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 Internet Protocol (IP) output module. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Donald Becker, <becker@super.org> 11 * Alan Cox, <Alan.Cox@linux.org> 12 * Richard Underwood 13 * Stefan Becker, <stefanb@yello.ping.de> 14 * Jorge Cwik, <jorge@laser.satlink.net> 15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 16 * Hirokazu Takahashi, <taka@valinux.co.jp> 17 * 18 * See ip_input.c for original log 19 * 20 * Fixes: 21 * Alan Cox : Missing nonblock feature in ip_build_xmit. 22 * Mike Kilburn : htons() missing in ip_build_xmit. 23 * Bradford Johnson: Fix faulty handling of some frames when 24 * no route is found. 25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit 26 * (in case if packet not accepted by 27 * output firewall rules) 28 * Mike McLagan : Routing by source 29 * Alexey Kuznetsov: use new route cache 30 * Andi Kleen: Fix broken PMTU recovery and remove 31 * some redundant tests. 32 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 33 * Andi Kleen : Replace ip_reply with ip_send_reply. 34 * Andi Kleen : Split fast and slow ip_build_xmit path 35 * for decreased register pressure on x86 36 * and more readibility. 37 * Marc Boucher : When call_out_firewall returns FW_QUEUE, 38 * silently drop skb instead of failing with -EPERM. 39 * Detlev Wengorz : Copy protocol for fragments. 40 * Hirokazu Takahashi: HW checksumming for outgoing UDP 41 * datagrams. 42 * Hirokazu Takahashi: sendfile() on UDP works now. 43 */ 44 45 #include <asm/uaccess.h> 46 #include <linux/module.h> 47 #include <linux/types.h> 48 #include <linux/kernel.h> 49 #include <linux/mm.h> 50 #include <linux/string.h> 51 #include <linux/errno.h> 52 #include <linux/highmem.h> 53 #include <linux/slab.h> 54 55 #include <linux/socket.h> 56 #include <linux/sockios.h> 57 #include <linux/in.h> 58 #include <linux/inet.h> 59 #include <linux/netdevice.h> 60 #include <linux/etherdevice.h> 61 #include <linux/proc_fs.h> 62 #include <linux/stat.h> 63 #include <linux/init.h> 64 65 #include <net/snmp.h> 66 #include <net/ip.h> 67 #include <net/protocol.h> 68 #include <net/route.h> 69 #include <net/xfrm.h> 70 #include <linux/skbuff.h> 71 #include <net/sock.h> 72 #include <net/arp.h> 73 #include <net/icmp.h> 74 #include <net/checksum.h> 75 #include <net/inetpeer.h> 76 #include <linux/igmp.h> 77 #include <linux/netfilter_ipv4.h> 78 #include <linux/netfilter_bridge.h> 79 #include <linux/mroute.h> 80 #include <linux/netlink.h> 81 #include <linux/tcp.h> 82 83 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL; 84 EXPORT_SYMBOL(sysctl_ip_default_ttl); 85 86 /* Generate a checksum for an outgoing IP datagram. */ 87 void ip_send_check(struct iphdr *iph) 88 { 89 iph->check = 0; 90 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 91 } 92 EXPORT_SYMBOL(ip_send_check); 93 94 int __ip_local_out(struct sk_buff *skb) 95 { 96 struct iphdr *iph = ip_hdr(skb); 97 98 iph->tot_len = htons(skb->len); 99 ip_send_check(iph); 100 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL, 101 skb_dst(skb)->dev, dst_output); 102 } 103 104 int ip_local_out_sk(struct sock *sk, struct sk_buff *skb) 105 { 106 int err; 107 108 err = __ip_local_out(skb); 109 if (likely(err == 1)) 110 err = dst_output_sk(sk, skb); 111 112 return err; 113 } 114 EXPORT_SYMBOL_GPL(ip_local_out_sk); 115 116 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) 117 { 118 int ttl = inet->uc_ttl; 119 120 if (ttl < 0) 121 ttl = ip4_dst_hoplimit(dst); 122 return ttl; 123 } 124 125 /* 126 * Add an ip header to a skbuff and send it out. 127 * 128 */ 129 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk, 130 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) 131 { 132 struct inet_sock *inet = inet_sk(sk); 133 struct rtable *rt = skb_rtable(skb); 134 struct iphdr *iph; 135 136 /* Build the IP header. */ 137 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); 138 skb_reset_network_header(skb); 139 iph = ip_hdr(skb); 140 iph->version = 4; 141 iph->ihl = 5; 142 iph->tos = inet->tos; 143 if (ip_dont_fragment(sk, &rt->dst)) 144 iph->frag_off = htons(IP_DF); 145 else 146 iph->frag_off = 0; 147 iph->ttl = ip_select_ttl(inet, &rt->dst); 148 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); 149 iph->saddr = saddr; 150 iph->protocol = sk->sk_protocol; 151 ip_select_ident(skb, &rt->dst, sk); 152 153 if (opt && opt->opt.optlen) { 154 iph->ihl += opt->opt.optlen>>2; 155 ip_options_build(skb, &opt->opt, daddr, rt, 0); 156 } 157 158 skb->priority = sk->sk_priority; 159 skb->mark = sk->sk_mark; 160 161 /* Send it out. */ 162 return ip_local_out(skb); 163 } 164 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); 165 166 static inline int ip_finish_output2(struct sk_buff *skb) 167 { 168 struct dst_entry *dst = skb_dst(skb); 169 struct rtable *rt = (struct rtable *)dst; 170 struct net_device *dev = dst->dev; 171 unsigned int hh_len = LL_RESERVED_SPACE(dev); 172 struct neighbour *neigh; 173 u32 nexthop; 174 175 if (rt->rt_type == RTN_MULTICAST) { 176 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len); 177 } else if (rt->rt_type == RTN_BROADCAST) 178 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len); 179 180 /* Be paranoid, rather than too clever. */ 181 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 182 struct sk_buff *skb2; 183 184 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 185 if (skb2 == NULL) { 186 kfree_skb(skb); 187 return -ENOMEM; 188 } 189 if (skb->sk) 190 skb_set_owner_w(skb2, skb->sk); 191 consume_skb(skb); 192 skb = skb2; 193 } 194 195 rcu_read_lock_bh(); 196 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr); 197 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 198 if (unlikely(!neigh)) 199 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 200 if (!IS_ERR(neigh)) { 201 int res = dst_neigh_output(dst, neigh, skb); 202 203 rcu_read_unlock_bh(); 204 return res; 205 } 206 rcu_read_unlock_bh(); 207 208 net_dbg_ratelimited("%s: No header cache and no neighbour!\n", 209 __func__); 210 kfree_skb(skb); 211 return -EINVAL; 212 } 213 214 static int ip_finish_output(struct sk_buff *skb) 215 { 216 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) 217 /* Policy lookup after SNAT yielded a new policy */ 218 if (skb_dst(skb)->xfrm != NULL) { 219 IPCB(skb)->flags |= IPSKB_REROUTED; 220 return dst_output(skb); 221 } 222 #endif 223 if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb)) 224 return ip_fragment(skb, ip_finish_output2); 225 else 226 return ip_finish_output2(skb); 227 } 228 229 int ip_mc_output(struct sock *sk, struct sk_buff *skb) 230 { 231 struct rtable *rt = skb_rtable(skb); 232 struct net_device *dev = rt->dst.dev; 233 234 /* 235 * If the indicated interface is up and running, send the packet. 236 */ 237 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); 238 239 skb->dev = dev; 240 skb->protocol = htons(ETH_P_IP); 241 242 /* 243 * Multicasts are looped back for other local users 244 */ 245 246 if (rt->rt_flags&RTCF_MULTICAST) { 247 if (sk_mc_loop(sk) 248 #ifdef CONFIG_IP_MROUTE 249 /* Small optimization: do not loopback not local frames, 250 which returned after forwarding; they will be dropped 251 by ip_mr_input in any case. 252 Note, that local frames are looped back to be delivered 253 to local recipients. 254 255 This check is duplicated in ip_mr_input at the moment. 256 */ 257 && 258 ((rt->rt_flags & RTCF_LOCAL) || 259 !(IPCB(skb)->flags & IPSKB_FORWARDED)) 260 #endif 261 ) { 262 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 263 if (newskb) 264 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 265 newskb, NULL, newskb->dev, 266 dev_loopback_xmit); 267 } 268 269 /* Multicasts with ttl 0 must not go beyond the host */ 270 271 if (ip_hdr(skb)->ttl == 0) { 272 kfree_skb(skb); 273 return 0; 274 } 275 } 276 277 if (rt->rt_flags&RTCF_BROADCAST) { 278 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 279 if (newskb) 280 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb, 281 NULL, newskb->dev, dev_loopback_xmit); 282 } 283 284 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, 285 skb->dev, ip_finish_output, 286 !(IPCB(skb)->flags & IPSKB_REROUTED)); 287 } 288 289 int ip_output(struct sock *sk, struct sk_buff *skb) 290 { 291 struct net_device *dev = skb_dst(skb)->dev; 292 293 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); 294 295 skb->dev = dev; 296 skb->protocol = htons(ETH_P_IP); 297 298 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev, 299 ip_finish_output, 300 !(IPCB(skb)->flags & IPSKB_REROUTED)); 301 } 302 303 /* 304 * copy saddr and daddr, possibly using 64bit load/stores 305 * Equivalent to : 306 * iph->saddr = fl4->saddr; 307 * iph->daddr = fl4->daddr; 308 */ 309 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) 310 { 311 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != 312 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); 313 memcpy(&iph->saddr, &fl4->saddr, 314 sizeof(fl4->saddr) + sizeof(fl4->daddr)); 315 } 316 317 /* Note: skb->sk can be different from sk, in case of tunnels */ 318 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) 319 { 320 struct inet_sock *inet = inet_sk(sk); 321 struct ip_options_rcu *inet_opt; 322 struct flowi4 *fl4; 323 struct rtable *rt; 324 struct iphdr *iph; 325 int res; 326 327 /* Skip all of this if the packet is already routed, 328 * f.e. by something like SCTP. 329 */ 330 rcu_read_lock(); 331 inet_opt = rcu_dereference(inet->inet_opt); 332 fl4 = &fl->u.ip4; 333 rt = skb_rtable(skb); 334 if (rt != NULL) 335 goto packet_routed; 336 337 /* Make sure we can route this packet. */ 338 rt = (struct rtable *)__sk_dst_check(sk, 0); 339 if (rt == NULL) { 340 __be32 daddr; 341 342 /* Use correct destination address if we have options. */ 343 daddr = inet->inet_daddr; 344 if (inet_opt && inet_opt->opt.srr) 345 daddr = inet_opt->opt.faddr; 346 347 /* If this fails, retransmit mechanism of transport layer will 348 * keep trying until route appears or the connection times 349 * itself out. 350 */ 351 rt = ip_route_output_ports(sock_net(sk), fl4, sk, 352 daddr, inet->inet_saddr, 353 inet->inet_dport, 354 inet->inet_sport, 355 sk->sk_protocol, 356 RT_CONN_FLAGS(sk), 357 sk->sk_bound_dev_if); 358 if (IS_ERR(rt)) 359 goto no_route; 360 sk_setup_caps(sk, &rt->dst); 361 } 362 skb_dst_set_noref(skb, &rt->dst); 363 364 packet_routed: 365 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) 366 goto no_route; 367 368 /* OK, we know where to send it, allocate and build IP header. */ 369 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 370 skb_reset_network_header(skb); 371 iph = ip_hdr(skb); 372 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); 373 if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df) 374 iph->frag_off = htons(IP_DF); 375 else 376 iph->frag_off = 0; 377 iph->ttl = ip_select_ttl(inet, &rt->dst); 378 iph->protocol = sk->sk_protocol; 379 ip_copy_addrs(iph, fl4); 380 381 /* Transport layer set skb->h.foo itself. */ 382 383 if (inet_opt && inet_opt->opt.optlen) { 384 iph->ihl += inet_opt->opt.optlen >> 2; 385 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 386 } 387 388 ip_select_ident_more(skb, &rt->dst, sk, 389 (skb_shinfo(skb)->gso_segs ?: 1) - 1); 390 391 /* TODO : should we use skb->sk here instead of sk ? */ 392 skb->priority = sk->sk_priority; 393 skb->mark = sk->sk_mark; 394 395 res = ip_local_out(skb); 396 rcu_read_unlock(); 397 return res; 398 399 no_route: 400 rcu_read_unlock(); 401 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); 402 kfree_skb(skb); 403 return -EHOSTUNREACH; 404 } 405 EXPORT_SYMBOL(ip_queue_xmit); 406 407 408 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) 409 { 410 to->pkt_type = from->pkt_type; 411 to->priority = from->priority; 412 to->protocol = from->protocol; 413 skb_dst_drop(to); 414 skb_dst_copy(to, from); 415 to->dev = from->dev; 416 to->mark = from->mark; 417 418 /* Copy the flags to each fragment. */ 419 IPCB(to)->flags = IPCB(from)->flags; 420 421 #ifdef CONFIG_NET_SCHED 422 to->tc_index = from->tc_index; 423 #endif 424 nf_copy(to, from); 425 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 426 to->ipvs_property = from->ipvs_property; 427 #endif 428 skb_copy_secmark(to, from); 429 } 430 431 /* 432 * This IP datagram is too large to be sent in one piece. Break it up into 433 * smaller pieces (each of size equal to IP header plus 434 * a block of the data of the original IP data part) that will yet fit in a 435 * single device frame, and queue such a frame for sending. 436 */ 437 438 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)) 439 { 440 struct iphdr *iph; 441 int ptr; 442 struct net_device *dev; 443 struct sk_buff *skb2; 444 unsigned int mtu, hlen, left, len, ll_rs; 445 int offset; 446 __be16 not_last_frag; 447 struct rtable *rt = skb_rtable(skb); 448 int err = 0; 449 450 dev = rt->dst.dev; 451 452 /* 453 * Point into the IP datagram header. 454 */ 455 456 iph = ip_hdr(skb); 457 458 mtu = ip_skb_dst_mtu(skb); 459 if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) || 460 (IPCB(skb)->frag_max_size && 461 IPCB(skb)->frag_max_size > mtu))) { 462 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 463 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 464 htonl(mtu)); 465 kfree_skb(skb); 466 return -EMSGSIZE; 467 } 468 469 /* 470 * Setup starting values. 471 */ 472 473 hlen = iph->ihl * 4; 474 mtu = mtu - hlen; /* Size of data space */ 475 #ifdef CONFIG_BRIDGE_NETFILTER 476 if (skb->nf_bridge) 477 mtu -= nf_bridge_mtu_reduction(skb); 478 #endif 479 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 480 481 /* When frag_list is given, use it. First, check its validity: 482 * some transformers could create wrong frag_list or break existing 483 * one, it is not prohibited. In this case fall back to copying. 484 * 485 * LATER: this step can be merged to real generation of fragments, 486 * we can switch to copy when see the first bad fragment. 487 */ 488 if (skb_has_frag_list(skb)) { 489 struct sk_buff *frag, *frag2; 490 int first_len = skb_pagelen(skb); 491 492 if (first_len - hlen > mtu || 493 ((first_len - hlen) & 7) || 494 ip_is_fragment(iph) || 495 skb_cloned(skb)) 496 goto slow_path; 497 498 skb_walk_frags(skb, frag) { 499 /* Correct geometry. */ 500 if (frag->len > mtu || 501 ((frag->len & 7) && frag->next) || 502 skb_headroom(frag) < hlen) 503 goto slow_path_clean; 504 505 /* Partially cloned skb? */ 506 if (skb_shared(frag)) 507 goto slow_path_clean; 508 509 BUG_ON(frag->sk); 510 if (skb->sk) { 511 frag->sk = skb->sk; 512 frag->destructor = sock_wfree; 513 } 514 skb->truesize -= frag->truesize; 515 } 516 517 /* Everything is OK. Generate! */ 518 519 err = 0; 520 offset = 0; 521 frag = skb_shinfo(skb)->frag_list; 522 skb_frag_list_init(skb); 523 skb->data_len = first_len - skb_headlen(skb); 524 skb->len = first_len; 525 iph->tot_len = htons(first_len); 526 iph->frag_off = htons(IP_MF); 527 ip_send_check(iph); 528 529 for (;;) { 530 /* Prepare header of the next frame, 531 * before previous one went down. */ 532 if (frag) { 533 frag->ip_summed = CHECKSUM_NONE; 534 skb_reset_transport_header(frag); 535 __skb_push(frag, hlen); 536 skb_reset_network_header(frag); 537 memcpy(skb_network_header(frag), iph, hlen); 538 iph = ip_hdr(frag); 539 iph->tot_len = htons(frag->len); 540 ip_copy_metadata(frag, skb); 541 if (offset == 0) 542 ip_options_fragment(frag); 543 offset += skb->len - hlen; 544 iph->frag_off = htons(offset>>3); 545 if (frag->next != NULL) 546 iph->frag_off |= htons(IP_MF); 547 /* Ready, complete checksum */ 548 ip_send_check(iph); 549 } 550 551 err = output(skb); 552 553 if (!err) 554 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); 555 if (err || !frag) 556 break; 557 558 skb = frag; 559 frag = skb->next; 560 skb->next = NULL; 561 } 562 563 if (err == 0) { 564 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); 565 return 0; 566 } 567 568 while (frag) { 569 skb = frag->next; 570 kfree_skb(frag); 571 frag = skb; 572 } 573 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 574 return err; 575 576 slow_path_clean: 577 skb_walk_frags(skb, frag2) { 578 if (frag2 == frag) 579 break; 580 frag2->sk = NULL; 581 frag2->destructor = NULL; 582 skb->truesize += frag2->truesize; 583 } 584 } 585 586 slow_path: 587 /* for offloaded checksums cleanup checksum before fragmentation */ 588 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb)) 589 goto fail; 590 iph = ip_hdr(skb); 591 592 left = skb->len - hlen; /* Space per frame */ 593 ptr = hlen; /* Where to start from */ 594 595 /* for bridged IP traffic encapsulated inside f.e. a vlan header, 596 * we need to make room for the encapsulating header 597 */ 598 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb)); 599 600 /* 601 * Fragment the datagram. 602 */ 603 604 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 605 not_last_frag = iph->frag_off & htons(IP_MF); 606 607 /* 608 * Keep copying data until we run out. 609 */ 610 611 while (left > 0) { 612 len = left; 613 /* IF: it doesn't fit, use 'mtu' - the data space left */ 614 if (len > mtu) 615 len = mtu; 616 /* IF: we are not sending up to and including the packet end 617 then align the next start on an eight byte boundary */ 618 if (len < left) { 619 len &= ~7; 620 } 621 /* 622 * Allocate buffer. 623 */ 624 625 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { 626 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); 627 err = -ENOMEM; 628 goto fail; 629 } 630 631 /* 632 * Set up data on packet 633 */ 634 635 ip_copy_metadata(skb2, skb); 636 skb_reserve(skb2, ll_rs); 637 skb_put(skb2, len + hlen); 638 skb_reset_network_header(skb2); 639 skb2->transport_header = skb2->network_header + hlen; 640 641 /* 642 * Charge the memory for the fragment to any owner 643 * it might possess 644 */ 645 646 if (skb->sk) 647 skb_set_owner_w(skb2, skb->sk); 648 649 /* 650 * Copy the packet header into the new buffer. 651 */ 652 653 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); 654 655 /* 656 * Copy a block of the IP datagram. 657 */ 658 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) 659 BUG(); 660 left -= len; 661 662 /* 663 * Fill in the new header fields. 664 */ 665 iph = ip_hdr(skb2); 666 iph->frag_off = htons((offset >> 3)); 667 668 /* ANK: dirty, but effective trick. Upgrade options only if 669 * the segment to be fragmented was THE FIRST (otherwise, 670 * options are already fixed) and make it ONCE 671 * on the initial skb, so that all the following fragments 672 * will inherit fixed options. 673 */ 674 if (offset == 0) 675 ip_options_fragment(skb); 676 677 /* 678 * Added AC : If we are fragmenting a fragment that's not the 679 * last fragment then keep MF on each bit 680 */ 681 if (left > 0 || not_last_frag) 682 iph->frag_off |= htons(IP_MF); 683 ptr += len; 684 offset += len; 685 686 /* 687 * Put this fragment into the sending queue. 688 */ 689 iph->tot_len = htons(len + hlen); 690 691 ip_send_check(iph); 692 693 err = output(skb2); 694 if (err) 695 goto fail; 696 697 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); 698 } 699 consume_skb(skb); 700 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); 701 return err; 702 703 fail: 704 kfree_skb(skb); 705 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 706 return err; 707 } 708 EXPORT_SYMBOL(ip_fragment); 709 710 int 711 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 712 { 713 struct iovec *iov = from; 714 715 if (skb->ip_summed == CHECKSUM_PARTIAL) { 716 if (memcpy_fromiovecend(to, iov, offset, len) < 0) 717 return -EFAULT; 718 } else { 719 __wsum csum = 0; 720 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) 721 return -EFAULT; 722 skb->csum = csum_block_add(skb->csum, csum, odd); 723 } 724 return 0; 725 } 726 EXPORT_SYMBOL(ip_generic_getfrag); 727 728 static inline __wsum 729 csum_page(struct page *page, int offset, int copy) 730 { 731 char *kaddr; 732 __wsum csum; 733 kaddr = kmap(page); 734 csum = csum_partial(kaddr + offset, copy, 0); 735 kunmap(page); 736 return csum; 737 } 738 739 static inline int ip_ufo_append_data(struct sock *sk, 740 struct sk_buff_head *queue, 741 int getfrag(void *from, char *to, int offset, int len, 742 int odd, struct sk_buff *skb), 743 void *from, int length, int hh_len, int fragheaderlen, 744 int transhdrlen, int maxfraglen, unsigned int flags) 745 { 746 struct sk_buff *skb; 747 int err; 748 749 /* There is support for UDP fragmentation offload by network 750 * device, so create one single skb packet containing complete 751 * udp datagram 752 */ 753 if ((skb = skb_peek_tail(queue)) == NULL) { 754 skb = sock_alloc_send_skb(sk, 755 hh_len + fragheaderlen + transhdrlen + 20, 756 (flags & MSG_DONTWAIT), &err); 757 758 if (skb == NULL) 759 return err; 760 761 /* reserve space for Hardware header */ 762 skb_reserve(skb, hh_len); 763 764 /* create space for UDP/IP header */ 765 skb_put(skb, fragheaderlen + transhdrlen); 766 767 /* initialize network header pointer */ 768 skb_reset_network_header(skb); 769 770 /* initialize protocol header pointer */ 771 skb->transport_header = skb->network_header + fragheaderlen; 772 773 skb->csum = 0; 774 775 776 __skb_queue_tail(queue, skb); 777 } else if (skb_is_gso(skb)) { 778 goto append; 779 } 780 781 skb->ip_summed = CHECKSUM_PARTIAL; 782 /* specify the length of each IP datagram fragment */ 783 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen; 784 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 785 786 append: 787 return skb_append_datato_frags(sk, skb, getfrag, from, 788 (length - transhdrlen)); 789 } 790 791 static int __ip_append_data(struct sock *sk, 792 struct flowi4 *fl4, 793 struct sk_buff_head *queue, 794 struct inet_cork *cork, 795 struct page_frag *pfrag, 796 int getfrag(void *from, char *to, int offset, 797 int len, int odd, struct sk_buff *skb), 798 void *from, int length, int transhdrlen, 799 unsigned int flags) 800 { 801 struct inet_sock *inet = inet_sk(sk); 802 struct sk_buff *skb; 803 804 struct ip_options *opt = cork->opt; 805 int hh_len; 806 int exthdrlen; 807 int mtu; 808 int copy; 809 int err; 810 int offset = 0; 811 unsigned int maxfraglen, fragheaderlen, maxnonfragsize; 812 int csummode = CHECKSUM_NONE; 813 struct rtable *rt = (struct rtable *)cork->dst; 814 815 skb = skb_peek_tail(queue); 816 817 exthdrlen = !skb ? rt->dst.header_len : 0; 818 mtu = cork->fragsize; 819 820 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 821 822 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 823 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 824 maxnonfragsize = ip_sk_local_df(sk) ? 0xFFFF : mtu; 825 826 if (cork->length + length > maxnonfragsize - fragheaderlen) { 827 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 828 mtu - (opt ? opt->optlen : 0)); 829 return -EMSGSIZE; 830 } 831 832 /* 833 * transhdrlen > 0 means that this is the first fragment and we wish 834 * it won't be fragmented in the future. 835 */ 836 if (transhdrlen && 837 length + fragheaderlen <= mtu && 838 rt->dst.dev->features & NETIF_F_V4_CSUM && 839 !exthdrlen) 840 csummode = CHECKSUM_PARTIAL; 841 842 cork->length += length; 843 if (((length > mtu) || (skb && skb_is_gso(skb))) && 844 (sk->sk_protocol == IPPROTO_UDP) && 845 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) { 846 err = ip_ufo_append_data(sk, queue, getfrag, from, length, 847 hh_len, fragheaderlen, transhdrlen, 848 maxfraglen, flags); 849 if (err) 850 goto error; 851 return 0; 852 } 853 854 /* So, what's going on in the loop below? 855 * 856 * We use calculated fragment length to generate chained skb, 857 * each of segments is IP fragment ready for sending to network after 858 * adding appropriate IP header. 859 */ 860 861 if (!skb) 862 goto alloc_new_skb; 863 864 while (length > 0) { 865 /* Check if the remaining data fits into current packet. */ 866 copy = mtu - skb->len; 867 if (copy < length) 868 copy = maxfraglen - skb->len; 869 if (copy <= 0) { 870 char *data; 871 unsigned int datalen; 872 unsigned int fraglen; 873 unsigned int fraggap; 874 unsigned int alloclen; 875 struct sk_buff *skb_prev; 876 alloc_new_skb: 877 skb_prev = skb; 878 if (skb_prev) 879 fraggap = skb_prev->len - maxfraglen; 880 else 881 fraggap = 0; 882 883 /* 884 * If remaining data exceeds the mtu, 885 * we know we need more fragment(s). 886 */ 887 datalen = length + fraggap; 888 if (datalen > mtu - fragheaderlen) 889 datalen = maxfraglen - fragheaderlen; 890 fraglen = datalen + fragheaderlen; 891 892 if ((flags & MSG_MORE) && 893 !(rt->dst.dev->features&NETIF_F_SG)) 894 alloclen = mtu; 895 else 896 alloclen = fraglen; 897 898 alloclen += exthdrlen; 899 900 /* The last fragment gets additional space at tail. 901 * Note, with MSG_MORE we overallocate on fragments, 902 * because we have no idea what fragment will be 903 * the last. 904 */ 905 if (datalen == length + fraggap) 906 alloclen += rt->dst.trailer_len; 907 908 if (transhdrlen) { 909 skb = sock_alloc_send_skb(sk, 910 alloclen + hh_len + 15, 911 (flags & MSG_DONTWAIT), &err); 912 } else { 913 skb = NULL; 914 if (atomic_read(&sk->sk_wmem_alloc) <= 915 2 * sk->sk_sndbuf) 916 skb = sock_wmalloc(sk, 917 alloclen + hh_len + 15, 1, 918 sk->sk_allocation); 919 if (unlikely(skb == NULL)) 920 err = -ENOBUFS; 921 else 922 /* only the initial fragment is 923 time stamped */ 924 cork->tx_flags = 0; 925 } 926 if (skb == NULL) 927 goto error; 928 929 /* 930 * Fill in the control structures 931 */ 932 skb->ip_summed = csummode; 933 skb->csum = 0; 934 skb_reserve(skb, hh_len); 935 skb_shinfo(skb)->tx_flags = cork->tx_flags; 936 937 /* 938 * Find where to start putting bytes. 939 */ 940 data = skb_put(skb, fraglen + exthdrlen); 941 skb_set_network_header(skb, exthdrlen); 942 skb->transport_header = (skb->network_header + 943 fragheaderlen); 944 data += fragheaderlen + exthdrlen; 945 946 if (fraggap) { 947 skb->csum = skb_copy_and_csum_bits( 948 skb_prev, maxfraglen, 949 data + transhdrlen, fraggap, 0); 950 skb_prev->csum = csum_sub(skb_prev->csum, 951 skb->csum); 952 data += fraggap; 953 pskb_trim_unique(skb_prev, maxfraglen); 954 } 955 956 copy = datalen - transhdrlen - fraggap; 957 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 958 err = -EFAULT; 959 kfree_skb(skb); 960 goto error; 961 } 962 963 offset += copy; 964 length -= datalen - fraggap; 965 transhdrlen = 0; 966 exthdrlen = 0; 967 csummode = CHECKSUM_NONE; 968 969 /* 970 * Put the packet on the pending queue. 971 */ 972 __skb_queue_tail(queue, skb); 973 continue; 974 } 975 976 if (copy > length) 977 copy = length; 978 979 if (!(rt->dst.dev->features&NETIF_F_SG)) { 980 unsigned int off; 981 982 off = skb->len; 983 if (getfrag(from, skb_put(skb, copy), 984 offset, copy, off, skb) < 0) { 985 __skb_trim(skb, off); 986 err = -EFAULT; 987 goto error; 988 } 989 } else { 990 int i = skb_shinfo(skb)->nr_frags; 991 992 err = -ENOMEM; 993 if (!sk_page_frag_refill(sk, pfrag)) 994 goto error; 995 996 if (!skb_can_coalesce(skb, i, pfrag->page, 997 pfrag->offset)) { 998 err = -EMSGSIZE; 999 if (i == MAX_SKB_FRAGS) 1000 goto error; 1001 1002 __skb_fill_page_desc(skb, i, pfrag->page, 1003 pfrag->offset, 0); 1004 skb_shinfo(skb)->nr_frags = ++i; 1005 get_page(pfrag->page); 1006 } 1007 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1008 if (getfrag(from, 1009 page_address(pfrag->page) + pfrag->offset, 1010 offset, copy, skb->len, skb) < 0) 1011 goto error_efault; 1012 1013 pfrag->offset += copy; 1014 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1015 skb->len += copy; 1016 skb->data_len += copy; 1017 skb->truesize += copy; 1018 atomic_add(copy, &sk->sk_wmem_alloc); 1019 } 1020 offset += copy; 1021 length -= copy; 1022 } 1023 1024 return 0; 1025 1026 error_efault: 1027 err = -EFAULT; 1028 error: 1029 cork->length -= length; 1030 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1031 return err; 1032 } 1033 1034 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 1035 struct ipcm_cookie *ipc, struct rtable **rtp) 1036 { 1037 struct ip_options_rcu *opt; 1038 struct rtable *rt; 1039 1040 /* 1041 * setup for corking. 1042 */ 1043 opt = ipc->opt; 1044 if (opt) { 1045 if (cork->opt == NULL) { 1046 cork->opt = kmalloc(sizeof(struct ip_options) + 40, 1047 sk->sk_allocation); 1048 if (unlikely(cork->opt == NULL)) 1049 return -ENOBUFS; 1050 } 1051 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 1052 cork->flags |= IPCORK_OPT; 1053 cork->addr = ipc->addr; 1054 } 1055 rt = *rtp; 1056 if (unlikely(!rt)) 1057 return -EFAULT; 1058 /* 1059 * We steal reference to this route, caller should not release it 1060 */ 1061 *rtp = NULL; 1062 cork->fragsize = ip_sk_use_pmtu(sk) ? 1063 dst_mtu(&rt->dst) : rt->dst.dev->mtu; 1064 cork->dst = &rt->dst; 1065 cork->length = 0; 1066 cork->ttl = ipc->ttl; 1067 cork->tos = ipc->tos; 1068 cork->priority = ipc->priority; 1069 cork->tx_flags = ipc->tx_flags; 1070 1071 return 0; 1072 } 1073 1074 /* 1075 * ip_append_data() and ip_append_page() can make one large IP datagram 1076 * from many pieces of data. Each pieces will be holded on the socket 1077 * until ip_push_pending_frames() is called. Each piece can be a page 1078 * or non-page data. 1079 * 1080 * Not only UDP, other transport protocols - e.g. raw sockets - can use 1081 * this interface potentially. 1082 * 1083 * LATER: length must be adjusted by pad at tail, when it is required. 1084 */ 1085 int ip_append_data(struct sock *sk, struct flowi4 *fl4, 1086 int getfrag(void *from, char *to, int offset, int len, 1087 int odd, struct sk_buff *skb), 1088 void *from, int length, int transhdrlen, 1089 struct ipcm_cookie *ipc, struct rtable **rtp, 1090 unsigned int flags) 1091 { 1092 struct inet_sock *inet = inet_sk(sk); 1093 int err; 1094 1095 if (flags&MSG_PROBE) 1096 return 0; 1097 1098 if (skb_queue_empty(&sk->sk_write_queue)) { 1099 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 1100 if (err) 1101 return err; 1102 } else { 1103 transhdrlen = 0; 1104 } 1105 1106 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, 1107 sk_page_frag(sk), getfrag, 1108 from, length, transhdrlen, flags); 1109 } 1110 1111 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 1112 int offset, size_t size, int flags) 1113 { 1114 struct inet_sock *inet = inet_sk(sk); 1115 struct sk_buff *skb; 1116 struct rtable *rt; 1117 struct ip_options *opt = NULL; 1118 struct inet_cork *cork; 1119 int hh_len; 1120 int mtu; 1121 int len; 1122 int err; 1123 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; 1124 1125 if (inet->hdrincl) 1126 return -EPERM; 1127 1128 if (flags&MSG_PROBE) 1129 return 0; 1130 1131 if (skb_queue_empty(&sk->sk_write_queue)) 1132 return -EINVAL; 1133 1134 cork = &inet->cork.base; 1135 rt = (struct rtable *)cork->dst; 1136 if (cork->flags & IPCORK_OPT) 1137 opt = cork->opt; 1138 1139 if (!(rt->dst.dev->features&NETIF_F_SG)) 1140 return -EOPNOTSUPP; 1141 1142 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1143 mtu = cork->fragsize; 1144 1145 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1146 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1147 maxnonfragsize = ip_sk_local_df(sk) ? 0xFFFF : mtu; 1148 1149 if (cork->length + size > maxnonfragsize - fragheaderlen) { 1150 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 1151 mtu - (opt ? opt->optlen : 0)); 1152 return -EMSGSIZE; 1153 } 1154 1155 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) 1156 return -EINVAL; 1157 1158 cork->length += size; 1159 if ((size + skb->len > mtu) && 1160 (sk->sk_protocol == IPPROTO_UDP) && 1161 (rt->dst.dev->features & NETIF_F_UFO)) { 1162 skb_shinfo(skb)->gso_size = mtu - fragheaderlen; 1163 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 1164 } 1165 1166 1167 while (size > 0) { 1168 int i; 1169 1170 if (skb_is_gso(skb)) 1171 len = size; 1172 else { 1173 1174 /* Check if the remaining data fits into current packet. */ 1175 len = mtu - skb->len; 1176 if (len < size) 1177 len = maxfraglen - skb->len; 1178 } 1179 if (len <= 0) { 1180 struct sk_buff *skb_prev; 1181 int alloclen; 1182 1183 skb_prev = skb; 1184 fraggap = skb_prev->len - maxfraglen; 1185 1186 alloclen = fragheaderlen + hh_len + fraggap + 15; 1187 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 1188 if (unlikely(!skb)) { 1189 err = -ENOBUFS; 1190 goto error; 1191 } 1192 1193 /* 1194 * Fill in the control structures 1195 */ 1196 skb->ip_summed = CHECKSUM_NONE; 1197 skb->csum = 0; 1198 skb_reserve(skb, hh_len); 1199 1200 /* 1201 * Find where to start putting bytes. 1202 */ 1203 skb_put(skb, fragheaderlen + fraggap); 1204 skb_reset_network_header(skb); 1205 skb->transport_header = (skb->network_header + 1206 fragheaderlen); 1207 if (fraggap) { 1208 skb->csum = skb_copy_and_csum_bits(skb_prev, 1209 maxfraglen, 1210 skb_transport_header(skb), 1211 fraggap, 0); 1212 skb_prev->csum = csum_sub(skb_prev->csum, 1213 skb->csum); 1214 pskb_trim_unique(skb_prev, maxfraglen); 1215 } 1216 1217 /* 1218 * Put the packet on the pending queue. 1219 */ 1220 __skb_queue_tail(&sk->sk_write_queue, skb); 1221 continue; 1222 } 1223 1224 i = skb_shinfo(skb)->nr_frags; 1225 if (len > size) 1226 len = size; 1227 if (skb_can_coalesce(skb, i, page, offset)) { 1228 skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len); 1229 } else if (i < MAX_SKB_FRAGS) { 1230 get_page(page); 1231 skb_fill_page_desc(skb, i, page, offset, len); 1232 } else { 1233 err = -EMSGSIZE; 1234 goto error; 1235 } 1236 1237 if (skb->ip_summed == CHECKSUM_NONE) { 1238 __wsum csum; 1239 csum = csum_page(page, offset, len); 1240 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1241 } 1242 1243 skb->len += len; 1244 skb->data_len += len; 1245 skb->truesize += len; 1246 atomic_add(len, &sk->sk_wmem_alloc); 1247 offset += len; 1248 size -= len; 1249 } 1250 return 0; 1251 1252 error: 1253 cork->length -= size; 1254 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1255 return err; 1256 } 1257 1258 static void ip_cork_release(struct inet_cork *cork) 1259 { 1260 cork->flags &= ~IPCORK_OPT; 1261 kfree(cork->opt); 1262 cork->opt = NULL; 1263 dst_release(cork->dst); 1264 cork->dst = NULL; 1265 } 1266 1267 /* 1268 * Combined all pending IP fragments on the socket as one IP datagram 1269 * and push them out. 1270 */ 1271 struct sk_buff *__ip_make_skb(struct sock *sk, 1272 struct flowi4 *fl4, 1273 struct sk_buff_head *queue, 1274 struct inet_cork *cork) 1275 { 1276 struct sk_buff *skb, *tmp_skb; 1277 struct sk_buff **tail_skb; 1278 struct inet_sock *inet = inet_sk(sk); 1279 struct net *net = sock_net(sk); 1280 struct ip_options *opt = NULL; 1281 struct rtable *rt = (struct rtable *)cork->dst; 1282 struct iphdr *iph; 1283 __be16 df = 0; 1284 __u8 ttl; 1285 1286 if ((skb = __skb_dequeue(queue)) == NULL) 1287 goto out; 1288 tail_skb = &(skb_shinfo(skb)->frag_list); 1289 1290 /* move skb->data to ip header from ext header */ 1291 if (skb->data < skb_network_header(skb)) 1292 __skb_pull(skb, skb_network_offset(skb)); 1293 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1294 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1295 *tail_skb = tmp_skb; 1296 tail_skb = &(tmp_skb->next); 1297 skb->len += tmp_skb->len; 1298 skb->data_len += tmp_skb->len; 1299 skb->truesize += tmp_skb->truesize; 1300 tmp_skb->destructor = NULL; 1301 tmp_skb->sk = NULL; 1302 } 1303 1304 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 1305 * to fragment the frame generated here. No matter, what transforms 1306 * how transforms change size of the packet, it will come out. 1307 */ 1308 skb->local_df = ip_sk_local_df(sk); 1309 1310 /* DF bit is set when we want to see DF on outgoing frames. 1311 * If local_df is set too, we still allow to fragment this frame 1312 * locally. */ 1313 if (inet->pmtudisc == IP_PMTUDISC_DO || 1314 inet->pmtudisc == IP_PMTUDISC_PROBE || 1315 (skb->len <= dst_mtu(&rt->dst) && 1316 ip_dont_fragment(sk, &rt->dst))) 1317 df = htons(IP_DF); 1318 1319 if (cork->flags & IPCORK_OPT) 1320 opt = cork->opt; 1321 1322 if (cork->ttl != 0) 1323 ttl = cork->ttl; 1324 else if (rt->rt_type == RTN_MULTICAST) 1325 ttl = inet->mc_ttl; 1326 else 1327 ttl = ip_select_ttl(inet, &rt->dst); 1328 1329 iph = ip_hdr(skb); 1330 iph->version = 4; 1331 iph->ihl = 5; 1332 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; 1333 iph->frag_off = df; 1334 iph->ttl = ttl; 1335 iph->protocol = sk->sk_protocol; 1336 ip_copy_addrs(iph, fl4); 1337 ip_select_ident(skb, &rt->dst, sk); 1338 1339 if (opt) { 1340 iph->ihl += opt->optlen>>2; 1341 ip_options_build(skb, opt, cork->addr, rt, 0); 1342 } 1343 1344 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; 1345 skb->mark = sk->sk_mark; 1346 /* 1347 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1348 * on dst refcount 1349 */ 1350 cork->dst = NULL; 1351 skb_dst_set(skb, &rt->dst); 1352 1353 if (iph->protocol == IPPROTO_ICMP) 1354 icmp_out_count(net, ((struct icmphdr *) 1355 skb_transport_header(skb))->type); 1356 1357 ip_cork_release(cork); 1358 out: 1359 return skb; 1360 } 1361 1362 int ip_send_skb(struct net *net, struct sk_buff *skb) 1363 { 1364 int err; 1365 1366 err = ip_local_out(skb); 1367 if (err) { 1368 if (err > 0) 1369 err = net_xmit_errno(err); 1370 if (err) 1371 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 1372 } 1373 1374 return err; 1375 } 1376 1377 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 1378 { 1379 struct sk_buff *skb; 1380 1381 skb = ip_finish_skb(sk, fl4); 1382 if (!skb) 1383 return 0; 1384 1385 /* Netfilter gets whole the not fragmented skb. */ 1386 return ip_send_skb(sock_net(sk), skb); 1387 } 1388 1389 /* 1390 * Throw away all pending data on the socket. 1391 */ 1392 static void __ip_flush_pending_frames(struct sock *sk, 1393 struct sk_buff_head *queue, 1394 struct inet_cork *cork) 1395 { 1396 struct sk_buff *skb; 1397 1398 while ((skb = __skb_dequeue_tail(queue)) != NULL) 1399 kfree_skb(skb); 1400 1401 ip_cork_release(cork); 1402 } 1403 1404 void ip_flush_pending_frames(struct sock *sk) 1405 { 1406 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 1407 } 1408 1409 struct sk_buff *ip_make_skb(struct sock *sk, 1410 struct flowi4 *fl4, 1411 int getfrag(void *from, char *to, int offset, 1412 int len, int odd, struct sk_buff *skb), 1413 void *from, int length, int transhdrlen, 1414 struct ipcm_cookie *ipc, struct rtable **rtp, 1415 unsigned int flags) 1416 { 1417 struct inet_cork cork; 1418 struct sk_buff_head queue; 1419 int err; 1420 1421 if (flags & MSG_PROBE) 1422 return NULL; 1423 1424 __skb_queue_head_init(&queue); 1425 1426 cork.flags = 0; 1427 cork.addr = 0; 1428 cork.opt = NULL; 1429 err = ip_setup_cork(sk, &cork, ipc, rtp); 1430 if (err) 1431 return ERR_PTR(err); 1432 1433 err = __ip_append_data(sk, fl4, &queue, &cork, 1434 ¤t->task_frag, getfrag, 1435 from, length, transhdrlen, flags); 1436 if (err) { 1437 __ip_flush_pending_frames(sk, &queue, &cork); 1438 return ERR_PTR(err); 1439 } 1440 1441 return __ip_make_skb(sk, fl4, &queue, &cork); 1442 } 1443 1444 /* 1445 * Fetch data from kernel space and fill in checksum if needed. 1446 */ 1447 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 1448 int len, int odd, struct sk_buff *skb) 1449 { 1450 __wsum csum; 1451 1452 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); 1453 skb->csum = csum_block_add(skb->csum, csum, odd); 1454 return 0; 1455 } 1456 1457 /* 1458 * Generic function to send a packet as reply to another packet. 1459 * Used to send some TCP resets/acks so far. 1460 * 1461 * Use a fake percpu inet socket to avoid false sharing and contention. 1462 */ 1463 static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = { 1464 .sk = { 1465 .__sk_common = { 1466 .skc_refcnt = ATOMIC_INIT(1), 1467 }, 1468 .sk_wmem_alloc = ATOMIC_INIT(1), 1469 .sk_allocation = GFP_ATOMIC, 1470 .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE), 1471 }, 1472 .pmtudisc = IP_PMTUDISC_WANT, 1473 .uc_ttl = -1, 1474 }; 1475 1476 void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr, 1477 __be32 saddr, const struct ip_reply_arg *arg, 1478 unsigned int len) 1479 { 1480 struct ip_options_data replyopts; 1481 struct ipcm_cookie ipc; 1482 struct flowi4 fl4; 1483 struct rtable *rt = skb_rtable(skb); 1484 struct sk_buff *nskb; 1485 struct sock *sk; 1486 struct inet_sock *inet; 1487 1488 if (ip_options_echo(&replyopts.opt.opt, skb)) 1489 return; 1490 1491 ipc.addr = daddr; 1492 ipc.opt = NULL; 1493 ipc.tx_flags = 0; 1494 ipc.ttl = 0; 1495 ipc.tos = -1; 1496 1497 if (replyopts.opt.opt.optlen) { 1498 ipc.opt = &replyopts.opt; 1499 1500 if (replyopts.opt.opt.srr) 1501 daddr = replyopts.opt.opt.faddr; 1502 } 1503 1504 flowi4_init_output(&fl4, arg->bound_dev_if, 0, 1505 RT_TOS(arg->tos), 1506 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, 1507 ip_reply_arg_flowi_flags(arg), 1508 daddr, saddr, 1509 tcp_hdr(skb)->source, tcp_hdr(skb)->dest); 1510 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 1511 rt = ip_route_output_key(net, &fl4); 1512 if (IS_ERR(rt)) 1513 return; 1514 1515 inet = &get_cpu_var(unicast_sock); 1516 1517 inet->tos = arg->tos; 1518 sk = &inet->sk; 1519 sk->sk_priority = skb->priority; 1520 sk->sk_protocol = ip_hdr(skb)->protocol; 1521 sk->sk_bound_dev_if = arg->bound_dev_if; 1522 sock_net_set(sk, net); 1523 __skb_queue_head_init(&sk->sk_write_queue); 1524 sk->sk_sndbuf = sysctl_wmem_default; 1525 ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0, 1526 &ipc, &rt, MSG_DONTWAIT); 1527 nskb = skb_peek(&sk->sk_write_queue); 1528 if (nskb) { 1529 if (arg->csumoffset >= 0) 1530 *((__sum16 *)skb_transport_header(nskb) + 1531 arg->csumoffset) = csum_fold(csum_add(nskb->csum, 1532 arg->csum)); 1533 nskb->ip_summed = CHECKSUM_NONE; 1534 skb_orphan(nskb); 1535 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb)); 1536 ip_push_pending_frames(sk, &fl4); 1537 } 1538 1539 put_cpu_var(unicast_sock); 1540 1541 ip_rt_put(rt); 1542 } 1543 1544 void __init ip_init(void) 1545 { 1546 ip_rt_init(); 1547 inet_initpeers(); 1548 1549 #if defined(CONFIG_IP_MULTICAST) 1550 igmp_mc_init(); 1551 #endif 1552 } 1553