1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The Internet Protocol (IP) output module. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Donald Becker, <becker@super.org> 11 * Alan Cox, <Alan.Cox@linux.org> 12 * Richard Underwood 13 * Stefan Becker, <stefanb@yello.ping.de> 14 * Jorge Cwik, <jorge@laser.satlink.net> 15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 16 * Hirokazu Takahashi, <taka@valinux.co.jp> 17 * 18 * See ip_input.c for original log 19 * 20 * Fixes: 21 * Alan Cox : Missing nonblock feature in ip_build_xmit. 22 * Mike Kilburn : htons() missing in ip_build_xmit. 23 * Bradford Johnson: Fix faulty handling of some frames when 24 * no route is found. 25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit 26 * (in case if packet not accepted by 27 * output firewall rules) 28 * Mike McLagan : Routing by source 29 * Alexey Kuznetsov: use new route cache 30 * Andi Kleen: Fix broken PMTU recovery and remove 31 * some redundant tests. 32 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 33 * Andi Kleen : Replace ip_reply with ip_send_reply. 34 * Andi Kleen : Split fast and slow ip_build_xmit path 35 * for decreased register pressure on x86 36 * and more readibility. 37 * Marc Boucher : When call_out_firewall returns FW_QUEUE, 38 * silently drop skb instead of failing with -EPERM. 39 * Detlev Wengorz : Copy protocol for fragments. 40 * Hirokazu Takahashi: HW checksumming for outgoing UDP 41 * datagrams. 42 * Hirokazu Takahashi: sendfile() on UDP works now. 43 */ 44 45 #include <asm/uaccess.h> 46 #include <linux/module.h> 47 #include <linux/types.h> 48 #include <linux/kernel.h> 49 #include <linux/mm.h> 50 #include <linux/string.h> 51 #include <linux/errno.h> 52 #include <linux/highmem.h> 53 #include <linux/slab.h> 54 55 #include <linux/socket.h> 56 #include <linux/sockios.h> 57 #include <linux/in.h> 58 #include <linux/inet.h> 59 #include <linux/netdevice.h> 60 #include <linux/etherdevice.h> 61 #include <linux/proc_fs.h> 62 #include <linux/stat.h> 63 #include <linux/init.h> 64 65 #include <net/snmp.h> 66 #include <net/ip.h> 67 #include <net/protocol.h> 68 #include <net/route.h> 69 #include <net/xfrm.h> 70 #include <linux/skbuff.h> 71 #include <net/sock.h> 72 #include <net/arp.h> 73 #include <net/icmp.h> 74 #include <net/checksum.h> 75 #include <net/inetpeer.h> 76 #include <linux/igmp.h> 77 #include <linux/netfilter_ipv4.h> 78 #include <linux/netfilter_bridge.h> 79 #include <linux/mroute.h> 80 #include <linux/netlink.h> 81 #include <linux/tcp.h> 82 83 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL; 84 EXPORT_SYMBOL(sysctl_ip_default_ttl); 85 86 /* Generate a checksum for an outgoing IP datagram. */ 87 void ip_send_check(struct iphdr *iph) 88 { 89 iph->check = 0; 90 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 91 } 92 EXPORT_SYMBOL(ip_send_check); 93 94 int __ip_local_out(struct sk_buff *skb) 95 { 96 struct iphdr *iph = ip_hdr(skb); 97 98 iph->tot_len = htons(skb->len); 99 ip_send_check(iph); 100 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL, 101 skb_dst(skb)->dev, dst_output); 102 } 103 104 int ip_local_out(struct sk_buff *skb) 105 { 106 int err; 107 108 err = __ip_local_out(skb); 109 if (likely(err == 1)) 110 err = dst_output(skb); 111 112 return err; 113 } 114 EXPORT_SYMBOL_GPL(ip_local_out); 115 116 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) 117 { 118 int ttl = inet->uc_ttl; 119 120 if (ttl < 0) 121 ttl = ip4_dst_hoplimit(dst); 122 return ttl; 123 } 124 125 /* 126 * Add an ip header to a skbuff and send it out. 127 * 128 */ 129 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk, 130 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) 131 { 132 struct inet_sock *inet = inet_sk(sk); 133 struct rtable *rt = skb_rtable(skb); 134 struct iphdr *iph; 135 136 /* Build the IP header. */ 137 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); 138 skb_reset_network_header(skb); 139 iph = ip_hdr(skb); 140 iph->version = 4; 141 iph->ihl = 5; 142 iph->tos = inet->tos; 143 if (ip_dont_fragment(sk, &rt->dst)) 144 iph->frag_off = htons(IP_DF); 145 else 146 iph->frag_off = 0; 147 iph->ttl = ip_select_ttl(inet, &rt->dst); 148 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); 149 iph->saddr = saddr; 150 iph->protocol = sk->sk_protocol; 151 ip_select_ident(iph, &rt->dst, sk); 152 153 if (opt && opt->opt.optlen) { 154 iph->ihl += opt->opt.optlen>>2; 155 ip_options_build(skb, &opt->opt, daddr, rt, 0); 156 } 157 158 skb->priority = sk->sk_priority; 159 skb->mark = sk->sk_mark; 160 161 /* Send it out. */ 162 return ip_local_out(skb); 163 } 164 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); 165 166 static inline int ip_finish_output2(struct sk_buff *skb) 167 { 168 struct dst_entry *dst = skb_dst(skb); 169 struct rtable *rt = (struct rtable *)dst; 170 struct net_device *dev = dst->dev; 171 unsigned int hh_len = LL_RESERVED_SPACE(dev); 172 struct neighbour *neigh; 173 u32 nexthop; 174 175 if (rt->rt_type == RTN_MULTICAST) { 176 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len); 177 } else if (rt->rt_type == RTN_BROADCAST) 178 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len); 179 180 /* Be paranoid, rather than too clever. */ 181 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 182 struct sk_buff *skb2; 183 184 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); 185 if (skb2 == NULL) { 186 kfree_skb(skb); 187 return -ENOMEM; 188 } 189 if (skb->sk) 190 skb_set_owner_w(skb2, skb->sk); 191 consume_skb(skb); 192 skb = skb2; 193 } 194 195 rcu_read_lock_bh(); 196 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr); 197 neigh = __ipv4_neigh_lookup_noref(dev, nexthop); 198 if (unlikely(!neigh)) 199 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); 200 if (!IS_ERR(neigh)) { 201 int res = dst_neigh_output(dst, neigh, skb); 202 203 rcu_read_unlock_bh(); 204 return res; 205 } 206 rcu_read_unlock_bh(); 207 208 net_dbg_ratelimited("%s: No header cache and no neighbour!\n", 209 __func__); 210 kfree_skb(skb); 211 return -EINVAL; 212 } 213 214 static inline int ip_skb_dst_mtu(struct sk_buff *skb) 215 { 216 struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL; 217 218 return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ? 219 skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb)); 220 } 221 222 static int ip_finish_output(struct sk_buff *skb) 223 { 224 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) 225 /* Policy lookup after SNAT yielded a new policy */ 226 if (skb_dst(skb)->xfrm != NULL) { 227 IPCB(skb)->flags |= IPSKB_REROUTED; 228 return dst_output(skb); 229 } 230 #endif 231 if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb)) 232 return ip_fragment(skb, ip_finish_output2); 233 else 234 return ip_finish_output2(skb); 235 } 236 237 int ip_mc_output(struct sk_buff *skb) 238 { 239 struct sock *sk = skb->sk; 240 struct rtable *rt = skb_rtable(skb); 241 struct net_device *dev = rt->dst.dev; 242 243 /* 244 * If the indicated interface is up and running, send the packet. 245 */ 246 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); 247 248 skb->dev = dev; 249 skb->protocol = htons(ETH_P_IP); 250 251 /* 252 * Multicasts are looped back for other local users 253 */ 254 255 if (rt->rt_flags&RTCF_MULTICAST) { 256 if (sk_mc_loop(sk) 257 #ifdef CONFIG_IP_MROUTE 258 /* Small optimization: do not loopback not local frames, 259 which returned after forwarding; they will be dropped 260 by ip_mr_input in any case. 261 Note, that local frames are looped back to be delivered 262 to local recipients. 263 264 This check is duplicated in ip_mr_input at the moment. 265 */ 266 && 267 ((rt->rt_flags & RTCF_LOCAL) || 268 !(IPCB(skb)->flags & IPSKB_FORWARDED)) 269 #endif 270 ) { 271 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 272 if (newskb) 273 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, 274 newskb, NULL, newskb->dev, 275 dev_loopback_xmit); 276 } 277 278 /* Multicasts with ttl 0 must not go beyond the host */ 279 280 if (ip_hdr(skb)->ttl == 0) { 281 kfree_skb(skb); 282 return 0; 283 } 284 } 285 286 if (rt->rt_flags&RTCF_BROADCAST) { 287 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 288 if (newskb) 289 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb, 290 NULL, newskb->dev, dev_loopback_xmit); 291 } 292 293 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, 294 skb->dev, ip_finish_output, 295 !(IPCB(skb)->flags & IPSKB_REROUTED)); 296 } 297 298 int ip_output(struct sk_buff *skb) 299 { 300 struct net_device *dev = skb_dst(skb)->dev; 301 302 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); 303 304 skb->dev = dev; 305 skb->protocol = htons(ETH_P_IP); 306 307 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev, 308 ip_finish_output, 309 !(IPCB(skb)->flags & IPSKB_REROUTED)); 310 } 311 312 /* 313 * copy saddr and daddr, possibly using 64bit load/stores 314 * Equivalent to : 315 * iph->saddr = fl4->saddr; 316 * iph->daddr = fl4->daddr; 317 */ 318 static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) 319 { 320 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != 321 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); 322 memcpy(&iph->saddr, &fl4->saddr, 323 sizeof(fl4->saddr) + sizeof(fl4->daddr)); 324 } 325 326 int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl) 327 { 328 struct sock *sk = skb->sk; 329 struct inet_sock *inet = inet_sk(sk); 330 struct ip_options_rcu *inet_opt; 331 struct flowi4 *fl4; 332 struct rtable *rt; 333 struct iphdr *iph; 334 int res; 335 336 /* Skip all of this if the packet is already routed, 337 * f.e. by something like SCTP. 338 */ 339 rcu_read_lock(); 340 inet_opt = rcu_dereference(inet->inet_opt); 341 fl4 = &fl->u.ip4; 342 rt = skb_rtable(skb); 343 if (rt != NULL) 344 goto packet_routed; 345 346 /* Make sure we can route this packet. */ 347 rt = (struct rtable *)__sk_dst_check(sk, 0); 348 if (rt == NULL) { 349 __be32 daddr; 350 351 /* Use correct destination address if we have options. */ 352 daddr = inet->inet_daddr; 353 if (inet_opt && inet_opt->opt.srr) 354 daddr = inet_opt->opt.faddr; 355 356 /* If this fails, retransmit mechanism of transport layer will 357 * keep trying until route appears or the connection times 358 * itself out. 359 */ 360 rt = ip_route_output_ports(sock_net(sk), fl4, sk, 361 daddr, inet->inet_saddr, 362 inet->inet_dport, 363 inet->inet_sport, 364 sk->sk_protocol, 365 RT_CONN_FLAGS(sk), 366 sk->sk_bound_dev_if); 367 if (IS_ERR(rt)) 368 goto no_route; 369 sk_setup_caps(sk, &rt->dst); 370 } 371 skb_dst_set_noref(skb, &rt->dst); 372 373 packet_routed: 374 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) 375 goto no_route; 376 377 /* OK, we know where to send it, allocate and build IP header. */ 378 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); 379 skb_reset_network_header(skb); 380 iph = ip_hdr(skb); 381 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); 382 if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df) 383 iph->frag_off = htons(IP_DF); 384 else 385 iph->frag_off = 0; 386 iph->ttl = ip_select_ttl(inet, &rt->dst); 387 iph->protocol = sk->sk_protocol; 388 ip_copy_addrs(iph, fl4); 389 390 /* Transport layer set skb->h.foo itself. */ 391 392 if (inet_opt && inet_opt->opt.optlen) { 393 iph->ihl += inet_opt->opt.optlen >> 2; 394 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); 395 } 396 397 ip_select_ident_more(iph, &rt->dst, sk, 398 (skb_shinfo(skb)->gso_segs ?: 1) - 1); 399 400 skb->priority = sk->sk_priority; 401 skb->mark = sk->sk_mark; 402 403 res = ip_local_out(skb); 404 rcu_read_unlock(); 405 return res; 406 407 no_route: 408 rcu_read_unlock(); 409 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); 410 kfree_skb(skb); 411 return -EHOSTUNREACH; 412 } 413 EXPORT_SYMBOL(ip_queue_xmit); 414 415 416 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) 417 { 418 to->pkt_type = from->pkt_type; 419 to->priority = from->priority; 420 to->protocol = from->protocol; 421 skb_dst_drop(to); 422 skb_dst_copy(to, from); 423 to->dev = from->dev; 424 to->mark = from->mark; 425 426 /* Copy the flags to each fragment. */ 427 IPCB(to)->flags = IPCB(from)->flags; 428 429 #ifdef CONFIG_NET_SCHED 430 to->tc_index = from->tc_index; 431 #endif 432 nf_copy(to, from); 433 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) 434 to->nf_trace = from->nf_trace; 435 #endif 436 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 437 to->ipvs_property = from->ipvs_property; 438 #endif 439 skb_copy_secmark(to, from); 440 } 441 442 /* 443 * This IP datagram is too large to be sent in one piece. Break it up into 444 * smaller pieces (each of size equal to IP header plus 445 * a block of the data of the original IP data part) that will yet fit in a 446 * single device frame, and queue such a frame for sending. 447 */ 448 449 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)) 450 { 451 struct iphdr *iph; 452 int ptr; 453 struct net_device *dev; 454 struct sk_buff *skb2; 455 unsigned int mtu, hlen, left, len, ll_rs; 456 int offset; 457 __be16 not_last_frag; 458 struct rtable *rt = skb_rtable(skb); 459 int err = 0; 460 461 dev = rt->dst.dev; 462 463 /* 464 * Point into the IP datagram header. 465 */ 466 467 iph = ip_hdr(skb); 468 469 if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) || 470 (IPCB(skb)->frag_max_size && 471 IPCB(skb)->frag_max_size > dst_mtu(&rt->dst)))) { 472 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 473 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, 474 htonl(ip_skb_dst_mtu(skb))); 475 kfree_skb(skb); 476 return -EMSGSIZE; 477 } 478 479 /* 480 * Setup starting values. 481 */ 482 483 hlen = iph->ihl * 4; 484 mtu = dst_mtu(&rt->dst) - hlen; /* Size of data space */ 485 #ifdef CONFIG_BRIDGE_NETFILTER 486 if (skb->nf_bridge) 487 mtu -= nf_bridge_mtu_reduction(skb); 488 #endif 489 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; 490 491 /* When frag_list is given, use it. First, check its validity: 492 * some transformers could create wrong frag_list or break existing 493 * one, it is not prohibited. In this case fall back to copying. 494 * 495 * LATER: this step can be merged to real generation of fragments, 496 * we can switch to copy when see the first bad fragment. 497 */ 498 if (skb_has_frag_list(skb)) { 499 struct sk_buff *frag, *frag2; 500 int first_len = skb_pagelen(skb); 501 502 if (first_len - hlen > mtu || 503 ((first_len - hlen) & 7) || 504 ip_is_fragment(iph) || 505 skb_cloned(skb)) 506 goto slow_path; 507 508 skb_walk_frags(skb, frag) { 509 /* Correct geometry. */ 510 if (frag->len > mtu || 511 ((frag->len & 7) && frag->next) || 512 skb_headroom(frag) < hlen) 513 goto slow_path_clean; 514 515 /* Partially cloned skb? */ 516 if (skb_shared(frag)) 517 goto slow_path_clean; 518 519 BUG_ON(frag->sk); 520 if (skb->sk) { 521 frag->sk = skb->sk; 522 frag->destructor = sock_wfree; 523 } 524 skb->truesize -= frag->truesize; 525 } 526 527 /* Everything is OK. Generate! */ 528 529 err = 0; 530 offset = 0; 531 frag = skb_shinfo(skb)->frag_list; 532 skb_frag_list_init(skb); 533 skb->data_len = first_len - skb_headlen(skb); 534 skb->len = first_len; 535 iph->tot_len = htons(first_len); 536 iph->frag_off = htons(IP_MF); 537 ip_send_check(iph); 538 539 for (;;) { 540 /* Prepare header of the next frame, 541 * before previous one went down. */ 542 if (frag) { 543 frag->ip_summed = CHECKSUM_NONE; 544 skb_reset_transport_header(frag); 545 __skb_push(frag, hlen); 546 skb_reset_network_header(frag); 547 memcpy(skb_network_header(frag), iph, hlen); 548 iph = ip_hdr(frag); 549 iph->tot_len = htons(frag->len); 550 ip_copy_metadata(frag, skb); 551 if (offset == 0) 552 ip_options_fragment(frag); 553 offset += skb->len - hlen; 554 iph->frag_off = htons(offset>>3); 555 if (frag->next != NULL) 556 iph->frag_off |= htons(IP_MF); 557 /* Ready, complete checksum */ 558 ip_send_check(iph); 559 } 560 561 err = output(skb); 562 563 if (!err) 564 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); 565 if (err || !frag) 566 break; 567 568 skb = frag; 569 frag = skb->next; 570 skb->next = NULL; 571 } 572 573 if (err == 0) { 574 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); 575 return 0; 576 } 577 578 while (frag) { 579 skb = frag->next; 580 kfree_skb(frag); 581 frag = skb; 582 } 583 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 584 return err; 585 586 slow_path_clean: 587 skb_walk_frags(skb, frag2) { 588 if (frag2 == frag) 589 break; 590 frag2->sk = NULL; 591 frag2->destructor = NULL; 592 skb->truesize += frag2->truesize; 593 } 594 } 595 596 slow_path: 597 /* for offloaded checksums cleanup checksum before fragmentation */ 598 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb)) 599 goto fail; 600 iph = ip_hdr(skb); 601 602 left = skb->len - hlen; /* Space per frame */ 603 ptr = hlen; /* Where to start from */ 604 605 /* for bridged IP traffic encapsulated inside f.e. a vlan header, 606 * we need to make room for the encapsulating header 607 */ 608 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb)); 609 610 /* 611 * Fragment the datagram. 612 */ 613 614 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; 615 not_last_frag = iph->frag_off & htons(IP_MF); 616 617 /* 618 * Keep copying data until we run out. 619 */ 620 621 while (left > 0) { 622 len = left; 623 /* IF: it doesn't fit, use 'mtu' - the data space left */ 624 if (len > mtu) 625 len = mtu; 626 /* IF: we are not sending up to and including the packet end 627 then align the next start on an eight byte boundary */ 628 if (len < left) { 629 len &= ~7; 630 } 631 /* 632 * Allocate buffer. 633 */ 634 635 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { 636 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); 637 err = -ENOMEM; 638 goto fail; 639 } 640 641 /* 642 * Set up data on packet 643 */ 644 645 ip_copy_metadata(skb2, skb); 646 skb_reserve(skb2, ll_rs); 647 skb_put(skb2, len + hlen); 648 skb_reset_network_header(skb2); 649 skb2->transport_header = skb2->network_header + hlen; 650 651 /* 652 * Charge the memory for the fragment to any owner 653 * it might possess 654 */ 655 656 if (skb->sk) 657 skb_set_owner_w(skb2, skb->sk); 658 659 /* 660 * Copy the packet header into the new buffer. 661 */ 662 663 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); 664 665 /* 666 * Copy a block of the IP datagram. 667 */ 668 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) 669 BUG(); 670 left -= len; 671 672 /* 673 * Fill in the new header fields. 674 */ 675 iph = ip_hdr(skb2); 676 iph->frag_off = htons((offset >> 3)); 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 (offset == 0) 685 ip_options_fragment(skb); 686 687 /* 688 * Added AC : If we are fragmenting a fragment that's not the 689 * last fragment then keep MF on each bit 690 */ 691 if (left > 0 || not_last_frag) 692 iph->frag_off |= htons(IP_MF); 693 ptr += len; 694 offset += len; 695 696 /* 697 * Put this fragment into the sending queue. 698 */ 699 iph->tot_len = htons(len + hlen); 700 701 ip_send_check(iph); 702 703 err = output(skb2); 704 if (err) 705 goto fail; 706 707 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); 708 } 709 consume_skb(skb); 710 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); 711 return err; 712 713 fail: 714 kfree_skb(skb); 715 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); 716 return err; 717 } 718 EXPORT_SYMBOL(ip_fragment); 719 720 int 721 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) 722 { 723 struct iovec *iov = from; 724 725 if (skb->ip_summed == CHECKSUM_PARTIAL) { 726 if (memcpy_fromiovecend(to, iov, offset, len) < 0) 727 return -EFAULT; 728 } else { 729 __wsum csum = 0; 730 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) 731 return -EFAULT; 732 skb->csum = csum_block_add(skb->csum, csum, odd); 733 } 734 return 0; 735 } 736 EXPORT_SYMBOL(ip_generic_getfrag); 737 738 static inline __wsum 739 csum_page(struct page *page, int offset, int copy) 740 { 741 char *kaddr; 742 __wsum csum; 743 kaddr = kmap(page); 744 csum = csum_partial(kaddr + offset, copy, 0); 745 kunmap(page); 746 return csum; 747 } 748 749 static inline int ip_ufo_append_data(struct sock *sk, 750 struct sk_buff_head *queue, 751 int getfrag(void *from, char *to, int offset, int len, 752 int odd, struct sk_buff *skb), 753 void *from, int length, int hh_len, int fragheaderlen, 754 int transhdrlen, int maxfraglen, unsigned int flags) 755 { 756 struct sk_buff *skb; 757 int err; 758 759 /* There is support for UDP fragmentation offload by network 760 * device, so create one single skb packet containing complete 761 * udp datagram 762 */ 763 if ((skb = skb_peek_tail(queue)) == NULL) { 764 skb = sock_alloc_send_skb(sk, 765 hh_len + fragheaderlen + transhdrlen + 20, 766 (flags & MSG_DONTWAIT), &err); 767 768 if (skb == NULL) 769 return err; 770 771 /* reserve space for Hardware header */ 772 skb_reserve(skb, hh_len); 773 774 /* create space for UDP/IP header */ 775 skb_put(skb, fragheaderlen + transhdrlen); 776 777 /* initialize network header pointer */ 778 skb_reset_network_header(skb); 779 780 /* initialize protocol header pointer */ 781 skb->transport_header = skb->network_header + fragheaderlen; 782 783 skb->ip_summed = CHECKSUM_PARTIAL; 784 skb->csum = 0; 785 786 /* specify the length of each IP datagram fragment */ 787 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen; 788 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 789 __skb_queue_tail(queue, skb); 790 } 791 792 return skb_append_datato_frags(sk, skb, getfrag, from, 793 (length - transhdrlen)); 794 } 795 796 static int __ip_append_data(struct sock *sk, 797 struct flowi4 *fl4, 798 struct sk_buff_head *queue, 799 struct inet_cork *cork, 800 struct page_frag *pfrag, 801 int getfrag(void *from, char *to, int offset, 802 int len, int odd, struct sk_buff *skb), 803 void *from, int length, int transhdrlen, 804 unsigned int flags) 805 { 806 struct inet_sock *inet = inet_sk(sk); 807 struct sk_buff *skb; 808 809 struct ip_options *opt = cork->opt; 810 int hh_len; 811 int exthdrlen; 812 int mtu; 813 int copy; 814 int err; 815 int offset = 0; 816 unsigned int maxfraglen, fragheaderlen; 817 int csummode = CHECKSUM_NONE; 818 struct rtable *rt = (struct rtable *)cork->dst; 819 820 skb = skb_peek_tail(queue); 821 822 exthdrlen = !skb ? rt->dst.header_len : 0; 823 mtu = cork->fragsize; 824 825 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 826 827 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 828 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 829 830 if (cork->length + length > 0xFFFF - fragheaderlen) { 831 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, 832 mtu-exthdrlen); 833 return -EMSGSIZE; 834 } 835 836 /* 837 * transhdrlen > 0 means that this is the first fragment and we wish 838 * it won't be fragmented in the future. 839 */ 840 if (transhdrlen && 841 length + fragheaderlen <= mtu && 842 rt->dst.dev->features & NETIF_F_V4_CSUM && 843 !exthdrlen) 844 csummode = CHECKSUM_PARTIAL; 845 846 cork->length += length; 847 if (((length > mtu) || (skb && skb_is_gso(skb))) && 848 (sk->sk_protocol == IPPROTO_UDP) && 849 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) { 850 err = ip_ufo_append_data(sk, queue, getfrag, from, length, 851 hh_len, fragheaderlen, transhdrlen, 852 maxfraglen, flags); 853 if (err) 854 goto error; 855 return 0; 856 } 857 858 /* So, what's going on in the loop below? 859 * 860 * We use calculated fragment length to generate chained skb, 861 * each of segments is IP fragment ready for sending to network after 862 * adding appropriate IP header. 863 */ 864 865 if (!skb) 866 goto alloc_new_skb; 867 868 while (length > 0) { 869 /* Check if the remaining data fits into current packet. */ 870 copy = mtu - skb->len; 871 if (copy < length) 872 copy = maxfraglen - skb->len; 873 if (copy <= 0) { 874 char *data; 875 unsigned int datalen; 876 unsigned int fraglen; 877 unsigned int fraggap; 878 unsigned int alloclen; 879 struct sk_buff *skb_prev; 880 alloc_new_skb: 881 skb_prev = skb; 882 if (skb_prev) 883 fraggap = skb_prev->len - maxfraglen; 884 else 885 fraggap = 0; 886 887 /* 888 * If remaining data exceeds the mtu, 889 * we know we need more fragment(s). 890 */ 891 datalen = length + fraggap; 892 if (datalen > mtu - fragheaderlen) 893 datalen = maxfraglen - fragheaderlen; 894 fraglen = datalen + fragheaderlen; 895 896 if ((flags & MSG_MORE) && 897 !(rt->dst.dev->features&NETIF_F_SG)) 898 alloclen = mtu; 899 else 900 alloclen = fraglen; 901 902 alloclen += exthdrlen; 903 904 /* The last fragment gets additional space at tail. 905 * Note, with MSG_MORE we overallocate on fragments, 906 * because we have no idea what fragment will be 907 * the last. 908 */ 909 if (datalen == length + fraggap) 910 alloclen += rt->dst.trailer_len; 911 912 if (transhdrlen) { 913 skb = sock_alloc_send_skb(sk, 914 alloclen + hh_len + 15, 915 (flags & MSG_DONTWAIT), &err); 916 } else { 917 skb = NULL; 918 if (atomic_read(&sk->sk_wmem_alloc) <= 919 2 * sk->sk_sndbuf) 920 skb = sock_wmalloc(sk, 921 alloclen + hh_len + 15, 1, 922 sk->sk_allocation); 923 if (unlikely(skb == NULL)) 924 err = -ENOBUFS; 925 else 926 /* only the initial fragment is 927 time stamped */ 928 cork->tx_flags = 0; 929 } 930 if (skb == NULL) 931 goto error; 932 933 /* 934 * Fill in the control structures 935 */ 936 skb->ip_summed = csummode; 937 skb->csum = 0; 938 skb_reserve(skb, hh_len); 939 skb_shinfo(skb)->tx_flags = cork->tx_flags; 940 941 /* 942 * Find where to start putting bytes. 943 */ 944 data = skb_put(skb, fraglen + exthdrlen); 945 skb_set_network_header(skb, exthdrlen); 946 skb->transport_header = (skb->network_header + 947 fragheaderlen); 948 data += fragheaderlen + exthdrlen; 949 950 if (fraggap) { 951 skb->csum = skb_copy_and_csum_bits( 952 skb_prev, maxfraglen, 953 data + transhdrlen, fraggap, 0); 954 skb_prev->csum = csum_sub(skb_prev->csum, 955 skb->csum); 956 data += fraggap; 957 pskb_trim_unique(skb_prev, maxfraglen); 958 } 959 960 copy = datalen - transhdrlen - fraggap; 961 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 962 err = -EFAULT; 963 kfree_skb(skb); 964 goto error; 965 } 966 967 offset += copy; 968 length -= datalen - fraggap; 969 transhdrlen = 0; 970 exthdrlen = 0; 971 csummode = CHECKSUM_NONE; 972 973 /* 974 * Put the packet on the pending queue. 975 */ 976 __skb_queue_tail(queue, skb); 977 continue; 978 } 979 980 if (copy > length) 981 copy = length; 982 983 if (!(rt->dst.dev->features&NETIF_F_SG)) { 984 unsigned int off; 985 986 off = skb->len; 987 if (getfrag(from, skb_put(skb, copy), 988 offset, copy, off, skb) < 0) { 989 __skb_trim(skb, off); 990 err = -EFAULT; 991 goto error; 992 } 993 } else { 994 int i = skb_shinfo(skb)->nr_frags; 995 996 err = -ENOMEM; 997 if (!sk_page_frag_refill(sk, pfrag)) 998 goto error; 999 1000 if (!skb_can_coalesce(skb, i, pfrag->page, 1001 pfrag->offset)) { 1002 err = -EMSGSIZE; 1003 if (i == MAX_SKB_FRAGS) 1004 goto error; 1005 1006 __skb_fill_page_desc(skb, i, pfrag->page, 1007 pfrag->offset, 0); 1008 skb_shinfo(skb)->nr_frags = ++i; 1009 get_page(pfrag->page); 1010 } 1011 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1012 if (getfrag(from, 1013 page_address(pfrag->page) + pfrag->offset, 1014 offset, copy, skb->len, skb) < 0) 1015 goto error_efault; 1016 1017 pfrag->offset += copy; 1018 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1019 skb->len += copy; 1020 skb->data_len += copy; 1021 skb->truesize += copy; 1022 atomic_add(copy, &sk->sk_wmem_alloc); 1023 } 1024 offset += copy; 1025 length -= copy; 1026 } 1027 1028 return 0; 1029 1030 error_efault: 1031 err = -EFAULT; 1032 error: 1033 cork->length -= length; 1034 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1035 return err; 1036 } 1037 1038 static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, 1039 struct ipcm_cookie *ipc, struct rtable **rtp) 1040 { 1041 struct inet_sock *inet = inet_sk(sk); 1042 struct ip_options_rcu *opt; 1043 struct rtable *rt; 1044 1045 /* 1046 * setup for corking. 1047 */ 1048 opt = ipc->opt; 1049 if (opt) { 1050 if (cork->opt == NULL) { 1051 cork->opt = kmalloc(sizeof(struct ip_options) + 40, 1052 sk->sk_allocation); 1053 if (unlikely(cork->opt == NULL)) 1054 return -ENOBUFS; 1055 } 1056 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); 1057 cork->flags |= IPCORK_OPT; 1058 cork->addr = ipc->addr; 1059 } 1060 rt = *rtp; 1061 if (unlikely(!rt)) 1062 return -EFAULT; 1063 /* 1064 * We steal reference to this route, caller should not release it 1065 */ 1066 *rtp = NULL; 1067 cork->fragsize = inet->pmtudisc == IP_PMTUDISC_PROBE ? 1068 rt->dst.dev->mtu : dst_mtu(&rt->dst); 1069 cork->dst = &rt->dst; 1070 cork->length = 0; 1071 cork->tx_flags = ipc->tx_flags; 1072 1073 return 0; 1074 } 1075 1076 /* 1077 * ip_append_data() and ip_append_page() can make one large IP datagram 1078 * from many pieces of data. Each pieces will be holded on the socket 1079 * until ip_push_pending_frames() is called. Each piece can be a page 1080 * or non-page data. 1081 * 1082 * Not only UDP, other transport protocols - e.g. raw sockets - can use 1083 * this interface potentially. 1084 * 1085 * LATER: length must be adjusted by pad at tail, when it is required. 1086 */ 1087 int ip_append_data(struct sock *sk, struct flowi4 *fl4, 1088 int getfrag(void *from, char *to, int offset, int len, 1089 int odd, struct sk_buff *skb), 1090 void *from, int length, int transhdrlen, 1091 struct ipcm_cookie *ipc, struct rtable **rtp, 1092 unsigned int flags) 1093 { 1094 struct inet_sock *inet = inet_sk(sk); 1095 int err; 1096 1097 if (flags&MSG_PROBE) 1098 return 0; 1099 1100 if (skb_queue_empty(&sk->sk_write_queue)) { 1101 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); 1102 if (err) 1103 return err; 1104 } else { 1105 transhdrlen = 0; 1106 } 1107 1108 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, 1109 sk_page_frag(sk), getfrag, 1110 from, length, transhdrlen, flags); 1111 } 1112 1113 ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, 1114 int offset, size_t size, int flags) 1115 { 1116 struct inet_sock *inet = inet_sk(sk); 1117 struct sk_buff *skb; 1118 struct rtable *rt; 1119 struct ip_options *opt = NULL; 1120 struct inet_cork *cork; 1121 int hh_len; 1122 int mtu; 1123 int len; 1124 int err; 1125 unsigned int maxfraglen, fragheaderlen, fraggap; 1126 1127 if (inet->hdrincl) 1128 return -EPERM; 1129 1130 if (flags&MSG_PROBE) 1131 return 0; 1132 1133 if (skb_queue_empty(&sk->sk_write_queue)) 1134 return -EINVAL; 1135 1136 cork = &inet->cork.base; 1137 rt = (struct rtable *)cork->dst; 1138 if (cork->flags & IPCORK_OPT) 1139 opt = cork->opt; 1140 1141 if (!(rt->dst.dev->features&NETIF_F_SG)) 1142 return -EOPNOTSUPP; 1143 1144 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1145 mtu = cork->fragsize; 1146 1147 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); 1148 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; 1149 1150 if (cork->length + size > 0xFFFF - fragheaderlen) { 1151 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, mtu); 1152 return -EMSGSIZE; 1153 } 1154 1155 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) 1156 return -EINVAL; 1157 1158 cork->length += size; 1159 if ((size + skb->len > mtu) && 1160 (sk->sk_protocol == IPPROTO_UDP) && 1161 (rt->dst.dev->features & NETIF_F_UFO)) { 1162 skb_shinfo(skb)->gso_size = mtu - fragheaderlen; 1163 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 1164 } 1165 1166 1167 while (size > 0) { 1168 int i; 1169 1170 if (skb_is_gso(skb)) 1171 len = size; 1172 else { 1173 1174 /* Check if the remaining data fits into current packet. */ 1175 len = mtu - skb->len; 1176 if (len < size) 1177 len = maxfraglen - skb->len; 1178 } 1179 if (len <= 0) { 1180 struct sk_buff *skb_prev; 1181 int alloclen; 1182 1183 skb_prev = skb; 1184 fraggap = skb_prev->len - maxfraglen; 1185 1186 alloclen = fragheaderlen + hh_len + fraggap + 15; 1187 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); 1188 if (unlikely(!skb)) { 1189 err = -ENOBUFS; 1190 goto error; 1191 } 1192 1193 /* 1194 * Fill in the control structures 1195 */ 1196 skb->ip_summed = CHECKSUM_NONE; 1197 skb->csum = 0; 1198 skb_reserve(skb, hh_len); 1199 1200 /* 1201 * Find where to start putting bytes. 1202 */ 1203 skb_put(skb, fragheaderlen + fraggap); 1204 skb_reset_network_header(skb); 1205 skb->transport_header = (skb->network_header + 1206 fragheaderlen); 1207 if (fraggap) { 1208 skb->csum = skb_copy_and_csum_bits(skb_prev, 1209 maxfraglen, 1210 skb_transport_header(skb), 1211 fraggap, 0); 1212 skb_prev->csum = csum_sub(skb_prev->csum, 1213 skb->csum); 1214 pskb_trim_unique(skb_prev, maxfraglen); 1215 } 1216 1217 /* 1218 * Put the packet on the pending queue. 1219 */ 1220 __skb_queue_tail(&sk->sk_write_queue, skb); 1221 continue; 1222 } 1223 1224 i = skb_shinfo(skb)->nr_frags; 1225 if (len > size) 1226 len = size; 1227 if (skb_can_coalesce(skb, i, page, offset)) { 1228 skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len); 1229 } else if (i < MAX_SKB_FRAGS) { 1230 get_page(page); 1231 skb_fill_page_desc(skb, i, page, offset, len); 1232 } else { 1233 err = -EMSGSIZE; 1234 goto error; 1235 } 1236 1237 if (skb->ip_summed == CHECKSUM_NONE) { 1238 __wsum csum; 1239 csum = csum_page(page, offset, len); 1240 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1241 } 1242 1243 skb->len += len; 1244 skb->data_len += len; 1245 skb->truesize += len; 1246 atomic_add(len, &sk->sk_wmem_alloc); 1247 offset += len; 1248 size -= len; 1249 } 1250 return 0; 1251 1252 error: 1253 cork->length -= size; 1254 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); 1255 return err; 1256 } 1257 1258 static void ip_cork_release(struct inet_cork *cork) 1259 { 1260 cork->flags &= ~IPCORK_OPT; 1261 kfree(cork->opt); 1262 cork->opt = NULL; 1263 dst_release(cork->dst); 1264 cork->dst = NULL; 1265 } 1266 1267 /* 1268 * Combined all pending IP fragments on the socket as one IP datagram 1269 * and push them out. 1270 */ 1271 struct sk_buff *__ip_make_skb(struct sock *sk, 1272 struct flowi4 *fl4, 1273 struct sk_buff_head *queue, 1274 struct inet_cork *cork) 1275 { 1276 struct sk_buff *skb, *tmp_skb; 1277 struct sk_buff **tail_skb; 1278 struct inet_sock *inet = inet_sk(sk); 1279 struct net *net = sock_net(sk); 1280 struct ip_options *opt = NULL; 1281 struct rtable *rt = (struct rtable *)cork->dst; 1282 struct iphdr *iph; 1283 __be16 df = 0; 1284 __u8 ttl; 1285 1286 if ((skb = __skb_dequeue(queue)) == NULL) 1287 goto out; 1288 tail_skb = &(skb_shinfo(skb)->frag_list); 1289 1290 /* move skb->data to ip header from ext header */ 1291 if (skb->data < skb_network_header(skb)) 1292 __skb_pull(skb, skb_network_offset(skb)); 1293 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1294 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1295 *tail_skb = tmp_skb; 1296 tail_skb = &(tmp_skb->next); 1297 skb->len += tmp_skb->len; 1298 skb->data_len += tmp_skb->len; 1299 skb->truesize += tmp_skb->truesize; 1300 tmp_skb->destructor = NULL; 1301 tmp_skb->sk = NULL; 1302 } 1303 1304 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow 1305 * to fragment the frame generated here. No matter, what transforms 1306 * how transforms change size of the packet, it will come out. 1307 */ 1308 if (inet->pmtudisc < IP_PMTUDISC_DO) 1309 skb->local_df = 1; 1310 1311 /* DF bit is set when we want to see DF on outgoing frames. 1312 * If local_df is set too, we still allow to fragment this frame 1313 * locally. */ 1314 if (inet->pmtudisc >= IP_PMTUDISC_DO || 1315 (skb->len <= dst_mtu(&rt->dst) && 1316 ip_dont_fragment(sk, &rt->dst))) 1317 df = htons(IP_DF); 1318 1319 if (cork->flags & IPCORK_OPT) 1320 opt = cork->opt; 1321 1322 if (rt->rt_type == RTN_MULTICAST) 1323 ttl = inet->mc_ttl; 1324 else 1325 ttl = ip_select_ttl(inet, &rt->dst); 1326 1327 iph = (struct iphdr *)skb->data; 1328 iph->version = 4; 1329 iph->ihl = 5; 1330 iph->tos = inet->tos; 1331 iph->frag_off = df; 1332 iph->ttl = ttl; 1333 iph->protocol = sk->sk_protocol; 1334 ip_copy_addrs(iph, fl4); 1335 ip_select_ident(iph, &rt->dst, sk); 1336 1337 if (opt) { 1338 iph->ihl += opt->optlen>>2; 1339 ip_options_build(skb, opt, cork->addr, rt, 0); 1340 } 1341 1342 skb->priority = sk->sk_priority; 1343 skb->mark = sk->sk_mark; 1344 /* 1345 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec 1346 * on dst refcount 1347 */ 1348 cork->dst = NULL; 1349 skb_dst_set(skb, &rt->dst); 1350 1351 if (iph->protocol == IPPROTO_ICMP) 1352 icmp_out_count(net, ((struct icmphdr *) 1353 skb_transport_header(skb))->type); 1354 1355 ip_cork_release(cork); 1356 out: 1357 return skb; 1358 } 1359 1360 int ip_send_skb(struct net *net, struct sk_buff *skb) 1361 { 1362 int err; 1363 1364 err = ip_local_out(skb); 1365 if (err) { 1366 if (err > 0) 1367 err = net_xmit_errno(err); 1368 if (err) 1369 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); 1370 } 1371 1372 return err; 1373 } 1374 1375 int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) 1376 { 1377 struct sk_buff *skb; 1378 1379 skb = ip_finish_skb(sk, fl4); 1380 if (!skb) 1381 return 0; 1382 1383 /* Netfilter gets whole the not fragmented skb. */ 1384 return ip_send_skb(sock_net(sk), skb); 1385 } 1386 1387 /* 1388 * Throw away all pending data on the socket. 1389 */ 1390 static void __ip_flush_pending_frames(struct sock *sk, 1391 struct sk_buff_head *queue, 1392 struct inet_cork *cork) 1393 { 1394 struct sk_buff *skb; 1395 1396 while ((skb = __skb_dequeue_tail(queue)) != NULL) 1397 kfree_skb(skb); 1398 1399 ip_cork_release(cork); 1400 } 1401 1402 void ip_flush_pending_frames(struct sock *sk) 1403 { 1404 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); 1405 } 1406 1407 struct sk_buff *ip_make_skb(struct sock *sk, 1408 struct flowi4 *fl4, 1409 int getfrag(void *from, char *to, int offset, 1410 int len, int odd, struct sk_buff *skb), 1411 void *from, int length, int transhdrlen, 1412 struct ipcm_cookie *ipc, struct rtable **rtp, 1413 unsigned int flags) 1414 { 1415 struct inet_cork cork; 1416 struct sk_buff_head queue; 1417 int err; 1418 1419 if (flags & MSG_PROBE) 1420 return NULL; 1421 1422 __skb_queue_head_init(&queue); 1423 1424 cork.flags = 0; 1425 cork.addr = 0; 1426 cork.opt = NULL; 1427 err = ip_setup_cork(sk, &cork, ipc, rtp); 1428 if (err) 1429 return ERR_PTR(err); 1430 1431 err = __ip_append_data(sk, fl4, &queue, &cork, 1432 ¤t->task_frag, getfrag, 1433 from, length, transhdrlen, flags); 1434 if (err) { 1435 __ip_flush_pending_frames(sk, &queue, &cork); 1436 return ERR_PTR(err); 1437 } 1438 1439 return __ip_make_skb(sk, fl4, &queue, &cork); 1440 } 1441 1442 /* 1443 * Fetch data from kernel space and fill in checksum if needed. 1444 */ 1445 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 1446 int len, int odd, struct sk_buff *skb) 1447 { 1448 __wsum csum; 1449 1450 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); 1451 skb->csum = csum_block_add(skb->csum, csum, odd); 1452 return 0; 1453 } 1454 1455 /* 1456 * Generic function to send a packet as reply to another packet. 1457 * Used to send some TCP resets/acks so far. 1458 * 1459 * Use a fake percpu inet socket to avoid false sharing and contention. 1460 */ 1461 static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = { 1462 .sk = { 1463 .__sk_common = { 1464 .skc_refcnt = ATOMIC_INIT(1), 1465 }, 1466 .sk_wmem_alloc = ATOMIC_INIT(1), 1467 .sk_allocation = GFP_ATOMIC, 1468 .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE), 1469 }, 1470 .pmtudisc = IP_PMTUDISC_WANT, 1471 .uc_ttl = -1, 1472 }; 1473 1474 void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr, 1475 __be32 saddr, const struct ip_reply_arg *arg, 1476 unsigned int len) 1477 { 1478 struct ip_options_data replyopts; 1479 struct ipcm_cookie ipc; 1480 struct flowi4 fl4; 1481 struct rtable *rt = skb_rtable(skb); 1482 struct sk_buff *nskb; 1483 struct sock *sk; 1484 struct inet_sock *inet; 1485 1486 if (ip_options_echo(&replyopts.opt.opt, skb)) 1487 return; 1488 1489 ipc.addr = daddr; 1490 ipc.opt = NULL; 1491 ipc.tx_flags = 0; 1492 1493 if (replyopts.opt.opt.optlen) { 1494 ipc.opt = &replyopts.opt; 1495 1496 if (replyopts.opt.opt.srr) 1497 daddr = replyopts.opt.opt.faddr; 1498 } 1499 1500 flowi4_init_output(&fl4, arg->bound_dev_if, 0, 1501 RT_TOS(arg->tos), 1502 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, 1503 ip_reply_arg_flowi_flags(arg), 1504 daddr, saddr, 1505 tcp_hdr(skb)->source, tcp_hdr(skb)->dest); 1506 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 1507 rt = ip_route_output_key(net, &fl4); 1508 if (IS_ERR(rt)) 1509 return; 1510 1511 inet = &get_cpu_var(unicast_sock); 1512 1513 inet->tos = arg->tos; 1514 sk = &inet->sk; 1515 sk->sk_priority = skb->priority; 1516 sk->sk_protocol = ip_hdr(skb)->protocol; 1517 sk->sk_bound_dev_if = arg->bound_dev_if; 1518 sock_net_set(sk, net); 1519 __skb_queue_head_init(&sk->sk_write_queue); 1520 sk->sk_sndbuf = sysctl_wmem_default; 1521 ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0, 1522 &ipc, &rt, MSG_DONTWAIT); 1523 nskb = skb_peek(&sk->sk_write_queue); 1524 if (nskb) { 1525 if (arg->csumoffset >= 0) 1526 *((__sum16 *)skb_transport_header(nskb) + 1527 arg->csumoffset) = csum_fold(csum_add(nskb->csum, 1528 arg->csum)); 1529 nskb->ip_summed = CHECKSUM_NONE; 1530 skb_orphan(nskb); 1531 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb)); 1532 ip_push_pending_frames(sk, &fl4); 1533 } 1534 1535 put_cpu_var(unicast_sock); 1536 1537 ip_rt_put(rt); 1538 } 1539 1540 void __init ip_init(void) 1541 { 1542 ip_rt_init(); 1543 inet_initpeers(); 1544 1545 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS) 1546 igmp_mc_proc_init(); 1547 #endif 1548 } 1549