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 } 415 416 if (udp_sk(sk)->gro_enabled) 417 udp_cmsg_recv(msg, sk, skb); 418 419 if (np->rxopt.all) 420 ip6_datagram_recv_common_ctl(sk, msg, skb); 421 422 if (is_udp4) { 423 if (inet_cmsg_flags(inet)) 424 ip_cmsg_recv_offset(msg, sk, skb, 425 sizeof(struct udphdr), off); 426 } else { 427 if (np->rxopt.all) 428 ip6_datagram_recv_specific_ctl(sk, msg, skb); 429 } 430 431 err = copied; 432 if (flags & MSG_TRUNC) 433 err = ulen; 434 435 skb_consume_udp(sk, skb, peeking ? -err : err); 436 return err; 437 438 csum_copy_err: 439 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 440 udp_skb_destructor)) { 441 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 442 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 443 } 444 kfree_skb(skb); 445 446 /* starting over for a new packet, but check if we need to yield */ 447 cond_resched(); 448 msg->msg_flags &= ~MSG_TRUNC; 449 goto try_again; 450 } 451 452 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 453 void udpv6_encap_enable(void) 454 { 455 static_branch_inc(&udpv6_encap_needed_key); 456 } 457 EXPORT_SYMBOL(udpv6_encap_enable); 458 459 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 460 * through error handlers in encapsulations looking for a match. 461 */ 462 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 463 struct inet6_skb_parm *opt, 464 u8 type, u8 code, int offset, __be32 info) 465 { 466 int i; 467 468 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 469 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 470 u8 type, u8 code, int offset, __be32 info); 471 const struct ip6_tnl_encap_ops *encap; 472 473 encap = rcu_dereference(ip6tun_encaps[i]); 474 if (!encap) 475 continue; 476 handler = encap->err_handler; 477 if (handler && !handler(skb, opt, type, code, offset, info)) 478 return 0; 479 } 480 481 return -ENOENT; 482 } 483 484 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 485 * reversing source and destination port: this will match tunnels that force the 486 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 487 * lwtunnels might actually break this assumption by being configured with 488 * different destination ports on endpoints, in this case we won't be able to 489 * trace ICMP messages back to them. 490 * 491 * If this doesn't match any socket, probe tunnels with arbitrary destination 492 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 493 * we've sent packets to won't necessarily match the local destination port. 494 * 495 * Then ask the tunnel implementation to match the error against a valid 496 * association. 497 * 498 * Return an error if we can't find a match, the socket if we need further 499 * processing, zero otherwise. 500 */ 501 static struct sock *__udp6_lib_err_encap(struct net *net, 502 const struct ipv6hdr *hdr, int offset, 503 struct udphdr *uh, 504 struct udp_table *udptable, 505 struct sock *sk, 506 struct sk_buff *skb, 507 struct inet6_skb_parm *opt, 508 u8 type, u8 code, __be32 info) 509 { 510 int (*lookup)(struct sock *sk, struct sk_buff *skb); 511 int network_offset, transport_offset; 512 struct udp_sock *up; 513 514 network_offset = skb_network_offset(skb); 515 transport_offset = skb_transport_offset(skb); 516 517 /* Network header needs to point to the outer IPv6 header inside ICMP */ 518 skb_reset_network_header(skb); 519 520 /* Transport header needs to point to the UDP header */ 521 skb_set_transport_header(skb, offset); 522 523 if (sk) { 524 up = udp_sk(sk); 525 526 lookup = READ_ONCE(up->encap_err_lookup); 527 if (lookup && lookup(sk, skb)) 528 sk = NULL; 529 530 goto out; 531 } 532 533 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 534 &hdr->saddr, uh->dest, 535 inet6_iif(skb), 0, udptable, skb); 536 if (sk) { 537 up = udp_sk(sk); 538 539 lookup = READ_ONCE(up->encap_err_lookup); 540 if (!lookup || lookup(sk, skb)) 541 sk = NULL; 542 } 543 544 out: 545 if (!sk) { 546 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 547 offset, info)); 548 } 549 550 skb_set_transport_header(skb, transport_offset); 551 skb_set_network_header(skb, network_offset); 552 553 return sk; 554 } 555 556 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 557 u8 type, u8 code, int offset, __be32 info, 558 struct udp_table *udptable) 559 { 560 struct ipv6_pinfo *np; 561 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 562 const struct in6_addr *saddr = &hdr->saddr; 563 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr; 564 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 565 bool tunnel = false; 566 struct sock *sk; 567 int harderr; 568 int err; 569 struct net *net = dev_net(skb->dev); 570 571 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 572 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 573 574 if (!sk || udp_sk(sk)->encap_type) { 575 /* No socket for error: try tunnels before discarding */ 576 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 577 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 578 udptable, sk, skb, 579 opt, type, code, info); 580 if (!sk) 581 return 0; 582 } else 583 sk = ERR_PTR(-ENOENT); 584 585 if (IS_ERR(sk)) { 586 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 587 ICMP6_MIB_INERRORS); 588 return PTR_ERR(sk); 589 } 590 591 tunnel = true; 592 } 593 594 harderr = icmpv6_err_convert(type, code, &err); 595 np = inet6_sk(sk); 596 597 if (type == ICMPV6_PKT_TOOBIG) { 598 if (!ip6_sk_accept_pmtu(sk)) 599 goto out; 600 ip6_sk_update_pmtu(skb, sk, info); 601 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 602 harderr = 1; 603 } 604 if (type == NDISC_REDIRECT) { 605 if (tunnel) { 606 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 607 READ_ONCE(sk->sk_mark), sk->sk_uid); 608 } else { 609 ip6_sk_redirect(skb, sk); 610 } 611 goto out; 612 } 613 614 /* Tunnels don't have an application socket: don't pass errors back */ 615 if (tunnel) { 616 if (udp_sk(sk)->encap_err_rcv) 617 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, 618 ntohl(info), (u8 *)(uh+1)); 619 goto out; 620 } 621 622 if (!np->recverr) { 623 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 624 goto out; 625 } else { 626 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 627 } 628 629 sk->sk_err = err; 630 sk_error_report(sk); 631 out: 632 return 0; 633 } 634 635 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 636 { 637 int rc; 638 639 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 640 sock_rps_save_rxhash(sk, skb); 641 sk_mark_napi_id(sk, skb); 642 sk_incoming_cpu_update(sk); 643 } else { 644 sk_mark_napi_id_once(sk, skb); 645 } 646 647 rc = __udp_enqueue_schedule_skb(sk, skb); 648 if (rc < 0) { 649 int is_udplite = IS_UDPLITE(sk); 650 enum skb_drop_reason drop_reason; 651 652 /* Note that an ENOMEM error is charged twice */ 653 if (rc == -ENOMEM) { 654 UDP6_INC_STATS(sock_net(sk), 655 UDP_MIB_RCVBUFERRORS, is_udplite); 656 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 657 } else { 658 UDP6_INC_STATS(sock_net(sk), 659 UDP_MIB_MEMERRORS, is_udplite); 660 drop_reason = SKB_DROP_REASON_PROTO_MEM; 661 } 662 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 663 kfree_skb_reason(skb, drop_reason); 664 trace_udp_fail_queue_rcv_skb(rc, sk); 665 return -1; 666 } 667 668 return 0; 669 } 670 671 static __inline__ int udpv6_err(struct sk_buff *skb, 672 struct inet6_skb_parm *opt, u8 type, 673 u8 code, int offset, __be32 info) 674 { 675 return __udp6_lib_err(skb, opt, type, code, offset, info, 676 dev_net(skb->dev)->ipv4.udp_table); 677 } 678 679 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 680 { 681 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 682 struct udp_sock *up = udp_sk(sk); 683 int is_udplite = IS_UDPLITE(sk); 684 685 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 686 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 687 goto drop; 688 } 689 nf_reset_ct(skb); 690 691 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 692 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 693 694 /* 695 * This is an encapsulation socket so pass the skb to 696 * the socket's udp_encap_rcv() hook. Otherwise, just 697 * fall through and pass this up the UDP socket. 698 * up->encap_rcv() returns the following value: 699 * =0 if skb was successfully passed to the encap 700 * handler or was discarded by it. 701 * >0 if skb should be passed on to UDP. 702 * <0 if skb should be resubmitted as proto -N 703 */ 704 705 /* if we're overly short, let UDP handle it */ 706 encap_rcv = READ_ONCE(up->encap_rcv); 707 if (encap_rcv) { 708 int ret; 709 710 /* Verify checksum before giving to encap */ 711 if (udp_lib_checksum_complete(skb)) 712 goto csum_error; 713 714 ret = encap_rcv(sk, skb); 715 if (ret <= 0) { 716 __UDP6_INC_STATS(sock_net(sk), 717 UDP_MIB_INDATAGRAMS, 718 is_udplite); 719 return -ret; 720 } 721 } 722 723 /* FALLTHROUGH -- it's a UDP Packet */ 724 } 725 726 /* 727 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 728 */ 729 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 730 731 if (up->pcrlen == 0) { /* full coverage was set */ 732 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 733 UDP_SKB_CB(skb)->cscov, skb->len); 734 goto drop; 735 } 736 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 737 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 738 UDP_SKB_CB(skb)->cscov, up->pcrlen); 739 goto drop; 740 } 741 } 742 743 prefetch(&sk->sk_rmem_alloc); 744 if (rcu_access_pointer(sk->sk_filter) && 745 udp_lib_checksum_complete(skb)) 746 goto csum_error; 747 748 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 749 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 750 goto drop; 751 } 752 753 udp_csum_pull_header(skb); 754 755 skb_dst_drop(skb); 756 757 return __udpv6_queue_rcv_skb(sk, skb); 758 759 csum_error: 760 drop_reason = SKB_DROP_REASON_UDP_CSUM; 761 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 762 drop: 763 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 764 atomic_inc(&sk->sk_drops); 765 kfree_skb_reason(skb, drop_reason); 766 return -1; 767 } 768 769 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 770 { 771 struct sk_buff *next, *segs; 772 int ret; 773 774 if (likely(!udp_unexpected_gso(sk, skb))) 775 return udpv6_queue_rcv_one_skb(sk, skb); 776 777 __skb_push(skb, -skb_mac_offset(skb)); 778 segs = udp_rcv_segment(sk, skb, false); 779 skb_list_walk_safe(segs, skb, next) { 780 __skb_pull(skb, skb_transport_offset(skb)); 781 782 udp_post_segment_fix_csum(skb); 783 ret = udpv6_queue_rcv_one_skb(sk, skb); 784 if (ret > 0) 785 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 786 true); 787 } 788 return 0; 789 } 790 791 static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk, 792 __be16 loc_port, const struct in6_addr *loc_addr, 793 __be16 rmt_port, const struct in6_addr *rmt_addr, 794 int dif, int sdif, unsigned short hnum) 795 { 796 const struct inet_sock *inet = inet_sk(sk); 797 798 if (!net_eq(sock_net(sk), net)) 799 return false; 800 801 if (udp_sk(sk)->udp_port_hash != hnum || 802 sk->sk_family != PF_INET6 || 803 (inet->inet_dport && inet->inet_dport != rmt_port) || 804 (!ipv6_addr_any(&sk->sk_v6_daddr) && 805 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 806 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) || 807 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 808 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 809 return false; 810 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 811 return false; 812 return true; 813 } 814 815 static void udp6_csum_zero_error(struct sk_buff *skb) 816 { 817 /* RFC 2460 section 8.1 says that we SHOULD log 818 * this error. Well, it is reasonable. 819 */ 820 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 821 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 822 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 823 } 824 825 /* 826 * Note: called only from the BH handler context, 827 * so we don't need to lock the hashes. 828 */ 829 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 830 const struct in6_addr *saddr, const struct in6_addr *daddr, 831 struct udp_table *udptable, int proto) 832 { 833 struct sock *sk, *first = NULL; 834 const struct udphdr *uh = udp_hdr(skb); 835 unsigned short hnum = ntohs(uh->dest); 836 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 837 unsigned int offset = offsetof(typeof(*sk), sk_node); 838 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 839 int dif = inet6_iif(skb); 840 int sdif = inet6_sdif(skb); 841 struct hlist_node *node; 842 struct sk_buff *nskb; 843 844 if (use_hash2) { 845 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 846 udptable->mask; 847 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 848 start_lookup: 849 hslot = &udptable->hash2[hash2]; 850 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 851 } 852 853 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 854 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 855 uh->source, saddr, dif, sdif, 856 hnum)) 857 continue; 858 /* If zero checksum and no_check is not on for 859 * the socket then skip it. 860 */ 861 if (!uh->check && !udp_sk(sk)->no_check6_rx) 862 continue; 863 if (!first) { 864 first = sk; 865 continue; 866 } 867 nskb = skb_clone(skb, GFP_ATOMIC); 868 if (unlikely(!nskb)) { 869 atomic_inc(&sk->sk_drops); 870 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 871 IS_UDPLITE(sk)); 872 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 873 IS_UDPLITE(sk)); 874 continue; 875 } 876 877 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 878 consume_skb(nskb); 879 } 880 881 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 882 if (use_hash2 && hash2 != hash2_any) { 883 hash2 = hash2_any; 884 goto start_lookup; 885 } 886 887 if (first) { 888 if (udpv6_queue_rcv_skb(first, skb) > 0) 889 consume_skb(skb); 890 } else { 891 kfree_skb(skb); 892 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 893 proto == IPPROTO_UDPLITE); 894 } 895 return 0; 896 } 897 898 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 899 { 900 if (udp_sk_rx_dst_set(sk, dst)) { 901 const struct rt6_info *rt = (const struct rt6_info *)dst; 902 903 sk->sk_rx_dst_cookie = rt6_get_cookie(rt); 904 } 905 } 906 907 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 908 * return code conversion for ip layer consumption 909 */ 910 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 911 struct udphdr *uh) 912 { 913 int ret; 914 915 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 916 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 917 918 ret = udpv6_queue_rcv_skb(sk, skb); 919 920 /* a return value > 0 means to resubmit the input */ 921 if (ret > 0) 922 return ret; 923 return 0; 924 } 925 926 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 927 int proto) 928 { 929 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 930 const struct in6_addr *saddr, *daddr; 931 struct net *net = dev_net(skb->dev); 932 struct udphdr *uh; 933 struct sock *sk; 934 bool refcounted; 935 u32 ulen = 0; 936 937 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 938 goto discard; 939 940 saddr = &ipv6_hdr(skb)->saddr; 941 daddr = &ipv6_hdr(skb)->daddr; 942 uh = udp_hdr(skb); 943 944 ulen = ntohs(uh->len); 945 if (ulen > skb->len) 946 goto short_packet; 947 948 if (proto == IPPROTO_UDP) { 949 /* UDP validates ulen. */ 950 951 /* Check for jumbo payload */ 952 if (ulen == 0) 953 ulen = skb->len; 954 955 if (ulen < sizeof(*uh)) 956 goto short_packet; 957 958 if (ulen < skb->len) { 959 if (pskb_trim_rcsum(skb, ulen)) 960 goto short_packet; 961 saddr = &ipv6_hdr(skb)->saddr; 962 daddr = &ipv6_hdr(skb)->daddr; 963 uh = udp_hdr(skb); 964 } 965 } 966 967 if (udp6_csum_init(skb, uh, proto)) 968 goto csum_error; 969 970 /* Check if the socket is already available, e.g. due to early demux */ 971 sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 972 &refcounted, udp6_ehashfn); 973 if (IS_ERR(sk)) 974 goto no_sk; 975 976 if (sk) { 977 struct dst_entry *dst = skb_dst(skb); 978 int ret; 979 980 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 981 udp6_sk_rx_dst_set(sk, dst); 982 983 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 984 if (refcounted) 985 sock_put(sk); 986 goto report_csum_error; 987 } 988 989 ret = udp6_unicast_rcv_skb(sk, skb, uh); 990 if (refcounted) 991 sock_put(sk); 992 return ret; 993 } 994 995 /* 996 * Multicast receive code 997 */ 998 if (ipv6_addr_is_multicast(daddr)) 999 return __udp6_lib_mcast_deliver(net, skb, 1000 saddr, daddr, udptable, proto); 1001 1002 /* Unicast */ 1003 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1004 if (sk) { 1005 if (!uh->check && !udp_sk(sk)->no_check6_rx) 1006 goto report_csum_error; 1007 return udp6_unicast_rcv_skb(sk, skb, uh); 1008 } 1009 no_sk: 1010 reason = SKB_DROP_REASON_NO_SOCKET; 1011 1012 if (!uh->check) 1013 goto report_csum_error; 1014 1015 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1016 goto discard; 1017 nf_reset_ct(skb); 1018 1019 if (udp_lib_checksum_complete(skb)) 1020 goto csum_error; 1021 1022 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1023 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 1024 1025 kfree_skb_reason(skb, reason); 1026 return 0; 1027 1028 short_packet: 1029 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1030 reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1031 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 1032 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1033 saddr, ntohs(uh->source), 1034 ulen, skb->len, 1035 daddr, ntohs(uh->dest)); 1036 goto discard; 1037 1038 report_csum_error: 1039 udp6_csum_zero_error(skb); 1040 csum_error: 1041 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1042 reason = SKB_DROP_REASON_UDP_CSUM; 1043 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1044 discard: 1045 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1046 kfree_skb_reason(skb, reason); 1047 return 0; 1048 } 1049 1050 1051 static struct sock *__udp6_lib_demux_lookup(struct net *net, 1052 __be16 loc_port, const struct in6_addr *loc_addr, 1053 __be16 rmt_port, const struct in6_addr *rmt_addr, 1054 int dif, int sdif) 1055 { 1056 struct udp_table *udptable = net->ipv4.udp_table; 1057 unsigned short hnum = ntohs(loc_port); 1058 unsigned int hash2, slot2; 1059 struct udp_hslot *hslot2; 1060 __portpair ports; 1061 struct sock *sk; 1062 1063 hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1064 slot2 = hash2 & udptable->mask; 1065 hslot2 = &udptable->hash2[slot2]; 1066 ports = INET_COMBINED_PORTS(rmt_port, hnum); 1067 1068 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1069 if (sk->sk_state == TCP_ESTABLISHED && 1070 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif)) 1071 return sk; 1072 /* Only check first socket in chain */ 1073 break; 1074 } 1075 return NULL; 1076 } 1077 1078 void udp_v6_early_demux(struct sk_buff *skb) 1079 { 1080 struct net *net = dev_net(skb->dev); 1081 const struct udphdr *uh; 1082 struct sock *sk; 1083 struct dst_entry *dst; 1084 int dif = skb->dev->ifindex; 1085 int sdif = inet6_sdif(skb); 1086 1087 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1088 sizeof(struct udphdr))) 1089 return; 1090 1091 uh = udp_hdr(skb); 1092 1093 if (skb->pkt_type == PACKET_HOST) 1094 sk = __udp6_lib_demux_lookup(net, uh->dest, 1095 &ipv6_hdr(skb)->daddr, 1096 uh->source, &ipv6_hdr(skb)->saddr, 1097 dif, sdif); 1098 else 1099 return; 1100 1101 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1102 return; 1103 1104 skb->sk = sk; 1105 skb->destructor = sock_efree; 1106 dst = rcu_dereference(sk->sk_rx_dst); 1107 1108 if (dst) 1109 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1110 if (dst) { 1111 /* set noref for now. 1112 * any place which wants to hold dst has to call 1113 * dst_hold_safe() 1114 */ 1115 skb_dst_set_noref(skb, dst); 1116 } 1117 } 1118 1119 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1120 { 1121 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 1122 } 1123 1124 /* 1125 * Throw away all pending data and cancel the corking. Socket is locked. 1126 */ 1127 static void udp_v6_flush_pending_frames(struct sock *sk) 1128 { 1129 struct udp_sock *up = udp_sk(sk); 1130 1131 if (up->pending == AF_INET) 1132 udp_flush_pending_frames(sk); 1133 else if (up->pending) { 1134 up->len = 0; 1135 up->pending = 0; 1136 ip6_flush_pending_frames(sk); 1137 } 1138 } 1139 1140 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1141 int addr_len) 1142 { 1143 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1144 return -EINVAL; 1145 /* The following checks are replicated from __ip6_datagram_connect() 1146 * and intended to prevent BPF program called below from accessing 1147 * bytes that are out of the bound specified by user in addr_len. 1148 */ 1149 if (uaddr->sa_family == AF_INET) { 1150 if (ipv6_only_sock(sk)) 1151 return -EAFNOSUPPORT; 1152 return udp_pre_connect(sk, uaddr, addr_len); 1153 } 1154 1155 if (addr_len < SIN6_LEN_RFC2133) 1156 return -EINVAL; 1157 1158 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1159 } 1160 1161 /** 1162 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1163 * @sk: socket we are sending on 1164 * @skb: sk_buff containing the filled-in UDP header 1165 * (checksum field must be zeroed out) 1166 * @saddr: source address 1167 * @daddr: destination address 1168 * @len: length of packet 1169 */ 1170 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1171 const struct in6_addr *saddr, 1172 const struct in6_addr *daddr, int len) 1173 { 1174 unsigned int offset; 1175 struct udphdr *uh = udp_hdr(skb); 1176 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1177 __wsum csum = 0; 1178 1179 if (!frags) { 1180 /* Only one fragment on the socket. */ 1181 skb->csum_start = skb_transport_header(skb) - skb->head; 1182 skb->csum_offset = offsetof(struct udphdr, check); 1183 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1184 } else { 1185 /* 1186 * HW-checksum won't work as there are two or more 1187 * fragments on the socket so that all csums of sk_buffs 1188 * should be together 1189 */ 1190 offset = skb_transport_offset(skb); 1191 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1192 csum = skb->csum; 1193 1194 skb->ip_summed = CHECKSUM_NONE; 1195 1196 do { 1197 csum = csum_add(csum, frags->csum); 1198 } while ((frags = frags->next)); 1199 1200 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1201 csum); 1202 if (uh->check == 0) 1203 uh->check = CSUM_MANGLED_0; 1204 } 1205 } 1206 1207 /* 1208 * Sending 1209 */ 1210 1211 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1212 struct inet_cork *cork) 1213 { 1214 struct sock *sk = skb->sk; 1215 struct udphdr *uh; 1216 int err = 0; 1217 int is_udplite = IS_UDPLITE(sk); 1218 __wsum csum = 0; 1219 int offset = skb_transport_offset(skb); 1220 int len = skb->len - offset; 1221 int datalen = len - sizeof(*uh); 1222 1223 /* 1224 * Create a UDP header 1225 */ 1226 uh = udp_hdr(skb); 1227 uh->source = fl6->fl6_sport; 1228 uh->dest = fl6->fl6_dport; 1229 uh->len = htons(len); 1230 uh->check = 0; 1231 1232 if (cork->gso_size) { 1233 const int hlen = skb_network_header_len(skb) + 1234 sizeof(struct udphdr); 1235 1236 if (hlen + cork->gso_size > cork->fragsize) { 1237 kfree_skb(skb); 1238 return -EINVAL; 1239 } 1240 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1241 kfree_skb(skb); 1242 return -EINVAL; 1243 } 1244 if (udp_sk(sk)->no_check6_tx) { 1245 kfree_skb(skb); 1246 return -EINVAL; 1247 } 1248 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1249 dst_xfrm(skb_dst(skb))) { 1250 kfree_skb(skb); 1251 return -EIO; 1252 } 1253 1254 if (datalen > cork->gso_size) { 1255 skb_shinfo(skb)->gso_size = cork->gso_size; 1256 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1257 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1258 cork->gso_size); 1259 } 1260 goto csum_partial; 1261 } 1262 1263 if (is_udplite) 1264 csum = udplite_csum(skb); 1265 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1266 skb->ip_summed = CHECKSUM_NONE; 1267 goto send; 1268 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1269 csum_partial: 1270 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1271 goto send; 1272 } else 1273 csum = udp_csum(skb); 1274 1275 /* add protocol-dependent pseudo-header */ 1276 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1277 len, fl6->flowi6_proto, csum); 1278 if (uh->check == 0) 1279 uh->check = CSUM_MANGLED_0; 1280 1281 send: 1282 err = ip6_send_skb(skb); 1283 if (err) { 1284 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1285 UDP6_INC_STATS(sock_net(sk), 1286 UDP_MIB_SNDBUFERRORS, is_udplite); 1287 err = 0; 1288 } 1289 } else { 1290 UDP6_INC_STATS(sock_net(sk), 1291 UDP_MIB_OUTDATAGRAMS, is_udplite); 1292 } 1293 return err; 1294 } 1295 1296 static int udp_v6_push_pending_frames(struct sock *sk) 1297 { 1298 struct sk_buff *skb; 1299 struct udp_sock *up = udp_sk(sk); 1300 int err = 0; 1301 1302 if (up->pending == AF_INET) 1303 return udp_push_pending_frames(sk); 1304 1305 skb = ip6_finish_skb(sk); 1306 if (!skb) 1307 goto out; 1308 1309 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6, 1310 &inet_sk(sk)->cork.base); 1311 out: 1312 up->len = 0; 1313 up->pending = 0; 1314 return err; 1315 } 1316 1317 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1318 { 1319 struct ipv6_txoptions opt_space; 1320 struct udp_sock *up = udp_sk(sk); 1321 struct inet_sock *inet = inet_sk(sk); 1322 struct ipv6_pinfo *np = inet6_sk(sk); 1323 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1324 struct in6_addr *daddr, *final_p, final; 1325 struct ipv6_txoptions *opt = NULL; 1326 struct ipv6_txoptions *opt_to_free = NULL; 1327 struct ip6_flowlabel *flowlabel = NULL; 1328 struct inet_cork_full cork; 1329 struct flowi6 *fl6 = &cork.fl.u.ip6; 1330 struct dst_entry *dst; 1331 struct ipcm6_cookie ipc6; 1332 int addr_len = msg->msg_namelen; 1333 bool connected = false; 1334 int ulen = len; 1335 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1336 int err; 1337 int is_udplite = IS_UDPLITE(sk); 1338 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1339 1340 ipcm6_init(&ipc6); 1341 ipc6.gso_size = READ_ONCE(up->gso_size); 1342 ipc6.sockc.tsflags = sk->sk_tsflags; 1343 ipc6.sockc.mark = READ_ONCE(sk->sk_mark); 1344 1345 /* destination address check */ 1346 if (sin6) { 1347 if (addr_len < offsetof(struct sockaddr, sa_data)) 1348 return -EINVAL; 1349 1350 switch (sin6->sin6_family) { 1351 case AF_INET6: 1352 if (addr_len < SIN6_LEN_RFC2133) 1353 return -EINVAL; 1354 daddr = &sin6->sin6_addr; 1355 if (ipv6_addr_any(daddr) && 1356 ipv6_addr_v4mapped(&np->saddr)) 1357 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1358 daddr); 1359 break; 1360 case AF_INET: 1361 goto do_udp_sendmsg; 1362 case AF_UNSPEC: 1363 msg->msg_name = sin6 = NULL; 1364 msg->msg_namelen = addr_len = 0; 1365 daddr = NULL; 1366 break; 1367 default: 1368 return -EINVAL; 1369 } 1370 } else if (!up->pending) { 1371 if (sk->sk_state != TCP_ESTABLISHED) 1372 return -EDESTADDRREQ; 1373 daddr = &sk->sk_v6_daddr; 1374 } else 1375 daddr = NULL; 1376 1377 if (daddr) { 1378 if (ipv6_addr_v4mapped(daddr)) { 1379 struct sockaddr_in sin; 1380 sin.sin_family = AF_INET; 1381 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1382 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1383 msg->msg_name = &sin; 1384 msg->msg_namelen = sizeof(sin); 1385 do_udp_sendmsg: 1386 err = ipv6_only_sock(sk) ? 1387 -ENETUNREACH : udp_sendmsg(sk, msg, len); 1388 msg->msg_name = sin6; 1389 msg->msg_namelen = addr_len; 1390 return err; 1391 } 1392 } 1393 1394 /* Rough check on arithmetic overflow, 1395 better check is made in ip6_append_data(). 1396 */ 1397 if (len > INT_MAX - sizeof(struct udphdr)) 1398 return -EMSGSIZE; 1399 1400 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1401 if (up->pending) { 1402 if (up->pending == AF_INET) 1403 return udp_sendmsg(sk, msg, len); 1404 /* 1405 * There are pending frames. 1406 * The socket lock must be held while it's corked. 1407 */ 1408 lock_sock(sk); 1409 if (likely(up->pending)) { 1410 if (unlikely(up->pending != AF_INET6)) { 1411 release_sock(sk); 1412 return -EAFNOSUPPORT; 1413 } 1414 dst = NULL; 1415 goto do_append_data; 1416 } 1417 release_sock(sk); 1418 } 1419 ulen += sizeof(struct udphdr); 1420 1421 memset(fl6, 0, sizeof(*fl6)); 1422 1423 if (sin6) { 1424 if (sin6->sin6_port == 0) 1425 return -EINVAL; 1426 1427 fl6->fl6_dport = sin6->sin6_port; 1428 daddr = &sin6->sin6_addr; 1429 1430 if (np->sndflow) { 1431 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1432 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) { 1433 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1434 if (IS_ERR(flowlabel)) 1435 return -EINVAL; 1436 } 1437 } 1438 1439 /* 1440 * Otherwise it will be difficult to maintain 1441 * sk->sk_dst_cache. 1442 */ 1443 if (sk->sk_state == TCP_ESTABLISHED && 1444 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1445 daddr = &sk->sk_v6_daddr; 1446 1447 if (addr_len >= sizeof(struct sockaddr_in6) && 1448 sin6->sin6_scope_id && 1449 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1450 fl6->flowi6_oif = sin6->sin6_scope_id; 1451 } else { 1452 if (sk->sk_state != TCP_ESTABLISHED) 1453 return -EDESTADDRREQ; 1454 1455 fl6->fl6_dport = inet->inet_dport; 1456 daddr = &sk->sk_v6_daddr; 1457 fl6->flowlabel = np->flow_label; 1458 connected = true; 1459 } 1460 1461 if (!fl6->flowi6_oif) 1462 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if); 1463 1464 if (!fl6->flowi6_oif) 1465 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1466 1467 fl6->flowi6_uid = sk->sk_uid; 1468 1469 if (msg->msg_controllen) { 1470 opt = &opt_space; 1471 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1472 opt->tot_len = sizeof(*opt); 1473 ipc6.opt = opt; 1474 1475 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1476 if (err > 0) 1477 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6, 1478 &ipc6); 1479 if (err < 0) { 1480 fl6_sock_release(flowlabel); 1481 return err; 1482 } 1483 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1484 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1485 if (IS_ERR(flowlabel)) 1486 return -EINVAL; 1487 } 1488 if (!(opt->opt_nflen|opt->opt_flen)) 1489 opt = NULL; 1490 connected = false; 1491 } 1492 if (!opt) { 1493 opt = txopt_get(np); 1494 opt_to_free = opt; 1495 } 1496 if (flowlabel) 1497 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1498 opt = ipv6_fixup_options(&opt_space, opt); 1499 ipc6.opt = opt; 1500 1501 fl6->flowi6_proto = sk->sk_protocol; 1502 fl6->flowi6_mark = ipc6.sockc.mark; 1503 fl6->daddr = *daddr; 1504 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr)) 1505 fl6->saddr = np->saddr; 1506 fl6->fl6_sport = inet->inet_sport; 1507 1508 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1509 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1510 (struct sockaddr *)sin6, 1511 &fl6->saddr); 1512 if (err) 1513 goto out_no_dst; 1514 if (sin6) { 1515 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1516 /* BPF program rewrote IPv6-only by IPv4-mapped 1517 * IPv6. It's currently unsupported. 1518 */ 1519 err = -ENOTSUPP; 1520 goto out_no_dst; 1521 } 1522 if (sin6->sin6_port == 0) { 1523 /* BPF program set invalid port. Reject it. */ 1524 err = -EINVAL; 1525 goto out_no_dst; 1526 } 1527 fl6->fl6_dport = sin6->sin6_port; 1528 fl6->daddr = sin6->sin6_addr; 1529 } 1530 } 1531 1532 if (ipv6_addr_any(&fl6->daddr)) 1533 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1534 1535 final_p = fl6_update_dst(fl6, opt, &final); 1536 if (final_p) 1537 connected = false; 1538 1539 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) { 1540 fl6->flowi6_oif = np->mcast_oif; 1541 connected = false; 1542 } else if (!fl6->flowi6_oif) 1543 fl6->flowi6_oif = np->ucast_oif; 1544 1545 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6)); 1546 1547 if (ipc6.tclass < 0) 1548 ipc6.tclass = np->tclass; 1549 1550 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel); 1551 1552 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected); 1553 if (IS_ERR(dst)) { 1554 err = PTR_ERR(dst); 1555 dst = NULL; 1556 goto out; 1557 } 1558 1559 if (ipc6.hlimit < 0) 1560 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst); 1561 1562 if (msg->msg_flags&MSG_CONFIRM) 1563 goto do_confirm; 1564 back_from_confirm: 1565 1566 /* Lockless fast path for the non-corking case */ 1567 if (!corkreq) { 1568 struct sk_buff *skb; 1569 1570 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1571 sizeof(struct udphdr), &ipc6, 1572 (struct rt6_info *)dst, 1573 msg->msg_flags, &cork); 1574 err = PTR_ERR(skb); 1575 if (!IS_ERR_OR_NULL(skb)) 1576 err = udp_v6_send_skb(skb, fl6, &cork.base); 1577 /* ip6_make_skb steals dst reference */ 1578 goto out_no_dst; 1579 } 1580 1581 lock_sock(sk); 1582 if (unlikely(up->pending)) { 1583 /* The socket is already corked while preparing it. */ 1584 /* ... which is an evident application bug. --ANK */ 1585 release_sock(sk); 1586 1587 net_dbg_ratelimited("udp cork app bug 2\n"); 1588 err = -EINVAL; 1589 goto out; 1590 } 1591 1592 up->pending = AF_INET6; 1593 1594 do_append_data: 1595 if (ipc6.dontfrag < 0) 1596 ipc6.dontfrag = np->dontfrag; 1597 up->len += ulen; 1598 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1599 &ipc6, fl6, (struct rt6_info *)dst, 1600 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1601 if (err) 1602 udp_v6_flush_pending_frames(sk); 1603 else if (!corkreq) 1604 err = udp_v6_push_pending_frames(sk); 1605 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1606 up->pending = 0; 1607 1608 if (err > 0) 1609 err = np->recverr ? net_xmit_errno(err) : 0; 1610 release_sock(sk); 1611 1612 out: 1613 dst_release(dst); 1614 out_no_dst: 1615 fl6_sock_release(flowlabel); 1616 txopt_put(opt_to_free); 1617 if (!err) 1618 return len; 1619 /* 1620 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1621 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1622 * we don't have a good statistic (IpOutDiscards but it can be too many 1623 * things). We could add another new stat but at least for now that 1624 * seems like overkill. 1625 */ 1626 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1627 UDP6_INC_STATS(sock_net(sk), 1628 UDP_MIB_SNDBUFERRORS, is_udplite); 1629 } 1630 return err; 1631 1632 do_confirm: 1633 if (msg->msg_flags & MSG_PROBE) 1634 dst_confirm_neigh(dst, &fl6->daddr); 1635 if (!(msg->msg_flags&MSG_PROBE) || len) 1636 goto back_from_confirm; 1637 err = 0; 1638 goto out; 1639 } 1640 EXPORT_SYMBOL(udpv6_sendmsg); 1641 1642 static void udpv6_splice_eof(struct socket *sock) 1643 { 1644 struct sock *sk = sock->sk; 1645 struct udp_sock *up = udp_sk(sk); 1646 1647 if (!up->pending || READ_ONCE(up->corkflag)) 1648 return; 1649 1650 lock_sock(sk); 1651 if (up->pending && !READ_ONCE(up->corkflag)) 1652 udp_v6_push_pending_frames(sk); 1653 release_sock(sk); 1654 } 1655 1656 void udpv6_destroy_sock(struct sock *sk) 1657 { 1658 struct udp_sock *up = udp_sk(sk); 1659 lock_sock(sk); 1660 1661 /* protects from races with udp_abort() */ 1662 sock_set_flag(sk, SOCK_DEAD); 1663 udp_v6_flush_pending_frames(sk); 1664 release_sock(sk); 1665 1666 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1667 if (up->encap_type) { 1668 void (*encap_destroy)(struct sock *sk); 1669 encap_destroy = READ_ONCE(up->encap_destroy); 1670 if (encap_destroy) 1671 encap_destroy(sk); 1672 } 1673 if (up->encap_enabled) { 1674 static_branch_dec(&udpv6_encap_needed_key); 1675 udp_encap_disable(); 1676 } 1677 } 1678 } 1679 1680 /* 1681 * Socket option code for UDP 1682 */ 1683 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1684 unsigned int optlen) 1685 { 1686 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 1687 return udp_lib_setsockopt(sk, level, optname, 1688 optval, optlen, 1689 udp_v6_push_pending_frames); 1690 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1691 } 1692 1693 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1694 char __user *optval, int __user *optlen) 1695 { 1696 if (level == SOL_UDP || level == SOL_UDPLITE) 1697 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1698 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1699 } 1700 1701 static const struct inet6_protocol udpv6_protocol = { 1702 .handler = udpv6_rcv, 1703 .err_handler = udpv6_err, 1704 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1705 }; 1706 1707 /* ------------------------------------------------------------------------ */ 1708 #ifdef CONFIG_PROC_FS 1709 int udp6_seq_show(struct seq_file *seq, void *v) 1710 { 1711 if (v == SEQ_START_TOKEN) { 1712 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1713 } else { 1714 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1715 const struct inet_sock *inet = inet_sk((const struct sock *)v); 1716 __u16 srcp = ntohs(inet->inet_sport); 1717 __u16 destp = ntohs(inet->inet_dport); 1718 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1719 udp_rqueue_get(v), bucket); 1720 } 1721 return 0; 1722 } 1723 1724 const struct seq_operations udp6_seq_ops = { 1725 .start = udp_seq_start, 1726 .next = udp_seq_next, 1727 .stop = udp_seq_stop, 1728 .show = udp6_seq_show, 1729 }; 1730 EXPORT_SYMBOL(udp6_seq_ops); 1731 1732 static struct udp_seq_afinfo udp6_seq_afinfo = { 1733 .family = AF_INET6, 1734 .udp_table = NULL, 1735 }; 1736 1737 int __net_init udp6_proc_init(struct net *net) 1738 { 1739 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1740 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1741 return -ENOMEM; 1742 return 0; 1743 } 1744 1745 void udp6_proc_exit(struct net *net) 1746 { 1747 remove_proc_entry("udp6", net->proc_net); 1748 } 1749 #endif /* CONFIG_PROC_FS */ 1750 1751 /* ------------------------------------------------------------------------ */ 1752 1753 struct proto udpv6_prot = { 1754 .name = "UDPv6", 1755 .owner = THIS_MODULE, 1756 .close = udp_lib_close, 1757 .pre_connect = udpv6_pre_connect, 1758 .connect = ip6_datagram_connect, 1759 .disconnect = udp_disconnect, 1760 .ioctl = udp_ioctl, 1761 .init = udpv6_init_sock, 1762 .destroy = udpv6_destroy_sock, 1763 .setsockopt = udpv6_setsockopt, 1764 .getsockopt = udpv6_getsockopt, 1765 .sendmsg = udpv6_sendmsg, 1766 .recvmsg = udpv6_recvmsg, 1767 .splice_eof = udpv6_splice_eof, 1768 .release_cb = ip6_datagram_release_cb, 1769 .hash = udp_lib_hash, 1770 .unhash = udp_lib_unhash, 1771 .rehash = udp_v6_rehash, 1772 .get_port = udp_v6_get_port, 1773 .put_port = udp_lib_unhash, 1774 #ifdef CONFIG_BPF_SYSCALL 1775 .psock_update_sk_prot = udp_bpf_update_proto, 1776 #endif 1777 1778 .memory_allocated = &udp_memory_allocated, 1779 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 1780 1781 .sysctl_mem = sysctl_udp_mem, 1782 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1783 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1784 .obj_size = sizeof(struct udp6_sock), 1785 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6), 1786 .h.udp_table = NULL, 1787 .diag_destroy = udp_abort, 1788 }; 1789 1790 static struct inet_protosw udpv6_protosw = { 1791 .type = SOCK_DGRAM, 1792 .protocol = IPPROTO_UDP, 1793 .prot = &udpv6_prot, 1794 .ops = &inet6_dgram_ops, 1795 .flags = INET_PROTOSW_PERMANENT, 1796 }; 1797 1798 int __init udpv6_init(void) 1799 { 1800 int ret; 1801 1802 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1803 if (ret) 1804 goto out; 1805 1806 ret = inet6_register_protosw(&udpv6_protosw); 1807 if (ret) 1808 goto out_udpv6_protocol; 1809 out: 1810 return ret; 1811 1812 out_udpv6_protocol: 1813 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1814 goto out; 1815 } 1816 1817 void udpv6_exit(void) 1818 { 1819 inet6_unregister_protosw(&udpv6_protosw); 1820 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1821 } 1822