err.no Git - linux-2.6/blob - net/ipv6/netfilter/nf_conntrack_reasm.c

   1 /*
   2  * IPv6 fragment reassembly for connection tracking
   3  *
   4  * Copyright (C)2004 USAGI/WIDE Project
   5  *
   6  * Author:
   7  *      Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
   8  *
   9  * Based on: net/ipv6/reassembly.c
  10  *
  11  * This program is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU General Public License
  13  * as published by the Free Software Foundation; either version
  14  * 2 of the License, or (at your option) any later version.
  15  */
  16
  17 #include <linux/errno.h>
  18 #include <linux/types.h>
  19 #include <linux/string.h>
  20 #include <linux/socket.h>
  21 #include <linux/sockios.h>
  22 #include <linux/jiffies.h>
  23 #include <linux/net.h>
  24 #include <linux/list.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/in6.h>
  27 #include <linux/ipv6.h>
  28 #include <linux/icmpv6.h>
  29 #include <linux/random.h>
  30 #include <linux/jhash.h>
  31
  32 #include <net/sock.h>
  33 #include <net/snmp.h>
  34 #include <net/inet_frag.h>
  35
  36 #include <net/ipv6.h>
  37 #include <net/protocol.h>
  38 #include <net/transp_v6.h>
  39 #include <net/rawv6.h>
  40 #include <net/ndisc.h>
  41 #include <net/addrconf.h>
  42 #include <linux/sysctl.h>
  43 #include <linux/netfilter.h>
  44 #include <linux/netfilter_ipv6.h>
  45 #include <linux/kernel.h>
  46 #include <linux/module.h>
  47
  48 #define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */
  49 #define NF_CT_FRAG6_LOW_THRESH 196608  /* == 192*1024 */
  50 #define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT
  51
  52 struct nf_ct_frag6_skb_cb
  53 {
  54         struct inet6_skb_parm   h;
  55         int                     offset;
  56         struct sk_buff          *orig;
  57 };
  58
  59 #define NFCT_FRAG6_CB(skb)      ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
  60
  61 struct nf_ct_frag6_queue
  62 {
  63         struct inet_frag_queue  q;
  64
  65         __be32                  id;             /* fragment id          */
  66         struct in6_addr         saddr;
  67         struct in6_addr         daddr;
  68
  69         unsigned int            csum;
  70         __u16                   nhoffset;
  71 };
  72
  73 struct inet_frags_ctl nf_frags_ctl __read_mostly = {
  74         .high_thresh     = 256 * 1024,
  75         .low_thresh      = 192 * 1024,
  76         .timeout         = IPV6_FRAG_TIMEOUT,
  77         .secret_interval = 10 * 60 * HZ,
  78 };
  79
  80 static struct inet_frags nf_frags;
  81
  82 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
  83                                struct in6_addr *daddr)
  84 {
  85         u32 a, b, c;
  86
  87         a = (__force u32)saddr->s6_addr32[0];
  88         b = (__force u32)saddr->s6_addr32[1];
  89         c = (__force u32)saddr->s6_addr32[2];
  90
  91         a += JHASH_GOLDEN_RATIO;
  92         b += JHASH_GOLDEN_RATIO;
  93         c += nf_frags.rnd;
  94         __jhash_mix(a, b, c);
  95
  96         a += (__force u32)saddr->s6_addr32[3];
  97         b += (__force u32)daddr->s6_addr32[0];
  98         c += (__force u32)daddr->s6_addr32[1];
  99         __jhash_mix(a, b, c);
 100
 101         a += (__force u32)daddr->s6_addr32[2];
 102         b += (__force u32)daddr->s6_addr32[3];
 103         c += (__force u32)id;
 104         __jhash_mix(a, b, c);
 105
 106         return c & (INETFRAGS_HASHSZ - 1);
 107 }
 108
 109 static unsigned int nf_hashfn(struct inet_frag_queue *q)
 110 {
 111         struct nf_ct_frag6_queue *nq;
 112
 113         nq = container_of(q, struct nf_ct_frag6_queue, q);
 114         return ip6qhashfn(nq->id, &nq->saddr, &nq->daddr);
 115 }
 116
 117 static void nf_skb_free(struct sk_buff *skb)
 118 {
 119         if (NFCT_FRAG6_CB(skb)->orig)
 120                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
 121 }
 122
 123 /* Memory Tracking Functions. */
 124 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
 125 {
 126         if (work)
 127                 *work -= skb->truesize;
 128         atomic_sub(skb->truesize, &nf_frags.mem);
 129         nf_skb_free(skb);
 130         kfree_skb(skb);
 131 }
 132
 133 static void nf_frag_free(struct inet_frag_queue *q)
 134 {
 135         kfree(container_of(q, struct nf_ct_frag6_queue, q));
 136 }
 137
 138 static inline struct nf_ct_frag6_queue *frag_alloc_queue(void)
 139 {
 140         struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
 141
 142         if (!fq)
 143                 return NULL;
 144         atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_frags.mem);
 145         return fq;
 146 }
 147
 148 /* Destruction primitives. */
 149
 150 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq, unsigned int *work)
 151 {
 152         if (atomic_dec_and_test(&fq->q.refcnt))
 153                 inet_frag_destroy(&fq->q, &nf_frags, work);
 154 }
 155
 156 /* Kill fq entry. It is not destroyed immediately,
 157  * because caller (and someone more) holds reference count.
 158  */
 159 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
 160 {
 161         inet_frag_kill(&fq->q, &nf_frags);
 162 }
 163
 164 static void nf_ct_frag6_evictor(void)
 165 {
 166         struct nf_ct_frag6_queue *fq;
 167         struct list_head *tmp;
 168         unsigned int work;
 169
 170         work = atomic_read(&nf_frags.mem);
 171         if (work <= nf_frags_ctl.low_thresh)
 172                 return;
 173
 174         work -= nf_frags_ctl.low_thresh;
 175         while (work > 0) {
 176                 read_lock(&nf_frags.lock);
 177                 if (list_empty(&nf_frags.lru_list)) {
 178                         read_unlock(&nf_frags.lock);
 179                         return;
 180                 }
 181                 tmp = nf_frags.lru_list.next;
 182                 BUG_ON(tmp == NULL);
 183                 fq = list_entry(tmp, struct nf_ct_frag6_queue, q.lru_list);
 184                 atomic_inc(&fq->q.refcnt);
 185                 read_unlock(&nf_frags.lock);
 186
 187                 spin_lock(&fq->q.lock);
 188                 if (!(fq->q.last_in&COMPLETE))
 189                         fq_kill(fq);
 190                 spin_unlock(&fq->q.lock);
 191
 192                 fq_put(fq, &work);
 193         }
 194 }
 195
 196 static void nf_ct_frag6_expire(unsigned long data)
 197 {
 198         struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
 199
 200         spin_lock(&fq->q.lock);
 201
 202         if (fq->q.last_in & COMPLETE)
 203                 goto out;
 204
 205         fq_kill(fq);
 206
 207 out:
 208         spin_unlock(&fq->q.lock);
 209         fq_put(fq, NULL);
 210 }
 211
 212 /* Creation primitives. */
 213
 214 static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
 215                                           struct nf_ct_frag6_queue *fq_in)
 216 {
 217         struct nf_ct_frag6_queue *fq;
 218 #ifdef CONFIG_SMP
 219         struct hlist_node *n;
 220 #endif
 221
 222         write_lock(&nf_frags.lock);
 223 #ifdef CONFIG_SMP
 224         hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
 225                 if (fq->id == fq_in->id &&
 226                     ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
 227                     ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
 228                         atomic_inc(&fq->q.refcnt);
 229                         write_unlock(&nf_frags.lock);
 230                         fq_in->q.last_in |= COMPLETE;
 231                         fq_put(fq_in, NULL);
 232                         return fq;
 233                 }
 234         }
 235 #endif
 236         fq = fq_in;
 237
 238         if (!mod_timer(&fq->q.timer, jiffies + nf_frags_ctl.timeout))
 239                 atomic_inc(&fq->q.refcnt);
 240
 241         atomic_inc(&fq->q.refcnt);
 242         hlist_add_head(&fq->q.list, &nf_frags.hash[hash]);
 243         INIT_LIST_HEAD(&fq->q.lru_list);
 244         list_add_tail(&fq->q.lru_list, &nf_frags.lru_list);
 245         nf_frags.nqueues++;
 246         write_unlock(&nf_frags.lock);
 247         return fq;
 248 }
 249
 250
 251 static struct nf_ct_frag6_queue *
 252 nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src,                             struct in6_addr *dst)
 253 {
 254         struct nf_ct_frag6_queue *fq;
 255
 256         if ((fq = frag_alloc_queue()) == NULL) {
 257                 pr_debug("Can't alloc new queue\n");
 258                 goto oom;
 259         }
 260
 261         memset(fq, 0, sizeof(struct nf_ct_frag6_queue));
 262
 263         fq->id = id;
 264         ipv6_addr_copy(&fq->saddr, src);
 265         ipv6_addr_copy(&fq->daddr, dst);
 266
 267         setup_timer(&fq->q.timer, nf_ct_frag6_expire, (unsigned long)fq);
 268         spin_lock_init(&fq->q.lock);
 269         atomic_set(&fq->q.refcnt, 1);
 270
 271         return nf_ct_frag6_intern(hash, fq);
 272
 273 oom:
 274         return NULL;
 275 }
 276
 277 static __inline__ struct nf_ct_frag6_queue *
 278 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
 279 {
 280         struct nf_ct_frag6_queue *fq;
 281         struct hlist_node *n;
 282         unsigned int hash = ip6qhashfn(id, src, dst);
 283
 284         read_lock(&nf_frags.lock);
 285         hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
 286                 if (fq->id == id &&
 287                     ipv6_addr_equal(src, &fq->saddr) &&
 288                     ipv6_addr_equal(dst, &fq->daddr)) {
 289                         atomic_inc(&fq->q.refcnt);
 290                         read_unlock(&nf_frags.lock);
 291                         return fq;
 292                 }
 293         }
 294         read_unlock(&nf_frags.lock);
 295
 296         return nf_ct_frag6_create(hash, id, src, dst);
 297 }
 298
 299
 300 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
 301                              struct frag_hdr *fhdr, int nhoff)
 302 {
 303         struct sk_buff *prev, *next;
 304         int offset, end;
 305
 306         if (fq->q.last_in & COMPLETE) {
 307                 pr_debug("Allready completed\n");
 308                 goto err;
 309         }
 310
 311         offset = ntohs(fhdr->frag_off) & ~0x7;
 312         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 313                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 314
 315         if ((unsigned int)end > IPV6_MAXPLEN) {
 316                 pr_debug("offset is too large.\n");
 317                 return -1;
 318         }
 319
 320         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 321                 const unsigned char *nh = skb_network_header(skb);
 322                 skb->csum = csum_sub(skb->csum,
 323                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 324                                                   0));
 325         }
 326
 327         /* Is this the final fragment? */
 328         if (!(fhdr->frag_off & htons(IP6_MF))) {
 329                 /* If we already have some bits beyond end
 330                  * or have different end, the segment is corrupted.
 331                  */
 332                 if (end < fq->q.len ||
 333                     ((fq->q.last_in & LAST_IN) && end != fq->q.len)) {
 334                         pr_debug("already received last fragment\n");
 335                         goto err;
 336                 }
 337                 fq->q.last_in |= LAST_IN;
 338                 fq->q.len = end;
 339         } else {
 340                 /* Check if the fragment is rounded to 8 bytes.
 341                  * Required by the RFC.
 342                  */
 343                 if (end & 0x7) {
 344                         /* RFC2460 says always send parameter problem in
 345                          * this case. -DaveM
 346                          */
 347                         pr_debug("end of fragment not rounded to 8 bytes.\n");
 348                         return -1;
 349                 }
 350                 if (end > fq->q.len) {
 351                         /* Some bits beyond end -> corruption. */
 352                         if (fq->q.last_in & LAST_IN) {
 353                                 pr_debug("last packet already reached.\n");
 354                                 goto err;
 355                         }
 356                         fq->q.len = end;
 357                 }
 358         }
 359
 360         if (end == offset)
 361                 goto err;
 362
 363         /* Point into the IP datagram 'data' part. */
 364         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
 365                 pr_debug("queue: message is too short.\n");
 366                 goto err;
 367         }
 368         if (pskb_trim_rcsum(skb, end - offset)) {
 369                 pr_debug("Can't trim\n");
 370                 goto err;
 371         }
 372
 373         /* Find out which fragments are in front and at the back of us
 374          * in the chain of fragments so far.  We must know where to put
 375          * this fragment, right?
 376          */
 377         prev = NULL;
 378         for (next = fq->q.fragments; next != NULL; next = next->next) {
 379                 if (NFCT_FRAG6_CB(next)->offset >= offset)
 380                         break;  /* bingo! */
 381                 prev = next;
 382         }
 383
 384         /* We found where to put this one.  Check for overlap with
 385          * preceding fragment, and, if needed, align things so that
 386          * any overlaps are eliminated.
 387          */
 388         if (prev) {
 389                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
 390
 391                 if (i > 0) {
 392                         offset += i;
 393                         if (end <= offset) {
 394                                 pr_debug("overlap\n");
 395                                 goto err;
 396                         }
 397                         if (!pskb_pull(skb, i)) {
 398                                 pr_debug("Can't pull\n");
 399                                 goto err;
 400                         }
 401                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 402                                 skb->ip_summed = CHECKSUM_NONE;
 403                 }
 404         }
 405
 406         /* Look for overlap with succeeding segments.
 407          * If we can merge fragments, do it.
 408          */
 409         while (next && NFCT_FRAG6_CB(next)->offset < end) {
 410                 /* overlap is 'i' bytes */
 411                 int i = end - NFCT_FRAG6_CB(next)->offset;
 412
 413                 if (i < next->len) {
 414                         /* Eat head of the next overlapped fragment
 415                          * and leave the loop. The next ones cannot overlap.
 416                          */
 417                         pr_debug("Eat head of the overlapped parts.: %d", i);
 418                         if (!pskb_pull(next, i))
 419                                 goto err;
 420
 421                         /* next fragment */
 422                         NFCT_FRAG6_CB(next)->offset += i;
 423                         fq->q.meat -= i;
 424                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 425                                 next->ip_summed = CHECKSUM_NONE;
 426                         break;
 427                 } else {
 428                         struct sk_buff *free_it = next;
 429
 430                         /* Old fragmnet is completely overridden with
 431                          * new one drop it.
 432                          */
 433                         next = next->next;
 434
 435                         if (prev)
 436                                 prev->next = next;
 437                         else
 438                                 fq->q.fragments = next;
 439
 440                         fq->q.meat -= free_it->len;
 441                         frag_kfree_skb(free_it, NULL);
 442                 }
 443         }
 444
 445         NFCT_FRAG6_CB(skb)->offset = offset;
 446
 447         /* Insert this fragment in the chain of fragments. */
 448         skb->next = next;
 449         if (prev)
 450                 prev->next = skb;
 451         else
 452                 fq->q.fragments = skb;
 453
 454         skb->dev = NULL;
 455         fq->q.stamp = skb->tstamp;
 456         fq->q.meat += skb->len;
 457         atomic_add(skb->truesize, &nf_frags.mem);
 458
 459         /* The first fragment.
 460          * nhoffset is obtained from the first fragment, of course.
 461          */
 462         if (offset == 0) {
 463                 fq->nhoffset = nhoff;
 464                 fq->q.last_in |= FIRST_IN;
 465         }
 466         write_lock(&nf_frags.lock);
 467         list_move_tail(&fq->q.lru_list, &nf_frags.lru_list);
 468         write_unlock(&nf_frags.lock);
 469         return 0;
 470
 471 err:
 472         return -1;
 473 }
 474
 475 /*
 476  *      Check if this packet is complete.
 477  *      Returns NULL on failure by any reason, and pointer
 478  *      to current nexthdr field in reassembled frame.
 479  *
 480  *      It is called with locked fq, and caller must check that
 481  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 482  *      the last and the first frames arrived and all the bits are here.
 483  */
 484 static struct sk_buff *
 485 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
 486 {
 487         struct sk_buff *fp, *op, *head = fq->q.fragments;
 488         int    payload_len;
 489
 490         fq_kill(fq);
 491
 492         BUG_TRAP(head != NULL);
 493         BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
 494
 495         /* Unfragmented part is taken from the first segment. */
 496         payload_len = ((head->data - skb_network_header(head)) -
 497                        sizeof(struct ipv6hdr) + fq->q.len -
 498                        sizeof(struct frag_hdr));
 499         if (payload_len > IPV6_MAXPLEN) {
 500                 pr_debug("payload len is too large.\n");
 501                 goto out_oversize;
 502         }
 503
 504         /* Head of list must not be cloned. */
 505         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
 506                 pr_debug("skb is cloned but can't expand head");
 507                 goto out_oom;
 508         }
 509
 510         /* If the first fragment is fragmented itself, we split
 511          * it to two chunks: the first with data and paged part
 512          * and the second, holding only fragments. */
 513         if (skb_shinfo(head)->frag_list) {
 514                 struct sk_buff *clone;
 515                 int i, plen = 0;
 516
 517                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
 518                         pr_debug("Can't alloc skb\n");
 519                         goto out_oom;
 520                 }
 521                 clone->next = head->next;
 522                 head->next = clone;
 523                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 524                 skb_shinfo(head)->frag_list = NULL;
 525                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 526                         plen += skb_shinfo(head)->frags[i].size;
 527                 clone->len = clone->data_len = head->data_len - plen;
 528                 head->data_len -= clone->len;
 529                 head->len -= clone->len;
 530                 clone->csum = 0;
 531                 clone->ip_summed = head->ip_summed;
 532
 533                 NFCT_FRAG6_CB(clone)->orig = NULL;
 534                 atomic_add(clone->truesize, &nf_frags.mem);
 535         }
 536
 537         /* We have to remove fragment header from datagram and to relocate
 538          * header in order to calculate ICV correctly. */
 539         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
 540         memmove(head->head + sizeof(struct frag_hdr), head->head,
 541                 (head->data - head->head) - sizeof(struct frag_hdr));
 542         head->mac_header += sizeof(struct frag_hdr);
 543         head->network_header += sizeof(struct frag_hdr);
 544
 545         skb_shinfo(head)->frag_list = head->next;
 546         skb_reset_transport_header(head);
 547         skb_push(head, head->data - skb_network_header(head));
 548         atomic_sub(head->truesize, &nf_frags.mem);
 549
 550         for (fp=head->next; fp; fp = fp->next) {
 551                 head->data_len += fp->len;
 552                 head->len += fp->len;
 553                 if (head->ip_summed != fp->ip_summed)
 554                         head->ip_summed = CHECKSUM_NONE;
 555                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 556                         head->csum = csum_add(head->csum, fp->csum);
 557                 head->truesize += fp->truesize;
 558                 atomic_sub(fp->truesize, &nf_frags.mem);
 559         }
 560
 561         head->next = NULL;
 562         head->dev = dev;
 563         head->tstamp = fq->q.stamp;
 564         ipv6_hdr(head)->payload_len = htons(payload_len);
 565
 566         /* Yes, and fold redundant checksum back. 8) */
 567         if (head->ip_summed == CHECKSUM_COMPLETE)
 568                 head->csum = csum_partial(skb_network_header(head),
 569                                           skb_network_header_len(head),
 570                                           head->csum);
 571
 572         fq->q.fragments = NULL;
 573
 574         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
 575         fp = skb_shinfo(head)->frag_list;
 576         if (NFCT_FRAG6_CB(fp)->orig == NULL)
 577                 /* at above code, head skb is divided into two skbs. */
 578                 fp = fp->next;
 579
 580         op = NFCT_FRAG6_CB(head)->orig;
 581         for (; fp; fp = fp->next) {
 582                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
 583
 584                 op->next = orig;
 585                 op = orig;
 586                 NFCT_FRAG6_CB(fp)->orig = NULL;
 587         }
 588
 589         return head;
 590
 591 out_oversize:
 592         if (net_ratelimit())
 593                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
 594         goto out_fail;
 595 out_oom:
 596         if (net_ratelimit())
 597                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
 598 out_fail:
 599         return NULL;
 600 }
 601
 602 /*
 603  * find the header just before Fragment Header.
 604  *
 605  * if success return 0 and set ...
 606  * (*prevhdrp): the value of "Next Header Field" in the header
 607  *              just before Fragment Header.
 608  * (*prevhoff): the offset of "Next Header Field" in the header
 609  *              just before Fragment Header.
 610  * (*fhoff)   : the offset of Fragment Header.
 611  *
 612  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
 613  *
 614  */
 615 static int
 616 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
 617 {
 618         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 619         const int netoff = skb_network_offset(skb);
 620         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
 621         int start = netoff + sizeof(struct ipv6hdr);
 622         int len = skb->len - start;
 623         u8 prevhdr = NEXTHDR_IPV6;
 624
 625         while (nexthdr != NEXTHDR_FRAGMENT) {
 626                 struct ipv6_opt_hdr hdr;
 627                 int hdrlen;
 628
 629                 if (!ipv6_ext_hdr(nexthdr)) {
 630                         return -1;
 631                 }
 632                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
 633                         pr_debug("too short\n");
 634                         return -1;
 635                 }
 636                 if (nexthdr == NEXTHDR_NONE) {
 637                         pr_debug("next header is none\n");
 638                         return -1;
 639                 }
 640                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
 641                         BUG();
 642                 if (nexthdr == NEXTHDR_AUTH)
 643                         hdrlen = (hdr.hdrlen+2)<<2;
 644                 else
 645                         hdrlen = ipv6_optlen(&hdr);
 646
 647                 prevhdr = nexthdr;
 648                 prev_nhoff = start;
 649
 650                 nexthdr = hdr.nexthdr;
 651                 len -= hdrlen;
 652                 start += hdrlen;
 653         }
 654
 655         if (len < 0)
 656                 return -1;
 657
 658         *prevhdrp = prevhdr;
 659         *prevhoff = prev_nhoff;
 660         *fhoff = start;
 661
 662         return 0;
 663 }
 664
 665 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
 666 {
 667         struct sk_buff *clone;
 668         struct net_device *dev = skb->dev;
 669         struct frag_hdr *fhdr;
 670         struct nf_ct_frag6_queue *fq;
 671         struct ipv6hdr *hdr;
 672         int fhoff, nhoff;
 673         u8 prevhdr;
 674         struct sk_buff *ret_skb = NULL;
 675
 676         /* Jumbo payload inhibits frag. header */
 677         if (ipv6_hdr(skb)->payload_len == 0) {
 678                 pr_debug("payload len = 0\n");
 679                 return skb;
 680         }
 681
 682         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
 683                 return skb;
 684
 685         clone = skb_clone(skb, GFP_ATOMIC);
 686         if (clone == NULL) {
 687                 pr_debug("Can't clone skb\n");
 688                 return skb;
 689         }
 690
 691         NFCT_FRAG6_CB(clone)->orig = skb;
 692
 693         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
 694                 pr_debug("message is too short.\n");
 695                 goto ret_orig;
 696         }
 697
 698         skb_set_transport_header(clone, fhoff);
 699         hdr = ipv6_hdr(clone);
 700         fhdr = (struct frag_hdr *)skb_transport_header(clone);
 701
 702         if (!(fhdr->frag_off & htons(0xFFF9))) {
 703                 pr_debug("Invalid fragment offset\n");
 704                 /* It is not a fragmented frame */
 705                 goto ret_orig;
 706         }
 707
 708         if (atomic_read(&nf_frags.mem) > nf_frags_ctl.high_thresh)
 709                 nf_ct_frag6_evictor();
 710
 711         fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
 712         if (fq == NULL) {
 713                 pr_debug("Can't find and can't create new queue\n");
 714                 goto ret_orig;
 715         }
 716
 717         spin_lock(&fq->q.lock);
 718
 719         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
 720                 spin_unlock(&fq->q.lock);
 721                 pr_debug("Can't insert skb to queue\n");
 722                 fq_put(fq, NULL);
 723                 goto ret_orig;
 724         }
 725
 726         if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) {
 727                 ret_skb = nf_ct_frag6_reasm(fq, dev);
 728                 if (ret_skb == NULL)
 729                         pr_debug("Can't reassemble fragmented packets\n");
 730         }
 731         spin_unlock(&fq->q.lock);
 732
 733         fq_put(fq, NULL);
 734         return ret_skb;
 735
 736 ret_orig:
 737         kfree_skb(clone);
 738         return skb;
 739 }
 740
 741 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
 742                         struct net_device *in, struct net_device *out,
 743                         int (*okfn)(struct sk_buff *))
 744 {
 745         struct sk_buff *s, *s2;
 746
 747         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
 748                 nf_conntrack_put_reasm(s->nfct_reasm);
 749                 nf_conntrack_get_reasm(skb);
 750                 s->nfct_reasm = skb;
 751
 752                 s2 = s->next;
 753                 s->next = NULL;
 754
 755                 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
 756                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
 757                 s = s2;
 758         }
 759         nf_conntrack_put_reasm(skb);
 760 }
 761
 762 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
 763 {
 764         struct sk_buff *s, *s2;
 765
 766         for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
 767
 768                 s2 = s->next;
 769                 kfree_skb(s);
 770         }
 771
 772         kfree_skb(skb);
 773
 774         return 0;
 775 }
 776
 777 int nf_ct_frag6_init(void)
 778 {
 779         nf_frags.ctl = &nf_frags_ctl;
 780         nf_frags.hashfn = nf_hashfn;
 781         nf_frags.destructor = nf_frag_free;
 782         nf_frags.skb_free = nf_skb_free;
 783         nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
 784         inet_frags_init(&nf_frags);
 785
 786         return 0;
 787 }
 788
 789 void nf_ct_frag6_cleanup(void)
 790 {
 791         inet_frags_fini(&nf_frags);
 792
 793         nf_frags_ctl.low_thresh = 0;
 794         nf_ct_frag6_evictor();
 795 }