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 <net/lwtunnel.h> 77 #include <linux/igmp.h> 78 #include <linux/netfilter_ipv4.h> 79 #include <linux/netfilter_bridge.h> 80 #include <linux/netlink.h> 81 #include <linux/tcp.h> 82 83 static int 84 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 85 unsigned int mtu, 86 int (*output)(struct net *, struct sock *, struct sk_buff *)); 87 88 /* Generate a checksum for an outgoing IP datagram. */ 89 void ip_send_check(struct iphdr *iph) 90 { 91 iph->check = 0; 92 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 93 } 94 EXPORT_SYMBOL(ip_send_check); 95 96 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 97 { 98 struct iphdr *iph = ip_hdr(skb); 99 100 iph->tot_len = htons(skb->len); 101 ip_send_check(iph); 102 103 /* if egress device is enslaved to an L3 master device pass the 104 * skb to its handler for processing 105 */ 106 skb = l3mdev_ip_out(sk, skb); 107 if (unlikely(!skb)) 108 return 0; 109 110 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, 111 net, sk, skb, NULL, skb_dst(skb)->dev, 112 dst_output); 113 } 114 115 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 116 { 117 int err; 118 119 err = __ip_local_out(net, sk, skb); 120 if (likely(err == 1)) 121 err = dst_output(net, sk, skb); 122 123 return err; 124 } 125 EXPORT_SYMBOL_GPL(ip_local_out); 126 127 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) 128 { 129 int ttl = inet->uc_ttl; 130 131 if (ttl < 0) 132 ttl = ip4_dst_hoplimit(dst); 133 return ttl; 134 } 135 136 /* 137 * Add an ip header to a skbuff and send it out. 138 * 139 */ 140 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk, 141 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) 142 { 143 struct inet_sock *inet = inet_sk(sk); 144 struct rtable *rt = skb_rtable(skb); 145 struct net *net = sock_net(sk); 146 struct iphdr *iph; 147 148 /* Build the IP header. */ 149 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); 150 skb_reset_network_header(skb); 151 iph = ip_hdr(skb); 152 iph->version = 4; 153 iph->ihl = 5; 154 iph->tos = inet->tos; 155 iph->ttl = ip_select_ttl(inet, &rt->dst); 156 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); 157 iph->saddr = saddr; 158 iph->protocol = sk->sk_protocol; 159 if (ip_dont_fragment(sk, &rt->dst)) { 160 iph->frag_off = htons(IP_DF); 161 iph->id = 0; 162 } else { 163 iph->frag_off = 0; 164 __ip_select_ident(net, iph, 1); 165 } 166 167 if (opt && opt->opt.optlen) { 168 iph->ihl += opt->opt.optlen>>2; 169 ip_options_build(skb, &opt->opt, daddr, rt, 0); 170 } 171 172 skb->priority = sk->sk_priority; 173 skb->mark = sk->sk_mark; 174 175 /* Send it out. */ 176 return ip_local_out(net, skb->sk, skb); 177 } 178 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); 179 180 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) 181 { 182 struct dst_entry *dst = skb_dst(skb); 183 struct rtable *rt = (struct rtable *)dst; 184 struct net_device *dev = dst->dev; 185 unsigned int hh_len = LL_RESERVED_SPACE(dev); 186 struct neighbour *neigh; 187 u32 nexthop; 188 189 if (rt->rt_type == RTN_MULTICAST) { 190 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len); 191 } else if (rt->rt_type == RTN_BROADCAST) 192 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len); 193 194 /* Be paranoid, rather than too clever. */ 195 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 196 struct sk_buff *skb2; 197 198 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 199 if (!skb2) { 200 kfree_skb(skb); 201 return -ENOMEM; 202 } 203 if (skb->sk) 204 skb_set_owner_w(skb2, skb->sk); 205 consume_skb(skb); 206 skb = skb2; 207 } 208 209 if (lwtunnel_xmit_redirect(dst->lwtstate)) { 210 int res = lwtunnel_xmit(skb); 211 212 if (res < 0 || res == LWTUNNEL_XMIT_DONE) 213 return res; 214 } 215 216 rcu_read_lock_bh(); 217 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr); 218 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 219 if (unlikely(!neigh)) 220 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 221 if (!IS_ERR(neigh)) { 222 int res = dst_neigh_output(dst, neigh, skb); 223 224 rcu_read_unlock_bh(); 225 return res; 226 } 227 rcu_read_unlock_bh(); 228 229 net_dbg_ratelimited("%s: No header cache and no neighbour!\n", 230 __func__); 231 kfree_skb(skb); 232 return -EINVAL; 233 } 234 235 static int ip_finish_output_gso(struct net *net, struct sock *sk, 236 struct sk_buff *skb, unsigned int mtu) 237 { 238 netdev_features_t features; 239 struct sk_buff *segs; 240 int ret = 0; 241 242 /* common case: fragmentation of segments is not allowed, 243 * or seglen is <= mtu 244 */ 245 if (((IPCB(skb)->flags & IPSKB_FRAG_SEGS) == 0) || 246 skb_gso_validate_mtu(skb, mtu)) 247 return ip_finish_output2(net, sk, skb); 248 249 /* Slowpath - GSO segment length is exceeding the dst MTU. 250 * 251 * This can happen in two cases: 252 * 1) TCP GRO packet, DF bit not set 253 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly 254 * from host network stack. 255 */ 256 features = netif_skb_features(skb); 257 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET); 258 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); 259 if (IS_ERR_OR_NULL(segs)) { 260 kfree_skb(skb); 261 return -ENOMEM; 262 } 263 264 consume_skb(skb); 265 266 do { 267 struct sk_buff *nskb = segs->next; 268 int err; 269 270 segs->next = NULL; 271 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2); 272 273 if (err && ret == 0) 274 ret = err; 275 segs = nskb; 276 } while (segs); 277 278 return ret; 279 } 280 281 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 282 { 283 unsigned int mtu; 284 285 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) 286 /* Policy lookup after SNAT yielded a new policy */ 287 if (skb_dst(skb)->xfrm) { 288 IPCB(skb)->flags |= IPSKB_REROUTED; 289 return dst_output(net, sk, skb); 290 } 291 #endif 292 mtu = ip_skb_dst_mtu(sk, skb); 293 if (skb_is_gso(skb)) 294 return ip_finish_output_gso(net, sk, skb, mtu); 295 296 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU)) 297 return ip_fragment(net, sk, skb, mtu, ip_finish_output2); 298 299 return ip_finish_output2(net, sk, skb); 300 } 301 302 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb) 303 { 304 struct rtable *rt = skb_rtable(skb); 305 struct net_device *dev = rt->dst.dev; 306 307 /* 308 * If the indicated interface is up and running, send the packet. 309 */ 310 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 311 312 skb->dev = dev; 313 skb->protocol = htons(ETH_P_IP); 314 315 /* 316 * Multicasts are looped back for other local users 317 */ 318 319 if (rt->rt_flags&RTCF_MULTICAST) { 320 if (sk_mc_loop(sk) 321 #ifdef CONFIG_IP_MROUTE 322 /* Small optimization: do not loopback not local frames, 323 which returned after forwarding; they will be dropped 324 by ip_mr_input in any case. 325 Note, that local frames are looped back to be delivered 326 to local recipients. 327 328 This check is duplicated in ip_mr_input at the moment. 329 */ 330 && 331 ((rt->rt_flags & RTCF_LOCAL) || 332 !(IPCB(skb)->flags & IPSKB_FORWARDED)) 333 #endif 334 ) { 335 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 336 if (newskb) 337 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 338 net, sk, newskb, NULL, newskb->dev, 339 dev_loopback_xmit); 340 } 341 342 /* Multicasts with ttl 0 must not go beyond the host */ 343 344 if (ip_hdr(skb)->ttl == 0) { 345 kfree_skb(skb); 346 return 0; 347 } 348 } 349 350 if (rt->rt_flags&RTCF_BROADCAST) { 351 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 352 if (newskb) 353 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 354 net, sk, newskb, NULL, newskb->dev, 355 dev_loopback_xmit); 356 } 357 358 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 359 net, sk, skb, NULL, skb->dev, 360 ip_finish_output, 361 !(IPCB(skb)->flags & IPSKB_REROUTED)); 362 } 363 364 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb) 365 { 366 struct net_device *dev = skb_dst(skb)->dev; 367 368 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 369 370 skb->dev = dev; 371 skb->protocol = htons(ETH_P_IP); 372 373 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 374 net, sk, skb, NULL, dev, 375 ip_finish_output, 376 !(IPCB(skb)->flags & IPSKB_REROUTED)); 377 } 378 379 /* 380 * copy saddr and daddr, possibly using 64bit load/stores 381 * Equivalent to : 382 * iph->saddr = fl4->saddr; 383 * iph->daddr = fl4->daddr; 384 */ 385 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) 386 { 387 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != 388 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); 389 memcpy(&iph->saddr, &fl4->saddr, 390 sizeof(fl4->saddr) + sizeof(fl4->daddr)); 391 } 392 393 /* Note: skb->sk can be different from sk, in case of tunnels */ 394 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) 395 { 396 struct inet_sock *inet = inet_sk(sk); 397 struct net *net = sock_net(sk); 398 struct ip_options_rcu *inet_opt; 399 struct flowi4 *fl4; 400 struct rtable *rt; 401 struct iphdr *iph; 402 int res; 403 404 /* Skip all of this if the packet is already routed, 405 * f.e. by something like SCTP. 406 */ 407 rcu_read_lock(); 408 inet_opt = rcu_dereference(inet->inet_opt); 409 fl4 = &fl->u.ip4; 410 rt = skb_rtable(skb); 411 if (rt) 412 goto packet_routed; 413 414 /* Make sure we can route this packet. */ 415 rt = (struct rtable *)__sk_dst_check(sk, 0); 416 if (!rt) { 417 __be32 daddr; 418 419 /* Use correct destination address if we have options. */ 420 daddr = inet->inet_daddr; 421 if (inet_opt && inet_opt->opt.srr) 422 daddr = inet_opt->opt.faddr; 423 424 /* If this fails, retransmit mechanism of transport layer will 425 * keep trying until route appears or the connection times 426 * itself out. 427 */ 428 rt = ip_route_output_ports(net, fl4, sk, 429 daddr, inet->inet_saddr, 430 inet->inet_dport, 431 inet->inet_sport, 432 sk->sk_protocol, 433 RT_CONN_FLAGS(sk), 434 sk->sk_bound_dev_if); 435 if (IS_ERR(rt)) 436 goto no_route; 437 sk_setup_caps(sk, &rt->dst); 438 } 439 skb_dst_set_noref(skb, &rt->dst); 440 441 packet_routed: 442 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) 443 goto no_route; 444 445 /* OK, we know where to send it, allocate and build IP header. */ 446 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 447 skb_reset_network_header(skb); 448 iph = ip_hdr(skb); 449 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); 450 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) 451 iph->frag_off = htons(IP_DF); 452 else 453 iph->frag_off = 0; 454 iph->ttl = ip_select_ttl(inet, &rt->dst); 455 iph->protocol = sk->sk_protocol; 456 ip_copy_addrs(iph, fl4); 457 458 /* Transport layer set skb->h.foo itself. */ 459 460 if (inet_opt && inet_opt->opt.optlen) { 461 iph->ihl += inet_opt->opt.optlen >> 2; 462 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 463 } 464 465 ip_select_ident_segs(net, skb, sk, 466 skb_shinfo(skb)->gso_segs ?: 1); 467 468 /* TODO : should we use skb->sk here instead of sk ? */ 469 skb->priority = sk->sk_priority; 470 skb->mark = sk->sk_mark; 471 472 res = ip_local_out(net, sk, skb); 473 rcu_read_unlock(); 474 return res; 475 476 no_route: 477 rcu_read_unlock(); 478 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 479 kfree_skb(skb); 480 return -EHOSTUNREACH; 481 } 482 EXPORT_SYMBOL(ip_queue_xmit); 483 484 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) 485 { 486 to->pkt_type = from->pkt_type; 487 to->priority = from->priority; 488 to->protocol = from->protocol; 489 skb_dst_drop(to); 490 skb_dst_copy(to, from); 491 to->dev = from->dev; 492 to->mark = from->mark; 493 494 /* Copy the flags to each fragment. */ 495 IPCB(to)->flags = IPCB(from)->flags; 496 497 #ifdef CONFIG_NET_SCHED 498 to->tc_index = from->tc_index; 499 #endif 500 nf_copy(to, from); 501 #if IS_ENABLED(CONFIG_IP_VS) 502 to->ipvs_property = from->ipvs_property; 503 #endif 504 skb_copy_secmark(to, from); 505 } 506 507 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 508 unsigned int mtu, 509 int (*output)(struct net *, struct sock *, struct sk_buff *)) 510 { 511 struct iphdr *iph = ip_hdr(skb); 512 513 if ((iph->frag_off & htons(IP_DF)) == 0) 514 return ip_do_fragment(net, sk, skb, output); 515 516 if (unlikely(!skb->ignore_df || 517 (IPCB(skb)->frag_max_size && 518 IPCB(skb)->frag_max_size > mtu))) { 519 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 520 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 521 htonl(mtu)); 522 kfree_skb(skb); 523 return -EMSGSIZE; 524 } 525 526 return ip_do_fragment(net, sk, skb, output); 527 } 528 529 /* 530 * This IP datagram is too large to be sent in one piece. Break it up into 531 * smaller pieces (each of size equal to IP header plus 532 * a block of the data of the original IP data part) that will yet fit in a 533 * single device frame, and queue such a frame for sending. 534 */ 535 536 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 537 int (*output)(struct net *, struct sock *, struct sk_buff *)) 538 { 539 struct iphdr *iph; 540 int ptr; 541 struct sk_buff *skb2; 542 unsigned int mtu, hlen, left, len, ll_rs; 543 int offset; 544 __be16 not_last_frag; 545 struct rtable *rt = skb_rtable(skb); 546 int err = 0; 547 548 /* for offloaded checksums cleanup checksum before fragmentation */ 549 if (skb->ip_summed == CHECKSUM_PARTIAL && 550 (err = skb_checksum_help(skb))) 551 goto fail; 552 553 /* 554 * Point into the IP datagram header. 555 */ 556 557 iph = ip_hdr(skb); 558 559 mtu = ip_skb_dst_mtu(sk, skb); 560 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) 561 mtu = IPCB(skb)->frag_max_size; 562 563 /* 564 * Setup starting values. 565 */ 566 567 hlen = iph->ihl * 4; 568 mtu = mtu - hlen; /* Size of data space */ 569 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 570 571 /* When frag_list is given, use it. First, check its validity: 572 * some transformers could create wrong frag_list or break existing 573 * one, it is not prohibited. In this case fall back to copying. 574 * 575 * LATER: this step can be merged to real generation of fragments, 576 * we can switch to copy when see the first bad fragment. 577 */ 578 if (skb_has_frag_list(skb)) { 579 struct sk_buff *frag, *frag2; 580 int first_len = skb_pagelen(skb); 581 582 if (first_len - hlen > mtu || 583 ((first_len - hlen) & 7) || 584 ip_is_fragment(iph) || 585 skb_cloned(skb)) 586 goto slow_path; 587 588 skb_walk_frags(skb, frag) { 589 /* Correct geometry. */ 590 if (frag->len > mtu || 591 ((frag->len & 7) && frag->next) || 592 skb_headroom(frag) < hlen) 593 goto slow_path_clean; 594 595 /* Partially cloned skb? */ 596 if (skb_shared(frag)) 597 goto slow_path_clean; 598 599 BUG_ON(frag->sk); 600 if (skb->sk) { 601 frag->sk = skb->sk; 602 frag->destructor = sock_wfree; 603 } 604 skb->truesize -= frag->truesize; 605 } 606 607 /* Everything is OK. Generate! */ 608 609 err = 0; 610 offset = 0; 611 frag = skb_shinfo(skb)->frag_list; 612 skb_frag_list_init(skb); 613 skb->data_len = first_len - skb_headlen(skb); 614 skb->len = first_len; 615 iph->tot_len = htons(first_len); 616 iph->frag_off = htons(IP_MF); 617 ip_send_check(iph); 618 619 for (;;) { 620 /* Prepare header of the next frame, 621 * before previous one went down. */ 622 if (frag) { 623 frag->ip_summed = CHECKSUM_NONE; 624 skb_reset_transport_header(frag); 625 __skb_push(frag, hlen); 626 skb_reset_network_header(frag); 627 memcpy(skb_network_header(frag), iph, hlen); 628 iph = ip_hdr(frag); 629 iph->tot_len = htons(frag->len); 630 ip_copy_metadata(frag, skb); 631 if (offset == 0) 632 ip_options_fragment(frag); 633 offset += skb->len - hlen; 634 iph->frag_off = htons(offset>>3); 635 if (frag->next) 636 iph->frag_off |= htons(IP_MF); 637 /* Ready, complete checksum */ 638 ip_send_check(iph); 639 } 640 641 err = output(net, sk, skb); 642 643 if (!err) 644 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 645 if (err || !frag) 646 break; 647 648 skb = frag; 649 frag = skb->next; 650 skb->next = NULL; 651 } 652 653 if (err == 0) { 654 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 655 return 0; 656 } 657 658 while (frag) { 659 skb = frag->next; 660 kfree_skb(frag); 661 frag = skb; 662 } 663 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 664 return err; 665 666 slow_path_clean: 667 skb_walk_frags(skb, frag2) { 668 if (frag2 == frag) 669 break; 670 frag2->sk = NULL; 671 frag2->destructor = NULL; 672 skb->truesize += frag2->truesize; 673 } 674 } 675 676 slow_path: 677 iph = ip_hdr(skb); 678 679 left = skb->len - hlen; /* Space per frame */ 680 ptr = hlen; /* Where to start from */ 681 682 ll_rs = LL_RESERVED_SPACE(rt->dst.dev); 683 684 /* 685 * Fragment the datagram. 686 */ 687 688 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 689 not_last_frag = iph->frag_off & htons(IP_MF); 690 691 /* 692 * Keep copying data until we run out. 693 */ 694 695 while (left > 0) { 696 len = left; 697 /* IF: it doesn't fit, use 'mtu' - the data space left */ 698 if (len > mtu) 699 len = mtu; 700 /* IF: we are not sending up to and including the packet end 701 then align the next start on an eight byte boundary */ 702 if (len < left) { 703 len &= ~7; 704 } 705 706 /* Allocate buffer */ 707 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC); 708 if (!skb2) { 709 err = -ENOMEM; 710 goto fail; 711 } 712 713 /* 714 * Set up data on packet 715 */ 716 717 ip_copy_metadata(skb2, skb); 718 skb_reserve(skb2, ll_rs); 719 skb_put(skb2, len + hlen); 720 skb_reset_network_header(skb2); 721 skb2->transport_header = skb2->network_header + hlen; 722 723 /* 724 * Charge the memory for the fragment to any owner 725 * it might possess 726 */ 727 728 if (skb->sk) 729 skb_set_owner_w(skb2, skb->sk); 730 731 /* 732 * Copy the packet header into the new buffer. 733 */ 734 735 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); 736 737 /* 738 * Copy a block of the IP datagram. 739 */ 740 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) 741 BUG(); 742 left -= len; 743 744 /* 745 * Fill in the new header fields. 746 */ 747 iph = ip_hdr(skb2); 748 iph->frag_off = htons((offset >> 3)); 749 750 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU) 751 iph->frag_off |= htons(IP_DF); 752 753 /* ANK: dirty, but effective trick. Upgrade options only if 754 * the segment to be fragmented was THE FIRST (otherwise, 755 * options are already fixed) and make it ONCE 756 * on the initial skb, so that all the following fragments 757 * will inherit fixed options. 758 */ 759 if (offset == 0) 760 ip_options_fragment(skb); 761 762 /* 763 * Added AC : If we are fragmenting a fragment that's not the 764 * last fragment then keep MF on each bit 765 */ 766 if (left > 0 || not_last_frag) 767 iph->frag_off |= htons(IP_MF); 768 ptr += len; 769 offset += len; 770 771 /* 772 * Put this fragment into the sending queue. 773 */ 774 iph->tot_len = htons(len + hlen); 775 776 ip_send_check(iph); 777 778 err = output(net, sk, skb2); 779 if (err) 780 goto fail; 781 782 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 783 } 784 consume_skb(skb); 785 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 786 return err; 787 788 fail: 789 kfree_skb(skb); 790 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 791 return err; 792 } 793 EXPORT_SYMBOL(ip_do_fragment); 794 795 int 796 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 797 { 798 struct msghdr *msg = from; 799 800 if (skb->ip_summed == CHECKSUM_PARTIAL) { 801 if (copy_from_iter(to, len, &msg->msg_iter) != len) 802 return -EFAULT; 803 } else { 804 __wsum csum = 0; 805 if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len) 806 return -EFAULT; 807 skb->csum = csum_block_add(skb->csum, csum, odd); 808 } 809 return 0; 810 } 811 EXPORT_SYMBOL(ip_generic_getfrag); 812 813 static inline __wsum 814 csum_page(struct page *page, int offset, int copy) 815 { 816 char *kaddr; 817 __wsum csum; 818 kaddr = kmap(page); 819 csum = csum_partial(kaddr + offset, copy, 0); 820 kunmap(page); 821 return csum; 822 } 823 824 static inline int ip_ufo_append_data(struct sock *sk, 825 struct sk_buff_head *queue, 826 int getfrag(void *from, char *to, int offset, int len, 827 int odd, struct sk_buff *skb), 828 void *from, int length, int hh_len, int fragheaderlen, 829 int transhdrlen, int maxfraglen, unsigned int flags) 830 { 831 struct sk_buff *skb; 832 int err; 833 834 /* There is support for UDP fragmentation offload by network 835 * device, so create one single skb packet containing complete 836 * udp datagram 837 */ 838 skb = skb_peek_tail(queue); 839 if (!skb) { 840 skb = sock_alloc_send_skb(sk, 841 hh_len + fragheaderlen + transhdrlen + 20, 842 (flags & MSG_DONTWAIT), &err); 843 844 if (!skb) 845 return err; 846 847 /* reserve space for Hardware header */ 848 skb_reserve(skb, hh_len); 849 850 /* create space for UDP/IP header */ 851 skb_put(skb, fragheaderlen + transhdrlen); 852 853 /* initialize network header pointer */ 854 skb_reset_network_header(skb); 855 856 /* initialize protocol header pointer */ 857 skb->transport_header = skb->network_header + fragheaderlen; 858 859 skb->csum = 0; 860 861 __skb_queue_tail(queue, skb); 862 } else if (skb_is_gso(skb)) { 863 goto append; 864 } 865 866 skb->ip_summed = CHECKSUM_PARTIAL; 867 /* specify the length of each IP datagram fragment */ 868 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen; 869 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 870 871 append: 872 return skb_append_datato_frags(sk, skb, getfrag, from, 873 (length - transhdrlen)); 874 } 875 876 static int __ip_append_data(struct sock *sk, 877 struct flowi4 *fl4, 878 struct sk_buff_head *queue, 879 struct inet_cork *cork, 880 struct page_frag *pfrag, 881 int getfrag(void *from, char *to, int offset, 882 int len, int odd, struct sk_buff *skb), 883 void *from, int length, int transhdrlen, 884 unsigned int flags) 885 { 886 struct inet_sock *inet = inet_sk(sk); 887 struct sk_buff *skb; 888 889 struct ip_options *opt = cork->opt; 890 int hh_len; 891 int exthdrlen; 892 int mtu; 893 int copy; 894 int err; 895 int offset = 0; 896 unsigned int maxfraglen, fragheaderlen, maxnonfragsize; 897 int csummode = CHECKSUM_NONE; 898 struct rtable *rt = (struct rtable *)cork->dst; 899 u32 tskey = 0; 900 901 skb = skb_peek_tail(queue); 902 903 exthdrlen = !skb ? rt->dst.header_len : 0; 904 mtu = cork->fragsize; 905 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && 906 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) 907 tskey = sk->sk_tskey++; 908 909 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 910 911 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 912 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 913 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 914 915 if (cork->length + length > maxnonfragsize - fragheaderlen) { 916 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 917 mtu - (opt ? opt->optlen : 0)); 918 return -EMSGSIZE; 919 } 920 921 /* 922 * transhdrlen > 0 means that this is the first fragment and we wish 923 * it won't be fragmented in the future. 924 */ 925 if (transhdrlen && 926 length + fragheaderlen <= mtu && 927 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) && 928 !(flags & MSG_MORE) && 929 !exthdrlen) 930 csummode = CHECKSUM_PARTIAL; 931 932 cork->length += length; 933 if (((length > mtu) || (skb && skb_is_gso(skb))) && 934 (sk->sk_protocol == IPPROTO_UDP) && 935 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len && 936 (sk->sk_type == SOCK_DGRAM) && !sk->sk_no_check_tx) { 937 err = ip_ufo_append_data(sk, queue, getfrag, from, length, 938 hh_len, fragheaderlen, transhdrlen, 939 maxfraglen, flags); 940 if (err) 941 goto error; 942 return 0; 943 } 944 945 /* So, what's going on in the loop below? 946 * 947 * We use calculated fragment length to generate chained skb, 948 * each of segments is IP fragment ready for sending to network after 949 * adding appropriate IP header. 950 */ 951 952 if (!skb) 953 goto alloc_new_skb; 954 955 while (length > 0) { 956 /* Check if the remaining data fits into current packet. */ 957 copy = mtu - skb->len; 958 if (copy < length) 959 copy = maxfraglen - skb->len; 960 if (copy <= 0) { 961 char *data; 962 unsigned int datalen; 963 unsigned int fraglen; 964 unsigned int fraggap; 965 unsigned int alloclen; 966 struct sk_buff *skb_prev; 967 alloc_new_skb: 968 skb_prev = skb; 969 if (skb_prev) 970 fraggap = skb_prev->len - maxfraglen; 971 else 972 fraggap = 0; 973 974 /* 975 * If remaining data exceeds the mtu, 976 * we know we need more fragment(s). 977 */ 978 datalen = length + fraggap; 979 if (datalen > mtu - fragheaderlen) 980 datalen = maxfraglen - fragheaderlen; 981 fraglen = datalen + fragheaderlen; 982 983 if ((flags & MSG_MORE) && 984 !(rt->dst.dev->features&NETIF_F_SG)) 985 alloclen = mtu; 986 else 987 alloclen = fraglen; 988 989 alloclen += exthdrlen; 990 991 /* The last fragment gets additional space at tail. 992 * Note, with MSG_MORE we overallocate on fragments, 993 * because we have no idea what fragment will be 994 * the last. 995 */ 996 if (datalen == length + fraggap) 997 alloclen += rt->dst.trailer_len; 998 999 if (transhdrlen) { 1000 skb = sock_alloc_send_skb(sk, 1001 alloclen + hh_len + 15, 1002 (flags & MSG_DONTWAIT), &err); 1003 } else { 1004 skb = NULL; 1005 if (atomic_read(&sk->sk_wmem_alloc) <= 1006 2 * sk->sk_sndbuf) 1007 skb = sock_wmalloc(sk, 1008 alloclen + hh_len + 15, 1, 1009 sk->sk_allocation); 1010 if (unlikely(!skb)) 1011 err = -ENOBUFS; 1012 } 1013 if (!skb) 1014 goto error; 1015 1016 /* 1017 * Fill in the control structures 1018 */ 1019 skb->ip_summed = csummode; 1020 skb->csum = 0; 1021 skb_reserve(skb, hh_len); 1022 1023 /* only the initial fragment is time stamped */ 1024 skb_shinfo(skb)->tx_flags = cork->tx_flags; 1025 cork->tx_flags = 0; 1026 skb_shinfo(skb)->tskey = tskey; 1027 tskey = 0; 1028 1029 /* 1030 * Find where to start putting bytes. 1031 */ 1032 data = skb_put(skb, fraglen + exthdrlen); 1033 skb_set_network_header(skb, exthdrlen); 1034 skb->transport_header = (skb->network_header + 1035 fragheaderlen); 1036 data += fragheaderlen + exthdrlen; 1037 1038 if (fraggap) { 1039 skb->csum = skb_copy_and_csum_bits( 1040 skb_prev, maxfraglen, 1041 data + transhdrlen, fraggap, 0); 1042 skb_prev->csum = csum_sub(skb_prev->csum, 1043 skb->csum); 1044 data += fraggap; 1045 pskb_trim_unique(skb_prev, maxfraglen); 1046 } 1047 1048 copy = datalen - transhdrlen - fraggap; 1049 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 1050 err = -EFAULT; 1051 kfree_skb(skb); 1052 goto error; 1053 } 1054 1055 offset += copy; 1056 length -= datalen - fraggap; 1057 transhdrlen = 0; 1058 exthdrlen = 0; 1059 csummode = CHECKSUM_NONE; 1060 1061 /* 1062 * Put the packet on the pending queue. 1063 */ 1064 __skb_queue_tail(queue, skb); 1065 continue; 1066 } 1067 1068 if (copy > length) 1069 copy = length; 1070 1071 if (!(rt->dst.dev->features&NETIF_F_SG)) { 1072 unsigned int off; 1073 1074 off = skb->len; 1075 if (getfrag(from, skb_put(skb, copy), 1076 offset, copy, off, skb) < 0) { 1077 __skb_trim(skb, off); 1078 err = -EFAULT; 1079 goto error; 1080 } 1081 } else { 1082 int i = skb_shinfo(skb)->nr_frags; 1083 1084 err = -ENOMEM; 1085 if (!sk_page_frag_refill(sk, pfrag)) 1086 goto error; 1087 1088 if (!skb_can_coalesce(skb, i, pfrag->page, 1089 pfrag->offset)) { 1090 err = -EMSGSIZE; 1091 if (i == MAX_SKB_FRAGS) 1092 goto error; 1093 1094 __skb_fill_page_desc(skb, i, pfrag->page, 1095 pfrag->offset, 0); 1096 skb_shinfo(skb)->nr_frags = ++i; 1097 get_page(pfrag->page); 1098 } 1099 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1100 if (getfrag(from, 1101 page_address(pfrag->page) + pfrag->offset, 1102 offset, copy, skb->len, skb) < 0) 1103 goto error_efault; 1104 1105 pfrag->offset += copy; 1106 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1107 skb->len += copy; 1108 skb->data_len += copy; 1109 skb->truesize += copy; 1110 atomic_add(copy, &sk->sk_wmem_alloc); 1111 } 1112 offset += copy; 1113 length -= copy; 1114 } 1115 1116 return 0; 1117 1118 error_efault: 1119 err = -EFAULT; 1120 error: 1121 cork->length -= length; 1122 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1123 return err; 1124 } 1125 1126 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 1127 struct ipcm_cookie *ipc, struct rtable **rtp) 1128 { 1129 struct ip_options_rcu *opt; 1130 struct rtable *rt; 1131 1132 /* 1133 * setup for corking. 1134 */ 1135 opt = ipc->opt; 1136 if (opt) { 1137 if (!cork->opt) { 1138 cork->opt = kmalloc(sizeof(struct ip_options) + 40, 1139 sk->sk_allocation); 1140 if (unlikely(!cork->opt)) 1141 return -ENOBUFS; 1142 } 1143 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 1144 cork->flags |= IPCORK_OPT; 1145 cork->addr = ipc->addr; 1146 } 1147 rt = *rtp; 1148 if (unlikely(!rt)) 1149 return -EFAULT; 1150 /* 1151 * We steal reference to this route, caller should not release it 1152 */ 1153 *rtp = NULL; 1154 cork->fragsize = ip_sk_use_pmtu(sk) ? 1155 dst_mtu(&rt->dst) : rt->dst.dev->mtu; 1156 cork->dst = &rt->dst; 1157 cork->length = 0; 1158 cork->ttl = ipc->ttl; 1159 cork->tos = ipc->tos; 1160 cork->priority = ipc->priority; 1161 cork->tx_flags = ipc->tx_flags; 1162 1163 return 0; 1164 } 1165 1166 /* 1167 * ip_append_data() and ip_append_page() can make one large IP datagram 1168 * from many pieces of data. Each pieces will be holded on the socket 1169 * until ip_push_pending_frames() is called. Each piece can be a page 1170 * or non-page data. 1171 * 1172 * Not only UDP, other transport protocols - e.g. raw sockets - can use 1173 * this interface potentially. 1174 * 1175 * LATER: length must be adjusted by pad at tail, when it is required. 1176 */ 1177 int ip_append_data(struct sock *sk, struct flowi4 *fl4, 1178 int getfrag(void *from, char *to, int offset, int len, 1179 int odd, struct sk_buff *skb), 1180 void *from, int length, int transhdrlen, 1181 struct ipcm_cookie *ipc, struct rtable **rtp, 1182 unsigned int flags) 1183 { 1184 struct inet_sock *inet = inet_sk(sk); 1185 int err; 1186 1187 if (flags&MSG_PROBE) 1188 return 0; 1189 1190 if (skb_queue_empty(&sk->sk_write_queue)) { 1191 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 1192 if (err) 1193 return err; 1194 } else { 1195 transhdrlen = 0; 1196 } 1197 1198 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, 1199 sk_page_frag(sk), getfrag, 1200 from, length, transhdrlen, flags); 1201 } 1202 1203 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 1204 int offset, size_t size, int flags) 1205 { 1206 struct inet_sock *inet = inet_sk(sk); 1207 struct sk_buff *skb; 1208 struct rtable *rt; 1209 struct ip_options *opt = NULL; 1210 struct inet_cork *cork; 1211 int hh_len; 1212 int mtu; 1213 int len; 1214 int err; 1215 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; 1216 1217 if (inet->hdrincl) 1218 return -EPERM; 1219 1220 if (flags&MSG_PROBE) 1221 return 0; 1222 1223 if (skb_queue_empty(&sk->sk_write_queue)) 1224 return -EINVAL; 1225 1226 cork = &inet->cork.base; 1227 rt = (struct rtable *)cork->dst; 1228 if (cork->flags & IPCORK_OPT) 1229 opt = cork->opt; 1230 1231 if (!(rt->dst.dev->features&NETIF_F_SG)) 1232 return -EOPNOTSUPP; 1233 1234 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1235 mtu = cork->fragsize; 1236 1237 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1238 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1239 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 1240 1241 if (cork->length + size > maxnonfragsize - fragheaderlen) { 1242 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 1243 mtu - (opt ? opt->optlen : 0)); 1244 return -EMSGSIZE; 1245 } 1246 1247 skb = skb_peek_tail(&sk->sk_write_queue); 1248 if (!skb) 1249 return -EINVAL; 1250 1251 if ((size + skb->len > mtu) && 1252 (sk->sk_protocol == IPPROTO_UDP) && 1253 (rt->dst.dev->features & NETIF_F_UFO)) { 1254 if (skb->ip_summed != CHECKSUM_PARTIAL) 1255 return -EOPNOTSUPP; 1256 1257 skb_shinfo(skb)->gso_size = mtu - fragheaderlen; 1258 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 1259 } 1260 cork->length += size; 1261 1262 while (size > 0) { 1263 if (skb_is_gso(skb)) { 1264 len = size; 1265 } else { 1266 1267 /* Check if the remaining data fits into current packet. */ 1268 len = mtu - skb->len; 1269 if (len < size) 1270 len = maxfraglen - skb->len; 1271 } 1272 if (len <= 0) { 1273 struct sk_buff *skb_prev; 1274 int alloclen; 1275 1276 skb_prev = skb; 1277 fraggap = skb_prev->len - maxfraglen; 1278 1279 alloclen = fragheaderlen + hh_len + fraggap + 15; 1280 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 1281 if (unlikely(!skb)) { 1282 err = -ENOBUFS; 1283 goto error; 1284 } 1285 1286 /* 1287 * Fill in the control structures 1288 */ 1289 skb->ip_summed = CHECKSUM_NONE; 1290 skb->csum = 0; 1291 skb_reserve(skb, hh_len); 1292 1293 /* 1294 * Find where to start putting bytes. 1295 */ 1296 skb_put(skb, fragheaderlen + fraggap); 1297 skb_reset_network_header(skb); 1298 skb->transport_header = (skb->network_header + 1299 fragheaderlen); 1300 if (fraggap) { 1301 skb->csum = skb_copy_and_csum_bits(skb_prev, 1302 maxfraglen, 1303 skb_transport_header(skb), 1304 fraggap, 0); 1305 skb_prev->csum = csum_sub(skb_prev->csum, 1306 skb->csum); 1307 pskb_trim_unique(skb_prev, maxfraglen); 1308 } 1309 1310 /* 1311 * Put the packet on the pending queue. 1312 */ 1313 __skb_queue_tail(&sk->sk_write_queue, skb); 1314 continue; 1315 } 1316 1317 if (len > size) 1318 len = size; 1319 1320 if (skb_append_pagefrags(skb, page, offset, len)) { 1321 err = -EMSGSIZE; 1322 goto error; 1323 } 1324 1325 if (skb->ip_summed == CHECKSUM_NONE) { 1326 __wsum csum; 1327 csum = csum_page(page, offset, len); 1328 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1329 } 1330 1331 skb->len += len; 1332 skb->data_len += len; 1333 skb->truesize += len; 1334 atomic_add(len, &sk->sk_wmem_alloc); 1335 offset += len; 1336 size -= len; 1337 } 1338 return 0; 1339 1340 error: 1341 cork->length -= size; 1342 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1343 return err; 1344 } 1345 1346 static void ip_cork_release(struct inet_cork *cork) 1347 { 1348 cork->flags &= ~IPCORK_OPT; 1349 kfree(cork->opt); 1350 cork->opt = NULL; 1351 dst_release(cork->dst); 1352 cork->dst = NULL; 1353 } 1354 1355 /* 1356 * Combined all pending IP fragments on the socket as one IP datagram 1357 * and push them out. 1358 */ 1359 struct sk_buff *__ip_make_skb(struct sock *sk, 1360 struct flowi4 *fl4, 1361 struct sk_buff_head *queue, 1362 struct inet_cork *cork) 1363 { 1364 struct sk_buff *skb, *tmp_skb; 1365 struct sk_buff **tail_skb; 1366 struct inet_sock *inet = inet_sk(sk); 1367 struct net *net = sock_net(sk); 1368 struct ip_options *opt = NULL; 1369 struct rtable *rt = (struct rtable *)cork->dst; 1370 struct iphdr *iph; 1371 __be16 df = 0; 1372 __u8 ttl; 1373 1374 skb = __skb_dequeue(queue); 1375 if (!skb) 1376 goto out; 1377 tail_skb = &(skb_shinfo(skb)->frag_list); 1378 1379 /* move skb->data to ip header from ext header */ 1380 if (skb->data < skb_network_header(skb)) 1381 __skb_pull(skb, skb_network_offset(skb)); 1382 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1383 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1384 *tail_skb = tmp_skb; 1385 tail_skb = &(tmp_skb->next); 1386 skb->len += tmp_skb->len; 1387 skb->data_len += tmp_skb->len; 1388 skb->truesize += tmp_skb->truesize; 1389 tmp_skb->destructor = NULL; 1390 tmp_skb->sk = NULL; 1391 } 1392 1393 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 1394 * to fragment the frame generated here. No matter, what transforms 1395 * how transforms change size of the packet, it will come out. 1396 */ 1397 skb->ignore_df = ip_sk_ignore_df(sk); 1398 1399 /* DF bit is set when we want to see DF on outgoing frames. 1400 * If ignore_df is set too, we still allow to fragment this frame 1401 * locally. */ 1402 if (inet->pmtudisc == IP_PMTUDISC_DO || 1403 inet->pmtudisc == IP_PMTUDISC_PROBE || 1404 (skb->len <= dst_mtu(&rt->dst) && 1405 ip_dont_fragment(sk, &rt->dst))) 1406 df = htons(IP_DF); 1407 1408 if (cork->flags & IPCORK_OPT) 1409 opt = cork->opt; 1410 1411 if (cork->ttl != 0) 1412 ttl = cork->ttl; 1413 else if (rt->rt_type == RTN_MULTICAST) 1414 ttl = inet->mc_ttl; 1415 else 1416 ttl = ip_select_ttl(inet, &rt->dst); 1417 1418 iph = ip_hdr(skb); 1419 iph->version = 4; 1420 iph->ihl = 5; 1421 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; 1422 iph->frag_off = df; 1423 iph->ttl = ttl; 1424 iph->protocol = sk->sk_protocol; 1425 ip_copy_addrs(iph, fl4); 1426 ip_select_ident(net, skb, sk); 1427 1428 if (opt) { 1429 iph->ihl += opt->optlen>>2; 1430 ip_options_build(skb, opt, cork->addr, rt, 0); 1431 } 1432 1433 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; 1434 skb->mark = sk->sk_mark; 1435 /* 1436 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1437 * on dst refcount 1438 */ 1439 cork->dst = NULL; 1440 skb_dst_set(skb, &rt->dst); 1441 1442 if (iph->protocol == IPPROTO_ICMP) 1443 icmp_out_count(net, ((struct icmphdr *) 1444 skb_transport_header(skb))->type); 1445 1446 ip_cork_release(cork); 1447 out: 1448 return skb; 1449 } 1450 1451 int ip_send_skb(struct net *net, struct sk_buff *skb) 1452 { 1453 int err; 1454 1455 err = ip_local_out(net, skb->sk, skb); 1456 if (err) { 1457 if (err > 0) 1458 err = net_xmit_errno(err); 1459 if (err) 1460 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 1461 } 1462 1463 return err; 1464 } 1465 1466 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 1467 { 1468 struct sk_buff *skb; 1469 1470 skb = ip_finish_skb(sk, fl4); 1471 if (!skb) 1472 return 0; 1473 1474 /* Netfilter gets whole the not fragmented skb. */ 1475 return ip_send_skb(sock_net(sk), skb); 1476 } 1477 1478 /* 1479 * Throw away all pending data on the socket. 1480 */ 1481 static void __ip_flush_pending_frames(struct sock *sk, 1482 struct sk_buff_head *queue, 1483 struct inet_cork *cork) 1484 { 1485 struct sk_buff *skb; 1486 1487 while ((skb = __skb_dequeue_tail(queue)) != NULL) 1488 kfree_skb(skb); 1489 1490 ip_cork_release(cork); 1491 } 1492 1493 void ip_flush_pending_frames(struct sock *sk) 1494 { 1495 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 1496 } 1497 1498 struct sk_buff *ip_make_skb(struct sock *sk, 1499 struct flowi4 *fl4, 1500 int getfrag(void *from, char *to, int offset, 1501 int len, int odd, struct sk_buff *skb), 1502 void *from, int length, int transhdrlen, 1503 struct ipcm_cookie *ipc, struct rtable **rtp, 1504 unsigned int flags) 1505 { 1506 struct inet_cork cork; 1507 struct sk_buff_head queue; 1508 int err; 1509 1510 if (flags & MSG_PROBE) 1511 return NULL; 1512 1513 __skb_queue_head_init(&queue); 1514 1515 cork.flags = 0; 1516 cork.addr = 0; 1517 cork.opt = NULL; 1518 err = ip_setup_cork(sk, &cork, ipc, rtp); 1519 if (err) 1520 return ERR_PTR(err); 1521 1522 err = __ip_append_data(sk, fl4, &queue, &cork, 1523 ¤t->task_frag, getfrag, 1524 from, length, transhdrlen, flags); 1525 if (err) { 1526 __ip_flush_pending_frames(sk, &queue, &cork); 1527 return ERR_PTR(err); 1528 } 1529 1530 return __ip_make_skb(sk, fl4, &queue, &cork); 1531 } 1532 1533 /* 1534 * Fetch data from kernel space and fill in checksum if needed. 1535 */ 1536 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 1537 int len, int odd, struct sk_buff *skb) 1538 { 1539 __wsum csum; 1540 1541 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); 1542 skb->csum = csum_block_add(skb->csum, csum, odd); 1543 return 0; 1544 } 1545 1546 /* 1547 * Generic function to send a packet as reply to another packet. 1548 * Used to send some TCP resets/acks so far. 1549 */ 1550 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, 1551 const struct ip_options *sopt, 1552 __be32 daddr, __be32 saddr, 1553 const struct ip_reply_arg *arg, 1554 unsigned int len) 1555 { 1556 struct ip_options_data replyopts; 1557 struct ipcm_cookie ipc; 1558 struct flowi4 fl4; 1559 struct rtable *rt = skb_rtable(skb); 1560 struct net *net = sock_net(sk); 1561 struct sk_buff *nskb; 1562 int err; 1563 int oif; 1564 1565 if (__ip_options_echo(&replyopts.opt.opt, skb, sopt)) 1566 return; 1567 1568 ipc.addr = daddr; 1569 ipc.opt = NULL; 1570 ipc.tx_flags = 0; 1571 ipc.ttl = 0; 1572 ipc.tos = -1; 1573 1574 if (replyopts.opt.opt.optlen) { 1575 ipc.opt = &replyopts.opt; 1576 1577 if (replyopts.opt.opt.srr) 1578 daddr = replyopts.opt.opt.faddr; 1579 } 1580 1581 oif = arg->bound_dev_if; 1582 oif = oif ? : skb->skb_iif; 1583 1584 flowi4_init_output(&fl4, oif, 1585 IP4_REPLY_MARK(net, skb->mark), 1586 RT_TOS(arg->tos), 1587 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, 1588 ip_reply_arg_flowi_flags(arg), 1589 daddr, saddr, 1590 tcp_hdr(skb)->source, tcp_hdr(skb)->dest); 1591 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 1592 rt = ip_route_output_key(net, &fl4); 1593 if (IS_ERR(rt)) 1594 return; 1595 1596 inet_sk(sk)->tos = arg->tos; 1597 1598 sk->sk_priority = skb->priority; 1599 sk->sk_protocol = ip_hdr(skb)->protocol; 1600 sk->sk_bound_dev_if = arg->bound_dev_if; 1601 sk->sk_sndbuf = sysctl_wmem_default; 1602 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, 1603 len, 0, &ipc, &rt, MSG_DONTWAIT); 1604 if (unlikely(err)) { 1605 ip_flush_pending_frames(sk); 1606 goto out; 1607 } 1608 1609 nskb = skb_peek(&sk->sk_write_queue); 1610 if (nskb) { 1611 if (arg->csumoffset >= 0) 1612 *((__sum16 *)skb_transport_header(nskb) + 1613 arg->csumoffset) = csum_fold(csum_add(nskb->csum, 1614 arg->csum)); 1615 nskb->ip_summed = CHECKSUM_NONE; 1616 ip_push_pending_frames(sk, &fl4); 1617 } 1618 out: 1619 ip_rt_put(rt); 1620 } 1621 1622 void __init ip_init(void) 1623 { 1624 ip_rt_init(); 1625 inet_initpeers(); 1626 1627 #if defined(CONFIG_IP_MULTICAST) 1628 igmp_mc_init(); 1629 #endif 1630 } 1631