1 /* 2 * IPv6 output functions 3 * Linux INET6 implementation 4 * 5 * Authors: 6 * Pedro Roque <roque@di.fc.ul.pt> 7 * 8 * Based on linux/net/ipv4/ip_output.c 9 * 10 * This program is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU General Public License 12 * as published by the Free Software Foundation; either version 13 * 2 of the License, or (at your option) any later version. 14 * 15 * Changes: 16 * A.N.Kuznetsov : airthmetics in fragmentation. 17 * extension headers are implemented. 18 * route changes now work. 19 * ip6_forward does not confuse sniffers. 20 * etc. 21 * 22 * H. von Brand : Added missing #include <linux/string.h> 23 * Imran Patel : frag id should be in NBO 24 * Kazunori MIYAZAWA @USAGI 25 * : add ip6_append_data and related functions 26 * for datagram xmit 27 */ 28 29 #include <linux/errno.h> 30 #include <linux/kernel.h> 31 #include <linux/string.h> 32 #include <linux/socket.h> 33 #include <linux/net.h> 34 #include <linux/netdevice.h> 35 #include <linux/if_arp.h> 36 #include <linux/in6.h> 37 #include <linux/tcp.h> 38 #include <linux/route.h> 39 #include <linux/module.h> 40 #include <linux/slab.h> 41 42 #include <linux/bpf-cgroup.h> 43 #include <linux/netfilter.h> 44 #include <linux/netfilter_ipv6.h> 45 46 #include <net/sock.h> 47 #include <net/snmp.h> 48 49 #include <net/ipv6.h> 50 #include <net/ndisc.h> 51 #include <net/protocol.h> 52 #include <net/ip6_route.h> 53 #include <net/addrconf.h> 54 #include <net/rawv6.h> 55 #include <net/icmp.h> 56 #include <net/xfrm.h> 57 #include <net/checksum.h> 58 #include <linux/mroute6.h> 59 #include <net/l3mdev.h> 60 #include <net/lwtunnel.h> 61 62 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) 63 { 64 struct dst_entry *dst = skb_dst(skb); 65 struct net_device *dev = dst->dev; 66 struct neighbour *neigh; 67 struct in6_addr *nexthop; 68 int ret; 69 70 skb->protocol = htons(ETH_P_IPV6); 71 skb->dev = dev; 72 73 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) { 74 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb)); 75 76 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) && 77 ((mroute6_socket(net, skb) && 78 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) || 79 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr, 80 &ipv6_hdr(skb)->saddr))) { 81 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 82 83 /* Do not check for IFF_ALLMULTI; multicast routing 84 is not supported in any case. 85 */ 86 if (newskb) 87 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, 88 net, sk, newskb, NULL, newskb->dev, 89 dev_loopback_xmit); 90 91 if (ipv6_hdr(skb)->hop_limit == 0) { 92 IP6_INC_STATS(net, idev, 93 IPSTATS_MIB_OUTDISCARDS); 94 kfree_skb(skb); 95 return 0; 96 } 97 } 98 99 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len); 100 101 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <= 102 IPV6_ADDR_SCOPE_NODELOCAL && 103 !(dev->flags & IFF_LOOPBACK)) { 104 kfree_skb(skb); 105 return 0; 106 } 107 } 108 109 if (lwtunnel_xmit_redirect(dst->lwtstate)) { 110 int res = lwtunnel_xmit(skb); 111 112 if (res < 0 || res == LWTUNNEL_XMIT_DONE) 113 return res; 114 } 115 116 rcu_read_lock_bh(); 117 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 118 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 119 if (unlikely(!neigh)) 120 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 121 if (!IS_ERR(neigh)) { 122 sock_confirm_neigh(skb, neigh); 123 ret = neigh_output(neigh, skb); 124 rcu_read_unlock_bh(); 125 return ret; 126 } 127 rcu_read_unlock_bh(); 128 129 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 130 kfree_skb(skb); 131 return -EINVAL; 132 } 133 134 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 135 { 136 int ret; 137 138 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); 139 if (ret) { 140 kfree_skb(skb); 141 return ret; 142 } 143 144 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) || 145 dst_allfrag(skb_dst(skb)) || 146 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size)) 147 return ip6_fragment(net, sk, skb, ip6_finish_output2); 148 else 149 return ip6_finish_output2(net, sk, skb); 150 } 151 152 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb) 153 { 154 struct net_device *dev = skb_dst(skb)->dev; 155 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb)); 156 157 if (unlikely(idev->cnf.disable_ipv6)) { 158 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS); 159 kfree_skb(skb); 160 return 0; 161 } 162 163 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 164 net, sk, skb, NULL, dev, 165 ip6_finish_output, 166 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 167 } 168 169 /* 170 * xmit an sk_buff (used by TCP, SCTP and DCCP) 171 * Note : socket lock is not held for SYNACK packets, but might be modified 172 * by calls to skb_set_owner_w() and ipv6_local_error(), 173 * which are using proper atomic operations or spinlocks. 174 */ 175 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6, 176 __u32 mark, struct ipv6_txoptions *opt, int tclass) 177 { 178 struct net *net = sock_net(sk); 179 const struct ipv6_pinfo *np = inet6_sk(sk); 180 struct in6_addr *first_hop = &fl6->daddr; 181 struct dst_entry *dst = skb_dst(skb); 182 struct ipv6hdr *hdr; 183 u8 proto = fl6->flowi6_proto; 184 int seg_len = skb->len; 185 int hlimit = -1; 186 u32 mtu; 187 188 if (opt) { 189 unsigned int head_room; 190 191 /* First: exthdrs may take lots of space (~8K for now) 192 MAX_HEADER is not enough. 193 */ 194 head_room = opt->opt_nflen + opt->opt_flen; 195 seg_len += head_room; 196 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev); 197 198 if (skb_headroom(skb) < head_room) { 199 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room); 200 if (!skb2) { 201 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 202 IPSTATS_MIB_OUTDISCARDS); 203 kfree_skb(skb); 204 return -ENOBUFS; 205 } 206 consume_skb(skb); 207 skb = skb2; 208 /* skb_set_owner_w() changes sk->sk_wmem_alloc atomically, 209 * it is safe to call in our context (socket lock not held) 210 */ 211 skb_set_owner_w(skb, (struct sock *)sk); 212 } 213 if (opt->opt_flen) 214 ipv6_push_frag_opts(skb, opt, &proto); 215 if (opt->opt_nflen) 216 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop, 217 &fl6->saddr); 218 } 219 220 skb_push(skb, sizeof(struct ipv6hdr)); 221 skb_reset_network_header(skb); 222 hdr = ipv6_hdr(skb); 223 224 /* 225 * Fill in the IPv6 header 226 */ 227 if (np) 228 hlimit = np->hop_limit; 229 if (hlimit < 0) 230 hlimit = ip6_dst_hoplimit(dst); 231 232 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel, 233 np->autoflowlabel, fl6)); 234 235 hdr->payload_len = htons(seg_len); 236 hdr->nexthdr = proto; 237 hdr->hop_limit = hlimit; 238 239 hdr->saddr = fl6->saddr; 240 hdr->daddr = *first_hop; 241 242 skb->protocol = htons(ETH_P_IPV6); 243 skb->priority = sk->sk_priority; 244 skb->mark = mark; 245 246 mtu = dst_mtu(dst); 247 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) { 248 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)), 249 IPSTATS_MIB_OUT, skb->len); 250 251 /* if egress device is enslaved to an L3 master device pass the 252 * skb to its handler for processing 253 */ 254 skb = l3mdev_ip6_out((struct sock *)sk, skb); 255 if (unlikely(!skb)) 256 return 0; 257 258 /* hooks should never assume socket lock is held. 259 * we promote our socket to non const 260 */ 261 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, 262 net, (struct sock *)sk, skb, NULL, dst->dev, 263 dst_output); 264 } 265 266 skb->dev = dst->dev; 267 /* ipv6_local_error() does not require socket lock, 268 * we promote our socket to non const 269 */ 270 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu); 271 272 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS); 273 kfree_skb(skb); 274 return -EMSGSIZE; 275 } 276 EXPORT_SYMBOL(ip6_xmit); 277 278 static int ip6_call_ra_chain(struct sk_buff *skb, int sel) 279 { 280 struct ip6_ra_chain *ra; 281 struct sock *last = NULL; 282 283 read_lock(&ip6_ra_lock); 284 for (ra = ip6_ra_chain; ra; ra = ra->next) { 285 struct sock *sk = ra->sk; 286 if (sk && ra->sel == sel && 287 (!sk->sk_bound_dev_if || 288 sk->sk_bound_dev_if == skb->dev->ifindex)) { 289 if (last) { 290 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 291 if (skb2) 292 rawv6_rcv(last, skb2); 293 } 294 last = sk; 295 } 296 } 297 298 if (last) { 299 rawv6_rcv(last, skb); 300 read_unlock(&ip6_ra_lock); 301 return 1; 302 } 303 read_unlock(&ip6_ra_lock); 304 return 0; 305 } 306 307 static int ip6_forward_proxy_check(struct sk_buff *skb) 308 { 309 struct ipv6hdr *hdr = ipv6_hdr(skb); 310 u8 nexthdr = hdr->nexthdr; 311 __be16 frag_off; 312 int offset; 313 314 if (ipv6_ext_hdr(nexthdr)) { 315 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off); 316 if (offset < 0) 317 return 0; 318 } else 319 offset = sizeof(struct ipv6hdr); 320 321 if (nexthdr == IPPROTO_ICMPV6) { 322 struct icmp6hdr *icmp6; 323 324 if (!pskb_may_pull(skb, (skb_network_header(skb) + 325 offset + 1 - skb->data))) 326 return 0; 327 328 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset); 329 330 switch (icmp6->icmp6_type) { 331 case NDISC_ROUTER_SOLICITATION: 332 case NDISC_ROUTER_ADVERTISEMENT: 333 case NDISC_NEIGHBOUR_SOLICITATION: 334 case NDISC_NEIGHBOUR_ADVERTISEMENT: 335 case NDISC_REDIRECT: 336 /* For reaction involving unicast neighbor discovery 337 * message destined to the proxied address, pass it to 338 * input function. 339 */ 340 return 1; 341 default: 342 break; 343 } 344 } 345 346 /* 347 * The proxying router can't forward traffic sent to a link-local 348 * address, so signal the sender and discard the packet. This 349 * behavior is clarified by the MIPv6 specification. 350 */ 351 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) { 352 dst_link_failure(skb); 353 return -1; 354 } 355 356 return 0; 357 } 358 359 static inline int ip6_forward_finish(struct net *net, struct sock *sk, 360 struct sk_buff *skb) 361 { 362 return dst_output(net, sk, skb); 363 } 364 365 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst) 366 { 367 unsigned int mtu; 368 struct inet6_dev *idev; 369 370 if (dst_metric_locked(dst, RTAX_MTU)) { 371 mtu = dst_metric_raw(dst, RTAX_MTU); 372 if (mtu) 373 return mtu; 374 } 375 376 mtu = IPV6_MIN_MTU; 377 rcu_read_lock(); 378 idev = __in6_dev_get(dst->dev); 379 if (idev) 380 mtu = idev->cnf.mtu6; 381 rcu_read_unlock(); 382 383 return mtu; 384 } 385 386 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu) 387 { 388 if (skb->len <= mtu) 389 return false; 390 391 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */ 392 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu) 393 return true; 394 395 if (skb->ignore_df) 396 return false; 397 398 if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu)) 399 return false; 400 401 return true; 402 } 403 404 int ip6_forward(struct sk_buff *skb) 405 { 406 struct dst_entry *dst = skb_dst(skb); 407 struct ipv6hdr *hdr = ipv6_hdr(skb); 408 struct inet6_skb_parm *opt = IP6CB(skb); 409 struct net *net = dev_net(dst->dev); 410 u32 mtu; 411 412 if (net->ipv6.devconf_all->forwarding == 0) 413 goto error; 414 415 if (skb->pkt_type != PACKET_HOST) 416 goto drop; 417 418 if (unlikely(skb->sk)) 419 goto drop; 420 421 if (skb_warn_if_lro(skb)) 422 goto drop; 423 424 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) { 425 __IP6_INC_STATS(net, ip6_dst_idev(dst), 426 IPSTATS_MIB_INDISCARDS); 427 goto drop; 428 } 429 430 skb_forward_csum(skb); 431 432 /* 433 * We DO NOT make any processing on 434 * RA packets, pushing them to user level AS IS 435 * without ane WARRANTY that application will be able 436 * to interpret them. The reason is that we 437 * cannot make anything clever here. 438 * 439 * We are not end-node, so that if packet contains 440 * AH/ESP, we cannot make anything. 441 * Defragmentation also would be mistake, RA packets 442 * cannot be fragmented, because there is no warranty 443 * that different fragments will go along one path. --ANK 444 */ 445 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) { 446 if (ip6_call_ra_chain(skb, ntohs(opt->ra))) 447 return 0; 448 } 449 450 /* 451 * check and decrement ttl 452 */ 453 if (hdr->hop_limit <= 1) { 454 /* Force OUTPUT device used as source address */ 455 skb->dev = dst->dev; 456 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0); 457 __IP6_INC_STATS(net, ip6_dst_idev(dst), 458 IPSTATS_MIB_INHDRERRORS); 459 460 kfree_skb(skb); 461 return -ETIMEDOUT; 462 } 463 464 /* XXX: idev->cnf.proxy_ndp? */ 465 if (net->ipv6.devconf_all->proxy_ndp && 466 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) { 467 int proxied = ip6_forward_proxy_check(skb); 468 if (proxied > 0) 469 return ip6_input(skb); 470 else if (proxied < 0) { 471 __IP6_INC_STATS(net, ip6_dst_idev(dst), 472 IPSTATS_MIB_INDISCARDS); 473 goto drop; 474 } 475 } 476 477 if (!xfrm6_route_forward(skb)) { 478 __IP6_INC_STATS(net, ip6_dst_idev(dst), 479 IPSTATS_MIB_INDISCARDS); 480 goto drop; 481 } 482 dst = skb_dst(skb); 483 484 /* IPv6 specs say nothing about it, but it is clear that we cannot 485 send redirects to source routed frames. 486 We don't send redirects to frames decapsulated from IPsec. 487 */ 488 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) { 489 struct in6_addr *target = NULL; 490 struct inet_peer *peer; 491 struct rt6_info *rt; 492 493 /* 494 * incoming and outgoing devices are the same 495 * send a redirect. 496 */ 497 498 rt = (struct rt6_info *) dst; 499 if (rt->rt6i_flags & RTF_GATEWAY) 500 target = &rt->rt6i_gateway; 501 else 502 target = &hdr->daddr; 503 504 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1); 505 506 /* Limit redirects both by destination (here) 507 and by source (inside ndisc_send_redirect) 508 */ 509 if (inet_peer_xrlim_allow(peer, 1*HZ)) 510 ndisc_send_redirect(skb, target); 511 if (peer) 512 inet_putpeer(peer); 513 } else { 514 int addrtype = ipv6_addr_type(&hdr->saddr); 515 516 /* This check is security critical. */ 517 if (addrtype == IPV6_ADDR_ANY || 518 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK)) 519 goto error; 520 if (addrtype & IPV6_ADDR_LINKLOCAL) { 521 icmpv6_send(skb, ICMPV6_DEST_UNREACH, 522 ICMPV6_NOT_NEIGHBOUR, 0); 523 goto error; 524 } 525 } 526 527 mtu = ip6_dst_mtu_forward(dst); 528 if (mtu < IPV6_MIN_MTU) 529 mtu = IPV6_MIN_MTU; 530 531 if (ip6_pkt_too_big(skb, mtu)) { 532 /* Again, force OUTPUT device used as source address */ 533 skb->dev = dst->dev; 534 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); 535 __IP6_INC_STATS(net, ip6_dst_idev(dst), 536 IPSTATS_MIB_INTOOBIGERRORS); 537 __IP6_INC_STATS(net, ip6_dst_idev(dst), 538 IPSTATS_MIB_FRAGFAILS); 539 kfree_skb(skb); 540 return -EMSGSIZE; 541 } 542 543 if (skb_cow(skb, dst->dev->hard_header_len)) { 544 __IP6_INC_STATS(net, ip6_dst_idev(dst), 545 IPSTATS_MIB_OUTDISCARDS); 546 goto drop; 547 } 548 549 hdr = ipv6_hdr(skb); 550 551 /* Mangling hops number delayed to point after skb COW */ 552 553 hdr->hop_limit--; 554 555 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS); 556 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len); 557 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, 558 net, NULL, skb, skb->dev, dst->dev, 559 ip6_forward_finish); 560 561 error: 562 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS); 563 drop: 564 kfree_skb(skb); 565 return -EINVAL; 566 } 567 568 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from) 569 { 570 to->pkt_type = from->pkt_type; 571 to->priority = from->priority; 572 to->protocol = from->protocol; 573 skb_dst_drop(to); 574 skb_dst_set(to, dst_clone(skb_dst(from))); 575 to->dev = from->dev; 576 to->mark = from->mark; 577 578 #ifdef CONFIG_NET_SCHED 579 to->tc_index = from->tc_index; 580 #endif 581 nf_copy(to, from); 582 skb_copy_secmark(to, from); 583 } 584 585 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 586 int (*output)(struct net *, struct sock *, struct sk_buff *)) 587 { 588 struct sk_buff *frag; 589 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb); 590 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ? 591 inet6_sk(skb->sk) : NULL; 592 struct ipv6hdr *tmp_hdr; 593 struct frag_hdr *fh; 594 unsigned int mtu, hlen, left, len; 595 int hroom, troom; 596 __be32 frag_id; 597 int ptr, offset = 0, err = 0; 598 u8 *prevhdr, nexthdr = 0; 599 600 err = ip6_find_1stfragopt(skb, &prevhdr); 601 if (err < 0) 602 goto fail; 603 hlen = err; 604 nexthdr = *prevhdr; 605 606 mtu = ip6_skb_dst_mtu(skb); 607 608 /* We must not fragment if the socket is set to force MTU discovery 609 * or if the skb it not generated by a local socket. 610 */ 611 if (unlikely(!skb->ignore_df && skb->len > mtu)) 612 goto fail_toobig; 613 614 if (IP6CB(skb)->frag_max_size) { 615 if (IP6CB(skb)->frag_max_size > mtu) 616 goto fail_toobig; 617 618 /* don't send fragments larger than what we received */ 619 mtu = IP6CB(skb)->frag_max_size; 620 if (mtu < IPV6_MIN_MTU) 621 mtu = IPV6_MIN_MTU; 622 } 623 624 if (np && np->frag_size < mtu) { 625 if (np->frag_size) 626 mtu = np->frag_size; 627 } 628 if (mtu < hlen + sizeof(struct frag_hdr) + 8) 629 goto fail_toobig; 630 mtu -= hlen + sizeof(struct frag_hdr); 631 632 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr, 633 &ipv6_hdr(skb)->saddr); 634 635 if (skb->ip_summed == CHECKSUM_PARTIAL && 636 (err = skb_checksum_help(skb))) 637 goto fail; 638 639 hroom = LL_RESERVED_SPACE(rt->dst.dev); 640 if (skb_has_frag_list(skb)) { 641 unsigned int first_len = skb_pagelen(skb); 642 struct sk_buff *frag2; 643 644 if (first_len - hlen > mtu || 645 ((first_len - hlen) & 7) || 646 skb_cloned(skb) || 647 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr))) 648 goto slow_path; 649 650 skb_walk_frags(skb, frag) { 651 /* Correct geometry. */ 652 if (frag->len > mtu || 653 ((frag->len & 7) && frag->next) || 654 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr))) 655 goto slow_path_clean; 656 657 /* Partially cloned skb? */ 658 if (skb_shared(frag)) 659 goto slow_path_clean; 660 661 BUG_ON(frag->sk); 662 if (skb->sk) { 663 frag->sk = skb->sk; 664 frag->destructor = sock_wfree; 665 } 666 skb->truesize -= frag->truesize; 667 } 668 669 err = 0; 670 offset = 0; 671 /* BUILD HEADER */ 672 673 *prevhdr = NEXTHDR_FRAGMENT; 674 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC); 675 if (!tmp_hdr) { 676 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 677 IPSTATS_MIB_FRAGFAILS); 678 err = -ENOMEM; 679 goto fail; 680 } 681 frag = skb_shinfo(skb)->frag_list; 682 skb_frag_list_init(skb); 683 684 __skb_pull(skb, hlen); 685 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr)); 686 __skb_push(skb, hlen); 687 skb_reset_network_header(skb); 688 memcpy(skb_network_header(skb), tmp_hdr, hlen); 689 690 fh->nexthdr = nexthdr; 691 fh->reserved = 0; 692 fh->frag_off = htons(IP6_MF); 693 fh->identification = frag_id; 694 695 first_len = skb_pagelen(skb); 696 skb->data_len = first_len - skb_headlen(skb); 697 skb->len = first_len; 698 ipv6_hdr(skb)->payload_len = htons(first_len - 699 sizeof(struct ipv6hdr)); 700 701 dst_hold(&rt->dst); 702 703 for (;;) { 704 /* Prepare header of the next frame, 705 * before previous one went down. */ 706 if (frag) { 707 frag->ip_summed = CHECKSUM_NONE; 708 skb_reset_transport_header(frag); 709 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr)); 710 __skb_push(frag, hlen); 711 skb_reset_network_header(frag); 712 memcpy(skb_network_header(frag), tmp_hdr, 713 hlen); 714 offset += skb->len - hlen - sizeof(struct frag_hdr); 715 fh->nexthdr = nexthdr; 716 fh->reserved = 0; 717 fh->frag_off = htons(offset); 718 if (frag->next) 719 fh->frag_off |= htons(IP6_MF); 720 fh->identification = frag_id; 721 ipv6_hdr(frag)->payload_len = 722 htons(frag->len - 723 sizeof(struct ipv6hdr)); 724 ip6_copy_metadata(frag, skb); 725 } 726 727 err = output(net, sk, skb); 728 if (!err) 729 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst), 730 IPSTATS_MIB_FRAGCREATES); 731 732 if (err || !frag) 733 break; 734 735 skb = frag; 736 frag = skb->next; 737 skb->next = NULL; 738 } 739 740 kfree(tmp_hdr); 741 742 if (err == 0) { 743 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst), 744 IPSTATS_MIB_FRAGOKS); 745 ip6_rt_put(rt); 746 return 0; 747 } 748 749 kfree_skb_list(frag); 750 751 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst), 752 IPSTATS_MIB_FRAGFAILS); 753 ip6_rt_put(rt); 754 return err; 755 756 slow_path_clean: 757 skb_walk_frags(skb, frag2) { 758 if (frag2 == frag) 759 break; 760 frag2->sk = NULL; 761 frag2->destructor = NULL; 762 skb->truesize += frag2->truesize; 763 } 764 } 765 766 slow_path: 767 left = skb->len - hlen; /* Space per frame */ 768 ptr = hlen; /* Where to start from */ 769 770 /* 771 * Fragment the datagram. 772 */ 773 774 troom = rt->dst.dev->needed_tailroom; 775 776 /* 777 * Keep copying data until we run out. 778 */ 779 while (left > 0) { 780 u8 *fragnexthdr_offset; 781 782 len = left; 783 /* IF: it doesn't fit, use 'mtu' - the data space left */ 784 if (len > mtu) 785 len = mtu; 786 /* IF: we are not sending up to and including the packet end 787 then align the next start on an eight byte boundary */ 788 if (len < left) { 789 len &= ~7; 790 } 791 792 /* Allocate buffer */ 793 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) + 794 hroom + troom, GFP_ATOMIC); 795 if (!frag) { 796 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 797 IPSTATS_MIB_FRAGFAILS); 798 err = -ENOMEM; 799 goto fail; 800 } 801 802 /* 803 * Set up data on packet 804 */ 805 806 ip6_copy_metadata(frag, skb); 807 skb_reserve(frag, hroom); 808 skb_put(frag, len + hlen + sizeof(struct frag_hdr)); 809 skb_reset_network_header(frag); 810 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen); 811 frag->transport_header = (frag->network_header + hlen + 812 sizeof(struct frag_hdr)); 813 814 /* 815 * Charge the memory for the fragment to any owner 816 * it might possess 817 */ 818 if (skb->sk) 819 skb_set_owner_w(frag, skb->sk); 820 821 /* 822 * Copy the packet header into the new buffer. 823 */ 824 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen); 825 826 fragnexthdr_offset = skb_network_header(frag); 827 fragnexthdr_offset += prevhdr - skb_network_header(skb); 828 *fragnexthdr_offset = NEXTHDR_FRAGMENT; 829 830 /* 831 * Build fragment header. 832 */ 833 fh->nexthdr = nexthdr; 834 fh->reserved = 0; 835 fh->identification = frag_id; 836 837 /* 838 * Copy a block of the IP datagram. 839 */ 840 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag), 841 len)); 842 left -= len; 843 844 fh->frag_off = htons(offset); 845 if (left > 0) 846 fh->frag_off |= htons(IP6_MF); 847 ipv6_hdr(frag)->payload_len = htons(frag->len - 848 sizeof(struct ipv6hdr)); 849 850 ptr += len; 851 offset += len; 852 853 /* 854 * Put this fragment into the sending queue. 855 */ 856 err = output(net, sk, frag); 857 if (err) 858 goto fail; 859 860 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 861 IPSTATS_MIB_FRAGCREATES); 862 } 863 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 864 IPSTATS_MIB_FRAGOKS); 865 consume_skb(skb); 866 return err; 867 868 fail_toobig: 869 if (skb->sk && dst_allfrag(skb_dst(skb))) 870 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK); 871 872 skb->dev = skb_dst(skb)->dev; 873 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); 874 err = -EMSGSIZE; 875 876 fail: 877 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 878 IPSTATS_MIB_FRAGFAILS); 879 kfree_skb(skb); 880 return err; 881 } 882 883 static inline int ip6_rt_check(const struct rt6key *rt_key, 884 const struct in6_addr *fl_addr, 885 const struct in6_addr *addr_cache) 886 { 887 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) && 888 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache)); 889 } 890 891 static struct dst_entry *ip6_sk_dst_check(struct sock *sk, 892 struct dst_entry *dst, 893 const struct flowi6 *fl6) 894 { 895 struct ipv6_pinfo *np = inet6_sk(sk); 896 struct rt6_info *rt; 897 898 if (!dst) 899 goto out; 900 901 if (dst->ops->family != AF_INET6) { 902 dst_release(dst); 903 return NULL; 904 } 905 906 rt = (struct rt6_info *)dst; 907 /* Yes, checking route validity in not connected 908 * case is not very simple. Take into account, 909 * that we do not support routing by source, TOS, 910 * and MSG_DONTROUTE --ANK (980726) 911 * 912 * 1. ip6_rt_check(): If route was host route, 913 * check that cached destination is current. 914 * If it is network route, we still may 915 * check its validity using saved pointer 916 * to the last used address: daddr_cache. 917 * We do not want to save whole address now, 918 * (because main consumer of this service 919 * is tcp, which has not this problem), 920 * so that the last trick works only on connected 921 * sockets. 922 * 2. oif also should be the same. 923 */ 924 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) || 925 #ifdef CONFIG_IPV6_SUBTREES 926 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) || 927 #endif 928 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) && 929 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) { 930 dst_release(dst); 931 dst = NULL; 932 } 933 934 out: 935 return dst; 936 } 937 938 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk, 939 struct dst_entry **dst, struct flowi6 *fl6) 940 { 941 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD 942 struct neighbour *n; 943 struct rt6_info *rt; 944 #endif 945 int err; 946 int flags = 0; 947 948 /* The correct way to handle this would be to do 949 * ip6_route_get_saddr, and then ip6_route_output; however, 950 * the route-specific preferred source forces the 951 * ip6_route_output call _before_ ip6_route_get_saddr. 952 * 953 * In source specific routing (no src=any default route), 954 * ip6_route_output will fail given src=any saddr, though, so 955 * that's why we try it again later. 956 */ 957 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) { 958 struct rt6_info *rt; 959 bool had_dst = *dst != NULL; 960 961 if (!had_dst) 962 *dst = ip6_route_output(net, sk, fl6); 963 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst; 964 err = ip6_route_get_saddr(net, rt, &fl6->daddr, 965 sk ? inet6_sk(sk)->srcprefs : 0, 966 &fl6->saddr); 967 if (err) 968 goto out_err_release; 969 970 /* If we had an erroneous initial result, pretend it 971 * never existed and let the SA-enabled version take 972 * over. 973 */ 974 if (!had_dst && (*dst)->error) { 975 dst_release(*dst); 976 *dst = NULL; 977 } 978 979 if (fl6->flowi6_oif) 980 flags |= RT6_LOOKUP_F_IFACE; 981 } 982 983 if (!*dst) 984 *dst = ip6_route_output_flags(net, sk, fl6, flags); 985 986 err = (*dst)->error; 987 if (err) 988 goto out_err_release; 989 990 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD 991 /* 992 * Here if the dst entry we've looked up 993 * has a neighbour entry that is in the INCOMPLETE 994 * state and the src address from the flow is 995 * marked as OPTIMISTIC, we release the found 996 * dst entry and replace it instead with the 997 * dst entry of the nexthop router 998 */ 999 rt = (struct rt6_info *) *dst; 1000 rcu_read_lock_bh(); 1001 n = __ipv6_neigh_lookup_noref(rt->dst.dev, 1002 rt6_nexthop(rt, &fl6->daddr)); 1003 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0; 1004 rcu_read_unlock_bh(); 1005 1006 if (err) { 1007 struct inet6_ifaddr *ifp; 1008 struct flowi6 fl_gw6; 1009 int redirect; 1010 1011 ifp = ipv6_get_ifaddr(net, &fl6->saddr, 1012 (*dst)->dev, 1); 1013 1014 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC); 1015 if (ifp) 1016 in6_ifa_put(ifp); 1017 1018 if (redirect) { 1019 /* 1020 * We need to get the dst entry for the 1021 * default router instead 1022 */ 1023 dst_release(*dst); 1024 memcpy(&fl_gw6, fl6, sizeof(struct flowi6)); 1025 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr)); 1026 *dst = ip6_route_output(net, sk, &fl_gw6); 1027 err = (*dst)->error; 1028 if (err) 1029 goto out_err_release; 1030 } 1031 } 1032 #endif 1033 if (ipv6_addr_v4mapped(&fl6->saddr) && 1034 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) { 1035 err = -EAFNOSUPPORT; 1036 goto out_err_release; 1037 } 1038 1039 return 0; 1040 1041 out_err_release: 1042 dst_release(*dst); 1043 *dst = NULL; 1044 1045 if (err == -ENETUNREACH) 1046 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES); 1047 return err; 1048 } 1049 1050 /** 1051 * ip6_dst_lookup - perform route lookup on flow 1052 * @sk: socket which provides route info 1053 * @dst: pointer to dst_entry * for result 1054 * @fl6: flow to lookup 1055 * 1056 * This function performs a route lookup on the given flow. 1057 * 1058 * It returns zero on success, or a standard errno code on error. 1059 */ 1060 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst, 1061 struct flowi6 *fl6) 1062 { 1063 *dst = NULL; 1064 return ip6_dst_lookup_tail(net, sk, dst, fl6); 1065 } 1066 EXPORT_SYMBOL_GPL(ip6_dst_lookup); 1067 1068 /** 1069 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec 1070 * @sk: socket which provides route info 1071 * @fl6: flow to lookup 1072 * @final_dst: final destination address for ipsec lookup 1073 * 1074 * This function performs a route lookup on the given flow. 1075 * 1076 * It returns a valid dst pointer on success, or a pointer encoded 1077 * error code. 1078 */ 1079 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6, 1080 const struct in6_addr *final_dst) 1081 { 1082 struct dst_entry *dst = NULL; 1083 int err; 1084 1085 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6); 1086 if (err) 1087 return ERR_PTR(err); 1088 if (final_dst) 1089 fl6->daddr = *final_dst; 1090 1091 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0); 1092 } 1093 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow); 1094 1095 /** 1096 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow 1097 * @sk: socket which provides the dst cache and route info 1098 * @fl6: flow to lookup 1099 * @final_dst: final destination address for ipsec lookup 1100 * 1101 * This function performs a route lookup on the given flow with the 1102 * possibility of using the cached route in the socket if it is valid. 1103 * It will take the socket dst lock when operating on the dst cache. 1104 * As a result, this function can only be used in process context. 1105 * 1106 * It returns a valid dst pointer on success, or a pointer encoded 1107 * error code. 1108 */ 1109 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6, 1110 const struct in6_addr *final_dst) 1111 { 1112 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie); 1113 1114 dst = ip6_sk_dst_check(sk, dst, fl6); 1115 if (!dst) 1116 dst = ip6_dst_lookup_flow(sk, fl6, final_dst); 1117 1118 return dst; 1119 } 1120 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow); 1121 1122 static inline int ip6_ufo_append_data(struct sock *sk, 1123 struct sk_buff_head *queue, 1124 int getfrag(void *from, char *to, int offset, int len, 1125 int odd, struct sk_buff *skb), 1126 void *from, int length, int hh_len, int fragheaderlen, 1127 int exthdrlen, int transhdrlen, int mtu, 1128 unsigned int flags, const struct flowi6 *fl6) 1129 1130 { 1131 struct sk_buff *skb; 1132 int err; 1133 1134 /* There is support for UDP large send offload by network 1135 * device, so create one single skb packet containing complete 1136 * udp datagram 1137 */ 1138 skb = skb_peek_tail(queue); 1139 if (!skb) { 1140 skb = sock_alloc_send_skb(sk, 1141 hh_len + fragheaderlen + transhdrlen + 20, 1142 (flags & MSG_DONTWAIT), &err); 1143 if (!skb) 1144 return err; 1145 1146 /* reserve space for Hardware header */ 1147 skb_reserve(skb, hh_len); 1148 1149 /* create space for UDP/IP header */ 1150 skb_put(skb, fragheaderlen + transhdrlen); 1151 1152 /* initialize network header pointer */ 1153 skb_set_network_header(skb, exthdrlen); 1154 1155 /* initialize protocol header pointer */ 1156 skb->transport_header = skb->network_header + fragheaderlen; 1157 1158 skb->protocol = htons(ETH_P_IPV6); 1159 skb->csum = 0; 1160 1161 if (flags & MSG_CONFIRM) 1162 skb_set_dst_pending_confirm(skb, 1); 1163 1164 __skb_queue_tail(queue, skb); 1165 } else if (skb_is_gso(skb)) { 1166 goto append; 1167 } 1168 1169 skb->ip_summed = CHECKSUM_PARTIAL; 1170 /* Specify the length of each IPv6 datagram fragment. 1171 * It has to be a multiple of 8. 1172 */ 1173 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen - 1174 sizeof(struct frag_hdr)) & ~7; 1175 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 1176 skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk), 1177 &fl6->daddr, 1178 &fl6->saddr); 1179 1180 append: 1181 return skb_append_datato_frags(sk, skb, getfrag, from, 1182 (length - transhdrlen)); 1183 } 1184 1185 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src, 1186 gfp_t gfp) 1187 { 1188 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL; 1189 } 1190 1191 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src, 1192 gfp_t gfp) 1193 { 1194 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL; 1195 } 1196 1197 static void ip6_append_data_mtu(unsigned int *mtu, 1198 int *maxfraglen, 1199 unsigned int fragheaderlen, 1200 struct sk_buff *skb, 1201 struct rt6_info *rt, 1202 unsigned int orig_mtu) 1203 { 1204 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) { 1205 if (!skb) { 1206 /* first fragment, reserve header_len */ 1207 *mtu = orig_mtu - rt->dst.header_len; 1208 1209 } else { 1210 /* 1211 * this fragment is not first, the headers 1212 * space is regarded as data space. 1213 */ 1214 *mtu = orig_mtu; 1215 } 1216 *maxfraglen = ((*mtu - fragheaderlen) & ~7) 1217 + fragheaderlen - sizeof(struct frag_hdr); 1218 } 1219 } 1220 1221 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork, 1222 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6, 1223 struct rt6_info *rt, struct flowi6 *fl6) 1224 { 1225 struct ipv6_pinfo *np = inet6_sk(sk); 1226 unsigned int mtu; 1227 struct ipv6_txoptions *opt = ipc6->opt; 1228 1229 /* 1230 * setup for corking 1231 */ 1232 if (opt) { 1233 if (WARN_ON(v6_cork->opt)) 1234 return -EINVAL; 1235 1236 v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation); 1237 if (unlikely(!v6_cork->opt)) 1238 return -ENOBUFS; 1239 1240 v6_cork->opt->tot_len = opt->tot_len; 1241 v6_cork->opt->opt_flen = opt->opt_flen; 1242 v6_cork->opt->opt_nflen = opt->opt_nflen; 1243 1244 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt, 1245 sk->sk_allocation); 1246 if (opt->dst0opt && !v6_cork->opt->dst0opt) 1247 return -ENOBUFS; 1248 1249 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt, 1250 sk->sk_allocation); 1251 if (opt->dst1opt && !v6_cork->opt->dst1opt) 1252 return -ENOBUFS; 1253 1254 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt, 1255 sk->sk_allocation); 1256 if (opt->hopopt && !v6_cork->opt->hopopt) 1257 return -ENOBUFS; 1258 1259 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt, 1260 sk->sk_allocation); 1261 if (opt->srcrt && !v6_cork->opt->srcrt) 1262 return -ENOBUFS; 1263 1264 /* need source address above miyazawa*/ 1265 } 1266 dst_hold(&rt->dst); 1267 cork->base.dst = &rt->dst; 1268 cork->fl.u.ip6 = *fl6; 1269 v6_cork->hop_limit = ipc6->hlimit; 1270 v6_cork->tclass = ipc6->tclass; 1271 if (rt->dst.flags & DST_XFRM_TUNNEL) 1272 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ? 1273 rt->dst.dev->mtu : dst_mtu(&rt->dst); 1274 else 1275 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ? 1276 rt->dst.dev->mtu : dst_mtu(rt->dst.path); 1277 if (np->frag_size < mtu) { 1278 if (np->frag_size) 1279 mtu = np->frag_size; 1280 } 1281 cork->base.fragsize = mtu; 1282 if (dst_allfrag(rt->dst.path)) 1283 cork->base.flags |= IPCORK_ALLFRAG; 1284 cork->base.length = 0; 1285 1286 return 0; 1287 } 1288 1289 static int __ip6_append_data(struct sock *sk, 1290 struct flowi6 *fl6, 1291 struct sk_buff_head *queue, 1292 struct inet_cork *cork, 1293 struct inet6_cork *v6_cork, 1294 struct page_frag *pfrag, 1295 int getfrag(void *from, char *to, int offset, 1296 int len, int odd, struct sk_buff *skb), 1297 void *from, int length, int transhdrlen, 1298 unsigned int flags, struct ipcm6_cookie *ipc6, 1299 const struct sockcm_cookie *sockc) 1300 { 1301 struct sk_buff *skb, *skb_prev = NULL; 1302 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu; 1303 int exthdrlen = 0; 1304 int dst_exthdrlen = 0; 1305 int hh_len; 1306 int copy; 1307 int err; 1308 int offset = 0; 1309 __u8 tx_flags = 0; 1310 u32 tskey = 0; 1311 struct rt6_info *rt = (struct rt6_info *)cork->dst; 1312 struct ipv6_txoptions *opt = v6_cork->opt; 1313 int csummode = CHECKSUM_NONE; 1314 unsigned int maxnonfragsize, headersize; 1315 1316 skb = skb_peek_tail(queue); 1317 if (!skb) { 1318 exthdrlen = opt ? opt->opt_flen : 0; 1319 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len; 1320 } 1321 1322 mtu = cork->fragsize; 1323 orig_mtu = mtu; 1324 1325 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1326 1327 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len + 1328 (opt ? opt->opt_nflen : 0); 1329 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - 1330 sizeof(struct frag_hdr); 1331 1332 headersize = sizeof(struct ipv6hdr) + 1333 (opt ? opt->opt_flen + opt->opt_nflen : 0) + 1334 (dst_allfrag(&rt->dst) ? 1335 sizeof(struct frag_hdr) : 0) + 1336 rt->rt6i_nfheader_len; 1337 1338 if (cork->length + length > mtu - headersize && ipc6->dontfrag && 1339 (sk->sk_protocol == IPPROTO_UDP || 1340 sk->sk_protocol == IPPROTO_RAW)) { 1341 ipv6_local_rxpmtu(sk, fl6, mtu - headersize + 1342 sizeof(struct ipv6hdr)); 1343 goto emsgsize; 1344 } 1345 1346 if (ip6_sk_ignore_df(sk)) 1347 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN; 1348 else 1349 maxnonfragsize = mtu; 1350 1351 if (cork->length + length > maxnonfragsize - headersize) { 1352 emsgsize: 1353 ipv6_local_error(sk, EMSGSIZE, fl6, 1354 mtu - headersize + 1355 sizeof(struct ipv6hdr)); 1356 return -EMSGSIZE; 1357 } 1358 1359 /* CHECKSUM_PARTIAL only with no extension headers and when 1360 * we are not going to fragment 1361 */ 1362 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP && 1363 headersize == sizeof(struct ipv6hdr) && 1364 length <= mtu - headersize && 1365 !(flags & MSG_MORE) && 1366 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM)) 1367 csummode = CHECKSUM_PARTIAL; 1368 1369 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) { 1370 sock_tx_timestamp(sk, sockc->tsflags, &tx_flags); 1371 if (tx_flags & SKBTX_ANY_SW_TSTAMP && 1372 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) 1373 tskey = sk->sk_tskey++; 1374 } 1375 1376 /* 1377 * Let's try using as much space as possible. 1378 * Use MTU if total length of the message fits into the MTU. 1379 * Otherwise, we need to reserve fragment header and 1380 * fragment alignment (= 8-15 octects, in total). 1381 * 1382 * Note that we may need to "move" the data from the tail of 1383 * of the buffer to the new fragment when we split 1384 * the message. 1385 * 1386 * FIXME: It may be fragmented into multiple chunks 1387 * at once if non-fragmentable extension headers 1388 * are too large. 1389 * --yoshfuji 1390 */ 1391 1392 cork->length += length; 1393 if ((((length + fragheaderlen) > mtu) || 1394 (skb && skb_is_gso(skb))) && 1395 (sk->sk_protocol == IPPROTO_UDP) && 1396 (rt->dst.dev->features & NETIF_F_UFO) && !dst_xfrm(&rt->dst) && 1397 (sk->sk_type == SOCK_DGRAM) && !udp_get_no_check6_tx(sk)) { 1398 err = ip6_ufo_append_data(sk, queue, getfrag, from, length, 1399 hh_len, fragheaderlen, exthdrlen, 1400 transhdrlen, mtu, flags, fl6); 1401 if (err) 1402 goto error; 1403 return 0; 1404 } 1405 1406 if (!skb) 1407 goto alloc_new_skb; 1408 1409 while (length > 0) { 1410 /* Check if the remaining data fits into current packet. */ 1411 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len; 1412 if (copy < length) 1413 copy = maxfraglen - skb->len; 1414 1415 if (copy <= 0) { 1416 char *data; 1417 unsigned int datalen; 1418 unsigned int fraglen; 1419 unsigned int fraggap; 1420 unsigned int alloclen; 1421 alloc_new_skb: 1422 /* There's no room in the current skb */ 1423 if (skb) 1424 fraggap = skb->len - maxfraglen; 1425 else 1426 fraggap = 0; 1427 /* update mtu and maxfraglen if necessary */ 1428 if (!skb || !skb_prev) 1429 ip6_append_data_mtu(&mtu, &maxfraglen, 1430 fragheaderlen, skb, rt, 1431 orig_mtu); 1432 1433 skb_prev = skb; 1434 1435 /* 1436 * If remaining data exceeds the mtu, 1437 * we know we need more fragment(s). 1438 */ 1439 datalen = length + fraggap; 1440 1441 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen) 1442 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len; 1443 if ((flags & MSG_MORE) && 1444 !(rt->dst.dev->features&NETIF_F_SG)) 1445 alloclen = mtu; 1446 else 1447 alloclen = datalen + fragheaderlen; 1448 1449 alloclen += dst_exthdrlen; 1450 1451 if (datalen != length + fraggap) { 1452 /* 1453 * this is not the last fragment, the trailer 1454 * space is regarded as data space. 1455 */ 1456 datalen += rt->dst.trailer_len; 1457 } 1458 1459 alloclen += rt->dst.trailer_len; 1460 fraglen = datalen + fragheaderlen; 1461 1462 /* 1463 * We just reserve space for fragment header. 1464 * Note: this may be overallocation if the message 1465 * (without MSG_MORE) fits into the MTU. 1466 */ 1467 alloclen += sizeof(struct frag_hdr); 1468 1469 copy = datalen - transhdrlen - fraggap; 1470 if (copy < 0) { 1471 err = -EINVAL; 1472 goto error; 1473 } 1474 if (transhdrlen) { 1475 skb = sock_alloc_send_skb(sk, 1476 alloclen + hh_len, 1477 (flags & MSG_DONTWAIT), &err); 1478 } else { 1479 skb = NULL; 1480 if (atomic_read(&sk->sk_wmem_alloc) <= 1481 2 * sk->sk_sndbuf) 1482 skb = sock_wmalloc(sk, 1483 alloclen + hh_len, 1, 1484 sk->sk_allocation); 1485 if (unlikely(!skb)) 1486 err = -ENOBUFS; 1487 } 1488 if (!skb) 1489 goto error; 1490 /* 1491 * Fill in the control structures 1492 */ 1493 skb->protocol = htons(ETH_P_IPV6); 1494 skb->ip_summed = csummode; 1495 skb->csum = 0; 1496 /* reserve for fragmentation and ipsec header */ 1497 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) + 1498 dst_exthdrlen); 1499 1500 /* Only the initial fragment is time stamped */ 1501 skb_shinfo(skb)->tx_flags = tx_flags; 1502 tx_flags = 0; 1503 skb_shinfo(skb)->tskey = tskey; 1504 tskey = 0; 1505 1506 /* 1507 * Find where to start putting bytes 1508 */ 1509 data = skb_put(skb, fraglen); 1510 skb_set_network_header(skb, exthdrlen); 1511 data += fragheaderlen; 1512 skb->transport_header = (skb->network_header + 1513 fragheaderlen); 1514 if (fraggap) { 1515 skb->csum = skb_copy_and_csum_bits( 1516 skb_prev, maxfraglen, 1517 data + transhdrlen, fraggap, 0); 1518 skb_prev->csum = csum_sub(skb_prev->csum, 1519 skb->csum); 1520 data += fraggap; 1521 pskb_trim_unique(skb_prev, maxfraglen); 1522 } 1523 if (copy > 0 && 1524 getfrag(from, data + transhdrlen, offset, 1525 copy, fraggap, skb) < 0) { 1526 err = -EFAULT; 1527 kfree_skb(skb); 1528 goto error; 1529 } 1530 1531 offset += copy; 1532 length -= datalen - fraggap; 1533 transhdrlen = 0; 1534 exthdrlen = 0; 1535 dst_exthdrlen = 0; 1536 1537 if ((flags & MSG_CONFIRM) && !skb_prev) 1538 skb_set_dst_pending_confirm(skb, 1); 1539 1540 /* 1541 * Put the packet on the pending queue 1542 */ 1543 __skb_queue_tail(queue, skb); 1544 continue; 1545 } 1546 1547 if (copy > length) 1548 copy = length; 1549 1550 if (!(rt->dst.dev->features&NETIF_F_SG)) { 1551 unsigned int off; 1552 1553 off = skb->len; 1554 if (getfrag(from, skb_put(skb, copy), 1555 offset, copy, off, skb) < 0) { 1556 __skb_trim(skb, off); 1557 err = -EFAULT; 1558 goto error; 1559 } 1560 } else { 1561 int i = skb_shinfo(skb)->nr_frags; 1562 1563 err = -ENOMEM; 1564 if (!sk_page_frag_refill(sk, pfrag)) 1565 goto error; 1566 1567 if (!skb_can_coalesce(skb, i, pfrag->page, 1568 pfrag->offset)) { 1569 err = -EMSGSIZE; 1570 if (i == MAX_SKB_FRAGS) 1571 goto error; 1572 1573 __skb_fill_page_desc(skb, i, pfrag->page, 1574 pfrag->offset, 0); 1575 skb_shinfo(skb)->nr_frags = ++i; 1576 get_page(pfrag->page); 1577 } 1578 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1579 if (getfrag(from, 1580 page_address(pfrag->page) + pfrag->offset, 1581 offset, copy, skb->len, skb) < 0) 1582 goto error_efault; 1583 1584 pfrag->offset += copy; 1585 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1586 skb->len += copy; 1587 skb->data_len += copy; 1588 skb->truesize += copy; 1589 atomic_add(copy, &sk->sk_wmem_alloc); 1590 } 1591 offset += copy; 1592 length -= copy; 1593 } 1594 1595 return 0; 1596 1597 error_efault: 1598 err = -EFAULT; 1599 error: 1600 cork->length -= length; 1601 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); 1602 return err; 1603 } 1604 1605 int ip6_append_data(struct sock *sk, 1606 int getfrag(void *from, char *to, int offset, int len, 1607 int odd, struct sk_buff *skb), 1608 void *from, int length, int transhdrlen, 1609 struct ipcm6_cookie *ipc6, struct flowi6 *fl6, 1610 struct rt6_info *rt, unsigned int flags, 1611 const struct sockcm_cookie *sockc) 1612 { 1613 struct inet_sock *inet = inet_sk(sk); 1614 struct ipv6_pinfo *np = inet6_sk(sk); 1615 int exthdrlen; 1616 int err; 1617 1618 if (flags&MSG_PROBE) 1619 return 0; 1620 if (skb_queue_empty(&sk->sk_write_queue)) { 1621 /* 1622 * setup for corking 1623 */ 1624 err = ip6_setup_cork(sk, &inet->cork, &np->cork, 1625 ipc6, rt, fl6); 1626 if (err) 1627 return err; 1628 1629 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0); 1630 length += exthdrlen; 1631 transhdrlen += exthdrlen; 1632 } else { 1633 fl6 = &inet->cork.fl.u.ip6; 1634 transhdrlen = 0; 1635 } 1636 1637 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base, 1638 &np->cork, sk_page_frag(sk), getfrag, 1639 from, length, transhdrlen, flags, ipc6, sockc); 1640 } 1641 EXPORT_SYMBOL_GPL(ip6_append_data); 1642 1643 static void ip6_cork_release(struct inet_cork_full *cork, 1644 struct inet6_cork *v6_cork) 1645 { 1646 if (v6_cork->opt) { 1647 kfree(v6_cork->opt->dst0opt); 1648 kfree(v6_cork->opt->dst1opt); 1649 kfree(v6_cork->opt->hopopt); 1650 kfree(v6_cork->opt->srcrt); 1651 kfree(v6_cork->opt); 1652 v6_cork->opt = NULL; 1653 } 1654 1655 if (cork->base.dst) { 1656 dst_release(cork->base.dst); 1657 cork->base.dst = NULL; 1658 cork->base.flags &= ~IPCORK_ALLFRAG; 1659 } 1660 memset(&cork->fl, 0, sizeof(cork->fl)); 1661 } 1662 1663 struct sk_buff *__ip6_make_skb(struct sock *sk, 1664 struct sk_buff_head *queue, 1665 struct inet_cork_full *cork, 1666 struct inet6_cork *v6_cork) 1667 { 1668 struct sk_buff *skb, *tmp_skb; 1669 struct sk_buff **tail_skb; 1670 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf; 1671 struct ipv6_pinfo *np = inet6_sk(sk); 1672 struct net *net = sock_net(sk); 1673 struct ipv6hdr *hdr; 1674 struct ipv6_txoptions *opt = v6_cork->opt; 1675 struct rt6_info *rt = (struct rt6_info *)cork->base.dst; 1676 struct flowi6 *fl6 = &cork->fl.u.ip6; 1677 unsigned char proto = fl6->flowi6_proto; 1678 1679 skb = __skb_dequeue(queue); 1680 if (!skb) 1681 goto out; 1682 tail_skb = &(skb_shinfo(skb)->frag_list); 1683 1684 /* move skb->data to ip header from ext header */ 1685 if (skb->data < skb_network_header(skb)) 1686 __skb_pull(skb, skb_network_offset(skb)); 1687 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1688 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1689 *tail_skb = tmp_skb; 1690 tail_skb = &(tmp_skb->next); 1691 skb->len += tmp_skb->len; 1692 skb->data_len += tmp_skb->len; 1693 skb->truesize += tmp_skb->truesize; 1694 tmp_skb->destructor = NULL; 1695 tmp_skb->sk = NULL; 1696 } 1697 1698 /* Allow local fragmentation. */ 1699 skb->ignore_df = ip6_sk_ignore_df(sk); 1700 1701 *final_dst = fl6->daddr; 1702 __skb_pull(skb, skb_network_header_len(skb)); 1703 if (opt && opt->opt_flen) 1704 ipv6_push_frag_opts(skb, opt, &proto); 1705 if (opt && opt->opt_nflen) 1706 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr); 1707 1708 skb_push(skb, sizeof(struct ipv6hdr)); 1709 skb_reset_network_header(skb); 1710 hdr = ipv6_hdr(skb); 1711 1712 ip6_flow_hdr(hdr, v6_cork->tclass, 1713 ip6_make_flowlabel(net, skb, fl6->flowlabel, 1714 np->autoflowlabel, fl6)); 1715 hdr->hop_limit = v6_cork->hop_limit; 1716 hdr->nexthdr = proto; 1717 hdr->saddr = fl6->saddr; 1718 hdr->daddr = *final_dst; 1719 1720 skb->priority = sk->sk_priority; 1721 skb->mark = sk->sk_mark; 1722 1723 skb_dst_set(skb, dst_clone(&rt->dst)); 1724 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len); 1725 if (proto == IPPROTO_ICMPV6) { 1726 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb)); 1727 1728 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type); 1729 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS); 1730 } 1731 1732 ip6_cork_release(cork, v6_cork); 1733 out: 1734 return skb; 1735 } 1736 1737 int ip6_send_skb(struct sk_buff *skb) 1738 { 1739 struct net *net = sock_net(skb->sk); 1740 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb); 1741 int err; 1742 1743 err = ip6_local_out(net, skb->sk, skb); 1744 if (err) { 1745 if (err > 0) 1746 err = net_xmit_errno(err); 1747 if (err) 1748 IP6_INC_STATS(net, rt->rt6i_idev, 1749 IPSTATS_MIB_OUTDISCARDS); 1750 } 1751 1752 return err; 1753 } 1754 1755 int ip6_push_pending_frames(struct sock *sk) 1756 { 1757 struct sk_buff *skb; 1758 1759 skb = ip6_finish_skb(sk); 1760 if (!skb) 1761 return 0; 1762 1763 return ip6_send_skb(skb); 1764 } 1765 EXPORT_SYMBOL_GPL(ip6_push_pending_frames); 1766 1767 static void __ip6_flush_pending_frames(struct sock *sk, 1768 struct sk_buff_head *queue, 1769 struct inet_cork_full *cork, 1770 struct inet6_cork *v6_cork) 1771 { 1772 struct sk_buff *skb; 1773 1774 while ((skb = __skb_dequeue_tail(queue)) != NULL) { 1775 if (skb_dst(skb)) 1776 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)), 1777 IPSTATS_MIB_OUTDISCARDS); 1778 kfree_skb(skb); 1779 } 1780 1781 ip6_cork_release(cork, v6_cork); 1782 } 1783 1784 void ip6_flush_pending_frames(struct sock *sk) 1785 { 1786 __ip6_flush_pending_frames(sk, &sk->sk_write_queue, 1787 &inet_sk(sk)->cork, &inet6_sk(sk)->cork); 1788 } 1789 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames); 1790 1791 struct sk_buff *ip6_make_skb(struct sock *sk, 1792 int getfrag(void *from, char *to, int offset, 1793 int len, int odd, struct sk_buff *skb), 1794 void *from, int length, int transhdrlen, 1795 struct ipcm6_cookie *ipc6, struct flowi6 *fl6, 1796 struct rt6_info *rt, unsigned int flags, 1797 const struct sockcm_cookie *sockc) 1798 { 1799 struct inet_cork_full cork; 1800 struct inet6_cork v6_cork; 1801 struct sk_buff_head queue; 1802 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0); 1803 int err; 1804 1805 if (flags & MSG_PROBE) 1806 return NULL; 1807 1808 __skb_queue_head_init(&queue); 1809 1810 cork.base.flags = 0; 1811 cork.base.addr = 0; 1812 cork.base.opt = NULL; 1813 v6_cork.opt = NULL; 1814 err = ip6_setup_cork(sk, &cork, &v6_cork, ipc6, rt, fl6); 1815 if (err) 1816 return ERR_PTR(err); 1817 1818 if (ipc6->dontfrag < 0) 1819 ipc6->dontfrag = inet6_sk(sk)->dontfrag; 1820 1821 err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork, 1822 ¤t->task_frag, getfrag, from, 1823 length + exthdrlen, transhdrlen + exthdrlen, 1824 flags, ipc6, sockc); 1825 if (err) { 1826 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork); 1827 return ERR_PTR(err); 1828 } 1829 1830 return __ip6_make_skb(sk, &queue, &cork, &v6_cork); 1831 } 1832