1 /* 2 * IPv6 output functions 3 * Linux INET6 implementation 4 * 5 * Authors: 6 * Pedro Roque <roque@di.fc.ul.pt> 7 * 8 * $Id: ip6_output.c,v 1.34 2002/02/01 22:01:04 davem Exp $ 9 * 10 * Based on linux/net/ipv4/ip_output.c 11 * 12 * This program is free software; you can redistribute it and/or 13 * modify it under the terms of the GNU General Public License 14 * as published by the Free Software Foundation; either version 15 * 2 of the License, or (at your option) any later version. 16 * 17 * Changes: 18 * A.N.Kuznetsov : airthmetics in fragmentation. 19 * extension headers are implemented. 20 * route changes now work. 21 * ip6_forward does not confuse sniffers. 22 * etc. 23 * 24 * H. von Brand : Added missing #include <linux/string.h> 25 * Imran Patel : frag id should be in NBO 26 * Kazunori MIYAZAWA @USAGI 27 * : add ip6_append_data and related functions 28 * for datagram xmit 29 */ 30 31 #include <linux/errno.h> 32 #include <linux/types.h> 33 #include <linux/string.h> 34 #include <linux/socket.h> 35 #include <linux/net.h> 36 #include <linux/netdevice.h> 37 #include <linux/if_arp.h> 38 #include <linux/in6.h> 39 #include <linux/tcp.h> 40 #include <linux/route.h> 41 #include <linux/module.h> 42 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 59 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)); 60 61 static __inline__ void ipv6_select_ident(struct sk_buff *skb, struct frag_hdr *fhdr) 62 { 63 static u32 ipv6_fragmentation_id = 1; 64 static DEFINE_SPINLOCK(ip6_id_lock); 65 66 spin_lock_bh(&ip6_id_lock); 67 fhdr->identification = htonl(ipv6_fragmentation_id); 68 if (++ipv6_fragmentation_id == 0) 69 ipv6_fragmentation_id = 1; 70 spin_unlock_bh(&ip6_id_lock); 71 } 72 73 static inline int ip6_output_finish(struct sk_buff *skb) 74 { 75 struct dst_entry *dst = skb->dst; 76 77 if (dst->hh) 78 return neigh_hh_output(dst->hh, skb); 79 else if (dst->neighbour) 80 return dst->neighbour->output(skb); 81 82 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 83 kfree_skb(skb); 84 return -EINVAL; 85 86 } 87 88 /* dev_loopback_xmit for use with netfilter. */ 89 static int ip6_dev_loopback_xmit(struct sk_buff *newskb) 90 { 91 skb_reset_mac_header(newskb); 92 __skb_pull(newskb, skb_network_offset(newskb)); 93 newskb->pkt_type = PACKET_LOOPBACK; 94 newskb->ip_summed = CHECKSUM_UNNECESSARY; 95 BUG_TRAP(newskb->dst); 96 97 netif_rx(newskb); 98 return 0; 99 } 100 101 102 static int ip6_output2(struct sk_buff *skb) 103 { 104 struct dst_entry *dst = skb->dst; 105 struct net_device *dev = dst->dev; 106 107 skb->protocol = htons(ETH_P_IPV6); 108 skb->dev = dev; 109 110 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) { 111 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL; 112 struct inet6_dev *idev = ip6_dst_idev(skb->dst); 113 114 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) && 115 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr, 116 &ipv6_hdr(skb)->saddr)) { 117 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); 118 119 /* Do not check for IFF_ALLMULTI; multicast routing 120 is not supported in any case. 121 */ 122 if (newskb) 123 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL, 124 newskb->dev, 125 ip6_dev_loopback_xmit); 126 127 if (ipv6_hdr(skb)->hop_limit == 0) { 128 IP6_INC_STATS(idev, IPSTATS_MIB_OUTDISCARDS); 129 kfree_skb(skb); 130 return 0; 131 } 132 } 133 134 IP6_INC_STATS(idev, IPSTATS_MIB_OUTMCASTPKTS); 135 } 136 137 return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish); 138 } 139 140 static inline int ip6_skb_dst_mtu(struct sk_buff *skb) 141 { 142 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL; 143 144 return (np && np->pmtudisc == IPV6_PMTUDISC_PROBE) ? 145 skb->dst->dev->mtu : dst_mtu(skb->dst); 146 } 147 148 int ip6_output(struct sk_buff *skb) 149 { 150 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) || 151 dst_allfrag(skb->dst)) 152 return ip6_fragment(skb, ip6_output2); 153 else 154 return ip6_output2(skb); 155 } 156 157 /* 158 * xmit an sk_buff (used by TCP) 159 */ 160 161 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl, 162 struct ipv6_txoptions *opt, int ipfragok) 163 { 164 struct ipv6_pinfo *np = inet6_sk(sk); 165 struct in6_addr *first_hop = &fl->fl6_dst; 166 struct dst_entry *dst = skb->dst; 167 struct ipv6hdr *hdr; 168 u8 proto = fl->proto; 169 int seg_len = skb->len; 170 int hlimit, tclass; 171 u32 mtu; 172 173 if (opt) { 174 int head_room; 175 176 /* First: exthdrs may take lots of space (~8K for now) 177 MAX_HEADER is not enough. 178 */ 179 head_room = opt->opt_nflen + opt->opt_flen; 180 seg_len += head_room; 181 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev); 182 183 if (skb_headroom(skb) < head_room) { 184 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room); 185 if (skb2 == NULL) { 186 IP6_INC_STATS(ip6_dst_idev(skb->dst), 187 IPSTATS_MIB_OUTDISCARDS); 188 kfree_skb(skb); 189 return -ENOBUFS; 190 } 191 kfree_skb(skb); 192 skb = skb2; 193 if (sk) 194 skb_set_owner_w(skb, sk); 195 } 196 if (opt->opt_flen) 197 ipv6_push_frag_opts(skb, opt, &proto); 198 if (opt->opt_nflen) 199 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop); 200 } 201 202 skb_push(skb, sizeof(struct ipv6hdr)); 203 skb_reset_network_header(skb); 204 hdr = ipv6_hdr(skb); 205 206 /* 207 * Fill in the IPv6 header 208 */ 209 210 hlimit = -1; 211 if (np) 212 hlimit = np->hop_limit; 213 if (hlimit < 0) 214 hlimit = dst_metric(dst, RTAX_HOPLIMIT); 215 if (hlimit < 0) 216 hlimit = ipv6_get_hoplimit(dst->dev); 217 218 tclass = -1; 219 if (np) 220 tclass = np->tclass; 221 if (tclass < 0) 222 tclass = 0; 223 224 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel; 225 226 hdr->payload_len = htons(seg_len); 227 hdr->nexthdr = proto; 228 hdr->hop_limit = hlimit; 229 230 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src); 231 ipv6_addr_copy(&hdr->daddr, first_hop); 232 233 skb->priority = sk->sk_priority; 234 235 mtu = dst_mtu(dst); 236 if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) { 237 IP6_INC_STATS(ip6_dst_idev(skb->dst), 238 IPSTATS_MIB_OUTREQUESTS); 239 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev, 240 dst_output); 241 } 242 243 if (net_ratelimit()) 244 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n"); 245 skb->dev = dst->dev; 246 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); 247 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS); 248 kfree_skb(skb); 249 return -EMSGSIZE; 250 } 251 252 EXPORT_SYMBOL(ip6_xmit); 253 254 /* 255 * To avoid extra problems ND packets are send through this 256 * routine. It's code duplication but I really want to avoid 257 * extra checks since ipv6_build_header is used by TCP (which 258 * is for us performance critical) 259 */ 260 261 int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev, 262 struct in6_addr *saddr, struct in6_addr *daddr, 263 int proto, int len) 264 { 265 struct ipv6_pinfo *np = inet6_sk(sk); 266 struct ipv6hdr *hdr; 267 int totlen; 268 269 skb->protocol = htons(ETH_P_IPV6); 270 skb->dev = dev; 271 272 totlen = len + sizeof(struct ipv6hdr); 273 274 skb_reset_network_header(skb); 275 skb_put(skb, sizeof(struct ipv6hdr)); 276 hdr = ipv6_hdr(skb); 277 278 *(__be32*)hdr = htonl(0x60000000); 279 280 hdr->payload_len = htons(len); 281 hdr->nexthdr = proto; 282 hdr->hop_limit = np->hop_limit; 283 284 ipv6_addr_copy(&hdr->saddr, saddr); 285 ipv6_addr_copy(&hdr->daddr, daddr); 286 287 return 0; 288 } 289 290 static int ip6_call_ra_chain(struct sk_buff *skb, int sel) 291 { 292 struct ip6_ra_chain *ra; 293 struct sock *last = NULL; 294 295 read_lock(&ip6_ra_lock); 296 for (ra = ip6_ra_chain; ra; ra = ra->next) { 297 struct sock *sk = ra->sk; 298 if (sk && ra->sel == sel && 299 (!sk->sk_bound_dev_if || 300 sk->sk_bound_dev_if == skb->dev->ifindex)) { 301 if (last) { 302 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 303 if (skb2) 304 rawv6_rcv(last, skb2); 305 } 306 last = sk; 307 } 308 } 309 310 if (last) { 311 rawv6_rcv(last, skb); 312 read_unlock(&ip6_ra_lock); 313 return 1; 314 } 315 read_unlock(&ip6_ra_lock); 316 return 0; 317 } 318 319 static int ip6_forward_proxy_check(struct sk_buff *skb) 320 { 321 struct ipv6hdr *hdr = ipv6_hdr(skb); 322 u8 nexthdr = hdr->nexthdr; 323 int offset; 324 325 if (ipv6_ext_hdr(nexthdr)) { 326 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr); 327 if (offset < 0) 328 return 0; 329 } else 330 offset = sizeof(struct ipv6hdr); 331 332 if (nexthdr == IPPROTO_ICMPV6) { 333 struct icmp6hdr *icmp6; 334 335 if (!pskb_may_pull(skb, (skb_network_header(skb) + 336 offset + 1 - skb->data))) 337 return 0; 338 339 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset); 340 341 switch (icmp6->icmp6_type) { 342 case NDISC_ROUTER_SOLICITATION: 343 case NDISC_ROUTER_ADVERTISEMENT: 344 case NDISC_NEIGHBOUR_SOLICITATION: 345 case NDISC_NEIGHBOUR_ADVERTISEMENT: 346 case NDISC_REDIRECT: 347 /* For reaction involving unicast neighbor discovery 348 * message destined to the proxied address, pass it to 349 * input function. 350 */ 351 return 1; 352 default: 353 break; 354 } 355 } 356 357 /* 358 * The proxying router can't forward traffic sent to a link-local 359 * address, so signal the sender and discard the packet. This 360 * behavior is clarified by the MIPv6 specification. 361 */ 362 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) { 363 dst_link_failure(skb); 364 return -1; 365 } 366 367 return 0; 368 } 369 370 static inline int ip6_forward_finish(struct sk_buff *skb) 371 { 372 return dst_output(skb); 373 } 374 375 int ip6_forward(struct sk_buff *skb) 376 { 377 struct dst_entry *dst = skb->dst; 378 struct ipv6hdr *hdr = ipv6_hdr(skb); 379 struct inet6_skb_parm *opt = IP6CB(skb); 380 381 if (ipv6_devconf.forwarding == 0) 382 goto error; 383 384 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) { 385 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS); 386 goto drop; 387 } 388 389 skb_forward_csum(skb); 390 391 /* 392 * We DO NOT make any processing on 393 * RA packets, pushing them to user level AS IS 394 * without ane WARRANTY that application will be able 395 * to interpret them. The reason is that we 396 * cannot make anything clever here. 397 * 398 * We are not end-node, so that if packet contains 399 * AH/ESP, we cannot make anything. 400 * Defragmentation also would be mistake, RA packets 401 * cannot be fragmented, because there is no warranty 402 * that different fragments will go along one path. --ANK 403 */ 404 if (opt->ra) { 405 u8 *ptr = skb_network_header(skb) + opt->ra; 406 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3])) 407 return 0; 408 } 409 410 /* 411 * check and decrement ttl 412 */ 413 if (hdr->hop_limit <= 1) { 414 /* Force OUTPUT device used as source address */ 415 skb->dev = dst->dev; 416 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 417 0, skb->dev); 418 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS); 419 420 kfree_skb(skb); 421 return -ETIMEDOUT; 422 } 423 424 /* XXX: idev->cnf.proxy_ndp? */ 425 if (ipv6_devconf.proxy_ndp && 426 pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) { 427 int proxied = ip6_forward_proxy_check(skb); 428 if (proxied > 0) 429 return ip6_input(skb); 430 else if (proxied < 0) { 431 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS); 432 goto drop; 433 } 434 } 435 436 if (!xfrm6_route_forward(skb)) { 437 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS); 438 goto drop; 439 } 440 dst = skb->dst; 441 442 /* IPv6 specs say nothing about it, but it is clear that we cannot 443 send redirects to source routed frames. 444 */ 445 if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) { 446 struct in6_addr *target = NULL; 447 struct rt6_info *rt; 448 struct neighbour *n = dst->neighbour; 449 450 /* 451 * incoming and outgoing devices are the same 452 * send a redirect. 453 */ 454 455 rt = (struct rt6_info *) dst; 456 if ((rt->rt6i_flags & RTF_GATEWAY)) 457 target = (struct in6_addr*)&n->primary_key; 458 else 459 target = &hdr->daddr; 460 461 /* Limit redirects both by destination (here) 462 and by source (inside ndisc_send_redirect) 463 */ 464 if (xrlim_allow(dst, 1*HZ)) 465 ndisc_send_redirect(skb, n, target); 466 } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK 467 |IPV6_ADDR_LINKLOCAL)) { 468 /* This check is security critical. */ 469 goto error; 470 } 471 472 if (skb->len > dst_mtu(dst)) { 473 /* Again, force OUTPUT device used as source address */ 474 skb->dev = dst->dev; 475 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev); 476 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS); 477 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS); 478 kfree_skb(skb); 479 return -EMSGSIZE; 480 } 481 482 if (skb_cow(skb, dst->dev->hard_header_len)) { 483 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS); 484 goto drop; 485 } 486 487 hdr = ipv6_hdr(skb); 488 489 /* Mangling hops number delayed to point after skb COW */ 490 491 hdr->hop_limit--; 492 493 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS); 494 return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish); 495 496 error: 497 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS); 498 drop: 499 kfree_skb(skb); 500 return -EINVAL; 501 } 502 503 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from) 504 { 505 to->pkt_type = from->pkt_type; 506 to->priority = from->priority; 507 to->protocol = from->protocol; 508 dst_release(to->dst); 509 to->dst = dst_clone(from->dst); 510 to->dev = from->dev; 511 to->mark = from->mark; 512 513 #ifdef CONFIG_NET_SCHED 514 to->tc_index = from->tc_index; 515 #endif 516 nf_copy(to, from); 517 skb_copy_secmark(to, from); 518 } 519 520 int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr) 521 { 522 u16 offset = sizeof(struct ipv6hdr); 523 struct ipv6_opt_hdr *exthdr = 524 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1); 525 unsigned int packet_len = skb->tail - skb->network_header; 526 int found_rhdr = 0; 527 *nexthdr = &ipv6_hdr(skb)->nexthdr; 528 529 while (offset + 1 <= packet_len) { 530 531 switch (**nexthdr) { 532 533 case NEXTHDR_HOP: 534 break; 535 case NEXTHDR_ROUTING: 536 found_rhdr = 1; 537 break; 538 case NEXTHDR_DEST: 539 #ifdef CONFIG_IPV6_MIP6 540 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0) 541 break; 542 #endif 543 if (found_rhdr) 544 return offset; 545 break; 546 default : 547 return offset; 548 } 549 550 offset += ipv6_optlen(exthdr); 551 *nexthdr = &exthdr->nexthdr; 552 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) + 553 offset); 554 } 555 556 return offset; 557 } 558 EXPORT_SYMBOL_GPL(ip6_find_1stfragopt); 559 560 static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)) 561 { 562 struct net_device *dev; 563 struct sk_buff *frag; 564 struct rt6_info *rt = (struct rt6_info*)skb->dst; 565 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL; 566 struct ipv6hdr *tmp_hdr; 567 struct frag_hdr *fh; 568 unsigned int mtu, hlen, left, len; 569 __be32 frag_id = 0; 570 int ptr, offset = 0, err=0; 571 u8 *prevhdr, nexthdr = 0; 572 573 dev = rt->u.dst.dev; 574 hlen = ip6_find_1stfragopt(skb, &prevhdr); 575 nexthdr = *prevhdr; 576 577 mtu = ip6_skb_dst_mtu(skb); 578 579 /* We must not fragment if the socket is set to force MTU discovery 580 * or if the skb it not generated by a local socket. (This last 581 * check should be redundant, but it's free.) 582 */ 583 if (!np || np->pmtudisc >= IPV6_PMTUDISC_DO) { 584 skb->dev = skb->dst->dev; 585 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev); 586 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS); 587 kfree_skb(skb); 588 return -EMSGSIZE; 589 } 590 591 if (np && np->frag_size < mtu) { 592 if (np->frag_size) 593 mtu = np->frag_size; 594 } 595 mtu -= hlen + sizeof(struct frag_hdr); 596 597 if (skb_shinfo(skb)->frag_list) { 598 int first_len = skb_pagelen(skb); 599 600 if (first_len - hlen > mtu || 601 ((first_len - hlen) & 7) || 602 skb_cloned(skb)) 603 goto slow_path; 604 605 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) { 606 /* Correct geometry. */ 607 if (frag->len > mtu || 608 ((frag->len & 7) && frag->next) || 609 skb_headroom(frag) < hlen) 610 goto slow_path; 611 612 /* Partially cloned skb? */ 613 if (skb_shared(frag)) 614 goto slow_path; 615 616 BUG_ON(frag->sk); 617 if (skb->sk) { 618 sock_hold(skb->sk); 619 frag->sk = skb->sk; 620 frag->destructor = sock_wfree; 621 skb->truesize -= frag->truesize; 622 } 623 } 624 625 err = 0; 626 offset = 0; 627 frag = skb_shinfo(skb)->frag_list; 628 skb_shinfo(skb)->frag_list = NULL; 629 /* BUILD HEADER */ 630 631 *prevhdr = NEXTHDR_FRAGMENT; 632 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC); 633 if (!tmp_hdr) { 634 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS); 635 return -ENOMEM; 636 } 637 638 __skb_pull(skb, hlen); 639 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr)); 640 __skb_push(skb, hlen); 641 skb_reset_network_header(skb); 642 memcpy(skb_network_header(skb), tmp_hdr, hlen); 643 644 ipv6_select_ident(skb, fh); 645 fh->nexthdr = nexthdr; 646 fh->reserved = 0; 647 fh->frag_off = htons(IP6_MF); 648 frag_id = fh->identification; 649 650 first_len = skb_pagelen(skb); 651 skb->data_len = first_len - skb_headlen(skb); 652 skb->len = first_len; 653 ipv6_hdr(skb)->payload_len = htons(first_len - 654 sizeof(struct ipv6hdr)); 655 656 dst_hold(&rt->u.dst); 657 658 for (;;) { 659 /* Prepare header of the next frame, 660 * before previous one went down. */ 661 if (frag) { 662 frag->ip_summed = CHECKSUM_NONE; 663 skb_reset_transport_header(frag); 664 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr)); 665 __skb_push(frag, hlen); 666 skb_reset_network_header(frag); 667 memcpy(skb_network_header(frag), tmp_hdr, 668 hlen); 669 offset += skb->len - hlen - sizeof(struct frag_hdr); 670 fh->nexthdr = nexthdr; 671 fh->reserved = 0; 672 fh->frag_off = htons(offset); 673 if (frag->next != NULL) 674 fh->frag_off |= htons(IP6_MF); 675 fh->identification = frag_id; 676 ipv6_hdr(frag)->payload_len = 677 htons(frag->len - 678 sizeof(struct ipv6hdr)); 679 ip6_copy_metadata(frag, skb); 680 } 681 682 err = output(skb); 683 if(!err) 684 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES); 685 686 if (err || !frag) 687 break; 688 689 skb = frag; 690 frag = skb->next; 691 skb->next = NULL; 692 } 693 694 kfree(tmp_hdr); 695 696 if (err == 0) { 697 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS); 698 dst_release(&rt->u.dst); 699 return 0; 700 } 701 702 while (frag) { 703 skb = frag->next; 704 kfree_skb(frag); 705 frag = skb; 706 } 707 708 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS); 709 dst_release(&rt->u.dst); 710 return err; 711 } 712 713 slow_path: 714 left = skb->len - hlen; /* Space per frame */ 715 ptr = hlen; /* Where to start from */ 716 717 /* 718 * Fragment the datagram. 719 */ 720 721 *prevhdr = NEXTHDR_FRAGMENT; 722 723 /* 724 * Keep copying data until we run out. 725 */ 726 while(left > 0) { 727 len = left; 728 /* IF: it doesn't fit, use 'mtu' - the data space left */ 729 if (len > mtu) 730 len = mtu; 731 /* IF: we are not sending upto and including the packet end 732 then align the next start on an eight byte boundary */ 733 if (len < left) { 734 len &= ~7; 735 } 736 /* 737 * Allocate buffer. 738 */ 739 740 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) { 741 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n"); 742 IP6_INC_STATS(ip6_dst_idev(skb->dst), 743 IPSTATS_MIB_FRAGFAILS); 744 err = -ENOMEM; 745 goto fail; 746 } 747 748 /* 749 * Set up data on packet 750 */ 751 752 ip6_copy_metadata(frag, skb); 753 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev)); 754 skb_put(frag, len + hlen + sizeof(struct frag_hdr)); 755 skb_reset_network_header(frag); 756 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen); 757 frag->transport_header = (frag->network_header + hlen + 758 sizeof(struct frag_hdr)); 759 760 /* 761 * Charge the memory for the fragment to any owner 762 * it might possess 763 */ 764 if (skb->sk) 765 skb_set_owner_w(frag, skb->sk); 766 767 /* 768 * Copy the packet header into the new buffer. 769 */ 770 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen); 771 772 /* 773 * Build fragment header. 774 */ 775 fh->nexthdr = nexthdr; 776 fh->reserved = 0; 777 if (!frag_id) { 778 ipv6_select_ident(skb, fh); 779 frag_id = fh->identification; 780 } else 781 fh->identification = frag_id; 782 783 /* 784 * Copy a block of the IP datagram. 785 */ 786 if (skb_copy_bits(skb, ptr, skb_transport_header(skb), len)) 787 BUG(); 788 left -= len; 789 790 fh->frag_off = htons(offset); 791 if (left > 0) 792 fh->frag_off |= htons(IP6_MF); 793 ipv6_hdr(frag)->payload_len = htons(frag->len - 794 sizeof(struct ipv6hdr)); 795 796 ptr += len; 797 offset += len; 798 799 /* 800 * Put this fragment into the sending queue. 801 */ 802 err = output(frag); 803 if (err) 804 goto fail; 805 806 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES); 807 } 808 IP6_INC_STATS(ip6_dst_idev(skb->dst), 809 IPSTATS_MIB_FRAGOKS); 810 kfree_skb(skb); 811 return err; 812 813 fail: 814 IP6_INC_STATS(ip6_dst_idev(skb->dst), 815 IPSTATS_MIB_FRAGFAILS); 816 kfree_skb(skb); 817 return err; 818 } 819 820 static inline int ip6_rt_check(struct rt6key *rt_key, 821 struct in6_addr *fl_addr, 822 struct in6_addr *addr_cache) 823 { 824 return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) && 825 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache))); 826 } 827 828 static struct dst_entry *ip6_sk_dst_check(struct sock *sk, 829 struct dst_entry *dst, 830 struct flowi *fl) 831 { 832 struct ipv6_pinfo *np = inet6_sk(sk); 833 struct rt6_info *rt = (struct rt6_info *)dst; 834 835 if (!dst) 836 goto out; 837 838 /* Yes, checking route validity in not connected 839 * case is not very simple. Take into account, 840 * that we do not support routing by source, TOS, 841 * and MSG_DONTROUTE --ANK (980726) 842 * 843 * 1. ip6_rt_check(): If route was host route, 844 * check that cached destination is current. 845 * If it is network route, we still may 846 * check its validity using saved pointer 847 * to the last used address: daddr_cache. 848 * We do not want to save whole address now, 849 * (because main consumer of this service 850 * is tcp, which has not this problem), 851 * so that the last trick works only on connected 852 * sockets. 853 * 2. oif also should be the same. 854 */ 855 if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) || 856 #ifdef CONFIG_IPV6_SUBTREES 857 ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) || 858 #endif 859 (fl->oif && fl->oif != dst->dev->ifindex)) { 860 dst_release(dst); 861 dst = NULL; 862 } 863 864 out: 865 return dst; 866 } 867 868 static int ip6_dst_lookup_tail(struct sock *sk, 869 struct dst_entry **dst, struct flowi *fl) 870 { 871 int err; 872 873 if (*dst == NULL) 874 *dst = ip6_route_output(sk, fl); 875 876 if ((err = (*dst)->error)) 877 goto out_err_release; 878 879 if (ipv6_addr_any(&fl->fl6_src)) { 880 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src); 881 if (err) 882 goto out_err_release; 883 } 884 885 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD 886 /* 887 * Here if the dst entry we've looked up 888 * has a neighbour entry that is in the INCOMPLETE 889 * state and the src address from the flow is 890 * marked as OPTIMISTIC, we release the found 891 * dst entry and replace it instead with the 892 * dst entry of the nexthop router 893 */ 894 if (!((*dst)->neighbour->nud_state & NUD_VALID)) { 895 struct inet6_ifaddr *ifp; 896 struct flowi fl_gw; 897 int redirect; 898 899 ifp = ipv6_get_ifaddr(&fl->fl6_src, (*dst)->dev, 1); 900 901 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC); 902 if (ifp) 903 in6_ifa_put(ifp); 904 905 if (redirect) { 906 /* 907 * We need to get the dst entry for the 908 * default router instead 909 */ 910 dst_release(*dst); 911 memcpy(&fl_gw, fl, sizeof(struct flowi)); 912 memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr)); 913 *dst = ip6_route_output(sk, &fl_gw); 914 if ((err = (*dst)->error)) 915 goto out_err_release; 916 } 917 } 918 #endif 919 920 return 0; 921 922 out_err_release: 923 dst_release(*dst); 924 *dst = NULL; 925 return err; 926 } 927 928 /** 929 * ip6_dst_lookup - perform route lookup on flow 930 * @sk: socket which provides route info 931 * @dst: pointer to dst_entry * for result 932 * @fl: flow to lookup 933 * 934 * This function performs a route lookup on the given flow. 935 * 936 * It returns zero on success, or a standard errno code on error. 937 */ 938 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl) 939 { 940 *dst = NULL; 941 return ip6_dst_lookup_tail(sk, dst, fl); 942 } 943 EXPORT_SYMBOL_GPL(ip6_dst_lookup); 944 945 /** 946 * ip6_sk_dst_lookup - perform socket cached route lookup on flow 947 * @sk: socket which provides the dst cache and route info 948 * @dst: pointer to dst_entry * for result 949 * @fl: flow to lookup 950 * 951 * This function performs a route lookup on the given flow with the 952 * possibility of using the cached route in the socket if it is valid. 953 * It will take the socket dst lock when operating on the dst cache. 954 * As a result, this function can only be used in process context. 955 * 956 * It returns zero on success, or a standard errno code on error. 957 */ 958 int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl) 959 { 960 *dst = NULL; 961 if (sk) { 962 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie); 963 *dst = ip6_sk_dst_check(sk, *dst, fl); 964 } 965 966 return ip6_dst_lookup_tail(sk, dst, fl); 967 } 968 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup); 969 970 static inline int ip6_ufo_append_data(struct sock *sk, 971 int getfrag(void *from, char *to, int offset, int len, 972 int odd, struct sk_buff *skb), 973 void *from, int length, int hh_len, int fragheaderlen, 974 int transhdrlen, int mtu,unsigned int flags) 975 976 { 977 struct sk_buff *skb; 978 int err; 979 980 /* There is support for UDP large send offload by network 981 * device, so create one single skb packet containing complete 982 * udp datagram 983 */ 984 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) { 985 skb = sock_alloc_send_skb(sk, 986 hh_len + fragheaderlen + transhdrlen + 20, 987 (flags & MSG_DONTWAIT), &err); 988 if (skb == NULL) 989 return -ENOMEM; 990 991 /* reserve space for Hardware header */ 992 skb_reserve(skb, hh_len); 993 994 /* create space for UDP/IP header */ 995 skb_put(skb,fragheaderlen + transhdrlen); 996 997 /* initialize network header pointer */ 998 skb_reset_network_header(skb); 999 1000 /* initialize protocol header pointer */ 1001 skb->transport_header = skb->network_header + fragheaderlen; 1002 1003 skb->ip_summed = CHECKSUM_PARTIAL; 1004 skb->csum = 0; 1005 sk->sk_sndmsg_off = 0; 1006 } 1007 1008 err = skb_append_datato_frags(sk,skb, getfrag, from, 1009 (length - transhdrlen)); 1010 if (!err) { 1011 struct frag_hdr fhdr; 1012 1013 /* specify the length of each IP datagram fragment*/ 1014 skb_shinfo(skb)->gso_size = mtu - fragheaderlen - 1015 sizeof(struct frag_hdr); 1016 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 1017 ipv6_select_ident(skb, &fhdr); 1018 skb_shinfo(skb)->ip6_frag_id = fhdr.identification; 1019 __skb_queue_tail(&sk->sk_write_queue, skb); 1020 1021 return 0; 1022 } 1023 /* There is not enough support do UPD LSO, 1024 * so follow normal path 1025 */ 1026 kfree_skb(skb); 1027 1028 return err; 1029 } 1030 1031 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to, 1032 int offset, int len, int odd, struct sk_buff *skb), 1033 void *from, int length, int transhdrlen, 1034 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl, 1035 struct rt6_info *rt, unsigned int flags) 1036 { 1037 struct inet_sock *inet = inet_sk(sk); 1038 struct ipv6_pinfo *np = inet6_sk(sk); 1039 struct sk_buff *skb; 1040 unsigned int maxfraglen, fragheaderlen; 1041 int exthdrlen; 1042 int hh_len; 1043 int mtu; 1044 int copy; 1045 int err; 1046 int offset = 0; 1047 int csummode = CHECKSUM_NONE; 1048 1049 if (flags&MSG_PROBE) 1050 return 0; 1051 if (skb_queue_empty(&sk->sk_write_queue)) { 1052 /* 1053 * setup for corking 1054 */ 1055 if (opt) { 1056 if (np->cork.opt == NULL) { 1057 np->cork.opt = kmalloc(opt->tot_len, 1058 sk->sk_allocation); 1059 if (unlikely(np->cork.opt == NULL)) 1060 return -ENOBUFS; 1061 } else if (np->cork.opt->tot_len < opt->tot_len) { 1062 printk(KERN_DEBUG "ip6_append_data: invalid option length\n"); 1063 return -EINVAL; 1064 } 1065 memcpy(np->cork.opt, opt, opt->tot_len); 1066 inet->cork.flags |= IPCORK_OPT; 1067 /* need source address above miyazawa*/ 1068 } 1069 dst_hold(&rt->u.dst); 1070 np->cork.rt = rt; 1071 inet->cork.fl = *fl; 1072 np->cork.hop_limit = hlimit; 1073 np->cork.tclass = tclass; 1074 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ? 1075 rt->u.dst.dev->mtu : dst_mtu(rt->u.dst.path); 1076 if (np->frag_size < mtu) { 1077 if (np->frag_size) 1078 mtu = np->frag_size; 1079 } 1080 inet->cork.fragsize = mtu; 1081 if (dst_allfrag(rt->u.dst.path)) 1082 inet->cork.flags |= IPCORK_ALLFRAG; 1083 inet->cork.length = 0; 1084 sk->sk_sndmsg_page = NULL; 1085 sk->sk_sndmsg_off = 0; 1086 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0); 1087 length += exthdrlen; 1088 transhdrlen += exthdrlen; 1089 } else { 1090 rt = np->cork.rt; 1091 fl = &inet->cork.fl; 1092 if (inet->cork.flags & IPCORK_OPT) 1093 opt = np->cork.opt; 1094 transhdrlen = 0; 1095 exthdrlen = 0; 1096 mtu = inet->cork.fragsize; 1097 } 1098 1099 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); 1100 1101 fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0); 1102 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr); 1103 1104 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) { 1105 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) { 1106 ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen); 1107 return -EMSGSIZE; 1108 } 1109 } 1110 1111 /* 1112 * Let's try using as much space as possible. 1113 * Use MTU if total length of the message fits into the MTU. 1114 * Otherwise, we need to reserve fragment header and 1115 * fragment alignment (= 8-15 octects, in total). 1116 * 1117 * Note that we may need to "move" the data from the tail of 1118 * of the buffer to the new fragment when we split 1119 * the message. 1120 * 1121 * FIXME: It may be fragmented into multiple chunks 1122 * at once if non-fragmentable extension headers 1123 * are too large. 1124 * --yoshfuji 1125 */ 1126 1127 inet->cork.length += length; 1128 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) && 1129 (rt->u.dst.dev->features & NETIF_F_UFO)) { 1130 1131 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len, 1132 fragheaderlen, transhdrlen, mtu, 1133 flags); 1134 if (err) 1135 goto error; 1136 return 0; 1137 } 1138 1139 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) 1140 goto alloc_new_skb; 1141 1142 while (length > 0) { 1143 /* Check if the remaining data fits into current packet. */ 1144 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len; 1145 if (copy < length) 1146 copy = maxfraglen - skb->len; 1147 1148 if (copy <= 0) { 1149 char *data; 1150 unsigned int datalen; 1151 unsigned int fraglen; 1152 unsigned int fraggap; 1153 unsigned int alloclen; 1154 struct sk_buff *skb_prev; 1155 alloc_new_skb: 1156 skb_prev = skb; 1157 1158 /* There's no room in the current skb */ 1159 if (skb_prev) 1160 fraggap = skb_prev->len - maxfraglen; 1161 else 1162 fraggap = 0; 1163 1164 /* 1165 * If remaining data exceeds the mtu, 1166 * we know we need more fragment(s). 1167 */ 1168 datalen = length + fraggap; 1169 if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen) 1170 datalen = maxfraglen - fragheaderlen; 1171 1172 fraglen = datalen + fragheaderlen; 1173 if ((flags & MSG_MORE) && 1174 !(rt->u.dst.dev->features&NETIF_F_SG)) 1175 alloclen = mtu; 1176 else 1177 alloclen = datalen + fragheaderlen; 1178 1179 /* 1180 * The last fragment gets additional space at tail. 1181 * Note: we overallocate on fragments with MSG_MODE 1182 * because we have no idea if we're the last one. 1183 */ 1184 if (datalen == length + fraggap) 1185 alloclen += rt->u.dst.trailer_len; 1186 1187 /* 1188 * We just reserve space for fragment header. 1189 * Note: this may be overallocation if the message 1190 * (without MSG_MORE) fits into the MTU. 1191 */ 1192 alloclen += sizeof(struct frag_hdr); 1193 1194 if (transhdrlen) { 1195 skb = sock_alloc_send_skb(sk, 1196 alloclen + hh_len, 1197 (flags & MSG_DONTWAIT), &err); 1198 } else { 1199 skb = NULL; 1200 if (atomic_read(&sk->sk_wmem_alloc) <= 1201 2 * sk->sk_sndbuf) 1202 skb = sock_wmalloc(sk, 1203 alloclen + hh_len, 1, 1204 sk->sk_allocation); 1205 if (unlikely(skb == NULL)) 1206 err = -ENOBUFS; 1207 } 1208 if (skb == NULL) 1209 goto error; 1210 /* 1211 * Fill in the control structures 1212 */ 1213 skb->ip_summed = csummode; 1214 skb->csum = 0; 1215 /* reserve for fragmentation */ 1216 skb_reserve(skb, hh_len+sizeof(struct frag_hdr)); 1217 1218 /* 1219 * Find where to start putting bytes 1220 */ 1221 data = skb_put(skb, fraglen); 1222 skb_set_network_header(skb, exthdrlen); 1223 data += fragheaderlen; 1224 skb->transport_header = (skb->network_header + 1225 fragheaderlen); 1226 if (fraggap) { 1227 skb->csum = skb_copy_and_csum_bits( 1228 skb_prev, maxfraglen, 1229 data + transhdrlen, fraggap, 0); 1230 skb_prev->csum = csum_sub(skb_prev->csum, 1231 skb->csum); 1232 data += fraggap; 1233 pskb_trim_unique(skb_prev, maxfraglen); 1234 } 1235 copy = datalen - transhdrlen - fraggap; 1236 if (copy < 0) { 1237 err = -EINVAL; 1238 kfree_skb(skb); 1239 goto error; 1240 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { 1241 err = -EFAULT; 1242 kfree_skb(skb); 1243 goto error; 1244 } 1245 1246 offset += copy; 1247 length -= datalen - fraggap; 1248 transhdrlen = 0; 1249 exthdrlen = 0; 1250 csummode = CHECKSUM_NONE; 1251 1252 /* 1253 * Put the packet on the pending queue 1254 */ 1255 __skb_queue_tail(&sk->sk_write_queue, skb); 1256 continue; 1257 } 1258 1259 if (copy > length) 1260 copy = length; 1261 1262 if (!(rt->u.dst.dev->features&NETIF_F_SG)) { 1263 unsigned int off; 1264 1265 off = skb->len; 1266 if (getfrag(from, skb_put(skb, copy), 1267 offset, copy, off, skb) < 0) { 1268 __skb_trim(skb, off); 1269 err = -EFAULT; 1270 goto error; 1271 } 1272 } else { 1273 int i = skb_shinfo(skb)->nr_frags; 1274 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1]; 1275 struct page *page = sk->sk_sndmsg_page; 1276 int off = sk->sk_sndmsg_off; 1277 unsigned int left; 1278 1279 if (page && (left = PAGE_SIZE - off) > 0) { 1280 if (copy >= left) 1281 copy = left; 1282 if (page != frag->page) { 1283 if (i == MAX_SKB_FRAGS) { 1284 err = -EMSGSIZE; 1285 goto error; 1286 } 1287 get_page(page); 1288 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0); 1289 frag = &skb_shinfo(skb)->frags[i]; 1290 } 1291 } else if(i < MAX_SKB_FRAGS) { 1292 if (copy > PAGE_SIZE) 1293 copy = PAGE_SIZE; 1294 page = alloc_pages(sk->sk_allocation, 0); 1295 if (page == NULL) { 1296 err = -ENOMEM; 1297 goto error; 1298 } 1299 sk->sk_sndmsg_page = page; 1300 sk->sk_sndmsg_off = 0; 1301 1302 skb_fill_page_desc(skb, i, page, 0, 0); 1303 frag = &skb_shinfo(skb)->frags[i]; 1304 skb->truesize += PAGE_SIZE; 1305 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); 1306 } else { 1307 err = -EMSGSIZE; 1308 goto error; 1309 } 1310 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) { 1311 err = -EFAULT; 1312 goto error; 1313 } 1314 sk->sk_sndmsg_off += copy; 1315 frag->size += copy; 1316 skb->len += copy; 1317 skb->data_len += copy; 1318 } 1319 offset += copy; 1320 length -= copy; 1321 } 1322 return 0; 1323 error: 1324 inet->cork.length -= length; 1325 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); 1326 return err; 1327 } 1328 1329 int ip6_push_pending_frames(struct sock *sk) 1330 { 1331 struct sk_buff *skb, *tmp_skb; 1332 struct sk_buff **tail_skb; 1333 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf; 1334 struct inet_sock *inet = inet_sk(sk); 1335 struct ipv6_pinfo *np = inet6_sk(sk); 1336 struct ipv6hdr *hdr; 1337 struct ipv6_txoptions *opt = np->cork.opt; 1338 struct rt6_info *rt = np->cork.rt; 1339 struct flowi *fl = &inet->cork.fl; 1340 unsigned char proto = fl->proto; 1341 int err = 0; 1342 1343 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL) 1344 goto out; 1345 tail_skb = &(skb_shinfo(skb)->frag_list); 1346 1347 /* move skb->data to ip header from ext header */ 1348 if (skb->data < skb_network_header(skb)) 1349 __skb_pull(skb, skb_network_offset(skb)); 1350 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { 1351 __skb_pull(tmp_skb, skb_network_header_len(skb)); 1352 *tail_skb = tmp_skb; 1353 tail_skb = &(tmp_skb->next); 1354 skb->len += tmp_skb->len; 1355 skb->data_len += tmp_skb->len; 1356 skb->truesize += tmp_skb->truesize; 1357 __sock_put(tmp_skb->sk); 1358 tmp_skb->destructor = NULL; 1359 tmp_skb->sk = NULL; 1360 } 1361 1362 ipv6_addr_copy(final_dst, &fl->fl6_dst); 1363 __skb_pull(skb, skb_network_header_len(skb)); 1364 if (opt && opt->opt_flen) 1365 ipv6_push_frag_opts(skb, opt, &proto); 1366 if (opt && opt->opt_nflen) 1367 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst); 1368 1369 skb_push(skb, sizeof(struct ipv6hdr)); 1370 skb_reset_network_header(skb); 1371 hdr = ipv6_hdr(skb); 1372 1373 *(__be32*)hdr = fl->fl6_flowlabel | 1374 htonl(0x60000000 | ((int)np->cork.tclass << 20)); 1375 1376 if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) 1377 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 1378 else 1379 hdr->payload_len = 0; 1380 hdr->hop_limit = np->cork.hop_limit; 1381 hdr->nexthdr = proto; 1382 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src); 1383 ipv6_addr_copy(&hdr->daddr, final_dst); 1384 1385 skb->priority = sk->sk_priority; 1386 1387 skb->dst = dst_clone(&rt->u.dst); 1388 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS); 1389 err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output); 1390 if (err) { 1391 if (err > 0) 1392 err = np->recverr ? net_xmit_errno(err) : 0; 1393 if (err) 1394 goto error; 1395 } 1396 1397 out: 1398 inet->cork.flags &= ~IPCORK_OPT; 1399 kfree(np->cork.opt); 1400 np->cork.opt = NULL; 1401 if (np->cork.rt) { 1402 dst_release(&np->cork.rt->u.dst); 1403 np->cork.rt = NULL; 1404 inet->cork.flags &= ~IPCORK_ALLFRAG; 1405 } 1406 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl)); 1407 return err; 1408 error: 1409 goto out; 1410 } 1411 1412 void ip6_flush_pending_frames(struct sock *sk) 1413 { 1414 struct inet_sock *inet = inet_sk(sk); 1415 struct ipv6_pinfo *np = inet6_sk(sk); 1416 struct sk_buff *skb; 1417 1418 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) { 1419 IP6_INC_STATS(ip6_dst_idev(skb->dst), 1420 IPSTATS_MIB_OUTDISCARDS); 1421 kfree_skb(skb); 1422 } 1423 1424 inet->cork.flags &= ~IPCORK_OPT; 1425 1426 kfree(np->cork.opt); 1427 np->cork.opt = NULL; 1428 if (np->cork.rt) { 1429 dst_release(&np->cork.rt->u.dst); 1430 np->cork.rt = NULL; 1431 inet->cork.flags &= ~IPCORK_ALLFRAG; 1432 } 1433 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl)); 1434 } 1435