1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The IP fragmentation functionality. 7 * 8 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG> 9 * Alan Cox <alan@lxorguk.ukuu.org.uk> 10 * 11 * Fixes: 12 * Alan Cox : Split from ip.c , see ip_input.c for history. 13 * David S. Miller : Begin massive cleanup... 14 * Andi Kleen : Add sysctls. 15 * xxxx : Overlapfrag bug. 16 * Ultima : ip_expire() kernel panic. 17 * Bill Hawes : Frag accounting and evictor fixes. 18 * John McDonald : 0 length frag bug. 19 * Alexey Kuznetsov: SMP races, threading, cleanup. 20 * Patrick McHardy : LRU queue of frag heads for evictor. 21 */ 22 23 #define pr_fmt(fmt) "IPv4: " fmt 24 25 #include <linux/compiler.h> 26 #include <linux/module.h> 27 #include <linux/types.h> 28 #include <linux/mm.h> 29 #include <linux/jiffies.h> 30 #include <linux/skbuff.h> 31 #include <linux/list.h> 32 #include <linux/ip.h> 33 #include <linux/icmp.h> 34 #include <linux/netdevice.h> 35 #include <linux/jhash.h> 36 #include <linux/random.h> 37 #include <linux/slab.h> 38 #include <net/route.h> 39 #include <net/dst.h> 40 #include <net/sock.h> 41 #include <net/ip.h> 42 #include <net/icmp.h> 43 #include <net/checksum.h> 44 #include <net/inetpeer.h> 45 #include <net/inet_frag.h> 46 #include <linux/tcp.h> 47 #include <linux/udp.h> 48 #include <linux/inet.h> 49 #include <linux/netfilter_ipv4.h> 50 #include <net/inet_ecn.h> 51 52 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 53 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c 54 * as well. Or notify me, at least. --ANK 55 */ 56 57 static int sysctl_ipfrag_max_dist __read_mostly = 64; 58 59 struct ipfrag_skb_cb 60 { 61 struct inet_skb_parm h; 62 int offset; 63 }; 64 65 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb)) 66 67 /* Describe an entry in the "incomplete datagrams" queue. */ 68 struct ipq { 69 struct inet_frag_queue q; 70 71 u32 user; 72 __be32 saddr; 73 __be32 daddr; 74 __be16 id; 75 u8 protocol; 76 u8 ecn; /* RFC3168 support */ 77 int iif; 78 unsigned int rid; 79 struct inet_peer *peer; 80 }; 81 82 /* RFC 3168 support : 83 * We want to check ECN values of all fragments, do detect invalid combinations. 84 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value. 85 */ 86 #define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */ 87 #define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */ 88 #define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */ 89 #define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */ 90 91 static inline u8 ip4_frag_ecn(u8 tos) 92 { 93 return 1 << (tos & INET_ECN_MASK); 94 } 95 96 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements 97 * Value : 0xff if frame should be dropped. 98 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field 99 */ 100 static const u8 ip4_frag_ecn_table[16] = { 101 /* at least one fragment had CE, and others ECT_0 or ECT_1 */ 102 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE, 103 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE, 104 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE, 105 106 /* invalid combinations : drop frame */ 107 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff, 108 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff, 109 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff, 110 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, 111 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff, 112 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff, 113 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff, 114 }; 115 116 static struct inet_frags ip4_frags; 117 118 int ip_frag_nqueues(struct net *net) 119 { 120 return net->ipv4.frags.nqueues; 121 } 122 123 int ip_frag_mem(struct net *net) 124 { 125 return sum_frag_mem_limit(&net->ipv4.frags); 126 } 127 128 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, 129 struct net_device *dev); 130 131 struct ip4_create_arg { 132 struct iphdr *iph; 133 u32 user; 134 }; 135 136 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) 137 { 138 return jhash_3words((__force u32)id << 16 | prot, 139 (__force u32)saddr, (__force u32)daddr, 140 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1); 141 } 142 143 static unsigned int ip4_hashfn(struct inet_frag_queue *q) 144 { 145 struct ipq *ipq; 146 147 ipq = container_of(q, struct ipq, q); 148 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol); 149 } 150 151 static bool ip4_frag_match(struct inet_frag_queue *q, void *a) 152 { 153 struct ipq *qp; 154 struct ip4_create_arg *arg = a; 155 156 qp = container_of(q, struct ipq, q); 157 return qp->id == arg->iph->id && 158 qp->saddr == arg->iph->saddr && 159 qp->daddr == arg->iph->daddr && 160 qp->protocol == arg->iph->protocol && 161 qp->user == arg->user; 162 } 163 164 static void ip4_frag_init(struct inet_frag_queue *q, void *a) 165 { 166 struct ipq *qp = container_of(q, struct ipq, q); 167 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4, 168 frags); 169 struct net *net = container_of(ipv4, struct net, ipv4); 170 171 struct ip4_create_arg *arg = a; 172 173 qp->protocol = arg->iph->protocol; 174 qp->id = arg->iph->id; 175 qp->ecn = ip4_frag_ecn(arg->iph->tos); 176 qp->saddr = arg->iph->saddr; 177 qp->daddr = arg->iph->daddr; 178 qp->user = arg->user; 179 qp->peer = sysctl_ipfrag_max_dist ? 180 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL; 181 } 182 183 static __inline__ void ip4_frag_free(struct inet_frag_queue *q) 184 { 185 struct ipq *qp; 186 187 qp = container_of(q, struct ipq, q); 188 if (qp->peer) 189 inet_putpeer(qp->peer); 190 } 191 192 193 /* Destruction primitives. */ 194 195 static __inline__ void ipq_put(struct ipq *ipq) 196 { 197 inet_frag_put(&ipq->q, &ip4_frags); 198 } 199 200 /* Kill ipq entry. It is not destroyed immediately, 201 * because caller (and someone more) holds reference count. 202 */ 203 static void ipq_kill(struct ipq *ipq) 204 { 205 inet_frag_kill(&ipq->q, &ip4_frags); 206 } 207 208 /* Memory limiting on fragments. Evictor trashes the oldest 209 * fragment queue until we are back under the threshold. 210 */ 211 static void ip_evictor(struct net *net) 212 { 213 int evicted; 214 215 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false); 216 if (evicted) 217 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted); 218 } 219 220 /* 221 * Oops, a fragment queue timed out. Kill it and send an ICMP reply. 222 */ 223 static void ip_expire(unsigned long arg) 224 { 225 struct ipq *qp; 226 struct net *net; 227 228 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q); 229 net = container_of(qp->q.net, struct net, ipv4.frags); 230 231 spin_lock(&qp->q.lock); 232 233 if (qp->q.last_in & INET_FRAG_COMPLETE) 234 goto out; 235 236 ipq_kill(qp); 237 238 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT); 239 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); 240 241 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) { 242 struct sk_buff *head = qp->q.fragments; 243 const struct iphdr *iph; 244 int err; 245 246 rcu_read_lock(); 247 head->dev = dev_get_by_index_rcu(net, qp->iif); 248 if (!head->dev) 249 goto out_rcu_unlock; 250 251 /* skb dst is stale, drop it, and perform route lookup again */ 252 skb_dst_drop(head); 253 iph = ip_hdr(head); 254 err = ip_route_input_noref(head, iph->daddr, iph->saddr, 255 iph->tos, head->dev); 256 if (err) 257 goto out_rcu_unlock; 258 259 /* 260 * Only an end host needs to send an ICMP 261 * "Fragment Reassembly Timeout" message, per RFC792. 262 */ 263 if (qp->user == IP_DEFRAG_AF_PACKET || 264 (qp->user == IP_DEFRAG_CONNTRACK_IN && 265 skb_rtable(head)->rt_type != RTN_LOCAL)) 266 goto out_rcu_unlock; 267 268 269 /* Send an ICMP "Fragment Reassembly Timeout" message. */ 270 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); 271 out_rcu_unlock: 272 rcu_read_unlock(); 273 } 274 out: 275 spin_unlock(&qp->q.lock); 276 ipq_put(qp); 277 } 278 279 /* Find the correct entry in the "incomplete datagrams" queue for 280 * this IP datagram, and create new one, if nothing is found. 281 */ 282 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) 283 { 284 struct inet_frag_queue *q; 285 struct ip4_create_arg arg; 286 unsigned int hash; 287 288 arg.iph = iph; 289 arg.user = user; 290 291 read_lock(&ip4_frags.lock); 292 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); 293 294 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); 295 if (q == NULL) 296 goto out_nomem; 297 298 return container_of(q, struct ipq, q); 299 300 out_nomem: 301 LIMIT_NETDEBUG(KERN_ERR pr_fmt("ip_frag_create: no memory left !\n")); 302 return NULL; 303 } 304 305 /* Is the fragment too far ahead to be part of ipq? */ 306 static inline int ip_frag_too_far(struct ipq *qp) 307 { 308 struct inet_peer *peer = qp->peer; 309 unsigned int max = sysctl_ipfrag_max_dist; 310 unsigned int start, end; 311 312 int rc; 313 314 if (!peer || !max) 315 return 0; 316 317 start = qp->rid; 318 end = atomic_inc_return(&peer->rid); 319 qp->rid = end; 320 321 rc = qp->q.fragments && (end - start) > max; 322 323 if (rc) { 324 struct net *net; 325 326 net = container_of(qp->q.net, struct net, ipv4.frags); 327 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); 328 } 329 330 return rc; 331 } 332 333 static int ip_frag_reinit(struct ipq *qp) 334 { 335 struct sk_buff *fp; 336 unsigned int sum_truesize = 0; 337 338 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { 339 atomic_inc(&qp->q.refcnt); 340 return -ETIMEDOUT; 341 } 342 343 fp = qp->q.fragments; 344 do { 345 struct sk_buff *xp = fp->next; 346 347 sum_truesize += fp->truesize; 348 kfree_skb(fp); 349 fp = xp; 350 } while (fp); 351 sub_frag_mem_limit(&qp->q, sum_truesize); 352 353 qp->q.last_in = 0; 354 qp->q.len = 0; 355 qp->q.meat = 0; 356 qp->q.fragments = NULL; 357 qp->q.fragments_tail = NULL; 358 qp->iif = 0; 359 qp->ecn = 0; 360 361 return 0; 362 } 363 364 /* Add new segment to existing queue. */ 365 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) 366 { 367 struct sk_buff *prev, *next; 368 struct net_device *dev; 369 int flags, offset; 370 int ihl, end; 371 int err = -ENOENT; 372 u8 ecn; 373 374 if (qp->q.last_in & INET_FRAG_COMPLETE) 375 goto err; 376 377 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && 378 unlikely(ip_frag_too_far(qp)) && 379 unlikely(err = ip_frag_reinit(qp))) { 380 ipq_kill(qp); 381 goto err; 382 } 383 384 ecn = ip4_frag_ecn(ip_hdr(skb)->tos); 385 offset = ntohs(ip_hdr(skb)->frag_off); 386 flags = offset & ~IP_OFFSET; 387 offset &= IP_OFFSET; 388 offset <<= 3; /* offset is in 8-byte chunks */ 389 ihl = ip_hdrlen(skb); 390 391 /* Determine the position of this fragment. */ 392 end = offset + skb->len - ihl; 393 err = -EINVAL; 394 395 /* Is this the final fragment? */ 396 if ((flags & IP_MF) == 0) { 397 /* If we already have some bits beyond end 398 * or have different end, the segment is corrupted. 399 */ 400 if (end < qp->q.len || 401 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len)) 402 goto err; 403 qp->q.last_in |= INET_FRAG_LAST_IN; 404 qp->q.len = end; 405 } else { 406 if (end&7) { 407 end &= ~7; 408 if (skb->ip_summed != CHECKSUM_UNNECESSARY) 409 skb->ip_summed = CHECKSUM_NONE; 410 } 411 if (end > qp->q.len) { 412 /* Some bits beyond end -> corruption. */ 413 if (qp->q.last_in & INET_FRAG_LAST_IN) 414 goto err; 415 qp->q.len = end; 416 } 417 } 418 if (end == offset) 419 goto err; 420 421 err = -ENOMEM; 422 if (pskb_pull(skb, ihl) == NULL) 423 goto err; 424 425 err = pskb_trim_rcsum(skb, end - offset); 426 if (err) 427 goto err; 428 429 /* Find out which fragments are in front and at the back of us 430 * in the chain of fragments so far. We must know where to put 431 * this fragment, right? 432 */ 433 prev = qp->q.fragments_tail; 434 if (!prev || FRAG_CB(prev)->offset < offset) { 435 next = NULL; 436 goto found; 437 } 438 prev = NULL; 439 for (next = qp->q.fragments; next != NULL; next = next->next) { 440 if (FRAG_CB(next)->offset >= offset) 441 break; /* bingo! */ 442 prev = next; 443 } 444 445 found: 446 /* We found where to put this one. Check for overlap with 447 * preceding fragment, and, if needed, align things so that 448 * any overlaps are eliminated. 449 */ 450 if (prev) { 451 int i = (FRAG_CB(prev)->offset + prev->len) - offset; 452 453 if (i > 0) { 454 offset += i; 455 err = -EINVAL; 456 if (end <= offset) 457 goto err; 458 err = -ENOMEM; 459 if (!pskb_pull(skb, i)) 460 goto err; 461 if (skb->ip_summed != CHECKSUM_UNNECESSARY) 462 skb->ip_summed = CHECKSUM_NONE; 463 } 464 } 465 466 err = -ENOMEM; 467 468 while (next && FRAG_CB(next)->offset < end) { 469 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ 470 471 if (i < next->len) { 472 /* Eat head of the next overlapped fragment 473 * and leave the loop. The next ones cannot overlap. 474 */ 475 if (!pskb_pull(next, i)) 476 goto err; 477 FRAG_CB(next)->offset += i; 478 qp->q.meat -= i; 479 if (next->ip_summed != CHECKSUM_UNNECESSARY) 480 next->ip_summed = CHECKSUM_NONE; 481 break; 482 } else { 483 struct sk_buff *free_it = next; 484 485 /* Old fragment is completely overridden with 486 * new one drop it. 487 */ 488 next = next->next; 489 490 if (prev) 491 prev->next = next; 492 else 493 qp->q.fragments = next; 494 495 qp->q.meat -= free_it->len; 496 sub_frag_mem_limit(&qp->q, free_it->truesize); 497 kfree_skb(free_it); 498 } 499 } 500 501 FRAG_CB(skb)->offset = offset; 502 503 /* Insert this fragment in the chain of fragments. */ 504 skb->next = next; 505 if (!next) 506 qp->q.fragments_tail = skb; 507 if (prev) 508 prev->next = skb; 509 else 510 qp->q.fragments = skb; 511 512 dev = skb->dev; 513 if (dev) { 514 qp->iif = dev->ifindex; 515 skb->dev = NULL; 516 } 517 qp->q.stamp = skb->tstamp; 518 qp->q.meat += skb->len; 519 qp->ecn |= ecn; 520 add_frag_mem_limit(&qp->q, skb->truesize); 521 if (offset == 0) 522 qp->q.last_in |= INET_FRAG_FIRST_IN; 523 524 if (ip_hdr(skb)->frag_off & htons(IP_DF) && 525 skb->len + ihl > qp->q.max_size) 526 qp->q.max_size = skb->len + ihl; 527 528 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && 529 qp->q.meat == qp->q.len) 530 return ip_frag_reasm(qp, prev, dev); 531 532 inet_frag_lru_move(&qp->q); 533 return -EINPROGRESS; 534 535 err: 536 kfree_skb(skb); 537 return err; 538 } 539 540 541 /* Build a new IP datagram from all its fragments. */ 542 543 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, 544 struct net_device *dev) 545 { 546 struct net *net = container_of(qp->q.net, struct net, ipv4.frags); 547 struct iphdr *iph; 548 struct sk_buff *fp, *head = qp->q.fragments; 549 int len; 550 int ihlen; 551 int err; 552 int sum_truesize; 553 u8 ecn; 554 555 ipq_kill(qp); 556 557 ecn = ip4_frag_ecn_table[qp->ecn]; 558 if (unlikely(ecn == 0xff)) { 559 err = -EINVAL; 560 goto out_fail; 561 } 562 /* Make the one we just received the head. */ 563 if (prev) { 564 head = prev->next; 565 fp = skb_clone(head, GFP_ATOMIC); 566 if (!fp) 567 goto out_nomem; 568 569 fp->next = head->next; 570 if (!fp->next) 571 qp->q.fragments_tail = fp; 572 prev->next = fp; 573 574 skb_morph(head, qp->q.fragments); 575 head->next = qp->q.fragments->next; 576 577 consume_skb(qp->q.fragments); 578 qp->q.fragments = head; 579 } 580 581 WARN_ON(head == NULL); 582 WARN_ON(FRAG_CB(head)->offset != 0); 583 584 /* Allocate a new buffer for the datagram. */ 585 ihlen = ip_hdrlen(head); 586 len = ihlen + qp->q.len; 587 588 err = -E2BIG; 589 if (len > 65535) 590 goto out_oversize; 591 592 /* Head of list must not be cloned. */ 593 if (skb_unclone(head, GFP_ATOMIC)) 594 goto out_nomem; 595 596 /* If the first fragment is fragmented itself, we split 597 * it to two chunks: the first with data and paged part 598 * and the second, holding only fragments. */ 599 if (skb_has_frag_list(head)) { 600 struct sk_buff *clone; 601 int i, plen = 0; 602 603 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) 604 goto out_nomem; 605 clone->next = head->next; 606 head->next = clone; 607 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; 608 skb_frag_list_init(head); 609 for (i = 0; i < skb_shinfo(head)->nr_frags; i++) 610 plen += skb_frag_size(&skb_shinfo(head)->frags[i]); 611 clone->len = clone->data_len = head->data_len - plen; 612 head->data_len -= clone->len; 613 head->len -= clone->len; 614 clone->csum = 0; 615 clone->ip_summed = head->ip_summed; 616 add_frag_mem_limit(&qp->q, clone->truesize); 617 } 618 619 skb_push(head, head->data - skb_network_header(head)); 620 621 sum_truesize = head->truesize; 622 for (fp = head->next; fp;) { 623 bool headstolen; 624 int delta; 625 struct sk_buff *next = fp->next; 626 627 sum_truesize += fp->truesize; 628 if (head->ip_summed != fp->ip_summed) 629 head->ip_summed = CHECKSUM_NONE; 630 else if (head->ip_summed == CHECKSUM_COMPLETE) 631 head->csum = csum_add(head->csum, fp->csum); 632 633 if (skb_try_coalesce(head, fp, &headstolen, &delta)) { 634 kfree_skb_partial(fp, headstolen); 635 } else { 636 if (!skb_shinfo(head)->frag_list) 637 skb_shinfo(head)->frag_list = fp; 638 head->data_len += fp->len; 639 head->len += fp->len; 640 head->truesize += fp->truesize; 641 } 642 fp = next; 643 } 644 sub_frag_mem_limit(&qp->q, sum_truesize); 645 646 head->next = NULL; 647 head->dev = dev; 648 head->tstamp = qp->q.stamp; 649 IPCB(head)->frag_max_size = qp->q.max_size; 650 651 iph = ip_hdr(head); 652 /* max_size != 0 implies at least one fragment had IP_DF set */ 653 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0; 654 iph->tot_len = htons(len); 655 iph->tos |= ecn; 656 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS); 657 qp->q.fragments = NULL; 658 qp->q.fragments_tail = NULL; 659 return 0; 660 661 out_nomem: 662 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"), 663 qp); 664 err = -ENOMEM; 665 goto out_fail; 666 out_oversize: 667 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr); 668 out_fail: 669 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); 670 return err; 671 } 672 673 /* Process an incoming IP datagram fragment. */ 674 int ip_defrag(struct sk_buff *skb, u32 user) 675 { 676 struct ipq *qp; 677 struct net *net; 678 679 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev); 680 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); 681 682 /* Start by cleaning up the memory. */ 683 ip_evictor(net); 684 685 /* Lookup (or create) queue header */ 686 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { 687 int ret; 688 689 spin_lock(&qp->q.lock); 690 691 ret = ip_frag_queue(qp, skb); 692 693 spin_unlock(&qp->q.lock); 694 ipq_put(qp); 695 return ret; 696 } 697 698 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); 699 kfree_skb(skb); 700 return -ENOMEM; 701 } 702 EXPORT_SYMBOL(ip_defrag); 703 704 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user) 705 { 706 struct iphdr iph; 707 u32 len; 708 709 if (skb->protocol != htons(ETH_P_IP)) 710 return skb; 711 712 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph))) 713 return skb; 714 715 if (iph.ihl < 5 || iph.version != 4) 716 return skb; 717 718 len = ntohs(iph.tot_len); 719 if (skb->len < len || len < (iph.ihl * 4)) 720 return skb; 721 722 if (ip_is_fragment(&iph)) { 723 skb = skb_share_check(skb, GFP_ATOMIC); 724 if (skb) { 725 if (!pskb_may_pull(skb, iph.ihl*4)) 726 return skb; 727 if (pskb_trim_rcsum(skb, len)) 728 return skb; 729 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 730 if (ip_defrag(skb, user)) 731 return NULL; 732 skb->rxhash = 0; 733 } 734 } 735 return skb; 736 } 737 EXPORT_SYMBOL(ip_check_defrag); 738 739 #ifdef CONFIG_SYSCTL 740 static int zero; 741 742 static struct ctl_table ip4_frags_ns_ctl_table[] = { 743 { 744 .procname = "ipfrag_high_thresh", 745 .data = &init_net.ipv4.frags.high_thresh, 746 .maxlen = sizeof(int), 747 .mode = 0644, 748 .proc_handler = proc_dointvec 749 }, 750 { 751 .procname = "ipfrag_low_thresh", 752 .data = &init_net.ipv4.frags.low_thresh, 753 .maxlen = sizeof(int), 754 .mode = 0644, 755 .proc_handler = proc_dointvec 756 }, 757 { 758 .procname = "ipfrag_time", 759 .data = &init_net.ipv4.frags.timeout, 760 .maxlen = sizeof(int), 761 .mode = 0644, 762 .proc_handler = proc_dointvec_jiffies, 763 }, 764 { } 765 }; 766 767 static struct ctl_table ip4_frags_ctl_table[] = { 768 { 769 .procname = "ipfrag_secret_interval", 770 .data = &ip4_frags.secret_interval, 771 .maxlen = sizeof(int), 772 .mode = 0644, 773 .proc_handler = proc_dointvec_jiffies, 774 }, 775 { 776 .procname = "ipfrag_max_dist", 777 .data = &sysctl_ipfrag_max_dist, 778 .maxlen = sizeof(int), 779 .mode = 0644, 780 .proc_handler = proc_dointvec_minmax, 781 .extra1 = &zero 782 }, 783 { } 784 }; 785 786 static int __net_init ip4_frags_ns_ctl_register(struct net *net) 787 { 788 struct ctl_table *table; 789 struct ctl_table_header *hdr; 790 791 table = ip4_frags_ns_ctl_table; 792 if (!net_eq(net, &init_net)) { 793 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); 794 if (table == NULL) 795 goto err_alloc; 796 797 table[0].data = &net->ipv4.frags.high_thresh; 798 table[1].data = &net->ipv4.frags.low_thresh; 799 table[2].data = &net->ipv4.frags.timeout; 800 801 /* Don't export sysctls to unprivileged users */ 802 if (net->user_ns != &init_user_ns) 803 table[0].procname = NULL; 804 } 805 806 hdr = register_net_sysctl(net, "net/ipv4", table); 807 if (hdr == NULL) 808 goto err_reg; 809 810 net->ipv4.frags_hdr = hdr; 811 return 0; 812 813 err_reg: 814 if (!net_eq(net, &init_net)) 815 kfree(table); 816 err_alloc: 817 return -ENOMEM; 818 } 819 820 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) 821 { 822 struct ctl_table *table; 823 824 table = net->ipv4.frags_hdr->ctl_table_arg; 825 unregister_net_sysctl_table(net->ipv4.frags_hdr); 826 kfree(table); 827 } 828 829 static void ip4_frags_ctl_register(void) 830 { 831 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); 832 } 833 #else 834 static inline int ip4_frags_ns_ctl_register(struct net *net) 835 { 836 return 0; 837 } 838 839 static inline void ip4_frags_ns_ctl_unregister(struct net *net) 840 { 841 } 842 843 static inline void ip4_frags_ctl_register(void) 844 { 845 } 846 #endif 847 848 static int __net_init ipv4_frags_init_net(struct net *net) 849 { 850 /* Fragment cache limits. 851 * 852 * The fragment memory accounting code, (tries to) account for 853 * the real memory usage, by measuring both the size of frag 854 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) 855 * and the SKB's truesize. 856 * 857 * A 64K fragment consumes 129736 bytes (44*2944)+200 858 * (1500 truesize == 2944, sizeof(struct ipq) == 200) 859 * 860 * We will commit 4MB at one time. Should we cross that limit 861 * we will prune down to 3MB, making room for approx 8 big 64K 862 * fragments 8x128k. 863 */ 864 net->ipv4.frags.high_thresh = 4 * 1024 * 1024; 865 net->ipv4.frags.low_thresh = 3 * 1024 * 1024; 866 /* 867 * Important NOTE! Fragment queue must be destroyed before MSL expires. 868 * RFC791 is wrong proposing to prolongate timer each fragment arrival 869 * by TTL. 870 */ 871 net->ipv4.frags.timeout = IP_FRAG_TIME; 872 873 inet_frags_init_net(&net->ipv4.frags); 874 875 return ip4_frags_ns_ctl_register(net); 876 } 877 878 static void __net_exit ipv4_frags_exit_net(struct net *net) 879 { 880 ip4_frags_ns_ctl_unregister(net); 881 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); 882 } 883 884 static struct pernet_operations ip4_frags_ops = { 885 .init = ipv4_frags_init_net, 886 .exit = ipv4_frags_exit_net, 887 }; 888 889 void __init ipfrag_init(void) 890 { 891 ip4_frags_ctl_register(); 892 register_pernet_subsys(&ip4_frags_ops); 893 ip4_frags.hashfn = ip4_hashfn; 894 ip4_frags.constructor = ip4_frag_init; 895 ip4_frags.destructor = ip4_frag_free; 896 ip4_frags.skb_free = NULL; 897 ip4_frags.qsize = sizeof(struct ipq); 898 ip4_frags.match = ip4_frag_match; 899 ip4_frags.frag_expire = ip_expire; 900 ip4_frags.secret_interval = 10 * 60 * HZ; 901 inet_frags_init(&ip4_frags); 902 } 903