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 #include <linux/export.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 static struct inet_frags ip6_frags;
  69
  70 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
  71                           struct net_device *dev);
  72
  73 /*
  74  * callers should be careful not to use the hash value outside the ipfrag_lock
  75  * as doing so could race with ipfrag_hash_rnd being recalculated.
  76  */
  77 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
  78                              const struct in6_addr *daddr, u32 rnd)
  79 {
  80         u32 c;
  81
  82         c = jhash_3words((__force u32)saddr->s6_addr32[0],
  83                          (__force u32)saddr->s6_addr32[1],
  84                          (__force u32)saddr->s6_addr32[2],
  85                          rnd);
  86
  87         c = jhash_3words((__force u32)saddr->s6_addr32[3],
  88                          (__force u32)daddr->s6_addr32[0],
  89                          (__force u32)daddr->s6_addr32[1],
  90                          c);
  91
  92         c =  jhash_3words((__force u32)daddr->s6_addr32[2],
  93                           (__force u32)daddr->s6_addr32[3],
  94                           (__force u32)id,
  95                           c);
  96
  97         return c & (INETFRAGS_HASHSZ - 1);
  98 }
  99 EXPORT_SYMBOL_GPL(inet6_hash_frag);
 100
 101 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 102 {
 103         struct frag_queue *fq;
 104
 105         fq = container_of(q, struct frag_queue, q);
 106         return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
 107 }
 108
 109 bool ip6_frag_match(struct inet_frag_queue *q, void *a)
 110 {
 111         struct frag_queue *fq;
 112         struct ip6_create_arg *arg = a;
 113
 114         fq = container_of(q, struct frag_queue, q);
 115         return  fq->id == arg->id &&
 116                 fq->user == arg->user &&
 117                 ipv6_addr_equal(&fq->saddr, arg->src) &&
 118                 ipv6_addr_equal(&fq->daddr, arg->dst);
 119 }
 120 EXPORT_SYMBOL(ip6_frag_match);
 121
 122 void ip6_frag_init(struct inet_frag_queue *q, void *a)
 123 {
 124         struct frag_queue *fq = container_of(q, struct frag_queue, q);
 125         struct ip6_create_arg *arg = a;
 126
 127         fq->id = arg->id;
 128         fq->user = arg->user;
 129         fq->saddr = *arg->src;
 130         fq->daddr = *arg->dst;
 131 }
 132 EXPORT_SYMBOL(ip6_frag_init);
 133
 134 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 135 {
 136         int evicted;
 137
 138         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
 139         if (evicted)
 140                 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
 141 }
 142
 143 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq,
 144                            struct inet_frags *frags)
 145 {
 146         struct net_device *dev = NULL;
 147
 148         spin_lock(&fq->q.lock);
 149
 150         if (fq->q.last_in & INET_FRAG_COMPLETE)
 151                 goto out;
 152
 153         inet_frag_kill(&fq->q, frags);
 154
 155         rcu_read_lock();
 156         dev = dev_get_by_index_rcu(net, fq->iif);
 157         if (!dev)
 158                 goto out_rcu_unlock;
 159
 160         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 161         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 162
 163         /* Don't send error if the first segment did not arrive. */
 164         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
 165                 goto out_rcu_unlock;
 166
 167         /*
 168            But use as source device on which LAST ARRIVED
 169            segment was received. And do not use fq->dev
 170            pointer directly, device might already disappeared.
 171          */
 172         fq->q.fragments->dev = dev;
 173         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
 174 out_rcu_unlock:
 175         rcu_read_unlock();
 176 out:
 177         spin_unlock(&fq->q.lock);
 178         inet_frag_put(&fq->q, frags);
 179 }
 180 EXPORT_SYMBOL(ip6_expire_frag_queue);
 181
 182 static void ip6_frag_expire(unsigned long data)
 183 {
 184         struct frag_queue *fq;
 185         struct net *net;
 186
 187         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 188         net = container_of(fq->q.net, struct net, ipv6.frags);
 189
 190         ip6_expire_frag_queue(net, fq, &ip6_frags);
 191 }
 192
 193 static __inline__ struct frag_queue *
 194 fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
 195 {
 196         struct inet_frag_queue *q;
 197         struct ip6_create_arg arg;
 198         unsigned int hash;
 199
 200         arg.id = id;
 201         arg.user = IP6_DEFRAG_LOCAL_DELIVER;
 202         arg.src = src;
 203         arg.dst = dst;
 204
 205         read_lock(&ip6_frags.lock);
 206         hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
 207
 208         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
 209         if (q == NULL)
 210                 return NULL;
 211
 212         return container_of(q, struct frag_queue, q);
 213 }
 214
 215 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 216                            struct frag_hdr *fhdr, int nhoff)
 217 {
 218         struct sk_buff *prev, *next;
 219         struct net_device *dev;
 220         int offset, end;
 221         struct net *net = dev_net(skb_dst(skb)->dev);
 222
 223         if (fq->q.last_in & INET_FRAG_COMPLETE)
 224                 goto err;
 225
 226         offset = ntohs(fhdr->frag_off) & ~0x7;
 227         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 228                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 229
 230         if ((unsigned int)end > IPV6_MAXPLEN) {
 231                 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 232                                  IPSTATS_MIB_INHDRERRORS);
 233                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 234                                   ((u8 *)&fhdr->frag_off -
 235                                    skb_network_header(skb)));
 236                 return -1;
 237         }
 238
 239         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 240                 const unsigned char *nh = skb_network_header(skb);
 241                 skb->csum = csum_sub(skb->csum,
 242                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 243                                                   0));
 244         }
 245
 246         /* Is this the final fragment? */
 247         if (!(fhdr->frag_off & htons(IP6_MF))) {
 248                 /* If we already have some bits beyond end
 249                  * or have different end, the segment is corrupted.
 250                  */
 251                 if (end < fq->q.len ||
 252                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
 253                         goto err;
 254                 fq->q.last_in |= INET_FRAG_LAST_IN;
 255                 fq->q.len = end;
 256         } else {
 257                 /* Check if the fragment is rounded to 8 bytes.
 258                  * Required by the RFC.
 259                  */
 260                 if (end & 0x7) {
 261                         /* RFC2460 says always send parameter problem in
 262                          * this case. -DaveM
 263                          */
 264                         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 265                                          IPSTATS_MIB_INHDRERRORS);
 266                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 267                                           offsetof(struct ipv6hdr, payload_len));
 268                         return -1;
 269                 }
 270                 if (end > fq->q.len) {
 271                         /* Some bits beyond end -> corruption. */
 272                         if (fq->q.last_in & INET_FRAG_LAST_IN)
 273                                 goto err;
 274                         fq->q.len = end;
 275                 }
 276         }
 277
 278         if (end == offset)
 279                 goto err;
 280
 281         /* Point into the IP datagram 'data' part. */
 282         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 283                 goto err;
 284
 285         if (pskb_trim_rcsum(skb, end - offset))
 286                 goto err;
 287
 288         /* Find out which fragments are in front and at the back of us
 289          * in the chain of fragments so far.  We must know where to put
 290          * this fragment, right?
 291          */
 292         prev = fq->q.fragments_tail;
 293         if (!prev || FRAG6_CB(prev)->offset < offset) {
 294                 next = NULL;
 295                 goto found;
 296         }
 297         prev = NULL;
 298         for(next = fq->q.fragments; next != NULL; next = next->next) {
 299                 if (FRAG6_CB(next)->offset >= offset)
 300                         break;  /* bingo! */
 301                 prev = next;
 302         }
 303
 304 found:
 305         /* RFC5722, Section 4, amended by Errata ID : 3089
 306          *                          When reassembling an IPv6 datagram, if
 307          *   one or more its constituent fragments is determined to be an
 308          *   overlapping fragment, the entire datagram (and any constituent
 309          *   fragments) MUST be silently discarded.
 310          */
 311
 312         /* Check for overlap with preceding fragment. */
 313         if (prev &&
 314             (FRAG6_CB(prev)->offset + prev->len) > offset)
 315                 goto discard_fq;
 316
 317         /* Look for overlap with succeeding segment. */
 318         if (next && FRAG6_CB(next)->offset < end)
 319                 goto discard_fq;
 320
 321         FRAG6_CB(skb)->offset = offset;
 322
 323         /* Insert this fragment in the chain of fragments. */
 324         skb->next = next;
 325         if (!next)
 326                 fq->q.fragments_tail = skb;
 327         if (prev)
 328                 prev->next = skb;
 329         else
 330                 fq->q.fragments = skb;
 331
 332         dev = skb->dev;
 333         if (dev) {
 334                 fq->iif = dev->ifindex;
 335                 skb->dev = NULL;
 336         }
 337         fq->q.stamp = skb->tstamp;
 338         fq->q.meat += skb->len;
 339         atomic_add(skb->truesize, &fq->q.net->mem);
 340
 341         /* The first fragment.
 342          * nhoffset is obtained from the first fragment, of course.
 343          */
 344         if (offset == 0) {
 345                 fq->nhoffset = nhoff;
 346                 fq->q.last_in |= INET_FRAG_FIRST_IN;
 347         }
 348
 349         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 350             fq->q.meat == fq->q.len)
 351                 return ip6_frag_reasm(fq, prev, dev);
 352
 353         write_lock(&ip6_frags.lock);
 354         list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
 355         write_unlock(&ip6_frags.lock);
 356         return -1;
 357
 358 discard_fq:
 359         inet_frag_kill(&fq->q, &ip6_frags);
 360 err:
 361         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 362                       IPSTATS_MIB_REASMFAILS);
 363         kfree_skb(skb);
 364         return -1;
 365 }
 366
 367 /*
 368  *      Check if this packet is complete.
 369  *      Returns NULL on failure by any reason, and pointer
 370  *      to current nexthdr field in reassembled frame.
 371  *
 372  *      It is called with locked fq, and caller must check that
 373  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 374  *      the last and the first frames arrived and all the bits are here.
 375  */
 376 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 377                           struct net_device *dev)
 378 {
 379         struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
 380         struct sk_buff *fp, *head = fq->q.fragments;
 381         int    payload_len;
 382         unsigned int nhoff;
 383         int sum_truesize;
 384
 385         inet_frag_kill(&fq->q, &ip6_frags);
 386
 387         /* Make the one we just received the head. */
 388         if (prev) {
 389                 head = prev->next;
 390                 fp = skb_clone(head, GFP_ATOMIC);
 391
 392                 if (!fp)
 393                         goto out_oom;
 394
 395                 fp->next = head->next;
 396                 if (!fp->next)
 397                         fq->q.fragments_tail = fp;
 398                 prev->next = fp;
 399
 400                 skb_morph(head, fq->q.fragments);
 401                 head->next = fq->q.fragments->next;
 402
 403                 consume_skb(fq->q.fragments);
 404                 fq->q.fragments = head;
 405         }
 406
 407         WARN_ON(head == NULL);
 408         WARN_ON(FRAG6_CB(head)->offset != 0);
 409
 410         /* Unfragmented part is taken from the first segment. */
 411         payload_len = ((head->data - skb_network_header(head)) -
 412                        sizeof(struct ipv6hdr) + fq->q.len -
 413                        sizeof(struct frag_hdr));
 414         if (payload_len > IPV6_MAXPLEN)
 415                 goto out_oversize;
 416
 417         /* Head of list must not be cloned. */
 418         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 419                 goto out_oom;
 420
 421         /* If the first fragment is fragmented itself, we split
 422          * it to two chunks: the first with data and paged part
 423          * and the second, holding only fragments. */
 424         if (skb_has_frag_list(head)) {
 425                 struct sk_buff *clone;
 426                 int i, plen = 0;
 427
 428                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 429                         goto out_oom;
 430                 clone->next = head->next;
 431                 head->next = clone;
 432                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 433                 skb_frag_list_init(head);
 434                 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
 435                         plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
 436                 clone->len = clone->data_len = head->data_len - plen;
 437                 head->data_len -= clone->len;
 438                 head->len -= clone->len;
 439                 clone->csum = 0;
 440                 clone->ip_summed = head->ip_summed;
 441                 atomic_add(clone->truesize, &fq->q.net->mem);
 442         }
 443
 444         /* We have to remove fragment header from datagram and to relocate
 445          * header in order to calculate ICV correctly. */
 446         nhoff = fq->nhoffset;
 447         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 448         memmove(head->head + sizeof(struct frag_hdr), head->head,
 449                 (head->data - head->head) - sizeof(struct frag_hdr));
 450         head->mac_header += sizeof(struct frag_hdr);
 451         head->network_header += sizeof(struct frag_hdr);
 452
 453         skb_reset_transport_header(head);
 454         skb_push(head, head->data - skb_network_header(head));
 455
 456         sum_truesize = head->truesize;
 457         for (fp = head->next; fp;) {
 458                 bool headstolen;
 459                 int delta;
 460                 struct sk_buff *next = fp->next;
 461
 462                 sum_truesize += fp->truesize;
 463                 if (head->ip_summed != fp->ip_summed)
 464                         head->ip_summed = CHECKSUM_NONE;
 465                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 466                         head->csum = csum_add(head->csum, fp->csum);
 467
 468                 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
 469                         kfree_skb_partial(fp, headstolen);
 470                 } else {
 471                         if (!skb_shinfo(head)->frag_list)
 472                                 skb_shinfo(head)->frag_list = fp;
 473                         head->data_len += fp->len;
 474                         head->len += fp->len;
 475                         head->truesize += fp->truesize;
 476                 }
 477                 fp = next;
 478         }
 479         atomic_sub(sum_truesize, &fq->q.net->mem);
 480
 481         head->next = NULL;
 482         head->dev = dev;
 483         head->tstamp = fq->q.stamp;
 484         ipv6_hdr(head)->payload_len = htons(payload_len);
 485         IP6CB(head)->nhoff = nhoff;
 486
 487         /* Yes, and fold redundant checksum back. 8) */
 488         if (head->ip_summed == CHECKSUM_COMPLETE)
 489                 head->csum = csum_partial(skb_network_header(head),
 490                                           skb_network_header_len(head),
 491                                           head->csum);
 492
 493         rcu_read_lock();
 494         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 495         rcu_read_unlock();
 496         fq->q.fragments = NULL;
 497         fq->q.fragments_tail = NULL;
 498         return 1;
 499
 500 out_oversize:
 501         net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
 502         goto out_fail;
 503 out_oom:
 504         net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
 505 out_fail:
 506         rcu_read_lock();
 507         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 508         rcu_read_unlock();
 509         return -1;
 510 }
 511
 512 static int ipv6_frag_rcv(struct sk_buff *skb)
 513 {
 514         struct frag_hdr *fhdr;
 515         struct frag_queue *fq;
 516         const struct ipv6hdr *hdr = ipv6_hdr(skb);
 517         struct net *net = dev_net(skb_dst(skb)->dev);
 518
 519         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
 520
 521         /* Jumbo payload inhibits frag. header */
 522         if (hdr->payload_len==0)
 523                 goto fail_hdr;
 524
 525         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 526                                  sizeof(struct frag_hdr))))
 527                 goto fail_hdr;
 528
 529         hdr = ipv6_hdr(skb);
 530         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 531
 532         if (!(fhdr->frag_off & htons(0xFFF9))) {
 533                 /* It is not a fragmented frame */
 534                 skb->transport_header += sizeof(struct frag_hdr);
 535                 IP6_INC_STATS_BH(net,
 536                                  ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
 537
 538                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 539                 return 1;
 540         }
 541
 542         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
 543                 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
 544
 545         fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
 546         if (fq != NULL) {
 547                 int ret;
 548
 549                 spin_lock(&fq->q.lock);
 550
 551                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 552
 553                 spin_unlock(&fq->q.lock);
 554                 inet_frag_put(&fq->q, &ip6_frags);
 555                 return ret;
 556         }
 557
 558         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
 559         kfree_skb(skb);
 560         return -1;
 561
 562 fail_hdr:
 563         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
 564         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
 565         return -1;
 566 }
 567
 568 static const struct inet6_protocol frag_protocol =
 569 {
 570         .handler        =       ipv6_frag_rcv,
 571         .flags          =       INET6_PROTO_NOPOLICY,
 572 };
 573
 574 #ifdef CONFIG_SYSCTL
 575 static struct ctl_table ip6_frags_ns_ctl_table[] = {
 576         {
 577                 .procname       = "ip6frag_high_thresh",
 578                 .data           = &init_net.ipv6.frags.high_thresh,
 579                 .maxlen         = sizeof(int),
 580                 .mode           = 0644,
 581                 .proc_handler   = proc_dointvec
 582         },
 583         {
 584                 .procname       = "ip6frag_low_thresh",
 585                 .data           = &init_net.ipv6.frags.low_thresh,
 586                 .maxlen         = sizeof(int),
 587                 .mode           = 0644,
 588                 .proc_handler   = proc_dointvec
 589         },
 590         {
 591                 .procname       = "ip6frag_time",
 592                 .data           = &init_net.ipv6.frags.timeout,
 593                 .maxlen         = sizeof(int),
 594                 .mode           = 0644,
 595                 .proc_handler   = proc_dointvec_jiffies,
 596         },
 597         { }
 598 };
 599
 600 static struct ctl_table ip6_frags_ctl_table[] = {
 601         {
 602                 .procname       = "ip6frag_secret_interval",
 603                 .data           = &ip6_frags.secret_interval,
 604                 .maxlen         = sizeof(int),
 605                 .mode           = 0644,
 606                 .proc_handler   = proc_dointvec_jiffies,
 607         },
 608         { }
 609 };
 610
 611 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
 612 {
 613         struct ctl_table *table;
 614         struct ctl_table_header *hdr;
 615
 616         table = ip6_frags_ns_ctl_table;
 617         if (!net_eq(net, &init_net)) {
 618                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
 619                 if (table == NULL)
 620                         goto err_alloc;
 621
 622                 table[0].data = &net->ipv6.frags.high_thresh;
 623                 table[1].data = &net->ipv6.frags.low_thresh;
 624                 table[2].data = &net->ipv6.frags.timeout;
 625         }
 626
 627         hdr = register_net_sysctl(net, "net/ipv6", table);
 628         if (hdr == NULL)
 629                 goto err_reg;
 630
 631         net->ipv6.sysctl.frags_hdr = hdr;
 632         return 0;
 633
 634 err_reg:
 635         if (!net_eq(net, &init_net))
 636                 kfree(table);
 637 err_alloc:
 638         return -ENOMEM;
 639 }
 640
 641 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
 642 {
 643         struct ctl_table *table;
 644
 645         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
 646         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
 647         if (!net_eq(net, &init_net))
 648                 kfree(table);
 649 }
 650
 651 static struct ctl_table_header *ip6_ctl_header;
 652
 653 static int ip6_frags_sysctl_register(void)
 654 {
 655         ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
 656                         ip6_frags_ctl_table);
 657         return ip6_ctl_header == NULL ? -ENOMEM : 0;
 658 }
 659
 660 static void ip6_frags_sysctl_unregister(void)
 661 {
 662         unregister_net_sysctl_table(ip6_ctl_header);
 663 }
 664 #else
 665 static inline int ip6_frags_ns_sysctl_register(struct net *net)
 666 {
 667         return 0;
 668 }
 669
 670 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
 671 {
 672 }
 673
 674 static inline int ip6_frags_sysctl_register(void)
 675 {
 676         return 0;
 677 }
 678
 679 static inline void ip6_frags_sysctl_unregister(void)
 680 {
 681 }
 682 #endif
 683
 684 static int __net_init ipv6_frags_init_net(struct net *net)
 685 {
 686         net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
 687         net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
 688         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
 689
 690         inet_frags_init_net(&net->ipv6.frags);
 691
 692         return ip6_frags_ns_sysctl_register(net);
 693 }
 694
 695 static void __net_exit ipv6_frags_exit_net(struct net *net)
 696 {
 697         ip6_frags_ns_sysctl_unregister(net);
 698         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 699 }
 700
 701 static struct pernet_operations ip6_frags_ops = {
 702         .init = ipv6_frags_init_net,
 703         .exit = ipv6_frags_exit_net,
 704 };
 705
 706 int __init ipv6_frag_init(void)
 707 {
 708         int ret;
 709
 710         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 711         if (ret)
 712                 goto out;
 713
 714         ret = ip6_frags_sysctl_register();
 715         if (ret)
 716                 goto err_sysctl;
 717
 718         ret = register_pernet_subsys(&ip6_frags_ops);
 719         if (ret)
 720                 goto err_pernet;
 721
 722         ip6_frags.hashfn = ip6_hashfn;
 723         ip6_frags.constructor = ip6_frag_init;
 724         ip6_frags.destructor = NULL;
 725         ip6_frags.skb_free = NULL;
 726         ip6_frags.qsize = sizeof(struct frag_queue);
 727         ip6_frags.match = ip6_frag_match;
 728         ip6_frags.frag_expire = ip6_frag_expire;
 729         ip6_frags.secret_interval = 10 * 60 * HZ;
 730         inet_frags_init(&ip6_frags);
 731 out:
 732         return ret;
 733
 734 err_pernet:
 735         ip6_frags_sysctl_unregister();
 736 err_sysctl:
 737         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 738         goto out;
 739 }
 740
 741 void ipv6_frag_exit(void)
 742 {
 743         inet_frags_fini(&ip6_frags);
 744         ip6_frags_sysctl_unregister();
 745         unregister_pernet_subsys(&ip6_frags_ops);
 746         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 747 }