2 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
29 * $KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
39 #include <sys/domain.h>
40 #include <sys/protosw.h>
41 #include <sys/socket.h>
42 #include <sys/errno.h>
44 #include <sys/kernel.h>
45 #include <sys/syslog.h>
46 #include <sys/vimage.h>
49 #include <net/route.h>
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 */
60 * Define it to get a correct behavior on per-interface statistics.
61 * You will need to perform an extra routing table lookup, per fragment,
62 * to do it. This may, or may not be, a performance hit.
64 #define IN6_IFSTAT_STRICT
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 *);
72 static struct mtx ip6qlock;
74 * These fields all protected by ip6qlock.
76 static u_int frag6_nfragpackets;
77 static u_int frag6_nfrags;
78 static struct ip6q ip6q; /* ip6 reassemble queue */
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)
86 static MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header");
89 * Initialise reassembly queue and fragment identifier.
92 frag6_change(void *tag)
94 INIT_VNET_INET6(curvnet);
96 V_ip6_maxfragpackets = nmbclusters / 4;
97 V_ip6_maxfrags = nmbclusters / 4;
103 INIT_VNET_INET6(curvnet);
105 V_ip6_maxfragpackets = nmbclusters / 4;
106 V_ip6_maxfrags = nmbclusters / 4;
107 EVENTHANDLER_REGISTER(nmbclusters_change,
108 frag6_change, NULL, EVENTHANDLER_PRI_ANY);
112 V_ip6q.ip6q_next = V_ip6q.ip6q_prev = &V_ip6q;
116 * In RFC2460, fragment and reassembly rule do not agree with each other,
117 * in terms of next header field handling in fragment header.
118 * While the sender will use the same value for all of the fragmented packets,
119 * receiver is suggested not to check the consistency.
121 * fragment rule (p20):
122 * (2) A Fragment header containing:
123 * The Next Header value that identifies the first header of
124 * the Fragmentable Part of the original packet.
125 * -> next header field is same for all fragments
127 * reassembly rule (p21):
128 * The Next Header field of the last header of the Unfragmentable
129 * Part is obtained from the Next Header field of the first
130 * fragment's Fragment header.
131 * -> should grab it from the first fragment only
133 * The following note also contradicts with fragment rule - noone is going to
134 * send different fragment with different next header field.
136 * additional note (p22):
137 * The Next Header values in the Fragment headers of different
138 * fragments of the same original packet may differ. Only the value
139 * from the Offset zero fragment packet is used for reassembly.
140 * -> should grab it from the first fragment only
142 * There is no explicit reason given in the RFC. Historical reason maybe?
148 frag6_input(struct mbuf **mp, int *offp, int proto)
150 INIT_VNET_INET6(curvnet);
151 struct mbuf *m = *mp, *t;
153 struct ip6_frag *ip6f;
155 struct ip6asfrag *af6, *ip6af, *af6dwn;
156 #ifdef IN6_IFSTAT_STRICT
157 struct in6_ifaddr *ia;
159 int offset = *offp, nxt, i, next;
161 int fragoff, frgpartlen; /* must be larger than u_int16_t */
162 struct ifnet *dstifp;
165 char ip6buf[INET6_ADDRSTRLEN];
168 ip6 = mtod(m, struct ip6_hdr *);
169 #ifndef PULLDOWN_TEST
170 IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
171 ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
173 IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
175 return (IPPROTO_DONE);
179 #ifdef IN6_IFSTAT_STRICT
180 /* find the destination interface of the packet. */
181 if ((ia = ip6_getdstifaddr(m)) != NULL)
184 /* we are violating the spec, this is not the destination interface */
185 if ((m->m_flags & M_PKTHDR) != 0)
186 dstifp = m->m_pkthdr.rcvif;
189 /* jumbo payload can't contain a fragment header */
190 if (ip6->ip6_plen == 0) {
191 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
192 in6_ifstat_inc(dstifp, ifs6_reass_fail);
197 * check whether fragment packet's fragment length is
198 * multiple of 8 octets.
199 * sizeof(struct ip6_frag) == 8
200 * sizeof(struct ip6_hdr) = 40
202 if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
203 (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
204 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
205 offsetof(struct ip6_hdr, ip6_plen));
206 in6_ifstat_inc(dstifp, ifs6_reass_fail);
210 V_ip6stat.ip6s_fragments++;
211 in6_ifstat_inc(dstifp, ifs6_reass_reqd);
213 /* offset now points to data portion */
214 offset += sizeof(struct ip6_frag);
219 * Enforce upper bound on number of fragments.
220 * If maxfrag is 0, never accept fragments.
221 * If maxfrag is -1, accept all fragments without limitation.
223 if (V_ip6_maxfrags < 0)
225 else if (V_frag6_nfrags >= (u_int)V_ip6_maxfrags)
228 for (q6 = V_ip6q.ip6q_next; q6 != &V_ip6q; q6 = q6->ip6q_next)
229 if (ip6f->ip6f_ident == q6->ip6q_ident &&
230 IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
231 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst))
236 * the first fragment to arrive, create a reassembly queue.
241 * Enforce upper bound on number of fragmented packets
242 * for which we attempt reassembly;
243 * If maxfragpackets is 0, never accept fragments.
244 * If maxfragpackets is -1, accept all fragments without
247 if (V_ip6_maxfragpackets < 0)
249 else if (V_frag6_nfragpackets >= (u_int)V_ip6_maxfragpackets)
251 V_frag6_nfragpackets++;
252 q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
256 bzero(q6, sizeof(*q6));
258 frag6_insque(q6, &V_ip6q);
260 /* ip6q_nxt will be filled afterwards, from 1st fragment */
261 q6->ip6q_down = q6->ip6q_up = (struct ip6asfrag *)q6;
263 q6->ip6q_nxtp = (u_char *)nxtp;
265 q6->ip6q_ident = ip6f->ip6f_ident;
266 q6->ip6q_ttl = IPV6_FRAGTTL;
267 q6->ip6q_src = ip6->ip6_src;
268 q6->ip6q_dst = ip6->ip6_dst;
270 (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
271 q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */
277 * If it's the 1st fragment, record the length of the
278 * unfragmentable part and the next header of the fragment header.
280 fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
282 q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
283 sizeof(struct ip6_frag);
284 q6->ip6q_nxt = ip6f->ip6f_nxt;
288 * Check that the reassembled packet would not exceed 65535 bytes
290 * If it would exceed, discard the fragment and return an ICMP error.
292 frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
293 if (q6->ip6q_unfrglen >= 0) {
294 /* The 1st fragment has already arrived. */
295 if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
296 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
297 offset - sizeof(struct ip6_frag) +
298 offsetof(struct ip6_frag, ip6f_offlg));
300 return (IPPROTO_DONE);
302 } else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
303 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
304 offset - sizeof(struct ip6_frag) +
305 offsetof(struct ip6_frag, ip6f_offlg));
307 return (IPPROTO_DONE);
310 * If it's the first fragment, do the above check for each
311 * fragment already stored in the reassembly queue.
314 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
316 af6dwn = af6->ip6af_down;
318 if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
320 struct mbuf *merr = IP6_REASS_MBUF(af6);
321 struct ip6_hdr *ip6err;
322 int erroff = af6->ip6af_offset;
324 /* dequeue the fragment. */
328 /* adjust pointer. */
329 ip6err = mtod(merr, struct ip6_hdr *);
332 * Restore source and destination addresses
333 * in the erroneous IPv6 header.
335 ip6err->ip6_src = q6->ip6q_src;
336 ip6err->ip6_dst = q6->ip6q_dst;
338 icmp6_error(merr, ICMP6_PARAM_PROB,
339 ICMP6_PARAMPROB_HEADER,
340 erroff - sizeof(struct ip6_frag) +
341 offsetof(struct ip6_frag, ip6f_offlg));
346 ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
350 bzero(ip6af, sizeof(*ip6af));
351 ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
352 ip6af->ip6af_off = fragoff;
353 ip6af->ip6af_frglen = frgpartlen;
354 ip6af->ip6af_offset = offset;
355 IP6_REASS_MBUF(ip6af) = m;
358 af6 = (struct ip6asfrag *)q6;
363 * Handle ECN by comparing this segment with the first one;
364 * if CE is set, do not lose CE.
365 * drop if CE and not-ECT are mixed for the same packet.
367 ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
369 if (ecn == IPTOS_ECN_CE) {
370 if (ecn0 == IPTOS_ECN_NOTECT) {
371 free(ip6af, M_FTABLE);
374 if (ecn0 != IPTOS_ECN_CE)
375 q6->ip6q_ecn = IPTOS_ECN_CE;
377 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
378 free(ip6af, M_FTABLE);
383 * Find a segment which begins after this one does.
385 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
386 af6 = af6->ip6af_down)
387 if (af6->ip6af_off > ip6af->ip6af_off)
392 * If there is a preceding segment, it may provide some of
393 * our data already. If so, drop the data from the incoming
394 * segment. If it provides all of our data, drop us.
396 if (af6->ip6af_up != (struct ip6asfrag *)q6) {
397 i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
400 if (i >= ip6af->ip6af_frglen)
402 m_adj(IP6_REASS_MBUF(ip6af), i);
403 ip6af->ip6af_off += i;
404 ip6af->ip6af_frglen -= i;
409 * While we overlap succeeding segments trim them or,
410 * if they are completely covered, dequeue them.
412 while (af6 != (struct ip6asfrag *)q6 &&
413 ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) {
414 i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
415 if (i < af6->ip6af_frglen) {
416 af6->ip6af_frglen -= i;
418 m_adj(IP6_REASS_MBUF(af6), i);
421 af6 = af6->ip6af_down;
422 m_freem(IP6_REASS_MBUF(af6->ip6af_up));
423 frag6_deq(af6->ip6af_up);
427 * If the incoming framgent overlaps some existing fragments in
428 * the reassembly queue, drop it, since it is dangerous to override
429 * existing fragments from a security point of view.
430 * We don't know which fragment is the bad guy - here we trust
431 * fragment that came in earlier, with no real reason.
433 * Note: due to changes after disabling this part, mbuf passed to
434 * m_adj() below now does not meet the requirement.
436 if (af6->ip6af_up != (struct ip6asfrag *)q6) {
437 i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
440 #if 0 /* suppress the noisy log */
441 log(LOG_ERR, "%d bytes of a fragment from %s "
442 "overlaps the previous fragment\n",
443 i, ip6_sprintf(ip6buf, &q6->ip6q_src));
445 free(ip6af, M_FTABLE);
449 if (af6 != (struct ip6asfrag *)q6) {
450 i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
452 #if 0 /* suppress the noisy log */
453 log(LOG_ERR, "%d bytes of a fragment from %s "
454 "overlaps the succeeding fragment",
455 i, ip6_sprintf(ip6buf, &q6->ip6q_src));
457 free(ip6af, M_FTABLE);
466 * Stick new segment in its place;
467 * check for complete reassembly.
468 * Move to front of packet queue, as we are
469 * the most recently active fragmented packet.
471 frag6_enq(ip6af, af6->ip6af_up);
475 if (q6 != V_ip6q.ip6q_next) {
477 frag6_insque(q6, &V_ip6q);
481 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
482 af6 = af6->ip6af_down) {
483 if (af6->ip6af_off != next) {
487 next += af6->ip6af_frglen;
489 if (af6->ip6af_up->ip6af_mff) {
495 * Reassembly is complete; concatenate fragments.
497 ip6af = q6->ip6q_down;
498 t = m = IP6_REASS_MBUF(ip6af);
499 af6 = ip6af->ip6af_down;
501 while (af6 != (struct ip6asfrag *)q6) {
502 af6dwn = af6->ip6af_down;
506 t->m_next = IP6_REASS_MBUF(af6);
507 m_adj(t->m_next, af6->ip6af_offset);
512 /* adjust offset to point where the original next header starts */
513 offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
514 free(ip6af, M_FTABLE);
515 ip6 = mtod(m, struct ip6_hdr *);
516 ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr));
517 if (q6->ip6q_ecn == IPTOS_ECN_CE)
518 ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
521 *q6->ip6q_nxtp = (u_char)(nxt & 0xff);
524 /* Delete frag6 header */
525 if (m->m_len >= offset + sizeof(struct ip6_frag)) {
526 /* This is the only possible case with !PULLDOWN_TEST */
527 ovbcopy((caddr_t)ip6, (caddr_t)ip6 + sizeof(struct ip6_frag),
529 m->m_data += sizeof(struct ip6_frag);
530 m->m_len -= sizeof(struct ip6_frag);
532 /* this comes with no copy if the boundary is on cluster */
533 if ((t = m_split(m, offset, M_DONTWAIT)) == NULL) {
535 V_frag6_nfrags -= q6->ip6q_nfrag;
537 V_frag6_nfragpackets--;
540 m_adj(t, sizeof(struct ip6_frag));
545 * Store NXT to the original.
548 char *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
553 V_frag6_nfrags -= q6->ip6q_nfrag;
555 V_frag6_nfragpackets--;
557 if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
559 for (t = m; t; t = t->m_next)
561 m->m_pkthdr.len = plen;
564 V_ip6stat.ip6s_reassembled++;
565 in6_ifstat_inc(dstifp, ifs6_reass_ok);
568 * Tell launch routine the next header
579 in6_ifstat_inc(dstifp, ifs6_reass_fail);
580 V_ip6stat.ip6s_fragdropped++;
586 * Free a fragment reassembly header and all
587 * associated datagrams.
590 frag6_freef(struct ip6q *q6)
592 INIT_VNET_INET6(curvnet);
593 struct ip6asfrag *af6, *down6;
597 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
599 struct mbuf *m = IP6_REASS_MBUF(af6);
601 down6 = af6->ip6af_down;
605 * Return ICMP time exceeded error for the 1st fragment.
606 * Just free other fragments.
608 if (af6->ip6af_off == 0) {
612 ip6 = mtod(m, struct ip6_hdr *);
614 /* restore source and destination addresses */
615 ip6->ip6_src = q6->ip6q_src;
616 ip6->ip6_dst = q6->ip6q_dst;
618 icmp6_error(m, ICMP6_TIME_EXCEEDED,
619 ICMP6_TIME_EXCEED_REASSEMBLY, 0);
625 V_frag6_nfrags -= q6->ip6q_nfrag;
627 V_frag6_nfragpackets--;
631 * Put an ip fragment on a reassembly chain.
632 * Like insque, but pointers in middle of structure.
635 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
641 af6->ip6af_down = up6->ip6af_down;
642 up6->ip6af_down->ip6af_up = af6;
643 up6->ip6af_down = af6;
647 * To frag6_enq as remque is to insque.
650 frag6_deq(struct ip6asfrag *af6)
655 af6->ip6af_up->ip6af_down = af6->ip6af_down;
656 af6->ip6af_down->ip6af_up = af6->ip6af_up;
660 frag6_insque(struct ip6q *new, struct ip6q *old)
665 new->ip6q_prev = old;
666 new->ip6q_next = old->ip6q_next;
667 old->ip6q_next->ip6q_prev= new;
668 old->ip6q_next = new;
672 frag6_remque(struct ip6q *p6)
677 p6->ip6q_prev->ip6q_next = p6->ip6q_next;
678 p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
682 * IPv6 reassembling timer processing;
683 * if a timer expires on a reassembly
689 VNET_ITERATOR_DECL(vnet_iter);
694 VNET_FOREACH(vnet_iter) {
695 CURVNET_SET(vnet_iter);
696 INIT_VNET_INET6(vnet_iter);
697 q6 = V_ip6q.ip6q_next;
699 while (q6 != &V_ip6q) {
702 if (q6->ip6q_prev->ip6q_ttl == 0) {
703 V_ip6stat.ip6s_fragtimeout++;
704 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
705 frag6_freef(q6->ip6q_prev);
709 * If we are over the maximum number of fragments
710 * (due to the limit being lowered), drain off
711 * enough to get down to the new limit.
713 while (V_frag6_nfragpackets > (u_int)V_ip6_maxfragpackets &&
715 V_ip6stat.ip6s_fragoverflow++;
716 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
717 frag6_freef(V_ip6q.ip6q_prev);
726 * Routing changes might produce a better route than we last used;
727 * make sure we notice eventually, even if forwarding only for one
728 * destination and the cache is never replaced.
730 if (V_ip6_forward_rt.ro_rt) {
731 RTFREE(V_ip6_forward_rt.ro_rt);
732 V_ip6_forward_rt.ro_rt = 0;
734 if (ipsrcchk_rt.ro_rt) {
735 RTFREE(ipsrcchk_rt.ro_rt);
736 ipsrcchk_rt.ro_rt = 0;
742 * Drain off all datagram fragments.
747 VNET_ITERATOR_DECL(vnet_iter);
749 if (IP6Q_TRYLOCK() == 0)
752 VNET_FOREACH(vnet_iter) {
753 CURVNET_SET(vnet_iter);
754 INIT_VNET_INET6(vnet_iter);
755 while (V_ip6q.ip6q_next != &V_ip6q) {
756 V_ip6stat.ip6s_fragdropped++;
757 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
758 frag6_freef(V_ip6q.ip6q_next);