net/ipv6/reassembly.c

   1 /*
   2  *      IPv6 fragment reassembly
   3  *      Linux INET6 implementation
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      Based on: net/ipv4/ip_fragment.c
   9  *
  10  *      This program is free software; you can redistribute it and/or
  11  *      modify it under the terms of the GNU General Public License
  12  *      as published by the Free Software Foundation; either version
  13  *      2 of the License, or (at your option) any later version.
  14  */
  15
  16 /*
  17  *      Fixes:
  18  *      Andi Kleen      Make it work with multiple hosts.
  19  *                      More RFC compliance.
  20  *
  21  *      Horst von Brand Add missing #include <linux/string.h>
  22  *      Alexey Kuznetsov        SMP races, threading, cleanup.
  23  *      Patrick McHardy         LRU queue of frag heads for evictor.
  24  *      Mitsuru KANDA @USAGI    Register inet6_protocol{}.
  25  *      David Stevens and
  26  *      YOSHIFUJI,H. @USAGI     Always remove fragment header to
  27  *                              calculate ICV correctly.
  28  */
  29 #include <linux/errno.h>
  30 #include <linux/types.h>
  31 #include <linux/string.h>
  32 #include <linux/socket.h>
  33 #include <linux/sockios.h>
  34 #include <linux/jiffies.h>
  35 #include <linux/net.h>
  36 #include <linux/list.h>
  37 #include <linux/netdevice.h>
  38 #include <linux/in6.h>
  39 #include <linux/ipv6.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/random.h>
  42 #include <linux/jhash.h>
  43 #include <linux/skbuff.h>
  44 #include <linux/slab.h>
  45
  46 #include <net/sock.h>
  47 #include <net/snmp.h>
  48
  49 #include <net/ipv6.h>
  50 #include <net/ip6_route.h>
  51 #include <net/protocol.h>
  52 #include <net/transp_v6.h>
  53 #include <net/rawv6.h>
  54 #include <net/ndisc.h>
  55 #include <net/addrconf.h>
  56 #include <net/inet_frag.h>
  57
  58 struct ip6frag_skb_cb
  59 {
  60         struct inet6_skb_parm   h;
  61         int                     offset;
  62 };
  63
  64 #define FRAG6_CB(skb)   ((struct ip6frag_skb_cb*)((skb)->cb))
  65
  66
  67 /*
  68  *      Equivalent of ipv4 struct ipq
  69  */
  70
  71 struct frag_queue
  72 {
  73         struct inet_frag_queue  q;
  74
  75         __be32                  id;             /* fragment id          */
  76         u32                     user;
  77         struct in6_addr         saddr;
  78         struct in6_addr         daddr;
  79
  80         int                     iif;
  81         unsigned int            csum;
  82         __u16                   nhoffset;
  83 };
  84
  85 static struct inet_frags ip6_frags;
  86
  87 int ip6_frag_nqueues(struct net *net)
  88 {
  89         return net->ipv6.frags.nqueues;
  90 }
  91
  92 int ip6_frag_mem(struct net *net)
  93 {
  94         return atomic_read(&net->ipv6.frags.mem);
  95 }
  96
  97 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
  98                           struct net_device *dev);
  99
 100 /*
 101  * callers should be careful not to use the hash value outside the ipfrag_lock
 102  * as doing so could race with ipfrag_hash_rnd being recalculated.
 103  */
 104 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
 105                              const struct in6_addr *daddr, u32 rnd)
 106 {
 107         u32 a, b, c;
 108
 109         a = (__force u32)saddr->s6_addr32[0];
 110         b = (__force u32)saddr->s6_addr32[1];
 111         c = (__force u32)saddr->s6_addr32[2];
 112
 113         a += JHASH_GOLDEN_RATIO;
 114         b += JHASH_GOLDEN_RATIO;
 115         c += rnd;
 116         __jhash_mix(a, b, c);
 117
 118         a += (__force u32)saddr->s6_addr32[3];
 119         b += (__force u32)daddr->s6_addr32[0];
 120         c += (__force u32)daddr->s6_addr32[1];
 121         __jhash_mix(a, b, c);
 122
 123         a += (__force u32)daddr->s6_addr32[2];
 124         b += (__force u32)daddr->s6_addr32[3];
 125         c += (__force u32)id;
 126         __jhash_mix(a, b, c);
 127
 128         return c & (INETFRAGS_HASHSZ - 1);
 129 }
 130 EXPORT_SYMBOL_GPL(inet6_hash_frag);
 131
 132 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 133 {
 134         struct frag_queue *fq;
 135
 136         fq = container_of(q, struct frag_queue, q);
 137         return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
 138 }
 139
 140 int ip6_frag_match(struct inet_frag_queue *q, void *a)
 141 {
 142         struct frag_queue *fq;
 143         struct ip6_create_arg *arg = a;
 144
 145         fq = container_of(q, struct frag_queue, q);
 146         return (fq->id == arg->id && fq->user == arg->user &&
 147                         ipv6_addr_equal(&fq->saddr, arg->src) &&
 148                         ipv6_addr_equal(&fq->daddr, arg->dst));
 149 }
 150 EXPORT_SYMBOL(ip6_frag_match);
 151
 152 /* Memory Tracking Functions. */
 153 static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
 154 {
 155         atomic_sub(skb->truesize, &nf->mem);
 156         kfree_skb(skb);
 157 }
 158
 159 void ip6_frag_init(struct inet_frag_queue *q, void *a)
 160 {
 161         struct frag_queue *fq = container_of(q, struct frag_queue, q);
 162         struct ip6_create_arg *arg = a;
 163
 164         fq->id = arg->id;
 165         fq->user = arg->user;
 166         ipv6_addr_copy(&fq->saddr, arg->src);
 167         ipv6_addr_copy(&fq->daddr, arg->dst);
 168 }
 169 EXPORT_SYMBOL(ip6_frag_init);
 170
 171 /* Destruction primitives. */
 172
 173 static __inline__ void fq_put(struct frag_queue *fq)
 174 {
 175         inet_frag_put(&fq->q, &ip6_frags);
 176 }
 177
 178 /* Kill fq entry. It is not destroyed immediately,
 179  * because caller (and someone more) holds reference count.
 180  */
 181 static __inline__ void fq_kill(struct frag_queue *fq)
 182 {
 183         inet_frag_kill(&fq->q, &ip6_frags);
 184 }
 185
 186 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 187 {
 188         int evicted;
 189
 190         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
 191         if (evicted)
 192                 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
 193 }
 194
 195 static void ip6_frag_expire(unsigned long data)
 196 {
 197         struct frag_queue *fq;
 198         struct net_device *dev = NULL;
 199         struct net *net;
 200
 201         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 202
 203         spin_lock(&fq->q.lock);
 204
 205         if (fq->q.last_in & INET_FRAG_COMPLETE)
 206                 goto out;
 207
 208         fq_kill(fq);
 209
 210         net = container_of(fq->q.net, struct net, ipv6.frags);
 211         rcu_read_lock();
 212         dev = dev_get_by_index_rcu(net, fq->iif);
 213         if (!dev)
 214                 goto out_rcu_unlock;
 215
 216         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 217         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 218
 219         /* Don't send error if the first segment did not arrive. */
 220         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
 221                 goto out_rcu_unlock;
 222
 223         /*
 224            But use as source device on which LAST ARRIVED
 225            segment was received. And do not use fq->dev
 226            pointer directly, device might already disappeared.
 227          */
 228         fq->q.fragments->dev = dev;
 229         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
 230 out_rcu_unlock:
 231         rcu_read_unlock();
 232 out:
 233         spin_unlock(&fq->q.lock);
 234         fq_put(fq);
 235 }
 236
 237 static __inline__ struct frag_queue *
 238 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst)
 239 {
 240         struct inet_frag_queue *q;
 241         struct ip6_create_arg arg;
 242         unsigned int hash;
 243
 244         arg.id = id;
 245         arg.user = IP6_DEFRAG_LOCAL_DELIVER;
 246         arg.src = src;
 247         arg.dst = dst;
 248
 249         read_lock(&ip6_frags.lock);
 250         hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
 251
 252         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
 253         if (q == NULL)
 254                 return NULL;
 255
 256         return container_of(q, struct frag_queue, q);
 257 }
 258
 259 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 260                            struct frag_hdr *fhdr, int nhoff)
 261 {
 262         struct sk_buff *prev, *next;
 263         struct net_device *dev;
 264         int offset, end;
 265         struct net *net = dev_net(skb_dst(skb)->dev);
 266
 267         if (fq->q.last_in & INET_FRAG_COMPLETE)
 268                 goto err;
 269
 270         offset = ntohs(fhdr->frag_off) & ~0x7;
 271         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 272                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 273
 274         if ((unsigned int)end > IPV6_MAXPLEN) {
 275                 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 276                                  IPSTATS_MIB_INHDRERRORS);
 277                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 278                                   ((u8 *)&fhdr->frag_off -
 279                                    skb_network_header(skb)));
 280                 return -1;
 281         }
 282
 283         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 284                 const unsigned char *nh = skb_network_header(skb);
 285                 skb->csum = csum_sub(skb->csum,
 286                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 287                                                   0));
 288         }
 289
 290         /* Is this the final fragment? */
 291         if (!(fhdr->frag_off & htons(IP6_MF))) {
 292                 /* If we already have some bits beyond end
 293                  * or have different end, the segment is corrupted.
 294                  */
 295                 if (end < fq->q.len ||
 296                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
 297                         goto err;
 298                 fq->q.last_in |= INET_FRAG_LAST_IN;
 299                 fq->q.len = end;
 300         } else {
 301                 /* Check if the fragment is rounded to 8 bytes.
 302                  * Required by the RFC.
 303                  */
 304                 if (end & 0x7) {
 305                         /* RFC2460 says always send parameter problem in
 306                          * this case. -DaveM
 307                          */
 308                         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 309                                          IPSTATS_MIB_INHDRERRORS);
 310                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 311                                           offsetof(struct ipv6hdr, payload_len));
 312                         return -1;
 313                 }
 314                 if (end > fq->q.len) {
 315                         /* Some bits beyond end -> corruption. */
 316                         if (fq->q.last_in & INET_FRAG_LAST_IN)
 317                                 goto err;
 318                         fq->q.len = end;
 319                 }
 320         }
 321
 322         if (end == offset)
 323                 goto err;
 324
 325         /* Point into the IP datagram 'data' part. */
 326         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 327                 goto err;
 328
 329         if (pskb_trim_rcsum(skb, end - offset))
 330                 goto err;
 331
 332         /* Find out which fragments are in front and at the back of us
 333          * in the chain of fragments so far.  We must know where to put
 334          * this fragment, right?
 335          */
 336         prev = fq->q.fragments_tail;
 337         if (!prev || FRAG6_CB(prev)->offset < offset) {
 338                 next = NULL;
 339                 goto found;
 340         }
 341         prev = NULL;
 342         for(next = fq->q.fragments; next != NULL; next = next->next) {
 343                 if (FRAG6_CB(next)->offset >= offset)
 344                         break;  /* bingo! */
 345                 prev = next;
 346         }
 347
 348 found:
 349         /* We found where to put this one.  Check for overlap with
 350          * preceding fragment, and, if needed, align things so that
 351          * any overlaps are eliminated.
 352          */
 353         if (prev) {
 354                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
 355
 356                 if (i > 0) {
 357                         offset += i;
 358                         if (end <= offset)
 359                                 goto err;
 360                         if (!pskb_pull(skb, i))
 361                                 goto err;
 362                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 363                                 skb->ip_summed = CHECKSUM_NONE;
 364                 }
 365         }
 366
 367         /* Look for overlap with succeeding segments.
 368          * If we can merge fragments, do it.
 369          */
 370         while (next && FRAG6_CB(next)->offset < end) {
 371                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
 372
 373                 if (i < next->len) {
 374                         /* Eat head of the next overlapped fragment
 375                          * and leave the loop. The next ones cannot overlap.
 376                          */
 377                         if (!pskb_pull(next, i))
 378                                 goto err;
 379                         FRAG6_CB(next)->offset += i;    /* next fragment */
 380                         fq->q.meat -= i;
 381                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 382                                 next->ip_summed = CHECKSUM_NONE;
 383                         break;
 384                 } else {
 385                         struct sk_buff *free_it = next;
 386
 387                         /* Old fragment is completely overridden with
 388                          * new one drop it.
 389                          */
 390                         next = next->next;
 391
 392                         if (prev)
 393                                 prev->next = next;
 394                         else
 395                                 fq->q.fragments = next;
 396
 397                         fq->q.meat -= free_it->len;
 398                         frag_kfree_skb(fq->q.net, free_it);
 399                 }
 400         }
 401
 402         FRAG6_CB(skb)->offset = offset;
 403
 404         /* Insert this fragment in the chain of fragments. */
 405         skb->next = next;
 406         if (!next)
 407                 fq->q.fragments_tail = skb;
 408         if (prev)
 409                 prev->next = skb;
 410         else
 411                 fq->q.fragments = skb;
 412
 413         dev = skb->dev;
 414         if (dev) {
 415                 fq->iif = dev->ifindex;
 416                 skb->dev = NULL;
 417         }
 418         fq->q.stamp = skb->tstamp;
 419         fq->q.meat += skb->len;
 420         atomic_add(skb->truesize, &fq->q.net->mem);
 421
 422         /* The first fragment.
 423          * nhoffset is obtained from the first fragment, of course.
 424          */
 425         if (offset == 0) {
 426                 fq->nhoffset = nhoff;
 427                 fq->q.last_in |= INET_FRAG_FIRST_IN;
 428         }
 429
 430         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 431             fq->q.meat == fq->q.len)
 432                 return ip6_frag_reasm(fq, prev, dev);
 433
 434         write_lock(&ip6_frags.lock);
 435         list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
 436         write_unlock(&ip6_frags.lock);
 437         return -1;
 438
 439 err:
 440         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 441                       IPSTATS_MIB_REASMFAILS);
 442         kfree_skb(skb);
 443         return -1;
 444 }
 445
 446 /*
 447  *      Check if this packet is complete.
 448  *      Returns NULL on failure by any reason, and pointer
 449  *      to current nexthdr field in reassembled frame.
 450  *
 451  *      It is called with locked fq, and caller must check that
 452  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 453  *      the last and the first frames arrived and all the bits are here.
 454  */
 455 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 456                           struct net_device *dev)
 457 {
 458         struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
 459         struct sk_buff *fp, *head = fq->q.fragments;
 460         int    payload_len;
 461         unsigned int nhoff;
 462
 463         fq_kill(fq);
 464
 465         /* Make the one we just received the head. */
 466         if (prev) {
 467                 head = prev->next;
 468                 fp = skb_clone(head, GFP_ATOMIC);
 469
 470                 if (!fp)
 471                         goto out_oom;
 472
 473                 fp->next = head->next;
 474                 if (!fp->next)
 475                         fq->q.fragments_tail = fp;
 476                 prev->next = fp;
 477
 478                 skb_morph(head, fq->q.fragments);
 479                 head->next = fq->q.fragments->next;
 480
 481                 kfree_skb(fq->q.fragments);
 482                 fq->q.fragments = head;
 483         }
 484
 485         WARN_ON(head == NULL);
 486         WARN_ON(FRAG6_CB(head)->offset != 0);
 487
 488         /* Unfragmented part is taken from the first segment. */
 489         payload_len = ((head->data - skb_network_header(head)) -
 490                        sizeof(struct ipv6hdr) + fq->q.len -
 491                        sizeof(struct frag_hdr));
 492         if (payload_len > IPV6_MAXPLEN)
 493                 goto out_oversize;
 494
 495         /* Head of list must not be cloned. */
 496         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 497                 goto out_oom;
 498
 499         /* If the first fragment is fragmented itself, we split
 500          * it to two chunks: the first with data and paged part
 501          * and the second, holding only fragments. */
 502         if (skb_has_frags(head)) {
 503                 struct sk_buff *clone;
 504                 int i, plen = 0;
 505
 506                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 507                         goto out_oom;
 508                 clone->next = head->next;
 509                 head->next = clone;
 510                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 511                 skb_frag_list_init(head);
 512                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 513                         plen += skb_shinfo(head)->frags[i].size;
 514                 clone->len = clone->data_len = head->data_len - plen;
 515                 head->data_len -= clone->len;
 516                 head->len -= clone->len;
 517                 clone->csum = 0;
 518                 clone->ip_summed = head->ip_summed;
 519                 atomic_add(clone->truesize, &fq->q.net->mem);
 520         }
 521
 522         /* We have to remove fragment header from datagram and to relocate
 523          * header in order to calculate ICV correctly. */
 524         nhoff = fq->nhoffset;
 525         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 526         memmove(head->head + sizeof(struct frag_hdr), head->head,
 527                 (head->data - head->head) - sizeof(struct frag_hdr));
 528         head->mac_header += sizeof(struct frag_hdr);
 529         head->network_header += sizeof(struct frag_hdr);
 530
 531         skb_shinfo(head)->frag_list = head->next;
 532         skb_reset_transport_header(head);
 533         skb_push(head, head->data - skb_network_header(head));
 534
 535         for (fp=head->next; fp; fp = fp->next) {
 536                 head->data_len += fp->len;
 537                 head->len += fp->len;
 538                 if (head->ip_summed != fp->ip_summed)
 539                         head->ip_summed = CHECKSUM_NONE;
 540                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 541                         head->csum = csum_add(head->csum, fp->csum);
 542                 head->truesize += fp->truesize;
 543         }
 544         atomic_sub(head->truesize, &fq->q.net->mem);
 545
 546         head->next = NULL;
 547         head->dev = dev;
 548         head->tstamp = fq->q.stamp;
 549         ipv6_hdr(head)->payload_len = htons(payload_len);
 550         IP6CB(head)->nhoff = nhoff;
 551
 552         /* Yes, and fold redundant checksum back. 8) */
 553         if (head->ip_summed == CHECKSUM_COMPLETE)
 554                 head->csum = csum_partial(skb_network_header(head),
 555                                           skb_network_header_len(head),
 556                                           head->csum);
 557
 558         rcu_read_lock();
 559         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 560         rcu_read_unlock();
 561         fq->q.fragments = NULL;
 562         fq->q.fragments_tail = NULL;
 563         return 1;
 564
 565 out_oversize:
 566         if (net_ratelimit())
 567                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
 568         goto out_fail;
 569 out_oom:
 570         if (net_ratelimit())
 571                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
 572 out_fail:
 573         rcu_read_lock();
 574         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 575         rcu_read_unlock();
 576         return -1;
 577 }
 578
 579 static int ipv6_frag_rcv(struct sk_buff *skb)
 580 {
 581         struct frag_hdr *fhdr;
 582         struct frag_queue *fq;
 583         struct ipv6hdr *hdr = ipv6_hdr(skb);
 584         struct net *net = dev_net(skb_dst(skb)->dev);
 585
 586         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
 587
 588         /* Jumbo payload inhibits frag. header */
 589         if (hdr->payload_len==0)
 590                 goto fail_hdr;
 591
 592         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 593                                  sizeof(struct frag_hdr))))
 594                 goto fail_hdr;
 595
 596         hdr = ipv6_hdr(skb);
 597         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 598
 599         if (!(fhdr->frag_off & htons(0xFFF9))) {
 600                 /* It is not a fragmented frame */
 601                 skb->transport_header += sizeof(struct frag_hdr);
 602                 IP6_INC_STATS_BH(net,
 603                                  ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
 604
 605                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 606                 return 1;
 607         }
 608
 609         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
 610                 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
 611
 612         fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
 613         if (fq != NULL) {
 614                 int ret;
 615
 616                 spin_lock(&fq->q.lock);
 617
 618                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 619
 620                 spin_unlock(&fq->q.lock);
 621                 fq_put(fq);
 622                 return ret;
 623         }
 624
 625         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
 626         kfree_skb(skb);
 627         return -1;
 628
 629 fail_hdr:
 630         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
 631         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
 632         return -1;
 633 }
 634
 635 static const struct inet6_protocol frag_protocol =
 636 {
 637         .handler        =       ipv6_frag_rcv,
 638         .flags          =       INET6_PROTO_NOPOLICY,
 639 };
 640
 641 #ifdef CONFIG_SYSCTL
 642 static struct ctl_table ip6_frags_ns_ctl_table[] = {
 643         {
 644                 .procname       = "ip6frag_high_thresh",
 645                 .data           = &init_net.ipv6.frags.high_thresh,
 646                 .maxlen         = sizeof(int),
 647                 .mode           = 0644,
 648                 .proc_handler   = proc_dointvec
 649         },
 650         {
 651                 .procname       = "ip6frag_low_thresh",
 652                 .data           = &init_net.ipv6.frags.low_thresh,
 653                 .maxlen         = sizeof(int),
 654                 .mode           = 0644,
 655                 .proc_handler   = proc_dointvec
 656         },
 657         {
 658                 .procname       = "ip6frag_time",
 659                 .data           = &init_net.ipv6.frags.timeout,
 660                 .maxlen         = sizeof(int),
 661                 .mode           = 0644,
 662                 .proc_handler   = proc_dointvec_jiffies,
 663         },
 664         { }
 665 };
 666
 667 static struct ctl_table ip6_frags_ctl_table[] = {
 668         {
 669                 .procname       = "ip6frag_secret_interval",
 670                 .data           = &ip6_frags.secret_interval,
 671                 .maxlen         = sizeof(int),
 672                 .mode           = 0644,
 673                 .proc_handler   = proc_dointvec_jiffies,
 674         },
 675         { }
 676 };
 677
 678 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
 679 {
 680         struct ctl_table *table;
 681         struct ctl_table_header *hdr;
 682
 683         table = ip6_frags_ns_ctl_table;
 684         if (!net_eq(net, &init_net)) {
 685                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
 686                 if (table == NULL)
 687                         goto err_alloc;
 688
 689                 table[0].data = &net->ipv6.frags.high_thresh;
 690                 table[1].data = &net->ipv6.frags.low_thresh;
 691                 table[2].data = &net->ipv6.frags.timeout;
 692         }
 693
 694         hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
 695         if (hdr == NULL)
 696                 goto err_reg;
 697
 698         net->ipv6.sysctl.frags_hdr = hdr;
 699         return 0;
 700
 701 err_reg:
 702         if (!net_eq(net, &init_net))
 703                 kfree(table);
 704 err_alloc:
 705         return -ENOMEM;
 706 }
 707
 708 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
 709 {
 710         struct ctl_table *table;
 711
 712         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
 713         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
 714         if (!net_eq(net, &init_net))
 715                 kfree(table);
 716 }
 717
 718 static struct ctl_table_header *ip6_ctl_header;
 719
 720 static int ip6_frags_sysctl_register(void)
 721 {
 722         ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
 723                         ip6_frags_ctl_table);
 724         return ip6_ctl_header == NULL ? -ENOMEM : 0;
 725 }
 726
 727 static void ip6_frags_sysctl_unregister(void)
 728 {
 729         unregister_net_sysctl_table(ip6_ctl_header);
 730 }
 731 #else
 732 static inline int ip6_frags_ns_sysctl_register(struct net *net)
 733 {
 734         return 0;
 735 }
 736
 737 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
 738 {
 739 }
 740
 741 static inline int ip6_frags_sysctl_register(void)
 742 {
 743         return 0;
 744 }
 745
 746 static inline void ip6_frags_sysctl_unregister(void)
 747 {
 748 }
 749 #endif
 750
 751 static int __net_init ipv6_frags_init_net(struct net *net)
 752 {
 753         net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
 754         net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
 755         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
 756
 757         inet_frags_init_net(&net->ipv6.frags);
 758
 759         return ip6_frags_ns_sysctl_register(net);
 760 }
 761
 762 static void __net_exit ipv6_frags_exit_net(struct net *net)
 763 {
 764         ip6_frags_ns_sysctl_unregister(net);
 765         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 766 }
 767
 768 static struct pernet_operations ip6_frags_ops = {
 769         .init = ipv6_frags_init_net,
 770         .exit = ipv6_frags_exit_net,
 771 };
 772
 773 int __init ipv6_frag_init(void)
 774 {
 775         int ret;
 776
 777         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 778         if (ret)
 779                 goto out;
 780
 781         ret = ip6_frags_sysctl_register();
 782         if (ret)
 783                 goto err_sysctl;
 784
 785         ret = register_pernet_subsys(&ip6_frags_ops);
 786         if (ret)
 787                 goto err_pernet;
 788
 789         ip6_frags.hashfn = ip6_hashfn;
 790         ip6_frags.constructor = ip6_frag_init;
 791         ip6_frags.destructor = NULL;
 792         ip6_frags.skb_free = NULL;
 793         ip6_frags.qsize = sizeof(struct frag_queue);
 794         ip6_frags.match = ip6_frag_match;
 795         ip6_frags.frag_expire = ip6_frag_expire;
 796         ip6_frags.secret_interval = 10 * 60 * HZ;
 797         inet_frags_init(&ip6_frags);
 798 out:
 799         return ret;
 800
 801 err_pernet:
 802         ip6_frags_sysctl_unregister();
 803 err_sysctl:
 804         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 805         goto out;
 806 }
 807
 808 void ipv6_frag_exit(void)
 809 {
 810         inet_frags_fini(&ip6_frags);
 811         ip6_frags_sysctl_unregister();
 812         unregister_pernet_subsys(&ip6_frags_ops);
 813         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 814 }