1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5 * All rights reserved.
6 * Copyright (c) 2019 Netflix, Inc.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
19 *
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
30 * SUCH DAMAGE.
31 *
32 * $KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $
33 */
34
35 #include <sys/cdefs.h>
36 #include "opt_rss.h"
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/domain.h>
41 #include <sys/eventhandler.h>
42 #include <sys/hash.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/protosw.h>
47 #include <sys/queue.h>
48 #include <sys/socket.h>
49 #include <sys/sysctl.h>
50 #include <sys/syslog.h>
51
52 #include <net/if.h>
53 #include <net/if_var.h>
54 #include <net/if_private.h>
55 #include <net/netisr.h>
56 #include <net/route.h>
57 #include <net/vnet.h>
58
59 #include <netinet/in.h>
60 #include <netinet/in_var.h>
61 #include <netinet/ip6.h>
62 #include <netinet6/ip6_var.h>
63 #include <netinet/icmp6.h>
64 #include <netinet/in_systm.h> /* For ECN definitions. */
65 #include <netinet/ip.h> /* For ECN definitions. */
66
67 #ifdef MAC
68 #include <security/mac/mac_framework.h>
69 #endif
70
71 /*
72 * A "big picture" of how IPv6 fragment queues are all linked together.
73 *
74 * struct ip6qbucket ip6qb[...]; hashed buckets
75 * ||||||||
76 * |
77 * +--- TAILQ(struct ip6q, packets) *q6; tailq entries holding
78 * |||||||| fragmented packets
79 * | (1 per original packet)
80 * |
81 * +--- TAILQ(struct ip6asfrag, ip6q_frags) *af6; tailq entries of IPv6
82 * | *ip6af;fragment packets
83 * | for one original packet
84 * + *mbuf
85 */
86
87 /* Reassembly headers are stored in hash buckets. */
88 #define IP6REASS_NHASH_LOG2 10
89 #define IP6REASS_NHASH (1 << IP6REASS_NHASH_LOG2)
90 #define IP6REASS_HMASK (IP6REASS_NHASH - 1)
91
92 TAILQ_HEAD(ip6qhead, ip6q);
93 struct ip6qbucket {
94 struct ip6qhead packets;
95 struct mtx lock;
96 int count;
97 };
98
99 struct ip6asfrag {
100 TAILQ_ENTRY(ip6asfrag) ip6af_tq;
101 struct mbuf *ip6af_m;
102 int ip6af_offset; /* Offset in ip6af_m to next header. */
103 int ip6af_frglen; /* Fragmentable part length. */
104 int ip6af_off; /* Fragment offset. */
105 bool ip6af_mff; /* More fragment bit in frag off. */
106 };
107
108 static MALLOC_DEFINE(M_FRAG6, "frag6", "IPv6 fragment reassembly header");
109
110 #ifdef VIMAGE
111 /* A flag to indicate if IPv6 fragmentation is initialized. */
112 VNET_DEFINE_STATIC(bool, frag6_on);
113 #define V_frag6_on VNET(frag6_on)
114 #endif
115
116 /* System wide (global) maximum and count of packets in reassembly queues. */
117 static int ip6_maxfrags;
118 static u_int __exclusive_cache_line frag6_nfrags;
119
120 /* Maximum and current packets in per-VNET reassembly queue. */
121 VNET_DEFINE_STATIC(int, ip6_maxfragpackets);
122 VNET_DEFINE_STATIC(volatile u_int, frag6_nfragpackets);
123 #define V_ip6_maxfragpackets VNET(ip6_maxfragpackets)
124 #define V_frag6_nfragpackets VNET(frag6_nfragpackets)
125
126 /* Maximum per-VNET reassembly timeout (milliseconds) */
127 VNET_DEFINE_STATIC(u_int, ip6_fraglifetime) = IPV6_DEFFRAGTTL;
128 #define V_ip6_fraglifetime VNET(ip6_fraglifetime)
129
130 /* Maximum per-VNET reassembly queues per bucket and fragments per packet. */
131 VNET_DEFINE_STATIC(int, ip6_maxfragbucketsize);
132 VNET_DEFINE_STATIC(int, ip6_maxfragsperpacket);
133 #define V_ip6_maxfragbucketsize VNET(ip6_maxfragbucketsize)
134 #define V_ip6_maxfragsperpacket VNET(ip6_maxfragsperpacket)
135
136 /* Per-VNET reassembly queue buckets. */
137 VNET_DEFINE_STATIC(struct ip6qbucket, ip6qb[IP6REASS_NHASH]);
138 VNET_DEFINE_STATIC(uint32_t, ip6qb_hashseed);
139 #define V_ip6qb VNET(ip6qb)
140 #define V_ip6qb_hashseed VNET(ip6qb_hashseed)
141
142 #define IP6QB_LOCK(_b) mtx_lock(&V_ip6qb[(_b)].lock)
143 #define IP6QB_TRYLOCK(_b) mtx_trylock(&V_ip6qb[(_b)].lock)
144 #define IP6QB_LOCK_ASSERT(_b) mtx_assert(&V_ip6qb[(_b)].lock, MA_OWNED)
145 #define IP6QB_UNLOCK(_b) mtx_unlock(&V_ip6qb[(_b)].lock)
146 #define IP6QB_HEAD(_b) (&V_ip6qb[(_b)].packets)
147
148 /*
149 * By default, limit the number of IP6 fragments across all reassembly
150 * queues to 1/32 of the total number of mbuf clusters.
151 *
152 * Limit the total number of reassembly queues per VNET to the
153 * IP6 fragment limit, but ensure the limit will not allow any bucket
154 * to grow above 100 items. (The bucket limit is
155 * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct
156 * multiplier to reach a 100-item limit.)
157 * The 100-item limit was chosen as brief testing seems to show that
158 * this produces "reasonable" performance on some subset of systems
159 * under DoS attack.
160 */
161 #define IP6_MAXFRAGS (nmbclusters / 32)
162 #define IP6_MAXFRAGPACKETS (imin(IP6_MAXFRAGS, IP6REASS_NHASH * 50))
163
164 /* Interval between periodic reassembly queue inspections */
165 #define IP6_CALLOUT_INTERVAL_MS 500
166
167 /*
168 * Sysctls and helper function.
169 */
170 SYSCTL_DECL(_net_inet6_ip6);
171
172 SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfrags,
173 CTLFLAG_RD, &frag6_nfrags, 0,
174 "Global number of IPv6 fragments across all reassembly queues.");
175
176 static void
frag6_set_bucketsize(void)177 frag6_set_bucketsize(void)
178 {
179 int i;
180
181 if ((i = V_ip6_maxfragpackets) > 0)
182 V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1);
183 }
184
185 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGS, maxfrags,
186 CTLFLAG_RW, &ip6_maxfrags, 0,
187 "Maximum allowed number of outstanding IPv6 packet fragments. "
188 "A value of 0 means no fragmented packets will be accepted, while "
189 "a value of -1 means no limit");
190
191 static int
sysctl_ip6_maxfragpackets(SYSCTL_HANDLER_ARGS)192 sysctl_ip6_maxfragpackets(SYSCTL_HANDLER_ARGS)
193 {
194 int error, val;
195
196 val = V_ip6_maxfragpackets;
197 error = sysctl_handle_int(oidp, &val, 0, req);
198 if (error != 0 || !req->newptr)
199 return (error);
200 V_ip6_maxfragpackets = val;
201 frag6_set_bucketsize();
202 return (0);
203 }
204 SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGPACKETS, maxfragpackets,
205 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
206 NULL, 0, sysctl_ip6_maxfragpackets, "I",
207 "Default maximum number of outstanding fragmented IPv6 packets. "
208 "A value of 0 means no fragmented packets will be accepted, while a "
209 "a value of -1 means no limit");
210 SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfragpackets,
211 CTLFLAG_VNET | CTLFLAG_RD,
212 __DEVOLATILE(u_int *, &VNET_NAME(frag6_nfragpackets)), 0,
213 "Per-VNET number of IPv6 fragments across all reassembly queues.");
214 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGSPERPACKET, maxfragsperpacket,
215 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragsperpacket), 0,
216 "Maximum allowed number of fragments per packet");
217 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGBUCKETSIZE, maxfragbucketsize,
218 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragbucketsize), 0,
219 "Maximum number of reassembly queues per hash bucket");
220
221 static int
frag6_milli_to_callout_ticks(int ms)222 frag6_milli_to_callout_ticks(int ms)
223 {
224 return (ms / IP6_CALLOUT_INTERVAL_MS);
225 }
226
227 static int
frag6_callout_ticks_to_milli(int ms)228 frag6_callout_ticks_to_milli(int ms)
229 {
230 return (ms * IP6_CALLOUT_INTERVAL_MS);
231 }
232
233 _Static_assert(sizeof(((struct ip6q *)NULL)->ip6q_ttl) >= 2,
234 "ip6q_ttl field is not large enough");
235
236 static int
sysctl_ip6_fraglifetime(SYSCTL_HANDLER_ARGS)237 sysctl_ip6_fraglifetime(SYSCTL_HANDLER_ARGS)
238 {
239 int error, val;
240
241 val = V_ip6_fraglifetime;
242 error = sysctl_handle_int(oidp, &val, 0, req);
243 if (error != 0 || !req->newptr)
244 return (error);
245 if (val <= 0)
246 val = IPV6_DEFFRAGTTL;
247
248 if (frag6_milli_to_callout_ticks(val) >= 65536)
249 val = frag6_callout_ticks_to_milli(65535);
250 #ifdef VIMAGE
251 if (!IS_DEFAULT_VNET(curvnet)) {
252 CURVNET_SET(vnet0);
253 int host_val = V_ip6_fraglifetime;
254 CURVNET_RESTORE();
255
256 if (val > host_val)
257 val = host_val;
258 }
259 #endif
260 V_ip6_fraglifetime = val;
261 return (0);
262 }
263 SYSCTL_PROC(_net_inet6_ip6, OID_AUTO, fraglifetime_ms,
264 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
265 NULL, 0, sysctl_ip6_fraglifetime, "I",
266 "Fragment lifetime, in milliseconds");
267
268 /*
269 * Remove the IPv6 fragmentation header from the mbuf.
270 */
271 int
ip6_deletefraghdr(struct mbuf * m,int offset,int wait __unused)272 ip6_deletefraghdr(struct mbuf *m, int offset, int wait __unused)
273 {
274 struct ip6_hdr *ip6;
275
276 KASSERT(m->m_len >= offset + sizeof(struct ip6_frag),
277 ("%s: ext headers not contigous in mbuf %p m_len %d >= "
278 "offset %d + %zu\n", __func__, m, m->m_len, offset,
279 sizeof(struct ip6_frag)));
280
281 /* Delete frag6 header. */
282 ip6 = mtod(m, struct ip6_hdr *);
283 bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag), offset);
284 m->m_data += sizeof(struct ip6_frag);
285 m->m_len -= sizeof(struct ip6_frag);
286 m->m_flags |= M_FRAGMENTED;
287
288 return (0);
289 }
290
291 static void
frag6_rmqueue(struct ip6q * q6,uint32_t bucket)292 frag6_rmqueue(struct ip6q *q6, uint32_t bucket)
293 {
294 IP6QB_LOCK_ASSERT(bucket);
295
296 TAILQ_REMOVE(IP6QB_HEAD(bucket), q6, ip6q_tq);
297 V_ip6qb[bucket].count--;
298 #ifdef MAC
299 mac_ip6q_destroy(q6);
300 #endif
301 free(q6, M_FRAG6);
302 atomic_subtract_int(&V_frag6_nfragpackets, 1);
303 }
304
305 /*
306 * Free a fragment reassembly header and all associated datagrams.
307 */
308 static void
frag6_freef(struct ip6q * q6,uint32_t bucket)309 frag6_freef(struct ip6q *q6, uint32_t bucket)
310 {
311 struct ip6_hdr *ip6;
312 struct ip6asfrag *af6;
313 struct mbuf *m;
314
315 IP6QB_LOCK_ASSERT(bucket);
316
317 while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) {
318 m = af6->ip6af_m;
319 TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq);
320
321 /*
322 * Return ICMP time exceeded error for the 1st fragment.
323 * Just free other fragments.
324 */
325 if (af6->ip6af_off == 0 && m->m_pkthdr.rcvif != NULL) {
326 /* Adjust pointer. */
327 ip6 = mtod(m, struct ip6_hdr *);
328
329 /* Restore source and destination addresses. */
330 ip6->ip6_src = q6->ip6q_src;
331 ip6->ip6_dst = q6->ip6q_dst;
332
333 icmp6_error(m, ICMP6_TIME_EXCEEDED,
334 ICMP6_TIME_EXCEED_REASSEMBLY, 0);
335 } else
336 m_freem(m);
337
338 free(af6, M_FRAG6);
339 }
340
341 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
342 frag6_rmqueue(q6, bucket);
343 }
344
345 /*
346 * Drain off all datagram fragments belonging to
347 * the given network interface.
348 */
349 static void
frag6_cleanup(void * arg __unused,struct ifnet * ifp)350 frag6_cleanup(void *arg __unused, struct ifnet *ifp)
351 {
352 struct ip6qhead *head;
353 struct ip6q *q6;
354 struct ip6asfrag *af6;
355 uint32_t bucket;
356
357 KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__));
358
359 CURVNET_SET_QUIET(ifp->if_vnet);
360 #ifdef VIMAGE
361 /*
362 * Skip processing if IPv6 reassembly is not initialised or
363 * torn down by frag6_destroy().
364 */
365 if (!V_frag6_on) {
366 CURVNET_RESTORE();
367 return;
368 }
369 #endif
370
371 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
372 IP6QB_LOCK(bucket);
373 head = IP6QB_HEAD(bucket);
374 /* Scan fragment list. */
375 TAILQ_FOREACH(q6, head, ip6q_tq) {
376 TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) {
377 /* Clear no longer valid rcvif pointer. */
378 if (af6->ip6af_m->m_pkthdr.rcvif == ifp)
379 af6->ip6af_m->m_pkthdr.rcvif = NULL;
380 }
381 }
382 IP6QB_UNLOCK(bucket);
383 }
384 CURVNET_RESTORE();
385 }
386 EVENTHANDLER_DEFINE(ifnet_departure_event, frag6_cleanup, NULL, 0);
387
388 /*
389 * Like in RFC2460, in RFC8200, fragment and reassembly rules do not agree with
390 * each other, in terms of next header field handling in fragment header.
391 * While the sender will use the same value for all of the fragmented packets,
392 * receiver is suggested not to check for consistency.
393 *
394 * Fragment rules (p18,p19):
395 * (2) A Fragment header containing:
396 * The Next Header value that identifies the first header
397 * after the Per-Fragment headers of the original packet.
398 * -> next header field is same for all fragments
399 *
400 * Reassembly rule (p20):
401 * The Next Header field of the last header of the Per-Fragment
402 * headers is obtained from the Next Header field of the first
403 * fragment's Fragment header.
404 * -> should grab it from the first fragment only
405 *
406 * The following note also contradicts with fragment rule - no one is going to
407 * send different fragment with different next header field.
408 *
409 * Additional note (p22) [not an error]:
410 * The Next Header values in the Fragment headers of different
411 * fragments of the same original packet may differ. Only the value
412 * from the Offset zero fragment packet is used for reassembly.
413 * -> should grab it from the first fragment only
414 *
415 * There is no explicit reason given in the RFC. Historical reason maybe?
416 */
417 /*
418 * Fragment input.
419 */
420 int
frag6_input(struct mbuf ** mp,int * offp,int proto)421 frag6_input(struct mbuf **mp, int *offp, int proto)
422 {
423 struct mbuf *m, *t;
424 struct ip6_hdr *ip6;
425 struct ip6_frag *ip6f;
426 struct ip6qhead *head;
427 struct ip6q *q6;
428 struct ip6asfrag *af6, *ip6af, *af6tmp;
429 struct in6_ifaddr *ia6;
430 struct ifnet *dstifp, *srcifp;
431 uint32_t hashkey[(sizeof(struct in6_addr) * 2 +
432 sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)];
433 uint32_t bucket, *hashkeyp;
434 int fragoff, frgpartlen; /* Must be larger than uint16_t. */
435 int nxt, offset, plen;
436 uint8_t ecn, ecn0;
437 bool only_frag;
438 #ifdef RSS
439 struct ip6_direct_ctx *ip6dc;
440 struct m_tag *mtag;
441 #endif
442
443 m = *mp;
444 offset = *offp;
445
446 M_ASSERTPKTHDR(m);
447
448 if (m->m_len < offset + sizeof(struct ip6_frag)) {
449 m = m_pullup(m, offset + sizeof(struct ip6_frag));
450 if (m == NULL) {
451 IP6STAT_INC(ip6s_exthdrtoolong);
452 *mp = NULL;
453 return (IPPROTO_DONE);
454 }
455 }
456 ip6 = mtod(m, struct ip6_hdr *);
457
458 dstifp = NULL;
459 /* Find the destination interface of the packet. */
460 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
461 if (ia6 != NULL)
462 dstifp = ia6->ia_ifp;
463
464 /* Jumbo payload cannot contain a fragment header. */
465 if (ip6->ip6_plen == 0) {
466 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset);
467 in6_ifstat_inc(dstifp, ifs6_reass_fail);
468 *mp = NULL;
469 return (IPPROTO_DONE);
470 }
471
472 /*
473 * Check whether fragment packet's fragment length is a
474 * multiple of 8 octets (unless it is the last one).
475 * sizeof(struct ip6_frag) == 8
476 * sizeof(struct ip6_hdr) = 40
477 */
478 ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset);
479 if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) &&
480 (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) {
481 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
482 offsetof(struct ip6_hdr, ip6_plen));
483 in6_ifstat_inc(dstifp, ifs6_reass_fail);
484 *mp = NULL;
485 return (IPPROTO_DONE);
486 }
487
488 IP6STAT_INC(ip6s_fragments);
489 in6_ifstat_inc(dstifp, ifs6_reass_reqd);
490
491 /*
492 * Handle "atomic" fragments (offset and m bit set to 0) upfront,
493 * unrelated to any reassembly. We need to remove the frag hdr
494 * which is ugly.
495 * See RFC 6946 and section 4.5 of RFC 8200.
496 */
497 if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
498 IP6STAT_INC(ip6s_atomicfrags);
499 nxt = ip6f->ip6f_nxt;
500 /*
501 * Set nxt(-hdr field value) to the original value.
502 * We cannot just set ip6->ip6_nxt as there might be
503 * an unfragmentable part with extension headers and
504 * we must update the last one.
505 */
506 m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t),
507 (caddr_t)&nxt);
508 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) -
509 sizeof(struct ip6_frag));
510 if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0)
511 goto dropfrag2;
512 m->m_pkthdr.len -= sizeof(struct ip6_frag);
513 in6_ifstat_inc(dstifp, ifs6_reass_ok);
514 *mp = m;
515 return (nxt);
516 }
517
518 /* Offset now points to data portion. */
519 offset += sizeof(struct ip6_frag);
520
521 /* Get fragment length and discard 0-byte fragments. */
522 frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset;
523 if (frgpartlen == 0) {
524 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
525 offsetof(struct ip6_hdr, ip6_plen));
526 in6_ifstat_inc(dstifp, ifs6_reass_fail);
527 IP6STAT_INC(ip6s_fragdropped);
528 *mp = NULL;
529 return (IPPROTO_DONE);
530 }
531
532 /*
533 * Enforce upper bound on number of fragments for the entire system.
534 * If maxfrag is 0, never accept fragments.
535 * If maxfrag is -1, accept all fragments without limitation.
536 */
537 if (ip6_maxfrags < 0)
538 ;
539 else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags)
540 goto dropfrag2;
541
542 /*
543 * Validate that a full header chain to the ULP is present in the
544 * packet containing the first fragment as per RFC RFC7112 and
545 * RFC 8200 pages 18,19:
546 * The first fragment packet is composed of:
547 * (3) Extension headers, if any, and the Upper-Layer header. These
548 * headers must be in the first fragment. ...
549 */
550 fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK);
551 /* XXX TODO. thj has D16851 open for this. */
552 /* Send ICMPv6 4,3 in case of violation. */
553
554 /* Store receive network interface pointer for later. */
555 srcifp = m->m_pkthdr.rcvif;
556
557 /* Generate a hash value for fragment bucket selection. */
558 hashkeyp = hashkey;
559 memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr));
560 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
561 memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr));
562 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp);
563 *hashkeyp = ip6f->ip6f_ident;
564 bucket = jenkins_hash32(hashkey, nitems(hashkey), V_ip6qb_hashseed);
565 bucket &= IP6REASS_HMASK;
566 IP6QB_LOCK(bucket);
567 head = IP6QB_HEAD(bucket);
568
569 TAILQ_FOREACH(q6, head, ip6q_tq)
570 if (ip6f->ip6f_ident == q6->ip6q_ident &&
571 IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) &&
572 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst)
573 #ifdef MAC
574 && mac_ip6q_match(m, q6)
575 #endif
576 )
577 break;
578
579 only_frag = false;
580 if (q6 == NULL) {
581 /* A first fragment to arrive creates a reassembly queue. */
582 only_frag = true;
583
584 /*
585 * Enforce upper bound on number of fragmented packets
586 * for which we attempt reassembly;
587 * If maxfragpackets is 0, never accept fragments.
588 * If maxfragpackets is -1, accept all fragments without
589 * limitation.
590 */
591 if (V_ip6_maxfragpackets < 0)
592 ;
593 else if (V_ip6qb[bucket].count >= V_ip6_maxfragbucketsize ||
594 atomic_load_int(&V_frag6_nfragpackets) >=
595 (u_int)V_ip6_maxfragpackets)
596 goto dropfrag;
597
598 /* Allocate IPv6 fragement packet queue entry. */
599 q6 = malloc(sizeof(struct ip6q), M_FRAG6, M_NOWAIT | M_ZERO);
600 if (q6 == NULL)
601 goto dropfrag;
602 #ifdef MAC
603 if (mac_ip6q_init(q6, M_NOWAIT) != 0) {
604 free(q6, M_FRAG6);
605 goto dropfrag;
606 }
607 mac_ip6q_create(m, q6);
608 #endif
609 atomic_add_int(&V_frag6_nfragpackets, 1);
610
611 /* ip6q_nxt will be filled afterwards, from 1st fragment. */
612 TAILQ_INIT(&q6->ip6q_frags);
613 q6->ip6q_ident = ip6f->ip6f_ident;
614 q6->ip6q_ttl = frag6_milli_to_callout_ticks(V_ip6_fraglifetime);
615 q6->ip6q_src = ip6->ip6_src;
616 q6->ip6q_dst = ip6->ip6_dst;
617 q6->ip6q_ecn = IPV6_ECN(ip6);
618 q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */
619
620 /* Add the fragemented packet to the bucket. */
621 TAILQ_INSERT_HEAD(head, q6, ip6q_tq);
622 V_ip6qb[bucket].count++;
623 }
624
625 /*
626 * If it is the 1st fragment, record the length of the
627 * unfragmentable part and the next header of the fragment header.
628 * Assume the first 1st fragement to arrive will be correct.
629 * We do not have any duplicate checks here yet so another packet
630 * with fragoff == 0 could come and overwrite the ip6q_unfrglen
631 * and worse, the next header, at any time.
632 */
633 if (fragoff == 0 && q6->ip6q_unfrglen == -1) {
634 q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) -
635 sizeof(struct ip6_frag);
636 q6->ip6q_nxt = ip6f->ip6f_nxt;
637 /* XXX ECN? */
638 }
639
640 /*
641 * Check that the reassembled packet would not exceed 65535 bytes
642 * in size.
643 * If it would exceed, discard the fragment and return an ICMP error.
644 */
645 if (q6->ip6q_unfrglen >= 0) {
646 /* The 1st fragment has already arrived. */
647 if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) {
648 if (only_frag)
649 frag6_rmqueue(q6, bucket);
650 IP6QB_UNLOCK(bucket);
651 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
652 offset - sizeof(struct ip6_frag) +
653 offsetof(struct ip6_frag, ip6f_offlg));
654 *mp = NULL;
655 return (IPPROTO_DONE);
656 }
657 } else if (fragoff + frgpartlen > IPV6_MAXPACKET) {
658 if (only_frag)
659 frag6_rmqueue(q6, bucket);
660 IP6QB_UNLOCK(bucket);
661 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER,
662 offset - sizeof(struct ip6_frag) +
663 offsetof(struct ip6_frag, ip6f_offlg));
664 *mp = NULL;
665 return (IPPROTO_DONE);
666 }
667
668 /*
669 * If it is the first fragment, do the above check for each
670 * fragment already stored in the reassembly queue.
671 */
672 if (fragoff == 0 && !only_frag) {
673 TAILQ_FOREACH_SAFE(af6, &q6->ip6q_frags, ip6af_tq, af6tmp) {
674 if (q6->ip6q_unfrglen + af6->ip6af_off +
675 af6->ip6af_frglen > IPV6_MAXPACKET) {
676 struct ip6_hdr *ip6err;
677 struct mbuf *merr;
678 int erroff;
679
680 merr = af6->ip6af_m;
681 erroff = af6->ip6af_offset;
682
683 /* Dequeue the fragment. */
684 TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq);
685 q6->ip6q_nfrag--;
686 atomic_subtract_int(&frag6_nfrags, 1);
687 free(af6, M_FRAG6);
688
689 /* Set a valid receive interface pointer. */
690 merr->m_pkthdr.rcvif = srcifp;
691
692 /* Adjust pointer. */
693 ip6err = mtod(merr, struct ip6_hdr *);
694
695 /*
696 * Restore source and destination addresses
697 * in the erroneous IPv6 header.
698 */
699 ip6err->ip6_src = q6->ip6q_src;
700 ip6err->ip6_dst = q6->ip6q_dst;
701
702 icmp6_error(merr, ICMP6_PARAM_PROB,
703 ICMP6_PARAMPROB_HEADER,
704 erroff - sizeof(struct ip6_frag) +
705 offsetof(struct ip6_frag, ip6f_offlg));
706 }
707 }
708 }
709
710 /* Allocate an IPv6 fragement queue entry for this fragmented part. */
711 ip6af = malloc(sizeof(struct ip6asfrag), M_FRAG6, M_NOWAIT | M_ZERO);
712 if (ip6af == NULL)
713 goto dropfrag;
714 ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) ? true : false;
715 ip6af->ip6af_off = fragoff;
716 ip6af->ip6af_frglen = frgpartlen;
717 ip6af->ip6af_offset = offset;
718 ip6af->ip6af_m = m;
719
720 if (only_frag) {
721 /*
722 * Do a manual insert rather than a hard-to-understand cast
723 * to a different type relying on data structure order to work.
724 */
725 TAILQ_INSERT_HEAD(&q6->ip6q_frags, ip6af, ip6af_tq);
726 goto postinsert;
727 }
728
729 /* Do duplicate, condition, and boundry checks. */
730 /*
731 * Handle ECN by comparing this segment with the first one;
732 * if CE is set, do not lose CE.
733 * Drop if CE and not-ECT are mixed for the same packet.
734 */
735 ecn = IPV6_ECN(ip6);
736 ecn0 = q6->ip6q_ecn;
737 if (ecn == IPTOS_ECN_CE) {
738 if (ecn0 == IPTOS_ECN_NOTECT) {
739 free(ip6af, M_FRAG6);
740 goto dropfrag;
741 }
742 if (ecn0 != IPTOS_ECN_CE)
743 q6->ip6q_ecn = IPTOS_ECN_CE;
744 }
745 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
746 free(ip6af, M_FRAG6);
747 goto dropfrag;
748 }
749
750 /* Find a fragmented part which begins after this one does. */
751 TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq)
752 if (af6->ip6af_off > ip6af->ip6af_off)
753 break;
754
755 /*
756 * If the incoming framgent overlaps some existing fragments in
757 * the reassembly queue, drop both the new fragment and the
758 * entire reassembly queue. However, if the new fragment
759 * is an exact duplicate of an existing fragment, only silently
760 * drop the existing fragment and leave the fragmentation queue
761 * unchanged, as allowed by the RFC. (RFC 8200, 4.5)
762 */
763 if (af6 != NULL)
764 af6tmp = TAILQ_PREV(af6, ip6fraghead, ip6af_tq);
765 else
766 af6tmp = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead);
767 if (af6tmp != NULL) {
768 if (af6tmp->ip6af_off + af6tmp->ip6af_frglen -
769 ip6af->ip6af_off > 0) {
770 if (af6tmp->ip6af_off != ip6af->ip6af_off ||
771 af6tmp->ip6af_frglen != ip6af->ip6af_frglen)
772 frag6_freef(q6, bucket);
773 free(ip6af, M_FRAG6);
774 goto dropfrag;
775 }
776 }
777 if (af6 != NULL) {
778 if (ip6af->ip6af_off + ip6af->ip6af_frglen -
779 af6->ip6af_off > 0) {
780 if (af6->ip6af_off != ip6af->ip6af_off ||
781 af6->ip6af_frglen != ip6af->ip6af_frglen)
782 frag6_freef(q6, bucket);
783 free(ip6af, M_FRAG6);
784 goto dropfrag;
785 }
786 }
787
788 #ifdef MAC
789 mac_ip6q_update(m, q6);
790 #endif
791
792 /*
793 * Stick new segment in its place; check for complete reassembly.
794 * If not complete, check fragment limit. Move to front of packet
795 * queue, as we are the most recently active fragmented packet.
796 */
797 if (af6 != NULL)
798 TAILQ_INSERT_BEFORE(af6, ip6af, ip6af_tq);
799 else
800 TAILQ_INSERT_TAIL(&q6->ip6q_frags, ip6af, ip6af_tq);
801 postinsert:
802 atomic_add_int(&frag6_nfrags, 1);
803 q6->ip6q_nfrag++;
804
805 plen = 0;
806 TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) {
807 if (af6->ip6af_off != plen) {
808 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
809 IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
810 frag6_freef(q6, bucket);
811 }
812 IP6QB_UNLOCK(bucket);
813 *mp = NULL;
814 return (IPPROTO_DONE);
815 }
816 plen += af6->ip6af_frglen;
817 }
818 af6 = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead);
819 if (af6->ip6af_mff) {
820 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) {
821 IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag);
822 frag6_freef(q6, bucket);
823 }
824 IP6QB_UNLOCK(bucket);
825 *mp = NULL;
826 return (IPPROTO_DONE);
827 }
828
829 /* Reassembly is complete; concatenate fragments. */
830 ip6af = TAILQ_FIRST(&q6->ip6q_frags);
831 t = m = ip6af->ip6af_m;
832 TAILQ_REMOVE(&q6->ip6q_frags, ip6af, ip6af_tq);
833 while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) {
834 m->m_pkthdr.csum_flags &=
835 af6->ip6af_m->m_pkthdr.csum_flags;
836 m->m_pkthdr.csum_data +=
837 af6->ip6af_m->m_pkthdr.csum_data;
838
839 TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq);
840 t = m_last(t);
841 m_adj(af6->ip6af_m, af6->ip6af_offset);
842 m_demote_pkthdr(af6->ip6af_m);
843 m_cat(t, af6->ip6af_m);
844 free(af6, M_FRAG6);
845 }
846
847 while (m->m_pkthdr.csum_data & 0xffff0000)
848 m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) +
849 (m->m_pkthdr.csum_data >> 16);
850
851 /* Adjust offset to point where the original next header starts. */
852 offset = ip6af->ip6af_offset - sizeof(struct ip6_frag);
853 free(ip6af, M_FRAG6);
854 if ((u_int)plen + (u_int)offset - sizeof(struct ip6_hdr) >
855 IPV6_MAXPACKET) {
856 frag6_freef(q6, bucket);
857 goto dropfrag;
858 }
859 ip6 = mtod(m, struct ip6_hdr *);
860 ip6->ip6_plen = htons((u_short)plen + offset - sizeof(struct ip6_hdr));
861 if (q6->ip6q_ecn == IPTOS_ECN_CE)
862 ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20);
863 nxt = q6->ip6q_nxt;
864
865 ip6_deletefraghdr(m, offset, M_NOWAIT);
866
867 /* Set nxt(-hdr field value) to the original value. */
868 m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t),
869 (caddr_t)&nxt);
870
871 #ifdef MAC
872 mac_ip6q_reassemble(q6, m);
873 #endif
874 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag);
875 frag6_rmqueue(q6, bucket);
876
877 if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */
878
879 plen = 0;
880 for (t = m; t; t = t->m_next)
881 plen += t->m_len;
882 m->m_pkthdr.len = plen;
883 /* Set a valid receive interface pointer. */
884 m->m_pkthdr.rcvif = srcifp;
885 }
886
887 #ifdef RSS
888 mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc),
889 M_NOWAIT);
890 if (mtag == NULL)
891 goto dropfrag;
892
893 ip6dc = (struct ip6_direct_ctx *)(mtag + 1);
894 ip6dc->ip6dc_nxt = nxt;
895 ip6dc->ip6dc_off = offset;
896
897 m_tag_prepend(m, mtag);
898 #endif
899
900 IP6QB_UNLOCK(bucket);
901 IP6STAT_INC(ip6s_reassembled);
902 in6_ifstat_inc(dstifp, ifs6_reass_ok);
903
904 #ifdef RSS
905 /* Queue/dispatch for reprocessing. */
906 netisr_dispatch(NETISR_IPV6_DIRECT, m);
907 *mp = NULL;
908 return (IPPROTO_DONE);
909 #endif
910
911 /* Tell launch routine the next header. */
912 *mp = m;
913 *offp = offset;
914
915 return (nxt);
916
917 dropfrag:
918 IP6QB_UNLOCK(bucket);
919 dropfrag2:
920 in6_ifstat_inc(dstifp, ifs6_reass_fail);
921 IP6STAT_INC(ip6s_fragdropped);
922 m_freem(m);
923 *mp = NULL;
924 return (IPPROTO_DONE);
925 }
926
927 /*
928 * IPv6 reassembling timer processing;
929 * if a timer expires on a reassembly queue, discard it.
930 */
931 static struct callout frag6_callout;
932 static void
frag6_slowtimo(void * arg __unused)933 frag6_slowtimo(void *arg __unused)
934 {
935 VNET_ITERATOR_DECL(vnet_iter);
936 struct ip6qhead *head;
937 struct ip6q *q6, *q6tmp;
938 uint32_t bucket;
939
940 if (atomic_load_int(&frag6_nfrags) == 0)
941 goto done;
942
943 VNET_LIST_RLOCK_NOSLEEP();
944 VNET_FOREACH(vnet_iter) {
945 CURVNET_SET(vnet_iter);
946 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
947 if (V_ip6qb[bucket].count == 0)
948 continue;
949 IP6QB_LOCK(bucket);
950 head = IP6QB_HEAD(bucket);
951 TAILQ_FOREACH_SAFE(q6, head, ip6q_tq, q6tmp)
952 if (--q6->ip6q_ttl == 0) {
953 IP6STAT_ADD(ip6s_fragtimeout,
954 q6->ip6q_nfrag);
955 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
956 frag6_freef(q6, bucket);
957 }
958 /*
959 * If we are over the maximum number of fragments
960 * (due to the limit being lowered), drain off
961 * enough to get down to the new limit.
962 * Note that we drain all reassembly queues if
963 * maxfragpackets is 0 (fragmentation is disabled),
964 * and do not enforce a limit when maxfragpackets
965 * is negative.
966 */
967 while ((V_ip6_maxfragpackets == 0 ||
968 (V_ip6_maxfragpackets > 0 &&
969 V_ip6qb[bucket].count > V_ip6_maxfragbucketsize)) &&
970 (q6 = TAILQ_LAST(head, ip6qhead)) != NULL) {
971 IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag);
972 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
973 frag6_freef(q6, bucket);
974 }
975 IP6QB_UNLOCK(bucket);
976 }
977 /*
978 * If we are still over the maximum number of fragmented
979 * packets, drain off enough to get down to the new limit.
980 */
981 bucket = 0;
982 while (V_ip6_maxfragpackets >= 0 &&
983 atomic_load_int(&V_frag6_nfragpackets) >
984 (u_int)V_ip6_maxfragpackets) {
985 IP6QB_LOCK(bucket);
986 q6 = TAILQ_LAST(IP6QB_HEAD(bucket), ip6qhead);
987 if (q6 != NULL) {
988 IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag);
989 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
990 frag6_freef(q6, bucket);
991 }
992 IP6QB_UNLOCK(bucket);
993 bucket = (bucket + 1) % IP6REASS_NHASH;
994 }
995 CURVNET_RESTORE();
996 }
997 VNET_LIST_RUNLOCK_NOSLEEP();
998 done:
999 callout_reset_sbt(&frag6_callout, SBT_1MS * IP6_CALLOUT_INTERVAL_MS,
1000 SBT_1MS * 10, frag6_slowtimo, NULL, 0);
1001 }
1002
1003 static void
frag6_slowtimo_init(void * arg __unused)1004 frag6_slowtimo_init(void *arg __unused)
1005 {
1006
1007 callout_init(&frag6_callout, 1);
1008 callout_reset_sbt(&frag6_callout, SBT_1MS * IP6_CALLOUT_INTERVAL_MS,
1009 SBT_1MS * 10, frag6_slowtimo, NULL, 0);
1010 }
1011 SYSINIT(frag6, SI_SUB_VNET_DONE, SI_ORDER_ANY, frag6_slowtimo_init, NULL);
1012
1013 /*
1014 * Eventhandler to adjust limits in case nmbclusters change.
1015 */
1016 static void
frag6_change(void * tag)1017 frag6_change(void *tag)
1018 {
1019 VNET_ITERATOR_DECL(vnet_iter);
1020
1021 ip6_maxfrags = IP6_MAXFRAGS;
1022 VNET_LIST_RLOCK_NOSLEEP();
1023 VNET_FOREACH(vnet_iter) {
1024 CURVNET_SET(vnet_iter);
1025 V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
1026 frag6_set_bucketsize();
1027 CURVNET_RESTORE();
1028 }
1029 VNET_LIST_RUNLOCK_NOSLEEP();
1030 }
1031
1032 /*
1033 * Initialise reassembly queue and fragment identifier.
1034 */
1035 void
frag6_init(void)1036 frag6_init(void)
1037 {
1038 uint32_t bucket;
1039
1040 V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS;
1041 frag6_set_bucketsize();
1042 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
1043 TAILQ_INIT(IP6QB_HEAD(bucket));
1044 mtx_init(&V_ip6qb[bucket].lock, "ip6qb", NULL, MTX_DEF);
1045 V_ip6qb[bucket].count = 0;
1046 }
1047 V_ip6qb_hashseed = arc4random();
1048 V_ip6_maxfragsperpacket = 64;
1049 #ifdef VIMAGE
1050 V_frag6_on = true;
1051 #endif
1052 if (!IS_DEFAULT_VNET(curvnet))
1053 return;
1054
1055 ip6_maxfrags = IP6_MAXFRAGS;
1056 EVENTHANDLER_REGISTER(nmbclusters_change,
1057 frag6_change, NULL, EVENTHANDLER_PRI_ANY);
1058 }
1059
1060 /*
1061 * Drain off all datagram fragments.
1062 */
1063 static void
frag6_drain_one(void)1064 frag6_drain_one(void)
1065 {
1066 struct ip6q *q6;
1067 uint32_t bucket;
1068
1069 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
1070 IP6QB_LOCK(bucket);
1071 while ((q6 = TAILQ_FIRST(IP6QB_HEAD(bucket))) != NULL) {
1072 IP6STAT_INC(ip6s_fragdropped);
1073 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */
1074 frag6_freef(q6, bucket);
1075 }
1076 IP6QB_UNLOCK(bucket);
1077 }
1078 }
1079
1080 void
frag6_drain(void)1081 frag6_drain(void)
1082 {
1083 VNET_ITERATOR_DECL(vnet_iter);
1084
1085 VNET_LIST_RLOCK_NOSLEEP();
1086 VNET_FOREACH(vnet_iter) {
1087 CURVNET_SET(vnet_iter);
1088 frag6_drain_one();
1089 CURVNET_RESTORE();
1090 }
1091 VNET_LIST_RUNLOCK_NOSLEEP();
1092 }
1093
1094 #ifdef VIMAGE
1095 /*
1096 * Clear up IPv6 reassembly structures.
1097 */
1098 void
frag6_destroy(void)1099 frag6_destroy(void)
1100 {
1101 uint32_t bucket;
1102
1103 frag6_drain_one();
1104 V_frag6_on = false;
1105 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) {
1106 KASSERT(V_ip6qb[bucket].count == 0,
1107 ("%s: V_ip6qb[%d] (%p) count not 0 (%d)", __func__,
1108 bucket, &V_ip6qb[bucket], V_ip6qb[bucket].count));
1109 mtx_destroy(&V_ip6qb[bucket].lock);
1110 }
1111 }
1112 #endif
1113