1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * IPv6 output functions 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on linux/net/ipv4/ip_output.c 10 * 11 * Changes: 12 * A.N.Kuznetsov : airthmetics in fragmentation. 13 * extension headers are implemented. 14 * route changes now work. 15 * ip6_forward does not confuse sniffers. 16 * etc. 17 * 18 * H. von Brand : Added missing #include <linux/string.h> 19 * Imran Patel : frag id should be in NBO 20 * Kazunori MIYAZAWA @USAGI 21 * : add ip6_append_data and related functions 22 * for datagram xmit 23 */ 24 25 #include <linux/errno.h> 26 #include <linux/kernel.h> 27 #include <linux/string.h> 28 #include <linux/socket.h> 29 #include <linux/net.h> 30 #include <linux/netdevice.h> 31 #include <linux/if_arp.h> 32 #include <linux/in6.h> 33 #include <linux/tcp.h> 34 #include <linux/route.h> 35 #include <linux/module.h> 36 #include <linux/slab.h> 37 38 #include <linux/bpf-cgroup.h> 39 #include <linux/netfilter.h> 40 #include <linux/netfilter_ipv6.h> 41 42 #include <net/sock.h> 43 #include <net/snmp.h> 44 45 #include <net/ipv6.h> 46 #include <net/ndisc.h> 47 #include <net/protocol.h> 48 #include <net/ip6_route.h> 49 #include <net/addrconf.h> 50 #include <net/rawv6.h> 51 #include <net/icmp.h> 52 #include <net/xfrm.h> 53 #include <net/checksum.h> 54 #include <linux/mroute6.h> 55 #include <net/l3mdev.h> 56 #include <net/lwtunnel.h> 57 58 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) 59 { 60 struct dst_entry *dst = skb_dst(skb); 61 struct net_device *dev = dst->dev; 62 const struct in6_addr *nexthop; 63 struct neighbour *neigh; 64 int ret; 65 66 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) { 67 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb)); 68 69 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) && 70 ((mroute6_is_socket(net, skb) && 71 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) || 72 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr, 73 &ipv6_hdr(skb)->saddr))) { 74 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 75 76 /* Do not check for IFF_ALLMULTI; multicast routing 77 is not supported in any case. 78 */ 79 if (newskb) 80 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING, 81 net, sk, newskb, NULL, newskb->dev, 82 dev_loopback_xmit); 83 84 if (ipv6_hdr(skb)->hop_limit == 0) { 85 IP6_INC_STATS(net, idev, 86 IPSTATS_MIB_OUTDISCARDS); 87 kfree_skb(skb); 88 return 0; 89 } 90 } 91 92 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len); 93 94 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <= 95 IPV6_ADDR_SCOPE_NODELOCAL && 96 !(dev->flags & IFF_LOOPBACK)) { 97 kfree_skb(skb); 98 return 0; 99 } 100 } 101 102 if (lwtunnel_xmit_redirect(dst->lwtstate)) { 103 int res = lwtunnel_xmit(skb); 104 105 if (res < 0 || res == LWTUNNEL_XMIT_DONE) 106 return res; 107 } 108 109 rcu_read_lock_bh(); 110 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr); 111 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop); 112 if (unlikely(!neigh)) 113 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false); 114 if (!IS_ERR(neigh)) { 115 sock_confirm_neigh(skb, neigh); 116 ret = neigh_output(neigh, skb, false); 117 rcu_read_unlock_bh(); 118 return ret; 119 } 120 rcu_read_unlock_bh(); 121 122 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 123 kfree_skb(skb); 124 return -EINVAL; 125 } 126 127 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 128 { 129 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) 130 /* Policy lookup after SNAT yielded a new policy */ 131 if (skb_dst(skb)->xfrm) { 132 IPCB(skb)->flags |= IPSKB_REROUTED; 133 return dst_output(net, sk, skb); 134 } 135 #endif 136 137 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) || 138 dst_allfrag(skb_dst(skb)) || 139 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size)) 140 return ip6_fragment(net, sk, skb, ip6_finish_output2); 141 else 142 return ip6_finish_output2(net, sk, skb); 143 } 144 145 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) 146 { 147 int ret; 148 149 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); 150 switch (ret) { 151 case NET_XMIT_SUCCESS: 152 return __ip6_finish_output(net, sk, skb); 153 case NET_XMIT_CN: 154 return __ip6_finish_output(net, sk, skb) ? : ret; 155 default: 156 kfree_skb(skb); 157 return ret; 158 } 159 } 160 161 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb) 162 { 163 struct net_device *dev = skb_dst(skb)->dev; 164 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb)); 165 166 skb->protocol = htons(ETH_P_IPV6); 167 skb->dev = dev; 168 169 if (unlikely(idev->cnf.disable_ipv6)) { 170 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS); 171 kfree_skb(skb); 172 return 0; 173 } 174 175 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, 176 net, sk, skb, NULL, dev, 177 ip6_finish_output, 178 !(IP6CB(skb)->flags & IP6SKB_REROUTED)); 179 } 180 181 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np) 182 { 183 if (!np->autoflowlabel_set) 184 return ip6_default_np_autolabel(net); 185 else 186 return np->autoflowlabel; 187 } 188 189 /* 190 * xmit an sk_buff (used by TCP, SCTP and DCCP) 191 * Note : socket lock is not held for SYNACK packets, but might be modified 192 * by calls to skb_set_owner_w() and ipv6_local_error(), 193 * which are using proper atomic operations or spinlocks. 194 */ 195 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6, 196 __u32 mark, struct ipv6_txoptions *opt, int tclass) 197 { 198 struct net *net = sock_net(sk); 199 const struct ipv6_pinfo *np = inet6_sk(sk); 200 struct in6_addr *first_hop = &fl6->daddr; 201 struct dst_entry *dst = skb_dst(skb); 202 unsigned int head_room; 203 struct ipv6hdr *hdr; 204 u8 proto = fl6->flowi6_proto; 205 int seg_len = skb->len; 206 int hlimit = -1; 207 u32 mtu; 208 209 head_room = sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev); 210 if (opt) 211 head_room += opt->opt_nflen + opt->opt_flen; 212 213 if (unlikely(skb_headroom(skb) < head_room)) { 214 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room); 215 if (!skb2) { 216 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 217 IPSTATS_MIB_OUTDISCARDS); 218 kfree_skb(skb); 219 return -ENOBUFS; 220 } 221 if (skb->sk) 222 skb_set_owner_w(skb2, skb->sk); 223 consume_skb(skb); 224 skb = skb2; 225 } 226 227 if (opt) { 228 seg_len += opt->opt_nflen + opt->opt_flen; 229 230 if (opt->opt_flen) 231 ipv6_push_frag_opts(skb, opt, &proto); 232 233 if (opt->opt_nflen) 234 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop, 235 &fl6->saddr); 236 } 237 238 skb_push(skb, sizeof(struct ipv6hdr)); 239 skb_reset_network_header(skb); 240 hdr = ipv6_hdr(skb); 241 242 /* 243 * Fill in the IPv6 header 244 */ 245 if (np) 246 hlimit = np->hop_limit; 247 if (hlimit < 0) 248 hlimit = ip6_dst_hoplimit(dst); 249 250 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel, 251 ip6_autoflowlabel(net, np), fl6)); 252 253 hdr->payload_len = htons(seg_len); 254 hdr->nexthdr = proto; 255 hdr->hop_limit = hlimit; 256 257 hdr->saddr = fl6->saddr; 258 hdr->daddr = *first_hop; 259 260 skb->protocol = htons(ETH_P_IPV6); 261 skb->priority = sk->sk_priority; 262 skb->mark = mark; 263 264 mtu = dst_mtu(dst); 265 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) { 266 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)), 267 IPSTATS_MIB_OUT, skb->len); 268 269 /* if egress device is enslaved to an L3 master device pass the 270 * skb to its handler for processing 271 */ 272 skb = l3mdev_ip6_out((struct sock *)sk, skb); 273 if (unlikely(!skb)) 274 return 0; 275 276 /* hooks should never assume socket lock is held. 277 * we promote our socket to non const 278 */ 279 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, 280 net, (struct sock *)sk, skb, NULL, dst->dev, 281 dst_output); 282 } 283 284 skb->dev = dst->dev; 285 /* ipv6_local_error() does not require socket lock, 286 * we promote our socket to non const 287 */ 288 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu); 289 290 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS); 291 kfree_skb(skb); 292 return -EMSGSIZE; 293 } 294 EXPORT_SYMBOL(ip6_xmit); 295 296 static int ip6_call_ra_chain(struct sk_buff *skb, int sel) 297 { 298 struct ip6_ra_chain *ra; 299 struct sock *last = NULL; 300 301 read_lock(&ip6_ra_lock); 302 for (ra = ip6_ra_chain; ra; ra = ra->next) { 303 struct sock *sk = ra->sk; 304 if (sk && ra->sel == sel && 305 (!sk->sk_bound_dev_if || 306 sk->sk_bound_dev_if == skb->dev->ifindex)) { 307 struct ipv6_pinfo *np = inet6_sk(sk); 308 309 if (np && np->rtalert_isolate && 310 !net_eq(sock_net(sk), dev_net(skb->dev))) { 311 continue; 312 } 313 if (last) { 314 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 315 if (skb2) 316 rawv6_rcv(last, skb2); 317 } 318 last = sk; 319 } 320 } 321 322 if (last) { 323 rawv6_rcv(last, skb); 324 read_unlock(&ip6_ra_lock); 325 return 1; 326 } 327 read_unlock(&ip6_ra_lock); 328 return 0; 329 } 330 331 static int ip6_forward_proxy_check(struct sk_buff *skb) 332 { 333 struct ipv6hdr *hdr = ipv6_hdr(skb); 334 u8 nexthdr = hdr->nexthdr; 335 __be16 frag_off; 336 int offset; 337 338 if (ipv6_ext_hdr(nexthdr)) { 339 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off); 340 if (offset < 0) 341 return 0; 342 } else 343 offset = sizeof(struct ipv6hdr); 344 345 if (nexthdr == IPPROTO_ICMPV6) { 346 struct icmp6hdr *icmp6; 347 348 if (!pskb_may_pull(skb, (skb_network_header(skb) + 349 offset + 1 - skb->data))) 350 return 0; 351 352 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset); 353 354 switch (icmp6->icmp6_type) { 355 case NDISC_ROUTER_SOLICITATION: 356 case NDISC_ROUTER_ADVERTISEMENT: 357 case NDISC_NEIGHBOUR_SOLICITATION: 358 case NDISC_NEIGHBOUR_ADVERTISEMENT: 359 case NDISC_REDIRECT: 360 /* For reaction involving unicast neighbor discovery 361 * message destined to the proxied address, pass it to 362 * input function. 363 */ 364 return 1; 365 default: 366 break; 367 } 368 } 369 370 /* 371 * The proxying router can't forward traffic sent to a link-local 372 * address, so signal the sender and discard the packet. This 373 * behavior is clarified by the MIPv6 specification. 374 */ 375 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) { 376 dst_link_failure(skb); 377 return -1; 378 } 379 380 return 0; 381 } 382 383 static inline int ip6_forward_finish(struct net *net, struct sock *sk, 384 struct sk_buff *skb) 385 { 386 struct dst_entry *dst = skb_dst(skb); 387 388 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS); 389 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len); 390 391 #ifdef CONFIG_NET_SWITCHDEV 392 if (skb->offload_l3_fwd_mark) { 393 consume_skb(skb); 394 return 0; 395 } 396 #endif 397 398 skb->tstamp = 0; 399 return dst_output(net, sk, skb); 400 } 401 402 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu) 403 { 404 if (skb->len <= mtu) 405 return false; 406 407 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */ 408 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu) 409 return true; 410 411 if (skb->ignore_df) 412 return false; 413 414 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu)) 415 return false; 416 417 return true; 418 } 419 420 int ip6_forward(struct sk_buff *skb) 421 { 422 struct inet6_dev *idev = __in6_dev_get_safely(skb->dev); 423 struct dst_entry *dst = skb_dst(skb); 424 struct ipv6hdr *hdr = ipv6_hdr(skb); 425 struct inet6_skb_parm *opt = IP6CB(skb); 426 struct net *net = dev_net(dst->dev); 427 u32 mtu; 428 429 if (net->ipv6.devconf_all->forwarding == 0) 430 goto error; 431 432 if (skb->pkt_type != PACKET_HOST) 433 goto drop; 434 435 if (unlikely(skb->sk)) 436 goto drop; 437 438 if (skb_warn_if_lro(skb)) 439 goto drop; 440 441 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) { 442 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS); 443 goto drop; 444 } 445 446 skb_forward_csum(skb); 447 448 /* 449 * We DO NOT make any processing on 450 * RA packets, pushing them to user level AS IS 451 * without ane WARRANTY that application will be able 452 * to interpret them. The reason is that we 453 * cannot make anything clever here. 454 * 455 * We are not end-node, so that if packet contains 456 * AH/ESP, we cannot make anything. 457 * Defragmentation also would be mistake, RA packets 458 * cannot be fragmented, because there is no warranty 459 * that different fragments will go along one path. --ANK 460 */ 461 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) { 462 if (ip6_call_ra_chain(skb, ntohs(opt->ra))) 463 return 0; 464 } 465 466 /* 467 * check and decrement ttl 468 */ 469 if (hdr->hop_limit <= 1) { 470 /* Force OUTPUT device used as source address */ 471 skb->dev = dst->dev; 472 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0); 473 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS); 474 475 kfree_skb(skb); 476 return -ETIMEDOUT; 477 } 478 479 /* XXX: idev->cnf.proxy_ndp? */ 480 if (net->ipv6.devconf_all->proxy_ndp && 481 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) { 482 int proxied = ip6_forward_proxy_check(skb); 483 if (proxied > 0) 484 return ip6_input(skb); 485 else if (proxied < 0) { 486 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS); 487 goto drop; 488 } 489 } 490 491 if (!xfrm6_route_forward(skb)) { 492 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS); 493 goto drop; 494 } 495 dst = skb_dst(skb); 496 497 /* IPv6 specs say nothing about it, but it is clear that we cannot 498 send redirects to source routed frames. 499 We don't send redirects to frames decapsulated from IPsec. 500 */ 501 if (IP6CB(skb)->iif == dst->dev->ifindex && 502 opt->srcrt == 0 && !skb_sec_path(skb)) { 503 struct in6_addr *target = NULL; 504 struct inet_peer *peer; 505 struct rt6_info *rt; 506 507 /* 508 * incoming and outgoing devices are the same 509 * send a redirect. 510 */ 511 512 rt = (struct rt6_info *) dst; 513 if (rt->rt6i_flags & RTF_GATEWAY) 514 target = &rt->rt6i_gateway; 515 else 516 target = &hdr->daddr; 517 518 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1); 519 520 /* Limit redirects both by destination (here) 521 and by source (inside ndisc_send_redirect) 522 */ 523 if (inet_peer_xrlim_allow(peer, 1*HZ)) 524 ndisc_send_redirect(skb, target); 525 if (peer) 526 inet_putpeer(peer); 527 } else { 528 int addrtype = ipv6_addr_type(&hdr->saddr); 529 530 /* This check is security critical. */ 531 if (addrtype == IPV6_ADDR_ANY || 532 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK)) 533 goto error; 534 if (addrtype & IPV6_ADDR_LINKLOCAL) { 535 icmpv6_send(skb, ICMPV6_DEST_UNREACH, 536 ICMPV6_NOT_NEIGHBOUR, 0); 537 goto error; 538 } 539 } 540 541 mtu = ip6_dst_mtu_forward(dst); 542 if (mtu < IPV6_MIN_MTU) 543 mtu = IPV6_MIN_MTU; 544 545 if (ip6_pkt_too_big(skb, mtu)) { 546 /* Again, force OUTPUT device used as source address */ 547 skb->dev = dst->dev; 548 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); 549 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS); 550 __IP6_INC_STATS(net, ip6_dst_idev(dst), 551 IPSTATS_MIB_FRAGFAILS); 552 kfree_skb(skb); 553 return -EMSGSIZE; 554 } 555 556 if (skb_cow(skb, dst->dev->hard_header_len)) { 557 __IP6_INC_STATS(net, ip6_dst_idev(dst), 558 IPSTATS_MIB_OUTDISCARDS); 559 goto drop; 560 } 561 562 hdr = ipv6_hdr(skb); 563 564 /* Mangling hops number delayed to point after skb COW */ 565 566 hdr->hop_limit--; 567 568 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, 569 net, NULL, skb, skb->dev, dst->dev, 570 ip6_forward_finish); 571 572 error: 573 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS); 574 drop: 575 kfree_skb(skb); 576 return -EINVAL; 577 } 578 579 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from) 580 { 581 to->pkt_type = from->pkt_type; 582 to->priority = from->priority; 583 to->protocol = from->protocol; 584 skb_dst_drop(to); 585 skb_dst_set(to, dst_clone(skb_dst(from))); 586 to->dev = from->dev; 587 to->mark = from->mark; 588 589 skb_copy_hash(to, from); 590 591 #ifdef CONFIG_NET_SCHED 592 to->tc_index = from->tc_index; 593 #endif 594 nf_copy(to, from); 595 skb_ext_copy(to, from); 596 skb_copy_secmark(to, from); 597 } 598 599 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr, 600 u8 nexthdr, __be32 frag_id, 601 struct ip6_fraglist_iter *iter) 602 { 603 unsigned int first_len; 604 struct frag_hdr *fh; 605 606 /* BUILD HEADER */ 607 *prevhdr = NEXTHDR_FRAGMENT; 608 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC); 609 if (!iter->tmp_hdr) 610 return -ENOMEM; 611 612 iter->frag = skb_shinfo(skb)->frag_list; 613 skb_frag_list_init(skb); 614 615 iter->offset = 0; 616 iter->hlen = hlen; 617 iter->frag_id = frag_id; 618 iter->nexthdr = nexthdr; 619 620 __skb_pull(skb, hlen); 621 fh = __skb_push(skb, sizeof(struct frag_hdr)); 622 __skb_push(skb, hlen); 623 skb_reset_network_header(skb); 624 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen); 625 626 fh->nexthdr = nexthdr; 627 fh->reserved = 0; 628 fh->frag_off = htons(IP6_MF); 629 fh->identification = frag_id; 630 631 first_len = skb_pagelen(skb); 632 skb->data_len = first_len - skb_headlen(skb); 633 skb->len = first_len; 634 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr)); 635 636 return 0; 637 } 638 EXPORT_SYMBOL(ip6_fraglist_init); 639 640 void ip6_fraglist_prepare(struct sk_buff *skb, 641 struct ip6_fraglist_iter *iter) 642 { 643 struct sk_buff *frag = iter->frag; 644 unsigned int hlen = iter->hlen; 645 struct frag_hdr *fh; 646 647 frag->ip_summed = CHECKSUM_NONE; 648 skb_reset_transport_header(frag); 649 fh = __skb_push(frag, sizeof(struct frag_hdr)); 650 __skb_push(frag, hlen); 651 skb_reset_network_header(frag); 652 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen); 653 iter->offset += skb->len - hlen - sizeof(struct frag_hdr); 654 fh->nexthdr = iter->nexthdr; 655 fh->reserved = 0; 656 fh->frag_off = htons(iter->offset); 657 if (frag->next) 658 fh->frag_off |= htons(IP6_MF); 659 fh->identification = iter->frag_id; 660 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr)); 661 ip6_copy_metadata(frag, skb); 662 } 663 EXPORT_SYMBOL(ip6_fraglist_prepare); 664 665 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu, 666 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr, 667 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state) 668 { 669 state->prevhdr = prevhdr; 670 state->nexthdr = nexthdr; 671 state->frag_id = frag_id; 672 673 state->hlen = hlen; 674 state->mtu = mtu; 675 676 state->left = skb->len - hlen; /* Space per frame */ 677 state->ptr = hlen; /* Where to start from */ 678 679 state->hroom = hdr_room; 680 state->troom = needed_tailroom; 681 682 state->offset = 0; 683 } 684 EXPORT_SYMBOL(ip6_frag_init); 685 686 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state) 687 { 688 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset; 689 struct sk_buff *frag; 690 struct frag_hdr *fh; 691 unsigned int len; 692 693 len = state->left; 694 /* IF: it doesn't fit, use 'mtu' - the data space left */ 695 if (len > state->mtu) 696 len = state->mtu; 697 /* IF: we are not sending up to and including the packet end 698 then align the next start on an eight byte boundary */ 699 if (len < state->left) 700 len &= ~7; 701 702 /* Allocate buffer */ 703 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) + 704 state->hroom + state->troom, GFP_ATOMIC); 705 if (!frag) 706 return ERR_PTR(-ENOMEM); 707 708 /* 709 * Set up data on packet 710 */ 711 712 ip6_copy_metadata(frag, skb); 713 skb_reserve(frag, state->hroom); 714 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr)); 715 skb_reset_network_header(frag); 716 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen); 717 frag->transport_header = (frag->network_header + state->hlen + 718 sizeof(struct frag_hdr)); 719 720 /* 721 * Charge the memory for the fragment to any owner 722 * it might possess 723 */ 724 if (skb->sk) 725 skb_set_owner_w(frag, skb->sk); 726 727 /* 728 * Copy the packet header into the new buffer. 729 */ 730 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen); 731 732 fragnexthdr_offset = skb_network_header(frag); 733 fragnexthdr_offset += prevhdr - skb_network_header(skb); 734 *fragnexthdr_offset = NEXTHDR_FRAGMENT; 735 736 /* 737 * Build fragment header. 738 */ 739 fh->nexthdr = state->nexthdr; 740 fh->reserved = 0; 741 fh->identification = state->frag_id; 742 743 /* 744 * Copy a block of the IP datagram. 745 */ 746 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag), 747 len)); 748 state->left -= len; 749 750 fh->frag_off = htons(state->offset); 751 if (state->left > 0) 752 fh->frag_off |= htons(IP6_MF); 753 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr)); 754 755 state->ptr += len; 756 state->offset += len; 757 758 return frag; 759 } 760 EXPORT_SYMBOL(ip6_frag_next); 761 762 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, 763 int (*output)(struct net *, struct sock *, struct sk_buff *)) 764 { 765 struct sk_buff *frag; 766 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb); 767 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ? 768 inet6_sk(skb->sk) : NULL; 769 struct ip6_frag_state state; 770 unsigned int mtu, hlen, nexthdr_offset; 771 int hroom, err = 0; 772 __be32 frag_id; 773 u8 *prevhdr, nexthdr = 0; 774 775 err = ip6_find_1stfragopt(skb, &prevhdr); 776 if (err < 0) 777 goto fail; 778 hlen = err; 779 nexthdr = *prevhdr; 780 nexthdr_offset = prevhdr - skb_network_header(skb); 781 782 mtu = ip6_skb_dst_mtu(skb); 783 784 /* We must not fragment if the socket is set to force MTU discovery 785 * or if the skb it not generated by a local socket. 786 */ 787 if (unlikely(!skb->ignore_df && skb->len > mtu)) 788 goto fail_toobig; 789 790 if (IP6CB(skb)->frag_max_size) { 791 if (IP6CB(skb)->frag_max_size > mtu) 792 goto fail_toobig; 793 794 /* don't send fragments larger than what we received */ 795 mtu = IP6CB(skb)->frag_max_size; 796 if (mtu < IPV6_MIN_MTU) 797 mtu = IPV6_MIN_MTU; 798 } 799 800 if (np && np->frag_size < mtu) { 801 if (np->frag_size) 802 mtu = np->frag_size; 803 } 804 if (mtu < hlen + sizeof(struct frag_hdr) + 8) 805 goto fail_toobig; 806 mtu -= hlen + sizeof(struct frag_hdr); 807 808 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr, 809 &ipv6_hdr(skb)->saddr); 810 811 if (skb->ip_summed == CHECKSUM_PARTIAL && 812 (err = skb_checksum_help(skb))) 813 goto fail; 814 815 prevhdr = skb_network_header(skb) + nexthdr_offset; 816 hroom = LL_RESERVED_SPACE(rt->dst.dev); 817 if (skb_has_frag_list(skb)) { 818 unsigned int first_len = skb_pagelen(skb); 819 struct ip6_fraglist_iter iter; 820 struct sk_buff *frag2; 821 822 if (first_len - hlen > mtu || 823 ((first_len - hlen) & 7) || 824 skb_cloned(skb) || 825 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr))) 826 goto slow_path; 827 828 skb_walk_frags(skb, frag) { 829 /* Correct geometry. */ 830 if (frag->len > mtu || 831 ((frag->len & 7) && frag->next) || 832 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr))) 833 goto slow_path_clean; 834 835 /* Partially cloned skb? */ 836 if (skb_shared(frag)) 837 goto slow_path_clean; 838 839 BUG_ON(frag->sk); 840 if (skb->sk) { 841 frag->sk = skb->sk; 842 frag->destructor = sock_wfree; 843 } 844 skb->truesize -= frag->truesize; 845 } 846 847 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id, 848 &iter); 849 if (err < 0) 850 goto fail; 851 852 for (;;) { 853 /* Prepare header of the next frame, 854 * before previous one went down. */ 855 if (iter.frag) 856 ip6_fraglist_prepare(skb, &iter); 857 858 err = output(net, sk, skb); 859 if (!err) 860 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst), 861 IPSTATS_MIB_FRAGCREATES); 862 863 if (err || !iter.frag) 864 break; 865 866 skb = ip6_fraglist_next(&iter); 867 } 868 869 kfree(iter.tmp_hdr); 870 871 if (err == 0) { 872 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst), 873 IPSTATS_MIB_FRAGOKS); 874 return 0; 875 } 876 877 kfree_skb_list(iter.frag); 878 879 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst), 880 IPSTATS_MIB_FRAGFAILS); 881 return err; 882 883 slow_path_clean: 884 skb_walk_frags(skb, frag2) { 885 if (frag2 == frag) 886 break; 887 frag2->sk = NULL; 888 frag2->destructor = NULL; 889 skb->truesize += frag2->truesize; 890 } 891 } 892 893 slow_path: 894 /* 895 * Fragment the datagram. 896 */ 897 898 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom, 899 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id, 900 &state); 901 902 /* 903 * Keep copying data until we run out. 904 */ 905 906 while (state.left > 0) { 907 frag = ip6_frag_next(skb, &state); 908 if (IS_ERR(frag)) { 909 err = PTR_ERR(frag); 910 goto fail; 911 } 912 913 /* 914 * Put this fragment into the sending queue. 915 */ 916 err = output(net, sk, frag); 917 if (err) 918 goto fail; 919 920 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 921 IPSTATS_MIB_FRAGCREATES); 922 } 923 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 924 IPSTATS_MIB_FRAGOKS); 925 consume_skb(skb); 926 return err; 927 928 fail_toobig: 929 if (skb->sk && dst_allfrag(skb_dst(skb))) 930 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK); 931 932 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); 933 err = -EMSGSIZE; 934 935 fail: 936 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), 937 IPSTATS_MIB_FRAGFAILS); 938 kfree_skb(skb); 939 return err; 940 } 941 942 static inline int ip6_rt_check(const struct rt6key *rt_key, 943 const struct in6_addr *fl_addr, 944 const struct in6_addr *addr_cache) 945 { 946 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) && 947 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache)); 948 } 949 950 static struct dst_entry *ip6_sk_dst_check(struct sock *sk, 951 struct dst_entry *dst, 952 const struct flowi6 *fl6) 953 { 954 struct ipv6_pinfo *np = inet6_sk(sk); 955 struct rt6_info *rt; 956 957 if (!dst) 958 goto out; 959 960 if (dst->ops->family != AF_INET6) { 961 dst_release(dst); 962 return NULL; 963 } 964 965 rt = (struct rt6_info *)dst; 966 /* Yes, checking route validity in not connected 967 * case is not very simple. Take into account, 968 * that we do not support routing by source, TOS, 969 * and MSG_DONTROUTE --ANK (980726) 970 * 971 * 1. ip6_rt_check(): If route was host route, 972 * check that cached destination is current. 973 * If it is network route, we still may 974 * check its validity using saved pointer 975 * to the last used address: daddr_cache. 976 * We do not want to save whole address now, 977 * (because main consumer of this service 978 * is tcp, which has not this problem), 979 * so that the last trick works only on connected 980 * sockets. 981 * 2. oif also should be the same. 982 */ 983 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) || 984 #ifdef CONFIG_IPV6_SUBTREES 985 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) || 986 #endif 987 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) && 988 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) { 989 dst_release(dst); 990 dst = NULL; 991 } 992 993 out: 994 return dst; 995 } 996 997 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk, 998 struct dst_entry **dst, struct flowi6 *fl6) 999 { 1000 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD 1001 struct neighbour *n; 1002 struct rt6_info *rt; 1003 #endif 1004 int err; 1005 int flags = 0; 1006 1007 /* The correct way to handle this would be to do 1008 * ip6_route_get_saddr, and then ip6_route_output; however, 1009 * the route-specific preferred source forces the 1010 * ip6_route_output call _before_ ip6_route_get_saddr. 1011 * 1012 * In source specific routing (no src=any default route), 1013 * ip6_route_output will fail given src=any saddr, though, so 1014 * that's why we try it again later. 1015 */ 1016 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) { 1017 struct fib6_info *from; 1018 struct rt6_info *rt; 1019 bool had_dst = *dst != NULL; 1020 1021 if (!had_dst) 1022 *dst = ip6_route_output(net, sk, fl6); 1023 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst; 1024 1025 rcu_read_lock(); 1026 from = rt ? rcu_dereference(rt->from) : NULL; 1027 err = ip6_route_get_saddr(net, from, &fl6->daddr, 1028 sk ? inet6_sk(sk)->srcprefs : 0, 1029 &fl6->saddr); 1030 rcu_read_unlock(); 1031 1032 if (err) 1033 goto out_err_release; 1034 1035 /* If we had an erroneous initial result, pretend it 1036 * never existed and let the SA-enabled version take 1037 * over. 1038 */ 1039 if (!had_dst && (*dst)->error) { 1040 dst_release(*dst); 1041 *dst = NULL; 1042 } 1043 1044 if (fl6->flowi6_oif) 1045 flags |= RT6_LOOKUP_F_IFACE; 1046 } 1047 1048 if (!*dst) 1049 *dst = ip6_route_output_flags(net, sk, fl6, flags); 1050 1051 err = (*dst)->error; 1052 if (err) 1053 goto out_err_release; 1054 1055 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD 1056 /* 1057 * Here if the dst entry we've looked up 1058 * has a neighbour entry that is in the INCOMPLETE 1059 * state and the src address from the flow is 1060 * marked as OPTIMISTIC, we release the found 1061 * dst entry and replace it instead with the 1062 * dst entry of the nexthop router 1063 */ 1064 rt = (struct rt6_info *) *dst; 1065 rcu_read_lock_bh(); 1066 n = __ipv6_neigh_lookup_noref(rt->dst.dev, 1067 rt6_nexthop(rt, &fl6->daddr)); 1068 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0; 1069 rcu_read_unlock_bh(); 1070 1071 if (err) { 1072 struct inet6_ifaddr *ifp; 1073 struct flowi6 fl_gw6; 1074 int redirect; 1075 1076 ifp = ipv6_get_ifaddr(net, &fl6->saddr, 1077 (*dst)->dev, 1); 1078 1079 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC); 1080 if (ifp) 1081 in6_ifa_put(ifp); 1082 1083 if (redirect) { 1084 /* 1085 * We need to get the dst entry for the 1086 * default router instead 1087 */ 1088 dst_release(*dst); 1089 memcpy(&fl_gw6, fl6, sizeof(struct flowi6)); 1090 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr)); 1091 *dst = ip6_route_output(net, sk, &fl_gw6); 1092 err = (*dst)->error; 1093 if (err) 1094 goto out_err_release; 1095 } 1096 } 1097 #endif 1098 if (ipv6_addr_v4mapped(&fl6->saddr) && 1099 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) { 1100 err = -EAFNOSUPPORT; 1101 goto out_err_release; 1102 } 1103 1104 return 0; 1105 1106 out_err_release: 1107 dst_release(*dst); 1108 *dst = NULL; 1109 1110 if (err == -ENETUNREACH) 1111 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES); 1112 return err; 1113 } 1114 1115 /** 1116 * ip6_dst_lookup - perform route lookup on flow 1117 * @sk: socket which provides route info 1118 * @dst: pointer to dst_entry * for result 1119 * @fl6: flow to lookup 1120 * 1121 * This function performs a route lookup on the given flow. 1122 * 1123 * It returns zero on success, or a standard errno code on error. 1124 */ 1125 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst, 1126 struct flowi6 *fl6) 1127 { 1128 *dst = NULL; 1129 return ip6_dst_lookup_tail(net, sk, dst, fl6); 1130 } 1131 EXPORT_SYMBOL_GPL(ip6_dst_lookup); 1132 1133 /** 1134 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec 1135 * @sk: socket which provides route info 1136 * @fl6: flow to lookup 1137 * @final_dst: final destination address for ipsec lookup 1138 * 1139 * This function performs a route lookup on the given flow. 1140 * 1141 * It returns a valid dst pointer on success, or a pointer encoded 1142 * error code. 1143 */ 1144 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6, 1145 const struct in6_addr *final_dst) 1146 { 1147 struct dst_entry *dst = NULL; 1148 int err; 1149 1150 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6); 1151 if (err) 1152 return ERR_PTR(err); 1153 if (final_dst) 1154 fl6->daddr = *final_dst; 1155 1156 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0); 1157 } 1158 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow); 1159 1160 /** 1161 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow 1162 * @sk: socket which provides the dst cache and route info 1163 * @fl6: flow to lookup 1164 * @final_dst: final destination address for ipsec lookup 1165 * @connected: whether @sk is connected or not 1166 * 1167 * This function performs a route lookup on the given flow with the 1168 * possibility of using the cached route in the socket if it is valid. 1169 * It will take the socket dst lock when operating on the dst cache. 1170 * As a result, this function can only be used in process context. 1171 * 1172 * In addition, for a connected socket, cache the dst in the socket 1173 * if the current cache is not valid. 1174 * 1175 * It returns a valid dst pointer on success, or a pointer encoded 1176 * error code. 1177 */ 1178 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6, 1179 const struct in6_addr *final_dst, 1180 bool connected) 1181 { 1182 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie); 1183 1184 dst = ip6_sk_dst_check(sk, dst, fl6); 1185 if (dst) 1186 return dst; 1187 1188 dst = ip6_dst_lookup_flow(sk, fl6, final_dst); 1189 if (connected && !IS_ERR(dst)) 1190 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6); 1191 1192 return dst; 1193 } 1194 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow); 1195 1196 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src, 1197 gfp_t gfp) 1198 { 1199 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL; 1200 } 1201 1202 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src, 1203 gfp_t gfp) 1204 { 1205 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL; 1206 } 1207 1208 static void ip6_append_data_mtu(unsigned int *mtu, 1209 int *maxfraglen, 1210 unsigned int fragheaderlen, 1211 struct sk_buff *skb, 1212 struct rt6_info *rt, 1213 unsigned int orig_mtu) 1214 { 1215 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) { 1216 if (!skb) { 1217 /* first fragment, reserve header_len */ 1218 *mtu = orig_mtu - rt->dst.header_len; 1219 1220 } else { 1221 /* 1222 * this fragment is not first, the headers 1223 * space is regarded as data space. 1224 */ 1225 *mtu = orig_mtu; 1226 } 1227 *maxfraglen = ((*mtu - fragheaderlen) & ~7) 1228 + fragheaderlen - sizeof(struct frag_hdr); 1229 } 1230 } 1231 1232 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork, 1233 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6, 1234 struct rt6_info *rt, struct flowi6 *fl6) 1235 { 1236 struct ipv6_pinfo *np = inet6_sk(sk); 1237 unsigned int mtu; 1238 struct ipv6_txoptions *opt = ipc6->opt; 1239 1240 /* 1241 * setup for corking 1242 */ 1243 if (opt) { 1244 if (WARN_ON(v6_cork->opt)) 1245 return -EINVAL; 1246 1247 v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation); 1248 if (unlikely(!v6_cork->opt)) 1249 return -ENOBUFS; 1250 1251 v6_cork->opt->tot_len = sizeof(*opt); 1252 v6_cork->opt->opt_flen = opt->opt_flen; 1253 v6_cork->opt->opt_nflen = opt->opt_nflen; 1254 1255 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt, 1256 sk->sk_allocation); 1257 if (opt->dst0opt && !v6_cork->opt->dst0opt) 1258 return -ENOBUFS; 1259 1260 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt, 1261 sk->sk_allocation); 1262 if (opt->dst1opt && !v6_cork->opt->dst1opt) 1263 return -ENOBUFS; 1264 1265 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt, 1266 sk->sk_allocation); 1267 if (opt->hopopt && !v6_cork->opt->hopopt) 1268 return -ENOBUFS; 1269 1270 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt, 1271 sk->sk_allocation); 1272 if (opt->srcrt && !v6_cork->opt->srcrt) 1273 return -ENOBUFS; 1274 1275 /* need source address above miyazawa*/ 1276 } 1277 dst_hold(&rt->dst); 1278 cork->base.dst = &rt->dst; 1279 cork->fl.u.ip6 = *fl6; 1280 v6_cork->hop_limit = ipc6->hlimit; 1281 v6_cork->tclass = ipc6->tclass; 1282 if (rt->dst.flags & DST_XFRM_TUNNEL) 1283 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ? 1284 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst); 1285 else 1286 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ? 1287 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst)); 1288 if (np->frag_size < mtu) { 1289 if (np->frag_size) 1290 mtu = np->frag_size; 1291 } 1292 if (mtu < IPV6_MIN_MTU) 1293 return -EINVAL; 1294 cork->base.fragsize = mtu; 1295 cork->base.gso_size = ipc6->gso_size; 1296 cork->base.tx_flags = 0; 1297 sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags); 1298 1299 if (dst_allfrag(xfrm_dst_path(&rt->dst))) 1300 cork->base.flags |= IPCORK_ALLFRAG; 1301 cork->base.length = 0; 1302 1303 cork->base.transmit_time = ipc6->sockc.transmit_time; 1304 1305 return 0; 1306 } 1307 1308 static int __ip6_append_data(struct sock *sk, 1309 struct flowi6 *fl6, 1310 struct sk_buff_head *queue, 1311 struct inet_cork *cork, 1312 struct inet6_cork *v6_cork, 1313 struct page_frag *pfrag, 1314 int getfrag(void *from, char *to, int offset, 1315 int len, int odd, struct sk_buff *skb), 1316 void *from, int length, int transhdrlen, 1317 unsigned int flags, struct ipcm6_cookie *ipc6) 1318 { 1319 struct sk_buff *skb, *skb_prev = NULL; 1320 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu; 1321 struct ubuf_info *uarg = NULL; 1322 int exthdrlen = 0; 1323 int dst_exthdrlen = 0; 1324 int hh_len; 1325 int copy; 1326 int err; 1327 int offset = 0; 1328 u32 tskey = 0; 1329 struct rt6_info *rt = (struct rt6_info *)cork->dst; 1330 struct ipv6_txoptions *opt = v6_cork->opt; 1331 int csummode = CHECKSUM_NONE; 1332 unsigned int maxnonfragsize, headersize; 1333 unsigned int wmem_alloc_delta = 0; 1334 bool paged, extra_uref = false; 1335 1336 skb = skb_peek_tail(queue); 1337 if (!skb) { 1338 exthdrlen = opt ? opt->opt_flen : 0; 1339 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len; 1340 } 1341 1342 paged = !!cork->gso_size; 1343 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize; 1344 orig_mtu = mtu; 1345 1346 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && 1347 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) 1348 tskey = sk->sk_tskey++; 1349 1350 hh_len = LL_RESERVED_SPACE(rt->dst.dev); 1351 1352 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len + 1353 (opt ? opt->opt_nflen : 0); 1354 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - 1355 sizeof(struct frag_hdr); 1356 1357 headersize = sizeof(struct ipv6hdr) + 1358 (opt ? opt->opt_flen + opt->opt_nflen : 0) + 1359 (dst_allfrag(&rt->dst) ? 1360 sizeof(struct frag_hdr) : 0) + 1361 rt->rt6i_nfheader_len; 1362 1363 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit 1364 * the first fragment 1365 */ 1366 if (headersize + transhdrlen > mtu) 1367 goto emsgsize; 1368 1369 if (cork->length + length > mtu - headersize && ipc6->dontfrag && 1370 (sk->sk_protocol == IPPROTO_UDP || 1371 sk->sk_protocol == IPPROTO_RAW)) { 1372 ipv6_local_rxpmtu(sk, fl6, mtu - headersize + 1373 sizeof(struct ipv6hdr)); 1374 goto emsgsize; 1375 } 1376 1377 if (ip6_sk_ignore_df(sk)) 1378 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN; 1379 else 1380 maxnonfragsize = mtu; 1381 1382 if (cork->length + length > maxnonfragsize - headersize) { 1383 emsgsize: 1384 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0); 1385 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu); 1386 return -EMSGSIZE; 1387 } 1388 1389 /* CHECKSUM_PARTIAL only with no extension headers and when 1390 * we are not going to fragment 1391 */ 1392 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP && 1393 headersize == sizeof(struct ipv6hdr) && 1394 length <= mtu - headersize && 1395 (!(flags & MSG_MORE) || cork->gso_size) && 1396 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM)) 1397 csummode = CHECKSUM_PARTIAL; 1398 1399 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) { 1400 uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb)); 1401 if (!uarg) 1402 return -ENOBUFS; 1403 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */ 1404 if (rt->dst.dev->features & NETIF_F_SG && 1405 csummode == CHECKSUM_PARTIAL) { 1406 paged = true; 1407 } else { 1408 uarg->zerocopy = 0; 1409 skb_zcopy_set(skb, uarg, &extra_uref); 1410 } 1411 } 1412 1413 /* 1414 * Let's try using as much space as possible. 1415 * Use MTU if total length of the message fits into the MTU. 1416 * Otherwise, we need to reserve fragment header and 1417 * fragment alignment (= 8-15 octects, in total). 1418 * 1419 * Note that we may need to "move" the data from the tail of 1420 * of the buffer to the new fragment when we split 1421 * the message. 1422 * 1423 * FIXME: It may be fragmented into multiple chunks 1424 * at once if non-fragmentable extension headers 1425 * are too large. 1426 * --yoshfuji 1427 */ 1428 1429 cork->length += length; 1430 if (!skb) 1431 goto alloc_new_skb; 1432 1433 while (length > 0) { 1434 /* Check if the remaining data fits into current packet. */ 1435 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len; 1436 if (copy < length) 1437 copy = maxfraglen - skb->len; 1438 1439 if (copy <= 0) { 1440 char *data; 1441 unsigned int datalen; 1442 unsigned int fraglen; 1443 unsigned int fraggap; 1444 unsigned int alloclen; 1445 unsigned int pagedlen; 1446 alloc_new_skb: 1447 /* There's no room in the current skb */ 1448 if (skb) 1449 fraggap = skb->len - maxfraglen; 1450 else 1451 fraggap = 0; 1452 /* update mtu and maxfraglen if necessary */ 1453 if (!skb || !skb_prev) 1454 ip6_append_data_mtu(&mtu, &maxfraglen, 1455 fragheaderlen, skb, rt, 1456 orig_mtu); 1457 1458 skb_prev = skb; 1459 1460 /* 1461 * If remaining data exceeds the mtu, 1462 * we know we need more fragment(s). 1463 */ 1464 datalen = length + fraggap; 1465 1466 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen) 1467 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len; 1468 fraglen = datalen + fragheaderlen; 1469 pagedlen = 0; 1470 1471 if ((flags & MSG_MORE) && 1472 !(rt->dst.dev->features&NETIF_F_SG)) 1473 alloclen = mtu; 1474 else if (!paged) 1475 alloclen = fraglen; 1476 else { 1477 alloclen = min_t(int, fraglen, MAX_HEADER); 1478 pagedlen = fraglen - alloclen; 1479 } 1480 1481 alloclen += dst_exthdrlen; 1482 1483 if (datalen != length + fraggap) { 1484 /* 1485 * this is not the last fragment, the trailer 1486 * space is regarded as data space. 1487 */ 1488 datalen += rt->dst.trailer_len; 1489 } 1490 1491 alloclen += rt->dst.trailer_len; 1492 fraglen = datalen + fragheaderlen; 1493 1494 /* 1495 * We just reserve space for fragment header. 1496 * Note: this may be overallocation if the message 1497 * (without MSG_MORE) fits into the MTU. 1498 */ 1499 alloclen += sizeof(struct frag_hdr); 1500 1501 copy = datalen - transhdrlen - fraggap - pagedlen; 1502 if (copy < 0) { 1503 err = -EINVAL; 1504 goto error; 1505 } 1506 if (transhdrlen) { 1507 skb = sock_alloc_send_skb(sk, 1508 alloclen + hh_len, 1509 (flags & MSG_DONTWAIT), &err); 1510 } else { 1511 skb = NULL; 1512 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <= 1513 2 * sk->sk_sndbuf) 1514 skb = alloc_skb(alloclen + hh_len, 1515 sk->sk_allocation); 1516 if (unlikely(!skb)) 1517 err = -ENOBUFS; 1518 } 1519 if (!skb) 1520 goto error; 1521 /* 1522 * Fill in the control structures 1523 */ 1524 skb->protocol = htons(ETH_P_IPV6); 1525 skb->ip_summed = csummode; 1526 skb->csum = 0; 1527 /* reserve for fragmentation and ipsec header */ 1528 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) + 1529 dst_exthdrlen); 1530 1531 /* 1532 * Find where to start putting bytes 1533 */ 1534 data = skb_put(skb, fraglen - pagedlen); 1535 skb_set_network_header(skb, exthdrlen); 1536 data += fragheaderlen; 1537 skb->transport_header = (skb->network_header + 1538 fragheaderlen); 1539 if (fraggap) { 1540 skb->csum = skb_copy_and_csum_bits( 1541 skb_prev, maxfraglen, 1542 data + transhdrlen, fraggap, 0); 1543 skb_prev->csum = csum_sub(skb_prev->csum, 1544 skb->csum); 1545 data += fraggap; 1546 pskb_trim_unique(skb_prev, maxfraglen); 1547 } 1548 if (copy > 0 && 1549 getfrag(from, data + transhdrlen, offset, 1550 copy, fraggap, skb) < 0) { 1551 err = -EFAULT; 1552 kfree_skb(skb); 1553 goto error; 1554 } 1555 1556 offset += copy; 1557 length -= copy + transhdrlen; 1558 transhdrlen = 0; 1559 exthdrlen = 0; 1560 dst_exthdrlen = 0; 1561 1562 /* Only the initial fragment is time stamped */ 1563 skb_shinfo(skb)->tx_flags = cork->tx_flags; 1564 cork->tx_flags = 0; 1565 skb_shinfo(skb)->tskey = tskey; 1566 tskey = 0; 1567 skb_zcopy_set(skb, uarg, &extra_uref); 1568 1569 if ((flags & MSG_CONFIRM) && !skb_prev) 1570 skb_set_dst_pending_confirm(skb, 1); 1571 1572 /* 1573 * Put the packet on the pending queue 1574 */ 1575 if (!skb->destructor) { 1576 skb->destructor = sock_wfree; 1577 skb->sk = sk; 1578 wmem_alloc_delta += skb->truesize; 1579 } 1580 __skb_queue_tail(queue, skb); 1581 continue; 1582 } 1583 1584 if (copy > length) 1585 copy = length; 1586 1587 if (!(rt->dst.dev->features&NETIF_F_SG) && 1588 skb_tailroom(skb) >= copy) { 1589 unsigned int off; 1590 1591 off = skb->len; 1592 if (getfrag(from, skb_put(skb, copy), 1593 offset, copy, off, skb) < 0) { 1594 __skb_trim(skb, off); 1595 err = -EFAULT; 1596 goto error; 1597 } 1598 } else if (!uarg || !uarg->zerocopy) { 1599 int i = skb_shinfo(skb)->nr_frags; 1600 1601 err = -ENOMEM; 1602 if (!sk_page_frag_refill(sk, pfrag)) 1603 goto error; 1604 1605 if (!skb_can_coalesce(skb, i, pfrag->page, 1606 pfrag->offset)) { 1607 err = -EMSGSIZE; 1608 if (i == MAX_SKB_FRAGS) 1609 goto error; 1610 1611 __skb_fill_page_desc(skb, i, pfrag->page, 1612 pfrag->offset, 0); 1613 skb_shinfo(skb)->nr_frags = ++i; 1614 get_page(pfrag->page); 1615 } 1616 copy = min_t(int, copy, pfrag->size - pfrag->offset); 1617 if (getfrag(from, 1618 page_address(pfrag->page) + pfrag->offset, 1619 offset, copy, skb->len, skb) < 0) 1620 goto error_efault; 1621 1622 pfrag->offset += copy; 1623 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); 1624 skb->len += copy; 1625 skb->data_len += copy; 1626 skb->truesize += copy; 1627 wmem_alloc_delta += copy; 1628 } else { 1629 err = skb_zerocopy_iter_dgram(skb, from, copy); 1630 if (err < 0) 1631 goto error; 1632 } 1633 offset += copy; 1634 length -= copy; 1635 } 1636 1637 if (wmem_alloc_delta) 1638 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1639 return 0; 1640 1641 error_efault: 1642 err = -EFAULT; 1643 error: 1644 if (uarg) 1645 sock_zerocopy_put_abort(uarg, extra_uref); 1646 cork->length -= length; 1647 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); 1648 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); 1649 return err; 1650 } 1651 1652 int ip6_append_data(struct sock *sk, 1653 int getfrag(void *from, char *to, int offset, int len, 1654 int odd, struct sk_buff *skb), 1655 void *from, int length, int transhdrlen, 1656 struct ipcm6_cookie *ipc6, struct flowi6 *fl6, 1657 struct rt6_info *rt, unsigned int flags) 1658 { 1659 struct inet_sock *inet = inet_sk(sk); 1660 struct ipv6_pinfo *np = inet6_sk(sk); 1661 int exthdrlen; 1662 int err; 1663 1664 if (flags&MSG_PROBE) 1665 return 0; 1666 if (skb_queue_empty(&sk->sk_write_queue)) { 1667 /* 1668 * setup for corking 1669 */ 1670 err = ip6_setup_cork(sk, &inet->cork, &np->cork, 1671 ipc6, rt, fl6); 1672 if (err) 1673 return err; 1674 1675 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0); 1676 length += exthdrlen; 1677 transhdrlen += exthdrlen; 1678 } else { 1679 fl6 = &inet->cork.fl.u.ip6; 1680 transhdrlen = 0; 1681 } 1682 1683 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base, 1684 &np->cork, sk_page_frag(sk), getfrag, 1685 from, length, transhdrlen, flags, ipc6); 1686 } 1687 EXPORT_SYMBOL_GPL(ip6_append_data); 1688 1689 static void ip6_cork_release(struct inet_cork_full *cork, 1690 struct inet6_cork *v6_cork) 1691 { 1692 if (v6_cork->opt) { 1693 kfree(v6_cork->opt->dst0opt); 1694 kfree(v6_cork->opt->dst1opt); 1695 kfree(v6_cork->opt->hopopt); 1696 kfree(v6_cork->opt->srcrt); 1697 kfree(v6_cork->opt); 1698 v6_cork->opt = NULL; 1699 } 1700 1701 if (cork->base.dst) { 1702 dst_release(cork->base.dst); 1703 cork->base.dst = NULL; 1704 cork->base.flags &= ~IPCORK_ALLFRAG; 1705 } 1706 memset(&cork->fl, 0, sizeof(cork->fl)); 1707 } 1708 1709 struct sk_buff *__ip6_make_skb(struct sock *sk, 1710 struct sk_buff_head *queue, 1711 struct inet_cork_full *cork, 1712 struct inet6_cork *v6_cork) 1713 { 1714 struct sk_buff *skb, *tmp_skb; 1715 struct sk_buff **tail_skb; 1716 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf; 1717 struct ipv6_pinfo *np = inet6_sk(sk); 1718 struct net *net = sock_net(sk); 1719 struct ipv6hdr *hdr; 1720 struct ipv6_txoptions *opt = v6_cork->opt; 1721 struct rt6_info *rt = (struct rt6_info *)cork->base.dst; 1722 struct flowi6 *fl6 = &cork->fl.u.ip6; 1723 unsigned char proto = fl6->flowi6_proto; 1724 1725 skb = __skb_dequeue(queue); 1726 if (!skb) 1727 goto out; 1728 tail_skb = &(skb_shinfo(skb)->frag_list); 1729 1730 /* move skb->data to ip header from ext header */ 1731 if (skb->data < skb_network_header(skb)) 1732 __skb_pull(skb, skb_network_offset(skb)); 1733 while ((tmp_skb = __skb_dequeue(queue)) != NULL) { 1734 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1735 *tail_skb = tmp_skb; 1736 tail_skb = &(tmp_skb->next); 1737 skb->len += tmp_skb->len; 1738 skb->data_len += tmp_skb->len; 1739 skb->truesize += tmp_skb->truesize; 1740 tmp_skb->destructor = NULL; 1741 tmp_skb->sk = NULL; 1742 } 1743 1744 /* Allow local fragmentation. */ 1745 skb->ignore_df = ip6_sk_ignore_df(sk); 1746 1747 *final_dst = fl6->daddr; 1748 __skb_pull(skb, skb_network_header_len(skb)); 1749 if (opt && opt->opt_flen) 1750 ipv6_push_frag_opts(skb, opt, &proto); 1751 if (opt && opt->opt_nflen) 1752 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr); 1753 1754 skb_push(skb, sizeof(struct ipv6hdr)); 1755 skb_reset_network_header(skb); 1756 hdr = ipv6_hdr(skb); 1757 1758 ip6_flow_hdr(hdr, v6_cork->tclass, 1759 ip6_make_flowlabel(net, skb, fl6->flowlabel, 1760 ip6_autoflowlabel(net, np), fl6)); 1761 hdr->hop_limit = v6_cork->hop_limit; 1762 hdr->nexthdr = proto; 1763 hdr->saddr = fl6->saddr; 1764 hdr->daddr = *final_dst; 1765 1766 skb->priority = sk->sk_priority; 1767 skb->mark = sk->sk_mark; 1768 1769 skb->tstamp = cork->base.transmit_time; 1770 1771 skb_dst_set(skb, dst_clone(&rt->dst)); 1772 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len); 1773 if (proto == IPPROTO_ICMPV6) { 1774 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb)); 1775 1776 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type); 1777 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS); 1778 } 1779 1780 ip6_cork_release(cork, v6_cork); 1781 out: 1782 return skb; 1783 } 1784 1785 int ip6_send_skb(struct sk_buff *skb) 1786 { 1787 struct net *net = sock_net(skb->sk); 1788 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb); 1789 int err; 1790 1791 err = ip6_local_out(net, skb->sk, skb); 1792 if (err) { 1793 if (err > 0) 1794 err = net_xmit_errno(err); 1795 if (err) 1796 IP6_INC_STATS(net, rt->rt6i_idev, 1797 IPSTATS_MIB_OUTDISCARDS); 1798 } 1799 1800 return err; 1801 } 1802 1803 int ip6_push_pending_frames(struct sock *sk) 1804 { 1805 struct sk_buff *skb; 1806 1807 skb = ip6_finish_skb(sk); 1808 if (!skb) 1809 return 0; 1810 1811 return ip6_send_skb(skb); 1812 } 1813 EXPORT_SYMBOL_GPL(ip6_push_pending_frames); 1814 1815 static void __ip6_flush_pending_frames(struct sock *sk, 1816 struct sk_buff_head *queue, 1817 struct inet_cork_full *cork, 1818 struct inet6_cork *v6_cork) 1819 { 1820 struct sk_buff *skb; 1821 1822 while ((skb = __skb_dequeue_tail(queue)) != NULL) { 1823 if (skb_dst(skb)) 1824 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)), 1825 IPSTATS_MIB_OUTDISCARDS); 1826 kfree_skb(skb); 1827 } 1828 1829 ip6_cork_release(cork, v6_cork); 1830 } 1831 1832 void ip6_flush_pending_frames(struct sock *sk) 1833 { 1834 __ip6_flush_pending_frames(sk, &sk->sk_write_queue, 1835 &inet_sk(sk)->cork, &inet6_sk(sk)->cork); 1836 } 1837 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames); 1838 1839 struct sk_buff *ip6_make_skb(struct sock *sk, 1840 int getfrag(void *from, char *to, int offset, 1841 int len, int odd, struct sk_buff *skb), 1842 void *from, int length, int transhdrlen, 1843 struct ipcm6_cookie *ipc6, struct flowi6 *fl6, 1844 struct rt6_info *rt, unsigned int flags, 1845 struct inet_cork_full *cork) 1846 { 1847 struct inet6_cork v6_cork; 1848 struct sk_buff_head queue; 1849 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0); 1850 int err; 1851 1852 if (flags & MSG_PROBE) 1853 return NULL; 1854 1855 __skb_queue_head_init(&queue); 1856 1857 cork->base.flags = 0; 1858 cork->base.addr = 0; 1859 cork->base.opt = NULL; 1860 cork->base.dst = NULL; 1861 v6_cork.opt = NULL; 1862 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6); 1863 if (err) { 1864 ip6_cork_release(cork, &v6_cork); 1865 return ERR_PTR(err); 1866 } 1867 if (ipc6->dontfrag < 0) 1868 ipc6->dontfrag = inet6_sk(sk)->dontfrag; 1869 1870 err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork, 1871 ¤t->task_frag, getfrag, from, 1872 length + exthdrlen, transhdrlen + exthdrlen, 1873 flags, ipc6); 1874 if (err) { 1875 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork); 1876 return ERR_PTR(err); 1877 } 1878 1879 return __ip6_make_skb(sk, &queue, cork, &v6_cork); 1880 } 1881