xref: /titanic_41/usr/src/uts/common/inet/ip/ip6_output.c (revision 44b099c4d944a196d124a02c7403ad891223139e)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 /* Copyright (c) 1990 Mentat Inc. */
27 
28 #include <sys/types.h>
29 #include <sys/stream.h>
30 #include <sys/strsubr.h>
31 #include <sys/dlpi.h>
32 #include <sys/strsun.h>
33 #include <sys/zone.h>
34 #include <sys/ddi.h>
35 #include <sys/sunddi.h>
36 #include <sys/cmn_err.h>
37 #include <sys/debug.h>
38 #include <sys/atomic.h>
39 
40 #include <sys/systm.h>
41 #include <sys/param.h>
42 #include <sys/kmem.h>
43 #include <sys/sdt.h>
44 #include <sys/socket.h>
45 #include <sys/mac.h>
46 #include <net/if.h>
47 #include <net/if_arp.h>
48 #include <net/route.h>
49 #include <sys/sockio.h>
50 #include <netinet/in.h>
51 #include <net/if_dl.h>
52 
53 #include <inet/common.h>
54 #include <inet/mi.h>
55 #include <inet/mib2.h>
56 #include <inet/nd.h>
57 #include <inet/arp.h>
58 #include <inet/snmpcom.h>
59 #include <inet/kstatcom.h>
60 
61 #include <netinet/igmp_var.h>
62 #include <netinet/ip6.h>
63 #include <netinet/icmp6.h>
64 #include <netinet/sctp.h>
65 
66 #include <inet/ip.h>
67 #include <inet/ip_impl.h>
68 #include <inet/ip6.h>
69 #include <inet/ip6_asp.h>
70 #include <inet/tcp.h>
71 #include <inet/ip_multi.h>
72 #include <inet/ip_if.h>
73 #include <inet/ip_ire.h>
74 #include <inet/ip_ftable.h>
75 #include <inet/ip_rts.h>
76 #include <inet/optcom.h>
77 #include <inet/ip_ndp.h>
78 #include <inet/ip_listutils.h>
79 #include <netinet/igmp.h>
80 #include <netinet/ip_mroute.h>
81 #include <inet/ipp_common.h>
82 
83 #include <net/pfkeyv2.h>
84 #include <inet/sadb.h>
85 #include <inet/ipsec_impl.h>
86 #include <inet/ipdrop.h>
87 #include <inet/ip_netinfo.h>
88 
89 #include <sys/pattr.h>
90 #include <inet/ipclassifier.h>
91 #include <inet/sctp_ip.h>
92 #include <inet/sctp/sctp_impl.h>
93 #include <inet/udp_impl.h>
94 #include <sys/sunddi.h>
95 
96 #include <sys/tsol/label.h>
97 #include <sys/tsol/tnet.h>
98 
99 #ifdef	DEBUG
100 extern boolean_t skip_sctp_cksum;
101 #endif
102 
103 int
ip_output_simple_v6(mblk_t * mp,ip_xmit_attr_t * ixa)104 ip_output_simple_v6(mblk_t *mp, ip_xmit_attr_t *ixa)
105 {
106 	ip6_t		*ip6h;
107 	in6_addr_t	firsthop; /* In IP header */
108 	in6_addr_t	dst;	/* End of source route, or ip6_dst if none */
109 	ire_t		*ire;
110 	in6_addr_t	setsrc;
111 	int		error;
112 	ill_t		*ill = NULL;
113 	dce_t		*dce = NULL;
114 	nce_t		*nce;
115 	iaflags_t	ixaflags = ixa->ixa_flags;
116 	ip_stack_t	*ipst = ixa->ixa_ipst;
117 	uint8_t		*nexthdrp;
118 	boolean_t	repeat = B_FALSE;
119 	boolean_t	multirt = B_FALSE;
120 	uint_t		ifindex;
121 	int64_t		now;
122 
123 	ip6h = (ip6_t *)mp->b_rptr;
124 	ASSERT(IPH_HDR_VERSION(ip6h) == IPV6_VERSION);
125 
126 	ASSERT(ixa->ixa_nce == NULL);
127 
128 	ixa->ixa_pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
129 	ASSERT(ixa->ixa_pktlen == msgdsize(mp));
130 	if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ixa->ixa_ip_hdr_length,
131 	    &nexthdrp)) {
132 		/* Malformed packet */
133 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsHCOutRequests);
134 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
135 		ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
136 		freemsg(mp);
137 		return (EINVAL);
138 	}
139 	ixa->ixa_protocol = *nexthdrp;
140 
141 	/*
142 	 * Assumes that source routed packets have already been massaged by
143 	 * the ULP (ip_massage_options_v6) and as a result ip6_dst is the next
144 	 * hop in the source route. The final destination is used for IPsec
145 	 * policy and DCE lookup.
146 	 */
147 	firsthop = ip6h->ip6_dst;
148 	dst = ip_get_dst_v6(ip6h, mp, NULL);
149 
150 repeat_ire:
151 	error = 0;
152 	setsrc = ipv6_all_zeros;
153 	ire = ip_select_route_v6(&firsthop, ip6h->ip6_src, ixa, NULL, &setsrc,
154 	    &error, &multirt);
155 	ASSERT(ire != NULL);	/* IRE_NOROUTE if none found */
156 	if (error != 0) {
157 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsHCOutRequests);
158 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
159 		ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
160 		freemsg(mp);
161 		goto done;
162 	}
163 
164 	if (ire->ire_flags & (RTF_BLACKHOLE|RTF_REJECT)) {
165 		/* ire_ill might be NULL hence need to skip some code */
166 		if (ixaflags & IXAF_SET_SOURCE)
167 			ip6h->ip6_src = ipv6_loopback;
168 		ixa->ixa_fragsize = IP_MAXPACKET;
169 		ire->ire_ob_pkt_count++;
170 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsHCOutRequests);
171 		/* No dce yet; use default one */
172 		error = (ire->ire_sendfn)(ire, mp, ip6h, ixa,
173 		    &ipst->ips_dce_default->dce_ident);
174 		goto done;
175 	}
176 
177 	/* Note that ip6_dst is only used for IRE_MULTICAST */
178 	nce = ire_to_nce(ire, INADDR_ANY, &ip6h->ip6_dst);
179 	if (nce == NULL) {
180 		/* Allocation failure? */
181 		ip_drop_output("ire_to_nce", mp, ill);
182 		freemsg(mp);
183 		error = ENOBUFS;
184 		goto done;
185 	}
186 	if (nce->nce_is_condemned) {
187 		nce_t *nce1;
188 
189 		nce1 = ire_handle_condemned_nce(nce, ire, NULL, ip6h, B_TRUE);
190 		nce_refrele(nce);
191 		if (nce1 == NULL) {
192 			if (!repeat) {
193 				/* Try finding a better IRE */
194 				repeat = B_TRUE;
195 				ire_refrele(ire);
196 				goto repeat_ire;
197 			}
198 			/* Tried twice - drop packet */
199 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
200 			ip_drop_output("No nce", mp, ill);
201 			freemsg(mp);
202 			error = ENOBUFS;
203 			goto done;
204 		}
205 		nce = nce1;
206 	}
207 	/*
208 	 * For multicast with multirt we have a flag passed back from
209 	 * ire_lookup_multi_ill_v6 since we don't have an IRE for each
210 	 * possible multicast address.
211 	 * We also need a flag for multicast since we can't check
212 	 * whether RTF_MULTIRT is set in ixa_ire for multicast.
213 	 */
214 	if (multirt) {
215 		ixa->ixa_postfragfn = ip_postfrag_multirt_v6;
216 		ixa->ixa_flags |= IXAF_MULTIRT_MULTICAST;
217 	} else {
218 		ixa->ixa_postfragfn = ire->ire_postfragfn;
219 		ixa->ixa_flags &= ~IXAF_MULTIRT_MULTICAST;
220 	}
221 	ASSERT(ixa->ixa_nce == NULL);
222 	ixa->ixa_nce = nce;
223 
224 	/*
225 	 * Check for a dce_t with a path mtu.
226 	 */
227 	ifindex = 0;
228 	if (IN6_IS_ADDR_LINKSCOPE(&dst))
229 		ifindex = nce->nce_common->ncec_ill->ill_phyint->phyint_ifindex;
230 
231 	dce = dce_lookup_v6(&dst, ifindex, ipst, NULL);
232 	ASSERT(dce != NULL);
233 
234 	if (!(ixaflags & IXAF_PMTU_DISCOVERY)) {
235 		ixa->ixa_fragsize = IPV6_MIN_MTU;
236 	} else if (dce->dce_flags & DCEF_PMTU) {
237 		/*
238 		 * To avoid a periodic timer to increase the path MTU we
239 		 * look at dce_last_change_time each time we send a packet.
240 		 */
241 		now = ddi_get_lbolt64();
242 		if (TICK_TO_SEC(now) - dce->dce_last_change_time >
243 		    ipst->ips_ip_pathmtu_interval) {
244 			/*
245 			 * Older than 20 minutes. Drop the path MTU information.
246 			 */
247 			mutex_enter(&dce->dce_lock);
248 			dce->dce_flags &= ~(DCEF_PMTU|DCEF_TOO_SMALL_PMTU);
249 			dce->dce_last_change_time = TICK_TO_SEC(now);
250 			mutex_exit(&dce->dce_lock);
251 			dce_increment_generation(dce);
252 			ixa->ixa_fragsize = ip_get_base_mtu(nce->nce_ill, ire);
253 		} else {
254 			uint_t fragsize;
255 
256 			fragsize = ip_get_base_mtu(nce->nce_ill, ire);
257 			if (fragsize > dce->dce_pmtu)
258 				fragsize = dce->dce_pmtu;
259 			ixa->ixa_fragsize = fragsize;
260 		}
261 	} else {
262 		ixa->ixa_fragsize = ip_get_base_mtu(nce->nce_ill, ire);
263 	}
264 
265 	/*
266 	 * We use use ire_nexthop_ill (and not ncec_ill) to avoid the under ipmp
267 	 * interface for source address selection.
268 	 */
269 	ill = ire_nexthop_ill(ire);
270 
271 	if (ixaflags & IXAF_SET_SOURCE) {
272 		in6_addr_t	src;
273 
274 		/*
275 		 * We use the final destination to get
276 		 * correct selection for source routed packets
277 		 */
278 
279 		/* If unreachable we have no ill but need some source */
280 		if (ill == NULL) {
281 			src = ipv6_loopback;
282 			error = 0;
283 		} else {
284 			error = ip_select_source_v6(ill, &setsrc, &dst,
285 			    ixa->ixa_zoneid, ipst, B_FALSE,
286 			    ixa->ixa_src_preferences, &src, NULL, NULL);
287 		}
288 		if (error != 0) {
289 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutRequests);
290 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
291 			ip_drop_output("ipIfStatsOutDiscards - no source",
292 			    mp, ill);
293 			freemsg(mp);
294 			goto done;
295 		}
296 		ip6h->ip6_src = src;
297 	} else if (ixaflags & IXAF_VERIFY_SOURCE) {
298 		/* Check if the IP source is assigned to the host. */
299 		if (!ip_verify_src(mp, ixa, NULL)) {
300 			/* Don't send a packet with a source that isn't ours */
301 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsHCOutRequests);
302 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
303 			ip_drop_output("ipIfStatsOutDiscards - invalid source",
304 			    mp, ill);
305 			freemsg(mp);
306 			error = EADDRNOTAVAIL;
307 			goto done;
308 		}
309 	}
310 
311 	/*
312 	 * Check against global IPsec policy to set the AH/ESP attributes.
313 	 * IPsec will set IXAF_IPSEC_* and ixa_ipsec_* as appropriate.
314 	 */
315 	if (!(ixaflags & (IXAF_NO_IPSEC|IXAF_IPSEC_SECURE))) {
316 		ASSERT(ixa->ixa_ipsec_policy == NULL);
317 		mp = ip_output_attach_policy(mp, NULL, ip6h, NULL, ixa);
318 		if (mp == NULL) {
319 			/* MIB and ip_drop_packet already done */
320 			return (EHOSTUNREACH);	/* IPsec policy failure */
321 		}
322 	}
323 
324 	if (ill != NULL) {
325 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutRequests);
326 	} else {
327 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsHCOutRequests);
328 	}
329 
330 	/*
331 	 * We update the statistics on the most specific IRE i.e., the first
332 	 * one we found.
333 	 * We don't have an IRE when we fragment, hence ire_ob_pkt_count
334 	 * can only count the use prior to fragmentation. However the MIB
335 	 * counters on the ill will be incremented in post fragmentation.
336 	 */
337 	ire->ire_ob_pkt_count++;
338 
339 	/*
340 	 * Based on ire_type and ire_flags call one of:
341 	 *	ire_send_local_v6 - for IRE_LOCAL and IRE_LOOPBACK
342 	 *	ire_send_multirt_v6 - if RTF_MULTIRT
343 	 *	ire_send_noroute_v6 - if RTF_REJECT or RTF_BLACHOLE
344 	 *	ire_send_multicast_v6 - for IRE_MULTICAST
345 	 *	ire_send_wire_v6 - for the rest.
346 	 */
347 	error = (ire->ire_sendfn)(ire, mp, ip6h, ixa, &dce->dce_ident);
348 done:
349 	ire_refrele(ire);
350 	if (dce != NULL)
351 		dce_refrele(dce);
352 	if (ill != NULL)
353 		ill_refrele(ill);
354 	if (ixa->ixa_nce != NULL)
355 		nce_refrele(ixa->ixa_nce);
356 	ixa->ixa_nce = NULL;
357 	return (error);
358 }
359 
360 /*
361  * ire_sendfn() functions.
362  * These functions use the following xmit_attr:
363  *  - ixa_fragsize - read to determine whether or not to fragment
364  *  - IXAF_IPSEC_SECURE - to determine whether or not to invoke IPsec
365  *  - ixa_ipsec_*  are used inside IPsec
366  *  - IXAF_LOOPBACK_COPY - for multicast
367  */
368 
369 
370 /*
371  * ire_sendfn for IRE_LOCAL and IRE_LOOPBACK
372  *
373  * The checks for restrict_interzone_loopback are done in ire_route_recursive.
374  */
375 /* ARGSUSED4 */
376 int
ire_send_local_v6(ire_t * ire,mblk_t * mp,void * iph_arg,ip_xmit_attr_t * ixa,uint32_t * identp)377 ire_send_local_v6(ire_t *ire, mblk_t *mp, void *iph_arg,
378     ip_xmit_attr_t *ixa, uint32_t *identp)
379 {
380 	ip6_t		*ip6h = (ip6_t *)iph_arg;
381 	ip_stack_t	*ipst = ixa->ixa_ipst;
382 	ill_t		*ill = ire->ire_ill;
383 	ip_recv_attr_t	iras;	/* NOTE: No bzero for performance */
384 	uint_t		pktlen = ixa->ixa_pktlen;
385 
386 	/*
387 	 * No fragmentation, no nce, and no application of IPsec.
388 	 *
389 	 *
390 	 * Note different order between IP provider and FW_HOOKS than in
391 	 * send_wire case.
392 	 */
393 
394 	/*
395 	 * DTrace this as ip:::send.  A packet blocked by FW_HOOKS will fire the
396 	 * send probe, but not the receive probe.
397 	 */
398 	DTRACE_IP7(send, mblk_t *, mp, conn_t *, NULL, void_ip_t *,
399 	    ip6h, __dtrace_ipsr_ill_t *, ill, ipha_t *, NULL, ip6_t *, ip6h,
400 	    int, 1);
401 
402 	DTRACE_PROBE4(ip6__loopback__out__start,
403 	    ill_t *, NULL, ill_t *, ill,
404 	    ip6_t *, ip6h, mblk_t *, mp);
405 
406 	if (HOOKS6_INTERESTED_LOOPBACK_OUT(ipst)) {
407 		int	error;
408 
409 		FW_HOOKS(ipst->ips_ip6_loopback_out_event,
410 		    ipst->ips_ipv6firewall_loopback_out,
411 		    NULL, ill, ip6h, mp, mp, 0, ipst, error);
412 
413 		DTRACE_PROBE1(ip6__loopback__out__end, mblk_t *, mp);
414 		if (mp == NULL)
415 			return (error);
416 
417 		/*
418 		 * Even if the destination was changed by the filter we use the
419 		 * forwarding decision that was made based on the address
420 		 * in ip_output/ip_set_destination.
421 		 */
422 		/* Length could be different */
423 		ip6h = (ip6_t *)mp->b_rptr;
424 		pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
425 	}
426 
427 	/*
428 	 * If a callback is enabled then we need to know the
429 	 * source and destination zoneids for the packet. We already
430 	 * have those handy.
431 	 */
432 	if (ipst->ips_ip6_observe.he_interested) {
433 		zoneid_t szone, dzone;
434 		zoneid_t stackzoneid;
435 
436 		stackzoneid = netstackid_to_zoneid(
437 		    ipst->ips_netstack->netstack_stackid);
438 
439 		if (stackzoneid == GLOBAL_ZONEID) {
440 			/* Shared-IP zone */
441 			dzone = ire->ire_zoneid;
442 			szone = ixa->ixa_zoneid;
443 		} else {
444 			szone = dzone = stackzoneid;
445 		}
446 		ipobs_hook(mp, IPOBS_HOOK_LOCAL, szone, dzone, ill, ipst);
447 	}
448 
449 	/* Handle lo0 stats */
450 	ipst->ips_loopback_packets++;
451 
452 	/*
453 	 * Update output mib stats. Note that we can't move into the icmp
454 	 * sender (icmp_output etc) since they don't know the ill and the
455 	 * stats are per ill.
456 	 */
457 	if (ixa->ixa_protocol == IPPROTO_ICMPV6) {
458 		icmp6_t		*icmp6;
459 
460 		icmp6 = (icmp6_t *)((uchar_t *)ip6h + ixa->ixa_ip_hdr_length);
461 		icmp_update_out_mib_v6(ill, icmp6);
462 	}
463 
464 	DTRACE_PROBE4(ip6__loopback__in__start,
465 	    ill_t *, ill, ill_t *, NULL,
466 	    ip6_t *, ip6h, mblk_t *, mp);
467 
468 	if (HOOKS6_INTERESTED_LOOPBACK_IN(ipst)) {
469 		int	error;
470 
471 		FW_HOOKS(ipst->ips_ip6_loopback_in_event,
472 		    ipst->ips_ipv6firewall_loopback_in,
473 		    ill, NULL, ip6h, mp, mp, 0, ipst, error);
474 
475 		DTRACE_PROBE1(ip6__loopback__in__end, mblk_t *, mp);
476 		if (mp == NULL)
477 			return (error);
478 
479 		/*
480 		 * Even if the destination was changed by the filter we use the
481 		 * forwarding decision that was made based on the address
482 		 * in ip_output/ip_set_destination.
483 		 */
484 		/* Length could be different */
485 		ip6h = (ip6_t *)mp->b_rptr;
486 		pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
487 	}
488 
489 	DTRACE_IP7(receive, mblk_t *, mp, conn_t *, NULL, void_ip_t *,
490 	    ip6h, __dtrace_ipsr_ill_t *, ill, ipha_t *, NULL, ip6_t *, ip6h,
491 	    int, 1);
492 
493 	/* Map ixa to ira including IPsec policies */
494 	ipsec_out_to_in(ixa, ill, &iras);
495 	iras.ira_pktlen = pktlen;
496 
497 	ire->ire_ib_pkt_count++;
498 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInReceives);
499 	UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCInOctets, pktlen);
500 
501 	/* Destined to ire_zoneid - use that for fanout */
502 	iras.ira_zoneid = ire->ire_zoneid;
503 
504 	if (is_system_labeled()) {
505 		iras.ira_flags |= IRAF_SYSTEM_LABELED;
506 
507 		/*
508 		 * This updates ira_cred, ira_tsl and ira_free_flags based
509 		 * on the label. We don't expect this to ever fail for
510 		 * loopback packets, so we silently drop the packet should it
511 		 * fail.
512 		 */
513 		if (!tsol_get_pkt_label(mp, IPV6_VERSION, &iras)) {
514 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
515 			ip_drop_input("tsol_get_pkt_label", mp, ill);
516 			freemsg(mp);
517 			return (0);
518 		}
519 		ASSERT(iras.ira_tsl != NULL);
520 
521 		/* tsol_get_pkt_label sometimes does pullupmsg */
522 		ip6h = (ip6_t *)mp->b_rptr;
523 	}
524 
525 	ip_fanout_v6(mp, ip6h, &iras);
526 
527 	/* We moved any IPsec refs from ixa to iras */
528 	ira_cleanup(&iras, B_FALSE);
529 	return (0);
530 }
531 
532 static void
multirt_check_v6(ire_t * ire,ip6_t * ip6h,ip_xmit_attr_t * ixa)533 multirt_check_v6(ire_t *ire, ip6_t *ip6h, ip_xmit_attr_t *ixa)
534 {
535 	ip_stack_t *ipst = ixa->ixa_ipst;
536 
537 	/* Limit the TTL on multirt packets. Do this even if IPV6_HOPLIMIT */
538 	if (ire->ire_type & IRE_MULTICAST) {
539 		if (ip6h->ip6_hops > 1) {
540 			ip2dbg(("ire_send_multirt_v6: forcing multicast "
541 			    "multirt TTL to 1 (was %d)\n", ip6h->ip6_hops));
542 			ip6h->ip6_hops = 1;
543 		}
544 		ixa->ixa_flags |= IXAF_NO_TTL_CHANGE;
545 	} else if ((ipst->ips_ip_multirt_ttl > 0) &&
546 	    (ip6h->ip6_hops > ipst->ips_ip_multirt_ttl)) {
547 		ip6h->ip6_hops = ipst->ips_ip_multirt_ttl;
548 		/*
549 		 * Need to ensure we don't increase the ttl should we go through
550 		 * ire_send_multicast.
551 		 */
552 		ixa->ixa_flags |= IXAF_NO_TTL_CHANGE;
553 	}
554 
555 	/* For IPv6 this also needs to insert a fragment header */
556 	ixa->ixa_flags |= IXAF_IPV6_ADD_FRAGHDR;
557 }
558 
559 /*
560  * ire_sendfn for IRE_MULTICAST
561  *
562  * Note that we do path MTU discovery by default for IPv6 multicast. But
563  * since unconnected UDP and RAW sockets don't set IXAF_PMTU_DISCOVERY
564  * only connected sockets get this by default.
565  */
566 int
ire_send_multicast_v6(ire_t * ire,mblk_t * mp,void * iph_arg,ip_xmit_attr_t * ixa,uint32_t * identp)567 ire_send_multicast_v6(ire_t *ire, mblk_t *mp, void *iph_arg,
568     ip_xmit_attr_t *ixa, uint32_t *identp)
569 {
570 	ip6_t		*ip6h = (ip6_t *)iph_arg;
571 	ip_stack_t	*ipst = ixa->ixa_ipst;
572 	ill_t		*ill = ire->ire_ill;
573 	iaflags_t	ixaflags = ixa->ixa_flags;
574 
575 	/*
576 	 * The IRE_MULTICAST is the same whether or not multirt is in use.
577 	 * Hence we need special-case code.
578 	 */
579 	if (ixaflags & IXAF_MULTIRT_MULTICAST)
580 		multirt_check_v6(ire, ip6h, ixa);
581 
582 	/*
583 	 * Check if anything in ip_input_v6 wants a copy of the transmitted
584 	 * packet (after IPsec and fragmentation)
585 	 *
586 	 * 1. Multicast routers always need a copy unless SO_DONTROUTE is set
587 	 *    RSVP and the rsvp daemon is an example of a
588 	 *    protocol and user level process that
589 	 *    handles it's own routing. Hence, it uses the
590 	 *    SO_DONTROUTE option to accomplish this.
591 	 * 2. If the sender has set IP_MULTICAST_LOOP, then we just
592 	 *    check whether there are any receivers for the group on the ill
593 	 *    (ignoring the zoneid).
594 	 * 3. If IP_MULTICAST_LOOP is not set, then we check if there are
595 	 *    any members in other shared-IP zones.
596 	 *    If such members exist, then we indicate that the sending zone
597 	 *    shouldn't get a loopback copy to preserve the IP_MULTICAST_LOOP
598 	 *    behavior.
599 	 *
600 	 * When we loopback we skip hardware checksum to make sure loopback
601 	 * copy is checksumed.
602 	 *
603 	 * Note that ire_ill is the upper in the case of IPMP.
604 	 */
605 	ixa->ixa_flags &= ~(IXAF_LOOPBACK_COPY | IXAF_NO_HW_CKSUM);
606 	if (ipst->ips_ip_g_mrouter && ill->ill_mrouter_cnt > 0 &&
607 	    !(ixaflags & IXAF_DONTROUTE)) {
608 		ixa->ixa_flags |= IXAF_LOOPBACK_COPY | IXAF_NO_HW_CKSUM;
609 	} else if (ixaflags & IXAF_MULTICAST_LOOP) {
610 		/*
611 		 * If this zone or any other zone has members then loopback
612 		 * a copy.
613 		 */
614 		if (ill_hasmembers_v6(ill, &ip6h->ip6_dst))
615 			ixa->ixa_flags |= IXAF_LOOPBACK_COPY | IXAF_NO_HW_CKSUM;
616 	} else if (ipst->ips_netstack->netstack_numzones > 1) {
617 		/*
618 		 * This zone should not have a copy. But there are some other
619 		 * zones which might have members.
620 		 */
621 		if (ill_hasmembers_otherzones_v6(ill, &ip6h->ip6_dst,
622 		    ixa->ixa_zoneid)) {
623 			ixa->ixa_flags |= IXAF_NO_LOOP_ZONEID_SET;
624 			ixa->ixa_no_loop_zoneid = ixa->ixa_zoneid;
625 			ixa->ixa_flags |= IXAF_LOOPBACK_COPY | IXAF_NO_HW_CKSUM;
626 		}
627 	}
628 
629 	/*
630 	 * Unless IPV6_HOPLIMIT or ire_send_multirt_v6 already set a ttl,
631 	 * force the ttl to the IP_MULTICAST_TTL value
632 	 */
633 	if (!(ixaflags & IXAF_NO_TTL_CHANGE)) {
634 		ip6h->ip6_hops = ixa->ixa_multicast_ttl;
635 	}
636 
637 	return (ire_send_wire_v6(ire, mp, ip6h, ixa, identp));
638 }
639 
640 /*
641  * ire_sendfn for IREs with RTF_MULTIRT
642  */
643 int
ire_send_multirt_v6(ire_t * ire,mblk_t * mp,void * iph_arg,ip_xmit_attr_t * ixa,uint32_t * identp)644 ire_send_multirt_v6(ire_t *ire, mblk_t *mp, void *iph_arg,
645     ip_xmit_attr_t *ixa, uint32_t *identp)
646 {
647 	ip6_t		*ip6h = (ip6_t *)iph_arg;
648 
649 	multirt_check_v6(ire, ip6h, ixa);
650 
651 	if (ire->ire_type & IRE_MULTICAST)
652 		return (ire_send_multicast_v6(ire, mp, ip6h, ixa, identp));
653 	else
654 		return (ire_send_wire_v6(ire, mp, ip6h, ixa, identp));
655 }
656 
657 /*
658  * ire_sendfn for IREs with RTF_REJECT/RTF_BLACKHOLE, including IRE_NOROUTE
659  */
660 /* ARGSUSED4 */
661 int
ire_send_noroute_v6(ire_t * ire,mblk_t * mp,void * iph_arg,ip_xmit_attr_t * ixa,uint32_t * identp)662 ire_send_noroute_v6(ire_t *ire, mblk_t *mp, void *iph_arg,
663     ip_xmit_attr_t *ixa, uint32_t *identp)
664 {
665 	ip6_t		*ip6h = (ip6_t *)iph_arg;
666 	ip_stack_t	*ipst = ixa->ixa_ipst;
667 	ill_t		*ill;
668 	ip_recv_attr_t	iras;
669 	boolean_t	dummy;
670 
671 	BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutNoRoutes);
672 
673 	if (ire->ire_type & IRE_NOROUTE) {
674 		/* A lack of a route as opposed to RTF_REJECT|BLACKHOLE */
675 		ip_rts_change_v6(RTM_MISS, &ip6h->ip6_dst, 0, 0, 0, 0, 0, 0,
676 		    RTA_DST, ipst);
677 	}
678 
679 	if (ire->ire_flags & RTF_BLACKHOLE) {
680 		ip_drop_output("ipIfStatsOutNoRoutes RTF_BLACKHOLE", mp, NULL);
681 		freemsg(mp);
682 		/* No error even for local senders - silent blackhole */
683 		return (0);
684 	}
685 	ip_drop_output("ipIfStatsOutNoRoutes RTF_REJECT", mp, NULL);
686 
687 	/*
688 	 * We need an ill_t for the ip_recv_attr_t even though this packet
689 	 * was never received and icmp_unreachable doesn't currently use
690 	 * ira_ill.
691 	 */
692 	ill = ill_lookup_on_name("lo0", B_FALSE,
693 	    !(ixa->ixa_flags & IRAF_IS_IPV4), &dummy, ipst);
694 	if (ill == NULL) {
695 		freemsg(mp);
696 		return (EHOSTUNREACH);
697 	}
698 
699 	bzero(&iras, sizeof (iras));
700 	/* Map ixa to ira including IPsec policies */
701 	ipsec_out_to_in(ixa, ill, &iras);
702 
703 	icmp_unreachable_v6(mp, ICMP6_DST_UNREACH_NOROUTE, B_FALSE, &iras);
704 	/* We moved any IPsec refs from ixa to iras */
705 	ira_cleanup(&iras, B_FALSE);
706 
707 	ill_refrele(ill);
708 	return (EHOSTUNREACH);
709 }
710 
711 /*
712  * Calculate a checksum ignoring any hardware capabilities
713  *
714  * Returns B_FALSE if the packet was too short for the checksum. Caller
715  * should free and do stats.
716  */
717 static boolean_t
ip_output_sw_cksum_v6(mblk_t * mp,ip6_t * ip6h,ip_xmit_attr_t * ixa)718 ip_output_sw_cksum_v6(mblk_t *mp, ip6_t *ip6h, ip_xmit_attr_t *ixa)
719 {
720 	ip_stack_t	*ipst = ixa->ixa_ipst;
721 	uint_t		pktlen = ixa->ixa_pktlen;
722 	uint16_t	*cksump;
723 	uint32_t	cksum;
724 	uint8_t		protocol = ixa->ixa_protocol;
725 	uint16_t	ip_hdr_length = ixa->ixa_ip_hdr_length;
726 
727 #define	iphs    ((uint16_t *)ip6h)
728 
729 	/* Just in case it contained garbage */
730 	DB_CKSUMFLAGS(mp) &= ~HCK_FLAGS;
731 
732 	/*
733 	 * Calculate ULP checksum
734 	 */
735 	if (protocol == IPPROTO_TCP) {
736 		cksump = IPH_TCPH_CHECKSUMP(ip6h, ip_hdr_length);
737 		cksum = IP_TCP_CSUM_COMP;
738 	} else if (protocol == IPPROTO_UDP) {
739 		cksump = IPH_UDPH_CHECKSUMP(ip6h, ip_hdr_length);
740 		cksum = IP_UDP_CSUM_COMP;
741 	} else if (protocol == IPPROTO_SCTP) {
742 		sctp_hdr_t	*sctph;
743 
744 		ASSERT(MBLKL(mp) >= (ip_hdr_length + sizeof (*sctph)));
745 		sctph = (sctp_hdr_t *)(mp->b_rptr + ip_hdr_length);
746 		/*
747 		 * Zero out the checksum field to ensure proper
748 		 * checksum calculation.
749 		 */
750 		sctph->sh_chksum = 0;
751 #ifdef	DEBUG
752 		if (!skip_sctp_cksum)
753 #endif
754 			sctph->sh_chksum = sctp_cksum(mp, ip_hdr_length);
755 		return (B_TRUE);
756 	} else if (ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
757 		/*
758 		 * icmp has placed length and routing
759 		 * header adjustment in the checksum field.
760 		 */
761 		cksump = (uint16_t *)(((uint8_t *)ip6h) + ip_hdr_length +
762 		    ixa->ixa_raw_cksum_offset);
763 		cksum = htons(protocol);
764 	} else if (protocol == IPPROTO_ICMPV6) {
765 		cksump = IPH_ICMPV6_CHECKSUMP(ip6h, ip_hdr_length);
766 		cksum = IP_ICMPV6_CSUM_COMP;	/* Pseudo-header cksum */
767 	} else {
768 		return (B_TRUE);
769 	}
770 
771 	/* ULP puts the checksum field is in the first mblk */
772 	ASSERT(((uchar_t *)cksump) + sizeof (uint16_t) <= mp->b_wptr);
773 
774 	/*
775 	 * We accumulate the pseudo header checksum in cksum.
776 	 * This is pretty hairy code, so watch close.  One
777 	 * thing to keep in mind is that UDP and TCP have
778 	 * stored their respective datagram lengths in their
779 	 * checksum fields.  This lines things up real nice.
780 	 */
781 	cksum += iphs[4] + iphs[5] + iphs[6] + iphs[7] +
782 	    iphs[8] + iphs[9] + iphs[10] + iphs[11] +
783 	    iphs[12] + iphs[13] + iphs[14] + iphs[15] +
784 	    iphs[16] + iphs[17] + iphs[18] + iphs[19];
785 	cksum = IP_CSUM(mp, ip_hdr_length, cksum);
786 
787 	/*
788 	 * For UDP/IPv6 a zero UDP checksum is not allowed.
789 	 * Change to 0xffff
790 	 */
791 	if (protocol == IPPROTO_UDP && cksum == 0)
792 		*cksump = ~cksum;
793 	else
794 		*cksump = cksum;
795 
796 	IP6_STAT(ipst, ip6_out_sw_cksum);
797 	IP6_STAT_UPDATE(ipst, ip6_out_sw_cksum_bytes, pktlen);
798 
799 	/* No IP header checksum for IPv6 */
800 
801 	return (B_TRUE);
802 #undef	iphs
803 }
804 
805 /* There are drivers that can't do partial checksum for ICMPv6 */
806 int nxge_cksum_workaround = 1;
807 
808 /*
809  * Calculate the ULP checksum - try to use hardware.
810  * In the case of MULTIRT or multicast the
811  * IXAF_NO_HW_CKSUM is set in which case we use software.
812  *
813  * Returns B_FALSE if the packet was too short for the checksum. Caller
814  * should free and do stats.
815  */
816 static boolean_t
ip_output_cksum_v6(iaflags_t ixaflags,mblk_t * mp,ip6_t * ip6h,ip_xmit_attr_t * ixa,ill_t * ill)817 ip_output_cksum_v6(iaflags_t ixaflags, mblk_t *mp, ip6_t *ip6h,
818     ip_xmit_attr_t *ixa, ill_t *ill)
819 {
820 	uint_t		pktlen = ixa->ixa_pktlen;
821 	uint16_t	*cksump;
822 	uint16_t	hck_flags;
823 	uint32_t	cksum;
824 	uint8_t		protocol = ixa->ixa_protocol;
825 	uint16_t	ip_hdr_length = ixa->ixa_ip_hdr_length;
826 
827 #define	iphs    ((uint16_t *)ip6h)
828 
829 	if ((ixaflags & IXAF_NO_HW_CKSUM) || !ILL_HCKSUM_CAPABLE(ill) ||
830 	    !dohwcksum) {
831 		return (ip_output_sw_cksum_v6(mp, ip6h, ixa));
832 	}
833 
834 	/*
835 	 * Calculate ULP checksum. Note that we don't use cksump and cksum
836 	 * if the ill has FULL support.
837 	 */
838 	if (protocol == IPPROTO_TCP) {
839 		cksump = IPH_TCPH_CHECKSUMP(ip6h, ip_hdr_length);
840 		cksum = IP_TCP_CSUM_COMP;	/* Pseudo-header cksum */
841 	} else if (protocol == IPPROTO_UDP) {
842 		cksump = IPH_UDPH_CHECKSUMP(ip6h, ip_hdr_length);
843 		cksum = IP_UDP_CSUM_COMP;	/* Pseudo-header cksum */
844 	} else if (protocol == IPPROTO_SCTP) {
845 		sctp_hdr_t	*sctph;
846 
847 		ASSERT(MBLKL(mp) >= (ip_hdr_length + sizeof (*sctph)));
848 		sctph = (sctp_hdr_t *)(mp->b_rptr + ip_hdr_length);
849 		/*
850 		 * Zero out the checksum field to ensure proper
851 		 * checksum calculation.
852 		 */
853 		sctph->sh_chksum = 0;
854 #ifdef	DEBUG
855 		if (!skip_sctp_cksum)
856 #endif
857 			sctph->sh_chksum = sctp_cksum(mp, ip_hdr_length);
858 		goto ip_hdr_cksum;
859 	} else if (ixa->ixa_flags & IXAF_SET_RAW_CKSUM) {
860 		/*
861 		 * icmp has placed length and routing
862 		 * header adjustment in the checksum field.
863 		 */
864 		cksump = (uint16_t *)(((uint8_t *)ip6h) + ip_hdr_length +
865 		    ixa->ixa_raw_cksum_offset);
866 		cksum = htons(protocol);
867 	} else if (protocol == IPPROTO_ICMPV6) {
868 		cksump = IPH_ICMPV6_CHECKSUMP(ip6h, ip_hdr_length);
869 		cksum = IP_ICMPV6_CSUM_COMP;	/* Pseudo-header cksum */
870 	} else {
871 	ip_hdr_cksum:
872 		/* No IP header checksum for IPv6 */
873 		return (B_TRUE);
874 	}
875 
876 	/* ULP puts the checksum field is in the first mblk */
877 	ASSERT(((uchar_t *)cksump) + sizeof (uint16_t) <= mp->b_wptr);
878 
879 	/*
880 	 * Underlying interface supports hardware checksum offload for
881 	 * the payload; leave the payload checksum for the hardware to
882 	 * calculate.  N.B: We only need to set up checksum info on the
883 	 * first mblk.
884 	 */
885 	hck_flags = ill->ill_hcksum_capab->ill_hcksum_txflags;
886 
887 	DB_CKSUMFLAGS(mp) &= ~HCK_FLAGS;
888 	if (hck_flags & HCKSUM_INET_FULL_V6) {
889 		/*
890 		 * Hardware calculates pseudo-header, header and the
891 		 * payload checksums, so clear the checksum field in
892 		 * the protocol header.
893 		 */
894 		*cksump = 0;
895 		DB_CKSUMFLAGS(mp) |= HCK_FULLCKSUM;
896 		return (B_TRUE);
897 	}
898 	if (((hck_flags) & HCKSUM_INET_PARTIAL) &&
899 	    (protocol != IPPROTO_ICMPV6 || !nxge_cksum_workaround)) {
900 		/*
901 		 * Partial checksum offload has been enabled.  Fill
902 		 * the checksum field in the protocol header with the
903 		 * pseudo-header checksum value.
904 		 *
905 		 * We accumulate the pseudo header checksum in cksum.
906 		 * This is pretty hairy code, so watch close.  One
907 		 * thing to keep in mind is that UDP and TCP have
908 		 * stored their respective datagram lengths in their
909 		 * checksum fields.  This lines things up real nice.
910 		 */
911 		cksum += iphs[4] + iphs[5] + iphs[6] + iphs[7] +
912 		    iphs[8] + iphs[9] + iphs[10] + iphs[11] +
913 		    iphs[12] + iphs[13] + iphs[14] + iphs[15] +
914 		    iphs[16] + iphs[17] + iphs[18] + iphs[19];
915 		cksum += *(cksump);
916 		cksum = (cksum & 0xFFFF) + (cksum >> 16);
917 		*(cksump) = (cksum & 0xFFFF) + (cksum >> 16);
918 
919 		/*
920 		 * Offsets are relative to beginning of IP header.
921 		 */
922 		DB_CKSUMSTART(mp) = ip_hdr_length;
923 		DB_CKSUMSTUFF(mp) = (uint8_t *)cksump - (uint8_t *)ip6h;
924 		DB_CKSUMEND(mp) = pktlen;
925 		DB_CKSUMFLAGS(mp) |= HCK_PARTIALCKSUM;
926 		return (B_TRUE);
927 	}
928 	/* Hardware capabilities include neither full nor partial IPv6 */
929 	return (ip_output_sw_cksum_v6(mp, ip6h, ixa));
930 #undef	iphs
931 }
932 
933 /*
934  * ire_sendfn for offlink and onlink destinations.
935  * Also called from the multicast, and multirt send functions.
936  *
937  * Assumes that the caller has a hold on the ire.
938  *
939  * This function doesn't care if the IRE just became condemned since that
940  * can happen at any time.
941  */
942 /* ARGSUSED */
943 int
ire_send_wire_v6(ire_t * ire,mblk_t * mp,void * iph_arg,ip_xmit_attr_t * ixa,uint32_t * identp)944 ire_send_wire_v6(ire_t *ire, mblk_t *mp, void *iph_arg,
945     ip_xmit_attr_t *ixa, uint32_t *identp)
946 {
947 	ip_stack_t	*ipst = ixa->ixa_ipst;
948 	ip6_t		*ip6h = (ip6_t *)iph_arg;
949 	iaflags_t	ixaflags = ixa->ixa_flags;
950 	ill_t		*ill;
951 	uint32_t	pktlen = ixa->ixa_pktlen;
952 
953 	ASSERT(ixa->ixa_nce != NULL);
954 	ill = ixa->ixa_nce->nce_ill;
955 
956 	/*
957 	 * Update output mib stats. Note that we can't move into the icmp
958 	 * sender (icmp_output etc) since they don't know the ill and the
959 	 * stats are per ill.
960 	 *
961 	 * With IPMP we record the stats on the upper ill.
962 	 */
963 	if (ixa->ixa_protocol == IPPROTO_ICMPV6) {
964 		icmp6_t		*icmp6;
965 
966 		icmp6 = (icmp6_t *)((uchar_t *)ip6h + ixa->ixa_ip_hdr_length);
967 		icmp_update_out_mib_v6(ixa->ixa_nce->nce_common->ncec_ill,
968 		    icmp6);
969 	}
970 
971 	if (ixaflags & IXAF_DONTROUTE)
972 		ip6h->ip6_hops = 1;
973 
974 	/*
975 	 * This might set b_band, thus the IPsec and fragmentation
976 	 * code in IP ensures that b_band is updated in the first mblk.
977 	 */
978 	if (IPP_ENABLED(IPP_LOCAL_OUT, ipst)) {
979 		/* ip_process translates an IS_UNDER_IPMP */
980 		mp = ip_process(IPP_LOCAL_OUT, mp, ill, ill);
981 		if (mp == NULL) {
982 			/* ip_drop_packet and MIB done */
983 			return (0);	/* Might just be delayed */
984 		}
985 	}
986 
987 	/*
988 	 * To handle IPsec/iptun's labeling needs we need to tag packets
989 	 * while we still have ixa_tsl
990 	 */
991 	if (is_system_labeled() && ixa->ixa_tsl != NULL &&
992 	    (ill->ill_mactype == DL_6TO4 || ill->ill_mactype == DL_IPV4 ||
993 	    ill->ill_mactype == DL_IPV6)) {
994 		cred_t *newcr;
995 
996 		newcr = copycred_from_tslabel(ixa->ixa_cred, ixa->ixa_tsl,
997 		    KM_NOSLEEP);
998 		if (newcr == NULL) {
999 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1000 			ip_drop_output("ipIfStatsOutDiscards - newcr",
1001 			    mp, ill);
1002 			freemsg(mp);
1003 			return (ENOBUFS);
1004 		}
1005 		mblk_setcred(mp, newcr, NOPID);
1006 		crfree(newcr);	/* mblk_setcred did its own crhold */
1007 	}
1008 
1009 	/*
1010 	 * IXAF_IPV6_ADD_FRAGHDR is set for CGTP so that we will add a
1011 	 * fragment header without fragmenting. CGTP on the receiver will
1012 	 * filter duplicates on the ident field.
1013 	 */
1014 	if (pktlen > ixa->ixa_fragsize ||
1015 	    (ixaflags & (IXAF_IPSEC_SECURE|IXAF_IPV6_ADD_FRAGHDR))) {
1016 		uint32_t ident;
1017 
1018 		if (ixaflags & IXAF_IPSEC_SECURE)
1019 			pktlen += ipsec_out_extra_length(ixa);
1020 
1021 		if (pktlen > IP_MAXPACKET)
1022 			return (EMSGSIZE);
1023 
1024 		if (ixaflags & IXAF_SET_ULP_CKSUM) {
1025 			/*
1026 			 * Compute ULP checksum using software
1027 			 */
1028 			if (!ip_output_sw_cksum_v6(mp, ip6h, ixa)) {
1029 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1030 				ip_drop_output("ipIfStatsOutDiscards", mp, ill);
1031 				freemsg(mp);
1032 				return (EINVAL);
1033 			}
1034 			/* Avoid checksum again below if we only add fraghdr */
1035 			ixaflags &= ~IXAF_SET_ULP_CKSUM;
1036 		}
1037 
1038 		/*
1039 		 * If we need a fragment header, pick the ident and insert
1040 		 * the header before IPsec to we have a place to store
1041 		 * the ident value.
1042 		 */
1043 		if ((ixaflags & IXAF_IPV6_ADD_FRAGHDR) ||
1044 		    pktlen > ixa->ixa_fragsize) {
1045 			/*
1046 			 * If this packet would generate a icmp_frag_needed
1047 			 * message, we need to handle it before we do the IPsec
1048 			 * processing. Otherwise, we need to strip the IPsec
1049 			 * headers before we send up the message to the ULPs
1050 			 * which becomes messy and difficult.
1051 			 */
1052 			if ((pktlen > ixa->ixa_fragsize) &&
1053 			    (ixaflags & IXAF_DONTFRAG)) {
1054 				/* Generate ICMP and return error */
1055 				ip_recv_attr_t	iras;
1056 
1057 				DTRACE_PROBE4(ip6__fragsize__fail,
1058 				    uint_t, pktlen, uint_t, ixa->ixa_fragsize,
1059 				    uint_t, ixa->ixa_pktlen,
1060 				    uint_t, ixa->ixa_pmtu);
1061 
1062 				bzero(&iras, sizeof (iras));
1063 				/* Map ixa to ira including IPsec policies */
1064 				ipsec_out_to_in(ixa, ill, &iras);
1065 
1066 				ip_drop_output("ICMP6_PKT_TOO_BIG", mp, ill);
1067 				icmp_pkt2big_v6(mp, ixa->ixa_fragsize, B_TRUE,
1068 				    &iras);
1069 				/* We moved any IPsec refs from ixa to iras */
1070 				ira_cleanup(&iras, B_FALSE);
1071 				return (EMSGSIZE);
1072 			}
1073 			DTRACE_PROBE4(ip6__fragsize__ok, uint_t, pktlen,
1074 			    uint_t, ixa->ixa_fragsize, uint_t, ixa->ixa_pktlen,
1075 			    uint_t, ixa->ixa_pmtu);
1076 			/*
1077 			 * Assign an ident value for this packet. There could
1078 			 * be other threads targeting the same destination, so
1079 			 * we have to arrange for a atomic increment.
1080 			 * Normally ixa_extra_ident is 0, but in the case of
1081 			 * LSO it will be the number of TCP segments  that the
1082 			 * driver/hardware will extraly construct.
1083 			 *
1084 			 * Note that cl_inet_ipident has only been used for
1085 			 * IPv4. We don't use it here.
1086 			 */
1087 			ident = atomic_add_32_nv(identp, ixa->ixa_extra_ident +
1088 			    1);
1089 			ixa->ixa_ident = ident;	/* In case we do IPsec */
1090 		}
1091 		if (ixaflags & IXAF_IPSEC_SECURE) {
1092 			/*
1093 			 * Pass in sufficient information so that
1094 			 * IPsec can determine whether to fragment, and
1095 			 * which function to call after fragmentation.
1096 			 */
1097 			return (ipsec_out_process(mp, ixa));
1098 		}
1099 
1100 		mp = ip_fraghdr_add_v6(mp, ident, ixa);
1101 		if (mp == NULL) {
1102 			/* MIB and ip_drop_output already done */
1103 			return (ENOMEM);
1104 		}
1105 		ASSERT(pktlen == ixa->ixa_pktlen);
1106 		pktlen += sizeof (ip6_frag_t);
1107 
1108 		if (pktlen > ixa->ixa_fragsize) {
1109 			return (ip_fragment_v6(mp, ixa->ixa_nce, ixaflags,
1110 			    pktlen, ixa->ixa_fragsize,
1111 			    ixa->ixa_xmit_hint, ixa->ixa_zoneid,
1112 			    ixa->ixa_no_loop_zoneid, ixa->ixa_postfragfn,
1113 			    &ixa->ixa_cookie));
1114 		}
1115 	}
1116 	if (ixaflags & IXAF_SET_ULP_CKSUM) {
1117 		/* Compute ULP checksum and IP header checksum */
1118 		/* An IS_UNDER_IPMP ill is ok here */
1119 		if (!ip_output_cksum_v6(ixaflags, mp, ip6h, ixa, ill)) {
1120 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1121 			ip_drop_output("ipIfStatsOutDiscards", mp, ill);
1122 			freemsg(mp);
1123 			return (EINVAL);
1124 		}
1125 	}
1126 	return ((ixa->ixa_postfragfn)(mp, ixa->ixa_nce, ixaflags,
1127 	    pktlen, ixa->ixa_xmit_hint, ixa->ixa_zoneid,
1128 	    ixa->ixa_no_loop_zoneid, &ixa->ixa_cookie));
1129 }
1130 
1131 /*
1132  * Post fragmentation function for RTF_MULTIRT routes.
1133  * Since IRE_MULTICASTs might have RTF_MULTIRT, this function
1134  * checks IXAF_LOOPBACK_COPY.
1135  *
1136  * If no packet is sent due to failures then we return an errno, but if at
1137  * least one succeeded we return zero.
1138  */
1139 int
ip_postfrag_multirt_v6(mblk_t * mp,nce_t * nce,iaflags_t ixaflags,uint_t pkt_len,uint32_t xmit_hint,zoneid_t szone,zoneid_t nolzid,uintptr_t * ixacookie)1140 ip_postfrag_multirt_v6(mblk_t *mp, nce_t *nce, iaflags_t ixaflags,
1141     uint_t pkt_len, uint32_t xmit_hint, zoneid_t szone, zoneid_t nolzid,
1142     uintptr_t *ixacookie)
1143 {
1144 	irb_t		*irb;
1145 	ip6_t		*ip6h = (ip6_t *)mp->b_rptr;
1146 	ire_t		*ire;
1147 	ire_t		*ire1;
1148 	mblk_t		*mp1;
1149 	nce_t		*nce1;
1150 	ill_t		*ill = nce->nce_ill;
1151 	ill_t		*ill1;
1152 	ip_stack_t	*ipst = ill->ill_ipst;
1153 	int		error = 0;
1154 	int		num_sent = 0;
1155 	int		err;
1156 	uint_t		ire_type;
1157 	in6_addr_t	nexthop;
1158 
1159 	ASSERT(!(ixaflags & IXAF_IS_IPV4));
1160 
1161 	/* Check for IXAF_LOOPBACK_COPY */
1162 	if (ixaflags & IXAF_LOOPBACK_COPY) {
1163 		mblk_t *mp1;
1164 
1165 		mp1 = copymsg(mp);
1166 		if (mp1 == NULL) {
1167 			/* Failed to deliver the loopback copy. */
1168 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1169 			ip_drop_output("ipIfStatsOutDiscards", mp, ill);
1170 			error = ENOBUFS;
1171 		} else {
1172 			ip_postfrag_loopback(mp1, nce, ixaflags, pkt_len,
1173 			    nolzid);
1174 		}
1175 	}
1176 
1177 	/*
1178 	 * Loop over RTF_MULTIRT for ip6_dst in the same bucket. Send
1179 	 * a copy to each one.
1180 	 * Use the nce (nexthop) and ip6_dst to find the ire.
1181 	 *
1182 	 * MULTIRT is not designed to work with shared-IP zones thus we don't
1183 	 * need to pass a zoneid or a label to the IRE lookup.
1184 	 */
1185 	if (IN6_ARE_ADDR_EQUAL(&nce->nce_addr, &ip6h->ip6_dst)) {
1186 		/* Broadcast and multicast case */
1187 		ire = ire_ftable_lookup_v6(&ip6h->ip6_dst, 0, 0, 0, NULL,
1188 		    ALL_ZONES, NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
1189 	} else {
1190 		/* Unicast case */
1191 		ire = ire_ftable_lookup_v6(&ip6h->ip6_dst, 0, &nce->nce_addr,
1192 		    0, NULL, ALL_ZONES, NULL, MATCH_IRE_GW, 0, ipst, NULL);
1193 	}
1194 
1195 	if (ire == NULL ||
1196 	    (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
1197 	    !(ire->ire_flags & RTF_MULTIRT)) {
1198 		/* Drop */
1199 		ip_drop_output("ip_postfrag_multirt didn't find route",
1200 		    mp, nce->nce_ill);
1201 		if (ire != NULL)
1202 			ire_refrele(ire);
1203 		return (ENETUNREACH);
1204 	}
1205 
1206 	irb = ire->ire_bucket;
1207 	irb_refhold(irb);
1208 	for (ire1 = irb->irb_ire; ire1 != NULL; ire1 = ire1->ire_next) {
1209 		if (IRE_IS_CONDEMNED(ire1) ||
1210 		    !(ire1->ire_flags & RTF_MULTIRT))
1211 			continue;
1212 
1213 		/* Note: When IPv6 uses radix tree we don't need this check */
1214 		if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, &ire1->ire_addr_v6))
1215 			continue;
1216 
1217 		/* Do the ire argument one after the loop */
1218 		if (ire1 == ire)
1219 			continue;
1220 
1221 		ill1 = ire_nexthop_ill(ire1);
1222 		if (ill1 == NULL) {
1223 			/*
1224 			 * This ire might not have been picked by
1225 			 * ire_route_recursive, in which case ire_dep might
1226 			 * not have been setup yet.
1227 			 * We kick ire_route_recursive to try to resolve
1228 			 * starting at ire1.
1229 			 */
1230 			ire_t *ire2;
1231 			uint_t match_flags = MATCH_IRE_DSTONLY;
1232 
1233 			if (ire1->ire_ill != NULL)
1234 				match_flags |= MATCH_IRE_ILL;
1235 			ire2 = ire_route_recursive_impl_v6(ire1,
1236 			    &ire1->ire_addr_v6, ire1->ire_type, ire1->ire_ill,
1237 			    ire1->ire_zoneid, NULL, match_flags,
1238 			    IRR_ALLOCATE, 0, ipst, NULL, NULL, NULL);
1239 			if (ire2 != NULL)
1240 				ire_refrele(ire2);
1241 			ill1 = ire_nexthop_ill(ire1);
1242 		}
1243 		if (ill1 == NULL) {
1244 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1245 			ip_drop_output("ipIfStatsOutDiscards - no ill",
1246 			    mp, ill);
1247 			error = ENETUNREACH;
1248 			continue;
1249 		}
1250 		/* Pick the addr and type to use for ndp_nce_init */
1251 		if (nce->nce_common->ncec_flags & NCE_F_MCAST) {
1252 			ire_type = IRE_MULTICAST;
1253 			nexthop = ip6h->ip6_dst;
1254 		} else {
1255 			ire_type = ire1->ire_type;	/* Doesn't matter */
1256 			nexthop = ire1->ire_gateway_addr_v6;
1257 		}
1258 
1259 		/* If IPMP meta or under, then we just drop */
1260 		if (ill1->ill_grp != NULL) {
1261 			BUMP_MIB(ill1->ill_ip_mib, ipIfStatsOutDiscards);
1262 			ip_drop_output("ipIfStatsOutDiscards - IPMP",
1263 			    mp, ill1);
1264 			ill_refrele(ill1);
1265 			error = ENETUNREACH;
1266 			continue;
1267 		}
1268 
1269 		nce1 = ndp_nce_init(ill1, &nexthop, ire_type);
1270 		if (nce1 == NULL) {
1271 			BUMP_MIB(ill1->ill_ip_mib, ipIfStatsOutDiscards);
1272 			ip_drop_output("ipIfStatsOutDiscards - no nce",
1273 			    mp, ill1);
1274 			ill_refrele(ill1);
1275 			error = ENOBUFS;
1276 			continue;
1277 		}
1278 		mp1 = copymsg(mp);
1279 		if (mp1 == NULL) {
1280 			BUMP_MIB(ill1->ill_ip_mib, ipIfStatsOutDiscards);
1281 			ip_drop_output("ipIfStatsOutDiscards", mp, ill1);
1282 			nce_refrele(nce1);
1283 			ill_refrele(ill1);
1284 			error = ENOBUFS;
1285 			continue;
1286 		}
1287 		/* Preserve HW checksum for this copy */
1288 		DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
1289 		DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
1290 		DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
1291 		DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
1292 		DB_LSOMSS(mp1) = DB_LSOMSS(mp);
1293 
1294 		ire1->ire_ob_pkt_count++;
1295 		err = ip_xmit(mp1, nce1, ixaflags, pkt_len, xmit_hint, szone,
1296 		    0, ixacookie);
1297 		if (err == 0)
1298 			num_sent++;
1299 		else
1300 			error = err;
1301 		nce_refrele(nce1);
1302 		ill_refrele(ill1);
1303 	}
1304 	irb_refrele(irb);
1305 	ire_refrele(ire);
1306 	/* Finally, the main one */
1307 	err = ip_xmit(mp, nce, ixaflags, pkt_len, xmit_hint, szone, 0,
1308 	    ixacookie);
1309 	if (err == 0)
1310 		num_sent++;
1311 	else
1312 		error = err;
1313 	if (num_sent > 0)
1314 		return (0);
1315 	else
1316 		return (error);
1317 }
1318