SAFE public projects git trees. - safe/jmp/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 int sysctl_ip6frag_high_thresh __read_mostly = 256*1024;
  60 int sysctl_ip6frag_low_thresh __read_mostly = 192*1024;
  61
  62 int sysctl_ip6frag_time __read_mostly = IPV6_FRAG_TIMEOUT;
  63
  64 struct ip6frag_skb_cb
  65 {
  66         struct inet6_skb_parm   h;
  67         int                     offset;
  68 };
  69
  70 #define FRAG6_CB(skb)   ((struct ip6frag_skb_cb*)((skb)->cb))
  71
  72
  73 /*
  74  *      Equivalent of ipv4 struct ipq
  75  */
  76
  77 struct frag_queue
  78 {
  79         struct inet_frag_queue  q;
  80
  81         __be32                  id;             /* fragment id          */
  82         struct in6_addr         saddr;
  83         struct in6_addr         daddr;
  84
  85         int                     iif;
  86         unsigned int            csum;
  87         __u16                   nhoffset;
  88 };
  89
  90 static struct inet_frags ip6_frags;
  91
  92 int ip6_frag_nqueues(void)
  93 {
  94         return ip6_frags.nqueues;
  95 }
  96
  97 int ip6_frag_mem(void)
  98 {
  99         return atomic_read(&ip6_frags.mem);
 100 }
 101
 102 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 103                           struct net_device *dev);
 104
 105 static __inline__ void __fq_unlink(struct frag_queue *fq)
 106 {
 107         hlist_del(&fq->q.list);
 108         list_del(&fq->q.lru_list);
 109         ip6_frags.nqueues--;
 110 }
 111
 112 static __inline__ void fq_unlink(struct frag_queue *fq)
 113 {
 114         write_lock(&ip6_frags.lock);
 115         __fq_unlink(fq);
 116         write_unlock(&ip6_frags.lock);
 117 }
 118
 119 /*
 120  * callers should be careful not to use the hash value outside the ipfrag_lock
 121  * as doing so could race with ipfrag_hash_rnd being recalculated.
 122  */
 123 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
 124                                struct in6_addr *daddr)
 125 {
 126         u32 a, b, c;
 127
 128         a = (__force u32)saddr->s6_addr32[0];
 129         b = (__force u32)saddr->s6_addr32[1];
 130         c = (__force u32)saddr->s6_addr32[2];
 131
 132         a += JHASH_GOLDEN_RATIO;
 133         b += JHASH_GOLDEN_RATIO;
 134         c += ip6_frags.rnd;
 135         __jhash_mix(a, b, c);
 136
 137         a += (__force u32)saddr->s6_addr32[3];
 138         b += (__force u32)daddr->s6_addr32[0];
 139         c += (__force u32)daddr->s6_addr32[1];
 140         __jhash_mix(a, b, c);
 141
 142         a += (__force u32)daddr->s6_addr32[2];
 143         b += (__force u32)daddr->s6_addr32[3];
 144         c += (__force u32)id;
 145         __jhash_mix(a, b, c);
 146
 147         return c & (INETFRAGS_HASHSZ - 1);
 148 }
 149
 150 int sysctl_ip6frag_secret_interval __read_mostly = 10 * 60 * HZ;
 151
 152 static void ip6_frag_secret_rebuild(unsigned long dummy)
 153 {
 154         unsigned long now = jiffies;
 155         int i;
 156
 157         write_lock(&ip6_frags.lock);
 158         get_random_bytes(&ip6_frags.rnd, sizeof(u32));
 159         for (i = 0; i < INETFRAGS_HASHSZ; i++) {
 160                 struct frag_queue *q;
 161                 struct hlist_node *p, *n;
 162
 163                 hlist_for_each_entry_safe(q, p, n, &ip6_frags.hash[i], q.list) {
 164                         unsigned int hval = ip6qhashfn(q->id,
 165                                                        &q->saddr,
 166                                                        &q->daddr);
 167
 168                         if (hval != i) {
 169                                 hlist_del(&q->q.list);
 170
 171                                 /* Relink to new hash chain. */
 172                                 hlist_add_head(&q->q.list,
 173                                                &ip6_frags.hash[hval]);
 174
 175                         }
 176                 }
 177         }
 178         write_unlock(&ip6_frags.lock);
 179
 180         mod_timer(&ip6_frags.secret_timer, now + sysctl_ip6frag_secret_interval);
 181 }
 182
 183 /* Memory Tracking Functions. */
 184 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
 185 {
 186         if (work)
 187                 *work -= skb->truesize;
 188         atomic_sub(skb->truesize, &ip6_frags.mem);
 189         kfree_skb(skb);
 190 }
 191
 192 static inline void frag_free_queue(struct frag_queue *fq, int *work)
 193 {
 194         if (work)
 195                 *work -= sizeof(struct frag_queue);
 196         atomic_sub(sizeof(struct frag_queue), &ip6_frags.mem);
 197         kfree(fq);
 198 }
 199
 200 static inline struct frag_queue *frag_alloc_queue(void)
 201 {
 202         struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
 203
 204         if(!fq)
 205                 return NULL;
 206         atomic_add(sizeof(struct frag_queue), &ip6_frags.mem);
 207         return fq;
 208 }
 209
 210 /* Destruction primitives. */
 211
 212 /* Complete destruction of fq. */
 213 static void ip6_frag_destroy(struct frag_queue *fq, int *work)
 214 {
 215         struct sk_buff *fp;
 216
 217         BUG_TRAP(fq->q.last_in&COMPLETE);
 218         BUG_TRAP(del_timer(&fq->q.timer) == 0);
 219
 220         /* Release all fragment data. */
 221         fp = fq->q.fragments;
 222         while (fp) {
 223                 struct sk_buff *xp = fp->next;
 224
 225                 frag_kfree_skb(fp, work);
 226                 fp = xp;
 227         }
 228
 229         frag_free_queue(fq, work);
 230 }
 231
 232 static __inline__ void fq_put(struct frag_queue *fq, int *work)
 233 {
 234         if (atomic_dec_and_test(&fq->q.refcnt))
 235                 ip6_frag_destroy(fq, work);
 236 }
 237
 238 /* Kill fq entry. It is not destroyed immediately,
 239  * because caller (and someone more) holds reference count.
 240  */
 241 static __inline__ void fq_kill(struct frag_queue *fq)
 242 {
 243         if (del_timer(&fq->q.timer))
 244                 atomic_dec(&fq->q.refcnt);
 245
 246         if (!(fq->q.last_in & COMPLETE)) {
 247                 fq_unlink(fq);
 248                 atomic_dec(&fq->q.refcnt);
 249                 fq->q.last_in |= COMPLETE;
 250         }
 251 }
 252
 253 static void ip6_evictor(struct inet6_dev *idev)
 254 {
 255         struct frag_queue *fq;
 256         struct list_head *tmp;
 257         int work;
 258
 259         work = atomic_read(&ip6_frags.mem) - sysctl_ip6frag_low_thresh;
 260         if (work <= 0)
 261                 return;
 262
 263         while(work > 0) {
 264                 read_lock(&ip6_frags.lock);
 265                 if (list_empty(&ip6_frags.lru_list)) {
 266                         read_unlock(&ip6_frags.lock);
 267                         return;
 268                 }
 269                 tmp = ip6_frags.lru_list.next;
 270                 fq = list_entry(tmp, struct frag_queue, q.lru_list);
 271                 atomic_inc(&fq->q.refcnt);
 272                 read_unlock(&ip6_frags.lock);
 273
 274                 spin_lock(&fq->q.lock);
 275                 if (!(fq->q.last_in&COMPLETE))
 276                         fq_kill(fq);
 277                 spin_unlock(&fq->q.lock);
 278
 279                 fq_put(fq, &work);
 280                 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
 281         }
 282 }
 283
 284 static void ip6_frag_expire(unsigned long data)
 285 {
 286         struct frag_queue *fq = (struct frag_queue *) data;
 287         struct net_device *dev = NULL;
 288
 289         spin_lock(&fq->q.lock);
 290
 291         if (fq->q.last_in & COMPLETE)
 292                 goto out;
 293
 294         fq_kill(fq);
 295
 296         dev = dev_get_by_index(&init_net, fq->iif);
 297         if (!dev)
 298                 goto out;
 299
 300         rcu_read_lock();
 301         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
 302         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 303         rcu_read_unlock();
 304
 305         /* Don't send error if the first segment did not arrive. */
 306         if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
 307                 goto out;
 308
 309         /*
 310            But use as source device on which LAST ARRIVED
 311            segment was received. And do not use fq->dev
 312            pointer directly, device might already disappeared.
 313          */
 314         fq->q.fragments->dev = dev;
 315         icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
 316 out:
 317         if (dev)
 318                 dev_put(dev);
 319         spin_unlock(&fq->q.lock);
 320         fq_put(fq, NULL);
 321 }
 322
 323 /* Creation primitives. */
 324
 325
 326 static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in)
 327 {
 328         struct frag_queue *fq;
 329         unsigned int hash;
 330 #ifdef CONFIG_SMP
 331         struct hlist_node *n;
 332 #endif
 333
 334         write_lock(&ip6_frags.lock);
 335         hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr);
 336 #ifdef CONFIG_SMP
 337         hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) {
 338                 if (fq->id == fq_in->id &&
 339                     ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
 340                     ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
 341                         atomic_inc(&fq->q.refcnt);
 342                         write_unlock(&ip6_frags.lock);
 343                         fq_in->q.last_in |= COMPLETE;
 344                         fq_put(fq_in, NULL);
 345                         return fq;
 346                 }
 347         }
 348 #endif
 349         fq = fq_in;
 350
 351         if (!mod_timer(&fq->q.timer, jiffies + sysctl_ip6frag_time))
 352                 atomic_inc(&fq->q.refcnt);
 353
 354         atomic_inc(&fq->q.refcnt);
 355         hlist_add_head(&fq->q.list, &ip6_frags.hash[hash]);
 356         INIT_LIST_HEAD(&fq->q.lru_list);
 357         list_add_tail(&fq->q.lru_list, &ip6_frags.lru_list);
 358         ip6_frags.nqueues++;
 359         write_unlock(&ip6_frags.lock);
 360         return fq;
 361 }
 362
 363
 364 static struct frag_queue *
 365 ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst,
 366                 struct inet6_dev *idev)
 367 {
 368         struct frag_queue *fq;
 369
 370         if ((fq = frag_alloc_queue()) == NULL)
 371                 goto oom;
 372
 373         fq->id = id;
 374         ipv6_addr_copy(&fq->saddr, src);
 375         ipv6_addr_copy(&fq->daddr, dst);
 376
 377         init_timer(&fq->q.timer);
 378         fq->q.timer.function = ip6_frag_expire;
 379         fq->q.timer.data = (long) fq;
 380         spin_lock_init(&fq->q.lock);
 381         atomic_set(&fq->q.refcnt, 1);
 382
 383         return ip6_frag_intern(fq);
 384
 385 oom:
 386         IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
 387         return NULL;
 388 }
 389
 390 static __inline__ struct frag_queue *
 391 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
 392         struct inet6_dev *idev)
 393 {
 394         struct frag_queue *fq;
 395         struct hlist_node *n;
 396         unsigned int hash;
 397
 398         read_lock(&ip6_frags.lock);
 399         hash = ip6qhashfn(id, src, dst);
 400         hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) {
 401                 if (fq->id == id &&
 402                     ipv6_addr_equal(src, &fq->saddr) &&
 403                     ipv6_addr_equal(dst, &fq->daddr)) {
 404                         atomic_inc(&fq->q.refcnt);
 405                         read_unlock(&ip6_frags.lock);
 406                         return fq;
 407                 }
 408         }
 409         read_unlock(&ip6_frags.lock);
 410
 411         return ip6_frag_create(id, src, dst, idev);
 412 }
 413
 414
 415 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 416                            struct frag_hdr *fhdr, int nhoff)
 417 {
 418         struct sk_buff *prev, *next;
 419         struct net_device *dev;
 420         int offset, end;
 421
 422         if (fq->q.last_in & COMPLETE)
 423                 goto err;
 424
 425         offset = ntohs(fhdr->frag_off) & ~0x7;
 426         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 427                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 428
 429         if ((unsigned int)end > IPV6_MAXPLEN) {
 430                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
 431                                  IPSTATS_MIB_INHDRERRORS);
 432                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 433                                   ((u8 *)&fhdr->frag_off -
 434                                    skb_network_header(skb)));
 435                 return -1;
 436         }
 437
 438         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 439                 const unsigned char *nh = skb_network_header(skb);
 440                 skb->csum = csum_sub(skb->csum,
 441                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 442                                                   0));
 443         }
 444
 445         /* Is this the final fragment? */
 446         if (!(fhdr->frag_off & htons(IP6_MF))) {
 447                 /* If we already have some bits beyond end
 448                  * or have different end, the segment is corrupted.
 449                  */
 450                 if (end < fq->q.len ||
 451                     ((fq->q.last_in & LAST_IN) && end != fq->q.len))
 452                         goto err;
 453                 fq->q.last_in |= LAST_IN;
 454                 fq->q.len = end;
 455         } else {
 456                 /* Check if the fragment is rounded to 8 bytes.
 457                  * Required by the RFC.
 458                  */
 459                 if (end & 0x7) {
 460                         /* RFC2460 says always send parameter problem in
 461                          * this case. -DaveM
 462                          */
 463                         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
 464                                          IPSTATS_MIB_INHDRERRORS);
 465                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 466                                           offsetof(struct ipv6hdr, payload_len));
 467                         return -1;
 468                 }
 469                 if (end > fq->q.len) {
 470                         /* Some bits beyond end -> corruption. */
 471                         if (fq->q.last_in & LAST_IN)
 472                                 goto err;
 473                         fq->q.len = end;
 474                 }
 475         }
 476
 477         if (end == offset)
 478                 goto err;
 479
 480         /* Point into the IP datagram 'data' part. */
 481         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 482                 goto err;
 483
 484         if (pskb_trim_rcsum(skb, end - offset))
 485                 goto err;
 486
 487         /* Find out which fragments are in front and at the back of us
 488          * in the chain of fragments so far.  We must know where to put
 489          * this fragment, right?
 490          */
 491         prev = NULL;
 492         for(next = fq->q.fragments; next != NULL; next = next->next) {
 493                 if (FRAG6_CB(next)->offset >= offset)
 494                         break;  /* bingo! */
 495                 prev = next;
 496         }
 497
 498         /* We found where to put this one.  Check for overlap with
 499          * preceding fragment, and, if needed, align things so that
 500          * any overlaps are eliminated.
 501          */
 502         if (prev) {
 503                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
 504
 505                 if (i > 0) {
 506                         offset += i;
 507                         if (end <= offset)
 508                                 goto err;
 509                         if (!pskb_pull(skb, i))
 510                                 goto err;
 511                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 512                                 skb->ip_summed = CHECKSUM_NONE;
 513                 }
 514         }
 515
 516         /* Look for overlap with succeeding segments.
 517          * If we can merge fragments, do it.
 518          */
 519         while (next && FRAG6_CB(next)->offset < end) {
 520                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
 521
 522                 if (i < next->len) {
 523                         /* Eat head of the next overlapped fragment
 524                          * and leave the loop. The next ones cannot overlap.
 525                          */
 526                         if (!pskb_pull(next, i))
 527                                 goto err;
 528                         FRAG6_CB(next)->offset += i;    /* next fragment */
 529                         fq->q.meat -= i;
 530                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 531                                 next->ip_summed = CHECKSUM_NONE;
 532                         break;
 533                 } else {
 534                         struct sk_buff *free_it = next;
 535
 536                         /* Old fragment is completely overridden with
 537                          * new one drop it.
 538                          */
 539                         next = next->next;
 540
 541                         if (prev)
 542                                 prev->next = next;
 543                         else
 544                                 fq->q.fragments = next;
 545
 546                         fq->q.meat -= free_it->len;
 547                         frag_kfree_skb(free_it, NULL);
 548                 }
 549         }
 550
 551         FRAG6_CB(skb)->offset = offset;
 552
 553         /* Insert this fragment in the chain of fragments. */
 554         skb->next = next;
 555         if (prev)
 556                 prev->next = skb;
 557         else
 558                 fq->q.fragments = skb;
 559
 560         dev = skb->dev;
 561         if (dev) {
 562                 fq->iif = dev->ifindex;
 563                 skb->dev = NULL;
 564         }
 565         fq->q.stamp = skb->tstamp;
 566         fq->q.meat += skb->len;
 567         atomic_add(skb->truesize, &ip6_frags.mem);
 568
 569         /* The first fragment.
 570          * nhoffset is obtained from the first fragment, of course.
 571          */
 572         if (offset == 0) {
 573                 fq->nhoffset = nhoff;
 574                 fq->q.last_in |= FIRST_IN;
 575         }
 576
 577         if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
 578                 return ip6_frag_reasm(fq, prev, dev);
 579
 580         write_lock(&ip6_frags.lock);
 581         list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
 582         write_unlock(&ip6_frags.lock);
 583         return -1;
 584
 585 err:
 586         IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
 587         kfree_skb(skb);
 588         return -1;
 589 }
 590
 591 /*
 592  *      Check if this packet is complete.
 593  *      Returns NULL on failure by any reason, and pointer
 594  *      to current nexthdr field in reassembled frame.
 595  *
 596  *      It is called with locked fq, and caller must check that
 597  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 598  *      the last and the first frames arrived and all the bits are here.
 599  */
 600 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
 601                           struct net_device *dev)
 602 {
 603         struct sk_buff *fp, *head = fq->q.fragments;
 604         int    payload_len;
 605         unsigned int nhoff;
 606
 607         fq_kill(fq);
 608
 609         /* Make the one we just received the head. */
 610         if (prev) {
 611                 head = prev->next;
 612                 fp = skb_clone(head, GFP_ATOMIC);
 613
 614                 if (!fp)
 615                         goto out_oom;
 616
 617                 fp->next = head->next;
 618                 prev->next = fp;
 619
 620                 skb_morph(head, fq->q.fragments);
 621                 head->next = fq->q.fragments->next;
 622
 623                 kfree_skb(fq->q.fragments);
 624                 fq->q.fragments = head;
 625         }
 626
 627         BUG_TRAP(head != NULL);
 628         BUG_TRAP(FRAG6_CB(head)->offset == 0);
 629
 630         /* Unfragmented part is taken from the first segment. */
 631         payload_len = ((head->data - skb_network_header(head)) -
 632                        sizeof(struct ipv6hdr) + fq->q.len -
 633                        sizeof(struct frag_hdr));
 634         if (payload_len > IPV6_MAXPLEN)
 635                 goto out_oversize;
 636
 637         /* Head of list must not be cloned. */
 638         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 639                 goto out_oom;
 640
 641         /* If the first fragment is fragmented itself, we split
 642          * it to two chunks: the first with data and paged part
 643          * and the second, holding only fragments. */
 644         if (skb_shinfo(head)->frag_list) {
 645                 struct sk_buff *clone;
 646                 int i, plen = 0;
 647
 648                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 649                         goto out_oom;
 650                 clone->next = head->next;
 651                 head->next = clone;
 652                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 653                 skb_shinfo(head)->frag_list = NULL;
 654                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 655                         plen += skb_shinfo(head)->frags[i].size;
 656                 clone->len = clone->data_len = head->data_len - plen;
 657                 head->data_len -= clone->len;
 658                 head->len -= clone->len;
 659                 clone->csum = 0;
 660                 clone->ip_summed = head->ip_summed;
 661                 atomic_add(clone->truesize, &ip6_frags.mem);
 662         }
 663
 664         /* We have to remove fragment header from datagram and to relocate
 665          * header in order to calculate ICV correctly. */
 666         nhoff = fq->nhoffset;
 667         skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
 668         memmove(head->head + sizeof(struct frag_hdr), head->head,
 669                 (head->data - head->head) - sizeof(struct frag_hdr));
 670         head->mac_header += sizeof(struct frag_hdr);
 671         head->network_header += sizeof(struct frag_hdr);
 672
 673         skb_shinfo(head)->frag_list = head->next;
 674         skb_reset_transport_header(head);
 675         skb_push(head, head->data - skb_network_header(head));
 676         atomic_sub(head->truesize, &ip6_frags.mem);
 677
 678         for (fp=head->next; fp; fp = fp->next) {
 679                 head->data_len += fp->len;
 680                 head->len += fp->len;
 681                 if (head->ip_summed != fp->ip_summed)
 682                         head->ip_summed = CHECKSUM_NONE;
 683                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 684                         head->csum = csum_add(head->csum, fp->csum);
 685                 head->truesize += fp->truesize;
 686                 atomic_sub(fp->truesize, &ip6_frags.mem);
 687         }
 688
 689         head->next = NULL;
 690         head->dev = dev;
 691         head->tstamp = fq->q.stamp;
 692         ipv6_hdr(head)->payload_len = htons(payload_len);
 693         IP6CB(head)->nhoff = nhoff;
 694
 695         /* Yes, and fold redundant checksum back. 8) */
 696         if (head->ip_summed == CHECKSUM_COMPLETE)
 697                 head->csum = csum_partial(skb_network_header(head),
 698                                           skb_network_header_len(head),
 699                                           head->csum);
 700
 701         rcu_read_lock();
 702         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
 703         rcu_read_unlock();
 704         fq->q.fragments = NULL;
 705         return 1;
 706
 707 out_oversize:
 708         if (net_ratelimit())
 709                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
 710         goto out_fail;
 711 out_oom:
 712         if (net_ratelimit())
 713                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
 714 out_fail:
 715         rcu_read_lock();
 716         IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 717         rcu_read_unlock();
 718         return -1;
 719 }
 720
 721 static int ipv6_frag_rcv(struct sk_buff **skbp)
 722 {
 723         struct sk_buff *skb = *skbp;
 724         struct frag_hdr *fhdr;
 725         struct frag_queue *fq;
 726         struct ipv6hdr *hdr = ipv6_hdr(skb);
 727
 728         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
 729
 730         /* Jumbo payload inhibits frag. header */
 731         if (hdr->payload_len==0) {
 732                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
 733                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 734                                   skb_network_header_len(skb));
 735                 return -1;
 736         }
 737         if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
 738                                  sizeof(struct frag_hdr)))) {
 739                 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
 740                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 741                                   skb_network_header_len(skb));
 742                 return -1;
 743         }
 744
 745         hdr = ipv6_hdr(skb);
 746         fhdr = (struct frag_hdr *)skb_transport_header(skb);
 747
 748         if (!(fhdr->frag_off & htons(0xFFF9))) {
 749                 /* It is not a fragmented frame */
 750                 skb->transport_header += sizeof(struct frag_hdr);
 751                 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
 752
 753                 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
 754                 return 1;
 755         }
 756
 757         if (atomic_read(&ip6_frags.mem) > sysctl_ip6frag_high_thresh)
 758                 ip6_evictor(ip6_dst_idev(skb->dst));
 759
 760         if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
 761                           ip6_dst_idev(skb->dst))) != NULL) {
 762                 int ret;
 763
 764                 spin_lock(&fq->q.lock);
 765
 766                 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 767
 768                 spin_unlock(&fq->q.lock);
 769                 fq_put(fq, NULL);
 770                 return ret;
 771         }
 772
 773         IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
 774         kfree_skb(skb);
 775         return -1;
 776 }
 777
 778 static struct inet6_protocol frag_protocol =
 779 {
 780         .handler        =       ipv6_frag_rcv,
 781         .flags          =       INET6_PROTO_NOPOLICY,
 782 };
 783
 784 void __init ipv6_frag_init(void)
 785 {
 786         if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
 787                 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
 788
 789         init_timer(&ip6_frags.secret_timer);
 790         ip6_frags.secret_timer.function = ip6_frag_secret_rebuild;
 791         ip6_frags.secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval;
 792         add_timer(&ip6_frags.secret_timer);
 793
 794         inet_frags_init(&ip6_frags);
 795 }