net/ipv6/netfilter/nf_conntrack_reasm.c

   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 static const char nf_frags_cache_name[] = "nf-frags";
  54
  55 static struct inet_frags nf_frags;
  56
  57 #ifdef CONFIG_SYSCTL
  58
  59 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
  60         {
  61                 .procname       = "nf_conntrack_frag6_timeout",
  62                 .data           = &init_net.nf_frag.frags.timeout,
  63                 .maxlen         = sizeof(unsigned int),
  64                 .mode           = 0644,
  65                 .proc_handler   = proc_dointvec_jiffies,
  66         },
  67         {
  68                 .procname       = "nf_conntrack_frag6_low_thresh",
  69                 .data           = &init_net.nf_frag.frags.low_thresh,
  70                 .maxlen         = sizeof(unsigned long),
  71                 .mode           = 0644,
  72                 .proc_handler   = proc_doulongvec_minmax,
  73                 .extra2         = &init_net.nf_frag.frags.high_thresh
  74         },
  75         {
  76                 .procname       = "nf_conntrack_frag6_high_thresh",
  77                 .data           = &init_net.nf_frag.frags.high_thresh,
  78                 .maxlen         = sizeof(unsigned long),
  79                 .mode           = 0644,
  80                 .proc_handler   = proc_doulongvec_minmax,
  81                 .extra1         = &init_net.nf_frag.frags.low_thresh
  82         },
  83         { }
  84 };
  85
  86 static int nf_ct_frag6_sysctl_register(struct net *net)
  87 {
  88         struct ctl_table *table;
  89         struct ctl_table_header *hdr;
  90
  91         table = nf_ct_frag6_sysctl_table;
  92         if (!net_eq(net, &init_net)) {
  93                 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
  94                                 GFP_KERNEL);
  95                 if (table == NULL)
  96                         goto err_alloc;
  97
  98                 table[0].data = &net->nf_frag.frags.timeout;
  99                 table[1].data = &net->nf_frag.frags.low_thresh;
 100                 table[1].extra2 = &net->nf_frag.frags.high_thresh;
 101                 table[2].data = &net->nf_frag.frags.high_thresh;
 102                 table[2].extra1 = &net->nf_frag.frags.low_thresh;
 103                 table[2].extra2 = &init_net.nf_frag.frags.high_thresh;
 104         }
 105
 106         hdr = register_net_sysctl(net, "net/netfilter", table);
 107         if (hdr == NULL)
 108                 goto err_reg;
 109
 110         net->nf_frag_frags_hdr = hdr;
 111         return 0;
 112
 113 err_reg:
 114         if (!net_eq(net, &init_net))
 115                 kfree(table);
 116 err_alloc:
 117         return -ENOMEM;
 118 }
 119
 120 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
 121 {
 122         struct ctl_table *table;
 123
 124         table = net->nf_frag_frags_hdr->ctl_table_arg;
 125         unregister_net_sysctl_table(net->nf_frag_frags_hdr);
 126         if (!net_eq(net, &init_net))
 127                 kfree(table);
 128 }
 129
 130 #else
 131 static int nf_ct_frag6_sysctl_register(struct net *net)
 132 {
 133         return 0;
 134 }
 135 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
 136 {
 137 }
 138 #endif
 139
 140 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
 141 {
 142         return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
 143 }
 144
 145 static void nf_ct_frag6_expire(struct timer_list *t)
 146 {
 147         struct inet_frag_queue *frag = from_timer(frag, t, timer);
 148         struct frag_queue *fq;
 149         struct net *net;
 150
 151         fq = container_of(frag, struct frag_queue, q);
 152         net = container_of(fq->q.net, struct net, nf_frag.frags);
 153
 154         ip6_expire_frag_queue(net, fq);
 155 }
 156
 157 /* Creation primitives. */
 158 static struct frag_queue *fq_find(struct net *net, __be32 id, u32 user,
 159                                   const struct ipv6hdr *hdr, int iif)
 160 {
 161         struct frag_v6_compare_key key = {
 162                 .id = id,
 163                 .saddr = hdr->saddr,
 164                 .daddr = hdr->daddr,
 165                 .user = user,
 166                 .iif = iif,
 167         };
 168         struct inet_frag_queue *q;
 169
 170         q = inet_frag_find(&net->nf_frag.frags, &key);
 171         if (!q)
 172                 return NULL;
 173
 174         return container_of(q, struct frag_queue, q);
 175 }
 176
 177
 178 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
 179                              const struct frag_hdr *fhdr, int nhoff)
 180 {
 181         struct sk_buff *prev, *next;
 182         unsigned int payload_len;
 183         int offset, end;
 184         u8 ecn;
 185
 186         if (fq->q.flags & INET_FRAG_COMPLETE) {
 187                 pr_debug("Already completed\n");
 188                 goto err;
 189         }
 190
 191         payload_len = ntohs(ipv6_hdr(skb)->payload_len);
 192
 193         offset = ntohs(fhdr->frag_off) & ~0x7;
 194         end = offset + (payload_len -
 195                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 196
 197         if ((unsigned int)end > IPV6_MAXPLEN) {
 198                 pr_debug("offset is too large.\n");
 199                 return -EINVAL;
 200         }
 201
 202         ecn = ip6_frag_ecn(ipv6_hdr(skb));
 203
 204         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 205                 const unsigned char *nh = skb_network_header(skb);
 206                 skb->csum = csum_sub(skb->csum,
 207                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 208                                                   0));
 209         }
 210
 211         /* Is this the final fragment? */
 212         if (!(fhdr->frag_off & htons(IP6_MF))) {
 213                 /* If we already have some bits beyond end
 214                  * or have different end, the segment is corrupted.
 215                  */
 216                 if (end < fq->q.len ||
 217                     ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
 218                         pr_debug("already received last fragment\n");
 219                         goto err;
 220                 }
 221                 fq->q.flags |= INET_FRAG_LAST_IN;
 222                 fq->q.len = end;
 223         } else {
 224                 /* Check if the fragment is rounded to 8 bytes.
 225                  * Required by the RFC.
 226                  */
 227                 if (end & 0x7) {
 228                         /* RFC2460 says always send parameter problem in
 229                          * this case. -DaveM
 230                          */
 231                         pr_debug("end of fragment not rounded to 8 bytes.\n");
 232                         inet_frag_kill(&fq->q);
 233                         return -EPROTO;
 234                 }
 235                 if (end > fq->q.len) {
 236                         /* Some bits beyond end -> corruption. */
 237                         if (fq->q.flags & INET_FRAG_LAST_IN) {
 238                                 pr_debug("last packet already reached.\n");
 239                                 goto err;
 240                         }
 241                         fq->q.len = end;
 242                 }
 243         }
 244
 245         if (end == offset)
 246                 goto err;
 247
 248         /* Point into the IP datagram 'data' part. */
 249         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
 250                 pr_debug("queue: message is too short.\n");
 251                 goto err;
 252         }
 253         if (pskb_trim_rcsum(skb, end - offset)) {
 254                 pr_debug("Can't trim\n");
 255                 goto err;
 256         }
 257
 258         /* Find out which fragments are in front and at the back of us
 259          * in the chain of fragments so far.  We must know where to put
 260          * this fragment, right?
 261          */
 262         prev = fq->q.fragments_tail;
 263         if (!prev || prev->ip_defrag_offset < offset) {
 264                 next = NULL;
 265                 goto found;
 266         }
 267         prev = NULL;
 268         for (next = fq->q.fragments; next != NULL; next = next->next) {
 269                 if (next->ip_defrag_offset >= offset)
 270                         break;  /* bingo! */
 271                 prev = next;
 272         }
 273
 274 found:
 275         /* RFC5722, Section 4:
 276          *                                  When reassembling an IPv6 datagram, if
 277          *   one or more its constituent fragments is determined to be an
 278          *   overlapping fragment, the entire datagram (and any constituent
 279          *   fragments, including those not yet received) MUST be silently
 280          *   discarded.
 281          */
 282
 283         /* Check for overlap with preceding fragment. */
 284         if (prev &&
 285             (prev->ip_defrag_offset + prev->len) > offset)
 286                 goto discard_fq;
 287
 288         /* Look for overlap with succeeding segment. */
 289         if (next && next->ip_defrag_offset < end)
 290                 goto discard_fq;
 291
 292         /* Note : skb->ip_defrag_offset and skb->dev share the same location */
 293         if (skb->dev)
 294                 fq->iif = skb->dev->ifindex;
 295         /* Makes sure compiler wont do silly aliasing games */
 296         barrier();
 297         skb->ip_defrag_offset = offset;
 298
 299         /* Insert this fragment in the chain of fragments. */
 300         skb->next = next;
 301         if (!next)
 302                 fq->q.fragments_tail = skb;
 303         if (prev)
 304                 prev->next = skb;
 305         else
 306                 fq->q.fragments = skb;
 307
 308         fq->q.stamp = skb->tstamp;
 309         fq->q.meat += skb->len;
 310         fq->ecn |= ecn;
 311         if (payload_len > fq->q.max_size)
 312                 fq->q.max_size = payload_len;
 313         add_frag_mem_limit(fq->q.net, skb->truesize);
 314
 315         /* The first fragment.
 316          * nhoffset is obtained from the first fragment, of course.
 317          */
 318         if (offset == 0) {
 319                 fq->nhoffset = nhoff;
 320                 fq->q.flags |= INET_FRAG_FIRST_IN;
 321         }
 322
 323         return 0;
 324
 325 discard_fq:
 326         inet_frag_kill(&fq->q);
 327 err:
 328         return -EINVAL;
 329 }
 330
 331 /*
 332  *      Check if this packet is complete.
 333  *
 334  *      It is called with locked fq, and caller must check that
 335  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 336  *      the last and the first frames arrived and all the bits are here.
 337  *
 338  *      returns true if *prev skb has been transformed into the reassembled
 339  *      skb, false otherwise.
 340  */
 341 static bool
 342 nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *prev,  struct net_device *dev)
 343 {
 344         struct sk_buff *fp, *head = fq->q.fragments;
 345         int    payload_len;
 346         u8 ecn;
 347
 348         inet_frag_kill(&fq->q);
 349
 350         WARN_ON(head == NULL);
 351         WARN_ON(head->ip_defrag_offset != 0);
 352
 353         ecn = ip_frag_ecn_table[fq->ecn];
 354         if (unlikely(ecn == 0xff))
 355                 return false;
 356
 357         /* Unfragmented part is taken from the first segment. */
 358         payload_len = ((head->data - skb_network_header(head)) -
 359                        sizeof(struct ipv6hdr) + fq->q.len -
 360                        sizeof(struct frag_hdr));
 361         if (payload_len > IPV6_MAXPLEN) {
 362                 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
 363                                     payload_len);
 364                 return false;
 365         }
 366
 367         /* Head of list must not be cloned. */
 368         if (skb_unclone(head, GFP_ATOMIC))
 369                 return false;
 370
 371         /* If the first fragment is fragmented itself, we split
 372          * it to two chunks: the first with data and paged part
 373          * and the second, holding only fragments. */
 374         if (skb_has_frag_list(head)) {
 375                 struct sk_buff *clone;
 376                 int i, plen = 0;
 377
 378                 clone = alloc_skb(0, GFP_ATOMIC);
 379                 if (clone == NULL)
 380                         return false;
 381
 382                 clone->next = head->next;
 383                 head->next = clone;
 384                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 385                 skb_frag_list_init(head);
 386                 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
 387                         plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
 388                 clone->len = clone->data_len = head->data_len - plen;
 389                 head->data_len -= clone->len;
 390                 head->len -= clone->len;
 391                 clone->csum = 0;
 392                 clone->ip_summed = head->ip_summed;
 393
 394                 add_frag_mem_limit(fq->q.net, clone->truesize);
 395         }
 396
 397         /* morph head into last received skb: prev.
 398          *
 399          * This allows callers of ipv6 conntrack defrag to continue
 400          * to use the last skb(frag) passed into the reasm engine.
 401          * The last skb frag 'silently' turns into the full reassembled skb.
 402          *
 403          * Since prev is also part of q->fragments we have to clone it first.
 404          */
 405         if (head != prev) {
 406                 struct sk_buff *iter;
 407
 408                 fp = skb_clone(prev, GFP_ATOMIC);
 409                 if (!fp)
 410                         return false;
 411
 412                 fp->next = prev->next;
 413
 414                 iter = head;
 415                 while (iter) {
 416                         if (iter->next == prev) {
 417                                 iter->next = fp;
 418                                 break;
 419                         }
 420                         iter = iter->next;
 421                 }
 422
 423                 skb_morph(prev, head);
 424                 prev->next = head->next;
 425                 consume_skb(head);
 426                 head = prev;
 427         }
 428
 429         /* We have to remove fragment header from datagram and to relocate
 430          * header in order to calculate ICV correctly. */
 431         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
 432         memmove(head->head + sizeof(struct frag_hdr), head->head,
 433                 (head->data - head->head) - sizeof(struct frag_hdr));
 434         head->mac_header += sizeof(struct frag_hdr);
 435         head->network_header += sizeof(struct frag_hdr);
 436
 437         skb_shinfo(head)->frag_list = head->next;
 438         skb_reset_transport_header(head);
 439         skb_push(head, head->data - skb_network_header(head));
 440
 441         for (fp = head->next; fp; fp = fp->next) {
 442                 head->data_len += fp->len;
 443                 head->len += fp->len;
 444                 if (head->ip_summed != fp->ip_summed)
 445                         head->ip_summed = CHECKSUM_NONE;
 446                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 447                         head->csum = csum_add(head->csum, fp->csum);
 448                 head->truesize += fp->truesize;
 449         }
 450         sub_frag_mem_limit(fq->q.net, head->truesize);
 451
 452         head->ignore_df = 1;
 453         head->next = NULL;
 454         head->dev = dev;
 455         head->tstamp = fq->q.stamp;
 456         ipv6_hdr(head)->payload_len = htons(payload_len);
 457         ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
 458         IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
 459
 460         /* Yes, and fold redundant checksum back. 8) */
 461         if (head->ip_summed == CHECKSUM_COMPLETE)
 462                 head->csum = csum_partial(skb_network_header(head),
 463                                           skb_network_header_len(head),
 464                                           head->csum);
 465
 466         fq->q.fragments = NULL;
 467         fq->q.fragments_tail = NULL;
 468
 469         return true;
 470 }
 471
 472 /*
 473  * find the header just before Fragment Header.
 474  *
 475  * if success return 0 and set ...
 476  * (*prevhdrp): the value of "Next Header Field" in the header
 477  *              just before Fragment Header.
 478  * (*prevhoff): the offset of "Next Header Field" in the header
 479  *              just before Fragment Header.
 480  * (*fhoff)   : the offset of Fragment Header.
 481  *
 482  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
 483  *
 484  */
 485 static int
 486 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
 487 {
 488         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 489         const int netoff = skb_network_offset(skb);
 490         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
 491         int start = netoff + sizeof(struct ipv6hdr);
 492         int len = skb->len - start;
 493         u8 prevhdr = NEXTHDR_IPV6;
 494
 495         while (nexthdr != NEXTHDR_FRAGMENT) {
 496                 struct ipv6_opt_hdr hdr;
 497                 int hdrlen;
 498
 499                 if (!ipv6_ext_hdr(nexthdr)) {
 500                         return -1;
 501                 }
 502                 if (nexthdr == NEXTHDR_NONE) {
 503                         pr_debug("next header is none\n");
 504                         return -1;
 505                 }
 506                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
 507                         pr_debug("too short\n");
 508                         return -1;
 509                 }
 510                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
 511                         BUG();
 512                 if (nexthdr == NEXTHDR_AUTH)
 513                         hdrlen = (hdr.hdrlen+2)<<2;
 514                 else
 515                         hdrlen = ipv6_optlen(&hdr);
 516
 517                 prevhdr = nexthdr;
 518                 prev_nhoff = start;
 519
 520                 nexthdr = hdr.nexthdr;
 521                 len -= hdrlen;
 522                 start += hdrlen;
 523         }
 524
 525         if (len < 0)
 526                 return -1;
 527
 528         *prevhdrp = prevhdr;
 529         *prevhoff = prev_nhoff;
 530         *fhoff = start;
 531
 532         return 0;
 533 }
 534
 535 int nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
 536 {
 537         u16 savethdr = skb->transport_header;
 538         struct net_device *dev = skb->dev;
 539         int fhoff, nhoff, ret;
 540         struct frag_hdr *fhdr;
 541         struct frag_queue *fq;
 542         struct ipv6hdr *hdr;
 543         u8 prevhdr;
 544
 545         /* Jumbo payload inhibits frag. header */
 546         if (ipv6_hdr(skb)->payload_len == 0) {
 547                 pr_debug("payload len = 0\n");
 548                 return 0;
 549         }
 550
 551         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
 552                 return 0;
 553
 554         if (!pskb_may_pull(skb, fhoff + sizeof(*fhdr)))
 555                 return -ENOMEM;
 556
 557         skb_set_transport_header(skb, fhoff);
 558         hdr = ipv6_hdr(skb);
 559         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 560
 561         skb_orphan(skb);
 562         fq = fq_find(net, fhdr->identification, user, hdr,
 563                      skb->dev ? skb->dev->ifindex : 0);
 564         if (fq == NULL) {
 565                 pr_debug("Can't find and can't create new queue\n");
 566                 return -ENOMEM;
 567         }
 568
 569         spin_lock_bh(&fq->q.lock);
 570
 571         ret = nf_ct_frag6_queue(fq, skb, fhdr, nhoff);
 572         if (ret < 0) {
 573                 if (ret == -EPROTO) {
 574                         skb->transport_header = savethdr;
 575                         ret = 0;
 576                 }
 577                 goto out_unlock;
 578         }
 579
 580         /* after queue has assumed skb ownership, only 0 or -EINPROGRESS
 581          * must be returned.
 582          */
 583         ret = -EINPROGRESS;
 584         if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 585             fq->q.meat == fq->q.len &&
 586             nf_ct_frag6_reasm(fq, skb, dev))
 587                 ret = 0;
 588         else
 589                 skb_dst_drop(skb);
 590
 591 out_unlock:
 592         spin_unlock_bh(&fq->q.lock);
 593         inet_frag_put(&fq->q);
 594         return ret;
 595 }
 596 EXPORT_SYMBOL_GPL(nf_ct_frag6_gather);
 597
 598 static int nf_ct_net_init(struct net *net)
 599 {
 600         int res;
 601
 602         net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
 603         net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
 604         net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
 605         net->nf_frag.frags.f = &nf_frags;
 606
 607         res = inet_frags_init_net(&net->nf_frag.frags);
 608         if (res < 0)
 609                 return res;
 610         res = nf_ct_frag6_sysctl_register(net);
 611         if (res < 0)
 612                 inet_frags_exit_net(&net->nf_frag.frags);
 613         return res;
 614 }
 615
 616 static void nf_ct_net_exit(struct net *net)
 617 {
 618         nf_ct_frags6_sysctl_unregister(net);
 619         inet_frags_exit_net(&net->nf_frag.frags);
 620 }
 621
 622 static struct pernet_operations nf_ct_net_ops = {
 623         .init = nf_ct_net_init,
 624         .exit = nf_ct_net_exit,
 625 };
 626
 627 int nf_ct_frag6_init(void)
 628 {
 629         int ret = 0;
 630
 631         nf_frags.constructor = ip6_frag_init;
 632         nf_frags.destructor = NULL;
 633         nf_frags.qsize = sizeof(struct frag_queue);
 634         nf_frags.frag_expire = nf_ct_frag6_expire;
 635         nf_frags.frags_cache_name = nf_frags_cache_name;
 636         nf_frags.rhash_params = ip6_rhash_params;
 637         ret = inet_frags_init(&nf_frags);
 638         if (ret)
 639                 goto out;
 640         ret = register_pernet_subsys(&nf_ct_net_ops);
 641         if (ret)
 642                 inet_frags_fini(&nf_frags);
 643
 644 out:
 645         return ret;
 646 }
 647
 648 void nf_ct_frag6_cleanup(void)
 649 {
 650         unregister_pernet_subsys(&nf_ct_net_ops);
 651         inet_frags_fini(&nf_frags);
 652 }