nf_conntrack_reasm.c revision 6aafeef03b9d9ecf255f3a80ed85ee070260e1ae
1/* 2 * IPv6 fragment reassembly for connection tracking 3 * 4 * Copyright (C)2004 USAGI/WIDE Project 5 * 6 * Author: 7 * Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp> 8 * 9 * Based on: net/ipv6/reassembly.c 10 * 11 * This program is free software; you can redistribute it and/or 12 * modify it under the terms of the GNU General Public License 13 * as published by the Free Software Foundation; either version 14 * 2 of the License, or (at your option) any later version. 15 */ 16 17#define pr_fmt(fmt) "IPv6-nf: " fmt 18 19#include <linux/errno.h> 20#include <linux/types.h> 21#include <linux/string.h> 22#include <linux/socket.h> 23#include <linux/sockios.h> 24#include <linux/jiffies.h> 25#include <linux/net.h> 26#include <linux/list.h> 27#include <linux/netdevice.h> 28#include <linux/in6.h> 29#include <linux/ipv6.h> 30#include <linux/icmpv6.h> 31#include <linux/random.h> 32#include <linux/slab.h> 33 34#include <net/sock.h> 35#include <net/snmp.h> 36#include <net/inet_frag.h> 37 38#include <net/ipv6.h> 39#include <net/protocol.h> 40#include <net/transp_v6.h> 41#include <net/rawv6.h> 42#include <net/ndisc.h> 43#include <net/addrconf.h> 44#include <net/inet_ecn.h> 45#include <net/netfilter/ipv6/nf_conntrack_ipv6.h> 46#include <linux/sysctl.h> 47#include <linux/netfilter.h> 48#include <linux/netfilter_ipv6.h> 49#include <linux/kernel.h> 50#include <linux/module.h> 51#include <net/netfilter/ipv6/nf_defrag_ipv6.h> 52 53 54struct nf_ct_frag6_skb_cb 55{ 56 struct inet6_skb_parm h; 57 int offset; 58 struct sk_buff *orig; 59}; 60 61#define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb)) 62 63static struct inet_frags nf_frags; 64 65#ifdef CONFIG_SYSCTL 66static struct ctl_table nf_ct_frag6_sysctl_table[] = { 67 { 68 .procname = "nf_conntrack_frag6_timeout", 69 .data = &init_net.nf_frag.frags.timeout, 70 .maxlen = sizeof(unsigned int), 71 .mode = 0644, 72 .proc_handler = proc_dointvec_jiffies, 73 }, 74 { 75 .procname = "nf_conntrack_frag6_low_thresh", 76 .data = &init_net.nf_frag.frags.low_thresh, 77 .maxlen = sizeof(unsigned int), 78 .mode = 0644, 79 .proc_handler = proc_dointvec, 80 }, 81 { 82 .procname = "nf_conntrack_frag6_high_thresh", 83 .data = &init_net.nf_frag.frags.high_thresh, 84 .maxlen = sizeof(unsigned int), 85 .mode = 0644, 86 .proc_handler = proc_dointvec, 87 }, 88 { } 89}; 90 91static int nf_ct_frag6_sysctl_register(struct net *net) 92{ 93 struct ctl_table *table; 94 struct ctl_table_header *hdr; 95 96 table = nf_ct_frag6_sysctl_table; 97 if (!net_eq(net, &init_net)) { 98 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table), 99 GFP_KERNEL); 100 if (table == NULL) 101 goto err_alloc; 102 103 table[0].data = &net->nf_frag.frags.timeout; 104 table[1].data = &net->nf_frag.frags.low_thresh; 105 table[2].data = &net->nf_frag.frags.high_thresh; 106 } 107 108 hdr = register_net_sysctl(net, "net/netfilter", table); 109 if (hdr == NULL) 110 goto err_reg; 111 112 net->nf_frag.sysctl.frags_hdr = hdr; 113 return 0; 114 115err_reg: 116 if (!net_eq(net, &init_net)) 117 kfree(table); 118err_alloc: 119 return -ENOMEM; 120} 121 122static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net) 123{ 124 struct ctl_table *table; 125 126 table = net->nf_frag.sysctl.frags_hdr->ctl_table_arg; 127 unregister_net_sysctl_table(net->nf_frag.sysctl.frags_hdr); 128 if (!net_eq(net, &init_net)) 129 kfree(table); 130} 131 132#else 133static int nf_ct_frag6_sysctl_register(struct net *net) 134{ 135 return 0; 136} 137static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net) 138{ 139} 140#endif 141 142static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h) 143{ 144 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK); 145} 146 147static unsigned int nf_hash_frag(__be32 id, const struct in6_addr *saddr, 148 const struct in6_addr *daddr) 149{ 150 u32 c; 151 152 net_get_random_once(&nf_frags.rnd, sizeof(nf_frags.rnd)); 153 c = jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr), 154 (__force u32)id, nf_frags.rnd); 155 return c & (INETFRAGS_HASHSZ - 1); 156} 157 158 159static unsigned int nf_hashfn(struct inet_frag_queue *q) 160{ 161 const struct frag_queue *nq; 162 163 nq = container_of(q, struct frag_queue, q); 164 return nf_hash_frag(nq->id, &nq->saddr, &nq->daddr); 165} 166 167static void nf_skb_free(struct sk_buff *skb) 168{ 169 if (NFCT_FRAG6_CB(skb)->orig) 170 kfree_skb(NFCT_FRAG6_CB(skb)->orig); 171} 172 173static void nf_ct_frag6_expire(unsigned long data) 174{ 175 struct frag_queue *fq; 176 struct net *net; 177 178 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q); 179 net = container_of(fq->q.net, struct net, nf_frag.frags); 180 181 ip6_expire_frag_queue(net, fq, &nf_frags); 182} 183 184/* Creation primitives. */ 185static inline struct frag_queue *fq_find(struct net *net, __be32 id, 186 u32 user, struct in6_addr *src, 187 struct in6_addr *dst, u8 ecn) 188{ 189 struct inet_frag_queue *q; 190 struct ip6_create_arg arg; 191 unsigned int hash; 192 193 arg.id = id; 194 arg.user = user; 195 arg.src = src; 196 arg.dst = dst; 197 arg.ecn = ecn; 198 199 read_lock_bh(&nf_frags.lock); 200 hash = nf_hash_frag(id, src, dst); 201 202 q = inet_frag_find(&net->nf_frag.frags, &nf_frags, &arg, hash); 203 local_bh_enable(); 204 if (IS_ERR_OR_NULL(q)) { 205 inet_frag_maybe_warn_overflow(q, pr_fmt()); 206 return NULL; 207 } 208 return container_of(q, struct frag_queue, q); 209} 210 211 212static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb, 213 const struct frag_hdr *fhdr, int nhoff) 214{ 215 struct sk_buff *prev, *next; 216 unsigned int payload_len; 217 int offset, end; 218 u8 ecn; 219 220 if (fq->q.last_in & INET_FRAG_COMPLETE) { 221 pr_debug("Already completed\n"); 222 goto err; 223 } 224 225 payload_len = ntohs(ipv6_hdr(skb)->payload_len); 226 227 offset = ntohs(fhdr->frag_off) & ~0x7; 228 end = offset + (payload_len - 229 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); 230 231 if ((unsigned int)end > IPV6_MAXPLEN) { 232 pr_debug("offset is too large.\n"); 233 return -1; 234 } 235 236 ecn = ip6_frag_ecn(ipv6_hdr(skb)); 237 238 if (skb->ip_summed == CHECKSUM_COMPLETE) { 239 const unsigned char *nh = skb_network_header(skb); 240 skb->csum = csum_sub(skb->csum, 241 csum_partial(nh, (u8 *)(fhdr + 1) - nh, 242 0)); 243 } 244 245 /* Is this the final fragment? */ 246 if (!(fhdr->frag_off & htons(IP6_MF))) { 247 /* If we already have some bits beyond end 248 * or have different end, the segment is corrupted. 249 */ 250 if (end < fq->q.len || 251 ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) { 252 pr_debug("already received last fragment\n"); 253 goto err; 254 } 255 fq->q.last_in |= INET_FRAG_LAST_IN; 256 fq->q.len = end; 257 } else { 258 /* Check if the fragment is rounded to 8 bytes. 259 * Required by the RFC. 260 */ 261 if (end & 0x7) { 262 /* RFC2460 says always send parameter problem in 263 * this case. -DaveM 264 */ 265 pr_debug("end of fragment not rounded to 8 bytes.\n"); 266 return -1; 267 } 268 if (end > fq->q.len) { 269 /* Some bits beyond end -> corruption. */ 270 if (fq->q.last_in & INET_FRAG_LAST_IN) { 271 pr_debug("last packet already reached.\n"); 272 goto err; 273 } 274 fq->q.len = end; 275 } 276 } 277 278 if (end == offset) 279 goto err; 280 281 /* Point into the IP datagram 'data' part. */ 282 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) { 283 pr_debug("queue: message is too short.\n"); 284 goto err; 285 } 286 if (pskb_trim_rcsum(skb, end - offset)) { 287 pr_debug("Can't trim\n"); 288 goto err; 289 } 290 291 /* Find out which fragments are in front and at the back of us 292 * in the chain of fragments so far. We must know where to put 293 * this fragment, right? 294 */ 295 prev = fq->q.fragments_tail; 296 if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) { 297 next = NULL; 298 goto found; 299 } 300 prev = NULL; 301 for (next = fq->q.fragments; next != NULL; next = next->next) { 302 if (NFCT_FRAG6_CB(next)->offset >= offset) 303 break; /* bingo! */ 304 prev = next; 305 } 306 307found: 308 /* RFC5722, Section 4: 309 * When reassembling an IPv6 datagram, if 310 * one or more its constituent fragments is determined to be an 311 * overlapping fragment, the entire datagram (and any constituent 312 * fragments, including those not yet received) MUST be silently 313 * discarded. 314 */ 315 316 /* Check for overlap with preceding fragment. */ 317 if (prev && 318 (NFCT_FRAG6_CB(prev)->offset + prev->len) > offset) 319 goto discard_fq; 320 321 /* Look for overlap with succeeding segment. */ 322 if (next && NFCT_FRAG6_CB(next)->offset < end) 323 goto discard_fq; 324 325 NFCT_FRAG6_CB(skb)->offset = offset; 326 327 /* Insert this fragment in the chain of fragments. */ 328 skb->next = next; 329 if (!next) 330 fq->q.fragments_tail = skb; 331 if (prev) 332 prev->next = skb; 333 else 334 fq->q.fragments = skb; 335 336 if (skb->dev) { 337 fq->iif = skb->dev->ifindex; 338 skb->dev = NULL; 339 } 340 fq->q.stamp = skb->tstamp; 341 fq->q.meat += skb->len; 342 fq->ecn |= ecn; 343 if (payload_len > fq->q.max_size) 344 fq->q.max_size = payload_len; 345 add_frag_mem_limit(&fq->q, skb->truesize); 346 347 /* The first fragment. 348 * nhoffset is obtained from the first fragment, of course. 349 */ 350 if (offset == 0) { 351 fq->nhoffset = nhoff; 352 fq->q.last_in |= INET_FRAG_FIRST_IN; 353 } 354 355 inet_frag_lru_move(&fq->q); 356 return 0; 357 358discard_fq: 359 inet_frag_kill(&fq->q, &nf_frags); 360err: 361 return -1; 362} 363 364/* 365 * Check if this packet is complete. 366 * Returns NULL on failure by any reason, and pointer 367 * to current nexthdr field in reassembled frame. 368 * 369 * It is called with locked fq, and caller must check that 370 * queue is eligible for reassembly i.e. it is not COMPLETE, 371 * the last and the first frames arrived and all the bits are here. 372 */ 373static struct sk_buff * 374nf_ct_frag6_reasm(struct frag_queue *fq, struct net_device *dev) 375{ 376 struct sk_buff *fp, *op, *head = fq->q.fragments; 377 int payload_len; 378 u8 ecn; 379 380 inet_frag_kill(&fq->q, &nf_frags); 381 382 WARN_ON(head == NULL); 383 WARN_ON(NFCT_FRAG6_CB(head)->offset != 0); 384 385 ecn = ip_frag_ecn_table[fq->ecn]; 386 if (unlikely(ecn == 0xff)) 387 goto out_fail; 388 389 /* Unfragmented part is taken from the first segment. */ 390 payload_len = ((head->data - skb_network_header(head)) - 391 sizeof(struct ipv6hdr) + fq->q.len - 392 sizeof(struct frag_hdr)); 393 if (payload_len > IPV6_MAXPLEN) { 394 pr_debug("payload len is too large.\n"); 395 goto out_oversize; 396 } 397 398 /* Head of list must not be cloned. */ 399 if (skb_unclone(head, GFP_ATOMIC)) { 400 pr_debug("skb is cloned but can't expand head"); 401 goto out_oom; 402 } 403 404 /* If the first fragment is fragmented itself, we split 405 * it to two chunks: the first with data and paged part 406 * and the second, holding only fragments. */ 407 if (skb_has_frag_list(head)) { 408 struct sk_buff *clone; 409 int i, plen = 0; 410 411 clone = alloc_skb(0, GFP_ATOMIC); 412 if (clone == NULL) 413 goto out_oom; 414 415 clone->next = head->next; 416 head->next = clone; 417 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; 418 skb_frag_list_init(head); 419 for (i = 0; i < skb_shinfo(head)->nr_frags; i++) 420 plen += skb_frag_size(&skb_shinfo(head)->frags[i]); 421 clone->len = clone->data_len = head->data_len - plen; 422 head->data_len -= clone->len; 423 head->len -= clone->len; 424 clone->csum = 0; 425 clone->ip_summed = head->ip_summed; 426 427 NFCT_FRAG6_CB(clone)->orig = NULL; 428 add_frag_mem_limit(&fq->q, clone->truesize); 429 } 430 431 /* We have to remove fragment header from datagram and to relocate 432 * header in order to calculate ICV correctly. */ 433 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0]; 434 memmove(head->head + sizeof(struct frag_hdr), head->head, 435 (head->data - head->head) - sizeof(struct frag_hdr)); 436 head->mac_header += sizeof(struct frag_hdr); 437 head->network_header += sizeof(struct frag_hdr); 438 439 skb_shinfo(head)->frag_list = head->next; 440 skb_reset_transport_header(head); 441 skb_push(head, head->data - skb_network_header(head)); 442 443 for (fp=head->next; fp; fp = fp->next) { 444 head->data_len += fp->len; 445 head->len += fp->len; 446 if (head->ip_summed != fp->ip_summed) 447 head->ip_summed = CHECKSUM_NONE; 448 else if (head->ip_summed == CHECKSUM_COMPLETE) 449 head->csum = csum_add(head->csum, fp->csum); 450 head->truesize += fp->truesize; 451 } 452 sub_frag_mem_limit(&fq->q, head->truesize); 453 454 head->local_df = 1; 455 head->next = NULL; 456 head->dev = dev; 457 head->tstamp = fq->q.stamp; 458 ipv6_hdr(head)->payload_len = htons(payload_len); 459 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn); 460 IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size; 461 462 /* Yes, and fold redundant checksum back. 8) */ 463 if (head->ip_summed == CHECKSUM_COMPLETE) 464 head->csum = csum_partial(skb_network_header(head), 465 skb_network_header_len(head), 466 head->csum); 467 468 fq->q.fragments = NULL; 469 fq->q.fragments_tail = NULL; 470 471 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */ 472 fp = skb_shinfo(head)->frag_list; 473 if (fp && NFCT_FRAG6_CB(fp)->orig == NULL) 474 /* at above code, head skb is divided into two skbs. */ 475 fp = fp->next; 476 477 op = NFCT_FRAG6_CB(head)->orig; 478 for (; fp; fp = fp->next) { 479 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig; 480 481 op->next = orig; 482 op = orig; 483 NFCT_FRAG6_CB(fp)->orig = NULL; 484 } 485 486 return head; 487 488out_oversize: 489 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n", 490 payload_len); 491 goto out_fail; 492out_oom: 493 net_dbg_ratelimited("nf_ct_frag6_reasm: no memory for reassembly\n"); 494out_fail: 495 return NULL; 496} 497 498/* 499 * find the header just before Fragment Header. 500 * 501 * if success return 0 and set ... 502 * (*prevhdrp): the value of "Next Header Field" in the header 503 * just before Fragment Header. 504 * (*prevhoff): the offset of "Next Header Field" in the header 505 * just before Fragment Header. 506 * (*fhoff) : the offset of Fragment Header. 507 * 508 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c 509 * 510 */ 511static int 512find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff) 513{ 514 u8 nexthdr = ipv6_hdr(skb)->nexthdr; 515 const int netoff = skb_network_offset(skb); 516 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr); 517 int start = netoff + sizeof(struct ipv6hdr); 518 int len = skb->len - start; 519 u8 prevhdr = NEXTHDR_IPV6; 520 521 while (nexthdr != NEXTHDR_FRAGMENT) { 522 struct ipv6_opt_hdr hdr; 523 int hdrlen; 524 525 if (!ipv6_ext_hdr(nexthdr)) { 526 return -1; 527 } 528 if (nexthdr == NEXTHDR_NONE) { 529 pr_debug("next header is none\n"); 530 return -1; 531 } 532 if (len < (int)sizeof(struct ipv6_opt_hdr)) { 533 pr_debug("too short\n"); 534 return -1; 535 } 536 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr))) 537 BUG(); 538 if (nexthdr == NEXTHDR_AUTH) 539 hdrlen = (hdr.hdrlen+2)<<2; 540 else 541 hdrlen = ipv6_optlen(&hdr); 542 543 prevhdr = nexthdr; 544 prev_nhoff = start; 545 546 nexthdr = hdr.nexthdr; 547 len -= hdrlen; 548 start += hdrlen; 549 } 550 551 if (len < 0) 552 return -1; 553 554 *prevhdrp = prevhdr; 555 *prevhoff = prev_nhoff; 556 *fhoff = start; 557 558 return 0; 559} 560 561struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user) 562{ 563 struct sk_buff *clone; 564 struct net_device *dev = skb->dev; 565 struct net *net = skb_dst(skb) ? dev_net(skb_dst(skb)->dev) 566 : dev_net(skb->dev); 567 struct frag_hdr *fhdr; 568 struct frag_queue *fq; 569 struct ipv6hdr *hdr; 570 int fhoff, nhoff; 571 u8 prevhdr; 572 struct sk_buff *ret_skb = NULL; 573 574 /* Jumbo payload inhibits frag. header */ 575 if (ipv6_hdr(skb)->payload_len == 0) { 576 pr_debug("payload len = 0\n"); 577 return skb; 578 } 579 580 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0) 581 return skb; 582 583 clone = skb_clone(skb, GFP_ATOMIC); 584 if (clone == NULL) { 585 pr_debug("Can't clone skb\n"); 586 return skb; 587 } 588 589 NFCT_FRAG6_CB(clone)->orig = skb; 590 591 if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) { 592 pr_debug("message is too short.\n"); 593 goto ret_orig; 594 } 595 596 skb_set_transport_header(clone, fhoff); 597 hdr = ipv6_hdr(clone); 598 fhdr = (struct frag_hdr *)skb_transport_header(clone); 599 600 local_bh_disable(); 601 inet_frag_evictor(&net->nf_frag.frags, &nf_frags, false); 602 local_bh_enable(); 603 604 fq = fq_find(net, fhdr->identification, user, &hdr->saddr, &hdr->daddr, 605 ip6_frag_ecn(hdr)); 606 if (fq == NULL) { 607 pr_debug("Can't find and can't create new queue\n"); 608 goto ret_orig; 609 } 610 611 spin_lock_bh(&fq->q.lock); 612 613 if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) { 614 spin_unlock_bh(&fq->q.lock); 615 pr_debug("Can't insert skb to queue\n"); 616 inet_frag_put(&fq->q, &nf_frags); 617 goto ret_orig; 618 } 619 620 if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && 621 fq->q.meat == fq->q.len) { 622 ret_skb = nf_ct_frag6_reasm(fq, dev); 623 if (ret_skb == NULL) 624 pr_debug("Can't reassemble fragmented packets\n"); 625 } 626 spin_unlock_bh(&fq->q.lock); 627 628 inet_frag_put(&fq->q, &nf_frags); 629 return ret_skb; 630 631ret_orig: 632 kfree_skb(clone); 633 return skb; 634} 635 636void nf_ct_frag6_consume_orig(struct sk_buff *skb) 637{ 638 struct sk_buff *s, *s2; 639 640 for (s = NFCT_FRAG6_CB(skb)->orig; s;) { 641 s2 = s->next; 642 s->next = NULL; 643 consume_skb(s); 644 s = s2; 645 } 646} 647 648static int nf_ct_net_init(struct net *net) 649{ 650 net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH; 651 net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH; 652 net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT; 653 inet_frags_init_net(&net->nf_frag.frags); 654 655 return nf_ct_frag6_sysctl_register(net); 656} 657 658static void nf_ct_net_exit(struct net *net) 659{ 660 nf_ct_frags6_sysctl_unregister(net); 661 inet_frags_exit_net(&net->nf_frag.frags, &nf_frags); 662} 663 664static struct pernet_operations nf_ct_net_ops = { 665 .init = nf_ct_net_init, 666 .exit = nf_ct_net_exit, 667}; 668 669int nf_ct_frag6_init(void) 670{ 671 int ret = 0; 672 673 nf_frags.hashfn = nf_hashfn; 674 nf_frags.constructor = ip6_frag_init; 675 nf_frags.destructor = NULL; 676 nf_frags.skb_free = nf_skb_free; 677 nf_frags.qsize = sizeof(struct frag_queue); 678 nf_frags.match = ip6_frag_match; 679 nf_frags.frag_expire = nf_ct_frag6_expire; 680 nf_frags.secret_interval = 10 * 60 * HZ; 681 inet_frags_init(&nf_frags); 682 683 ret = register_pernet_subsys(&nf_ct_net_ops); 684 if (ret) 685 inet_frags_fini(&nf_frags); 686 687 return ret; 688} 689 690void nf_ct_frag6_cleanup(void) 691{ 692 unregister_pernet_subsys(&nf_ct_net_ops); 693 inet_frags_fini(&nf_frags); 694} 695