SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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  *      $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
   9  *
  10  *      Based on: net/ipv4/ip_fragment.c
  11  *
  12  *      This program is free software; you can redistribute it and/or
  13  *      modify it under the terms of the GNU General Public License
  14  *      as published by the Free Software Foundation; either version
  15  *      2 of the License, or (at your option) any later version.
  16  */
  17
  18 /*
  19  *      Fixes:
  20  *      Andi Kleen      Make it work with multiple hosts.
  21  *                      More RFC compliance.
  22  *
  23  *      Horst von Brand Add missing #include <linux/string.h>
  24  *      Alexey Kuznetsov        SMP races, threading, cleanup.
  25  *      Patrick McHardy         LRU queue of frag heads for evictor.
  26  *      Mitsuru KANDA @USAGI    Register inet6_protocol{}.
  27  *      David Stevens and
  28  *      YOSHIFUJI,H. @USAGI     Always remove fragment header to
  29  *                              calculate ICV correctly.
  30  */
  31 #include <linux/errno.h>
  32 #include <linux/types.h>
  33 #include <linux/string.h>
  34 #include <linux/socket.h>
  35 #include <linux/sockios.h>
  36 #include <linux/jiffies.h>
  37 #include <linux/net.h>
  38 #include <linux/list.h>
  39 #include <linux/netdevice.h>
  40 #include <linux/in6.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/icmpv6.h>
  43 #include <linux/random.h>
  44 #include <linux/jhash.h>
  45 #include <linux/skbuff.h>
  46
  47 #include <net/sock.h>
  48 #include <net/snmp.h>
  49
  50 #include <net/ipv6.h>
  51 #include <net/ip6_route.h>
  52 #include <net/protocol.h>
  53 #include <net/transp_v6.h>
  54 #include <net/rawv6.h>
  55 #include <net/ndisc.h>
  56 #include <net/addrconf.h>
  57 #include <net/inet_frag.h>
  58
  59 struct ip6frag_skb_cb
  60 {
  61         struct inet6_skb_parm   h;
  62         int                     offset;
  63 };
  64
  65 #define FRAG6_CB(skb)   ((struct ip6frag_skb_cb*)((skb)->cb))
  66
  67
  68 /*
  69  *      Equivalent of ipv4 struct ipq
  70  */
  71
  72 struct frag_queue
  73 {
  74         struct inet_frag_queue  q;
  75
  76         __be32                  id;             /* fragment id          */
  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 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
 105                                struct in6_addr *daddr)
 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 += ip6_frags.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
 131 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 132 {
 133         struct frag_queue *fq;
 134
 135         fq = container_of(q, struct frag_queue, q);
 136         return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
 137 }
 138
 139 int ip6_frag_match(struct inet_frag_queue *q, void *a)
 140 {
 141         struct frag_queue *fq;
 142         struct ip6_create_arg *arg = a;
 143
 144         fq = container_of(q, struct frag_queue, q);
 145         return (fq->id == arg->id &&
 146                         ipv6_addr_equal(&fq->saddr, arg->src) &&
 147                         ipv6_addr_equal(&fq->daddr, arg->dst));
 148 }
 149 EXPORT_SYMBOL(ip6_frag_match);
 150
 151 /* Memory Tracking Functions. */
 152 static inline void frag_kfree_skb(struct netns_frags *nf,
 153                 struct sk_buff *skb, int *work)
 154 {
 155         if (work)
 156                 *work -= skb->truesize;
 157         atomic_sub(skb->truesize, &nf->mem);
 158         kfree_skb(skb);
 159 }
 160
 161 void ip6_frag_init(struct inet_frag_queue *q, void *a)
 162 {
 163         struct frag_queue *fq = container_of(q, struct frag_queue, q);
 164         struct ip6_create_arg *arg = a;
 165
 166         fq->id = arg->id;
 167         ipv6_addr_copy(&fq->saddr, arg->src);
 168         ipv6_addr_copy(&fq->daddr, arg->dst);
 169 }
 170 EXPORT_SYMBOL(ip6_frag_init);
 171
 172 /* Destruction primitives. */
 173
 174 static __inline__ void fq_put(struct frag_queue *fq)
 175 {
 176         inet_frag_put(&fq->q, &ip6_frags);
 177 }
 178
 179 /* Kill fq entry. It is not destroyed immediately,
 180  * because caller (and someone more) holds reference count.
 181  */
 182 static __inline__ void fq_kill(struct frag_queue *fq)
 183 {
 184         inet_frag_kill(&fq->q, &ip6_frags);
 185 }
 186
 187 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 188 {
 189         int evicted;
 190
 191         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
 192         if (evicted)
 193                 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
 194 }
 195
 196 static void ip6_frag_expire(unsigned long data)
 197 {
 198         struct frag_queue *fq;
 199         struct net_device *dev = NULL;
 200         struct net *net;
 201
 202         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 203
 204         spin_lock(&fq->q.lock);
 205
 206         if (fq->q.last_in & INET_FRAG_COMPLETE)
 207                 goto out;
 208
 209         fq_kill(fq);
 210
 211         net = container_of(fq->q.net, struct net, ipv6.frags);
 212         dev = dev_get_by_index(net, fq->iif);
 213         if (!dev)
 214                 goto out;
 215
 216         rcu_read_lock();
 217         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 218         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 219         rcu_read_unlock();
 220
 221         /* Don't send error if the first segment did not arrive. */
 222         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
 223                 goto out;
 224
 225         /*
 226            But use as source device on which LAST ARRIVED
 227            segment was received. And do not use fq->dev
 228            pointer directly, device might already disappeared.
 229          */
 230         fq->q.fragments->dev = dev;
 231         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
 232 out:
 233         if (dev)
 234                 dev_put(dev);
 235         spin_unlock(&fq->q.lock);
 236         fq_put(fq);
 237 }
 238
 239 static __inline__ struct frag_queue *
 240 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
 241         struct inet6_dev *idev)
 242 {
 243         struct inet_frag_queue *q;
 244         struct ip6_create_arg arg;
 245         unsigned int hash;
 246
 247         arg.id = id;
 248         arg.src = src;
 249         arg.dst = dst;
 250
 251         read_lock(&ip6_frags.lock);
 252         hash = ip6qhashfn(id, src, dst);
 253
 254         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
 255         if (q == NULL)
 256                 goto oom;
 257
 258         return container_of(q, struct frag_queue, q);
 259
 260 oom:
 261         IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
 262         return NULL;
 263 }
 264
 265 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 266                            struct frag_hdr *fhdr, int nhoff)
 267 {
 268         struct sk_buff *prev, *next;
 269         struct net_device *dev;
 270         int offset, end;
 271
 272         if (fq->q.last_in & INET_FRAG_COMPLETE)
 273                 goto err;
 274
 275         offset = ntohs(fhdr->frag_off) & ~0x7;
 276         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 277                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 278
 279         if ((unsigned int)end > IPV6_MAXPLEN) {
 280                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
 281                                  IPSTATS_MIB_INHDRERRORS);
 282                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 283                                   ((u8 *)&fhdr->frag_off -
 284                                    skb_network_header(skb)));
 285                 return -1;
 286         }
 287
 288         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 289                 const unsigned char *nh = skb_network_header(skb);
 290                 skb->csum = csum_sub(skb->csum,
 291                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 292                                                   0));
 293         }
 294
 295         /* Is this the final fragment? */
 296         if (!(fhdr->frag_off & htons(IP6_MF))) {
 297                 /* If we already have some bits beyond end
 298                  * or have different end, the segment is corrupted.
 299                  */
 300                 if (end < fq->q.len ||
 301                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
 302                         goto err;
 303                 fq->q.last_in |= INET_FRAG_LAST_IN;
 304                 fq->q.len = end;
 305         } else {
 306                 /* Check if the fragment is rounded to 8 bytes.
 307                  * Required by the RFC.
 308                  */
 309                 if (end & 0x7) {
 310                         /* RFC2460 says always send parameter problem in
 311                          * this case. -DaveM
 312                          */
 313                         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
 314                                          IPSTATS_MIB_INHDRERRORS);
 315                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 316                                           offsetof(struct ipv6hdr, payload_len));
 317                         return -1;
 318                 }
 319                 if (end > fq->q.len) {
 320                         /* Some bits beyond end -> corruption. */
 321                         if (fq->q.last_in & INET_FRAG_LAST_IN)
 322                                 goto err;
 323                         fq->q.len = end;
 324                 }
 325         }
 326
 327         if (end == offset)
 328                 goto err;
 329
 330         /* Point into the IP datagram 'data' part. */
 331         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 332                 goto err;
 333
 334         if (pskb_trim_rcsum(skb, end - offset))
 335                 goto err;
 336
 337         /* Find out which fragments are in front and at the back of us
 338          * in the chain of fragments so far.  We must know where to put
 339          * this fragment, right?
 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         /* We found where to put this one.  Check for overlap with
 349          * preceding fragment, and, if needed, align things so that
 350          * any overlaps are eliminated.
 351          */
 352         if (prev) {
 353                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
 354
 355                 if (i > 0) {
 356                         offset += i;
 357                         if (end <= offset)
 358                                 goto err;
 359                         if (!pskb_pull(skb, i))
 360                                 goto err;
 361                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 362                                 skb->ip_summed = CHECKSUM_NONE;
 363                 }
 364         }
 365
 366         /* Look for overlap with succeeding segments.
 367          * If we can merge fragments, do it.
 368          */
 369         while (next && FRAG6_CB(next)->offset < end) {
 370                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
 371
 372                 if (i < next->len) {
 373                         /* Eat head of the next overlapped fragment
 374                          * and leave the loop. The next ones cannot overlap.
 375                          */
 376                         if (!pskb_pull(next, i))
 377                                 goto err;
 378                         FRAG6_CB(next)->offset += i;    /* next fragment */
 379                         fq->q.meat -= i;
 380                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 381                                 next->ip_summed = CHECKSUM_NONE;
 382                         break;
 383                 } else {
 384                         struct sk_buff *free_it = next;
 385
 386                         /* Old fragment is completely overridden with
 387                          * new one drop it.
 388                          */
 389                         next = next->next;
 390
 391                         if (prev)
 392                                 prev->next = next;
 393                         else
 394                                 fq->q.fragments = next;
 395
 396                         fq->q.meat -= free_it->len;
 397                         frag_kfree_skb(fq->q.net, free_it, NULL);
 398                 }
 399         }
 400
 401         FRAG6_CB(skb)->offset = offset;
 402
 403         /* Insert this fragment in the chain of fragments. */
 404         skb->next = next;
 405         if (prev)
 406                 prev->next = skb;
 407         else
 408                 fq->q.fragments = skb;
 409
 410         dev = skb->dev;
 411         if (dev) {
 412                 fq->iif = dev->ifindex;
 413                 skb->dev = NULL;
 414         }
 415         fq->q.stamp = skb->tstamp;
 416         fq->q.meat += skb->len;
 417         atomic_add(skb->truesize, &fq->q.net->mem);
 418
 419         /* The first fragment.
 420          * nhoffset is obtained from the first fragment, of course.
 421          */
 422         if (offset == 0) {
 423                 fq->nhoffset = nhoff;
 424                 fq->q.last_in |= INET_FRAG_FIRST_IN;
 425         }
 426
 427         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 428             fq->q.meat == fq->q.len)
 429                 return ip6_frag_reasm(fq, prev, dev);
 430
 431         write_lock(&ip6_frags.lock);
 432         list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
 433         write_unlock(&ip6_frags.lock);
 434         return -1;
 435
 436 err:
 437         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
 438         kfree_skb(skb);
 439         return -1;
 440 }
 441
 442 /*
 443  *      Check if this packet is complete.
 444  *      Returns NULL on failure by any reason, and pointer
 445  *      to current nexthdr field in reassembled frame.
 446  *
 447  *      It is called with locked fq, and caller must check that
 448  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 449  *      the last and the first frames arrived and all the bits are here.
 450  */
 451 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 452                           struct net_device *dev)
 453 {
 454         struct sk_buff *fp, *head = fq->q.fragments;
 455         int    payload_len;
 456         unsigned int nhoff;
 457
 458         fq_kill(fq);
 459
 460         /* Make the one we just received the head. */
 461         if (prev) {
 462                 head = prev->next;
 463                 fp = skb_clone(head, GFP_ATOMIC);
 464
 465                 if (!fp)
 466                         goto out_oom;
 467
 468                 fp->next = head->next;
 469                 prev->next = fp;
 470
 471                 skb_morph(head, fq->q.fragments);
 472                 head->next = fq->q.fragments->next;
 473
 474                 kfree_skb(fq->q.fragments);
 475                 fq->q.fragments = head;
 476         }
 477
 478         BUG_TRAP(head != NULL);
 479         BUG_TRAP(FRAG6_CB(head)->offset == 0);
 480
 481         /* Unfragmented part is taken from the first segment. */
 482         payload_len = ((head->data - skb_network_header(head)) -
 483                        sizeof(struct ipv6hdr) + fq->q.len -
 484                        sizeof(struct frag_hdr));
 485         if (payload_len > IPV6_MAXPLEN)
 486                 goto out_oversize;
 487
 488         /* Head of list must not be cloned. */
 489         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 490                 goto out_oom;
 491
 492         /* If the first fragment is fragmented itself, we split
 493          * it to two chunks: the first with data and paged part
 494          * and the second, holding only fragments. */
 495         if (skb_shinfo(head)->frag_list) {
 496                 struct sk_buff *clone;
 497                 int i, plen = 0;
 498
 499                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 500                         goto out_oom;
 501                 clone->next = head->next;
 502                 head->next = clone;
 503                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 504                 skb_shinfo(head)->frag_list = NULL;
 505                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 506                         plen += skb_shinfo(head)->frags[i].size;
 507                 clone->len = clone->data_len = head->data_len - plen;
 508                 head->data_len -= clone->len;
 509                 head->len -= clone->len;
 510                 clone->csum = 0;
 511                 clone->ip_summed = head->ip_summed;
 512                 atomic_add(clone->truesize, &fq->q.net->mem);
 513         }
 514
 515         /* We have to remove fragment header from datagram and to relocate
 516          * header in order to calculate ICV correctly. */
 517         nhoff = fq->nhoffset;
 518         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 519         memmove(head->head + sizeof(struct frag_hdr), head->head,
 520                 (head->data - head->head) - sizeof(struct frag_hdr));
 521         head->mac_header += sizeof(struct frag_hdr);
 522         head->network_header += sizeof(struct frag_hdr);
 523
 524         skb_shinfo(head)->frag_list = head->next;
 525         skb_reset_transport_header(head);
 526         skb_push(head, head->data - skb_network_header(head));
 527         atomic_sub(head->truesize, &fq->q.net->mem);
 528
 529         for (fp=head->next; fp; fp = fp->next) {
 530                 head->data_len += fp->len;
 531                 head->len += fp->len;
 532                 if (head->ip_summed != fp->ip_summed)
 533                         head->ip_summed = CHECKSUM_NONE;
 534                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 535                         head->csum = csum_add(head->csum, fp->csum);
 536                 head->truesize += fp->truesize;
 537                 atomic_sub(fp->truesize, &fq->q.net->mem);
 538         }
 539
 540         head->next = NULL;
 541         head->dev = dev;
 542         head->tstamp = fq->q.stamp;
 543         ipv6_hdr(head)->payload_len = htons(payload_len);
 544         IP6CB(head)->nhoff = nhoff;
 545
 546         /* Yes, and fold redundant checksum back. 8) */
 547         if (head->ip_summed == CHECKSUM_COMPLETE)
 548                 head->csum = csum_partial(skb_network_header(head),
 549                                           skb_network_header_len(head),
 550                                           head->csum);
 551
 552         rcu_read_lock();
 553         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 554         rcu_read_unlock();
 555         fq->q.fragments = NULL;
 556         return 1;
 557
 558 out_oversize:
 559         if (net_ratelimit())
 560                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
 561         goto out_fail;
 562 out_oom:
 563         if (net_ratelimit())
 564                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
 565 out_fail:
 566         rcu_read_lock();
 567         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 568         rcu_read_unlock();
 569         return -1;
 570 }
 571
 572 static int ipv6_frag_rcv(struct sk_buff *skb)
 573 {
 574         struct frag_hdr *fhdr;
 575         struct frag_queue *fq;
 576         struct ipv6hdr *hdr = ipv6_hdr(skb);
 577         struct net *net;
 578
 579         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
 580
 581         /* Jumbo payload inhibits frag. header */
 582         if (hdr->payload_len==0) {
 583                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
 584                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 585                                   skb_network_header_len(skb));
 586                 return -1;
 587         }
 588         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 589                                  sizeof(struct frag_hdr)))) {
 590                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
 591                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 592                                   skb_network_header_len(skb));
 593                 return -1;
 594         }
 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(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
 603
 604                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 605                 return 1;
 606         }
 607
 608         net = dev_net(skb->dev);
 609         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
 610                 ip6_evictor(net, ip6_dst_idev(skb->dst));
 611
 612         if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
 613                           ip6_dst_idev(skb->dst))) != 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(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
 626         kfree_skb(skb);
 627         return -1;
 628 }
 629
 630 static struct inet6_protocol frag_protocol =
 631 {
 632         .handler        =       ipv6_frag_rcv,
 633         .flags          =       INET6_PROTO_NOPOLICY,
 634 };
 635
 636 #ifdef CONFIG_SYSCTL
 637 static struct ctl_table ip6_frags_ctl_table[] = {
 638         {
 639                 .ctl_name       = NET_IPV6_IP6FRAG_HIGH_THRESH,
 640                 .procname       = "ip6frag_high_thresh",
 641                 .data           = &init_net.ipv6.frags.high_thresh,
 642                 .maxlen         = sizeof(int),
 643                 .mode           = 0644,
 644                 .proc_handler   = &proc_dointvec
 645         },
 646         {
 647                 .ctl_name       = NET_IPV6_IP6FRAG_LOW_THRESH,
 648                 .procname       = "ip6frag_low_thresh",
 649                 .data           = &init_net.ipv6.frags.low_thresh,
 650                 .maxlen         = sizeof(int),
 651                 .mode           = 0644,
 652                 .proc_handler   = &proc_dointvec
 653         },
 654         {
 655                 .ctl_name       = NET_IPV6_IP6FRAG_TIME,
 656                 .procname       = "ip6frag_time",
 657                 .data           = &init_net.ipv6.frags.timeout,
 658                 .maxlen         = sizeof(int),
 659                 .mode           = 0644,
 660                 .proc_handler   = &proc_dointvec_jiffies,
 661                 .strategy       = &sysctl_jiffies,
 662         },
 663         {
 664                 .ctl_name       = NET_IPV6_IP6FRAG_SECRET_INTERVAL,
 665                 .procname       = "ip6frag_secret_interval",
 666                 .data           = &ip6_frags.secret_interval,
 667                 .maxlen         = sizeof(int),
 668                 .mode           = 0644,
 669                 .proc_handler   = &proc_dointvec_jiffies,
 670                 .strategy       = &sysctl_jiffies
 671         },
 672         { }
 673 };
 674
 675 static int ip6_frags_sysctl_register(struct net *net)
 676 {
 677         struct ctl_table *table;
 678         struct ctl_table_header *hdr;
 679
 680         table = ip6_frags_ctl_table;
 681         if (net != &init_net) {
 682                 table = kmemdup(table, sizeof(ip6_frags_ctl_table), GFP_KERNEL);
 683                 if (table == NULL)
 684                         goto err_alloc;
 685
 686                 table[0].data = &net->ipv6.frags.high_thresh;
 687                 table[1].data = &net->ipv6.frags.low_thresh;
 688                 table[2].data = &net->ipv6.frags.timeout;
 689                 table[3].mode &= ~0222;
 690         }
 691
 692         hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
 693         if (hdr == NULL)
 694                 goto err_reg;
 695
 696         net->ipv6.sysctl.frags_hdr = hdr;
 697         return 0;
 698
 699 err_reg:
 700         if (net != &init_net)
 701                 kfree(table);
 702 err_alloc:
 703         return -ENOMEM;
 704 }
 705
 706 static void ip6_frags_sysctl_unregister(struct net *net)
 707 {
 708         struct ctl_table *table;
 709
 710         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
 711         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
 712         kfree(table);
 713 }
 714 #else
 715 static inline int ip6_frags_sysctl_register(struct net *net)
 716 {
 717         return 0;
 718 }
 719
 720 static inline void ip6_frags_sysctl_unregister(struct net *net)
 721 {
 722 }
 723 #endif
 724
 725 static int ipv6_frags_init_net(struct net *net)
 726 {
 727         net->ipv6.frags.high_thresh = 256 * 1024;
 728         net->ipv6.frags.low_thresh = 192 * 1024;
 729         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
 730
 731         inet_frags_init_net(&net->ipv6.frags);
 732
 733         return ip6_frags_sysctl_register(net);
 734 }
 735
 736 static void ipv6_frags_exit_net(struct net *net)
 737 {
 738         ip6_frags_sysctl_unregister(net);
 739         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 740 }
 741
 742 static struct pernet_operations ip6_frags_ops = {
 743         .init = ipv6_frags_init_net,
 744         .exit = ipv6_frags_exit_net,
 745 };
 746
 747 int __init ipv6_frag_init(void)
 748 {
 749         int ret;
 750
 751         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 752         if (ret)
 753                 goto out;
 754
 755         register_pernet_subsys(&ip6_frags_ops);
 756
 757         ip6_frags.hashfn = ip6_hashfn;
 758         ip6_frags.constructor = ip6_frag_init;
 759         ip6_frags.destructor = NULL;
 760         ip6_frags.skb_free = NULL;
 761         ip6_frags.qsize = sizeof(struct frag_queue);
 762         ip6_frags.match = ip6_frag_match;
 763         ip6_frags.frag_expire = ip6_frag_expire;
 764         ip6_frags.secret_interval = 10 * 60 * HZ;
 765         inet_frags_init(&ip6_frags);
 766 out:
 767         return ret;
 768 }
 769
 770 void ipv6_frag_exit(void)
 771 {
 772         inet_frags_fini(&ip6_frags);
 773         unregister_pernet_subsys(&ip6_frags_ops);
 774         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 775 }