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 <linux/uaccess.h> 39 40 #include <net/addrconf.h> 41 #include <net/ndisc.h> 42 #include <net/protocol.h> 43 #include <net/transp_v6.h> 44 #include <net/ip6_route.h> 45 #include <net/raw.h> 46 #include <net/tcp_states.h> 47 #include <net/ip6_checksum.h> 48 #include <net/xfrm.h> 49 #include <net/inet_hashtables.h> 50 #include <net/inet6_hashtables.h> 51 #include <net/busy_poll.h> 52 #include <net/sock_reuseport.h> 53 54 #include <linux/proc_fs.h> 55 #include <linux/seq_file.h> 56 #include <trace/events/skb.h> 57 #include "udp_impl.h" 58 59 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb) 60 { 61 #if defined(CONFIG_NET_L3_MASTER_DEV) 62 if (!net->ipv4.sysctl_udp_l3mdev_accept && 63 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags)) 64 return true; 65 #endif 66 return false; 67 } 68 69 static u32 udp6_ehashfn(const struct net *net, 70 const struct in6_addr *laddr, 71 const u16 lport, 72 const struct in6_addr *faddr, 73 const __be16 fport) 74 { 75 static u32 udp6_ehash_secret __read_mostly; 76 static u32 udp_ipv6_hash_secret __read_mostly; 77 78 u32 lhash, fhash; 79 80 net_get_random_once(&udp6_ehash_secret, 81 sizeof(udp6_ehash_secret)); 82 net_get_random_once(&udp_ipv6_hash_secret, 83 sizeof(udp_ipv6_hash_secret)); 84 85 lhash = (__force u32)laddr->s6_addr32[3]; 86 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 87 88 return __inet6_ehashfn(lhash, lport, fhash, fport, 89 udp_ipv6_hash_secret + net_hash_mix(net)); 90 } 91 92 static u32 udp6_portaddr_hash(const struct net *net, 93 const struct in6_addr *addr6, 94 unsigned int port) 95 { 96 unsigned int hash, mix = net_hash_mix(net); 97 98 if (ipv6_addr_any(addr6)) 99 hash = jhash_1word(0, mix); 100 else if (ipv6_addr_v4mapped(addr6)) 101 hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix); 102 else 103 hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix); 104 105 return hash ^ port; 106 } 107 108 int udp_v6_get_port(struct sock *sk, unsigned short snum) 109 { 110 unsigned int hash2_nulladdr = 111 udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 112 unsigned int hash2_partial = 113 udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 114 115 /* precompute partial secondary hash */ 116 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 117 return udp_lib_get_port(sk, snum, hash2_nulladdr); 118 } 119 120 static void udp_v6_rehash(struct sock *sk) 121 { 122 u16 new_hash = udp6_portaddr_hash(sock_net(sk), 123 &sk->sk_v6_rcv_saddr, 124 inet_sk(sk)->inet_num); 125 126 udp_lib_rehash(sk, new_hash); 127 } 128 129 static int compute_score(struct sock *sk, struct net *net, 130 const struct in6_addr *saddr, __be16 sport, 131 const struct in6_addr *daddr, unsigned short hnum, 132 int dif, bool exact_dif) 133 { 134 int score; 135 struct inet_sock *inet; 136 137 if (!net_eq(sock_net(sk), net) || 138 udp_sk(sk)->udp_port_hash != hnum || 139 sk->sk_family != PF_INET6) 140 return -1; 141 142 score = 0; 143 inet = inet_sk(sk); 144 145 if (inet->inet_dport) { 146 if (inet->inet_dport != sport) 147 return -1; 148 score++; 149 } 150 151 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { 152 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 153 return -1; 154 score++; 155 } 156 157 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 158 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 159 return -1; 160 score++; 161 } 162 163 if (sk->sk_bound_dev_if || exact_dif) { 164 if (sk->sk_bound_dev_if != dif) 165 return -1; 166 score++; 167 } 168 169 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 170 score++; 171 172 return score; 173 } 174 175 /* called with rcu_read_lock() */ 176 static struct sock *udp6_lib_lookup2(struct net *net, 177 const struct in6_addr *saddr, __be16 sport, 178 const struct in6_addr *daddr, unsigned int hnum, int dif, 179 bool exact_dif, struct udp_hslot *hslot2, 180 struct sk_buff *skb) 181 { 182 struct sock *sk, *result; 183 int score, badness, matches = 0, reuseport = 0; 184 u32 hash = 0; 185 186 result = NULL; 187 badness = -1; 188 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 189 score = compute_score(sk, net, saddr, sport, 190 daddr, hnum, dif, exact_dif); 191 if (score > badness) { 192 reuseport = sk->sk_reuseport; 193 if (reuseport) { 194 hash = udp6_ehashfn(net, daddr, hnum, 195 saddr, sport); 196 197 result = reuseport_select_sock(sk, hash, skb, 198 sizeof(struct udphdr)); 199 if (result) 200 return result; 201 matches = 1; 202 } 203 result = sk; 204 badness = score; 205 } else if (score == badness && reuseport) { 206 matches++; 207 if (reciprocal_scale(hash, matches) == 0) 208 result = sk; 209 hash = next_pseudo_random32(hash); 210 } 211 } 212 return result; 213 } 214 215 /* rcu_read_lock() must be held */ 216 struct sock *__udp6_lib_lookup(struct net *net, 217 const struct in6_addr *saddr, __be16 sport, 218 const struct in6_addr *daddr, __be16 dport, 219 int dif, struct udp_table *udptable, 220 struct sk_buff *skb) 221 { 222 struct sock *sk, *result; 223 unsigned short hnum = ntohs(dport); 224 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 225 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 226 bool exact_dif = udp6_lib_exact_dif_match(net, skb); 227 int score, badness, matches = 0, reuseport = 0; 228 u32 hash = 0; 229 230 if (hslot->count > 10) { 231 hash2 = udp6_portaddr_hash(net, daddr, hnum); 232 slot2 = hash2 & udptable->mask; 233 hslot2 = &udptable->hash2[slot2]; 234 if (hslot->count < hslot2->count) 235 goto begin; 236 237 result = udp6_lib_lookup2(net, saddr, sport, 238 daddr, hnum, dif, exact_dif, 239 hslot2, skb); 240 if (!result) { 241 unsigned int old_slot2 = slot2; 242 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum); 243 slot2 = hash2 & udptable->mask; 244 /* avoid searching the same slot again. */ 245 if (unlikely(slot2 == old_slot2)) 246 return result; 247 248 hslot2 = &udptable->hash2[slot2]; 249 if (hslot->count < hslot2->count) 250 goto begin; 251 252 result = udp6_lib_lookup2(net, saddr, sport, 253 daddr, hnum, dif, 254 exact_dif, hslot2, 255 skb); 256 } 257 return result; 258 } 259 begin: 260 result = NULL; 261 badness = -1; 262 sk_for_each_rcu(sk, &hslot->head) { 263 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif, 264 exact_dif); 265 if (score > badness) { 266 reuseport = sk->sk_reuseport; 267 if (reuseport) { 268 hash = udp6_ehashfn(net, daddr, hnum, 269 saddr, sport); 270 result = reuseport_select_sock(sk, hash, skb, 271 sizeof(struct udphdr)); 272 if (result) 273 return result; 274 matches = 1; 275 } 276 result = sk; 277 badness = score; 278 } else if (score == badness && reuseport) { 279 matches++; 280 if (reciprocal_scale(hash, matches) == 0) 281 result = sk; 282 hash = next_pseudo_random32(hash); 283 } 284 } 285 return result; 286 } 287 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 288 289 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 290 __be16 sport, __be16 dport, 291 struct udp_table *udptable) 292 { 293 const struct ipv6hdr *iph = ipv6_hdr(skb); 294 295 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 296 &iph->daddr, dport, inet6_iif(skb), 297 udptable, skb); 298 } 299 300 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, 301 __be16 sport, __be16 dport) 302 { 303 const struct ipv6hdr *iph = ipv6_hdr(skb); 304 305 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 306 &iph->daddr, dport, inet6_iif(skb), 307 &udp_table, skb); 308 } 309 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); 310 311 /* Must be called under rcu_read_lock(). 312 * Does increment socket refcount. 313 */ 314 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \ 315 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \ 316 IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 317 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 318 const struct in6_addr *daddr, __be16 dport, int dif) 319 { 320 struct sock *sk; 321 322 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 323 dif, &udp_table, NULL); 324 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 325 sk = NULL; 326 return sk; 327 } 328 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 329 #endif 330 331 /* do not use the scratch area len for jumbogram: their length execeeds the 332 * scratch area space; note that the IP6CB flags is still in the first 333 * cacheline, so checking for jumbograms is cheap 334 */ 335 static int udp6_skb_len(struct sk_buff *skb) 336 { 337 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 338 } 339 340 /* 341 * This should be easy, if there is something there we 342 * return it, otherwise we block. 343 */ 344 345 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 346 int noblock, int flags, int *addr_len) 347 { 348 struct ipv6_pinfo *np = inet6_sk(sk); 349 struct inet_sock *inet = inet_sk(sk); 350 struct sk_buff *skb; 351 unsigned int ulen, copied; 352 int peeked, peeking, off; 353 int err; 354 int is_udplite = IS_UDPLITE(sk); 355 bool checksum_valid = false; 356 int is_udp4; 357 358 if (flags & MSG_ERRQUEUE) 359 return ipv6_recv_error(sk, msg, len, addr_len); 360 361 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 362 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 363 364 try_again: 365 peeking = off = sk_peek_offset(sk, flags); 366 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); 367 if (!skb) 368 return err; 369 370 ulen = udp6_skb_len(skb); 371 copied = len; 372 if (copied > ulen - off) 373 copied = ulen - off; 374 else if (copied < ulen) 375 msg->msg_flags |= MSG_TRUNC; 376 377 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 378 379 /* 380 * If checksum is needed at all, try to do it while copying the 381 * data. If the data is truncated, or if we only want a partial 382 * coverage checksum (UDP-Lite), do it before the copy. 383 */ 384 385 if (copied < ulen || peeking || 386 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 387 checksum_valid = udp_skb_csum_unnecessary(skb) || 388 !__udp_lib_checksum_complete(skb); 389 if (!checksum_valid) 390 goto csum_copy_err; 391 } 392 393 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 394 if (udp_skb_is_linear(skb)) 395 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 396 else 397 err = skb_copy_datagram_msg(skb, off, msg, copied); 398 } else { 399 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 400 if (err == -EINVAL) 401 goto csum_copy_err; 402 } 403 if (unlikely(err)) { 404 if (!peeked) { 405 atomic_inc(&sk->sk_drops); 406 if (is_udp4) 407 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 408 is_udplite); 409 else 410 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 411 is_udplite); 412 } 413 kfree_skb(skb); 414 return err; 415 } 416 if (!peeked) { 417 if (is_udp4) 418 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, 419 is_udplite); 420 else 421 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, 422 is_udplite); 423 } 424 425 sock_recv_ts_and_drops(msg, sk, skb); 426 427 /* Copy the address. */ 428 if (msg->msg_name) { 429 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 430 sin6->sin6_family = AF_INET6; 431 sin6->sin6_port = udp_hdr(skb)->source; 432 sin6->sin6_flowinfo = 0; 433 434 if (is_udp4) { 435 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 436 &sin6->sin6_addr); 437 sin6->sin6_scope_id = 0; 438 } else { 439 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 440 sin6->sin6_scope_id = 441 ipv6_iface_scope_id(&sin6->sin6_addr, 442 inet6_iif(skb)); 443 } 444 *addr_len = sizeof(*sin6); 445 } 446 447 if (np->rxopt.all) 448 ip6_datagram_recv_common_ctl(sk, msg, skb); 449 450 if (is_udp4) { 451 if (inet->cmsg_flags) 452 ip_cmsg_recv_offset(msg, sk, skb, 453 sizeof(struct udphdr), off); 454 } else { 455 if (np->rxopt.all) 456 ip6_datagram_recv_specific_ctl(sk, msg, skb); 457 } 458 459 err = copied; 460 if (flags & MSG_TRUNC) 461 err = ulen; 462 463 skb_consume_udp(sk, skb, peeking ? -err : err); 464 return err; 465 466 csum_copy_err: 467 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 468 udp_skb_destructor)) { 469 if (is_udp4) { 470 UDP_INC_STATS(sock_net(sk), 471 UDP_MIB_CSUMERRORS, is_udplite); 472 UDP_INC_STATS(sock_net(sk), 473 UDP_MIB_INERRORS, is_udplite); 474 } else { 475 UDP6_INC_STATS(sock_net(sk), 476 UDP_MIB_CSUMERRORS, is_udplite); 477 UDP6_INC_STATS(sock_net(sk), 478 UDP_MIB_INERRORS, is_udplite); 479 } 480 } 481 kfree_skb(skb); 482 483 /* starting over for a new packet, but check if we need to yield */ 484 cond_resched(); 485 msg->msg_flags &= ~MSG_TRUNC; 486 goto try_again; 487 } 488 489 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 490 u8 type, u8 code, int offset, __be32 info, 491 struct udp_table *udptable) 492 { 493 struct ipv6_pinfo *np; 494 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 495 const struct in6_addr *saddr = &hdr->saddr; 496 const struct in6_addr *daddr = &hdr->daddr; 497 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 498 struct sock *sk; 499 int harderr; 500 int err; 501 struct net *net = dev_net(skb->dev); 502 503 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 504 inet6_iif(skb), udptable, skb); 505 if (!sk) { 506 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 507 ICMP6_MIB_INERRORS); 508 return; 509 } 510 511 harderr = icmpv6_err_convert(type, code, &err); 512 np = inet6_sk(sk); 513 514 if (type == ICMPV6_PKT_TOOBIG) { 515 if (!ip6_sk_accept_pmtu(sk)) 516 goto out; 517 ip6_sk_update_pmtu(skb, sk, info); 518 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 519 harderr = 1; 520 } 521 if (type == NDISC_REDIRECT) { 522 ip6_sk_redirect(skb, sk); 523 goto out; 524 } 525 526 if (!np->recverr) { 527 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 528 goto out; 529 } else { 530 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 531 } 532 533 sk->sk_err = err; 534 sk->sk_error_report(sk); 535 out: 536 return; 537 } 538 539 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 540 { 541 int rc; 542 543 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 544 sock_rps_save_rxhash(sk, skb); 545 sk_mark_napi_id(sk, skb); 546 sk_incoming_cpu_update(sk); 547 } else { 548 sk_mark_napi_id_once(sk, skb); 549 } 550 551 rc = __udp_enqueue_schedule_skb(sk, skb); 552 if (rc < 0) { 553 int is_udplite = IS_UDPLITE(sk); 554 555 /* Note that an ENOMEM error is charged twice */ 556 if (rc == -ENOMEM) 557 UDP6_INC_STATS(sock_net(sk), 558 UDP_MIB_RCVBUFERRORS, is_udplite); 559 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 560 kfree_skb(skb); 561 return -1; 562 } 563 564 return 0; 565 } 566 567 static __inline__ void udpv6_err(struct sk_buff *skb, 568 struct inet6_skb_parm *opt, u8 type, 569 u8 code, int offset, __be32 info) 570 { 571 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 572 } 573 574 static struct static_key udpv6_encap_needed __read_mostly; 575 void udpv6_encap_enable(void) 576 { 577 if (!static_key_enabled(&udpv6_encap_needed)) 578 static_key_slow_inc(&udpv6_encap_needed); 579 } 580 EXPORT_SYMBOL(udpv6_encap_enable); 581 582 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 583 { 584 struct udp_sock *up = udp_sk(sk); 585 int is_udplite = IS_UDPLITE(sk); 586 587 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 588 goto drop; 589 590 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 591 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 592 593 /* 594 * This is an encapsulation socket so pass the skb to 595 * the socket's udp_encap_rcv() hook. Otherwise, just 596 * fall through and pass this up the UDP socket. 597 * up->encap_rcv() returns the following value: 598 * =0 if skb was successfully passed to the encap 599 * handler or was discarded by it. 600 * >0 if skb should be passed on to UDP. 601 * <0 if skb should be resubmitted as proto -N 602 */ 603 604 /* if we're overly short, let UDP handle it */ 605 encap_rcv = ACCESS_ONCE(up->encap_rcv); 606 if (encap_rcv) { 607 int ret; 608 609 /* Verify checksum before giving to encap */ 610 if (udp_lib_checksum_complete(skb)) 611 goto csum_error; 612 613 ret = encap_rcv(sk, skb); 614 if (ret <= 0) { 615 __UDP_INC_STATS(sock_net(sk), 616 UDP_MIB_INDATAGRAMS, 617 is_udplite); 618 return -ret; 619 } 620 } 621 622 /* FALLTHROUGH -- it's a UDP Packet */ 623 } 624 625 /* 626 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 627 */ 628 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 629 630 if (up->pcrlen == 0) { /* full coverage was set */ 631 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 632 UDP_SKB_CB(skb)->cscov, skb->len); 633 goto drop; 634 } 635 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 636 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 637 UDP_SKB_CB(skb)->cscov, up->pcrlen); 638 goto drop; 639 } 640 } 641 642 prefetch(&sk->sk_rmem_alloc); 643 if (rcu_access_pointer(sk->sk_filter) && 644 udp_lib_checksum_complete(skb)) 645 goto csum_error; 646 647 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 648 goto drop; 649 650 udp_csum_pull_header(skb); 651 652 skb_dst_drop(skb); 653 654 return __udpv6_queue_rcv_skb(sk, skb); 655 656 csum_error: 657 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 658 drop: 659 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 660 atomic_inc(&sk->sk_drops); 661 kfree_skb(skb); 662 return -1; 663 } 664 665 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 666 __be16 loc_port, const struct in6_addr *loc_addr, 667 __be16 rmt_port, const struct in6_addr *rmt_addr, 668 int dif, unsigned short hnum) 669 { 670 struct inet_sock *inet = inet_sk(sk); 671 672 if (!net_eq(sock_net(sk), net)) 673 return false; 674 675 if (udp_sk(sk)->udp_port_hash != hnum || 676 sk->sk_family != PF_INET6 || 677 (inet->inet_dport && inet->inet_dport != rmt_port) || 678 (!ipv6_addr_any(&sk->sk_v6_daddr) && 679 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 680 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) || 681 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 682 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 683 return false; 684 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 685 return false; 686 return true; 687 } 688 689 static void udp6_csum_zero_error(struct sk_buff *skb) 690 { 691 /* RFC 2460 section 8.1 says that we SHOULD log 692 * this error. Well, it is reasonable. 693 */ 694 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 695 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 696 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 697 } 698 699 /* 700 * Note: called only from the BH handler context, 701 * so we don't need to lock the hashes. 702 */ 703 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 704 const struct in6_addr *saddr, const struct in6_addr *daddr, 705 struct udp_table *udptable, int proto) 706 { 707 struct sock *sk, *first = NULL; 708 const struct udphdr *uh = udp_hdr(skb); 709 unsigned short hnum = ntohs(uh->dest); 710 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 711 unsigned int offset = offsetof(typeof(*sk), sk_node); 712 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 713 int dif = inet6_iif(skb); 714 struct hlist_node *node; 715 struct sk_buff *nskb; 716 717 if (use_hash2) { 718 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) & 719 udptable->mask; 720 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udptable->mask; 721 start_lookup: 722 hslot = &udptable->hash2[hash2]; 723 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 724 } 725 726 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 727 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 728 uh->source, saddr, dif, hnum)) 729 continue; 730 /* If zero checksum and no_check is not on for 731 * the socket then skip it. 732 */ 733 if (!uh->check && !udp_sk(sk)->no_check6_rx) 734 continue; 735 if (!first) { 736 first = sk; 737 continue; 738 } 739 nskb = skb_clone(skb, GFP_ATOMIC); 740 if (unlikely(!nskb)) { 741 atomic_inc(&sk->sk_drops); 742 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 743 IS_UDPLITE(sk)); 744 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 745 IS_UDPLITE(sk)); 746 continue; 747 } 748 749 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 750 consume_skb(nskb); 751 } 752 753 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 754 if (use_hash2 && hash2 != hash2_any) { 755 hash2 = hash2_any; 756 goto start_lookup; 757 } 758 759 if (first) { 760 if (udpv6_queue_rcv_skb(first, skb) > 0) 761 consume_skb(skb); 762 } else { 763 kfree_skb(skb); 764 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 765 proto == IPPROTO_UDPLITE); 766 } 767 return 0; 768 } 769 770 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 771 int proto) 772 { 773 const struct in6_addr *saddr, *daddr; 774 struct net *net = dev_net(skb->dev); 775 struct udphdr *uh; 776 struct sock *sk; 777 u32 ulen = 0; 778 779 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 780 goto discard; 781 782 saddr = &ipv6_hdr(skb)->saddr; 783 daddr = &ipv6_hdr(skb)->daddr; 784 uh = udp_hdr(skb); 785 786 ulen = ntohs(uh->len); 787 if (ulen > skb->len) 788 goto short_packet; 789 790 if (proto == IPPROTO_UDP) { 791 /* UDP validates ulen. */ 792 793 /* Check for jumbo payload */ 794 if (ulen == 0) 795 ulen = skb->len; 796 797 if (ulen < sizeof(*uh)) 798 goto short_packet; 799 800 if (ulen < skb->len) { 801 if (pskb_trim_rcsum(skb, ulen)) 802 goto short_packet; 803 saddr = &ipv6_hdr(skb)->saddr; 804 daddr = &ipv6_hdr(skb)->daddr; 805 uh = udp_hdr(skb); 806 } 807 } 808 809 if (udp6_csum_init(skb, uh, proto)) 810 goto csum_error; 811 812 /* Check if the socket is already available, e.g. due to early demux */ 813 sk = skb_steal_sock(skb); 814 if (sk) { 815 struct dst_entry *dst = skb_dst(skb); 816 int ret; 817 818 if (unlikely(sk->sk_rx_dst != dst)) 819 udp_sk_rx_dst_set(sk, dst); 820 821 ret = udpv6_queue_rcv_skb(sk, skb); 822 sock_put(sk); 823 824 /* a return value > 0 means to resubmit the input */ 825 if (ret > 0) 826 return ret; 827 return 0; 828 } 829 830 /* 831 * Multicast receive code 832 */ 833 if (ipv6_addr_is_multicast(daddr)) 834 return __udp6_lib_mcast_deliver(net, skb, 835 saddr, daddr, udptable, proto); 836 837 /* Unicast */ 838 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 839 if (sk) { 840 int ret; 841 842 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 843 udp6_csum_zero_error(skb); 844 goto csum_error; 845 } 846 847 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 848 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 849 ip6_compute_pseudo); 850 851 ret = udpv6_queue_rcv_skb(sk, skb); 852 853 /* a return value > 0 means to resubmit the input */ 854 if (ret > 0) 855 return ret; 856 857 return 0; 858 } 859 860 if (!uh->check) { 861 udp6_csum_zero_error(skb); 862 goto csum_error; 863 } 864 865 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 866 goto discard; 867 868 if (udp_lib_checksum_complete(skb)) 869 goto csum_error; 870 871 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 872 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 873 874 kfree_skb(skb); 875 return 0; 876 877 short_packet: 878 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 879 proto == IPPROTO_UDPLITE ? "-Lite" : "", 880 saddr, ntohs(uh->source), 881 ulen, skb->len, 882 daddr, ntohs(uh->dest)); 883 goto discard; 884 csum_error: 885 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 886 discard: 887 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 888 kfree_skb(skb); 889 return 0; 890 } 891 892 893 static struct sock *__udp6_lib_demux_lookup(struct net *net, 894 __be16 loc_port, const struct in6_addr *loc_addr, 895 __be16 rmt_port, const struct in6_addr *rmt_addr, 896 int dif) 897 { 898 unsigned short hnum = ntohs(loc_port); 899 unsigned int hash2 = udp6_portaddr_hash(net, loc_addr, hnum); 900 unsigned int slot2 = hash2 & udp_table.mask; 901 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 902 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 903 struct sock *sk; 904 905 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 906 if (sk->sk_state == TCP_ESTABLISHED && 907 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif)) 908 return sk; 909 /* Only check first socket in chain */ 910 break; 911 } 912 return NULL; 913 } 914 915 static void udp_v6_early_demux(struct sk_buff *skb) 916 { 917 struct net *net = dev_net(skb->dev); 918 const struct udphdr *uh; 919 struct sock *sk; 920 struct dst_entry *dst; 921 int dif = skb->dev->ifindex; 922 923 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 924 sizeof(struct udphdr))) 925 return; 926 927 uh = udp_hdr(skb); 928 929 if (skb->pkt_type == PACKET_HOST) 930 sk = __udp6_lib_demux_lookup(net, uh->dest, 931 &ipv6_hdr(skb)->daddr, 932 uh->source, &ipv6_hdr(skb)->saddr, 933 dif); 934 else 935 return; 936 937 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 938 return; 939 940 skb->sk = sk; 941 skb->destructor = sock_efree; 942 dst = READ_ONCE(sk->sk_rx_dst); 943 944 if (dst) 945 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 946 if (dst) { 947 /* set noref for now. 948 * any place which wants to hold dst has to call 949 * dst_hold_safe() 950 */ 951 skb_dst_set_noref(skb, dst); 952 } 953 } 954 955 static __inline__ int udpv6_rcv(struct sk_buff *skb) 956 { 957 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 958 } 959 960 /* 961 * Throw away all pending data and cancel the corking. Socket is locked. 962 */ 963 static void udp_v6_flush_pending_frames(struct sock *sk) 964 { 965 struct udp_sock *up = udp_sk(sk); 966 967 if (up->pending == AF_INET) 968 udp_flush_pending_frames(sk); 969 else if (up->pending) { 970 up->len = 0; 971 up->pending = 0; 972 ip6_flush_pending_frames(sk); 973 } 974 } 975 976 /** 977 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 978 * @sk: socket we are sending on 979 * @skb: sk_buff containing the filled-in UDP header 980 * (checksum field must be zeroed out) 981 */ 982 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 983 const struct in6_addr *saddr, 984 const struct in6_addr *daddr, int len) 985 { 986 unsigned int offset; 987 struct udphdr *uh = udp_hdr(skb); 988 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 989 __wsum csum = 0; 990 991 if (!frags) { 992 /* Only one fragment on the socket. */ 993 skb->csum_start = skb_transport_header(skb) - skb->head; 994 skb->csum_offset = offsetof(struct udphdr, check); 995 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 996 } else { 997 /* 998 * HW-checksum won't work as there are two or more 999 * fragments on the socket so that all csums of sk_buffs 1000 * should be together 1001 */ 1002 offset = skb_transport_offset(skb); 1003 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1004 1005 skb->ip_summed = CHECKSUM_NONE; 1006 1007 do { 1008 csum = csum_add(csum, frags->csum); 1009 } while ((frags = frags->next)); 1010 1011 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1012 csum); 1013 if (uh->check == 0) 1014 uh->check = CSUM_MANGLED_0; 1015 } 1016 } 1017 1018 /* 1019 * Sending 1020 */ 1021 1022 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6) 1023 { 1024 struct sock *sk = skb->sk; 1025 struct udphdr *uh; 1026 int err = 0; 1027 int is_udplite = IS_UDPLITE(sk); 1028 __wsum csum = 0; 1029 int offset = skb_transport_offset(skb); 1030 int len = skb->len - offset; 1031 1032 /* 1033 * Create a UDP header 1034 */ 1035 uh = udp_hdr(skb); 1036 uh->source = fl6->fl6_sport; 1037 uh->dest = fl6->fl6_dport; 1038 uh->len = htons(len); 1039 uh->check = 0; 1040 1041 if (is_udplite) 1042 csum = udplite_csum(skb); 1043 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1044 skb->ip_summed = CHECKSUM_NONE; 1045 goto send; 1046 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1047 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1048 goto send; 1049 } else 1050 csum = udp_csum(skb); 1051 1052 /* add protocol-dependent pseudo-header */ 1053 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1054 len, fl6->flowi6_proto, csum); 1055 if (uh->check == 0) 1056 uh->check = CSUM_MANGLED_0; 1057 1058 send: 1059 err = ip6_send_skb(skb); 1060 if (err) { 1061 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1062 UDP6_INC_STATS(sock_net(sk), 1063 UDP_MIB_SNDBUFERRORS, is_udplite); 1064 err = 0; 1065 } 1066 } else { 1067 UDP6_INC_STATS(sock_net(sk), 1068 UDP_MIB_OUTDATAGRAMS, is_udplite); 1069 } 1070 return err; 1071 } 1072 1073 static int udp_v6_push_pending_frames(struct sock *sk) 1074 { 1075 struct sk_buff *skb; 1076 struct udp_sock *up = udp_sk(sk); 1077 struct flowi6 fl6; 1078 int err = 0; 1079 1080 if (up->pending == AF_INET) 1081 return udp_push_pending_frames(sk); 1082 1083 /* ip6_finish_skb will release the cork, so make a copy of 1084 * fl6 here. 1085 */ 1086 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1087 1088 skb = ip6_finish_skb(sk); 1089 if (!skb) 1090 goto out; 1091 1092 err = udp_v6_send_skb(skb, &fl6); 1093 1094 out: 1095 up->len = 0; 1096 up->pending = 0; 1097 return err; 1098 } 1099 1100 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1101 { 1102 struct ipv6_txoptions opt_space; 1103 struct udp_sock *up = udp_sk(sk); 1104 struct inet_sock *inet = inet_sk(sk); 1105 struct ipv6_pinfo *np = inet6_sk(sk); 1106 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1107 struct in6_addr *daddr, *final_p, final; 1108 struct ipv6_txoptions *opt = NULL; 1109 struct ipv6_txoptions *opt_to_free = NULL; 1110 struct ip6_flowlabel *flowlabel = NULL; 1111 struct flowi6 fl6; 1112 struct dst_entry *dst; 1113 struct ipcm6_cookie ipc6; 1114 int addr_len = msg->msg_namelen; 1115 int ulen = len; 1116 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1117 int err; 1118 int connected = 0; 1119 int is_udplite = IS_UDPLITE(sk); 1120 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1121 struct sockcm_cookie sockc; 1122 1123 ipc6.hlimit = -1; 1124 ipc6.tclass = -1; 1125 ipc6.dontfrag = -1; 1126 sockc.tsflags = sk->sk_tsflags; 1127 1128 /* destination address check */ 1129 if (sin6) { 1130 if (addr_len < offsetof(struct sockaddr, sa_data)) 1131 return -EINVAL; 1132 1133 switch (sin6->sin6_family) { 1134 case AF_INET6: 1135 if (addr_len < SIN6_LEN_RFC2133) 1136 return -EINVAL; 1137 daddr = &sin6->sin6_addr; 1138 if (ipv6_addr_any(daddr) && 1139 ipv6_addr_v4mapped(&np->saddr)) 1140 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1141 daddr); 1142 break; 1143 case AF_INET: 1144 goto do_udp_sendmsg; 1145 case AF_UNSPEC: 1146 msg->msg_name = sin6 = NULL; 1147 msg->msg_namelen = addr_len = 0; 1148 daddr = NULL; 1149 break; 1150 default: 1151 return -EINVAL; 1152 } 1153 } else if (!up->pending) { 1154 if (sk->sk_state != TCP_ESTABLISHED) 1155 return -EDESTADDRREQ; 1156 daddr = &sk->sk_v6_daddr; 1157 } else 1158 daddr = NULL; 1159 1160 if (daddr) { 1161 if (ipv6_addr_v4mapped(daddr)) { 1162 struct sockaddr_in sin; 1163 sin.sin_family = AF_INET; 1164 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1165 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1166 msg->msg_name = &sin; 1167 msg->msg_namelen = sizeof(sin); 1168 do_udp_sendmsg: 1169 if (__ipv6_only_sock(sk)) 1170 return -ENETUNREACH; 1171 return udp_sendmsg(sk, msg, len); 1172 } 1173 } 1174 1175 if (up->pending == AF_INET) 1176 return udp_sendmsg(sk, msg, len); 1177 1178 /* Rough check on arithmetic overflow, 1179 better check is made in ip6_append_data(). 1180 */ 1181 if (len > INT_MAX - sizeof(struct udphdr)) 1182 return -EMSGSIZE; 1183 1184 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1185 if (up->pending) { 1186 /* 1187 * There are pending frames. 1188 * The socket lock must be held while it's corked. 1189 */ 1190 lock_sock(sk); 1191 if (likely(up->pending)) { 1192 if (unlikely(up->pending != AF_INET6)) { 1193 release_sock(sk); 1194 return -EAFNOSUPPORT; 1195 } 1196 dst = NULL; 1197 goto do_append_data; 1198 } 1199 release_sock(sk); 1200 } 1201 ulen += sizeof(struct udphdr); 1202 1203 memset(&fl6, 0, sizeof(fl6)); 1204 1205 if (sin6) { 1206 if (sin6->sin6_port == 0) 1207 return -EINVAL; 1208 1209 fl6.fl6_dport = sin6->sin6_port; 1210 daddr = &sin6->sin6_addr; 1211 1212 if (np->sndflow) { 1213 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1214 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1215 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1216 if (!flowlabel) 1217 return -EINVAL; 1218 } 1219 } 1220 1221 /* 1222 * Otherwise it will be difficult to maintain 1223 * sk->sk_dst_cache. 1224 */ 1225 if (sk->sk_state == TCP_ESTABLISHED && 1226 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1227 daddr = &sk->sk_v6_daddr; 1228 1229 if (addr_len >= sizeof(struct sockaddr_in6) && 1230 sin6->sin6_scope_id && 1231 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1232 fl6.flowi6_oif = sin6->sin6_scope_id; 1233 } else { 1234 if (sk->sk_state != TCP_ESTABLISHED) 1235 return -EDESTADDRREQ; 1236 1237 fl6.fl6_dport = inet->inet_dport; 1238 daddr = &sk->sk_v6_daddr; 1239 fl6.flowlabel = np->flow_label; 1240 connected = 1; 1241 } 1242 1243 if (!fl6.flowi6_oif) 1244 fl6.flowi6_oif = sk->sk_bound_dev_if; 1245 1246 if (!fl6.flowi6_oif) 1247 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1248 1249 fl6.flowi6_mark = sk->sk_mark; 1250 fl6.flowi6_uid = sk->sk_uid; 1251 1252 if (msg->msg_controllen) { 1253 opt = &opt_space; 1254 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1255 opt->tot_len = sizeof(*opt); 1256 ipc6.opt = opt; 1257 1258 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc); 1259 if (err < 0) { 1260 fl6_sock_release(flowlabel); 1261 return err; 1262 } 1263 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1264 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1265 if (!flowlabel) 1266 return -EINVAL; 1267 } 1268 if (!(opt->opt_nflen|opt->opt_flen)) 1269 opt = NULL; 1270 connected = 0; 1271 } 1272 if (!opt) { 1273 opt = txopt_get(np); 1274 opt_to_free = opt; 1275 } 1276 if (flowlabel) 1277 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1278 opt = ipv6_fixup_options(&opt_space, opt); 1279 ipc6.opt = opt; 1280 1281 fl6.flowi6_proto = sk->sk_protocol; 1282 if (!ipv6_addr_any(daddr)) 1283 fl6.daddr = *daddr; 1284 else 1285 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1286 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1287 fl6.saddr = np->saddr; 1288 fl6.fl6_sport = inet->inet_sport; 1289 1290 final_p = fl6_update_dst(&fl6, opt, &final); 1291 if (final_p) 1292 connected = 0; 1293 1294 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1295 fl6.flowi6_oif = np->mcast_oif; 1296 connected = 0; 1297 } else if (!fl6.flowi6_oif) 1298 fl6.flowi6_oif = np->ucast_oif; 1299 1300 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1301 1302 if (ipc6.tclass < 0) 1303 ipc6.tclass = np->tclass; 1304 1305 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1306 1307 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p); 1308 if (IS_ERR(dst)) { 1309 err = PTR_ERR(dst); 1310 dst = NULL; 1311 goto out; 1312 } 1313 1314 if (ipc6.hlimit < 0) 1315 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1316 1317 if (msg->msg_flags&MSG_CONFIRM) 1318 goto do_confirm; 1319 back_from_confirm: 1320 1321 /* Lockless fast path for the non-corking case */ 1322 if (!corkreq) { 1323 struct sk_buff *skb; 1324 1325 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1326 sizeof(struct udphdr), &ipc6, 1327 &fl6, (struct rt6_info *)dst, 1328 msg->msg_flags, &sockc); 1329 err = PTR_ERR(skb); 1330 if (!IS_ERR_OR_NULL(skb)) 1331 err = udp_v6_send_skb(skb, &fl6); 1332 goto release_dst; 1333 } 1334 1335 lock_sock(sk); 1336 if (unlikely(up->pending)) { 1337 /* The socket is already corked while preparing it. */ 1338 /* ... which is an evident application bug. --ANK */ 1339 release_sock(sk); 1340 1341 net_dbg_ratelimited("udp cork app bug 2\n"); 1342 err = -EINVAL; 1343 goto out; 1344 } 1345 1346 up->pending = AF_INET6; 1347 1348 do_append_data: 1349 if (ipc6.dontfrag < 0) 1350 ipc6.dontfrag = np->dontfrag; 1351 up->len += ulen; 1352 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1353 &ipc6, &fl6, (struct rt6_info *)dst, 1354 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc); 1355 if (err) 1356 udp_v6_flush_pending_frames(sk); 1357 else if (!corkreq) 1358 err = udp_v6_push_pending_frames(sk); 1359 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1360 up->pending = 0; 1361 1362 if (err > 0) 1363 err = np->recverr ? net_xmit_errno(err) : 0; 1364 release_sock(sk); 1365 1366 release_dst: 1367 if (dst) { 1368 if (connected) { 1369 ip6_dst_store(sk, dst, 1370 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ? 1371 &sk->sk_v6_daddr : NULL, 1372 #ifdef CONFIG_IPV6_SUBTREES 1373 ipv6_addr_equal(&fl6.saddr, &np->saddr) ? 1374 &np->saddr : 1375 #endif 1376 NULL); 1377 } else { 1378 dst_release(dst); 1379 } 1380 dst = NULL; 1381 } 1382 1383 out: 1384 dst_release(dst); 1385 fl6_sock_release(flowlabel); 1386 txopt_put(opt_to_free); 1387 if (!err) 1388 return len; 1389 /* 1390 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1391 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1392 * we don't have a good statistic (IpOutDiscards but it can be too many 1393 * things). We could add another new stat but at least for now that 1394 * seems like overkill. 1395 */ 1396 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1397 UDP6_INC_STATS(sock_net(sk), 1398 UDP_MIB_SNDBUFERRORS, is_udplite); 1399 } 1400 return err; 1401 1402 do_confirm: 1403 if (msg->msg_flags & MSG_PROBE) 1404 dst_confirm_neigh(dst, &fl6.daddr); 1405 if (!(msg->msg_flags&MSG_PROBE) || len) 1406 goto back_from_confirm; 1407 err = 0; 1408 goto out; 1409 } 1410 1411 void udpv6_destroy_sock(struct sock *sk) 1412 { 1413 struct udp_sock *up = udp_sk(sk); 1414 lock_sock(sk); 1415 udp_v6_flush_pending_frames(sk); 1416 release_sock(sk); 1417 1418 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 1419 void (*encap_destroy)(struct sock *sk); 1420 encap_destroy = ACCESS_ONCE(up->encap_destroy); 1421 if (encap_destroy) 1422 encap_destroy(sk); 1423 } 1424 1425 inet6_destroy_sock(sk); 1426 } 1427 1428 /* 1429 * Socket option code for UDP 1430 */ 1431 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1432 char __user *optval, unsigned int optlen) 1433 { 1434 if (level == SOL_UDP || level == SOL_UDPLITE) 1435 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1436 udp_v6_push_pending_frames); 1437 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1438 } 1439 1440 #ifdef CONFIG_COMPAT 1441 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1442 char __user *optval, unsigned int optlen) 1443 { 1444 if (level == SOL_UDP || level == SOL_UDPLITE) 1445 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1446 udp_v6_push_pending_frames); 1447 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1448 } 1449 #endif 1450 1451 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1452 char __user *optval, int __user *optlen) 1453 { 1454 if (level == SOL_UDP || level == SOL_UDPLITE) 1455 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1456 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1457 } 1458 1459 #ifdef CONFIG_COMPAT 1460 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1461 char __user *optval, int __user *optlen) 1462 { 1463 if (level == SOL_UDP || level == SOL_UDPLITE) 1464 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1465 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1466 } 1467 #endif 1468 1469 static struct inet6_protocol udpv6_protocol = { 1470 .early_demux = udp_v6_early_demux, 1471 .early_demux_handler = udp_v6_early_demux, 1472 .handler = udpv6_rcv, 1473 .err_handler = udpv6_err, 1474 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1475 }; 1476 1477 /* ------------------------------------------------------------------------ */ 1478 #ifdef CONFIG_PROC_FS 1479 int udp6_seq_show(struct seq_file *seq, void *v) 1480 { 1481 if (v == SEQ_START_TOKEN) { 1482 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1483 } else { 1484 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1485 struct inet_sock *inet = inet_sk(v); 1486 __u16 srcp = ntohs(inet->inet_sport); 1487 __u16 destp = ntohs(inet->inet_dport); 1488 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket); 1489 } 1490 return 0; 1491 } 1492 1493 static const struct file_operations udp6_afinfo_seq_fops = { 1494 .owner = THIS_MODULE, 1495 .open = udp_seq_open, 1496 .read = seq_read, 1497 .llseek = seq_lseek, 1498 .release = seq_release_net 1499 }; 1500 1501 static struct udp_seq_afinfo udp6_seq_afinfo = { 1502 .name = "udp6", 1503 .family = AF_INET6, 1504 .udp_table = &udp_table, 1505 .seq_fops = &udp6_afinfo_seq_fops, 1506 .seq_ops = { 1507 .show = udp6_seq_show, 1508 }, 1509 }; 1510 1511 int __net_init udp6_proc_init(struct net *net) 1512 { 1513 return udp_proc_register(net, &udp6_seq_afinfo); 1514 } 1515 1516 void udp6_proc_exit(struct net *net) 1517 { 1518 udp_proc_unregister(net, &udp6_seq_afinfo); 1519 } 1520 #endif /* CONFIG_PROC_FS */ 1521 1522 /* ------------------------------------------------------------------------ */ 1523 1524 struct proto udpv6_prot = { 1525 .name = "UDPv6", 1526 .owner = THIS_MODULE, 1527 .close = udp_lib_close, 1528 .connect = ip6_datagram_connect, 1529 .disconnect = udp_disconnect, 1530 .ioctl = udp_ioctl, 1531 .init = udp_init_sock, 1532 .destroy = udpv6_destroy_sock, 1533 .setsockopt = udpv6_setsockopt, 1534 .getsockopt = udpv6_getsockopt, 1535 .sendmsg = udpv6_sendmsg, 1536 .recvmsg = udpv6_recvmsg, 1537 .release_cb = ip6_datagram_release_cb, 1538 .hash = udp_lib_hash, 1539 .unhash = udp_lib_unhash, 1540 .rehash = udp_v6_rehash, 1541 .get_port = udp_v6_get_port, 1542 .memory_allocated = &udp_memory_allocated, 1543 .sysctl_mem = sysctl_udp_mem, 1544 .sysctl_wmem = &sysctl_udp_wmem_min, 1545 .sysctl_rmem = &sysctl_udp_rmem_min, 1546 .obj_size = sizeof(struct udp6_sock), 1547 .h.udp_table = &udp_table, 1548 #ifdef CONFIG_COMPAT 1549 .compat_setsockopt = compat_udpv6_setsockopt, 1550 .compat_getsockopt = compat_udpv6_getsockopt, 1551 #endif 1552 .diag_destroy = udp_abort, 1553 }; 1554 1555 static struct inet_protosw udpv6_protosw = { 1556 .type = SOCK_DGRAM, 1557 .protocol = IPPROTO_UDP, 1558 .prot = &udpv6_prot, 1559 .ops = &inet6_dgram_ops, 1560 .flags = INET_PROTOSW_PERMANENT, 1561 }; 1562 1563 int __init udpv6_init(void) 1564 { 1565 int ret; 1566 1567 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1568 if (ret) 1569 goto out; 1570 1571 ret = inet6_register_protosw(&udpv6_protosw); 1572 if (ret) 1573 goto out_udpv6_protocol; 1574 out: 1575 return ret; 1576 1577 out_udpv6_protocol: 1578 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1579 goto out; 1580 } 1581 1582 void udpv6_exit(void) 1583 { 1584 inet6_unregister_protosw(&udpv6_protosw); 1585 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1586 } 1587