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 <linux/slab.h> 38 #include <asm/uaccess.h> 39 40 #include <net/ndisc.h> 41 #include <net/protocol.h> 42 #include <net/transp_v6.h> 43 #include <net/ip6_route.h> 44 #include <net/raw.h> 45 #include <net/tcp_states.h> 46 #include <net/ip6_checksum.h> 47 #include <net/xfrm.h> 48 #include <net/inet6_hashtables.h> 49 #include <net/busy_poll.h> 50 #include <net/sock_reuseport.h> 51 52 #include <linux/proc_fs.h> 53 #include <linux/seq_file.h> 54 #include <trace/events/skb.h> 55 #include "udp_impl.h" 56 57 static u32 udp6_ehashfn(const struct net *net, 58 const struct in6_addr *laddr, 59 const u16 lport, 60 const struct in6_addr *faddr, 61 const __be16 fport) 62 { 63 static u32 udp6_ehash_secret __read_mostly; 64 static u32 udp_ipv6_hash_secret __read_mostly; 65 66 u32 lhash, fhash; 67 68 net_get_random_once(&udp6_ehash_secret, 69 sizeof(udp6_ehash_secret)); 70 net_get_random_once(&udp_ipv6_hash_secret, 71 sizeof(udp_ipv6_hash_secret)); 72 73 lhash = (__force u32)laddr->s6_addr32[3]; 74 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 75 76 return __inet6_ehashfn(lhash, lport, fhash, fport, 77 udp_ipv6_hash_secret + net_hash_mix(net)); 78 } 79 80 /* match_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6 81 * only, and any IPv4 addresses if not IPv6 only 82 * match_wildcard == false: addresses must be exactly the same, i.e. 83 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY, 84 * and 0.0.0.0 equals to 0.0.0.0 only 85 */ 86 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2, 87 bool match_wildcard) 88 { 89 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2); 90 int sk2_ipv6only = inet_v6_ipv6only(sk2); 91 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr); 92 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED; 93 94 /* if both are mapped, treat as IPv4 */ 95 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) { 96 if (!sk2_ipv6only) { 97 if (sk->sk_rcv_saddr == sk2->sk_rcv_saddr) 98 return 1; 99 if (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr) 100 return match_wildcard; 101 } 102 return 0; 103 } 104 105 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY) 106 return 1; 107 108 if (addr_type2 == IPV6_ADDR_ANY && match_wildcard && 109 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED)) 110 return 1; 111 112 if (addr_type == IPV6_ADDR_ANY && match_wildcard && 113 !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED)) 114 return 1; 115 116 if (sk2_rcv_saddr6 && 117 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6)) 118 return 1; 119 120 return 0; 121 } 122 123 static u32 udp6_portaddr_hash(const struct net *net, 124 const struct in6_addr *addr6, 125 unsigned int port) 126 { 127 unsigned int hash, mix = net_hash_mix(net); 128 129 if (ipv6_addr_any(addr6)) 130 hash = jhash_1word(0, mix); 131 else if (ipv6_addr_v4mapped(addr6)) 132 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix); 133 else 134 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix); 135 136 return hash ^ port; 137 } 138 139 int udp_v6_get_port(struct sock *sk, unsigned short snum) 140 { 141 unsigned int hash2_nulladdr = 142 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 143 unsigned int hash2_partial = 144 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 145 146 /* precompute partial secondary hash */ 147 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 148 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr); 149 } 150 151 static void udp_v6_rehash(struct sock *sk) 152 { 153 u16 new_hash = udp6_portaddr_hash(sock_net(sk), 154 &sk->sk_v6_rcv_saddr, 155 inet_sk(sk)->inet_num); 156 157 udp_lib_rehash(sk, new_hash); 158 } 159 160 static inline int compute_score(struct sock *sk, struct net *net, 161 unsigned short hnum, 162 const struct in6_addr *saddr, __be16 sport, 163 const struct in6_addr *daddr, __be16 dport, 164 int dif) 165 { 166 int score; 167 struct inet_sock *inet; 168 169 if (!net_eq(sock_net(sk), net) || 170 udp_sk(sk)->udp_port_hash != hnum || 171 sk->sk_family != PF_INET6) 172 return -1; 173 174 score = 0; 175 inet = inet_sk(sk); 176 177 if (inet->inet_dport) { 178 if (inet->inet_dport != sport) 179 return -1; 180 score++; 181 } 182 183 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { 184 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 185 return -1; 186 score++; 187 } 188 189 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 190 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 191 return -1; 192 score++; 193 } 194 195 if (sk->sk_bound_dev_if) { 196 if (sk->sk_bound_dev_if != dif) 197 return -1; 198 score++; 199 } 200 201 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 202 score++; 203 204 return score; 205 } 206 207 static inline int compute_score2(struct sock *sk, struct net *net, 208 const struct in6_addr *saddr, __be16 sport, 209 const struct in6_addr *daddr, 210 unsigned short hnum, int dif) 211 { 212 int score; 213 struct inet_sock *inet; 214 215 if (!net_eq(sock_net(sk), net) || 216 udp_sk(sk)->udp_port_hash != hnum || 217 sk->sk_family != PF_INET6) 218 return -1; 219 220 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 221 return -1; 222 223 score = 0; 224 inet = inet_sk(sk); 225 226 if (inet->inet_dport) { 227 if (inet->inet_dport != sport) 228 return -1; 229 score++; 230 } 231 232 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 233 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 234 return -1; 235 score++; 236 } 237 238 if (sk->sk_bound_dev_if) { 239 if (sk->sk_bound_dev_if != dif) 240 return -1; 241 score++; 242 } 243 244 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 245 score++; 246 247 return score; 248 } 249 250 /* called with read_rcu_lock() */ 251 static struct sock *udp6_lib_lookup2(struct net *net, 252 const struct in6_addr *saddr, __be16 sport, 253 const struct in6_addr *daddr, unsigned int hnum, int dif, 254 struct udp_hslot *hslot2, unsigned int slot2, 255 struct sk_buff *skb) 256 { 257 struct sock *sk, *result; 258 struct hlist_nulls_node *node; 259 int score, badness, matches = 0, reuseport = 0; 260 u32 hash = 0; 261 262 begin: 263 result = NULL; 264 badness = -1; 265 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 266 score = compute_score2(sk, net, saddr, sport, 267 daddr, hnum, dif); 268 if (score > badness) { 269 result = sk; 270 badness = score; 271 reuseport = sk->sk_reuseport; 272 if (reuseport) { 273 struct sock *sk2; 274 hash = udp6_ehashfn(net, daddr, hnum, 275 saddr, sport); 276 sk2 = reuseport_select_sock(sk, hash, skb, 277 sizeof(struct udphdr)); 278 if (sk2) { 279 result = sk2; 280 goto found; 281 } 282 matches = 1; 283 } 284 } else if (score == badness && reuseport) { 285 matches++; 286 if (reciprocal_scale(hash, matches) == 0) 287 result = sk; 288 hash = next_pseudo_random32(hash); 289 } 290 } 291 /* 292 * if the nulls value we got at the end of this lookup is 293 * not the expected one, we must restart lookup. 294 * We probably met an item that was moved to another chain. 295 */ 296 if (get_nulls_value(node) != slot2) 297 goto begin; 298 299 if (result) { 300 found: 301 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 302 result = NULL; 303 else if (unlikely(compute_score2(result, net, saddr, sport, 304 daddr, hnum, dif) < badness)) { 305 sock_put(result); 306 goto begin; 307 } 308 } 309 return result; 310 } 311 312 struct sock *__udp6_lib_lookup(struct net *net, 313 const struct in6_addr *saddr, __be16 sport, 314 const struct in6_addr *daddr, __be16 dport, 315 int dif, struct udp_table *udptable, 316 struct sk_buff *skb) 317 { 318 struct sock *sk, *result; 319 struct hlist_nulls_node *node; 320 unsigned short hnum = ntohs(dport); 321 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 322 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 323 int score, badness, matches = 0, reuseport = 0; 324 u32 hash = 0; 325 326 rcu_read_lock(); 327 if (hslot->count > 10) { 328 hash2 = udp6_portaddr_hash(net, daddr, hnum); 329 slot2 = hash2 & udptable->mask; 330 hslot2 = &udptable->hash2[slot2]; 331 if (hslot->count < hslot2->count) 332 goto begin; 333 334 result = udp6_lib_lookup2(net, saddr, sport, 335 daddr, hnum, dif, 336 hslot2, slot2, skb); 337 if (!result) { 338 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum); 339 slot2 = hash2 & udptable->mask; 340 hslot2 = &udptable->hash2[slot2]; 341 if (hslot->count < hslot2->count) 342 goto begin; 343 344 result = udp6_lib_lookup2(net, saddr, sport, 345 &in6addr_any, hnum, dif, 346 hslot2, slot2, skb); 347 } 348 rcu_read_unlock(); 349 return result; 350 } 351 begin: 352 result = NULL; 353 badness = -1; 354 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 355 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif); 356 if (score > badness) { 357 result = sk; 358 badness = score; 359 reuseport = sk->sk_reuseport; 360 if (reuseport) { 361 struct sock *sk2; 362 hash = udp6_ehashfn(net, daddr, hnum, 363 saddr, sport); 364 sk2 = reuseport_select_sock(sk, hash, skb, 365 sizeof(struct udphdr)); 366 if (sk2) { 367 result = sk2; 368 goto found; 369 } 370 matches = 1; 371 } 372 } else if (score == badness && reuseport) { 373 matches++; 374 if (reciprocal_scale(hash, matches) == 0) 375 result = sk; 376 hash = next_pseudo_random32(hash); 377 } 378 } 379 /* 380 * if the nulls value we got at the end of this lookup is 381 * not the expected one, we must restart lookup. 382 * We probably met an item that was moved to another chain. 383 */ 384 if (get_nulls_value(node) != slot) 385 goto begin; 386 387 if (result) { 388 found: 389 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 390 result = NULL; 391 else if (unlikely(compute_score(result, net, hnum, saddr, sport, 392 daddr, dport, dif) < badness)) { 393 sock_put(result); 394 goto begin; 395 } 396 } 397 rcu_read_unlock(); 398 return result; 399 } 400 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 401 402 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 403 __be16 sport, __be16 dport, 404 struct udp_table *udptable) 405 { 406 struct sock *sk; 407 const struct ipv6hdr *iph = ipv6_hdr(skb); 408 409 sk = skb_steal_sock(skb); 410 if (unlikely(sk)) 411 return sk; 412 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport, 413 &iph->daddr, dport, inet6_iif(skb), 414 udptable, skb); 415 } 416 417 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 418 const struct in6_addr *daddr, __be16 dport, int dif) 419 { 420 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table, NULL); 421 } 422 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 423 424 /* 425 * This should be easy, if there is something there we 426 * return it, otherwise we block. 427 */ 428 429 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 430 int noblock, int flags, int *addr_len) 431 { 432 struct ipv6_pinfo *np = inet6_sk(sk); 433 struct inet_sock *inet = inet_sk(sk); 434 struct sk_buff *skb; 435 unsigned int ulen, copied; 436 int peeked, off = 0; 437 int err; 438 int is_udplite = IS_UDPLITE(sk); 439 bool checksum_valid = false; 440 int is_udp4; 441 bool slow; 442 443 if (flags & MSG_ERRQUEUE) 444 return ipv6_recv_error(sk, msg, len, addr_len); 445 446 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 447 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 448 449 try_again: 450 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 451 &peeked, &off, &err); 452 if (!skb) 453 goto out; 454 455 ulen = skb->len - sizeof(struct udphdr); 456 copied = len; 457 if (copied > ulen) 458 copied = ulen; 459 else if (copied < ulen) 460 msg->msg_flags |= MSG_TRUNC; 461 462 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 463 464 /* 465 * If checksum is needed at all, try to do it while copying the 466 * data. If the data is truncated, or if we only want a partial 467 * coverage checksum (UDP-Lite), do it before the copy. 468 */ 469 470 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 471 checksum_valid = !udp_lib_checksum_complete(skb); 472 if (!checksum_valid) 473 goto csum_copy_err; 474 } 475 476 if (checksum_valid || skb_csum_unnecessary(skb)) 477 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 478 msg, copied); 479 else { 480 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), msg); 481 if (err == -EINVAL) 482 goto csum_copy_err; 483 } 484 if (unlikely(err)) { 485 trace_kfree_skb(skb, udpv6_recvmsg); 486 if (!peeked) { 487 atomic_inc(&sk->sk_drops); 488 if (is_udp4) 489 UDP_INC_STATS_USER(sock_net(sk), 490 UDP_MIB_INERRORS, 491 is_udplite); 492 else 493 UDP6_INC_STATS_USER(sock_net(sk), 494 UDP_MIB_INERRORS, 495 is_udplite); 496 } 497 goto out_free; 498 } 499 if (!peeked) { 500 if (is_udp4) 501 UDP_INC_STATS_USER(sock_net(sk), 502 UDP_MIB_INDATAGRAMS, is_udplite); 503 else 504 UDP6_INC_STATS_USER(sock_net(sk), 505 UDP_MIB_INDATAGRAMS, is_udplite); 506 } 507 508 sock_recv_ts_and_drops(msg, sk, skb); 509 510 /* Copy the address. */ 511 if (msg->msg_name) { 512 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 513 sin6->sin6_family = AF_INET6; 514 sin6->sin6_port = udp_hdr(skb)->source; 515 sin6->sin6_flowinfo = 0; 516 517 if (is_udp4) { 518 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 519 &sin6->sin6_addr); 520 sin6->sin6_scope_id = 0; 521 } else { 522 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 523 sin6->sin6_scope_id = 524 ipv6_iface_scope_id(&sin6->sin6_addr, 525 inet6_iif(skb)); 526 } 527 *addr_len = sizeof(*sin6); 528 } 529 530 if (np->rxopt.all) 531 ip6_datagram_recv_common_ctl(sk, msg, skb); 532 533 if (is_udp4) { 534 if (inet->cmsg_flags) 535 ip_cmsg_recv(msg, skb); 536 } else { 537 if (np->rxopt.all) 538 ip6_datagram_recv_specific_ctl(sk, msg, skb); 539 } 540 541 err = copied; 542 if (flags & MSG_TRUNC) 543 err = ulen; 544 545 out_free: 546 skb_free_datagram_locked(sk, skb); 547 out: 548 return err; 549 550 csum_copy_err: 551 slow = lock_sock_fast(sk); 552 if (!skb_kill_datagram(sk, skb, flags)) { 553 if (is_udp4) { 554 UDP_INC_STATS_USER(sock_net(sk), 555 UDP_MIB_CSUMERRORS, is_udplite); 556 UDP_INC_STATS_USER(sock_net(sk), 557 UDP_MIB_INERRORS, is_udplite); 558 } else { 559 UDP6_INC_STATS_USER(sock_net(sk), 560 UDP_MIB_CSUMERRORS, is_udplite); 561 UDP6_INC_STATS_USER(sock_net(sk), 562 UDP_MIB_INERRORS, is_udplite); 563 } 564 } 565 unlock_sock_fast(sk, slow); 566 567 /* starting over for a new packet, but check if we need to yield */ 568 cond_resched(); 569 msg->msg_flags &= ~MSG_TRUNC; 570 goto try_again; 571 } 572 573 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 574 u8 type, u8 code, int offset, __be32 info, 575 struct udp_table *udptable) 576 { 577 struct ipv6_pinfo *np; 578 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 579 const struct in6_addr *saddr = &hdr->saddr; 580 const struct in6_addr *daddr = &hdr->daddr; 581 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 582 struct sock *sk; 583 int err; 584 struct net *net = dev_net(skb->dev); 585 586 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 587 inet6_iif(skb), udptable, skb); 588 if (!sk) { 589 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), 590 ICMP6_MIB_INERRORS); 591 return; 592 } 593 594 if (type == ICMPV6_PKT_TOOBIG) { 595 if (!ip6_sk_accept_pmtu(sk)) 596 goto out; 597 ip6_sk_update_pmtu(skb, sk, info); 598 } 599 if (type == NDISC_REDIRECT) { 600 ip6_sk_redirect(skb, sk); 601 goto out; 602 } 603 604 np = inet6_sk(sk); 605 606 if (!icmpv6_err_convert(type, code, &err) && !np->recverr) 607 goto out; 608 609 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr) 610 goto out; 611 612 if (np->recverr) 613 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 614 615 sk->sk_err = err; 616 sk->sk_error_report(sk); 617 out: 618 sock_put(sk); 619 } 620 621 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 622 { 623 int rc; 624 625 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 626 sock_rps_save_rxhash(sk, skb); 627 sk_mark_napi_id(sk, skb); 628 sk_incoming_cpu_update(sk); 629 } 630 631 rc = sock_queue_rcv_skb(sk, skb); 632 if (rc < 0) { 633 int is_udplite = IS_UDPLITE(sk); 634 635 /* Note that an ENOMEM error is charged twice */ 636 if (rc == -ENOMEM) 637 UDP6_INC_STATS_BH(sock_net(sk), 638 UDP_MIB_RCVBUFERRORS, is_udplite); 639 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 640 kfree_skb(skb); 641 return -1; 642 } 643 return 0; 644 } 645 646 static __inline__ void udpv6_err(struct sk_buff *skb, 647 struct inet6_skb_parm *opt, u8 type, 648 u8 code, int offset, __be32 info) 649 { 650 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 651 } 652 653 static struct static_key udpv6_encap_needed __read_mostly; 654 void udpv6_encap_enable(void) 655 { 656 if (!static_key_enabled(&udpv6_encap_needed)) 657 static_key_slow_inc(&udpv6_encap_needed); 658 } 659 EXPORT_SYMBOL(udpv6_encap_enable); 660 661 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 662 { 663 struct udp_sock *up = udp_sk(sk); 664 int rc; 665 int is_udplite = IS_UDPLITE(sk); 666 667 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 668 goto drop; 669 670 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 671 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 672 673 /* 674 * This is an encapsulation socket so pass the skb to 675 * the socket's udp_encap_rcv() hook. Otherwise, just 676 * fall through and pass this up the UDP socket. 677 * up->encap_rcv() returns the following value: 678 * =0 if skb was successfully passed to the encap 679 * handler or was discarded by it. 680 * >0 if skb should be passed on to UDP. 681 * <0 if skb should be resubmitted as proto -N 682 */ 683 684 /* if we're overly short, let UDP handle it */ 685 encap_rcv = ACCESS_ONCE(up->encap_rcv); 686 if (skb->len > sizeof(struct udphdr) && encap_rcv) { 687 int ret; 688 689 /* Verify checksum before giving to encap */ 690 if (udp_lib_checksum_complete(skb)) 691 goto csum_error; 692 693 ret = encap_rcv(sk, skb); 694 if (ret <= 0) { 695 UDP_INC_STATS_BH(sock_net(sk), 696 UDP_MIB_INDATAGRAMS, 697 is_udplite); 698 return -ret; 699 } 700 } 701 702 /* FALLTHROUGH -- it's a UDP Packet */ 703 } 704 705 /* 706 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 707 */ 708 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 709 710 if (up->pcrlen == 0) { /* full coverage was set */ 711 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 712 UDP_SKB_CB(skb)->cscov, skb->len); 713 goto drop; 714 } 715 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 716 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 717 UDP_SKB_CB(skb)->cscov, up->pcrlen); 718 goto drop; 719 } 720 } 721 722 if (rcu_access_pointer(sk->sk_filter)) { 723 if (udp_lib_checksum_complete(skb)) 724 goto csum_error; 725 } 726 727 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 728 UDP6_INC_STATS_BH(sock_net(sk), 729 UDP_MIB_RCVBUFERRORS, is_udplite); 730 goto drop; 731 } 732 733 skb_dst_drop(skb); 734 735 bh_lock_sock(sk); 736 rc = 0; 737 if (!sock_owned_by_user(sk)) 738 rc = __udpv6_queue_rcv_skb(sk, skb); 739 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 740 bh_unlock_sock(sk); 741 goto drop; 742 } 743 bh_unlock_sock(sk); 744 745 return rc; 746 747 csum_error: 748 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 749 drop: 750 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 751 atomic_inc(&sk->sk_drops); 752 kfree_skb(skb); 753 return -1; 754 } 755 756 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 757 __be16 loc_port, const struct in6_addr *loc_addr, 758 __be16 rmt_port, const struct in6_addr *rmt_addr, 759 int dif, unsigned short hnum) 760 { 761 struct inet_sock *inet = inet_sk(sk); 762 763 if (!net_eq(sock_net(sk), net)) 764 return false; 765 766 if (udp_sk(sk)->udp_port_hash != hnum || 767 sk->sk_family != PF_INET6 || 768 (inet->inet_dport && inet->inet_dport != rmt_port) || 769 (!ipv6_addr_any(&sk->sk_v6_daddr) && 770 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 771 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) || 772 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 773 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 774 return false; 775 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 776 return false; 777 return true; 778 } 779 780 static void flush_stack(struct sock **stack, unsigned int count, 781 struct sk_buff *skb, unsigned int final) 782 { 783 struct sk_buff *skb1 = NULL; 784 struct sock *sk; 785 unsigned int i; 786 787 for (i = 0; i < count; i++) { 788 sk = stack[i]; 789 if (likely(!skb1)) 790 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 791 if (!skb1) { 792 atomic_inc(&sk->sk_drops); 793 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 794 IS_UDPLITE(sk)); 795 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 796 IS_UDPLITE(sk)); 797 } 798 799 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0) 800 skb1 = NULL; 801 sock_put(sk); 802 } 803 if (unlikely(skb1)) 804 kfree_skb(skb1); 805 } 806 807 static void udp6_csum_zero_error(struct sk_buff *skb) 808 { 809 /* RFC 2460 section 8.1 says that we SHOULD log 810 * this error. Well, it is reasonable. 811 */ 812 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 813 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 814 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 815 } 816 817 /* 818 * Note: called only from the BH handler context, 819 * so we don't need to lock the hashes. 820 */ 821 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 822 const struct in6_addr *saddr, const struct in6_addr *daddr, 823 struct udp_table *udptable, int proto) 824 { 825 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 826 const struct udphdr *uh = udp_hdr(skb); 827 struct hlist_nulls_node *node; 828 unsigned short hnum = ntohs(uh->dest); 829 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 830 int dif = inet6_iif(skb); 831 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 832 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 833 bool inner_flushed = false; 834 835 if (use_hash2) { 836 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) & 837 udp_table.mask; 838 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask; 839 start_lookup: 840 hslot = &udp_table.hash2[hash2]; 841 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 842 } 843 844 spin_lock(&hslot->lock); 845 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 846 if (__udp_v6_is_mcast_sock(net, sk, 847 uh->dest, daddr, 848 uh->source, saddr, 849 dif, hnum) && 850 /* If zero checksum and no_check is not on for 851 * the socket then skip it. 852 */ 853 (uh->check || udp_sk(sk)->no_check6_rx)) { 854 if (unlikely(count == ARRAY_SIZE(stack))) { 855 flush_stack(stack, count, skb, ~0); 856 inner_flushed = true; 857 count = 0; 858 } 859 stack[count++] = sk; 860 sock_hold(sk); 861 } 862 } 863 864 spin_unlock(&hslot->lock); 865 866 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 867 if (use_hash2 && hash2 != hash2_any) { 868 hash2 = hash2_any; 869 goto start_lookup; 870 } 871 872 if (count) { 873 flush_stack(stack, count, skb, count - 1); 874 } else { 875 if (!inner_flushed) 876 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 877 proto == IPPROTO_UDPLITE); 878 consume_skb(skb); 879 } 880 return 0; 881 } 882 883 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 884 int proto) 885 { 886 struct net *net = dev_net(skb->dev); 887 struct sock *sk; 888 struct udphdr *uh; 889 const struct in6_addr *saddr, *daddr; 890 u32 ulen = 0; 891 892 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 893 goto discard; 894 895 saddr = &ipv6_hdr(skb)->saddr; 896 daddr = &ipv6_hdr(skb)->daddr; 897 uh = udp_hdr(skb); 898 899 ulen = ntohs(uh->len); 900 if (ulen > skb->len) 901 goto short_packet; 902 903 if (proto == IPPROTO_UDP) { 904 /* UDP validates ulen. */ 905 906 /* Check for jumbo payload */ 907 if (ulen == 0) 908 ulen = skb->len; 909 910 if (ulen < sizeof(*uh)) 911 goto short_packet; 912 913 if (ulen < skb->len) { 914 if (pskb_trim_rcsum(skb, ulen)) 915 goto short_packet; 916 saddr = &ipv6_hdr(skb)->saddr; 917 daddr = &ipv6_hdr(skb)->daddr; 918 uh = udp_hdr(skb); 919 } 920 } 921 922 if (udp6_csum_init(skb, uh, proto)) 923 goto csum_error; 924 925 /* 926 * Multicast receive code 927 */ 928 if (ipv6_addr_is_multicast(daddr)) 929 return __udp6_lib_mcast_deliver(net, skb, 930 saddr, daddr, udptable, proto); 931 932 /* Unicast */ 933 934 /* 935 * check socket cache ... must talk to Alan about his plans 936 * for sock caches... i'll skip this for now. 937 */ 938 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 939 if (sk) { 940 int ret; 941 942 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 943 sock_put(sk); 944 udp6_csum_zero_error(skb); 945 goto csum_error; 946 } 947 948 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 949 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 950 ip6_compute_pseudo); 951 952 ret = udpv6_queue_rcv_skb(sk, skb); 953 sock_put(sk); 954 955 /* a return value > 0 means to resubmit the input, but 956 * it wants the return to be -protocol, or 0 957 */ 958 if (ret > 0) 959 return -ret; 960 961 return 0; 962 } 963 964 if (!uh->check) { 965 udp6_csum_zero_error(skb); 966 goto csum_error; 967 } 968 969 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 970 goto discard; 971 972 if (udp_lib_checksum_complete(skb)) 973 goto csum_error; 974 975 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 976 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 977 978 kfree_skb(skb); 979 return 0; 980 981 short_packet: 982 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 983 proto == IPPROTO_UDPLITE ? "-Lite" : "", 984 saddr, ntohs(uh->source), 985 ulen, skb->len, 986 daddr, ntohs(uh->dest)); 987 goto discard; 988 csum_error: 989 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 990 discard: 991 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 992 kfree_skb(skb); 993 return 0; 994 } 995 996 static __inline__ int udpv6_rcv(struct sk_buff *skb) 997 { 998 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 999 } 1000 1001 /* 1002 * Throw away all pending data and cancel the corking. Socket is locked. 1003 */ 1004 static void udp_v6_flush_pending_frames(struct sock *sk) 1005 { 1006 struct udp_sock *up = udp_sk(sk); 1007 1008 if (up->pending == AF_INET) 1009 udp_flush_pending_frames(sk); 1010 else if (up->pending) { 1011 up->len = 0; 1012 up->pending = 0; 1013 ip6_flush_pending_frames(sk); 1014 } 1015 } 1016 1017 /** 1018 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1019 * @sk: socket we are sending on 1020 * @skb: sk_buff containing the filled-in UDP header 1021 * (checksum field must be zeroed out) 1022 */ 1023 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1024 const struct in6_addr *saddr, 1025 const struct in6_addr *daddr, int len) 1026 { 1027 unsigned int offset; 1028 struct udphdr *uh = udp_hdr(skb); 1029 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1030 __wsum csum = 0; 1031 1032 if (!frags) { 1033 /* Only one fragment on the socket. */ 1034 skb->csum_start = skb_transport_header(skb) - skb->head; 1035 skb->csum_offset = offsetof(struct udphdr, check); 1036 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1037 } else { 1038 /* 1039 * HW-checksum won't work as there are two or more 1040 * fragments on the socket so that all csums of sk_buffs 1041 * should be together 1042 */ 1043 offset = skb_transport_offset(skb); 1044 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1045 1046 skb->ip_summed = CHECKSUM_NONE; 1047 1048 do { 1049 csum = csum_add(csum, frags->csum); 1050 } while ((frags = frags->next)); 1051 1052 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1053 csum); 1054 if (uh->check == 0) 1055 uh->check = CSUM_MANGLED_0; 1056 } 1057 } 1058 1059 /* 1060 * Sending 1061 */ 1062 1063 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6) 1064 { 1065 struct sock *sk = skb->sk; 1066 struct udphdr *uh; 1067 int err = 0; 1068 int is_udplite = IS_UDPLITE(sk); 1069 __wsum csum = 0; 1070 int offset = skb_transport_offset(skb); 1071 int len = skb->len - offset; 1072 1073 /* 1074 * Create a UDP header 1075 */ 1076 uh = udp_hdr(skb); 1077 uh->source = fl6->fl6_sport; 1078 uh->dest = fl6->fl6_dport; 1079 uh->len = htons(len); 1080 uh->check = 0; 1081 1082 if (is_udplite) 1083 csum = udplite_csum(skb); 1084 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1085 skb->ip_summed = CHECKSUM_NONE; 1086 goto send; 1087 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1088 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1089 goto send; 1090 } else 1091 csum = udp_csum(skb); 1092 1093 /* add protocol-dependent pseudo-header */ 1094 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1095 len, fl6->flowi6_proto, csum); 1096 if (uh->check == 0) 1097 uh->check = CSUM_MANGLED_0; 1098 1099 send: 1100 err = ip6_send_skb(skb); 1101 if (err) { 1102 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1103 UDP6_INC_STATS_USER(sock_net(sk), 1104 UDP_MIB_SNDBUFERRORS, is_udplite); 1105 err = 0; 1106 } 1107 } else 1108 UDP6_INC_STATS_USER(sock_net(sk), 1109 UDP_MIB_OUTDATAGRAMS, is_udplite); 1110 return err; 1111 } 1112 1113 static int udp_v6_push_pending_frames(struct sock *sk) 1114 { 1115 struct sk_buff *skb; 1116 struct udp_sock *up = udp_sk(sk); 1117 struct flowi6 fl6; 1118 int err = 0; 1119 1120 if (up->pending == AF_INET) 1121 return udp_push_pending_frames(sk); 1122 1123 /* ip6_finish_skb will release the cork, so make a copy of 1124 * fl6 here. 1125 */ 1126 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1127 1128 skb = ip6_finish_skb(sk); 1129 if (!skb) 1130 goto out; 1131 1132 err = udp_v6_send_skb(skb, &fl6); 1133 1134 out: 1135 up->len = 0; 1136 up->pending = 0; 1137 return err; 1138 } 1139 1140 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1141 { 1142 struct ipv6_txoptions opt_space; 1143 struct udp_sock *up = udp_sk(sk); 1144 struct inet_sock *inet = inet_sk(sk); 1145 struct ipv6_pinfo *np = inet6_sk(sk); 1146 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1147 struct in6_addr *daddr, *final_p, final; 1148 struct ipv6_txoptions *opt = NULL; 1149 struct ipv6_txoptions *opt_to_free = NULL; 1150 struct ip6_flowlabel *flowlabel = NULL; 1151 struct flowi6 fl6; 1152 struct dst_entry *dst; 1153 int addr_len = msg->msg_namelen; 1154 int ulen = len; 1155 int hlimit = -1; 1156 int tclass = -1; 1157 int dontfrag = -1; 1158 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1159 int err; 1160 int connected = 0; 1161 int is_udplite = IS_UDPLITE(sk); 1162 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1163 1164 /* destination address check */ 1165 if (sin6) { 1166 if (addr_len < offsetof(struct sockaddr, sa_data)) 1167 return -EINVAL; 1168 1169 switch (sin6->sin6_family) { 1170 case AF_INET6: 1171 if (addr_len < SIN6_LEN_RFC2133) 1172 return -EINVAL; 1173 daddr = &sin6->sin6_addr; 1174 break; 1175 case AF_INET: 1176 goto do_udp_sendmsg; 1177 case AF_UNSPEC: 1178 msg->msg_name = sin6 = NULL; 1179 msg->msg_namelen = addr_len = 0; 1180 daddr = NULL; 1181 break; 1182 default: 1183 return -EINVAL; 1184 } 1185 } else if (!up->pending) { 1186 if (sk->sk_state != TCP_ESTABLISHED) 1187 return -EDESTADDRREQ; 1188 daddr = &sk->sk_v6_daddr; 1189 } else 1190 daddr = NULL; 1191 1192 if (daddr) { 1193 if (ipv6_addr_v4mapped(daddr)) { 1194 struct sockaddr_in sin; 1195 sin.sin_family = AF_INET; 1196 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1197 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1198 msg->msg_name = &sin; 1199 msg->msg_namelen = sizeof(sin); 1200 do_udp_sendmsg: 1201 if (__ipv6_only_sock(sk)) 1202 return -ENETUNREACH; 1203 return udp_sendmsg(sk, msg, len); 1204 } 1205 } 1206 1207 if (up->pending == AF_INET) 1208 return udp_sendmsg(sk, msg, len); 1209 1210 /* Rough check on arithmetic overflow, 1211 better check is made in ip6_append_data(). 1212 */ 1213 if (len > INT_MAX - sizeof(struct udphdr)) 1214 return -EMSGSIZE; 1215 1216 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1217 if (up->pending) { 1218 /* 1219 * There are pending frames. 1220 * The socket lock must be held while it's corked. 1221 */ 1222 lock_sock(sk); 1223 if (likely(up->pending)) { 1224 if (unlikely(up->pending != AF_INET6)) { 1225 release_sock(sk); 1226 return -EAFNOSUPPORT; 1227 } 1228 dst = NULL; 1229 goto do_append_data; 1230 } 1231 release_sock(sk); 1232 } 1233 ulen += sizeof(struct udphdr); 1234 1235 memset(&fl6, 0, sizeof(fl6)); 1236 1237 if (sin6) { 1238 if (sin6->sin6_port == 0) 1239 return -EINVAL; 1240 1241 fl6.fl6_dport = sin6->sin6_port; 1242 daddr = &sin6->sin6_addr; 1243 1244 if (np->sndflow) { 1245 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1246 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1247 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1248 if (!flowlabel) 1249 return -EINVAL; 1250 } 1251 } 1252 1253 /* 1254 * Otherwise it will be difficult to maintain 1255 * sk->sk_dst_cache. 1256 */ 1257 if (sk->sk_state == TCP_ESTABLISHED && 1258 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1259 daddr = &sk->sk_v6_daddr; 1260 1261 if (addr_len >= sizeof(struct sockaddr_in6) && 1262 sin6->sin6_scope_id && 1263 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1264 fl6.flowi6_oif = sin6->sin6_scope_id; 1265 } else { 1266 if (sk->sk_state != TCP_ESTABLISHED) 1267 return -EDESTADDRREQ; 1268 1269 fl6.fl6_dport = inet->inet_dport; 1270 daddr = &sk->sk_v6_daddr; 1271 fl6.flowlabel = np->flow_label; 1272 connected = 1; 1273 } 1274 1275 if (!fl6.flowi6_oif) 1276 fl6.flowi6_oif = sk->sk_bound_dev_if; 1277 1278 if (!fl6.flowi6_oif) 1279 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1280 1281 fl6.flowi6_mark = sk->sk_mark; 1282 1283 if (msg->msg_controllen) { 1284 opt = &opt_space; 1285 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1286 opt->tot_len = sizeof(*opt); 1287 1288 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt, 1289 &hlimit, &tclass, &dontfrag); 1290 if (err < 0) { 1291 fl6_sock_release(flowlabel); 1292 return err; 1293 } 1294 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1295 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1296 if (!flowlabel) 1297 return -EINVAL; 1298 } 1299 if (!(opt->opt_nflen|opt->opt_flen)) 1300 opt = NULL; 1301 connected = 0; 1302 } 1303 if (!opt) { 1304 opt = txopt_get(np); 1305 opt_to_free = opt; 1306 } 1307 if (flowlabel) 1308 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1309 opt = ipv6_fixup_options(&opt_space, opt); 1310 1311 fl6.flowi6_proto = sk->sk_protocol; 1312 if (!ipv6_addr_any(daddr)) 1313 fl6.daddr = *daddr; 1314 else 1315 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1316 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1317 fl6.saddr = np->saddr; 1318 fl6.fl6_sport = inet->inet_sport; 1319 1320 final_p = fl6_update_dst(&fl6, opt, &final); 1321 if (final_p) 1322 connected = 0; 1323 1324 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1325 fl6.flowi6_oif = np->mcast_oif; 1326 connected = 0; 1327 } else if (!fl6.flowi6_oif) 1328 fl6.flowi6_oif = np->ucast_oif; 1329 1330 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1331 1332 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p); 1333 if (IS_ERR(dst)) { 1334 err = PTR_ERR(dst); 1335 dst = NULL; 1336 goto out; 1337 } 1338 1339 if (hlimit < 0) 1340 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1341 1342 if (tclass < 0) 1343 tclass = np->tclass; 1344 1345 if (msg->msg_flags&MSG_CONFIRM) 1346 goto do_confirm; 1347 back_from_confirm: 1348 1349 /* Lockless fast path for the non-corking case */ 1350 if (!corkreq) { 1351 struct sk_buff *skb; 1352 1353 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1354 sizeof(struct udphdr), hlimit, tclass, opt, 1355 &fl6, (struct rt6_info *)dst, 1356 msg->msg_flags, dontfrag); 1357 err = PTR_ERR(skb); 1358 if (!IS_ERR_OR_NULL(skb)) 1359 err = udp_v6_send_skb(skb, &fl6); 1360 goto release_dst; 1361 } 1362 1363 lock_sock(sk); 1364 if (unlikely(up->pending)) { 1365 /* The socket is already corked while preparing it. */ 1366 /* ... which is an evident application bug. --ANK */ 1367 release_sock(sk); 1368 1369 net_dbg_ratelimited("udp cork app bug 2\n"); 1370 err = -EINVAL; 1371 goto out; 1372 } 1373 1374 up->pending = AF_INET6; 1375 1376 do_append_data: 1377 if (dontfrag < 0) 1378 dontfrag = np->dontfrag; 1379 up->len += ulen; 1380 err = ip6_append_data(sk, getfrag, msg, ulen, 1381 sizeof(struct udphdr), hlimit, tclass, opt, &fl6, 1382 (struct rt6_info *)dst, 1383 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag); 1384 if (err) 1385 udp_v6_flush_pending_frames(sk); 1386 else if (!corkreq) 1387 err = udp_v6_push_pending_frames(sk); 1388 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1389 up->pending = 0; 1390 1391 if (err > 0) 1392 err = np->recverr ? net_xmit_errno(err) : 0; 1393 release_sock(sk); 1394 1395 release_dst: 1396 if (dst) { 1397 if (connected) { 1398 ip6_dst_store(sk, dst, 1399 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ? 1400 &sk->sk_v6_daddr : NULL, 1401 #ifdef CONFIG_IPV6_SUBTREES 1402 ipv6_addr_equal(&fl6.saddr, &np->saddr) ? 1403 &np->saddr : 1404 #endif 1405 NULL); 1406 } else { 1407 dst_release(dst); 1408 } 1409 dst = NULL; 1410 } 1411 1412 out: 1413 dst_release(dst); 1414 fl6_sock_release(flowlabel); 1415 txopt_put(opt_to_free); 1416 if (!err) 1417 return len; 1418 /* 1419 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1420 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1421 * we don't have a good statistic (IpOutDiscards but it can be too many 1422 * things). We could add another new stat but at least for now that 1423 * seems like overkill. 1424 */ 1425 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1426 UDP6_INC_STATS_USER(sock_net(sk), 1427 UDP_MIB_SNDBUFERRORS, is_udplite); 1428 } 1429 return err; 1430 1431 do_confirm: 1432 dst_confirm(dst); 1433 if (!(msg->msg_flags&MSG_PROBE) || len) 1434 goto back_from_confirm; 1435 err = 0; 1436 goto out; 1437 } 1438 1439 void udpv6_destroy_sock(struct sock *sk) 1440 { 1441 struct udp_sock *up = udp_sk(sk); 1442 lock_sock(sk); 1443 udp_v6_flush_pending_frames(sk); 1444 release_sock(sk); 1445 1446 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 1447 void (*encap_destroy)(struct sock *sk); 1448 encap_destroy = ACCESS_ONCE(up->encap_destroy); 1449 if (encap_destroy) 1450 encap_destroy(sk); 1451 } 1452 1453 inet6_destroy_sock(sk); 1454 } 1455 1456 /* 1457 * Socket option code for UDP 1458 */ 1459 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1460 char __user *optval, unsigned int optlen) 1461 { 1462 if (level == SOL_UDP || level == SOL_UDPLITE) 1463 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1464 udp_v6_push_pending_frames); 1465 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1466 } 1467 1468 #ifdef CONFIG_COMPAT 1469 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1470 char __user *optval, unsigned int optlen) 1471 { 1472 if (level == SOL_UDP || level == SOL_UDPLITE) 1473 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1474 udp_v6_push_pending_frames); 1475 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1476 } 1477 #endif 1478 1479 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1480 char __user *optval, int __user *optlen) 1481 { 1482 if (level == SOL_UDP || level == SOL_UDPLITE) 1483 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1484 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1485 } 1486 1487 #ifdef CONFIG_COMPAT 1488 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1489 char __user *optval, int __user *optlen) 1490 { 1491 if (level == SOL_UDP || level == SOL_UDPLITE) 1492 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1493 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1494 } 1495 #endif 1496 1497 static const struct inet6_protocol udpv6_protocol = { 1498 .handler = udpv6_rcv, 1499 .err_handler = udpv6_err, 1500 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1501 }; 1502 1503 /* ------------------------------------------------------------------------ */ 1504 #ifdef CONFIG_PROC_FS 1505 int udp6_seq_show(struct seq_file *seq, void *v) 1506 { 1507 if (v == SEQ_START_TOKEN) { 1508 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1509 } else { 1510 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1511 struct inet_sock *inet = inet_sk(v); 1512 __u16 srcp = ntohs(inet->inet_sport); 1513 __u16 destp = ntohs(inet->inet_dport); 1514 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket); 1515 } 1516 return 0; 1517 } 1518 1519 static const struct file_operations udp6_afinfo_seq_fops = { 1520 .owner = THIS_MODULE, 1521 .open = udp_seq_open, 1522 .read = seq_read, 1523 .llseek = seq_lseek, 1524 .release = seq_release_net 1525 }; 1526 1527 static struct udp_seq_afinfo udp6_seq_afinfo = { 1528 .name = "udp6", 1529 .family = AF_INET6, 1530 .udp_table = &udp_table, 1531 .seq_fops = &udp6_afinfo_seq_fops, 1532 .seq_ops = { 1533 .show = udp6_seq_show, 1534 }, 1535 }; 1536 1537 int __net_init udp6_proc_init(struct net *net) 1538 { 1539 return udp_proc_register(net, &udp6_seq_afinfo); 1540 } 1541 1542 void udp6_proc_exit(struct net *net) 1543 { 1544 udp_proc_unregister(net, &udp6_seq_afinfo); 1545 } 1546 #endif /* CONFIG_PROC_FS */ 1547 1548 void udp_v6_clear_sk(struct sock *sk, int size) 1549 { 1550 struct inet_sock *inet = inet_sk(sk); 1551 1552 /* we do not want to clear pinet6 field, because of RCU lookups */ 1553 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6)); 1554 1555 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6); 1556 memset(&inet->pinet6 + 1, 0, size); 1557 } 1558 1559 /* ------------------------------------------------------------------------ */ 1560 1561 struct proto udpv6_prot = { 1562 .name = "UDPv6", 1563 .owner = THIS_MODULE, 1564 .close = udp_lib_close, 1565 .connect = ip6_datagram_connect, 1566 .disconnect = udp_disconnect, 1567 .ioctl = udp_ioctl, 1568 .destroy = udpv6_destroy_sock, 1569 .setsockopt = udpv6_setsockopt, 1570 .getsockopt = udpv6_getsockopt, 1571 .sendmsg = udpv6_sendmsg, 1572 .recvmsg = udpv6_recvmsg, 1573 .backlog_rcv = __udpv6_queue_rcv_skb, 1574 .hash = udp_lib_hash, 1575 .unhash = udp_lib_unhash, 1576 .rehash = udp_v6_rehash, 1577 .get_port = udp_v6_get_port, 1578 .memory_allocated = &udp_memory_allocated, 1579 .sysctl_mem = sysctl_udp_mem, 1580 .sysctl_wmem = &sysctl_udp_wmem_min, 1581 .sysctl_rmem = &sysctl_udp_rmem_min, 1582 .obj_size = sizeof(struct udp6_sock), 1583 .slab_flags = SLAB_DESTROY_BY_RCU, 1584 .h.udp_table = &udp_table, 1585 #ifdef CONFIG_COMPAT 1586 .compat_setsockopt = compat_udpv6_setsockopt, 1587 .compat_getsockopt = compat_udpv6_getsockopt, 1588 #endif 1589 .clear_sk = udp_v6_clear_sk, 1590 }; 1591 1592 static struct inet_protosw udpv6_protosw = { 1593 .type = SOCK_DGRAM, 1594 .protocol = IPPROTO_UDP, 1595 .prot = &udpv6_prot, 1596 .ops = &inet6_dgram_ops, 1597 .flags = INET_PROTOSW_PERMANENT, 1598 }; 1599 1600 int __init udpv6_init(void) 1601 { 1602 int ret; 1603 1604 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1605 if (ret) 1606 goto out; 1607 1608 ret = inet6_register_protosw(&udpv6_protosw); 1609 if (ret) 1610 goto out_udpv6_protocol; 1611 out: 1612 return ret; 1613 1614 out_udpv6_protocol: 1615 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1616 goto out; 1617 } 1618 1619 void udpv6_exit(void) 1620 { 1621 inet6_unregister_protosw(&udpv6_protosw); 1622 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1623 } 1624