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