SAFE public projects git trees. - safe/jmp/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 unsigned int nf_ct_frag6_high_thresh __read_mostly = 256*1024;
  53 unsigned int nf_ct_frag6_low_thresh __read_mostly = 192*1024;
  54 unsigned long nf_ct_frag6_timeout __read_mostly = IPV6_FRAG_TIMEOUT;
  55
  56 struct nf_ct_frag6_skb_cb
  57 {
  58         struct inet6_skb_parm   h;
  59         int                     offset;
  60         struct sk_buff          *orig;
  61 };
  62
  63 #define NFCT_FRAG6_CB(skb)      ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
  64
  65 struct nf_ct_frag6_queue
  66 {
  67         struct inet_frag_queue  q;
  68
  69         __be32                  id;             /* fragment id          */
  70         struct in6_addr         saddr;
  71         struct in6_addr         daddr;
  72
  73         unsigned int            csum;
  74         __u16                   nhoffset;
  75 };
  76
  77 /* Hash table. */
  78
  79 #define FRAG6Q_HASHSZ   64
  80
  81 static struct hlist_head nf_ct_frag6_hash[FRAG6Q_HASHSZ];
  82 static DEFINE_RWLOCK(nf_ct_frag6_lock);
  83 static u32 nf_ct_frag6_hash_rnd;
  84 static LIST_HEAD(nf_ct_frag6_lru_list);
  85 int nf_ct_frag6_nqueues = 0;
  86
  87 static __inline__ void __fq_unlink(struct nf_ct_frag6_queue *fq)
  88 {
  89         hlist_del(&fq->q.list);
  90         list_del(&fq->q.lru_list);
  91         nf_ct_frag6_nqueues--;
  92 }
  93
  94 static __inline__ void fq_unlink(struct nf_ct_frag6_queue *fq)
  95 {
  96         write_lock(&nf_ct_frag6_lock);
  97         __fq_unlink(fq);
  98         write_unlock(&nf_ct_frag6_lock);
  99 }
 100
 101 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
 102                                struct in6_addr *daddr)
 103 {
 104         u32 a, b, c;
 105
 106         a = (__force u32)saddr->s6_addr32[0];
 107         b = (__force u32)saddr->s6_addr32[1];
 108         c = (__force u32)saddr->s6_addr32[2];
 109
 110         a += JHASH_GOLDEN_RATIO;
 111         b += JHASH_GOLDEN_RATIO;
 112         c += nf_ct_frag6_hash_rnd;
 113         __jhash_mix(a, b, c);
 114
 115         a += (__force u32)saddr->s6_addr32[3];
 116         b += (__force u32)daddr->s6_addr32[0];
 117         c += (__force u32)daddr->s6_addr32[1];
 118         __jhash_mix(a, b, c);
 119
 120         a += (__force u32)daddr->s6_addr32[2];
 121         b += (__force u32)daddr->s6_addr32[3];
 122         c += (__force u32)id;
 123         __jhash_mix(a, b, c);
 124
 125         return c & (FRAG6Q_HASHSZ - 1);
 126 }
 127
 128 static struct timer_list nf_ct_frag6_secret_timer;
 129 int nf_ct_frag6_secret_interval = 10 * 60 * HZ;
 130
 131 static void nf_ct_frag6_secret_rebuild(unsigned long dummy)
 132 {
 133         unsigned long now = jiffies;
 134         int i;
 135
 136         write_lock(&nf_ct_frag6_lock);
 137         get_random_bytes(&nf_ct_frag6_hash_rnd, sizeof(u32));
 138         for (i = 0; i < FRAG6Q_HASHSZ; i++) {
 139                 struct nf_ct_frag6_queue *q;
 140                 struct hlist_node *p, *n;
 141
 142                 hlist_for_each_entry_safe(q, p, n, &nf_ct_frag6_hash[i], q.list) {
 143                         unsigned int hval = ip6qhashfn(q->id,
 144                                                        &q->saddr,
 145                                                        &q->daddr);
 146                         if (hval != i) {
 147                                 hlist_del(&q->q.list);
 148                                 /* Relink to new hash chain. */
 149                                 hlist_add_head(&q->q.list,
 150                                                &nf_ct_frag6_hash[hval]);
 151                         }
 152                 }
 153         }
 154         write_unlock(&nf_ct_frag6_lock);
 155
 156         mod_timer(&nf_ct_frag6_secret_timer, now + nf_ct_frag6_secret_interval);
 157 }
 158
 159 atomic_t nf_ct_frag6_mem = ATOMIC_INIT(0);
 160
 161 /* Memory Tracking Functions. */
 162 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
 163 {
 164         if (work)
 165                 *work -= skb->truesize;
 166         atomic_sub(skb->truesize, &nf_ct_frag6_mem);
 167         if (NFCT_FRAG6_CB(skb)->orig)
 168                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
 169
 170         kfree_skb(skb);
 171 }
 172
 173 static inline void frag_free_queue(struct nf_ct_frag6_queue *fq,
 174                                    unsigned int *work)
 175 {
 176         if (work)
 177                 *work -= sizeof(struct nf_ct_frag6_queue);
 178         atomic_sub(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
 179         kfree(fq);
 180 }
 181
 182 static inline struct nf_ct_frag6_queue *frag_alloc_queue(void)
 183 {
 184         struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
 185
 186         if (!fq)
 187                 return NULL;
 188         atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
 189         return fq;
 190 }
 191
 192 /* Destruction primitives. */
 193
 194 /* Complete destruction of fq. */
 195 static void nf_ct_frag6_destroy(struct nf_ct_frag6_queue *fq,
 196                                 unsigned int *work)
 197 {
 198         struct sk_buff *fp;
 199
 200         BUG_TRAP(fq->q.last_in&COMPLETE);
 201         BUG_TRAP(del_timer(&fq->q.timer) == 0);
 202
 203         /* Release all fragment data. */
 204         fp = fq->q.fragments;
 205         while (fp) {
 206                 struct sk_buff *xp = fp->next;
 207
 208                 frag_kfree_skb(fp, work);
 209                 fp = xp;
 210         }
 211
 212         frag_free_queue(fq, work);
 213 }
 214
 215 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq, unsigned int *work)
 216 {
 217         if (atomic_dec_and_test(&fq->q.refcnt))
 218                 nf_ct_frag6_destroy(fq, work);
 219 }
 220
 221 /* Kill fq entry. It is not destroyed immediately,
 222  * because caller (and someone more) holds reference count.
 223  */
 224 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
 225 {
 226         if (del_timer(&fq->q.timer))
 227                 atomic_dec(&fq->q.refcnt);
 228
 229         if (!(fq->q.last_in & COMPLETE)) {
 230                 fq_unlink(fq);
 231                 atomic_dec(&fq->q.refcnt);
 232                 fq->q.last_in |= COMPLETE;
 233         }
 234 }
 235
 236 static void nf_ct_frag6_evictor(void)
 237 {
 238         struct nf_ct_frag6_queue *fq;
 239         struct list_head *tmp;
 240         unsigned int work;
 241
 242         work = atomic_read(&nf_ct_frag6_mem);
 243         if (work <= nf_ct_frag6_low_thresh)
 244                 return;
 245
 246         work -= nf_ct_frag6_low_thresh;
 247         while (work > 0) {
 248                 read_lock(&nf_ct_frag6_lock);
 249                 if (list_empty(&nf_ct_frag6_lru_list)) {
 250                         read_unlock(&nf_ct_frag6_lock);
 251                         return;
 252                 }
 253                 tmp = nf_ct_frag6_lru_list.next;
 254                 BUG_ON(tmp == NULL);
 255                 fq = list_entry(tmp, struct nf_ct_frag6_queue, q.lru_list);
 256                 atomic_inc(&fq->q.refcnt);
 257                 read_unlock(&nf_ct_frag6_lock);
 258
 259                 spin_lock(&fq->q.lock);
 260                 if (!(fq->q.last_in&COMPLETE))
 261                         fq_kill(fq);
 262                 spin_unlock(&fq->q.lock);
 263
 264                 fq_put(fq, &work);
 265         }
 266 }
 267
 268 static void nf_ct_frag6_expire(unsigned long data)
 269 {
 270         struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
 271
 272         spin_lock(&fq->q.lock);
 273
 274         if (fq->q.last_in & COMPLETE)
 275                 goto out;
 276
 277         fq_kill(fq);
 278
 279 out:
 280         spin_unlock(&fq->q.lock);
 281         fq_put(fq, NULL);
 282 }
 283
 284 /* Creation primitives. */
 285
 286 static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
 287                                           struct nf_ct_frag6_queue *fq_in)
 288 {
 289         struct nf_ct_frag6_queue *fq;
 290 #ifdef CONFIG_SMP
 291         struct hlist_node *n;
 292 #endif
 293
 294         write_lock(&nf_ct_frag6_lock);
 295 #ifdef CONFIG_SMP
 296         hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], q.list) {
 297                 if (fq->id == fq_in->id &&
 298                     ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
 299                     ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
 300                         atomic_inc(&fq->q.refcnt);
 301                         write_unlock(&nf_ct_frag6_lock);
 302                         fq_in->q.last_in |= COMPLETE;
 303                         fq_put(fq_in, NULL);
 304                         return fq;
 305                 }
 306         }
 307 #endif
 308         fq = fq_in;
 309
 310         if (!mod_timer(&fq->q.timer, jiffies + nf_ct_frag6_timeout))
 311                 atomic_inc(&fq->q.refcnt);
 312
 313         atomic_inc(&fq->q.refcnt);
 314         hlist_add_head(&fq->q.list, &nf_ct_frag6_hash[hash]);
 315         INIT_LIST_HEAD(&fq->q.lru_list);
 316         list_add_tail(&fq->q.lru_list, &nf_ct_frag6_lru_list);
 317         nf_ct_frag6_nqueues++;
 318         write_unlock(&nf_ct_frag6_lock);
 319         return fq;
 320 }
 321
 322
 323 static struct nf_ct_frag6_queue *
 324 nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src,                             struct in6_addr *dst)
 325 {
 326         struct nf_ct_frag6_queue *fq;
 327
 328         if ((fq = frag_alloc_queue()) == NULL) {
 329                 pr_debug("Can't alloc new queue\n");
 330                 goto oom;
 331         }
 332
 333         memset(fq, 0, sizeof(struct nf_ct_frag6_queue));
 334
 335         fq->id = id;
 336         ipv6_addr_copy(&fq->saddr, src);
 337         ipv6_addr_copy(&fq->daddr, dst);
 338
 339         setup_timer(&fq->q.timer, nf_ct_frag6_expire, (unsigned long)fq);
 340         spin_lock_init(&fq->q.lock);
 341         atomic_set(&fq->q.refcnt, 1);
 342
 343         return nf_ct_frag6_intern(hash, fq);
 344
 345 oom:
 346         return NULL;
 347 }
 348
 349 static __inline__ struct nf_ct_frag6_queue *
 350 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
 351 {
 352         struct nf_ct_frag6_queue *fq;
 353         struct hlist_node *n;
 354         unsigned int hash = ip6qhashfn(id, src, dst);
 355
 356         read_lock(&nf_ct_frag6_lock);
 357         hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], q.list) {
 358                 if (fq->id == id &&
 359                     ipv6_addr_equal(src, &fq->saddr) &&
 360                     ipv6_addr_equal(dst, &fq->daddr)) {
 361                         atomic_inc(&fq->q.refcnt);
 362                         read_unlock(&nf_ct_frag6_lock);
 363                         return fq;
 364                 }
 365         }
 366         read_unlock(&nf_ct_frag6_lock);
 367
 368         return nf_ct_frag6_create(hash, id, src, dst);
 369 }
 370
 371
 372 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
 373                              struct frag_hdr *fhdr, int nhoff)
 374 {
 375         struct sk_buff *prev, *next;
 376         int offset, end;
 377
 378         if (fq->q.last_in & COMPLETE) {
 379                 pr_debug("Allready completed\n");
 380                 goto err;
 381         }
 382
 383         offset = ntohs(fhdr->frag_off) & ~0x7;
 384         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 385                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 386
 387         if ((unsigned int)end > IPV6_MAXPLEN) {
 388                 pr_debug("offset is too large.\n");
 389                 return -1;
 390         }
 391
 392         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 393                 const unsigned char *nh = skb_network_header(skb);
 394                 skb->csum = csum_sub(skb->csum,
 395                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 396                                                   0));
 397         }
 398
 399         /* Is this the final fragment? */
 400         if (!(fhdr->frag_off & htons(IP6_MF))) {
 401                 /* If we already have some bits beyond end
 402                  * or have different end, the segment is corrupted.
 403                  */
 404                 if (end < fq->q.len ||
 405                     ((fq->q.last_in & LAST_IN) && end != fq->q.len)) {
 406                         pr_debug("already received last fragment\n");
 407                         goto err;
 408                 }
 409                 fq->q.last_in |= LAST_IN;
 410                 fq->q.len = end;
 411         } else {
 412                 /* Check if the fragment is rounded to 8 bytes.
 413                  * Required by the RFC.
 414                  */
 415                 if (end & 0x7) {
 416                         /* RFC2460 says always send parameter problem in
 417                          * this case. -DaveM
 418                          */
 419                         pr_debug("end of fragment not rounded to 8 bytes.\n");
 420                         return -1;
 421                 }
 422                 if (end > fq->q.len) {
 423                         /* Some bits beyond end -> corruption. */
 424                         if (fq->q.last_in & LAST_IN) {
 425                                 pr_debug("last packet already reached.\n");
 426                                 goto err;
 427                         }
 428                         fq->q.len = end;
 429                 }
 430         }
 431
 432         if (end == offset)
 433                 goto err;
 434
 435         /* Point into the IP datagram 'data' part. */
 436         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
 437                 pr_debug("queue: message is too short.\n");
 438                 goto err;
 439         }
 440         if (pskb_trim_rcsum(skb, end - offset)) {
 441                 pr_debug("Can't trim\n");
 442                 goto err;
 443         }
 444
 445         /* Find out which fragments are in front and at the back of us
 446          * in the chain of fragments so far.  We must know where to put
 447          * this fragment, right?
 448          */
 449         prev = NULL;
 450         for (next = fq->q.fragments; next != NULL; next = next->next) {
 451                 if (NFCT_FRAG6_CB(next)->offset >= offset)
 452                         break;  /* bingo! */
 453                 prev = next;
 454         }
 455
 456         /* We found where to put this one.  Check for overlap with
 457          * preceding fragment, and, if needed, align things so that
 458          * any overlaps are eliminated.
 459          */
 460         if (prev) {
 461                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
 462
 463                 if (i > 0) {
 464                         offset += i;
 465                         if (end <= offset) {
 466                                 pr_debug("overlap\n");
 467                                 goto err;
 468                         }
 469                         if (!pskb_pull(skb, i)) {
 470                                 pr_debug("Can't pull\n");
 471                                 goto err;
 472                         }
 473                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 474                                 skb->ip_summed = CHECKSUM_NONE;
 475                 }
 476         }
 477
 478         /* Look for overlap with succeeding segments.
 479          * If we can merge fragments, do it.
 480          */
 481         while (next && NFCT_FRAG6_CB(next)->offset < end) {
 482                 /* overlap is 'i' bytes */
 483                 int i = end - NFCT_FRAG6_CB(next)->offset;
 484
 485                 if (i < next->len) {
 486                         /* Eat head of the next overlapped fragment
 487                          * and leave the loop. The next ones cannot overlap.
 488                          */
 489                         pr_debug("Eat head of the overlapped parts.: %d", i);
 490                         if (!pskb_pull(next, i))
 491                                 goto err;
 492
 493                         /* next fragment */
 494                         NFCT_FRAG6_CB(next)->offset += i;
 495                         fq->q.meat -= i;
 496                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 497                                 next->ip_summed = CHECKSUM_NONE;
 498                         break;
 499                 } else {
 500                         struct sk_buff *free_it = next;
 501
 502                         /* Old fragmnet is completely overridden with
 503                          * new one drop it.
 504                          */
 505                         next = next->next;
 506
 507                         if (prev)
 508                                 prev->next = next;
 509                         else
 510                                 fq->q.fragments = next;
 511
 512                         fq->q.meat -= free_it->len;
 513                         frag_kfree_skb(free_it, NULL);
 514                 }
 515         }
 516
 517         NFCT_FRAG6_CB(skb)->offset = offset;
 518
 519         /* Insert this fragment in the chain of fragments. */
 520         skb->next = next;
 521         if (prev)
 522                 prev->next = skb;
 523         else
 524                 fq->q.fragments = skb;
 525
 526         skb->dev = NULL;
 527         fq->q.stamp = skb->tstamp;
 528         fq->q.meat += skb->len;
 529         atomic_add(skb->truesize, &nf_ct_frag6_mem);
 530
 531         /* The first fragment.
 532          * nhoffset is obtained from the first fragment, of course.
 533          */
 534         if (offset == 0) {
 535                 fq->nhoffset = nhoff;
 536                 fq->q.last_in |= FIRST_IN;
 537         }
 538         write_lock(&nf_ct_frag6_lock);
 539         list_move_tail(&fq->q.lru_list, &nf_ct_frag6_lru_list);
 540         write_unlock(&nf_ct_frag6_lock);
 541         return 0;
 542
 543 err:
 544         return -1;
 545 }
 546
 547 /*
 548  *      Check if this packet is complete.
 549  *      Returns NULL on failure by any reason, and pointer
 550  *      to current nexthdr field in reassembled frame.
 551  *
 552  *      It is called with locked fq, and caller must check that
 553  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 554  *      the last and the first frames arrived and all the bits are here.
 555  */
 556 static struct sk_buff *
 557 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
 558 {
 559         struct sk_buff *fp, *op, *head = fq->q.fragments;
 560         int    payload_len;
 561
 562         fq_kill(fq);
 563
 564         BUG_TRAP(head != NULL);
 565         BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
 566
 567         /* Unfragmented part is taken from the first segment. */
 568         payload_len = ((head->data - skb_network_header(head)) -
 569                        sizeof(struct ipv6hdr) + fq->q.len -
 570                        sizeof(struct frag_hdr));
 571         if (payload_len > IPV6_MAXPLEN) {
 572                 pr_debug("payload len is too large.\n");
 573                 goto out_oversize;
 574         }
 575
 576         /* Head of list must not be cloned. */
 577         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
 578                 pr_debug("skb is cloned but can't expand head");
 579                 goto out_oom;
 580         }
 581
 582         /* If the first fragment is fragmented itself, we split
 583          * it to two chunks: the first with data and paged part
 584          * and the second, holding only fragments. */
 585         if (skb_shinfo(head)->frag_list) {
 586                 struct sk_buff *clone;
 587                 int i, plen = 0;
 588
 589                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
 590                         pr_debug("Can't alloc skb\n");
 591                         goto out_oom;
 592                 }
 593                 clone->next = head->next;
 594                 head->next = clone;
 595                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 596                 skb_shinfo(head)->frag_list = NULL;
 597                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 598                         plen += skb_shinfo(head)->frags[i].size;
 599                 clone->len = clone->data_len = head->data_len - plen;
 600                 head->data_len -= clone->len;
 601                 head->len -= clone->len;
 602                 clone->csum = 0;
 603                 clone->ip_summed = head->ip_summed;
 604
 605                 NFCT_FRAG6_CB(clone)->orig = NULL;
 606                 atomic_add(clone->truesize, &nf_ct_frag6_mem);
 607         }
 608
 609         /* We have to remove fragment header from datagram and to relocate
 610          * header in order to calculate ICV correctly. */
 611         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
 612         memmove(head->head + sizeof(struct frag_hdr), head->head,
 613                 (head->data - head->head) - sizeof(struct frag_hdr));
 614         head->mac_header += sizeof(struct frag_hdr);
 615         head->network_header += sizeof(struct frag_hdr);
 616
 617         skb_shinfo(head)->frag_list = head->next;
 618         skb_reset_transport_header(head);
 619         skb_push(head, head->data - skb_network_header(head));
 620         atomic_sub(head->truesize, &nf_ct_frag6_mem);
 621
 622         for (fp=head->next; fp; fp = fp->next) {
 623                 head->data_len += fp->len;
 624                 head->len += fp->len;
 625                 if (head->ip_summed != fp->ip_summed)
 626                         head->ip_summed = CHECKSUM_NONE;
 627                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 628                         head->csum = csum_add(head->csum, fp->csum);
 629                 head->truesize += fp->truesize;
 630                 atomic_sub(fp->truesize, &nf_ct_frag6_mem);
 631         }
 632
 633         head->next = NULL;
 634         head->dev = dev;
 635         head->tstamp = fq->q.stamp;
 636         ipv6_hdr(head)->payload_len = htons(payload_len);
 637
 638         /* Yes, and fold redundant checksum back. 8) */
 639         if (head->ip_summed == CHECKSUM_COMPLETE)
 640                 head->csum = csum_partial(skb_network_header(head),
 641                                           skb_network_header_len(head),
 642                                           head->csum);
 643
 644         fq->q.fragments = NULL;
 645
 646         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
 647         fp = skb_shinfo(head)->frag_list;
 648         if (NFCT_FRAG6_CB(fp)->orig == NULL)
 649                 /* at above code, head skb is divided into two skbs. */
 650                 fp = fp->next;
 651
 652         op = NFCT_FRAG6_CB(head)->orig;
 653         for (; fp; fp = fp->next) {
 654                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
 655
 656                 op->next = orig;
 657                 op = orig;
 658                 NFCT_FRAG6_CB(fp)->orig = NULL;
 659         }
 660
 661         return head;
 662
 663 out_oversize:
 664         if (net_ratelimit())
 665                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
 666         goto out_fail;
 667 out_oom:
 668         if (net_ratelimit())
 669                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
 670 out_fail:
 671         return NULL;
 672 }
 673
 674 /*
 675  * find the header just before Fragment Header.
 676  *
 677  * if success return 0 and set ...
 678  * (*prevhdrp): the value of "Next Header Field" in the header
 679  *              just before Fragment Header.
 680  * (*prevhoff): the offset of "Next Header Field" in the header
 681  *              just before Fragment Header.
 682  * (*fhoff)   : the offset of Fragment Header.
 683  *
 684  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
 685  *
 686  */
 687 static int
 688 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
 689 {
 690         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 691         const int netoff = skb_network_offset(skb);
 692         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
 693         int start = netoff + sizeof(struct ipv6hdr);
 694         int len = skb->len - start;
 695         u8 prevhdr = NEXTHDR_IPV6;
 696
 697         while (nexthdr != NEXTHDR_FRAGMENT) {
 698                 struct ipv6_opt_hdr hdr;
 699                 int hdrlen;
 700
 701                 if (!ipv6_ext_hdr(nexthdr)) {
 702                         return -1;
 703                 }
 704                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
 705                         pr_debug("too short\n");
 706                         return -1;
 707                 }
 708                 if (nexthdr == NEXTHDR_NONE) {
 709                         pr_debug("next header is none\n");
 710                         return -1;
 711                 }
 712                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
 713                         BUG();
 714                 if (nexthdr == NEXTHDR_AUTH)
 715                         hdrlen = (hdr.hdrlen+2)<<2;
 716                 else
 717                         hdrlen = ipv6_optlen(&hdr);
 718
 719                 prevhdr = nexthdr;
 720                 prev_nhoff = start;
 721
 722                 nexthdr = hdr.nexthdr;
 723                 len -= hdrlen;
 724                 start += hdrlen;
 725         }
 726
 727         if (len < 0)
 728                 return -1;
 729
 730         *prevhdrp = prevhdr;
 731         *prevhoff = prev_nhoff;
 732         *fhoff = start;
 733
 734         return 0;
 735 }
 736
 737 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
 738 {
 739         struct sk_buff *clone;
 740         struct net_device *dev = skb->dev;
 741         struct frag_hdr *fhdr;
 742         struct nf_ct_frag6_queue *fq;
 743         struct ipv6hdr *hdr;
 744         int fhoff, nhoff;
 745         u8 prevhdr;
 746         struct sk_buff *ret_skb = NULL;
 747
 748         /* Jumbo payload inhibits frag. header */
 749         if (ipv6_hdr(skb)->payload_len == 0) {
 750                 pr_debug("payload len = 0\n");
 751                 return skb;
 752         }
 753
 754         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
 755                 return skb;
 756
 757         clone = skb_clone(skb, GFP_ATOMIC);
 758         if (clone == NULL) {
 759                 pr_debug("Can't clone skb\n");
 760                 return skb;
 761         }
 762
 763         NFCT_FRAG6_CB(clone)->orig = skb;
 764
 765         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
 766                 pr_debug("message is too short.\n");
 767                 goto ret_orig;
 768         }
 769
 770         skb_set_transport_header(clone, fhoff);
 771         hdr = ipv6_hdr(clone);
 772         fhdr = (struct frag_hdr *)skb_transport_header(clone);
 773
 774         if (!(fhdr->frag_off & htons(0xFFF9))) {
 775                 pr_debug("Invalid fragment offset\n");
 776                 /* It is not a fragmented frame */
 777                 goto ret_orig;
 778         }
 779
 780         if (atomic_read(&nf_ct_frag6_mem) > nf_ct_frag6_high_thresh)
 781                 nf_ct_frag6_evictor();
 782
 783         fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
 784         if (fq == NULL) {
 785                 pr_debug("Can't find and can't create new queue\n");
 786                 goto ret_orig;
 787         }
 788
 789         spin_lock(&fq->q.lock);
 790
 791         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
 792                 spin_unlock(&fq->q.lock);
 793                 pr_debug("Can't insert skb to queue\n");
 794                 fq_put(fq, NULL);
 795                 goto ret_orig;
 796         }
 797
 798         if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) {
 799                 ret_skb = nf_ct_frag6_reasm(fq, dev);
 800                 if (ret_skb == NULL)
 801                         pr_debug("Can't reassemble fragmented packets\n");
 802         }
 803         spin_unlock(&fq->q.lock);
 804
 805         fq_put(fq, NULL);
 806         return ret_skb;
 807
 808 ret_orig:
 809         kfree_skb(clone);
 810         return skb;
 811 }
 812
 813 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
 814                         struct net_device *in, struct net_device *out,
 815                         int (*okfn)(struct sk_buff *))
 816 {
 817         struct sk_buff *s, *s2;
 818
 819         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
 820                 nf_conntrack_put_reasm(s->nfct_reasm);
 821                 nf_conntrack_get_reasm(skb);
 822                 s->nfct_reasm = skb;
 823
 824                 s2 = s->next;
 825                 s->next = NULL;
 826
 827                 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
 828                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
 829                 s = s2;
 830         }
 831         nf_conntrack_put_reasm(skb);
 832 }
 833
 834 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
 835 {
 836         struct sk_buff *s, *s2;
 837
 838         for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
 839
 840                 s2 = s->next;
 841                 kfree_skb(s);
 842         }
 843
 844         kfree_skb(skb);
 845
 846         return 0;
 847 }
 848
 849 int nf_ct_frag6_init(void)
 850 {
 851         nf_ct_frag6_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
 852                                    (jiffies ^ (jiffies >> 6)));
 853
 854         setup_timer(&nf_ct_frag6_secret_timer, nf_ct_frag6_secret_rebuild, 0);
 855         nf_ct_frag6_secret_timer.expires = jiffies
 856                                            + nf_ct_frag6_secret_interval;
 857         add_timer(&nf_ct_frag6_secret_timer);
 858
 859         return 0;
 860 }
 861
 862 void nf_ct_frag6_cleanup(void)
 863 {
 864         del_timer(&nf_ct_frag6_secret_timer);
 865         nf_ct_frag6_low_thresh = 0;
 866         nf_ct_frag6_evictor();
 867 }