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/config.h>
  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
  47 #include <net/sock.h>
  48 #include <net/snmp.h>
  49
  50 #include <net/ipv6.h>
  51 #include <net/protocol.h>
  52 #include <net/transp_v6.h>
  53 #include <net/rawv6.h>
  54 #include <net/ndisc.h>
  55 #include <net/addrconf.h>
  56
  57 int sysctl_ip6frag_high_thresh = 256*1024;
  58 int sysctl_ip6frag_low_thresh = 192*1024;
  59
  60 int sysctl_ip6frag_time = IPV6_FRAG_TIMEOUT;
  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 hlist_node       list;
  78         struct list_head lru_list;              /* lru list member      */
  79
  80         __u32                   id;             /* fragment id          */
  81         struct in6_addr         saddr;
  82         struct in6_addr         daddr;
  83
  84         spinlock_t              lock;
  85         atomic_t                refcnt;
  86         struct timer_list       timer;          /* expire timer         */
  87         struct sk_buff          *fragments;
  88         int                     len;
  89         int                     meat;
  90         int                     iif;
  91         struct timeval          stamp;
  92         unsigned int            csum;
  93         __u8                    last_in;        /* has first/last segment arrived? */
  94 #define COMPLETE                4
  95 #define FIRST_IN                2
  96 #define LAST_IN                 1
  97         __u16                   nhoffset;
  98 };
  99
 100 /* Hash table. */
 101
 102 #define IP6Q_HASHSZ     64
 103
 104 static struct hlist_head ip6_frag_hash[IP6Q_HASHSZ];
 105 static DEFINE_RWLOCK(ip6_frag_lock);
 106 static u32 ip6_frag_hash_rnd;
 107 static LIST_HEAD(ip6_frag_lru_list);
 108 int ip6_frag_nqueues = 0;
 109
 110 static __inline__ void __fq_unlink(struct frag_queue *fq)
 111 {
 112         hlist_del(&fq->list);
 113         list_del(&fq->lru_list);
 114         ip6_frag_nqueues--;
 115 }
 116
 117 static __inline__ void fq_unlink(struct frag_queue *fq)
 118 {
 119         write_lock(&ip6_frag_lock);
 120         __fq_unlink(fq);
 121         write_unlock(&ip6_frag_lock);
 122 }
 123
 124 /*
 125  * callers should be careful not to use the hash value outside the ipfrag_lock
 126  * as doing so could race with ipfrag_hash_rnd being recalculated.
 127  */
 128 static unsigned int ip6qhashfn(u32 id, struct in6_addr *saddr,
 129                                struct in6_addr *daddr)
 130 {
 131         u32 a, b, c;
 132
 133         a = saddr->s6_addr32[0];
 134         b = saddr->s6_addr32[1];
 135         c = saddr->s6_addr32[2];
 136
 137         a += JHASH_GOLDEN_RATIO;
 138         b += JHASH_GOLDEN_RATIO;
 139         c += ip6_frag_hash_rnd;
 140         __jhash_mix(a, b, c);
 141
 142         a += saddr->s6_addr32[3];
 143         b += daddr->s6_addr32[0];
 144         c += daddr->s6_addr32[1];
 145         __jhash_mix(a, b, c);
 146
 147         a += daddr->s6_addr32[2];
 148         b += daddr->s6_addr32[3];
 149         c += id;
 150         __jhash_mix(a, b, c);
 151
 152         return c & (IP6Q_HASHSZ - 1);
 153 }
 154
 155 static struct timer_list ip6_frag_secret_timer;
 156 int sysctl_ip6frag_secret_interval = 10 * 60 * HZ;
 157
 158 static void ip6_frag_secret_rebuild(unsigned long dummy)
 159 {
 160         unsigned long now = jiffies;
 161         int i;
 162
 163         write_lock(&ip6_frag_lock);
 164         get_random_bytes(&ip6_frag_hash_rnd, sizeof(u32));
 165         for (i = 0; i < IP6Q_HASHSZ; i++) {
 166                 struct frag_queue *q;
 167                 struct hlist_node *p, *n;
 168
 169                 hlist_for_each_entry_safe(q, p, n, &ip6_frag_hash[i], list) {
 170                         unsigned int hval = ip6qhashfn(q->id,
 171                                                        &q->saddr,
 172                                                        &q->daddr);
 173
 174                         if (hval != i) {
 175                                 hlist_del(&q->list);
 176
 177                                 /* Relink to new hash chain. */
 178                                 hlist_add_head(&q->list,
 179                                                &ip6_frag_hash[hval]);
 180
 181                         }
 182                 }
 183         }
 184         write_unlock(&ip6_frag_lock);
 185
 186         mod_timer(&ip6_frag_secret_timer, now + sysctl_ip6frag_secret_interval);
 187 }
 188
 189 atomic_t ip6_frag_mem = ATOMIC_INIT(0);
 190
 191 /* Memory Tracking Functions. */
 192 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
 193 {
 194         if (work)
 195                 *work -= skb->truesize;
 196         atomic_sub(skb->truesize, &ip6_frag_mem);
 197         kfree_skb(skb);
 198 }
 199
 200 static inline void frag_free_queue(struct frag_queue *fq, int *work)
 201 {
 202         if (work)
 203                 *work -= sizeof(struct frag_queue);
 204         atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem);
 205         kfree(fq);
 206 }
 207
 208 static inline struct frag_queue *frag_alloc_queue(void)
 209 {
 210         struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
 211
 212         if(!fq)
 213                 return NULL;
 214         atomic_add(sizeof(struct frag_queue), &ip6_frag_mem);
 215         return fq;
 216 }
 217
 218 /* Destruction primitives. */
 219
 220 /* Complete destruction of fq. */
 221 static void ip6_frag_destroy(struct frag_queue *fq, int *work)
 222 {
 223         struct sk_buff *fp;
 224
 225         BUG_TRAP(fq->last_in&COMPLETE);
 226         BUG_TRAP(del_timer(&fq->timer) == 0);
 227
 228         /* Release all fragment data. */
 229         fp = fq->fragments;
 230         while (fp) {
 231                 struct sk_buff *xp = fp->next;
 232
 233                 frag_kfree_skb(fp, work);
 234                 fp = xp;
 235         }
 236
 237         frag_free_queue(fq, work);
 238 }
 239
 240 static __inline__ void fq_put(struct frag_queue *fq, int *work)
 241 {
 242         if (atomic_dec_and_test(&fq->refcnt))
 243                 ip6_frag_destroy(fq, work);
 244 }
 245
 246 /* Kill fq entry. It is not destroyed immediately,
 247  * because caller (and someone more) holds reference count.
 248  */
 249 static __inline__ void fq_kill(struct frag_queue *fq)
 250 {
 251         if (del_timer(&fq->timer))
 252                 atomic_dec(&fq->refcnt);
 253
 254         if (!(fq->last_in & COMPLETE)) {
 255                 fq_unlink(fq);
 256                 atomic_dec(&fq->refcnt);
 257                 fq->last_in |= COMPLETE;
 258         }
 259 }
 260
 261 static void ip6_evictor(void)
 262 {
 263         struct frag_queue *fq;
 264         struct list_head *tmp;
 265         int work;
 266
 267         work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh;
 268         if (work <= 0)
 269                 return;
 270
 271         while(work > 0) {
 272                 read_lock(&ip6_frag_lock);
 273                 if (list_empty(&ip6_frag_lru_list)) {
 274                         read_unlock(&ip6_frag_lock);
 275                         return;
 276                 }
 277                 tmp = ip6_frag_lru_list.next;
 278                 fq = list_entry(tmp, struct frag_queue, lru_list);
 279                 atomic_inc(&fq->refcnt);
 280                 read_unlock(&ip6_frag_lock);
 281
 282                 spin_lock(&fq->lock);
 283                 if (!(fq->last_in&COMPLETE))
 284                         fq_kill(fq);
 285                 spin_unlock(&fq->lock);
 286
 287                 fq_put(fq, &work);
 288                 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 289         }
 290 }
 291
 292 static void ip6_frag_expire(unsigned long data)
 293 {
 294         struct frag_queue *fq = (struct frag_queue *) data;
 295         struct net_device *dev;
 296
 297         spin_lock(&fq->lock);
 298
 299         if (fq->last_in & COMPLETE)
 300                 goto out;
 301
 302         fq_kill(fq);
 303
 304         IP6_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
 305         IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 306
 307         /* Don't send error if the first segment did not arrive. */
 308         if (!(fq->last_in&FIRST_IN) || !fq->fragments)
 309                 goto out;
 310
 311         dev = dev_get_by_index(fq->iif);
 312         if (!dev)
 313                 goto out;
 314
 315         /*
 316            But use as source device on which LAST ARRIVED
 317            segment was received. And do not use fq->dev
 318            pointer directly, device might already disappeared.
 319          */
 320         fq->fragments->dev = dev;
 321         icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
 322         dev_put(dev);
 323 out:
 324         spin_unlock(&fq->lock);
 325         fq_put(fq, NULL);
 326 }
 327
 328 /* Creation primitives. */
 329
 330
 331 static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in)
 332 {
 333         struct frag_queue *fq;
 334         unsigned int hash;
 335 #ifdef CONFIG_SMP
 336         struct hlist_node *n;
 337 #endif
 338
 339         write_lock(&ip6_frag_lock);
 340         hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr);
 341 #ifdef CONFIG_SMP
 342         hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) {
 343                 if (fq->id == fq_in->id &&
 344                     ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
 345                     ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
 346                         atomic_inc(&fq->refcnt);
 347                         write_unlock(&ip6_frag_lock);
 348                         fq_in->last_in |= COMPLETE;
 349                         fq_put(fq_in, NULL);
 350                         return fq;
 351                 }
 352         }
 353 #endif
 354         fq = fq_in;
 355
 356         if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time))
 357                 atomic_inc(&fq->refcnt);
 358
 359         atomic_inc(&fq->refcnt);
 360         hlist_add_head(&fq->list, &ip6_frag_hash[hash]);
 361         INIT_LIST_HEAD(&fq->lru_list);
 362         list_add_tail(&fq->lru_list, &ip6_frag_lru_list);
 363         ip6_frag_nqueues++;
 364         write_unlock(&ip6_frag_lock);
 365         return fq;
 366 }
 367
 368
 369 static struct frag_queue *
 370 ip6_frag_create(u32 id, struct in6_addr *src, struct in6_addr *dst)
 371 {
 372         struct frag_queue *fq;
 373
 374         if ((fq = frag_alloc_queue()) == NULL)
 375                 goto oom;
 376
 377         fq->id = id;
 378         ipv6_addr_copy(&fq->saddr, src);
 379         ipv6_addr_copy(&fq->daddr, dst);
 380
 381         init_timer(&fq->timer);
 382         fq->timer.function = ip6_frag_expire;
 383         fq->timer.data = (long) fq;
 384         spin_lock_init(&fq->lock);
 385         atomic_set(&fq->refcnt, 1);
 386
 387         return ip6_frag_intern(fq);
 388
 389 oom:
 390         IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 391         return NULL;
 392 }
 393
 394 static __inline__ struct frag_queue *
 395 fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst)
 396 {
 397         struct frag_queue *fq;
 398         struct hlist_node *n;
 399         unsigned int hash;
 400
 401         read_lock(&ip6_frag_lock);
 402         hash = ip6qhashfn(id, src, dst);
 403         hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) {
 404                 if (fq->id == id &&
 405                     ipv6_addr_equal(src, &fq->saddr) &&
 406                     ipv6_addr_equal(dst, &fq->daddr)) {
 407                         atomic_inc(&fq->refcnt);
 408                         read_unlock(&ip6_frag_lock);
 409                         return fq;
 410                 }
 411         }
 412         read_unlock(&ip6_frag_lock);
 413
 414         return ip6_frag_create(id, src, dst);
 415 }
 416
 417
 418 static void ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
 419                            struct frag_hdr *fhdr, int nhoff)
 420 {
 421         struct sk_buff *prev, *next;
 422         int offset, end;
 423
 424         if (fq->last_in & COMPLETE)
 425                 goto err;
 426
 427         offset = ntohs(fhdr->frag_off) & ~0x7;
 428         end = offset + (ntohs(skb->nh.ipv6h->payload_len) -
 429                         ((u8 *) (fhdr + 1) - (u8 *) (skb->nh.ipv6h + 1)));
 430
 431         if ((unsigned int)end > IPV6_MAXPLEN) {
 432                 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 433                 icmpv6_param_prob(skb,ICMPV6_HDR_FIELD, (u8*)&fhdr->frag_off - skb->nh.raw);
 434                 return;
 435         }
 436
 437         if (skb->ip_summed == CHECKSUM_HW)
 438                 skb->csum = csum_sub(skb->csum,
 439                                      csum_partial(skb->nh.raw, (u8*)(fhdr+1)-skb->nh.raw, 0));
 440
 441         /* Is this the final fragment? */
 442         if (!(fhdr->frag_off & htons(IP6_MF))) {
 443                 /* If we already have some bits beyond end
 444                  * or have different end, the segment is corrupted.
 445                  */
 446                 if (end < fq->len ||
 447                     ((fq->last_in & LAST_IN) && end != fq->len))
 448                         goto err;
 449                 fq->last_in |= LAST_IN;
 450                 fq->len = end;
 451         } else {
 452                 /* Check if the fragment is rounded to 8 bytes.
 453                  * Required by the RFC.
 454                  */
 455                 if (end & 0x7) {
 456                         /* RFC2460 says always send parameter problem in
 457                          * this case. -DaveM
 458                          */
 459                         IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
 460                         icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
 461                                           offsetof(struct ipv6hdr, payload_len));
 462                         return;
 463                 }
 464                 if (end > fq->len) {
 465                         /* Some bits beyond end -> corruption. */
 466                         if (fq->last_in & LAST_IN)
 467                                 goto err;
 468                         fq->len = end;
 469                 }
 470         }
 471
 472         if (end == offset)
 473                 goto err;
 474
 475         /* Point into the IP datagram 'data' part. */
 476         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
 477                 goto err;
 478
 479         if (pskb_trim_rcsum(skb, end - offset))
 480                 goto err;
 481
 482         /* Find out which fragments are in front and at the back of us
 483          * in the chain of fragments so far.  We must know where to put
 484          * this fragment, right?
 485          */
 486         prev = NULL;
 487         for(next = fq->fragments; next != NULL; next = next->next) {
 488                 if (FRAG6_CB(next)->offset >= offset)
 489                         break;  /* bingo! */
 490                 prev = next;
 491         }
 492
 493         /* We found where to put this one.  Check for overlap with
 494          * preceding fragment, and, if needed, align things so that
 495          * any overlaps are eliminated.
 496          */
 497         if (prev) {
 498                 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
 499
 500                 if (i > 0) {
 501                         offset += i;
 502                         if (end <= offset)
 503                                 goto err;
 504                         if (!pskb_pull(skb, i))
 505                                 goto err;
 506                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 507                                 skb->ip_summed = CHECKSUM_NONE;
 508                 }
 509         }
 510
 511         /* Look for overlap with succeeding segments.
 512          * If we can merge fragments, do it.
 513          */
 514         while (next && FRAG6_CB(next)->offset < end) {
 515                 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
 516
 517                 if (i < next->len) {
 518                         /* Eat head of the next overlapped fragment
 519                          * and leave the loop. The next ones cannot overlap.
 520                          */
 521                         if (!pskb_pull(next, i))
 522                                 goto err;
 523                         FRAG6_CB(next)->offset += i;    /* next fragment */
 524                         fq->meat -= i;
 525                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 526                                 next->ip_summed = CHECKSUM_NONE;
 527                         break;
 528                 } else {
 529                         struct sk_buff *free_it = next;
 530
 531                         /* Old fragment is completely overridden with
 532                          * new one drop it.
 533                          */
 534                         next = next->next;
 535
 536                         if (prev)
 537                                 prev->next = next;
 538                         else
 539                                 fq->fragments = next;
 540
 541                         fq->meat -= free_it->len;
 542                         frag_kfree_skb(free_it, NULL);
 543                 }
 544         }
 545
 546         FRAG6_CB(skb)->offset = offset;
 547
 548         /* Insert this fragment in the chain of fragments. */
 549         skb->next = next;
 550         if (prev)
 551                 prev->next = skb;
 552         else
 553                 fq->fragments = skb;
 554
 555         if (skb->dev)
 556                 fq->iif = skb->dev->ifindex;
 557         skb->dev = NULL;
 558         skb_get_timestamp(skb, &fq->stamp);
 559         fq->meat += skb->len;
 560         atomic_add(skb->truesize, &ip6_frag_mem);
 561
 562         /* The first fragment.
 563          * nhoffset is obtained from the first fragment, of course.
 564          */
 565         if (offset == 0) {
 566                 fq->nhoffset = nhoff;
 567                 fq->last_in |= FIRST_IN;
 568         }
 569         write_lock(&ip6_frag_lock);
 570         list_move_tail(&fq->lru_list, &ip6_frag_lru_list);
 571         write_unlock(&ip6_frag_lock);
 572         return;
 573
 574 err:
 575         IP6_INC_STATS(IPSTATS_MIB_REASMFAILS);
 576         kfree_skb(skb);
 577 }
 578
 579 /*
 580  *      Check if this packet is complete.
 581  *      Returns NULL on failure by any reason, and pointer
 582  *      to current nexthdr field in reassembled frame.
 583  *
 584  *      It is called with locked fq, and caller must check that
 585  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 586  *      the last and the first frames arrived and all the bits are here.
 587  */
 588 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff **skb_in,
 589                           struct net_device *dev)
 590 {
 591         struct sk_buff *fp, *head = fq->fragments;
 592         int    payload_len;
 593         unsigned int nhoff;
 594
 595         fq_kill(fq);
 596
 597         BUG_TRAP(head != NULL);
 598         BUG_TRAP(FRAG6_CB(head)->offset == 0);
 599
 600         /* Unfragmented part is taken from the first segment. */
 601         payload_len = (head->data - head->nh.raw) - sizeof(struct ipv6hdr) + fq->len - sizeof(struct frag_hdr);
 602         if (payload_len > IPV6_MAXPLEN)
 603                 goto out_oversize;
 604
 605         /* Head of list must not be cloned. */
 606         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 607                 goto out_oom;
 608
 609         /* If the first fragment is fragmented itself, we split
 610          * it to two chunks: the first with data and paged part
 611          * and the second, holding only fragments. */
 612         if (skb_shinfo(head)->frag_list) {
 613                 struct sk_buff *clone;
 614                 int i, plen = 0;
 615
 616                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 617                         goto out_oom;
 618                 clone->next = head->next;
 619                 head->next = clone;
 620                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 621                 skb_shinfo(head)->frag_list = NULL;
 622                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 623                         plen += skb_shinfo(head)->frags[i].size;
 624                 clone->len = clone->data_len = head->data_len - plen;
 625                 head->data_len -= clone->len;
 626                 head->len -= clone->len;
 627                 clone->csum = 0;
 628                 clone->ip_summed = head->ip_summed;
 629                 atomic_add(clone->truesize, &ip6_frag_mem);
 630         }
 631
 632         /* We have to remove fragment header from datagram and to relocate
 633          * header in order to calculate ICV correctly. */
 634         nhoff = fq->nhoffset;
 635         head->nh.raw[nhoff] = head->h.raw[0];
 636         memmove(head->head + sizeof(struct frag_hdr), head->head,
 637                 (head->data - head->head) - sizeof(struct frag_hdr));
 638         head->mac.raw += sizeof(struct frag_hdr);
 639         head->nh.raw += sizeof(struct frag_hdr);
 640
 641         skb_shinfo(head)->frag_list = head->next;
 642         head->h.raw = head->data;
 643         skb_push(head, head->data - head->nh.raw);
 644         atomic_sub(head->truesize, &ip6_frag_mem);
 645
 646         for (fp=head->next; fp; fp = fp->next) {
 647                 head->data_len += fp->len;
 648                 head->len += fp->len;
 649                 if (head->ip_summed != fp->ip_summed)
 650                         head->ip_summed = CHECKSUM_NONE;
 651                 else if (head->ip_summed == CHECKSUM_HW)
 652                         head->csum = csum_add(head->csum, fp->csum);
 653                 head->truesize += fp->truesize;
 654                 atomic_sub(fp->truesize, &ip6_frag_mem);
 655         }
 656
 657         head->next = NULL;
 658         head->dev = dev;
 659         skb_set_timestamp(head, &fq->stamp);
 660         head->nh.ipv6h->payload_len = htons(payload_len);
 661         IP6CB(head)->nhoff = nhoff;
 662
 663         *skb_in = head;
 664
 665         /* Yes, and fold redundant checksum back. 8) */
 666         if (head->ip_summed == CHECKSUM_HW)
 667                 head->csum = csum_partial(head->nh.raw, head->h.raw-head->nh.raw, head->csum);
 668
 669         IP6_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
 670         fq->fragments = NULL;
 671         return 1;
 672
 673 out_oversize:
 674         if (net_ratelimit())
 675                 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
 676         goto out_fail;
 677 out_oom:
 678         if (net_ratelimit())
 679                 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
 680 out_fail:
 681         IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 682         return -1;
 683 }
 684
 685 static int ipv6_frag_rcv(struct sk_buff **skbp)
 686 {
 687         struct sk_buff *skb = *skbp;
 688         struct net_device *dev = skb->dev;
 689         struct frag_hdr *fhdr;
 690         struct frag_queue *fq;
 691         struct ipv6hdr *hdr;
 692
 693         hdr = skb->nh.ipv6h;
 694
 695         IP6_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
 696
 697         /* Jumbo payload inhibits frag. header */
 698         if (hdr->payload_len==0) {
 699                 IP6_INC_STATS(IPSTATS_MIB_INHDRERRORS);
 700                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
 701                 return -1;
 702         }
 703         if (!pskb_may_pull(skb, (skb->h.raw-skb->data)+sizeof(struct frag_hdr))) {
 704                 IP6_INC_STATS(IPSTATS_MIB_INHDRERRORS);
 705                 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
 706                 return -1;
 707         }
 708
 709         hdr = skb->nh.ipv6h;
 710         fhdr = (struct frag_hdr *)skb->h.raw;
 711
 712         if (!(fhdr->frag_off & htons(0xFFF9))) {
 713                 /* It is not a fragmented frame */
 714                 skb->h.raw += sizeof(struct frag_hdr);
 715                 IP6_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
 716
 717                 IP6CB(skb)->nhoff = (u8*)fhdr - skb->nh.raw;
 718                 return 1;
 719         }
 720
 721         if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh)
 722                 ip6_evictor();
 723
 724         if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr)) != NULL) {
 725                 int ret = -1;
 726
 727                 spin_lock(&fq->lock);
 728
 729                 ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 730
 731                 if (fq->last_in == (FIRST_IN|LAST_IN) &&
 732                     fq->meat == fq->len)
 733                         ret = ip6_frag_reasm(fq, skbp, dev);
 734
 735                 spin_unlock(&fq->lock);
 736                 fq_put(fq, NULL);
 737                 return ret;
 738         }
 739
 740         IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 741         kfree_skb(skb);
 742         return -1;
 743 }
 744
 745 static struct inet6_protocol frag_protocol =
 746 {
 747         .handler        =       ipv6_frag_rcv,
 748         .flags          =       INET6_PROTO_NOPOLICY,
 749 };
 750
 751 void __init ipv6_frag_init(void)
 752 {
 753         if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
 754                 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
 755
 756         ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
 757                                    (jiffies ^ (jiffies >> 6)));
 758
 759         init_timer(&ip6_frag_secret_timer);
 760         ip6_frag_secret_timer.function = ip6_frag_secret_rebuild;
 761         ip6_frag_secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval;
 762         add_timer(&ip6_frag_secret_timer);
 763 }