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 skb_mark_not_on_list(segs); 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 __u8 tos) 428 { 429 struct inet_sock *inet = inet_sk(sk); 430 struct net *net = sock_net(sk); 431 struct ip_options_rcu *inet_opt; 432 struct flowi4 *fl4; 433 struct rtable *rt; 434 struct iphdr *iph; 435 int res; 436 437 /* Skip all of this if the packet is already routed, 438 * f.e. by something like SCTP. 439 */ 440 rcu_read_lock(); 441 inet_opt = rcu_dereference(inet->inet_opt); 442 fl4 = &fl->u.ip4; 443 rt = skb_rtable(skb); 444 if (rt) 445 goto packet_routed; 446 447 /* Make sure we can route this packet. */ 448 rt = (struct rtable *)__sk_dst_check(sk, 0); 449 if (!rt) { 450 __be32 daddr; 451 452 /* Use correct destination address if we have options. */ 453 daddr = inet->inet_daddr; 454 if (inet_opt && inet_opt->opt.srr) 455 daddr = inet_opt->opt.faddr; 456 457 /* If this fails, retransmit mechanism of transport layer will 458 * keep trying until route appears or the connection times 459 * itself out. 460 */ 461 rt = ip_route_output_ports(net, fl4, sk, 462 daddr, inet->inet_saddr, 463 inet->inet_dport, 464 inet->inet_sport, 465 sk->sk_protocol, 466 RT_CONN_FLAGS_TOS(sk, tos), 467 sk->sk_bound_dev_if); 468 if (IS_ERR(rt)) 469 goto no_route; 470 sk_setup_caps(sk, &rt->dst); 471 } 472 skb_dst_set_noref(skb, &rt->dst); 473 474 packet_routed: 475 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) 476 goto no_route; 477 478 /* OK, we know where to send it, allocate and build IP header. */ 479 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 480 skb_reset_network_header(skb); 481 iph = ip_hdr(skb); 482 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff)); 483 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) 484 iph->frag_off = htons(IP_DF); 485 else 486 iph->frag_off = 0; 487 iph->ttl = ip_select_ttl(inet, &rt->dst); 488 iph->protocol = sk->sk_protocol; 489 ip_copy_addrs(iph, fl4); 490 491 /* Transport layer set skb->h.foo itself. */ 492 493 if (inet_opt && inet_opt->opt.optlen) { 494 iph->ihl += inet_opt->opt.optlen >> 2; 495 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 496 } 497 498 ip_select_ident_segs(net, skb, sk, 499 skb_shinfo(skb)->gso_segs ?: 1); 500 501 /* TODO : should we use skb->sk here instead of sk ? */ 502 skb->priority = sk->sk_priority; 503 skb->mark = sk->sk_mark; 504 505 res = ip_local_out(net, sk, skb); 506 rcu_read_unlock(); 507 return res; 508 509 no_route: 510 rcu_read_unlock(); 511 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 512 kfree_skb(skb); 513 return -EHOSTUNREACH; 514 } 515 EXPORT_SYMBOL(__ip_queue_xmit); 516 517 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) 518 { 519 to->pkt_type = from->pkt_type; 520 to->priority = from->priority; 521 to->protocol = from->protocol; 522 to->skb_iif = from->skb_iif; 523 skb_dst_drop(to); 524 skb_dst_copy(to, from); 525 to->dev = from->dev; 526 to->mark = from->mark; 527 528 skb_copy_hash(to, from); 529 530 /* Copy the flags to each fragment. */ 531 IPCB(to)->flags = IPCB(from)->flags; 532 533 #ifdef CONFIG_NET_SCHED 534 to->tc_index = from->tc_index; 535 #endif 536 nf_copy(to, from); 537 skb_ext_copy(to, from); 538 #if IS_ENABLED(CONFIG_IP_VS) 539 to->ipvs_property = from->ipvs_property; 540 #endif 541 skb_copy_secmark(to, from); 542 } 543 544 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 545 unsigned int mtu, 546 int (*output)(struct net *, struct sock *, struct sk_buff *)) 547 { 548 struct iphdr *iph = ip_hdr(skb); 549 550 if ((iph->frag_off & htons(IP_DF)) == 0) 551 return ip_do_fragment(net, sk, skb, output); 552 553 if (unlikely(!skb->ignore_df || 554 (IPCB(skb)->frag_max_size && 555 IPCB(skb)->frag_max_size > mtu))) { 556 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 557 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 558 htonl(mtu)); 559 kfree_skb(skb); 560 return -EMSGSIZE; 561 } 562 563 return ip_do_fragment(net, sk, skb, output); 564 } 565 566 /* 567 * This IP datagram is too large to be sent in one piece. Break it up into 568 * smaller pieces (each of size equal to IP header plus 569 * a block of the data of the original IP data part) that will yet fit in a 570 * single device frame, and queue such a frame for sending. 571 */ 572 573 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 574 int (*output)(struct net *, struct sock *, struct sk_buff *)) 575 { 576 struct iphdr *iph; 577 int ptr; 578 struct sk_buff *skb2; 579 unsigned int mtu, hlen, left, len, ll_rs; 580 int offset; 581 __be16 not_last_frag; 582 struct rtable *rt = skb_rtable(skb); 583 int err = 0; 584 585 /* for offloaded checksums cleanup checksum before fragmentation */ 586 if (skb->ip_summed == CHECKSUM_PARTIAL && 587 (err = skb_checksum_help(skb))) 588 goto fail; 589 590 /* 591 * Point into the IP datagram header. 592 */ 593 594 iph = ip_hdr(skb); 595 596 mtu = ip_skb_dst_mtu(sk, skb); 597 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) 598 mtu = IPCB(skb)->frag_max_size; 599 600 /* 601 * Setup starting values. 602 */ 603 604 hlen = iph->ihl * 4; 605 mtu = mtu - hlen; /* Size of data space */ 606 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 607 ll_rs = LL_RESERVED_SPACE(rt->dst.dev); 608 609 /* When frag_list is given, use it. First, check its validity: 610 * some transformers could create wrong frag_list or break existing 611 * one, it is not prohibited. In this case fall back to copying. 612 * 613 * LATER: this step can be merged to real generation of fragments, 614 * we can switch to copy when see the first bad fragment. 615 */ 616 if (skb_has_frag_list(skb)) { 617 struct sk_buff *frag, *frag2; 618 unsigned int first_len = skb_pagelen(skb); 619 620 if (first_len - hlen > mtu || 621 ((first_len - hlen) & 7) || 622 ip_is_fragment(iph) || 623 skb_cloned(skb) || 624 skb_headroom(skb) < ll_rs) 625 goto slow_path; 626 627 skb_walk_frags(skb, frag) { 628 /* Correct geometry. */ 629 if (frag->len > mtu || 630 ((frag->len & 7) && frag->next) || 631 skb_headroom(frag) < hlen + ll_rs) 632 goto slow_path_clean; 633 634 /* Partially cloned skb? */ 635 if (skb_shared(frag)) 636 goto slow_path_clean; 637 638 BUG_ON(frag->sk); 639 if (skb->sk) { 640 frag->sk = skb->sk; 641 frag->destructor = sock_wfree; 642 } 643 skb->truesize -= frag->truesize; 644 } 645 646 /* Everything is OK. Generate! */ 647 648 err = 0; 649 offset = 0; 650 frag = skb_shinfo(skb)->frag_list; 651 skb_frag_list_init(skb); 652 skb->data_len = first_len - skb_headlen(skb); 653 skb->len = first_len; 654 iph->tot_len = htons(first_len); 655 iph->frag_off = htons(IP_MF); 656 ip_send_check(iph); 657 658 for (;;) { 659 /* Prepare header of the next frame, 660 * before previous one went down. */ 661 if (frag) { 662 frag->ip_summed = CHECKSUM_NONE; 663 skb_reset_transport_header(frag); 664 __skb_push(frag, hlen); 665 skb_reset_network_header(frag); 666 memcpy(skb_network_header(frag), iph, hlen); 667 iph = ip_hdr(frag); 668 iph->tot_len = htons(frag->len); 669 ip_copy_metadata(frag, skb); 670 if (offset == 0) 671 ip_options_fragment(frag); 672 offset += skb->len - hlen; 673 iph->frag_off = htons(offset>>3); 674 if (frag->next) 675 iph->frag_off |= htons(IP_MF); 676 /* Ready, complete checksum */ 677 ip_send_check(iph); 678 } 679 680 err = output(net, sk, skb); 681 682 if (!err) 683 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 684 if (err || !frag) 685 break; 686 687 skb = frag; 688 frag = skb->next; 689 skb_mark_not_on_list(skb); 690 } 691 692 if (err == 0) { 693 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 694 return 0; 695 } 696 697 while (frag) { 698 skb = frag->next; 699 kfree_skb(frag); 700 frag = skb; 701 } 702 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 703 return err; 704 705 slow_path_clean: 706 skb_walk_frags(skb, frag2) { 707 if (frag2 == frag) 708 break; 709 frag2->sk = NULL; 710 frag2->destructor = NULL; 711 skb->truesize += frag2->truesize; 712 } 713 } 714 715 slow_path: 716 iph = ip_hdr(skb); 717 718 left = skb->len - hlen; /* Space per frame */ 719 ptr = hlen; /* Where to start from */ 720 721 /* 722 * Fragment the datagram. 723 */ 724 725 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 726 not_last_frag = iph->frag_off & htons(IP_MF); 727 728 /* 729 * Keep copying data until we run out. 730 */ 731 732 while (left > 0) { 733 len = left; 734 /* IF: it doesn't fit, use 'mtu' - the data space left */ 735 if (len > mtu) 736 len = mtu; 737 /* IF: we are not sending up to and including the packet end 738 then align the next start on an eight byte boundary */ 739 if (len < left) { 740 len &= ~7; 741 } 742 743 /* Allocate buffer */ 744 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC); 745 if (!skb2) { 746 err = -ENOMEM; 747 goto fail; 748 } 749 750 /* 751 * Set up data on packet 752 */ 753 754 ip_copy_metadata(skb2, skb); 755 skb_reserve(skb2, ll_rs); 756 skb_put(skb2, len + hlen); 757 skb_reset_network_header(skb2); 758 skb2->transport_header = skb2->network_header + hlen; 759 760 /* 761 * Charge the memory for the fragment to any owner 762 * it might possess 763 */ 764 765 if (skb->sk) 766 skb_set_owner_w(skb2, skb->sk); 767 768 /* 769 * Copy the packet header into the new buffer. 770 */ 771 772 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); 773 774 /* 775 * Copy a block of the IP datagram. 776 */ 777 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) 778 BUG(); 779 left -= len; 780 781 /* 782 * Fill in the new header fields. 783 */ 784 iph = ip_hdr(skb2); 785 iph->frag_off = htons((offset >> 3)); 786 787 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU) 788 iph->frag_off |= htons(IP_DF); 789 790 /* ANK: dirty, but effective trick. Upgrade options only if 791 * the segment to be fragmented was THE FIRST (otherwise, 792 * options are already fixed) and make it ONCE 793 * on the initial skb, so that all the following fragments 794 * will inherit fixed options. 795 */ 796 if (offset == 0) 797 ip_options_fragment(skb); 798 799 /* 800 * Added AC : If we are fragmenting a fragment that's not the 801 * last fragment then keep MF on each bit 802 */ 803 if (left > 0 || not_last_frag) 804 iph->frag_off |= htons(IP_MF); 805 ptr += len; 806 offset += len; 807 808 /* 809 * Put this fragment into the sending queue. 810 */ 811 iph->tot_len = htons(len + hlen); 812 813 ip_send_check(iph); 814 815 err = output(net, sk, skb2); 816 if (err) 817 goto fail; 818 819 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 820 } 821 consume_skb(skb); 822 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 823 return err; 824 825 fail: 826 kfree_skb(skb); 827 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 828 return err; 829 } 830 EXPORT_SYMBOL(ip_do_fragment); 831 832 int 833 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 834 { 835 struct msghdr *msg = from; 836 837 if (skb->ip_summed == CHECKSUM_PARTIAL) { 838 if (!copy_from_iter_full(to, len, &msg->msg_iter)) 839 return -EFAULT; 840 } else { 841 __wsum csum = 0; 842 if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter)) 843 return -EFAULT; 844 skb->csum = csum_block_add(skb->csum, csum, odd); 845 } 846 return 0; 847 } 848 EXPORT_SYMBOL(ip_generic_getfrag); 849 850 static inline __wsum 851 csum_page(struct page *page, int offset, int copy) 852 { 853 char *kaddr; 854 __wsum csum; 855 kaddr = kmap(page); 856 csum = csum_partial(kaddr + offset, copy, 0); 857 kunmap(page); 858 return csum; 859 } 860 861 static int __ip_append_data(struct sock *sk, 862 struct flowi4 *fl4, 863 struct sk_buff_head *queue, 864 struct inet_cork *cork, 865 struct page_frag *pfrag, 866 int getfrag(void *from, char *to, int offset, 867 int len, int odd, struct sk_buff *skb), 868 void *from, int length, int transhdrlen, 869 unsigned int flags) 870 { 871 struct inet_sock *inet = inet_sk(sk); 872 struct ubuf_info *uarg = NULL; 873 struct sk_buff *skb; 874 875 struct ip_options *opt = cork->opt; 876 int hh_len; 877 int exthdrlen; 878 int mtu; 879 int copy; 880 int err; 881 int offset = 0; 882 unsigned int maxfraglen, fragheaderlen, maxnonfragsize; 883 int csummode = CHECKSUM_NONE; 884 struct rtable *rt = (struct rtable *)cork->dst; 885 unsigned int wmem_alloc_delta = 0; 886 bool paged, extra_uref; 887 u32 tskey = 0; 888 889 skb = skb_peek_tail(queue); 890 891 exthdrlen = !skb ? rt->dst.header_len : 0; 892 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; 893 paged = !!cork->gso_size; 894 895 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && 896 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) 897 tskey = sk->sk_tskey++; 898 899 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 900 901 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 902 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 903 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 904 905 if (cork->length + length > maxnonfragsize - fragheaderlen) { 906 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 907 mtu - (opt ? opt->optlen : 0)); 908 return -EMSGSIZE; 909 } 910 911 /* 912 * transhdrlen > 0 means that this is the first fragment and we wish 913 * it won't be fragmented in the future. 914 */ 915 if (transhdrlen && 916 length + fragheaderlen <= mtu && 917 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) && 918 (!(flags & MSG_MORE) || cork->gso_size) && 919 (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM))) 920 csummode = CHECKSUM_PARTIAL; 921 922 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) { 923 uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb)); 924 if (!uarg) 925 return -ENOBUFS; 926 extra_uref = true; 927 if (rt->dst.dev->features & NETIF_F_SG && 928 csummode == CHECKSUM_PARTIAL) { 929 paged = true; 930 } else { 931 uarg->zerocopy = 0; 932 skb_zcopy_set(skb, uarg, &extra_uref); 933 } 934 } 935 936 cork->length += length; 937 938 /* So, what's going on in the loop below? 939 * 940 * We use calculated fragment length to generate chained skb, 941 * each of segments is IP fragment ready for sending to network after 942 * adding appropriate IP header. 943 */ 944 945 if (!skb) 946 goto alloc_new_skb; 947 948 while (length > 0) { 949 /* Check if the remaining data fits into current packet. */ 950 copy = mtu - skb->len; 951 if (copy < length) 952 copy = maxfraglen - skb->len; 953 if (copy <= 0) { 954 char *data; 955 unsigned int datalen; 956 unsigned int fraglen; 957 unsigned int fraggap; 958 unsigned int alloclen; 959 unsigned int pagedlen; 960 struct sk_buff *skb_prev; 961 alloc_new_skb: 962 skb_prev = skb; 963 if (skb_prev) 964 fraggap = skb_prev->len - maxfraglen; 965 else 966 fraggap = 0; 967 968 /* 969 * If remaining data exceeds the mtu, 970 * we know we need more fragment(s). 971 */ 972 datalen = length + fraggap; 973 if (datalen > mtu - fragheaderlen) 974 datalen = maxfraglen - fragheaderlen; 975 fraglen = datalen + fragheaderlen; 976 pagedlen = 0; 977 978 if ((flags & MSG_MORE) && 979 !(rt->dst.dev->features&NETIF_F_SG)) 980 alloclen = mtu; 981 else if (!paged) 982 alloclen = fraglen; 983 else { 984 alloclen = min_t(int, fraglen, MAX_HEADER); 985 pagedlen = fraglen - alloclen; 986 } 987 988 alloclen += exthdrlen; 989 990 /* The last fragment gets additional space at tail. 991 * Note, with MSG_MORE we overallocate on fragments, 992 * because we have no idea what fragment will be 993 * the last. 994 */ 995 if (datalen == length + fraggap) 996 alloclen += rt->dst.trailer_len; 997 998 if (transhdrlen) { 999 skb = sock_alloc_send_skb(sk, 1000 alloclen + hh_len + 15, 1001 (flags & MSG_DONTWAIT), &err); 1002 } else { 1003 skb = NULL; 1004 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <= 1005 2 * sk->sk_sndbuf) 1006 skb = alloc_skb(alloclen + hh_len + 15, 1007 sk->sk_allocation); 1008 if (unlikely(!skb)) 1009 err = -ENOBUFS; 1010 } 1011 if (!skb) 1012 goto error; 1013 1014 /* 1015 * Fill in the control structures 1016 */ 1017 skb->ip_summed = csummode; 1018 skb->csum = 0; 1019 skb_reserve(skb, hh_len); 1020 1021 /* 1022 * Find where to start putting bytes. 1023 */ 1024 data = skb_put(skb, fraglen + exthdrlen - pagedlen); 1025 skb_set_network_header(skb, exthdrlen); 1026 skb->transport_header = (skb->network_header + 1027 fragheaderlen); 1028 data += fragheaderlen + exthdrlen; 1029 1030 if (fraggap) { 1031 skb->csum = skb_copy_and_csum_bits( 1032 skb_prev, maxfraglen, 1033 data + transhdrlen, fraggap, 0); 1034 skb_prev->csum = csum_sub(skb_prev->csum, 1035 skb->csum); 1036 data += fraggap; 1037 pskb_trim_unique(skb_prev, maxfraglen); 1038 } 1039 1040 copy = datalen - transhdrlen - fraggap - pagedlen; 1041 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 1042 err = -EFAULT; 1043 kfree_skb(skb); 1044 goto error; 1045 } 1046 1047 offset += copy; 1048 length -= copy + transhdrlen; 1049 transhdrlen = 0; 1050 exthdrlen = 0; 1051 csummode = CHECKSUM_NONE; 1052 1053 /* only the initial fragment is time stamped */ 1054 skb_shinfo(skb)->tx_flags = cork->tx_flags; 1055 cork->tx_flags = 0; 1056 skb_shinfo(skb)->tskey = tskey; 1057 tskey = 0; 1058 skb_zcopy_set(skb, uarg, &extra_uref); 1059 1060 if ((flags & MSG_CONFIRM) && !skb_prev) 1061 skb_set_dst_pending_confirm(skb, 1); 1062 1063 /* 1064 * Put the packet on the pending queue. 1065 */ 1066 if (!skb->destructor) { 1067 skb->destructor = sock_wfree; 1068 skb->sk = sk; 1069 wmem_alloc_delta += skb->truesize; 1070 } 1071 __skb_queue_tail(queue, skb); 1072 continue; 1073 } 1074 1075 if (copy > length) 1076 copy = length; 1077 1078 if (!(rt->dst.dev->features&NETIF_F_SG) && 1079 skb_tailroom(skb) >= copy) { 1080 unsigned int off; 1081 1082 off = skb->len; 1083 if (getfrag(from, skb_put(skb, copy), 1084 offset, copy, off, skb) < 0) { 1085 __skb_trim(skb, off); 1086 err = -EFAULT; 1087 goto error; 1088 } 1089 } else if (!uarg || !uarg->zerocopy) { 1090 int i = skb_shinfo(skb)->nr_frags; 1091 1092 err = -ENOMEM; 1093 if (!sk_page_frag_refill(sk, pfrag)) 1094 goto error; 1095 1096 if (!skb_can_coalesce(skb, i, pfrag->page, 1097 pfrag->offset)) { 1098 err = -EMSGSIZE; 1099 if (i == MAX_SKB_FRAGS) 1100 goto error; 1101 1102 __skb_fill_page_desc(skb, i, pfrag->page, 1103 pfrag->offset, 0); 1104 skb_shinfo(skb)->nr_frags = ++i; 1105 get_page(pfrag->page); 1106 } 1107 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1108 if (getfrag(from, 1109 page_address(pfrag->page) + pfrag->offset, 1110 offset, copy, skb->len, skb) < 0) 1111 goto error_efault; 1112 1113 pfrag->offset += copy; 1114 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1115 skb->len += copy; 1116 skb->data_len += copy; 1117 skb->truesize += copy; 1118 wmem_alloc_delta += copy; 1119 } else { 1120 err = skb_zerocopy_iter_dgram(skb, from, copy); 1121 if (err < 0) 1122 goto error; 1123 } 1124 offset += copy; 1125 length -= copy; 1126 } 1127 1128 if (wmem_alloc_delta) 1129 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1130 return 0; 1131 1132 error_efault: 1133 err = -EFAULT; 1134 error: 1135 if (uarg) 1136 sock_zerocopy_put_abort(uarg, extra_uref); 1137 cork->length -= length; 1138 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1139 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1140 return err; 1141 } 1142 1143 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 1144 struct ipcm_cookie *ipc, struct rtable **rtp) 1145 { 1146 struct ip_options_rcu *opt; 1147 struct rtable *rt; 1148 1149 rt = *rtp; 1150 if (unlikely(!rt)) 1151 return -EFAULT; 1152 1153 /* 1154 * setup for corking. 1155 */ 1156 opt = ipc->opt; 1157 if (opt) { 1158 if (!cork->opt) { 1159 cork->opt = kmalloc(sizeof(struct ip_options) + 40, 1160 sk->sk_allocation); 1161 if (unlikely(!cork->opt)) 1162 return -ENOBUFS; 1163 } 1164 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 1165 cork->flags |= IPCORK_OPT; 1166 cork->addr = ipc->addr; 1167 } 1168 1169 /* 1170 * We steal reference to this route, caller should not release it 1171 */ 1172 *rtp = NULL; 1173 cork->fragsize = ip_sk_use_pmtu(sk) ? 1174 dst_mtu(&rt->dst) : rt->dst.dev->mtu; 1175 1176 cork->gso_size = ipc->gso_size; 1177 cork->dst = &rt->dst; 1178 cork->length = 0; 1179 cork->ttl = ipc->ttl; 1180 cork->tos = ipc->tos; 1181 cork->priority = ipc->priority; 1182 cork->transmit_time = ipc->sockc.transmit_time; 1183 cork->tx_flags = 0; 1184 sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags); 1185 1186 return 0; 1187 } 1188 1189 /* 1190 * ip_append_data() and ip_append_page() can make one large IP datagram 1191 * from many pieces of data. Each pieces will be holded on the socket 1192 * until ip_push_pending_frames() is called. Each piece can be a page 1193 * or non-page data. 1194 * 1195 * Not only UDP, other transport protocols - e.g. raw sockets - can use 1196 * this interface potentially. 1197 * 1198 * LATER: length must be adjusted by pad at tail, when it is required. 1199 */ 1200 int ip_append_data(struct sock *sk, struct flowi4 *fl4, 1201 int getfrag(void *from, char *to, int offset, int len, 1202 int odd, struct sk_buff *skb), 1203 void *from, int length, int transhdrlen, 1204 struct ipcm_cookie *ipc, struct rtable **rtp, 1205 unsigned int flags) 1206 { 1207 struct inet_sock *inet = inet_sk(sk); 1208 int err; 1209 1210 if (flags&MSG_PROBE) 1211 return 0; 1212 1213 if (skb_queue_empty(&sk->sk_write_queue)) { 1214 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 1215 if (err) 1216 return err; 1217 } else { 1218 transhdrlen = 0; 1219 } 1220 1221 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, 1222 sk_page_frag(sk), getfrag, 1223 from, length, transhdrlen, flags); 1224 } 1225 1226 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 1227 int offset, size_t size, int flags) 1228 { 1229 struct inet_sock *inet = inet_sk(sk); 1230 struct sk_buff *skb; 1231 struct rtable *rt; 1232 struct ip_options *opt = NULL; 1233 struct inet_cork *cork; 1234 int hh_len; 1235 int mtu; 1236 int len; 1237 int err; 1238 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; 1239 1240 if (inet->hdrincl) 1241 return -EPERM; 1242 1243 if (flags&MSG_PROBE) 1244 return 0; 1245 1246 if (skb_queue_empty(&sk->sk_write_queue)) 1247 return -EINVAL; 1248 1249 cork = &inet->cork.base; 1250 rt = (struct rtable *)cork->dst; 1251 if (cork->flags & IPCORK_OPT) 1252 opt = cork->opt; 1253 1254 if (!(rt->dst.dev->features&NETIF_F_SG)) 1255 return -EOPNOTSUPP; 1256 1257 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1258 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; 1259 1260 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1261 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1262 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 1263 1264 if (cork->length + size > maxnonfragsize - fragheaderlen) { 1265 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 1266 mtu - (opt ? opt->optlen : 0)); 1267 return -EMSGSIZE; 1268 } 1269 1270 skb = skb_peek_tail(&sk->sk_write_queue); 1271 if (!skb) 1272 return -EINVAL; 1273 1274 cork->length += size; 1275 1276 while (size > 0) { 1277 /* Check if the remaining data fits into current packet. */ 1278 len = mtu - skb->len; 1279 if (len < size) 1280 len = maxfraglen - skb->len; 1281 1282 if (len <= 0) { 1283 struct sk_buff *skb_prev; 1284 int alloclen; 1285 1286 skb_prev = skb; 1287 fraggap = skb_prev->len - maxfraglen; 1288 1289 alloclen = fragheaderlen + hh_len + fraggap + 15; 1290 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 1291 if (unlikely(!skb)) { 1292 err = -ENOBUFS; 1293 goto error; 1294 } 1295 1296 /* 1297 * Fill in the control structures 1298 */ 1299 skb->ip_summed = CHECKSUM_NONE; 1300 skb->csum = 0; 1301 skb_reserve(skb, hh_len); 1302 1303 /* 1304 * Find where to start putting bytes. 1305 */ 1306 skb_put(skb, fragheaderlen + fraggap); 1307 skb_reset_network_header(skb); 1308 skb->transport_header = (skb->network_header + 1309 fragheaderlen); 1310 if (fraggap) { 1311 skb->csum = skb_copy_and_csum_bits(skb_prev, 1312 maxfraglen, 1313 skb_transport_header(skb), 1314 fraggap, 0); 1315 skb_prev->csum = csum_sub(skb_prev->csum, 1316 skb->csum); 1317 pskb_trim_unique(skb_prev, maxfraglen); 1318 } 1319 1320 /* 1321 * Put the packet on the pending queue. 1322 */ 1323 __skb_queue_tail(&sk->sk_write_queue, skb); 1324 continue; 1325 } 1326 1327 if (len > size) 1328 len = size; 1329 1330 if (skb_append_pagefrags(skb, page, offset, len)) { 1331 err = -EMSGSIZE; 1332 goto error; 1333 } 1334 1335 if (skb->ip_summed == CHECKSUM_NONE) { 1336 __wsum csum; 1337 csum = csum_page(page, offset, len); 1338 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1339 } 1340 1341 skb->len += len; 1342 skb->data_len += len; 1343 skb->truesize += len; 1344 refcount_add(len, &sk->sk_wmem_alloc); 1345 offset += len; 1346 size -= len; 1347 } 1348 return 0; 1349 1350 error: 1351 cork->length -= size; 1352 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1353 return err; 1354 } 1355 1356 static void ip_cork_release(struct inet_cork *cork) 1357 { 1358 cork->flags &= ~IPCORK_OPT; 1359 kfree(cork->opt); 1360 cork->opt = NULL; 1361 dst_release(cork->dst); 1362 cork->dst = NULL; 1363 } 1364 1365 /* 1366 * Combined all pending IP fragments on the socket as one IP datagram 1367 * and push them out. 1368 */ 1369 struct sk_buff *__ip_make_skb(struct sock *sk, 1370 struct flowi4 *fl4, 1371 struct sk_buff_head *queue, 1372 struct inet_cork *cork) 1373 { 1374 struct sk_buff *skb, *tmp_skb; 1375 struct sk_buff **tail_skb; 1376 struct inet_sock *inet = inet_sk(sk); 1377 struct net *net = sock_net(sk); 1378 struct ip_options *opt = NULL; 1379 struct rtable *rt = (struct rtable *)cork->dst; 1380 struct iphdr *iph; 1381 __be16 df = 0; 1382 __u8 ttl; 1383 1384 skb = __skb_dequeue(queue); 1385 if (!skb) 1386 goto out; 1387 tail_skb = &(skb_shinfo(skb)->frag_list); 1388 1389 /* move skb->data to ip header from ext header */ 1390 if (skb->data < skb_network_header(skb)) 1391 __skb_pull(skb, skb_network_offset(skb)); 1392 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1393 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1394 *tail_skb = tmp_skb; 1395 tail_skb = &(tmp_skb->next); 1396 skb->len += tmp_skb->len; 1397 skb->data_len += tmp_skb->len; 1398 skb->truesize += tmp_skb->truesize; 1399 tmp_skb->destructor = NULL; 1400 tmp_skb->sk = NULL; 1401 } 1402 1403 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 1404 * to fragment the frame generated here. No matter, what transforms 1405 * how transforms change size of the packet, it will come out. 1406 */ 1407 skb->ignore_df = ip_sk_ignore_df(sk); 1408 1409 /* DF bit is set when we want to see DF on outgoing frames. 1410 * If ignore_df is set too, we still allow to fragment this frame 1411 * locally. */ 1412 if (inet->pmtudisc == IP_PMTUDISC_DO || 1413 inet->pmtudisc == IP_PMTUDISC_PROBE || 1414 (skb->len <= dst_mtu(&rt->dst) && 1415 ip_dont_fragment(sk, &rt->dst))) 1416 df = htons(IP_DF); 1417 1418 if (cork->flags & IPCORK_OPT) 1419 opt = cork->opt; 1420 1421 if (cork->ttl != 0) 1422 ttl = cork->ttl; 1423 else if (rt->rt_type == RTN_MULTICAST) 1424 ttl = inet->mc_ttl; 1425 else 1426 ttl = ip_select_ttl(inet, &rt->dst); 1427 1428 iph = ip_hdr(skb); 1429 iph->version = 4; 1430 iph->ihl = 5; 1431 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; 1432 iph->frag_off = df; 1433 iph->ttl = ttl; 1434 iph->protocol = sk->sk_protocol; 1435 ip_copy_addrs(iph, fl4); 1436 ip_select_ident(net, skb, sk); 1437 1438 if (opt) { 1439 iph->ihl += opt->optlen>>2; 1440 ip_options_build(skb, opt, cork->addr, rt, 0); 1441 } 1442 1443 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; 1444 skb->mark = sk->sk_mark; 1445 skb->tstamp = cork->transmit_time; 1446 /* 1447 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1448 * on dst refcount 1449 */ 1450 cork->dst = NULL; 1451 skb_dst_set(skb, &rt->dst); 1452 1453 if (iph->protocol == IPPROTO_ICMP) 1454 icmp_out_count(net, ((struct icmphdr *) 1455 skb_transport_header(skb))->type); 1456 1457 ip_cork_release(cork); 1458 out: 1459 return skb; 1460 } 1461 1462 int ip_send_skb(struct net *net, struct sk_buff *skb) 1463 { 1464 int err; 1465 1466 err = ip_local_out(net, skb->sk, skb); 1467 if (err) { 1468 if (err > 0) 1469 err = net_xmit_errno(err); 1470 if (err) 1471 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 1472 } 1473 1474 return err; 1475 } 1476 1477 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 1478 { 1479 struct sk_buff *skb; 1480 1481 skb = ip_finish_skb(sk, fl4); 1482 if (!skb) 1483 return 0; 1484 1485 /* Netfilter gets whole the not fragmented skb. */ 1486 return ip_send_skb(sock_net(sk), skb); 1487 } 1488 1489 /* 1490 * Throw away all pending data on the socket. 1491 */ 1492 static void __ip_flush_pending_frames(struct sock *sk, 1493 struct sk_buff_head *queue, 1494 struct inet_cork *cork) 1495 { 1496 struct sk_buff *skb; 1497 1498 while ((skb = __skb_dequeue_tail(queue)) != NULL) 1499 kfree_skb(skb); 1500 1501 ip_cork_release(cork); 1502 } 1503 1504 void ip_flush_pending_frames(struct sock *sk) 1505 { 1506 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 1507 } 1508 1509 struct sk_buff *ip_make_skb(struct sock *sk, 1510 struct flowi4 *fl4, 1511 int getfrag(void *from, char *to, int offset, 1512 int len, int odd, struct sk_buff *skb), 1513 void *from, int length, int transhdrlen, 1514 struct ipcm_cookie *ipc, struct rtable **rtp, 1515 struct inet_cork *cork, unsigned int flags) 1516 { 1517 struct sk_buff_head queue; 1518 int err; 1519 1520 if (flags & MSG_PROBE) 1521 return NULL; 1522 1523 __skb_queue_head_init(&queue); 1524 1525 cork->flags = 0; 1526 cork->addr = 0; 1527 cork->opt = NULL; 1528 err = ip_setup_cork(sk, cork, ipc, rtp); 1529 if (err) 1530 return ERR_PTR(err); 1531 1532 err = __ip_append_data(sk, fl4, &queue, cork, 1533 ¤t->task_frag, getfrag, 1534 from, length, transhdrlen, flags); 1535 if (err) { 1536 __ip_flush_pending_frames(sk, &queue, cork); 1537 return ERR_PTR(err); 1538 } 1539 1540 return __ip_make_skb(sk, fl4, &queue, cork); 1541 } 1542 1543 /* 1544 * Fetch data from kernel space and fill in checksum if needed. 1545 */ 1546 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 1547 int len, int odd, struct sk_buff *skb) 1548 { 1549 __wsum csum; 1550 1551 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); 1552 skb->csum = csum_block_add(skb->csum, csum, odd); 1553 return 0; 1554 } 1555 1556 /* 1557 * Generic function to send a packet as reply to another packet. 1558 * Used to send some TCP resets/acks so far. 1559 */ 1560 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, 1561 const struct ip_options *sopt, 1562 __be32 daddr, __be32 saddr, 1563 const struct ip_reply_arg *arg, 1564 unsigned int len) 1565 { 1566 struct ip_options_data replyopts; 1567 struct ipcm_cookie ipc; 1568 struct flowi4 fl4; 1569 struct rtable *rt = skb_rtable(skb); 1570 struct net *net = sock_net(sk); 1571 struct sk_buff *nskb; 1572 int err; 1573 int oif; 1574 1575 if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt)) 1576 return; 1577 1578 ipcm_init(&ipc); 1579 ipc.addr = daddr; 1580 1581 if (replyopts.opt.opt.optlen) { 1582 ipc.opt = &replyopts.opt; 1583 1584 if (replyopts.opt.opt.srr) 1585 daddr = replyopts.opt.opt.faddr; 1586 } 1587 1588 oif = arg->bound_dev_if; 1589 if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) 1590 oif = skb->skb_iif; 1591 1592 flowi4_init_output(&fl4, oif, 1593 IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark, 1594 RT_TOS(arg->tos), 1595 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, 1596 ip_reply_arg_flowi_flags(arg), 1597 daddr, saddr, 1598 tcp_hdr(skb)->source, tcp_hdr(skb)->dest, 1599 arg->uid); 1600 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 1601 rt = ip_route_output_key(net, &fl4); 1602 if (IS_ERR(rt)) 1603 return; 1604 1605 inet_sk(sk)->tos = arg->tos; 1606 1607 sk->sk_priority = skb->priority; 1608 sk->sk_protocol = ip_hdr(skb)->protocol; 1609 sk->sk_bound_dev_if = arg->bound_dev_if; 1610 sk->sk_sndbuf = sysctl_wmem_default; 1611 sk->sk_mark = fl4.flowi4_mark; 1612 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, 1613 len, 0, &ipc, &rt, MSG_DONTWAIT); 1614 if (unlikely(err)) { 1615 ip_flush_pending_frames(sk); 1616 goto out; 1617 } 1618 1619 nskb = skb_peek(&sk->sk_write_queue); 1620 if (nskb) { 1621 if (arg->csumoffset >= 0) 1622 *((__sum16 *)skb_transport_header(nskb) + 1623 arg->csumoffset) = csum_fold(csum_add(nskb->csum, 1624 arg->csum)); 1625 nskb->ip_summed = CHECKSUM_NONE; 1626 ip_push_pending_frames(sk, &fl4); 1627 } 1628 out: 1629 ip_rt_put(rt); 1630 } 1631 1632 void __init ip_init(void) 1633 { 1634 ip_rt_init(); 1635 inet_initpeers(); 1636 1637 #if defined(CONFIG_IP_MULTICAST) 1638 igmp_mc_init(); 1639 #endif 1640 } 1641