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 bool select_ok = true; 261 u32 hash = 0; 262 263 begin: 264 result = NULL; 265 badness = -1; 266 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 267 score = compute_score2(sk, net, saddr, sport, 268 daddr, hnum, dif); 269 if (score > badness) { 270 result = sk; 271 badness = score; 272 reuseport = sk->sk_reuseport; 273 if (reuseport) { 274 hash = udp6_ehashfn(net, daddr, hnum, 275 saddr, sport); 276 if (select_ok) { 277 struct sock *sk2; 278 279 sk2 = reuseport_select_sock(sk, hash, skb, 280 sizeof(struct udphdr)); 281 if (sk2) { 282 result = sk2; 283 select_ok = false; 284 goto found; 285 } 286 } 287 matches = 1; 288 } 289 } else if (score == badness && reuseport) { 290 matches++; 291 if (reciprocal_scale(hash, matches) == 0) 292 result = sk; 293 hash = next_pseudo_random32(hash); 294 } 295 } 296 /* 297 * if the nulls value we got at the end of this lookup is 298 * not the expected one, we must restart lookup. 299 * We probably met an item that was moved to another chain. 300 */ 301 if (get_nulls_value(node) != slot2) 302 goto begin; 303 304 if (result) { 305 found: 306 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 307 result = NULL; 308 else if (unlikely(compute_score2(result, net, saddr, sport, 309 daddr, hnum, dif) < badness)) { 310 sock_put(result); 311 goto begin; 312 } 313 } 314 return result; 315 } 316 317 struct sock *__udp6_lib_lookup(struct net *net, 318 const struct in6_addr *saddr, __be16 sport, 319 const struct in6_addr *daddr, __be16 dport, 320 int dif, struct udp_table *udptable, 321 struct sk_buff *skb) 322 { 323 struct sock *sk, *result; 324 struct hlist_nulls_node *node; 325 unsigned short hnum = ntohs(dport); 326 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 327 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 328 int score, badness, matches = 0, reuseport = 0; 329 bool select_ok = true; 330 u32 hash = 0; 331 332 rcu_read_lock(); 333 if (hslot->count > 10) { 334 hash2 = udp6_portaddr_hash(net, daddr, hnum); 335 slot2 = hash2 & udptable->mask; 336 hslot2 = &udptable->hash2[slot2]; 337 if (hslot->count < hslot2->count) 338 goto begin; 339 340 result = udp6_lib_lookup2(net, saddr, sport, 341 daddr, hnum, dif, 342 hslot2, slot2, skb); 343 if (!result) { 344 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum); 345 slot2 = hash2 & udptable->mask; 346 hslot2 = &udptable->hash2[slot2]; 347 if (hslot->count < hslot2->count) 348 goto begin; 349 350 result = udp6_lib_lookup2(net, saddr, sport, 351 &in6addr_any, hnum, dif, 352 hslot2, slot2, skb); 353 } 354 rcu_read_unlock(); 355 return result; 356 } 357 begin: 358 result = NULL; 359 badness = -1; 360 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 361 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif); 362 if (score > badness) { 363 result = sk; 364 badness = score; 365 reuseport = sk->sk_reuseport; 366 if (reuseport) { 367 hash = udp6_ehashfn(net, daddr, hnum, 368 saddr, sport); 369 if (select_ok) { 370 struct sock *sk2; 371 372 sk2 = reuseport_select_sock(sk, hash, skb, 373 sizeof(struct udphdr)); 374 if (sk2) { 375 result = sk2; 376 select_ok = false; 377 goto found; 378 } 379 } 380 matches = 1; 381 } 382 } else if (score == badness && reuseport) { 383 matches++; 384 if (reciprocal_scale(hash, matches) == 0) 385 result = sk; 386 hash = next_pseudo_random32(hash); 387 } 388 } 389 /* 390 * if the nulls value we got at the end of this lookup is 391 * not the expected one, we must restart lookup. 392 * We probably met an item that was moved to another chain. 393 */ 394 if (get_nulls_value(node) != slot) 395 goto begin; 396 397 if (result) { 398 found: 399 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 400 result = NULL; 401 else if (unlikely(compute_score(result, net, hnum, saddr, sport, 402 daddr, dport, dif) < badness)) { 403 sock_put(result); 404 goto begin; 405 } 406 } 407 rcu_read_unlock(); 408 return result; 409 } 410 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 411 412 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 413 __be16 sport, __be16 dport, 414 struct udp_table *udptable) 415 { 416 struct sock *sk; 417 const struct ipv6hdr *iph = ipv6_hdr(skb); 418 419 sk = skb_steal_sock(skb); 420 if (unlikely(sk)) 421 return sk; 422 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport, 423 &iph->daddr, dport, inet6_iif(skb), 424 udptable, skb); 425 } 426 427 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 428 const struct in6_addr *daddr, __be16 dport, int dif) 429 { 430 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table, NULL); 431 } 432 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 433 434 /* 435 * This should be easy, if there is something there we 436 * return it, otherwise we block. 437 */ 438 439 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 440 int noblock, int flags, int *addr_len) 441 { 442 struct ipv6_pinfo *np = inet6_sk(sk); 443 struct inet_sock *inet = inet_sk(sk); 444 struct sk_buff *skb; 445 unsigned int ulen, copied; 446 int peeked, off = 0; 447 int err; 448 int is_udplite = IS_UDPLITE(sk); 449 bool checksum_valid = false; 450 int is_udp4; 451 bool slow; 452 453 if (flags & MSG_ERRQUEUE) 454 return ipv6_recv_error(sk, msg, len, addr_len); 455 456 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 457 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 458 459 try_again: 460 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 461 &peeked, &off, &err); 462 if (!skb) 463 goto out; 464 465 ulen = skb->len - sizeof(struct udphdr); 466 copied = len; 467 if (copied > ulen) 468 copied = ulen; 469 else if (copied < ulen) 470 msg->msg_flags |= MSG_TRUNC; 471 472 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 473 474 /* 475 * If checksum is needed at all, try to do it while copying the 476 * data. If the data is truncated, or if we only want a partial 477 * coverage checksum (UDP-Lite), do it before the copy. 478 */ 479 480 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 481 checksum_valid = !udp_lib_checksum_complete(skb); 482 if (!checksum_valid) 483 goto csum_copy_err; 484 } 485 486 if (checksum_valid || skb_csum_unnecessary(skb)) 487 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 488 msg, copied); 489 else { 490 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), msg); 491 if (err == -EINVAL) 492 goto csum_copy_err; 493 } 494 if (unlikely(err)) { 495 trace_kfree_skb(skb, udpv6_recvmsg); 496 if (!peeked) { 497 atomic_inc(&sk->sk_drops); 498 if (is_udp4) 499 UDP_INC_STATS_USER(sock_net(sk), 500 UDP_MIB_INERRORS, 501 is_udplite); 502 else 503 UDP6_INC_STATS_USER(sock_net(sk), 504 UDP_MIB_INERRORS, 505 is_udplite); 506 } 507 goto out_free; 508 } 509 if (!peeked) { 510 if (is_udp4) 511 UDP_INC_STATS_USER(sock_net(sk), 512 UDP_MIB_INDATAGRAMS, is_udplite); 513 else 514 UDP6_INC_STATS_USER(sock_net(sk), 515 UDP_MIB_INDATAGRAMS, is_udplite); 516 } 517 518 sock_recv_ts_and_drops(msg, sk, skb); 519 520 /* Copy the address. */ 521 if (msg->msg_name) { 522 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 523 sin6->sin6_family = AF_INET6; 524 sin6->sin6_port = udp_hdr(skb)->source; 525 sin6->sin6_flowinfo = 0; 526 527 if (is_udp4) { 528 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 529 &sin6->sin6_addr); 530 sin6->sin6_scope_id = 0; 531 } else { 532 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 533 sin6->sin6_scope_id = 534 ipv6_iface_scope_id(&sin6->sin6_addr, 535 inet6_iif(skb)); 536 } 537 *addr_len = sizeof(*sin6); 538 } 539 540 if (np->rxopt.all) 541 ip6_datagram_recv_common_ctl(sk, msg, skb); 542 543 if (is_udp4) { 544 if (inet->cmsg_flags) 545 ip_cmsg_recv(msg, skb); 546 } else { 547 if (np->rxopt.all) 548 ip6_datagram_recv_specific_ctl(sk, msg, skb); 549 } 550 551 err = copied; 552 if (flags & MSG_TRUNC) 553 err = ulen; 554 555 out_free: 556 skb_free_datagram_locked(sk, skb); 557 out: 558 return err; 559 560 csum_copy_err: 561 slow = lock_sock_fast(sk); 562 if (!skb_kill_datagram(sk, skb, flags)) { 563 if (is_udp4) { 564 UDP_INC_STATS_USER(sock_net(sk), 565 UDP_MIB_CSUMERRORS, is_udplite); 566 UDP_INC_STATS_USER(sock_net(sk), 567 UDP_MIB_INERRORS, is_udplite); 568 } else { 569 UDP6_INC_STATS_USER(sock_net(sk), 570 UDP_MIB_CSUMERRORS, is_udplite); 571 UDP6_INC_STATS_USER(sock_net(sk), 572 UDP_MIB_INERRORS, is_udplite); 573 } 574 } 575 unlock_sock_fast(sk, slow); 576 577 /* starting over for a new packet, but check if we need to yield */ 578 cond_resched(); 579 msg->msg_flags &= ~MSG_TRUNC; 580 goto try_again; 581 } 582 583 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 584 u8 type, u8 code, int offset, __be32 info, 585 struct udp_table *udptable) 586 { 587 struct ipv6_pinfo *np; 588 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 589 const struct in6_addr *saddr = &hdr->saddr; 590 const struct in6_addr *daddr = &hdr->daddr; 591 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 592 struct sock *sk; 593 int err; 594 struct net *net = dev_net(skb->dev); 595 596 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 597 inet6_iif(skb), udptable, skb); 598 if (!sk) { 599 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), 600 ICMP6_MIB_INERRORS); 601 return; 602 } 603 604 if (type == ICMPV6_PKT_TOOBIG) { 605 if (!ip6_sk_accept_pmtu(sk)) 606 goto out; 607 ip6_sk_update_pmtu(skb, sk, info); 608 } 609 if (type == NDISC_REDIRECT) { 610 ip6_sk_redirect(skb, sk); 611 goto out; 612 } 613 614 np = inet6_sk(sk); 615 616 if (!icmpv6_err_convert(type, code, &err) && !np->recverr) 617 goto out; 618 619 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr) 620 goto out; 621 622 if (np->recverr) 623 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 624 625 sk->sk_err = err; 626 sk->sk_error_report(sk); 627 out: 628 sock_put(sk); 629 } 630 631 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 632 { 633 int rc; 634 635 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 636 sock_rps_save_rxhash(sk, skb); 637 sk_mark_napi_id(sk, skb); 638 sk_incoming_cpu_update(sk); 639 } 640 641 rc = sock_queue_rcv_skb(sk, skb); 642 if (rc < 0) { 643 int is_udplite = IS_UDPLITE(sk); 644 645 /* Note that an ENOMEM error is charged twice */ 646 if (rc == -ENOMEM) 647 UDP6_INC_STATS_BH(sock_net(sk), 648 UDP_MIB_RCVBUFERRORS, is_udplite); 649 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 650 kfree_skb(skb); 651 return -1; 652 } 653 return 0; 654 } 655 656 static __inline__ void udpv6_err(struct sk_buff *skb, 657 struct inet6_skb_parm *opt, u8 type, 658 u8 code, int offset, __be32 info) 659 { 660 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 661 } 662 663 static struct static_key udpv6_encap_needed __read_mostly; 664 void udpv6_encap_enable(void) 665 { 666 if (!static_key_enabled(&udpv6_encap_needed)) 667 static_key_slow_inc(&udpv6_encap_needed); 668 } 669 EXPORT_SYMBOL(udpv6_encap_enable); 670 671 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 672 { 673 struct udp_sock *up = udp_sk(sk); 674 int rc; 675 int is_udplite = IS_UDPLITE(sk); 676 677 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 678 goto drop; 679 680 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 681 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 682 683 /* 684 * This is an encapsulation socket so pass the skb to 685 * the socket's udp_encap_rcv() hook. Otherwise, just 686 * fall through and pass this up the UDP socket. 687 * up->encap_rcv() returns the following value: 688 * =0 if skb was successfully passed to the encap 689 * handler or was discarded by it. 690 * >0 if skb should be passed on to UDP. 691 * <0 if skb should be resubmitted as proto -N 692 */ 693 694 /* if we're overly short, let UDP handle it */ 695 encap_rcv = ACCESS_ONCE(up->encap_rcv); 696 if (skb->len > sizeof(struct udphdr) && encap_rcv) { 697 int ret; 698 699 /* Verify checksum before giving to encap */ 700 if (udp_lib_checksum_complete(skb)) 701 goto csum_error; 702 703 ret = encap_rcv(sk, skb); 704 if (ret <= 0) { 705 UDP_INC_STATS_BH(sock_net(sk), 706 UDP_MIB_INDATAGRAMS, 707 is_udplite); 708 return -ret; 709 } 710 } 711 712 /* FALLTHROUGH -- it's a UDP Packet */ 713 } 714 715 /* 716 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 717 */ 718 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 719 720 if (up->pcrlen == 0) { /* full coverage was set */ 721 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 722 UDP_SKB_CB(skb)->cscov, skb->len); 723 goto drop; 724 } 725 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 726 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 727 UDP_SKB_CB(skb)->cscov, up->pcrlen); 728 goto drop; 729 } 730 } 731 732 if (rcu_access_pointer(sk->sk_filter)) { 733 if (udp_lib_checksum_complete(skb)) 734 goto csum_error; 735 } 736 737 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 738 UDP6_INC_STATS_BH(sock_net(sk), 739 UDP_MIB_RCVBUFERRORS, is_udplite); 740 goto drop; 741 } 742 743 skb_dst_drop(skb); 744 745 bh_lock_sock(sk); 746 rc = 0; 747 if (!sock_owned_by_user(sk)) 748 rc = __udpv6_queue_rcv_skb(sk, skb); 749 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 750 bh_unlock_sock(sk); 751 goto drop; 752 } 753 bh_unlock_sock(sk); 754 755 return rc; 756 757 csum_error: 758 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 759 drop: 760 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 761 atomic_inc(&sk->sk_drops); 762 kfree_skb(skb); 763 return -1; 764 } 765 766 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 767 __be16 loc_port, const struct in6_addr *loc_addr, 768 __be16 rmt_port, const struct in6_addr *rmt_addr, 769 int dif, unsigned short hnum) 770 { 771 struct inet_sock *inet = inet_sk(sk); 772 773 if (!net_eq(sock_net(sk), net)) 774 return false; 775 776 if (udp_sk(sk)->udp_port_hash != hnum || 777 sk->sk_family != PF_INET6 || 778 (inet->inet_dport && inet->inet_dport != rmt_port) || 779 (!ipv6_addr_any(&sk->sk_v6_daddr) && 780 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 781 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) || 782 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 783 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 784 return false; 785 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 786 return false; 787 return true; 788 } 789 790 static void flush_stack(struct sock **stack, unsigned int count, 791 struct sk_buff *skb, unsigned int final) 792 { 793 struct sk_buff *skb1 = NULL; 794 struct sock *sk; 795 unsigned int i; 796 797 for (i = 0; i < count; i++) { 798 sk = stack[i]; 799 if (likely(!skb1)) 800 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 801 if (!skb1) { 802 atomic_inc(&sk->sk_drops); 803 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 804 IS_UDPLITE(sk)); 805 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 806 IS_UDPLITE(sk)); 807 } 808 809 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0) 810 skb1 = NULL; 811 sock_put(sk); 812 } 813 if (unlikely(skb1)) 814 kfree_skb(skb1); 815 } 816 817 static void udp6_csum_zero_error(struct sk_buff *skb) 818 { 819 /* RFC 2460 section 8.1 says that we SHOULD log 820 * this error. Well, it is reasonable. 821 */ 822 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 823 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 824 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 825 } 826 827 /* 828 * Note: called only from the BH handler context, 829 * so we don't need to lock the hashes. 830 */ 831 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 832 const struct in6_addr *saddr, const struct in6_addr *daddr, 833 struct udp_table *udptable, int proto) 834 { 835 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 836 const struct udphdr *uh = udp_hdr(skb); 837 struct hlist_nulls_node *node; 838 unsigned short hnum = ntohs(uh->dest); 839 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 840 int dif = inet6_iif(skb); 841 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 842 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 843 bool inner_flushed = false; 844 845 if (use_hash2) { 846 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) & 847 udp_table.mask; 848 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask; 849 start_lookup: 850 hslot = &udp_table.hash2[hash2]; 851 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 852 } 853 854 spin_lock(&hslot->lock); 855 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 856 if (__udp_v6_is_mcast_sock(net, sk, 857 uh->dest, daddr, 858 uh->source, saddr, 859 dif, hnum) && 860 /* If zero checksum and no_check is not on for 861 * the socket then skip it. 862 */ 863 (uh->check || udp_sk(sk)->no_check6_rx)) { 864 if (unlikely(count == ARRAY_SIZE(stack))) { 865 flush_stack(stack, count, skb, ~0); 866 inner_flushed = true; 867 count = 0; 868 } 869 stack[count++] = sk; 870 sock_hold(sk); 871 } 872 } 873 874 spin_unlock(&hslot->lock); 875 876 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 877 if (use_hash2 && hash2 != hash2_any) { 878 hash2 = hash2_any; 879 goto start_lookup; 880 } 881 882 if (count) { 883 flush_stack(stack, count, skb, count - 1); 884 } else { 885 if (!inner_flushed) 886 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 887 proto == IPPROTO_UDPLITE); 888 consume_skb(skb); 889 } 890 return 0; 891 } 892 893 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 894 int proto) 895 { 896 struct net *net = dev_net(skb->dev); 897 struct sock *sk; 898 struct udphdr *uh; 899 const struct in6_addr *saddr, *daddr; 900 u32 ulen = 0; 901 902 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 903 goto discard; 904 905 saddr = &ipv6_hdr(skb)->saddr; 906 daddr = &ipv6_hdr(skb)->daddr; 907 uh = udp_hdr(skb); 908 909 ulen = ntohs(uh->len); 910 if (ulen > skb->len) 911 goto short_packet; 912 913 if (proto == IPPROTO_UDP) { 914 /* UDP validates ulen. */ 915 916 /* Check for jumbo payload */ 917 if (ulen == 0) 918 ulen = skb->len; 919 920 if (ulen < sizeof(*uh)) 921 goto short_packet; 922 923 if (ulen < skb->len) { 924 if (pskb_trim_rcsum(skb, ulen)) 925 goto short_packet; 926 saddr = &ipv6_hdr(skb)->saddr; 927 daddr = &ipv6_hdr(skb)->daddr; 928 uh = udp_hdr(skb); 929 } 930 } 931 932 if (udp6_csum_init(skb, uh, proto)) 933 goto csum_error; 934 935 /* 936 * Multicast receive code 937 */ 938 if (ipv6_addr_is_multicast(daddr)) 939 return __udp6_lib_mcast_deliver(net, skb, 940 saddr, daddr, udptable, proto); 941 942 /* Unicast */ 943 944 /* 945 * check socket cache ... must talk to Alan about his plans 946 * for sock caches... i'll skip this for now. 947 */ 948 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 949 if (sk) { 950 int ret; 951 952 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 953 sock_put(sk); 954 udp6_csum_zero_error(skb); 955 goto csum_error; 956 } 957 958 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 959 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 960 ip6_compute_pseudo); 961 962 ret = udpv6_queue_rcv_skb(sk, skb); 963 sock_put(sk); 964 965 /* a return value > 0 means to resubmit the input */ 966 if (ret > 0) 967 return ret; 968 969 return 0; 970 } 971 972 if (!uh->check) { 973 udp6_csum_zero_error(skb); 974 goto csum_error; 975 } 976 977 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 978 goto discard; 979 980 if (udp_lib_checksum_complete(skb)) 981 goto csum_error; 982 983 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 984 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 985 986 kfree_skb(skb); 987 return 0; 988 989 short_packet: 990 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 991 proto == IPPROTO_UDPLITE ? "-Lite" : "", 992 saddr, ntohs(uh->source), 993 ulen, skb->len, 994 daddr, ntohs(uh->dest)); 995 goto discard; 996 csum_error: 997 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 998 discard: 999 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1000 kfree_skb(skb); 1001 return 0; 1002 } 1003 1004 static __inline__ int udpv6_rcv(struct sk_buff *skb) 1005 { 1006 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1007 } 1008 1009 /* 1010 * Throw away all pending data and cancel the corking. Socket is locked. 1011 */ 1012 static void udp_v6_flush_pending_frames(struct sock *sk) 1013 { 1014 struct udp_sock *up = udp_sk(sk); 1015 1016 if (up->pending == AF_INET) 1017 udp_flush_pending_frames(sk); 1018 else if (up->pending) { 1019 up->len = 0; 1020 up->pending = 0; 1021 ip6_flush_pending_frames(sk); 1022 } 1023 } 1024 1025 /** 1026 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1027 * @sk: socket we are sending on 1028 * @skb: sk_buff containing the filled-in UDP header 1029 * (checksum field must be zeroed out) 1030 */ 1031 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1032 const struct in6_addr *saddr, 1033 const struct in6_addr *daddr, int len) 1034 { 1035 unsigned int offset; 1036 struct udphdr *uh = udp_hdr(skb); 1037 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1038 __wsum csum = 0; 1039 1040 if (!frags) { 1041 /* Only one fragment on the socket. */ 1042 skb->csum_start = skb_transport_header(skb) - skb->head; 1043 skb->csum_offset = offsetof(struct udphdr, check); 1044 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1045 } else { 1046 /* 1047 * HW-checksum won't work as there are two or more 1048 * fragments on the socket so that all csums of sk_buffs 1049 * should be together 1050 */ 1051 offset = skb_transport_offset(skb); 1052 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1053 1054 skb->ip_summed = CHECKSUM_NONE; 1055 1056 do { 1057 csum = csum_add(csum, frags->csum); 1058 } while ((frags = frags->next)); 1059 1060 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1061 csum); 1062 if (uh->check == 0) 1063 uh->check = CSUM_MANGLED_0; 1064 } 1065 } 1066 1067 /* 1068 * Sending 1069 */ 1070 1071 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6) 1072 { 1073 struct sock *sk = skb->sk; 1074 struct udphdr *uh; 1075 int err = 0; 1076 int is_udplite = IS_UDPLITE(sk); 1077 __wsum csum = 0; 1078 int offset = skb_transport_offset(skb); 1079 int len = skb->len - offset; 1080 1081 /* 1082 * Create a UDP header 1083 */ 1084 uh = udp_hdr(skb); 1085 uh->source = fl6->fl6_sport; 1086 uh->dest = fl6->fl6_dport; 1087 uh->len = htons(len); 1088 uh->check = 0; 1089 1090 if (is_udplite) 1091 csum = udplite_csum(skb); 1092 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1093 skb->ip_summed = CHECKSUM_NONE; 1094 goto send; 1095 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1096 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1097 goto send; 1098 } else 1099 csum = udp_csum(skb); 1100 1101 /* add protocol-dependent pseudo-header */ 1102 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1103 len, fl6->flowi6_proto, csum); 1104 if (uh->check == 0) 1105 uh->check = CSUM_MANGLED_0; 1106 1107 send: 1108 err = ip6_send_skb(skb); 1109 if (err) { 1110 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1111 UDP6_INC_STATS_USER(sock_net(sk), 1112 UDP_MIB_SNDBUFERRORS, is_udplite); 1113 err = 0; 1114 } 1115 } else 1116 UDP6_INC_STATS_USER(sock_net(sk), 1117 UDP_MIB_OUTDATAGRAMS, is_udplite); 1118 return err; 1119 } 1120 1121 static int udp_v6_push_pending_frames(struct sock *sk) 1122 { 1123 struct sk_buff *skb; 1124 struct udp_sock *up = udp_sk(sk); 1125 struct flowi6 fl6; 1126 int err = 0; 1127 1128 if (up->pending == AF_INET) 1129 return udp_push_pending_frames(sk); 1130 1131 /* ip6_finish_skb will release the cork, so make a copy of 1132 * fl6 here. 1133 */ 1134 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1135 1136 skb = ip6_finish_skb(sk); 1137 if (!skb) 1138 goto out; 1139 1140 err = udp_v6_send_skb(skb, &fl6); 1141 1142 out: 1143 up->len = 0; 1144 up->pending = 0; 1145 return err; 1146 } 1147 1148 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1149 { 1150 struct ipv6_txoptions opt_space; 1151 struct udp_sock *up = udp_sk(sk); 1152 struct inet_sock *inet = inet_sk(sk); 1153 struct ipv6_pinfo *np = inet6_sk(sk); 1154 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1155 struct in6_addr *daddr, *final_p, final; 1156 struct ipv6_txoptions *opt = NULL; 1157 struct ipv6_txoptions *opt_to_free = NULL; 1158 struct ip6_flowlabel *flowlabel = NULL; 1159 struct flowi6 fl6; 1160 struct dst_entry *dst; 1161 int addr_len = msg->msg_namelen; 1162 int ulen = len; 1163 int hlimit = -1; 1164 int tclass = -1; 1165 int dontfrag = -1; 1166 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1167 int err; 1168 int connected = 0; 1169 int is_udplite = IS_UDPLITE(sk); 1170 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1171 1172 /* destination address check */ 1173 if (sin6) { 1174 if (addr_len < offsetof(struct sockaddr, sa_data)) 1175 return -EINVAL; 1176 1177 switch (sin6->sin6_family) { 1178 case AF_INET6: 1179 if (addr_len < SIN6_LEN_RFC2133) 1180 return -EINVAL; 1181 daddr = &sin6->sin6_addr; 1182 break; 1183 case AF_INET: 1184 goto do_udp_sendmsg; 1185 case AF_UNSPEC: 1186 msg->msg_name = sin6 = NULL; 1187 msg->msg_namelen = addr_len = 0; 1188 daddr = NULL; 1189 break; 1190 default: 1191 return -EINVAL; 1192 } 1193 } else if (!up->pending) { 1194 if (sk->sk_state != TCP_ESTABLISHED) 1195 return -EDESTADDRREQ; 1196 daddr = &sk->sk_v6_daddr; 1197 } else 1198 daddr = NULL; 1199 1200 if (daddr) { 1201 if (ipv6_addr_v4mapped(daddr)) { 1202 struct sockaddr_in sin; 1203 sin.sin_family = AF_INET; 1204 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1205 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1206 msg->msg_name = &sin; 1207 msg->msg_namelen = sizeof(sin); 1208 do_udp_sendmsg: 1209 if (__ipv6_only_sock(sk)) 1210 return -ENETUNREACH; 1211 return udp_sendmsg(sk, msg, len); 1212 } 1213 } 1214 1215 if (up->pending == AF_INET) 1216 return udp_sendmsg(sk, msg, len); 1217 1218 /* Rough check on arithmetic overflow, 1219 better check is made in ip6_append_data(). 1220 */ 1221 if (len > INT_MAX - sizeof(struct udphdr)) 1222 return -EMSGSIZE; 1223 1224 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1225 if (up->pending) { 1226 /* 1227 * There are pending frames. 1228 * The socket lock must be held while it's corked. 1229 */ 1230 lock_sock(sk); 1231 if (likely(up->pending)) { 1232 if (unlikely(up->pending != AF_INET6)) { 1233 release_sock(sk); 1234 return -EAFNOSUPPORT; 1235 } 1236 dst = NULL; 1237 goto do_append_data; 1238 } 1239 release_sock(sk); 1240 } 1241 ulen += sizeof(struct udphdr); 1242 1243 memset(&fl6, 0, sizeof(fl6)); 1244 1245 if (sin6) { 1246 if (sin6->sin6_port == 0) 1247 return -EINVAL; 1248 1249 fl6.fl6_dport = sin6->sin6_port; 1250 daddr = &sin6->sin6_addr; 1251 1252 if (np->sndflow) { 1253 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1254 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1255 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1256 if (!flowlabel) 1257 return -EINVAL; 1258 } 1259 } 1260 1261 /* 1262 * Otherwise it will be difficult to maintain 1263 * sk->sk_dst_cache. 1264 */ 1265 if (sk->sk_state == TCP_ESTABLISHED && 1266 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1267 daddr = &sk->sk_v6_daddr; 1268 1269 if (addr_len >= sizeof(struct sockaddr_in6) && 1270 sin6->sin6_scope_id && 1271 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1272 fl6.flowi6_oif = sin6->sin6_scope_id; 1273 } else { 1274 if (sk->sk_state != TCP_ESTABLISHED) 1275 return -EDESTADDRREQ; 1276 1277 fl6.fl6_dport = inet->inet_dport; 1278 daddr = &sk->sk_v6_daddr; 1279 fl6.flowlabel = np->flow_label; 1280 connected = 1; 1281 } 1282 1283 if (!fl6.flowi6_oif) 1284 fl6.flowi6_oif = sk->sk_bound_dev_if; 1285 1286 if (!fl6.flowi6_oif) 1287 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1288 1289 fl6.flowi6_mark = sk->sk_mark; 1290 1291 if (msg->msg_controllen) { 1292 opt = &opt_space; 1293 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1294 opt->tot_len = sizeof(*opt); 1295 1296 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt, 1297 &hlimit, &tclass, &dontfrag); 1298 if (err < 0) { 1299 fl6_sock_release(flowlabel); 1300 return err; 1301 } 1302 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1303 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1304 if (!flowlabel) 1305 return -EINVAL; 1306 } 1307 if (!(opt->opt_nflen|opt->opt_flen)) 1308 opt = NULL; 1309 connected = 0; 1310 } 1311 if (!opt) { 1312 opt = txopt_get(np); 1313 opt_to_free = opt; 1314 } 1315 if (flowlabel) 1316 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1317 opt = ipv6_fixup_options(&opt_space, opt); 1318 1319 fl6.flowi6_proto = sk->sk_protocol; 1320 if (!ipv6_addr_any(daddr)) 1321 fl6.daddr = *daddr; 1322 else 1323 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1324 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1325 fl6.saddr = np->saddr; 1326 fl6.fl6_sport = inet->inet_sport; 1327 1328 final_p = fl6_update_dst(&fl6, opt, &final); 1329 if (final_p) 1330 connected = 0; 1331 1332 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1333 fl6.flowi6_oif = np->mcast_oif; 1334 connected = 0; 1335 } else if (!fl6.flowi6_oif) 1336 fl6.flowi6_oif = np->ucast_oif; 1337 1338 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1339 1340 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p); 1341 if (IS_ERR(dst)) { 1342 err = PTR_ERR(dst); 1343 dst = NULL; 1344 goto out; 1345 } 1346 1347 if (hlimit < 0) 1348 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1349 1350 if (tclass < 0) 1351 tclass = np->tclass; 1352 1353 if (msg->msg_flags&MSG_CONFIRM) 1354 goto do_confirm; 1355 back_from_confirm: 1356 1357 /* Lockless fast path for the non-corking case */ 1358 if (!corkreq) { 1359 struct sk_buff *skb; 1360 1361 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1362 sizeof(struct udphdr), hlimit, tclass, opt, 1363 &fl6, (struct rt6_info *)dst, 1364 msg->msg_flags, dontfrag); 1365 err = PTR_ERR(skb); 1366 if (!IS_ERR_OR_NULL(skb)) 1367 err = udp_v6_send_skb(skb, &fl6); 1368 goto release_dst; 1369 } 1370 1371 lock_sock(sk); 1372 if (unlikely(up->pending)) { 1373 /* The socket is already corked while preparing it. */ 1374 /* ... which is an evident application bug. --ANK */ 1375 release_sock(sk); 1376 1377 net_dbg_ratelimited("udp cork app bug 2\n"); 1378 err = -EINVAL; 1379 goto out; 1380 } 1381 1382 up->pending = AF_INET6; 1383 1384 do_append_data: 1385 if (dontfrag < 0) 1386 dontfrag = np->dontfrag; 1387 up->len += ulen; 1388 err = ip6_append_data(sk, getfrag, msg, ulen, 1389 sizeof(struct udphdr), hlimit, tclass, opt, &fl6, 1390 (struct rt6_info *)dst, 1391 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag); 1392 if (err) 1393 udp_v6_flush_pending_frames(sk); 1394 else if (!corkreq) 1395 err = udp_v6_push_pending_frames(sk); 1396 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1397 up->pending = 0; 1398 1399 if (err > 0) 1400 err = np->recverr ? net_xmit_errno(err) : 0; 1401 release_sock(sk); 1402 1403 release_dst: 1404 if (dst) { 1405 if (connected) { 1406 ip6_dst_store(sk, dst, 1407 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ? 1408 &sk->sk_v6_daddr : NULL, 1409 #ifdef CONFIG_IPV6_SUBTREES 1410 ipv6_addr_equal(&fl6.saddr, &np->saddr) ? 1411 &np->saddr : 1412 #endif 1413 NULL); 1414 } else { 1415 dst_release(dst); 1416 } 1417 dst = NULL; 1418 } 1419 1420 out: 1421 dst_release(dst); 1422 fl6_sock_release(flowlabel); 1423 txopt_put(opt_to_free); 1424 if (!err) 1425 return len; 1426 /* 1427 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1428 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1429 * we don't have a good statistic (IpOutDiscards but it can be too many 1430 * things). We could add another new stat but at least for now that 1431 * seems like overkill. 1432 */ 1433 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1434 UDP6_INC_STATS_USER(sock_net(sk), 1435 UDP_MIB_SNDBUFERRORS, is_udplite); 1436 } 1437 return err; 1438 1439 do_confirm: 1440 dst_confirm(dst); 1441 if (!(msg->msg_flags&MSG_PROBE) || len) 1442 goto back_from_confirm; 1443 err = 0; 1444 goto out; 1445 } 1446 1447 void udpv6_destroy_sock(struct sock *sk) 1448 { 1449 struct udp_sock *up = udp_sk(sk); 1450 lock_sock(sk); 1451 udp_v6_flush_pending_frames(sk); 1452 release_sock(sk); 1453 1454 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 1455 void (*encap_destroy)(struct sock *sk); 1456 encap_destroy = ACCESS_ONCE(up->encap_destroy); 1457 if (encap_destroy) 1458 encap_destroy(sk); 1459 } 1460 1461 inet6_destroy_sock(sk); 1462 } 1463 1464 /* 1465 * Socket option code for UDP 1466 */ 1467 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1468 char __user *optval, unsigned int optlen) 1469 { 1470 if (level == SOL_UDP || level == SOL_UDPLITE) 1471 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1472 udp_v6_push_pending_frames); 1473 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1474 } 1475 1476 #ifdef CONFIG_COMPAT 1477 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1478 char __user *optval, unsigned int optlen) 1479 { 1480 if (level == SOL_UDP || level == SOL_UDPLITE) 1481 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1482 udp_v6_push_pending_frames); 1483 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1484 } 1485 #endif 1486 1487 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1488 char __user *optval, int __user *optlen) 1489 { 1490 if (level == SOL_UDP || level == SOL_UDPLITE) 1491 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1492 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1493 } 1494 1495 #ifdef CONFIG_COMPAT 1496 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1497 char __user *optval, int __user *optlen) 1498 { 1499 if (level == SOL_UDP || level == SOL_UDPLITE) 1500 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1501 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1502 } 1503 #endif 1504 1505 static const struct inet6_protocol udpv6_protocol = { 1506 .handler = udpv6_rcv, 1507 .err_handler = udpv6_err, 1508 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1509 }; 1510 1511 /* ------------------------------------------------------------------------ */ 1512 #ifdef CONFIG_PROC_FS 1513 int udp6_seq_show(struct seq_file *seq, void *v) 1514 { 1515 if (v == SEQ_START_TOKEN) { 1516 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1517 } else { 1518 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1519 struct inet_sock *inet = inet_sk(v); 1520 __u16 srcp = ntohs(inet->inet_sport); 1521 __u16 destp = ntohs(inet->inet_dport); 1522 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket); 1523 } 1524 return 0; 1525 } 1526 1527 static const struct file_operations udp6_afinfo_seq_fops = { 1528 .owner = THIS_MODULE, 1529 .open = udp_seq_open, 1530 .read = seq_read, 1531 .llseek = seq_lseek, 1532 .release = seq_release_net 1533 }; 1534 1535 static struct udp_seq_afinfo udp6_seq_afinfo = { 1536 .name = "udp6", 1537 .family = AF_INET6, 1538 .udp_table = &udp_table, 1539 .seq_fops = &udp6_afinfo_seq_fops, 1540 .seq_ops = { 1541 .show = udp6_seq_show, 1542 }, 1543 }; 1544 1545 int __net_init udp6_proc_init(struct net *net) 1546 { 1547 return udp_proc_register(net, &udp6_seq_afinfo); 1548 } 1549 1550 void udp6_proc_exit(struct net *net) 1551 { 1552 udp_proc_unregister(net, &udp6_seq_afinfo); 1553 } 1554 #endif /* CONFIG_PROC_FS */ 1555 1556 void udp_v6_clear_sk(struct sock *sk, int size) 1557 { 1558 struct inet_sock *inet = inet_sk(sk); 1559 1560 /* we do not want to clear pinet6 field, because of RCU lookups */ 1561 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6)); 1562 1563 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6); 1564 memset(&inet->pinet6 + 1, 0, size); 1565 } 1566 1567 /* ------------------------------------------------------------------------ */ 1568 1569 struct proto udpv6_prot = { 1570 .name = "UDPv6", 1571 .owner = THIS_MODULE, 1572 .close = udp_lib_close, 1573 .connect = ip6_datagram_connect, 1574 .disconnect = udp_disconnect, 1575 .ioctl = udp_ioctl, 1576 .destroy = udpv6_destroy_sock, 1577 .setsockopt = udpv6_setsockopt, 1578 .getsockopt = udpv6_getsockopt, 1579 .sendmsg = udpv6_sendmsg, 1580 .recvmsg = udpv6_recvmsg, 1581 .backlog_rcv = __udpv6_queue_rcv_skb, 1582 .hash = udp_lib_hash, 1583 .unhash = udp_lib_unhash, 1584 .rehash = udp_v6_rehash, 1585 .get_port = udp_v6_get_port, 1586 .memory_allocated = &udp_memory_allocated, 1587 .sysctl_mem = sysctl_udp_mem, 1588 .sysctl_wmem = &sysctl_udp_wmem_min, 1589 .sysctl_rmem = &sysctl_udp_rmem_min, 1590 .obj_size = sizeof(struct udp6_sock), 1591 .slab_flags = SLAB_DESTROY_BY_RCU, 1592 .h.udp_table = &udp_table, 1593 #ifdef CONFIG_COMPAT 1594 .compat_setsockopt = compat_udpv6_setsockopt, 1595 .compat_getsockopt = compat_udpv6_getsockopt, 1596 #endif 1597 .clear_sk = udp_v6_clear_sk, 1598 }; 1599 1600 static struct inet_protosw udpv6_protosw = { 1601 .type = SOCK_DGRAM, 1602 .protocol = IPPROTO_UDP, 1603 .prot = &udpv6_prot, 1604 .ops = &inet6_dgram_ops, 1605 .flags = INET_PROTOSW_PERMANENT, 1606 }; 1607 1608 int __init udpv6_init(void) 1609 { 1610 int ret; 1611 1612 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1613 if (ret) 1614 goto out; 1615 1616 ret = inet6_register_protosw(&udpv6_protosw); 1617 if (ret) 1618 goto out_udpv6_protocol; 1619 out: 1620 return ret; 1621 1622 out_udpv6_protocol: 1623 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1624 goto out; 1625 } 1626 1627 void udpv6_exit(void) 1628 { 1629 inet6_unregister_protosw(&udpv6_protosw); 1630 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1631 } 1632