1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * $KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $ 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_rss.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/hash.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/domain.h> 45 #include <sys/eventhandler.h> 46 #include <sys/protosw.h> 47 #include <sys/socket.h> 48 #include <sys/errno.h> 49 #include <sys/time.h> 50 #include <sys/kernel.h> 51 #include <sys/syslog.h> 52 53 #include <machine/atomic.h> 54 55 #include <net/if.h> 56 #include <net/if_var.h> 57 #include <net/netisr.h> 58 #include <net/route.h> 59 #include <net/vnet.h> 60 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 */ 68 69 #include <security/mac/mac_framework.h> 70 71 /* 72 * Reassembly headers are stored in hash buckets. 73 */ 74 #define IP6REASS_NHASH_LOG2 10 75 #define IP6REASS_NHASH (1 << IP6REASS_NHASH_LOG2) 76 #define IP6REASS_HMASK (IP6REASS_NHASH - 1) 77 78 static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *, 79 uint32_t bucket __unused); 80 static void frag6_deq(struct ip6asfrag *, uint32_t bucket __unused); 81 static void frag6_insque_head(struct ip6q *, struct ip6q *, 82 uint32_t bucket); 83 static void frag6_remque(struct ip6q *, uint32_t bucket); 84 static void frag6_freef(struct ip6q *, uint32_t bucket); 85 86 struct ip6qbucket { 87 struct ip6q ip6q; 88 struct mtx lock; 89 int count; 90 }; 91 92 VNET_DEFINE_STATIC(volatile u_int, frag6_nfragpackets); 93 volatile u_int frag6_nfrags = 0; 94 VNET_DEFINE_STATIC(struct ip6qbucket, ip6q[IP6REASS_NHASH]); 95 VNET_DEFINE_STATIC(uint32_t, ip6q_hashseed); 96 97 #define V_frag6_nfragpackets VNET(frag6_nfragpackets) 98 #define V_ip6q VNET(ip6q) 99 #define V_ip6q_hashseed VNET(ip6q_hashseed) 100 101 #define IP6Q_LOCK(i) mtx_lock(&V_ip6q[(i)].lock) 102 #define IP6Q_TRYLOCK(i) mtx_trylock(&V_ip6q[(i)].lock) 103 #define IP6Q_LOCK_ASSERT(i) mtx_assert(&V_ip6q[(i)].lock, MA_OWNED) 104 #define IP6Q_UNLOCK(i) mtx_unlock(&V_ip6q[(i)].lock) 105 #define IP6Q_HEAD(i) (&V_ip6q[(i)].ip6q) 106 107 static MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header"); 108 109 /* 110 * By default, limit the number of IP6 fragments across all reassembly 111 * queues to 1/32 of the total number of mbuf clusters. 112 * 113 * Limit the total number of reassembly queues per VNET to the 114 * IP6 fragment limit, but ensure the limit will not allow any bucket 115 * to grow above 100 items. (The bucket limit is 116 * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct 117 * multiplier to reach a 100-item limit.) 118 * The 100-item limit was chosen as brief testing seems to show that 119 * this produces "reasonable" performance on some subset of systems 120 * under DoS attack. 121 */ 122 #define IP6_MAXFRAGS (nmbclusters / 32) 123 #define IP6_MAXFRAGPACKETS (imin(IP6_MAXFRAGS, IP6REASS_NHASH * 50)) 124 125 /* 126 * Initialise reassembly queue and fragment identifier. 127 */ 128 void 129 frag6_set_bucketsize() 130 { 131 int i; 132 133 if ((i = V_ip6_maxfragpackets) > 0) 134 V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1); 135 } 136 137 static void 138 frag6_change(void *tag) 139 { 140 VNET_ITERATOR_DECL(vnet_iter); 141 142 ip6_maxfrags = IP6_MAXFRAGS; 143 VNET_LIST_RLOCK_NOSLEEP(); 144 VNET_FOREACH(vnet_iter) { 145 CURVNET_SET(vnet_iter); 146 V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 147 frag6_set_bucketsize(); 148 CURVNET_RESTORE(); 149 } 150 VNET_LIST_RUNLOCK_NOSLEEP(); 151 } 152 153 void 154 frag6_init(void) 155 { 156 struct ip6q *q6; 157 int i; 158 159 V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 160 frag6_set_bucketsize(); 161 for (i = 0; i < IP6REASS_NHASH; i++) { 162 q6 = IP6Q_HEAD(i); 163 q6->ip6q_next = q6->ip6q_prev = q6; 164 mtx_init(&V_ip6q[i].lock, "ip6qlock", NULL, MTX_DEF); 165 V_ip6q[i].count = 0; 166 } 167 V_ip6q_hashseed = arc4random(); 168 V_ip6_maxfragsperpacket = 64; 169 if (!IS_DEFAULT_VNET(curvnet)) 170 return; 171 172 ip6_maxfrags = IP6_MAXFRAGS; 173 EVENTHANDLER_REGISTER(nmbclusters_change, 174 frag6_change, NULL, EVENTHANDLER_PRI_ANY); 175 } 176 177 /* 178 * In RFC2460, fragment and reassembly rule do not agree with each other, 179 * in terms of next header field handling in fragment header. 180 * While the sender will use the same value for all of the fragmented packets, 181 * receiver is suggested not to check the consistency. 182 * 183 * fragment rule (p20): 184 * (2) A Fragment header containing: 185 * The Next Header value that identifies the first header of 186 * the Fragmentable Part of the original packet. 187 * -> next header field is same for all fragments 188 * 189 * reassembly rule (p21): 190 * The Next Header field of the last header of the Unfragmentable 191 * Part is obtained from the Next Header field of the first 192 * fragment's Fragment header. 193 * -> should grab it from the first fragment only 194 * 195 * The following note also contradicts with fragment rule - no one is going to 196 * send different fragment with different next header field. 197 * 198 * additional note (p22): 199 * The Next Header values in the Fragment headers of different 200 * fragments of the same original packet may differ. Only the value 201 * from the Offset zero fragment packet is used for reassembly. 202 * -> should grab it from the first fragment only 203 * 204 * There is no explicit reason given in the RFC. Historical reason maybe? 205 */ 206 /* 207 * Fragment input 208 */ 209 int 210 frag6_input(struct mbuf **mp, int *offp, int proto) 211 { 212 struct mbuf *m = *mp, *t; 213 struct ip6_hdr *ip6; 214 struct ip6_frag *ip6f; 215 struct ip6q *head, *q6; 216 struct ip6asfrag *af6, *ip6af, *af6dwn; 217 struct in6_ifaddr *ia; 218 int offset = *offp, nxt, i, next; 219 int first_frag = 0; 220 int fragoff, frgpartlen; /* must be larger than u_int16_t */ 221 uint32_t hash, hashkey[sizeof(struct in6_addr) * 2 + 1], *hashkeyp; 222 struct ifnet *dstifp; 223 u_int8_t ecn, ecn0; 224 #ifdef RSS 225 struct m_tag *mtag; 226 struct ip6_direct_ctx *ip6dc; 227 #endif 228 229 #if 0 230 char ip6buf[INET6_ADDRSTRLEN]; 231 #endif 232 233 ip6 = mtod(m, struct ip6_hdr *); 234 #ifndef PULLDOWN_TEST 235 IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE); 236 ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset); 237 #else 238 IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f)); 239 if (ip6f == NULL) 240 return (IPPROTO_DONE); 241 #endif 242 243 dstifp = NULL; 244 /* find the destination interface of the packet. */ 245 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 246 if (ia != NULL) { 247 dstifp = ia->ia_ifp; 248 ifa_free(&ia->ia_ifa); 249 } 250 /* jumbo payload can't contain a fragment header */ 251 if (ip6->ip6_plen == 0) { 252 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset); 253 in6_ifstat_inc(dstifp, ifs6_reass_fail); 254 return IPPROTO_DONE; 255 } 256 257 /* 258 * check whether fragment packet's fragment length is 259 * multiple of 8 octets. 260 * sizeof(struct ip6_frag) == 8 261 * sizeof(struct ip6_hdr) = 40 262 */ 263 if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) && 264 (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) { 265 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 266 offsetof(struct ip6_hdr, ip6_plen)); 267 in6_ifstat_inc(dstifp, ifs6_reass_fail); 268 return IPPROTO_DONE; 269 } 270 271 IP6STAT_INC(ip6s_fragments); 272 in6_ifstat_inc(dstifp, ifs6_reass_reqd); 273 274 /* offset now points to data portion */ 275 offset += sizeof(struct ip6_frag); 276 277 /* 278 * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0) 279 * upfront, unrelated to any reassembly. Just skip the fragment header. 280 */ 281 if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) { 282 /* XXX-BZ we want dedicated counters for this. */ 283 IP6STAT_INC(ip6s_reassembled); 284 in6_ifstat_inc(dstifp, ifs6_reass_ok); 285 *offp = offset; 286 m->m_flags |= M_FRAGMENTED; 287 return (ip6f->ip6f_nxt); 288 } 289 290 /* Get fragment length and discard 0-byte fragments. */ 291 frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset; 292 if (frgpartlen == 0) { 293 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 294 offsetof(struct ip6_hdr, ip6_plen)); 295 in6_ifstat_inc(dstifp, ifs6_reass_fail); 296 IP6STAT_INC(ip6s_fragdropped); 297 return IPPROTO_DONE; 298 } 299 300 hashkeyp = hashkey; 301 memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr)); 302 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 303 memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr)); 304 hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 305 *hashkeyp = ip6f->ip6f_ident; 306 hash = jenkins_hash32(hashkey, nitems(hashkey), V_ip6q_hashseed); 307 hash &= IP6REASS_HMASK; 308 head = IP6Q_HEAD(hash); 309 IP6Q_LOCK(hash); 310 311 /* 312 * Enforce upper bound on number of fragments. 313 * If maxfrag is 0, never accept fragments. 314 * If maxfrag is -1, accept all fragments without limitation. 315 */ 316 if (ip6_maxfrags < 0) 317 ; 318 else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags) 319 goto dropfrag; 320 321 for (q6 = head->ip6q_next; q6 != head; q6 = q6->ip6q_next) 322 if (ip6f->ip6f_ident == q6->ip6q_ident && 323 IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) && 324 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst) 325 #ifdef MAC 326 && mac_ip6q_match(m, q6) 327 #endif 328 ) 329 break; 330 331 if (q6 == head) { 332 /* 333 * the first fragment to arrive, create a reassembly queue. 334 */ 335 first_frag = 1; 336 337 /* 338 * Enforce upper bound on number of fragmented packets 339 * for which we attempt reassembly; 340 * If maxfragpackets is 0, never accept fragments. 341 * If maxfragpackets is -1, accept all fragments without 342 * limitation. 343 */ 344 if (V_ip6_maxfragpackets < 0) 345 ; 346 else if (V_ip6q[hash].count >= V_ip6_maxfragbucketsize || 347 atomic_load_int(&V_frag6_nfragpackets) >= 348 (u_int)V_ip6_maxfragpackets) 349 goto dropfrag; 350 atomic_add_int(&V_frag6_nfragpackets, 1); 351 q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE, 352 M_NOWAIT); 353 if (q6 == NULL) 354 goto dropfrag; 355 bzero(q6, sizeof(*q6)); 356 #ifdef MAC 357 if (mac_ip6q_init(q6, M_NOWAIT) != 0) { 358 free(q6, M_FTABLE); 359 goto dropfrag; 360 } 361 mac_ip6q_create(m, q6); 362 #endif 363 frag6_insque_head(q6, head, hash); 364 365 /* ip6q_nxt will be filled afterwards, from 1st fragment */ 366 q6->ip6q_down = q6->ip6q_up = (struct ip6asfrag *)q6; 367 #ifdef notyet 368 q6->ip6q_nxtp = (u_char *)nxtp; 369 #endif 370 q6->ip6q_ident = ip6f->ip6f_ident; 371 q6->ip6q_ttl = IPV6_FRAGTTL; 372 q6->ip6q_src = ip6->ip6_src; 373 q6->ip6q_dst = ip6->ip6_dst; 374 q6->ip6q_ecn = 375 (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 376 q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */ 377 378 q6->ip6q_nfrag = 0; 379 } 380 381 /* 382 * If it's the 1st fragment, record the length of the 383 * unfragmentable part and the next header of the fragment header. 384 */ 385 fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK); 386 if (fragoff == 0) { 387 q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) - 388 sizeof(struct ip6_frag); 389 q6->ip6q_nxt = ip6f->ip6f_nxt; 390 } 391 392 /* 393 * Check that the reassembled packet would not exceed 65535 bytes 394 * in size. 395 * If it would exceed, discard the fragment and return an ICMP error. 396 */ 397 if (q6->ip6q_unfrglen >= 0) { 398 /* The 1st fragment has already arrived. */ 399 if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) { 400 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 401 offset - sizeof(struct ip6_frag) + 402 offsetof(struct ip6_frag, ip6f_offlg)); 403 IP6Q_UNLOCK(hash); 404 return (IPPROTO_DONE); 405 } 406 } else if (fragoff + frgpartlen > IPV6_MAXPACKET) { 407 icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 408 offset - sizeof(struct ip6_frag) + 409 offsetof(struct ip6_frag, ip6f_offlg)); 410 IP6Q_UNLOCK(hash); 411 return (IPPROTO_DONE); 412 } 413 /* 414 * If it's the first fragment, do the above check for each 415 * fragment already stored in the reassembly queue. 416 */ 417 if (fragoff == 0) { 418 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; 419 af6 = af6dwn) { 420 af6dwn = af6->ip6af_down; 421 422 if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen > 423 IPV6_MAXPACKET) { 424 struct mbuf *merr = IP6_REASS_MBUF(af6); 425 struct ip6_hdr *ip6err; 426 int erroff = af6->ip6af_offset; 427 428 /* dequeue the fragment. */ 429 frag6_deq(af6, hash); 430 free(af6, M_FTABLE); 431 432 /* adjust pointer. */ 433 ip6err = mtod(merr, struct ip6_hdr *); 434 435 /* 436 * Restore source and destination addresses 437 * in the erroneous IPv6 header. 438 */ 439 ip6err->ip6_src = q6->ip6q_src; 440 ip6err->ip6_dst = q6->ip6q_dst; 441 442 icmp6_error(merr, ICMP6_PARAM_PROB, 443 ICMP6_PARAMPROB_HEADER, 444 erroff - sizeof(struct ip6_frag) + 445 offsetof(struct ip6_frag, ip6f_offlg)); 446 } 447 } 448 } 449 450 ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE, 451 M_NOWAIT); 452 if (ip6af == NULL) 453 goto dropfrag; 454 bzero(ip6af, sizeof(*ip6af)); 455 ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG; 456 ip6af->ip6af_off = fragoff; 457 ip6af->ip6af_frglen = frgpartlen; 458 ip6af->ip6af_offset = offset; 459 IP6_REASS_MBUF(ip6af) = m; 460 461 if (first_frag) { 462 af6 = (struct ip6asfrag *)q6; 463 goto insert; 464 } 465 466 /* 467 * Handle ECN by comparing this segment with the first one; 468 * if CE is set, do not lose CE. 469 * drop if CE and not-ECT are mixed for the same packet. 470 */ 471 ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 472 ecn0 = q6->ip6q_ecn; 473 if (ecn == IPTOS_ECN_CE) { 474 if (ecn0 == IPTOS_ECN_NOTECT) { 475 free(ip6af, M_FTABLE); 476 goto dropfrag; 477 } 478 if (ecn0 != IPTOS_ECN_CE) 479 q6->ip6q_ecn = IPTOS_ECN_CE; 480 } 481 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) { 482 free(ip6af, M_FTABLE); 483 goto dropfrag; 484 } 485 486 /* 487 * Find a segment which begins after this one does. 488 */ 489 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; 490 af6 = af6->ip6af_down) 491 if (af6->ip6af_off > ip6af->ip6af_off) 492 break; 493 494 #if 0 495 /* 496 * If there is a preceding segment, it may provide some of 497 * our data already. If so, drop the data from the incoming 498 * segment. If it provides all of our data, drop us. 499 */ 500 if (af6->ip6af_up != (struct ip6asfrag *)q6) { 501 i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen 502 - ip6af->ip6af_off; 503 if (i > 0) { 504 if (i >= ip6af->ip6af_frglen) 505 goto dropfrag; 506 m_adj(IP6_REASS_MBUF(ip6af), i); 507 ip6af->ip6af_off += i; 508 ip6af->ip6af_frglen -= i; 509 } 510 } 511 512 /* 513 * While we overlap succeeding segments trim them or, 514 * if they are completely covered, dequeue them. 515 */ 516 while (af6 != (struct ip6asfrag *)q6 && 517 ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) { 518 i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off; 519 if (i < af6->ip6af_frglen) { 520 af6->ip6af_frglen -= i; 521 af6->ip6af_off += i; 522 m_adj(IP6_REASS_MBUF(af6), i); 523 break; 524 } 525 af6 = af6->ip6af_down; 526 m_freem(IP6_REASS_MBUF(af6->ip6af_up)); 527 frag6_deq(af6->ip6af_up, hash); 528 } 529 #else 530 /* 531 * If the incoming framgent overlaps some existing fragments in 532 * the reassembly queue, drop it, since it is dangerous to override 533 * existing fragments from a security point of view. 534 * We don't know which fragment is the bad guy - here we trust 535 * fragment that came in earlier, with no real reason. 536 * 537 * Note: due to changes after disabling this part, mbuf passed to 538 * m_adj() below now does not meet the requirement. 539 */ 540 if (af6->ip6af_up != (struct ip6asfrag *)q6) { 541 i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen 542 - ip6af->ip6af_off; 543 if (i > 0) { 544 #if 0 /* suppress the noisy log */ 545 log(LOG_ERR, "%d bytes of a fragment from %s " 546 "overlaps the previous fragment\n", 547 i, ip6_sprintf(ip6buf, &q6->ip6q_src)); 548 #endif 549 free(ip6af, M_FTABLE); 550 goto dropfrag; 551 } 552 } 553 if (af6 != (struct ip6asfrag *)q6) { 554 i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off; 555 if (i > 0) { 556 #if 0 /* suppress the noisy log */ 557 log(LOG_ERR, "%d bytes of a fragment from %s " 558 "overlaps the succeeding fragment", 559 i, ip6_sprintf(ip6buf, &q6->ip6q_src)); 560 #endif 561 free(ip6af, M_FTABLE); 562 goto dropfrag; 563 } 564 } 565 #endif 566 567 insert: 568 #ifdef MAC 569 if (!first_frag) 570 mac_ip6q_update(m, q6); 571 #endif 572 573 /* 574 * Stick new segment in its place; 575 * check for complete reassembly. 576 * If not complete, check fragment limit. 577 * Move to front of packet queue, as we are 578 * the most recently active fragmented packet. 579 */ 580 frag6_enq(ip6af, af6->ip6af_up, hash); 581 atomic_add_int(&frag6_nfrags, 1); 582 q6->ip6q_nfrag++; 583 #if 0 /* xxx */ 584 if (q6 != head->ip6q_next) { 585 frag6_remque(q6, hash); 586 frag6_insque_head(q6, head, hash); 587 } 588 #endif 589 next = 0; 590 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; 591 af6 = af6->ip6af_down) { 592 if (af6->ip6af_off != next) { 593 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 594 IP6STAT_INC(ip6s_fragdropped); 595 frag6_freef(q6, hash); 596 } 597 IP6Q_UNLOCK(hash); 598 return IPPROTO_DONE; 599 } 600 next += af6->ip6af_frglen; 601 } 602 if (af6->ip6af_up->ip6af_mff) { 603 if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 604 IP6STAT_INC(ip6s_fragdropped); 605 frag6_freef(q6, hash); 606 } 607 IP6Q_UNLOCK(hash); 608 return IPPROTO_DONE; 609 } 610 611 /* 612 * Reassembly is complete; concatenate fragments. 613 */ 614 ip6af = q6->ip6q_down; 615 t = m = IP6_REASS_MBUF(ip6af); 616 af6 = ip6af->ip6af_down; 617 frag6_deq(ip6af, hash); 618 while (af6 != (struct ip6asfrag *)q6) { 619 m->m_pkthdr.csum_flags &= 620 IP6_REASS_MBUF(af6)->m_pkthdr.csum_flags; 621 m->m_pkthdr.csum_data += 622 IP6_REASS_MBUF(af6)->m_pkthdr.csum_data; 623 624 af6dwn = af6->ip6af_down; 625 frag6_deq(af6, hash); 626 while (t->m_next) 627 t = t->m_next; 628 m_adj(IP6_REASS_MBUF(af6), af6->ip6af_offset); 629 m_demote_pkthdr(IP6_REASS_MBUF(af6)); 630 m_cat(t, IP6_REASS_MBUF(af6)); 631 free(af6, M_FTABLE); 632 af6 = af6dwn; 633 } 634 635 while (m->m_pkthdr.csum_data & 0xffff0000) 636 m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 637 (m->m_pkthdr.csum_data >> 16); 638 639 /* adjust offset to point where the original next header starts */ 640 offset = ip6af->ip6af_offset - sizeof(struct ip6_frag); 641 free(ip6af, M_FTABLE); 642 ip6 = mtod(m, struct ip6_hdr *); 643 ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr)); 644 if (q6->ip6q_ecn == IPTOS_ECN_CE) 645 ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20); 646 nxt = q6->ip6q_nxt; 647 #ifdef notyet 648 *q6->ip6q_nxtp = (u_char)(nxt & 0xff); 649 #endif 650 651 if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) { 652 frag6_remque(q6, hash); 653 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 654 #ifdef MAC 655 mac_ip6q_destroy(q6); 656 #endif 657 free(q6, M_FTABLE); 658 atomic_subtract_int(&V_frag6_nfragpackets, 1); 659 660 goto dropfrag; 661 } 662 663 /* 664 * Store NXT to the original. 665 */ 666 m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), 667 (caddr_t)&nxt); 668 669 frag6_remque(q6, hash); 670 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 671 #ifdef MAC 672 mac_ip6q_reassemble(q6, m); 673 mac_ip6q_destroy(q6); 674 #endif 675 free(q6, M_FTABLE); 676 atomic_subtract_int(&V_frag6_nfragpackets, 1); 677 678 if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */ 679 int plen = 0; 680 for (t = m; t; t = t->m_next) 681 plen += t->m_len; 682 m->m_pkthdr.len = plen; 683 } 684 685 #ifdef RSS 686 mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc), 687 M_NOWAIT); 688 if (mtag == NULL) 689 goto dropfrag; 690 691 ip6dc = (struct ip6_direct_ctx *)(mtag + 1); 692 ip6dc->ip6dc_nxt = nxt; 693 ip6dc->ip6dc_off = offset; 694 695 m_tag_prepend(m, mtag); 696 #endif 697 698 IP6Q_UNLOCK(hash); 699 IP6STAT_INC(ip6s_reassembled); 700 in6_ifstat_inc(dstifp, ifs6_reass_ok); 701 702 #ifdef RSS 703 /* 704 * Queue/dispatch for reprocessing. 705 */ 706 netisr_dispatch(NETISR_IPV6_DIRECT, m); 707 return IPPROTO_DONE; 708 #endif 709 710 /* 711 * Tell launch routine the next header 712 */ 713 714 *mp = m; 715 *offp = offset; 716 717 return nxt; 718 719 dropfrag: 720 IP6Q_UNLOCK(hash); 721 in6_ifstat_inc(dstifp, ifs6_reass_fail); 722 IP6STAT_INC(ip6s_fragdropped); 723 m_freem(m); 724 return IPPROTO_DONE; 725 } 726 727 /* 728 * Free a fragment reassembly header and all 729 * associated datagrams. 730 */ 731 static void 732 frag6_freef(struct ip6q *q6, uint32_t bucket) 733 { 734 struct ip6asfrag *af6, *down6; 735 736 IP6Q_LOCK_ASSERT(bucket); 737 738 for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; 739 af6 = down6) { 740 struct mbuf *m = IP6_REASS_MBUF(af6); 741 742 down6 = af6->ip6af_down; 743 frag6_deq(af6, bucket); 744 745 /* 746 * Return ICMP time exceeded error for the 1st fragment. 747 * Just free other fragments. 748 */ 749 if (af6->ip6af_off == 0) { 750 struct ip6_hdr *ip6; 751 752 /* adjust pointer */ 753 ip6 = mtod(m, struct ip6_hdr *); 754 755 /* restore source and destination addresses */ 756 ip6->ip6_src = q6->ip6q_src; 757 ip6->ip6_dst = q6->ip6q_dst; 758 759 icmp6_error(m, ICMP6_TIME_EXCEEDED, 760 ICMP6_TIME_EXCEED_REASSEMBLY, 0); 761 } else 762 m_freem(m); 763 free(af6, M_FTABLE); 764 } 765 frag6_remque(q6, bucket); 766 atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 767 #ifdef MAC 768 mac_ip6q_destroy(q6); 769 #endif 770 free(q6, M_FTABLE); 771 atomic_subtract_int(&V_frag6_nfragpackets, 1); 772 } 773 774 /* 775 * Put an ip fragment on a reassembly chain. 776 * Like insque, but pointers in middle of structure. 777 */ 778 static void 779 frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6, 780 uint32_t bucket __unused) 781 { 782 783 IP6Q_LOCK_ASSERT(bucket); 784 785 af6->ip6af_up = up6; 786 af6->ip6af_down = up6->ip6af_down; 787 up6->ip6af_down->ip6af_up = af6; 788 up6->ip6af_down = af6; 789 } 790 791 /* 792 * To frag6_enq as remque is to insque. 793 */ 794 static void 795 frag6_deq(struct ip6asfrag *af6, uint32_t bucket __unused) 796 { 797 798 IP6Q_LOCK_ASSERT(bucket); 799 800 af6->ip6af_up->ip6af_down = af6->ip6af_down; 801 af6->ip6af_down->ip6af_up = af6->ip6af_up; 802 } 803 804 static void 805 frag6_insque_head(struct ip6q *new, struct ip6q *old, uint32_t bucket) 806 { 807 808 IP6Q_LOCK_ASSERT(bucket); 809 KASSERT(IP6Q_HEAD(bucket) == old, 810 ("%s: attempt to insert at head of wrong bucket" 811 " (bucket=%u, old=%p)", __func__, bucket, old)); 812 813 new->ip6q_prev = old; 814 new->ip6q_next = old->ip6q_next; 815 old->ip6q_next->ip6q_prev= new; 816 old->ip6q_next = new; 817 V_ip6q[bucket].count++; 818 } 819 820 static void 821 frag6_remque(struct ip6q *p6, uint32_t bucket) 822 { 823 824 IP6Q_LOCK_ASSERT(bucket); 825 826 p6->ip6q_prev->ip6q_next = p6->ip6q_next; 827 p6->ip6q_next->ip6q_prev = p6->ip6q_prev; 828 V_ip6q[bucket].count--; 829 } 830 831 /* 832 * IPv6 reassembling timer processing; 833 * if a timer expires on a reassembly 834 * queue, discard it. 835 */ 836 void 837 frag6_slowtimo(void) 838 { 839 VNET_ITERATOR_DECL(vnet_iter); 840 struct ip6q *head, *q6; 841 int i; 842 843 VNET_LIST_RLOCK_NOSLEEP(); 844 VNET_FOREACH(vnet_iter) { 845 CURVNET_SET(vnet_iter); 846 for (i = 0; i < IP6REASS_NHASH; i++) { 847 IP6Q_LOCK(i); 848 head = IP6Q_HEAD(i); 849 q6 = head->ip6q_next; 850 if (q6 == NULL) { 851 /* 852 * XXXJTL: This should never happen. This 853 * should turn into an assertion. 854 */ 855 IP6Q_UNLOCK(i); 856 continue; 857 } 858 while (q6 != head) { 859 --q6->ip6q_ttl; 860 q6 = q6->ip6q_next; 861 if (q6->ip6q_prev->ip6q_ttl == 0) { 862 IP6STAT_INC(ip6s_fragtimeout); 863 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 864 frag6_freef(q6->ip6q_prev, i); 865 } 866 } 867 /* 868 * If we are over the maximum number of fragments 869 * (due to the limit being lowered), drain off 870 * enough to get down to the new limit. 871 * Note that we drain all reassembly queues if 872 * maxfragpackets is 0 (fragmentation is disabled), 873 * and don't enforce a limit when maxfragpackets 874 * is negative. 875 */ 876 while ((V_ip6_maxfragpackets == 0 || 877 (V_ip6_maxfragpackets > 0 && 878 V_ip6q[i].count > V_ip6_maxfragbucketsize)) && 879 head->ip6q_prev != head) { 880 IP6STAT_INC(ip6s_fragoverflow); 881 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 882 frag6_freef(head->ip6q_prev, i); 883 } 884 IP6Q_UNLOCK(i); 885 } 886 /* 887 * If we are still over the maximum number of fragmented 888 * packets, drain off enough to get down to the new limit. 889 */ 890 i = 0; 891 while (V_ip6_maxfragpackets >= 0 && 892 atomic_load_int(&V_frag6_nfragpackets) > 893 (u_int)V_ip6_maxfragpackets) { 894 IP6Q_LOCK(i); 895 head = IP6Q_HEAD(i); 896 if (head->ip6q_prev != head) { 897 IP6STAT_INC(ip6s_fragoverflow); 898 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 899 frag6_freef(head->ip6q_prev, i); 900 } 901 IP6Q_UNLOCK(i); 902 i = (i + 1) % IP6REASS_NHASH; 903 } 904 CURVNET_RESTORE(); 905 } 906 VNET_LIST_RUNLOCK_NOSLEEP(); 907 } 908 909 /* 910 * Drain off all datagram fragments. 911 */ 912 void 913 frag6_drain(void) 914 { 915 VNET_ITERATOR_DECL(vnet_iter); 916 struct ip6q *head; 917 int i; 918 919 VNET_LIST_RLOCK_NOSLEEP(); 920 VNET_FOREACH(vnet_iter) { 921 CURVNET_SET(vnet_iter); 922 for (i = 0; i < IP6REASS_NHASH; i++) { 923 if (IP6Q_TRYLOCK(i) == 0) 924 continue; 925 head = IP6Q_HEAD(i); 926 while (head->ip6q_next != head) { 927 IP6STAT_INC(ip6s_fragdropped); 928 /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 929 frag6_freef(head->ip6q_next, i); 930 } 931 IP6Q_UNLOCK(i); 932 } 933 CURVNET_RESTORE(); 934 } 935 VNET_LIST_RUNLOCK_NOSLEEP(); 936 } 937 938 int 939 ip6_deletefraghdr(struct mbuf *m, int offset, int wait) 940 { 941 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 942 struct mbuf *t; 943 944 /* Delete frag6 header. */ 945 if (m->m_len >= offset + sizeof(struct ip6_frag)) { 946 /* This is the only possible case with !PULLDOWN_TEST. */ 947 bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag), 948 offset); 949 m->m_data += sizeof(struct ip6_frag); 950 m->m_len -= sizeof(struct ip6_frag); 951 } else { 952 /* This comes with no copy if the boundary is on cluster. */ 953 if ((t = m_split(m, offset, wait)) == NULL) 954 return (ENOMEM); 955 m_adj(t, sizeof(struct ip6_frag)); 956 m_cat(m, t); 957 } 958 959 m->m_flags |= M_FRAGMENTED; 960 return (0); 961 } 962