xref: /freebsd/sys/netinet6/frag6.c (revision febdb468801f35e51c6c5c22221cfce9197c6f3b)
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_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 - noone 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 #ifdef IN6_IFSTAT_STRICT
163 	struct in6_ifaddr *ia;
164 #endif
165 	int offset = *offp, nxt, i, next;
166 	int first_frag = 0;
167 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
168 	struct ifnet *dstifp;
169 	u_int8_t ecn, ecn0;
170 #if 0
171 	char ip6buf[INET6_ADDRSTRLEN];
172 #endif
173 
174 	ip6 = mtod(m, struct ip6_hdr *);
175 #ifndef PULLDOWN_TEST
176 	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
177 	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
178 #else
179 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
180 	if (ip6f == NULL)
181 		return (IPPROTO_DONE);
182 #endif
183 
184 	dstifp = NULL;
185 #ifdef IN6_IFSTAT_STRICT
186 	/* find the destination interface of the packet. */
187 	if ((ia = ip6_getdstifaddr(m)) != NULL) {
188 		dstifp = ia->ia_ifp;
189 		ifa_free(&ia->ia_ifa);
190 	}
191 #else
192 	/* we are violating the spec, this is not the destination interface */
193 	if ((m->m_flags & M_PKTHDR) != 0)
194 		dstifp = m->m_pkthdr.rcvif;
195 #endif
196 
197 	/* jumbo payload can't contain a fragment header */
198 	if (ip6->ip6_plen == 0) {
199 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
200 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
201 		return IPPROTO_DONE;
202 	}
203 
204 	/*
205 	 * check whether fragment packet's fragment length is
206 	 * multiple of 8 octets.
207 	 * sizeof(struct ip6_frag) == 8
208 	 * sizeof(struct ip6_hdr) = 40
209 	 */
210 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
211 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
212 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
213 		    offsetof(struct ip6_hdr, ip6_plen));
214 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
215 		return IPPROTO_DONE;
216 	}
217 
218 	IP6STAT_INC(ip6s_fragments);
219 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
220 
221 	/* offset now points to data portion */
222 	offset += sizeof(struct ip6_frag);
223 
224 	/*
225 	 * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0)
226 	 * upfront, unrelated to any reassembly.  Just skip the fragment header.
227 	 */
228 	if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
229 		/* XXX-BZ we want dedicated counters for this. */
230 		IP6STAT_INC(ip6s_reassembled);
231 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
232 		*offp = offset;
233 		return (ip6f->ip6f_nxt);
234 	}
235 
236 	IP6Q_LOCK();
237 
238 	/*
239 	 * Enforce upper bound on number of fragments.
240 	 * If maxfrag is 0, never accept fragments.
241 	 * If maxfrag is -1, accept all fragments without limitation.
242 	 */
243 	if (V_ip6_maxfrags < 0)
244 		;
245 	else if (V_frag6_nfrags >= (u_int)V_ip6_maxfrags)
246 		goto dropfrag;
247 
248 	for (q6 = V_ip6q.ip6q_next; q6 != &V_ip6q; q6 = q6->ip6q_next)
249 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
250 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
251 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
252 #ifdef MAC
253 		    && mac_ip6q_match(m, q6)
254 #endif
255 		    )
256 			break;
257 
258 	if (q6 == &V_ip6q) {
259 		/*
260 		 * the first fragment to arrive, create a reassembly queue.
261 		 */
262 		first_frag = 1;
263 
264 		/*
265 		 * Enforce upper bound on number of fragmented packets
266 		 * for which we attempt reassembly;
267 		 * If maxfragpackets is 0, never accept fragments.
268 		 * If maxfragpackets is -1, accept all fragments without
269 		 * limitation.
270 		 */
271 		if (V_ip6_maxfragpackets < 0)
272 			;
273 		else if (V_frag6_nfragpackets >= (u_int)V_ip6_maxfragpackets)
274 			goto dropfrag;
275 		V_frag6_nfragpackets++;
276 		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
277 		    M_NOWAIT);
278 		if (q6 == NULL)
279 			goto dropfrag;
280 		bzero(q6, sizeof(*q6));
281 #ifdef MAC
282 		if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
283 			free(q6, M_FTABLE);
284 			goto dropfrag;
285 		}
286 		mac_ip6q_create(m, q6);
287 #endif
288 		frag6_insque(q6, &V_ip6q);
289 
290 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
291 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
292 #ifdef notyet
293 		q6->ip6q_nxtp	= (u_char *)nxtp;
294 #endif
295 		q6->ip6q_ident	= ip6f->ip6f_ident;
296 		q6->ip6q_ttl	= IPV6_FRAGTTL;
297 		q6->ip6q_src	= ip6->ip6_src;
298 		q6->ip6q_dst	= ip6->ip6_dst;
299 		q6->ip6q_ecn	=
300 		    (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
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 = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
377 	    M_NOWAIT);
378 	if (ip6af == NULL)
379 		goto dropfrag;
380 	bzero(ip6af, sizeof(*ip6af));
381 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
382 	ip6af->ip6af_off = fragoff;
383 	ip6af->ip6af_frglen = frgpartlen;
384 	ip6af->ip6af_offset = offset;
385 	IP6_REASS_MBUF(ip6af) = m;
386 
387 	if (first_frag) {
388 		af6 = (struct ip6asfrag *)q6;
389 		goto insert;
390 	}
391 
392 	/*
393 	 * Handle ECN by comparing this segment with the first one;
394 	 * if CE is set, do not lose CE.
395 	 * drop if CE and not-ECT are mixed for the same packet.
396 	 */
397 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
398 	ecn0 = q6->ip6q_ecn;
399 	if (ecn == IPTOS_ECN_CE) {
400 		if (ecn0 == IPTOS_ECN_NOTECT) {
401 			free(ip6af, M_FTABLE);
402 			goto dropfrag;
403 		}
404 		if (ecn0 != IPTOS_ECN_CE)
405 			q6->ip6q_ecn = IPTOS_ECN_CE;
406 	}
407 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
408 		free(ip6af, M_FTABLE);
409 		goto dropfrag;
410 	}
411 
412 	/*
413 	 * Find a segment which begins after this one does.
414 	 */
415 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
416 	     af6 = af6->ip6af_down)
417 		if (af6->ip6af_off > ip6af->ip6af_off)
418 			break;
419 
420 #if 0
421 	/*
422 	 * If there is a preceding segment, it may provide some of
423 	 * our data already.  If so, drop the data from the incoming
424 	 * segment.  If it provides all of our data, drop us.
425 	 */
426 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
427 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
428 			- ip6af->ip6af_off;
429 		if (i > 0) {
430 			if (i >= ip6af->ip6af_frglen)
431 				goto dropfrag;
432 			m_adj(IP6_REASS_MBUF(ip6af), i);
433 			ip6af->ip6af_off += i;
434 			ip6af->ip6af_frglen -= i;
435 		}
436 	}
437 
438 	/*
439 	 * While we overlap succeeding segments trim them or,
440 	 * if they are completely covered, dequeue them.
441 	 */
442 	while (af6 != (struct ip6asfrag *)q6 &&
443 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
444 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
445 		if (i < af6->ip6af_frglen) {
446 			af6->ip6af_frglen -= i;
447 			af6->ip6af_off += i;
448 			m_adj(IP6_REASS_MBUF(af6), i);
449 			break;
450 		}
451 		af6 = af6->ip6af_down;
452 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
453 		frag6_deq(af6->ip6af_up);
454 	}
455 #else
456 	/*
457 	 * If the incoming framgent overlaps some existing fragments in
458 	 * the reassembly queue, drop it, since it is dangerous to override
459 	 * existing fragments from a security point of view.
460 	 * We don't know which fragment is the bad guy - here we trust
461 	 * fragment that came in earlier, with no real reason.
462 	 *
463 	 * Note: due to changes after disabling this part, mbuf passed to
464 	 * m_adj() below now does not meet the requirement.
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(ip6buf, &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(ip6buf, &q6->ip6q_src));
486 #endif
487 			free(ip6af, M_FTABLE);
488 			goto dropfrag;
489 		}
490 	}
491 #endif
492 
493 insert:
494 #ifdef MAC
495 	if (!first_frag)
496 		mac_ip6q_update(m, q6);
497 #endif
498 
499 	/*
500 	 * Stick new segment in its place;
501 	 * check for complete reassembly.
502 	 * Move to front of packet queue, as we are
503 	 * the most recently active fragmented packet.
504 	 */
505 	frag6_enq(ip6af, af6->ip6af_up);
506 	V_frag6_nfrags++;
507 	q6->ip6q_nfrag++;
508 #if 0 /* xxx */
509 	if (q6 != V_ip6q.ip6q_next) {
510 		frag6_remque(q6);
511 		frag6_insque(q6, &V_ip6q);
512 	}
513 #endif
514 	next = 0;
515 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
516 	     af6 = af6->ip6af_down) {
517 		if (af6->ip6af_off != next) {
518 			IP6Q_UNLOCK();
519 			return IPPROTO_DONE;
520 		}
521 		next += af6->ip6af_frglen;
522 	}
523 	if (af6->ip6af_up->ip6af_mff) {
524 		IP6Q_UNLOCK();
525 		return IPPROTO_DONE;
526 	}
527 
528 	/*
529 	 * Reassembly is complete; concatenate fragments.
530 	 */
531 	ip6af = q6->ip6q_down;
532 	t = m = IP6_REASS_MBUF(ip6af);
533 	af6 = ip6af->ip6af_down;
534 	frag6_deq(ip6af);
535 	while (af6 != (struct ip6asfrag *)q6) {
536 		af6dwn = af6->ip6af_down;
537 		frag6_deq(af6);
538 		while (t->m_next)
539 			t = t->m_next;
540 		t->m_next = IP6_REASS_MBUF(af6);
541 		m_adj(t->m_next, af6->ip6af_offset);
542 		free(af6, M_FTABLE);
543 		af6 = af6dwn;
544 	}
545 
546 	/* adjust offset to point where the original next header starts */
547 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
548 	free(ip6af, M_FTABLE);
549 	ip6 = mtod(m, struct ip6_hdr *);
550 	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
551 	if (q6->ip6q_ecn == IPTOS_ECN_CE)
552 		ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
553 	nxt = q6->ip6q_nxt;
554 #ifdef notyet
555 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
556 #endif
557 
558 	/* Delete frag6 header */
559 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
560 		/* This is the only possible case with !PULLDOWN_TEST */
561 		ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
562 		    offset);
563 		m->m_data += sizeof(struct ip6_frag);
564 		m->m_len -= sizeof(struct ip6_frag);
565 	} else {
566 		/* this comes with no copy if the boundary is on cluster */
567 		if ((t = m_split(m, offset, M_NOWAIT)) == NULL) {
568 			frag6_remque(q6);
569 			V_frag6_nfrags -= q6->ip6q_nfrag;
570 #ifdef MAC
571 			mac_ip6q_destroy(q6);
572 #endif
573 			free(q6, M_FTABLE);
574 			V_frag6_nfragpackets--;
575 			goto dropfrag;
576 		}
577 		m_adj(t, sizeof(struct ip6_frag));
578 		m_cat(m, t);
579 	}
580 
581 	/*
582 	 * Store NXT to the original.
583 	 */
584 	{
585 		char *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
586 		*prvnxtp = nxt;
587 	}
588 
589 	frag6_remque(q6);
590 	V_frag6_nfrags -= q6->ip6q_nfrag;
591 #ifdef MAC
592 	mac_ip6q_reassemble(q6, m);
593 	mac_ip6q_destroy(q6);
594 #endif
595 	free(q6, M_FTABLE);
596 	V_frag6_nfragpackets--;
597 
598 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
599 		int plen = 0;
600 		for (t = m; t; t = t->m_next)
601 			plen += t->m_len;
602 		m->m_pkthdr.len = plen;
603 	}
604 
605 	IP6STAT_INC(ip6s_reassembled);
606 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
607 
608 	/*
609 	 * Tell launch routine the next header
610 	 */
611 
612 	*mp = m;
613 	*offp = offset;
614 
615 	IP6Q_UNLOCK();
616 	return nxt;
617 
618  dropfrag:
619 	IP6Q_UNLOCK();
620 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
621 	IP6STAT_INC(ip6s_fragdropped);
622 	m_freem(m);
623 	return IPPROTO_DONE;
624 }
625 
626 /*
627  * Free a fragment reassembly header and all
628  * associated datagrams.
629  */
630 void
631 frag6_freef(struct ip6q *q6)
632 {
633 	struct ip6asfrag *af6, *down6;
634 
635 	IP6Q_LOCK_ASSERT();
636 
637 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
638 	     af6 = down6) {
639 		struct mbuf *m = IP6_REASS_MBUF(af6);
640 
641 		down6 = af6->ip6af_down;
642 		frag6_deq(af6);
643 
644 		/*
645 		 * Return ICMP time exceeded error for the 1st fragment.
646 		 * Just free other fragments.
647 		 */
648 		if (af6->ip6af_off == 0) {
649 			struct ip6_hdr *ip6;
650 
651 			/* adjust pointer */
652 			ip6 = mtod(m, struct ip6_hdr *);
653 
654 			/* restore source and destination addresses */
655 			ip6->ip6_src = q6->ip6q_src;
656 			ip6->ip6_dst = q6->ip6q_dst;
657 
658 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
659 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
660 		} else
661 			m_freem(m);
662 		free(af6, M_FTABLE);
663 	}
664 	frag6_remque(q6);
665 	V_frag6_nfrags -= q6->ip6q_nfrag;
666 #ifdef MAC
667 	mac_ip6q_destroy(q6);
668 #endif
669 	free(q6, M_FTABLE);
670 	V_frag6_nfragpackets--;
671 }
672 
673 /*
674  * Put an ip fragment on a reassembly chain.
675  * Like insque, but pointers in middle of structure.
676  */
677 void
678 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
679 {
680 
681 	IP6Q_LOCK_ASSERT();
682 
683 	af6->ip6af_up = up6;
684 	af6->ip6af_down = up6->ip6af_down;
685 	up6->ip6af_down->ip6af_up = af6;
686 	up6->ip6af_down = af6;
687 }
688 
689 /*
690  * To frag6_enq as remque is to insque.
691  */
692 void
693 frag6_deq(struct ip6asfrag *af6)
694 {
695 
696 	IP6Q_LOCK_ASSERT();
697 
698 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
699 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
700 }
701 
702 void
703 frag6_insque(struct ip6q *new, struct ip6q *old)
704 {
705 
706 	IP6Q_LOCK_ASSERT();
707 
708 	new->ip6q_prev = old;
709 	new->ip6q_next = old->ip6q_next;
710 	old->ip6q_next->ip6q_prev= new;
711 	old->ip6q_next = new;
712 }
713 
714 void
715 frag6_remque(struct ip6q *p6)
716 {
717 
718 	IP6Q_LOCK_ASSERT();
719 
720 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
721 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
722 }
723 
724 /*
725  * IPv6 reassembling timer processing;
726  * if a timer expires on a reassembly
727  * queue, discard it.
728  */
729 void
730 frag6_slowtimo(void)
731 {
732 	VNET_ITERATOR_DECL(vnet_iter);
733 	struct ip6q *q6;
734 
735 	VNET_LIST_RLOCK_NOSLEEP();
736 	IP6Q_LOCK();
737 	VNET_FOREACH(vnet_iter) {
738 		CURVNET_SET(vnet_iter);
739 		q6 = V_ip6q.ip6q_next;
740 		if (q6)
741 			while (q6 != &V_ip6q) {
742 				--q6->ip6q_ttl;
743 				q6 = q6->ip6q_next;
744 				if (q6->ip6q_prev->ip6q_ttl == 0) {
745 					IP6STAT_INC(ip6s_fragtimeout);
746 					/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
747 					frag6_freef(q6->ip6q_prev);
748 				}
749 			}
750 		/*
751 		 * If we are over the maximum number of fragments
752 		 * (due to the limit being lowered), drain off
753 		 * enough to get down to the new limit.
754 		 */
755 		while (V_frag6_nfragpackets > (u_int)V_ip6_maxfragpackets &&
756 		    V_ip6q.ip6q_prev) {
757 			IP6STAT_INC(ip6s_fragoverflow);
758 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
759 			frag6_freef(V_ip6q.ip6q_prev);
760 		}
761 		CURVNET_RESTORE();
762 	}
763 	IP6Q_UNLOCK();
764 	VNET_LIST_RUNLOCK_NOSLEEP();
765 }
766 
767 /*
768  * Drain off all datagram fragments.
769  */
770 void
771 frag6_drain(void)
772 {
773 	VNET_ITERATOR_DECL(vnet_iter);
774 
775 	VNET_LIST_RLOCK_NOSLEEP();
776 	if (IP6Q_TRYLOCK() == 0) {
777 		VNET_LIST_RUNLOCK_NOSLEEP();
778 		return;
779 	}
780 	VNET_FOREACH(vnet_iter) {
781 		CURVNET_SET(vnet_iter);
782 		while (V_ip6q.ip6q_next != &V_ip6q) {
783 			IP6STAT_INC(ip6s_fragdropped);
784 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
785 			frag6_freef(V_ip6q.ip6q_next);
786 		}
787 		CURVNET_RESTORE();
788 	}
789 	IP6Q_UNLOCK();
790 	VNET_LIST_RUNLOCK_NOSLEEP();
791 }
792