2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6 * Copyright (c) 2019 Netflix, Inc.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
32 * $KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/domain.h>
43 #include <sys/eventhandler.h>
45 #include <sys/kernel.h>
46 #include <sys/malloc.h>
48 #include <sys/kernel.h>
49 #include <sys/protosw.h>
50 #include <sys/queue.h>
51 #include <sys/socket.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
56 #include <net/if_var.h>
57 #include <net/netisr.h>
58 #include <net/route.h>
61 #include <netinet/in.h>
62 #include <netinet/in_var.h>
63 #include <netinet/ip6.h>
64 #include <netinet6/ip6_var.h>
65 #include <netinet/icmp6.h>
66 #include <netinet/in_systm.h> /* For ECN definitions. */
67 #include <netinet/ip.h> /* For ECN definitions. */
70 #include <security/mac/mac_framework.h>
74 * A "big picture" of how IPv6 fragment queues are all linked together.
76 * struct ip6qbucket ip6qb[...]; hashed buckets
79 * +--- TAILQ(struct ip6q, packets) *q6; tailq entries holding
80 * |||||||| fragmented packets
81 * | (1 per original packet)
83 * +--- TAILQ(struct ip6asfrag, ip6q_frags) *af6; tailq entries of IPv6
84 * | *ip6af;fragment packets
85 * | for one original packet
89 /* Reassembly headers are stored in hash buckets. */
90 #define IP6REASS_NHASH_LOG2 10
91 #define IP6REASS_NHASH (1 << IP6REASS_NHASH_LOG2)
92 #define IP6REASS_HMASK (IP6REASS_NHASH - 1)
94 TAILQ_HEAD(ip6qhead, ip6q);
96 struct ip6qhead packets;
102 TAILQ_ENTRY(ip6asfrag) ip6af_tq;
103 struct mbuf *ip6af_m;
104 int ip6af_offset; /* Offset in ip6af_m to next header. */
105 int ip6af_frglen; /* Fragmentable part length. */
106 int ip6af_off; /* Fragment offset. */
107 bool ip6af_mff; /* More fragment bit in frag off. */
110 static MALLOC_DEFINE(M_FRAG6, "frag6", "IPv6 fragment reassembly header");
113 /* A flag to indicate if IPv6 fragmentation is initialized. */
114 VNET_DEFINE_STATIC(bool, frag6_on);
115 #define V_frag6_on VNET(frag6_on)
118 /* System wide (global) maximum and count of packets in reassembly queues. */
119 static int ip6_maxfrags;
120 static volatile u_int frag6_nfrags = 0;
122 /* Maximum and current packets in per-VNET reassembly queue. */
123 VNET_DEFINE_STATIC(int, ip6_maxfragpackets);
124 VNET_DEFINE_STATIC(volatile u_int, frag6_nfragpackets);
125 #define V_ip6_maxfragpackets VNET(ip6_maxfragpackets)
126 #define V_frag6_nfragpackets VNET(frag6_nfragpackets)
128 /* Maximum per-VNET reassembly queues per bucket and fragments per packet. */
129 VNET_DEFINE_STATIC(int, ip6_maxfragbucketsize);
130 VNET_DEFINE_STATIC(int, ip6_maxfragsperpacket);
131 #define V_ip6_maxfragbucketsize VNET(ip6_maxfragbucketsize)
132 #define V_ip6_maxfragsperpacket VNET(ip6_maxfragsperpacket)
134 /* Per-VNET reassembly queue buckets. */
135 VNET_DEFINE_STATIC(struct ip6qbucket, ip6qb[IP6REASS_NHASH]);
136 VNET_DEFINE_STATIC(uint32_t, ip6qb_hashseed);
137 #define V_ip6qb VNET(ip6qb)
138 #define V_ip6qb_hashseed VNET(ip6qb_hashseed)
140 #define IP6QB_LOCK(_b) mtx_lock(&V_ip6qb[(_b)].lock)
141 #define IP6QB_TRYLOCK(_b) mtx_trylock(&V_ip6qb[(_b)].lock)
142 #define IP6QB_LOCK_ASSERT(_b) mtx_assert(&V_ip6qb[(_b)].lock, MA_OWNED)
143 #define IP6QB_UNLOCK(_b) mtx_unlock(&V_ip6qb[(_b)].lock)
144 #define IP6QB_HEAD(_b) (&V_ip6qb[(_b)].packets)
147 * By default, limit the number of IP6 fragments across all reassembly
148 * queues to 1/32 of the total number of mbuf clusters.
150 * Limit the total number of reassembly queues per VNET to the
151 * IP6 fragment limit, but ensure the limit will not allow any bucket
152 * to grow above 100 items. (The bucket limit is
153 * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct
154 * multiplier to reach a 100-item limit.)
155 * The 100-item limit was chosen as brief testing seems to show that
156 * this produces "reasonable" performance on some subset of systems
159 #define IP6_MAXFRAGS (nmbclusters / 32)
160 #define IP6_MAXFRAGPACKETS (imin(IP6_MAXFRAGS, IP6REASS_NHASH * 50))
164 * Sysctls and helper function.
166 SYSCTL_DECL(_net_inet6_ip6);
168 SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfrags,
169 CTLFLAG_RD, __DEVOLATILE(u_int *, &frag6_nfrags), 0,
170 "Global number of IPv6 fragments across all reassembly queues.");
173 frag6_set_bucketsize(void)
177 if ((i = V_ip6_maxfragpackets) > 0)
178 V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1);
181 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGS, maxfrags,
182 CTLFLAG_RW, &ip6_maxfrags, 0,
183 "Maximum allowed number of outstanding IPv6 packet fragments. "
184 "A value of 0 means no fragmented packets will be accepted, while a "
185 "a value of -1 means no limit");
188 sysctl_ip6_maxfragpackets(SYSCTL_HANDLER_ARGS)
192 val = V_ip6_maxfragpackets;
193 error = sysctl_handle_int(oidp, &val, 0, req);
194 if (error != 0 || !req->newptr)
196 V_ip6_maxfragpackets = val;
197 frag6_set_bucketsize();
200 SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGPACKETS, maxfragpackets,
201 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, NULL, 0,
202 sysctl_ip6_maxfragpackets, "I",
203 "Default maximum number of outstanding fragmented IPv6 packets. "
204 "A value of 0 means no fragmented packets will be accepted, while a "
205 "a value of -1 means no limit");
206 SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfragpackets,
207 CTLFLAG_VNET | CTLFLAG_RD,
208 __DEVOLATILE(u_int *, &VNET_NAME(frag6_nfragpackets)), 0,
209 "Per-VNET number of IPv6 fragments across all reassembly queues.");
210 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGSPERPACKET, maxfragsperpacket,
211 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragsperpacket), 0,
212 "Maximum allowed number of fragments per packet");
213 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGBUCKETSIZE, maxfragbucketsize,
214 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragbucketsize), 0,
215 "Maximum number of reassembly queues per hash bucket");
219 * Remove the IPv6 fragmentation header from the mbuf.
222 ip6_deletefraghdr(struct mbuf *m, int offset, int wait)
227 /* Delete frag6 header. */
228 if (m->m_len >= offset + sizeof(struct ip6_frag)) {
230 /* This is the only possible case with !PULLDOWN_TEST. */
231 ip6 = mtod(m, struct ip6_hdr *);
232 bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag),
234 m->m_data += sizeof(struct ip6_frag);
235 m->m_len -= sizeof(struct ip6_frag);
238 /* This comes with no copy if the boundary is on cluster. */
239 if ((t = m_split(m, offset, wait)) == NULL)
241 m_adj(t, sizeof(struct ip6_frag));
245 m->m_flags |= M_FRAGMENTED;
250 * Free a fragment reassembly header and all associated datagrams.
253 frag6_freef(struct ip6q *q6, uint32_t bucket)
256 struct ip6asfrag *af6;
259 IP6QB_LOCK_ASSERT(bucket);
261 while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) {
264 TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq);
267 * Return ICMP time exceeded error for the 1st fragment.
268 * Just free other fragments.
270 if (af6->ip6af_off == 0 && m->m_pkthdr.rcvif != NULL) {
272 /* Adjust pointer. */
273 ip6 = mtod(m, struct ip6_hdr *);
275 /* Restore source and destination addresses. */
276 ip6->ip6_src = q6->ip6q_src;
277 ip6->ip6_dst = q6->ip6q_dst;
279 icmp6_error(m, ICMP6_TIME_EXCEEDED,
280 ICMP6_TIME_EXCEED_REASSEMBLY, 0);
287 TAILQ_REMOVE(IP6QB_HEAD(bucket), q6, ip6q_tq);
288 V_ip6qb[bucket].count--;
289 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
291 mac_ip6q_destroy(q6);
294 atomic_subtract_int(&V_frag6_nfragpackets, 1);
298 * Drain off all datagram fragments belonging to
299 * the given network interface.
302 frag6_cleanup(void *arg __unused, struct ifnet *ifp)
304 struct ip6qhead *head;
306 struct ip6asfrag *af6;
309 KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__));
311 CURVNET_SET_QUIET(ifp->if_vnet);
314 * Skip processing if IPv6 reassembly is not initialised or
315 * torn down by frag6_destroy().
323 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
325 head = IP6QB_HEAD(bucket);
326 /* Scan fragment list. */
327 TAILQ_FOREACH(q6, head, ip6q_tq) {
328 TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) {
330 /* Clear no longer valid rcvif pointer. */
331 if (af6->ip6af_m->m_pkthdr.rcvif == ifp)
332 af6->ip6af_m->m_pkthdr.rcvif = NULL;
335 IP6QB_UNLOCK(bucket);
339 EVENTHANDLER_DEFINE(ifnet_departure_event, frag6_cleanup, NULL, 0);
342 * Like in RFC2460, in RFC8200, fragment and reassembly rules do not agree with
343 * each other, in terms of next header field handling in fragment header.
344 * While the sender will use the same value for all of the fragmented packets,
345 * receiver is suggested not to check for consistency.
347 * Fragment rules (p18,p19):
348 * (2) A Fragment header containing:
349 * The Next Header value that identifies the first header
350 * after the Per-Fragment headers of the original packet.
351 * -> next header field is same for all fragments
353 * Reassembly rule (p20):
354 * The Next Header field of the last header of the Per-Fragment
355 * headers is obtained from the Next Header field of the first
356 * fragment's Fragment header.
357 * -> should grab it from the first fragment only
359 * The following note also contradicts with fragment rule - no one is going to
360 * send different fragment with different next header field.
362 * Additional note (p22) [not an error]:
363 * The Next Header values in the Fragment headers of different
364 * fragments of the same original packet may differ. Only the value
365 * from the Offset zero fragment packet is used for reassembly.
366 * -> should grab it from the first fragment only
368 * There is no explicit reason given in the RFC. Historical reason maybe?
374 frag6_input(struct mbuf **mp, int *offp, int proto)
378 struct ip6_frag *ip6f;
379 struct ip6qhead *head;
381 struct ip6asfrag *af6, *ip6af, *af6tmp;
382 struct in6_ifaddr *ia6;
383 struct ifnet *dstifp, *srcifp;
384 uint32_t hashkey[(sizeof(struct in6_addr) * 2 +
385 sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)];
386 uint32_t bucket, *hashkeyp;
387 int fragoff, frgpartlen; /* Must be larger than uint16_t. */
388 int nxt, offset, plen;
392 struct ip6_direct_ctx *ip6dc;
399 ip6 = mtod(m, struct ip6_hdr *);
400 #ifndef PULLDOWN_TEST
401 IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE);
402 ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
404 IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f));
406 return (IPPROTO_DONE);
410 /* Find the destination interface of the packet. */
411 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
413 dstifp = ia6->ia_ifp;
414 ifa_free(&ia6->ia_ifa);
417 /* Jumbo payload cannot contain a fragment header. */
418 if (ip6->ip6_plen == 0) {
419 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
420 in6_ifstat_inc(dstifp, ifs6_reass_fail);
421 return (IPPROTO_DONE);
425 * Check whether fragment packet's fragment length is a
426 * multiple of 8 octets (unless it is the last one).
427 * sizeof(struct ip6_frag) == 8
428 * sizeof(struct ip6_hdr) = 40
430 if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
431 (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
432 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
433 offsetof(struct ip6_hdr, ip6_plen));
434 in6_ifstat_inc(dstifp, ifs6_reass_fail);
435 return (IPPROTO_DONE);
438 IP6STAT_INC(ip6s_fragments);
439 in6_ifstat_inc(dstifp, ifs6_reass_reqd);
441 /* Offset now points to data portion. */
442 offset += sizeof(struct ip6_frag);
445 * Handle "atomic" fragments (offset and m bit set to 0) upfront,
446 * unrelated to any reassembly. Still need to remove the frag hdr.
447 * See RFC 6946 and section 4.5 of RFC 8200.
449 if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
450 /* XXX-BZ we want dedicated counters for this. */
451 IP6STAT_INC(ip6s_reassembled);
452 /* XXX-BZ handle correctly. */
453 in6_ifstat_inc(dstifp, ifs6_reass_ok);
455 m->m_flags |= M_FRAGMENTED;
456 return (ip6f->ip6f_nxt);
459 /* Get fragment length and discard 0-byte fragments. */
460 frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
461 if (frgpartlen == 0) {
462 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
463 offsetof(struct ip6_hdr, ip6_plen));
464 in6_ifstat_inc(dstifp, ifs6_reass_fail);
465 IP6STAT_INC(ip6s_fragdropped);
466 return (IPPROTO_DONE);
470 * Enforce upper bound on number of fragments for the entire system.
471 * If maxfrag is 0, never accept fragments.
472 * If maxfrag is -1, accept all fragments without limitation.
474 if (ip6_maxfrags < 0)
476 else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags)
480 * Validate that a full header chain to the ULP is present in the
481 * packet containing the first fragment as per RFC RFC7112 and
482 * RFC 8200 pages 18,19:
483 * The first fragment packet is composed of:
484 * (3) Extension headers, if any, and the Upper-Layer header. These
485 * headers must be in the first fragment. ...
487 fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
488 /* XXX TODO. thj has D16851 open for this. */
489 /* Send ICMPv6 4,3 in case of violation. */
491 /* Store receive network interface pointer for later. */
492 srcifp = m->m_pkthdr.rcvif;
494 /* Generate a hash value for fragment bucket selection. */
496 memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr));
497 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
498 memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr));
499 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
500 *hashkeyp = ip6f->ip6f_ident;
501 bucket = jenkins_hash32(hashkey, nitems(hashkey), V_ip6qb_hashseed);
502 bucket &= IP6REASS_HMASK;
504 head = IP6QB_HEAD(bucket);
506 TAILQ_FOREACH(q6, head, ip6q_tq)
507 if (ip6f->ip6f_ident == q6->ip6q_ident &&
508 IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
509 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
511 && mac_ip6q_match(m, q6)
519 /* A first fragment to arrive creates a reassembly queue. */
523 * Enforce upper bound on number of fragmented packets
524 * for which we attempt reassembly;
525 * If maxfragpackets is 0, never accept fragments.
526 * If maxfragpackets is -1, accept all fragments without
529 if (V_ip6_maxfragpackets < 0)
531 else if (V_ip6qb[bucket].count >= V_ip6_maxfragbucketsize ||
532 atomic_load_int(&V_frag6_nfragpackets) >=
533 (u_int)V_ip6_maxfragpackets)
536 /* Allocate IPv6 fragement packet queue entry. */
537 q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FRAG6,
542 if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
546 mac_ip6q_create(m, q6);
548 atomic_add_int(&V_frag6_nfragpackets, 1);
550 /* ip6q_nxt will be filled afterwards, from 1st fragment. */
551 TAILQ_INIT(&q6->ip6q_frags);
552 q6->ip6q_ident = ip6f->ip6f_ident;
553 q6->ip6q_ttl = IPV6_FRAGTTL;
554 q6->ip6q_src = ip6->ip6_src;
555 q6->ip6q_dst = ip6->ip6_dst;
557 (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
558 q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */
560 /* Add the fragemented packet to the bucket. */
561 TAILQ_INSERT_HEAD(head, q6, ip6q_tq);
562 V_ip6qb[bucket].count++;
566 * If it is the 1st fragment, record the length of the
567 * unfragmentable part and the next header of the fragment header.
568 * Assume the first 1st fragement to arrive will be correct.
569 * We do not have any duplicate checks here yet so another packet
570 * with fragoff == 0 could come and overwrite the ip6q_unfrglen
571 * and worse, the next header, at any time.
573 if (fragoff == 0 && q6->ip6q_unfrglen == -1) {
574 q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
575 sizeof(struct ip6_frag);
576 q6->ip6q_nxt = ip6f->ip6f_nxt;
581 * Check that the reassembled packet would not exceed 65535 bytes
583 * If it would exceed, discard the fragment and return an ICMP error.
585 if (q6->ip6q_unfrglen >= 0) {
586 /* The 1st fragment has already arrived. */
587 if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
589 TAILQ_REMOVE(head, q6, ip6q_tq);
590 V_ip6qb[bucket].count--;
591 atomic_subtract_int(&V_frag6_nfragpackets, 1);
593 mac_ip6q_destroy(q6);
597 IP6QB_UNLOCK(bucket);
598 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
599 offset - sizeof(struct ip6_frag) +
600 offsetof(struct ip6_frag, ip6f_offlg));
601 return (IPPROTO_DONE);
603 } else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
605 TAILQ_REMOVE(head, q6, ip6q_tq);
606 V_ip6qb[bucket].count--;
607 atomic_subtract_int(&V_frag6_nfragpackets, 1);
609 mac_ip6q_destroy(q6);
613 IP6QB_UNLOCK(bucket);
614 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
615 offset - sizeof(struct ip6_frag) +
616 offsetof(struct ip6_frag, ip6f_offlg));
617 return (IPPROTO_DONE);
621 * If it is the first fragment, do the above check for each
622 * fragment already stored in the reassembly queue.
624 if (fragoff == 0 && !only_frag) {
625 TAILQ_FOREACH_SAFE(af6, &q6->ip6q_frags, ip6af_tq, af6tmp) {
627 if (q6->ip6q_unfrglen + af6->ip6af_off +
628 af6->ip6af_frglen > IPV6_MAXPACKET) {
629 struct ip6_hdr *ip6err;
634 erroff = af6->ip6af_offset;
636 /* Dequeue the fragment. */
637 TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq);
639 atomic_subtract_int(&frag6_nfrags, 1);
642 /* Set a valid receive interface pointer. */
643 merr->m_pkthdr.rcvif = srcifp;
645 /* Adjust pointer. */
646 ip6err = mtod(merr, struct ip6_hdr *);
649 * Restore source and destination addresses
650 * in the erroneous IPv6 header.
652 ip6err->ip6_src = q6->ip6q_src;
653 ip6err->ip6_dst = q6->ip6q_dst;
655 icmp6_error(merr, ICMP6_PARAM_PROB,
656 ICMP6_PARAMPROB_HEADER,
657 erroff - sizeof(struct ip6_frag) +
658 offsetof(struct ip6_frag, ip6f_offlg));
663 /* Allocate an IPv6 fragement queue entry for this fragmented part. */
664 ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FRAG6,
668 ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) ? true : false;
669 ip6af->ip6af_off = fragoff;
670 ip6af->ip6af_frglen = frgpartlen;
671 ip6af->ip6af_offset = offset;
676 * Do a manual insert rather than a hard-to-understand cast
677 * to a different type relying on data structure order to work.
679 TAILQ_INSERT_HEAD(&q6->ip6q_frags, ip6af, ip6af_tq);
683 /* Do duplicate, condition, and boundry checks. */
685 * Handle ECN by comparing this segment with the first one;
686 * if CE is set, do not lose CE.
687 * Drop if CE and not-ECT are mixed for the same packet.
689 ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK;
691 if (ecn == IPTOS_ECN_CE) {
692 if (ecn0 == IPTOS_ECN_NOTECT) {
693 free(ip6af, M_FRAG6);
696 if (ecn0 != IPTOS_ECN_CE)
697 q6->ip6q_ecn = IPTOS_ECN_CE;
699 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
700 free(ip6af, M_FRAG6);
704 /* Find a fragmented part which begins after this one does. */
705 TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq)
706 if (af6->ip6af_off > ip6af->ip6af_off)
710 * If the incoming framgent overlaps some existing fragments in
711 * the reassembly queue, drop both the new fragment and the
712 * entire reassembly queue. However, if the new fragment
713 * is an exact duplicate of an existing fragment, only silently
714 * drop the existing fragment and leave the fragmentation queue
715 * unchanged, as allowed by the RFC. (RFC 8200, 4.5)
718 af6tmp = TAILQ_PREV(af6, ip6fraghead, ip6af_tq);
720 af6tmp = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead);
721 if (af6tmp != NULL) {
722 if (af6tmp->ip6af_off + af6tmp->ip6af_frglen -
723 ip6af->ip6af_off > 0) {
724 if (af6tmp->ip6af_off != ip6af->ip6af_off ||
725 af6tmp->ip6af_frglen != ip6af->ip6af_frglen)
726 frag6_freef(q6, bucket);
727 free(ip6af, M_FRAG6);
732 if (ip6af->ip6af_off + ip6af->ip6af_frglen -
733 af6->ip6af_off > 0) {
734 if (af6->ip6af_off != ip6af->ip6af_off ||
735 af6->ip6af_frglen != ip6af->ip6af_frglen)
736 frag6_freef(q6, bucket);
737 free(ip6af, M_FRAG6);
743 mac_ip6q_update(m, q6);
747 * Stick new segment in its place; check for complete reassembly.
748 * If not complete, check fragment limit. Move to front of packet
749 * queue, as we are the most recently active fragmented packet.
752 TAILQ_INSERT_BEFORE(af6, ip6af, ip6af_tq);
754 TAILQ_INSERT_TAIL(&q6->ip6q_frags, ip6af, ip6af_tq);
756 atomic_add_int(&frag6_nfrags, 1);
760 TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) {
761 if (af6->ip6af_off != plen) {
762 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
763 IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
764 frag6_freef(q6, bucket);
766 IP6QB_UNLOCK(bucket);
767 return (IPPROTO_DONE);
769 plen += af6->ip6af_frglen;
771 af6 = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead);
772 if (af6->ip6af_mff) {
773 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
774 IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
775 frag6_freef(q6, bucket);
777 IP6QB_UNLOCK(bucket);
778 return (IPPROTO_DONE);
781 /* Reassembly is complete; concatenate fragments. */
782 ip6af = TAILQ_FIRST(&q6->ip6q_frags);
783 t = m = ip6af->ip6af_m;
784 TAILQ_REMOVE(&q6->ip6q_frags, ip6af, ip6af_tq);
785 while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) {
786 m->m_pkthdr.csum_flags &=
787 af6->ip6af_m->m_pkthdr.csum_flags;
788 m->m_pkthdr.csum_data +=
789 af6->ip6af_m->m_pkthdr.csum_data;
791 TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq);
793 m_adj(af6->ip6af_m, af6->ip6af_offset);
794 m_demote_pkthdr(af6->ip6af_m);
795 m_cat(t, af6->ip6af_m);
799 while (m->m_pkthdr.csum_data & 0xffff0000)
800 m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
801 (m->m_pkthdr.csum_data >> 16);
803 /* Adjust offset to point where the original next header starts. */
804 offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
805 free(ip6af, M_FRAG6);
806 ip6 = mtod(m, struct ip6_hdr *);
807 ip6->ip6_plen = htons((u_short)plen + offset - sizeof(struct ip6_hdr));
808 if (q6->ip6q_ecn == IPTOS_ECN_CE)
809 ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
812 TAILQ_REMOVE(head, q6, ip6q_tq);
813 V_ip6qb[bucket].count--;
814 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
816 if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) {
818 mac_ip6q_destroy(q6);
821 atomic_subtract_int(&V_frag6_nfragpackets, 1);
826 /* Set nxt(-hdr field value) to the original value. */
827 m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t),
831 mac_ip6q_reassemble(q6, m);
832 mac_ip6q_destroy(q6);
835 atomic_subtract_int(&V_frag6_nfragpackets, 1);
837 if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
840 for (t = m; t; t = t->m_next)
842 m->m_pkthdr.len = plen;
843 /* Set a valid receive interface pointer. */
844 m->m_pkthdr.rcvif = srcifp;
848 mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc),
853 ip6dc = (struct ip6_direct_ctx *)(mtag + 1);
854 ip6dc->ip6dc_nxt = nxt;
855 ip6dc->ip6dc_off = offset;
857 m_tag_prepend(m, mtag);
860 IP6QB_UNLOCK(bucket);
861 IP6STAT_INC(ip6s_reassembled);
862 in6_ifstat_inc(dstifp, ifs6_reass_ok);
865 /* Queue/dispatch for reprocessing. */
866 netisr_dispatch(NETISR_IPV6_DIRECT, m);
867 return (IPPROTO_DONE);
870 /* Tell launch routine the next header. */
877 IP6QB_UNLOCK(bucket);
879 in6_ifstat_inc(dstifp, ifs6_reass_fail);
880 IP6STAT_INC(ip6s_fragdropped);
882 return (IPPROTO_DONE);
886 * IPv6 reassembling timer processing;
887 * if a timer expires on a reassembly queue, discard it.
892 VNET_ITERATOR_DECL(vnet_iter);
893 struct ip6qhead *head;
894 struct ip6q *q6, *q6tmp;
897 VNET_LIST_RLOCK_NOSLEEP();
898 VNET_FOREACH(vnet_iter) {
899 CURVNET_SET(vnet_iter);
900 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
902 head = IP6QB_HEAD(bucket);
903 TAILQ_FOREACH_SAFE(q6, head, ip6q_tq, q6tmp)
904 if (--q6->ip6q_ttl == 0) {
905 IP6STAT_ADD(ip6s_fragtimeout,
907 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
908 frag6_freef(q6, bucket);
911 * If we are over the maximum number of fragments
912 * (due to the limit being lowered), drain off
913 * enough to get down to the new limit.
914 * Note that we drain all reassembly queues if
915 * maxfragpackets is 0 (fragmentation is disabled),
916 * and do not enforce a limit when maxfragpackets
919 while ((V_ip6_maxfragpackets == 0 ||
920 (V_ip6_maxfragpackets > 0 &&
921 V_ip6qb[bucket].count > V_ip6_maxfragbucketsize)) &&
922 (q6 = TAILQ_LAST(head, ip6qhead)) != NULL) {
923 IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag);
924 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
925 frag6_freef(q6, bucket);
927 IP6QB_UNLOCK(bucket);
930 * If we are still over the maximum number of fragmented
931 * packets, drain off enough to get down to the new limit.
934 while (V_ip6_maxfragpackets >= 0 &&
935 atomic_load_int(&V_frag6_nfragpackets) >
936 (u_int)V_ip6_maxfragpackets) {
938 q6 = TAILQ_LAST(IP6QB_HEAD(bucket), ip6qhead);
940 IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag);
941 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
942 frag6_freef(q6, bucket);
944 IP6QB_UNLOCK(bucket);
945 bucket = (bucket + 1) % IP6REASS_NHASH;
949 VNET_LIST_RUNLOCK_NOSLEEP();
953 * Eventhandler to adjust limits in case nmbclusters change.
956 frag6_change(void *tag)
958 VNET_ITERATOR_DECL(vnet_iter);
960 ip6_maxfrags = IP6_MAXFRAGS;
961 VNET_LIST_RLOCK_NOSLEEP();
962 VNET_FOREACH(vnet_iter) {
963 CURVNET_SET(vnet_iter);
964 V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
965 frag6_set_bucketsize();
968 VNET_LIST_RUNLOCK_NOSLEEP();
972 * Initialise reassembly queue and fragment identifier.
979 V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
980 frag6_set_bucketsize();
981 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
982 TAILQ_INIT(IP6QB_HEAD(bucket));
983 mtx_init(&V_ip6qb[bucket].lock, "ip6qb", NULL, MTX_DEF);
984 V_ip6qb[bucket].count = 0;
986 V_ip6qb_hashseed = arc4random();
987 V_ip6_maxfragsperpacket = 64;
991 if (!IS_DEFAULT_VNET(curvnet))
994 ip6_maxfrags = IP6_MAXFRAGS;
995 EVENTHANDLER_REGISTER(nmbclusters_change,
996 frag6_change, NULL, EVENTHANDLER_PRI_ANY);
1000 * Drain off all datagram fragments.
1003 frag6_drain_one(void)
1008 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
1010 while ((q6 = TAILQ_FIRST(IP6QB_HEAD(bucket))) != NULL) {
1011 IP6STAT_INC(ip6s_fragdropped);
1012 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
1013 frag6_freef(q6, bucket);
1015 IP6QB_UNLOCK(bucket);
1022 VNET_ITERATOR_DECL(vnet_iter);
1024 VNET_LIST_RLOCK_NOSLEEP();
1025 VNET_FOREACH(vnet_iter) {
1026 CURVNET_SET(vnet_iter);
1030 VNET_LIST_RUNLOCK_NOSLEEP();
1035 * Clear up IPv6 reassembly structures.
1044 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
1045 KASSERT(V_ip6qb[bucket].count == 0,
1046 ("%s: V_ip6qb[%d] (%p) count not 0 (%d)", __func__,
1047 bucket, &V_ip6qb[bucket], V_ip6qb[bucket].count));
1048 mtx_destroy(&V_ip6qb[bucket].lock);