xref: /freebsd/sys/netinet6/frag6.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
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 "opt_rss.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/domain.h>
42 #include <sys/eventhandler.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/errno.h>
46 #include <sys/time.h>
47 #include <sys/kernel.h>
48 #include <sys/syslog.h>
49 
50 #include <net/if.h>
51 #include <net/if_var.h>
52 #include <net/netisr.h>
53 #include <net/route.h>
54 #include <net/vnet.h>
55 
56 #include <netinet/in.h>
57 #include <netinet/in_var.h>
58 #include <netinet/ip6.h>
59 #include <netinet6/ip6_var.h>
60 #include <netinet/icmp6.h>
61 #include <netinet/in_systm.h>	/* for ECN definitions */
62 #include <netinet/ip.h>		/* for ECN definitions */
63 
64 #include <security/mac/mac_framework.h>
65 
66 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *);
67 static void frag6_deq(struct ip6asfrag *);
68 static void frag6_insque(struct ip6q *, struct ip6q *);
69 static void frag6_remque(struct ip6q *);
70 static void frag6_freef(struct ip6q *);
71 
72 static struct mtx ip6qlock;
73 /*
74  * These fields all protected by ip6qlock.
75  */
76 static VNET_DEFINE(u_int, frag6_nfragpackets);
77 static VNET_DEFINE(u_int, frag6_nfrags);
78 static VNET_DEFINE(struct ip6q, ip6q);	/* ip6 reassemble queue */
79 
80 #define	V_frag6_nfragpackets		VNET(frag6_nfragpackets)
81 #define	V_frag6_nfrags			VNET(frag6_nfrags)
82 #define	V_ip6q				VNET(ip6q)
83 
84 #define	IP6Q_LOCK_INIT()	mtx_init(&ip6qlock, "ip6qlock", NULL, MTX_DEF);
85 #define	IP6Q_LOCK()		mtx_lock(&ip6qlock)
86 #define	IP6Q_TRYLOCK()		mtx_trylock(&ip6qlock)
87 #define	IP6Q_LOCK_ASSERT()	mtx_assert(&ip6qlock, MA_OWNED)
88 #define	IP6Q_UNLOCK()		mtx_unlock(&ip6qlock)
89 
90 static MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header");
91 
92 /*
93  * Initialise reassembly queue and fragment identifier.
94  */
95 static void
96 frag6_change(void *tag)
97 {
98 
99 	V_ip6_maxfragpackets = nmbclusters / 4;
100 	V_ip6_maxfrags = nmbclusters / 4;
101 }
102 
103 void
104 frag6_init(void)
105 {
106 
107 	V_ip6_maxfragpackets = nmbclusters / 4;
108 	V_ip6_maxfrags = nmbclusters / 4;
109 	V_ip6q.ip6q_next = V_ip6q.ip6q_prev = &V_ip6q;
110 
111 	if (!IS_DEFAULT_VNET(curvnet))
112 		return;
113 
114 	EVENTHANDLER_REGISTER(nmbclusters_change,
115 	    frag6_change, NULL, EVENTHANDLER_PRI_ANY);
116 
117 	IP6Q_LOCK_INIT();
118 }
119 
120 /*
121  * In RFC2460, fragment and reassembly rule do not agree with each other,
122  * in terms of next header field handling in fragment header.
123  * While the sender will use the same value for all of the fragmented packets,
124  * receiver is suggested not to check the consistency.
125  *
126  * fragment rule (p20):
127  *	(2) A Fragment header containing:
128  *	The Next Header value that identifies the first header of
129  *	the Fragmentable Part of the original packet.
130  *		-> next header field is same for all fragments
131  *
132  * reassembly rule (p21):
133  *	The Next Header field of the last header of the Unfragmentable
134  *	Part is obtained from the Next Header field of the first
135  *	fragment's Fragment header.
136  *		-> should grab it from the first fragment only
137  *
138  * The following note also contradicts with fragment rule - no one is going to
139  * send different fragment with different next header field.
140  *
141  * additional note (p22):
142  *	The Next Header values in the Fragment headers of different
143  *	fragments of the same original packet may differ.  Only the value
144  *	from the Offset zero fragment packet is used for reassembly.
145  *		-> should grab it from the first fragment only
146  *
147  * There is no explicit reason given in the RFC.  Historical reason maybe?
148  */
149 /*
150  * Fragment input
151  */
152 int
153 frag6_input(struct mbuf **mp, int *offp, int proto)
154 {
155 	struct mbuf *m = *mp, *t;
156 	struct ip6_hdr *ip6;
157 	struct ip6_frag *ip6f;
158 	struct ip6q *q6;
159 	struct ip6asfrag *af6, *ip6af, *af6dwn;
160 	struct in6_ifaddr *ia;
161 	int offset = *offp, nxt, i, next;
162 	int first_frag = 0;
163 	int fragoff, frgpartlen;	/* must be larger than u_int16_t */
164 	struct ifnet *dstifp;
165 	u_int8_t ecn, ecn0;
166 #ifdef RSS
167 	struct m_tag *mtag;
168 	struct ip6_direct_ctx *ip6dc;
169 #endif
170 
171 #if 0
172 	char ip6buf[INET6_ADDRSTRLEN];
173 #endif
174 
175 	ip6 = mtod(m, struct ip6_hdr *);
176 #ifndef PULLDOWN_TEST
177 	IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
178 	ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
179 #else
180 	IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
181 	if (ip6f == NULL)
182 		return (IPPROTO_DONE);
183 #endif
184 
185 	dstifp = NULL;
186 	/* find the destination interface of the packet. */
187 	ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
188 	if (ia != NULL) {
189 		dstifp = ia->ia_ifp;
190 		ifa_free(&ia->ia_ifa);
191 	}
192 	/* jumbo payload can't contain a fragment header */
193 	if (ip6->ip6_plen == 0) {
194 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
195 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
196 		return IPPROTO_DONE;
197 	}
198 
199 	/*
200 	 * check whether fragment packet's fragment length is
201 	 * multiple of 8 octets.
202 	 * sizeof(struct ip6_frag) == 8
203 	 * sizeof(struct ip6_hdr) = 40
204 	 */
205 	if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
206 	    (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
207 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
208 		    offsetof(struct ip6_hdr, ip6_plen));
209 		in6_ifstat_inc(dstifp, ifs6_reass_fail);
210 		return IPPROTO_DONE;
211 	}
212 
213 	IP6STAT_INC(ip6s_fragments);
214 	in6_ifstat_inc(dstifp, ifs6_reass_reqd);
215 
216 	/* offset now points to data portion */
217 	offset += sizeof(struct ip6_frag);
218 
219 	/*
220 	 * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0)
221 	 * upfront, unrelated to any reassembly.  Just skip the fragment header.
222 	 */
223 	if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
224 		/* XXX-BZ we want dedicated counters for this. */
225 		IP6STAT_INC(ip6s_reassembled);
226 		in6_ifstat_inc(dstifp, ifs6_reass_ok);
227 		*offp = offset;
228 		return (ip6f->ip6f_nxt);
229 	}
230 
231 	IP6Q_LOCK();
232 
233 	/*
234 	 * Enforce upper bound on number of fragments.
235 	 * If maxfrag is 0, never accept fragments.
236 	 * If maxfrag is -1, accept all fragments without limitation.
237 	 */
238 	if (V_ip6_maxfrags < 0)
239 		;
240 	else if (V_frag6_nfrags >= (u_int)V_ip6_maxfrags)
241 		goto dropfrag;
242 
243 	for (q6 = V_ip6q.ip6q_next; q6 != &V_ip6q; q6 = q6->ip6q_next)
244 		if (ip6f->ip6f_ident == q6->ip6q_ident &&
245 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
246 		    IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
247 #ifdef MAC
248 		    && mac_ip6q_match(m, q6)
249 #endif
250 		    )
251 			break;
252 
253 	if (q6 == &V_ip6q) {
254 		/*
255 		 * the first fragment to arrive, create a reassembly queue.
256 		 */
257 		first_frag = 1;
258 
259 		/*
260 		 * Enforce upper bound on number of fragmented packets
261 		 * for which we attempt reassembly;
262 		 * If maxfragpackets is 0, never accept fragments.
263 		 * If maxfragpackets is -1, accept all fragments without
264 		 * limitation.
265 		 */
266 		if (V_ip6_maxfragpackets < 0)
267 			;
268 		else if (V_frag6_nfragpackets >= (u_int)V_ip6_maxfragpackets)
269 			goto dropfrag;
270 		V_frag6_nfragpackets++;
271 		q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
272 		    M_NOWAIT);
273 		if (q6 == NULL)
274 			goto dropfrag;
275 		bzero(q6, sizeof(*q6));
276 #ifdef MAC
277 		if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
278 			free(q6, M_FTABLE);
279 			goto dropfrag;
280 		}
281 		mac_ip6q_create(m, q6);
282 #endif
283 		frag6_insque(q6, &V_ip6q);
284 
285 		/* ip6q_nxt will be filled afterwards, from 1st fragment */
286 		q6->ip6q_down	= q6->ip6q_up = (struct ip6asfrag *)q6;
287 #ifdef notyet
288 		q6->ip6q_nxtp	= (u_char *)nxtp;
289 #endif
290 		q6->ip6q_ident	= ip6f->ip6f_ident;
291 		q6->ip6q_ttl	= IPV6_FRAGTTL;
292 		q6->ip6q_src	= ip6->ip6_src;
293 		q6->ip6q_dst	= ip6->ip6_dst;
294 		q6->ip6q_ecn	=
295 		    (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
296 		q6->ip6q_unfrglen = -1;	/* The 1st fragment has not arrived. */
297 
298 		q6->ip6q_nfrag = 0;
299 	}
300 
301 	/*
302 	 * If it's the 1st fragment, record the length of the
303 	 * unfragmentable part and the next header of the fragment header.
304 	 */
305 	fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
306 	if (fragoff == 0) {
307 		q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
308 		    sizeof(struct ip6_frag);
309 		q6->ip6q_nxt = ip6f->ip6f_nxt;
310 	}
311 
312 	/*
313 	 * Check that the reassembled packet would not exceed 65535 bytes
314 	 * in size.
315 	 * If it would exceed, discard the fragment and return an ICMP error.
316 	 */
317 	frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
318 	if (q6->ip6q_unfrglen >= 0) {
319 		/* The 1st fragment has already arrived. */
320 		if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
321 			icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
322 			    offset - sizeof(struct ip6_frag) +
323 			    offsetof(struct ip6_frag, ip6f_offlg));
324 			IP6Q_UNLOCK();
325 			return (IPPROTO_DONE);
326 		}
327 	} else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
328 		icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
329 		    offset - sizeof(struct ip6_frag) +
330 		    offsetof(struct ip6_frag, ip6f_offlg));
331 		IP6Q_UNLOCK();
332 		return (IPPROTO_DONE);
333 	}
334 	/*
335 	 * If it's the first fragment, do the above check for each
336 	 * fragment already stored in the reassembly queue.
337 	 */
338 	if (fragoff == 0) {
339 		for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
340 		     af6 = af6dwn) {
341 			af6dwn = af6->ip6af_down;
342 
343 			if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
344 			    IPV6_MAXPACKET) {
345 				struct mbuf *merr = IP6_REASS_MBUF(af6);
346 				struct ip6_hdr *ip6err;
347 				int erroff = af6->ip6af_offset;
348 
349 				/* dequeue the fragment. */
350 				frag6_deq(af6);
351 				free(af6, M_FTABLE);
352 
353 				/* adjust pointer. */
354 				ip6err = mtod(merr, struct ip6_hdr *);
355 
356 				/*
357 				 * Restore source and destination addresses
358 				 * in the erroneous IPv6 header.
359 				 */
360 				ip6err->ip6_src = q6->ip6q_src;
361 				ip6err->ip6_dst = q6->ip6q_dst;
362 
363 				icmp6_error(merr, ICMP6_PARAM_PROB,
364 				    ICMP6_PARAMPROB_HEADER,
365 				    erroff - sizeof(struct ip6_frag) +
366 				    offsetof(struct ip6_frag, ip6f_offlg));
367 			}
368 		}
369 	}
370 
371 	ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
372 	    M_NOWAIT);
373 	if (ip6af == NULL)
374 		goto dropfrag;
375 	bzero(ip6af, sizeof(*ip6af));
376 	ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
377 	ip6af->ip6af_off = fragoff;
378 	ip6af->ip6af_frglen = frgpartlen;
379 	ip6af->ip6af_offset = offset;
380 	IP6_REASS_MBUF(ip6af) = m;
381 
382 	if (first_frag) {
383 		af6 = (struct ip6asfrag *)q6;
384 		goto insert;
385 	}
386 
387 	/*
388 	 * Handle ECN by comparing this segment with the first one;
389 	 * if CE is set, do not lose CE.
390 	 * drop if CE and not-ECT are mixed for the same packet.
391 	 */
392 	ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
393 	ecn0 = q6->ip6q_ecn;
394 	if (ecn == IPTOS_ECN_CE) {
395 		if (ecn0 == IPTOS_ECN_NOTECT) {
396 			free(ip6af, M_FTABLE);
397 			goto dropfrag;
398 		}
399 		if (ecn0 != IPTOS_ECN_CE)
400 			q6->ip6q_ecn = IPTOS_ECN_CE;
401 	}
402 	if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
403 		free(ip6af, M_FTABLE);
404 		goto dropfrag;
405 	}
406 
407 	/*
408 	 * Find a segment which begins after this one does.
409 	 */
410 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
411 	     af6 = af6->ip6af_down)
412 		if (af6->ip6af_off > ip6af->ip6af_off)
413 			break;
414 
415 #if 0
416 	/*
417 	 * If there is a preceding segment, it may provide some of
418 	 * our data already.  If so, drop the data from the incoming
419 	 * segment.  If it provides all of our data, drop us.
420 	 */
421 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
422 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
423 			- ip6af->ip6af_off;
424 		if (i > 0) {
425 			if (i >= ip6af->ip6af_frglen)
426 				goto dropfrag;
427 			m_adj(IP6_REASS_MBUF(ip6af), i);
428 			ip6af->ip6af_off += i;
429 			ip6af->ip6af_frglen -= i;
430 		}
431 	}
432 
433 	/*
434 	 * While we overlap succeeding segments trim them or,
435 	 * if they are completely covered, dequeue them.
436 	 */
437 	while (af6 != (struct ip6asfrag *)q6 &&
438 	       ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
439 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
440 		if (i < af6->ip6af_frglen) {
441 			af6->ip6af_frglen -= i;
442 			af6->ip6af_off += i;
443 			m_adj(IP6_REASS_MBUF(af6), i);
444 			break;
445 		}
446 		af6 = af6->ip6af_down;
447 		m_freem(IP6_REASS_MBUF(af6->ip6af_up));
448 		frag6_deq(af6->ip6af_up);
449 	}
450 #else
451 	/*
452 	 * If the incoming framgent overlaps some existing fragments in
453 	 * the reassembly queue, drop it, since it is dangerous to override
454 	 * existing fragments from a security point of view.
455 	 * We don't know which fragment is the bad guy - here we trust
456 	 * fragment that came in earlier, with no real reason.
457 	 *
458 	 * Note: due to changes after disabling this part, mbuf passed to
459 	 * m_adj() below now does not meet the requirement.
460 	 */
461 	if (af6->ip6af_up != (struct ip6asfrag *)q6) {
462 		i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
463 			- ip6af->ip6af_off;
464 		if (i > 0) {
465 #if 0				/* suppress the noisy log */
466 			log(LOG_ERR, "%d bytes of a fragment from %s "
467 			    "overlaps the previous fragment\n",
468 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
469 #endif
470 			free(ip6af, M_FTABLE);
471 			goto dropfrag;
472 		}
473 	}
474 	if (af6 != (struct ip6asfrag *)q6) {
475 		i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
476 		if (i > 0) {
477 #if 0				/* suppress the noisy log */
478 			log(LOG_ERR, "%d bytes of a fragment from %s "
479 			    "overlaps the succeeding fragment",
480 			    i, ip6_sprintf(ip6buf, &q6->ip6q_src));
481 #endif
482 			free(ip6af, M_FTABLE);
483 			goto dropfrag;
484 		}
485 	}
486 #endif
487 
488 insert:
489 #ifdef MAC
490 	if (!first_frag)
491 		mac_ip6q_update(m, q6);
492 #endif
493 
494 	/*
495 	 * Stick new segment in its place;
496 	 * check for complete reassembly.
497 	 * Move to front of packet queue, as we are
498 	 * the most recently active fragmented packet.
499 	 */
500 	frag6_enq(ip6af, af6->ip6af_up);
501 	V_frag6_nfrags++;
502 	q6->ip6q_nfrag++;
503 #if 0 /* xxx */
504 	if (q6 != V_ip6q.ip6q_next) {
505 		frag6_remque(q6);
506 		frag6_insque(q6, &V_ip6q);
507 	}
508 #endif
509 	next = 0;
510 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
511 	     af6 = af6->ip6af_down) {
512 		if (af6->ip6af_off != next) {
513 			IP6Q_UNLOCK();
514 			return IPPROTO_DONE;
515 		}
516 		next += af6->ip6af_frglen;
517 	}
518 	if (af6->ip6af_up->ip6af_mff) {
519 		IP6Q_UNLOCK();
520 		return IPPROTO_DONE;
521 	}
522 
523 	/*
524 	 * Reassembly is complete; concatenate fragments.
525 	 */
526 	ip6af = q6->ip6q_down;
527 	t = m = IP6_REASS_MBUF(ip6af);
528 	af6 = ip6af->ip6af_down;
529 	frag6_deq(ip6af);
530 	while (af6 != (struct ip6asfrag *)q6) {
531 		m->m_pkthdr.csum_flags &=
532 		    IP6_REASS_MBUF(af6)->m_pkthdr.csum_flags;
533 		m->m_pkthdr.csum_data +=
534 		    IP6_REASS_MBUF(af6)->m_pkthdr.csum_data;
535 
536 		af6dwn = af6->ip6af_down;
537 		frag6_deq(af6);
538 		while (t->m_next)
539 			t = t->m_next;
540 		m_adj(IP6_REASS_MBUF(af6), af6->ip6af_offset);
541 		m_cat(t, IP6_REASS_MBUF(af6));
542 		free(af6, M_FTABLE);
543 		af6 = af6dwn;
544 	}
545 
546 	while (m->m_pkthdr.csum_data & 0xffff0000)
547 		m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
548 		    (m->m_pkthdr.csum_data >> 16);
549 
550 	/* adjust offset to point where the original next header starts */
551 	offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
552 	free(ip6af, M_FTABLE);
553 	ip6 = mtod(m, struct ip6_hdr *);
554 	ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
555 	if (q6->ip6q_ecn == IPTOS_ECN_CE)
556 		ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
557 	nxt = q6->ip6q_nxt;
558 #ifdef notyet
559 	*q6->ip6q_nxtp = (u_char)(nxt & 0xff);
560 #endif
561 
562 	if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) {
563 		frag6_remque(q6);
564 		V_frag6_nfrags -= q6->ip6q_nfrag;
565 #ifdef MAC
566 		mac_ip6q_destroy(q6);
567 #endif
568 		free(q6, M_FTABLE);
569 		V_frag6_nfragpackets--;
570 
571 		goto dropfrag;
572 	}
573 
574 	/*
575 	 * Store NXT to the original.
576 	 */
577 	{
578 		char *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
579 		*prvnxtp = nxt;
580 	}
581 
582 	frag6_remque(q6);
583 	V_frag6_nfrags -= q6->ip6q_nfrag;
584 #ifdef MAC
585 	mac_ip6q_reassemble(q6, m);
586 	mac_ip6q_destroy(q6);
587 #endif
588 	free(q6, M_FTABLE);
589 	V_frag6_nfragpackets--;
590 
591 	if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
592 		int plen = 0;
593 		for (t = m; t; t = t->m_next)
594 			plen += t->m_len;
595 		m->m_pkthdr.len = plen;
596 	}
597 
598 #ifdef RSS
599 	mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc),
600 	    M_NOWAIT);
601 	if (mtag == NULL)
602 		goto dropfrag;
603 
604 	ip6dc = (struct ip6_direct_ctx *)(mtag + 1);
605 	ip6dc->ip6dc_nxt = nxt;
606 	ip6dc->ip6dc_off = offset;
607 
608 	m_tag_prepend(m, mtag);
609 #endif
610 
611 	IP6Q_UNLOCK();
612 	IP6STAT_INC(ip6s_reassembled);
613 	in6_ifstat_inc(dstifp, ifs6_reass_ok);
614 
615 #ifdef RSS
616 	/*
617 	 * Queue/dispatch for reprocessing.
618 	 */
619 	netisr_dispatch(NETISR_IPV6_DIRECT, m);
620 	return IPPROTO_DONE;
621 #endif
622 
623 	/*
624 	 * Tell launch routine the next header
625 	 */
626 
627 	*mp = m;
628 	*offp = offset;
629 
630 	return nxt;
631 
632  dropfrag:
633 	IP6Q_UNLOCK();
634 	in6_ifstat_inc(dstifp, ifs6_reass_fail);
635 	IP6STAT_INC(ip6s_fragdropped);
636 	m_freem(m);
637 	return IPPROTO_DONE;
638 }
639 
640 /*
641  * Free a fragment reassembly header and all
642  * associated datagrams.
643  */
644 void
645 frag6_freef(struct ip6q *q6)
646 {
647 	struct ip6asfrag *af6, *down6;
648 
649 	IP6Q_LOCK_ASSERT();
650 
651 	for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
652 	     af6 = down6) {
653 		struct mbuf *m = IP6_REASS_MBUF(af6);
654 
655 		down6 = af6->ip6af_down;
656 		frag6_deq(af6);
657 
658 		/*
659 		 * Return ICMP time exceeded error for the 1st fragment.
660 		 * Just free other fragments.
661 		 */
662 		if (af6->ip6af_off == 0) {
663 			struct ip6_hdr *ip6;
664 
665 			/* adjust pointer */
666 			ip6 = mtod(m, struct ip6_hdr *);
667 
668 			/* restore source and destination addresses */
669 			ip6->ip6_src = q6->ip6q_src;
670 			ip6->ip6_dst = q6->ip6q_dst;
671 
672 			icmp6_error(m, ICMP6_TIME_EXCEEDED,
673 				    ICMP6_TIME_EXCEED_REASSEMBLY, 0);
674 		} else
675 			m_freem(m);
676 		free(af6, M_FTABLE);
677 	}
678 	frag6_remque(q6);
679 	V_frag6_nfrags -= q6->ip6q_nfrag;
680 #ifdef MAC
681 	mac_ip6q_destroy(q6);
682 #endif
683 	free(q6, M_FTABLE);
684 	V_frag6_nfragpackets--;
685 }
686 
687 /*
688  * Put an ip fragment on a reassembly chain.
689  * Like insque, but pointers in middle of structure.
690  */
691 void
692 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
693 {
694 
695 	IP6Q_LOCK_ASSERT();
696 
697 	af6->ip6af_up = up6;
698 	af6->ip6af_down = up6->ip6af_down;
699 	up6->ip6af_down->ip6af_up = af6;
700 	up6->ip6af_down = af6;
701 }
702 
703 /*
704  * To frag6_enq as remque is to insque.
705  */
706 void
707 frag6_deq(struct ip6asfrag *af6)
708 {
709 
710 	IP6Q_LOCK_ASSERT();
711 
712 	af6->ip6af_up->ip6af_down = af6->ip6af_down;
713 	af6->ip6af_down->ip6af_up = af6->ip6af_up;
714 }
715 
716 void
717 frag6_insque(struct ip6q *new, struct ip6q *old)
718 {
719 
720 	IP6Q_LOCK_ASSERT();
721 
722 	new->ip6q_prev = old;
723 	new->ip6q_next = old->ip6q_next;
724 	old->ip6q_next->ip6q_prev= new;
725 	old->ip6q_next = new;
726 }
727 
728 void
729 frag6_remque(struct ip6q *p6)
730 {
731 
732 	IP6Q_LOCK_ASSERT();
733 
734 	p6->ip6q_prev->ip6q_next = p6->ip6q_next;
735 	p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
736 }
737 
738 /*
739  * IPv6 reassembling timer processing;
740  * if a timer expires on a reassembly
741  * queue, discard it.
742  */
743 void
744 frag6_slowtimo(void)
745 {
746 	VNET_ITERATOR_DECL(vnet_iter);
747 	struct ip6q *q6;
748 
749 	VNET_LIST_RLOCK_NOSLEEP();
750 	IP6Q_LOCK();
751 	VNET_FOREACH(vnet_iter) {
752 		CURVNET_SET(vnet_iter);
753 		q6 = V_ip6q.ip6q_next;
754 		if (q6)
755 			while (q6 != &V_ip6q) {
756 				--q6->ip6q_ttl;
757 				q6 = q6->ip6q_next;
758 				if (q6->ip6q_prev->ip6q_ttl == 0) {
759 					IP6STAT_INC(ip6s_fragtimeout);
760 					/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
761 					frag6_freef(q6->ip6q_prev);
762 				}
763 			}
764 		/*
765 		 * If we are over the maximum number of fragments
766 		 * (due to the limit being lowered), drain off
767 		 * enough to get down to the new limit.
768 		 */
769 		while (V_frag6_nfragpackets > (u_int)V_ip6_maxfragpackets &&
770 		    V_ip6q.ip6q_prev) {
771 			IP6STAT_INC(ip6s_fragoverflow);
772 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
773 			frag6_freef(V_ip6q.ip6q_prev);
774 		}
775 		CURVNET_RESTORE();
776 	}
777 	IP6Q_UNLOCK();
778 	VNET_LIST_RUNLOCK_NOSLEEP();
779 }
780 
781 /*
782  * Drain off all datagram fragments.
783  */
784 void
785 frag6_drain(void)
786 {
787 	VNET_ITERATOR_DECL(vnet_iter);
788 
789 	VNET_LIST_RLOCK_NOSLEEP();
790 	if (IP6Q_TRYLOCK() == 0) {
791 		VNET_LIST_RUNLOCK_NOSLEEP();
792 		return;
793 	}
794 	VNET_FOREACH(vnet_iter) {
795 		CURVNET_SET(vnet_iter);
796 		while (V_ip6q.ip6q_next != &V_ip6q) {
797 			IP6STAT_INC(ip6s_fragdropped);
798 			/* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
799 			frag6_freef(V_ip6q.ip6q_next);
800 		}
801 		CURVNET_RESTORE();
802 	}
803 	IP6Q_UNLOCK();
804 	VNET_LIST_RUNLOCK_NOSLEEP();
805 }
806 
807 int
808 ip6_deletefraghdr(struct mbuf *m, int offset, int wait)
809 {
810 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
811 	struct mbuf *t;
812 
813 	/* Delete frag6 header. */
814 	if (m->m_len >= offset + sizeof(struct ip6_frag)) {
815 		/* This is the only possible case with !PULLDOWN_TEST. */
816 		bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag),
817 		    offset);
818 		m->m_data += sizeof(struct ip6_frag);
819 		m->m_len -= sizeof(struct ip6_frag);
820 	} else {
821 		/* This comes with no copy if the boundary is on cluster. */
822 		if ((t = m_split(m, offset, wait)) == NULL)
823 			return (ENOMEM);
824 		m_adj(t, sizeof(struct ip6_frag));
825 		m_cat(m, t);
826 	}
827 
828 	return (0);
829 }
830