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/errno.h> 21 #include <linux/types.h> 22 #include <linux/socket.h> 23 #include <linux/sockios.h> 24 #include <linux/net.h> 25 #include <linux/in6.h> 26 #include <linux/netdevice.h> 27 #include <linux/if_arp.h> 28 #include <linux/ipv6.h> 29 #include <linux/icmpv6.h> 30 #include <linux/init.h> 31 #include <linux/module.h> 32 #include <linux/skbuff.h> 33 #include <linux/slab.h> 34 #include <linux/uaccess.h> 35 #include <linux/indirect_call_wrapper.h> 36 37 #include <net/addrconf.h> 38 #include <net/ndisc.h> 39 #include <net/protocol.h> 40 #include <net/transp_v6.h> 41 #include <net/ip6_route.h> 42 #include <net/raw.h> 43 #include <net/tcp_states.h> 44 #include <net/ip6_checksum.h> 45 #include <net/ip6_tunnel.h> 46 #include <net/xfrm.h> 47 #include <net/inet_hashtables.h> 48 #include <net/inet6_hashtables.h> 49 #include <net/busy_poll.h> 50 #include <net/sock_reuseport.h> 51 52 #include <linux/proc_fs.h> 53 #include <linux/seq_file.h> 54 #include <trace/events/skb.h> 55 #include "udp_impl.h" 56 57 static u32 udp6_ehashfn(const struct net *net, 58 const struct in6_addr *laddr, 59 const u16 lport, 60 const struct in6_addr *faddr, 61 const __be16 fport) 62 { 63 static u32 udp6_ehash_secret __read_mostly; 64 static u32 udp_ipv6_hash_secret __read_mostly; 65 66 u32 lhash, fhash; 67 68 net_get_random_once(&udp6_ehash_secret, 69 sizeof(udp6_ehash_secret)); 70 net_get_random_once(&udp_ipv6_hash_secret, 71 sizeof(udp_ipv6_hash_secret)); 72 73 lhash = (__force u32)laddr->s6_addr32[3]; 74 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 75 76 return __inet6_ehashfn(lhash, lport, fhash, fport, 77 udp_ipv6_hash_secret + net_hash_mix(net)); 78 } 79 80 int udp_v6_get_port(struct sock *sk, unsigned short snum) 81 { 82 unsigned int hash2_nulladdr = 83 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 84 unsigned int hash2_partial = 85 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 86 87 /* precompute partial secondary hash */ 88 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 89 return udp_lib_get_port(sk, snum, hash2_nulladdr); 90 } 91 92 void udp_v6_rehash(struct sock *sk) 93 { 94 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 95 &sk->sk_v6_rcv_saddr, 96 inet_sk(sk)->inet_num); 97 98 udp_lib_rehash(sk, new_hash); 99 } 100 101 static int compute_score(struct sock *sk, struct net *net, 102 const struct in6_addr *saddr, __be16 sport, 103 const struct in6_addr *daddr, unsigned short hnum, 104 int dif, int sdif) 105 { 106 int score; 107 struct inet_sock *inet; 108 bool dev_match; 109 110 if (!net_eq(sock_net(sk), net) || 111 udp_sk(sk)->udp_port_hash != hnum || 112 sk->sk_family != PF_INET6) 113 return -1; 114 115 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 116 return -1; 117 118 score = 0; 119 inet = inet_sk(sk); 120 121 if (inet->inet_dport) { 122 if (inet->inet_dport != sport) 123 return -1; 124 score++; 125 } 126 127 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 128 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 129 return -1; 130 score++; 131 } 132 133 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif); 134 if (!dev_match) 135 return -1; 136 score++; 137 138 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 139 score++; 140 141 return score; 142 } 143 144 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 145 struct sk_buff *skb, 146 const struct in6_addr *saddr, 147 __be16 sport, 148 const struct in6_addr *daddr, 149 unsigned int hnum) 150 { 151 struct sock *reuse_sk = NULL; 152 u32 hash; 153 154 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 155 hash = udp6_ehashfn(net, daddr, hnum, saddr, sport); 156 reuse_sk = reuseport_select_sock(sk, hash, skb, 157 sizeof(struct udphdr)); 158 } 159 return reuse_sk; 160 } 161 162 /* called with rcu_read_lock() */ 163 static struct sock *udp6_lib_lookup2(struct net *net, 164 const struct in6_addr *saddr, __be16 sport, 165 const struct in6_addr *daddr, unsigned int hnum, 166 int dif, int sdif, struct udp_hslot *hslot2, 167 struct sk_buff *skb) 168 { 169 struct sock *sk, *result; 170 int score, badness; 171 172 result = NULL; 173 badness = -1; 174 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 175 score = compute_score(sk, net, saddr, sport, 176 daddr, hnum, dif, sdif); 177 if (score > badness) { 178 result = lookup_reuseport(net, sk, skb, 179 saddr, sport, daddr, hnum); 180 /* Fall back to scoring if group has connections */ 181 if (result && !reuseport_has_conns(sk, false)) 182 return result; 183 184 result = result ? : sk; 185 badness = score; 186 } 187 } 188 return result; 189 } 190 191 static inline struct sock *udp6_lookup_run_bpf(struct net *net, 192 struct udp_table *udptable, 193 struct sk_buff *skb, 194 const struct in6_addr *saddr, 195 __be16 sport, 196 const struct in6_addr *daddr, 197 u16 hnum) 198 { 199 struct sock *sk, *reuse_sk; 200 bool no_reuseport; 201 202 if (udptable != &udp_table) 203 return NULL; /* only UDP is supported */ 204 205 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_UDP, 206 saddr, sport, daddr, hnum, &sk); 207 if (no_reuseport || IS_ERR_OR_NULL(sk)) 208 return sk; 209 210 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 211 if (reuse_sk) 212 sk = reuse_sk; 213 return sk; 214 } 215 216 /* rcu_read_lock() must be held */ 217 struct sock *__udp6_lib_lookup(struct net *net, 218 const struct in6_addr *saddr, __be16 sport, 219 const struct in6_addr *daddr, __be16 dport, 220 int dif, int sdif, struct udp_table *udptable, 221 struct sk_buff *skb) 222 { 223 unsigned short hnum = ntohs(dport); 224 unsigned int hash2, slot2; 225 struct udp_hslot *hslot2; 226 struct sock *result, *sk; 227 228 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 229 slot2 = hash2 & udptable->mask; 230 hslot2 = &udptable->hash2[slot2]; 231 232 /* Lookup connected or non-wildcard sockets */ 233 result = udp6_lib_lookup2(net, saddr, sport, 234 daddr, hnum, dif, sdif, 235 hslot2, skb); 236 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 237 goto done; 238 239 /* Lookup redirect from BPF */ 240 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { 241 sk = udp6_lookup_run_bpf(net, udptable, skb, 242 saddr, sport, daddr, hnum); 243 if (sk) { 244 result = sk; 245 goto done; 246 } 247 } 248 249 /* Got non-wildcard socket or error on first lookup */ 250 if (result) 251 goto done; 252 253 /* Lookup wildcard sockets */ 254 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 255 slot2 = hash2 & udptable->mask; 256 hslot2 = &udptable->hash2[slot2]; 257 258 result = udp6_lib_lookup2(net, saddr, sport, 259 &in6addr_any, hnum, dif, sdif, 260 hslot2, skb); 261 done: 262 if (IS_ERR(result)) 263 return NULL; 264 return result; 265 } 266 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 267 268 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 269 __be16 sport, __be16 dport, 270 struct udp_table *udptable) 271 { 272 const struct ipv6hdr *iph = ipv6_hdr(skb); 273 274 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 275 &iph->daddr, dport, inet6_iif(skb), 276 inet6_sdif(skb), udptable, skb); 277 } 278 279 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, 280 __be16 sport, __be16 dport) 281 { 282 const struct ipv6hdr *iph = ipv6_hdr(skb); 283 284 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 285 &iph->daddr, dport, inet6_iif(skb), 286 inet6_sdif(skb), &udp_table, NULL); 287 } 288 289 /* Must be called under rcu_read_lock(). 290 * Does increment socket refcount. 291 */ 292 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 293 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 294 const struct in6_addr *daddr, __be16 dport, int dif) 295 { 296 struct sock *sk; 297 298 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 299 dif, 0, &udp_table, NULL); 300 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 301 sk = NULL; 302 return sk; 303 } 304 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 305 #endif 306 307 /* do not use the scratch area len for jumbogram: their length execeeds the 308 * scratch area space; note that the IP6CB flags is still in the first 309 * cacheline, so checking for jumbograms is cheap 310 */ 311 static int udp6_skb_len(struct sk_buff *skb) 312 { 313 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 314 } 315 316 /* 317 * This should be easy, if there is something there we 318 * return it, otherwise we block. 319 */ 320 321 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 322 int noblock, int flags, int *addr_len) 323 { 324 struct ipv6_pinfo *np = inet6_sk(sk); 325 struct inet_sock *inet = inet_sk(sk); 326 struct sk_buff *skb; 327 unsigned int ulen, copied; 328 int off, err, peeking = flags & MSG_PEEK; 329 int is_udplite = IS_UDPLITE(sk); 330 struct udp_mib __percpu *mib; 331 bool checksum_valid = false; 332 int is_udp4; 333 334 if (flags & MSG_ERRQUEUE) 335 return ipv6_recv_error(sk, msg, len, addr_len); 336 337 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 338 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 339 340 try_again: 341 off = sk_peek_offset(sk, flags); 342 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 343 if (!skb) 344 return err; 345 346 ulen = udp6_skb_len(skb); 347 copied = len; 348 if (copied > ulen - off) 349 copied = ulen - off; 350 else if (copied < ulen) 351 msg->msg_flags |= MSG_TRUNC; 352 353 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 354 mib = __UDPX_MIB(sk, is_udp4); 355 356 /* 357 * If checksum is needed at all, try to do it while copying the 358 * data. If the data is truncated, or if we only want a partial 359 * coverage checksum (UDP-Lite), do it before the copy. 360 */ 361 362 if (copied < ulen || peeking || 363 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 364 checksum_valid = udp_skb_csum_unnecessary(skb) || 365 !__udp_lib_checksum_complete(skb); 366 if (!checksum_valid) 367 goto csum_copy_err; 368 } 369 370 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 371 if (udp_skb_is_linear(skb)) 372 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 373 else 374 err = skb_copy_datagram_msg(skb, off, msg, copied); 375 } else { 376 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 377 if (err == -EINVAL) 378 goto csum_copy_err; 379 } 380 if (unlikely(err)) { 381 if (!peeking) { 382 atomic_inc(&sk->sk_drops); 383 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 384 } 385 kfree_skb(skb); 386 return err; 387 } 388 if (!peeking) 389 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 390 391 sock_recv_ts_and_drops(msg, sk, skb); 392 393 /* Copy the address. */ 394 if (msg->msg_name) { 395 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 396 sin6->sin6_family = AF_INET6; 397 sin6->sin6_port = udp_hdr(skb)->source; 398 sin6->sin6_flowinfo = 0; 399 400 if (is_udp4) { 401 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 402 &sin6->sin6_addr); 403 sin6->sin6_scope_id = 0; 404 } else { 405 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 406 sin6->sin6_scope_id = 407 ipv6_iface_scope_id(&sin6->sin6_addr, 408 inet6_iif(skb)); 409 } 410 *addr_len = sizeof(*sin6); 411 412 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 413 (struct sockaddr *)sin6); 414 } 415 416 if (udp_sk(sk)->gro_enabled) 417 udp_cmsg_recv(msg, sk, skb); 418 419 if (np->rxopt.all) 420 ip6_datagram_recv_common_ctl(sk, msg, skb); 421 422 if (is_udp4) { 423 if (inet->cmsg_flags) 424 ip_cmsg_recv_offset(msg, sk, skb, 425 sizeof(struct udphdr), off); 426 } else { 427 if (np->rxopt.all) 428 ip6_datagram_recv_specific_ctl(sk, msg, skb); 429 } 430 431 err = copied; 432 if (flags & MSG_TRUNC) 433 err = ulen; 434 435 skb_consume_udp(sk, skb, peeking ? -err : err); 436 return err; 437 438 csum_copy_err: 439 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 440 udp_skb_destructor)) { 441 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 442 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 443 } 444 kfree_skb(skb); 445 446 /* starting over for a new packet, but check if we need to yield */ 447 cond_resched(); 448 msg->msg_flags &= ~MSG_TRUNC; 449 goto try_again; 450 } 451 452 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 453 void udpv6_encap_enable(void) 454 { 455 static_branch_inc(&udpv6_encap_needed_key); 456 } 457 EXPORT_SYMBOL(udpv6_encap_enable); 458 459 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 460 * through error handlers in encapsulations looking for a match. 461 */ 462 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 463 struct inet6_skb_parm *opt, 464 u8 type, u8 code, int offset, __be32 info) 465 { 466 int i; 467 468 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 469 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 470 u8 type, u8 code, int offset, __be32 info); 471 const struct ip6_tnl_encap_ops *encap; 472 473 encap = rcu_dereference(ip6tun_encaps[i]); 474 if (!encap) 475 continue; 476 handler = encap->err_handler; 477 if (handler && !handler(skb, opt, type, code, offset, info)) 478 return 0; 479 } 480 481 return -ENOENT; 482 } 483 484 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 485 * reversing source and destination port: this will match tunnels that force the 486 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 487 * lwtunnels might actually break this assumption by being configured with 488 * different destination ports on endpoints, in this case we won't be able to 489 * trace ICMP messages back to them. 490 * 491 * If this doesn't match any socket, probe tunnels with arbitrary destination 492 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 493 * we've sent packets to won't necessarily match the local destination port. 494 * 495 * Then ask the tunnel implementation to match the error against a valid 496 * association. 497 * 498 * Return an error if we can't find a match, the socket if we need further 499 * processing, zero otherwise. 500 */ 501 static struct sock *__udp6_lib_err_encap(struct net *net, 502 const struct ipv6hdr *hdr, int offset, 503 struct udphdr *uh, 504 struct udp_table *udptable, 505 struct sock *sk, 506 struct sk_buff *skb, 507 struct inet6_skb_parm *opt, 508 u8 type, u8 code, __be32 info) 509 { 510 int (*lookup)(struct sock *sk, struct sk_buff *skb); 511 int network_offset, transport_offset; 512 struct udp_sock *up; 513 514 network_offset = skb_network_offset(skb); 515 transport_offset = skb_transport_offset(skb); 516 517 /* Network header needs to point to the outer IPv6 header inside ICMP */ 518 skb_reset_network_header(skb); 519 520 /* Transport header needs to point to the UDP header */ 521 skb_set_transport_header(skb, offset); 522 523 if (sk) { 524 up = udp_sk(sk); 525 526 lookup = READ_ONCE(up->encap_err_lookup); 527 if (lookup && lookup(sk, skb)) 528 sk = NULL; 529 530 goto out; 531 } 532 533 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 534 &hdr->saddr, uh->dest, 535 inet6_iif(skb), 0, udptable, skb); 536 if (sk) { 537 up = udp_sk(sk); 538 539 lookup = READ_ONCE(up->encap_err_lookup); 540 if (!lookup || lookup(sk, skb)) 541 sk = NULL; 542 } 543 544 out: 545 if (!sk) { 546 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 547 offset, info)); 548 } 549 550 skb_set_transport_header(skb, transport_offset); 551 skb_set_network_header(skb, network_offset); 552 553 return sk; 554 } 555 556 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 557 u8 type, u8 code, int offset, __be32 info, 558 struct udp_table *udptable) 559 { 560 struct ipv6_pinfo *np; 561 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 562 const struct in6_addr *saddr = &hdr->saddr; 563 const struct in6_addr *daddr = &hdr->daddr; 564 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 565 bool tunnel = false; 566 struct sock *sk; 567 int harderr; 568 int err; 569 struct net *net = dev_net(skb->dev); 570 571 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 572 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 573 574 if (!sk || udp_sk(sk)->encap_type) { 575 /* No socket for error: try tunnels before discarding */ 576 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 577 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 578 udptable, sk, skb, 579 opt, type, code, info); 580 if (!sk) 581 return 0; 582 } else 583 sk = ERR_PTR(-ENOENT); 584 585 if (IS_ERR(sk)) { 586 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 587 ICMP6_MIB_INERRORS); 588 return PTR_ERR(sk); 589 } 590 591 tunnel = true; 592 } 593 594 harderr = icmpv6_err_convert(type, code, &err); 595 np = inet6_sk(sk); 596 597 if (type == ICMPV6_PKT_TOOBIG) { 598 if (!ip6_sk_accept_pmtu(sk)) 599 goto out; 600 ip6_sk_update_pmtu(skb, sk, info); 601 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 602 harderr = 1; 603 } 604 if (type == NDISC_REDIRECT) { 605 if (tunnel) { 606 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 607 sk->sk_mark, sk->sk_uid); 608 } else { 609 ip6_sk_redirect(skb, sk); 610 } 611 goto out; 612 } 613 614 /* Tunnels don't have an application socket: don't pass errors back */ 615 if (tunnel) 616 goto out; 617 618 if (!np->recverr) { 619 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 620 goto out; 621 } else { 622 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 623 } 624 625 sk->sk_err = err; 626 sk_error_report(sk); 627 out: 628 return 0; 629 } 630 631 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 632 { 633 int rc; 634 635 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 636 sock_rps_save_rxhash(sk, skb); 637 sk_mark_napi_id(sk, skb); 638 sk_incoming_cpu_update(sk); 639 } else { 640 sk_mark_napi_id_once(sk, skb); 641 } 642 643 rc = __udp_enqueue_schedule_skb(sk, skb); 644 if (rc < 0) { 645 int is_udplite = IS_UDPLITE(sk); 646 647 /* Note that an ENOMEM error is charged twice */ 648 if (rc == -ENOMEM) 649 UDP6_INC_STATS(sock_net(sk), 650 UDP_MIB_RCVBUFERRORS, is_udplite); 651 else 652 UDP6_INC_STATS(sock_net(sk), 653 UDP_MIB_MEMERRORS, is_udplite); 654 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 655 kfree_skb(skb); 656 return -1; 657 } 658 659 return 0; 660 } 661 662 static __inline__ int udpv6_err(struct sk_buff *skb, 663 struct inet6_skb_parm *opt, u8 type, 664 u8 code, int offset, __be32 info) 665 { 666 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 667 } 668 669 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 670 { 671 struct udp_sock *up = udp_sk(sk); 672 int is_udplite = IS_UDPLITE(sk); 673 674 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 675 goto drop; 676 677 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 678 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 679 680 /* 681 * This is an encapsulation socket so pass the skb to 682 * the socket's udp_encap_rcv() hook. Otherwise, just 683 * fall through and pass this up the UDP socket. 684 * up->encap_rcv() returns the following value: 685 * =0 if skb was successfully passed to the encap 686 * handler or was discarded by it. 687 * >0 if skb should be passed on to UDP. 688 * <0 if skb should be resubmitted as proto -N 689 */ 690 691 /* if we're overly short, let UDP handle it */ 692 encap_rcv = READ_ONCE(up->encap_rcv); 693 if (encap_rcv) { 694 int ret; 695 696 /* Verify checksum before giving to encap */ 697 if (udp_lib_checksum_complete(skb)) 698 goto csum_error; 699 700 ret = encap_rcv(sk, skb); 701 if (ret <= 0) { 702 __UDP_INC_STATS(sock_net(sk), 703 UDP_MIB_INDATAGRAMS, 704 is_udplite); 705 return -ret; 706 } 707 } 708 709 /* FALLTHROUGH -- it's a UDP Packet */ 710 } 711 712 /* 713 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 714 */ 715 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 716 717 if (up->pcrlen == 0) { /* full coverage was set */ 718 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 719 UDP_SKB_CB(skb)->cscov, skb->len); 720 goto drop; 721 } 722 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 723 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 724 UDP_SKB_CB(skb)->cscov, up->pcrlen); 725 goto drop; 726 } 727 } 728 729 prefetch(&sk->sk_rmem_alloc); 730 if (rcu_access_pointer(sk->sk_filter) && 731 udp_lib_checksum_complete(skb)) 732 goto csum_error; 733 734 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 735 goto drop; 736 737 udp_csum_pull_header(skb); 738 739 skb_dst_drop(skb); 740 741 return __udpv6_queue_rcv_skb(sk, skb); 742 743 csum_error: 744 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 745 drop: 746 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 747 atomic_inc(&sk->sk_drops); 748 kfree_skb(skb); 749 return -1; 750 } 751 752 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 753 { 754 struct sk_buff *next, *segs; 755 int ret; 756 757 if (likely(!udp_unexpected_gso(sk, skb))) 758 return udpv6_queue_rcv_one_skb(sk, skb); 759 760 __skb_push(skb, -skb_mac_offset(skb)); 761 segs = udp_rcv_segment(sk, skb, false); 762 skb_list_walk_safe(segs, skb, next) { 763 __skb_pull(skb, skb_transport_offset(skb)); 764 765 udp_post_segment_fix_csum(skb); 766 ret = udpv6_queue_rcv_one_skb(sk, skb); 767 if (ret > 0) 768 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 769 true); 770 } 771 return 0; 772 } 773 774 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 775 __be16 loc_port, const struct in6_addr *loc_addr, 776 __be16 rmt_port, const struct in6_addr *rmt_addr, 777 int dif, int sdif, unsigned short hnum) 778 { 779 struct inet_sock *inet = inet_sk(sk); 780 781 if (!net_eq(sock_net(sk), net)) 782 return false; 783 784 if (udp_sk(sk)->udp_port_hash != hnum || 785 sk->sk_family != PF_INET6 || 786 (inet->inet_dport && inet->inet_dport != rmt_port) || 787 (!ipv6_addr_any(&sk->sk_v6_daddr) && 788 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 789 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) || 790 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 791 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 792 return false; 793 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 794 return false; 795 return true; 796 } 797 798 static void udp6_csum_zero_error(struct sk_buff *skb) 799 { 800 /* RFC 2460 section 8.1 says that we SHOULD log 801 * this error. Well, it is reasonable. 802 */ 803 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 804 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 805 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 806 } 807 808 /* 809 * Note: called only from the BH handler context, 810 * so we don't need to lock the hashes. 811 */ 812 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 813 const struct in6_addr *saddr, const struct in6_addr *daddr, 814 struct udp_table *udptable, int proto) 815 { 816 struct sock *sk, *first = NULL; 817 const struct udphdr *uh = udp_hdr(skb); 818 unsigned short hnum = ntohs(uh->dest); 819 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 820 unsigned int offset = offsetof(typeof(*sk), sk_node); 821 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 822 int dif = inet6_iif(skb); 823 int sdif = inet6_sdif(skb); 824 struct hlist_node *node; 825 struct sk_buff *nskb; 826 827 if (use_hash2) { 828 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 829 udptable->mask; 830 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 831 start_lookup: 832 hslot = &udptable->hash2[hash2]; 833 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 834 } 835 836 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 837 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 838 uh->source, saddr, dif, sdif, 839 hnum)) 840 continue; 841 /* If zero checksum and no_check is not on for 842 * the socket then skip it. 843 */ 844 if (!uh->check && !udp_sk(sk)->no_check6_rx) 845 continue; 846 if (!first) { 847 first = sk; 848 continue; 849 } 850 nskb = skb_clone(skb, GFP_ATOMIC); 851 if (unlikely(!nskb)) { 852 atomic_inc(&sk->sk_drops); 853 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 854 IS_UDPLITE(sk)); 855 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 856 IS_UDPLITE(sk)); 857 continue; 858 } 859 860 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 861 consume_skb(nskb); 862 } 863 864 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 865 if (use_hash2 && hash2 != hash2_any) { 866 hash2 = hash2_any; 867 goto start_lookup; 868 } 869 870 if (first) { 871 if (udpv6_queue_rcv_skb(first, skb) > 0) 872 consume_skb(skb); 873 } else { 874 kfree_skb(skb); 875 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 876 proto == IPPROTO_UDPLITE); 877 } 878 return 0; 879 } 880 881 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 882 { 883 if (udp_sk_rx_dst_set(sk, dst)) { 884 const struct rt6_info *rt = (const struct rt6_info *)dst; 885 886 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 887 } 888 } 889 890 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 891 * return code conversion for ip layer consumption 892 */ 893 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 894 struct udphdr *uh) 895 { 896 int ret; 897 898 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 899 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 900 901 ret = udpv6_queue_rcv_skb(sk, skb); 902 903 /* a return value > 0 means to resubmit the input */ 904 if (ret > 0) 905 return ret; 906 return 0; 907 } 908 909 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 910 int proto) 911 { 912 const struct in6_addr *saddr, *daddr; 913 struct net *net = dev_net(skb->dev); 914 struct udphdr *uh; 915 struct sock *sk; 916 bool refcounted; 917 u32 ulen = 0; 918 919 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 920 goto discard; 921 922 saddr = &ipv6_hdr(skb)->saddr; 923 daddr = &ipv6_hdr(skb)->daddr; 924 uh = udp_hdr(skb); 925 926 ulen = ntohs(uh->len); 927 if (ulen > skb->len) 928 goto short_packet; 929 930 if (proto == IPPROTO_UDP) { 931 /* UDP validates ulen. */ 932 933 /* Check for jumbo payload */ 934 if (ulen == 0) 935 ulen = skb->len; 936 937 if (ulen < sizeof(*uh)) 938 goto short_packet; 939 940 if (ulen < skb->len) { 941 if (pskb_trim_rcsum(skb, ulen)) 942 goto short_packet; 943 saddr = &ipv6_hdr(skb)->saddr; 944 daddr = &ipv6_hdr(skb)->daddr; 945 uh = udp_hdr(skb); 946 } 947 } 948 949 if (udp6_csum_init(skb, uh, proto)) 950 goto csum_error; 951 952 /* Check if the socket is already available, e.g. due to early demux */ 953 sk = skb_steal_sock(skb, &refcounted); 954 if (sk) { 955 struct dst_entry *dst = skb_dst(skb); 956 int ret; 957 958 if (unlikely(sk->sk_rx_dst != dst)) 959 udp6_sk_rx_dst_set(sk, dst); 960 961 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 962 if (refcounted) 963 sock_put(sk); 964 goto report_csum_error; 965 } 966 967 ret = udp6_unicast_rcv_skb(sk, skb, uh); 968 if (refcounted) 969 sock_put(sk); 970 return ret; 971 } 972 973 /* 974 * Multicast receive code 975 */ 976 if (ipv6_addr_is_multicast(daddr)) 977 return __udp6_lib_mcast_deliver(net, skb, 978 saddr, daddr, udptable, proto); 979 980 /* Unicast */ 981 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 982 if (sk) { 983 if (!uh->check && !udp_sk(sk)->no_check6_rx) 984 goto report_csum_error; 985 return udp6_unicast_rcv_skb(sk, skb, uh); 986 } 987 988 if (!uh->check) 989 goto report_csum_error; 990 991 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 992 goto discard; 993 994 if (udp_lib_checksum_complete(skb)) 995 goto csum_error; 996 997 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 998 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 999 1000 kfree_skb(skb); 1001 return 0; 1002 1003 short_packet: 1004 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 1005 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1006 saddr, ntohs(uh->source), 1007 ulen, skb->len, 1008 daddr, ntohs(uh->dest)); 1009 goto discard; 1010 1011 report_csum_error: 1012 udp6_csum_zero_error(skb); 1013 csum_error: 1014 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1015 discard: 1016 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1017 kfree_skb(skb); 1018 return 0; 1019 } 1020 1021 1022 static struct sock *__udp6_lib_demux_lookup(struct net *net, 1023 __be16 loc_port, const struct in6_addr *loc_addr, 1024 __be16 rmt_port, const struct in6_addr *rmt_addr, 1025 int dif, int sdif) 1026 { 1027 unsigned short hnum = ntohs(loc_port); 1028 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1029 unsigned int slot2 = hash2 & udp_table.mask; 1030 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1031 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1032 struct sock *sk; 1033 1034 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1035 if (sk->sk_state == TCP_ESTABLISHED && 1036 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 1037 return sk; 1038 /* Only check first socket in chain */ 1039 break; 1040 } 1041 return NULL; 1042 } 1043 1044 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb) 1045 { 1046 struct net *net = dev_net(skb->dev); 1047 const struct udphdr *uh; 1048 struct sock *sk; 1049 struct dst_entry *dst; 1050 int dif = skb->dev->ifindex; 1051 int sdif = inet6_sdif(skb); 1052 1053 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1054 sizeof(struct udphdr))) 1055 return; 1056 1057 uh = udp_hdr(skb); 1058 1059 if (skb->pkt_type == PACKET_HOST) 1060 sk = __udp6_lib_demux_lookup(net, uh->dest, 1061 &ipv6_hdr(skb)->daddr, 1062 uh->source, &ipv6_hdr(skb)->saddr, 1063 dif, sdif); 1064 else 1065 return; 1066 1067 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1068 return; 1069 1070 skb->sk = sk; 1071 skb->destructor = sock_efree; 1072 dst = READ_ONCE(sk->sk_rx_dst); 1073 1074 if (dst) 1075 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 1076 if (dst) { 1077 /* set noref for now. 1078 * any place which wants to hold dst has to call 1079 * dst_hold_safe() 1080 */ 1081 skb_dst_set_noref(skb, dst); 1082 } 1083 } 1084 1085 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1086 { 1087 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1088 } 1089 1090 /* 1091 * Throw away all pending data and cancel the corking. Socket is locked. 1092 */ 1093 static void udp_v6_flush_pending_frames(struct sock *sk) 1094 { 1095 struct udp_sock *up = udp_sk(sk); 1096 1097 if (up->pending == AF_INET) 1098 udp_flush_pending_frames(sk); 1099 else if (up->pending) { 1100 up->len = 0; 1101 up->pending = 0; 1102 ip6_flush_pending_frames(sk); 1103 } 1104 } 1105 1106 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1107 int addr_len) 1108 { 1109 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1110 return -EINVAL; 1111 /* The following checks are replicated from __ip6_datagram_connect() 1112 * and intended to prevent BPF program called below from accessing 1113 * bytes that are out of the bound specified by user in addr_len. 1114 */ 1115 if (uaddr->sa_family == AF_INET) { 1116 if (__ipv6_only_sock(sk)) 1117 return -EAFNOSUPPORT; 1118 return udp_pre_connect(sk, uaddr, addr_len); 1119 } 1120 1121 if (addr_len < SIN6_LEN_RFC2133) 1122 return -EINVAL; 1123 1124 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1125 } 1126 1127 /** 1128 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1129 * @sk: socket we are sending on 1130 * @skb: sk_buff containing the filled-in UDP header 1131 * (checksum field must be zeroed out) 1132 * @saddr: source address 1133 * @daddr: destination address 1134 * @len: length of packet 1135 */ 1136 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1137 const struct in6_addr *saddr, 1138 const struct in6_addr *daddr, int len) 1139 { 1140 unsigned int offset; 1141 struct udphdr *uh = udp_hdr(skb); 1142 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1143 __wsum csum = 0; 1144 1145 if (!frags) { 1146 /* Only one fragment on the socket. */ 1147 skb->csum_start = skb_transport_header(skb) - skb->head; 1148 skb->csum_offset = offsetof(struct udphdr, check); 1149 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1150 } else { 1151 /* 1152 * HW-checksum won't work as there are two or more 1153 * fragments on the socket so that all csums of sk_buffs 1154 * should be together 1155 */ 1156 offset = skb_transport_offset(skb); 1157 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1158 csum = skb->csum; 1159 1160 skb->ip_summed = CHECKSUM_NONE; 1161 1162 do { 1163 csum = csum_add(csum, frags->csum); 1164 } while ((frags = frags->next)); 1165 1166 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1167 csum); 1168 if (uh->check == 0) 1169 uh->check = CSUM_MANGLED_0; 1170 } 1171 } 1172 1173 /* 1174 * Sending 1175 */ 1176 1177 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1178 struct inet_cork *cork) 1179 { 1180 struct sock *sk = skb->sk; 1181 struct udphdr *uh; 1182 int err = 0; 1183 int is_udplite = IS_UDPLITE(sk); 1184 __wsum csum = 0; 1185 int offset = skb_transport_offset(skb); 1186 int len = skb->len - offset; 1187 int datalen = len - sizeof(*uh); 1188 1189 /* 1190 * Create a UDP header 1191 */ 1192 uh = udp_hdr(skb); 1193 uh->source = fl6->fl6_sport; 1194 uh->dest = fl6->fl6_dport; 1195 uh->len = htons(len); 1196 uh->check = 0; 1197 1198 if (cork->gso_size) { 1199 const int hlen = skb_network_header_len(skb) + 1200 sizeof(struct udphdr); 1201 1202 if (hlen + cork->gso_size > cork->fragsize) { 1203 kfree_skb(skb); 1204 return -EINVAL; 1205 } 1206 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 1207 kfree_skb(skb); 1208 return -EINVAL; 1209 } 1210 if (udp_sk(sk)->no_check6_tx) { 1211 kfree_skb(skb); 1212 return -EINVAL; 1213 } 1214 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1215 dst_xfrm(skb_dst(skb))) { 1216 kfree_skb(skb); 1217 return -EIO; 1218 } 1219 1220 if (datalen > cork->gso_size) { 1221 skb_shinfo(skb)->gso_size = cork->gso_size; 1222 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1223 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1224 cork->gso_size); 1225 } 1226 goto csum_partial; 1227 } 1228 1229 if (is_udplite) 1230 csum = udplite_csum(skb); 1231 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1232 skb->ip_summed = CHECKSUM_NONE; 1233 goto send; 1234 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1235 csum_partial: 1236 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1237 goto send; 1238 } else 1239 csum = udp_csum(skb); 1240 1241 /* add protocol-dependent pseudo-header */ 1242 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1243 len, fl6->flowi6_proto, csum); 1244 if (uh->check == 0) 1245 uh->check = CSUM_MANGLED_0; 1246 1247 send: 1248 err = ip6_send_skb(skb); 1249 if (err) { 1250 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1251 UDP6_INC_STATS(sock_net(sk), 1252 UDP_MIB_SNDBUFERRORS, is_udplite); 1253 err = 0; 1254 } 1255 } else { 1256 UDP6_INC_STATS(sock_net(sk), 1257 UDP_MIB_OUTDATAGRAMS, is_udplite); 1258 } 1259 return err; 1260 } 1261 1262 static int udp_v6_push_pending_frames(struct sock *sk) 1263 { 1264 struct sk_buff *skb; 1265 struct udp_sock *up = udp_sk(sk); 1266 struct flowi6 fl6; 1267 int err = 0; 1268 1269 if (up->pending == AF_INET) 1270 return udp_push_pending_frames(sk); 1271 1272 /* ip6_finish_skb will release the cork, so make a copy of 1273 * fl6 here. 1274 */ 1275 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1276 1277 skb = ip6_finish_skb(sk); 1278 if (!skb) 1279 goto out; 1280 1281 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1282 1283 out: 1284 up->len = 0; 1285 up->pending = 0; 1286 return err; 1287 } 1288 1289 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1290 { 1291 struct ipv6_txoptions opt_space; 1292 struct udp_sock *up = udp_sk(sk); 1293 struct inet_sock *inet = inet_sk(sk); 1294 struct ipv6_pinfo *np = inet6_sk(sk); 1295 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1296 struct in6_addr *daddr, *final_p, final; 1297 struct ipv6_txoptions *opt = NULL; 1298 struct ipv6_txoptions *opt_to_free = NULL; 1299 struct ip6_flowlabel *flowlabel = NULL; 1300 struct flowi6 fl6; 1301 struct dst_entry *dst; 1302 struct ipcm6_cookie ipc6; 1303 int addr_len = msg->msg_namelen; 1304 bool connected = false; 1305 int ulen = len; 1306 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1307 int err; 1308 int is_udplite = IS_UDPLITE(sk); 1309 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1310 1311 ipcm6_init(&ipc6); 1312 ipc6.gso_size = READ_ONCE(up->gso_size); 1313 ipc6.sockc.tsflags = sk->sk_tsflags; 1314 ipc6.sockc.mark = sk->sk_mark; 1315 1316 /* destination address check */ 1317 if (sin6) { 1318 if (addr_len < offsetof(struct sockaddr, sa_data)) 1319 return -EINVAL; 1320 1321 switch (sin6->sin6_family) { 1322 case AF_INET6: 1323 if (addr_len < SIN6_LEN_RFC2133) 1324 return -EINVAL; 1325 daddr = &sin6->sin6_addr; 1326 if (ipv6_addr_any(daddr) && 1327 ipv6_addr_v4mapped(&np->saddr)) 1328 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1329 daddr); 1330 break; 1331 case AF_INET: 1332 goto do_udp_sendmsg; 1333 case AF_UNSPEC: 1334 msg->msg_name = sin6 = NULL; 1335 msg->msg_namelen = addr_len = 0; 1336 daddr = NULL; 1337 break; 1338 default: 1339 return -EINVAL; 1340 } 1341 } else if (!up->pending) { 1342 if (sk->sk_state != TCP_ESTABLISHED) 1343 return -EDESTADDRREQ; 1344 daddr = &sk->sk_v6_daddr; 1345 } else 1346 daddr = NULL; 1347 1348 if (daddr) { 1349 if (ipv6_addr_v4mapped(daddr)) { 1350 struct sockaddr_in sin; 1351 sin.sin_family = AF_INET; 1352 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1353 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1354 msg->msg_name = &sin; 1355 msg->msg_namelen = sizeof(sin); 1356 do_udp_sendmsg: 1357 if (__ipv6_only_sock(sk)) 1358 return -ENETUNREACH; 1359 return udp_sendmsg(sk, msg, len); 1360 } 1361 } 1362 1363 if (up->pending == AF_INET) 1364 return udp_sendmsg(sk, msg, len); 1365 1366 /* Rough check on arithmetic overflow, 1367 better check is made in ip6_append_data(). 1368 */ 1369 if (len > INT_MAX - sizeof(struct udphdr)) 1370 return -EMSGSIZE; 1371 1372 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1373 if (up->pending) { 1374 /* 1375 * There are pending frames. 1376 * The socket lock must be held while it's corked. 1377 */ 1378 lock_sock(sk); 1379 if (likely(up->pending)) { 1380 if (unlikely(up->pending != AF_INET6)) { 1381 release_sock(sk); 1382 return -EAFNOSUPPORT; 1383 } 1384 dst = NULL; 1385 goto do_append_data; 1386 } 1387 release_sock(sk); 1388 } 1389 ulen += sizeof(struct udphdr); 1390 1391 memset(&fl6, 0, sizeof(fl6)); 1392 1393 if (sin6) { 1394 if (sin6->sin6_port == 0) 1395 return -EINVAL; 1396 1397 fl6.fl6_dport = sin6->sin6_port; 1398 daddr = &sin6->sin6_addr; 1399 1400 if (np->sndflow) { 1401 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1402 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1403 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1404 if (IS_ERR(flowlabel)) 1405 return -EINVAL; 1406 } 1407 } 1408 1409 /* 1410 * Otherwise it will be difficult to maintain 1411 * sk->sk_dst_cache. 1412 */ 1413 if (sk->sk_state == TCP_ESTABLISHED && 1414 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1415 daddr = &sk->sk_v6_daddr; 1416 1417 if (addr_len >= sizeof(struct sockaddr_in6) && 1418 sin6->sin6_scope_id && 1419 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1420 fl6.flowi6_oif = sin6->sin6_scope_id; 1421 } else { 1422 if (sk->sk_state != TCP_ESTABLISHED) 1423 return -EDESTADDRREQ; 1424 1425 fl6.fl6_dport = inet->inet_dport; 1426 daddr = &sk->sk_v6_daddr; 1427 fl6.flowlabel = np->flow_label; 1428 connected = true; 1429 } 1430 1431 if (!fl6.flowi6_oif) 1432 fl6.flowi6_oif = sk->sk_bound_dev_if; 1433 1434 if (!fl6.flowi6_oif) 1435 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1436 1437 fl6.flowi6_mark = ipc6.sockc.mark; 1438 fl6.flowi6_uid = sk->sk_uid; 1439 1440 if (msg->msg_controllen) { 1441 opt = &opt_space; 1442 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1443 opt->tot_len = sizeof(*opt); 1444 ipc6.opt = opt; 1445 1446 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1447 if (err > 0) 1448 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1449 &ipc6); 1450 if (err < 0) { 1451 fl6_sock_release(flowlabel); 1452 return err; 1453 } 1454 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1455 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1456 if (IS_ERR(flowlabel)) 1457 return -EINVAL; 1458 } 1459 if (!(opt->opt_nflen|opt->opt_flen)) 1460 opt = NULL; 1461 connected = false; 1462 } 1463 if (!opt) { 1464 opt = txopt_get(np); 1465 opt_to_free = opt; 1466 } 1467 if (flowlabel) 1468 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1469 opt = ipv6_fixup_options(&opt_space, opt); 1470 ipc6.opt = opt; 1471 1472 fl6.flowi6_proto = sk->sk_protocol; 1473 fl6.daddr = *daddr; 1474 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1475 fl6.saddr = np->saddr; 1476 fl6.fl6_sport = inet->inet_sport; 1477 1478 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1479 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1480 (struct sockaddr *)sin6, &fl6.saddr); 1481 if (err) 1482 goto out_no_dst; 1483 if (sin6) { 1484 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1485 /* BPF program rewrote IPv6-only by IPv4-mapped 1486 * IPv6. It's currently unsupported. 1487 */ 1488 err = -ENOTSUPP; 1489 goto out_no_dst; 1490 } 1491 if (sin6->sin6_port == 0) { 1492 /* BPF program set invalid port. Reject it. */ 1493 err = -EINVAL; 1494 goto out_no_dst; 1495 } 1496 fl6.fl6_dport = sin6->sin6_port; 1497 fl6.daddr = sin6->sin6_addr; 1498 } 1499 } 1500 1501 if (ipv6_addr_any(&fl6.daddr)) 1502 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1503 1504 final_p = fl6_update_dst(&fl6, opt, &final); 1505 if (final_p) 1506 connected = false; 1507 1508 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1509 fl6.flowi6_oif = np->mcast_oif; 1510 connected = false; 1511 } else if (!fl6.flowi6_oif) 1512 fl6.flowi6_oif = np->ucast_oif; 1513 1514 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); 1515 1516 if (ipc6.tclass < 0) 1517 ipc6.tclass = np->tclass; 1518 1519 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1520 1521 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1522 if (IS_ERR(dst)) { 1523 err = PTR_ERR(dst); 1524 dst = NULL; 1525 goto out; 1526 } 1527 1528 if (ipc6.hlimit < 0) 1529 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1530 1531 if (msg->msg_flags&MSG_CONFIRM) 1532 goto do_confirm; 1533 back_from_confirm: 1534 1535 /* Lockless fast path for the non-corking case */ 1536 if (!corkreq) { 1537 struct inet_cork_full cork; 1538 struct sk_buff *skb; 1539 1540 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1541 sizeof(struct udphdr), &ipc6, 1542 &fl6, (struct rt6_info *)dst, 1543 msg->msg_flags, &cork); 1544 err = PTR_ERR(skb); 1545 if (!IS_ERR_OR_NULL(skb)) 1546 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1547 goto out; 1548 } 1549 1550 lock_sock(sk); 1551 if (unlikely(up->pending)) { 1552 /* The socket is already corked while preparing it. */ 1553 /* ... which is an evident application bug. --ANK */ 1554 release_sock(sk); 1555 1556 net_dbg_ratelimited("udp cork app bug 2\n"); 1557 err = -EINVAL; 1558 goto out; 1559 } 1560 1561 up->pending = AF_INET6; 1562 1563 do_append_data: 1564 if (ipc6.dontfrag < 0) 1565 ipc6.dontfrag = np->dontfrag; 1566 up->len += ulen; 1567 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1568 &ipc6, &fl6, (struct rt6_info *)dst, 1569 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1570 if (err) 1571 udp_v6_flush_pending_frames(sk); 1572 else if (!corkreq) 1573 err = udp_v6_push_pending_frames(sk); 1574 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1575 up->pending = 0; 1576 1577 if (err > 0) 1578 err = np->recverr ? net_xmit_errno(err) : 0; 1579 release_sock(sk); 1580 1581 out: 1582 dst_release(dst); 1583 out_no_dst: 1584 fl6_sock_release(flowlabel); 1585 txopt_put(opt_to_free); 1586 if (!err) 1587 return len; 1588 /* 1589 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1590 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1591 * we don't have a good statistic (IpOutDiscards but it can be too many 1592 * things). We could add another new stat but at least for now that 1593 * seems like overkill. 1594 */ 1595 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1596 UDP6_INC_STATS(sock_net(sk), 1597 UDP_MIB_SNDBUFERRORS, is_udplite); 1598 } 1599 return err; 1600 1601 do_confirm: 1602 if (msg->msg_flags & MSG_PROBE) 1603 dst_confirm_neigh(dst, &fl6.daddr); 1604 if (!(msg->msg_flags&MSG_PROBE) || len) 1605 goto back_from_confirm; 1606 err = 0; 1607 goto out; 1608 } 1609 1610 void udpv6_destroy_sock(struct sock *sk) 1611 { 1612 struct udp_sock *up = udp_sk(sk); 1613 lock_sock(sk); 1614 1615 /* protects from races with udp_abort() */ 1616 sock_set_flag(sk, SOCK_DEAD); 1617 udp_v6_flush_pending_frames(sk); 1618 release_sock(sk); 1619 1620 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1621 if (up->encap_type) { 1622 void (*encap_destroy)(struct sock *sk); 1623 encap_destroy = READ_ONCE(up->encap_destroy); 1624 if (encap_destroy) 1625 encap_destroy(sk); 1626 } 1627 if (up->encap_enabled) { 1628 static_branch_dec(&udpv6_encap_needed_key); 1629 udp_encap_disable(); 1630 } 1631 } 1632 1633 inet6_destroy_sock(sk); 1634 } 1635 1636 /* 1637 * Socket option code for UDP 1638 */ 1639 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1640 unsigned int optlen) 1641 { 1642 if (level == SOL_UDP || level == SOL_UDPLITE) 1643 return udp_lib_setsockopt(sk, level, optname, 1644 optval, optlen, 1645 udp_v6_push_pending_frames); 1646 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1647 } 1648 1649 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1650 char __user *optval, int __user *optlen) 1651 { 1652 if (level == SOL_UDP || level == SOL_UDPLITE) 1653 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1654 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1655 } 1656 1657 /* thinking of making this const? Don't. 1658 * early_demux can change based on sysctl. 1659 */ 1660 static struct inet6_protocol udpv6_protocol = { 1661 .early_demux = udp_v6_early_demux, 1662 .early_demux_handler = udp_v6_early_demux, 1663 .handler = udpv6_rcv, 1664 .err_handler = udpv6_err, 1665 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1666 }; 1667 1668 /* ------------------------------------------------------------------------ */ 1669 #ifdef CONFIG_PROC_FS 1670 int udp6_seq_show(struct seq_file *seq, void *v) 1671 { 1672 if (v == SEQ_START_TOKEN) { 1673 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1674 } else { 1675 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1676 struct inet_sock *inet = inet_sk(v); 1677 __u16 srcp = ntohs(inet->inet_sport); 1678 __u16 destp = ntohs(inet->inet_dport); 1679 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1680 udp_rqueue_get(v), bucket); 1681 } 1682 return 0; 1683 } 1684 1685 const struct seq_operations udp6_seq_ops = { 1686 .start = udp_seq_start, 1687 .next = udp_seq_next, 1688 .stop = udp_seq_stop, 1689 .show = udp6_seq_show, 1690 }; 1691 EXPORT_SYMBOL(udp6_seq_ops); 1692 1693 static struct udp_seq_afinfo udp6_seq_afinfo = { 1694 .family = AF_INET6, 1695 .udp_table = &udp_table, 1696 }; 1697 1698 int __net_init udp6_proc_init(struct net *net) 1699 { 1700 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1701 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1702 return -ENOMEM; 1703 return 0; 1704 } 1705 1706 void udp6_proc_exit(struct net *net) 1707 { 1708 remove_proc_entry("udp6", net->proc_net); 1709 } 1710 #endif /* CONFIG_PROC_FS */ 1711 1712 /* ------------------------------------------------------------------------ */ 1713 1714 struct proto udpv6_prot = { 1715 .name = "UDPv6", 1716 .owner = THIS_MODULE, 1717 .close = udp_lib_close, 1718 .pre_connect = udpv6_pre_connect, 1719 .connect = ip6_datagram_connect, 1720 .disconnect = udp_disconnect, 1721 .ioctl = udp_ioctl, 1722 .init = udp_init_sock, 1723 .destroy = udpv6_destroy_sock, 1724 .setsockopt = udpv6_setsockopt, 1725 .getsockopt = udpv6_getsockopt, 1726 .sendmsg = udpv6_sendmsg, 1727 .recvmsg = udpv6_recvmsg, 1728 .release_cb = ip6_datagram_release_cb, 1729 .hash = udp_lib_hash, 1730 .unhash = udp_lib_unhash, 1731 .rehash = udp_v6_rehash, 1732 .get_port = udp_v6_get_port, 1733 #ifdef CONFIG_BPF_SYSCALL 1734 .psock_update_sk_prot = udp_bpf_update_proto, 1735 #endif 1736 .memory_allocated = &udp_memory_allocated, 1737 .sysctl_mem = sysctl_udp_mem, 1738 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1739 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1740 .obj_size = sizeof(struct udp6_sock), 1741 .h.udp_table = &udp_table, 1742 .diag_destroy = udp_abort, 1743 }; 1744 1745 static struct inet_protosw udpv6_protosw = { 1746 .type = SOCK_DGRAM, 1747 .protocol = IPPROTO_UDP, 1748 .prot = &udpv6_prot, 1749 .ops = &inet6_dgram_ops, 1750 .flags = INET_PROTOSW_PERMANENT, 1751 }; 1752 1753 int __init udpv6_init(void) 1754 { 1755 int ret; 1756 1757 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1758 if (ret) 1759 goto out; 1760 1761 ret = inet6_register_protosw(&udpv6_protosw); 1762 if (ret) 1763 goto out_udpv6_protocol; 1764 out: 1765 return ret; 1766 1767 out_udpv6_protocol: 1768 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1769 goto out; 1770 } 1771 1772 void udpv6_exit(void) 1773 { 1774 inet6_unregister_protosw(&udpv6_protosw); 1775 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1776 } 1777