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
  35 #include <net/ipv6.h>
  36 #include <net/protocol.h>
  37 #include <net/transp_v6.h>
  38 #include <net/rawv6.h>
  39 #include <net/ndisc.h>
  40 #include <net/addrconf.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/netfilter.h>
  43 #include <linux/netfilter_ipv6.h>
  44 #include <linux/kernel.h>
  45 #include <linux/module.h>
  46
  47 #if 0
  48 #define DEBUGP printk
  49 #else
  50 #define DEBUGP(format, args...)
  51 #endif
  52
  53 #define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */
  54 #define NF_CT_FRAG6_LOW_THRESH 196608  /* == 192*1024 */
  55 #define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT
  56
  57 unsigned int nf_ct_frag6_high_thresh __read_mostly = 256*1024;
  58 unsigned int nf_ct_frag6_low_thresh __read_mostly = 192*1024;
  59 unsigned long nf_ct_frag6_timeout __read_mostly = IPV6_FRAG_TIMEOUT;
  60
  61 struct nf_ct_frag6_skb_cb
  62 {
  63         struct inet6_skb_parm   h;
  64         int                     offset;
  65         struct sk_buff          *orig;
  66 };
  67
  68 #define NFCT_FRAG6_CB(skb)      ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
  69
  70 struct nf_ct_frag6_queue
  71 {
  72         struct hlist_node       list;
  73         struct list_head        lru_list;       /* lru list member      */
  74
  75         __be32                  id;             /* fragment id          */
  76         struct in6_addr         saddr;
  77         struct in6_addr         daddr;
  78
  79         spinlock_t              lock;
  80         atomic_t                refcnt;
  81         struct timer_list       timer;          /* expire timer         */
  82         struct sk_buff          *fragments;
  83         int                     len;
  84         int                     meat;
  85         ktime_t                 stamp;
  86         unsigned int            csum;
  87         __u8                    last_in;        /* has first/last segment arrived? */
  88 #define COMPLETE                4
  89 #define FIRST_IN                2
  90 #define LAST_IN                 1
  91         __u16                   nhoffset;
  92 };
  93
  94 /* Hash table. */
  95
  96 #define FRAG6Q_HASHSZ   64
  97
  98 static struct hlist_head nf_ct_frag6_hash[FRAG6Q_HASHSZ];
  99 static DEFINE_RWLOCK(nf_ct_frag6_lock);
 100 static u32 nf_ct_frag6_hash_rnd;
 101 static LIST_HEAD(nf_ct_frag6_lru_list);
 102 int nf_ct_frag6_nqueues = 0;
 103
 104 static __inline__ void __fq_unlink(struct nf_ct_frag6_queue *fq)
 105 {
 106         hlist_del(&fq->list);
 107         list_del(&fq->lru_list);
 108         nf_ct_frag6_nqueues--;
 109 }
 110
 111 static __inline__ void fq_unlink(struct nf_ct_frag6_queue *fq)
 112 {
 113         write_lock(&nf_ct_frag6_lock);
 114         __fq_unlink(fq);
 115         write_unlock(&nf_ct_frag6_lock);
 116 }
 117
 118 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
 119                                struct in6_addr *daddr)
 120 {
 121         u32 a, b, c;
 122
 123         a = (__force u32)saddr->s6_addr32[0];
 124         b = (__force u32)saddr->s6_addr32[1];
 125         c = (__force u32)saddr->s6_addr32[2];
 126
 127         a += JHASH_GOLDEN_RATIO;
 128         b += JHASH_GOLDEN_RATIO;
 129         c += nf_ct_frag6_hash_rnd;
 130         __jhash_mix(a, b, c);
 131
 132         a += (__force u32)saddr->s6_addr32[3];
 133         b += (__force u32)daddr->s6_addr32[0];
 134         c += (__force u32)daddr->s6_addr32[1];
 135         __jhash_mix(a, b, c);
 136
 137         a += (__force u32)daddr->s6_addr32[2];
 138         b += (__force u32)daddr->s6_addr32[3];
 139         c += (__force u32)id;
 140         __jhash_mix(a, b, c);
 141
 142         return c & (FRAG6Q_HASHSZ - 1);
 143 }
 144
 145 static struct timer_list nf_ct_frag6_secret_timer;
 146 int nf_ct_frag6_secret_interval = 10 * 60 * HZ;
 147
 148 static void nf_ct_frag6_secret_rebuild(unsigned long dummy)
 149 {
 150         unsigned long now = jiffies;
 151         int i;
 152
 153         write_lock(&nf_ct_frag6_lock);
 154         get_random_bytes(&nf_ct_frag6_hash_rnd, sizeof(u32));
 155         for (i = 0; i < FRAG6Q_HASHSZ; i++) {
 156                 struct nf_ct_frag6_queue *q;
 157                 struct hlist_node *p, *n;
 158
 159                 hlist_for_each_entry_safe(q, p, n, &nf_ct_frag6_hash[i], list) {
 160                         unsigned int hval = ip6qhashfn(q->id,
 161                                                        &q->saddr,
 162                                                        &q->daddr);
 163                         if (hval != i) {
 164                                 hlist_del(&q->list);
 165                                 /* Relink to new hash chain. */
 166                                 hlist_add_head(&q->list,
 167                                                &nf_ct_frag6_hash[hval]);
 168                         }
 169                 }
 170         }
 171         write_unlock(&nf_ct_frag6_lock);
 172
 173         mod_timer(&nf_ct_frag6_secret_timer, now + nf_ct_frag6_secret_interval);
 174 }
 175
 176 atomic_t nf_ct_frag6_mem = ATOMIC_INIT(0);
 177
 178 /* Memory Tracking Functions. */
 179 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
 180 {
 181         if (work)
 182                 *work -= skb->truesize;
 183         atomic_sub(skb->truesize, &nf_ct_frag6_mem);
 184         if (NFCT_FRAG6_CB(skb)->orig)
 185                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
 186
 187         kfree_skb(skb);
 188 }
 189
 190 static inline void frag_free_queue(struct nf_ct_frag6_queue *fq,
 191                                    unsigned int *work)
 192 {
 193         if (work)
 194                 *work -= sizeof(struct nf_ct_frag6_queue);
 195         atomic_sub(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
 196         kfree(fq);
 197 }
 198
 199 static inline struct nf_ct_frag6_queue *frag_alloc_queue(void)
 200 {
 201         struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
 202
 203         if (!fq)
 204                 return NULL;
 205         atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
 206         return fq;
 207 }
 208
 209 /* Destruction primitives. */
 210
 211 /* Complete destruction of fq. */
 212 static void nf_ct_frag6_destroy(struct nf_ct_frag6_queue *fq,
 213                                 unsigned int *work)
 214 {
 215         struct sk_buff *fp;
 216
 217         BUG_TRAP(fq->last_in&COMPLETE);
 218         BUG_TRAP(del_timer(&fq->timer) == 0);
 219
 220         /* Release all fragment data. */
 221         fp = fq->fragments;
 222         while (fp) {
 223                 struct sk_buff *xp = fp->next;
 224
 225                 frag_kfree_skb(fp, work);
 226                 fp = xp;
 227         }
 228
 229         frag_free_queue(fq, work);
 230 }
 231
 232 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq, unsigned int *work)
 233 {
 234         if (atomic_dec_and_test(&fq->refcnt))
 235                 nf_ct_frag6_destroy(fq, work);
 236 }
 237
 238 /* Kill fq entry. It is not destroyed immediately,
 239  * because caller (and someone more) holds reference count.
 240  */
 241 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
 242 {
 243         if (del_timer(&fq->timer))
 244                 atomic_dec(&fq->refcnt);
 245
 246         if (!(fq->last_in & COMPLETE)) {
 247                 fq_unlink(fq);
 248                 atomic_dec(&fq->refcnt);
 249                 fq->last_in |= COMPLETE;
 250         }
 251 }
 252
 253 static void nf_ct_frag6_evictor(void)
 254 {
 255         struct nf_ct_frag6_queue *fq;
 256         struct list_head *tmp;
 257         unsigned int work;
 258
 259         work = atomic_read(&nf_ct_frag6_mem);
 260         if (work <= nf_ct_frag6_low_thresh)
 261                 return;
 262
 263         work -= nf_ct_frag6_low_thresh;
 264         while (work > 0) {
 265                 read_lock(&nf_ct_frag6_lock);
 266                 if (list_empty(&nf_ct_frag6_lru_list)) {
 267                         read_unlock(&nf_ct_frag6_lock);
 268                         return;
 269                 }
 270                 tmp = nf_ct_frag6_lru_list.next;
 271                 BUG_ON(tmp == NULL);
 272                 fq = list_entry(tmp, struct nf_ct_frag6_queue, lru_list);
 273                 atomic_inc(&fq->refcnt);
 274                 read_unlock(&nf_ct_frag6_lock);
 275
 276                 spin_lock(&fq->lock);
 277                 if (!(fq->last_in&COMPLETE))
 278                         fq_kill(fq);
 279                 spin_unlock(&fq->lock);
 280
 281                 fq_put(fq, &work);
 282         }
 283 }
 284
 285 static void nf_ct_frag6_expire(unsigned long data)
 286 {
 287         struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
 288
 289         spin_lock(&fq->lock);
 290
 291         if (fq->last_in & COMPLETE)
 292                 goto out;
 293
 294         fq_kill(fq);
 295
 296 out:
 297         spin_unlock(&fq->lock);
 298         fq_put(fq, NULL);
 299 }
 300
 301 /* Creation primitives. */
 302
 303 static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
 304                                           struct nf_ct_frag6_queue *fq_in)
 305 {
 306         struct nf_ct_frag6_queue *fq;
 307 #ifdef CONFIG_SMP
 308         struct hlist_node *n;
 309 #endif
 310
 311         write_lock(&nf_ct_frag6_lock);
 312 #ifdef CONFIG_SMP
 313         hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], list) {
 314                 if (fq->id == fq_in->id &&
 315                     ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
 316                     ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
 317                         atomic_inc(&fq->refcnt);
 318                         write_unlock(&nf_ct_frag6_lock);
 319                         fq_in->last_in |= COMPLETE;
 320                         fq_put(fq_in, NULL);
 321                         return fq;
 322                 }
 323         }
 324 #endif
 325         fq = fq_in;
 326
 327         if (!mod_timer(&fq->timer, jiffies + nf_ct_frag6_timeout))
 328                 atomic_inc(&fq->refcnt);
 329
 330         atomic_inc(&fq->refcnt);
 331         hlist_add_head(&fq->list, &nf_ct_frag6_hash[hash]);
 332         INIT_LIST_HEAD(&fq->lru_list);
 333         list_add_tail(&fq->lru_list, &nf_ct_frag6_lru_list);
 334         nf_ct_frag6_nqueues++;
 335         write_unlock(&nf_ct_frag6_lock);
 336         return fq;
 337 }
 338
 339
 340 static struct nf_ct_frag6_queue *
 341 nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src,                             struct in6_addr *dst)
 342 {
 343         struct nf_ct_frag6_queue *fq;
 344
 345         if ((fq = frag_alloc_queue()) == NULL) {
 346                 DEBUGP("Can't alloc new queue\n");
 347                 goto oom;
 348         }
 349
 350         memset(fq, 0, sizeof(struct nf_ct_frag6_queue));
 351
 352         fq->id = id;
 353         ipv6_addr_copy(&fq->saddr, src);
 354         ipv6_addr_copy(&fq->daddr, dst);
 355
 356         setup_timer(&fq->timer, nf_ct_frag6_expire, (unsigned long)fq);
 357         spin_lock_init(&fq->lock);
 358         atomic_set(&fq->refcnt, 1);
 359
 360         return nf_ct_frag6_intern(hash, fq);
 361
 362 oom:
 363         return NULL;
 364 }
 365
 366 static __inline__ struct nf_ct_frag6_queue *
 367 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
 368 {
 369         struct nf_ct_frag6_queue *fq;
 370         struct hlist_node *n;
 371         unsigned int hash = ip6qhashfn(id, src, dst);
 372
 373         read_lock(&nf_ct_frag6_lock);
 374         hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], list) {
 375                 if (fq->id == id &&
 376                     ipv6_addr_equal(src, &fq->saddr) &&
 377                     ipv6_addr_equal(dst, &fq->daddr)) {
 378                         atomic_inc(&fq->refcnt);
 379                         read_unlock(&nf_ct_frag6_lock);
 380                         return fq;
 381                 }
 382         }
 383         read_unlock(&nf_ct_frag6_lock);
 384
 385         return nf_ct_frag6_create(hash, id, src, dst);
 386 }
 387
 388
 389 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
 390                              struct frag_hdr *fhdr, int nhoff)
 391 {
 392         struct sk_buff *prev, *next;
 393         int offset, end;
 394
 395         if (fq->last_in & COMPLETE) {
 396                 DEBUGP("Allready completed\n");
 397                 goto err;
 398         }
 399
 400         offset = ntohs(fhdr->frag_off) & ~0x7;
 401         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
 402                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 403
 404         if ((unsigned int)end > IPV6_MAXPLEN) {
 405                 DEBUGP("offset is too large.\n");
 406                 return -1;
 407         }
 408
 409         if (skb->ip_summed == CHECKSUM_COMPLETE) {
 410                 const unsigned char *nh = skb_network_header(skb);
 411                 skb->csum = csum_sub(skb->csum,
 412                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
 413                                                   0));
 414         }
 415
 416         /* Is this the final fragment? */
 417         if (!(fhdr->frag_off & htons(IP6_MF))) {
 418                 /* If we already have some bits beyond end
 419                  * or have different end, the segment is corrupted.
 420                  */
 421                 if (end < fq->len ||
 422                     ((fq->last_in & LAST_IN) && end != fq->len)) {
 423                         DEBUGP("already received last fragment\n");
 424                         goto err;
 425                 }
 426                 fq->last_in |= LAST_IN;
 427                 fq->len = end;
 428         } else {
 429                 /* Check if the fragment is rounded to 8 bytes.
 430                  * Required by the RFC.
 431                  */
 432                 if (end & 0x7) {
 433                         /* RFC2460 says always send parameter problem in
 434                          * this case. -DaveM
 435                          */
 436                         DEBUGP("the end of this fragment is not rounded to 8 bytes.\n");
 437                         return -1;
 438                 }
 439                 if (end > fq->len) {
 440                         /* Some bits beyond end -> corruption. */
 441                         if (fq->last_in & LAST_IN) {
 442                                 DEBUGP("last packet already reached.\n");
 443                                 goto err;
 444                         }
 445                         fq->len = end;
 446                 }
 447         }
 448
 449         if (end == offset)
 450                 goto err;
 451
 452         /* Point into the IP datagram 'data' part. */
 453         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
 454                 DEBUGP("queue: message is too short.\n");
 455                 goto err;
 456         }
 457         if (pskb_trim_rcsum(skb, end - offset)) {
 458                 DEBUGP("Can't trim\n");
 459                 goto err;
 460         }
 461
 462         /* Find out which fragments are in front and at the back of us
 463          * in the chain of fragments so far.  We must know where to put
 464          * this fragment, right?
 465          */
 466         prev = NULL;
 467         for (next = fq->fragments; next != NULL; next = next->next) {
 468                 if (NFCT_FRAG6_CB(next)->offset >= offset)
 469                         break;  /* bingo! */
 470                 prev = next;
 471         }
 472
 473         /* We found where to put this one.  Check for overlap with
 474          * preceding fragment, and, if needed, align things so that
 475          * any overlaps are eliminated.
 476          */
 477         if (prev) {
 478                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
 479
 480                 if (i > 0) {
 481                         offset += i;
 482                         if (end <= offset) {
 483                                 DEBUGP("overlap\n");
 484                                 goto err;
 485                         }
 486                         if (!pskb_pull(skb, i)) {
 487                                 DEBUGP("Can't pull\n");
 488                                 goto err;
 489                         }
 490                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 491                                 skb->ip_summed = CHECKSUM_NONE;
 492                 }
 493         }
 494
 495         /* Look for overlap with succeeding segments.
 496          * If we can merge fragments, do it.
 497          */
 498         while (next && NFCT_FRAG6_CB(next)->offset < end) {
 499                 /* overlap is 'i' bytes */
 500                 int i = end - NFCT_FRAG6_CB(next)->offset;
 501
 502                 if (i < next->len) {
 503                         /* Eat head of the next overlapped fragment
 504                          * and leave the loop. The next ones cannot overlap.
 505                          */
 506                         DEBUGP("Eat head of the overlapped parts.: %d", i);
 507                         if (!pskb_pull(next, i))
 508                                 goto err;
 509
 510                         /* next fragment */
 511                         NFCT_FRAG6_CB(next)->offset += i;
 512                         fq->meat -= i;
 513                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 514                                 next->ip_summed = CHECKSUM_NONE;
 515                         break;
 516                 } else {
 517                         struct sk_buff *free_it = next;
 518
 519                         /* Old fragmnet is completely overridden with
 520                          * new one drop it.
 521                          */
 522                         next = next->next;
 523
 524                         if (prev)
 525                                 prev->next = next;
 526                         else
 527                                 fq->fragments = next;
 528
 529                         fq->meat -= free_it->len;
 530                         frag_kfree_skb(free_it, NULL);
 531                 }
 532         }
 533
 534         NFCT_FRAG6_CB(skb)->offset = offset;
 535
 536         /* Insert this fragment in the chain of fragments. */
 537         skb->next = next;
 538         if (prev)
 539                 prev->next = skb;
 540         else
 541                 fq->fragments = skb;
 542
 543         skb->dev = NULL;
 544         fq->stamp = skb->tstamp;
 545         fq->meat += skb->len;
 546         atomic_add(skb->truesize, &nf_ct_frag6_mem);
 547
 548         /* The first fragment.
 549          * nhoffset is obtained from the first fragment, of course.
 550          */
 551         if (offset == 0) {
 552                 fq->nhoffset = nhoff;
 553                 fq->last_in |= FIRST_IN;
 554         }
 555         write_lock(&nf_ct_frag6_lock);
 556         list_move_tail(&fq->lru_list, &nf_ct_frag6_lru_list);
 557         write_unlock(&nf_ct_frag6_lock);
 558         return 0;
 559
 560 err:
 561         return -1;
 562 }
 563
 564 /*
 565  *      Check if this packet is complete.
 566  *      Returns NULL on failure by any reason, and pointer
 567  *      to current nexthdr field in reassembled frame.
 568  *
 569  *      It is called with locked fq, and caller must check that
 570  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 571  *      the last and the first frames arrived and all the bits are here.
 572  */
 573 static struct sk_buff *
 574 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
 575 {
 576         struct sk_buff *fp, *op, *head = fq->fragments;
 577         int    payload_len;
 578
 579         fq_kill(fq);
 580
 581         BUG_TRAP(head != NULL);
 582         BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
 583
 584         /* Unfragmented part is taken from the first segment. */
 585         payload_len = ((head->data - skb_network_header(head)) -
 586                        sizeof(struct ipv6hdr) + fq->len -
 587                        sizeof(struct frag_hdr));
 588         if (payload_len > IPV6_MAXPLEN) {
 589                 DEBUGP("payload len is too large.\n");
 590                 goto out_oversize;
 591         }
 592
 593         /* Head of list must not be cloned. */
 594         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
 595                 DEBUGP("skb is cloned but can't expand head");
 596                 goto out_oom;
 597         }
 598
 599         /* If the first fragment is fragmented itself, we split
 600          * it to two chunks: the first with data and paged part
 601          * and the second, holding only fragments. */
 602         if (skb_shinfo(head)->frag_list) {
 603                 struct sk_buff *clone;
 604                 int i, plen = 0;
 605
 606                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
 607                         DEBUGP("Can't alloc skb\n");
 608                         goto out_oom;
 609                 }
 610                 clone->next = head->next;
 611                 head->next = clone;
 612                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 613                 skb_shinfo(head)->frag_list = NULL;
 614                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 615                         plen += skb_shinfo(head)->frags[i].size;
 616                 clone->len = clone->data_len = head->data_len - plen;
 617                 head->data_len -= clone->len;
 618                 head->len -= clone->len;
 619                 clone->csum = 0;
 620                 clone->ip_summed = head->ip_summed;
 621
 622                 NFCT_FRAG6_CB(clone)->orig = NULL;
 623                 atomic_add(clone->truesize, &nf_ct_frag6_mem);
 624         }
 625
 626         /* We have to remove fragment header from datagram and to relocate
 627          * header in order to calculate ICV correctly. */
 628         skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
 629         memmove(head->head + sizeof(struct frag_hdr), head->head,
 630                 (head->data - head->head) - sizeof(struct frag_hdr));
 631         head->mac_header += sizeof(struct frag_hdr);
 632         head->network_header += sizeof(struct frag_hdr);
 633
 634         skb_shinfo(head)->frag_list = head->next;
 635         skb_reset_transport_header(head);
 636         skb_push(head, head->data - skb_network_header(head));
 637         atomic_sub(head->truesize, &nf_ct_frag6_mem);
 638
 639         for (fp=head->next; fp; fp = fp->next) {
 640                 head->data_len += fp->len;
 641                 head->len += fp->len;
 642                 if (head->ip_summed != fp->ip_summed)
 643                         head->ip_summed = CHECKSUM_NONE;
 644                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 645                         head->csum = csum_add(head->csum, fp->csum);
 646                 head->truesize += fp->truesize;
 647                 atomic_sub(fp->truesize, &nf_ct_frag6_mem);
 648         }
 649
 650         head->next = NULL;
 651         head->dev = dev;
 652         head->tstamp = fq->stamp;
 653         ipv6_hdr(head)->payload_len = htons(payload_len);
 654
 655         /* Yes, and fold redundant checksum back. 8) */
 656         if (head->ip_summed == CHECKSUM_COMPLETE)
 657                 head->csum = csum_partial(skb_network_header(head),
 658                                           skb_network_header_len(head),
 659                                           head->csum);
 660
 661         fq->fragments = NULL;
 662
 663         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
 664         fp = skb_shinfo(head)->frag_list;
 665         if (NFCT_FRAG6_CB(fp)->orig == NULL)
 666                 /* at above code, head skb is divided into two skbs. */
 667                 fp = fp->next;
 668
 669         op = NFCT_FRAG6_CB(head)->orig;
 670         for (; fp; fp = fp->next) {
 671                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
 672
 673                 op->next = orig;
 674                 op = orig;
 675                 NFCT_FRAG6_CB(fp)->orig = NULL;
 676         }
 677
 678         return head;
 679
 680 out_oversize:
 681         if (net_ratelimit())
 682                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
 683         goto out_fail;
 684 out_oom:
 685         if (net_ratelimit())
 686                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
 687 out_fail:
 688         return NULL;
 689 }
 690
 691 /*
 692  * find the header just before Fragment Header.
 693  *
 694  * if success return 0 and set ...
 695  * (*prevhdrp): the value of "Next Header Field" in the header
 696  *              just before Fragment Header.
 697  * (*prevhoff): the offset of "Next Header Field" in the header
 698  *              just before Fragment Header.
 699  * (*fhoff)   : the offset of Fragment Header.
 700  *
 701  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
 702  *
 703  */
 704 static int
 705 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
 706 {
 707         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 708         const int netoff = skb_network_offset(skb);
 709         u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
 710         int start = netoff + sizeof(struct ipv6hdr);
 711         int len = skb->len - start;
 712         u8 prevhdr = NEXTHDR_IPV6;
 713
 714         while (nexthdr != NEXTHDR_FRAGMENT) {
 715                 struct ipv6_opt_hdr hdr;
 716                 int hdrlen;
 717
 718                 if (!ipv6_ext_hdr(nexthdr)) {
 719                         return -1;
 720                 }
 721                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
 722                         DEBUGP("too short\n");
 723                         return -1;
 724                 }
 725                 if (nexthdr == NEXTHDR_NONE) {
 726                         DEBUGP("next header is none\n");
 727                         return -1;
 728                 }
 729                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
 730                         BUG();
 731                 if (nexthdr == NEXTHDR_AUTH)
 732                         hdrlen = (hdr.hdrlen+2)<<2;
 733                 else
 734                         hdrlen = ipv6_optlen(&hdr);
 735
 736                 prevhdr = nexthdr;
 737                 prev_nhoff = start;
 738
 739                 nexthdr = hdr.nexthdr;
 740                 len -= hdrlen;
 741                 start += hdrlen;
 742         }
 743
 744         if (len < 0)
 745                 return -1;
 746
 747         *prevhdrp = prevhdr;
 748         *prevhoff = prev_nhoff;
 749         *fhoff = start;
 750
 751         return 0;
 752 }
 753
 754 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
 755 {
 756         struct sk_buff *clone;
 757         struct net_device *dev = skb->dev;
 758         struct frag_hdr *fhdr;
 759         struct nf_ct_frag6_queue *fq;
 760         struct ipv6hdr *hdr;
 761         int fhoff, nhoff;
 762         u8 prevhdr;
 763         struct sk_buff *ret_skb = NULL;
 764
 765         /* Jumbo payload inhibits frag. header */
 766         if (ipv6_hdr(skb)->payload_len == 0) {
 767                 DEBUGP("payload len = 0\n");
 768                 return skb;
 769         }
 770
 771         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
 772                 return skb;
 773
 774         clone = skb_clone(skb, GFP_ATOMIC);
 775         if (clone == NULL) {
 776                 DEBUGP("Can't clone skb\n");
 777                 return skb;
 778         }
 779
 780         NFCT_FRAG6_CB(clone)->orig = skb;
 781
 782         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
 783                 DEBUGP("message is too short.\n");
 784                 goto ret_orig;
 785         }
 786
 787         skb_set_transport_header(clone, fhoff);
 788         hdr = ipv6_hdr(clone);
 789         fhdr = (struct frag_hdr *)skb_transport_header(clone);
 790
 791         if (!(fhdr->frag_off & htons(0xFFF9))) {
 792                 DEBUGP("Invalid fragment offset\n");
 793                 /* It is not a fragmented frame */
 794                 goto ret_orig;
 795         }
 796
 797         if (atomic_read(&nf_ct_frag6_mem) > nf_ct_frag6_high_thresh)
 798                 nf_ct_frag6_evictor();
 799
 800         fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
 801         if (fq == NULL) {
 802                 DEBUGP("Can't find and can't create new queue\n");
 803                 goto ret_orig;
 804         }
 805
 806         spin_lock(&fq->lock);
 807
 808         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
 809                 spin_unlock(&fq->lock);
 810                 DEBUGP("Can't insert skb to queue\n");
 811                 fq_put(fq, NULL);
 812                 goto ret_orig;
 813         }
 814
 815         if (fq->last_in == (FIRST_IN|LAST_IN) && fq->meat == fq->len) {
 816                 ret_skb = nf_ct_frag6_reasm(fq, dev);
 817                 if (ret_skb == NULL)
 818                         DEBUGP("Can't reassemble fragmented packets\n");
 819         }
 820         spin_unlock(&fq->lock);
 821
 822         fq_put(fq, NULL);
 823         return ret_skb;
 824
 825 ret_orig:
 826         kfree_skb(clone);
 827         return skb;
 828 }
 829
 830 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
 831                         struct net_device *in, struct net_device *out,
 832                         int (*okfn)(struct sk_buff *))
 833 {
 834         struct sk_buff *s, *s2;
 835
 836         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
 837                 nf_conntrack_put_reasm(s->nfct_reasm);
 838                 nf_conntrack_get_reasm(skb);
 839                 s->nfct_reasm = skb;
 840
 841                 s2 = s->next;
 842                 s->next = NULL;
 843
 844                 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
 845                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
 846                 s = s2;
 847         }
 848         nf_conntrack_put_reasm(skb);
 849 }
 850
 851 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
 852 {
 853         struct sk_buff *s, *s2;
 854
 855         for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
 856
 857                 s2 = s->next;
 858                 kfree_skb(s);
 859         }
 860
 861         kfree_skb(skb);
 862
 863         return 0;
 864 }
 865
 866 int nf_ct_frag6_init(void)
 867 {
 868         nf_ct_frag6_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
 869                                    (jiffies ^ (jiffies >> 6)));
 870
 871         setup_timer(&nf_ct_frag6_secret_timer, nf_ct_frag6_secret_rebuild, 0);
 872         nf_ct_frag6_secret_timer.expires = jiffies
 873                                            + nf_ct_frag6_secret_interval;
 874         add_timer(&nf_ct_frag6_secret_timer);
 875
 876         return 0;
 877 }
 878
 879 void nf_ct_frag6_cleanup(void)
 880 {
 881         del_timer(&nf_ct_frag6_secret_timer);
 882         nf_ct_frag6_low_thresh = 0;
 883         nf_ct_frag6_evictor();
 884 }