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 38 #include <net/addrconf.h> 39 #include <net/ndisc.h> 40 #include <net/protocol.h> 41 #include <net/transp_v6.h> 42 #include <net/ip6_route.h> 43 #include <net/raw.h> 44 #include <net/seg6.h> 45 #include <net/tcp_states.h> 46 #include <net/ip6_checksum.h> 47 #include <net/ip6_tunnel.h> 48 #include <trace/events/udp.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 #include <net/gro.h> 55 56 #include <linux/proc_fs.h> 57 #include <linux/seq_file.h> 58 #include <trace/events/skb.h> 59 #include "udp_impl.h" 60 61 static void udpv6_destruct_sock(struct sock *sk) 62 { 63 udp_destruct_common(sk); 64 inet6_sock_destruct(sk); 65 } 66 67 int udpv6_init_sock(struct sock *sk) 68 { 69 udp_lib_init_sock(sk); 70 sk->sk_destruct = udpv6_destruct_sock; 71 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 72 return 0; 73 } 74 75 INDIRECT_CALLABLE_SCOPE 76 u32 udp6_ehashfn(const struct net *net, 77 const struct in6_addr *laddr, 78 const u16 lport, 79 const struct in6_addr *faddr, 80 const __be16 fport) 81 { 82 static u32 udp6_ehash_secret __read_mostly; 83 static u32 udp_ipv6_hash_secret __read_mostly; 84 85 u32 lhash, fhash; 86 87 net_get_random_once(&udp6_ehash_secret, 88 sizeof(udp6_ehash_secret)); 89 net_get_random_once(&udp_ipv6_hash_secret, 90 sizeof(udp_ipv6_hash_secret)); 91 92 lhash = (__force u32)laddr->s6_addr32[3]; 93 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 94 95 return __inet6_ehashfn(lhash, lport, fhash, fport, 96 udp6_ehash_secret + net_hash_mix(net)); 97 } 98 99 int udp_v6_get_port(struct sock *sk, unsigned short snum) 100 { 101 unsigned int hash2_nulladdr = 102 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 103 unsigned int hash2_partial = 104 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 105 106 /* precompute partial secondary hash */ 107 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 108 return udp_lib_get_port(sk, snum, hash2_nulladdr); 109 } 110 111 void udp_v6_rehash(struct sock *sk) 112 { 113 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 114 &sk->sk_v6_rcv_saddr, 115 inet_sk(sk)->inet_num); 116 117 udp_lib_rehash(sk, new_hash); 118 } 119 120 static int compute_score(struct sock *sk, struct net *net, 121 const struct in6_addr *saddr, __be16 sport, 122 const struct in6_addr *daddr, unsigned short hnum, 123 int dif, int sdif) 124 { 125 int bound_dev_if, score; 126 struct inet_sock *inet; 127 bool dev_match; 128 129 if (!net_eq(sock_net(sk), net) || 130 udp_sk(sk)->udp_port_hash != hnum || 131 sk->sk_family != PF_INET6) 132 return -1; 133 134 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 135 return -1; 136 137 score = 0; 138 inet = inet_sk(sk); 139 140 if (inet->inet_dport) { 141 if (inet->inet_dport != sport) 142 return -1; 143 score++; 144 } 145 146 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 147 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 148 return -1; 149 score++; 150 } 151 152 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 153 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif); 154 if (!dev_match) 155 return -1; 156 if (bound_dev_if) 157 score++; 158 159 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 160 score++; 161 162 return score; 163 } 164 165 /* called with rcu_read_lock() */ 166 static struct sock *udp6_lib_lookup2(struct net *net, 167 const struct in6_addr *saddr, __be16 sport, 168 const struct in6_addr *daddr, unsigned int hnum, 169 int dif, int sdif, struct udp_hslot *hslot2, 170 struct sk_buff *skb) 171 { 172 struct sock *sk, *result; 173 int score, badness; 174 175 result = NULL; 176 badness = -1; 177 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 178 score = compute_score(sk, net, saddr, sport, 179 daddr, hnum, dif, sdif); 180 if (score > badness) { 181 badness = score; 182 183 if (sk->sk_state == TCP_ESTABLISHED) { 184 result = sk; 185 continue; 186 } 187 188 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 189 saddr, sport, daddr, hnum, udp6_ehashfn); 190 if (!result) { 191 result = sk; 192 continue; 193 } 194 195 /* Fall back to scoring if group has connections */ 196 if (!reuseport_has_conns(sk)) 197 return result; 198 199 /* Reuseport logic returned an error, keep original score. */ 200 if (IS_ERR(result)) 201 continue; 202 203 badness = compute_score(sk, net, saddr, sport, 204 daddr, hnum, dif, sdif); 205 } 206 } 207 return result; 208 } 209 210 /* rcu_read_lock() must be held */ 211 struct sock *__udp6_lib_lookup(struct net *net, 212 const struct in6_addr *saddr, __be16 sport, 213 const struct in6_addr *daddr, __be16 dport, 214 int dif, int sdif, struct udp_table *udptable, 215 struct sk_buff *skb) 216 { 217 unsigned short hnum = ntohs(dport); 218 unsigned int hash2, slot2; 219 struct udp_hslot *hslot2; 220 struct sock *result, *sk; 221 222 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 223 slot2 = hash2 & udptable->mask; 224 hslot2 = &udptable->hash2[slot2]; 225 226 /* Lookup connected or non-wildcard sockets */ 227 result = udp6_lib_lookup2(net, saddr, sport, 228 daddr, hnum, dif, sdif, 229 hslot2, skb); 230 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 231 goto done; 232 233 /* Lookup redirect from BPF */ 234 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 235 udptable == net->ipv4.udp_table) { 236 sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 237 saddr, sport, daddr, hnum, dif, 238 udp6_ehashfn); 239 if (sk) { 240 result = sk; 241 goto done; 242 } 243 } 244 245 /* Got non-wildcard socket or error on first lookup */ 246 if (result) 247 goto done; 248 249 /* Lookup wildcard sockets */ 250 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 251 slot2 = hash2 & udptable->mask; 252 hslot2 = &udptable->hash2[slot2]; 253 254 result = udp6_lib_lookup2(net, saddr, sport, 255 &in6addr_any, hnum, dif, sdif, 256 hslot2, skb); 257 done: 258 if (IS_ERR(result)) 259 return NULL; 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(const struct sk_buff *skb, 276 __be16 sport, __be16 dport) 277 { 278 const struct ipv6hdr *iph = ipv6_hdr(skb); 279 struct net *net = dev_net(skb->dev); 280 int iif, sdif; 281 282 inet6_get_iif_sdif(skb, &iif, &sdif); 283 284 return __udp6_lib_lookup(net, &iph->saddr, sport, 285 &iph->daddr, dport, iif, 286 sdif, net->ipv4.udp_table, NULL); 287 } 288 289 /* Must be called under rcu_read_lock(). 290 * Does increment socket refcount. 291 */ 292 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 293 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 294 const struct in6_addr *daddr, __be16 dport, int dif) 295 { 296 struct sock *sk; 297 298 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 299 dif, 0, net->ipv4.udp_table, NULL); 300 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 301 sk = NULL; 302 return sk; 303 } 304 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 305 #endif 306 307 /* do not use the scratch area len for jumbogram: their length execeeds the 308 * scratch area space; note that the IP6CB flags is still in the first 309 * cacheline, so checking for jumbograms is cheap 310 */ 311 static int udp6_skb_len(struct sk_buff *skb) 312 { 313 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 314 } 315 316 /* 317 * This should be easy, if there is something there we 318 * return it, otherwise we block. 319 */ 320 321 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 322 int flags, int *addr_len) 323 { 324 struct ipv6_pinfo *np = inet6_sk(sk); 325 struct inet_sock *inet = inet_sk(sk); 326 struct sk_buff *skb; 327 unsigned int ulen, copied; 328 int off, err, peeking = flags & MSG_PEEK; 329 int is_udplite = IS_UDPLITE(sk); 330 struct udp_mib __percpu *mib; 331 bool checksum_valid = false; 332 int is_udp4; 333 334 if (flags & MSG_ERRQUEUE) 335 return ipv6_recv_error(sk, msg, len, addr_len); 336 337 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 338 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 339 340 try_again: 341 off = sk_peek_offset(sk, flags); 342 skb = __skb_recv_udp(sk, flags, &off, &err); 343 if (!skb) 344 return err; 345 346 ulen = udp6_skb_len(skb); 347 copied = len; 348 if (copied > ulen - off) 349 copied = ulen - off; 350 else if (copied < ulen) 351 msg->msg_flags |= MSG_TRUNC; 352 353 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 354 mib = __UDPX_MIB(sk, is_udp4); 355 356 /* 357 * If checksum is needed at all, try to do it while copying the 358 * data. If the data is truncated, or if we only want a partial 359 * coverage checksum (UDP-Lite), do it before the copy. 360 */ 361 362 if (copied < ulen || peeking || 363 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 364 checksum_valid = udp_skb_csum_unnecessary(skb) || 365 !__udp_lib_checksum_complete(skb); 366 if (!checksum_valid) 367 goto csum_copy_err; 368 } 369 370 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 371 if (udp_skb_is_linear(skb)) 372 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 373 else 374 err = skb_copy_datagram_msg(skb, off, msg, copied); 375 } else { 376 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 377 if (err == -EINVAL) 378 goto csum_copy_err; 379 } 380 if (unlikely(err)) { 381 if (!peeking) { 382 atomic_inc(&sk->sk_drops); 383 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 384 } 385 kfree_skb(skb); 386 return err; 387 } 388 if (!peeking) 389 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 390 391 sock_recv_cmsgs(msg, sk, skb); 392 393 /* Copy the address. */ 394 if (msg->msg_name) { 395 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 396 sin6->sin6_family = AF_INET6; 397 sin6->sin6_port = udp_hdr(skb)->source; 398 sin6->sin6_flowinfo = 0; 399 400 if (is_udp4) { 401 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 402 &sin6->sin6_addr); 403 sin6->sin6_scope_id = 0; 404 } else { 405 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 406 sin6->sin6_scope_id = 407 ipv6_iface_scope_id(&sin6->sin6_addr, 408 inet6_iif(skb)); 409 } 410 *addr_len = sizeof(*sin6); 411 412 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 413 (struct sockaddr *)sin6, 414 addr_len); 415 } 416 417 if (udp_test_bit(GRO_ENABLED, sk)) 418 udp_cmsg_recv(msg, sk, skb); 419 420 if (np->rxopt.all) 421 ip6_datagram_recv_common_ctl(sk, msg, skb); 422 423 if (is_udp4) { 424 if (inet_cmsg_flags(inet)) 425 ip_cmsg_recv_offset(msg, sk, skb, 426 sizeof(struct udphdr), off); 427 } else { 428 if (np->rxopt.all) 429 ip6_datagram_recv_specific_ctl(sk, msg, skb); 430 } 431 432 err = copied; 433 if (flags & MSG_TRUNC) 434 err = ulen; 435 436 skb_consume_udp(sk, skb, peeking ? -err : err); 437 return err; 438 439 csum_copy_err: 440 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 441 udp_skb_destructor)) { 442 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 443 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 444 } 445 kfree_skb(skb); 446 447 /* starting over for a new packet, but check if we need to yield */ 448 cond_resched(); 449 msg->msg_flags &= ~MSG_TRUNC; 450 goto try_again; 451 } 452 453 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 454 void udpv6_encap_enable(void) 455 { 456 static_branch_inc(&udpv6_encap_needed_key); 457 } 458 EXPORT_SYMBOL(udpv6_encap_enable); 459 460 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 461 * through error handlers in encapsulations looking for a match. 462 */ 463 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 464 struct inet6_skb_parm *opt, 465 u8 type, u8 code, int offset, __be32 info) 466 { 467 int i; 468 469 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 470 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 471 u8 type, u8 code, int offset, __be32 info); 472 const struct ip6_tnl_encap_ops *encap; 473 474 encap = rcu_dereference(ip6tun_encaps[i]); 475 if (!encap) 476 continue; 477 handler = encap->err_handler; 478 if (handler && !handler(skb, opt, type, code, offset, info)) 479 return 0; 480 } 481 482 return -ENOENT; 483 } 484 485 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 486 * reversing source and destination port: this will match tunnels that force the 487 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 488 * lwtunnels might actually break this assumption by being configured with 489 * different destination ports on endpoints, in this case we won't be able to 490 * trace ICMP messages back to them. 491 * 492 * If this doesn't match any socket, probe tunnels with arbitrary destination 493 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 494 * we've sent packets to won't necessarily match the local destination port. 495 * 496 * Then ask the tunnel implementation to match the error against a valid 497 * association. 498 * 499 * Return an error if we can't find a match, the socket if we need further 500 * processing, zero otherwise. 501 */ 502 static struct sock *__udp6_lib_err_encap(struct net *net, 503 const struct ipv6hdr *hdr, int offset, 504 struct udphdr *uh, 505 struct udp_table *udptable, 506 struct sock *sk, 507 struct sk_buff *skb, 508 struct inet6_skb_parm *opt, 509 u8 type, u8 code, __be32 info) 510 { 511 int (*lookup)(struct sock *sk, struct sk_buff *skb); 512 int network_offset, transport_offset; 513 struct udp_sock *up; 514 515 network_offset = skb_network_offset(skb); 516 transport_offset = skb_transport_offset(skb); 517 518 /* Network header needs to point to the outer IPv6 header inside ICMP */ 519 skb_reset_network_header(skb); 520 521 /* Transport header needs to point to the UDP header */ 522 skb_set_transport_header(skb, offset); 523 524 if (sk) { 525 up = udp_sk(sk); 526 527 lookup = READ_ONCE(up->encap_err_lookup); 528 if (lookup && lookup(sk, skb)) 529 sk = NULL; 530 531 goto out; 532 } 533 534 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 535 &hdr->saddr, uh->dest, 536 inet6_iif(skb), 0, udptable, skb); 537 if (sk) { 538 up = udp_sk(sk); 539 540 lookup = READ_ONCE(up->encap_err_lookup); 541 if (!lookup || lookup(sk, skb)) 542 sk = NULL; 543 } 544 545 out: 546 if (!sk) { 547 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 548 offset, info)); 549 } 550 551 skb_set_transport_header(skb, transport_offset); 552 skb_set_network_header(skb, network_offset); 553 554 return sk; 555 } 556 557 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 558 u8 type, u8 code, int offset, __be32 info, 559 struct udp_table *udptable) 560 { 561 struct ipv6_pinfo *np; 562 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 563 const struct in6_addr *saddr = &hdr->saddr; 564 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr; 565 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 566 bool tunnel = false; 567 struct sock *sk; 568 int harderr; 569 int err; 570 struct net *net = dev_net(skb->dev); 571 572 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 573 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 574 575 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) { 576 /* No socket for error: try tunnels before discarding */ 577 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 578 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 579 udptable, sk, skb, 580 opt, type, code, info); 581 if (!sk) 582 return 0; 583 } else 584 sk = ERR_PTR(-ENOENT); 585 586 if (IS_ERR(sk)) { 587 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 588 ICMP6_MIB_INERRORS); 589 return PTR_ERR(sk); 590 } 591 592 tunnel = true; 593 } 594 595 harderr = icmpv6_err_convert(type, code, &err); 596 np = inet6_sk(sk); 597 598 if (type == ICMPV6_PKT_TOOBIG) { 599 if (!ip6_sk_accept_pmtu(sk)) 600 goto out; 601 ip6_sk_update_pmtu(skb, sk, info); 602 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 603 harderr = 1; 604 } 605 if (type == NDISC_REDIRECT) { 606 if (tunnel) { 607 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 608 READ_ONCE(sk->sk_mark), sk->sk_uid); 609 } else { 610 ip6_sk_redirect(skb, sk); 611 } 612 goto out; 613 } 614 615 /* Tunnels don't have an application socket: don't pass errors back */ 616 if (tunnel) { 617 if (udp_sk(sk)->encap_err_rcv) 618 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, 619 ntohl(info), (u8 *)(uh+1)); 620 goto out; 621 } 622 623 if (!np->recverr) { 624 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 625 goto out; 626 } else { 627 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 628 } 629 630 sk->sk_err = err; 631 sk_error_report(sk); 632 out: 633 return 0; 634 } 635 636 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 637 { 638 int rc; 639 640 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 641 sock_rps_save_rxhash(sk, skb); 642 sk_mark_napi_id(sk, skb); 643 sk_incoming_cpu_update(sk); 644 } else { 645 sk_mark_napi_id_once(sk, skb); 646 } 647 648 rc = __udp_enqueue_schedule_skb(sk, skb); 649 if (rc < 0) { 650 int is_udplite = IS_UDPLITE(sk); 651 enum skb_drop_reason drop_reason; 652 653 /* Note that an ENOMEM error is charged twice */ 654 if (rc == -ENOMEM) { 655 UDP6_INC_STATS(sock_net(sk), 656 UDP_MIB_RCVBUFERRORS, is_udplite); 657 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 658 } else { 659 UDP6_INC_STATS(sock_net(sk), 660 UDP_MIB_MEMERRORS, is_udplite); 661 drop_reason = SKB_DROP_REASON_PROTO_MEM; 662 } 663 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 664 kfree_skb_reason(skb, drop_reason); 665 trace_udp_fail_queue_rcv_skb(rc, sk); 666 return -1; 667 } 668 669 return 0; 670 } 671 672 static __inline__ int udpv6_err(struct sk_buff *skb, 673 struct inet6_skb_parm *opt, u8 type, 674 u8 code, int offset, __be32 info) 675 { 676 return __udp6_lib_err(skb, opt, type, code, offset, info, 677 dev_net(skb->dev)->ipv4.udp_table); 678 } 679 680 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 681 { 682 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 683 struct udp_sock *up = udp_sk(sk); 684 int is_udplite = IS_UDPLITE(sk); 685 686 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 687 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 688 goto drop; 689 } 690 nf_reset_ct(skb); 691 692 if (static_branch_unlikely(&udpv6_encap_needed_key) && 693 READ_ONCE(up->encap_type)) { 694 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 695 696 /* 697 * This is an encapsulation socket so pass the skb to 698 * the socket's udp_encap_rcv() hook. Otherwise, just 699 * fall through and pass this up the UDP socket. 700 * up->encap_rcv() returns the following value: 701 * =0 if skb was successfully passed to the encap 702 * handler or was discarded by it. 703 * >0 if skb should be passed on to UDP. 704 * <0 if skb should be resubmitted as proto -N 705 */ 706 707 /* if we're overly short, let UDP handle it */ 708 encap_rcv = READ_ONCE(up->encap_rcv); 709 if (encap_rcv) { 710 int ret; 711 712 /* Verify checksum before giving to encap */ 713 if (udp_lib_checksum_complete(skb)) 714 goto csum_error; 715 716 ret = encap_rcv(sk, skb); 717 if (ret <= 0) { 718 __UDP6_INC_STATS(sock_net(sk), 719 UDP_MIB_INDATAGRAMS, 720 is_udplite); 721 return -ret; 722 } 723 } 724 725 /* FALLTHROUGH -- it's a UDP Packet */ 726 } 727 728 /* 729 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 730 */ 731 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) { 732 u16 pcrlen = READ_ONCE(up->pcrlen); 733 734 if (pcrlen == 0) { /* full coverage was set */ 735 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 736 UDP_SKB_CB(skb)->cscov, skb->len); 737 goto drop; 738 } 739 if (UDP_SKB_CB(skb)->cscov < pcrlen) { 740 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 741 UDP_SKB_CB(skb)->cscov, pcrlen); 742 goto drop; 743 } 744 } 745 746 prefetch(&sk->sk_rmem_alloc); 747 if (rcu_access_pointer(sk->sk_filter) && 748 udp_lib_checksum_complete(skb)) 749 goto csum_error; 750 751 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 752 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 753 goto drop; 754 } 755 756 udp_csum_pull_header(skb); 757 758 skb_dst_drop(skb); 759 760 return __udpv6_queue_rcv_skb(sk, skb); 761 762 csum_error: 763 drop_reason = SKB_DROP_REASON_UDP_CSUM; 764 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 765 drop: 766 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 767 atomic_inc(&sk->sk_drops); 768 kfree_skb_reason(skb, drop_reason); 769 return -1; 770 } 771 772 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 773 { 774 struct sk_buff *next, *segs; 775 int ret; 776 777 if (likely(!udp_unexpected_gso(sk, skb))) 778 return udpv6_queue_rcv_one_skb(sk, skb); 779 780 __skb_push(skb, -skb_mac_offset(skb)); 781 segs = udp_rcv_segment(sk, skb, false); 782 skb_list_walk_safe(segs, skb, next) { 783 __skb_pull(skb, skb_transport_offset(skb)); 784 785 udp_post_segment_fix_csum(skb); 786 ret = udpv6_queue_rcv_one_skb(sk, skb); 787 if (ret > 0) 788 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 789 true); 790 } 791 return 0; 792 } 793 794 static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk, 795 __be16 loc_port, const struct in6_addr *loc_addr, 796 __be16 rmt_port, const struct in6_addr *rmt_addr, 797 int dif, int sdif, unsigned short hnum) 798 { 799 const struct inet_sock *inet = inet_sk(sk); 800 801 if (!net_eq(sock_net(sk), net)) 802 return false; 803 804 if (udp_sk(sk)->udp_port_hash != hnum || 805 sk->sk_family != PF_INET6 || 806 (inet->inet_dport && inet->inet_dport != rmt_port) || 807 (!ipv6_addr_any(&sk->sk_v6_daddr) && 808 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 809 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) || 810 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 811 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 812 return false; 813 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 814 return false; 815 return true; 816 } 817 818 static void udp6_csum_zero_error(struct sk_buff *skb) 819 { 820 /* RFC 2460 section 8.1 says that we SHOULD log 821 * this error. Well, it is reasonable. 822 */ 823 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 824 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 825 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 826 } 827 828 /* 829 * Note: called only from the BH handler context, 830 * so we don't need to lock the hashes. 831 */ 832 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 833 const struct in6_addr *saddr, const struct in6_addr *daddr, 834 struct udp_table *udptable, int proto) 835 { 836 struct sock *sk, *first = NULL; 837 const struct udphdr *uh = udp_hdr(skb); 838 unsigned short hnum = ntohs(uh->dest); 839 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 840 unsigned int offset = offsetof(typeof(*sk), sk_node); 841 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 842 int dif = inet6_iif(skb); 843 int sdif = inet6_sdif(skb); 844 struct hlist_node *node; 845 struct sk_buff *nskb; 846 847 if (use_hash2) { 848 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 849 udptable->mask; 850 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 851 start_lookup: 852 hslot = &udptable->hash2[hash2]; 853 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 854 } 855 856 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 857 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 858 uh->source, saddr, dif, sdif, 859 hnum)) 860 continue; 861 /* If zero checksum and no_check is not on for 862 * the socket then skip it. 863 */ 864 if (!uh->check && !udp_get_no_check6_rx(sk)) 865 continue; 866 if (!first) { 867 first = sk; 868 continue; 869 } 870 nskb = skb_clone(skb, GFP_ATOMIC); 871 if (unlikely(!nskb)) { 872 atomic_inc(&sk->sk_drops); 873 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 874 IS_UDPLITE(sk)); 875 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 876 IS_UDPLITE(sk)); 877 continue; 878 } 879 880 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 881 consume_skb(nskb); 882 } 883 884 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 885 if (use_hash2 && hash2 != hash2_any) { 886 hash2 = hash2_any; 887 goto start_lookup; 888 } 889 890 if (first) { 891 if (udpv6_queue_rcv_skb(first, skb) > 0) 892 consume_skb(skb); 893 } else { 894 kfree_skb(skb); 895 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 896 proto == IPPROTO_UDPLITE); 897 } 898 return 0; 899 } 900 901 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 902 { 903 if (udp_sk_rx_dst_set(sk, dst)) { 904 const struct rt6_info *rt = (const struct rt6_info *)dst; 905 906 sk->sk_rx_dst_cookie = rt6_get_cookie(rt); 907 } 908 } 909 910 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 911 * return code conversion for ip layer consumption 912 */ 913 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 914 struct udphdr *uh) 915 { 916 int ret; 917 918 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 919 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 920 921 ret = udpv6_queue_rcv_skb(sk, skb); 922 923 /* a return value > 0 means to resubmit the input */ 924 if (ret > 0) 925 return ret; 926 return 0; 927 } 928 929 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 930 int proto) 931 { 932 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 933 const struct in6_addr *saddr, *daddr; 934 struct net *net = dev_net(skb->dev); 935 struct udphdr *uh; 936 struct sock *sk; 937 bool refcounted; 938 u32 ulen = 0; 939 940 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 941 goto discard; 942 943 saddr = &ipv6_hdr(skb)->saddr; 944 daddr = &ipv6_hdr(skb)->daddr; 945 uh = udp_hdr(skb); 946 947 ulen = ntohs(uh->len); 948 if (ulen > skb->len) 949 goto short_packet; 950 951 if (proto == IPPROTO_UDP) { 952 /* UDP validates ulen. */ 953 954 /* Check for jumbo payload */ 955 if (ulen == 0) 956 ulen = skb->len; 957 958 if (ulen < sizeof(*uh)) 959 goto short_packet; 960 961 if (ulen < skb->len) { 962 if (pskb_trim_rcsum(skb, ulen)) 963 goto short_packet; 964 saddr = &ipv6_hdr(skb)->saddr; 965 daddr = &ipv6_hdr(skb)->daddr; 966 uh = udp_hdr(skb); 967 } 968 } 969 970 if (udp6_csum_init(skb, uh, proto)) 971 goto csum_error; 972 973 /* Check if the socket is already available, e.g. due to early demux */ 974 sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 975 &refcounted, udp6_ehashfn); 976 if (IS_ERR(sk)) 977 goto no_sk; 978 979 if (sk) { 980 struct dst_entry *dst = skb_dst(skb); 981 int ret; 982 983 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 984 udp6_sk_rx_dst_set(sk, dst); 985 986 if (!uh->check && !udp_get_no_check6_rx(sk)) { 987 if (refcounted) 988 sock_put(sk); 989 goto report_csum_error; 990 } 991 992 ret = udp6_unicast_rcv_skb(sk, skb, uh); 993 if (refcounted) 994 sock_put(sk); 995 return ret; 996 } 997 998 /* 999 * Multicast receive code 1000 */ 1001 if (ipv6_addr_is_multicast(daddr)) 1002 return __udp6_lib_mcast_deliver(net, skb, 1003 saddr, daddr, udptable, proto); 1004 1005 /* Unicast */ 1006 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1007 if (sk) { 1008 if (!uh->check && !udp_get_no_check6_rx(sk)) 1009 goto report_csum_error; 1010 return udp6_unicast_rcv_skb(sk, skb, uh); 1011 } 1012 no_sk: 1013 reason = SKB_DROP_REASON_NO_SOCKET; 1014 1015 if (!uh->check) 1016 goto report_csum_error; 1017 1018 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1019 goto discard; 1020 nf_reset_ct(skb); 1021 1022 if (udp_lib_checksum_complete(skb)) 1023 goto csum_error; 1024 1025 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1026 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 1027 1028 kfree_skb_reason(skb, reason); 1029 return 0; 1030 1031 short_packet: 1032 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1033 reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1034 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 1035 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1036 saddr, ntohs(uh->source), 1037 ulen, skb->len, 1038 daddr, ntohs(uh->dest)); 1039 goto discard; 1040 1041 report_csum_error: 1042 udp6_csum_zero_error(skb); 1043 csum_error: 1044 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1045 reason = SKB_DROP_REASON_UDP_CSUM; 1046 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1047 discard: 1048 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1049 kfree_skb_reason(skb, reason); 1050 return 0; 1051 } 1052 1053 1054 static struct sock *__udp6_lib_demux_lookup(struct net *net, 1055 __be16 loc_port, const struct in6_addr *loc_addr, 1056 __be16 rmt_port, const struct in6_addr *rmt_addr, 1057 int dif, int sdif) 1058 { 1059 struct udp_table *udptable = net->ipv4.udp_table; 1060 unsigned short hnum = ntohs(loc_port); 1061 unsigned int hash2, slot2; 1062 struct udp_hslot *hslot2; 1063 __portpair ports; 1064 struct sock *sk; 1065 1066 hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1067 slot2 = hash2 & udptable->mask; 1068 hslot2 = &udptable->hash2[slot2]; 1069 ports = INET_COMBINED_PORTS(rmt_port, hnum); 1070 1071 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1072 if (sk->sk_state == TCP_ESTABLISHED && 1073 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif)) 1074 return sk; 1075 /* Only check first socket in chain */ 1076 break; 1077 } 1078 return NULL; 1079 } 1080 1081 void udp_v6_early_demux(struct sk_buff *skb) 1082 { 1083 struct net *net = dev_net(skb->dev); 1084 const struct udphdr *uh; 1085 struct sock *sk; 1086 struct dst_entry *dst; 1087 int dif = skb->dev->ifindex; 1088 int sdif = inet6_sdif(skb); 1089 1090 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1091 sizeof(struct udphdr))) 1092 return; 1093 1094 uh = udp_hdr(skb); 1095 1096 if (skb->pkt_type == PACKET_HOST) 1097 sk = __udp6_lib_demux_lookup(net, uh->dest, 1098 &ipv6_hdr(skb)->daddr, 1099 uh->source, &ipv6_hdr(skb)->saddr, 1100 dif, sdif); 1101 else 1102 return; 1103 1104 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1105 return; 1106 1107 skb->sk = sk; 1108 skb->destructor = sock_efree; 1109 dst = rcu_dereference(sk->sk_rx_dst); 1110 1111 if (dst) 1112 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1113 if (dst) { 1114 /* set noref for now. 1115 * any place which wants to hold dst has to call 1116 * dst_hold_safe() 1117 */ 1118 skb_dst_set_noref(skb, dst); 1119 } 1120 } 1121 1122 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1123 { 1124 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 1125 } 1126 1127 /* 1128 * Throw away all pending data and cancel the corking. Socket is locked. 1129 */ 1130 static void udp_v6_flush_pending_frames(struct sock *sk) 1131 { 1132 struct udp_sock *up = udp_sk(sk); 1133 1134 if (up->pending == AF_INET) 1135 udp_flush_pending_frames(sk); 1136 else if (up->pending) { 1137 up->len = 0; 1138 WRITE_ONCE(up->pending, 0); 1139 ip6_flush_pending_frames(sk); 1140 } 1141 } 1142 1143 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1144 int addr_len) 1145 { 1146 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1147 return -EINVAL; 1148 /* The following checks are replicated from __ip6_datagram_connect() 1149 * and intended to prevent BPF program called below from accessing 1150 * bytes that are out of the bound specified by user in addr_len. 1151 */ 1152 if (uaddr->sa_family == AF_INET) { 1153 if (ipv6_only_sock(sk)) 1154 return -EAFNOSUPPORT; 1155 return udp_pre_connect(sk, uaddr, addr_len); 1156 } 1157 1158 if (addr_len < SIN6_LEN_RFC2133) 1159 return -EINVAL; 1160 1161 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len); 1162 } 1163 1164 /** 1165 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1166 * @sk: socket we are sending on 1167 * @skb: sk_buff containing the filled-in UDP header 1168 * (checksum field must be zeroed out) 1169 * @saddr: source address 1170 * @daddr: destination address 1171 * @len: length of packet 1172 */ 1173 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1174 const struct in6_addr *saddr, 1175 const struct in6_addr *daddr, int len) 1176 { 1177 unsigned int offset; 1178 struct udphdr *uh = udp_hdr(skb); 1179 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1180 __wsum csum = 0; 1181 1182 if (!frags) { 1183 /* Only one fragment on the socket. */ 1184 skb->csum_start = skb_transport_header(skb) - skb->head; 1185 skb->csum_offset = offsetof(struct udphdr, check); 1186 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1187 } else { 1188 /* 1189 * HW-checksum won't work as there are two or more 1190 * fragments on the socket so that all csums of sk_buffs 1191 * should be together 1192 */ 1193 offset = skb_transport_offset(skb); 1194 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1195 csum = skb->csum; 1196 1197 skb->ip_summed = CHECKSUM_NONE; 1198 1199 do { 1200 csum = csum_add(csum, frags->csum); 1201 } while ((frags = frags->next)); 1202 1203 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1204 csum); 1205 if (uh->check == 0) 1206 uh->check = CSUM_MANGLED_0; 1207 } 1208 } 1209 1210 /* 1211 * Sending 1212 */ 1213 1214 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1215 struct inet_cork *cork) 1216 { 1217 struct sock *sk = skb->sk; 1218 struct udphdr *uh; 1219 int err = 0; 1220 int is_udplite = IS_UDPLITE(sk); 1221 __wsum csum = 0; 1222 int offset = skb_transport_offset(skb); 1223 int len = skb->len - offset; 1224 int datalen = len - sizeof(*uh); 1225 1226 /* 1227 * Create a UDP header 1228 */ 1229 uh = udp_hdr(skb); 1230 uh->source = fl6->fl6_sport; 1231 uh->dest = fl6->fl6_dport; 1232 uh->len = htons(len); 1233 uh->check = 0; 1234 1235 if (cork->gso_size) { 1236 const int hlen = skb_network_header_len(skb) + 1237 sizeof(struct udphdr); 1238 1239 if (hlen + cork->gso_size > cork->fragsize) { 1240 kfree_skb(skb); 1241 return -EINVAL; 1242 } 1243 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1244 kfree_skb(skb); 1245 return -EINVAL; 1246 } 1247 if (udp_get_no_check6_tx(sk)) { 1248 kfree_skb(skb); 1249 return -EINVAL; 1250 } 1251 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1252 dst_xfrm(skb_dst(skb))) { 1253 kfree_skb(skb); 1254 return -EIO; 1255 } 1256 1257 if (datalen > cork->gso_size) { 1258 skb_shinfo(skb)->gso_size = cork->gso_size; 1259 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1260 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1261 cork->gso_size); 1262 } 1263 goto csum_partial; 1264 } 1265 1266 if (is_udplite) 1267 csum = udplite_csum(skb); 1268 else if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */ 1269 skb->ip_summed = CHECKSUM_NONE; 1270 goto send; 1271 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1272 csum_partial: 1273 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1274 goto send; 1275 } else 1276 csum = udp_csum(skb); 1277 1278 /* add protocol-dependent pseudo-header */ 1279 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1280 len, fl6->flowi6_proto, csum); 1281 if (uh->check == 0) 1282 uh->check = CSUM_MANGLED_0; 1283 1284 send: 1285 err = ip6_send_skb(skb); 1286 if (err) { 1287 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1288 UDP6_INC_STATS(sock_net(sk), 1289 UDP_MIB_SNDBUFERRORS, is_udplite); 1290 err = 0; 1291 } 1292 } else { 1293 UDP6_INC_STATS(sock_net(sk), 1294 UDP_MIB_OUTDATAGRAMS, is_udplite); 1295 } 1296 return err; 1297 } 1298 1299 static int udp_v6_push_pending_frames(struct sock *sk) 1300 { 1301 struct sk_buff *skb; 1302 struct udp_sock *up = udp_sk(sk); 1303 int err = 0; 1304 1305 if (up->pending == AF_INET) 1306 return udp_push_pending_frames(sk); 1307 1308 skb = ip6_finish_skb(sk); 1309 if (!skb) 1310 goto out; 1311 1312 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6, 1313 &inet_sk(sk)->cork.base); 1314 out: 1315 up->len = 0; 1316 WRITE_ONCE(up->pending, 0); 1317 return err; 1318 } 1319 1320 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1321 { 1322 struct ipv6_txoptions opt_space; 1323 struct udp_sock *up = udp_sk(sk); 1324 struct inet_sock *inet = inet_sk(sk); 1325 struct ipv6_pinfo *np = inet6_sk(sk); 1326 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1327 struct in6_addr *daddr, *final_p, final; 1328 struct ipv6_txoptions *opt = NULL; 1329 struct ipv6_txoptions *opt_to_free = NULL; 1330 struct ip6_flowlabel *flowlabel = NULL; 1331 struct inet_cork_full cork; 1332 struct flowi6 *fl6 = &cork.fl.u.ip6; 1333 struct dst_entry *dst; 1334 struct ipcm6_cookie ipc6; 1335 int addr_len = msg->msg_namelen; 1336 bool connected = false; 1337 int ulen = len; 1338 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE; 1339 int err; 1340 int is_udplite = IS_UDPLITE(sk); 1341 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1342 1343 ipcm6_init(&ipc6); 1344 ipc6.gso_size = READ_ONCE(up->gso_size); 1345 ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags); 1346 ipc6.sockc.mark = READ_ONCE(sk->sk_mark); 1347 1348 /* destination address check */ 1349 if (sin6) { 1350 if (addr_len < offsetof(struct sockaddr, sa_data)) 1351 return -EINVAL; 1352 1353 switch (sin6->sin6_family) { 1354 case AF_INET6: 1355 if (addr_len < SIN6_LEN_RFC2133) 1356 return -EINVAL; 1357 daddr = &sin6->sin6_addr; 1358 if (ipv6_addr_any(daddr) && 1359 ipv6_addr_v4mapped(&np->saddr)) 1360 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1361 daddr); 1362 break; 1363 case AF_INET: 1364 goto do_udp_sendmsg; 1365 case AF_UNSPEC: 1366 msg->msg_name = sin6 = NULL; 1367 msg->msg_namelen = addr_len = 0; 1368 daddr = NULL; 1369 break; 1370 default: 1371 return -EINVAL; 1372 } 1373 } else if (!READ_ONCE(up->pending)) { 1374 if (sk->sk_state != TCP_ESTABLISHED) 1375 return -EDESTADDRREQ; 1376 daddr = &sk->sk_v6_daddr; 1377 } else 1378 daddr = NULL; 1379 1380 if (daddr) { 1381 if (ipv6_addr_v4mapped(daddr)) { 1382 struct sockaddr_in sin; 1383 sin.sin_family = AF_INET; 1384 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1385 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1386 msg->msg_name = &sin; 1387 msg->msg_namelen = sizeof(sin); 1388 do_udp_sendmsg: 1389 err = ipv6_only_sock(sk) ? 1390 -ENETUNREACH : udp_sendmsg(sk, msg, len); 1391 msg->msg_name = sin6; 1392 msg->msg_namelen = addr_len; 1393 return err; 1394 } 1395 } 1396 1397 /* Rough check on arithmetic overflow, 1398 better check is made in ip6_append_data(). 1399 */ 1400 if (len > INT_MAX - sizeof(struct udphdr)) 1401 return -EMSGSIZE; 1402 1403 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1404 if (READ_ONCE(up->pending)) { 1405 if (READ_ONCE(up->pending) == AF_INET) 1406 return udp_sendmsg(sk, msg, len); 1407 /* 1408 * There are pending frames. 1409 * The socket lock must be held while it's corked. 1410 */ 1411 lock_sock(sk); 1412 if (likely(up->pending)) { 1413 if (unlikely(up->pending != AF_INET6)) { 1414 release_sock(sk); 1415 return -EAFNOSUPPORT; 1416 } 1417 dst = NULL; 1418 goto do_append_data; 1419 } 1420 release_sock(sk); 1421 } 1422 ulen += sizeof(struct udphdr); 1423 1424 memset(fl6, 0, sizeof(*fl6)); 1425 1426 if (sin6) { 1427 if (sin6->sin6_port == 0) 1428 return -EINVAL; 1429 1430 fl6->fl6_dport = sin6->sin6_port; 1431 daddr = &sin6->sin6_addr; 1432 1433 if (np->sndflow) { 1434 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1435 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) { 1436 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1437 if (IS_ERR(flowlabel)) 1438 return -EINVAL; 1439 } 1440 } 1441 1442 /* 1443 * Otherwise it will be difficult to maintain 1444 * sk->sk_dst_cache. 1445 */ 1446 if (sk->sk_state == TCP_ESTABLISHED && 1447 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1448 daddr = &sk->sk_v6_daddr; 1449 1450 if (addr_len >= sizeof(struct sockaddr_in6) && 1451 sin6->sin6_scope_id && 1452 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1453 fl6->flowi6_oif = sin6->sin6_scope_id; 1454 } else { 1455 if (sk->sk_state != TCP_ESTABLISHED) 1456 return -EDESTADDRREQ; 1457 1458 fl6->fl6_dport = inet->inet_dport; 1459 daddr = &sk->sk_v6_daddr; 1460 fl6->flowlabel = np->flow_label; 1461 connected = true; 1462 } 1463 1464 if (!fl6->flowi6_oif) 1465 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if); 1466 1467 if (!fl6->flowi6_oif) 1468 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1469 1470 fl6->flowi6_uid = sk->sk_uid; 1471 1472 if (msg->msg_controllen) { 1473 opt = &opt_space; 1474 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1475 opt->tot_len = sizeof(*opt); 1476 ipc6.opt = opt; 1477 1478 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1479 if (err > 0) 1480 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6, 1481 &ipc6); 1482 if (err < 0) { 1483 fl6_sock_release(flowlabel); 1484 return err; 1485 } 1486 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1487 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1488 if (IS_ERR(flowlabel)) 1489 return -EINVAL; 1490 } 1491 if (!(opt->opt_nflen|opt->opt_flen)) 1492 opt = NULL; 1493 connected = false; 1494 } 1495 if (!opt) { 1496 opt = txopt_get(np); 1497 opt_to_free = opt; 1498 } 1499 if (flowlabel) 1500 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1501 opt = ipv6_fixup_options(&opt_space, opt); 1502 ipc6.opt = opt; 1503 1504 fl6->flowi6_proto = sk->sk_protocol; 1505 fl6->flowi6_mark = ipc6.sockc.mark; 1506 fl6->daddr = *daddr; 1507 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr)) 1508 fl6->saddr = np->saddr; 1509 fl6->fl6_sport = inet->inet_sport; 1510 1511 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1512 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1513 (struct sockaddr *)sin6, 1514 &addr_len, 1515 &fl6->saddr); 1516 if (err) 1517 goto out_no_dst; 1518 if (sin6) { 1519 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1520 /* BPF program rewrote IPv6-only by IPv4-mapped 1521 * IPv6. It's currently unsupported. 1522 */ 1523 err = -ENOTSUPP; 1524 goto out_no_dst; 1525 } 1526 if (sin6->sin6_port == 0) { 1527 /* BPF program set invalid port. Reject it. */ 1528 err = -EINVAL; 1529 goto out_no_dst; 1530 } 1531 fl6->fl6_dport = sin6->sin6_port; 1532 fl6->daddr = sin6->sin6_addr; 1533 } 1534 } 1535 1536 if (ipv6_addr_any(&fl6->daddr)) 1537 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1538 1539 final_p = fl6_update_dst(fl6, opt, &final); 1540 if (final_p) 1541 connected = false; 1542 1543 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) { 1544 fl6->flowi6_oif = np->mcast_oif; 1545 connected = false; 1546 } else if (!fl6->flowi6_oif) 1547 fl6->flowi6_oif = np->ucast_oif; 1548 1549 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6)); 1550 1551 if (ipc6.tclass < 0) 1552 ipc6.tclass = np->tclass; 1553 1554 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel); 1555 1556 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected); 1557 if (IS_ERR(dst)) { 1558 err = PTR_ERR(dst); 1559 dst = NULL; 1560 goto out; 1561 } 1562 1563 if (ipc6.hlimit < 0) 1564 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst); 1565 1566 if (msg->msg_flags&MSG_CONFIRM) 1567 goto do_confirm; 1568 back_from_confirm: 1569 1570 /* Lockless fast path for the non-corking case */ 1571 if (!corkreq) { 1572 struct sk_buff *skb; 1573 1574 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1575 sizeof(struct udphdr), &ipc6, 1576 (struct rt6_info *)dst, 1577 msg->msg_flags, &cork); 1578 err = PTR_ERR(skb); 1579 if (!IS_ERR_OR_NULL(skb)) 1580 err = udp_v6_send_skb(skb, fl6, &cork.base); 1581 /* ip6_make_skb steals dst reference */ 1582 goto out_no_dst; 1583 } 1584 1585 lock_sock(sk); 1586 if (unlikely(up->pending)) { 1587 /* The socket is already corked while preparing it. */ 1588 /* ... which is an evident application bug. --ANK */ 1589 release_sock(sk); 1590 1591 net_dbg_ratelimited("udp cork app bug 2\n"); 1592 err = -EINVAL; 1593 goto out; 1594 } 1595 1596 WRITE_ONCE(up->pending, AF_INET6); 1597 1598 do_append_data: 1599 if (ipc6.dontfrag < 0) 1600 ipc6.dontfrag = np->dontfrag; 1601 up->len += ulen; 1602 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1603 &ipc6, fl6, (struct rt6_info *)dst, 1604 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1605 if (err) 1606 udp_v6_flush_pending_frames(sk); 1607 else if (!corkreq) 1608 err = udp_v6_push_pending_frames(sk); 1609 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1610 WRITE_ONCE(up->pending, 0); 1611 1612 if (err > 0) 1613 err = np->recverr ? net_xmit_errno(err) : 0; 1614 release_sock(sk); 1615 1616 out: 1617 dst_release(dst); 1618 out_no_dst: 1619 fl6_sock_release(flowlabel); 1620 txopt_put(opt_to_free); 1621 if (!err) 1622 return len; 1623 /* 1624 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1625 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1626 * we don't have a good statistic (IpOutDiscards but it can be too many 1627 * things). We could add another new stat but at least for now that 1628 * seems like overkill. 1629 */ 1630 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1631 UDP6_INC_STATS(sock_net(sk), 1632 UDP_MIB_SNDBUFERRORS, is_udplite); 1633 } 1634 return err; 1635 1636 do_confirm: 1637 if (msg->msg_flags & MSG_PROBE) 1638 dst_confirm_neigh(dst, &fl6->daddr); 1639 if (!(msg->msg_flags&MSG_PROBE) || len) 1640 goto back_from_confirm; 1641 err = 0; 1642 goto out; 1643 } 1644 EXPORT_SYMBOL(udpv6_sendmsg); 1645 1646 static void udpv6_splice_eof(struct socket *sock) 1647 { 1648 struct sock *sk = sock->sk; 1649 struct udp_sock *up = udp_sk(sk); 1650 1651 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk)) 1652 return; 1653 1654 lock_sock(sk); 1655 if (up->pending && !udp_test_bit(CORK, sk)) 1656 udp_v6_push_pending_frames(sk); 1657 release_sock(sk); 1658 } 1659 1660 void udpv6_destroy_sock(struct sock *sk) 1661 { 1662 struct udp_sock *up = udp_sk(sk); 1663 lock_sock(sk); 1664 1665 /* protects from races with udp_abort() */ 1666 sock_set_flag(sk, SOCK_DEAD); 1667 udp_v6_flush_pending_frames(sk); 1668 release_sock(sk); 1669 1670 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1671 if (up->encap_type) { 1672 void (*encap_destroy)(struct sock *sk); 1673 encap_destroy = READ_ONCE(up->encap_destroy); 1674 if (encap_destroy) 1675 encap_destroy(sk); 1676 } 1677 if (udp_test_bit(ENCAP_ENABLED, sk)) { 1678 static_branch_dec(&udpv6_encap_needed_key); 1679 udp_encap_disable(); 1680 } 1681 } 1682 } 1683 1684 /* 1685 * Socket option code for UDP 1686 */ 1687 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1688 unsigned int optlen) 1689 { 1690 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 1691 return udp_lib_setsockopt(sk, level, optname, 1692 optval, optlen, 1693 udp_v6_push_pending_frames); 1694 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1695 } 1696 1697 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1698 char __user *optval, int __user *optlen) 1699 { 1700 if (level == SOL_UDP || level == SOL_UDPLITE) 1701 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1702 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1703 } 1704 1705 static const struct inet6_protocol udpv6_protocol = { 1706 .handler = udpv6_rcv, 1707 .err_handler = udpv6_err, 1708 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1709 }; 1710 1711 /* ------------------------------------------------------------------------ */ 1712 #ifdef CONFIG_PROC_FS 1713 int udp6_seq_show(struct seq_file *seq, void *v) 1714 { 1715 if (v == SEQ_START_TOKEN) { 1716 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1717 } else { 1718 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1719 const struct inet_sock *inet = inet_sk((const struct sock *)v); 1720 __u16 srcp = ntohs(inet->inet_sport); 1721 __u16 destp = ntohs(inet->inet_dport); 1722 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1723 udp_rqueue_get(v), bucket); 1724 } 1725 return 0; 1726 } 1727 1728 const struct seq_operations udp6_seq_ops = { 1729 .start = udp_seq_start, 1730 .next = udp_seq_next, 1731 .stop = udp_seq_stop, 1732 .show = udp6_seq_show, 1733 }; 1734 EXPORT_SYMBOL(udp6_seq_ops); 1735 1736 static struct udp_seq_afinfo udp6_seq_afinfo = { 1737 .family = AF_INET6, 1738 .udp_table = NULL, 1739 }; 1740 1741 int __net_init udp6_proc_init(struct net *net) 1742 { 1743 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1744 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1745 return -ENOMEM; 1746 return 0; 1747 } 1748 1749 void udp6_proc_exit(struct net *net) 1750 { 1751 remove_proc_entry("udp6", net->proc_net); 1752 } 1753 #endif /* CONFIG_PROC_FS */ 1754 1755 /* ------------------------------------------------------------------------ */ 1756 1757 struct proto udpv6_prot = { 1758 .name = "UDPv6", 1759 .owner = THIS_MODULE, 1760 .close = udp_lib_close, 1761 .pre_connect = udpv6_pre_connect, 1762 .connect = ip6_datagram_connect, 1763 .disconnect = udp_disconnect, 1764 .ioctl = udp_ioctl, 1765 .init = udpv6_init_sock, 1766 .destroy = udpv6_destroy_sock, 1767 .setsockopt = udpv6_setsockopt, 1768 .getsockopt = udpv6_getsockopt, 1769 .sendmsg = udpv6_sendmsg, 1770 .recvmsg = udpv6_recvmsg, 1771 .splice_eof = udpv6_splice_eof, 1772 .release_cb = ip6_datagram_release_cb, 1773 .hash = udp_lib_hash, 1774 .unhash = udp_lib_unhash, 1775 .rehash = udp_v6_rehash, 1776 .get_port = udp_v6_get_port, 1777 .put_port = udp_lib_unhash, 1778 #ifdef CONFIG_BPF_SYSCALL 1779 .psock_update_sk_prot = udp_bpf_update_proto, 1780 #endif 1781 1782 .memory_allocated = &udp_memory_allocated, 1783 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 1784 1785 .sysctl_mem = sysctl_udp_mem, 1786 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1787 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1788 .obj_size = sizeof(struct udp6_sock), 1789 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6), 1790 .h.udp_table = NULL, 1791 .diag_destroy = udp_abort, 1792 }; 1793 1794 static struct inet_protosw udpv6_protosw = { 1795 .type = SOCK_DGRAM, 1796 .protocol = IPPROTO_UDP, 1797 .prot = &udpv6_prot, 1798 .ops = &inet6_dgram_ops, 1799 .flags = INET_PROTOSW_PERMANENT, 1800 }; 1801 1802 int __init udpv6_init(void) 1803 { 1804 int ret; 1805 1806 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1807 if (ret) 1808 goto out; 1809 1810 ret = inet6_register_protosw(&udpv6_protosw); 1811 if (ret) 1812 goto out_udpv6_protocol; 1813 out: 1814 return ret; 1815 1816 out_udpv6_protocol: 1817 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1818 goto out; 1819 } 1820 1821 void udpv6_exit(void) 1822 { 1823 inet6_unregister_protosw(&udpv6_protosw); 1824 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1825 } 1826