1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * UDP over IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on linux/ipv4/udp.c 10 * 11 * Fixes: 12 * Hideaki YOSHIFUJI : sin6_scope_id support 13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 15 * a single port at the same time. 16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. 18 */ 19 20 #include <linux/bpf-cgroup.h> 21 #include <linux/errno.h> 22 #include <linux/types.h> 23 #include <linux/socket.h> 24 #include <linux/sockios.h> 25 #include <linux/net.h> 26 #include <linux/in6.h> 27 #include <linux/netdevice.h> 28 #include <linux/if_arp.h> 29 #include <linux/ipv6.h> 30 #include <linux/icmpv6.h> 31 #include <linux/init.h> 32 #include <linux/module.h> 33 #include <linux/skbuff.h> 34 #include <linux/slab.h> 35 #include <linux/uaccess.h> 36 #include <linux/indirect_call_wrapper.h> 37 38 #include <net/addrconf.h> 39 #include <net/ndisc.h> 40 #include <net/protocol.h> 41 #include <net/transp_v6.h> 42 #include <net/ip6_route.h> 43 #include <net/raw.h> 44 #include <net/seg6.h> 45 #include <net/tcp_states.h> 46 #include <net/ip6_checksum.h> 47 #include <net/ip6_tunnel.h> 48 #include <net/xfrm.h> 49 #include <net/inet_hashtables.h> 50 #include <net/inet6_hashtables.h> 51 #include <net/busy_poll.h> 52 #include <net/sock_reuseport.h> 53 54 #include <linux/proc_fs.h> 55 #include <linux/seq_file.h> 56 #include <trace/events/skb.h> 57 #include "udp_impl.h" 58 59 static void udpv6_destruct_sock(struct sock *sk) 60 { 61 udp_destruct_common(sk); 62 inet6_sock_destruct(sk); 63 } 64 65 int udpv6_init_sock(struct sock *sk) 66 { 67 skb_queue_head_init(&udp_sk(sk)->reader_queue); 68 sk->sk_destruct = udpv6_destruct_sock; 69 return 0; 70 } 71 72 static u32 udp6_ehashfn(const struct net *net, 73 const struct in6_addr *laddr, 74 const u16 lport, 75 const struct in6_addr *faddr, 76 const __be16 fport) 77 { 78 static u32 udp6_ehash_secret __read_mostly; 79 static u32 udp_ipv6_hash_secret __read_mostly; 80 81 u32 lhash, fhash; 82 83 net_get_random_once(&udp6_ehash_secret, 84 sizeof(udp6_ehash_secret)); 85 net_get_random_once(&udp_ipv6_hash_secret, 86 sizeof(udp_ipv6_hash_secret)); 87 88 lhash = (__force u32)laddr->s6_addr32[3]; 89 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 90 91 return __inet6_ehashfn(lhash, lport, fhash, fport, 92 udp_ipv6_hash_secret + net_hash_mix(net)); 93 } 94 95 int udp_v6_get_port(struct sock *sk, unsigned short snum) 96 { 97 unsigned int hash2_nulladdr = 98 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 99 unsigned int hash2_partial = 100 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 101 102 /* precompute partial secondary hash */ 103 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 104 return udp_lib_get_port(sk, snum, hash2_nulladdr); 105 } 106 107 void udp_v6_rehash(struct sock *sk) 108 { 109 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 110 &sk->sk_v6_rcv_saddr, 111 inet_sk(sk)->inet_num); 112 113 udp_lib_rehash(sk, new_hash); 114 } 115 116 static int compute_score(struct sock *sk, struct net *net, 117 const struct in6_addr *saddr, __be16 sport, 118 const struct in6_addr *daddr, unsigned short hnum, 119 int dif, int sdif) 120 { 121 int bound_dev_if, score; 122 struct inet_sock *inet; 123 bool dev_match; 124 125 if (!net_eq(sock_net(sk), net) || 126 udp_sk(sk)->udp_port_hash != hnum || 127 sk->sk_family != PF_INET6) 128 return -1; 129 130 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 131 return -1; 132 133 score = 0; 134 inet = inet_sk(sk); 135 136 if (inet->inet_dport) { 137 if (inet->inet_dport != sport) 138 return -1; 139 score++; 140 } 141 142 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 143 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 144 return -1; 145 score++; 146 } 147 148 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 149 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif); 150 if (!dev_match) 151 return -1; 152 if (bound_dev_if) 153 score++; 154 155 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 156 score++; 157 158 return score; 159 } 160 161 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 162 struct sk_buff *skb, 163 const struct in6_addr *saddr, 164 __be16 sport, 165 const struct in6_addr *daddr, 166 unsigned int hnum) 167 { 168 struct sock *reuse_sk = NULL; 169 u32 hash; 170 171 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 172 hash = udp6_ehashfn(net, daddr, hnum, saddr, sport); 173 reuse_sk = reuseport_select_sock(sk, hash, skb, 174 sizeof(struct udphdr)); 175 } 176 return reuse_sk; 177 } 178 179 /* called with rcu_read_lock() */ 180 static struct sock *udp6_lib_lookup2(struct net *net, 181 const struct in6_addr *saddr, __be16 sport, 182 const struct in6_addr *daddr, unsigned int hnum, 183 int dif, int sdif, struct udp_hslot *hslot2, 184 struct sk_buff *skb) 185 { 186 struct sock *sk, *result; 187 int score, badness; 188 189 result = NULL; 190 badness = -1; 191 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 192 score = compute_score(sk, net, saddr, sport, 193 daddr, hnum, dif, sdif); 194 if (score > badness) { 195 result = lookup_reuseport(net, sk, skb, 196 saddr, sport, daddr, hnum); 197 /* Fall back to scoring if group has connections */ 198 if (result && !reuseport_has_conns(sk, false)) 199 return result; 200 201 result = result ? : sk; 202 badness = score; 203 } 204 } 205 return result; 206 } 207 208 static inline struct sock *udp6_lookup_run_bpf(struct net *net, 209 struct udp_table *udptable, 210 struct sk_buff *skb, 211 const struct in6_addr *saddr, 212 __be16 sport, 213 const struct in6_addr *daddr, 214 u16 hnum, const int dif) 215 { 216 struct sock *sk, *reuse_sk; 217 bool no_reuseport; 218 219 if (udptable != &udp_table) 220 return NULL; /* only UDP is supported */ 221 222 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_UDP, saddr, sport, 223 daddr, hnum, dif, &sk); 224 if (no_reuseport || IS_ERR_OR_NULL(sk)) 225 return sk; 226 227 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 228 if (reuse_sk) 229 sk = reuse_sk; 230 return sk; 231 } 232 233 /* rcu_read_lock() must be held */ 234 struct sock *__udp6_lib_lookup(struct net *net, 235 const struct in6_addr *saddr, __be16 sport, 236 const struct in6_addr *daddr, __be16 dport, 237 int dif, int sdif, struct udp_table *udptable, 238 struct sk_buff *skb) 239 { 240 unsigned short hnum = ntohs(dport); 241 unsigned int hash2, slot2; 242 struct udp_hslot *hslot2; 243 struct sock *result, *sk; 244 245 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 246 slot2 = hash2 & udptable->mask; 247 hslot2 = &udptable->hash2[slot2]; 248 249 /* Lookup connected or non-wildcard sockets */ 250 result = udp6_lib_lookup2(net, saddr, sport, 251 daddr, hnum, dif, sdif, 252 hslot2, skb); 253 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 254 goto done; 255 256 /* Lookup redirect from BPF */ 257 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { 258 sk = udp6_lookup_run_bpf(net, udptable, skb, 259 saddr, sport, daddr, hnum, dif); 260 if (sk) { 261 result = sk; 262 goto done; 263 } 264 } 265 266 /* Got non-wildcard socket or error on first lookup */ 267 if (result) 268 goto done; 269 270 /* Lookup wildcard sockets */ 271 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 272 slot2 = hash2 & udptable->mask; 273 hslot2 = &udptable->hash2[slot2]; 274 275 result = udp6_lib_lookup2(net, saddr, sport, 276 &in6addr_any, hnum, dif, sdif, 277 hslot2, skb); 278 done: 279 if (IS_ERR(result)) 280 return NULL; 281 return result; 282 } 283 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 284 285 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 286 __be16 sport, __be16 dport, 287 struct udp_table *udptable) 288 { 289 const struct ipv6hdr *iph = ipv6_hdr(skb); 290 291 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 292 &iph->daddr, dport, inet6_iif(skb), 293 inet6_sdif(skb), udptable, skb); 294 } 295 296 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, 297 __be16 sport, __be16 dport) 298 { 299 const struct ipv6hdr *iph = ipv6_hdr(skb); 300 301 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 302 &iph->daddr, dport, inet6_iif(skb), 303 inet6_sdif(skb), &udp_table, NULL); 304 } 305 306 /* Must be called under rcu_read_lock(). 307 * Does increment socket refcount. 308 */ 309 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 310 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 311 const struct in6_addr *daddr, __be16 dport, int dif) 312 { 313 struct sock *sk; 314 315 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 316 dif, 0, &udp_table, NULL); 317 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 318 sk = NULL; 319 return sk; 320 } 321 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 322 #endif 323 324 /* do not use the scratch area len for jumbogram: their length execeeds the 325 * scratch area space; note that the IP6CB flags is still in the first 326 * cacheline, so checking for jumbograms is cheap 327 */ 328 static int udp6_skb_len(struct sk_buff *skb) 329 { 330 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 331 } 332 333 /* 334 * This should be easy, if there is something there we 335 * return it, otherwise we block. 336 */ 337 338 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 339 int flags, int *addr_len) 340 { 341 struct ipv6_pinfo *np = inet6_sk(sk); 342 struct inet_sock *inet = inet_sk(sk); 343 struct sk_buff *skb; 344 unsigned int ulen, copied; 345 int off, err, peeking = flags & MSG_PEEK; 346 int is_udplite = IS_UDPLITE(sk); 347 struct udp_mib __percpu *mib; 348 bool checksum_valid = false; 349 int is_udp4; 350 351 if (flags & MSG_ERRQUEUE) 352 return ipv6_recv_error(sk, msg, len, addr_len); 353 354 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 355 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 356 357 try_again: 358 off = sk_peek_offset(sk, flags); 359 skb = __skb_recv_udp(sk, flags, &off, &err); 360 if (!skb) 361 return err; 362 363 ulen = udp6_skb_len(skb); 364 copied = len; 365 if (copied > ulen - off) 366 copied = ulen - off; 367 else if (copied < ulen) 368 msg->msg_flags |= MSG_TRUNC; 369 370 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 371 mib = __UDPX_MIB(sk, is_udp4); 372 373 /* 374 * If checksum is needed at all, try to do it while copying the 375 * data. If the data is truncated, or if we only want a partial 376 * coverage checksum (UDP-Lite), do it before the copy. 377 */ 378 379 if (copied < ulen || peeking || 380 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 381 checksum_valid = udp_skb_csum_unnecessary(skb) || 382 !__udp_lib_checksum_complete(skb); 383 if (!checksum_valid) 384 goto csum_copy_err; 385 } 386 387 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 388 if (udp_skb_is_linear(skb)) 389 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 390 else 391 err = skb_copy_datagram_msg(skb, off, msg, copied); 392 } else { 393 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 394 if (err == -EINVAL) 395 goto csum_copy_err; 396 } 397 if (unlikely(err)) { 398 if (!peeking) { 399 atomic_inc(&sk->sk_drops); 400 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 401 } 402 kfree_skb(skb); 403 return err; 404 } 405 if (!peeking) 406 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 407 408 sock_recv_cmsgs(msg, sk, skb); 409 410 /* Copy the address. */ 411 if (msg->msg_name) { 412 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 413 sin6->sin6_family = AF_INET6; 414 sin6->sin6_port = udp_hdr(skb)->source; 415 sin6->sin6_flowinfo = 0; 416 417 if (is_udp4) { 418 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 419 &sin6->sin6_addr); 420 sin6->sin6_scope_id = 0; 421 } else { 422 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 423 sin6->sin6_scope_id = 424 ipv6_iface_scope_id(&sin6->sin6_addr, 425 inet6_iif(skb)); 426 } 427 *addr_len = sizeof(*sin6); 428 429 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 430 (struct sockaddr *)sin6); 431 } 432 433 if (udp_sk(sk)->gro_enabled) 434 udp_cmsg_recv(msg, sk, skb); 435 436 if (np->rxopt.all) 437 ip6_datagram_recv_common_ctl(sk, msg, skb); 438 439 if (is_udp4) { 440 if (inet->cmsg_flags) 441 ip_cmsg_recv_offset(msg, sk, skb, 442 sizeof(struct udphdr), off); 443 } else { 444 if (np->rxopt.all) 445 ip6_datagram_recv_specific_ctl(sk, msg, skb); 446 } 447 448 err = copied; 449 if (flags & MSG_TRUNC) 450 err = ulen; 451 452 skb_consume_udp(sk, skb, peeking ? -err : err); 453 return err; 454 455 csum_copy_err: 456 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 457 udp_skb_destructor)) { 458 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 459 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 460 } 461 kfree_skb(skb); 462 463 /* starting over for a new packet, but check if we need to yield */ 464 cond_resched(); 465 msg->msg_flags &= ~MSG_TRUNC; 466 goto try_again; 467 } 468 469 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 470 void udpv6_encap_enable(void) 471 { 472 static_branch_inc(&udpv6_encap_needed_key); 473 } 474 EXPORT_SYMBOL(udpv6_encap_enable); 475 476 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 477 * through error handlers in encapsulations looking for a match. 478 */ 479 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 480 struct inet6_skb_parm *opt, 481 u8 type, u8 code, int offset, __be32 info) 482 { 483 int i; 484 485 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 486 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 487 u8 type, u8 code, int offset, __be32 info); 488 const struct ip6_tnl_encap_ops *encap; 489 490 encap = rcu_dereference(ip6tun_encaps[i]); 491 if (!encap) 492 continue; 493 handler = encap->err_handler; 494 if (handler && !handler(skb, opt, type, code, offset, info)) 495 return 0; 496 } 497 498 return -ENOENT; 499 } 500 501 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 502 * reversing source and destination port: this will match tunnels that force the 503 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 504 * lwtunnels might actually break this assumption by being configured with 505 * different destination ports on endpoints, in this case we won't be able to 506 * trace ICMP messages back to them. 507 * 508 * If this doesn't match any socket, probe tunnels with arbitrary destination 509 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 510 * we've sent packets to won't necessarily match the local destination port. 511 * 512 * Then ask the tunnel implementation to match the error against a valid 513 * association. 514 * 515 * Return an error if we can't find a match, the socket if we need further 516 * processing, zero otherwise. 517 */ 518 static struct sock *__udp6_lib_err_encap(struct net *net, 519 const struct ipv6hdr *hdr, int offset, 520 struct udphdr *uh, 521 struct udp_table *udptable, 522 struct sock *sk, 523 struct sk_buff *skb, 524 struct inet6_skb_parm *opt, 525 u8 type, u8 code, __be32 info) 526 { 527 int (*lookup)(struct sock *sk, struct sk_buff *skb); 528 int network_offset, transport_offset; 529 struct udp_sock *up; 530 531 network_offset = skb_network_offset(skb); 532 transport_offset = skb_transport_offset(skb); 533 534 /* Network header needs to point to the outer IPv6 header inside ICMP */ 535 skb_reset_network_header(skb); 536 537 /* Transport header needs to point to the UDP header */ 538 skb_set_transport_header(skb, offset); 539 540 if (sk) { 541 up = udp_sk(sk); 542 543 lookup = READ_ONCE(up->encap_err_lookup); 544 if (lookup && lookup(sk, skb)) 545 sk = NULL; 546 547 goto out; 548 } 549 550 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 551 &hdr->saddr, uh->dest, 552 inet6_iif(skb), 0, udptable, skb); 553 if (sk) { 554 up = udp_sk(sk); 555 556 lookup = READ_ONCE(up->encap_err_lookup); 557 if (!lookup || lookup(sk, skb)) 558 sk = NULL; 559 } 560 561 out: 562 if (!sk) { 563 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 564 offset, info)); 565 } 566 567 skb_set_transport_header(skb, transport_offset); 568 skb_set_network_header(skb, network_offset); 569 570 return sk; 571 } 572 573 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 574 u8 type, u8 code, int offset, __be32 info, 575 struct udp_table *udptable) 576 { 577 struct ipv6_pinfo *np; 578 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 579 const struct in6_addr *saddr = &hdr->saddr; 580 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr; 581 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 582 bool tunnel = false; 583 struct sock *sk; 584 int harderr; 585 int err; 586 struct net *net = dev_net(skb->dev); 587 588 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 589 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 590 591 if (!sk || udp_sk(sk)->encap_type) { 592 /* No socket for error: try tunnels before discarding */ 593 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 594 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 595 udptable, sk, skb, 596 opt, type, code, info); 597 if (!sk) 598 return 0; 599 } else 600 sk = ERR_PTR(-ENOENT); 601 602 if (IS_ERR(sk)) { 603 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 604 ICMP6_MIB_INERRORS); 605 return PTR_ERR(sk); 606 } 607 608 tunnel = true; 609 } 610 611 harderr = icmpv6_err_convert(type, code, &err); 612 np = inet6_sk(sk); 613 614 if (type == ICMPV6_PKT_TOOBIG) { 615 if (!ip6_sk_accept_pmtu(sk)) 616 goto out; 617 ip6_sk_update_pmtu(skb, sk, info); 618 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 619 harderr = 1; 620 } 621 if (type == NDISC_REDIRECT) { 622 if (tunnel) { 623 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 624 sk->sk_mark, sk->sk_uid); 625 } else { 626 ip6_sk_redirect(skb, sk); 627 } 628 goto out; 629 } 630 631 /* Tunnels don't have an application socket: don't pass errors back */ 632 if (tunnel) { 633 if (udp_sk(sk)->encap_err_rcv) 634 udp_sk(sk)->encap_err_rcv(sk, skb, offset); 635 goto out; 636 } 637 638 if (!np->recverr) { 639 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 640 goto out; 641 } else { 642 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 643 } 644 645 sk->sk_err = err; 646 sk_error_report(sk); 647 out: 648 return 0; 649 } 650 651 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 652 { 653 int rc; 654 655 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 656 sock_rps_save_rxhash(sk, skb); 657 sk_mark_napi_id(sk, skb); 658 sk_incoming_cpu_update(sk); 659 } else { 660 sk_mark_napi_id_once(sk, skb); 661 } 662 663 rc = __udp_enqueue_schedule_skb(sk, skb); 664 if (rc < 0) { 665 int is_udplite = IS_UDPLITE(sk); 666 enum skb_drop_reason drop_reason; 667 668 /* Note that an ENOMEM error is charged twice */ 669 if (rc == -ENOMEM) { 670 UDP6_INC_STATS(sock_net(sk), 671 UDP_MIB_RCVBUFERRORS, is_udplite); 672 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 673 } else { 674 UDP6_INC_STATS(sock_net(sk), 675 UDP_MIB_MEMERRORS, is_udplite); 676 drop_reason = SKB_DROP_REASON_PROTO_MEM; 677 } 678 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 679 kfree_skb_reason(skb, drop_reason); 680 return -1; 681 } 682 683 return 0; 684 } 685 686 static __inline__ int udpv6_err(struct sk_buff *skb, 687 struct inet6_skb_parm *opt, u8 type, 688 u8 code, int offset, __be32 info) 689 { 690 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 691 } 692 693 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 694 { 695 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 696 struct udp_sock *up = udp_sk(sk); 697 int is_udplite = IS_UDPLITE(sk); 698 699 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 700 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 701 goto drop; 702 } 703 704 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 705 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 706 707 /* 708 * This is an encapsulation socket so pass the skb to 709 * the socket's udp_encap_rcv() hook. Otherwise, just 710 * fall through and pass this up the UDP socket. 711 * up->encap_rcv() returns the following value: 712 * =0 if skb was successfully passed to the encap 713 * handler or was discarded by it. 714 * >0 if skb should be passed on to UDP. 715 * <0 if skb should be resubmitted as proto -N 716 */ 717 718 /* if we're overly short, let UDP handle it */ 719 encap_rcv = READ_ONCE(up->encap_rcv); 720 if (encap_rcv) { 721 int ret; 722 723 /* Verify checksum before giving to encap */ 724 if (udp_lib_checksum_complete(skb)) 725 goto csum_error; 726 727 ret = encap_rcv(sk, skb); 728 if (ret <= 0) { 729 __UDP6_INC_STATS(sock_net(sk), 730 UDP_MIB_INDATAGRAMS, 731 is_udplite); 732 return -ret; 733 } 734 } 735 736 /* FALLTHROUGH -- it's a UDP Packet */ 737 } 738 739 /* 740 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 741 */ 742 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 743 744 if (up->pcrlen == 0) { /* full coverage was set */ 745 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 746 UDP_SKB_CB(skb)->cscov, skb->len); 747 goto drop; 748 } 749 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 750 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 751 UDP_SKB_CB(skb)->cscov, up->pcrlen); 752 goto drop; 753 } 754 } 755 756 prefetch(&sk->sk_rmem_alloc); 757 if (rcu_access_pointer(sk->sk_filter) && 758 udp_lib_checksum_complete(skb)) 759 goto csum_error; 760 761 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 762 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 763 goto drop; 764 } 765 766 udp_csum_pull_header(skb); 767 768 skb_dst_drop(skb); 769 770 return __udpv6_queue_rcv_skb(sk, skb); 771 772 csum_error: 773 drop_reason = SKB_DROP_REASON_UDP_CSUM; 774 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 775 drop: 776 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 777 atomic_inc(&sk->sk_drops); 778 kfree_skb_reason(skb, drop_reason); 779 return -1; 780 } 781 782 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 783 { 784 struct sk_buff *next, *segs; 785 int ret; 786 787 if (likely(!udp_unexpected_gso(sk, skb))) 788 return udpv6_queue_rcv_one_skb(sk, skb); 789 790 __skb_push(skb, -skb_mac_offset(skb)); 791 segs = udp_rcv_segment(sk, skb, false); 792 skb_list_walk_safe(segs, skb, next) { 793 __skb_pull(skb, skb_transport_offset(skb)); 794 795 udp_post_segment_fix_csum(skb); 796 ret = udpv6_queue_rcv_one_skb(sk, skb); 797 if (ret > 0) 798 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 799 true); 800 } 801 return 0; 802 } 803 804 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 805 __be16 loc_port, const struct in6_addr *loc_addr, 806 __be16 rmt_port, const struct in6_addr *rmt_addr, 807 int dif, int sdif, unsigned short hnum) 808 { 809 struct inet_sock *inet = inet_sk(sk); 810 811 if (!net_eq(sock_net(sk), net)) 812 return false; 813 814 if (udp_sk(sk)->udp_port_hash != hnum || 815 sk->sk_family != PF_INET6 || 816 (inet->inet_dport && inet->inet_dport != rmt_port) || 817 (!ipv6_addr_any(&sk->sk_v6_daddr) && 818 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 819 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) || 820 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 821 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 822 return false; 823 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 824 return false; 825 return true; 826 } 827 828 static void udp6_csum_zero_error(struct sk_buff *skb) 829 { 830 /* RFC 2460 section 8.1 says that we SHOULD log 831 * this error. Well, it is reasonable. 832 */ 833 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 834 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 835 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 836 } 837 838 /* 839 * Note: called only from the BH handler context, 840 * so we don't need to lock the hashes. 841 */ 842 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 843 const struct in6_addr *saddr, const struct in6_addr *daddr, 844 struct udp_table *udptable, int proto) 845 { 846 struct sock *sk, *first = NULL; 847 const struct udphdr *uh = udp_hdr(skb); 848 unsigned short hnum = ntohs(uh->dest); 849 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 850 unsigned int offset = offsetof(typeof(*sk), sk_node); 851 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 852 int dif = inet6_iif(skb); 853 int sdif = inet6_sdif(skb); 854 struct hlist_node *node; 855 struct sk_buff *nskb; 856 857 if (use_hash2) { 858 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 859 udptable->mask; 860 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 861 start_lookup: 862 hslot = &udptable->hash2[hash2]; 863 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 864 } 865 866 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 867 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 868 uh->source, saddr, dif, sdif, 869 hnum)) 870 continue; 871 /* If zero checksum and no_check is not on for 872 * the socket then skip it. 873 */ 874 if (!uh->check && !udp_sk(sk)->no_check6_rx) 875 continue; 876 if (!first) { 877 first = sk; 878 continue; 879 } 880 nskb = skb_clone(skb, GFP_ATOMIC); 881 if (unlikely(!nskb)) { 882 atomic_inc(&sk->sk_drops); 883 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 884 IS_UDPLITE(sk)); 885 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 886 IS_UDPLITE(sk)); 887 continue; 888 } 889 890 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 891 consume_skb(nskb); 892 } 893 894 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 895 if (use_hash2 && hash2 != hash2_any) { 896 hash2 = hash2_any; 897 goto start_lookup; 898 } 899 900 if (first) { 901 if (udpv6_queue_rcv_skb(first, skb) > 0) 902 consume_skb(skb); 903 } else { 904 kfree_skb(skb); 905 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 906 proto == IPPROTO_UDPLITE); 907 } 908 return 0; 909 } 910 911 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 912 { 913 if (udp_sk_rx_dst_set(sk, dst)) { 914 const struct rt6_info *rt = (const struct rt6_info *)dst; 915 916 sk->sk_rx_dst_cookie = rt6_get_cookie(rt); 917 } 918 } 919 920 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 921 * return code conversion for ip layer consumption 922 */ 923 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 924 struct udphdr *uh) 925 { 926 int ret; 927 928 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 929 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 930 931 ret = udpv6_queue_rcv_skb(sk, skb); 932 933 /* a return value > 0 means to resubmit the input */ 934 if (ret > 0) 935 return ret; 936 return 0; 937 } 938 939 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 940 int proto) 941 { 942 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 943 const struct in6_addr *saddr, *daddr; 944 struct net *net = dev_net(skb->dev); 945 struct udphdr *uh; 946 struct sock *sk; 947 bool refcounted; 948 u32 ulen = 0; 949 950 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 951 goto discard; 952 953 saddr = &ipv6_hdr(skb)->saddr; 954 daddr = &ipv6_hdr(skb)->daddr; 955 uh = udp_hdr(skb); 956 957 ulen = ntohs(uh->len); 958 if (ulen > skb->len) 959 goto short_packet; 960 961 if (proto == IPPROTO_UDP) { 962 /* UDP validates ulen. */ 963 964 /* Check for jumbo payload */ 965 if (ulen == 0) 966 ulen = skb->len; 967 968 if (ulen < sizeof(*uh)) 969 goto short_packet; 970 971 if (ulen < skb->len) { 972 if (pskb_trim_rcsum(skb, ulen)) 973 goto short_packet; 974 saddr = &ipv6_hdr(skb)->saddr; 975 daddr = &ipv6_hdr(skb)->daddr; 976 uh = udp_hdr(skb); 977 } 978 } 979 980 if (udp6_csum_init(skb, uh, proto)) 981 goto csum_error; 982 983 /* Check if the socket is already available, e.g. due to early demux */ 984 sk = skb_steal_sock(skb, &refcounted); 985 if (sk) { 986 struct dst_entry *dst = skb_dst(skb); 987 int ret; 988 989 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 990 udp6_sk_rx_dst_set(sk, dst); 991 992 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 993 if (refcounted) 994 sock_put(sk); 995 goto report_csum_error; 996 } 997 998 ret = udp6_unicast_rcv_skb(sk, skb, uh); 999 if (refcounted) 1000 sock_put(sk); 1001 return ret; 1002 } 1003 1004 /* 1005 * Multicast receive code 1006 */ 1007 if (ipv6_addr_is_multicast(daddr)) 1008 return __udp6_lib_mcast_deliver(net, skb, 1009 saddr, daddr, udptable, proto); 1010 1011 /* Unicast */ 1012 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1013 if (sk) { 1014 if (!uh->check && !udp_sk(sk)->no_check6_rx) 1015 goto report_csum_error; 1016 return udp6_unicast_rcv_skb(sk, skb, uh); 1017 } 1018 1019 reason = SKB_DROP_REASON_NO_SOCKET; 1020 1021 if (!uh->check) 1022 goto report_csum_error; 1023 1024 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1025 goto discard; 1026 1027 if (udp_lib_checksum_complete(skb)) 1028 goto csum_error; 1029 1030 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1031 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 1032 1033 kfree_skb_reason(skb, reason); 1034 return 0; 1035 1036 short_packet: 1037 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1038 reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1039 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 1040 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1041 saddr, ntohs(uh->source), 1042 ulen, skb->len, 1043 daddr, ntohs(uh->dest)); 1044 goto discard; 1045 1046 report_csum_error: 1047 udp6_csum_zero_error(skb); 1048 csum_error: 1049 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1050 reason = SKB_DROP_REASON_UDP_CSUM; 1051 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1052 discard: 1053 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1054 kfree_skb_reason(skb, reason); 1055 return 0; 1056 } 1057 1058 1059 static struct sock *__udp6_lib_demux_lookup(struct net *net, 1060 __be16 loc_port, const struct in6_addr *loc_addr, 1061 __be16 rmt_port, const struct in6_addr *rmt_addr, 1062 int dif, int sdif) 1063 { 1064 unsigned short hnum = ntohs(loc_port); 1065 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1066 unsigned int slot2 = hash2 & udp_table.mask; 1067 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1068 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1069 struct sock *sk; 1070 1071 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1072 if (sk->sk_state == TCP_ESTABLISHED && 1073 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif)) 1074 return sk; 1075 /* Only check first socket in chain */ 1076 break; 1077 } 1078 return NULL; 1079 } 1080 1081 void udp_v6_early_demux(struct sk_buff *skb) 1082 { 1083 struct net *net = dev_net(skb->dev); 1084 const struct udphdr *uh; 1085 struct sock *sk; 1086 struct dst_entry *dst; 1087 int dif = skb->dev->ifindex; 1088 int sdif = inet6_sdif(skb); 1089 1090 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1091 sizeof(struct udphdr))) 1092 return; 1093 1094 uh = udp_hdr(skb); 1095 1096 if (skb->pkt_type == PACKET_HOST) 1097 sk = __udp6_lib_demux_lookup(net, uh->dest, 1098 &ipv6_hdr(skb)->daddr, 1099 uh->source, &ipv6_hdr(skb)->saddr, 1100 dif, sdif); 1101 else 1102 return; 1103 1104 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1105 return; 1106 1107 skb->sk = sk; 1108 skb->destructor = sock_efree; 1109 dst = rcu_dereference(sk->sk_rx_dst); 1110 1111 if (dst) 1112 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1113 if (dst) { 1114 /* set noref for now. 1115 * any place which wants to hold dst has to call 1116 * dst_hold_safe() 1117 */ 1118 skb_dst_set_noref(skb, dst); 1119 } 1120 } 1121 1122 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1123 { 1124 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1125 } 1126 1127 /* 1128 * Throw away all pending data and cancel the corking. Socket is locked. 1129 */ 1130 static void udp_v6_flush_pending_frames(struct sock *sk) 1131 { 1132 struct udp_sock *up = udp_sk(sk); 1133 1134 if (up->pending == AF_INET) 1135 udp_flush_pending_frames(sk); 1136 else if (up->pending) { 1137 up->len = 0; 1138 up->pending = 0; 1139 ip6_flush_pending_frames(sk); 1140 } 1141 } 1142 1143 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1144 int addr_len) 1145 { 1146 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1147 return -EINVAL; 1148 /* The following checks are replicated from __ip6_datagram_connect() 1149 * and intended to prevent BPF program called below from accessing 1150 * bytes that are out of the bound specified by user in addr_len. 1151 */ 1152 if (uaddr->sa_family == AF_INET) { 1153 if (ipv6_only_sock(sk)) 1154 return -EAFNOSUPPORT; 1155 return udp_pre_connect(sk, uaddr, addr_len); 1156 } 1157 1158 if (addr_len < SIN6_LEN_RFC2133) 1159 return -EINVAL; 1160 1161 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1162 } 1163 1164 /** 1165 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1166 * @sk: socket we are sending on 1167 * @skb: sk_buff containing the filled-in UDP header 1168 * (checksum field must be zeroed out) 1169 * @saddr: source address 1170 * @daddr: destination address 1171 * @len: length of packet 1172 */ 1173 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1174 const struct in6_addr *saddr, 1175 const struct in6_addr *daddr, int len) 1176 { 1177 unsigned int offset; 1178 struct udphdr *uh = udp_hdr(skb); 1179 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1180 __wsum csum = 0; 1181 1182 if (!frags) { 1183 /* Only one fragment on the socket. */ 1184 skb->csum_start = skb_transport_header(skb) - skb->head; 1185 skb->csum_offset = offsetof(struct udphdr, check); 1186 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1187 } else { 1188 /* 1189 * HW-checksum won't work as there are two or more 1190 * fragments on the socket so that all csums of sk_buffs 1191 * should be together 1192 */ 1193 offset = skb_transport_offset(skb); 1194 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1195 csum = skb->csum; 1196 1197 skb->ip_summed = CHECKSUM_NONE; 1198 1199 do { 1200 csum = csum_add(csum, frags->csum); 1201 } while ((frags = frags->next)); 1202 1203 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1204 csum); 1205 if (uh->check == 0) 1206 uh->check = CSUM_MANGLED_0; 1207 } 1208 } 1209 1210 /* 1211 * Sending 1212 */ 1213 1214 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1215 struct inet_cork *cork) 1216 { 1217 struct sock *sk = skb->sk; 1218 struct udphdr *uh; 1219 int err = 0; 1220 int is_udplite = IS_UDPLITE(sk); 1221 __wsum csum = 0; 1222 int offset = skb_transport_offset(skb); 1223 int len = skb->len - offset; 1224 int datalen = len - sizeof(*uh); 1225 1226 /* 1227 * Create a UDP header 1228 */ 1229 uh = udp_hdr(skb); 1230 uh->source = fl6->fl6_sport; 1231 uh->dest = fl6->fl6_dport; 1232 uh->len = htons(len); 1233 uh->check = 0; 1234 1235 if (cork->gso_size) { 1236 const int hlen = skb_network_header_len(skb) + 1237 sizeof(struct udphdr); 1238 1239 if (hlen + cork->gso_size > cork->fragsize) { 1240 kfree_skb(skb); 1241 return -EINVAL; 1242 } 1243 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1244 kfree_skb(skb); 1245 return -EINVAL; 1246 } 1247 if (udp_sk(sk)->no_check6_tx) { 1248 kfree_skb(skb); 1249 return -EINVAL; 1250 } 1251 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1252 dst_xfrm(skb_dst(skb))) { 1253 kfree_skb(skb); 1254 return -EIO; 1255 } 1256 1257 if (datalen > cork->gso_size) { 1258 skb_shinfo(skb)->gso_size = cork->gso_size; 1259 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1260 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1261 cork->gso_size); 1262 } 1263 goto csum_partial; 1264 } 1265 1266 if (is_udplite) 1267 csum = udplite_csum(skb); 1268 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1269 skb->ip_summed = CHECKSUM_NONE; 1270 goto send; 1271 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1272 csum_partial: 1273 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1274 goto send; 1275 } else 1276 csum = udp_csum(skb); 1277 1278 /* add protocol-dependent pseudo-header */ 1279 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1280 len, fl6->flowi6_proto, csum); 1281 if (uh->check == 0) 1282 uh->check = CSUM_MANGLED_0; 1283 1284 send: 1285 err = ip6_send_skb(skb); 1286 if (err) { 1287 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1288 UDP6_INC_STATS(sock_net(sk), 1289 UDP_MIB_SNDBUFERRORS, is_udplite); 1290 err = 0; 1291 } 1292 } else { 1293 UDP6_INC_STATS(sock_net(sk), 1294 UDP_MIB_OUTDATAGRAMS, is_udplite); 1295 } 1296 return err; 1297 } 1298 1299 static int udp_v6_push_pending_frames(struct sock *sk) 1300 { 1301 struct sk_buff *skb; 1302 struct udp_sock *up = udp_sk(sk); 1303 int err = 0; 1304 1305 if (up->pending == AF_INET) 1306 return udp_push_pending_frames(sk); 1307 1308 skb = ip6_finish_skb(sk); 1309 if (!skb) 1310 goto out; 1311 1312 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6, 1313 &inet_sk(sk)->cork.base); 1314 out: 1315 up->len = 0; 1316 up->pending = 0; 1317 return err; 1318 } 1319 1320 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1321 { 1322 struct ipv6_txoptions opt_space; 1323 struct udp_sock *up = udp_sk(sk); 1324 struct inet_sock *inet = inet_sk(sk); 1325 struct ipv6_pinfo *np = inet6_sk(sk); 1326 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1327 struct in6_addr *daddr, *final_p, final; 1328 struct ipv6_txoptions *opt = NULL; 1329 struct ipv6_txoptions *opt_to_free = NULL; 1330 struct ip6_flowlabel *flowlabel = NULL; 1331 struct inet_cork_full cork; 1332 struct flowi6 *fl6 = &cork.fl.u.ip6; 1333 struct dst_entry *dst; 1334 struct ipcm6_cookie ipc6; 1335 int addr_len = msg->msg_namelen; 1336 bool connected = false; 1337 int ulen = len; 1338 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1339 int err; 1340 int is_udplite = IS_UDPLITE(sk); 1341 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1342 1343 ipcm6_init(&ipc6); 1344 ipc6.gso_size = READ_ONCE(up->gso_size); 1345 ipc6.sockc.tsflags = sk->sk_tsflags; 1346 ipc6.sockc.mark = sk->sk_mark; 1347 1348 /* destination address check */ 1349 if (sin6) { 1350 if (addr_len < offsetof(struct sockaddr, sa_data)) 1351 return -EINVAL; 1352 1353 switch (sin6->sin6_family) { 1354 case AF_INET6: 1355 if (addr_len < SIN6_LEN_RFC2133) 1356 return -EINVAL; 1357 daddr = &sin6->sin6_addr; 1358 if (ipv6_addr_any(daddr) && 1359 ipv6_addr_v4mapped(&np->saddr)) 1360 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1361 daddr); 1362 break; 1363 case AF_INET: 1364 goto do_udp_sendmsg; 1365 case AF_UNSPEC: 1366 msg->msg_name = sin6 = NULL; 1367 msg->msg_namelen = addr_len = 0; 1368 daddr = NULL; 1369 break; 1370 default: 1371 return -EINVAL; 1372 } 1373 } else if (!up->pending) { 1374 if (sk->sk_state != TCP_ESTABLISHED) 1375 return -EDESTADDRREQ; 1376 daddr = &sk->sk_v6_daddr; 1377 } else 1378 daddr = NULL; 1379 1380 if (daddr) { 1381 if (ipv6_addr_v4mapped(daddr)) { 1382 struct sockaddr_in sin; 1383 sin.sin_family = AF_INET; 1384 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1385 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1386 msg->msg_name = &sin; 1387 msg->msg_namelen = sizeof(sin); 1388 do_udp_sendmsg: 1389 if (ipv6_only_sock(sk)) 1390 return -ENETUNREACH; 1391 return udp_sendmsg(sk, msg, len); 1392 } 1393 } 1394 1395 /* Rough check on arithmetic overflow, 1396 better check is made in ip6_append_data(). 1397 */ 1398 if (len > INT_MAX - sizeof(struct udphdr)) 1399 return -EMSGSIZE; 1400 1401 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1402 if (up->pending) { 1403 if (up->pending == AF_INET) 1404 return udp_sendmsg(sk, msg, len); 1405 /* 1406 * There are pending frames. 1407 * The socket lock must be held while it's corked. 1408 */ 1409 lock_sock(sk); 1410 if (likely(up->pending)) { 1411 if (unlikely(up->pending != AF_INET6)) { 1412 release_sock(sk); 1413 return -EAFNOSUPPORT; 1414 } 1415 dst = NULL; 1416 goto do_append_data; 1417 } 1418 release_sock(sk); 1419 } 1420 ulen += sizeof(struct udphdr); 1421 1422 memset(fl6, 0, sizeof(*fl6)); 1423 1424 if (sin6) { 1425 if (sin6->sin6_port == 0) 1426 return -EINVAL; 1427 1428 fl6->fl6_dport = sin6->sin6_port; 1429 daddr = &sin6->sin6_addr; 1430 1431 if (np->sndflow) { 1432 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1433 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) { 1434 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1435 if (IS_ERR(flowlabel)) 1436 return -EINVAL; 1437 } 1438 } 1439 1440 /* 1441 * Otherwise it will be difficult to maintain 1442 * sk->sk_dst_cache. 1443 */ 1444 if (sk->sk_state == TCP_ESTABLISHED && 1445 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1446 daddr = &sk->sk_v6_daddr; 1447 1448 if (addr_len >= sizeof(struct sockaddr_in6) && 1449 sin6->sin6_scope_id && 1450 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1451 fl6->flowi6_oif = sin6->sin6_scope_id; 1452 } else { 1453 if (sk->sk_state != TCP_ESTABLISHED) 1454 return -EDESTADDRREQ; 1455 1456 fl6->fl6_dport = inet->inet_dport; 1457 daddr = &sk->sk_v6_daddr; 1458 fl6->flowlabel = np->flow_label; 1459 connected = true; 1460 } 1461 1462 if (!fl6->flowi6_oif) 1463 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if); 1464 1465 if (!fl6->flowi6_oif) 1466 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1467 1468 fl6->flowi6_uid = sk->sk_uid; 1469 1470 if (msg->msg_controllen) { 1471 opt = &opt_space; 1472 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1473 opt->tot_len = sizeof(*opt); 1474 ipc6.opt = opt; 1475 1476 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1477 if (err > 0) 1478 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6, 1479 &ipc6); 1480 if (err < 0) { 1481 fl6_sock_release(flowlabel); 1482 return err; 1483 } 1484 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1485 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1486 if (IS_ERR(flowlabel)) 1487 return -EINVAL; 1488 } 1489 if (!(opt->opt_nflen|opt->opt_flen)) 1490 opt = NULL; 1491 connected = false; 1492 } 1493 if (!opt) { 1494 opt = txopt_get(np); 1495 opt_to_free = opt; 1496 } 1497 if (flowlabel) 1498 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1499 opt = ipv6_fixup_options(&opt_space, opt); 1500 ipc6.opt = opt; 1501 1502 fl6->flowi6_proto = sk->sk_protocol; 1503 fl6->flowi6_mark = ipc6.sockc.mark; 1504 fl6->daddr = *daddr; 1505 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr)) 1506 fl6->saddr = np->saddr; 1507 fl6->fl6_sport = inet->inet_sport; 1508 1509 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1510 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1511 (struct sockaddr *)sin6, 1512 &fl6->saddr); 1513 if (err) 1514 goto out_no_dst; 1515 if (sin6) { 1516 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1517 /* BPF program rewrote IPv6-only by IPv4-mapped 1518 * IPv6. It's currently unsupported. 1519 */ 1520 err = -ENOTSUPP; 1521 goto out_no_dst; 1522 } 1523 if (sin6->sin6_port == 0) { 1524 /* BPF program set invalid port. Reject it. */ 1525 err = -EINVAL; 1526 goto out_no_dst; 1527 } 1528 fl6->fl6_dport = sin6->sin6_port; 1529 fl6->daddr = sin6->sin6_addr; 1530 } 1531 } 1532 1533 if (ipv6_addr_any(&fl6->daddr)) 1534 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1535 1536 final_p = fl6_update_dst(fl6, opt, &final); 1537 if (final_p) 1538 connected = false; 1539 1540 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) { 1541 fl6->flowi6_oif = np->mcast_oif; 1542 connected = false; 1543 } else if (!fl6->flowi6_oif) 1544 fl6->flowi6_oif = np->ucast_oif; 1545 1546 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6)); 1547 1548 if (ipc6.tclass < 0) 1549 ipc6.tclass = np->tclass; 1550 1551 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel); 1552 1553 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected); 1554 if (IS_ERR(dst)) { 1555 err = PTR_ERR(dst); 1556 dst = NULL; 1557 goto out; 1558 } 1559 1560 if (ipc6.hlimit < 0) 1561 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst); 1562 1563 if (msg->msg_flags&MSG_CONFIRM) 1564 goto do_confirm; 1565 back_from_confirm: 1566 1567 /* Lockless fast path for the non-corking case */ 1568 if (!corkreq) { 1569 struct sk_buff *skb; 1570 1571 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1572 sizeof(struct udphdr), &ipc6, 1573 (struct rt6_info *)dst, 1574 msg->msg_flags, &cork); 1575 err = PTR_ERR(skb); 1576 if (!IS_ERR_OR_NULL(skb)) 1577 err = udp_v6_send_skb(skb, fl6, &cork.base); 1578 /* ip6_make_skb steals dst reference */ 1579 goto out_no_dst; 1580 } 1581 1582 lock_sock(sk); 1583 if (unlikely(up->pending)) { 1584 /* The socket is already corked while preparing it. */ 1585 /* ... which is an evident application bug. --ANK */ 1586 release_sock(sk); 1587 1588 net_dbg_ratelimited("udp cork app bug 2\n"); 1589 err = -EINVAL; 1590 goto out; 1591 } 1592 1593 up->pending = AF_INET6; 1594 1595 do_append_data: 1596 if (ipc6.dontfrag < 0) 1597 ipc6.dontfrag = np->dontfrag; 1598 up->len += ulen; 1599 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1600 &ipc6, fl6, (struct rt6_info *)dst, 1601 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1602 if (err) 1603 udp_v6_flush_pending_frames(sk); 1604 else if (!corkreq) 1605 err = udp_v6_push_pending_frames(sk); 1606 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1607 up->pending = 0; 1608 1609 if (err > 0) 1610 err = np->recverr ? net_xmit_errno(err) : 0; 1611 release_sock(sk); 1612 1613 out: 1614 dst_release(dst); 1615 out_no_dst: 1616 fl6_sock_release(flowlabel); 1617 txopt_put(opt_to_free); 1618 if (!err) 1619 return len; 1620 /* 1621 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1622 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1623 * we don't have a good statistic (IpOutDiscards but it can be too many 1624 * things). We could add another new stat but at least for now that 1625 * seems like overkill. 1626 */ 1627 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1628 UDP6_INC_STATS(sock_net(sk), 1629 UDP_MIB_SNDBUFERRORS, is_udplite); 1630 } 1631 return err; 1632 1633 do_confirm: 1634 if (msg->msg_flags & MSG_PROBE) 1635 dst_confirm_neigh(dst, &fl6->daddr); 1636 if (!(msg->msg_flags&MSG_PROBE) || len) 1637 goto back_from_confirm; 1638 err = 0; 1639 goto out; 1640 } 1641 1642 void udpv6_destroy_sock(struct sock *sk) 1643 { 1644 struct udp_sock *up = udp_sk(sk); 1645 lock_sock(sk); 1646 1647 /* protects from races with udp_abort() */ 1648 sock_set_flag(sk, SOCK_DEAD); 1649 udp_v6_flush_pending_frames(sk); 1650 release_sock(sk); 1651 1652 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1653 if (up->encap_type) { 1654 void (*encap_destroy)(struct sock *sk); 1655 encap_destroy = READ_ONCE(up->encap_destroy); 1656 if (encap_destroy) 1657 encap_destroy(sk); 1658 } 1659 if (up->encap_enabled) { 1660 static_branch_dec(&udpv6_encap_needed_key); 1661 udp_encap_disable(); 1662 } 1663 } 1664 1665 inet6_destroy_sock(sk); 1666 } 1667 1668 /* 1669 * Socket option code for UDP 1670 */ 1671 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1672 unsigned int optlen) 1673 { 1674 if (level == SOL_UDP || level == SOL_UDPLITE) 1675 return udp_lib_setsockopt(sk, level, optname, 1676 optval, optlen, 1677 udp_v6_push_pending_frames); 1678 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1679 } 1680 1681 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1682 char __user *optval, int __user *optlen) 1683 { 1684 if (level == SOL_UDP || level == SOL_UDPLITE) 1685 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1686 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1687 } 1688 1689 static const struct inet6_protocol udpv6_protocol = { 1690 .handler = udpv6_rcv, 1691 .err_handler = udpv6_err, 1692 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1693 }; 1694 1695 /* ------------------------------------------------------------------------ */ 1696 #ifdef CONFIG_PROC_FS 1697 int udp6_seq_show(struct seq_file *seq, void *v) 1698 { 1699 if (v == SEQ_START_TOKEN) { 1700 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1701 } else { 1702 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1703 struct inet_sock *inet = inet_sk(v); 1704 __u16 srcp = ntohs(inet->inet_sport); 1705 __u16 destp = ntohs(inet->inet_dport); 1706 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1707 udp_rqueue_get(v), bucket); 1708 } 1709 return 0; 1710 } 1711 1712 const struct seq_operations udp6_seq_ops = { 1713 .start = udp_seq_start, 1714 .next = udp_seq_next, 1715 .stop = udp_seq_stop, 1716 .show = udp6_seq_show, 1717 }; 1718 EXPORT_SYMBOL(udp6_seq_ops); 1719 1720 static struct udp_seq_afinfo udp6_seq_afinfo = { 1721 .family = AF_INET6, 1722 .udp_table = &udp_table, 1723 }; 1724 1725 int __net_init udp6_proc_init(struct net *net) 1726 { 1727 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1728 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1729 return -ENOMEM; 1730 return 0; 1731 } 1732 1733 void udp6_proc_exit(struct net *net) 1734 { 1735 remove_proc_entry("udp6", net->proc_net); 1736 } 1737 #endif /* CONFIG_PROC_FS */ 1738 1739 /* ------------------------------------------------------------------------ */ 1740 1741 struct proto udpv6_prot = { 1742 .name = "UDPv6", 1743 .owner = THIS_MODULE, 1744 .close = udp_lib_close, 1745 .pre_connect = udpv6_pre_connect, 1746 .connect = ip6_datagram_connect, 1747 .disconnect = udp_disconnect, 1748 .ioctl = udp_ioctl, 1749 .init = udpv6_init_sock, 1750 .destroy = udpv6_destroy_sock, 1751 .setsockopt = udpv6_setsockopt, 1752 .getsockopt = udpv6_getsockopt, 1753 .sendmsg = udpv6_sendmsg, 1754 .recvmsg = udpv6_recvmsg, 1755 .release_cb = ip6_datagram_release_cb, 1756 .hash = udp_lib_hash, 1757 .unhash = udp_lib_unhash, 1758 .rehash = udp_v6_rehash, 1759 .get_port = udp_v6_get_port, 1760 .put_port = udp_lib_unhash, 1761 #ifdef CONFIG_BPF_SYSCALL 1762 .psock_update_sk_prot = udp_bpf_update_proto, 1763 #endif 1764 1765 .memory_allocated = &udp_memory_allocated, 1766 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 1767 1768 .sysctl_mem = sysctl_udp_mem, 1769 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1770 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1771 .obj_size = sizeof(struct udp6_sock), 1772 .h.udp_table = &udp_table, 1773 .diag_destroy = udp_abort, 1774 }; 1775 1776 static struct inet_protosw udpv6_protosw = { 1777 .type = SOCK_DGRAM, 1778 .protocol = IPPROTO_UDP, 1779 .prot = &udpv6_prot, 1780 .ops = &inet6_dgram_ops, 1781 .flags = INET_PROTOSW_PERMANENT, 1782 }; 1783 1784 int __init udpv6_init(void) 1785 { 1786 int ret; 1787 1788 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1789 if (ret) 1790 goto out; 1791 1792 ret = inet6_register_protosw(&udpv6_protosw); 1793 if (ret) 1794 goto out_udpv6_protocol; 1795 out: 1796 return ret; 1797 1798 out_udpv6_protocol: 1799 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1800 goto out; 1801 } 1802 1803 void udpv6_exit(void) 1804 { 1805 inet6_unregister_protosw(&udpv6_protosw); 1806 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1807 } 1808