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/tcp_states.h> 45 #include <net/ip6_checksum.h> 46 #include <net/ip6_tunnel.h> 47 #include <net/xfrm.h> 48 #include <net/inet_hashtables.h> 49 #include <net/inet6_hashtables.h> 50 #include <net/busy_poll.h> 51 #include <net/sock_reuseport.h> 52 53 #include <linux/proc_fs.h> 54 #include <linux/seq_file.h> 55 #include <trace/events/skb.h> 56 #include "udp_impl.h" 57 58 static u32 udp6_ehashfn(const struct net *net, 59 const struct in6_addr *laddr, 60 const u16 lport, 61 const struct in6_addr *faddr, 62 const __be16 fport) 63 { 64 static u32 udp6_ehash_secret __read_mostly; 65 static u32 udp_ipv6_hash_secret __read_mostly; 66 67 u32 lhash, fhash; 68 69 net_get_random_once(&udp6_ehash_secret, 70 sizeof(udp6_ehash_secret)); 71 net_get_random_once(&udp_ipv6_hash_secret, 72 sizeof(udp_ipv6_hash_secret)); 73 74 lhash = (__force u32)laddr->s6_addr32[3]; 75 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 76 77 return __inet6_ehashfn(lhash, lport, fhash, fport, 78 udp_ipv6_hash_secret + net_hash_mix(net)); 79 } 80 81 int udp_v6_get_port(struct sock *sk, unsigned short snum) 82 { 83 unsigned int hash2_nulladdr = 84 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 85 unsigned int hash2_partial = 86 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 87 88 /* precompute partial secondary hash */ 89 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 90 return udp_lib_get_port(sk, snum, hash2_nulladdr); 91 } 92 93 void udp_v6_rehash(struct sock *sk) 94 { 95 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 96 &sk->sk_v6_rcv_saddr, 97 inet_sk(sk)->inet_num); 98 99 udp_lib_rehash(sk, new_hash); 100 } 101 102 static int compute_score(struct sock *sk, struct net *net, 103 const struct in6_addr *saddr, __be16 sport, 104 const struct in6_addr *daddr, unsigned short hnum, 105 int dif, int sdif) 106 { 107 int score; 108 struct inet_sock *inet; 109 bool dev_match; 110 111 if (!net_eq(sock_net(sk), net) || 112 udp_sk(sk)->udp_port_hash != hnum || 113 sk->sk_family != PF_INET6) 114 return -1; 115 116 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 117 return -1; 118 119 score = 0; 120 inet = inet_sk(sk); 121 122 if (inet->inet_dport) { 123 if (inet->inet_dport != sport) 124 return -1; 125 score++; 126 } 127 128 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 129 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 130 return -1; 131 score++; 132 } 133 134 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif); 135 if (!dev_match) 136 return -1; 137 if (sk->sk_bound_dev_if) 138 score++; 139 140 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 141 score++; 142 143 return score; 144 } 145 146 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 147 struct sk_buff *skb, 148 const struct in6_addr *saddr, 149 __be16 sport, 150 const struct in6_addr *daddr, 151 unsigned int hnum) 152 { 153 struct sock *reuse_sk = NULL; 154 u32 hash; 155 156 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 157 hash = udp6_ehashfn(net, daddr, hnum, saddr, sport); 158 reuse_sk = reuseport_select_sock(sk, hash, skb, 159 sizeof(struct udphdr)); 160 } 161 return reuse_sk; 162 } 163 164 /* called with rcu_read_lock() */ 165 static struct sock *udp6_lib_lookup2(struct net *net, 166 const struct in6_addr *saddr, __be16 sport, 167 const struct in6_addr *daddr, unsigned int hnum, 168 int dif, int sdif, struct udp_hslot *hslot2, 169 struct sk_buff *skb) 170 { 171 struct sock *sk, *result; 172 int score, badness; 173 174 result = NULL; 175 badness = -1; 176 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 177 score = compute_score(sk, net, saddr, sport, 178 daddr, hnum, dif, sdif); 179 if (score > badness) { 180 result = lookup_reuseport(net, sk, skb, 181 saddr, sport, daddr, hnum); 182 /* Fall back to scoring if group has connections */ 183 if (result && !reuseport_has_conns(sk, false)) 184 return result; 185 186 result = result ? : sk; 187 badness = score; 188 } 189 } 190 return result; 191 } 192 193 static inline struct sock *udp6_lookup_run_bpf(struct net *net, 194 struct udp_table *udptable, 195 struct sk_buff *skb, 196 const struct in6_addr *saddr, 197 __be16 sport, 198 const struct in6_addr *daddr, 199 u16 hnum, const int dif) 200 { 201 struct sock *sk, *reuse_sk; 202 bool no_reuseport; 203 204 if (udptable != &udp_table) 205 return NULL; /* only UDP is supported */ 206 207 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_UDP, saddr, sport, 208 daddr, hnum, dif, &sk); 209 if (no_reuseport || IS_ERR_OR_NULL(sk)) 210 return sk; 211 212 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 213 if (reuse_sk) 214 sk = reuse_sk; 215 return sk; 216 } 217 218 /* rcu_read_lock() must be held */ 219 struct sock *__udp6_lib_lookup(struct net *net, 220 const struct in6_addr *saddr, __be16 sport, 221 const struct in6_addr *daddr, __be16 dport, 222 int dif, int sdif, struct udp_table *udptable, 223 struct sk_buff *skb) 224 { 225 unsigned short hnum = ntohs(dport); 226 unsigned int hash2, slot2; 227 struct udp_hslot *hslot2; 228 struct sock *result, *sk; 229 230 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 231 slot2 = hash2 & udptable->mask; 232 hslot2 = &udptable->hash2[slot2]; 233 234 /* Lookup connected or non-wildcard sockets */ 235 result = udp6_lib_lookup2(net, saddr, sport, 236 daddr, hnum, dif, sdif, 237 hslot2, skb); 238 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 239 goto done; 240 241 /* Lookup redirect from BPF */ 242 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { 243 sk = udp6_lookup_run_bpf(net, udptable, skb, 244 saddr, sport, daddr, hnum, dif); 245 if (sk) { 246 result = sk; 247 goto done; 248 } 249 } 250 251 /* Got non-wildcard socket or error on first lookup */ 252 if (result) 253 goto done; 254 255 /* Lookup wildcard sockets */ 256 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 257 slot2 = hash2 & udptable->mask; 258 hslot2 = &udptable->hash2[slot2]; 259 260 result = udp6_lib_lookup2(net, saddr, sport, 261 &in6addr_any, hnum, dif, sdif, 262 hslot2, skb); 263 done: 264 if (IS_ERR(result)) 265 return NULL; 266 return result; 267 } 268 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 269 270 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 271 __be16 sport, __be16 dport, 272 struct udp_table *udptable) 273 { 274 const struct ipv6hdr *iph = ipv6_hdr(skb); 275 276 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 277 &iph->daddr, dport, inet6_iif(skb), 278 inet6_sdif(skb), udptable, skb); 279 } 280 281 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, 282 __be16 sport, __be16 dport) 283 { 284 const struct ipv6hdr *iph = ipv6_hdr(skb); 285 286 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 287 &iph->daddr, dport, inet6_iif(skb), 288 inet6_sdif(skb), &udp_table, NULL); 289 } 290 291 /* Must be called under rcu_read_lock(). 292 * Does increment socket refcount. 293 */ 294 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 295 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 296 const struct in6_addr *daddr, __be16 dport, int dif) 297 { 298 struct sock *sk; 299 300 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 301 dif, 0, &udp_table, NULL); 302 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 303 sk = NULL; 304 return sk; 305 } 306 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 307 #endif 308 309 /* do not use the scratch area len for jumbogram: their length execeeds the 310 * scratch area space; note that the IP6CB flags is still in the first 311 * cacheline, so checking for jumbograms is cheap 312 */ 313 static int udp6_skb_len(struct sk_buff *skb) 314 { 315 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 316 } 317 318 /* 319 * This should be easy, if there is something there we 320 * return it, otherwise we block. 321 */ 322 323 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 324 int noblock, int flags, int *addr_len) 325 { 326 struct ipv6_pinfo *np = inet6_sk(sk); 327 struct inet_sock *inet = inet_sk(sk); 328 struct sk_buff *skb; 329 unsigned int ulen, copied; 330 int off, err, peeking = flags & MSG_PEEK; 331 int is_udplite = IS_UDPLITE(sk); 332 struct udp_mib __percpu *mib; 333 bool checksum_valid = false; 334 int is_udp4; 335 336 if (flags & MSG_ERRQUEUE) 337 return ipv6_recv_error(sk, msg, len, addr_len); 338 339 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 340 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 341 342 try_again: 343 off = sk_peek_offset(sk, flags); 344 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 345 if (!skb) 346 return err; 347 348 ulen = udp6_skb_len(skb); 349 copied = len; 350 if (copied > ulen - off) 351 copied = ulen - off; 352 else if (copied < ulen) 353 msg->msg_flags |= MSG_TRUNC; 354 355 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 356 mib = __UDPX_MIB(sk, is_udp4); 357 358 /* 359 * If checksum is needed at all, try to do it while copying the 360 * data. If the data is truncated, or if we only want a partial 361 * coverage checksum (UDP-Lite), do it before the copy. 362 */ 363 364 if (copied < ulen || peeking || 365 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 366 checksum_valid = udp_skb_csum_unnecessary(skb) || 367 !__udp_lib_checksum_complete(skb); 368 if (!checksum_valid) 369 goto csum_copy_err; 370 } 371 372 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 373 if (udp_skb_is_linear(skb)) 374 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 375 else 376 err = skb_copy_datagram_msg(skb, off, msg, copied); 377 } else { 378 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 379 if (err == -EINVAL) 380 goto csum_copy_err; 381 } 382 if (unlikely(err)) { 383 if (!peeking) { 384 atomic_inc(&sk->sk_drops); 385 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 386 } 387 kfree_skb(skb); 388 return err; 389 } 390 if (!peeking) 391 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 392 393 sock_recv_ts_and_drops(msg, sk, skb); 394 395 /* Copy the address. */ 396 if (msg->msg_name) { 397 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 398 sin6->sin6_family = AF_INET6; 399 sin6->sin6_port = udp_hdr(skb)->source; 400 sin6->sin6_flowinfo = 0; 401 402 if (is_udp4) { 403 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 404 &sin6->sin6_addr); 405 sin6->sin6_scope_id = 0; 406 } else { 407 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 408 sin6->sin6_scope_id = 409 ipv6_iface_scope_id(&sin6->sin6_addr, 410 inet6_iif(skb)); 411 } 412 *addr_len = sizeof(*sin6); 413 414 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 415 (struct sockaddr *)sin6); 416 } 417 418 if (udp_sk(sk)->gro_enabled) 419 udp_cmsg_recv(msg, sk, skb); 420 421 if (np->rxopt.all) 422 ip6_datagram_recv_common_ctl(sk, msg, skb); 423 424 if (is_udp4) { 425 if (inet->cmsg_flags) 426 ip_cmsg_recv_offset(msg, sk, skb, 427 sizeof(struct udphdr), off); 428 } else { 429 if (np->rxopt.all) 430 ip6_datagram_recv_specific_ctl(sk, msg, skb); 431 } 432 433 err = copied; 434 if (flags & MSG_TRUNC) 435 err = ulen; 436 437 skb_consume_udp(sk, skb, peeking ? -err : err); 438 return err; 439 440 csum_copy_err: 441 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 442 udp_skb_destructor)) { 443 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 444 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 445 } 446 kfree_skb(skb); 447 448 /* starting over for a new packet, but check if we need to yield */ 449 cond_resched(); 450 msg->msg_flags &= ~MSG_TRUNC; 451 goto try_again; 452 } 453 454 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 455 void udpv6_encap_enable(void) 456 { 457 static_branch_inc(&udpv6_encap_needed_key); 458 } 459 EXPORT_SYMBOL(udpv6_encap_enable); 460 461 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 462 * through error handlers in encapsulations looking for a match. 463 */ 464 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 465 struct inet6_skb_parm *opt, 466 u8 type, u8 code, int offset, __be32 info) 467 { 468 int i; 469 470 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 471 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 472 u8 type, u8 code, int offset, __be32 info); 473 const struct ip6_tnl_encap_ops *encap; 474 475 encap = rcu_dereference(ip6tun_encaps[i]); 476 if (!encap) 477 continue; 478 handler = encap->err_handler; 479 if (handler && !handler(skb, opt, type, code, offset, info)) 480 return 0; 481 } 482 483 return -ENOENT; 484 } 485 486 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 487 * reversing source and destination port: this will match tunnels that force the 488 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 489 * lwtunnels might actually break this assumption by being configured with 490 * different destination ports on endpoints, in this case we won't be able to 491 * trace ICMP messages back to them. 492 * 493 * If this doesn't match any socket, probe tunnels with arbitrary destination 494 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 495 * we've sent packets to won't necessarily match the local destination port. 496 * 497 * Then ask the tunnel implementation to match the error against a valid 498 * association. 499 * 500 * Return an error if we can't find a match, the socket if we need further 501 * processing, zero otherwise. 502 */ 503 static struct sock *__udp6_lib_err_encap(struct net *net, 504 const struct ipv6hdr *hdr, int offset, 505 struct udphdr *uh, 506 struct udp_table *udptable, 507 struct sock *sk, 508 struct sk_buff *skb, 509 struct inet6_skb_parm *opt, 510 u8 type, u8 code, __be32 info) 511 { 512 int (*lookup)(struct sock *sk, struct sk_buff *skb); 513 int network_offset, transport_offset; 514 struct udp_sock *up; 515 516 network_offset = skb_network_offset(skb); 517 transport_offset = skb_transport_offset(skb); 518 519 /* Network header needs to point to the outer IPv6 header inside ICMP */ 520 skb_reset_network_header(skb); 521 522 /* Transport header needs to point to the UDP header */ 523 skb_set_transport_header(skb, offset); 524 525 if (sk) { 526 up = udp_sk(sk); 527 528 lookup = READ_ONCE(up->encap_err_lookup); 529 if (lookup && lookup(sk, skb)) 530 sk = NULL; 531 532 goto out; 533 } 534 535 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 536 &hdr->saddr, uh->dest, 537 inet6_iif(skb), 0, udptable, skb); 538 if (sk) { 539 up = udp_sk(sk); 540 541 lookup = READ_ONCE(up->encap_err_lookup); 542 if (!lookup || lookup(sk, skb)) 543 sk = NULL; 544 } 545 546 out: 547 if (!sk) { 548 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 549 offset, info)); 550 } 551 552 skb_set_transport_header(skb, transport_offset); 553 skb_set_network_header(skb, network_offset); 554 555 return sk; 556 } 557 558 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 559 u8 type, u8 code, int offset, __be32 info, 560 struct udp_table *udptable) 561 { 562 struct ipv6_pinfo *np; 563 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 564 const struct in6_addr *saddr = &hdr->saddr; 565 const struct in6_addr *daddr = &hdr->daddr; 566 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 567 bool tunnel = false; 568 struct sock *sk; 569 int harderr; 570 int err; 571 struct net *net = dev_net(skb->dev); 572 573 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 574 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 575 576 if (!sk || udp_sk(sk)->encap_type) { 577 /* No socket for error: try tunnels before discarding */ 578 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 579 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 580 udptable, sk, skb, 581 opt, type, code, info); 582 if (!sk) 583 return 0; 584 } else 585 sk = ERR_PTR(-ENOENT); 586 587 if (IS_ERR(sk)) { 588 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 589 ICMP6_MIB_INERRORS); 590 return PTR_ERR(sk); 591 } 592 593 tunnel = true; 594 } 595 596 harderr = icmpv6_err_convert(type, code, &err); 597 np = inet6_sk(sk); 598 599 if (type == ICMPV6_PKT_TOOBIG) { 600 if (!ip6_sk_accept_pmtu(sk)) 601 goto out; 602 ip6_sk_update_pmtu(skb, sk, info); 603 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 604 harderr = 1; 605 } 606 if (type == NDISC_REDIRECT) { 607 if (tunnel) { 608 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 609 sk->sk_mark, sk->sk_uid); 610 } else { 611 ip6_sk_redirect(skb, sk); 612 } 613 goto out; 614 } 615 616 /* Tunnels don't have an application socket: don't pass errors back */ 617 if (tunnel) 618 goto out; 619 620 if (!np->recverr) { 621 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 622 goto out; 623 } else { 624 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 625 } 626 627 sk->sk_err = err; 628 sk_error_report(sk); 629 out: 630 return 0; 631 } 632 633 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 634 { 635 int rc; 636 637 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 638 sock_rps_save_rxhash(sk, skb); 639 sk_mark_napi_id(sk, skb); 640 sk_incoming_cpu_update(sk); 641 } else { 642 sk_mark_napi_id_once(sk, skb); 643 } 644 645 rc = __udp_enqueue_schedule_skb(sk, skb); 646 if (rc < 0) { 647 int is_udplite = IS_UDPLITE(sk); 648 649 /* Note that an ENOMEM error is charged twice */ 650 if (rc == -ENOMEM) 651 UDP6_INC_STATS(sock_net(sk), 652 UDP_MIB_RCVBUFERRORS, is_udplite); 653 else 654 UDP6_INC_STATS(sock_net(sk), 655 UDP_MIB_MEMERRORS, is_udplite); 656 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 657 kfree_skb(skb); 658 return -1; 659 } 660 661 return 0; 662 } 663 664 static __inline__ int udpv6_err(struct sk_buff *skb, 665 struct inet6_skb_parm *opt, u8 type, 666 u8 code, int offset, __be32 info) 667 { 668 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 669 } 670 671 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 672 { 673 struct udp_sock *up = udp_sk(sk); 674 int is_udplite = IS_UDPLITE(sk); 675 676 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 677 goto drop; 678 679 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 680 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 681 682 /* 683 * This is an encapsulation socket so pass the skb to 684 * the socket's udp_encap_rcv() hook. Otherwise, just 685 * fall through and pass this up the UDP socket. 686 * up->encap_rcv() returns the following value: 687 * =0 if skb was successfully passed to the encap 688 * handler or was discarded by it. 689 * >0 if skb should be passed on to UDP. 690 * <0 if skb should be resubmitted as proto -N 691 */ 692 693 /* if we're overly short, let UDP handle it */ 694 encap_rcv = READ_ONCE(up->encap_rcv); 695 if (encap_rcv) { 696 int ret; 697 698 /* Verify checksum before giving to encap */ 699 if (udp_lib_checksum_complete(skb)) 700 goto csum_error; 701 702 ret = encap_rcv(sk, skb); 703 if (ret <= 0) { 704 __UDP6_INC_STATS(sock_net(sk), 705 UDP_MIB_INDATAGRAMS, 706 is_udplite); 707 return -ret; 708 } 709 } 710 711 /* FALLTHROUGH -- it's a UDP Packet */ 712 } 713 714 /* 715 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 716 */ 717 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 718 719 if (up->pcrlen == 0) { /* full coverage was set */ 720 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 721 UDP_SKB_CB(skb)->cscov, skb->len); 722 goto drop; 723 } 724 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 725 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 726 UDP_SKB_CB(skb)->cscov, up->pcrlen); 727 goto drop; 728 } 729 } 730 731 prefetch(&sk->sk_rmem_alloc); 732 if (rcu_access_pointer(sk->sk_filter) && 733 udp_lib_checksum_complete(skb)) 734 goto csum_error; 735 736 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 737 goto drop; 738 739 udp_csum_pull_header(skb); 740 741 skb_dst_drop(skb); 742 743 return __udpv6_queue_rcv_skb(sk, skb); 744 745 csum_error: 746 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 747 drop: 748 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 749 atomic_inc(&sk->sk_drops); 750 kfree_skb(skb); 751 return -1; 752 } 753 754 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 755 { 756 struct sk_buff *next, *segs; 757 int ret; 758 759 if (likely(!udp_unexpected_gso(sk, skb))) 760 return udpv6_queue_rcv_one_skb(sk, skb); 761 762 __skb_push(skb, -skb_mac_offset(skb)); 763 segs = udp_rcv_segment(sk, skb, false); 764 skb_list_walk_safe(segs, skb, next) { 765 __skb_pull(skb, skb_transport_offset(skb)); 766 767 udp_post_segment_fix_csum(skb); 768 ret = udpv6_queue_rcv_one_skb(sk, skb); 769 if (ret > 0) 770 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 771 true); 772 } 773 return 0; 774 } 775 776 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 777 __be16 loc_port, const struct in6_addr *loc_addr, 778 __be16 rmt_port, const struct in6_addr *rmt_addr, 779 int dif, int sdif, unsigned short hnum) 780 { 781 struct inet_sock *inet = inet_sk(sk); 782 783 if (!net_eq(sock_net(sk), net)) 784 return false; 785 786 if (udp_sk(sk)->udp_port_hash != hnum || 787 sk->sk_family != PF_INET6 || 788 (inet->inet_dport && inet->inet_dport != rmt_port) || 789 (!ipv6_addr_any(&sk->sk_v6_daddr) && 790 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 791 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) || 792 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 793 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 794 return false; 795 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 796 return false; 797 return true; 798 } 799 800 static void udp6_csum_zero_error(struct sk_buff *skb) 801 { 802 /* RFC 2460 section 8.1 says that we SHOULD log 803 * this error. Well, it is reasonable. 804 */ 805 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 806 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 807 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 808 } 809 810 /* 811 * Note: called only from the BH handler context, 812 * so we don't need to lock the hashes. 813 */ 814 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 815 const struct in6_addr *saddr, const struct in6_addr *daddr, 816 struct udp_table *udptable, int proto) 817 { 818 struct sock *sk, *first = NULL; 819 const struct udphdr *uh = udp_hdr(skb); 820 unsigned short hnum = ntohs(uh->dest); 821 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 822 unsigned int offset = offsetof(typeof(*sk), sk_node); 823 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 824 int dif = inet6_iif(skb); 825 int sdif = inet6_sdif(skb); 826 struct hlist_node *node; 827 struct sk_buff *nskb; 828 829 if (use_hash2) { 830 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 831 udptable->mask; 832 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 833 start_lookup: 834 hslot = &udptable->hash2[hash2]; 835 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 836 } 837 838 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 839 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 840 uh->source, saddr, dif, sdif, 841 hnum)) 842 continue; 843 /* If zero checksum and no_check is not on for 844 * the socket then skip it. 845 */ 846 if (!uh->check && !udp_sk(sk)->no_check6_rx) 847 continue; 848 if (!first) { 849 first = sk; 850 continue; 851 } 852 nskb = skb_clone(skb, GFP_ATOMIC); 853 if (unlikely(!nskb)) { 854 atomic_inc(&sk->sk_drops); 855 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 856 IS_UDPLITE(sk)); 857 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 858 IS_UDPLITE(sk)); 859 continue; 860 } 861 862 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 863 consume_skb(nskb); 864 } 865 866 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 867 if (use_hash2 && hash2 != hash2_any) { 868 hash2 = hash2_any; 869 goto start_lookup; 870 } 871 872 if (first) { 873 if (udpv6_queue_rcv_skb(first, skb) > 0) 874 consume_skb(skb); 875 } else { 876 kfree_skb(skb); 877 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 878 proto == IPPROTO_UDPLITE); 879 } 880 return 0; 881 } 882 883 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 884 { 885 if (udp_sk_rx_dst_set(sk, dst)) { 886 const struct rt6_info *rt = (const struct rt6_info *)dst; 887 888 sk->sk_rx_dst_cookie = rt6_get_cookie(rt); 889 } 890 } 891 892 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 893 * return code conversion for ip layer consumption 894 */ 895 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 896 struct udphdr *uh) 897 { 898 int ret; 899 900 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 901 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 902 903 ret = udpv6_queue_rcv_skb(sk, skb); 904 905 /* a return value > 0 means to resubmit the input */ 906 if (ret > 0) 907 return ret; 908 return 0; 909 } 910 911 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 912 int proto) 913 { 914 const struct in6_addr *saddr, *daddr; 915 struct net *net = dev_net(skb->dev); 916 struct udphdr *uh; 917 struct sock *sk; 918 bool refcounted; 919 u32 ulen = 0; 920 921 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 922 goto discard; 923 924 saddr = &ipv6_hdr(skb)->saddr; 925 daddr = &ipv6_hdr(skb)->daddr; 926 uh = udp_hdr(skb); 927 928 ulen = ntohs(uh->len); 929 if (ulen > skb->len) 930 goto short_packet; 931 932 if (proto == IPPROTO_UDP) { 933 /* UDP validates ulen. */ 934 935 /* Check for jumbo payload */ 936 if (ulen == 0) 937 ulen = skb->len; 938 939 if (ulen < sizeof(*uh)) 940 goto short_packet; 941 942 if (ulen < skb->len) { 943 if (pskb_trim_rcsum(skb, ulen)) 944 goto short_packet; 945 saddr = &ipv6_hdr(skb)->saddr; 946 daddr = &ipv6_hdr(skb)->daddr; 947 uh = udp_hdr(skb); 948 } 949 } 950 951 if (udp6_csum_init(skb, uh, proto)) 952 goto csum_error; 953 954 /* Check if the socket is already available, e.g. due to early demux */ 955 sk = skb_steal_sock(skb, &refcounted); 956 if (sk) { 957 struct dst_entry *dst = skb_dst(skb); 958 int ret; 959 960 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 961 udp6_sk_rx_dst_set(sk, dst); 962 963 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 964 if (refcounted) 965 sock_put(sk); 966 goto report_csum_error; 967 } 968 969 ret = udp6_unicast_rcv_skb(sk, skb, uh); 970 if (refcounted) 971 sock_put(sk); 972 return ret; 973 } 974 975 /* 976 * Multicast receive code 977 */ 978 if (ipv6_addr_is_multicast(daddr)) 979 return __udp6_lib_mcast_deliver(net, skb, 980 saddr, daddr, udptable, proto); 981 982 /* Unicast */ 983 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 984 if (sk) { 985 if (!uh->check && !udp_sk(sk)->no_check6_rx) 986 goto report_csum_error; 987 return udp6_unicast_rcv_skb(sk, skb, uh); 988 } 989 990 if (!uh->check) 991 goto report_csum_error; 992 993 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 994 goto discard; 995 996 if (udp_lib_checksum_complete(skb)) 997 goto csum_error; 998 999 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1000 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 1001 1002 kfree_skb(skb); 1003 return 0; 1004 1005 short_packet: 1006 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 1007 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1008 saddr, ntohs(uh->source), 1009 ulen, skb->len, 1010 daddr, ntohs(uh->dest)); 1011 goto discard; 1012 1013 report_csum_error: 1014 udp6_csum_zero_error(skb); 1015 csum_error: 1016 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1017 discard: 1018 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1019 kfree_skb(skb); 1020 return 0; 1021 } 1022 1023 1024 static struct sock *__udp6_lib_demux_lookup(struct net *net, 1025 __be16 loc_port, const struct in6_addr *loc_addr, 1026 __be16 rmt_port, const struct in6_addr *rmt_addr, 1027 int dif, int sdif) 1028 { 1029 unsigned short hnum = ntohs(loc_port); 1030 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1031 unsigned int slot2 = hash2 & udp_table.mask; 1032 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1033 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1034 struct sock *sk; 1035 1036 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1037 if (sk->sk_state == TCP_ESTABLISHED && 1038 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 1039 return sk; 1040 /* Only check first socket in chain */ 1041 break; 1042 } 1043 return NULL; 1044 } 1045 1046 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb) 1047 { 1048 struct net *net = dev_net(skb->dev); 1049 const struct udphdr *uh; 1050 struct sock *sk; 1051 struct dst_entry *dst; 1052 int dif = skb->dev->ifindex; 1053 int sdif = inet6_sdif(skb); 1054 1055 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1056 sizeof(struct udphdr))) 1057 return; 1058 1059 uh = udp_hdr(skb); 1060 1061 if (skb->pkt_type == PACKET_HOST) 1062 sk = __udp6_lib_demux_lookup(net, uh->dest, 1063 &ipv6_hdr(skb)->daddr, 1064 uh->source, &ipv6_hdr(skb)->saddr, 1065 dif, sdif); 1066 else 1067 return; 1068 1069 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1070 return; 1071 1072 skb->sk = sk; 1073 skb->destructor = sock_efree; 1074 dst = rcu_dereference(sk->sk_rx_dst); 1075 1076 if (dst) 1077 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1078 if (dst) { 1079 /* set noref for now. 1080 * any place which wants to hold dst has to call 1081 * dst_hold_safe() 1082 */ 1083 skb_dst_set_noref(skb, dst); 1084 } 1085 } 1086 1087 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1088 { 1089 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1090 } 1091 1092 /* 1093 * Throw away all pending data and cancel the corking. Socket is locked. 1094 */ 1095 static void udp_v6_flush_pending_frames(struct sock *sk) 1096 { 1097 struct udp_sock *up = udp_sk(sk); 1098 1099 if (up->pending == AF_INET) 1100 udp_flush_pending_frames(sk); 1101 else if (up->pending) { 1102 up->len = 0; 1103 up->pending = 0; 1104 ip6_flush_pending_frames(sk); 1105 } 1106 } 1107 1108 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1109 int addr_len) 1110 { 1111 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1112 return -EINVAL; 1113 /* The following checks are replicated from __ip6_datagram_connect() 1114 * and intended to prevent BPF program called below from accessing 1115 * bytes that are out of the bound specified by user in addr_len. 1116 */ 1117 if (uaddr->sa_family == AF_INET) { 1118 if (__ipv6_only_sock(sk)) 1119 return -EAFNOSUPPORT; 1120 return udp_pre_connect(sk, uaddr, addr_len); 1121 } 1122 1123 if (addr_len < SIN6_LEN_RFC2133) 1124 return -EINVAL; 1125 1126 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1127 } 1128 1129 /** 1130 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1131 * @sk: socket we are sending on 1132 * @skb: sk_buff containing the filled-in UDP header 1133 * (checksum field must be zeroed out) 1134 * @saddr: source address 1135 * @daddr: destination address 1136 * @len: length of packet 1137 */ 1138 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1139 const struct in6_addr *saddr, 1140 const struct in6_addr *daddr, int len) 1141 { 1142 unsigned int offset; 1143 struct udphdr *uh = udp_hdr(skb); 1144 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1145 __wsum csum = 0; 1146 1147 if (!frags) { 1148 /* Only one fragment on the socket. */ 1149 skb->csum_start = skb_transport_header(skb) - skb->head; 1150 skb->csum_offset = offsetof(struct udphdr, check); 1151 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1152 } else { 1153 /* 1154 * HW-checksum won't work as there are two or more 1155 * fragments on the socket so that all csums of sk_buffs 1156 * should be together 1157 */ 1158 offset = skb_transport_offset(skb); 1159 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1160 csum = skb->csum; 1161 1162 skb->ip_summed = CHECKSUM_NONE; 1163 1164 do { 1165 csum = csum_add(csum, frags->csum); 1166 } while ((frags = frags->next)); 1167 1168 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1169 csum); 1170 if (uh->check == 0) 1171 uh->check = CSUM_MANGLED_0; 1172 } 1173 } 1174 1175 /* 1176 * Sending 1177 */ 1178 1179 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1180 struct inet_cork *cork) 1181 { 1182 struct sock *sk = skb->sk; 1183 struct udphdr *uh; 1184 int err = 0; 1185 int is_udplite = IS_UDPLITE(sk); 1186 __wsum csum = 0; 1187 int offset = skb_transport_offset(skb); 1188 int len = skb->len - offset; 1189 int datalen = len - sizeof(*uh); 1190 1191 /* 1192 * Create a UDP header 1193 */ 1194 uh = udp_hdr(skb); 1195 uh->source = fl6->fl6_sport; 1196 uh->dest = fl6->fl6_dport; 1197 uh->len = htons(len); 1198 uh->check = 0; 1199 1200 if (cork->gso_size) { 1201 const int hlen = skb_network_header_len(skb) + 1202 sizeof(struct udphdr); 1203 1204 if (hlen + cork->gso_size > cork->fragsize) { 1205 kfree_skb(skb); 1206 return -EINVAL; 1207 } 1208 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1209 kfree_skb(skb); 1210 return -EINVAL; 1211 } 1212 if (udp_sk(sk)->no_check6_tx) { 1213 kfree_skb(skb); 1214 return -EINVAL; 1215 } 1216 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1217 dst_xfrm(skb_dst(skb))) { 1218 kfree_skb(skb); 1219 return -EIO; 1220 } 1221 1222 if (datalen > cork->gso_size) { 1223 skb_shinfo(skb)->gso_size = cork->gso_size; 1224 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1225 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1226 cork->gso_size); 1227 } 1228 goto csum_partial; 1229 } 1230 1231 if (is_udplite) 1232 csum = udplite_csum(skb); 1233 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1234 skb->ip_summed = CHECKSUM_NONE; 1235 goto send; 1236 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1237 csum_partial: 1238 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1239 goto send; 1240 } else 1241 csum = udp_csum(skb); 1242 1243 /* add protocol-dependent pseudo-header */ 1244 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1245 len, fl6->flowi6_proto, csum); 1246 if (uh->check == 0) 1247 uh->check = CSUM_MANGLED_0; 1248 1249 send: 1250 err = ip6_send_skb(skb); 1251 if (err) { 1252 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1253 UDP6_INC_STATS(sock_net(sk), 1254 UDP_MIB_SNDBUFERRORS, is_udplite); 1255 err = 0; 1256 } 1257 } else { 1258 UDP6_INC_STATS(sock_net(sk), 1259 UDP_MIB_OUTDATAGRAMS, is_udplite); 1260 } 1261 return err; 1262 } 1263 1264 static int udp_v6_push_pending_frames(struct sock *sk) 1265 { 1266 struct sk_buff *skb; 1267 struct udp_sock *up = udp_sk(sk); 1268 struct flowi6 fl6; 1269 int err = 0; 1270 1271 if (up->pending == AF_INET) 1272 return udp_push_pending_frames(sk); 1273 1274 /* ip6_finish_skb will release the cork, so make a copy of 1275 * fl6 here. 1276 */ 1277 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1278 1279 skb = ip6_finish_skb(sk); 1280 if (!skb) 1281 goto out; 1282 1283 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1284 1285 out: 1286 up->len = 0; 1287 up->pending = 0; 1288 return err; 1289 } 1290 1291 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1292 { 1293 struct ipv6_txoptions opt_space; 1294 struct udp_sock *up = udp_sk(sk); 1295 struct inet_sock *inet = inet_sk(sk); 1296 struct ipv6_pinfo *np = inet6_sk(sk); 1297 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1298 struct in6_addr *daddr, *final_p, final; 1299 struct ipv6_txoptions *opt = NULL; 1300 struct ipv6_txoptions *opt_to_free = NULL; 1301 struct ip6_flowlabel *flowlabel = NULL; 1302 struct flowi6 fl6; 1303 struct dst_entry *dst; 1304 struct ipcm6_cookie ipc6; 1305 int addr_len = msg->msg_namelen; 1306 bool connected = false; 1307 int ulen = len; 1308 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1309 int err; 1310 int is_udplite = IS_UDPLITE(sk); 1311 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1312 1313 ipcm6_init(&ipc6); 1314 ipc6.gso_size = READ_ONCE(up->gso_size); 1315 ipc6.sockc.tsflags = sk->sk_tsflags; 1316 ipc6.sockc.mark = sk->sk_mark; 1317 1318 /* destination address check */ 1319 if (sin6) { 1320 if (addr_len < offsetof(struct sockaddr, sa_data)) 1321 return -EINVAL; 1322 1323 switch (sin6->sin6_family) { 1324 case AF_INET6: 1325 if (addr_len < SIN6_LEN_RFC2133) 1326 return -EINVAL; 1327 daddr = &sin6->sin6_addr; 1328 if (ipv6_addr_any(daddr) && 1329 ipv6_addr_v4mapped(&np->saddr)) 1330 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1331 daddr); 1332 break; 1333 case AF_INET: 1334 goto do_udp_sendmsg; 1335 case AF_UNSPEC: 1336 msg->msg_name = sin6 = NULL; 1337 msg->msg_namelen = addr_len = 0; 1338 daddr = NULL; 1339 break; 1340 default: 1341 return -EINVAL; 1342 } 1343 } else if (!up->pending) { 1344 if (sk->sk_state != TCP_ESTABLISHED) 1345 return -EDESTADDRREQ; 1346 daddr = &sk->sk_v6_daddr; 1347 } else 1348 daddr = NULL; 1349 1350 if (daddr) { 1351 if (ipv6_addr_v4mapped(daddr)) { 1352 struct sockaddr_in sin; 1353 sin.sin_family = AF_INET; 1354 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1355 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1356 msg->msg_name = &sin; 1357 msg->msg_namelen = sizeof(sin); 1358 do_udp_sendmsg: 1359 if (__ipv6_only_sock(sk)) 1360 return -ENETUNREACH; 1361 return udp_sendmsg(sk, msg, len); 1362 } 1363 } 1364 1365 if (up->pending == AF_INET) 1366 return udp_sendmsg(sk, msg, len); 1367 1368 /* Rough check on arithmetic overflow, 1369 better check is made in ip6_append_data(). 1370 */ 1371 if (len > INT_MAX - sizeof(struct udphdr)) 1372 return -EMSGSIZE; 1373 1374 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1375 if (up->pending) { 1376 /* 1377 * There are pending frames. 1378 * The socket lock must be held while it's corked. 1379 */ 1380 lock_sock(sk); 1381 if (likely(up->pending)) { 1382 if (unlikely(up->pending != AF_INET6)) { 1383 release_sock(sk); 1384 return -EAFNOSUPPORT; 1385 } 1386 dst = NULL; 1387 goto do_append_data; 1388 } 1389 release_sock(sk); 1390 } 1391 ulen += sizeof(struct udphdr); 1392 1393 memset(&fl6, 0, sizeof(fl6)); 1394 1395 if (sin6) { 1396 if (sin6->sin6_port == 0) 1397 return -EINVAL; 1398 1399 fl6.fl6_dport = sin6->sin6_port; 1400 daddr = &sin6->sin6_addr; 1401 1402 if (np->sndflow) { 1403 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1404 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1405 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1406 if (IS_ERR(flowlabel)) 1407 return -EINVAL; 1408 } 1409 } 1410 1411 /* 1412 * Otherwise it will be difficult to maintain 1413 * sk->sk_dst_cache. 1414 */ 1415 if (sk->sk_state == TCP_ESTABLISHED && 1416 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1417 daddr = &sk->sk_v6_daddr; 1418 1419 if (addr_len >= sizeof(struct sockaddr_in6) && 1420 sin6->sin6_scope_id && 1421 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1422 fl6.flowi6_oif = sin6->sin6_scope_id; 1423 } else { 1424 if (sk->sk_state != TCP_ESTABLISHED) 1425 return -EDESTADDRREQ; 1426 1427 fl6.fl6_dport = inet->inet_dport; 1428 daddr = &sk->sk_v6_daddr; 1429 fl6.flowlabel = np->flow_label; 1430 connected = true; 1431 } 1432 1433 if (!fl6.flowi6_oif) 1434 fl6.flowi6_oif = sk->sk_bound_dev_if; 1435 1436 if (!fl6.flowi6_oif) 1437 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1438 1439 fl6.flowi6_uid = sk->sk_uid; 1440 1441 if (msg->msg_controllen) { 1442 opt = &opt_space; 1443 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1444 opt->tot_len = sizeof(*opt); 1445 ipc6.opt = opt; 1446 1447 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1448 if (err > 0) 1449 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1450 &ipc6); 1451 if (err < 0) { 1452 fl6_sock_release(flowlabel); 1453 return err; 1454 } 1455 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1456 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1457 if (IS_ERR(flowlabel)) 1458 return -EINVAL; 1459 } 1460 if (!(opt->opt_nflen|opt->opt_flen)) 1461 opt = NULL; 1462 connected = false; 1463 } 1464 if (!opt) { 1465 opt = txopt_get(np); 1466 opt_to_free = opt; 1467 } 1468 if (flowlabel) 1469 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1470 opt = ipv6_fixup_options(&opt_space, opt); 1471 ipc6.opt = opt; 1472 1473 fl6.flowi6_proto = sk->sk_protocol; 1474 fl6.flowi6_mark = ipc6.sockc.mark; 1475 fl6.daddr = *daddr; 1476 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1477 fl6.saddr = np->saddr; 1478 fl6.fl6_sport = inet->inet_sport; 1479 1480 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1481 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1482 (struct sockaddr *)sin6, &fl6.saddr); 1483 if (err) 1484 goto out_no_dst; 1485 if (sin6) { 1486 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1487 /* BPF program rewrote IPv6-only by IPv4-mapped 1488 * IPv6. It's currently unsupported. 1489 */ 1490 err = -ENOTSUPP; 1491 goto out_no_dst; 1492 } 1493 if (sin6->sin6_port == 0) { 1494 /* BPF program set invalid port. Reject it. */ 1495 err = -EINVAL; 1496 goto out_no_dst; 1497 } 1498 fl6.fl6_dport = sin6->sin6_port; 1499 fl6.daddr = sin6->sin6_addr; 1500 } 1501 } 1502 1503 if (ipv6_addr_any(&fl6.daddr)) 1504 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1505 1506 final_p = fl6_update_dst(&fl6, opt, &final); 1507 if (final_p) 1508 connected = false; 1509 1510 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1511 fl6.flowi6_oif = np->mcast_oif; 1512 connected = false; 1513 } else if (!fl6.flowi6_oif) 1514 fl6.flowi6_oif = np->ucast_oif; 1515 1516 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); 1517 1518 if (ipc6.tclass < 0) 1519 ipc6.tclass = np->tclass; 1520 1521 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1522 1523 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1524 if (IS_ERR(dst)) { 1525 err = PTR_ERR(dst); 1526 dst = NULL; 1527 goto out; 1528 } 1529 1530 if (ipc6.hlimit < 0) 1531 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1532 1533 if (msg->msg_flags&MSG_CONFIRM) 1534 goto do_confirm; 1535 back_from_confirm: 1536 1537 /* Lockless fast path for the non-corking case */ 1538 if (!corkreq) { 1539 struct inet_cork_full cork; 1540 struct sk_buff *skb; 1541 1542 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1543 sizeof(struct udphdr), &ipc6, 1544 &fl6, (struct rt6_info *)dst, 1545 msg->msg_flags, &cork); 1546 err = PTR_ERR(skb); 1547 if (!IS_ERR_OR_NULL(skb)) 1548 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1549 goto out; 1550 } 1551 1552 lock_sock(sk); 1553 if (unlikely(up->pending)) { 1554 /* The socket is already corked while preparing it. */ 1555 /* ... which is an evident application bug. --ANK */ 1556 release_sock(sk); 1557 1558 net_dbg_ratelimited("udp cork app bug 2\n"); 1559 err = -EINVAL; 1560 goto out; 1561 } 1562 1563 up->pending = AF_INET6; 1564 1565 do_append_data: 1566 if (ipc6.dontfrag < 0) 1567 ipc6.dontfrag = np->dontfrag; 1568 up->len += ulen; 1569 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1570 &ipc6, &fl6, (struct rt6_info *)dst, 1571 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1572 if (err) 1573 udp_v6_flush_pending_frames(sk); 1574 else if (!corkreq) 1575 err = udp_v6_push_pending_frames(sk); 1576 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1577 up->pending = 0; 1578 1579 if (err > 0) 1580 err = np->recverr ? net_xmit_errno(err) : 0; 1581 release_sock(sk); 1582 1583 out: 1584 dst_release(dst); 1585 out_no_dst: 1586 fl6_sock_release(flowlabel); 1587 txopt_put(opt_to_free); 1588 if (!err) 1589 return len; 1590 /* 1591 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1592 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1593 * we don't have a good statistic (IpOutDiscards but it can be too many 1594 * things). We could add another new stat but at least for now that 1595 * seems like overkill. 1596 */ 1597 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1598 UDP6_INC_STATS(sock_net(sk), 1599 UDP_MIB_SNDBUFERRORS, is_udplite); 1600 } 1601 return err; 1602 1603 do_confirm: 1604 if (msg->msg_flags & MSG_PROBE) 1605 dst_confirm_neigh(dst, &fl6.daddr); 1606 if (!(msg->msg_flags&MSG_PROBE) || len) 1607 goto back_from_confirm; 1608 err = 0; 1609 goto out; 1610 } 1611 1612 void udpv6_destroy_sock(struct sock *sk) 1613 { 1614 struct udp_sock *up = udp_sk(sk); 1615 lock_sock(sk); 1616 1617 /* protects from races with udp_abort() */ 1618 sock_set_flag(sk, SOCK_DEAD); 1619 udp_v6_flush_pending_frames(sk); 1620 release_sock(sk); 1621 1622 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1623 if (up->encap_type) { 1624 void (*encap_destroy)(struct sock *sk); 1625 encap_destroy = READ_ONCE(up->encap_destroy); 1626 if (encap_destroy) 1627 encap_destroy(sk); 1628 } 1629 if (up->encap_enabled) { 1630 static_branch_dec(&udpv6_encap_needed_key); 1631 udp_encap_disable(); 1632 } 1633 } 1634 1635 inet6_destroy_sock(sk); 1636 } 1637 1638 /* 1639 * Socket option code for UDP 1640 */ 1641 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1642 unsigned int optlen) 1643 { 1644 if (level == SOL_UDP || level == SOL_UDPLITE) 1645 return udp_lib_setsockopt(sk, level, optname, 1646 optval, optlen, 1647 udp_v6_push_pending_frames); 1648 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1649 } 1650 1651 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1652 char __user *optval, int __user *optlen) 1653 { 1654 if (level == SOL_UDP || level == SOL_UDPLITE) 1655 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1656 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1657 } 1658 1659 /* thinking of making this const? Don't. 1660 * early_demux can change based on sysctl. 1661 */ 1662 static struct inet6_protocol udpv6_protocol = { 1663 .early_demux = udp_v6_early_demux, 1664 .early_demux_handler = udp_v6_early_demux, 1665 .handler = udpv6_rcv, 1666 .err_handler = udpv6_err, 1667 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1668 }; 1669 1670 /* ------------------------------------------------------------------------ */ 1671 #ifdef CONFIG_PROC_FS 1672 int udp6_seq_show(struct seq_file *seq, void *v) 1673 { 1674 if (v == SEQ_START_TOKEN) { 1675 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1676 } else { 1677 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1678 struct inet_sock *inet = inet_sk(v); 1679 __u16 srcp = ntohs(inet->inet_sport); 1680 __u16 destp = ntohs(inet->inet_dport); 1681 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1682 udp_rqueue_get(v), bucket); 1683 } 1684 return 0; 1685 } 1686 1687 const struct seq_operations udp6_seq_ops = { 1688 .start = udp_seq_start, 1689 .next = udp_seq_next, 1690 .stop = udp_seq_stop, 1691 .show = udp6_seq_show, 1692 }; 1693 EXPORT_SYMBOL(udp6_seq_ops); 1694 1695 static struct udp_seq_afinfo udp6_seq_afinfo = { 1696 .family = AF_INET6, 1697 .udp_table = &udp_table, 1698 }; 1699 1700 int __net_init udp6_proc_init(struct net *net) 1701 { 1702 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1703 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1704 return -ENOMEM; 1705 return 0; 1706 } 1707 1708 void udp6_proc_exit(struct net *net) 1709 { 1710 remove_proc_entry("udp6", net->proc_net); 1711 } 1712 #endif /* CONFIG_PROC_FS */ 1713 1714 /* ------------------------------------------------------------------------ */ 1715 1716 struct proto udpv6_prot = { 1717 .name = "UDPv6", 1718 .owner = THIS_MODULE, 1719 .close = udp_lib_close, 1720 .pre_connect = udpv6_pre_connect, 1721 .connect = ip6_datagram_connect, 1722 .disconnect = udp_disconnect, 1723 .ioctl = udp_ioctl, 1724 .init = udp_init_sock, 1725 .destroy = udpv6_destroy_sock, 1726 .setsockopt = udpv6_setsockopt, 1727 .getsockopt = udpv6_getsockopt, 1728 .sendmsg = udpv6_sendmsg, 1729 .recvmsg = udpv6_recvmsg, 1730 .release_cb = ip6_datagram_release_cb, 1731 .hash = udp_lib_hash, 1732 .unhash = udp_lib_unhash, 1733 .rehash = udp_v6_rehash, 1734 .get_port = udp_v6_get_port, 1735 .put_port = udp_lib_unhash, 1736 #ifdef CONFIG_BPF_SYSCALL 1737 .psock_update_sk_prot = udp_bpf_update_proto, 1738 #endif 1739 .memory_allocated = &udp_memory_allocated, 1740 .sysctl_mem = sysctl_udp_mem, 1741 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1742 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1743 .obj_size = sizeof(struct udp6_sock), 1744 .h.udp_table = &udp_table, 1745 .diag_destroy = udp_abort, 1746 }; 1747 1748 static struct inet_protosw udpv6_protosw = { 1749 .type = SOCK_DGRAM, 1750 .protocol = IPPROTO_UDP, 1751 .prot = &udpv6_prot, 1752 .ops = &inet6_dgram_ops, 1753 .flags = INET_PROTOSW_PERMANENT, 1754 }; 1755 1756 int __init udpv6_init(void) 1757 { 1758 int ret; 1759 1760 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1761 if (ret) 1762 goto out; 1763 1764 ret = inet6_register_protosw(&udpv6_protosw); 1765 if (ret) 1766 goto out_udpv6_protocol; 1767 out: 1768 return ret; 1769 1770 out_udpv6_protocol: 1771 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1772 goto out; 1773 } 1774 1775 void udpv6_exit(void) 1776 { 1777 inet6_unregister_protosw(&udpv6_protosw); 1778 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1779 } 1780