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 /*
  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         u32                     user;
  78         struct in6_addr         saddr;
  79         struct in6_addr         daddr;
  80
  81         int                     iif;
  82         unsigned int            csum;
  83         __u16                   nhoffset;
  84 };
  85
  86 static struct inet_frags ip6_frags;
  87
  88 int ip6_frag_nqueues(struct net *net)
  89 {
  90         return net->ipv6.frags.nqueues;
  91 }
  92
  93 int ip6_frag_mem(struct net *net)
  94 {
  95         return atomic_read(&net->ipv6.frags.mem);
  96 }
  97
  98 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
  99                           struct net_device *dev);
 100
 101 /*
 102  * callers should be careful not to use the hash value outside the ipfrag_lock
 103  * as doing so could race with ipfrag_hash_rnd being recalculated.
 104  */
 105 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
 106                              const struct in6_addr *daddr, u32 rnd)
 107 {
 108         u32 c;
 109
 110         c = jhash_3words((__force u32)saddr->s6_addr32[0],
 111                          (__force u32)saddr->s6_addr32[1],
 112                          (__force u32)saddr->s6_addr32[2],
 113                          rnd);
 114
 115         c = jhash_3words((__force u32)saddr->s6_addr32[3],
 116                          (__force u32)daddr->s6_addr32[0],
 117                          (__force u32)daddr->s6_addr32[1],
 118                          c);
 119
 120         c =  jhash_3words((__force u32)daddr->s6_addr32[2],
 121                           (__force u32)daddr->s6_addr32[3],
 122                           (__force u32)id,
 123                           c);
 124
 125         return c & (INETFRAGS_HASHSZ - 1);
 126 }
 127 EXPORT_SYMBOL_GPL(inet6_hash_frag);
 128
 129 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 130 {
 131         struct frag_queue *fq;
 132
 133         fq = container_of(q, struct frag_queue, q);
 134         return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
 135 }
 136
 137 bool ip6_frag_match(struct inet_frag_queue *q, void *a)
 138 {
 139         struct frag_queue *fq;
 140         struct ip6_create_arg *arg = a;
 141
 142         fq = container_of(q, struct frag_queue, q);
 143         return  fq->id == arg->id &&
 144                 fq->user == arg->user &&
 145                 ipv6_addr_equal(&fq->saddr, arg->src) &&
 146                 ipv6_addr_equal(&fq->daddr, arg->dst);
 147 }
 148 EXPORT_SYMBOL(ip6_frag_match);
 149
 150 void ip6_frag_init(struct inet_frag_queue *q, void *a)
 151 {
 152         struct frag_queue *fq = container_of(q, struct frag_queue, q);
 153         struct ip6_create_arg *arg = a;
 154
 155         fq->id = arg->id;
 156         fq->user = arg->user;
 157         fq->saddr = *arg->src;
 158         fq->daddr = *arg->dst;
 159 }
 160 EXPORT_SYMBOL(ip6_frag_init);
 161
 162 /* Destruction primitives. */
 163
 164 static __inline__ void fq_put(struct frag_queue *fq)
 165 {
 166         inet_frag_put(&fq->q, &ip6_frags);
 167 }
 168
 169 /* Kill fq entry. It is not destroyed immediately,
 170  * because caller (and someone more) holds reference count.
 171  */
 172 static __inline__ void fq_kill(struct frag_queue *fq)
 173 {
 174         inet_frag_kill(&fq->q, &ip6_frags);
 175 }
 176
 177 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 178 {
 179         int evicted;
 180
 181         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
 182         if (evicted)
 183                 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
 184 }
 185
 186 static void ip6_frag_expire(unsigned long data)
 187 {
 188         struct frag_queue *fq;
 189         struct net_device *dev = NULL;
 190         struct net *net;
 191
 192         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 193
 194         spin_lock(&fq->q.lock);
 195
 196         if (fq->q.last_in & INET_FRAG_COMPLETE)
 197                 goto out;
 198
 199         fq_kill(fq);
 200
 201         net = container_of(fq->q.net, struct net, ipv6.frags);
 202         rcu_read_lock();
 203         dev = dev_get_by_index_rcu(net, fq->iif);
 204         if (!dev)
 205                 goto out_rcu_unlock;
 206
 207         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 208         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 209
 210         /* Don't send error if the first segment did not arrive. */
 211         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
 212                 goto out_rcu_unlock;
 213
 214         /*
 215            But use as source device on which LAST ARRIVED
 216            segment was received. And do not use fq->dev
 217            pointer directly, device might already disappeared.
 218          */
 219         fq->q.fragments->dev = dev;
 220         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
 221 out_rcu_unlock:
 222         rcu_read_unlock();
 223 out:
 224         spin_unlock(&fq->q.lock);
 225         fq_put(fq);
 226 }
 227
 228 static __inline__ struct frag_queue *
 229 fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
 230 {
 231         struct inet_frag_queue *q;
 232         struct ip6_create_arg arg;
 233         unsigned int hash;
 234
 235         arg.id = id;
 236         arg.user = IP6_DEFRAG_LOCAL_DELIVER;
 237         arg.src = src;
 238         arg.dst = dst;
 239
 240         read_lock(&ip6_frags.lock);
 241         hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
 242
 243         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
 244         if (q == NULL)
 245                 return NULL;
 246
 247         return container_of(q, struct frag_queue, q);
 248 }
 249
 250 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 251                            struct frag_hdr *fhdr, int nhoff)
 252 {
 253         struct sk_buff *prev, *next;
 254         struct net_device *dev;
 255         int offset, end;
 256         struct net *net = dev_net(skb_dst(skb)->dev);
 257
 258         if (fq->q.last_in & INET_FRAG_COMPLETE)
 259                 goto err;
 260
 261         offset = ntohs(fhdr->frag_off) & ~0x7;
 262         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 263                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 264
 265         if ((unsigned int)end > IPV6_MAXPLEN) {
 266                 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 267                                  IPSTATS_MIB_INHDRERRORS);
 268                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 269                                   ((u8 *)&fhdr->frag_off -
 270                                    skb_network_header(skb)));
 271                 return -1;
 272         }
 273
 274         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 275                 const unsigned char *nh = skb_network_header(skb);
 276                 skb->csum = csum_sub(skb->csum,
 277                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 278                                                   0));
 279         }
 280
 281         /* Is this the final fragment? */
 282         if (!(fhdr->frag_off & htons(IP6_MF))) {
 283                 /* If we already have some bits beyond end
 284                  * or have different end, the segment is corrupted.
 285                  */
 286                 if (end < fq->q.len ||
 287                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
 288                         goto err;
 289                 fq->q.last_in |= INET_FRAG_LAST_IN;
 290                 fq->q.len = end;
 291         } else {
 292                 /* Check if the fragment is rounded to 8 bytes.
 293                  * Required by the RFC.
 294                  */
 295                 if (end & 0x7) {
 296                         /* RFC2460 says always send parameter problem in
 297                          * this case. -DaveM
 298                          */
 299                         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 300                                          IPSTATS_MIB_INHDRERRORS);
 301                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 302                                           offsetof(struct ipv6hdr, payload_len));
 303                         return -1;
 304                 }
 305                 if (end > fq->q.len) {
 306                         /* Some bits beyond end -> corruption. */
 307                         if (fq->q.last_in & INET_FRAG_LAST_IN)
 308                                 goto err;
 309                         fq->q.len = end;
 310                 }
 311         }
 312
 313         if (end == offset)
 314                 goto err;
 315
 316         /* Point into the IP datagram 'data' part. */
 317         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 318                 goto err;
 319
 320         if (pskb_trim_rcsum(skb, end - offset))
 321                 goto err;
 322
 323         /* Find out which fragments are in front and at the back of us
 324          * in the chain of fragments so far.  We must know where to put
 325          * this fragment, right?
 326          */
 327         prev = fq->q.fragments_tail;
 328         if (!prev || FRAG6_CB(prev)->offset < offset) {
 329                 next = NULL;
 330                 goto found;
 331         }
 332         prev = NULL;
 333         for(next = fq->q.fragments; next != NULL; next = next->next) {
 334                 if (FRAG6_CB(next)->offset >= offset)
 335                         break;  /* bingo! */
 336                 prev = next;
 337         }
 338
 339 found:
 340         /* RFC5722, Section 4, amended by Errata ID : 3089
 341          *                          When reassembling an IPv6 datagram, if
 342          *   one or more its constituent fragments is determined to be an
 343          *   overlapping fragment, the entire datagram (and any constituent
 344          *   fragments) MUST be silently discarded.
 345          */
 346
 347         /* Check for overlap with preceding fragment. */
 348         if (prev &&
 349             (FRAG6_CB(prev)->offset + prev->len) > offset)
 350                 goto discard_fq;
 351
 352         /* Look for overlap with succeeding segment. */
 353         if (next && FRAG6_CB(next)->offset < end)
 354                 goto discard_fq;
 355
 356         FRAG6_CB(skb)->offset = offset;
 357
 358         /* Insert this fragment in the chain of fragments. */
 359         skb->next = next;
 360         if (!next)
 361                 fq->q.fragments_tail = skb;
 362         if (prev)
 363                 prev->next = skb;
 364         else
 365                 fq->q.fragments = skb;
 366
 367         dev = skb->dev;
 368         if (dev) {
 369                 fq->iif = dev->ifindex;
 370                 skb->dev = NULL;
 371         }
 372         fq->q.stamp = skb->tstamp;
 373         fq->q.meat += skb->len;
 374         atomic_add(skb->truesize, &fq->q.net->mem);
 375
 376         /* The first fragment.
 377          * nhoffset is obtained from the first fragment, of course.
 378          */
 379         if (offset == 0) {
 380                 fq->nhoffset = nhoff;
 381                 fq->q.last_in |= INET_FRAG_FIRST_IN;
 382         }
 383
 384         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 385             fq->q.meat == fq->q.len)
 386                 return ip6_frag_reasm(fq, prev, dev);
 387
 388         write_lock(&ip6_frags.lock);
 389         list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
 390         write_unlock(&ip6_frags.lock);
 391         return -1;
 392
 393 discard_fq:
 394         fq_kill(fq);
 395 err:
 396         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 397                       IPSTATS_MIB_REASMFAILS);
 398         kfree_skb(skb);
 399         return -1;
 400 }
 401
 402 /*
 403  *      Check if this packet is complete.
 404  *      Returns NULL on failure by any reason, and pointer
 405  *      to current nexthdr field in reassembled frame.
 406  *
 407  *      It is called with locked fq, and caller must check that
 408  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 409  *      the last and the first frames arrived and all the bits are here.
 410  */
 411 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 412                           struct net_device *dev)
 413 {
 414         struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
 415         struct sk_buff *fp, *head = fq->q.fragments;
 416         int    payload_len;
 417         unsigned int nhoff;
 418
 419         fq_kill(fq);
 420
 421         /* Make the one we just received the head. */
 422         if (prev) {
 423                 head = prev->next;
 424                 fp = skb_clone(head, GFP_ATOMIC);
 425
 426                 if (!fp)
 427                         goto out_oom;
 428
 429                 fp->next = head->next;
 430                 if (!fp->next)
 431                         fq->q.fragments_tail = fp;
 432                 prev->next = fp;
 433
 434                 skb_morph(head, fq->q.fragments);
 435                 head->next = fq->q.fragments->next;
 436
 437                 consume_skb(fq->q.fragments);
 438                 fq->q.fragments = head;
 439         }
 440
 441         WARN_ON(head == NULL);
 442         WARN_ON(FRAG6_CB(head)->offset != 0);
 443
 444         /* Unfragmented part is taken from the first segment. */
 445         payload_len = ((head->data - skb_network_header(head)) -
 446                        sizeof(struct ipv6hdr) + fq->q.len -
 447                        sizeof(struct frag_hdr));
 448         if (payload_len > IPV6_MAXPLEN)
 449                 goto out_oversize;
 450
 451         /* Head of list must not be cloned. */
 452         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 453                 goto out_oom;
 454
 455         /* If the first fragment is fragmented itself, we split
 456          * it to two chunks: the first with data and paged part
 457          * and the second, holding only fragments. */
 458         if (skb_has_frag_list(head)) {
 459                 struct sk_buff *clone;
 460                 int i, plen = 0;
 461
 462                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 463                         goto out_oom;
 464                 clone->next = head->next;
 465                 head->next = clone;
 466                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 467                 skb_frag_list_init(head);
 468                 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
 469                         plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
 470                 clone->len = clone->data_len = head->data_len - plen;
 471                 head->data_len -= clone->len;
 472                 head->len -= clone->len;
 473                 clone->csum = 0;
 474                 clone->ip_summed = head->ip_summed;
 475                 atomic_add(clone->truesize, &fq->q.net->mem);
 476         }
 477
 478         /* We have to remove fragment header from datagram and to relocate
 479          * header in order to calculate ICV correctly. */
 480         nhoff = fq->nhoffset;
 481         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 482         memmove(head->head + sizeof(struct frag_hdr), head->head,
 483                 (head->data - head->head) - sizeof(struct frag_hdr));
 484         head->mac_header += sizeof(struct frag_hdr);
 485         head->network_header += sizeof(struct frag_hdr);
 486
 487         skb_shinfo(head)->frag_list = head->next;
 488         skb_reset_transport_header(head);
 489         skb_push(head, head->data - skb_network_header(head));
 490
 491         for (fp=head->next; fp; fp = fp->next) {
 492                 head->data_len += fp->len;
 493                 head->len += fp->len;
 494                 if (head->ip_summed != fp->ip_summed)
 495                         head->ip_summed = CHECKSUM_NONE;
 496                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 497                         head->csum = csum_add(head->csum, fp->csum);
 498                 head->truesize += fp->truesize;
 499         }
 500         atomic_sub(head->truesize, &fq->q.net->mem);
 501
 502         head->next = NULL;
 503         head->dev = dev;
 504         head->tstamp = fq->q.stamp;
 505         ipv6_hdr(head)->payload_len = htons(payload_len);
 506         IP6CB(head)->nhoff = nhoff;
 507
 508         /* Yes, and fold redundant checksum back. 8) */
 509         if (head->ip_summed == CHECKSUM_COMPLETE)
 510                 head->csum = csum_partial(skb_network_header(head),
 511                                           skb_network_header_len(head),
 512                                           head->csum);
 513
 514         rcu_read_lock();
 515         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 516         rcu_read_unlock();
 517         fq->q.fragments = NULL;
 518         fq->q.fragments_tail = NULL;
 519         return 1;
 520
 521 out_oversize:
 522         net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
 523         goto out_fail;
 524 out_oom:
 525         net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
 526 out_fail:
 527         rcu_read_lock();
 528         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 529         rcu_read_unlock();
 530         return -1;
 531 }
 532
 533 static int ipv6_frag_rcv(struct sk_buff *skb)
 534 {
 535         struct frag_hdr *fhdr;
 536         struct frag_queue *fq;
 537         const struct ipv6hdr *hdr = ipv6_hdr(skb);
 538         struct net *net = dev_net(skb_dst(skb)->dev);
 539
 540         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
 541
 542         /* Jumbo payload inhibits frag. header */
 543         if (hdr->payload_len==0)
 544                 goto fail_hdr;
 545
 546         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 547                                  sizeof(struct frag_hdr))))
 548                 goto fail_hdr;
 549
 550         hdr = ipv6_hdr(skb);
 551         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 552
 553         if (!(fhdr->frag_off & htons(0xFFF9))) {
 554                 /* It is not a fragmented frame */
 555                 skb->transport_header += sizeof(struct frag_hdr);
 556                 IP6_INC_STATS_BH(net,
 557                                  ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
 558
 559                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 560                 return 1;
 561         }
 562
 563         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
 564                 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
 565
 566         fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
 567         if (fq != NULL) {
 568                 int ret;
 569
 570                 spin_lock(&fq->q.lock);
 571
 572                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 573
 574                 spin_unlock(&fq->q.lock);
 575                 fq_put(fq);
 576                 return ret;
 577         }
 578
 579         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
 580         kfree_skb(skb);
 581         return -1;
 582
 583 fail_hdr:
 584         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
 585         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
 586         return -1;
 587 }
 588
 589 static const struct inet6_protocol frag_protocol =
 590 {
 591         .handler        =       ipv6_frag_rcv,
 592         .flags          =       INET6_PROTO_NOPOLICY,
 593 };
 594
 595 #ifdef CONFIG_SYSCTL
 596 static struct ctl_table ip6_frags_ns_ctl_table[] = {
 597         {
 598                 .procname       = "ip6frag_high_thresh",
 599                 .data           = &init_net.ipv6.frags.high_thresh,
 600                 .maxlen         = sizeof(int),
 601                 .mode           = 0644,
 602                 .proc_handler   = proc_dointvec
 603         },
 604         {
 605                 .procname       = "ip6frag_low_thresh",
 606                 .data           = &init_net.ipv6.frags.low_thresh,
 607                 .maxlen         = sizeof(int),
 608                 .mode           = 0644,
 609                 .proc_handler   = proc_dointvec
 610         },
 611         {
 612                 .procname       = "ip6frag_time",
 613                 .data           = &init_net.ipv6.frags.timeout,
 614                 .maxlen         = sizeof(int),
 615                 .mode           = 0644,
 616                 .proc_handler   = proc_dointvec_jiffies,
 617         },
 618         { }
 619 };
 620
 621 static struct ctl_table ip6_frags_ctl_table[] = {
 622         {
 623                 .procname       = "ip6frag_secret_interval",
 624                 .data           = &ip6_frags.secret_interval,
 625                 .maxlen         = sizeof(int),
 626                 .mode           = 0644,
 627                 .proc_handler   = proc_dointvec_jiffies,
 628         },
 629         { }
 630 };
 631
 632 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
 633 {
 634         struct ctl_table *table;
 635         struct ctl_table_header *hdr;
 636
 637         table = ip6_frags_ns_ctl_table;
 638         if (!net_eq(net, &init_net)) {
 639                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
 640                 if (table == NULL)
 641                         goto err_alloc;
 642
 643                 table[0].data = &net->ipv6.frags.high_thresh;
 644                 table[1].data = &net->ipv6.frags.low_thresh;
 645                 table[2].data = &net->ipv6.frags.timeout;
 646         }
 647
 648         hdr = register_net_sysctl(net, "net/ipv6", table);
 649         if (hdr == NULL)
 650                 goto err_reg;
 651
 652         net->ipv6.sysctl.frags_hdr = hdr;
 653         return 0;
 654
 655 err_reg:
 656         if (!net_eq(net, &init_net))
 657                 kfree(table);
 658 err_alloc:
 659         return -ENOMEM;
 660 }
 661
 662 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
 663 {
 664         struct ctl_table *table;
 665
 666         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
 667         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
 668         if (!net_eq(net, &init_net))
 669                 kfree(table);
 670 }
 671
 672 static struct ctl_table_header *ip6_ctl_header;
 673
 674 static int ip6_frags_sysctl_register(void)
 675 {
 676         ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
 677                         ip6_frags_ctl_table);
 678         return ip6_ctl_header == NULL ? -ENOMEM : 0;
 679 }
 680
 681 static void ip6_frags_sysctl_unregister(void)
 682 {
 683         unregister_net_sysctl_table(ip6_ctl_header);
 684 }
 685 #else
 686 static inline int ip6_frags_ns_sysctl_register(struct net *net)
 687 {
 688         return 0;
 689 }
 690
 691 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
 692 {
 693 }
 694
 695 static inline int ip6_frags_sysctl_register(void)
 696 {
 697         return 0;
 698 }
 699
 700 static inline void ip6_frags_sysctl_unregister(void)
 701 {
 702 }
 703 #endif
 704
 705 static int __net_init ipv6_frags_init_net(struct net *net)
 706 {
 707         net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
 708         net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
 709         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
 710
 711         inet_frags_init_net(&net->ipv6.frags);
 712
 713         return ip6_frags_ns_sysctl_register(net);
 714 }
 715
 716 static void __net_exit ipv6_frags_exit_net(struct net *net)
 717 {
 718         ip6_frags_ns_sysctl_unregister(net);
 719         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 720 }
 721
 722 static struct pernet_operations ip6_frags_ops = {
 723         .init = ipv6_frags_init_net,
 724         .exit = ipv6_frags_exit_net,
 725 };
 726
 727 int __init ipv6_frag_init(void)
 728 {
 729         int ret;
 730
 731         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 732         if (ret)
 733                 goto out;
 734
 735         ret = ip6_frags_sysctl_register();
 736         if (ret)
 737                 goto err_sysctl;
 738
 739         ret = register_pernet_subsys(&ip6_frags_ops);
 740         if (ret)
 741                 goto err_pernet;
 742
 743         ip6_frags.hashfn = ip6_hashfn;
 744         ip6_frags.constructor = ip6_frag_init;
 745         ip6_frags.destructor = NULL;
 746         ip6_frags.skb_free = NULL;
 747         ip6_frags.qsize = sizeof(struct frag_queue);
 748         ip6_frags.match = ip6_frag_match;
 749         ip6_frags.frag_expire = ip6_frag_expire;
 750         ip6_frags.secret_interval = 10 * 60 * HZ;
 751         inet_frags_init(&ip6_frags);
 752 out:
 753         return ret;
 754
 755 err_pernet:
 756         ip6_frags_sysctl_unregister();
 757 err_sysctl:
 758         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 759         goto out;
 760 }
 761
 762 void ipv6_frag_exit(void)
 763 {
 764         inet_frags_fini(&ip6_frags);
 765         ip6_frags_sysctl_unregister();
 766         unregister_pernet_subsys(&ip6_frags_ops);
 767         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 768 }