1 /* 2 * UDP over IPv6 3 * Linux INET6 implementation 4 * 5 * Authors: 6 * Pedro Roque <roque@di.fc.ul.pt> 7 * 8 * Based on linux/ipv4/udp.c 9 * 10 * Fixes: 11 * Hideaki YOSHIFUJI : sin6_scope_id support 12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 14 * a single port at the same time. 15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. 17 * 18 * This program is free software; you can redistribute it and/or 19 * modify it under the terms of the GNU General Public License 20 * as published by the Free Software Foundation; either version 21 * 2 of the License, or (at your option) any later version. 22 */ 23 24 #include <linux/errno.h> 25 #include <linux/types.h> 26 #include <linux/socket.h> 27 #include <linux/sockios.h> 28 #include <linux/net.h> 29 #include <linux/in6.h> 30 #include <linux/netdevice.h> 31 #include <linux/if_arp.h> 32 #include <linux/ipv6.h> 33 #include <linux/icmpv6.h> 34 #include <linux/init.h> 35 #include <linux/module.h> 36 #include <linux/skbuff.h> 37 #include <asm/uaccess.h> 38 39 #include <net/ndisc.h> 40 #include <net/protocol.h> 41 #include <net/transp_v6.h> 42 #include <net/ip6_route.h> 43 #include <net/raw.h> 44 #include <net/tcp_states.h> 45 #include <net/ip6_checksum.h> 46 #include <net/xfrm.h> 47 48 #include <linux/proc_fs.h> 49 #include <linux/seq_file.h> 50 #include "udp_impl.h" 51 52 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2) 53 { 54 const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr; 55 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2); 56 __be32 sk_rcv_saddr = inet_sk(sk)->rcv_saddr; 57 __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2); 58 int sk_ipv6only = ipv6_only_sock(sk); 59 int sk2_ipv6only = inet_v6_ipv6only(sk2); 60 int addr_type = ipv6_addr_type(sk_rcv_saddr6); 61 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED; 62 63 /* if both are mapped, treat as IPv4 */ 64 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) 65 return (!sk2_ipv6only && 66 (!sk_rcv_saddr || !sk2_rcv_saddr || 67 sk_rcv_saddr == sk2_rcv_saddr)); 68 69 if (addr_type2 == IPV6_ADDR_ANY && 70 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED)) 71 return 1; 72 73 if (addr_type == IPV6_ADDR_ANY && 74 !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED)) 75 return 1; 76 77 if (sk2_rcv_saddr6 && 78 ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6)) 79 return 1; 80 81 return 0; 82 } 83 84 int udp_v6_get_port(struct sock *sk, unsigned short snum) 85 { 86 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal); 87 } 88 89 static inline int compute_score(struct sock *sk, struct net *net, 90 unsigned short hnum, 91 struct in6_addr *saddr, __be16 sport, 92 struct in6_addr *daddr, __be16 dport, 93 int dif) 94 { 95 int score = -1; 96 97 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum && 98 sk->sk_family == PF_INET6) { 99 struct ipv6_pinfo *np = inet6_sk(sk); 100 struct inet_sock *inet = inet_sk(sk); 101 102 score = 0; 103 if (inet->dport) { 104 if (inet->dport != sport) 105 return -1; 106 score++; 107 } 108 if (!ipv6_addr_any(&np->rcv_saddr)) { 109 if (!ipv6_addr_equal(&np->rcv_saddr, daddr)) 110 return -1; 111 score++; 112 } 113 if (!ipv6_addr_any(&np->daddr)) { 114 if (!ipv6_addr_equal(&np->daddr, saddr)) 115 return -1; 116 score++; 117 } 118 if (sk->sk_bound_dev_if) { 119 if (sk->sk_bound_dev_if != dif) 120 return -1; 121 score++; 122 } 123 } 124 return score; 125 } 126 127 static struct sock *__udp6_lib_lookup(struct net *net, 128 struct in6_addr *saddr, __be16 sport, 129 struct in6_addr *daddr, __be16 dport, 130 int dif, struct udp_table *udptable) 131 { 132 struct sock *sk, *result; 133 struct hlist_nulls_node *node; 134 unsigned short hnum = ntohs(dport); 135 unsigned int hash = udp_hashfn(net, hnum); 136 struct udp_hslot *hslot = &udptable->hash[hash]; 137 int score, badness; 138 139 rcu_read_lock(); 140 begin: 141 result = NULL; 142 badness = -1; 143 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 144 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif); 145 if (score > badness) { 146 result = sk; 147 badness = score; 148 } 149 } 150 /* 151 * if the nulls value we got at the end of this lookup is 152 * not the expected one, we must restart lookup. 153 * We probably met an item that was moved to another chain. 154 */ 155 if (get_nulls_value(node) != hash) 156 goto begin; 157 158 if (result) { 159 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) 160 result = NULL; 161 else if (unlikely(compute_score(result, net, hnum, saddr, sport, 162 daddr, dport, dif) < badness)) { 163 sock_put(result); 164 goto begin; 165 } 166 } 167 rcu_read_unlock(); 168 return result; 169 } 170 171 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 172 __be16 sport, __be16 dport, 173 struct udp_table *udptable) 174 { 175 struct sock *sk; 176 struct ipv6hdr *iph = ipv6_hdr(skb); 177 178 if (unlikely(sk = skb_steal_sock(skb))) 179 return sk; 180 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport, 181 &iph->daddr, dport, inet6_iif(skb), 182 udptable); 183 } 184 185 /* 186 * This should be easy, if there is something there we 187 * return it, otherwise we block. 188 */ 189 190 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk, 191 struct msghdr *msg, size_t len, 192 int noblock, int flags, int *addr_len) 193 { 194 struct ipv6_pinfo *np = inet6_sk(sk); 195 struct inet_sock *inet = inet_sk(sk); 196 struct sk_buff *skb; 197 unsigned int ulen, copied; 198 int peeked; 199 int err; 200 int is_udplite = IS_UDPLITE(sk); 201 int is_udp4; 202 203 if (addr_len) 204 *addr_len=sizeof(struct sockaddr_in6); 205 206 if (flags & MSG_ERRQUEUE) 207 return ipv6_recv_error(sk, msg, len); 208 209 try_again: 210 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 211 &peeked, &err); 212 if (!skb) 213 goto out; 214 215 ulen = skb->len - sizeof(struct udphdr); 216 copied = len; 217 if (copied > ulen) 218 copied = ulen; 219 else if (copied < ulen) 220 msg->msg_flags |= MSG_TRUNC; 221 222 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 223 224 /* 225 * If checksum is needed at all, try to do it while copying the 226 * data. If the data is truncated, or if we only want a partial 227 * coverage checksum (UDP-Lite), do it before the copy. 228 */ 229 230 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 231 if (udp_lib_checksum_complete(skb)) 232 goto csum_copy_err; 233 } 234 235 if (skb_csum_unnecessary(skb)) 236 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 237 msg->msg_iov, copied ); 238 else { 239 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 240 if (err == -EINVAL) 241 goto csum_copy_err; 242 } 243 if (err) 244 goto out_free; 245 246 if (!peeked) { 247 if (is_udp4) 248 UDP_INC_STATS_USER(sock_net(sk), 249 UDP_MIB_INDATAGRAMS, is_udplite); 250 else 251 UDP6_INC_STATS_USER(sock_net(sk), 252 UDP_MIB_INDATAGRAMS, is_udplite); 253 } 254 255 sock_recv_timestamp(msg, sk, skb); 256 257 /* Copy the address. */ 258 if (msg->msg_name) { 259 struct sockaddr_in6 *sin6; 260 261 sin6 = (struct sockaddr_in6 *) msg->msg_name; 262 sin6->sin6_family = AF_INET6; 263 sin6->sin6_port = udp_hdr(skb)->source; 264 sin6->sin6_flowinfo = 0; 265 sin6->sin6_scope_id = 0; 266 267 if (is_udp4) 268 ipv6_addr_set(&sin6->sin6_addr, 0, 0, 269 htonl(0xffff), ip_hdr(skb)->saddr); 270 else { 271 ipv6_addr_copy(&sin6->sin6_addr, 272 &ipv6_hdr(skb)->saddr); 273 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL) 274 sin6->sin6_scope_id = IP6CB(skb)->iif; 275 } 276 277 } 278 if (is_udp4) { 279 if (inet->cmsg_flags) 280 ip_cmsg_recv(msg, skb); 281 } else { 282 if (np->rxopt.all) 283 datagram_recv_ctl(sk, msg, skb); 284 } 285 286 err = copied; 287 if (flags & MSG_TRUNC) 288 err = ulen; 289 290 out_free: 291 skb_free_datagram_locked(sk, skb); 292 out: 293 return err; 294 295 csum_copy_err: 296 lock_sock(sk); 297 if (!skb_kill_datagram(sk, skb, flags)) { 298 if (is_udp4) 299 UDP_INC_STATS_USER(sock_net(sk), 300 UDP_MIB_INERRORS, is_udplite); 301 else 302 UDP6_INC_STATS_USER(sock_net(sk), 303 UDP_MIB_INERRORS, is_udplite); 304 } 305 release_sock(sk); 306 307 if (flags & MSG_DONTWAIT) 308 return -EAGAIN; 309 goto try_again; 310 } 311 312 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 313 u8 type, u8 code, int offset, __be32 info, 314 struct udp_table *udptable) 315 { 316 struct ipv6_pinfo *np; 317 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data; 318 struct in6_addr *saddr = &hdr->saddr; 319 struct in6_addr *daddr = &hdr->daddr; 320 struct udphdr *uh = (struct udphdr*)(skb->data+offset); 321 struct sock *sk; 322 int err; 323 324 sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest, 325 saddr, uh->source, inet6_iif(skb), udptable); 326 if (sk == NULL) 327 return; 328 329 np = inet6_sk(sk); 330 331 if (!icmpv6_err_convert(type, code, &err) && !np->recverr) 332 goto out; 333 334 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr) 335 goto out; 336 337 if (np->recverr) 338 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 339 340 sk->sk_err = err; 341 sk->sk_error_report(sk); 342 out: 343 sock_put(sk); 344 } 345 346 static __inline__ void udpv6_err(struct sk_buff *skb, 347 struct inet6_skb_parm *opt, u8 type, 348 u8 code, int offset, __be32 info ) 349 { 350 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 351 } 352 353 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 354 { 355 struct udp_sock *up = udp_sk(sk); 356 int rc; 357 int is_udplite = IS_UDPLITE(sk); 358 359 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 360 goto drop; 361 362 /* 363 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 364 */ 365 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 366 367 if (up->pcrlen == 0) { /* full coverage was set */ 368 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage" 369 " %d while full coverage %d requested\n", 370 UDP_SKB_CB(skb)->cscov, skb->len); 371 goto drop; 372 } 373 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 374 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d " 375 "too small, need min %d\n", 376 UDP_SKB_CB(skb)->cscov, up->pcrlen); 377 goto drop; 378 } 379 } 380 381 if (sk->sk_filter) { 382 if (udp_lib_checksum_complete(skb)) 383 goto drop; 384 } 385 386 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) { 387 /* Note that an ENOMEM error is charged twice */ 388 if (rc == -ENOMEM) { 389 UDP6_INC_STATS_BH(sock_net(sk), 390 UDP_MIB_RCVBUFERRORS, is_udplite); 391 atomic_inc(&sk->sk_drops); 392 } 393 goto drop; 394 } 395 396 return 0; 397 drop: 398 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 399 kfree_skb(skb); 400 return -1; 401 } 402 403 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk, 404 __be16 loc_port, struct in6_addr *loc_addr, 405 __be16 rmt_port, struct in6_addr *rmt_addr, 406 int dif) 407 { 408 struct hlist_nulls_node *node; 409 struct sock *s = sk; 410 unsigned short num = ntohs(loc_port); 411 412 sk_nulls_for_each_from(s, node) { 413 struct inet_sock *inet = inet_sk(s); 414 415 if (!net_eq(sock_net(s), net)) 416 continue; 417 418 if (s->sk_hash == num && s->sk_family == PF_INET6) { 419 struct ipv6_pinfo *np = inet6_sk(s); 420 if (inet->dport) { 421 if (inet->dport != rmt_port) 422 continue; 423 } 424 if (!ipv6_addr_any(&np->daddr) && 425 !ipv6_addr_equal(&np->daddr, rmt_addr)) 426 continue; 427 428 if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif) 429 continue; 430 431 if (!ipv6_addr_any(&np->rcv_saddr)) { 432 if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr)) 433 continue; 434 } 435 if (!inet6_mc_check(s, loc_addr, rmt_addr)) 436 continue; 437 return s; 438 } 439 } 440 return NULL; 441 } 442 443 /* 444 * Note: called only from the BH handler context, 445 * so we don't need to lock the hashes. 446 */ 447 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 448 struct in6_addr *saddr, struct in6_addr *daddr, 449 struct udp_table *udptable) 450 { 451 struct sock *sk, *sk2; 452 const struct udphdr *uh = udp_hdr(skb); 453 struct udp_hslot *hslot = &udptable->hash[udp_hashfn(net, ntohs(uh->dest))]; 454 int dif; 455 456 spin_lock(&hslot->lock); 457 sk = sk_nulls_head(&hslot->head); 458 dif = inet6_iif(skb); 459 sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 460 if (!sk) { 461 kfree_skb(skb); 462 goto out; 463 } 464 465 sk2 = sk; 466 while ((sk2 = udp_v6_mcast_next(net, sk_nulls_next(sk2), uh->dest, daddr, 467 uh->source, saddr, dif))) { 468 struct sk_buff *buff = skb_clone(skb, GFP_ATOMIC); 469 if (buff) { 470 bh_lock_sock(sk2); 471 if (!sock_owned_by_user(sk2)) 472 udpv6_queue_rcv_skb(sk2, buff); 473 else 474 sk_add_backlog(sk2, buff); 475 bh_unlock_sock(sk2); 476 } 477 } 478 bh_lock_sock(sk); 479 if (!sock_owned_by_user(sk)) 480 udpv6_queue_rcv_skb(sk, skb); 481 else 482 sk_add_backlog(sk, skb); 483 bh_unlock_sock(sk); 484 out: 485 spin_unlock(&hslot->lock); 486 return 0; 487 } 488 489 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh, 490 int proto) 491 { 492 int err; 493 494 UDP_SKB_CB(skb)->partial_cov = 0; 495 UDP_SKB_CB(skb)->cscov = skb->len; 496 497 if (proto == IPPROTO_UDPLITE) { 498 err = udplite_checksum_init(skb, uh); 499 if (err) 500 return err; 501 } 502 503 if (uh->check == 0) { 504 /* RFC 2460 section 8.1 says that we SHOULD log 505 this error. Well, it is reasonable. 506 */ 507 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n"); 508 return 1; 509 } 510 if (skb->ip_summed == CHECKSUM_COMPLETE && 511 !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr, 512 skb->len, proto, skb->csum)) 513 skb->ip_summed = CHECKSUM_UNNECESSARY; 514 515 if (!skb_csum_unnecessary(skb)) 516 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 517 &ipv6_hdr(skb)->daddr, 518 skb->len, proto, 0)); 519 520 return 0; 521 } 522 523 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 524 int proto) 525 { 526 struct sock *sk; 527 struct udphdr *uh; 528 struct net_device *dev = skb->dev; 529 struct in6_addr *saddr, *daddr; 530 u32 ulen = 0; 531 struct net *net = dev_net(skb->dev); 532 533 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 534 goto short_packet; 535 536 saddr = &ipv6_hdr(skb)->saddr; 537 daddr = &ipv6_hdr(skb)->daddr; 538 uh = udp_hdr(skb); 539 540 ulen = ntohs(uh->len); 541 if (ulen > skb->len) 542 goto short_packet; 543 544 if (proto == IPPROTO_UDP) { 545 /* UDP validates ulen. */ 546 547 /* Check for jumbo payload */ 548 if (ulen == 0) 549 ulen = skb->len; 550 551 if (ulen < sizeof(*uh)) 552 goto short_packet; 553 554 if (ulen < skb->len) { 555 if (pskb_trim_rcsum(skb, ulen)) 556 goto short_packet; 557 saddr = &ipv6_hdr(skb)->saddr; 558 daddr = &ipv6_hdr(skb)->daddr; 559 uh = udp_hdr(skb); 560 } 561 } 562 563 if (udp6_csum_init(skb, uh, proto)) 564 goto discard; 565 566 /* 567 * Multicast receive code 568 */ 569 if (ipv6_addr_is_multicast(daddr)) 570 return __udp6_lib_mcast_deliver(net, skb, 571 saddr, daddr, udptable); 572 573 /* Unicast */ 574 575 /* 576 * check socket cache ... must talk to Alan about his plans 577 * for sock caches... i'll skip this for now. 578 */ 579 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 580 581 if (sk == NULL) { 582 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 583 goto discard; 584 585 if (udp_lib_checksum_complete(skb)) 586 goto discard; 587 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, 588 proto == IPPROTO_UDPLITE); 589 590 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev); 591 592 kfree_skb(skb); 593 return 0; 594 } 595 596 /* deliver */ 597 598 bh_lock_sock(sk); 599 if (!sock_owned_by_user(sk)) 600 udpv6_queue_rcv_skb(sk, skb); 601 else 602 sk_add_backlog(sk, skb); 603 bh_unlock_sock(sk); 604 sock_put(sk); 605 return 0; 606 607 short_packet: 608 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: %d/%u\n", 609 proto == IPPROTO_UDPLITE ? "-Lite" : "", 610 ulen, skb->len); 611 612 discard: 613 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 614 kfree_skb(skb); 615 return 0; 616 } 617 618 static __inline__ int udpv6_rcv(struct sk_buff *skb) 619 { 620 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 621 } 622 623 /* 624 * Throw away all pending data and cancel the corking. Socket is locked. 625 */ 626 static void udp_v6_flush_pending_frames(struct sock *sk) 627 { 628 struct udp_sock *up = udp_sk(sk); 629 630 if (up->pending == AF_INET) 631 udp_flush_pending_frames(sk); 632 else if (up->pending) { 633 up->len = 0; 634 up->pending = 0; 635 ip6_flush_pending_frames(sk); 636 } 637 } 638 639 /** 640 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 641 * @sk: socket we are sending on 642 * @skb: sk_buff containing the filled-in UDP header 643 * (checksum field must be zeroed out) 644 */ 645 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 646 const struct in6_addr *saddr, 647 const struct in6_addr *daddr, int len) 648 { 649 unsigned int offset; 650 struct udphdr *uh = udp_hdr(skb); 651 __wsum csum = 0; 652 653 if (skb_queue_len(&sk->sk_write_queue) == 1) { 654 /* Only one fragment on the socket. */ 655 skb->csum_start = skb_transport_header(skb) - skb->head; 656 skb->csum_offset = offsetof(struct udphdr, check); 657 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 658 } else { 659 /* 660 * HW-checksum won't work as there are two or more 661 * fragments on the socket so that all csums of sk_buffs 662 * should be together 663 */ 664 offset = skb_transport_offset(skb); 665 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 666 667 skb->ip_summed = CHECKSUM_NONE; 668 669 skb_queue_walk(&sk->sk_write_queue, skb) { 670 csum = csum_add(csum, skb->csum); 671 } 672 673 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 674 csum); 675 if (uh->check == 0) 676 uh->check = CSUM_MANGLED_0; 677 } 678 } 679 680 /* 681 * Sending 682 */ 683 684 static int udp_v6_push_pending_frames(struct sock *sk) 685 { 686 struct sk_buff *skb; 687 struct udphdr *uh; 688 struct udp_sock *up = udp_sk(sk); 689 struct inet_sock *inet = inet_sk(sk); 690 struct flowi *fl = &inet->cork.fl; 691 int err = 0; 692 int is_udplite = IS_UDPLITE(sk); 693 __wsum csum = 0; 694 695 /* Grab the skbuff where UDP header space exists. */ 696 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 697 goto out; 698 699 /* 700 * Create a UDP header 701 */ 702 uh = udp_hdr(skb); 703 uh->source = fl->fl_ip_sport; 704 uh->dest = fl->fl_ip_dport; 705 uh->len = htons(up->len); 706 uh->check = 0; 707 708 if (is_udplite) 709 csum = udplite_csum_outgoing(sk, skb); 710 else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 711 udp6_hwcsum_outgoing(sk, skb, &fl->fl6_src, &fl->fl6_dst, 712 up->len); 713 goto send; 714 } else 715 csum = udp_csum_outgoing(sk, skb); 716 717 /* add protocol-dependent pseudo-header */ 718 uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst, 719 up->len, fl->proto, csum ); 720 if (uh->check == 0) 721 uh->check = CSUM_MANGLED_0; 722 723 send: 724 err = ip6_push_pending_frames(sk); 725 if (err) { 726 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 727 UDP6_INC_STATS_USER(sock_net(sk), 728 UDP_MIB_SNDBUFERRORS, is_udplite); 729 err = 0; 730 } 731 } else 732 UDP6_INC_STATS_USER(sock_net(sk), 733 UDP_MIB_OUTDATAGRAMS, is_udplite); 734 out: 735 up->len = 0; 736 up->pending = 0; 737 return err; 738 } 739 740 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk, 741 struct msghdr *msg, size_t len) 742 { 743 struct ipv6_txoptions opt_space; 744 struct udp_sock *up = udp_sk(sk); 745 struct inet_sock *inet = inet_sk(sk); 746 struct ipv6_pinfo *np = inet6_sk(sk); 747 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name; 748 struct in6_addr *daddr, *final_p = NULL, final; 749 struct ipv6_txoptions *opt = NULL; 750 struct ip6_flowlabel *flowlabel = NULL; 751 struct flowi fl; 752 struct dst_entry *dst; 753 int addr_len = msg->msg_namelen; 754 int ulen = len; 755 int hlimit = -1; 756 int tclass = -1; 757 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 758 int err; 759 int connected = 0; 760 int is_udplite = IS_UDPLITE(sk); 761 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 762 763 /* destination address check */ 764 if (sin6) { 765 if (addr_len < offsetof(struct sockaddr, sa_data)) 766 return -EINVAL; 767 768 switch (sin6->sin6_family) { 769 case AF_INET6: 770 if (addr_len < SIN6_LEN_RFC2133) 771 return -EINVAL; 772 daddr = &sin6->sin6_addr; 773 break; 774 case AF_INET: 775 goto do_udp_sendmsg; 776 case AF_UNSPEC: 777 msg->msg_name = sin6 = NULL; 778 msg->msg_namelen = addr_len = 0; 779 daddr = NULL; 780 break; 781 default: 782 return -EINVAL; 783 } 784 } else if (!up->pending) { 785 if (sk->sk_state != TCP_ESTABLISHED) 786 return -EDESTADDRREQ; 787 daddr = &np->daddr; 788 } else 789 daddr = NULL; 790 791 if (daddr) { 792 if (ipv6_addr_v4mapped(daddr)) { 793 struct sockaddr_in sin; 794 sin.sin_family = AF_INET; 795 sin.sin_port = sin6 ? sin6->sin6_port : inet->dport; 796 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 797 msg->msg_name = &sin; 798 msg->msg_namelen = sizeof(sin); 799 do_udp_sendmsg: 800 if (__ipv6_only_sock(sk)) 801 return -ENETUNREACH; 802 return udp_sendmsg(iocb, sk, msg, len); 803 } 804 } 805 806 if (up->pending == AF_INET) 807 return udp_sendmsg(iocb, sk, msg, len); 808 809 /* Rough check on arithmetic overflow, 810 better check is made in ip6_append_data(). 811 */ 812 if (len > INT_MAX - sizeof(struct udphdr)) 813 return -EMSGSIZE; 814 815 if (up->pending) { 816 /* 817 * There are pending frames. 818 * The socket lock must be held while it's corked. 819 */ 820 lock_sock(sk); 821 if (likely(up->pending)) { 822 if (unlikely(up->pending != AF_INET6)) { 823 release_sock(sk); 824 return -EAFNOSUPPORT; 825 } 826 dst = NULL; 827 goto do_append_data; 828 } 829 release_sock(sk); 830 } 831 ulen += sizeof(struct udphdr); 832 833 memset(&fl, 0, sizeof(fl)); 834 835 if (sin6) { 836 if (sin6->sin6_port == 0) 837 return -EINVAL; 838 839 fl.fl_ip_dport = sin6->sin6_port; 840 daddr = &sin6->sin6_addr; 841 842 if (np->sndflow) { 843 fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 844 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) { 845 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel); 846 if (flowlabel == NULL) 847 return -EINVAL; 848 daddr = &flowlabel->dst; 849 } 850 } 851 852 /* 853 * Otherwise it will be difficult to maintain 854 * sk->sk_dst_cache. 855 */ 856 if (sk->sk_state == TCP_ESTABLISHED && 857 ipv6_addr_equal(daddr, &np->daddr)) 858 daddr = &np->daddr; 859 860 if (addr_len >= sizeof(struct sockaddr_in6) && 861 sin6->sin6_scope_id && 862 ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL) 863 fl.oif = sin6->sin6_scope_id; 864 } else { 865 if (sk->sk_state != TCP_ESTABLISHED) 866 return -EDESTADDRREQ; 867 868 fl.fl_ip_dport = inet->dport; 869 daddr = &np->daddr; 870 fl.fl6_flowlabel = np->flow_label; 871 connected = 1; 872 } 873 874 if (!fl.oif) 875 fl.oif = sk->sk_bound_dev_if; 876 877 if (!fl.oif) 878 fl.oif = np->sticky_pktinfo.ipi6_ifindex; 879 880 if (msg->msg_controllen) { 881 opt = &opt_space; 882 memset(opt, 0, sizeof(struct ipv6_txoptions)); 883 opt->tot_len = sizeof(*opt); 884 885 err = datagram_send_ctl(sock_net(sk), msg, &fl, opt, &hlimit, &tclass); 886 if (err < 0) { 887 fl6_sock_release(flowlabel); 888 return err; 889 } 890 if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 891 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel); 892 if (flowlabel == NULL) 893 return -EINVAL; 894 } 895 if (!(opt->opt_nflen|opt->opt_flen)) 896 opt = NULL; 897 connected = 0; 898 } 899 if (opt == NULL) 900 opt = np->opt; 901 if (flowlabel) 902 opt = fl6_merge_options(&opt_space, flowlabel, opt); 903 opt = ipv6_fixup_options(&opt_space, opt); 904 905 fl.proto = sk->sk_protocol; 906 if (!ipv6_addr_any(daddr)) 907 ipv6_addr_copy(&fl.fl6_dst, daddr); 908 else 909 fl.fl6_dst.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 910 if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr)) 911 ipv6_addr_copy(&fl.fl6_src, &np->saddr); 912 fl.fl_ip_sport = inet->sport; 913 914 /* merge ip6_build_xmit from ip6_output */ 915 if (opt && opt->srcrt) { 916 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt; 917 ipv6_addr_copy(&final, &fl.fl6_dst); 918 ipv6_addr_copy(&fl.fl6_dst, rt0->addr); 919 final_p = &final; 920 connected = 0; 921 } 922 923 if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) { 924 fl.oif = np->mcast_oif; 925 connected = 0; 926 } 927 928 security_sk_classify_flow(sk, &fl); 929 930 err = ip6_sk_dst_lookup(sk, &dst, &fl); 931 if (err) 932 goto out; 933 if (final_p) 934 ipv6_addr_copy(&fl.fl6_dst, final_p); 935 936 err = __xfrm_lookup(sock_net(sk), &dst, &fl, sk, XFRM_LOOKUP_WAIT); 937 if (err < 0) { 938 if (err == -EREMOTE) 939 err = ip6_dst_blackhole(sk, &dst, &fl); 940 if (err < 0) 941 goto out; 942 } 943 944 if (hlimit < 0) { 945 if (ipv6_addr_is_multicast(&fl.fl6_dst)) 946 hlimit = np->mcast_hops; 947 else 948 hlimit = np->hop_limit; 949 if (hlimit < 0) 950 hlimit = ip6_dst_hoplimit(dst); 951 } 952 953 if (tclass < 0) 954 tclass = np->tclass; 955 956 if (msg->msg_flags&MSG_CONFIRM) 957 goto do_confirm; 958 back_from_confirm: 959 960 lock_sock(sk); 961 if (unlikely(up->pending)) { 962 /* The socket is already corked while preparing it. */ 963 /* ... which is an evident application bug. --ANK */ 964 release_sock(sk); 965 966 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 967 err = -EINVAL; 968 goto out; 969 } 970 971 up->pending = AF_INET6; 972 973 do_append_data: 974 up->len += ulen; 975 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 976 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen, 977 sizeof(struct udphdr), hlimit, tclass, opt, &fl, 978 (struct rt6_info*)dst, 979 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 980 if (err) 981 udp_v6_flush_pending_frames(sk); 982 else if (!corkreq) 983 err = udp_v6_push_pending_frames(sk); 984 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 985 up->pending = 0; 986 987 if (dst) { 988 if (connected) { 989 ip6_dst_store(sk, dst, 990 ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ? 991 &np->daddr : NULL, 992 #ifdef CONFIG_IPV6_SUBTREES 993 ipv6_addr_equal(&fl.fl6_src, &np->saddr) ? 994 &np->saddr : 995 #endif 996 NULL); 997 } else { 998 dst_release(dst); 999 } 1000 dst = NULL; 1001 } 1002 1003 if (err > 0) 1004 err = np->recverr ? net_xmit_errno(err) : 0; 1005 release_sock(sk); 1006 out: 1007 dst_release(dst); 1008 fl6_sock_release(flowlabel); 1009 if (!err) 1010 return len; 1011 /* 1012 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1013 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1014 * we don't have a good statistic (IpOutDiscards but it can be too many 1015 * things). We could add another new stat but at least for now that 1016 * seems like overkill. 1017 */ 1018 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1019 UDP6_INC_STATS_USER(sock_net(sk), 1020 UDP_MIB_SNDBUFERRORS, is_udplite); 1021 } 1022 return err; 1023 1024 do_confirm: 1025 dst_confirm(dst); 1026 if (!(msg->msg_flags&MSG_PROBE) || len) 1027 goto back_from_confirm; 1028 err = 0; 1029 goto out; 1030 } 1031 1032 void udpv6_destroy_sock(struct sock *sk) 1033 { 1034 lock_sock(sk); 1035 udp_v6_flush_pending_frames(sk); 1036 release_sock(sk); 1037 1038 inet6_destroy_sock(sk); 1039 } 1040 1041 /* 1042 * Socket option code for UDP 1043 */ 1044 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1045 char __user *optval, unsigned int optlen) 1046 { 1047 if (level == SOL_UDP || level == SOL_UDPLITE) 1048 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1049 udp_v6_push_pending_frames); 1050 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1051 } 1052 1053 #ifdef CONFIG_COMPAT 1054 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1055 char __user *optval, unsigned int optlen) 1056 { 1057 if (level == SOL_UDP || level == SOL_UDPLITE) 1058 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1059 udp_v6_push_pending_frames); 1060 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1061 } 1062 #endif 1063 1064 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1065 char __user *optval, int __user *optlen) 1066 { 1067 if (level == SOL_UDP || level == SOL_UDPLITE) 1068 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1069 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1070 } 1071 1072 #ifdef CONFIG_COMPAT 1073 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1074 char __user *optval, int __user *optlen) 1075 { 1076 if (level == SOL_UDP || level == SOL_UDPLITE) 1077 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1078 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1079 } 1080 #endif 1081 1082 static int udp6_ufo_send_check(struct sk_buff *skb) 1083 { 1084 struct ipv6hdr *ipv6h; 1085 struct udphdr *uh; 1086 1087 if (!pskb_may_pull(skb, sizeof(*uh))) 1088 return -EINVAL; 1089 1090 ipv6h = ipv6_hdr(skb); 1091 uh = udp_hdr(skb); 1092 1093 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len, 1094 IPPROTO_UDP, 0); 1095 skb->csum_start = skb_transport_header(skb) - skb->head; 1096 skb->csum_offset = offsetof(struct udphdr, check); 1097 skb->ip_summed = CHECKSUM_PARTIAL; 1098 return 0; 1099 } 1100 1101 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, int features) 1102 { 1103 struct sk_buff *segs = ERR_PTR(-EINVAL); 1104 unsigned int mss; 1105 unsigned int unfrag_ip6hlen, unfrag_len; 1106 struct frag_hdr *fptr; 1107 u8 *mac_start, *prevhdr; 1108 u8 nexthdr; 1109 u8 frag_hdr_sz = sizeof(struct frag_hdr); 1110 int offset; 1111 __wsum csum; 1112 1113 mss = skb_shinfo(skb)->gso_size; 1114 if (unlikely(skb->len <= mss)) 1115 goto out; 1116 1117 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 1118 /* Packet is from an untrusted source, reset gso_segs. */ 1119 int type = skb_shinfo(skb)->gso_type; 1120 1121 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || 1122 !(type & (SKB_GSO_UDP)))) 1123 goto out; 1124 1125 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 1126 1127 segs = NULL; 1128 goto out; 1129 } 1130 1131 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot 1132 * do checksum of UDP packets sent as multiple IP fragments. 1133 */ 1134 offset = skb->csum_start - skb_headroom(skb); 1135 csum = skb_checksum(skb, offset, skb->len- offset, 0); 1136 offset += skb->csum_offset; 1137 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 1138 skb->ip_summed = CHECKSUM_NONE; 1139 1140 /* Check if there is enough headroom to insert fragment header. */ 1141 if ((skb_headroom(skb) < frag_hdr_sz) && 1142 pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC)) 1143 goto out; 1144 1145 /* Find the unfragmentable header and shift it left by frag_hdr_sz 1146 * bytes to insert fragment header. 1147 */ 1148 unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr); 1149 nexthdr = *prevhdr; 1150 *prevhdr = NEXTHDR_FRAGMENT; 1151 unfrag_len = skb_network_header(skb) - skb_mac_header(skb) + 1152 unfrag_ip6hlen; 1153 mac_start = skb_mac_header(skb); 1154 memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len); 1155 1156 skb->mac_header -= frag_hdr_sz; 1157 skb->network_header -= frag_hdr_sz; 1158 1159 fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen); 1160 fptr->nexthdr = nexthdr; 1161 fptr->reserved = 0; 1162 ipv6_select_ident(fptr); 1163 1164 /* Fragment the skb. ipv6 header and the remaining fields of the 1165 * fragment header are updated in ipv6_gso_segment() 1166 */ 1167 segs = skb_segment(skb, features); 1168 1169 out: 1170 return segs; 1171 } 1172 1173 static const struct inet6_protocol udpv6_protocol = { 1174 .handler = udpv6_rcv, 1175 .err_handler = udpv6_err, 1176 .gso_send_check = udp6_ufo_send_check, 1177 .gso_segment = udp6_ufo_fragment, 1178 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1179 }; 1180 1181 /* ------------------------------------------------------------------------ */ 1182 #ifdef CONFIG_PROC_FS 1183 1184 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket) 1185 { 1186 struct inet_sock *inet = inet_sk(sp); 1187 struct ipv6_pinfo *np = inet6_sk(sp); 1188 struct in6_addr *dest, *src; 1189 __u16 destp, srcp; 1190 1191 dest = &np->daddr; 1192 src = &np->rcv_saddr; 1193 destp = ntohs(inet->dport); 1194 srcp = ntohs(inet->sport); 1195 seq_printf(seq, 1196 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " 1197 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n", 1198 bucket, 1199 src->s6_addr32[0], src->s6_addr32[1], 1200 src->s6_addr32[2], src->s6_addr32[3], srcp, 1201 dest->s6_addr32[0], dest->s6_addr32[1], 1202 dest->s6_addr32[2], dest->s6_addr32[3], destp, 1203 sp->sk_state, 1204 sk_wmem_alloc_get(sp), 1205 sk_rmem_alloc_get(sp), 1206 0, 0L, 0, 1207 sock_i_uid(sp), 0, 1208 sock_i_ino(sp), 1209 atomic_read(&sp->sk_refcnt), sp, 1210 atomic_read(&sp->sk_drops)); 1211 } 1212 1213 int udp6_seq_show(struct seq_file *seq, void *v) 1214 { 1215 if (v == SEQ_START_TOKEN) 1216 seq_printf(seq, 1217 " sl " 1218 "local_address " 1219 "remote_address " 1220 "st tx_queue rx_queue tr tm->when retrnsmt" 1221 " uid timeout inode ref pointer drops\n"); 1222 else 1223 udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket); 1224 return 0; 1225 } 1226 1227 static struct udp_seq_afinfo udp6_seq_afinfo = { 1228 .name = "udp6", 1229 .family = AF_INET6, 1230 .udp_table = &udp_table, 1231 .seq_fops = { 1232 .owner = THIS_MODULE, 1233 }, 1234 .seq_ops = { 1235 .show = udp6_seq_show, 1236 }, 1237 }; 1238 1239 int udp6_proc_init(struct net *net) 1240 { 1241 return udp_proc_register(net, &udp6_seq_afinfo); 1242 } 1243 1244 void udp6_proc_exit(struct net *net) { 1245 udp_proc_unregister(net, &udp6_seq_afinfo); 1246 } 1247 #endif /* CONFIG_PROC_FS */ 1248 1249 /* ------------------------------------------------------------------------ */ 1250 1251 struct proto udpv6_prot = { 1252 .name = "UDPv6", 1253 .owner = THIS_MODULE, 1254 .close = udp_lib_close, 1255 .connect = ip6_datagram_connect, 1256 .disconnect = udp_disconnect, 1257 .ioctl = udp_ioctl, 1258 .destroy = udpv6_destroy_sock, 1259 .setsockopt = udpv6_setsockopt, 1260 .getsockopt = udpv6_getsockopt, 1261 .sendmsg = udpv6_sendmsg, 1262 .recvmsg = udpv6_recvmsg, 1263 .backlog_rcv = udpv6_queue_rcv_skb, 1264 .hash = udp_lib_hash, 1265 .unhash = udp_lib_unhash, 1266 .get_port = udp_v6_get_port, 1267 .memory_allocated = &udp_memory_allocated, 1268 .sysctl_mem = sysctl_udp_mem, 1269 .sysctl_wmem = &sysctl_udp_wmem_min, 1270 .sysctl_rmem = &sysctl_udp_rmem_min, 1271 .obj_size = sizeof(struct udp6_sock), 1272 .slab_flags = SLAB_DESTROY_BY_RCU, 1273 .h.udp_table = &udp_table, 1274 #ifdef CONFIG_COMPAT 1275 .compat_setsockopt = compat_udpv6_setsockopt, 1276 .compat_getsockopt = compat_udpv6_getsockopt, 1277 #endif 1278 }; 1279 1280 static struct inet_protosw udpv6_protosw = { 1281 .type = SOCK_DGRAM, 1282 .protocol = IPPROTO_UDP, 1283 .prot = &udpv6_prot, 1284 .ops = &inet6_dgram_ops, 1285 .capability =-1, 1286 .no_check = UDP_CSUM_DEFAULT, 1287 .flags = INET_PROTOSW_PERMANENT, 1288 }; 1289 1290 1291 int __init udpv6_init(void) 1292 { 1293 int ret; 1294 1295 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1296 if (ret) 1297 goto out; 1298 1299 ret = inet6_register_protosw(&udpv6_protosw); 1300 if (ret) 1301 goto out_udpv6_protocol; 1302 out: 1303 return ret; 1304 1305 out_udpv6_protocol: 1306 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1307 goto out; 1308 } 1309 1310 void udpv6_exit(void) 1311 { 1312 inet6_unregister_protosw(&udpv6_protosw); 1313 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1314 } 1315