xref: /freebsd/sys/netinet6/frag6.c (revision f95c930e6701aa675377bc5871ea490dd565eeba)
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/if_var.h>
49 #include <net/route.h>
50 #include <net/vnet.h>
51 
52 #include <netinet/in.h>
53 #include <netinet/in_var.h>
54 #include <netinet/ip6.h>
55 #include <netinet6/ip6_var.h>
56 #include <netinet/icmp6.h>
57 #include <netinet/in_systm.h>	/* for ECN definitions */
58 #include <netinet/ip.h>		/* for ECN definitions */
59 
60 #include <security/mac/mac_framework.h>
61 
62 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *);
63 static void frag6_deq(struct ip6asfrag *);
64 static void frag6_insque(struct ip6q *, struct ip6q *);
65 static void frag6_remque(struct ip6q *);
66 static void frag6_freef(struct ip6q *);
67 
68 static struct mtx ip6qlock;
69 /*
70  * These fields all protected by ip6qlock.
71  */
72 static VNET_DEFINE(u_int, frag6_nfragpackets);
73 static VNET_DEFINE(u_int, frag6_nfrags);
74 static VNET_DEFINE(struct ip6q, ip6q);	/* ip6 reassemble queue */
75 
76 #define	V_frag6_nfragpackets		VNET(frag6_nfragpackets)
77 #define	V_frag6_nfrags			VNET(frag6_nfrags)
78 #define	V_ip6q				VNET(ip6q)
79 
80 #define	IP6Q_LOCK_INIT()	mtx_init(&ip6qlock, "ip6qlock", NULL, MTX_DEF);
81 #define	IP6Q_LOCK()		mtx_lock(&ip6qlock)
82 #define	IP6Q_TRYLOCK()		mtx_trylock(&ip6qlock)
83 #define	IP6Q_LOCK_ASSERT()	mtx_assert(&ip6qlock, MA_OWNED)
84 #define	IP6Q_UNLOCK()		mtx_unlock(&ip6qlock)
85 
86 static MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header");
87 
88 /*
89  * Initialise reassembly queue and fragment identifier.
90  */
91 static void
92 frag6_change(void *tag)
93 {
94 
95 	V_ip6_maxfragpackets = nmbclusters / 4;
96 	V_ip6_maxfrags = nmbclusters / 4;
97 }
98 
99 void
100 frag6_init(void)
101 {
102 
103 	V_ip6_maxfragpackets = nmbclusters / 4;
104 	V_ip6_maxfrags = nmbclusters / 4;
105 	V_ip6q.ip6q_next = V_ip6q.ip6q_prev = &V_ip6q;
106 
107 	if (!IS_DEFAULT_VNET(curvnet))
108 		return;
109 
110 	EVENTHANDLER_REGISTER(nmbclusters_change,
111 	    frag6_change, NULL, EVENTHANDLER_PRI_ANY);
112 
113 	IP6Q_LOCK_INIT();
114 }
115 
116 /*
117  * In RFC2460, fragment and reassembly rule do not agree with each other,
118  * in terms of next header field handling in fragment header.
119  * While the sender will use the same value for all of the fragmented packets,
120  * receiver is suggested not to check the consistency.
121  *
122  * fragment rule (p20):
123  *	(2) A Fragment header containing:
124  *	The Next Header value that identifies the first header of
125  *	the Fragmentable Part of the original packet.
126  *		-> next header field is same for all fragments
127  *
128  * reassembly rule (p21):
129  *	The Next Header field of the last header of the Unfragmentable
130  *	Part is obtained from the Next Header field of the first
131  *	fragment's Fragment header.
132  *		-> should grab it from the first fragment only
133  *
134  * The following note also contradicts with fragment rule - noone is going to
135  * send different fragment with different next header field.
136  *
137  * additional note (p22):
138  *	The Next Header values in the Fragment headers of different
139  *	fragments of the same original packet may differ.  Only the value
140  *	from the Offset zero fragment packet is used for reassembly.
141  *		-> should grab it from the first fragment only
142  *
143  * There is no explicit reason given in the RFC.  Historical reason maybe?
144  */
145 /*
146  * Fragment input
147  */
148 int
149 frag6_input(struct mbuf **mp, int *offp, int proto)
150 {
151 	struct mbuf *m = *mp, *t;
152 	struct ip6_hdr *ip6;
153 	struct ip6_frag *ip6f;
154 	struct ip6q *q6;
155 	struct ip6asfrag *af6, *ip6af, *af6dwn;
156 	struct in6_ifaddr *ia;
157 	int offset = *offp, nxt, i, next;
158 	int first_frag = 0;
159 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
160 	struct ifnet *dstifp;
161 	u_int8_t ecn, ecn0;
162 #if 0
163 	char ip6buf[INET6_ADDRSTRLEN];
164 #endif
165 
166 	ip6 = mtod(m, struct ip6_hdr *);
167 #ifndef PULLDOWN_TEST
168 	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
169 	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
170 #else
171 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
172 	if (ip6f == NULL)
173 		return (IPPROTO_DONE);
174 #endif
175 
176 	dstifp = NULL;
177 	/* find the destination interface of the packet. */
178 	ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
179 	if (ia != NULL) {
180 		dstifp = ia->ia_ifp;
181 		ifa_free(&ia->ia_ifa);
182 	}
183 	/* jumbo payload can't contain a fragment header */
184 	if (ip6->ip6_plen == 0) {
185 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
186 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
187 		return IPPROTO_DONE;
188 	}
189 
190 	/*
191 	 * check whether fragment packet's fragment length is
192 	 * multiple of 8 octets.
193 	 * sizeof(struct ip6_frag) == 8
194 	 * sizeof(struct ip6_hdr) = 40
195 	 */
196 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
197 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
198 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
199 		    offsetof(struct ip6_hdr, ip6_plen));
200 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
201 		return IPPROTO_DONE;
202 	}
203 
204 	IP6STAT_INC(ip6s_fragments);
205 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
206 
207 	/* offset now points to data portion */
208 	offset += sizeof(struct ip6_frag);
209 
210 	/*
211 	 * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0)
212 	 * upfront, unrelated to any reassembly.  Just skip the fragment header.
213 	 */
214 	if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
215 		/* XXX-BZ we want dedicated counters for this. */
216 		IP6STAT_INC(ip6s_reassembled);
217 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
218 		*offp = offset;
219 		return (ip6f->ip6f_nxt);
220 	}
221 
222 	IP6Q_LOCK();
223 
224 	/*
225 	 * Enforce upper bound on number of fragments.
226 	 * If maxfrag is 0, never accept fragments.
227 	 * If maxfrag is -1, accept all fragments without limitation.
228 	 */
229 	if (V_ip6_maxfrags < 0)
230 		;
231 	else if (V_frag6_nfrags >= (u_int)V_ip6_maxfrags)
232 		goto dropfrag;
233 
234 	for (q6 = V_ip6q.ip6q_next; q6 != &V_ip6q; q6 = q6->ip6q_next)
235 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
236 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
237 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
238 #ifdef MAC
239 		    && mac_ip6q_match(m, q6)
240 #endif
241 		    )
242 			break;
243 
244 	if (q6 == &V_ip6q) {
245 		/*
246 		 * the first fragment to arrive, create a reassembly queue.
247 		 */
248 		first_frag = 1;
249 
250 		/*
251 		 * Enforce upper bound on number of fragmented packets
252 		 * for which we attempt reassembly;
253 		 * If maxfragpackets is 0, never accept fragments.
254 		 * If maxfragpackets is -1, accept all fragments without
255 		 * limitation.
256 		 */
257 		if (V_ip6_maxfragpackets < 0)
258 			;
259 		else if (V_frag6_nfragpackets >= (u_int)V_ip6_maxfragpackets)
260 			goto dropfrag;
261 		V_frag6_nfragpackets++;
262 		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
263 		    M_NOWAIT);
264 		if (q6 == NULL)
265 			goto dropfrag;
266 		bzero(q6, sizeof(*q6));
267 #ifdef MAC
268 		if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
269 			free(q6, M_FTABLE);
270 			goto dropfrag;
271 		}
272 		mac_ip6q_create(m, q6);
273 #endif
274 		frag6_insque(q6, &V_ip6q);
275 
276 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
277 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
278 #ifdef notyet
279 		q6->ip6q_nxtp	= (u_char *)nxtp;
280 #endif
281 		q6->ip6q_ident	= ip6f->ip6f_ident;
282 		q6->ip6q_ttl	= IPV6_FRAGTTL;
283 		q6->ip6q_src	= ip6->ip6_src;
284 		q6->ip6q_dst	= ip6->ip6_dst;
285 		q6->ip6q_ecn	=
286 		    (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
287 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
288 
289 		q6->ip6q_nfrag = 0;
290 	}
291 
292 	/*
293 	 * If it's the 1st fragment, record the length of the
294 	 * unfragmentable part and the next header of the fragment header.
295 	 */
296 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
297 	if (fragoff == 0) {
298 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
299 		    sizeof(struct ip6_frag);
300 		q6->ip6q_nxt = ip6f->ip6f_nxt;
301 	}
302 
303 	/*
304 	 * Check that the reassembled packet would not exceed 65535 bytes
305 	 * in size.
306 	 * If it would exceed, discard the fragment and return an ICMP error.
307 	 */
308 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
309 	if (q6->ip6q_unfrglen >= 0) {
310 		/* The 1st fragment has already arrived. */
311 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
312 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
313 			    offset - sizeof(struct ip6_frag) +
314 			    offsetof(struct ip6_frag, ip6f_offlg));
315 			IP6Q_UNLOCK();
316 			return (IPPROTO_DONE);
317 		}
318 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
319 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
320 		    offset - sizeof(struct ip6_frag) +
321 		    offsetof(struct ip6_frag, ip6f_offlg));
322 		IP6Q_UNLOCK();
323 		return (IPPROTO_DONE);
324 	}
325 	/*
326 	 * If it's the first fragment, do the above check for each
327 	 * fragment already stored in the reassembly queue.
328 	 */
329 	if (fragoff == 0) {
330 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
331 		     af6 = af6dwn) {
332 			af6dwn = af6->ip6af_down;
333 
334 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
335 			    IPV6_MAXPACKET) {
336 				struct mbuf *merr = IP6_REASS_MBUF(af6);
337 				struct ip6_hdr *ip6err;
338 				int erroff = af6->ip6af_offset;
339 
340 				/* dequeue the fragment. */
341 				frag6_deq(af6);
342 				free(af6, M_FTABLE);
343 
344 				/* adjust pointer. */
345 				ip6err = mtod(merr, struct ip6_hdr *);
346 
347 				/*
348 				 * Restore source and destination addresses
349 				 * in the erroneous IPv6 header.
350 				 */
351 				ip6err->ip6_src = q6->ip6q_src;
352 				ip6err->ip6_dst = q6->ip6q_dst;
353 
354 				icmp6_error(merr, ICMP6_PARAM_PROB,
355 				    ICMP6_PARAMPROB_HEADER,
356 				    erroff - sizeof(struct ip6_frag) +
357 				    offsetof(struct ip6_frag, ip6f_offlg));
358 			}
359 		}
360 	}
361 
362 	ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
363 	    M_NOWAIT);
364 	if (ip6af == NULL)
365 		goto dropfrag;
366 	bzero(ip6af, sizeof(*ip6af));
367 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
368 	ip6af->ip6af_off = fragoff;
369 	ip6af->ip6af_frglen = frgpartlen;
370 	ip6af->ip6af_offset = offset;
371 	IP6_REASS_MBUF(ip6af) = m;
372 
373 	if (first_frag) {
374 		af6 = (struct ip6asfrag *)q6;
375 		goto insert;
376 	}
377 
378 	/*
379 	 * Handle ECN by comparing this segment with the first one;
380 	 * if CE is set, do not lose CE.
381 	 * drop if CE and not-ECT are mixed for the same packet.
382 	 */
383 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
384 	ecn0 = q6->ip6q_ecn;
385 	if (ecn == IPTOS_ECN_CE) {
386 		if (ecn0 == IPTOS_ECN_NOTECT) {
387 			free(ip6af, M_FTABLE);
388 			goto dropfrag;
389 		}
390 		if (ecn0 != IPTOS_ECN_CE)
391 			q6->ip6q_ecn = IPTOS_ECN_CE;
392 	}
393 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
394 		free(ip6af, M_FTABLE);
395 		goto dropfrag;
396 	}
397 
398 	/*
399 	 * Find a segment which begins after this one does.
400 	 */
401 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
402 	     af6 = af6->ip6af_down)
403 		if (af6->ip6af_off > ip6af->ip6af_off)
404 			break;
405 
406 #if 0
407 	/*
408 	 * If there is a preceding segment, it may provide some of
409 	 * our data already.  If so, drop the data from the incoming
410 	 * segment.  If it provides all of our data, drop us.
411 	 */
412 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
413 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
414 			- ip6af->ip6af_off;
415 		if (i > 0) {
416 			if (i >= ip6af->ip6af_frglen)
417 				goto dropfrag;
418 			m_adj(IP6_REASS_MBUF(ip6af), i);
419 			ip6af->ip6af_off += i;
420 			ip6af->ip6af_frglen -= i;
421 		}
422 	}
423 
424 	/*
425 	 * While we overlap succeeding segments trim them or,
426 	 * if they are completely covered, dequeue them.
427 	 */
428 	while (af6 != (struct ip6asfrag *)q6 &&
429 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
430 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
431 		if (i < af6->ip6af_frglen) {
432 			af6->ip6af_frglen -= i;
433 			af6->ip6af_off += i;
434 			m_adj(IP6_REASS_MBUF(af6), i);
435 			break;
436 		}
437 		af6 = af6->ip6af_down;
438 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
439 		frag6_deq(af6->ip6af_up);
440 	}
441 #else
442 	/*
443 	 * If the incoming framgent overlaps some existing fragments in
444 	 * the reassembly queue, drop it, since it is dangerous to override
445 	 * existing fragments from a security point of view.
446 	 * We don't know which fragment is the bad guy - here we trust
447 	 * fragment that came in earlier, with no real reason.
448 	 *
449 	 * Note: due to changes after disabling this part, mbuf passed to
450 	 * m_adj() below now does not meet the requirement.
451 	 */
452 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
453 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
454 			- ip6af->ip6af_off;
455 		if (i > 0) {
456 #if 0				/* suppress the noisy log */
457 			log(LOG_ERR, "%d bytes of a fragment from %s "
458 			    "overlaps the previous fragment\n",
459 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
460 #endif
461 			free(ip6af, M_FTABLE);
462 			goto dropfrag;
463 		}
464 	}
465 	if (af6 != (struct ip6asfrag *)q6) {
466 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
467 		if (i > 0) {
468 #if 0				/* suppress the noisy log */
469 			log(LOG_ERR, "%d bytes of a fragment from %s "
470 			    "overlaps the succeeding fragment",
471 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
472 #endif
473 			free(ip6af, M_FTABLE);
474 			goto dropfrag;
475 		}
476 	}
477 #endif
478 
479 insert:
480 #ifdef MAC
481 	if (!first_frag)
482 		mac_ip6q_update(m, q6);
483 #endif
484 
485 	/*
486 	 * Stick new segment in its place;
487 	 * check for complete reassembly.
488 	 * Move to front of packet queue, as we are
489 	 * the most recently active fragmented packet.
490 	 */
491 	frag6_enq(ip6af, af6->ip6af_up);
492 	V_frag6_nfrags++;
493 	q6->ip6q_nfrag++;
494 #if 0 /* xxx */
495 	if (q6 != V_ip6q.ip6q_next) {
496 		frag6_remque(q6);
497 		frag6_insque(q6, &V_ip6q);
498 	}
499 #endif
500 	next = 0;
501 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
502 	     af6 = af6->ip6af_down) {
503 		if (af6->ip6af_off != next) {
504 			IP6Q_UNLOCK();
505 			return IPPROTO_DONE;
506 		}
507 		next += af6->ip6af_frglen;
508 	}
509 	if (af6->ip6af_up->ip6af_mff) {
510 		IP6Q_UNLOCK();
511 		return IPPROTO_DONE;
512 	}
513 
514 	/*
515 	 * Reassembly is complete; concatenate fragments.
516 	 */
517 	ip6af = q6->ip6q_down;
518 	t = m = IP6_REASS_MBUF(ip6af);
519 	af6 = ip6af->ip6af_down;
520 	frag6_deq(ip6af);
521 	while (af6 != (struct ip6asfrag *)q6) {
522 		af6dwn = af6->ip6af_down;
523 		frag6_deq(af6);
524 		while (t->m_next)
525 			t = t->m_next;
526 		t->m_next = IP6_REASS_MBUF(af6);
527 		m_adj(t->m_next, af6->ip6af_offset);
528 		free(af6, M_FTABLE);
529 		af6 = af6dwn;
530 	}
531 
532 	/* adjust offset to point where the original next header starts */
533 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
534 	free(ip6af, M_FTABLE);
535 	ip6 = mtod(m, struct ip6_hdr *);
536 	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
537 	if (q6->ip6q_ecn == IPTOS_ECN_CE)
538 		ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
539 	nxt = q6->ip6q_nxt;
540 #ifdef notyet
541 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
542 #endif
543 
544 	if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) {
545 		frag6_remque(q6);
546 		V_frag6_nfrags -= q6->ip6q_nfrag;
547 #ifdef MAC
548 		mac_ip6q_destroy(q6);
549 #endif
550 		free(q6, M_FTABLE);
551 		V_frag6_nfragpackets--;
552 
553 		goto dropfrag;
554 	}
555 
556 	/*
557 	 * Store NXT to the original.
558 	 */
559 	{
560 		char *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
561 		*prvnxtp = nxt;
562 	}
563 
564 	frag6_remque(q6);
565 	V_frag6_nfrags -= q6->ip6q_nfrag;
566 #ifdef MAC
567 	mac_ip6q_reassemble(q6, m);
568 	mac_ip6q_destroy(q6);
569 #endif
570 	free(q6, M_FTABLE);
571 	V_frag6_nfragpackets--;
572 
573 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
574 		int plen = 0;
575 		for (t = m; t; t = t->m_next)
576 			plen += t->m_len;
577 		m->m_pkthdr.len = plen;
578 	}
579 
580 	IP6STAT_INC(ip6s_reassembled);
581 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
582 
583 	/*
584 	 * Tell launch routine the next header
585 	 */
586 
587 	*mp = m;
588 	*offp = offset;
589 
590 	IP6Q_UNLOCK();
591 	return nxt;
592 
593  dropfrag:
594 	IP6Q_UNLOCK();
595 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
596 	IP6STAT_INC(ip6s_fragdropped);
597 	m_freem(m);
598 	return IPPROTO_DONE;
599 }
600 
601 /*
602  * Free a fragment reassembly header and all
603  * associated datagrams.
604  */
605 void
606 frag6_freef(struct ip6q *q6)
607 {
608 	struct ip6asfrag *af6, *down6;
609 
610 	IP6Q_LOCK_ASSERT();
611 
612 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
613 	     af6 = down6) {
614 		struct mbuf *m = IP6_REASS_MBUF(af6);
615 
616 		down6 = af6->ip6af_down;
617 		frag6_deq(af6);
618 
619 		/*
620 		 * Return ICMP time exceeded error for the 1st fragment.
621 		 * Just free other fragments.
622 		 */
623 		if (af6->ip6af_off == 0) {
624 			struct ip6_hdr *ip6;
625 
626 			/* adjust pointer */
627 			ip6 = mtod(m, struct ip6_hdr *);
628 
629 			/* restore source and destination addresses */
630 			ip6->ip6_src = q6->ip6q_src;
631 			ip6->ip6_dst = q6->ip6q_dst;
632 
633 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
634 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
635 		} else
636 			m_freem(m);
637 		free(af6, M_FTABLE);
638 	}
639 	frag6_remque(q6);
640 	V_frag6_nfrags -= q6->ip6q_nfrag;
641 #ifdef MAC
642 	mac_ip6q_destroy(q6);
643 #endif
644 	free(q6, M_FTABLE);
645 	V_frag6_nfragpackets--;
646 }
647 
648 /*
649  * Put an ip fragment on a reassembly chain.
650  * Like insque, but pointers in middle of structure.
651  */
652 void
653 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
654 {
655 
656 	IP6Q_LOCK_ASSERT();
657 
658 	af6->ip6af_up = up6;
659 	af6->ip6af_down = up6->ip6af_down;
660 	up6->ip6af_down->ip6af_up = af6;
661 	up6->ip6af_down = af6;
662 }
663 
664 /*
665  * To frag6_enq as remque is to insque.
666  */
667 void
668 frag6_deq(struct ip6asfrag *af6)
669 {
670 
671 	IP6Q_LOCK_ASSERT();
672 
673 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
674 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
675 }
676 
677 void
678 frag6_insque(struct ip6q *new, struct ip6q *old)
679 {
680 
681 	IP6Q_LOCK_ASSERT();
682 
683 	new->ip6q_prev = old;
684 	new->ip6q_next = old->ip6q_next;
685 	old->ip6q_next->ip6q_prev= new;
686 	old->ip6q_next = new;
687 }
688 
689 void
690 frag6_remque(struct ip6q *p6)
691 {
692 
693 	IP6Q_LOCK_ASSERT();
694 
695 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
696 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
697 }
698 
699 /*
700  * IPv6 reassembling timer processing;
701  * if a timer expires on a reassembly
702  * queue, discard it.
703  */
704 void
705 frag6_slowtimo(void)
706 {
707 	VNET_ITERATOR_DECL(vnet_iter);
708 	struct ip6q *q6;
709 
710 	VNET_LIST_RLOCK_NOSLEEP();
711 	IP6Q_LOCK();
712 	VNET_FOREACH(vnet_iter) {
713 		CURVNET_SET(vnet_iter);
714 		q6 = V_ip6q.ip6q_next;
715 		if (q6)
716 			while (q6 != &V_ip6q) {
717 				--q6->ip6q_ttl;
718 				q6 = q6->ip6q_next;
719 				if (q6->ip6q_prev->ip6q_ttl == 0) {
720 					IP6STAT_INC(ip6s_fragtimeout);
721 					/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
722 					frag6_freef(q6->ip6q_prev);
723 				}
724 			}
725 		/*
726 		 * If we are over the maximum number of fragments
727 		 * (due to the limit being lowered), drain off
728 		 * enough to get down to the new limit.
729 		 */
730 		while (V_frag6_nfragpackets > (u_int)V_ip6_maxfragpackets &&
731 		    V_ip6q.ip6q_prev) {
732 			IP6STAT_INC(ip6s_fragoverflow);
733 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
734 			frag6_freef(V_ip6q.ip6q_prev);
735 		}
736 		CURVNET_RESTORE();
737 	}
738 	IP6Q_UNLOCK();
739 	VNET_LIST_RUNLOCK_NOSLEEP();
740 }
741 
742 /*
743  * Drain off all datagram fragments.
744  */
745 void
746 frag6_drain(void)
747 {
748 	VNET_ITERATOR_DECL(vnet_iter);
749 
750 	VNET_LIST_RLOCK_NOSLEEP();
751 	if (IP6Q_TRYLOCK() == 0) {
752 		VNET_LIST_RUNLOCK_NOSLEEP();
753 		return;
754 	}
755 	VNET_FOREACH(vnet_iter) {
756 		CURVNET_SET(vnet_iter);
757 		while (V_ip6q.ip6q_next != &V_ip6q) {
758 			IP6STAT_INC(ip6s_fragdropped);
759 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
760 			frag6_freef(V_ip6q.ip6q_next);
761 		}
762 		CURVNET_RESTORE();
763 	}
764 	IP6Q_UNLOCK();
765 	VNET_LIST_RUNLOCK_NOSLEEP();
766 }
767 
768 int
769 ip6_deletefraghdr(struct mbuf *m, int offset, int wait)
770 {
771 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
772 	struct mbuf *t;
773 
774 	/* Delete frag6 header. */
775 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
776 		/* This is the only possible case with !PULLDOWN_TEST. */
777 		bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag),
778 		    offset);
779 		m->m_data += sizeof(struct ip6_frag);
780 		m->m_len -= sizeof(struct ip6_frag);
781 	} else {
782 		/* This comes with no copy if the boundary is on cluster. */
783 		if ((t = m_split(m, offset, wait)) == NULL)
784 			return (ENOMEM);
785 		m_adj(t, sizeof(struct ip6_frag));
786 		m_cat(m, t);
787 	}
788 
789 	return (0);
790 }
791