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/reserve.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 inline void frag_kfree_skb(struct netns_frags *nf,
 154                 struct sk_buff *skb, int *work)
 155 {
 156         if (work)
 157                 *work -= skb->truesize;
 158         atomic_sub(skb->truesize, &nf->mem);
 159         kfree_skb(skb);
 160 }
 161
 162 void ip6_frag_init(struct inet_frag_queue *q, void *a)
 163 {
 164         struct frag_queue *fq = container_of(q, struct frag_queue, q);
 165         struct ip6_create_arg *arg = a;
 166
 167         fq->id = arg->id;
 168         fq->user = arg->user;
 169         ipv6_addr_copy(&fq->saddr, arg->src);
 170         ipv6_addr_copy(&fq->daddr, arg->dst);
 171 }
 172 EXPORT_SYMBOL(ip6_frag_init);
 173
 174 /* Destruction primitives. */
 175
 176 static __inline__ void fq_put(struct frag_queue *fq)
 177 {
 178         inet_frag_put(&fq->q, &ip6_frags);
 179 }
 180
 181 /* Kill fq entry. It is not destroyed immediately,
 182  * because caller (and someone more) holds reference count.
 183  */
 184 static __inline__ void fq_kill(struct frag_queue *fq)
 185 {
 186         inet_frag_kill(&fq->q, &ip6_frags);
 187 }
 188
 189 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 190 {
 191         int evicted;
 192
 193         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
 194         if (evicted)
 195                 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
 196 }
 197
 198 static void ip6_frag_expire(unsigned long data)
 199 {
 200         struct frag_queue *fq;
 201         struct net_device *dev = NULL;
 202         struct net *net;
 203
 204         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 205
 206         spin_lock(&fq->q.lock);
 207
 208         if (fq->q.last_in & INET_FRAG_COMPLETE)
 209                 goto out;
 210
 211         fq_kill(fq);
 212
 213         net = container_of(fq->q.net, struct net, ipv6.frags);
 214         rcu_read_lock();
 215         dev = dev_get_by_index_rcu(net, fq->iif);
 216         if (!dev)
 217                 goto out_rcu_unlock;
 218
 219         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 220         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 221
 222         /* Don't send error if the first segment did not arrive. */
 223         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
 224                 goto out_rcu_unlock;
 225
 226         /*
 227            But use as source device on which LAST ARRIVED
 228            segment was received. And do not use fq->dev
 229            pointer directly, device might already disappeared.
 230          */
 231         fq->q.fragments->dev = dev;
 232         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
 233 out_rcu_unlock:
 234         rcu_read_unlock();
 235 out:
 236         spin_unlock(&fq->q.lock);
 237         fq_put(fq);
 238 }
 239
 240 static __inline__ struct frag_queue *
 241 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
 242         struct inet6_dev *idev)
 243 {
 244         struct inet_frag_queue *q;
 245         struct ip6_create_arg arg;
 246         unsigned int hash;
 247
 248         arg.id = id;
 249         arg.user = IP6_DEFRAG_LOCAL_DELIVER;
 250         arg.src = src;
 251         arg.dst = dst;
 252
 253         read_lock(&ip6_frags.lock);
 254         hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
 255
 256         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
 257         if (q == NULL)
 258                 goto oom;
 259
 260         return container_of(q, struct frag_queue, q);
 261
 262 oom:
 263         IP6_INC_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS);
 264         return NULL;
 265 }
 266
 267 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 268                            struct frag_hdr *fhdr, int nhoff)
 269 {
 270         struct sk_buff *prev, *next;
 271         struct net_device *dev;
 272         int offset, end;
 273         struct net *net = dev_net(skb_dst(skb)->dev);
 274
 275         if (fq->q.last_in & INET_FRAG_COMPLETE)
 276                 goto err;
 277
 278         offset = ntohs(fhdr->frag_off) & ~0x7;
 279         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 280                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 281
 282         if ((unsigned int)end > IPV6_MAXPLEN) {
 283                 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 284                                  IPSTATS_MIB_INHDRERRORS);
 285                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 286                                   ((u8 *)&fhdr->frag_off -
 287                                    skb_network_header(skb)));
 288                 return -1;
 289         }
 290
 291         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 292                 const unsigned char *nh = skb_network_header(skb);
 293                 skb->csum = csum_sub(skb->csum,
 294                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 295                                                   0));
 296         }
 297
 298         /* Is this the final fragment? */
 299         if (!(fhdr->frag_off & htons(IP6_MF))) {
 300                 /* If we already have some bits beyond end
 301                  * or have different end, the segment is corrupted.
 302                  */
 303                 if (end < fq->q.len ||
 304                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
 305                         goto err;
 306                 fq->q.last_in |= INET_FRAG_LAST_IN;
 307                 fq->q.len = end;
 308         } else {
 309                 /* Check if the fragment is rounded to 8 bytes.
 310                  * Required by the RFC.
 311                  */
 312                 if (end & 0x7) {
 313                         /* RFC2460 says always send parameter problem in
 314                          * this case. -DaveM
 315                          */
 316                         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 317                                          IPSTATS_MIB_INHDRERRORS);
 318                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 319                                           offsetof(struct ipv6hdr, payload_len));
 320                         return -1;
 321                 }
 322                 if (end > fq->q.len) {
 323                         /* Some bits beyond end -> corruption. */
 324                         if (fq->q.last_in & INET_FRAG_LAST_IN)
 325                                 goto err;
 326                         fq->q.len = end;
 327                 }
 328         }
 329
 330         if (end == offset)
 331                 goto err;
 332
 333         /* Point into the IP datagram 'data' part. */
 334         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 335                 goto err;
 336
 337         if (pskb_trim_rcsum(skb, end - offset))
 338                 goto err;
 339
 340         /* Find out which fragments are in front and at the back of us
 341          * in the chain of fragments so far.  We must know where to put
 342          * this fragment, right?
 343          */
 344         prev = NULL;
 345         for(next = fq->q.fragments; next != NULL; next = next->next) {
 346                 if (FRAG6_CB(next)->offset >= offset)
 347                         break;  /* bingo! */
 348                 prev = next;
 349         }
 350
 351         /* We found where to put this one.  Check for overlap with
 352          * preceding fragment, and, if needed, align things so that
 353          * any overlaps are eliminated.
 354          */
 355         if (prev) {
 356                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
 357
 358                 if (i > 0) {
 359                         offset += i;
 360                         if (end <= offset)
 361                                 goto err;
 362                         if (!pskb_pull(skb, i))
 363                                 goto err;
 364                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 365                                 skb->ip_summed = CHECKSUM_NONE;
 366                 }
 367         }
 368
 369         /* Look for overlap with succeeding segments.
 370          * If we can merge fragments, do it.
 371          */
 372         while (next && FRAG6_CB(next)->offset < end) {
 373                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
 374
 375                 if (i < next->len) {
 376                         /* Eat head of the next overlapped fragment
 377                          * and leave the loop. The next ones cannot overlap.
 378                          */
 379                         if (!pskb_pull(next, i))
 380                                 goto err;
 381                         FRAG6_CB(next)->offset += i;    /* next fragment */
 382                         fq->q.meat -= i;
 383                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 384                                 next->ip_summed = CHECKSUM_NONE;
 385                         break;
 386                 } else {
 387                         struct sk_buff *free_it = next;
 388
 389                         /* Old fragment is completely overridden with
 390                          * new one drop it.
 391                          */
 392                         next = next->next;
 393
 394                         if (prev)
 395                                 prev->next = next;
 396                         else
 397                                 fq->q.fragments = next;
 398
 399                         fq->q.meat -= free_it->len;
 400                         frag_kfree_skb(fq->q.net, free_it, NULL);
 401                 }
 402         }
 403
 404         FRAG6_CB(skb)->offset = offset;
 405
 406         /* Insert this fragment in the chain of fragments. */
 407         skb->next = next;
 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                 prev->next = fp;
 475
 476                 skb_morph(head, fq->q.fragments);
 477                 head->next = fq->q.fragments->next;
 478
 479                 kfree_skb(fq->q.fragments);
 480                 fq->q.fragments = head;
 481         }
 482
 483         WARN_ON(head == NULL);
 484         WARN_ON(FRAG6_CB(head)->offset != 0);
 485
 486         /* Unfragmented part is taken from the first segment. */
 487         payload_len = ((head->data - skb_network_header(head)) -
 488                        sizeof(struct ipv6hdr) + fq->q.len -
 489                        sizeof(struct frag_hdr));
 490         if (payload_len > IPV6_MAXPLEN)
 491                 goto out_oversize;
 492
 493         /* Head of list must not be cloned. */
 494         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 495                 goto out_oom;
 496
 497         /* If the first fragment is fragmented itself, we split
 498          * it to two chunks: the first with data and paged part
 499          * and the second, holding only fragments. */
 500         if (skb_has_frags(head)) {
 501                 struct sk_buff *clone;
 502                 int i, plen = 0;
 503
 504                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 505                         goto out_oom;
 506                 clone->next = head->next;
 507                 head->next = clone;
 508                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 509                 skb_frag_list_init(head);
 510                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 511                         plen += skb_shinfo(head)->frags[i].size;
 512                 clone->len = clone->data_len = head->data_len - plen;
 513                 head->data_len -= clone->len;
 514                 head->len -= clone->len;
 515                 clone->csum = 0;
 516                 clone->ip_summed = head->ip_summed;
 517                 atomic_add(clone->truesize, &fq->q.net->mem);
 518         }
 519
 520         /* We have to remove fragment header from datagram and to relocate
 521          * header in order to calculate ICV correctly. */
 522         nhoff = fq->nhoffset;
 523         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 524         memmove(head->head + sizeof(struct frag_hdr), head->head,
 525                 (head->data - head->head) - sizeof(struct frag_hdr));
 526         head->mac_header += sizeof(struct frag_hdr);
 527         head->network_header += sizeof(struct frag_hdr);
 528
 529         skb_shinfo(head)->frag_list = head->next;
 530         skb_reset_transport_header(head);
 531         skb_push(head, head->data - skb_network_header(head));
 532         atomic_sub(head->truesize, &fq->q.net->mem);
 533
 534         for (fp=head->next; fp; fp = fp->next) {
 535                 head->data_len += fp->len;
 536                 head->len += fp->len;
 537                 if (head->ip_summed != fp->ip_summed)
 538                         head->ip_summed = CHECKSUM_NONE;
 539                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 540                         head->csum = csum_add(head->csum, fp->csum);
 541                 head->truesize += fp->truesize;
 542                 atomic_sub(fp->truesize, &fq->q.net->mem);
 543         }
 544
 545         head->next = NULL;
 546         head->dev = dev;
 547         head->tstamp = fq->q.stamp;
 548         ipv6_hdr(head)->payload_len = htons(payload_len);
 549         IP6CB(head)->nhoff = nhoff;
 550
 551         /* Yes, and fold redundant checksum back. 8) */
 552         if (head->ip_summed == CHECKSUM_COMPLETE)
 553                 head->csum = csum_partial(skb_network_header(head),
 554                                           skb_network_header_len(head),
 555                                           head->csum);
 556
 557         rcu_read_lock();
 558         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 559         rcu_read_unlock();
 560         fq->q.fragments = NULL;
 561         return 1;
 562
 563 out_oversize:
 564         if (net_ratelimit())
 565                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
 566         goto out_fail;
 567 out_oom:
 568         if (net_ratelimit())
 569                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
 570 out_fail:
 571         rcu_read_lock();
 572         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 573         rcu_read_unlock();
 574         return -1;
 575 }
 576
 577 static int ipv6_frag_rcv(struct sk_buff *skb)
 578 {
 579         struct frag_hdr *fhdr;
 580         struct frag_queue *fq;
 581         struct ipv6hdr *hdr = ipv6_hdr(skb);
 582         struct net *net = dev_net(skb_dst(skb)->dev);
 583
 584         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
 585
 586         /* Jumbo payload inhibits frag. header */
 587         if (hdr->payload_len==0)
 588                 goto fail_hdr;
 589
 590         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 591                                  sizeof(struct frag_hdr))))
 592                 goto fail_hdr;
 593
 594         hdr = ipv6_hdr(skb);
 595         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 596
 597         if (!(fhdr->frag_off & htons(0xFFF9))) {
 598                 /* It is not a fragmented frame */
 599                 skb->transport_header += sizeof(struct frag_hdr);
 600                 IP6_INC_STATS_BH(net,
 601                                  ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
 602
 603                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 604                 return 1;
 605         }
 606
 607         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
 608                 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
 609
 610         if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
 611                           ip6_dst_idev(skb_dst(skb)))) != NULL) {
 612                 int ret;
 613
 614                 spin_lock(&fq->q.lock);
 615
 616                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 617
 618                 spin_unlock(&fq->q.lock);
 619                 fq_put(fq);
 620                 return ret;
 621         }
 622
 623         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
 624         kfree_skb(skb);
 625         return -1;
 626
 627 fail_hdr:
 628         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
 629         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
 630         return -1;
 631 }
 632
 633 static const struct inet6_protocol frag_protocol =
 634 {
 635         .handler        =       ipv6_frag_rcv,
 636         .flags          =       INET6_PROTO_NOPOLICY,
 637 };
 638
 639 #ifdef CONFIG_SYSCTL
 640 static int
 641 proc_dointvec_fragment(struct ctl_table *table, int write,
 642                 void __user *buffer, size_t *lenp, loff_t *ppos)
 643 {
 644         struct net *net = container_of(table->data, struct net,
 645                                        ipv6.frags.high_thresh);
 646         ctl_table tmp = *table;
 647         int new_bytes, ret;
 648
 649         mutex_lock(&net->ipv6.frags.lock);
 650         if (write) {
 651                 tmp.data = &new_bytes;
 652                 table = &tmp;
 653         }
 654
 655         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 656
 657         if (!ret && write) {
 658                 ret = mem_reserve_kmalloc_set(&net->ipv6.frags.reserve,
 659                                               new_bytes);
 660                 if (!ret)
 661                         net->ipv6.frags.high_thresh = new_bytes;
 662         }
 663         mutex_unlock(&net->ipv6.frags.lock);
 664
 665         return ret;
 666 }
 667
 668 static struct ctl_table ip6_frags_ns_ctl_table[] = {
 669         {
 670                 .procname       = "ip6frag_high_thresh",
 671                 .data           = &init_net.ipv6.frags.high_thresh,
 672                 .maxlen         = sizeof(int),
 673                 .mode           = 0644,
 674                 .proc_handler   = proc_dointvec_fragment,
 675         },
 676         {
 677                 .procname       = "ip6frag_low_thresh",
 678                 .data           = &init_net.ipv6.frags.low_thresh,
 679                 .maxlen         = sizeof(int),
 680                 .mode           = 0644,
 681                 .proc_handler   = proc_dointvec
 682         },
 683         {
 684                 .procname       = "ip6frag_time",
 685                 .data           = &init_net.ipv6.frags.timeout,
 686                 .maxlen         = sizeof(int),
 687                 .mode           = 0644,
 688                 .proc_handler   = proc_dointvec_jiffies,
 689         },
 690         { }
 691 };
 692
 693 static struct ctl_table ip6_frags_ctl_table[] = {
 694         {
 695                 .procname       = "ip6frag_secret_interval",
 696                 .data           = &ip6_frags.secret_interval,
 697                 .maxlen         = sizeof(int),
 698                 .mode           = 0644,
 699                 .proc_handler   = proc_dointvec_jiffies,
 700         },
 701         { }
 702 };
 703
 704 static int ip6_frags_ns_sysctl_register(struct net *net)
 705 {
 706         struct ctl_table *table;
 707         struct ctl_table_header *hdr;
 708
 709         table = ip6_frags_ns_ctl_table;
 710         if (!net_eq(net, &init_net)) {
 711                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
 712                 if (table == NULL)
 713                         goto err_alloc;
 714
 715                 table[0].data = &net->ipv6.frags.high_thresh;
 716                 table[1].data = &net->ipv6.frags.low_thresh;
 717                 table[2].data = &net->ipv6.frags.timeout;
 718         }
 719
 720         hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
 721         if (hdr == NULL)
 722                 goto err_reg;
 723
 724         net->ipv6.sysctl.frags_hdr = hdr;
 725         return 0;
 726
 727 err_reg:
 728         if (!net_eq(net, &init_net))
 729                 kfree(table);
 730 err_alloc:
 731         return -ENOMEM;
 732 }
 733
 734 static void ip6_frags_ns_sysctl_unregister(struct net *net)
 735 {
 736         struct ctl_table *table;
 737
 738         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
 739         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
 740         if (!net_eq(net, &init_net))
 741                 kfree(table);
 742 }
 743
 744 static struct ctl_table_header *ip6_ctl_header;
 745
 746 static int ip6_frags_sysctl_register(void)
 747 {
 748         ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
 749                         ip6_frags_ctl_table);
 750         return ip6_ctl_header == NULL ? -ENOMEM : 0;
 751 }
 752
 753 static void ip6_frags_sysctl_unregister(void)
 754 {
 755         unregister_net_sysctl_table(ip6_ctl_header);
 756 }
 757 #else
 758 static inline int ip6_frags_ns_sysctl_register(struct net *net)
 759 {
 760         return 0;
 761 }
 762
 763 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
 764 {
 765 }
 766
 767 static inline int ip6_frags_sysctl_register(void)
 768 {
 769         return 0;
 770 }
 771
 772 static inline void ip6_frags_sysctl_unregister(void)
 773 {
 774 }
 775 #endif
 776
 777 static int ipv6_frags_init_net(struct net *net)
 778 {
 779         int ret;
 780
 781         net->ipv6.frags.high_thresh = 256 * 1024;
 782         net->ipv6.frags.low_thresh = 192 * 1024;
 783         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
 784
 785         inet_frags_init_net(&net->ipv6.frags);
 786
 787         ret = ip6_frags_ns_sysctl_register(net);
 788         if (ret)
 789                 goto out_reg;
 790
 791         mem_reserve_init(&net->ipv6.frags.reserve, "IPv6 fragment cache",
 792                          &net_skb_reserve);
 793         ret = mem_reserve_kmalloc_set(&net->ipv6.frags.reserve,
 794                                       net->ipv6.frags.high_thresh);
 795         if (ret)
 796                 goto out_reserve;
 797
 798         return 0;
 799
 800 out_reserve:
 801         mem_reserve_disconnect(&net->ipv6.frags.reserve);
 802         ip6_frags_ns_sysctl_unregister(net);
 803 out_reg:
 804         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 805
 806         return ret;
 807 }
 808
 809 static void ipv6_frags_exit_net(struct net *net)
 810 {
 811         mem_reserve_disconnect(&net->ipv6.frags.reserve);
 812         ip6_frags_ns_sysctl_unregister(net);
 813         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 814 }
 815
 816 static struct pernet_operations ip6_frags_ops = {
 817         .init = ipv6_frags_init_net,
 818         .exit = ipv6_frags_exit_net,
 819 };
 820
 821 int __init ipv6_frag_init(void)
 822 {
 823         int ret;
 824
 825         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 826         if (ret)
 827                 goto out;
 828
 829         ret = ip6_frags_sysctl_register();
 830         if (ret)
 831                 goto err_sysctl;
 832
 833         ret = register_pernet_subsys(&ip6_frags_ops);
 834         if (ret)
 835                 goto err_pernet;
 836
 837         ip6_frags.hashfn = ip6_hashfn;
 838         ip6_frags.constructor = ip6_frag_init;
 839         ip6_frags.destructor = NULL;
 840         ip6_frags.skb_free = NULL;
 841         ip6_frags.qsize = sizeof(struct frag_queue);
 842         ip6_frags.match = ip6_frag_match;
 843         ip6_frags.frag_expire = ip6_frag_expire;
 844         ip6_frags.secret_interval = 10 * 60 * HZ;
 845         inet_frags_init(&ip6_frags);
 846 out:
 847         return ret;
 848
 849 err_pernet:
 850         ip6_frags_sysctl_unregister();
 851 err_sysctl:
 852         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 853         goto out;
 854 }
 855
 856 void ipv6_frag_exit(void)
 857 {
 858         inet_frags_fini(&ip6_frags);
 859         ip6_frags_sysctl_unregister();
 860         unregister_pernet_subsys(&ip6_frags_ops);
 861         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 862 }