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 c;
 108
 109         c = jhash_3words((__force u32)saddr->s6_addr32[0],
 110                          (__force u32)saddr->s6_addr32[1],
 111                          (__force u32)saddr->s6_addr32[2],
 112                          rnd);
 113
 114         c = jhash_3words((__force u32)saddr->s6_addr32[3],
 115                          (__force u32)daddr->s6_addr32[0],
 116                          (__force u32)daddr->s6_addr32[1],
 117                          c);
 118
 119         c =  jhash_3words((__force u32)daddr->s6_addr32[2],
 120                           (__force u32)daddr->s6_addr32[3],
 121                           (__force u32)id,
 122                           c);
 123
 124         return c & (INETFRAGS_HASHSZ - 1);
 125 }
 126 EXPORT_SYMBOL_GPL(inet6_hash_frag);
 127
 128 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 129 {
 130         struct frag_queue *fq;
 131
 132         fq = container_of(q, struct frag_queue, q);
 133         return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
 134 }
 135
 136 int ip6_frag_match(struct inet_frag_queue *q, void *a)
 137 {
 138         struct frag_queue *fq;
 139         struct ip6_create_arg *arg = a;
 140
 141         fq = container_of(q, struct frag_queue, q);
 142         return (fq->id == arg->id && fq->user == arg->user &&
 143                         ipv6_addr_equal(&fq->saddr, arg->src) &&
 144                         ipv6_addr_equal(&fq->daddr, arg->dst));
 145 }
 146 EXPORT_SYMBOL(ip6_frag_match);
 147
 148 void ip6_frag_init(struct inet_frag_queue *q, void *a)
 149 {
 150         struct frag_queue *fq = container_of(q, struct frag_queue, q);
 151         struct ip6_create_arg *arg = a;
 152
 153         fq->id = arg->id;
 154         fq->user = arg->user;
 155         ipv6_addr_copy(&fq->saddr, arg->src);
 156         ipv6_addr_copy(&fq->daddr, arg->dst);
 157 }
 158 EXPORT_SYMBOL(ip6_frag_init);
 159
 160 /* Destruction primitives. */
 161
 162 static __inline__ void fq_put(struct frag_queue *fq)
 163 {
 164         inet_frag_put(&fq->q, &ip6_frags);
 165 }
 166
 167 /* Kill fq entry. It is not destroyed immediately,
 168  * because caller (and someone more) holds reference count.
 169  */
 170 static __inline__ void fq_kill(struct frag_queue *fq)
 171 {
 172         inet_frag_kill(&fq->q, &ip6_frags);
 173 }
 174
 175 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 176 {
 177         int evicted;
 178
 179         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
 180         if (evicted)
 181                 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
 182 }
 183
 184 static void ip6_frag_expire(unsigned long data)
 185 {
 186         struct frag_queue *fq;
 187         struct net_device *dev = NULL;
 188         struct net *net;
 189
 190         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 191
 192         spin_lock(&fq->q.lock);
 193
 194         if (fq->q.last_in & INET_FRAG_COMPLETE)
 195                 goto out;
 196
 197         fq_kill(fq);
 198
 199         net = container_of(fq->q.net, struct net, ipv6.frags);
 200         rcu_read_lock();
 201         dev = dev_get_by_index_rcu(net, fq->iif);
 202         if (!dev)
 203                 goto out_rcu_unlock;
 204
 205         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 206         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 207
 208         /* Don't send error if the first segment did not arrive. */
 209         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
 210                 goto out_rcu_unlock;
 211
 212         /*
 213            But use as source device on which LAST ARRIVED
 214            segment was received. And do not use fq->dev
 215            pointer directly, device might already disappeared.
 216          */
 217         fq->q.fragments->dev = dev;
 218         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
 219 out_rcu_unlock:
 220         rcu_read_unlock();
 221 out:
 222         spin_unlock(&fq->q.lock);
 223         fq_put(fq);
 224 }
 225
 226 static __inline__ struct frag_queue *
 227 fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
 228 {
 229         struct inet_frag_queue *q;
 230         struct ip6_create_arg arg;
 231         unsigned int hash;
 232
 233         arg.id = id;
 234         arg.user = IP6_DEFRAG_LOCAL_DELIVER;
 235         arg.src = src;
 236         arg.dst = dst;
 237
 238         read_lock(&ip6_frags.lock);
 239         hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
 240
 241         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
 242         if (q == NULL)
 243                 return NULL;
 244
 245         return container_of(q, struct frag_queue, q);
 246 }
 247
 248 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 249                            struct frag_hdr *fhdr, int nhoff)
 250 {
 251         struct sk_buff *prev, *next;
 252         struct net_device *dev;
 253         int offset, end;
 254         struct net *net = dev_net(skb_dst(skb)->dev);
 255
 256         if (fq->q.last_in & INET_FRAG_COMPLETE)
 257                 goto err;
 258
 259         offset = ntohs(fhdr->frag_off) & ~0x7;
 260         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 261                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 262
 263         if ((unsigned int)end > IPV6_MAXPLEN) {
 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                                   ((u8 *)&fhdr->frag_off -
 268                                    skb_network_header(skb)));
 269                 return -1;
 270         }
 271
 272         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 273                 const unsigned char *nh = skb_network_header(skb);
 274                 skb->csum = csum_sub(skb->csum,
 275                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 276                                                   0));
 277         }
 278
 279         /* Is this the final fragment? */
 280         if (!(fhdr->frag_off & htons(IP6_MF))) {
 281                 /* If we already have some bits beyond end
 282                  * or have different end, the segment is corrupted.
 283                  */
 284                 if (end < fq->q.len ||
 285                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
 286                         goto err;
 287                 fq->q.last_in |= INET_FRAG_LAST_IN;
 288                 fq->q.len = end;
 289         } else {
 290                 /* Check if the fragment is rounded to 8 bytes.
 291                  * Required by the RFC.
 292                  */
 293                 if (end & 0x7) {
 294                         /* RFC2460 says always send parameter problem in
 295                          * this case. -DaveM
 296                          */
 297                         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 298                                          IPSTATS_MIB_INHDRERRORS);
 299                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 300                                           offsetof(struct ipv6hdr, payload_len));
 301                         return -1;
 302                 }
 303                 if (end > fq->q.len) {
 304                         /* Some bits beyond end -> corruption. */
 305                         if (fq->q.last_in & INET_FRAG_LAST_IN)
 306                                 goto err;
 307                         fq->q.len = end;
 308                 }
 309         }
 310
 311         if (end == offset)
 312                 goto err;
 313
 314         /* Point into the IP datagram 'data' part. */
 315         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 316                 goto err;
 317
 318         if (pskb_trim_rcsum(skb, end - offset))
 319                 goto err;
 320
 321         /* Find out which fragments are in front and at the back of us
 322          * in the chain of fragments so far.  We must know where to put
 323          * this fragment, right?
 324          */
 325         prev = fq->q.fragments_tail;
 326         if (!prev || FRAG6_CB(prev)->offset < offset) {
 327                 next = NULL;
 328                 goto found;
 329         }
 330         prev = NULL;
 331         for(next = fq->q.fragments; next != NULL; next = next->next) {
 332                 if (FRAG6_CB(next)->offset >= offset)
 333                         break;  /* bingo! */
 334                 prev = next;
 335         }
 336
 337 found:
 338         /* RFC5722, Section 4:
 339          *                                  When reassembling an IPv6 datagram, if
 340          *   one or more its constituent fragments is determined to be an
 341          *   overlapping fragment, the entire datagram (and any constituent
 342          *   fragments, including those not yet received) MUST be silently
 343          *   discarded.
 344          */
 345
 346         /* Check for overlap with preceding fragment. */
 347         if (prev &&
 348             (FRAG6_CB(prev)->offset + prev->len) > offset)
 349                 goto discard_fq;
 350
 351         /* Look for overlap with succeeding segment. */
 352         if (next && FRAG6_CB(next)->offset < end)
 353                 goto discard_fq;
 354
 355         FRAG6_CB(skb)->offset = offset;
 356
 357         /* Insert this fragment in the chain of fragments. */
 358         skb->next = next;
 359         if (!next)
 360                 fq->q.fragments_tail = skb;
 361         if (prev)
 362                 prev->next = skb;
 363         else
 364                 fq->q.fragments = skb;
 365
 366         dev = skb->dev;
 367         if (dev) {
 368                 fq->iif = dev->ifindex;
 369                 skb->dev = NULL;
 370         }
 371         fq->q.stamp = skb->tstamp;
 372         fq->q.meat += skb->len;
 373         atomic_add(skb->truesize, &fq->q.net->mem);
 374
 375         /* The first fragment.
 376          * nhoffset is obtained from the first fragment, of course.
 377          */
 378         if (offset == 0) {
 379                 fq->nhoffset = nhoff;
 380                 fq->q.last_in |= INET_FRAG_FIRST_IN;
 381         }
 382
 383         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 384             fq->q.meat == fq->q.len)
 385                 return ip6_frag_reasm(fq, prev, dev);
 386
 387         write_lock(&ip6_frags.lock);
 388         list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
 389         write_unlock(&ip6_frags.lock);
 390         return -1;
 391
 392 discard_fq:
 393         fq_kill(fq);
 394 err:
 395         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 396                       IPSTATS_MIB_REASMFAILS);
 397         kfree_skb(skb);
 398         return -1;
 399 }
 400
 401 /*
 402  *      Check if this packet is complete.
 403  *      Returns NULL on failure by any reason, and pointer
 404  *      to current nexthdr field in reassembled frame.
 405  *
 406  *      It is called with locked fq, and caller must check that
 407  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 408  *      the last and the first frames arrived and all the bits are here.
 409  */
 410 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 411                           struct net_device *dev)
 412 {
 413         struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
 414         struct sk_buff *fp, *head = fq->q.fragments;
 415         int    payload_len;
 416         unsigned int nhoff;
 417
 418         fq_kill(fq);
 419
 420         /* Make the one we just received the head. */
 421         if (prev) {
 422                 head = prev->next;
 423                 fp = skb_clone(head, GFP_ATOMIC);
 424
 425                 if (!fp)
 426                         goto out_oom;
 427
 428                 fp->next = head->next;
 429                 if (!fp->next)
 430                         fq->q.fragments_tail = fp;
 431                 prev->next = fp;
 432
 433                 skb_morph(head, fq->q.fragments);
 434                 head->next = fq->q.fragments->next;
 435
 436                 kfree_skb(fq->q.fragments);
 437                 fq->q.fragments = head;
 438         }
 439
 440         WARN_ON(head == NULL);
 441         WARN_ON(FRAG6_CB(head)->offset != 0);
 442
 443         /* Unfragmented part is taken from the first segment. */
 444         payload_len = ((head->data - skb_network_header(head)) -
 445                        sizeof(struct ipv6hdr) + fq->q.len -
 446                        sizeof(struct frag_hdr));
 447         if (payload_len > IPV6_MAXPLEN)
 448                 goto out_oversize;
 449
 450         /* Head of list must not be cloned. */
 451         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 452                 goto out_oom;
 453
 454         /* If the first fragment is fragmented itself, we split
 455          * it to two chunks: the first with data and paged part
 456          * and the second, holding only fragments. */
 457         if (skb_has_frag_list(head)) {
 458                 struct sk_buff *clone;
 459                 int i, plen = 0;
 460
 461                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 462                         goto out_oom;
 463                 clone->next = head->next;
 464                 head->next = clone;
 465                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 466                 skb_frag_list_init(head);
 467                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 468                         plen += skb_shinfo(head)->frags[i].size;
 469                 clone->len = clone->data_len = head->data_len - plen;
 470                 head->data_len -= clone->len;
 471                 head->len -= clone->len;
 472                 clone->csum = 0;
 473                 clone->ip_summed = head->ip_summed;
 474                 atomic_add(clone->truesize, &fq->q.net->mem);
 475         }
 476
 477         /* We have to remove fragment header from datagram and to relocate
 478          * header in order to calculate ICV correctly. */
 479         nhoff = fq->nhoffset;
 480         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 481         memmove(head->head + sizeof(struct frag_hdr), head->head,
 482                 (head->data - head->head) - sizeof(struct frag_hdr));
 483         head->mac_header += sizeof(struct frag_hdr);
 484         head->network_header += sizeof(struct frag_hdr);
 485
 486         skb_shinfo(head)->frag_list = head->next;
 487         skb_reset_transport_header(head);
 488         skb_push(head, head->data - skb_network_header(head));
 489
 490         for (fp=head->next; fp; fp = fp->next) {
 491                 head->data_len += fp->len;
 492                 head->len += fp->len;
 493                 if (head->ip_summed != fp->ip_summed)
 494                         head->ip_summed = CHECKSUM_NONE;
 495                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 496                         head->csum = csum_add(head->csum, fp->csum);
 497                 head->truesize += fp->truesize;
 498         }
 499         atomic_sub(head->truesize, &fq->q.net->mem);
 500
 501         head->next = NULL;
 502         head->dev = dev;
 503         head->tstamp = fq->q.stamp;
 504         ipv6_hdr(head)->payload_len = htons(payload_len);
 505         IP6CB(head)->nhoff = nhoff;
 506
 507         /* Yes, and fold redundant checksum back. 8) */
 508         if (head->ip_summed == CHECKSUM_COMPLETE)
 509                 head->csum = csum_partial(skb_network_header(head),
 510                                           skb_network_header_len(head),
 511                                           head->csum);
 512
 513         rcu_read_lock();
 514         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 515         rcu_read_unlock();
 516         fq->q.fragments = NULL;
 517         fq->q.fragments_tail = NULL;
 518         return 1;
 519
 520 out_oversize:
 521         if (net_ratelimit())
 522                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
 523         goto out_fail;
 524 out_oom:
 525         if (net_ratelimit())
 526                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
 527 out_fail:
 528         rcu_read_lock();
 529         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 530         rcu_read_unlock();
 531         return -1;
 532 }
 533
 534 static int ipv6_frag_rcv(struct sk_buff *skb)
 535 {
 536         struct frag_hdr *fhdr;
 537         struct frag_queue *fq;
 538         const struct ipv6hdr *hdr = ipv6_hdr(skb);
 539         struct net *net = dev_net(skb_dst(skb)->dev);
 540
 541         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
 542
 543         /* Jumbo payload inhibits frag. header */
 544         if (hdr->payload_len==0)
 545                 goto fail_hdr;
 546
 547         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 548                                  sizeof(struct frag_hdr))))
 549                 goto fail_hdr;
 550
 551         hdr = ipv6_hdr(skb);
 552         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 553
 554         if (!(fhdr->frag_off & htons(0xFFF9))) {
 555                 /* It is not a fragmented frame */
 556                 skb->transport_header += sizeof(struct frag_hdr);
 557                 IP6_INC_STATS_BH(net,
 558                                  ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
 559
 560                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 561                 return 1;
 562         }
 563
 564         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
 565                 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
 566
 567         fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
 568         if (fq != NULL) {
 569                 int ret;
 570
 571                 spin_lock(&fq->q.lock);
 572
 573                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 574
 575                 spin_unlock(&fq->q.lock);
 576                 fq_put(fq);
 577                 return ret;
 578         }
 579
 580         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
 581         kfree_skb(skb);
 582         return -1;
 583
 584 fail_hdr:
 585         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
 586         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
 587         return -1;
 588 }
 589
 590 static const struct inet6_protocol frag_protocol =
 591 {
 592         .handler        =       ipv6_frag_rcv,
 593         .flags          =       INET6_PROTO_NOPOLICY,
 594 };
 595
 596 #ifdef CONFIG_SYSCTL
 597 static struct ctl_table ip6_frags_ns_ctl_table[] = {
 598         {
 599                 .procname       = "ip6frag_high_thresh",
 600                 .data           = &init_net.ipv6.frags.high_thresh,
 601                 .maxlen         = sizeof(int),
 602                 .mode           = 0644,
 603                 .proc_handler   = proc_dointvec
 604         },
 605         {
 606                 .procname       = "ip6frag_low_thresh",
 607                 .data           = &init_net.ipv6.frags.low_thresh,
 608                 .maxlen         = sizeof(int),
 609                 .mode           = 0644,
 610                 .proc_handler   = proc_dointvec
 611         },
 612         {
 613                 .procname       = "ip6frag_time",
 614                 .data           = &init_net.ipv6.frags.timeout,
 615                 .maxlen         = sizeof(int),
 616                 .mode           = 0644,
 617                 .proc_handler   = proc_dointvec_jiffies,
 618         },
 619         { }
 620 };
 621
 622 static struct ctl_table ip6_frags_ctl_table[] = {
 623         {
 624                 .procname       = "ip6frag_secret_interval",
 625                 .data           = &ip6_frags.secret_interval,
 626                 .maxlen         = sizeof(int),
 627                 .mode           = 0644,
 628                 .proc_handler   = proc_dointvec_jiffies,
 629         },
 630         { }
 631 };
 632
 633 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
 634 {
 635         struct ctl_table *table;
 636         struct ctl_table_header *hdr;
 637
 638         table = ip6_frags_ns_ctl_table;
 639         if (!net_eq(net, &init_net)) {
 640                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
 641                 if (table == NULL)
 642                         goto err_alloc;
 643
 644                 table[0].data = &net->ipv6.frags.high_thresh;
 645                 table[1].data = &net->ipv6.frags.low_thresh;
 646                 table[2].data = &net->ipv6.frags.timeout;
 647         }
 648
 649         hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
 650         if (hdr == NULL)
 651                 goto err_reg;
 652
 653         net->ipv6.sysctl.frags_hdr = hdr;
 654         return 0;
 655
 656 err_reg:
 657         if (!net_eq(net, &init_net))
 658                 kfree(table);
 659 err_alloc:
 660         return -ENOMEM;
 661 }
 662
 663 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
 664 {
 665         struct ctl_table *table;
 666
 667         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
 668         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
 669         if (!net_eq(net, &init_net))
 670                 kfree(table);
 671 }
 672
 673 static struct ctl_table_header *ip6_ctl_header;
 674
 675 static int ip6_frags_sysctl_register(void)
 676 {
 677         ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
 678                         ip6_frags_ctl_table);
 679         return ip6_ctl_header == NULL ? -ENOMEM : 0;
 680 }
 681
 682 static void ip6_frags_sysctl_unregister(void)
 683 {
 684         unregister_net_sysctl_table(ip6_ctl_header);
 685 }
 686 #else
 687 static inline int ip6_frags_ns_sysctl_register(struct net *net)
 688 {
 689         return 0;
 690 }
 691
 692 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
 693 {
 694 }
 695
 696 static inline int ip6_frags_sysctl_register(void)
 697 {
 698         return 0;
 699 }
 700
 701 static inline void ip6_frags_sysctl_unregister(void)
 702 {
 703 }
 704 #endif
 705
 706 static int __net_init ipv6_frags_init_net(struct net *net)
 707 {
 708         net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
 709         net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
 710         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
 711
 712         inet_frags_init_net(&net->ipv6.frags);
 713
 714         return ip6_frags_ns_sysctl_register(net);
 715 }
 716
 717 static void __net_exit ipv6_frags_exit_net(struct net *net)
 718 {
 719         ip6_frags_ns_sysctl_unregister(net);
 720         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 721 }
 722
 723 static struct pernet_operations ip6_frags_ops = {
 724         .init = ipv6_frags_init_net,
 725         .exit = ipv6_frags_exit_net,
 726 };
 727
 728 int __init ipv6_frag_init(void)
 729 {
 730         int ret;
 731
 732         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 733         if (ret)
 734                 goto out;
 735
 736         ret = ip6_frags_sysctl_register();
 737         if (ret)
 738                 goto err_sysctl;
 739
 740         ret = register_pernet_subsys(&ip6_frags_ops);
 741         if (ret)
 742                 goto err_pernet;
 743
 744         ip6_frags.hashfn = ip6_hashfn;
 745         ip6_frags.constructor = ip6_frag_init;
 746         ip6_frags.destructor = NULL;
 747         ip6_frags.skb_free = NULL;
 748         ip6_frags.qsize = sizeof(struct frag_queue);
 749         ip6_frags.match = ip6_frag_match;
 750         ip6_frags.frag_expire = ip6_frag_expire;
 751         ip6_frags.secret_interval = 10 * 60 * HZ;
 752         inet_frags_init(&ip6_frags);
 753 out:
 754         return ret;
 755
 756 err_pernet:
 757         ip6_frags_sysctl_unregister();
 758 err_sysctl:
 759         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 760         goto out;
 761 }
 762
 763 void ipv6_frag_exit(void)
 764 {
 765         inet_frags_fini(&ip6_frags);
 766         ip6_frags_sysctl_unregister();
 767         unregister_pernet_subsys(&ip6_frags_ops);
 768         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 769 }