xref: /illumos-gate/usr/src/uts/common/inet/ip/icmp.c (revision 9d468d1aa2ff370e3e4b0437df3e0b065fb645ea)
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 /* Default structure copied into T_INFO_ACK messages */
207 static struct T_info_ack icmp_g_t_info_ack = {
208 	T_INFO_ACK,
209 	IP_MAXPACKET,	 /* TSDU_size.  icmp allows maximum size messages. */
210 	T_INVALID,	/* ETSDU_size.  icmp does not support expedited data. */
211 	T_INVALID,	/* CDATA_size. icmp does not support connect data. */
212 	T_INVALID,	/* DDATA_size. icmp does not support disconnect data. */
213 	0,		/* ADDR_size - filled in later. */
214 	0,		/* OPT_size - not initialized here */
215 	IP_MAXPACKET,	/* TIDU_size.  icmp allows maximum size messages. */
216 	T_CLTS,		/* SERV_type.  icmp supports connection-less. */
217 	TS_UNBND,	/* CURRENT_state.  This is set from icmp_state. */
218 	(XPG4_1|SENDZERO) /* PROVIDER_flag */
219 };
220 
221 /*
222  * Table of ND variables supported by icmp.  These are loaded into is_nd
223  * when the stack instance is created.
224  * All of these are alterable, within the min/max values given, at run time.
225  */
226 static icmpparam_t	icmp_param_arr[] = {
227 	/* min	max	value	name */
228 	{ 0,	128,	32,	"icmp_wroff_extra" },
229 	{ 1,	255,	255,	"icmp_ipv4_ttl" },
230 	{ 0, IPV6_MAX_HOPS, IPV6_DEFAULT_HOPS,	"icmp_ipv6_hoplimit"},
231 	{ 0,	1,	1,	"icmp_bsd_compat" },
232 	{ 4096,	65536,	8192,	"icmp_xmit_hiwat"},
233 	{ 0,	65536,	1024,	"icmp_xmit_lowat"},
234 	{ 4096,	65536,	8192,	"icmp_recv_hiwat"},
235 	{ 65536, 1024*1024*1024, 256*1024,	"icmp_max_buf"},
236 	{ 0,	1,	0,	"icmp_pmtu_discovery" },
237 	{ 0,	1,	0,	"icmp_sendto_ignerr" },
238 };
239 #define	is_wroff_extra			is_param_arr[0].icmp_param_value
240 #define	is_ipv4_ttl			is_param_arr[1].icmp_param_value
241 #define	is_ipv6_hoplimit		is_param_arr[2].icmp_param_value
242 #define	is_bsd_compat			is_param_arr[3].icmp_param_value
243 #define	is_xmit_hiwat			is_param_arr[4].icmp_param_value
244 #define	is_xmit_lowat			is_param_arr[5].icmp_param_value
245 #define	is_recv_hiwat			is_param_arr[6].icmp_param_value
246 #define	is_max_buf			is_param_arr[7].icmp_param_value
247 #define	is_pmtu_discovery		is_param_arr[8].icmp_param_value
248 #define	is_sendto_ignerr		is_param_arr[9].icmp_param_value
249 
250 typedef union T_primitives *t_primp_t;
251 
252 /*
253  * This routine is called to handle each O_T_BIND_REQ/T_BIND_REQ message
254  * passed to icmp_wput.
255  * It calls IP to verify the local IP address, and calls IP to insert
256  * the conn_t in the fanout table.
257  * If everything is ok it then sends the T_BIND_ACK back up.
258  */
259 static void
260 icmp_tpi_bind(queue_t *q, mblk_t *mp)
261 {
262 	int	error;
263 	struct sockaddr *sa;
264 	struct T_bind_req *tbr;
265 	socklen_t	len;
266 	sin_t	*sin;
267 	sin6_t	*sin6;
268 	icmp_t		*icmp;
269 	conn_t	*connp = Q_TO_CONN(q);
270 	mblk_t *mp1;
271 	cred_t *cr;
272 
273 	/*
274 	 * All Solaris components should pass a db_credp
275 	 * for this TPI message, hence we ASSERT.
276 	 * But in case there is some other M_PROTO that looks
277 	 * like a TPI message sent by some other kernel
278 	 * component, we check and return an error.
279 	 */
280 	cr = msg_getcred(mp, NULL);
281 	ASSERT(cr != NULL);
282 	if (cr == NULL) {
283 		icmp_err_ack(q, mp, TSYSERR, EINVAL);
284 		return;
285 	}
286 
287 	icmp = connp->conn_icmp;
288 	if ((mp->b_wptr - mp->b_rptr) < sizeof (*tbr)) {
289 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
290 		    "icmp_bind: bad req, len %u",
291 		    (uint_t)(mp->b_wptr - mp->b_rptr));
292 		icmp_err_ack(q, mp, TPROTO, 0);
293 		return;
294 	}
295 
296 	if (icmp->icmp_state != TS_UNBND) {
297 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
298 		    "icmp_bind: bad state, %u", icmp->icmp_state);
299 		icmp_err_ack(q, mp, TOUTSTATE, 0);
300 		return;
301 	}
302 
303 	/*
304 	 * Reallocate the message to make sure we have enough room for an
305 	 * address.
306 	 */
307 	mp1 = reallocb(mp, sizeof (struct T_bind_ack) + sizeof (sin6_t), 1);
308 	if (mp1 == NULL) {
309 		icmp_err_ack(q, mp, TSYSERR, ENOMEM);
310 		return;
311 	}
312 	mp = mp1;
313 
314 	/* Reset the message type in preparation for shipping it back. */
315 	DB_TYPE(mp) = M_PCPROTO;
316 	tbr = (struct T_bind_req *)mp->b_rptr;
317 	len = tbr->ADDR_length;
318 	switch (len) {
319 	case 0:	/* request for a generic port */
320 		tbr->ADDR_offset = sizeof (struct T_bind_req);
321 		if (connp->conn_family == AF_INET) {
322 			tbr->ADDR_length = sizeof (sin_t);
323 			sin = (sin_t *)&tbr[1];
324 			*sin = sin_null;
325 			sin->sin_family = AF_INET;
326 			mp->b_wptr = (uchar_t *)&sin[1];
327 			sa = (struct sockaddr *)sin;
328 			len = sizeof (sin_t);
329 		} else {
330 			ASSERT(connp->conn_family == AF_INET6);
331 			tbr->ADDR_length = sizeof (sin6_t);
332 			sin6 = (sin6_t *)&tbr[1];
333 			*sin6 = sin6_null;
334 			sin6->sin6_family = AF_INET6;
335 			mp->b_wptr = (uchar_t *)&sin6[1];
336 			sa = (struct sockaddr *)sin6;
337 			len = sizeof (sin6_t);
338 		}
339 		break;
340 
341 	case sizeof (sin_t):	/* Complete IPv4 address */
342 		sa = (struct sockaddr *)mi_offset_param(mp, tbr->ADDR_offset,
343 		    sizeof (sin_t));
344 		break;
345 
346 	case sizeof (sin6_t):	/* Complete IPv6 address */
347 		sa = (struct sockaddr *)mi_offset_param(mp,
348 		    tbr->ADDR_offset, sizeof (sin6_t));
349 		break;
350 
351 	default:
352 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
353 		    "icmp_bind: bad ADDR_length %u", tbr->ADDR_length);
354 		icmp_err_ack(q, mp, TBADADDR, 0);
355 		return;
356 	}
357 
358 	error = rawip_do_bind(connp, sa, len);
359 	if (error != 0) {
360 		if (error > 0) {
361 			icmp_err_ack(q, mp, TSYSERR, error);
362 		} else {
363 			icmp_err_ack(q, mp, -error, 0);
364 		}
365 	} else {
366 		tbr->PRIM_type = T_BIND_ACK;
367 		qreply(q, mp);
368 	}
369 }
370 
371 static int
372 rawip_do_bind(conn_t *connp, struct sockaddr *sa, socklen_t len)
373 {
374 	sin_t		*sin;
375 	sin6_t		*sin6;
376 	icmp_t		*icmp = connp->conn_icmp;
377 	int		error = 0;
378 	ip_laddr_t	laddr_type = IPVL_UNICAST_UP;	/* INADDR_ANY */
379 	in_port_t	lport;		/* Network byte order */
380 	ipaddr_t	v4src;		/* Set if AF_INET */
381 	in6_addr_t	v6src;
382 	uint_t		scopeid = 0;
383 	zoneid_t	zoneid = IPCL_ZONEID(connp);
384 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
385 
386 	if (sa == NULL || !OK_32PTR((char *)sa)) {
387 		return (EINVAL);
388 	}
389 
390 	switch (len) {
391 	case sizeof (sin_t):    /* Complete IPv4 address */
392 		sin = (sin_t *)sa;
393 		if (sin->sin_family != AF_INET ||
394 		    connp->conn_family != AF_INET) {
395 			/* TSYSERR, EAFNOSUPPORT */
396 			return (EAFNOSUPPORT);
397 		}
398 		v4src = sin->sin_addr.s_addr;
399 		IN6_IPADDR_TO_V4MAPPED(v4src, &v6src);
400 		if (v4src != INADDR_ANY) {
401 			laddr_type = ip_laddr_verify_v4(v4src, zoneid, ipst,
402 			    B_TRUE);
403 		}
404 		lport = sin->sin_port;
405 		break;
406 	case sizeof (sin6_t): /* Complete IPv6 address */
407 		sin6 = (sin6_t *)sa;
408 		if (sin6->sin6_family != AF_INET6 ||
409 		    connp->conn_family != AF_INET6) {
410 			/* TSYSERR, EAFNOSUPPORT */
411 			return (EAFNOSUPPORT);
412 		}
413 		/* No support for mapped addresses on raw sockets */
414 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
415 			/* TSYSERR, EADDRNOTAVAIL */
416 			return (EADDRNOTAVAIL);
417 		}
418 		v6src = sin6->sin6_addr;
419 		if (!IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
420 			if (IN6_IS_ADDR_LINKSCOPE(&v6src))
421 				scopeid = sin6->sin6_scope_id;
422 			laddr_type = ip_laddr_verify_v6(&v6src, zoneid, ipst,
423 			    B_TRUE, scopeid);
424 		}
425 		lport = sin6->sin6_port;
426 		break;
427 
428 	default:
429 		/* TBADADDR */
430 		return (EADDRNOTAVAIL);
431 	}
432 
433 	/* Is the local address a valid unicast, multicast, or broadcast? */
434 	if (laddr_type == IPVL_BAD)
435 		return (EADDRNOTAVAIL);
436 
437 	/*
438 	 * The state must be TS_UNBND.
439 	 */
440 	mutex_enter(&connp->conn_lock);
441 	if (icmp->icmp_state != TS_UNBND) {
442 		mutex_exit(&connp->conn_lock);
443 		return (-TOUTSTATE);
444 	}
445 
446 	/*
447 	 * Copy the source address into our icmp structure.  This address
448 	 * may still be zero; if so, ip will fill in the correct address
449 	 * each time an outbound packet is passed to it.
450 	 * If we are binding to a broadcast or multicast address then
451 	 * we just set the conn_bound_addr since we don't want to use
452 	 * that as the source address when sending.
453 	 */
454 	connp->conn_bound_addr_v6 = v6src;
455 	connp->conn_laddr_v6 = v6src;
456 	if (scopeid != 0) {
457 		connp->conn_ixa->ixa_flags |= IXAF_SCOPEID_SET;
458 		connp->conn_ixa->ixa_scopeid = scopeid;
459 		connp->conn_incoming_ifindex = scopeid;
460 	} else {
461 		connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
462 		connp->conn_incoming_ifindex = connp->conn_bound_if;
463 	}
464 
465 	switch (laddr_type) {
466 	case IPVL_UNICAST_UP:
467 	case IPVL_UNICAST_DOWN:
468 		connp->conn_saddr_v6 = v6src;
469 		connp->conn_mcbc_bind = B_FALSE;
470 		break;
471 	case IPVL_MCAST:
472 	case IPVL_BCAST:
473 		/* ip_set_destination will pick a source address later */
474 		connp->conn_saddr_v6 = ipv6_all_zeros;
475 		connp->conn_mcbc_bind = B_TRUE;
476 		break;
477 	}
478 
479 	/* Any errors after this point should use late_error */
480 
481 	/*
482 	 * Use sin_port/sin6_port since applications like psh use SOCK_RAW
483 	 * with IPPROTO_TCP.
484 	 */
485 	connp->conn_lport = lport;
486 	connp->conn_fport = 0;
487 
488 	if (connp->conn_family == AF_INET) {
489 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
490 	} else {
491 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
492 	}
493 
494 	icmp->icmp_state = TS_IDLE;
495 
496 	/*
497 	 * We create an initial header template here to make a subsequent
498 	 * sendto have a starting point. Since conn_last_dst is zero the
499 	 * first sendto will always follow the 'dst changed' code path.
500 	 * Note that we defer massaging options and the related checksum
501 	 * adjustment until we have a destination address.
502 	 */
503 	error = icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
504 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
505 	if (error != 0) {
506 		mutex_exit(&connp->conn_lock);
507 		goto late_error;
508 	}
509 	/* Just in case */
510 	connp->conn_faddr_v6 = ipv6_all_zeros;
511 	connp->conn_v6lastdst = ipv6_all_zeros;
512 	mutex_exit(&connp->conn_lock);
513 
514 	error = ip_laddr_fanout_insert(connp);
515 	if (error != 0)
516 		goto late_error;
517 
518 	/* Bind succeeded */
519 	return (0);
520 
521 late_error:
522 	mutex_enter(&connp->conn_lock);
523 	connp->conn_saddr_v6 = ipv6_all_zeros;
524 	connp->conn_bound_addr_v6 = ipv6_all_zeros;
525 	connp->conn_laddr_v6 = ipv6_all_zeros;
526 	if (scopeid != 0) {
527 		connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
528 		connp->conn_incoming_ifindex = connp->conn_bound_if;
529 	}
530 	icmp->icmp_state = TS_UNBND;
531 	connp->conn_v6lastdst = ipv6_all_zeros;
532 	connp->conn_lport = 0;
533 
534 	/* Restore the header that was built above - different source address */
535 	(void) icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
536 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
537 	mutex_exit(&connp->conn_lock);
538 	return (error);
539 }
540 
541 /*
542  * Tell IP to just bind to the protocol.
543  */
544 static void
545 icmp_bind_proto(icmp_t *icmp)
546 {
547 	conn_t	*connp = icmp->icmp_connp;
548 
549 	mutex_enter(&connp->conn_lock);
550 	connp->conn_saddr_v6 = ipv6_all_zeros;
551 	connp->conn_laddr_v6 = ipv6_all_zeros;
552 	connp->conn_faddr_v6 = ipv6_all_zeros;
553 	connp->conn_v6lastdst = ipv6_all_zeros;
554 	mutex_exit(&connp->conn_lock);
555 
556 	(void) ip_laddr_fanout_insert(connp);
557 }
558 
559 /*
560  * This routine handles each T_CONN_REQ message passed to icmp.  It
561  * associates a default destination address with the stream.
562  *
563  * After various error checks are completed, icmp_connect() lays
564  * the target address and port into the composite header template.
565  * Then we ask IP for information, including a source address if we didn't
566  * already have one. Finally we send up the T_OK_ACK reply message.
567  */
568 static void
569 icmp_tpi_connect(queue_t *q, mblk_t *mp)
570 {
571 	conn_t	*connp = Q_TO_CONN(q);
572 	struct T_conn_req	*tcr;
573 	struct sockaddr *sa;
574 	socklen_t len;
575 	int error;
576 	cred_t *cr;
577 	pid_t pid;
578 	/*
579 	 * All Solaris components should pass a db_credp
580 	 * for this TPI message, hence we ASSERT.
581 	 * But in case there is some other M_PROTO that looks
582 	 * like a TPI message sent by some other kernel
583 	 * component, we check and return an error.
584 	 */
585 	cr = msg_getcred(mp, &pid);
586 	ASSERT(cr != NULL);
587 	if (cr == NULL) {
588 		icmp_err_ack(q, mp, TSYSERR, EINVAL);
589 		return;
590 	}
591 
592 	tcr = (struct T_conn_req *)mp->b_rptr;
593 	/* Sanity checks */
594 	if ((mp->b_wptr - mp->b_rptr) < sizeof (struct T_conn_req)) {
595 		icmp_err_ack(q, mp, TPROTO, 0);
596 		return;
597 	}
598 
599 	if (tcr->OPT_length != 0) {
600 		icmp_err_ack(q, mp, TBADOPT, 0);
601 		return;
602 	}
603 
604 	len = tcr->DEST_length;
605 
606 	switch (len) {
607 	default:
608 		icmp_err_ack(q, mp, TBADADDR, 0);
609 		return;
610 	case sizeof (sin_t):
611 		sa = (struct sockaddr *)mi_offset_param(mp, tcr->DEST_offset,
612 		    sizeof (sin_t));
613 		break;
614 	case sizeof (sin6_t):
615 		sa = (struct sockaddr *)mi_offset_param(mp,
616 		    tcr->DEST_offset, sizeof (sin6_t));
617 		break;
618 	}
619 
620 	error = proto_verify_ip_addr(connp->conn_family, sa, len);
621 	if (error != 0) {
622 		icmp_err_ack(q, mp, TSYSERR, error);
623 		return;
624 	}
625 
626 	error = rawip_do_connect(connp, sa, len, cr, pid);
627 	if (error != 0) {
628 		if (error < 0) {
629 			icmp_err_ack(q, mp, -error, 0);
630 		} else {
631 			icmp_err_ack(q, mp, 0, error);
632 		}
633 	} else {
634 		mblk_t *mp1;
635 
636 		/*
637 		 * We have to send a connection confirmation to
638 		 * keep TLI happy.
639 		 */
640 		if (connp->conn_family == AF_INET) {
641 			mp1 = mi_tpi_conn_con(NULL, (char *)sa,
642 			    sizeof (sin_t), NULL, 0);
643 		} else {
644 			ASSERT(connp->conn_family == AF_INET6);
645 			mp1 = mi_tpi_conn_con(NULL, (char *)sa,
646 			    sizeof (sin6_t), NULL, 0);
647 		}
648 		if (mp1 == NULL) {
649 			icmp_err_ack(q, mp, TSYSERR, ENOMEM);
650 			return;
651 		}
652 
653 		/*
654 		 * Send ok_ack for T_CONN_REQ
655 		 */
656 		mp = mi_tpi_ok_ack_alloc(mp);
657 		if (mp == NULL) {
658 			/* Unable to reuse the T_CONN_REQ for the ack. */
659 			icmp_err_ack_prim(q, mp1, T_CONN_REQ, TSYSERR, ENOMEM);
660 			return;
661 		}
662 		putnext(connp->conn_rq, mp);
663 		putnext(connp->conn_rq, mp1);
664 	}
665 }
666 
667 static int
668 rawip_do_connect(conn_t *connp, const struct sockaddr *sa, socklen_t len,
669     cred_t *cr, pid_t pid)
670 {
671 	icmp_t		*icmp;
672 	sin_t		*sin;
673 	sin6_t		*sin6;
674 	int		error;
675 	uint16_t 	dstport;
676 	ipaddr_t	v4dst;
677 	in6_addr_t	v6dst;
678 	uint32_t	flowinfo;
679 	ip_xmit_attr_t	*ixa;
680 	uint_t		scopeid = 0;
681 	uint_t		srcid = 0;
682 	in6_addr_t	v6src = connp->conn_saddr_v6;
683 
684 	icmp = connp->conn_icmp;
685 
686 	if (sa == NULL || !OK_32PTR((char *)sa)) {
687 		return (EINVAL);
688 	}
689 
690 	ASSERT(sa != NULL && len != 0);
691 
692 	/*
693 	 * Determine packet type based on type of address passed in
694 	 * the request should contain an IPv4 or IPv6 address.
695 	 * Make sure that address family matches the type of
696 	 * family of the address passed down.
697 	 */
698 	switch (len) {
699 	case sizeof (sin_t):
700 		sin = (sin_t *)sa;
701 
702 		v4dst = sin->sin_addr.s_addr;
703 		dstport = sin->sin_port;
704 		IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
705 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
706 		break;
707 
708 	case sizeof (sin6_t):
709 		sin6 = (sin6_t *)sa;
710 
711 		/* No support for mapped addresses on raw sockets */
712 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
713 			return (EADDRNOTAVAIL);
714 		}
715 		v6dst = sin6->sin6_addr;
716 		dstport = sin6->sin6_port;
717 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
718 		flowinfo = sin6->sin6_flowinfo;
719 		if (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr))
720 			scopeid = sin6->sin6_scope_id;
721 		srcid = sin6->__sin6_src_id;
722 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
723 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
724 			    connp->conn_netstack);
725 		}
726 		break;
727 	}
728 
729 	/*
730 	 * If there is a different thread using conn_ixa then we get a new
731 	 * copy and cut the old one loose from conn_ixa. Otherwise we use
732 	 * conn_ixa and prevent any other thread from using/changing it.
733 	 * Once connect() is done other threads can use conn_ixa since the
734 	 * refcnt will be back at one.
735 	 */
736 	ixa = conn_get_ixa(connp, B_TRUE);
737 	if (ixa == NULL)
738 		return (ENOMEM);
739 
740 	ASSERT(ixa->ixa_refcnt >= 2);
741 	ASSERT(ixa == connp->conn_ixa);
742 
743 	mutex_enter(&connp->conn_lock);
744 	/*
745 	 * This icmp_t must have bound already before doing a connect.
746 	 * Reject if a connect is in progress (we drop conn_lock during
747 	 * rawip_do_connect).
748 	 */
749 	if (icmp->icmp_state == TS_UNBND || icmp->icmp_state == TS_WCON_CREQ) {
750 		mutex_exit(&connp->conn_lock);
751 		ixa_refrele(ixa);
752 		return (-TOUTSTATE);
753 	}
754 
755 	if (icmp->icmp_state == TS_DATA_XFER) {
756 		/* Already connected - clear out state */
757 		if (connp->conn_mcbc_bind)
758 			connp->conn_saddr_v6 = ipv6_all_zeros;
759 		else
760 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
761 		connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
762 		connp->conn_faddr_v6 = ipv6_all_zeros;
763 		icmp->icmp_state = TS_IDLE;
764 	}
765 
766 	/*
767 	 * Use sin_port/sin6_port since applications like psh use SOCK_RAW
768 	 * with IPPROTO_TCP.
769 	 */
770 	connp->conn_fport = dstport;
771 	if (connp->conn_ipversion == IPV4_VERSION) {
772 		/*
773 		 * Interpret a zero destination to mean loopback.
774 		 * Update the T_CONN_REQ (sin/sin6) since it is used to
775 		 * generate the T_CONN_CON.
776 		 */
777 		if (v4dst == INADDR_ANY) {
778 			v4dst = htonl(INADDR_LOOPBACK);
779 			IN6_IPADDR_TO_V4MAPPED(v4dst, &v6dst);
780 			ASSERT(connp->conn_family == AF_INET);
781 			sin->sin_addr.s_addr = v4dst;
782 		}
783 		connp->conn_faddr_v6 = v6dst;
784 		connp->conn_flowinfo = 0;
785 	} else {
786 		ASSERT(connp->conn_ipversion == IPV6_VERSION);
787 		/*
788 		 * Interpret a zero destination to mean loopback.
789 		 * Update the T_CONN_REQ (sin/sin6) since it is used to
790 		 * generate the T_CONN_CON.
791 		 */
792 		if (IN6_IS_ADDR_UNSPECIFIED(&v6dst)) {
793 			v6dst = ipv6_loopback;
794 			sin6->sin6_addr = v6dst;
795 		}
796 		connp->conn_faddr_v6 = v6dst;
797 		connp->conn_flowinfo = flowinfo;
798 	}
799 
800 	/*
801 	 * We update our cred/cpid based on the caller of connect
802 	 */
803 	if (connp->conn_cred != cr) {
804 		crhold(cr);
805 		crfree(connp->conn_cred);
806 		connp->conn_cred = cr;
807 	}
808 	connp->conn_cpid = pid;
809 	ixa->ixa_cred = cr;
810 	ixa->ixa_cpid = pid;
811 	if (is_system_labeled()) {
812 		/* We need to restart with a label based on the cred */
813 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
814 	}
815 
816 	if (scopeid != 0) {
817 		ixa->ixa_flags |= IXAF_SCOPEID_SET;
818 		ixa->ixa_scopeid = scopeid;
819 		connp->conn_incoming_ifindex = scopeid;
820 	} else {
821 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
822 		connp->conn_incoming_ifindex = connp->conn_bound_if;
823 	}
824 
825 	/*
826 	 * conn_connect will drop conn_lock and reacquire it.
827 	 * To prevent a send* from messing with this icmp_t while the lock
828 	 * is dropped we set icmp_state and clear conn_v6lastdst.
829 	 * That will make all send* fail with EISCONN.
830 	 */
831 	connp->conn_v6lastdst = ipv6_all_zeros;
832 	icmp->icmp_state = TS_WCON_CREQ;
833 
834 	error = conn_connect(connp, NULL, IPDF_ALLOW_MCBC);
835 	mutex_exit(&connp->conn_lock);
836 	if (error != 0)
837 		goto connect_failed;
838 
839 	/*
840 	 * The addresses have been verified. Time to insert in
841 	 * the correct fanout list.
842 	 */
843 	error = ipcl_conn_insert(connp);
844 	if (error != 0)
845 		goto connect_failed;
846 
847 	mutex_enter(&connp->conn_lock);
848 	error = icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
849 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
850 	if (error != 0) {
851 		mutex_exit(&connp->conn_lock);
852 		goto connect_failed;
853 	}
854 
855 	icmp->icmp_state = TS_DATA_XFER;
856 	/* Record this as the "last" send even though we haven't sent any */
857 	connp->conn_v6lastdst = connp->conn_faddr_v6;
858 	connp->conn_lastipversion = connp->conn_ipversion;
859 	connp->conn_lastdstport = connp->conn_fport;
860 	connp->conn_lastflowinfo = connp->conn_flowinfo;
861 	connp->conn_lastscopeid = scopeid;
862 	connp->conn_lastsrcid = srcid;
863 	/* Also remember a source to use together with lastdst */
864 	connp->conn_v6lastsrc = v6src;
865 	mutex_exit(&connp->conn_lock);
866 
867 	ixa_refrele(ixa);
868 	return (0);
869 
870 connect_failed:
871 	if (ixa != NULL)
872 		ixa_refrele(ixa);
873 	mutex_enter(&connp->conn_lock);
874 	icmp->icmp_state = TS_IDLE;
875 	/* In case the source address was set above */
876 	if (connp->conn_mcbc_bind)
877 		connp->conn_saddr_v6 = ipv6_all_zeros;
878 	else
879 		connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
880 	connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
881 	connp->conn_faddr_v6 = ipv6_all_zeros;
882 	connp->conn_v6lastdst = ipv6_all_zeros;
883 	connp->conn_flowinfo = 0;
884 
885 	(void) icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
886 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
887 	mutex_exit(&connp->conn_lock);
888 	return (error);
889 }
890 
891 static void
892 rawip_do_close(conn_t *connp)
893 {
894 	ASSERT(connp != NULL && IPCL_IS_RAWIP(connp));
895 
896 	ip_quiesce_conn(connp);
897 
898 	if (!IPCL_IS_NONSTR(connp)) {
899 		qprocsoff(connp->conn_rq);
900 	}
901 
902 	icmp_close_free(connp);
903 
904 	/*
905 	 * Now we are truly single threaded on this stream, and can
906 	 * delete the things hanging off the connp, and finally the connp.
907 	 * We removed this connp from the fanout list, it cannot be
908 	 * accessed thru the fanouts, and we already waited for the
909 	 * conn_ref to drop to 0. We are already in close, so
910 	 * there cannot be any other thread from the top. qprocsoff
911 	 * has completed, and service has completed or won't run in
912 	 * future.
913 	 */
914 	ASSERT(connp->conn_ref == 1);
915 
916 	if (!IPCL_IS_NONSTR(connp)) {
917 		inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
918 	} else {
919 		ip_free_helper_stream(connp);
920 	}
921 
922 	connp->conn_ref--;
923 	ipcl_conn_destroy(connp);
924 }
925 
926 static int
927 icmp_close(queue_t *q, int flags)
928 {
929 	conn_t  *connp;
930 
931 	if (flags & SO_FALLBACK) {
932 		/*
933 		 * stream is being closed while in fallback
934 		 * simply free the resources that were allocated
935 		 */
936 		inet_minor_free(WR(q)->q_ptr, (dev_t)(RD(q)->q_ptr));
937 		qprocsoff(q);
938 		goto done;
939 	}
940 
941 	connp = Q_TO_CONN(q);
942 	(void) rawip_do_close(connp);
943 done:
944 	q->q_ptr = WR(q)->q_ptr = NULL;
945 	return (0);
946 }
947 
948 static void
949 icmp_close_free(conn_t *connp)
950 {
951 	icmp_t *icmp = connp->conn_icmp;
952 
953 	if (icmp->icmp_filter != NULL) {
954 		kmem_free(icmp->icmp_filter, sizeof (icmp6_filter_t));
955 		icmp->icmp_filter = NULL;
956 	}
957 
958 	/*
959 	 * Clear any fields which the kmem_cache constructor clears.
960 	 * Only icmp_connp needs to be preserved.
961 	 * TBD: We should make this more efficient to avoid clearing
962 	 * everything.
963 	 */
964 	ASSERT(icmp->icmp_connp == connp);
965 	bzero(icmp, sizeof (icmp_t));
966 	icmp->icmp_connp = connp;
967 }
968 
969 /*
970  * This routine handles each T_DISCON_REQ message passed to icmp
971  * as an indicating that ICMP is no longer connected. This results
972  * in telling IP to restore the binding to just the local address.
973  */
974 static int
975 icmp_do_disconnect(conn_t *connp)
976 {
977 	icmp_t	*icmp = connp->conn_icmp;
978 	int	error;
979 
980 	mutex_enter(&connp->conn_lock);
981 	if (icmp->icmp_state != TS_DATA_XFER) {
982 		mutex_exit(&connp->conn_lock);
983 		return (-TOUTSTATE);
984 	}
985 	if (connp->conn_mcbc_bind)
986 		connp->conn_saddr_v6 = ipv6_all_zeros;
987 	else
988 		connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
989 	connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
990 	connp->conn_faddr_v6 = ipv6_all_zeros;
991 	icmp->icmp_state = TS_IDLE;
992 
993 	connp->conn_v6lastdst = ipv6_all_zeros;
994 	error = icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
995 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
996 	mutex_exit(&connp->conn_lock);
997 	if (error != 0)
998 		return (error);
999 
1000 	/*
1001 	 * Tell IP to remove the full binding and revert
1002 	 * to the local address binding.
1003 	 */
1004 	return (ip_laddr_fanout_insert(connp));
1005 }
1006 
1007 static void
1008 icmp_tpi_disconnect(queue_t *q, mblk_t *mp)
1009 {
1010 	conn_t	*connp = Q_TO_CONN(q);
1011 	int	error;
1012 
1013 	/*
1014 	 * Allocate the largest primitive we need to send back
1015 	 * T_error_ack is > than T_ok_ack
1016 	 */
1017 	mp = reallocb(mp, sizeof (struct T_error_ack), 1);
1018 	if (mp == NULL) {
1019 		/* Unable to reuse the T_DISCON_REQ for the ack. */
1020 		icmp_err_ack_prim(q, mp, T_DISCON_REQ, TSYSERR, ENOMEM);
1021 		return;
1022 	}
1023 
1024 	error = icmp_do_disconnect(connp);
1025 
1026 	if (error != 0) {
1027 		if (error > 0) {
1028 			icmp_err_ack(q, mp, 0, error);
1029 		} else {
1030 			icmp_err_ack(q, mp, -error, 0);
1031 		}
1032 	} else {
1033 		mp = mi_tpi_ok_ack_alloc(mp);
1034 		ASSERT(mp != NULL);
1035 		qreply(q, mp);
1036 	}
1037 }
1038 
1039 static int
1040 icmp_disconnect(conn_t *connp)
1041 {
1042 	int	error;
1043 
1044 	connp->conn_dgram_errind = B_FALSE;
1045 
1046 	error = icmp_do_disconnect(connp);
1047 
1048 	if (error < 0)
1049 		error = proto_tlitosyserr(-error);
1050 	return (error);
1051 }
1052 
1053 /* This routine creates a T_ERROR_ACK message and passes it upstream. */
1054 static void
1055 icmp_err_ack(queue_t *q, mblk_t *mp, t_scalar_t t_error, int sys_error)
1056 {
1057 	if ((mp = mi_tpi_err_ack_alloc(mp, t_error, sys_error)) != NULL)
1058 		qreply(q, mp);
1059 }
1060 
1061 /* Shorthand to generate and send TPI error acks to our client */
1062 static void
1063 icmp_err_ack_prim(queue_t *q, mblk_t *mp, t_scalar_t primitive,
1064     t_scalar_t t_error, int sys_error)
1065 {
1066 	struct T_error_ack	*teackp;
1067 
1068 	if ((mp = tpi_ack_alloc(mp, sizeof (struct T_error_ack),
1069 	    M_PCPROTO, T_ERROR_ACK)) != NULL) {
1070 		teackp = (struct T_error_ack *)mp->b_rptr;
1071 		teackp->ERROR_prim = primitive;
1072 		teackp->TLI_error = t_error;
1073 		teackp->UNIX_error = sys_error;
1074 		qreply(q, mp);
1075 	}
1076 }
1077 
1078 /*
1079  * icmp_icmp_input is called as conn_recvicmp to process ICMP messages.
1080  * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
1081  * Assumes that IP has pulled up everything up to and including the ICMP header.
1082  */
1083 /* ARGSUSED2 */
1084 static void
1085 icmp_icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
1086 {
1087 	conn_t		*connp = (conn_t *)arg1;
1088 	icmp_t		*icmp = connp->conn_icmp;
1089 	icmph_t		*icmph;
1090 	ipha_t		*ipha;
1091 	int		iph_hdr_length;
1092 	sin_t		sin;
1093 	mblk_t		*mp1;
1094 	int		error = 0;
1095 
1096 	ipha = (ipha_t *)mp->b_rptr;
1097 
1098 	ASSERT(OK_32PTR(mp->b_rptr));
1099 
1100 	if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
1101 		ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
1102 		icmp_icmp_error_ipv6(connp, mp, ira);
1103 		return;
1104 	}
1105 	ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
1106 
1107 	/* Skip past the outer IP and ICMP headers */
1108 	ASSERT(IPH_HDR_LENGTH(ipha) == ira->ira_ip_hdr_length);
1109 	iph_hdr_length = ira->ira_ip_hdr_length;
1110 	icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1111 	ipha = (ipha_t *)&icmph[1];	/* Inner IP header */
1112 
1113 	iph_hdr_length = IPH_HDR_LENGTH(ipha);
1114 
1115 	switch (icmph->icmph_type) {
1116 	case ICMP_DEST_UNREACHABLE:
1117 		switch (icmph->icmph_code) {
1118 		case ICMP_FRAGMENTATION_NEEDED: {
1119 			ipha_t		*ipha;
1120 			ip_xmit_attr_t	*ixa;
1121 			/*
1122 			 * IP has already adjusted the path MTU.
1123 			 * But we need to adjust DF for IPv4.
1124 			 */
1125 			if (connp->conn_ipversion != IPV4_VERSION)
1126 				break;
1127 
1128 			ixa = conn_get_ixa(connp, B_FALSE);
1129 			if (ixa == NULL || ixa->ixa_ire == NULL) {
1130 				/*
1131 				 * Some other thread holds conn_ixa. We will
1132 				 * redo this on the next ICMP too big.
1133 				 */
1134 				if (ixa != NULL)
1135 					ixa_refrele(ixa);
1136 				break;
1137 			}
1138 			(void) ip_get_pmtu(ixa);
1139 
1140 			mutex_enter(&connp->conn_lock);
1141 			ipha = (ipha_t *)connp->conn_ht_iphc;
1142 			if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
1143 				ipha->ipha_fragment_offset_and_flags |=
1144 				    IPH_DF_HTONS;
1145 			} else {
1146 				ipha->ipha_fragment_offset_and_flags &=
1147 				    ~IPH_DF_HTONS;
1148 			}
1149 			mutex_exit(&connp->conn_lock);
1150 			ixa_refrele(ixa);
1151 			break;
1152 		}
1153 		case ICMP_PORT_UNREACHABLE:
1154 		case ICMP_PROTOCOL_UNREACHABLE:
1155 			error = ECONNREFUSED;
1156 			break;
1157 		default:
1158 			/* Transient errors */
1159 			break;
1160 		}
1161 		break;
1162 	default:
1163 		/* Transient errors */
1164 		break;
1165 	}
1166 	if (error == 0) {
1167 		freemsg(mp);
1168 		return;
1169 	}
1170 
1171 	/*
1172 	 * Deliver T_UDERROR_IND when the application has asked for it.
1173 	 * The socket layer enables this automatically when connected.
1174 	 */
1175 	if (!connp->conn_dgram_errind) {
1176 		freemsg(mp);
1177 		return;
1178 	}
1179 
1180 	sin = sin_null;
1181 	sin.sin_family = AF_INET;
1182 	sin.sin_addr.s_addr = ipha->ipha_dst;
1183 
1184 	if (IPCL_IS_NONSTR(connp)) {
1185 		mutex_enter(&connp->conn_lock);
1186 		if (icmp->icmp_state == TS_DATA_XFER) {
1187 			if (sin.sin_addr.s_addr == connp->conn_faddr_v4) {
1188 				mutex_exit(&connp->conn_lock);
1189 				(*connp->conn_upcalls->su_set_error)
1190 				    (connp->conn_upper_handle, error);
1191 				goto done;
1192 			}
1193 		} else {
1194 			icmp->icmp_delayed_error = error;
1195 			*((sin_t *)&icmp->icmp_delayed_addr) = sin;
1196 		}
1197 		mutex_exit(&connp->conn_lock);
1198 	} else {
1199 		mp1 = mi_tpi_uderror_ind((char *)&sin, sizeof (sin_t), NULL, 0,
1200 		    error);
1201 		if (mp1 != NULL)
1202 			putnext(connp->conn_rq, mp1);
1203 	}
1204 done:
1205 	freemsg(mp);
1206 }
1207 
1208 /*
1209  * icmp_icmp_error_ipv6 is called by icmp_icmp_error to process ICMP for IPv6.
1210  * Generates the appropriate T_UDERROR_IND for permanent (non-transient) errors.
1211  * Assumes that IP has pulled up all the extension headers as well as the
1212  * ICMPv6 header.
1213  */
1214 static void
1215 icmp_icmp_error_ipv6(conn_t *connp, mblk_t *mp, ip_recv_attr_t *ira)
1216 {
1217 	icmp6_t		*icmp6;
1218 	ip6_t		*ip6h, *outer_ip6h;
1219 	uint16_t	iph_hdr_length;
1220 	uint8_t		*nexthdrp;
1221 	sin6_t		sin6;
1222 	mblk_t		*mp1;
1223 	int		error = 0;
1224 	icmp_t		*icmp = connp->conn_icmp;
1225 
1226 	outer_ip6h = (ip6_t *)mp->b_rptr;
1227 #ifdef DEBUG
1228 	if (outer_ip6h->ip6_nxt != IPPROTO_ICMPV6)
1229 		iph_hdr_length = ip_hdr_length_v6(mp, outer_ip6h);
1230 	else
1231 		iph_hdr_length = IPV6_HDR_LEN;
1232 	ASSERT(iph_hdr_length == ira->ira_ip_hdr_length);
1233 #endif
1234 	/* Skip past the outer IP and ICMP headers */
1235 	iph_hdr_length = ira->ira_ip_hdr_length;
1236 	icmp6 = (icmp6_t *)&mp->b_rptr[iph_hdr_length];
1237 
1238 	ip6h = (ip6_t *)&icmp6[1];	/* Inner IP header */
1239 	if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length, &nexthdrp)) {
1240 		freemsg(mp);
1241 		return;
1242 	}
1243 
1244 	switch (icmp6->icmp6_type) {
1245 	case ICMP6_DST_UNREACH:
1246 		switch (icmp6->icmp6_code) {
1247 		case ICMP6_DST_UNREACH_NOPORT:
1248 			error = ECONNREFUSED;
1249 			break;
1250 		case ICMP6_DST_UNREACH_ADMIN:
1251 		case ICMP6_DST_UNREACH_NOROUTE:
1252 		case ICMP6_DST_UNREACH_BEYONDSCOPE:
1253 		case ICMP6_DST_UNREACH_ADDR:
1254 			/* Transient errors */
1255 			break;
1256 		default:
1257 			break;
1258 		}
1259 		break;
1260 	case ICMP6_PACKET_TOO_BIG: {
1261 		struct T_unitdata_ind	*tudi;
1262 		struct T_opthdr		*toh;
1263 		size_t			udi_size;
1264 		mblk_t			*newmp;
1265 		t_scalar_t		opt_length = sizeof (struct T_opthdr) +
1266 		    sizeof (struct ip6_mtuinfo);
1267 		sin6_t			*sin6;
1268 		struct ip6_mtuinfo	*mtuinfo;
1269 
1270 		/*
1271 		 * If the application has requested to receive path mtu
1272 		 * information, send up an empty message containing an
1273 		 * IPV6_PATHMTU ancillary data item.
1274 		 */
1275 		if (!connp->conn_ipv6_recvpathmtu)
1276 			break;
1277 
1278 		udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t) +
1279 		    opt_length;
1280 		if ((newmp = allocb(udi_size, BPRI_MED)) == NULL) {
1281 			BUMP_MIB(&icmp->icmp_is->is_rawip_mib, rawipInErrors);
1282 			break;
1283 		}
1284 
1285 		/*
1286 		 * newmp->b_cont is left to NULL on purpose.  This is an
1287 		 * empty message containing only ancillary data.
1288 		 */
1289 		newmp->b_datap->db_type = M_PROTO;
1290 		tudi = (struct T_unitdata_ind *)newmp->b_rptr;
1291 		newmp->b_wptr = (uchar_t *)tudi + udi_size;
1292 		tudi->PRIM_type = T_UNITDATA_IND;
1293 		tudi->SRC_length = sizeof (sin6_t);
1294 		tudi->SRC_offset = sizeof (struct T_unitdata_ind);
1295 		tudi->OPT_offset = tudi->SRC_offset + sizeof (sin6_t);
1296 		tudi->OPT_length = opt_length;
1297 
1298 		sin6 = (sin6_t *)&tudi[1];
1299 		bzero(sin6, sizeof (sin6_t));
1300 		sin6->sin6_family = AF_INET6;
1301 		sin6->sin6_addr = connp->conn_faddr_v6;
1302 
1303 		toh = (struct T_opthdr *)&sin6[1];
1304 		toh->level = IPPROTO_IPV6;
1305 		toh->name = IPV6_PATHMTU;
1306 		toh->len = opt_length;
1307 		toh->status = 0;
1308 
1309 		mtuinfo = (struct ip6_mtuinfo *)&toh[1];
1310 		bzero(mtuinfo, sizeof (struct ip6_mtuinfo));
1311 		mtuinfo->ip6m_addr.sin6_family = AF_INET6;
1312 		mtuinfo->ip6m_addr.sin6_addr = ip6h->ip6_dst;
1313 		mtuinfo->ip6m_mtu = icmp6->icmp6_mtu;
1314 		/*
1315 		 * We've consumed everything we need from the original
1316 		 * message.  Free it, then send our empty message.
1317 		 */
1318 		freemsg(mp);
1319 		icmp_ulp_recv(connp, newmp, msgdsize(newmp));
1320 		return;
1321 	}
1322 	case ICMP6_TIME_EXCEEDED:
1323 		/* Transient errors */
1324 		break;
1325 	case ICMP6_PARAM_PROB:
1326 		/* If this corresponds to an ICMP_PROTOCOL_UNREACHABLE */
1327 		if (icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER &&
1328 		    (uchar_t *)ip6h + icmp6->icmp6_pptr ==
1329 		    (uchar_t *)nexthdrp) {
1330 			error = ECONNREFUSED;
1331 			break;
1332 		}
1333 		break;
1334 	}
1335 	if (error == 0) {
1336 		freemsg(mp);
1337 		return;
1338 	}
1339 
1340 	/*
1341 	 * Deliver T_UDERROR_IND when the application has asked for it.
1342 	 * The socket layer enables this automatically when connected.
1343 	 */
1344 	if (!connp->conn_dgram_errind) {
1345 		freemsg(mp);
1346 		return;
1347 	}
1348 
1349 	sin6 = sin6_null;
1350 	sin6.sin6_family = AF_INET6;
1351 	sin6.sin6_addr = ip6h->ip6_dst;
1352 	sin6.sin6_flowinfo = ip6h->ip6_vcf & ~IPV6_VERS_AND_FLOW_MASK;
1353 	if (IPCL_IS_NONSTR(connp)) {
1354 		mutex_enter(&connp->conn_lock);
1355 		if (icmp->icmp_state == TS_DATA_XFER) {
1356 			if (IN6_ARE_ADDR_EQUAL(&sin6.sin6_addr,
1357 			    &connp->conn_faddr_v6)) {
1358 				mutex_exit(&connp->conn_lock);
1359 				(*connp->conn_upcalls->su_set_error)
1360 				    (connp->conn_upper_handle, error);
1361 				goto done;
1362 			}
1363 		} else {
1364 			icmp->icmp_delayed_error = error;
1365 			*((sin6_t *)&icmp->icmp_delayed_addr) = sin6;
1366 		}
1367 		mutex_exit(&connp->conn_lock);
1368 	} else {
1369 		mp1 = mi_tpi_uderror_ind((char *)&sin6, sizeof (sin6_t),
1370 		    NULL, 0, error);
1371 		if (mp1 != NULL)
1372 			putnext(connp->conn_rq, mp1);
1373 	}
1374 done:
1375 	freemsg(mp);
1376 }
1377 
1378 /*
1379  * This routine responds to T_ADDR_REQ messages.  It is called by icmp_wput.
1380  * The local address is filled in if endpoint is bound. The remote address
1381  * is filled in if remote address has been precified ("connected endpoint")
1382  * (The concept of connected CLTS sockets is alien to published TPI
1383  *  but we support it anyway).
1384  */
1385 static void
1386 icmp_addr_req(queue_t *q, mblk_t *mp)
1387 {
1388 	struct sockaddr *sa;
1389 	mblk_t	*ackmp;
1390 	struct T_addr_ack *taa;
1391 	icmp_t	*icmp = Q_TO_ICMP(q);
1392 	conn_t	*connp = icmp->icmp_connp;
1393 	uint_t	addrlen;
1394 
1395 	/* Make it large enough for worst case */
1396 	ackmp = reallocb(mp, sizeof (struct T_addr_ack) +
1397 	    2 * sizeof (sin6_t), 1);
1398 	if (ackmp == NULL) {
1399 		icmp_err_ack(q, mp, TSYSERR, ENOMEM);
1400 		return;
1401 	}
1402 	taa = (struct T_addr_ack *)ackmp->b_rptr;
1403 
1404 	bzero(taa, sizeof (struct T_addr_ack));
1405 	ackmp->b_wptr = (uchar_t *)&taa[1];
1406 
1407 	taa->PRIM_type = T_ADDR_ACK;
1408 	ackmp->b_datap->db_type = M_PCPROTO;
1409 
1410 	if (connp->conn_family == AF_INET)
1411 		addrlen = sizeof (sin_t);
1412 	else
1413 		addrlen = sizeof (sin6_t);
1414 
1415 	mutex_enter(&connp->conn_lock);
1416 	/*
1417 	 * Note: Following code assumes 32 bit alignment of basic
1418 	 * data structures like sin_t and struct T_addr_ack.
1419 	 */
1420 	if (icmp->icmp_state != TS_UNBND) {
1421 		/*
1422 		 * Fill in local address first
1423 		 */
1424 		taa->LOCADDR_offset = sizeof (*taa);
1425 		taa->LOCADDR_length = addrlen;
1426 		sa = (struct sockaddr *)&taa[1];
1427 		(void) conn_getsockname(connp, sa, &addrlen);
1428 		ackmp->b_wptr += addrlen;
1429 	}
1430 	if (icmp->icmp_state == TS_DATA_XFER) {
1431 		/*
1432 		 * connected, fill remote address too
1433 		 */
1434 		taa->REMADDR_length = addrlen;
1435 		/* assumed 32-bit alignment */
1436 		taa->REMADDR_offset = taa->LOCADDR_offset + taa->LOCADDR_length;
1437 		sa = (struct sockaddr *)(ackmp->b_rptr + taa->REMADDR_offset);
1438 		(void) conn_getpeername(connp, sa, &addrlen);
1439 		ackmp->b_wptr += addrlen;
1440 	}
1441 	mutex_exit(&connp->conn_lock);
1442 	ASSERT(ackmp->b_wptr <= ackmp->b_datap->db_lim);
1443 	qreply(q, ackmp);
1444 }
1445 
1446 static void
1447 icmp_copy_info(struct T_info_ack *tap, icmp_t *icmp)
1448 {
1449 	conn_t		*connp = icmp->icmp_connp;
1450 
1451 	*tap = icmp_g_t_info_ack;
1452 
1453 	if (connp->conn_family == AF_INET6)
1454 		tap->ADDR_size = sizeof (sin6_t);
1455 	else
1456 		tap->ADDR_size = sizeof (sin_t);
1457 	tap->CURRENT_state = icmp->icmp_state;
1458 	tap->OPT_size = icmp_max_optsize;
1459 }
1460 
1461 static void
1462 icmp_do_capability_ack(icmp_t *icmp, struct T_capability_ack *tcap,
1463     t_uscalar_t cap_bits1)
1464 {
1465 	tcap->CAP_bits1 = 0;
1466 
1467 	if (cap_bits1 & TC1_INFO) {
1468 		icmp_copy_info(&tcap->INFO_ack, icmp);
1469 		tcap->CAP_bits1 |= TC1_INFO;
1470 	}
1471 }
1472 
1473 /*
1474  * This routine responds to T_CAPABILITY_REQ messages.  It is called by
1475  * icmp_wput.  Much of the T_CAPABILITY_ACK information is copied from
1476  * icmp_g_t_info_ack.  The current state of the stream is copied from
1477  * icmp_state.
1478  */
1479 static void
1480 icmp_capability_req(queue_t *q, mblk_t *mp)
1481 {
1482 	icmp_t			*icmp = Q_TO_ICMP(q);
1483 	t_uscalar_t		cap_bits1;
1484 	struct T_capability_ack	*tcap;
1485 
1486 	cap_bits1 = ((struct T_capability_req *)mp->b_rptr)->CAP_bits1;
1487 
1488 	mp = tpi_ack_alloc(mp, sizeof (struct T_capability_ack),
1489 	    mp->b_datap->db_type, T_CAPABILITY_ACK);
1490 	if (!mp)
1491 		return;
1492 
1493 	tcap = (struct T_capability_ack *)mp->b_rptr;
1494 
1495 	icmp_do_capability_ack(icmp, tcap, cap_bits1);
1496 
1497 	qreply(q, mp);
1498 }
1499 
1500 /*
1501  * This routine responds to T_INFO_REQ messages.  It is called by icmp_wput.
1502  * Most of the T_INFO_ACK information is copied from icmp_g_t_info_ack.
1503  * The current state of the stream is copied from icmp_state.
1504  */
1505 static void
1506 icmp_info_req(queue_t *q, mblk_t *mp)
1507 {
1508 	icmp_t	*icmp = Q_TO_ICMP(q);
1509 
1510 	/* Create a T_INFO_ACK message. */
1511 	mp = tpi_ack_alloc(mp, sizeof (struct T_info_ack), M_PCPROTO,
1512 	    T_INFO_ACK);
1513 	if (!mp)
1514 		return;
1515 	icmp_copy_info((struct T_info_ack *)mp->b_rptr, icmp);
1516 	qreply(q, mp);
1517 }
1518 
1519 static int
1520 icmp_tpi_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
1521     int family)
1522 {
1523 	conn_t *connp;
1524 	dev_t	conn_dev;
1525 	int	error;
1526 
1527 	/* If the stream is already open, return immediately. */
1528 	if (q->q_ptr != NULL)
1529 		return (0);
1530 
1531 	if (sflag == MODOPEN)
1532 		return (EINVAL);
1533 
1534 	/*
1535 	 * Since ICMP is not used so heavily, allocating from the small
1536 	 * arena should be sufficient.
1537 	 */
1538 	if ((conn_dev = inet_minor_alloc(ip_minor_arena_sa)) == 0) {
1539 		return (EBUSY);
1540 	}
1541 
1542 	if (flag & SO_FALLBACK) {
1543 		/*
1544 		 * Non streams socket needs a stream to fallback to
1545 		 */
1546 		RD(q)->q_ptr = (void *)conn_dev;
1547 		WR(q)->q_qinfo = &icmp_fallback_sock_winit;
1548 		WR(q)->q_ptr = (void *)ip_minor_arena_sa;
1549 		qprocson(q);
1550 		return (0);
1551 	}
1552 
1553 	connp = rawip_do_open(family, credp, &error, KM_SLEEP);
1554 	if (connp == NULL) {
1555 		ASSERT(error != 0);
1556 		inet_minor_free(ip_minor_arena_sa, connp->conn_dev);
1557 		return (error);
1558 	}
1559 
1560 	*devp = makedevice(getemajor(*devp), (minor_t)conn_dev);
1561 	connp->conn_dev = conn_dev;
1562 	connp->conn_minor_arena = ip_minor_arena_sa;
1563 
1564 	/*
1565 	 * Initialize the icmp_t structure for this stream.
1566 	 */
1567 	q->q_ptr = connp;
1568 	WR(q)->q_ptr = connp;
1569 	connp->conn_rq = q;
1570 	connp->conn_wq = WR(q);
1571 
1572 	WR(q)->q_hiwat = connp->conn_sndbuf;
1573 	WR(q)->q_lowat = connp->conn_sndlowat;
1574 
1575 	qprocson(q);
1576 
1577 	/* Set the Stream head write offset. */
1578 	(void) proto_set_tx_wroff(q, connp, connp->conn_wroff);
1579 	(void) proto_set_rx_hiwat(connp->conn_rq, connp, connp->conn_rcvbuf);
1580 
1581 	mutex_enter(&connp->conn_lock);
1582 	connp->conn_state_flags &= ~CONN_INCIPIENT;
1583 	mutex_exit(&connp->conn_lock);
1584 
1585 	icmp_bind_proto(connp->conn_icmp);
1586 
1587 	return (0);
1588 }
1589 
1590 /* For /dev/icmp aka AF_INET open */
1591 static int
1592 icmp_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1593 {
1594 	return (icmp_tpi_open(q, devp, flag, sflag, credp, AF_INET));
1595 }
1596 
1597 /* For /dev/icmp6 aka AF_INET6 open */
1598 static int
1599 icmp_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
1600 {
1601 	return (icmp_tpi_open(q, devp, flag, sflag, credp, AF_INET6));
1602 }
1603 
1604 /*
1605  * This is the open routine for icmp.  It allocates a icmp_t structure for
1606  * the stream and, on the first open of the module, creates an ND table.
1607  */
1608 static conn_t *
1609 rawip_do_open(int family, cred_t *credp, int *err, int flags)
1610 {
1611 	icmp_t	*icmp;
1612 	conn_t *connp;
1613 	zoneid_t zoneid;
1614 	netstack_t *ns;
1615 	icmp_stack_t *is;
1616 	int len;
1617 	boolean_t isv6 = B_FALSE;
1618 
1619 	*err = secpolicy_net_icmpaccess(credp);
1620 	if (*err != 0)
1621 		return (NULL);
1622 
1623 	if (family == AF_INET6)
1624 		isv6 = B_TRUE;
1625 
1626 	ns = netstack_find_by_cred(credp);
1627 	ASSERT(ns != NULL);
1628 	is = ns->netstack_icmp;
1629 	ASSERT(is != NULL);
1630 
1631 	/*
1632 	 * For exclusive stacks we set the zoneid to zero
1633 	 * to make ICMP operate as if in the global zone.
1634 	 */
1635 	if (ns->netstack_stackid != GLOBAL_NETSTACKID)
1636 		zoneid = GLOBAL_ZONEID;
1637 	else
1638 		zoneid = crgetzoneid(credp);
1639 
1640 	ASSERT(flags == KM_SLEEP || flags == KM_NOSLEEP);
1641 
1642 	connp = ipcl_conn_create(IPCL_RAWIPCONN, flags, ns);
1643 	icmp = connp->conn_icmp;
1644 
1645 	/*
1646 	 * ipcl_conn_create did a netstack_hold. Undo the hold that was
1647 	 * done by netstack_find_by_cred()
1648 	 */
1649 	netstack_rele(ns);
1650 
1651 	/*
1652 	 * Since this conn_t/icmp_t is not yet visible to anybody else we don't
1653 	 * need to lock anything.
1654 	 */
1655 	ASSERT(connp->conn_proto == IPPROTO_ICMP);
1656 	ASSERT(connp->conn_icmp == icmp);
1657 	ASSERT(icmp->icmp_connp == connp);
1658 
1659 	/* Set the initial state of the stream and the privilege status. */
1660 	icmp->icmp_state = TS_UNBND;
1661 	connp->conn_ixa->ixa_flags |= IXAF_VERIFY_SOURCE;
1662 	if (isv6) {
1663 		connp->conn_family = AF_INET6;
1664 		connp->conn_ipversion = IPV6_VERSION;
1665 		connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
1666 		connp->conn_proto = IPPROTO_ICMPV6;
1667 		/* May be changed by a SO_PROTOTYPE socket option. */
1668 		connp->conn_proto = IPPROTO_ICMPV6;
1669 		connp->conn_ixa->ixa_protocol = connp->conn_proto;
1670 		connp->conn_ixa->ixa_raw_cksum_offset = 2;
1671 		connp->conn_default_ttl = is->is_ipv6_hoplimit;
1672 		len = sizeof (ip6_t);
1673 	} else {
1674 		connp->conn_family = AF_INET;
1675 		connp->conn_ipversion = IPV4_VERSION;
1676 		connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
1677 		/* May be changed by a SO_PROTOTYPE socket option. */
1678 		connp->conn_proto = IPPROTO_ICMP;
1679 		connp->conn_ixa->ixa_protocol = connp->conn_proto;
1680 		connp->conn_default_ttl = is->is_ipv4_ttl;
1681 		len = sizeof (ipha_t);
1682 	}
1683 	connp->conn_xmit_ipp.ipp_unicast_hops = connp->conn_default_ttl;
1684 
1685 	connp->conn_ixa->ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1686 
1687 	/*
1688 	 * For the socket of protocol IPPROTO_RAW or when IP_HDRINCL is set,
1689 	 * the checksum is provided in the pre-built packet. We clear
1690 	 * IXAF_SET_ULP_CKSUM to tell IP that the application has sent a
1691 	 * complete IP header and not to compute the transport checksum.
1692 	 */
1693 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
1694 	/* conn_allzones can not be set this early, hence no IPCL_ZONEID */
1695 	connp->conn_ixa->ixa_zoneid = zoneid;
1696 
1697 	connp->conn_zoneid = zoneid;
1698 
1699 	/*
1700 	 * If the caller has the process-wide flag set, then default to MAC
1701 	 * exempt mode.  This allows read-down to unlabeled hosts.
1702 	 */
1703 	if (getpflags(NET_MAC_AWARE, credp) != 0)
1704 		connp->conn_mac_mode = CONN_MAC_AWARE;
1705 
1706 	connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
1707 
1708 	icmp->icmp_is = is;
1709 
1710 	connp->conn_rcvbuf = is->is_recv_hiwat;
1711 	connp->conn_sndbuf = is->is_xmit_hiwat;
1712 	connp->conn_sndlowat = is->is_xmit_lowat;
1713 	connp->conn_rcvlowat = icmp_mod_info.mi_lowat;
1714 
1715 	connp->conn_wroff = len + is->is_wroff_extra;
1716 	connp->conn_so_type = SOCK_RAW;
1717 
1718 	connp->conn_recv = icmp_input;
1719 	connp->conn_recvicmp = icmp_icmp_input;
1720 	crhold(credp);
1721 	connp->conn_cred = credp;
1722 	connp->conn_cpid = curproc->p_pid;
1723 	connp->conn_open_time = ddi_get_lbolt64();
1724 	/* Cache things in ixa without an extra refhold */
1725 	connp->conn_ixa->ixa_cred = connp->conn_cred;
1726 	connp->conn_ixa->ixa_cpid = connp->conn_cpid;
1727 	if (is_system_labeled())
1728 		connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
1729 
1730 	connp->conn_flow_cntrld = B_FALSE;
1731 
1732 	if (is->is_pmtu_discovery)
1733 		connp->conn_ixa->ixa_flags |= IXAF_PMTU_DISCOVERY;
1734 
1735 	return (connp);
1736 }
1737 
1738 /*
1739  * Which ICMP options OK to set through T_UNITDATA_REQ...
1740  */
1741 /* ARGSUSED */
1742 static boolean_t
1743 icmp_opt_allow_udr_set(t_scalar_t level, t_scalar_t name)
1744 {
1745 	return (B_TRUE);
1746 }
1747 
1748 /*
1749  * This routine gets default values of certain options whose default
1750  * values are maintained by protcol specific code
1751  */
1752 int
1753 icmp_opt_default(queue_t *q, t_scalar_t level, t_scalar_t name, uchar_t *ptr)
1754 {
1755 	icmp_t *icmp = Q_TO_ICMP(q);
1756 	icmp_stack_t *is = icmp->icmp_is;
1757 	int *i1 = (int *)ptr;
1758 
1759 	switch (level) {
1760 	case IPPROTO_IP:
1761 		switch (name) {
1762 		case IP_MULTICAST_TTL:
1763 			*ptr = (uchar_t)IP_DEFAULT_MULTICAST_TTL;
1764 			return (sizeof (uchar_t));
1765 		case IP_MULTICAST_LOOP:
1766 			*ptr = (uchar_t)IP_DEFAULT_MULTICAST_LOOP;
1767 			return (sizeof (uchar_t));
1768 		}
1769 		break;
1770 	case IPPROTO_IPV6:
1771 		switch (name) {
1772 		case IPV6_MULTICAST_HOPS:
1773 			*i1 = IP_DEFAULT_MULTICAST_TTL;
1774 			return (sizeof (int));
1775 		case IPV6_MULTICAST_LOOP:
1776 			*i1 = IP_DEFAULT_MULTICAST_LOOP;
1777 			return (sizeof (int));
1778 		case IPV6_UNICAST_HOPS:
1779 			*i1 = is->is_ipv6_hoplimit;
1780 			return (sizeof (int));
1781 		}
1782 		break;
1783 	case IPPROTO_ICMPV6:
1784 		switch (name) {
1785 		case ICMP6_FILTER:
1786 			/* Make it look like "pass all" */
1787 			ICMP6_FILTER_SETPASSALL((icmp6_filter_t *)ptr);
1788 			return (sizeof (icmp6_filter_t));
1789 		}
1790 		break;
1791 	}
1792 	return (-1);
1793 }
1794 
1795 /*
1796  * This routine retrieves the current status of socket options.
1797  * It returns the size of the option retrieved, or -1.
1798  */
1799 int
1800 icmp_opt_get(conn_t *connp, int level, int name, uchar_t *ptr)
1801 {
1802 	icmp_t		*icmp = connp->conn_icmp;
1803 	int		*i1 = (int *)ptr;
1804 	conn_opt_arg_t	coas;
1805 	int		retval;
1806 
1807 	coas.coa_connp = connp;
1808 	coas.coa_ixa = connp->conn_ixa;
1809 	coas.coa_ipp = &connp->conn_xmit_ipp;
1810 	coas.coa_ancillary = B_FALSE;
1811 	coas.coa_changed = 0;
1812 
1813 	/*
1814 	 * We assume that the optcom framework has checked for the set
1815 	 * of levels and names that are supported, hence we don't worry
1816 	 * about rejecting based on that.
1817 	 * First check for ICMP specific handling, then pass to common routine.
1818 	 */
1819 	switch (level) {
1820 	case IPPROTO_IP:
1821 		/*
1822 		 * Only allow IPv4 option processing on IPv4 sockets.
1823 		 */
1824 		if (connp->conn_family != AF_INET)
1825 			return (-1);
1826 
1827 		switch (name) {
1828 		case IP_OPTIONS:
1829 		case T_IP_OPTIONS:
1830 			/* Options are passed up with each packet */
1831 			return (0);
1832 		case IP_HDRINCL:
1833 			mutex_enter(&connp->conn_lock);
1834 			*i1 = (int)icmp->icmp_hdrincl;
1835 			mutex_exit(&connp->conn_lock);
1836 			return (sizeof (int));
1837 		}
1838 		break;
1839 
1840 	case IPPROTO_IPV6:
1841 		/*
1842 		 * Only allow IPv6 option processing on native IPv6 sockets.
1843 		 */
1844 		if (connp->conn_family != AF_INET6)
1845 			return (-1);
1846 
1847 		switch (name) {
1848 		case IPV6_CHECKSUM:
1849 			/*
1850 			 * Return offset or -1 if no checksum offset.
1851 			 * Does not apply to IPPROTO_ICMPV6
1852 			 */
1853 			if (connp->conn_proto == IPPROTO_ICMPV6)
1854 				return (-1);
1855 
1856 			mutex_enter(&connp->conn_lock);
1857 			if (connp->conn_ixa->ixa_flags & IXAF_SET_RAW_CKSUM)
1858 				*i1 = connp->conn_ixa->ixa_raw_cksum_offset;
1859 			else
1860 				*i1 = -1;
1861 			mutex_exit(&connp->conn_lock);
1862 			return (sizeof (int));
1863 		}
1864 		break;
1865 
1866 	case IPPROTO_ICMPV6:
1867 		/*
1868 		 * Only allow IPv6 option processing on native IPv6 sockets.
1869 		 */
1870 		if (connp->conn_family != AF_INET6)
1871 			return (-1);
1872 
1873 		if (connp->conn_proto != IPPROTO_ICMPV6)
1874 			return (-1);
1875 
1876 		switch (name) {
1877 		case ICMP6_FILTER:
1878 			mutex_enter(&connp->conn_lock);
1879 			if (icmp->icmp_filter == NULL) {
1880 				/* Make it look like "pass all" */
1881 				ICMP6_FILTER_SETPASSALL((icmp6_filter_t *)ptr);
1882 			} else {
1883 				(void) bcopy(icmp->icmp_filter, ptr,
1884 				    sizeof (icmp6_filter_t));
1885 			}
1886 			mutex_exit(&connp->conn_lock);
1887 			return (sizeof (icmp6_filter_t));
1888 		}
1889 	}
1890 	mutex_enter(&connp->conn_lock);
1891 	retval = conn_opt_get(&coas, level, name, ptr);
1892 	mutex_exit(&connp->conn_lock);
1893 	return (retval);
1894 }
1895 
1896 /*
1897  * This routine retrieves the current status of socket options.
1898  * It returns the size of the option retrieved, or -1.
1899  */
1900 int
1901 icmp_tpi_opt_get(queue_t *q, int level, int name, uchar_t *ptr)
1902 {
1903 	conn_t		*connp = Q_TO_CONN(q);
1904 	int 		err;
1905 
1906 	err = icmp_opt_get(connp, level, name, ptr);
1907 	return (err);
1908 }
1909 
1910 /*
1911  * This routine sets socket options.
1912  */
1913 int
1914 icmp_do_opt_set(conn_opt_arg_t *coa, int level, int name,
1915     uint_t inlen, uchar_t *invalp, cred_t *cr, boolean_t checkonly)
1916 {
1917 	conn_t		*connp = coa->coa_connp;
1918 	ip_xmit_attr_t	*ixa = coa->coa_ixa;
1919 	icmp_t		*icmp = connp->conn_icmp;
1920 	icmp_stack_t	*is = icmp->icmp_is;
1921 	int		*i1 = (int *)invalp;
1922 	boolean_t	onoff = (*i1 == 0) ? 0 : 1;
1923 	int		error;
1924 
1925 	ASSERT(MUTEX_NOT_HELD(&coa->coa_connp->conn_lock));
1926 
1927 	/*
1928 	 * For fixed length options, no sanity check
1929 	 * of passed in length is done. It is assumed *_optcom_req()
1930 	 * routines do the right thing.
1931 	 */
1932 
1933 	switch (level) {
1934 	case SOL_SOCKET:
1935 		switch (name) {
1936 		case SO_PROTOTYPE:
1937 			if ((*i1 & 0xFF) != IPPROTO_ICMP &&
1938 			    (*i1 & 0xFF) != IPPROTO_ICMPV6 &&
1939 			    secpolicy_net_rawaccess(cr) != 0) {
1940 				return (EACCES);
1941 			}
1942 			if (checkonly)
1943 				break;
1944 
1945 			mutex_enter(&connp->conn_lock);
1946 			connp->conn_proto = *i1 & 0xFF;
1947 			ixa->ixa_protocol = connp->conn_proto;
1948 			if ((connp->conn_proto == IPPROTO_RAW ||
1949 			    connp->conn_proto == IPPROTO_IGMP) &&
1950 			    connp->conn_family == AF_INET) {
1951 				icmp->icmp_hdrincl = 1;
1952 				ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
1953 			} else if (connp->conn_proto == IPPROTO_UDP ||
1954 			    connp->conn_proto == IPPROTO_TCP ||
1955 			    connp->conn_proto == IPPROTO_SCTP) {
1956 				/* Used by test applications like psh */
1957 				icmp->icmp_hdrincl = 0;
1958 				ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
1959 			} else {
1960 				icmp->icmp_hdrincl = 0;
1961 				ixa->ixa_flags |= IXAF_SET_ULP_CKSUM;
1962 			}
1963 
1964 			if (connp->conn_family == AF_INET6 &&
1965 			    connp->conn_proto == IPPROTO_ICMPV6) {
1966 				/* Set offset for icmp6_cksum */
1967 				ixa->ixa_flags &= ~IXAF_SET_RAW_CKSUM;
1968 				ixa->ixa_raw_cksum_offset = 2;
1969 			}
1970 			if (icmp->icmp_filter != NULL &&
1971 			    connp->conn_proto != IPPROTO_ICMPV6) {
1972 				kmem_free(icmp->icmp_filter,
1973 				    sizeof (icmp6_filter_t));
1974 				icmp->icmp_filter = NULL;
1975 			}
1976 			mutex_exit(&connp->conn_lock);
1977 
1978 			coa->coa_changed |= COA_HEADER_CHANGED;
1979 			/*
1980 			 * For SCTP, we don't use icmp_bind_proto() for
1981 			 * raw socket binding.
1982 			 */
1983 			if (connp->conn_proto == IPPROTO_SCTP)
1984 				return (0);
1985 
1986 			coa->coa_changed |= COA_ICMP_BIND_NEEDED;
1987 			return (0);
1988 
1989 		case SO_SNDBUF:
1990 			if (*i1 > is->is_max_buf) {
1991 				return (ENOBUFS);
1992 			}
1993 			break;
1994 		case SO_RCVBUF:
1995 			if (*i1 > is->is_max_buf) {
1996 				return (ENOBUFS);
1997 			}
1998 			break;
1999 		}
2000 		break;
2001 
2002 	case IPPROTO_IP:
2003 		/*
2004 		 * Only allow IPv4 option processing on IPv4 sockets.
2005 		 */
2006 		if (connp->conn_family != AF_INET)
2007 			return (EINVAL);
2008 
2009 		switch (name) {
2010 		case IP_HDRINCL:
2011 			if (!checkonly) {
2012 				mutex_enter(&connp->conn_lock);
2013 				icmp->icmp_hdrincl = onoff;
2014 				if (onoff)
2015 					ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
2016 				else
2017 					ixa->ixa_flags |= IXAF_SET_ULP_CKSUM;
2018 				mutex_exit(&connp->conn_lock);
2019 			}
2020 			break;
2021 		}
2022 		break;
2023 
2024 	case IPPROTO_IPV6:
2025 		if (connp->conn_family != AF_INET6)
2026 			return (EINVAL);
2027 
2028 		switch (name) {
2029 		case IPV6_CHECKSUM:
2030 			/*
2031 			 * Integer offset into the user data of where the
2032 			 * checksum is located.
2033 			 * Offset of -1 disables option.
2034 			 * Does not apply to IPPROTO_ICMPV6.
2035 			 */
2036 			if (connp->conn_proto == IPPROTO_ICMPV6 ||
2037 			    coa->coa_ancillary) {
2038 				return (EINVAL);
2039 			}
2040 			if ((*i1 != -1) && ((*i1 < 0) || (*i1 & 0x1) != 0)) {
2041 				/* Negative or not 16 bit aligned offset */
2042 				return (EINVAL);
2043 			}
2044 			if (checkonly)
2045 				break;
2046 
2047 			mutex_enter(&connp->conn_lock);
2048 			if (*i1 == -1) {
2049 				ixa->ixa_flags &= ~IXAF_SET_RAW_CKSUM;
2050 				ixa->ixa_raw_cksum_offset = 0;
2051 				ixa->ixa_flags &= ~IXAF_SET_ULP_CKSUM;
2052 			} else {
2053 				ixa->ixa_flags |= IXAF_SET_RAW_CKSUM;
2054 				ixa->ixa_raw_cksum_offset = *i1;
2055 				ixa->ixa_flags |= IXAF_SET_ULP_CKSUM;
2056 			}
2057 			mutex_exit(&connp->conn_lock);
2058 			break;
2059 		}
2060 		break;
2061 
2062 	case IPPROTO_ICMPV6:
2063 		/*
2064 		 * Only allow IPv6 option processing on IPv6 sockets.
2065 		 */
2066 		if (connp->conn_family != AF_INET6)
2067 			return (EINVAL);
2068 		if (connp->conn_proto != IPPROTO_ICMPV6)
2069 			return (EINVAL);
2070 
2071 		switch (name) {
2072 		case ICMP6_FILTER:
2073 			if (checkonly)
2074 				break;
2075 
2076 			if ((inlen != 0) &&
2077 			    (inlen != sizeof (icmp6_filter_t)))
2078 				return (EINVAL);
2079 
2080 			mutex_enter(&connp->conn_lock);
2081 			if (inlen == 0) {
2082 				if (icmp->icmp_filter != NULL) {
2083 					kmem_free(icmp->icmp_filter,
2084 					    sizeof (icmp6_filter_t));
2085 					icmp->icmp_filter = NULL;
2086 				}
2087 			} else {
2088 				if (icmp->icmp_filter == NULL) {
2089 					icmp->icmp_filter = kmem_alloc(
2090 					    sizeof (icmp6_filter_t),
2091 					    KM_NOSLEEP);
2092 					if (icmp->icmp_filter == NULL) {
2093 						mutex_exit(&connp->conn_lock);
2094 						return (ENOBUFS);
2095 					}
2096 				}
2097 				(void) bcopy(invalp, icmp->icmp_filter, inlen);
2098 			}
2099 			mutex_exit(&connp->conn_lock);
2100 			break;
2101 		}
2102 		break;
2103 	}
2104 	error = conn_opt_set(coa, level, name, inlen, invalp,
2105 	    checkonly, cr);
2106 	return (error);
2107 }
2108 
2109 /*
2110  * This routine sets socket options.
2111  */
2112 int
2113 icmp_opt_set(conn_t *connp, uint_t optset_context, int level, int name,
2114     uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
2115     void *thisdg_attrs, cred_t *cr)
2116 {
2117 	icmp_t		*icmp = connp->conn_icmp;
2118 	int		err;
2119 	conn_opt_arg_t	coas, *coa;
2120 	boolean_t	checkonly;
2121 	icmp_stack_t	*is = icmp->icmp_is;
2122 
2123 	switch (optset_context) {
2124 	case SETFN_OPTCOM_CHECKONLY:
2125 		checkonly = B_TRUE;
2126 		/*
2127 		 * Note: Implies T_CHECK semantics for T_OPTCOM_REQ
2128 		 * inlen != 0 implies value supplied and
2129 		 * 	we have to "pretend" to set it.
2130 		 * inlen == 0 implies that there is no
2131 		 * 	value part in T_CHECK request and just validation
2132 		 * done elsewhere should be enough, we just return here.
2133 		 */
2134 		if (inlen == 0) {
2135 			*outlenp = 0;
2136 			return (0);
2137 		}
2138 		break;
2139 	case SETFN_OPTCOM_NEGOTIATE:
2140 		checkonly = B_FALSE;
2141 		break;
2142 	case SETFN_UD_NEGOTIATE:
2143 	case SETFN_CONN_NEGOTIATE:
2144 		checkonly = B_FALSE;
2145 		/*
2146 		 * Negotiating local and "association-related" options
2147 		 * through T_UNITDATA_REQ.
2148 		 *
2149 		 * Following routine can filter out ones we do not
2150 		 * want to be "set" this way.
2151 		 */
2152 		if (!icmp_opt_allow_udr_set(level, name)) {
2153 			*outlenp = 0;
2154 			return (EINVAL);
2155 		}
2156 		break;
2157 	default:
2158 		/*
2159 		 * We should never get here
2160 		 */
2161 		*outlenp = 0;
2162 		return (EINVAL);
2163 	}
2164 
2165 	ASSERT((optset_context != SETFN_OPTCOM_CHECKONLY) ||
2166 	    (optset_context == SETFN_OPTCOM_CHECKONLY && inlen != 0));
2167 
2168 	if (thisdg_attrs != NULL) {
2169 		/* Options from T_UNITDATA_REQ */
2170 		coa = (conn_opt_arg_t *)thisdg_attrs;
2171 		ASSERT(coa->coa_connp == connp);
2172 		ASSERT(coa->coa_ixa != NULL);
2173 		ASSERT(coa->coa_ipp != NULL);
2174 		ASSERT(coa->coa_ancillary);
2175 	} else {
2176 		coa = &coas;
2177 		coas.coa_connp = connp;
2178 		/* Get a reference on conn_ixa to prevent concurrent mods */
2179 		coas.coa_ixa = conn_get_ixa(connp, B_TRUE);
2180 		if (coas.coa_ixa == NULL) {
2181 			*outlenp = 0;
2182 			return (ENOMEM);
2183 		}
2184 		coas.coa_ipp = &connp->conn_xmit_ipp;
2185 		coas.coa_ancillary = B_FALSE;
2186 		coas.coa_changed = 0;
2187 	}
2188 
2189 	err = icmp_do_opt_set(coa, level, name, inlen, invalp,
2190 	    cr, checkonly);
2191 	if (err != 0) {
2192 errout:
2193 		if (!coa->coa_ancillary)
2194 			ixa_refrele(coa->coa_ixa);
2195 		*outlenp = 0;
2196 		return (err);
2197 	}
2198 
2199 	/*
2200 	 * Common case of OK return with outval same as inval.
2201 	 */
2202 	if (invalp != outvalp) {
2203 		/* don't trust bcopy for identical src/dst */
2204 		(void) bcopy(invalp, outvalp, inlen);
2205 	}
2206 	*outlenp = inlen;
2207 
2208 	/*
2209 	 * If this was not ancillary data, then we rebuild the headers,
2210 	 * update the IRE/NCE, and IPsec as needed.
2211 	 * Since the label depends on the destination we go through
2212 	 * ip_set_destination first.
2213 	 */
2214 	if (coa->coa_ancillary) {
2215 		return (0);
2216 	}
2217 
2218 	if (coa->coa_changed & COA_ROUTE_CHANGED) {
2219 		in6_addr_t saddr, faddr, nexthop;
2220 		in_port_t fport;
2221 
2222 		/*
2223 		 * We clear lastdst to make sure we pick up the change
2224 		 * next time sending.
2225 		 * If we are connected we re-cache the information.
2226 		 * We ignore errors to preserve BSD behavior.
2227 		 * Note that we don't redo IPsec policy lookup here
2228 		 * since the final destination (or source) didn't change.
2229 		 */
2230 		mutex_enter(&connp->conn_lock);
2231 		connp->conn_v6lastdst = ipv6_all_zeros;
2232 
2233 		ip_attr_nexthop(coa->coa_ipp, coa->coa_ixa,
2234 		    &connp->conn_faddr_v6, &nexthop);
2235 		saddr = connp->conn_saddr_v6;
2236 		faddr = connp->conn_faddr_v6;
2237 		fport = connp->conn_fport;
2238 		mutex_exit(&connp->conn_lock);
2239 
2240 		if (!IN6_IS_ADDR_UNSPECIFIED(&faddr) &&
2241 		    !IN6_IS_ADDR_V4MAPPED_ANY(&faddr)) {
2242 			(void) ip_attr_connect(connp, coa->coa_ixa,
2243 			    &saddr, &faddr, &nexthop, fport, NULL, NULL,
2244 			    IPDF_ALLOW_MCBC | IPDF_VERIFY_DST);
2245 		}
2246 	}
2247 
2248 	ixa_refrele(coa->coa_ixa);
2249 
2250 	if (coa->coa_changed & COA_HEADER_CHANGED) {
2251 		/*
2252 		 * Rebuild the header template if we are connected.
2253 		 * Otherwise clear conn_v6lastdst so we rebuild the header
2254 		 * in the data path.
2255 		 */
2256 		mutex_enter(&connp->conn_lock);
2257 		if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) &&
2258 		    !IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_faddr_v6)) {
2259 			err = icmp_build_hdr_template(connp,
2260 			    &connp->conn_saddr_v6, &connp->conn_faddr_v6,
2261 			    connp->conn_flowinfo);
2262 			if (err != 0) {
2263 				mutex_exit(&connp->conn_lock);
2264 				return (err);
2265 			}
2266 		} else {
2267 			connp->conn_v6lastdst = ipv6_all_zeros;
2268 		}
2269 		mutex_exit(&connp->conn_lock);
2270 	}
2271 	if (coa->coa_changed & COA_RCVBUF_CHANGED) {
2272 		(void) proto_set_rx_hiwat(connp->conn_rq, connp,
2273 		    connp->conn_rcvbuf);
2274 	}
2275 	if ((coa->coa_changed & COA_SNDBUF_CHANGED) && !IPCL_IS_NONSTR(connp)) {
2276 		connp->conn_wq->q_hiwat = connp->conn_sndbuf;
2277 	}
2278 	if (coa->coa_changed & COA_WROFF_CHANGED) {
2279 		/* Increase wroff if needed */
2280 		uint_t wroff;
2281 
2282 		mutex_enter(&connp->conn_lock);
2283 		wroff = connp->conn_ht_iphc_allocated + is->is_wroff_extra;
2284 		if (wroff > connp->conn_wroff) {
2285 			connp->conn_wroff = wroff;
2286 			mutex_exit(&connp->conn_lock);
2287 			(void) proto_set_tx_wroff(connp->conn_rq, connp, wroff);
2288 		} else {
2289 			mutex_exit(&connp->conn_lock);
2290 		}
2291 	}
2292 	if (coa->coa_changed & COA_ICMP_BIND_NEEDED) {
2293 		icmp_bind_proto(icmp);
2294 	}
2295 	return (err);
2296 }
2297 
2298 /* This routine sets socket options. */
2299 int
2300 icmp_tpi_opt_set(queue_t *q, uint_t optset_context, int level, int name,
2301     uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
2302     void *thisdg_attrs, cred_t *cr)
2303 {
2304 	conn_t	*connp = Q_TO_CONN(q);
2305 	int error;
2306 
2307 	error = icmp_opt_set(connp, optset_context, level, name, inlen, invalp,
2308 	    outlenp, outvalp, thisdg_attrs, cr);
2309 	return (error);
2310 }
2311 
2312 /*
2313  * Setup IP headers.
2314  *
2315  * Note that IP_HDRINCL has ipha_protocol that is different than conn_proto,
2316  * but icmp_output_hdrincl restores ipha_protocol once we return.
2317  */
2318 mblk_t *
2319 icmp_prepend_hdr(conn_t *connp, ip_xmit_attr_t *ixa, const ip_pkt_t *ipp,
2320     const in6_addr_t *v6src, const in6_addr_t *v6dst, uint32_t flowinfo,
2321     mblk_t *data_mp, int *errorp)
2322 {
2323 	mblk_t		*mp;
2324 	icmp_stack_t	*is = connp->conn_netstack->netstack_icmp;
2325 	uint_t		data_len;
2326 	uint32_t	cksum;
2327 
2328 	data_len = msgdsize(data_mp);
2329 	mp = conn_prepend_hdr(ixa, ipp, v6src, v6dst, connp->conn_proto,
2330 	    flowinfo, 0, data_mp, data_len, is->is_wroff_extra, &cksum, errorp);
2331 	if (mp == NULL) {
2332 		ASSERT(*errorp != 0);
2333 		return (NULL);
2334 	}
2335 
2336 	ixa->ixa_pktlen = data_len + ixa->ixa_ip_hdr_length;
2337 
2338 	/*
2339 	 * If there was a routing option/header then conn_prepend_hdr
2340 	 * has massaged it and placed the pseudo-header checksum difference
2341 	 * in the cksum argument.
2342 	 *
2343 	 * Prepare for ICMPv6 checksum done in IP.
2344 	 *
2345 	 * We make it easy for IP to include our pseudo header
2346 	 * by putting our length (and any routing header adjustment)
2347 	 * in the ICMPv6 checksum field.
2348 	 * The IP source, destination, and length have already been set by
2349 	 * conn_prepend_hdr.
2350 	 */
2351 	cksum += data_len;
2352 	cksum = (cksum >> 16) + (cksum & 0xFFFF);
2353 	ASSERT(cksum < 0x10000);
2354 
2355 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
2356 		ipha_t	*ipha = (ipha_t *)mp->b_rptr;
2357 
2358 		ASSERT(ntohs(ipha->ipha_length) == ixa->ixa_pktlen);
2359 	} else {
2360 		ip6_t	*ip6h = (ip6_t *)mp->b_rptr;
2361 		uint_t	cksum_offset = 0;
2362 
2363 		ASSERT(ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN == ixa->ixa_pktlen);
2364 
2365 		if (ixa->ixa_flags & IXAF_SET_ULP_CKSUM) {
2366 			if (connp->conn_proto == IPPROTO_ICMPV6) {
2367 				cksum_offset = ixa->ixa_ip_hdr_length +
2368 				    offsetof(icmp6_t, icmp6_cksum);
2369 			} else if (ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
2370 				cksum_offset = ixa->ixa_ip_hdr_length +
2371 				    ixa->ixa_raw_cksum_offset;
2372 			}
2373 		}
2374 		if (cksum_offset != 0) {
2375 			uint16_t *ptr;
2376 
2377 			/* Make sure the checksum fits in the first mblk */
2378 			if (cksum_offset + sizeof (short) > MBLKL(mp)) {
2379 				mblk_t *mp1;
2380 
2381 				mp1 = msgpullup(mp,
2382 				    cksum_offset + sizeof (short));
2383 				freemsg(mp);
2384 				if (mp1 == NULL) {
2385 					*errorp = ENOMEM;
2386 					return (NULL);
2387 				}
2388 				mp = mp1;
2389 				ip6h = (ip6_t *)mp->b_rptr;
2390 			}
2391 			ptr = (uint16_t *)(mp->b_rptr + cksum_offset);
2392 			*ptr = htons(cksum);
2393 		}
2394 	}
2395 
2396 	/* Note that we don't try to update wroff due to ancillary data */
2397 	return (mp);
2398 }
2399 
2400 static int
2401 icmp_build_hdr_template(conn_t *connp, const in6_addr_t *v6src,
2402     const in6_addr_t *v6dst, uint32_t flowinfo)
2403 {
2404 	int		error;
2405 
2406 	ASSERT(MUTEX_HELD(&connp->conn_lock));
2407 	/*
2408 	 * We clear lastdst to make sure we don't use the lastdst path
2409 	 * next time sending since we might not have set v6dst yet.
2410 	 */
2411 	connp->conn_v6lastdst = ipv6_all_zeros;
2412 
2413 	error = conn_build_hdr_template(connp, 0, 0, v6src, v6dst, flowinfo);
2414 	if (error != 0)
2415 		return (error);
2416 
2417 	/*
2418 	 * Any routing header/option has been massaged. The checksum difference
2419 	 * is stored in conn_sum.
2420 	 */
2421 	return (0);
2422 }
2423 
2424 /*
2425  * This routine retrieves the value of an ND variable in a icmpparam_t
2426  * structure.  It is called through nd_getset when a user reads the
2427  * variable.
2428  */
2429 /* ARGSUSED */
2430 static int
2431 icmp_param_get(queue_t *q, mblk_t *mp, caddr_t cp, cred_t *cr)
2432 {
2433 	icmpparam_t	*icmppa = (icmpparam_t *)cp;
2434 
2435 	(void) mi_mpprintf(mp, "%d", icmppa->icmp_param_value);
2436 	return (0);
2437 }
2438 
2439 /*
2440  * Walk through the param array specified registering each element with the
2441  * named dispatch (ND) handler.
2442  */
2443 static boolean_t
2444 icmp_param_register(IDP *ndp, icmpparam_t *icmppa, int cnt)
2445 {
2446 	for (; cnt-- > 0; icmppa++) {
2447 		if (icmppa->icmp_param_name && icmppa->icmp_param_name[0]) {
2448 			if (!nd_load(ndp, icmppa->icmp_param_name,
2449 			    icmp_param_get, icmp_param_set,
2450 			    (caddr_t)icmppa)) {
2451 				nd_free(ndp);
2452 				return (B_FALSE);
2453 			}
2454 		}
2455 	}
2456 	return (B_TRUE);
2457 }
2458 
2459 /* This routine sets an ND variable in a icmpparam_t structure. */
2460 /* ARGSUSED */
2461 static int
2462 icmp_param_set(queue_t *q, mblk_t *mp, char *value, caddr_t cp, cred_t *cr)
2463 {
2464 	long		new_value;
2465 	icmpparam_t	*icmppa = (icmpparam_t *)cp;
2466 
2467 	/*
2468 	 * Fail the request if the new value does not lie within the
2469 	 * required bounds.
2470 	 */
2471 	if (ddi_strtol(value, NULL, 10, &new_value) != 0 ||
2472 	    new_value < icmppa->icmp_param_min ||
2473 	    new_value > icmppa->icmp_param_max) {
2474 		return (EINVAL);
2475 	}
2476 	/* Set the new value */
2477 	icmppa->icmp_param_value = new_value;
2478 	return (0);
2479 }
2480 
2481 static mblk_t *
2482 icmp_queue_fallback(icmp_t *icmp, mblk_t *mp)
2483 {
2484 	ASSERT(MUTEX_HELD(&icmp->icmp_recv_lock));
2485 	if (IPCL_IS_NONSTR(icmp->icmp_connp)) {
2486 		/*
2487 		 * fallback has started but messages have not been moved yet
2488 		 */
2489 		if (icmp->icmp_fallback_queue_head == NULL) {
2490 			ASSERT(icmp->icmp_fallback_queue_tail == NULL);
2491 			icmp->icmp_fallback_queue_head = mp;
2492 			icmp->icmp_fallback_queue_tail = mp;
2493 		} else {
2494 			ASSERT(icmp->icmp_fallback_queue_tail != NULL);
2495 			icmp->icmp_fallback_queue_tail->b_next = mp;
2496 			icmp->icmp_fallback_queue_tail = mp;
2497 		}
2498 		return (NULL);
2499 	} else {
2500 		/*
2501 		 * Fallback completed, let the caller putnext() the mblk.
2502 		 */
2503 		return (mp);
2504 	}
2505 }
2506 
2507 /*
2508  * Deliver data to ULP. In case we have a socket, and it's falling back to
2509  * TPI, then we'll queue the mp for later processing.
2510  */
2511 static void
2512 icmp_ulp_recv(conn_t *connp, mblk_t *mp, uint_t len)
2513 {
2514 	if (IPCL_IS_NONSTR(connp)) {
2515 		icmp_t *icmp = connp->conn_icmp;
2516 		int error;
2517 
2518 		ASSERT(len == msgdsize(mp));
2519 		if ((*connp->conn_upcalls->su_recv)
2520 		    (connp->conn_upper_handle, mp, len, 0, &error, NULL) < 0) {
2521 			mutex_enter(&icmp->icmp_recv_lock);
2522 			if (error == ENOSPC) {
2523 				/*
2524 				 * let's confirm while holding the lock
2525 				 */
2526 				if ((*connp->conn_upcalls->su_recv)
2527 				    (connp->conn_upper_handle, NULL, 0, 0,
2528 				    &error, NULL) < 0) {
2529 					ASSERT(error == ENOSPC);
2530 					if (error == ENOSPC) {
2531 						connp->conn_flow_cntrld =
2532 						    B_TRUE;
2533 					}
2534 				}
2535 				mutex_exit(&icmp->icmp_recv_lock);
2536 			} else {
2537 				ASSERT(error == EOPNOTSUPP);
2538 				mp = icmp_queue_fallback(icmp, mp);
2539 				mutex_exit(&icmp->icmp_recv_lock);
2540 				if (mp != NULL)
2541 					putnext(connp->conn_rq, mp);
2542 			}
2543 		}
2544 		ASSERT(MUTEX_NOT_HELD(&icmp->icmp_recv_lock));
2545 	} else {
2546 		putnext(connp->conn_rq, mp);
2547 	}
2548 }
2549 
2550 /*
2551  * This is the inbound data path.
2552  * IP has already pulled up the IP headers and verified alignment
2553  * etc.
2554  */
2555 /* ARGSUSED2 */
2556 static void
2557 icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
2558 {
2559 	conn_t			*connp = (conn_t *)arg1;
2560 	struct T_unitdata_ind	*tudi;
2561 	uchar_t			*rptr;		/* Pointer to IP header */
2562 	int			ip_hdr_length;
2563 	int			udi_size;	/* Size of T_unitdata_ind */
2564 	int			pkt_len;
2565 	icmp_t			*icmp;
2566 	ip_pkt_t		ipps;
2567 	ip6_t			*ip6h;
2568 	mblk_t			*mp1;
2569 	crb_t			recv_ancillary;
2570 	icmp_stack_t		*is;
2571 	sin_t			*sin;
2572 	sin6_t			*sin6;
2573 	ipha_t			*ipha;
2574 
2575 	ASSERT(connp->conn_flags & IPCL_RAWIPCONN);
2576 
2577 	icmp = connp->conn_icmp;
2578 	is = icmp->icmp_is;
2579 	rptr = mp->b_rptr;
2580 
2581 	ASSERT(DB_TYPE(mp) == M_DATA);
2582 	ASSERT(OK_32PTR(rptr));
2583 	ASSERT(ira->ira_pktlen == msgdsize(mp));
2584 	pkt_len = ira->ira_pktlen;
2585 
2586 	/*
2587 	 * Get a snapshot of these and allow other threads to change
2588 	 * them after that. We need the same recv_ancillary when determining
2589 	 * the size as when adding the ancillary data items.
2590 	 */
2591 	mutex_enter(&connp->conn_lock);
2592 	recv_ancillary = connp->conn_recv_ancillary;
2593 	mutex_exit(&connp->conn_lock);
2594 
2595 	ip_hdr_length = ira->ira_ip_hdr_length;
2596 	ASSERT(MBLKL(mp) >= ip_hdr_length);	/* IP did a pullup */
2597 
2598 	/* Initialize regardless of IP version */
2599 	ipps.ipp_fields = 0;
2600 
2601 	if (ira->ira_flags & IRAF_IS_IPV4) {
2602 		ASSERT(IPH_HDR_VERSION(rptr) == IPV4_VERSION);
2603 		ASSERT(MBLKL(mp) >= sizeof (ipha_t));
2604 		ASSERT(ira->ira_ip_hdr_length == IPH_HDR_LENGTH(rptr));
2605 
2606 		ipha = (ipha_t *)mp->b_rptr;
2607 		if (recv_ancillary.crb_all != 0)
2608 			(void) ip_find_hdr_v4(ipha, &ipps, B_FALSE);
2609 
2610 		/*
2611 		 * BSD for some reason adjusts ipha_length to exclude the
2612 		 * IP header length. We do the same.
2613 		 */
2614 		if (is->is_bsd_compat) {
2615 			ushort_t len;
2616 
2617 			len = ntohs(ipha->ipha_length);
2618 			if (mp->b_datap->db_ref > 1) {
2619 				/*
2620 				 * Allocate a new IP header so that we can
2621 				 * modify ipha_length.
2622 				 */
2623 				mblk_t	*mp1;
2624 
2625 				mp1 = allocb(ip_hdr_length, BPRI_MED);
2626 				if (mp1 == NULL) {
2627 					freemsg(mp);
2628 					BUMP_MIB(&is->is_rawip_mib,
2629 					    rawipInErrors);
2630 					return;
2631 				}
2632 				bcopy(rptr, mp1->b_rptr, ip_hdr_length);
2633 				mp->b_rptr = rptr + ip_hdr_length;
2634 				rptr = mp1->b_rptr;
2635 				ipha = (ipha_t *)rptr;
2636 				mp1->b_cont = mp;
2637 				mp1->b_wptr = rptr + ip_hdr_length;
2638 				mp = mp1;
2639 			}
2640 			len -= ip_hdr_length;
2641 			ipha->ipha_length = htons(len);
2642 		}
2643 
2644 		/*
2645 		 * For RAW sockets we not pass ICMP/IPv4 packets to AF_INET6
2646 		 * sockets. This is ensured by icmp_bind and the IP fanout code.
2647 		 */
2648 		ASSERT(connp->conn_family == AF_INET);
2649 
2650 		/*
2651 		 * This is the inbound data path.  Packets are passed upstream
2652 		 * as T_UNITDATA_IND messages with full IPv4 headers still
2653 		 * attached.
2654 		 */
2655 
2656 		/*
2657 		 * Normally only send up the source address.
2658 		 * If any ancillary data items are wanted we add those.
2659 		 */
2660 		udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin_t);
2661 		if (recv_ancillary.crb_all != 0) {
2662 			udi_size += conn_recvancillary_size(connp,
2663 			    recv_ancillary, ira, mp, &ipps);
2664 		}
2665 
2666 		/* Allocate a message block for the T_UNITDATA_IND structure. */
2667 		mp1 = allocb(udi_size, BPRI_MED);
2668 		if (mp1 == NULL) {
2669 			freemsg(mp);
2670 			BUMP_MIB(&is->is_rawip_mib, rawipInErrors);
2671 			return;
2672 		}
2673 		mp1->b_cont = mp;
2674 		tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2675 		mp1->b_datap->db_type = M_PROTO;
2676 		mp1->b_wptr = (uchar_t *)tudi + udi_size;
2677 		tudi->PRIM_type = T_UNITDATA_IND;
2678 		tudi->SRC_length = sizeof (sin_t);
2679 		tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2680 		sin = (sin_t *)&tudi[1];
2681 		*sin = sin_null;
2682 		sin->sin_family = AF_INET;
2683 		sin->sin_addr.s_addr = ipha->ipha_src;
2684 		*(uint32_t *)&sin->sin_zero[0] = 0;
2685 		*(uint32_t *)&sin->sin_zero[4] = 0;
2686 		tudi->OPT_offset =  sizeof (struct T_unitdata_ind) +
2687 		    sizeof (sin_t);
2688 		udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin_t));
2689 		tudi->OPT_length = udi_size;
2690 
2691 		/*
2692 		 * Add options if IP_RECVIF etc is set
2693 		 */
2694 		if (udi_size != 0) {
2695 			conn_recvancillary_add(connp, recv_ancillary, ira,
2696 			    &ipps, (uchar_t *)&sin[1], udi_size);
2697 		}
2698 		goto deliver;
2699 	}
2700 
2701 	ASSERT(IPH_HDR_VERSION(rptr) == IPV6_VERSION);
2702 	/*
2703 	 * IPv6 packets can only be received by applications
2704 	 * that are prepared to receive IPv6 addresses.
2705 	 * The IP fanout must ensure this.
2706 	 */
2707 	ASSERT(connp->conn_family == AF_INET6);
2708 
2709 	/*
2710 	 * Handle IPv6 packets. We don't pass up the IP headers with the
2711 	 * payload for IPv6.
2712 	 */
2713 
2714 	ip6h = (ip6_t *)rptr;
2715 	if (recv_ancillary.crb_all != 0) {
2716 		/*
2717 		 * Call on ip_find_hdr_v6 which gets individual lenghts of
2718 		 * extension headers (and pointers to them).
2719 		 */
2720 		uint8_t		nexthdr;
2721 
2722 		/* We don't care about the length or nextheader. */
2723 		(void) ip_find_hdr_v6(mp, ip6h, B_TRUE, &ipps, &nexthdr);
2724 
2725 		/*
2726 		 * We do not pass up hop-by-hop options or any other
2727 		 * extension header as part of the packet. Applications
2728 		 * that want to see them have to specify IPV6_RECV* socket
2729 		 * options. And conn_recvancillary_size/add explicitly
2730 		 * drops the TX option from IPV6_HOPOPTS as it does for UDP.
2731 		 *
2732 		 * If we had multilevel ICMP sockets, then we'd want to
2733 		 * modify conn_recvancillary_size/add to
2734 		 * allow the user to see the label.
2735 		 */
2736 	}
2737 
2738 	/*
2739 	 * Check a filter for ICMPv6 types if needed.
2740 	 * Verify raw checksums if needed.
2741 	 */
2742 	mutex_enter(&connp->conn_lock);
2743 	if (icmp->icmp_filter != NULL) {
2744 		int type;
2745 
2746 		/* Assumes that IP has done the pullupmsg */
2747 		type = mp->b_rptr[ip_hdr_length];
2748 
2749 		ASSERT(mp->b_rptr + ip_hdr_length <= mp->b_wptr);
2750 		if (ICMP6_FILTER_WILLBLOCK(type, icmp->icmp_filter)) {
2751 			mutex_exit(&connp->conn_lock);
2752 			freemsg(mp);
2753 			return;
2754 		}
2755 	}
2756 	if (connp->conn_ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
2757 		/* Checksum */
2758 		uint16_t	*up;
2759 		uint32_t	sum;
2760 		int		remlen;
2761 
2762 		up = (uint16_t *)&ip6h->ip6_src;
2763 
2764 		remlen = msgdsize(mp) - ip_hdr_length;
2765 		sum = htons(connp->conn_proto + remlen)
2766 		    + up[0] + up[1] + up[2] + up[3]
2767 		    + up[4] + up[5] + up[6] + up[7]
2768 		    + up[8] + up[9] + up[10] + up[11]
2769 		    + up[12] + up[13] + up[14] + up[15];
2770 		sum = (sum & 0xffff) + (sum >> 16);
2771 		sum = IP_CSUM(mp, ip_hdr_length, sum);
2772 		if (sum != 0) {
2773 			/* IPv6 RAW checksum failed */
2774 			ip0dbg(("icmp_rput: RAW checksum failed %x\n", sum));
2775 			mutex_exit(&connp->conn_lock);
2776 			freemsg(mp);
2777 			BUMP_MIB(&is->is_rawip_mib, rawipInCksumErrs);
2778 			return;
2779 		}
2780 	}
2781 	mutex_exit(&connp->conn_lock);
2782 
2783 	udi_size = sizeof (struct T_unitdata_ind) + sizeof (sin6_t);
2784 
2785 	if (recv_ancillary.crb_all != 0) {
2786 		udi_size += conn_recvancillary_size(connp,
2787 		    recv_ancillary, ira, mp, &ipps);
2788 	}
2789 
2790 	mp1 = allocb(udi_size, BPRI_MED);
2791 	if (mp1 == NULL) {
2792 		freemsg(mp);
2793 		BUMP_MIB(&is->is_rawip_mib, rawipInErrors);
2794 		return;
2795 	}
2796 	mp1->b_cont = mp;
2797 	mp1->b_datap->db_type = M_PROTO;
2798 	tudi = (struct T_unitdata_ind *)mp1->b_rptr;
2799 	mp1->b_wptr = (uchar_t *)tudi + udi_size;
2800 	tudi->PRIM_type = T_UNITDATA_IND;
2801 	tudi->SRC_length = sizeof (sin6_t);
2802 	tudi->SRC_offset = sizeof (struct T_unitdata_ind);
2803 	tudi->OPT_offset = sizeof (struct T_unitdata_ind) + sizeof (sin6_t);
2804 	udi_size -= (sizeof (struct T_unitdata_ind) + sizeof (sin6_t));
2805 	tudi->OPT_length = udi_size;
2806 	sin6 = (sin6_t *)&tudi[1];
2807 	*sin6 = sin6_null;
2808 	sin6->sin6_port = 0;
2809 	sin6->sin6_family = AF_INET6;
2810 
2811 	sin6->sin6_addr = ip6h->ip6_src;
2812 	/* No sin6_flowinfo per API */
2813 	sin6->sin6_flowinfo = 0;
2814 	/* For link-scope pass up scope id */
2815 	if (IN6_IS_ADDR_LINKSCOPE(&ip6h->ip6_src))
2816 		sin6->sin6_scope_id = ira->ira_ruifindex;
2817 	else
2818 		sin6->sin6_scope_id = 0;
2819 	sin6->__sin6_src_id = ip_srcid_find_addr(&ip6h->ip6_dst,
2820 	    IPCL_ZONEID(connp), is->is_netstack);
2821 
2822 	if (udi_size != 0) {
2823 		conn_recvancillary_add(connp, recv_ancillary, ira,
2824 		    &ipps, (uchar_t *)&sin6[1], udi_size);
2825 	}
2826 
2827 	/* Skip all the IPv6 headers per API */
2828 	mp->b_rptr += ip_hdr_length;
2829 	pkt_len -= ip_hdr_length;
2830 
2831 deliver:
2832 	BUMP_MIB(&is->is_rawip_mib, rawipInDatagrams);
2833 	icmp_ulp_recv(connp, mp1, pkt_len);
2834 }
2835 
2836 /*
2837  * return SNMP stuff in buffer in mpdata. We don't hold any lock and report
2838  * information that can be changing beneath us.
2839  */
2840 mblk_t *
2841 icmp_snmp_get(queue_t *q, mblk_t *mpctl)
2842 {
2843 	mblk_t			*mpdata;
2844 	struct opthdr		*optp;
2845 	conn_t			*connp = Q_TO_CONN(q);
2846 	icmp_stack_t		*is = connp->conn_netstack->netstack_icmp;
2847 	mblk_t			*mp2ctl;
2848 
2849 	/*
2850 	 * make a copy of the original message
2851 	 */
2852 	mp2ctl = copymsg(mpctl);
2853 
2854 	if (mpctl == NULL ||
2855 	    (mpdata = mpctl->b_cont) == NULL) {
2856 		freemsg(mpctl);
2857 		freemsg(mp2ctl);
2858 		return (0);
2859 	}
2860 
2861 	/* fixed length structure for IPv4 and IPv6 counters */
2862 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
2863 	optp->level = EXPER_RAWIP;
2864 	optp->name = 0;
2865 	(void) snmp_append_data(mpdata, (char *)&is->is_rawip_mib,
2866 	    sizeof (is->is_rawip_mib));
2867 	optp->len = msgdsize(mpdata);
2868 	qreply(q, mpctl);
2869 
2870 	return (mp2ctl);
2871 }
2872 
2873 /*
2874  * Return 0 if invalid set request, 1 otherwise, including non-rawip requests.
2875  * TODO:  If this ever actually tries to set anything, it needs to be
2876  * to do the appropriate locking.
2877  */
2878 /* ARGSUSED */
2879 int
2880 icmp_snmp_set(queue_t *q, t_scalar_t level, t_scalar_t name,
2881     uchar_t *ptr, int len)
2882 {
2883 	switch (level) {
2884 	case EXPER_RAWIP:
2885 		return (0);
2886 	default:
2887 		return (1);
2888 	}
2889 }
2890 
2891 /*
2892  * This routine creates a T_UDERROR_IND message and passes it upstream.
2893  * The address and options are copied from the T_UNITDATA_REQ message
2894  * passed in mp.  This message is freed.
2895  */
2896 static void
2897 icmp_ud_err(queue_t *q, mblk_t *mp, t_scalar_t err)
2898 {
2899 	struct T_unitdata_req *tudr;
2900 	mblk_t	*mp1;
2901 	uchar_t *destaddr;
2902 	t_scalar_t destlen;
2903 	uchar_t	*optaddr;
2904 	t_scalar_t optlen;
2905 
2906 	if ((mp->b_wptr < mp->b_rptr) ||
2907 	    (MBLKL(mp)) < sizeof (struct T_unitdata_req)) {
2908 		goto done;
2909 	}
2910 	tudr = (struct T_unitdata_req *)mp->b_rptr;
2911 	destaddr = mp->b_rptr + tudr->DEST_offset;
2912 	if (destaddr < mp->b_rptr || destaddr >= mp->b_wptr ||
2913 	    destaddr + tudr->DEST_length < mp->b_rptr ||
2914 	    destaddr + tudr->DEST_length > mp->b_wptr) {
2915 		goto done;
2916 	}
2917 	optaddr = mp->b_rptr + tudr->OPT_offset;
2918 	if (optaddr < mp->b_rptr || optaddr >= mp->b_wptr ||
2919 	    optaddr + tudr->OPT_length < mp->b_rptr ||
2920 	    optaddr + tudr->OPT_length > mp->b_wptr) {
2921 		goto done;
2922 	}
2923 	destlen = tudr->DEST_length;
2924 	optlen = tudr->OPT_length;
2925 
2926 	mp1 = mi_tpi_uderror_ind((char *)destaddr, destlen,
2927 	    (char *)optaddr, optlen, err);
2928 	if (mp1 != NULL)
2929 		qreply(q, mp1);
2930 
2931 done:
2932 	freemsg(mp);
2933 }
2934 
2935 static int
2936 rawip_do_unbind(conn_t *connp)
2937 {
2938 	icmp_t	*icmp = connp->conn_icmp;
2939 
2940 	mutex_enter(&connp->conn_lock);
2941 	/* If a bind has not been done, we can't unbind. */
2942 	if (icmp->icmp_state == TS_UNBND) {
2943 		mutex_exit(&connp->conn_lock);
2944 		return (-TOUTSTATE);
2945 	}
2946 	connp->conn_saddr_v6 = ipv6_all_zeros;
2947 	connp->conn_bound_addr_v6 = ipv6_all_zeros;
2948 	connp->conn_laddr_v6 = ipv6_all_zeros;
2949 	connp->conn_mcbc_bind = B_FALSE;
2950 	connp->conn_lport = 0;
2951 	connp->conn_fport = 0;
2952 	/* In case we were also connected */
2953 	connp->conn_faddr_v6 = ipv6_all_zeros;
2954 	connp->conn_v6lastdst = ipv6_all_zeros;
2955 
2956 	icmp->icmp_state = TS_UNBND;
2957 
2958 	(void) icmp_build_hdr_template(connp, &connp->conn_saddr_v6,
2959 	    &connp->conn_faddr_v6, connp->conn_flowinfo);
2960 	mutex_exit(&connp->conn_lock);
2961 
2962 	ip_unbind(connp);
2963 	return (0);
2964 }
2965 
2966 /*
2967  * This routine is called by icmp_wput to handle T_UNBIND_REQ messages.
2968  * After some error checking, the message is passed downstream to ip.
2969  */
2970 static void
2971 icmp_tpi_unbind(queue_t *q, mblk_t *mp)
2972 {
2973 	conn_t	*connp = Q_TO_CONN(q);
2974 	int	error;
2975 
2976 	ASSERT(mp->b_cont == NULL);
2977 	error = rawip_do_unbind(connp);
2978 	if (error) {
2979 		if (error < 0) {
2980 			icmp_err_ack(q, mp, -error, 0);
2981 		} else {
2982 			icmp_err_ack(q, mp, 0, error);
2983 		}
2984 		return;
2985 	}
2986 
2987 	/*
2988 	 * Convert mp into a T_OK_ACK
2989 	 */
2990 
2991 	mp = mi_tpi_ok_ack_alloc(mp);
2992 
2993 	/*
2994 	 * should not happen in practice... T_OK_ACK is smaller than the
2995 	 * original message.
2996 	 */
2997 	ASSERT(mp != NULL);
2998 	ASSERT(((struct T_ok_ack *)mp->b_rptr)->PRIM_type == T_OK_ACK);
2999 	qreply(q, mp);
3000 }
3001 
3002 /*
3003  * Process IPv4 packets that already include an IP header.
3004  * Used when IP_HDRINCL has been set (implicit for IPPROTO_RAW and
3005  * IPPROTO_IGMP).
3006  * In this case we ignore the address and any options in the T_UNITDATA_REQ.
3007  *
3008  * The packet is assumed to have a base (20 byte) IP header followed
3009  * by the upper-layer protocol. We include any IP_OPTIONS including a
3010  * CIPSO label but otherwise preserve the base IP header.
3011  */
3012 static int
3013 icmp_output_hdrincl(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid)
3014 {
3015 	icmp_t		*icmp = connp->conn_icmp;
3016 	icmp_stack_t	*is = icmp->icmp_is;
3017 	ipha_t		iphas;
3018 	ipha_t		*ipha;
3019 	int		ip_hdr_length;
3020 	int		tp_hdr_len;
3021 	ip_xmit_attr_t	*ixa;
3022 	ip_pkt_t	*ipp;
3023 	in6_addr_t	v6src;
3024 	in6_addr_t	v6dst;
3025 	in6_addr_t	v6nexthop;
3026 	int		error;
3027 	boolean_t	do_ipsec;
3028 
3029 	/*
3030 	 * We need an exclusive copy of conn_ixa since the included IP
3031 	 * header could have any destination.
3032 	 * That copy has no pointers hence we
3033 	 * need to set them up once we've parsed the ancillary data.
3034 	 */
3035 	ixa = conn_get_ixa_exclusive(connp);
3036 	if (ixa == NULL) {
3037 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3038 		freemsg(mp);
3039 		return (ENOMEM);
3040 	}
3041 	ASSERT(cr != NULL);
3042 	/*
3043 	 * Caller has a reference on cr; from db_credp or because we
3044 	 * are running in process context.
3045 	 */
3046 	ixa->ixa_cred = cr;
3047 	ixa->ixa_cpid = pid;
3048 	if (is_system_labeled()) {
3049 		/* We need to restart with a label based on the cred */
3050 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
3051 	}
3052 
3053 	/* In case previous destination was multicast or multirt */
3054 	ip_attr_newdst(ixa);
3055 
3056 	/* Get a copy of conn_xmit_ipp since the TX label might change it */
3057 	ipp = kmem_zalloc(sizeof (*ipp), KM_NOSLEEP);
3058 	if (ipp == NULL) {
3059 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3060 		ixa->ixa_cpid = connp->conn_cpid;
3061 		ixa_refrele(ixa);
3062 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3063 		freemsg(mp);
3064 		return (ENOMEM);
3065 	}
3066 	mutex_enter(&connp->conn_lock);
3067 	error = ip_pkt_copy(&connp->conn_xmit_ipp, ipp, KM_NOSLEEP);
3068 	mutex_exit(&connp->conn_lock);
3069 	if (error != 0) {
3070 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3071 		freemsg(mp);
3072 		goto done;
3073 	}
3074 
3075 	/* Sanity check length of packet */
3076 	ipha = (ipha_t *)mp->b_rptr;
3077 
3078 	ip_hdr_length = IP_SIMPLE_HDR_LENGTH;
3079 	if ((mp->b_wptr - mp->b_rptr) < IP_SIMPLE_HDR_LENGTH) {
3080 		if (!pullupmsg(mp, IP_SIMPLE_HDR_LENGTH)) {
3081 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3082 			freemsg(mp);
3083 			goto done;
3084 		}
3085 		ipha = (ipha_t *)mp->b_rptr;
3086 	}
3087 	ipha->ipha_version_and_hdr_length =
3088 	    (IP_VERSION<<4) | (ip_hdr_length>>2);
3089 
3090 	/*
3091 	 * We set IXAF_DONTFRAG if the application set DF which makes
3092 	 * IP not fragment.
3093 	 */
3094 	ipha->ipha_fragment_offset_and_flags &= htons(IPH_DF);
3095 	if (ipha->ipha_fragment_offset_and_flags & htons(IPH_DF))
3096 		ixa->ixa_flags |= (IXAF_DONTFRAG | IXAF_PMTU_IPV4_DF);
3097 	else
3098 		ixa->ixa_flags &= ~(IXAF_DONTFRAG | IXAF_PMTU_IPV4_DF);
3099 
3100 	/* Even for multicast and broadcast we honor the apps ttl */
3101 	ixa->ixa_flags |= IXAF_NO_TTL_CHANGE;
3102 
3103 	/*
3104 	 * No source verification for non-local addresses
3105 	 */
3106 	if (ipha->ipha_src != INADDR_ANY &&
3107 	    ip_laddr_verify_v4(ipha->ipha_src, ixa->ixa_zoneid,
3108 	    is->is_netstack->netstack_ip, B_FALSE)
3109 	    != IPVL_UNICAST_UP) {
3110 		ixa->ixa_flags &= ~IXAF_VERIFY_SOURCE;
3111 	}
3112 
3113 	if (ipha->ipha_dst == INADDR_ANY)
3114 		ipha->ipha_dst = htonl(INADDR_LOOPBACK);
3115 
3116 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6src);
3117 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &v6dst);
3118 
3119 	/* Defer IPsec if it might need to look at ICMP type/code */
3120 	do_ipsec = ipha->ipha_protocol != IPPROTO_ICMP;
3121 	ixa->ixa_flags |= IXAF_IS_IPV4;
3122 
3123 	ip_attr_nexthop(ipp, ixa, &v6dst, &v6nexthop);
3124 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop,
3125 	    connp->conn_fport, &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
3126 	    (do_ipsec ? IPDF_IPSEC : 0));
3127 	switch (error) {
3128 	case 0:
3129 		break;
3130 	case EADDRNOTAVAIL:
3131 		/*
3132 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3133 		 * Don't have the application see that errno
3134 		 */
3135 		error = ENETUNREACH;
3136 		goto failed;
3137 	case ENETDOWN:
3138 		/*
3139 		 * Have !ipif_addr_ready address; drop packet silently
3140 		 * until we can get applications to not send until we
3141 		 * are ready.
3142 		 */
3143 		error = 0;
3144 		goto failed;
3145 	case EHOSTUNREACH:
3146 	case ENETUNREACH:
3147 		if (ixa->ixa_ire != NULL) {
3148 			/*
3149 			 * Let conn_ip_output/ire_send_noroute return
3150 			 * the error and send any local ICMP error.
3151 			 */
3152 			error = 0;
3153 			break;
3154 		}
3155 		/* FALLTHRU */
3156 	default:
3157 	failed:
3158 		freemsg(mp);
3159 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3160 		goto done;
3161 	}
3162 	if (ipha->ipha_src == INADDR_ANY)
3163 		IN6_V4MAPPED_TO_IPADDR(&v6src, ipha->ipha_src);
3164 
3165 	/*
3166 	 * We might be going to a different destination than last time,
3167 	 * thus check that TX allows the communication and compute any
3168 	 * needed label.
3169 	 *
3170 	 * TSOL Note: We have an exclusive ipp and ixa for this thread so we
3171 	 * don't have to worry about concurrent threads.
3172 	 */
3173 	if (is_system_labeled()) {
3174 		/*
3175 		 * Check whether Trusted Solaris policy allows communication
3176 		 * with this host, and pretend that the destination is
3177 		 * unreachable if not.
3178 		 * Compute any needed label and place it in ipp_label_v4/v6.
3179 		 *
3180 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
3181 		 * ipp_label_v4/v6 to form the packet.
3182 		 *
3183 		 * Tsol note: We have ipp structure local to this thread so
3184 		 * no locking is needed.
3185 		 */
3186 		error = conn_update_label(connp, ixa, &v6dst, ipp);
3187 		if (error != 0) {
3188 			freemsg(mp);
3189 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3190 			goto done;
3191 		}
3192 	}
3193 
3194 	/*
3195 	 * Save away a copy of the IPv4 header the application passed down
3196 	 * and then prepend an IPv4 header complete with any IP options
3197 	 * including label.
3198 	 * We need a struct copy since icmp_prepend_hdr will reuse the available
3199 	 * space in the mblk.
3200 	 */
3201 	iphas = *ipha;
3202 	mp->b_rptr += IP_SIMPLE_HDR_LENGTH;
3203 
3204 	mp = icmp_prepend_hdr(connp, ixa, ipp, &v6src, &v6dst, 0, mp, &error);
3205 	if (mp == NULL) {
3206 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3207 		ASSERT(error != 0);
3208 		goto done;
3209 	}
3210 	if (ixa->ixa_pktlen > IP_MAXPACKET) {
3211 		error = EMSGSIZE;
3212 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3213 		freemsg(mp);
3214 		goto done;
3215 	}
3216 	/* Restore key parts of the header that the application passed down */
3217 	ipha = (ipha_t *)mp->b_rptr;
3218 	ipha->ipha_type_of_service = iphas.ipha_type_of_service;
3219 	ipha->ipha_ident = iphas.ipha_ident;
3220 	ipha->ipha_fragment_offset_and_flags =
3221 	    iphas.ipha_fragment_offset_and_flags;
3222 	ipha->ipha_ttl = iphas.ipha_ttl;
3223 	ipha->ipha_protocol = iphas.ipha_protocol;
3224 	ipha->ipha_src = iphas.ipha_src;
3225 	ipha->ipha_dst = iphas.ipha_dst;
3226 
3227 	ixa->ixa_protocol = ipha->ipha_protocol;
3228 
3229 	/*
3230 	 * Make sure that the IP header plus any transport header that is
3231 	 * checksumed by ip_output is in the first mblk. (ip_output assumes
3232 	 * that at least the checksum field is in the first mblk.)
3233 	 */
3234 	switch (ipha->ipha_protocol) {
3235 	case IPPROTO_UDP:
3236 		tp_hdr_len = 8;
3237 		break;
3238 	case IPPROTO_TCP:
3239 		tp_hdr_len = 20;
3240 		break;
3241 	default:
3242 		tp_hdr_len = 0;
3243 		break;
3244 	}
3245 	ip_hdr_length = IPH_HDR_LENGTH(ipha);
3246 	if (mp->b_wptr - mp->b_rptr < ip_hdr_length + tp_hdr_len) {
3247 		if (!pullupmsg(mp, ip_hdr_length + tp_hdr_len)) {
3248 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3249 			if (mp->b_cont == NULL)
3250 				error = EINVAL;
3251 			else
3252 				error = ENOMEM;
3253 			freemsg(mp);
3254 			goto done;
3255 		}
3256 	}
3257 
3258 	if (!do_ipsec) {
3259 		/* Policy might differ for different ICMP type/code */
3260 		if (ixa->ixa_ipsec_policy != NULL) {
3261 			IPPOL_REFRELE(ixa->ixa_ipsec_policy);
3262 			ixa->ixa_ipsec_policy = NULL;
3263 			ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
3264 		}
3265 		mp = ip_output_attach_policy(mp, ipha, NULL, connp, ixa);
3266 		if (mp == NULL) {
3267 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3268 			error = EHOSTUNREACH;	/* IPsec policy failure */
3269 			goto done;
3270 		}
3271 	}
3272 
3273 	/* We're done.  Pass the packet to ip. */
3274 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3275 
3276 	error = conn_ip_output(mp, ixa);
3277 	/* No rawipOutErrors if an error since IP increases its error counter */
3278 	switch (error) {
3279 	case 0:
3280 		break;
3281 	case EWOULDBLOCK:
3282 		(void) ixa_check_drain_insert(connp, ixa);
3283 		error = 0;
3284 		break;
3285 	case EADDRNOTAVAIL:
3286 		/*
3287 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3288 		 * Don't have the application see that errno
3289 		 */
3290 		error = ENETUNREACH;
3291 		break;
3292 	}
3293 done:
3294 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3295 	ixa->ixa_cpid = connp->conn_cpid;
3296 	ixa_refrele(ixa);
3297 	ip_pkt_free(ipp);
3298 	kmem_free(ipp, sizeof (*ipp));
3299 	return (error);
3300 }
3301 
3302 static mblk_t *
3303 icmp_output_attach_policy(mblk_t *mp, conn_t *connp, ip_xmit_attr_t *ixa)
3304 {
3305 	ipha_t	*ipha = NULL;
3306 	ip6_t	*ip6h = NULL;
3307 
3308 	if (ixa->ixa_flags & IXAF_IS_IPV4)
3309 		ipha = (ipha_t *)mp->b_rptr;
3310 	else
3311 		ip6h = (ip6_t *)mp->b_rptr;
3312 
3313 	if (ixa->ixa_ipsec_policy != NULL) {
3314 		IPPOL_REFRELE(ixa->ixa_ipsec_policy);
3315 		ixa->ixa_ipsec_policy = NULL;
3316 		ixa->ixa_flags &= ~IXAF_IPSEC_SECURE;
3317 	}
3318 	return (ip_output_attach_policy(mp, ipha, ip6h, connp, ixa));
3319 }
3320 
3321 /*
3322  * Handle T_UNITDATA_REQ with options. Both IPv4 and IPv6
3323  * Either tudr_mp or msg is set. If tudr_mp we take ancillary data from
3324  * the TPI options, otherwise we take them from msg_control.
3325  * If both sin and sin6 is set it is a connected socket and we use conn_faddr.
3326  * Always consumes mp; never consumes tudr_mp.
3327  */
3328 static int
3329 icmp_output_ancillary(conn_t *connp, sin_t *sin, sin6_t *sin6, mblk_t *mp,
3330     mblk_t *tudr_mp, struct nmsghdr *msg, cred_t *cr, pid_t pid)
3331 {
3332 	icmp_t		*icmp = connp->conn_icmp;
3333 	icmp_stack_t	*is = icmp->icmp_is;
3334 	int		error;
3335 	ip_xmit_attr_t	*ixa;
3336 	ip_pkt_t	*ipp;
3337 	in6_addr_t	v6src;
3338 	in6_addr_t	v6dst;
3339 	in6_addr_t	v6nexthop;
3340 	in_port_t	dstport;
3341 	uint32_t	flowinfo;
3342 	uint_t		srcid;
3343 	int		is_absreq_failure = 0;
3344 	conn_opt_arg_t	coas, *coa;
3345 
3346 	ASSERT(tudr_mp != NULL || msg != NULL);
3347 
3348 	/*
3349 	 * Get ixa before checking state to handle a disconnect race.
3350 	 *
3351 	 * We need an exclusive copy of conn_ixa since the ancillary data
3352 	 * options might modify it. That copy has no pointers hence we
3353 	 * need to set them up once we've parsed the ancillary data.
3354 	 */
3355 	ixa = conn_get_ixa_exclusive(connp);
3356 	if (ixa == NULL) {
3357 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3358 		freemsg(mp);
3359 		return (ENOMEM);
3360 	}
3361 	ASSERT(cr != NULL);
3362 	ixa->ixa_cred = cr;
3363 	ixa->ixa_cpid = pid;
3364 	if (is_system_labeled()) {
3365 		/* We need to restart with a label based on the cred */
3366 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
3367 	}
3368 
3369 	/* In case previous destination was multicast or multirt */
3370 	ip_attr_newdst(ixa);
3371 
3372 	/* Get a copy of conn_xmit_ipp since the options might change it */
3373 	ipp = kmem_zalloc(sizeof (*ipp), KM_NOSLEEP);
3374 	if (ipp == NULL) {
3375 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3376 		ixa->ixa_cpid = connp->conn_cpid;
3377 		ixa_refrele(ixa);
3378 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3379 		freemsg(mp);
3380 		return (ENOMEM);
3381 	}
3382 	mutex_enter(&connp->conn_lock);
3383 	error = ip_pkt_copy(&connp->conn_xmit_ipp, ipp, KM_NOSLEEP);
3384 	mutex_exit(&connp->conn_lock);
3385 	if (error != 0) {
3386 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3387 		freemsg(mp);
3388 		goto done;
3389 	}
3390 
3391 	/*
3392 	 * Parse the options and update ixa and ipp as a result.
3393 	 */
3394 
3395 	coa = &coas;
3396 	coa->coa_connp = connp;
3397 	coa->coa_ixa = ixa;
3398 	coa->coa_ipp = ipp;
3399 	coa->coa_ancillary = B_TRUE;
3400 	coa->coa_changed = 0;
3401 
3402 	if (msg != NULL) {
3403 		error = process_auxiliary_options(connp, msg->msg_control,
3404 		    msg->msg_controllen, coa, &icmp_opt_obj, icmp_opt_set, cr);
3405 	} else {
3406 		struct T_unitdata_req *tudr;
3407 
3408 		tudr = (struct T_unitdata_req *)tudr_mp->b_rptr;
3409 		ASSERT(tudr->PRIM_type == T_UNITDATA_REQ);
3410 		error = tpi_optcom_buf(connp->conn_wq, tudr_mp,
3411 		    &tudr->OPT_length, tudr->OPT_offset, cr, &icmp_opt_obj,
3412 		    coa, &is_absreq_failure);
3413 	}
3414 	if (error != 0) {
3415 		/*
3416 		 * Note: No special action needed in this
3417 		 * module for "is_absreq_failure"
3418 		 */
3419 		freemsg(mp);
3420 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3421 		goto done;
3422 	}
3423 	ASSERT(is_absreq_failure == 0);
3424 
3425 	mutex_enter(&connp->conn_lock);
3426 	/*
3427 	 * If laddr is unspecified then we look at sin6_src_id.
3428 	 * We will give precedence to a source address set with IPV6_PKTINFO
3429 	 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
3430 	 * want ip_attr_connect to select a source (since it can fail) when
3431 	 * IPV6_PKTINFO is specified.
3432 	 * If this doesn't result in a source address then we get a source
3433 	 * from ip_attr_connect() below.
3434 	 */
3435 	v6src = connp->conn_saddr_v6;
3436 	if (sin != NULL) {
3437 		IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
3438 		dstport = sin->sin_port;
3439 		flowinfo = 0;
3440 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3441 		ixa->ixa_flags |= IXAF_IS_IPV4;
3442 	} else if (sin6 != NULL) {
3443 		v6dst = sin6->sin6_addr;
3444 		dstport = sin6->sin6_port;
3445 		flowinfo = sin6->sin6_flowinfo;
3446 		srcid = sin6->__sin6_src_id;
3447 		if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
3448 			ixa->ixa_scopeid = sin6->sin6_scope_id;
3449 			ixa->ixa_flags |= IXAF_SCOPEID_SET;
3450 		} else {
3451 			ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
3452 		}
3453 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
3454 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
3455 			    connp->conn_netstack);
3456 		}
3457 		if (IN6_IS_ADDR_V4MAPPED(&v6dst))
3458 			ixa->ixa_flags |= IXAF_IS_IPV4;
3459 		else
3460 			ixa->ixa_flags &= ~IXAF_IS_IPV4;
3461 	} else {
3462 		/* Connected case */
3463 		v6dst = connp->conn_faddr_v6;
3464 		flowinfo = connp->conn_flowinfo;
3465 	}
3466 	mutex_exit(&connp->conn_lock);
3467 	/* Handle IPV6_PKTINFO setting source address. */
3468 	if (IN6_IS_ADDR_UNSPECIFIED(&v6src) &&
3469 	    (ipp->ipp_fields & IPPF_ADDR)) {
3470 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
3471 			if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3472 				v6src = ipp->ipp_addr;
3473 		} else {
3474 			if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
3475 				v6src = ipp->ipp_addr;
3476 		}
3477 	}
3478 	/*
3479 	 * Allow source not assigned to the system
3480 	 * only if it is not a local addresses
3481 	 */
3482 	if (!V6_OR_V4_INADDR_ANY(v6src)) {
3483 		ip_laddr_t laddr_type;
3484 
3485 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
3486 			ipaddr_t v4src;
3487 
3488 			IN6_V4MAPPED_TO_IPADDR(&v6src, v4src);
3489 			laddr_type = ip_laddr_verify_v4(v4src, ixa->ixa_zoneid,
3490 			    is->is_netstack->netstack_ip, B_FALSE);
3491 		} else {
3492 			laddr_type = ip_laddr_verify_v6(&v6src, ixa->ixa_zoneid,
3493 			    is->is_netstack->netstack_ip, B_FALSE, B_FALSE);
3494 		}
3495 		if (laddr_type != IPVL_UNICAST_UP)
3496 			ixa->ixa_flags &= ~IXAF_VERIFY_SOURCE;
3497 	}
3498 
3499 	ip_attr_nexthop(ipp, ixa, &v6dst, &v6nexthop);
3500 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
3501 	    &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST);
3502 
3503 	switch (error) {
3504 	case 0:
3505 		break;
3506 	case EADDRNOTAVAIL:
3507 		/*
3508 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3509 		 * Don't have the application see that errno
3510 		 */
3511 		error = ENETUNREACH;
3512 		goto failed;
3513 	case ENETDOWN:
3514 		/*
3515 		 * Have !ipif_addr_ready address; drop packet silently
3516 		 * until we can get applications to not send until we
3517 		 * are ready.
3518 		 */
3519 		error = 0;
3520 		goto failed;
3521 	case EHOSTUNREACH:
3522 	case ENETUNREACH:
3523 		if (ixa->ixa_ire != NULL) {
3524 			/*
3525 			 * Let conn_ip_output/ire_send_noroute return
3526 			 * the error and send any local ICMP error.
3527 			 */
3528 			error = 0;
3529 			break;
3530 		}
3531 		/* FALLTHRU */
3532 	default:
3533 	failed:
3534 		freemsg(mp);
3535 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3536 		goto done;
3537 	}
3538 
3539 	/*
3540 	 * We might be going to a different destination than last time,
3541 	 * thus check that TX allows the communication and compute any
3542 	 * needed label.
3543 	 *
3544 	 * TSOL Note: We have an exclusive ipp and ixa for this thread so we
3545 	 * don't have to worry about concurrent threads.
3546 	 */
3547 	if (is_system_labeled()) {
3548 		/*
3549 		 * Check whether Trusted Solaris policy allows communication
3550 		 * with this host, and pretend that the destination is
3551 		 * unreachable if not.
3552 		 * Compute any needed label and place it in ipp_label_v4/v6.
3553 		 *
3554 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
3555 		 * ipp_label_v4/v6 to form the packet.
3556 		 *
3557 		 * Tsol note: We have ipp structure local to this thread so
3558 		 * no locking is needed.
3559 		 */
3560 		error = conn_update_label(connp, ixa, &v6dst, ipp);
3561 		if (error != 0) {
3562 			freemsg(mp);
3563 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3564 			goto done;
3565 		}
3566 	}
3567 	mp = icmp_prepend_hdr(connp, ixa, ipp, &v6src, &v6dst, flowinfo, mp,
3568 	    &error);
3569 	if (mp == NULL) {
3570 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3571 		ASSERT(error != 0);
3572 		goto done;
3573 	}
3574 	if (ixa->ixa_pktlen > IP_MAXPACKET) {
3575 		error = EMSGSIZE;
3576 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3577 		freemsg(mp);
3578 		goto done;
3579 	}
3580 
3581 	/* Policy might differ for different ICMP type/code */
3582 	mp = icmp_output_attach_policy(mp, connp, ixa);
3583 	if (mp == NULL) {
3584 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3585 		error = EHOSTUNREACH;	/* IPsec policy failure */
3586 		goto done;
3587 	}
3588 
3589 	/* We're done.  Pass the packet to ip. */
3590 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3591 
3592 	error = conn_ip_output(mp, ixa);
3593 	if (!connp->conn_unspec_src)
3594 		ixa->ixa_flags |= IXAF_VERIFY_SOURCE;
3595 	/* No rawipOutErrors if an error since IP increases its error counter */
3596 	switch (error) {
3597 	case 0:
3598 		break;
3599 	case EWOULDBLOCK:
3600 		(void) ixa_check_drain_insert(connp, ixa);
3601 		error = 0;
3602 		break;
3603 	case EADDRNOTAVAIL:
3604 		/*
3605 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3606 		 * Don't have the application see that errno
3607 		 */
3608 		error = ENETUNREACH;
3609 		/* FALLTHRU */
3610 	default:
3611 		mutex_enter(&connp->conn_lock);
3612 		/*
3613 		 * Clear the source and v6lastdst so we call ip_attr_connect
3614 		 * for the next packet and try to pick a better source.
3615 		 */
3616 		if (connp->conn_mcbc_bind)
3617 			connp->conn_saddr_v6 = ipv6_all_zeros;
3618 		else
3619 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
3620 		connp->conn_v6lastdst = ipv6_all_zeros;
3621 		mutex_exit(&connp->conn_lock);
3622 		break;
3623 	}
3624 done:
3625 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3626 	ixa->ixa_cpid = connp->conn_cpid;
3627 	ixa_refrele(ixa);
3628 	ip_pkt_free(ipp);
3629 	kmem_free(ipp, sizeof (*ipp));
3630 	return (error);
3631 }
3632 
3633 /*
3634  * Handle sending an M_DATA for a connected socket.
3635  * Handles both IPv4 and IPv6.
3636  */
3637 int
3638 icmp_output_connected(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid)
3639 {
3640 	icmp_t		*icmp = connp->conn_icmp;
3641 	icmp_stack_t	*is = icmp->icmp_is;
3642 	int		error;
3643 	ip_xmit_attr_t	*ixa;
3644 	boolean_t	do_ipsec;
3645 
3646 	/*
3647 	 * If no other thread is using conn_ixa this just gets a reference to
3648 	 * conn_ixa. Otherwise we get a safe copy of conn_ixa.
3649 	 */
3650 	ixa = conn_get_ixa(connp, B_FALSE);
3651 	if (ixa == NULL) {
3652 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3653 		freemsg(mp);
3654 		return (ENOMEM);
3655 	}
3656 
3657 	ASSERT(cr != NULL);
3658 	ixa->ixa_cred = cr;
3659 	ixa->ixa_cpid = pid;
3660 
3661 	/* Defer IPsec if it might need to look at ICMP type/code */
3662 	switch (ixa->ixa_protocol) {
3663 	case IPPROTO_ICMP:
3664 	case IPPROTO_ICMPV6:
3665 		do_ipsec = B_FALSE;
3666 		break;
3667 	default:
3668 		do_ipsec = B_TRUE;
3669 	}
3670 
3671 	mutex_enter(&connp->conn_lock);
3672 	mp = icmp_prepend_header_template(connp, ixa, mp,
3673 	    &connp->conn_saddr_v6, connp->conn_flowinfo, &error);
3674 
3675 	if (mp == NULL) {
3676 		ASSERT(error != 0);
3677 		mutex_exit(&connp->conn_lock);
3678 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3679 		ixa->ixa_cpid = connp->conn_cpid;
3680 		ixa_refrele(ixa);
3681 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3682 		freemsg(mp);
3683 		return (error);
3684 	}
3685 
3686 	if (!do_ipsec) {
3687 		/* Policy might differ for different ICMP type/code */
3688 		mp = icmp_output_attach_policy(mp, connp, ixa);
3689 		if (mp == NULL) {
3690 			mutex_exit(&connp->conn_lock);
3691 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3692 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3693 			ixa->ixa_cpid = connp->conn_cpid;
3694 			ixa_refrele(ixa);
3695 			return (EHOSTUNREACH);	/* IPsec policy failure */
3696 		}
3697 	}
3698 
3699 	/*
3700 	 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
3701 	 * safe copy, then we need to fill in any pointers in it.
3702 	 */
3703 	if (ixa->ixa_ire == NULL) {
3704 		in6_addr_t	faddr, saddr;
3705 		in6_addr_t	nexthop;
3706 		in_port_t	fport;
3707 
3708 		saddr = connp->conn_saddr_v6;
3709 		faddr = connp->conn_faddr_v6;
3710 		fport = connp->conn_fport;
3711 		ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &faddr, &nexthop);
3712 		mutex_exit(&connp->conn_lock);
3713 
3714 		error = ip_attr_connect(connp, ixa, &saddr, &faddr, &nexthop,
3715 		    fport, NULL, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
3716 		    (do_ipsec ? IPDF_IPSEC : 0));
3717 		switch (error) {
3718 		case 0:
3719 			break;
3720 		case EADDRNOTAVAIL:
3721 			/*
3722 			 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3723 			 * Don't have the application see that errno
3724 			 */
3725 			error = ENETUNREACH;
3726 			goto failed;
3727 		case ENETDOWN:
3728 			/*
3729 			 * Have !ipif_addr_ready address; drop packet silently
3730 			 * until we can get applications to not send until we
3731 			 * are ready.
3732 			 */
3733 			error = 0;
3734 			goto failed;
3735 		case EHOSTUNREACH:
3736 		case ENETUNREACH:
3737 			if (ixa->ixa_ire != NULL) {
3738 				/*
3739 				 * Let conn_ip_output/ire_send_noroute return
3740 				 * the error and send any local ICMP error.
3741 				 */
3742 				error = 0;
3743 				break;
3744 			}
3745 			/* FALLTHRU */
3746 		default:
3747 		failed:
3748 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3749 			ixa->ixa_cpid = connp->conn_cpid;
3750 			ixa_refrele(ixa);
3751 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3752 			freemsg(mp);
3753 			return (error);
3754 		}
3755 	} else {
3756 		/* Done with conn_t */
3757 		mutex_exit(&connp->conn_lock);
3758 	}
3759 
3760 	/* We're done.  Pass the packet to ip. */
3761 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3762 
3763 	error = conn_ip_output(mp, ixa);
3764 	/* No rawipOutErrors if an error since IP increases its error counter */
3765 	switch (error) {
3766 	case 0:
3767 		break;
3768 	case EWOULDBLOCK:
3769 		(void) ixa_check_drain_insert(connp, ixa);
3770 		error = 0;
3771 		break;
3772 	case EADDRNOTAVAIL:
3773 		/*
3774 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3775 		 * Don't have the application see that errno
3776 		 */
3777 		error = ENETUNREACH;
3778 		break;
3779 	}
3780 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3781 	ixa->ixa_cpid = connp->conn_cpid;
3782 	ixa_refrele(ixa);
3783 	return (error);
3784 }
3785 
3786 /*
3787  * Handle sending an M_DATA to the last destination.
3788  * Handles both IPv4 and IPv6.
3789  *
3790  * NOTE: The caller must hold conn_lock and we drop it here.
3791  */
3792 int
3793 icmp_output_lastdst(conn_t *connp, mblk_t *mp, cred_t *cr, pid_t pid,
3794     ip_xmit_attr_t *ixa)
3795 {
3796 	icmp_t		*icmp = connp->conn_icmp;
3797 	icmp_stack_t	*is = icmp->icmp_is;
3798 	int		error;
3799 	boolean_t	do_ipsec;
3800 
3801 	ASSERT(MUTEX_HELD(&connp->conn_lock));
3802 	ASSERT(ixa != NULL);
3803 
3804 	ASSERT(cr != NULL);
3805 	ixa->ixa_cred = cr;
3806 	ixa->ixa_cpid = pid;
3807 
3808 	/* Defer IPsec if it might need to look at ICMP type/code */
3809 	switch (ixa->ixa_protocol) {
3810 	case IPPROTO_ICMP:
3811 	case IPPROTO_ICMPV6:
3812 		do_ipsec = B_FALSE;
3813 		break;
3814 	default:
3815 		do_ipsec = B_TRUE;
3816 	}
3817 
3818 
3819 	mp = icmp_prepend_header_template(connp, ixa, mp,
3820 	    &connp->conn_v6lastsrc, connp->conn_lastflowinfo, &error);
3821 
3822 	if (mp == NULL) {
3823 		ASSERT(error != 0);
3824 		mutex_exit(&connp->conn_lock);
3825 		ixa->ixa_cred = connp->conn_cred;	/* Restore */
3826 		ixa->ixa_cpid = connp->conn_cpid;
3827 		ixa_refrele(ixa);
3828 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3829 		freemsg(mp);
3830 		return (error);
3831 	}
3832 
3833 	if (!do_ipsec) {
3834 		/* Policy might differ for different ICMP type/code */
3835 		mp = icmp_output_attach_policy(mp, connp, ixa);
3836 		if (mp == NULL) {
3837 			mutex_exit(&connp->conn_lock);
3838 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3839 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3840 			ixa->ixa_cpid = connp->conn_cpid;
3841 			ixa_refrele(ixa);
3842 			return (EHOSTUNREACH);	/* IPsec policy failure */
3843 		}
3844 	}
3845 
3846 	/*
3847 	 * In case we got a safe copy of conn_ixa, or if opt_set made us a new
3848 	 * safe copy, then we need to fill in any pointers in it.
3849 	 */
3850 	if (ixa->ixa_ire == NULL) {
3851 		in6_addr_t	lastdst, lastsrc;
3852 		in6_addr_t	nexthop;
3853 		in_port_t	lastport;
3854 
3855 		lastsrc = connp->conn_v6lastsrc;
3856 		lastdst = connp->conn_v6lastdst;
3857 		lastport = connp->conn_lastdstport;
3858 		ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &lastdst, &nexthop);
3859 		mutex_exit(&connp->conn_lock);
3860 
3861 		error = ip_attr_connect(connp, ixa, &lastsrc, &lastdst,
3862 		    &nexthop, lastport, NULL, NULL, IPDF_ALLOW_MCBC |
3863 		    IPDF_VERIFY_DST | (do_ipsec ? IPDF_IPSEC : 0));
3864 		switch (error) {
3865 		case 0:
3866 			break;
3867 		case EADDRNOTAVAIL:
3868 			/*
3869 			 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3870 			 * Don't have the application see that errno
3871 			 */
3872 			error = ENETUNREACH;
3873 			goto failed;
3874 		case ENETDOWN:
3875 			/*
3876 			 * Have !ipif_addr_ready address; drop packet silently
3877 			 * until we can get applications to not send until we
3878 			 * are ready.
3879 			 */
3880 			error = 0;
3881 			goto failed;
3882 		case EHOSTUNREACH:
3883 		case ENETUNREACH:
3884 			if (ixa->ixa_ire != NULL) {
3885 				/*
3886 				 * Let conn_ip_output/ire_send_noroute return
3887 				 * the error and send any local ICMP error.
3888 				 */
3889 				error = 0;
3890 				break;
3891 			}
3892 			/* FALLTHRU */
3893 		default:
3894 		failed:
3895 			ixa->ixa_cred = connp->conn_cred;	/* Restore */
3896 			ixa->ixa_cpid = connp->conn_cpid;
3897 			ixa_refrele(ixa);
3898 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
3899 			freemsg(mp);
3900 			return (error);
3901 		}
3902 	} else {
3903 		/* Done with conn_t */
3904 		mutex_exit(&connp->conn_lock);
3905 	}
3906 
3907 	/* We're done.  Pass the packet to ip. */
3908 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
3909 	error = conn_ip_output(mp, ixa);
3910 	/* No rawipOutErrors if an error since IP increases its error counter */
3911 	switch (error) {
3912 	case 0:
3913 		break;
3914 	case EWOULDBLOCK:
3915 		(void) ixa_check_drain_insert(connp, ixa);
3916 		error = 0;
3917 		break;
3918 	case EADDRNOTAVAIL:
3919 		/*
3920 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
3921 		 * Don't have the application see that errno
3922 		 */
3923 		error = ENETUNREACH;
3924 		/* FALLTHRU */
3925 	default:
3926 		mutex_enter(&connp->conn_lock);
3927 		/*
3928 		 * Clear the source and v6lastdst so we call ip_attr_connect
3929 		 * for the next packet and try to pick a better source.
3930 		 */
3931 		if (connp->conn_mcbc_bind)
3932 			connp->conn_saddr_v6 = ipv6_all_zeros;
3933 		else
3934 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
3935 		connp->conn_v6lastdst = ipv6_all_zeros;
3936 		mutex_exit(&connp->conn_lock);
3937 		break;
3938 	}
3939 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
3940 	ixa->ixa_cpid = connp->conn_cpid;
3941 	ixa_refrele(ixa);
3942 	return (error);
3943 }
3944 
3945 
3946 /*
3947  * Prepend the header template and then fill in the source and
3948  * flowinfo. The caller needs to handle the destination address since
3949  * it's setting is different if rthdr or source route.
3950  *
3951  * Returns NULL is allocation failed or if the packet would exceed IP_MAXPACKET.
3952  * When it returns NULL it sets errorp.
3953  */
3954 static mblk_t *
3955 icmp_prepend_header_template(conn_t *connp, ip_xmit_attr_t *ixa, mblk_t *mp,
3956     const in6_addr_t *v6src, uint32_t flowinfo, int *errorp)
3957 {
3958 	icmp_t		*icmp = connp->conn_icmp;
3959 	icmp_stack_t	*is = icmp->icmp_is;
3960 	uint_t		pktlen;
3961 	uint_t		copylen;
3962 	uint8_t		*iph;
3963 	uint_t		ip_hdr_length;
3964 	uint32_t	cksum;
3965 	ip_pkt_t	*ipp;
3966 
3967 	ASSERT(MUTEX_HELD(&connp->conn_lock));
3968 
3969 	/*
3970 	 * Copy the header template.
3971 	 */
3972 	copylen = connp->conn_ht_iphc_len;
3973 	pktlen = copylen + msgdsize(mp);
3974 	if (pktlen > IP_MAXPACKET) {
3975 		freemsg(mp);
3976 		*errorp = EMSGSIZE;
3977 		return (NULL);
3978 	}
3979 	ixa->ixa_pktlen = pktlen;
3980 
3981 	/* check/fix buffer config, setup pointers into it */
3982 	iph = mp->b_rptr - copylen;
3983 	if (DB_REF(mp) != 1 || iph < DB_BASE(mp) || !OK_32PTR(iph)) {
3984 		mblk_t *mp1;
3985 
3986 		mp1 = allocb(copylen + is->is_wroff_extra, BPRI_MED);
3987 		if (mp1 == NULL) {
3988 			freemsg(mp);
3989 			*errorp = ENOMEM;
3990 			return (NULL);
3991 		}
3992 		mp1->b_wptr = DB_LIM(mp1);
3993 		mp1->b_cont = mp;
3994 		mp = mp1;
3995 		iph = (mp->b_wptr - copylen);
3996 	}
3997 	mp->b_rptr = iph;
3998 	bcopy(connp->conn_ht_iphc, iph, copylen);
3999 	ip_hdr_length = (uint_t)(connp->conn_ht_ulp - connp->conn_ht_iphc);
4000 
4001 	ixa->ixa_ip_hdr_length = ip_hdr_length;
4002 
4003 	/*
4004 	 * Prepare for ICMPv6 checksum done in IP.
4005 	 *
4006 	 * icmp_build_hdr_template has already massaged any routing header
4007 	 * and placed the result in conn_sum.
4008 	 *
4009 	 * We make it easy for IP to include our pseudo header
4010 	 * by putting our length (and any routing header adjustment)
4011 	 * in the ICMPv6 checksum field.
4012 	 */
4013 	cksum = pktlen - ip_hdr_length;
4014 
4015 	cksum += connp->conn_sum;
4016 	cksum = (cksum >> 16) + (cksum & 0xFFFF);
4017 	ASSERT(cksum < 0x10000);
4018 
4019 	ipp = &connp->conn_xmit_ipp;
4020 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
4021 		ipha_t	*ipha = (ipha_t *)iph;
4022 
4023 		ipha->ipha_length = htons((uint16_t)pktlen);
4024 
4025 		/* if IP_PKTINFO specified an addres it wins over bind() */
4026 		if ((ipp->ipp_fields & IPPF_ADDR) &&
4027 		    IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
4028 			ASSERT(ipp->ipp_addr_v4 != INADDR_ANY);
4029 			ipha->ipha_src = ipp->ipp_addr_v4;
4030 		} else {
4031 			IN6_V4MAPPED_TO_IPADDR(v6src, ipha->ipha_src);
4032 		}
4033 	} else {
4034 		ip6_t *ip6h = (ip6_t *)iph;
4035 		uint_t	cksum_offset = 0;
4036 
4037 		ip6h->ip6_plen =  htons((uint16_t)(pktlen - IPV6_HDR_LEN));
4038 
4039 		/* if IP_PKTINFO specified an addres it wins over bind() */
4040 		if ((ipp->ipp_fields & IPPF_ADDR) &&
4041 		    !IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr)) {
4042 			ASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ipp->ipp_addr));
4043 			ip6h->ip6_src = ipp->ipp_addr;
4044 		} else {
4045 			ip6h->ip6_src = *v6src;
4046 		}
4047 		ip6h->ip6_vcf =
4048 		    (IPV6_DEFAULT_VERS_AND_FLOW & IPV6_VERS_AND_FLOW_MASK) |
4049 		    (flowinfo & ~IPV6_VERS_AND_FLOW_MASK);
4050 		if (ipp->ipp_fields & IPPF_TCLASS) {
4051 			/* Overrides the class part of flowinfo */
4052 			ip6h->ip6_vcf = IPV6_TCLASS_FLOW(ip6h->ip6_vcf,
4053 			    ipp->ipp_tclass);
4054 		}
4055 
4056 		if (ixa->ixa_flags & IXAF_SET_ULP_CKSUM) {
4057 			if (connp->conn_proto == IPPROTO_ICMPV6) {
4058 				cksum_offset = ixa->ixa_ip_hdr_length +
4059 				    offsetof(icmp6_t, icmp6_cksum);
4060 			} else if (ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
4061 				cksum_offset = ixa->ixa_ip_hdr_length +
4062 				    ixa->ixa_raw_cksum_offset;
4063 			}
4064 		}
4065 		if (cksum_offset != 0) {
4066 			uint16_t *ptr;
4067 
4068 			/* Make sure the checksum fits in the first mblk */
4069 			if (cksum_offset + sizeof (short) > MBLKL(mp)) {
4070 				mblk_t *mp1;
4071 
4072 				mp1 = msgpullup(mp,
4073 				    cksum_offset + sizeof (short));
4074 				freemsg(mp);
4075 				if (mp1 == NULL) {
4076 					*errorp = ENOMEM;
4077 					return (NULL);
4078 				}
4079 				mp = mp1;
4080 				iph = mp->b_rptr;
4081 				ip6h = (ip6_t *)iph;
4082 			}
4083 			ptr = (uint16_t *)(mp->b_rptr + cksum_offset);
4084 			*ptr = htons(cksum);
4085 		}
4086 	}
4087 
4088 	return (mp);
4089 }
4090 
4091 /*
4092  * This routine handles all messages passed downstream.  It either
4093  * consumes the message or passes it downstream; it never queues a
4094  * a message.
4095  */
4096 void
4097 icmp_wput(queue_t *q, mblk_t *mp)
4098 {
4099 	sin6_t		*sin6;
4100 	sin_t		*sin = NULL;
4101 	uint_t		srcid;
4102 	conn_t		*connp = Q_TO_CONN(q);
4103 	icmp_t		*icmp = connp->conn_icmp;
4104 	int		error = 0;
4105 	struct sockaddr	*addr = NULL;
4106 	socklen_t	addrlen;
4107 	icmp_stack_t	*is = icmp->icmp_is;
4108 	struct T_unitdata_req *tudr;
4109 	mblk_t		*data_mp;
4110 	cred_t		*cr;
4111 	pid_t		pid;
4112 
4113 	/*
4114 	 * We directly handle several cases here: T_UNITDATA_REQ message
4115 	 * coming down as M_PROTO/M_PCPROTO and M_DATA messages for connected
4116 	 * socket.
4117 	 */
4118 	switch (DB_TYPE(mp)) {
4119 	case M_DATA:
4120 		/* sockfs never sends down M_DATA */
4121 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4122 		freemsg(mp);
4123 		return;
4124 
4125 	case M_PROTO:
4126 	case M_PCPROTO:
4127 		tudr = (struct T_unitdata_req *)mp->b_rptr;
4128 		if (MBLKL(mp) < sizeof (*tudr) ||
4129 		    ((t_primp_t)mp->b_rptr)->type != T_UNITDATA_REQ) {
4130 			icmp_wput_other(q, mp);
4131 			return;
4132 		}
4133 		break;
4134 
4135 	default:
4136 		icmp_wput_other(q, mp);
4137 		return;
4138 	}
4139 
4140 	/* Handle valid T_UNITDATA_REQ here */
4141 	data_mp = mp->b_cont;
4142 	if (data_mp == NULL) {
4143 		error = EPROTO;
4144 		goto ud_error2;
4145 	}
4146 	mp->b_cont = NULL;
4147 
4148 	if (!MBLKIN(mp, 0, tudr->DEST_offset + tudr->DEST_length)) {
4149 		error = EADDRNOTAVAIL;
4150 		goto ud_error2;
4151 	}
4152 
4153 	/*
4154 	 * All Solaris components should pass a db_credp
4155 	 * for this message, hence we ASSERT.
4156 	 * On production kernels we return an error to be robust against
4157 	 * random streams modules sitting on top of us.
4158 	 */
4159 	cr = msg_getcred(mp, &pid);
4160 	ASSERT(cr != NULL);
4161 	if (cr == NULL) {
4162 		error = EINVAL;
4163 		goto ud_error2;
4164 	}
4165 
4166 	/*
4167 	 * If a port has not been bound to the stream, fail.
4168 	 * This is not a problem when sockfs is directly
4169 	 * above us, because it will ensure that the socket
4170 	 * is first bound before allowing data to be sent.
4171 	 */
4172 	if (icmp->icmp_state == TS_UNBND) {
4173 		error = EPROTO;
4174 		goto ud_error2;
4175 	}
4176 	addr = (struct sockaddr *)&mp->b_rptr[tudr->DEST_offset];
4177 	addrlen = tudr->DEST_length;
4178 
4179 	switch (connp->conn_family) {
4180 	case AF_INET6:
4181 		sin6 = (sin6_t *)addr;
4182 		if (!OK_32PTR((char *)sin6) || (addrlen != sizeof (sin6_t)) ||
4183 		    (sin6->sin6_family != AF_INET6)) {
4184 			error = EADDRNOTAVAIL;
4185 			goto ud_error2;
4186 		}
4187 
4188 		/* No support for mapped addresses on raw sockets */
4189 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
4190 			error = EADDRNOTAVAIL;
4191 			goto ud_error2;
4192 		}
4193 		srcid = sin6->__sin6_src_id;
4194 
4195 		/*
4196 		 * If the local address is a mapped address return
4197 		 * an error.
4198 		 * It would be possible to send an IPv6 packet but the
4199 		 * response would never make it back to the application
4200 		 * since it is bound to a mapped address.
4201 		 */
4202 		if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
4203 			error = EADDRNOTAVAIL;
4204 			goto ud_error2;
4205 		}
4206 
4207 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
4208 			sin6->sin6_addr = ipv6_loopback;
4209 
4210 		if (tudr->OPT_length != 0) {
4211 			/*
4212 			 * If we are connected then the destination needs to be
4213 			 * the same as the connected one.
4214 			 */
4215 			if (icmp->icmp_state == TS_DATA_XFER &&
4216 			    !conn_same_as_last_v6(connp, sin6)) {
4217 				error = EISCONN;
4218 				goto ud_error2;
4219 			}
4220 			error = icmp_output_ancillary(connp, NULL, sin6,
4221 			    data_mp, mp, NULL, cr, pid);
4222 		} else {
4223 			ip_xmit_attr_t *ixa;
4224 
4225 			/*
4226 			 * We have to allocate an ip_xmit_attr_t before we grab
4227 			 * conn_lock and we need to hold conn_lock once we've
4228 			 * checked conn_same_as_last_v6 to handle concurrent
4229 			 * send* calls on a socket.
4230 			 */
4231 			ixa = conn_get_ixa(connp, B_FALSE);
4232 			if (ixa == NULL) {
4233 				error = ENOMEM;
4234 				goto ud_error2;
4235 			}
4236 			mutex_enter(&connp->conn_lock);
4237 
4238 			if (conn_same_as_last_v6(connp, sin6) &&
4239 			    connp->conn_lastsrcid == srcid &&
4240 			    ipsec_outbound_policy_current(ixa)) {
4241 				/* icmp_output_lastdst drops conn_lock */
4242 				error = icmp_output_lastdst(connp, data_mp, cr,
4243 				    pid, ixa);
4244 			} else {
4245 				/* icmp_output_newdst drops conn_lock */
4246 				error = icmp_output_newdst(connp, data_mp, NULL,
4247 				    sin6, cr, pid, ixa);
4248 			}
4249 			ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
4250 		}
4251 		if (error == 0) {
4252 			freeb(mp);
4253 			return;
4254 		}
4255 		break;
4256 
4257 	case AF_INET:
4258 		sin = (sin_t *)addr;
4259 		if ((!OK_32PTR((char *)sin) || addrlen != sizeof (sin_t)) ||
4260 		    (sin->sin_family != AF_INET)) {
4261 			error = EADDRNOTAVAIL;
4262 			goto ud_error2;
4263 		}
4264 		if (sin->sin_addr.s_addr == INADDR_ANY)
4265 			sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
4266 
4267 		/* Protocol 255 contains full IP headers */
4268 		/* Read without holding lock */
4269 		if (icmp->icmp_hdrincl) {
4270 			if (MBLKL(data_mp) < IP_SIMPLE_HDR_LENGTH) {
4271 				if (!pullupmsg(data_mp, IP_SIMPLE_HDR_LENGTH)) {
4272 					error = EINVAL;
4273 					goto ud_error2;
4274 				}
4275 			}
4276 			error = icmp_output_hdrincl(connp, data_mp, cr, pid);
4277 			if (error == 0) {
4278 				freeb(mp);
4279 				return;
4280 			}
4281 			/* data_mp consumed above */
4282 			data_mp = NULL;
4283 			goto ud_error2;
4284 		}
4285 
4286 		if (tudr->OPT_length != 0) {
4287 			/*
4288 			 * If we are connected then the destination needs to be
4289 			 * the same as the connected one.
4290 			 */
4291 			if (icmp->icmp_state == TS_DATA_XFER &&
4292 			    !conn_same_as_last_v4(connp, sin)) {
4293 				error = EISCONN;
4294 				goto ud_error2;
4295 			}
4296 			error = icmp_output_ancillary(connp, sin, NULL,
4297 			    data_mp, mp, NULL, cr, pid);
4298 		} else {
4299 			ip_xmit_attr_t *ixa;
4300 
4301 			/*
4302 			 * We have to allocate an ip_xmit_attr_t before we grab
4303 			 * conn_lock and we need to hold conn_lock once we've
4304 			 * checked conn_same_as_last_v4 to handle concurrent
4305 			 * send* calls on a socket.
4306 			 */
4307 			ixa = conn_get_ixa(connp, B_FALSE);
4308 			if (ixa == NULL) {
4309 				error = ENOMEM;
4310 				goto ud_error2;
4311 			}
4312 			mutex_enter(&connp->conn_lock);
4313 
4314 			if (conn_same_as_last_v4(connp, sin) &&
4315 			    ipsec_outbound_policy_current(ixa)) {
4316 				/* icmp_output_lastdst drops conn_lock */
4317 				error = icmp_output_lastdst(connp, data_mp, cr,
4318 				    pid, ixa);
4319 			} else {
4320 				/* icmp_output_newdst drops conn_lock */
4321 				error = icmp_output_newdst(connp, data_mp, sin,
4322 				    NULL, cr, pid, ixa);
4323 			}
4324 			ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
4325 		}
4326 		if (error == 0) {
4327 			freeb(mp);
4328 			return;
4329 		}
4330 		break;
4331 	}
4332 	ASSERT(mp != NULL);
4333 	/* mp is freed by the following routine */
4334 	icmp_ud_err(q, mp, (t_scalar_t)error);
4335 	return;
4336 
4337 ud_error2:
4338 	BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4339 	freemsg(data_mp);
4340 	ASSERT(mp != NULL);
4341 	/* mp is freed by the following routine */
4342 	icmp_ud_err(q, mp, (t_scalar_t)error);
4343 }
4344 
4345 /*
4346  * Handle the case of the IP address or flow label being different
4347  * for both IPv4 and IPv6.
4348  *
4349  * NOTE: The caller must hold conn_lock and we drop it here.
4350  */
4351 static int
4352 icmp_output_newdst(conn_t *connp, mblk_t *data_mp, sin_t *sin, sin6_t *sin6,
4353     cred_t *cr, pid_t pid, ip_xmit_attr_t *ixa)
4354 {
4355 	icmp_t		*icmp = connp->conn_icmp;
4356 	icmp_stack_t	*is = icmp->icmp_is;
4357 	int		error;
4358 	ip_xmit_attr_t	*oldixa;
4359 	boolean_t	do_ipsec;
4360 	uint_t		srcid;
4361 	uint32_t	flowinfo;
4362 	in6_addr_t	v6src;
4363 	in6_addr_t	v6dst;
4364 	in6_addr_t	v6nexthop;
4365 	in_port_t	dstport;
4366 
4367 	ASSERT(MUTEX_HELD(&connp->conn_lock));
4368 	ASSERT(ixa != NULL);
4369 
4370 	/*
4371 	 * We hold conn_lock across all the use and modifications of
4372 	 * the conn_lastdst, conn_ixa, and conn_xmit_ipp to ensure that they
4373 	 * stay consistent.
4374 	 */
4375 
4376 	ASSERT(cr != NULL);
4377 	ixa->ixa_cred = cr;
4378 	ixa->ixa_cpid = pid;
4379 	if (is_system_labeled()) {
4380 		/* We need to restart with a label based on the cred */
4381 		ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
4382 	}
4383 	/*
4384 	 * If we are connected then the destination needs to be the
4385 	 * same as the connected one, which is not the case here since we
4386 	 * checked for that above.
4387 	 */
4388 	if (icmp->icmp_state == TS_DATA_XFER) {
4389 		mutex_exit(&connp->conn_lock);
4390 		error = EISCONN;
4391 		goto ud_error;
4392 	}
4393 
4394 	/* In case previous destination was multicast or multirt */
4395 	ip_attr_newdst(ixa);
4396 
4397 	/*
4398 	 * If laddr is unspecified then we look at sin6_src_id.
4399 	 * We will give precedence to a source address set with IPV6_PKTINFO
4400 	 * (aka IPPF_ADDR) but that is handled in build_hdrs. However, we don't
4401 	 * want ip_attr_connect to select a source (since it can fail) when
4402 	 * IPV6_PKTINFO is specified.
4403 	 * If this doesn't result in a source address then we get a source
4404 	 * from ip_attr_connect() below.
4405 	 */
4406 	v6src = connp->conn_saddr_v6;
4407 	if (sin != NULL) {
4408 		IN6_IPADDR_TO_V4MAPPED(sin->sin_addr.s_addr, &v6dst);
4409 		dstport = sin->sin_port;
4410 		flowinfo = 0;
4411 		srcid = 0;
4412 		ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
4413 		if (srcid != 0 && V4_PART_OF_V6(&v6src) == INADDR_ANY) {
4414 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
4415 			    connp->conn_netstack);
4416 		}
4417 		ixa->ixa_flags |= IXAF_IS_IPV4;
4418 	} else {
4419 		v6dst = sin6->sin6_addr;
4420 		dstport = sin6->sin6_port;
4421 		flowinfo = sin6->sin6_flowinfo;
4422 		srcid = sin6->__sin6_src_id;
4423 		if (IN6_IS_ADDR_LINKSCOPE(&v6dst) && sin6->sin6_scope_id != 0) {
4424 			ixa->ixa_scopeid = sin6->sin6_scope_id;
4425 			ixa->ixa_flags |= IXAF_SCOPEID_SET;
4426 		} else {
4427 			ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
4428 		}
4429 		if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
4430 			ip_srcid_find_id(srcid, &v6src, IPCL_ZONEID(connp),
4431 			    connp->conn_netstack);
4432 		}
4433 		if (IN6_IS_ADDR_V4MAPPED(&v6dst))
4434 			ixa->ixa_flags |= IXAF_IS_IPV4;
4435 		else
4436 			ixa->ixa_flags &= ~IXAF_IS_IPV4;
4437 	}
4438 	/* Handle IPV6_PKTINFO setting source address. */
4439 	if (IN6_IS_ADDR_UNSPECIFIED(&v6src) &&
4440 	    (connp->conn_xmit_ipp.ipp_fields & IPPF_ADDR)) {
4441 		ip_pkt_t *ipp = &connp->conn_xmit_ipp;
4442 
4443 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
4444 			if (IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
4445 				v6src = ipp->ipp_addr;
4446 		} else {
4447 			if (!IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
4448 				v6src = ipp->ipp_addr;
4449 		}
4450 	}
4451 
4452 	/* Defer IPsec if it might need to look at ICMP type/code */
4453 	switch (ixa->ixa_protocol) {
4454 	case IPPROTO_ICMP:
4455 	case IPPROTO_ICMPV6:
4456 		do_ipsec = B_FALSE;
4457 		break;
4458 	default:
4459 		do_ipsec = B_TRUE;
4460 	}
4461 
4462 	ip_attr_nexthop(&connp->conn_xmit_ipp, ixa, &v6dst, &v6nexthop);
4463 	mutex_exit(&connp->conn_lock);
4464 
4465 	error = ip_attr_connect(connp, ixa, &v6src, &v6dst, &v6nexthop, dstport,
4466 	    &v6src, NULL, IPDF_ALLOW_MCBC | IPDF_VERIFY_DST |
4467 	    (do_ipsec ? IPDF_IPSEC : 0));
4468 	switch (error) {
4469 	case 0:
4470 		break;
4471 	case EADDRNOTAVAIL:
4472 		/*
4473 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
4474 		 * Don't have the application see that errno
4475 		 */
4476 		error = ENETUNREACH;
4477 		goto failed;
4478 	case ENETDOWN:
4479 		/*
4480 		 * Have !ipif_addr_ready address; drop packet silently
4481 		 * until we can get applications to not send until we
4482 		 * are ready.
4483 		 */
4484 		error = 0;
4485 		goto failed;
4486 	case EHOSTUNREACH:
4487 	case ENETUNREACH:
4488 		if (ixa->ixa_ire != NULL) {
4489 			/*
4490 			 * Let conn_ip_output/ire_send_noroute return
4491 			 * the error and send any local ICMP error.
4492 			 */
4493 			error = 0;
4494 			break;
4495 		}
4496 		/* FALLTHRU */
4497 	default:
4498 	failed:
4499 		goto ud_error;
4500 	}
4501 
4502 	mutex_enter(&connp->conn_lock);
4503 	/*
4504 	 * While we dropped the lock some other thread might have connected
4505 	 * this socket. If so we bail out with EISCONN to ensure that the
4506 	 * connecting thread is the one that updates conn_ixa, conn_ht_*
4507 	 * and conn_*last*.
4508 	 */
4509 	if (icmp->icmp_state == TS_DATA_XFER) {
4510 		mutex_exit(&connp->conn_lock);
4511 		error = EISCONN;
4512 		goto ud_error;
4513 	}
4514 
4515 	/*
4516 	 * We need to rebuild the headers if
4517 	 *  - we are labeling packets (could be different for different
4518 	 *    destinations)
4519 	 *  - we have a source route (or routing header) since we need to
4520 	 *    massage that to get the pseudo-header checksum
4521 	 *  - a socket option with COA_HEADER_CHANGED has been set which
4522 	 *    set conn_v6lastdst to zero.
4523 	 *
4524 	 * Otherwise the prepend function will just update the src, dst,
4525 	 * and flow label.
4526 	 */
4527 	if (is_system_labeled()) {
4528 		/* TX MLP requires SCM_UCRED and don't have that here */
4529 		if (connp->conn_mlp_type != mlptSingle) {
4530 			mutex_exit(&connp->conn_lock);
4531 			error = ECONNREFUSED;
4532 			goto ud_error;
4533 		}
4534 		/*
4535 		 * Check whether Trusted Solaris policy allows communication
4536 		 * with this host, and pretend that the destination is
4537 		 * unreachable if not.
4538 		 * Compute any needed label and place it in ipp_label_v4/v6.
4539 		 *
4540 		 * Later conn_build_hdr_template/conn_prepend_hdr takes
4541 		 * ipp_label_v4/v6 to form the packet.
4542 		 *
4543 		 * Tsol note: Since we hold conn_lock we know no other
4544 		 * thread manipulates conn_xmit_ipp.
4545 		 */
4546 		error = conn_update_label(connp, ixa, &v6dst,
4547 		    &connp->conn_xmit_ipp);
4548 		if (error != 0) {
4549 			mutex_exit(&connp->conn_lock);
4550 			goto ud_error;
4551 		}
4552 		/* Rebuild the header template */
4553 		error = icmp_build_hdr_template(connp, &v6src, &v6dst,
4554 		    flowinfo);
4555 		if (error != 0) {
4556 			mutex_exit(&connp->conn_lock);
4557 			goto ud_error;
4558 		}
4559 	} else if (connp->conn_xmit_ipp.ipp_fields &
4560 	    (IPPF_IPV4_OPTIONS|IPPF_RTHDR) ||
4561 	    IN6_IS_ADDR_UNSPECIFIED(&connp->conn_v6lastdst)) {
4562 		/* Rebuild the header template */
4563 		error = icmp_build_hdr_template(connp, &v6src, &v6dst,
4564 		    flowinfo);
4565 		if (error != 0) {
4566 			mutex_exit(&connp->conn_lock);
4567 			goto ud_error;
4568 		}
4569 	} else {
4570 		/* Simply update the destination address if no source route */
4571 		if (ixa->ixa_flags & IXAF_IS_IPV4) {
4572 			ipha_t	*ipha = (ipha_t *)connp->conn_ht_iphc;
4573 
4574 			IN6_V4MAPPED_TO_IPADDR(&v6dst, ipha->ipha_dst);
4575 			if (ixa->ixa_flags & IXAF_PMTU_IPV4_DF) {
4576 				ipha->ipha_fragment_offset_and_flags |=
4577 				    IPH_DF_HTONS;
4578 			} else {
4579 				ipha->ipha_fragment_offset_and_flags &=
4580 				    ~IPH_DF_HTONS;
4581 			}
4582 		} else {
4583 			ip6_t *ip6h = (ip6_t *)connp->conn_ht_iphc;
4584 			ip6h->ip6_dst = v6dst;
4585 		}
4586 	}
4587 
4588 	/*
4589 	 * Remember the dst etc which corresponds to the built header
4590 	 * template and conn_ixa.
4591 	 */
4592 	oldixa = conn_replace_ixa(connp, ixa);
4593 	connp->conn_v6lastdst = v6dst;
4594 	connp->conn_lastflowinfo = flowinfo;
4595 	connp->conn_lastscopeid = ixa->ixa_scopeid;
4596 	connp->conn_lastsrcid = srcid;
4597 	/* Also remember a source to use together with lastdst */
4598 	connp->conn_v6lastsrc = v6src;
4599 
4600 	data_mp = icmp_prepend_header_template(connp, ixa, data_mp, &v6src,
4601 	    flowinfo, &error);
4602 
4603 	/* Done with conn_t */
4604 	mutex_exit(&connp->conn_lock);
4605 	ixa_refrele(oldixa);
4606 
4607 	if (data_mp == NULL) {
4608 		ASSERT(error != 0);
4609 		goto ud_error;
4610 	}
4611 
4612 	if (!do_ipsec) {
4613 		/* Policy might differ for different ICMP type/code */
4614 		data_mp = icmp_output_attach_policy(data_mp, connp, ixa);
4615 		if (data_mp == NULL) {
4616 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4617 			error = EHOSTUNREACH;	/* IPsec policy failure */
4618 			goto done;
4619 		}
4620 	}
4621 
4622 	/* We're done.  Pass the packet to ip. */
4623 	BUMP_MIB(&is->is_rawip_mib, rawipOutDatagrams);
4624 
4625 	error = conn_ip_output(data_mp, ixa);
4626 	/* No rawipOutErrors if an error since IP increases its error counter */
4627 	switch (error) {
4628 	case 0:
4629 		break;
4630 	case EWOULDBLOCK:
4631 		(void) ixa_check_drain_insert(connp, ixa);
4632 		error = 0;
4633 		break;
4634 	case EADDRNOTAVAIL:
4635 		/*
4636 		 * IXAF_VERIFY_SOURCE tells us to pick a better source.
4637 		 * Don't have the application see that errno
4638 		 */
4639 		error = ENETUNREACH;
4640 		/* FALLTHRU */
4641 	default:
4642 		mutex_enter(&connp->conn_lock);
4643 		/*
4644 		 * Clear the source and v6lastdst so we call ip_attr_connect
4645 		 * for the next packet and try to pick a better source.
4646 		 */
4647 		if (connp->conn_mcbc_bind)
4648 			connp->conn_saddr_v6 = ipv6_all_zeros;
4649 		else
4650 			connp->conn_saddr_v6 = connp->conn_bound_addr_v6;
4651 		connp->conn_v6lastdst = ipv6_all_zeros;
4652 		mutex_exit(&connp->conn_lock);
4653 		break;
4654 	}
4655 done:
4656 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
4657 	ixa->ixa_cpid = connp->conn_cpid;
4658 	ixa_refrele(ixa);
4659 	return (error);
4660 
4661 ud_error:
4662 	ixa->ixa_cred = connp->conn_cred;	/* Restore */
4663 	ixa->ixa_cpid = connp->conn_cpid;
4664 	ixa_refrele(ixa);
4665 
4666 	BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
4667 	freemsg(data_mp);
4668 	return (error);
4669 }
4670 
4671 /* ARGSUSED */
4672 static void
4673 icmp_wput_fallback(queue_t *q, mblk_t *mp)
4674 {
4675 #ifdef DEBUG
4676 	cmn_err(CE_CONT, "icmp_wput_fallback: Message during fallback \n");
4677 #endif
4678 	freemsg(mp);
4679 }
4680 
4681 static void
4682 icmp_wput_other(queue_t *q, mblk_t *mp)
4683 {
4684 	uchar_t	*rptr = mp->b_rptr;
4685 	struct iocblk *iocp;
4686 	conn_t	*connp = Q_TO_CONN(q);
4687 	icmp_t	*icmp = connp->conn_icmp;
4688 	icmp_stack_t *is = icmp->icmp_is;
4689 	cred_t *cr;
4690 
4691 	switch (mp->b_datap->db_type) {
4692 	case M_PROTO:
4693 	case M_PCPROTO:
4694 		if (mp->b_wptr - rptr < sizeof (t_scalar_t)) {
4695 			/*
4696 			 * If the message does not contain a PRIM_type,
4697 			 * throw it away.
4698 			 */
4699 			freemsg(mp);
4700 			return;
4701 		}
4702 		switch (((t_primp_t)rptr)->type) {
4703 		case T_ADDR_REQ:
4704 			icmp_addr_req(q, mp);
4705 			return;
4706 		case O_T_BIND_REQ:
4707 		case T_BIND_REQ:
4708 			icmp_tpi_bind(q, mp);
4709 			return;
4710 		case T_CONN_REQ:
4711 			icmp_tpi_connect(q, mp);
4712 			return;
4713 		case T_CAPABILITY_REQ:
4714 			icmp_capability_req(q, mp);
4715 			return;
4716 		case T_INFO_REQ:
4717 			icmp_info_req(q, mp);
4718 			return;
4719 		case T_UNITDATA_REQ:
4720 			/*
4721 			 * If a T_UNITDATA_REQ gets here, the address must
4722 			 * be bad.  Valid T_UNITDATA_REQs are handled
4723 			 * in icmp_wput.
4724 			 */
4725 			icmp_ud_err(q, mp, EADDRNOTAVAIL);
4726 			return;
4727 		case T_UNBIND_REQ:
4728 			icmp_tpi_unbind(q, mp);
4729 			return;
4730 		case T_SVR4_OPTMGMT_REQ:
4731 			/*
4732 			 * All Solaris components should pass a db_credp
4733 			 * for this TPI message, hence we ASSERT.
4734 			 * But in case there is some other M_PROTO that looks
4735 			 * like a TPI message sent by some other kernel
4736 			 * component, we check and return an error.
4737 			 */
4738 			cr = msg_getcred(mp, NULL);
4739 			ASSERT(cr != NULL);
4740 			if (cr == NULL) {
4741 				icmp_err_ack(q, mp, TSYSERR, EINVAL);
4742 				return;
4743 			}
4744 
4745 			if (!snmpcom_req(q, mp, icmp_snmp_set, ip_snmp_get,
4746 			    cr)) {
4747 				svr4_optcom_req(q, mp, cr, &icmp_opt_obj);
4748 			}
4749 			return;
4750 
4751 		case T_OPTMGMT_REQ:
4752 			/*
4753 			 * All Solaris components should pass a db_credp
4754 			 * for this TPI message, hence we ASSERT.
4755 			 * But in case there is some other M_PROTO that looks
4756 			 * like a TPI message sent by some other kernel
4757 			 * component, we check and return an error.
4758 			 */
4759 			cr = msg_getcred(mp, NULL);
4760 			ASSERT(cr != NULL);
4761 			if (cr == NULL) {
4762 				icmp_err_ack(q, mp, TSYSERR, EINVAL);
4763 				return;
4764 			}
4765 			tpi_optcom_req(q, mp, cr, &icmp_opt_obj);
4766 			return;
4767 
4768 		case T_DISCON_REQ:
4769 			icmp_tpi_disconnect(q, mp);
4770 			return;
4771 
4772 		/* The following TPI message is not supported by icmp. */
4773 		case O_T_CONN_RES:
4774 		case T_CONN_RES:
4775 			icmp_err_ack(q, mp, TNOTSUPPORT, 0);
4776 			return;
4777 
4778 		/* The following 3 TPI requests are illegal for icmp. */
4779 		case T_DATA_REQ:
4780 		case T_EXDATA_REQ:
4781 		case T_ORDREL_REQ:
4782 			icmp_err_ack(q, mp, TNOTSUPPORT, 0);
4783 			return;
4784 		default:
4785 			break;
4786 		}
4787 		break;
4788 	case M_FLUSH:
4789 		if (*rptr & FLUSHW)
4790 			flushq(q, FLUSHDATA);
4791 		break;
4792 	case M_IOCTL:
4793 		iocp = (struct iocblk *)mp->b_rptr;
4794 		switch (iocp->ioc_cmd) {
4795 		case TI_GETPEERNAME:
4796 			if (icmp->icmp_state != TS_DATA_XFER) {
4797 				/*
4798 				 * If a default destination address has not
4799 				 * been associated with the stream, then we
4800 				 * don't know the peer's name.
4801 				 */
4802 				iocp->ioc_error = ENOTCONN;
4803 				iocp->ioc_count = 0;
4804 				mp->b_datap->db_type = M_IOCACK;
4805 				qreply(q, mp);
4806 				return;
4807 			}
4808 			/* FALLTHRU */
4809 		case TI_GETMYNAME:
4810 			/*
4811 			 * For TI_GETPEERNAME and TI_GETMYNAME, we first
4812 			 * need to copyin the user's strbuf structure.
4813 			 * Processing will continue in the M_IOCDATA case
4814 			 * below.
4815 			 */
4816 			mi_copyin(q, mp, NULL,
4817 			    SIZEOF_STRUCT(strbuf, iocp->ioc_flag));
4818 			return;
4819 		case ND_SET:
4820 			/* nd_getset performs the necessary checking */
4821 		case ND_GET:
4822 			if (nd_getset(q, is->is_nd, mp)) {
4823 				qreply(q, mp);
4824 				return;
4825 			}
4826 			break;
4827 		default:
4828 			break;
4829 		}
4830 		break;
4831 	case M_IOCDATA:
4832 		icmp_wput_iocdata(q, mp);
4833 		return;
4834 	default:
4835 		/* Unrecognized messages are passed through without change. */
4836 		break;
4837 	}
4838 	ip_wput_nondata(q, mp);
4839 }
4840 
4841 /*
4842  * icmp_wput_iocdata is called by icmp_wput_other to handle all M_IOCDATA
4843  * messages.
4844  */
4845 static void
4846 icmp_wput_iocdata(queue_t *q, mblk_t *mp)
4847 {
4848 	mblk_t		*mp1;
4849 	STRUCT_HANDLE(strbuf, sb);
4850 	uint_t		addrlen;
4851 	conn_t		*connp = Q_TO_CONN(q);
4852 	icmp_t		*icmp = connp->conn_icmp;
4853 
4854 	/* Make sure it is one of ours. */
4855 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4856 	case TI_GETMYNAME:
4857 	case TI_GETPEERNAME:
4858 		break;
4859 	default:
4860 		ip_wput_nondata(q, mp);
4861 		return;
4862 	}
4863 
4864 	switch (mi_copy_state(q, mp, &mp1)) {
4865 	case -1:
4866 		return;
4867 	case MI_COPY_CASE(MI_COPY_IN, 1):
4868 		break;
4869 	case MI_COPY_CASE(MI_COPY_OUT, 1):
4870 		/*
4871 		 * The address has been copied out, so now
4872 		 * copyout the strbuf.
4873 		 */
4874 		mi_copyout(q, mp);
4875 		return;
4876 	case MI_COPY_CASE(MI_COPY_OUT, 2):
4877 		/*
4878 		 * The address and strbuf have been copied out.
4879 		 * We're done, so just acknowledge the original
4880 		 * M_IOCTL.
4881 		 */
4882 		mi_copy_done(q, mp, 0);
4883 		return;
4884 	default:
4885 		/*
4886 		 * Something strange has happened, so acknowledge
4887 		 * the original M_IOCTL with an EPROTO error.
4888 		 */
4889 		mi_copy_done(q, mp, EPROTO);
4890 		return;
4891 	}
4892 
4893 	/*
4894 	 * Now we have the strbuf structure for TI_GETMYNAME
4895 	 * and TI_GETPEERNAME.  Next we copyout the requested
4896 	 * address and then we'll copyout the strbuf.
4897 	 */
4898 	STRUCT_SET_HANDLE(sb, ((struct iocblk *)mp->b_rptr)->ioc_flag,
4899 	    (void *)mp1->b_rptr);
4900 
4901 	if (connp->conn_family == AF_INET)
4902 		addrlen = sizeof (sin_t);
4903 	else
4904 		addrlen = sizeof (sin6_t);
4905 
4906 	if (STRUCT_FGET(sb, maxlen) < addrlen) {
4907 		mi_copy_done(q, mp, EINVAL);
4908 		return;
4909 	}
4910 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4911 	case TI_GETMYNAME:
4912 		break;
4913 	case TI_GETPEERNAME:
4914 		if (icmp->icmp_state != TS_DATA_XFER) {
4915 			mi_copy_done(q, mp, ENOTCONN);
4916 			return;
4917 		}
4918 		break;
4919 	default:
4920 		mi_copy_done(q, mp, EPROTO);
4921 		return;
4922 	}
4923 	mp1 = mi_copyout_alloc(q, mp, STRUCT_FGETP(sb, buf), addrlen, B_TRUE);
4924 	if (!mp1)
4925 		return;
4926 
4927 	STRUCT_FSET(sb, len, addrlen);
4928 	switch (((struct iocblk *)mp->b_rptr)->ioc_cmd) {
4929 	case TI_GETMYNAME:
4930 		(void) conn_getsockname(connp, (struct sockaddr *)mp1->b_wptr,
4931 		    &addrlen);
4932 		break;
4933 	case TI_GETPEERNAME:
4934 		(void) conn_getpeername(connp, (struct sockaddr *)mp1->b_wptr,
4935 		    &addrlen);
4936 		break;
4937 	}
4938 	mp1->b_wptr += addrlen;
4939 	/* Copy out the address */
4940 	mi_copyout(q, mp);
4941 }
4942 
4943 void
4944 icmp_ddi_g_init(void)
4945 {
4946 	icmp_max_optsize = optcom_max_optsize(icmp_opt_obj.odb_opt_des_arr,
4947 	    icmp_opt_obj.odb_opt_arr_cnt);
4948 
4949 	/*
4950 	 * We want to be informed each time a stack is created or
4951 	 * destroyed in the kernel, so we can maintain the
4952 	 * set of icmp_stack_t's.
4953 	 */
4954 	netstack_register(NS_ICMP, rawip_stack_init, NULL, rawip_stack_fini);
4955 }
4956 
4957 void
4958 icmp_ddi_g_destroy(void)
4959 {
4960 	netstack_unregister(NS_ICMP);
4961 }
4962 
4963 #define	INET_NAME	"ip"
4964 
4965 /*
4966  * Initialize the ICMP stack instance.
4967  */
4968 static void *
4969 rawip_stack_init(netstackid_t stackid, netstack_t *ns)
4970 {
4971 	icmp_stack_t	*is;
4972 	icmpparam_t	*pa;
4973 	int		error = 0;
4974 	major_t		major;
4975 
4976 	is = (icmp_stack_t *)kmem_zalloc(sizeof (*is), KM_SLEEP);
4977 	is->is_netstack = ns;
4978 
4979 	pa = (icmpparam_t *)kmem_alloc(sizeof (icmp_param_arr), KM_SLEEP);
4980 	is->is_param_arr = pa;
4981 	bcopy(icmp_param_arr, is->is_param_arr, sizeof (icmp_param_arr));
4982 
4983 	(void) icmp_param_register(&is->is_nd,
4984 	    is->is_param_arr, A_CNT(icmp_param_arr));
4985 	is->is_ksp = rawip_kstat_init(stackid);
4986 
4987 	major = mod_name_to_major(INET_NAME);
4988 	error = ldi_ident_from_major(major, &is->is_ldi_ident);
4989 	ASSERT(error == 0);
4990 	return (is);
4991 }
4992 
4993 /*
4994  * Free the ICMP stack instance.
4995  */
4996 static void
4997 rawip_stack_fini(netstackid_t stackid, void *arg)
4998 {
4999 	icmp_stack_t *is = (icmp_stack_t *)arg;
5000 
5001 	nd_free(&is->is_nd);
5002 	kmem_free(is->is_param_arr, sizeof (icmp_param_arr));
5003 	is->is_param_arr = NULL;
5004 
5005 	rawip_kstat_fini(stackid, is->is_ksp);
5006 	is->is_ksp = NULL;
5007 	ldi_ident_release(is->is_ldi_ident);
5008 	kmem_free(is, sizeof (*is));
5009 }
5010 
5011 static void *
5012 rawip_kstat_init(netstackid_t stackid) {
5013 	kstat_t	*ksp;
5014 
5015 	rawip_named_kstat_t template = {
5016 		{ "inDatagrams",	KSTAT_DATA_UINT32, 0 },
5017 		{ "inCksumErrs",	KSTAT_DATA_UINT32, 0 },
5018 		{ "inErrors",		KSTAT_DATA_UINT32, 0 },
5019 		{ "outDatagrams",	KSTAT_DATA_UINT32, 0 },
5020 		{ "outErrors",		KSTAT_DATA_UINT32, 0 },
5021 	};
5022 
5023 	ksp = kstat_create_netstack("icmp", 0, "rawip", "mib2",
5024 					KSTAT_TYPE_NAMED,
5025 					NUM_OF_FIELDS(rawip_named_kstat_t),
5026 					0, stackid);
5027 	if (ksp == NULL || ksp->ks_data == NULL)
5028 		return (NULL);
5029 
5030 	bcopy(&template, ksp->ks_data, sizeof (template));
5031 	ksp->ks_update = rawip_kstat_update;
5032 	ksp->ks_private = (void *)(uintptr_t)stackid;
5033 
5034 	kstat_install(ksp);
5035 	return (ksp);
5036 }
5037 
5038 static void
5039 rawip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
5040 {
5041 	if (ksp != NULL) {
5042 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
5043 		kstat_delete_netstack(ksp, stackid);
5044 	}
5045 }
5046 
5047 static int
5048 rawip_kstat_update(kstat_t *ksp, int rw)
5049 {
5050 	rawip_named_kstat_t *rawipkp;
5051 	netstackid_t	stackid = (netstackid_t)(uintptr_t)ksp->ks_private;
5052 	netstack_t	*ns;
5053 	icmp_stack_t	*is;
5054 
5055 	if ((ksp == NULL) || (ksp->ks_data == NULL))
5056 		return (EIO);
5057 
5058 	if (rw == KSTAT_WRITE)
5059 		return (EACCES);
5060 
5061 	rawipkp = (rawip_named_kstat_t *)ksp->ks_data;
5062 
5063 	ns = netstack_find_by_stackid(stackid);
5064 	if (ns == NULL)
5065 		return (-1);
5066 	is = ns->netstack_icmp;
5067 	if (is == NULL) {
5068 		netstack_rele(ns);
5069 		return (-1);
5070 	}
5071 	rawipkp->inDatagrams.value.ui32 =  is->is_rawip_mib.rawipInDatagrams;
5072 	rawipkp->inCksumErrs.value.ui32 =  is->is_rawip_mib.rawipInCksumErrs;
5073 	rawipkp->inErrors.value.ui32 =	   is->is_rawip_mib.rawipInErrors;
5074 	rawipkp->outDatagrams.value.ui32 = is->is_rawip_mib.rawipOutDatagrams;
5075 	rawipkp->outErrors.value.ui32 =	   is->is_rawip_mib.rawipOutErrors;
5076 	netstack_rele(ns);
5077 	return (0);
5078 }
5079 
5080 /* ARGSUSED */
5081 int
5082 rawip_accept(sock_lower_handle_t lproto_handle,
5083     sock_lower_handle_t eproto_handle, sock_upper_handle_t sock_handle,
5084     cred_t *cr)
5085 {
5086 	return (EOPNOTSUPP);
5087 }
5088 
5089 /* ARGSUSED */
5090 int
5091 rawip_bind(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5092     socklen_t len, cred_t *cr)
5093 {
5094 	conn_t  *connp = (conn_t *)proto_handle;
5095 	int	error;
5096 
5097 	/* All Solaris components should pass a cred for this operation. */
5098 	ASSERT(cr != NULL);
5099 
5100 	/* Binding to a NULL address really means unbind */
5101 	if (sa == NULL)
5102 		error = rawip_do_unbind(connp);
5103 	else
5104 		error = rawip_do_bind(connp, sa, len);
5105 
5106 	if (error < 0) {
5107 		if (error == -TOUTSTATE)
5108 			error = EINVAL;
5109 		else
5110 			error = proto_tlitosyserr(-error);
5111 	}
5112 	return (error);
5113 }
5114 
5115 static int
5116 rawip_implicit_bind(conn_t *connp)
5117 {
5118 	sin6_t sin6addr;
5119 	sin_t *sin;
5120 	sin6_t *sin6;
5121 	socklen_t len;
5122 	int error;
5123 
5124 	if (connp->conn_family == AF_INET) {
5125 		len = sizeof (struct sockaddr_in);
5126 		sin = (sin_t *)&sin6addr;
5127 		*sin = sin_null;
5128 		sin->sin_family = AF_INET;
5129 		sin->sin_addr.s_addr = INADDR_ANY;
5130 	} else {
5131 		ASSERT(connp->conn_family == AF_INET6);
5132 		len = sizeof (sin6_t);
5133 		sin6 = (sin6_t *)&sin6addr;
5134 		*sin6 = sin6_null;
5135 		sin6->sin6_family = AF_INET6;
5136 		V6_SET_ZERO(sin6->sin6_addr);
5137 	}
5138 
5139 	error = rawip_do_bind(connp, (struct sockaddr *)&sin6addr, len);
5140 
5141 	return ((error < 0) ? proto_tlitosyserr(-error) : error);
5142 }
5143 
5144 static int
5145 rawip_unbind(conn_t *connp)
5146 {
5147 	int error;
5148 
5149 	error = rawip_do_unbind(connp);
5150 	if (error < 0) {
5151 		error = proto_tlitosyserr(-error);
5152 	}
5153 	return (error);
5154 }
5155 
5156 /* ARGSUSED */
5157 int
5158 rawip_listen(sock_lower_handle_t proto_handle, int backlog, cred_t *cr)
5159 {
5160 	return (EOPNOTSUPP);
5161 }
5162 
5163 int
5164 rawip_connect(sock_lower_handle_t proto_handle, const struct sockaddr *sa,
5165     socklen_t len, sock_connid_t *id, cred_t *cr)
5166 {
5167 	conn_t	*connp = (conn_t *)proto_handle;
5168 	icmp_t *icmp = connp->conn_icmp;
5169 	int	error;
5170 	boolean_t did_bind = B_FALSE;
5171 	pid_t	pid = curproc->p_pid;
5172 
5173 	/* All Solaris components should pass a cred for this operation. */
5174 	ASSERT(cr != NULL);
5175 
5176 	if (sa == NULL) {
5177 		/*
5178 		 * Disconnect
5179 		 * Make sure we are connected
5180 		 */
5181 		if (icmp->icmp_state != TS_DATA_XFER)
5182 			return (EINVAL);
5183 
5184 		error = icmp_disconnect(connp);
5185 		return (error);
5186 	}
5187 
5188 	error = proto_verify_ip_addr(connp->conn_family, sa, len);
5189 	if (error != 0)
5190 		return (error);
5191 
5192 	/* do an implicit bind if necessary */
5193 	if (icmp->icmp_state == TS_UNBND) {
5194 		error = rawip_implicit_bind(connp);
5195 		/*
5196 		 * We could be racing with an actual bind, in which case
5197 		 * we would see EPROTO. We cross our fingers and try
5198 		 * to connect.
5199 		 */
5200 		if (!(error == 0 || error == EPROTO))
5201 			return (error);
5202 		did_bind = B_TRUE;
5203 	}
5204 
5205 	/*
5206 	 * set SO_DGRAM_ERRIND
5207 	 */
5208 	connp->conn_dgram_errind = B_TRUE;
5209 
5210 	error = rawip_do_connect(connp, sa, len, cr, pid);
5211 	if (error != 0 && did_bind) {
5212 		int unbind_err;
5213 
5214 		unbind_err = rawip_unbind(connp);
5215 		ASSERT(unbind_err == 0);
5216 	}
5217 
5218 	if (error == 0) {
5219 		*id = 0;
5220 		(*connp->conn_upcalls->su_connected)(connp->conn_upper_handle,
5221 		    0, NULL, -1);
5222 	} else if (error < 0) {
5223 		error = proto_tlitosyserr(-error);
5224 	}
5225 	return (error);
5226 }
5227 
5228 /* ARGSUSED2 */
5229 int
5230 rawip_fallback(sock_lower_handle_t proto_handle, queue_t *q,
5231     boolean_t direct_sockfs, so_proto_quiesced_cb_t quiesced_cb)
5232 {
5233 	conn_t  *connp = (conn_t *)proto_handle;
5234 	icmp_t	*icmp;
5235 	struct T_capability_ack tca;
5236 	struct sockaddr_in6 laddr, faddr;
5237 	socklen_t laddrlen, faddrlen;
5238 	short opts;
5239 	struct stroptions *stropt;
5240 	mblk_t *stropt_mp;
5241 	int error;
5242 
5243 	icmp = connp->conn_icmp;
5244 
5245 	stropt_mp = allocb_wait(sizeof (*stropt), BPRI_HI, STR_NOSIG, NULL);
5246 
5247 	/*
5248 	 * setup the fallback stream that was allocated
5249 	 */
5250 	connp->conn_dev = (dev_t)RD(q)->q_ptr;
5251 	connp->conn_minor_arena = WR(q)->q_ptr;
5252 
5253 	RD(q)->q_ptr = WR(q)->q_ptr = connp;
5254 
5255 	WR(q)->q_qinfo = &icmpwinit;
5256 
5257 	connp->conn_rq = RD(q);
5258 	connp->conn_wq = WR(q);
5259 
5260 	/* Notify stream head about options before sending up data */
5261 	stropt_mp->b_datap->db_type = M_SETOPTS;
5262 	stropt_mp->b_wptr += sizeof (*stropt);
5263 	stropt = (struct stroptions *)stropt_mp->b_rptr;
5264 	stropt->so_flags = SO_WROFF | SO_HIWAT;
5265 	stropt->so_wroff = connp->conn_wroff;
5266 	stropt->so_hiwat = connp->conn_rcvbuf;
5267 	putnext(RD(q), stropt_mp);
5268 
5269 	/*
5270 	 * free helper stream
5271 	 */
5272 	ip_free_helper_stream(connp);
5273 
5274 	/*
5275 	 * Collect the information needed to sync with the sonode
5276 	 */
5277 	icmp_do_capability_ack(icmp, &tca, TC1_INFO);
5278 
5279 	laddrlen = faddrlen = sizeof (sin6_t);
5280 	(void) rawip_getsockname((sock_lower_handle_t)connp,
5281 	    (struct sockaddr *)&laddr, &laddrlen, CRED());
5282 	error = rawip_getpeername((sock_lower_handle_t)connp,
5283 	    (struct sockaddr *)&faddr, &faddrlen, CRED());
5284 	if (error != 0)
5285 		faddrlen = 0;
5286 	opts = 0;
5287 	if (connp->conn_dgram_errind)
5288 		opts |= SO_DGRAM_ERRIND;
5289 	if (connp->conn_ixa->ixa_flags & IXAF_DONTROUTE)
5290 		opts |= SO_DONTROUTE;
5291 
5292 	(*quiesced_cb)(connp->conn_upper_handle, q, &tca,
5293 	    (struct sockaddr *)&laddr, laddrlen,
5294 	    (struct sockaddr *)&faddr, faddrlen, opts);
5295 
5296 	/*
5297 	 * Attempts to send data up during fallback will result in it being
5298 	 * queued in icmp_t. Now we push up any queued packets.
5299 	 */
5300 	mutex_enter(&icmp->icmp_recv_lock);
5301 	while (icmp->icmp_fallback_queue_head != NULL) {
5302 		mblk_t	*mp;
5303 
5304 		mp = icmp->icmp_fallback_queue_head;
5305 		icmp->icmp_fallback_queue_head = mp->b_next;
5306 		mp->b_next = NULL;
5307 		mutex_exit(&icmp->icmp_recv_lock);
5308 		putnext(RD(q), mp);
5309 		mutex_enter(&icmp->icmp_recv_lock);
5310 	}
5311 	icmp->icmp_fallback_queue_tail = icmp->icmp_fallback_queue_head;
5312 
5313 	/*
5314 	 * No longer a streams less socket
5315 	 */
5316 	mutex_enter(&connp->conn_lock);
5317 	connp->conn_flags &= ~IPCL_NONSTR;
5318 	mutex_exit(&connp->conn_lock);
5319 
5320 	mutex_exit(&icmp->icmp_recv_lock);
5321 
5322 	ASSERT(icmp->icmp_fallback_queue_head == NULL &&
5323 	    icmp->icmp_fallback_queue_tail == NULL);
5324 
5325 	ASSERT(connp->conn_ref >= 1);
5326 
5327 	return (0);
5328 }
5329 
5330 /* ARGSUSED2 */
5331 sock_lower_handle_t
5332 rawip_create(int family, int type, int proto, sock_downcalls_t **sock_downcalls,
5333     uint_t *smodep, int *errorp, int flags, cred_t *credp)
5334 {
5335 	conn_t *connp;
5336 
5337 	if (type != SOCK_RAW || (family != AF_INET && family != AF_INET6)) {
5338 		*errorp = EPROTONOSUPPORT;
5339 		return (NULL);
5340 	}
5341 
5342 	connp = rawip_do_open(family, credp, errorp, flags);
5343 	if (connp != NULL) {
5344 		connp->conn_flags |= IPCL_NONSTR;
5345 
5346 		mutex_enter(&connp->conn_lock);
5347 		connp->conn_state_flags &= ~CONN_INCIPIENT;
5348 		mutex_exit(&connp->conn_lock);
5349 		*sock_downcalls = &sock_rawip_downcalls;
5350 		*smodep = SM_ATOMIC;
5351 	} else {
5352 		ASSERT(*errorp != 0);
5353 	}
5354 
5355 	return ((sock_lower_handle_t)connp);
5356 }
5357 
5358 /* ARGSUSED3 */
5359 void
5360 rawip_activate(sock_lower_handle_t proto_handle,
5361     sock_upper_handle_t sock_handle, sock_upcalls_t *sock_upcalls, int flags,
5362     cred_t *cr)
5363 {
5364 	conn_t 			*connp = (conn_t *)proto_handle;
5365 	struct sock_proto_props sopp;
5366 
5367 	/* All Solaris components should pass a cred for this operation. */
5368 	ASSERT(cr != NULL);
5369 
5370 	connp->conn_upcalls = sock_upcalls;
5371 	connp->conn_upper_handle = sock_handle;
5372 
5373 	sopp.sopp_flags = SOCKOPT_WROFF | SOCKOPT_RCVHIWAT | SOCKOPT_RCVLOWAT |
5374 	    SOCKOPT_MAXBLK | SOCKOPT_MAXPSZ | SOCKOPT_MINPSZ;
5375 	sopp.sopp_wroff = connp->conn_wroff;
5376 	sopp.sopp_rxhiwat = connp->conn_rcvbuf;
5377 	sopp.sopp_rxlowat = connp->conn_rcvlowat;
5378 	sopp.sopp_maxblk = INFPSZ;
5379 	sopp.sopp_maxpsz = IP_MAXPACKET;
5380 	sopp.sopp_minpsz = (icmp_mod_info.mi_minpsz == 1) ? 0 :
5381 	    icmp_mod_info.mi_minpsz;
5382 
5383 	(*connp->conn_upcalls->su_set_proto_props)
5384 	    (connp->conn_upper_handle, &sopp);
5385 
5386 	icmp_bind_proto(connp->conn_icmp);
5387 }
5388 
5389 /* ARGSUSED3 */
5390 int
5391 rawip_getpeername(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5392     socklen_t *salenp, cred_t *cr)
5393 {
5394 	conn_t  *connp = (conn_t *)proto_handle;
5395 	icmp_t  *icmp = connp->conn_icmp;
5396 	int	error;
5397 
5398 	/* All Solaris components should pass a cred for this operation. */
5399 	ASSERT(cr != NULL);
5400 
5401 	mutex_enter(&connp->conn_lock);
5402 	if (icmp->icmp_state != TS_DATA_XFER)
5403 		error = ENOTCONN;
5404 	else
5405 		error = conn_getpeername(connp, sa, salenp);
5406 	mutex_exit(&connp->conn_lock);
5407 	return (error);
5408 }
5409 
5410 /* ARGSUSED3 */
5411 int
5412 rawip_getsockname(sock_lower_handle_t proto_handle, struct sockaddr *sa,
5413     socklen_t *salenp, cred_t *cr)
5414 {
5415 	conn_t  *connp = (conn_t *)proto_handle;
5416 	int	error;
5417 
5418 	/* All Solaris components should pass a cred for this operation. */
5419 	ASSERT(cr != NULL);
5420 
5421 	mutex_enter(&connp->conn_lock);
5422 	error = conn_getsockname(connp, sa, salenp);
5423 	mutex_exit(&connp->conn_lock);
5424 	return (error);
5425 }
5426 
5427 int
5428 rawip_setsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
5429     const void *optvalp, socklen_t optlen, cred_t *cr)
5430 {
5431 	conn_t	*connp = (conn_t *)proto_handle;
5432 	int error;
5433 
5434 	/* All Solaris components should pass a cred for this operation. */
5435 	ASSERT(cr != NULL);
5436 
5437 	error = proto_opt_check(level, option_name, optlen, NULL,
5438 	    icmp_opt_obj.odb_opt_des_arr,
5439 	    icmp_opt_obj.odb_opt_arr_cnt,
5440 	    B_TRUE, B_FALSE, cr);
5441 
5442 	if (error != 0) {
5443 		/*
5444 		 * option not recognized
5445 		 */
5446 		if (error < 0) {
5447 			error = proto_tlitosyserr(-error);
5448 		}
5449 		return (error);
5450 	}
5451 
5452 	error = icmp_opt_set(connp, SETFN_OPTCOM_NEGOTIATE, level,
5453 	    option_name, optlen, (uchar_t *)optvalp, (uint_t *)&optlen,
5454 	    (uchar_t *)optvalp, NULL, cr);
5455 
5456 	ASSERT(error >= 0);
5457 
5458 	return (error);
5459 }
5460 
5461 int
5462 rawip_getsockopt(sock_lower_handle_t proto_handle, int level, int option_name,
5463     void *optvalp, socklen_t *optlen, cred_t *cr)
5464 {
5465 	int		error;
5466 	conn_t		*connp = (conn_t *)proto_handle;
5467 	t_uscalar_t	max_optbuf_len;
5468 	void		*optvalp_buf;
5469 	int		len;
5470 
5471 	/* All Solaris components should pass a cred for this operation. */
5472 	ASSERT(cr != NULL);
5473 
5474 	error = proto_opt_check(level, option_name, *optlen, &max_optbuf_len,
5475 	    icmp_opt_obj.odb_opt_des_arr,
5476 	    icmp_opt_obj.odb_opt_arr_cnt,
5477 	    B_FALSE, B_TRUE, cr);
5478 
5479 	if (error != 0) {
5480 		if (error < 0) {
5481 			error = proto_tlitosyserr(-error);
5482 		}
5483 		return (error);
5484 	}
5485 
5486 	optvalp_buf = kmem_alloc(max_optbuf_len, KM_SLEEP);
5487 	len = icmp_opt_get(connp, level, option_name, optvalp_buf);
5488 	if (len == -1) {
5489 		kmem_free(optvalp_buf, max_optbuf_len);
5490 		return (EINVAL);
5491 	}
5492 
5493 	/*
5494 	 * update optlen and copy option value
5495 	 */
5496 	t_uscalar_t size = MIN(len, *optlen);
5497 
5498 	bcopy(optvalp_buf, optvalp, size);
5499 	bcopy(&size, optlen, sizeof (size));
5500 
5501 	kmem_free(optvalp_buf, max_optbuf_len);
5502 	return (0);
5503 }
5504 
5505 /* ARGSUSED1 */
5506 int
5507 rawip_close(sock_lower_handle_t proto_handle, int flags, cred_t *cr)
5508 {
5509 	conn_t	*connp = (conn_t *)proto_handle;
5510 
5511 	/* All Solaris components should pass a cred for this operation. */
5512 	ASSERT(cr != NULL);
5513 
5514 	(void) rawip_do_close(connp);
5515 	return (0);
5516 }
5517 
5518 /* ARGSUSED2 */
5519 int
5520 rawip_shutdown(sock_lower_handle_t proto_handle, int how, cred_t *cr)
5521 {
5522 	conn_t  *connp = (conn_t *)proto_handle;
5523 
5524 	/* All Solaris components should pass a cred for this operation. */
5525 	ASSERT(cr != NULL);
5526 
5527 	/* shut down the send side */
5528 	if (how != SHUT_RD)
5529 		(*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
5530 		    SOCK_OPCTL_SHUT_SEND, 0);
5531 	/* shut down the recv side */
5532 	if (how != SHUT_WR)
5533 		(*connp->conn_upcalls->su_opctl)(connp->conn_upper_handle,
5534 		    SOCK_OPCTL_SHUT_RECV, 0);
5535 	return (0);
5536 }
5537 
5538 void
5539 rawip_clr_flowctrl(sock_lower_handle_t proto_handle)
5540 {
5541 	conn_t  *connp = (conn_t *)proto_handle;
5542 	icmp_t	*icmp = connp->conn_icmp;
5543 
5544 	mutex_enter(&icmp->icmp_recv_lock);
5545 	connp->conn_flow_cntrld = B_FALSE;
5546 	mutex_exit(&icmp->icmp_recv_lock);
5547 }
5548 
5549 int
5550 rawip_ioctl(sock_lower_handle_t proto_handle, int cmd, intptr_t arg,
5551     int mode, int32_t *rvalp, cred_t *cr)
5552 {
5553 	conn_t  	*connp = (conn_t *)proto_handle;
5554 	int		error;
5555 
5556 	/* All Solaris components should pass a cred for this operation. */
5557 	ASSERT(cr != NULL);
5558 
5559 	/*
5560 	 * If we don't have a helper stream then create one.
5561 	 * ip_create_helper_stream takes care of locking the conn_t,
5562 	 * so this check for NULL is just a performance optimization.
5563 	 */
5564 	if (connp->conn_helper_info == NULL) {
5565 		icmp_stack_t *is = connp->conn_icmp->icmp_is;
5566 
5567 		ASSERT(is->is_ldi_ident != NULL);
5568 
5569 		/*
5570 		 * Create a helper stream for non-STREAMS socket.
5571 		 */
5572 		error = ip_create_helper_stream(connp, is->is_ldi_ident);
5573 		if (error != 0) {
5574 			ip0dbg(("rawip_ioctl: create of IP helper stream "
5575 			    "failed %d\n", error));
5576 			return (error);
5577 		}
5578 	}
5579 
5580 	switch (cmd) {
5581 	case ND_SET:
5582 	case ND_GET:
5583 	case _SIOCSOCKFALLBACK:
5584 	case TI_GETPEERNAME:
5585 	case TI_GETMYNAME:
5586 #ifdef DEBUG
5587 		cmn_err(CE_CONT, "icmp_ioctl cmd 0x%x on non streams"
5588 		    " socket", cmd);
5589 #endif
5590 		error = EINVAL;
5591 		break;
5592 	default:
5593 		/*
5594 		 * Pass on to IP using helper stream
5595 		 */
5596 		error = ldi_ioctl(connp->conn_helper_info->iphs_handle,
5597 		    cmd, arg, mode, cr, rvalp);
5598 		break;
5599 	}
5600 	return (error);
5601 }
5602 
5603 int
5604 rawip_send(sock_lower_handle_t proto_handle, mblk_t *mp, struct nmsghdr *msg,
5605     cred_t *cr)
5606 {
5607 	sin6_t		*sin6;
5608 	sin_t		*sin = NULL;
5609 	uint_t		srcid;
5610 	conn_t		*connp = (conn_t *)proto_handle;
5611 	icmp_t		*icmp = connp->conn_icmp;
5612 	int		error = 0;
5613 	icmp_stack_t	*is = icmp->icmp_is;
5614 	pid_t		pid = curproc->p_pid;
5615 	ip_xmit_attr_t	*ixa;
5616 
5617 	ASSERT(DB_TYPE(mp) == M_DATA);
5618 
5619 	/* All Solaris components should pass a cred for this operation. */
5620 	ASSERT(cr != NULL);
5621 
5622 	/* do an implicit bind if necessary */
5623 	if (icmp->icmp_state == TS_UNBND) {
5624 		error = rawip_implicit_bind(connp);
5625 		/*
5626 		 * We could be racing with an actual bind, in which case
5627 		 * we would see EPROTO. We cross our fingers and try
5628 		 * to connect.
5629 		 */
5630 		if (!(error == 0 || error == EPROTO)) {
5631 			freemsg(mp);
5632 			return (error);
5633 		}
5634 	}
5635 
5636 	/* Protocol 255 contains full IP headers */
5637 	/* Read without holding lock */
5638 	if (icmp->icmp_hdrincl) {
5639 		ASSERT(connp->conn_ipversion == IPV4_VERSION);
5640 		if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH) {
5641 			if (!pullupmsg(mp, IP_SIMPLE_HDR_LENGTH)) {
5642 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5643 				freemsg(mp);
5644 				return (EINVAL);
5645 			}
5646 		}
5647 		error = icmp_output_hdrincl(connp, mp, cr, pid);
5648 		if (is->is_sendto_ignerr)
5649 			return (0);
5650 		else
5651 			return (error);
5652 	}
5653 
5654 	/* Connected? */
5655 	if (msg->msg_name == NULL) {
5656 		if (icmp->icmp_state != TS_DATA_XFER) {
5657 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5658 			return (EDESTADDRREQ);
5659 		}
5660 		if (msg->msg_controllen != 0) {
5661 			error = icmp_output_ancillary(connp, NULL, NULL, mp,
5662 			    NULL, msg, cr, pid);
5663 		} else {
5664 			error = icmp_output_connected(connp, mp, cr, pid);
5665 		}
5666 		if (is->is_sendto_ignerr)
5667 			return (0);
5668 		else
5669 			return (error);
5670 	}
5671 	if (icmp->icmp_state == TS_DATA_XFER) {
5672 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5673 		return (EISCONN);
5674 	}
5675 	error = proto_verify_ip_addr(connp->conn_family,
5676 	    (struct sockaddr *)msg->msg_name, msg->msg_namelen);
5677 	if (error != 0) {
5678 		BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5679 		return (error);
5680 	}
5681 	switch (connp->conn_family) {
5682 	case AF_INET6:
5683 		sin6 = (sin6_t *)msg->msg_name;
5684 
5685 		/* No support for mapped addresses on raw sockets */
5686 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
5687 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5688 			return (EADDRNOTAVAIL);
5689 		}
5690 		srcid = sin6->__sin6_src_id;
5691 
5692 		/*
5693 		 * If the local address is a mapped address return
5694 		 * an error.
5695 		 * It would be possible to send an IPv6 packet but the
5696 		 * response would never make it back to the application
5697 		 * since it is bound to a mapped address.
5698 		 */
5699 		if (IN6_IS_ADDR_V4MAPPED(&connp->conn_saddr_v6)) {
5700 			BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5701 			return (EADDRNOTAVAIL);
5702 		}
5703 
5704 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
5705 			sin6->sin6_addr = ipv6_loopback;
5706 
5707 		/*
5708 		 * We have to allocate an ip_xmit_attr_t before we grab
5709 		 * conn_lock and we need to hold conn_lock once we've check
5710 		 * conn_same_as_last_v6 to handle concurrent send* calls on a
5711 		 * socket.
5712 		 */
5713 		if (msg->msg_controllen == 0) {
5714 			ixa = conn_get_ixa(connp, B_FALSE);
5715 			if (ixa == NULL) {
5716 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5717 				return (ENOMEM);
5718 			}
5719 		} else {
5720 			ixa = NULL;
5721 		}
5722 		mutex_enter(&connp->conn_lock);
5723 		if (icmp->icmp_delayed_error != 0) {
5724 			sin6_t  *sin2 = (sin6_t *)&icmp->icmp_delayed_addr;
5725 
5726 			error = icmp->icmp_delayed_error;
5727 			icmp->icmp_delayed_error = 0;
5728 
5729 			/* Compare IP address and family */
5730 
5731 			if (IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr,
5732 			    &sin2->sin6_addr) &&
5733 			    sin6->sin6_family == sin2->sin6_family) {
5734 				mutex_exit(&connp->conn_lock);
5735 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5736 				if (ixa != NULL)
5737 					ixa_refrele(ixa);
5738 				return (error);
5739 			}
5740 		}
5741 		if (msg->msg_controllen != 0) {
5742 			mutex_exit(&connp->conn_lock);
5743 			ASSERT(ixa == NULL);
5744 			error = icmp_output_ancillary(connp, NULL, sin6, mp,
5745 			    NULL, msg, cr, pid);
5746 		} else if (conn_same_as_last_v6(connp, sin6) &&
5747 		    connp->conn_lastsrcid == srcid &&
5748 		    ipsec_outbound_policy_current(ixa)) {
5749 			/* icmp_output_lastdst drops conn_lock */
5750 			error = icmp_output_lastdst(connp, mp, cr, pid, ixa);
5751 		} else {
5752 			/* icmp_output_newdst drops conn_lock */
5753 			error = icmp_output_newdst(connp, mp, NULL, sin6, cr,
5754 			    pid, ixa);
5755 		}
5756 		ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
5757 		if (is->is_sendto_ignerr)
5758 			return (0);
5759 		else
5760 			return (error);
5761 	case AF_INET:
5762 		sin = (sin_t *)msg->msg_name;
5763 
5764 		if (sin->sin_addr.s_addr == INADDR_ANY)
5765 			sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
5766 
5767 		/*
5768 		 * We have to allocate an ip_xmit_attr_t before we grab
5769 		 * conn_lock and we need to hold conn_lock once we've check
5770 		 * conn_same_as_last_v6 to handle concurrent send* on a socket.
5771 		 */
5772 		if (msg->msg_controllen == 0) {
5773 			ixa = conn_get_ixa(connp, B_FALSE);
5774 			if (ixa == NULL) {
5775 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5776 				return (ENOMEM);
5777 			}
5778 		} else {
5779 			ixa = NULL;
5780 		}
5781 		mutex_enter(&connp->conn_lock);
5782 		if (icmp->icmp_delayed_error != 0) {
5783 			sin_t  *sin2 = (sin_t *)&icmp->icmp_delayed_addr;
5784 
5785 			error = icmp->icmp_delayed_error;
5786 			icmp->icmp_delayed_error = 0;
5787 
5788 			/* Compare IP address */
5789 
5790 			if (sin->sin_addr.s_addr == sin2->sin_addr.s_addr) {
5791 				mutex_exit(&connp->conn_lock);
5792 				BUMP_MIB(&is->is_rawip_mib, rawipOutErrors);
5793 				if (ixa != NULL)
5794 					ixa_refrele(ixa);
5795 				return (error);
5796 			}
5797 		}
5798 
5799 		if (msg->msg_controllen != 0) {
5800 			mutex_exit(&connp->conn_lock);
5801 			ASSERT(ixa == NULL);
5802 			error = icmp_output_ancillary(connp, sin, NULL, mp,
5803 			    NULL, msg, cr, pid);
5804 		} else if (conn_same_as_last_v4(connp, sin) &&
5805 		    ipsec_outbound_policy_current(ixa)) {
5806 			/* icmp_output_lastdst drops conn_lock */
5807 			error = icmp_output_lastdst(connp, mp, cr, pid, ixa);
5808 		} else {
5809 			/* icmp_output_newdst drops conn_lock */
5810 			error = icmp_output_newdst(connp, mp, sin, NULL, cr,
5811 			    pid, ixa);
5812 		}
5813 		ASSERT(MUTEX_NOT_HELD(&connp->conn_lock));
5814 		if (is->is_sendto_ignerr)
5815 			return (0);
5816 		else
5817 			return (error);
5818 	default:
5819 		return (EINVAL);
5820 	}
5821 }
5822 
5823 sock_downcalls_t sock_rawip_downcalls = {
5824 	rawip_activate,
5825 	rawip_accept,
5826 	rawip_bind,
5827 	rawip_listen,
5828 	rawip_connect,
5829 	rawip_getpeername,
5830 	rawip_getsockname,
5831 	rawip_getsockopt,
5832 	rawip_setsockopt,
5833 	rawip_send,
5834 	NULL,
5835 	NULL,
5836 	NULL,
5837 	rawip_shutdown,
5838 	rawip_clr_flowctrl,
5839 	rawip_ioctl,
5840 	rawip_close
5841 };
5842