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 <linux/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 <net/lwtunnel.h> 77 #include <linux/bpf-cgroup.h> 78 #include <linux/igmp.h> 79 #include <linux/netfilter_ipv4.h> 80 #include <linux/netfilter_bridge.h> 81 #include <linux/netlink.h> 82 #include <linux/tcp.h> 83 84 static int 85 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 86 unsigned int mtu, 87 int (*output)(struct net *, struct sock *, struct sk_buff *)); 88 89 /* Generate a checksum for an outgoing IP datagram. */ 90 void ip_send_check(struct iphdr *iph) 91 { 92 iph->check = 0; 93 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 94 } 95 EXPORT_SYMBOL(ip_send_check); 96 97 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 98 { 99 struct iphdr *iph = ip_hdr(skb); 100 101 iph->tot_len = htons(skb->len); 102 ip_send_check(iph); 103 104 /* if egress device is enslaved to an L3 master device pass the 105 * skb to its handler for processing 106 */ 107 skb = l3mdev_ip_out(sk, skb); 108 if (unlikely(!skb)) 109 return 0; 110 111 skb->protocol = htons(ETH_P_IP); 112 113 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, 114 net, sk, skb, NULL, skb_dst(skb)->dev, 115 dst_output); 116 } 117 118 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 119 { 120 int err; 121 122 err = __ip_local_out(net, sk, skb); 123 if (likely(err == 1)) 124 err = dst_output(net, sk, skb); 125 126 return err; 127 } 128 EXPORT_SYMBOL_GPL(ip_local_out); 129 130 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) 131 { 132 int ttl = inet->uc_ttl; 133 134 if (ttl < 0) 135 ttl = ip4_dst_hoplimit(dst); 136 return ttl; 137 } 138 139 /* 140 * Add an ip header to a skbuff and send it out. 141 * 142 */ 143 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk, 144 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) 145 { 146 struct inet_sock *inet = inet_sk(sk); 147 struct rtable *rt = skb_rtable(skb); 148 struct net *net = sock_net(sk); 149 struct iphdr *iph; 150 151 /* Build the IP header. */ 152 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); 153 skb_reset_network_header(skb); 154 iph = ip_hdr(skb); 155 iph->version = 4; 156 iph->ihl = 5; 157 iph->tos = inet->tos; 158 iph->ttl = ip_select_ttl(inet, &rt->dst); 159 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); 160 iph->saddr = saddr; 161 iph->protocol = sk->sk_protocol; 162 if (ip_dont_fragment(sk, &rt->dst)) { 163 iph->frag_off = htons(IP_DF); 164 iph->id = 0; 165 } else { 166 iph->frag_off = 0; 167 __ip_select_ident(net, iph, 1); 168 } 169 170 if (opt && opt->opt.optlen) { 171 iph->ihl += opt->opt.optlen>>2; 172 ip_options_build(skb, &opt->opt, daddr, rt, 0); 173 } 174 175 skb->priority = sk->sk_priority; 176 if (!skb->mark) 177 skb->mark = sk->sk_mark; 178 179 /* Send it out. */ 180 return ip_local_out(net, skb->sk, skb); 181 } 182 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); 183 184 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) 185 { 186 struct dst_entry *dst = skb_dst(skb); 187 struct rtable *rt = (struct rtable *)dst; 188 struct net_device *dev = dst->dev; 189 unsigned int hh_len = LL_RESERVED_SPACE(dev); 190 struct neighbour *neigh; 191 u32 nexthop; 192 193 if (rt->rt_type == RTN_MULTICAST) { 194 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len); 195 } else if (rt->rt_type == RTN_BROADCAST) 196 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len); 197 198 /* Be paranoid, rather than too clever. */ 199 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 200 struct sk_buff *skb2; 201 202 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 203 if (!skb2) { 204 kfree_skb(skb); 205 return -ENOMEM; 206 } 207 if (skb->sk) 208 skb_set_owner_w(skb2, skb->sk); 209 consume_skb(skb); 210 skb = skb2; 211 } 212 213 if (lwtunnel_xmit_redirect(dst->lwtstate)) { 214 int res = lwtunnel_xmit(skb); 215 216 if (res < 0 || res == LWTUNNEL_XMIT_DONE) 217 return res; 218 } 219 220 rcu_read_lock_bh(); 221 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr); 222 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 223 if (unlikely(!neigh)) 224 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 225 if (!IS_ERR(neigh)) { 226 int res; 227 228 sock_confirm_neigh(skb, neigh); 229 res = neigh_output(neigh, skb); 230 231 rcu_read_unlock_bh(); 232 return res; 233 } 234 rcu_read_unlock_bh(); 235 236 net_dbg_ratelimited("%s: No header cache and no neighbour!\n", 237 __func__); 238 kfree_skb(skb); 239 return -EINVAL; 240 } 241 242 static int ip_finish_output_gso(struct net *net, struct sock *sk, 243 struct sk_buff *skb, unsigned int mtu) 244 { 245 netdev_features_t features; 246 struct sk_buff *segs; 247 int ret = 0; 248 249 /* common case: seglen is <= mtu 250 */ 251 if (skb_gso_validate_network_len(skb, mtu)) 252 return ip_finish_output2(net, sk, skb); 253 254 /* Slowpath - GSO segment length exceeds the egress MTU. 255 * 256 * This can happen in several cases: 257 * - Forwarding of a TCP GRO skb, when DF flag is not set. 258 * - Forwarding of an skb that arrived on a virtualization interface 259 * (virtio-net/vhost/tap) with TSO/GSO size set by other network 260 * stack. 261 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an 262 * interface with a smaller MTU. 263 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is 264 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an 265 * insufficent MTU. 266 */ 267 features = netif_skb_features(skb); 268 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET); 269 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); 270 if (IS_ERR_OR_NULL(segs)) { 271 kfree_skb(skb); 272 return -ENOMEM; 273 } 274 275 consume_skb(skb); 276 277 do { 278 struct sk_buff *nskb = segs->next; 279 int err; 280 281 segs->next = NULL; 282 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2); 283 284 if (err && ret == 0) 285 ret = err; 286 segs = nskb; 287 } while (segs); 288 289 return ret; 290 } 291 292 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 293 { 294 unsigned int mtu; 295 int ret; 296 297 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); 298 if (ret) { 299 kfree_skb(skb); 300 return ret; 301 } 302 303 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) 304 /* Policy lookup after SNAT yielded a new policy */ 305 if (skb_dst(skb)->xfrm) { 306 IPCB(skb)->flags |= IPSKB_REROUTED; 307 return dst_output(net, sk, skb); 308 } 309 #endif 310 mtu = ip_skb_dst_mtu(sk, skb); 311 if (skb_is_gso(skb)) 312 return ip_finish_output_gso(net, sk, skb, mtu); 313 314 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU)) 315 return ip_fragment(net, sk, skb, mtu, ip_finish_output2); 316 317 return ip_finish_output2(net, sk, skb); 318 } 319 320 static int ip_mc_finish_output(struct net *net, struct sock *sk, 321 struct sk_buff *skb) 322 { 323 int ret; 324 325 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); 326 if (ret) { 327 kfree_skb(skb); 328 return ret; 329 } 330 331 return dev_loopback_xmit(net, sk, skb); 332 } 333 334 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb) 335 { 336 struct rtable *rt = skb_rtable(skb); 337 struct net_device *dev = rt->dst.dev; 338 339 /* 340 * If the indicated interface is up and running, send the packet. 341 */ 342 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 343 344 skb->dev = dev; 345 skb->protocol = htons(ETH_P_IP); 346 347 /* 348 * Multicasts are looped back for other local users 349 */ 350 351 if (rt->rt_flags&RTCF_MULTICAST) { 352 if (sk_mc_loop(sk) 353 #ifdef CONFIG_IP_MROUTE 354 /* Small optimization: do not loopback not local frames, 355 which returned after forwarding; they will be dropped 356 by ip_mr_input in any case. 357 Note, that local frames are looped back to be delivered 358 to local recipients. 359 360 This check is duplicated in ip_mr_input at the moment. 361 */ 362 && 363 ((rt->rt_flags & RTCF_LOCAL) || 364 !(IPCB(skb)->flags & IPSKB_FORWARDED)) 365 #endif 366 ) { 367 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 368 if (newskb) 369 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 370 net, sk, newskb, NULL, newskb->dev, 371 ip_mc_finish_output); 372 } 373 374 /* Multicasts with ttl 0 must not go beyond the host */ 375 376 if (ip_hdr(skb)->ttl == 0) { 377 kfree_skb(skb); 378 return 0; 379 } 380 } 381 382 if (rt->rt_flags&RTCF_BROADCAST) { 383 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 384 if (newskb) 385 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 386 net, sk, newskb, NULL, newskb->dev, 387 ip_mc_finish_output); 388 } 389 390 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 391 net, sk, skb, NULL, skb->dev, 392 ip_finish_output, 393 !(IPCB(skb)->flags & IPSKB_REROUTED)); 394 } 395 396 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb) 397 { 398 struct net_device *dev = skb_dst(skb)->dev; 399 400 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 401 402 skb->dev = dev; 403 skb->protocol = htons(ETH_P_IP); 404 405 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 406 net, sk, skb, NULL, dev, 407 ip_finish_output, 408 !(IPCB(skb)->flags & IPSKB_REROUTED)); 409 } 410 411 /* 412 * copy saddr and daddr, possibly using 64bit load/stores 413 * Equivalent to : 414 * iph->saddr = fl4->saddr; 415 * iph->daddr = fl4->daddr; 416 */ 417 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) 418 { 419 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != 420 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); 421 memcpy(&iph->saddr, &fl4->saddr, 422 sizeof(fl4->saddr) + sizeof(fl4->daddr)); 423 } 424 425 /* Note: skb->sk can be different from sk, in case of tunnels */ 426 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) 427 { 428 struct inet_sock *inet = inet_sk(sk); 429 struct net *net = sock_net(sk); 430 struct ip_options_rcu *inet_opt; 431 struct flowi4 *fl4; 432 struct rtable *rt; 433 struct iphdr *iph; 434 int res; 435 436 /* Skip all of this if the packet is already routed, 437 * f.e. by something like SCTP. 438 */ 439 rcu_read_lock(); 440 inet_opt = rcu_dereference(inet->inet_opt); 441 fl4 = &fl->u.ip4; 442 rt = skb_rtable(skb); 443 if (rt) 444 goto packet_routed; 445 446 /* Make sure we can route this packet. */ 447 rt = (struct rtable *)__sk_dst_check(sk, 0); 448 if (!rt) { 449 __be32 daddr; 450 451 /* Use correct destination address if we have options. */ 452 daddr = inet->inet_daddr; 453 if (inet_opt && inet_opt->opt.srr) 454 daddr = inet_opt->opt.faddr; 455 456 /* If this fails, retransmit mechanism of transport layer will 457 * keep trying until route appears or the connection times 458 * itself out. 459 */ 460 rt = ip_route_output_ports(net, fl4, sk, 461 daddr, inet->inet_saddr, 462 inet->inet_dport, 463 inet->inet_sport, 464 sk->sk_protocol, 465 RT_CONN_FLAGS(sk), 466 sk->sk_bound_dev_if); 467 if (IS_ERR(rt)) 468 goto no_route; 469 sk_setup_caps(sk, &rt->dst); 470 } 471 skb_dst_set_noref(skb, &rt->dst); 472 473 packet_routed: 474 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) 475 goto no_route; 476 477 /* OK, we know where to send it, allocate and build IP header. */ 478 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 479 skb_reset_network_header(skb); 480 iph = ip_hdr(skb); 481 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); 482 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) 483 iph->frag_off = htons(IP_DF); 484 else 485 iph->frag_off = 0; 486 iph->ttl = ip_select_ttl(inet, &rt->dst); 487 iph->protocol = sk->sk_protocol; 488 ip_copy_addrs(iph, fl4); 489 490 /* Transport layer set skb->h.foo itself. */ 491 492 if (inet_opt && inet_opt->opt.optlen) { 493 iph->ihl += inet_opt->opt.optlen >> 2; 494 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 495 } 496 497 ip_select_ident_segs(net, skb, sk, 498 skb_shinfo(skb)->gso_segs ?: 1); 499 500 /* TODO : should we use skb->sk here instead of sk ? */ 501 skb->priority = sk->sk_priority; 502 skb->mark = sk->sk_mark; 503 504 res = ip_local_out(net, sk, skb); 505 rcu_read_unlock(); 506 return res; 507 508 no_route: 509 rcu_read_unlock(); 510 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 511 kfree_skb(skb); 512 return -EHOSTUNREACH; 513 } 514 EXPORT_SYMBOL(ip_queue_xmit); 515 516 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) 517 { 518 to->pkt_type = from->pkt_type; 519 to->priority = from->priority; 520 to->protocol = from->protocol; 521 skb_dst_drop(to); 522 skb_dst_copy(to, from); 523 to->dev = from->dev; 524 to->mark = from->mark; 525 526 /* Copy the flags to each fragment. */ 527 IPCB(to)->flags = IPCB(from)->flags; 528 529 #ifdef CONFIG_NET_SCHED 530 to->tc_index = from->tc_index; 531 #endif 532 nf_copy(to, from); 533 #if IS_ENABLED(CONFIG_IP_VS) 534 to->ipvs_property = from->ipvs_property; 535 #endif 536 skb_copy_secmark(to, from); 537 } 538 539 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 540 unsigned int mtu, 541 int (*output)(struct net *, struct sock *, struct sk_buff *)) 542 { 543 struct iphdr *iph = ip_hdr(skb); 544 545 if ((iph->frag_off & htons(IP_DF)) == 0) 546 return ip_do_fragment(net, sk, skb, output); 547 548 if (unlikely(!skb->ignore_df || 549 (IPCB(skb)->frag_max_size && 550 IPCB(skb)->frag_max_size > mtu))) { 551 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 552 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 553 htonl(mtu)); 554 kfree_skb(skb); 555 return -EMSGSIZE; 556 } 557 558 return ip_do_fragment(net, sk, skb, output); 559 } 560 561 /* 562 * This IP datagram is too large to be sent in one piece. Break it up into 563 * smaller pieces (each of size equal to IP header plus 564 * a block of the data of the original IP data part) that will yet fit in a 565 * single device frame, and queue such a frame for sending. 566 */ 567 568 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 569 int (*output)(struct net *, struct sock *, struct sk_buff *)) 570 { 571 struct iphdr *iph; 572 int ptr; 573 struct sk_buff *skb2; 574 unsigned int mtu, hlen, left, len, ll_rs; 575 int offset; 576 __be16 not_last_frag; 577 struct rtable *rt = skb_rtable(skb); 578 int err = 0; 579 580 /* for offloaded checksums cleanup checksum before fragmentation */ 581 if (skb->ip_summed == CHECKSUM_PARTIAL && 582 (err = skb_checksum_help(skb))) 583 goto fail; 584 585 /* 586 * Point into the IP datagram header. 587 */ 588 589 iph = ip_hdr(skb); 590 591 mtu = ip_skb_dst_mtu(sk, skb); 592 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) 593 mtu = IPCB(skb)->frag_max_size; 594 595 /* 596 * Setup starting values. 597 */ 598 599 hlen = iph->ihl * 4; 600 mtu = mtu - hlen; /* Size of data space */ 601 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 602 ll_rs = LL_RESERVED_SPACE(rt->dst.dev); 603 604 /* When frag_list is given, use it. First, check its validity: 605 * some transformers could create wrong frag_list or break existing 606 * one, it is not prohibited. In this case fall back to copying. 607 * 608 * LATER: this step can be merged to real generation of fragments, 609 * we can switch to copy when see the first bad fragment. 610 */ 611 if (skb_has_frag_list(skb)) { 612 struct sk_buff *frag, *frag2; 613 unsigned int first_len = skb_pagelen(skb); 614 615 if (first_len - hlen > mtu || 616 ((first_len - hlen) & 7) || 617 ip_is_fragment(iph) || 618 skb_cloned(skb) || 619 skb_headroom(skb) < ll_rs) 620 goto slow_path; 621 622 skb_walk_frags(skb, frag) { 623 /* Correct geometry. */ 624 if (frag->len > mtu || 625 ((frag->len & 7) && frag->next) || 626 skb_headroom(frag) < hlen + ll_rs) 627 goto slow_path_clean; 628 629 /* Partially cloned skb? */ 630 if (skb_shared(frag)) 631 goto slow_path_clean; 632 633 BUG_ON(frag->sk); 634 if (skb->sk) { 635 frag->sk = skb->sk; 636 frag->destructor = sock_wfree; 637 } 638 skb->truesize -= frag->truesize; 639 } 640 641 /* Everything is OK. Generate! */ 642 643 err = 0; 644 offset = 0; 645 frag = skb_shinfo(skb)->frag_list; 646 skb_frag_list_init(skb); 647 skb->data_len = first_len - skb_headlen(skb); 648 skb->len = first_len; 649 iph->tot_len = htons(first_len); 650 iph->frag_off = htons(IP_MF); 651 ip_send_check(iph); 652 653 for (;;) { 654 /* Prepare header of the next frame, 655 * before previous one went down. */ 656 if (frag) { 657 frag->ip_summed = CHECKSUM_NONE; 658 skb_reset_transport_header(frag); 659 __skb_push(frag, hlen); 660 skb_reset_network_header(frag); 661 memcpy(skb_network_header(frag), iph, hlen); 662 iph = ip_hdr(frag); 663 iph->tot_len = htons(frag->len); 664 ip_copy_metadata(frag, skb); 665 if (offset == 0) 666 ip_options_fragment(frag); 667 offset += skb->len - hlen; 668 iph->frag_off = htons(offset>>3); 669 if (frag->next) 670 iph->frag_off |= htons(IP_MF); 671 /* Ready, complete checksum */ 672 ip_send_check(iph); 673 } 674 675 err = output(net, sk, skb); 676 677 if (!err) 678 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 679 if (err || !frag) 680 break; 681 682 skb = frag; 683 frag = skb->next; 684 skb->next = NULL; 685 } 686 687 if (err == 0) { 688 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 689 return 0; 690 } 691 692 while (frag) { 693 skb = frag->next; 694 kfree_skb(frag); 695 frag = skb; 696 } 697 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 698 return err; 699 700 slow_path_clean: 701 skb_walk_frags(skb, frag2) { 702 if (frag2 == frag) 703 break; 704 frag2->sk = NULL; 705 frag2->destructor = NULL; 706 skb->truesize += frag2->truesize; 707 } 708 } 709 710 slow_path: 711 iph = ip_hdr(skb); 712 713 left = skb->len - hlen; /* Space per frame */ 714 ptr = hlen; /* Where to start from */ 715 716 /* 717 * Fragment the datagram. 718 */ 719 720 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 721 not_last_frag = iph->frag_off & htons(IP_MF); 722 723 /* 724 * Keep copying data until we run out. 725 */ 726 727 while (left > 0) { 728 len = left; 729 /* IF: it doesn't fit, use 'mtu' - the data space left */ 730 if (len > mtu) 731 len = mtu; 732 /* IF: we are not sending up to and including the packet end 733 then align the next start on an eight byte boundary */ 734 if (len < left) { 735 len &= ~7; 736 } 737 738 /* Allocate buffer */ 739 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC); 740 if (!skb2) { 741 err = -ENOMEM; 742 goto fail; 743 } 744 745 /* 746 * Set up data on packet 747 */ 748 749 ip_copy_metadata(skb2, skb); 750 skb_reserve(skb2, ll_rs); 751 skb_put(skb2, len + hlen); 752 skb_reset_network_header(skb2); 753 skb2->transport_header = skb2->network_header + hlen; 754 755 /* 756 * Charge the memory for the fragment to any owner 757 * it might possess 758 */ 759 760 if (skb->sk) 761 skb_set_owner_w(skb2, skb->sk); 762 763 /* 764 * Copy the packet header into the new buffer. 765 */ 766 767 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); 768 769 /* 770 * Copy a block of the IP datagram. 771 */ 772 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) 773 BUG(); 774 left -= len; 775 776 /* 777 * Fill in the new header fields. 778 */ 779 iph = ip_hdr(skb2); 780 iph->frag_off = htons((offset >> 3)); 781 782 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU) 783 iph->frag_off |= htons(IP_DF); 784 785 /* ANK: dirty, but effective trick. Upgrade options only if 786 * the segment to be fragmented was THE FIRST (otherwise, 787 * options are already fixed) and make it ONCE 788 * on the initial skb, so that all the following fragments 789 * will inherit fixed options. 790 */ 791 if (offset == 0) 792 ip_options_fragment(skb); 793 794 /* 795 * Added AC : If we are fragmenting a fragment that's not the 796 * last fragment then keep MF on each bit 797 */ 798 if (left > 0 || not_last_frag) 799 iph->frag_off |= htons(IP_MF); 800 ptr += len; 801 offset += len; 802 803 /* 804 * Put this fragment into the sending queue. 805 */ 806 iph->tot_len = htons(len + hlen); 807 808 ip_send_check(iph); 809 810 err = output(net, sk, skb2); 811 if (err) 812 goto fail; 813 814 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 815 } 816 consume_skb(skb); 817 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 818 return err; 819 820 fail: 821 kfree_skb(skb); 822 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 823 return err; 824 } 825 EXPORT_SYMBOL(ip_do_fragment); 826 827 int 828 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 829 { 830 struct msghdr *msg = from; 831 832 if (skb->ip_summed == CHECKSUM_PARTIAL) { 833 if (!copy_from_iter_full(to, len, &msg->msg_iter)) 834 return -EFAULT; 835 } else { 836 __wsum csum = 0; 837 if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter)) 838 return -EFAULT; 839 skb->csum = csum_block_add(skb->csum, csum, odd); 840 } 841 return 0; 842 } 843 EXPORT_SYMBOL(ip_generic_getfrag); 844 845 static inline __wsum 846 csum_page(struct page *page, int offset, int copy) 847 { 848 char *kaddr; 849 __wsum csum; 850 kaddr = kmap(page); 851 csum = csum_partial(kaddr + offset, copy, 0); 852 kunmap(page); 853 return csum; 854 } 855 856 static int __ip_append_data(struct sock *sk, 857 struct flowi4 *fl4, 858 struct sk_buff_head *queue, 859 struct inet_cork *cork, 860 struct page_frag *pfrag, 861 int getfrag(void *from, char *to, int offset, 862 int len, int odd, struct sk_buff *skb), 863 void *from, int length, int transhdrlen, 864 unsigned int flags) 865 { 866 struct inet_sock *inet = inet_sk(sk); 867 struct sk_buff *skb; 868 869 struct ip_options *opt = cork->opt; 870 int hh_len; 871 int exthdrlen; 872 int mtu; 873 int copy; 874 int err; 875 int offset = 0; 876 unsigned int maxfraglen, fragheaderlen, maxnonfragsize; 877 int csummode = CHECKSUM_NONE; 878 struct rtable *rt = (struct rtable *)cork->dst; 879 unsigned int wmem_alloc_delta = 0; 880 u32 tskey = 0; 881 882 skb = skb_peek_tail(queue); 883 884 exthdrlen = !skb ? rt->dst.header_len : 0; 885 mtu = cork->fragsize; 886 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && 887 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) 888 tskey = sk->sk_tskey++; 889 890 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 891 892 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 893 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 894 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 895 896 if (cork->length + length > maxnonfragsize - fragheaderlen) { 897 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 898 mtu - (opt ? opt->optlen : 0)); 899 return -EMSGSIZE; 900 } 901 902 /* 903 * transhdrlen > 0 means that this is the first fragment and we wish 904 * it won't be fragmented in the future. 905 */ 906 if (transhdrlen && 907 length + fragheaderlen <= mtu && 908 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) && 909 !(flags & MSG_MORE) && 910 !exthdrlen) 911 csummode = CHECKSUM_PARTIAL; 912 913 cork->length += length; 914 915 /* So, what's going on in the loop below? 916 * 917 * We use calculated fragment length to generate chained skb, 918 * each of segments is IP fragment ready for sending to network after 919 * adding appropriate IP header. 920 */ 921 922 if (!skb) 923 goto alloc_new_skb; 924 925 while (length > 0) { 926 /* Check if the remaining data fits into current packet. */ 927 copy = mtu - skb->len; 928 if (copy < length) 929 copy = maxfraglen - skb->len; 930 if (copy <= 0) { 931 char *data; 932 unsigned int datalen; 933 unsigned int fraglen; 934 unsigned int fraggap; 935 unsigned int alloclen; 936 struct sk_buff *skb_prev; 937 alloc_new_skb: 938 skb_prev = skb; 939 if (skb_prev) 940 fraggap = skb_prev->len - maxfraglen; 941 else 942 fraggap = 0; 943 944 /* 945 * If remaining data exceeds the mtu, 946 * we know we need more fragment(s). 947 */ 948 datalen = length + fraggap; 949 if (datalen > mtu - fragheaderlen) 950 datalen = maxfraglen - fragheaderlen; 951 fraglen = datalen + fragheaderlen; 952 953 if ((flags & MSG_MORE) && 954 !(rt->dst.dev->features&NETIF_F_SG)) 955 alloclen = mtu; 956 else 957 alloclen = fraglen; 958 959 alloclen += exthdrlen; 960 961 /* The last fragment gets additional space at tail. 962 * Note, with MSG_MORE we overallocate on fragments, 963 * because we have no idea what fragment will be 964 * the last. 965 */ 966 if (datalen == length + fraggap) 967 alloclen += rt->dst.trailer_len; 968 969 if (transhdrlen) { 970 skb = sock_alloc_send_skb(sk, 971 alloclen + hh_len + 15, 972 (flags & MSG_DONTWAIT), &err); 973 } else { 974 skb = NULL; 975 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <= 976 2 * sk->sk_sndbuf) 977 skb = alloc_skb(alloclen + hh_len + 15, 978 sk->sk_allocation); 979 if (unlikely(!skb)) 980 err = -ENOBUFS; 981 } 982 if (!skb) 983 goto error; 984 985 /* 986 * Fill in the control structures 987 */ 988 skb->ip_summed = csummode; 989 skb->csum = 0; 990 skb_reserve(skb, hh_len); 991 992 /* only the initial fragment is time stamped */ 993 skb_shinfo(skb)->tx_flags = cork->tx_flags; 994 cork->tx_flags = 0; 995 skb_shinfo(skb)->tskey = tskey; 996 tskey = 0; 997 998 /* 999 * Find where to start putting bytes. 1000 */ 1001 data = skb_put(skb, fraglen + exthdrlen); 1002 skb_set_network_header(skb, exthdrlen); 1003 skb->transport_header = (skb->network_header + 1004 fragheaderlen); 1005 data += fragheaderlen + exthdrlen; 1006 1007 if (fraggap) { 1008 skb->csum = skb_copy_and_csum_bits( 1009 skb_prev, maxfraglen, 1010 data + transhdrlen, fraggap, 0); 1011 skb_prev->csum = csum_sub(skb_prev->csum, 1012 skb->csum); 1013 data += fraggap; 1014 pskb_trim_unique(skb_prev, maxfraglen); 1015 } 1016 1017 copy = datalen - transhdrlen - fraggap; 1018 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 1019 err = -EFAULT; 1020 kfree_skb(skb); 1021 goto error; 1022 } 1023 1024 offset += copy; 1025 length -= datalen - fraggap; 1026 transhdrlen = 0; 1027 exthdrlen = 0; 1028 csummode = CHECKSUM_NONE; 1029 1030 if ((flags & MSG_CONFIRM) && !skb_prev) 1031 skb_set_dst_pending_confirm(skb, 1); 1032 1033 /* 1034 * Put the packet on the pending queue. 1035 */ 1036 if (!skb->destructor) { 1037 skb->destructor = sock_wfree; 1038 skb->sk = sk; 1039 wmem_alloc_delta += skb->truesize; 1040 } 1041 __skb_queue_tail(queue, skb); 1042 continue; 1043 } 1044 1045 if (copy > length) 1046 copy = length; 1047 1048 if (!(rt->dst.dev->features&NETIF_F_SG)) { 1049 unsigned int off; 1050 1051 off = skb->len; 1052 if (getfrag(from, skb_put(skb, copy), 1053 offset, copy, off, skb) < 0) { 1054 __skb_trim(skb, off); 1055 err = -EFAULT; 1056 goto error; 1057 } 1058 } else { 1059 int i = skb_shinfo(skb)->nr_frags; 1060 1061 err = -ENOMEM; 1062 if (!sk_page_frag_refill(sk, pfrag)) 1063 goto error; 1064 1065 if (!skb_can_coalesce(skb, i, pfrag->page, 1066 pfrag->offset)) { 1067 err = -EMSGSIZE; 1068 if (i == MAX_SKB_FRAGS) 1069 goto error; 1070 1071 __skb_fill_page_desc(skb, i, pfrag->page, 1072 pfrag->offset, 0); 1073 skb_shinfo(skb)->nr_frags = ++i; 1074 get_page(pfrag->page); 1075 } 1076 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1077 if (getfrag(from, 1078 page_address(pfrag->page) + pfrag->offset, 1079 offset, copy, skb->len, skb) < 0) 1080 goto error_efault; 1081 1082 pfrag->offset += copy; 1083 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1084 skb->len += copy; 1085 skb->data_len += copy; 1086 skb->truesize += copy; 1087 wmem_alloc_delta += copy; 1088 } 1089 offset += copy; 1090 length -= copy; 1091 } 1092 1093 if (wmem_alloc_delta) 1094 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1095 return 0; 1096 1097 error_efault: 1098 err = -EFAULT; 1099 error: 1100 cork->length -= length; 1101 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1102 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1103 return err; 1104 } 1105 1106 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 1107 struct ipcm_cookie *ipc, struct rtable **rtp) 1108 { 1109 struct ip_options_rcu *opt; 1110 struct rtable *rt; 1111 1112 /* 1113 * setup for corking. 1114 */ 1115 opt = ipc->opt; 1116 if (opt) { 1117 if (!cork->opt) { 1118 cork->opt = kmalloc(sizeof(struct ip_options) + 40, 1119 sk->sk_allocation); 1120 if (unlikely(!cork->opt)) 1121 return -ENOBUFS; 1122 } 1123 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 1124 cork->flags |= IPCORK_OPT; 1125 cork->addr = ipc->addr; 1126 } 1127 rt = *rtp; 1128 if (unlikely(!rt)) 1129 return -EFAULT; 1130 /* 1131 * We steal reference to this route, caller should not release it 1132 */ 1133 *rtp = NULL; 1134 cork->fragsize = ip_sk_use_pmtu(sk) ? 1135 dst_mtu(&rt->dst) : rt->dst.dev->mtu; 1136 cork->dst = &rt->dst; 1137 cork->length = 0; 1138 cork->ttl = ipc->ttl; 1139 cork->tos = ipc->tos; 1140 cork->priority = ipc->priority; 1141 cork->tx_flags = ipc->tx_flags; 1142 1143 return 0; 1144 } 1145 1146 /* 1147 * ip_append_data() and ip_append_page() can make one large IP datagram 1148 * from many pieces of data. Each pieces will be holded on the socket 1149 * until ip_push_pending_frames() is called. Each piece can be a page 1150 * or non-page data. 1151 * 1152 * Not only UDP, other transport protocols - e.g. raw sockets - can use 1153 * this interface potentially. 1154 * 1155 * LATER: length must be adjusted by pad at tail, when it is required. 1156 */ 1157 int ip_append_data(struct sock *sk, struct flowi4 *fl4, 1158 int getfrag(void *from, char *to, int offset, int len, 1159 int odd, struct sk_buff *skb), 1160 void *from, int length, int transhdrlen, 1161 struct ipcm_cookie *ipc, struct rtable **rtp, 1162 unsigned int flags) 1163 { 1164 struct inet_sock *inet = inet_sk(sk); 1165 int err; 1166 1167 if (flags&MSG_PROBE) 1168 return 0; 1169 1170 if (skb_queue_empty(&sk->sk_write_queue)) { 1171 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 1172 if (err) 1173 return err; 1174 } else { 1175 transhdrlen = 0; 1176 } 1177 1178 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, 1179 sk_page_frag(sk), getfrag, 1180 from, length, transhdrlen, flags); 1181 } 1182 1183 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 1184 int offset, size_t size, int flags) 1185 { 1186 struct inet_sock *inet = inet_sk(sk); 1187 struct sk_buff *skb; 1188 struct rtable *rt; 1189 struct ip_options *opt = NULL; 1190 struct inet_cork *cork; 1191 int hh_len; 1192 int mtu; 1193 int len; 1194 int err; 1195 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; 1196 1197 if (inet->hdrincl) 1198 return -EPERM; 1199 1200 if (flags&MSG_PROBE) 1201 return 0; 1202 1203 if (skb_queue_empty(&sk->sk_write_queue)) 1204 return -EINVAL; 1205 1206 cork = &inet->cork.base; 1207 rt = (struct rtable *)cork->dst; 1208 if (cork->flags & IPCORK_OPT) 1209 opt = cork->opt; 1210 1211 if (!(rt->dst.dev->features&NETIF_F_SG)) 1212 return -EOPNOTSUPP; 1213 1214 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1215 mtu = cork->fragsize; 1216 1217 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1218 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1219 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 1220 1221 if (cork->length + size > maxnonfragsize - fragheaderlen) { 1222 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 1223 mtu - (opt ? opt->optlen : 0)); 1224 return -EMSGSIZE; 1225 } 1226 1227 skb = skb_peek_tail(&sk->sk_write_queue); 1228 if (!skb) 1229 return -EINVAL; 1230 1231 cork->length += size; 1232 1233 while (size > 0) { 1234 /* Check if the remaining data fits into current packet. */ 1235 len = mtu - skb->len; 1236 if (len < size) 1237 len = maxfraglen - skb->len; 1238 1239 if (len <= 0) { 1240 struct sk_buff *skb_prev; 1241 int alloclen; 1242 1243 skb_prev = skb; 1244 fraggap = skb_prev->len - maxfraglen; 1245 1246 alloclen = fragheaderlen + hh_len + fraggap + 15; 1247 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 1248 if (unlikely(!skb)) { 1249 err = -ENOBUFS; 1250 goto error; 1251 } 1252 1253 /* 1254 * Fill in the control structures 1255 */ 1256 skb->ip_summed = CHECKSUM_NONE; 1257 skb->csum = 0; 1258 skb_reserve(skb, hh_len); 1259 1260 /* 1261 * Find where to start putting bytes. 1262 */ 1263 skb_put(skb, fragheaderlen + fraggap); 1264 skb_reset_network_header(skb); 1265 skb->transport_header = (skb->network_header + 1266 fragheaderlen); 1267 if (fraggap) { 1268 skb->csum = skb_copy_and_csum_bits(skb_prev, 1269 maxfraglen, 1270 skb_transport_header(skb), 1271 fraggap, 0); 1272 skb_prev->csum = csum_sub(skb_prev->csum, 1273 skb->csum); 1274 pskb_trim_unique(skb_prev, maxfraglen); 1275 } 1276 1277 /* 1278 * Put the packet on the pending queue. 1279 */ 1280 __skb_queue_tail(&sk->sk_write_queue, skb); 1281 continue; 1282 } 1283 1284 if (len > size) 1285 len = size; 1286 1287 if (skb_append_pagefrags(skb, page, offset, len)) { 1288 err = -EMSGSIZE; 1289 goto error; 1290 } 1291 1292 if (skb->ip_summed == CHECKSUM_NONE) { 1293 __wsum csum; 1294 csum = csum_page(page, offset, len); 1295 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1296 } 1297 1298 skb->len += len; 1299 skb->data_len += len; 1300 skb->truesize += len; 1301 refcount_add(len, &sk->sk_wmem_alloc); 1302 offset += len; 1303 size -= len; 1304 } 1305 return 0; 1306 1307 error: 1308 cork->length -= size; 1309 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1310 return err; 1311 } 1312 1313 static void ip_cork_release(struct inet_cork *cork) 1314 { 1315 cork->flags &= ~IPCORK_OPT; 1316 kfree(cork->opt); 1317 cork->opt = NULL; 1318 dst_release(cork->dst); 1319 cork->dst = NULL; 1320 } 1321 1322 /* 1323 * Combined all pending IP fragments on the socket as one IP datagram 1324 * and push them out. 1325 */ 1326 struct sk_buff *__ip_make_skb(struct sock *sk, 1327 struct flowi4 *fl4, 1328 struct sk_buff_head *queue, 1329 struct inet_cork *cork) 1330 { 1331 struct sk_buff *skb, *tmp_skb; 1332 struct sk_buff **tail_skb; 1333 struct inet_sock *inet = inet_sk(sk); 1334 struct net *net = sock_net(sk); 1335 struct ip_options *opt = NULL; 1336 struct rtable *rt = (struct rtable *)cork->dst; 1337 struct iphdr *iph; 1338 __be16 df = 0; 1339 __u8 ttl; 1340 1341 skb = __skb_dequeue(queue); 1342 if (!skb) 1343 goto out; 1344 tail_skb = &(skb_shinfo(skb)->frag_list); 1345 1346 /* move skb->data to ip header from ext header */ 1347 if (skb->data < skb_network_header(skb)) 1348 __skb_pull(skb, skb_network_offset(skb)); 1349 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1350 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1351 *tail_skb = tmp_skb; 1352 tail_skb = &(tmp_skb->next); 1353 skb->len += tmp_skb->len; 1354 skb->data_len += tmp_skb->len; 1355 skb->truesize += tmp_skb->truesize; 1356 tmp_skb->destructor = NULL; 1357 tmp_skb->sk = NULL; 1358 } 1359 1360 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 1361 * to fragment the frame generated here. No matter, what transforms 1362 * how transforms change size of the packet, it will come out. 1363 */ 1364 skb->ignore_df = ip_sk_ignore_df(sk); 1365 1366 /* DF bit is set when we want to see DF on outgoing frames. 1367 * If ignore_df is set too, we still allow to fragment this frame 1368 * locally. */ 1369 if (inet->pmtudisc == IP_PMTUDISC_DO || 1370 inet->pmtudisc == IP_PMTUDISC_PROBE || 1371 (skb->len <= dst_mtu(&rt->dst) && 1372 ip_dont_fragment(sk, &rt->dst))) 1373 df = htons(IP_DF); 1374 1375 if (cork->flags & IPCORK_OPT) 1376 opt = cork->opt; 1377 1378 if (cork->ttl != 0) 1379 ttl = cork->ttl; 1380 else if (rt->rt_type == RTN_MULTICAST) 1381 ttl = inet->mc_ttl; 1382 else 1383 ttl = ip_select_ttl(inet, &rt->dst); 1384 1385 iph = ip_hdr(skb); 1386 iph->version = 4; 1387 iph->ihl = 5; 1388 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; 1389 iph->frag_off = df; 1390 iph->ttl = ttl; 1391 iph->protocol = sk->sk_protocol; 1392 ip_copy_addrs(iph, fl4); 1393 ip_select_ident(net, skb, sk); 1394 1395 if (opt) { 1396 iph->ihl += opt->optlen>>2; 1397 ip_options_build(skb, opt, cork->addr, rt, 0); 1398 } 1399 1400 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; 1401 skb->mark = sk->sk_mark; 1402 /* 1403 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1404 * on dst refcount 1405 */ 1406 cork->dst = NULL; 1407 skb_dst_set(skb, &rt->dst); 1408 1409 if (iph->protocol == IPPROTO_ICMP) 1410 icmp_out_count(net, ((struct icmphdr *) 1411 skb_transport_header(skb))->type); 1412 1413 ip_cork_release(cork); 1414 out: 1415 return skb; 1416 } 1417 1418 int ip_send_skb(struct net *net, struct sk_buff *skb) 1419 { 1420 int err; 1421 1422 err = ip_local_out(net, skb->sk, skb); 1423 if (err) { 1424 if (err > 0) 1425 err = net_xmit_errno(err); 1426 if (err) 1427 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 1428 } 1429 1430 return err; 1431 } 1432 1433 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 1434 { 1435 struct sk_buff *skb; 1436 1437 skb = ip_finish_skb(sk, fl4); 1438 if (!skb) 1439 return 0; 1440 1441 /* Netfilter gets whole the not fragmented skb. */ 1442 return ip_send_skb(sock_net(sk), skb); 1443 } 1444 1445 /* 1446 * Throw away all pending data on the socket. 1447 */ 1448 static void __ip_flush_pending_frames(struct sock *sk, 1449 struct sk_buff_head *queue, 1450 struct inet_cork *cork) 1451 { 1452 struct sk_buff *skb; 1453 1454 while ((skb = __skb_dequeue_tail(queue)) != NULL) 1455 kfree_skb(skb); 1456 1457 ip_cork_release(cork); 1458 } 1459 1460 void ip_flush_pending_frames(struct sock *sk) 1461 { 1462 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 1463 } 1464 1465 struct sk_buff *ip_make_skb(struct sock *sk, 1466 struct flowi4 *fl4, 1467 int getfrag(void *from, char *to, int offset, 1468 int len, int odd, struct sk_buff *skb), 1469 void *from, int length, int transhdrlen, 1470 struct ipcm_cookie *ipc, struct rtable **rtp, 1471 unsigned int flags) 1472 { 1473 struct inet_cork cork; 1474 struct sk_buff_head queue; 1475 int err; 1476 1477 if (flags & MSG_PROBE) 1478 return NULL; 1479 1480 __skb_queue_head_init(&queue); 1481 1482 cork.flags = 0; 1483 cork.addr = 0; 1484 cork.opt = NULL; 1485 err = ip_setup_cork(sk, &cork, ipc, rtp); 1486 if (err) 1487 return ERR_PTR(err); 1488 1489 err = __ip_append_data(sk, fl4, &queue, &cork, 1490 ¤t->task_frag, getfrag, 1491 from, length, transhdrlen, flags); 1492 if (err) { 1493 __ip_flush_pending_frames(sk, &queue, &cork); 1494 return ERR_PTR(err); 1495 } 1496 1497 return __ip_make_skb(sk, fl4, &queue, &cork); 1498 } 1499 1500 /* 1501 * Fetch data from kernel space and fill in checksum if needed. 1502 */ 1503 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 1504 int len, int odd, struct sk_buff *skb) 1505 { 1506 __wsum csum; 1507 1508 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); 1509 skb->csum = csum_block_add(skb->csum, csum, odd); 1510 return 0; 1511 } 1512 1513 /* 1514 * Generic function to send a packet as reply to another packet. 1515 * Used to send some TCP resets/acks so far. 1516 */ 1517 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, 1518 const struct ip_options *sopt, 1519 __be32 daddr, __be32 saddr, 1520 const struct ip_reply_arg *arg, 1521 unsigned int len) 1522 { 1523 struct ip_options_data replyopts; 1524 struct ipcm_cookie ipc; 1525 struct flowi4 fl4; 1526 struct rtable *rt = skb_rtable(skb); 1527 struct net *net = sock_net(sk); 1528 struct sk_buff *nskb; 1529 int err; 1530 int oif; 1531 1532 if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt)) 1533 return; 1534 1535 ipc.addr = daddr; 1536 ipc.opt = NULL; 1537 ipc.tx_flags = 0; 1538 ipc.ttl = 0; 1539 ipc.tos = -1; 1540 1541 if (replyopts.opt.opt.optlen) { 1542 ipc.opt = &replyopts.opt; 1543 1544 if (replyopts.opt.opt.srr) 1545 daddr = replyopts.opt.opt.faddr; 1546 } 1547 1548 oif = arg->bound_dev_if; 1549 if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) 1550 oif = skb->skb_iif; 1551 1552 flowi4_init_output(&fl4, oif, 1553 IP4_REPLY_MARK(net, skb->mark), 1554 RT_TOS(arg->tos), 1555 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, 1556 ip_reply_arg_flowi_flags(arg), 1557 daddr, saddr, 1558 tcp_hdr(skb)->source, tcp_hdr(skb)->dest, 1559 arg->uid); 1560 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 1561 rt = ip_route_output_key(net, &fl4); 1562 if (IS_ERR(rt)) 1563 return; 1564 1565 inet_sk(sk)->tos = arg->tos; 1566 1567 sk->sk_priority = skb->priority; 1568 sk->sk_protocol = ip_hdr(skb)->protocol; 1569 sk->sk_bound_dev_if = arg->bound_dev_if; 1570 sk->sk_sndbuf = sysctl_wmem_default; 1571 sk->sk_mark = fl4.flowi4_mark; 1572 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, 1573 len, 0, &ipc, &rt, MSG_DONTWAIT); 1574 if (unlikely(err)) { 1575 ip_flush_pending_frames(sk); 1576 goto out; 1577 } 1578 1579 nskb = skb_peek(&sk->sk_write_queue); 1580 if (nskb) { 1581 if (arg->csumoffset >= 0) 1582 *((__sum16 *)skb_transport_header(nskb) + 1583 arg->csumoffset) = csum_fold(csum_add(nskb->csum, 1584 arg->csum)); 1585 nskb->ip_summed = CHECKSUM_NONE; 1586 ip_push_pending_frames(sk, &fl4); 1587 } 1588 out: 1589 ip_rt_put(rt); 1590 } 1591 1592 void __init ip_init(void) 1593 { 1594 ip_rt_init(); 1595 inet_initpeers(); 1596 1597 #if defined(CONFIG_IP_MULTICAST) 1598 igmp_mc_init(); 1599 #endif 1600 } 1601