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; 152 struct inet_sock *inet; 153 154 if (!net_eq(sock_net(sk), net) || 155 udp_sk(sk)->udp_port_hash != hnum || 156 sk->sk_family != PF_INET6) 157 return -1; 158 159 score = 0; 160 inet = inet_sk(sk); 161 162 if (inet->inet_dport) { 163 if (inet->inet_dport != sport) 164 return -1; 165 score++; 166 } 167 168 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { 169 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 170 return -1; 171 score++; 172 } 173 174 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 175 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 176 return -1; 177 score++; 178 } 179 180 if (sk->sk_bound_dev_if) { 181 if (sk->sk_bound_dev_if != dif) 182 return -1; 183 score++; 184 } 185 186 return score; 187 } 188 189 #define SCORE2_MAX (1 + 1 + 1) 190 static inline int compute_score2(struct sock *sk, struct net *net, 191 const struct in6_addr *saddr, __be16 sport, 192 const struct in6_addr *daddr, 193 unsigned short hnum, int dif) 194 { 195 int score; 196 struct inet_sock *inet; 197 198 if (!net_eq(sock_net(sk), net) || 199 udp_sk(sk)->udp_port_hash != hnum || 200 sk->sk_family != PF_INET6) 201 return -1; 202 203 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 204 return -1; 205 206 score = 0; 207 inet = inet_sk(sk); 208 209 if (inet->inet_dport) { 210 if (inet->inet_dport != sport) 211 return -1; 212 score++; 213 } 214 215 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 216 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 217 return -1; 218 score++; 219 } 220 221 if (sk->sk_bound_dev_if) { 222 if (sk->sk_bound_dev_if != dif) 223 return -1; 224 score++; 225 } 226 227 return score; 228 } 229 230 /* called with read_rcu_lock() */ 231 static struct sock *udp6_lib_lookup2(struct net *net, 232 const struct in6_addr *saddr, __be16 sport, 233 const struct in6_addr *daddr, unsigned int hnum, int dif, 234 struct udp_hslot *hslot2, unsigned int slot2) 235 { 236 struct sock *sk, *result; 237 struct hlist_nulls_node *node; 238 int score, badness, matches = 0, reuseport = 0; 239 u32 hash = 0; 240 241 begin: 242 result = NULL; 243 badness = -1; 244 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 245 score = compute_score2(sk, net, saddr, sport, 246 daddr, hnum, dif); 247 if (score > badness) { 248 result = sk; 249 badness = score; 250 reuseport = sk->sk_reuseport; 251 if (reuseport) { 252 hash = udp6_ehashfn(net, daddr, hnum, 253 saddr, sport); 254 matches = 1; 255 } else if (score == SCORE2_MAX) 256 goto exact_match; 257 } else if (score == badness && reuseport) { 258 matches++; 259 if (reciprocal_scale(hash, matches) == 0) 260 result = sk; 261 hash = next_pseudo_random32(hash); 262 } 263 } 264 /* 265 * if the nulls value we got at the end of this lookup is 266 * not the expected one, we must restart lookup. 267 * We probably met an item that was moved to another chain. 268 */ 269 if (get_nulls_value(node) != slot2) 270 goto begin; 271 272 if (result) { 273 exact_match: 274 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 275 result = NULL; 276 else if (unlikely(compute_score2(result, net, saddr, sport, 277 daddr, hnum, dif) < badness)) { 278 sock_put(result); 279 goto begin; 280 } 281 } 282 return result; 283 } 284 285 struct sock *__udp6_lib_lookup(struct net *net, 286 const struct in6_addr *saddr, __be16 sport, 287 const struct in6_addr *daddr, __be16 dport, 288 int dif, struct udp_table *udptable) 289 { 290 struct sock *sk, *result; 291 struct hlist_nulls_node *node; 292 unsigned short hnum = ntohs(dport); 293 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 294 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 295 int score, badness, matches = 0, reuseport = 0; 296 u32 hash = 0; 297 298 rcu_read_lock(); 299 if (hslot->count > 10) { 300 hash2 = udp6_portaddr_hash(net, daddr, hnum); 301 slot2 = hash2 & udptable->mask; 302 hslot2 = &udptable->hash2[slot2]; 303 if (hslot->count < hslot2->count) 304 goto begin; 305 306 result = udp6_lib_lookup2(net, saddr, sport, 307 daddr, hnum, dif, 308 hslot2, slot2); 309 if (!result) { 310 hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum); 311 slot2 = hash2 & udptable->mask; 312 hslot2 = &udptable->hash2[slot2]; 313 if (hslot->count < hslot2->count) 314 goto begin; 315 316 result = udp6_lib_lookup2(net, saddr, sport, 317 &in6addr_any, hnum, dif, 318 hslot2, slot2); 319 } 320 rcu_read_unlock(); 321 return result; 322 } 323 begin: 324 result = NULL; 325 badness = -1; 326 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 327 score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif); 328 if (score > badness) { 329 result = sk; 330 badness = score; 331 reuseport = sk->sk_reuseport; 332 if (reuseport) { 333 hash = udp6_ehashfn(net, daddr, hnum, 334 saddr, sport); 335 matches = 1; 336 } 337 } else if (score == badness && reuseport) { 338 matches++; 339 if (reciprocal_scale(hash, matches) == 0) 340 result = sk; 341 hash = next_pseudo_random32(hash); 342 } 343 } 344 /* 345 * if the nulls value we got at the end of this lookup is 346 * not the expected one, we must restart lookup. 347 * We probably met an item that was moved to another chain. 348 */ 349 if (get_nulls_value(node) != slot) 350 goto begin; 351 352 if (result) { 353 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 354 result = NULL; 355 else if (unlikely(compute_score(result, net, hnum, saddr, sport, 356 daddr, dport, dif) < badness)) { 357 sock_put(result); 358 goto begin; 359 } 360 } 361 rcu_read_unlock(); 362 return result; 363 } 364 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 365 366 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 367 __be16 sport, __be16 dport, 368 struct udp_table *udptable) 369 { 370 struct sock *sk; 371 const struct ipv6hdr *iph = ipv6_hdr(skb); 372 373 sk = skb_steal_sock(skb); 374 if (unlikely(sk)) 375 return sk; 376 return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport, 377 &iph->daddr, dport, inet6_iif(skb), 378 udptable); 379 } 380 381 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 382 const struct in6_addr *daddr, __be16 dport, int dif) 383 { 384 return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 385 } 386 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 387 388 389 /* 390 * This should be easy, if there is something there we 391 * return it, otherwise we block. 392 */ 393 394 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk, 395 struct msghdr *msg, size_t len, 396 int noblock, int flags, int *addr_len) 397 { 398 struct ipv6_pinfo *np = inet6_sk(sk); 399 struct inet_sock *inet = inet_sk(sk); 400 struct sk_buff *skb; 401 unsigned int ulen, copied; 402 int peeked, off = 0; 403 int err; 404 int is_udplite = IS_UDPLITE(sk); 405 int is_udp4; 406 bool slow; 407 408 if (flags & MSG_ERRQUEUE) 409 return ipv6_recv_error(sk, msg, len, addr_len); 410 411 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 412 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 413 414 try_again: 415 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 416 &peeked, &off, &err); 417 if (!skb) 418 goto out; 419 420 ulen = skb->len - sizeof(struct udphdr); 421 copied = len; 422 if (copied > ulen) 423 copied = ulen; 424 else if (copied < ulen) 425 msg->msg_flags |= MSG_TRUNC; 426 427 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 428 429 /* 430 * If checksum is needed at all, try to do it while copying the 431 * data. If the data is truncated, or if we only want a partial 432 * coverage checksum (UDP-Lite), do it before the copy. 433 */ 434 435 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 436 if (udp_lib_checksum_complete(skb)) 437 goto csum_copy_err; 438 } 439 440 if (skb_csum_unnecessary(skb)) 441 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 442 msg, copied); 443 else { 444 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), msg); 445 if (err == -EINVAL) 446 goto csum_copy_err; 447 } 448 if (unlikely(err)) { 449 trace_kfree_skb(skb, udpv6_recvmsg); 450 if (!peeked) { 451 atomic_inc(&sk->sk_drops); 452 if (is_udp4) 453 UDP_INC_STATS_USER(sock_net(sk), 454 UDP_MIB_INERRORS, 455 is_udplite); 456 else 457 UDP6_INC_STATS_USER(sock_net(sk), 458 UDP_MIB_INERRORS, 459 is_udplite); 460 } 461 goto out_free; 462 } 463 if (!peeked) { 464 if (is_udp4) 465 UDP_INC_STATS_USER(sock_net(sk), 466 UDP_MIB_INDATAGRAMS, is_udplite); 467 else 468 UDP6_INC_STATS_USER(sock_net(sk), 469 UDP_MIB_INDATAGRAMS, is_udplite); 470 } 471 472 sock_recv_ts_and_drops(msg, sk, skb); 473 474 /* Copy the address. */ 475 if (msg->msg_name) { 476 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 477 sin6->sin6_family = AF_INET6; 478 sin6->sin6_port = udp_hdr(skb)->source; 479 sin6->sin6_flowinfo = 0; 480 481 if (is_udp4) { 482 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 483 &sin6->sin6_addr); 484 sin6->sin6_scope_id = 0; 485 } else { 486 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 487 sin6->sin6_scope_id = 488 ipv6_iface_scope_id(&sin6->sin6_addr, 489 inet6_iif(skb)); 490 } 491 *addr_len = sizeof(*sin6); 492 } 493 494 if (np->rxopt.all) 495 ip6_datagram_recv_common_ctl(sk, msg, skb); 496 497 if (is_udp4) { 498 if (inet->cmsg_flags) 499 ip_cmsg_recv(msg, skb); 500 } else { 501 if (np->rxopt.all) 502 ip6_datagram_recv_specific_ctl(sk, msg, skb); 503 } 504 505 err = copied; 506 if (flags & MSG_TRUNC) 507 err = ulen; 508 509 out_free: 510 skb_free_datagram_locked(sk, skb); 511 out: 512 return err; 513 514 csum_copy_err: 515 slow = lock_sock_fast(sk); 516 if (!skb_kill_datagram(sk, skb, flags)) { 517 if (is_udp4) { 518 UDP_INC_STATS_USER(sock_net(sk), 519 UDP_MIB_CSUMERRORS, is_udplite); 520 UDP_INC_STATS_USER(sock_net(sk), 521 UDP_MIB_INERRORS, is_udplite); 522 } else { 523 UDP6_INC_STATS_USER(sock_net(sk), 524 UDP_MIB_CSUMERRORS, is_udplite); 525 UDP6_INC_STATS_USER(sock_net(sk), 526 UDP_MIB_INERRORS, is_udplite); 527 } 528 } 529 unlock_sock_fast(sk, slow); 530 531 if (noblock) 532 return -EAGAIN; 533 534 /* starting over for a new packet */ 535 msg->msg_flags &= ~MSG_TRUNC; 536 goto try_again; 537 } 538 539 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 540 u8 type, u8 code, int offset, __be32 info, 541 struct udp_table *udptable) 542 { 543 struct ipv6_pinfo *np; 544 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 545 const struct in6_addr *saddr = &hdr->saddr; 546 const struct in6_addr *daddr = &hdr->daddr; 547 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 548 struct sock *sk; 549 int err; 550 struct net *net = dev_net(skb->dev); 551 552 sk = __udp6_lib_lookup(net, daddr, uh->dest, 553 saddr, uh->source, inet6_iif(skb), udptable); 554 if (sk == NULL) { 555 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), 556 ICMP6_MIB_INERRORS); 557 return; 558 } 559 560 if (type == ICMPV6_PKT_TOOBIG) { 561 if (!ip6_sk_accept_pmtu(sk)) 562 goto out; 563 ip6_sk_update_pmtu(skb, sk, info); 564 } 565 if (type == NDISC_REDIRECT) { 566 ip6_sk_redirect(skb, sk); 567 goto out; 568 } 569 570 np = inet6_sk(sk); 571 572 if (!icmpv6_err_convert(type, code, &err) && !np->recverr) 573 goto out; 574 575 if (sk->sk_state != TCP_ESTABLISHED && !np->recverr) 576 goto out; 577 578 if (np->recverr) 579 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 580 581 sk->sk_err = err; 582 sk->sk_error_report(sk); 583 out: 584 sock_put(sk); 585 } 586 587 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 588 { 589 int rc; 590 591 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 592 sock_rps_save_rxhash(sk, skb); 593 sk_mark_napi_id(sk, skb); 594 sk_incoming_cpu_update(sk); 595 } 596 597 rc = sock_queue_rcv_skb(sk, skb); 598 if (rc < 0) { 599 int is_udplite = IS_UDPLITE(sk); 600 601 /* Note that an ENOMEM error is charged twice */ 602 if (rc == -ENOMEM) 603 UDP6_INC_STATS_BH(sock_net(sk), 604 UDP_MIB_RCVBUFERRORS, is_udplite); 605 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 606 kfree_skb(skb); 607 return -1; 608 } 609 return 0; 610 } 611 612 static __inline__ void udpv6_err(struct sk_buff *skb, 613 struct inet6_skb_parm *opt, u8 type, 614 u8 code, int offset, __be32 info) 615 { 616 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 617 } 618 619 static struct static_key udpv6_encap_needed __read_mostly; 620 void udpv6_encap_enable(void) 621 { 622 if (!static_key_enabled(&udpv6_encap_needed)) 623 static_key_slow_inc(&udpv6_encap_needed); 624 } 625 EXPORT_SYMBOL(udpv6_encap_enable); 626 627 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 628 { 629 struct udp_sock *up = udp_sk(sk); 630 int rc; 631 int is_udplite = IS_UDPLITE(sk); 632 633 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 634 goto drop; 635 636 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 637 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 638 639 /* 640 * This is an encapsulation socket so pass the skb to 641 * the socket's udp_encap_rcv() hook. Otherwise, just 642 * fall through and pass this up the UDP socket. 643 * up->encap_rcv() returns the following value: 644 * =0 if skb was successfully passed to the encap 645 * handler or was discarded by it. 646 * >0 if skb should be passed on to UDP. 647 * <0 if skb should be resubmitted as proto -N 648 */ 649 650 /* if we're overly short, let UDP handle it */ 651 encap_rcv = ACCESS_ONCE(up->encap_rcv); 652 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { 653 int ret; 654 655 /* Verify checksum before giving to encap */ 656 if (udp_lib_checksum_complete(skb)) 657 goto csum_error; 658 659 ret = encap_rcv(sk, skb); 660 if (ret <= 0) { 661 UDP_INC_STATS_BH(sock_net(sk), 662 UDP_MIB_INDATAGRAMS, 663 is_udplite); 664 return -ret; 665 } 666 } 667 668 /* FALLTHROUGH -- it's a UDP Packet */ 669 } 670 671 /* 672 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 673 */ 674 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 675 676 if (up->pcrlen == 0) { /* full coverage was set */ 677 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 678 UDP_SKB_CB(skb)->cscov, skb->len); 679 goto drop; 680 } 681 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 682 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 683 UDP_SKB_CB(skb)->cscov, up->pcrlen); 684 goto drop; 685 } 686 } 687 688 if (rcu_access_pointer(sk->sk_filter)) { 689 if (udp_lib_checksum_complete(skb)) 690 goto csum_error; 691 } 692 693 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 694 UDP6_INC_STATS_BH(sock_net(sk), 695 UDP_MIB_RCVBUFERRORS, is_udplite); 696 goto drop; 697 } 698 699 skb_dst_drop(skb); 700 701 bh_lock_sock(sk); 702 rc = 0; 703 if (!sock_owned_by_user(sk)) 704 rc = __udpv6_queue_rcv_skb(sk, skb); 705 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 706 bh_unlock_sock(sk); 707 goto drop; 708 } 709 bh_unlock_sock(sk); 710 711 return rc; 712 713 csum_error: 714 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 715 drop: 716 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 717 atomic_inc(&sk->sk_drops); 718 kfree_skb(skb); 719 return -1; 720 } 721 722 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 723 __be16 loc_port, const struct in6_addr *loc_addr, 724 __be16 rmt_port, const struct in6_addr *rmt_addr, 725 int dif, unsigned short hnum) 726 { 727 struct inet_sock *inet = inet_sk(sk); 728 729 if (!net_eq(sock_net(sk), net)) 730 return false; 731 732 if (udp_sk(sk)->udp_port_hash != hnum || 733 sk->sk_family != PF_INET6 || 734 (inet->inet_dport && inet->inet_dport != rmt_port) || 735 (!ipv6_addr_any(&sk->sk_v6_daddr) && 736 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 737 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 738 return false; 739 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 740 return false; 741 return true; 742 } 743 744 static void flush_stack(struct sock **stack, unsigned int count, 745 struct sk_buff *skb, unsigned int final) 746 { 747 struct sk_buff *skb1 = NULL; 748 struct sock *sk; 749 unsigned int i; 750 751 for (i = 0; i < count; i++) { 752 sk = stack[i]; 753 if (likely(skb1 == NULL)) 754 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 755 if (!skb1) { 756 atomic_inc(&sk->sk_drops); 757 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 758 IS_UDPLITE(sk)); 759 UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 760 IS_UDPLITE(sk)); 761 } 762 763 if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0) 764 skb1 = NULL; 765 sock_put(sk); 766 } 767 if (unlikely(skb1)) 768 kfree_skb(skb1); 769 } 770 771 static void udp6_csum_zero_error(struct sk_buff *skb) 772 { 773 /* RFC 2460 section 8.1 says that we SHOULD log 774 * this error. Well, it is reasonable. 775 */ 776 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 777 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 778 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 779 } 780 781 /* 782 * Note: called only from the BH handler context, 783 * so we don't need to lock the hashes. 784 */ 785 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 786 const struct in6_addr *saddr, const struct in6_addr *daddr, 787 struct udp_table *udptable, int proto) 788 { 789 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 790 const struct udphdr *uh = udp_hdr(skb); 791 struct hlist_nulls_node *node; 792 unsigned short hnum = ntohs(uh->dest); 793 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 794 int dif = inet6_iif(skb); 795 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 796 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 797 bool inner_flushed = false; 798 799 if (use_hash2) { 800 hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) & 801 udp_table.mask; 802 hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask; 803 start_lookup: 804 hslot = &udp_table.hash2[hash2]; 805 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 806 } 807 808 spin_lock(&hslot->lock); 809 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 810 if (__udp_v6_is_mcast_sock(net, sk, 811 uh->dest, daddr, 812 uh->source, saddr, 813 dif, hnum) && 814 /* If zero checksum and no_check is not on for 815 * the socket then skip it. 816 */ 817 (uh->check || udp_sk(sk)->no_check6_rx)) { 818 if (unlikely(count == ARRAY_SIZE(stack))) { 819 flush_stack(stack, count, skb, ~0); 820 inner_flushed = true; 821 count = 0; 822 } 823 stack[count++] = sk; 824 sock_hold(sk); 825 } 826 } 827 828 spin_unlock(&hslot->lock); 829 830 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 831 if (use_hash2 && hash2 != hash2_any) { 832 hash2 = hash2_any; 833 goto start_lookup; 834 } 835 836 if (count) { 837 flush_stack(stack, count, skb, count - 1); 838 } else { 839 if (!inner_flushed) 840 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 841 proto == IPPROTO_UDPLITE); 842 consume_skb(skb); 843 } 844 return 0; 845 } 846 847 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 848 int proto) 849 { 850 struct net *net = dev_net(skb->dev); 851 struct sock *sk; 852 struct udphdr *uh; 853 const struct in6_addr *saddr, *daddr; 854 u32 ulen = 0; 855 856 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 857 goto discard; 858 859 saddr = &ipv6_hdr(skb)->saddr; 860 daddr = &ipv6_hdr(skb)->daddr; 861 uh = udp_hdr(skb); 862 863 ulen = ntohs(uh->len); 864 if (ulen > skb->len) 865 goto short_packet; 866 867 if (proto == IPPROTO_UDP) { 868 /* UDP validates ulen. */ 869 870 /* Check for jumbo payload */ 871 if (ulen == 0) 872 ulen = skb->len; 873 874 if (ulen < sizeof(*uh)) 875 goto short_packet; 876 877 if (ulen < skb->len) { 878 if (pskb_trim_rcsum(skb, ulen)) 879 goto short_packet; 880 saddr = &ipv6_hdr(skb)->saddr; 881 daddr = &ipv6_hdr(skb)->daddr; 882 uh = udp_hdr(skb); 883 } 884 } 885 886 if (udp6_csum_init(skb, uh, proto)) 887 goto csum_error; 888 889 /* 890 * Multicast receive code 891 */ 892 if (ipv6_addr_is_multicast(daddr)) 893 return __udp6_lib_mcast_deliver(net, skb, 894 saddr, daddr, udptable, proto); 895 896 /* Unicast */ 897 898 /* 899 * check socket cache ... must talk to Alan about his plans 900 * for sock caches... i'll skip this for now. 901 */ 902 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 903 if (sk != NULL) { 904 int ret; 905 906 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 907 sock_put(sk); 908 udp6_csum_zero_error(skb); 909 goto csum_error; 910 } 911 912 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 913 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 914 ip6_compute_pseudo); 915 916 ret = udpv6_queue_rcv_skb(sk, skb); 917 sock_put(sk); 918 919 /* a return value > 0 means to resubmit the input, but 920 * it wants the return to be -protocol, or 0 921 */ 922 if (ret > 0) 923 return -ret; 924 925 return 0; 926 } 927 928 if (!uh->check) { 929 udp6_csum_zero_error(skb); 930 goto csum_error; 931 } 932 933 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 934 goto discard; 935 936 if (udp_lib_checksum_complete(skb)) 937 goto csum_error; 938 939 UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 940 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 941 942 kfree_skb(skb); 943 return 0; 944 945 short_packet: 946 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 947 proto == IPPROTO_UDPLITE ? "-Lite" : "", 948 saddr, ntohs(uh->source), 949 ulen, skb->len, 950 daddr, ntohs(uh->dest)); 951 goto discard; 952 csum_error: 953 UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 954 discard: 955 UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 956 kfree_skb(skb); 957 return 0; 958 } 959 960 static __inline__ int udpv6_rcv(struct sk_buff *skb) 961 { 962 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 963 } 964 965 /* 966 * Throw away all pending data and cancel the corking. Socket is locked. 967 */ 968 static void udp_v6_flush_pending_frames(struct sock *sk) 969 { 970 struct udp_sock *up = udp_sk(sk); 971 972 if (up->pending == AF_INET) 973 udp_flush_pending_frames(sk); 974 else if (up->pending) { 975 up->len = 0; 976 up->pending = 0; 977 ip6_flush_pending_frames(sk); 978 } 979 } 980 981 /** 982 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 983 * @sk: socket we are sending on 984 * @skb: sk_buff containing the filled-in UDP header 985 * (checksum field must be zeroed out) 986 */ 987 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 988 const struct in6_addr *saddr, 989 const struct in6_addr *daddr, int len) 990 { 991 unsigned int offset; 992 struct udphdr *uh = udp_hdr(skb); 993 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 994 __wsum csum = 0; 995 996 if (!frags) { 997 /* Only one fragment on the socket. */ 998 skb->csum_start = skb_transport_header(skb) - skb->head; 999 skb->csum_offset = offsetof(struct udphdr, check); 1000 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1001 } else { 1002 /* 1003 * HW-checksum won't work as there are two or more 1004 * fragments on the socket so that all csums of sk_buffs 1005 * should be together 1006 */ 1007 offset = skb_transport_offset(skb); 1008 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1009 1010 skb->ip_summed = CHECKSUM_NONE; 1011 1012 do { 1013 csum = csum_add(csum, frags->csum); 1014 } while ((frags = frags->next)); 1015 1016 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1017 csum); 1018 if (uh->check == 0) 1019 uh->check = CSUM_MANGLED_0; 1020 } 1021 } 1022 1023 /* 1024 * Sending 1025 */ 1026 1027 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6) 1028 { 1029 struct sock *sk = skb->sk; 1030 struct udphdr *uh; 1031 int err = 0; 1032 int is_udplite = IS_UDPLITE(sk); 1033 __wsum csum = 0; 1034 int offset = skb_transport_offset(skb); 1035 int len = skb->len - offset; 1036 1037 /* 1038 * Create a UDP header 1039 */ 1040 uh = udp_hdr(skb); 1041 uh->source = fl6->fl6_sport; 1042 uh->dest = fl6->fl6_dport; 1043 uh->len = htons(len); 1044 uh->check = 0; 1045 1046 if (is_udplite) 1047 csum = udplite_csum(skb); 1048 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1049 skb->ip_summed = CHECKSUM_NONE; 1050 goto send; 1051 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1052 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1053 goto send; 1054 } else 1055 csum = udp_csum(skb); 1056 1057 /* add protocol-dependent pseudo-header */ 1058 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1059 len, fl6->flowi6_proto, csum); 1060 if (uh->check == 0) 1061 uh->check = CSUM_MANGLED_0; 1062 1063 send: 1064 err = ip6_send_skb(skb); 1065 if (err) { 1066 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1067 UDP6_INC_STATS_USER(sock_net(sk), 1068 UDP_MIB_SNDBUFERRORS, is_udplite); 1069 err = 0; 1070 } 1071 } else 1072 UDP6_INC_STATS_USER(sock_net(sk), 1073 UDP_MIB_OUTDATAGRAMS, is_udplite); 1074 return err; 1075 } 1076 1077 static int udp_v6_push_pending_frames(struct sock *sk) 1078 { 1079 struct sk_buff *skb; 1080 struct udp_sock *up = udp_sk(sk); 1081 struct flowi6 fl6; 1082 int err = 0; 1083 1084 if (up->pending == AF_INET) 1085 return udp_push_pending_frames(sk); 1086 1087 /* ip6_finish_skb will release the cork, so make a copy of 1088 * fl6 here. 1089 */ 1090 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1091 1092 skb = ip6_finish_skb(sk); 1093 if (!skb) 1094 goto out; 1095 1096 err = udp_v6_send_skb(skb, &fl6); 1097 1098 out: 1099 up->len = 0; 1100 up->pending = 0; 1101 return err; 1102 } 1103 1104 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk, 1105 struct msghdr *msg, size_t len) 1106 { 1107 struct ipv6_txoptions opt_space; 1108 struct udp_sock *up = udp_sk(sk); 1109 struct inet_sock *inet = inet_sk(sk); 1110 struct ipv6_pinfo *np = inet6_sk(sk); 1111 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1112 struct in6_addr *daddr, *final_p, final; 1113 struct ipv6_txoptions *opt = NULL; 1114 struct ip6_flowlabel *flowlabel = NULL; 1115 struct flowi6 fl6; 1116 struct dst_entry *dst; 1117 int addr_len = msg->msg_namelen; 1118 int ulen = len; 1119 int hlimit = -1; 1120 int tclass = -1; 1121 int dontfrag = -1; 1122 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1123 int err; 1124 int connected = 0; 1125 int is_udplite = IS_UDPLITE(sk); 1126 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 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 break; 1139 case AF_INET: 1140 goto do_udp_sendmsg; 1141 case AF_UNSPEC: 1142 msg->msg_name = sin6 = NULL; 1143 msg->msg_namelen = addr_len = 0; 1144 daddr = NULL; 1145 break; 1146 default: 1147 return -EINVAL; 1148 } 1149 } else if (!up->pending) { 1150 if (sk->sk_state != TCP_ESTABLISHED) 1151 return -EDESTADDRREQ; 1152 daddr = &sk->sk_v6_daddr; 1153 } else 1154 daddr = NULL; 1155 1156 if (daddr) { 1157 if (ipv6_addr_v4mapped(daddr)) { 1158 struct sockaddr_in sin; 1159 sin.sin_family = AF_INET; 1160 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1161 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1162 msg->msg_name = &sin; 1163 msg->msg_namelen = sizeof(sin); 1164 do_udp_sendmsg: 1165 if (__ipv6_only_sock(sk)) 1166 return -ENETUNREACH; 1167 return udp_sendmsg(iocb, sk, msg, len); 1168 } 1169 } 1170 1171 if (up->pending == AF_INET) 1172 return udp_sendmsg(iocb, sk, msg, len); 1173 1174 /* Rough check on arithmetic overflow, 1175 better check is made in ip6_append_data(). 1176 */ 1177 if (len > INT_MAX - sizeof(struct udphdr)) 1178 return -EMSGSIZE; 1179 1180 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1181 if (up->pending) { 1182 /* 1183 * There are pending frames. 1184 * The socket lock must be held while it's corked. 1185 */ 1186 lock_sock(sk); 1187 if (likely(up->pending)) { 1188 if (unlikely(up->pending != AF_INET6)) { 1189 release_sock(sk); 1190 return -EAFNOSUPPORT; 1191 } 1192 dst = NULL; 1193 goto do_append_data; 1194 } 1195 release_sock(sk); 1196 } 1197 ulen += sizeof(struct udphdr); 1198 1199 memset(&fl6, 0, sizeof(fl6)); 1200 1201 if (sin6) { 1202 if (sin6->sin6_port == 0) 1203 return -EINVAL; 1204 1205 fl6.fl6_dport = sin6->sin6_port; 1206 daddr = &sin6->sin6_addr; 1207 1208 if (np->sndflow) { 1209 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1210 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1211 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1212 if (flowlabel == NULL) 1213 return -EINVAL; 1214 } 1215 } 1216 1217 /* 1218 * Otherwise it will be difficult to maintain 1219 * sk->sk_dst_cache. 1220 */ 1221 if (sk->sk_state == TCP_ESTABLISHED && 1222 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1223 daddr = &sk->sk_v6_daddr; 1224 1225 if (addr_len >= sizeof(struct sockaddr_in6) && 1226 sin6->sin6_scope_id && 1227 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1228 fl6.flowi6_oif = sin6->sin6_scope_id; 1229 } else { 1230 if (sk->sk_state != TCP_ESTABLISHED) 1231 return -EDESTADDRREQ; 1232 1233 fl6.fl6_dport = inet->inet_dport; 1234 daddr = &sk->sk_v6_daddr; 1235 fl6.flowlabel = np->flow_label; 1236 connected = 1; 1237 } 1238 1239 if (!fl6.flowi6_oif) 1240 fl6.flowi6_oif = sk->sk_bound_dev_if; 1241 1242 if (!fl6.flowi6_oif) 1243 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1244 1245 fl6.flowi6_mark = sk->sk_mark; 1246 1247 if (msg->msg_controllen) { 1248 opt = &opt_space; 1249 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1250 opt->tot_len = sizeof(*opt); 1251 1252 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt, 1253 &hlimit, &tclass, &dontfrag); 1254 if (err < 0) { 1255 fl6_sock_release(flowlabel); 1256 return err; 1257 } 1258 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1259 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1260 if (flowlabel == NULL) 1261 return -EINVAL; 1262 } 1263 if (!(opt->opt_nflen|opt->opt_flen)) 1264 opt = NULL; 1265 connected = 0; 1266 } 1267 if (opt == NULL) 1268 opt = np->opt; 1269 if (flowlabel) 1270 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1271 opt = ipv6_fixup_options(&opt_space, opt); 1272 1273 fl6.flowi6_proto = sk->sk_protocol; 1274 if (!ipv6_addr_any(daddr)) 1275 fl6.daddr = *daddr; 1276 else 1277 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1278 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1279 fl6.saddr = np->saddr; 1280 fl6.fl6_sport = inet->inet_sport; 1281 1282 final_p = fl6_update_dst(&fl6, opt, &final); 1283 if (final_p) 1284 connected = 0; 1285 1286 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1287 fl6.flowi6_oif = np->mcast_oif; 1288 connected = 0; 1289 } else if (!fl6.flowi6_oif) 1290 fl6.flowi6_oif = np->ucast_oif; 1291 1292 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1293 1294 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p); 1295 if (IS_ERR(dst)) { 1296 err = PTR_ERR(dst); 1297 dst = NULL; 1298 goto out; 1299 } 1300 1301 if (hlimit < 0) 1302 hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1303 1304 if (tclass < 0) 1305 tclass = np->tclass; 1306 1307 if (msg->msg_flags&MSG_CONFIRM) 1308 goto do_confirm; 1309 back_from_confirm: 1310 1311 /* Lockless fast path for the non-corking case */ 1312 if (!corkreq) { 1313 struct sk_buff *skb; 1314 1315 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1316 sizeof(struct udphdr), hlimit, tclass, opt, 1317 &fl6, (struct rt6_info *)dst, 1318 msg->msg_flags, dontfrag); 1319 err = PTR_ERR(skb); 1320 if (!IS_ERR_OR_NULL(skb)) 1321 err = udp_v6_send_skb(skb, &fl6); 1322 goto release_dst; 1323 } 1324 1325 lock_sock(sk); 1326 if (unlikely(up->pending)) { 1327 /* The socket is already corked while preparing it. */ 1328 /* ... which is an evident application bug. --ANK */ 1329 release_sock(sk); 1330 1331 net_dbg_ratelimited("udp cork app bug 2\n"); 1332 err = -EINVAL; 1333 goto out; 1334 } 1335 1336 up->pending = AF_INET6; 1337 1338 do_append_data: 1339 if (dontfrag < 0) 1340 dontfrag = np->dontfrag; 1341 up->len += ulen; 1342 err = ip6_append_data(sk, getfrag, msg, ulen, 1343 sizeof(struct udphdr), hlimit, tclass, opt, &fl6, 1344 (struct rt6_info *)dst, 1345 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag); 1346 if (err) 1347 udp_v6_flush_pending_frames(sk); 1348 else if (!corkreq) 1349 err = udp_v6_push_pending_frames(sk); 1350 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1351 up->pending = 0; 1352 1353 if (err > 0) 1354 err = np->recverr ? net_xmit_errno(err) : 0; 1355 release_sock(sk); 1356 1357 release_dst: 1358 if (dst) { 1359 if (connected) { 1360 ip6_dst_store(sk, dst, 1361 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ? 1362 &sk->sk_v6_daddr : NULL, 1363 #ifdef CONFIG_IPV6_SUBTREES 1364 ipv6_addr_equal(&fl6.saddr, &np->saddr) ? 1365 &np->saddr : 1366 #endif 1367 NULL); 1368 } else { 1369 dst_release(dst); 1370 } 1371 dst = NULL; 1372 } 1373 1374 out: 1375 dst_release(dst); 1376 fl6_sock_release(flowlabel); 1377 if (!err) 1378 return len; 1379 /* 1380 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1381 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1382 * we don't have a good statistic (IpOutDiscards but it can be too many 1383 * things). We could add another new stat but at least for now that 1384 * seems like overkill. 1385 */ 1386 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1387 UDP6_INC_STATS_USER(sock_net(sk), 1388 UDP_MIB_SNDBUFERRORS, is_udplite); 1389 } 1390 return err; 1391 1392 do_confirm: 1393 dst_confirm(dst); 1394 if (!(msg->msg_flags&MSG_PROBE) || len) 1395 goto back_from_confirm; 1396 err = 0; 1397 goto out; 1398 } 1399 1400 void udpv6_destroy_sock(struct sock *sk) 1401 { 1402 struct udp_sock *up = udp_sk(sk); 1403 lock_sock(sk); 1404 udp_v6_flush_pending_frames(sk); 1405 release_sock(sk); 1406 1407 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 1408 void (*encap_destroy)(struct sock *sk); 1409 encap_destroy = ACCESS_ONCE(up->encap_destroy); 1410 if (encap_destroy) 1411 encap_destroy(sk); 1412 } 1413 1414 inet6_destroy_sock(sk); 1415 } 1416 1417 /* 1418 * Socket option code for UDP 1419 */ 1420 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1421 char __user *optval, unsigned int optlen) 1422 { 1423 if (level == SOL_UDP || level == SOL_UDPLITE) 1424 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1425 udp_v6_push_pending_frames); 1426 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1427 } 1428 1429 #ifdef CONFIG_COMPAT 1430 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1431 char __user *optval, unsigned int optlen) 1432 { 1433 if (level == SOL_UDP || level == SOL_UDPLITE) 1434 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1435 udp_v6_push_pending_frames); 1436 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1437 } 1438 #endif 1439 1440 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1441 char __user *optval, int __user *optlen) 1442 { 1443 if (level == SOL_UDP || level == SOL_UDPLITE) 1444 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1445 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1446 } 1447 1448 #ifdef CONFIG_COMPAT 1449 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1450 char __user *optval, int __user *optlen) 1451 { 1452 if (level == SOL_UDP || level == SOL_UDPLITE) 1453 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1454 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1455 } 1456 #endif 1457 1458 static const struct inet6_protocol udpv6_protocol = { 1459 .handler = udpv6_rcv, 1460 .err_handler = udpv6_err, 1461 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1462 }; 1463 1464 /* ------------------------------------------------------------------------ */ 1465 #ifdef CONFIG_PROC_FS 1466 int udp6_seq_show(struct seq_file *seq, void *v) 1467 { 1468 if (v == SEQ_START_TOKEN) { 1469 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1470 } else { 1471 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1472 struct inet_sock *inet = inet_sk(v); 1473 __u16 srcp = ntohs(inet->inet_sport); 1474 __u16 destp = ntohs(inet->inet_dport); 1475 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket); 1476 } 1477 return 0; 1478 } 1479 1480 static const struct file_operations udp6_afinfo_seq_fops = { 1481 .owner = THIS_MODULE, 1482 .open = udp_seq_open, 1483 .read = seq_read, 1484 .llseek = seq_lseek, 1485 .release = seq_release_net 1486 }; 1487 1488 static struct udp_seq_afinfo udp6_seq_afinfo = { 1489 .name = "udp6", 1490 .family = AF_INET6, 1491 .udp_table = &udp_table, 1492 .seq_fops = &udp6_afinfo_seq_fops, 1493 .seq_ops = { 1494 .show = udp6_seq_show, 1495 }, 1496 }; 1497 1498 int __net_init udp6_proc_init(struct net *net) 1499 { 1500 return udp_proc_register(net, &udp6_seq_afinfo); 1501 } 1502 1503 void udp6_proc_exit(struct net *net) { 1504 udp_proc_unregister(net, &udp6_seq_afinfo); 1505 } 1506 #endif /* CONFIG_PROC_FS */ 1507 1508 void udp_v6_clear_sk(struct sock *sk, int size) 1509 { 1510 struct inet_sock *inet = inet_sk(sk); 1511 1512 /* we do not want to clear pinet6 field, because of RCU lookups */ 1513 sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6)); 1514 1515 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6); 1516 memset(&inet->pinet6 + 1, 0, size); 1517 } 1518 1519 /* ------------------------------------------------------------------------ */ 1520 1521 struct proto udpv6_prot = { 1522 .name = "UDPv6", 1523 .owner = THIS_MODULE, 1524 .close = udp_lib_close, 1525 .connect = ip6_datagram_connect, 1526 .disconnect = udp_disconnect, 1527 .ioctl = udp_ioctl, 1528 .destroy = udpv6_destroy_sock, 1529 .setsockopt = udpv6_setsockopt, 1530 .getsockopt = udpv6_getsockopt, 1531 .sendmsg = udpv6_sendmsg, 1532 .recvmsg = udpv6_recvmsg, 1533 .backlog_rcv = __udpv6_queue_rcv_skb, 1534 .hash = udp_lib_hash, 1535 .unhash = udp_lib_unhash, 1536 .rehash = udp_v6_rehash, 1537 .get_port = udp_v6_get_port, 1538 .memory_allocated = &udp_memory_allocated, 1539 .sysctl_mem = sysctl_udp_mem, 1540 .sysctl_wmem = &sysctl_udp_wmem_min, 1541 .sysctl_rmem = &sysctl_udp_rmem_min, 1542 .obj_size = sizeof(struct udp6_sock), 1543 .slab_flags = SLAB_DESTROY_BY_RCU, 1544 .h.udp_table = &udp_table, 1545 #ifdef CONFIG_COMPAT 1546 .compat_setsockopt = compat_udpv6_setsockopt, 1547 .compat_getsockopt = compat_udpv6_getsockopt, 1548 #endif 1549 .clear_sk = udp_v6_clear_sk, 1550 }; 1551 1552 static struct inet_protosw udpv6_protosw = { 1553 .type = SOCK_DGRAM, 1554 .protocol = IPPROTO_UDP, 1555 .prot = &udpv6_prot, 1556 .ops = &inet6_dgram_ops, 1557 .flags = INET_PROTOSW_PERMANENT, 1558 }; 1559 1560 1561 int __init udpv6_init(void) 1562 { 1563 int ret; 1564 1565 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1566 if (ret) 1567 goto out; 1568 1569 ret = inet6_register_protosw(&udpv6_protosw); 1570 if (ret) 1571 goto out_udpv6_protocol; 1572 out: 1573 return ret; 1574 1575 out_udpv6_protocol: 1576 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1577 goto out; 1578 } 1579 1580 void udpv6_exit(void) 1581 { 1582 inet6_unregister_protosw(&udpv6_protosw); 1583 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1584 } 1585