1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * The Internet Protocol (IP) output module. 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Donald Becker, <becker@super.org> 12 * Alan Cox, <Alan.Cox@linux.org> 13 * Richard Underwood 14 * Stefan Becker, <stefanb@yello.ping.de> 15 * Jorge Cwik, <jorge@laser.satlink.net> 16 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 17 * Hirokazu Takahashi, <taka@valinux.co.jp> 18 * 19 * See ip_input.c for original log 20 * 21 * Fixes: 22 * Alan Cox : Missing nonblock feature in ip_build_xmit. 23 * Mike Kilburn : htons() missing in ip_build_xmit. 24 * Bradford Johnson: Fix faulty handling of some frames when 25 * no route is found. 26 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit 27 * (in case if packet not accepted by 28 * output firewall rules) 29 * Mike McLagan : Routing by source 30 * Alexey Kuznetsov: use new route cache 31 * Andi Kleen: Fix broken PMTU recovery and remove 32 * some redundant tests. 33 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 34 * Andi Kleen : Replace ip_reply with ip_send_reply. 35 * Andi Kleen : Split fast and slow ip_build_xmit path 36 * for decreased register pressure on x86 37 * and more readibility. 38 * Marc Boucher : When call_out_firewall returns FW_QUEUE, 39 * silently drop skb instead of failing with -EPERM. 40 * Detlev Wengorz : Copy protocol for fragments. 41 * Hirokazu Takahashi: HW checksumming for outgoing UDP 42 * datagrams. 43 * Hirokazu Takahashi: sendfile() on UDP works now. 44 */ 45 46 #include <linux/uaccess.h> 47 #include <linux/module.h> 48 #include <linux/types.h> 49 #include <linux/kernel.h> 50 #include <linux/mm.h> 51 #include <linux/string.h> 52 #include <linux/errno.h> 53 #include <linux/highmem.h> 54 #include <linux/slab.h> 55 56 #include <linux/socket.h> 57 #include <linux/sockios.h> 58 #include <linux/in.h> 59 #include <linux/inet.h> 60 #include <linux/netdevice.h> 61 #include <linux/etherdevice.h> 62 #include <linux/proc_fs.h> 63 #include <linux/stat.h> 64 #include <linux/init.h> 65 66 #include <net/snmp.h> 67 #include <net/ip.h> 68 #include <net/protocol.h> 69 #include <net/route.h> 70 #include <net/xfrm.h> 71 #include <linux/skbuff.h> 72 #include <net/sock.h> 73 #include <net/arp.h> 74 #include <net/icmp.h> 75 #include <net/checksum.h> 76 #include <net/inetpeer.h> 77 #include <net/lwtunnel.h> 78 #include <linux/bpf-cgroup.h> 79 #include <linux/igmp.h> 80 #include <linux/netfilter_ipv4.h> 81 #include <linux/netfilter_bridge.h> 82 #include <linux/netlink.h> 83 #include <linux/tcp.h> 84 85 static int 86 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 87 unsigned int mtu, 88 int (*output)(struct net *, struct sock *, struct sk_buff *)); 89 90 /* Generate a checksum for an outgoing IP datagram. */ 91 void ip_send_check(struct iphdr *iph) 92 { 93 iph->check = 0; 94 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 95 } 96 EXPORT_SYMBOL(ip_send_check); 97 98 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 99 { 100 struct iphdr *iph = ip_hdr(skb); 101 102 iph->tot_len = htons(skb->len); 103 ip_send_check(iph); 104 105 /* if egress device is enslaved to an L3 master device pass the 106 * skb to its handler for processing 107 */ 108 skb = l3mdev_ip_out(sk, skb); 109 if (unlikely(!skb)) 110 return 0; 111 112 skb->protocol = htons(ETH_P_IP); 113 114 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, 115 net, sk, skb, NULL, skb_dst(skb)->dev, 116 dst_output); 117 } 118 119 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 120 { 121 int err; 122 123 err = __ip_local_out(net, sk, skb); 124 if (likely(err == 1)) 125 err = dst_output(net, sk, skb); 126 127 return err; 128 } 129 EXPORT_SYMBOL_GPL(ip_local_out); 130 131 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) 132 { 133 int ttl = inet->uc_ttl; 134 135 if (ttl < 0) 136 ttl = ip4_dst_hoplimit(dst); 137 return ttl; 138 } 139 140 /* 141 * Add an ip header to a skbuff and send it out. 142 * 143 */ 144 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk, 145 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) 146 { 147 struct inet_sock *inet = inet_sk(sk); 148 struct rtable *rt = skb_rtable(skb); 149 struct net *net = sock_net(sk); 150 struct iphdr *iph; 151 152 /* Build the IP header. */ 153 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); 154 skb_reset_network_header(skb); 155 iph = ip_hdr(skb); 156 iph->version = 4; 157 iph->ihl = 5; 158 iph->tos = inet->tos; 159 iph->ttl = ip_select_ttl(inet, &rt->dst); 160 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); 161 iph->saddr = saddr; 162 iph->protocol = sk->sk_protocol; 163 if (ip_dont_fragment(sk, &rt->dst)) { 164 iph->frag_off = htons(IP_DF); 165 iph->id = 0; 166 } else { 167 iph->frag_off = 0; 168 __ip_select_ident(net, iph, 1); 169 } 170 171 if (opt && opt->opt.optlen) { 172 iph->ihl += opt->opt.optlen>>2; 173 ip_options_build(skb, &opt->opt, daddr, rt, 0); 174 } 175 176 skb->priority = sk->sk_priority; 177 if (!skb->mark) 178 skb->mark = sk->sk_mark; 179 180 /* Send it out. */ 181 return ip_local_out(net, skb->sk, skb); 182 } 183 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); 184 185 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) 186 { 187 struct dst_entry *dst = skb_dst(skb); 188 struct rtable *rt = (struct rtable *)dst; 189 struct net_device *dev = dst->dev; 190 unsigned int hh_len = LL_RESERVED_SPACE(dev); 191 struct neighbour *neigh; 192 bool is_v6gw = false; 193 194 if (rt->rt_type == RTN_MULTICAST) { 195 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len); 196 } else if (rt->rt_type == RTN_BROADCAST) 197 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len); 198 199 /* Be paranoid, rather than too clever. */ 200 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 201 struct sk_buff *skb2; 202 203 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 204 if (!skb2) { 205 kfree_skb(skb); 206 return -ENOMEM; 207 } 208 if (skb->sk) 209 skb_set_owner_w(skb2, skb->sk); 210 consume_skb(skb); 211 skb = skb2; 212 } 213 214 if (lwtunnel_xmit_redirect(dst->lwtstate)) { 215 int res = lwtunnel_xmit(skb); 216 217 if (res < 0 || res == LWTUNNEL_XMIT_DONE) 218 return res; 219 } 220 221 rcu_read_lock_bh(); 222 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 223 if (!IS_ERR(neigh)) { 224 int res; 225 226 sock_confirm_neigh(skb, neigh); 227 /* if crossing protocols, can not use the cached header */ 228 res = neigh_output(neigh, skb, is_v6gw); 229 rcu_read_unlock_bh(); 230 return res; 231 } 232 rcu_read_unlock_bh(); 233 234 net_dbg_ratelimited("%s: No header cache and no neighbour!\n", 235 __func__); 236 kfree_skb(skb); 237 return -EINVAL; 238 } 239 240 static int ip_finish_output_gso(struct net *net, struct sock *sk, 241 struct sk_buff *skb, unsigned int mtu) 242 { 243 struct sk_buff *segs, *nskb; 244 netdev_features_t features; 245 int ret = 0; 246 247 /* common case: seglen is <= mtu 248 */ 249 if (skb_gso_validate_network_len(skb, mtu)) 250 return ip_finish_output2(net, sk, skb); 251 252 /* Slowpath - GSO segment length exceeds the egress MTU. 253 * 254 * This can happen in several cases: 255 * - Forwarding of a TCP GRO skb, when DF flag is not set. 256 * - Forwarding of an skb that arrived on a virtualization interface 257 * (virtio-net/vhost/tap) with TSO/GSO size set by other network 258 * stack. 259 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an 260 * interface with a smaller MTU. 261 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is 262 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an 263 * insufficent MTU. 264 */ 265 features = netif_skb_features(skb); 266 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET); 267 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); 268 if (IS_ERR_OR_NULL(segs)) { 269 kfree_skb(skb); 270 return -ENOMEM; 271 } 272 273 consume_skb(skb); 274 275 skb_list_walk_safe(segs, segs, nskb) { 276 int err; 277 278 skb_mark_not_on_list(segs); 279 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2); 280 281 if (err && ret == 0) 282 ret = err; 283 } 284 285 return ret; 286 } 287 288 static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 289 { 290 unsigned int mtu; 291 292 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) 293 /* Policy lookup after SNAT yielded a new policy */ 294 if (skb_dst(skb)->xfrm) { 295 IPCB(skb)->flags |= IPSKB_REROUTED; 296 return dst_output(net, sk, skb); 297 } 298 #endif 299 mtu = ip_skb_dst_mtu(sk, skb); 300 if (skb_is_gso(skb)) 301 return ip_finish_output_gso(net, sk, skb, mtu); 302 303 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU)) 304 return ip_fragment(net, sk, skb, mtu, ip_finish_output2); 305 306 return ip_finish_output2(net, sk, skb); 307 } 308 309 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 310 { 311 int ret; 312 313 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); 314 switch (ret) { 315 case NET_XMIT_SUCCESS: 316 return __ip_finish_output(net, sk, skb); 317 case NET_XMIT_CN: 318 return __ip_finish_output(net, sk, skb) ? : ret; 319 default: 320 kfree_skb(skb); 321 return ret; 322 } 323 } 324 325 static int ip_mc_finish_output(struct net *net, struct sock *sk, 326 struct sk_buff *skb) 327 { 328 struct rtable *new_rt; 329 bool do_cn = false; 330 int ret, err; 331 332 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); 333 switch (ret) { 334 case NET_XMIT_CN: 335 do_cn = true; 336 fallthrough; 337 case NET_XMIT_SUCCESS: 338 break; 339 default: 340 kfree_skb(skb); 341 return ret; 342 } 343 344 /* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting 345 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten, 346 * see ipv4_pktinfo_prepare(). 347 */ 348 new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb)); 349 if (new_rt) { 350 new_rt->rt_iif = 0; 351 skb_dst_drop(skb); 352 skb_dst_set(skb, &new_rt->dst); 353 } 354 355 err = dev_loopback_xmit(net, sk, skb); 356 return (do_cn && err) ? ret : err; 357 } 358 359 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb) 360 { 361 struct rtable *rt = skb_rtable(skb); 362 struct net_device *dev = rt->dst.dev; 363 364 /* 365 * If the indicated interface is up and running, send the packet. 366 */ 367 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 368 369 skb->dev = dev; 370 skb->protocol = htons(ETH_P_IP); 371 372 /* 373 * Multicasts are looped back for other local users 374 */ 375 376 if (rt->rt_flags&RTCF_MULTICAST) { 377 if (sk_mc_loop(sk) 378 #ifdef CONFIG_IP_MROUTE 379 /* Small optimization: do not loopback not local frames, 380 which returned after forwarding; they will be dropped 381 by ip_mr_input in any case. 382 Note, that local frames are looped back to be delivered 383 to local recipients. 384 385 This check is duplicated in ip_mr_input at the moment. 386 */ 387 && 388 ((rt->rt_flags & RTCF_LOCAL) || 389 !(IPCB(skb)->flags & IPSKB_FORWARDED)) 390 #endif 391 ) { 392 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 393 if (newskb) 394 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 395 net, sk, newskb, NULL, newskb->dev, 396 ip_mc_finish_output); 397 } 398 399 /* Multicasts with ttl 0 must not go beyond the host */ 400 401 if (ip_hdr(skb)->ttl == 0) { 402 kfree_skb(skb); 403 return 0; 404 } 405 } 406 407 if (rt->rt_flags&RTCF_BROADCAST) { 408 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 409 if (newskb) 410 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 411 net, sk, newskb, NULL, newskb->dev, 412 ip_mc_finish_output); 413 } 414 415 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 416 net, sk, skb, NULL, skb->dev, 417 ip_finish_output, 418 !(IPCB(skb)->flags & IPSKB_REROUTED)); 419 } 420 421 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb) 422 { 423 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev; 424 425 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 426 427 skb->dev = dev; 428 skb->protocol = htons(ETH_P_IP); 429 430 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 431 net, sk, skb, indev, dev, 432 ip_finish_output, 433 !(IPCB(skb)->flags & IPSKB_REROUTED)); 434 } 435 436 /* 437 * copy saddr and daddr, possibly using 64bit load/stores 438 * Equivalent to : 439 * iph->saddr = fl4->saddr; 440 * iph->daddr = fl4->daddr; 441 */ 442 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) 443 { 444 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != 445 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); 446 memcpy(&iph->saddr, &fl4->saddr, 447 sizeof(fl4->saddr) + sizeof(fl4->daddr)); 448 } 449 450 /* Note: skb->sk can be different from sk, in case of tunnels */ 451 int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl, 452 __u8 tos) 453 { 454 struct inet_sock *inet = inet_sk(sk); 455 struct net *net = sock_net(sk); 456 struct ip_options_rcu *inet_opt; 457 struct flowi4 *fl4; 458 struct rtable *rt; 459 struct iphdr *iph; 460 int res; 461 462 /* Skip all of this if the packet is already routed, 463 * f.e. by something like SCTP. 464 */ 465 rcu_read_lock(); 466 inet_opt = rcu_dereference(inet->inet_opt); 467 fl4 = &fl->u.ip4; 468 rt = skb_rtable(skb); 469 if (rt) 470 goto packet_routed; 471 472 /* Make sure we can route this packet. */ 473 rt = (struct rtable *)__sk_dst_check(sk, 0); 474 if (!rt) { 475 __be32 daddr; 476 477 /* Use correct destination address if we have options. */ 478 daddr = inet->inet_daddr; 479 if (inet_opt && inet_opt->opt.srr) 480 daddr = inet_opt->opt.faddr; 481 482 /* If this fails, retransmit mechanism of transport layer will 483 * keep trying until route appears or the connection times 484 * itself out. 485 */ 486 rt = ip_route_output_ports(net, fl4, sk, 487 daddr, inet->inet_saddr, 488 inet->inet_dport, 489 inet->inet_sport, 490 sk->sk_protocol, 491 RT_CONN_FLAGS_TOS(sk, tos), 492 sk->sk_bound_dev_if); 493 if (IS_ERR(rt)) 494 goto no_route; 495 sk_setup_caps(sk, &rt->dst); 496 } 497 skb_dst_set_noref(skb, &rt->dst); 498 499 packet_routed: 500 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) 501 goto no_route; 502 503 /* OK, we know where to send it, allocate and build IP header. */ 504 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 505 skb_reset_network_header(skb); 506 iph = ip_hdr(skb); 507 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff)); 508 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) 509 iph->frag_off = htons(IP_DF); 510 else 511 iph->frag_off = 0; 512 iph->ttl = ip_select_ttl(inet, &rt->dst); 513 iph->protocol = sk->sk_protocol; 514 ip_copy_addrs(iph, fl4); 515 516 /* Transport layer set skb->h.foo itself. */ 517 518 if (inet_opt && inet_opt->opt.optlen) { 519 iph->ihl += inet_opt->opt.optlen >> 2; 520 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 521 } 522 523 ip_select_ident_segs(net, skb, sk, 524 skb_shinfo(skb)->gso_segs ?: 1); 525 526 /* TODO : should we use skb->sk here instead of sk ? */ 527 skb->priority = sk->sk_priority; 528 skb->mark = sk->sk_mark; 529 530 res = ip_local_out(net, sk, skb); 531 rcu_read_unlock(); 532 return res; 533 534 no_route: 535 rcu_read_unlock(); 536 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 537 kfree_skb(skb); 538 return -EHOSTUNREACH; 539 } 540 EXPORT_SYMBOL(__ip_queue_xmit); 541 542 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) 543 { 544 to->pkt_type = from->pkt_type; 545 to->priority = from->priority; 546 to->protocol = from->protocol; 547 to->skb_iif = from->skb_iif; 548 skb_dst_drop(to); 549 skb_dst_copy(to, from); 550 to->dev = from->dev; 551 to->mark = from->mark; 552 553 skb_copy_hash(to, from); 554 555 #ifdef CONFIG_NET_SCHED 556 to->tc_index = from->tc_index; 557 #endif 558 nf_copy(to, from); 559 skb_ext_copy(to, from); 560 #if IS_ENABLED(CONFIG_IP_VS) 561 to->ipvs_property = from->ipvs_property; 562 #endif 563 skb_copy_secmark(to, from); 564 } 565 566 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 567 unsigned int mtu, 568 int (*output)(struct net *, struct sock *, struct sk_buff *)) 569 { 570 struct iphdr *iph = ip_hdr(skb); 571 572 if ((iph->frag_off & htons(IP_DF)) == 0) 573 return ip_do_fragment(net, sk, skb, output); 574 575 if (unlikely(!skb->ignore_df || 576 (IPCB(skb)->frag_max_size && 577 IPCB(skb)->frag_max_size > mtu))) { 578 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 579 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 580 htonl(mtu)); 581 kfree_skb(skb); 582 return -EMSGSIZE; 583 } 584 585 return ip_do_fragment(net, sk, skb, output); 586 } 587 588 void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph, 589 unsigned int hlen, struct ip_fraglist_iter *iter) 590 { 591 unsigned int first_len = skb_pagelen(skb); 592 593 iter->frag = skb_shinfo(skb)->frag_list; 594 skb_frag_list_init(skb); 595 596 iter->offset = 0; 597 iter->iph = iph; 598 iter->hlen = hlen; 599 600 skb->data_len = first_len - skb_headlen(skb); 601 skb->len = first_len; 602 iph->tot_len = htons(first_len); 603 iph->frag_off = htons(IP_MF); 604 ip_send_check(iph); 605 } 606 EXPORT_SYMBOL(ip_fraglist_init); 607 608 static void ip_fraglist_ipcb_prepare(struct sk_buff *skb, 609 struct ip_fraglist_iter *iter) 610 { 611 struct sk_buff *to = iter->frag; 612 613 /* Copy the flags to each fragment. */ 614 IPCB(to)->flags = IPCB(skb)->flags; 615 616 if (iter->offset == 0) 617 ip_options_fragment(to); 618 } 619 620 void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter) 621 { 622 unsigned int hlen = iter->hlen; 623 struct iphdr *iph = iter->iph; 624 struct sk_buff *frag; 625 626 frag = iter->frag; 627 frag->ip_summed = CHECKSUM_NONE; 628 skb_reset_transport_header(frag); 629 __skb_push(frag, hlen); 630 skb_reset_network_header(frag); 631 memcpy(skb_network_header(frag), iph, hlen); 632 iter->iph = ip_hdr(frag); 633 iph = iter->iph; 634 iph->tot_len = htons(frag->len); 635 ip_copy_metadata(frag, skb); 636 iter->offset += skb->len - hlen; 637 iph->frag_off = htons(iter->offset >> 3); 638 if (frag->next) 639 iph->frag_off |= htons(IP_MF); 640 /* Ready, complete checksum */ 641 ip_send_check(iph); 642 } 643 EXPORT_SYMBOL(ip_fraglist_prepare); 644 645 void ip_frag_init(struct sk_buff *skb, unsigned int hlen, 646 unsigned int ll_rs, unsigned int mtu, bool DF, 647 struct ip_frag_state *state) 648 { 649 struct iphdr *iph = ip_hdr(skb); 650 651 state->DF = DF; 652 state->hlen = hlen; 653 state->ll_rs = ll_rs; 654 state->mtu = mtu; 655 656 state->left = skb->len - hlen; /* Space per frame */ 657 state->ptr = hlen; /* Where to start from */ 658 659 state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 660 state->not_last_frag = iph->frag_off & htons(IP_MF); 661 } 662 EXPORT_SYMBOL(ip_frag_init); 663 664 static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to, 665 bool first_frag, struct ip_frag_state *state) 666 { 667 /* Copy the flags to each fragment. */ 668 IPCB(to)->flags = IPCB(from)->flags; 669 670 /* ANK: dirty, but effective trick. Upgrade options only if 671 * the segment to be fragmented was THE FIRST (otherwise, 672 * options are already fixed) and make it ONCE 673 * on the initial skb, so that all the following fragments 674 * will inherit fixed options. 675 */ 676 if (first_frag) 677 ip_options_fragment(from); 678 } 679 680 struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state) 681 { 682 unsigned int len = state->left; 683 struct sk_buff *skb2; 684 struct iphdr *iph; 685 686 len = state->left; 687 /* IF: it doesn't fit, use 'mtu' - the data space left */ 688 if (len > state->mtu) 689 len = state->mtu; 690 /* IF: we are not sending up to and including the packet end 691 then align the next start on an eight byte boundary */ 692 if (len < state->left) { 693 len &= ~7; 694 } 695 696 /* Allocate buffer */ 697 skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC); 698 if (!skb2) 699 return ERR_PTR(-ENOMEM); 700 701 /* 702 * Set up data on packet 703 */ 704 705 ip_copy_metadata(skb2, skb); 706 skb_reserve(skb2, state->ll_rs); 707 skb_put(skb2, len + state->hlen); 708 skb_reset_network_header(skb2); 709 skb2->transport_header = skb2->network_header + state->hlen; 710 711 /* 712 * Charge the memory for the fragment to any owner 713 * it might possess 714 */ 715 716 if (skb->sk) 717 skb_set_owner_w(skb2, skb->sk); 718 719 /* 720 * Copy the packet header into the new buffer. 721 */ 722 723 skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen); 724 725 /* 726 * Copy a block of the IP datagram. 727 */ 728 if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len)) 729 BUG(); 730 state->left -= len; 731 732 /* 733 * Fill in the new header fields. 734 */ 735 iph = ip_hdr(skb2); 736 iph->frag_off = htons((state->offset >> 3)); 737 if (state->DF) 738 iph->frag_off |= htons(IP_DF); 739 740 /* 741 * Added AC : If we are fragmenting a fragment that's not the 742 * last fragment then keep MF on each bit 743 */ 744 if (state->left > 0 || state->not_last_frag) 745 iph->frag_off |= htons(IP_MF); 746 state->ptr += len; 747 state->offset += len; 748 749 iph->tot_len = htons(len + state->hlen); 750 751 ip_send_check(iph); 752 753 return skb2; 754 } 755 EXPORT_SYMBOL(ip_frag_next); 756 757 /* 758 * This IP datagram is too large to be sent in one piece. Break it up into 759 * smaller pieces (each of size equal to IP header plus 760 * a block of the data of the original IP data part) that will yet fit in a 761 * single device frame, and queue such a frame for sending. 762 */ 763 764 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 765 int (*output)(struct net *, struct sock *, struct sk_buff *)) 766 { 767 struct iphdr *iph; 768 struct sk_buff *skb2; 769 struct rtable *rt = skb_rtable(skb); 770 unsigned int mtu, hlen, ll_rs; 771 struct ip_fraglist_iter iter; 772 ktime_t tstamp = skb->tstamp; 773 struct ip_frag_state state; 774 int err = 0; 775 776 /* for offloaded checksums cleanup checksum before fragmentation */ 777 if (skb->ip_summed == CHECKSUM_PARTIAL && 778 (err = skb_checksum_help(skb))) 779 goto fail; 780 781 /* 782 * Point into the IP datagram header. 783 */ 784 785 iph = ip_hdr(skb); 786 787 mtu = ip_skb_dst_mtu(sk, skb); 788 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) 789 mtu = IPCB(skb)->frag_max_size; 790 791 /* 792 * Setup starting values. 793 */ 794 795 hlen = iph->ihl * 4; 796 mtu = mtu - hlen; /* Size of data space */ 797 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 798 ll_rs = LL_RESERVED_SPACE(rt->dst.dev); 799 800 /* When frag_list is given, use it. First, check its validity: 801 * some transformers could create wrong frag_list or break existing 802 * one, it is not prohibited. In this case fall back to copying. 803 * 804 * LATER: this step can be merged to real generation of fragments, 805 * we can switch to copy when see the first bad fragment. 806 */ 807 if (skb_has_frag_list(skb)) { 808 struct sk_buff *frag, *frag2; 809 unsigned int first_len = skb_pagelen(skb); 810 811 if (first_len - hlen > mtu || 812 ((first_len - hlen) & 7) || 813 ip_is_fragment(iph) || 814 skb_cloned(skb) || 815 skb_headroom(skb) < ll_rs) 816 goto slow_path; 817 818 skb_walk_frags(skb, frag) { 819 /* Correct geometry. */ 820 if (frag->len > mtu || 821 ((frag->len & 7) && frag->next) || 822 skb_headroom(frag) < hlen + ll_rs) 823 goto slow_path_clean; 824 825 /* Partially cloned skb? */ 826 if (skb_shared(frag)) 827 goto slow_path_clean; 828 829 BUG_ON(frag->sk); 830 if (skb->sk) { 831 frag->sk = skb->sk; 832 frag->destructor = sock_wfree; 833 } 834 skb->truesize -= frag->truesize; 835 } 836 837 /* Everything is OK. Generate! */ 838 ip_fraglist_init(skb, iph, hlen, &iter); 839 840 for (;;) { 841 /* Prepare header of the next frame, 842 * before previous one went down. */ 843 if (iter.frag) { 844 ip_fraglist_ipcb_prepare(skb, &iter); 845 ip_fraglist_prepare(skb, &iter); 846 } 847 848 skb->tstamp = tstamp; 849 err = output(net, sk, skb); 850 851 if (!err) 852 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 853 if (err || !iter.frag) 854 break; 855 856 skb = ip_fraglist_next(&iter); 857 } 858 859 if (err == 0) { 860 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 861 return 0; 862 } 863 864 kfree_skb_list(iter.frag); 865 866 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 867 return err; 868 869 slow_path_clean: 870 skb_walk_frags(skb, frag2) { 871 if (frag2 == frag) 872 break; 873 frag2->sk = NULL; 874 frag2->destructor = NULL; 875 skb->truesize += frag2->truesize; 876 } 877 } 878 879 slow_path: 880 /* 881 * Fragment the datagram. 882 */ 883 884 ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU, 885 &state); 886 887 /* 888 * Keep copying data until we run out. 889 */ 890 891 while (state.left > 0) { 892 bool first_frag = (state.offset == 0); 893 894 skb2 = ip_frag_next(skb, &state); 895 if (IS_ERR(skb2)) { 896 err = PTR_ERR(skb2); 897 goto fail; 898 } 899 ip_frag_ipcb(skb, skb2, first_frag, &state); 900 901 /* 902 * Put this fragment into the sending queue. 903 */ 904 skb2->tstamp = tstamp; 905 err = output(net, sk, skb2); 906 if (err) 907 goto fail; 908 909 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 910 } 911 consume_skb(skb); 912 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 913 return err; 914 915 fail: 916 kfree_skb(skb); 917 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 918 return err; 919 } 920 EXPORT_SYMBOL(ip_do_fragment); 921 922 int 923 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 924 { 925 struct msghdr *msg = from; 926 927 if (skb->ip_summed == CHECKSUM_PARTIAL) { 928 if (!copy_from_iter_full(to, len, &msg->msg_iter)) 929 return -EFAULT; 930 } else { 931 __wsum csum = 0; 932 if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter)) 933 return -EFAULT; 934 skb->csum = csum_block_add(skb->csum, csum, odd); 935 } 936 return 0; 937 } 938 EXPORT_SYMBOL(ip_generic_getfrag); 939 940 static inline __wsum 941 csum_page(struct page *page, int offset, int copy) 942 { 943 char *kaddr; 944 __wsum csum; 945 kaddr = kmap(page); 946 csum = csum_partial(kaddr + offset, copy, 0); 947 kunmap(page); 948 return csum; 949 } 950 951 static int __ip_append_data(struct sock *sk, 952 struct flowi4 *fl4, 953 struct sk_buff_head *queue, 954 struct inet_cork *cork, 955 struct page_frag *pfrag, 956 int getfrag(void *from, char *to, int offset, 957 int len, int odd, struct sk_buff *skb), 958 void *from, int length, int transhdrlen, 959 unsigned int flags) 960 { 961 struct inet_sock *inet = inet_sk(sk); 962 struct ubuf_info *uarg = NULL; 963 struct sk_buff *skb; 964 965 struct ip_options *opt = cork->opt; 966 int hh_len; 967 int exthdrlen; 968 int mtu; 969 int copy; 970 int err; 971 int offset = 0; 972 unsigned int maxfraglen, fragheaderlen, maxnonfragsize; 973 int csummode = CHECKSUM_NONE; 974 struct rtable *rt = (struct rtable *)cork->dst; 975 unsigned int wmem_alloc_delta = 0; 976 bool paged, extra_uref = false; 977 u32 tskey = 0; 978 979 skb = skb_peek_tail(queue); 980 981 exthdrlen = !skb ? rt->dst.header_len : 0; 982 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; 983 paged = !!cork->gso_size; 984 985 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && 986 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) 987 tskey = sk->sk_tskey++; 988 989 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 990 991 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 992 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 993 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 994 995 if (cork->length + length > maxnonfragsize - fragheaderlen) { 996 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 997 mtu - (opt ? opt->optlen : 0)); 998 return -EMSGSIZE; 999 } 1000 1001 /* 1002 * transhdrlen > 0 means that this is the first fragment and we wish 1003 * it won't be fragmented in the future. 1004 */ 1005 if (transhdrlen && 1006 length + fragheaderlen <= mtu && 1007 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) && 1008 (!(flags & MSG_MORE) || cork->gso_size) && 1009 (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM))) 1010 csummode = CHECKSUM_PARTIAL; 1011 1012 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) { 1013 uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb)); 1014 if (!uarg) 1015 return -ENOBUFS; 1016 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */ 1017 if (rt->dst.dev->features & NETIF_F_SG && 1018 csummode == CHECKSUM_PARTIAL) { 1019 paged = true; 1020 } else { 1021 uarg->zerocopy = 0; 1022 skb_zcopy_set(skb, uarg, &extra_uref); 1023 } 1024 } 1025 1026 cork->length += length; 1027 1028 /* So, what's going on in the loop below? 1029 * 1030 * We use calculated fragment length to generate chained skb, 1031 * each of segments is IP fragment ready for sending to network after 1032 * adding appropriate IP header. 1033 */ 1034 1035 if (!skb) 1036 goto alloc_new_skb; 1037 1038 while (length > 0) { 1039 /* Check if the remaining data fits into current packet. */ 1040 copy = mtu - skb->len; 1041 if (copy < length) 1042 copy = maxfraglen - skb->len; 1043 if (copy <= 0) { 1044 char *data; 1045 unsigned int datalen; 1046 unsigned int fraglen; 1047 unsigned int fraggap; 1048 unsigned int alloclen; 1049 unsigned int pagedlen; 1050 struct sk_buff *skb_prev; 1051 alloc_new_skb: 1052 skb_prev = skb; 1053 if (skb_prev) 1054 fraggap = skb_prev->len - maxfraglen; 1055 else 1056 fraggap = 0; 1057 1058 /* 1059 * If remaining data exceeds the mtu, 1060 * we know we need more fragment(s). 1061 */ 1062 datalen = length + fraggap; 1063 if (datalen > mtu - fragheaderlen) 1064 datalen = maxfraglen - fragheaderlen; 1065 fraglen = datalen + fragheaderlen; 1066 pagedlen = 0; 1067 1068 if ((flags & MSG_MORE) && 1069 !(rt->dst.dev->features&NETIF_F_SG)) 1070 alloclen = mtu; 1071 else if (!paged) 1072 alloclen = fraglen; 1073 else { 1074 alloclen = min_t(int, fraglen, MAX_HEADER); 1075 pagedlen = fraglen - alloclen; 1076 } 1077 1078 alloclen += exthdrlen; 1079 1080 /* The last fragment gets additional space at tail. 1081 * Note, with MSG_MORE we overallocate on fragments, 1082 * because we have no idea what fragment will be 1083 * the last. 1084 */ 1085 if (datalen == length + fraggap) 1086 alloclen += rt->dst.trailer_len; 1087 1088 if (transhdrlen) { 1089 skb = sock_alloc_send_skb(sk, 1090 alloclen + hh_len + 15, 1091 (flags & MSG_DONTWAIT), &err); 1092 } else { 1093 skb = NULL; 1094 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <= 1095 2 * sk->sk_sndbuf) 1096 skb = alloc_skb(alloclen + hh_len + 15, 1097 sk->sk_allocation); 1098 if (unlikely(!skb)) 1099 err = -ENOBUFS; 1100 } 1101 if (!skb) 1102 goto error; 1103 1104 /* 1105 * Fill in the control structures 1106 */ 1107 skb->ip_summed = csummode; 1108 skb->csum = 0; 1109 skb_reserve(skb, hh_len); 1110 1111 /* 1112 * Find where to start putting bytes. 1113 */ 1114 data = skb_put(skb, fraglen + exthdrlen - pagedlen); 1115 skb_set_network_header(skb, exthdrlen); 1116 skb->transport_header = (skb->network_header + 1117 fragheaderlen); 1118 data += fragheaderlen + exthdrlen; 1119 1120 if (fraggap) { 1121 skb->csum = skb_copy_and_csum_bits( 1122 skb_prev, maxfraglen, 1123 data + transhdrlen, fraggap, 0); 1124 skb_prev->csum = csum_sub(skb_prev->csum, 1125 skb->csum); 1126 data += fraggap; 1127 pskb_trim_unique(skb_prev, maxfraglen); 1128 } 1129 1130 copy = datalen - transhdrlen - fraggap - pagedlen; 1131 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 1132 err = -EFAULT; 1133 kfree_skb(skb); 1134 goto error; 1135 } 1136 1137 offset += copy; 1138 length -= copy + transhdrlen; 1139 transhdrlen = 0; 1140 exthdrlen = 0; 1141 csummode = CHECKSUM_NONE; 1142 1143 /* only the initial fragment is time stamped */ 1144 skb_shinfo(skb)->tx_flags = cork->tx_flags; 1145 cork->tx_flags = 0; 1146 skb_shinfo(skb)->tskey = tskey; 1147 tskey = 0; 1148 skb_zcopy_set(skb, uarg, &extra_uref); 1149 1150 if ((flags & MSG_CONFIRM) && !skb_prev) 1151 skb_set_dst_pending_confirm(skb, 1); 1152 1153 /* 1154 * Put the packet on the pending queue. 1155 */ 1156 if (!skb->destructor) { 1157 skb->destructor = sock_wfree; 1158 skb->sk = sk; 1159 wmem_alloc_delta += skb->truesize; 1160 } 1161 __skb_queue_tail(queue, skb); 1162 continue; 1163 } 1164 1165 if (copy > length) 1166 copy = length; 1167 1168 if (!(rt->dst.dev->features&NETIF_F_SG) && 1169 skb_tailroom(skb) >= copy) { 1170 unsigned int off; 1171 1172 off = skb->len; 1173 if (getfrag(from, skb_put(skb, copy), 1174 offset, copy, off, skb) < 0) { 1175 __skb_trim(skb, off); 1176 err = -EFAULT; 1177 goto error; 1178 } 1179 } else if (!uarg || !uarg->zerocopy) { 1180 int i = skb_shinfo(skb)->nr_frags; 1181 1182 err = -ENOMEM; 1183 if (!sk_page_frag_refill(sk, pfrag)) 1184 goto error; 1185 1186 if (!skb_can_coalesce(skb, i, pfrag->page, 1187 pfrag->offset)) { 1188 err = -EMSGSIZE; 1189 if (i == MAX_SKB_FRAGS) 1190 goto error; 1191 1192 __skb_fill_page_desc(skb, i, pfrag->page, 1193 pfrag->offset, 0); 1194 skb_shinfo(skb)->nr_frags = ++i; 1195 get_page(pfrag->page); 1196 } 1197 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1198 if (getfrag(from, 1199 page_address(pfrag->page) + pfrag->offset, 1200 offset, copy, skb->len, skb) < 0) 1201 goto error_efault; 1202 1203 pfrag->offset += copy; 1204 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1205 skb->len += copy; 1206 skb->data_len += copy; 1207 skb->truesize += copy; 1208 wmem_alloc_delta += copy; 1209 } else { 1210 err = skb_zerocopy_iter_dgram(skb, from, copy); 1211 if (err < 0) 1212 goto error; 1213 } 1214 offset += copy; 1215 length -= copy; 1216 } 1217 1218 if (wmem_alloc_delta) 1219 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1220 return 0; 1221 1222 error_efault: 1223 err = -EFAULT; 1224 error: 1225 if (uarg) 1226 sock_zerocopy_put_abort(uarg, extra_uref); 1227 cork->length -= length; 1228 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1229 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1230 return err; 1231 } 1232 1233 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 1234 struct ipcm_cookie *ipc, struct rtable **rtp) 1235 { 1236 struct ip_options_rcu *opt; 1237 struct rtable *rt; 1238 1239 rt = *rtp; 1240 if (unlikely(!rt)) 1241 return -EFAULT; 1242 1243 /* 1244 * setup for corking. 1245 */ 1246 opt = ipc->opt; 1247 if (opt) { 1248 if (!cork->opt) { 1249 cork->opt = kmalloc(sizeof(struct ip_options) + 40, 1250 sk->sk_allocation); 1251 if (unlikely(!cork->opt)) 1252 return -ENOBUFS; 1253 } 1254 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 1255 cork->flags |= IPCORK_OPT; 1256 cork->addr = ipc->addr; 1257 } 1258 1259 cork->fragsize = ip_sk_use_pmtu(sk) ? 1260 dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu); 1261 1262 if (!inetdev_valid_mtu(cork->fragsize)) 1263 return -ENETUNREACH; 1264 1265 cork->gso_size = ipc->gso_size; 1266 1267 cork->dst = &rt->dst; 1268 /* We stole this route, caller should not release it. */ 1269 *rtp = NULL; 1270 1271 cork->length = 0; 1272 cork->ttl = ipc->ttl; 1273 cork->tos = ipc->tos; 1274 cork->mark = ipc->sockc.mark; 1275 cork->priority = ipc->priority; 1276 cork->transmit_time = ipc->sockc.transmit_time; 1277 cork->tx_flags = 0; 1278 sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags); 1279 1280 return 0; 1281 } 1282 1283 /* 1284 * ip_append_data() and ip_append_page() can make one large IP datagram 1285 * from many pieces of data. Each pieces will be holded on the socket 1286 * until ip_push_pending_frames() is called. Each piece can be a page 1287 * or non-page data. 1288 * 1289 * Not only UDP, other transport protocols - e.g. raw sockets - can use 1290 * this interface potentially. 1291 * 1292 * LATER: length must be adjusted by pad at tail, when it is required. 1293 */ 1294 int ip_append_data(struct sock *sk, struct flowi4 *fl4, 1295 int getfrag(void *from, char *to, int offset, int len, 1296 int odd, struct sk_buff *skb), 1297 void *from, int length, int transhdrlen, 1298 struct ipcm_cookie *ipc, struct rtable **rtp, 1299 unsigned int flags) 1300 { 1301 struct inet_sock *inet = inet_sk(sk); 1302 int err; 1303 1304 if (flags&MSG_PROBE) 1305 return 0; 1306 1307 if (skb_queue_empty(&sk->sk_write_queue)) { 1308 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 1309 if (err) 1310 return err; 1311 } else { 1312 transhdrlen = 0; 1313 } 1314 1315 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, 1316 sk_page_frag(sk), getfrag, 1317 from, length, transhdrlen, flags); 1318 } 1319 1320 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 1321 int offset, size_t size, int flags) 1322 { 1323 struct inet_sock *inet = inet_sk(sk); 1324 struct sk_buff *skb; 1325 struct rtable *rt; 1326 struct ip_options *opt = NULL; 1327 struct inet_cork *cork; 1328 int hh_len; 1329 int mtu; 1330 int len; 1331 int err; 1332 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; 1333 1334 if (inet->hdrincl) 1335 return -EPERM; 1336 1337 if (flags&MSG_PROBE) 1338 return 0; 1339 1340 if (skb_queue_empty(&sk->sk_write_queue)) 1341 return -EINVAL; 1342 1343 cork = &inet->cork.base; 1344 rt = (struct rtable *)cork->dst; 1345 if (cork->flags & IPCORK_OPT) 1346 opt = cork->opt; 1347 1348 if (!(rt->dst.dev->features&NETIF_F_SG)) 1349 return -EOPNOTSUPP; 1350 1351 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1352 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; 1353 1354 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1355 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1356 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 1357 1358 if (cork->length + size > maxnonfragsize - fragheaderlen) { 1359 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 1360 mtu - (opt ? opt->optlen : 0)); 1361 return -EMSGSIZE; 1362 } 1363 1364 skb = skb_peek_tail(&sk->sk_write_queue); 1365 if (!skb) 1366 return -EINVAL; 1367 1368 cork->length += size; 1369 1370 while (size > 0) { 1371 /* Check if the remaining data fits into current packet. */ 1372 len = mtu - skb->len; 1373 if (len < size) 1374 len = maxfraglen - skb->len; 1375 1376 if (len <= 0) { 1377 struct sk_buff *skb_prev; 1378 int alloclen; 1379 1380 skb_prev = skb; 1381 fraggap = skb_prev->len - maxfraglen; 1382 1383 alloclen = fragheaderlen + hh_len + fraggap + 15; 1384 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 1385 if (unlikely(!skb)) { 1386 err = -ENOBUFS; 1387 goto error; 1388 } 1389 1390 /* 1391 * Fill in the control structures 1392 */ 1393 skb->ip_summed = CHECKSUM_NONE; 1394 skb->csum = 0; 1395 skb_reserve(skb, hh_len); 1396 1397 /* 1398 * Find where to start putting bytes. 1399 */ 1400 skb_put(skb, fragheaderlen + fraggap); 1401 skb_reset_network_header(skb); 1402 skb->transport_header = (skb->network_header + 1403 fragheaderlen); 1404 if (fraggap) { 1405 skb->csum = skb_copy_and_csum_bits(skb_prev, 1406 maxfraglen, 1407 skb_transport_header(skb), 1408 fraggap, 0); 1409 skb_prev->csum = csum_sub(skb_prev->csum, 1410 skb->csum); 1411 pskb_trim_unique(skb_prev, maxfraglen); 1412 } 1413 1414 /* 1415 * Put the packet on the pending queue. 1416 */ 1417 __skb_queue_tail(&sk->sk_write_queue, skb); 1418 continue; 1419 } 1420 1421 if (len > size) 1422 len = size; 1423 1424 if (skb_append_pagefrags(skb, page, offset, len)) { 1425 err = -EMSGSIZE; 1426 goto error; 1427 } 1428 1429 if (skb->ip_summed == CHECKSUM_NONE) { 1430 __wsum csum; 1431 csum = csum_page(page, offset, len); 1432 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1433 } 1434 1435 skb->len += len; 1436 skb->data_len += len; 1437 skb->truesize += len; 1438 refcount_add(len, &sk->sk_wmem_alloc); 1439 offset += len; 1440 size -= len; 1441 } 1442 return 0; 1443 1444 error: 1445 cork->length -= size; 1446 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1447 return err; 1448 } 1449 1450 static void ip_cork_release(struct inet_cork *cork) 1451 { 1452 cork->flags &= ~IPCORK_OPT; 1453 kfree(cork->opt); 1454 cork->opt = NULL; 1455 dst_release(cork->dst); 1456 cork->dst = NULL; 1457 } 1458 1459 /* 1460 * Combined all pending IP fragments on the socket as one IP datagram 1461 * and push them out. 1462 */ 1463 struct sk_buff *__ip_make_skb(struct sock *sk, 1464 struct flowi4 *fl4, 1465 struct sk_buff_head *queue, 1466 struct inet_cork *cork) 1467 { 1468 struct sk_buff *skb, *tmp_skb; 1469 struct sk_buff **tail_skb; 1470 struct inet_sock *inet = inet_sk(sk); 1471 struct net *net = sock_net(sk); 1472 struct ip_options *opt = NULL; 1473 struct rtable *rt = (struct rtable *)cork->dst; 1474 struct iphdr *iph; 1475 __be16 df = 0; 1476 __u8 ttl; 1477 1478 skb = __skb_dequeue(queue); 1479 if (!skb) 1480 goto out; 1481 tail_skb = &(skb_shinfo(skb)->frag_list); 1482 1483 /* move skb->data to ip header from ext header */ 1484 if (skb->data < skb_network_header(skb)) 1485 __skb_pull(skb, skb_network_offset(skb)); 1486 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1487 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1488 *tail_skb = tmp_skb; 1489 tail_skb = &(tmp_skb->next); 1490 skb->len += tmp_skb->len; 1491 skb->data_len += tmp_skb->len; 1492 skb->truesize += tmp_skb->truesize; 1493 tmp_skb->destructor = NULL; 1494 tmp_skb->sk = NULL; 1495 } 1496 1497 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 1498 * to fragment the frame generated here. No matter, what transforms 1499 * how transforms change size of the packet, it will come out. 1500 */ 1501 skb->ignore_df = ip_sk_ignore_df(sk); 1502 1503 /* DF bit is set when we want to see DF on outgoing frames. 1504 * If ignore_df is set too, we still allow to fragment this frame 1505 * locally. */ 1506 if (inet->pmtudisc == IP_PMTUDISC_DO || 1507 inet->pmtudisc == IP_PMTUDISC_PROBE || 1508 (skb->len <= dst_mtu(&rt->dst) && 1509 ip_dont_fragment(sk, &rt->dst))) 1510 df = htons(IP_DF); 1511 1512 if (cork->flags & IPCORK_OPT) 1513 opt = cork->opt; 1514 1515 if (cork->ttl != 0) 1516 ttl = cork->ttl; 1517 else if (rt->rt_type == RTN_MULTICAST) 1518 ttl = inet->mc_ttl; 1519 else 1520 ttl = ip_select_ttl(inet, &rt->dst); 1521 1522 iph = ip_hdr(skb); 1523 iph->version = 4; 1524 iph->ihl = 5; 1525 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; 1526 iph->frag_off = df; 1527 iph->ttl = ttl; 1528 iph->protocol = sk->sk_protocol; 1529 ip_copy_addrs(iph, fl4); 1530 ip_select_ident(net, skb, sk); 1531 1532 if (opt) { 1533 iph->ihl += opt->optlen>>2; 1534 ip_options_build(skb, opt, cork->addr, rt, 0); 1535 } 1536 1537 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; 1538 skb->mark = cork->mark; 1539 skb->tstamp = cork->transmit_time; 1540 /* 1541 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1542 * on dst refcount 1543 */ 1544 cork->dst = NULL; 1545 skb_dst_set(skb, &rt->dst); 1546 1547 if (iph->protocol == IPPROTO_ICMP) 1548 icmp_out_count(net, ((struct icmphdr *) 1549 skb_transport_header(skb))->type); 1550 1551 ip_cork_release(cork); 1552 out: 1553 return skb; 1554 } 1555 1556 int ip_send_skb(struct net *net, struct sk_buff *skb) 1557 { 1558 int err; 1559 1560 err = ip_local_out(net, skb->sk, skb); 1561 if (err) { 1562 if (err > 0) 1563 err = net_xmit_errno(err); 1564 if (err) 1565 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 1566 } 1567 1568 return err; 1569 } 1570 1571 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 1572 { 1573 struct sk_buff *skb; 1574 1575 skb = ip_finish_skb(sk, fl4); 1576 if (!skb) 1577 return 0; 1578 1579 /* Netfilter gets whole the not fragmented skb. */ 1580 return ip_send_skb(sock_net(sk), skb); 1581 } 1582 1583 /* 1584 * Throw away all pending data on the socket. 1585 */ 1586 static void __ip_flush_pending_frames(struct sock *sk, 1587 struct sk_buff_head *queue, 1588 struct inet_cork *cork) 1589 { 1590 struct sk_buff *skb; 1591 1592 while ((skb = __skb_dequeue_tail(queue)) != NULL) 1593 kfree_skb(skb); 1594 1595 ip_cork_release(cork); 1596 } 1597 1598 void ip_flush_pending_frames(struct sock *sk) 1599 { 1600 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 1601 } 1602 1603 struct sk_buff *ip_make_skb(struct sock *sk, 1604 struct flowi4 *fl4, 1605 int getfrag(void *from, char *to, int offset, 1606 int len, int odd, struct sk_buff *skb), 1607 void *from, int length, int transhdrlen, 1608 struct ipcm_cookie *ipc, struct rtable **rtp, 1609 struct inet_cork *cork, unsigned int flags) 1610 { 1611 struct sk_buff_head queue; 1612 int err; 1613 1614 if (flags & MSG_PROBE) 1615 return NULL; 1616 1617 __skb_queue_head_init(&queue); 1618 1619 cork->flags = 0; 1620 cork->addr = 0; 1621 cork->opt = NULL; 1622 err = ip_setup_cork(sk, cork, ipc, rtp); 1623 if (err) 1624 return ERR_PTR(err); 1625 1626 err = __ip_append_data(sk, fl4, &queue, cork, 1627 ¤t->task_frag, getfrag, 1628 from, length, transhdrlen, flags); 1629 if (err) { 1630 __ip_flush_pending_frames(sk, &queue, cork); 1631 return ERR_PTR(err); 1632 } 1633 1634 return __ip_make_skb(sk, fl4, &queue, cork); 1635 } 1636 1637 /* 1638 * Fetch data from kernel space and fill in checksum if needed. 1639 */ 1640 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 1641 int len, int odd, struct sk_buff *skb) 1642 { 1643 __wsum csum; 1644 1645 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); 1646 skb->csum = csum_block_add(skb->csum, csum, odd); 1647 return 0; 1648 } 1649 1650 /* 1651 * Generic function to send a packet as reply to another packet. 1652 * Used to send some TCP resets/acks so far. 1653 */ 1654 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, 1655 const struct ip_options *sopt, 1656 __be32 daddr, __be32 saddr, 1657 const struct ip_reply_arg *arg, 1658 unsigned int len, u64 transmit_time) 1659 { 1660 struct ip_options_data replyopts; 1661 struct ipcm_cookie ipc; 1662 struct flowi4 fl4; 1663 struct rtable *rt = skb_rtable(skb); 1664 struct net *net = sock_net(sk); 1665 struct sk_buff *nskb; 1666 int err; 1667 int oif; 1668 1669 if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt)) 1670 return; 1671 1672 ipcm_init(&ipc); 1673 ipc.addr = daddr; 1674 ipc.sockc.transmit_time = transmit_time; 1675 1676 if (replyopts.opt.opt.optlen) { 1677 ipc.opt = &replyopts.opt; 1678 1679 if (replyopts.opt.opt.srr) 1680 daddr = replyopts.opt.opt.faddr; 1681 } 1682 1683 oif = arg->bound_dev_if; 1684 if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) 1685 oif = skb->skb_iif; 1686 1687 flowi4_init_output(&fl4, oif, 1688 IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark, 1689 RT_TOS(arg->tos), 1690 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, 1691 ip_reply_arg_flowi_flags(arg), 1692 daddr, saddr, 1693 tcp_hdr(skb)->source, tcp_hdr(skb)->dest, 1694 arg->uid); 1695 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 1696 rt = ip_route_output_key(net, &fl4); 1697 if (IS_ERR(rt)) 1698 return; 1699 1700 inet_sk(sk)->tos = arg->tos; 1701 1702 sk->sk_protocol = ip_hdr(skb)->protocol; 1703 sk->sk_bound_dev_if = arg->bound_dev_if; 1704 sk->sk_sndbuf = sysctl_wmem_default; 1705 sk->sk_mark = fl4.flowi4_mark; 1706 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, 1707 len, 0, &ipc, &rt, MSG_DONTWAIT); 1708 if (unlikely(err)) { 1709 ip_flush_pending_frames(sk); 1710 goto out; 1711 } 1712 1713 nskb = skb_peek(&sk->sk_write_queue); 1714 if (nskb) { 1715 if (arg->csumoffset >= 0) 1716 *((__sum16 *)skb_transport_header(nskb) + 1717 arg->csumoffset) = csum_fold(csum_add(nskb->csum, 1718 arg->csum)); 1719 nskb->ip_summed = CHECKSUM_NONE; 1720 ip_push_pending_frames(sk, &fl4); 1721 } 1722 out: 1723 ip_rt_put(rt); 1724 } 1725 1726 void __init ip_init(void) 1727 { 1728 ip_rt_init(); 1729 inet_initpeers(); 1730 1731 #if defined(CONFIG_IP_MULTICAST) 1732 igmp_mc_init(); 1733 #endif 1734 } 1735