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 has no dst, perform route lookup again */ 252 iph = ip_hdr(head); 253 err = ip_route_input_noref(head, iph->daddr, iph->saddr, 254 iph->tos, head->dev); 255 if (err) 256 goto out_rcu_unlock; 257 258 /* 259 * Only an end host needs to send an ICMP 260 * "Fragment Reassembly Timeout" message, per RFC792. 261 */ 262 if (qp->user == IP_DEFRAG_AF_PACKET || 263 (qp->user == IP_DEFRAG_CONNTRACK_IN && 264 skb_rtable(head)->rt_type != RTN_LOCAL)) 265 goto out_rcu_unlock; 266 267 268 /* Send an ICMP "Fragment Reassembly Timeout" message. */ 269 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); 270 out_rcu_unlock: 271 rcu_read_unlock(); 272 } 273 out: 274 spin_unlock(&qp->q.lock); 275 ipq_put(qp); 276 } 277 278 /* Find the correct entry in the "incomplete datagrams" queue for 279 * this IP datagram, and create new one, if nothing is found. 280 */ 281 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user) 282 { 283 struct inet_frag_queue *q; 284 struct ip4_create_arg arg; 285 unsigned int hash; 286 287 arg.iph = iph; 288 arg.user = user; 289 290 read_lock(&ip4_frags.lock); 291 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol); 292 293 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash); 294 if (IS_ERR_OR_NULL(q)) { 295 inet_frag_maybe_warn_overflow(q, pr_fmt()); 296 return NULL; 297 } 298 return container_of(q, struct ipq, q); 299 } 300 301 /* Is the fragment too far ahead to be part of ipq? */ 302 static inline int ip_frag_too_far(struct ipq *qp) 303 { 304 struct inet_peer *peer = qp->peer; 305 unsigned int max = sysctl_ipfrag_max_dist; 306 unsigned int start, end; 307 308 int rc; 309 310 if (!peer || !max) 311 return 0; 312 313 start = qp->rid; 314 end = atomic_inc_return(&peer->rid); 315 qp->rid = end; 316 317 rc = qp->q.fragments && (end - start) > max; 318 319 if (rc) { 320 struct net *net; 321 322 net = container_of(qp->q.net, struct net, ipv4.frags); 323 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); 324 } 325 326 return rc; 327 } 328 329 static int ip_frag_reinit(struct ipq *qp) 330 { 331 struct sk_buff *fp; 332 unsigned int sum_truesize = 0; 333 334 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { 335 atomic_inc(&qp->q.refcnt); 336 return -ETIMEDOUT; 337 } 338 339 fp = qp->q.fragments; 340 do { 341 struct sk_buff *xp = fp->next; 342 343 sum_truesize += fp->truesize; 344 kfree_skb(fp); 345 fp = xp; 346 } while (fp); 347 sub_frag_mem_limit(&qp->q, sum_truesize); 348 349 qp->q.last_in = 0; 350 qp->q.len = 0; 351 qp->q.meat = 0; 352 qp->q.fragments = NULL; 353 qp->q.fragments_tail = NULL; 354 qp->iif = 0; 355 qp->ecn = 0; 356 357 return 0; 358 } 359 360 /* Add new segment to existing queue. */ 361 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) 362 { 363 struct sk_buff *prev, *next; 364 struct net_device *dev; 365 int flags, offset; 366 int ihl, end; 367 int err = -ENOENT; 368 u8 ecn; 369 370 if (qp->q.last_in & INET_FRAG_COMPLETE) 371 goto err; 372 373 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && 374 unlikely(ip_frag_too_far(qp)) && 375 unlikely(err = ip_frag_reinit(qp))) { 376 ipq_kill(qp); 377 goto err; 378 } 379 380 ecn = ip4_frag_ecn(ip_hdr(skb)->tos); 381 offset = ntohs(ip_hdr(skb)->frag_off); 382 flags = offset & ~IP_OFFSET; 383 offset &= IP_OFFSET; 384 offset <<= 3; /* offset is in 8-byte chunks */ 385 ihl = ip_hdrlen(skb); 386 387 /* Determine the position of this fragment. */ 388 end = offset + skb->len - ihl; 389 err = -EINVAL; 390 391 /* Is this the final fragment? */ 392 if ((flags & IP_MF) == 0) { 393 /* If we already have some bits beyond end 394 * or have different end, the segment is corrupted. 395 */ 396 if (end < qp->q.len || 397 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len)) 398 goto err; 399 qp->q.last_in |= INET_FRAG_LAST_IN; 400 qp->q.len = end; 401 } else { 402 if (end&7) { 403 end &= ~7; 404 if (skb->ip_summed != CHECKSUM_UNNECESSARY) 405 skb->ip_summed = CHECKSUM_NONE; 406 } 407 if (end > qp->q.len) { 408 /* Some bits beyond end -> corruption. */ 409 if (qp->q.last_in & INET_FRAG_LAST_IN) 410 goto err; 411 qp->q.len = end; 412 } 413 } 414 if (end == offset) 415 goto err; 416 417 err = -ENOMEM; 418 if (pskb_pull(skb, ihl) == NULL) 419 goto err; 420 421 err = pskb_trim_rcsum(skb, end - offset); 422 if (err) 423 goto err; 424 425 /* Find out which fragments are in front and at the back of us 426 * in the chain of fragments so far. We must know where to put 427 * this fragment, right? 428 */ 429 prev = qp->q.fragments_tail; 430 if (!prev || FRAG_CB(prev)->offset < offset) { 431 next = NULL; 432 goto found; 433 } 434 prev = NULL; 435 for (next = qp->q.fragments; next != NULL; next = next->next) { 436 if (FRAG_CB(next)->offset >= offset) 437 break; /* bingo! */ 438 prev = next; 439 } 440 441 found: 442 /* We found where to put this one. Check for overlap with 443 * preceding fragment, and, if needed, align things so that 444 * any overlaps are eliminated. 445 */ 446 if (prev) { 447 int i = (FRAG_CB(prev)->offset + prev->len) - offset; 448 449 if (i > 0) { 450 offset += i; 451 err = -EINVAL; 452 if (end <= offset) 453 goto err; 454 err = -ENOMEM; 455 if (!pskb_pull(skb, i)) 456 goto err; 457 if (skb->ip_summed != CHECKSUM_UNNECESSARY) 458 skb->ip_summed = CHECKSUM_NONE; 459 } 460 } 461 462 err = -ENOMEM; 463 464 while (next && FRAG_CB(next)->offset < end) { 465 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ 466 467 if (i < next->len) { 468 /* Eat head of the next overlapped fragment 469 * and leave the loop. The next ones cannot overlap. 470 */ 471 if (!pskb_pull(next, i)) 472 goto err; 473 FRAG_CB(next)->offset += i; 474 qp->q.meat -= i; 475 if (next->ip_summed != CHECKSUM_UNNECESSARY) 476 next->ip_summed = CHECKSUM_NONE; 477 break; 478 } else { 479 struct sk_buff *free_it = next; 480 481 /* Old fragment is completely overridden with 482 * new one drop it. 483 */ 484 next = next->next; 485 486 if (prev) 487 prev->next = next; 488 else 489 qp->q.fragments = next; 490 491 qp->q.meat -= free_it->len; 492 sub_frag_mem_limit(&qp->q, free_it->truesize); 493 kfree_skb(free_it); 494 } 495 } 496 497 FRAG_CB(skb)->offset = offset; 498 499 /* Insert this fragment in the chain of fragments. */ 500 skb->next = next; 501 if (!next) 502 qp->q.fragments_tail = skb; 503 if (prev) 504 prev->next = skb; 505 else 506 qp->q.fragments = skb; 507 508 dev = skb->dev; 509 if (dev) { 510 qp->iif = dev->ifindex; 511 skb->dev = NULL; 512 } 513 qp->q.stamp = skb->tstamp; 514 qp->q.meat += skb->len; 515 qp->ecn |= ecn; 516 add_frag_mem_limit(&qp->q, skb->truesize); 517 if (offset == 0) 518 qp->q.last_in |= INET_FRAG_FIRST_IN; 519 520 if (ip_hdr(skb)->frag_off & htons(IP_DF) && 521 skb->len + ihl > qp->q.max_size) 522 qp->q.max_size = skb->len + ihl; 523 524 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && 525 qp->q.meat == qp->q.len) { 526 unsigned long orefdst = skb->_skb_refdst; 527 528 skb->_skb_refdst = 0UL; 529 err = ip_frag_reasm(qp, prev, dev); 530 skb->_skb_refdst = orefdst; 531 return err; 532 } 533 534 skb_dst_drop(skb); 535 inet_frag_lru_move(&qp->q); 536 return -EINPROGRESS; 537 538 err: 539 kfree_skb(skb); 540 return err; 541 } 542 543 544 /* Build a new IP datagram from all its fragments. */ 545 546 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, 547 struct net_device *dev) 548 { 549 struct net *net = container_of(qp->q.net, struct net, ipv4.frags); 550 struct iphdr *iph; 551 struct sk_buff *fp, *head = qp->q.fragments; 552 int len; 553 int ihlen; 554 int err; 555 int sum_truesize; 556 u8 ecn; 557 558 ipq_kill(qp); 559 560 ecn = ip4_frag_ecn_table[qp->ecn]; 561 if (unlikely(ecn == 0xff)) { 562 err = -EINVAL; 563 goto out_fail; 564 } 565 /* Make the one we just received the head. */ 566 if (prev) { 567 head = prev->next; 568 fp = skb_clone(head, GFP_ATOMIC); 569 if (!fp) 570 goto out_nomem; 571 572 fp->next = head->next; 573 if (!fp->next) 574 qp->q.fragments_tail = fp; 575 prev->next = fp; 576 577 skb_morph(head, qp->q.fragments); 578 head->next = qp->q.fragments->next; 579 580 consume_skb(qp->q.fragments); 581 qp->q.fragments = head; 582 } 583 584 WARN_ON(head == NULL); 585 WARN_ON(FRAG_CB(head)->offset != 0); 586 587 /* Allocate a new buffer for the datagram. */ 588 ihlen = ip_hdrlen(head); 589 len = ihlen + qp->q.len; 590 591 err = -E2BIG; 592 if (len > 65535) 593 goto out_oversize; 594 595 /* Head of list must not be cloned. */ 596 if (skb_unclone(head, GFP_ATOMIC)) 597 goto out_nomem; 598 599 /* If the first fragment is fragmented itself, we split 600 * it to two chunks: the first with data and paged part 601 * and the second, holding only fragments. */ 602 if (skb_has_frag_list(head)) { 603 struct sk_buff *clone; 604 int i, plen = 0; 605 606 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) 607 goto out_nomem; 608 clone->next = head->next; 609 head->next = clone; 610 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; 611 skb_frag_list_init(head); 612 for (i = 0; i < skb_shinfo(head)->nr_frags; i++) 613 plen += skb_frag_size(&skb_shinfo(head)->frags[i]); 614 clone->len = clone->data_len = head->data_len - plen; 615 head->data_len -= clone->len; 616 head->len -= clone->len; 617 clone->csum = 0; 618 clone->ip_summed = head->ip_summed; 619 add_frag_mem_limit(&qp->q, clone->truesize); 620 } 621 622 skb_push(head, head->data - skb_network_header(head)); 623 624 sum_truesize = head->truesize; 625 for (fp = head->next; fp;) { 626 bool headstolen; 627 int delta; 628 struct sk_buff *next = fp->next; 629 630 sum_truesize += fp->truesize; 631 if (head->ip_summed != fp->ip_summed) 632 head->ip_summed = CHECKSUM_NONE; 633 else if (head->ip_summed == CHECKSUM_COMPLETE) 634 head->csum = csum_add(head->csum, fp->csum); 635 636 if (skb_try_coalesce(head, fp, &headstolen, &delta)) { 637 kfree_skb_partial(fp, headstolen); 638 } else { 639 if (!skb_shinfo(head)->frag_list) 640 skb_shinfo(head)->frag_list = fp; 641 head->data_len += fp->len; 642 head->len += fp->len; 643 head->truesize += fp->truesize; 644 } 645 fp = next; 646 } 647 sub_frag_mem_limit(&qp->q, sum_truesize); 648 649 head->next = NULL; 650 head->dev = dev; 651 head->tstamp = qp->q.stamp; 652 IPCB(head)->frag_max_size = qp->q.max_size; 653 654 iph = ip_hdr(head); 655 /* max_size != 0 implies at least one fragment had IP_DF set */ 656 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0; 657 iph->tot_len = htons(len); 658 iph->tos |= ecn; 659 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS); 660 qp->q.fragments = NULL; 661 qp->q.fragments_tail = NULL; 662 return 0; 663 664 out_nomem: 665 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"), 666 qp); 667 err = -ENOMEM; 668 goto out_fail; 669 out_oversize: 670 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr); 671 out_fail: 672 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); 673 return err; 674 } 675 676 /* Process an incoming IP datagram fragment. */ 677 int ip_defrag(struct sk_buff *skb, u32 user) 678 { 679 struct ipq *qp; 680 struct net *net; 681 682 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev); 683 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS); 684 685 /* Start by cleaning up the memory. */ 686 ip_evictor(net); 687 688 /* Lookup (or create) queue header */ 689 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) { 690 int ret; 691 692 spin_lock(&qp->q.lock); 693 694 ret = ip_frag_queue(qp, skb); 695 696 spin_unlock(&qp->q.lock); 697 ipq_put(qp); 698 return ret; 699 } 700 701 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS); 702 kfree_skb(skb); 703 return -ENOMEM; 704 } 705 EXPORT_SYMBOL(ip_defrag); 706 707 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user) 708 { 709 struct iphdr iph; 710 u32 len; 711 712 if (skb->protocol != htons(ETH_P_IP)) 713 return skb; 714 715 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph))) 716 return skb; 717 718 if (iph.ihl < 5 || iph.version != 4) 719 return skb; 720 721 len = ntohs(iph.tot_len); 722 if (skb->len < len || len < (iph.ihl * 4)) 723 return skb; 724 725 if (ip_is_fragment(&iph)) { 726 skb = skb_share_check(skb, GFP_ATOMIC); 727 if (skb) { 728 if (!pskb_may_pull(skb, iph.ihl*4)) 729 return skb; 730 if (pskb_trim_rcsum(skb, len)) 731 return skb; 732 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 733 if (ip_defrag(skb, user)) 734 return NULL; 735 skb->rxhash = 0; 736 } 737 } 738 return skb; 739 } 740 EXPORT_SYMBOL(ip_check_defrag); 741 742 #ifdef CONFIG_SYSCTL 743 static int zero; 744 745 static struct ctl_table ip4_frags_ns_ctl_table[] = { 746 { 747 .procname = "ipfrag_high_thresh", 748 .data = &init_net.ipv4.frags.high_thresh, 749 .maxlen = sizeof(int), 750 .mode = 0644, 751 .proc_handler = proc_dointvec 752 }, 753 { 754 .procname = "ipfrag_low_thresh", 755 .data = &init_net.ipv4.frags.low_thresh, 756 .maxlen = sizeof(int), 757 .mode = 0644, 758 .proc_handler = proc_dointvec 759 }, 760 { 761 .procname = "ipfrag_time", 762 .data = &init_net.ipv4.frags.timeout, 763 .maxlen = sizeof(int), 764 .mode = 0644, 765 .proc_handler = proc_dointvec_jiffies, 766 }, 767 { } 768 }; 769 770 static struct ctl_table ip4_frags_ctl_table[] = { 771 { 772 .procname = "ipfrag_secret_interval", 773 .data = &ip4_frags.secret_interval, 774 .maxlen = sizeof(int), 775 .mode = 0644, 776 .proc_handler = proc_dointvec_jiffies, 777 }, 778 { 779 .procname = "ipfrag_max_dist", 780 .data = &sysctl_ipfrag_max_dist, 781 .maxlen = sizeof(int), 782 .mode = 0644, 783 .proc_handler = proc_dointvec_minmax, 784 .extra1 = &zero 785 }, 786 { } 787 }; 788 789 static int __net_init ip4_frags_ns_ctl_register(struct net *net) 790 { 791 struct ctl_table *table; 792 struct ctl_table_header *hdr; 793 794 table = ip4_frags_ns_ctl_table; 795 if (!net_eq(net, &init_net)) { 796 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); 797 if (table == NULL) 798 goto err_alloc; 799 800 table[0].data = &net->ipv4.frags.high_thresh; 801 table[1].data = &net->ipv4.frags.low_thresh; 802 table[2].data = &net->ipv4.frags.timeout; 803 804 /* Don't export sysctls to unprivileged users */ 805 if (net->user_ns != &init_user_ns) 806 table[0].procname = NULL; 807 } 808 809 hdr = register_net_sysctl(net, "net/ipv4", table); 810 if (hdr == NULL) 811 goto err_reg; 812 813 net->ipv4.frags_hdr = hdr; 814 return 0; 815 816 err_reg: 817 if (!net_eq(net, &init_net)) 818 kfree(table); 819 err_alloc: 820 return -ENOMEM; 821 } 822 823 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) 824 { 825 struct ctl_table *table; 826 827 table = net->ipv4.frags_hdr->ctl_table_arg; 828 unregister_net_sysctl_table(net->ipv4.frags_hdr); 829 kfree(table); 830 } 831 832 static void ip4_frags_ctl_register(void) 833 { 834 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); 835 } 836 #else 837 static inline int ip4_frags_ns_ctl_register(struct net *net) 838 { 839 return 0; 840 } 841 842 static inline void ip4_frags_ns_ctl_unregister(struct net *net) 843 { 844 } 845 846 static inline void ip4_frags_ctl_register(void) 847 { 848 } 849 #endif 850 851 static int __net_init ipv4_frags_init_net(struct net *net) 852 { 853 /* Fragment cache limits. 854 * 855 * The fragment memory accounting code, (tries to) account for 856 * the real memory usage, by measuring both the size of frag 857 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) 858 * and the SKB's truesize. 859 * 860 * A 64K fragment consumes 129736 bytes (44*2944)+200 861 * (1500 truesize == 2944, sizeof(struct ipq) == 200) 862 * 863 * We will commit 4MB at one time. Should we cross that limit 864 * we will prune down to 3MB, making room for approx 8 big 64K 865 * fragments 8x128k. 866 */ 867 net->ipv4.frags.high_thresh = 4 * 1024 * 1024; 868 net->ipv4.frags.low_thresh = 3 * 1024 * 1024; 869 /* 870 * Important NOTE! Fragment queue must be destroyed before MSL expires. 871 * RFC791 is wrong proposing to prolongate timer each fragment arrival 872 * by TTL. 873 */ 874 net->ipv4.frags.timeout = IP_FRAG_TIME; 875 876 inet_frags_init_net(&net->ipv4.frags); 877 878 return ip4_frags_ns_ctl_register(net); 879 } 880 881 static void __net_exit ipv4_frags_exit_net(struct net *net) 882 { 883 ip4_frags_ns_ctl_unregister(net); 884 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags); 885 } 886 887 static struct pernet_operations ip4_frags_ops = { 888 .init = ipv4_frags_init_net, 889 .exit = ipv4_frags_exit_net, 890 }; 891 892 void __init ipfrag_init(void) 893 { 894 ip4_frags_ctl_register(); 895 register_pernet_subsys(&ip4_frags_ops); 896 ip4_frags.hashfn = ip4_hashfn; 897 ip4_frags.constructor = ip4_frag_init; 898 ip4_frags.destructor = ip4_frag_free; 899 ip4_frags.skb_free = NULL; 900 ip4_frags.qsize = sizeof(struct ipq); 901 ip4_frags.match = ip4_frag_match; 902 ip4_frags.frag_expire = ip_expire; 903 ip4_frags.secret_interval = 10 * 60 * HZ; 904 inet_frags_init(&ip4_frags); 905 } 906