
2770 		    + up[4] + up[5] + up[6] + up[7]
2771 		    + up[8] + up[9] + up[10] + up[11]
2772 		    + up[12] + up[13] + up[14] + up[15];
2773 		sum = (sum & 0xffff) + (sum >> 16);
2774 		sum = IP_CSUM(mp, ip_hdr_length, sum);
2775 		if (sum != 0) {
2776 			/* IPv6 RAW checksum failed */
2777 			ip0dbg(("icmp_rput: RAW checksum failed %x\n", sum));
2778 			mutex_exit(&connp->conn_lock);
2779 			freemsg(mp);
2780 			BUMP_MIB(&is->is_rawip_mib, rawipInCksumErrs);
2781 			return;
2782 		}
2783 	}
2784 	mutex_exit(&connp->conn_lock);
2785 
2786 	udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t);
2787 
2788 	if (recv_ancillary.crb_all != 0) {
2789 		udi_size += conn_recvancillary_size(connp,
2790 		    recv_ancillary, ira, mp, &ipps);
2791 	}
2792 
2793 	mp1 = allocb(udi_size, BPRI_MED);
2794 	if (mp1 == NULL) {
2795 		freemsg(mp);
2796 		BUMP_MIB(&is->is_rawip_mib, rawipInErrors);
2797 		return;
2798 	}
2799 	mp1->b_cont = mp;
2800 	mp1->b_datap->db_type = M_PROTO;
2801 	tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2802 	mp1->b_wptr = (uchar_t *)tudi + udi_size;
2803 	tudi->PRIM_type = T_UNITDATA_IND;
2804 	tudi->SRC_length = sizeof (sin6_t);
2805 	tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2806 	tudi->OPT_offset = sizeof (struct T_unitdata_ind) + sizeof (sin6_t);
2807 	udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin6_t));
2808 	tudi->OPT_length = udi_size;
2809 	sin6 = (sin6_t *)&tudi[1];
2810 	*sin6 = sin6_null;
2811 	sin6->sin6_port = 0;
2812 	sin6->sin6_family = AF_INET6;
2813 
2814 	sin6->sin6_addr = ip6h->ip6_src;
2815 	/* No sin6_flowinfo per API */
2816 	sin6->sin6_flowinfo = 0;
2817 	/* For link-scope pass up scope id */
2818 	if (IN6_IS_ADDR_LINKSCOPE(&ip6h->ip6_src))
2819 		sin6->sin6_scope_id = ira->ira_ruifindex;
2820 	else
2821 		sin6->sin6_scope_id = 0;
2822 	sin6->__sin6_src_id = ip_srcid_find_addr(&ip6h->ip6_dst,
2823 	    IPCL_ZONEID(connp), is->is_netstack);
2824 
2825 	if (udi_size != 0) {
2826 		conn_recvancillary_add(connp, recv_ancillary, ira,
2827 		    &ipps, (uchar_t *)&sin6[1], udi_size);
2828 	}
2829 
2830 	/* Skip all the IPv6 headers per API */
2831 	mp->b_rptr += ip_hdr_length;
2832 	pkt_len -= ip_hdr_length;
2833 
2834 deliver:
2835 	BUMP_MIB(&is->is_rawip_mib, rawipInDatagrams);
2836 	icmp_ulp_recv(connp, mp1, pkt_len);
2837 }
2838 
2839 /*
2840  * return SNMP stuff in buffer in mpdata. We don't hold any lock and report
2841  * information that can be changing beneath us.
2842  */
2843 mblk_t *
2844 icmp_snmp_get(queue_t *q, mblk_t *mpctl)
2845 {
2846 	mblk_t			*mpdata;
2847 	struct opthdr		*optp;
2848 	conn_t			*connp = Q_TO_CONN(q);
2849 	icmp_stack_t		*is = connp->conn_netstack->netstack_icmp;
2850 	mblk_t			*mp2ctl;
2851 
2852 	/*
2853 	 * make a copy of the original message
2854 	 */
2855 	mp2ctl = copymsg(mpctl);
2856 
2857 	if (mpctl == NULL ||
2858 	    (mpdata = mpctl->b_cont) == NULL) {
2859 		freemsg(mpctl);
2860 		freemsg(mp2ctl);
2861 		return (0);
2862 	}
2863 
2864 	/* fixed length structure for IPv4 and IPv6 counters */
2865 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
2866 	optp->level = EXPER_RAWIP;
2867 	optp->name = 0;
2868 	(void) snmp_append_data(mpdata, (char *)&is->is_rawip_mib,
2869 	    sizeof (is->is_rawip_mib));
2870 	optp->len = msgdsize(mpdata);
2871 	qreply(q, mpctl);
2872 
2873 	return (mp2ctl);
2874 }
2875 
2876 /*
2877  * Return 0 if invalid set request, 1 otherwise, including non-rawip requests.
2878  * TODO:  If this ever actually tries to set anything, it needs to be
2879  * to do the appropriate locking.
2880  */
2881 /* ARGSUSED */
2882 int
2883 icmp_snmp_set(queue_t *q, t_scalar_t level, t_scalar_t name,
2884     uchar_t *ptr, int len)
2885 {
2886 	switch (level) {
2887 	case EXPER_RAWIP:
2888 		return (0);
2889 	default:
2890 		return (1);
2891 	}
2892 }
2893 
2894 /*
2895  * This routine creates a T_UDERROR_IND message and passes it upstream.
2896  * The address and options are copied from the T_UNITDATA_REQ message
2897  * passed in mp.  This message is freed.
2898  */
2899 static void
2900 icmp_ud_err(queue_t *q, mblk_t *mp, t_scalar_t err)
2901 {
2902 	struct T_unitdata_req *tudr;
2903 	mblk_t	*mp1;
2904 	uchar_t *destaddr;
2905 	t_scalar_t destlen;
2906 	uchar_t	*optaddr;
2907 	t_scalar_t optlen;
2908 
2909 	if ((mp->b_wptr < mp->b_rptr) ||
2910 	    (MBLKL(mp)) < sizeof (struct T_unitdata_req)) {
2911 		goto done;
2912 	}
2913 	tudr = (struct T_unitdata_req *)mp->b_rptr;
2914 	destaddr = mp->b_rptr + tudr->DEST_offset;
2915 	if (destaddr < mp->b_rptr || destaddr >= mp->b_wptr ||
2916 	    destaddr + tudr->DEST_length < mp->b_rptr ||
2917 	    destaddr + tudr->DEST_length > mp->b_wptr) {
2918 		goto done;
2919 	}
2920 	optaddr = mp->b_rptr + tudr->OPT_offset;
2921 	if (optaddr < mp->b_rptr || optaddr >= mp->b_wptr ||
2922 	    optaddr + tudr->OPT_length < mp->b_rptr ||
2923 	    optaddr + tudr->OPT_length > mp->b_wptr) {
2924 		goto done;
2925 	}
2926 	destlen = tudr->DEST_length;
2927 	optlen = tudr->OPT_length;
2928 
2929 	mp1 = mi_tpi_uderror_ind((char *)destaddr, destlen,
2930 	    (char *)optaddr, optlen, err);
2931 	if (mp1 != NULL)
2932 		qreply(q, mp1);
2933 
2934 done:
2935 	freemsg(mp);
2936 }
2937 
2938 static int
2939 rawip_do_unbind(conn_t *connp)
2940 {
2941 	icmp_t	*icmp = connp->conn_icmp;
2942 
2943 	mutex_enter(&connp->conn_lock);
2944 	/* If a bind has not been done, we can't unbind. */
2945 	if (icmp->icmp_state == TS_UNBND) {
2946 		mutex_exit(&connp->conn_lock);
2947 		return (-TOUTSTATE);
2948 	}
2949 	connp->conn_saddr_v6 = ipv6_all_zeros;
2950 	connp->conn_bound_addr_v6 = ipv6_all_zeros;
2951 	connp->conn_laddr_v6 = ipv6_all_zeros;
2952 	connp->conn_mcbc_bind = B_FALSE;
2953 	connp->conn_lport = 0;
2954 	connp->conn_fport = 0;
2955 	/* In case we were also connected */
2956 	connp->conn_faddr_v6 = ipv6_all_zeros;
2957 	connp->conn_v6lastdst = ipv6_all_zeros;
2958 
2959 	icmp->icmp_state = TS_UNBND;
2960 
2961 	(void) icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
2962 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
2963 	mutex_exit(&connp->conn_lock);
2964 
2965 	ip_unbind(connp);
2966 	return (0);
2967 }
2968 
2969 /*
2970  * This routine is called by icmp_wput to handle T_UNBIND_REQ messages.
2971  * After some error checking, the message is passed downstream to ip.
2972  */
2973 static void
2974 icmp_tpi_unbind(queue_t *q, mblk_t *mp)
2975 {
2976 	conn_t	*connp = Q_TO_CONN(q);
2977 	int	error;
2978 
2979 	ASSERT(mp->b_cont == NULL);
2980 	error = rawip_do_unbind(connp);
2981 	if (error) {
2982 		if (error < 0) {
2983 			icmp_err_ack(q, mp, -error, 0);
2984 		} else {
2985 			icmp_err_ack(q, mp, 0, error);
2986 		}
2987 		return;
2988 	}
2989 
2990 	/*
2991 	 * Convert mp into a T_OK_ACK
2992 	 */
2993 
2994 	mp = mi_tpi_ok_ack_alloc(mp);
2995 
2996 	/*
2997 	 * should not happen in practice... T_OK_ACK is smaller than the
2998 	 * original message.
2999 	 */
3000 	ASSERT(mp != NULL);
3001 	ASSERT(((struct T_ok_ack *)mp->b_rptr)->PRIM_type == T_OK_ACK);
3002 	qreply(q, mp);
3003 }
3004 
3005 /*
3006  * Process IPv4 packets that already include an IP header.
3007  * Used when IP_HDRINCL has been set (implicit for IPPROTO_RAW and
3008  * IPPROTO_IGMP).
3009  * In this case we ignore the address and any options in the T_UNITDATA_REQ.
3010  *
3011  * The packet is assumed to have a base (20 byte) IP header followed
3012  * by the upper-layer protocol. We include any IP_OPTIONS including a
3013  * CIPSO label but otherwise preserve the base IP header.
3014  */
3015 static int
3016 icmp_output_hdrincl(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid)
3017 {
3018 	icmp_t		*icmp = connp->conn_icmp;
3019 	icmp_stack_t	*is = icmp->icmp_is;
3020 	ipha_t		iphas;
3021 	ipha_t		*ipha;
3022 	int		ip_hdr_length;
3023 	int		tp_hdr_len;
3024 	ip_xmit_attr_t	*ixa;
3025 	ip_pkt_t	*ipp;
3026 	in6_addr_t	v6src;
3027 	in6_addr_t	v6dst;
3028 	in6_addr_t	v6nexthop;
3029 	int		error;
3030 	boolean_t	do_ipsec;
3031 
3032 	/*
3033 	 * We need an exclusive copy of conn_ixa since the included IP
3034 	 * header could have any destination.
3035 	 * That copy has no pointers hence we
3036 	 * need to set them up once we've parsed the ancillary data.
3037 	 */
3038 	ixa = conn_get_ixa_exclusive(connp);
3039 	if (ixa == NULL) {
3040 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3041 		freemsg(mp);
3042 		return (ENOMEM);
3043 	}
3044 	ASSERT(cr != NULL);
3045 	/*
3046 	 * Caller has a reference on cr; from db_credp or because we
3047 	 * are running in process context.
3048 	 */
3049 	ixa->ixa_cred = cr;
3050 	ixa->ixa_cpid = pid;
3051 	if (is_system_labeled()) {
3052 		/* We need to restart with a label based on the cred */
3053 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
3054 	}
3055 
3056 	/* In case previous destination was multicast or multirt */
3057 	ip_attr_newdst(ixa);
3058 
3059 	/* Get a copy of conn_xmit_ipp since the TX label might change it */
3060 	ipp = kmem_zalloc(sizeof (*ipp), KM_NOSLEEP);
3061 	if (ipp == NULL) {
3062 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3063 		ixa->ixa_cpid = connp->conn_cpid;
3064 		ixa_refrele(ixa);
3065 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3066 		freemsg(mp);
3067 		return (ENOMEM);
3068 	}
3069 	mutex_enter(&connp->conn_lock);
3070 	error = ip_pkt_copy(&connp->conn_xmit_ipp, ipp, KM_NOSLEEP);
3071 	mutex_exit(&connp->conn_lock);
3072 	if (error != 0) {
3073 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3074 		freemsg(mp);
3075 		goto done;
3076 	}
3077 
3078 	/* Sanity check length of packet */
3079 	ipha = (ipha_t *)mp->b_rptr;
3080 
3081 	ip_hdr_length = IP_SIMPLE_HDR_LENGTH;
3082 	if ((mp->b_wptr - mp->b_rptr) < IP_SIMPLE_HDR_LENGTH) {
3083 		if (!pullupmsg(mp, IP_SIMPLE_HDR_LENGTH)) {
3084 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3085 			freemsg(mp);
3086 			goto done;
3087 		}
3088 		ipha = (ipha_t *)mp->b_rptr;
3089 	}
3090 	ipha->ipha_version_and_hdr_length =
3091 	    (IP_VERSION<<4) | (ip_hdr_length>>2);
3092 
3093 	/*
3094 	 * We set IXAF_DONTFRAG if the application set DF which makes
3095 	 * IP not fragment.
3096 	 */
3097 	ipha->ipha_fragment_offset_and_flags &= htons(IPH_DF);
3098 	if (ipha->ipha_fragment_offset_and_flags & htons(IPH_DF))
3099 		ixa->ixa_flags |= (IXAF_DONTFRAG | IXAF_PMTU_IPV4_DF);
3100 	else
3101 		ixa->ixa_flags &= ~(IXAF_DONTFRAG | IXAF_PMTU_IPV4_DF);
3102 
3103 	/* Even for multicast and broadcast we honor the apps ttl */
3104 	ixa->ixa_flags |= IXAF_NO_TTL_CHANGE;
3105 
3106 	/*
3107 	 * No source verification for non-local addresses
3108 	 */
3109 	if (ipha->ipha_src != INADDR_ANY &&
3110 	    ip_laddr_verify_v4(ipha->ipha_src, ixa->ixa_zoneid,
3111 	    is->is_netstack->netstack_ip, B_FALSE)
3112 	    != IPVL_UNICAST_UP) {
3113 		ixa->ixa_flags &= ~IXAF_VERIFY_SOURCE;
3114 	}
3115 
3116 	if (ipha->ipha_dst == INADDR_ANY)
3117 		ipha->ipha_dst = htonl(INADDR_LOOPBACK);
3118 
3119 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6src);
3120 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &v6dst);
3121 
3122 	/* Defer IPsec if it might need to look at ICMP type/code */
3123 	do_ipsec = ipha->ipha_protocol != IPPROTO_ICMP;
3124 	ixa->ixa_flags |= IXAF_IS_IPV4;
3125 
3126 	ip_attr_nexthop(ipp, ixa, &v6dst, &v6nexthop);
3127 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop,
3128 	    connp->conn_fport, &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
3129 	    (do_ipsec ? IPDF_IPSEC : 0));
3130 	switch (error) {
3131 	case 0:
3132 		break;
3133 	case EADDRNOTAVAIL:
3134 		/*
3135 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3136 		 * Don't have the application see that errno
3137 		 */
3138 		error = ENETUNREACH;
3139 		goto failed;
3140 	case ENETDOWN:
3141 		/*
3142 		 * Have !ipif_addr_ready address; drop packet silently
3143 		 * until we can get applications to not send until we
3144 		 * are ready.
3145 		 */
3146 		error = 0;
3147 		goto failed;
3148 	case EHOSTUNREACH:
3149 	case ENETUNREACH:
3150 		if (ixa->ixa_ire != NULL) {
3151 			/*
3152 			 * Let conn_ip_output/ire_send_noroute return
3153 			 * the error and send any local ICMP error.
3154 			 */
3155 			error = 0;
3156 			break;
3157 		}
3158 		/* FALLTHRU */
3159 	default:
3160 	failed:
3161 		freemsg(mp);
3162 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3163 		goto done;
3164 	}
3165 	if (ipha->ipha_src == INADDR_ANY)
3166 		IN6_V4MAPPED_TO_IPADDR(&v6src, ipha->ipha_src);
3167 
3168 	/*
3169 	 * We might be going to a different destination than last time,
3170 	 * thus check that TX allows the communication and compute any
3171 	 * needed label.
3172 	 *
3173 	 * TSOL Note: We have an exclusive ipp and ixa for this thread so we
3174 	 * don't have to worry about concurrent threads.
3175 	 */
3176 	if (is_system_labeled()) {
3177 		/*
3178 		 * Check whether Trusted Solaris policy allows communication
3179 		 * with this host, and pretend that the destination is
3180 		 * unreachable if not.
3181 		 * Compute any needed label and place it in ipp_label_v4/v6.
3182 		 *
3183 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
3184 		 * ipp_label_v4/v6 to form the packet.
3185 		 *
3186 		 * Tsol note: We have ipp structure local to this thread so
3187 		 * no locking is needed.
3188 		 */
3189 		error = conn_update_label(connp, ixa, &v6dst, ipp);
3190 		if (error != 0) {
3191 			freemsg(mp);
3192 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3193 			goto done;
3194 		}
3195 	}
3196 
3197 	/*
3198 	 * Save away a copy of the IPv4 header the application passed down
3199 	 * and then prepend an IPv4 header complete with any IP options
3200 	 * including label.
3201 	 * We need a struct copy since icmp_prepend_hdr will reuse the available
3202 	 * space in the mblk.
3203 	 */
3204 	iphas = *ipha;
3205 	mp->b_rptr += IP_SIMPLE_HDR_LENGTH;
3206 
3207 	mp = icmp_prepend_hdr(connp, ixa, ipp, &v6src, &v6dst, 0, mp, &error);
3208 	if (mp == NULL) {
3209 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3210 		ASSERT(error != 0);
3211 		goto done;
3212 	}
3213 	if (ixa->ixa_pktlen > IP_MAXPACKET) {
3214 		error = EMSGSIZE;
3215 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3216 		freemsg(mp);
3217 		goto done;
3218 	}
3219 	/* Restore key parts of the header that the application passed down */
3220 	ipha = (ipha_t *)mp->b_rptr;
3221 	ipha->ipha_type_of_service = iphas.ipha_type_of_service;
3222 	ipha->ipha_ident = iphas.ipha_ident;
3223 	ipha->ipha_fragment_offset_and_flags =
3224 	    iphas.ipha_fragment_offset_and_flags;
3225 	ipha->ipha_ttl = iphas.ipha_ttl;
3226 	ipha->ipha_protocol = iphas.ipha_protocol;
3227 	ipha->ipha_src = iphas.ipha_src;
3228 	ipha->ipha_dst = iphas.ipha_dst;
3229 
3230 	ixa->ixa_protocol = ipha->ipha_protocol;
3231 
3232 	/*
3233 	 * Make sure that the IP header plus any transport header that is
3234 	 * checksumed by ip_output is in the first mblk. (ip_output assumes
3235 	 * that at least the checksum field is in the first mblk.)
3236 	 */
3237 	switch (ipha->ipha_protocol) {
3238 	case IPPROTO_UDP:
3239 		tp_hdr_len = 8;
3240 		break;
3241 	case IPPROTO_TCP:
3242 		tp_hdr_len = 20;
3243 		break;
3244 	default:
3245 		tp_hdr_len = 0;
3246 		break;
3247 	}
3248 	ip_hdr_length = IPH_HDR_LENGTH(ipha);
3249 	if (mp->b_wptr - mp->b_rptr < ip_hdr_length + tp_hdr_len) {
3250 		if (!pullupmsg(mp, ip_hdr_length + tp_hdr_len)) {
3251 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3252 			if (mp->b_cont == NULL)
3253 				error = EINVAL;
3254 			else
3255 				error = ENOMEM;
3256 			freemsg(mp);
3257 			goto done;
3258 		}
3259 	}
3260 
3261 	if (!do_ipsec) {
3262 		/* Policy might differ for different ICMP type/code */
3263 		if (ixa->ixa_ipsec_policy != NULL) {
3264 			IPPOL_REFRELE(ixa->ixa_ipsec_policy);
3265 			ixa->ixa_ipsec_policy = NULL;
3266 			ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
3267 		}
3268 		mp = ip_output_attach_policy(mp, ipha, NULL, connp, ixa);
3269 		if (mp == NULL) {
3270 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3271 			error = EHOSTUNREACH;	/* IPsec policy failure */
3272 			goto done;
3273 		}
3274 	}
3275 
3276 	/* We're done.  Pass the packet to ip. */
3277 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3278 
3279 	error = conn_ip_output(mp, ixa);
3280 	/* No rawipOutErrors if an error since IP increases its error counter */
3281 	switch (error) {
3282 	case 0:
3283 		break;
3284 	case EWOULDBLOCK:
3285 		(void) ixa_check_drain_insert(connp, ixa);
3286 		error = 0;
3287 		break;
3288 	case EADDRNOTAVAIL:
3289 		/*
3290 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3291 		 * Don't have the application see that errno
3292 		 */
3293 		error = ENETUNREACH;
3294 		break;
3295 	}
3296 done:
3297 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3298 	ixa->ixa_cpid = connp->conn_cpid;
3299 	ixa_refrele(ixa);
3300 	ip_pkt_free(ipp);
3301 	kmem_free(ipp, sizeof (*ipp));
3302 	return (error);
3303 }
3304 
3305 static mblk_t *
3306 icmp_output_attach_policy(mblk_t *mp, conn_t *connp, ip_xmit_attr_t *ixa)
3307 {
3308 	ipha_t	*ipha = NULL;
3309 	ip6_t	*ip6h = NULL;
3310 
3311 	if (ixa->ixa_flags & IXAF_IS_IPV4)
3312 		ipha = (ipha_t *)mp->b_rptr;
3313 	else
3314 		ip6h = (ip6_t *)mp->b_rptr;
3315 
3316 	if (ixa->ixa_ipsec_policy != NULL) {
3317 		IPPOL_REFRELE(ixa->ixa_ipsec_policy);
3318 		ixa->ixa_ipsec_policy = NULL;
3319 		ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
3320 	}
3321 	return (ip_output_attach_policy(mp, ipha, ip6h, connp, ixa));
3322 }
3323 
3324 /*
3325  * Handle T_UNITDATA_REQ with options. Both IPv4 and IPv6
3326  * Either tudr_mp or msg is set. If tudr_mp we take ancillary data from
3327  * the TPI options, otherwise we take them from msg_control.
3328  * If both sin and sin6 is set it is a connected socket and we use conn_faddr.
3329  * Always consumes mp; never consumes tudr_mp.
3330  */
3331 static int
3332 icmp_output_ancillary(conn_t *connp, sin_t *sin, sin6_t *sin6, mblk_t *mp,
3333     mblk_t *tudr_mp, struct nmsghdr *msg, cred_t *cr, pid_t pid)
3334 {
3335 	icmp_t		*icmp = connp->conn_icmp;
3336 	icmp_stack_t	*is = icmp->icmp_is;
3337 	int		error;
3338 	ip_xmit_attr_t	*ixa;
3339 	ip_pkt_t	*ipp;
3340 	in6_addr_t	v6src;
3341 	in6_addr_t	v6dst;
3342 	in6_addr_t	v6nexthop;
3343 	in_port_t	dstport;
3344 	uint32_t	flowinfo;
3345 	uint_t		srcid;
3346 	int		is_absreq_failure = 0;
3347 	conn_opt_arg_t	coas, *coa;
3348 
3349 	ASSERT(tudr_mp != NULL || msg != NULL);
3350 
3351 	/*
3352 	 * Get ixa before checking state to handle a disconnect race.
3353 	 *
3354 	 * We need an exclusive copy of conn_ixa since the ancillary data
3355 	 * options might modify it. That copy has no pointers hence we
3356 	 * need to set them up once we've parsed the ancillary data.
3357 	 */
3358 	ixa = conn_get_ixa_exclusive(connp);
3359 	if (ixa == NULL) {
3360 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3361 		freemsg(mp);
3362 		return (ENOMEM);
3363 	}
3364 	ASSERT(cr != NULL);
3365 	ixa->ixa_cred = cr;
3366 	ixa->ixa_cpid = pid;
3367 	if (is_system_labeled()) {
3368 		/* We need to restart with a label based on the cred */
3369 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
3370 	}
3371 
3372 	/* In case previous destination was multicast or multirt */
3373 	ip_attr_newdst(ixa);
3374 
3375 	/* Get a copy of conn_xmit_ipp since the options might change it */
3376 	ipp = kmem_zalloc(sizeof (*ipp), KM_NOSLEEP);
3377 	if (ipp == NULL) {
3378 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3379 		ixa->ixa_cpid = connp->conn_cpid;
3380 		ixa_refrele(ixa);
3381 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3382 		freemsg(mp);
3383 		return (ENOMEM);
3384 	}
3385 	mutex_enter(&connp->conn_lock);
3386 	error = ip_pkt_copy(&connp->conn_xmit_ipp, ipp, KM_NOSLEEP);
3387 	mutex_exit(&connp->conn_lock);
3388 	if (error != 0) {
3389 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3390 		freemsg(mp);
3391 		goto done;
3392 	}
3393 
3394 	/*
3395 	 * Parse the options and update ixa and ipp as a result.
3396 	 */
3397 
3398 	coa = &coas;
3399 	coa->coa_connp = connp;
3400 	coa->coa_ixa = ixa;
3401 	coa->coa_ipp = ipp;
3402 	coa->coa_ancillary = B_TRUE;
3403 	coa->coa_changed = 0;
3404 
3405 	if (msg != NULL) {
3406 		error = process_auxiliary_options(connp, msg->msg_control,
3407 		    msg->msg_controllen, coa, &icmp_opt_obj, icmp_opt_set, cr);
3408 	} else {
3409 		struct T_unitdata_req *tudr;
3410 
3411 		tudr = (struct T_unitdata_req *)tudr_mp->b_rptr;
3412 		ASSERT(tudr->PRIM_type == T_UNITDATA_REQ);
3413 		error = tpi_optcom_buf(connp->conn_wq, tudr_mp,
3414 		    &tudr->OPT_length, tudr->OPT_offset, cr, &icmp_opt_obj,
3415 		    coa, &is_absreq_failure);
3416 	}
3417 	if (error != 0) {
3418 		/*
3419 		 * Note: No special action needed in this
3420 		 * module for "is_absreq_failure"
3421 		 */
3422 		freemsg(mp);
3423 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3424 		goto done;
3425 	}
3426 	ASSERT(is_absreq_failure == 0);
3427 
3428 	mutex_enter(&connp->conn_lock);
3429 	/*
3430 	 * If laddr is unspecified then we look at sin6_src_id.
3431 	 * We will give precedence to a source address set with IPV6_PKTINFO
3432 	 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
3433 	 * want ip_attr_connect to select a source (since it can fail) when
3434 	 * IPV6_PKTINFO is specified.
3435 	 * If this doesn't result in a source address then we get a source
3436 	 * from ip_attr_connect() below.
3437 	 */
3438 	v6src = connp->conn_saddr_v6;
3439 	if (sin != NULL) {
3440 		IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
3441 		dstport = sin->sin_port;
3442 		flowinfo = 0;
3443 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3444 		ixa->ixa_flags |= IXAF_IS_IPV4;
3445 	} else if (sin6 != NULL) {
3446 		v6dst = sin6->sin6_addr;
3447 		dstport = sin6->sin6_port;
3448 		flowinfo = sin6->sin6_flowinfo;
3449 		srcid = sin6->__sin6_src_id;
3450 		if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
3451 			ixa->ixa_scopeid = sin6->sin6_scope_id;
3452 			ixa->ixa_flags |= IXAF_SCOPEID_SET;
3453 		} else {
3454 			ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3455 		}
3456 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
3457 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
3458 			    connp->conn_netstack);
3459 		}
3460 		if (IN6_IS_ADDR_V4MAPPED(&v6dst))
3461 			ixa->ixa_flags |= IXAF_IS_IPV4;
3462 		else
3463 			ixa->ixa_flags &= ~IXAF_IS_IPV4;
3464 	} else {
3465 		/* Connected case */
3466 		v6dst = connp->conn_faddr_v6;
3467 		flowinfo = connp->conn_flowinfo;
3468 	}
3469 	mutex_exit(&connp->conn_lock);
3470 	/* Handle IPV6_PKTINFO setting source address. */
3471 	if (IN6_IS_ADDR_UNSPECIFIED(&v6src) &&
3472 	    (ipp->ipp_fields & IPPF_ADDR)) {
3473 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
3474 			if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3475 				v6src = ipp->ipp_addr;
3476 		} else {
3477 			if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3478 				v6src = ipp->ipp_addr;
3479 		}
3480 	}
3481 	/*
3482 	 * Allow source not assigned to the system
3483 	 * only if it is not a local addresses
3484 	 */
3485 	if (!V6_OR_V4_INADDR_ANY(v6src)) {
3486 		ip_laddr_t laddr_type;
3487 
3488 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
3489 			ipaddr_t v4src;
3490 
3491 			IN6_V4MAPPED_TO_IPADDR(&v6src, v4src);
3492 			laddr_type = ip_laddr_verify_v4(v4src, ixa->ixa_zoneid,
3493 			    is->is_netstack->netstack_ip, B_FALSE);
3494 		} else {
3495 			laddr_type = ip_laddr_verify_v6(&v6src, ixa->ixa_zoneid,
3496 			    is->is_netstack->netstack_ip, B_FALSE, B_FALSE);
3497 		}
3498 		if (laddr_type != IPVL_UNICAST_UP)
3499 			ixa->ixa_flags &= ~IXAF_VERIFY_SOURCE;
3500 	}
3501 
3502 	ip_attr_nexthop(ipp, ixa, &v6dst, &v6nexthop);
3503 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
3504 	    &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST);
3505 
3506 	switch (error) {
3507 	case 0:
3508 		break;
3509 	case EADDRNOTAVAIL:
3510 		/*
3511 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3512 		 * Don't have the application see that errno
3513 		 */
3514 		error = ENETUNREACH;
3515 		goto failed;
3516 	case ENETDOWN:
3517 		/*
3518 		 * Have !ipif_addr_ready address; drop packet silently
3519 		 * until we can get applications to not send until we
3520 		 * are ready.
3521 		 */
3522 		error = 0;
3523 		goto failed;
3524 	case EHOSTUNREACH:
3525 	case ENETUNREACH:
3526 		if (ixa->ixa_ire != NULL) {
3527 			/*
3528 			 * Let conn_ip_output/ire_send_noroute return
3529 			 * the error and send any local ICMP error.
3530 			 */
3531 			error = 0;
3532 			break;
3533 		}
3534 		/* FALLTHRU */
3535 	default:
3536 	failed:
3537 		freemsg(mp);
3538 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3539 		goto done;
3540 	}
3541 
3542 	/*
3543 	 * We might be going to a different destination than last time,
3544 	 * thus check that TX allows the communication and compute any
3545 	 * needed label.
3546 	 *
3547 	 * TSOL Note: We have an exclusive ipp and ixa for this thread so we
3548 	 * don't have to worry about concurrent threads.
3549 	 */
3550 	if (is_system_labeled()) {
3551 		/*
3552 		 * Check whether Trusted Solaris policy allows communication
3553 		 * with this host, and pretend that the destination is
3554 		 * unreachable if not.
3555 		 * Compute any needed label and place it in ipp_label_v4/v6.
3556 		 *
3557 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
3558 		 * ipp_label_v4/v6 to form the packet.
3559 		 *
3560 		 * Tsol note: We have ipp structure local to this thread so
3561 		 * no locking is needed.
3562 		 */
3563 		error = conn_update_label(connp, ixa, &v6dst, ipp);
3564 		if (error != 0) {
3565 			freemsg(mp);
3566 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3567 			goto done;
3568 		}
3569 	}
3570 	mp = icmp_prepend_hdr(connp, ixa, ipp, &v6src, &v6dst, flowinfo, mp,
3571 	    &error);
3572 	if (mp == NULL) {
3573 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3574 		ASSERT(error != 0);
3575 		goto done;
3576 	}
3577 	if (ixa->ixa_pktlen > IP_MAXPACKET) {
3578 		error = EMSGSIZE;
3579 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3580 		freemsg(mp);
3581 		goto done;
3582 	}
3583 
3584 	/* Policy might differ for different ICMP type/code */
3585 	mp = icmp_output_attach_policy(mp, connp, ixa);
3586 	if (mp == NULL) {
3587 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3588 		error = EHOSTUNREACH;	/* IPsec policy failure */
3589 		goto done;
3590 	}
3591 
3592 	/* We're done.  Pass the packet to ip. */
3593 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3594 
3595 	error = conn_ip_output(mp, ixa);
3596 	if (!connp->conn_unspec_src)
3597 		ixa->ixa_flags |= IXAF_VERIFY_SOURCE;
3598 	/* No rawipOutErrors if an error since IP increases its error counter */
3599 	switch (error) {
3600 	case 0:
3601 		break;
3602 	case EWOULDBLOCK:
3603 		(void) ixa_check_drain_insert(connp, ixa);
3604 		error = 0;
3605 		break;
3606 	case EADDRNOTAVAIL:
3607 		/*
3608 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3609 		 * Don't have the application see that errno
3610 		 */
3611 		error = ENETUNREACH;
3612 		/* FALLTHRU */
3613 	default:
3614 		mutex_enter(&connp->conn_lock);
3615 		/*
3616 		 * Clear the source and v6lastdst so we call ip_attr_connect
3617 		 * for the next packet and try to pick a better source.
3618 		 */
3619 		if (connp->conn_mcbc_bind)
3620 			connp->conn_saddr_v6 = ipv6_all_zeros;
3621 		else
3622 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
3623 		connp->conn_v6lastdst = ipv6_all_zeros;
3624 		mutex_exit(&connp->conn_lock);
3625 		break;
3626 	}
3627 done:
3628 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3629 	ixa->ixa_cpid = connp->conn_cpid;
3630 	ixa_refrele(ixa);
3631 	ip_pkt_free(ipp);
3632 	kmem_free(ipp, sizeof (*ipp));
3633 	return (error);
3634 }
3635 
3636 /*
3637  * Handle sending an M_DATA for a connected socket.
3638  * Handles both IPv4 and IPv6.
3639  */
3640 int
3641 icmp_output_connected(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid)
3642 {
3643 	icmp_t		*icmp = connp->conn_icmp;
3644 	icmp_stack_t	*is = icmp->icmp_is;
3645 	int		error;
3646 	ip_xmit_attr_t	*ixa;
3647 	boolean_t	do_ipsec;
3648 
3649 	/*
3650 	 * If no other thread is using conn_ixa this just gets a reference to
3651 	 * conn_ixa. Otherwise we get a safe copy of conn_ixa.
3652 	 */
3653 	ixa = conn_get_ixa(connp, B_FALSE);
3654 	if (ixa == NULL) {
3655 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3656 		freemsg(mp);
3657 		return (ENOMEM);
3658 	}
3659 
3660 	ASSERT(cr != NULL);
3661 	ixa->ixa_cred = cr;
3662 	ixa->ixa_cpid = pid;
3663 
3664 	/* Defer IPsec if it might need to look at ICMP type/code */
3665 	switch (ixa->ixa_protocol) {
3666 	case IPPROTO_ICMP:
3667 	case IPPROTO_ICMPV6:
3668 		do_ipsec = B_FALSE;
3669 		break;
3670 	default:
3671 		do_ipsec = B_TRUE;
3672 	}
3673 
3674 	mutex_enter(&connp->conn_lock);
3675 	mp = icmp_prepend_header_template(connp, ixa, mp,
3676 	    &connp->conn_saddr_v6, connp->conn_flowinfo, &error);
3677 
3678 	if (mp == NULL) {
3679 		ASSERT(error != 0);
3680 		mutex_exit(&connp->conn_lock);
3681 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3682 		ixa->ixa_cpid = connp->conn_cpid;
3683 		ixa_refrele(ixa);
3684 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3685 		freemsg(mp);
3686 		return (error);
3687 	}
3688 
3689 	if (!do_ipsec) {
3690 		/* Policy might differ for different ICMP type/code */
3691 		mp = icmp_output_attach_policy(mp, connp, ixa);
3692 		if (mp == NULL) {
3693 			mutex_exit(&connp->conn_lock);
3694 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3695 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3696 			ixa->ixa_cpid = connp->conn_cpid;
3697 			ixa_refrele(ixa);
3698 			return (EHOSTUNREACH);	/* IPsec policy failure */
3699 		}
3700 	}
3701 
3702 	/*
3703 	 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
3704 	 * safe copy, then we need to fill in any pointers in it.
3705 	 */
3706 	if (ixa->ixa_ire == NULL) {
3707 		in6_addr_t	faddr, saddr;
3708 		in6_addr_t	nexthop;
3709 		in_port_t	fport;
3710 
3711 		saddr = connp->conn_saddr_v6;
3712 		faddr = connp->conn_faddr_v6;
3713 		fport = connp->conn_fport;
3714 		ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &faddr, &nexthop);
3715 		mutex_exit(&connp->conn_lock);
3716 
3717 		error = ip_attr_connect(connp, ixa, &saddr, &faddr, &nexthop,
3718 		    fport, NULL, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
3719 		    (do_ipsec ? IPDF_IPSEC : 0));
3720 		switch (error) {
3721 		case 0:
3722 			break;
3723 		case EADDRNOTAVAIL:
3724 			/*
3725 			 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3726 			 * Don't have the application see that errno
3727 			 */
3728 			error = ENETUNREACH;
3729 			goto failed;
3730 		case ENETDOWN:
3731 			/*
3732 			 * Have !ipif_addr_ready address; drop packet silently
3733 			 * until we can get applications to not send until we
3734 			 * are ready.
3735 			 */
3736 			error = 0;
3737 			goto failed;
3738 		case EHOSTUNREACH:
3739 		case ENETUNREACH:
3740 			if (ixa->ixa_ire != NULL) {
3741 				/*
3742 				 * Let conn_ip_output/ire_send_noroute return
3743 				 * the error and send any local ICMP error.
3744 				 */
3745 				error = 0;
3746 				break;
3747 			}
3748 			/* FALLTHRU */
3749 		default:
3750 		failed:
3751 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3752 			ixa->ixa_cpid = connp->conn_cpid;
3753 			ixa_refrele(ixa);
3754 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3755 			freemsg(mp);
3756 			return (error);
3757 		}
3758 	} else {
3759 		/* Done with conn_t */
3760 		mutex_exit(&connp->conn_lock);
3761 	}
3762 
3763 	/* We're done.  Pass the packet to ip. */
3764 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3765 
3766 	error = conn_ip_output(mp, ixa);
3767 	/* No rawipOutErrors if an error since IP increases its error counter */
3768 	switch (error) {
3769 	case 0:
3770 		break;
3771 	case EWOULDBLOCK:
3772 		(void) ixa_check_drain_insert(connp, ixa);
3773 		error = 0;
3774 		break;
3775 	case EADDRNOTAVAIL:
3776 		/*
3777 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3778 		 * Don't have the application see that errno
3779 		 */
3780 		error = ENETUNREACH;
3781 		break;
3782 	}
3783 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3784 	ixa->ixa_cpid = connp->conn_cpid;
3785 	ixa_refrele(ixa);
3786 	return (error);
3787 }
3788 
3789 /*
3790  * Handle sending an M_DATA to the last destination.
3791  * Handles both IPv4 and IPv6.
3792  *
3793  * NOTE: The caller must hold conn_lock and we drop it here.
3794  */
3795 int
3796 icmp_output_lastdst(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid,
3797     ip_xmit_attr_t *ixa)
3798 {
3799 	icmp_t		*icmp = connp->conn_icmp;
3800 	icmp_stack_t	*is = icmp->icmp_is;
3801 	int		error;
3802 	boolean_t	do_ipsec;
3803 
3804 	ASSERT(MUTEX_HELD(&connp->conn_lock));
3805 	ASSERT(ixa != NULL);
3806 
3807 	ASSERT(cr != NULL);
3808 	ixa->ixa_cred = cr;
3809 	ixa->ixa_cpid = pid;
3810 
3811 	/* Defer IPsec if it might need to look at ICMP type/code */
3812 	switch (ixa->ixa_protocol) {
3813 	case IPPROTO_ICMP:
3814 	case IPPROTO_ICMPV6:
3815 		do_ipsec = B_FALSE;
3816 		break;
3817 	default:
3818 		do_ipsec = B_TRUE;
3819 	}
3820 
3821 
3822 	mp = icmp_prepend_header_template(connp, ixa, mp,
3823 	    &connp->conn_v6lastsrc, connp->conn_lastflowinfo, &error);
3824 
3825 	if (mp == NULL) {
3826 		ASSERT(error != 0);
3827 		mutex_exit(&connp->conn_lock);
3828 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3829 		ixa->ixa_cpid = connp->conn_cpid;
3830 		ixa_refrele(ixa);
3831 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3832 		freemsg(mp);
3833 		return (error);
3834 	}
3835 
3836 	if (!do_ipsec) {
3837 		/* Policy might differ for different ICMP type/code */
3838 		mp = icmp_output_attach_policy(mp, connp, ixa);
3839 		if (mp == NULL) {
3840 			mutex_exit(&connp->conn_lock);
3841 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3842 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3843 			ixa->ixa_cpid = connp->conn_cpid;
3844 			ixa_refrele(ixa);
3845 			return (EHOSTUNREACH);	/* IPsec policy failure */
3846 		}
3847 	}
3848 
3849 	/*
3850 	 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
3851 	 * safe copy, then we need to fill in any pointers in it.
3852 	 */
3853 	if (ixa->ixa_ire == NULL) {
3854 		in6_addr_t	lastdst, lastsrc;
3855 		in6_addr_t	nexthop;
3856 		in_port_t	lastport;
3857 
3858 		lastsrc = connp->conn_v6lastsrc;
3859 		lastdst = connp->conn_v6lastdst;
3860 		lastport = connp->conn_lastdstport;
3861 		ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &lastdst, &nexthop);
3862 		mutex_exit(&connp->conn_lock);
3863 
3864 		error = ip_attr_connect(connp, ixa, &lastsrc, &lastdst,
3865 		    &nexthop, lastport, NULL, NULL, IPDF_ALLOW_MCBC |
3866 		    IPDF_VERIFY_DST | (do_ipsec ? IPDF_IPSEC : 0));
3867 		switch (error) {
3868 		case 0:
3869 			break;
3870 		case EADDRNOTAVAIL:
3871 			/*
3872 			 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3873 			 * Don't have the application see that errno
3874 			 */
3875 			error = ENETUNREACH;
3876 			goto failed;
3877 		case ENETDOWN:
3878 			/*
3879 			 * Have !ipif_addr_ready address; drop packet silently
3880 			 * until we can get applications to not send until we
3881 			 * are ready.
3882 			 */
3883 			error = 0;
3884 			goto failed;
3885 		case EHOSTUNREACH:
3886 		case ENETUNREACH:
3887 			if (ixa->ixa_ire != NULL) {
3888 				/*
3889 				 * Let conn_ip_output/ire_send_noroute return
3890 				 * the error and send any local ICMP error.
3891 				 */
3892 				error = 0;
3893 				break;
3894 			}
3895 			/* FALLTHRU */
3896 		default:
3897 		failed:
3898 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3899 			ixa->ixa_cpid = connp->conn_cpid;
3900 			ixa_refrele(ixa);
3901 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3902 			freemsg(mp);
3903 			return (error);
3904 		}
3905 	} else {
3906 		/* Done with conn_t */
3907 		mutex_exit(&connp->conn_lock);
3908 	}
3909 
3910 	/* We're done.  Pass the packet to ip. */
3911 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3912 	error = conn_ip_output(mp, ixa);
3913 	/* No rawipOutErrors if an error since IP increases its error counter */
3914 	switch (error) {
3915 	case 0:
3916 		break;
3917 	case EWOULDBLOCK:
3918 		(void) ixa_check_drain_insert(connp, ixa);
3919 		error = 0;
3920 		break;
3921 	case EADDRNOTAVAIL:
3922 		/*
3923 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3924 		 * Don't have the application see that errno
3925 		 */
3926 		error = ENETUNREACH;
3927 		/* FALLTHRU */
3928 	default:
3929 		mutex_enter(&connp->conn_lock);
3930 		/*
3931 		 * Clear the source and v6lastdst so we call ip_attr_connect
3932 		 * for the next packet and try to pick a better source.
3933 		 */
3934 		if (connp->conn_mcbc_bind)
3935 			connp->conn_saddr_v6 = ipv6_all_zeros;
3936 		else
3937 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
3938 		connp->conn_v6lastdst = ipv6_all_zeros;
3939 		mutex_exit(&connp->conn_lock);
3940 		break;
3941 	}
3942 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3943 	ixa->ixa_cpid = connp->conn_cpid;
3944 	ixa_refrele(ixa);
3945 	return (error);
3946 }
3947 
3948 
3949 /*
3950  * Prepend the header template and then fill in the source and
3951  * flowinfo. The caller needs to handle the destination address since
3952  * it's setting is different if rthdr or source route.
3953  *
3954  * Returns NULL is allocation failed or if the packet would exceed IP_MAXPACKET.
3955  * When it returns NULL it sets errorp.
3956  */
3957 static mblk_t *
3958 icmp_prepend_header_template(conn_t *connp, ip_xmit_attr_t *ixa, mblk_t *mp,
3959     const in6_addr_t *v6src, uint32_t flowinfo, int *errorp)
3960 {
3961 	icmp_t		*icmp = connp->conn_icmp;
3962 	icmp_stack_t	*is = icmp->icmp_is;
3963 	uint_t		pktlen;
3964 	uint_t		copylen;
3965 	uint8_t		*iph;
3966 	uint_t		ip_hdr_length;
3967 	uint32_t	cksum;
3968 	ip_pkt_t	*ipp;
3969 
3970 	ASSERT(MUTEX_HELD(&connp->conn_lock));
3971 
3972 	/*
3973 	 * Copy the header template.
3974 	 */
3975 	copylen = connp->conn_ht_iphc_len;
3976 	pktlen = copylen + msgdsize(mp);
3977 	if (pktlen > IP_MAXPACKET) {
3978 		freemsg(mp);
3979 		*errorp = EMSGSIZE;
3980 		return (NULL);
3981 	}
3982 	ixa->ixa_pktlen = pktlen;
3983 
3984 	/* check/fix buffer config, setup pointers into it */
3985 	iph = mp->b_rptr - copylen;
3986 	if (DB_REF(mp) != 1 || iph < DB_BASE(mp) || !OK_32PTR(iph)) {
3987 		mblk_t *mp1;
3988 
3989 		mp1 = allocb(copylen + is->is_wroff_extra, BPRI_MED);
3990 		if (mp1 == NULL) {
3991 			freemsg(mp);
3992 			*errorp = ENOMEM;
3993 			return (NULL);
3994 		}
3995 		mp1->b_wptr = DB_LIM(mp1);
3996 		mp1->b_cont = mp;
3997 		mp = mp1;
3998 		iph = (mp->b_wptr - copylen);
3999 	}
4000 	mp->b_rptr = iph;
4001 	bcopy(connp->conn_ht_iphc, iph, copylen);
4002 	ip_hdr_length = (uint_t)(connp->conn_ht_ulp - connp->conn_ht_iphc);
4003 
4004 	ixa->ixa_ip_hdr_length = ip_hdr_length;
4005 
4006 	/*
4007 	 * Prepare for ICMPv6 checksum done in IP.
4008 	 *
4009 	 * icmp_build_hdr_template has already massaged any routing header
4010 	 * and placed the result in conn_sum.
4011 	 *
4012 	 * We make it easy for IP to include our pseudo header
4013 	 * by putting our length (and any routing header adjustment)
4014 	 * in the ICMPv6 checksum field.
4015 	 */
4016 	cksum = pktlen - ip_hdr_length;
4017 
4018 	cksum += connp->conn_sum;
4019 	cksum = (cksum >> 16) + (cksum & 0xFFFF);
4020 	ASSERT(cksum < 0x10000);
4021 
4022 	ipp = &connp->conn_xmit_ipp;
4023 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
4024 		ipha_t	*ipha = (ipha_t *)iph;
4025 
4026 		ipha->ipha_length = htons((uint16_t)pktlen);
4027 
4028 		/* if IP_PKTINFO specified an addres it wins over bind() */
4029 		if ((ipp->ipp_fields & IPPF_ADDR) &&
4030 		    IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
4031 			ASSERT(ipp->ipp_addr_v4 != INADDR_ANY);
4032 			ipha->ipha_src = ipp->ipp_addr_v4;
4033 		} else {
4034 			IN6_V4MAPPED_TO_IPADDR(v6src, ipha->ipha_src);
4035 		}
4036 	} else {
4037 		ip6_t *ip6h = (ip6_t *)iph;
4038 		uint_t	cksum_offset = 0;
4039 
4040 		ip6h->ip6_plen =  htons((uint16_t)(pktlen - IPV6_HDR_LEN));
4041 
4042 		/* if IP_PKTINFO specified an addres it wins over bind() */
4043 		if ((ipp->ipp_fields & IPPF_ADDR) &&
4044 		    !IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
4045 			ASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ipp->ipp_addr));
4046 			ip6h->ip6_src = ipp->ipp_addr;
4047 		} else {
4048 			ip6h->ip6_src = *v6src;
4049 		}
4050 		ip6h->ip6_vcf =
4051 		    (IPV6_DEFAULT_VERS_AND_FLOW & IPV6_VERS_AND_FLOW_MASK) |
4052 		    (flowinfo & ~IPV6_VERS_AND_FLOW_MASK);
4053 		if (ipp->ipp_fields & IPPF_TCLASS) {
4054 			/* Overrides the class part of flowinfo */
4055 			ip6h->ip6_vcf = IPV6_TCLASS_FLOW(ip6h->ip6_vcf,
4056 			    ipp->ipp_tclass);
4057 		}
4058 
4059 		if (ixa->ixa_flags & IXAF_SET_ULP_CKSUM) {
4060 			if (connp->conn_proto == IPPROTO_ICMPV6) {
4061 				cksum_offset = ixa->ixa_ip_hdr_length +
4062 				    offsetof(icmp6_t, icmp6_cksum);
4063 			} else if (ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
4064 				cksum_offset = ixa->ixa_ip_hdr_length +
4065 				    ixa->ixa_raw_cksum_offset;
4066 			}
4067 		}
4068 		if (cksum_offset != 0) {
4069 			uint16_t *ptr;
4070 
4071 			/* Make sure the checksum fits in the first mblk */
4072 			if (cksum_offset + sizeof (short) > MBLKL(mp)) {
4073 				mblk_t *mp1;
4074 
4075 				mp1 = msgpullup(mp,
4076 				    cksum_offset + sizeof (short));
4077 				freemsg(mp);
4078 				if (mp1 == NULL) {
4079 					*errorp = ENOMEM;
4080 					return (NULL);
4081 				}
4082 				mp = mp1;
4083 				iph = mp->b_rptr;
4084 				ip6h = (ip6_t *)iph;
4085 			}
4086 			ptr = (uint16_t *)(mp->b_rptr + cksum_offset);
4087 			*ptr = htons(cksum);
4088 		}
4089 	}
4090 
4091 	return (mp);
4092 }
4093 
4094 /*
4095  * This routine handles all messages passed downstream.  It either
4096  * consumes the message or passes it downstream; it never queues a
4097  * a message.
4098  */
4099 void
4100 icmp_wput(queue_t *q, mblk_t *mp)
4101 {
4102 	sin6_t		*sin6;
4103 	sin_t		*sin = NULL;
4104 	uint_t		srcid;
4105 	conn_t		*connp = Q_TO_CONN(q);
4106 	icmp_t		*icmp = connp->conn_icmp;
4107 	int		error = 0;
4108 	struct sockaddr	*addr = NULL;
4109 	socklen_t	addrlen;
4110 	icmp_stack_t	*is = icmp->icmp_is;
4111 	struct T_unitdata_req *tudr;
4112 	mblk_t		*data_mp;
4113 	cred_t		*cr;
4114 	pid_t		pid;
4115 
4116 	/*
4117 	 * We directly handle several cases here: T_UNITDATA_REQ message
4118 	 * coming down as M_PROTO/M_PCPROTO and M_DATA messages for connected
4119 	 * socket.
4120 	 */
4121 	switch (DB_TYPE(mp)) {
4122 	case M_DATA:
4123 		/* sockfs never sends down M_DATA */
4124 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4125 		freemsg(mp);
4126 		return;
4127 
4128 	case M_PROTO:
4129 	case M_PCPROTO:
4130 		tudr = (struct T_unitdata_req *)mp->b_rptr;
4131 		if (MBLKL(mp) < sizeof (*tudr) ||
4132 		    ((t_primp_t)mp->b_rptr)->type != T_UNITDATA_REQ) {
4133 			icmp_wput_other(q, mp);
4134 			return;
4135 		}
4136 		break;
4137 
4138 	default:
4139 		icmp_wput_other(q, mp);
4140 		return;
4141 	}
4142 
4143 	/* Handle valid T_UNITDATA_REQ here */
4144 	data_mp = mp->b_cont;
4145 	if (data_mp == NULL) {
4146 		error = EPROTO;
4147 		goto ud_error2;
4148 	}
4149 	mp->b_cont = NULL;
4150 
4151 	if (!MBLKIN(mp, 0, tudr->DEST_offset + tudr->DEST_length)) {
4152 		error = EADDRNOTAVAIL;
4153 		goto ud_error2;
4154 	}
4155 
4156 	/*
4157 	 * All Solaris components should pass a db_credp
4158 	 * for this message, hence we ASSERT.
4159 	 * On production kernels we return an error to be robust against
4160 	 * random streams modules sitting on top of us.
4161 	 */
4162 	cr = msg_getcred(mp, &pid);
4163 	ASSERT(cr != NULL);
4164 	if (cr == NULL) {
4165 		error = EINVAL;
4166 		goto ud_error2;
4167 	}
4168 
4169 	/*
4170 	 * If a port has not been bound to the stream, fail.
4171 	 * This is not a problem when sockfs is directly
4172 	 * above us, because it will ensure that the socket
4173 	 * is first bound before allowing data to be sent.
4174 	 */
4175 	if (icmp->icmp_state == TS_UNBND) {
4176 		error = EPROTO;
4177 		goto ud_error2;
4178 	}
4179 	addr = (struct sockaddr *)&mp->b_rptr[tudr->DEST_offset];
4180 	addrlen = tudr->DEST_length;
4181 
4182 	switch (connp->conn_family) {
4183 	case AF_INET6:
4184 		sin6 = (sin6_t *)addr;
4185 		if (!OK_32PTR((char *)sin6) || (addrlen != sizeof (sin6_t)) ||
4186 		    (sin6->sin6_family != AF_INET6)) {
4187 			error = EADDRNOTAVAIL;
4188 			goto ud_error2;
4189 		}
4190 
4191 		/* No support for mapped addresses on raw sockets */
4192 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
4193 			error = EADDRNOTAVAIL;
4194 			goto ud_error2;
4195 		}
4196 		srcid = sin6->__sin6_src_id;
4197 
4198 		/*
4199 		 * If the local address is a mapped address return
4200 		 * an error.
4201 		 * It would be possible to send an IPv6 packet but the
4202 		 * response would never make it back to the application
4203 		 * since it is bound to a mapped address.
4204 		 */
4205 		if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
4206 			error = EADDRNOTAVAIL;
4207 			goto ud_error2;
4208 		}
4209 
4210 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
4211 			sin6->sin6_addr = ipv6_loopback;
4212 
4213 		if (tudr->OPT_length != 0) {
4214 			/*
4215 			 * If we are connected then the destination needs to be
4216 			 * the same as the connected one.
4217 			 */
4218 			if (icmp->icmp_state == TS_DATA_XFER &&
4219 			    !conn_same_as_last_v6(connp, sin6)) {
4220 				error = EISCONN;
4221 				goto ud_error2;
4222 			}
4223 			error = icmp_output_ancillary(connp, NULL, sin6,
4224 			    data_mp, mp, NULL, cr, pid);
4225 		} else {
4226 			ip_xmit_attr_t *ixa;
4227 
4228 			/*
4229 			 * We have to allocate an ip_xmit_attr_t before we grab
4230 			 * conn_lock and we need to hold conn_lock once we've
4231 			 * checked conn_same_as_last_v6 to handle concurrent
4232 			 * send* calls on a socket.
4233 			 */
4234 			ixa = conn_get_ixa(connp, B_FALSE);
4235 			if (ixa == NULL) {
4236 				error = ENOMEM;
4237 				goto ud_error2;
4238 			}
4239 			mutex_enter(&connp->conn_lock);
4240 
4241 			if (conn_same_as_last_v6(connp, sin6) &&
4242 			    connp->conn_lastsrcid == srcid &&
4243 			    ipsec_outbound_policy_current(ixa)) {
4244 				/* icmp_output_lastdst drops conn_lock */
4245 				error = icmp_output_lastdst(connp, data_mp, cr,
4246 				    pid, ixa);
4247 			} else {
4248 				/* icmp_output_newdst drops conn_lock */
4249 				error = icmp_output_newdst(connp, data_mp, NULL,
4250 				    sin6, cr, pid, ixa);
4251 			}
4252 			ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
4253 		}
4254 		if (error == 0) {
4255 			freeb(mp);
4256 			return;
4257 		}
4258 		break;
4259 
4260 	case AF_INET:
4261 		sin = (sin_t *)addr;
4262 		if ((!OK_32PTR((char *)sin) || addrlen != sizeof (sin_t)) ||
4263 		    (sin->sin_family != AF_INET)) {
4264 			error = EADDRNOTAVAIL;
4265 			goto ud_error2;
4266 		}
4267 		if (sin->sin_addr.s_addr == INADDR_ANY)
4268 			sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
4269 
4270 		/* Protocol 255 contains full IP headers */
4271 		/* Read without holding lock */
4272 		if (icmp->icmp_hdrincl) {
4273 			if (MBLKL(data_mp) < IP_SIMPLE_HDR_LENGTH) {
4274 				if (!pullupmsg(data_mp, IP_SIMPLE_HDR_LENGTH)) {
4275 					error = EINVAL;
4276 					goto ud_error2;
4277 				}
4278 			}
4279 			error = icmp_output_hdrincl(connp, data_mp, cr, pid);
4280 			if (error == 0) {
4281 				freeb(mp);
4282 				return;
4283 			}
4284 			/* data_mp consumed above */
4285 			data_mp = NULL;
4286 			goto ud_error2;
4287 		}
4288 
4289 		if (tudr->OPT_length != 0) {
4290 			/*
4291 			 * If we are connected then the destination needs to be
4292 			 * the same as the connected one.
4293 			 */
4294 			if (icmp->icmp_state == TS_DATA_XFER &&
4295 			    !conn_same_as_last_v4(connp, sin)) {
4296 				error = EISCONN;
4297 				goto ud_error2;
4298 			}
4299 			error = icmp_output_ancillary(connp, sin, NULL,
4300 			    data_mp, mp, NULL, cr, pid);
4301 		} else {
4302 			ip_xmit_attr_t *ixa;
4303 
4304 			/*
4305 			 * We have to allocate an ip_xmit_attr_t before we grab
4306 			 * conn_lock and we need to hold conn_lock once we've
4307 			 * checked conn_same_as_last_v4 to handle concurrent
4308 			 * send* calls on a socket.
4309 			 */
4310 			ixa = conn_get_ixa(connp, B_FALSE);
4311 			if (ixa == NULL) {
4312 				error = ENOMEM;
4313 				goto ud_error2;
4314 			}
4315 			mutex_enter(&connp->conn_lock);
4316 
4317 			if (conn_same_as_last_v4(connp, sin) &&
4318 			    ipsec_outbound_policy_current(ixa)) {
4319 				/* icmp_output_lastdst drops conn_lock */
4320 				error = icmp_output_lastdst(connp, data_mp, cr,
4321 				    pid, ixa);
4322 			} else {
4323 				/* icmp_output_newdst drops conn_lock */
4324 				error = icmp_output_newdst(connp, data_mp, sin,
4325 				    NULL, cr, pid, ixa);
4326 			}
4327 			ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
4328 		}
4329 		if (error == 0) {
4330 			freeb(mp);
4331 			return;
4332 		}
4333 		break;
4334 	}
4335 	ASSERT(mp != NULL);
4336 	/* mp is freed by the following routine */
4337 	icmp_ud_err(q, mp, (t_scalar_t)error);
4338 	return;
4339 
4340 ud_error2:
4341 	BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4342 	freemsg(data_mp);
4343 	ASSERT(mp != NULL);
4344 	/* mp is freed by the following routine */
4345 	icmp_ud_err(q, mp, (t_scalar_t)error);
4346 }
4347 
4348 /*
4349  * Handle the case of the IP address or flow label being different
4350  * for both IPv4 and IPv6.
4351  *
4352  * NOTE: The caller must hold conn_lock and we drop it here.
4353  */
4354 static int
4355 icmp_output_newdst(conn_t *connp, mblk_t *data_mp, sin_t *sin, sin6_t *sin6,
4356     cred_t *cr, pid_t pid, ip_xmit_attr_t *ixa)
4357 {
4358 	icmp_t		*icmp = connp->conn_icmp;
4359 	icmp_stack_t	*is = icmp->icmp_is;
4360 	int		error;
4361 	ip_xmit_attr_t	*oldixa;
4362 	boolean_t	do_ipsec;
4363 	uint_t		srcid;
4364 	uint32_t	flowinfo;
4365 	in6_addr_t	v6src;
4366 	in6_addr_t	v6dst;
4367 	in6_addr_t	v6nexthop;
4368 	in_port_t	dstport;
4369 
4370 	ASSERT(MUTEX_HELD(&connp->conn_lock));
4371 	ASSERT(ixa != NULL);
4372 
4373 	/*
4374 	 * We hold conn_lock across all the use and modifications of
4375 	 * the conn_lastdst, conn_ixa, and conn_xmit_ipp to ensure that they
4376 	 * stay consistent.
4377 	 */
4378 
4379 	ASSERT(cr != NULL);
4380 	ixa->ixa_cred = cr;
4381 	ixa->ixa_cpid = pid;
4382 	if (is_system_labeled()) {
4383 		/* We need to restart with a label based on the cred */
4384 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
4385 	}
4386 	/*
4387 	 * If we are connected then the destination needs to be the
4388 	 * same as the connected one, which is not the case here since we
4389 	 * checked for that above.
4390 	 */
4391 	if (icmp->icmp_state == TS_DATA_XFER) {
4392 		mutex_exit(&connp->conn_lock);
4393 		error = EISCONN;
4394 		goto ud_error;
4395 	}
4396 
4397 	/* In case previous destination was multicast or multirt */
4398 	ip_attr_newdst(ixa);
4399 
4400 	/*
4401 	 * If laddr is unspecified then we look at sin6_src_id.
4402 	 * We will give precedence to a source address set with IPV6_PKTINFO
4403 	 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
4404 	 * want ip_attr_connect to select a source (since it can fail) when
4405 	 * IPV6_PKTINFO is specified.
4406 	 * If this doesn't result in a source address then we get a source
4407 	 * from ip_attr_connect() below.
4408 	 */
4409 	v6src = connp->conn_saddr_v6;
4410 	if (sin != NULL) {
4411 		IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
4412 		dstport = sin->sin_port;
4413 		flowinfo = 0;
4414 		srcid = 0;
4415 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
4416 		if (srcid != 0 && V4_PART_OF_V6(&v6src) == INADDR_ANY) {
4417 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
4418 			    connp->conn_netstack);
4419 		}
4420 		ixa->ixa_flags |= IXAF_IS_IPV4;
4421 	} else {
4422 		v6dst = sin6->sin6_addr;
4423 		dstport = sin6->sin6_port;
4424 		flowinfo = sin6->sin6_flowinfo;
4425 		srcid = sin6->__sin6_src_id;
4426 		if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
4427 			ixa->ixa_scopeid = sin6->sin6_scope_id;
4428 			ixa->ixa_flags |= IXAF_SCOPEID_SET;
4429 		} else {
4430 			ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
4431 		}
4432 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
4433 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
4434 			    connp->conn_netstack);
4435 		}
4436 		if (IN6_IS_ADDR_V4MAPPED(&v6dst))
4437 			ixa->ixa_flags |= IXAF_IS_IPV4;
4438 		else
4439 			ixa->ixa_flags &= ~IXAF_IS_IPV4;
4440 	}
4441 	/* Handle IPV6_PKTINFO setting source address. */
4442 	if (IN6_IS_ADDR_UNSPECIFIED(&v6src) &&
4443 	    (connp->conn_xmit_ipp.ipp_fields & IPPF_ADDR)) {
4444 		ip_pkt_t *ipp = &connp->conn_xmit_ipp;
4445 
4446 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
4447 			if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
4448 				v6src = ipp->ipp_addr;
4449 		} else {
4450 			if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
4451 				v6src = ipp->ipp_addr;
4452 		}
4453 	}
4454 
4455 	/* Defer IPsec if it might need to look at ICMP type/code */
4456 	switch (ixa->ixa_protocol) {
4457 	case IPPROTO_ICMP:
4458 	case IPPROTO_ICMPV6:
4459 		do_ipsec = B_FALSE;
4460 		break;
4461 	default:
4462 		do_ipsec = B_TRUE;
4463 	}
4464 
4465 	ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &v6dst, &v6nexthop);
4466 	mutex_exit(&connp->conn_lock);
4467 
4468 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
4469 	    &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
4470 	    (do_ipsec ? IPDF_IPSEC : 0));
4471 	switch (error) {
4472 	case 0:
4473 		break;
4474 	case EADDRNOTAVAIL:
4475 		/*
4476 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
4477 		 * Don't have the application see that errno
4478 		 */
4479 		error = ENETUNREACH;
4480 		goto failed;
4481 	case ENETDOWN:
4482 		/*
4483 		 * Have !ipif_addr_ready address; drop packet silently
4484 		 * until we can get applications to not send until we
4485 		 * are ready.
4486 		 */
4487 		error = 0;
4488 		goto failed;
4489 	case EHOSTUNREACH:
4490 	case ENETUNREACH:
4491 		if (ixa->ixa_ire != NULL) {
4492 			/*
4493 			 * Let conn_ip_output/ire_send_noroute return
4494 			 * the error and send any local ICMP error.
4495 			 */
4496 			error = 0;
4497 			break;
4498 		}
4499 		/* FALLTHRU */
4500 	default:
4501 	failed:
4502 		goto ud_error;
4503 	}
4504 
4505 	mutex_enter(&connp->conn_lock);
4506 	/*
4507 	 * While we dropped the lock some other thread might have connected
4508 	 * this socket. If so we bail out with EISCONN to ensure that the
4509 	 * connecting thread is the one that updates conn_ixa, conn_ht_*
4510 	 * and conn_*last*.
4511 	 */
4512 	if (icmp->icmp_state == TS_DATA_XFER) {
4513 		mutex_exit(&connp->conn_lock);
4514 		error = EISCONN;
4515 		goto ud_error;
4516 	}
4517 
4518 	/*
4519 	 * We need to rebuild the headers if
4520 	 *  - we are labeling packets (could be different for different
4521 	 *    destinations)
4522 	 *  - we have a source route (or routing header) since we need to
4523 	 *    massage that to get the pseudo-header checksum
4524 	 *  - a socket option with COA_HEADER_CHANGED has been set which
4525 	 *    set conn_v6lastdst to zero.
4526 	 *
4527 	 * Otherwise the prepend function will just update the src, dst,
4528 	 * and flow label.
4529 	 */
4530 	if (is_system_labeled()) {
4531 		/* TX MLP requires SCM_UCRED and don't have that here */
4532 		if (connp->conn_mlp_type != mlptSingle) {
4533 			mutex_exit(&connp->conn_lock);
4534 			error = ECONNREFUSED;
4535 			goto ud_error;
4536 		}
4537 		/*
4538 		 * Check whether Trusted Solaris policy allows communication
4539 		 * with this host, and pretend that the destination is
4540 		 * unreachable if not.
4541 		 * Compute any needed label and place it in ipp_label_v4/v6.
4542 		 *
4543 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
4544 		 * ipp_label_v4/v6 to form the packet.
4545 		 *
4546 		 * Tsol note: Since we hold conn_lock we know no other
4547 		 * thread manipulates conn_xmit_ipp.
4548 		 */
4549 		error = conn_update_label(connp, ixa, &v6dst,
4550 		    &connp->conn_xmit_ipp);
4551 		if (error != 0) {
4552 			mutex_exit(&connp->conn_lock);
4553 			goto ud_error;
4554 		}
4555 		/* Rebuild the header template */
4556 		error = icmp_build_hdr_template(connp, &v6src, &v6dst,
4557 		    flowinfo);
4558 		if (error != 0) {
4559 			mutex_exit(&connp->conn_lock);
4560 			goto ud_error;
4561 		}
4562 	} else if (connp->conn_xmit_ipp.ipp_fields &
4563 	    (IPPF_IPV4_OPTIONS|IPPF_RTHDR) ||
4564 	    IN6_IS_ADDR_UNSPECIFIED(&connp->conn_v6lastdst)) {
4565 		/* Rebuild the header template */
4566 		error = icmp_build_hdr_template(connp, &v6src, &v6dst,
4567 		    flowinfo);
4568 		if (error != 0) {
4569 			mutex_exit(&connp->conn_lock);
4570 			goto ud_error;
4571 		}
4572 	} else {
4573 		/* Simply update the destination address if no source route */
4574 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
4575 			ipha_t	*ipha = (ipha_t *)connp->conn_ht_iphc;
4576 
4577 			IN6_V4MAPPED_TO_IPADDR(&v6dst, ipha->ipha_dst);
4578 			if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
4579 				ipha->ipha_fragment_offset_and_flags |=
4580 				    IPH_DF_HTONS;
4581 			} else {
4582 				ipha->ipha_fragment_offset_and_flags &=
4583 				    ~IPH_DF_HTONS;
4584 			}
4585 		} else {
4586 			ip6_t *ip6h = (ip6_t *)connp->conn_ht_iphc;
4587 			ip6h->ip6_dst = v6dst;
4588 		}
4589 	}
4590 
4591 	/*
4592 	 * Remember the dst etc which corresponds to the built header
4593 	 * template and conn_ixa.
4594 	 */
4595 	oldixa = conn_replace_ixa(connp, ixa);
4596 	connp->conn_v6lastdst = v6dst;
4597 	connp->conn_lastflowinfo = flowinfo;
4598 	connp->conn_lastscopeid = ixa->ixa_scopeid;
4599 	connp->conn_lastsrcid = srcid;
4600 	/* Also remember a source to use together with lastdst */
4601 	connp->conn_v6lastsrc = v6src;
4602 
4603 	data_mp = icmp_prepend_header_template(connp, ixa, data_mp, &v6src,
4604 	    flowinfo, &error);
4605 
4606 	/* Done with conn_t */
4607 	mutex_exit(&connp->conn_lock);
4608 	ixa_refrele(oldixa);
4609 
4610 	if (data_mp == NULL) {
4611 		ASSERT(error != 0);
4612 		goto ud_error;
4613 	}
4614 
4615 	if (!do_ipsec) {
4616 		/* Policy might differ for different ICMP type/code */
4617 		data_mp = icmp_output_attach_policy(data_mp, connp, ixa);
4618 		if (data_mp == NULL) {
4619 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4620 			error = EHOSTUNREACH;	/* IPsec policy failure */
4621 			goto done;
4622 		}
4623 	}
4624 
4625 	/* We're done.  Pass the packet to ip. */
4626 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
4627 
4628 	error = conn_ip_output(data_mp, ixa);
4629 	/* No rawipOutErrors if an error since IP increases its error counter */
4630 	switch (error) {
4631 	case 0:
4632 		break;
4633 	case EWOULDBLOCK:
4634 		(void) ixa_check_drain_insert(connp, ixa);
4635 		error = 0;
4636 		break;
4637 	case EADDRNOTAVAIL:
4638 		/*
4639 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
4640 		 * Don't have the application see that errno
4641 		 */
4642 		error = ENETUNREACH;
4643 		/* FALLTHRU */
4644 	default:
4645 		mutex_enter(&connp->conn_lock);
4646 		/*
4647 		 * Clear the source and v6lastdst so we call ip_attr_connect
4648 		 * for the next packet and try to pick a better source.
4649 		 */
4650 		if (connp->conn_mcbc_bind)
4651 			connp->conn_saddr_v6 = ipv6_all_zeros;
4652 		else
4653 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
4654 		connp->conn_v6lastdst = ipv6_all_zeros;
4655 		mutex_exit(&connp->conn_lock);
4656 		break;
4657 	}
4658 done:
4659 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
4660 	ixa->ixa_cpid = connp->conn_cpid;
4661 	ixa_refrele(ixa);
4662 	return (error);
4663 
4664 ud_error:
4665 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
4666 	ixa->ixa_cpid = connp->conn_cpid;
4667 	ixa_refrele(ixa);
4668 
4669 	BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4670 	freemsg(data_mp);
4671 	return (error);
4672 }
4673 
4674 /* ARGSUSED */
4675 static void
4676 icmp_wput_fallback(queue_t *q, mblk_t *mp)
4677 {
4678 #ifdef DEBUG
4679 	cmn_err(CE_CONT, "icmp_wput_fallback: Message during fallback \n");
4680 #endif
4681 	freemsg(mp);
4682 }
4683 
4684 static void
4685 icmp_wput_other(queue_t *q, mblk_t *mp)
4686 {
4687 	uchar_t	*rptr = mp->b_rptr;
4688 	struct iocblk *iocp;
4689 	conn_t	*connp = Q_TO_CONN(q);
4690 	icmp_t	*icmp = connp->conn_icmp;
4691 	icmp_stack_t *is = icmp->icmp_is;
4692 	cred_t *cr;
4693 
4694 	switch (mp->b_datap->db_type) {
4695 	case M_PROTO:
4696 	case M_PCPROTO:
4697 		if (mp->b_wptr - rptr < sizeof (t_scalar_t)) {
4698 			/*
4699 			 * If the message does not contain a PRIM_type,
4700 			 * throw it away.
4701 			 */
4702 			freemsg(mp);
4703 			return;
4704 		}
4705 		switch (((t_primp_t)rptr)->type) {
4706 		case T_ADDR_REQ:
4707 			icmp_addr_req(q, mp);
4708 			return;
4709 		case O_T_BIND_REQ:
4710 		case T_BIND_REQ:
4711 			icmp_tpi_bind(q, mp);
4712 			return;
4713 		case T_CONN_REQ:
4714 			icmp_tpi_connect(q, mp);
4715 			return;
4716 		case T_CAPABILITY_REQ:
4717 			icmp_capability_req(q, mp);
4718 			return;
4719 		case T_INFO_REQ:
4720 			icmp_info_req(q, mp);
4721 			return;
4722 		case T_UNITDATA_REQ:
4723 			/*
4724 			 * If a T_UNITDATA_REQ gets here, the address must
4725 			 * be bad.  Valid T_UNITDATA_REQs are handled
4726 			 * in icmp_wput.
4727 			 */
4728 			icmp_ud_err(q, mp, EADDRNOTAVAIL);
4729 			return;
4730 		case T_UNBIND_REQ:
4731 			icmp_tpi_unbind(q, mp);
4732 			return;
4733 		case T_SVR4_OPTMGMT_REQ:
4734 			/*
4735 			 * All Solaris components should pass a db_credp
4736 			 * for this TPI message, hence we ASSERT.
4737 			 * But in case there is some other M_PROTO that looks
4738 			 * like a TPI message sent by some other kernel
4739 			 * component, we check and return an error.
4740 			 */
4741 			cr = msg_getcred(mp, NULL);
4742 			ASSERT(cr != NULL);
4743 			if (cr == NULL) {
4744 				icmp_err_ack(q, mp, TSYSERR, EINVAL);
4745 				return;
4746 			}
4747 
4748 			if (!snmpcom_req(q, mp, icmp_snmp_set, ip_snmp_get,
4749 			    cr)) {
4750 				svr4_optcom_req(q, mp, cr, &icmp_opt_obj);
4751 			}
4752 			return;
4753 
4754 		case T_OPTMGMT_REQ:
4755 			/*
4756 			 * All Solaris components should pass a db_credp
4757 			 * for this TPI message, hence we ASSERT.
4758 			 * But in case there is some other M_PROTO that looks
4759 			 * like a TPI message sent by some other kernel
4760 			 * component, we check and return an error.
4761 			 */
4762 			cr = msg_getcred(mp, NULL);
4763 			ASSERT(cr != NULL);
4764 			if (cr == NULL) {
4765 				icmp_err_ack(q, mp, TSYSERR, EINVAL);
4766 				return;
4767 			}
4768 			tpi_optcom_req(q, mp, cr, &icmp_opt_obj);
4769 			return;
4770 
4771 		case T_DISCON_REQ:
4772 			icmp_tpi_disconnect(q, mp);
4773 			return;
4774 
4775 		/* The following TPI message is not supported by icmp. */
4776 		case O_T_CONN_RES:
4777 		case T_CONN_RES:
4778 			icmp_err_ack(q, mp, TNOTSUPPORT, 0);
4779 			return;
4780 
4781 		/* The following 3 TPI requests are illegal for icmp. */
4782 		case T_DATA_REQ:
4783 		case T_EXDATA_REQ:
4784 		case T_ORDREL_REQ:
4785 			icmp_err_ack(q, mp, TNOTSUPPORT, 0);
4786 			return;
4787 		default:
4788 			break;
4789 		}
4790 		break;
4791 	case M_FLUSH:
4792 		if (*rptr & FLUSHW)
4793 			flushq(q, FLUSHDATA);
4794 		break;
4795 	case M_IOCTL:
4796 		iocp = (struct iocblk *)mp->b_rptr;
4797 		switch (iocp->ioc_cmd) {
4798 		case TI_GETPEERNAME:
4799 			if (icmp->icmp_state != TS_DATA_XFER) {
4800 				/*
4801 				 * If a default destination address has not
4802 				 * been associated with the stream, then we
4803 				 * don't know the peer's name.
4804 				 */
4805 				iocp->ioc_error = ENOTCONN;
4806 				iocp->ioc_count = 0;
4807 				mp->b_datap->db_type = M_IOCACK;
4808 				qreply(q, mp);
4809 				return;
4810 			}
4811 			/* FALLTHRU */
4812 		case TI_GETMYNAME:
4813 			/*
4814 			 * For TI_GETPEERNAME and TI_GETMYNAME, we first
4815 			 * need to copyin the user's strbuf structure.
4816 			 * Processing will continue in the M_IOCDATA case
4817 			 * below.
4818 			 */
4819 			mi_copyin(q, mp, NULL,
4820 			    SIZEOF_STRUCT(strbuf, iocp->ioc_flag));
4821 			return;
4822 		case ND_SET:
4823 			/* nd_getset performs the necessary checking */
4824 		case ND_GET:
4825 			if (nd_getset(q, is->is_nd, mp)) {
4826 				qreply(q, mp);
4827 				return;
4828 			}
4829 			break;
4830 		default:
4831 			break;
4832 		}
4833 		break;
4834 	case M_IOCDATA:
4835 		icmp_wput_iocdata(q, mp);
4836 		return;
4837 	default:
4838 		/* Unrecognized messages are passed through without change. */
4839 		break;
4840 	}
4841 	ip_wput_nondata(q, mp);
4842 }
4843 
4844 /*
4845  * icmp_wput_iocdata is called by icmp_wput_other to handle all M_IOCDATA
4846  * messages.
4847  */
4848 static void
4849 icmp_wput_iocdata(queue_t *q, mblk_t *mp)
4850 {
4851 	mblk_t		*mp1;
4852 	STRUCT_HANDLE(strbuf, sb);
4853 	uint_t		addrlen;
4854 	conn_t		*connp = Q_TO_CONN(q);
4855 	icmp_t		*icmp = connp->conn_icmp;
4856 
4857 	/* Make sure it is one of ours. */
4858 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4859 	case TI_GETMYNAME:
4860 	case TI_GETPEERNAME:
4861 		break;
4862 	default:
4863 		ip_wput_nondata(q, mp);
4864 		return;
4865 	}
4866 
4867 	switch (mi_copy_state(q, mp, &mp1)) {
4868 	case -1:
4869 		return;
4870 	case MI_COPY_CASE(MI_COPY_IN, 1):
4871 		break;
4872 	case MI_COPY_CASE(MI_COPY_OUT, 1):
4873 		/*
4874 		 * The address has been copied out, so now
4875 		 * copyout the strbuf.
4876 		 */
4877 		mi_copyout(q, mp);
4878 		return;
4879 	case MI_COPY_CASE(MI_COPY_OUT, 2):
4880 		/*
4881 		 * The address and strbuf have been copied out.
4882 		 * We're done, so just acknowledge the original
4883 		 * M_IOCTL.
4884 		 */
4885 		mi_copy_done(q, mp, 0);
4886 		return;
4887 	default:
4888 		/*
4889 		 * Something strange has happened, so acknowledge
4890 		 * the original M_IOCTL with an EPROTO error.
4891 		 */
4892 		mi_copy_done(q, mp, EPROTO);
4893 		return;
4894 	}
4895 
4896 	/*
4897 	 * Now we have the strbuf structure for TI_GETMYNAME
4898 	 * and TI_GETPEERNAME.  Next we copyout the requested
4899 	 * address and then we'll copyout the strbuf.
4900 	 */
4901 	STRUCT_SET_HANDLE(sb, ((struct iocblk *)mp->b_rptr)->ioc_flag,
4902 	    (void *)mp1->b_rptr);
4903 
4904 	if (connp->conn_family == AF_INET)
4905 		addrlen = sizeof (sin_t);
4906 	else
4907 		addrlen = sizeof (sin6_t);
4908 
4909 	if (STRUCT_FGET(sb, maxlen) < addrlen) {
4910 		mi_copy_done(q, mp, EINVAL);
4911 		return;
4912 	}
4913 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4914 	case TI_GETMYNAME:
4915 		break;
4916 	case TI_GETPEERNAME:
4917 		if (icmp->icmp_state != TS_DATA_XFER) {
4918 			mi_copy_done(q, mp, ENOTCONN);
4919 			return;
4920 		}
4921 		break;
4922 	default:
4923 		mi_copy_done(q, mp, EPROTO);
4924 		return;
4925 	}
4926 	mp1 = mi_copyout_alloc(q, mp, STRUCT_FGETP(sb, buf), addrlen, B_TRUE);
4927 	if (!mp1)
4928 		return;
4929 
4930 	STRUCT_FSET(sb, len, addrlen);
4931 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4932 	case TI_GETMYNAME:
4933 		(void) conn_getsockname(connp, (struct sockaddr *)mp1->b_wptr,
4934 		    &addrlen);
4935 		break;
4936 	case TI_GETPEERNAME:
4937 		(void) conn_getpeername(connp, (struct sockaddr *)mp1->b_wptr,
4938 		    &addrlen);
4939 		break;
4940 	}
4941 	mp1->b_wptr += addrlen;
4942 	/* Copy out the address */
4943 	mi_copyout(q, mp);
4944 }
4945 
4946 void
4947 icmp_ddi_g_init(void)
4948 {
4949 	icmp_max_optsize = optcom_max_optsize(icmp_opt_obj.odb_opt_des_arr,
4950 	    icmp_opt_obj.odb_opt_arr_cnt);
4951 
4952 	/*
4953 	 * We want to be informed each time a stack is created or
4954 	 * destroyed in the kernel, so we can maintain the
4955 	 * set of icmp_stack_t's.
4956 	 */
4957 	netstack_register(NS_ICMP, rawip_stack_init, NULL, rawip_stack_fini);
4958 }
4959 
4960 void
4961 icmp_ddi_g_destroy(void)
4962 {
4963 	netstack_unregister(NS_ICMP);
4964 }
4965 
4966 #define	INET_NAME	"ip"
4967 
4968 /*
4969  * Initialize the ICMP stack instance.
4970  */
4971 static void *
4972 rawip_stack_init(netstackid_t stackid, netstack_t *ns)
4973 {
4974 	icmp_stack_t	*is;
4975 	icmpparam_t	*pa;
4976 	int		error = 0;
4977 	major_t		major;
4978 
4979 	is = (icmp_stack_t *)kmem_zalloc(sizeof (*is), KM_SLEEP);
4980 	is->is_netstack = ns;
4981 
4982 	pa = (icmpparam_t *)kmem_alloc(sizeof (icmp_param_arr), KM_SLEEP);
4983 	is->is_param_arr = pa;
4984 	bcopy(icmp_param_arr, is->is_param_arr, sizeof (icmp_param_arr));
4985 
4986 	(void) icmp_param_register(&is->is_nd,
4987 	    is->is_param_arr, A_CNT(icmp_param_arr));
4988 	is->is_ksp = rawip_kstat_init(stackid);
4989 
4990 	major = mod_name_to_major(INET_NAME);
4991 	error = ldi_ident_from_major(major, &is->is_ldi_ident);
4992 	ASSERT(error == 0);
4993 	return (is);
4994 }
4995 
4996 /*
4997  * Free the ICMP stack instance.
4998  */
4999 static void
5000 rawip_stack_fini(netstackid_t stackid, void *arg)
5001 {
5002 	icmp_stack_t *is = (icmp_stack_t *)arg;
5003 
5004 	nd_free(&is->is_nd);
5005 	kmem_free(is->is_param_arr, sizeof (icmp_param_arr));
5006 	is->is_param_arr = NULL;
5007 
5008 	rawip_kstat_fini(stackid, is->is_ksp);
5009 	is->is_ksp = NULL;
5010 	ldi_ident_release(is->is_ldi_ident);
5011 	kmem_free(is, sizeof (*is));
5012 }
5013 
5014 static void *
5015 rawip_kstat_init(netstackid_t stackid) {
5016 	kstat_t	*ksp;
5017 
5018 	rawip_named_kstat_t template = {
5019 		{ "inDatagrams",	KSTAT_DATA_UINT32, 0 },
5020 		{ "inCksumErrs",	KSTAT_DATA_UINT32, 0 },
5021 		{ "inErrors",		KSTAT_DATA_UINT32, 0 },
5022 		{ "outDatagrams",	KSTAT_DATA_UINT32, 0 },
5023 		{ "outErrors",		KSTAT_DATA_UINT32, 0 },
5024 	};
5025 
5026 	ksp = kstat_create_netstack("icmp", 0, "rawip", "mib2",
5027 					KSTAT_TYPE_NAMED,
5028 					NUM_OF_FIELDS(rawip_named_kstat_t),
5029 					0, stackid);
5030 	if (ksp == NULL || ksp->ks_data == NULL)
5031 		return (NULL);
5032 
5033 	bcopy(&template, ksp->ks_data, sizeof (template));
5034 	ksp->ks_update = rawip_kstat_update;
5035 	ksp->ks_private = (void *)(uintptr_t)stackid;
5036 
5037 	kstat_install(ksp);
5038 	return (ksp);
5039 }
5040 
5041 static void
5042 rawip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
5043 {
5044 	if (ksp != NULL) {
5045 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
5046 		kstat_delete_netstack(ksp, stackid);
5047 	}
5048 }
5049 
5050 static int
5051 rawip_kstat_update(kstat_t *ksp, int rw)
5052 {
5053 	rawip_named_kstat_t *rawipkp;
5054 	netstackid_t	stackid = (netstackid_t)(uintptr_t)ksp->ks_private;
5055 	netstack_t	*ns;
5056 	icmp_stack_t	*is;
5057 
5058 	if ((ksp == NULL) || (ksp->ks_data == NULL))
5059 		return (EIO);
5060 
5061 	if (rw == KSTAT_WRITE)
5062 		return (EACCES);
5063 
5064 	rawipkp = (rawip_named_kstat_t *)ksp->ks_data;
5065 
5066 	ns = netstack_find_by_stackid(stackid);
5067 	if (ns == NULL)
5068 		return (-1);
5069 	is = ns->netstack_icmp;
5070 	if (is == NULL) {
5071 		netstack_rele(ns);
5072 		return (-1);
5073 	}
5074 	rawipkp->inDatagrams.value.ui32 =  is->is_rawip_mib.rawipInDatagrams;
5075 	rawipkp->inCksumErrs.value.ui32 =  is->is_rawip_mib.rawipInCksumErrs;
5076 	rawipkp->inErrors.value.ui32 =	   is->is_rawip_mib.rawipInErrors;
5077 	rawipkp->outDatagrams.value.ui32 = is->is_rawip_mib.rawipOutDatagrams;
5078 	rawipkp->outErrors.value.ui32 =	   is->is_rawip_mib.rawipOutErrors;
5079 	netstack_rele(ns);
5080 	return (0);
5081 }
5082 
5083 /* ARGSUSED */
5084 int
5085 rawip_accept(sock_lower_handle_t lproto_handle,
5086     sock_lower_handle_t eproto_handle, sock_upper_handle_t sock_handle,
5087     cred_t *cr)
5088 {
5089 	return (EOPNOTSUPP);
5090 }
5091 
5092 /* ARGSUSED */
5093 int
5094 rawip_bind(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5095     socklen_t len, cred_t *cr)
5096 {
5097 	conn_t  *connp = (conn_t *)proto_handle;
5098 	int	error;
5099 
5100 	/* All Solaris components should pass a cred for this operation. */
5101 	ASSERT(cr != NULL);
5102 
5103 	/* Binding to a NULL address really means unbind */
5104 	if (sa == NULL)
5105 		error = rawip_do_unbind(connp);
5106 	else
5107 		error = rawip_do_bind(connp, sa, len);
5108 
5109 	if (error < 0) {
5110 		if (error == -TOUTSTATE)
5111 			error = EINVAL;
5112 		else
5113 			error = proto_tlitosyserr(-error);
5114 	}
5115 	return (error);
5116 }
5117 
5118 static int
5119 rawip_implicit_bind(conn_t *connp)
5120 {
5121 	sin6_t sin6addr;
5122 	sin_t *sin;
5123 	sin6_t *sin6;
5124 	socklen_t len;
5125 	int error;
5126 
5127 	if (connp->conn_family == AF_INET) {
5128 		len = sizeof (struct sockaddr_in);
5129 		sin = (sin_t *)&sin6addr;
5130 		*sin = sin_null;
5131 		sin->sin_family = AF_INET;
5132 		sin->sin_addr.s_addr = INADDR_ANY;
5133 	} else {
5134 		ASSERT(connp->conn_family == AF_INET6);
5135 		len = sizeof (sin6_t);
5136 		sin6 = (sin6_t *)&sin6addr;
5137 		*sin6 = sin6_null;
5138 		sin6->sin6_family = AF_INET6;
5139 		V6_SET_ZERO(sin6->sin6_addr);
5140 	}
5141 
5142 	error = rawip_do_bind(connp, (struct sockaddr *)&sin6addr, len);
5143 
5144 	return ((error < 0) ? proto_tlitosyserr(-error) : error);
5145 }
5146 
5147 static int
5148 rawip_unbind(conn_t *connp)
5149 {
5150 	int error;
5151 
5152 	error = rawip_do_unbind(connp);
5153 	if (error < 0) {
5154 		error = proto_tlitosyserr(-error);
5155 	}
5156 	return (error);
5157 }
5158 
5159 /* ARGSUSED */
5160 int
5161 rawip_listen(sock_lower_handle_t proto_handle, int backlog, cred_t *cr)
5162 {
5163 	return (EOPNOTSUPP);
5164 }
5165 
5166 int
5167 rawip_connect(sock_lower_handle_t proto_handle, const struct sockaddr *sa,
5168     socklen_t len, sock_connid_t *id, cred_t *cr)
5169 {
5170 	conn_t	*connp = (conn_t *)proto_handle;
5171 	icmp_t *icmp = connp->conn_icmp;
5172 	int	error;
5173 	boolean_t did_bind = B_FALSE;
5174 	pid_t	pid = curproc->p_pid;
5175 
5176 	/* All Solaris components should pass a cred for this operation. */
5177 	ASSERT(cr != NULL);
5178 
5179 	if (sa == NULL) {
5180 		/*
5181 		 * Disconnect
5182 		 * Make sure we are connected
5183 		 */
5184 		if (icmp->icmp_state != TS_DATA_XFER)
5185 			return (EINVAL);
5186 
5187 		error = icmp_disconnect(connp);
5188 		return (error);
5189 	}
5190 
5191 	error = proto_verify_ip_addr(connp->conn_family, sa, len);
5192 	if (error != 0)
5193 		return (error);
5194 
5195 	/* do an implicit bind if necessary */
5196 	if (icmp->icmp_state == TS_UNBND) {
5197 		error = rawip_implicit_bind(connp);
5198 		/*
5199 		 * We could be racing with an actual bind, in which case
5200 		 * we would see EPROTO. We cross our fingers and try
5201 		 * to connect.
5202 		 */
5203 		if (!(error == 0 || error == EPROTO))
5204 			return (error);
5205 		did_bind = B_TRUE;
5206 	}
5207 
5208 	/*
5209 	 * set SO_DGRAM_ERRIND
5210 	 */
5211 	connp->conn_dgram_errind = B_TRUE;
5212 
5213 	error = rawip_do_connect(connp, sa, len, cr, pid);
5214 	if (error != 0 && did_bind) {
5215 		int unbind_err;
5216 
5217 		unbind_err = rawip_unbind(connp);
5218 		ASSERT(unbind_err == 0);
5219 	}
5220 
5221 	if (error == 0) {
5222 		*id = 0;
5223 		(*connp->conn_upcalls->su_connected)(connp->conn_upper_handle,
5224 		    0, NULL, -1);
5225 	} else if (error < 0) {
5226 		error = proto_tlitosyserr(-error);
5227 	}
5228 	return (error);
5229 }
5230 
5231 /* ARGSUSED2 */
5232 int
5233 rawip_fallback(sock_lower_handle_t proto_handle, queue_t *q,
5234     boolean_t direct_sockfs, so_proto_quiesced_cb_t quiesced_cb)
5235 {
5236 	conn_t  *connp = (conn_t *)proto_handle;
5237 	icmp_t	*icmp;
5238 	struct T_capability_ack tca;
5239 	struct sockaddr_in6 laddr, faddr;
5240 	socklen_t laddrlen, faddrlen;
5241 	short opts;
5242 	struct stroptions *stropt;
5243 	mblk_t *stropt_mp;
5244 	int error;
5245 
5246 	icmp = connp->conn_icmp;
5247 
5248 	stropt_mp = allocb_wait(sizeof (*stropt), BPRI_HI, STR_NOSIG, NULL);
5249 
5250 	/*
5251 	 * setup the fallback stream that was allocated
5252 	 */
5253 	connp->conn_dev = (dev_t)RD(q)->q_ptr;
5254 	connp->conn_minor_arena = WR(q)->q_ptr;
5255 
5256 	RD(q)->q_ptr = WR(q)->q_ptr = connp;
5257 
5258 	WR(q)->q_qinfo = &icmpwinit;
5259 
5260 	connp->conn_rq = RD(q);
5261 	connp->conn_wq = WR(q);
5262 
5263 	/* Notify stream head about options before sending up data */
5264 	stropt_mp->b_datap->db_type = M_SETOPTS;
5265 	stropt_mp->b_wptr += sizeof (*stropt);
5266 	stropt = (struct stroptions *)stropt_mp->b_rptr;
5267 	stropt->so_flags = SO_WROFF | SO_HIWAT;
5268 	stropt->so_wroff = connp->conn_wroff;
5269 	stropt->so_hiwat = connp->conn_rcvbuf;
5270 	putnext(RD(q), stropt_mp);
5271 
5272 	/*
5273 	 * free helper stream
5274 	 */
5275 	ip_free_helper_stream(connp);
5276 
5277 	/*
5278 	 * Collect the information needed to sync with the sonode
5279 	 */
5280 	icmp_do_capability_ack(icmp, &tca, TC1_INFO);
5281 
5282 	laddrlen = faddrlen = sizeof (sin6_t);
5283 	(void) rawip_getsockname((sock_lower_handle_t)connp,
5284 	    (struct sockaddr *)&laddr, &laddrlen, CRED());
5285 	error = rawip_getpeername((sock_lower_handle_t)connp,
5286 	    (struct sockaddr *)&faddr, &faddrlen, CRED());
5287 	if (error != 0)
5288 		faddrlen = 0;
5289 	opts = 0;
5290 	if (connp->conn_dgram_errind)
5291 		opts |= SO_DGRAM_ERRIND;
5292 	if (connp->conn_ixa->ixa_flags & IXAF_DONTROUTE)
5293 		opts |= SO_DONTROUTE;
5294 
5295 	(*quiesced_cb)(connp->conn_upper_handle, q, &tca,
5296 	    (struct sockaddr *)&laddr, laddrlen,
5297 	    (struct sockaddr *)&faddr, faddrlen, opts);
5298 
5299 	/*
5300 	 * Attempts to send data up during fallback will result in it being
5301 	 * queued in icmp_t. Now we push up any queued packets.
5302 	 */
5303 	mutex_enter(&icmp->icmp_recv_lock);
5304 	while (icmp->icmp_fallback_queue_head != NULL) {
5305 		mblk_t	*mp;
5306 
5307 		mp = icmp->icmp_fallback_queue_head;
5308 		icmp->icmp_fallback_queue_head = mp->b_next;
5309 		mp->b_next = NULL;
5310 		mutex_exit(&icmp->icmp_recv_lock);
5311 		putnext(RD(q), mp);
5312 		mutex_enter(&icmp->icmp_recv_lock);
5313 	}
5314 	icmp->icmp_fallback_queue_tail = icmp->icmp_fallback_queue_head;
5315 
5316 	/*
5317 	 * No longer a streams less socket
5318 	 */
5319 	mutex_enter(&connp->conn_lock);
5320 	connp->conn_flags &= ~IPCL_NONSTR;
5321 	mutex_exit(&connp->conn_lock);
5322 
5323 	mutex_exit(&icmp->icmp_recv_lock);
5324 
5325 	ASSERT(icmp->icmp_fallback_queue_head == NULL &&
5326 	    icmp->icmp_fallback_queue_tail == NULL);
5327 
5328 	ASSERT(connp->conn_ref >= 1);
5329 
5330 	return (0);
5331 }
5332 
5333 /* ARGSUSED2 */
5334 sock_lower_handle_t
5335 rawip_create(int family, int type, int proto, sock_downcalls_t **sock_downcalls,
5336     uint_t *smodep, int *errorp, int flags, cred_t *credp)
5337 {
5338 	conn_t *connp;
5339 
5340 	if (type != SOCK_RAW || (family != AF_INET && family != AF_INET6)) {
5341 		*errorp = EPROTONOSUPPORT;
5342 		return (NULL);
5343 	}
5344 
5345 	connp = rawip_do_open(family, credp, errorp, flags);
5346 	if (connp != NULL) {
5347 		connp->conn_flags |= IPCL_NONSTR;
5348 
5349 		mutex_enter(&connp->conn_lock);
5350 		connp->conn_state_flags &= ~CONN_INCIPIENT;
5351 		mutex_exit(&connp->conn_lock);
5352 		*sock_downcalls = &sock_rawip_downcalls;
5353 		*smodep = SM_ATOMIC;
5354 	} else {
5355 		ASSERT(*errorp != 0);
5356 	}
5357 
5358 	return ((sock_lower_handle_t)connp);
5359 }
5360 
5361 /* ARGSUSED3 */
5362 void
5363 rawip_activate(sock_lower_handle_t proto_handle,
5364     sock_upper_handle_t sock_handle, sock_upcalls_t *sock_upcalls, int flags,
5365     cred_t *cr)
5366 {
5367 	conn_t 			*connp = (conn_t *)proto_handle;
5368 	struct sock_proto_props sopp;
5369 
5370 	/* All Solaris components should pass a cred for this operation. */
5371 	ASSERT(cr != NULL);
5372 
5373 	connp->conn_upcalls = sock_upcalls;
5374 	connp->conn_upper_handle = sock_handle;
5375 
5376 	sopp.sopp_flags = SOCKOPT_WROFF | SOCKOPT_RCVHIWAT | SOCKOPT_RCVLOWAT |
5377 	    SOCKOPT_MAXBLK | SOCKOPT_MAXPSZ | SOCKOPT_MINPSZ;
5378 	sopp.sopp_wroff = connp->conn_wroff;
5379 	sopp.sopp_rxhiwat = connp->conn_rcvbuf;
5380 	sopp.sopp_rxlowat = connp->conn_rcvlowat;
5381 	sopp.sopp_maxblk = INFPSZ;
5382 	sopp.sopp_maxpsz = IP_MAXPACKET;
5383 	sopp.sopp_minpsz = (icmp_mod_info.mi_minpsz == 1) ? 0 :
5384 	    icmp_mod_info.mi_minpsz;
5385 
5386 	(*connp->conn_upcalls->su_set_proto_props)
5387 	    (connp->conn_upper_handle, &sopp);
5388 
5389 	icmp_bind_proto(connp->conn_icmp);
5390 }
5391 
5392 /* ARGSUSED3 */
5393 int
5394 rawip_getpeername(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5395     socklen_t *salenp, cred_t *cr)
5396 {
5397 	conn_t  *connp = (conn_t *)proto_handle;
5398 	icmp_t  *icmp = connp->conn_icmp;
5399 	int	error;
5400 
5401 	/* All Solaris components should pass a cred for this operation. */
5402 	ASSERT(cr != NULL);
5403 
5404 	mutex_enter(&connp->conn_lock);
5405 	if (icmp->icmp_state != TS_DATA_XFER)
5406 		error = ENOTCONN;
5407 	else
5408 		error = conn_getpeername(connp, sa, salenp);
5409 	mutex_exit(&connp->conn_lock);
5410 	return (error);
5411 }
5412 
5413 /* ARGSUSED3 */
5414 int
5415 rawip_getsockname(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5416     socklen_t *salenp, cred_t *cr)
5417 {
5418 	conn_t  *connp = (conn_t *)proto_handle;
5419 	int	error;
5420 
5421 	/* All Solaris components should pass a cred for this operation. */
5422 	ASSERT(cr != NULL);
5423 
5424 	mutex_enter(&connp->conn_lock);
5425 	error = conn_getsockname(connp, sa, salenp);
5426 	mutex_exit(&connp->conn_lock);
5427 	return (error);
5428 }
5429 
5430 int
5431 rawip_setsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
5432     const void *optvalp, socklen_t optlen, cred_t *cr)
5433 {
5434 	conn_t	*connp = (conn_t *)proto_handle;
5435 	int error;
5436 
5437 	/* All Solaris components should pass a cred for this operation. */
5438 	ASSERT(cr != NULL);
5439 
5440 	error = proto_opt_check(level, option_name, optlen, NULL,
5441 	    icmp_opt_obj.odb_opt_des_arr,
5442 	    icmp_opt_obj.odb_opt_arr_cnt,
5443 	    B_TRUE, B_FALSE, cr);
5444 
5445 	if (error != 0) {
5446 		/*
5447 		 * option not recognized
5448 		 */
5449 		if (error < 0) {
5450 			error = proto_tlitosyserr(-error);
5451 		}
5452 		return (error);
5453 	}
5454 
5455 	error = icmp_opt_set(connp, SETFN_OPTCOM_NEGOTIATE, level,
5456 	    option_name, optlen, (uchar_t *)optvalp, (uint_t *)&optlen,
5457 	    (uchar_t *)optvalp, NULL, cr);
5458 
5459 	ASSERT(error >= 0);
5460 
5461 	return (error);
5462 }
5463 
5464 int
5465 rawip_getsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
5466     void *optvalp, socklen_t *optlen, cred_t *cr)
5467 {
5468 	int		error;
5469 	conn_t		*connp = (conn_t *)proto_handle;
5470 	t_uscalar_t	max_optbuf_len;
5471 	void		*optvalp_buf;
5472 	int		len;
5473 
5474 	/* All Solaris components should pass a cred for this operation. */
5475 	ASSERT(cr != NULL);
5476 
5477 	error = proto_opt_check(level, option_name, *optlen, &max_optbuf_len,
5478 	    icmp_opt_obj.odb_opt_des_arr,
5479 	    icmp_opt_obj.odb_opt_arr_cnt,
5480 	    B_FALSE, B_TRUE, cr);
5481 
5482 	if (error != 0) {
5483 		if (error < 0) {
5484 			error = proto_tlitosyserr(-error);
5485 		}
5486 		return (error);
5487 	}
5488 
5489 	optvalp_buf = kmem_alloc(max_optbuf_len, KM_SLEEP);
5490 	len = icmp_opt_get(connp, level, option_name, optvalp_buf);
5491 	if (len == -1) {
5492 		kmem_free(optvalp_buf, max_optbuf_len);
5493 		return (EINVAL);
5494 	}
5495 
5496 	/*
5497 	 * update optlen and copy option value
5498 	 */
5499 	t_uscalar_t size = MIN(len, *optlen);
5500 
5501 	bcopy(optvalp_buf, optvalp, size);
5502 	bcopy(&size, optlen, sizeof (size));
5503 
5504 	kmem_free(optvalp_buf, max_optbuf_len);
5505 	return (0);
5506 }
5507 
5508 /* ARGSUSED1 */
5509 int
5510 rawip_close(sock_lower_handle_t proto_handle, int flags, cred_t *cr)
5511 {
5512 	conn_t	*connp = (conn_t *)proto_handle;
5513 
5514 	/* All Solaris components should pass a cred for this operation. */
5515 	ASSERT(cr != NULL);
5516 
5517 	(void) rawip_do_close(connp);
5518 	return (0);
5519 }
5520 
5521 /* ARGSUSED2 */
5522 int
5523 rawip_shutdown(sock_lower_handle_t proto_handle, int how, cred_t *cr)
5524 {
5525 	conn_t  *connp = (conn_t *)proto_handle;
5526 
5527 	/* All Solaris components should pass a cred for this operation. */
5528 	ASSERT(cr != NULL);
5529 
5530 	/* shut down the send side */
5531 	if (how != SHUT_RD)
5532 		(*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
5533 		    SOCK_OPCTL_SHUT_SEND, 0);
5534 	/* shut down the recv side */
5535 	if (how != SHUT_WR)
5536 		(*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
5537 		    SOCK_OPCTL_SHUT_RECV, 0);
5538 	return (0);
5539 }
5540 
5541 void
5542 rawip_clr_flowctrl(sock_lower_handle_t proto_handle)
5543 {
5544 	conn_t  *connp = (conn_t *)proto_handle;
5545 	icmp_t	*icmp = connp->conn_icmp;
5546 
5547 	mutex_enter(&icmp->icmp_recv_lock);
5548 	connp->conn_flow_cntrld = B_FALSE;
5549 	mutex_exit(&icmp->icmp_recv_lock);
5550 }
5551 
5552 int
5553 rawip_ioctl(sock_lower_handle_t proto_handle, int cmd, intptr_t arg,
5554     int mode, int32_t *rvalp, cred_t *cr)
5555 {
5556 	conn_t  	*connp = (conn_t *)proto_handle;
5557 	int		error;
5558 
5559 	/* All Solaris components should pass a cred for this operation. */
5560 	ASSERT(cr != NULL);
5561 
5562 	/*
5563 	 * If we don't have a helper stream then create one.
5564 	 * ip_create_helper_stream takes care of locking the conn_t,
5565 	 * so this check for NULL is just a performance optimization.
5566 	 */
5567 	if (connp->conn_helper_info == NULL) {
5568 		icmp_stack_t *is = connp->conn_icmp->icmp_is;
5569 
5570 		ASSERT(is->is_ldi_ident != NULL);
5571 
5572 		/*
5573 		 * Create a helper stream for non-STREAMS socket.
5574 		 */
5575 		error = ip_create_helper_stream(connp, is->is_ldi_ident);
5576 		if (error != 0) {
5577 			ip0dbg(("rawip_ioctl: create of IP helper stream "
5578 			    "failed %d\n", error));
5579 			return (error);
5580 		}
5581 	}
5582 
5583 	switch (cmd) {
5584 	case ND_SET:
5585 	case ND_GET:
5586 	case _SIOCSOCKFALLBACK:
5587 	case TI_GETPEERNAME:
5588 	case TI_GETMYNAME:
5589 #ifdef DEBUG
5590 		cmn_err(CE_CONT, "icmp_ioctl cmd 0x%x on non streams"
5591 		    " socket", cmd);
5592 #endif
5593 		error = EINVAL;
5594 		break;
5595 	default:
5596 		/*
5597 		 * Pass on to IP using helper stream
5598 		 */
5599 		error = ldi_ioctl(connp->conn_helper_info->iphs_handle,
5600 		    cmd, arg, mode, cr, rvalp);
5601 		break;
5602 	}
5603 	return (error);
5604 }
5605 
5606 int
5607 rawip_send(sock_lower_handle_t proto_handle, mblk_t *mp, struct nmsghdr *msg,
5608     cred_t *cr)
5609 {
5610 	sin6_t		*sin6;
5611 	sin_t		*sin = NULL;
5612 	uint_t		srcid;
5613 	conn_t		*connp = (conn_t *)proto_handle;
5614 	icmp_t		*icmp = connp->conn_icmp;
5615 	int		error = 0;
5616 	icmp_stack_t	*is = icmp->icmp_is;
5617 	pid_t		pid = curproc->p_pid;
5618 	ip_xmit_attr_t	*ixa;
5619 
5620 	ASSERT(DB_TYPE(mp) == M_DATA);
5621 
5622 	/* All Solaris components should pass a cred for this operation. */
5623 	ASSERT(cr != NULL);
5624 
5625 	/* do an implicit bind if necessary */
5626 	if (icmp->icmp_state == TS_UNBND) {
5627 		error = rawip_implicit_bind(connp);
5628 		/*
5629 		 * We could be racing with an actual bind, in which case
5630 		 * we would see EPROTO. We cross our fingers and try
5631 		 * to connect.
5632 		 */
5633 		if (!(error == 0 || error == EPROTO)) {
5634 			freemsg(mp);
5635 			return (error);
5636 		}
5637 	}
5638 
5639 	/* Protocol 255 contains full IP headers */
5640 	/* Read without holding lock */
5641 	if (icmp->icmp_hdrincl) {
5642 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
5643 		if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH) {
5644 			if (!pullupmsg(mp, IP_SIMPLE_HDR_LENGTH)) {
5645 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5646 				freemsg(mp);
5647 				return (EINVAL);
5648 			}
5649 		}
5650 		error = icmp_output_hdrincl(connp, mp, cr, pid);
5651 		if (is->is_sendto_ignerr)
5652 			return (0);
5653 		else
5654 			return (error);
5655 	}
5656 
5657 	/* Connected? */
5658 	if (msg->msg_name == NULL) {
5659 		if (icmp->icmp_state != TS_DATA_XFER) {
5660 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5661 			return (EDESTADDRREQ);
5662 		}
5663 		if (msg->msg_controllen != 0) {
5664 			error = icmp_output_ancillary(connp, NULL, NULL, mp,
5665 			    NULL, msg, cr, pid);
5666 		} else {
5667 			error = icmp_output_connected(connp, mp, cr, pid);
5668 		}
5669 		if (is->is_sendto_ignerr)
5670 			return (0);
5671 		else
5672 			return (error);
5673 	}
5674 	if (icmp->icmp_state == TS_DATA_XFER) {
5675 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5676 		return (EISCONN);
5677 	}
5678 	error = proto_verify_ip_addr(connp->conn_family,
5679 	    (struct sockaddr *)msg->msg_name, msg->msg_namelen);
5680 	if (error != 0) {
5681 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5682 		return (error);
5683 	}
5684 	switch (connp->conn_family) {
5685 	case AF_INET6:
5686 		sin6 = (sin6_t *)msg->msg_name;
5687 
5688 		/* No support for mapped addresses on raw sockets */
5689 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
5690 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5691 			return (EADDRNOTAVAIL);
5692 		}
5693 		srcid = sin6->__sin6_src_id;
5694 
5695 		/*
5696 		 * If the local address is a mapped address return
5697 		 * an error.
5698 		 * It would be possible to send an IPv6 packet but the
5699 		 * response would never make it back to the application
5700 		 * since it is bound to a mapped address.
5701 		 */
5702 		if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
5703 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5704 			return (EADDRNOTAVAIL);
5705 		}
5706 
5707 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
5708 			sin6->sin6_addr = ipv6_loopback;
5709 
5710 		/*
5711 		 * We have to allocate an ip_xmit_attr_t before we grab
5712 		 * conn_lock and we need to hold conn_lock once we've check
5713 		 * conn_same_as_last_v6 to handle concurrent send* calls on a
5714 		 * socket.
5715 		 */
5716 		if (msg->msg_controllen == 0) {
5717 			ixa = conn_get_ixa(connp, B_FALSE);
5718 			if (ixa == NULL) {
5719 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5720 				return (ENOMEM);
5721 			}
5722 		} else {
5723 			ixa = NULL;
5724 		}
5725 		mutex_enter(&connp->conn_lock);
5726 		if (icmp->icmp_delayed_error != 0) {
5727 			sin6_t  *sin2 = (sin6_t *)&icmp->icmp_delayed_addr;
5728 
5729 			error = icmp->icmp_delayed_error;
5730 			icmp->icmp_delayed_error = 0;
5731 
5732 			/* Compare IP address and family */
5733 
5734 			if (IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr,
5735 			    &sin2->sin6_addr) &&
5736 			    sin6->sin6_family == sin2->sin6_family) {
5737 				mutex_exit(&connp->conn_lock);
5738 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5739 				if (ixa != NULL)
5740 					ixa_refrele(ixa);
5741 				return (error);
5742 			}
5743 		}
5744 		if (msg->msg_controllen != 0) {
5745 			mutex_exit(&connp->conn_lock);
5746 			ASSERT(ixa == NULL);
5747 			error = icmp_output_ancillary(connp, NULL, sin6, mp,
5748 			    NULL, msg, cr, pid);
5749 		} else if (conn_same_as_last_v6(connp, sin6) &&
5750 		    connp->conn_lastsrcid == srcid &&
5751 		    ipsec_outbound_policy_current(ixa)) {
5752 			/* icmp_output_lastdst drops conn_lock */
5753 			error = icmp_output_lastdst(connp, mp, cr, pid, ixa);
5754 		} else {
5755 			/* icmp_output_newdst drops conn_lock */
5756 			error = icmp_output_newdst(connp, mp, NULL, sin6, cr,
5757 			    pid, ixa);
5758 		}
5759 		ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
5760 		if (is->is_sendto_ignerr)
5761 			return (0);
5762 		else
5763 			return (error);
5764 	case AF_INET:
5765 		sin = (sin_t *)msg->msg_name;
5766 
5767 		if (sin->sin_addr.s_addr == INADDR_ANY)
5768 			sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
5769 
5770 		/*
5771 		 * We have to allocate an ip_xmit_attr_t before we grab
5772 		 * conn_lock and we need to hold conn_lock once we've check
5773 		 * conn_same_as_last_v6 to handle concurrent send* on a socket.
5774 		 */
5775 		if (msg->msg_controllen == 0) {
5776 			ixa = conn_get_ixa(connp, B_FALSE);
5777 			if (ixa == NULL) {
5778 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5779 				return (ENOMEM);
5780 			}
5781 		} else {
5782 			ixa = NULL;
5783 		}
5784 		mutex_enter(&connp->conn_lock);
5785 		if (icmp->icmp_delayed_error != 0) {
5786 			sin_t  *sin2 = (sin_t *)&icmp->icmp_delayed_addr;
5787 
5788 			error = icmp->icmp_delayed_error;
5789 			icmp->icmp_delayed_error = 0;
5790 
5791 			/* Compare IP address */
5792 
5793 			if (sin->sin_addr.s_addr == sin2->sin_addr.s_addr) {
5794 				mutex_exit(&connp->conn_lock);
5795 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5796 				if (ixa != NULL)
5797 					ixa_refrele(ixa);
5798 				return (error);
5799 			}
5800 		}
5801 
5802 		if (msg->msg_controllen != 0) {
5803 			mutex_exit(&connp->conn_lock);
5804 			ASSERT(ixa == NULL);
5805 			error = icmp_output_ancillary(connp, sin, NULL, mp,
5806 			    NULL, msg, cr, pid);
5807 		} else if (conn_same_as_last_v4(connp, sin) &&
5808 		    ipsec_outbound_policy_current(ixa)) {
5809 			/* icmp_output_lastdst drops conn_lock */
5810 			error = icmp_output_lastdst(connp, mp, cr, pid, ixa);
5811 		} else {
5812 			/* icmp_output_newdst drops conn_lock */
5813 			error = icmp_output_newdst(connp, mp, sin, NULL, cr,
5814 			    pid, ixa);
5815 		}
5816 		ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
5817 		if (is->is_sendto_ignerr)
5818 			return (0);
5819 		else
5820 			return (error);
5821 	default:
5822 		return (EINVAL);
5823 	}
5824 }
5825 
5826 sock_downcalls_t sock_rawip_downcalls = {
5827 	rawip_activate,
5828 	rawip_accept,
5829 	rawip_bind,
5830 	rawip_listen,
5831 	rawip_connect,
5832 	rawip_getpeername,
5833 	rawip_getsockname,
5834 	rawip_getsockopt,
5835 	rawip_setsockopt,
5836 	rawip_send,
5837 	NULL,
5838 	NULL,
5839 	NULL,
5840 	rawip_shutdown,
5841 	rawip_clr_flowctrl,
5842 	rawip_ioctl,
5843 	rawip_close
5844 };
5845