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 __FBSDID("$FreeBSD$"); 37 38 #include "opt_rss.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/domain.h> 43 #include <sys/eventhandler.h> 44 #include <sys/hash.h> 45 #include <sys/kernel.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/protosw.h> 49 #include <sys/queue.h> 50 #include <sys/socket.h> 51 #include <sys/sysctl.h> 52 #include <sys/syslog.h> 53 54 #include <net/if.h> 55 #include <net/if_var.h> 56 #include <net/netisr.h> 57 #include <net/route.h> 58 #include <net/vnet.h> 59 60 #include <netinet/in.h> 61 #include <netinet/in_var.h> 62 #include <netinet/ip6.h> 63 #include <netinet6/ip6_var.h> 64 #include <netinet/icmp6.h> 65 #include <netinet/in_systm.h> /* For ECN definitions. */ 66 #include <netinet/ip.h> /* For ECN definitions. */ 67 68 #ifdef MAC 69 #include <security/mac/mac_framework.h> 70 #endif 71 72 /* 73 * A "big picture" of how IPv6 fragment queues are all linked together. 74 * 75 * struct ip6qbucket ip6qb[...]; hashed buckets 76 * |||||||| 77 * | 78 * +--- TAILQ(struct ip6q, packets) *q6; tailq entries holding 79 * |||||||| fragmented packets 80 * | (1 per original packet) 81 * | 82 * +--- TAILQ(struct ip6asfrag, ip6q_frags) *af6; tailq entries of IPv6 83 * | *ip6af;fragment packets 84 * | for one original packet 85 * + *mbuf 86 */ 87 88 /* Reassembly headers are stored in hash buckets. */ 89 #define IP6REASS_NHASH_LOG2 10 90 #define IP6REASS_NHASH (1 << IP6REASS_NHASH_LOG2) 91 #define IP6REASS_HMASK (IP6REASS_NHASH - 1) 92 93 TAILQ_HEAD(ip6qhead, ip6q); 94 struct ip6qbucket { 95 struct ip6qhead packets; 96 struct mtx lock; 97 int count; 98 }; 99 100 struct ip6asfrag { 101 TAILQ_ENTRY(ip6asfrag) ip6af_tq; 102 struct mbuf *ip6af_m; 103 int ip6af_offset; /* Offset in ip6af_m to next header. */ 104 int ip6af_frglen; /* Fragmentable part length. */ 105 int ip6af_off; /* Fragment offset. */ 106 bool ip6af_mff; /* More fragment bit in frag off. */ 107 }; 108 109 static MALLOC_DEFINE(M_FRAG6, "frag6", "IPv6 fragment reassembly header"); 110 111 #ifdef VIMAGE 112 /* A flag to indicate if IPv6 fragmentation is initialized. */ 113 VNET_DEFINE_STATIC(bool, frag6_on); 114 #define V_frag6_on VNET(frag6_on) 115 #endif 116 117 /* System wide (global) maximum and count of packets in reassembly queues. */ 118 static int ip6_maxfrags; 119 static volatile u_int frag6_nfrags = 0; 120 121 /* Maximum and current packets in per-VNET reassembly queue. */ 122 VNET_DEFINE_STATIC(int, ip6_maxfragpackets); 123 VNET_DEFINE_STATIC(volatile u_int, frag6_nfragpackets); 124 #define V_ip6_maxfragpackets VNET(ip6_maxfragpackets) 125 #define V_frag6_nfragpackets VNET(frag6_nfragpackets) 126 127 /* Maximum per-VNET reassembly queues per bucket and fragments per packet. */ 128 VNET_DEFINE_STATIC(int, ip6_maxfragbucketsize); 129 VNET_DEFINE_STATIC(int, ip6_maxfragsperpacket); 130 #define V_ip6_maxfragbucketsize VNET(ip6_maxfragbucketsize) 131 #define V_ip6_maxfragsperpacket VNET(ip6_maxfragsperpacket) 132 133 /* Per-VNET reassembly queue buckets. */ 134 VNET_DEFINE_STATIC(struct ip6qbucket, ip6qb[IP6REASS_NHASH]); 135 VNET_DEFINE_STATIC(uint32_t, ip6qb_hashseed); 136 #define V_ip6qb VNET(ip6qb) 137 #define V_ip6qb_hashseed VNET(ip6qb_hashseed) 138 139 #define IP6QB_LOCK(_b) mtx_lock(&V_ip6qb[(_b)].lock) 140 #define IP6QB_TRYLOCK(_b) mtx_trylock(&V_ip6qb[(_b)].lock) 141 #define IP6QB_LOCK_ASSERT(_b) mtx_assert(&V_ip6qb[(_b)].lock, MA_OWNED) 142 #define IP6QB_UNLOCK(_b) mtx_unlock(&V_ip6qb[(_b)].lock) 143 #define IP6QB_HEAD(_b) (&V_ip6qb[(_b)].packets) 144 145 /* 146 * By default, limit the number of IP6 fragments across all reassembly 147 * queues to 1/32 of the total number of mbuf clusters. 148 * 149 * Limit the total number of reassembly queues per VNET to the 150 * IP6 fragment limit, but ensure the limit will not allow any bucket 151 * to grow above 100 items. (The bucket limit is 152 * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct 153 * multiplier to reach a 100-item limit.) 154 * The 100-item limit was chosen as brief testing seems to show that 155 * this produces "reasonable" performance on some subset of systems 156 * under DoS attack. 157 */ 158 #define IP6_MAXFRAGS (nmbclusters / 32) 159 #define IP6_MAXFRAGPACKETS (imin(IP6_MAXFRAGS, IP6REASS_NHASH * 50)) 160 161 /* 162 * Sysctls and helper function. 163 */ 164 SYSCTL_DECL(_net_inet6_ip6); 165 166 SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfrags, 167 CTLFLAG_RD, __DEVOLATILE(u_int *, &frag6_nfrags), 0, 168 "Global number of IPv6 fragments across all reassembly queues."); 169 170 static void 171 frag6_set_bucketsize(void) 172 { 173 int i; 174 175 if ((i = V_ip6_maxfragpackets) > 0) 176 V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1); 177 } 178 179 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGS, maxfrags, 180 CTLFLAG_RW, &ip6_maxfrags, 0, 181 "Maximum allowed number of outstanding IPv6 packet fragments. " 182 "A value of 0 means no fragmented packets will be accepted, while a " 183 "a value of -1 means no limit"); 184 185 static int 186 sysctl_ip6_maxfragpackets(SYSCTL_HANDLER_ARGS) 187 { 188 int error, val; 189 190 val = V_ip6_maxfragpackets; 191 error = sysctl_handle_int(oidp, &val, 0, req); 192 if (error != 0 || !req->newptr) 193 return (error); 194 V_ip6_maxfragpackets = val; 195 frag6_set_bucketsize(); 196 return (0); 197 } 198 SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGPACKETS, maxfragpackets, 199 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 200 NULL, 0, sysctl_ip6_maxfragpackets, "I", 201 "Default maximum number of outstanding fragmented IPv6 packets. " 202 "A value of 0 means no fragmented packets will be accepted, while a " 203 "a value of -1 means no limit"); 204 SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfragpackets, 205 CTLFLAG_VNET | CTLFLAG_RD, 206 __DEVOLATILE(u_int *, &VNET_NAME(frag6_nfragpackets)), 0, 207 "Per-VNET number of IPv6 fragments across all reassembly queues."); 208 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGSPERPACKET, maxfragsperpacket, 209 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragsperpacket), 0, 210 "Maximum allowed number of fragments per packet"); 211 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGBUCKETSIZE, maxfragbucketsize, 212 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragbucketsize), 0, 213 "Maximum number of reassembly queues per hash bucket"); 214 215 /* 216 * Remove the IPv6 fragmentation header from the mbuf. 217 */ 218 int 219 ip6_deletefraghdr(struct mbuf *m, int offset, int wait __unused) 220 { 221 struct ip6_hdr *ip6; 222 223 KASSERT(m->m_len >= offset + sizeof(struct ip6_frag), 224 ("%s: ext headers not contigous in mbuf %p m_len %d >= " 225 "offset %d + %zu\n", __func__, m, m->m_len, offset, 226 sizeof(struct ip6_frag))); 227 228 /* Delete frag6 header. */ 229 ip6 = mtod(m, struct ip6_hdr *); 230 bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag), offset); 231 m->m_data += sizeof(struct ip6_frag); 232 m->m_len -= sizeof(struct ip6_frag); 233 m->m_flags |= M_FRAGMENTED; 234 235 return (0); 236 } 237 238 /* 239 * Free a fragment reassembly header and all associated datagrams. 240 */ 241 static void 242 frag6_freef(struct ip6q *q6, uint32_t bucket) 243 { 244 struct ip6_hdr *ip6; 245 struct ip6asfrag *af6; 246 struct mbuf *m; 247 248 IP6QB_LOCK_ASSERT(bucket); 249 250 while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 251 m = af6->ip6af_m; 252 TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 253 254 /* 255 * Return ICMP time exceeded error for the 1st fragment. 256 * Just free other fragments. 257 */ 258 if (af6->ip6af_off == 0 && m->m_pkthdr.rcvif != NULL) { 259 /* Adjust pointer. */ 260 ip6 = mtod(m, struct ip6_hdr *); 261 262 /* Restore source and destination addresses. */ 263 ip6->ip6_src = q6->ip6q_src; 264 ip6->ip6_dst = q6->ip6q_dst; 265 266 icmp6_error(m, ICMP6_TIME_EXCEEDED, 267 ICMP6_TIME_EXCEED_REASSEMBLY, 0); 268 } else 269 m_freem(m); 270 271 free(af6, M_FRAG6); 272 } 273 274 TAILQ_REMOVE(IP6QB_HEAD(bucket), q6, ip6q_tq); 275 V_ip6qb[bucket].count--; 276 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 277 #ifdef MAC 278 mac_ip6q_destroy(q6); 279 #endif 280 free(q6, M_FRAG6); 281 atomic_subtract_int(&V_frag6_nfragpackets, 1); 282 } 283 284 /* 285 * Drain off all datagram fragments belonging to 286 * the given network interface. 287 */ 288 static void 289 frag6_cleanup(void *arg __unused, struct ifnet *ifp) 290 { 291 struct ip6qhead *head; 292 struct ip6q *q6; 293 struct ip6asfrag *af6; 294 uint32_t bucket; 295 296 KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__)); 297 298 CURVNET_SET_QUIET(ifp->if_vnet); 299 #ifdef VIMAGE 300 /* 301 * Skip processing if IPv6 reassembly is not initialised or 302 * torn down by frag6_destroy(). 303 */ 304 if (!V_frag6_on) { 305 CURVNET_RESTORE(); 306 return; 307 } 308 #endif 309 310 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 311 IP6QB_LOCK(bucket); 312 head = IP6QB_HEAD(bucket); 313 /* Scan fragment list. */ 314 TAILQ_FOREACH(q6, head, ip6q_tq) { 315 TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 316 /* Clear no longer valid rcvif pointer. */ 317 if (af6->ip6af_m->m_pkthdr.rcvif == ifp) 318 af6->ip6af_m->m_pkthdr.rcvif = NULL; 319 } 320 } 321 IP6QB_UNLOCK(bucket); 322 } 323 CURVNET_RESTORE(); 324 } 325 EVENTHANDLER_DEFINE(ifnet_departure_event, frag6_cleanup, NULL, 0); 326 327 /* 328 * Like in RFC2460, in RFC8200, fragment and reassembly rules do not agree with 329 * each other, in terms of next header field handling in fragment header. 330 * While the sender will use the same value for all of the fragmented packets, 331 * receiver is suggested not to check for consistency. 332 * 333 * Fragment rules (p18,p19): 334 * (2) A Fragment header containing: 335 * The Next Header value that identifies the first header 336 * after the Per-Fragment headers of the original packet. 337 * -> next header field is same for all fragments 338 * 339 * Reassembly rule (p20): 340 * The Next Header field of the last header of the Per-Fragment 341 * headers is obtained from the Next Header field of the first 342 * fragment's Fragment header. 343 * -> should grab it from the first fragment only 344 * 345 * The following note also contradicts with fragment rule - no one is going to 346 * send different fragment with different next header field. 347 * 348 * Additional note (p22) [not an error]: 349 * The Next Header values in the Fragment headers of different 350 * fragments of the same original packet may differ. Only the value 351 * from the Offset zero fragment packet is used for reassembly. 352 * -> should grab it from the first fragment only 353 * 354 * There is no explicit reason given in the RFC. Historical reason maybe? 355 */ 356 /* 357 * Fragment input. 358 */ 359 int 360 frag6_input(struct mbuf **mp, int *offp, int proto) 361 { 362 struct mbuf *m, *t; 363 struct ip6_hdr *ip6; 364 struct ip6_frag *ip6f; 365 struct ip6qhead *head; 366 struct ip6q *q6; 367 struct ip6asfrag *af6, *ip6af, *af6tmp; 368 struct in6_ifaddr *ia6; 369 struct ifnet *dstifp, *srcifp; 370 uint32_t hashkey[(sizeof(struct in6_addr) * 2 + 371 sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)]; 372 uint32_t bucket, *hashkeyp; 373 int fragoff, frgpartlen; /* Must be larger than uint16_t. */ 374 int nxt, offset, plen; 375 uint8_t ecn, ecn0; 376 bool only_frag; 377 #ifdef RSS 378 struct ip6_direct_ctx *ip6dc; 379 struct m_tag *mtag; 380 #endif 381 382 m = *mp; 383 offset = *offp; 384 385 M_ASSERTPKTHDR(m); 386 387 if (m->m_len < offset + sizeof(struct ip6_frag)) { 388 m = m_pullup(m, offset + sizeof(struct ip6_frag)); 389 if (m == NULL) { 390 IP6STAT_INC(ip6s_exthdrtoolong); 391 *mp = NULL; 392 return (IPPROTO_DONE); 393 } 394 } 395 ip6 = mtod(m, struct ip6_hdr *); 396 397 dstifp = NULL; 398 /* Find the destination interface of the packet. */ 399 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 400 if (ia6 != NULL) { 401 dstifp = ia6->ia_ifp; 402 ifa_free(&ia6->ia_ifa); 403 } 404 405 /* Jumbo payload cannot contain a fragment header. */ 406 if (ip6->ip6_plen == 0) { 407 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset); 408 in6_ifstat_inc(dstifp, ifs6_reass_fail); 409 *mp = NULL; 410 return (IPPROTO_DONE); 411 } 412 413 /* 414 * Check whether fragment packet's fragment length is a 415 * multiple of 8 octets (unless it is the last one). 416 * sizeof(struct ip6_frag) == 8 417 * sizeof(struct ip6_hdr) = 40 418 */ 419 ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset); 420 if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) && 421 (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) { 422 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 423 offsetof(struct ip6_hdr, ip6_plen)); 424 in6_ifstat_inc(dstifp, ifs6_reass_fail); 425 *mp = NULL; 426 return (IPPROTO_DONE); 427 } 428 429 IP6STAT_INC(ip6s_fragments); 430 in6_ifstat_inc(dstifp, ifs6_reass_reqd); 431 432 /* 433 * Handle "atomic" fragments (offset and m bit set to 0) upfront, 434 * unrelated to any reassembly. We need to remove the frag hdr 435 * which is ugly. 436 * See RFC 6946 and section 4.5 of RFC 8200. 437 */ 438 if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) { 439 IP6STAT_INC(ip6s_atomicfrags); 440 nxt = ip6f->ip6f_nxt; 441 /* 442 * Set nxt(-hdr field value) to the original value. 443 * We cannot just set ip6->ip6_nxt as there might be 444 * an unfragmentable part with extension headers and 445 * we must update the last one. 446 */ 447 m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), 448 (caddr_t)&nxt); 449 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - 450 sizeof(struct ip6_frag)); 451 if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) 452 goto dropfrag2; 453 m->m_pkthdr.len -= sizeof(struct ip6_frag); 454 in6_ifstat_inc(dstifp, ifs6_reass_ok); 455 *mp = m; 456 return (nxt); 457 } 458 459 /* Offset now points to data portion. */ 460 offset += sizeof(struct ip6_frag); 461 462 /* Get fragment length and discard 0-byte fragments. */ 463 frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset; 464 if (frgpartlen == 0) { 465 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 466 offsetof(struct ip6_hdr, ip6_plen)); 467 in6_ifstat_inc(dstifp, ifs6_reass_fail); 468 IP6STAT_INC(ip6s_fragdropped); 469 *mp = NULL; 470 return (IPPROTO_DONE); 471 } 472 473 /* 474 * Enforce upper bound on number of fragments for the entire system. 475 * If maxfrag is 0, never accept fragments. 476 * If maxfrag is -1, accept all fragments without limitation. 477 */ 478 if (ip6_maxfrags < 0) 479 ; 480 else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags) 481 goto dropfrag2; 482 483 /* 484 * Validate that a full header chain to the ULP is present in the 485 * packet containing the first fragment as per RFC RFC7112 and 486 * RFC 8200 pages 18,19: 487 * The first fragment packet is composed of: 488 * (3) Extension headers, if any, and the Upper-Layer header. These 489 * headers must be in the first fragment. ... 490 */ 491 fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK); 492 /* XXX TODO. thj has D16851 open for this. */ 493 /* Send ICMPv6 4,3 in case of violation. */ 494 495 /* Store receive network interface pointer for later. */ 496 srcifp = m->m_pkthdr.rcvif; 497 498 /* Generate a hash value for fragment bucket selection. */ 499 hashkeyp = hashkey; 500 memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr)); 501 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 502 memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr)); 503 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 504 *hashkeyp = ip6f->ip6f_ident; 505 bucket = jenkins_hash32(hashkey, nitems(hashkey), V_ip6qb_hashseed); 506 bucket &= IP6REASS_HMASK; 507 IP6QB_LOCK(bucket); 508 head = IP6QB_HEAD(bucket); 509 510 TAILQ_FOREACH(q6, head, ip6q_tq) 511 if (ip6f->ip6f_ident == q6->ip6q_ident && 512 IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) && 513 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst) 514 #ifdef MAC 515 && mac_ip6q_match(m, q6) 516 #endif 517 ) 518 break; 519 520 only_frag = false; 521 if (q6 == NULL) { 522 /* A first fragment to arrive creates a reassembly queue. */ 523 only_frag = true; 524 525 /* 526 * Enforce upper bound on number of fragmented packets 527 * for which we attempt reassembly; 528 * If maxfragpackets is 0, never accept fragments. 529 * If maxfragpackets is -1, accept all fragments without 530 * limitation. 531 */ 532 if (V_ip6_maxfragpackets < 0) 533 ; 534 else if (V_ip6qb[bucket].count >= V_ip6_maxfragbucketsize || 535 atomic_load_int(&V_frag6_nfragpackets) >= 536 (u_int)V_ip6_maxfragpackets) 537 goto dropfrag; 538 539 /* Allocate IPv6 fragement packet queue entry. */ 540 q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FRAG6, 541 M_NOWAIT | M_ZERO); 542 if (q6 == NULL) 543 goto dropfrag; 544 #ifdef MAC 545 if (mac_ip6q_init(q6, M_NOWAIT) != 0) { 546 free(q6, M_FRAG6); 547 goto dropfrag; 548 } 549 mac_ip6q_create(m, q6); 550 #endif 551 atomic_add_int(&V_frag6_nfragpackets, 1); 552 553 /* ip6q_nxt will be filled afterwards, from 1st fragment. */ 554 TAILQ_INIT(&q6->ip6q_frags); 555 q6->ip6q_ident = ip6f->ip6f_ident; 556 q6->ip6q_ttl = IPV6_FRAGTTL; 557 q6->ip6q_src = ip6->ip6_src; 558 q6->ip6q_dst = ip6->ip6_dst; 559 q6->ip6q_ecn = 560 (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 561 q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */ 562 563 /* Add the fragemented packet to the bucket. */ 564 TAILQ_INSERT_HEAD(head, q6, ip6q_tq); 565 V_ip6qb[bucket].count++; 566 } 567 568 /* 569 * If it is the 1st fragment, record the length of the 570 * unfragmentable part and the next header of the fragment header. 571 * Assume the first 1st fragement to arrive will be correct. 572 * We do not have any duplicate checks here yet so another packet 573 * with fragoff == 0 could come and overwrite the ip6q_unfrglen 574 * and worse, the next header, at any time. 575 */ 576 if (fragoff == 0 && q6->ip6q_unfrglen == -1) { 577 q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) - 578 sizeof(struct ip6_frag); 579 q6->ip6q_nxt = ip6f->ip6f_nxt; 580 /* XXX ECN? */ 581 } 582 583 /* 584 * Check that the reassembled packet would not exceed 65535 bytes 585 * in size. 586 * If it would exceed, discard the fragment and return an ICMP error. 587 */ 588 if (q6->ip6q_unfrglen >= 0) { 589 /* The 1st fragment has already arrived. */ 590 if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) { 591 if (only_frag) { 592 TAILQ_REMOVE(head, q6, ip6q_tq); 593 V_ip6qb[bucket].count--; 594 atomic_subtract_int(&V_frag6_nfragpackets, 1); 595 #ifdef MAC 596 mac_ip6q_destroy(q6); 597 #endif 598 free(q6, M_FRAG6); 599 } 600 IP6QB_UNLOCK(bucket); 601 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 602 offset - sizeof(struct ip6_frag) + 603 offsetof(struct ip6_frag, ip6f_offlg)); 604 *mp = NULL; 605 return (IPPROTO_DONE); 606 } 607 } else if (fragoff + frgpartlen > IPV6_MAXPACKET) { 608 if (only_frag) { 609 TAILQ_REMOVE(head, q6, ip6q_tq); 610 V_ip6qb[bucket].count--; 611 atomic_subtract_int(&V_frag6_nfragpackets, 1); 612 #ifdef MAC 613 mac_ip6q_destroy(q6); 614 #endif 615 free(q6, M_FRAG6); 616 } 617 IP6QB_UNLOCK(bucket); 618 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 619 offset - sizeof(struct ip6_frag) + 620 offsetof(struct ip6_frag, ip6f_offlg)); 621 *mp = NULL; 622 return (IPPROTO_DONE); 623 } 624 625 /* 626 * If it is the first fragment, do the above check for each 627 * fragment already stored in the reassembly queue. 628 */ 629 if (fragoff == 0 && !only_frag) { 630 TAILQ_FOREACH_SAFE(af6, &q6->ip6q_frags, ip6af_tq, af6tmp) { 631 if (q6->ip6q_unfrglen + af6->ip6af_off + 632 af6->ip6af_frglen > IPV6_MAXPACKET) { 633 struct ip6_hdr *ip6err; 634 struct mbuf *merr; 635 int erroff; 636 637 merr = af6->ip6af_m; 638 erroff = af6->ip6af_offset; 639 640 /* Dequeue the fragment. */ 641 TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 642 q6->ip6q_nfrag--; 643 atomic_subtract_int(&frag6_nfrags, 1); 644 free(af6, M_FRAG6); 645 646 /* Set a valid receive interface pointer. */ 647 merr->m_pkthdr.rcvif = srcifp; 648 649 /* Adjust pointer. */ 650 ip6err = mtod(merr, struct ip6_hdr *); 651 652 /* 653 * Restore source and destination addresses 654 * in the erroneous IPv6 header. 655 */ 656 ip6err->ip6_src = q6->ip6q_src; 657 ip6err->ip6_dst = q6->ip6q_dst; 658 659 icmp6_error(merr, ICMP6_PARAM_PROB, 660 ICMP6_PARAMPROB_HEADER, 661 erroff - sizeof(struct ip6_frag) + 662 offsetof(struct ip6_frag, ip6f_offlg)); 663 } 664 } 665 } 666 667 /* Allocate an IPv6 fragement queue entry for this fragmented part. */ 668 ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FRAG6, 669 M_NOWAIT | M_ZERO); 670 if (ip6af == NULL) 671 goto dropfrag; 672 ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) ? true : false; 673 ip6af->ip6af_off = fragoff; 674 ip6af->ip6af_frglen = frgpartlen; 675 ip6af->ip6af_offset = offset; 676 ip6af->ip6af_m = m; 677 678 if (only_frag) { 679 /* 680 * Do a manual insert rather than a hard-to-understand cast 681 * to a different type relying on data structure order to work. 682 */ 683 TAILQ_INSERT_HEAD(&q6->ip6q_frags, ip6af, ip6af_tq); 684 goto postinsert; 685 } 686 687 /* Do duplicate, condition, and boundry checks. */ 688 /* 689 * Handle ECN by comparing this segment with the first one; 690 * if CE is set, do not lose CE. 691 * Drop if CE and not-ECT are mixed for the same packet. 692 */ 693 ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 694 ecn0 = q6->ip6q_ecn; 695 if (ecn == IPTOS_ECN_CE) { 696 if (ecn0 == IPTOS_ECN_NOTECT) { 697 free(ip6af, M_FRAG6); 698 goto dropfrag; 699 } 700 if (ecn0 != IPTOS_ECN_CE) 701 q6->ip6q_ecn = IPTOS_ECN_CE; 702 } 703 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) { 704 free(ip6af, M_FRAG6); 705 goto dropfrag; 706 } 707 708 /* Find a fragmented part which begins after this one does. */ 709 TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) 710 if (af6->ip6af_off > ip6af->ip6af_off) 711 break; 712 713 /* 714 * If the incoming framgent overlaps some existing fragments in 715 * the reassembly queue, drop both the new fragment and the 716 * entire reassembly queue. However, if the new fragment 717 * is an exact duplicate of an existing fragment, only silently 718 * drop the existing fragment and leave the fragmentation queue 719 * unchanged, as allowed by the RFC. (RFC 8200, 4.5) 720 */ 721 if (af6 != NULL) 722 af6tmp = TAILQ_PREV(af6, ip6fraghead, ip6af_tq); 723 else 724 af6tmp = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 725 if (af6tmp != NULL) { 726 if (af6tmp->ip6af_off + af6tmp->ip6af_frglen - 727 ip6af->ip6af_off > 0) { 728 if (af6tmp->ip6af_off != ip6af->ip6af_off || 729 af6tmp->ip6af_frglen != ip6af->ip6af_frglen) 730 frag6_freef(q6, bucket); 731 free(ip6af, M_FRAG6); 732 goto dropfrag; 733 } 734 } 735 if (af6 != NULL) { 736 if (ip6af->ip6af_off + ip6af->ip6af_frglen - 737 af6->ip6af_off > 0) { 738 if (af6->ip6af_off != ip6af->ip6af_off || 739 af6->ip6af_frglen != ip6af->ip6af_frglen) 740 frag6_freef(q6, bucket); 741 free(ip6af, M_FRAG6); 742 goto dropfrag; 743 } 744 } 745 746 #ifdef MAC 747 mac_ip6q_update(m, q6); 748 #endif 749 750 /* 751 * Stick new segment in its place; check for complete reassembly. 752 * If not complete, check fragment limit. Move to front of packet 753 * queue, as we are the most recently active fragmented packet. 754 */ 755 if (af6 != NULL) 756 TAILQ_INSERT_BEFORE(af6, ip6af, ip6af_tq); 757 else 758 TAILQ_INSERT_TAIL(&q6->ip6q_frags, ip6af, ip6af_tq); 759 postinsert: 760 atomic_add_int(&frag6_nfrags, 1); 761 q6->ip6q_nfrag++; 762 763 plen = 0; 764 TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 765 if (af6->ip6af_off != plen) { 766 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 767 IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 768 frag6_freef(q6, bucket); 769 } 770 IP6QB_UNLOCK(bucket); 771 *mp = NULL; 772 return (IPPROTO_DONE); 773 } 774 plen += af6->ip6af_frglen; 775 } 776 af6 = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 777 if (af6->ip6af_mff) { 778 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 779 IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 780 frag6_freef(q6, bucket); 781 } 782 IP6QB_UNLOCK(bucket); 783 *mp = NULL; 784 return (IPPROTO_DONE); 785 } 786 787 /* Reassembly is complete; concatenate fragments. */ 788 ip6af = TAILQ_FIRST(&q6->ip6q_frags); 789 t = m = ip6af->ip6af_m; 790 TAILQ_REMOVE(&q6->ip6q_frags, ip6af, ip6af_tq); 791 while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 792 m->m_pkthdr.csum_flags &= 793 af6->ip6af_m->m_pkthdr.csum_flags; 794 m->m_pkthdr.csum_data += 795 af6->ip6af_m->m_pkthdr.csum_data; 796 797 TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 798 t = m_last(t); 799 m_adj(af6->ip6af_m, af6->ip6af_offset); 800 m_demote_pkthdr(af6->ip6af_m); 801 m_cat(t, af6->ip6af_m); 802 free(af6, M_FRAG6); 803 } 804 805 while (m->m_pkthdr.csum_data & 0xffff0000) 806 m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 807 (m->m_pkthdr.csum_data >> 16); 808 809 /* Adjust offset to point where the original next header starts. */ 810 offset = ip6af->ip6af_offset - sizeof(struct ip6_frag); 811 free(ip6af, M_FRAG6); 812 ip6 = mtod(m, struct ip6_hdr *); 813 ip6->ip6_plen = htons((u_short)plen + offset - sizeof(struct ip6_hdr)); 814 if (q6->ip6q_ecn == IPTOS_ECN_CE) 815 ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20); 816 nxt = q6->ip6q_nxt; 817 818 TAILQ_REMOVE(head, q6, ip6q_tq); 819 V_ip6qb[bucket].count--; 820 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 821 822 ip6_deletefraghdr(m, offset, M_NOWAIT); 823 824 /* Set nxt(-hdr field value) to the original value. */ 825 m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), 826 (caddr_t)&nxt); 827 828 #ifdef MAC 829 mac_ip6q_reassemble(q6, m); 830 mac_ip6q_destroy(q6); 831 #endif 832 free(q6, M_FRAG6); 833 atomic_subtract_int(&V_frag6_nfragpackets, 1); 834 835 if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */ 836 837 plen = 0; 838 for (t = m; t; t = t->m_next) 839 plen += t->m_len; 840 m->m_pkthdr.len = plen; 841 /* Set a valid receive interface pointer. */ 842 m->m_pkthdr.rcvif = srcifp; 843 } 844 845 #ifdef RSS 846 mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc), 847 M_NOWAIT); 848 if (mtag == NULL) 849 goto dropfrag; 850 851 ip6dc = (struct ip6_direct_ctx *)(mtag + 1); 852 ip6dc->ip6dc_nxt = nxt; 853 ip6dc->ip6dc_off = offset; 854 855 m_tag_prepend(m, mtag); 856 #endif 857 858 IP6QB_UNLOCK(bucket); 859 IP6STAT_INC(ip6s_reassembled); 860 in6_ifstat_inc(dstifp, ifs6_reass_ok); 861 862 #ifdef RSS 863 /* Queue/dispatch for reprocessing. */ 864 netisr_dispatch(NETISR_IPV6_DIRECT, m); 865 *mp = NULL; 866 return (IPPROTO_DONE); 867 #endif 868 869 /* Tell launch routine the next header. */ 870 *mp = m; 871 *offp = offset; 872 873 return (nxt); 874 875 dropfrag: 876 IP6QB_UNLOCK(bucket); 877 dropfrag2: 878 in6_ifstat_inc(dstifp, ifs6_reass_fail); 879 IP6STAT_INC(ip6s_fragdropped); 880 m_freem(m); 881 *mp = NULL; 882 return (IPPROTO_DONE); 883 } 884 885 /* 886 * IPv6 reassembling timer processing; 887 * if a timer expires on a reassembly queue, discard it. 888 */ 889 void 890 frag6_slowtimo(void) 891 { 892 VNET_ITERATOR_DECL(vnet_iter); 893 struct ip6qhead *head; 894 struct ip6q *q6, *q6tmp; 895 uint32_t bucket; 896 897 VNET_LIST_RLOCK_NOSLEEP(); 898 VNET_FOREACH(vnet_iter) { 899 CURVNET_SET(vnet_iter); 900 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 901 IP6QB_LOCK(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, 906 q6->ip6q_nfrag); 907 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 908 frag6_freef(q6, bucket); 909 } 910 /* 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 917 * is negative. 918 */ 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); 926 } 927 IP6QB_UNLOCK(bucket); 928 } 929 /* 930 * If we are still over the maximum number of fragmented 931 * packets, drain off enough to get down to the new limit. 932 */ 933 bucket = 0; 934 while (V_ip6_maxfragpackets >= 0 && 935 atomic_load_int(&V_frag6_nfragpackets) > 936 (u_int)V_ip6_maxfragpackets) { 937 IP6QB_LOCK(bucket); 938 q6 = TAILQ_LAST(IP6QB_HEAD(bucket), ip6qhead); 939 if (q6 != NULL) { 940 IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag); 941 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 942 frag6_freef(q6, bucket); 943 } 944 IP6QB_UNLOCK(bucket); 945 bucket = (bucket + 1) % IP6REASS_NHASH; 946 } 947 CURVNET_RESTORE(); 948 } 949 VNET_LIST_RUNLOCK_NOSLEEP(); 950 } 951 952 /* 953 * Eventhandler to adjust limits in case nmbclusters change. 954 */ 955 static void 956 frag6_change(void *tag) 957 { 958 VNET_ITERATOR_DECL(vnet_iter); 959 960 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(); 966 CURVNET_RESTORE(); 967 } 968 VNET_LIST_RUNLOCK_NOSLEEP(); 969 } 970 971 /* 972 * Initialise reassembly queue and fragment identifier. 973 */ 974 void 975 frag6_init(void) 976 { 977 uint32_t bucket; 978 979 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; 985 } 986 V_ip6qb_hashseed = arc4random(); 987 V_ip6_maxfragsperpacket = 64; 988 #ifdef VIMAGE 989 V_frag6_on = true; 990 #endif 991 if (!IS_DEFAULT_VNET(curvnet)) 992 return; 993 994 ip6_maxfrags = IP6_MAXFRAGS; 995 EVENTHANDLER_REGISTER(nmbclusters_change, 996 frag6_change, NULL, EVENTHANDLER_PRI_ANY); 997 } 998 999 /* 1000 * Drain off all datagram fragments. 1001 */ 1002 static void 1003 frag6_drain_one(void) 1004 { 1005 struct ip6q *q6; 1006 uint32_t bucket; 1007 1008 for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 1009 IP6QB_LOCK(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); 1014 } 1015 IP6QB_UNLOCK(bucket); 1016 } 1017 } 1018 1019 void 1020 frag6_drain(void) 1021 { 1022 VNET_ITERATOR_DECL(vnet_iter); 1023 1024 VNET_LIST_RLOCK_NOSLEEP(); 1025 VNET_FOREACH(vnet_iter) { 1026 CURVNET_SET(vnet_iter); 1027 frag6_drain_one(); 1028 CURVNET_RESTORE(); 1029 } 1030 VNET_LIST_RUNLOCK_NOSLEEP(); 1031 } 1032 1033 #ifdef VIMAGE 1034 /* 1035 * Clear up IPv6 reassembly structures. 1036 */ 1037 void 1038 frag6_destroy(void) 1039 { 1040 uint32_t bucket; 1041 1042 frag6_drain_one(); 1043 V_frag6_on = false; 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); 1049 } 1050 } 1051 #endif 1052