xref: /netbsd-src/sys/netinet6/frag6.c (revision 16dce51364ebe8aeafbae46bc5aa167b8115bc45) 
1 /*	$NetBSD: frag6.c,v 1.71 2018/04/13 11:32:44 maxv 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/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: frag6.c,v 1.71 2018/04/13 11:32:44 maxv Exp $");
35 
36 #ifdef _KERNEL_OPT
37 #include "opt_net_mpsafe.h"
38 #endif
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/mbuf.h>
43 #include <sys/errno.h>
44 #include <sys/time.h>
45 #include <sys/kmem.h>
46 #include <sys/kernel.h>
47 #include <sys/syslog.h>
48 
49 #include <net/if.h>
50 #include <net/route.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 <netinet6/ip6_private.h>
57 #include <netinet/icmp6.h>
58 
59 #include <net/net_osdep.h>
60 
61 /*
62  * IPv6 reassembly queue structure. Each fragment being reassembled is
63  * attached to one of these structures.
64  *
65  * XXX: Would be better to use TAILQ.
66  */
67 struct	ip6q {
68 	u_int32_t	ip6q_head;
69 	u_int16_t	ip6q_len;
70 	u_int8_t	ip6q_nxt;	/* ip6f_nxt in first fragment */
71 	u_int8_t	ip6q_hlim;
72 	struct ip6asfrag *ip6q_down;
73 	struct ip6asfrag *ip6q_up;
74 	u_int32_t	ip6q_ident;
75 	u_int8_t	ip6q_ttl;
76 	struct in6_addr	ip6q_src, ip6q_dst;
77 	struct ip6q	*ip6q_next;
78 	struct ip6q	*ip6q_prev;
79 	int		ip6q_unfrglen;	/* len of unfragmentable part */
80 	int		ip6q_nfrag;	/* # of fragments */
81 };
82 
83 struct	ip6asfrag {
84 	u_int32_t	ip6af_head;
85 	u_int16_t	ip6af_len;
86 	u_int8_t	ip6af_nxt;
87 	u_int8_t	ip6af_hlim;
88 	/* must not override the above members during reassembling */
89 	struct ip6asfrag *ip6af_down;
90 	struct ip6asfrag *ip6af_up;
91 	struct mbuf	*ip6af_m;
92 	int		ip6af_offset;	/* offset in ip6af_m to next header */
93 	int		ip6af_frglen;	/* fragmentable part length */
94 	int		ip6af_off;	/* fragment offset */
95 	bool		ip6af_mff;	/* more fragment bit in frag off */
96 };
97 
98 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *);
99 static void frag6_deq(struct ip6asfrag *);
100 static void frag6_insque(struct ip6q *, struct ip6q *);
101 static void frag6_remque(struct ip6q *);
102 static void frag6_freef(struct ip6q *);
103 
104 static int frag6_drainwanted;
105 
106 static u_int frag6_nfragpackets;
107 static u_int frag6_nfrags;
108 static struct ip6q ip6q;	/* ip6 reassembly queue */
109 
110 /* Protects ip6q */
111 static kmutex_t	frag6_lock __cacheline_aligned;
112 
113 /*
114  * Initialise reassembly queue and fragment identifier.
115  */
116 void
117 frag6_init(void)
118 {
119 
120 	ip6q.ip6q_next = ip6q.ip6q_prev = &ip6q;
121 	mutex_init(&frag6_lock, MUTEX_DEFAULT, IPL_NET);
122 }
123 
124 /*
125  * IPv6 fragment input.
126  *
127  * In RFC2460, fragment and reassembly rule do not agree with each other,
128  * in terms of next header field handling in fragment header.
129  * While the sender will use the same value for all of the fragmented packets,
130  * receiver is suggested not to check the consistency.
131  *
132  * fragment rule (p20):
133  *	(2) A Fragment header containing:
134  *	The Next Header value that identifies the first header of
135  *	the Fragmentable Part of the original packet.
136  *		-> next header field is same for all fragments
137  *
138  * reassembly rule (p21):
139  *	The Next Header field of the last header of the Unfragmentable
140  *	Part is obtained from the Next Header field of the first
141  *	fragment's Fragment header.
142  *		-> should grab it from the first fragment only
143  *
144  * The following note also contradicts with fragment rule - noone is going to
145  * send different fragment with different next header field.
146  *
147  * additional note (p22):
148  *	The Next Header values in the Fragment headers of different
149  *	fragments of the same original packet may differ.  Only the value
150  *	from the Offset zero fragment packet is used for reassembly.
151  *		-> should grab it from the first fragment only
152  *
153  * There is no explicit reason given in the RFC.  Historical reason maybe?
154  *
155  * XXX: It would be better to use a pool, rather than kmem.
156  */
157 int
158 frag6_input(struct mbuf **mp, int *offp, int proto)
159 {
160 	struct rtentry *rt;
161 	struct mbuf *m = *mp, *t;
162 	struct ip6_hdr *ip6;
163 	struct ip6_frag *ip6f;
164 	struct ip6q *q6;
165 	struct ip6asfrag *af6, *ip6af, *af6dwn;
166 	int offset = *offp, nxt, i, next;
167 	int first_frag = 0;
168 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
169 	struct ifnet *dstifp;
170 	static struct route ro;
171 	union {
172 		struct sockaddr		dst;
173 		struct sockaddr_in6	dst6;
174 	} u;
175 
176 	ip6 = mtod(m, struct ip6_hdr *);
177 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
178 	if (ip6f == NULL)
179 		return IPPROTO_DONE;
180 
181 	dstifp = NULL;
182 	/* find the destination interface of the packet. */
183 	sockaddr_in6_init(&u.dst6, &ip6->ip6_dst, 0, 0, 0);
184 	if ((rt = rtcache_lookup(&ro, &u.dst)) != NULL && rt->rt_ifa != NULL)
185 		dstifp = ((struct in6_ifaddr *)rt->rt_ifa)->ia_ifp;
186 
187 	/* jumbo payload can't contain a fragment header */
188 	if (ip6->ip6_plen == 0) {
189 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
190 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
191 		goto done;
192 	}
193 
194 	/*
195 	 * Check whether fragment packet's fragment length is non-zero and
196 	 * multiple of 8 octets.
197 	 * sizeof(struct ip6_frag) == 8
198 	 * sizeof(struct ip6_hdr) = 40
199 	 */
200 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
201 	    (((ntohs(ip6->ip6_plen) - offset) == 0) ||
202 	     ((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
203 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
204 		    offsetof(struct ip6_hdr, ip6_plen));
205 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
206 		goto done;
207 	}
208 
209 	IP6_STATINC(IP6_STAT_FRAGMENTS);
210 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
211 
212 	/* offset now points to data portion */
213 	offset += sizeof(struct ip6_frag);
214 
215 	/*
216 	 * RFC6946: A host that receives an IPv6 packet which includes
217 	 * a Fragment Header with the "Fragment Offset" equal to 0 and
218 	 * the "M" bit equal to 0 MUST process such packet in isolation
219 	 * from any other packets/fragments.
220 	 *
221 	 * XXX: Would be better to remove this fragment header entirely,
222 	 * for us not to get confused later when looking back at the
223 	 * previous headers in the chain.
224 	 */
225 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
226 	if (fragoff == 0 && !(ip6f->ip6f_offlg & IP6F_MORE_FRAG)) {
227 		IP6_STATINC(IP6_STAT_REASSEMBLED);
228 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
229 		*offp = offset;
230 		rtcache_unref(rt, &ro);
231 		return ip6f->ip6f_nxt;
232 	}
233 
234 	mutex_enter(&frag6_lock);
235 
236 	/*
237 	 * Enforce upper bound on number of fragments.
238 	 * If maxfrag is 0, never accept fragments.
239 	 * If maxfrag is -1, accept all fragments without limitation.
240 	 */
241 	if (ip6_maxfrags < 0)
242 		;
243 	else if (frag6_nfrags >= (u_int)ip6_maxfrags)
244 		goto dropfrag;
245 
246 	for (q6 = ip6q.ip6q_next; q6 != &ip6q; q6 = q6->ip6q_next)
247 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
248 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
249 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
250 			break;
251 
252 	if (q6 == &ip6q) {
253 		/*
254 		 * the first fragment to arrive, create a reassembly queue.
255 		 */
256 		first_frag = 1;
257 
258 		/*
259 		 * Enforce upper bound on number of fragmented packets
260 		 * for which we attempt reassembly;
261 		 * If maxfragpackets is 0, never accept fragments.
262 		 * If maxfragpackets is -1, accept all fragments without
263 		 * limitation.
264 		 */
265 		if (ip6_maxfragpackets < 0)
266 			;
267 		else if (frag6_nfragpackets >= (u_int)ip6_maxfragpackets)
268 			goto dropfrag;
269 		frag6_nfragpackets++;
270 
271 		q6 = kmem_intr_zalloc(sizeof(struct ip6q), KM_NOSLEEP);
272 		if (q6 == NULL) {
273 			goto dropfrag;
274 		}
275 		frag6_insque(q6, &ip6q);
276 
277 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
278 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
279 		q6->ip6q_ident	= ip6f->ip6f_ident;
280 		q6->ip6q_ttl 	= IPV6_FRAGTTL;
281 		q6->ip6q_src	= ip6->ip6_src;
282 		q6->ip6q_dst	= ip6->ip6_dst;
283 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
284 
285 		q6->ip6q_nfrag = 0;
286 	}
287 
288 	/*
289 	 * If it's the 1st fragment, record the length of the
290 	 * unfragmentable part and the next header of the fragment header.
291 	 */
292 	if (fragoff == 0) {
293 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
294 		    sizeof(struct ip6_frag);
295 		q6->ip6q_nxt = ip6f->ip6f_nxt;
296 	}
297 
298 	/*
299 	 * Check that the reassembled packet would not exceed 65535 bytes
300 	 * in size. If it would exceed, discard the fragment and return an
301 	 * ICMP error.
302 	 */
303 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
304 	if (q6->ip6q_unfrglen >= 0) {
305 		/* The 1st fragment has already arrived. */
306 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
307 			mutex_exit(&frag6_lock);
308 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
309 			    offset - sizeof(struct ip6_frag) +
310 			    offsetof(struct ip6_frag, ip6f_offlg));
311 			goto done;
312 		}
313 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
314 		mutex_exit(&frag6_lock);
315 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
316 		    offset - sizeof(struct ip6_frag) +
317 		    offsetof(struct ip6_frag, ip6f_offlg));
318 		goto done;
319 	}
320 
321 	/*
322 	 * If it's the first fragment, do the above check for each
323 	 * fragment already stored in the reassembly queue.
324 	 */
325 	if (fragoff == 0) {
326 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
327 		     af6 = af6dwn) {
328 			af6dwn = af6->ip6af_down;
329 
330 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
331 			    IPV6_MAXPACKET) {
332 				struct mbuf *merr = af6->ip6af_m;
333 				struct ip6_hdr *ip6err;
334 				int erroff = af6->ip6af_offset;
335 
336 				/* dequeue the fragment. */
337 				frag6_deq(af6);
338 				kmem_intr_free(af6, sizeof(struct ip6asfrag));
339 
340 				/* adjust pointer. */
341 				ip6err = mtod(merr, struct ip6_hdr *);
342 
343 				/*
344 				 * Restore source and destination addresses
345 				 * in the erroneous IPv6 header.
346 				 */
347 				ip6err->ip6_src = q6->ip6q_src;
348 				ip6err->ip6_dst = q6->ip6q_dst;
349 
350 				icmp6_error(merr, ICMP6_PARAM_PROB,
351 				    ICMP6_PARAMPROB_HEADER,
352 				    erroff - sizeof(struct ip6_frag) +
353 				    offsetof(struct ip6_frag, ip6f_offlg));
354 			}
355 		}
356 	}
357 
358 	ip6af = kmem_intr_zalloc(sizeof(struct ip6asfrag), KM_NOSLEEP);
359 	if (ip6af == NULL) {
360 		goto dropfrag;
361 	}
362 	ip6af->ip6af_head = ip6->ip6_flow;
363 	ip6af->ip6af_len = ip6->ip6_plen;
364 	ip6af->ip6af_nxt = ip6->ip6_nxt;
365 	ip6af->ip6af_hlim = ip6->ip6_hlim;
366 	ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) != 0;
367 	ip6af->ip6af_off = fragoff;
368 	ip6af->ip6af_frglen = frgpartlen;
369 	ip6af->ip6af_offset = offset;
370 	ip6af->ip6af_m = m;
371 
372 	if (first_frag) {
373 		af6 = (struct ip6asfrag *)q6;
374 		goto insert;
375 	}
376 
377 	/*
378 	 * Find a segment which begins after this one does.
379 	 */
380 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
381 	     af6 = af6->ip6af_down)
382 		if (af6->ip6af_off > ip6af->ip6af_off)
383 			break;
384 
385 	/*
386 	 * If the incoming fragment overlaps some existing fragments in
387 	 * the reassembly queue - drop it as per RFC 5722.
388 	 */
389 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
390 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
391 			- ip6af->ip6af_off;
392 		if (i > 0) {
393 			kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
394 			goto dropfrag;
395 		}
396 	}
397 	if (af6 != (struct ip6asfrag *)q6) {
398 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
399 		if (i > 0) {
400 			kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
401 			goto dropfrag;
402 		}
403 	}
404 
405 insert:
406 	/*
407 	 * Stick new segment in its place.
408 	 */
409 	frag6_enq(ip6af, af6->ip6af_up);
410 	frag6_nfrags++;
411 	q6->ip6q_nfrag++;
412 
413 	/*
414 	 * Check for complete reassembly.
415 	 */
416 	next = 0;
417 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
418 	     af6 = af6->ip6af_down) {
419 		if (af6->ip6af_off != next) {
420 			mutex_exit(&frag6_lock);
421 			goto done;
422 		}
423 		next += af6->ip6af_frglen;
424 	}
425 	if (af6->ip6af_up->ip6af_mff) {
426 		mutex_exit(&frag6_lock);
427 		goto done;
428 	}
429 
430 	/*
431 	 * Reassembly is complete; concatenate fragments.
432 	 */
433 	ip6af = q6->ip6q_down;
434 	t = m = ip6af->ip6af_m;
435 	af6 = ip6af->ip6af_down;
436 	frag6_deq(ip6af);
437 	while (af6 != (struct ip6asfrag *)q6) {
438 		af6dwn = af6->ip6af_down;
439 		frag6_deq(af6);
440 		while (t->m_next)
441 			t = t->m_next;
442 		t->m_next = af6->ip6af_m;
443 		m_adj(t->m_next, af6->ip6af_offset);
444 		m_pkthdr_remove(t->m_next);
445 		kmem_intr_free(af6, sizeof(struct ip6asfrag));
446 		af6 = af6dwn;
447 	}
448 
449 	/* adjust offset to point where the original next header starts */
450 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
451 	kmem_intr_free(ip6af, sizeof(struct ip6asfrag));
452 	ip6 = mtod(m, struct ip6_hdr *);
453 	ip6->ip6_plen = htons(next + offset - sizeof(struct ip6_hdr));
454 	ip6->ip6_src = q6->ip6q_src;
455 	ip6->ip6_dst = q6->ip6q_dst;
456 	nxt = q6->ip6q_nxt;
457 
458 	/*
459 	 * Delete frag6 header.
460 	 */
461 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
462 		memmove((char *)ip6 + sizeof(struct ip6_frag), ip6, offset);
463 		m->m_data += sizeof(struct ip6_frag);
464 		m->m_len -= sizeof(struct ip6_frag);
465 	} else {
466 		/* this comes with no copy if the boundary is on cluster */
467 		if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
468 			frag6_remque(q6);
469 			frag6_nfrags -= q6->ip6q_nfrag;
470 			kmem_intr_free(q6, sizeof(struct ip6q));
471 			frag6_nfragpackets--;
472 			goto dropfrag;
473 		}
474 		m_adj(t, sizeof(struct ip6_frag));
475 		m_cat(m, t);
476 	}
477 
478 	frag6_remque(q6);
479 	frag6_nfrags -= q6->ip6q_nfrag;
480 	kmem_intr_free(q6, sizeof(struct ip6q));
481 	frag6_nfragpackets--;
482 
483 	{
484 		KASSERT(m->m_flags & M_PKTHDR);
485 		int plen = 0;
486 		for (t = m; t; t = t->m_next) {
487 			plen += t->m_len;
488 		}
489 		m->m_pkthdr.len = plen;
490 		/* XXX XXX: clear csum_flags? */
491 	}
492 
493 	/*
494 	 * Restore NXT to the original.
495 	 */
496 	{
497 		const int prvnxt = ip6_get_prevhdr(m, offset);
498 		uint8_t *prvnxtp;
499 
500 		IP6_EXTHDR_GET(prvnxtp, uint8_t *, m, prvnxt,
501 		    sizeof(*prvnxtp));
502 		if (prvnxtp == NULL) {
503 			goto dropfrag;
504 		}
505 		*prvnxtp = nxt;
506 	}
507 
508 	IP6_STATINC(IP6_STAT_REASSEMBLED);
509 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
510 	rtcache_unref(rt, &ro);
511 	mutex_exit(&frag6_lock);
512 
513 	/*
514 	 * Tell launch routine the next header.
515 	 */
516 	*mp = m;
517 	*offp = offset;
518 	return nxt;
519 
520  dropfrag:
521 	mutex_exit(&frag6_lock);
522 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
523 	IP6_STATINC(IP6_STAT_FRAGDROPPED);
524 	m_freem(m);
525  done:
526 	rtcache_unref(rt, &ro);
527 	return IPPROTO_DONE;
528 }
529 
530 int
531 ip6_reass_packet(struct mbuf **mp, int offset)
532 {
533 
534 	if (frag6_input(mp, &offset, IPPROTO_IPV6) == IPPROTO_DONE) {
535 		*mp = NULL;
536 		return EINVAL;
537 	}
538 	return 0;
539 }
540 
541 /*
542  * Free a fragment reassembly header and all
543  * associated datagrams.
544  */
545 static void
546 frag6_freef(struct ip6q *q6)
547 {
548 	struct ip6asfrag *af6, *down6;
549 
550 	KASSERT(mutex_owned(&frag6_lock));
551 
552 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
553 	     af6 = down6) {
554 		struct mbuf *m = af6->ip6af_m;
555 
556 		down6 = af6->ip6af_down;
557 		frag6_deq(af6);
558 
559 		/*
560 		 * Return ICMP time exceeded error for the 1st fragment.
561 		 * Just free other fragments.
562 		 */
563 		if (af6->ip6af_off == 0) {
564 			struct ip6_hdr *ip6;
565 
566 			/* adjust pointer */
567 			ip6 = mtod(m, struct ip6_hdr *);
568 
569 			/* restore source and destination addresses */
570 			ip6->ip6_src = q6->ip6q_src;
571 			ip6->ip6_dst = q6->ip6q_dst;
572 
573 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
574 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
575 		} else {
576 			m_freem(m);
577 		}
578 		kmem_intr_free(af6, sizeof(struct ip6asfrag));
579 	}
580 
581 	frag6_remque(q6);
582 	frag6_nfrags -= q6->ip6q_nfrag;
583 	kmem_intr_free(q6, sizeof(struct ip6q));
584 	frag6_nfragpackets--;
585 }
586 
587 /*
588  * Put an ip fragment on a reassembly chain.
589  * Like insque, but pointers in middle of structure.
590  */
591 void
592 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
593 {
594 
595 	KASSERT(mutex_owned(&frag6_lock));
596 
597 	af6->ip6af_up = up6;
598 	af6->ip6af_down = up6->ip6af_down;
599 	up6->ip6af_down->ip6af_up = af6;
600 	up6->ip6af_down = af6;
601 }
602 
603 /*
604  * To frag6_enq as remque is to insque.
605  */
606 void
607 frag6_deq(struct ip6asfrag *af6)
608 {
609 
610 	KASSERT(mutex_owned(&frag6_lock));
611 
612 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
613 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
614 }
615 
616 /*
617  * Insert newq after oldq.
618  */
619 void
620 frag6_insque(struct ip6q *newq, struct ip6q *oldq)
621 {
622 
623 	KASSERT(mutex_owned(&frag6_lock));
624 
625 	newq->ip6q_prev = oldq;
626 	newq->ip6q_next = oldq->ip6q_next;
627 	oldq->ip6q_next->ip6q_prev = newq;
628 	oldq->ip6q_next = newq;
629 }
630 
631 /*
632  * Unlink p6.
633  */
634 void
635 frag6_remque(struct ip6q *p6)
636 {
637 
638 	KASSERT(mutex_owned(&frag6_lock));
639 
640 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
641 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
642 }
643 
644 void
645 frag6_fasttimo(void)
646 {
647 
648 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
649 
650 	if (frag6_drainwanted) {
651 		frag6_drain();
652 		frag6_drainwanted = 0;
653 	}
654 
655 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
656 }
657 
658 /*
659  * IPv6 reassembling timer processing;
660  * if a timer expires on a reassembly
661  * queue, discard it.
662  */
663 void
664 frag6_slowtimo(void)
665 {
666 	struct ip6q *q6;
667 
668 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
669 
670 	mutex_enter(&frag6_lock);
671 	q6 = ip6q.ip6q_next;
672 	if (q6) {
673 		while (q6 != &ip6q) {
674 			--q6->ip6q_ttl;
675 			q6 = q6->ip6q_next;
676 			if (q6->ip6q_prev->ip6q_ttl == 0) {
677 				IP6_STATINC(IP6_STAT_FRAGTIMEOUT);
678 				/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
679 				frag6_freef(q6->ip6q_prev);
680 			}
681 		}
682 	}
683 
684 	/*
685 	 * If we are over the maximum number of fragments
686 	 * (due to the limit being lowered), drain off
687 	 * enough to get down to the new limit.
688 	 */
689 	while (frag6_nfragpackets > (u_int)ip6_maxfragpackets &&
690 	    ip6q.ip6q_prev) {
691 		IP6_STATINC(IP6_STAT_FRAGOVERFLOW);
692 		/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
693 		frag6_freef(ip6q.ip6q_prev);
694 	}
695 	mutex_exit(&frag6_lock);
696 
697 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
698 
699 #if 0
700 	/*
701 	 * Routing changes might produce a better route than we last used;
702 	 * make sure we notice eventually, even if forwarding only for one
703 	 * destination and the cache is never replaced.
704 	 */
705 	rtcache_free(&ip6_forward_rt);
706 	rtcache_free(&ipsrcchk_rt);
707 #endif
708 }
709 
710 void
711 frag6_drainstub(void)
712 {
713 	frag6_drainwanted = 1;
714 }
715 
716 /*
717  * Drain off all datagram fragments.
718  */
719 void
720 frag6_drain(void)
721 {
722 
723 	if (mutex_tryenter(&frag6_lock)) {
724 		while (ip6q.ip6q_next != &ip6q) {
725 			IP6_STATINC(IP6_STAT_FRAGDROPPED);
726 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
727 			frag6_freef(ip6q.ip6q_next);
728 		}
729 		mutex_exit(&frag6_lock);
730 	}
731 }
732