1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * UDP over IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on linux/ipv4/udp.c 10 * 11 * Fixes: 12 * Hideaki YOSHIFUJI : sin6_scope_id support 13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 15 * a single port at the same time. 16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. 18 */ 19 20 #include <linux/bpf-cgroup.h> 21 #include <linux/errno.h> 22 #include <linux/types.h> 23 #include <linux/socket.h> 24 #include <linux/sockios.h> 25 #include <linux/net.h> 26 #include <linux/in6.h> 27 #include <linux/netdevice.h> 28 #include <linux/if_arp.h> 29 #include <linux/ipv6.h> 30 #include <linux/icmpv6.h> 31 #include <linux/init.h> 32 #include <linux/module.h> 33 #include <linux/skbuff.h> 34 #include <linux/slab.h> 35 #include <linux/uaccess.h> 36 #include <linux/indirect_call_wrapper.h> 37 #include <trace/events/udp.h> 38 39 #include <net/addrconf.h> 40 #include <net/ndisc.h> 41 #include <net/protocol.h> 42 #include <net/transp_v6.h> 43 #include <net/ip6_route.h> 44 #include <net/raw.h> 45 #include <net/seg6.h> 46 #include <net/tcp_states.h> 47 #include <net/ip6_checksum.h> 48 #include <net/ip6_tunnel.h> 49 #include <trace/events/udp.h> 50 #include <net/xfrm.h> 51 #include <net/inet_hashtables.h> 52 #include <net/inet6_hashtables.h> 53 #include <net/busy_poll.h> 54 #include <net/sock_reuseport.h> 55 #include <net/gro.h> 56 57 #include <linux/proc_fs.h> 58 #include <linux/seq_file.h> 59 #include <trace/events/skb.h> 60 #include "udp_impl.h" 61 62 static void udpv6_destruct_sock(struct sock *sk) 63 { 64 udp_destruct_common(sk); 65 inet6_sock_destruct(sk); 66 } 67 68 int udpv6_init_sock(struct sock *sk) 69 { 70 udp_lib_init_sock(sk); 71 sk->sk_destruct = udpv6_destruct_sock; 72 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 73 return 0; 74 } 75 76 INDIRECT_CALLABLE_SCOPE 77 u32 udp6_ehashfn(const struct net *net, 78 const struct in6_addr *laddr, 79 const u16 lport, 80 const struct in6_addr *faddr, 81 const __be16 fport) 82 { 83 u32 lhash, fhash; 84 85 net_get_random_once(&udp6_ehash_secret, 86 sizeof(udp6_ehash_secret)); 87 net_get_random_once(&udp_ipv6_hash_secret, 88 sizeof(udp_ipv6_hash_secret)); 89 90 lhash = (__force u32)laddr->s6_addr32[3]; 91 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 92 93 return __inet6_ehashfn(lhash, lport, fhash, fport, 94 udp6_ehash_secret + net_hash_mix(net)); 95 } 96 97 int udp_v6_get_port(struct sock *sk, unsigned short snum) 98 { 99 unsigned int hash2_nulladdr = 100 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 101 unsigned int hash2_partial = 102 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 103 104 /* precompute partial secondary hash */ 105 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 106 return udp_lib_get_port(sk, snum, hash2_nulladdr); 107 } 108 109 void udp_v6_rehash(struct sock *sk) 110 { 111 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 112 &sk->sk_v6_rcv_saddr, 113 inet_sk(sk)->inet_num); 114 115 udp_lib_rehash(sk, new_hash); 116 } 117 118 static int compute_score(struct sock *sk, struct net *net, 119 const struct in6_addr *saddr, __be16 sport, 120 const struct in6_addr *daddr, unsigned short hnum, 121 int dif, int sdif) 122 { 123 int bound_dev_if, score; 124 struct inet_sock *inet; 125 bool dev_match; 126 127 if (!net_eq(sock_net(sk), net) || 128 udp_sk(sk)->udp_port_hash != hnum || 129 sk->sk_family != PF_INET6) 130 return -1; 131 132 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 133 return -1; 134 135 score = 0; 136 inet = inet_sk(sk); 137 138 if (inet->inet_dport) { 139 if (inet->inet_dport != sport) 140 return -1; 141 score++; 142 } 143 144 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 145 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 146 return -1; 147 score++; 148 } 149 150 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 151 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif); 152 if (!dev_match) 153 return -1; 154 if (bound_dev_if) 155 score++; 156 157 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 158 score++; 159 160 return score; 161 } 162 163 /* called with rcu_read_lock() */ 164 static struct sock *udp6_lib_lookup2(struct net *net, 165 const struct in6_addr *saddr, __be16 sport, 166 const struct in6_addr *daddr, unsigned int hnum, 167 int dif, int sdif, struct udp_hslot *hslot2, 168 struct sk_buff *skb) 169 { 170 struct sock *sk, *result; 171 int score, badness; 172 bool need_rescore; 173 174 result = NULL; 175 badness = -1; 176 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 177 need_rescore = false; 178 rescore: 179 score = compute_score(need_rescore ? result : sk, net, saddr, 180 sport, daddr, hnum, dif, sdif); 181 if (score > badness) { 182 badness = score; 183 184 if (need_rescore) 185 continue; 186 187 if (sk->sk_state == TCP_ESTABLISHED) { 188 result = sk; 189 continue; 190 } 191 192 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 193 saddr, sport, daddr, hnum, udp6_ehashfn); 194 if (!result) { 195 result = sk; 196 continue; 197 } 198 199 /* Fall back to scoring if group has connections */ 200 if (!reuseport_has_conns(sk)) 201 return result; 202 203 /* Reuseport logic returned an error, keep original score. */ 204 if (IS_ERR(result)) 205 continue; 206 207 /* compute_score is too long of a function to be 208 * inlined, and calling it again here yields 209 * measureable overhead for some 210 * workloads. Work around it by jumping 211 * backwards to rescore 'result'. 212 */ 213 need_rescore = true; 214 goto rescore; 215 } 216 } 217 return result; 218 } 219 220 /* rcu_read_lock() must be held */ 221 struct sock *__udp6_lib_lookup(struct net *net, 222 const struct in6_addr *saddr, __be16 sport, 223 const struct in6_addr *daddr, __be16 dport, 224 int dif, int sdif, struct udp_table *udptable, 225 struct sk_buff *skb) 226 { 227 unsigned short hnum = ntohs(dport); 228 unsigned int hash2, slot2; 229 struct udp_hslot *hslot2; 230 struct sock *result, *sk; 231 232 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 233 slot2 = hash2 & udptable->mask; 234 hslot2 = &udptable->hash2[slot2]; 235 236 /* Lookup connected or non-wildcard sockets */ 237 result = udp6_lib_lookup2(net, saddr, sport, 238 daddr, hnum, dif, sdif, 239 hslot2, skb); 240 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 241 goto done; 242 243 /* Lookup redirect from BPF */ 244 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 245 udptable == net->ipv4.udp_table) { 246 sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 247 saddr, sport, daddr, hnum, dif, 248 udp6_ehashfn); 249 if (sk) { 250 result = sk; 251 goto done; 252 } 253 } 254 255 /* Got non-wildcard socket or error on first lookup */ 256 if (result) 257 goto done; 258 259 /* Lookup wildcard sockets */ 260 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 261 slot2 = hash2 & udptable->mask; 262 hslot2 = &udptable->hash2[slot2]; 263 264 result = udp6_lib_lookup2(net, saddr, sport, 265 &in6addr_any, hnum, dif, sdif, 266 hslot2, skb); 267 done: 268 if (IS_ERR(result)) 269 return NULL; 270 return result; 271 } 272 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 273 274 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 275 __be16 sport, __be16 dport, 276 struct udp_table *udptable) 277 { 278 const struct ipv6hdr *iph = ipv6_hdr(skb); 279 280 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 281 &iph->daddr, dport, inet6_iif(skb), 282 inet6_sdif(skb), udptable, skb); 283 } 284 285 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, 286 __be16 sport, __be16 dport) 287 { 288 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation]; 289 const struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + offset); 290 struct net *net = dev_net(skb->dev); 291 int iif, sdif; 292 293 inet6_get_iif_sdif(skb, &iif, &sdif); 294 295 return __udp6_lib_lookup(net, &iph->saddr, sport, 296 &iph->daddr, dport, iif, 297 sdif, net->ipv4.udp_table, NULL); 298 } 299 300 /* Must be called under rcu_read_lock(). 301 * Does increment socket refcount. 302 */ 303 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 304 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 305 const struct in6_addr *daddr, __be16 dport, int dif) 306 { 307 struct sock *sk; 308 309 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 310 dif, 0, net->ipv4.udp_table, NULL); 311 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 312 sk = NULL; 313 return sk; 314 } 315 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 316 #endif 317 318 /* do not use the scratch area len for jumbogram: their length execeeds the 319 * scratch area space; note that the IP6CB flags is still in the first 320 * cacheline, so checking for jumbograms is cheap 321 */ 322 static int udp6_skb_len(struct sk_buff *skb) 323 { 324 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 325 } 326 327 /* 328 * This should be easy, if there is something there we 329 * return it, otherwise we block. 330 */ 331 332 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 333 int flags, int *addr_len) 334 { 335 struct ipv6_pinfo *np = inet6_sk(sk); 336 struct inet_sock *inet = inet_sk(sk); 337 struct sk_buff *skb; 338 unsigned int ulen, copied; 339 int off, err, peeking = flags & MSG_PEEK; 340 int is_udplite = IS_UDPLITE(sk); 341 struct udp_mib __percpu *mib; 342 bool checksum_valid = false; 343 int is_udp4; 344 345 if (flags & MSG_ERRQUEUE) 346 return ipv6_recv_error(sk, msg, len, addr_len); 347 348 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 349 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 350 351 try_again: 352 off = sk_peek_offset(sk, flags); 353 skb = __skb_recv_udp(sk, flags, &off, &err); 354 if (!skb) 355 return err; 356 357 ulen = udp6_skb_len(skb); 358 copied = len; 359 if (copied > ulen - off) 360 copied = ulen - off; 361 else if (copied < ulen) 362 msg->msg_flags |= MSG_TRUNC; 363 364 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 365 mib = __UDPX_MIB(sk, is_udp4); 366 367 /* 368 * If checksum is needed at all, try to do it while copying the 369 * data. If the data is truncated, or if we only want a partial 370 * coverage checksum (UDP-Lite), do it before the copy. 371 */ 372 373 if (copied < ulen || peeking || 374 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 375 checksum_valid = udp_skb_csum_unnecessary(skb) || 376 !__udp_lib_checksum_complete(skb); 377 if (!checksum_valid) 378 goto csum_copy_err; 379 } 380 381 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 382 if (udp_skb_is_linear(skb)) 383 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 384 else 385 err = skb_copy_datagram_msg(skb, off, msg, copied); 386 } else { 387 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 388 if (err == -EINVAL) 389 goto csum_copy_err; 390 } 391 if (unlikely(err)) { 392 if (!peeking) { 393 atomic_inc(&sk->sk_drops); 394 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 395 } 396 kfree_skb(skb); 397 return err; 398 } 399 if (!peeking) 400 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 401 402 sock_recv_cmsgs(msg, sk, skb); 403 404 /* Copy the address. */ 405 if (msg->msg_name) { 406 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 407 sin6->sin6_family = AF_INET6; 408 sin6->sin6_port = udp_hdr(skb)->source; 409 sin6->sin6_flowinfo = 0; 410 411 if (is_udp4) { 412 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 413 &sin6->sin6_addr); 414 sin6->sin6_scope_id = 0; 415 } else { 416 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 417 sin6->sin6_scope_id = 418 ipv6_iface_scope_id(&sin6->sin6_addr, 419 inet6_iif(skb)); 420 } 421 *addr_len = sizeof(*sin6); 422 423 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 424 (struct sockaddr *)sin6, 425 addr_len); 426 } 427 428 if (udp_test_bit(GRO_ENABLED, sk)) 429 udp_cmsg_recv(msg, sk, skb); 430 431 if (np->rxopt.all) 432 ip6_datagram_recv_common_ctl(sk, msg, skb); 433 434 if (is_udp4) { 435 if (inet_cmsg_flags(inet)) 436 ip_cmsg_recv_offset(msg, sk, skb, 437 sizeof(struct udphdr), off); 438 } else { 439 if (np->rxopt.all) 440 ip6_datagram_recv_specific_ctl(sk, msg, skb); 441 } 442 443 err = copied; 444 if (flags & MSG_TRUNC) 445 err = ulen; 446 447 skb_consume_udp(sk, skb, peeking ? -err : err); 448 return err; 449 450 csum_copy_err: 451 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 452 udp_skb_destructor)) { 453 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 454 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 455 } 456 kfree_skb(skb); 457 458 /* starting over for a new packet, but check if we need to yield */ 459 cond_resched(); 460 msg->msg_flags &= ~MSG_TRUNC; 461 goto try_again; 462 } 463 464 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 465 void udpv6_encap_enable(void) 466 { 467 static_branch_inc(&udpv6_encap_needed_key); 468 } 469 EXPORT_SYMBOL(udpv6_encap_enable); 470 471 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 472 * through error handlers in encapsulations looking for a match. 473 */ 474 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 475 struct inet6_skb_parm *opt, 476 u8 type, u8 code, int offset, __be32 info) 477 { 478 int i; 479 480 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 481 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 482 u8 type, u8 code, int offset, __be32 info); 483 const struct ip6_tnl_encap_ops *encap; 484 485 encap = rcu_dereference(ip6tun_encaps[i]); 486 if (!encap) 487 continue; 488 handler = encap->err_handler; 489 if (handler && !handler(skb, opt, type, code, offset, info)) 490 return 0; 491 } 492 493 return -ENOENT; 494 } 495 496 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 497 * reversing source and destination port: this will match tunnels that force the 498 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 499 * lwtunnels might actually break this assumption by being configured with 500 * different destination ports on endpoints, in this case we won't be able to 501 * trace ICMP messages back to them. 502 * 503 * If this doesn't match any socket, probe tunnels with arbitrary destination 504 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 505 * we've sent packets to won't necessarily match the local destination port. 506 * 507 * Then ask the tunnel implementation to match the error against a valid 508 * association. 509 * 510 * Return an error if we can't find a match, the socket if we need further 511 * processing, zero otherwise. 512 */ 513 static struct sock *__udp6_lib_err_encap(struct net *net, 514 const struct ipv6hdr *hdr, int offset, 515 struct udphdr *uh, 516 struct udp_table *udptable, 517 struct sock *sk, 518 struct sk_buff *skb, 519 struct inet6_skb_parm *opt, 520 u8 type, u8 code, __be32 info) 521 { 522 int (*lookup)(struct sock *sk, struct sk_buff *skb); 523 int network_offset, transport_offset; 524 struct udp_sock *up; 525 526 network_offset = skb_network_offset(skb); 527 transport_offset = skb_transport_offset(skb); 528 529 /* Network header needs to point to the outer IPv6 header inside ICMP */ 530 skb_reset_network_header(skb); 531 532 /* Transport header needs to point to the UDP header */ 533 skb_set_transport_header(skb, offset); 534 535 if (sk) { 536 up = udp_sk(sk); 537 538 lookup = READ_ONCE(up->encap_err_lookup); 539 if (lookup && lookup(sk, skb)) 540 sk = NULL; 541 542 goto out; 543 } 544 545 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 546 &hdr->saddr, uh->dest, 547 inet6_iif(skb), 0, udptable, skb); 548 if (sk) { 549 up = udp_sk(sk); 550 551 lookup = READ_ONCE(up->encap_err_lookup); 552 if (!lookup || lookup(sk, skb)) 553 sk = NULL; 554 } 555 556 out: 557 if (!sk) { 558 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 559 offset, info)); 560 } 561 562 skb_set_transport_header(skb, transport_offset); 563 skb_set_network_header(skb, network_offset); 564 565 return sk; 566 } 567 568 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 569 u8 type, u8 code, int offset, __be32 info, 570 struct udp_table *udptable) 571 { 572 struct ipv6_pinfo *np; 573 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 574 const struct in6_addr *saddr = &hdr->saddr; 575 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr; 576 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 577 bool tunnel = false; 578 struct sock *sk; 579 int harderr; 580 int err; 581 struct net *net = dev_net(skb->dev); 582 583 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 584 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 585 586 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) { 587 /* No socket for error: try tunnels before discarding */ 588 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 589 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 590 udptable, sk, skb, 591 opt, type, code, info); 592 if (!sk) 593 return 0; 594 } else 595 sk = ERR_PTR(-ENOENT); 596 597 if (IS_ERR(sk)) { 598 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 599 ICMP6_MIB_INERRORS); 600 return PTR_ERR(sk); 601 } 602 603 tunnel = true; 604 } 605 606 harderr = icmpv6_err_convert(type, code, &err); 607 np = inet6_sk(sk); 608 609 if (type == ICMPV6_PKT_TOOBIG) { 610 if (!ip6_sk_accept_pmtu(sk)) 611 goto out; 612 ip6_sk_update_pmtu(skb, sk, info); 613 if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT) 614 harderr = 1; 615 } 616 if (type == NDISC_REDIRECT) { 617 if (tunnel) { 618 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 619 READ_ONCE(sk->sk_mark), sk->sk_uid); 620 } else { 621 ip6_sk_redirect(skb, sk); 622 } 623 goto out; 624 } 625 626 /* Tunnels don't have an application socket: don't pass errors back */ 627 if (tunnel) { 628 if (udp_sk(sk)->encap_err_rcv) 629 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, 630 ntohl(info), (u8 *)(uh+1)); 631 goto out; 632 } 633 634 if (!inet6_test_bit(RECVERR6, sk)) { 635 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 636 goto out; 637 } else { 638 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 639 } 640 641 sk->sk_err = err; 642 sk_error_report(sk); 643 out: 644 return 0; 645 } 646 647 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 648 { 649 int rc; 650 651 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 652 sock_rps_save_rxhash(sk, skb); 653 sk_mark_napi_id(sk, skb); 654 sk_incoming_cpu_update(sk); 655 } else { 656 sk_mark_napi_id_once(sk, skb); 657 } 658 659 rc = __udp_enqueue_schedule_skb(sk, skb); 660 if (rc < 0) { 661 int is_udplite = IS_UDPLITE(sk); 662 enum skb_drop_reason drop_reason; 663 664 /* Note that an ENOMEM error is charged twice */ 665 if (rc == -ENOMEM) { 666 UDP6_INC_STATS(sock_net(sk), 667 UDP_MIB_RCVBUFERRORS, is_udplite); 668 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 669 } else { 670 UDP6_INC_STATS(sock_net(sk), 671 UDP_MIB_MEMERRORS, is_udplite); 672 drop_reason = SKB_DROP_REASON_PROTO_MEM; 673 } 674 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 675 trace_udp_fail_queue_rcv_skb(rc, sk, skb); 676 sk_skb_reason_drop(sk, skb, drop_reason); 677 return -1; 678 } 679 680 return 0; 681 } 682 683 static __inline__ int udpv6_err(struct sk_buff *skb, 684 struct inet6_skb_parm *opt, u8 type, 685 u8 code, int offset, __be32 info) 686 { 687 return __udp6_lib_err(skb, opt, type, code, offset, info, 688 dev_net(skb->dev)->ipv4.udp_table); 689 } 690 691 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 692 { 693 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 694 struct udp_sock *up = udp_sk(sk); 695 int is_udplite = IS_UDPLITE(sk); 696 697 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 698 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 699 goto drop; 700 } 701 nf_reset_ct(skb); 702 703 if (static_branch_unlikely(&udpv6_encap_needed_key) && 704 READ_ONCE(up->encap_type)) { 705 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 706 707 /* 708 * This is an encapsulation socket so pass the skb to 709 * the socket's udp_encap_rcv() hook. Otherwise, just 710 * fall through and pass this up the UDP socket. 711 * up->encap_rcv() returns the following value: 712 * =0 if skb was successfully passed to the encap 713 * handler or was discarded by it. 714 * >0 if skb should be passed on to UDP. 715 * <0 if skb should be resubmitted as proto -N 716 */ 717 718 /* if we're overly short, let UDP handle it */ 719 encap_rcv = READ_ONCE(up->encap_rcv); 720 if (encap_rcv) { 721 int ret; 722 723 /* Verify checksum before giving to encap */ 724 if (udp_lib_checksum_complete(skb)) 725 goto csum_error; 726 727 ret = encap_rcv(sk, skb); 728 if (ret <= 0) { 729 __UDP6_INC_STATS(sock_net(sk), 730 UDP_MIB_INDATAGRAMS, 731 is_udplite); 732 return -ret; 733 } 734 } 735 736 /* FALLTHROUGH -- it's a UDP Packet */ 737 } 738 739 /* 740 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 741 */ 742 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) { 743 u16 pcrlen = READ_ONCE(up->pcrlen); 744 745 if (pcrlen == 0) { /* full coverage was set */ 746 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 747 UDP_SKB_CB(skb)->cscov, skb->len); 748 goto drop; 749 } 750 if (UDP_SKB_CB(skb)->cscov < pcrlen) { 751 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 752 UDP_SKB_CB(skb)->cscov, pcrlen); 753 goto drop; 754 } 755 } 756 757 prefetch(&sk->sk_rmem_alloc); 758 if (rcu_access_pointer(sk->sk_filter) && 759 udp_lib_checksum_complete(skb)) 760 goto csum_error; 761 762 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 763 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 764 goto drop; 765 } 766 767 udp_csum_pull_header(skb); 768 769 skb_dst_drop(skb); 770 771 return __udpv6_queue_rcv_skb(sk, skb); 772 773 csum_error: 774 drop_reason = SKB_DROP_REASON_UDP_CSUM; 775 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 776 drop: 777 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 778 atomic_inc(&sk->sk_drops); 779 sk_skb_reason_drop(sk, skb, drop_reason); 780 return -1; 781 } 782 783 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 784 { 785 struct sk_buff *next, *segs; 786 int ret; 787 788 if (likely(!udp_unexpected_gso(sk, skb))) 789 return udpv6_queue_rcv_one_skb(sk, skb); 790 791 __skb_push(skb, -skb_mac_offset(skb)); 792 segs = udp_rcv_segment(sk, skb, false); 793 skb_list_walk_safe(segs, skb, next) { 794 __skb_pull(skb, skb_transport_offset(skb)); 795 796 udp_post_segment_fix_csum(skb); 797 ret = udpv6_queue_rcv_one_skb(sk, skb); 798 if (ret > 0) 799 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 800 true); 801 } 802 return 0; 803 } 804 805 static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk, 806 __be16 loc_port, const struct in6_addr *loc_addr, 807 __be16 rmt_port, const struct in6_addr *rmt_addr, 808 int dif, int sdif, unsigned short hnum) 809 { 810 const struct inet_sock *inet = inet_sk(sk); 811 812 if (!net_eq(sock_net(sk), net)) 813 return false; 814 815 if (udp_sk(sk)->udp_port_hash != hnum || 816 sk->sk_family != PF_INET6 || 817 (inet->inet_dport && inet->inet_dport != rmt_port) || 818 (!ipv6_addr_any(&sk->sk_v6_daddr) && 819 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 820 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) || 821 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 822 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 823 return false; 824 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 825 return false; 826 return true; 827 } 828 829 static void udp6_csum_zero_error(struct sk_buff *skb) 830 { 831 /* RFC 2460 section 8.1 says that we SHOULD log 832 * this error. Well, it is reasonable. 833 */ 834 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 835 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 836 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 837 } 838 839 /* 840 * Note: called only from the BH handler context, 841 * so we don't need to lock the hashes. 842 */ 843 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 844 const struct in6_addr *saddr, const struct in6_addr *daddr, 845 struct udp_table *udptable, int proto) 846 { 847 struct sock *sk, *first = NULL; 848 const struct udphdr *uh = udp_hdr(skb); 849 unsigned short hnum = ntohs(uh->dest); 850 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 851 unsigned int offset = offsetof(typeof(*sk), sk_node); 852 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 853 int dif = inet6_iif(skb); 854 int sdif = inet6_sdif(skb); 855 struct hlist_node *node; 856 struct sk_buff *nskb; 857 858 if (use_hash2) { 859 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 860 udptable->mask; 861 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 862 start_lookup: 863 hslot = &udptable->hash2[hash2]; 864 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 865 } 866 867 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 868 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 869 uh->source, saddr, dif, sdif, 870 hnum)) 871 continue; 872 /* If zero checksum and no_check is not on for 873 * the socket then skip it. 874 */ 875 if (!uh->check && !udp_get_no_check6_rx(sk)) 876 continue; 877 if (!first) { 878 first = sk; 879 continue; 880 } 881 nskb = skb_clone(skb, GFP_ATOMIC); 882 if (unlikely(!nskb)) { 883 atomic_inc(&sk->sk_drops); 884 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 885 IS_UDPLITE(sk)); 886 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 887 IS_UDPLITE(sk)); 888 continue; 889 } 890 891 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 892 consume_skb(nskb); 893 } 894 895 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 896 if (use_hash2 && hash2 != hash2_any) { 897 hash2 = hash2_any; 898 goto start_lookup; 899 } 900 901 if (first) { 902 if (udpv6_queue_rcv_skb(first, skb) > 0) 903 consume_skb(skb); 904 } else { 905 kfree_skb(skb); 906 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 907 proto == IPPROTO_UDPLITE); 908 } 909 return 0; 910 } 911 912 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 913 { 914 if (udp_sk_rx_dst_set(sk, dst)) 915 sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst)); 916 } 917 918 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 919 * return code conversion for ip layer consumption 920 */ 921 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 922 struct udphdr *uh) 923 { 924 int ret; 925 926 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 927 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 928 929 ret = udpv6_queue_rcv_skb(sk, skb); 930 931 /* a return value > 0 means to resubmit the input */ 932 if (ret > 0) 933 return ret; 934 return 0; 935 } 936 937 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 938 int proto) 939 { 940 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 941 const struct in6_addr *saddr, *daddr; 942 struct net *net = dev_net(skb->dev); 943 struct sock *sk = NULL; 944 struct udphdr *uh; 945 bool refcounted; 946 u32 ulen = 0; 947 948 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 949 goto discard; 950 951 saddr = &ipv6_hdr(skb)->saddr; 952 daddr = &ipv6_hdr(skb)->daddr; 953 uh = udp_hdr(skb); 954 955 ulen = ntohs(uh->len); 956 if (ulen > skb->len) 957 goto short_packet; 958 959 if (proto == IPPROTO_UDP) { 960 /* UDP validates ulen. */ 961 962 /* Check for jumbo payload */ 963 if (ulen == 0) 964 ulen = skb->len; 965 966 if (ulen < sizeof(*uh)) 967 goto short_packet; 968 969 if (ulen < skb->len) { 970 if (pskb_trim_rcsum(skb, ulen)) 971 goto short_packet; 972 saddr = &ipv6_hdr(skb)->saddr; 973 daddr = &ipv6_hdr(skb)->daddr; 974 uh = udp_hdr(skb); 975 } 976 } 977 978 if (udp6_csum_init(skb, uh, proto)) 979 goto csum_error; 980 981 /* Check if the socket is already available, e.g. due to early demux */ 982 sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 983 &refcounted, udp6_ehashfn); 984 if (IS_ERR(sk)) 985 goto no_sk; 986 987 if (sk) { 988 struct dst_entry *dst = skb_dst(skb); 989 int ret; 990 991 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 992 udp6_sk_rx_dst_set(sk, dst); 993 994 if (!uh->check && !udp_get_no_check6_rx(sk)) { 995 if (refcounted) 996 sock_put(sk); 997 goto report_csum_error; 998 } 999 1000 ret = udp6_unicast_rcv_skb(sk, skb, uh); 1001 if (refcounted) 1002 sock_put(sk); 1003 return ret; 1004 } 1005 1006 /* 1007 * Multicast receive code 1008 */ 1009 if (ipv6_addr_is_multicast(daddr)) 1010 return __udp6_lib_mcast_deliver(net, skb, 1011 saddr, daddr, udptable, proto); 1012 1013 /* Unicast */ 1014 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1015 if (sk) { 1016 if (!uh->check && !udp_get_no_check6_rx(sk)) 1017 goto report_csum_error; 1018 return udp6_unicast_rcv_skb(sk, skb, uh); 1019 } 1020 no_sk: 1021 reason = SKB_DROP_REASON_NO_SOCKET; 1022 1023 if (!uh->check) 1024 goto report_csum_error; 1025 1026 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1027 goto discard; 1028 nf_reset_ct(skb); 1029 1030 if (udp_lib_checksum_complete(skb)) 1031 goto csum_error; 1032 1033 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1034 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 1035 1036 sk_skb_reason_drop(sk, skb, reason); 1037 return 0; 1038 1039 short_packet: 1040 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1041 reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1042 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 1043 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1044 saddr, ntohs(uh->source), 1045 ulen, skb->len, 1046 daddr, ntohs(uh->dest)); 1047 goto discard; 1048 1049 report_csum_error: 1050 udp6_csum_zero_error(skb); 1051 csum_error: 1052 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1053 reason = SKB_DROP_REASON_UDP_CSUM; 1054 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1055 discard: 1056 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1057 sk_skb_reason_drop(sk, skb, reason); 1058 return 0; 1059 } 1060 1061 1062 static struct sock *__udp6_lib_demux_lookup(struct net *net, 1063 __be16 loc_port, const struct in6_addr *loc_addr, 1064 __be16 rmt_port, const struct in6_addr *rmt_addr, 1065 int dif, int sdif) 1066 { 1067 struct udp_table *udptable = net->ipv4.udp_table; 1068 unsigned short hnum = ntohs(loc_port); 1069 unsigned int hash2, slot2; 1070 struct udp_hslot *hslot2; 1071 __portpair ports; 1072 struct sock *sk; 1073 1074 hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1075 slot2 = hash2 & udptable->mask; 1076 hslot2 = &udptable->hash2[slot2]; 1077 ports = INET_COMBINED_PORTS(rmt_port, hnum); 1078 1079 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1080 if (sk->sk_state == TCP_ESTABLISHED && 1081 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif)) 1082 return sk; 1083 /* Only check first socket in chain */ 1084 break; 1085 } 1086 return NULL; 1087 } 1088 1089 void udp_v6_early_demux(struct sk_buff *skb) 1090 { 1091 struct net *net = dev_net(skb->dev); 1092 const struct udphdr *uh; 1093 struct sock *sk; 1094 struct dst_entry *dst; 1095 int dif = skb->dev->ifindex; 1096 int sdif = inet6_sdif(skb); 1097 1098 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1099 sizeof(struct udphdr))) 1100 return; 1101 1102 uh = udp_hdr(skb); 1103 1104 if (skb->pkt_type == PACKET_HOST) 1105 sk = __udp6_lib_demux_lookup(net, uh->dest, 1106 &ipv6_hdr(skb)->daddr, 1107 uh->source, &ipv6_hdr(skb)->saddr, 1108 dif, sdif); 1109 else 1110 return; 1111 1112 if (!sk) 1113 return; 1114 1115 skb->sk = sk; 1116 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk)); 1117 skb->destructor = sock_pfree; 1118 dst = rcu_dereference(sk->sk_rx_dst); 1119 1120 if (dst) 1121 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1122 if (dst) { 1123 /* set noref for now. 1124 * any place which wants to hold dst has to call 1125 * dst_hold_safe() 1126 */ 1127 skb_dst_set_noref(skb, dst); 1128 } 1129 } 1130 1131 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1132 { 1133 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 1134 } 1135 1136 /* 1137 * Throw away all pending data and cancel the corking. Socket is locked. 1138 */ 1139 static void udp_v6_flush_pending_frames(struct sock *sk) 1140 { 1141 struct udp_sock *up = udp_sk(sk); 1142 1143 if (up->pending == AF_INET) 1144 udp_flush_pending_frames(sk); 1145 else if (up->pending) { 1146 up->len = 0; 1147 WRITE_ONCE(up->pending, 0); 1148 ip6_flush_pending_frames(sk); 1149 } 1150 } 1151 1152 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1153 int addr_len) 1154 { 1155 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1156 return -EINVAL; 1157 /* The following checks are replicated from __ip6_datagram_connect() 1158 * and intended to prevent BPF program called below from accessing 1159 * bytes that are out of the bound specified by user in addr_len. 1160 */ 1161 if (uaddr->sa_family == AF_INET) { 1162 if (ipv6_only_sock(sk)) 1163 return -EAFNOSUPPORT; 1164 return udp_pre_connect(sk, uaddr, addr_len); 1165 } 1166 1167 if (addr_len < SIN6_LEN_RFC2133) 1168 return -EINVAL; 1169 1170 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len); 1171 } 1172 1173 /** 1174 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1175 * @sk: socket we are sending on 1176 * @skb: sk_buff containing the filled-in UDP header 1177 * (checksum field must be zeroed out) 1178 * @saddr: source address 1179 * @daddr: destination address 1180 * @len: length of packet 1181 */ 1182 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1183 const struct in6_addr *saddr, 1184 const struct in6_addr *daddr, int len) 1185 { 1186 unsigned int offset; 1187 struct udphdr *uh = udp_hdr(skb); 1188 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1189 __wsum csum = 0; 1190 1191 if (!frags) { 1192 /* Only one fragment on the socket. */ 1193 skb->csum_start = skb_transport_header(skb) - skb->head; 1194 skb->csum_offset = offsetof(struct udphdr, check); 1195 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1196 } else { 1197 /* 1198 * HW-checksum won't work as there are two or more 1199 * fragments on the socket so that all csums of sk_buffs 1200 * should be together 1201 */ 1202 offset = skb_transport_offset(skb); 1203 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1204 csum = skb->csum; 1205 1206 skb->ip_summed = CHECKSUM_NONE; 1207 1208 do { 1209 csum = csum_add(csum, frags->csum); 1210 } while ((frags = frags->next)); 1211 1212 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1213 csum); 1214 if (uh->check == 0) 1215 uh->check = CSUM_MANGLED_0; 1216 } 1217 } 1218 1219 /* 1220 * Sending 1221 */ 1222 1223 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1224 struct inet_cork *cork) 1225 { 1226 struct sock *sk = skb->sk; 1227 struct udphdr *uh; 1228 int err = 0; 1229 int is_udplite = IS_UDPLITE(sk); 1230 __wsum csum = 0; 1231 int offset = skb_transport_offset(skb); 1232 int len = skb->len - offset; 1233 int datalen = len - sizeof(*uh); 1234 1235 /* 1236 * Create a UDP header 1237 */ 1238 uh = udp_hdr(skb); 1239 uh->source = fl6->fl6_sport; 1240 uh->dest = fl6->fl6_dport; 1241 uh->len = htons(len); 1242 uh->check = 0; 1243 1244 if (cork->gso_size) { 1245 const int hlen = skb_network_header_len(skb) + 1246 sizeof(struct udphdr); 1247 1248 if (hlen + cork->gso_size > cork->fragsize) { 1249 kfree_skb(skb); 1250 return -EINVAL; 1251 } 1252 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1253 kfree_skb(skb); 1254 return -EINVAL; 1255 } 1256 if (udp_get_no_check6_tx(sk)) { 1257 kfree_skb(skb); 1258 return -EINVAL; 1259 } 1260 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1261 dst_xfrm(skb_dst(skb))) { 1262 kfree_skb(skb); 1263 return -EIO; 1264 } 1265 1266 if (datalen > cork->gso_size) { 1267 skb_shinfo(skb)->gso_size = cork->gso_size; 1268 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1269 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1270 cork->gso_size); 1271 } 1272 goto csum_partial; 1273 } 1274 1275 if (is_udplite) 1276 csum = udplite_csum(skb); 1277 else if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */ 1278 skb->ip_summed = CHECKSUM_NONE; 1279 goto send; 1280 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1281 csum_partial: 1282 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1283 goto send; 1284 } else 1285 csum = udp_csum(skb); 1286 1287 /* add protocol-dependent pseudo-header */ 1288 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1289 len, fl6->flowi6_proto, csum); 1290 if (uh->check == 0) 1291 uh->check = CSUM_MANGLED_0; 1292 1293 send: 1294 err = ip6_send_skb(skb); 1295 if (err) { 1296 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) { 1297 UDP6_INC_STATS(sock_net(sk), 1298 UDP_MIB_SNDBUFERRORS, is_udplite); 1299 err = 0; 1300 } 1301 } else { 1302 UDP6_INC_STATS(sock_net(sk), 1303 UDP_MIB_OUTDATAGRAMS, is_udplite); 1304 } 1305 return err; 1306 } 1307 1308 static int udp_v6_push_pending_frames(struct sock *sk) 1309 { 1310 struct sk_buff *skb; 1311 struct udp_sock *up = udp_sk(sk); 1312 int err = 0; 1313 1314 if (up->pending == AF_INET) 1315 return udp_push_pending_frames(sk); 1316 1317 skb = ip6_finish_skb(sk); 1318 if (!skb) 1319 goto out; 1320 1321 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6, 1322 &inet_sk(sk)->cork.base); 1323 out: 1324 up->len = 0; 1325 WRITE_ONCE(up->pending, 0); 1326 return err; 1327 } 1328 1329 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1330 { 1331 struct ipv6_txoptions opt_space; 1332 struct udp_sock *up = udp_sk(sk); 1333 struct inet_sock *inet = inet_sk(sk); 1334 struct ipv6_pinfo *np = inet6_sk(sk); 1335 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1336 struct in6_addr *daddr, *final_p, final; 1337 struct ipv6_txoptions *opt = NULL; 1338 struct ipv6_txoptions *opt_to_free = NULL; 1339 struct ip6_flowlabel *flowlabel = NULL; 1340 struct inet_cork_full cork; 1341 struct flowi6 *fl6 = &cork.fl.u.ip6; 1342 struct dst_entry *dst; 1343 struct ipcm6_cookie ipc6; 1344 int addr_len = msg->msg_namelen; 1345 bool connected = false; 1346 int ulen = len; 1347 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE; 1348 int err; 1349 int is_udplite = IS_UDPLITE(sk); 1350 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1351 1352 ipcm6_init(&ipc6); 1353 ipc6.gso_size = READ_ONCE(up->gso_size); 1354 ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags); 1355 ipc6.sockc.mark = READ_ONCE(sk->sk_mark); 1356 1357 /* destination address check */ 1358 if (sin6) { 1359 if (addr_len < offsetof(struct sockaddr, sa_data)) 1360 return -EINVAL; 1361 1362 switch (sin6->sin6_family) { 1363 case AF_INET6: 1364 if (addr_len < SIN6_LEN_RFC2133) 1365 return -EINVAL; 1366 daddr = &sin6->sin6_addr; 1367 if (ipv6_addr_any(daddr) && 1368 ipv6_addr_v4mapped(&np->saddr)) 1369 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1370 daddr); 1371 break; 1372 case AF_INET: 1373 goto do_udp_sendmsg; 1374 case AF_UNSPEC: 1375 msg->msg_name = sin6 = NULL; 1376 msg->msg_namelen = addr_len = 0; 1377 daddr = NULL; 1378 break; 1379 default: 1380 return -EINVAL; 1381 } 1382 } else if (!READ_ONCE(up->pending)) { 1383 if (sk->sk_state != TCP_ESTABLISHED) 1384 return -EDESTADDRREQ; 1385 daddr = &sk->sk_v6_daddr; 1386 } else 1387 daddr = NULL; 1388 1389 if (daddr) { 1390 if (ipv6_addr_v4mapped(daddr)) { 1391 struct sockaddr_in sin; 1392 sin.sin_family = AF_INET; 1393 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1394 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1395 msg->msg_name = &sin; 1396 msg->msg_namelen = sizeof(sin); 1397 do_udp_sendmsg: 1398 err = ipv6_only_sock(sk) ? 1399 -ENETUNREACH : udp_sendmsg(sk, msg, len); 1400 msg->msg_name = sin6; 1401 msg->msg_namelen = addr_len; 1402 return err; 1403 } 1404 } 1405 1406 /* Rough check on arithmetic overflow, 1407 better check is made in ip6_append_data(). 1408 */ 1409 if (len > INT_MAX - sizeof(struct udphdr)) 1410 return -EMSGSIZE; 1411 1412 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1413 if (READ_ONCE(up->pending)) { 1414 if (READ_ONCE(up->pending) == AF_INET) 1415 return udp_sendmsg(sk, msg, len); 1416 /* 1417 * There are pending frames. 1418 * The socket lock must be held while it's corked. 1419 */ 1420 lock_sock(sk); 1421 if (likely(up->pending)) { 1422 if (unlikely(up->pending != AF_INET6)) { 1423 release_sock(sk); 1424 return -EAFNOSUPPORT; 1425 } 1426 dst = NULL; 1427 goto do_append_data; 1428 } 1429 release_sock(sk); 1430 } 1431 ulen += sizeof(struct udphdr); 1432 1433 memset(fl6, 0, sizeof(*fl6)); 1434 1435 if (sin6) { 1436 if (sin6->sin6_port == 0) 1437 return -EINVAL; 1438 1439 fl6->fl6_dport = sin6->sin6_port; 1440 daddr = &sin6->sin6_addr; 1441 1442 if (inet6_test_bit(SNDFLOW, sk)) { 1443 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1444 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) { 1445 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1446 if (IS_ERR(flowlabel)) 1447 return -EINVAL; 1448 } 1449 } 1450 1451 /* 1452 * Otherwise it will be difficult to maintain 1453 * sk->sk_dst_cache. 1454 */ 1455 if (sk->sk_state == TCP_ESTABLISHED && 1456 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1457 daddr = &sk->sk_v6_daddr; 1458 1459 if (addr_len >= sizeof(struct sockaddr_in6) && 1460 sin6->sin6_scope_id && 1461 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1462 fl6->flowi6_oif = sin6->sin6_scope_id; 1463 } else { 1464 if (sk->sk_state != TCP_ESTABLISHED) 1465 return -EDESTADDRREQ; 1466 1467 fl6->fl6_dport = inet->inet_dport; 1468 daddr = &sk->sk_v6_daddr; 1469 fl6->flowlabel = np->flow_label; 1470 connected = true; 1471 } 1472 1473 if (!fl6->flowi6_oif) 1474 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if); 1475 1476 if (!fl6->flowi6_oif) 1477 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1478 1479 fl6->flowi6_uid = sk->sk_uid; 1480 1481 if (msg->msg_controllen) { 1482 opt = &opt_space; 1483 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1484 opt->tot_len = sizeof(*opt); 1485 ipc6.opt = opt; 1486 1487 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1488 if (err > 0) { 1489 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6, 1490 &ipc6); 1491 connected = false; 1492 } 1493 if (err < 0) { 1494 fl6_sock_release(flowlabel); 1495 return err; 1496 } 1497 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1498 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1499 if (IS_ERR(flowlabel)) 1500 return -EINVAL; 1501 } 1502 if (!(opt->opt_nflen|opt->opt_flen)) 1503 opt = NULL; 1504 } 1505 if (!opt) { 1506 opt = txopt_get(np); 1507 opt_to_free = opt; 1508 } 1509 if (flowlabel) 1510 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1511 opt = ipv6_fixup_options(&opt_space, opt); 1512 ipc6.opt = opt; 1513 1514 fl6->flowi6_proto = sk->sk_protocol; 1515 fl6->flowi6_mark = ipc6.sockc.mark; 1516 fl6->daddr = *daddr; 1517 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr)) 1518 fl6->saddr = np->saddr; 1519 fl6->fl6_sport = inet->inet_sport; 1520 1521 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1522 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1523 (struct sockaddr *)sin6, 1524 &addr_len, 1525 &fl6->saddr); 1526 if (err) 1527 goto out_no_dst; 1528 if (sin6) { 1529 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1530 /* BPF program rewrote IPv6-only by IPv4-mapped 1531 * IPv6. It's currently unsupported. 1532 */ 1533 err = -ENOTSUPP; 1534 goto out_no_dst; 1535 } 1536 if (sin6->sin6_port == 0) { 1537 /* BPF program set invalid port. Reject it. */ 1538 err = -EINVAL; 1539 goto out_no_dst; 1540 } 1541 fl6->fl6_dport = sin6->sin6_port; 1542 fl6->daddr = sin6->sin6_addr; 1543 } 1544 } 1545 1546 if (ipv6_addr_any(&fl6->daddr)) 1547 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1548 1549 final_p = fl6_update_dst(fl6, opt, &final); 1550 if (final_p) 1551 connected = false; 1552 1553 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) { 1554 fl6->flowi6_oif = READ_ONCE(np->mcast_oif); 1555 connected = false; 1556 } else if (!fl6->flowi6_oif) 1557 fl6->flowi6_oif = READ_ONCE(np->ucast_oif); 1558 1559 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6)); 1560 1561 if (ipc6.tclass < 0) 1562 ipc6.tclass = np->tclass; 1563 1564 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel); 1565 1566 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected); 1567 if (IS_ERR(dst)) { 1568 err = PTR_ERR(dst); 1569 dst = NULL; 1570 goto out; 1571 } 1572 1573 if (ipc6.hlimit < 0) 1574 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst); 1575 1576 if (msg->msg_flags&MSG_CONFIRM) 1577 goto do_confirm; 1578 back_from_confirm: 1579 1580 /* Lockless fast path for the non-corking case */ 1581 if (!corkreq) { 1582 struct sk_buff *skb; 1583 1584 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1585 sizeof(struct udphdr), &ipc6, 1586 dst_rt6_info(dst), 1587 msg->msg_flags, &cork); 1588 err = PTR_ERR(skb); 1589 if (!IS_ERR_OR_NULL(skb)) 1590 err = udp_v6_send_skb(skb, fl6, &cork.base); 1591 /* ip6_make_skb steals dst reference */ 1592 goto out_no_dst; 1593 } 1594 1595 lock_sock(sk); 1596 if (unlikely(up->pending)) { 1597 /* The socket is already corked while preparing it. */ 1598 /* ... which is an evident application bug. --ANK */ 1599 release_sock(sk); 1600 1601 net_dbg_ratelimited("udp cork app bug 2\n"); 1602 err = -EINVAL; 1603 goto out; 1604 } 1605 1606 WRITE_ONCE(up->pending, AF_INET6); 1607 1608 do_append_data: 1609 if (ipc6.dontfrag < 0) 1610 ipc6.dontfrag = inet6_test_bit(DONTFRAG, sk); 1611 up->len += ulen; 1612 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1613 &ipc6, fl6, dst_rt6_info(dst), 1614 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1615 if (err) 1616 udp_v6_flush_pending_frames(sk); 1617 else if (!corkreq) 1618 err = udp_v6_push_pending_frames(sk); 1619 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1620 WRITE_ONCE(up->pending, 0); 1621 1622 if (err > 0) 1623 err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0; 1624 release_sock(sk); 1625 1626 out: 1627 dst_release(dst); 1628 out_no_dst: 1629 fl6_sock_release(flowlabel); 1630 txopt_put(opt_to_free); 1631 if (!err) 1632 return len; 1633 /* 1634 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1635 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1636 * we don't have a good statistic (IpOutDiscards but it can be too many 1637 * things). We could add another new stat but at least for now that 1638 * seems like overkill. 1639 */ 1640 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1641 UDP6_INC_STATS(sock_net(sk), 1642 UDP_MIB_SNDBUFERRORS, is_udplite); 1643 } 1644 return err; 1645 1646 do_confirm: 1647 if (msg->msg_flags & MSG_PROBE) 1648 dst_confirm_neigh(dst, &fl6->daddr); 1649 if (!(msg->msg_flags&MSG_PROBE) || len) 1650 goto back_from_confirm; 1651 err = 0; 1652 goto out; 1653 } 1654 EXPORT_SYMBOL(udpv6_sendmsg); 1655 1656 static void udpv6_splice_eof(struct socket *sock) 1657 { 1658 struct sock *sk = sock->sk; 1659 struct udp_sock *up = udp_sk(sk); 1660 1661 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk)) 1662 return; 1663 1664 lock_sock(sk); 1665 if (up->pending && !udp_test_bit(CORK, sk)) 1666 udp_v6_push_pending_frames(sk); 1667 release_sock(sk); 1668 } 1669 1670 void udpv6_destroy_sock(struct sock *sk) 1671 { 1672 struct udp_sock *up = udp_sk(sk); 1673 lock_sock(sk); 1674 1675 /* protects from races with udp_abort() */ 1676 sock_set_flag(sk, SOCK_DEAD); 1677 udp_v6_flush_pending_frames(sk); 1678 release_sock(sk); 1679 1680 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1681 if (up->encap_type) { 1682 void (*encap_destroy)(struct sock *sk); 1683 encap_destroy = READ_ONCE(up->encap_destroy); 1684 if (encap_destroy) 1685 encap_destroy(sk); 1686 } 1687 if (udp_test_bit(ENCAP_ENABLED, sk)) { 1688 static_branch_dec(&udpv6_encap_needed_key); 1689 udp_encap_disable(); 1690 } 1691 } 1692 } 1693 1694 /* 1695 * Socket option code for UDP 1696 */ 1697 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1698 unsigned int optlen) 1699 { 1700 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 1701 return udp_lib_setsockopt(sk, level, optname, 1702 optval, optlen, 1703 udp_v6_push_pending_frames); 1704 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1705 } 1706 1707 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1708 char __user *optval, int __user *optlen) 1709 { 1710 if (level == SOL_UDP || level == SOL_UDPLITE) 1711 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1712 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1713 } 1714 1715 1716 /* ------------------------------------------------------------------------ */ 1717 #ifdef CONFIG_PROC_FS 1718 int udp6_seq_show(struct seq_file *seq, void *v) 1719 { 1720 if (v == SEQ_START_TOKEN) { 1721 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1722 } else { 1723 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1724 const struct inet_sock *inet = inet_sk((const struct sock *)v); 1725 __u16 srcp = ntohs(inet->inet_sport); 1726 __u16 destp = ntohs(inet->inet_dport); 1727 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1728 udp_rqueue_get(v), bucket); 1729 } 1730 return 0; 1731 } 1732 1733 const struct seq_operations udp6_seq_ops = { 1734 .start = udp_seq_start, 1735 .next = udp_seq_next, 1736 .stop = udp_seq_stop, 1737 .show = udp6_seq_show, 1738 }; 1739 EXPORT_SYMBOL(udp6_seq_ops); 1740 1741 static struct udp_seq_afinfo udp6_seq_afinfo = { 1742 .family = AF_INET6, 1743 .udp_table = NULL, 1744 }; 1745 1746 int __net_init udp6_proc_init(struct net *net) 1747 { 1748 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1749 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1750 return -ENOMEM; 1751 return 0; 1752 } 1753 1754 void udp6_proc_exit(struct net *net) 1755 { 1756 remove_proc_entry("udp6", net->proc_net); 1757 } 1758 #endif /* CONFIG_PROC_FS */ 1759 1760 /* ------------------------------------------------------------------------ */ 1761 1762 struct proto udpv6_prot = { 1763 .name = "UDPv6", 1764 .owner = THIS_MODULE, 1765 .close = udp_lib_close, 1766 .pre_connect = udpv6_pre_connect, 1767 .connect = ip6_datagram_connect, 1768 .disconnect = udp_disconnect, 1769 .ioctl = udp_ioctl, 1770 .init = udpv6_init_sock, 1771 .destroy = udpv6_destroy_sock, 1772 .setsockopt = udpv6_setsockopt, 1773 .getsockopt = udpv6_getsockopt, 1774 .sendmsg = udpv6_sendmsg, 1775 .recvmsg = udpv6_recvmsg, 1776 .splice_eof = udpv6_splice_eof, 1777 .release_cb = ip6_datagram_release_cb, 1778 .hash = udp_lib_hash, 1779 .unhash = udp_lib_unhash, 1780 .rehash = udp_v6_rehash, 1781 .get_port = udp_v6_get_port, 1782 .put_port = udp_lib_unhash, 1783 #ifdef CONFIG_BPF_SYSCALL 1784 .psock_update_sk_prot = udp_bpf_update_proto, 1785 #endif 1786 1787 .memory_allocated = &udp_memory_allocated, 1788 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 1789 1790 .sysctl_mem = sysctl_udp_mem, 1791 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1792 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1793 .obj_size = sizeof(struct udp6_sock), 1794 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6), 1795 .h.udp_table = NULL, 1796 .diag_destroy = udp_abort, 1797 }; 1798 1799 static struct inet_protosw udpv6_protosw = { 1800 .type = SOCK_DGRAM, 1801 .protocol = IPPROTO_UDP, 1802 .prot = &udpv6_prot, 1803 .ops = &inet6_dgram_ops, 1804 .flags = INET_PROTOSW_PERMANENT, 1805 }; 1806 1807 int __init udpv6_init(void) 1808 { 1809 int ret; 1810 1811 net_hotdata.udpv6_protocol = (struct inet6_protocol) { 1812 .handler = udpv6_rcv, 1813 .err_handler = udpv6_err, 1814 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL, 1815 }; 1816 ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1817 if (ret) 1818 goto out; 1819 1820 ret = inet6_register_protosw(&udpv6_protosw); 1821 if (ret) 1822 goto out_udpv6_protocol; 1823 out: 1824 return ret; 1825 1826 out_udpv6_protocol: 1827 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1828 goto out; 1829 } 1830 1831 void udpv6_exit(void) 1832 { 1833 inet6_unregister_protosw(&udpv6_protosw); 1834 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1835 } 1836