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 hlist_node       list;
  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 };
  94
  95 /* Hash table. */
  96
  97 #define FRAG6Q_HASHSZ   64
  98
  99 static struct hlist_head nf_ct_frag6_hash[FRAG6Q_HASHSZ];
 100 static DEFINE_RWLOCK(nf_ct_frag6_lock);
 101 static u32 nf_ct_frag6_hash_rnd;
 102 static LIST_HEAD(nf_ct_frag6_lru_list);
 103 int nf_ct_frag6_nqueues = 0;
 104
 105 static __inline__ void __fq_unlink(struct nf_ct_frag6_queue *fq)
 106 {
 107         hlist_del(&fq->list);
 108         list_del(&fq->lru_list);
 109         nf_ct_frag6_nqueues--;
 110 }
 111
 112 static __inline__ void fq_unlink(struct nf_ct_frag6_queue *fq)
 113 {
 114         write_lock(&nf_ct_frag6_lock);
 115         __fq_unlink(fq);
 116         write_unlock(&nf_ct_frag6_lock);
 117 }
 118
 119 static unsigned int ip6qhashfn(u32 id, struct in6_addr *saddr,
 120                                struct in6_addr *daddr)
 121 {
 122         u32 a, b, c;
 123
 124         a = saddr->s6_addr32[0];
 125         b = saddr->s6_addr32[1];
 126         c = saddr->s6_addr32[2];
 127
 128         a += JHASH_GOLDEN_RATIO;
 129         b += JHASH_GOLDEN_RATIO;
 130         c += nf_ct_frag6_hash_rnd;
 131         __jhash_mix(a, b, c);
 132
 133         a += saddr->s6_addr32[3];
 134         b += daddr->s6_addr32[0];
 135         c += daddr->s6_addr32[1];
 136         __jhash_mix(a, b, c);
 137
 138         a += daddr->s6_addr32[2];
 139         b += daddr->s6_addr32[3];
 140         c += id;
 141         __jhash_mix(a, b, c);
 142
 143         return c & (FRAG6Q_HASHSZ - 1);
 144 }
 145
 146 static struct timer_list nf_ct_frag6_secret_timer;
 147 int nf_ct_frag6_secret_interval = 10 * 60 * HZ;
 148
 149 static void nf_ct_frag6_secret_rebuild(unsigned long dummy)
 150 {
 151         unsigned long now = jiffies;
 152         int i;
 153
 154         write_lock(&nf_ct_frag6_lock);
 155         get_random_bytes(&nf_ct_frag6_hash_rnd, sizeof(u32));
 156         for (i = 0; i < FRAG6Q_HASHSZ; i++) {
 157                 struct nf_ct_frag6_queue *q;
 158                 struct hlist_node *p, *n;
 159
 160                 hlist_for_each_entry_safe(q, p, n, &nf_ct_frag6_hash[i], list) {
 161                         unsigned int hval = ip6qhashfn(q->id,
 162                                                        &q->saddr,
 163                                                        &q->daddr);
 164                         if (hval != i) {
 165                                 hlist_del(&q->list);
 166                                 /* Relink to new hash chain. */
 167                                 hlist_add_head(&q->list,
 168                                                &nf_ct_frag6_hash[hval]);
 169                         }
 170                 }
 171         }
 172         write_unlock(&nf_ct_frag6_lock);
 173
 174         mod_timer(&nf_ct_frag6_secret_timer, now + nf_ct_frag6_secret_interval);
 175 }
 176
 177 atomic_t nf_ct_frag6_mem = ATOMIC_INIT(0);
 178
 179 /* Memory Tracking Functions. */
 180 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
 181 {
 182         if (work)
 183                 *work -= skb->truesize;
 184         atomic_sub(skb->truesize, &nf_ct_frag6_mem);
 185         if (NFCT_FRAG6_CB(skb)->orig)
 186                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
 187
 188         kfree_skb(skb);
 189 }
 190
 191 static inline void frag_free_queue(struct nf_ct_frag6_queue *fq,
 192                                    unsigned int *work)
 193 {
 194         if (work)
 195                 *work -= sizeof(struct nf_ct_frag6_queue);
 196         atomic_sub(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
 197         kfree(fq);
 198 }
 199
 200 static inline struct nf_ct_frag6_queue *frag_alloc_queue(void)
 201 {
 202         struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
 203
 204         if (!fq)
 205                 return NULL;
 206         atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
 207         return fq;
 208 }
 209
 210 /* Destruction primitives. */
 211
 212 /* Complete destruction of fq. */
 213 static void nf_ct_frag6_destroy(struct nf_ct_frag6_queue *fq,
 214                                 unsigned int *work)
 215 {
 216         struct sk_buff *fp;
 217
 218         BUG_TRAP(fq->last_in&COMPLETE);
 219         BUG_TRAP(del_timer(&fq->timer) == 0);
 220
 221         /* Release all fragment data. */
 222         fp = fq->fragments;
 223         while (fp) {
 224                 struct sk_buff *xp = fp->next;
 225
 226                 frag_kfree_skb(fp, work);
 227                 fp = xp;
 228         }
 229
 230         frag_free_queue(fq, work);
 231 }
 232
 233 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq, unsigned int *work)
 234 {
 235         if (atomic_dec_and_test(&fq->refcnt))
 236                 nf_ct_frag6_destroy(fq, work);
 237 }
 238
 239 /* Kill fq entry. It is not destroyed immediately,
 240  * because caller (and someone more) holds reference count.
 241  */
 242 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
 243 {
 244         if (del_timer(&fq->timer))
 245                 atomic_dec(&fq->refcnt);
 246
 247         if (!(fq->last_in & COMPLETE)) {
 248                 fq_unlink(fq);
 249                 atomic_dec(&fq->refcnt);
 250                 fq->last_in |= COMPLETE;
 251         }
 252 }
 253
 254 static void nf_ct_frag6_evictor(void)
 255 {
 256         struct nf_ct_frag6_queue *fq;
 257         struct list_head *tmp;
 258         unsigned int work;
 259
 260         work = atomic_read(&nf_ct_frag6_mem);
 261         if (work <= nf_ct_frag6_low_thresh)
 262                 return;
 263
 264         work -= nf_ct_frag6_low_thresh;
 265         while (work > 0) {
 266                 read_lock(&nf_ct_frag6_lock);
 267                 if (list_empty(&nf_ct_frag6_lru_list)) {
 268                         read_unlock(&nf_ct_frag6_lock);
 269                         return;
 270                 }
 271                 tmp = nf_ct_frag6_lru_list.next;
 272                 BUG_ON(tmp == NULL);
 273                 fq = list_entry(tmp, struct nf_ct_frag6_queue, lru_list);
 274                 atomic_inc(&fq->refcnt);
 275                 read_unlock(&nf_ct_frag6_lock);
 276
 277                 spin_lock(&fq->lock);
 278                 if (!(fq->last_in&COMPLETE))
 279                         fq_kill(fq);
 280                 spin_unlock(&fq->lock);
 281
 282                 fq_put(fq, &work);
 283         }
 284 }
 285
 286 static void nf_ct_frag6_expire(unsigned long data)
 287 {
 288         struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
 289
 290         spin_lock(&fq->lock);
 291
 292         if (fq->last_in & COMPLETE)
 293                 goto out;
 294
 295         fq_kill(fq);
 296
 297 out:
 298         spin_unlock(&fq->lock);
 299         fq_put(fq, NULL);
 300 }
 301
 302 /* Creation primitives. */
 303
 304 static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
 305                                           struct nf_ct_frag6_queue *fq_in)
 306 {
 307         struct nf_ct_frag6_queue *fq;
 308 #ifdef CONFIG_SMP
 309         struct hlist_node *n;
 310 #endif
 311
 312         write_lock(&nf_ct_frag6_lock);
 313 #ifdef CONFIG_SMP
 314         hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], list) {
 315                 if (fq->id == fq_in->id &&
 316                     ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
 317                     ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
 318                         atomic_inc(&fq->refcnt);
 319                         write_unlock(&nf_ct_frag6_lock);
 320                         fq_in->last_in |= COMPLETE;
 321                         fq_put(fq_in, NULL);
 322                         return fq;
 323                 }
 324         }
 325 #endif
 326         fq = fq_in;
 327
 328         if (!mod_timer(&fq->timer, jiffies + nf_ct_frag6_timeout))
 329                 atomic_inc(&fq->refcnt);
 330
 331         atomic_inc(&fq->refcnt);
 332         hlist_add_head(&fq->list, &nf_ct_frag6_hash[hash]);
 333         INIT_LIST_HEAD(&fq->lru_list);
 334         list_add_tail(&fq->lru_list, &nf_ct_frag6_lru_list);
 335         nf_ct_frag6_nqueues++;
 336         write_unlock(&nf_ct_frag6_lock);
 337         return fq;
 338 }
 339
 340
 341 static struct nf_ct_frag6_queue *
 342 nf_ct_frag6_create(unsigned int hash, u32 id, struct in6_addr *src,                                struct in6_addr *dst)
 343 {
 344         struct nf_ct_frag6_queue *fq;
 345
 346         if ((fq = frag_alloc_queue()) == NULL) {
 347                 DEBUGP("Can't alloc new queue\n");
 348                 goto oom;
 349         }
 350
 351         memset(fq, 0, sizeof(struct nf_ct_frag6_queue));
 352
 353         fq->id = id;
 354         ipv6_addr_copy(&fq->saddr, src);
 355         ipv6_addr_copy(&fq->daddr, dst);
 356
 357         init_timer(&fq->timer);
 358         fq->timer.function = nf_ct_frag6_expire;
 359         fq->timer.data = (long) fq;
 360         spin_lock_init(&fq->lock);
 361         atomic_set(&fq->refcnt, 1);
 362
 363         return nf_ct_frag6_intern(hash, fq);
 364
 365 oom:
 366         return NULL;
 367 }
 368
 369 static __inline__ struct nf_ct_frag6_queue *
 370 fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst)
 371 {
 372         struct nf_ct_frag6_queue *fq;
 373         struct hlist_node *n;
 374         unsigned int hash = ip6qhashfn(id, src, dst);
 375
 376         read_lock(&nf_ct_frag6_lock);
 377         hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], list) {
 378                 if (fq->id == id &&
 379                     ipv6_addr_equal(src, &fq->saddr) &&
 380                     ipv6_addr_equal(dst, &fq->daddr)) {
 381                         atomic_inc(&fq->refcnt);
 382                         read_unlock(&nf_ct_frag6_lock);
 383                         return fq;
 384                 }
 385         }
 386         read_unlock(&nf_ct_frag6_lock);
 387
 388         return nf_ct_frag6_create(hash, id, src, dst);
 389 }
 390
 391
 392 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
 393                              struct frag_hdr *fhdr, int nhoff)
 394 {
 395         struct sk_buff *prev, *next;
 396         int offset, end;
 397
 398         if (fq->last_in & COMPLETE) {
 399                 DEBUGP("Allready completed\n");
 400                 goto err;
 401         }
 402
 403         offset = ntohs(fhdr->frag_off) & ~0x7;
 404         end = offset + (ntohs(skb->nh.ipv6h->payload_len) -
 405                         ((u8 *) (fhdr + 1) - (u8 *) (skb->nh.ipv6h + 1)));
 406
 407         if ((unsigned int)end > IPV6_MAXPLEN) {
 408                 DEBUGP("offset is too large.\n");
 409                 return -1;
 410         }
 411
 412         if (skb->ip_summed == CHECKSUM_HW)
 413                 skb->csum = csum_sub(skb->csum,
 414                                      csum_partial(skb->nh.raw,
 415                                                   (u8*)(fhdr + 1) - skb->nh.raw,
 416                                                   0));
 417
 418         /* Is this the final fragment? */
 419         if (!(fhdr->frag_off & htons(IP6_MF))) {
 420                 /* If we already have some bits beyond end
 421                  * or have different end, the segment is corrupted.
 422                  */
 423                 if (end < fq->len ||
 424                     ((fq->last_in & LAST_IN) && end != fq->len)) {
 425                         DEBUGP("already received last fragment\n");
 426                         goto err;
 427                 }
 428                 fq->last_in |= LAST_IN;
 429                 fq->len = end;
 430         } else {
 431                 /* Check if the fragment is rounded to 8 bytes.
 432                  * Required by the RFC.
 433                  */
 434                 if (end & 0x7) {
 435                         /* RFC2460 says always send parameter problem in
 436                          * this case. -DaveM
 437                          */
 438                         DEBUGP("the end of this fragment is not rounded to 8 bytes.\n");
 439                         return -1;
 440                 }
 441                 if (end > fq->len) {
 442                         /* Some bits beyond end -> corruption. */
 443                         if (fq->last_in & LAST_IN) {
 444                                 DEBUGP("last packet already reached.\n");
 445                                 goto err;
 446                         }
 447                         fq->len = end;
 448                 }
 449         }
 450
 451         if (end == offset)
 452                 goto err;
 453
 454         /* Point into the IP datagram 'data' part. */
 455         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
 456                 DEBUGP("queue: message is too short.\n");
 457                 goto err;
 458         }
 459         if (end-offset < skb->len) {
 460                 if (pskb_trim(skb, end - offset)) {
 461                         DEBUGP("Can't trim\n");
 462                         goto err;
 463                 }
 464                 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 465                         skb->ip_summed = CHECKSUM_NONE;
 466         }
 467
 468         /* Find out which fragments are in front and at the back of us
 469          * in the chain of fragments so far.  We must know where to put
 470          * this fragment, right?
 471          */
 472         prev = NULL;
 473         for (next = fq->fragments; next != NULL; next = next->next) {
 474                 if (NFCT_FRAG6_CB(next)->offset >= offset)
 475                         break;  /* bingo! */
 476                 prev = next;
 477         }
 478
 479         /* We found where to put this one.  Check for overlap with
 480          * preceding fragment, and, if needed, align things so that
 481          * any overlaps are eliminated.
 482          */
 483         if (prev) {
 484                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
 485
 486                 if (i > 0) {
 487                         offset += i;
 488                         if (end <= offset) {
 489                                 DEBUGP("overlap\n");
 490                                 goto err;
 491                         }
 492                         if (!pskb_pull(skb, i)) {
 493                                 DEBUGP("Can't pull\n");
 494                                 goto err;
 495                         }
 496                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 497                                 skb->ip_summed = CHECKSUM_NONE;
 498                 }
 499         }
 500
 501         /* Look for overlap with succeeding segments.
 502          * If we can merge fragments, do it.
 503          */
 504         while (next && NFCT_FRAG6_CB(next)->offset < end) {
 505                 /* overlap is 'i' bytes */
 506                 int i = end - NFCT_FRAG6_CB(next)->offset;
 507
 508                 if (i < next->len) {
 509                         /* Eat head of the next overlapped fragment
 510                          * and leave the loop. The next ones cannot overlap.
 511                          */
 512                         DEBUGP("Eat head of the overlapped parts.: %d", i);
 513                         if (!pskb_pull(next, i))
 514                                 goto err;
 515
 516                         /* next fragment */
 517                         NFCT_FRAG6_CB(next)->offset += i;
 518                         fq->meat -= i;
 519                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 520                                 next->ip_summed = CHECKSUM_NONE;
 521                         break;
 522                 } else {
 523                         struct sk_buff *free_it = next;
 524
 525                         /* Old fragmnet is completely overridden with
 526                          * new one drop it.
 527                          */
 528                         next = next->next;
 529
 530                         if (prev)
 531                                 prev->next = next;
 532                         else
 533                                 fq->fragments = next;
 534
 535                         fq->meat -= free_it->len;
 536                         frag_kfree_skb(free_it, NULL);
 537                 }
 538         }
 539
 540         NFCT_FRAG6_CB(skb)->offset = offset;
 541
 542         /* Insert this fragment in the chain of fragments. */
 543         skb->next = next;
 544         if (prev)
 545                 prev->next = skb;
 546         else
 547                 fq->fragments = skb;
 548
 549         skb->dev = NULL;
 550         skb_get_timestamp(skb, &fq->stamp);
 551         fq->meat += skb->len;
 552         atomic_add(skb->truesize, &nf_ct_frag6_mem);
 553
 554         /* The first fragment.
 555          * nhoffset is obtained from the first fragment, of course.
 556          */
 557         if (offset == 0) {
 558                 fq->nhoffset = nhoff;
 559                 fq->last_in |= FIRST_IN;
 560         }
 561         write_lock(&nf_ct_frag6_lock);
 562         list_move_tail(&fq->lru_list, &nf_ct_frag6_lru_list);
 563         write_unlock(&nf_ct_frag6_lock);
 564         return 0;
 565
 566 err:
 567         return -1;
 568 }
 569
 570 /*
 571  *      Check if this packet is complete.
 572  *      Returns NULL on failure by any reason, and pointer
 573  *      to current nexthdr field in reassembled frame.
 574  *
 575  *      It is called with locked fq, and caller must check that
 576  *      queue is eligible for reassembly i.e. it is not COMPLETE,
 577  *      the last and the first frames arrived and all the bits are here.
 578  */
 579 static struct sk_buff *
 580 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
 581 {
 582         struct sk_buff *fp, *op, *head = fq->fragments;
 583         int    payload_len;
 584
 585         fq_kill(fq);
 586
 587         BUG_TRAP(head != NULL);
 588         BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
 589
 590         /* Unfragmented part is taken from the first segment. */
 591         payload_len = (head->data - head->nh.raw) - sizeof(struct ipv6hdr) + fq->len - sizeof(struct frag_hdr);
 592         if (payload_len > IPV6_MAXPLEN) {
 593                 DEBUGP("payload len is too large.\n");
 594                 goto out_oversize;
 595         }
 596
 597         /* Head of list must not be cloned. */
 598         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
 599                 DEBUGP("skb is cloned but can't expand head");
 600                 goto out_oom;
 601         }
 602
 603         /* If the first fragment is fragmented itself, we split
 604          * it to two chunks: the first with data and paged part
 605          * and the second, holding only fragments. */
 606         if (skb_shinfo(head)->frag_list) {
 607                 struct sk_buff *clone;
 608                 int i, plen = 0;
 609
 610                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
 611                         DEBUGP("Can't alloc skb\n");
 612                         goto out_oom;
 613                 }
 614                 clone->next = head->next;
 615                 head->next = clone;
 616                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 617                 skb_shinfo(head)->frag_list = NULL;
 618                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 619                         plen += skb_shinfo(head)->frags[i].size;
 620                 clone->len = clone->data_len = head->data_len - plen;
 621                 head->data_len -= clone->len;
 622                 head->len -= clone->len;
 623                 clone->csum = 0;
 624                 clone->ip_summed = head->ip_summed;
 625
 626                 NFCT_FRAG6_CB(clone)->orig = NULL;
 627                 atomic_add(clone->truesize, &nf_ct_frag6_mem);
 628         }
 629
 630         /* We have to remove fragment header from datagram and to relocate
 631          * header in order to calculate ICV correctly. */
 632         head->nh.raw[fq->nhoffset] = head->h.raw[0];
 633         memmove(head->head + sizeof(struct frag_hdr), head->head,
 634                 (head->data - head->head) - sizeof(struct frag_hdr));
 635         head->mac.raw += sizeof(struct frag_hdr);
 636         head->nh.raw += sizeof(struct frag_hdr);
 637
 638         skb_shinfo(head)->frag_list = head->next;
 639         head->h.raw = head->data;
 640         skb_push(head, head->data - head->nh.raw);
 641         atomic_sub(head->truesize, &nf_ct_frag6_mem);
 642
 643         for (fp=head->next; fp; fp = fp->next) {
 644                 head->data_len += fp->len;
 645                 head->len += fp->len;
 646                 if (head->ip_summed != fp->ip_summed)
 647                         head->ip_summed = CHECKSUM_NONE;
 648                 else if (head->ip_summed == CHECKSUM_HW)
 649                         head->csum = csum_add(head->csum, fp->csum);
 650                 head->truesize += fp->truesize;
 651                 atomic_sub(fp->truesize, &nf_ct_frag6_mem);
 652         }
 653
 654         head->next = NULL;
 655         head->dev = dev;
 656         skb_set_timestamp(head, &fq->stamp);
 657         head->nh.ipv6h->payload_len = htons(payload_len);
 658
 659         /* Yes, and fold redundant checksum back. 8) */
 660         if (head->ip_summed == CHECKSUM_HW)
 661                 head->csum = csum_partial(head->nh.raw, head->h.raw-head->nh.raw, head->csum);
 662
 663         fq->fragments = NULL;
 664
 665         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
 666         fp = skb_shinfo(head)->frag_list;
 667         if (NFCT_FRAG6_CB(fp)->orig == NULL)
 668                 /* at above code, head skb is divided into two skbs. */
 669                 fp = fp->next;
 670
 671         op = NFCT_FRAG6_CB(head)->orig;
 672         for (; fp; fp = fp->next) {
 673                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
 674
 675                 op->next = orig;
 676                 op = orig;
 677                 NFCT_FRAG6_CB(fp)->orig = NULL;
 678         }
 679
 680         return head;
 681
 682 out_oversize:
 683         if (net_ratelimit())
 684                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
 685         goto out_fail;
 686 out_oom:
 687         if (net_ratelimit())
 688                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
 689 out_fail:
 690         return NULL;
 691 }
 692
 693 /*
 694  * find the header just before Fragment Header.
 695  *
 696  * if success return 0 and set ...
 697  * (*prevhdrp): the value of "Next Header Field" in the header
 698  *              just before Fragment Header.
 699  * (*prevhoff): the offset of "Next Header Field" in the header
 700  *              just before Fragment Header.
 701  * (*fhoff)   : the offset of Fragment Header.
 702  *
 703  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
 704  *
 705  */
 706 static int
 707 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
 708 {
 709         u8 nexthdr = skb->nh.ipv6h->nexthdr;
 710         u8 prev_nhoff = (u8 *)&skb->nh.ipv6h->nexthdr - skb->data;
 711         int start = (u8 *)(skb->nh.ipv6h+1) - skb->data;
 712         int len = skb->len - start;
 713         u8 prevhdr = NEXTHDR_IPV6;
 714
 715         while (nexthdr != NEXTHDR_FRAGMENT) {
 716                 struct ipv6_opt_hdr hdr;
 717                 int hdrlen;
 718
 719                 if (!ipv6_ext_hdr(nexthdr)) {
 720                         return -1;
 721                 }
 722                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
 723                         DEBUGP("too short\n");
 724                         return -1;
 725                 }
 726                 if (nexthdr == NEXTHDR_NONE) {
 727                         DEBUGP("next header is none\n");
 728                         return -1;
 729                 }
 730                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
 731                         BUG();
 732                 if (nexthdr == NEXTHDR_AUTH)
 733                         hdrlen = (hdr.hdrlen+2)<<2;
 734                 else
 735                         hdrlen = ipv6_optlen(&hdr);
 736
 737                 prevhdr = nexthdr;
 738                 prev_nhoff = start;
 739
 740                 nexthdr = hdr.nexthdr;
 741                 len -= hdrlen;
 742                 start += hdrlen;
 743         }
 744
 745         if (len < 0)
 746                 return -1;
 747
 748         *prevhdrp = prevhdr;
 749         *prevhoff = prev_nhoff;
 750         *fhoff = start;
 751
 752         return 0;
 753 }
 754
 755 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
 756 {
 757         struct sk_buff *clone;
 758         struct net_device *dev = skb->dev;
 759         struct frag_hdr *fhdr;
 760         struct nf_ct_frag6_queue *fq;
 761         struct ipv6hdr *hdr;
 762         int fhoff, nhoff;
 763         u8 prevhdr;
 764         struct sk_buff *ret_skb = NULL;
 765
 766         /* Jumbo payload inhibits frag. header */
 767         if (skb->nh.ipv6h->payload_len == 0) {
 768                 DEBUGP("payload len = 0\n");
 769                 return skb;
 770         }
 771
 772         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
 773                 return skb;
 774
 775         clone = skb_clone(skb, GFP_ATOMIC);
 776         if (clone == NULL) {
 777                 DEBUGP("Can't clone skb\n");
 778                 return skb;
 779         }
 780
 781         NFCT_FRAG6_CB(clone)->orig = skb;
 782
 783         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
 784                 DEBUGP("message is too short.\n");
 785                 goto ret_orig;
 786         }
 787
 788         clone->h.raw = clone->data + fhoff;
 789         hdr = clone->nh.ipv6h;
 790         fhdr = (struct frag_hdr *)clone->h.raw;
 791
 792         if (!(fhdr->frag_off & htons(0xFFF9))) {
 793                 DEBUGP("Invalid fragment offset\n");
 794                 /* It is not a fragmented frame */
 795                 goto ret_orig;
 796         }
 797
 798         if (atomic_read(&nf_ct_frag6_mem) > nf_ct_frag6_high_thresh)
 799                 nf_ct_frag6_evictor();
 800
 801         fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
 802         if (fq == NULL) {
 803                 DEBUGP("Can't find and can't create new queue\n");
 804                 goto ret_orig;
 805         }
 806
 807         spin_lock(&fq->lock);
 808
 809         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
 810                 spin_unlock(&fq->lock);
 811                 DEBUGP("Can't insert skb to queue\n");
 812                 fq_put(fq, NULL);
 813                 goto ret_orig;
 814         }
 815
 816         if (fq->last_in == (FIRST_IN|LAST_IN) && fq->meat == fq->len) {
 817                 ret_skb = nf_ct_frag6_reasm(fq, dev);
 818                 if (ret_skb == NULL)
 819                         DEBUGP("Can't reassemble fragmented packets\n");
 820         }
 821         spin_unlock(&fq->lock);
 822
 823         fq_put(fq, NULL);
 824         return ret_skb;
 825
 826 ret_orig:
 827         kfree_skb(clone);
 828         return skb;
 829 }
 830
 831 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
 832                         struct net_device *in, struct net_device *out,
 833                         int (*okfn)(struct sk_buff *))
 834 {
 835         struct sk_buff *s, *s2;
 836
 837         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
 838                 nf_conntrack_put_reasm(s->nfct_reasm);
 839                 nf_conntrack_get_reasm(skb);
 840                 s->nfct_reasm = skb;
 841
 842                 s2 = s->next;
 843                 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
 844                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
 845                 s = s2;
 846         }
 847         nf_conntrack_put_reasm(skb);
 848 }
 849
 850 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
 851 {
 852         struct sk_buff *s, *s2;
 853
 854         for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
 855
 856                 s2 = s->next;
 857                 kfree_skb(s);
 858         }
 859
 860         kfree_skb(skb);
 861
 862         return 0;
 863 }
 864
 865 int nf_ct_frag6_init(void)
 866 {
 867         nf_ct_frag6_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
 868                                    (jiffies ^ (jiffies >> 6)));
 869
 870         init_timer(&nf_ct_frag6_secret_timer);
 871         nf_ct_frag6_secret_timer.function = nf_ct_frag6_secret_rebuild;
 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 }