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 u32 lhash, fhash; 83 84 net_get_random_once(&udp6_ehash_secret, 85 sizeof(udp6_ehash_secret)); 86 net_get_random_once(&udp_ipv6_hash_secret, 87 sizeof(udp_ipv6_hash_secret)); 88 89 lhash = (__force u32)laddr->s6_addr32[3]; 90 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 91 92 return __inet6_ehashfn(lhash, lport, fhash, fport, 93 udp6_ehash_secret + net_hash_mix(net)); 94 } 95 96 int udp_v6_get_port(struct sock *sk, unsigned short snum) 97 { 98 unsigned int hash2_nulladdr = 99 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 100 unsigned int hash2_partial = 101 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 102 103 /* precompute partial secondary hash */ 104 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 105 return udp_lib_get_port(sk, snum, hash2_nulladdr); 106 } 107 108 void udp_v6_rehash(struct sock *sk) 109 { 110 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 111 &sk->sk_v6_rcv_saddr, 112 inet_sk(sk)->inet_num); 113 114 udp_lib_rehash(sk, new_hash); 115 } 116 117 static int compute_score(struct sock *sk, struct net *net, 118 const struct in6_addr *saddr, __be16 sport, 119 const struct in6_addr *daddr, unsigned short hnum, 120 int dif, int sdif) 121 { 122 int bound_dev_if, score; 123 struct inet_sock *inet; 124 bool dev_match; 125 126 if (!net_eq(sock_net(sk), net) || 127 udp_sk(sk)->udp_port_hash != hnum || 128 sk->sk_family != PF_INET6) 129 return -1; 130 131 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 132 return -1; 133 134 score = 0; 135 inet = inet_sk(sk); 136 137 if (inet->inet_dport) { 138 if (inet->inet_dport != sport) 139 return -1; 140 score++; 141 } 142 143 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 144 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 145 return -1; 146 score++; 147 } 148 149 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 150 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif); 151 if (!dev_match) 152 return -1; 153 if (bound_dev_if) 154 score++; 155 156 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 157 score++; 158 159 return score; 160 } 161 162 /* called with rcu_read_lock() */ 163 static struct sock *udp6_lib_lookup2(struct net *net, 164 const struct in6_addr *saddr, __be16 sport, 165 const struct in6_addr *daddr, unsigned int hnum, 166 int dif, int sdif, struct udp_hslot *hslot2, 167 struct sk_buff *skb) 168 { 169 struct sock *sk, *result; 170 int score, badness; 171 172 result = NULL; 173 badness = -1; 174 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 175 score = compute_score(sk, net, saddr, sport, 176 daddr, hnum, dif, sdif); 177 if (score > badness) { 178 badness = score; 179 180 if (sk->sk_state == TCP_ESTABLISHED) { 181 result = sk; 182 continue; 183 } 184 185 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 186 saddr, sport, daddr, hnum, udp6_ehashfn); 187 if (!result) { 188 result = sk; 189 continue; 190 } 191 192 /* Fall back to scoring if group has connections */ 193 if (!reuseport_has_conns(sk)) 194 return result; 195 196 /* Reuseport logic returned an error, keep original score. */ 197 if (IS_ERR(result)) 198 continue; 199 200 badness = compute_score(sk, net, saddr, sport, 201 daddr, hnum, dif, sdif); 202 } 203 } 204 return result; 205 } 206 207 /* rcu_read_lock() must be held */ 208 struct sock *__udp6_lib_lookup(struct net *net, 209 const struct in6_addr *saddr, __be16 sport, 210 const struct in6_addr *daddr, __be16 dport, 211 int dif, int sdif, struct udp_table *udptable, 212 struct sk_buff *skb) 213 { 214 unsigned short hnum = ntohs(dport); 215 unsigned int hash2, slot2; 216 struct udp_hslot *hslot2; 217 struct sock *result, *sk; 218 219 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 220 slot2 = hash2 & udptable->mask; 221 hslot2 = &udptable->hash2[slot2]; 222 223 /* Lookup connected or non-wildcard sockets */ 224 result = udp6_lib_lookup2(net, saddr, sport, 225 daddr, hnum, dif, sdif, 226 hslot2, skb); 227 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 228 goto done; 229 230 /* Lookup redirect from BPF */ 231 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 232 udptable == net->ipv4.udp_table) { 233 sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 234 saddr, sport, daddr, hnum, dif, 235 udp6_ehashfn); 236 if (sk) { 237 result = sk; 238 goto done; 239 } 240 } 241 242 /* Got non-wildcard socket or error on first lookup */ 243 if (result) 244 goto done; 245 246 /* Lookup wildcard sockets */ 247 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 248 slot2 = hash2 & udptable->mask; 249 hslot2 = &udptable->hash2[slot2]; 250 251 result = udp6_lib_lookup2(net, saddr, sport, 252 &in6addr_any, hnum, dif, sdif, 253 hslot2, skb); 254 done: 255 if (IS_ERR(result)) 256 return NULL; 257 return result; 258 } 259 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 260 261 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 262 __be16 sport, __be16 dport, 263 struct udp_table *udptable) 264 { 265 const struct ipv6hdr *iph = ipv6_hdr(skb); 266 267 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 268 &iph->daddr, dport, inet6_iif(skb), 269 inet6_sdif(skb), udptable, skb); 270 } 271 272 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, 273 __be16 sport, __be16 dport) 274 { 275 const struct ipv6hdr *iph = ipv6_hdr(skb); 276 struct net *net = dev_net(skb->dev); 277 int iif, sdif; 278 279 inet6_get_iif_sdif(skb, &iif, &sdif); 280 281 return __udp6_lib_lookup(net, &iph->saddr, sport, 282 &iph->daddr, dport, iif, 283 sdif, net->ipv4.udp_table, NULL); 284 } 285 286 /* Must be called under rcu_read_lock(). 287 * Does increment socket refcount. 288 */ 289 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 290 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 291 const struct in6_addr *daddr, __be16 dport, int dif) 292 { 293 struct sock *sk; 294 295 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 296 dif, 0, net->ipv4.udp_table, NULL); 297 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 298 sk = NULL; 299 return sk; 300 } 301 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 302 #endif 303 304 /* do not use the scratch area len for jumbogram: their length execeeds the 305 * scratch area space; note that the IP6CB flags is still in the first 306 * cacheline, so checking for jumbograms is cheap 307 */ 308 static int udp6_skb_len(struct sk_buff *skb) 309 { 310 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 311 } 312 313 /* 314 * This should be easy, if there is something there we 315 * return it, otherwise we block. 316 */ 317 318 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 319 int flags, int *addr_len) 320 { 321 struct ipv6_pinfo *np = inet6_sk(sk); 322 struct inet_sock *inet = inet_sk(sk); 323 struct sk_buff *skb; 324 unsigned int ulen, copied; 325 int off, err, peeking = flags & MSG_PEEK; 326 int is_udplite = IS_UDPLITE(sk); 327 struct udp_mib __percpu *mib; 328 bool checksum_valid = false; 329 int is_udp4; 330 331 if (flags & MSG_ERRQUEUE) 332 return ipv6_recv_error(sk, msg, len, addr_len); 333 334 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 335 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 336 337 try_again: 338 off = sk_peek_offset(sk, flags); 339 skb = __skb_recv_udp(sk, flags, &off, &err); 340 if (!skb) 341 return err; 342 343 ulen = udp6_skb_len(skb); 344 copied = len; 345 if (copied > ulen - off) 346 copied = ulen - off; 347 else if (copied < ulen) 348 msg->msg_flags |= MSG_TRUNC; 349 350 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 351 mib = __UDPX_MIB(sk, is_udp4); 352 353 /* 354 * If checksum is needed at all, try to do it while copying the 355 * data. If the data is truncated, or if we only want a partial 356 * coverage checksum (UDP-Lite), do it before the copy. 357 */ 358 359 if (copied < ulen || peeking || 360 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 361 checksum_valid = udp_skb_csum_unnecessary(skb) || 362 !__udp_lib_checksum_complete(skb); 363 if (!checksum_valid) 364 goto csum_copy_err; 365 } 366 367 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 368 if (udp_skb_is_linear(skb)) 369 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 370 else 371 err = skb_copy_datagram_msg(skb, off, msg, copied); 372 } else { 373 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 374 if (err == -EINVAL) 375 goto csum_copy_err; 376 } 377 if (unlikely(err)) { 378 if (!peeking) { 379 atomic_inc(&sk->sk_drops); 380 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 381 } 382 kfree_skb(skb); 383 return err; 384 } 385 if (!peeking) 386 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 387 388 sock_recv_cmsgs(msg, sk, skb); 389 390 /* Copy the address. */ 391 if (msg->msg_name) { 392 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 393 sin6->sin6_family = AF_INET6; 394 sin6->sin6_port = udp_hdr(skb)->source; 395 sin6->sin6_flowinfo = 0; 396 397 if (is_udp4) { 398 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 399 &sin6->sin6_addr); 400 sin6->sin6_scope_id = 0; 401 } else { 402 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 403 sin6->sin6_scope_id = 404 ipv6_iface_scope_id(&sin6->sin6_addr, 405 inet6_iif(skb)); 406 } 407 *addr_len = sizeof(*sin6); 408 409 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 410 (struct sockaddr *)sin6, 411 addr_len); 412 } 413 414 if (udp_test_bit(GRO_ENABLED, sk)) 415 udp_cmsg_recv(msg, sk, skb); 416 417 if (np->rxopt.all) 418 ip6_datagram_recv_common_ctl(sk, msg, skb); 419 420 if (is_udp4) { 421 if (inet_cmsg_flags(inet)) 422 ip_cmsg_recv_offset(msg, sk, skb, 423 sizeof(struct udphdr), off); 424 } else { 425 if (np->rxopt.all) 426 ip6_datagram_recv_specific_ctl(sk, msg, skb); 427 } 428 429 err = copied; 430 if (flags & MSG_TRUNC) 431 err = ulen; 432 433 skb_consume_udp(sk, skb, peeking ? -err : err); 434 return err; 435 436 csum_copy_err: 437 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 438 udp_skb_destructor)) { 439 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 440 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 441 } 442 kfree_skb(skb); 443 444 /* starting over for a new packet, but check if we need to yield */ 445 cond_resched(); 446 msg->msg_flags &= ~MSG_TRUNC; 447 goto try_again; 448 } 449 450 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 451 void udpv6_encap_enable(void) 452 { 453 static_branch_inc(&udpv6_encap_needed_key); 454 } 455 EXPORT_SYMBOL(udpv6_encap_enable); 456 457 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 458 * through error handlers in encapsulations looking for a match. 459 */ 460 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 461 struct inet6_skb_parm *opt, 462 u8 type, u8 code, int offset, __be32 info) 463 { 464 int i; 465 466 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 467 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 468 u8 type, u8 code, int offset, __be32 info); 469 const struct ip6_tnl_encap_ops *encap; 470 471 encap = rcu_dereference(ip6tun_encaps[i]); 472 if (!encap) 473 continue; 474 handler = encap->err_handler; 475 if (handler && !handler(skb, opt, type, code, offset, info)) 476 return 0; 477 } 478 479 return -ENOENT; 480 } 481 482 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 483 * reversing source and destination port: this will match tunnels that force the 484 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 485 * lwtunnels might actually break this assumption by being configured with 486 * different destination ports on endpoints, in this case we won't be able to 487 * trace ICMP messages back to them. 488 * 489 * If this doesn't match any socket, probe tunnels with arbitrary destination 490 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 491 * we've sent packets to won't necessarily match the local destination port. 492 * 493 * Then ask the tunnel implementation to match the error against a valid 494 * association. 495 * 496 * Return an error if we can't find a match, the socket if we need further 497 * processing, zero otherwise. 498 */ 499 static struct sock *__udp6_lib_err_encap(struct net *net, 500 const struct ipv6hdr *hdr, int offset, 501 struct udphdr *uh, 502 struct udp_table *udptable, 503 struct sock *sk, 504 struct sk_buff *skb, 505 struct inet6_skb_parm *opt, 506 u8 type, u8 code, __be32 info) 507 { 508 int (*lookup)(struct sock *sk, struct sk_buff *skb); 509 int network_offset, transport_offset; 510 struct udp_sock *up; 511 512 network_offset = skb_network_offset(skb); 513 transport_offset = skb_transport_offset(skb); 514 515 /* Network header needs to point to the outer IPv6 header inside ICMP */ 516 skb_reset_network_header(skb); 517 518 /* Transport header needs to point to the UDP header */ 519 skb_set_transport_header(skb, offset); 520 521 if (sk) { 522 up = udp_sk(sk); 523 524 lookup = READ_ONCE(up->encap_err_lookup); 525 if (lookup && lookup(sk, skb)) 526 sk = NULL; 527 528 goto out; 529 } 530 531 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 532 &hdr->saddr, uh->dest, 533 inet6_iif(skb), 0, udptable, skb); 534 if (sk) { 535 up = udp_sk(sk); 536 537 lookup = READ_ONCE(up->encap_err_lookup); 538 if (!lookup || lookup(sk, skb)) 539 sk = NULL; 540 } 541 542 out: 543 if (!sk) { 544 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 545 offset, info)); 546 } 547 548 skb_set_transport_header(skb, transport_offset); 549 skb_set_network_header(skb, network_offset); 550 551 return sk; 552 } 553 554 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 555 u8 type, u8 code, int offset, __be32 info, 556 struct udp_table *udptable) 557 { 558 struct ipv6_pinfo *np; 559 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 560 const struct in6_addr *saddr = &hdr->saddr; 561 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr; 562 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 563 bool tunnel = false; 564 struct sock *sk; 565 int harderr; 566 int err; 567 struct net *net = dev_net(skb->dev); 568 569 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 570 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 571 572 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) { 573 /* No socket for error: try tunnels before discarding */ 574 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 575 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 576 udptable, sk, skb, 577 opt, type, code, info); 578 if (!sk) 579 return 0; 580 } else 581 sk = ERR_PTR(-ENOENT); 582 583 if (IS_ERR(sk)) { 584 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 585 ICMP6_MIB_INERRORS); 586 return PTR_ERR(sk); 587 } 588 589 tunnel = true; 590 } 591 592 harderr = icmpv6_err_convert(type, code, &err); 593 np = inet6_sk(sk); 594 595 if (type == ICMPV6_PKT_TOOBIG) { 596 if (!ip6_sk_accept_pmtu(sk)) 597 goto out; 598 ip6_sk_update_pmtu(skb, sk, info); 599 if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT) 600 harderr = 1; 601 } 602 if (type == NDISC_REDIRECT) { 603 if (tunnel) { 604 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 605 READ_ONCE(sk->sk_mark), sk->sk_uid); 606 } else { 607 ip6_sk_redirect(skb, sk); 608 } 609 goto out; 610 } 611 612 /* Tunnels don't have an application socket: don't pass errors back */ 613 if (tunnel) { 614 if (udp_sk(sk)->encap_err_rcv) 615 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, 616 ntohl(info), (u8 *)(uh+1)); 617 goto out; 618 } 619 620 if (!inet6_test_bit(RECVERR6, sk)) { 621 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 622 goto out; 623 } else { 624 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 625 } 626 627 sk->sk_err = err; 628 sk_error_report(sk); 629 out: 630 return 0; 631 } 632 633 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 634 { 635 int rc; 636 637 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 638 sock_rps_save_rxhash(sk, skb); 639 sk_mark_napi_id(sk, skb); 640 sk_incoming_cpu_update(sk); 641 } else { 642 sk_mark_napi_id_once(sk, skb); 643 } 644 645 rc = __udp_enqueue_schedule_skb(sk, skb); 646 if (rc < 0) { 647 int is_udplite = IS_UDPLITE(sk); 648 enum skb_drop_reason drop_reason; 649 650 /* Note that an ENOMEM error is charged twice */ 651 if (rc == -ENOMEM) { 652 UDP6_INC_STATS(sock_net(sk), 653 UDP_MIB_RCVBUFERRORS, is_udplite); 654 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 655 } else { 656 UDP6_INC_STATS(sock_net(sk), 657 UDP_MIB_MEMERRORS, is_udplite); 658 drop_reason = SKB_DROP_REASON_PROTO_MEM; 659 } 660 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 661 kfree_skb_reason(skb, drop_reason); 662 trace_udp_fail_queue_rcv_skb(rc, sk); 663 return -1; 664 } 665 666 return 0; 667 } 668 669 static __inline__ int udpv6_err(struct sk_buff *skb, 670 struct inet6_skb_parm *opt, u8 type, 671 u8 code, int offset, __be32 info) 672 { 673 return __udp6_lib_err(skb, opt, type, code, offset, info, 674 dev_net(skb->dev)->ipv4.udp_table); 675 } 676 677 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 678 { 679 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 680 struct udp_sock *up = udp_sk(sk); 681 int is_udplite = IS_UDPLITE(sk); 682 683 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 684 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 685 goto drop; 686 } 687 nf_reset_ct(skb); 688 689 if (static_branch_unlikely(&udpv6_encap_needed_key) && 690 READ_ONCE(up->encap_type)) { 691 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 692 693 /* 694 * This is an encapsulation socket so pass the skb to 695 * the socket's udp_encap_rcv() hook. Otherwise, just 696 * fall through and pass this up the UDP socket. 697 * up->encap_rcv() returns the following value: 698 * =0 if skb was successfully passed to the encap 699 * handler or was discarded by it. 700 * >0 if skb should be passed on to UDP. 701 * <0 if skb should be resubmitted as proto -N 702 */ 703 704 /* if we're overly short, let UDP handle it */ 705 encap_rcv = READ_ONCE(up->encap_rcv); 706 if (encap_rcv) { 707 int ret; 708 709 /* Verify checksum before giving to encap */ 710 if (udp_lib_checksum_complete(skb)) 711 goto csum_error; 712 713 ret = encap_rcv(sk, skb); 714 if (ret <= 0) { 715 __UDP6_INC_STATS(sock_net(sk), 716 UDP_MIB_INDATAGRAMS, 717 is_udplite); 718 return -ret; 719 } 720 } 721 722 /* FALLTHROUGH -- it's a UDP Packet */ 723 } 724 725 /* 726 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 727 */ 728 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) { 729 u16 pcrlen = READ_ONCE(up->pcrlen); 730 731 if (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 < pcrlen) { 737 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 738 UDP_SKB_CB(skb)->cscov, 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_get_no_check6_rx(sk)) 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_get_no_check6_rx(sk)) { 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_get_no_check6_rx(sk)) 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) 1102 return; 1103 1104 skb->sk = sk; 1105 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk)); 1106 skb->destructor = sock_pfree; 1107 dst = rcu_dereference(sk->sk_rx_dst); 1108 1109 if (dst) 1110 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1111 if (dst) { 1112 /* set noref for now. 1113 * any place which wants to hold dst has to call 1114 * dst_hold_safe() 1115 */ 1116 skb_dst_set_noref(skb, dst); 1117 } 1118 } 1119 1120 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1121 { 1122 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 1123 } 1124 1125 /* 1126 * Throw away all pending data and cancel the corking. Socket is locked. 1127 */ 1128 static void udp_v6_flush_pending_frames(struct sock *sk) 1129 { 1130 struct udp_sock *up = udp_sk(sk); 1131 1132 if (up->pending == AF_INET) 1133 udp_flush_pending_frames(sk); 1134 else if (up->pending) { 1135 up->len = 0; 1136 WRITE_ONCE(up->pending, 0); 1137 ip6_flush_pending_frames(sk); 1138 } 1139 } 1140 1141 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1142 int addr_len) 1143 { 1144 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1145 return -EINVAL; 1146 /* The following checks are replicated from __ip6_datagram_connect() 1147 * and intended to prevent BPF program called below from accessing 1148 * bytes that are out of the bound specified by user in addr_len. 1149 */ 1150 if (uaddr->sa_family == AF_INET) { 1151 if (ipv6_only_sock(sk)) 1152 return -EAFNOSUPPORT; 1153 return udp_pre_connect(sk, uaddr, addr_len); 1154 } 1155 1156 if (addr_len < SIN6_LEN_RFC2133) 1157 return -EINVAL; 1158 1159 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len); 1160 } 1161 1162 /** 1163 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1164 * @sk: socket we are sending on 1165 * @skb: sk_buff containing the filled-in UDP header 1166 * (checksum field must be zeroed out) 1167 * @saddr: source address 1168 * @daddr: destination address 1169 * @len: length of packet 1170 */ 1171 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1172 const struct in6_addr *saddr, 1173 const struct in6_addr *daddr, int len) 1174 { 1175 unsigned int offset; 1176 struct udphdr *uh = udp_hdr(skb); 1177 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1178 __wsum csum = 0; 1179 1180 if (!frags) { 1181 /* Only one fragment on the socket. */ 1182 skb->csum_start = skb_transport_header(skb) - skb->head; 1183 skb->csum_offset = offsetof(struct udphdr, check); 1184 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1185 } else { 1186 /* 1187 * HW-checksum won't work as there are two or more 1188 * fragments on the socket so that all csums of sk_buffs 1189 * should be together 1190 */ 1191 offset = skb_transport_offset(skb); 1192 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1193 csum = skb->csum; 1194 1195 skb->ip_summed = CHECKSUM_NONE; 1196 1197 do { 1198 csum = csum_add(csum, frags->csum); 1199 } while ((frags = frags->next)); 1200 1201 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1202 csum); 1203 if (uh->check == 0) 1204 uh->check = CSUM_MANGLED_0; 1205 } 1206 } 1207 1208 /* 1209 * Sending 1210 */ 1211 1212 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1213 struct inet_cork *cork) 1214 { 1215 struct sock *sk = skb->sk; 1216 struct udphdr *uh; 1217 int err = 0; 1218 int is_udplite = IS_UDPLITE(sk); 1219 __wsum csum = 0; 1220 int offset = skb_transport_offset(skb); 1221 int len = skb->len - offset; 1222 int datalen = len - sizeof(*uh); 1223 1224 /* 1225 * Create a UDP header 1226 */ 1227 uh = udp_hdr(skb); 1228 uh->source = fl6->fl6_sport; 1229 uh->dest = fl6->fl6_dport; 1230 uh->len = htons(len); 1231 uh->check = 0; 1232 1233 if (cork->gso_size) { 1234 const int hlen = skb_network_header_len(skb) + 1235 sizeof(struct udphdr); 1236 1237 if (hlen + cork->gso_size > cork->fragsize) { 1238 kfree_skb(skb); 1239 return -EINVAL; 1240 } 1241 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1242 kfree_skb(skb); 1243 return -EINVAL; 1244 } 1245 if (udp_get_no_check6_tx(sk)) { 1246 kfree_skb(skb); 1247 return -EINVAL; 1248 } 1249 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1250 dst_xfrm(skb_dst(skb))) { 1251 kfree_skb(skb); 1252 return -EIO; 1253 } 1254 1255 if (datalen > cork->gso_size) { 1256 skb_shinfo(skb)->gso_size = cork->gso_size; 1257 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1258 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1259 cork->gso_size); 1260 } 1261 goto csum_partial; 1262 } 1263 1264 if (is_udplite) 1265 csum = udplite_csum(skb); 1266 else if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */ 1267 skb->ip_summed = CHECKSUM_NONE; 1268 goto send; 1269 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1270 csum_partial: 1271 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1272 goto send; 1273 } else 1274 csum = udp_csum(skb); 1275 1276 /* add protocol-dependent pseudo-header */ 1277 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1278 len, fl6->flowi6_proto, csum); 1279 if (uh->check == 0) 1280 uh->check = CSUM_MANGLED_0; 1281 1282 send: 1283 err = ip6_send_skb(skb); 1284 if (err) { 1285 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) { 1286 UDP6_INC_STATS(sock_net(sk), 1287 UDP_MIB_SNDBUFERRORS, is_udplite); 1288 err = 0; 1289 } 1290 } else { 1291 UDP6_INC_STATS(sock_net(sk), 1292 UDP_MIB_OUTDATAGRAMS, is_udplite); 1293 } 1294 return err; 1295 } 1296 1297 static int udp_v6_push_pending_frames(struct sock *sk) 1298 { 1299 struct sk_buff *skb; 1300 struct udp_sock *up = udp_sk(sk); 1301 int err = 0; 1302 1303 if (up->pending == AF_INET) 1304 return udp_push_pending_frames(sk); 1305 1306 skb = ip6_finish_skb(sk); 1307 if (!skb) 1308 goto out; 1309 1310 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6, 1311 &inet_sk(sk)->cork.base); 1312 out: 1313 up->len = 0; 1314 WRITE_ONCE(up->pending, 0); 1315 return err; 1316 } 1317 1318 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1319 { 1320 struct ipv6_txoptions opt_space; 1321 struct udp_sock *up = udp_sk(sk); 1322 struct inet_sock *inet = inet_sk(sk); 1323 struct ipv6_pinfo *np = inet6_sk(sk); 1324 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1325 struct in6_addr *daddr, *final_p, final; 1326 struct ipv6_txoptions *opt = NULL; 1327 struct ipv6_txoptions *opt_to_free = NULL; 1328 struct ip6_flowlabel *flowlabel = NULL; 1329 struct inet_cork_full cork; 1330 struct flowi6 *fl6 = &cork.fl.u.ip6; 1331 struct dst_entry *dst; 1332 struct ipcm6_cookie ipc6; 1333 int addr_len = msg->msg_namelen; 1334 bool connected = false; 1335 int ulen = len; 1336 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE; 1337 int err; 1338 int is_udplite = IS_UDPLITE(sk); 1339 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1340 1341 ipcm6_init(&ipc6); 1342 ipc6.gso_size = READ_ONCE(up->gso_size); 1343 ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags); 1344 ipc6.sockc.mark = READ_ONCE(sk->sk_mark); 1345 1346 /* destination address check */ 1347 if (sin6) { 1348 if (addr_len < offsetof(struct sockaddr, sa_data)) 1349 return -EINVAL; 1350 1351 switch (sin6->sin6_family) { 1352 case AF_INET6: 1353 if (addr_len < SIN6_LEN_RFC2133) 1354 return -EINVAL; 1355 daddr = &sin6->sin6_addr; 1356 if (ipv6_addr_any(daddr) && 1357 ipv6_addr_v4mapped(&np->saddr)) 1358 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1359 daddr); 1360 break; 1361 case AF_INET: 1362 goto do_udp_sendmsg; 1363 case AF_UNSPEC: 1364 msg->msg_name = sin6 = NULL; 1365 msg->msg_namelen = addr_len = 0; 1366 daddr = NULL; 1367 break; 1368 default: 1369 return -EINVAL; 1370 } 1371 } else if (!READ_ONCE(up->pending)) { 1372 if (sk->sk_state != TCP_ESTABLISHED) 1373 return -EDESTADDRREQ; 1374 daddr = &sk->sk_v6_daddr; 1375 } else 1376 daddr = NULL; 1377 1378 if (daddr) { 1379 if (ipv6_addr_v4mapped(daddr)) { 1380 struct sockaddr_in sin; 1381 sin.sin_family = AF_INET; 1382 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1383 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1384 msg->msg_name = &sin; 1385 msg->msg_namelen = sizeof(sin); 1386 do_udp_sendmsg: 1387 err = ipv6_only_sock(sk) ? 1388 -ENETUNREACH : udp_sendmsg(sk, msg, len); 1389 msg->msg_name = sin6; 1390 msg->msg_namelen = addr_len; 1391 return err; 1392 } 1393 } 1394 1395 /* Rough check on arithmetic overflow, 1396 better check is made in ip6_append_data(). 1397 */ 1398 if (len > INT_MAX - sizeof(struct udphdr)) 1399 return -EMSGSIZE; 1400 1401 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1402 if (READ_ONCE(up->pending)) { 1403 if (READ_ONCE(up->pending) == AF_INET) 1404 return udp_sendmsg(sk, msg, len); 1405 /* 1406 * There are pending frames. 1407 * The socket lock must be held while it's corked. 1408 */ 1409 lock_sock(sk); 1410 if (likely(up->pending)) { 1411 if (unlikely(up->pending != AF_INET6)) { 1412 release_sock(sk); 1413 return -EAFNOSUPPORT; 1414 } 1415 dst = NULL; 1416 goto do_append_data; 1417 } 1418 release_sock(sk); 1419 } 1420 ulen += sizeof(struct udphdr); 1421 1422 memset(fl6, 0, sizeof(*fl6)); 1423 1424 if (sin6) { 1425 if (sin6->sin6_port == 0) 1426 return -EINVAL; 1427 1428 fl6->fl6_dport = sin6->sin6_port; 1429 daddr = &sin6->sin6_addr; 1430 1431 if (inet6_test_bit(SNDFLOW, sk)) { 1432 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1433 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) { 1434 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1435 if (IS_ERR(flowlabel)) 1436 return -EINVAL; 1437 } 1438 } 1439 1440 /* 1441 * Otherwise it will be difficult to maintain 1442 * sk->sk_dst_cache. 1443 */ 1444 if (sk->sk_state == TCP_ESTABLISHED && 1445 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1446 daddr = &sk->sk_v6_daddr; 1447 1448 if (addr_len >= sizeof(struct sockaddr_in6) && 1449 sin6->sin6_scope_id && 1450 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1451 fl6->flowi6_oif = sin6->sin6_scope_id; 1452 } else { 1453 if (sk->sk_state != TCP_ESTABLISHED) 1454 return -EDESTADDRREQ; 1455 1456 fl6->fl6_dport = inet->inet_dport; 1457 daddr = &sk->sk_v6_daddr; 1458 fl6->flowlabel = np->flow_label; 1459 connected = true; 1460 } 1461 1462 if (!fl6->flowi6_oif) 1463 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if); 1464 1465 if (!fl6->flowi6_oif) 1466 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1467 1468 fl6->flowi6_uid = sk->sk_uid; 1469 1470 if (msg->msg_controllen) { 1471 opt = &opt_space; 1472 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1473 opt->tot_len = sizeof(*opt); 1474 ipc6.opt = opt; 1475 1476 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1477 if (err > 0) 1478 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6, 1479 &ipc6); 1480 if (err < 0) { 1481 fl6_sock_release(flowlabel); 1482 return err; 1483 } 1484 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1485 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1486 if (IS_ERR(flowlabel)) 1487 return -EINVAL; 1488 } 1489 if (!(opt->opt_nflen|opt->opt_flen)) 1490 opt = NULL; 1491 connected = false; 1492 } 1493 if (!opt) { 1494 opt = txopt_get(np); 1495 opt_to_free = opt; 1496 } 1497 if (flowlabel) 1498 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1499 opt = ipv6_fixup_options(&opt_space, opt); 1500 ipc6.opt = opt; 1501 1502 fl6->flowi6_proto = sk->sk_protocol; 1503 fl6->flowi6_mark = ipc6.sockc.mark; 1504 fl6->daddr = *daddr; 1505 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr)) 1506 fl6->saddr = np->saddr; 1507 fl6->fl6_sport = inet->inet_sport; 1508 1509 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1510 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1511 (struct sockaddr *)sin6, 1512 &addr_len, 1513 &fl6->saddr); 1514 if (err) 1515 goto out_no_dst; 1516 if (sin6) { 1517 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1518 /* BPF program rewrote IPv6-only by IPv4-mapped 1519 * IPv6. It's currently unsupported. 1520 */ 1521 err = -ENOTSUPP; 1522 goto out_no_dst; 1523 } 1524 if (sin6->sin6_port == 0) { 1525 /* BPF program set invalid port. Reject it. */ 1526 err = -EINVAL; 1527 goto out_no_dst; 1528 } 1529 fl6->fl6_dport = sin6->sin6_port; 1530 fl6->daddr = sin6->sin6_addr; 1531 } 1532 } 1533 1534 if (ipv6_addr_any(&fl6->daddr)) 1535 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1536 1537 final_p = fl6_update_dst(fl6, opt, &final); 1538 if (final_p) 1539 connected = false; 1540 1541 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) { 1542 fl6->flowi6_oif = READ_ONCE(np->mcast_oif); 1543 connected = false; 1544 } else if (!fl6->flowi6_oif) 1545 fl6->flowi6_oif = READ_ONCE(np->ucast_oif); 1546 1547 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6)); 1548 1549 if (ipc6.tclass < 0) 1550 ipc6.tclass = np->tclass; 1551 1552 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel); 1553 1554 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected); 1555 if (IS_ERR(dst)) { 1556 err = PTR_ERR(dst); 1557 dst = NULL; 1558 goto out; 1559 } 1560 1561 if (ipc6.hlimit < 0) 1562 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst); 1563 1564 if (msg->msg_flags&MSG_CONFIRM) 1565 goto do_confirm; 1566 back_from_confirm: 1567 1568 /* Lockless fast path for the non-corking case */ 1569 if (!corkreq) { 1570 struct sk_buff *skb; 1571 1572 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1573 sizeof(struct udphdr), &ipc6, 1574 (struct rt6_info *)dst, 1575 msg->msg_flags, &cork); 1576 err = PTR_ERR(skb); 1577 if (!IS_ERR_OR_NULL(skb)) 1578 err = udp_v6_send_skb(skb, fl6, &cork.base); 1579 /* ip6_make_skb steals dst reference */ 1580 goto out_no_dst; 1581 } 1582 1583 lock_sock(sk); 1584 if (unlikely(up->pending)) { 1585 /* The socket is already corked while preparing it. */ 1586 /* ... which is an evident application bug. --ANK */ 1587 release_sock(sk); 1588 1589 net_dbg_ratelimited("udp cork app bug 2\n"); 1590 err = -EINVAL; 1591 goto out; 1592 } 1593 1594 WRITE_ONCE(up->pending, AF_INET6); 1595 1596 do_append_data: 1597 if (ipc6.dontfrag < 0) 1598 ipc6.dontfrag = inet6_test_bit(DONTFRAG, sk); 1599 up->len += ulen; 1600 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1601 &ipc6, fl6, (struct rt6_info *)dst, 1602 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1603 if (err) 1604 udp_v6_flush_pending_frames(sk); 1605 else if (!corkreq) 1606 err = udp_v6_push_pending_frames(sk); 1607 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1608 WRITE_ONCE(up->pending, 0); 1609 1610 if (err > 0) 1611 err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0; 1612 release_sock(sk); 1613 1614 out: 1615 dst_release(dst); 1616 out_no_dst: 1617 fl6_sock_release(flowlabel); 1618 txopt_put(opt_to_free); 1619 if (!err) 1620 return len; 1621 /* 1622 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1623 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1624 * we don't have a good statistic (IpOutDiscards but it can be too many 1625 * things). We could add another new stat but at least for now that 1626 * seems like overkill. 1627 */ 1628 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1629 UDP6_INC_STATS(sock_net(sk), 1630 UDP_MIB_SNDBUFERRORS, is_udplite); 1631 } 1632 return err; 1633 1634 do_confirm: 1635 if (msg->msg_flags & MSG_PROBE) 1636 dst_confirm_neigh(dst, &fl6->daddr); 1637 if (!(msg->msg_flags&MSG_PROBE) || len) 1638 goto back_from_confirm; 1639 err = 0; 1640 goto out; 1641 } 1642 EXPORT_SYMBOL(udpv6_sendmsg); 1643 1644 static void udpv6_splice_eof(struct socket *sock) 1645 { 1646 struct sock *sk = sock->sk; 1647 struct udp_sock *up = udp_sk(sk); 1648 1649 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk)) 1650 return; 1651 1652 lock_sock(sk); 1653 if (up->pending && !udp_test_bit(CORK, sk)) 1654 udp_v6_push_pending_frames(sk); 1655 release_sock(sk); 1656 } 1657 1658 void udpv6_destroy_sock(struct sock *sk) 1659 { 1660 struct udp_sock *up = udp_sk(sk); 1661 lock_sock(sk); 1662 1663 /* protects from races with udp_abort() */ 1664 sock_set_flag(sk, SOCK_DEAD); 1665 udp_v6_flush_pending_frames(sk); 1666 release_sock(sk); 1667 1668 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1669 if (up->encap_type) { 1670 void (*encap_destroy)(struct sock *sk); 1671 encap_destroy = READ_ONCE(up->encap_destroy); 1672 if (encap_destroy) 1673 encap_destroy(sk); 1674 } 1675 if (udp_test_bit(ENCAP_ENABLED, sk)) { 1676 static_branch_dec(&udpv6_encap_needed_key); 1677 udp_encap_disable(); 1678 } 1679 } 1680 } 1681 1682 /* 1683 * Socket option code for UDP 1684 */ 1685 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1686 unsigned int optlen) 1687 { 1688 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 1689 return udp_lib_setsockopt(sk, level, optname, 1690 optval, optlen, 1691 udp_v6_push_pending_frames); 1692 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1693 } 1694 1695 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1696 char __user *optval, int __user *optlen) 1697 { 1698 if (level == SOL_UDP || level == SOL_UDPLITE) 1699 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1700 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1701 } 1702 1703 1704 /* ------------------------------------------------------------------------ */ 1705 #ifdef CONFIG_PROC_FS 1706 int udp6_seq_show(struct seq_file *seq, void *v) 1707 { 1708 if (v == SEQ_START_TOKEN) { 1709 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1710 } else { 1711 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1712 const struct inet_sock *inet = inet_sk((const struct sock *)v); 1713 __u16 srcp = ntohs(inet->inet_sport); 1714 __u16 destp = ntohs(inet->inet_dport); 1715 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1716 udp_rqueue_get(v), bucket); 1717 } 1718 return 0; 1719 } 1720 1721 const struct seq_operations udp6_seq_ops = { 1722 .start = udp_seq_start, 1723 .next = udp_seq_next, 1724 .stop = udp_seq_stop, 1725 .show = udp6_seq_show, 1726 }; 1727 EXPORT_SYMBOL(udp6_seq_ops); 1728 1729 static struct udp_seq_afinfo udp6_seq_afinfo = { 1730 .family = AF_INET6, 1731 .udp_table = NULL, 1732 }; 1733 1734 int __net_init udp6_proc_init(struct net *net) 1735 { 1736 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1737 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1738 return -ENOMEM; 1739 return 0; 1740 } 1741 1742 void udp6_proc_exit(struct net *net) 1743 { 1744 remove_proc_entry("udp6", net->proc_net); 1745 } 1746 #endif /* CONFIG_PROC_FS */ 1747 1748 /* ------------------------------------------------------------------------ */ 1749 1750 struct proto udpv6_prot = { 1751 .name = "UDPv6", 1752 .owner = THIS_MODULE, 1753 .close = udp_lib_close, 1754 .pre_connect = udpv6_pre_connect, 1755 .connect = ip6_datagram_connect, 1756 .disconnect = udp_disconnect, 1757 .ioctl = udp_ioctl, 1758 .init = udpv6_init_sock, 1759 .destroy = udpv6_destroy_sock, 1760 .setsockopt = udpv6_setsockopt, 1761 .getsockopt = udpv6_getsockopt, 1762 .sendmsg = udpv6_sendmsg, 1763 .recvmsg = udpv6_recvmsg, 1764 .splice_eof = udpv6_splice_eof, 1765 .release_cb = ip6_datagram_release_cb, 1766 .hash = udp_lib_hash, 1767 .unhash = udp_lib_unhash, 1768 .rehash = udp_v6_rehash, 1769 .get_port = udp_v6_get_port, 1770 .put_port = udp_lib_unhash, 1771 #ifdef CONFIG_BPF_SYSCALL 1772 .psock_update_sk_prot = udp_bpf_update_proto, 1773 #endif 1774 1775 .memory_allocated = &udp_memory_allocated, 1776 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 1777 1778 .sysctl_mem = sysctl_udp_mem, 1779 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1780 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1781 .obj_size = sizeof(struct udp6_sock), 1782 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6), 1783 .h.udp_table = NULL, 1784 .diag_destroy = udp_abort, 1785 }; 1786 1787 static struct inet_protosw udpv6_protosw = { 1788 .type = SOCK_DGRAM, 1789 .protocol = IPPROTO_UDP, 1790 .prot = &udpv6_prot, 1791 .ops = &inet6_dgram_ops, 1792 .flags = INET_PROTOSW_PERMANENT, 1793 }; 1794 1795 int __init udpv6_init(void) 1796 { 1797 int ret; 1798 1799 net_hotdata.udpv6_protocol = (struct inet6_protocol) { 1800 .handler = udpv6_rcv, 1801 .err_handler = udpv6_err, 1802 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL, 1803 }; 1804 ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1805 if (ret) 1806 goto out; 1807 1808 ret = inet6_register_protosw(&udpv6_protosw); 1809 if (ret) 1810 goto out_udpv6_protocol; 1811 out: 1812 return ret; 1813 1814 out_udpv6_protocol: 1815 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1816 goto out; 1817 } 1818 1819 void udpv6_exit(void) 1820 { 1821 inet6_unregister_protosw(&udpv6_protosw); 1822 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1823 } 1824