git.oblomov.eu Git - linux-2.6/blob - 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  *      $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
   9  *
  10  *      Based on: net/ipv4/ip_fragment.c
  11  *
  12  *      This program is free software; you can redistribute it and/or
  13  *      modify it under the terms of the GNU General Public License
  14  *      as published by the Free Software Foundation; either version
  15  *      2 of the License, or (at your option) any later version.
  16  */
  17
  18 /*
  19  *      Fixes:
  20  *      Andi Kleen      Make it work with multiple hosts.
  21  *                      More RFC compliance.
  22  *
  23  *      Horst von Brand Add missing #include <linux/string.h>
  24  *      Alexey Kuznetsov        SMP races, threading, cleanup.
  25  *      Patrick McHardy         LRU queue of frag heads for evictor.
  26  *      Mitsuru KANDA @USAGI    Register inet6_protocol{}.
  27  *      David Stevens and
  28  *      YOSHIFUJI,H. @USAGI     Always remove fragment header to
  29  *                              calculate ICV correctly.
  30  */
  31 #include <linux/errno.h>
  32 #include <linux/types.h>
  33 #include <linux/string.h>
  34 #include <linux/socket.h>
  35 #include <linux/sockios.h>
  36 #include <linux/jiffies.h>
  37 #include <linux/net.h>
  38 #include <linux/list.h>
  39 #include <linux/netdevice.h>
  40 #include <linux/in6.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/icmpv6.h>
  43 #include <linux/random.h>
  44 #include <linux/jhash.h>
  45 #include <linux/skbuff.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         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 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
 105                                struct in6_addr *daddr)
 106 {
 107         u32 a, b, c;
 108
 109         a = (__force u32)saddr->s6_addr32[0];
 110         b = (__force u32)saddr->s6_addr32[1];
 111         c = (__force u32)saddr->s6_addr32[2];
 112
 113         a += JHASH_GOLDEN_RATIO;
 114         b += JHASH_GOLDEN_RATIO;
 115         c += ip6_frags.rnd;
 116         __jhash_mix(a, b, c);
 117
 118         a += (__force u32)saddr->s6_addr32[3];
 119         b += (__force u32)daddr->s6_addr32[0];
 120         c += (__force u32)daddr->s6_addr32[1];
 121         __jhash_mix(a, b, c);
 122
 123         a += (__force u32)daddr->s6_addr32[2];
 124         b += (__force u32)daddr->s6_addr32[3];
 125         c += (__force u32)id;
 126         __jhash_mix(a, b, c);
 127
 128         return c & (INETFRAGS_HASHSZ - 1);
 129 }
 130
 131 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 132 {
 133         struct frag_queue *fq;
 134
 135         fq = container_of(q, struct frag_queue, q);
 136         return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
 137 }
 138
 139 int ip6_frag_match(struct inet_frag_queue *q, void *a)
 140 {
 141         struct frag_queue *fq;
 142         struct ip6_create_arg *arg = a;
 143
 144         fq = container_of(q, struct frag_queue, q);
 145         return (fq->id == arg->id &&
 146                         ipv6_addr_equal(&fq->saddr, arg->src) &&
 147                         ipv6_addr_equal(&fq->daddr, arg->dst));
 148 }
 149 EXPORT_SYMBOL(ip6_frag_match);
 150
 151 /* Memory Tracking Functions. */
 152 static inline void frag_kfree_skb(struct netns_frags *nf,
 153                 struct sk_buff *skb, int *work)
 154 {
 155         if (work)
 156                 *work -= skb->truesize;
 157         atomic_sub(skb->truesize, &nf->mem);
 158         kfree_skb(skb);
 159 }
 160
 161 void ip6_frag_init(struct inet_frag_queue *q, void *a)
 162 {
 163         struct frag_queue *fq = container_of(q, struct frag_queue, q);
 164         struct ip6_create_arg *arg = a;
 165
 166         fq->id = arg->id;
 167         ipv6_addr_copy(&fq->saddr, arg->src);
 168         ipv6_addr_copy(&fq->daddr, arg->dst);
 169 }
 170 EXPORT_SYMBOL(ip6_frag_init);
 171
 172 /* Destruction primitives. */
 173
 174 static __inline__ void fq_put(struct frag_queue *fq)
 175 {
 176         inet_frag_put(&fq->q, &ip6_frags);
 177 }
 178
 179 /* Kill fq entry. It is not destroyed immediately,
 180  * because caller (and someone more) holds reference count.
 181  */
 182 static __inline__ void fq_kill(struct frag_queue *fq)
 183 {
 184         inet_frag_kill(&fq->q, &ip6_frags);
 185 }
 186
 187 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 188 {
 189         int evicted;
 190
 191         evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
 192         if (evicted)
 193                 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
 194 }
 195
 196 static void ip6_frag_expire(unsigned long data)
 197 {
 198         struct frag_queue *fq;
 199         struct net_device *dev = NULL;
 200         struct net *net;
 201
 202         fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 203
 204         spin_lock(&fq->q.lock);
 205
 206         if (fq->q.last_in & INET_FRAG_COMPLETE)
 207                 goto out;
 208
 209         fq_kill(fq);
 210
 211         net = container_of(fq->q.net, struct net, ipv6.frags);
 212         dev = dev_get_by_index(net, fq->iif);
 213         if (!dev)
 214                 goto out;
 215
 216         rcu_read_lock();
 217         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 218         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 219         rcu_read_unlock();
 220
 221         /* Don't send error if the first segment did not arrive. */
 222         if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
 223                 goto out;
 224
 225         /*
 226            But use as source device on which LAST ARRIVED
 227            segment was received. And do not use fq->dev
 228            pointer directly, device might already disappeared.
 229          */
 230         fq->q.fragments->dev = dev;
 231         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
 232 out:
 233         if (dev)
 234                 dev_put(dev);
 235         spin_unlock(&fq->q.lock);
 236         fq_put(fq);
 237 }
 238
 239 static __inline__ struct frag_queue *
 240 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
 241         struct inet6_dev *idev)
 242 {
 243         struct inet_frag_queue *q;
 244         struct ip6_create_arg arg;
 245         unsigned int hash;
 246
 247         arg.id = id;
 248         arg.src = src;
 249         arg.dst = dst;
 250         hash = ip6qhashfn(id, src, dst);
 251
 252         q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
 253         if (q == NULL)
 254                 goto oom;
 255
 256         return container_of(q, struct frag_queue, q);
 257
 258 oom:
 259         IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
 260         return NULL;
 261 }
 262
 263 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 264                            struct frag_hdr *fhdr, int nhoff)
 265 {
 266         struct sk_buff *prev, *next;
 267         struct net_device *dev;
 268         int offset, end;
 269
 270         if (fq->q.last_in & INET_FRAG_COMPLETE)
 271                 goto err;
 272
 273         offset = ntohs(fhdr->frag_off) & ~0x7;
 274         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 275                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 276
 277         if ((unsigned int)end > IPV6_MAXPLEN) {
 278                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
 279                                  IPSTATS_MIB_INHDRERRORS);
 280                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 281                                   ((u8 *)&fhdr->frag_off -
 282                                    skb_network_header(skb)));
 283                 return -1;
 284         }
 285
 286         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 287                 const unsigned char *nh = skb_network_header(skb);
 288                 skb->csum = csum_sub(skb->csum,
 289                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 290                                                   0));
 291         }
 292
 293         /* Is this the final fragment? */
 294         if (!(fhdr->frag_off & htons(IP6_MF))) {
 295                 /* If we already have some bits beyond end
 296                  * or have different end, the segment is corrupted.
 297                  */
 298                 if (end < fq->q.len ||
 299                     ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
 300                         goto err;
 301                 fq->q.last_in |= INET_FRAG_LAST_IN;
 302                 fq->q.len = end;
 303         } else {
 304                 /* Check if the fragment is rounded to 8 bytes.
 305                  * Required by the RFC.
 306                  */
 307                 if (end & 0x7) {
 308                         /* RFC2460 says always send parameter problem in
 309                          * this case. -DaveM
 310                          */
 311                         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
 312                                          IPSTATS_MIB_INHDRERRORS);
 313                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 314                                           offsetof(struct ipv6hdr, payload_len));
 315                         return -1;
 316                 }
 317                 if (end > fq->q.len) {
 318                         /* Some bits beyond end -> corruption. */
 319                         if (fq->q.last_in & INET_FRAG_LAST_IN)
 320                                 goto err;
 321                         fq->q.len = end;
 322                 }
 323         }
 324
 325         if (end == offset)
 326                 goto err;
 327
 328         /* Point into the IP datagram 'data' part. */
 329         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 330                 goto err;
 331
 332         if (pskb_trim_rcsum(skb, end - offset))
 333                 goto err;
 334
 335         /* Find out which fragments are in front and at the back of us
 336          * in the chain of fragments so far.  We must know where to put
 337          * this fragment, right?
 338          */
 339         prev = NULL;
 340         for(next = fq->q.fragments; next != NULL; next = next->next) {
 341                 if (FRAG6_CB(next)->offset >= offset)
 342                         break;  /* bingo! */
 343                 prev = next;
 344         }
 345
 346         /* We found where to put this one.  Check for overlap with
 347          * preceding fragment, and, if needed, align things so that
 348          * any overlaps are eliminated.
 349          */
 350         if (prev) {
 351                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
 352
 353                 if (i > 0) {
 354                         offset += i;
 355                         if (end <= offset)
 356                                 goto err;
 357                         if (!pskb_pull(skb, i))
 358                                 goto err;
 359                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 360                                 skb->ip_summed = CHECKSUM_NONE;
 361                 }
 362         }
 363
 364         /* Look for overlap with succeeding segments.
 365          * If we can merge fragments, do it.
 366          */
 367         while (next && FRAG6_CB(next)->offset < end) {
 368                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
 369
 370                 if (i < next->len) {
 371                         /* Eat head of the next overlapped fragment
 372                          * and leave the loop. The next ones cannot overlap.
 373                          */
 374                         if (!pskb_pull(next, i))
 375                                 goto err;
 376                         FRAG6_CB(next)->offset += i;    /* next fragment */
 377                         fq->q.meat -= i;
 378                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 379                                 next->ip_summed = CHECKSUM_NONE;
 380                         break;
 381                 } else {
 382                         struct sk_buff *free_it = next;
 383
 384                         /* Old fragment is completely overridden with
 385                          * new one drop it.
 386                          */
 387                         next = next->next;
 388
 389                         if (prev)
 390                                 prev->next = next;
 391                         else
 392                                 fq->q.fragments = next;
 393
 394                         fq->q.meat -= free_it->len;
 395                         frag_kfree_skb(fq->q.net, free_it, NULL);
 396                 }
 397         }
 398
 399         FRAG6_CB(skb)->offset = offset;
 400
 401         /* Insert this fragment in the chain of fragments. */
 402         skb->next = next;
 403         if (prev)
 404                 prev->next = skb;
 405         else
 406                 fq->q.fragments = skb;
 407
 408         dev = skb->dev;
 409         if (dev) {
 410                 fq->iif = dev->ifindex;
 411                 skb->dev = NULL;
 412         }
 413         fq->q.stamp = skb->tstamp;
 414         fq->q.meat += skb->len;
 415         atomic_add(skb->truesize, &fq->q.net->mem);
 416
 417         /* The first fragment.
 418          * nhoffset is obtained from the first fragment, of course.
 419          */
 420         if (offset == 0) {
 421                 fq->nhoffset = nhoff;
 422                 fq->q.last_in |= INET_FRAG_FIRST_IN;
 423         }
 424
 425         if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 426             fq->q.meat == fq->q.len)
 427                 return ip6_frag_reasm(fq, prev, dev);
 428
 429         write_lock(&ip6_frags.lock);
 430         list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
 431         write_unlock(&ip6_frags.lock);
 432         return -1;
 433
 434 err:
 435         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
 436         kfree_skb(skb);
 437         return -1;
 438 }
 439
 440 /*
 441  *      Check if this packet is complete.
 442  *      Returns NULL on failure by any reason, and pointer
 443  *      to current nexthdr field in reassembled frame.
 444  *
 445  *      It is called with locked fq, and caller must check that
 446  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 447  *      the last and the first frames arrived and all the bits are here.
 448  */
 449 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 450                           struct net_device *dev)
 451 {
 452         struct sk_buff *fp, *head = fq->q.fragments;
 453         int    payload_len;
 454         unsigned int nhoff;
 455
 456         fq_kill(fq);
 457
 458         /* Make the one we just received the head. */
 459         if (prev) {
 460                 head = prev->next;
 461                 fp = skb_clone(head, GFP_ATOMIC);
 462
 463                 if (!fp)
 464                         goto out_oom;
 465
 466                 fp->next = head->next;
 467                 prev->next = fp;
 468
 469                 skb_morph(head, fq->q.fragments);
 470                 head->next = fq->q.fragments->next;
 471
 472                 kfree_skb(fq->q.fragments);
 473                 fq->q.fragments = head;
 474         }
 475
 476         BUG_TRAP(head != NULL);
 477         BUG_TRAP(FRAG6_CB(head)->offset == 0);
 478
 479         /* Unfragmented part is taken from the first segment. */
 480         payload_len = ((head->data - skb_network_header(head)) -
 481                        sizeof(struct ipv6hdr) + fq->q.len -
 482                        sizeof(struct frag_hdr));
 483         if (payload_len > IPV6_MAXPLEN)
 484                 goto out_oversize;
 485
 486         /* Head of list must not be cloned. */
 487         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 488                 goto out_oom;
 489
 490         /* If the first fragment is fragmented itself, we split
 491          * it to two chunks: the first with data and paged part
 492          * and the second, holding only fragments. */
 493         if (skb_shinfo(head)->frag_list) {
 494                 struct sk_buff *clone;
 495                 int i, plen = 0;
 496
 497                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 498                         goto out_oom;
 499                 clone->next = head->next;
 500                 head->next = clone;
 501                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 502                 skb_shinfo(head)->frag_list = NULL;
 503                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 504                         plen += skb_shinfo(head)->frags[i].size;
 505                 clone->len = clone->data_len = head->data_len - plen;
 506                 head->data_len -= clone->len;
 507                 head->len -= clone->len;
 508                 clone->csum = 0;
 509                 clone->ip_summed = head->ip_summed;
 510                 atomic_add(clone->truesize, &fq->q.net->mem);
 511         }
 512
 513         /* We have to remove fragment header from datagram and to relocate
 514          * header in order to calculate ICV correctly. */
 515         nhoff = fq->nhoffset;
 516         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 517         memmove(head->head + sizeof(struct frag_hdr), head->head,
 518                 (head->data - head->head) - sizeof(struct frag_hdr));
 519         head->mac_header += sizeof(struct frag_hdr);
 520         head->network_header += sizeof(struct frag_hdr);
 521
 522         skb_shinfo(head)->frag_list = head->next;
 523         skb_reset_transport_header(head);
 524         skb_push(head, head->data - skb_network_header(head));
 525         atomic_sub(head->truesize, &fq->q.net->mem);
 526
 527         for (fp=head->next; fp; fp = fp->next) {
 528                 head->data_len += fp->len;
 529                 head->len += fp->len;
 530                 if (head->ip_summed != fp->ip_summed)
 531                         head->ip_summed = CHECKSUM_NONE;
 532                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 533                         head->csum = csum_add(head->csum, fp->csum);
 534                 head->truesize += fp->truesize;
 535                 atomic_sub(fp->truesize, &fq->q.net->mem);
 536         }
 537
 538         head->next = NULL;
 539         head->dev = dev;
 540         head->tstamp = fq->q.stamp;
 541         ipv6_hdr(head)->payload_len = htons(payload_len);
 542         IP6CB(head)->nhoff = nhoff;
 543
 544         /* Yes, and fold redundant checksum back. 8) */
 545         if (head->ip_summed == CHECKSUM_COMPLETE)
 546                 head->csum = csum_partial(skb_network_header(head),
 547                                           skb_network_header_len(head),
 548                                           head->csum);
 549
 550         rcu_read_lock();
 551         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 552         rcu_read_unlock();
 553         fq->q.fragments = NULL;
 554         return 1;
 555
 556 out_oversize:
 557         if (net_ratelimit())
 558                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
 559         goto out_fail;
 560 out_oom:
 561         if (net_ratelimit())
 562                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
 563 out_fail:
 564         rcu_read_lock();
 565         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 566         rcu_read_unlock();
 567         return -1;
 568 }
 569
 570 static int ipv6_frag_rcv(struct sk_buff *skb)
 571 {
 572         struct frag_hdr *fhdr;
 573         struct frag_queue *fq;
 574         struct ipv6hdr *hdr = ipv6_hdr(skb);
 575         struct net *net;
 576
 577         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
 578
 579         /* Jumbo payload inhibits frag. header */
 580         if (hdr->payload_len==0) {
 581                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
 582                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 583                                   skb_network_header_len(skb));
 584                 return -1;
 585         }
 586         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 587                                  sizeof(struct frag_hdr)))) {
 588                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
 589                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 590                                   skb_network_header_len(skb));
 591                 return -1;
 592         }
 593
 594         hdr = ipv6_hdr(skb);
 595         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 596
 597         if (!(fhdr->frag_off & htons(0xFFF9))) {
 598                 /* It is not a fragmented frame */
 599                 skb->transport_header += sizeof(struct frag_hdr);
 600                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
 601
 602                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 603                 return 1;
 604         }
 605
 606         net = dev_net(skb->dev);
 607         if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
 608                 ip6_evictor(net, ip6_dst_idev(skb->dst));
 609
 610         if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
 611                           ip6_dst_idev(skb->dst))) != NULL) {
 612                 int ret;
 613
 614                 spin_lock(&fq->q.lock);
 615
 616                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 617
 618                 spin_unlock(&fq->q.lock);
 619                 fq_put(fq);
 620                 return ret;
 621         }
 622
 623         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
 624         kfree_skb(skb);
 625         return -1;
 626 }
 627
 628 static struct inet6_protocol frag_protocol =
 629 {
 630         .handler        =       ipv6_frag_rcv,
 631         .flags          =       INET6_PROTO_NOPOLICY,
 632 };
 633
 634 #ifdef CONFIG_SYSCTL
 635 static struct ctl_table ip6_frags_ns_ctl_table[] = {
 636         {
 637                 .ctl_name       = NET_IPV6_IP6FRAG_HIGH_THRESH,
 638                 .procname       = "ip6frag_high_thresh",
 639                 .data           = &init_net.ipv6.frags.high_thresh,
 640                 .maxlen         = sizeof(int),
 641                 .mode           = 0644,
 642                 .proc_handler   = &proc_dointvec
 643         },
 644         {
 645                 .ctl_name       = NET_IPV6_IP6FRAG_LOW_THRESH,
 646                 .procname       = "ip6frag_low_thresh",
 647                 .data           = &init_net.ipv6.frags.low_thresh,
 648                 .maxlen         = sizeof(int),
 649                 .mode           = 0644,
 650                 .proc_handler   = &proc_dointvec
 651         },
 652         {
 653                 .ctl_name       = NET_IPV6_IP6FRAG_TIME,
 654                 .procname       = "ip6frag_time",
 655                 .data           = &init_net.ipv6.frags.timeout,
 656                 .maxlen         = sizeof(int),
 657                 .mode           = 0644,
 658                 .proc_handler   = &proc_dointvec_jiffies,
 659                 .strategy       = &sysctl_jiffies,
 660         },
 661         { }
 662 };
 663
 664 static struct ctl_table ip6_frags_ctl_table[] = {
 665         {
 666                 .ctl_name       = NET_IPV6_IP6FRAG_SECRET_INTERVAL,
 667                 .procname       = "ip6frag_secret_interval",
 668                 .data           = &ip6_frags.secret_interval,
 669                 .maxlen         = sizeof(int),
 670                 .mode           = 0644,
 671                 .proc_handler   = &proc_dointvec_jiffies,
 672                 .strategy       = &sysctl_jiffies
 673         },
 674         { }
 675 };
 676
 677 static int ip6_frags_ns_sysctl_register(struct net *net)
 678 {
 679         struct ctl_table *table;
 680         struct ctl_table_header *hdr;
 681
 682         table = ip6_frags_ns_ctl_table;
 683         if (net != &init_net) {
 684                 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
 685                 if (table == NULL)
 686                         goto err_alloc;
 687
 688                 table[0].data = &net->ipv6.frags.high_thresh;
 689                 table[1].data = &net->ipv6.frags.low_thresh;
 690                 table[2].data = &net->ipv6.frags.timeout;
 691         }
 692
 693         hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
 694         if (hdr == NULL)
 695                 goto err_reg;
 696
 697         net->ipv6.sysctl.frags_hdr = hdr;
 698         return 0;
 699
 700 err_reg:
 701         if (net != &init_net)
 702                 kfree(table);
 703 err_alloc:
 704         return -ENOMEM;
 705 }
 706
 707 static void ip6_frags_ns_sysctl_unregister(struct net *net)
 708 {
 709         struct ctl_table *table;
 710
 711         table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
 712         unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
 713         kfree(table);
 714 }
 715
 716 static struct ctl_table_header *ip6_ctl_header;
 717
 718 static int ip6_frags_sysctl_register(void)
 719 {
 720         ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
 721                         ip6_frags_ctl_table);
 722         return ip6_ctl_header == NULL ? -ENOMEM : 0;
 723 }
 724
 725 static void ip6_frags_sysctl_unregister(void)
 726 {
 727         unregister_net_sysctl_table(ip6_ctl_header);
 728 }
 729 #else
 730 static inline int ip6_frags_ns_sysctl_register(struct net *net)
 731 {
 732         return 0;
 733 }
 734
 735 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
 736 {
 737 }
 738
 739 static inline int ip6_frags_sysctl_register(void)
 740 {
 741         return 0;
 742 }
 743
 744 static inline void ip6_frags_sysctl_unregister(void)
 745 {
 746 }
 747 #endif
 748
 749 static int ipv6_frags_init_net(struct net *net)
 750 {
 751         net->ipv6.frags.high_thresh = 256 * 1024;
 752         net->ipv6.frags.low_thresh = 192 * 1024;
 753         net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
 754
 755         inet_frags_init_net(&net->ipv6.frags);
 756
 757         return ip6_frags_ns_sysctl_register(net);
 758 }
 759
 760 static void ipv6_frags_exit_net(struct net *net)
 761 {
 762         ip6_frags_ns_sysctl_unregister(net);
 763         inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 764 }
 765
 766 static struct pernet_operations ip6_frags_ops = {
 767         .init = ipv6_frags_init_net,
 768         .exit = ipv6_frags_exit_net,
 769 };
 770
 771 int __init ipv6_frag_init(void)
 772 {
 773         int ret;
 774
 775         ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 776         if (ret)
 777                 goto out;
 778
 779         ret = ip6_frags_sysctl_register();
 780         if (ret)
 781                 goto err_sysctl;
 782
 783         ret = register_pernet_subsys(&ip6_frags_ops);
 784         if (ret)
 785                 goto err_pernet;
 786
 787         ip6_frags.hashfn = ip6_hashfn;
 788         ip6_frags.constructor = ip6_frag_init;
 789         ip6_frags.destructor = NULL;
 790         ip6_frags.skb_free = NULL;
 791         ip6_frags.qsize = sizeof(struct frag_queue);
 792         ip6_frags.match = ip6_frag_match;
 793         ip6_frags.frag_expire = ip6_frag_expire;
 794         ip6_frags.secret_interval = 10 * 60 * HZ;
 795         inet_frags_init(&ip6_frags);
 796 out:
 797         return ret;
 798
 799 err_pernet:
 800         ip6_frags_sysctl_unregister();
 801 err_sysctl:
 802         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 803         goto out;
 804 }
 805
 806 void ipv6_frag_exit(void)
 807 {
 808         inet_frags_fini(&ip6_frags);
 809         ip6_frags_sysctl_unregister();
 810         unregister_pernet_subsys(&ip6_frags_ops);
 811         inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 812 }