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 /* called with rcu_read_lock() */ 145 static struct sock *udp6_lib_lookup2(struct net *net, 146 const struct in6_addr *saddr, __be16 sport, 147 const struct in6_addr *daddr, unsigned int hnum, 148 int dif, int sdif, struct udp_hslot *hslot2, 149 struct sk_buff *skb) 150 { 151 struct sock *sk, *result; 152 int score, badness; 153 u32 hash = 0; 154 155 result = NULL; 156 badness = -1; 157 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 158 score = compute_score(sk, net, saddr, sport, 159 daddr, hnum, dif, sdif); 160 if (score > badness) { 161 if (sk->sk_reuseport && 162 sk->sk_state != TCP_ESTABLISHED) { 163 hash = udp6_ehashfn(net, daddr, hnum, 164 saddr, sport); 165 166 result = reuseport_select_sock(sk, hash, skb, 167 sizeof(struct udphdr)); 168 if (result && !reuseport_has_conns(sk, false)) 169 return result; 170 } 171 result = sk; 172 badness = score; 173 } 174 } 175 return result; 176 } 177 178 /* rcu_read_lock() must be held */ 179 struct sock *__udp6_lib_lookup(struct net *net, 180 const struct in6_addr *saddr, __be16 sport, 181 const struct in6_addr *daddr, __be16 dport, 182 int dif, int sdif, struct udp_table *udptable, 183 struct sk_buff *skb) 184 { 185 unsigned short hnum = ntohs(dport); 186 unsigned int hash2, slot2; 187 struct udp_hslot *hslot2; 188 struct sock *result; 189 190 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 191 slot2 = hash2 & udptable->mask; 192 hslot2 = &udptable->hash2[slot2]; 193 194 result = udp6_lib_lookup2(net, saddr, sport, 195 daddr, hnum, dif, sdif, 196 hslot2, skb); 197 if (!result) { 198 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 199 slot2 = hash2 & udptable->mask; 200 201 hslot2 = &udptable->hash2[slot2]; 202 203 result = udp6_lib_lookup2(net, saddr, sport, 204 &in6addr_any, hnum, dif, sdif, 205 hslot2, skb); 206 } 207 if (IS_ERR(result)) 208 return NULL; 209 return result; 210 } 211 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 212 213 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 214 __be16 sport, __be16 dport, 215 struct udp_table *udptable) 216 { 217 const struct ipv6hdr *iph = ipv6_hdr(skb); 218 219 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 220 &iph->daddr, dport, inet6_iif(skb), 221 inet6_sdif(skb), udptable, skb); 222 } 223 224 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, 225 __be16 sport, __be16 dport) 226 { 227 const struct ipv6hdr *iph = ipv6_hdr(skb); 228 229 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 230 &iph->daddr, dport, inet6_iif(skb), 231 inet6_sdif(skb), &udp_table, NULL); 232 } 233 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); 234 235 /* Must be called under rcu_read_lock(). 236 * Does increment socket refcount. 237 */ 238 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 239 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 240 const struct in6_addr *daddr, __be16 dport, int dif) 241 { 242 struct sock *sk; 243 244 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 245 dif, 0, &udp_table, NULL); 246 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 247 sk = NULL; 248 return sk; 249 } 250 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 251 #endif 252 253 /* do not use the scratch area len for jumbogram: their length execeeds the 254 * scratch area space; note that the IP6CB flags is still in the first 255 * cacheline, so checking for jumbograms is cheap 256 */ 257 static int udp6_skb_len(struct sk_buff *skb) 258 { 259 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 260 } 261 262 /* 263 * This should be easy, if there is something there we 264 * return it, otherwise we block. 265 */ 266 267 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 268 int noblock, int flags, int *addr_len) 269 { 270 struct ipv6_pinfo *np = inet6_sk(sk); 271 struct inet_sock *inet = inet_sk(sk); 272 struct sk_buff *skb; 273 unsigned int ulen, copied; 274 int off, err, peeking = flags & MSG_PEEK; 275 int is_udplite = IS_UDPLITE(sk); 276 struct udp_mib __percpu *mib; 277 bool checksum_valid = false; 278 int is_udp4; 279 280 if (flags & MSG_ERRQUEUE) 281 return ipv6_recv_error(sk, msg, len, addr_len); 282 283 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 284 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 285 286 try_again: 287 off = sk_peek_offset(sk, flags); 288 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 289 if (!skb) 290 return err; 291 292 ulen = udp6_skb_len(skb); 293 copied = len; 294 if (copied > ulen - off) 295 copied = ulen - off; 296 else if (copied < ulen) 297 msg->msg_flags |= MSG_TRUNC; 298 299 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 300 mib = __UDPX_MIB(sk, is_udp4); 301 302 /* 303 * If checksum is needed at all, try to do it while copying the 304 * data. If the data is truncated, or if we only want a partial 305 * coverage checksum (UDP-Lite), do it before the copy. 306 */ 307 308 if (copied < ulen || peeking || 309 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 310 checksum_valid = udp_skb_csum_unnecessary(skb) || 311 !__udp_lib_checksum_complete(skb); 312 if (!checksum_valid) 313 goto csum_copy_err; 314 } 315 316 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 317 if (udp_skb_is_linear(skb)) 318 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 319 else 320 err = skb_copy_datagram_msg(skb, off, msg, copied); 321 } else { 322 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 323 if (err == -EINVAL) 324 goto csum_copy_err; 325 } 326 if (unlikely(err)) { 327 if (!peeking) { 328 atomic_inc(&sk->sk_drops); 329 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 330 } 331 kfree_skb(skb); 332 return err; 333 } 334 if (!peeking) 335 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 336 337 sock_recv_ts_and_drops(msg, sk, skb); 338 339 /* Copy the address. */ 340 if (msg->msg_name) { 341 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 342 sin6->sin6_family = AF_INET6; 343 sin6->sin6_port = udp_hdr(skb)->source; 344 sin6->sin6_flowinfo = 0; 345 346 if (is_udp4) { 347 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 348 &sin6->sin6_addr); 349 sin6->sin6_scope_id = 0; 350 } else { 351 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 352 sin6->sin6_scope_id = 353 ipv6_iface_scope_id(&sin6->sin6_addr, 354 inet6_iif(skb)); 355 } 356 *addr_len = sizeof(*sin6); 357 358 if (cgroup_bpf_enabled) 359 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 360 (struct sockaddr *)sin6); 361 } 362 363 if (udp_sk(sk)->gro_enabled) 364 udp_cmsg_recv(msg, sk, skb); 365 366 if (np->rxopt.all) 367 ip6_datagram_recv_common_ctl(sk, msg, skb); 368 369 if (is_udp4) { 370 if (inet->cmsg_flags) 371 ip_cmsg_recv_offset(msg, sk, skb, 372 sizeof(struct udphdr), off); 373 } else { 374 if (np->rxopt.all) 375 ip6_datagram_recv_specific_ctl(sk, msg, skb); 376 } 377 378 err = copied; 379 if (flags & MSG_TRUNC) 380 err = ulen; 381 382 skb_consume_udp(sk, skb, peeking ? -err : err); 383 return err; 384 385 csum_copy_err: 386 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 387 udp_skb_destructor)) { 388 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 389 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 390 } 391 kfree_skb(skb); 392 393 /* starting over for a new packet, but check if we need to yield */ 394 cond_resched(); 395 msg->msg_flags &= ~MSG_TRUNC; 396 goto try_again; 397 } 398 399 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 400 void udpv6_encap_enable(void) 401 { 402 static_branch_inc(&udpv6_encap_needed_key); 403 } 404 EXPORT_SYMBOL(udpv6_encap_enable); 405 406 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 407 * through error handlers in encapsulations looking for a match. 408 */ 409 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 410 struct inet6_skb_parm *opt, 411 u8 type, u8 code, int offset, __be32 info) 412 { 413 int i; 414 415 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 416 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 417 u8 type, u8 code, int offset, __be32 info); 418 const struct ip6_tnl_encap_ops *encap; 419 420 encap = rcu_dereference(ip6tun_encaps[i]); 421 if (!encap) 422 continue; 423 handler = encap->err_handler; 424 if (handler && !handler(skb, opt, type, code, offset, info)) 425 return 0; 426 } 427 428 return -ENOENT; 429 } 430 431 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 432 * reversing source and destination port: this will match tunnels that force the 433 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 434 * lwtunnels might actually break this assumption by being configured with 435 * different destination ports on endpoints, in this case we won't be able to 436 * trace ICMP messages back to them. 437 * 438 * If this doesn't match any socket, probe tunnels with arbitrary destination 439 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 440 * we've sent packets to won't necessarily match the local destination port. 441 * 442 * Then ask the tunnel implementation to match the error against a valid 443 * association. 444 * 445 * Return an error if we can't find a match, the socket if we need further 446 * processing, zero otherwise. 447 */ 448 static struct sock *__udp6_lib_err_encap(struct net *net, 449 const struct ipv6hdr *hdr, int offset, 450 struct udphdr *uh, 451 struct udp_table *udptable, 452 struct sk_buff *skb, 453 struct inet6_skb_parm *opt, 454 u8 type, u8 code, __be32 info) 455 { 456 int network_offset, transport_offset; 457 struct sock *sk; 458 459 network_offset = skb_network_offset(skb); 460 transport_offset = skb_transport_offset(skb); 461 462 /* Network header needs to point to the outer IPv6 header inside ICMP */ 463 skb_reset_network_header(skb); 464 465 /* Transport header needs to point to the UDP header */ 466 skb_set_transport_header(skb, offset); 467 468 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 469 &hdr->saddr, uh->dest, 470 inet6_iif(skb), 0, udptable, skb); 471 if (sk) { 472 int (*lookup)(struct sock *sk, struct sk_buff *skb); 473 struct udp_sock *up = udp_sk(sk); 474 475 lookup = READ_ONCE(up->encap_err_lookup); 476 if (!lookup || lookup(sk, skb)) 477 sk = NULL; 478 } 479 480 if (!sk) { 481 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 482 offset, info)); 483 } 484 485 skb_set_transport_header(skb, transport_offset); 486 skb_set_network_header(skb, network_offset); 487 488 return sk; 489 } 490 491 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 492 u8 type, u8 code, int offset, __be32 info, 493 struct udp_table *udptable) 494 { 495 struct ipv6_pinfo *np; 496 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 497 const struct in6_addr *saddr = &hdr->saddr; 498 const struct in6_addr *daddr = &hdr->daddr; 499 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 500 bool tunnel = false; 501 struct sock *sk; 502 int harderr; 503 int err; 504 struct net *net = dev_net(skb->dev); 505 506 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 507 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 508 if (!sk) { 509 /* No socket for error: try tunnels before discarding */ 510 sk = ERR_PTR(-ENOENT); 511 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 512 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 513 udptable, skb, 514 opt, type, code, info); 515 if (!sk) 516 return 0; 517 } 518 519 if (IS_ERR(sk)) { 520 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 521 ICMP6_MIB_INERRORS); 522 return PTR_ERR(sk); 523 } 524 525 tunnel = true; 526 } 527 528 harderr = icmpv6_err_convert(type, code, &err); 529 np = inet6_sk(sk); 530 531 if (type == ICMPV6_PKT_TOOBIG) { 532 if (!ip6_sk_accept_pmtu(sk)) 533 goto out; 534 ip6_sk_update_pmtu(skb, sk, info); 535 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 536 harderr = 1; 537 } 538 if (type == NDISC_REDIRECT) { 539 if (tunnel) { 540 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 541 sk->sk_mark, sk->sk_uid); 542 } else { 543 ip6_sk_redirect(skb, sk); 544 } 545 goto out; 546 } 547 548 /* Tunnels don't have an application socket: don't pass errors back */ 549 if (tunnel) 550 goto out; 551 552 if (!np->recverr) { 553 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 554 goto out; 555 } else { 556 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 557 } 558 559 sk->sk_err = err; 560 sk->sk_error_report(sk); 561 out: 562 return 0; 563 } 564 565 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 566 { 567 int rc; 568 569 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 570 sock_rps_save_rxhash(sk, skb); 571 sk_mark_napi_id(sk, skb); 572 sk_incoming_cpu_update(sk); 573 } else { 574 sk_mark_napi_id_once(sk, skb); 575 } 576 577 rc = __udp_enqueue_schedule_skb(sk, skb); 578 if (rc < 0) { 579 int is_udplite = IS_UDPLITE(sk); 580 581 /* Note that an ENOMEM error is charged twice */ 582 if (rc == -ENOMEM) 583 UDP6_INC_STATS(sock_net(sk), 584 UDP_MIB_RCVBUFERRORS, is_udplite); 585 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 586 kfree_skb(skb); 587 return -1; 588 } 589 590 return 0; 591 } 592 593 static __inline__ int udpv6_err(struct sk_buff *skb, 594 struct inet6_skb_parm *opt, u8 type, 595 u8 code, int offset, __be32 info) 596 { 597 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 598 } 599 600 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 601 { 602 struct udp_sock *up = udp_sk(sk); 603 int is_udplite = IS_UDPLITE(sk); 604 605 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 606 goto drop; 607 608 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 609 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 610 611 /* 612 * This is an encapsulation socket so pass the skb to 613 * the socket's udp_encap_rcv() hook. Otherwise, just 614 * fall through and pass this up the UDP socket. 615 * up->encap_rcv() returns the following value: 616 * =0 if skb was successfully passed to the encap 617 * handler or was discarded by it. 618 * >0 if skb should be passed on to UDP. 619 * <0 if skb should be resubmitted as proto -N 620 */ 621 622 /* if we're overly short, let UDP handle it */ 623 encap_rcv = READ_ONCE(up->encap_rcv); 624 if (encap_rcv) { 625 int ret; 626 627 /* Verify checksum before giving to encap */ 628 if (udp_lib_checksum_complete(skb)) 629 goto csum_error; 630 631 ret = encap_rcv(sk, skb); 632 if (ret <= 0) { 633 __UDP_INC_STATS(sock_net(sk), 634 UDP_MIB_INDATAGRAMS, 635 is_udplite); 636 return -ret; 637 } 638 } 639 640 /* FALLTHROUGH -- it's a UDP Packet */ 641 } 642 643 /* 644 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 645 */ 646 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 647 648 if (up->pcrlen == 0) { /* full coverage was set */ 649 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 650 UDP_SKB_CB(skb)->cscov, skb->len); 651 goto drop; 652 } 653 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 654 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 655 UDP_SKB_CB(skb)->cscov, up->pcrlen); 656 goto drop; 657 } 658 } 659 660 prefetch(&sk->sk_rmem_alloc); 661 if (rcu_access_pointer(sk->sk_filter) && 662 udp_lib_checksum_complete(skb)) 663 goto csum_error; 664 665 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 666 goto drop; 667 668 udp_csum_pull_header(skb); 669 670 skb_dst_drop(skb); 671 672 return __udpv6_queue_rcv_skb(sk, skb); 673 674 csum_error: 675 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 676 drop: 677 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 678 atomic_inc(&sk->sk_drops); 679 kfree_skb(skb); 680 return -1; 681 } 682 683 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 684 { 685 struct sk_buff *next, *segs; 686 int ret; 687 688 if (likely(!udp_unexpected_gso(sk, skb))) 689 return udpv6_queue_rcv_one_skb(sk, skb); 690 691 __skb_push(skb, -skb_mac_offset(skb)); 692 segs = udp_rcv_segment(sk, skb, false); 693 skb_list_walk_safe(segs, skb, next) { 694 __skb_pull(skb, skb_transport_offset(skb)); 695 696 ret = udpv6_queue_rcv_one_skb(sk, skb); 697 if (ret > 0) 698 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 699 true); 700 } 701 return 0; 702 } 703 704 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 705 __be16 loc_port, const struct in6_addr *loc_addr, 706 __be16 rmt_port, const struct in6_addr *rmt_addr, 707 int dif, int sdif, unsigned short hnum) 708 { 709 struct inet_sock *inet = inet_sk(sk); 710 711 if (!net_eq(sock_net(sk), net)) 712 return false; 713 714 if (udp_sk(sk)->udp_port_hash != hnum || 715 sk->sk_family != PF_INET6 || 716 (inet->inet_dport && inet->inet_dport != rmt_port) || 717 (!ipv6_addr_any(&sk->sk_v6_daddr) && 718 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 719 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) || 720 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 721 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 722 return false; 723 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 724 return false; 725 return true; 726 } 727 728 static void udp6_csum_zero_error(struct sk_buff *skb) 729 { 730 /* RFC 2460 section 8.1 says that we SHOULD log 731 * this error. Well, it is reasonable. 732 */ 733 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 734 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 735 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 736 } 737 738 /* 739 * Note: called only from the BH handler context, 740 * so we don't need to lock the hashes. 741 */ 742 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 743 const struct in6_addr *saddr, const struct in6_addr *daddr, 744 struct udp_table *udptable, int proto) 745 { 746 struct sock *sk, *first = NULL; 747 const struct udphdr *uh = udp_hdr(skb); 748 unsigned short hnum = ntohs(uh->dest); 749 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 750 unsigned int offset = offsetof(typeof(*sk), sk_node); 751 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 752 int dif = inet6_iif(skb); 753 int sdif = inet6_sdif(skb); 754 struct hlist_node *node; 755 struct sk_buff *nskb; 756 757 if (use_hash2) { 758 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 759 udptable->mask; 760 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 761 start_lookup: 762 hslot = &udptable->hash2[hash2]; 763 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 764 } 765 766 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 767 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 768 uh->source, saddr, dif, sdif, 769 hnum)) 770 continue; 771 /* If zero checksum and no_check is not on for 772 * the socket then skip it. 773 */ 774 if (!uh->check && !udp_sk(sk)->no_check6_rx) 775 continue; 776 if (!first) { 777 first = sk; 778 continue; 779 } 780 nskb = skb_clone(skb, GFP_ATOMIC); 781 if (unlikely(!nskb)) { 782 atomic_inc(&sk->sk_drops); 783 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 784 IS_UDPLITE(sk)); 785 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 786 IS_UDPLITE(sk)); 787 continue; 788 } 789 790 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 791 consume_skb(nskb); 792 } 793 794 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 795 if (use_hash2 && hash2 != hash2_any) { 796 hash2 = hash2_any; 797 goto start_lookup; 798 } 799 800 if (first) { 801 if (udpv6_queue_rcv_skb(first, skb) > 0) 802 consume_skb(skb); 803 } else { 804 kfree_skb(skb); 805 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 806 proto == IPPROTO_UDPLITE); 807 } 808 return 0; 809 } 810 811 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 812 { 813 if (udp_sk_rx_dst_set(sk, dst)) { 814 const struct rt6_info *rt = (const struct rt6_info *)dst; 815 816 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 817 } 818 } 819 820 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 821 * return code conversion for ip layer consumption 822 */ 823 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 824 struct udphdr *uh) 825 { 826 int ret; 827 828 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 829 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 830 831 ret = udpv6_queue_rcv_skb(sk, skb); 832 833 /* a return value > 0 means to resubmit the input */ 834 if (ret > 0) 835 return ret; 836 return 0; 837 } 838 839 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 840 int proto) 841 { 842 const struct in6_addr *saddr, *daddr; 843 struct net *net = dev_net(skb->dev); 844 struct udphdr *uh; 845 struct sock *sk; 846 bool refcounted; 847 u32 ulen = 0; 848 849 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 850 goto discard; 851 852 saddr = &ipv6_hdr(skb)->saddr; 853 daddr = &ipv6_hdr(skb)->daddr; 854 uh = udp_hdr(skb); 855 856 ulen = ntohs(uh->len); 857 if (ulen > skb->len) 858 goto short_packet; 859 860 if (proto == IPPROTO_UDP) { 861 /* UDP validates ulen. */ 862 863 /* Check for jumbo payload */ 864 if (ulen == 0) 865 ulen = skb->len; 866 867 if (ulen < sizeof(*uh)) 868 goto short_packet; 869 870 if (ulen < skb->len) { 871 if (pskb_trim_rcsum(skb, ulen)) 872 goto short_packet; 873 saddr = &ipv6_hdr(skb)->saddr; 874 daddr = &ipv6_hdr(skb)->daddr; 875 uh = udp_hdr(skb); 876 } 877 } 878 879 if (udp6_csum_init(skb, uh, proto)) 880 goto csum_error; 881 882 /* Check if the socket is already available, e.g. due to early demux */ 883 sk = skb_steal_sock(skb, &refcounted); 884 if (sk) { 885 struct dst_entry *dst = skb_dst(skb); 886 int ret; 887 888 if (unlikely(sk->sk_rx_dst != dst)) 889 udp6_sk_rx_dst_set(sk, dst); 890 891 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 892 if (refcounted) 893 sock_put(sk); 894 goto report_csum_error; 895 } 896 897 ret = udp6_unicast_rcv_skb(sk, skb, uh); 898 if (refcounted) 899 sock_put(sk); 900 return ret; 901 } 902 903 /* 904 * Multicast receive code 905 */ 906 if (ipv6_addr_is_multicast(daddr)) 907 return __udp6_lib_mcast_deliver(net, skb, 908 saddr, daddr, udptable, proto); 909 910 /* Unicast */ 911 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 912 if (sk) { 913 if (!uh->check && !udp_sk(sk)->no_check6_rx) 914 goto report_csum_error; 915 return udp6_unicast_rcv_skb(sk, skb, uh); 916 } 917 918 if (!uh->check) 919 goto report_csum_error; 920 921 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 922 goto discard; 923 924 if (udp_lib_checksum_complete(skb)) 925 goto csum_error; 926 927 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 928 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 929 930 kfree_skb(skb); 931 return 0; 932 933 short_packet: 934 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 935 proto == IPPROTO_UDPLITE ? "-Lite" : "", 936 saddr, ntohs(uh->source), 937 ulen, skb->len, 938 daddr, ntohs(uh->dest)); 939 goto discard; 940 941 report_csum_error: 942 udp6_csum_zero_error(skb); 943 csum_error: 944 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 945 discard: 946 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 947 kfree_skb(skb); 948 return 0; 949 } 950 951 952 static struct sock *__udp6_lib_demux_lookup(struct net *net, 953 __be16 loc_port, const struct in6_addr *loc_addr, 954 __be16 rmt_port, const struct in6_addr *rmt_addr, 955 int dif, int sdif) 956 { 957 unsigned short hnum = ntohs(loc_port); 958 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 959 unsigned int slot2 = hash2 & udp_table.mask; 960 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 961 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 962 struct sock *sk; 963 964 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 965 if (sk->sk_state == TCP_ESTABLISHED && 966 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 967 return sk; 968 /* Only check first socket in chain */ 969 break; 970 } 971 return NULL; 972 } 973 974 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb) 975 { 976 struct net *net = dev_net(skb->dev); 977 const struct udphdr *uh; 978 struct sock *sk; 979 struct dst_entry *dst; 980 int dif = skb->dev->ifindex; 981 int sdif = inet6_sdif(skb); 982 983 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 984 sizeof(struct udphdr))) 985 return; 986 987 uh = udp_hdr(skb); 988 989 if (skb->pkt_type == PACKET_HOST) 990 sk = __udp6_lib_demux_lookup(net, uh->dest, 991 &ipv6_hdr(skb)->daddr, 992 uh->source, &ipv6_hdr(skb)->saddr, 993 dif, sdif); 994 else 995 return; 996 997 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 998 return; 999 1000 skb->sk = sk; 1001 skb->destructor = sock_efree; 1002 dst = READ_ONCE(sk->sk_rx_dst); 1003 1004 if (dst) 1005 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 1006 if (dst) { 1007 /* set noref for now. 1008 * any place which wants to hold dst has to call 1009 * dst_hold_safe() 1010 */ 1011 skb_dst_set_noref(skb, dst); 1012 } 1013 } 1014 1015 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1016 { 1017 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1018 } 1019 1020 /* 1021 * Throw away all pending data and cancel the corking. Socket is locked. 1022 */ 1023 static void udp_v6_flush_pending_frames(struct sock *sk) 1024 { 1025 struct udp_sock *up = udp_sk(sk); 1026 1027 if (up->pending == AF_INET) 1028 udp_flush_pending_frames(sk); 1029 else if (up->pending) { 1030 up->len = 0; 1031 up->pending = 0; 1032 ip6_flush_pending_frames(sk); 1033 } 1034 } 1035 1036 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1037 int addr_len) 1038 { 1039 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1040 return -EINVAL; 1041 /* The following checks are replicated from __ip6_datagram_connect() 1042 * and intended to prevent BPF program called below from accessing 1043 * bytes that are out of the bound specified by user in addr_len. 1044 */ 1045 if (uaddr->sa_family == AF_INET) { 1046 if (__ipv6_only_sock(sk)) 1047 return -EAFNOSUPPORT; 1048 return udp_pre_connect(sk, uaddr, addr_len); 1049 } 1050 1051 if (addr_len < SIN6_LEN_RFC2133) 1052 return -EINVAL; 1053 1054 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1055 } 1056 1057 /** 1058 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1059 * @sk: socket we are sending on 1060 * @skb: sk_buff containing the filled-in UDP header 1061 * (checksum field must be zeroed out) 1062 */ 1063 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1064 const struct in6_addr *saddr, 1065 const struct in6_addr *daddr, int len) 1066 { 1067 unsigned int offset; 1068 struct udphdr *uh = udp_hdr(skb); 1069 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1070 __wsum csum = 0; 1071 1072 if (!frags) { 1073 /* Only one fragment on the socket. */ 1074 skb->csum_start = skb_transport_header(skb) - skb->head; 1075 skb->csum_offset = offsetof(struct udphdr, check); 1076 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1077 } else { 1078 /* 1079 * HW-checksum won't work as there are two or more 1080 * fragments on the socket so that all csums of sk_buffs 1081 * should be together 1082 */ 1083 offset = skb_transport_offset(skb); 1084 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1085 csum = skb->csum; 1086 1087 skb->ip_summed = CHECKSUM_NONE; 1088 1089 do { 1090 csum = csum_add(csum, frags->csum); 1091 } while ((frags = frags->next)); 1092 1093 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1094 csum); 1095 if (uh->check == 0) 1096 uh->check = CSUM_MANGLED_0; 1097 } 1098 } 1099 1100 /* 1101 * Sending 1102 */ 1103 1104 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1105 struct inet_cork *cork) 1106 { 1107 struct sock *sk = skb->sk; 1108 struct udphdr *uh; 1109 int err = 0; 1110 int is_udplite = IS_UDPLITE(sk); 1111 __wsum csum = 0; 1112 int offset = skb_transport_offset(skb); 1113 int len = skb->len - offset; 1114 int datalen = len - sizeof(*uh); 1115 1116 /* 1117 * Create a UDP header 1118 */ 1119 uh = udp_hdr(skb); 1120 uh->source = fl6->fl6_sport; 1121 uh->dest = fl6->fl6_dport; 1122 uh->len = htons(len); 1123 uh->check = 0; 1124 1125 if (cork->gso_size) { 1126 const int hlen = skb_network_header_len(skb) + 1127 sizeof(struct udphdr); 1128 1129 if (hlen + cork->gso_size > cork->fragsize) { 1130 kfree_skb(skb); 1131 return -EINVAL; 1132 } 1133 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 1134 kfree_skb(skb); 1135 return -EINVAL; 1136 } 1137 if (udp_sk(sk)->no_check6_tx) { 1138 kfree_skb(skb); 1139 return -EINVAL; 1140 } 1141 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1142 dst_xfrm(skb_dst(skb))) { 1143 kfree_skb(skb); 1144 return -EIO; 1145 } 1146 1147 if (datalen > cork->gso_size) { 1148 skb_shinfo(skb)->gso_size = cork->gso_size; 1149 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1150 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1151 cork->gso_size); 1152 } 1153 goto csum_partial; 1154 } 1155 1156 if (is_udplite) 1157 csum = udplite_csum(skb); 1158 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1159 skb->ip_summed = CHECKSUM_NONE; 1160 goto send; 1161 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1162 csum_partial: 1163 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1164 goto send; 1165 } else 1166 csum = udp_csum(skb); 1167 1168 /* add protocol-dependent pseudo-header */ 1169 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1170 len, fl6->flowi6_proto, csum); 1171 if (uh->check == 0) 1172 uh->check = CSUM_MANGLED_0; 1173 1174 send: 1175 err = ip6_send_skb(skb); 1176 if (err) { 1177 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1178 UDP6_INC_STATS(sock_net(sk), 1179 UDP_MIB_SNDBUFERRORS, is_udplite); 1180 err = 0; 1181 } 1182 } else { 1183 UDP6_INC_STATS(sock_net(sk), 1184 UDP_MIB_OUTDATAGRAMS, is_udplite); 1185 } 1186 return err; 1187 } 1188 1189 static int udp_v6_push_pending_frames(struct sock *sk) 1190 { 1191 struct sk_buff *skb; 1192 struct udp_sock *up = udp_sk(sk); 1193 struct flowi6 fl6; 1194 int err = 0; 1195 1196 if (up->pending == AF_INET) 1197 return udp_push_pending_frames(sk); 1198 1199 /* ip6_finish_skb will release the cork, so make a copy of 1200 * fl6 here. 1201 */ 1202 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1203 1204 skb = ip6_finish_skb(sk); 1205 if (!skb) 1206 goto out; 1207 1208 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1209 1210 out: 1211 up->len = 0; 1212 up->pending = 0; 1213 return err; 1214 } 1215 1216 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1217 { 1218 struct ipv6_txoptions opt_space; 1219 struct udp_sock *up = udp_sk(sk); 1220 struct inet_sock *inet = inet_sk(sk); 1221 struct ipv6_pinfo *np = inet6_sk(sk); 1222 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1223 struct in6_addr *daddr, *final_p, final; 1224 struct ipv6_txoptions *opt = NULL; 1225 struct ipv6_txoptions *opt_to_free = NULL; 1226 struct ip6_flowlabel *flowlabel = NULL; 1227 struct flowi6 fl6; 1228 struct dst_entry *dst; 1229 struct ipcm6_cookie ipc6; 1230 int addr_len = msg->msg_namelen; 1231 bool connected = false; 1232 int ulen = len; 1233 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1234 int err; 1235 int is_udplite = IS_UDPLITE(sk); 1236 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1237 1238 ipcm6_init(&ipc6); 1239 ipc6.gso_size = up->gso_size; 1240 ipc6.sockc.tsflags = sk->sk_tsflags; 1241 ipc6.sockc.mark = sk->sk_mark; 1242 1243 /* destination address check */ 1244 if (sin6) { 1245 if (addr_len < offsetof(struct sockaddr, sa_data)) 1246 return -EINVAL; 1247 1248 switch (sin6->sin6_family) { 1249 case AF_INET6: 1250 if (addr_len < SIN6_LEN_RFC2133) 1251 return -EINVAL; 1252 daddr = &sin6->sin6_addr; 1253 if (ipv6_addr_any(daddr) && 1254 ipv6_addr_v4mapped(&np->saddr)) 1255 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1256 daddr); 1257 break; 1258 case AF_INET: 1259 goto do_udp_sendmsg; 1260 case AF_UNSPEC: 1261 msg->msg_name = sin6 = NULL; 1262 msg->msg_namelen = addr_len = 0; 1263 daddr = NULL; 1264 break; 1265 default: 1266 return -EINVAL; 1267 } 1268 } else if (!up->pending) { 1269 if (sk->sk_state != TCP_ESTABLISHED) 1270 return -EDESTADDRREQ; 1271 daddr = &sk->sk_v6_daddr; 1272 } else 1273 daddr = NULL; 1274 1275 if (daddr) { 1276 if (ipv6_addr_v4mapped(daddr)) { 1277 struct sockaddr_in sin; 1278 sin.sin_family = AF_INET; 1279 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1280 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1281 msg->msg_name = &sin; 1282 msg->msg_namelen = sizeof(sin); 1283 do_udp_sendmsg: 1284 if (__ipv6_only_sock(sk)) 1285 return -ENETUNREACH; 1286 return udp_sendmsg(sk, msg, len); 1287 } 1288 } 1289 1290 if (up->pending == AF_INET) 1291 return udp_sendmsg(sk, msg, len); 1292 1293 /* Rough check on arithmetic overflow, 1294 better check is made in ip6_append_data(). 1295 */ 1296 if (len > INT_MAX - sizeof(struct udphdr)) 1297 return -EMSGSIZE; 1298 1299 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1300 if (up->pending) { 1301 /* 1302 * There are pending frames. 1303 * The socket lock must be held while it's corked. 1304 */ 1305 lock_sock(sk); 1306 if (likely(up->pending)) { 1307 if (unlikely(up->pending != AF_INET6)) { 1308 release_sock(sk); 1309 return -EAFNOSUPPORT; 1310 } 1311 dst = NULL; 1312 goto do_append_data; 1313 } 1314 release_sock(sk); 1315 } 1316 ulen += sizeof(struct udphdr); 1317 1318 memset(&fl6, 0, sizeof(fl6)); 1319 1320 if (sin6) { 1321 if (sin6->sin6_port == 0) 1322 return -EINVAL; 1323 1324 fl6.fl6_dport = sin6->sin6_port; 1325 daddr = &sin6->sin6_addr; 1326 1327 if (np->sndflow) { 1328 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1329 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1330 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1331 if (IS_ERR(flowlabel)) 1332 return -EINVAL; 1333 } 1334 } 1335 1336 /* 1337 * Otherwise it will be difficult to maintain 1338 * sk->sk_dst_cache. 1339 */ 1340 if (sk->sk_state == TCP_ESTABLISHED && 1341 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1342 daddr = &sk->sk_v6_daddr; 1343 1344 if (addr_len >= sizeof(struct sockaddr_in6) && 1345 sin6->sin6_scope_id && 1346 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1347 fl6.flowi6_oif = sin6->sin6_scope_id; 1348 } else { 1349 if (sk->sk_state != TCP_ESTABLISHED) 1350 return -EDESTADDRREQ; 1351 1352 fl6.fl6_dport = inet->inet_dport; 1353 daddr = &sk->sk_v6_daddr; 1354 fl6.flowlabel = np->flow_label; 1355 connected = true; 1356 } 1357 1358 if (!fl6.flowi6_oif) 1359 fl6.flowi6_oif = sk->sk_bound_dev_if; 1360 1361 if (!fl6.flowi6_oif) 1362 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1363 1364 fl6.flowi6_mark = ipc6.sockc.mark; 1365 fl6.flowi6_uid = sk->sk_uid; 1366 1367 if (msg->msg_controllen) { 1368 opt = &opt_space; 1369 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1370 opt->tot_len = sizeof(*opt); 1371 ipc6.opt = opt; 1372 1373 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1374 if (err > 0) 1375 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1376 &ipc6); 1377 if (err < 0) { 1378 fl6_sock_release(flowlabel); 1379 return err; 1380 } 1381 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1382 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1383 if (IS_ERR(flowlabel)) 1384 return -EINVAL; 1385 } 1386 if (!(opt->opt_nflen|opt->opt_flen)) 1387 opt = NULL; 1388 connected = false; 1389 } 1390 if (!opt) { 1391 opt = txopt_get(np); 1392 opt_to_free = opt; 1393 } 1394 if (flowlabel) 1395 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1396 opt = ipv6_fixup_options(&opt_space, opt); 1397 ipc6.opt = opt; 1398 1399 fl6.flowi6_proto = sk->sk_protocol; 1400 fl6.daddr = *daddr; 1401 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1402 fl6.saddr = np->saddr; 1403 fl6.fl6_sport = inet->inet_sport; 1404 1405 if (cgroup_bpf_enabled && !connected) { 1406 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1407 (struct sockaddr *)sin6, &fl6.saddr); 1408 if (err) 1409 goto out_no_dst; 1410 if (sin6) { 1411 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1412 /* BPF program rewrote IPv6-only by IPv4-mapped 1413 * IPv6. It's currently unsupported. 1414 */ 1415 err = -ENOTSUPP; 1416 goto out_no_dst; 1417 } 1418 if (sin6->sin6_port == 0) { 1419 /* BPF program set invalid port. Reject it. */ 1420 err = -EINVAL; 1421 goto out_no_dst; 1422 } 1423 fl6.fl6_dport = sin6->sin6_port; 1424 fl6.daddr = sin6->sin6_addr; 1425 } 1426 } 1427 1428 if (ipv6_addr_any(&fl6.daddr)) 1429 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1430 1431 final_p = fl6_update_dst(&fl6, opt, &final); 1432 if (final_p) 1433 connected = false; 1434 1435 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1436 fl6.flowi6_oif = np->mcast_oif; 1437 connected = false; 1438 } else if (!fl6.flowi6_oif) 1439 fl6.flowi6_oif = np->ucast_oif; 1440 1441 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1442 1443 if (ipc6.tclass < 0) 1444 ipc6.tclass = np->tclass; 1445 1446 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1447 1448 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1449 if (IS_ERR(dst)) { 1450 err = PTR_ERR(dst); 1451 dst = NULL; 1452 goto out; 1453 } 1454 1455 if (ipc6.hlimit < 0) 1456 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1457 1458 if (msg->msg_flags&MSG_CONFIRM) 1459 goto do_confirm; 1460 back_from_confirm: 1461 1462 /* Lockless fast path for the non-corking case */ 1463 if (!corkreq) { 1464 struct inet_cork_full cork; 1465 struct sk_buff *skb; 1466 1467 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1468 sizeof(struct udphdr), &ipc6, 1469 &fl6, (struct rt6_info *)dst, 1470 msg->msg_flags, &cork); 1471 err = PTR_ERR(skb); 1472 if (!IS_ERR_OR_NULL(skb)) 1473 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1474 goto out; 1475 } 1476 1477 lock_sock(sk); 1478 if (unlikely(up->pending)) { 1479 /* The socket is already corked while preparing it. */ 1480 /* ... which is an evident application bug. --ANK */ 1481 release_sock(sk); 1482 1483 net_dbg_ratelimited("udp cork app bug 2\n"); 1484 err = -EINVAL; 1485 goto out; 1486 } 1487 1488 up->pending = AF_INET6; 1489 1490 do_append_data: 1491 if (ipc6.dontfrag < 0) 1492 ipc6.dontfrag = np->dontfrag; 1493 up->len += ulen; 1494 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1495 &ipc6, &fl6, (struct rt6_info *)dst, 1496 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1497 if (err) 1498 udp_v6_flush_pending_frames(sk); 1499 else if (!corkreq) 1500 err = udp_v6_push_pending_frames(sk); 1501 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1502 up->pending = 0; 1503 1504 if (err > 0) 1505 err = np->recverr ? net_xmit_errno(err) : 0; 1506 release_sock(sk); 1507 1508 out: 1509 dst_release(dst); 1510 out_no_dst: 1511 fl6_sock_release(flowlabel); 1512 txopt_put(opt_to_free); 1513 if (!err) 1514 return len; 1515 /* 1516 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1517 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1518 * we don't have a good statistic (IpOutDiscards but it can be too many 1519 * things). We could add another new stat but at least for now that 1520 * seems like overkill. 1521 */ 1522 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1523 UDP6_INC_STATS(sock_net(sk), 1524 UDP_MIB_SNDBUFERRORS, is_udplite); 1525 } 1526 return err; 1527 1528 do_confirm: 1529 if (msg->msg_flags & MSG_PROBE) 1530 dst_confirm_neigh(dst, &fl6.daddr); 1531 if (!(msg->msg_flags&MSG_PROBE) || len) 1532 goto back_from_confirm; 1533 err = 0; 1534 goto out; 1535 } 1536 1537 void udpv6_destroy_sock(struct sock *sk) 1538 { 1539 struct udp_sock *up = udp_sk(sk); 1540 lock_sock(sk); 1541 udp_v6_flush_pending_frames(sk); 1542 release_sock(sk); 1543 1544 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1545 if (up->encap_type) { 1546 void (*encap_destroy)(struct sock *sk); 1547 encap_destroy = READ_ONCE(up->encap_destroy); 1548 if (encap_destroy) 1549 encap_destroy(sk); 1550 } 1551 if (up->encap_enabled) 1552 static_branch_dec(&udpv6_encap_needed_key); 1553 } 1554 1555 inet6_destroy_sock(sk); 1556 } 1557 1558 /* 1559 * Socket option code for UDP 1560 */ 1561 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1562 char __user *optval, unsigned int optlen) 1563 { 1564 if (level == SOL_UDP || level == SOL_UDPLITE) 1565 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1566 udp_v6_push_pending_frames); 1567 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1568 } 1569 1570 #ifdef CONFIG_COMPAT 1571 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1572 char __user *optval, unsigned int optlen) 1573 { 1574 if (level == SOL_UDP || level == SOL_UDPLITE) 1575 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1576 udp_v6_push_pending_frames); 1577 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1578 } 1579 #endif 1580 1581 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1582 char __user *optval, int __user *optlen) 1583 { 1584 if (level == SOL_UDP || level == SOL_UDPLITE) 1585 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1586 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1587 } 1588 1589 #ifdef CONFIG_COMPAT 1590 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1591 char __user *optval, int __user *optlen) 1592 { 1593 if (level == SOL_UDP || level == SOL_UDPLITE) 1594 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1595 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1596 } 1597 #endif 1598 1599 /* thinking of making this const? Don't. 1600 * early_demux can change based on sysctl. 1601 */ 1602 static struct inet6_protocol udpv6_protocol = { 1603 .early_demux = udp_v6_early_demux, 1604 .early_demux_handler = udp_v6_early_demux, 1605 .handler = udpv6_rcv, 1606 .err_handler = udpv6_err, 1607 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1608 }; 1609 1610 /* ------------------------------------------------------------------------ */ 1611 #ifdef CONFIG_PROC_FS 1612 int udp6_seq_show(struct seq_file *seq, void *v) 1613 { 1614 if (v == SEQ_START_TOKEN) { 1615 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1616 } else { 1617 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1618 struct inet_sock *inet = inet_sk(v); 1619 __u16 srcp = ntohs(inet->inet_sport); 1620 __u16 destp = ntohs(inet->inet_dport); 1621 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1622 udp_rqueue_get(v), bucket); 1623 } 1624 return 0; 1625 } 1626 1627 const struct seq_operations udp6_seq_ops = { 1628 .start = udp_seq_start, 1629 .next = udp_seq_next, 1630 .stop = udp_seq_stop, 1631 .show = udp6_seq_show, 1632 }; 1633 EXPORT_SYMBOL(udp6_seq_ops); 1634 1635 static struct udp_seq_afinfo udp6_seq_afinfo = { 1636 .family = AF_INET6, 1637 .udp_table = &udp_table, 1638 }; 1639 1640 int __net_init udp6_proc_init(struct net *net) 1641 { 1642 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1643 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1644 return -ENOMEM; 1645 return 0; 1646 } 1647 1648 void udp6_proc_exit(struct net *net) 1649 { 1650 remove_proc_entry("udp6", net->proc_net); 1651 } 1652 #endif /* CONFIG_PROC_FS */ 1653 1654 /* ------------------------------------------------------------------------ */ 1655 1656 struct proto udpv6_prot = { 1657 .name = "UDPv6", 1658 .owner = THIS_MODULE, 1659 .close = udp_lib_close, 1660 .pre_connect = udpv6_pre_connect, 1661 .connect = ip6_datagram_connect, 1662 .disconnect = udp_disconnect, 1663 .ioctl = udp_ioctl, 1664 .init = udp_init_sock, 1665 .destroy = udpv6_destroy_sock, 1666 .setsockopt = udpv6_setsockopt, 1667 .getsockopt = udpv6_getsockopt, 1668 .sendmsg = udpv6_sendmsg, 1669 .recvmsg = udpv6_recvmsg, 1670 .release_cb = ip6_datagram_release_cb, 1671 .hash = udp_lib_hash, 1672 .unhash = udp_lib_unhash, 1673 .rehash = udp_v6_rehash, 1674 .get_port = udp_v6_get_port, 1675 .memory_allocated = &udp_memory_allocated, 1676 .sysctl_mem = sysctl_udp_mem, 1677 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1678 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1679 .obj_size = sizeof(struct udp6_sock), 1680 .h.udp_table = &udp_table, 1681 #ifdef CONFIG_COMPAT 1682 .compat_setsockopt = compat_udpv6_setsockopt, 1683 .compat_getsockopt = compat_udpv6_getsockopt, 1684 #endif 1685 .diag_destroy = udp_abort, 1686 }; 1687 1688 static struct inet_protosw udpv6_protosw = { 1689 .type = SOCK_DGRAM, 1690 .protocol = IPPROTO_UDP, 1691 .prot = &udpv6_prot, 1692 .ops = &inet6_dgram_ops, 1693 .flags = INET_PROTOSW_PERMANENT, 1694 }; 1695 1696 int __init udpv6_init(void) 1697 { 1698 int ret; 1699 1700 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1701 if (ret) 1702 goto out; 1703 1704 ret = inet6_register_protosw(&udpv6_protosw); 1705 if (ret) 1706 goto out_udpv6_protocol; 1707 out: 1708 return ret; 1709 1710 out_udpv6_protocol: 1711 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1712 goto out; 1713 } 1714 1715 void udpv6_exit(void) 1716 { 1717 inet6_unregister_protosw(&udpv6_protosw); 1718 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1719 } 1720