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
  30 #define pr_fmt(fmt) "IPv6: " fmt
  31
  32 #include <linux/errno.h>
  33 #include <linux/types.h>
  34 #include <linux/string.h>
  35 #include <linux/socket.h>
  36 #include <linux/sockios.h>
  37 #include <linux/jiffies.h>
  38 #include <linux/net.h>
  39 #include <linux/list.h>
  40 #include <linux/netdevice.h>
  41 #include <linux/in6.h>
  42 #include <linux/ipv6.h>
  43 #include <linux/icmpv6.h>
  44 #include <linux/random.h>
  45 #include <linux/jhash.h>
  46 #include <linux/skbuff.h>
  47 #include <linux/slab.h>
  48 #include <linux/export.h>
  49
  50 #include <net/sock.h>
  51 #include <net/snmp.h>
  52
  53 #include <net/ipv6.h>
  54 #include <net/ip6_route.h>
  55 #include <net/protocol.h>
  56 #include <net/transp_v6.h>
  57 #include <net/rawv6.h>
  58 #include <net/ndisc.h>
  59 #include <net/addrconf.h>
  60 #include <net/inet_frag.h>
  61
  62 struct ip6frag_skb_cb
  63 {
  64         struct inet6_skb_parm   h;
  65         int                     offset;
  66 };
  67
  68 #define FRAG6_CB(skb)   ((struct ip6frag_skb_cb*)((skb)->cb))
  69
  70
  71 static struct inet_frags ip6_frags;
  72
  73 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
  74                           struct net_device *dev);
  75
  76 /*
  77  * callers should be careful not to use the hash value outside the ipfrag_lock
  78  * as doing so could race with ipfrag_hash_rnd being recalculated.
  79  */
  80 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
  81                              const struct in6_addr *daddr, u32 rnd)
  82 {
  83         u32 c;
  84
  85         c = jhash_3words((__force u32)saddr->s6_addr32[0],
  86                          (__force u32)saddr->s6_addr32[1],
  87                          (__force u32)saddr->s6_addr32[2],
  88                          rnd);
  89
  90         c = jhash_3words((__force u32)saddr->s6_addr32[3],
  91                          (__force u32)daddr->s6_addr32[0],
  92                          (__force u32)daddr->s6_addr32[1],
  93                          c);
  94
  95         c =  jhash_3words((__force u32)daddr->s6_addr32[2],
  96                           (__force u32)daddr->s6_addr32[3],
  97                           (__force u32)id,
  98                           c);
  99
 100         return c & (INETFRAGS_HASHSZ - 1);
 101 }
 102 EXPORT_SYMBOL_GPL(inet6_hash_frag);
 103
 104 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 105 {
 106         struct frag_queue *fq;
 107
 108         fq = container_of(q, struct frag_queue, q);
 109         return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
 110 }
 111
 112 bool ip6_frag_match(struct inet_frag_queue *q, void *a)
 113 {
 114         struct frag_queue *fq;
 115         struct ip6_create_arg *arg = a;
 116
 117         fq = container_of(q, struct frag_queue, q);
 118         return  fq->id == arg->id &&
 119                 fq->user == arg->user &&
 120                 ipv6_addr_equal(&fq->saddr, arg->src) &&
 121                 ipv6_addr_equal(&fq->daddr, arg->dst);
 122 }
 123 EXPORT_SYMBOL(ip6_frag_match);
 124
 125 void ip6_frag_init(struct inet_frag_queue *q, void *a)
 126 {
 127         struct frag_queue *fq = container_of(q, struct frag_queue, q);
 128         struct ip6_create_arg *arg = a;
 129
 130         fq->id = arg->id;
 131         fq->user = arg->user;
 132         fq->saddr = *arg->src;
 133         fq->daddr = *arg->dst;
 134 }
 135 EXPORT_SYMBOL(ip6_frag_init);
 136
 137 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq,
 138                            struct inet_frags *frags)
 139 {
 140         struct net_device *dev = NULL;
 141
 142         spin_lock(&fq->q.lock);
 143
 144         if (fq->q.last_in & INET_FRAG_COMPLETE)
 145                 goto out;
 146
 147         inet_frag_kill(&fq->q, frags);
 148
 149         rcu_read_lock();
 150         dev = dev_get_by_index_rcu(net, fq->iif);
 151         if (!dev)
 152                 goto out_rcu_unlock;
 153
 154         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 155         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 156
 157         /* Don't send error if the first segment did not arrive. */
 158         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
 159                 goto out_rcu_unlock;
 160
 161         /*
 162            But use as source device on which LAST ARRIVED
 163            segment was received. And do not use fq->dev
 164            pointer directly, device might already disappeared.
 165          */
 166         fq->q.fragments->dev = dev;
 167         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
 168 out_rcu_unlock:
 169         rcu_read_unlock();
 170 out:
 171         spin_unlock(&fq->q.lock);
 172         inet_frag_put(&fq->q, frags);
 173 }
 174 EXPORT_SYMBOL(ip6_expire_frag_queue);
 175
 176 static void ip6_frag_expire(unsigned long data)
 177 {
 178         struct frag_queue *fq;
 179         struct net *net;
 180
 181         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 182         net = container_of(fq->q.net, struct net, ipv6.frags);
 183
 184         ip6_expire_frag_queue(net, fq, &ip6_frags);
 185 }
 186
 187 static __inline__ struct frag_queue *
 188 fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
 189 {
 190         struct inet_frag_queue *q;
 191         struct ip6_create_arg arg;
 192         unsigned int hash;
 193
 194         arg.id = id;
 195         arg.user = IP6_DEFRAG_LOCAL_DELIVER;
 196         arg.src = src;
 197         arg.dst = dst;
 198
 199         read_lock(&ip6_frags.lock);
 200         hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
 201
 202         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
 203         if (IS_ERR_OR_NULL(q)) {
 204                 inet_frag_maybe_warn_overflow(q, pr_fmt());
 205                 return NULL;
 206         }
 207         return container_of(q, struct frag_queue, q);
 208 }
 209
 210 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 211                            struct frag_hdr *fhdr, int nhoff)
 212 {
 213         struct sk_buff *prev, *next;
 214         struct net_device *dev;
 215         int offset, end;
 216         struct net *net = dev_net(skb_dst(skb)->dev);
 217
 218         if (fq->q.last_in & INET_FRAG_COMPLETE)
 219                 goto err;
 220
 221         offset = ntohs(fhdr->frag_off) & ~0x7;
 222         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 223                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 224
 225         if ((unsigned int)end > IPV6_MAXPLEN) {
 226                 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 227                                  IPSTATS_MIB_INHDRERRORS);
 228                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 229                                   ((u8 *)&fhdr->frag_off -
 230                                    skb_network_header(skb)));
 231                 return -1;
 232         }
 233
 234         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 235                 const unsigned char *nh = skb_network_header(skb);
 236                 skb->csum = csum_sub(skb->csum,
 237                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 238                                                   0));
 239         }
 240
 241         /* Is this the final fragment? */
 242         if (!(fhdr->frag_off & htons(IP6_MF))) {
 243                 /* If we already have some bits beyond end
 244                  * or have different end, the segment is corrupted.
 245                  */
 246                 if (end < fq->q.len ||
 247                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
 248                         goto err;
 249                 fq->q.last_in |= INET_FRAG_LAST_IN;
 250                 fq->q.len = end;
 251         } else {
 252                 /* Check if the fragment is rounded to 8 bytes.
 253                  * Required by the RFC.
 254                  */
 255                 if (end & 0x7) {
 256                         /* RFC2460 says always send parameter problem in
 257                          * this case. -DaveM
 258                          */
 259                         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 260                                          IPSTATS_MIB_INHDRERRORS);
 261                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 262                                           offsetof(struct ipv6hdr, payload_len));
 263                         return -1;
 264                 }
 265                 if (end > fq->q.len) {
 266                         /* Some bits beyond end -> corruption. */
 267                         if (fq->q.last_in & INET_FRAG_LAST_IN)
 268                                 goto err;
 269                         fq->q.len = end;
 270                 }
 271         }
 272
 273         if (end == offset)
 274                 goto err;
 275
 276         /* Point into the IP datagram 'data' part. */
 277         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 278                 goto err;
 279
 280         if (pskb_trim_rcsum(skb, end - offset))
 281                 goto err;
 282
 283         /* Find out which fragments are in front and at the back of us
 284          * in the chain of fragments so far.  We must know where to put
 285          * this fragment, right?
 286          */
 287         prev = fq->q.fragments_tail;
 288         if (!prev || FRAG6_CB(prev)->offset < offset) {
 289                 next = NULL;
 290                 goto found;
 291         }
 292         prev = NULL;
 293         for(next = fq->q.fragments; next != NULL; next = next->next) {
 294                 if (FRAG6_CB(next)->offset >= offset)
 295                         break;  /* bingo! */
 296                 prev = next;
 297         }
 298
 299 found:
 300         /* RFC5722, Section 4, amended by Errata ID : 3089
 301          *                          When reassembling an IPv6 datagram, if
 302          *   one or more its constituent fragments is determined to be an
 303          *   overlapping fragment, the entire datagram (and any constituent
 304          *   fragments) MUST be silently discarded.
 305          */
 306
 307         /* Check for overlap with preceding fragment. */
 308         if (prev &&
 309             (FRAG6_CB(prev)->offset + prev->len) > offset)
 310                 goto discard_fq;
 311
 312         /* Look for overlap with succeeding segment. */
 313         if (next && FRAG6_CB(next)->offset < end)
 314                 goto discard_fq;
 315
 316         FRAG6_CB(skb)->offset = offset;
 317
 318         /* Insert this fragment in the chain of fragments. */
 319         skb->next = next;
 320         if (!next)
 321                 fq->q.fragments_tail = skb;
 322         if (prev)
 323                 prev->next = skb;
 324         else
 325                 fq->q.fragments = skb;
 326
 327         dev = skb->dev;
 328         if (dev) {
 329                 fq->iif = dev->ifindex;
 330                 skb->dev = NULL;
 331         }
 332         fq->q.stamp = skb->tstamp;
 333         fq->q.meat += skb->len;
 334         atomic_add(skb->truesize, &fq->q.net->mem);
 335
 336         /* The first fragment.
 337          * nhoffset is obtained from the first fragment, of course.
 338          */
 339         if (offset == 0) {
 340                 fq->nhoffset = nhoff;
 341                 fq->q.last_in |= INET_FRAG_FIRST_IN;
 342         }
 343
 344         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 345             fq->q.meat == fq->q.len)
 346                 return ip6_frag_reasm(fq, prev, dev);
 347
 348         write_lock(&ip6_frags.lock);
 349         list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
 350         write_unlock(&ip6_frags.lock);
 351         return -1;
 352
 353 discard_fq:
 354         inet_frag_kill(&fq->q, &ip6_frags);
 355 err:
 356         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 357                       IPSTATS_MIB_REASMFAILS);
 358         kfree_skb(skb);
 359         return -1;
 360 }
 361
 362 /*
 363  *      Check if this packet is complete.
 364  *      Returns NULL on failure by any reason, and pointer
 365  *      to current nexthdr field in reassembled frame.
 366  *
 367  *      It is called with locked fq, and caller must check that
 368  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 369  *      the last and the first frames arrived and all the bits are here.
 370  */
 371 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 372                           struct net_device *dev)
 373 {
 374         struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
 375         struct sk_buff *fp, *head = fq->q.fragments;
 376         int    payload_len;
 377         unsigned int nhoff;
 378         int sum_truesize;
 379
 380         inet_frag_kill(&fq->q, &ip6_frags);
 381
 382         /* Make the one we just received the head. */
 383         if (prev) {
 384                 head = prev->next;
 385                 fp = skb_clone(head, GFP_ATOMIC);
 386
 387                 if (!fp)
 388                         goto out_oom;
 389
 390                 fp->next = head->next;
 391                 if (!fp->next)
 392                         fq->q.fragments_tail = fp;
 393                 prev->next = fp;
 394
 395                 skb_morph(head, fq->q.fragments);
 396                 head->next = fq->q.fragments->next;
 397
 398                 consume_skb(fq->q.fragments);
 399                 fq->q.fragments = head;
 400         }
 401
 402         WARN_ON(head == NULL);
 403         WARN_ON(FRAG6_CB(head)->offset != 0);
 404
 405         /* Unfragmented part is taken from the first segment. */
 406         payload_len = ((head->data - skb_network_header(head)) -
 407                        sizeof(struct ipv6hdr) + fq->q.len -
 408                        sizeof(struct frag_hdr));
 409         if (payload_len > IPV6_MAXPLEN)
 410                 goto out_oversize;
 411
 412         /* Head of list must not be cloned. */
 413         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 414                 goto out_oom;
 415
 416         /* If the first fragment is fragmented itself, we split
 417          * it to two chunks: the first with data and paged part
 418          * and the second, holding only fragments. */
 419         if (skb_has_frag_list(head)) {
 420                 struct sk_buff *clone;
 421                 int i, plen = 0;
 422
 423                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 424                         goto out_oom;
 425                 clone->next = head->next;
 426                 head->next = clone;
 427                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 428                 skb_frag_list_init(head);
 429                 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
 430                         plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
 431                 clone->len = clone->data_len = head->data_len - plen;
 432                 head->data_len -= clone->len;
 433                 head->len -= clone->len;
 434                 clone->csum = 0;
 435                 clone->ip_summed = head->ip_summed;
 436                 atomic_add(clone->truesize, &fq->q.net->mem);
 437         }
 438
 439         /* We have to remove fragment header from datagram and to relocate
 440          * header in order to calculate ICV correctly. */
 441         nhoff = fq->nhoffset;
 442         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 443         memmove(head->head + sizeof(struct frag_hdr), head->head,
 444                 (head->data - head->head) - sizeof(struct frag_hdr));
 445         head->mac_header += sizeof(struct frag_hdr);
 446         head->network_header += sizeof(struct frag_hdr);
 447
 448         skb_reset_transport_header(head);
 449         skb_push(head, head->data - skb_network_header(head));
 450
 451         sum_truesize = head->truesize;
 452         for (fp = head->next; fp;) {
 453                 bool headstolen;
 454                 int delta;
 455                 struct sk_buff *next = fp->next;
 456
 457                 sum_truesize += fp->truesize;
 458                 if (head->ip_summed != fp->ip_summed)
 459                         head->ip_summed = CHECKSUM_NONE;
 460                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 461                         head->csum = csum_add(head->csum, fp->csum);
 462
 463                 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
 464                         kfree_skb_partial(fp, headstolen);
 465                 } else {
 466                         if (!skb_shinfo(head)->frag_list)
 467                                 skb_shinfo(head)->frag_list = fp;
 468                         head->data_len += fp->len;
 469                         head->len += fp->len;
 470                         head->truesize += fp->truesize;
 471                 }
 472                 fp = next;
 473         }
 474         atomic_sub(sum_truesize, &fq->q.net->mem);
 475
 476         head->next = NULL;
 477         head->dev = dev;
 478         head->tstamp = fq->q.stamp;
 479         ipv6_hdr(head)->payload_len = htons(payload_len);
 480         IP6CB(head)->nhoff = nhoff;
 481
 482         /* Yes, and fold redundant checksum back. 8) */
 483         if (head->ip_summed == CHECKSUM_COMPLETE)
 484                 head->csum = csum_partial(skb_network_header(head),
 485                                           skb_network_header_len(head),
 486                                           head->csum);
 487
 488         rcu_read_lock();
 489         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 490         rcu_read_unlock();
 491         fq->q.fragments = NULL;
 492         fq->q.fragments_tail = NULL;
 493         return 1;
 494
 495 out_oversize:
 496         net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
 497         goto out_fail;
 498 out_oom:
 499         net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
 500 out_fail:
 501         rcu_read_lock();
 502         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 503         rcu_read_unlock();
 504         return -1;
 505 }
 506
 507 static int ipv6_frag_rcv(struct sk_buff *skb)
 508 {
 509         struct frag_hdr *fhdr;
 510         struct frag_queue *fq;
 511         const struct ipv6hdr *hdr = ipv6_hdr(skb);
 512         struct net *net = dev_net(skb_dst(skb)->dev);
 513         int evicted;
 514
 515         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
 516
 517         /* Jumbo payload inhibits frag. header */
 518         if (hdr->payload_len==0)
 519                 goto fail_hdr;
 520
 521         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 522                                  sizeof(struct frag_hdr))))
 523                 goto fail_hdr;
 524
 525         hdr = ipv6_hdr(skb);
 526         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 527
 528         if (!(fhdr->frag_off & htons(0xFFF9))) {
 529                 /* It is not a fragmented frame */
 530                 skb->transport_header += sizeof(struct frag_hdr);
 531                 IP6_INC_STATS_BH(net,
 532                                  ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
 533
 534                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 535                 return 1;
 536         }
 537
 538         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags, false);
 539         if (evicted)
 540                 IP6_ADD_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 541                                  IPSTATS_MIB_REASMFAILS, evicted);
 542
 543         fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
 544         if (fq != NULL) {
 545                 int ret;
 546
 547                 spin_lock(&fq->q.lock);
 548
 549                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 550
 551                 spin_unlock(&fq->q.lock);
 552                 inet_frag_put(&fq->q, &ip6_frags);
 553                 return ret;
 554         }
 555
 556         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
 557         kfree_skb(skb);
 558         return -1;
 559
 560 fail_hdr:
 561         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
 562         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
 563         return -1;
 564 }
 565
 566 static const struct inet6_protocol frag_protocol =
 567 {
 568         .handler        =       ipv6_frag_rcv,
 569         .flags          =       INET6_PROTO_NOPOLICY,
 570 };
 571
 572 #ifdef CONFIG_SYSCTL
 573 static struct ctl_table ip6_frags_ns_ctl_table[] = {
 574         {
 575                 .procname       = "ip6frag_high_thresh",
 576                 .data           = &init_net.ipv6.frags.high_thresh,
 577                 .maxlen         = sizeof(int),
 578                 .mode           = 0644,
 579                 .proc_handler   = proc_dointvec
 580         },
 581         {
 582                 .procname       = "ip6frag_low_thresh",
 583                 .data           = &init_net.ipv6.frags.low_thresh,
 584                 .maxlen         = sizeof(int),
 585                 .mode           = 0644,
 586                 .proc_handler   = proc_dointvec
 587         },
 588         {
 589                 .procname       = "ip6frag_time",
 590                 .data           = &init_net.ipv6.frags.timeout,
 591                 .maxlen         = sizeof(int),
 592                 .mode           = 0644,
 593                 .proc_handler   = proc_dointvec_jiffies,
 594         },
 595         { }
 596 };
 597
 598 static struct ctl_table ip6_frags_ctl_table[] = {
 599         {
 600                 .procname       = "ip6frag_secret_interval",
 601                 .data           = &ip6_frags.secret_interval,
 602                 .maxlen         = sizeof(int),
 603                 .mode           = 0644,
 604                 .proc_handler   = proc_dointvec_jiffies,
 605         },
 606         { }
 607 };
 608
 609 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
 610 {
 611         struct ctl_table *table;
 612         struct ctl_table_header *hdr;
 613
 614         table = ip6_frags_ns_ctl_table;
 615         if (!net_eq(net, &init_net)) {
 616                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
 617                 if (table == NULL)
 618                         goto err_alloc;
 619
 620                 table[0].data = &net->ipv6.frags.high_thresh;
 621                 table[1].data = &net->ipv6.frags.low_thresh;
 622                 table[2].data = &net->ipv6.frags.timeout;
 623
 624                 /* Don't export sysctls to unprivileged users */
 625                 if (net->user_ns != &init_user_ns)
 626                         table[0].procname = NULL;
 627         }
 628
 629         hdr = register_net_sysctl(net, "net/ipv6", table);
 630         if (hdr == NULL)
 631                 goto err_reg;
 632
 633         net->ipv6.sysctl.frags_hdr = hdr;
 634         return 0;
 635
 636 err_reg:
 637         if (!net_eq(net, &init_net))
 638                 kfree(table);
 639 err_alloc:
 640         return -ENOMEM;
 641 }
 642
 643 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
 644 {
 645         struct ctl_table *table;
 646
 647         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
 648         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
 649         if (!net_eq(net, &init_net))
 650                 kfree(table);
 651 }
 652
 653 static struct ctl_table_header *ip6_ctl_header;
 654
 655 static int ip6_frags_sysctl_register(void)
 656 {
 657         ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
 658                         ip6_frags_ctl_table);
 659         return ip6_ctl_header == NULL ? -ENOMEM : 0;
 660 }
 661
 662 static void ip6_frags_sysctl_unregister(void)
 663 {
 664         unregister_net_sysctl_table(ip6_ctl_header);
 665 }
 666 #else
 667 static inline int ip6_frags_ns_sysctl_register(struct net *net)
 668 {
 669         return 0;
 670 }
 671
 672 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
 673 {
 674 }
 675
 676 static inline int ip6_frags_sysctl_register(void)
 677 {
 678         return 0;
 679 }
 680
 681 static inline void ip6_frags_sysctl_unregister(void)
 682 {
 683 }
 684 #endif
 685
 686 static int __net_init ipv6_frags_init_net(struct net *net)
 687 {
 688         net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
 689         net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
 690         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
 691
 692         inet_frags_init_net(&net->ipv6.frags);
 693
 694         return ip6_frags_ns_sysctl_register(net);
 695 }
 696
 697 static void __net_exit ipv6_frags_exit_net(struct net *net)
 698 {
 699         ip6_frags_ns_sysctl_unregister(net);
 700         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 701 }
 702
 703 static struct pernet_operations ip6_frags_ops = {
 704         .init = ipv6_frags_init_net,
 705         .exit = ipv6_frags_exit_net,
 706 };
 707
 708 int __init ipv6_frag_init(void)
 709 {
 710         int ret;
 711
 712         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 713         if (ret)
 714                 goto out;
 715
 716         ret = ip6_frags_sysctl_register();
 717         if (ret)
 718                 goto err_sysctl;
 719
 720         ret = register_pernet_subsys(&ip6_frags_ops);
 721         if (ret)
 722                 goto err_pernet;
 723
 724         ip6_frags.hashfn = ip6_hashfn;
 725         ip6_frags.constructor = ip6_frag_init;
 726         ip6_frags.destructor = NULL;
 727         ip6_frags.skb_free = NULL;
 728         ip6_frags.qsize = sizeof(struct frag_queue);
 729         ip6_frags.match = ip6_frag_match;
 730         ip6_frags.frag_expire = ip6_frag_expire;
 731         ip6_frags.secret_interval = 10 * 60 * HZ;
 732         inet_frags_init(&ip6_frags);
 733 out:
 734         return ret;
 735
 736 err_pernet:
 737         ip6_frags_sysctl_unregister();
 738 err_sysctl:
 739         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 740         goto out;
 741 }
 742
 743 void ipv6_frag_exit(void)
 744 {
 745         inet_frags_fini(&ip6_frags);
 746         ip6_frags_sysctl_unregister();
 747         unregister_pernet_subsys(&ip6_frags_ops);
 748         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 749 }