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>
48 #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 */
59 #include <security/mac/mac_framework.h>
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.
66 #define IN6_IFSTAT_STRICT
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 *);
74 static struct mtx ip6qlock;
76 * These fields all protected by ip6qlock.
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 */
82 #define V_frag6_nfragpackets VNET(frag6_nfragpackets)
83 #define V_frag6_nfrags VNET(frag6_nfrags)
84 #define V_ip6q VNET(ip6q)
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)
92 static MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header");
95 * Initialise reassembly queue and fragment identifier.
98 frag6_change(void *tag)
101 V_ip6_maxfragpackets = nmbclusters / 4;
102 V_ip6_maxfrags = nmbclusters / 4;
109 V_ip6_maxfragpackets = nmbclusters / 4;
110 V_ip6_maxfrags = nmbclusters / 4;
111 V_ip6q.ip6q_next = V_ip6q.ip6q_prev = &V_ip6q;
113 if (!IS_DEFAULT_VNET(curvnet))
116 EVENTHANDLER_REGISTER(nmbclusters_change,
117 frag6_change, NULL, EVENTHANDLER_PRI_ANY);
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.
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
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
140 * The following note also contradicts with fragment rule - noone is going to
141 * send different fragment with different next header field.
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
149 * There is no explicit reason given in the RFC. Historical reason maybe?
155 frag6_input(struct mbuf **mp, int *offp, int proto)
157 struct mbuf *m = *mp, *t;
159 struct ip6_frag *ip6f;
161 struct ip6asfrag *af6, *ip6af, *af6dwn;
162 #ifdef IN6_IFSTAT_STRICT
163 struct in6_ifaddr *ia;
165 int offset = *offp, nxt, i, next;
167 int fragoff, frgpartlen; /* must be larger than u_int16_t */
168 struct ifnet *dstifp;
171 char ip6buf[INET6_ADDRSTRLEN];
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);
179 IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
181 return (IPPROTO_DONE);
185 #ifdef IN6_IFSTAT_STRICT
186 /* find the destination interface of the packet. */
187 if ((ia = ip6_getdstifaddr(m)) != NULL) {
189 ifa_free(&ia->ia_ifa);
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;
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);
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
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);
218 IP6STAT_INC(ip6s_fragments);
219 in6_ifstat_inc(dstifp, ifs6_reass_reqd);
221 /* offset now points to data portion */
222 offset += sizeof(struct ip6_frag);
225 * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0)
226 * upfront, unrelated to any reassembly. Just skip the fragment header.
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);
233 return (ip6f->ip6f_nxt);
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.
243 if (V_ip6_maxfrags < 0)
245 else if (V_frag6_nfrags >= (u_int)V_ip6_maxfrags)
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)
253 && mac_ip6q_match(m, q6)
260 * the first fragment to arrive, create a reassembly queue.
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
271 if (V_ip6_maxfragpackets < 0)
273 else if (V_frag6_nfragpackets >= (u_int)V_ip6_maxfragpackets)
275 V_frag6_nfragpackets++;
276 q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE,
280 bzero(q6, sizeof(*q6));
282 if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
286 mac_ip6q_create(m, q6);
288 frag6_insque(q6, &V_ip6q);
290 /* ip6q_nxt will be filled afterwards, from 1st fragment */
291 q6->ip6q_down = q6->ip6q_up = (struct ip6asfrag *)q6;
293 q6->ip6q_nxtp = (u_char *)nxtp;
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;
300 (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
301 q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */
307 * If it's the 1st fragment, record the length of the
308 * unfragmentable part and the next header of the fragment header.
310 fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
312 q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
313 sizeof(struct ip6_frag);
314 q6->ip6q_nxt = ip6f->ip6f_nxt;
318 * Check that the reassembled packet would not exceed 65535 bytes
320 * If it would exceed, discard the fragment and return an ICMP error.
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));
330 return (IPPROTO_DONE);
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));
337 return (IPPROTO_DONE);
340 * If it's the first fragment, do the above check for each
341 * fragment already stored in the reassembly queue.
344 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
346 af6dwn = af6->ip6af_down;
348 if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
350 struct mbuf *merr = IP6_REASS_MBUF(af6);
351 struct ip6_hdr *ip6err;
352 int erroff = af6->ip6af_offset;
354 /* dequeue the fragment. */
358 /* adjust pointer. */
359 ip6err = mtod(merr, struct ip6_hdr *);
362 * Restore source and destination addresses
363 * in the erroneous IPv6 header.
365 ip6err->ip6_src = q6->ip6q_src;
366 ip6err->ip6_dst = q6->ip6q_dst;
368 icmp6_error(merr, ICMP6_PARAM_PROB,
369 ICMP6_PARAMPROB_HEADER,
370 erroff - sizeof(struct ip6_frag) +
371 offsetof(struct ip6_frag, ip6f_offlg));
376 ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE,
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;
388 af6 = (struct ip6asfrag *)q6;
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.
397 ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
399 if (ecn == IPTOS_ECN_CE) {
400 if (ecn0 == IPTOS_ECN_NOTECT) {
401 free(ip6af, M_FTABLE);
404 if (ecn0 != IPTOS_ECN_CE)
405 q6->ip6q_ecn = IPTOS_ECN_CE;
407 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
408 free(ip6af, M_FTABLE);
413 * Find a segment which begins after this one does.
415 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
416 af6 = af6->ip6af_down)
417 if (af6->ip6af_off > ip6af->ip6af_off)
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.
426 if (af6->ip6af_up != (struct ip6asfrag *)q6) {
427 i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
430 if (i >= ip6af->ip6af_frglen)
432 m_adj(IP6_REASS_MBUF(ip6af), i);
433 ip6af->ip6af_off += i;
434 ip6af->ip6af_frglen -= i;
439 * While we overlap succeeding segments trim them or,
440 * if they are completely covered, dequeue them.
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;
448 m_adj(IP6_REASS_MBUF(af6), i);
451 af6 = af6->ip6af_down;
452 m_freem(IP6_REASS_MBUF(af6->ip6af_up));
453 frag6_deq(af6->ip6af_up);
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.
463 * Note: due to changes after disabling this part, mbuf passed to
464 * m_adj() below now does not meet the requirement.
466 if (af6->ip6af_up != (struct ip6asfrag *)q6) {
467 i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen
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));
475 free(ip6af, M_FTABLE);
479 if (af6 != (struct ip6asfrag *)q6) {
480 i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off;
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));
487 free(ip6af, M_FTABLE);
496 mac_ip6q_update(m, q6);
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.
505 frag6_enq(ip6af, af6->ip6af_up);
509 if (q6 != V_ip6q.ip6q_next) {
511 frag6_insque(q6, &V_ip6q);
515 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
516 af6 = af6->ip6af_down) {
517 if (af6->ip6af_off != next) {
521 next += af6->ip6af_frglen;
523 if (af6->ip6af_up->ip6af_mff) {
529 * Reassembly is complete; concatenate fragments.
531 ip6af = q6->ip6q_down;
532 t = m = IP6_REASS_MBUF(ip6af);
533 af6 = ip6af->ip6af_down;
535 while (af6 != (struct ip6asfrag *)q6) {
536 af6dwn = af6->ip6af_down;
540 t->m_next = IP6_REASS_MBUF(af6);
541 m_adj(t->m_next, af6->ip6af_offset);
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);
555 *q6->ip6q_nxtp = (u_char)(nxt & 0xff);
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),
563 m->m_data += sizeof(struct ip6_frag);
564 m->m_len -= sizeof(struct ip6_frag);
566 /* this comes with no copy if the boundary is on cluster */
567 if ((t = m_split(m, offset, M_NOWAIT)) == NULL) {
569 V_frag6_nfrags -= q6->ip6q_nfrag;
571 mac_ip6q_destroy(q6);
574 V_frag6_nfragpackets--;
577 m_adj(t, sizeof(struct ip6_frag));
582 * Store NXT to the original.
585 char *prvnxtp = ip6_get_prevhdr(m, offset); /* XXX */
590 V_frag6_nfrags -= q6->ip6q_nfrag;
592 mac_ip6q_reassemble(q6, m);
593 mac_ip6q_destroy(q6);
596 V_frag6_nfragpackets--;
598 if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
600 for (t = m; t; t = t->m_next)
602 m->m_pkthdr.len = plen;
605 IP6STAT_INC(ip6s_reassembled);
606 in6_ifstat_inc(dstifp, ifs6_reass_ok);
609 * Tell launch routine the next header
620 in6_ifstat_inc(dstifp, ifs6_reass_fail);
621 IP6STAT_INC(ip6s_fragdropped);
627 * Free a fragment reassembly header and all
628 * associated datagrams.
631 frag6_freef(struct ip6q *q6)
633 struct ip6asfrag *af6, *down6;
637 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
639 struct mbuf *m = IP6_REASS_MBUF(af6);
641 down6 = af6->ip6af_down;
645 * Return ICMP time exceeded error for the 1st fragment.
646 * Just free other fragments.
648 if (af6->ip6af_off == 0) {
652 ip6 = mtod(m, struct ip6_hdr *);
654 /* restore source and destination addresses */
655 ip6->ip6_src = q6->ip6q_src;
656 ip6->ip6_dst = q6->ip6q_dst;
658 icmp6_error(m, ICMP6_TIME_EXCEEDED,
659 ICMP6_TIME_EXCEED_REASSEMBLY, 0);
665 V_frag6_nfrags -= q6->ip6q_nfrag;
667 mac_ip6q_destroy(q6);
670 V_frag6_nfragpackets--;
674 * Put an ip fragment on a reassembly chain.
675 * Like insque, but pointers in middle of structure.
678 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6)
684 af6->ip6af_down = up6->ip6af_down;
685 up6->ip6af_down->ip6af_up = af6;
686 up6->ip6af_down = af6;
690 * To frag6_enq as remque is to insque.
693 frag6_deq(struct ip6asfrag *af6)
698 af6->ip6af_up->ip6af_down = af6->ip6af_down;
699 af6->ip6af_down->ip6af_up = af6->ip6af_up;
703 frag6_insque(struct ip6q *new, struct ip6q *old)
708 new->ip6q_prev = old;
709 new->ip6q_next = old->ip6q_next;
710 old->ip6q_next->ip6q_prev= new;
711 old->ip6q_next = new;
715 frag6_remque(struct ip6q *p6)
720 p6->ip6q_prev->ip6q_next = p6->ip6q_next;
721 p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
725 * IPv6 reassembling timer processing;
726 * if a timer expires on a reassembly
732 VNET_ITERATOR_DECL(vnet_iter);
735 VNET_LIST_RLOCK_NOSLEEP();
737 VNET_FOREACH(vnet_iter) {
738 CURVNET_SET(vnet_iter);
739 q6 = V_ip6q.ip6q_next;
741 while (q6 != &V_ip6q) {
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);
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.
755 while (V_frag6_nfragpackets > (u_int)V_ip6_maxfragpackets &&
757 IP6STAT_INC(ip6s_fragoverflow);
758 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
759 frag6_freef(V_ip6q.ip6q_prev);
764 VNET_LIST_RUNLOCK_NOSLEEP();
768 * Drain off all datagram fragments.
773 VNET_ITERATOR_DECL(vnet_iter);
775 VNET_LIST_RLOCK_NOSLEEP();
776 if (IP6Q_TRYLOCK() == 0) {
777 VNET_LIST_RUNLOCK_NOSLEEP();
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);
790 VNET_LIST_RUNLOCK_NOSLEEP();