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/config.h>
  18 #include <linux/errno.h>
  19 #include <linux/types.h>
  20 #include <linux/string.h>
  21 #include <linux/socket.h>
  22 #include <linux/sockios.h>
  23 #include <linux/jiffies.h>
  24 #include <linux/net.h>
  25 #include <linux/list.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/in6.h>
  28 #include <linux/ipv6.h>
  29 #include <linux/icmpv6.h>
  30 #include <linux/random.h>
  31 #include <linux/jhash.h>
  32
  33 #include <net/sock.h>
  34 #include <net/snmp.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 #if 0
  49 #define DEBUGP printk
  50 #else
  51 #define DEBUGP(format, args...)
  52 #endif
  53
  54 #define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */
  55 #define NF_CT_FRAG6_LOW_THRESH 196608  /* == 192*1024 */
  56 #define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT
  57
  58 unsigned int nf_ct_frag6_high_thresh = 256*1024;
  59 unsigned int nf_ct_frag6_low_thresh = 192*1024;
  60 unsigned long nf_ct_frag6_timeout = IPV6_FRAG_TIMEOUT;
  61
  62 struct nf_ct_frag6_skb_cb
  63 {
  64         struct inet6_skb_parm   h;
  65         int                     offset;
  66         struct sk_buff          *orig;
  67 };
  68
  69 #define NFCT_FRAG6_CB(skb)      ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
  70
  71 struct nf_ct_frag6_queue
  72 {
  73         struct nf_ct_frag6_queue        *next;
  74         struct list_head lru_list;              /* lru list member      */
  75
  76         __u32                   id;             /* fragment id          */
  77         struct in6_addr         saddr;
  78         struct in6_addr         daddr;
  79
  80         spinlock_t              lock;
  81         atomic_t                refcnt;
  82         struct timer_list       timer;          /* expire timer         */
  83         struct sk_buff          *fragments;
  84         int                     len;
  85         int                     meat;
  86         struct timeval          stamp;
  87         unsigned int            csum;
  88         __u8                    last_in;        /* has first/last segment arrived? */
  89 #define COMPLETE                4
  90 #define FIRST_IN                2
  91 #define LAST_IN                 1
  92         __u16                   nhoffset;
  93         struct nf_ct_frag6_queue        **pprev;
  94 };
  95
  96 /* Hash table. */
  97
  98 #define FRAG6Q_HASHSZ   64
  99
 100 static struct nf_ct_frag6_queue *nf_ct_frag6_hash[FRAG6Q_HASHSZ];
 101 static DEFINE_RWLOCK(nf_ct_frag6_lock);
 102 static u32 nf_ct_frag6_hash_rnd;
 103 static LIST_HEAD(nf_ct_frag6_lru_list);
 104 int nf_ct_frag6_nqueues = 0;
 105
 106 static __inline__ void __fq_unlink(struct nf_ct_frag6_queue *fq)
 107 {
 108         if (fq->next)
 109                 fq->next->pprev = fq->pprev;
 110         *fq->pprev = fq->next;
 111         list_del(&fq->lru_list);
 112         nf_ct_frag6_nqueues--;
 113 }
 114
 115 static __inline__ void fq_unlink(struct nf_ct_frag6_queue *fq)
 116 {
 117         write_lock(&nf_ct_frag6_lock);
 118         __fq_unlink(fq);
 119         write_unlock(&nf_ct_frag6_lock);
 120 }
 121
 122 static unsigned int ip6qhashfn(u32 id, struct in6_addr *saddr,
 123                                struct in6_addr *daddr)
 124 {
 125         u32 a, b, c;
 126
 127         a = saddr->s6_addr32[0];
 128         b = saddr->s6_addr32[1];
 129         c = saddr->s6_addr32[2];
 130
 131         a += JHASH_GOLDEN_RATIO;
 132         b += JHASH_GOLDEN_RATIO;
 133         c += nf_ct_frag6_hash_rnd;
 134         __jhash_mix(a, b, c);
 135
 136         a += saddr->s6_addr32[3];
 137         b += daddr->s6_addr32[0];
 138         c += daddr->s6_addr32[1];
 139         __jhash_mix(a, b, c);
 140
 141         a += daddr->s6_addr32[2];
 142         b += daddr->s6_addr32[3];
 143         c += id;
 144         __jhash_mix(a, b, c);
 145
 146         return c & (FRAG6Q_HASHSZ - 1);
 147 }
 148
 149 static struct timer_list nf_ct_frag6_secret_timer;
 150 int nf_ct_frag6_secret_interval = 10 * 60 * HZ;
 151
 152 static void nf_ct_frag6_secret_rebuild(unsigned long dummy)
 153 {
 154         unsigned long now = jiffies;
 155         int i;
 156
 157         write_lock(&nf_ct_frag6_lock);
 158         get_random_bytes(&nf_ct_frag6_hash_rnd, sizeof(u32));
 159         for (i = 0; i < FRAG6Q_HASHSZ; i++) {
 160                 struct nf_ct_frag6_queue *q;
 161
 162                 q = nf_ct_frag6_hash[i];
 163                 while (q) {
 164                         struct nf_ct_frag6_queue *next = q->next;
 165                         unsigned int hval = ip6qhashfn(q->id,
 166                                                        &q->saddr,
 167                                                        &q->daddr);
 168
 169                         if (hval != i) {
 170                                 /* Unlink. */
 171                                 if (q->next)
 172                                         q->next->pprev = q->pprev;
 173                                 *q->pprev = q->next;
 174
 175                                 /* Relink to new hash chain. */
 176                                 if ((q->next = nf_ct_frag6_hash[hval]) != NULL)
 177                                         q->next->pprev = &q->next;
 178                                 nf_ct_frag6_hash[hval] = q;
 179                                 q->pprev = &nf_ct_frag6_hash[hval];
 180                         }
 181
 182                         q = next;
 183                 }
 184         }
 185         write_unlock(&nf_ct_frag6_lock);
 186
 187         mod_timer(&nf_ct_frag6_secret_timer, now + nf_ct_frag6_secret_interval);
 188 }
 189
 190 atomic_t nf_ct_frag6_mem = ATOMIC_INIT(0);
 191
 192 /* Memory Tracking Functions. */
 193 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
 194 {
 195         if (work)
 196                 *work -= skb->truesize;
 197         atomic_sub(skb->truesize, &nf_ct_frag6_mem);
 198         if (NFCT_FRAG6_CB(skb)->orig)
 199                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
 200
 201         kfree_skb(skb);
 202 }
 203
 204 static inline void frag_free_queue(struct nf_ct_frag6_queue *fq,
 205                                    unsigned int *work)
 206 {
 207         if (work)
 208                 *work -= sizeof(struct nf_ct_frag6_queue);
 209         atomic_sub(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
 210         kfree(fq);
 211 }
 212
 213 static inline struct nf_ct_frag6_queue *frag_alloc_queue(void)
 214 {
 215         struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
 216
 217         if (!fq)
 218                 return NULL;
 219         atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
 220         return fq;
 221 }
 222
 223 /* Destruction primitives. */
 224
 225 /* Complete destruction of fq. */
 226 static void nf_ct_frag6_destroy(struct nf_ct_frag6_queue *fq,
 227                                 unsigned int *work)
 228 {
 229         struct sk_buff *fp;
 230
 231         BUG_TRAP(fq->last_in&COMPLETE);
 232         BUG_TRAP(del_timer(&fq->timer) == 0);
 233
 234         /* Release all fragment data. */
 235         fp = fq->fragments;
 236         while (fp) {
 237                 struct sk_buff *xp = fp->next;
 238
 239                 frag_kfree_skb(fp, work);
 240                 fp = xp;
 241         }
 242
 243         frag_free_queue(fq, work);
 244 }
 245
 246 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq, unsigned int *work)
 247 {
 248         if (atomic_dec_and_test(&fq->refcnt))
 249                 nf_ct_frag6_destroy(fq, work);
 250 }
 251
 252 /* Kill fq entry. It is not destroyed immediately,
 253  * because caller (and someone more) holds reference count.
 254  */
 255 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
 256 {
 257         if (del_timer(&fq->timer))
 258                 atomic_dec(&fq->refcnt);
 259
 260         if (!(fq->last_in & COMPLETE)) {
 261                 fq_unlink(fq);
 262                 atomic_dec(&fq->refcnt);
 263                 fq->last_in |= COMPLETE;
 264         }
 265 }
 266
 267 static void nf_ct_frag6_evictor(void)
 268 {
 269         struct nf_ct_frag6_queue *fq;
 270         struct list_head *tmp;
 271         unsigned int work;
 272
 273         work = atomic_read(&nf_ct_frag6_mem);
 274         if (work <= nf_ct_frag6_low_thresh)
 275                 return;
 276
 277         work -= nf_ct_frag6_low_thresh;
 278         while (work > 0) {
 279                 read_lock(&nf_ct_frag6_lock);
 280                 if (list_empty(&nf_ct_frag6_lru_list)) {
 281                         read_unlock(&nf_ct_frag6_lock);
 282                         return;
 283                 }
 284                 tmp = nf_ct_frag6_lru_list.next;
 285                 BUG_ON(tmp == NULL);
 286                 fq = list_entry(tmp, struct nf_ct_frag6_queue, lru_list);
 287                 atomic_inc(&fq->refcnt);
 288                 read_unlock(&nf_ct_frag6_lock);
 289
 290                 spin_lock(&fq->lock);
 291                 if (!(fq->last_in&COMPLETE))
 292                         fq_kill(fq);
 293                 spin_unlock(&fq->lock);
 294
 295                 fq_put(fq, &work);
 296         }
 297 }
 298
 299 static void nf_ct_frag6_expire(unsigned long data)
 300 {
 301         struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
 302
 303         spin_lock(&fq->lock);
 304
 305         if (fq->last_in & COMPLETE)
 306                 goto out;
 307
 308         fq_kill(fq);
 309
 310 out:
 311         spin_unlock(&fq->lock);
 312         fq_put(fq, NULL);
 313 }
 314
 315 /* Creation primitives. */
 316
 317
 318 static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
 319                                           struct nf_ct_frag6_queue *fq_in)
 320 {
 321         struct nf_ct_frag6_queue *fq;
 322
 323         write_lock(&nf_ct_frag6_lock);
 324 #ifdef CONFIG_SMP
 325         for (fq = nf_ct_frag6_hash[hash]; fq; fq = fq->next) {
 326                 if (fq->id == fq_in->id &&
 327                     !ipv6_addr_cmp(&fq_in->saddr, &fq->saddr) &&
 328                     !ipv6_addr_cmp(&fq_in->daddr, &fq->daddr)) {
 329                         atomic_inc(&fq->refcnt);
 330                         write_unlock(&nf_ct_frag6_lock);
 331                         fq_in->last_in |= COMPLETE;
 332                         fq_put(fq_in, NULL);
 333                         return fq;
 334                 }
 335         }
 336 #endif
 337         fq = fq_in;
 338
 339         if (!mod_timer(&fq->timer, jiffies + nf_ct_frag6_timeout))
 340                 atomic_inc(&fq->refcnt);
 341
 342         atomic_inc(&fq->refcnt);
 343         if ((fq->next = nf_ct_frag6_hash[hash]) != NULL)
 344                 fq->next->pprev = &fq->next;
 345         nf_ct_frag6_hash[hash] = fq;
 346         fq->pprev = &nf_ct_frag6_hash[hash];
 347         INIT_LIST_HEAD(&fq->lru_list);
 348         list_add_tail(&fq->lru_list, &nf_ct_frag6_lru_list);
 349         nf_ct_frag6_nqueues++;
 350         write_unlock(&nf_ct_frag6_lock);
 351         return fq;
 352 }
 353
 354
 355 static struct nf_ct_frag6_queue *
 356 nf_ct_frag6_create(unsigned int hash, u32 id, struct in6_addr *src,                                struct in6_addr *dst)
 357 {
 358         struct nf_ct_frag6_queue *fq;
 359
 360         if ((fq = frag_alloc_queue()) == NULL) {
 361                 DEBUGP("Can't alloc new queue\n");
 362                 goto oom;
 363         }
 364
 365         memset(fq, 0, sizeof(struct nf_ct_frag6_queue));
 366
 367         fq->id = id;
 368         ipv6_addr_copy(&fq->saddr, src);
 369         ipv6_addr_copy(&fq->daddr, dst);
 370
 371         init_timer(&fq->timer);
 372         fq->timer.function = nf_ct_frag6_expire;
 373         fq->timer.data = (long) fq;
 374         spin_lock_init(&fq->lock);
 375         atomic_set(&fq->refcnt, 1);
 376
 377         return nf_ct_frag6_intern(hash, fq);
 378
 379 oom:
 380         return NULL;
 381 }
 382
 383 static __inline__ struct nf_ct_frag6_queue *
 384 fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst)
 385 {
 386         struct nf_ct_frag6_queue *fq;
 387         unsigned int hash = ip6qhashfn(id, src, dst);
 388
 389         read_lock(&nf_ct_frag6_lock);
 390         for (fq = nf_ct_frag6_hash[hash]; fq; fq = fq->next) {
 391                 if (fq->id == id &&
 392                     !ipv6_addr_cmp(src, &fq->saddr) &&
 393                     !ipv6_addr_cmp(dst, &fq->daddr)) {
 394                         atomic_inc(&fq->refcnt);
 395                         read_unlock(&nf_ct_frag6_lock);
 396                         return fq;
 397                 }
 398         }
 399         read_unlock(&nf_ct_frag6_lock);
 400
 401         return nf_ct_frag6_create(hash, id, src, dst);
 402 }
 403
 404
 405 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
 406                              struct frag_hdr *fhdr, int nhoff)
 407 {
 408         struct sk_buff *prev, *next;
 409         int offset, end;
 410
 411         if (fq->last_in & COMPLETE) {
 412                 DEBUGP("Allready completed\n");
 413                 goto err;
 414         }
 415
 416         offset = ntohs(fhdr->frag_off) & ~0x7;
 417         end = offset + (ntohs(skb->nh.ipv6h->payload_len) -
 418                         ((u8 *) (fhdr + 1) - (u8 *) (skb->nh.ipv6h + 1)));
 419
 420         if ((unsigned int)end > IPV6_MAXPLEN) {
 421                 DEBUGP("offset is too large.\n");
 422                 return -1;
 423         }
 424
 425         if (skb->ip_summed == CHECKSUM_HW)
 426                 skb->csum = csum_sub(skb->csum,
 427                                      csum_partial(skb->nh.raw,
 428                                                   (u8*)(fhdr + 1) - skb->nh.raw,
 429                                                   0));
 430
 431         /* Is this the final fragment? */
 432         if (!(fhdr->frag_off & htons(IP6_MF))) {
 433                 /* If we already have some bits beyond end
 434                  * or have different end, the segment is corrupted.
 435                  */
 436                 if (end < fq->len ||
 437                     ((fq->last_in & LAST_IN) && end != fq->len)) {
 438                         DEBUGP("already received last fragment\n");
 439                         goto err;
 440                 }
 441                 fq->last_in |= LAST_IN;
 442                 fq->len = end;
 443         } else {
 444                 /* Check if the fragment is rounded to 8 bytes.
 445                  * Required by the RFC.
 446                  */
 447                 if (end & 0x7) {
 448                         /* RFC2460 says always send parameter problem in
 449                          * this case. -DaveM
 450                          */
 451                         DEBUGP("the end of this fragment is not rounded to 8 bytes.\n");
 452                         return -1;
 453                 }
 454                 if (end > fq->len) {
 455                         /* Some bits beyond end -> corruption. */
 456                         if (fq->last_in & LAST_IN) {
 457                                 DEBUGP("last packet already reached.\n");
 458                                 goto err;
 459                         }
 460                         fq->len = end;
 461                 }
 462         }
 463
 464         if (end == offset)
 465                 goto err;
 466
 467         /* Point into the IP datagram 'data' part. */
 468         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
 469                 DEBUGP("queue: message is too short.\n");
 470                 goto err;
 471         }
 472         if (end-offset < skb->len) {
 473                 if (pskb_trim(skb, end - offset)) {
 474                         DEBUGP("Can't trim\n");
 475                         goto err;
 476                 }
 477                 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 478                         skb->ip_summed = CHECKSUM_NONE;
 479         }
 480
 481         /* Find out which fragments are in front and at the back of us
 482          * in the chain of fragments so far.  We must know where to put
 483          * this fragment, right?
 484          */
 485         prev = NULL;
 486         for (next = fq->fragments; next != NULL; next = next->next) {
 487                 if (NFCT_FRAG6_CB(next)->offset >= offset)
 488                         break;  /* bingo! */
 489                 prev = next;
 490         }
 491
 492         /* We found where to put this one.  Check for overlap with
 493          * preceding fragment, and, if needed, align things so that
 494          * any overlaps are eliminated.
 495          */
 496         if (prev) {
 497                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
 498
 499                 if (i > 0) {
 500                         offset += i;
 501                         if (end <= offset) {
 502                                 DEBUGP("overlap\n");
 503                                 goto err;
 504                         }
 505                         if (!pskb_pull(skb, i)) {
 506                                 DEBUGP("Can't pull\n");
 507                                 goto err;
 508                         }
 509                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 510                                 skb->ip_summed = CHECKSUM_NONE;
 511                 }
 512         }
 513
 514         /* Look for overlap with succeeding segments.
 515          * If we can merge fragments, do it.
 516          */
 517         while (next && NFCT_FRAG6_CB(next)->offset < end) {
 518                 /* overlap is 'i' bytes */
 519                 int i = end - NFCT_FRAG6_CB(next)->offset;
 520
 521                 if (i < next->len) {
 522                         /* Eat head of the next overlapped fragment
 523                          * and leave the loop. The next ones cannot overlap.
 524                          */
 525                         DEBUGP("Eat head of the overlapped parts.: %d", i);
 526                         if (!pskb_pull(next, i))
 527                                 goto err;
 528
 529                         /* next fragment */
 530                         NFCT_FRAG6_CB(next)->offset += i;
 531                         fq->meat -= i;
 532                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 533                                 next->ip_summed = CHECKSUM_NONE;
 534                         break;
 535                 } else {
 536                         struct sk_buff *free_it = next;
 537
 538                         /* Old fragmnet is completely overridden with
 539                          * new one drop it.
 540                          */
 541                         next = next->next;
 542
 543                         if (prev)
 544                                 prev->next = next;
 545                         else
 546                                 fq->fragments = next;
 547
 548                         fq->meat -= free_it->len;
 549                         frag_kfree_skb(free_it, NULL);
 550                 }
 551         }
 552
 553         NFCT_FRAG6_CB(skb)->offset = offset;
 554
 555         /* Insert this fragment in the chain of fragments. */
 556         skb->next = next;
 557         if (prev)
 558                 prev->next = skb;
 559         else
 560                 fq->fragments = skb;
 561
 562         skb->dev = NULL;
 563         skb_get_timestamp(skb, &fq->stamp);
 564         fq->meat += skb->len;
 565         atomic_add(skb->truesize, &nf_ct_frag6_mem);
 566
 567         /* The first fragment.
 568          * nhoffset is obtained from the first fragment, of course.
 569          */
 570         if (offset == 0) {
 571                 fq->nhoffset = nhoff;
 572                 fq->last_in |= FIRST_IN;
 573         }
 574         write_lock(&nf_ct_frag6_lock);
 575         list_move_tail(&fq->lru_list, &nf_ct_frag6_lru_list);
 576         write_unlock(&nf_ct_frag6_lock);
 577         return 0;
 578
 579 err:
 580         return -1;
 581 }
 582
 583 /*
 584  *      Check if this packet is complete.
 585  *      Returns NULL on failure by any reason, and pointer
 586  *      to current nexthdr field in reassembled frame.
 587  *
 588  *      It is called with locked fq, and caller must check that
 589  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 590  *      the last and the first frames arrived and all the bits are here.
 591  */
 592 static struct sk_buff *
 593 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
 594 {
 595         struct sk_buff *fp, *op, *head = fq->fragments;
 596         int    payload_len;
 597
 598         fq_kill(fq);
 599
 600         BUG_TRAP(head != NULL);
 601         BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
 602
 603         /* Unfragmented part is taken from the first segment. */
 604         payload_len = (head->data - head->nh.raw) - sizeof(struct ipv6hdr) + fq->len - sizeof(struct frag_hdr);
 605         if (payload_len > IPV6_MAXPLEN) {
 606                 DEBUGP("payload len is too large.\n");
 607                 goto out_oversize;
 608         }
 609
 610         /* Head of list must not be cloned. */
 611         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
 612                 DEBUGP("skb is cloned but can't expand head");
 613                 goto out_oom;
 614         }
 615
 616         /* If the first fragment is fragmented itself, we split
 617          * it to two chunks: the first with data and paged part
 618          * and the second, holding only fragments. */
 619         if (skb_shinfo(head)->frag_list) {
 620                 struct sk_buff *clone;
 621                 int i, plen = 0;
 622
 623                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
 624                         DEBUGP("Can't alloc skb\n");
 625                         goto out_oom;
 626                 }
 627                 clone->next = head->next;
 628                 head->next = clone;
 629                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 630                 skb_shinfo(head)->frag_list = NULL;
 631                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 632                         plen += skb_shinfo(head)->frags[i].size;
 633                 clone->len = clone->data_len = head->data_len - plen;
 634                 head->data_len -= clone->len;
 635                 head->len -= clone->len;
 636                 clone->csum = 0;
 637                 clone->ip_summed = head->ip_summed;
 638
 639                 NFCT_FRAG6_CB(clone)->orig = NULL;
 640                 atomic_add(clone->truesize, &nf_ct_frag6_mem);
 641         }
 642
 643         /* We have to remove fragment header from datagram and to relocate
 644          * header in order to calculate ICV correctly. */
 645         head->nh.raw[fq->nhoffset] = head->h.raw[0];
 646         memmove(head->head + sizeof(struct frag_hdr), head->head,
 647                 (head->data - head->head) - sizeof(struct frag_hdr));
 648         head->mac.raw += sizeof(struct frag_hdr);
 649         head->nh.raw += sizeof(struct frag_hdr);
 650
 651         skb_shinfo(head)->frag_list = head->next;
 652         head->h.raw = head->data;
 653         skb_push(head, head->data - head->nh.raw);
 654         atomic_sub(head->truesize, &nf_ct_frag6_mem);
 655
 656         for (fp=head->next; fp; fp = fp->next) {
 657                 head->data_len += fp->len;
 658                 head->len += fp->len;
 659                 if (head->ip_summed != fp->ip_summed)
 660                         head->ip_summed = CHECKSUM_NONE;
 661                 else if (head->ip_summed == CHECKSUM_HW)
 662                         head->csum = csum_add(head->csum, fp->csum);
 663                 head->truesize += fp->truesize;
 664                 atomic_sub(fp->truesize, &nf_ct_frag6_mem);
 665         }
 666
 667         head->next = NULL;
 668         head->dev = dev;
 669         skb_set_timestamp(head, &fq->stamp);
 670         head->nh.ipv6h->payload_len = htons(payload_len);
 671
 672         /* Yes, and fold redundant checksum back. 8) */
 673         if (head->ip_summed == CHECKSUM_HW)
 674                 head->csum = csum_partial(head->nh.raw, head->h.raw-head->nh.raw, head->csum);
 675
 676         fq->fragments = NULL;
 677
 678         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
 679         fp = skb_shinfo(head)->frag_list;
 680         if (NFCT_FRAG6_CB(fp)->orig == NULL)
 681                 /* at above code, head skb is divided into two skbs. */
 682                 fp = fp->next;
 683
 684         op = NFCT_FRAG6_CB(head)->orig;
 685         for (; fp; fp = fp->next) {
 686                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
 687
 688                 op->next = orig;
 689                 op = orig;
 690                 NFCT_FRAG6_CB(fp)->orig = NULL;
 691         }
 692
 693         return head;
 694
 695 out_oversize:
 696         if (net_ratelimit())
 697                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
 698         goto out_fail;
 699 out_oom:
 700         if (net_ratelimit())
 701                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
 702 out_fail:
 703         return NULL;
 704 }
 705
 706 /*
 707  * find the header just before Fragment Header.
 708  *
 709  * if success return 0 and set ...
 710  * (*prevhdrp): the value of "Next Header Field" in the header
 711  *              just before Fragment Header.
 712  * (*prevhoff): the offset of "Next Header Field" in the header
 713  *              just before Fragment Header.
 714  * (*fhoff)   : the offset of Fragment Header.
 715  *
 716  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
 717  *
 718  */
 719 static int
 720 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
 721 {
 722         u8 nexthdr = skb->nh.ipv6h->nexthdr;
 723         u8 prev_nhoff = (u8 *)&skb->nh.ipv6h->nexthdr - skb->data;
 724         int start = (u8 *)(skb->nh.ipv6h+1) - skb->data;
 725         int len = skb->len - start;
 726         u8 prevhdr = NEXTHDR_IPV6;
 727
 728         while (nexthdr != NEXTHDR_FRAGMENT) {
 729                 struct ipv6_opt_hdr hdr;
 730                 int hdrlen;
 731
 732                 if (!ipv6_ext_hdr(nexthdr)) {
 733                         return -1;
 734                 }
 735                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
 736                         DEBUGP("too short\n");
 737                         return -1;
 738                 }
 739                 if (nexthdr == NEXTHDR_NONE) {
 740                         DEBUGP("next header is none\n");
 741                         return -1;
 742                 }
 743                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
 744                         BUG();
 745                 if (nexthdr == NEXTHDR_AUTH)
 746                         hdrlen = (hdr.hdrlen+2)<<2;
 747                 else
 748                         hdrlen = ipv6_optlen(&hdr);
 749
 750                 prevhdr = nexthdr;
 751                 prev_nhoff = start;
 752
 753                 nexthdr = hdr.nexthdr;
 754                 len -= hdrlen;
 755                 start += hdrlen;
 756         }
 757
 758         if (len < 0)
 759                 return -1;
 760
 761         *prevhdrp = prevhdr;
 762         *prevhoff = prev_nhoff;
 763         *fhoff = start;
 764
 765         return 0;
 766 }
 767
 768 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
 769 {
 770         struct sk_buff *clone;
 771         struct net_device *dev = skb->dev;
 772         struct frag_hdr *fhdr;
 773         struct nf_ct_frag6_queue *fq;
 774         struct ipv6hdr *hdr;
 775         int fhoff, nhoff;
 776         u8 prevhdr;
 777         struct sk_buff *ret_skb = NULL;
 778
 779         /* Jumbo payload inhibits frag. header */
 780         if (skb->nh.ipv6h->payload_len == 0) {
 781                 DEBUGP("payload len = 0\n");
 782                 return skb;
 783         }
 784
 785         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
 786                 return skb;
 787
 788         clone = skb_clone(skb, GFP_ATOMIC);
 789         if (clone == NULL) {
 790                 DEBUGP("Can't clone skb\n");
 791                 return skb;
 792         }
 793
 794         NFCT_FRAG6_CB(clone)->orig = skb;
 795
 796         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
 797                 DEBUGP("message is too short.\n");
 798                 goto ret_orig;
 799         }
 800
 801         clone->h.raw = clone->data + fhoff;
 802         hdr = clone->nh.ipv6h;
 803         fhdr = (struct frag_hdr *)clone->h.raw;
 804
 805         if (!(fhdr->frag_off & htons(0xFFF9))) {
 806                 DEBUGP("Invalid fragment offset\n");
 807                 /* It is not a fragmented frame */
 808                 goto ret_orig;
 809         }
 810
 811         if (atomic_read(&nf_ct_frag6_mem) > nf_ct_frag6_high_thresh)
 812                 nf_ct_frag6_evictor();
 813
 814         fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
 815         if (fq == NULL) {
 816                 DEBUGP("Can't find and can't create new queue\n");
 817                 goto ret_orig;
 818         }
 819
 820         spin_lock(&fq->lock);
 821
 822         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
 823                 spin_unlock(&fq->lock);
 824                 DEBUGP("Can't insert skb to queue\n");
 825                 fq_put(fq, NULL);
 826                 goto ret_orig;
 827         }
 828
 829         if (fq->last_in == (FIRST_IN|LAST_IN) && fq->meat == fq->len) {
 830                 ret_skb = nf_ct_frag6_reasm(fq, dev);
 831                 if (ret_skb == NULL)
 832                         DEBUGP("Can't reassemble fragmented packets\n");
 833         }
 834         spin_unlock(&fq->lock);
 835
 836         fq_put(fq, NULL);
 837         return ret_skb;
 838
 839 ret_orig:
 840         kfree_skb(clone);
 841         return skb;
 842 }
 843
 844 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
 845                         struct net_device *in, struct net_device *out,
 846                         int (*okfn)(struct sk_buff *))
 847 {
 848         struct sk_buff *s, *s2;
 849
 850         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
 851                 nf_conntrack_put_reasm(s->nfct_reasm);
 852                 nf_conntrack_get_reasm(skb);
 853                 s->nfct_reasm = skb;
 854
 855                 s2 = s->next;
 856                 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
 857                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
 858                 s = s2;
 859         }
 860         nf_conntrack_put_reasm(skb);
 861 }
 862
 863 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
 864 {
 865         struct sk_buff *s, *s2;
 866
 867         for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
 868
 869                 s2 = s->next;
 870                 kfree_skb(s);
 871         }
 872
 873         kfree_skb(skb);
 874
 875         return 0;
 876 }
 877
 878 int nf_ct_frag6_init(void)
 879 {
 880         nf_ct_frag6_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
 881                                    (jiffies ^ (jiffies >> 6)));
 882
 883         init_timer(&nf_ct_frag6_secret_timer);
 884         nf_ct_frag6_secret_timer.function = nf_ct_frag6_secret_rebuild;
 885         nf_ct_frag6_secret_timer.expires = jiffies
 886                                            + nf_ct_frag6_secret_interval;
 887         add_timer(&nf_ct_frag6_secret_timer);
 888
 889         return 0;
 890 }
 891
 892 void nf_ct_frag6_cleanup(void)
 893 {
 894         del_timer(&nf_ct_frag6_secret_timer);
 895         nf_ct_frag6_low_thresh = 0;
 896         nf_ct_frag6_evictor();
 897 }