1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The Internet Protocol (IP) output module. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Donald Becker, <becker@super.org> 11 * Alan Cox, <Alan.Cox@linux.org> 12 * Richard Underwood 13 * Stefan Becker, <stefanb@yello.ping.de> 14 * Jorge Cwik, <jorge@laser.satlink.net> 15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 16 * Hirokazu Takahashi, <taka@valinux.co.jp> 17 * 18 * See ip_input.c for original log 19 * 20 * Fixes: 21 * Alan Cox : Missing nonblock feature in ip_build_xmit. 22 * Mike Kilburn : htons() missing in ip_build_xmit. 23 * Bradford Johnson: Fix faulty handling of some frames when 24 * no route is found. 25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit 26 * (in case if packet not accepted by 27 * output firewall rules) 28 * Mike McLagan : Routing by source 29 * Alexey Kuznetsov: use new route cache 30 * Andi Kleen: Fix broken PMTU recovery and remove 31 * some redundant tests. 32 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 33 * Andi Kleen : Replace ip_reply with ip_send_reply. 34 * Andi Kleen : Split fast and slow ip_build_xmit path 35 * for decreased register pressure on x86 36 * and more readibility. 37 * Marc Boucher : When call_out_firewall returns FW_QUEUE, 38 * silently drop skb instead of failing with -EPERM. 39 * Detlev Wengorz : Copy protocol for fragments. 40 * Hirokazu Takahashi: HW checksumming for outgoing UDP 41 * datagrams. 42 * Hirokazu Takahashi: sendfile() on UDP works now. 43 */ 44 45 #include <asm/uaccess.h> 46 #include <linux/module.h> 47 #include <linux/types.h> 48 #include <linux/kernel.h> 49 #include <linux/mm.h> 50 #include <linux/string.h> 51 #include <linux/errno.h> 52 #include <linux/highmem.h> 53 #include <linux/slab.h> 54 55 #include <linux/socket.h> 56 #include <linux/sockios.h> 57 #include <linux/in.h> 58 #include <linux/inet.h> 59 #include <linux/netdevice.h> 60 #include <linux/etherdevice.h> 61 #include <linux/proc_fs.h> 62 #include <linux/stat.h> 63 #include <linux/init.h> 64 65 #include <net/snmp.h> 66 #include <net/ip.h> 67 #include <net/protocol.h> 68 #include <net/route.h> 69 #include <net/xfrm.h> 70 #include <linux/skbuff.h> 71 #include <net/sock.h> 72 #include <net/arp.h> 73 #include <net/icmp.h> 74 #include <net/checksum.h> 75 #include <net/inetpeer.h> 76 #include <linux/igmp.h> 77 #include <linux/netfilter_ipv4.h> 78 #include <linux/netfilter_bridge.h> 79 #include <linux/netlink.h> 80 #include <linux/tcp.h> 81 82 static int 83 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 84 unsigned int mtu, 85 int (*output)(struct net *, struct sock *, struct sk_buff *)); 86 87 /* Generate a checksum for an outgoing IP datagram. */ 88 void ip_send_check(struct iphdr *iph) 89 { 90 iph->check = 0; 91 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 92 } 93 EXPORT_SYMBOL(ip_send_check); 94 95 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 96 { 97 struct iphdr *iph = ip_hdr(skb); 98 99 iph->tot_len = htons(skb->len); 100 ip_send_check(iph); 101 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, 102 net, sk, skb, NULL, skb_dst(skb)->dev, 103 dst_output); 104 } 105 106 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 107 { 108 int err; 109 110 err = __ip_local_out(net, sk, skb); 111 if (likely(err == 1)) 112 err = dst_output(net, sk, skb); 113 114 return err; 115 } 116 EXPORT_SYMBOL_GPL(ip_local_out); 117 118 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) 119 { 120 int ttl = inet->uc_ttl; 121 122 if (ttl < 0) 123 ttl = ip4_dst_hoplimit(dst); 124 return ttl; 125 } 126 127 /* 128 * Add an ip header to a skbuff and send it out. 129 * 130 */ 131 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk, 132 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) 133 { 134 struct inet_sock *inet = inet_sk(sk); 135 struct rtable *rt = skb_rtable(skb); 136 struct net *net = sock_net(sk); 137 struct iphdr *iph; 138 139 /* Build the IP header. */ 140 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); 141 skb_reset_network_header(skb); 142 iph = ip_hdr(skb); 143 iph->version = 4; 144 iph->ihl = 5; 145 iph->tos = inet->tos; 146 iph->ttl = ip_select_ttl(inet, &rt->dst); 147 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); 148 iph->saddr = saddr; 149 iph->protocol = sk->sk_protocol; 150 if (ip_dont_fragment(sk, &rt->dst)) { 151 iph->frag_off = htons(IP_DF); 152 iph->id = 0; 153 } else { 154 iph->frag_off = 0; 155 __ip_select_ident(net, iph, 1); 156 } 157 158 if (opt && opt->opt.optlen) { 159 iph->ihl += opt->opt.optlen>>2; 160 ip_options_build(skb, &opt->opt, daddr, rt, 0); 161 } 162 163 skb->priority = sk->sk_priority; 164 skb->mark = sk->sk_mark; 165 166 /* Send it out. */ 167 return ip_local_out(net, skb->sk, skb); 168 } 169 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); 170 171 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) 172 { 173 struct dst_entry *dst = skb_dst(skb); 174 struct rtable *rt = (struct rtable *)dst; 175 struct net_device *dev = dst->dev; 176 unsigned int hh_len = LL_RESERVED_SPACE(dev); 177 struct neighbour *neigh; 178 u32 nexthop; 179 180 if (rt->rt_type == RTN_MULTICAST) { 181 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len); 182 } else if (rt->rt_type == RTN_BROADCAST) 183 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len); 184 185 /* Be paranoid, rather than too clever. */ 186 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 187 struct sk_buff *skb2; 188 189 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 190 if (!skb2) { 191 kfree_skb(skb); 192 return -ENOMEM; 193 } 194 if (skb->sk) 195 skb_set_owner_w(skb2, skb->sk); 196 consume_skb(skb); 197 skb = skb2; 198 } 199 200 rcu_read_lock_bh(); 201 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr); 202 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 203 if (unlikely(!neigh)) 204 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 205 if (!IS_ERR(neigh)) { 206 int res = dst_neigh_output(dst, neigh, skb); 207 208 rcu_read_unlock_bh(); 209 return res; 210 } 211 rcu_read_unlock_bh(); 212 213 net_dbg_ratelimited("%s: No header cache and no neighbour!\n", 214 __func__); 215 kfree_skb(skb); 216 return -EINVAL; 217 } 218 219 static int ip_finish_output_gso(struct net *net, struct sock *sk, 220 struct sk_buff *skb, unsigned int mtu) 221 { 222 netdev_features_t features; 223 struct sk_buff *segs; 224 int ret = 0; 225 226 /* common case: locally created skb or seglen is <= mtu */ 227 if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) || 228 skb_gso_network_seglen(skb) <= mtu) 229 return ip_finish_output2(net, sk, skb); 230 231 /* Slowpath - GSO segment length is exceeding the dst MTU. 232 * 233 * This can happen in two cases: 234 * 1) TCP GRO packet, DF bit not set 235 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly 236 * from host network stack. 237 */ 238 features = netif_skb_features(skb); 239 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET); 240 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); 241 if (IS_ERR_OR_NULL(segs)) { 242 kfree_skb(skb); 243 return -ENOMEM; 244 } 245 246 consume_skb(skb); 247 248 do { 249 struct sk_buff *nskb = segs->next; 250 int err; 251 252 segs->next = NULL; 253 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2); 254 255 if (err && ret == 0) 256 ret = err; 257 segs = nskb; 258 } while (segs); 259 260 return ret; 261 } 262 263 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 264 { 265 unsigned int mtu; 266 267 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) 268 /* Policy lookup after SNAT yielded a new policy */ 269 if (skb_dst(skb)->xfrm) { 270 IPCB(skb)->flags |= IPSKB_REROUTED; 271 return dst_output(net, sk, skb); 272 } 273 #endif 274 mtu = ip_skb_dst_mtu(sk, skb); 275 if (skb_is_gso(skb)) 276 return ip_finish_output_gso(net, sk, skb, mtu); 277 278 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU)) 279 return ip_fragment(net, sk, skb, mtu, ip_finish_output2); 280 281 return ip_finish_output2(net, sk, skb); 282 } 283 284 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb) 285 { 286 struct rtable *rt = skb_rtable(skb); 287 struct net_device *dev = rt->dst.dev; 288 289 /* 290 * If the indicated interface is up and running, send the packet. 291 */ 292 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 293 294 skb->dev = dev; 295 skb->protocol = htons(ETH_P_IP); 296 297 /* 298 * Multicasts are looped back for other local users 299 */ 300 301 if (rt->rt_flags&RTCF_MULTICAST) { 302 if (sk_mc_loop(sk) 303 #ifdef CONFIG_IP_MROUTE 304 /* Small optimization: do not loopback not local frames, 305 which returned after forwarding; they will be dropped 306 by ip_mr_input in any case. 307 Note, that local frames are looped back to be delivered 308 to local recipients. 309 310 This check is duplicated in ip_mr_input at the moment. 311 */ 312 && 313 ((rt->rt_flags & RTCF_LOCAL) || 314 !(IPCB(skb)->flags & IPSKB_FORWARDED)) 315 #endif 316 ) { 317 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 318 if (newskb) 319 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 320 net, sk, newskb, NULL, newskb->dev, 321 dev_loopback_xmit); 322 } 323 324 /* Multicasts with ttl 0 must not go beyond the host */ 325 326 if (ip_hdr(skb)->ttl == 0) { 327 kfree_skb(skb); 328 return 0; 329 } 330 } 331 332 if (rt->rt_flags&RTCF_BROADCAST) { 333 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 334 if (newskb) 335 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 336 net, sk, newskb, NULL, newskb->dev, 337 dev_loopback_xmit); 338 } 339 340 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 341 net, sk, skb, NULL, skb->dev, 342 ip_finish_output, 343 !(IPCB(skb)->flags & IPSKB_REROUTED)); 344 } 345 346 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb) 347 { 348 struct net_device *dev = skb_dst(skb)->dev; 349 350 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 351 352 skb->dev = dev; 353 skb->protocol = htons(ETH_P_IP); 354 355 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 356 net, sk, skb, NULL, dev, 357 ip_finish_output, 358 !(IPCB(skb)->flags & IPSKB_REROUTED)); 359 } 360 361 /* 362 * copy saddr and daddr, possibly using 64bit load/stores 363 * Equivalent to : 364 * iph->saddr = fl4->saddr; 365 * iph->daddr = fl4->daddr; 366 */ 367 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) 368 { 369 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != 370 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); 371 memcpy(&iph->saddr, &fl4->saddr, 372 sizeof(fl4->saddr) + sizeof(fl4->daddr)); 373 } 374 375 /* Note: skb->sk can be different from sk, in case of tunnels */ 376 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) 377 { 378 struct inet_sock *inet = inet_sk(sk); 379 struct net *net = sock_net(sk); 380 struct ip_options_rcu *inet_opt; 381 struct flowi4 *fl4; 382 struct rtable *rt; 383 struct iphdr *iph; 384 int res; 385 386 /* Skip all of this if the packet is already routed, 387 * f.e. by something like SCTP. 388 */ 389 rcu_read_lock(); 390 inet_opt = rcu_dereference(inet->inet_opt); 391 fl4 = &fl->u.ip4; 392 rt = skb_rtable(skb); 393 if (rt) 394 goto packet_routed; 395 396 /* Make sure we can route this packet. */ 397 rt = (struct rtable *)__sk_dst_check(sk, 0); 398 if (!rt) { 399 __be32 daddr; 400 401 /* Use correct destination address if we have options. */ 402 daddr = inet->inet_daddr; 403 if (inet_opt && inet_opt->opt.srr) 404 daddr = inet_opt->opt.faddr; 405 406 /* If this fails, retransmit mechanism of transport layer will 407 * keep trying until route appears or the connection times 408 * itself out. 409 */ 410 rt = ip_route_output_ports(net, fl4, sk, 411 daddr, inet->inet_saddr, 412 inet->inet_dport, 413 inet->inet_sport, 414 sk->sk_protocol, 415 RT_CONN_FLAGS(sk), 416 sk->sk_bound_dev_if); 417 if (IS_ERR(rt)) 418 goto no_route; 419 sk_setup_caps(sk, &rt->dst); 420 } 421 skb_dst_set_noref(skb, &rt->dst); 422 423 packet_routed: 424 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) 425 goto no_route; 426 427 /* OK, we know where to send it, allocate and build IP header. */ 428 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 429 skb_reset_network_header(skb); 430 iph = ip_hdr(skb); 431 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); 432 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) 433 iph->frag_off = htons(IP_DF); 434 else 435 iph->frag_off = 0; 436 iph->ttl = ip_select_ttl(inet, &rt->dst); 437 iph->protocol = sk->sk_protocol; 438 ip_copy_addrs(iph, fl4); 439 440 /* Transport layer set skb->h.foo itself. */ 441 442 if (inet_opt && inet_opt->opt.optlen) { 443 iph->ihl += inet_opt->opt.optlen >> 2; 444 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 445 } 446 447 ip_select_ident_segs(net, skb, sk, 448 skb_shinfo(skb)->gso_segs ?: 1); 449 450 /* TODO : should we use skb->sk here instead of sk ? */ 451 skb->priority = sk->sk_priority; 452 skb->mark = sk->sk_mark; 453 454 res = ip_local_out(net, sk, skb); 455 rcu_read_unlock(); 456 return res; 457 458 no_route: 459 rcu_read_unlock(); 460 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 461 kfree_skb(skb); 462 return -EHOSTUNREACH; 463 } 464 EXPORT_SYMBOL(ip_queue_xmit); 465 466 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) 467 { 468 to->pkt_type = from->pkt_type; 469 to->priority = from->priority; 470 to->protocol = from->protocol; 471 skb_dst_drop(to); 472 skb_dst_copy(to, from); 473 to->dev = from->dev; 474 to->mark = from->mark; 475 476 /* Copy the flags to each fragment. */ 477 IPCB(to)->flags = IPCB(from)->flags; 478 479 #ifdef CONFIG_NET_SCHED 480 to->tc_index = from->tc_index; 481 #endif 482 nf_copy(to, from); 483 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 484 to->ipvs_property = from->ipvs_property; 485 #endif 486 skb_copy_secmark(to, from); 487 } 488 489 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 490 unsigned int mtu, 491 int (*output)(struct net *, struct sock *, struct sk_buff *)) 492 { 493 struct iphdr *iph = ip_hdr(skb); 494 495 if ((iph->frag_off & htons(IP_DF)) == 0) 496 return ip_do_fragment(net, sk, skb, output); 497 498 if (unlikely(!skb->ignore_df || 499 (IPCB(skb)->frag_max_size && 500 IPCB(skb)->frag_max_size > mtu))) { 501 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 502 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 503 htonl(mtu)); 504 kfree_skb(skb); 505 return -EMSGSIZE; 506 } 507 508 return ip_do_fragment(net, sk, skb, output); 509 } 510 511 /* 512 * This IP datagram is too large to be sent in one piece. Break it up into 513 * smaller pieces (each of size equal to IP header plus 514 * a block of the data of the original IP data part) that will yet fit in a 515 * single device frame, and queue such a frame for sending. 516 */ 517 518 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 519 int (*output)(struct net *, struct sock *, struct sk_buff *)) 520 { 521 struct iphdr *iph; 522 int ptr; 523 struct net_device *dev; 524 struct sk_buff *skb2; 525 unsigned int mtu, hlen, left, len, ll_rs; 526 int offset; 527 __be16 not_last_frag; 528 struct rtable *rt = skb_rtable(skb); 529 int err = 0; 530 531 dev = rt->dst.dev; 532 533 /* for offloaded checksums cleanup checksum before fragmentation */ 534 if (skb->ip_summed == CHECKSUM_PARTIAL && 535 (err = skb_checksum_help(skb))) 536 goto fail; 537 538 /* 539 * Point into the IP datagram header. 540 */ 541 542 iph = ip_hdr(skb); 543 544 mtu = ip_skb_dst_mtu(sk, skb); 545 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) 546 mtu = IPCB(skb)->frag_max_size; 547 548 /* 549 * Setup starting values. 550 */ 551 552 hlen = iph->ihl * 4; 553 mtu = mtu - hlen; /* Size of data space */ 554 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 555 556 /* When frag_list is given, use it. First, check its validity: 557 * some transformers could create wrong frag_list or break existing 558 * one, it is not prohibited. In this case fall back to copying. 559 * 560 * LATER: this step can be merged to real generation of fragments, 561 * we can switch to copy when see the first bad fragment. 562 */ 563 if (skb_has_frag_list(skb)) { 564 struct sk_buff *frag, *frag2; 565 int first_len = skb_pagelen(skb); 566 567 if (first_len - hlen > mtu || 568 ((first_len - hlen) & 7) || 569 ip_is_fragment(iph) || 570 skb_cloned(skb)) 571 goto slow_path; 572 573 skb_walk_frags(skb, frag) { 574 /* Correct geometry. */ 575 if (frag->len > mtu || 576 ((frag->len & 7) && frag->next) || 577 skb_headroom(frag) < hlen) 578 goto slow_path_clean; 579 580 /* Partially cloned skb? */ 581 if (skb_shared(frag)) 582 goto slow_path_clean; 583 584 BUG_ON(frag->sk); 585 if (skb->sk) { 586 frag->sk = skb->sk; 587 frag->destructor = sock_wfree; 588 } 589 skb->truesize -= frag->truesize; 590 } 591 592 /* Everything is OK. Generate! */ 593 594 err = 0; 595 offset = 0; 596 frag = skb_shinfo(skb)->frag_list; 597 skb_frag_list_init(skb); 598 skb->data_len = first_len - skb_headlen(skb); 599 skb->len = first_len; 600 iph->tot_len = htons(first_len); 601 iph->frag_off = htons(IP_MF); 602 ip_send_check(iph); 603 604 for (;;) { 605 /* Prepare header of the next frame, 606 * before previous one went down. */ 607 if (frag) { 608 frag->ip_summed = CHECKSUM_NONE; 609 skb_reset_transport_header(frag); 610 __skb_push(frag, hlen); 611 skb_reset_network_header(frag); 612 memcpy(skb_network_header(frag), iph, hlen); 613 iph = ip_hdr(frag); 614 iph->tot_len = htons(frag->len); 615 ip_copy_metadata(frag, skb); 616 if (offset == 0) 617 ip_options_fragment(frag); 618 offset += skb->len - hlen; 619 iph->frag_off = htons(offset>>3); 620 if (frag->next) 621 iph->frag_off |= htons(IP_MF); 622 /* Ready, complete checksum */ 623 ip_send_check(iph); 624 } 625 626 err = output(net, sk, skb); 627 628 if (!err) 629 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 630 if (err || !frag) 631 break; 632 633 skb = frag; 634 frag = skb->next; 635 skb->next = NULL; 636 } 637 638 if (err == 0) { 639 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 640 return 0; 641 } 642 643 while (frag) { 644 skb = frag->next; 645 kfree_skb(frag); 646 frag = skb; 647 } 648 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 649 return err; 650 651 slow_path_clean: 652 skb_walk_frags(skb, frag2) { 653 if (frag2 == frag) 654 break; 655 frag2->sk = NULL; 656 frag2->destructor = NULL; 657 skb->truesize += frag2->truesize; 658 } 659 } 660 661 slow_path: 662 iph = ip_hdr(skb); 663 664 left = skb->len - hlen; /* Space per frame */ 665 ptr = hlen; /* Where to start from */ 666 667 ll_rs = LL_RESERVED_SPACE(rt->dst.dev); 668 669 /* 670 * Fragment the datagram. 671 */ 672 673 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 674 not_last_frag = iph->frag_off & htons(IP_MF); 675 676 /* 677 * Keep copying data until we run out. 678 */ 679 680 while (left > 0) { 681 len = left; 682 /* IF: it doesn't fit, use 'mtu' - the data space left */ 683 if (len > mtu) 684 len = mtu; 685 /* IF: we are not sending up to and including the packet end 686 then align the next start on an eight byte boundary */ 687 if (len < left) { 688 len &= ~7; 689 } 690 691 /* Allocate buffer */ 692 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC); 693 if (!skb2) { 694 err = -ENOMEM; 695 goto fail; 696 } 697 698 /* 699 * Set up data on packet 700 */ 701 702 ip_copy_metadata(skb2, skb); 703 skb_reserve(skb2, ll_rs); 704 skb_put(skb2, len + hlen); 705 skb_reset_network_header(skb2); 706 skb2->transport_header = skb2->network_header + hlen; 707 708 /* 709 * Charge the memory for the fragment to any owner 710 * it might possess 711 */ 712 713 if (skb->sk) 714 skb_set_owner_w(skb2, skb->sk); 715 716 /* 717 * Copy the packet header into the new buffer. 718 */ 719 720 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); 721 722 /* 723 * Copy a block of the IP datagram. 724 */ 725 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) 726 BUG(); 727 left -= len; 728 729 /* 730 * Fill in the new header fields. 731 */ 732 iph = ip_hdr(skb2); 733 iph->frag_off = htons((offset >> 3)); 734 735 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU) 736 iph->frag_off |= htons(IP_DF); 737 738 /* ANK: dirty, but effective trick. Upgrade options only if 739 * the segment to be fragmented was THE FIRST (otherwise, 740 * options are already fixed) and make it ONCE 741 * on the initial skb, so that all the following fragments 742 * will inherit fixed options. 743 */ 744 if (offset == 0) 745 ip_options_fragment(skb); 746 747 /* 748 * Added AC : If we are fragmenting a fragment that's not the 749 * last fragment then keep MF on each bit 750 */ 751 if (left > 0 || not_last_frag) 752 iph->frag_off |= htons(IP_MF); 753 ptr += len; 754 offset += len; 755 756 /* 757 * Put this fragment into the sending queue. 758 */ 759 iph->tot_len = htons(len + hlen); 760 761 ip_send_check(iph); 762 763 err = output(net, sk, skb2); 764 if (err) 765 goto fail; 766 767 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 768 } 769 consume_skb(skb); 770 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 771 return err; 772 773 fail: 774 kfree_skb(skb); 775 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 776 return err; 777 } 778 EXPORT_SYMBOL(ip_do_fragment); 779 780 int 781 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 782 { 783 struct msghdr *msg = from; 784 785 if (skb->ip_summed == CHECKSUM_PARTIAL) { 786 if (copy_from_iter(to, len, &msg->msg_iter) != len) 787 return -EFAULT; 788 } else { 789 __wsum csum = 0; 790 if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len) 791 return -EFAULT; 792 skb->csum = csum_block_add(skb->csum, csum, odd); 793 } 794 return 0; 795 } 796 EXPORT_SYMBOL(ip_generic_getfrag); 797 798 static inline __wsum 799 csum_page(struct page *page, int offset, int copy) 800 { 801 char *kaddr; 802 __wsum csum; 803 kaddr = kmap(page); 804 csum = csum_partial(kaddr + offset, copy, 0); 805 kunmap(page); 806 return csum; 807 } 808 809 static inline int ip_ufo_append_data(struct sock *sk, 810 struct sk_buff_head *queue, 811 int getfrag(void *from, char *to, int offset, int len, 812 int odd, struct sk_buff *skb), 813 void *from, int length, int hh_len, int fragheaderlen, 814 int transhdrlen, int maxfraglen, unsigned int flags) 815 { 816 struct sk_buff *skb; 817 int err; 818 819 /* There is support for UDP fragmentation offload by network 820 * device, so create one single skb packet containing complete 821 * udp datagram 822 */ 823 skb = skb_peek_tail(queue); 824 if (!skb) { 825 skb = sock_alloc_send_skb(sk, 826 hh_len + fragheaderlen + transhdrlen + 20, 827 (flags & MSG_DONTWAIT), &err); 828 829 if (!skb) 830 return err; 831 832 /* reserve space for Hardware header */ 833 skb_reserve(skb, hh_len); 834 835 /* create space for UDP/IP header */ 836 skb_put(skb, fragheaderlen + transhdrlen); 837 838 /* initialize network header pointer */ 839 skb_reset_network_header(skb); 840 841 /* initialize protocol header pointer */ 842 skb->transport_header = skb->network_header + fragheaderlen; 843 844 skb->csum = 0; 845 846 __skb_queue_tail(queue, skb); 847 } else if (skb_is_gso(skb)) { 848 goto append; 849 } 850 851 skb->ip_summed = CHECKSUM_PARTIAL; 852 /* specify the length of each IP datagram fragment */ 853 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen; 854 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 855 856 append: 857 return skb_append_datato_frags(sk, skb, getfrag, from, 858 (length - transhdrlen)); 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 sk_buff *skb; 873 874 struct ip_options *opt = cork->opt; 875 int hh_len; 876 int exthdrlen; 877 int mtu; 878 int copy; 879 int err; 880 int offset = 0; 881 unsigned int maxfraglen, fragheaderlen, maxnonfragsize; 882 int csummode = CHECKSUM_NONE; 883 struct rtable *rt = (struct rtable *)cork->dst; 884 u32 tskey = 0; 885 886 skb = skb_peek_tail(queue); 887 888 exthdrlen = !skb ? rt->dst.header_len : 0; 889 mtu = cork->fragsize; 890 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && 891 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) 892 tskey = sk->sk_tskey++; 893 894 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 895 896 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 897 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 898 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 899 900 if (cork->length + length > maxnonfragsize - fragheaderlen) { 901 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 902 mtu - (opt ? opt->optlen : 0)); 903 return -EMSGSIZE; 904 } 905 906 /* 907 * transhdrlen > 0 means that this is the first fragment and we wish 908 * it won't be fragmented in the future. 909 */ 910 if (transhdrlen && 911 length + fragheaderlen <= mtu && 912 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) && 913 !(flags & MSG_MORE) && 914 !exthdrlen) 915 csummode = CHECKSUM_PARTIAL; 916 917 cork->length += length; 918 if (((length > mtu) || (skb && skb_is_gso(skb))) && 919 (sk->sk_protocol == IPPROTO_UDP) && 920 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len && 921 (sk->sk_type == SOCK_DGRAM) && !sk->sk_no_check_tx) { 922 err = ip_ufo_append_data(sk, queue, getfrag, from, length, 923 hh_len, fragheaderlen, transhdrlen, 924 maxfraglen, flags); 925 if (err) 926 goto error; 927 return 0; 928 } 929 930 /* So, what's going on in the loop below? 931 * 932 * We use calculated fragment length to generate chained skb, 933 * each of segments is IP fragment ready for sending to network after 934 * adding appropriate IP header. 935 */ 936 937 if (!skb) 938 goto alloc_new_skb; 939 940 while (length > 0) { 941 /* Check if the remaining data fits into current packet. */ 942 copy = mtu - skb->len; 943 if (copy < length) 944 copy = maxfraglen - skb->len; 945 if (copy <= 0) { 946 char *data; 947 unsigned int datalen; 948 unsigned int fraglen; 949 unsigned int fraggap; 950 unsigned int alloclen; 951 struct sk_buff *skb_prev; 952 alloc_new_skb: 953 skb_prev = skb; 954 if (skb_prev) 955 fraggap = skb_prev->len - maxfraglen; 956 else 957 fraggap = 0; 958 959 /* 960 * If remaining data exceeds the mtu, 961 * we know we need more fragment(s). 962 */ 963 datalen = length + fraggap; 964 if (datalen > mtu - fragheaderlen) 965 datalen = maxfraglen - fragheaderlen; 966 fraglen = datalen + fragheaderlen; 967 968 if ((flags & MSG_MORE) && 969 !(rt->dst.dev->features&NETIF_F_SG)) 970 alloclen = mtu; 971 else 972 alloclen = fraglen; 973 974 alloclen += exthdrlen; 975 976 /* The last fragment gets additional space at tail. 977 * Note, with MSG_MORE we overallocate on fragments, 978 * because we have no idea what fragment will be 979 * the last. 980 */ 981 if (datalen == length + fraggap) 982 alloclen += rt->dst.trailer_len; 983 984 if (transhdrlen) { 985 skb = sock_alloc_send_skb(sk, 986 alloclen + hh_len + 15, 987 (flags & MSG_DONTWAIT), &err); 988 } else { 989 skb = NULL; 990 if (atomic_read(&sk->sk_wmem_alloc) <= 991 2 * sk->sk_sndbuf) 992 skb = sock_wmalloc(sk, 993 alloclen + hh_len + 15, 1, 994 sk->sk_allocation); 995 if (unlikely(!skb)) 996 err = -ENOBUFS; 997 } 998 if (!skb) 999 goto error; 1000 1001 /* 1002 * Fill in the control structures 1003 */ 1004 skb->ip_summed = csummode; 1005 skb->csum = 0; 1006 skb_reserve(skb, hh_len); 1007 1008 /* only the initial fragment is time stamped */ 1009 skb_shinfo(skb)->tx_flags = cork->tx_flags; 1010 cork->tx_flags = 0; 1011 skb_shinfo(skb)->tskey = tskey; 1012 tskey = 0; 1013 1014 /* 1015 * Find where to start putting bytes. 1016 */ 1017 data = skb_put(skb, fraglen + exthdrlen); 1018 skb_set_network_header(skb, exthdrlen); 1019 skb->transport_header = (skb->network_header + 1020 fragheaderlen); 1021 data += fragheaderlen + exthdrlen; 1022 1023 if (fraggap) { 1024 skb->csum = skb_copy_and_csum_bits( 1025 skb_prev, maxfraglen, 1026 data + transhdrlen, fraggap, 0); 1027 skb_prev->csum = csum_sub(skb_prev->csum, 1028 skb->csum); 1029 data += fraggap; 1030 pskb_trim_unique(skb_prev, maxfraglen); 1031 } 1032 1033 copy = datalen - transhdrlen - fraggap; 1034 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 1035 err = -EFAULT; 1036 kfree_skb(skb); 1037 goto error; 1038 } 1039 1040 offset += copy; 1041 length -= datalen - fraggap; 1042 transhdrlen = 0; 1043 exthdrlen = 0; 1044 csummode = CHECKSUM_NONE; 1045 1046 /* 1047 * Put the packet on the pending queue. 1048 */ 1049 __skb_queue_tail(queue, skb); 1050 continue; 1051 } 1052 1053 if (copy > length) 1054 copy = length; 1055 1056 if (!(rt->dst.dev->features&NETIF_F_SG)) { 1057 unsigned int off; 1058 1059 off = skb->len; 1060 if (getfrag(from, skb_put(skb, copy), 1061 offset, copy, off, skb) < 0) { 1062 __skb_trim(skb, off); 1063 err = -EFAULT; 1064 goto error; 1065 } 1066 } else { 1067 int i = skb_shinfo(skb)->nr_frags; 1068 1069 err = -ENOMEM; 1070 if (!sk_page_frag_refill(sk, pfrag)) 1071 goto error; 1072 1073 if (!skb_can_coalesce(skb, i, pfrag->page, 1074 pfrag->offset)) { 1075 err = -EMSGSIZE; 1076 if (i == MAX_SKB_FRAGS) 1077 goto error; 1078 1079 __skb_fill_page_desc(skb, i, pfrag->page, 1080 pfrag->offset, 0); 1081 skb_shinfo(skb)->nr_frags = ++i; 1082 get_page(pfrag->page); 1083 } 1084 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1085 if (getfrag(from, 1086 page_address(pfrag->page) + pfrag->offset, 1087 offset, copy, skb->len, skb) < 0) 1088 goto error_efault; 1089 1090 pfrag->offset += copy; 1091 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1092 skb->len += copy; 1093 skb->data_len += copy; 1094 skb->truesize += copy; 1095 atomic_add(copy, &sk->sk_wmem_alloc); 1096 } 1097 offset += copy; 1098 length -= copy; 1099 } 1100 1101 return 0; 1102 1103 error_efault: 1104 err = -EFAULT; 1105 error: 1106 cork->length -= length; 1107 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1108 return err; 1109 } 1110 1111 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 1112 struct ipcm_cookie *ipc, struct rtable **rtp) 1113 { 1114 struct ip_options_rcu *opt; 1115 struct rtable *rt; 1116 1117 /* 1118 * setup for corking. 1119 */ 1120 opt = ipc->opt; 1121 if (opt) { 1122 if (!cork->opt) { 1123 cork->opt = kmalloc(sizeof(struct ip_options) + 40, 1124 sk->sk_allocation); 1125 if (unlikely(!cork->opt)) 1126 return -ENOBUFS; 1127 } 1128 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 1129 cork->flags |= IPCORK_OPT; 1130 cork->addr = ipc->addr; 1131 } 1132 rt = *rtp; 1133 if (unlikely(!rt)) 1134 return -EFAULT; 1135 /* 1136 * We steal reference to this route, caller should not release it 1137 */ 1138 *rtp = NULL; 1139 cork->fragsize = ip_sk_use_pmtu(sk) ? 1140 dst_mtu(&rt->dst) : rt->dst.dev->mtu; 1141 cork->dst = &rt->dst; 1142 cork->length = 0; 1143 cork->ttl = ipc->ttl; 1144 cork->tos = ipc->tos; 1145 cork->priority = ipc->priority; 1146 cork->tx_flags = ipc->tx_flags; 1147 1148 return 0; 1149 } 1150 1151 /* 1152 * ip_append_data() and ip_append_page() can make one large IP datagram 1153 * from many pieces of data. Each pieces will be holded on the socket 1154 * until ip_push_pending_frames() is called. Each piece can be a page 1155 * or non-page data. 1156 * 1157 * Not only UDP, other transport protocols - e.g. raw sockets - can use 1158 * this interface potentially. 1159 * 1160 * LATER: length must be adjusted by pad at tail, when it is required. 1161 */ 1162 int ip_append_data(struct sock *sk, struct flowi4 *fl4, 1163 int getfrag(void *from, char *to, int offset, int len, 1164 int odd, struct sk_buff *skb), 1165 void *from, int length, int transhdrlen, 1166 struct ipcm_cookie *ipc, struct rtable **rtp, 1167 unsigned int flags) 1168 { 1169 struct inet_sock *inet = inet_sk(sk); 1170 int err; 1171 1172 if (flags&MSG_PROBE) 1173 return 0; 1174 1175 if (skb_queue_empty(&sk->sk_write_queue)) { 1176 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 1177 if (err) 1178 return err; 1179 } else { 1180 transhdrlen = 0; 1181 } 1182 1183 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, 1184 sk_page_frag(sk), getfrag, 1185 from, length, transhdrlen, flags); 1186 } 1187 1188 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 1189 int offset, size_t size, int flags) 1190 { 1191 struct inet_sock *inet = inet_sk(sk); 1192 struct sk_buff *skb; 1193 struct rtable *rt; 1194 struct ip_options *opt = NULL; 1195 struct inet_cork *cork; 1196 int hh_len; 1197 int mtu; 1198 int len; 1199 int err; 1200 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; 1201 1202 if (inet->hdrincl) 1203 return -EPERM; 1204 1205 if (flags&MSG_PROBE) 1206 return 0; 1207 1208 if (skb_queue_empty(&sk->sk_write_queue)) 1209 return -EINVAL; 1210 1211 cork = &inet->cork.base; 1212 rt = (struct rtable *)cork->dst; 1213 if (cork->flags & IPCORK_OPT) 1214 opt = cork->opt; 1215 1216 if (!(rt->dst.dev->features&NETIF_F_SG)) 1217 return -EOPNOTSUPP; 1218 1219 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1220 mtu = cork->fragsize; 1221 1222 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1223 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1224 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 1225 1226 if (cork->length + size > maxnonfragsize - fragheaderlen) { 1227 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 1228 mtu - (opt ? opt->optlen : 0)); 1229 return -EMSGSIZE; 1230 } 1231 1232 skb = skb_peek_tail(&sk->sk_write_queue); 1233 if (!skb) 1234 return -EINVAL; 1235 1236 if ((size + skb->len > mtu) && 1237 (sk->sk_protocol == IPPROTO_UDP) && 1238 (rt->dst.dev->features & NETIF_F_UFO)) { 1239 if (skb->ip_summed != CHECKSUM_PARTIAL) 1240 return -EOPNOTSUPP; 1241 1242 skb_shinfo(skb)->gso_size = mtu - fragheaderlen; 1243 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 1244 } 1245 cork->length += size; 1246 1247 while (size > 0) { 1248 if (skb_is_gso(skb)) { 1249 len = size; 1250 } else { 1251 1252 /* Check if the remaining data fits into current packet. */ 1253 len = mtu - skb->len; 1254 if (len < size) 1255 len = maxfraglen - skb->len; 1256 } 1257 if (len <= 0) { 1258 struct sk_buff *skb_prev; 1259 int alloclen; 1260 1261 skb_prev = skb; 1262 fraggap = skb_prev->len - maxfraglen; 1263 1264 alloclen = fragheaderlen + hh_len + fraggap + 15; 1265 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 1266 if (unlikely(!skb)) { 1267 err = -ENOBUFS; 1268 goto error; 1269 } 1270 1271 /* 1272 * Fill in the control structures 1273 */ 1274 skb->ip_summed = CHECKSUM_NONE; 1275 skb->csum = 0; 1276 skb_reserve(skb, hh_len); 1277 1278 /* 1279 * Find where to start putting bytes. 1280 */ 1281 skb_put(skb, fragheaderlen + fraggap); 1282 skb_reset_network_header(skb); 1283 skb->transport_header = (skb->network_header + 1284 fragheaderlen); 1285 if (fraggap) { 1286 skb->csum = skb_copy_and_csum_bits(skb_prev, 1287 maxfraglen, 1288 skb_transport_header(skb), 1289 fraggap, 0); 1290 skb_prev->csum = csum_sub(skb_prev->csum, 1291 skb->csum); 1292 pskb_trim_unique(skb_prev, maxfraglen); 1293 } 1294 1295 /* 1296 * Put the packet on the pending queue. 1297 */ 1298 __skb_queue_tail(&sk->sk_write_queue, skb); 1299 continue; 1300 } 1301 1302 if (len > size) 1303 len = size; 1304 1305 if (skb_append_pagefrags(skb, page, offset, len)) { 1306 err = -EMSGSIZE; 1307 goto error; 1308 } 1309 1310 if (skb->ip_summed == CHECKSUM_NONE) { 1311 __wsum csum; 1312 csum = csum_page(page, offset, len); 1313 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1314 } 1315 1316 skb->len += len; 1317 skb->data_len += len; 1318 skb->truesize += len; 1319 atomic_add(len, &sk->sk_wmem_alloc); 1320 offset += len; 1321 size -= len; 1322 } 1323 return 0; 1324 1325 error: 1326 cork->length -= size; 1327 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1328 return err; 1329 } 1330 1331 static void ip_cork_release(struct inet_cork *cork) 1332 { 1333 cork->flags &= ~IPCORK_OPT; 1334 kfree(cork->opt); 1335 cork->opt = NULL; 1336 dst_release(cork->dst); 1337 cork->dst = NULL; 1338 } 1339 1340 /* 1341 * Combined all pending IP fragments on the socket as one IP datagram 1342 * and push them out. 1343 */ 1344 struct sk_buff *__ip_make_skb(struct sock *sk, 1345 struct flowi4 *fl4, 1346 struct sk_buff_head *queue, 1347 struct inet_cork *cork) 1348 { 1349 struct sk_buff *skb, *tmp_skb; 1350 struct sk_buff **tail_skb; 1351 struct inet_sock *inet = inet_sk(sk); 1352 struct net *net = sock_net(sk); 1353 struct ip_options *opt = NULL; 1354 struct rtable *rt = (struct rtable *)cork->dst; 1355 struct iphdr *iph; 1356 __be16 df = 0; 1357 __u8 ttl; 1358 1359 skb = __skb_dequeue(queue); 1360 if (!skb) 1361 goto out; 1362 tail_skb = &(skb_shinfo(skb)->frag_list); 1363 1364 /* move skb->data to ip header from ext header */ 1365 if (skb->data < skb_network_header(skb)) 1366 __skb_pull(skb, skb_network_offset(skb)); 1367 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1368 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1369 *tail_skb = tmp_skb; 1370 tail_skb = &(tmp_skb->next); 1371 skb->len += tmp_skb->len; 1372 skb->data_len += tmp_skb->len; 1373 skb->truesize += tmp_skb->truesize; 1374 tmp_skb->destructor = NULL; 1375 tmp_skb->sk = NULL; 1376 } 1377 1378 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 1379 * to fragment the frame generated here. No matter, what transforms 1380 * how transforms change size of the packet, it will come out. 1381 */ 1382 skb->ignore_df = ip_sk_ignore_df(sk); 1383 1384 /* DF bit is set when we want to see DF on outgoing frames. 1385 * If ignore_df is set too, we still allow to fragment this frame 1386 * locally. */ 1387 if (inet->pmtudisc == IP_PMTUDISC_DO || 1388 inet->pmtudisc == IP_PMTUDISC_PROBE || 1389 (skb->len <= dst_mtu(&rt->dst) && 1390 ip_dont_fragment(sk, &rt->dst))) 1391 df = htons(IP_DF); 1392 1393 if (cork->flags & IPCORK_OPT) 1394 opt = cork->opt; 1395 1396 if (cork->ttl != 0) 1397 ttl = cork->ttl; 1398 else if (rt->rt_type == RTN_MULTICAST) 1399 ttl = inet->mc_ttl; 1400 else 1401 ttl = ip_select_ttl(inet, &rt->dst); 1402 1403 iph = ip_hdr(skb); 1404 iph->version = 4; 1405 iph->ihl = 5; 1406 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; 1407 iph->frag_off = df; 1408 iph->ttl = ttl; 1409 iph->protocol = sk->sk_protocol; 1410 ip_copy_addrs(iph, fl4); 1411 ip_select_ident(net, skb, sk); 1412 1413 if (opt) { 1414 iph->ihl += opt->optlen>>2; 1415 ip_options_build(skb, opt, cork->addr, rt, 0); 1416 } 1417 1418 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; 1419 skb->mark = sk->sk_mark; 1420 /* 1421 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1422 * on dst refcount 1423 */ 1424 cork->dst = NULL; 1425 skb_dst_set(skb, &rt->dst); 1426 1427 if (iph->protocol == IPPROTO_ICMP) 1428 icmp_out_count(net, ((struct icmphdr *) 1429 skb_transport_header(skb))->type); 1430 1431 ip_cork_release(cork); 1432 out: 1433 return skb; 1434 } 1435 1436 int ip_send_skb(struct net *net, struct sk_buff *skb) 1437 { 1438 int err; 1439 1440 err = ip_local_out(net, skb->sk, skb); 1441 if (err) { 1442 if (err > 0) 1443 err = net_xmit_errno(err); 1444 if (err) 1445 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 1446 } 1447 1448 return err; 1449 } 1450 1451 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 1452 { 1453 struct sk_buff *skb; 1454 1455 skb = ip_finish_skb(sk, fl4); 1456 if (!skb) 1457 return 0; 1458 1459 /* Netfilter gets whole the not fragmented skb. */ 1460 return ip_send_skb(sock_net(sk), skb); 1461 } 1462 1463 /* 1464 * Throw away all pending data on the socket. 1465 */ 1466 static void __ip_flush_pending_frames(struct sock *sk, 1467 struct sk_buff_head *queue, 1468 struct inet_cork *cork) 1469 { 1470 struct sk_buff *skb; 1471 1472 while ((skb = __skb_dequeue_tail(queue)) != NULL) 1473 kfree_skb(skb); 1474 1475 ip_cork_release(cork); 1476 } 1477 1478 void ip_flush_pending_frames(struct sock *sk) 1479 { 1480 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 1481 } 1482 1483 struct sk_buff *ip_make_skb(struct sock *sk, 1484 struct flowi4 *fl4, 1485 int getfrag(void *from, char *to, int offset, 1486 int len, int odd, struct sk_buff *skb), 1487 void *from, int length, int transhdrlen, 1488 struct ipcm_cookie *ipc, struct rtable **rtp, 1489 unsigned int flags) 1490 { 1491 struct inet_cork cork; 1492 struct sk_buff_head queue; 1493 int err; 1494 1495 if (flags & MSG_PROBE) 1496 return NULL; 1497 1498 __skb_queue_head_init(&queue); 1499 1500 cork.flags = 0; 1501 cork.addr = 0; 1502 cork.opt = NULL; 1503 err = ip_setup_cork(sk, &cork, ipc, rtp); 1504 if (err) 1505 return ERR_PTR(err); 1506 1507 err = __ip_append_data(sk, fl4, &queue, &cork, 1508 ¤t->task_frag, getfrag, 1509 from, length, transhdrlen, flags); 1510 if (err) { 1511 __ip_flush_pending_frames(sk, &queue, &cork); 1512 return ERR_PTR(err); 1513 } 1514 1515 return __ip_make_skb(sk, fl4, &queue, &cork); 1516 } 1517 1518 /* 1519 * Fetch data from kernel space and fill in checksum if needed. 1520 */ 1521 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 1522 int len, int odd, struct sk_buff *skb) 1523 { 1524 __wsum csum; 1525 1526 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); 1527 skb->csum = csum_block_add(skb->csum, csum, odd); 1528 return 0; 1529 } 1530 1531 /* 1532 * Generic function to send a packet as reply to another packet. 1533 * Used to send some TCP resets/acks so far. 1534 */ 1535 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, 1536 const struct ip_options *sopt, 1537 __be32 daddr, __be32 saddr, 1538 const struct ip_reply_arg *arg, 1539 unsigned int len) 1540 { 1541 struct ip_options_data replyopts; 1542 struct ipcm_cookie ipc; 1543 struct flowi4 fl4; 1544 struct rtable *rt = skb_rtable(skb); 1545 struct net *net = sock_net(sk); 1546 struct sk_buff *nskb; 1547 int err; 1548 int oif; 1549 1550 if (__ip_options_echo(&replyopts.opt.opt, skb, sopt)) 1551 return; 1552 1553 ipc.addr = daddr; 1554 ipc.opt = NULL; 1555 ipc.tx_flags = 0; 1556 ipc.ttl = 0; 1557 ipc.tos = -1; 1558 1559 if (replyopts.opt.opt.optlen) { 1560 ipc.opt = &replyopts.opt; 1561 1562 if (replyopts.opt.opt.srr) 1563 daddr = replyopts.opt.opt.faddr; 1564 } 1565 1566 oif = arg->bound_dev_if; 1567 if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) 1568 oif = skb->skb_iif; 1569 1570 flowi4_init_output(&fl4, oif, 1571 IP4_REPLY_MARK(net, skb->mark), 1572 RT_TOS(arg->tos), 1573 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, 1574 ip_reply_arg_flowi_flags(arg), 1575 daddr, saddr, 1576 tcp_hdr(skb)->source, tcp_hdr(skb)->dest); 1577 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 1578 rt = ip_route_output_key(net, &fl4); 1579 if (IS_ERR(rt)) 1580 return; 1581 1582 inet_sk(sk)->tos = arg->tos; 1583 1584 sk->sk_priority = skb->priority; 1585 sk->sk_protocol = ip_hdr(skb)->protocol; 1586 sk->sk_bound_dev_if = arg->bound_dev_if; 1587 sk->sk_sndbuf = sysctl_wmem_default; 1588 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, 1589 len, 0, &ipc, &rt, MSG_DONTWAIT); 1590 if (unlikely(err)) { 1591 ip_flush_pending_frames(sk); 1592 goto out; 1593 } 1594 1595 nskb = skb_peek(&sk->sk_write_queue); 1596 if (nskb) { 1597 if (arg->csumoffset >= 0) 1598 *((__sum16 *)skb_transport_header(nskb) + 1599 arg->csumoffset) = csum_fold(csum_add(nskb->csum, 1600 arg->csum)); 1601 nskb->ip_summed = CHECKSUM_NONE; 1602 ip_push_pending_frames(sk, &fl4); 1603 } 1604 out: 1605 ip_rt_put(rt); 1606 } 1607 1608 void __init ip_init(void) 1609 { 1610 ip_rt_init(); 1611 inet_initpeers(); 1612 1613 #if defined(CONFIG_IP_MULTICAST) 1614 igmp_mc_init(); 1615 #endif 1616 } 1617