xref: /freebsd/sys/netinet6/frag6.c (revision eb69d1f144a6fcc765d1b9d44a5ae8082353e70b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_rss.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/domain.h>
44 #include <sys/eventhandler.h>
45 #include <sys/protosw.h>
46 #include <sys/socket.h>
47 #include <sys/errno.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/syslog.h>
51 
52 #include <net/if.h>
53 #include <net/if_var.h>
54 #include <net/netisr.h>
55 #include <net/route.h>
56 #include <net/vnet.h>
57 
58 #include <netinet/in.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip6.h>
61 #include <netinet6/ip6_var.h>
62 #include <netinet/icmp6.h>
63 #include <netinet/in_systm.h>	/* for ECN definitions */
64 #include <netinet/ip.h>		/* for ECN definitions */
65 
66 #include <security/mac/mac_framework.h>
67 
68 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *);
69 static void frag6_deq(struct ip6asfrag *);
70 static void frag6_insque(struct ip6q *, struct ip6q *);
71 static void frag6_remque(struct ip6q *);
72 static void frag6_freef(struct ip6q *);
73 
74 static struct mtx ip6qlock;
75 /*
76  * These fields all protected by ip6qlock.
77  */
78 static VNET_DEFINE(u_int, frag6_nfragpackets);
79 static VNET_DEFINE(u_int, frag6_nfrags);
80 static VNET_DEFINE(struct ip6q, ip6q);	/* ip6 reassemble queue */
81 
82 #define	V_frag6_nfragpackets		VNET(frag6_nfragpackets)
83 #define	V_frag6_nfrags			VNET(frag6_nfrags)
84 #define	V_ip6q				VNET(ip6q)
85 
86 #define	IP6Q_LOCK_INIT()	mtx_init(&ip6qlock, "ip6qlock", NULL, MTX_DEF);
87 #define	IP6Q_LOCK()		mtx_lock(&ip6qlock)
88 #define	IP6Q_TRYLOCK()		mtx_trylock(&ip6qlock)
89 #define	IP6Q_LOCK_ASSERT()	mtx_assert(&ip6qlock, MA_OWNED)
90 #define	IP6Q_UNLOCK()		mtx_unlock(&ip6qlock)
91 
92 static MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header");
93 
94 /*
95  * Initialise reassembly queue and fragment identifier.
96  */
97 static void
98 frag6_change(void *tag)
99 {
100 
101 	V_ip6_maxfragpackets = nmbclusters / 4;
102 	V_ip6_maxfrags = nmbclusters / 4;
103 }
104 
105 void
106 frag6_init(void)
107 {
108 
109 	V_ip6_maxfragpackets = nmbclusters / 4;
110 	V_ip6_maxfrags = nmbclusters / 4;
111 	V_ip6q.ip6q_next = V_ip6q.ip6q_prev = &V_ip6q;
112 
113 	if (!IS_DEFAULT_VNET(curvnet))
114 		return;
115 
116 	EVENTHANDLER_REGISTER(nmbclusters_change,
117 	    frag6_change, NULL, EVENTHANDLER_PRI_ANY);
118 
119 	IP6Q_LOCK_INIT();
120 }
121 
122 /*
123  * In RFC2460, fragment and reassembly rule do not agree with each other,
124  * in terms of next header field handling in fragment header.
125  * While the sender will use the same value for all of the fragmented packets,
126  * receiver is suggested not to check the consistency.
127  *
128  * fragment rule (p20):
129  *	(2) A Fragment header containing:
130  *	The Next Header value that identifies the first header of
131  *	the Fragmentable Part of the original packet.
132  *		-> next header field is same for all fragments
133  *
134  * reassembly rule (p21):
135  *	The Next Header field of the last header of the Unfragmentable
136  *	Part is obtained from the Next Header field of the first
137  *	fragment's Fragment header.
138  *		-> should grab it from the first fragment only
139  *
140  * The following note also contradicts with fragment rule - no one is going to
141  * send different fragment with different next header field.
142  *
143  * additional note (p22):
144  *	The Next Header values in the Fragment headers of different
145  *	fragments of the same original packet may differ.  Only the value
146  *	from the Offset zero fragment packet is used for reassembly.
147  *		-> should grab it from the first fragment only
148  *
149  * There is no explicit reason given in the RFC.  Historical reason maybe?
150  */
151 /*
152  * Fragment input
153  */
154 int
155 frag6_input(struct mbuf **mp, int *offp, int proto)
156 {
157 	struct mbuf *m = *mp, *t;
158 	struct ip6_hdr *ip6;
159 	struct ip6_frag *ip6f;
160 	struct ip6q *q6;
161 	struct ip6asfrag *af6, *ip6af, *af6dwn;
162 	struct in6_ifaddr *ia;
163 	int offset = *offp, nxt, i, next;
164 	int first_frag = 0;
165 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
166 	struct ifnet *dstifp;
167 	u_int8_t ecn, ecn0;
168 #ifdef RSS
169 	struct m_tag *mtag;
170 	struct ip6_direct_ctx *ip6dc;
171 #endif
172 
173 #if 0
174 	char ip6buf[INET6_ADDRSTRLEN];
175 #endif
176 
177 	ip6 = mtod(m, struct ip6_hdr *);
178 #ifndef PULLDOWN_TEST
179 	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
180 	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
181 #else
182 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
183 	if (ip6f == NULL)
184 		return (IPPROTO_DONE);
185 #endif
186 
187 	dstifp = NULL;
188 	/* find the destination interface of the packet. */
189 	ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
190 	if (ia != NULL) {
191 		dstifp = ia->ia_ifp;
192 		ifa_free(&ia->ia_ifa);
193 	}
194 	/* jumbo payload can't contain a fragment header */
195 	if (ip6->ip6_plen == 0) {
196 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
197 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
198 		return IPPROTO_DONE;
199 	}
200 
201 	/*
202 	 * check whether fragment packet's fragment length is
203 	 * multiple of 8 octets.
204 	 * sizeof(struct ip6_frag) == 8
205 	 * sizeof(struct ip6_hdr) = 40
206 	 */
207 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
208 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
209 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
210 		    offsetof(struct ip6_hdr, ip6_plen));
211 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
212 		return IPPROTO_DONE;
213 	}
214 
215 	IP6STAT_INC(ip6s_fragments);
216 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
217 
218 	/* offset now points to data portion */
219 	offset += sizeof(struct ip6_frag);
220 
221 	/*
222 	 * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0)
223 	 * upfront, unrelated to any reassembly.  Just skip the fragment header.
224 	 */
225 	if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
226 		/* XXX-BZ we want dedicated counters for this. */
227 		IP6STAT_INC(ip6s_reassembled);
228 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
229 		*offp = offset;
230 		m->m_flags |= M_FRAGMENTED;
231 		return (ip6f->ip6f_nxt);
232 	}
233 
234 	IP6Q_LOCK();
235 
236 	/*
237 	 * Enforce upper bound on number of fragments.
238 	 * If maxfrag is 0, never accept fragments.
239 	 * If maxfrag is -1, accept all fragments without limitation.
240 	 */
241 	if (V_ip6_maxfrags < 0)
242 		;
243 	else if (V_frag6_nfrags >= (u_int)V_ip6_maxfrags)
244 		goto dropfrag;
245 
246 	for (q6 = V_ip6q.ip6q_next; q6 != &V_ip6q; q6 = q6->ip6q_next)
247 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
248 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
249 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
250 #ifdef MAC
251 		    && mac_ip6q_match(m, q6)
252 #endif
253 		    )
254 			break;
255 
256 	if (q6 == &V_ip6q) {
257 		/*
258 		 * the first fragment to arrive, create a reassembly queue.
259 		 */
260 		first_frag = 1;
261 
262 		/*
263 		 * Enforce upper bound on number of fragmented packets
264 		 * for which we attempt reassembly;
265 		 * If maxfragpackets is 0, never accept fragments.
266 		 * If maxfragpackets is -1, accept all fragments without
267 		 * limitation.
268 		 */
269 		if (V_ip6_maxfragpackets < 0)
270 			;
271 		else if (V_frag6_nfragpackets >= (u_int)V_ip6_maxfragpackets)
272 			goto dropfrag;
273 		V_frag6_nfragpackets++;
274 		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
275 		    M_NOWAIT);
276 		if (q6 == NULL)
277 			goto dropfrag;
278 		bzero(q6, sizeof(*q6));
279 #ifdef MAC
280 		if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
281 			free(q6, M_FTABLE);
282 			goto dropfrag;
283 		}
284 		mac_ip6q_create(m, q6);
285 #endif
286 		frag6_insque(q6, &V_ip6q);
287 
288 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
289 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
290 #ifdef notyet
291 		q6->ip6q_nxtp	= (u_char *)nxtp;
292 #endif
293 		q6->ip6q_ident	= ip6f->ip6f_ident;
294 		q6->ip6q_ttl	= IPV6_FRAGTTL;
295 		q6->ip6q_src	= ip6->ip6_src;
296 		q6->ip6q_dst	= ip6->ip6_dst;
297 		q6->ip6q_ecn	=
298 		    (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
299 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
300 
301 		q6->ip6q_nfrag = 0;
302 	}
303 
304 	/*
305 	 * If it's the 1st fragment, record the length of the
306 	 * unfragmentable part and the next header of the fragment header.
307 	 */
308 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
309 	if (fragoff == 0) {
310 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
311 		    sizeof(struct ip6_frag);
312 		q6->ip6q_nxt = ip6f->ip6f_nxt;
313 	}
314 
315 	/*
316 	 * Check that the reassembled packet would not exceed 65535 bytes
317 	 * in size.
318 	 * If it would exceed, discard the fragment and return an ICMP error.
319 	 */
320 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
321 	if (q6->ip6q_unfrglen >= 0) {
322 		/* The 1st fragment has already arrived. */
323 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
324 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
325 			    offset - sizeof(struct ip6_frag) +
326 			    offsetof(struct ip6_frag, ip6f_offlg));
327 			IP6Q_UNLOCK();
328 			return (IPPROTO_DONE);
329 		}
330 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
331 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
332 		    offset - sizeof(struct ip6_frag) +
333 		    offsetof(struct ip6_frag, ip6f_offlg));
334 		IP6Q_UNLOCK();
335 		return (IPPROTO_DONE);
336 	}
337 	/*
338 	 * If it's the first fragment, do the above check for each
339 	 * fragment already stored in the reassembly queue.
340 	 */
341 	if (fragoff == 0) {
342 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
343 		     af6 = af6dwn) {
344 			af6dwn = af6->ip6af_down;
345 
346 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
347 			    IPV6_MAXPACKET) {
348 				struct mbuf *merr = IP6_REASS_MBUF(af6);
349 				struct ip6_hdr *ip6err;
350 				int erroff = af6->ip6af_offset;
351 
352 				/* dequeue the fragment. */
353 				frag6_deq(af6);
354 				free(af6, M_FTABLE);
355 
356 				/* adjust pointer. */
357 				ip6err = mtod(merr, struct ip6_hdr *);
358 
359 				/*
360 				 * Restore source and destination addresses
361 				 * in the erroneous IPv6 header.
362 				 */
363 				ip6err->ip6_src = q6->ip6q_src;
364 				ip6err->ip6_dst = q6->ip6q_dst;
365 
366 				icmp6_error(merr, ICMP6_PARAM_PROB,
367 				    ICMP6_PARAMPROB_HEADER,
368 				    erroff - sizeof(struct ip6_frag) +
369 				    offsetof(struct ip6_frag, ip6f_offlg));
370 			}
371 		}
372 	}
373 
374 	ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
375 	    M_NOWAIT);
376 	if (ip6af == NULL)
377 		goto dropfrag;
378 	bzero(ip6af, sizeof(*ip6af));
379 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
380 	ip6af->ip6af_off = fragoff;
381 	ip6af->ip6af_frglen = frgpartlen;
382 	ip6af->ip6af_offset = offset;
383 	IP6_REASS_MBUF(ip6af) = m;
384 
385 	if (first_frag) {
386 		af6 = (struct ip6asfrag *)q6;
387 		goto insert;
388 	}
389 
390 	/*
391 	 * Handle ECN by comparing this segment with the first one;
392 	 * if CE is set, do not lose CE.
393 	 * drop if CE and not-ECT are mixed for the same packet.
394 	 */
395 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
396 	ecn0 = q6->ip6q_ecn;
397 	if (ecn == IPTOS_ECN_CE) {
398 		if (ecn0 == IPTOS_ECN_NOTECT) {
399 			free(ip6af, M_FTABLE);
400 			goto dropfrag;
401 		}
402 		if (ecn0 != IPTOS_ECN_CE)
403 			q6->ip6q_ecn = IPTOS_ECN_CE;
404 	}
405 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
406 		free(ip6af, M_FTABLE);
407 		goto dropfrag;
408 	}
409 
410 	/*
411 	 * Find a segment which begins after this one does.
412 	 */
413 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
414 	     af6 = af6->ip6af_down)
415 		if (af6->ip6af_off > ip6af->ip6af_off)
416 			break;
417 
418 #if 0
419 	/*
420 	 * If there is a preceding segment, it may provide some of
421 	 * our data already.  If so, drop the data from the incoming
422 	 * segment.  If it provides all of our data, drop us.
423 	 */
424 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
425 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
426 			- ip6af->ip6af_off;
427 		if (i > 0) {
428 			if (i >= ip6af->ip6af_frglen)
429 				goto dropfrag;
430 			m_adj(IP6_REASS_MBUF(ip6af), i);
431 			ip6af->ip6af_off += i;
432 			ip6af->ip6af_frglen -= i;
433 		}
434 	}
435 
436 	/*
437 	 * While we overlap succeeding segments trim them or,
438 	 * if they are completely covered, dequeue them.
439 	 */
440 	while (af6 != (struct ip6asfrag *)q6 &&
441 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
442 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
443 		if (i < af6->ip6af_frglen) {
444 			af6->ip6af_frglen -= i;
445 			af6->ip6af_off += i;
446 			m_adj(IP6_REASS_MBUF(af6), i);
447 			break;
448 		}
449 		af6 = af6->ip6af_down;
450 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
451 		frag6_deq(af6->ip6af_up);
452 	}
453 #else
454 	/*
455 	 * If the incoming framgent overlaps some existing fragments in
456 	 * the reassembly queue, drop it, since it is dangerous to override
457 	 * existing fragments from a security point of view.
458 	 * We don't know which fragment is the bad guy - here we trust
459 	 * fragment that came in earlier, with no real reason.
460 	 *
461 	 * Note: due to changes after disabling this part, mbuf passed to
462 	 * m_adj() below now does not meet the requirement.
463 	 */
464 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
465 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
466 			- ip6af->ip6af_off;
467 		if (i > 0) {
468 #if 0				/* suppress the noisy log */
469 			log(LOG_ERR, "%d bytes of a fragment from %s "
470 			    "overlaps the previous fragment\n",
471 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
472 #endif
473 			free(ip6af, M_FTABLE);
474 			goto dropfrag;
475 		}
476 	}
477 	if (af6 != (struct ip6asfrag *)q6) {
478 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
479 		if (i > 0) {
480 #if 0				/* suppress the noisy log */
481 			log(LOG_ERR, "%d bytes of a fragment from %s "
482 			    "overlaps the succeeding fragment",
483 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
484 #endif
485 			free(ip6af, M_FTABLE);
486 			goto dropfrag;
487 		}
488 	}
489 #endif
490 
491 insert:
492 #ifdef MAC
493 	if (!first_frag)
494 		mac_ip6q_update(m, q6);
495 #endif
496 
497 	/*
498 	 * Stick new segment in its place;
499 	 * check for complete reassembly.
500 	 * Move to front of packet queue, as we are
501 	 * the most recently active fragmented packet.
502 	 */
503 	frag6_enq(ip6af, af6->ip6af_up);
504 	V_frag6_nfrags++;
505 	q6->ip6q_nfrag++;
506 #if 0 /* xxx */
507 	if (q6 != V_ip6q.ip6q_next) {
508 		frag6_remque(q6);
509 		frag6_insque(q6, &V_ip6q);
510 	}
511 #endif
512 	next = 0;
513 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
514 	     af6 = af6->ip6af_down) {
515 		if (af6->ip6af_off != next) {
516 			IP6Q_UNLOCK();
517 			return IPPROTO_DONE;
518 		}
519 		next += af6->ip6af_frglen;
520 	}
521 	if (af6->ip6af_up->ip6af_mff) {
522 		IP6Q_UNLOCK();
523 		return IPPROTO_DONE;
524 	}
525 
526 	/*
527 	 * Reassembly is complete; concatenate fragments.
528 	 */
529 	ip6af = q6->ip6q_down;
530 	t = m = IP6_REASS_MBUF(ip6af);
531 	af6 = ip6af->ip6af_down;
532 	frag6_deq(ip6af);
533 	while (af6 != (struct ip6asfrag *)q6) {
534 		m->m_pkthdr.csum_flags &=
535 		    IP6_REASS_MBUF(af6)->m_pkthdr.csum_flags;
536 		m->m_pkthdr.csum_data +=
537 		    IP6_REASS_MBUF(af6)->m_pkthdr.csum_data;
538 
539 		af6dwn = af6->ip6af_down;
540 		frag6_deq(af6);
541 		while (t->m_next)
542 			t = t->m_next;
543 		m_adj(IP6_REASS_MBUF(af6), af6->ip6af_offset);
544 		m_demote_pkthdr(IP6_REASS_MBUF(af6));
545 		m_cat(t, IP6_REASS_MBUF(af6));
546 		free(af6, M_FTABLE);
547 		af6 = af6dwn;
548 	}
549 
550 	while (m->m_pkthdr.csum_data & 0xffff0000)
551 		m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
552 		    (m->m_pkthdr.csum_data >> 16);
553 
554 	/* adjust offset to point where the original next header starts */
555 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
556 	free(ip6af, M_FTABLE);
557 	ip6 = mtod(m, struct ip6_hdr *);
558 	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
559 	if (q6->ip6q_ecn == IPTOS_ECN_CE)
560 		ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
561 	nxt = q6->ip6q_nxt;
562 #ifdef notyet
563 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
564 #endif
565 
566 	if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) {
567 		frag6_remque(q6);
568 		V_frag6_nfrags -= q6->ip6q_nfrag;
569 #ifdef MAC
570 		mac_ip6q_destroy(q6);
571 #endif
572 		free(q6, M_FTABLE);
573 		V_frag6_nfragpackets--;
574 
575 		goto dropfrag;
576 	}
577 
578 	/*
579 	 * Store NXT to the original.
580 	 */
581 	m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t),
582 	    (caddr_t)&nxt);
583 
584 	frag6_remque(q6);
585 	V_frag6_nfrags -= q6->ip6q_nfrag;
586 #ifdef MAC
587 	mac_ip6q_reassemble(q6, m);
588 	mac_ip6q_destroy(q6);
589 #endif
590 	free(q6, M_FTABLE);
591 	V_frag6_nfragpackets--;
592 
593 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
594 		int plen = 0;
595 		for (t = m; t; t = t->m_next)
596 			plen += t->m_len;
597 		m->m_pkthdr.len = plen;
598 	}
599 
600 #ifdef RSS
601 	mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc),
602 	    M_NOWAIT);
603 	if (mtag == NULL)
604 		goto dropfrag;
605 
606 	ip6dc = (struct ip6_direct_ctx *)(mtag + 1);
607 	ip6dc->ip6dc_nxt = nxt;
608 	ip6dc->ip6dc_off = offset;
609 
610 	m_tag_prepend(m, mtag);
611 #endif
612 
613 	IP6Q_UNLOCK();
614 	IP6STAT_INC(ip6s_reassembled);
615 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
616 
617 #ifdef RSS
618 	/*
619 	 * Queue/dispatch for reprocessing.
620 	 */
621 	netisr_dispatch(NETISR_IPV6_DIRECT, m);
622 	return IPPROTO_DONE;
623 #endif
624 
625 	/*
626 	 * Tell launch routine the next header
627 	 */
628 
629 	*mp = m;
630 	*offp = offset;
631 
632 	return nxt;
633 
634  dropfrag:
635 	IP6Q_UNLOCK();
636 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
637 	IP6STAT_INC(ip6s_fragdropped);
638 	m_freem(m);
639 	return IPPROTO_DONE;
640 }
641 
642 /*
643  * Free a fragment reassembly header and all
644  * associated datagrams.
645  */
646 void
647 frag6_freef(struct ip6q *q6)
648 {
649 	struct ip6asfrag *af6, *down6;
650 
651 	IP6Q_LOCK_ASSERT();
652 
653 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
654 	     af6 = down6) {
655 		struct mbuf *m = IP6_REASS_MBUF(af6);
656 
657 		down6 = af6->ip6af_down;
658 		frag6_deq(af6);
659 
660 		/*
661 		 * Return ICMP time exceeded error for the 1st fragment.
662 		 * Just free other fragments.
663 		 */
664 		if (af6->ip6af_off == 0) {
665 			struct ip6_hdr *ip6;
666 
667 			/* adjust pointer */
668 			ip6 = mtod(m, struct ip6_hdr *);
669 
670 			/* restore source and destination addresses */
671 			ip6->ip6_src = q6->ip6q_src;
672 			ip6->ip6_dst = q6->ip6q_dst;
673 
674 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
675 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
676 		} else
677 			m_freem(m);
678 		free(af6, M_FTABLE);
679 	}
680 	frag6_remque(q6);
681 	V_frag6_nfrags -= q6->ip6q_nfrag;
682 #ifdef MAC
683 	mac_ip6q_destroy(q6);
684 #endif
685 	free(q6, M_FTABLE);
686 	V_frag6_nfragpackets--;
687 }
688 
689 /*
690  * Put an ip fragment on a reassembly chain.
691  * Like insque, but pointers in middle of structure.
692  */
693 void
694 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
695 {
696 
697 	IP6Q_LOCK_ASSERT();
698 
699 	af6->ip6af_up = up6;
700 	af6->ip6af_down = up6->ip6af_down;
701 	up6->ip6af_down->ip6af_up = af6;
702 	up6->ip6af_down = af6;
703 }
704 
705 /*
706  * To frag6_enq as remque is to insque.
707  */
708 void
709 frag6_deq(struct ip6asfrag *af6)
710 {
711 
712 	IP6Q_LOCK_ASSERT();
713 
714 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
715 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
716 }
717 
718 void
719 frag6_insque(struct ip6q *new, struct ip6q *old)
720 {
721 
722 	IP6Q_LOCK_ASSERT();
723 
724 	new->ip6q_prev = old;
725 	new->ip6q_next = old->ip6q_next;
726 	old->ip6q_next->ip6q_prev= new;
727 	old->ip6q_next = new;
728 }
729 
730 void
731 frag6_remque(struct ip6q *p6)
732 {
733 
734 	IP6Q_LOCK_ASSERT();
735 
736 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
737 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
738 }
739 
740 /*
741  * IPv6 reassembling timer processing;
742  * if a timer expires on a reassembly
743  * queue, discard it.
744  */
745 void
746 frag6_slowtimo(void)
747 {
748 	VNET_ITERATOR_DECL(vnet_iter);
749 	struct ip6q *q6;
750 
751 	VNET_LIST_RLOCK_NOSLEEP();
752 	IP6Q_LOCK();
753 	VNET_FOREACH(vnet_iter) {
754 		CURVNET_SET(vnet_iter);
755 		q6 = V_ip6q.ip6q_next;
756 		if (q6)
757 			while (q6 != &V_ip6q) {
758 				--q6->ip6q_ttl;
759 				q6 = q6->ip6q_next;
760 				if (q6->ip6q_prev->ip6q_ttl == 0) {
761 					IP6STAT_INC(ip6s_fragtimeout);
762 					/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
763 					frag6_freef(q6->ip6q_prev);
764 				}
765 			}
766 		/*
767 		 * If we are over the maximum number of fragments
768 		 * (due to the limit being lowered), drain off
769 		 * enough to get down to the new limit.
770 		 */
771 		while (V_frag6_nfragpackets > (u_int)V_ip6_maxfragpackets &&
772 		    V_ip6q.ip6q_prev) {
773 			IP6STAT_INC(ip6s_fragoverflow);
774 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
775 			frag6_freef(V_ip6q.ip6q_prev);
776 		}
777 		CURVNET_RESTORE();
778 	}
779 	IP6Q_UNLOCK();
780 	VNET_LIST_RUNLOCK_NOSLEEP();
781 }
782 
783 /*
784  * Drain off all datagram fragments.
785  */
786 void
787 frag6_drain(void)
788 {
789 	VNET_ITERATOR_DECL(vnet_iter);
790 
791 	VNET_LIST_RLOCK_NOSLEEP();
792 	if (IP6Q_TRYLOCK() == 0) {
793 		VNET_LIST_RUNLOCK_NOSLEEP();
794 		return;
795 	}
796 	VNET_FOREACH(vnet_iter) {
797 		CURVNET_SET(vnet_iter);
798 		while (V_ip6q.ip6q_next != &V_ip6q) {
799 			IP6STAT_INC(ip6s_fragdropped);
800 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
801 			frag6_freef(V_ip6q.ip6q_next);
802 		}
803 		CURVNET_RESTORE();
804 	}
805 	IP6Q_UNLOCK();
806 	VNET_LIST_RUNLOCK_NOSLEEP();
807 }
808 
809 int
810 ip6_deletefraghdr(struct mbuf *m, int offset, int wait)
811 {
812 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
813 	struct mbuf *t;
814 
815 	/* Delete frag6 header. */
816 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
817 		/* This is the only possible case with !PULLDOWN_TEST. */
818 		bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag),
819 		    offset);
820 		m->m_data += sizeof(struct ip6_frag);
821 		m->m_len -= sizeof(struct ip6_frag);
822 	} else {
823 		/* This comes with no copy if the boundary is on cluster. */
824 		if ((t = m_split(m, offset, wait)) == NULL)
825 			return (ENOMEM);
826 		m_adj(t, sizeof(struct ip6_frag));
827 		m_cat(m, t);
828 	}
829 
830 	m->m_flags |= M_FRAGMENTED;
831 	return (0);
832 }
833