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/inet_ecn.h> 78 #include <net/lwtunnel.h> 79 #include <linux/bpf-cgroup.h> 80 #include <linux/igmp.h> 81 #include <linux/netfilter_ipv4.h> 82 #include <linux/netfilter_bridge.h> 83 #include <linux/netlink.h> 84 #include <linux/tcp.h> 85 86 static int 87 ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 88 unsigned int mtu, 89 int (*output)(struct net *, struct sock *, struct sk_buff *)); 90 91 /* Generate a checksum for an outgoing IP datagram. */ 92 void ip_send_check(struct iphdr *iph) 93 { 94 iph->check = 0; 95 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 96 } 97 EXPORT_SYMBOL(ip_send_check); 98 99 int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 100 { 101 struct iphdr *iph = ip_hdr(skb); 102 103 iph->tot_len = htons(skb->len); 104 ip_send_check(iph); 105 106 /* if egress device is enslaved to an L3 master device pass the 107 * skb to its handler for processing 108 */ 109 skb = l3mdev_ip_out(sk, skb); 110 if (unlikely(!skb)) 111 return 0; 112 113 skb->protocol = htons(ETH_P_IP); 114 115 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, 116 net, sk, skb, NULL, skb_dst(skb)->dev, 117 dst_output); 118 } 119 120 int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) 121 { 122 int err; 123 124 err = __ip_local_out(net, sk, skb); 125 if (likely(err == 1)) 126 err = dst_output(net, sk, skb); 127 128 return err; 129 } 130 EXPORT_SYMBOL_GPL(ip_local_out); 131 132 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) 133 { 134 int ttl = inet->uc_ttl; 135 136 if (ttl < 0) 137 ttl = ip4_dst_hoplimit(dst); 138 return ttl; 139 } 140 141 /* 142 * Add an ip header to a skbuff and send it out. 143 * 144 */ 145 int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk, 146 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt, 147 u8 tos) 148 { 149 struct inet_sock *inet = inet_sk(sk); 150 struct rtable *rt = skb_rtable(skb); 151 struct net *net = sock_net(sk); 152 struct iphdr *iph; 153 154 /* Build the IP header. */ 155 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); 156 skb_reset_network_header(skb); 157 iph = ip_hdr(skb); 158 iph->version = 4; 159 iph->ihl = 5; 160 iph->tos = tos; 161 iph->ttl = ip_select_ttl(inet, &rt->dst); 162 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); 163 iph->saddr = saddr; 164 iph->protocol = sk->sk_protocol; 165 if (ip_dont_fragment(sk, &rt->dst)) { 166 iph->frag_off = htons(IP_DF); 167 iph->id = 0; 168 } else { 169 iph->frag_off = 0; 170 __ip_select_ident(net, iph, 1); 171 } 172 173 if (opt && opt->opt.optlen) { 174 iph->ihl += opt->opt.optlen>>2; 175 ip_options_build(skb, &opt->opt, daddr, rt, 0); 176 } 177 178 skb->priority = sk->sk_priority; 179 if (!skb->mark) 180 skb->mark = sk->sk_mark; 181 182 /* Send it out. */ 183 return ip_local_out(net, skb->sk, skb); 184 } 185 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); 186 187 static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) 188 { 189 struct dst_entry *dst = skb_dst(skb); 190 struct rtable *rt = (struct rtable *)dst; 191 struct net_device *dev = dst->dev; 192 unsigned int hh_len = LL_RESERVED_SPACE(dev); 193 struct neighbour *neigh; 194 bool is_v6gw = false; 195 196 if (rt->rt_type == RTN_MULTICAST) { 197 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len); 198 } else if (rt->rt_type == RTN_BROADCAST) 199 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len); 200 201 /* Be paranoid, rather than too clever. */ 202 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 203 struct sk_buff *skb2; 204 205 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 206 if (!skb2) { 207 kfree_skb(skb); 208 return -ENOMEM; 209 } 210 if (skb->sk) 211 skb_set_owner_w(skb2, skb->sk); 212 consume_skb(skb); 213 skb = skb2; 214 } 215 216 if (lwtunnel_xmit_redirect(dst->lwtstate)) { 217 int res = lwtunnel_xmit(skb); 218 219 if (res < 0 || res == LWTUNNEL_XMIT_DONE) 220 return res; 221 } 222 223 rcu_read_lock_bh(); 224 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 225 if (!IS_ERR(neigh)) { 226 int res; 227 228 sock_confirm_neigh(skb, neigh); 229 /* if crossing protocols, can not use the cached header */ 230 res = neigh_output(neigh, skb, is_v6gw); 231 rcu_read_unlock_bh(); 232 return res; 233 } 234 rcu_read_unlock_bh(); 235 236 net_dbg_ratelimited("%s: No header cache and no neighbour!\n", 237 __func__); 238 kfree_skb(skb); 239 return -EINVAL; 240 } 241 242 static int ip_finish_output_gso(struct net *net, struct sock *sk, 243 struct sk_buff *skb, unsigned int mtu) 244 { 245 struct sk_buff *segs, *nskb; 246 netdev_features_t features; 247 int ret = 0; 248 249 /* common case: seglen is <= mtu 250 */ 251 if (skb_gso_validate_network_len(skb, mtu)) 252 return ip_finish_output2(net, sk, skb); 253 254 /* Slowpath - GSO segment length exceeds the egress MTU. 255 * 256 * This can happen in several cases: 257 * - Forwarding of a TCP GRO skb, when DF flag is not set. 258 * - Forwarding of an skb that arrived on a virtualization interface 259 * (virtio-net/vhost/tap) with TSO/GSO size set by other network 260 * stack. 261 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an 262 * interface with a smaller MTU. 263 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is 264 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an 265 * insufficent MTU. 266 */ 267 features = netif_skb_features(skb); 268 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET); 269 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); 270 if (IS_ERR_OR_NULL(segs)) { 271 kfree_skb(skb); 272 return -ENOMEM; 273 } 274 275 consume_skb(skb); 276 277 skb_list_walk_safe(segs, segs, nskb) { 278 int err; 279 280 skb_mark_not_on_list(segs); 281 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2); 282 283 if (err && ret == 0) 284 ret = err; 285 } 286 287 return ret; 288 } 289 290 static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 291 { 292 unsigned int mtu; 293 294 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) 295 /* Policy lookup after SNAT yielded a new policy */ 296 if (skb_dst(skb)->xfrm) { 297 IPCB(skb)->flags |= IPSKB_REROUTED; 298 return dst_output(net, sk, skb); 299 } 300 #endif 301 mtu = ip_skb_dst_mtu(sk, skb); 302 if (skb_is_gso(skb)) 303 return ip_finish_output_gso(net, sk, skb, mtu); 304 305 if (skb->len > mtu || IPCB(skb)->frag_max_size) 306 return ip_fragment(net, sk, skb, mtu, ip_finish_output2); 307 308 return ip_finish_output2(net, sk, skb); 309 } 310 311 static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 312 { 313 int ret; 314 315 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); 316 switch (ret) { 317 case NET_XMIT_SUCCESS: 318 return __ip_finish_output(net, sk, skb); 319 case NET_XMIT_CN: 320 return __ip_finish_output(net, sk, skb) ? : ret; 321 default: 322 kfree_skb(skb); 323 return ret; 324 } 325 } 326 327 static int ip_mc_finish_output(struct net *net, struct sock *sk, 328 struct sk_buff *skb) 329 { 330 struct rtable *new_rt; 331 bool do_cn = false; 332 int ret, err; 333 334 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); 335 switch (ret) { 336 case NET_XMIT_CN: 337 do_cn = true; 338 fallthrough; 339 case NET_XMIT_SUCCESS: 340 break; 341 default: 342 kfree_skb(skb); 343 return ret; 344 } 345 346 /* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting 347 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten, 348 * see ipv4_pktinfo_prepare(). 349 */ 350 new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb)); 351 if (new_rt) { 352 new_rt->rt_iif = 0; 353 skb_dst_drop(skb); 354 skb_dst_set(skb, &new_rt->dst); 355 } 356 357 err = dev_loopback_xmit(net, sk, skb); 358 return (do_cn && err) ? ret : err; 359 } 360 361 int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb) 362 { 363 struct rtable *rt = skb_rtable(skb); 364 struct net_device *dev = rt->dst.dev; 365 366 /* 367 * If the indicated interface is up and running, send the packet. 368 */ 369 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 370 371 skb->dev = dev; 372 skb->protocol = htons(ETH_P_IP); 373 374 /* 375 * Multicasts are looped back for other local users 376 */ 377 378 if (rt->rt_flags&RTCF_MULTICAST) { 379 if (sk_mc_loop(sk) 380 #ifdef CONFIG_IP_MROUTE 381 /* Small optimization: do not loopback not local frames, 382 which returned after forwarding; they will be dropped 383 by ip_mr_input in any case. 384 Note, that local frames are looped back to be delivered 385 to local recipients. 386 387 This check is duplicated in ip_mr_input at the moment. 388 */ 389 && 390 ((rt->rt_flags & RTCF_LOCAL) || 391 !(IPCB(skb)->flags & IPSKB_FORWARDED)) 392 #endif 393 ) { 394 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 395 if (newskb) 396 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 397 net, sk, newskb, NULL, newskb->dev, 398 ip_mc_finish_output); 399 } 400 401 /* Multicasts with ttl 0 must not go beyond the host */ 402 403 if (ip_hdr(skb)->ttl == 0) { 404 kfree_skb(skb); 405 return 0; 406 } 407 } 408 409 if (rt->rt_flags&RTCF_BROADCAST) { 410 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 411 if (newskb) 412 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 413 net, sk, newskb, NULL, newskb->dev, 414 ip_mc_finish_output); 415 } 416 417 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 418 net, sk, skb, NULL, skb->dev, 419 ip_finish_output, 420 !(IPCB(skb)->flags & IPSKB_REROUTED)); 421 } 422 423 int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb) 424 { 425 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev; 426 427 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); 428 429 skb->dev = dev; 430 skb->protocol = htons(ETH_P_IP); 431 432 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, 433 net, sk, skb, indev, dev, 434 ip_finish_output, 435 !(IPCB(skb)->flags & IPSKB_REROUTED)); 436 } 437 438 /* 439 * copy saddr and daddr, possibly using 64bit load/stores 440 * Equivalent to : 441 * iph->saddr = fl4->saddr; 442 * iph->daddr = fl4->daddr; 443 */ 444 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) 445 { 446 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != 447 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); 448 memcpy(&iph->saddr, &fl4->saddr, 449 sizeof(fl4->saddr) + sizeof(fl4->daddr)); 450 } 451 452 /* Note: skb->sk can be different from sk, in case of tunnels */ 453 int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl, 454 __u8 tos) 455 { 456 struct inet_sock *inet = inet_sk(sk); 457 struct net *net = sock_net(sk); 458 struct ip_options_rcu *inet_opt; 459 struct flowi4 *fl4; 460 struct rtable *rt; 461 struct iphdr *iph; 462 int res; 463 464 /* Skip all of this if the packet is already routed, 465 * f.e. by something like SCTP. 466 */ 467 rcu_read_lock(); 468 inet_opt = rcu_dereference(inet->inet_opt); 469 fl4 = &fl->u.ip4; 470 rt = skb_rtable(skb); 471 if (rt) 472 goto packet_routed; 473 474 /* Make sure we can route this packet. */ 475 rt = (struct rtable *)__sk_dst_check(sk, 0); 476 if (!rt) { 477 __be32 daddr; 478 479 /* Use correct destination address if we have options. */ 480 daddr = inet->inet_daddr; 481 if (inet_opt && inet_opt->opt.srr) 482 daddr = inet_opt->opt.faddr; 483 484 /* If this fails, retransmit mechanism of transport layer will 485 * keep trying until route appears or the connection times 486 * itself out. 487 */ 488 rt = ip_route_output_ports(net, fl4, sk, 489 daddr, inet->inet_saddr, 490 inet->inet_dport, 491 inet->inet_sport, 492 sk->sk_protocol, 493 RT_CONN_FLAGS_TOS(sk, tos), 494 sk->sk_bound_dev_if); 495 if (IS_ERR(rt)) 496 goto no_route; 497 sk_setup_caps(sk, &rt->dst); 498 } 499 skb_dst_set_noref(skb, &rt->dst); 500 501 packet_routed: 502 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) 503 goto no_route; 504 505 /* OK, we know where to send it, allocate and build IP header. */ 506 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 507 skb_reset_network_header(skb); 508 iph = ip_hdr(skb); 509 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff)); 510 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) 511 iph->frag_off = htons(IP_DF); 512 else 513 iph->frag_off = 0; 514 iph->ttl = ip_select_ttl(inet, &rt->dst); 515 iph->protocol = sk->sk_protocol; 516 ip_copy_addrs(iph, fl4); 517 518 /* Transport layer set skb->h.foo itself. */ 519 520 if (inet_opt && inet_opt->opt.optlen) { 521 iph->ihl += inet_opt->opt.optlen >> 2; 522 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 523 } 524 525 ip_select_ident_segs(net, skb, sk, 526 skb_shinfo(skb)->gso_segs ?: 1); 527 528 /* TODO : should we use skb->sk here instead of sk ? */ 529 skb->priority = sk->sk_priority; 530 skb->mark = sk->sk_mark; 531 532 res = ip_local_out(net, sk, skb); 533 rcu_read_unlock(); 534 return res; 535 536 no_route: 537 rcu_read_unlock(); 538 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 539 kfree_skb(skb); 540 return -EHOSTUNREACH; 541 } 542 EXPORT_SYMBOL(__ip_queue_xmit); 543 544 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) 545 { 546 return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos); 547 } 548 EXPORT_SYMBOL(ip_queue_xmit); 549 550 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) 551 { 552 to->pkt_type = from->pkt_type; 553 to->priority = from->priority; 554 to->protocol = from->protocol; 555 to->skb_iif = from->skb_iif; 556 skb_dst_drop(to); 557 skb_dst_copy(to, from); 558 to->dev = from->dev; 559 to->mark = from->mark; 560 561 skb_copy_hash(to, from); 562 563 #ifdef CONFIG_NET_SCHED 564 to->tc_index = from->tc_index; 565 #endif 566 nf_copy(to, from); 567 skb_ext_copy(to, from); 568 #if IS_ENABLED(CONFIG_IP_VS) 569 to->ipvs_property = from->ipvs_property; 570 #endif 571 skb_copy_secmark(to, from); 572 } 573 574 static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 575 unsigned int mtu, 576 int (*output)(struct net *, struct sock *, struct sk_buff *)) 577 { 578 struct iphdr *iph = ip_hdr(skb); 579 580 if ((iph->frag_off & htons(IP_DF)) == 0) 581 return ip_do_fragment(net, sk, skb, output); 582 583 if (unlikely(!skb->ignore_df || 584 (IPCB(skb)->frag_max_size && 585 IPCB(skb)->frag_max_size > mtu))) { 586 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 587 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 588 htonl(mtu)); 589 kfree_skb(skb); 590 return -EMSGSIZE; 591 } 592 593 return ip_do_fragment(net, sk, skb, output); 594 } 595 596 void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph, 597 unsigned int hlen, struct ip_fraglist_iter *iter) 598 { 599 unsigned int first_len = skb_pagelen(skb); 600 601 iter->frag = skb_shinfo(skb)->frag_list; 602 skb_frag_list_init(skb); 603 604 iter->offset = 0; 605 iter->iph = iph; 606 iter->hlen = hlen; 607 608 skb->data_len = first_len - skb_headlen(skb); 609 skb->len = first_len; 610 iph->tot_len = htons(first_len); 611 iph->frag_off = htons(IP_MF); 612 ip_send_check(iph); 613 } 614 EXPORT_SYMBOL(ip_fraglist_init); 615 616 static void ip_fraglist_ipcb_prepare(struct sk_buff *skb, 617 struct ip_fraglist_iter *iter) 618 { 619 struct sk_buff *to = iter->frag; 620 621 /* Copy the flags to each fragment. */ 622 IPCB(to)->flags = IPCB(skb)->flags; 623 624 if (iter->offset == 0) 625 ip_options_fragment(to); 626 } 627 628 void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter) 629 { 630 unsigned int hlen = iter->hlen; 631 struct iphdr *iph = iter->iph; 632 struct sk_buff *frag; 633 634 frag = iter->frag; 635 frag->ip_summed = CHECKSUM_NONE; 636 skb_reset_transport_header(frag); 637 __skb_push(frag, hlen); 638 skb_reset_network_header(frag); 639 memcpy(skb_network_header(frag), iph, hlen); 640 iter->iph = ip_hdr(frag); 641 iph = iter->iph; 642 iph->tot_len = htons(frag->len); 643 ip_copy_metadata(frag, skb); 644 iter->offset += skb->len - hlen; 645 iph->frag_off = htons(iter->offset >> 3); 646 if (frag->next) 647 iph->frag_off |= htons(IP_MF); 648 /* Ready, complete checksum */ 649 ip_send_check(iph); 650 } 651 EXPORT_SYMBOL(ip_fraglist_prepare); 652 653 void ip_frag_init(struct sk_buff *skb, unsigned int hlen, 654 unsigned int ll_rs, unsigned int mtu, bool DF, 655 struct ip_frag_state *state) 656 { 657 struct iphdr *iph = ip_hdr(skb); 658 659 state->DF = DF; 660 state->hlen = hlen; 661 state->ll_rs = ll_rs; 662 state->mtu = mtu; 663 664 state->left = skb->len - hlen; /* Space per frame */ 665 state->ptr = hlen; /* Where to start from */ 666 667 state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 668 state->not_last_frag = iph->frag_off & htons(IP_MF); 669 } 670 EXPORT_SYMBOL(ip_frag_init); 671 672 static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to, 673 bool first_frag, struct ip_frag_state *state) 674 { 675 /* Copy the flags to each fragment. */ 676 IPCB(to)->flags = IPCB(from)->flags; 677 678 /* ANK: dirty, but effective trick. Upgrade options only if 679 * the segment to be fragmented was THE FIRST (otherwise, 680 * options are already fixed) and make it ONCE 681 * on the initial skb, so that all the following fragments 682 * will inherit fixed options. 683 */ 684 if (first_frag) 685 ip_options_fragment(from); 686 } 687 688 struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state) 689 { 690 unsigned int len = state->left; 691 struct sk_buff *skb2; 692 struct iphdr *iph; 693 694 len = state->left; 695 /* IF: it doesn't fit, use 'mtu' - the data space left */ 696 if (len > state->mtu) 697 len = state->mtu; 698 /* IF: we are not sending up to and including the packet end 699 then align the next start on an eight byte boundary */ 700 if (len < state->left) { 701 len &= ~7; 702 } 703 704 /* Allocate buffer */ 705 skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC); 706 if (!skb2) 707 return ERR_PTR(-ENOMEM); 708 709 /* 710 * Set up data on packet 711 */ 712 713 ip_copy_metadata(skb2, skb); 714 skb_reserve(skb2, state->ll_rs); 715 skb_put(skb2, len + state->hlen); 716 skb_reset_network_header(skb2); 717 skb2->transport_header = skb2->network_header + state->hlen; 718 719 /* 720 * Charge the memory for the fragment to any owner 721 * it might possess 722 */ 723 724 if (skb->sk) 725 skb_set_owner_w(skb2, skb->sk); 726 727 /* 728 * Copy the packet header into the new buffer. 729 */ 730 731 skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen); 732 733 /* 734 * Copy a block of the IP datagram. 735 */ 736 if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len)) 737 BUG(); 738 state->left -= len; 739 740 /* 741 * Fill in the new header fields. 742 */ 743 iph = ip_hdr(skb2); 744 iph->frag_off = htons((state->offset >> 3)); 745 if (state->DF) 746 iph->frag_off |= htons(IP_DF); 747 748 /* 749 * Added AC : If we are fragmenting a fragment that's not the 750 * last fragment then keep MF on each bit 751 */ 752 if (state->left > 0 || state->not_last_frag) 753 iph->frag_off |= htons(IP_MF); 754 state->ptr += len; 755 state->offset += len; 756 757 iph->tot_len = htons(len + state->hlen); 758 759 ip_send_check(iph); 760 761 return skb2; 762 } 763 EXPORT_SYMBOL(ip_frag_next); 764 765 /* 766 * This IP datagram is too large to be sent in one piece. Break it up into 767 * smaller pieces (each of size equal to IP header plus 768 * a block of the data of the original IP data part) that will yet fit in a 769 * single device frame, and queue such a frame for sending. 770 */ 771 772 int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 773 int (*output)(struct net *, struct sock *, struct sk_buff *)) 774 { 775 struct iphdr *iph; 776 struct sk_buff *skb2; 777 struct rtable *rt = skb_rtable(skb); 778 unsigned int mtu, hlen, ll_rs; 779 struct ip_fraglist_iter iter; 780 ktime_t tstamp = skb->tstamp; 781 struct ip_frag_state state; 782 int err = 0; 783 784 /* for offloaded checksums cleanup checksum before fragmentation */ 785 if (skb->ip_summed == CHECKSUM_PARTIAL && 786 (err = skb_checksum_help(skb))) 787 goto fail; 788 789 /* 790 * Point into the IP datagram header. 791 */ 792 793 iph = ip_hdr(skb); 794 795 mtu = ip_skb_dst_mtu(sk, skb); 796 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) 797 mtu = IPCB(skb)->frag_max_size; 798 799 /* 800 * Setup starting values. 801 */ 802 803 hlen = iph->ihl * 4; 804 mtu = mtu - hlen; /* Size of data space */ 805 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 806 ll_rs = LL_RESERVED_SPACE(rt->dst.dev); 807 808 /* When frag_list is given, use it. First, check its validity: 809 * some transformers could create wrong frag_list or break existing 810 * one, it is not prohibited. In this case fall back to copying. 811 * 812 * LATER: this step can be merged to real generation of fragments, 813 * we can switch to copy when see the first bad fragment. 814 */ 815 if (skb_has_frag_list(skb)) { 816 struct sk_buff *frag, *frag2; 817 unsigned int first_len = skb_pagelen(skb); 818 819 if (first_len - hlen > mtu || 820 ((first_len - hlen) & 7) || 821 ip_is_fragment(iph) || 822 skb_cloned(skb) || 823 skb_headroom(skb) < ll_rs) 824 goto slow_path; 825 826 skb_walk_frags(skb, frag) { 827 /* Correct geometry. */ 828 if (frag->len > mtu || 829 ((frag->len & 7) && frag->next) || 830 skb_headroom(frag) < hlen + ll_rs) 831 goto slow_path_clean; 832 833 /* Partially cloned skb? */ 834 if (skb_shared(frag)) 835 goto slow_path_clean; 836 837 BUG_ON(frag->sk); 838 if (skb->sk) { 839 frag->sk = skb->sk; 840 frag->destructor = sock_wfree; 841 } 842 skb->truesize -= frag->truesize; 843 } 844 845 /* Everything is OK. Generate! */ 846 ip_fraglist_init(skb, iph, hlen, &iter); 847 848 for (;;) { 849 /* Prepare header of the next frame, 850 * before previous one went down. */ 851 if (iter.frag) { 852 ip_fraglist_ipcb_prepare(skb, &iter); 853 ip_fraglist_prepare(skb, &iter); 854 } 855 856 skb->tstamp = tstamp; 857 err = output(net, sk, skb); 858 859 if (!err) 860 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 861 if (err || !iter.frag) 862 break; 863 864 skb = ip_fraglist_next(&iter); 865 } 866 867 if (err == 0) { 868 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 869 return 0; 870 } 871 872 kfree_skb_list(iter.frag); 873 874 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 875 return err; 876 877 slow_path_clean: 878 skb_walk_frags(skb, frag2) { 879 if (frag2 == frag) 880 break; 881 frag2->sk = NULL; 882 frag2->destructor = NULL; 883 skb->truesize += frag2->truesize; 884 } 885 } 886 887 slow_path: 888 /* 889 * Fragment the datagram. 890 */ 891 892 ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU, 893 &state); 894 895 /* 896 * Keep copying data until we run out. 897 */ 898 899 while (state.left > 0) { 900 bool first_frag = (state.offset == 0); 901 902 skb2 = ip_frag_next(skb, &state); 903 if (IS_ERR(skb2)) { 904 err = PTR_ERR(skb2); 905 goto fail; 906 } 907 ip_frag_ipcb(skb, skb2, first_frag, &state); 908 909 /* 910 * Put this fragment into the sending queue. 911 */ 912 skb2->tstamp = tstamp; 913 err = output(net, sk, skb2); 914 if (err) 915 goto fail; 916 917 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); 918 } 919 consume_skb(skb); 920 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); 921 return err; 922 923 fail: 924 kfree_skb(skb); 925 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); 926 return err; 927 } 928 EXPORT_SYMBOL(ip_do_fragment); 929 930 int 931 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 932 { 933 struct msghdr *msg = from; 934 935 if (skb->ip_summed == CHECKSUM_PARTIAL) { 936 if (!copy_from_iter_full(to, len, &msg->msg_iter)) 937 return -EFAULT; 938 } else { 939 __wsum csum = 0; 940 if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter)) 941 return -EFAULT; 942 skb->csum = csum_block_add(skb->csum, csum, odd); 943 } 944 return 0; 945 } 946 EXPORT_SYMBOL(ip_generic_getfrag); 947 948 static inline __wsum 949 csum_page(struct page *page, int offset, int copy) 950 { 951 char *kaddr; 952 __wsum csum; 953 kaddr = kmap(page); 954 csum = csum_partial(kaddr + offset, copy, 0); 955 kunmap(page); 956 return csum; 957 } 958 959 static int __ip_append_data(struct sock *sk, 960 struct flowi4 *fl4, 961 struct sk_buff_head *queue, 962 struct inet_cork *cork, 963 struct page_frag *pfrag, 964 int getfrag(void *from, char *to, int offset, 965 int len, int odd, struct sk_buff *skb), 966 void *from, int length, int transhdrlen, 967 unsigned int flags) 968 { 969 struct inet_sock *inet = inet_sk(sk); 970 struct ubuf_info *uarg = NULL; 971 struct sk_buff *skb; 972 973 struct ip_options *opt = cork->opt; 974 int hh_len; 975 int exthdrlen; 976 int mtu; 977 int copy; 978 int err; 979 int offset = 0; 980 unsigned int maxfraglen, fragheaderlen, maxnonfragsize; 981 int csummode = CHECKSUM_NONE; 982 struct rtable *rt = (struct rtable *)cork->dst; 983 unsigned int wmem_alloc_delta = 0; 984 bool paged, extra_uref = false; 985 u32 tskey = 0; 986 987 skb = skb_peek_tail(queue); 988 989 exthdrlen = !skb ? rt->dst.header_len : 0; 990 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; 991 paged = !!cork->gso_size; 992 993 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && 994 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) 995 tskey = sk->sk_tskey++; 996 997 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 998 999 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1000 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1001 maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu; 1002 1003 if (cork->length + length > maxnonfragsize - fragheaderlen) { 1004 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 1005 mtu - (opt ? opt->optlen : 0)); 1006 return -EMSGSIZE; 1007 } 1008 1009 /* 1010 * transhdrlen > 0 means that this is the first fragment and we wish 1011 * it won't be fragmented in the future. 1012 */ 1013 if (transhdrlen && 1014 length + fragheaderlen <= mtu && 1015 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) && 1016 (!(flags & MSG_MORE) || cork->gso_size) && 1017 (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM))) 1018 csummode = CHECKSUM_PARTIAL; 1019 1020 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) { 1021 uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb)); 1022 if (!uarg) 1023 return -ENOBUFS; 1024 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */ 1025 if (rt->dst.dev->features & NETIF_F_SG && 1026 csummode == CHECKSUM_PARTIAL) { 1027 paged = true; 1028 } else { 1029 uarg->zerocopy = 0; 1030 skb_zcopy_set(skb, uarg, &extra_uref); 1031 } 1032 } 1033 1034 cork->length += length; 1035 1036 /* So, what's going on in the loop below? 1037 * 1038 * We use calculated fragment length to generate chained skb, 1039 * each of segments is IP fragment ready for sending to network after 1040 * adding appropriate IP header. 1041 */ 1042 1043 if (!skb) 1044 goto alloc_new_skb; 1045 1046 while (length > 0) { 1047 /* Check if the remaining data fits into current packet. */ 1048 copy = mtu - skb->len; 1049 if (copy < length) 1050 copy = maxfraglen - skb->len; 1051 if (copy <= 0) { 1052 char *data; 1053 unsigned int datalen; 1054 unsigned int fraglen; 1055 unsigned int fraggap; 1056 unsigned int alloclen; 1057 unsigned int pagedlen; 1058 struct sk_buff *skb_prev; 1059 alloc_new_skb: 1060 skb_prev = skb; 1061 if (skb_prev) 1062 fraggap = skb_prev->len - maxfraglen; 1063 else 1064 fraggap = 0; 1065 1066 /* 1067 * If remaining data exceeds the mtu, 1068 * we know we need more fragment(s). 1069 */ 1070 datalen = length + fraggap; 1071 if (datalen > mtu - fragheaderlen) 1072 datalen = maxfraglen - fragheaderlen; 1073 fraglen = datalen + fragheaderlen; 1074 pagedlen = 0; 1075 1076 if ((flags & MSG_MORE) && 1077 !(rt->dst.dev->features&NETIF_F_SG)) 1078 alloclen = mtu; 1079 else if (!paged) 1080 alloclen = fraglen; 1081 else { 1082 alloclen = min_t(int, fraglen, MAX_HEADER); 1083 pagedlen = fraglen - alloclen; 1084 } 1085 1086 alloclen += exthdrlen; 1087 1088 /* The last fragment gets additional space at tail. 1089 * Note, with MSG_MORE we overallocate on fragments, 1090 * because we have no idea what fragment will be 1091 * the last. 1092 */ 1093 if (datalen == length + fraggap) 1094 alloclen += rt->dst.trailer_len; 1095 1096 if (transhdrlen) { 1097 skb = sock_alloc_send_skb(sk, 1098 alloclen + hh_len + 15, 1099 (flags & MSG_DONTWAIT), &err); 1100 } else { 1101 skb = NULL; 1102 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <= 1103 2 * sk->sk_sndbuf) 1104 skb = alloc_skb(alloclen + hh_len + 15, 1105 sk->sk_allocation); 1106 if (unlikely(!skb)) 1107 err = -ENOBUFS; 1108 } 1109 if (!skb) 1110 goto error; 1111 1112 /* 1113 * Fill in the control structures 1114 */ 1115 skb->ip_summed = csummode; 1116 skb->csum = 0; 1117 skb_reserve(skb, hh_len); 1118 1119 /* 1120 * Find where to start putting bytes. 1121 */ 1122 data = skb_put(skb, fraglen + exthdrlen - pagedlen); 1123 skb_set_network_header(skb, exthdrlen); 1124 skb->transport_header = (skb->network_header + 1125 fragheaderlen); 1126 data += fragheaderlen + exthdrlen; 1127 1128 if (fraggap) { 1129 skb->csum = skb_copy_and_csum_bits( 1130 skb_prev, maxfraglen, 1131 data + transhdrlen, fraggap); 1132 skb_prev->csum = csum_sub(skb_prev->csum, 1133 skb->csum); 1134 data += fraggap; 1135 pskb_trim_unique(skb_prev, maxfraglen); 1136 } 1137 1138 copy = datalen - transhdrlen - fraggap - pagedlen; 1139 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 1140 err = -EFAULT; 1141 kfree_skb(skb); 1142 goto error; 1143 } 1144 1145 offset += copy; 1146 length -= copy + transhdrlen; 1147 transhdrlen = 0; 1148 exthdrlen = 0; 1149 csummode = CHECKSUM_NONE; 1150 1151 /* only the initial fragment is time stamped */ 1152 skb_shinfo(skb)->tx_flags = cork->tx_flags; 1153 cork->tx_flags = 0; 1154 skb_shinfo(skb)->tskey = tskey; 1155 tskey = 0; 1156 skb_zcopy_set(skb, uarg, &extra_uref); 1157 1158 if ((flags & MSG_CONFIRM) && !skb_prev) 1159 skb_set_dst_pending_confirm(skb, 1); 1160 1161 /* 1162 * Put the packet on the pending queue. 1163 */ 1164 if (!skb->destructor) { 1165 skb->destructor = sock_wfree; 1166 skb->sk = sk; 1167 wmem_alloc_delta += skb->truesize; 1168 } 1169 __skb_queue_tail(queue, skb); 1170 continue; 1171 } 1172 1173 if (copy > length) 1174 copy = length; 1175 1176 if (!(rt->dst.dev->features&NETIF_F_SG) && 1177 skb_tailroom(skb) >= copy) { 1178 unsigned int off; 1179 1180 off = skb->len; 1181 if (getfrag(from, skb_put(skb, copy), 1182 offset, copy, off, skb) < 0) { 1183 __skb_trim(skb, off); 1184 err = -EFAULT; 1185 goto error; 1186 } 1187 } else if (!uarg || !uarg->zerocopy) { 1188 int i = skb_shinfo(skb)->nr_frags; 1189 1190 err = -ENOMEM; 1191 if (!sk_page_frag_refill(sk, pfrag)) 1192 goto error; 1193 1194 if (!skb_can_coalesce(skb, i, pfrag->page, 1195 pfrag->offset)) { 1196 err = -EMSGSIZE; 1197 if (i == MAX_SKB_FRAGS) 1198 goto error; 1199 1200 __skb_fill_page_desc(skb, i, pfrag->page, 1201 pfrag->offset, 0); 1202 skb_shinfo(skb)->nr_frags = ++i; 1203 get_page(pfrag->page); 1204 } 1205 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1206 if (getfrag(from, 1207 page_address(pfrag->page) + pfrag->offset, 1208 offset, copy, skb->len, skb) < 0) 1209 goto error_efault; 1210 1211 pfrag->offset += copy; 1212 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1213 skb->len += copy; 1214 skb->data_len += copy; 1215 skb->truesize += copy; 1216 wmem_alloc_delta += copy; 1217 } else { 1218 err = skb_zerocopy_iter_dgram(skb, from, copy); 1219 if (err < 0) 1220 goto error; 1221 } 1222 offset += copy; 1223 length -= copy; 1224 } 1225 1226 if (wmem_alloc_delta) 1227 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1228 return 0; 1229 1230 error_efault: 1231 err = -EFAULT; 1232 error: 1233 if (uarg) 1234 sock_zerocopy_put_abort(uarg, extra_uref); 1235 cork->length -= length; 1236 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1237 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1238 return err; 1239 } 1240 1241 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 1242 struct ipcm_cookie *ipc, struct rtable **rtp) 1243 { 1244 struct ip_options_rcu *opt; 1245 struct rtable *rt; 1246 1247 rt = *rtp; 1248 if (unlikely(!rt)) 1249 return -EFAULT; 1250 1251 /* 1252 * setup for corking. 1253 */ 1254 opt = ipc->opt; 1255 if (opt) { 1256 if (!cork->opt) { 1257 cork->opt = kmalloc(sizeof(struct ip_options) + 40, 1258 sk->sk_allocation); 1259 if (unlikely(!cork->opt)) 1260 return -ENOBUFS; 1261 } 1262 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 1263 cork->flags |= IPCORK_OPT; 1264 cork->addr = ipc->addr; 1265 } 1266 1267 cork->fragsize = ip_sk_use_pmtu(sk) ? 1268 dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu); 1269 1270 if (!inetdev_valid_mtu(cork->fragsize)) 1271 return -ENETUNREACH; 1272 1273 cork->gso_size = ipc->gso_size; 1274 1275 cork->dst = &rt->dst; 1276 /* We stole this route, caller should not release it. */ 1277 *rtp = NULL; 1278 1279 cork->length = 0; 1280 cork->ttl = ipc->ttl; 1281 cork->tos = ipc->tos; 1282 cork->mark = ipc->sockc.mark; 1283 cork->priority = ipc->priority; 1284 cork->transmit_time = ipc->sockc.transmit_time; 1285 cork->tx_flags = 0; 1286 sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags); 1287 1288 return 0; 1289 } 1290 1291 /* 1292 * ip_append_data() and ip_append_page() can make one large IP datagram 1293 * from many pieces of data. Each pieces will be holded on the socket 1294 * until ip_push_pending_frames() is called. Each piece can be a page 1295 * or non-page data. 1296 * 1297 * Not only UDP, other transport protocols - e.g. raw sockets - can use 1298 * this interface potentially. 1299 * 1300 * LATER: length must be adjusted by pad at tail, when it is required. 1301 */ 1302 int ip_append_data(struct sock *sk, struct flowi4 *fl4, 1303 int getfrag(void *from, char *to, int offset, int len, 1304 int odd, struct sk_buff *skb), 1305 void *from, int length, int transhdrlen, 1306 struct ipcm_cookie *ipc, struct rtable **rtp, 1307 unsigned int flags) 1308 { 1309 struct inet_sock *inet = inet_sk(sk); 1310 int err; 1311 1312 if (flags&MSG_PROBE) 1313 return 0; 1314 1315 if (skb_queue_empty(&sk->sk_write_queue)) { 1316 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 1317 if (err) 1318 return err; 1319 } else { 1320 transhdrlen = 0; 1321 } 1322 1323 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, 1324 sk_page_frag(sk), getfrag, 1325 from, length, transhdrlen, flags); 1326 } 1327 1328 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 1329 int offset, size_t size, int flags) 1330 { 1331 struct inet_sock *inet = inet_sk(sk); 1332 struct sk_buff *skb; 1333 struct rtable *rt; 1334 struct ip_options *opt = NULL; 1335 struct inet_cork *cork; 1336 int hh_len; 1337 int mtu; 1338 int len; 1339 int err; 1340 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; 1341 1342 if (inet->hdrincl) 1343 return -EPERM; 1344 1345 if (flags&MSG_PROBE) 1346 return 0; 1347 1348 if (skb_queue_empty(&sk->sk_write_queue)) 1349 return -EINVAL; 1350 1351 cork = &inet->cork.base; 1352 rt = (struct rtable *)cork->dst; 1353 if (cork->flags & IPCORK_OPT) 1354 opt = cork->opt; 1355 1356 if (!(rt->dst.dev->features & NETIF_F_SG)) 1357 return -EOPNOTSUPP; 1358 1359 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1360 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize; 1361 1362 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1363 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1364 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; 1365 1366 if (cork->length + size > maxnonfragsize - fragheaderlen) { 1367 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 1368 mtu - (opt ? opt->optlen : 0)); 1369 return -EMSGSIZE; 1370 } 1371 1372 skb = skb_peek_tail(&sk->sk_write_queue); 1373 if (!skb) 1374 return -EINVAL; 1375 1376 cork->length += size; 1377 1378 while (size > 0) { 1379 /* Check if the remaining data fits into current packet. */ 1380 len = mtu - skb->len; 1381 if (len < size) 1382 len = maxfraglen - skb->len; 1383 1384 if (len <= 0) { 1385 struct sk_buff *skb_prev; 1386 int alloclen; 1387 1388 skb_prev = skb; 1389 fraggap = skb_prev->len - maxfraglen; 1390 1391 alloclen = fragheaderlen + hh_len + fraggap + 15; 1392 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 1393 if (unlikely(!skb)) { 1394 err = -ENOBUFS; 1395 goto error; 1396 } 1397 1398 /* 1399 * Fill in the control structures 1400 */ 1401 skb->ip_summed = CHECKSUM_NONE; 1402 skb->csum = 0; 1403 skb_reserve(skb, hh_len); 1404 1405 /* 1406 * Find where to start putting bytes. 1407 */ 1408 skb_put(skb, fragheaderlen + fraggap); 1409 skb_reset_network_header(skb); 1410 skb->transport_header = (skb->network_header + 1411 fragheaderlen); 1412 if (fraggap) { 1413 skb->csum = skb_copy_and_csum_bits(skb_prev, 1414 maxfraglen, 1415 skb_transport_header(skb), 1416 fraggap); 1417 skb_prev->csum = csum_sub(skb_prev->csum, 1418 skb->csum); 1419 pskb_trim_unique(skb_prev, maxfraglen); 1420 } 1421 1422 /* 1423 * Put the packet on the pending queue. 1424 */ 1425 __skb_queue_tail(&sk->sk_write_queue, skb); 1426 continue; 1427 } 1428 1429 if (len > size) 1430 len = size; 1431 1432 if (skb_append_pagefrags(skb, page, offset, len)) { 1433 err = -EMSGSIZE; 1434 goto error; 1435 } 1436 1437 if (skb->ip_summed == CHECKSUM_NONE) { 1438 __wsum csum; 1439 csum = csum_page(page, offset, len); 1440 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1441 } 1442 1443 skb->len += len; 1444 skb->data_len += len; 1445 skb->truesize += len; 1446 refcount_add(len, &sk->sk_wmem_alloc); 1447 offset += len; 1448 size -= len; 1449 } 1450 return 0; 1451 1452 error: 1453 cork->length -= size; 1454 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1455 return err; 1456 } 1457 1458 static void ip_cork_release(struct inet_cork *cork) 1459 { 1460 cork->flags &= ~IPCORK_OPT; 1461 kfree(cork->opt); 1462 cork->opt = NULL; 1463 dst_release(cork->dst); 1464 cork->dst = NULL; 1465 } 1466 1467 /* 1468 * Combined all pending IP fragments on the socket as one IP datagram 1469 * and push them out. 1470 */ 1471 struct sk_buff *__ip_make_skb(struct sock *sk, 1472 struct flowi4 *fl4, 1473 struct sk_buff_head *queue, 1474 struct inet_cork *cork) 1475 { 1476 struct sk_buff *skb, *tmp_skb; 1477 struct sk_buff **tail_skb; 1478 struct inet_sock *inet = inet_sk(sk); 1479 struct net *net = sock_net(sk); 1480 struct ip_options *opt = NULL; 1481 struct rtable *rt = (struct rtable *)cork->dst; 1482 struct iphdr *iph; 1483 __be16 df = 0; 1484 __u8 ttl; 1485 1486 skb = __skb_dequeue(queue); 1487 if (!skb) 1488 goto out; 1489 tail_skb = &(skb_shinfo(skb)->frag_list); 1490 1491 /* move skb->data to ip header from ext header */ 1492 if (skb->data < skb_network_header(skb)) 1493 __skb_pull(skb, skb_network_offset(skb)); 1494 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1495 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1496 *tail_skb = tmp_skb; 1497 tail_skb = &(tmp_skb->next); 1498 skb->len += tmp_skb->len; 1499 skb->data_len += tmp_skb->len; 1500 skb->truesize += tmp_skb->truesize; 1501 tmp_skb->destructor = NULL; 1502 tmp_skb->sk = NULL; 1503 } 1504 1505 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 1506 * to fragment the frame generated here. No matter, what transforms 1507 * how transforms change size of the packet, it will come out. 1508 */ 1509 skb->ignore_df = ip_sk_ignore_df(sk); 1510 1511 /* DF bit is set when we want to see DF on outgoing frames. 1512 * If ignore_df is set too, we still allow to fragment this frame 1513 * locally. */ 1514 if (inet->pmtudisc == IP_PMTUDISC_DO || 1515 inet->pmtudisc == IP_PMTUDISC_PROBE || 1516 (skb->len <= dst_mtu(&rt->dst) && 1517 ip_dont_fragment(sk, &rt->dst))) 1518 df = htons(IP_DF); 1519 1520 if (cork->flags & IPCORK_OPT) 1521 opt = cork->opt; 1522 1523 if (cork->ttl != 0) 1524 ttl = cork->ttl; 1525 else if (rt->rt_type == RTN_MULTICAST) 1526 ttl = inet->mc_ttl; 1527 else 1528 ttl = ip_select_ttl(inet, &rt->dst); 1529 1530 iph = ip_hdr(skb); 1531 iph->version = 4; 1532 iph->ihl = 5; 1533 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; 1534 iph->frag_off = df; 1535 iph->ttl = ttl; 1536 iph->protocol = sk->sk_protocol; 1537 ip_copy_addrs(iph, fl4); 1538 ip_select_ident(net, skb, sk); 1539 1540 if (opt) { 1541 iph->ihl += opt->optlen >> 2; 1542 ip_options_build(skb, opt, cork->addr, rt, 0); 1543 } 1544 1545 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; 1546 skb->mark = cork->mark; 1547 skb->tstamp = cork->transmit_time; 1548 /* 1549 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1550 * on dst refcount 1551 */ 1552 cork->dst = NULL; 1553 skb_dst_set(skb, &rt->dst); 1554 1555 if (iph->protocol == IPPROTO_ICMP) 1556 icmp_out_count(net, ((struct icmphdr *) 1557 skb_transport_header(skb))->type); 1558 1559 ip_cork_release(cork); 1560 out: 1561 return skb; 1562 } 1563 1564 int ip_send_skb(struct net *net, struct sk_buff *skb) 1565 { 1566 int err; 1567 1568 err = ip_local_out(net, skb->sk, skb); 1569 if (err) { 1570 if (err > 0) 1571 err = net_xmit_errno(err); 1572 if (err) 1573 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 1574 } 1575 1576 return err; 1577 } 1578 1579 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 1580 { 1581 struct sk_buff *skb; 1582 1583 skb = ip_finish_skb(sk, fl4); 1584 if (!skb) 1585 return 0; 1586 1587 /* Netfilter gets whole the not fragmented skb. */ 1588 return ip_send_skb(sock_net(sk), skb); 1589 } 1590 1591 /* 1592 * Throw away all pending data on the socket. 1593 */ 1594 static void __ip_flush_pending_frames(struct sock *sk, 1595 struct sk_buff_head *queue, 1596 struct inet_cork *cork) 1597 { 1598 struct sk_buff *skb; 1599 1600 while ((skb = __skb_dequeue_tail(queue)) != NULL) 1601 kfree_skb(skb); 1602 1603 ip_cork_release(cork); 1604 } 1605 1606 void ip_flush_pending_frames(struct sock *sk) 1607 { 1608 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 1609 } 1610 1611 struct sk_buff *ip_make_skb(struct sock *sk, 1612 struct flowi4 *fl4, 1613 int getfrag(void *from, char *to, int offset, 1614 int len, int odd, struct sk_buff *skb), 1615 void *from, int length, int transhdrlen, 1616 struct ipcm_cookie *ipc, struct rtable **rtp, 1617 struct inet_cork *cork, unsigned int flags) 1618 { 1619 struct sk_buff_head queue; 1620 int err; 1621 1622 if (flags & MSG_PROBE) 1623 return NULL; 1624 1625 __skb_queue_head_init(&queue); 1626 1627 cork->flags = 0; 1628 cork->addr = 0; 1629 cork->opt = NULL; 1630 err = ip_setup_cork(sk, cork, ipc, rtp); 1631 if (err) 1632 return ERR_PTR(err); 1633 1634 err = __ip_append_data(sk, fl4, &queue, cork, 1635 ¤t->task_frag, getfrag, 1636 from, length, transhdrlen, flags); 1637 if (err) { 1638 __ip_flush_pending_frames(sk, &queue, cork); 1639 return ERR_PTR(err); 1640 } 1641 1642 return __ip_make_skb(sk, fl4, &queue, cork); 1643 } 1644 1645 /* 1646 * Fetch data from kernel space and fill in checksum if needed. 1647 */ 1648 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 1649 int len, int odd, struct sk_buff *skb) 1650 { 1651 __wsum csum; 1652 1653 csum = csum_partial_copy_nocheck(dptr+offset, to, len); 1654 skb->csum = csum_block_add(skb->csum, csum, odd); 1655 return 0; 1656 } 1657 1658 /* 1659 * Generic function to send a packet as reply to another packet. 1660 * Used to send some TCP resets/acks so far. 1661 */ 1662 void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, 1663 const struct ip_options *sopt, 1664 __be32 daddr, __be32 saddr, 1665 const struct ip_reply_arg *arg, 1666 unsigned int len, u64 transmit_time) 1667 { 1668 struct ip_options_data replyopts; 1669 struct ipcm_cookie ipc; 1670 struct flowi4 fl4; 1671 struct rtable *rt = skb_rtable(skb); 1672 struct net *net = sock_net(sk); 1673 struct sk_buff *nskb; 1674 int err; 1675 int oif; 1676 1677 if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt)) 1678 return; 1679 1680 ipcm_init(&ipc); 1681 ipc.addr = daddr; 1682 ipc.sockc.transmit_time = transmit_time; 1683 1684 if (replyopts.opt.opt.optlen) { 1685 ipc.opt = &replyopts.opt; 1686 1687 if (replyopts.opt.opt.srr) 1688 daddr = replyopts.opt.opt.faddr; 1689 } 1690 1691 oif = arg->bound_dev_if; 1692 if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) 1693 oif = skb->skb_iif; 1694 1695 flowi4_init_output(&fl4, oif, 1696 IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark, 1697 RT_TOS(arg->tos), 1698 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, 1699 ip_reply_arg_flowi_flags(arg), 1700 daddr, saddr, 1701 tcp_hdr(skb)->source, tcp_hdr(skb)->dest, 1702 arg->uid); 1703 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4)); 1704 rt = ip_route_output_key(net, &fl4); 1705 if (IS_ERR(rt)) 1706 return; 1707 1708 inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK; 1709 1710 sk->sk_protocol = ip_hdr(skb)->protocol; 1711 sk->sk_bound_dev_if = arg->bound_dev_if; 1712 sk->sk_sndbuf = sysctl_wmem_default; 1713 ipc.sockc.mark = fl4.flowi4_mark; 1714 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, 1715 len, 0, &ipc, &rt, MSG_DONTWAIT); 1716 if (unlikely(err)) { 1717 ip_flush_pending_frames(sk); 1718 goto out; 1719 } 1720 1721 nskb = skb_peek(&sk->sk_write_queue); 1722 if (nskb) { 1723 if (arg->csumoffset >= 0) 1724 *((__sum16 *)skb_transport_header(nskb) + 1725 arg->csumoffset) = csum_fold(csum_add(nskb->csum, 1726 arg->csum)); 1727 nskb->ip_summed = CHECKSUM_NONE; 1728 ip_push_pending_frames(sk, &fl4); 1729 } 1730 out: 1731 ip_rt_put(rt); 1732 } 1733 1734 void __init ip_init(void) 1735 { 1736 ip_rt_init(); 1737 inet_initpeers(); 1738 1739 #if defined(CONFIG_IP_MULTICAST) 1740 igmp_mc_init(); 1741 #endif 1742 } 1743