xref: /freebsd/sys/netinet6/frag6.c (revision 830940567b49bb0c08dfaed40418999e76616909)
1 /*-
2  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the project nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	$KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
38 #include <sys/mbuf.h>
39 #include <sys/domain.h>
40 #include <sys/protosw.h>
41 #include <sys/socket.h>
42 #include <sys/errno.h>
43 #include <sys/time.h>
44 #include <sys/kernel.h>
45 #include <sys/syslog.h>
46 
47 #include <net/if.h>
48 #include <net/route.h>
49 #include <net/vnet.h>
50 
51 #include <netinet/in.h>
52 #include <netinet/in_var.h>
53 #include <netinet/ip6.h>
54 #include <netinet6/ip6_var.h>
55 #include <netinet/icmp6.h>
56 #include <netinet/in_systm.h>	/* for ECN definitions */
57 #include <netinet/ip.h>		/* for ECN definitions */
58 
59 #include <security/mac/mac_framework.h>
60 
61 /*
62  * Define it to get a correct behavior on per-interface statistics.
63  * You will need to perform an extra routing table lookup, per fragment,
64  * to do it.  This may, or may not be, a performance hit.
65  */
66 #define IN6_IFSTAT_STRICT
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_ip6q.ip6q_next = V_ip6q.ip6q_prev = &V_ip6q;
110 	V_ip6_maxfragpackets = nmbclusters / 4;
111 	V_ip6_maxfrags = nmbclusters / 4;
112 
113 	if (!IS_DEFAULT_VNET(curvnet))
114 		return;
115 
116 	IP6Q_LOCK_INIT();
117 	EVENTHANDLER_REGISTER(nmbclusters_change,
118 	    frag6_change, NULL, EVENTHANDLER_PRI_ANY);
119 }
120 
121 /*
122  * In RFC2460, fragment and reassembly rule do not agree with each other,
123  * in terms of next header field handling in fragment header.
124  * While the sender will use the same value for all of the fragmented packets,
125  * receiver is suggested not to check the consistency.
126  *
127  * fragment rule (p20):
128  *	(2) A Fragment header containing:
129  *	The Next Header value that identifies the first header of
130  *	the Fragmentable Part of the original packet.
131  *		-> next header field is same for all fragments
132  *
133  * reassembly rule (p21):
134  *	The Next Header field of the last header of the Unfragmentable
135  *	Part is obtained from the Next Header field of the first
136  *	fragment's Fragment header.
137  *		-> should grab it from the first fragment only
138  *
139  * The following note also contradicts with fragment rule - noone is going to
140  * send different fragment with different next header field.
141  *
142  * additional note (p22):
143  *	The Next Header values in the Fragment headers of different
144  *	fragments of the same original packet may differ.  Only the value
145  *	from the Offset zero fragment packet is used for reassembly.
146  *		-> should grab it from the first fragment only
147  *
148  * There is no explicit reason given in the RFC.  Historical reason maybe?
149  */
150 /*
151  * Fragment input
152  */
153 int
154 frag6_input(struct mbuf **mp, int *offp, int proto)
155 {
156 	struct mbuf *m = *mp, *t;
157 	struct ip6_hdr *ip6;
158 	struct ip6_frag *ip6f;
159 	struct ip6q *q6;
160 	struct ip6asfrag *af6, *ip6af, *af6dwn;
161 #ifdef IN6_IFSTAT_STRICT
162 	struct in6_ifaddr *ia;
163 #endif
164 	int offset = *offp, nxt, i, next;
165 	int first_frag = 0;
166 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
167 	struct ifnet *dstifp;
168 	u_int8_t ecn, ecn0;
169 #if 0
170 	char ip6buf[INET6_ADDRSTRLEN];
171 #endif
172 
173 	ip6 = mtod(m, struct ip6_hdr *);
174 #ifndef PULLDOWN_TEST
175 	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
176 	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
177 #else
178 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
179 	if (ip6f == NULL)
180 		return (IPPROTO_DONE);
181 #endif
182 
183 	dstifp = NULL;
184 #ifdef IN6_IFSTAT_STRICT
185 	/* find the destination interface of the packet. */
186 	if ((ia = ip6_getdstifaddr(m)) != NULL) {
187 		dstifp = ia->ia_ifp;
188 		ifa_free(&ia->ia_ifa);
189 	}
190 #else
191 	/* we are violating the spec, this is not the destination interface */
192 	if ((m->m_flags & M_PKTHDR) != 0)
193 		dstifp = m->m_pkthdr.rcvif;
194 #endif
195 
196 	/* jumbo payload can't contain a fragment header */
197 	if (ip6->ip6_plen == 0) {
198 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
199 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
200 		return IPPROTO_DONE;
201 	}
202 
203 	/*
204 	 * check whether fragment packet's fragment length is
205 	 * multiple of 8 octets.
206 	 * sizeof(struct ip6_frag) == 8
207 	 * sizeof(struct ip6_hdr) = 40
208 	 */
209 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
210 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
211 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
212 		    offsetof(struct ip6_hdr, ip6_plen));
213 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
214 		return IPPROTO_DONE;
215 	}
216 
217 	V_ip6stat.ip6s_fragments++;
218 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
219 
220 	/* offset now points to data portion */
221 	offset += sizeof(struct ip6_frag);
222 
223 	IP6Q_LOCK();
224 
225 	/*
226 	 * Enforce upper bound on number of fragments.
227 	 * If maxfrag is 0, never accept fragments.
228 	 * If maxfrag is -1, accept all fragments without limitation.
229 	 */
230 	if (V_ip6_maxfrags < 0)
231 		;
232 	else if (V_frag6_nfrags >= (u_int)V_ip6_maxfrags)
233 		goto dropfrag;
234 
235 	for (q6 = V_ip6q.ip6q_next; q6 != &V_ip6q; q6 = q6->ip6q_next)
236 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
237 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
238 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
239 #ifdef MAC
240 		    && mac_ip6q_match(m, q6)
241 #endif
242 		    )
243 			break;
244 
245 	if (q6 == &V_ip6q) {
246 		/*
247 		 * the first fragment to arrive, create a reassembly queue.
248 		 */
249 		first_frag = 1;
250 
251 		/*
252 		 * Enforce upper bound on number of fragmented packets
253 		 * for which we attempt reassembly;
254 		 * If maxfragpackets is 0, never accept fragments.
255 		 * If maxfragpackets is -1, accept all fragments without
256 		 * limitation.
257 		 */
258 		if (V_ip6_maxfragpackets < 0)
259 			;
260 		else if (V_frag6_nfragpackets >= (u_int)V_ip6_maxfragpackets)
261 			goto dropfrag;
262 		V_frag6_nfragpackets++;
263 		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
264 		    M_NOWAIT);
265 		if (q6 == NULL)
266 			goto dropfrag;
267 		bzero(q6, sizeof(*q6));
268 #ifdef MAC
269 		if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
270 			free(q6, M_FTABLE);
271 			goto dropfrag;
272 		}
273 		mac_ip6q_create(m, q6);
274 #endif
275 		frag6_insque(q6, &V_ip6q);
276 
277 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
278 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
279 #ifdef notyet
280 		q6->ip6q_nxtp	= (u_char *)nxtp;
281 #endif
282 		q6->ip6q_ident	= ip6f->ip6f_ident;
283 		q6->ip6q_ttl	= IPV6_FRAGTTL;
284 		q6->ip6q_src	= ip6->ip6_src;
285 		q6->ip6q_dst	= ip6->ip6_dst;
286 		q6->ip6q_ecn	=
287 		    (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
288 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
289 
290 		q6->ip6q_nfrag = 0;
291 	}
292 
293 	/*
294 	 * If it's the 1st fragment, record the length of the
295 	 * unfragmentable part and the next header of the fragment header.
296 	 */
297 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
298 	if (fragoff == 0) {
299 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
300 		    sizeof(struct ip6_frag);
301 		q6->ip6q_nxt = ip6f->ip6f_nxt;
302 	}
303 
304 	/*
305 	 * Check that the reassembled packet would not exceed 65535 bytes
306 	 * in size.
307 	 * If it would exceed, discard the fragment and return an ICMP error.
308 	 */
309 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
310 	if (q6->ip6q_unfrglen >= 0) {
311 		/* The 1st fragment has already arrived. */
312 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
313 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
314 			    offset - sizeof(struct ip6_frag) +
315 			    offsetof(struct ip6_frag, ip6f_offlg));
316 			IP6Q_UNLOCK();
317 			return (IPPROTO_DONE);
318 		}
319 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
320 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
321 		    offset - sizeof(struct ip6_frag) +
322 		    offsetof(struct ip6_frag, ip6f_offlg));
323 		IP6Q_UNLOCK();
324 		return (IPPROTO_DONE);
325 	}
326 	/*
327 	 * If it's the first fragment, do the above check for each
328 	 * fragment already stored in the reassembly queue.
329 	 */
330 	if (fragoff == 0) {
331 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
332 		     af6 = af6dwn) {
333 			af6dwn = af6->ip6af_down;
334 
335 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
336 			    IPV6_MAXPACKET) {
337 				struct mbuf *merr = IP6_REASS_MBUF(af6);
338 				struct ip6_hdr *ip6err;
339 				int erroff = af6->ip6af_offset;
340 
341 				/* dequeue the fragment. */
342 				frag6_deq(af6);
343 				free(af6, M_FTABLE);
344 
345 				/* adjust pointer. */
346 				ip6err = mtod(merr, struct ip6_hdr *);
347 
348 				/*
349 				 * Restore source and destination addresses
350 				 * in the erroneous IPv6 header.
351 				 */
352 				ip6err->ip6_src = q6->ip6q_src;
353 				ip6err->ip6_dst = q6->ip6q_dst;
354 
355 				icmp6_error(merr, ICMP6_PARAM_PROB,
356 				    ICMP6_PARAMPROB_HEADER,
357 				    erroff - sizeof(struct ip6_frag) +
358 				    offsetof(struct ip6_frag, ip6f_offlg));
359 			}
360 		}
361 	}
362 
363 	ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
364 	    M_NOWAIT);
365 	if (ip6af == NULL)
366 		goto dropfrag;
367 	bzero(ip6af, sizeof(*ip6af));
368 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
369 	ip6af->ip6af_off = fragoff;
370 	ip6af->ip6af_frglen = frgpartlen;
371 	ip6af->ip6af_offset = offset;
372 	IP6_REASS_MBUF(ip6af) = m;
373 
374 	if (first_frag) {
375 		af6 = (struct ip6asfrag *)q6;
376 		goto insert;
377 	}
378 
379 	/*
380 	 * Handle ECN by comparing this segment with the first one;
381 	 * if CE is set, do not lose CE.
382 	 * drop if CE and not-ECT are mixed for the same packet.
383 	 */
384 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
385 	ecn0 = q6->ip6q_ecn;
386 	if (ecn == IPTOS_ECN_CE) {
387 		if (ecn0 == IPTOS_ECN_NOTECT) {
388 			free(ip6af, M_FTABLE);
389 			goto dropfrag;
390 		}
391 		if (ecn0 != IPTOS_ECN_CE)
392 			q6->ip6q_ecn = IPTOS_ECN_CE;
393 	}
394 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
395 		free(ip6af, M_FTABLE);
396 		goto dropfrag;
397 	}
398 
399 	/*
400 	 * Find a segment which begins after this one does.
401 	 */
402 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
403 	     af6 = af6->ip6af_down)
404 		if (af6->ip6af_off > ip6af->ip6af_off)
405 			break;
406 
407 #if 0
408 	/*
409 	 * If there is a preceding segment, it may provide some of
410 	 * our data already.  If so, drop the data from the incoming
411 	 * segment.  If it provides all of our data, drop us.
412 	 */
413 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
414 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
415 			- ip6af->ip6af_off;
416 		if (i > 0) {
417 			if (i >= ip6af->ip6af_frglen)
418 				goto dropfrag;
419 			m_adj(IP6_REASS_MBUF(ip6af), i);
420 			ip6af->ip6af_off += i;
421 			ip6af->ip6af_frglen -= i;
422 		}
423 	}
424 
425 	/*
426 	 * While we overlap succeeding segments trim them or,
427 	 * if they are completely covered, dequeue them.
428 	 */
429 	while (af6 != (struct ip6asfrag *)q6 &&
430 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
431 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
432 		if (i < af6->ip6af_frglen) {
433 			af6->ip6af_frglen -= i;
434 			af6->ip6af_off += i;
435 			m_adj(IP6_REASS_MBUF(af6), i);
436 			break;
437 		}
438 		af6 = af6->ip6af_down;
439 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
440 		frag6_deq(af6->ip6af_up);
441 	}
442 #else
443 	/*
444 	 * If the incoming framgent overlaps some existing fragments in
445 	 * the reassembly queue, drop it, since it is dangerous to override
446 	 * existing fragments from a security point of view.
447 	 * We don't know which fragment is the bad guy - here we trust
448 	 * fragment that came in earlier, with no real reason.
449 	 *
450 	 * Note: due to changes after disabling this part, mbuf passed to
451 	 * m_adj() below now does not meet the requirement.
452 	 */
453 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
454 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
455 			- ip6af->ip6af_off;
456 		if (i > 0) {
457 #if 0				/* suppress the noisy log */
458 			log(LOG_ERR, "%d bytes of a fragment from %s "
459 			    "overlaps the previous fragment\n",
460 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
461 #endif
462 			free(ip6af, M_FTABLE);
463 			goto dropfrag;
464 		}
465 	}
466 	if (af6 != (struct ip6asfrag *)q6) {
467 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
468 		if (i > 0) {
469 #if 0				/* suppress the noisy log */
470 			log(LOG_ERR, "%d bytes of a fragment from %s "
471 			    "overlaps the succeeding fragment",
472 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
473 #endif
474 			free(ip6af, M_FTABLE);
475 			goto dropfrag;
476 		}
477 	}
478 #endif
479 
480 insert:
481 #ifdef MAC
482 	if (!first_frag)
483 		mac_ip6q_update(m, q6);
484 #endif
485 
486 	/*
487 	 * Stick new segment in its place;
488 	 * check for complete reassembly.
489 	 * Move to front of packet queue, as we are
490 	 * the most recently active fragmented packet.
491 	 */
492 	frag6_enq(ip6af, af6->ip6af_up);
493 	V_frag6_nfrags++;
494 	q6->ip6q_nfrag++;
495 #if 0 /* xxx */
496 	if (q6 != V_ip6q.ip6q_next) {
497 		frag6_remque(q6);
498 		frag6_insque(q6, &V_ip6q);
499 	}
500 #endif
501 	next = 0;
502 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
503 	     af6 = af6->ip6af_down) {
504 		if (af6->ip6af_off != next) {
505 			IP6Q_UNLOCK();
506 			return IPPROTO_DONE;
507 		}
508 		next += af6->ip6af_frglen;
509 	}
510 	if (af6->ip6af_up->ip6af_mff) {
511 		IP6Q_UNLOCK();
512 		return IPPROTO_DONE;
513 	}
514 
515 	/*
516 	 * Reassembly is complete; concatenate fragments.
517 	 */
518 	ip6af = q6->ip6q_down;
519 	t = m = IP6_REASS_MBUF(ip6af);
520 	af6 = ip6af->ip6af_down;
521 	frag6_deq(ip6af);
522 	while (af6 != (struct ip6asfrag *)q6) {
523 		af6dwn = af6->ip6af_down;
524 		frag6_deq(af6);
525 		while (t->m_next)
526 			t = t->m_next;
527 		t->m_next = IP6_REASS_MBUF(af6);
528 		m_adj(t->m_next, af6->ip6af_offset);
529 		free(af6, M_FTABLE);
530 		af6 = af6dwn;
531 	}
532 
533 	/* adjust offset to point where the original next header starts */
534 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
535 	free(ip6af, M_FTABLE);
536 	ip6 = mtod(m, struct ip6_hdr *);
537 	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
538 	if (q6->ip6q_ecn == IPTOS_ECN_CE)
539 		ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
540 	nxt = q6->ip6q_nxt;
541 #ifdef notyet
542 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
543 #endif
544 
545 	/* Delete frag6 header */
546 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
547 		/* This is the only possible case with !PULLDOWN_TEST */
548 		ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
549 		    offset);
550 		m->m_data += sizeof(struct ip6_frag);
551 		m->m_len -= sizeof(struct ip6_frag);
552 	} else {
553 		/* this comes with no copy if the boundary is on cluster */
554 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
555 			frag6_remque(q6);
556 			V_frag6_nfrags -= q6->ip6q_nfrag;
557 #ifdef MAC
558 			mac_ip6q_destroy(q6);
559 #endif
560 			free(q6, M_FTABLE);
561 			V_frag6_nfragpackets--;
562 			goto dropfrag;
563 		}
564 		m_adj(t, sizeof(struct ip6_frag));
565 		m_cat(m, t);
566 	}
567 
568 	/*
569 	 * Store NXT to the original.
570 	 */
571 	{
572 		char *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
573 		*prvnxtp = nxt;
574 	}
575 
576 	frag6_remque(q6);
577 	V_frag6_nfrags -= q6->ip6q_nfrag;
578 #ifdef MAC
579 	mac_ip6q_reassemble(q6, m);
580 	mac_ip6q_destroy(q6);
581 #endif
582 	free(q6, M_FTABLE);
583 	V_frag6_nfragpackets--;
584 
585 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
586 		int plen = 0;
587 		for (t = m; t; t = t->m_next)
588 			plen += t->m_len;
589 		m->m_pkthdr.len = plen;
590 	}
591 
592 	V_ip6stat.ip6s_reassembled++;
593 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
594 
595 	/*
596 	 * Tell launch routine the next header
597 	 */
598 
599 	*mp = m;
600 	*offp = offset;
601 
602 	IP6Q_UNLOCK();
603 	return nxt;
604 
605  dropfrag:
606 	IP6Q_UNLOCK();
607 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
608 	V_ip6stat.ip6s_fragdropped++;
609 	m_freem(m);
610 	return IPPROTO_DONE;
611 }
612 
613 /*
614  * Free a fragment reassembly header and all
615  * associated datagrams.
616  */
617 void
618 frag6_freef(struct ip6q *q6)
619 {
620 	struct ip6asfrag *af6, *down6;
621 
622 	IP6Q_LOCK_ASSERT();
623 
624 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
625 	     af6 = down6) {
626 		struct mbuf *m = IP6_REASS_MBUF(af6);
627 
628 		down6 = af6->ip6af_down;
629 		frag6_deq(af6);
630 
631 		/*
632 		 * Return ICMP time exceeded error for the 1st fragment.
633 		 * Just free other fragments.
634 		 */
635 		if (af6->ip6af_off == 0) {
636 			struct ip6_hdr *ip6;
637 
638 			/* adjust pointer */
639 			ip6 = mtod(m, struct ip6_hdr *);
640 
641 			/* restore source and destination addresses */
642 			ip6->ip6_src = q6->ip6q_src;
643 			ip6->ip6_dst = q6->ip6q_dst;
644 
645 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
646 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
647 		} else
648 			m_freem(m);
649 		free(af6, M_FTABLE);
650 	}
651 	frag6_remque(q6);
652 	V_frag6_nfrags -= q6->ip6q_nfrag;
653 #ifdef MAC
654 	mac_ip6q_destroy(q6);
655 #endif
656 	free(q6, M_FTABLE);
657 	V_frag6_nfragpackets--;
658 }
659 
660 /*
661  * Put an ip fragment on a reassembly chain.
662  * Like insque, but pointers in middle of structure.
663  */
664 void
665 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
666 {
667 
668 	IP6Q_LOCK_ASSERT();
669 
670 	af6->ip6af_up = up6;
671 	af6->ip6af_down = up6->ip6af_down;
672 	up6->ip6af_down->ip6af_up = af6;
673 	up6->ip6af_down = af6;
674 }
675 
676 /*
677  * To frag6_enq as remque is to insque.
678  */
679 void
680 frag6_deq(struct ip6asfrag *af6)
681 {
682 
683 	IP6Q_LOCK_ASSERT();
684 
685 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
686 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
687 }
688 
689 void
690 frag6_insque(struct ip6q *new, struct ip6q *old)
691 {
692 
693 	IP6Q_LOCK_ASSERT();
694 
695 	new->ip6q_prev = old;
696 	new->ip6q_next = old->ip6q_next;
697 	old->ip6q_next->ip6q_prev= new;
698 	old->ip6q_next = new;
699 }
700 
701 void
702 frag6_remque(struct ip6q *p6)
703 {
704 
705 	IP6Q_LOCK_ASSERT();
706 
707 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
708 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
709 }
710 
711 /*
712  * IPv6 reassembling timer processing;
713  * if a timer expires on a reassembly
714  * queue, discard it.
715  */
716 void
717 frag6_slowtimo(void)
718 {
719 	VNET_ITERATOR_DECL(vnet_iter);
720 	struct ip6q *q6;
721 
722 	VNET_LIST_RLOCK_NOSLEEP();
723 	IP6Q_LOCK();
724 	VNET_FOREACH(vnet_iter) {
725 		CURVNET_SET(vnet_iter);
726 		q6 = V_ip6q.ip6q_next;
727 		if (q6)
728 			while (q6 != &V_ip6q) {
729 				--q6->ip6q_ttl;
730 				q6 = q6->ip6q_next;
731 				if (q6->ip6q_prev->ip6q_ttl == 0) {
732 					V_ip6stat.ip6s_fragtimeout++;
733 					/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
734 					frag6_freef(q6->ip6q_prev);
735 				}
736 			}
737 		/*
738 		 * If we are over the maximum number of fragments
739 		 * (due to the limit being lowered), drain off
740 		 * enough to get down to the new limit.
741 		 */
742 		while (V_frag6_nfragpackets > (u_int)V_ip6_maxfragpackets &&
743 		    V_ip6q.ip6q_prev) {
744 			V_ip6stat.ip6s_fragoverflow++;
745 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
746 			frag6_freef(V_ip6q.ip6q_prev);
747 		}
748 		CURVNET_RESTORE();
749 	}
750 	IP6Q_UNLOCK();
751 	VNET_LIST_RUNLOCK_NOSLEEP();
752 }
753 
754 /*
755  * Drain off all datagram fragments.
756  */
757 void
758 frag6_drain(void)
759 {
760 	VNET_ITERATOR_DECL(vnet_iter);
761 
762 	VNET_LIST_RLOCK_NOSLEEP();
763 	if (IP6Q_TRYLOCK() == 0) {
764 		VNET_LIST_RUNLOCK_NOSLEEP();
765 		return;
766 	}
767 	VNET_FOREACH(vnet_iter) {
768 		CURVNET_SET(vnet_iter);
769 		while (V_ip6q.ip6q_next != &V_ip6q) {
770 			V_ip6stat.ip6s_fragdropped++;
771 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
772 			frag6_freef(V_ip6q.ip6q_next);
773 		}
774 		CURVNET_RESTORE();
775 	}
776 	IP6Q_UNLOCK();
777 	VNET_LIST_RUNLOCK_NOSLEEP();
778 }
779