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/ip6_tunnel.h> 49 #include <net/xfrm.h> 50 #include <net/inet_hashtables.h> 51 #include <net/inet6_hashtables.h> 52 #include <net/busy_poll.h> 53 #include <net/sock_reuseport.h> 54 55 #include <linux/proc_fs.h> 56 #include <linux/seq_file.h> 57 #include <trace/events/skb.h> 58 #include "udp_impl.h" 59 60 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb) 61 { 62 #if defined(CONFIG_NET_L3_MASTER_DEV) 63 if (!net->ipv4.sysctl_udp_l3mdev_accept && 64 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags)) 65 return true; 66 #endif 67 return false; 68 } 69 70 static u32 udp6_ehashfn(const struct net *net, 71 const struct in6_addr *laddr, 72 const u16 lport, 73 const struct in6_addr *faddr, 74 const __be16 fport) 75 { 76 static u32 udp6_ehash_secret __read_mostly; 77 static u32 udp_ipv6_hash_secret __read_mostly; 78 79 u32 lhash, fhash; 80 81 net_get_random_once(&udp6_ehash_secret, 82 sizeof(udp6_ehash_secret)); 83 net_get_random_once(&udp_ipv6_hash_secret, 84 sizeof(udp_ipv6_hash_secret)); 85 86 lhash = (__force u32)laddr->s6_addr32[3]; 87 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 88 89 return __inet6_ehashfn(lhash, lport, fhash, fport, 90 udp_ipv6_hash_secret + net_hash_mix(net)); 91 } 92 93 int udp_v6_get_port(struct sock *sk, unsigned short snum) 94 { 95 unsigned int hash2_nulladdr = 96 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 97 unsigned int hash2_partial = 98 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 99 100 /* precompute partial secondary hash */ 101 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 102 return udp_lib_get_port(sk, snum, hash2_nulladdr); 103 } 104 105 static void udp_v6_rehash(struct sock *sk) 106 { 107 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 108 &sk->sk_v6_rcv_saddr, 109 inet_sk(sk)->inet_num); 110 111 udp_lib_rehash(sk, new_hash); 112 } 113 114 static int compute_score(struct sock *sk, struct net *net, 115 const struct in6_addr *saddr, __be16 sport, 116 const struct in6_addr *daddr, unsigned short hnum, 117 int dif, int sdif, bool exact_dif) 118 { 119 int score; 120 struct inet_sock *inet; 121 bool dev_match; 122 123 if (!net_eq(sock_net(sk), net) || 124 udp_sk(sk)->udp_port_hash != hnum || 125 sk->sk_family != PF_INET6) 126 return -1; 127 128 score = 0; 129 inet = inet_sk(sk); 130 131 if (inet->inet_dport) { 132 if (inet->inet_dport != sport) 133 return -1; 134 score++; 135 } 136 137 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { 138 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 139 return -1; 140 score++; 141 } 142 143 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 144 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 145 return -1; 146 score++; 147 } 148 149 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif); 150 if (!dev_match) 151 return -1; 152 score++; 153 154 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 155 score++; 156 157 return score; 158 } 159 160 /* called with rcu_read_lock() */ 161 static struct sock *udp6_lib_lookup2(struct net *net, 162 const struct in6_addr *saddr, __be16 sport, 163 const struct in6_addr *daddr, unsigned int hnum, 164 int dif, int sdif, bool exact_dif, 165 struct udp_hslot *hslot2, struct sk_buff *skb) 166 { 167 struct sock *sk, *result; 168 int score, badness; 169 u32 hash = 0; 170 171 result = NULL; 172 badness = -1; 173 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 174 score = compute_score(sk, net, saddr, sport, 175 daddr, hnum, dif, sdif, exact_dif); 176 if (score > badness) { 177 if (sk->sk_reuseport) { 178 hash = udp6_ehashfn(net, daddr, hnum, 179 saddr, sport); 180 181 result = reuseport_select_sock(sk, hash, skb, 182 sizeof(struct udphdr)); 183 if (result) 184 return result; 185 } 186 result = sk; 187 badness = score; 188 } 189 } 190 return result; 191 } 192 193 /* rcu_read_lock() must be held */ 194 struct sock *__udp6_lib_lookup(struct net *net, 195 const struct in6_addr *saddr, __be16 sport, 196 const struct in6_addr *daddr, __be16 dport, 197 int dif, int sdif, struct udp_table *udptable, 198 struct sk_buff *skb) 199 { 200 struct sock *sk, *result; 201 unsigned short hnum = ntohs(dport); 202 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 203 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 204 bool exact_dif = udp6_lib_exact_dif_match(net, skb); 205 int score, badness; 206 u32 hash = 0; 207 208 if (hslot->count > 10) { 209 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 210 slot2 = hash2 & udptable->mask; 211 hslot2 = &udptable->hash2[slot2]; 212 if (hslot->count < hslot2->count) 213 goto begin; 214 215 result = udp6_lib_lookup2(net, saddr, sport, 216 daddr, hnum, dif, sdif, exact_dif, 217 hslot2, skb); 218 if (!result) { 219 unsigned int old_slot2 = slot2; 220 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 221 slot2 = hash2 & udptable->mask; 222 /* avoid searching the same slot again. */ 223 if (unlikely(slot2 == old_slot2)) 224 return result; 225 226 hslot2 = &udptable->hash2[slot2]; 227 if (hslot->count < hslot2->count) 228 goto begin; 229 230 result = udp6_lib_lookup2(net, saddr, sport, 231 daddr, hnum, dif, sdif, 232 exact_dif, hslot2, 233 skb); 234 } 235 if (unlikely(IS_ERR(result))) 236 return NULL; 237 return result; 238 } 239 begin: 240 result = NULL; 241 badness = -1; 242 sk_for_each_rcu(sk, &hslot->head) { 243 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif, 244 sdif, exact_dif); 245 if (score > badness) { 246 if (sk->sk_reuseport) { 247 hash = udp6_ehashfn(net, daddr, hnum, 248 saddr, sport); 249 result = reuseport_select_sock(sk, hash, skb, 250 sizeof(struct udphdr)); 251 if (unlikely(IS_ERR(result))) 252 return NULL; 253 if (result) 254 return result; 255 } 256 result = sk; 257 badness = score; 258 } 259 } 260 return result; 261 } 262 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 263 264 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 265 __be16 sport, __be16 dport, 266 struct udp_table *udptable) 267 { 268 const struct ipv6hdr *iph = ipv6_hdr(skb); 269 270 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 271 &iph->daddr, dport, inet6_iif(skb), 272 inet6_sdif(skb), udptable, skb); 273 } 274 275 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, 276 __be16 sport, __be16 dport) 277 { 278 const struct ipv6hdr *iph = ipv6_hdr(skb); 279 280 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 281 &iph->daddr, dport, inet6_iif(skb), 282 inet6_sdif(skb), &udp_table, skb); 283 } 284 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); 285 286 /* Must be called under rcu_read_lock(). 287 * Does increment socket refcount. 288 */ 289 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 290 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 291 const struct in6_addr *daddr, __be16 dport, int dif) 292 { 293 struct sock *sk; 294 295 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 296 dif, 0, &udp_table, NULL); 297 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 298 sk = NULL; 299 return sk; 300 } 301 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 302 #endif 303 304 /* do not use the scratch area len for jumbogram: their length execeeds the 305 * scratch area space; note that the IP6CB flags is still in the first 306 * cacheline, so checking for jumbograms is cheap 307 */ 308 static int udp6_skb_len(struct sk_buff *skb) 309 { 310 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 311 } 312 313 /* 314 * This should be easy, if there is something there we 315 * return it, otherwise we block. 316 */ 317 318 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 319 int noblock, int flags, int *addr_len) 320 { 321 struct ipv6_pinfo *np = inet6_sk(sk); 322 struct inet_sock *inet = inet_sk(sk); 323 struct sk_buff *skb; 324 unsigned int ulen, copied; 325 int peeked, peeking, off; 326 int err; 327 int is_udplite = IS_UDPLITE(sk); 328 bool checksum_valid = false; 329 struct udp_mib *mib; 330 int is_udp4; 331 332 if (flags & MSG_ERRQUEUE) 333 return ipv6_recv_error(sk, msg, len, addr_len); 334 335 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 336 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 337 338 try_again: 339 peeking = flags & MSG_PEEK; 340 off = sk_peek_offset(sk, flags); 341 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); 342 if (!skb) 343 return err; 344 345 ulen = udp6_skb_len(skb); 346 copied = len; 347 if (copied > ulen - off) 348 copied = ulen - off; 349 else if (copied < ulen) 350 msg->msg_flags |= MSG_TRUNC; 351 352 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 353 mib = __UDPX_MIB(sk, is_udp4); 354 355 /* 356 * If checksum is needed at all, try to do it while copying the 357 * data. If the data is truncated, or if we only want a partial 358 * coverage checksum (UDP-Lite), do it before the copy. 359 */ 360 361 if (copied < ulen || peeking || 362 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 363 checksum_valid = udp_skb_csum_unnecessary(skb) || 364 !__udp_lib_checksum_complete(skb); 365 if (!checksum_valid) 366 goto csum_copy_err; 367 } 368 369 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 370 if (udp_skb_is_linear(skb)) 371 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 372 else 373 err = skb_copy_datagram_msg(skb, off, msg, copied); 374 } else { 375 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 376 if (err == -EINVAL) 377 goto csum_copy_err; 378 } 379 if (unlikely(err)) { 380 if (!peeked) { 381 atomic_inc(&sk->sk_drops); 382 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 383 } 384 kfree_skb(skb); 385 return err; 386 } 387 if (!peeked) 388 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 389 390 sock_recv_ts_and_drops(msg, sk, skb); 391 392 /* Copy the address. */ 393 if (msg->msg_name) { 394 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 395 sin6->sin6_family = AF_INET6; 396 sin6->sin6_port = udp_hdr(skb)->source; 397 sin6->sin6_flowinfo = 0; 398 399 if (is_udp4) { 400 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 401 &sin6->sin6_addr); 402 sin6->sin6_scope_id = 0; 403 } else { 404 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 405 sin6->sin6_scope_id = 406 ipv6_iface_scope_id(&sin6->sin6_addr, 407 inet6_iif(skb)); 408 } 409 *addr_len = sizeof(*sin6); 410 } 411 412 if (udp_sk(sk)->gro_enabled) 413 udp_cmsg_recv(msg, sk, skb); 414 415 if (np->rxopt.all) 416 ip6_datagram_recv_common_ctl(sk, msg, skb); 417 418 if (is_udp4) { 419 if (inet->cmsg_flags) 420 ip_cmsg_recv_offset(msg, sk, skb, 421 sizeof(struct udphdr), off); 422 } else { 423 if (np->rxopt.all) 424 ip6_datagram_recv_specific_ctl(sk, msg, skb); 425 } 426 427 err = copied; 428 if (flags & MSG_TRUNC) 429 err = ulen; 430 431 skb_consume_udp(sk, skb, peeking ? -err : err); 432 return err; 433 434 csum_copy_err: 435 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 436 udp_skb_destructor)) { 437 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 438 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 439 } 440 kfree_skb(skb); 441 442 /* starting over for a new packet, but check if we need to yield */ 443 cond_resched(); 444 msg->msg_flags &= ~MSG_TRUNC; 445 goto try_again; 446 } 447 448 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 449 void udpv6_encap_enable(void) 450 { 451 static_branch_enable(&udpv6_encap_needed_key); 452 } 453 EXPORT_SYMBOL(udpv6_encap_enable); 454 455 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 456 * through error handlers in encapsulations looking for a match. 457 */ 458 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 459 struct inet6_skb_parm *opt, 460 u8 type, u8 code, int offset, u32 info) 461 { 462 int i; 463 464 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 465 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 466 u8 type, u8 code, int offset, u32 info); 467 468 if (!ip6tun_encaps[i]) 469 continue; 470 handler = rcu_dereference(ip6tun_encaps[i]->err_handler); 471 if (handler && !handler(skb, opt, type, code, offset, info)) 472 return 0; 473 } 474 475 return -ENOENT; 476 } 477 478 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 479 * reversing source and destination port: this will match tunnels that force the 480 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 481 * lwtunnels might actually break this assumption by being configured with 482 * different destination ports on endpoints, in this case we won't be able to 483 * trace ICMP messages back to them. 484 * 485 * If this doesn't match any socket, probe tunnels with arbitrary destination 486 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 487 * we've sent packets to won't necessarily match the local destination port. 488 * 489 * Then ask the tunnel implementation to match the error against a valid 490 * association. 491 * 492 * Return an error if we can't find a match, the socket if we need further 493 * processing, zero otherwise. 494 */ 495 static struct sock *__udp6_lib_err_encap(struct net *net, 496 const struct ipv6hdr *hdr, int offset, 497 struct udphdr *uh, 498 struct udp_table *udptable, 499 struct sk_buff *skb, 500 struct inet6_skb_parm *opt, 501 u8 type, u8 code, __be32 info) 502 { 503 int network_offset, transport_offset; 504 struct sock *sk; 505 506 network_offset = skb_network_offset(skb); 507 transport_offset = skb_transport_offset(skb); 508 509 /* Network header needs to point to the outer IPv6 header inside ICMP */ 510 skb_reset_network_header(skb); 511 512 /* Transport header needs to point to the UDP header */ 513 skb_set_transport_header(skb, offset); 514 515 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 516 &hdr->saddr, uh->dest, 517 inet6_iif(skb), 0, udptable, skb); 518 if (sk) { 519 int (*lookup)(struct sock *sk, struct sk_buff *skb); 520 struct udp_sock *up = udp_sk(sk); 521 522 lookup = READ_ONCE(up->encap_err_lookup); 523 if (!lookup || lookup(sk, skb)) 524 sk = NULL; 525 } 526 527 if (!sk) { 528 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 529 offset, info)); 530 } 531 532 skb_set_transport_header(skb, transport_offset); 533 skb_set_network_header(skb, network_offset); 534 535 return sk; 536 } 537 538 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 539 u8 type, u8 code, int offset, __be32 info, 540 struct udp_table *udptable) 541 { 542 struct ipv6_pinfo *np; 543 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 544 const struct in6_addr *saddr = &hdr->saddr; 545 const struct in6_addr *daddr = &hdr->daddr; 546 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 547 bool tunnel = false; 548 struct sock *sk; 549 int harderr; 550 int err; 551 struct net *net = dev_net(skb->dev); 552 553 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 554 inet6_iif(skb), inet6_sdif(skb), udptable, skb); 555 if (!sk) { 556 /* No socket for error: try tunnels before discarding */ 557 sk = ERR_PTR(-ENOENT); 558 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 559 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 560 udptable, skb, 561 opt, type, code, info); 562 if (!sk) 563 return 0; 564 } 565 566 if (IS_ERR(sk)) { 567 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 568 ICMP6_MIB_INERRORS); 569 return PTR_ERR(sk); 570 } 571 572 tunnel = true; 573 } 574 575 harderr = icmpv6_err_convert(type, code, &err); 576 np = inet6_sk(sk); 577 578 if (type == ICMPV6_PKT_TOOBIG) { 579 if (!ip6_sk_accept_pmtu(sk)) 580 goto out; 581 ip6_sk_update_pmtu(skb, sk, info); 582 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 583 harderr = 1; 584 } 585 if (type == NDISC_REDIRECT) { 586 if (tunnel) { 587 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 588 sk->sk_mark, sk->sk_uid); 589 } else { 590 ip6_sk_redirect(skb, sk); 591 } 592 goto out; 593 } 594 595 /* Tunnels don't have an application socket: don't pass errors back */ 596 if (tunnel) 597 goto out; 598 599 if (!np->recverr) { 600 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 601 goto out; 602 } else { 603 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 604 } 605 606 sk->sk_err = err; 607 sk->sk_error_report(sk); 608 out: 609 return 0; 610 } 611 612 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 613 { 614 int rc; 615 616 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 617 sock_rps_save_rxhash(sk, skb); 618 sk_mark_napi_id(sk, skb); 619 sk_incoming_cpu_update(sk); 620 } else { 621 sk_mark_napi_id_once(sk, skb); 622 } 623 624 rc = __udp_enqueue_schedule_skb(sk, skb); 625 if (rc < 0) { 626 int is_udplite = IS_UDPLITE(sk); 627 628 /* Note that an ENOMEM error is charged twice */ 629 if (rc == -ENOMEM) 630 UDP6_INC_STATS(sock_net(sk), 631 UDP_MIB_RCVBUFERRORS, is_udplite); 632 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 633 kfree_skb(skb); 634 return -1; 635 } 636 637 return 0; 638 } 639 640 static __inline__ int udpv6_err(struct sk_buff *skb, 641 struct inet6_skb_parm *opt, u8 type, 642 u8 code, int offset, __be32 info) 643 { 644 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 645 } 646 647 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 648 { 649 struct udp_sock *up = udp_sk(sk); 650 int is_udplite = IS_UDPLITE(sk); 651 652 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 653 goto drop; 654 655 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 656 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 657 658 /* 659 * This is an encapsulation socket so pass the skb to 660 * the socket's udp_encap_rcv() hook. Otherwise, just 661 * fall through and pass this up the UDP socket. 662 * up->encap_rcv() returns the following value: 663 * =0 if skb was successfully passed to the encap 664 * handler or was discarded by it. 665 * >0 if skb should be passed on to UDP. 666 * <0 if skb should be resubmitted as proto -N 667 */ 668 669 /* if we're overly short, let UDP handle it */ 670 encap_rcv = READ_ONCE(up->encap_rcv); 671 if (encap_rcv) { 672 int ret; 673 674 /* Verify checksum before giving to encap */ 675 if (udp_lib_checksum_complete(skb)) 676 goto csum_error; 677 678 ret = encap_rcv(sk, skb); 679 if (ret <= 0) { 680 __UDP_INC_STATS(sock_net(sk), 681 UDP_MIB_INDATAGRAMS, 682 is_udplite); 683 return -ret; 684 } 685 } 686 687 /* FALLTHROUGH -- it's a UDP Packet */ 688 } 689 690 /* 691 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 692 */ 693 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 694 695 if (up->pcrlen == 0) { /* full coverage was set */ 696 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 697 UDP_SKB_CB(skb)->cscov, skb->len); 698 goto drop; 699 } 700 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 701 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 702 UDP_SKB_CB(skb)->cscov, up->pcrlen); 703 goto drop; 704 } 705 } 706 707 prefetch(&sk->sk_rmem_alloc); 708 if (rcu_access_pointer(sk->sk_filter) && 709 udp_lib_checksum_complete(skb)) 710 goto csum_error; 711 712 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 713 goto drop; 714 715 udp_csum_pull_header(skb); 716 717 skb_dst_drop(skb); 718 719 return __udpv6_queue_rcv_skb(sk, skb); 720 721 csum_error: 722 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 723 drop: 724 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 725 atomic_inc(&sk->sk_drops); 726 kfree_skb(skb); 727 return -1; 728 } 729 730 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 731 { 732 struct sk_buff *next, *segs; 733 int ret; 734 735 if (likely(!udp_unexpected_gso(sk, skb))) 736 return udpv6_queue_rcv_one_skb(sk, skb); 737 738 __skb_push(skb, -skb_mac_offset(skb)); 739 segs = udp_rcv_segment(sk, skb, false); 740 for (skb = segs; skb; skb = next) { 741 next = skb->next; 742 __skb_pull(skb, skb_transport_offset(skb)); 743 744 ret = udpv6_queue_rcv_one_skb(sk, skb); 745 if (ret > 0) 746 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 747 true); 748 } 749 return 0; 750 } 751 752 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 753 __be16 loc_port, const struct in6_addr *loc_addr, 754 __be16 rmt_port, const struct in6_addr *rmt_addr, 755 int dif, int sdif, unsigned short hnum) 756 { 757 struct inet_sock *inet = inet_sk(sk); 758 759 if (!net_eq(sock_net(sk), net)) 760 return false; 761 762 if (udp_sk(sk)->udp_port_hash != hnum || 763 sk->sk_family != PF_INET6 || 764 (inet->inet_dport && inet->inet_dport != rmt_port) || 765 (!ipv6_addr_any(&sk->sk_v6_daddr) && 766 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 767 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) || 768 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 769 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 770 return false; 771 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 772 return false; 773 return true; 774 } 775 776 static void udp6_csum_zero_error(struct sk_buff *skb) 777 { 778 /* RFC 2460 section 8.1 says that we SHOULD log 779 * this error. Well, it is reasonable. 780 */ 781 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 782 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 783 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 784 } 785 786 /* 787 * Note: called only from the BH handler context, 788 * so we don't need to lock the hashes. 789 */ 790 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 791 const struct in6_addr *saddr, const struct in6_addr *daddr, 792 struct udp_table *udptable, int proto) 793 { 794 struct sock *sk, *first = NULL; 795 const struct udphdr *uh = udp_hdr(skb); 796 unsigned short hnum = ntohs(uh->dest); 797 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 798 unsigned int offset = offsetof(typeof(*sk), sk_node); 799 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 800 int dif = inet6_iif(skb); 801 int sdif = inet6_sdif(skb); 802 struct hlist_node *node; 803 struct sk_buff *nskb; 804 805 if (use_hash2) { 806 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 807 udptable->mask; 808 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 809 start_lookup: 810 hslot = &udptable->hash2[hash2]; 811 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 812 } 813 814 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 815 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 816 uh->source, saddr, dif, sdif, 817 hnum)) 818 continue; 819 /* If zero checksum and no_check is not on for 820 * the socket then skip it. 821 */ 822 if (!uh->check && !udp_sk(sk)->no_check6_rx) 823 continue; 824 if (!first) { 825 first = sk; 826 continue; 827 } 828 nskb = skb_clone(skb, GFP_ATOMIC); 829 if (unlikely(!nskb)) { 830 atomic_inc(&sk->sk_drops); 831 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 832 IS_UDPLITE(sk)); 833 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 834 IS_UDPLITE(sk)); 835 continue; 836 } 837 838 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 839 consume_skb(nskb); 840 } 841 842 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 843 if (use_hash2 && hash2 != hash2_any) { 844 hash2 = hash2_any; 845 goto start_lookup; 846 } 847 848 if (first) { 849 if (udpv6_queue_rcv_skb(first, skb) > 0) 850 consume_skb(skb); 851 } else { 852 kfree_skb(skb); 853 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 854 proto == IPPROTO_UDPLITE); 855 } 856 return 0; 857 } 858 859 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 860 { 861 if (udp_sk_rx_dst_set(sk, dst)) { 862 const struct rt6_info *rt = (const struct rt6_info *)dst; 863 864 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 865 } 866 } 867 868 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 869 * return code conversion for ip layer consumption 870 */ 871 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 872 struct udphdr *uh) 873 { 874 int ret; 875 876 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 877 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 878 ip6_compute_pseudo); 879 880 ret = udpv6_queue_rcv_skb(sk, skb); 881 882 /* a return value > 0 means to resubmit the input */ 883 if (ret > 0) 884 return ret; 885 return 0; 886 } 887 888 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 889 int proto) 890 { 891 const struct in6_addr *saddr, *daddr; 892 struct net *net = dev_net(skb->dev); 893 struct udphdr *uh; 894 struct sock *sk; 895 u32 ulen = 0; 896 897 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 898 goto discard; 899 900 saddr = &ipv6_hdr(skb)->saddr; 901 daddr = &ipv6_hdr(skb)->daddr; 902 uh = udp_hdr(skb); 903 904 ulen = ntohs(uh->len); 905 if (ulen > skb->len) 906 goto short_packet; 907 908 if (proto == IPPROTO_UDP) { 909 /* UDP validates ulen. */ 910 911 /* Check for jumbo payload */ 912 if (ulen == 0) 913 ulen = skb->len; 914 915 if (ulen < sizeof(*uh)) 916 goto short_packet; 917 918 if (ulen < skb->len) { 919 if (pskb_trim_rcsum(skb, ulen)) 920 goto short_packet; 921 saddr = &ipv6_hdr(skb)->saddr; 922 daddr = &ipv6_hdr(skb)->daddr; 923 uh = udp_hdr(skb); 924 } 925 } 926 927 if (udp6_csum_init(skb, uh, proto)) 928 goto csum_error; 929 930 /* Check if the socket is already available, e.g. due to early demux */ 931 sk = skb_steal_sock(skb); 932 if (sk) { 933 struct dst_entry *dst = skb_dst(skb); 934 int ret; 935 936 if (unlikely(sk->sk_rx_dst != dst)) 937 udp6_sk_rx_dst_set(sk, dst); 938 939 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 940 sock_put(sk); 941 goto report_csum_error; 942 } 943 944 ret = udp6_unicast_rcv_skb(sk, skb, uh); 945 sock_put(sk); 946 return ret; 947 } 948 949 /* 950 * Multicast receive code 951 */ 952 if (ipv6_addr_is_multicast(daddr)) 953 return __udp6_lib_mcast_deliver(net, skb, 954 saddr, daddr, udptable, proto); 955 956 /* Unicast */ 957 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 958 if (sk) { 959 if (!uh->check && !udp_sk(sk)->no_check6_rx) 960 goto report_csum_error; 961 return udp6_unicast_rcv_skb(sk, skb, uh); 962 } 963 964 if (!uh->check) 965 goto report_csum_error; 966 967 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 968 goto discard; 969 970 if (udp_lib_checksum_complete(skb)) 971 goto csum_error; 972 973 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 974 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 975 976 kfree_skb(skb); 977 return 0; 978 979 short_packet: 980 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 981 proto == IPPROTO_UDPLITE ? "-Lite" : "", 982 saddr, ntohs(uh->source), 983 ulen, skb->len, 984 daddr, ntohs(uh->dest)); 985 goto discard; 986 987 report_csum_error: 988 udp6_csum_zero_error(skb); 989 csum_error: 990 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 991 discard: 992 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 993 kfree_skb(skb); 994 return 0; 995 } 996 997 998 static struct sock *__udp6_lib_demux_lookup(struct net *net, 999 __be16 loc_port, const struct in6_addr *loc_addr, 1000 __be16 rmt_port, const struct in6_addr *rmt_addr, 1001 int dif, int sdif) 1002 { 1003 unsigned short hnum = ntohs(loc_port); 1004 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1005 unsigned int slot2 = hash2 & udp_table.mask; 1006 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1007 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1008 struct sock *sk; 1009 1010 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1011 if (sk->sk_state == TCP_ESTABLISHED && 1012 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 1013 return sk; 1014 /* Only check first socket in chain */ 1015 break; 1016 } 1017 return NULL; 1018 } 1019 1020 static void udp_v6_early_demux(struct sk_buff *skb) 1021 { 1022 struct net *net = dev_net(skb->dev); 1023 const struct udphdr *uh; 1024 struct sock *sk; 1025 struct dst_entry *dst; 1026 int dif = skb->dev->ifindex; 1027 int sdif = inet6_sdif(skb); 1028 1029 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1030 sizeof(struct udphdr))) 1031 return; 1032 1033 uh = udp_hdr(skb); 1034 1035 if (skb->pkt_type == PACKET_HOST) 1036 sk = __udp6_lib_demux_lookup(net, uh->dest, 1037 &ipv6_hdr(skb)->daddr, 1038 uh->source, &ipv6_hdr(skb)->saddr, 1039 dif, sdif); 1040 else 1041 return; 1042 1043 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1044 return; 1045 1046 skb->sk = sk; 1047 skb->destructor = sock_efree; 1048 dst = READ_ONCE(sk->sk_rx_dst); 1049 1050 if (dst) 1051 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 1052 if (dst) { 1053 /* set noref for now. 1054 * any place which wants to hold dst has to call 1055 * dst_hold_safe() 1056 */ 1057 skb_dst_set_noref(skb, dst); 1058 } 1059 } 1060 1061 static __inline__ int udpv6_rcv(struct sk_buff *skb) 1062 { 1063 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1064 } 1065 1066 /* 1067 * Throw away all pending data and cancel the corking. Socket is locked. 1068 */ 1069 static void udp_v6_flush_pending_frames(struct sock *sk) 1070 { 1071 struct udp_sock *up = udp_sk(sk); 1072 1073 if (up->pending == AF_INET) 1074 udp_flush_pending_frames(sk); 1075 else if (up->pending) { 1076 up->len = 0; 1077 up->pending = 0; 1078 ip6_flush_pending_frames(sk); 1079 } 1080 } 1081 1082 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1083 int addr_len) 1084 { 1085 /* The following checks are replicated from __ip6_datagram_connect() 1086 * and intended to prevent BPF program called below from accessing 1087 * bytes that are out of the bound specified by user in addr_len. 1088 */ 1089 if (uaddr->sa_family == AF_INET) { 1090 if (__ipv6_only_sock(sk)) 1091 return -EAFNOSUPPORT; 1092 return udp_pre_connect(sk, uaddr, addr_len); 1093 } 1094 1095 if (addr_len < SIN6_LEN_RFC2133) 1096 return -EINVAL; 1097 1098 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1099 } 1100 1101 /** 1102 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1103 * @sk: socket we are sending on 1104 * @skb: sk_buff containing the filled-in UDP header 1105 * (checksum field must be zeroed out) 1106 */ 1107 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1108 const struct in6_addr *saddr, 1109 const struct in6_addr *daddr, int len) 1110 { 1111 unsigned int offset; 1112 struct udphdr *uh = udp_hdr(skb); 1113 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1114 __wsum csum = 0; 1115 1116 if (!frags) { 1117 /* Only one fragment on the socket. */ 1118 skb->csum_start = skb_transport_header(skb) - skb->head; 1119 skb->csum_offset = offsetof(struct udphdr, check); 1120 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1121 } else { 1122 /* 1123 * HW-checksum won't work as there are two or more 1124 * fragments on the socket so that all csums of sk_buffs 1125 * should be together 1126 */ 1127 offset = skb_transport_offset(skb); 1128 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1129 csum = skb->csum; 1130 1131 skb->ip_summed = CHECKSUM_NONE; 1132 1133 do { 1134 csum = csum_add(csum, frags->csum); 1135 } while ((frags = frags->next)); 1136 1137 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1138 csum); 1139 if (uh->check == 0) 1140 uh->check = CSUM_MANGLED_0; 1141 } 1142 } 1143 1144 /* 1145 * Sending 1146 */ 1147 1148 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1149 struct inet_cork *cork) 1150 { 1151 struct sock *sk = skb->sk; 1152 struct udphdr *uh; 1153 int err = 0; 1154 int is_udplite = IS_UDPLITE(sk); 1155 __wsum csum = 0; 1156 int offset = skb_transport_offset(skb); 1157 int len = skb->len - offset; 1158 1159 /* 1160 * Create a UDP header 1161 */ 1162 uh = udp_hdr(skb); 1163 uh->source = fl6->fl6_sport; 1164 uh->dest = fl6->fl6_dport; 1165 uh->len = htons(len); 1166 uh->check = 0; 1167 1168 if (cork->gso_size) { 1169 const int hlen = skb_network_header_len(skb) + 1170 sizeof(struct udphdr); 1171 1172 if (hlen + cork->gso_size > cork->fragsize) 1173 return -EINVAL; 1174 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) 1175 return -EINVAL; 1176 if (udp_sk(sk)->no_check6_tx) 1177 return -EINVAL; 1178 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1179 dst_xfrm(skb_dst(skb))) 1180 return -EIO; 1181 1182 skb_shinfo(skb)->gso_size = cork->gso_size; 1183 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1184 goto csum_partial; 1185 } 1186 1187 if (is_udplite) 1188 csum = udplite_csum(skb); 1189 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1190 skb->ip_summed = CHECKSUM_NONE; 1191 goto send; 1192 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1193 csum_partial: 1194 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1195 goto send; 1196 } else 1197 csum = udp_csum(skb); 1198 1199 /* add protocol-dependent pseudo-header */ 1200 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1201 len, fl6->flowi6_proto, csum); 1202 if (uh->check == 0) 1203 uh->check = CSUM_MANGLED_0; 1204 1205 send: 1206 err = ip6_send_skb(skb); 1207 if (err) { 1208 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1209 UDP6_INC_STATS(sock_net(sk), 1210 UDP_MIB_SNDBUFERRORS, is_udplite); 1211 err = 0; 1212 } 1213 } else { 1214 UDP6_INC_STATS(sock_net(sk), 1215 UDP_MIB_OUTDATAGRAMS, is_udplite); 1216 } 1217 return err; 1218 } 1219 1220 static int udp_v6_push_pending_frames(struct sock *sk) 1221 { 1222 struct sk_buff *skb; 1223 struct udp_sock *up = udp_sk(sk); 1224 struct flowi6 fl6; 1225 int err = 0; 1226 1227 if (up->pending == AF_INET) 1228 return udp_push_pending_frames(sk); 1229 1230 /* ip6_finish_skb will release the cork, so make a copy of 1231 * fl6 here. 1232 */ 1233 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1234 1235 skb = ip6_finish_skb(sk); 1236 if (!skb) 1237 goto out; 1238 1239 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1240 1241 out: 1242 up->len = 0; 1243 up->pending = 0; 1244 return err; 1245 } 1246 1247 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1248 { 1249 struct ipv6_txoptions opt_space; 1250 struct udp_sock *up = udp_sk(sk); 1251 struct inet_sock *inet = inet_sk(sk); 1252 struct ipv6_pinfo *np = inet6_sk(sk); 1253 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1254 struct in6_addr *daddr, *final_p, final; 1255 struct ipv6_txoptions *opt = NULL; 1256 struct ipv6_txoptions *opt_to_free = NULL; 1257 struct ip6_flowlabel *flowlabel = NULL; 1258 struct flowi6 fl6; 1259 struct dst_entry *dst; 1260 struct ipcm6_cookie ipc6; 1261 int addr_len = msg->msg_namelen; 1262 bool connected = false; 1263 int ulen = len; 1264 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1265 int err; 1266 int is_udplite = IS_UDPLITE(sk); 1267 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1268 1269 ipcm6_init(&ipc6); 1270 ipc6.gso_size = up->gso_size; 1271 ipc6.sockc.tsflags = sk->sk_tsflags; 1272 1273 /* destination address check */ 1274 if (sin6) { 1275 if (addr_len < offsetof(struct sockaddr, sa_data)) 1276 return -EINVAL; 1277 1278 switch (sin6->sin6_family) { 1279 case AF_INET6: 1280 if (addr_len < SIN6_LEN_RFC2133) 1281 return -EINVAL; 1282 daddr = &sin6->sin6_addr; 1283 if (ipv6_addr_any(daddr) && 1284 ipv6_addr_v4mapped(&np->saddr)) 1285 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1286 daddr); 1287 break; 1288 case AF_INET: 1289 goto do_udp_sendmsg; 1290 case AF_UNSPEC: 1291 msg->msg_name = sin6 = NULL; 1292 msg->msg_namelen = addr_len = 0; 1293 daddr = NULL; 1294 break; 1295 default: 1296 return -EINVAL; 1297 } 1298 } else if (!up->pending) { 1299 if (sk->sk_state != TCP_ESTABLISHED) 1300 return -EDESTADDRREQ; 1301 daddr = &sk->sk_v6_daddr; 1302 } else 1303 daddr = NULL; 1304 1305 if (daddr) { 1306 if (ipv6_addr_v4mapped(daddr)) { 1307 struct sockaddr_in sin; 1308 sin.sin_family = AF_INET; 1309 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1310 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1311 msg->msg_name = &sin; 1312 msg->msg_namelen = sizeof(sin); 1313 do_udp_sendmsg: 1314 if (__ipv6_only_sock(sk)) 1315 return -ENETUNREACH; 1316 return udp_sendmsg(sk, msg, len); 1317 } 1318 } 1319 1320 if (up->pending == AF_INET) 1321 return udp_sendmsg(sk, msg, len); 1322 1323 /* Rough check on arithmetic overflow, 1324 better check is made in ip6_append_data(). 1325 */ 1326 if (len > INT_MAX - sizeof(struct udphdr)) 1327 return -EMSGSIZE; 1328 1329 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1330 if (up->pending) { 1331 /* 1332 * There are pending frames. 1333 * The socket lock must be held while it's corked. 1334 */ 1335 lock_sock(sk); 1336 if (likely(up->pending)) { 1337 if (unlikely(up->pending != AF_INET6)) { 1338 release_sock(sk); 1339 return -EAFNOSUPPORT; 1340 } 1341 dst = NULL; 1342 goto do_append_data; 1343 } 1344 release_sock(sk); 1345 } 1346 ulen += sizeof(struct udphdr); 1347 1348 memset(&fl6, 0, sizeof(fl6)); 1349 1350 if (sin6) { 1351 if (sin6->sin6_port == 0) 1352 return -EINVAL; 1353 1354 fl6.fl6_dport = sin6->sin6_port; 1355 daddr = &sin6->sin6_addr; 1356 1357 if (np->sndflow) { 1358 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1359 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1360 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1361 if (!flowlabel) 1362 return -EINVAL; 1363 } 1364 } 1365 1366 /* 1367 * Otherwise it will be difficult to maintain 1368 * sk->sk_dst_cache. 1369 */ 1370 if (sk->sk_state == TCP_ESTABLISHED && 1371 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1372 daddr = &sk->sk_v6_daddr; 1373 1374 if (addr_len >= sizeof(struct sockaddr_in6) && 1375 sin6->sin6_scope_id && 1376 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1377 fl6.flowi6_oif = sin6->sin6_scope_id; 1378 } else { 1379 if (sk->sk_state != TCP_ESTABLISHED) 1380 return -EDESTADDRREQ; 1381 1382 fl6.fl6_dport = inet->inet_dport; 1383 daddr = &sk->sk_v6_daddr; 1384 fl6.flowlabel = np->flow_label; 1385 connected = true; 1386 } 1387 1388 if (!fl6.flowi6_oif) 1389 fl6.flowi6_oif = sk->sk_bound_dev_if; 1390 1391 if (!fl6.flowi6_oif) 1392 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1393 1394 fl6.flowi6_mark = sk->sk_mark; 1395 fl6.flowi6_uid = sk->sk_uid; 1396 1397 if (msg->msg_controllen) { 1398 opt = &opt_space; 1399 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1400 opt->tot_len = sizeof(*opt); 1401 ipc6.opt = opt; 1402 1403 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1404 if (err > 0) 1405 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1406 &ipc6); 1407 if (err < 0) { 1408 fl6_sock_release(flowlabel); 1409 return err; 1410 } 1411 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1412 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1413 if (!flowlabel) 1414 return -EINVAL; 1415 } 1416 if (!(opt->opt_nflen|opt->opt_flen)) 1417 opt = NULL; 1418 connected = false; 1419 } 1420 if (!opt) { 1421 opt = txopt_get(np); 1422 opt_to_free = opt; 1423 } 1424 if (flowlabel) 1425 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1426 opt = ipv6_fixup_options(&opt_space, opt); 1427 ipc6.opt = opt; 1428 1429 fl6.flowi6_proto = sk->sk_protocol; 1430 if (!ipv6_addr_any(daddr)) 1431 fl6.daddr = *daddr; 1432 else 1433 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1434 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1435 fl6.saddr = np->saddr; 1436 fl6.fl6_sport = inet->inet_sport; 1437 1438 if (cgroup_bpf_enabled && !connected) { 1439 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1440 (struct sockaddr *)sin6, &fl6.saddr); 1441 if (err) 1442 goto out_no_dst; 1443 if (sin6) { 1444 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1445 /* BPF program rewrote IPv6-only by IPv4-mapped 1446 * IPv6. It's currently unsupported. 1447 */ 1448 err = -ENOTSUPP; 1449 goto out_no_dst; 1450 } 1451 if (sin6->sin6_port == 0) { 1452 /* BPF program set invalid port. Reject it. */ 1453 err = -EINVAL; 1454 goto out_no_dst; 1455 } 1456 fl6.fl6_dport = sin6->sin6_port; 1457 fl6.daddr = sin6->sin6_addr; 1458 } 1459 } 1460 1461 final_p = fl6_update_dst(&fl6, opt, &final); 1462 if (final_p) 1463 connected = false; 1464 1465 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1466 fl6.flowi6_oif = np->mcast_oif; 1467 connected = false; 1468 } else if (!fl6.flowi6_oif) 1469 fl6.flowi6_oif = np->ucast_oif; 1470 1471 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1472 1473 if (ipc6.tclass < 0) 1474 ipc6.tclass = np->tclass; 1475 1476 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1477 1478 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1479 if (IS_ERR(dst)) { 1480 err = PTR_ERR(dst); 1481 dst = NULL; 1482 goto out; 1483 } 1484 1485 if (ipc6.hlimit < 0) 1486 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1487 1488 if (msg->msg_flags&MSG_CONFIRM) 1489 goto do_confirm; 1490 back_from_confirm: 1491 1492 /* Lockless fast path for the non-corking case */ 1493 if (!corkreq) { 1494 struct inet_cork_full cork; 1495 struct sk_buff *skb; 1496 1497 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1498 sizeof(struct udphdr), &ipc6, 1499 &fl6, (struct rt6_info *)dst, 1500 msg->msg_flags, &cork); 1501 err = PTR_ERR(skb); 1502 if (!IS_ERR_OR_NULL(skb)) 1503 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1504 goto out; 1505 } 1506 1507 lock_sock(sk); 1508 if (unlikely(up->pending)) { 1509 /* The socket is already corked while preparing it. */ 1510 /* ... which is an evident application bug. --ANK */ 1511 release_sock(sk); 1512 1513 net_dbg_ratelimited("udp cork app bug 2\n"); 1514 err = -EINVAL; 1515 goto out; 1516 } 1517 1518 up->pending = AF_INET6; 1519 1520 do_append_data: 1521 if (ipc6.dontfrag < 0) 1522 ipc6.dontfrag = np->dontfrag; 1523 up->len += ulen; 1524 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1525 &ipc6, &fl6, (struct rt6_info *)dst, 1526 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1527 if (err) 1528 udp_v6_flush_pending_frames(sk); 1529 else if (!corkreq) 1530 err = udp_v6_push_pending_frames(sk); 1531 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1532 up->pending = 0; 1533 1534 if (err > 0) 1535 err = np->recverr ? net_xmit_errno(err) : 0; 1536 release_sock(sk); 1537 1538 out: 1539 dst_release(dst); 1540 out_no_dst: 1541 fl6_sock_release(flowlabel); 1542 txopt_put(opt_to_free); 1543 if (!err) 1544 return len; 1545 /* 1546 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1547 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1548 * we don't have a good statistic (IpOutDiscards but it can be too many 1549 * things). We could add another new stat but at least for now that 1550 * seems like overkill. 1551 */ 1552 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1553 UDP6_INC_STATS(sock_net(sk), 1554 UDP_MIB_SNDBUFERRORS, is_udplite); 1555 } 1556 return err; 1557 1558 do_confirm: 1559 if (msg->msg_flags & MSG_PROBE) 1560 dst_confirm_neigh(dst, &fl6.daddr); 1561 if (!(msg->msg_flags&MSG_PROBE) || len) 1562 goto back_from_confirm; 1563 err = 0; 1564 goto out; 1565 } 1566 1567 void udpv6_destroy_sock(struct sock *sk) 1568 { 1569 struct udp_sock *up = udp_sk(sk); 1570 lock_sock(sk); 1571 udp_v6_flush_pending_frames(sk); 1572 release_sock(sk); 1573 1574 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1575 if (up->encap_type) { 1576 void (*encap_destroy)(struct sock *sk); 1577 encap_destroy = READ_ONCE(up->encap_destroy); 1578 if (encap_destroy) 1579 encap_destroy(sk); 1580 } 1581 if (up->encap_enabled) 1582 static_branch_disable(&udpv6_encap_needed_key); 1583 } 1584 1585 inet6_destroy_sock(sk); 1586 } 1587 1588 /* 1589 * Socket option code for UDP 1590 */ 1591 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1592 char __user *optval, unsigned int optlen) 1593 { 1594 if (level == SOL_UDP || level == SOL_UDPLITE) 1595 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1596 udp_v6_push_pending_frames); 1597 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1598 } 1599 1600 #ifdef CONFIG_COMPAT 1601 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1602 char __user *optval, unsigned int optlen) 1603 { 1604 if (level == SOL_UDP || level == SOL_UDPLITE) 1605 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1606 udp_v6_push_pending_frames); 1607 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1608 } 1609 #endif 1610 1611 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1612 char __user *optval, int __user *optlen) 1613 { 1614 if (level == SOL_UDP || level == SOL_UDPLITE) 1615 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1616 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1617 } 1618 1619 #ifdef CONFIG_COMPAT 1620 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1621 char __user *optval, int __user *optlen) 1622 { 1623 if (level == SOL_UDP || level == SOL_UDPLITE) 1624 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1625 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1626 } 1627 #endif 1628 1629 /* thinking of making this const? Don't. 1630 * early_demux can change based on sysctl. 1631 */ 1632 static struct inet6_protocol udpv6_protocol = { 1633 .early_demux = udp_v6_early_demux, 1634 .early_demux_handler = udp_v6_early_demux, 1635 .handler = udpv6_rcv, 1636 .err_handler = udpv6_err, 1637 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1638 }; 1639 1640 /* ------------------------------------------------------------------------ */ 1641 #ifdef CONFIG_PROC_FS 1642 int udp6_seq_show(struct seq_file *seq, void *v) 1643 { 1644 if (v == SEQ_START_TOKEN) { 1645 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1646 } else { 1647 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1648 struct inet_sock *inet = inet_sk(v); 1649 __u16 srcp = ntohs(inet->inet_sport); 1650 __u16 destp = ntohs(inet->inet_dport); 1651 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1652 udp_rqueue_get(v), bucket); 1653 } 1654 return 0; 1655 } 1656 1657 const struct seq_operations udp6_seq_ops = { 1658 .start = udp_seq_start, 1659 .next = udp_seq_next, 1660 .stop = udp_seq_stop, 1661 .show = udp6_seq_show, 1662 }; 1663 EXPORT_SYMBOL(udp6_seq_ops); 1664 1665 static struct udp_seq_afinfo udp6_seq_afinfo = { 1666 .family = AF_INET6, 1667 .udp_table = &udp_table, 1668 }; 1669 1670 int __net_init udp6_proc_init(struct net *net) 1671 { 1672 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1673 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1674 return -ENOMEM; 1675 return 0; 1676 } 1677 1678 void udp6_proc_exit(struct net *net) 1679 { 1680 remove_proc_entry("udp6", net->proc_net); 1681 } 1682 #endif /* CONFIG_PROC_FS */ 1683 1684 /* ------------------------------------------------------------------------ */ 1685 1686 struct proto udpv6_prot = { 1687 .name = "UDPv6", 1688 .owner = THIS_MODULE, 1689 .close = udp_lib_close, 1690 .pre_connect = udpv6_pre_connect, 1691 .connect = ip6_datagram_connect, 1692 .disconnect = udp_disconnect, 1693 .ioctl = udp_ioctl, 1694 .init = udp_init_sock, 1695 .destroy = udpv6_destroy_sock, 1696 .setsockopt = udpv6_setsockopt, 1697 .getsockopt = udpv6_getsockopt, 1698 .sendmsg = udpv6_sendmsg, 1699 .recvmsg = udpv6_recvmsg, 1700 .release_cb = ip6_datagram_release_cb, 1701 .hash = udp_lib_hash, 1702 .unhash = udp_lib_unhash, 1703 .rehash = udp_v6_rehash, 1704 .get_port = udp_v6_get_port, 1705 .memory_allocated = &udp_memory_allocated, 1706 .sysctl_mem = sysctl_udp_mem, 1707 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1708 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1709 .obj_size = sizeof(struct udp6_sock), 1710 .h.udp_table = &udp_table, 1711 #ifdef CONFIG_COMPAT 1712 .compat_setsockopt = compat_udpv6_setsockopt, 1713 .compat_getsockopt = compat_udpv6_getsockopt, 1714 #endif 1715 .diag_destroy = udp_abort, 1716 }; 1717 1718 static struct inet_protosw udpv6_protosw = { 1719 .type = SOCK_DGRAM, 1720 .protocol = IPPROTO_UDP, 1721 .prot = &udpv6_prot, 1722 .ops = &inet6_dgram_ops, 1723 .flags = INET_PROTOSW_PERMANENT, 1724 }; 1725 1726 int __init udpv6_init(void) 1727 { 1728 int ret; 1729 1730 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1731 if (ret) 1732 goto out; 1733 1734 ret = inet6_register_protosw(&udpv6_protosw); 1735 if (ret) 1736 goto out_udpv6_protocol; 1737 out: 1738 return ret; 1739 1740 out_udpv6_protocol: 1741 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1742 goto out; 1743 } 1744 1745 void udpv6_exit(void) 1746 { 1747 inet6_unregister_protosw(&udpv6_protosw); 1748 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1749 } 1750