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 <net/xfrm.h> 49 #include <net/inet_hashtables.h> 50 #include <net/inet6_hashtables.h> 51 #include <net/busy_poll.h> 52 #include <net/sock_reuseport.h> 53 54 #include <linux/proc_fs.h> 55 #include <linux/seq_file.h> 56 #include <trace/events/skb.h> 57 #include "udp_impl.h" 58 59 static u32 udp6_ehashfn(const struct net *net, 60 const struct in6_addr *laddr, 61 const u16 lport, 62 const struct in6_addr *faddr, 63 const __be16 fport) 64 { 65 static u32 udp6_ehash_secret __read_mostly; 66 static u32 udp_ipv6_hash_secret __read_mostly; 67 68 u32 lhash, fhash; 69 70 net_get_random_once(&udp6_ehash_secret, 71 sizeof(udp6_ehash_secret)); 72 net_get_random_once(&udp_ipv6_hash_secret, 73 sizeof(udp_ipv6_hash_secret)); 74 75 lhash = (__force u32)laddr->s6_addr32[3]; 76 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 77 78 return __inet6_ehashfn(lhash, lport, fhash, fport, 79 udp_ipv6_hash_secret + net_hash_mix(net)); 80 } 81 82 int udp_v6_get_port(struct sock *sk, unsigned short snum) 83 { 84 unsigned int hash2_nulladdr = 85 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 86 unsigned int hash2_partial = 87 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 88 89 /* precompute partial secondary hash */ 90 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 91 return udp_lib_get_port(sk, snum, hash2_nulladdr); 92 } 93 94 void udp_v6_rehash(struct sock *sk) 95 { 96 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 97 &sk->sk_v6_rcv_saddr, 98 inet_sk(sk)->inet_num); 99 100 udp_lib_rehash(sk, new_hash); 101 } 102 103 static int compute_score(struct sock *sk, struct net *net, 104 const struct in6_addr *saddr, __be16 sport, 105 const struct in6_addr *daddr, unsigned short hnum, 106 int dif, int sdif) 107 { 108 int bound_dev_if, score; 109 struct inet_sock *inet; 110 bool dev_match; 111 112 if (!net_eq(sock_net(sk), net) || 113 udp_sk(sk)->udp_port_hash != hnum || 114 sk->sk_family != PF_INET6) 115 return -1; 116 117 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 118 return -1; 119 120 score = 0; 121 inet = inet_sk(sk); 122 123 if (inet->inet_dport) { 124 if (inet->inet_dport != sport) 125 return -1; 126 score++; 127 } 128 129 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 130 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 131 return -1; 132 score++; 133 } 134 135 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 136 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif); 137 if (!dev_match) 138 return -1; 139 if (bound_dev_if) 140 score++; 141 142 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 143 score++; 144 145 return score; 146 } 147 148 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 149 struct sk_buff *skb, 150 const struct in6_addr *saddr, 151 __be16 sport, 152 const struct in6_addr *daddr, 153 unsigned int hnum) 154 { 155 struct sock *reuse_sk = NULL; 156 u32 hash; 157 158 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 159 hash = udp6_ehashfn(net, daddr, hnum, saddr, sport); 160 reuse_sk = reuseport_select_sock(sk, hash, skb, 161 sizeof(struct udphdr)); 162 } 163 return reuse_sk; 164 } 165 166 /* called with rcu_read_lock() */ 167 static struct sock *udp6_lib_lookup2(struct net *net, 168 const struct in6_addr *saddr, __be16 sport, 169 const struct in6_addr *daddr, unsigned int hnum, 170 int dif, int sdif, struct udp_hslot *hslot2, 171 struct sk_buff *skb) 172 { 173 struct sock *sk, *result; 174 int score, badness; 175 176 result = NULL; 177 badness = -1; 178 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 179 score = compute_score(sk, net, saddr, sport, 180 daddr, hnum, dif, sdif); 181 if (score > badness) { 182 result = lookup_reuseport(net, sk, skb, 183 saddr, sport, daddr, hnum); 184 /* Fall back to scoring if group has connections */ 185 if (result && !reuseport_has_conns(sk, false)) 186 return result; 187 188 result = result ? : sk; 189 badness = score; 190 } 191 } 192 return result; 193 } 194 195 static inline struct sock *udp6_lookup_run_bpf(struct net *net, 196 struct udp_table *udptable, 197 struct sk_buff *skb, 198 const struct in6_addr *saddr, 199 __be16 sport, 200 const struct in6_addr *daddr, 201 u16 hnum, const int dif) 202 { 203 struct sock *sk, *reuse_sk; 204 bool no_reuseport; 205 206 if (udptable != &udp_table) 207 return NULL; /* only UDP is supported */ 208 209 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_UDP, saddr, sport, 210 daddr, hnum, dif, &sk); 211 if (no_reuseport || IS_ERR_OR_NULL(sk)) 212 return sk; 213 214 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 215 if (reuse_sk) 216 sk = reuse_sk; 217 return sk; 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 sk = udp6_lookup_run_bpf(net, udptable, skb, 246 saddr, sport, daddr, hnum, dif); 247 if (sk) { 248 result = sk; 249 goto done; 250 } 251 } 252 253 /* Got non-wildcard socket or error on first lookup */ 254 if (result) 255 goto done; 256 257 /* Lookup wildcard sockets */ 258 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 259 slot2 = hash2 & udptable->mask; 260 hslot2 = &udptable->hash2[slot2]; 261 262 result = udp6_lib_lookup2(net, saddr, sport, 263 &in6addr_any, hnum, dif, sdif, 264 hslot2, skb); 265 done: 266 if (IS_ERR(result)) 267 return NULL; 268 return result; 269 } 270 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 271 272 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 273 __be16 sport, __be16 dport, 274 struct udp_table *udptable) 275 { 276 const struct ipv6hdr *iph = ipv6_hdr(skb); 277 278 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 279 &iph->daddr, dport, inet6_iif(skb), 280 inet6_sdif(skb), udptable, skb); 281 } 282 283 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, 284 __be16 sport, __be16 dport) 285 { 286 const struct ipv6hdr *iph = ipv6_hdr(skb); 287 288 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 289 &iph->daddr, dport, inet6_iif(skb), 290 inet6_sdif(skb), &udp_table, NULL); 291 } 292 293 /* Must be called under rcu_read_lock(). 294 * Does increment socket refcount. 295 */ 296 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 297 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 298 const struct in6_addr *daddr, __be16 dport, int dif) 299 { 300 struct sock *sk; 301 302 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 303 dif, 0, &udp_table, NULL); 304 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 305 sk = NULL; 306 return sk; 307 } 308 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 309 #endif 310 311 /* do not use the scratch area len for jumbogram: their length execeeds the 312 * scratch area space; note that the IP6CB flags is still in the first 313 * cacheline, so checking for jumbograms is cheap 314 */ 315 static int udp6_skb_len(struct sk_buff *skb) 316 { 317 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 318 } 319 320 /* 321 * This should be easy, if there is something there we 322 * return it, otherwise we block. 323 */ 324 325 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 326 int flags, int *addr_len) 327 { 328 struct ipv6_pinfo *np = inet6_sk(sk); 329 struct inet_sock *inet = inet_sk(sk); 330 struct sk_buff *skb; 331 unsigned int ulen, copied; 332 int off, err, peeking = flags & MSG_PEEK; 333 int is_udplite = IS_UDPLITE(sk); 334 struct udp_mib __percpu *mib; 335 bool checksum_valid = false; 336 int is_udp4; 337 338 if (flags & MSG_ERRQUEUE) 339 return ipv6_recv_error(sk, msg, len, addr_len); 340 341 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 342 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 343 344 try_again: 345 off = sk_peek_offset(sk, flags); 346 skb = __skb_recv_udp(sk, flags, &off, &err); 347 if (!skb) 348 return err; 349 350 ulen = udp6_skb_len(skb); 351 copied = len; 352 if (copied > ulen - off) 353 copied = ulen - off; 354 else if (copied < ulen) 355 msg->msg_flags |= MSG_TRUNC; 356 357 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 358 mib = __UDPX_MIB(sk, is_udp4); 359 360 /* 361 * If checksum is needed at all, try to do it while copying the 362 * data. If the data is truncated, or if we only want a partial 363 * coverage checksum (UDP-Lite), do it before the copy. 364 */ 365 366 if (copied < ulen || peeking || 367 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 368 checksum_valid = udp_skb_csum_unnecessary(skb) || 369 !__udp_lib_checksum_complete(skb); 370 if (!checksum_valid) 371 goto csum_copy_err; 372 } 373 374 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 375 if (udp_skb_is_linear(skb)) 376 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 377 else 378 err = skb_copy_datagram_msg(skb, off, msg, copied); 379 } else { 380 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 381 if (err == -EINVAL) 382 goto csum_copy_err; 383 } 384 if (unlikely(err)) { 385 if (!peeking) { 386 atomic_inc(&sk->sk_drops); 387 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 388 } 389 kfree_skb(skb); 390 return err; 391 } 392 if (!peeking) 393 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 394 395 sock_recv_cmsgs(msg, sk, skb); 396 397 /* Copy the address. */ 398 if (msg->msg_name) { 399 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 400 sin6->sin6_family = AF_INET6; 401 sin6->sin6_port = udp_hdr(skb)->source; 402 sin6->sin6_flowinfo = 0; 403 404 if (is_udp4) { 405 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 406 &sin6->sin6_addr); 407 sin6->sin6_scope_id = 0; 408 } else { 409 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 410 sin6->sin6_scope_id = 411 ipv6_iface_scope_id(&sin6->sin6_addr, 412 inet6_iif(skb)); 413 } 414 *addr_len = sizeof(*sin6); 415 416 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 417 (struct sockaddr *)sin6); 418 } 419 420 if (udp_sk(sk)->gro_enabled) 421 udp_cmsg_recv(msg, sk, skb); 422 423 if (np->rxopt.all) 424 ip6_datagram_recv_common_ctl(sk, msg, skb); 425 426 if (is_udp4) { 427 if (inet->cmsg_flags) 428 ip_cmsg_recv_offset(msg, sk, skb, 429 sizeof(struct udphdr), off); 430 } else { 431 if (np->rxopt.all) 432 ip6_datagram_recv_specific_ctl(sk, msg, skb); 433 } 434 435 err = copied; 436 if (flags & MSG_TRUNC) 437 err = ulen; 438 439 skb_consume_udp(sk, skb, peeking ? -err : err); 440 return err; 441 442 csum_copy_err: 443 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 444 udp_skb_destructor)) { 445 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 446 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 447 } 448 kfree_skb(skb); 449 450 /* starting over for a new packet, but check if we need to yield */ 451 cond_resched(); 452 msg->msg_flags &= ~MSG_TRUNC; 453 goto try_again; 454 } 455 456 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 457 void udpv6_encap_enable(void) 458 { 459 static_branch_inc(&udpv6_encap_needed_key); 460 } 461 EXPORT_SYMBOL(udpv6_encap_enable); 462 463 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 464 * through error handlers in encapsulations looking for a match. 465 */ 466 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 467 struct inet6_skb_parm *opt, 468 u8 type, u8 code, int offset, __be32 info) 469 { 470 int i; 471 472 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 473 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 474 u8 type, u8 code, int offset, __be32 info); 475 const struct ip6_tnl_encap_ops *encap; 476 477 encap = rcu_dereference(ip6tun_encaps[i]); 478 if (!encap) 479 continue; 480 handler = encap->err_handler; 481 if (handler && !handler(skb, opt, type, code, offset, info)) 482 return 0; 483 } 484 485 return -ENOENT; 486 } 487 488 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 489 * reversing source and destination port: this will match tunnels that force the 490 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 491 * lwtunnels might actually break this assumption by being configured with 492 * different destination ports on endpoints, in this case we won't be able to 493 * trace ICMP messages back to them. 494 * 495 * If this doesn't match any socket, probe tunnels with arbitrary destination 496 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 497 * we've sent packets to won't necessarily match the local destination port. 498 * 499 * Then ask the tunnel implementation to match the error against a valid 500 * association. 501 * 502 * Return an error if we can't find a match, the socket if we need further 503 * processing, zero otherwise. 504 */ 505 static struct sock *__udp6_lib_err_encap(struct net *net, 506 const struct ipv6hdr *hdr, int offset, 507 struct udphdr *uh, 508 struct udp_table *udptable, 509 struct sock *sk, 510 struct sk_buff *skb, 511 struct inet6_skb_parm *opt, 512 u8 type, u8 code, __be32 info) 513 { 514 int (*lookup)(struct sock *sk, struct sk_buff *skb); 515 int network_offset, transport_offset; 516 struct udp_sock *up; 517 518 network_offset = skb_network_offset(skb); 519 transport_offset = skb_transport_offset(skb); 520 521 /* Network header needs to point to the outer IPv6 header inside ICMP */ 522 skb_reset_network_header(skb); 523 524 /* Transport header needs to point to the UDP header */ 525 skb_set_transport_header(skb, offset); 526 527 if (sk) { 528 up = udp_sk(sk); 529 530 lookup = READ_ONCE(up->encap_err_lookup); 531 if (lookup && lookup(sk, skb)) 532 sk = NULL; 533 534 goto out; 535 } 536 537 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 538 &hdr->saddr, uh->dest, 539 inet6_iif(skb), 0, udptable, skb); 540 if (sk) { 541 up = udp_sk(sk); 542 543 lookup = READ_ONCE(up->encap_err_lookup); 544 if (!lookup || lookup(sk, skb)) 545 sk = NULL; 546 } 547 548 out: 549 if (!sk) { 550 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 551 offset, info)); 552 } 553 554 skb_set_transport_header(skb, transport_offset); 555 skb_set_network_header(skb, network_offset); 556 557 return sk; 558 } 559 560 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 561 u8 type, u8 code, int offset, __be32 info, 562 struct udp_table *udptable) 563 { 564 struct ipv6_pinfo *np; 565 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 566 const struct in6_addr *saddr = &hdr->saddr; 567 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr; 568 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 569 bool tunnel = false; 570 struct sock *sk; 571 int harderr; 572 int err; 573 struct net *net = dev_net(skb->dev); 574 575 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 576 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 577 578 if (!sk || udp_sk(sk)->encap_type) { 579 /* No socket for error: try tunnels before discarding */ 580 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 581 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 582 udptable, sk, skb, 583 opt, type, code, info); 584 if (!sk) 585 return 0; 586 } else 587 sk = ERR_PTR(-ENOENT); 588 589 if (IS_ERR(sk)) { 590 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 591 ICMP6_MIB_INERRORS); 592 return PTR_ERR(sk); 593 } 594 595 tunnel = true; 596 } 597 598 harderr = icmpv6_err_convert(type, code, &err); 599 np = inet6_sk(sk); 600 601 if (type == ICMPV6_PKT_TOOBIG) { 602 if (!ip6_sk_accept_pmtu(sk)) 603 goto out; 604 ip6_sk_update_pmtu(skb, sk, info); 605 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 606 harderr = 1; 607 } 608 if (type == NDISC_REDIRECT) { 609 if (tunnel) { 610 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 611 sk->sk_mark, sk->sk_uid); 612 } else { 613 ip6_sk_redirect(skb, sk); 614 } 615 goto out; 616 } 617 618 /* Tunnels don't have an application socket: don't pass errors back */ 619 if (tunnel) 620 goto out; 621 622 if (!np->recverr) { 623 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 624 goto out; 625 } else { 626 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 627 } 628 629 sk->sk_err = err; 630 sk_error_report(sk); 631 out: 632 return 0; 633 } 634 635 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 636 { 637 int rc; 638 639 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 640 sock_rps_save_rxhash(sk, skb); 641 sk_mark_napi_id(sk, skb); 642 sk_incoming_cpu_update(sk); 643 } else { 644 sk_mark_napi_id_once(sk, skb); 645 } 646 647 rc = __udp_enqueue_schedule_skb(sk, skb); 648 if (rc < 0) { 649 int is_udplite = IS_UDPLITE(sk); 650 651 /* Note that an ENOMEM error is charged twice */ 652 if (rc == -ENOMEM) 653 UDP6_INC_STATS(sock_net(sk), 654 UDP_MIB_RCVBUFERRORS, is_udplite); 655 else 656 UDP6_INC_STATS(sock_net(sk), 657 UDP_MIB_MEMERRORS, is_udplite); 658 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 659 kfree_skb(skb); 660 return -1; 661 } 662 663 return 0; 664 } 665 666 static __inline__ int udpv6_err(struct sk_buff *skb, 667 struct inet6_skb_parm *opt, u8 type, 668 u8 code, int offset, __be32 info) 669 { 670 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 671 } 672 673 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 674 { 675 struct udp_sock *up = udp_sk(sk); 676 int is_udplite = IS_UDPLITE(sk); 677 678 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 679 goto drop; 680 681 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 682 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 683 684 /* 685 * This is an encapsulation socket so pass the skb to 686 * the socket's udp_encap_rcv() hook. Otherwise, just 687 * fall through and pass this up the UDP socket. 688 * up->encap_rcv() returns the following value: 689 * =0 if skb was successfully passed to the encap 690 * handler or was discarded by it. 691 * >0 if skb should be passed on to UDP. 692 * <0 if skb should be resubmitted as proto -N 693 */ 694 695 /* if we're overly short, let UDP handle it */ 696 encap_rcv = READ_ONCE(up->encap_rcv); 697 if (encap_rcv) { 698 int ret; 699 700 /* Verify checksum before giving to encap */ 701 if (udp_lib_checksum_complete(skb)) 702 goto csum_error; 703 704 ret = encap_rcv(sk, skb); 705 if (ret <= 0) { 706 __UDP6_INC_STATS(sock_net(sk), 707 UDP_MIB_INDATAGRAMS, 708 is_udplite); 709 return -ret; 710 } 711 } 712 713 /* FALLTHROUGH -- it's a UDP Packet */ 714 } 715 716 /* 717 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 718 */ 719 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 720 721 if (up->pcrlen == 0) { /* full coverage was set */ 722 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 723 UDP_SKB_CB(skb)->cscov, skb->len); 724 goto drop; 725 } 726 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 727 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 728 UDP_SKB_CB(skb)->cscov, up->pcrlen); 729 goto drop; 730 } 731 } 732 733 prefetch(&sk->sk_rmem_alloc); 734 if (rcu_access_pointer(sk->sk_filter) && 735 udp_lib_checksum_complete(skb)) 736 goto csum_error; 737 738 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 739 goto drop; 740 741 udp_csum_pull_header(skb); 742 743 skb_dst_drop(skb); 744 745 return __udpv6_queue_rcv_skb(sk, skb); 746 747 csum_error: 748 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 749 drop: 750 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 751 atomic_inc(&sk->sk_drops); 752 kfree_skb(skb); 753 return -1; 754 } 755 756 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 757 { 758 struct sk_buff *next, *segs; 759 int ret; 760 761 if (likely(!udp_unexpected_gso(sk, skb))) 762 return udpv6_queue_rcv_one_skb(sk, skb); 763 764 __skb_push(skb, -skb_mac_offset(skb)); 765 segs = udp_rcv_segment(sk, skb, false); 766 skb_list_walk_safe(segs, skb, next) { 767 __skb_pull(skb, skb_transport_offset(skb)); 768 769 udp_post_segment_fix_csum(skb); 770 ret = udpv6_queue_rcv_one_skb(sk, skb); 771 if (ret > 0) 772 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 773 true); 774 } 775 return 0; 776 } 777 778 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 779 __be16 loc_port, const struct in6_addr *loc_addr, 780 __be16 rmt_port, const struct in6_addr *rmt_addr, 781 int dif, int sdif, unsigned short hnum) 782 { 783 struct inet_sock *inet = inet_sk(sk); 784 785 if (!net_eq(sock_net(sk), net)) 786 return false; 787 788 if (udp_sk(sk)->udp_port_hash != hnum || 789 sk->sk_family != PF_INET6 || 790 (inet->inet_dport && inet->inet_dport != rmt_port) || 791 (!ipv6_addr_any(&sk->sk_v6_daddr) && 792 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 793 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) || 794 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 795 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 796 return false; 797 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 798 return false; 799 return true; 800 } 801 802 static void udp6_csum_zero_error(struct sk_buff *skb) 803 { 804 /* RFC 2460 section 8.1 says that we SHOULD log 805 * this error. Well, it is reasonable. 806 */ 807 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 808 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 809 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 810 } 811 812 /* 813 * Note: called only from the BH handler context, 814 * so we don't need to lock the hashes. 815 */ 816 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 817 const struct in6_addr *saddr, const struct in6_addr *daddr, 818 struct udp_table *udptable, int proto) 819 { 820 struct sock *sk, *first = NULL; 821 const struct udphdr *uh = udp_hdr(skb); 822 unsigned short hnum = ntohs(uh->dest); 823 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 824 unsigned int offset = offsetof(typeof(*sk), sk_node); 825 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 826 int dif = inet6_iif(skb); 827 int sdif = inet6_sdif(skb); 828 struct hlist_node *node; 829 struct sk_buff *nskb; 830 831 if (use_hash2) { 832 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 833 udptable->mask; 834 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 835 start_lookup: 836 hslot = &udptable->hash2[hash2]; 837 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 838 } 839 840 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 841 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 842 uh->source, saddr, dif, sdif, 843 hnum)) 844 continue; 845 /* If zero checksum and no_check is not on for 846 * the socket then skip it. 847 */ 848 if (!uh->check && !udp_sk(sk)->no_check6_rx) 849 continue; 850 if (!first) { 851 first = sk; 852 continue; 853 } 854 nskb = skb_clone(skb, GFP_ATOMIC); 855 if (unlikely(!nskb)) { 856 atomic_inc(&sk->sk_drops); 857 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 858 IS_UDPLITE(sk)); 859 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 860 IS_UDPLITE(sk)); 861 continue; 862 } 863 864 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 865 consume_skb(nskb); 866 } 867 868 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 869 if (use_hash2 && hash2 != hash2_any) { 870 hash2 = hash2_any; 871 goto start_lookup; 872 } 873 874 if (first) { 875 if (udpv6_queue_rcv_skb(first, skb) > 0) 876 consume_skb(skb); 877 } else { 878 kfree_skb(skb); 879 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 880 proto == IPPROTO_UDPLITE); 881 } 882 return 0; 883 } 884 885 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 886 { 887 if (udp_sk_rx_dst_set(sk, dst)) { 888 const struct rt6_info *rt = (const struct rt6_info *)dst; 889 890 sk->sk_rx_dst_cookie = rt6_get_cookie(rt); 891 } 892 } 893 894 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 895 * return code conversion for ip layer consumption 896 */ 897 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 898 struct udphdr *uh) 899 { 900 int ret; 901 902 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 903 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 904 905 ret = udpv6_queue_rcv_skb(sk, skb); 906 907 /* a return value > 0 means to resubmit the input */ 908 if (ret > 0) 909 return ret; 910 return 0; 911 } 912 913 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 914 int proto) 915 { 916 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 917 const struct in6_addr *saddr, *daddr; 918 struct net *net = dev_net(skb->dev); 919 struct udphdr *uh; 920 struct sock *sk; 921 bool refcounted; 922 u32 ulen = 0; 923 924 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 925 goto discard; 926 927 saddr = &ipv6_hdr(skb)->saddr; 928 daddr = &ipv6_hdr(skb)->daddr; 929 uh = udp_hdr(skb); 930 931 ulen = ntohs(uh->len); 932 if (ulen > skb->len) 933 goto short_packet; 934 935 if (proto == IPPROTO_UDP) { 936 /* UDP validates ulen. */ 937 938 /* Check for jumbo payload */ 939 if (ulen == 0) 940 ulen = skb->len; 941 942 if (ulen < sizeof(*uh)) 943 goto short_packet; 944 945 if (ulen < skb->len) { 946 if (pskb_trim_rcsum(skb, ulen)) 947 goto short_packet; 948 saddr = &ipv6_hdr(skb)->saddr; 949 daddr = &ipv6_hdr(skb)->daddr; 950 uh = udp_hdr(skb); 951 } 952 } 953 954 if (udp6_csum_init(skb, uh, proto)) 955 goto csum_error; 956 957 /* Check if the socket is already available, e.g. due to early demux */ 958 sk = skb_steal_sock(skb, &refcounted); 959 if (sk) { 960 struct dst_entry *dst = skb_dst(skb); 961 int ret; 962 963 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 964 udp6_sk_rx_dst_set(sk, dst); 965 966 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 967 if (refcounted) 968 sock_put(sk); 969 goto report_csum_error; 970 } 971 972 ret = udp6_unicast_rcv_skb(sk, skb, uh); 973 if (refcounted) 974 sock_put(sk); 975 return ret; 976 } 977 978 /* 979 * Multicast receive code 980 */ 981 if (ipv6_addr_is_multicast(daddr)) 982 return __udp6_lib_mcast_deliver(net, skb, 983 saddr, daddr, udptable, proto); 984 985 /* Unicast */ 986 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 987 if (sk) { 988 if (!uh->check && !udp_sk(sk)->no_check6_rx) 989 goto report_csum_error; 990 return udp6_unicast_rcv_skb(sk, skb, uh); 991 } 992 993 reason = SKB_DROP_REASON_NO_SOCKET; 994 995 if (!uh->check) 996 goto report_csum_error; 997 998 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 999 goto discard; 1000 1001 if (udp_lib_checksum_complete(skb)) 1002 goto csum_error; 1003 1004 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1005 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 1006 1007 kfree_skb_reason(skb, reason); 1008 return 0; 1009 1010 short_packet: 1011 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1012 reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1013 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 1014 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1015 saddr, ntohs(uh->source), 1016 ulen, skb->len, 1017 daddr, ntohs(uh->dest)); 1018 goto discard; 1019 1020 report_csum_error: 1021 udp6_csum_zero_error(skb); 1022 csum_error: 1023 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1024 reason = SKB_DROP_REASON_UDP_CSUM; 1025 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1026 discard: 1027 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1028 kfree_skb_reason(skb, reason); 1029 return 0; 1030 } 1031 1032 1033 static struct sock *__udp6_lib_demux_lookup(struct net *net, 1034 __be16 loc_port, const struct in6_addr *loc_addr, 1035 __be16 rmt_port, const struct in6_addr *rmt_addr, 1036 int dif, int sdif) 1037 { 1038 unsigned short hnum = ntohs(loc_port); 1039 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1040 unsigned int slot2 = hash2 & udp_table.mask; 1041 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1042 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1043 struct sock *sk; 1044 1045 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1046 if (sk->sk_state == TCP_ESTABLISHED && 1047 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif)) 1048 return sk; 1049 /* Only check first socket in chain */ 1050 break; 1051 } 1052 return NULL; 1053 } 1054 1055 void udp_v6_early_demux(struct sk_buff *skb) 1056 { 1057 struct net *net = dev_net(skb->dev); 1058 const struct udphdr *uh; 1059 struct sock *sk; 1060 struct dst_entry *dst; 1061 int dif = skb->dev->ifindex; 1062 int sdif = inet6_sdif(skb); 1063 1064 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1065 sizeof(struct udphdr))) 1066 return; 1067 1068 uh = udp_hdr(skb); 1069 1070 if (skb->pkt_type == PACKET_HOST) 1071 sk = __udp6_lib_demux_lookup(net, uh->dest, 1072 &ipv6_hdr(skb)->daddr, 1073 uh->source, &ipv6_hdr(skb)->saddr, 1074 dif, sdif); 1075 else 1076 return; 1077 1078 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1079 return; 1080 1081 skb->sk = sk; 1082 skb->destructor = sock_efree; 1083 dst = rcu_dereference(sk->sk_rx_dst); 1084 1085 if (dst) 1086 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1087 if (dst) { 1088 /* set noref for now. 1089 * any place which wants to hold dst has to call 1090 * dst_hold_safe() 1091 */ 1092 skb_dst_set_noref(skb, dst); 1093 } 1094 } 1095 1096 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1097 { 1098 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1099 } 1100 1101 /* 1102 * Throw away all pending data and cancel the corking. Socket is locked. 1103 */ 1104 static void udp_v6_flush_pending_frames(struct sock *sk) 1105 { 1106 struct udp_sock *up = udp_sk(sk); 1107 1108 if (up->pending == AF_INET) 1109 udp_flush_pending_frames(sk); 1110 else if (up->pending) { 1111 up->len = 0; 1112 up->pending = 0; 1113 ip6_flush_pending_frames(sk); 1114 } 1115 } 1116 1117 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1118 int addr_len) 1119 { 1120 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1121 return -EINVAL; 1122 /* The following checks are replicated from __ip6_datagram_connect() 1123 * and intended to prevent BPF program called below from accessing 1124 * bytes that are out of the bound specified by user in addr_len. 1125 */ 1126 if (uaddr->sa_family == AF_INET) { 1127 if (ipv6_only_sock(sk)) 1128 return -EAFNOSUPPORT; 1129 return udp_pre_connect(sk, uaddr, addr_len); 1130 } 1131 1132 if (addr_len < SIN6_LEN_RFC2133) 1133 return -EINVAL; 1134 1135 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1136 } 1137 1138 /** 1139 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1140 * @sk: socket we are sending on 1141 * @skb: sk_buff containing the filled-in UDP header 1142 * (checksum field must be zeroed out) 1143 * @saddr: source address 1144 * @daddr: destination address 1145 * @len: length of packet 1146 */ 1147 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1148 const struct in6_addr *saddr, 1149 const struct in6_addr *daddr, int len) 1150 { 1151 unsigned int offset; 1152 struct udphdr *uh = udp_hdr(skb); 1153 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1154 __wsum csum = 0; 1155 1156 if (!frags) { 1157 /* Only one fragment on the socket. */ 1158 skb->csum_start = skb_transport_header(skb) - skb->head; 1159 skb->csum_offset = offsetof(struct udphdr, check); 1160 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1161 } else { 1162 /* 1163 * HW-checksum won't work as there are two or more 1164 * fragments on the socket so that all csums of sk_buffs 1165 * should be together 1166 */ 1167 offset = skb_transport_offset(skb); 1168 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1169 csum = skb->csum; 1170 1171 skb->ip_summed = CHECKSUM_NONE; 1172 1173 do { 1174 csum = csum_add(csum, frags->csum); 1175 } while ((frags = frags->next)); 1176 1177 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1178 csum); 1179 if (uh->check == 0) 1180 uh->check = CSUM_MANGLED_0; 1181 } 1182 } 1183 1184 /* 1185 * Sending 1186 */ 1187 1188 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1189 struct inet_cork *cork) 1190 { 1191 struct sock *sk = skb->sk; 1192 struct udphdr *uh; 1193 int err = 0; 1194 int is_udplite = IS_UDPLITE(sk); 1195 __wsum csum = 0; 1196 int offset = skb_transport_offset(skb); 1197 int len = skb->len - offset; 1198 int datalen = len - sizeof(*uh); 1199 1200 /* 1201 * Create a UDP header 1202 */ 1203 uh = udp_hdr(skb); 1204 uh->source = fl6->fl6_sport; 1205 uh->dest = fl6->fl6_dport; 1206 uh->len = htons(len); 1207 uh->check = 0; 1208 1209 if (cork->gso_size) { 1210 const int hlen = skb_network_header_len(skb) + 1211 sizeof(struct udphdr); 1212 1213 if (hlen + cork->gso_size > cork->fragsize) { 1214 kfree_skb(skb); 1215 return -EINVAL; 1216 } 1217 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1218 kfree_skb(skb); 1219 return -EINVAL; 1220 } 1221 if (udp_sk(sk)->no_check6_tx) { 1222 kfree_skb(skb); 1223 return -EINVAL; 1224 } 1225 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1226 dst_xfrm(skb_dst(skb))) { 1227 kfree_skb(skb); 1228 return -EIO; 1229 } 1230 1231 if (datalen > cork->gso_size) { 1232 skb_shinfo(skb)->gso_size = cork->gso_size; 1233 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1234 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1235 cork->gso_size); 1236 } 1237 goto csum_partial; 1238 } 1239 1240 if (is_udplite) 1241 csum = udplite_csum(skb); 1242 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1243 skb->ip_summed = CHECKSUM_NONE; 1244 goto send; 1245 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1246 csum_partial: 1247 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1248 goto send; 1249 } else 1250 csum = udp_csum(skb); 1251 1252 /* add protocol-dependent pseudo-header */ 1253 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1254 len, fl6->flowi6_proto, csum); 1255 if (uh->check == 0) 1256 uh->check = CSUM_MANGLED_0; 1257 1258 send: 1259 err = ip6_send_skb(skb); 1260 if (err) { 1261 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1262 UDP6_INC_STATS(sock_net(sk), 1263 UDP_MIB_SNDBUFERRORS, is_udplite); 1264 err = 0; 1265 } 1266 } else { 1267 UDP6_INC_STATS(sock_net(sk), 1268 UDP_MIB_OUTDATAGRAMS, is_udplite); 1269 } 1270 return err; 1271 } 1272 1273 static int udp_v6_push_pending_frames(struct sock *sk) 1274 { 1275 struct sk_buff *skb; 1276 struct udp_sock *up = udp_sk(sk); 1277 int err = 0; 1278 1279 if (up->pending == AF_INET) 1280 return udp_push_pending_frames(sk); 1281 1282 skb = ip6_finish_skb(sk); 1283 if (!skb) 1284 goto out; 1285 1286 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6, 1287 &inet_sk(sk)->cork.base); 1288 out: 1289 up->len = 0; 1290 up->pending = 0; 1291 return err; 1292 } 1293 1294 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1295 { 1296 struct ipv6_txoptions opt_space; 1297 struct udp_sock *up = udp_sk(sk); 1298 struct inet_sock *inet = inet_sk(sk); 1299 struct ipv6_pinfo *np = inet6_sk(sk); 1300 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1301 struct in6_addr *daddr, *final_p, final; 1302 struct ipv6_txoptions *opt = NULL; 1303 struct ipv6_txoptions *opt_to_free = NULL; 1304 struct ip6_flowlabel *flowlabel = NULL; 1305 struct inet_cork_full cork; 1306 struct flowi6 *fl6 = &cork.fl.u.ip6; 1307 struct dst_entry *dst; 1308 struct ipcm6_cookie ipc6; 1309 int addr_len = msg->msg_namelen; 1310 bool connected = false; 1311 int ulen = len; 1312 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1313 int err; 1314 int is_udplite = IS_UDPLITE(sk); 1315 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1316 1317 ipcm6_init(&ipc6); 1318 ipc6.gso_size = READ_ONCE(up->gso_size); 1319 ipc6.sockc.tsflags = sk->sk_tsflags; 1320 ipc6.sockc.mark = sk->sk_mark; 1321 1322 /* destination address check */ 1323 if (sin6) { 1324 if (addr_len < offsetof(struct sockaddr, sa_data)) 1325 return -EINVAL; 1326 1327 switch (sin6->sin6_family) { 1328 case AF_INET6: 1329 if (addr_len < SIN6_LEN_RFC2133) 1330 return -EINVAL; 1331 daddr = &sin6->sin6_addr; 1332 if (ipv6_addr_any(daddr) && 1333 ipv6_addr_v4mapped(&np->saddr)) 1334 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1335 daddr); 1336 break; 1337 case AF_INET: 1338 goto do_udp_sendmsg; 1339 case AF_UNSPEC: 1340 msg->msg_name = sin6 = NULL; 1341 msg->msg_namelen = addr_len = 0; 1342 daddr = NULL; 1343 break; 1344 default: 1345 return -EINVAL; 1346 } 1347 } else if (!up->pending) { 1348 if (sk->sk_state != TCP_ESTABLISHED) 1349 return -EDESTADDRREQ; 1350 daddr = &sk->sk_v6_daddr; 1351 } else 1352 daddr = NULL; 1353 1354 if (daddr) { 1355 if (ipv6_addr_v4mapped(daddr)) { 1356 struct sockaddr_in sin; 1357 sin.sin_family = AF_INET; 1358 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1359 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1360 msg->msg_name = &sin; 1361 msg->msg_namelen = sizeof(sin); 1362 do_udp_sendmsg: 1363 if (ipv6_only_sock(sk)) 1364 return -ENETUNREACH; 1365 return udp_sendmsg(sk, msg, len); 1366 } 1367 } 1368 1369 /* Rough check on arithmetic overflow, 1370 better check is made in ip6_append_data(). 1371 */ 1372 if (len > INT_MAX - sizeof(struct udphdr)) 1373 return -EMSGSIZE; 1374 1375 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1376 if (up->pending) { 1377 if (up->pending == AF_INET) 1378 return udp_sendmsg(sk, msg, len); 1379 /* 1380 * There are pending frames. 1381 * The socket lock must be held while it's corked. 1382 */ 1383 lock_sock(sk); 1384 if (likely(up->pending)) { 1385 if (unlikely(up->pending != AF_INET6)) { 1386 release_sock(sk); 1387 return -EAFNOSUPPORT; 1388 } 1389 dst = NULL; 1390 goto do_append_data; 1391 } 1392 release_sock(sk); 1393 } 1394 ulen += sizeof(struct udphdr); 1395 1396 memset(fl6, 0, sizeof(*fl6)); 1397 1398 if (sin6) { 1399 if (sin6->sin6_port == 0) 1400 return -EINVAL; 1401 1402 fl6->fl6_dport = sin6->sin6_port; 1403 daddr = &sin6->sin6_addr; 1404 1405 if (np->sndflow) { 1406 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1407 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) { 1408 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1409 if (IS_ERR(flowlabel)) 1410 return -EINVAL; 1411 } 1412 } 1413 1414 /* 1415 * Otherwise it will be difficult to maintain 1416 * sk->sk_dst_cache. 1417 */ 1418 if (sk->sk_state == TCP_ESTABLISHED && 1419 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1420 daddr = &sk->sk_v6_daddr; 1421 1422 if (addr_len >= sizeof(struct sockaddr_in6) && 1423 sin6->sin6_scope_id && 1424 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1425 fl6->flowi6_oif = sin6->sin6_scope_id; 1426 } else { 1427 if (sk->sk_state != TCP_ESTABLISHED) 1428 return -EDESTADDRREQ; 1429 1430 fl6->fl6_dport = inet->inet_dport; 1431 daddr = &sk->sk_v6_daddr; 1432 fl6->flowlabel = np->flow_label; 1433 connected = true; 1434 } 1435 1436 if (!fl6->flowi6_oif) 1437 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if); 1438 1439 if (!fl6->flowi6_oif) 1440 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1441 1442 fl6->flowi6_uid = sk->sk_uid; 1443 1444 if (msg->msg_controllen) { 1445 opt = &opt_space; 1446 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1447 opt->tot_len = sizeof(*opt); 1448 ipc6.opt = opt; 1449 1450 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1451 if (err > 0) 1452 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6, 1453 &ipc6); 1454 if (err < 0) { 1455 fl6_sock_release(flowlabel); 1456 return err; 1457 } 1458 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1459 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1460 if (IS_ERR(flowlabel)) 1461 return -EINVAL; 1462 } 1463 if (!(opt->opt_nflen|opt->opt_flen)) 1464 opt = NULL; 1465 connected = false; 1466 } 1467 if (!opt) { 1468 opt = txopt_get(np); 1469 opt_to_free = opt; 1470 } 1471 if (flowlabel) 1472 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1473 opt = ipv6_fixup_options(&opt_space, opt); 1474 ipc6.opt = opt; 1475 1476 fl6->flowi6_proto = sk->sk_protocol; 1477 fl6->flowi6_mark = ipc6.sockc.mark; 1478 fl6->daddr = *daddr; 1479 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr)) 1480 fl6->saddr = np->saddr; 1481 fl6->fl6_sport = inet->inet_sport; 1482 1483 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1484 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1485 (struct sockaddr *)sin6, 1486 &fl6->saddr); 1487 if (err) 1488 goto out_no_dst; 1489 if (sin6) { 1490 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1491 /* BPF program rewrote IPv6-only by IPv4-mapped 1492 * IPv6. It's currently unsupported. 1493 */ 1494 err = -ENOTSUPP; 1495 goto out_no_dst; 1496 } 1497 if (sin6->sin6_port == 0) { 1498 /* BPF program set invalid port. Reject it. */ 1499 err = -EINVAL; 1500 goto out_no_dst; 1501 } 1502 fl6->fl6_dport = sin6->sin6_port; 1503 fl6->daddr = sin6->sin6_addr; 1504 } 1505 } 1506 1507 if (ipv6_addr_any(&fl6->daddr)) 1508 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1509 1510 final_p = fl6_update_dst(fl6, opt, &final); 1511 if (final_p) 1512 connected = false; 1513 1514 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) { 1515 fl6->flowi6_oif = np->mcast_oif; 1516 connected = false; 1517 } else if (!fl6->flowi6_oif) 1518 fl6->flowi6_oif = np->ucast_oif; 1519 1520 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6)); 1521 1522 if (ipc6.tclass < 0) 1523 ipc6.tclass = np->tclass; 1524 1525 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel); 1526 1527 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected); 1528 if (IS_ERR(dst)) { 1529 err = PTR_ERR(dst); 1530 dst = NULL; 1531 goto out; 1532 } 1533 1534 if (ipc6.hlimit < 0) 1535 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst); 1536 1537 if (msg->msg_flags&MSG_CONFIRM) 1538 goto do_confirm; 1539 back_from_confirm: 1540 1541 /* Lockless fast path for the non-corking case */ 1542 if (!corkreq) { 1543 struct sk_buff *skb; 1544 1545 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1546 sizeof(struct udphdr), &ipc6, 1547 (struct rt6_info *)dst, 1548 msg->msg_flags, &cork); 1549 err = PTR_ERR(skb); 1550 if (!IS_ERR_OR_NULL(skb)) 1551 err = udp_v6_send_skb(skb, fl6, &cork.base); 1552 /* ip6_make_skb steals dst reference */ 1553 goto out_no_dst; 1554 } 1555 1556 lock_sock(sk); 1557 if (unlikely(up->pending)) { 1558 /* The socket is already corked while preparing it. */ 1559 /* ... which is an evident application bug. --ANK */ 1560 release_sock(sk); 1561 1562 net_dbg_ratelimited("udp cork app bug 2\n"); 1563 err = -EINVAL; 1564 goto out; 1565 } 1566 1567 up->pending = AF_INET6; 1568 1569 do_append_data: 1570 if (ipc6.dontfrag < 0) 1571 ipc6.dontfrag = np->dontfrag; 1572 up->len += ulen; 1573 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1574 &ipc6, fl6, (struct rt6_info *)dst, 1575 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1576 if (err) 1577 udp_v6_flush_pending_frames(sk); 1578 else if (!corkreq) 1579 err = udp_v6_push_pending_frames(sk); 1580 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1581 up->pending = 0; 1582 1583 if (err > 0) 1584 err = np->recverr ? net_xmit_errno(err) : 0; 1585 release_sock(sk); 1586 1587 out: 1588 dst_release(dst); 1589 out_no_dst: 1590 fl6_sock_release(flowlabel); 1591 txopt_put(opt_to_free); 1592 if (!err) 1593 return len; 1594 /* 1595 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1596 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1597 * we don't have a good statistic (IpOutDiscards but it can be too many 1598 * things). We could add another new stat but at least for now that 1599 * seems like overkill. 1600 */ 1601 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1602 UDP6_INC_STATS(sock_net(sk), 1603 UDP_MIB_SNDBUFERRORS, is_udplite); 1604 } 1605 return err; 1606 1607 do_confirm: 1608 if (msg->msg_flags & MSG_PROBE) 1609 dst_confirm_neigh(dst, &fl6->daddr); 1610 if (!(msg->msg_flags&MSG_PROBE) || len) 1611 goto back_from_confirm; 1612 err = 0; 1613 goto out; 1614 } 1615 1616 void udpv6_destroy_sock(struct sock *sk) 1617 { 1618 struct udp_sock *up = udp_sk(sk); 1619 lock_sock(sk); 1620 1621 /* protects from races with udp_abort() */ 1622 sock_set_flag(sk, SOCK_DEAD); 1623 udp_v6_flush_pending_frames(sk); 1624 release_sock(sk); 1625 1626 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1627 if (up->encap_type) { 1628 void (*encap_destroy)(struct sock *sk); 1629 encap_destroy = READ_ONCE(up->encap_destroy); 1630 if (encap_destroy) 1631 encap_destroy(sk); 1632 } 1633 if (up->encap_enabled) { 1634 static_branch_dec(&udpv6_encap_needed_key); 1635 udp_encap_disable(); 1636 } 1637 } 1638 1639 inet6_destroy_sock(sk); 1640 } 1641 1642 /* 1643 * Socket option code for UDP 1644 */ 1645 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1646 unsigned int optlen) 1647 { 1648 if (level == SOL_UDP || level == SOL_UDPLITE) 1649 return udp_lib_setsockopt(sk, level, optname, 1650 optval, optlen, 1651 udp_v6_push_pending_frames); 1652 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1653 } 1654 1655 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1656 char __user *optval, int __user *optlen) 1657 { 1658 if (level == SOL_UDP || level == SOL_UDPLITE) 1659 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1660 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1661 } 1662 1663 static const struct inet6_protocol udpv6_protocol = { 1664 .handler = udpv6_rcv, 1665 .err_handler = udpv6_err, 1666 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1667 }; 1668 1669 /* ------------------------------------------------------------------------ */ 1670 #ifdef CONFIG_PROC_FS 1671 int udp6_seq_show(struct seq_file *seq, void *v) 1672 { 1673 if (v == SEQ_START_TOKEN) { 1674 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1675 } else { 1676 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1677 struct inet_sock *inet = inet_sk(v); 1678 __u16 srcp = ntohs(inet->inet_sport); 1679 __u16 destp = ntohs(inet->inet_dport); 1680 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1681 udp_rqueue_get(v), bucket); 1682 } 1683 return 0; 1684 } 1685 1686 const struct seq_operations udp6_seq_ops = { 1687 .start = udp_seq_start, 1688 .next = udp_seq_next, 1689 .stop = udp_seq_stop, 1690 .show = udp6_seq_show, 1691 }; 1692 EXPORT_SYMBOL(udp6_seq_ops); 1693 1694 static struct udp_seq_afinfo udp6_seq_afinfo = { 1695 .family = AF_INET6, 1696 .udp_table = &udp_table, 1697 }; 1698 1699 int __net_init udp6_proc_init(struct net *net) 1700 { 1701 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1702 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1703 return -ENOMEM; 1704 return 0; 1705 } 1706 1707 void udp6_proc_exit(struct net *net) 1708 { 1709 remove_proc_entry("udp6", net->proc_net); 1710 } 1711 #endif /* CONFIG_PROC_FS */ 1712 1713 /* ------------------------------------------------------------------------ */ 1714 1715 struct proto udpv6_prot = { 1716 .name = "UDPv6", 1717 .owner = THIS_MODULE, 1718 .close = udp_lib_close, 1719 .pre_connect = udpv6_pre_connect, 1720 .connect = ip6_datagram_connect, 1721 .disconnect = udp_disconnect, 1722 .ioctl = udp_ioctl, 1723 .init = udp_init_sock, 1724 .destroy = udpv6_destroy_sock, 1725 .setsockopt = udpv6_setsockopt, 1726 .getsockopt = udpv6_getsockopt, 1727 .sendmsg = udpv6_sendmsg, 1728 .recvmsg = udpv6_recvmsg, 1729 .release_cb = ip6_datagram_release_cb, 1730 .hash = udp_lib_hash, 1731 .unhash = udp_lib_unhash, 1732 .rehash = udp_v6_rehash, 1733 .get_port = udp_v6_get_port, 1734 .put_port = udp_lib_unhash, 1735 #ifdef CONFIG_BPF_SYSCALL 1736 .psock_update_sk_prot = udp_bpf_update_proto, 1737 #endif 1738 1739 .memory_allocated = &udp_memory_allocated, 1740 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 1741 1742 .sysctl_mem = sysctl_udp_mem, 1743 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1744 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1745 .obj_size = sizeof(struct udp6_sock), 1746 .h.udp_table = &udp_table, 1747 .diag_destroy = udp_abort, 1748 }; 1749 1750 static struct inet_protosw udpv6_protosw = { 1751 .type = SOCK_DGRAM, 1752 .protocol = IPPROTO_UDP, 1753 .prot = &udpv6_prot, 1754 .ops = &inet6_dgram_ops, 1755 .flags = INET_PROTOSW_PERMANENT, 1756 }; 1757 1758 int __init udpv6_init(void) 1759 { 1760 int ret; 1761 1762 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1763 if (ret) 1764 goto out; 1765 1766 ret = inet6_register_protosw(&udpv6_protosw); 1767 if (ret) 1768 goto out_udpv6_protocol; 1769 out: 1770 return ret; 1771 1772 out_udpv6_protocol: 1773 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1774 goto out; 1775 } 1776 1777 void udpv6_exit(void) 1778 { 1779 inet6_unregister_protosw(&udpv6_protosw); 1780 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1781 } 1782