1 /* 2 * UDP over IPv6 3 * Linux INET6 implementation 4 * 5 * Authors: 6 * Pedro Roque <roque@di.fc.ul.pt> 7 * 8 * Based on linux/ipv4/udp.c 9 * 10 * Fixes: 11 * Hideaki YOSHIFUJI : sin6_scope_id support 12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 14 * a single port at the same time. 15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. 17 * 18 * This program is free software; you can redistribute it and/or 19 * modify it under the terms of the GNU General Public License 20 * as published by the Free Software Foundation; either version 21 * 2 of the License, or (at your option) any later version. 22 */ 23 24 #include <linux/errno.h> 25 #include <linux/types.h> 26 #include <linux/socket.h> 27 #include <linux/sockios.h> 28 #include <linux/net.h> 29 #include <linux/in6.h> 30 #include <linux/netdevice.h> 31 #include <linux/if_arp.h> 32 #include <linux/ipv6.h> 33 #include <linux/icmpv6.h> 34 #include <linux/init.h> 35 #include <linux/module.h> 36 #include <linux/skbuff.h> 37 #include <linux/slab.h> 38 #include <linux/uaccess.h> 39 #include <linux/indirect_call_wrapper.h> 40 41 #include <net/addrconf.h> 42 #include <net/ndisc.h> 43 #include <net/protocol.h> 44 #include <net/transp_v6.h> 45 #include <net/ip6_route.h> 46 #include <net/raw.h> 47 #include <net/tcp_states.h> 48 #include <net/ip6_checksum.h> 49 #include <net/ip6_tunnel.h> 50 #include <net/xfrm.h> 51 #include <net/inet_hashtables.h> 52 #include <net/inet6_hashtables.h> 53 #include <net/busy_poll.h> 54 #include <net/sock_reuseport.h> 55 56 #include <linux/proc_fs.h> 57 #include <linux/seq_file.h> 58 #include <trace/events/skb.h> 59 #include "udp_impl.h" 60 61 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb) 62 { 63 #if defined(CONFIG_NET_L3_MASTER_DEV) 64 if (!net->ipv4.sysctl_udp_l3mdev_accept && 65 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags)) 66 return true; 67 #endif 68 return false; 69 } 70 71 static u32 udp6_ehashfn(const struct net *net, 72 const struct in6_addr *laddr, 73 const u16 lport, 74 const struct in6_addr *faddr, 75 const __be16 fport) 76 { 77 static u32 udp6_ehash_secret __read_mostly; 78 static u32 udp_ipv6_hash_secret __read_mostly; 79 80 u32 lhash, fhash; 81 82 net_get_random_once(&udp6_ehash_secret, 83 sizeof(udp6_ehash_secret)); 84 net_get_random_once(&udp_ipv6_hash_secret, 85 sizeof(udp_ipv6_hash_secret)); 86 87 lhash = (__force u32)laddr->s6_addr32[3]; 88 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 89 90 return __inet6_ehashfn(lhash, lport, fhash, fport, 91 udp_ipv6_hash_secret + net_hash_mix(net)); 92 } 93 94 int udp_v6_get_port(struct sock *sk, unsigned short snum) 95 { 96 unsigned int hash2_nulladdr = 97 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 98 unsigned int hash2_partial = 99 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 100 101 /* precompute partial secondary hash */ 102 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 103 return udp_lib_get_port(sk, snum, hash2_nulladdr); 104 } 105 106 void udp_v6_rehash(struct sock *sk) 107 { 108 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 109 &sk->sk_v6_rcv_saddr, 110 inet_sk(sk)->inet_num); 111 112 udp_lib_rehash(sk, new_hash); 113 } 114 115 static int compute_score(struct sock *sk, struct net *net, 116 const struct in6_addr *saddr, __be16 sport, 117 const struct in6_addr *daddr, unsigned short hnum, 118 int dif, int sdif, bool exact_dif) 119 { 120 int score; 121 struct inet_sock *inet; 122 bool dev_match; 123 124 if (!net_eq(sock_net(sk), net) || 125 udp_sk(sk)->udp_port_hash != hnum || 126 sk->sk_family != PF_INET6) 127 return -1; 128 129 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 130 return -1; 131 132 score = 0; 133 inet = inet_sk(sk); 134 135 if (inet->inet_dport) { 136 if (inet->inet_dport != sport) 137 return -1; 138 score++; 139 } 140 141 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 142 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 143 return -1; 144 score++; 145 } 146 147 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif); 148 if (!dev_match) 149 return -1; 150 score++; 151 152 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 153 score++; 154 155 return score; 156 } 157 158 /* called with rcu_read_lock() */ 159 static struct sock *udp6_lib_lookup2(struct net *net, 160 const struct in6_addr *saddr, __be16 sport, 161 const struct in6_addr *daddr, unsigned int hnum, 162 int dif, int sdif, bool exact_dif, 163 struct udp_hslot *hslot2, struct sk_buff *skb) 164 { 165 struct sock *sk, *result; 166 int score, badness; 167 u32 hash = 0; 168 169 result = NULL; 170 badness = -1; 171 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 172 score = compute_score(sk, net, saddr, sport, 173 daddr, hnum, dif, sdif, exact_dif); 174 if (score > badness) { 175 if (sk->sk_reuseport) { 176 hash = udp6_ehashfn(net, daddr, hnum, 177 saddr, sport); 178 179 result = reuseport_select_sock(sk, hash, skb, 180 sizeof(struct udphdr)); 181 if (result) 182 return result; 183 } 184 result = sk; 185 badness = score; 186 } 187 } 188 return result; 189 } 190 191 /* rcu_read_lock() must be held */ 192 struct sock *__udp6_lib_lookup(struct net *net, 193 const struct in6_addr *saddr, __be16 sport, 194 const struct in6_addr *daddr, __be16 dport, 195 int dif, int sdif, struct udp_table *udptable, 196 struct sk_buff *skb) 197 { 198 unsigned short hnum = ntohs(dport); 199 unsigned int hash2, slot2; 200 struct udp_hslot *hslot2; 201 struct sock *result; 202 bool exact_dif = udp6_lib_exact_dif_match(net, skb); 203 204 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 205 slot2 = hash2 & udptable->mask; 206 hslot2 = &udptable->hash2[slot2]; 207 208 result = udp6_lib_lookup2(net, saddr, sport, 209 daddr, hnum, dif, sdif, exact_dif, 210 hslot2, skb); 211 if (!result) { 212 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 213 slot2 = hash2 & udptable->mask; 214 215 hslot2 = &udptable->hash2[slot2]; 216 217 result = udp6_lib_lookup2(net, saddr, sport, 218 &in6addr_any, hnum, dif, sdif, 219 exact_dif, hslot2, 220 skb); 221 } 222 if (unlikely(IS_ERR(result))) 223 return NULL; 224 return result; 225 } 226 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 227 228 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 229 __be16 sport, __be16 dport, 230 struct udp_table *udptable) 231 { 232 const struct ipv6hdr *iph = ipv6_hdr(skb); 233 234 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 235 &iph->daddr, dport, inet6_iif(skb), 236 inet6_sdif(skb), udptable, skb); 237 } 238 239 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, 240 __be16 sport, __be16 dport) 241 { 242 const struct ipv6hdr *iph = ipv6_hdr(skb); 243 244 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 245 &iph->daddr, dport, inet6_iif(skb), 246 inet6_sdif(skb), &udp_table, NULL); 247 } 248 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); 249 250 /* Must be called under rcu_read_lock(). 251 * Does increment socket refcount. 252 */ 253 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 254 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 255 const struct in6_addr *daddr, __be16 dport, int dif) 256 { 257 struct sock *sk; 258 259 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 260 dif, 0, &udp_table, NULL); 261 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 262 sk = NULL; 263 return sk; 264 } 265 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 266 #endif 267 268 /* do not use the scratch area len for jumbogram: their length execeeds the 269 * scratch area space; note that the IP6CB flags is still in the first 270 * cacheline, so checking for jumbograms is cheap 271 */ 272 static int udp6_skb_len(struct sk_buff *skb) 273 { 274 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 275 } 276 277 /* 278 * This should be easy, if there is something there we 279 * return it, otherwise we block. 280 */ 281 282 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 283 int noblock, int flags, int *addr_len) 284 { 285 struct ipv6_pinfo *np = inet6_sk(sk); 286 struct inet_sock *inet = inet_sk(sk); 287 struct sk_buff *skb; 288 unsigned int ulen, copied; 289 int off, err, peeking = flags & MSG_PEEK; 290 int is_udplite = IS_UDPLITE(sk); 291 struct udp_mib __percpu *mib; 292 bool checksum_valid = false; 293 int is_udp4; 294 295 if (flags & MSG_ERRQUEUE) 296 return ipv6_recv_error(sk, msg, len, addr_len); 297 298 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 299 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 300 301 try_again: 302 off = sk_peek_offset(sk, flags); 303 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 304 if (!skb) 305 return err; 306 307 ulen = udp6_skb_len(skb); 308 copied = len; 309 if (copied > ulen - off) 310 copied = ulen - off; 311 else if (copied < ulen) 312 msg->msg_flags |= MSG_TRUNC; 313 314 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 315 mib = __UDPX_MIB(sk, is_udp4); 316 317 /* 318 * If checksum is needed at all, try to do it while copying the 319 * data. If the data is truncated, or if we only want a partial 320 * coverage checksum (UDP-Lite), do it before the copy. 321 */ 322 323 if (copied < ulen || peeking || 324 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 325 checksum_valid = udp_skb_csum_unnecessary(skb) || 326 !__udp_lib_checksum_complete(skb); 327 if (!checksum_valid) 328 goto csum_copy_err; 329 } 330 331 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 332 if (udp_skb_is_linear(skb)) 333 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 334 else 335 err = skb_copy_datagram_msg(skb, off, msg, copied); 336 } else { 337 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 338 if (err == -EINVAL) 339 goto csum_copy_err; 340 } 341 if (unlikely(err)) { 342 if (!peeking) { 343 atomic_inc(&sk->sk_drops); 344 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 345 } 346 kfree_skb(skb); 347 return err; 348 } 349 if (!peeking) 350 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 351 352 sock_recv_ts_and_drops(msg, sk, skb); 353 354 /* Copy the address. */ 355 if (msg->msg_name) { 356 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 357 sin6->sin6_family = AF_INET6; 358 sin6->sin6_port = udp_hdr(skb)->source; 359 sin6->sin6_flowinfo = 0; 360 361 if (is_udp4) { 362 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 363 &sin6->sin6_addr); 364 sin6->sin6_scope_id = 0; 365 } else { 366 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 367 sin6->sin6_scope_id = 368 ipv6_iface_scope_id(&sin6->sin6_addr, 369 inet6_iif(skb)); 370 } 371 *addr_len = sizeof(*sin6); 372 373 if (cgroup_bpf_enabled) 374 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 375 (struct sockaddr *)sin6); 376 } 377 378 if (udp_sk(sk)->gro_enabled) 379 udp_cmsg_recv(msg, sk, skb); 380 381 if (np->rxopt.all) 382 ip6_datagram_recv_common_ctl(sk, msg, skb); 383 384 if (is_udp4) { 385 if (inet->cmsg_flags) 386 ip_cmsg_recv_offset(msg, sk, skb, 387 sizeof(struct udphdr), off); 388 } else { 389 if (np->rxopt.all) 390 ip6_datagram_recv_specific_ctl(sk, msg, skb); 391 } 392 393 err = copied; 394 if (flags & MSG_TRUNC) 395 err = ulen; 396 397 skb_consume_udp(sk, skb, peeking ? -err : err); 398 return err; 399 400 csum_copy_err: 401 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 402 udp_skb_destructor)) { 403 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 404 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 405 } 406 kfree_skb(skb); 407 408 /* starting over for a new packet, but check if we need to yield */ 409 cond_resched(); 410 msg->msg_flags &= ~MSG_TRUNC; 411 goto try_again; 412 } 413 414 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 415 void udpv6_encap_enable(void) 416 { 417 static_branch_inc(&udpv6_encap_needed_key); 418 } 419 EXPORT_SYMBOL(udpv6_encap_enable); 420 421 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 422 * through error handlers in encapsulations looking for a match. 423 */ 424 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 425 struct inet6_skb_parm *opt, 426 u8 type, u8 code, int offset, __be32 info) 427 { 428 int i; 429 430 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 431 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 432 u8 type, u8 code, int offset, __be32 info); 433 const struct ip6_tnl_encap_ops *encap; 434 435 encap = rcu_dereference(ip6tun_encaps[i]); 436 if (!encap) 437 continue; 438 handler = encap->err_handler; 439 if (handler && !handler(skb, opt, type, code, offset, info)) 440 return 0; 441 } 442 443 return -ENOENT; 444 } 445 446 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 447 * reversing source and destination port: this will match tunnels that force the 448 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 449 * lwtunnels might actually break this assumption by being configured with 450 * different destination ports on endpoints, in this case we won't be able to 451 * trace ICMP messages back to them. 452 * 453 * If this doesn't match any socket, probe tunnels with arbitrary destination 454 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 455 * we've sent packets to won't necessarily match the local destination port. 456 * 457 * Then ask the tunnel implementation to match the error against a valid 458 * association. 459 * 460 * Return an error if we can't find a match, the socket if we need further 461 * processing, zero otherwise. 462 */ 463 static struct sock *__udp6_lib_err_encap(struct net *net, 464 const struct ipv6hdr *hdr, int offset, 465 struct udphdr *uh, 466 struct udp_table *udptable, 467 struct sk_buff *skb, 468 struct inet6_skb_parm *opt, 469 u8 type, u8 code, __be32 info) 470 { 471 int network_offset, transport_offset; 472 struct sock *sk; 473 474 network_offset = skb_network_offset(skb); 475 transport_offset = skb_transport_offset(skb); 476 477 /* Network header needs to point to the outer IPv6 header inside ICMP */ 478 skb_reset_network_header(skb); 479 480 /* Transport header needs to point to the UDP header */ 481 skb_set_transport_header(skb, offset); 482 483 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 484 &hdr->saddr, uh->dest, 485 inet6_iif(skb), 0, udptable, skb); 486 if (sk) { 487 int (*lookup)(struct sock *sk, struct sk_buff *skb); 488 struct udp_sock *up = udp_sk(sk); 489 490 lookup = READ_ONCE(up->encap_err_lookup); 491 if (!lookup || lookup(sk, skb)) 492 sk = NULL; 493 } 494 495 if (!sk) { 496 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 497 offset, info)); 498 } 499 500 skb_set_transport_header(skb, transport_offset); 501 skb_set_network_header(skb, network_offset); 502 503 return sk; 504 } 505 506 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 507 u8 type, u8 code, int offset, __be32 info, 508 struct udp_table *udptable) 509 { 510 struct ipv6_pinfo *np; 511 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 512 const struct in6_addr *saddr = &hdr->saddr; 513 const struct in6_addr *daddr = &hdr->daddr; 514 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 515 bool tunnel = false; 516 struct sock *sk; 517 int harderr; 518 int err; 519 struct net *net = dev_net(skb->dev); 520 521 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 522 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 523 if (!sk) { 524 /* No socket for error: try tunnels before discarding */ 525 sk = ERR_PTR(-ENOENT); 526 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 527 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 528 udptable, skb, 529 opt, type, code, info); 530 if (!sk) 531 return 0; 532 } 533 534 if (IS_ERR(sk)) { 535 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 536 ICMP6_MIB_INERRORS); 537 return PTR_ERR(sk); 538 } 539 540 tunnel = true; 541 } 542 543 harderr = icmpv6_err_convert(type, code, &err); 544 np = inet6_sk(sk); 545 546 if (type == ICMPV6_PKT_TOOBIG) { 547 if (!ip6_sk_accept_pmtu(sk)) 548 goto out; 549 ip6_sk_update_pmtu(skb, sk, info); 550 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 551 harderr = 1; 552 } 553 if (type == NDISC_REDIRECT) { 554 if (tunnel) { 555 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 556 sk->sk_mark, sk->sk_uid); 557 } else { 558 ip6_sk_redirect(skb, sk); 559 } 560 goto out; 561 } 562 563 /* Tunnels don't have an application socket: don't pass errors back */ 564 if (tunnel) 565 goto out; 566 567 if (!np->recverr) { 568 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 569 goto out; 570 } else { 571 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 572 } 573 574 sk->sk_err = err; 575 sk->sk_error_report(sk); 576 out: 577 return 0; 578 } 579 580 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 581 { 582 int rc; 583 584 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 585 sock_rps_save_rxhash(sk, skb); 586 sk_mark_napi_id(sk, skb); 587 sk_incoming_cpu_update(sk); 588 } else { 589 sk_mark_napi_id_once(sk, skb); 590 } 591 592 rc = __udp_enqueue_schedule_skb(sk, skb); 593 if (rc < 0) { 594 int is_udplite = IS_UDPLITE(sk); 595 596 /* Note that an ENOMEM error is charged twice */ 597 if (rc == -ENOMEM) 598 UDP6_INC_STATS(sock_net(sk), 599 UDP_MIB_RCVBUFERRORS, is_udplite); 600 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 601 kfree_skb(skb); 602 return -1; 603 } 604 605 return 0; 606 } 607 608 static __inline__ int udpv6_err(struct sk_buff *skb, 609 struct inet6_skb_parm *opt, u8 type, 610 u8 code, int offset, __be32 info) 611 { 612 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 613 } 614 615 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 616 { 617 struct udp_sock *up = udp_sk(sk); 618 int is_udplite = IS_UDPLITE(sk); 619 620 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 621 goto drop; 622 623 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 624 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 625 626 /* 627 * This is an encapsulation socket so pass the skb to 628 * the socket's udp_encap_rcv() hook. Otherwise, just 629 * fall through and pass this up the UDP socket. 630 * up->encap_rcv() returns the following value: 631 * =0 if skb was successfully passed to the encap 632 * handler or was discarded by it. 633 * >0 if skb should be passed on to UDP. 634 * <0 if skb should be resubmitted as proto -N 635 */ 636 637 /* if we're overly short, let UDP handle it */ 638 encap_rcv = READ_ONCE(up->encap_rcv); 639 if (encap_rcv) { 640 int ret; 641 642 /* Verify checksum before giving to encap */ 643 if (udp_lib_checksum_complete(skb)) 644 goto csum_error; 645 646 ret = encap_rcv(sk, skb); 647 if (ret <= 0) { 648 __UDP_INC_STATS(sock_net(sk), 649 UDP_MIB_INDATAGRAMS, 650 is_udplite); 651 return -ret; 652 } 653 } 654 655 /* FALLTHROUGH -- it's a UDP Packet */ 656 } 657 658 /* 659 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 660 */ 661 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 662 663 if (up->pcrlen == 0) { /* full coverage was set */ 664 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 665 UDP_SKB_CB(skb)->cscov, skb->len); 666 goto drop; 667 } 668 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 669 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 670 UDP_SKB_CB(skb)->cscov, up->pcrlen); 671 goto drop; 672 } 673 } 674 675 prefetch(&sk->sk_rmem_alloc); 676 if (rcu_access_pointer(sk->sk_filter) && 677 udp_lib_checksum_complete(skb)) 678 goto csum_error; 679 680 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 681 goto drop; 682 683 udp_csum_pull_header(skb); 684 685 skb_dst_drop(skb); 686 687 return __udpv6_queue_rcv_skb(sk, skb); 688 689 csum_error: 690 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 691 drop: 692 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 693 atomic_inc(&sk->sk_drops); 694 kfree_skb(skb); 695 return -1; 696 } 697 698 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 699 { 700 struct sk_buff *next, *segs; 701 int ret; 702 703 if (likely(!udp_unexpected_gso(sk, skb))) 704 return udpv6_queue_rcv_one_skb(sk, skb); 705 706 __skb_push(skb, -skb_mac_offset(skb)); 707 segs = udp_rcv_segment(sk, skb, false); 708 for (skb = segs; skb; skb = next) { 709 next = skb->next; 710 __skb_pull(skb, skb_transport_offset(skb)); 711 712 ret = udpv6_queue_rcv_one_skb(sk, skb); 713 if (ret > 0) 714 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 715 true); 716 } 717 return 0; 718 } 719 720 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 721 __be16 loc_port, const struct in6_addr *loc_addr, 722 __be16 rmt_port, const struct in6_addr *rmt_addr, 723 int dif, int sdif, unsigned short hnum) 724 { 725 struct inet_sock *inet = inet_sk(sk); 726 727 if (!net_eq(sock_net(sk), net)) 728 return false; 729 730 if (udp_sk(sk)->udp_port_hash != hnum || 731 sk->sk_family != PF_INET6 || 732 (inet->inet_dport && inet->inet_dport != rmt_port) || 733 (!ipv6_addr_any(&sk->sk_v6_daddr) && 734 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 735 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) || 736 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 737 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 738 return false; 739 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 740 return false; 741 return true; 742 } 743 744 static void udp6_csum_zero_error(struct sk_buff *skb) 745 { 746 /* RFC 2460 section 8.1 says that we SHOULD log 747 * this error. Well, it is reasonable. 748 */ 749 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 750 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 751 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 752 } 753 754 /* 755 * Note: called only from the BH handler context, 756 * so we don't need to lock the hashes. 757 */ 758 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 759 const struct in6_addr *saddr, const struct in6_addr *daddr, 760 struct udp_table *udptable, int proto) 761 { 762 struct sock *sk, *first = NULL; 763 const struct udphdr *uh = udp_hdr(skb); 764 unsigned short hnum = ntohs(uh->dest); 765 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 766 unsigned int offset = offsetof(typeof(*sk), sk_node); 767 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 768 int dif = inet6_iif(skb); 769 int sdif = inet6_sdif(skb); 770 struct hlist_node *node; 771 struct sk_buff *nskb; 772 773 if (use_hash2) { 774 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 775 udptable->mask; 776 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 777 start_lookup: 778 hslot = &udptable->hash2[hash2]; 779 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 780 } 781 782 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 783 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 784 uh->source, saddr, dif, sdif, 785 hnum)) 786 continue; 787 /* If zero checksum and no_check is not on for 788 * the socket then skip it. 789 */ 790 if (!uh->check && !udp_sk(sk)->no_check6_rx) 791 continue; 792 if (!first) { 793 first = sk; 794 continue; 795 } 796 nskb = skb_clone(skb, GFP_ATOMIC); 797 if (unlikely(!nskb)) { 798 atomic_inc(&sk->sk_drops); 799 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 800 IS_UDPLITE(sk)); 801 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 802 IS_UDPLITE(sk)); 803 continue; 804 } 805 806 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 807 consume_skb(nskb); 808 } 809 810 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 811 if (use_hash2 && hash2 != hash2_any) { 812 hash2 = hash2_any; 813 goto start_lookup; 814 } 815 816 if (first) { 817 if (udpv6_queue_rcv_skb(first, skb) > 0) 818 consume_skb(skb); 819 } else { 820 kfree_skb(skb); 821 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 822 proto == IPPROTO_UDPLITE); 823 } 824 return 0; 825 } 826 827 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 828 { 829 if (udp_sk_rx_dst_set(sk, dst)) { 830 const struct rt6_info *rt = (const struct rt6_info *)dst; 831 832 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 833 } 834 } 835 836 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 837 * return code conversion for ip layer consumption 838 */ 839 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 840 struct udphdr *uh) 841 { 842 int ret; 843 844 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 845 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 846 ip6_compute_pseudo); 847 848 ret = udpv6_queue_rcv_skb(sk, skb); 849 850 /* a return value > 0 means to resubmit the input */ 851 if (ret > 0) 852 return ret; 853 return 0; 854 } 855 856 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 857 int proto) 858 { 859 const struct in6_addr *saddr, *daddr; 860 struct net *net = dev_net(skb->dev); 861 struct udphdr *uh; 862 struct sock *sk; 863 u32 ulen = 0; 864 865 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 866 goto discard; 867 868 saddr = &ipv6_hdr(skb)->saddr; 869 daddr = &ipv6_hdr(skb)->daddr; 870 uh = udp_hdr(skb); 871 872 ulen = ntohs(uh->len); 873 if (ulen > skb->len) 874 goto short_packet; 875 876 if (proto == IPPROTO_UDP) { 877 /* UDP validates ulen. */ 878 879 /* Check for jumbo payload */ 880 if (ulen == 0) 881 ulen = skb->len; 882 883 if (ulen < sizeof(*uh)) 884 goto short_packet; 885 886 if (ulen < skb->len) { 887 if (pskb_trim_rcsum(skb, ulen)) 888 goto short_packet; 889 saddr = &ipv6_hdr(skb)->saddr; 890 daddr = &ipv6_hdr(skb)->daddr; 891 uh = udp_hdr(skb); 892 } 893 } 894 895 if (udp6_csum_init(skb, uh, proto)) 896 goto csum_error; 897 898 /* Check if the socket is already available, e.g. due to early demux */ 899 sk = skb_steal_sock(skb); 900 if (sk) { 901 struct dst_entry *dst = skb_dst(skb); 902 int ret; 903 904 if (unlikely(sk->sk_rx_dst != dst)) 905 udp6_sk_rx_dst_set(sk, dst); 906 907 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 908 sock_put(sk); 909 goto report_csum_error; 910 } 911 912 ret = udp6_unicast_rcv_skb(sk, skb, uh); 913 sock_put(sk); 914 return ret; 915 } 916 917 /* 918 * Multicast receive code 919 */ 920 if (ipv6_addr_is_multicast(daddr)) 921 return __udp6_lib_mcast_deliver(net, skb, 922 saddr, daddr, udptable, proto); 923 924 /* Unicast */ 925 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 926 if (sk) { 927 if (!uh->check && !udp_sk(sk)->no_check6_rx) 928 goto report_csum_error; 929 return udp6_unicast_rcv_skb(sk, skb, uh); 930 } 931 932 if (!uh->check) 933 goto report_csum_error; 934 935 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 936 goto discard; 937 938 if (udp_lib_checksum_complete(skb)) 939 goto csum_error; 940 941 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 942 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 943 944 kfree_skb(skb); 945 return 0; 946 947 short_packet: 948 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 949 proto == IPPROTO_UDPLITE ? "-Lite" : "", 950 saddr, ntohs(uh->source), 951 ulen, skb->len, 952 daddr, ntohs(uh->dest)); 953 goto discard; 954 955 report_csum_error: 956 udp6_csum_zero_error(skb); 957 csum_error: 958 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 959 discard: 960 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 961 kfree_skb(skb); 962 return 0; 963 } 964 965 966 static struct sock *__udp6_lib_demux_lookup(struct net *net, 967 __be16 loc_port, const struct in6_addr *loc_addr, 968 __be16 rmt_port, const struct in6_addr *rmt_addr, 969 int dif, int sdif) 970 { 971 unsigned short hnum = ntohs(loc_port); 972 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 973 unsigned int slot2 = hash2 & udp_table.mask; 974 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 975 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 976 struct sock *sk; 977 978 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 979 if (sk->sk_state == TCP_ESTABLISHED && 980 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 981 return sk; 982 /* Only check first socket in chain */ 983 break; 984 } 985 return NULL; 986 } 987 988 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb) 989 { 990 struct net *net = dev_net(skb->dev); 991 const struct udphdr *uh; 992 struct sock *sk; 993 struct dst_entry *dst; 994 int dif = skb->dev->ifindex; 995 int sdif = inet6_sdif(skb); 996 997 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 998 sizeof(struct udphdr))) 999 return; 1000 1001 uh = udp_hdr(skb); 1002 1003 if (skb->pkt_type == PACKET_HOST) 1004 sk = __udp6_lib_demux_lookup(net, uh->dest, 1005 &ipv6_hdr(skb)->daddr, 1006 uh->source, &ipv6_hdr(skb)->saddr, 1007 dif, sdif); 1008 else 1009 return; 1010 1011 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1012 return; 1013 1014 skb->sk = sk; 1015 skb->destructor = sock_efree; 1016 dst = READ_ONCE(sk->sk_rx_dst); 1017 1018 if (dst) 1019 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 1020 if (dst) { 1021 /* set noref for now. 1022 * any place which wants to hold dst has to call 1023 * dst_hold_safe() 1024 */ 1025 skb_dst_set_noref(skb, dst); 1026 } 1027 } 1028 1029 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1030 { 1031 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1032 } 1033 1034 /* 1035 * Throw away all pending data and cancel the corking. Socket is locked. 1036 */ 1037 static void udp_v6_flush_pending_frames(struct sock *sk) 1038 { 1039 struct udp_sock *up = udp_sk(sk); 1040 1041 if (up->pending == AF_INET) 1042 udp_flush_pending_frames(sk); 1043 else if (up->pending) { 1044 up->len = 0; 1045 up->pending = 0; 1046 ip6_flush_pending_frames(sk); 1047 } 1048 } 1049 1050 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1051 int addr_len) 1052 { 1053 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1054 return -EINVAL; 1055 /* The following checks are replicated from __ip6_datagram_connect() 1056 * and intended to prevent BPF program called below from accessing 1057 * bytes that are out of the bound specified by user in addr_len. 1058 */ 1059 if (uaddr->sa_family == AF_INET) { 1060 if (__ipv6_only_sock(sk)) 1061 return -EAFNOSUPPORT; 1062 return udp_pre_connect(sk, uaddr, addr_len); 1063 } 1064 1065 if (addr_len < SIN6_LEN_RFC2133) 1066 return -EINVAL; 1067 1068 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1069 } 1070 1071 /** 1072 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1073 * @sk: socket we are sending on 1074 * @skb: sk_buff containing the filled-in UDP header 1075 * (checksum field must be zeroed out) 1076 */ 1077 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1078 const struct in6_addr *saddr, 1079 const struct in6_addr *daddr, int len) 1080 { 1081 unsigned int offset; 1082 struct udphdr *uh = udp_hdr(skb); 1083 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1084 __wsum csum = 0; 1085 1086 if (!frags) { 1087 /* Only one fragment on the socket. */ 1088 skb->csum_start = skb_transport_header(skb) - skb->head; 1089 skb->csum_offset = offsetof(struct udphdr, check); 1090 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1091 } else { 1092 /* 1093 * HW-checksum won't work as there are two or more 1094 * fragments on the socket so that all csums of sk_buffs 1095 * should be together 1096 */ 1097 offset = skb_transport_offset(skb); 1098 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1099 csum = skb->csum; 1100 1101 skb->ip_summed = CHECKSUM_NONE; 1102 1103 do { 1104 csum = csum_add(csum, frags->csum); 1105 } while ((frags = frags->next)); 1106 1107 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1108 csum); 1109 if (uh->check == 0) 1110 uh->check = CSUM_MANGLED_0; 1111 } 1112 } 1113 1114 /* 1115 * Sending 1116 */ 1117 1118 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1119 struct inet_cork *cork) 1120 { 1121 struct sock *sk = skb->sk; 1122 struct udphdr *uh; 1123 int err = 0; 1124 int is_udplite = IS_UDPLITE(sk); 1125 __wsum csum = 0; 1126 int offset = skb_transport_offset(skb); 1127 int len = skb->len - offset; 1128 1129 /* 1130 * Create a UDP header 1131 */ 1132 uh = udp_hdr(skb); 1133 uh->source = fl6->fl6_sport; 1134 uh->dest = fl6->fl6_dport; 1135 uh->len = htons(len); 1136 uh->check = 0; 1137 1138 if (cork->gso_size) { 1139 const int hlen = skb_network_header_len(skb) + 1140 sizeof(struct udphdr); 1141 1142 if (hlen + cork->gso_size > cork->fragsize) { 1143 kfree_skb(skb); 1144 return -EINVAL; 1145 } 1146 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 1147 kfree_skb(skb); 1148 return -EINVAL; 1149 } 1150 if (udp_sk(sk)->no_check6_tx) { 1151 kfree_skb(skb); 1152 return -EINVAL; 1153 } 1154 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1155 dst_xfrm(skb_dst(skb))) { 1156 kfree_skb(skb); 1157 return -EIO; 1158 } 1159 1160 skb_shinfo(skb)->gso_size = cork->gso_size; 1161 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1162 goto csum_partial; 1163 } 1164 1165 if (is_udplite) 1166 csum = udplite_csum(skb); 1167 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1168 skb->ip_summed = CHECKSUM_NONE; 1169 goto send; 1170 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1171 csum_partial: 1172 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1173 goto send; 1174 } else 1175 csum = udp_csum(skb); 1176 1177 /* add protocol-dependent pseudo-header */ 1178 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1179 len, fl6->flowi6_proto, csum); 1180 if (uh->check == 0) 1181 uh->check = CSUM_MANGLED_0; 1182 1183 send: 1184 err = ip6_send_skb(skb); 1185 if (err) { 1186 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1187 UDP6_INC_STATS(sock_net(sk), 1188 UDP_MIB_SNDBUFERRORS, is_udplite); 1189 err = 0; 1190 } 1191 } else { 1192 UDP6_INC_STATS(sock_net(sk), 1193 UDP_MIB_OUTDATAGRAMS, is_udplite); 1194 } 1195 return err; 1196 } 1197 1198 static int udp_v6_push_pending_frames(struct sock *sk) 1199 { 1200 struct sk_buff *skb; 1201 struct udp_sock *up = udp_sk(sk); 1202 struct flowi6 fl6; 1203 int err = 0; 1204 1205 if (up->pending == AF_INET) 1206 return udp_push_pending_frames(sk); 1207 1208 /* ip6_finish_skb will release the cork, so make a copy of 1209 * fl6 here. 1210 */ 1211 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1212 1213 skb = ip6_finish_skb(sk); 1214 if (!skb) 1215 goto out; 1216 1217 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1218 1219 out: 1220 up->len = 0; 1221 up->pending = 0; 1222 return err; 1223 } 1224 1225 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1226 { 1227 struct ipv6_txoptions opt_space; 1228 struct udp_sock *up = udp_sk(sk); 1229 struct inet_sock *inet = inet_sk(sk); 1230 struct ipv6_pinfo *np = inet6_sk(sk); 1231 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1232 struct in6_addr *daddr, *final_p, final; 1233 struct ipv6_txoptions *opt = NULL; 1234 struct ipv6_txoptions *opt_to_free = NULL; 1235 struct ip6_flowlabel *flowlabel = NULL; 1236 struct flowi6 fl6; 1237 struct dst_entry *dst; 1238 struct ipcm6_cookie ipc6; 1239 int addr_len = msg->msg_namelen; 1240 bool connected = false; 1241 int ulen = len; 1242 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1243 int err; 1244 int is_udplite = IS_UDPLITE(sk); 1245 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1246 1247 ipcm6_init(&ipc6); 1248 ipc6.gso_size = up->gso_size; 1249 ipc6.sockc.tsflags = sk->sk_tsflags; 1250 1251 /* destination address check */ 1252 if (sin6) { 1253 if (addr_len < offsetof(struct sockaddr, sa_data)) 1254 return -EINVAL; 1255 1256 switch (sin6->sin6_family) { 1257 case AF_INET6: 1258 if (addr_len < SIN6_LEN_RFC2133) 1259 return -EINVAL; 1260 daddr = &sin6->sin6_addr; 1261 if (ipv6_addr_any(daddr) && 1262 ipv6_addr_v4mapped(&np->saddr)) 1263 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1264 daddr); 1265 break; 1266 case AF_INET: 1267 goto do_udp_sendmsg; 1268 case AF_UNSPEC: 1269 msg->msg_name = sin6 = NULL; 1270 msg->msg_namelen = addr_len = 0; 1271 daddr = NULL; 1272 break; 1273 default: 1274 return -EINVAL; 1275 } 1276 } else if (!up->pending) { 1277 if (sk->sk_state != TCP_ESTABLISHED) 1278 return -EDESTADDRREQ; 1279 daddr = &sk->sk_v6_daddr; 1280 } else 1281 daddr = NULL; 1282 1283 if (daddr) { 1284 if (ipv6_addr_v4mapped(daddr)) { 1285 struct sockaddr_in sin; 1286 sin.sin_family = AF_INET; 1287 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1288 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1289 msg->msg_name = &sin; 1290 msg->msg_namelen = sizeof(sin); 1291 do_udp_sendmsg: 1292 if (__ipv6_only_sock(sk)) 1293 return -ENETUNREACH; 1294 return udp_sendmsg(sk, msg, len); 1295 } 1296 } 1297 1298 if (up->pending == AF_INET) 1299 return udp_sendmsg(sk, msg, len); 1300 1301 /* Rough check on arithmetic overflow, 1302 better check is made in ip6_append_data(). 1303 */ 1304 if (len > INT_MAX - sizeof(struct udphdr)) 1305 return -EMSGSIZE; 1306 1307 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1308 if (up->pending) { 1309 /* 1310 * There are pending frames. 1311 * The socket lock must be held while it's corked. 1312 */ 1313 lock_sock(sk); 1314 if (likely(up->pending)) { 1315 if (unlikely(up->pending != AF_INET6)) { 1316 release_sock(sk); 1317 return -EAFNOSUPPORT; 1318 } 1319 dst = NULL; 1320 goto do_append_data; 1321 } 1322 release_sock(sk); 1323 } 1324 ulen += sizeof(struct udphdr); 1325 1326 memset(&fl6, 0, sizeof(fl6)); 1327 1328 if (sin6) { 1329 if (sin6->sin6_port == 0) 1330 return -EINVAL; 1331 1332 fl6.fl6_dport = sin6->sin6_port; 1333 daddr = &sin6->sin6_addr; 1334 1335 if (np->sndflow) { 1336 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1337 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1338 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1339 if (!flowlabel) 1340 return -EINVAL; 1341 } 1342 } 1343 1344 /* 1345 * Otherwise it will be difficult to maintain 1346 * sk->sk_dst_cache. 1347 */ 1348 if (sk->sk_state == TCP_ESTABLISHED && 1349 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1350 daddr = &sk->sk_v6_daddr; 1351 1352 if (addr_len >= sizeof(struct sockaddr_in6) && 1353 sin6->sin6_scope_id && 1354 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1355 fl6.flowi6_oif = sin6->sin6_scope_id; 1356 } else { 1357 if (sk->sk_state != TCP_ESTABLISHED) 1358 return -EDESTADDRREQ; 1359 1360 fl6.fl6_dport = inet->inet_dport; 1361 daddr = &sk->sk_v6_daddr; 1362 fl6.flowlabel = np->flow_label; 1363 connected = true; 1364 } 1365 1366 if (!fl6.flowi6_oif) 1367 fl6.flowi6_oif = sk->sk_bound_dev_if; 1368 1369 if (!fl6.flowi6_oif) 1370 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1371 1372 fl6.flowi6_mark = sk->sk_mark; 1373 fl6.flowi6_uid = sk->sk_uid; 1374 1375 if (msg->msg_controllen) { 1376 opt = &opt_space; 1377 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1378 opt->tot_len = sizeof(*opt); 1379 ipc6.opt = opt; 1380 1381 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1382 if (err > 0) 1383 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1384 &ipc6); 1385 if (err < 0) { 1386 fl6_sock_release(flowlabel); 1387 return err; 1388 } 1389 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1390 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1391 if (!flowlabel) 1392 return -EINVAL; 1393 } 1394 if (!(opt->opt_nflen|opt->opt_flen)) 1395 opt = NULL; 1396 connected = false; 1397 } 1398 if (!opt) { 1399 opt = txopt_get(np); 1400 opt_to_free = opt; 1401 } 1402 if (flowlabel) 1403 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1404 opt = ipv6_fixup_options(&opt_space, opt); 1405 ipc6.opt = opt; 1406 1407 fl6.flowi6_proto = sk->sk_protocol; 1408 fl6.daddr = *daddr; 1409 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1410 fl6.saddr = np->saddr; 1411 fl6.fl6_sport = inet->inet_sport; 1412 1413 if (cgroup_bpf_enabled && !connected) { 1414 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1415 (struct sockaddr *)sin6, &fl6.saddr); 1416 if (err) 1417 goto out_no_dst; 1418 if (sin6) { 1419 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1420 /* BPF program rewrote IPv6-only by IPv4-mapped 1421 * IPv6. It's currently unsupported. 1422 */ 1423 err = -ENOTSUPP; 1424 goto out_no_dst; 1425 } 1426 if (sin6->sin6_port == 0) { 1427 /* BPF program set invalid port. Reject it. */ 1428 err = -EINVAL; 1429 goto out_no_dst; 1430 } 1431 fl6.fl6_dport = sin6->sin6_port; 1432 fl6.daddr = sin6->sin6_addr; 1433 } 1434 } 1435 1436 if (ipv6_addr_any(&fl6.daddr)) 1437 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1438 1439 final_p = fl6_update_dst(&fl6, opt, &final); 1440 if (final_p) 1441 connected = false; 1442 1443 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1444 fl6.flowi6_oif = np->mcast_oif; 1445 connected = false; 1446 } else if (!fl6.flowi6_oif) 1447 fl6.flowi6_oif = np->ucast_oif; 1448 1449 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1450 1451 if (ipc6.tclass < 0) 1452 ipc6.tclass = np->tclass; 1453 1454 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1455 1456 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1457 if (IS_ERR(dst)) { 1458 err = PTR_ERR(dst); 1459 dst = NULL; 1460 goto out; 1461 } 1462 1463 if (ipc6.hlimit < 0) 1464 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1465 1466 if (msg->msg_flags&MSG_CONFIRM) 1467 goto do_confirm; 1468 back_from_confirm: 1469 1470 /* Lockless fast path for the non-corking case */ 1471 if (!corkreq) { 1472 struct inet_cork_full cork; 1473 struct sk_buff *skb; 1474 1475 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1476 sizeof(struct udphdr), &ipc6, 1477 &fl6, (struct rt6_info *)dst, 1478 msg->msg_flags, &cork); 1479 err = PTR_ERR(skb); 1480 if (!IS_ERR_OR_NULL(skb)) 1481 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1482 goto out; 1483 } 1484 1485 lock_sock(sk); 1486 if (unlikely(up->pending)) { 1487 /* The socket is already corked while preparing it. */ 1488 /* ... which is an evident application bug. --ANK */ 1489 release_sock(sk); 1490 1491 net_dbg_ratelimited("udp cork app bug 2\n"); 1492 err = -EINVAL; 1493 goto out; 1494 } 1495 1496 up->pending = AF_INET6; 1497 1498 do_append_data: 1499 if (ipc6.dontfrag < 0) 1500 ipc6.dontfrag = np->dontfrag; 1501 up->len += ulen; 1502 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1503 &ipc6, &fl6, (struct rt6_info *)dst, 1504 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1505 if (err) 1506 udp_v6_flush_pending_frames(sk); 1507 else if (!corkreq) 1508 err = udp_v6_push_pending_frames(sk); 1509 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1510 up->pending = 0; 1511 1512 if (err > 0) 1513 err = np->recverr ? net_xmit_errno(err) : 0; 1514 release_sock(sk); 1515 1516 out: 1517 dst_release(dst); 1518 out_no_dst: 1519 fl6_sock_release(flowlabel); 1520 txopt_put(opt_to_free); 1521 if (!err) 1522 return len; 1523 /* 1524 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1525 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1526 * we don't have a good statistic (IpOutDiscards but it can be too many 1527 * things). We could add another new stat but at least for now that 1528 * seems like overkill. 1529 */ 1530 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1531 UDP6_INC_STATS(sock_net(sk), 1532 UDP_MIB_SNDBUFERRORS, is_udplite); 1533 } 1534 return err; 1535 1536 do_confirm: 1537 if (msg->msg_flags & MSG_PROBE) 1538 dst_confirm_neigh(dst, &fl6.daddr); 1539 if (!(msg->msg_flags&MSG_PROBE) || len) 1540 goto back_from_confirm; 1541 err = 0; 1542 goto out; 1543 } 1544 1545 void udpv6_destroy_sock(struct sock *sk) 1546 { 1547 struct udp_sock *up = udp_sk(sk); 1548 lock_sock(sk); 1549 udp_v6_flush_pending_frames(sk); 1550 release_sock(sk); 1551 1552 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1553 if (up->encap_type) { 1554 void (*encap_destroy)(struct sock *sk); 1555 encap_destroy = READ_ONCE(up->encap_destroy); 1556 if (encap_destroy) 1557 encap_destroy(sk); 1558 } 1559 if (up->encap_enabled) 1560 static_branch_dec(&udpv6_encap_needed_key); 1561 } 1562 1563 inet6_destroy_sock(sk); 1564 } 1565 1566 /* 1567 * Socket option code for UDP 1568 */ 1569 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1570 char __user *optval, unsigned int optlen) 1571 { 1572 if (level == SOL_UDP || level == SOL_UDPLITE) 1573 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1574 udp_v6_push_pending_frames); 1575 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1576 } 1577 1578 #ifdef CONFIG_COMPAT 1579 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1580 char __user *optval, unsigned int optlen) 1581 { 1582 if (level == SOL_UDP || level == SOL_UDPLITE) 1583 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1584 udp_v6_push_pending_frames); 1585 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1586 } 1587 #endif 1588 1589 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1590 char __user *optval, int __user *optlen) 1591 { 1592 if (level == SOL_UDP || level == SOL_UDPLITE) 1593 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1594 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1595 } 1596 1597 #ifdef CONFIG_COMPAT 1598 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1599 char __user *optval, int __user *optlen) 1600 { 1601 if (level == SOL_UDP || level == SOL_UDPLITE) 1602 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1603 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1604 } 1605 #endif 1606 1607 /* thinking of making this const? Don't. 1608 * early_demux can change based on sysctl. 1609 */ 1610 static struct inet6_protocol udpv6_protocol = { 1611 .early_demux = udp_v6_early_demux, 1612 .early_demux_handler = udp_v6_early_demux, 1613 .handler = udpv6_rcv, 1614 .err_handler = udpv6_err, 1615 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1616 }; 1617 1618 /* ------------------------------------------------------------------------ */ 1619 #ifdef CONFIG_PROC_FS 1620 int udp6_seq_show(struct seq_file *seq, void *v) 1621 { 1622 if (v == SEQ_START_TOKEN) { 1623 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1624 } else { 1625 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1626 struct inet_sock *inet = inet_sk(v); 1627 __u16 srcp = ntohs(inet->inet_sport); 1628 __u16 destp = ntohs(inet->inet_dport); 1629 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1630 udp_rqueue_get(v), bucket); 1631 } 1632 return 0; 1633 } 1634 1635 const struct seq_operations udp6_seq_ops = { 1636 .start = udp_seq_start, 1637 .next = udp_seq_next, 1638 .stop = udp_seq_stop, 1639 .show = udp6_seq_show, 1640 }; 1641 EXPORT_SYMBOL(udp6_seq_ops); 1642 1643 static struct udp_seq_afinfo udp6_seq_afinfo = { 1644 .family = AF_INET6, 1645 .udp_table = &udp_table, 1646 }; 1647 1648 int __net_init udp6_proc_init(struct net *net) 1649 { 1650 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1651 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1652 return -ENOMEM; 1653 return 0; 1654 } 1655 1656 void udp6_proc_exit(struct net *net) 1657 { 1658 remove_proc_entry("udp6", net->proc_net); 1659 } 1660 #endif /* CONFIG_PROC_FS */ 1661 1662 /* ------------------------------------------------------------------------ */ 1663 1664 struct proto udpv6_prot = { 1665 .name = "UDPv6", 1666 .owner = THIS_MODULE, 1667 .close = udp_lib_close, 1668 .pre_connect = udpv6_pre_connect, 1669 .connect = ip6_datagram_connect, 1670 .disconnect = udp_disconnect, 1671 .ioctl = udp_ioctl, 1672 .init = udp_init_sock, 1673 .destroy = udpv6_destroy_sock, 1674 .setsockopt = udpv6_setsockopt, 1675 .getsockopt = udpv6_getsockopt, 1676 .sendmsg = udpv6_sendmsg, 1677 .recvmsg = udpv6_recvmsg, 1678 .release_cb = ip6_datagram_release_cb, 1679 .hash = udp_lib_hash, 1680 .unhash = udp_lib_unhash, 1681 .rehash = udp_v6_rehash, 1682 .get_port = udp_v6_get_port, 1683 .memory_allocated = &udp_memory_allocated, 1684 .sysctl_mem = sysctl_udp_mem, 1685 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1686 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1687 .obj_size = sizeof(struct udp6_sock), 1688 .h.udp_table = &udp_table, 1689 #ifdef CONFIG_COMPAT 1690 .compat_setsockopt = compat_udpv6_setsockopt, 1691 .compat_getsockopt = compat_udpv6_getsockopt, 1692 #endif 1693 .diag_destroy = udp_abort, 1694 }; 1695 1696 static struct inet_protosw udpv6_protosw = { 1697 .type = SOCK_DGRAM, 1698 .protocol = IPPROTO_UDP, 1699 .prot = &udpv6_prot, 1700 .ops = &inet6_dgram_ops, 1701 .flags = INET_PROTOSW_PERMANENT, 1702 }; 1703 1704 int __init udpv6_init(void) 1705 { 1706 int ret; 1707 1708 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1709 if (ret) 1710 goto out; 1711 1712 ret = inet6_register_protosw(&udpv6_protosw); 1713 if (ret) 1714 goto out_udpv6_protocol; 1715 out: 1716 return ret; 1717 1718 out_udpv6_protocol: 1719 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1720 goto out; 1721 } 1722 1723 void udpv6_exit(void) 1724 { 1725 inet6_unregister_protosw(&udpv6_protosw); 1726 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1727 } 1728