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 /*
  72  *      Equivalent of ipv4 struct ipq
  73  */
  74
  75 struct frag_queue
  76 {
  77         struct inet_frag_queue  q;
  78
  79         __be32                  id;             /* fragment id          */
  80         u32                     user;
  81         struct in6_addr         saddr;
  82         struct in6_addr         daddr;
  83
  84         int                     iif;
  85         unsigned int            csum;
  86         __u16                   nhoffset;
  87 };
  88
  89 static struct inet_frags ip6_frags;
  90
  91 int ip6_frag_nqueues(struct net *net)
  92 {
  93         return net->ipv6.frags.nqueues;
  94 }
  95
  96 int ip6_frag_mem(struct net *net)
  97 {
  98         return atomic_read(&net->ipv6.frags.mem);
  99 }
 100
 101 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 102                           struct net_device *dev);
 103
 104 /*
 105  * callers should be careful not to use the hash value outside the ipfrag_lock
 106  * as doing so could race with ipfrag_hash_rnd being recalculated.
 107  */
 108 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
 109                              const struct in6_addr *daddr, u32 rnd)
 110 {
 111         u32 c;
 112
 113         c = jhash_3words((__force u32)saddr->s6_addr32[0],
 114                          (__force u32)saddr->s6_addr32[1],
 115                          (__force u32)saddr->s6_addr32[2],
 116                          rnd);
 117
 118         c = jhash_3words((__force u32)saddr->s6_addr32[3],
 119                          (__force u32)daddr->s6_addr32[0],
 120                          (__force u32)daddr->s6_addr32[1],
 121                          c);
 122
 123         c =  jhash_3words((__force u32)daddr->s6_addr32[2],
 124                           (__force u32)daddr->s6_addr32[3],
 125                           (__force u32)id,
 126                           c);
 127
 128         return c & (INETFRAGS_HASHSZ - 1);
 129 }
 130 EXPORT_SYMBOL_GPL(inet6_hash_frag);
 131
 132 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 133 {
 134         struct frag_queue *fq;
 135
 136         fq = container_of(q, struct frag_queue, q);
 137         return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
 138 }
 139
 140 int ip6_frag_match(struct inet_frag_queue *q, void *a)
 141 {
 142         struct frag_queue *fq;
 143         struct ip6_create_arg *arg = a;
 144
 145         fq = container_of(q, struct frag_queue, q);
 146         return (fq->id == arg->id && fq->user == arg->user &&
 147                 ipv6_addr_equal(&fq->saddr, arg->src) &&
 148                 ipv6_addr_equal(&fq->daddr, arg->dst) &&
 149                 (arg->iif == fq->iif ||
 150                  !(ipv6_addr_type(arg->dst) & (IPV6_ADDR_MULTICAST |
 151                                                IPV6_ADDR_LINKLOCAL))));
 152 }
 153 EXPORT_SYMBOL(ip6_frag_match);
 154
 155 void ip6_frag_init(struct inet_frag_queue *q, void *a)
 156 {
 157         struct frag_queue *fq = container_of(q, struct frag_queue, q);
 158         struct ip6_create_arg *arg = a;
 159
 160         fq->id = arg->id;
 161         fq->user = arg->user;
 162         ipv6_addr_copy(&fq->saddr, arg->src);
 163         ipv6_addr_copy(&fq->daddr, arg->dst);
 164 }
 165 EXPORT_SYMBOL(ip6_frag_init);
 166
 167 /* Destruction primitives. */
 168
 169 static __inline__ void fq_put(struct frag_queue *fq)
 170 {
 171         inet_frag_put(&fq->q, &ip6_frags);
 172 }
 173
 174 /* Kill fq entry. It is not destroyed immediately,
 175  * because caller (and someone more) holds reference count.
 176  */
 177 static __inline__ void fq_kill(struct frag_queue *fq)
 178 {
 179         inet_frag_kill(&fq->q, &ip6_frags);
 180 }
 181
 182 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 183 {
 184         int evicted;
 185
 186         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
 187         if (evicted)
 188                 IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
 189 }
 190
 191 static void ip6_frag_expire(unsigned long data)
 192 {
 193         struct frag_queue *fq;
 194         struct net_device *dev = NULL;
 195         struct net *net;
 196
 197         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 198
 199         spin_lock(&fq->q.lock);
 200
 201         if (fq->q.last_in & INET_FRAG_COMPLETE)
 202                 goto out;
 203
 204         fq_kill(fq);
 205
 206         net = container_of(fq->q.net, struct net, ipv6.frags);
 207         rcu_read_lock();
 208         dev = dev_get_by_index_rcu(net, fq->iif);
 209         if (!dev)
 210                 goto out_rcu_unlock;
 211
 212         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 213         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 214
 215         /* Don't send error if the first segment did not arrive. */
 216         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
 217                 goto out_rcu_unlock;
 218
 219         /*
 220            But use as source device on which LAST ARRIVED
 221            segment was received. And do not use fq->dev
 222            pointer directly, device might already disappeared.
 223          */
 224         fq->q.fragments->dev = dev;
 225         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
 226 out_rcu_unlock:
 227         rcu_read_unlock();
 228 out:
 229         spin_unlock(&fq->q.lock);
 230         fq_put(fq);
 231 }
 232
 233 static __inline__ struct frag_queue *
 234 fq_find(struct net *net, __be32 id, const struct in6_addr *src,
 235         const struct in6_addr *dst, int iif)
 236 {
 237         struct inet_frag_queue *q;
 238         struct ip6_create_arg arg;
 239         unsigned int hash;
 240
 241         arg.id = id;
 242         arg.user = IP6_DEFRAG_LOCAL_DELIVER;
 243         arg.src = src;
 244         arg.dst = dst;
 245         arg.iif = iif;
 246
 247         read_lock(&ip6_frags.lock);
 248         hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
 249
 250         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
 251         if (IS_ERR_OR_NULL(q)) {
 252                 inet_frag_maybe_warn_overflow(q, pr_fmt());
 253                 return NULL;
 254         }
 255         return container_of(q, struct frag_queue, q);
 256 }
 257
 258 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 259                            struct frag_hdr *fhdr, int nhoff)
 260 {
 261         struct sk_buff *prev, *next;
 262         struct net_device *dev;
 263         int offset, end;
 264         struct net *net = dev_net(skb_dst(skb)->dev);
 265
 266         if (fq->q.last_in & INET_FRAG_COMPLETE)
 267                 goto err;
 268
 269         offset = ntohs(fhdr->frag_off) & ~0x7;
 270         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 271                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 272
 273         if ((unsigned int)end > IPV6_MAXPLEN) {
 274                 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 275                                  IPSTATS_MIB_INHDRERRORS);
 276                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 277                                   ((u8 *)&fhdr->frag_off -
 278                                    skb_network_header(skb)));
 279                 return -1;
 280         }
 281
 282         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 283                 const unsigned char *nh = skb_network_header(skb);
 284                 skb->csum = csum_sub(skb->csum,
 285                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 286                                                   0));
 287         }
 288
 289         /* Is this the final fragment? */
 290         if (!(fhdr->frag_off & htons(IP6_MF))) {
 291                 /* If we already have some bits beyond end
 292                  * or have different end, the segment is corrupted.
 293                  */
 294                 if (end < fq->q.len ||
 295                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
 296                         goto err;
 297                 fq->q.last_in |= INET_FRAG_LAST_IN;
 298                 fq->q.len = end;
 299         } else {
 300                 /* Check if the fragment is rounded to 8 bytes.
 301                  * Required by the RFC.
 302                  */
 303                 if (end & 0x7) {
 304                         /* RFC2460 says always send parameter problem in
 305                          * this case. -DaveM
 306                          */
 307                         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 308                                          IPSTATS_MIB_INHDRERRORS);
 309                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 310                                           offsetof(struct ipv6hdr, payload_len));
 311                         return -1;
 312                 }
 313                 if (end > fq->q.len) {
 314                         /* Some bits beyond end -> corruption. */
 315                         if (fq->q.last_in & INET_FRAG_LAST_IN)
 316                                 goto err;
 317                         fq->q.len = end;
 318                 }
 319         }
 320
 321         if (end == offset)
 322                 goto err;
 323
 324         /* Point into the IP datagram 'data' part. */
 325         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 326                 goto err;
 327
 328         if (pskb_trim_rcsum(skb, end - offset))
 329                 goto err;
 330
 331         /* Find out which fragments are in front and at the back of us
 332          * in the chain of fragments so far.  We must know where to put
 333          * this fragment, right?
 334          */
 335         prev = fq->q.fragments_tail;
 336         if (!prev || FRAG6_CB(prev)->offset < offset) {
 337                 next = NULL;
 338                 goto found;
 339         }
 340         prev = NULL;
 341         for(next = fq->q.fragments; next != NULL; next = next->next) {
 342                 if (FRAG6_CB(next)->offset >= offset)
 343                         break;  /* bingo! */
 344                 prev = next;
 345         }
 346
 347 found:
 348         /* RFC5722, Section 4:
 349          *                                  When reassembling an IPv6 datagram, if
 350          *   one or more its constituent fragments is determined to be an
 351          *   overlapping fragment, the entire datagram (and any constituent
 352          *   fragments, including those not yet received) MUST be silently
 353          *   discarded.
 354          */
 355
 356         /* Check for overlap with preceding fragment. */
 357         if (prev &&
 358             (FRAG6_CB(prev)->offset + prev->len) > offset)
 359                 goto discard_fq;
 360
 361         /* Look for overlap with succeeding segment. */
 362         if (next && FRAG6_CB(next)->offset < end)
 363                 goto discard_fq;
 364
 365         FRAG6_CB(skb)->offset = offset;
 366
 367         /* Insert this fragment in the chain of fragments. */
 368         skb->next = next;
 369         if (!next)
 370                 fq->q.fragments_tail = skb;
 371         if (prev)
 372                 prev->next = skb;
 373         else
 374                 fq->q.fragments = skb;
 375
 376         dev = skb->dev;
 377         if (dev) {
 378                 fq->iif = dev->ifindex;
 379                 skb->dev = NULL;
 380         }
 381         fq->q.stamp = skb->tstamp;
 382         fq->q.meat += skb->len;
 383         atomic_add(skb->truesize, &fq->q.net->mem);
 384
 385         /* The first fragment.
 386          * nhoffset is obtained from the first fragment, of course.
 387          */
 388         if (offset == 0) {
 389                 fq->nhoffset = nhoff;
 390                 fq->q.last_in |= INET_FRAG_FIRST_IN;
 391         }
 392
 393         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 394             fq->q.meat == fq->q.len) {
 395                 int res;
 396                 unsigned long orefdst = skb->_skb_refdst;
 397
 398                 skb->_skb_refdst = 0UL;
 399                 res = ip6_frag_reasm(fq, prev, dev);
 400                 skb->_skb_refdst = orefdst;
 401                 return res;
 402         }
 403
 404         skb_dst_drop(skb);
 405
 406         write_lock(&ip6_frags.lock);
 407         list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
 408         write_unlock(&ip6_frags.lock);
 409         return -1;
 410
 411 discard_fq:
 412         fq_kill(fq);
 413 err:
 414         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
 415                       IPSTATS_MIB_REASMFAILS);
 416         kfree_skb(skb);
 417         return -1;
 418 }
 419
 420 /*
 421  *      Check if this packet is complete.
 422  *      Returns NULL on failure by any reason, and pointer
 423  *      to current nexthdr field in reassembled frame.
 424  *
 425  *      It is called with locked fq, and caller must check that
 426  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 427  *      the last and the first frames arrived and all the bits are here.
 428  */
 429 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 430                           struct net_device *dev)
 431 {
 432         struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
 433         struct sk_buff *fp, *head = fq->q.fragments;
 434         int    payload_len;
 435         unsigned int nhoff;
 436
 437         fq_kill(fq);
 438
 439         /* Make the one we just received the head. */
 440         if (prev) {
 441                 head = prev->next;
 442                 fp = skb_clone(head, GFP_ATOMIC);
 443
 444                 if (!fp)
 445                         goto out_oom;
 446
 447                 fp->next = head->next;
 448                 if (!fp->next)
 449                         fq->q.fragments_tail = fp;
 450                 prev->next = fp;
 451
 452                 skb_morph(head, fq->q.fragments);
 453                 head->next = fq->q.fragments->next;
 454
 455                 kfree_skb(fq->q.fragments);
 456                 fq->q.fragments = head;
 457         }
 458
 459         WARN_ON(head == NULL);
 460         WARN_ON(FRAG6_CB(head)->offset != 0);
 461
 462         /* Unfragmented part is taken from the first segment. */
 463         payload_len = ((head->data - skb_network_header(head)) -
 464                        sizeof(struct ipv6hdr) + fq->q.len -
 465                        sizeof(struct frag_hdr));
 466         if (payload_len > IPV6_MAXPLEN)
 467                 goto out_oversize;
 468
 469         /* Head of list must not be cloned. */
 470         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 471                 goto out_oom;
 472
 473         /* If the first fragment is fragmented itself, we split
 474          * it to two chunks: the first with data and paged part
 475          * and the second, holding only fragments. */
 476         if (skb_has_frag_list(head)) {
 477                 struct sk_buff *clone;
 478                 int i, plen = 0;
 479
 480                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 481                         goto out_oom;
 482                 clone->next = head->next;
 483                 head->next = clone;
 484                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 485                 skb_frag_list_init(head);
 486                 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
 487                         plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
 488                 clone->len = clone->data_len = head->data_len - plen;
 489                 head->data_len -= clone->len;
 490                 head->len -= clone->len;
 491                 clone->csum = 0;
 492                 clone->ip_summed = head->ip_summed;
 493                 atomic_add(clone->truesize, &fq->q.net->mem);
 494         }
 495
 496         /* We have to remove fragment header from datagram and to relocate
 497          * header in order to calculate ICV correctly. */
 498         nhoff = fq->nhoffset;
 499         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 500         memmove(head->head + sizeof(struct frag_hdr), head->head,
 501                 (head->data - head->head) - sizeof(struct frag_hdr));
 502         head->mac_header += sizeof(struct frag_hdr);
 503         head->network_header += sizeof(struct frag_hdr);
 504
 505         skb_shinfo(head)->frag_list = head->next;
 506         skb_reset_transport_header(head);
 507         skb_push(head, head->data - skb_network_header(head));
 508
 509         for (fp=head->next; fp; fp = fp->next) {
 510                 head->data_len += fp->len;
 511                 head->len += fp->len;
 512                 if (head->ip_summed != fp->ip_summed)
 513                         head->ip_summed = CHECKSUM_NONE;
 514                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 515                         head->csum = csum_add(head->csum, fp->csum);
 516                 head->truesize += fp->truesize;
 517         }
 518         atomic_sub(head->truesize, &fq->q.net->mem);
 519
 520         head->next = NULL;
 521         head->dev = dev;
 522         head->tstamp = fq->q.stamp;
 523         ipv6_hdr(head)->payload_len = htons(payload_len);
 524         IP6CB(head)->nhoff = nhoff;
 525         IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
 526
 527         /* Yes, and fold redundant checksum back. 8) */
 528         if (head->ip_summed == CHECKSUM_COMPLETE)
 529                 head->csum = csum_partial(skb_network_header(head),
 530                                           skb_network_header_len(head),
 531                                           head->csum);
 532
 533         rcu_read_lock();
 534         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 535         rcu_read_unlock();
 536         fq->q.fragments = NULL;
 537         fq->q.fragments_tail = NULL;
 538         return 1;
 539
 540 out_oversize:
 541         if (net_ratelimit())
 542                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
 543         goto out_fail;
 544 out_oom:
 545         if (net_ratelimit())
 546                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
 547 out_fail:
 548         rcu_read_lock();
 549         IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 550         rcu_read_unlock();
 551         return -1;
 552 }
 553
 554 static int ipv6_frag_rcv(struct sk_buff *skb)
 555 {
 556         struct frag_hdr *fhdr;
 557         struct frag_queue *fq;
 558         const struct ipv6hdr *hdr = ipv6_hdr(skb);
 559         struct net *net = dev_net(skb_dst(skb)->dev);
 560
 561         if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
 562                 goto fail_hdr;
 563
 564         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
 565
 566         /* Jumbo payload inhibits frag. header */
 567         if (hdr->payload_len==0)
 568                 goto fail_hdr;
 569
 570         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 571                                  sizeof(struct frag_hdr))))
 572                 goto fail_hdr;
 573
 574         hdr = ipv6_hdr(skb);
 575         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 576
 577         if (!(fhdr->frag_off & htons(0xFFF9))) {
 578                 /* It is not a fragmented frame */
 579                 skb->transport_header += sizeof(struct frag_hdr);
 580                 IP6_INC_STATS_BH(net,
 581                                  ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
 582
 583                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 584                 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
 585                 return 1;
 586         }
 587
 588         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
 589                 ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
 590
 591         fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
 592                      skb->dev ? skb->dev->ifindex : 0);
 593         if (fq != NULL) {
 594                 int ret;
 595
 596                 spin_lock(&fq->q.lock);
 597
 598                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 599
 600                 spin_unlock(&fq->q.lock);
 601                 fq_put(fq);
 602                 return ret;
 603         }
 604
 605         IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
 606         kfree_skb(skb);
 607         return -1;
 608
 609 fail_hdr:
 610         IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
 611         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
 612         return -1;
 613 }
 614
 615 static const struct inet6_protocol frag_protocol =
 616 {
 617         .handler        =       ipv6_frag_rcv,
 618         .flags          =       INET6_PROTO_NOPOLICY,
 619 };
 620
 621 #ifdef CONFIG_SYSCTL
 622 static struct ctl_table ip6_frags_ns_ctl_table[] = {
 623         {
 624                 .procname       = "ip6frag_high_thresh",
 625                 .data           = &init_net.ipv6.frags.high_thresh,
 626                 .maxlen         = sizeof(int),
 627                 .mode           = 0644,
 628                 .proc_handler   = proc_dointvec
 629         },
 630         {
 631                 .procname       = "ip6frag_low_thresh",
 632                 .data           = &init_net.ipv6.frags.low_thresh,
 633                 .maxlen         = sizeof(int),
 634                 .mode           = 0644,
 635                 .proc_handler   = proc_dointvec
 636         },
 637         {
 638                 .procname       = "ip6frag_time",
 639                 .data           = &init_net.ipv6.frags.timeout,
 640                 .maxlen         = sizeof(int),
 641                 .mode           = 0644,
 642                 .proc_handler   = proc_dointvec_jiffies,
 643         },
 644         { }
 645 };
 646
 647 static struct ctl_table ip6_frags_ctl_table[] = {
 648         {
 649                 .procname       = "ip6frag_secret_interval",
 650                 .data           = &ip6_frags.secret_interval,
 651                 .maxlen         = sizeof(int),
 652                 .mode           = 0644,
 653                 .proc_handler   = proc_dointvec_jiffies,
 654         },
 655         { }
 656 };
 657
 658 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
 659 {
 660         struct ctl_table *table;
 661         struct ctl_table_header *hdr;
 662
 663         table = ip6_frags_ns_ctl_table;
 664         if (!net_eq(net, &init_net)) {
 665                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
 666                 if (table == NULL)
 667                         goto err_alloc;
 668
 669                 table[0].data = &net->ipv6.frags.high_thresh;
 670                 table[1].data = &net->ipv6.frags.low_thresh;
 671                 table[2].data = &net->ipv6.frags.timeout;
 672         }
 673
 674         hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
 675         if (hdr == NULL)
 676                 goto err_reg;
 677
 678         net->ipv6.sysctl.frags_hdr = hdr;
 679         return 0;
 680
 681 err_reg:
 682         if (!net_eq(net, &init_net))
 683                 kfree(table);
 684 err_alloc:
 685         return -ENOMEM;
 686 }
 687
 688 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
 689 {
 690         struct ctl_table *table;
 691
 692         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
 693         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
 694         if (!net_eq(net, &init_net))
 695                 kfree(table);
 696 }
 697
 698 static struct ctl_table_header *ip6_ctl_header;
 699
 700 static int ip6_frags_sysctl_register(void)
 701 {
 702         ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
 703                         ip6_frags_ctl_table);
 704         return ip6_ctl_header == NULL ? -ENOMEM : 0;
 705 }
 706
 707 static void ip6_frags_sysctl_unregister(void)
 708 {
 709         unregister_net_sysctl_table(ip6_ctl_header);
 710 }
 711 #else
 712 static inline int ip6_frags_ns_sysctl_register(struct net *net)
 713 {
 714         return 0;
 715 }
 716
 717 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
 718 {
 719 }
 720
 721 static inline int ip6_frags_sysctl_register(void)
 722 {
 723         return 0;
 724 }
 725
 726 static inline void ip6_frags_sysctl_unregister(void)
 727 {
 728 }
 729 #endif
 730
 731 static int __net_init ipv6_frags_init_net(struct net *net)
 732 {
 733         net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
 734         net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
 735         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
 736
 737         inet_frags_init_net(&net->ipv6.frags);
 738
 739         return ip6_frags_ns_sysctl_register(net);
 740 }
 741
 742 static void __net_exit ipv6_frags_exit_net(struct net *net)
 743 {
 744         ip6_frags_ns_sysctl_unregister(net);
 745         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 746 }
 747
 748 static struct pernet_operations ip6_frags_ops = {
 749         .init = ipv6_frags_init_net,
 750         .exit = ipv6_frags_exit_net,
 751 };
 752
 753 int __init ipv6_frag_init(void)
 754 {
 755         int ret;
 756
 757         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 758         if (ret)
 759                 goto out;
 760
 761         ret = ip6_frags_sysctl_register();
 762         if (ret)
 763                 goto err_sysctl;
 764
 765         ret = register_pernet_subsys(&ip6_frags_ops);
 766         if (ret)
 767                 goto err_pernet;
 768
 769         ip6_frags.hashfn = ip6_hashfn;
 770         ip6_frags.constructor = ip6_frag_init;
 771         ip6_frags.destructor = NULL;
 772         ip6_frags.skb_free = NULL;
 773         ip6_frags.qsize = sizeof(struct frag_queue);
 774         ip6_frags.match = ip6_frag_match;
 775         ip6_frags.frag_expire = ip6_frag_expire;
 776         ip6_frags.secret_interval = 10 * 60 * HZ;
 777         inet_frags_init(&ip6_frags);
 778 out:
 779         return ret;
 780
 781 err_pernet:
 782         ip6_frags_sysctl_unregister();
 783 err_sysctl:
 784         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 785         goto out;
 786 }
 787
 788 void ipv6_frag_exit(void)
 789 {
 790         inet_frags_fini(&ip6_frags);
 791         ip6_frags_sysctl_unregister();
 792         unregister_pernet_subsys(&ip6_frags_ops);
 793         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 794 }