1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * IPv6 fragment reassembly 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on: net/ipv4/ip_fragment.c 10 */ 11 12 /* 13 * Fixes: 14 * Andi Kleen Make it work with multiple hosts. 15 * More RFC compliance. 16 * 17 * Horst von Brand Add missing #include <linux/string.h> 18 * Alexey Kuznetsov SMP races, threading, cleanup. 19 * Patrick McHardy LRU queue of frag heads for evictor. 20 * Mitsuru KANDA @USAGI Register inet6_protocol{}. 21 * David Stevens and 22 * YOSHIFUJI,H. @USAGI Always remove fragment header to 23 * calculate ICV correctly. 24 */ 25 26 #define pr_fmt(fmt) "IPv6: " fmt 27 28 #include <linux/errno.h> 29 #include <linux/types.h> 30 #include <linux/string.h> 31 #include <linux/socket.h> 32 #include <linux/sockios.h> 33 #include <linux/jiffies.h> 34 #include <linux/net.h> 35 #include <linux/list.h> 36 #include <linux/netdevice.h> 37 #include <linux/in6.h> 38 #include <linux/ipv6.h> 39 #include <linux/icmpv6.h> 40 #include <linux/random.h> 41 #include <linux/jhash.h> 42 #include <linux/skbuff.h> 43 #include <linux/slab.h> 44 #include <linux/export.h> 45 #include <linux/tcp.h> 46 #include <linux/udp.h> 47 48 #include <net/sock.h> 49 #include <net/snmp.h> 50 51 #include <net/ipv6.h> 52 #include <net/ip6_route.h> 53 #include <net/protocol.h> 54 #include <net/transp_v6.h> 55 #include <net/rawv6.h> 56 #include <net/ndisc.h> 57 #include <net/addrconf.h> 58 #include <net/ipv6_frag.h> 59 #include <net/inet_ecn.h> 60 61 static const char ip6_frag_cache_name[] = "ip6-frags"; 62 ip6_frag_ecn(const struct ipv6hdr * ipv6h)63 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h) 64 { 65 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK); 66 } 67 68 static struct inet_frags ip6_frags; 69 70 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb, 71 struct sk_buff *prev_tail, struct net_device *dev); 72 ip6_frag_expire(struct timer_list * t)73 static void ip6_frag_expire(struct timer_list *t) 74 { 75 struct inet_frag_queue *frag = from_timer(frag, t, timer); 76 struct frag_queue *fq; 77 78 fq = container_of(frag, struct frag_queue, q); 79 80 ip6frag_expire_frag_queue(fq->q.fqdir->net, fq); 81 } 82 83 static struct frag_queue * fq_find(struct net * net,__be32 id,const struct ipv6hdr * hdr,int iif)84 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif) 85 { 86 struct frag_v6_compare_key key = { 87 .id = id, 88 .saddr = hdr->saddr, 89 .daddr = hdr->daddr, 90 .user = IP6_DEFRAG_LOCAL_DELIVER, 91 .iif = iif, 92 }; 93 struct inet_frag_queue *q; 94 95 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST | 96 IPV6_ADDR_LINKLOCAL))) 97 key.iif = 0; 98 99 q = inet_frag_find(net->ipv6.fqdir, &key); 100 if (!q) 101 return NULL; 102 103 return container_of(q, struct frag_queue, q); 104 } 105 ip6_frag_queue(struct frag_queue * fq,struct sk_buff * skb,struct frag_hdr * fhdr,int nhoff,u32 * prob_offset)106 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb, 107 struct frag_hdr *fhdr, int nhoff, 108 u32 *prob_offset) 109 { 110 struct net *net = dev_net(skb_dst(skb)->dev); 111 int offset, end, fragsize; 112 struct sk_buff *prev_tail; 113 struct net_device *dev; 114 int err = -ENOENT; 115 SKB_DR(reason); 116 u8 ecn; 117 118 /* If reassembly is already done, @skb must be a duplicate frag. */ 119 if (fq->q.flags & INET_FRAG_COMPLETE) { 120 SKB_DR_SET(reason, DUP_FRAG); 121 goto err; 122 } 123 124 err = -EINVAL; 125 offset = ntohs(fhdr->frag_off) & ~0x7; 126 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) - 127 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); 128 129 if ((unsigned int)end > IPV6_MAXPLEN) { 130 *prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb); 131 /* note that if prob_offset is set, the skb is freed elsewhere, 132 * we do not free it here. 133 */ 134 return -1; 135 } 136 137 ecn = ip6_frag_ecn(ipv6_hdr(skb)); 138 139 if (skb->ip_summed == CHECKSUM_COMPLETE) { 140 const unsigned char *nh = skb_network_header(skb); 141 skb->csum = csum_sub(skb->csum, 142 csum_partial(nh, (u8 *)(fhdr + 1) - nh, 143 0)); 144 } 145 146 /* Is this the final fragment? */ 147 if (!(fhdr->frag_off & htons(IP6_MF))) { 148 /* If we already have some bits beyond end 149 * or have different end, the segment is corrupted. 150 */ 151 if (end < fq->q.len || 152 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) 153 goto discard_fq; 154 fq->q.flags |= INET_FRAG_LAST_IN; 155 fq->q.len = end; 156 } else { 157 /* Check if the fragment is rounded to 8 bytes. 158 * Required by the RFC. 159 */ 160 if (end & 0x7) { 161 /* RFC2460 says always send parameter problem in 162 * this case. -DaveM 163 */ 164 *prob_offset = offsetof(struct ipv6hdr, payload_len); 165 return -1; 166 } 167 if (end > fq->q.len) { 168 /* Some bits beyond end -> corruption. */ 169 if (fq->q.flags & INET_FRAG_LAST_IN) 170 goto discard_fq; 171 fq->q.len = end; 172 } 173 } 174 175 if (end == offset) 176 goto discard_fq; 177 178 err = -ENOMEM; 179 /* Point into the IP datagram 'data' part. */ 180 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) 181 goto discard_fq; 182 183 err = pskb_trim_rcsum(skb, end - offset); 184 if (err) 185 goto discard_fq; 186 187 /* Note : skb->rbnode and skb->dev share the same location. */ 188 dev = skb->dev; 189 /* Makes sure compiler wont do silly aliasing games */ 190 barrier(); 191 192 prev_tail = fq->q.fragments_tail; 193 err = inet_frag_queue_insert(&fq->q, skb, offset, end); 194 if (err) 195 goto insert_error; 196 197 if (dev) 198 fq->iif = dev->ifindex; 199 200 fq->q.stamp = skb->tstamp; 201 fq->q.tstamp_type = skb->tstamp_type; 202 fq->q.meat += skb->len; 203 fq->ecn |= ecn; 204 add_frag_mem_limit(fq->q.fqdir, skb->truesize); 205 206 fragsize = -skb_network_offset(skb) + skb->len; 207 if (fragsize > fq->q.max_size) 208 fq->q.max_size = fragsize; 209 210 /* The first fragment. 211 * nhoffset is obtained from the first fragment, of course. 212 */ 213 if (offset == 0) { 214 fq->nhoffset = nhoff; 215 fq->q.flags |= INET_FRAG_FIRST_IN; 216 } 217 218 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && 219 fq->q.meat == fq->q.len) { 220 unsigned long orefdst = skb->_skb_refdst; 221 222 skb->_skb_refdst = 0UL; 223 err = ip6_frag_reasm(fq, skb, prev_tail, dev); 224 skb->_skb_refdst = orefdst; 225 return err; 226 } 227 228 skb_dst_drop(skb); 229 return -EINPROGRESS; 230 231 insert_error: 232 if (err == IPFRAG_DUP) { 233 SKB_DR_SET(reason, DUP_FRAG); 234 err = -EINVAL; 235 goto err; 236 } 237 err = -EINVAL; 238 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 239 IPSTATS_MIB_REASM_OVERLAPS); 240 discard_fq: 241 inet_frag_kill(&fq->q); 242 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 243 IPSTATS_MIB_REASMFAILS); 244 err: 245 kfree_skb_reason(skb, reason); 246 return err; 247 } 248 249 /* 250 * Check if this packet is complete. 251 * 252 * It is called with locked fq, and caller must check that 253 * queue is eligible for reassembly i.e. it is not COMPLETE, 254 * the last and the first frames arrived and all the bits are here. 255 */ ip6_frag_reasm(struct frag_queue * fq,struct sk_buff * skb,struct sk_buff * prev_tail,struct net_device * dev)256 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb, 257 struct sk_buff *prev_tail, struct net_device *dev) 258 { 259 struct net *net = fq->q.fqdir->net; 260 unsigned int nhoff; 261 void *reasm_data; 262 int payload_len; 263 u8 ecn; 264 265 inet_frag_kill(&fq->q); 266 267 ecn = ip_frag_ecn_table[fq->ecn]; 268 if (unlikely(ecn == 0xff)) 269 goto out_fail; 270 271 reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail); 272 if (!reasm_data) 273 goto out_oom; 274 275 payload_len = -skb_network_offset(skb) - 276 sizeof(struct ipv6hdr) + fq->q.len - 277 sizeof(struct frag_hdr); 278 if (payload_len > IPV6_MAXPLEN) 279 goto out_oversize; 280 281 /* We have to remove fragment header from datagram and to relocate 282 * header in order to calculate ICV correctly. */ 283 nhoff = fq->nhoffset; 284 skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0]; 285 memmove(skb->head + sizeof(struct frag_hdr), skb->head, 286 (skb->data - skb->head) - sizeof(struct frag_hdr)); 287 if (skb_mac_header_was_set(skb)) 288 skb->mac_header += sizeof(struct frag_hdr); 289 skb->network_header += sizeof(struct frag_hdr); 290 291 skb_reset_transport_header(skb); 292 293 inet_frag_reasm_finish(&fq->q, skb, reasm_data, true); 294 295 skb->dev = dev; 296 ipv6_hdr(skb)->payload_len = htons(payload_len); 297 ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn); 298 IP6CB(skb)->nhoff = nhoff; 299 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED; 300 IP6CB(skb)->frag_max_size = fq->q.max_size; 301 302 /* Yes, and fold redundant checksum back. 8) */ 303 skb_postpush_rcsum(skb, skb_network_header(skb), 304 skb_network_header_len(skb)); 305 306 rcu_read_lock(); 307 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMOKS); 308 rcu_read_unlock(); 309 fq->q.rb_fragments = RB_ROOT; 310 fq->q.fragments_tail = NULL; 311 fq->q.last_run_head = NULL; 312 return 1; 313 314 out_oversize: 315 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len); 316 goto out_fail; 317 out_oom: 318 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n"); 319 out_fail: 320 rcu_read_lock(); 321 __IP6_INC_STATS(net, __in6_dev_stats_get(dev, skb), IPSTATS_MIB_REASMFAILS); 322 rcu_read_unlock(); 323 inet_frag_kill(&fq->q); 324 return -1; 325 } 326 ipv6_frag_rcv(struct sk_buff * skb)327 static int ipv6_frag_rcv(struct sk_buff *skb) 328 { 329 struct frag_hdr *fhdr; 330 struct frag_queue *fq; 331 const struct ipv6hdr *hdr = ipv6_hdr(skb); 332 struct net *net = dev_net(skb_dst(skb)->dev); 333 u8 nexthdr; 334 int iif; 335 336 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED) 337 goto fail_hdr; 338 339 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS); 340 341 /* Jumbo payload inhibits frag. header */ 342 if (hdr->payload_len == 0) 343 goto fail_hdr; 344 345 if (!pskb_may_pull(skb, (skb_transport_offset(skb) + 346 sizeof(struct frag_hdr)))) 347 goto fail_hdr; 348 349 hdr = ipv6_hdr(skb); 350 fhdr = (struct frag_hdr *)skb_transport_header(skb); 351 352 if (!(fhdr->frag_off & htons(IP6_OFFSET | IP6_MF))) { 353 /* It is not a fragmented frame */ 354 skb->transport_header += sizeof(struct frag_hdr); 355 __IP6_INC_STATS(net, 356 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS); 357 358 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb); 359 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED; 360 IP6CB(skb)->frag_max_size = ntohs(hdr->payload_len) + 361 sizeof(struct ipv6hdr); 362 return 1; 363 } 364 365 /* RFC 8200, Section 4.5 Fragment Header: 366 * If the first fragment does not include all headers through an 367 * Upper-Layer header, then that fragment should be discarded and 368 * an ICMP Parameter Problem, Code 3, message should be sent to 369 * the source of the fragment, with the Pointer field set to zero. 370 */ 371 nexthdr = hdr->nexthdr; 372 if (ipv6frag_thdr_truncated(skb, skb_network_offset(skb) + sizeof(struct ipv6hdr), &nexthdr)) { 373 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev), 374 IPSTATS_MIB_INHDRERRORS); 375 icmpv6_param_prob(skb, ICMPV6_HDR_INCOMP, 0); 376 return -1; 377 } 378 379 iif = skb->dev ? skb->dev->ifindex : 0; 380 fq = fq_find(net, fhdr->identification, hdr, iif); 381 if (fq) { 382 u32 prob_offset = 0; 383 int ret; 384 385 spin_lock(&fq->q.lock); 386 387 fq->iif = iif; 388 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff, 389 &prob_offset); 390 391 spin_unlock(&fq->q.lock); 392 inet_frag_put(&fq->q); 393 if (prob_offset) { 394 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev), 395 IPSTATS_MIB_INHDRERRORS); 396 /* icmpv6_param_prob() calls kfree_skb(skb) */ 397 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset); 398 } 399 return ret; 400 } 401 402 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS); 403 kfree_skb(skb); 404 return -1; 405 406 fail_hdr: 407 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev), 408 IPSTATS_MIB_INHDRERRORS); 409 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb)); 410 return -1; 411 } 412 413 static const struct inet6_protocol frag_protocol = { 414 .handler = ipv6_frag_rcv, 415 .flags = INET6_PROTO_NOPOLICY, 416 }; 417 418 #ifdef CONFIG_SYSCTL 419 420 static struct ctl_table ip6_frags_ns_ctl_table[] = { 421 { 422 .procname = "ip6frag_high_thresh", 423 .maxlen = sizeof(unsigned long), 424 .mode = 0644, 425 .proc_handler = proc_doulongvec_minmax, 426 }, 427 { 428 .procname = "ip6frag_low_thresh", 429 .maxlen = sizeof(unsigned long), 430 .mode = 0644, 431 .proc_handler = proc_doulongvec_minmax, 432 }, 433 { 434 .procname = "ip6frag_time", 435 .maxlen = sizeof(int), 436 .mode = 0644, 437 .proc_handler = proc_dointvec_jiffies, 438 }, 439 }; 440 441 /* secret interval has been deprecated */ 442 static int ip6_frags_secret_interval_unused; 443 static struct ctl_table ip6_frags_ctl_table[] = { 444 { 445 .procname = "ip6frag_secret_interval", 446 .data = &ip6_frags_secret_interval_unused, 447 .maxlen = sizeof(int), 448 .mode = 0644, 449 .proc_handler = proc_dointvec_jiffies, 450 }, 451 }; 452 ip6_frags_ns_sysctl_register(struct net * net)453 static int __net_init ip6_frags_ns_sysctl_register(struct net *net) 454 { 455 struct ctl_table *table; 456 struct ctl_table_header *hdr; 457 458 table = ip6_frags_ns_ctl_table; 459 if (!net_eq(net, &init_net)) { 460 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL); 461 if (!table) 462 goto err_alloc; 463 464 } 465 table[0].data = &net->ipv6.fqdir->high_thresh; 466 table[0].extra1 = &net->ipv6.fqdir->low_thresh; 467 table[1].data = &net->ipv6.fqdir->low_thresh; 468 table[1].extra2 = &net->ipv6.fqdir->high_thresh; 469 table[2].data = &net->ipv6.fqdir->timeout; 470 471 hdr = register_net_sysctl_sz(net, "net/ipv6", table, 472 ARRAY_SIZE(ip6_frags_ns_ctl_table)); 473 if (!hdr) 474 goto err_reg; 475 476 net->ipv6.sysctl.frags_hdr = hdr; 477 return 0; 478 479 err_reg: 480 if (!net_eq(net, &init_net)) 481 kfree(table); 482 err_alloc: 483 return -ENOMEM; 484 } 485 ip6_frags_ns_sysctl_unregister(struct net * net)486 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net) 487 { 488 const struct ctl_table *table; 489 490 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg; 491 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr); 492 if (!net_eq(net, &init_net)) 493 kfree(table); 494 } 495 496 static struct ctl_table_header *ip6_ctl_header; 497 ip6_frags_sysctl_register(void)498 static int ip6_frags_sysctl_register(void) 499 { 500 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6", 501 ip6_frags_ctl_table); 502 return ip6_ctl_header == NULL ? -ENOMEM : 0; 503 } 504 ip6_frags_sysctl_unregister(void)505 static void ip6_frags_sysctl_unregister(void) 506 { 507 unregister_net_sysctl_table(ip6_ctl_header); 508 } 509 #else ip6_frags_ns_sysctl_register(struct net * net)510 static int ip6_frags_ns_sysctl_register(struct net *net) 511 { 512 return 0; 513 } 514 ip6_frags_ns_sysctl_unregister(struct net * net)515 static void ip6_frags_ns_sysctl_unregister(struct net *net) 516 { 517 } 518 ip6_frags_sysctl_register(void)519 static int ip6_frags_sysctl_register(void) 520 { 521 return 0; 522 } 523 ip6_frags_sysctl_unregister(void)524 static void ip6_frags_sysctl_unregister(void) 525 { 526 } 527 #endif 528 ipv6_frags_init_net(struct net * net)529 static int __net_init ipv6_frags_init_net(struct net *net) 530 { 531 int res; 532 533 res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net); 534 if (res < 0) 535 return res; 536 537 net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH; 538 net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH; 539 net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT; 540 541 res = ip6_frags_ns_sysctl_register(net); 542 if (res < 0) 543 fqdir_exit(net->ipv6.fqdir); 544 return res; 545 } 546 ipv6_frags_pre_exit_net(struct net * net)547 static void __net_exit ipv6_frags_pre_exit_net(struct net *net) 548 { 549 fqdir_pre_exit(net->ipv6.fqdir); 550 } 551 ipv6_frags_exit_net(struct net * net)552 static void __net_exit ipv6_frags_exit_net(struct net *net) 553 { 554 ip6_frags_ns_sysctl_unregister(net); 555 fqdir_exit(net->ipv6.fqdir); 556 } 557 558 static struct pernet_operations ip6_frags_ops = { 559 .init = ipv6_frags_init_net, 560 .pre_exit = ipv6_frags_pre_exit_net, 561 .exit = ipv6_frags_exit_net, 562 }; 563 564 static const struct rhashtable_params ip6_rhash_params = { 565 .head_offset = offsetof(struct inet_frag_queue, node), 566 .hashfn = ip6frag_key_hashfn, 567 .obj_hashfn = ip6frag_obj_hashfn, 568 .obj_cmpfn = ip6frag_obj_cmpfn, 569 .automatic_shrinking = true, 570 }; 571 ipv6_frag_init(void)572 int __init ipv6_frag_init(void) 573 { 574 int ret; 575 576 ip6_frags.constructor = ip6frag_init; 577 ip6_frags.destructor = NULL; 578 ip6_frags.qsize = sizeof(struct frag_queue); 579 ip6_frags.frag_expire = ip6_frag_expire; 580 ip6_frags.frags_cache_name = ip6_frag_cache_name; 581 ip6_frags.rhash_params = ip6_rhash_params; 582 ret = inet_frags_init(&ip6_frags); 583 if (ret) 584 goto out; 585 586 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT); 587 if (ret) 588 goto err_protocol; 589 590 ret = ip6_frags_sysctl_register(); 591 if (ret) 592 goto err_sysctl; 593 594 ret = register_pernet_subsys(&ip6_frags_ops); 595 if (ret) 596 goto err_pernet; 597 598 out: 599 return ret; 600 601 err_pernet: 602 ip6_frags_sysctl_unregister(); 603 err_sysctl: 604 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT); 605 err_protocol: 606 inet_frags_fini(&ip6_frags); 607 goto out; 608 } 609 ipv6_frag_exit(void)610 void ipv6_frag_exit(void) 611 { 612 ip6_frags_sysctl_unregister(); 613 unregister_pernet_subsys(&ip6_frags_ops); 614 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT); 615 inet_frags_fini(&ip6_frags); 616 } 617