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 51 #include <linux/proc_fs.h> 52 #include <linux/seq_file.h> 53 #include <trace/events/skb.h> 54 #include "udp_impl.h" 55 56 static unsigned int udp6_ehashfn(struct net *net, 57 const struct in6_addr *laddr, 58 const u16 lport, 59 const struct in6_addr *faddr, 60 const __be16 fport) 61 { 62 static u32 udp6_ehash_secret __read_mostly; 63 static u32 udp_ipv6_hash_secret __read_mostly; 64 65 u32 lhash, fhash; 66 67 net_get_random_once(&udp6_ehash_secret, 68 sizeof(udp6_ehash_secret)); 69 net_get_random_once(&udp_ipv6_hash_secret, 70 sizeof(udp_ipv6_hash_secret)); 71 72 lhash = (__force u32)laddr->s6_addr32[3]; 73 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 74 75 return __inet6_ehashfn(lhash, lport, fhash, fport, 76 udp_ipv6_hash_secret + net_hash_mix(net)); 77 } 78 79 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2) 80 { 81 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2); 82 int sk2_ipv6only = inet_v6_ipv6only(sk2); 83 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr); 84 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED; 85 86 /* if both are mapped, treat as IPv4 */ 87 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) 88 return (!sk2_ipv6only && 89 (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr || 90 sk->sk_rcv_saddr == sk2->sk_rcv_saddr)); 91 92 if (addr_type2 == IPV6_ADDR_ANY && 93 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED)) 94 return 1; 95 96 if (addr_type == IPV6_ADDR_ANY && 97 !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED)) 98 return 1; 99 100 if (sk2_rcv_saddr6 && 101 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6)) 102 return 1; 103 104 return 0; 105 } 106 107 static unsigned int udp6_portaddr_hash(struct net *net, 108 const struct in6_addr *addr6, 109 unsigned int port) 110 { 111 unsigned int hash, mix = net_hash_mix(net); 112 113 if (ipv6_addr_any(addr6)) 114 hash = jhash_1word(0, mix); 115 else if (ipv6_addr_v4mapped(addr6)) 116 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix); 117 else 118 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix); 119 120 return hash ^ port; 121 } 122 123 124 int udp_v6_get_port(struct sock *sk, unsigned short snum) 125 { 126 unsigned int hash2_nulladdr = 127 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 128 unsigned int hash2_partial = 129 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 130 131 /* precompute partial secondary hash */ 132 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 133 return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr); 134 } 135 136 static void udp_v6_rehash(struct sock *sk) 137 { 138 u16 new_hash = udp6_portaddr_hash(sock_net(sk), 139 &sk->sk_v6_rcv_saddr, 140 inet_sk(sk)->inet_num); 141 142 udp_lib_rehash(sk, new_hash); 143 } 144 145 static inline int compute_score(struct sock *sk, struct net *net, 146 unsigned short hnum, 147 const struct in6_addr *saddr, __be16 sport, 148 const struct in6_addr *daddr, __be16 dport, 149 int dif) 150 { 151 int score = -1; 152 153 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 154 sk->sk_family == PF_INET6) { 155 struct inet_sock *inet = inet_sk(sk); 156 157 score = 0; 158 if (inet->inet_dport) { 159 if (inet->inet_dport != sport) 160 return -1; 161 score++; 162 } 163 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { 164 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 165 return -1; 166 score++; 167 } 168 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 169 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 170 return -1; 171 score++; 172 } 173 if (sk->sk_bound_dev_if) { 174 if (sk->sk_bound_dev_if != dif) 175 return -1; 176 score++; 177 } 178 } 179 return score; 180 } 181 182 #define SCORE2_MAX (1 + 1 + 1) 183 static inline int compute_score2(struct sock *sk, struct net *net, 184 const struct in6_addr *saddr, __be16 sport, 185 const struct in6_addr *daddr, unsigned short hnum, 186 int dif) 187 { 188 int score = -1; 189 190 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 191 sk->sk_family == PF_INET6) { 192 struct inet_sock *inet = inet_sk(sk); 193 194 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 195 return -1; 196 score = 0; 197 if (inet->inet_dport) { 198 if (inet->inet_dport != sport) 199 return -1; 200 score++; 201 } 202 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 203 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 204 return -1; 205 score++; 206 } 207 if (sk->sk_bound_dev_if) { 208 if (sk->sk_bound_dev_if != dif) 209 return -1; 210 score++; 211 } 212 } 213 return score; 214 } 215 216 217 /* called with read_rcu_lock() */ 218 static struct sock *udp6_lib_lookup2(struct net *net, 219 const struct in6_addr *saddr, __be16 sport, 220 const struct in6_addr *daddr, unsigned int hnum, int dif, 221 struct udp_hslot *hslot2, unsigned int slot2) 222 { 223 struct sock *sk, *result; 224 struct hlist_nulls_node *node; 225 int score, badness, matches = 0, reuseport = 0; 226 u32 hash = 0; 227 228 begin: 229 result = NULL; 230 badness = -1; 231 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 232 score = compute_score2(sk, net, saddr, sport, 233 daddr, hnum, dif); 234 if (score > badness) { 235 result = sk; 236 badness = score; 237 reuseport = sk->sk_reuseport; 238 if (reuseport) { 239 hash = udp6_ehashfn(net, daddr, hnum, 240 saddr, sport); 241 matches = 1; 242 } else if (score == SCORE2_MAX) 243 goto exact_match; 244 } else if (score == badness && reuseport) { 245 matches++; 246 if (reciprocal_scale(hash, matches) == 0) 247 result = sk; 248 hash = next_pseudo_random32(hash); 249 } 250 } 251 /* 252 * if the nulls value we got at the end of this lookup is 253 * not the expected one, we must restart lookup. 254 * We probably met an item that was moved to another chain. 255 */ 256 if (get_nulls_value(node) != slot2) 257 goto begin; 258 259 if (result) { 260 exact_match: 261 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 262 result = NULL; 263 else if (unlikely(compute_score2(result, net, saddr, sport, 264 daddr, hnum, dif) < badness)) { 265 sock_put(result); 266 goto begin; 267 } 268 } 269 return result; 270 } 271 272 struct sock *__udp6_lib_lookup(struct net *net, 273 const struct in6_addr *saddr, __be16 sport, 274 const struct in6_addr *daddr, __be16 dport, 275 int dif, struct udp_table *udptable) 276 { 277 struct sock *sk, *result; 278 struct hlist_nulls_node *node; 279 unsigned short hnum = ntohs(dport); 280 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 281 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 282 int score, badness, matches = 0, reuseport = 0; 283 u32 hash = 0; 284 285 rcu_read_lock(); 286 if (hslot->count > 10) { 287 hash2 = udp6_portaddr_hash(net, daddr, hnum); 288 slot2 = hash2 & udptable->mask; 289 hslot2 = &udptable->hash2[slot2]; 290 if (hslot->count < hslot2->count) 291 goto begin; 292 293 result = udp6_lib_lookup2(net, saddr, sport, 294 daddr, hnum, dif, 295 hslot2, slot2); 296 if (!result) { 297 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum); 298 slot2 = hash2 & udptable->mask; 299 hslot2 = &udptable->hash2[slot2]; 300 if (hslot->count < hslot2->count) 301 goto begin; 302 303 result = udp6_lib_lookup2(net, saddr, sport, 304 &in6addr_any, hnum, dif, 305 hslot2, slot2); 306 } 307 rcu_read_unlock(); 308 return result; 309 } 310 begin: 311 result = NULL; 312 badness = -1; 313 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 314 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif); 315 if (score > badness) { 316 result = sk; 317 badness = score; 318 reuseport = sk->sk_reuseport; 319 if (reuseport) { 320 hash = udp6_ehashfn(net, daddr, hnum, 321 saddr, sport); 322 matches = 1; 323 } 324 } else if (score == badness && reuseport) { 325 matches++; 326 if (reciprocal_scale(hash, matches) == 0) 327 result = sk; 328 hash = next_pseudo_random32(hash); 329 } 330 } 331 /* 332 * if the nulls value we got at the end of this lookup is 333 * not the expected one, we must restart lookup. 334 * We probably met an item that was moved to another chain. 335 */ 336 if (get_nulls_value(node) != slot) 337 goto begin; 338 339 if (result) { 340 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 341 result = NULL; 342 else if (unlikely(compute_score(result, net, hnum, saddr, sport, 343 daddr, dport, dif) < badness)) { 344 sock_put(result); 345 goto begin; 346 } 347 } 348 rcu_read_unlock(); 349 return result; 350 } 351 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 352 353 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 354 __be16 sport, __be16 dport, 355 struct udp_table *udptable) 356 { 357 struct sock *sk; 358 const struct ipv6hdr *iph = ipv6_hdr(skb); 359 360 if (unlikely(sk = skb_steal_sock(skb))) 361 return sk; 362 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport, 363 &iph->daddr, dport, inet6_iif(skb), 364 udptable); 365 } 366 367 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 368 const struct in6_addr *daddr, __be16 dport, int dif) 369 { 370 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 371 } 372 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 373 374 375 /* 376 * This should be easy, if there is something there we 377 * return it, otherwise we block. 378 */ 379 380 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk, 381 struct msghdr *msg, size_t len, 382 int noblock, int flags, int *addr_len) 383 { 384 struct ipv6_pinfo *np = inet6_sk(sk); 385 struct inet_sock *inet = inet_sk(sk); 386 struct sk_buff *skb; 387 unsigned int ulen, copied; 388 int peeked, off = 0; 389 int err; 390 int is_udplite = IS_UDPLITE(sk); 391 int is_udp4; 392 bool slow; 393 394 if (flags & MSG_ERRQUEUE) 395 return ipv6_recv_error(sk, msg, len, addr_len); 396 397 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 398 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 399 400 try_again: 401 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 402 &peeked, &off, &err); 403 if (!skb) 404 goto out; 405 406 ulen = skb->len - sizeof(struct udphdr); 407 copied = len; 408 if (copied > ulen) 409 copied = ulen; 410 else if (copied < ulen) 411 msg->msg_flags |= MSG_TRUNC; 412 413 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 414 415 /* 416 * If checksum is needed at all, try to do it while copying the 417 * data. If the data is truncated, or if we only want a partial 418 * coverage checksum (UDP-Lite), do it before the copy. 419 */ 420 421 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 422 if (udp_lib_checksum_complete(skb)) 423 goto csum_copy_err; 424 } 425 426 if (skb_csum_unnecessary(skb)) 427 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 428 msg->msg_iov, copied); 429 else { 430 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 431 if (err == -EINVAL) 432 goto csum_copy_err; 433 } 434 if (unlikely(err)) { 435 trace_kfree_skb(skb, udpv6_recvmsg); 436 if (!peeked) { 437 atomic_inc(&sk->sk_drops); 438 if (is_udp4) 439 UDP_INC_STATS_USER(sock_net(sk), 440 UDP_MIB_INERRORS, 441 is_udplite); 442 else 443 UDP6_INC_STATS_USER(sock_net(sk), 444 UDP_MIB_INERRORS, 445 is_udplite); 446 } 447 goto out_free; 448 } 449 if (!peeked) { 450 if (is_udp4) 451 UDP_INC_STATS_USER(sock_net(sk), 452 UDP_MIB_INDATAGRAMS, is_udplite); 453 else 454 UDP6_INC_STATS_USER(sock_net(sk), 455 UDP_MIB_INDATAGRAMS, is_udplite); 456 } 457 458 sock_recv_ts_and_drops(msg, sk, skb); 459 460 /* Copy the address. */ 461 if (msg->msg_name) { 462 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 463 sin6->sin6_family = AF_INET6; 464 sin6->sin6_port = udp_hdr(skb)->source; 465 sin6->sin6_flowinfo = 0; 466 467 if (is_udp4) { 468 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 469 &sin6->sin6_addr); 470 sin6->sin6_scope_id = 0; 471 } else { 472 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 473 sin6->sin6_scope_id = 474 ipv6_iface_scope_id(&sin6->sin6_addr, 475 inet6_iif(skb)); 476 } 477 *addr_len = sizeof(*sin6); 478 } 479 480 if (np->rxopt.all) 481 ip6_datagram_recv_common_ctl(sk, msg, skb); 482 483 if (is_udp4) { 484 if (inet->cmsg_flags) 485 ip_cmsg_recv(msg, skb); 486 } else { 487 if (np->rxopt.all) 488 ip6_datagram_recv_specific_ctl(sk, msg, skb); 489 } 490 491 err = copied; 492 if (flags & MSG_TRUNC) 493 err = ulen; 494 495 out_free: 496 skb_free_datagram_locked(sk, skb); 497 out: 498 return err; 499 500 csum_copy_err: 501 slow = lock_sock_fast(sk); 502 if (!skb_kill_datagram(sk, skb, flags)) { 503 if (is_udp4) { 504 UDP_INC_STATS_USER(sock_net(sk), 505 UDP_MIB_CSUMERRORS, is_udplite); 506 UDP_INC_STATS_USER(sock_net(sk), 507 UDP_MIB_INERRORS, is_udplite); 508 } else { 509 UDP6_INC_STATS_USER(sock_net(sk), 510 UDP_MIB_CSUMERRORS, is_udplite); 511 UDP6_INC_STATS_USER(sock_net(sk), 512 UDP_MIB_INERRORS, is_udplite); 513 } 514 } 515 unlock_sock_fast(sk, slow); 516 517 if (noblock) 518 return -EAGAIN; 519 520 /* starting over for a new packet */ 521 msg->msg_flags &= ~MSG_TRUNC; 522 goto try_again; 523 } 524 525 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 526 u8 type, u8 code, int offset, __be32 info, 527 struct udp_table *udptable) 528 { 529 struct ipv6_pinfo *np; 530 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 531 const struct in6_addr *saddr = &hdr->saddr; 532 const struct in6_addr *daddr = &hdr->daddr; 533 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 534 struct sock *sk; 535 int err; 536 struct net *net = dev_net(skb->dev); 537 538 sk = __udp6_lib_lookup(net, daddr, uh->dest, 539 saddr, uh->source, inet6_iif(skb), udptable); 540 if (sk == NULL) { 541 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), 542 ICMP6_MIB_INERRORS); 543 return; 544 } 545 546 if (type == ICMPV6_PKT_TOOBIG) { 547 if (!ip6_sk_accept_pmtu(sk)) 548 goto out; 549 ip6_sk_update_pmtu(skb, sk, info); 550 } 551 if (type == NDISC_REDIRECT) { 552 ip6_sk_redirect(skb, sk); 553 goto out; 554 } 555 556 np = inet6_sk(sk); 557 558 if (!icmpv6_err_convert(type, code, &err) && !np->recverr) 559 goto out; 560 561 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr) 562 goto out; 563 564 if (np->recverr) 565 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 566 567 sk->sk_err = err; 568 sk->sk_error_report(sk); 569 out: 570 sock_put(sk); 571 } 572 573 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 574 { 575 int rc; 576 577 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 578 sock_rps_save_rxhash(sk, skb); 579 sk_mark_napi_id(sk, skb); 580 } 581 582 rc = sock_queue_rcv_skb(sk, skb); 583 if (rc < 0) { 584 int is_udplite = IS_UDPLITE(sk); 585 586 /* Note that an ENOMEM error is charged twice */ 587 if (rc == -ENOMEM) 588 UDP6_INC_STATS_BH(sock_net(sk), 589 UDP_MIB_RCVBUFERRORS, is_udplite); 590 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 591 kfree_skb(skb); 592 return -1; 593 } 594 return 0; 595 } 596 597 static __inline__ void udpv6_err(struct sk_buff *skb, 598 struct inet6_skb_parm *opt, u8 type, 599 u8 code, int offset, __be32 info) 600 { 601 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 602 } 603 604 static struct static_key udpv6_encap_needed __read_mostly; 605 void udpv6_encap_enable(void) 606 { 607 if (!static_key_enabled(&udpv6_encap_needed)) 608 static_key_slow_inc(&udpv6_encap_needed); 609 } 610 EXPORT_SYMBOL(udpv6_encap_enable); 611 612 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 613 { 614 struct udp_sock *up = udp_sk(sk); 615 int rc; 616 int is_udplite = IS_UDPLITE(sk); 617 618 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 619 goto drop; 620 621 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 622 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 623 624 /* 625 * This is an encapsulation socket so pass the skb to 626 * the socket's udp_encap_rcv() hook. Otherwise, just 627 * fall through and pass this up the UDP socket. 628 * up->encap_rcv() returns the following value: 629 * =0 if skb was successfully passed to the encap 630 * handler or was discarded by it. 631 * >0 if skb should be passed on to UDP. 632 * <0 if skb should be resubmitted as proto -N 633 */ 634 635 /* if we're overly short, let UDP handle it */ 636 encap_rcv = ACCESS_ONCE(up->encap_rcv); 637 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { 638 int ret; 639 640 /* Verify checksum before giving to encap */ 641 if (udp_lib_checksum_complete(skb)) 642 goto csum_error; 643 644 ret = encap_rcv(sk, skb); 645 if (ret <= 0) { 646 UDP_INC_STATS_BH(sock_net(sk), 647 UDP_MIB_INDATAGRAMS, 648 is_udplite); 649 return -ret; 650 } 651 } 652 653 /* FALLTHROUGH -- it's a UDP Packet */ 654 } 655 656 /* 657 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 658 */ 659 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 660 661 if (up->pcrlen == 0) { /* full coverage was set */ 662 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage" 663 " %d while full coverage %d requested\n", 664 UDP_SKB_CB(skb)->cscov, skb->len); 665 goto drop; 666 } 667 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 668 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d " 669 "too small, need min %d\n", 670 UDP_SKB_CB(skb)->cscov, up->pcrlen); 671 goto drop; 672 } 673 } 674 675 if (rcu_access_pointer(sk->sk_filter)) { 676 if (udp_lib_checksum_complete(skb)) 677 goto csum_error; 678 } 679 680 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 681 UDP6_INC_STATS_BH(sock_net(sk), 682 UDP_MIB_RCVBUFERRORS, is_udplite); 683 goto drop; 684 } 685 686 skb_dst_drop(skb); 687 688 bh_lock_sock(sk); 689 rc = 0; 690 if (!sock_owned_by_user(sk)) 691 rc = __udpv6_queue_rcv_skb(sk, skb); 692 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 693 bh_unlock_sock(sk); 694 goto drop; 695 } 696 bh_unlock_sock(sk); 697 698 return rc; 699 700 csum_error: 701 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 702 drop: 703 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 704 atomic_inc(&sk->sk_drops); 705 kfree_skb(skb); 706 return -1; 707 } 708 709 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 710 __be16 loc_port, const struct in6_addr *loc_addr, 711 __be16 rmt_port, const struct in6_addr *rmt_addr, 712 int dif, unsigned short hnum) 713 { 714 struct inet_sock *inet = inet_sk(sk); 715 716 if (!net_eq(sock_net(sk), net)) 717 return false; 718 719 if (udp_sk(sk)->udp_port_hash != hnum || 720 sk->sk_family != PF_INET6 || 721 (inet->inet_dport && inet->inet_dport != rmt_port) || 722 (!ipv6_addr_any(&sk->sk_v6_daddr) && 723 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 724 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 725 return false; 726 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 727 return false; 728 return true; 729 } 730 731 static void flush_stack(struct sock **stack, unsigned int count, 732 struct sk_buff *skb, unsigned int final) 733 { 734 struct sk_buff *skb1 = NULL; 735 struct sock *sk; 736 unsigned int i; 737 738 for (i = 0; i < count; i++) { 739 sk = stack[i]; 740 if (likely(skb1 == NULL)) 741 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 742 if (!skb1) { 743 atomic_inc(&sk->sk_drops); 744 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 745 IS_UDPLITE(sk)); 746 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 747 IS_UDPLITE(sk)); 748 } 749 750 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0) 751 skb1 = NULL; 752 sock_put(sk); 753 } 754 if (unlikely(skb1)) 755 kfree_skb(skb1); 756 } 757 758 static void udp6_csum_zero_error(struct sk_buff *skb) 759 { 760 /* RFC 2460 section 8.1 says that we SHOULD log 761 * this error. Well, it is reasonable. 762 */ 763 LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 764 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 765 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 766 } 767 768 /* 769 * Note: called only from the BH handler context, 770 * so we don't need to lock the hashes. 771 */ 772 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 773 const struct in6_addr *saddr, const struct in6_addr *daddr, 774 struct udp_table *udptable) 775 { 776 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 777 const struct udphdr *uh = udp_hdr(skb); 778 struct hlist_nulls_node *node; 779 unsigned short hnum = ntohs(uh->dest); 780 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 781 int dif = inet6_iif(skb); 782 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 783 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 784 785 if (use_hash2) { 786 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) & 787 udp_table.mask; 788 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask; 789 start_lookup: 790 hslot = &udp_table.hash2[hash2]; 791 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 792 } 793 794 spin_lock(&hslot->lock); 795 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 796 if (__udp_v6_is_mcast_sock(net, sk, 797 uh->dest, daddr, 798 uh->source, saddr, 799 dif, hnum) && 800 /* If zero checksum and no_check is not on for 801 * the socket then skip it. 802 */ 803 (uh->check || udp_sk(sk)->no_check6_rx)) { 804 if (unlikely(count == ARRAY_SIZE(stack))) { 805 flush_stack(stack, count, skb, ~0); 806 count = 0; 807 } 808 stack[count++] = sk; 809 sock_hold(sk); 810 } 811 } 812 813 spin_unlock(&hslot->lock); 814 815 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 816 if (use_hash2 && hash2 != hash2_any) { 817 hash2 = hash2_any; 818 goto start_lookup; 819 } 820 821 if (count) { 822 flush_stack(stack, count, skb, count - 1); 823 } else { 824 kfree_skb(skb); 825 } 826 return 0; 827 } 828 829 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 830 int proto) 831 { 832 struct net *net = dev_net(skb->dev); 833 struct sock *sk; 834 struct udphdr *uh; 835 const struct in6_addr *saddr, *daddr; 836 u32 ulen = 0; 837 838 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 839 goto discard; 840 841 saddr = &ipv6_hdr(skb)->saddr; 842 daddr = &ipv6_hdr(skb)->daddr; 843 uh = udp_hdr(skb); 844 845 ulen = ntohs(uh->len); 846 if (ulen > skb->len) 847 goto short_packet; 848 849 if (proto == IPPROTO_UDP) { 850 /* UDP validates ulen. */ 851 852 /* Check for jumbo payload */ 853 if (ulen == 0) 854 ulen = skb->len; 855 856 if (ulen < sizeof(*uh)) 857 goto short_packet; 858 859 if (ulen < skb->len) { 860 if (pskb_trim_rcsum(skb, ulen)) 861 goto short_packet; 862 saddr = &ipv6_hdr(skb)->saddr; 863 daddr = &ipv6_hdr(skb)->daddr; 864 uh = udp_hdr(skb); 865 } 866 } 867 868 if (udp6_csum_init(skb, uh, proto)) 869 goto csum_error; 870 871 /* 872 * Multicast receive code 873 */ 874 if (ipv6_addr_is_multicast(daddr)) 875 return __udp6_lib_mcast_deliver(net, skb, 876 saddr, daddr, udptable); 877 878 /* Unicast */ 879 880 /* 881 * check socket cache ... must talk to Alan about his plans 882 * for sock caches... i'll skip this for now. 883 */ 884 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 885 if (sk != NULL) { 886 int ret; 887 888 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 889 sock_put(sk); 890 udp6_csum_zero_error(skb); 891 goto csum_error; 892 } 893 894 if (udp_sk(sk)->convert_csum && uh->check && !IS_UDPLITE(sk)) 895 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 896 ip6_compute_pseudo); 897 898 ret = udpv6_queue_rcv_skb(sk, skb); 899 sock_put(sk); 900 901 /* a return value > 0 means to resubmit the input, but 902 * it wants the return to be -protocol, or 0 903 */ 904 if (ret > 0) 905 return -ret; 906 907 return 0; 908 } 909 910 if (!uh->check) { 911 udp6_csum_zero_error(skb); 912 goto csum_error; 913 } 914 915 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 916 goto discard; 917 918 if (udp_lib_checksum_complete(skb)) 919 goto csum_error; 920 921 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 922 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 923 924 kfree_skb(skb); 925 return 0; 926 927 short_packet: 928 LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 929 proto == IPPROTO_UDPLITE ? "-Lite" : "", 930 saddr, 931 ntohs(uh->source), 932 ulen, 933 skb->len, 934 daddr, 935 ntohs(uh->dest)); 936 goto discard; 937 csum_error: 938 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 939 discard: 940 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 941 kfree_skb(skb); 942 return 0; 943 } 944 945 static __inline__ int udpv6_rcv(struct sk_buff *skb) 946 { 947 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 948 } 949 950 /* 951 * Throw away all pending data and cancel the corking. Socket is locked. 952 */ 953 static void udp_v6_flush_pending_frames(struct sock *sk) 954 { 955 struct udp_sock *up = udp_sk(sk); 956 957 if (up->pending == AF_INET) 958 udp_flush_pending_frames(sk); 959 else if (up->pending) { 960 up->len = 0; 961 up->pending = 0; 962 ip6_flush_pending_frames(sk); 963 } 964 } 965 966 /** 967 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 968 * @sk: socket we are sending on 969 * @skb: sk_buff containing the filled-in UDP header 970 * (checksum field must be zeroed out) 971 */ 972 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 973 const struct in6_addr *saddr, 974 const struct in6_addr *daddr, int len) 975 { 976 unsigned int offset; 977 struct udphdr *uh = udp_hdr(skb); 978 __wsum csum = 0; 979 980 if (skb_queue_len(&sk->sk_write_queue) == 1) { 981 /* Only one fragment on the socket. */ 982 skb->csum_start = skb_transport_header(skb) - skb->head; 983 skb->csum_offset = offsetof(struct udphdr, check); 984 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 985 } else { 986 /* 987 * HW-checksum won't work as there are two or more 988 * fragments on the socket so that all csums of sk_buffs 989 * should be together 990 */ 991 offset = skb_transport_offset(skb); 992 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 993 994 skb->ip_summed = CHECKSUM_NONE; 995 996 skb_queue_walk(&sk->sk_write_queue, skb) { 997 csum = csum_add(csum, skb->csum); 998 } 999 1000 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1001 csum); 1002 if (uh->check == 0) 1003 uh->check = CSUM_MANGLED_0; 1004 } 1005 } 1006 1007 /* 1008 * Sending 1009 */ 1010 1011 static int udp_v6_push_pending_frames(struct sock *sk) 1012 { 1013 struct sk_buff *skb; 1014 struct udphdr *uh; 1015 struct udp_sock *up = udp_sk(sk); 1016 struct inet_sock *inet = inet_sk(sk); 1017 struct flowi6 *fl6; 1018 int err = 0; 1019 int is_udplite = IS_UDPLITE(sk); 1020 __wsum csum = 0; 1021 1022 if (up->pending == AF_INET) 1023 return udp_push_pending_frames(sk); 1024 1025 fl6 = &inet->cork.fl.u.ip6; 1026 1027 /* Grab the skbuff where UDP header space exists. */ 1028 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 1029 goto out; 1030 1031 /* 1032 * Create a UDP header 1033 */ 1034 uh = udp_hdr(skb); 1035 uh->source = fl6->fl6_sport; 1036 uh->dest = fl6->fl6_dport; 1037 uh->len = htons(up->len); 1038 uh->check = 0; 1039 1040 if (is_udplite) 1041 csum = udplite_csum_outgoing(sk, skb); 1042 else if (up->no_check6_tx) { /* UDP csum disabled */ 1043 skb->ip_summed = CHECKSUM_NONE; 1044 goto send; 1045 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1046 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, 1047 up->len); 1048 goto send; 1049 } else 1050 csum = udp_csum_outgoing(sk, skb); 1051 1052 /* add protocol-dependent pseudo-header */ 1053 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1054 up->len, fl6->flowi6_proto, csum); 1055 if (uh->check == 0) 1056 uh->check = CSUM_MANGLED_0; 1057 1058 send: 1059 err = ip6_push_pending_frames(sk); 1060 if (err) { 1061 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1062 UDP6_INC_STATS_USER(sock_net(sk), 1063 UDP_MIB_SNDBUFERRORS, is_udplite); 1064 err = 0; 1065 } 1066 } else 1067 UDP6_INC_STATS_USER(sock_net(sk), 1068 UDP_MIB_OUTDATAGRAMS, is_udplite); 1069 out: 1070 up->len = 0; 1071 up->pending = 0; 1072 return err; 1073 } 1074 1075 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk, 1076 struct msghdr *msg, size_t len) 1077 { 1078 struct ipv6_txoptions opt_space; 1079 struct udp_sock *up = udp_sk(sk); 1080 struct inet_sock *inet = inet_sk(sk); 1081 struct ipv6_pinfo *np = inet6_sk(sk); 1082 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1083 struct in6_addr *daddr, *final_p, final; 1084 struct ipv6_txoptions *opt = NULL; 1085 struct ip6_flowlabel *flowlabel = NULL; 1086 struct flowi6 fl6; 1087 struct dst_entry *dst; 1088 int addr_len = msg->msg_namelen; 1089 int ulen = len; 1090 int hlimit = -1; 1091 int tclass = -1; 1092 int dontfrag = -1; 1093 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1094 int err; 1095 int connected = 0; 1096 int is_udplite = IS_UDPLITE(sk); 1097 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1098 1099 /* destination address check */ 1100 if (sin6) { 1101 if (addr_len < offsetof(struct sockaddr, sa_data)) 1102 return -EINVAL; 1103 1104 switch (sin6->sin6_family) { 1105 case AF_INET6: 1106 if (addr_len < SIN6_LEN_RFC2133) 1107 return -EINVAL; 1108 daddr = &sin6->sin6_addr; 1109 break; 1110 case AF_INET: 1111 goto do_udp_sendmsg; 1112 case AF_UNSPEC: 1113 msg->msg_name = sin6 = NULL; 1114 msg->msg_namelen = addr_len = 0; 1115 daddr = NULL; 1116 break; 1117 default: 1118 return -EINVAL; 1119 } 1120 } else if (!up->pending) { 1121 if (sk->sk_state != TCP_ESTABLISHED) 1122 return -EDESTADDRREQ; 1123 daddr = &sk->sk_v6_daddr; 1124 } else 1125 daddr = NULL; 1126 1127 if (daddr) { 1128 if (ipv6_addr_v4mapped(daddr)) { 1129 struct sockaddr_in sin; 1130 sin.sin_family = AF_INET; 1131 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1132 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1133 msg->msg_name = &sin; 1134 msg->msg_namelen = sizeof(sin); 1135 do_udp_sendmsg: 1136 if (__ipv6_only_sock(sk)) 1137 return -ENETUNREACH; 1138 return udp_sendmsg(iocb, sk, msg, len); 1139 } 1140 } 1141 1142 if (up->pending == AF_INET) 1143 return udp_sendmsg(iocb, sk, msg, len); 1144 1145 /* Rough check on arithmetic overflow, 1146 better check is made in ip6_append_data(). 1147 */ 1148 if (len > INT_MAX - sizeof(struct udphdr)) 1149 return -EMSGSIZE; 1150 1151 if (up->pending) { 1152 /* 1153 * There are pending frames. 1154 * The socket lock must be held while it's corked. 1155 */ 1156 lock_sock(sk); 1157 if (likely(up->pending)) { 1158 if (unlikely(up->pending != AF_INET6)) { 1159 release_sock(sk); 1160 return -EAFNOSUPPORT; 1161 } 1162 dst = NULL; 1163 goto do_append_data; 1164 } 1165 release_sock(sk); 1166 } 1167 ulen += sizeof(struct udphdr); 1168 1169 memset(&fl6, 0, sizeof(fl6)); 1170 1171 if (sin6) { 1172 if (sin6->sin6_port == 0) 1173 return -EINVAL; 1174 1175 fl6.fl6_dport = sin6->sin6_port; 1176 daddr = &sin6->sin6_addr; 1177 1178 if (np->sndflow) { 1179 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1180 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1181 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1182 if (flowlabel == NULL) 1183 return -EINVAL; 1184 } 1185 } 1186 1187 /* 1188 * Otherwise it will be difficult to maintain 1189 * sk->sk_dst_cache. 1190 */ 1191 if (sk->sk_state == TCP_ESTABLISHED && 1192 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1193 daddr = &sk->sk_v6_daddr; 1194 1195 if (addr_len >= sizeof(struct sockaddr_in6) && 1196 sin6->sin6_scope_id && 1197 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1198 fl6.flowi6_oif = sin6->sin6_scope_id; 1199 } else { 1200 if (sk->sk_state != TCP_ESTABLISHED) 1201 return -EDESTADDRREQ; 1202 1203 fl6.fl6_dport = inet->inet_dport; 1204 daddr = &sk->sk_v6_daddr; 1205 fl6.flowlabel = np->flow_label; 1206 connected = 1; 1207 } 1208 1209 if (!fl6.flowi6_oif) 1210 fl6.flowi6_oif = sk->sk_bound_dev_if; 1211 1212 if (!fl6.flowi6_oif) 1213 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1214 1215 fl6.flowi6_mark = sk->sk_mark; 1216 1217 if (msg->msg_controllen) { 1218 opt = &opt_space; 1219 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1220 opt->tot_len = sizeof(*opt); 1221 1222 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt, 1223 &hlimit, &tclass, &dontfrag); 1224 if (err < 0) { 1225 fl6_sock_release(flowlabel); 1226 return err; 1227 } 1228 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1229 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1230 if (flowlabel == NULL) 1231 return -EINVAL; 1232 } 1233 if (!(opt->opt_nflen|opt->opt_flen)) 1234 opt = NULL; 1235 connected = 0; 1236 } 1237 if (opt == NULL) 1238 opt = np->opt; 1239 if (flowlabel) 1240 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1241 opt = ipv6_fixup_options(&opt_space, opt); 1242 1243 fl6.flowi6_proto = sk->sk_protocol; 1244 if (!ipv6_addr_any(daddr)) 1245 fl6.daddr = *daddr; 1246 else 1247 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1248 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1249 fl6.saddr = np->saddr; 1250 fl6.fl6_sport = inet->inet_sport; 1251 1252 final_p = fl6_update_dst(&fl6, opt, &final); 1253 if (final_p) 1254 connected = 0; 1255 1256 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1257 fl6.flowi6_oif = np->mcast_oif; 1258 connected = 0; 1259 } else if (!fl6.flowi6_oif) 1260 fl6.flowi6_oif = np->ucast_oif; 1261 1262 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1263 1264 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p); 1265 if (IS_ERR(dst)) { 1266 err = PTR_ERR(dst); 1267 dst = NULL; 1268 goto out; 1269 } 1270 1271 if (hlimit < 0) 1272 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1273 1274 if (tclass < 0) 1275 tclass = np->tclass; 1276 1277 if (msg->msg_flags&MSG_CONFIRM) 1278 goto do_confirm; 1279 back_from_confirm: 1280 1281 lock_sock(sk); 1282 if (unlikely(up->pending)) { 1283 /* The socket is already corked while preparing it. */ 1284 /* ... which is an evident application bug. --ANK */ 1285 release_sock(sk); 1286 1287 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 1288 err = -EINVAL; 1289 goto out; 1290 } 1291 1292 up->pending = AF_INET6; 1293 1294 do_append_data: 1295 if (dontfrag < 0) 1296 dontfrag = np->dontfrag; 1297 up->len += ulen; 1298 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1299 err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen, 1300 sizeof(struct udphdr), hlimit, tclass, opt, &fl6, 1301 (struct rt6_info *)dst, 1302 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag); 1303 if (err) 1304 udp_v6_flush_pending_frames(sk); 1305 else if (!corkreq) 1306 err = udp_v6_push_pending_frames(sk); 1307 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1308 up->pending = 0; 1309 1310 if (dst) { 1311 if (connected) { 1312 ip6_dst_store(sk, dst, 1313 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ? 1314 &sk->sk_v6_daddr : NULL, 1315 #ifdef CONFIG_IPV6_SUBTREES 1316 ipv6_addr_equal(&fl6.saddr, &np->saddr) ? 1317 &np->saddr : 1318 #endif 1319 NULL); 1320 } else { 1321 dst_release(dst); 1322 } 1323 dst = NULL; 1324 } 1325 1326 if (err > 0) 1327 err = np->recverr ? net_xmit_errno(err) : 0; 1328 release_sock(sk); 1329 out: 1330 dst_release(dst); 1331 fl6_sock_release(flowlabel); 1332 if (!err) 1333 return len; 1334 /* 1335 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1336 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1337 * we don't have a good statistic (IpOutDiscards but it can be too many 1338 * things). We could add another new stat but at least for now that 1339 * seems like overkill. 1340 */ 1341 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1342 UDP6_INC_STATS_USER(sock_net(sk), 1343 UDP_MIB_SNDBUFERRORS, is_udplite); 1344 } 1345 return err; 1346 1347 do_confirm: 1348 dst_confirm(dst); 1349 if (!(msg->msg_flags&MSG_PROBE) || len) 1350 goto back_from_confirm; 1351 err = 0; 1352 goto out; 1353 } 1354 1355 void udpv6_destroy_sock(struct sock *sk) 1356 { 1357 struct udp_sock *up = udp_sk(sk); 1358 lock_sock(sk); 1359 udp_v6_flush_pending_frames(sk); 1360 release_sock(sk); 1361 1362 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 1363 void (*encap_destroy)(struct sock *sk); 1364 encap_destroy = ACCESS_ONCE(up->encap_destroy); 1365 if (encap_destroy) 1366 encap_destroy(sk); 1367 } 1368 1369 inet6_destroy_sock(sk); 1370 } 1371 1372 /* 1373 * Socket option code for UDP 1374 */ 1375 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1376 char __user *optval, unsigned int optlen) 1377 { 1378 if (level == SOL_UDP || level == SOL_UDPLITE) 1379 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1380 udp_v6_push_pending_frames); 1381 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1382 } 1383 1384 #ifdef CONFIG_COMPAT 1385 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1386 char __user *optval, unsigned int optlen) 1387 { 1388 if (level == SOL_UDP || level == SOL_UDPLITE) 1389 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1390 udp_v6_push_pending_frames); 1391 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1392 } 1393 #endif 1394 1395 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1396 char __user *optval, int __user *optlen) 1397 { 1398 if (level == SOL_UDP || level == SOL_UDPLITE) 1399 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1400 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1401 } 1402 1403 #ifdef CONFIG_COMPAT 1404 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1405 char __user *optval, int __user *optlen) 1406 { 1407 if (level == SOL_UDP || level == SOL_UDPLITE) 1408 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1409 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1410 } 1411 #endif 1412 1413 static const struct inet6_protocol udpv6_protocol = { 1414 .handler = udpv6_rcv, 1415 .err_handler = udpv6_err, 1416 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1417 }; 1418 1419 /* ------------------------------------------------------------------------ */ 1420 #ifdef CONFIG_PROC_FS 1421 int udp6_seq_show(struct seq_file *seq, void *v) 1422 { 1423 if (v == SEQ_START_TOKEN) { 1424 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1425 } else { 1426 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1427 struct inet_sock *inet = inet_sk(v); 1428 __u16 srcp = ntohs(inet->inet_sport); 1429 __u16 destp = ntohs(inet->inet_dport); 1430 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket); 1431 } 1432 return 0; 1433 } 1434 1435 static const struct file_operations udp6_afinfo_seq_fops = { 1436 .owner = THIS_MODULE, 1437 .open = udp_seq_open, 1438 .read = seq_read, 1439 .llseek = seq_lseek, 1440 .release = seq_release_net 1441 }; 1442 1443 static struct udp_seq_afinfo udp6_seq_afinfo = { 1444 .name = "udp6", 1445 .family = AF_INET6, 1446 .udp_table = &udp_table, 1447 .seq_fops = &udp6_afinfo_seq_fops, 1448 .seq_ops = { 1449 .show = udp6_seq_show, 1450 }, 1451 }; 1452 1453 int __net_init udp6_proc_init(struct net *net) 1454 { 1455 return udp_proc_register(net, &udp6_seq_afinfo); 1456 } 1457 1458 void udp6_proc_exit(struct net *net) { 1459 udp_proc_unregister(net, &udp6_seq_afinfo); 1460 } 1461 #endif /* CONFIG_PROC_FS */ 1462 1463 void udp_v6_clear_sk(struct sock *sk, int size) 1464 { 1465 struct inet_sock *inet = inet_sk(sk); 1466 1467 /* we do not want to clear pinet6 field, because of RCU lookups */ 1468 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6)); 1469 1470 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6); 1471 memset(&inet->pinet6 + 1, 0, size); 1472 } 1473 1474 /* ------------------------------------------------------------------------ */ 1475 1476 struct proto udpv6_prot = { 1477 .name = "UDPv6", 1478 .owner = THIS_MODULE, 1479 .close = udp_lib_close, 1480 .connect = ip6_datagram_connect, 1481 .disconnect = udp_disconnect, 1482 .ioctl = udp_ioctl, 1483 .destroy = udpv6_destroy_sock, 1484 .setsockopt = udpv6_setsockopt, 1485 .getsockopt = udpv6_getsockopt, 1486 .sendmsg = udpv6_sendmsg, 1487 .recvmsg = udpv6_recvmsg, 1488 .backlog_rcv = __udpv6_queue_rcv_skb, 1489 .hash = udp_lib_hash, 1490 .unhash = udp_lib_unhash, 1491 .rehash = udp_v6_rehash, 1492 .get_port = udp_v6_get_port, 1493 .memory_allocated = &udp_memory_allocated, 1494 .sysctl_mem = sysctl_udp_mem, 1495 .sysctl_wmem = &sysctl_udp_wmem_min, 1496 .sysctl_rmem = &sysctl_udp_rmem_min, 1497 .obj_size = sizeof(struct udp6_sock), 1498 .slab_flags = SLAB_DESTROY_BY_RCU, 1499 .h.udp_table = &udp_table, 1500 #ifdef CONFIG_COMPAT 1501 .compat_setsockopt = compat_udpv6_setsockopt, 1502 .compat_getsockopt = compat_udpv6_getsockopt, 1503 #endif 1504 .clear_sk = udp_v6_clear_sk, 1505 }; 1506 1507 static struct inet_protosw udpv6_protosw = { 1508 .type = SOCK_DGRAM, 1509 .protocol = IPPROTO_UDP, 1510 .prot = &udpv6_prot, 1511 .ops = &inet6_dgram_ops, 1512 .flags = INET_PROTOSW_PERMANENT, 1513 }; 1514 1515 1516 int __init udpv6_init(void) 1517 { 1518 int ret; 1519 1520 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1521 if (ret) 1522 goto out; 1523 1524 ret = inet6_register_protosw(&udpv6_protosw); 1525 if (ret) 1526 goto out_udpv6_protocol; 1527 out: 1528 return ret; 1529 1530 out_udpv6_protocol: 1531 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1532 goto out; 1533 } 1534 1535 void udpv6_exit(void) 1536 { 1537 inet6_unregister_protosw(&udpv6_protosw); 1538 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1539 } 1540