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