xref: /openbsd-src/sys/netinet6/frag6.c (revision a28daedfc357b214be5c701aa8ba8adb29a7f1c2)
1 /*	$OpenBSD: frag6.c,v 1.27 2008/11/23 13:30:59 claudio Exp $	*/
2 /*	$KAME: frag6.c,v 1.40 2002/05/27 21:40:31 itojun Exp $	*/
3 
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/malloc.h>
36 #include <sys/mbuf.h>
37 #include <sys/domain.h>
38 #include <sys/protosw.h>
39 #include <sys/socket.h>
40 #include <sys/errno.h>
41 #include <sys/time.h>
42 #include <sys/kernel.h>
43 #include <sys/syslog.h>
44 
45 #include <net/if.h>
46 #include <net/route.h>
47 
48 #include <netinet/in.h>
49 #include <netinet/in_var.h>
50 #include <netinet/ip6.h>
51 #include <netinet6/ip6_var.h>
52 #include <netinet/icmp6.h>
53 #include <netinet/in_systm.h>	/* for ECN definitions */
54 #include <netinet/ip.h>		/* for ECN definitions */
55 
56 #include <dev/rndvar.h>
57 
58 /*
59  * Define it to get a correct behavior on per-interface statistics.
60  * You will need to perform an extra routing table lookup, per fragment,
61  * to do it.  This may, or may not be, a performance hit.
62  */
63 #define IN6_IFSTAT_STRICT
64 
65 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *);
66 static void frag6_deq(struct ip6asfrag *);
67 static void frag6_insque(struct ip6q *, struct ip6q *);
68 static void frag6_remque(struct ip6q *);
69 static void frag6_freef(struct ip6q *);
70 
71 static int ip6q_locked;
72 u_int frag6_nfragpackets;
73 u_int frag6_nfrags;
74 struct	ip6q ip6q;	/* ip6 reassemble queue */
75 
76 static __inline int ip6q_lock_try(void);
77 static __inline void ip6q_unlock(void);
78 
79 static __inline int
80 ip6q_lock_try()
81 {
82 	int s;
83 
84 	/* Use splvm() due to mbuf allocation. */
85 	s = splvm();
86 	if (ip6q_locked) {
87 		splx(s);
88 		return (0);
89 	}
90 	ip6q_locked = 1;
91 	splx(s);
92 	return (1);
93 }
94 
95 static __inline void
96 ip6q_unlock()
97 {
98 	int s;
99 
100 	s = splvm();
101 	ip6q_locked = 0;
102 	splx(s);
103 }
104 
105 #ifdef DIAGNOSTIC
106 #define	IP6Q_LOCK()							\
107 do {									\
108 	if (ip6q_lock_try() == 0) {					\
109 		printf("%s:%d: ip6q already locked\n", __FILE__, __LINE__); \
110 		panic("ip6q_lock");					\
111 	}								\
112 } while (0)
113 #define	IP6Q_LOCK_CHECK()						\
114 do {									\
115 	if (ip6q_locked == 0) {						\
116 		printf("%s:%d: ip6q lock not held\n", __FILE__, __LINE__); \
117 		panic("ip6q lock check");				\
118 	}								\
119 } while (0)
120 #else
121 #define	IP6Q_LOCK()		(void) ip6q_lock_try()
122 #define	IP6Q_LOCK_CHECK()	/* nothing */
123 #endif
124 
125 #define	IP6Q_UNLOCK()		ip6q_unlock()
126 
127 #ifndef offsetof		/* XXX */
128 #define	offsetof(type, member)	((size_t)(&((type *)0)->member))
129 #endif
130 
131 /*
132  * Initialise reassembly queue and fragment identifier.
133  */
134 void
135 frag6_init()
136 {
137 
138 	ip6q.ip6q_next = ip6q.ip6q_prev = &ip6q;
139 }
140 
141 /*
142  * In RFC2460, fragment and reassembly rule do not agree with each other,
143  * in terms of next header field handling in fragment header.
144  * While the sender will use the same value for all of the fragmented packets,
145  * receiver is suggested not to check the consistency.
146  *
147  * fragment rule (p20):
148  *	(2) A Fragment header containing:
149  *	The Next Header value that identifies the first header of
150  *	the Fragmentable Part of the original packet.
151  *		-> next header field is same for all fragments
152  *
153  * reassembly rule (p21):
154  *	The Next Header field of the last header of the Unfragmentable
155  *	Part is obtained from the Next Header field of the first
156  *	fragment's Fragment header.
157  *		-> should grab it from the first fragment only
158  *
159  * The following note also contradicts with fragment rule - noone is going to
160  * send different fragment with different next header field.
161  *
162  * additional note (p22):
163  *	The Next Header values in the Fragment headers of different
164  *	fragments of the same original packet may differ.  Only the value
165  *	from the Offset zero fragment packet is used for reassembly.
166  *		-> should grab it from the first fragment only
167  *
168  * There is no explicit reason given in the RFC.  Historical reason maybe?
169  */
170 /*
171  * Fragment input
172  */
173 int
174 frag6_input(struct mbuf **mp, int *offp, int proto)
175 {
176 	struct mbuf *m = *mp, *t;
177 	struct ip6_hdr *ip6;
178 	struct ip6_frag *ip6f;
179 	struct ip6q *q6;
180 	struct ip6asfrag *af6, *ip6af, *af6dwn;
181 	int offset = *offp, nxt, i, next;
182 	int first_frag = 0;
183 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
184 	struct ifnet *dstifp;
185 #ifdef IN6_IFSTAT_STRICT
186 	static struct route_in6 ro;
187 	struct sockaddr_in6 *dst;
188 #endif
189 	u_int8_t ecn, ecn0;
190 
191 	ip6 = mtod(m, struct ip6_hdr *);
192 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
193 	if (ip6f == NULL)
194 		return IPPROTO_DONE;
195 
196 	dstifp = NULL;
197 #ifdef IN6_IFSTAT_STRICT
198 	/* find the destination interface of the packet. */
199 	dst = (struct sockaddr_in6 *)&ro.ro_dst;
200 	if (ro.ro_rt && ((ro.ro_rt->rt_flags & RTF_UP) == 0
201 	  || !IN6_ARE_ADDR_EQUAL(&dst->sin6_addr, &ip6->ip6_dst))) {
202 		RTFREE(ro.ro_rt);
203 		ro.ro_rt = (struct rtentry *)0;
204 	}
205 	if (ro.ro_rt == NULL) {
206 		bzero(dst, sizeof(*dst));
207 		dst->sin6_family = AF_INET6;
208 		dst->sin6_len = sizeof(struct sockaddr_in6);
209 		dst->sin6_addr = ip6->ip6_dst;
210 	}
211 
212 	rtalloc_mpath((struct route *)&ro, &ip6->ip6_src.s6_addr32[0], 0);
213 
214 	if (ro.ro_rt != NULL && ro.ro_rt->rt_ifa != NULL)
215 		dstifp = ((struct in6_ifaddr *)ro.ro_rt->rt_ifa)->ia_ifp;
216 #else
217 	/* we are violating the spec, this is not the destination interface */
218 	if ((m->m_flags & M_PKTHDR) != 0)
219 		dstifp = m->m_pkthdr.rcvif;
220 #endif
221 
222 	/* jumbo payload can't contain a fragment header */
223 	if (ip6->ip6_plen == 0) {
224 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
225 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
226 		return IPPROTO_DONE;
227 	}
228 
229 	/*
230 	 * check whether fragment packet's fragment length is
231 	 * multiple of 8 octets.
232 	 * sizeof(struct ip6_frag) == 8
233 	 * sizeof(struct ip6_hdr) = 40
234 	 */
235 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
236 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
237 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
238 		    offsetof(struct ip6_hdr, ip6_plen));
239 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
240 		return IPPROTO_DONE;
241 	}
242 
243 	ip6stat.ip6s_fragments++;
244 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
245 
246 	/* offset now points to data portion */
247 	offset += sizeof(struct ip6_frag);
248 
249 	IP6Q_LOCK();
250 
251 	/*
252 	 * Enforce upper bound on number of fragments.
253 	 * If maxfrag is 0, never accept fragments.
254 	 * If maxfrag is -1, accept all fragments without limitation.
255 	 */
256 	if (ip6_maxfrags < 0)
257 		;
258 	else if (frag6_nfrags >= (u_int)ip6_maxfrags)
259 		goto dropfrag;
260 
261 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next)
262 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
263 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
264 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
265 			break;
266 
267 	if (q6 == &ip6q) {
268 		/*
269 		 * the first fragment to arrive, create a reassembly queue.
270 		 */
271 		first_frag = 1;
272 
273 		/*
274 		 * Enforce upper bound on number of fragmented packets
275 		 * for which we attempt reassembly;
276 		 * If maxfragpackets is 0, never accept fragments.
277 		 * If maxfragpackets is -1, accept all fragments without
278 		 * limitation.
279 		 */
280 		if (ip6_maxfragpackets < 0)
281 			;
282 		else if (frag6_nfragpackets >= (u_int)ip6_maxfragpackets)
283 			goto dropfrag;
284 		frag6_nfragpackets++;
285 		q6 = malloc(sizeof(*q6), M_FTABLE, M_DONTWAIT | M_ZERO);
286 		if (q6 == NULL)
287 			goto dropfrag;
288 
289 		frag6_insque(q6, &ip6q);
290 
291 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
292 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
293 #ifdef notyet
294 		q6->ip6q_nxtp	= (u_char *)nxtp;
295 #endif
296 		q6->ip6q_ident	= ip6f->ip6f_ident;
297 		q6->ip6q_arrive = 0; /* Is it used anywhere? */
298 		q6->ip6q_ttl 	= IPV6_FRAGTTL;
299 		q6->ip6q_src	= ip6->ip6_src;
300 		q6->ip6q_dst	= ip6->ip6_dst;
301 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
302 
303 		q6->ip6q_nfrag = 0;
304 	}
305 
306 	/*
307 	 * If it's the 1st fragment, record the length of the
308 	 * unfragmentable part and the next header of the fragment header.
309 	 */
310 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
311 	if (fragoff == 0) {
312 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
313 		    sizeof(struct ip6_frag);
314 		q6->ip6q_nxt = ip6f->ip6f_nxt;
315 	}
316 
317 	/*
318 	 * Check that the reassembled packet would not exceed 65535 bytes
319 	 * in size.
320 	 * If it would exceed, discard the fragment and return an ICMP error.
321 	 */
322 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
323 	if (q6->ip6q_unfrglen >= 0) {
324 		/* The 1st fragment has already arrived. */
325 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
326 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
327 			    offset - sizeof(struct ip6_frag) +
328 			    offsetof(struct ip6_frag, ip6f_offlg));
329 			IP6Q_UNLOCK();
330 			return (IPPROTO_DONE);
331 		}
332 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
333 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
334 			    offset - sizeof(struct ip6_frag) +
335 				offsetof(struct ip6_frag, ip6f_offlg));
336 		IP6Q_UNLOCK();
337 		return (IPPROTO_DONE);
338 	}
339 	/*
340 	 * If it's the first fragment, do the above check for each
341 	 * fragment already stored in the reassembly queue.
342 	 */
343 	if (fragoff == 0) {
344 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
345 		     af6 = af6dwn) {
346 			af6dwn = af6->ip6af_down;
347 
348 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
349 			    IPV6_MAXPACKET) {
350 				struct mbuf *merr = IP6_REASS_MBUF(af6);
351 				struct ip6_hdr *ip6err;
352 				int erroff = af6->ip6af_offset;
353 
354 				/* dequeue the fragment. */
355 				frag6_deq(af6);
356 				free(af6, M_FTABLE);
357 
358 				/* adjust pointer. */
359 				ip6err = mtod(merr, struct ip6_hdr *);
360 
361 				/*
362 				 * Restore source and destination addresses
363 				 * in the erroneous IPv6 header.
364 				 */
365 				ip6err->ip6_src = q6->ip6q_src;
366 				ip6err->ip6_dst = q6->ip6q_dst;
367 
368 				icmp6_error(merr, ICMP6_PARAM_PROB,
369 				    ICMP6_PARAMPROB_HEADER,
370 				    erroff - sizeof(struct ip6_frag) +
371 				    offsetof(struct ip6_frag, ip6f_offlg));
372 			}
373 		}
374 	}
375 
376 	ip6af = malloc(sizeof(*ip6af), M_FTABLE, M_DONTWAIT | M_ZERO);
377 	if (ip6af == NULL)
378 		goto dropfrag;
379 	ip6af->ip6af_head = ip6->ip6_flow;
380 	ip6af->ip6af_len = ip6->ip6_plen;
381 	ip6af->ip6af_nxt = ip6->ip6_nxt;
382 	ip6af->ip6af_hlim = ip6->ip6_hlim;
383 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
384 	ip6af->ip6af_off = fragoff;
385 	ip6af->ip6af_frglen = frgpartlen;
386 	ip6af->ip6af_offset = offset;
387 	IP6_REASS_MBUF(ip6af) = m;
388 
389 	if (first_frag) {
390 		af6 = (struct ip6asfrag *)q6;
391 		goto insert;
392 	}
393 
394 	/*
395 	 * Handle ECN by comparing this segment with the first one;
396 	 * if CE is set, do not lose CE.
397 	 * drop if CE and not-ECT are mixed for the same packet.
398 	 */
399 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
400 	ecn0 = (ntohl(q6->ip6q_down->ip6af_head) >> 20) & IPTOS_ECN_MASK;
401 	if (ecn == IPTOS_ECN_CE) {
402 		if (ecn0 == IPTOS_ECN_NOTECT) {
403 			free(ip6af, M_FTABLE);
404 			goto dropfrag;
405 		}
406 		if (ecn0 != IPTOS_ECN_CE)
407 			q6->ip6q_down->ip6af_head |= htonl(IPTOS_ECN_CE << 20);
408 	}
409 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
410 		free(ip6af, M_FTABLE);
411 		goto dropfrag;
412 	}
413 
414 	/*
415 	 * Find a segment which begins after this one does.
416 	 */
417 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
418 	     af6 = af6->ip6af_down)
419 		if (af6->ip6af_off > ip6af->ip6af_off)
420 			break;
421 
422 #if 0
423 	/*
424 	 * If there is a preceding segment, it may provide some of
425 	 * our data already.  If so, drop the data from the incoming
426 	 * segment.  If it provides all of our data, drop us.
427 	 */
428 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
429 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
430 			- ip6af->ip6af_off;
431 		if (i > 0) {
432 			if (i >= ip6af->ip6af_frglen)
433 				goto dropfrag;
434 			m_adj(IP6_REASS_MBUF(ip6af), i);
435 			ip6af->ip6af_off += i;
436 			ip6af->ip6af_frglen -= i;
437 		}
438 	}
439 
440 	/*
441 	 * While we overlap succeeding segments trim them or,
442 	 * if they are completely covered, dequeue them.
443 	 */
444 	while (af6 != (struct ip6asfrag *)q6 &&
445 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
446 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
447 		if (i < af6->ip6af_frglen) {
448 			af6->ip6af_frglen -= i;
449 			af6->ip6af_off += i;
450 			m_adj(IP6_REASS_MBUF(af6), i);
451 			break;
452 		}
453 		af6 = af6->ip6af_down;
454 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
455 		frag6_deq(af6->ip6af_up);
456 	}
457 #else
458 	/*
459 	 * If the incoming fragment overlaps some existing fragments in
460 	 * the reassembly queue, drop it, since it is dangerous to override
461 	 * existing fragments from a security point of view.
462 	 * We don't know which fragment is the bad guy - here we trust
463 	 * fragment that came in earlier, with no real reason.
464 	 */
465 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
466 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
467 			- ip6af->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 previous fragment\n",
472 			    i, ip6_sprintf(&q6->ip6q_src));
473 #endif
474 			free(ip6af, M_FTABLE);
475 			goto dropfrag;
476 		}
477 	}
478 	if (af6 != (struct ip6asfrag *)q6) {
479 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
480 		if (i > 0) {
481 #if 0				/* suppress the noisy log */
482 			log(LOG_ERR, "%d bytes of a fragment from %s "
483 			    "overlaps the succeeding fragment",
484 			    i, ip6_sprintf(&q6->ip6q_src));
485 #endif
486 			free(ip6af, M_FTABLE);
487 			goto dropfrag;
488 		}
489 	}
490 #endif
491 
492 insert:
493 
494 	/*
495 	 * Stick new segment in its place;
496 	 * check for complete reassembly.
497 	 * Move to front of packet queue, as we are
498 	 * the most recently active fragmented packet.
499 	 */
500 	frag6_enq(ip6af, af6->ip6af_up);
501 	frag6_nfrags++;
502 	q6->ip6q_nfrag++;
503 #if 0 /* xxx */
504 	if (q6 != ip6q.ip6q_next) {
505 		frag6_remque(q6);
506 		frag6_insque(q6, &ip6q);
507 	}
508 #endif
509 	next = 0;
510 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
511 	     af6 = af6->ip6af_down) {
512 		if (af6->ip6af_off != next) {
513 			IP6Q_UNLOCK();
514 			return IPPROTO_DONE;
515 		}
516 		next += af6->ip6af_frglen;
517 	}
518 	if (af6->ip6af_up->ip6af_mff) {
519 		IP6Q_UNLOCK();
520 		return IPPROTO_DONE;
521 	}
522 
523 	/*
524 	 * Reassembly is complete; concatenate fragments.
525 	 */
526 	ip6af = q6->ip6q_down;
527 	t = m = IP6_REASS_MBUF(ip6af);
528 	af6 = ip6af->ip6af_down;
529 	frag6_deq(ip6af);
530 	while (af6 != (struct ip6asfrag *)q6) {
531 		af6dwn = af6->ip6af_down;
532 		frag6_deq(af6);
533 		while (t->m_next)
534 			t = t->m_next;
535 		t->m_next = IP6_REASS_MBUF(af6);
536 		m_adj(t->m_next, af6->ip6af_offset);
537 		free(af6, M_FTABLE);
538 		af6 = af6dwn;
539 	}
540 
541 	/* adjust offset to point where the original next header starts */
542 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
543 	free(ip6af, M_FTABLE);
544 	ip6 = mtod(m, struct ip6_hdr *);
545 	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
546 	ip6->ip6_src = q6->ip6q_src;
547 	ip6->ip6_dst = q6->ip6q_dst;
548 	nxt = q6->ip6q_nxt;
549 #ifdef notyet
550 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
551 #endif
552 
553 	/*
554 	 * Delete frag6 header with as a few cost as possible.
555 	 */
556 	if (offset < m->m_len) {
557 		ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
558 			offset);
559 		m->m_data += sizeof(struct ip6_frag);
560 		m->m_len -= sizeof(struct ip6_frag);
561 	} else {
562 		/* this comes with no copy if the boundary is on cluster */
563 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
564 			frag6_remque(q6);
565 			frag6_nfrags -= q6->ip6q_nfrag;
566 			free(q6, M_FTABLE);
567 			frag6_nfragpackets--;
568 			goto dropfrag;
569 		}
570 		m_adj(t, sizeof(struct ip6_frag));
571 		m_cat(m, t);
572 	}
573 
574 	/*
575 	 * Store NXT to the original.
576 	 */
577 	{
578 		u_int8_t *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
579 		*prvnxtp = nxt;
580 	}
581 
582 	frag6_remque(q6);
583 	frag6_nfrags -= q6->ip6q_nfrag;
584 	free(q6, M_FTABLE);
585 	frag6_nfragpackets--;
586 
587 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
588 		int plen = 0;
589 		for (t = m; t; t = t->m_next)
590 			plen += t->m_len;
591 		m->m_pkthdr.len = plen;
592 	}
593 
594 	ip6stat.ip6s_reassembled++;
595 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
596 
597 	/*
598 	 * Tell launch routine the next header
599 	 */
600 
601 	*mp = m;
602 	*offp = offset;
603 
604 	IP6Q_UNLOCK();
605 	return nxt;
606 
607  dropfrag:
608 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
609 	ip6stat.ip6s_fragdropped++;
610 	m_freem(m);
611 	IP6Q_UNLOCK();
612 	return IPPROTO_DONE;
613 }
614 
615 /*
616  * Free a fragment reassembly header and all
617  * associated datagrams.
618  */
619 void
620 frag6_freef(struct ip6q *q6)
621 {
622 	struct ip6asfrag *af6, *down6;
623 
624 	IP6Q_LOCK_CHECK();
625 
626 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
627 	     af6 = down6) {
628 		struct mbuf *m = IP6_REASS_MBUF(af6);
629 
630 		down6 = af6->ip6af_down;
631 		frag6_deq(af6);
632 
633 		/*
634 		 * Return ICMP time exceeded error for the 1st fragment.
635 		 * Just free other fragments.
636 		 */
637 		if (af6->ip6af_off == 0) {
638 			struct ip6_hdr *ip6;
639 
640 			/* adjust pointer */
641 			ip6 = mtod(m, struct ip6_hdr *);
642 
643 			/* restoure source and destination addresses */
644 			ip6->ip6_src = q6->ip6q_src;
645 			ip6->ip6_dst = q6->ip6q_dst;
646 
647 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
648 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
649 		} else
650 			m_freem(m);
651 		free(af6, M_FTABLE);
652 	}
653 	frag6_remque(q6);
654 	frag6_nfrags -= q6->ip6q_nfrag;
655 	free(q6, M_FTABLE);
656 	frag6_nfragpackets--;
657 }
658 
659 /*
660  * Put an ip fragment on a reassembly chain.
661  * Like insque, but pointers in middle of structure.
662  */
663 void
664 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
665 {
666 
667 	IP6Q_LOCK_CHECK();
668 
669 	af6->ip6af_up = up6;
670 	af6->ip6af_down = up6->ip6af_down;
671 	up6->ip6af_down->ip6af_up = af6;
672 	up6->ip6af_down = af6;
673 }
674 
675 /*
676  * To frag6_enq as remque is to insque.
677  */
678 void
679 frag6_deq(struct ip6asfrag *af6)
680 {
681 
682 	IP6Q_LOCK_CHECK();
683 
684 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
685 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
686 }
687 
688 void
689 frag6_insque(struct ip6q *new, struct ip6q *old)
690 {
691 
692 	IP6Q_LOCK_CHECK();
693 
694 	new->ip6q_prev = old;
695 	new->ip6q_next = old->ip6q_next;
696 	old->ip6q_next->ip6q_prev= new;
697 	old->ip6q_next = new;
698 }
699 
700 void
701 frag6_remque(struct ip6q *p6)
702 {
703 
704 	IP6Q_LOCK_CHECK();
705 
706 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
707 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
708 }
709 
710 /*
711  * IPv6 reassembling timer processing;
712  * if a timer expires on a reassembly
713  * queue, discard it.
714  */
715 void
716 frag6_slowtimo()
717 {
718 	struct ip6q *q6;
719 	int s = splsoftnet();
720 
721 	IP6Q_LOCK();
722 	q6 = ip6q.ip6q_next;
723 	if (q6)
724 		while (q6 != &ip6q) {
725 			--q6->ip6q_ttl;
726 			q6 = q6->ip6q_next;
727 			if (q6->ip6q_prev->ip6q_ttl == 0) {
728 				ip6stat.ip6s_fragtimeout++;
729 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
730 				frag6_freef(q6->ip6q_prev);
731 			}
732 		}
733 	/*
734 	 * If we are over the maximum number of fragments
735 	 * (due to the limit being lowered), drain off
736 	 * enough to get down to the new limit.
737 	 */
738 	while (frag6_nfragpackets > (u_int)ip6_maxfragpackets &&
739 	    ip6q.ip6q_prev) {
740 		ip6stat.ip6s_fragoverflow++;
741 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
742 		frag6_freef(ip6q.ip6q_prev);
743 	}
744 	IP6Q_UNLOCK();
745 
746 #if 0
747 	/*
748 	 * Routing changes might produce a better route than we last used;
749 	 * make sure we notice eventually, even if forwarding only for one
750 	 * destination and the cache is never replaced.
751 	 */
752 	if (ip6_forward_rt.ro_rt) {
753 		RTFREE(ip6_forward_rt.ro_rt);
754 		ip6_forward_rt.ro_rt = 0;
755 	}
756 	if (ipsrcchk_rt.ro_rt) {
757 		RTFREE(ipsrcchk_rt.ro_rt);
758 		ipsrcchk_rt.ro_rt = 0;
759 	}
760 #endif
761 
762 	splx(s);
763 }
764 
765 /*
766  * Drain off all datagram fragments.
767  */
768 void
769 frag6_drain()
770 {
771 
772 	if (ip6q_lock_try() == 0)
773 		return;
774 	while (ip6q.ip6q_next != &ip6q) {
775 		ip6stat.ip6s_fragdropped++;
776 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
777 		frag6_freef(ip6q.ip6q_next);
778 	}
779 	IP6Q_UNLOCK();
780 }
781