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/ipv6_frag.h>
  37
  38 #include <net/protocol.h>
  39 #include <net/transp_v6.h>
  40 #include <net/rawv6.h>
  41 #include <net/ndisc.h>
  42 #include <net/addrconf.h>
  43 #include <net/inet_ecn.h>
  44 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
  45 #include <linux/sysctl.h>
  46 #include <linux/netfilter.h>
  47 #include <linux/netfilter_ipv6.h>
  48 #include <linux/kernel.h>
  49 #include <linux/module.h>
  50 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  51
  52 static const char nf_frags_cache_name[] = "nf-frags";
  53
  54 static struct inet_frags nf_frags;
  55
  56 #ifdef CONFIG_SYSCTL
  57
  58 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
  59         {
  60                 .procname       = "nf_conntrack_frag6_timeout",
  61                 .data           = &init_net.nf_frag.frags.timeout,
  62                 .maxlen         = sizeof(unsigned int),
  63                 .mode           = 0644,
  64                 .proc_handler   = proc_dointvec_jiffies,
  65         },
  66         {
  67                 .procname       = "nf_conntrack_frag6_low_thresh",
  68                 .data           = &init_net.nf_frag.frags.low_thresh,
  69                 .maxlen         = sizeof(unsigned long),
  70                 .mode           = 0644,
  71                 .proc_handler   = proc_doulongvec_minmax,
  72                 .extra2         = &init_net.nf_frag.frags.high_thresh
  73         },
  74         {
  75                 .procname       = "nf_conntrack_frag6_high_thresh",
  76                 .data           = &init_net.nf_frag.frags.high_thresh,
  77                 .maxlen         = sizeof(unsigned long),
  78                 .mode           = 0644,
  79                 .proc_handler   = proc_doulongvec_minmax,
  80                 .extra1         = &init_net.nf_frag.frags.low_thresh
  81         },
  82         { }
  83 };
  84
  85 static int nf_ct_frag6_sysctl_register(struct net *net)
  86 {
  87         struct ctl_table *table;
  88         struct ctl_table_header *hdr;
  89
  90         table = nf_ct_frag6_sysctl_table;
  91         if (!net_eq(net, &init_net)) {
  92                 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
  93                                 GFP_KERNEL);
  94                 if (table == NULL)
  95                         goto err_alloc;
  96
  97                 table[0].data = &net->nf_frag.frags.timeout;
  98                 table[1].data = &net->nf_frag.frags.low_thresh;
  99                 table[1].extra2 = &net->nf_frag.frags.high_thresh;
 100                 table[2].data = &net->nf_frag.frags.high_thresh;
 101                 table[2].extra1 = &net->nf_frag.frags.low_thresh;
 102                 table[2].extra2 = &init_net.nf_frag.frags.high_thresh;
 103         }
 104
 105         hdr = register_net_sysctl(net, "net/netfilter", table);
 106         if (hdr == NULL)
 107                 goto err_reg;
 108
 109         net->nf_frag_frags_hdr = hdr;
 110         return 0;
 111
 112 err_reg:
 113         if (!net_eq(net, &init_net))
 114                 kfree(table);
 115 err_alloc:
 116         return -ENOMEM;
 117 }
 118
 119 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
 120 {
 121         struct ctl_table *table;
 122
 123         table = net->nf_frag_frags_hdr->ctl_table_arg;
 124         unregister_net_sysctl_table(net->nf_frag_frags_hdr);
 125         if (!net_eq(net, &init_net))
 126                 kfree(table);
 127 }
 128
 129 #else
 130 static int nf_ct_frag6_sysctl_register(struct net *net)
 131 {
 132         return 0;
 133 }
 134 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
 135 {
 136 }
 137 #endif
 138
 139 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
 140                              struct sk_buff *prev_tail, struct net_device *dev);
 141
 142 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
 143 {
 144         return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
 145 }
 146
 147 static void nf_ct_frag6_expire(struct timer_list *t)
 148 {
 149         struct inet_frag_queue *frag = from_timer(frag, t, timer);
 150         struct frag_queue *fq;
 151         struct net *net;
 152
 153         fq = container_of(frag, struct frag_queue, q);
 154         net = container_of(fq->q.net, struct net, nf_frag.frags);
 155
 156         ip6frag_expire_frag_queue(net, fq);
 157 }
 158
 159 /* Creation primitives. */
 160 static struct frag_queue *fq_find(struct net *net, __be32 id, u32 user,
 161                                   const struct ipv6hdr *hdr, int iif)
 162 {
 163         struct frag_v6_compare_key key = {
 164                 .id = id,
 165                 .saddr = hdr->saddr,
 166                 .daddr = hdr->daddr,
 167                 .user = user,
 168                 .iif = iif,
 169         };
 170         struct inet_frag_queue *q;
 171
 172         q = inet_frag_find(&net->nf_frag.frags, &key);
 173         if (!q)
 174                 return NULL;
 175
 176         return container_of(q, struct frag_queue, q);
 177 }
 178
 179
 180 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
 181                              const struct frag_hdr *fhdr, int nhoff)
 182 {
 183         unsigned int payload_len;
 184         struct net_device *dev;
 185         struct sk_buff *prev;
 186         int offset, end, err;
 187         u8 ecn;
 188
 189         if (fq->q.flags & INET_FRAG_COMPLETE) {
 190                 pr_debug("Already completed\n");
 191                 goto err;
 192         }
 193
 194         payload_len = ntohs(ipv6_hdr(skb)->payload_len);
 195
 196         offset = ntohs(fhdr->frag_off) & ~0x7;
 197         end = offset + (payload_len -
 198                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 199
 200         if ((unsigned int)end > IPV6_MAXPLEN) {
 201                 pr_debug("offset is too large.\n");
 202                 return -EINVAL;
 203         }
 204
 205         ecn = ip6_frag_ecn(ipv6_hdr(skb));
 206
 207         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 208                 const unsigned char *nh = skb_network_header(skb);
 209                 skb->csum = csum_sub(skb->csum,
 210                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 211                                                   0));
 212         }
 213
 214         /* Is this the final fragment? */
 215         if (!(fhdr->frag_off & htons(IP6_MF))) {
 216                 /* If we already have some bits beyond end
 217                  * or have different end, the segment is corrupted.
 218                  */
 219                 if (end < fq->q.len ||
 220                     ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
 221                         pr_debug("already received last fragment\n");
 222                         goto err;
 223                 }
 224                 fq->q.flags |= INET_FRAG_LAST_IN;
 225                 fq->q.len = end;
 226         } else {
 227                 /* Check if the fragment is rounded to 8 bytes.
 228                  * Required by the RFC.
 229                  */
 230                 if (end & 0x7) {
 231                         /* RFC2460 says always send parameter problem in
 232                          * this case. -DaveM
 233                          */
 234                         pr_debug("end of fragment not rounded to 8 bytes.\n");
 235                         inet_frag_kill(&fq->q);
 236                         return -EPROTO;
 237                 }
 238                 if (end > fq->q.len) {
 239                         /* Some bits beyond end -> corruption. */
 240                         if (fq->q.flags & INET_FRAG_LAST_IN) {
 241                                 pr_debug("last packet already reached.\n");
 242                                 goto err;
 243                         }
 244                         fq->q.len = end;
 245                 }
 246         }
 247
 248         if (end == offset)
 249                 goto err;
 250
 251         /* Point into the IP datagram 'data' part. */
 252         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
 253                 pr_debug("queue: message is too short.\n");
 254                 goto err;
 255         }
 256         if (pskb_trim_rcsum(skb, end - offset)) {
 257                 pr_debug("Can't trim\n");
 258                 goto err;
 259         }
 260
 261         /* Note : skb->rbnode and skb->dev share the same location. */
 262         dev = skb->dev;
 263         /* Makes sure compiler wont do silly aliasing games */
 264         barrier();
 265
 266         prev = fq->q.fragments_tail;
 267         err = inet_frag_queue_insert(&fq->q, skb, offset, end);
 268         if (err) {
 269                 if (err == IPFRAG_DUP) {
 270                         /* No error for duplicates, pretend they got queued. */
 271                         kfree_skb(skb);
 272                         return -EINPROGRESS;
 273                 }
 274                 goto insert_error;
 275         }
 276
 277         if (dev)
 278                 fq->iif = dev->ifindex;
 279
 280         fq->q.stamp = skb->tstamp;
 281         fq->q.meat += skb->len;
 282         fq->ecn |= ecn;
 283         if (payload_len > fq->q.max_size)
 284                 fq->q.max_size = payload_len;
 285         add_frag_mem_limit(fq->q.net, skb->truesize);
 286
 287         /* The first fragment.
 288          * nhoffset is obtained from the first fragment, of course.
 289          */
 290         if (offset == 0) {
 291                 fq->nhoffset = nhoff;
 292                 fq->q.flags |= INET_FRAG_FIRST_IN;
 293         }
 294
 295         if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 296             fq->q.meat == fq->q.len) {
 297                 unsigned long orefdst = skb->_skb_refdst;
 298
 299                 skb->_skb_refdst = 0UL;
 300                 err = nf_ct_frag6_reasm(fq, skb, prev, dev);
 301                 skb->_skb_refdst = orefdst;
 302
 303                 /* After queue has assumed skb ownership, only 0 or
 304                  * -EINPROGRESS must be returned.
 305                  */
 306                 return err ? -EINPROGRESS : 0;
 307         }
 308
 309         skb_dst_drop(skb);
 310         return -EINPROGRESS;
 311
 312 insert_error:
 313         inet_frag_kill(&fq->q);
 314 err:
 315         skb_dst_drop(skb);
 316         return -EINVAL;
 317 }
 318
 319 /*
 320  *      Check if this packet is complete.
 321  *
 322  *      It is called with locked fq, and caller must check that
 323  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 324  *      the last and the first frames arrived and all the bits are here.
 325  */
 326 static int nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *skb,
 327                              struct sk_buff *prev_tail, struct net_device *dev)
 328 {
 329         void *reasm_data;
 330         int payload_len;
 331         u8 ecn;
 332
 333         inet_frag_kill(&fq->q);
 334
 335         ecn = ip_frag_ecn_table[fq->ecn];
 336         if (unlikely(ecn == 0xff))
 337                 goto err;
 338
 339         reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
 340         if (!reasm_data)
 341                 goto err;
 342
 343         payload_len = ((skb->data - skb_network_header(skb)) -
 344                        sizeof(struct ipv6hdr) + fq->q.len -
 345                        sizeof(struct frag_hdr));
 346         if (payload_len > IPV6_MAXPLEN) {
 347                 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
 348                                     payload_len);
 349                 goto err;
 350         }
 351
 352         /* We have to remove fragment header from datagram and to relocate
 353          * header in order to calculate ICV correctly. */
 354         skb_network_header(skb)[fq->nhoffset] = skb_transport_header(skb)[0];
 355         memmove(skb->head + sizeof(struct frag_hdr), skb->head,
 356                 (skb->data - skb->head) - sizeof(struct frag_hdr));
 357         skb->mac_header += sizeof(struct frag_hdr);
 358         skb->network_header += sizeof(struct frag_hdr);
 359
 360         skb_reset_transport_header(skb);
 361
 362         inet_frag_reasm_finish(&fq->q, skb, reasm_data);
 363
 364         skb->ignore_df = 1;
 365         skb->dev = dev;
 366         ipv6_hdr(skb)->payload_len = htons(payload_len);
 367         ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
 368         IP6CB(skb)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
 369
 370         /* Yes, and fold redundant checksum back. 8) */
 371         if (skb->ip_summed == CHECKSUM_COMPLETE)
 372                 skb->csum = csum_partial(skb_network_header(skb),
 373                                          skb_network_header_len(skb),
 374                                          skb->csum);
 375
 376         fq->q.fragments = NULL;
 377         fq->q.rb_fragments = RB_ROOT;
 378         fq->q.fragments_tail = NULL;
 379         fq->q.last_run_head = NULL;
 380
 381         return 0;
 382
 383 err:
 384         inet_frag_kill(&fq->q);
 385         return -EINVAL;
 386 }
 387
 388 /*
 389  * find the header just before Fragment Header.
 390  *
 391  * if success return 0 and set ...
 392  * (*prevhdrp): the value of "Next Header Field" in the header
 393  *              just before Fragment Header.
 394  * (*prevhoff): the offset of "Next Header Field" in the header
 395  *              just before Fragment Header.
 396  * (*fhoff)   : the offset of Fragment Header.
 397  *
 398  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
 399  *
 400  */
 401 static int
 402 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
 403 {
 404         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 405         const int netoff = skb_network_offset(skb);
 406         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
 407         int start = netoff + sizeof(struct ipv6hdr);
 408         int len = skb->len - start;
 409         u8 prevhdr = NEXTHDR_IPV6;
 410
 411         while (nexthdr != NEXTHDR_FRAGMENT) {
 412                 struct ipv6_opt_hdr hdr;
 413                 int hdrlen;
 414
 415                 if (!ipv6_ext_hdr(nexthdr)) {
 416                         return -1;
 417                 }
 418                 if (nexthdr == NEXTHDR_NONE) {
 419                         pr_debug("next header is none\n");
 420                         return -1;
 421                 }
 422                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
 423                         pr_debug("too short\n");
 424                         return -1;
 425                 }
 426                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
 427                         BUG();
 428                 if (nexthdr == NEXTHDR_AUTH)
 429                         hdrlen = (hdr.hdrlen+2)<<2;
 430                 else
 431                         hdrlen = ipv6_optlen(&hdr);
 432
 433                 prevhdr = nexthdr;
 434                 prev_nhoff = start;
 435
 436                 nexthdr = hdr.nexthdr;
 437                 len -= hdrlen;
 438                 start += hdrlen;
 439         }
 440
 441         if (len < 0)
 442                 return -1;
 443
 444         *prevhdrp = prevhdr;
 445         *prevhoff = prev_nhoff;
 446         *fhoff = start;
 447
 448         return 0;
 449 }
 450
 451 int nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
 452 {
 453         u16 savethdr = skb->transport_header;
 454         int fhoff, nhoff, ret;
 455         struct frag_hdr *fhdr;
 456         struct frag_queue *fq;
 457         struct ipv6hdr *hdr;
 458         u8 prevhdr;
 459
 460         /* Jumbo payload inhibits frag. header */
 461         if (ipv6_hdr(skb)->payload_len == 0) {
 462                 pr_debug("payload len = 0\n");
 463                 return 0;
 464         }
 465
 466         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
 467                 return 0;
 468
 469         if (!pskb_may_pull(skb, fhoff + sizeof(*fhdr)))
 470                 return -ENOMEM;
 471
 472         skb_set_transport_header(skb, fhoff);
 473         hdr = ipv6_hdr(skb);
 474         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 475
 476         skb_orphan(skb);
 477         fq = fq_find(net, fhdr->identification, user, hdr,
 478                      skb->dev ? skb->dev->ifindex : 0);
 479         if (fq == NULL) {
 480                 pr_debug("Can't find and can't create new queue\n");
 481                 return -ENOMEM;
 482         }
 483
 484         spin_lock_bh(&fq->q.lock);
 485
 486         ret = nf_ct_frag6_queue(fq, skb, fhdr, nhoff);
 487         if (ret == -EPROTO) {
 488                 skb->transport_header = savethdr;
 489                 ret = 0;
 490         }
 491
 492         spin_unlock_bh(&fq->q.lock);
 493         inet_frag_put(&fq->q);
 494         return ret;
 495 }
 496 EXPORT_SYMBOL_GPL(nf_ct_frag6_gather);
 497
 498 static int nf_ct_net_init(struct net *net)
 499 {
 500         int res;
 501
 502         net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
 503         net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
 504         net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
 505         net->nf_frag.frags.f = &nf_frags;
 506
 507         res = inet_frags_init_net(&net->nf_frag.frags);
 508         if (res < 0)
 509                 return res;
 510         res = nf_ct_frag6_sysctl_register(net);
 511         if (res < 0)
 512                 inet_frags_exit_net(&net->nf_frag.frags);
 513         return res;
 514 }
 515
 516 static void nf_ct_net_exit(struct net *net)
 517 {
 518         nf_ct_frags6_sysctl_unregister(net);
 519         inet_frags_exit_net(&net->nf_frag.frags);
 520 }
 521
 522 static struct pernet_operations nf_ct_net_ops = {
 523         .init = nf_ct_net_init,
 524         .exit = nf_ct_net_exit,
 525 };
 526
 527 static const struct rhashtable_params nfct_rhash_params = {
 528         .head_offset            = offsetof(struct inet_frag_queue, node),
 529         .hashfn                 = ip6frag_key_hashfn,
 530         .obj_hashfn             = ip6frag_obj_hashfn,
 531         .obj_cmpfn              = ip6frag_obj_cmpfn,
 532         .automatic_shrinking    = true,
 533 };
 534
 535 int nf_ct_frag6_init(void)
 536 {
 537         int ret = 0;
 538
 539         nf_frags.constructor = ip6frag_init;
 540         nf_frags.destructor = NULL;
 541         nf_frags.qsize = sizeof(struct frag_queue);
 542         nf_frags.frag_expire = nf_ct_frag6_expire;
 543         nf_frags.frags_cache_name = nf_frags_cache_name;
 544         nf_frags.rhash_params = nfct_rhash_params;
 545         ret = inet_frags_init(&nf_frags);
 546         if (ret)
 547                 goto out;
 548         ret = register_pernet_subsys(&nf_ct_net_ops);
 549         if (ret)
 550                 inet_frags_fini(&nf_frags);
 551
 552 out:
 553         return ret;
 554 }
 555
 556 void nf_ct_frag6_cleanup(void)
 557 {
 558         unregister_pernet_subsys(&nf_ct_net_ops);
 559         inet_frags_fini(&nf_frags);
 560 }