1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * UDP over IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on linux/ipv4/udp.c 10 * 11 * Fixes: 12 * Hideaki YOSHIFUJI : sin6_scope_id support 13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 15 * a single port at the same time. 16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. 18 */ 19 20 #include <linux/bpf-cgroup.h> 21 #include <linux/errno.h> 22 #include <linux/types.h> 23 #include <linux/socket.h> 24 #include <linux/sockios.h> 25 #include <linux/net.h> 26 #include <linux/in6.h> 27 #include <linux/netdevice.h> 28 #include <linux/if_arp.h> 29 #include <linux/ipv6.h> 30 #include <linux/icmpv6.h> 31 #include <linux/init.h> 32 #include <linux/module.h> 33 #include <linux/skbuff.h> 34 #include <linux/slab.h> 35 #include <linux/uaccess.h> 36 #include <linux/indirect_call_wrapper.h> 37 #include <trace/events/udp.h> 38 39 #include <net/addrconf.h> 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/seg6.h> 46 #include <net/tcp_states.h> 47 #include <net/ip6_checksum.h> 48 #include <net/ip6_tunnel.h> 49 #include <net/udp_tunnel.h> 50 #include <net/xfrm.h> 51 #include <net/inet_hashtables.h> 52 #include <net/inet6_hashtables.h> 53 #include <net/busy_poll.h> 54 #include <net/sock_reuseport.h> 55 #include <net/gro.h> 56 57 #include <linux/proc_fs.h> 58 #include <linux/seq_file.h> 59 #include <trace/events/skb.h> 60 #include "udp_impl.h" 61 62 static void udpv6_destruct_sock(struct sock *sk) 63 { 64 udp_destruct_common(sk); 65 inet6_sock_destruct(sk); 66 } 67 68 int udpv6_init_sock(struct sock *sk) 69 { 70 udp_lib_init_sock(sk); 71 sk->sk_destruct = udpv6_destruct_sock; 72 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 73 return 0; 74 } 75 76 INDIRECT_CALLABLE_SCOPE 77 u32 udp6_ehashfn(const struct net *net, 78 const struct in6_addr *laddr, 79 const u16 lport, 80 const struct in6_addr *faddr, 81 const __be16 fport) 82 { 83 u32 lhash, fhash; 84 85 net_get_random_once(&udp6_ehash_secret, 86 sizeof(udp6_ehash_secret)); 87 net_get_random_once(&udp_ipv6_hash_secret, 88 sizeof(udp_ipv6_hash_secret)); 89 90 lhash = (__force u32)laddr->s6_addr32[3]; 91 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 92 93 return __inet6_ehashfn(lhash, lport, fhash, fport, 94 udp6_ehash_secret + net_hash_mix(net)); 95 } 96 97 int udp_v6_get_port(struct sock *sk, unsigned short snum) 98 { 99 unsigned int hash2_nulladdr = 100 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 101 unsigned int hash2_partial = 102 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 103 104 /* precompute partial secondary hash */ 105 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 106 return udp_lib_get_port(sk, snum, hash2_nulladdr); 107 } 108 109 void udp_v6_rehash(struct sock *sk) 110 { 111 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 112 &sk->sk_v6_rcv_saddr, 113 inet_sk(sk)->inet_num); 114 u16 new_hash4; 115 116 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) { 117 new_hash4 = udp_ehashfn(sock_net(sk), 118 sk->sk_rcv_saddr, sk->sk_num, 119 sk->sk_daddr, sk->sk_dport); 120 } else { 121 new_hash4 = udp6_ehashfn(sock_net(sk), 122 &sk->sk_v6_rcv_saddr, sk->sk_num, 123 &sk->sk_v6_daddr, sk->sk_dport); 124 } 125 126 udp_lib_rehash(sk, new_hash, new_hash4); 127 } 128 129 static int compute_score(struct sock *sk, const struct net *net, 130 const struct in6_addr *saddr, __be16 sport, 131 const struct in6_addr *daddr, unsigned short hnum, 132 int dif, int sdif) 133 { 134 int bound_dev_if, score; 135 struct inet_sock *inet; 136 bool dev_match; 137 138 if (!net_eq(sock_net(sk), net) || 139 udp_sk(sk)->udp_port_hash != hnum || 140 sk->sk_family != PF_INET6) 141 return -1; 142 143 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 144 return -1; 145 146 score = 0; 147 inet = inet_sk(sk); 148 149 if (inet->inet_dport) { 150 if (inet->inet_dport != sport) 151 return -1; 152 score++; 153 } 154 155 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 156 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 157 return -1; 158 score++; 159 } 160 161 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 162 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif); 163 if (!dev_match) 164 return -1; 165 if (bound_dev_if) 166 score++; 167 168 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 169 score++; 170 171 return score; 172 } 173 174 /** 175 * udp6_lib_lookup1() - Simplified lookup using primary hash (destination port) 176 * @net: Network namespace 177 * @saddr: Source address, network order 178 * @sport: Source port, network order 179 * @daddr: Destination address, network order 180 * @hnum: Destination port, host order 181 * @dif: Destination interface index 182 * @sdif: Destination bridge port index, if relevant 183 * @udptable: Set of UDP hash tables 184 * 185 * Simplified lookup to be used as fallback if no sockets are found due to a 186 * potential race between (receive) address change, and lookup happening before 187 * the rehash operation. This function ignores SO_REUSEPORT groups while scoring 188 * result sockets, because if we have one, we don't need the fallback at all. 189 * 190 * Called under rcu_read_lock(). 191 * 192 * Return: socket with highest matching score if any, NULL if none 193 */ 194 static struct sock *udp6_lib_lookup1(const struct net *net, 195 const struct in6_addr *saddr, __be16 sport, 196 const struct in6_addr *daddr, 197 unsigned int hnum, int dif, int sdif, 198 const struct udp_table *udptable) 199 { 200 unsigned int slot = udp_hashfn(net, hnum, udptable->mask); 201 struct udp_hslot *hslot = &udptable->hash[slot]; 202 struct sock *sk, *result = NULL; 203 int score, badness = 0; 204 205 sk_for_each_rcu(sk, &hslot->head) { 206 score = compute_score(sk, net, 207 saddr, sport, daddr, hnum, dif, sdif); 208 if (score > badness) { 209 result = sk; 210 badness = score; 211 } 212 } 213 214 return result; 215 } 216 217 /* called with rcu_read_lock() */ 218 static struct sock *udp6_lib_lookup2(const struct net *net, 219 const struct in6_addr *saddr, __be16 sport, 220 const struct in6_addr *daddr, unsigned int hnum, 221 int dif, int sdif, struct udp_hslot *hslot2, 222 struct sk_buff *skb) 223 { 224 struct sock *sk, *result; 225 int score, badness; 226 bool need_rescore; 227 228 result = NULL; 229 badness = -1; 230 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 231 need_rescore = false; 232 rescore: 233 score = compute_score(need_rescore ? result : sk, net, saddr, 234 sport, daddr, hnum, dif, sdif); 235 if (score > badness) { 236 badness = score; 237 238 if (need_rescore) 239 continue; 240 241 if (sk->sk_state == TCP_ESTABLISHED) { 242 result = sk; 243 continue; 244 } 245 246 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 247 saddr, sport, daddr, hnum, udp6_ehashfn); 248 if (!result) { 249 result = sk; 250 continue; 251 } 252 253 /* Fall back to scoring if group has connections */ 254 if (!reuseport_has_conns(sk)) 255 return result; 256 257 /* Reuseport logic returned an error, keep original score. */ 258 if (IS_ERR(result)) 259 continue; 260 261 /* compute_score is too long of a function to be 262 * inlined, and calling it again here yields 263 * measureable overhead for some 264 * workloads. Work around it by jumping 265 * backwards to rescore 'result'. 266 */ 267 need_rescore = true; 268 goto rescore; 269 } 270 } 271 return result; 272 } 273 274 #if IS_ENABLED(CONFIG_BASE_SMALL) 275 static struct sock *udp6_lib_lookup4(const struct net *net, 276 const struct in6_addr *saddr, __be16 sport, 277 const struct in6_addr *daddr, 278 unsigned int hnum, int dif, int sdif, 279 struct udp_table *udptable) 280 { 281 return NULL; 282 } 283 284 static void udp6_hash4(struct sock *sk) 285 { 286 } 287 #else /* !CONFIG_BASE_SMALL */ 288 static struct sock *udp6_lib_lookup4(const struct net *net, 289 const struct in6_addr *saddr, __be16 sport, 290 const struct in6_addr *daddr, 291 unsigned int hnum, int dif, int sdif, 292 struct udp_table *udptable) 293 { 294 const __portpair ports = INET_COMBINED_PORTS(sport, hnum); 295 const struct hlist_nulls_node *node; 296 struct udp_hslot *hslot4; 297 unsigned int hash4, slot; 298 struct udp_sock *up; 299 struct sock *sk; 300 301 hash4 = udp6_ehashfn(net, daddr, hnum, saddr, sport); 302 slot = hash4 & udptable->mask; 303 hslot4 = &udptable->hash4[slot]; 304 305 begin: 306 udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) { 307 sk = (struct sock *)up; 308 if (inet6_match(net, sk, saddr, daddr, ports, dif, sdif)) 309 return sk; 310 } 311 312 /* if the nulls value we got at the end of this lookup is not the 313 * expected one, we must restart lookup. We probably met an item that 314 * was moved to another chain due to rehash. 315 */ 316 if (get_nulls_value(node) != slot) 317 goto begin; 318 319 return NULL; 320 } 321 322 static void udp6_hash4(struct sock *sk) 323 { 324 struct net *net = sock_net(sk); 325 unsigned int hash; 326 327 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) { 328 udp4_hash4(sk); 329 return; 330 } 331 332 if (sk_unhashed(sk) || ipv6_addr_any(&sk->sk_v6_rcv_saddr)) 333 return; 334 335 hash = udp6_ehashfn(net, &sk->sk_v6_rcv_saddr, sk->sk_num, 336 &sk->sk_v6_daddr, sk->sk_dport); 337 338 udp_lib_hash4(sk, hash); 339 } 340 #endif /* CONFIG_BASE_SMALL */ 341 342 /* rcu_read_lock() must be held */ 343 struct sock *__udp6_lib_lookup(const struct net *net, 344 const struct in6_addr *saddr, __be16 sport, 345 const struct in6_addr *daddr, __be16 dport, 346 int dif, int sdif, struct udp_table *udptable, 347 struct sk_buff *skb) 348 { 349 unsigned short hnum = ntohs(dport); 350 struct udp_hslot *hslot2; 351 struct sock *result, *sk; 352 unsigned int hash2; 353 354 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 355 hslot2 = udp_hashslot2(udptable, hash2); 356 357 if (udp_has_hash4(hslot2)) { 358 result = udp6_lib_lookup4(net, saddr, sport, daddr, hnum, 359 dif, sdif, udptable); 360 if (result) /* udp6_lib_lookup4 return sk or NULL */ 361 return result; 362 } 363 364 /* Lookup connected or non-wildcard sockets */ 365 result = udp6_lib_lookup2(net, saddr, sport, 366 daddr, hnum, dif, sdif, 367 hslot2, skb); 368 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 369 goto done; 370 371 /* Lookup redirect from BPF */ 372 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 373 udptable == net->ipv4.udp_table) { 374 sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 375 saddr, sport, daddr, hnum, dif, 376 udp6_ehashfn); 377 if (sk) { 378 result = sk; 379 goto done; 380 } 381 } 382 383 /* Got non-wildcard socket or error on first lookup */ 384 if (result) 385 goto done; 386 387 /* Lookup wildcard sockets */ 388 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 389 hslot2 = udp_hashslot2(udptable, hash2); 390 391 result = udp6_lib_lookup2(net, saddr, sport, 392 &in6addr_any, hnum, dif, sdif, 393 hslot2, skb); 394 if (!IS_ERR_OR_NULL(result)) 395 goto done; 396 397 /* Cover address change/lookup/rehash race: see __udp4_lib_lookup() */ 398 result = udp6_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif, 399 udptable); 400 401 done: 402 if (IS_ERR(result)) 403 return NULL; 404 return result; 405 } 406 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 407 408 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 409 __be16 sport, __be16 dport, 410 struct udp_table *udptable) 411 { 412 const struct ipv6hdr *iph = ipv6_hdr(skb); 413 414 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 415 &iph->daddr, dport, inet6_iif(skb), 416 inet6_sdif(skb), udptable, skb); 417 } 418 419 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, 420 __be16 sport, __be16 dport) 421 { 422 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation]; 423 const struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + offset); 424 struct net *net = dev_net(skb->dev); 425 int iif, sdif; 426 427 inet6_get_iif_sdif(skb, &iif, &sdif); 428 429 return __udp6_lib_lookup(net, &iph->saddr, sport, 430 &iph->daddr, dport, iif, 431 sdif, net->ipv4.udp_table, NULL); 432 } 433 434 /* Must be called under rcu_read_lock(). 435 * Does increment socket refcount. 436 */ 437 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 438 struct sock *udp6_lib_lookup(const struct net *net, const struct in6_addr *saddr, __be16 sport, 439 const struct in6_addr *daddr, __be16 dport, int dif) 440 { 441 struct sock *sk; 442 443 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 444 dif, 0, net->ipv4.udp_table, NULL); 445 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 446 sk = NULL; 447 return sk; 448 } 449 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 450 #endif 451 452 /* do not use the scratch area len for jumbogram: their length execeeds the 453 * scratch area space; note that the IP6CB flags is still in the first 454 * cacheline, so checking for jumbograms is cheap 455 */ 456 static int udp6_skb_len(struct sk_buff *skb) 457 { 458 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 459 } 460 461 /* 462 * This should be easy, if there is something there we 463 * return it, otherwise we block. 464 */ 465 466 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 467 int flags, int *addr_len) 468 { 469 struct ipv6_pinfo *np = inet6_sk(sk); 470 struct inet_sock *inet = inet_sk(sk); 471 struct sk_buff *skb; 472 unsigned int ulen, copied; 473 int off, err, peeking = flags & MSG_PEEK; 474 int is_udplite = IS_UDPLITE(sk); 475 struct udp_mib __percpu *mib; 476 bool checksum_valid = false; 477 int is_udp4; 478 479 if (flags & MSG_ERRQUEUE) 480 return ipv6_recv_error(sk, msg, len, addr_len); 481 482 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 483 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 484 485 try_again: 486 off = sk_peek_offset(sk, flags); 487 skb = __skb_recv_udp(sk, flags, &off, &err); 488 if (!skb) 489 return err; 490 491 ulen = udp6_skb_len(skb); 492 copied = len; 493 if (copied > ulen - off) 494 copied = ulen - off; 495 else if (copied < ulen) 496 msg->msg_flags |= MSG_TRUNC; 497 498 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 499 mib = __UDPX_MIB(sk, is_udp4); 500 501 /* 502 * If checksum is needed at all, try to do it while copying the 503 * data. If the data is truncated, or if we only want a partial 504 * coverage checksum (UDP-Lite), do it before the copy. 505 */ 506 507 if (copied < ulen || peeking || 508 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 509 checksum_valid = udp_skb_csum_unnecessary(skb) || 510 !__udp_lib_checksum_complete(skb); 511 if (!checksum_valid) 512 goto csum_copy_err; 513 } 514 515 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 516 if (udp_skb_is_linear(skb)) 517 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 518 else 519 err = skb_copy_datagram_msg(skb, off, msg, copied); 520 } else { 521 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 522 if (err == -EINVAL) 523 goto csum_copy_err; 524 } 525 if (unlikely(err)) { 526 if (!peeking) { 527 atomic_inc(&sk->sk_drops); 528 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 529 } 530 kfree_skb(skb); 531 return err; 532 } 533 if (!peeking) 534 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 535 536 sock_recv_cmsgs(msg, sk, skb); 537 538 /* Copy the address. */ 539 if (msg->msg_name) { 540 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 541 sin6->sin6_family = AF_INET6; 542 sin6->sin6_port = udp_hdr(skb)->source; 543 sin6->sin6_flowinfo = 0; 544 545 if (is_udp4) { 546 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 547 &sin6->sin6_addr); 548 sin6->sin6_scope_id = 0; 549 } else { 550 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 551 sin6->sin6_scope_id = 552 ipv6_iface_scope_id(&sin6->sin6_addr, 553 inet6_iif(skb)); 554 } 555 *addr_len = sizeof(*sin6); 556 557 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 558 (struct sockaddr *)sin6, 559 addr_len); 560 } 561 562 if (udp_test_bit(GRO_ENABLED, sk)) 563 udp_cmsg_recv(msg, sk, skb); 564 565 if (np->rxopt.all) 566 ip6_datagram_recv_common_ctl(sk, msg, skb); 567 568 if (is_udp4) { 569 if (inet_cmsg_flags(inet)) 570 ip_cmsg_recv_offset(msg, sk, skb, 571 sizeof(struct udphdr), off); 572 } else { 573 if (np->rxopt.all) 574 ip6_datagram_recv_specific_ctl(sk, msg, skb); 575 } 576 577 err = copied; 578 if (flags & MSG_TRUNC) 579 err = ulen; 580 581 skb_consume_udp(sk, skb, peeking ? -err : err); 582 return err; 583 584 csum_copy_err: 585 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 586 udp_skb_destructor)) { 587 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 588 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 589 } 590 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM); 591 592 /* starting over for a new packet, but check if we need to yield */ 593 cond_resched(); 594 msg->msg_flags &= ~MSG_TRUNC; 595 goto try_again; 596 } 597 598 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 599 void udpv6_encap_enable(void) 600 { 601 static_branch_inc(&udpv6_encap_needed_key); 602 } 603 EXPORT_SYMBOL(udpv6_encap_enable); 604 605 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 606 * through error handlers in encapsulations looking for a match. 607 */ 608 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 609 struct inet6_skb_parm *opt, 610 u8 type, u8 code, int offset, __be32 info) 611 { 612 int i; 613 614 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 615 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 616 u8 type, u8 code, int offset, __be32 info); 617 const struct ip6_tnl_encap_ops *encap; 618 619 encap = rcu_dereference(ip6tun_encaps[i]); 620 if (!encap) 621 continue; 622 handler = encap->err_handler; 623 if (handler && !handler(skb, opt, type, code, offset, info)) 624 return 0; 625 } 626 627 return -ENOENT; 628 } 629 630 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 631 * reversing source and destination port: this will match tunnels that force the 632 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 633 * lwtunnels might actually break this assumption by being configured with 634 * different destination ports on endpoints, in this case we won't be able to 635 * trace ICMP messages back to them. 636 * 637 * If this doesn't match any socket, probe tunnels with arbitrary destination 638 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 639 * we've sent packets to won't necessarily match the local destination port. 640 * 641 * Then ask the tunnel implementation to match the error against a valid 642 * association. 643 * 644 * Return an error if we can't find a match, the socket if we need further 645 * processing, zero otherwise. 646 */ 647 static struct sock *__udp6_lib_err_encap(struct net *net, 648 const struct ipv6hdr *hdr, int offset, 649 struct udphdr *uh, 650 struct udp_table *udptable, 651 struct sock *sk, 652 struct sk_buff *skb, 653 struct inet6_skb_parm *opt, 654 u8 type, u8 code, __be32 info) 655 { 656 int (*lookup)(struct sock *sk, struct sk_buff *skb); 657 int network_offset, transport_offset; 658 struct udp_sock *up; 659 660 network_offset = skb_network_offset(skb); 661 transport_offset = skb_transport_offset(skb); 662 663 /* Network header needs to point to the outer IPv6 header inside ICMP */ 664 skb_reset_network_header(skb); 665 666 /* Transport header needs to point to the UDP header */ 667 skb_set_transport_header(skb, offset); 668 669 if (sk) { 670 up = udp_sk(sk); 671 672 lookup = READ_ONCE(up->encap_err_lookup); 673 if (lookup && lookup(sk, skb)) 674 sk = NULL; 675 676 goto out; 677 } 678 679 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 680 &hdr->saddr, uh->dest, 681 inet6_iif(skb), 0, udptable, skb); 682 if (sk) { 683 up = udp_sk(sk); 684 685 lookup = READ_ONCE(up->encap_err_lookup); 686 if (!lookup || lookup(sk, skb)) 687 sk = NULL; 688 } 689 690 out: 691 if (!sk) { 692 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 693 offset, info)); 694 } 695 696 skb_set_transport_header(skb, transport_offset); 697 skb_set_network_header(skb, network_offset); 698 699 return sk; 700 } 701 702 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 703 u8 type, u8 code, int offset, __be32 info, 704 struct udp_table *udptable) 705 { 706 struct ipv6_pinfo *np; 707 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 708 const struct in6_addr *saddr = &hdr->saddr; 709 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr; 710 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 711 bool tunnel = false; 712 struct sock *sk; 713 int harderr; 714 int err; 715 struct net *net = dev_net(skb->dev); 716 717 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 718 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 719 720 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) { 721 /* No socket for error: try tunnels before discarding */ 722 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 723 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 724 udptable, sk, skb, 725 opt, type, code, info); 726 if (!sk) 727 return 0; 728 } else 729 sk = ERR_PTR(-ENOENT); 730 731 if (IS_ERR(sk)) { 732 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 733 ICMP6_MIB_INERRORS); 734 return PTR_ERR(sk); 735 } 736 737 tunnel = true; 738 } 739 740 harderr = icmpv6_err_convert(type, code, &err); 741 np = inet6_sk(sk); 742 743 if (type == ICMPV6_PKT_TOOBIG) { 744 if (!ip6_sk_accept_pmtu(sk)) 745 goto out; 746 ip6_sk_update_pmtu(skb, sk, info); 747 if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT) 748 harderr = 1; 749 } 750 if (type == NDISC_REDIRECT) { 751 if (tunnel) { 752 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 753 READ_ONCE(sk->sk_mark), 754 sk_uid(sk)); 755 } else { 756 ip6_sk_redirect(skb, sk); 757 } 758 goto out; 759 } 760 761 /* Tunnels don't have an application socket: don't pass errors back */ 762 if (tunnel) { 763 if (udp_sk(sk)->encap_err_rcv) 764 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, 765 ntohl(info), (u8 *)(uh+1)); 766 goto out; 767 } 768 769 if (!inet6_test_bit(RECVERR6, sk)) { 770 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 771 goto out; 772 } else { 773 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 774 } 775 776 sk->sk_err = err; 777 sk_error_report(sk); 778 out: 779 return 0; 780 } 781 782 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 783 { 784 int rc; 785 786 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 787 sock_rps_save_rxhash(sk, skb); 788 sk_mark_napi_id(sk, skb); 789 sk_incoming_cpu_update(sk); 790 } else { 791 sk_mark_napi_id_once(sk, skb); 792 } 793 794 rc = __udp_enqueue_schedule_skb(sk, skb); 795 if (rc < 0) { 796 int is_udplite = IS_UDPLITE(sk); 797 enum skb_drop_reason drop_reason; 798 799 /* Note that an ENOMEM error is charged twice */ 800 if (rc == -ENOMEM) { 801 UDP6_INC_STATS(sock_net(sk), 802 UDP_MIB_RCVBUFERRORS, is_udplite); 803 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 804 } else { 805 UDP6_INC_STATS(sock_net(sk), 806 UDP_MIB_MEMERRORS, is_udplite); 807 drop_reason = SKB_DROP_REASON_PROTO_MEM; 808 } 809 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 810 trace_udp_fail_queue_rcv_skb(rc, sk, skb); 811 sk_skb_reason_drop(sk, skb, drop_reason); 812 return -1; 813 } 814 815 return 0; 816 } 817 818 static __inline__ int udpv6_err(struct sk_buff *skb, 819 struct inet6_skb_parm *opt, u8 type, 820 u8 code, int offset, __be32 info) 821 { 822 return __udp6_lib_err(skb, opt, type, code, offset, info, 823 dev_net(skb->dev)->ipv4.udp_table); 824 } 825 826 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 827 { 828 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 829 struct udp_sock *up = udp_sk(sk); 830 int is_udplite = IS_UDPLITE(sk); 831 832 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 833 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 834 goto drop; 835 } 836 nf_reset_ct(skb); 837 838 if (static_branch_unlikely(&udpv6_encap_needed_key) && 839 READ_ONCE(up->encap_type)) { 840 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 841 842 /* 843 * This is an encapsulation socket so pass the skb to 844 * the socket's udp_encap_rcv() hook. Otherwise, just 845 * fall through and pass this up the UDP socket. 846 * up->encap_rcv() returns the following value: 847 * =0 if skb was successfully passed to the encap 848 * handler or was discarded by it. 849 * >0 if skb should be passed on to UDP. 850 * <0 if skb should be resubmitted as proto -N 851 */ 852 853 /* if we're overly short, let UDP handle it */ 854 encap_rcv = READ_ONCE(up->encap_rcv); 855 if (encap_rcv) { 856 int ret; 857 858 /* Verify checksum before giving to encap */ 859 if (udp_lib_checksum_complete(skb)) 860 goto csum_error; 861 862 ret = encap_rcv(sk, skb); 863 if (ret <= 0) { 864 __UDP6_INC_STATS(sock_net(sk), 865 UDP_MIB_INDATAGRAMS, 866 is_udplite); 867 return -ret; 868 } 869 } 870 871 /* FALLTHROUGH -- it's a UDP Packet */ 872 } 873 874 /* 875 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 876 */ 877 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) { 878 u16 pcrlen = READ_ONCE(up->pcrlen); 879 880 if (pcrlen == 0) { /* full coverage was set */ 881 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 882 UDP_SKB_CB(skb)->cscov, skb->len); 883 goto drop; 884 } 885 if (UDP_SKB_CB(skb)->cscov < pcrlen) { 886 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 887 UDP_SKB_CB(skb)->cscov, pcrlen); 888 goto drop; 889 } 890 } 891 892 prefetch(&sk->sk_rmem_alloc); 893 if (rcu_access_pointer(sk->sk_filter) && 894 udp_lib_checksum_complete(skb)) 895 goto csum_error; 896 897 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 898 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 899 goto drop; 900 } 901 902 udp_csum_pull_header(skb); 903 904 skb_dst_drop(skb); 905 906 return __udpv6_queue_rcv_skb(sk, skb); 907 908 csum_error: 909 drop_reason = SKB_DROP_REASON_UDP_CSUM; 910 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 911 drop: 912 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 913 atomic_inc(&sk->sk_drops); 914 sk_skb_reason_drop(sk, skb, drop_reason); 915 return -1; 916 } 917 918 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 919 { 920 struct sk_buff *next, *segs; 921 int ret; 922 923 if (likely(!udp_unexpected_gso(sk, skb))) 924 return udpv6_queue_rcv_one_skb(sk, skb); 925 926 __skb_push(skb, -skb_mac_offset(skb)); 927 segs = udp_rcv_segment(sk, skb, false); 928 skb_list_walk_safe(segs, skb, next) { 929 __skb_pull(skb, skb_transport_offset(skb)); 930 931 udp_post_segment_fix_csum(skb); 932 ret = udpv6_queue_rcv_one_skb(sk, skb); 933 if (ret > 0) 934 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 935 true); 936 } 937 return 0; 938 } 939 940 static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk, 941 __be16 loc_port, const struct in6_addr *loc_addr, 942 __be16 rmt_port, const struct in6_addr *rmt_addr, 943 int dif, int sdif, unsigned short hnum) 944 { 945 const struct inet_sock *inet = inet_sk(sk); 946 947 if (!net_eq(sock_net(sk), net)) 948 return false; 949 950 if (udp_sk(sk)->udp_port_hash != hnum || 951 sk->sk_family != PF_INET6 || 952 (inet->inet_dport && inet->inet_dport != rmt_port) || 953 (!ipv6_addr_any(&sk->sk_v6_daddr) && 954 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 955 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) || 956 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 957 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 958 return false; 959 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 960 return false; 961 return true; 962 } 963 964 static void udp6_csum_zero_error(struct sk_buff *skb) 965 { 966 /* RFC 2460 section 8.1 says that we SHOULD log 967 * this error. Well, it is reasonable. 968 */ 969 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 970 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 971 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 972 } 973 974 /* 975 * Note: called only from the BH handler context, 976 * so we don't need to lock the hashes. 977 */ 978 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 979 const struct in6_addr *saddr, const struct in6_addr *daddr, 980 struct udp_table *udptable, int proto) 981 { 982 struct sock *sk, *first = NULL; 983 const struct udphdr *uh = udp_hdr(skb); 984 unsigned short hnum = ntohs(uh->dest); 985 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 986 unsigned int offset = offsetof(typeof(*sk), sk_node); 987 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 988 int dif = inet6_iif(skb); 989 int sdif = inet6_sdif(skb); 990 struct hlist_node *node; 991 struct sk_buff *nskb; 992 993 if (use_hash2) { 994 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 995 udptable->mask; 996 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 997 start_lookup: 998 hslot = &udptable->hash2[hash2].hslot; 999 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1000 } 1001 1002 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 1003 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 1004 uh->source, saddr, dif, sdif, 1005 hnum)) 1006 continue; 1007 /* If zero checksum and no_check is not on for 1008 * the socket then skip it. 1009 */ 1010 if (!uh->check && !udp_get_no_check6_rx(sk)) 1011 continue; 1012 if (!first) { 1013 first = sk; 1014 continue; 1015 } 1016 nskb = skb_clone(skb, GFP_ATOMIC); 1017 if (unlikely(!nskb)) { 1018 atomic_inc(&sk->sk_drops); 1019 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 1020 IS_UDPLITE(sk)); 1021 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 1022 IS_UDPLITE(sk)); 1023 continue; 1024 } 1025 1026 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 1027 consume_skb(nskb); 1028 } 1029 1030 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1031 if (use_hash2 && hash2 != hash2_any) { 1032 hash2 = hash2_any; 1033 goto start_lookup; 1034 } 1035 1036 if (first) { 1037 if (udpv6_queue_rcv_skb(first, skb) > 0) 1038 consume_skb(skb); 1039 } else { 1040 kfree_skb(skb); 1041 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 1042 proto == IPPROTO_UDPLITE); 1043 } 1044 return 0; 1045 } 1046 1047 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1048 { 1049 if (udp_sk_rx_dst_set(sk, dst)) 1050 sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst)); 1051 } 1052 1053 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 1054 * return code conversion for ip layer consumption 1055 */ 1056 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 1057 struct udphdr *uh) 1058 { 1059 int ret; 1060 1061 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1062 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 1063 1064 ret = udpv6_queue_rcv_skb(sk, skb); 1065 1066 /* a return value > 0 means to resubmit the input */ 1067 if (ret > 0) 1068 return ret; 1069 return 0; 1070 } 1071 1072 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1073 int proto) 1074 { 1075 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 1076 const struct in6_addr *saddr, *daddr; 1077 struct net *net = dev_net(skb->dev); 1078 struct sock *sk = NULL; 1079 struct udphdr *uh; 1080 bool refcounted; 1081 u32 ulen = 0; 1082 1083 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1084 goto discard; 1085 1086 saddr = &ipv6_hdr(skb)->saddr; 1087 daddr = &ipv6_hdr(skb)->daddr; 1088 uh = udp_hdr(skb); 1089 1090 ulen = ntohs(uh->len); 1091 if (ulen > skb->len) 1092 goto short_packet; 1093 1094 if (proto == IPPROTO_UDP) { 1095 /* UDP validates ulen. */ 1096 1097 /* Check for jumbo payload */ 1098 if (ulen == 0) 1099 ulen = skb->len; 1100 1101 if (ulen < sizeof(*uh)) 1102 goto short_packet; 1103 1104 if (ulen < skb->len) { 1105 if (pskb_trim_rcsum(skb, ulen)) 1106 goto short_packet; 1107 saddr = &ipv6_hdr(skb)->saddr; 1108 daddr = &ipv6_hdr(skb)->daddr; 1109 uh = udp_hdr(skb); 1110 } 1111 } 1112 1113 if (udp6_csum_init(skb, uh, proto)) 1114 goto csum_error; 1115 1116 /* Check if the socket is already available, e.g. due to early demux */ 1117 sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 1118 &refcounted, udp6_ehashfn); 1119 if (IS_ERR(sk)) 1120 goto no_sk; 1121 1122 if (sk) { 1123 struct dst_entry *dst = skb_dst(skb); 1124 int ret; 1125 1126 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 1127 udp6_sk_rx_dst_set(sk, dst); 1128 1129 if (!uh->check && !udp_get_no_check6_rx(sk)) { 1130 if (refcounted) 1131 sock_put(sk); 1132 goto report_csum_error; 1133 } 1134 1135 ret = udp6_unicast_rcv_skb(sk, skb, uh); 1136 if (refcounted) 1137 sock_put(sk); 1138 return ret; 1139 } 1140 1141 /* 1142 * Multicast receive code 1143 */ 1144 if (ipv6_addr_is_multicast(daddr)) 1145 return __udp6_lib_mcast_deliver(net, skb, 1146 saddr, daddr, udptable, proto); 1147 1148 /* Unicast */ 1149 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1150 if (sk) { 1151 if (!uh->check && !udp_get_no_check6_rx(sk)) 1152 goto report_csum_error; 1153 return udp6_unicast_rcv_skb(sk, skb, uh); 1154 } 1155 no_sk: 1156 reason = SKB_DROP_REASON_NO_SOCKET; 1157 1158 if (!uh->check) 1159 goto report_csum_error; 1160 1161 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1162 goto discard; 1163 nf_reset_ct(skb); 1164 1165 if (udp_lib_checksum_complete(skb)) 1166 goto csum_error; 1167 1168 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1169 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 1170 1171 sk_skb_reason_drop(sk, skb, reason); 1172 return 0; 1173 1174 short_packet: 1175 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1176 reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1177 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 1178 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1179 saddr, ntohs(uh->source), 1180 ulen, skb->len, 1181 daddr, ntohs(uh->dest)); 1182 goto discard; 1183 1184 report_csum_error: 1185 udp6_csum_zero_error(skb); 1186 csum_error: 1187 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1188 reason = SKB_DROP_REASON_UDP_CSUM; 1189 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1190 discard: 1191 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1192 sk_skb_reason_drop(sk, skb, reason); 1193 return 0; 1194 } 1195 1196 1197 static struct sock *__udp6_lib_demux_lookup(struct net *net, 1198 __be16 loc_port, const struct in6_addr *loc_addr, 1199 __be16 rmt_port, const struct in6_addr *rmt_addr, 1200 int dif, int sdif) 1201 { 1202 struct udp_table *udptable = net->ipv4.udp_table; 1203 unsigned short hnum = ntohs(loc_port); 1204 struct udp_hslot *hslot2; 1205 unsigned int hash2; 1206 __portpair ports; 1207 struct sock *sk; 1208 1209 hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1210 hslot2 = udp_hashslot2(udptable, hash2); 1211 ports = INET_COMBINED_PORTS(rmt_port, hnum); 1212 1213 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1214 if (sk->sk_state == TCP_ESTABLISHED && 1215 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif)) 1216 return sk; 1217 /* Only check first socket in chain */ 1218 break; 1219 } 1220 return NULL; 1221 } 1222 1223 void udp_v6_early_demux(struct sk_buff *skb) 1224 { 1225 struct net *net = dev_net(skb->dev); 1226 const struct udphdr *uh; 1227 struct sock *sk; 1228 struct dst_entry *dst; 1229 int dif = skb->dev->ifindex; 1230 int sdif = inet6_sdif(skb); 1231 1232 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1233 sizeof(struct udphdr))) 1234 return; 1235 1236 uh = udp_hdr(skb); 1237 1238 if (skb->pkt_type == PACKET_HOST) 1239 sk = __udp6_lib_demux_lookup(net, uh->dest, 1240 &ipv6_hdr(skb)->daddr, 1241 uh->source, &ipv6_hdr(skb)->saddr, 1242 dif, sdif); 1243 else 1244 return; 1245 1246 if (!sk) 1247 return; 1248 1249 skb->sk = sk; 1250 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk)); 1251 skb->destructor = sock_pfree; 1252 dst = rcu_dereference(sk->sk_rx_dst); 1253 1254 if (dst) 1255 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1256 if (dst) { 1257 /* set noref for now. 1258 * any place which wants to hold dst has to call 1259 * dst_hold_safe() 1260 */ 1261 skb_dst_set_noref(skb, dst); 1262 } 1263 } 1264 1265 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1266 { 1267 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 1268 } 1269 1270 /* 1271 * Throw away all pending data and cancel the corking. Socket is locked. 1272 */ 1273 static void udp_v6_flush_pending_frames(struct sock *sk) 1274 { 1275 struct udp_sock *up = udp_sk(sk); 1276 1277 if (up->pending == AF_INET) 1278 udp_flush_pending_frames(sk); 1279 else if (up->pending) { 1280 up->len = 0; 1281 WRITE_ONCE(up->pending, 0); 1282 ip6_flush_pending_frames(sk); 1283 } 1284 } 1285 1286 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1287 int addr_len) 1288 { 1289 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1290 return -EINVAL; 1291 /* The following checks are replicated from __ip6_datagram_connect() 1292 * and intended to prevent BPF program called below from accessing 1293 * bytes that are out of the bound specified by user in addr_len. 1294 */ 1295 if (uaddr->sa_family == AF_INET) { 1296 if (ipv6_only_sock(sk)) 1297 return -EAFNOSUPPORT; 1298 return udp_pre_connect(sk, uaddr, addr_len); 1299 } 1300 1301 if (addr_len < SIN6_LEN_RFC2133) 1302 return -EINVAL; 1303 1304 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len); 1305 } 1306 1307 static int udpv6_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1308 { 1309 int res; 1310 1311 lock_sock(sk); 1312 res = __ip6_datagram_connect(sk, uaddr, addr_len); 1313 if (!res) 1314 udp6_hash4(sk); 1315 release_sock(sk); 1316 return res; 1317 } 1318 1319 /** 1320 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1321 * @sk: socket we are sending on 1322 * @skb: sk_buff containing the filled-in UDP header 1323 * (checksum field must be zeroed out) 1324 * @saddr: source address 1325 * @daddr: destination address 1326 * @len: length of packet 1327 */ 1328 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1329 const struct in6_addr *saddr, 1330 const struct in6_addr *daddr, int len) 1331 { 1332 unsigned int offset; 1333 struct udphdr *uh = udp_hdr(skb); 1334 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1335 __wsum csum = 0; 1336 1337 if (!frags) { 1338 /* Only one fragment on the socket. */ 1339 skb->csum_start = skb_transport_header(skb) - skb->head; 1340 skb->csum_offset = offsetof(struct udphdr, check); 1341 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1342 } else { 1343 /* 1344 * HW-checksum won't work as there are two or more 1345 * fragments on the socket so that all csums of sk_buffs 1346 * should be together 1347 */ 1348 offset = skb_transport_offset(skb); 1349 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1350 csum = skb->csum; 1351 1352 skb->ip_summed = CHECKSUM_NONE; 1353 1354 do { 1355 csum = csum_add(csum, frags->csum); 1356 } while ((frags = frags->next)); 1357 1358 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1359 csum); 1360 if (uh->check == 0) 1361 uh->check = CSUM_MANGLED_0; 1362 } 1363 } 1364 1365 /* 1366 * Sending 1367 */ 1368 1369 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1370 struct inet_cork *cork) 1371 { 1372 struct sock *sk = skb->sk; 1373 struct udphdr *uh; 1374 int err = 0; 1375 int is_udplite = IS_UDPLITE(sk); 1376 __wsum csum = 0; 1377 int offset = skb_transport_offset(skb); 1378 int len = skb->len - offset; 1379 int datalen = len - sizeof(*uh); 1380 1381 /* 1382 * Create a UDP header 1383 */ 1384 uh = udp_hdr(skb); 1385 uh->source = fl6->fl6_sport; 1386 uh->dest = fl6->fl6_dport; 1387 uh->len = htons(len); 1388 uh->check = 0; 1389 1390 if (cork->gso_size) { 1391 const int hlen = skb_network_header_len(skb) + 1392 sizeof(struct udphdr); 1393 1394 if (hlen + min(datalen, cork->gso_size) > cork->fragsize) { 1395 kfree_skb(skb); 1396 return -EMSGSIZE; 1397 } 1398 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1399 kfree_skb(skb); 1400 return -EINVAL; 1401 } 1402 if (udp_get_no_check6_tx(sk)) { 1403 kfree_skb(skb); 1404 return -EINVAL; 1405 } 1406 if (is_udplite || dst_xfrm(skb_dst(skb))) { 1407 kfree_skb(skb); 1408 return -EIO; 1409 } 1410 1411 if (datalen > cork->gso_size) { 1412 skb_shinfo(skb)->gso_size = cork->gso_size; 1413 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1414 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1415 cork->gso_size); 1416 1417 /* Don't checksum the payload, skb will get segmented */ 1418 goto csum_partial; 1419 } 1420 } 1421 1422 if (is_udplite) 1423 csum = udplite_csum(skb); 1424 else if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */ 1425 skb->ip_summed = CHECKSUM_NONE; 1426 goto send; 1427 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1428 csum_partial: 1429 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1430 goto send; 1431 } else 1432 csum = udp_csum(skb); 1433 1434 /* add protocol-dependent pseudo-header */ 1435 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1436 len, fl6->flowi6_proto, csum); 1437 if (uh->check == 0) 1438 uh->check = CSUM_MANGLED_0; 1439 1440 send: 1441 err = ip6_send_skb(skb); 1442 if (err) { 1443 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) { 1444 UDP6_INC_STATS(sock_net(sk), 1445 UDP_MIB_SNDBUFERRORS, is_udplite); 1446 err = 0; 1447 } 1448 } else { 1449 UDP6_INC_STATS(sock_net(sk), 1450 UDP_MIB_OUTDATAGRAMS, is_udplite); 1451 } 1452 return err; 1453 } 1454 1455 static int udp_v6_push_pending_frames(struct sock *sk) 1456 { 1457 struct sk_buff *skb; 1458 struct udp_sock *up = udp_sk(sk); 1459 int err = 0; 1460 1461 if (up->pending == AF_INET) 1462 return udp_push_pending_frames(sk); 1463 1464 skb = ip6_finish_skb(sk); 1465 if (!skb) 1466 goto out; 1467 1468 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6, 1469 &inet_sk(sk)->cork.base); 1470 out: 1471 up->len = 0; 1472 WRITE_ONCE(up->pending, 0); 1473 return err; 1474 } 1475 1476 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1477 { 1478 struct ipv6_txoptions opt_space; 1479 struct udp_sock *up = udp_sk(sk); 1480 struct inet_sock *inet = inet_sk(sk); 1481 struct ipv6_pinfo *np = inet6_sk(sk); 1482 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1483 struct in6_addr *daddr, *final_p, final; 1484 struct ipv6_txoptions *opt = NULL; 1485 struct ipv6_txoptions *opt_to_free = NULL; 1486 struct ip6_flowlabel *flowlabel = NULL; 1487 struct inet_cork_full cork; 1488 struct flowi6 *fl6 = &cork.fl.u.ip6; 1489 struct dst_entry *dst; 1490 struct ipcm6_cookie ipc6; 1491 int addr_len = msg->msg_namelen; 1492 bool connected = false; 1493 int ulen = len; 1494 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE; 1495 int err; 1496 int is_udplite = IS_UDPLITE(sk); 1497 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1498 1499 ipcm6_init_sk(&ipc6, sk); 1500 ipc6.gso_size = READ_ONCE(up->gso_size); 1501 1502 /* destination address check */ 1503 if (sin6) { 1504 if (addr_len < offsetof(struct sockaddr, sa_data)) 1505 return -EINVAL; 1506 1507 switch (sin6->sin6_family) { 1508 case AF_INET6: 1509 if (addr_len < SIN6_LEN_RFC2133) 1510 return -EINVAL; 1511 daddr = &sin6->sin6_addr; 1512 if (ipv6_addr_any(daddr) && 1513 ipv6_addr_v4mapped(&np->saddr)) 1514 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1515 daddr); 1516 break; 1517 case AF_INET: 1518 goto do_udp_sendmsg; 1519 case AF_UNSPEC: 1520 msg->msg_name = sin6 = NULL; 1521 msg->msg_namelen = addr_len = 0; 1522 daddr = NULL; 1523 break; 1524 default: 1525 return -EINVAL; 1526 } 1527 } else if (!READ_ONCE(up->pending)) { 1528 if (sk->sk_state != TCP_ESTABLISHED) 1529 return -EDESTADDRREQ; 1530 daddr = &sk->sk_v6_daddr; 1531 } else 1532 daddr = NULL; 1533 1534 if (daddr) { 1535 if (ipv6_addr_v4mapped(daddr)) { 1536 struct sockaddr_in sin; 1537 sin.sin_family = AF_INET; 1538 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1539 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1540 msg->msg_name = &sin; 1541 msg->msg_namelen = sizeof(sin); 1542 do_udp_sendmsg: 1543 err = ipv6_only_sock(sk) ? 1544 -ENETUNREACH : udp_sendmsg(sk, msg, len); 1545 msg->msg_name = sin6; 1546 msg->msg_namelen = addr_len; 1547 return err; 1548 } 1549 } 1550 1551 /* Rough check on arithmetic overflow, 1552 better check is made in ip6_append_data(). 1553 */ 1554 if (len > INT_MAX - sizeof(struct udphdr)) 1555 return -EMSGSIZE; 1556 1557 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1558 if (READ_ONCE(up->pending)) { 1559 if (READ_ONCE(up->pending) == AF_INET) 1560 return udp_sendmsg(sk, msg, len); 1561 /* 1562 * There are pending frames. 1563 * The socket lock must be held while it's corked. 1564 */ 1565 lock_sock(sk); 1566 if (likely(up->pending)) { 1567 if (unlikely(up->pending != AF_INET6)) { 1568 release_sock(sk); 1569 return -EAFNOSUPPORT; 1570 } 1571 dst = NULL; 1572 goto do_append_data; 1573 } 1574 release_sock(sk); 1575 } 1576 ulen += sizeof(struct udphdr); 1577 1578 memset(fl6, 0, sizeof(*fl6)); 1579 1580 if (sin6) { 1581 if (sin6->sin6_port == 0) 1582 return -EINVAL; 1583 1584 fl6->fl6_dport = sin6->sin6_port; 1585 daddr = &sin6->sin6_addr; 1586 1587 if (inet6_test_bit(SNDFLOW, sk)) { 1588 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1589 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) { 1590 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1591 if (IS_ERR(flowlabel)) 1592 return -EINVAL; 1593 } 1594 } 1595 1596 /* 1597 * Otherwise it will be difficult to maintain 1598 * sk->sk_dst_cache. 1599 */ 1600 if (sk->sk_state == TCP_ESTABLISHED && 1601 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1602 daddr = &sk->sk_v6_daddr; 1603 1604 if (addr_len >= sizeof(struct sockaddr_in6) && 1605 sin6->sin6_scope_id && 1606 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1607 fl6->flowi6_oif = sin6->sin6_scope_id; 1608 } else { 1609 if (sk->sk_state != TCP_ESTABLISHED) 1610 return -EDESTADDRREQ; 1611 1612 fl6->fl6_dport = inet->inet_dport; 1613 daddr = &sk->sk_v6_daddr; 1614 fl6->flowlabel = np->flow_label; 1615 connected = true; 1616 } 1617 1618 if (!fl6->flowi6_oif) 1619 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if); 1620 1621 if (!fl6->flowi6_oif) 1622 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1623 1624 fl6->flowi6_uid = sk_uid(sk); 1625 1626 if (msg->msg_controllen) { 1627 opt = &opt_space; 1628 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1629 opt->tot_len = sizeof(*opt); 1630 ipc6.opt = opt; 1631 1632 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1633 if (err > 0) { 1634 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6, 1635 &ipc6); 1636 connected = false; 1637 } 1638 if (err < 0) { 1639 fl6_sock_release(flowlabel); 1640 return err; 1641 } 1642 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1643 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1644 if (IS_ERR(flowlabel)) 1645 return -EINVAL; 1646 } 1647 if (!(opt->opt_nflen|opt->opt_flen)) 1648 opt = NULL; 1649 } 1650 if (!opt) { 1651 opt = txopt_get(np); 1652 opt_to_free = opt; 1653 } 1654 if (flowlabel) 1655 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1656 opt = ipv6_fixup_options(&opt_space, opt); 1657 ipc6.opt = opt; 1658 1659 fl6->flowi6_proto = sk->sk_protocol; 1660 fl6->flowi6_mark = ipc6.sockc.mark; 1661 fl6->daddr = *daddr; 1662 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr)) 1663 fl6->saddr = np->saddr; 1664 fl6->fl6_sport = inet->inet_sport; 1665 1666 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1667 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1668 (struct sockaddr *)sin6, 1669 &addr_len, 1670 &fl6->saddr); 1671 if (err) 1672 goto out_no_dst; 1673 if (sin6) { 1674 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1675 /* BPF program rewrote IPv6-only by IPv4-mapped 1676 * IPv6. It's currently unsupported. 1677 */ 1678 err = -ENOTSUPP; 1679 goto out_no_dst; 1680 } 1681 if (sin6->sin6_port == 0) { 1682 /* BPF program set invalid port. Reject it. */ 1683 err = -EINVAL; 1684 goto out_no_dst; 1685 } 1686 fl6->fl6_dport = sin6->sin6_port; 1687 fl6->daddr = sin6->sin6_addr; 1688 } 1689 } 1690 1691 if (ipv6_addr_any(&fl6->daddr)) 1692 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1693 1694 final_p = fl6_update_dst(fl6, opt, &final); 1695 if (final_p) 1696 connected = false; 1697 1698 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) { 1699 fl6->flowi6_oif = READ_ONCE(np->mcast_oif); 1700 connected = false; 1701 } else if (!fl6->flowi6_oif) 1702 fl6->flowi6_oif = READ_ONCE(np->ucast_oif); 1703 1704 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6)); 1705 1706 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel); 1707 1708 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected); 1709 if (IS_ERR(dst)) { 1710 err = PTR_ERR(dst); 1711 dst = NULL; 1712 goto out; 1713 } 1714 1715 if (ipc6.hlimit < 0) 1716 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst); 1717 1718 if (msg->msg_flags&MSG_CONFIRM) 1719 goto do_confirm; 1720 back_from_confirm: 1721 1722 /* Lockless fast path for the non-corking case */ 1723 if (!corkreq) { 1724 struct sk_buff *skb; 1725 1726 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1727 sizeof(struct udphdr), &ipc6, 1728 dst_rt6_info(dst), 1729 msg->msg_flags, &cork); 1730 err = PTR_ERR(skb); 1731 if (!IS_ERR_OR_NULL(skb)) 1732 err = udp_v6_send_skb(skb, fl6, &cork.base); 1733 /* ip6_make_skb steals dst reference */ 1734 goto out_no_dst; 1735 } 1736 1737 lock_sock(sk); 1738 if (unlikely(up->pending)) { 1739 /* The socket is already corked while preparing it. */ 1740 /* ... which is an evident application bug. --ANK */ 1741 release_sock(sk); 1742 1743 net_dbg_ratelimited("udp cork app bug 2\n"); 1744 err = -EINVAL; 1745 goto out; 1746 } 1747 1748 WRITE_ONCE(up->pending, AF_INET6); 1749 1750 do_append_data: 1751 up->len += ulen; 1752 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1753 &ipc6, fl6, dst_rt6_info(dst), 1754 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1755 if (err) 1756 udp_v6_flush_pending_frames(sk); 1757 else if (!corkreq) 1758 err = udp_v6_push_pending_frames(sk); 1759 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1760 WRITE_ONCE(up->pending, 0); 1761 1762 if (err > 0) 1763 err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0; 1764 release_sock(sk); 1765 1766 out: 1767 dst_release(dst); 1768 out_no_dst: 1769 fl6_sock_release(flowlabel); 1770 txopt_put(opt_to_free); 1771 if (!err) 1772 return len; 1773 /* 1774 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1775 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1776 * we don't have a good statistic (IpOutDiscards but it can be too many 1777 * things). We could add another new stat but at least for now that 1778 * seems like overkill. 1779 */ 1780 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1781 UDP6_INC_STATS(sock_net(sk), 1782 UDP_MIB_SNDBUFERRORS, is_udplite); 1783 } 1784 return err; 1785 1786 do_confirm: 1787 if (msg->msg_flags & MSG_PROBE) 1788 dst_confirm_neigh(dst, &fl6->daddr); 1789 if (!(msg->msg_flags&MSG_PROBE) || len) 1790 goto back_from_confirm; 1791 err = 0; 1792 goto out; 1793 } 1794 EXPORT_SYMBOL(udpv6_sendmsg); 1795 1796 static void udpv6_splice_eof(struct socket *sock) 1797 { 1798 struct sock *sk = sock->sk; 1799 struct udp_sock *up = udp_sk(sk); 1800 1801 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk)) 1802 return; 1803 1804 lock_sock(sk); 1805 if (up->pending && !udp_test_bit(CORK, sk)) 1806 udp_v6_push_pending_frames(sk); 1807 release_sock(sk); 1808 } 1809 1810 void udpv6_destroy_sock(struct sock *sk) 1811 { 1812 struct udp_sock *up = udp_sk(sk); 1813 lock_sock(sk); 1814 1815 /* protects from races with udp_abort() */ 1816 sock_set_flag(sk, SOCK_DEAD); 1817 udp_v6_flush_pending_frames(sk); 1818 release_sock(sk); 1819 1820 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1821 if (up->encap_type) { 1822 void (*encap_destroy)(struct sock *sk); 1823 encap_destroy = READ_ONCE(up->encap_destroy); 1824 if (encap_destroy) 1825 encap_destroy(sk); 1826 } 1827 if (udp_test_bit(ENCAP_ENABLED, sk)) { 1828 static_branch_dec(&udpv6_encap_needed_key); 1829 udp_encap_disable(); 1830 udp_tunnel_cleanup_gro(sk); 1831 } 1832 } 1833 } 1834 1835 /* 1836 * Socket option code for UDP 1837 */ 1838 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1839 unsigned int optlen) 1840 { 1841 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 1842 return udp_lib_setsockopt(sk, level, optname, 1843 optval, optlen, 1844 udp_v6_push_pending_frames); 1845 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1846 } 1847 1848 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1849 char __user *optval, int __user *optlen) 1850 { 1851 if (level == SOL_UDP || level == SOL_UDPLITE) 1852 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1853 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1854 } 1855 1856 1857 /* ------------------------------------------------------------------------ */ 1858 #ifdef CONFIG_PROC_FS 1859 int udp6_seq_show(struct seq_file *seq, void *v) 1860 { 1861 if (v == SEQ_START_TOKEN) { 1862 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1863 } else { 1864 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1865 const struct inet_sock *inet = inet_sk((const struct sock *)v); 1866 __u16 srcp = ntohs(inet->inet_sport); 1867 __u16 destp = ntohs(inet->inet_dport); 1868 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1869 udp_rqueue_get(v), bucket); 1870 } 1871 return 0; 1872 } 1873 1874 const struct seq_operations udp6_seq_ops = { 1875 .start = udp_seq_start, 1876 .next = udp_seq_next, 1877 .stop = udp_seq_stop, 1878 .show = udp6_seq_show, 1879 }; 1880 EXPORT_SYMBOL(udp6_seq_ops); 1881 1882 static struct udp_seq_afinfo udp6_seq_afinfo = { 1883 .family = AF_INET6, 1884 .udp_table = NULL, 1885 }; 1886 1887 int __net_init udp6_proc_init(struct net *net) 1888 { 1889 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1890 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1891 return -ENOMEM; 1892 return 0; 1893 } 1894 1895 void udp6_proc_exit(struct net *net) 1896 { 1897 remove_proc_entry("udp6", net->proc_net); 1898 } 1899 #endif /* CONFIG_PROC_FS */ 1900 1901 /* ------------------------------------------------------------------------ */ 1902 1903 struct proto udpv6_prot = { 1904 .name = "UDPv6", 1905 .owner = THIS_MODULE, 1906 .close = udp_lib_close, 1907 .pre_connect = udpv6_pre_connect, 1908 .connect = udpv6_connect, 1909 .disconnect = udp_disconnect, 1910 .ioctl = udp_ioctl, 1911 .init = udpv6_init_sock, 1912 .destroy = udpv6_destroy_sock, 1913 .setsockopt = udpv6_setsockopt, 1914 .getsockopt = udpv6_getsockopt, 1915 .sendmsg = udpv6_sendmsg, 1916 .recvmsg = udpv6_recvmsg, 1917 .splice_eof = udpv6_splice_eof, 1918 .release_cb = ip6_datagram_release_cb, 1919 .hash = udp_lib_hash, 1920 .unhash = udp_lib_unhash, 1921 .rehash = udp_v6_rehash, 1922 .get_port = udp_v6_get_port, 1923 .put_port = udp_lib_unhash, 1924 #ifdef CONFIG_BPF_SYSCALL 1925 .psock_update_sk_prot = udp_bpf_update_proto, 1926 #endif 1927 1928 .memory_allocated = &net_aligned_data.udp_memory_allocated, 1929 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 1930 1931 .sysctl_mem = sysctl_udp_mem, 1932 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1933 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1934 .obj_size = sizeof(struct udp6_sock), 1935 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6), 1936 .h.udp_table = NULL, 1937 .diag_destroy = udp_abort, 1938 }; 1939 1940 static struct inet_protosw udpv6_protosw = { 1941 .type = SOCK_DGRAM, 1942 .protocol = IPPROTO_UDP, 1943 .prot = &udpv6_prot, 1944 .ops = &inet6_dgram_ops, 1945 .flags = INET_PROTOSW_PERMANENT, 1946 }; 1947 1948 int __init udpv6_init(void) 1949 { 1950 int ret; 1951 1952 net_hotdata.udpv6_protocol = (struct inet6_protocol) { 1953 .handler = udpv6_rcv, 1954 .err_handler = udpv6_err, 1955 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL, 1956 }; 1957 ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1958 if (ret) 1959 goto out; 1960 1961 ret = inet6_register_protosw(&udpv6_protosw); 1962 if (ret) 1963 goto out_udpv6_protocol; 1964 out: 1965 return ret; 1966 1967 out_udpv6_protocol: 1968 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1969 goto out; 1970 } 1971 1972 void udpv6_exit(void) 1973 { 1974 inet6_unregister_protosw(&udpv6_protosw); 1975 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1976 } 1977