xref: /openbsd-src/sys/netinet6/frag6.c (revision 850e275390052b330d93020bf619a739a3c277ac)
1 /*	$OpenBSD: frag6.c,v 1.26 2008/06/11 19:00:50 mcbride 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
201 	 && ((ro.ro_rt->rt_flags & RTF_UP) == 0
202 	  || !IN6_ARE_ADDR_EQUAL(&dst->sin6_addr, &ip6->ip6_dst))) {
203 		RTFREE(ro.ro_rt);
204 		ro.ro_rt = (struct rtentry *)0;
205 	}
206 	if (ro.ro_rt == NULL) {
207 		bzero(dst, sizeof(*dst));
208 		dst->sin6_family = AF_INET6;
209 		dst->sin6_len = sizeof(struct sockaddr_in6);
210 		dst->sin6_addr = ip6->ip6_dst;
211 	}
212 
213 	rtalloc_mpath((struct route *)&ro, &ip6->ip6_src.s6_addr32[0], 0);
214 
215 	if (ro.ro_rt != NULL && ro.ro_rt->rt_ifa != NULL)
216 		dstifp = ((struct in6_ifaddr *)ro.ro_rt->rt_ifa)->ia_ifp;
217 #else
218 	/* we are violating the spec, this is not the destination interface */
219 	if ((m->m_flags & M_PKTHDR) != 0)
220 		dstifp = m->m_pkthdr.rcvif;
221 #endif
222 
223 	/* jumbo payload can't contain a fragment header */
224 	if (ip6->ip6_plen == 0) {
225 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
226 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
227 		return IPPROTO_DONE;
228 	}
229 
230 	/*
231 	 * check whether fragment packet's fragment length is
232 	 * multiple of 8 octets.
233 	 * sizeof(struct ip6_frag) == 8
234 	 * sizeof(struct ip6_hdr) = 40
235 	 */
236 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
237 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
238 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
239 		    offsetof(struct ip6_hdr, ip6_plen));
240 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
241 		return IPPROTO_DONE;
242 	}
243 
244 	ip6stat.ip6s_fragments++;
245 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
246 
247 	/* offset now points to data portion */
248 	offset += sizeof(struct ip6_frag);
249 
250 	IP6Q_LOCK();
251 
252 	/*
253 	 * Enforce upper bound on number of fragments.
254 	 * If maxfrag is 0, never accept fragments.
255 	 * If maxfrag is -1, accept all fragments without limitation.
256 	 */
257 	if (ip6_maxfrags < 0)
258 		;
259 	else if (frag6_nfrags >= (u_int)ip6_maxfrags)
260 		goto dropfrag;
261 
262 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next)
263 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
264 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
265 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
266 			break;
267 
268 	if (q6 == &ip6q) {
269 		/*
270 		 * the first fragment to arrive, create a reassembly queue.
271 		 */
272 		first_frag = 1;
273 
274 		/*
275 		 * Enforce upper bound on number of fragmented packets
276 		 * for which we attempt reassembly;
277 		 * If maxfragpackets is 0, never accept fragments.
278 		 * If maxfragpackets is -1, accept all fragments without
279 		 * limitation.
280 		 */
281 		if (ip6_maxfragpackets < 0)
282 			;
283 		else if (frag6_nfragpackets >= (u_int)ip6_maxfragpackets)
284 			goto dropfrag;
285 		frag6_nfragpackets++;
286 		q6 = malloc(sizeof(*q6), M_FTABLE, M_DONTWAIT | M_ZERO);
287 		if (q6 == NULL)
288 			goto dropfrag;
289 
290 		frag6_insque(q6, &ip6q);
291 
292 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
293 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
294 #ifdef notyet
295 		q6->ip6q_nxtp	= (u_char *)nxtp;
296 #endif
297 		q6->ip6q_ident	= ip6f->ip6f_ident;
298 		q6->ip6q_arrive = 0; /* Is it used anywhere? */
299 		q6->ip6q_ttl 	= IPV6_FRAGTTL;
300 		q6->ip6q_src	= ip6->ip6_src;
301 		q6->ip6q_dst	= ip6->ip6_dst;
302 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
303 
304 		q6->ip6q_nfrag = 0;
305 	}
306 
307 	/*
308 	 * If it's the 1st fragment, record the length of the
309 	 * unfragmentable part and the next header of the fragment header.
310 	 */
311 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
312 	if (fragoff == 0) {
313 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
314 		    sizeof(struct ip6_frag);
315 		q6->ip6q_nxt = ip6f->ip6f_nxt;
316 	}
317 
318 	/*
319 	 * Check that the reassembled packet would not exceed 65535 bytes
320 	 * in size.
321 	 * If it would exceed, discard the fragment and return an ICMP error.
322 	 */
323 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
324 	if (q6->ip6q_unfrglen >= 0) {
325 		/* The 1st fragment has already arrived. */
326 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
327 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
328 			    offset - sizeof(struct ip6_frag) +
329 			    offsetof(struct ip6_frag, ip6f_offlg));
330 			IP6Q_UNLOCK();
331 			return (IPPROTO_DONE);
332 		}
333 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
334 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
335 			    offset - sizeof(struct ip6_frag) +
336 				offsetof(struct ip6_frag, ip6f_offlg));
337 		IP6Q_UNLOCK();
338 		return (IPPROTO_DONE);
339 	}
340 	/*
341 	 * If it's the first fragment, do the above check for each
342 	 * fragment already stored in the reassembly queue.
343 	 */
344 	if (fragoff == 0) {
345 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
346 		     af6 = af6dwn) {
347 			af6dwn = af6->ip6af_down;
348 
349 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
350 			    IPV6_MAXPACKET) {
351 				struct mbuf *merr = IP6_REASS_MBUF(af6);
352 				struct ip6_hdr *ip6err;
353 				int erroff = af6->ip6af_offset;
354 
355 				/* dequeue the fragment. */
356 				frag6_deq(af6);
357 				free(af6, M_FTABLE);
358 
359 				/* adjust pointer. */
360 				ip6err = mtod(merr, struct ip6_hdr *);
361 
362 				/*
363 				 * Restore source and destination addresses
364 				 * in the erroneous IPv6 header.
365 				 */
366 				ip6err->ip6_src = q6->ip6q_src;
367 				ip6err->ip6_dst = q6->ip6q_dst;
368 
369 				icmp6_error(merr, ICMP6_PARAM_PROB,
370 				    ICMP6_PARAMPROB_HEADER,
371 				    erroff - sizeof(struct ip6_frag) +
372 				    offsetof(struct ip6_frag, ip6f_offlg));
373 			}
374 		}
375 	}
376 
377 	ip6af = malloc(sizeof(*ip6af), M_FTABLE, M_DONTWAIT | M_ZERO);
378 	if (ip6af == NULL)
379 		goto dropfrag;
380 	ip6af->ip6af_head = ip6->ip6_flow;
381 	ip6af->ip6af_len = ip6->ip6_plen;
382 	ip6af->ip6af_nxt = ip6->ip6_nxt;
383 	ip6af->ip6af_hlim = ip6->ip6_hlim;
384 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
385 	ip6af->ip6af_off = fragoff;
386 	ip6af->ip6af_frglen = frgpartlen;
387 	ip6af->ip6af_offset = offset;
388 	IP6_REASS_MBUF(ip6af) = m;
389 
390 	if (first_frag) {
391 		af6 = (struct ip6asfrag *)q6;
392 		goto insert;
393 	}
394 
395 	/*
396 	 * Handle ECN by comparing this segment with the first one;
397 	 * if CE is set, do not lose CE.
398 	 * drop if CE and not-ECT are mixed for the same packet.
399 	 */
400 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
401 	ecn0 = (ntohl(q6->ip6q_down->ip6af_head) >> 20) & IPTOS_ECN_MASK;
402 	if (ecn == IPTOS_ECN_CE) {
403 		if (ecn0 == IPTOS_ECN_NOTECT) {
404 			free(ip6af, M_FTABLE);
405 			goto dropfrag;
406 		}
407 		if (ecn0 != IPTOS_ECN_CE)
408 			q6->ip6q_down->ip6af_head |= htonl(IPTOS_ECN_CE << 20);
409 	}
410 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
411 		free(ip6af, M_FTABLE);
412 		goto dropfrag;
413 	}
414 
415 	/*
416 	 * Find a segment which begins after this one does.
417 	 */
418 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
419 	     af6 = af6->ip6af_down)
420 		if (af6->ip6af_off > ip6af->ip6af_off)
421 			break;
422 
423 #if 0
424 	/*
425 	 * If there is a preceding segment, it may provide some of
426 	 * our data already.  If so, drop the data from the incoming
427 	 * segment.  If it provides all of our data, drop us.
428 	 */
429 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
430 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
431 			- ip6af->ip6af_off;
432 		if (i > 0) {
433 			if (i >= ip6af->ip6af_frglen)
434 				goto dropfrag;
435 			m_adj(IP6_REASS_MBUF(ip6af), i);
436 			ip6af->ip6af_off += i;
437 			ip6af->ip6af_frglen -= i;
438 		}
439 	}
440 
441 	/*
442 	 * While we overlap succeeding segments trim them or,
443 	 * if they are completely covered, dequeue them.
444 	 */
445 	while (af6 != (struct ip6asfrag *)q6 &&
446 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
447 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
448 		if (i < af6->ip6af_frglen) {
449 			af6->ip6af_frglen -= i;
450 			af6->ip6af_off += i;
451 			m_adj(IP6_REASS_MBUF(af6), i);
452 			break;
453 		}
454 		af6 = af6->ip6af_down;
455 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
456 		frag6_deq(af6->ip6af_up);
457 	}
458 #else
459 	/*
460 	 * If the incoming fragment overlaps some existing fragments in
461 	 * the reassembly queue, drop it, since it is dangerous to override
462 	 * existing fragments from a security point of view.
463 	 * We don't know which fragment is the bad guy - here we trust
464 	 * fragment that came in earlier, with no real reason.
465 	 */
466 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
467 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
468 			- ip6af->ip6af_off;
469 		if (i > 0) {
470 #if 0				/* suppress the noisy log */
471 			log(LOG_ERR, "%d bytes of a fragment from %s "
472 			    "overlaps the previous fragment\n",
473 			    i, ip6_sprintf(&q6->ip6q_src));
474 #endif
475 			free(ip6af, M_FTABLE);
476 			goto dropfrag;
477 		}
478 	}
479 	if (af6 != (struct ip6asfrag *)q6) {
480 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
481 		if (i > 0) {
482 #if 0				/* suppress the noisy log */
483 			log(LOG_ERR, "%d bytes of a fragment from %s "
484 			    "overlaps the succeeding fragment",
485 			    i, ip6_sprintf(&q6->ip6q_src));
486 #endif
487 			free(ip6af, M_FTABLE);
488 			goto dropfrag;
489 		}
490 	}
491 #endif
492 
493 insert:
494 
495 	/*
496 	 * Stick new segment in its place;
497 	 * check for complete reassembly.
498 	 * Move to front of packet queue, as we are
499 	 * the most recently active fragmented packet.
500 	 */
501 	frag6_enq(ip6af, af6->ip6af_up);
502 	frag6_nfrags++;
503 	q6->ip6q_nfrag++;
504 #if 0 /* xxx */
505 	if (q6 != ip6q.ip6q_next) {
506 		frag6_remque(q6);
507 		frag6_insque(q6, &ip6q);
508 	}
509 #endif
510 	next = 0;
511 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
512 	     af6 = af6->ip6af_down) {
513 		if (af6->ip6af_off != next) {
514 			IP6Q_UNLOCK();
515 			return IPPROTO_DONE;
516 		}
517 		next += af6->ip6af_frglen;
518 	}
519 	if (af6->ip6af_up->ip6af_mff) {
520 		IP6Q_UNLOCK();
521 		return IPPROTO_DONE;
522 	}
523 
524 	/*
525 	 * Reassembly is complete; concatenate fragments.
526 	 */
527 	ip6af = q6->ip6q_down;
528 	t = m = IP6_REASS_MBUF(ip6af);
529 	af6 = ip6af->ip6af_down;
530 	frag6_deq(ip6af);
531 	while (af6 != (struct ip6asfrag *)q6) {
532 		af6dwn = af6->ip6af_down;
533 		frag6_deq(af6);
534 		while (t->m_next)
535 			t = t->m_next;
536 		t->m_next = IP6_REASS_MBUF(af6);
537 		m_adj(t->m_next, af6->ip6af_offset);
538 		free(af6, M_FTABLE);
539 		af6 = af6dwn;
540 	}
541 
542 	/* adjust offset to point where the original next header starts */
543 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
544 	free(ip6af, M_FTABLE);
545 	ip6 = mtod(m, struct ip6_hdr *);
546 	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
547 	ip6->ip6_src = q6->ip6q_src;
548 	ip6->ip6_dst = q6->ip6q_dst;
549 	nxt = q6->ip6q_nxt;
550 #ifdef notyet
551 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
552 #endif
553 
554 	/*
555 	 * Delete frag6 header with as a few cost as possible.
556 	 */
557 	if (offset < m->m_len) {
558 		ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
559 			offset);
560 		m->m_data += sizeof(struct ip6_frag);
561 		m->m_len -= sizeof(struct ip6_frag);
562 	} else {
563 		/* this comes with no copy if the boundary is on cluster */
564 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
565 			frag6_remque(q6);
566 			frag6_nfrags -= q6->ip6q_nfrag;
567 			free(q6, M_FTABLE);
568 			frag6_nfragpackets--;
569 			goto dropfrag;
570 		}
571 		m_adj(t, sizeof(struct ip6_frag));
572 		m_cat(m, t);
573 	}
574 
575 	/*
576 	 * Store NXT to the original.
577 	 */
578 	{
579 		u_int8_t *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
580 		*prvnxtp = nxt;
581 	}
582 
583 	frag6_remque(q6);
584 	frag6_nfrags -= q6->ip6q_nfrag;
585 	free(q6, M_FTABLE);
586 	frag6_nfragpackets--;
587 
588 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
589 		int plen = 0;
590 		for (t = m; t; t = t->m_next)
591 			plen += t->m_len;
592 		m->m_pkthdr.len = plen;
593 	}
594 
595 	ip6stat.ip6s_reassembled++;
596 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
597 
598 	/*
599 	 * Tell launch routine the next header
600 	 */
601 
602 	*mp = m;
603 	*offp = offset;
604 
605 	IP6Q_UNLOCK();
606 	return nxt;
607 
608  dropfrag:
609 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
610 	ip6stat.ip6s_fragdropped++;
611 	m_freem(m);
612 	IP6Q_UNLOCK();
613 	return IPPROTO_DONE;
614 }
615 
616 /*
617  * Free a fragment reassembly header and all
618  * associated datagrams.
619  */
620 void
621 frag6_freef(struct ip6q *q6)
622 {
623 	struct ip6asfrag *af6, *down6;
624 
625 	IP6Q_LOCK_CHECK();
626 
627 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
628 	     af6 = down6) {
629 		struct mbuf *m = IP6_REASS_MBUF(af6);
630 
631 		down6 = af6->ip6af_down;
632 		frag6_deq(af6);
633 
634 		/*
635 		 * Return ICMP time exceeded error for the 1st fragment.
636 		 * Just free other fragments.
637 		 */
638 		if (af6->ip6af_off == 0) {
639 			struct ip6_hdr *ip6;
640 
641 			/* adjust pointer */
642 			ip6 = mtod(m, struct ip6_hdr *);
643 
644 			/* restoure source and destination addresses */
645 			ip6->ip6_src = q6->ip6q_src;
646 			ip6->ip6_dst = q6->ip6q_dst;
647 
648 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
649 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
650 		} else
651 			m_freem(m);
652 		free(af6, M_FTABLE);
653 	}
654 	frag6_remque(q6);
655 	frag6_nfrags -= q6->ip6q_nfrag;
656 	free(q6, M_FTABLE);
657 	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_CHECK();
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_CHECK();
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_CHECK();
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_CHECK();
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()
718 {
719 	struct ip6q *q6;
720 	int s = splsoftnet();
721 
722 	IP6Q_LOCK();
723 	q6 = ip6q.ip6q_next;
724 	if (q6)
725 		while (q6 != &ip6q) {
726 			--q6->ip6q_ttl;
727 			q6 = q6->ip6q_next;
728 			if (q6->ip6q_prev->ip6q_ttl == 0) {
729 				ip6stat.ip6s_fragtimeout++;
730 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
731 				frag6_freef(q6->ip6q_prev);
732 			}
733 		}
734 	/*
735 	 * If we are over the maximum number of fragments
736 	 * (due to the limit being lowered), drain off
737 	 * enough to get down to the new limit.
738 	 */
739 	while (frag6_nfragpackets > (u_int)ip6_maxfragpackets &&
740 	    ip6q.ip6q_prev) {
741 		ip6stat.ip6s_fragoverflow++;
742 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
743 		frag6_freef(ip6q.ip6q_prev);
744 	}
745 	IP6Q_UNLOCK();
746 
747 #if 0
748 	/*
749 	 * Routing changes might produce a better route than we last used;
750 	 * make sure we notice eventually, even if forwarding only for one
751 	 * destination and the cache is never replaced.
752 	 */
753 	if (ip6_forward_rt.ro_rt) {
754 		RTFREE(ip6_forward_rt.ro_rt);
755 		ip6_forward_rt.ro_rt = 0;
756 	}
757 	if (ipsrcchk_rt.ro_rt) {
758 		RTFREE(ipsrcchk_rt.ro_rt);
759 		ipsrcchk_rt.ro_rt = 0;
760 	}
761 #endif
762 
763 	splx(s);
764 }
765 
766 /*
767  * Drain off all datagram fragments.
768  */
769 void
770 frag6_drain()
771 {
772 
773 	if (ip6q_lock_try() == 0)
774 		return;
775 	while (ip6q.ip6q_next != &ip6q) {
776 		ip6stat.ip6s_fragdropped++;
777 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
778 		frag6_freef(ip6q.ip6q_next);
779 	}
780 	IP6Q_UNLOCK();
781 }
782