1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * The IP fragmentation functionality. 8 * 9 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG> 10 * Alan Cox <alan@lxorguk.ukuu.org.uk> 11 * 12 * Fixes: 13 * Alan Cox : Split from ip.c , see ip_input.c for history. 14 * David S. Miller : Begin massive cleanup... 15 * Andi Kleen : Add sysctls. 16 * xxxx : Overlapfrag bug. 17 * Ultima : ip_expire() kernel panic. 18 * Bill Hawes : Frag accounting and evictor fixes. 19 * John McDonald : 0 length frag bug. 20 * Alexey Kuznetsov: SMP races, threading, cleanup. 21 * Patrick McHardy : LRU queue of frag heads for evictor. 22 */ 23 24 #define pr_fmt(fmt) "IPv4: " fmt 25 26 #include <linux/compiler.h> 27 #include <linux/module.h> 28 #include <linux/types.h> 29 #include <linux/mm.h> 30 #include <linux/jiffies.h> 31 #include <linux/skbuff.h> 32 #include <linux/list.h> 33 #include <linux/ip.h> 34 #include <linux/icmp.h> 35 #include <linux/netdevice.h> 36 #include <linux/jhash.h> 37 #include <linux/random.h> 38 #include <linux/slab.h> 39 #include <net/route.h> 40 #include <net/dst.h> 41 #include <net/sock.h> 42 #include <net/ip.h> 43 #include <net/icmp.h> 44 #include <net/checksum.h> 45 #include <net/inetpeer.h> 46 #include <net/inet_frag.h> 47 #include <linux/tcp.h> 48 #include <linux/udp.h> 49 #include <linux/inet.h> 50 #include <linux/netfilter_ipv4.h> 51 #include <net/inet_ecn.h> 52 #include <net/l3mdev.h> 53 54 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 55 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c 56 * as well. Or notify me, at least. --ANK 57 */ 58 static const char ip_frag_cache_name[] = "ip4-frags"; 59 60 /* Describe an entry in the "incomplete datagrams" queue. */ 61 struct ipq { 62 struct inet_frag_queue q; 63 64 u8 ecn; /* RFC3168 support */ 65 u16 max_df_size; /* largest frag with DF set seen */ 66 int iif; 67 unsigned int rid; 68 struct inet_peer *peer; 69 }; 70 71 static u8 ip4_frag_ecn(u8 tos) 72 { 73 return 1 << (tos & INET_ECN_MASK); 74 } 75 76 static struct inet_frags ip4_frags; 77 78 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, 79 struct net_device *dev); 80 81 82 static void ip4_frag_init(struct inet_frag_queue *q, const void *a) 83 { 84 struct ipq *qp = container_of(q, struct ipq, q); 85 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4, 86 frags); 87 struct net *net = container_of(ipv4, struct net, ipv4); 88 89 const struct frag_v4_compare_key *key = a; 90 91 q->key.v4 = *key; 92 qp->ecn = 0; 93 qp->peer = q->net->max_dist ? 94 inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) : 95 NULL; 96 } 97 98 static void ip4_frag_free(struct inet_frag_queue *q) 99 { 100 struct ipq *qp; 101 102 qp = container_of(q, struct ipq, q); 103 if (qp->peer) 104 inet_putpeer(qp->peer); 105 } 106 107 108 /* Destruction primitives. */ 109 110 static void ipq_put(struct ipq *ipq) 111 { 112 inet_frag_put(&ipq->q); 113 } 114 115 /* Kill ipq entry. It is not destroyed immediately, 116 * because caller (and someone more) holds reference count. 117 */ 118 static void ipq_kill(struct ipq *ipq) 119 { 120 inet_frag_kill(&ipq->q); 121 } 122 123 static bool frag_expire_skip_icmp(u32 user) 124 { 125 return user == IP_DEFRAG_AF_PACKET || 126 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN, 127 __IP_DEFRAG_CONNTRACK_IN_END) || 128 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN, 129 __IP_DEFRAG_CONNTRACK_BRIDGE_IN); 130 } 131 132 /* 133 * Oops, a fragment queue timed out. Kill it and send an ICMP reply. 134 */ 135 static void ip_expire(struct timer_list *t) 136 { 137 struct inet_frag_queue *frag = from_timer(frag, t, timer); 138 const struct iphdr *iph; 139 struct sk_buff *head; 140 struct net *net; 141 struct ipq *qp; 142 int err; 143 144 qp = container_of(frag, struct ipq, q); 145 net = container_of(qp->q.net, struct net, ipv4.frags); 146 147 rcu_read_lock(); 148 spin_lock(&qp->q.lock); 149 150 if (qp->q.flags & INET_FRAG_COMPLETE) 151 goto out; 152 153 ipq_kill(qp); 154 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); 155 156 head = qp->q.fragments; 157 158 __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT); 159 160 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !head) 161 goto out; 162 163 head->dev = dev_get_by_index_rcu(net, qp->iif); 164 if (!head->dev) 165 goto out; 166 167 168 /* skb has no dst, perform route lookup again */ 169 iph = ip_hdr(head); 170 err = ip_route_input_noref(head, iph->daddr, iph->saddr, 171 iph->tos, head->dev); 172 if (err) 173 goto out; 174 175 /* Only an end host needs to send an ICMP 176 * "Fragment Reassembly Timeout" message, per RFC792. 177 */ 178 if (frag_expire_skip_icmp(qp->q.key.v4.user) && 179 (skb_rtable(head)->rt_type != RTN_LOCAL)) 180 goto out; 181 182 skb_get(head); 183 spin_unlock(&qp->q.lock); 184 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); 185 kfree_skb(head); 186 goto out_rcu_unlock; 187 188 out: 189 spin_unlock(&qp->q.lock); 190 out_rcu_unlock: 191 rcu_read_unlock(); 192 ipq_put(qp); 193 } 194 195 /* Find the correct entry in the "incomplete datagrams" queue for 196 * this IP datagram, and create new one, if nothing is found. 197 */ 198 static struct ipq *ip_find(struct net *net, struct iphdr *iph, 199 u32 user, int vif) 200 { 201 struct frag_v4_compare_key key = { 202 .saddr = iph->saddr, 203 .daddr = iph->daddr, 204 .user = user, 205 .vif = vif, 206 .id = iph->id, 207 .protocol = iph->protocol, 208 }; 209 struct inet_frag_queue *q; 210 211 q = inet_frag_find(&net->ipv4.frags, &key); 212 if (!q) 213 return NULL; 214 215 return container_of(q, struct ipq, q); 216 } 217 218 /* Is the fragment too far ahead to be part of ipq? */ 219 static int ip_frag_too_far(struct ipq *qp) 220 { 221 struct inet_peer *peer = qp->peer; 222 unsigned int max = qp->q.net->max_dist; 223 unsigned int start, end; 224 225 int rc; 226 227 if (!peer || !max) 228 return 0; 229 230 start = qp->rid; 231 end = atomic_inc_return(&peer->rid); 232 qp->rid = end; 233 234 rc = qp->q.fragments && (end - start) > max; 235 236 if (rc) { 237 struct net *net; 238 239 net = container_of(qp->q.net, struct net, ipv4.frags); 240 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); 241 } 242 243 return rc; 244 } 245 246 static int ip_frag_reinit(struct ipq *qp) 247 { 248 struct sk_buff *fp; 249 unsigned int sum_truesize = 0; 250 251 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) { 252 refcount_inc(&qp->q.refcnt); 253 return -ETIMEDOUT; 254 } 255 256 fp = qp->q.fragments; 257 do { 258 struct sk_buff *xp = fp->next; 259 260 sum_truesize += fp->truesize; 261 kfree_skb(fp); 262 fp = xp; 263 } while (fp); 264 sub_frag_mem_limit(qp->q.net, sum_truesize); 265 266 qp->q.flags = 0; 267 qp->q.len = 0; 268 qp->q.meat = 0; 269 qp->q.fragments = NULL; 270 qp->q.fragments_tail = NULL; 271 qp->iif = 0; 272 qp->ecn = 0; 273 274 return 0; 275 } 276 277 /* Add new segment to existing queue. */ 278 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) 279 { 280 struct sk_buff *prev, *next; 281 struct net_device *dev; 282 unsigned int fragsize; 283 int flags, offset; 284 int ihl, end; 285 int err = -ENOENT; 286 u8 ecn; 287 288 if (qp->q.flags & INET_FRAG_COMPLETE) 289 goto err; 290 291 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && 292 unlikely(ip_frag_too_far(qp)) && 293 unlikely(err = ip_frag_reinit(qp))) { 294 ipq_kill(qp); 295 goto err; 296 } 297 298 ecn = ip4_frag_ecn(ip_hdr(skb)->tos); 299 offset = ntohs(ip_hdr(skb)->frag_off); 300 flags = offset & ~IP_OFFSET; 301 offset &= IP_OFFSET; 302 offset <<= 3; /* offset is in 8-byte chunks */ 303 ihl = ip_hdrlen(skb); 304 305 /* Determine the position of this fragment. */ 306 end = offset + skb->len - skb_network_offset(skb) - ihl; 307 err = -EINVAL; 308 309 /* Is this the final fragment? */ 310 if ((flags & IP_MF) == 0) { 311 /* If we already have some bits beyond end 312 * or have different end, the segment is corrupted. 313 */ 314 if (end < qp->q.len || 315 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len)) 316 goto err; 317 qp->q.flags |= INET_FRAG_LAST_IN; 318 qp->q.len = end; 319 } else { 320 if (end&7) { 321 end &= ~7; 322 if (skb->ip_summed != CHECKSUM_UNNECESSARY) 323 skb->ip_summed = CHECKSUM_NONE; 324 } 325 if (end > qp->q.len) { 326 /* Some bits beyond end -> corruption. */ 327 if (qp->q.flags & INET_FRAG_LAST_IN) 328 goto err; 329 qp->q.len = end; 330 } 331 } 332 if (end == offset) 333 goto err; 334 335 err = -ENOMEM; 336 if (!pskb_pull(skb, skb_network_offset(skb) + ihl)) 337 goto err; 338 339 err = pskb_trim_rcsum(skb, end - offset); 340 if (err) 341 goto err; 342 343 /* Find out which fragments are in front and at the back of us 344 * in the chain of fragments so far. We must know where to put 345 * this fragment, right? 346 */ 347 prev = qp->q.fragments_tail; 348 if (!prev || prev->ip_defrag_offset < offset) { 349 next = NULL; 350 goto found; 351 } 352 prev = NULL; 353 for (next = qp->q.fragments; next != NULL; next = next->next) { 354 if (next->ip_defrag_offset >= offset) 355 break; /* bingo! */ 356 prev = next; 357 } 358 359 found: 360 /* We found where to put this one. Check for overlap with 361 * preceding fragment, and, if needed, align things so that 362 * any overlaps are eliminated. 363 */ 364 if (prev) { 365 int i = (prev->ip_defrag_offset + prev->len) - offset; 366 367 if (i > 0) { 368 offset += i; 369 err = -EINVAL; 370 if (end <= offset) 371 goto err; 372 err = -ENOMEM; 373 if (!pskb_pull(skb, i)) 374 goto err; 375 if (skb->ip_summed != CHECKSUM_UNNECESSARY) 376 skb->ip_summed = CHECKSUM_NONE; 377 } 378 } 379 380 err = -ENOMEM; 381 382 while (next && next->ip_defrag_offset < end) { 383 int i = end - next->ip_defrag_offset; /* overlap is 'i' bytes */ 384 385 if (i < next->len) { 386 int delta = -next->truesize; 387 388 /* Eat head of the next overlapped fragment 389 * and leave the loop. The next ones cannot overlap. 390 */ 391 if (!pskb_pull(next, i)) 392 goto err; 393 delta += next->truesize; 394 if (delta) 395 add_frag_mem_limit(qp->q.net, delta); 396 next->ip_defrag_offset += i; 397 qp->q.meat -= i; 398 if (next->ip_summed != CHECKSUM_UNNECESSARY) 399 next->ip_summed = CHECKSUM_NONE; 400 break; 401 } else { 402 struct sk_buff *free_it = next; 403 404 /* Old fragment is completely overridden with 405 * new one drop it. 406 */ 407 next = next->next; 408 409 if (prev) 410 prev->next = next; 411 else 412 qp->q.fragments = next; 413 414 qp->q.meat -= free_it->len; 415 sub_frag_mem_limit(qp->q.net, free_it->truesize); 416 kfree_skb(free_it); 417 } 418 } 419 420 /* Note : skb->ip_defrag_offset and skb->dev share the same location */ 421 dev = skb->dev; 422 if (dev) 423 qp->iif = dev->ifindex; 424 /* Makes sure compiler wont do silly aliasing games */ 425 barrier(); 426 skb->ip_defrag_offset = offset; 427 428 /* Insert this fragment in the chain of fragments. */ 429 skb->next = next; 430 if (!next) 431 qp->q.fragments_tail = skb; 432 if (prev) 433 prev->next = skb; 434 else 435 qp->q.fragments = skb; 436 437 qp->q.stamp = skb->tstamp; 438 qp->q.meat += skb->len; 439 qp->ecn |= ecn; 440 add_frag_mem_limit(qp->q.net, skb->truesize); 441 if (offset == 0) 442 qp->q.flags |= INET_FRAG_FIRST_IN; 443 444 fragsize = skb->len + ihl; 445 446 if (fragsize > qp->q.max_size) 447 qp->q.max_size = fragsize; 448 449 if (ip_hdr(skb)->frag_off & htons(IP_DF) && 450 fragsize > qp->max_df_size) 451 qp->max_df_size = fragsize; 452 453 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) && 454 qp->q.meat == qp->q.len) { 455 unsigned long orefdst = skb->_skb_refdst; 456 457 skb->_skb_refdst = 0UL; 458 err = ip_frag_reasm(qp, prev, dev); 459 skb->_skb_refdst = orefdst; 460 return err; 461 } 462 463 skb_dst_drop(skb); 464 return -EINPROGRESS; 465 466 err: 467 kfree_skb(skb); 468 return err; 469 } 470 471 472 /* Build a new IP datagram from all its fragments. */ 473 474 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev, 475 struct net_device *dev) 476 { 477 struct net *net = container_of(qp->q.net, struct net, ipv4.frags); 478 struct iphdr *iph; 479 struct sk_buff *fp, *head = qp->q.fragments; 480 int len; 481 int ihlen; 482 int err; 483 u8 ecn; 484 485 ipq_kill(qp); 486 487 ecn = ip_frag_ecn_table[qp->ecn]; 488 if (unlikely(ecn == 0xff)) { 489 err = -EINVAL; 490 goto out_fail; 491 } 492 /* Make the one we just received the head. */ 493 if (prev) { 494 head = prev->next; 495 fp = skb_clone(head, GFP_ATOMIC); 496 if (!fp) 497 goto out_nomem; 498 499 fp->next = head->next; 500 if (!fp->next) 501 qp->q.fragments_tail = fp; 502 prev->next = fp; 503 504 skb_morph(head, qp->q.fragments); 505 head->next = qp->q.fragments->next; 506 507 consume_skb(qp->q.fragments); 508 qp->q.fragments = head; 509 } 510 511 WARN_ON(!head); 512 WARN_ON(head->ip_defrag_offset != 0); 513 514 /* Allocate a new buffer for the datagram. */ 515 ihlen = ip_hdrlen(head); 516 len = ihlen + qp->q.len; 517 518 err = -E2BIG; 519 if (len > 65535) 520 goto out_oversize; 521 522 /* Head of list must not be cloned. */ 523 if (skb_unclone(head, GFP_ATOMIC)) 524 goto out_nomem; 525 526 /* If the first fragment is fragmented itself, we split 527 * it to two chunks: the first with data and paged part 528 * and the second, holding only fragments. */ 529 if (skb_has_frag_list(head)) { 530 struct sk_buff *clone; 531 int i, plen = 0; 532 533 clone = alloc_skb(0, GFP_ATOMIC); 534 if (!clone) 535 goto out_nomem; 536 clone->next = head->next; 537 head->next = clone; 538 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; 539 skb_frag_list_init(head); 540 for (i = 0; i < skb_shinfo(head)->nr_frags; i++) 541 plen += skb_frag_size(&skb_shinfo(head)->frags[i]); 542 clone->len = clone->data_len = head->data_len - plen; 543 head->data_len -= clone->len; 544 head->len -= clone->len; 545 clone->csum = 0; 546 clone->ip_summed = head->ip_summed; 547 add_frag_mem_limit(qp->q.net, clone->truesize); 548 } 549 550 skb_shinfo(head)->frag_list = head->next; 551 skb_push(head, head->data - skb_network_header(head)); 552 553 for (fp=head->next; fp; fp = fp->next) { 554 head->data_len += fp->len; 555 head->len += fp->len; 556 if (head->ip_summed != fp->ip_summed) 557 head->ip_summed = CHECKSUM_NONE; 558 else if (head->ip_summed == CHECKSUM_COMPLETE) 559 head->csum = csum_add(head->csum, fp->csum); 560 head->truesize += fp->truesize; 561 } 562 sub_frag_mem_limit(qp->q.net, head->truesize); 563 564 head->next = NULL; 565 head->dev = dev; 566 head->tstamp = qp->q.stamp; 567 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size); 568 569 iph = ip_hdr(head); 570 iph->tot_len = htons(len); 571 iph->tos |= ecn; 572 573 /* When we set IP_DF on a refragmented skb we must also force a 574 * call to ip_fragment to avoid forwarding a DF-skb of size s while 575 * original sender only sent fragments of size f (where f < s). 576 * 577 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest 578 * frag seen to avoid sending tiny DF-fragments in case skb was built 579 * from one very small df-fragment and one large non-df frag. 580 */ 581 if (qp->max_df_size == qp->q.max_size) { 582 IPCB(head)->flags |= IPSKB_FRAG_PMTU; 583 iph->frag_off = htons(IP_DF); 584 } else { 585 iph->frag_off = 0; 586 } 587 588 ip_send_check(iph); 589 590 __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS); 591 qp->q.fragments = NULL; 592 qp->q.fragments_tail = NULL; 593 return 0; 594 595 out_nomem: 596 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp); 597 err = -ENOMEM; 598 goto out_fail; 599 out_oversize: 600 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr); 601 out_fail: 602 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); 603 return err; 604 } 605 606 /* Process an incoming IP datagram fragment. */ 607 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user) 608 { 609 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev; 610 int vif = l3mdev_master_ifindex_rcu(dev); 611 struct ipq *qp; 612 613 __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS); 614 skb_orphan(skb); 615 616 /* Lookup (or create) queue header */ 617 qp = ip_find(net, ip_hdr(skb), user, vif); 618 if (qp) { 619 int ret; 620 621 spin_lock(&qp->q.lock); 622 623 ret = ip_frag_queue(qp, skb); 624 625 spin_unlock(&qp->q.lock); 626 ipq_put(qp); 627 return ret; 628 } 629 630 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS); 631 kfree_skb(skb); 632 return -ENOMEM; 633 } 634 EXPORT_SYMBOL(ip_defrag); 635 636 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user) 637 { 638 struct iphdr iph; 639 int netoff; 640 u32 len; 641 642 if (skb->protocol != htons(ETH_P_IP)) 643 return skb; 644 645 netoff = skb_network_offset(skb); 646 647 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0) 648 return skb; 649 650 if (iph.ihl < 5 || iph.version != 4) 651 return skb; 652 653 len = ntohs(iph.tot_len); 654 if (skb->len < netoff + len || len < (iph.ihl * 4)) 655 return skb; 656 657 if (ip_is_fragment(&iph)) { 658 skb = skb_share_check(skb, GFP_ATOMIC); 659 if (skb) { 660 if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) 661 return skb; 662 if (pskb_trim_rcsum(skb, netoff + len)) 663 return skb; 664 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); 665 if (ip_defrag(net, skb, user)) 666 return NULL; 667 skb_clear_hash(skb); 668 } 669 } 670 return skb; 671 } 672 EXPORT_SYMBOL(ip_check_defrag); 673 674 #ifdef CONFIG_SYSCTL 675 static int dist_min; 676 677 static struct ctl_table ip4_frags_ns_ctl_table[] = { 678 { 679 .procname = "ipfrag_high_thresh", 680 .data = &init_net.ipv4.frags.high_thresh, 681 .maxlen = sizeof(unsigned long), 682 .mode = 0644, 683 .proc_handler = proc_doulongvec_minmax, 684 .extra1 = &init_net.ipv4.frags.low_thresh 685 }, 686 { 687 .procname = "ipfrag_low_thresh", 688 .data = &init_net.ipv4.frags.low_thresh, 689 .maxlen = sizeof(unsigned long), 690 .mode = 0644, 691 .proc_handler = proc_doulongvec_minmax, 692 .extra2 = &init_net.ipv4.frags.high_thresh 693 }, 694 { 695 .procname = "ipfrag_time", 696 .data = &init_net.ipv4.frags.timeout, 697 .maxlen = sizeof(int), 698 .mode = 0644, 699 .proc_handler = proc_dointvec_jiffies, 700 }, 701 { 702 .procname = "ipfrag_max_dist", 703 .data = &init_net.ipv4.frags.max_dist, 704 .maxlen = sizeof(int), 705 .mode = 0644, 706 .proc_handler = proc_dointvec_minmax, 707 .extra1 = &dist_min, 708 }, 709 { } 710 }; 711 712 /* secret interval has been deprecated */ 713 static int ip4_frags_secret_interval_unused; 714 static struct ctl_table ip4_frags_ctl_table[] = { 715 { 716 .procname = "ipfrag_secret_interval", 717 .data = &ip4_frags_secret_interval_unused, 718 .maxlen = sizeof(int), 719 .mode = 0644, 720 .proc_handler = proc_dointvec_jiffies, 721 }, 722 { } 723 }; 724 725 static int __net_init ip4_frags_ns_ctl_register(struct net *net) 726 { 727 struct ctl_table *table; 728 struct ctl_table_header *hdr; 729 730 table = ip4_frags_ns_ctl_table; 731 if (!net_eq(net, &init_net)) { 732 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL); 733 if (!table) 734 goto err_alloc; 735 736 table[0].data = &net->ipv4.frags.high_thresh; 737 table[0].extra1 = &net->ipv4.frags.low_thresh; 738 table[0].extra2 = &init_net.ipv4.frags.high_thresh; 739 table[1].data = &net->ipv4.frags.low_thresh; 740 table[1].extra2 = &net->ipv4.frags.high_thresh; 741 table[2].data = &net->ipv4.frags.timeout; 742 table[3].data = &net->ipv4.frags.max_dist; 743 } 744 745 hdr = register_net_sysctl(net, "net/ipv4", table); 746 if (!hdr) 747 goto err_reg; 748 749 net->ipv4.frags_hdr = hdr; 750 return 0; 751 752 err_reg: 753 if (!net_eq(net, &init_net)) 754 kfree(table); 755 err_alloc: 756 return -ENOMEM; 757 } 758 759 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net) 760 { 761 struct ctl_table *table; 762 763 table = net->ipv4.frags_hdr->ctl_table_arg; 764 unregister_net_sysctl_table(net->ipv4.frags_hdr); 765 kfree(table); 766 } 767 768 static void __init ip4_frags_ctl_register(void) 769 { 770 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table); 771 } 772 #else 773 static int ip4_frags_ns_ctl_register(struct net *net) 774 { 775 return 0; 776 } 777 778 static void ip4_frags_ns_ctl_unregister(struct net *net) 779 { 780 } 781 782 static void __init ip4_frags_ctl_register(void) 783 { 784 } 785 #endif 786 787 static int __net_init ipv4_frags_init_net(struct net *net) 788 { 789 int res; 790 791 /* Fragment cache limits. 792 * 793 * The fragment memory accounting code, (tries to) account for 794 * the real memory usage, by measuring both the size of frag 795 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue)) 796 * and the SKB's truesize. 797 * 798 * A 64K fragment consumes 129736 bytes (44*2944)+200 799 * (1500 truesize == 2944, sizeof(struct ipq) == 200) 800 * 801 * We will commit 4MB at one time. Should we cross that limit 802 * we will prune down to 3MB, making room for approx 8 big 64K 803 * fragments 8x128k. 804 */ 805 net->ipv4.frags.high_thresh = 4 * 1024 * 1024; 806 net->ipv4.frags.low_thresh = 3 * 1024 * 1024; 807 /* 808 * Important NOTE! Fragment queue must be destroyed before MSL expires. 809 * RFC791 is wrong proposing to prolongate timer each fragment arrival 810 * by TTL. 811 */ 812 net->ipv4.frags.timeout = IP_FRAG_TIME; 813 814 net->ipv4.frags.max_dist = 64; 815 net->ipv4.frags.f = &ip4_frags; 816 817 res = inet_frags_init_net(&net->ipv4.frags); 818 if (res < 0) 819 return res; 820 res = ip4_frags_ns_ctl_register(net); 821 if (res < 0) 822 inet_frags_exit_net(&net->ipv4.frags); 823 return res; 824 } 825 826 static void __net_exit ipv4_frags_exit_net(struct net *net) 827 { 828 ip4_frags_ns_ctl_unregister(net); 829 inet_frags_exit_net(&net->ipv4.frags); 830 } 831 832 static struct pernet_operations ip4_frags_ops = { 833 .init = ipv4_frags_init_net, 834 .exit = ipv4_frags_exit_net, 835 }; 836 837 838 static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed) 839 { 840 return jhash2(data, 841 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed); 842 } 843 844 static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed) 845 { 846 const struct inet_frag_queue *fq = data; 847 848 return jhash2((const u32 *)&fq->key.v4, 849 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed); 850 } 851 852 static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr) 853 { 854 const struct frag_v4_compare_key *key = arg->key; 855 const struct inet_frag_queue *fq = ptr; 856 857 return !!memcmp(&fq->key, key, sizeof(*key)); 858 } 859 860 static const struct rhashtable_params ip4_rhash_params = { 861 .head_offset = offsetof(struct inet_frag_queue, node), 862 .key_offset = offsetof(struct inet_frag_queue, key), 863 .key_len = sizeof(struct frag_v4_compare_key), 864 .hashfn = ip4_key_hashfn, 865 .obj_hashfn = ip4_obj_hashfn, 866 .obj_cmpfn = ip4_obj_cmpfn, 867 .automatic_shrinking = true, 868 }; 869 870 void __init ipfrag_init(void) 871 { 872 ip4_frags.constructor = ip4_frag_init; 873 ip4_frags.destructor = ip4_frag_free; 874 ip4_frags.qsize = sizeof(struct ipq); 875 ip4_frags.frag_expire = ip_expire; 876 ip4_frags.frags_cache_name = ip_frag_cache_name; 877 ip4_frags.rhash_params = ip4_rhash_params; 878 if (inet_frags_init(&ip4_frags)) 879 panic("IP: failed to allocate ip4_frags cache\n"); 880 ip4_frags_ctl_register(); 881 register_pernet_subsys(&ip4_frags_ops); 882 } 883