1 /* 2 * IPv6 fragment reassembly 3 * Linux INET6 implementation 4 * 5 * Authors: 6 * Pedro Roque <roque@di.fc.ul.pt> 7 * 8 * $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $ 9 * 10 * Based on: net/ipv4/ip_fragment.c 11 * 12 * This program is free software; you can redistribute it and/or 13 * modify it under the terms of the GNU General Public License 14 * as published by the Free Software Foundation; either version 15 * 2 of the License, or (at your option) any later version. 16 */ 17 18 /* 19 * Fixes: 20 * Andi Kleen Make it work with multiple hosts. 21 * More RFC compliance. 22 * 23 * Horst von Brand Add missing #include <linux/string.h> 24 * Alexey Kuznetsov SMP races, threading, cleanup. 25 * Patrick McHardy LRU queue of frag heads for evictor. 26 * Mitsuru KANDA @USAGI Register inet6_protocol{}. 27 * David Stevens and 28 * YOSHIFUJI,H. @USAGI Always remove fragment header to 29 * calculate ICV correctly. 30 */ 31 #include <linux/errno.h> 32 #include <linux/types.h> 33 #include <linux/string.h> 34 #include <linux/socket.h> 35 #include <linux/sockios.h> 36 #include <linux/jiffies.h> 37 #include <linux/net.h> 38 #include <linux/list.h> 39 #include <linux/netdevice.h> 40 #include <linux/in6.h> 41 #include <linux/ipv6.h> 42 #include <linux/icmpv6.h> 43 #include <linux/random.h> 44 #include <linux/jhash.h> 45 46 #include <net/sock.h> 47 #include <net/snmp.h> 48 49 #include <net/ipv6.h> 50 #include <net/ip6_route.h> 51 #include <net/protocol.h> 52 #include <net/transp_v6.h> 53 #include <net/rawv6.h> 54 #include <net/ndisc.h> 55 #include <net/addrconf.h> 56 57 int sysctl_ip6frag_high_thresh __read_mostly = 256*1024; 58 int sysctl_ip6frag_low_thresh __read_mostly = 192*1024; 59 60 int sysctl_ip6frag_time __read_mostly = IPV6_FRAG_TIMEOUT; 61 62 struct ip6frag_skb_cb 63 { 64 struct inet6_skb_parm h; 65 int offset; 66 }; 67 68 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb)) 69 70 71 /* 72 * Equivalent of ipv4 struct ipq 73 */ 74 75 struct frag_queue 76 { 77 struct hlist_node list; 78 struct list_head lru_list; /* lru list member */ 79 80 __be32 id; /* fragment id */ 81 struct in6_addr saddr; 82 struct in6_addr daddr; 83 84 spinlock_t lock; 85 atomic_t refcnt; 86 struct timer_list timer; /* expire timer */ 87 struct sk_buff *fragments; 88 int len; 89 int meat; 90 int iif; 91 ktime_t stamp; 92 unsigned int csum; 93 __u8 last_in; /* has first/last segment arrived? */ 94 #define COMPLETE 4 95 #define FIRST_IN 2 96 #define LAST_IN 1 97 __u16 nhoffset; 98 }; 99 100 /* Hash table. */ 101 102 #define IP6Q_HASHSZ 64 103 104 static struct hlist_head ip6_frag_hash[IP6Q_HASHSZ]; 105 static DEFINE_RWLOCK(ip6_frag_lock); 106 static u32 ip6_frag_hash_rnd; 107 static LIST_HEAD(ip6_frag_lru_list); 108 int ip6_frag_nqueues = 0; 109 110 static __inline__ void __fq_unlink(struct frag_queue *fq) 111 { 112 hlist_del(&fq->list); 113 list_del(&fq->lru_list); 114 ip6_frag_nqueues--; 115 } 116 117 static __inline__ void fq_unlink(struct frag_queue *fq) 118 { 119 write_lock(&ip6_frag_lock); 120 __fq_unlink(fq); 121 write_unlock(&ip6_frag_lock); 122 } 123 124 /* 125 * callers should be careful not to use the hash value outside the ipfrag_lock 126 * as doing so could race with ipfrag_hash_rnd being recalculated. 127 */ 128 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr, 129 struct in6_addr *daddr) 130 { 131 u32 a, b, c; 132 133 a = (__force u32)saddr->s6_addr32[0]; 134 b = (__force u32)saddr->s6_addr32[1]; 135 c = (__force u32)saddr->s6_addr32[2]; 136 137 a += JHASH_GOLDEN_RATIO; 138 b += JHASH_GOLDEN_RATIO; 139 c += ip6_frag_hash_rnd; 140 __jhash_mix(a, b, c); 141 142 a += (__force u32)saddr->s6_addr32[3]; 143 b += (__force u32)daddr->s6_addr32[0]; 144 c += (__force u32)daddr->s6_addr32[1]; 145 __jhash_mix(a, b, c); 146 147 a += (__force u32)daddr->s6_addr32[2]; 148 b += (__force u32)daddr->s6_addr32[3]; 149 c += (__force u32)id; 150 __jhash_mix(a, b, c); 151 152 return c & (IP6Q_HASHSZ - 1); 153 } 154 155 static struct timer_list ip6_frag_secret_timer; 156 int sysctl_ip6frag_secret_interval __read_mostly = 10 * 60 * HZ; 157 158 static void ip6_frag_secret_rebuild(unsigned long dummy) 159 { 160 unsigned long now = jiffies; 161 int i; 162 163 write_lock(&ip6_frag_lock); 164 get_random_bytes(&ip6_frag_hash_rnd, sizeof(u32)); 165 for (i = 0; i < IP6Q_HASHSZ; i++) { 166 struct frag_queue *q; 167 struct hlist_node *p, *n; 168 169 hlist_for_each_entry_safe(q, p, n, &ip6_frag_hash[i], list) { 170 unsigned int hval = ip6qhashfn(q->id, 171 &q->saddr, 172 &q->daddr); 173 174 if (hval != i) { 175 hlist_del(&q->list); 176 177 /* Relink to new hash chain. */ 178 hlist_add_head(&q->list, 179 &ip6_frag_hash[hval]); 180 181 } 182 } 183 } 184 write_unlock(&ip6_frag_lock); 185 186 mod_timer(&ip6_frag_secret_timer, now + sysctl_ip6frag_secret_interval); 187 } 188 189 atomic_t ip6_frag_mem = ATOMIC_INIT(0); 190 191 /* Memory Tracking Functions. */ 192 static inline void frag_kfree_skb(struct sk_buff *skb, int *work) 193 { 194 if (work) 195 *work -= skb->truesize; 196 atomic_sub(skb->truesize, &ip6_frag_mem); 197 kfree_skb(skb); 198 } 199 200 static inline void frag_free_queue(struct frag_queue *fq, int *work) 201 { 202 if (work) 203 *work -= sizeof(struct frag_queue); 204 atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem); 205 kfree(fq); 206 } 207 208 static inline struct frag_queue *frag_alloc_queue(void) 209 { 210 struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC); 211 212 if(!fq) 213 return NULL; 214 atomic_add(sizeof(struct frag_queue), &ip6_frag_mem); 215 return fq; 216 } 217 218 /* Destruction primitives. */ 219 220 /* Complete destruction of fq. */ 221 static void ip6_frag_destroy(struct frag_queue *fq, int *work) 222 { 223 struct sk_buff *fp; 224 225 BUG_TRAP(fq->last_in&COMPLETE); 226 BUG_TRAP(del_timer(&fq->timer) == 0); 227 228 /* Release all fragment data. */ 229 fp = fq->fragments; 230 while (fp) { 231 struct sk_buff *xp = fp->next; 232 233 frag_kfree_skb(fp, work); 234 fp = xp; 235 } 236 237 frag_free_queue(fq, work); 238 } 239 240 static __inline__ void fq_put(struct frag_queue *fq, int *work) 241 { 242 if (atomic_dec_and_test(&fq->refcnt)) 243 ip6_frag_destroy(fq, work); 244 } 245 246 /* Kill fq entry. It is not destroyed immediately, 247 * because caller (and someone more) holds reference count. 248 */ 249 static __inline__ void fq_kill(struct frag_queue *fq) 250 { 251 if (del_timer(&fq->timer)) 252 atomic_dec(&fq->refcnt); 253 254 if (!(fq->last_in & COMPLETE)) { 255 fq_unlink(fq); 256 atomic_dec(&fq->refcnt); 257 fq->last_in |= COMPLETE; 258 } 259 } 260 261 static void ip6_evictor(struct inet6_dev *idev) 262 { 263 struct frag_queue *fq; 264 struct list_head *tmp; 265 int work; 266 267 work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh; 268 if (work <= 0) 269 return; 270 271 while(work > 0) { 272 read_lock(&ip6_frag_lock); 273 if (list_empty(&ip6_frag_lru_list)) { 274 read_unlock(&ip6_frag_lock); 275 return; 276 } 277 tmp = ip6_frag_lru_list.next; 278 fq = list_entry(tmp, struct frag_queue, lru_list); 279 atomic_inc(&fq->refcnt); 280 read_unlock(&ip6_frag_lock); 281 282 spin_lock(&fq->lock); 283 if (!(fq->last_in&COMPLETE)) 284 fq_kill(fq); 285 spin_unlock(&fq->lock); 286 287 fq_put(fq, &work); 288 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); 289 } 290 } 291 292 static void ip6_frag_expire(unsigned long data) 293 { 294 struct frag_queue *fq = (struct frag_queue *) data; 295 struct net_device *dev = NULL; 296 297 spin_lock(&fq->lock); 298 299 if (fq->last_in & COMPLETE) 300 goto out; 301 302 fq_kill(fq); 303 304 dev = dev_get_by_index(fq->iif); 305 if (!dev) 306 goto out; 307 308 rcu_read_lock(); 309 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT); 310 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); 311 rcu_read_unlock(); 312 313 /* Don't send error if the first segment did not arrive. */ 314 if (!(fq->last_in&FIRST_IN) || !fq->fragments) 315 goto out; 316 317 /* 318 But use as source device on which LAST ARRIVED 319 segment was received. And do not use fq->dev 320 pointer directly, device might already disappeared. 321 */ 322 fq->fragments->dev = dev; 323 icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev); 324 out: 325 if (dev) 326 dev_put(dev); 327 spin_unlock(&fq->lock); 328 fq_put(fq, NULL); 329 } 330 331 /* Creation primitives. */ 332 333 334 static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in) 335 { 336 struct frag_queue *fq; 337 unsigned int hash; 338 #ifdef CONFIG_SMP 339 struct hlist_node *n; 340 #endif 341 342 write_lock(&ip6_frag_lock); 343 hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr); 344 #ifdef CONFIG_SMP 345 hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) { 346 if (fq->id == fq_in->id && 347 ipv6_addr_equal(&fq_in->saddr, &fq->saddr) && 348 ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) { 349 atomic_inc(&fq->refcnt); 350 write_unlock(&ip6_frag_lock); 351 fq_in->last_in |= COMPLETE; 352 fq_put(fq_in, NULL); 353 return fq; 354 } 355 } 356 #endif 357 fq = fq_in; 358 359 if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time)) 360 atomic_inc(&fq->refcnt); 361 362 atomic_inc(&fq->refcnt); 363 hlist_add_head(&fq->list, &ip6_frag_hash[hash]); 364 INIT_LIST_HEAD(&fq->lru_list); 365 list_add_tail(&fq->lru_list, &ip6_frag_lru_list); 366 ip6_frag_nqueues++; 367 write_unlock(&ip6_frag_lock); 368 return fq; 369 } 370 371 372 static struct frag_queue * 373 ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst, 374 struct inet6_dev *idev) 375 { 376 struct frag_queue *fq; 377 378 if ((fq = frag_alloc_queue()) == NULL) 379 goto oom; 380 381 fq->id = id; 382 ipv6_addr_copy(&fq->saddr, src); 383 ipv6_addr_copy(&fq->daddr, dst); 384 385 init_timer(&fq->timer); 386 fq->timer.function = ip6_frag_expire; 387 fq->timer.data = (long) fq; 388 spin_lock_init(&fq->lock); 389 atomic_set(&fq->refcnt, 1); 390 391 return ip6_frag_intern(fq); 392 393 oom: 394 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); 395 return NULL; 396 } 397 398 static __inline__ struct frag_queue * 399 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst, 400 struct inet6_dev *idev) 401 { 402 struct frag_queue *fq; 403 struct hlist_node *n; 404 unsigned int hash; 405 406 read_lock(&ip6_frag_lock); 407 hash = ip6qhashfn(id, src, dst); 408 hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) { 409 if (fq->id == id && 410 ipv6_addr_equal(src, &fq->saddr) && 411 ipv6_addr_equal(dst, &fq->daddr)) { 412 atomic_inc(&fq->refcnt); 413 read_unlock(&ip6_frag_lock); 414 return fq; 415 } 416 } 417 read_unlock(&ip6_frag_lock); 418 419 return ip6_frag_create(id, src, dst, idev); 420 } 421 422 423 static void ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb, 424 struct frag_hdr *fhdr, int nhoff) 425 { 426 struct sk_buff *prev, *next; 427 int offset, end; 428 429 if (fq->last_in & COMPLETE) 430 goto err; 431 432 offset = ntohs(fhdr->frag_off) & ~0x7; 433 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) - 434 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); 435 436 if ((unsigned int)end > IPV6_MAXPLEN) { 437 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), 438 IPSTATS_MIB_INHDRERRORS); 439 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, 440 ((u8 *)&fhdr->frag_off - 441 skb_network_header(skb))); 442 return; 443 } 444 445 if (skb->ip_summed == CHECKSUM_COMPLETE) { 446 const unsigned char *nh = skb_network_header(skb); 447 skb->csum = csum_sub(skb->csum, 448 csum_partial(nh, (u8 *)(fhdr + 1) - nh, 449 0)); 450 } 451 452 /* Is this the final fragment? */ 453 if (!(fhdr->frag_off & htons(IP6_MF))) { 454 /* If we already have some bits beyond end 455 * or have different end, the segment is corrupted. 456 */ 457 if (end < fq->len || 458 ((fq->last_in & LAST_IN) && end != fq->len)) 459 goto err; 460 fq->last_in |= LAST_IN; 461 fq->len = end; 462 } else { 463 /* Check if the fragment is rounded to 8 bytes. 464 * Required by the RFC. 465 */ 466 if (end & 0x7) { 467 /* RFC2460 says always send parameter problem in 468 * this case. -DaveM 469 */ 470 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), 471 IPSTATS_MIB_INHDRERRORS); 472 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, 473 offsetof(struct ipv6hdr, payload_len)); 474 return; 475 } 476 if (end > fq->len) { 477 /* Some bits beyond end -> corruption. */ 478 if (fq->last_in & LAST_IN) 479 goto err; 480 fq->len = end; 481 } 482 } 483 484 if (end == offset) 485 goto err; 486 487 /* Point into the IP datagram 'data' part. */ 488 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) 489 goto err; 490 491 if (pskb_trim_rcsum(skb, end - offset)) 492 goto err; 493 494 /* Find out which fragments are in front and at the back of us 495 * in the chain of fragments so far. We must know where to put 496 * this fragment, right? 497 */ 498 prev = NULL; 499 for(next = fq->fragments; next != NULL; next = next->next) { 500 if (FRAG6_CB(next)->offset >= offset) 501 break; /* bingo! */ 502 prev = next; 503 } 504 505 /* We found where to put this one. Check for overlap with 506 * preceding fragment, and, if needed, align things so that 507 * any overlaps are eliminated. 508 */ 509 if (prev) { 510 int i = (FRAG6_CB(prev)->offset + prev->len) - offset; 511 512 if (i > 0) { 513 offset += i; 514 if (end <= offset) 515 goto err; 516 if (!pskb_pull(skb, i)) 517 goto err; 518 if (skb->ip_summed != CHECKSUM_UNNECESSARY) 519 skb->ip_summed = CHECKSUM_NONE; 520 } 521 } 522 523 /* Look for overlap with succeeding segments. 524 * If we can merge fragments, do it. 525 */ 526 while (next && FRAG6_CB(next)->offset < end) { 527 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */ 528 529 if (i < next->len) { 530 /* Eat head of the next overlapped fragment 531 * and leave the loop. The next ones cannot overlap. 532 */ 533 if (!pskb_pull(next, i)) 534 goto err; 535 FRAG6_CB(next)->offset += i; /* next fragment */ 536 fq->meat -= i; 537 if (next->ip_summed != CHECKSUM_UNNECESSARY) 538 next->ip_summed = CHECKSUM_NONE; 539 break; 540 } else { 541 struct sk_buff *free_it = next; 542 543 /* Old fragment is completely overridden with 544 * new one drop it. 545 */ 546 next = next->next; 547 548 if (prev) 549 prev->next = next; 550 else 551 fq->fragments = next; 552 553 fq->meat -= free_it->len; 554 frag_kfree_skb(free_it, NULL); 555 } 556 } 557 558 FRAG6_CB(skb)->offset = offset; 559 560 /* Insert this fragment in the chain of fragments. */ 561 skb->next = next; 562 if (prev) 563 prev->next = skb; 564 else 565 fq->fragments = skb; 566 567 if (skb->dev) 568 fq->iif = skb->dev->ifindex; 569 skb->dev = NULL; 570 fq->stamp = skb->tstamp; 571 fq->meat += skb->len; 572 atomic_add(skb->truesize, &ip6_frag_mem); 573 574 /* The first fragment. 575 * nhoffset is obtained from the first fragment, of course. 576 */ 577 if (offset == 0) { 578 fq->nhoffset = nhoff; 579 fq->last_in |= FIRST_IN; 580 } 581 write_lock(&ip6_frag_lock); 582 list_move_tail(&fq->lru_list, &ip6_frag_lru_list); 583 write_unlock(&ip6_frag_lock); 584 return; 585 586 err: 587 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); 588 kfree_skb(skb); 589 } 590 591 /* 592 * Check if this packet is complete. 593 * Returns NULL on failure by any reason, and pointer 594 * to current nexthdr field in reassembled frame. 595 * 596 * It is called with locked fq, and caller must check that 597 * queue is eligible for reassembly i.e. it is not COMPLETE, 598 * the last and the first frames arrived and all the bits are here. 599 */ 600 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff **skb_in, 601 struct net_device *dev) 602 { 603 struct sk_buff *fp, *head = fq->fragments; 604 int payload_len; 605 unsigned int nhoff; 606 607 fq_kill(fq); 608 609 BUG_TRAP(head != NULL); 610 BUG_TRAP(FRAG6_CB(head)->offset == 0); 611 612 /* Unfragmented part is taken from the first segment. */ 613 payload_len = ((head->data - skb_network_header(head)) - 614 sizeof(struct ipv6hdr) + fq->len - 615 sizeof(struct frag_hdr)); 616 if (payload_len > IPV6_MAXPLEN) 617 goto out_oversize; 618 619 /* Head of list must not be cloned. */ 620 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) 621 goto out_oom; 622 623 /* If the first fragment is fragmented itself, we split 624 * it to two chunks: the first with data and paged part 625 * and the second, holding only fragments. */ 626 if (skb_shinfo(head)->frag_list) { 627 struct sk_buff *clone; 628 int i, plen = 0; 629 630 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) 631 goto out_oom; 632 clone->next = head->next; 633 head->next = clone; 634 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; 635 skb_shinfo(head)->frag_list = NULL; 636 for (i=0; i<skb_shinfo(head)->nr_frags; i++) 637 plen += skb_shinfo(head)->frags[i].size; 638 clone->len = clone->data_len = head->data_len - plen; 639 head->data_len -= clone->len; 640 head->len -= clone->len; 641 clone->csum = 0; 642 clone->ip_summed = head->ip_summed; 643 atomic_add(clone->truesize, &ip6_frag_mem); 644 } 645 646 /* We have to remove fragment header from datagram and to relocate 647 * header in order to calculate ICV correctly. */ 648 nhoff = fq->nhoffset; 649 skb_network_header(head)[nhoff] = skb_transport_header(head)[0]; 650 memmove(head->head + sizeof(struct frag_hdr), head->head, 651 (head->data - head->head) - sizeof(struct frag_hdr)); 652 head->mac_header += sizeof(struct frag_hdr); 653 head->network_header += sizeof(struct frag_hdr); 654 655 skb_shinfo(head)->frag_list = head->next; 656 skb_reset_transport_header(head); 657 skb_push(head, head->data - skb_network_header(head)); 658 atomic_sub(head->truesize, &ip6_frag_mem); 659 660 for (fp=head->next; fp; fp = fp->next) { 661 head->data_len += fp->len; 662 head->len += fp->len; 663 if (head->ip_summed != fp->ip_summed) 664 head->ip_summed = CHECKSUM_NONE; 665 else if (head->ip_summed == CHECKSUM_COMPLETE) 666 head->csum = csum_add(head->csum, fp->csum); 667 head->truesize += fp->truesize; 668 atomic_sub(fp->truesize, &ip6_frag_mem); 669 } 670 671 head->next = NULL; 672 head->dev = dev; 673 head->tstamp = fq->stamp; 674 ipv6_hdr(head)->payload_len = htons(payload_len); 675 IP6CB(head)->nhoff = nhoff; 676 677 *skb_in = head; 678 679 /* Yes, and fold redundant checksum back. 8) */ 680 if (head->ip_summed == CHECKSUM_COMPLETE) 681 head->csum = csum_partial(skb_network_header(head), 682 skb_network_header_len(head), 683 head->csum); 684 685 rcu_read_lock(); 686 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS); 687 rcu_read_unlock(); 688 fq->fragments = NULL; 689 return 1; 690 691 out_oversize: 692 if (net_ratelimit()) 693 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len); 694 goto out_fail; 695 out_oom: 696 if (net_ratelimit()) 697 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n"); 698 out_fail: 699 rcu_read_lock(); 700 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); 701 rcu_read_unlock(); 702 return -1; 703 } 704 705 static int ipv6_frag_rcv(struct sk_buff **skbp) 706 { 707 struct sk_buff *skb = *skbp; 708 struct net_device *dev = skb->dev; 709 struct frag_hdr *fhdr; 710 struct frag_queue *fq; 711 struct ipv6hdr *hdr = ipv6_hdr(skb); 712 713 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS); 714 715 /* Jumbo payload inhibits frag. header */ 716 if (hdr->payload_len==0) { 717 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); 718 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, 719 skb_network_header_len(skb)); 720 return -1; 721 } 722 if (!pskb_may_pull(skb, (skb_transport_offset(skb) + 723 sizeof(struct frag_hdr)))) { 724 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); 725 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, 726 skb_network_header_len(skb)); 727 return -1; 728 } 729 730 hdr = ipv6_hdr(skb); 731 fhdr = (struct frag_hdr *)skb_transport_header(skb); 732 733 if (!(fhdr->frag_off & htons(0xFFF9))) { 734 /* It is not a fragmented frame */ 735 skb->transport_header += sizeof(struct frag_hdr); 736 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS); 737 738 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb); 739 return 1; 740 } 741 742 if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh) 743 ip6_evictor(ip6_dst_idev(skb->dst)); 744 745 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr, 746 ip6_dst_idev(skb->dst))) != NULL) { 747 int ret = -1; 748 749 spin_lock(&fq->lock); 750 751 ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff); 752 753 if (fq->last_in == (FIRST_IN|LAST_IN) && 754 fq->meat == fq->len) 755 ret = ip6_frag_reasm(fq, skbp, dev); 756 757 spin_unlock(&fq->lock); 758 fq_put(fq, NULL); 759 return ret; 760 } 761 762 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); 763 kfree_skb(skb); 764 return -1; 765 } 766 767 static struct inet6_protocol frag_protocol = 768 { 769 .handler = ipv6_frag_rcv, 770 .flags = INET6_PROTO_NOPOLICY, 771 }; 772 773 void __init ipv6_frag_init(void) 774 { 775 if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0) 776 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n"); 777 778 ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ 779 (jiffies ^ (jiffies >> 6))); 780 781 init_timer(&ip6_frag_secret_timer); 782 ip6_frag_secret_timer.function = ip6_frag_secret_rebuild; 783 ip6_frag_secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval; 784 add_timer(&ip6_frag_secret_timer); 785 } 786