1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * UDP over IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on linux/ipv4/udp.c 10 * 11 * Fixes: 12 * Hideaki YOSHIFUJI : sin6_scope_id support 13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 15 * a single port at the same time. 16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. 18 */ 19 20 #include <linux/errno.h> 21 #include <linux/types.h> 22 #include <linux/socket.h> 23 #include <linux/sockios.h> 24 #include <linux/net.h> 25 #include <linux/in6.h> 26 #include <linux/netdevice.h> 27 #include <linux/if_arp.h> 28 #include <linux/ipv6.h> 29 #include <linux/icmpv6.h> 30 #include <linux/init.h> 31 #include <linux/module.h> 32 #include <linux/skbuff.h> 33 #include <linux/slab.h> 34 #include <linux/uaccess.h> 35 #include <linux/indirect_call_wrapper.h> 36 37 #include <net/addrconf.h> 38 #include <net/ndisc.h> 39 #include <net/protocol.h> 40 #include <net/transp_v6.h> 41 #include <net/ip6_route.h> 42 #include <net/raw.h> 43 #include <net/tcp_states.h> 44 #include <net/ip6_checksum.h> 45 #include <net/ip6_tunnel.h> 46 #include <net/xfrm.h> 47 #include <net/inet_hashtables.h> 48 #include <net/inet6_hashtables.h> 49 #include <net/busy_poll.h> 50 #include <net/sock_reuseport.h> 51 52 #include <linux/proc_fs.h> 53 #include <linux/seq_file.h> 54 #include <trace/events/skb.h> 55 #include "udp_impl.h" 56 57 static u32 udp6_ehashfn(const struct net *net, 58 const struct in6_addr *laddr, 59 const u16 lport, 60 const struct in6_addr *faddr, 61 const __be16 fport) 62 { 63 static u32 udp6_ehash_secret __read_mostly; 64 static u32 udp_ipv6_hash_secret __read_mostly; 65 66 u32 lhash, fhash; 67 68 net_get_random_once(&udp6_ehash_secret, 69 sizeof(udp6_ehash_secret)); 70 net_get_random_once(&udp_ipv6_hash_secret, 71 sizeof(udp_ipv6_hash_secret)); 72 73 lhash = (__force u32)laddr->s6_addr32[3]; 74 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 75 76 return __inet6_ehashfn(lhash, lport, fhash, fport, 77 udp_ipv6_hash_secret + net_hash_mix(net)); 78 } 79 80 int udp_v6_get_port(struct sock *sk, unsigned short snum) 81 { 82 unsigned int hash2_nulladdr = 83 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 84 unsigned int hash2_partial = 85 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 86 87 /* precompute partial secondary hash */ 88 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 89 return udp_lib_get_port(sk, snum, hash2_nulladdr); 90 } 91 92 void udp_v6_rehash(struct sock *sk) 93 { 94 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 95 &sk->sk_v6_rcv_saddr, 96 inet_sk(sk)->inet_num); 97 98 udp_lib_rehash(sk, new_hash); 99 } 100 101 static int compute_score(struct sock *sk, struct net *net, 102 const struct in6_addr *saddr, __be16 sport, 103 const struct in6_addr *daddr, unsigned short hnum, 104 int dif, int sdif) 105 { 106 int score; 107 struct inet_sock *inet; 108 bool dev_match; 109 110 if (!net_eq(sock_net(sk), net) || 111 udp_sk(sk)->udp_port_hash != hnum || 112 sk->sk_family != PF_INET6) 113 return -1; 114 115 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 116 return -1; 117 118 score = 0; 119 inet = inet_sk(sk); 120 121 if (inet->inet_dport) { 122 if (inet->inet_dport != sport) 123 return -1; 124 score++; 125 } 126 127 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 128 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 129 return -1; 130 score++; 131 } 132 133 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif); 134 if (!dev_match) 135 return -1; 136 score++; 137 138 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 139 score++; 140 141 return score; 142 } 143 144 /* called with rcu_read_lock() */ 145 static struct sock *udp6_lib_lookup2(struct net *net, 146 const struct in6_addr *saddr, __be16 sport, 147 const struct in6_addr *daddr, unsigned int hnum, 148 int dif, int sdif, struct udp_hslot *hslot2, 149 struct sk_buff *skb) 150 { 151 struct sock *sk, *result, *reuseport_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 reuseport_result = NULL; 162 163 if (sk->sk_reuseport && 164 sk->sk_state != TCP_ESTABLISHED) { 165 hash = udp6_ehashfn(net, daddr, hnum, 166 saddr, sport); 167 168 reuseport_result = reuseport_select_sock(sk, hash, skb, 169 sizeof(struct udphdr)); 170 if (reuseport_result && !reuseport_has_conns(sk, false)) 171 return reuseport_result; 172 } 173 174 result = reuseport_result ? : sk; 175 badness = score; 176 } 177 } 178 return result; 179 } 180 181 /* rcu_read_lock() must be held */ 182 struct sock *__udp6_lib_lookup(struct net *net, 183 const struct in6_addr *saddr, __be16 sport, 184 const struct in6_addr *daddr, __be16 dport, 185 int dif, int sdif, struct udp_table *udptable, 186 struct sk_buff *skb) 187 { 188 unsigned short hnum = ntohs(dport); 189 unsigned int hash2, slot2; 190 struct udp_hslot *hslot2; 191 struct sock *result; 192 193 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 194 slot2 = hash2 & udptable->mask; 195 hslot2 = &udptable->hash2[slot2]; 196 197 result = udp6_lib_lookup2(net, saddr, sport, 198 daddr, hnum, dif, sdif, 199 hslot2, skb); 200 if (!result) { 201 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 202 slot2 = hash2 & udptable->mask; 203 204 hslot2 = &udptable->hash2[slot2]; 205 206 result = udp6_lib_lookup2(net, saddr, sport, 207 &in6addr_any, hnum, dif, sdif, 208 hslot2, skb); 209 } 210 if (IS_ERR(result)) 211 return NULL; 212 return result; 213 } 214 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 215 216 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 217 __be16 sport, __be16 dport, 218 struct udp_table *udptable) 219 { 220 const struct ipv6hdr *iph = ipv6_hdr(skb); 221 222 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 223 &iph->daddr, dport, inet6_iif(skb), 224 inet6_sdif(skb), udptable, skb); 225 } 226 227 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, 228 __be16 sport, __be16 dport) 229 { 230 const struct ipv6hdr *iph = ipv6_hdr(skb); 231 232 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 233 &iph->daddr, dport, inet6_iif(skb), 234 inet6_sdif(skb), &udp_table, NULL); 235 } 236 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); 237 238 /* Must be called under rcu_read_lock(). 239 * Does increment socket refcount. 240 */ 241 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 242 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 243 const struct in6_addr *daddr, __be16 dport, int dif) 244 { 245 struct sock *sk; 246 247 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 248 dif, 0, &udp_table, NULL); 249 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 250 sk = NULL; 251 return sk; 252 } 253 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 254 #endif 255 256 /* do not use the scratch area len for jumbogram: their length execeeds the 257 * scratch area space; note that the IP6CB flags is still in the first 258 * cacheline, so checking for jumbograms is cheap 259 */ 260 static int udp6_skb_len(struct sk_buff *skb) 261 { 262 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 263 } 264 265 /* 266 * This should be easy, if there is something there we 267 * return it, otherwise we block. 268 */ 269 270 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 271 int noblock, int flags, int *addr_len) 272 { 273 struct ipv6_pinfo *np = inet6_sk(sk); 274 struct inet_sock *inet = inet_sk(sk); 275 struct sk_buff *skb; 276 unsigned int ulen, copied; 277 int off, err, peeking = flags & MSG_PEEK; 278 int is_udplite = IS_UDPLITE(sk); 279 struct udp_mib __percpu *mib; 280 bool checksum_valid = false; 281 int is_udp4; 282 283 if (flags & MSG_ERRQUEUE) 284 return ipv6_recv_error(sk, msg, len, addr_len); 285 286 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 287 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 288 289 try_again: 290 off = sk_peek_offset(sk, flags); 291 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 292 if (!skb) 293 return err; 294 295 ulen = udp6_skb_len(skb); 296 copied = len; 297 if (copied > ulen - off) 298 copied = ulen - off; 299 else if (copied < ulen) 300 msg->msg_flags |= MSG_TRUNC; 301 302 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 303 mib = __UDPX_MIB(sk, is_udp4); 304 305 /* 306 * If checksum is needed at all, try to do it while copying the 307 * data. If the data is truncated, or if we only want a partial 308 * coverage checksum (UDP-Lite), do it before the copy. 309 */ 310 311 if (copied < ulen || peeking || 312 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 313 checksum_valid = udp_skb_csum_unnecessary(skb) || 314 !__udp_lib_checksum_complete(skb); 315 if (!checksum_valid) 316 goto csum_copy_err; 317 } 318 319 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 320 if (udp_skb_is_linear(skb)) 321 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 322 else 323 err = skb_copy_datagram_msg(skb, off, msg, copied); 324 } else { 325 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 326 if (err == -EINVAL) 327 goto csum_copy_err; 328 } 329 if (unlikely(err)) { 330 if (!peeking) { 331 atomic_inc(&sk->sk_drops); 332 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 333 } 334 kfree_skb(skb); 335 return err; 336 } 337 if (!peeking) 338 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 339 340 sock_recv_ts_and_drops(msg, sk, skb); 341 342 /* Copy the address. */ 343 if (msg->msg_name) { 344 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 345 sin6->sin6_family = AF_INET6; 346 sin6->sin6_port = udp_hdr(skb)->source; 347 sin6->sin6_flowinfo = 0; 348 349 if (is_udp4) { 350 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 351 &sin6->sin6_addr); 352 sin6->sin6_scope_id = 0; 353 } else { 354 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 355 sin6->sin6_scope_id = 356 ipv6_iface_scope_id(&sin6->sin6_addr, 357 inet6_iif(skb)); 358 } 359 *addr_len = sizeof(*sin6); 360 361 if (cgroup_bpf_enabled) 362 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 363 (struct sockaddr *)sin6); 364 } 365 366 if (udp_sk(sk)->gro_enabled) 367 udp_cmsg_recv(msg, sk, skb); 368 369 if (np->rxopt.all) 370 ip6_datagram_recv_common_ctl(sk, msg, skb); 371 372 if (is_udp4) { 373 if (inet->cmsg_flags) 374 ip_cmsg_recv_offset(msg, sk, skb, 375 sizeof(struct udphdr), off); 376 } else { 377 if (np->rxopt.all) 378 ip6_datagram_recv_specific_ctl(sk, msg, skb); 379 } 380 381 err = copied; 382 if (flags & MSG_TRUNC) 383 err = ulen; 384 385 skb_consume_udp(sk, skb, peeking ? -err : err); 386 return err; 387 388 csum_copy_err: 389 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 390 udp_skb_destructor)) { 391 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 392 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 393 } 394 kfree_skb(skb); 395 396 /* starting over for a new packet, but check if we need to yield */ 397 cond_resched(); 398 msg->msg_flags &= ~MSG_TRUNC; 399 goto try_again; 400 } 401 402 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 403 void udpv6_encap_enable(void) 404 { 405 static_branch_inc(&udpv6_encap_needed_key); 406 } 407 EXPORT_SYMBOL(udpv6_encap_enable); 408 409 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 410 * through error handlers in encapsulations looking for a match. 411 */ 412 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 413 struct inet6_skb_parm *opt, 414 u8 type, u8 code, int offset, __be32 info) 415 { 416 int i; 417 418 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 419 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 420 u8 type, u8 code, int offset, __be32 info); 421 const struct ip6_tnl_encap_ops *encap; 422 423 encap = rcu_dereference(ip6tun_encaps[i]); 424 if (!encap) 425 continue; 426 handler = encap->err_handler; 427 if (handler && !handler(skb, opt, type, code, offset, info)) 428 return 0; 429 } 430 431 return -ENOENT; 432 } 433 434 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 435 * reversing source and destination port: this will match tunnels that force the 436 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 437 * lwtunnels might actually break this assumption by being configured with 438 * different destination ports on endpoints, in this case we won't be able to 439 * trace ICMP messages back to them. 440 * 441 * If this doesn't match any socket, probe tunnels with arbitrary destination 442 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 443 * we've sent packets to won't necessarily match the local destination port. 444 * 445 * Then ask the tunnel implementation to match the error against a valid 446 * association. 447 * 448 * Return an error if we can't find a match, the socket if we need further 449 * processing, zero otherwise. 450 */ 451 static struct sock *__udp6_lib_err_encap(struct net *net, 452 const struct ipv6hdr *hdr, int offset, 453 struct udphdr *uh, 454 struct udp_table *udptable, 455 struct sk_buff *skb, 456 struct inet6_skb_parm *opt, 457 u8 type, u8 code, __be32 info) 458 { 459 int network_offset, transport_offset; 460 struct sock *sk; 461 462 network_offset = skb_network_offset(skb); 463 transport_offset = skb_transport_offset(skb); 464 465 /* Network header needs to point to the outer IPv6 header inside ICMP */ 466 skb_reset_network_header(skb); 467 468 /* Transport header needs to point to the UDP header */ 469 skb_set_transport_header(skb, offset); 470 471 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 472 &hdr->saddr, uh->dest, 473 inet6_iif(skb), 0, udptable, skb); 474 if (sk) { 475 int (*lookup)(struct sock *sk, struct sk_buff *skb); 476 struct udp_sock *up = udp_sk(sk); 477 478 lookup = READ_ONCE(up->encap_err_lookup); 479 if (!lookup || lookup(sk, skb)) 480 sk = NULL; 481 } 482 483 if (!sk) { 484 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 485 offset, info)); 486 } 487 488 skb_set_transport_header(skb, transport_offset); 489 skb_set_network_header(skb, network_offset); 490 491 return sk; 492 } 493 494 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 495 u8 type, u8 code, int offset, __be32 info, 496 struct udp_table *udptable) 497 { 498 struct ipv6_pinfo *np; 499 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 500 const struct in6_addr *saddr = &hdr->saddr; 501 const struct in6_addr *daddr = &hdr->daddr; 502 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 503 bool tunnel = false; 504 struct sock *sk; 505 int harderr; 506 int err; 507 struct net *net = dev_net(skb->dev); 508 509 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 510 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 511 if (!sk) { 512 /* No socket for error: try tunnels before discarding */ 513 sk = ERR_PTR(-ENOENT); 514 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 515 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 516 udptable, skb, 517 opt, type, code, info); 518 if (!sk) 519 return 0; 520 } 521 522 if (IS_ERR(sk)) { 523 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 524 ICMP6_MIB_INERRORS); 525 return PTR_ERR(sk); 526 } 527 528 tunnel = true; 529 } 530 531 harderr = icmpv6_err_convert(type, code, &err); 532 np = inet6_sk(sk); 533 534 if (type == ICMPV6_PKT_TOOBIG) { 535 if (!ip6_sk_accept_pmtu(sk)) 536 goto out; 537 ip6_sk_update_pmtu(skb, sk, info); 538 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 539 harderr = 1; 540 } 541 if (type == NDISC_REDIRECT) { 542 if (tunnel) { 543 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 544 sk->sk_mark, sk->sk_uid); 545 } else { 546 ip6_sk_redirect(skb, sk); 547 } 548 goto out; 549 } 550 551 /* Tunnels don't have an application socket: don't pass errors back */ 552 if (tunnel) 553 goto out; 554 555 if (!np->recverr) { 556 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 557 goto out; 558 } else { 559 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 560 } 561 562 sk->sk_err = err; 563 sk->sk_error_report(sk); 564 out: 565 return 0; 566 } 567 568 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 569 { 570 int rc; 571 572 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 573 sock_rps_save_rxhash(sk, skb); 574 sk_mark_napi_id(sk, skb); 575 sk_incoming_cpu_update(sk); 576 } else { 577 sk_mark_napi_id_once(sk, skb); 578 } 579 580 rc = __udp_enqueue_schedule_skb(sk, skb); 581 if (rc < 0) { 582 int is_udplite = IS_UDPLITE(sk); 583 584 /* Note that an ENOMEM error is charged twice */ 585 if (rc == -ENOMEM) 586 UDP6_INC_STATS(sock_net(sk), 587 UDP_MIB_RCVBUFERRORS, is_udplite); 588 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 589 kfree_skb(skb); 590 return -1; 591 } 592 593 return 0; 594 } 595 596 static __inline__ int udpv6_err(struct sk_buff *skb, 597 struct inet6_skb_parm *opt, u8 type, 598 u8 code, int offset, __be32 info) 599 { 600 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 601 } 602 603 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 604 { 605 struct udp_sock *up = udp_sk(sk); 606 int is_udplite = IS_UDPLITE(sk); 607 608 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 609 goto drop; 610 611 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 612 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 613 614 /* 615 * This is an encapsulation socket so pass the skb to 616 * the socket's udp_encap_rcv() hook. Otherwise, just 617 * fall through and pass this up the UDP socket. 618 * up->encap_rcv() returns the following value: 619 * =0 if skb was successfully passed to the encap 620 * handler or was discarded by it. 621 * >0 if skb should be passed on to UDP. 622 * <0 if skb should be resubmitted as proto -N 623 */ 624 625 /* if we're overly short, let UDP handle it */ 626 encap_rcv = READ_ONCE(up->encap_rcv); 627 if (encap_rcv) { 628 int ret; 629 630 /* Verify checksum before giving to encap */ 631 if (udp_lib_checksum_complete(skb)) 632 goto csum_error; 633 634 ret = encap_rcv(sk, skb); 635 if (ret <= 0) { 636 __UDP_INC_STATS(sock_net(sk), 637 UDP_MIB_INDATAGRAMS, 638 is_udplite); 639 return -ret; 640 } 641 } 642 643 /* FALLTHROUGH -- it's a UDP Packet */ 644 } 645 646 /* 647 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 648 */ 649 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 650 651 if (up->pcrlen == 0) { /* full coverage was set */ 652 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 653 UDP_SKB_CB(skb)->cscov, skb->len); 654 goto drop; 655 } 656 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 657 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 658 UDP_SKB_CB(skb)->cscov, up->pcrlen); 659 goto drop; 660 } 661 } 662 663 prefetch(&sk->sk_rmem_alloc); 664 if (rcu_access_pointer(sk->sk_filter) && 665 udp_lib_checksum_complete(skb)) 666 goto csum_error; 667 668 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 669 goto drop; 670 671 udp_csum_pull_header(skb); 672 673 skb_dst_drop(skb); 674 675 return __udpv6_queue_rcv_skb(sk, skb); 676 677 csum_error: 678 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 679 drop: 680 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 681 atomic_inc(&sk->sk_drops); 682 kfree_skb(skb); 683 return -1; 684 } 685 686 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 687 { 688 struct sk_buff *next, *segs; 689 int ret; 690 691 if (likely(!udp_unexpected_gso(sk, skb))) 692 return udpv6_queue_rcv_one_skb(sk, skb); 693 694 __skb_push(skb, -skb_mac_offset(skb)); 695 segs = udp_rcv_segment(sk, skb, false); 696 skb_list_walk_safe(segs, skb, next) { 697 __skb_pull(skb, skb_transport_offset(skb)); 698 699 ret = udpv6_queue_rcv_one_skb(sk, skb); 700 if (ret > 0) 701 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 702 true); 703 } 704 return 0; 705 } 706 707 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 708 __be16 loc_port, const struct in6_addr *loc_addr, 709 __be16 rmt_port, const struct in6_addr *rmt_addr, 710 int dif, int sdif, unsigned short hnum) 711 { 712 struct inet_sock *inet = inet_sk(sk); 713 714 if (!net_eq(sock_net(sk), net)) 715 return false; 716 717 if (udp_sk(sk)->udp_port_hash != hnum || 718 sk->sk_family != PF_INET6 || 719 (inet->inet_dport && inet->inet_dport != rmt_port) || 720 (!ipv6_addr_any(&sk->sk_v6_daddr) && 721 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 722 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) || 723 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 724 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 725 return false; 726 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 727 return false; 728 return true; 729 } 730 731 static void udp6_csum_zero_error(struct sk_buff *skb) 732 { 733 /* RFC 2460 section 8.1 says that we SHOULD log 734 * this error. Well, it is reasonable. 735 */ 736 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 737 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 738 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 739 } 740 741 /* 742 * Note: called only from the BH handler context, 743 * so we don't need to lock the hashes. 744 */ 745 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 746 const struct in6_addr *saddr, const struct in6_addr *daddr, 747 struct udp_table *udptable, int proto) 748 { 749 struct sock *sk, *first = NULL; 750 const struct udphdr *uh = udp_hdr(skb); 751 unsigned short hnum = ntohs(uh->dest); 752 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 753 unsigned int offset = offsetof(typeof(*sk), sk_node); 754 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 755 int dif = inet6_iif(skb); 756 int sdif = inet6_sdif(skb); 757 struct hlist_node *node; 758 struct sk_buff *nskb; 759 760 if (use_hash2) { 761 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 762 udptable->mask; 763 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 764 start_lookup: 765 hslot = &udptable->hash2[hash2]; 766 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 767 } 768 769 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 770 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 771 uh->source, saddr, dif, sdif, 772 hnum)) 773 continue; 774 /* If zero checksum and no_check is not on for 775 * the socket then skip it. 776 */ 777 if (!uh->check && !udp_sk(sk)->no_check6_rx) 778 continue; 779 if (!first) { 780 first = sk; 781 continue; 782 } 783 nskb = skb_clone(skb, GFP_ATOMIC); 784 if (unlikely(!nskb)) { 785 atomic_inc(&sk->sk_drops); 786 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 787 IS_UDPLITE(sk)); 788 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 789 IS_UDPLITE(sk)); 790 continue; 791 } 792 793 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 794 consume_skb(nskb); 795 } 796 797 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 798 if (use_hash2 && hash2 != hash2_any) { 799 hash2 = hash2_any; 800 goto start_lookup; 801 } 802 803 if (first) { 804 if (udpv6_queue_rcv_skb(first, skb) > 0) 805 consume_skb(skb); 806 } else { 807 kfree_skb(skb); 808 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 809 proto == IPPROTO_UDPLITE); 810 } 811 return 0; 812 } 813 814 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 815 { 816 if (udp_sk_rx_dst_set(sk, dst)) { 817 const struct rt6_info *rt = (const struct rt6_info *)dst; 818 819 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 820 } 821 } 822 823 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 824 * return code conversion for ip layer consumption 825 */ 826 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 827 struct udphdr *uh) 828 { 829 int ret; 830 831 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 832 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 833 834 ret = udpv6_queue_rcv_skb(sk, skb); 835 836 /* a return value > 0 means to resubmit the input */ 837 if (ret > 0) 838 return ret; 839 return 0; 840 } 841 842 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 843 int proto) 844 { 845 const struct in6_addr *saddr, *daddr; 846 struct net *net = dev_net(skb->dev); 847 struct udphdr *uh; 848 struct sock *sk; 849 bool refcounted; 850 u32 ulen = 0; 851 852 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 853 goto discard; 854 855 saddr = &ipv6_hdr(skb)->saddr; 856 daddr = &ipv6_hdr(skb)->daddr; 857 uh = udp_hdr(skb); 858 859 ulen = ntohs(uh->len); 860 if (ulen > skb->len) 861 goto short_packet; 862 863 if (proto == IPPROTO_UDP) { 864 /* UDP validates ulen. */ 865 866 /* Check for jumbo payload */ 867 if (ulen == 0) 868 ulen = skb->len; 869 870 if (ulen < sizeof(*uh)) 871 goto short_packet; 872 873 if (ulen < skb->len) { 874 if (pskb_trim_rcsum(skb, ulen)) 875 goto short_packet; 876 saddr = &ipv6_hdr(skb)->saddr; 877 daddr = &ipv6_hdr(skb)->daddr; 878 uh = udp_hdr(skb); 879 } 880 } 881 882 if (udp6_csum_init(skb, uh, proto)) 883 goto csum_error; 884 885 /* Check if the socket is already available, e.g. due to early demux */ 886 sk = skb_steal_sock(skb, &refcounted); 887 if (sk) { 888 struct dst_entry *dst = skb_dst(skb); 889 int ret; 890 891 if (unlikely(sk->sk_rx_dst != dst)) 892 udp6_sk_rx_dst_set(sk, dst); 893 894 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 895 if (refcounted) 896 sock_put(sk); 897 goto report_csum_error; 898 } 899 900 ret = udp6_unicast_rcv_skb(sk, skb, uh); 901 if (refcounted) 902 sock_put(sk); 903 return ret; 904 } 905 906 /* 907 * Multicast receive code 908 */ 909 if (ipv6_addr_is_multicast(daddr)) 910 return __udp6_lib_mcast_deliver(net, skb, 911 saddr, daddr, udptable, proto); 912 913 /* Unicast */ 914 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 915 if (sk) { 916 if (!uh->check && !udp_sk(sk)->no_check6_rx) 917 goto report_csum_error; 918 return udp6_unicast_rcv_skb(sk, skb, uh); 919 } 920 921 if (!uh->check) 922 goto report_csum_error; 923 924 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 925 goto discard; 926 927 if (udp_lib_checksum_complete(skb)) 928 goto csum_error; 929 930 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 931 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 932 933 kfree_skb(skb); 934 return 0; 935 936 short_packet: 937 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 938 proto == IPPROTO_UDPLITE ? "-Lite" : "", 939 saddr, ntohs(uh->source), 940 ulen, skb->len, 941 daddr, ntohs(uh->dest)); 942 goto discard; 943 944 report_csum_error: 945 udp6_csum_zero_error(skb); 946 csum_error: 947 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 948 discard: 949 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 950 kfree_skb(skb); 951 return 0; 952 } 953 954 955 static struct sock *__udp6_lib_demux_lookup(struct net *net, 956 __be16 loc_port, const struct in6_addr *loc_addr, 957 __be16 rmt_port, const struct in6_addr *rmt_addr, 958 int dif, int sdif) 959 { 960 unsigned short hnum = ntohs(loc_port); 961 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 962 unsigned int slot2 = hash2 & udp_table.mask; 963 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 964 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 965 struct sock *sk; 966 967 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 968 if (sk->sk_state == TCP_ESTABLISHED && 969 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 970 return sk; 971 /* Only check first socket in chain */ 972 break; 973 } 974 return NULL; 975 } 976 977 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb) 978 { 979 struct net *net = dev_net(skb->dev); 980 const struct udphdr *uh; 981 struct sock *sk; 982 struct dst_entry *dst; 983 int dif = skb->dev->ifindex; 984 int sdif = inet6_sdif(skb); 985 986 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 987 sizeof(struct udphdr))) 988 return; 989 990 uh = udp_hdr(skb); 991 992 if (skb->pkt_type == PACKET_HOST) 993 sk = __udp6_lib_demux_lookup(net, uh->dest, 994 &ipv6_hdr(skb)->daddr, 995 uh->source, &ipv6_hdr(skb)->saddr, 996 dif, sdif); 997 else 998 return; 999 1000 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1001 return; 1002 1003 skb->sk = sk; 1004 skb->destructor = sock_efree; 1005 dst = READ_ONCE(sk->sk_rx_dst); 1006 1007 if (dst) 1008 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 1009 if (dst) { 1010 /* set noref for now. 1011 * any place which wants to hold dst has to call 1012 * dst_hold_safe() 1013 */ 1014 skb_dst_set_noref(skb, dst); 1015 } 1016 } 1017 1018 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1019 { 1020 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1021 } 1022 1023 /* 1024 * Throw away all pending data and cancel the corking. Socket is locked. 1025 */ 1026 static void udp_v6_flush_pending_frames(struct sock *sk) 1027 { 1028 struct udp_sock *up = udp_sk(sk); 1029 1030 if (up->pending == AF_INET) 1031 udp_flush_pending_frames(sk); 1032 else if (up->pending) { 1033 up->len = 0; 1034 up->pending = 0; 1035 ip6_flush_pending_frames(sk); 1036 } 1037 } 1038 1039 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1040 int addr_len) 1041 { 1042 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1043 return -EINVAL; 1044 /* The following checks are replicated from __ip6_datagram_connect() 1045 * and intended to prevent BPF program called below from accessing 1046 * bytes that are out of the bound specified by user in addr_len. 1047 */ 1048 if (uaddr->sa_family == AF_INET) { 1049 if (__ipv6_only_sock(sk)) 1050 return -EAFNOSUPPORT; 1051 return udp_pre_connect(sk, uaddr, addr_len); 1052 } 1053 1054 if (addr_len < SIN6_LEN_RFC2133) 1055 return -EINVAL; 1056 1057 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1058 } 1059 1060 /** 1061 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1062 * @sk: socket we are sending on 1063 * @skb: sk_buff containing the filled-in UDP header 1064 * (checksum field must be zeroed out) 1065 */ 1066 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1067 const struct in6_addr *saddr, 1068 const struct in6_addr *daddr, int len) 1069 { 1070 unsigned int offset; 1071 struct udphdr *uh = udp_hdr(skb); 1072 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1073 __wsum csum = 0; 1074 1075 if (!frags) { 1076 /* Only one fragment on the socket. */ 1077 skb->csum_start = skb_transport_header(skb) - skb->head; 1078 skb->csum_offset = offsetof(struct udphdr, check); 1079 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1080 } else { 1081 /* 1082 * HW-checksum won't work as there are two or more 1083 * fragments on the socket so that all csums of sk_buffs 1084 * should be together 1085 */ 1086 offset = skb_transport_offset(skb); 1087 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1088 csum = skb->csum; 1089 1090 skb->ip_summed = CHECKSUM_NONE; 1091 1092 do { 1093 csum = csum_add(csum, frags->csum); 1094 } while ((frags = frags->next)); 1095 1096 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1097 csum); 1098 if (uh->check == 0) 1099 uh->check = CSUM_MANGLED_0; 1100 } 1101 } 1102 1103 /* 1104 * Sending 1105 */ 1106 1107 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1108 struct inet_cork *cork) 1109 { 1110 struct sock *sk = skb->sk; 1111 struct udphdr *uh; 1112 int err = 0; 1113 int is_udplite = IS_UDPLITE(sk); 1114 __wsum csum = 0; 1115 int offset = skb_transport_offset(skb); 1116 int len = skb->len - offset; 1117 int datalen = len - sizeof(*uh); 1118 1119 /* 1120 * Create a UDP header 1121 */ 1122 uh = udp_hdr(skb); 1123 uh->source = fl6->fl6_sport; 1124 uh->dest = fl6->fl6_dport; 1125 uh->len = htons(len); 1126 uh->check = 0; 1127 1128 if (cork->gso_size) { 1129 const int hlen = skb_network_header_len(skb) + 1130 sizeof(struct udphdr); 1131 1132 if (hlen + cork->gso_size > cork->fragsize) { 1133 kfree_skb(skb); 1134 return -EINVAL; 1135 } 1136 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 1137 kfree_skb(skb); 1138 return -EINVAL; 1139 } 1140 if (udp_sk(sk)->no_check6_tx) { 1141 kfree_skb(skb); 1142 return -EINVAL; 1143 } 1144 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1145 dst_xfrm(skb_dst(skb))) { 1146 kfree_skb(skb); 1147 return -EIO; 1148 } 1149 1150 if (datalen > cork->gso_size) { 1151 skb_shinfo(skb)->gso_size = cork->gso_size; 1152 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1153 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1154 cork->gso_size); 1155 } 1156 goto csum_partial; 1157 } 1158 1159 if (is_udplite) 1160 csum = udplite_csum(skb); 1161 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1162 skb->ip_summed = CHECKSUM_NONE; 1163 goto send; 1164 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1165 csum_partial: 1166 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1167 goto send; 1168 } else 1169 csum = udp_csum(skb); 1170 1171 /* add protocol-dependent pseudo-header */ 1172 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1173 len, fl6->flowi6_proto, csum); 1174 if (uh->check == 0) 1175 uh->check = CSUM_MANGLED_0; 1176 1177 send: 1178 err = ip6_send_skb(skb); 1179 if (err) { 1180 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1181 UDP6_INC_STATS(sock_net(sk), 1182 UDP_MIB_SNDBUFERRORS, is_udplite); 1183 err = 0; 1184 } 1185 } else { 1186 UDP6_INC_STATS(sock_net(sk), 1187 UDP_MIB_OUTDATAGRAMS, is_udplite); 1188 } 1189 return err; 1190 } 1191 1192 static int udp_v6_push_pending_frames(struct sock *sk) 1193 { 1194 struct sk_buff *skb; 1195 struct udp_sock *up = udp_sk(sk); 1196 struct flowi6 fl6; 1197 int err = 0; 1198 1199 if (up->pending == AF_INET) 1200 return udp_push_pending_frames(sk); 1201 1202 /* ip6_finish_skb will release the cork, so make a copy of 1203 * fl6 here. 1204 */ 1205 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1206 1207 skb = ip6_finish_skb(sk); 1208 if (!skb) 1209 goto out; 1210 1211 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1212 1213 out: 1214 up->len = 0; 1215 up->pending = 0; 1216 return err; 1217 } 1218 1219 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1220 { 1221 struct ipv6_txoptions opt_space; 1222 struct udp_sock *up = udp_sk(sk); 1223 struct inet_sock *inet = inet_sk(sk); 1224 struct ipv6_pinfo *np = inet6_sk(sk); 1225 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1226 struct in6_addr *daddr, *final_p, final; 1227 struct ipv6_txoptions *opt = NULL; 1228 struct ipv6_txoptions *opt_to_free = NULL; 1229 struct ip6_flowlabel *flowlabel = NULL; 1230 struct flowi6 fl6; 1231 struct dst_entry *dst; 1232 struct ipcm6_cookie ipc6; 1233 int addr_len = msg->msg_namelen; 1234 bool connected = false; 1235 int ulen = len; 1236 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1237 int err; 1238 int is_udplite = IS_UDPLITE(sk); 1239 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1240 1241 ipcm6_init(&ipc6); 1242 ipc6.gso_size = up->gso_size; 1243 ipc6.sockc.tsflags = sk->sk_tsflags; 1244 ipc6.sockc.mark = sk->sk_mark; 1245 1246 /* destination address check */ 1247 if (sin6) { 1248 if (addr_len < offsetof(struct sockaddr, sa_data)) 1249 return -EINVAL; 1250 1251 switch (sin6->sin6_family) { 1252 case AF_INET6: 1253 if (addr_len < SIN6_LEN_RFC2133) 1254 return -EINVAL; 1255 daddr = &sin6->sin6_addr; 1256 if (ipv6_addr_any(daddr) && 1257 ipv6_addr_v4mapped(&np->saddr)) 1258 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1259 daddr); 1260 break; 1261 case AF_INET: 1262 goto do_udp_sendmsg; 1263 case AF_UNSPEC: 1264 msg->msg_name = sin6 = NULL; 1265 msg->msg_namelen = addr_len = 0; 1266 daddr = NULL; 1267 break; 1268 default: 1269 return -EINVAL; 1270 } 1271 } else if (!up->pending) { 1272 if (sk->sk_state != TCP_ESTABLISHED) 1273 return -EDESTADDRREQ; 1274 daddr = &sk->sk_v6_daddr; 1275 } else 1276 daddr = NULL; 1277 1278 if (daddr) { 1279 if (ipv6_addr_v4mapped(daddr)) { 1280 struct sockaddr_in sin; 1281 sin.sin_family = AF_INET; 1282 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1283 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1284 msg->msg_name = &sin; 1285 msg->msg_namelen = sizeof(sin); 1286 do_udp_sendmsg: 1287 if (__ipv6_only_sock(sk)) 1288 return -ENETUNREACH; 1289 return udp_sendmsg(sk, msg, len); 1290 } 1291 } 1292 1293 if (up->pending == AF_INET) 1294 return udp_sendmsg(sk, msg, len); 1295 1296 /* Rough check on arithmetic overflow, 1297 better check is made in ip6_append_data(). 1298 */ 1299 if (len > INT_MAX - sizeof(struct udphdr)) 1300 return -EMSGSIZE; 1301 1302 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1303 if (up->pending) { 1304 /* 1305 * There are pending frames. 1306 * The socket lock must be held while it's corked. 1307 */ 1308 lock_sock(sk); 1309 if (likely(up->pending)) { 1310 if (unlikely(up->pending != AF_INET6)) { 1311 release_sock(sk); 1312 return -EAFNOSUPPORT; 1313 } 1314 dst = NULL; 1315 goto do_append_data; 1316 } 1317 release_sock(sk); 1318 } 1319 ulen += sizeof(struct udphdr); 1320 1321 memset(&fl6, 0, sizeof(fl6)); 1322 1323 if (sin6) { 1324 if (sin6->sin6_port == 0) 1325 return -EINVAL; 1326 1327 fl6.fl6_dport = sin6->sin6_port; 1328 daddr = &sin6->sin6_addr; 1329 1330 if (np->sndflow) { 1331 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1332 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1333 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1334 if (IS_ERR(flowlabel)) 1335 return -EINVAL; 1336 } 1337 } 1338 1339 /* 1340 * Otherwise it will be difficult to maintain 1341 * sk->sk_dst_cache. 1342 */ 1343 if (sk->sk_state == TCP_ESTABLISHED && 1344 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1345 daddr = &sk->sk_v6_daddr; 1346 1347 if (addr_len >= sizeof(struct sockaddr_in6) && 1348 sin6->sin6_scope_id && 1349 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1350 fl6.flowi6_oif = sin6->sin6_scope_id; 1351 } else { 1352 if (sk->sk_state != TCP_ESTABLISHED) 1353 return -EDESTADDRREQ; 1354 1355 fl6.fl6_dport = inet->inet_dport; 1356 daddr = &sk->sk_v6_daddr; 1357 fl6.flowlabel = np->flow_label; 1358 connected = true; 1359 } 1360 1361 if (!fl6.flowi6_oif) 1362 fl6.flowi6_oif = sk->sk_bound_dev_if; 1363 1364 if (!fl6.flowi6_oif) 1365 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1366 1367 fl6.flowi6_mark = ipc6.sockc.mark; 1368 fl6.flowi6_uid = sk->sk_uid; 1369 1370 if (msg->msg_controllen) { 1371 opt = &opt_space; 1372 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1373 opt->tot_len = sizeof(*opt); 1374 ipc6.opt = opt; 1375 1376 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1377 if (err > 0) 1378 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1379 &ipc6); 1380 if (err < 0) { 1381 fl6_sock_release(flowlabel); 1382 return err; 1383 } 1384 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1385 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1386 if (IS_ERR(flowlabel)) 1387 return -EINVAL; 1388 } 1389 if (!(opt->opt_nflen|opt->opt_flen)) 1390 opt = NULL; 1391 connected = false; 1392 } 1393 if (!opt) { 1394 opt = txopt_get(np); 1395 opt_to_free = opt; 1396 } 1397 if (flowlabel) 1398 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1399 opt = ipv6_fixup_options(&opt_space, opt); 1400 ipc6.opt = opt; 1401 1402 fl6.flowi6_proto = sk->sk_protocol; 1403 fl6.daddr = *daddr; 1404 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1405 fl6.saddr = np->saddr; 1406 fl6.fl6_sport = inet->inet_sport; 1407 1408 if (cgroup_bpf_enabled && !connected) { 1409 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1410 (struct sockaddr *)sin6, &fl6.saddr); 1411 if (err) 1412 goto out_no_dst; 1413 if (sin6) { 1414 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1415 /* BPF program rewrote IPv6-only by IPv4-mapped 1416 * IPv6. It's currently unsupported. 1417 */ 1418 err = -ENOTSUPP; 1419 goto out_no_dst; 1420 } 1421 if (sin6->sin6_port == 0) { 1422 /* BPF program set invalid port. Reject it. */ 1423 err = -EINVAL; 1424 goto out_no_dst; 1425 } 1426 fl6.fl6_dport = sin6->sin6_port; 1427 fl6.daddr = sin6->sin6_addr; 1428 } 1429 } 1430 1431 if (ipv6_addr_any(&fl6.daddr)) 1432 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1433 1434 final_p = fl6_update_dst(&fl6, opt, &final); 1435 if (final_p) 1436 connected = false; 1437 1438 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1439 fl6.flowi6_oif = np->mcast_oif; 1440 connected = false; 1441 } else if (!fl6.flowi6_oif) 1442 fl6.flowi6_oif = np->ucast_oif; 1443 1444 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1445 1446 if (ipc6.tclass < 0) 1447 ipc6.tclass = np->tclass; 1448 1449 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1450 1451 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1452 if (IS_ERR(dst)) { 1453 err = PTR_ERR(dst); 1454 dst = NULL; 1455 goto out; 1456 } 1457 1458 if (ipc6.hlimit < 0) 1459 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1460 1461 if (msg->msg_flags&MSG_CONFIRM) 1462 goto do_confirm; 1463 back_from_confirm: 1464 1465 /* Lockless fast path for the non-corking case */ 1466 if (!corkreq) { 1467 struct inet_cork_full cork; 1468 struct sk_buff *skb; 1469 1470 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1471 sizeof(struct udphdr), &ipc6, 1472 &fl6, (struct rt6_info *)dst, 1473 msg->msg_flags, &cork); 1474 err = PTR_ERR(skb); 1475 if (!IS_ERR_OR_NULL(skb)) 1476 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1477 goto out; 1478 } 1479 1480 lock_sock(sk); 1481 if (unlikely(up->pending)) { 1482 /* The socket is already corked while preparing it. */ 1483 /* ... which is an evident application bug. --ANK */ 1484 release_sock(sk); 1485 1486 net_dbg_ratelimited("udp cork app bug 2\n"); 1487 err = -EINVAL; 1488 goto out; 1489 } 1490 1491 up->pending = AF_INET6; 1492 1493 do_append_data: 1494 if (ipc6.dontfrag < 0) 1495 ipc6.dontfrag = np->dontfrag; 1496 up->len += ulen; 1497 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1498 &ipc6, &fl6, (struct rt6_info *)dst, 1499 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1500 if (err) 1501 udp_v6_flush_pending_frames(sk); 1502 else if (!corkreq) 1503 err = udp_v6_push_pending_frames(sk); 1504 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1505 up->pending = 0; 1506 1507 if (err > 0) 1508 err = np->recverr ? net_xmit_errno(err) : 0; 1509 release_sock(sk); 1510 1511 out: 1512 dst_release(dst); 1513 out_no_dst: 1514 fl6_sock_release(flowlabel); 1515 txopt_put(opt_to_free); 1516 if (!err) 1517 return len; 1518 /* 1519 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1520 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1521 * we don't have a good statistic (IpOutDiscards but it can be too many 1522 * things). We could add another new stat but at least for now that 1523 * seems like overkill. 1524 */ 1525 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1526 UDP6_INC_STATS(sock_net(sk), 1527 UDP_MIB_SNDBUFERRORS, is_udplite); 1528 } 1529 return err; 1530 1531 do_confirm: 1532 if (msg->msg_flags & MSG_PROBE) 1533 dst_confirm_neigh(dst, &fl6.daddr); 1534 if (!(msg->msg_flags&MSG_PROBE) || len) 1535 goto back_from_confirm; 1536 err = 0; 1537 goto out; 1538 } 1539 1540 void udpv6_destroy_sock(struct sock *sk) 1541 { 1542 struct udp_sock *up = udp_sk(sk); 1543 lock_sock(sk); 1544 udp_v6_flush_pending_frames(sk); 1545 release_sock(sk); 1546 1547 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1548 if (up->encap_type) { 1549 void (*encap_destroy)(struct sock *sk); 1550 encap_destroy = READ_ONCE(up->encap_destroy); 1551 if (encap_destroy) 1552 encap_destroy(sk); 1553 } 1554 if (up->encap_enabled) 1555 static_branch_dec(&udpv6_encap_needed_key); 1556 } 1557 1558 inet6_destroy_sock(sk); 1559 } 1560 1561 /* 1562 * Socket option code for UDP 1563 */ 1564 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1565 char __user *optval, unsigned int optlen) 1566 { 1567 if (level == SOL_UDP || level == SOL_UDPLITE) 1568 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1569 udp_v6_push_pending_frames); 1570 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1571 } 1572 1573 #ifdef CONFIG_COMPAT 1574 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1575 char __user *optval, unsigned int optlen) 1576 { 1577 if (level == SOL_UDP || level == SOL_UDPLITE) 1578 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1579 udp_v6_push_pending_frames); 1580 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1581 } 1582 #endif 1583 1584 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1585 char __user *optval, int __user *optlen) 1586 { 1587 if (level == SOL_UDP || level == SOL_UDPLITE) 1588 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1589 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1590 } 1591 1592 #ifdef CONFIG_COMPAT 1593 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1594 char __user *optval, int __user *optlen) 1595 { 1596 if (level == SOL_UDP || level == SOL_UDPLITE) 1597 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1598 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1599 } 1600 #endif 1601 1602 /* thinking of making this const? Don't. 1603 * early_demux can change based on sysctl. 1604 */ 1605 static struct inet6_protocol udpv6_protocol = { 1606 .early_demux = udp_v6_early_demux, 1607 .early_demux_handler = udp_v6_early_demux, 1608 .handler = udpv6_rcv, 1609 .err_handler = udpv6_err, 1610 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1611 }; 1612 1613 /* ------------------------------------------------------------------------ */ 1614 #ifdef CONFIG_PROC_FS 1615 int udp6_seq_show(struct seq_file *seq, void *v) 1616 { 1617 if (v == SEQ_START_TOKEN) { 1618 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1619 } else { 1620 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1621 struct inet_sock *inet = inet_sk(v); 1622 __u16 srcp = ntohs(inet->inet_sport); 1623 __u16 destp = ntohs(inet->inet_dport); 1624 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1625 udp_rqueue_get(v), bucket); 1626 } 1627 return 0; 1628 } 1629 1630 const struct seq_operations udp6_seq_ops = { 1631 .start = udp_seq_start, 1632 .next = udp_seq_next, 1633 .stop = udp_seq_stop, 1634 .show = udp6_seq_show, 1635 }; 1636 EXPORT_SYMBOL(udp6_seq_ops); 1637 1638 static struct udp_seq_afinfo udp6_seq_afinfo = { 1639 .family = AF_INET6, 1640 .udp_table = &udp_table, 1641 }; 1642 1643 int __net_init udp6_proc_init(struct net *net) 1644 { 1645 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1646 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1647 return -ENOMEM; 1648 return 0; 1649 } 1650 1651 void udp6_proc_exit(struct net *net) 1652 { 1653 remove_proc_entry("udp6", net->proc_net); 1654 } 1655 #endif /* CONFIG_PROC_FS */ 1656 1657 /* ------------------------------------------------------------------------ */ 1658 1659 struct proto udpv6_prot = { 1660 .name = "UDPv6", 1661 .owner = THIS_MODULE, 1662 .close = udp_lib_close, 1663 .pre_connect = udpv6_pre_connect, 1664 .connect = ip6_datagram_connect, 1665 .disconnect = udp_disconnect, 1666 .ioctl = udp_ioctl, 1667 .init = udp_init_sock, 1668 .destroy = udpv6_destroy_sock, 1669 .setsockopt = udpv6_setsockopt, 1670 .getsockopt = udpv6_getsockopt, 1671 .sendmsg = udpv6_sendmsg, 1672 .recvmsg = udpv6_recvmsg, 1673 .release_cb = ip6_datagram_release_cb, 1674 .hash = udp_lib_hash, 1675 .unhash = udp_lib_unhash, 1676 .rehash = udp_v6_rehash, 1677 .get_port = udp_v6_get_port, 1678 .memory_allocated = &udp_memory_allocated, 1679 .sysctl_mem = sysctl_udp_mem, 1680 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1681 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1682 .obj_size = sizeof(struct udp6_sock), 1683 .h.udp_table = &udp_table, 1684 #ifdef CONFIG_COMPAT 1685 .compat_setsockopt = compat_udpv6_setsockopt, 1686 .compat_getsockopt = compat_udpv6_getsockopt, 1687 #endif 1688 .diag_destroy = udp_abort, 1689 }; 1690 1691 static struct inet_protosw udpv6_protosw = { 1692 .type = SOCK_DGRAM, 1693 .protocol = IPPROTO_UDP, 1694 .prot = &udpv6_prot, 1695 .ops = &inet6_dgram_ops, 1696 .flags = INET_PROTOSW_PERMANENT, 1697 }; 1698 1699 int __init udpv6_init(void) 1700 { 1701 int ret; 1702 1703 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1704 if (ret) 1705 goto out; 1706 1707 ret = inet6_register_protosw(&udpv6_protosw); 1708 if (ret) 1709 goto out_udpv6_protocol; 1710 out: 1711 return ret; 1712 1713 out_udpv6_protocol: 1714 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1715 goto out; 1716 } 1717 1718 void udpv6_exit(void) 1719 { 1720 inet6_unregister_protosw(&udpv6_protosw); 1721 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1722 } 1723