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(struct sk_buff *skb) 105 { 106 int err; 107 108 err = __ip_local_out(skb); 109 if (likely(err == 1)) 110 err = dst_output(skb); 111 112 return err; 113 } 114 EXPORT_SYMBOL_GPL(ip_local_out); 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 sk_buff *skb) 230 { 231 struct sock *sk = skb->sk; 232 struct rtable *rt = skb_rtable(skb); 233 struct net_device *dev = rt->dst.dev; 234 235 /* 236 * If the indicated interface is up and running, send the packet. 237 */ 238 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); 239 240 skb->dev = dev; 241 skb->protocol = htons(ETH_P_IP); 242 243 /* 244 * Multicasts are looped back for other local users 245 */ 246 247 if (rt->rt_flags&RTCF_MULTICAST) { 248 if (sk_mc_loop(sk) 249 #ifdef CONFIG_IP_MROUTE 250 /* Small optimization: do not loopback not local frames, 251 which returned after forwarding; they will be dropped 252 by ip_mr_input in any case. 253 Note, that local frames are looped back to be delivered 254 to local recipients. 255 256 This check is duplicated in ip_mr_input at the moment. 257 */ 258 && 259 ((rt->rt_flags & RTCF_LOCAL) || 260 !(IPCB(skb)->flags & IPSKB_FORWARDED)) 261 #endif 262 ) { 263 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 264 if (newskb) 265 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 266 newskb, NULL, newskb->dev, 267 dev_loopback_xmit); 268 } 269 270 /* Multicasts with ttl 0 must not go beyond the host */ 271 272 if (ip_hdr(skb)->ttl == 0) { 273 kfree_skb(skb); 274 return 0; 275 } 276 } 277 278 if (rt->rt_flags&RTCF_BROADCAST) { 279 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 280 if (newskb) 281 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb, 282 NULL, newskb->dev, dev_loopback_xmit); 283 } 284 285 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, 286 skb->dev, ip_finish_output, 287 !(IPCB(skb)->flags & IPSKB_REROUTED)); 288 } 289 290 int ip_output(struct sk_buff *skb) 291 { 292 struct net_device *dev = skb_dst(skb)->dev; 293 294 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); 295 296 skb->dev = dev; 297 skb->protocol = htons(ETH_P_IP); 298 299 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev, 300 ip_finish_output, 301 !(IPCB(skb)->flags & IPSKB_REROUTED)); 302 } 303 304 /* 305 * copy saddr and daddr, possibly using 64bit load/stores 306 * Equivalent to : 307 * iph->saddr = fl4->saddr; 308 * iph->daddr = fl4->daddr; 309 */ 310 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) 311 { 312 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != 313 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); 314 memcpy(&iph->saddr, &fl4->saddr, 315 sizeof(fl4->saddr) + sizeof(fl4->daddr)); 316 } 317 318 int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl) 319 { 320 struct sock *sk = skb->sk; 321 struct inet_sock *inet = inet_sk(sk); 322 struct ip_options_rcu *inet_opt; 323 struct flowi4 *fl4; 324 struct rtable *rt; 325 struct iphdr *iph; 326 int res; 327 328 /* Skip all of this if the packet is already routed, 329 * f.e. by something like SCTP. 330 */ 331 rcu_read_lock(); 332 inet_opt = rcu_dereference(inet->inet_opt); 333 fl4 = &fl->u.ip4; 334 rt = skb_rtable(skb); 335 if (rt != NULL) 336 goto packet_routed; 337 338 /* Make sure we can route this packet. */ 339 rt = (struct rtable *)__sk_dst_check(sk, 0); 340 if (rt == NULL) { 341 __be32 daddr; 342 343 /* Use correct destination address if we have options. */ 344 daddr = inet->inet_daddr; 345 if (inet_opt && inet_opt->opt.srr) 346 daddr = inet_opt->opt.faddr; 347 348 /* If this fails, retransmit mechanism of transport layer will 349 * keep trying until route appears or the connection times 350 * itself out. 351 */ 352 rt = ip_route_output_ports(sock_net(sk), fl4, sk, 353 daddr, inet->inet_saddr, 354 inet->inet_dport, 355 inet->inet_sport, 356 sk->sk_protocol, 357 RT_CONN_FLAGS(sk), 358 sk->sk_bound_dev_if); 359 if (IS_ERR(rt)) 360 goto no_route; 361 sk_setup_caps(sk, &rt->dst); 362 } 363 skb_dst_set_noref(skb, &rt->dst); 364 365 packet_routed: 366 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) 367 goto no_route; 368 369 /* OK, we know where to send it, allocate and build IP header. */ 370 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 371 skb_reset_network_header(skb); 372 iph = ip_hdr(skb); 373 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); 374 if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df) 375 iph->frag_off = htons(IP_DF); 376 else 377 iph->frag_off = 0; 378 iph->ttl = ip_select_ttl(inet, &rt->dst); 379 iph->protocol = sk->sk_protocol; 380 ip_copy_addrs(iph, fl4); 381 382 /* Transport layer set skb->h.foo itself. */ 383 384 if (inet_opt && inet_opt->opt.optlen) { 385 iph->ihl += inet_opt->opt.optlen >> 2; 386 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 387 } 388 389 ip_select_ident_more(skb, &rt->dst, sk, 390 (skb_shinfo(skb)->gso_segs ?: 1) - 1); 391 392 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 bool forwarding = IPCB(skb)->flags & IPSKB_FORWARDED; 450 451 dev = rt->dst.dev; 452 453 /* 454 * Point into the IP datagram header. 455 */ 456 457 iph = ip_hdr(skb); 458 459 mtu = ip_dst_mtu_maybe_forward(&rt->dst, forwarding); 460 if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) || 461 (IPCB(skb)->frag_max_size && 462 IPCB(skb)->frag_max_size > mtu))) { 463 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 464 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 465 htonl(mtu)); 466 kfree_skb(skb); 467 return -EMSGSIZE; 468 } 469 470 /* 471 * Setup starting values. 472 */ 473 474 hlen = iph->ihl * 4; 475 mtu = mtu - hlen; /* Size of data space */ 476 #ifdef CONFIG_BRIDGE_NETFILTER 477 if (skb->nf_bridge) 478 mtu -= nf_bridge_mtu_reduction(skb); 479 #endif 480 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 481 482 /* When frag_list is given, use it. First, check its validity: 483 * some transformers could create wrong frag_list or break existing 484 * one, it is not prohibited. In this case fall back to copying. 485 * 486 * LATER: this step can be merged to real generation of fragments, 487 * we can switch to copy when see the first bad fragment. 488 */ 489 if (skb_has_frag_list(skb)) { 490 struct sk_buff *frag, *frag2; 491 int first_len = skb_pagelen(skb); 492 493 if (first_len - hlen > mtu || 494 ((first_len - hlen) & 7) || 495 ip_is_fragment(iph) || 496 skb_cloned(skb)) 497 goto slow_path; 498 499 skb_walk_frags(skb, frag) { 500 /* Correct geometry. */ 501 if (frag->len > mtu || 502 ((frag->len & 7) && frag->next) || 503 skb_headroom(frag) < hlen) 504 goto slow_path_clean; 505 506 /* Partially cloned skb? */ 507 if (skb_shared(frag)) 508 goto slow_path_clean; 509 510 BUG_ON(frag->sk); 511 if (skb->sk) { 512 frag->sk = skb->sk; 513 frag->destructor = sock_wfree; 514 } 515 skb->truesize -= frag->truesize; 516 } 517 518 /* Everything is OK. Generate! */ 519 520 err = 0; 521 offset = 0; 522 frag = skb_shinfo(skb)->frag_list; 523 skb_frag_list_init(skb); 524 skb->data_len = first_len - skb_headlen(skb); 525 skb->len = first_len; 526 iph->tot_len = htons(first_len); 527 iph->frag_off = htons(IP_MF); 528 ip_send_check(iph); 529 530 for (;;) { 531 /* Prepare header of the next frame, 532 * before previous one went down. */ 533 if (frag) { 534 frag->ip_summed = CHECKSUM_NONE; 535 skb_reset_transport_header(frag); 536 __skb_push(frag, hlen); 537 skb_reset_network_header(frag); 538 memcpy(skb_network_header(frag), iph, hlen); 539 iph = ip_hdr(frag); 540 iph->tot_len = htons(frag->len); 541 ip_copy_metadata(frag, skb); 542 if (offset == 0) 543 ip_options_fragment(frag); 544 offset += skb->len - hlen; 545 iph->frag_off = htons(offset>>3); 546 if (frag->next != NULL) 547 iph->frag_off |= htons(IP_MF); 548 /* Ready, complete checksum */ 549 ip_send_check(iph); 550 } 551 552 err = output(skb); 553 554 if (!err) 555 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); 556 if (err || !frag) 557 break; 558 559 skb = frag; 560 frag = skb->next; 561 skb->next = NULL; 562 } 563 564 if (err == 0) { 565 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); 566 return 0; 567 } 568 569 while (frag) { 570 skb = frag->next; 571 kfree_skb(frag); 572 frag = skb; 573 } 574 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 575 return err; 576 577 slow_path_clean: 578 skb_walk_frags(skb, frag2) { 579 if (frag2 == frag) 580 break; 581 frag2->sk = NULL; 582 frag2->destructor = NULL; 583 skb->truesize += frag2->truesize; 584 } 585 } 586 587 slow_path: 588 /* for offloaded checksums cleanup checksum before fragmentation */ 589 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb)) 590 goto fail; 591 iph = ip_hdr(skb); 592 593 left = skb->len - hlen; /* Space per frame */ 594 ptr = hlen; /* Where to start from */ 595 596 /* for bridged IP traffic encapsulated inside f.e. a vlan header, 597 * we need to make room for the encapsulating header 598 */ 599 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb)); 600 601 /* 602 * Fragment the datagram. 603 */ 604 605 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 606 not_last_frag = iph->frag_off & htons(IP_MF); 607 608 /* 609 * Keep copying data until we run out. 610 */ 611 612 while (left > 0) { 613 len = left; 614 /* IF: it doesn't fit, use 'mtu' - the data space left */ 615 if (len > mtu) 616 len = mtu; 617 /* IF: we are not sending up to and including the packet end 618 then align the next start on an eight byte boundary */ 619 if (len < left) { 620 len &= ~7; 621 } 622 /* 623 * Allocate buffer. 624 */ 625 626 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { 627 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); 628 err = -ENOMEM; 629 goto fail; 630 } 631 632 /* 633 * Set up data on packet 634 */ 635 636 ip_copy_metadata(skb2, skb); 637 skb_reserve(skb2, ll_rs); 638 skb_put(skb2, len + hlen); 639 skb_reset_network_header(skb2); 640 skb2->transport_header = skb2->network_header + hlen; 641 642 /* 643 * Charge the memory for the fragment to any owner 644 * it might possess 645 */ 646 647 if (skb->sk) 648 skb_set_owner_w(skb2, skb->sk); 649 650 /* 651 * Copy the packet header into the new buffer. 652 */ 653 654 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); 655 656 /* 657 * Copy a block of the IP datagram. 658 */ 659 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) 660 BUG(); 661 left -= len; 662 663 /* 664 * Fill in the new header fields. 665 */ 666 iph = ip_hdr(skb2); 667 iph->frag_off = htons((offset >> 3)); 668 669 /* ANK: dirty, but effective trick. Upgrade options only if 670 * the segment to be fragmented was THE FIRST (otherwise, 671 * options are already fixed) and make it ONCE 672 * on the initial skb, so that all the following fragments 673 * will inherit fixed options. 674 */ 675 if (offset == 0) 676 ip_options_fragment(skb); 677 678 /* 679 * Added AC : If we are fragmenting a fragment that's not the 680 * last fragment then keep MF on each bit 681 */ 682 if (left > 0 || not_last_frag) 683 iph->frag_off |= htons(IP_MF); 684 ptr += len; 685 offset += len; 686 687 /* 688 * Put this fragment into the sending queue. 689 */ 690 iph->tot_len = htons(len + hlen); 691 692 ip_send_check(iph); 693 694 err = output(skb2); 695 if (err) 696 goto fail; 697 698 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); 699 } 700 consume_skb(skb); 701 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); 702 return err; 703 704 fail: 705 kfree_skb(skb); 706 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 707 return err; 708 } 709 EXPORT_SYMBOL(ip_fragment); 710 711 int 712 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 713 { 714 struct iovec *iov = from; 715 716 if (skb->ip_summed == CHECKSUM_PARTIAL) { 717 if (memcpy_fromiovecend(to, iov, offset, len) < 0) 718 return -EFAULT; 719 } else { 720 __wsum csum = 0; 721 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) 722 return -EFAULT; 723 skb->csum = csum_block_add(skb->csum, csum, odd); 724 } 725 return 0; 726 } 727 EXPORT_SYMBOL(ip_generic_getfrag); 728 729 static inline __wsum 730 csum_page(struct page *page, int offset, int copy) 731 { 732 char *kaddr; 733 __wsum csum; 734 kaddr = kmap(page); 735 csum = csum_partial(kaddr + offset, copy, 0); 736 kunmap(page); 737 return csum; 738 } 739 740 static inline int ip_ufo_append_data(struct sock *sk, 741 struct sk_buff_head *queue, 742 int getfrag(void *from, char *to, int offset, int len, 743 int odd, struct sk_buff *skb), 744 void *from, int length, int hh_len, int fragheaderlen, 745 int transhdrlen, int maxfraglen, unsigned int flags) 746 { 747 struct sk_buff *skb; 748 int err; 749 750 /* There is support for UDP fragmentation offload by network 751 * device, so create one single skb packet containing complete 752 * udp datagram 753 */ 754 if ((skb = skb_peek_tail(queue)) == NULL) { 755 skb = sock_alloc_send_skb(sk, 756 hh_len + fragheaderlen + transhdrlen + 20, 757 (flags & MSG_DONTWAIT), &err); 758 759 if (skb == NULL) 760 return err; 761 762 /* reserve space for Hardware header */ 763 skb_reserve(skb, hh_len); 764 765 /* create space for UDP/IP header */ 766 skb_put(skb, fragheaderlen + transhdrlen); 767 768 /* initialize network header pointer */ 769 skb_reset_network_header(skb); 770 771 /* initialize protocol header pointer */ 772 skb->transport_header = skb->network_header + fragheaderlen; 773 774 skb->csum = 0; 775 776 777 __skb_queue_tail(queue, skb); 778 } else if (skb_is_gso(skb)) { 779 goto append; 780 } 781 782 skb->ip_summed = CHECKSUM_PARTIAL; 783 /* specify the length of each IP datagram fragment */ 784 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen; 785 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 786 787 append: 788 return skb_append_datato_frags(sk, skb, getfrag, from, 789 (length - transhdrlen)); 790 } 791 792 static int __ip_append_data(struct sock *sk, 793 struct flowi4 *fl4, 794 struct sk_buff_head *queue, 795 struct inet_cork *cork, 796 struct page_frag *pfrag, 797 int getfrag(void *from, char *to, int offset, 798 int len, int odd, struct sk_buff *skb), 799 void *from, int length, int transhdrlen, 800 unsigned int flags) 801 { 802 struct inet_sock *inet = inet_sk(sk); 803 struct sk_buff *skb; 804 805 struct ip_options *opt = cork->opt; 806 int hh_len; 807 int exthdrlen; 808 int mtu; 809 int copy; 810 int err; 811 int offset = 0; 812 unsigned int maxfraglen, fragheaderlen, maxnonfragsize; 813 int csummode = CHECKSUM_NONE; 814 struct rtable *rt = (struct rtable *)cork->dst; 815 816 skb = skb_peek_tail(queue); 817 818 exthdrlen = !skb ? rt->dst.header_len : 0; 819 mtu = cork->fragsize; 820 821 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 822 823 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 824 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 825 maxnonfragsize = (inet->pmtudisc >= IP_PMTUDISC_DO) ? 826 mtu : 0xFFFF; 827 828 if (cork->length + length > maxnonfragsize - fragheaderlen) { 829 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 830 mtu - (opt ? opt->optlen : 0)); 831 return -EMSGSIZE; 832 } 833 834 /* 835 * transhdrlen > 0 means that this is the first fragment and we wish 836 * it won't be fragmented in the future. 837 */ 838 if (transhdrlen && 839 length + fragheaderlen <= mtu && 840 rt->dst.dev->features & NETIF_F_V4_CSUM && 841 !exthdrlen) 842 csummode = CHECKSUM_PARTIAL; 843 844 cork->length += length; 845 if (((length > mtu) || (skb && skb_is_gso(skb))) && 846 (sk->sk_protocol == IPPROTO_UDP) && 847 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) { 848 err = ip_ufo_append_data(sk, queue, getfrag, from, length, 849 hh_len, fragheaderlen, transhdrlen, 850 maxfraglen, flags); 851 if (err) 852 goto error; 853 return 0; 854 } 855 856 /* So, what's going on in the loop below? 857 * 858 * We use calculated fragment length to generate chained skb, 859 * each of segments is IP fragment ready for sending to network after 860 * adding appropriate IP header. 861 */ 862 863 if (!skb) 864 goto alloc_new_skb; 865 866 while (length > 0) { 867 /* Check if the remaining data fits into current packet. */ 868 copy = mtu - skb->len; 869 if (copy < length) 870 copy = maxfraglen - skb->len; 871 if (copy <= 0) { 872 char *data; 873 unsigned int datalen; 874 unsigned int fraglen; 875 unsigned int fraggap; 876 unsigned int alloclen; 877 struct sk_buff *skb_prev; 878 alloc_new_skb: 879 skb_prev = skb; 880 if (skb_prev) 881 fraggap = skb_prev->len - maxfraglen; 882 else 883 fraggap = 0; 884 885 /* 886 * If remaining data exceeds the mtu, 887 * we know we need more fragment(s). 888 */ 889 datalen = length + fraggap; 890 if (datalen > mtu - fragheaderlen) 891 datalen = maxfraglen - fragheaderlen; 892 fraglen = datalen + fragheaderlen; 893 894 if ((flags & MSG_MORE) && 895 !(rt->dst.dev->features&NETIF_F_SG)) 896 alloclen = mtu; 897 else 898 alloclen = fraglen; 899 900 alloclen += exthdrlen; 901 902 /* The last fragment gets additional space at tail. 903 * Note, with MSG_MORE we overallocate on fragments, 904 * because we have no idea what fragment will be 905 * the last. 906 */ 907 if (datalen == length + fraggap) 908 alloclen += rt->dst.trailer_len; 909 910 if (transhdrlen) { 911 skb = sock_alloc_send_skb(sk, 912 alloclen + hh_len + 15, 913 (flags & MSG_DONTWAIT), &err); 914 } else { 915 skb = NULL; 916 if (atomic_read(&sk->sk_wmem_alloc) <= 917 2 * sk->sk_sndbuf) 918 skb = sock_wmalloc(sk, 919 alloclen + hh_len + 15, 1, 920 sk->sk_allocation); 921 if (unlikely(skb == NULL)) 922 err = -ENOBUFS; 923 else 924 /* only the initial fragment is 925 time stamped */ 926 cork->tx_flags = 0; 927 } 928 if (skb == NULL) 929 goto error; 930 931 /* 932 * Fill in the control structures 933 */ 934 skb->ip_summed = csummode; 935 skb->csum = 0; 936 skb_reserve(skb, hh_len); 937 skb_shinfo(skb)->tx_flags = cork->tx_flags; 938 939 /* 940 * Find where to start putting bytes. 941 */ 942 data = skb_put(skb, fraglen + exthdrlen); 943 skb_set_network_header(skb, exthdrlen); 944 skb->transport_header = (skb->network_header + 945 fragheaderlen); 946 data += fragheaderlen + exthdrlen; 947 948 if (fraggap) { 949 skb->csum = skb_copy_and_csum_bits( 950 skb_prev, maxfraglen, 951 data + transhdrlen, fraggap, 0); 952 skb_prev->csum = csum_sub(skb_prev->csum, 953 skb->csum); 954 data += fraggap; 955 pskb_trim_unique(skb_prev, maxfraglen); 956 } 957 958 copy = datalen - transhdrlen - fraggap; 959 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 960 err = -EFAULT; 961 kfree_skb(skb); 962 goto error; 963 } 964 965 offset += copy; 966 length -= datalen - fraggap; 967 transhdrlen = 0; 968 exthdrlen = 0; 969 csummode = CHECKSUM_NONE; 970 971 /* 972 * Put the packet on the pending queue. 973 */ 974 __skb_queue_tail(queue, skb); 975 continue; 976 } 977 978 if (copy > length) 979 copy = length; 980 981 if (!(rt->dst.dev->features&NETIF_F_SG)) { 982 unsigned int off; 983 984 off = skb->len; 985 if (getfrag(from, skb_put(skb, copy), 986 offset, copy, off, skb) < 0) { 987 __skb_trim(skb, off); 988 err = -EFAULT; 989 goto error; 990 } 991 } else { 992 int i = skb_shinfo(skb)->nr_frags; 993 994 err = -ENOMEM; 995 if (!sk_page_frag_refill(sk, pfrag)) 996 goto error; 997 998 if (!skb_can_coalesce(skb, i, pfrag->page, 999 pfrag->offset)) { 1000 err = -EMSGSIZE; 1001 if (i == MAX_SKB_FRAGS) 1002 goto error; 1003 1004 __skb_fill_page_desc(skb, i, pfrag->page, 1005 pfrag->offset, 0); 1006 skb_shinfo(skb)->nr_frags = ++i; 1007 get_page(pfrag->page); 1008 } 1009 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1010 if (getfrag(from, 1011 page_address(pfrag->page) + pfrag->offset, 1012 offset, copy, skb->len, skb) < 0) 1013 goto error_efault; 1014 1015 pfrag->offset += copy; 1016 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1017 skb->len += copy; 1018 skb->data_len += copy; 1019 skb->truesize += copy; 1020 atomic_add(copy, &sk->sk_wmem_alloc); 1021 } 1022 offset += copy; 1023 length -= copy; 1024 } 1025 1026 return 0; 1027 1028 error_efault: 1029 err = -EFAULT; 1030 error: 1031 cork->length -= length; 1032 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1033 return err; 1034 } 1035 1036 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 1037 struct ipcm_cookie *ipc, struct rtable **rtp) 1038 { 1039 struct ip_options_rcu *opt; 1040 struct rtable *rt; 1041 1042 /* 1043 * setup for corking. 1044 */ 1045 opt = ipc->opt; 1046 if (opt) { 1047 if (cork->opt == NULL) { 1048 cork->opt = kmalloc(sizeof(struct ip_options) + 40, 1049 sk->sk_allocation); 1050 if (unlikely(cork->opt == NULL)) 1051 return -ENOBUFS; 1052 } 1053 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 1054 cork->flags |= IPCORK_OPT; 1055 cork->addr = ipc->addr; 1056 } 1057 rt = *rtp; 1058 if (unlikely(!rt)) 1059 return -EFAULT; 1060 /* 1061 * We steal reference to this route, caller should not release it 1062 */ 1063 *rtp = NULL; 1064 cork->fragsize = ip_sk_use_pmtu(sk) ? 1065 dst_mtu(&rt->dst) : rt->dst.dev->mtu; 1066 cork->dst = &rt->dst; 1067 cork->length = 0; 1068 cork->ttl = ipc->ttl; 1069 cork->tos = ipc->tos; 1070 cork->priority = ipc->priority; 1071 cork->tx_flags = ipc->tx_flags; 1072 1073 return 0; 1074 } 1075 1076 /* 1077 * ip_append_data() and ip_append_page() can make one large IP datagram 1078 * from many pieces of data. Each pieces will be holded on the socket 1079 * until ip_push_pending_frames() is called. Each piece can be a page 1080 * or non-page data. 1081 * 1082 * Not only UDP, other transport protocols - e.g. raw sockets - can use 1083 * this interface potentially. 1084 * 1085 * LATER: length must be adjusted by pad at tail, when it is required. 1086 */ 1087 int ip_append_data(struct sock *sk, struct flowi4 *fl4, 1088 int getfrag(void *from, char *to, int offset, int len, 1089 int odd, struct sk_buff *skb), 1090 void *from, int length, int transhdrlen, 1091 struct ipcm_cookie *ipc, struct rtable **rtp, 1092 unsigned int flags) 1093 { 1094 struct inet_sock *inet = inet_sk(sk); 1095 int err; 1096 1097 if (flags&MSG_PROBE) 1098 return 0; 1099 1100 if (skb_queue_empty(&sk->sk_write_queue)) { 1101 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 1102 if (err) 1103 return err; 1104 } else { 1105 transhdrlen = 0; 1106 } 1107 1108 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, 1109 sk_page_frag(sk), getfrag, 1110 from, length, transhdrlen, flags); 1111 } 1112 1113 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 1114 int offset, size_t size, int flags) 1115 { 1116 struct inet_sock *inet = inet_sk(sk); 1117 struct sk_buff *skb; 1118 struct rtable *rt; 1119 struct ip_options *opt = NULL; 1120 struct inet_cork *cork; 1121 int hh_len; 1122 int mtu; 1123 int len; 1124 int err; 1125 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; 1126 1127 if (inet->hdrincl) 1128 return -EPERM; 1129 1130 if (flags&MSG_PROBE) 1131 return 0; 1132 1133 if (skb_queue_empty(&sk->sk_write_queue)) 1134 return -EINVAL; 1135 1136 cork = &inet->cork.base; 1137 rt = (struct rtable *)cork->dst; 1138 if (cork->flags & IPCORK_OPT) 1139 opt = cork->opt; 1140 1141 if (!(rt->dst.dev->features&NETIF_F_SG)) 1142 return -EOPNOTSUPP; 1143 1144 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1145 mtu = cork->fragsize; 1146 1147 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1148 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1149 maxnonfragsize = (inet->pmtudisc >= IP_PMTUDISC_DO) ? 1150 mtu : 0xFFFF; 1151 1152 if (cork->length + size > maxnonfragsize - fragheaderlen) { 1153 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 1154 mtu - (opt ? opt->optlen : 0)); 1155 return -EMSGSIZE; 1156 } 1157 1158 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) 1159 return -EINVAL; 1160 1161 cork->length += size; 1162 if ((size + skb->len > mtu) && 1163 (sk->sk_protocol == IPPROTO_UDP) && 1164 (rt->dst.dev->features & NETIF_F_UFO)) { 1165 skb_shinfo(skb)->gso_size = mtu - fragheaderlen; 1166 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 1167 } 1168 1169 1170 while (size > 0) { 1171 int i; 1172 1173 if (skb_is_gso(skb)) 1174 len = size; 1175 else { 1176 1177 /* Check if the remaining data fits into current packet. */ 1178 len = mtu - skb->len; 1179 if (len < size) 1180 len = maxfraglen - skb->len; 1181 } 1182 if (len <= 0) { 1183 struct sk_buff *skb_prev; 1184 int alloclen; 1185 1186 skb_prev = skb; 1187 fraggap = skb_prev->len - maxfraglen; 1188 1189 alloclen = fragheaderlen + hh_len + fraggap + 15; 1190 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 1191 if (unlikely(!skb)) { 1192 err = -ENOBUFS; 1193 goto error; 1194 } 1195 1196 /* 1197 * Fill in the control structures 1198 */ 1199 skb->ip_summed = CHECKSUM_NONE; 1200 skb->csum = 0; 1201 skb_reserve(skb, hh_len); 1202 1203 /* 1204 * Find where to start putting bytes. 1205 */ 1206 skb_put(skb, fragheaderlen + fraggap); 1207 skb_reset_network_header(skb); 1208 skb->transport_header = (skb->network_header + 1209 fragheaderlen); 1210 if (fraggap) { 1211 skb->csum = skb_copy_and_csum_bits(skb_prev, 1212 maxfraglen, 1213 skb_transport_header(skb), 1214 fraggap, 0); 1215 skb_prev->csum = csum_sub(skb_prev->csum, 1216 skb->csum); 1217 pskb_trim_unique(skb_prev, maxfraglen); 1218 } 1219 1220 /* 1221 * Put the packet on the pending queue. 1222 */ 1223 __skb_queue_tail(&sk->sk_write_queue, skb); 1224 continue; 1225 } 1226 1227 i = skb_shinfo(skb)->nr_frags; 1228 if (len > size) 1229 len = size; 1230 if (skb_can_coalesce(skb, i, page, offset)) { 1231 skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len); 1232 } else if (i < MAX_SKB_FRAGS) { 1233 get_page(page); 1234 skb_fill_page_desc(skb, i, page, offset, len); 1235 } else { 1236 err = -EMSGSIZE; 1237 goto error; 1238 } 1239 1240 if (skb->ip_summed == CHECKSUM_NONE) { 1241 __wsum csum; 1242 csum = csum_page(page, offset, len); 1243 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1244 } 1245 1246 skb->len += len; 1247 skb->data_len += len; 1248 skb->truesize += len; 1249 atomic_add(len, &sk->sk_wmem_alloc); 1250 offset += len; 1251 size -= len; 1252 } 1253 return 0; 1254 1255 error: 1256 cork->length -= size; 1257 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1258 return err; 1259 } 1260 1261 static void ip_cork_release(struct inet_cork *cork) 1262 { 1263 cork->flags &= ~IPCORK_OPT; 1264 kfree(cork->opt); 1265 cork->opt = NULL; 1266 dst_release(cork->dst); 1267 cork->dst = NULL; 1268 } 1269 1270 /* 1271 * Combined all pending IP fragments on the socket as one IP datagram 1272 * and push them out. 1273 */ 1274 struct sk_buff *__ip_make_skb(struct sock *sk, 1275 struct flowi4 *fl4, 1276 struct sk_buff_head *queue, 1277 struct inet_cork *cork) 1278 { 1279 struct sk_buff *skb, *tmp_skb; 1280 struct sk_buff **tail_skb; 1281 struct inet_sock *inet = inet_sk(sk); 1282 struct net *net = sock_net(sk); 1283 struct ip_options *opt = NULL; 1284 struct rtable *rt = (struct rtable *)cork->dst; 1285 struct iphdr *iph; 1286 __be16 df = 0; 1287 __u8 ttl; 1288 1289 if ((skb = __skb_dequeue(queue)) == NULL) 1290 goto out; 1291 tail_skb = &(skb_shinfo(skb)->frag_list); 1292 1293 /* move skb->data to ip header from ext header */ 1294 if (skb->data < skb_network_header(skb)) 1295 __skb_pull(skb, skb_network_offset(skb)); 1296 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1297 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1298 *tail_skb = tmp_skb; 1299 tail_skb = &(tmp_skb->next); 1300 skb->len += tmp_skb->len; 1301 skb->data_len += tmp_skb->len; 1302 skb->truesize += tmp_skb->truesize; 1303 tmp_skb->destructor = NULL; 1304 tmp_skb->sk = NULL; 1305 } 1306 1307 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 1308 * to fragment the frame generated here. No matter, what transforms 1309 * how transforms change size of the packet, it will come out. 1310 */ 1311 if (inet->pmtudisc < IP_PMTUDISC_DO) 1312 skb->local_df = 1; 1313 1314 /* DF bit is set when we want to see DF on outgoing frames. 1315 * If local_df is set too, we still allow to fragment this frame 1316 * locally. */ 1317 if (inet->pmtudisc == IP_PMTUDISC_DO || 1318 inet->pmtudisc == IP_PMTUDISC_PROBE || 1319 (skb->len <= dst_mtu(&rt->dst) && 1320 ip_dont_fragment(sk, &rt->dst))) 1321 df = htons(IP_DF); 1322 1323 if (cork->flags & IPCORK_OPT) 1324 opt = cork->opt; 1325 1326 if (cork->ttl != 0) 1327 ttl = cork->ttl; 1328 else if (rt->rt_type == RTN_MULTICAST) 1329 ttl = inet->mc_ttl; 1330 else 1331 ttl = ip_select_ttl(inet, &rt->dst); 1332 1333 iph = ip_hdr(skb); 1334 iph->version = 4; 1335 iph->ihl = 5; 1336 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; 1337 iph->frag_off = df; 1338 iph->ttl = ttl; 1339 iph->protocol = sk->sk_protocol; 1340 ip_copy_addrs(iph, fl4); 1341 ip_select_ident(skb, &rt->dst, sk); 1342 1343 if (opt) { 1344 iph->ihl += opt->optlen>>2; 1345 ip_options_build(skb, opt, cork->addr, rt, 0); 1346 } 1347 1348 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; 1349 skb->mark = sk->sk_mark; 1350 /* 1351 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1352 * on dst refcount 1353 */ 1354 cork->dst = NULL; 1355 skb_dst_set(skb, &rt->dst); 1356 1357 if (iph->protocol == IPPROTO_ICMP) 1358 icmp_out_count(net, ((struct icmphdr *) 1359 skb_transport_header(skb))->type); 1360 1361 ip_cork_release(cork); 1362 out: 1363 return skb; 1364 } 1365 1366 int ip_send_skb(struct net *net, struct sk_buff *skb) 1367 { 1368 int err; 1369 1370 err = ip_local_out(skb); 1371 if (err) { 1372 if (err > 0) 1373 err = net_xmit_errno(err); 1374 if (err) 1375 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 1376 } 1377 1378 return err; 1379 } 1380 1381 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 1382 { 1383 struct sk_buff *skb; 1384 1385 skb = ip_finish_skb(sk, fl4); 1386 if (!skb) 1387 return 0; 1388 1389 /* Netfilter gets whole the not fragmented skb. */ 1390 return ip_send_skb(sock_net(sk), skb); 1391 } 1392 1393 /* 1394 * Throw away all pending data on the socket. 1395 */ 1396 static void __ip_flush_pending_frames(struct sock *sk, 1397 struct sk_buff_head *queue, 1398 struct inet_cork *cork) 1399 { 1400 struct sk_buff *skb; 1401 1402 while ((skb = __skb_dequeue_tail(queue)) != NULL) 1403 kfree_skb(skb); 1404 1405 ip_cork_release(cork); 1406 } 1407 1408 void ip_flush_pending_frames(struct sock *sk) 1409 { 1410 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 1411 } 1412 1413 struct sk_buff *ip_make_skb(struct sock *sk, 1414 struct flowi4 *fl4, 1415 int getfrag(void *from, char *to, int offset, 1416 int len, int odd, struct sk_buff *skb), 1417 void *from, int length, int transhdrlen, 1418 struct ipcm_cookie *ipc, struct rtable **rtp, 1419 unsigned int flags) 1420 { 1421 struct inet_cork cork; 1422 struct sk_buff_head queue; 1423 int err; 1424 1425 if (flags & MSG_PROBE) 1426 return NULL; 1427 1428 __skb_queue_head_init(&queue); 1429 1430 cork.flags = 0; 1431 cork.addr = 0; 1432 cork.opt = NULL; 1433 err = ip_setup_cork(sk, &cork, ipc, rtp); 1434 if (err) 1435 return ERR_PTR(err); 1436 1437 err = __ip_append_data(sk, fl4, &queue, &cork, 1438 ¤t->task_frag, getfrag, 1439 from, length, transhdrlen, flags); 1440 if (err) { 1441 __ip_flush_pending_frames(sk, &queue, &cork); 1442 return ERR_PTR(err); 1443 } 1444 1445 return __ip_make_skb(sk, fl4, &queue, &cork); 1446 } 1447 1448 /* 1449 * Fetch data from kernel space and fill in checksum if needed. 1450 */ 1451 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 1452 int len, int odd, struct sk_buff *skb) 1453 { 1454 __wsum csum; 1455 1456 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); 1457 skb->csum = csum_block_add(skb->csum, csum, odd); 1458 return 0; 1459 } 1460 1461 /* 1462 * Generic function to send a packet as reply to another packet. 1463 * Used to send some TCP resets/acks so far. 1464 * 1465 * Use a fake percpu inet socket to avoid false sharing and contention. 1466 */ 1467 static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = { 1468 .sk = { 1469 .__sk_common = { 1470 .skc_refcnt = ATOMIC_INIT(1), 1471 }, 1472 .sk_wmem_alloc = ATOMIC_INIT(1), 1473 .sk_allocation = GFP_ATOMIC, 1474 .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE), 1475 }, 1476 .pmtudisc = IP_PMTUDISC_WANT, 1477 .uc_ttl = -1, 1478 }; 1479 1480 void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr, 1481 __be32 saddr, const struct ip_reply_arg *arg, 1482 unsigned int len) 1483 { 1484 struct ip_options_data replyopts; 1485 struct ipcm_cookie ipc; 1486 struct flowi4 fl4; 1487 struct rtable *rt = skb_rtable(skb); 1488 struct sk_buff *nskb; 1489 struct sock *sk; 1490 struct inet_sock *inet; 1491 1492 if (ip_options_echo(&replyopts.opt.opt, skb)) 1493 return; 1494 1495 ipc.addr = daddr; 1496 ipc.opt = NULL; 1497 ipc.tx_flags = 0; 1498 ipc.ttl = 0; 1499 ipc.tos = -1; 1500 1501 if (replyopts.opt.opt.optlen) { 1502 ipc.opt = &replyopts.opt; 1503 1504 if (replyopts.opt.opt.srr) 1505 daddr = replyopts.opt.opt.faddr; 1506 } 1507 1508 flowi4_init_output(&fl4, arg->bound_dev_if, 0, 1509 RT_TOS(arg->tos), 1510 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, 1511 ip_reply_arg_flowi_flags(arg), 1512 daddr, saddr, 1513 tcp_hdr(skb)->source, tcp_hdr(skb)->dest); 1514 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 1515 rt = ip_route_output_key(net, &fl4); 1516 if (IS_ERR(rt)) 1517 return; 1518 1519 inet = &get_cpu_var(unicast_sock); 1520 1521 inet->tos = arg->tos; 1522 sk = &inet->sk; 1523 sk->sk_priority = skb->priority; 1524 sk->sk_protocol = ip_hdr(skb)->protocol; 1525 sk->sk_bound_dev_if = arg->bound_dev_if; 1526 sock_net_set(sk, net); 1527 __skb_queue_head_init(&sk->sk_write_queue); 1528 sk->sk_sndbuf = sysctl_wmem_default; 1529 ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0, 1530 &ipc, &rt, MSG_DONTWAIT); 1531 nskb = skb_peek(&sk->sk_write_queue); 1532 if (nskb) { 1533 if (arg->csumoffset >= 0) 1534 *((__sum16 *)skb_transport_header(nskb) + 1535 arg->csumoffset) = csum_fold(csum_add(nskb->csum, 1536 arg->csum)); 1537 nskb->ip_summed = CHECKSUM_NONE; 1538 skb_orphan(nskb); 1539 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb)); 1540 ip_push_pending_frames(sk, &fl4); 1541 } 1542 1543 put_cpu_var(unicast_sock); 1544 1545 ip_rt_put(rt); 1546 } 1547 1548 void __init ip_init(void) 1549 { 1550 ip_rt_init(); 1551 inet_initpeers(); 1552 1553 #if defined(CONFIG_IP_MULTICAST) 1554 igmp_mc_init(); 1555 #endif 1556 } 1557