2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the project nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * $KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/domain.h>
42 #include <sys/eventhandler.h>
44 #include <sys/kernel.h>
45 #include <sys/malloc.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/sysctl.h>
50 #include <sys/syslog.h>
53 #include <net/if_var.h>
54 #include <net/netisr.h>
55 #include <net/route.h>
58 #include <netinet/in.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip6.h>
61 #include <netinet6/ip6_var.h>
62 #include <netinet/icmp6.h>
63 #include <netinet/in_systm.h> /* For ECN definitions. */
64 #include <netinet/ip.h> /* For ECN definitions. */
67 #include <security/mac/mac_framework.h>
70 /* Reassembly headers are stored in hash buckets. */
71 #define IP6REASS_NHASH_LOG2 10
72 #define IP6REASS_NHASH (1 << IP6REASS_NHASH_LOG2)
73 #define IP6REASS_HMASK (IP6REASS_NHASH - 1)
75 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *,
76 uint32_t bucket __unused);
77 static void frag6_deq(struct ip6asfrag *, uint32_t bucket __unused);
78 static void frag6_insque_head(struct ip6q *, struct ip6q *,
80 static void frag6_remque(struct ip6q *, uint32_t bucket);
81 static void frag6_freef(struct ip6q *, uint32_t bucket);
89 static MALLOC_DEFINE(M_FRAG6, "frag6", "IPv6 fragment reassembly header");
91 /* System wide (global) maximum and count of packets in reassembly queues. */
92 static int ip6_maxfrags;
93 static volatile u_int frag6_nfrags = 0;
95 /* Maximum and current packets in per-VNET reassembly queue. */
96 VNET_DEFINE_STATIC(int, ip6_maxfragpackets);
97 VNET_DEFINE_STATIC(volatile u_int, frag6_nfragpackets);
98 #define V_ip6_maxfragpackets VNET(ip6_maxfragpackets)
99 #define V_frag6_nfragpackets VNET(frag6_nfragpackets)
101 /* Maximum per-VNET reassembly queues per bucket and fragments per packet. */
102 VNET_DEFINE_STATIC(int, ip6_maxfragbucketsize);
103 VNET_DEFINE_STATIC(int, ip6_maxfragsperpacket);
104 #define V_ip6_maxfragbucketsize VNET(ip6_maxfragbucketsize)
105 #define V_ip6_maxfragsperpacket VNET(ip6_maxfragsperpacket)
107 /* Per-VNET reassembly queue buckets. */
108 VNET_DEFINE_STATIC(struct ip6qbucket, ip6qb[IP6REASS_NHASH]);
109 VNET_DEFINE_STATIC(uint32_t, ip6qb_hashseed);
110 #define V_ip6qb VNET(ip6qb)
111 #define V_ip6qb_hashseed VNET(ip6qb_hashseed)
113 #define IP6QB_LOCK(_b) mtx_lock(&V_ip6qb[(_b)].lock)
114 #define IP6QB_TRYLOCK(_b) mtx_trylock(&V_ip6qb[(_b)].lock)
115 #define IP6QB_LOCK_ASSERT(_b) mtx_assert(&V_ip6qb[(_b)].lock, MA_OWNED)
116 #define IP6QB_UNLOCK(_b) mtx_unlock(&V_ip6qb[(_b)].lock)
117 #define IP6QB_HEAD(_b) (&V_ip6qb[(_b)].ip6q)
120 * By default, limit the number of IP6 fragments across all reassembly
121 * queues to 1/32 of the total number of mbuf clusters.
123 * Limit the total number of reassembly queues per VNET to the
124 * IP6 fragment limit, but ensure the limit will not allow any bucket
125 * to grow above 100 items. (The bucket limit is
126 * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct
127 * multiplier to reach a 100-item limit.)
128 * The 100-item limit was chosen as brief testing seems to show that
129 * this produces "reasonable" performance on some subset of systems
132 #define IP6_MAXFRAGS (nmbclusters / 32)
133 #define IP6_MAXFRAGPACKETS (imin(IP6_MAXFRAGS, IP6REASS_NHASH * 50))
137 * Sysctls and helper function.
139 SYSCTL_DECL(_net_inet6_ip6);
142 frag6_set_bucketsize(void)
146 if ((i = V_ip6_maxfragpackets) > 0)
147 V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1);
150 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGS, maxfrags,
151 CTLFLAG_RW, &ip6_maxfrags, 0,
152 "Maximum allowed number of outstanding IPv6 packet fragments. "
153 "A value of 0 means no fragmented packets will be accepted, while a "
154 "a value of -1 means no limit");
157 sysctl_ip6_maxfragpackets(SYSCTL_HANDLER_ARGS)
161 val = V_ip6_maxfragpackets;
162 error = sysctl_handle_int(oidp, &val, 0, req);
163 if (error != 0 || !req->newptr)
165 V_ip6_maxfragpackets = val;
166 frag6_set_bucketsize();
169 SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGPACKETS, maxfragpackets,
170 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, NULL, 0,
171 sysctl_ip6_maxfragpackets, "I",
172 "Default maximum number of outstanding fragmented IPv6 packets. "
173 "A value of 0 means no fragmented packets will be accepted, while a "
174 "a value of -1 means no limit");
175 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGSPERPACKET, maxfragsperpacket,
176 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragsperpacket), 0,
177 "Maximum allowed number of fragments per packet");
178 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGBUCKETSIZE, maxfragbucketsize,
179 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragbucketsize), 0,
180 "Maximum number of reassembly queues per hash bucket");
184 * Remove the IPv6 fragmentation header from the mbuf.
187 ip6_deletefraghdr(struct mbuf *m, int offset, int wait)
192 /* Delete frag6 header. */
193 if (m->m_len >= offset + sizeof(struct ip6_frag)) {
195 /* This is the only possible case with !PULLDOWN_TEST. */
196 ip6 = mtod(m, struct ip6_hdr *);
197 bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag),
199 m->m_data += sizeof(struct ip6_frag);
200 m->m_len -= sizeof(struct ip6_frag);
203 /* This comes with no copy if the boundary is on cluster. */
204 if ((t = m_split(m, offset, wait)) == NULL)
206 m_adj(t, sizeof(struct ip6_frag));
210 m->m_flags |= M_FRAGMENTED;
215 * Free a fragment reassembly header and all associated datagrams.
218 frag6_freef(struct ip6q *q6, uint32_t bucket)
221 struct ip6asfrag *af6, *down6;
224 IP6QB_LOCK_ASSERT(bucket);
226 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
229 m = IP6_REASS_MBUF(af6);
230 down6 = af6->ip6af_down;
231 frag6_deq(af6, bucket);
234 * Return ICMP time exceeded error for the 1st fragment.
235 * Just free other fragments.
237 if (af6->ip6af_off == 0) {
239 /* Adjust pointer. */
240 ip6 = mtod(m, struct ip6_hdr *);
242 /* Restore source and destination addresses. */
243 ip6->ip6_src = q6->ip6q_src;
244 ip6->ip6_dst = q6->ip6q_dst;
246 icmp6_error(m, ICMP6_TIME_EXCEEDED,
247 ICMP6_TIME_EXCEED_REASSEMBLY, 0);
253 frag6_remque(q6, bucket);
254 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
256 mac_ip6q_destroy(q6);
259 atomic_subtract_int(&V_frag6_nfragpackets, 1);
263 * Like in RFC2460, in RFC8200, fragment and reassembly rules do not agree with
264 * each other, in terms of next header field handling in fragment header.
265 * While the sender will use the same value for all of the fragmented packets,
266 * receiver is suggested not to check for consistency.
268 * Fragment rules (p18,p19):
269 * (2) A Fragment header containing:
270 * The Next Header value that identifies the first header
271 * after the Per-Fragment headers of the original packet.
272 * -> next header field is same for all fragments
274 * Reassembly rule (p20):
275 * The Next Header field of the last header of the Per-Fragment
276 * headers is obtained from the Next Header field of the first
277 * fragment's Fragment header.
278 * -> should grab it from the first fragment only
280 * The following note also contradicts with fragment rule - no one is going to
281 * send different fragment with different next header field.
283 * Additional note (p22) [not an error]:
284 * The Next Header values in the Fragment headers of different
285 * fragments of the same original packet may differ. Only the value
286 * from the Offset zero fragment packet is used for reassembly.
287 * -> should grab it from the first fragment only
289 * There is no explicit reason given in the RFC. Historical reason maybe?
295 frag6_input(struct mbuf **mp, int *offp, int proto)
297 struct ifnet *dstifp;
298 struct in6_ifaddr *ia6;
300 struct ip6_frag *ip6f;
301 struct ip6q *head, *q6;
302 struct ip6asfrag *af6, *af6dwn, *ip6af;
304 uint32_t hashkey[(sizeof(struct in6_addr) * 2 +
305 sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)];
306 uint32_t bucket, *hashkeyp;
307 int fragoff, frgpartlen; /* Must be larger than uint16_t. */
308 int nxt, offset, plen;
312 struct ip6_direct_ctx *ip6dc;
319 ip6 = mtod(m, struct ip6_hdr *);
320 #ifndef PULLDOWN_TEST
321 IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
322 ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
324 IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
326 return (IPPROTO_DONE);
330 /* Find the destination interface of the packet. */
331 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
333 dstifp = ia6->ia_ifp;
334 ifa_free(&ia6->ia_ifa);
337 /* Jumbo payload cannot contain a fragment header. */
338 if (ip6->ip6_plen == 0) {
339 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
340 in6_ifstat_inc(dstifp, ifs6_reass_fail);
341 return (IPPROTO_DONE);
345 * Check whether fragment packet's fragment length is a
346 * multiple of 8 octets (unless it is the last one).
347 * sizeof(struct ip6_frag) == 8
348 * sizeof(struct ip6_hdr) = 40
350 if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
351 (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
352 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
353 offsetof(struct ip6_hdr, ip6_plen));
354 in6_ifstat_inc(dstifp, ifs6_reass_fail);
355 return (IPPROTO_DONE);
358 IP6STAT_INC(ip6s_fragments);
359 in6_ifstat_inc(dstifp, ifs6_reass_reqd);
361 /* Offset now points to data portion. */
362 offset += sizeof(struct ip6_frag);
365 * Handle "atomic" fragments (offset and m bit set to 0) upfront,
366 * unrelated to any reassembly. Still need to remove the frag hdr.
367 * See RFC 6946 and section 4.5 of RFC 8200.
369 if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
370 IP6STAT_INC(ip6s_atomicfrags);
371 /* XXX-BZ handle correctly. */
372 in6_ifstat_inc(dstifp, ifs6_reass_ok);
374 m->m_flags |= M_FRAGMENTED;
375 return (ip6f->ip6f_nxt);
378 /* Get fragment length and discard 0-byte fragments. */
379 frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
380 if (frgpartlen == 0) {
381 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
382 offsetof(struct ip6_hdr, ip6_plen));
383 in6_ifstat_inc(dstifp, ifs6_reass_fail);
384 IP6STAT_INC(ip6s_fragdropped);
385 return (IPPROTO_DONE);
388 /* Generate a hash value for fragment bucket selection. */
390 memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr));
391 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
392 memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr));
393 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
394 *hashkeyp = ip6f->ip6f_ident;
395 bucket = jenkins_hash32(hashkey, nitems(hashkey), V_ip6qb_hashseed);
396 bucket &= IP6REASS_HMASK;
397 head = IP6QB_HEAD(bucket);
401 * Enforce upper bound on number of fragments for the entire system.
402 * If maxfrag is 0, never accept fragments.
403 * If maxfrag is -1, accept all fragments without limitation.
405 if (ip6_maxfrags < 0)
407 else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags)
410 for (q6 = head->ip6q_next; q6 != head; q6 = q6->ip6q_next)
411 if (ip6f->ip6f_ident == q6->ip6q_ident &&
412 IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
413 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
415 && mac_ip6q_match(m, q6)
423 /* A first fragment to arrive creates a reassembly queue. */
427 * Enforce upper bound on number of fragmented packets
428 * for which we attempt reassembly;
429 * If maxfragpackets is 0, never accept fragments.
430 * If maxfragpackets is -1, accept all fragments without
433 if (V_ip6_maxfragpackets < 0)
435 else if (V_ip6qb[bucket].count >= V_ip6_maxfragbucketsize ||
436 atomic_load_int(&V_frag6_nfragpackets) >=
437 (u_int)V_ip6_maxfragpackets)
439 atomic_add_int(&V_frag6_nfragpackets, 1);
441 /* Allocate IPv6 fragement packet queue entry. */
442 q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FRAG6,
447 if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
451 mac_ip6q_create(m, q6);
453 frag6_insque_head(q6, head, bucket);
455 /* ip6q_nxt will be filled afterwards, from 1st fragment. */
456 q6->ip6q_down = q6->ip6q_up = (struct ip6asfrag *)q6;
458 q6->ip6q_nxtp = (u_char *)nxtp;
460 q6->ip6q_ident = ip6f->ip6f_ident;
461 q6->ip6q_ttl = IPV6_FRAGTTL;
462 q6->ip6q_src = ip6->ip6_src;
463 q6->ip6q_dst = ip6->ip6_dst;
465 (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
466 q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */
472 * If it is the 1st fragment, record the length of the
473 * unfragmentable part and the next header of the fragment header.
475 fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
477 q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
478 sizeof(struct ip6_frag);
479 q6->ip6q_nxt = ip6f->ip6f_nxt;
483 * Check that the reassembled packet would not exceed 65535 bytes
485 * If it would exceed, discard the fragment and return an ICMP error.
487 if (q6->ip6q_unfrglen >= 0) {
488 /* The 1st fragment has already arrived. */
489 if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
490 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
491 offset - sizeof(struct ip6_frag) +
492 offsetof(struct ip6_frag, ip6f_offlg));
493 IP6QB_UNLOCK(bucket);
494 return (IPPROTO_DONE);
496 } else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
497 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
498 offset - sizeof(struct ip6_frag) +
499 offsetof(struct ip6_frag, ip6f_offlg));
500 IP6QB_UNLOCK(bucket);
501 return (IPPROTO_DONE);
504 * If it is the first fragment, do the above check for each
505 * fragment already stored in the reassembly queue.
508 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
510 af6dwn = af6->ip6af_down;
512 if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen >
514 struct ip6_hdr *ip6err;
518 merr = IP6_REASS_MBUF(af6);
519 erroff = af6->ip6af_offset;
521 /* Dequeue the fragment. */
522 frag6_deq(af6, bucket);
525 /* Adjust pointer. */
526 ip6err = mtod(merr, struct ip6_hdr *);
529 * Restore source and destination addresses
530 * in the erroneous IPv6 header.
532 ip6err->ip6_src = q6->ip6q_src;
533 ip6err->ip6_dst = q6->ip6q_dst;
535 icmp6_error(merr, ICMP6_PARAM_PROB,
536 ICMP6_PARAMPROB_HEADER,
537 erroff - sizeof(struct ip6_frag) +
538 offsetof(struct ip6_frag, ip6f_offlg));
543 /* Allocate an IPv6 fragement queue entry for this fragmented part. */
544 ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FRAG6,
548 ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG;
549 ip6af->ip6af_off = fragoff;
550 ip6af->ip6af_frglen = frgpartlen;
551 ip6af->ip6af_offset = offset;
552 IP6_REASS_MBUF(ip6af) = m;
555 af6 = (struct ip6asfrag *)q6;
559 /* Do duplicate, condition, and boundry checks. */
561 * Handle ECN by comparing this segment with the first one;
562 * if CE is set, do not lose CE.
563 * Drop if CE and not-ECT are mixed for the same packet.
565 ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
567 if (ecn == IPTOS_ECN_CE) {
568 if (ecn0 == IPTOS_ECN_NOTECT) {
569 free(ip6af, M_FRAG6);
572 if (ecn0 != IPTOS_ECN_CE)
573 q6->ip6q_ecn = IPTOS_ECN_CE;
575 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
576 free(ip6af, M_FRAG6);
580 /* Find a fragmented part which begins after this one does. */
581 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
582 af6 = af6->ip6af_down)
583 if (af6->ip6af_off > ip6af->ip6af_off)
587 * If the incoming framgent overlaps some existing fragments in
588 * the reassembly queue, drop both the new fragment and the
589 * entire reassembly queue. However, if the new fragment
590 * is an exact duplicate of an existing fragment, only silently
591 * drop the existing fragment and leave the fragmentation queue
592 * unchanged, as allowed by the RFC. (RFC 8200, 4.5)
594 if (af6->ip6af_up != (struct ip6asfrag *)q6) {
595 if (af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen -
596 ip6af->ip6af_off > 0) {
597 free(ip6af, M_FRAG6);
601 if (af6 != (struct ip6asfrag *)q6) {
602 if (ip6af->ip6af_off + ip6af->ip6af_frglen -
603 af6->ip6af_off > 0) {
604 free(ip6af, M_FRAG6);
612 mac_ip6q_update(m, q6);
616 * Stick new segment in its place; check for complete reassembly.
617 * If not complete, check fragment limit. Move to front of packet
618 * queue, as we are the most recently active fragmented packet.
620 frag6_enq(ip6af, af6->ip6af_up, bucket);
621 atomic_add_int(&frag6_nfrags, 1);
624 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6;
625 af6 = af6->ip6af_down) {
626 if (af6->ip6af_off != plen) {
627 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
628 IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
629 frag6_freef(q6, bucket);
631 IP6QB_UNLOCK(bucket);
632 return (IPPROTO_DONE);
634 plen += af6->ip6af_frglen;
636 if (af6->ip6af_up->ip6af_mff) {
637 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
638 IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
639 frag6_freef(q6, bucket);
641 IP6QB_UNLOCK(bucket);
642 return (IPPROTO_DONE);
645 /* Reassembly is complete; concatenate fragments. */
646 ip6af = q6->ip6q_down;
647 t = m = IP6_REASS_MBUF(ip6af);
648 af6 = ip6af->ip6af_down;
649 frag6_deq(ip6af, bucket);
650 while (af6 != (struct ip6asfrag *)q6) {
651 m->m_pkthdr.csum_flags &=
652 IP6_REASS_MBUF(af6)->m_pkthdr.csum_flags;
653 m->m_pkthdr.csum_data +=
654 IP6_REASS_MBUF(af6)->m_pkthdr.csum_data;
656 af6dwn = af6->ip6af_down;
657 frag6_deq(af6, bucket);
660 m_adj(IP6_REASS_MBUF(af6), af6->ip6af_offset);
661 m_demote_pkthdr(IP6_REASS_MBUF(af6));
662 m_cat(t, IP6_REASS_MBUF(af6));
667 while (m->m_pkthdr.csum_data & 0xffff0000)
668 m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
669 (m->m_pkthdr.csum_data >> 16);
671 /* Adjust offset to point where the original next header starts. */
672 offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
673 free(ip6af, M_FRAG6);
674 ip6 = mtod(m, struct ip6_hdr *);
675 ip6->ip6_plen = htons((u_short)plen + offset - sizeof(struct ip6_hdr));
676 if (q6->ip6q_ecn == IPTOS_ECN_CE)
677 ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
680 if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) {
681 frag6_remque(q6, bucket);
682 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
684 mac_ip6q_destroy(q6);
687 atomic_subtract_int(&V_frag6_nfragpackets, 1);
692 /* Set nxt(-hdr field value) to the original value. */
693 m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t),
696 frag6_remque(q6, bucket);
697 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
699 mac_ip6q_reassemble(q6, m);
700 mac_ip6q_destroy(q6);
703 atomic_subtract_int(&V_frag6_nfragpackets, 1);
705 if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
708 for (t = m; t; t = t->m_next)
710 m->m_pkthdr.len = plen;
714 mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc),
719 ip6dc = (struct ip6_direct_ctx *)(mtag + 1);
720 ip6dc->ip6dc_nxt = nxt;
721 ip6dc->ip6dc_off = offset;
723 m_tag_prepend(m, mtag);
726 IP6QB_UNLOCK(bucket);
727 IP6STAT_INC(ip6s_reassembled);
728 in6_ifstat_inc(dstifp, ifs6_reass_ok);
731 /* Queue/dispatch for reprocessing. */
732 netisr_dispatch(NETISR_IPV6_DIRECT, m);
733 return (IPPROTO_DONE);
736 /* Tell launch routine the next header. */
743 IP6QB_UNLOCK(bucket);
744 in6_ifstat_inc(dstifp, ifs6_reass_fail);
745 IP6STAT_INC(ip6s_fragdropped);
747 return (IPPROTO_DONE);
751 * IPv6 reassembling timer processing;
752 * if a timer expires on a reassembly queue, discard it.
757 VNET_ITERATOR_DECL(vnet_iter);
758 struct ip6q *head, *q6;
761 VNET_LIST_RLOCK_NOSLEEP();
762 VNET_FOREACH(vnet_iter) {
763 CURVNET_SET(vnet_iter);
764 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
766 head = IP6QB_HEAD(bucket);
767 q6 = head->ip6q_next;
770 * XXXJTL: This should never happen. This
771 * should turn into an assertion.
773 IP6QB_UNLOCK(bucket);
779 if (q6->ip6q_prev->ip6q_ttl == 0) {
780 IP6STAT_ADD(ip6s_fragtimeout,
781 q6->ip6q_prev->ip6q_nfrag);
782 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
783 frag6_freef(q6->ip6q_prev, bucket);
787 * If we are over the maximum number of fragments
788 * (due to the limit being lowered), drain off
789 * enough to get down to the new limit.
790 * Note that we drain all reassembly queues if
791 * maxfragpackets is 0 (fragmentation is disabled),
792 * and do not enforce a limit when maxfragpackets
795 while ((V_ip6_maxfragpackets == 0 ||
796 (V_ip6_maxfragpackets > 0 &&
797 V_ip6qb[bucket].count > V_ip6_maxfragbucketsize)) &&
798 head->ip6q_prev != head) {
799 IP6STAT_ADD(ip6s_fragoverflow,
800 q6->ip6q_prev->ip6q_nfrag);
801 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
802 frag6_freef(head->ip6q_prev, bucket);
804 IP6QB_UNLOCK(bucket);
807 * If we are still over the maximum number of fragmented
808 * packets, drain off enough to get down to the new limit.
811 while (V_ip6_maxfragpackets >= 0 &&
812 atomic_load_int(&V_frag6_nfragpackets) >
813 (u_int)V_ip6_maxfragpackets) {
815 head = IP6QB_HEAD(bucket);
816 if (head->ip6q_prev != head) {
817 IP6STAT_ADD(ip6s_fragoverflow,
818 q6->ip6q_prev->ip6q_nfrag);
819 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
820 frag6_freef(head->ip6q_prev, bucket);
822 IP6QB_UNLOCK(bucket);
823 bucket = (bucket + 1) % IP6REASS_NHASH;
827 VNET_LIST_RUNLOCK_NOSLEEP();
831 * Eventhandler to adjust limits in case nmbclusters change.
834 frag6_change(void *tag)
836 VNET_ITERATOR_DECL(vnet_iter);
838 ip6_maxfrags = IP6_MAXFRAGS;
839 VNET_LIST_RLOCK_NOSLEEP();
840 VNET_FOREACH(vnet_iter) {
841 CURVNET_SET(vnet_iter);
842 V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
843 frag6_set_bucketsize();
846 VNET_LIST_RUNLOCK_NOSLEEP();
850 * Initialise reassembly queue and fragment identifier.
858 V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
859 frag6_set_bucketsize();
860 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
861 q6 = IP6QB_HEAD(bucket);
862 q6->ip6q_next = q6->ip6q_prev = q6;
863 mtx_init(&V_ip6qb[bucket].lock, "ip6qlock", NULL, MTX_DEF);
864 V_ip6qb[bucket].count = 0;
866 V_ip6qb_hashseed = arc4random();
867 V_ip6_maxfragsperpacket = 64;
868 if (!IS_DEFAULT_VNET(curvnet))
871 ip6_maxfrags = IP6_MAXFRAGS;
872 EVENTHANDLER_REGISTER(nmbclusters_change,
873 frag6_change, NULL, EVENTHANDLER_PRI_ANY);
877 * Drain off all datagram fragments.
882 VNET_ITERATOR_DECL(vnet_iter);
886 VNET_LIST_RLOCK_NOSLEEP();
887 VNET_FOREACH(vnet_iter) {
888 CURVNET_SET(vnet_iter);
889 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
890 if (IP6QB_TRYLOCK(bucket) == 0)
892 head = IP6QB_HEAD(bucket);
893 while (head->ip6q_next != head) {
894 IP6STAT_INC(ip6s_fragdropped);
895 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
896 frag6_freef(head->ip6q_next, bucket);
898 IP6QB_UNLOCK(bucket);
902 VNET_LIST_RUNLOCK_NOSLEEP();
906 * Put an ip fragment on a reassembly chain.
907 * Like insque, but pointers in middle of structure.
910 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6,
911 uint32_t bucket __unused)
914 IP6QB_LOCK_ASSERT(bucket);
917 af6->ip6af_down = up6->ip6af_down;
918 up6->ip6af_down->ip6af_up = af6;
919 up6->ip6af_down = af6;
923 * To frag6_enq as remque is to insque.
926 frag6_deq(struct ip6asfrag *af6, uint32_t bucket __unused)
929 IP6QB_LOCK_ASSERT(bucket);
931 af6->ip6af_up->ip6af_down = af6->ip6af_down;
932 af6->ip6af_down->ip6af_up = af6->ip6af_up;
936 frag6_insque_head(struct ip6q *new, struct ip6q *old, uint32_t bucket)
939 IP6QB_LOCK_ASSERT(bucket);
940 KASSERT(IP6QB_HEAD(bucket) == old,
941 ("%s: attempt to insert at head of wrong bucket"
942 " (bucket=%u, old=%p)", __func__, bucket, old));
944 new->ip6q_prev = old;
945 new->ip6q_next = old->ip6q_next;
946 old->ip6q_next->ip6q_prev= new;
947 old->ip6q_next = new;
948 V_ip6qb[bucket].count++;
952 frag6_remque(struct ip6q *p6, uint32_t bucket)
955 IP6QB_LOCK_ASSERT(bucket);
957 p6->ip6q_prev->ip6q_next = p6->ip6q_next;
958 p6->ip6q_next->ip6q_prev = p6->ip6q_prev;
959 V_ip6qb[bucket].count--;