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;
 141
 142         fq = kzalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
 143         if (fq == NULL)
 144                 return NULL;
 145         atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_frags.mem);
 146         return fq;
 147 }
 148
 149 /* Destruction primitives. */
 150
 151 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
 152 {
 153         inet_frag_put(&fq->q, &nf_frags);
 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         inet_frag_evictor(&nf_frags);
 167 }
 168
 169 static void nf_ct_frag6_expire(unsigned long data)
 170 {
 171         struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
 172
 173         spin_lock(&fq->q.lock);
 174
 175         if (fq->q.last_in & COMPLETE)
 176                 goto out;
 177
 178         fq_kill(fq);
 179
 180 out:
 181         spin_unlock(&fq->q.lock);
 182         fq_put(fq);
 183 }
 184
 185 /* Creation primitives. */
 186
 187 static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
 188                                           struct nf_ct_frag6_queue *fq_in)
 189 {
 190         struct nf_ct_frag6_queue *fq;
 191 #ifdef CONFIG_SMP
 192         struct hlist_node *n;
 193 #endif
 194
 195         write_lock(&nf_frags.lock);
 196 #ifdef CONFIG_SMP
 197         hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
 198                 if (fq->id == fq_in->id &&
 199                     ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
 200                     ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
 201                         atomic_inc(&fq->q.refcnt);
 202                         write_unlock(&nf_frags.lock);
 203                         fq_in->q.last_in |= COMPLETE;
 204                         fq_put(fq_in);
 205                         return fq;
 206                 }
 207         }
 208 #endif
 209         fq = fq_in;
 210
 211         if (!mod_timer(&fq->q.timer, jiffies + nf_frags_ctl.timeout))
 212                 atomic_inc(&fq->q.refcnt);
 213
 214         atomic_inc(&fq->q.refcnt);
 215         hlist_add_head(&fq->q.list, &nf_frags.hash[hash]);
 216         INIT_LIST_HEAD(&fq->q.lru_list);
 217         list_add_tail(&fq->q.lru_list, &nf_frags.lru_list);
 218         nf_frags.nqueues++;
 219         write_unlock(&nf_frags.lock);
 220         return fq;
 221 }
 222
 223
 224 static struct nf_ct_frag6_queue *
 225 nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src,                             struct in6_addr *dst)
 226 {
 227         struct nf_ct_frag6_queue *fq;
 228
 229         if ((fq = frag_alloc_queue()) == NULL) {
 230                 pr_debug("Can't alloc new queue\n");
 231                 goto oom;
 232         }
 233
 234         fq->id = id;
 235         ipv6_addr_copy(&fq->saddr, src);
 236         ipv6_addr_copy(&fq->daddr, dst);
 237
 238         setup_timer(&fq->q.timer, nf_ct_frag6_expire, (unsigned long)fq);
 239         spin_lock_init(&fq->q.lock);
 240         atomic_set(&fq->q.refcnt, 1);
 241
 242         return nf_ct_frag6_intern(hash, fq);
 243
 244 oom:
 245         return NULL;
 246 }
 247
 248 static __inline__ struct nf_ct_frag6_queue *
 249 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
 250 {
 251         struct nf_ct_frag6_queue *fq;
 252         struct hlist_node *n;
 253         unsigned int hash = ip6qhashfn(id, src, dst);
 254
 255         read_lock(&nf_frags.lock);
 256         hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
 257                 if (fq->id == id &&
 258                     ipv6_addr_equal(src, &fq->saddr) &&
 259                     ipv6_addr_equal(dst, &fq->daddr)) {
 260                         atomic_inc(&fq->q.refcnt);
 261                         read_unlock(&nf_frags.lock);
 262                         return fq;
 263                 }
 264         }
 265         read_unlock(&nf_frags.lock);
 266
 267         return nf_ct_frag6_create(hash, id, src, dst);
 268 }
 269
 270
 271 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
 272                              struct frag_hdr *fhdr, int nhoff)
 273 {
 274         struct sk_buff *prev, *next;
 275         int offset, end;
 276
 277         if (fq->q.last_in & COMPLETE) {
 278                 pr_debug("Allready completed\n");
 279                 goto err;
 280         }
 281
 282         offset = ntohs(fhdr->frag_off) & ~0x7;
 283         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 284                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 285
 286         if ((unsigned int)end > IPV6_MAXPLEN) {
 287                 pr_debug("offset is too large.\n");
 288                 return -1;
 289         }
 290
 291         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 292                 const unsigned char *nh = skb_network_header(skb);
 293                 skb->csum = csum_sub(skb->csum,
 294                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 295                                                   0));
 296         }
 297
 298         /* Is this the final fragment? */
 299         if (!(fhdr->frag_off & htons(IP6_MF))) {
 300                 /* If we already have some bits beyond end
 301                  * or have different end, the segment is corrupted.
 302                  */
 303                 if (end < fq->q.len ||
 304                     ((fq->q.last_in & LAST_IN) && end != fq->q.len)) {
 305                         pr_debug("already received last fragment\n");
 306                         goto err;
 307                 }
 308                 fq->q.last_in |= LAST_IN;
 309                 fq->q.len = end;
 310         } else {
 311                 /* Check if the fragment is rounded to 8 bytes.
 312                  * Required by the RFC.
 313                  */
 314                 if (end & 0x7) {
 315                         /* RFC2460 says always send parameter problem in
 316                          * this case. -DaveM
 317                          */
 318                         pr_debug("end of fragment not rounded to 8 bytes.\n");
 319                         return -1;
 320                 }
 321                 if (end > fq->q.len) {
 322                         /* Some bits beyond end -> corruption. */
 323                         if (fq->q.last_in & LAST_IN) {
 324                                 pr_debug("last packet already reached.\n");
 325                                 goto err;
 326                         }
 327                         fq->q.len = end;
 328                 }
 329         }
 330
 331         if (end == offset)
 332                 goto err;
 333
 334         /* Point into the IP datagram 'data' part. */
 335         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
 336                 pr_debug("queue: message is too short.\n");
 337                 goto err;
 338         }
 339         if (pskb_trim_rcsum(skb, end - offset)) {
 340                 pr_debug("Can't trim\n");
 341                 goto err;
 342         }
 343
 344         /* Find out which fragments are in front and at the back of us
 345          * in the chain of fragments so far.  We must know where to put
 346          * this fragment, right?
 347          */
 348         prev = NULL;
 349         for (next = fq->q.fragments; next != NULL; next = next->next) {
 350                 if (NFCT_FRAG6_CB(next)->offset >= offset)
 351                         break;  /* bingo! */
 352                 prev = next;
 353         }
 354
 355         /* We found where to put this one.  Check for overlap with
 356          * preceding fragment, and, if needed, align things so that
 357          * any overlaps are eliminated.
 358          */
 359         if (prev) {
 360                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
 361
 362                 if (i > 0) {
 363                         offset += i;
 364                         if (end <= offset) {
 365                                 pr_debug("overlap\n");
 366                                 goto err;
 367                         }
 368                         if (!pskb_pull(skb, i)) {
 369                                 pr_debug("Can't pull\n");
 370                                 goto err;
 371                         }
 372                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 373                                 skb->ip_summed = CHECKSUM_NONE;
 374                 }
 375         }
 376
 377         /* Look for overlap with succeeding segments.
 378          * If we can merge fragments, do it.
 379          */
 380         while (next && NFCT_FRAG6_CB(next)->offset < end) {
 381                 /* overlap is 'i' bytes */
 382                 int i = end - NFCT_FRAG6_CB(next)->offset;
 383
 384                 if (i < next->len) {
 385                         /* Eat head of the next overlapped fragment
 386                          * and leave the loop. The next ones cannot overlap.
 387                          */
 388                         pr_debug("Eat head of the overlapped parts.: %d", i);
 389                         if (!pskb_pull(next, i))
 390                                 goto err;
 391
 392                         /* next fragment */
 393                         NFCT_FRAG6_CB(next)->offset += i;
 394                         fq->q.meat -= i;
 395                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 396                                 next->ip_summed = CHECKSUM_NONE;
 397                         break;
 398                 } else {
 399                         struct sk_buff *free_it = next;
 400
 401                         /* Old fragmnet is completely overridden with
 402                          * new one drop it.
 403                          */
 404                         next = next->next;
 405
 406                         if (prev)
 407                                 prev->next = next;
 408                         else
 409                                 fq->q.fragments = next;
 410
 411                         fq->q.meat -= free_it->len;
 412                         frag_kfree_skb(free_it, NULL);
 413                 }
 414         }
 415
 416         NFCT_FRAG6_CB(skb)->offset = offset;
 417
 418         /* Insert this fragment in the chain of fragments. */
 419         skb->next = next;
 420         if (prev)
 421                 prev->next = skb;
 422         else
 423                 fq->q.fragments = skb;
 424
 425         skb->dev = NULL;
 426         fq->q.stamp = skb->tstamp;
 427         fq->q.meat += skb->len;
 428         atomic_add(skb->truesize, &nf_frags.mem);
 429
 430         /* The first fragment.
 431          * nhoffset is obtained from the first fragment, of course.
 432          */
 433         if (offset == 0) {
 434                 fq->nhoffset = nhoff;
 435                 fq->q.last_in |= FIRST_IN;
 436         }
 437         write_lock(&nf_frags.lock);
 438         list_move_tail(&fq->q.lru_list, &nf_frags.lru_list);
 439         write_unlock(&nf_frags.lock);
 440         return 0;
 441
 442 err:
 443         return -1;
 444 }
 445
 446 /*
 447  *      Check if this packet is complete.
 448  *      Returns NULL on failure by any reason, and pointer
 449  *      to current nexthdr field in reassembled frame.
 450  *
 451  *      It is called with locked fq, and caller must check that
 452  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 453  *      the last and the first frames arrived and all the bits are here.
 454  */
 455 static struct sk_buff *
 456 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
 457 {
 458         struct sk_buff *fp, *op, *head = fq->q.fragments;
 459         int    payload_len;
 460
 461         fq_kill(fq);
 462
 463         BUG_TRAP(head != NULL);
 464         BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
 465
 466         /* Unfragmented part is taken from the first segment. */
 467         payload_len = ((head->data - skb_network_header(head)) -
 468                        sizeof(struct ipv6hdr) + fq->q.len -
 469                        sizeof(struct frag_hdr));
 470         if (payload_len > IPV6_MAXPLEN) {
 471                 pr_debug("payload len is too large.\n");
 472                 goto out_oversize;
 473         }
 474
 475         /* Head of list must not be cloned. */
 476         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
 477                 pr_debug("skb is cloned but can't expand head");
 478                 goto out_oom;
 479         }
 480
 481         /* If the first fragment is fragmented itself, we split
 482          * it to two chunks: the first with data and paged part
 483          * and the second, holding only fragments. */
 484         if (skb_shinfo(head)->frag_list) {
 485                 struct sk_buff *clone;
 486                 int i, plen = 0;
 487
 488                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
 489                         pr_debug("Can't alloc skb\n");
 490                         goto out_oom;
 491                 }
 492                 clone->next = head->next;
 493                 head->next = clone;
 494                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 495                 skb_shinfo(head)->frag_list = NULL;
 496                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 497                         plen += skb_shinfo(head)->frags[i].size;
 498                 clone->len = clone->data_len = head->data_len - plen;
 499                 head->data_len -= clone->len;
 500                 head->len -= clone->len;
 501                 clone->csum = 0;
 502                 clone->ip_summed = head->ip_summed;
 503
 504                 NFCT_FRAG6_CB(clone)->orig = NULL;
 505                 atomic_add(clone->truesize, &nf_frags.mem);
 506         }
 507
 508         /* We have to remove fragment header from datagram and to relocate
 509          * header in order to calculate ICV correctly. */
 510         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
 511         memmove(head->head + sizeof(struct frag_hdr), head->head,
 512                 (head->data - head->head) - sizeof(struct frag_hdr));
 513         head->mac_header += sizeof(struct frag_hdr);
 514         head->network_header += sizeof(struct frag_hdr);
 515
 516         skb_shinfo(head)->frag_list = head->next;
 517         skb_reset_transport_header(head);
 518         skb_push(head, head->data - skb_network_header(head));
 519         atomic_sub(head->truesize, &nf_frags.mem);
 520
 521         for (fp=head->next; fp; fp = fp->next) {
 522                 head->data_len += fp->len;
 523                 head->len += fp->len;
 524                 if (head->ip_summed != fp->ip_summed)
 525                         head->ip_summed = CHECKSUM_NONE;
 526                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 527                         head->csum = csum_add(head->csum, fp->csum);
 528                 head->truesize += fp->truesize;
 529                 atomic_sub(fp->truesize, &nf_frags.mem);
 530         }
 531
 532         head->next = NULL;
 533         head->dev = dev;
 534         head->tstamp = fq->q.stamp;
 535         ipv6_hdr(head)->payload_len = htons(payload_len);
 536
 537         /* Yes, and fold redundant checksum back. 8) */
 538         if (head->ip_summed == CHECKSUM_COMPLETE)
 539                 head->csum = csum_partial(skb_network_header(head),
 540                                           skb_network_header_len(head),
 541                                           head->csum);
 542
 543         fq->q.fragments = NULL;
 544
 545         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
 546         fp = skb_shinfo(head)->frag_list;
 547         if (NFCT_FRAG6_CB(fp)->orig == NULL)
 548                 /* at above code, head skb is divided into two skbs. */
 549                 fp = fp->next;
 550
 551         op = NFCT_FRAG6_CB(head)->orig;
 552         for (; fp; fp = fp->next) {
 553                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
 554
 555                 op->next = orig;
 556                 op = orig;
 557                 NFCT_FRAG6_CB(fp)->orig = NULL;
 558         }
 559
 560         return head;
 561
 562 out_oversize:
 563         if (net_ratelimit())
 564                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
 565         goto out_fail;
 566 out_oom:
 567         if (net_ratelimit())
 568                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
 569 out_fail:
 570         return NULL;
 571 }
 572
 573 /*
 574  * find the header just before Fragment Header.
 575  *
 576  * if success return 0 and set ...
 577  * (*prevhdrp): the value of "Next Header Field" in the header
 578  *              just before Fragment Header.
 579  * (*prevhoff): the offset of "Next Header Field" in the header
 580  *              just before Fragment Header.
 581  * (*fhoff)   : the offset of Fragment Header.
 582  *
 583  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
 584  *
 585  */
 586 static int
 587 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
 588 {
 589         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 590         const int netoff = skb_network_offset(skb);
 591         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
 592         int start = netoff + sizeof(struct ipv6hdr);
 593         int len = skb->len - start;
 594         u8 prevhdr = NEXTHDR_IPV6;
 595
 596         while (nexthdr != NEXTHDR_FRAGMENT) {
 597                 struct ipv6_opt_hdr hdr;
 598                 int hdrlen;
 599
 600                 if (!ipv6_ext_hdr(nexthdr)) {
 601                         return -1;
 602                 }
 603                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
 604                         pr_debug("too short\n");
 605                         return -1;
 606                 }
 607                 if (nexthdr == NEXTHDR_NONE) {
 608                         pr_debug("next header is none\n");
 609                         return -1;
 610                 }
 611                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
 612                         BUG();
 613                 if (nexthdr == NEXTHDR_AUTH)
 614                         hdrlen = (hdr.hdrlen+2)<<2;
 615                 else
 616                         hdrlen = ipv6_optlen(&hdr);
 617
 618                 prevhdr = nexthdr;
 619                 prev_nhoff = start;
 620
 621                 nexthdr = hdr.nexthdr;
 622                 len -= hdrlen;
 623                 start += hdrlen;
 624         }
 625
 626         if (len < 0)
 627                 return -1;
 628
 629         *prevhdrp = prevhdr;
 630         *prevhoff = prev_nhoff;
 631         *fhoff = start;
 632
 633         return 0;
 634 }
 635
 636 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
 637 {
 638         struct sk_buff *clone;
 639         struct net_device *dev = skb->dev;
 640         struct frag_hdr *fhdr;
 641         struct nf_ct_frag6_queue *fq;
 642         struct ipv6hdr *hdr;
 643         int fhoff, nhoff;
 644         u8 prevhdr;
 645         struct sk_buff *ret_skb = NULL;
 646
 647         /* Jumbo payload inhibits frag. header */
 648         if (ipv6_hdr(skb)->payload_len == 0) {
 649                 pr_debug("payload len = 0\n");
 650                 return skb;
 651         }
 652
 653         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
 654                 return skb;
 655
 656         clone = skb_clone(skb, GFP_ATOMIC);
 657         if (clone == NULL) {
 658                 pr_debug("Can't clone skb\n");
 659                 return skb;
 660         }
 661
 662         NFCT_FRAG6_CB(clone)->orig = skb;
 663
 664         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
 665                 pr_debug("message is too short.\n");
 666                 goto ret_orig;
 667         }
 668
 669         skb_set_transport_header(clone, fhoff);
 670         hdr = ipv6_hdr(clone);
 671         fhdr = (struct frag_hdr *)skb_transport_header(clone);
 672
 673         if (!(fhdr->frag_off & htons(0xFFF9))) {
 674                 pr_debug("Invalid fragment offset\n");
 675                 /* It is not a fragmented frame */
 676                 goto ret_orig;
 677         }
 678
 679         if (atomic_read(&nf_frags.mem) > nf_frags_ctl.high_thresh)
 680                 nf_ct_frag6_evictor();
 681
 682         fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
 683         if (fq == NULL) {
 684                 pr_debug("Can't find and can't create new queue\n");
 685                 goto ret_orig;
 686         }
 687
 688         spin_lock(&fq->q.lock);
 689
 690         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
 691                 spin_unlock(&fq->q.lock);
 692                 pr_debug("Can't insert skb to queue\n");
 693                 fq_put(fq);
 694                 goto ret_orig;
 695         }
 696
 697         if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) {
 698                 ret_skb = nf_ct_frag6_reasm(fq, dev);
 699                 if (ret_skb == NULL)
 700                         pr_debug("Can't reassemble fragmented packets\n");
 701         }
 702         spin_unlock(&fq->q.lock);
 703
 704         fq_put(fq);
 705         return ret_skb;
 706
 707 ret_orig:
 708         kfree_skb(clone);
 709         return skb;
 710 }
 711
 712 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
 713                         struct net_device *in, struct net_device *out,
 714                         int (*okfn)(struct sk_buff *))
 715 {
 716         struct sk_buff *s, *s2;
 717
 718         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
 719                 nf_conntrack_put_reasm(s->nfct_reasm);
 720                 nf_conntrack_get_reasm(skb);
 721                 s->nfct_reasm = skb;
 722
 723                 s2 = s->next;
 724                 s->next = NULL;
 725
 726                 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
 727                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
 728                 s = s2;
 729         }
 730         nf_conntrack_put_reasm(skb);
 731 }
 732
 733 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
 734 {
 735         struct sk_buff *s, *s2;
 736
 737         for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
 738
 739                 s2 = s->next;
 740                 kfree_skb(s);
 741         }
 742
 743         kfree_skb(skb);
 744
 745         return 0;
 746 }
 747
 748 int nf_ct_frag6_init(void)
 749 {
 750         nf_frags.ctl = &nf_frags_ctl;
 751         nf_frags.hashfn = nf_hashfn;
 752         nf_frags.destructor = nf_frag_free;
 753         nf_frags.skb_free = nf_skb_free;
 754         nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
 755         inet_frags_init(&nf_frags);
 756
 757         return 0;
 758 }
 759
 760 void nf_ct_frag6_cleanup(void)
 761 {
 762         inet_frags_fini(&nf_frags);
 763
 764         nf_frags_ctl.low_thresh = 0;
 765         nf_ct_frag6_evictor();
 766 }