1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * UDP over IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on linux/ipv4/udp.c 10 * 11 * Fixes: 12 * Hideaki YOSHIFUJI : sin6_scope_id support 13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 15 * a single port at the same time. 16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. 18 */ 19 20 #include <linux/errno.h> 21 #include <linux/types.h> 22 #include <linux/socket.h> 23 #include <linux/sockios.h> 24 #include <linux/net.h> 25 #include <linux/in6.h> 26 #include <linux/netdevice.h> 27 #include <linux/if_arp.h> 28 #include <linux/ipv6.h> 29 #include <linux/icmpv6.h> 30 #include <linux/init.h> 31 #include <linux/module.h> 32 #include <linux/skbuff.h> 33 #include <linux/slab.h> 34 #include <linux/uaccess.h> 35 #include <linux/indirect_call_wrapper.h> 36 37 #include <net/addrconf.h> 38 #include <net/ndisc.h> 39 #include <net/protocol.h> 40 #include <net/transp_v6.h> 41 #include <net/ip6_route.h> 42 #include <net/raw.h> 43 #include <net/tcp_states.h> 44 #include <net/ip6_checksum.h> 45 #include <net/ip6_tunnel.h> 46 #include <net/xfrm.h> 47 #include <net/inet_hashtables.h> 48 #include <net/inet6_hashtables.h> 49 #include <net/busy_poll.h> 50 #include <net/sock_reuseport.h> 51 52 #include <linux/proc_fs.h> 53 #include <linux/seq_file.h> 54 #include <trace/events/skb.h> 55 #include "udp_impl.h" 56 57 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb) 58 { 59 #if defined(CONFIG_NET_L3_MASTER_DEV) 60 if (!net->ipv4.sysctl_udp_l3mdev_accept && 61 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags)) 62 return true; 63 #endif 64 return false; 65 } 66 67 static u32 udp6_ehashfn(const struct net *net, 68 const struct in6_addr *laddr, 69 const u16 lport, 70 const struct in6_addr *faddr, 71 const __be16 fport) 72 { 73 static u32 udp6_ehash_secret __read_mostly; 74 static u32 udp_ipv6_hash_secret __read_mostly; 75 76 u32 lhash, fhash; 77 78 net_get_random_once(&udp6_ehash_secret, 79 sizeof(udp6_ehash_secret)); 80 net_get_random_once(&udp_ipv6_hash_secret, 81 sizeof(udp_ipv6_hash_secret)); 82 83 lhash = (__force u32)laddr->s6_addr32[3]; 84 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 85 86 return __inet6_ehashfn(lhash, lport, fhash, fport, 87 udp_ipv6_hash_secret + net_hash_mix(net)); 88 } 89 90 int udp_v6_get_port(struct sock *sk, unsigned short snum) 91 { 92 unsigned int hash2_nulladdr = 93 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 94 unsigned int hash2_partial = 95 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 96 97 /* precompute partial secondary hash */ 98 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 99 return udp_lib_get_port(sk, snum, hash2_nulladdr); 100 } 101 102 void udp_v6_rehash(struct sock *sk) 103 { 104 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 105 &sk->sk_v6_rcv_saddr, 106 inet_sk(sk)->inet_num); 107 108 udp_lib_rehash(sk, new_hash); 109 } 110 111 static int compute_score(struct sock *sk, struct net *net, 112 const struct in6_addr *saddr, __be16 sport, 113 const struct in6_addr *daddr, unsigned short hnum, 114 int dif, int sdif, bool exact_dif) 115 { 116 int score; 117 struct inet_sock *inet; 118 bool dev_match; 119 120 if (!net_eq(sock_net(sk), net) || 121 udp_sk(sk)->udp_port_hash != hnum || 122 sk->sk_family != PF_INET6) 123 return -1; 124 125 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 126 return -1; 127 128 score = 0; 129 inet = inet_sk(sk); 130 131 if (inet->inet_dport) { 132 if (inet->inet_dport != sport) 133 return -1; 134 score++; 135 } 136 137 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 138 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 139 return -1; 140 score++; 141 } 142 143 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif); 144 if (!dev_match) 145 return -1; 146 score++; 147 148 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 149 score++; 150 151 return score; 152 } 153 154 /* called with rcu_read_lock() */ 155 static struct sock *udp6_lib_lookup2(struct net *net, 156 const struct in6_addr *saddr, __be16 sport, 157 const struct in6_addr *daddr, unsigned int hnum, 158 int dif, int sdif, bool exact_dif, 159 struct udp_hslot *hslot2, struct sk_buff *skb) 160 { 161 struct sock *sk, *result; 162 int score, badness; 163 u32 hash = 0; 164 165 result = NULL; 166 badness = -1; 167 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 168 score = compute_score(sk, net, saddr, sport, 169 daddr, hnum, dif, sdif, exact_dif); 170 if (score > badness) { 171 if (sk->sk_reuseport) { 172 hash = udp6_ehashfn(net, daddr, hnum, 173 saddr, sport); 174 175 result = reuseport_select_sock(sk, hash, skb, 176 sizeof(struct udphdr)); 177 if (result) 178 return result; 179 } 180 result = sk; 181 badness = score; 182 } 183 } 184 return result; 185 } 186 187 /* rcu_read_lock() must be held */ 188 struct sock *__udp6_lib_lookup(struct net *net, 189 const struct in6_addr *saddr, __be16 sport, 190 const struct in6_addr *daddr, __be16 dport, 191 int dif, int sdif, struct udp_table *udptable, 192 struct sk_buff *skb) 193 { 194 unsigned short hnum = ntohs(dport); 195 unsigned int hash2, slot2; 196 struct udp_hslot *hslot2; 197 struct sock *result; 198 bool exact_dif = udp6_lib_exact_dif_match(net, skb); 199 200 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 201 slot2 = hash2 & udptable->mask; 202 hslot2 = &udptable->hash2[slot2]; 203 204 result = udp6_lib_lookup2(net, saddr, sport, 205 daddr, hnum, dif, sdif, exact_dif, 206 hslot2, skb); 207 if (!result) { 208 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 209 slot2 = hash2 & udptable->mask; 210 211 hslot2 = &udptable->hash2[slot2]; 212 213 result = udp6_lib_lookup2(net, saddr, sport, 214 &in6addr_any, hnum, dif, sdif, 215 exact_dif, hslot2, 216 skb); 217 } 218 if (unlikely(IS_ERR(result))) 219 return NULL; 220 return result; 221 } 222 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 223 224 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 225 __be16 sport, __be16 dport, 226 struct udp_table *udptable) 227 { 228 const struct ipv6hdr *iph = ipv6_hdr(skb); 229 230 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 231 &iph->daddr, dport, inet6_iif(skb), 232 inet6_sdif(skb), udptable, skb); 233 } 234 235 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, 236 __be16 sport, __be16 dport) 237 { 238 const struct ipv6hdr *iph = ipv6_hdr(skb); 239 240 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 241 &iph->daddr, dport, inet6_iif(skb), 242 inet6_sdif(skb), &udp_table, NULL); 243 } 244 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); 245 246 /* Must be called under rcu_read_lock(). 247 * Does increment socket refcount. 248 */ 249 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 250 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 251 const struct in6_addr *daddr, __be16 dport, int dif) 252 { 253 struct sock *sk; 254 255 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 256 dif, 0, &udp_table, NULL); 257 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 258 sk = NULL; 259 return sk; 260 } 261 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 262 #endif 263 264 /* do not use the scratch area len for jumbogram: their length execeeds the 265 * scratch area space; note that the IP6CB flags is still in the first 266 * cacheline, so checking for jumbograms is cheap 267 */ 268 static int udp6_skb_len(struct sk_buff *skb) 269 { 270 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 271 } 272 273 /* 274 * This should be easy, if there is something there we 275 * return it, otherwise we block. 276 */ 277 278 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 279 int noblock, int flags, int *addr_len) 280 { 281 struct ipv6_pinfo *np = inet6_sk(sk); 282 struct inet_sock *inet = inet_sk(sk); 283 struct sk_buff *skb; 284 unsigned int ulen, copied; 285 int off, err, peeking = flags & MSG_PEEK; 286 int is_udplite = IS_UDPLITE(sk); 287 struct udp_mib __percpu *mib; 288 bool checksum_valid = false; 289 int is_udp4; 290 291 if (flags & MSG_ERRQUEUE) 292 return ipv6_recv_error(sk, msg, len, addr_len); 293 294 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 295 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 296 297 try_again: 298 off = sk_peek_offset(sk, flags); 299 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 300 if (!skb) 301 return err; 302 303 ulen = udp6_skb_len(skb); 304 copied = len; 305 if (copied > ulen - off) 306 copied = ulen - off; 307 else if (copied < ulen) 308 msg->msg_flags |= MSG_TRUNC; 309 310 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 311 mib = __UDPX_MIB(sk, is_udp4); 312 313 /* 314 * If checksum is needed at all, try to do it while copying the 315 * data. If the data is truncated, or if we only want a partial 316 * coverage checksum (UDP-Lite), do it before the copy. 317 */ 318 319 if (copied < ulen || peeking || 320 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 321 checksum_valid = udp_skb_csum_unnecessary(skb) || 322 !__udp_lib_checksum_complete(skb); 323 if (!checksum_valid) 324 goto csum_copy_err; 325 } 326 327 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 328 if (udp_skb_is_linear(skb)) 329 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 330 else 331 err = skb_copy_datagram_msg(skb, off, msg, copied); 332 } else { 333 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 334 if (err == -EINVAL) 335 goto csum_copy_err; 336 } 337 if (unlikely(err)) { 338 if (!peeking) { 339 atomic_inc(&sk->sk_drops); 340 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 341 } 342 kfree_skb(skb); 343 return err; 344 } 345 if (!peeking) 346 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 347 348 sock_recv_ts_and_drops(msg, sk, skb); 349 350 /* Copy the address. */ 351 if (msg->msg_name) { 352 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 353 sin6->sin6_family = AF_INET6; 354 sin6->sin6_port = udp_hdr(skb)->source; 355 sin6->sin6_flowinfo = 0; 356 357 if (is_udp4) { 358 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 359 &sin6->sin6_addr); 360 sin6->sin6_scope_id = 0; 361 } else { 362 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 363 sin6->sin6_scope_id = 364 ipv6_iface_scope_id(&sin6->sin6_addr, 365 inet6_iif(skb)); 366 } 367 *addr_len = sizeof(*sin6); 368 369 if (cgroup_bpf_enabled) 370 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 371 (struct sockaddr *)sin6); 372 } 373 374 if (udp_sk(sk)->gro_enabled) 375 udp_cmsg_recv(msg, sk, skb); 376 377 if (np->rxopt.all) 378 ip6_datagram_recv_common_ctl(sk, msg, skb); 379 380 if (is_udp4) { 381 if (inet->cmsg_flags) 382 ip_cmsg_recv_offset(msg, sk, skb, 383 sizeof(struct udphdr), off); 384 } else { 385 if (np->rxopt.all) 386 ip6_datagram_recv_specific_ctl(sk, msg, skb); 387 } 388 389 err = copied; 390 if (flags & MSG_TRUNC) 391 err = ulen; 392 393 skb_consume_udp(sk, skb, peeking ? -err : err); 394 return err; 395 396 csum_copy_err: 397 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 398 udp_skb_destructor)) { 399 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 400 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 401 } 402 kfree_skb(skb); 403 404 /* starting over for a new packet, but check if we need to yield */ 405 cond_resched(); 406 msg->msg_flags &= ~MSG_TRUNC; 407 goto try_again; 408 } 409 410 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 411 void udpv6_encap_enable(void) 412 { 413 static_branch_inc(&udpv6_encap_needed_key); 414 } 415 EXPORT_SYMBOL(udpv6_encap_enable); 416 417 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 418 * through error handlers in encapsulations looking for a match. 419 */ 420 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 421 struct inet6_skb_parm *opt, 422 u8 type, u8 code, int offset, __be32 info) 423 { 424 int i; 425 426 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 427 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 428 u8 type, u8 code, int offset, __be32 info); 429 const struct ip6_tnl_encap_ops *encap; 430 431 encap = rcu_dereference(ip6tun_encaps[i]); 432 if (!encap) 433 continue; 434 handler = encap->err_handler; 435 if (handler && !handler(skb, opt, type, code, offset, info)) 436 return 0; 437 } 438 439 return -ENOENT; 440 } 441 442 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 443 * reversing source and destination port: this will match tunnels that force the 444 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 445 * lwtunnels might actually break this assumption by being configured with 446 * different destination ports on endpoints, in this case we won't be able to 447 * trace ICMP messages back to them. 448 * 449 * If this doesn't match any socket, probe tunnels with arbitrary destination 450 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 451 * we've sent packets to won't necessarily match the local destination port. 452 * 453 * Then ask the tunnel implementation to match the error against a valid 454 * association. 455 * 456 * Return an error if we can't find a match, the socket if we need further 457 * processing, zero otherwise. 458 */ 459 static struct sock *__udp6_lib_err_encap(struct net *net, 460 const struct ipv6hdr *hdr, int offset, 461 struct udphdr *uh, 462 struct udp_table *udptable, 463 struct sk_buff *skb, 464 struct inet6_skb_parm *opt, 465 u8 type, u8 code, __be32 info) 466 { 467 int network_offset, transport_offset; 468 struct sock *sk; 469 470 network_offset = skb_network_offset(skb); 471 transport_offset = skb_transport_offset(skb); 472 473 /* Network header needs to point to the outer IPv6 header inside ICMP */ 474 skb_reset_network_header(skb); 475 476 /* Transport header needs to point to the UDP header */ 477 skb_set_transport_header(skb, offset); 478 479 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 480 &hdr->saddr, uh->dest, 481 inet6_iif(skb), 0, udptable, skb); 482 if (sk) { 483 int (*lookup)(struct sock *sk, struct sk_buff *skb); 484 struct udp_sock *up = udp_sk(sk); 485 486 lookup = READ_ONCE(up->encap_err_lookup); 487 if (!lookup || lookup(sk, skb)) 488 sk = NULL; 489 } 490 491 if (!sk) { 492 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 493 offset, info)); 494 } 495 496 skb_set_transport_header(skb, transport_offset); 497 skb_set_network_header(skb, network_offset); 498 499 return sk; 500 } 501 502 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 503 u8 type, u8 code, int offset, __be32 info, 504 struct udp_table *udptable) 505 { 506 struct ipv6_pinfo *np; 507 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 508 const struct in6_addr *saddr = &hdr->saddr; 509 const struct in6_addr *daddr = &hdr->daddr; 510 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 511 bool tunnel = false; 512 struct sock *sk; 513 int harderr; 514 int err; 515 struct net *net = dev_net(skb->dev); 516 517 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 518 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 519 if (!sk) { 520 /* No socket for error: try tunnels before discarding */ 521 sk = ERR_PTR(-ENOENT); 522 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 523 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 524 udptable, skb, 525 opt, type, code, info); 526 if (!sk) 527 return 0; 528 } 529 530 if (IS_ERR(sk)) { 531 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 532 ICMP6_MIB_INERRORS); 533 return PTR_ERR(sk); 534 } 535 536 tunnel = true; 537 } 538 539 harderr = icmpv6_err_convert(type, code, &err); 540 np = inet6_sk(sk); 541 542 if (type == ICMPV6_PKT_TOOBIG) { 543 if (!ip6_sk_accept_pmtu(sk)) 544 goto out; 545 ip6_sk_update_pmtu(skb, sk, info); 546 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 547 harderr = 1; 548 } 549 if (type == NDISC_REDIRECT) { 550 if (tunnel) { 551 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 552 sk->sk_mark, sk->sk_uid); 553 } else { 554 ip6_sk_redirect(skb, sk); 555 } 556 goto out; 557 } 558 559 /* Tunnels don't have an application socket: don't pass errors back */ 560 if (tunnel) 561 goto out; 562 563 if (!np->recverr) { 564 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 565 goto out; 566 } else { 567 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 568 } 569 570 sk->sk_err = err; 571 sk->sk_error_report(sk); 572 out: 573 return 0; 574 } 575 576 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 577 { 578 int rc; 579 580 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 581 sock_rps_save_rxhash(sk, skb); 582 sk_mark_napi_id(sk, skb); 583 sk_incoming_cpu_update(sk); 584 } else { 585 sk_mark_napi_id_once(sk, skb); 586 } 587 588 rc = __udp_enqueue_schedule_skb(sk, skb); 589 if (rc < 0) { 590 int is_udplite = IS_UDPLITE(sk); 591 592 /* Note that an ENOMEM error is charged twice */ 593 if (rc == -ENOMEM) 594 UDP6_INC_STATS(sock_net(sk), 595 UDP_MIB_RCVBUFERRORS, is_udplite); 596 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 597 kfree_skb(skb); 598 return -1; 599 } 600 601 return 0; 602 } 603 604 static __inline__ int udpv6_err(struct sk_buff *skb, 605 struct inet6_skb_parm *opt, u8 type, 606 u8 code, int offset, __be32 info) 607 { 608 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 609 } 610 611 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 612 { 613 struct udp_sock *up = udp_sk(sk); 614 int is_udplite = IS_UDPLITE(sk); 615 616 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 617 goto drop; 618 619 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 620 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 621 622 /* 623 * This is an encapsulation socket so pass the skb to 624 * the socket's udp_encap_rcv() hook. Otherwise, just 625 * fall through and pass this up the UDP socket. 626 * up->encap_rcv() returns the following value: 627 * =0 if skb was successfully passed to the encap 628 * handler or was discarded by it. 629 * >0 if skb should be passed on to UDP. 630 * <0 if skb should be resubmitted as proto -N 631 */ 632 633 /* if we're overly short, let UDP handle it */ 634 encap_rcv = READ_ONCE(up->encap_rcv); 635 if (encap_rcv) { 636 int ret; 637 638 /* Verify checksum before giving to encap */ 639 if (udp_lib_checksum_complete(skb)) 640 goto csum_error; 641 642 ret = encap_rcv(sk, skb); 643 if (ret <= 0) { 644 __UDP_INC_STATS(sock_net(sk), 645 UDP_MIB_INDATAGRAMS, 646 is_udplite); 647 return -ret; 648 } 649 } 650 651 /* FALLTHROUGH -- it's a UDP Packet */ 652 } 653 654 /* 655 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 656 */ 657 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 658 659 if (up->pcrlen == 0) { /* full coverage was set */ 660 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 661 UDP_SKB_CB(skb)->cscov, skb->len); 662 goto drop; 663 } 664 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 665 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 666 UDP_SKB_CB(skb)->cscov, up->pcrlen); 667 goto drop; 668 } 669 } 670 671 prefetch(&sk->sk_rmem_alloc); 672 if (rcu_access_pointer(sk->sk_filter) && 673 udp_lib_checksum_complete(skb)) 674 goto csum_error; 675 676 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 677 goto drop; 678 679 udp_csum_pull_header(skb); 680 681 skb_dst_drop(skb); 682 683 return __udpv6_queue_rcv_skb(sk, skb); 684 685 csum_error: 686 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 687 drop: 688 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 689 atomic_inc(&sk->sk_drops); 690 kfree_skb(skb); 691 return -1; 692 } 693 694 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 695 { 696 struct sk_buff *next, *segs; 697 int ret; 698 699 if (likely(!udp_unexpected_gso(sk, skb))) 700 return udpv6_queue_rcv_one_skb(sk, skb); 701 702 __skb_push(skb, -skb_mac_offset(skb)); 703 segs = udp_rcv_segment(sk, skb, false); 704 for (skb = segs; skb; skb = next) { 705 next = skb->next; 706 __skb_pull(skb, skb_transport_offset(skb)); 707 708 ret = udpv6_queue_rcv_one_skb(sk, skb); 709 if (ret > 0) 710 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 711 true); 712 } 713 return 0; 714 } 715 716 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 717 __be16 loc_port, const struct in6_addr *loc_addr, 718 __be16 rmt_port, const struct in6_addr *rmt_addr, 719 int dif, int sdif, unsigned short hnum) 720 { 721 struct inet_sock *inet = inet_sk(sk); 722 723 if (!net_eq(sock_net(sk), net)) 724 return false; 725 726 if (udp_sk(sk)->udp_port_hash != hnum || 727 sk->sk_family != PF_INET6 || 728 (inet->inet_dport && inet->inet_dport != rmt_port) || 729 (!ipv6_addr_any(&sk->sk_v6_daddr) && 730 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 731 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) || 732 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 733 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 734 return false; 735 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 736 return false; 737 return true; 738 } 739 740 static void udp6_csum_zero_error(struct sk_buff *skb) 741 { 742 /* RFC 2460 section 8.1 says that we SHOULD log 743 * this error. Well, it is reasonable. 744 */ 745 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 746 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 747 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 748 } 749 750 /* 751 * Note: called only from the BH handler context, 752 * so we don't need to lock the hashes. 753 */ 754 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 755 const struct in6_addr *saddr, const struct in6_addr *daddr, 756 struct udp_table *udptable, int proto) 757 { 758 struct sock *sk, *first = NULL; 759 const struct udphdr *uh = udp_hdr(skb); 760 unsigned short hnum = ntohs(uh->dest); 761 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 762 unsigned int offset = offsetof(typeof(*sk), sk_node); 763 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 764 int dif = inet6_iif(skb); 765 int sdif = inet6_sdif(skb); 766 struct hlist_node *node; 767 struct sk_buff *nskb; 768 769 if (use_hash2) { 770 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 771 udptable->mask; 772 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 773 start_lookup: 774 hslot = &udptable->hash2[hash2]; 775 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 776 } 777 778 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 779 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 780 uh->source, saddr, dif, sdif, 781 hnum)) 782 continue; 783 /* If zero checksum and no_check is not on for 784 * the socket then skip it. 785 */ 786 if (!uh->check && !udp_sk(sk)->no_check6_rx) 787 continue; 788 if (!first) { 789 first = sk; 790 continue; 791 } 792 nskb = skb_clone(skb, GFP_ATOMIC); 793 if (unlikely(!nskb)) { 794 atomic_inc(&sk->sk_drops); 795 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 796 IS_UDPLITE(sk)); 797 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 798 IS_UDPLITE(sk)); 799 continue; 800 } 801 802 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 803 consume_skb(nskb); 804 } 805 806 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 807 if (use_hash2 && hash2 != hash2_any) { 808 hash2 = hash2_any; 809 goto start_lookup; 810 } 811 812 if (first) { 813 if (udpv6_queue_rcv_skb(first, skb) > 0) 814 consume_skb(skb); 815 } else { 816 kfree_skb(skb); 817 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 818 proto == IPPROTO_UDPLITE); 819 } 820 return 0; 821 } 822 823 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 824 { 825 if (udp_sk_rx_dst_set(sk, dst)) { 826 const struct rt6_info *rt = (const struct rt6_info *)dst; 827 828 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 829 } 830 } 831 832 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 833 * return code conversion for ip layer consumption 834 */ 835 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 836 struct udphdr *uh) 837 { 838 int ret; 839 840 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 841 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 842 ip6_compute_pseudo); 843 844 ret = udpv6_queue_rcv_skb(sk, skb); 845 846 /* a return value > 0 means to resubmit the input */ 847 if (ret > 0) 848 return ret; 849 return 0; 850 } 851 852 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 853 int proto) 854 { 855 const struct in6_addr *saddr, *daddr; 856 struct net *net = dev_net(skb->dev); 857 struct udphdr *uh; 858 struct sock *sk; 859 u32 ulen = 0; 860 861 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 862 goto discard; 863 864 saddr = &ipv6_hdr(skb)->saddr; 865 daddr = &ipv6_hdr(skb)->daddr; 866 uh = udp_hdr(skb); 867 868 ulen = ntohs(uh->len); 869 if (ulen > skb->len) 870 goto short_packet; 871 872 if (proto == IPPROTO_UDP) { 873 /* UDP validates ulen. */ 874 875 /* Check for jumbo payload */ 876 if (ulen == 0) 877 ulen = skb->len; 878 879 if (ulen < sizeof(*uh)) 880 goto short_packet; 881 882 if (ulen < skb->len) { 883 if (pskb_trim_rcsum(skb, ulen)) 884 goto short_packet; 885 saddr = &ipv6_hdr(skb)->saddr; 886 daddr = &ipv6_hdr(skb)->daddr; 887 uh = udp_hdr(skb); 888 } 889 } 890 891 if (udp6_csum_init(skb, uh, proto)) 892 goto csum_error; 893 894 /* Check if the socket is already available, e.g. due to early demux */ 895 sk = skb_steal_sock(skb); 896 if (sk) { 897 struct dst_entry *dst = skb_dst(skb); 898 int ret; 899 900 if (unlikely(sk->sk_rx_dst != dst)) 901 udp6_sk_rx_dst_set(sk, dst); 902 903 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 904 sock_put(sk); 905 goto report_csum_error; 906 } 907 908 ret = udp6_unicast_rcv_skb(sk, skb, uh); 909 sock_put(sk); 910 return ret; 911 } 912 913 /* 914 * Multicast receive code 915 */ 916 if (ipv6_addr_is_multicast(daddr)) 917 return __udp6_lib_mcast_deliver(net, skb, 918 saddr, daddr, udptable, proto); 919 920 /* Unicast */ 921 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 922 if (sk) { 923 if (!uh->check && !udp_sk(sk)->no_check6_rx) 924 goto report_csum_error; 925 return udp6_unicast_rcv_skb(sk, skb, uh); 926 } 927 928 if (!uh->check) 929 goto report_csum_error; 930 931 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 932 goto discard; 933 934 if (udp_lib_checksum_complete(skb)) 935 goto csum_error; 936 937 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 938 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 939 940 kfree_skb(skb); 941 return 0; 942 943 short_packet: 944 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 945 proto == IPPROTO_UDPLITE ? "-Lite" : "", 946 saddr, ntohs(uh->source), 947 ulen, skb->len, 948 daddr, ntohs(uh->dest)); 949 goto discard; 950 951 report_csum_error: 952 udp6_csum_zero_error(skb); 953 csum_error: 954 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 955 discard: 956 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 957 kfree_skb(skb); 958 return 0; 959 } 960 961 962 static struct sock *__udp6_lib_demux_lookup(struct net *net, 963 __be16 loc_port, const struct in6_addr *loc_addr, 964 __be16 rmt_port, const struct in6_addr *rmt_addr, 965 int dif, int sdif) 966 { 967 unsigned short hnum = ntohs(loc_port); 968 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 969 unsigned int slot2 = hash2 & udp_table.mask; 970 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 971 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 972 struct sock *sk; 973 974 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 975 if (sk->sk_state == TCP_ESTABLISHED && 976 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 977 return sk; 978 /* Only check first socket in chain */ 979 break; 980 } 981 return NULL; 982 } 983 984 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb) 985 { 986 struct net *net = dev_net(skb->dev); 987 const struct udphdr *uh; 988 struct sock *sk; 989 struct dst_entry *dst; 990 int dif = skb->dev->ifindex; 991 int sdif = inet6_sdif(skb); 992 993 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 994 sizeof(struct udphdr))) 995 return; 996 997 uh = udp_hdr(skb); 998 999 if (skb->pkt_type == PACKET_HOST) 1000 sk = __udp6_lib_demux_lookup(net, uh->dest, 1001 &ipv6_hdr(skb)->daddr, 1002 uh->source, &ipv6_hdr(skb)->saddr, 1003 dif, sdif); 1004 else 1005 return; 1006 1007 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 1008 return; 1009 1010 skb->sk = sk; 1011 skb->destructor = sock_efree; 1012 dst = READ_ONCE(sk->sk_rx_dst); 1013 1014 if (dst) 1015 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 1016 if (dst) { 1017 /* set noref for now. 1018 * any place which wants to hold dst has to call 1019 * dst_hold_safe() 1020 */ 1021 skb_dst_set_noref(skb, dst); 1022 } 1023 } 1024 1025 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1026 { 1027 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1028 } 1029 1030 /* 1031 * Throw away all pending data and cancel the corking. Socket is locked. 1032 */ 1033 static void udp_v6_flush_pending_frames(struct sock *sk) 1034 { 1035 struct udp_sock *up = udp_sk(sk); 1036 1037 if (up->pending == AF_INET) 1038 udp_flush_pending_frames(sk); 1039 else if (up->pending) { 1040 up->len = 0; 1041 up->pending = 0; 1042 ip6_flush_pending_frames(sk); 1043 } 1044 } 1045 1046 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1047 int addr_len) 1048 { 1049 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1050 return -EINVAL; 1051 /* The following checks are replicated from __ip6_datagram_connect() 1052 * and intended to prevent BPF program called below from accessing 1053 * bytes that are out of the bound specified by user in addr_len. 1054 */ 1055 if (uaddr->sa_family == AF_INET) { 1056 if (__ipv6_only_sock(sk)) 1057 return -EAFNOSUPPORT; 1058 return udp_pre_connect(sk, uaddr, addr_len); 1059 } 1060 1061 if (addr_len < SIN6_LEN_RFC2133) 1062 return -EINVAL; 1063 1064 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 1065 } 1066 1067 /** 1068 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1069 * @sk: socket we are sending on 1070 * @skb: sk_buff containing the filled-in UDP header 1071 * (checksum field must be zeroed out) 1072 */ 1073 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1074 const struct in6_addr *saddr, 1075 const struct in6_addr *daddr, int len) 1076 { 1077 unsigned int offset; 1078 struct udphdr *uh = udp_hdr(skb); 1079 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1080 __wsum csum = 0; 1081 1082 if (!frags) { 1083 /* Only one fragment on the socket. */ 1084 skb->csum_start = skb_transport_header(skb) - skb->head; 1085 skb->csum_offset = offsetof(struct udphdr, check); 1086 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1087 } else { 1088 /* 1089 * HW-checksum won't work as there are two or more 1090 * fragments on the socket so that all csums of sk_buffs 1091 * should be together 1092 */ 1093 offset = skb_transport_offset(skb); 1094 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1095 csum = skb->csum; 1096 1097 skb->ip_summed = CHECKSUM_NONE; 1098 1099 do { 1100 csum = csum_add(csum, frags->csum); 1101 } while ((frags = frags->next)); 1102 1103 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1104 csum); 1105 if (uh->check == 0) 1106 uh->check = CSUM_MANGLED_0; 1107 } 1108 } 1109 1110 /* 1111 * Sending 1112 */ 1113 1114 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1115 struct inet_cork *cork) 1116 { 1117 struct sock *sk = skb->sk; 1118 struct udphdr *uh; 1119 int err = 0; 1120 int is_udplite = IS_UDPLITE(sk); 1121 __wsum csum = 0; 1122 int offset = skb_transport_offset(skb); 1123 int len = skb->len - offset; 1124 1125 /* 1126 * Create a UDP header 1127 */ 1128 uh = udp_hdr(skb); 1129 uh->source = fl6->fl6_sport; 1130 uh->dest = fl6->fl6_dport; 1131 uh->len = htons(len); 1132 uh->check = 0; 1133 1134 if (cork->gso_size) { 1135 const int hlen = skb_network_header_len(skb) + 1136 sizeof(struct udphdr); 1137 1138 if (hlen + cork->gso_size > cork->fragsize) { 1139 kfree_skb(skb); 1140 return -EINVAL; 1141 } 1142 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 1143 kfree_skb(skb); 1144 return -EINVAL; 1145 } 1146 if (udp_sk(sk)->no_check6_tx) { 1147 kfree_skb(skb); 1148 return -EINVAL; 1149 } 1150 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1151 dst_xfrm(skb_dst(skb))) { 1152 kfree_skb(skb); 1153 return -EIO; 1154 } 1155 1156 skb_shinfo(skb)->gso_size = cork->gso_size; 1157 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1158 goto csum_partial; 1159 } 1160 1161 if (is_udplite) 1162 csum = udplite_csum(skb); 1163 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1164 skb->ip_summed = CHECKSUM_NONE; 1165 goto send; 1166 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1167 csum_partial: 1168 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1169 goto send; 1170 } else 1171 csum = udp_csum(skb); 1172 1173 /* add protocol-dependent pseudo-header */ 1174 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1175 len, fl6->flowi6_proto, csum); 1176 if (uh->check == 0) 1177 uh->check = CSUM_MANGLED_0; 1178 1179 send: 1180 err = ip6_send_skb(skb); 1181 if (err) { 1182 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1183 UDP6_INC_STATS(sock_net(sk), 1184 UDP_MIB_SNDBUFERRORS, is_udplite); 1185 err = 0; 1186 } 1187 } else { 1188 UDP6_INC_STATS(sock_net(sk), 1189 UDP_MIB_OUTDATAGRAMS, is_udplite); 1190 } 1191 return err; 1192 } 1193 1194 static int udp_v6_push_pending_frames(struct sock *sk) 1195 { 1196 struct sk_buff *skb; 1197 struct udp_sock *up = udp_sk(sk); 1198 struct flowi6 fl6; 1199 int err = 0; 1200 1201 if (up->pending == AF_INET) 1202 return udp_push_pending_frames(sk); 1203 1204 /* ip6_finish_skb will release the cork, so make a copy of 1205 * fl6 here. 1206 */ 1207 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1208 1209 skb = ip6_finish_skb(sk); 1210 if (!skb) 1211 goto out; 1212 1213 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1214 1215 out: 1216 up->len = 0; 1217 up->pending = 0; 1218 return err; 1219 } 1220 1221 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1222 { 1223 struct ipv6_txoptions opt_space; 1224 struct udp_sock *up = udp_sk(sk); 1225 struct inet_sock *inet = inet_sk(sk); 1226 struct ipv6_pinfo *np = inet6_sk(sk); 1227 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1228 struct in6_addr *daddr, *final_p, final; 1229 struct ipv6_txoptions *opt = NULL; 1230 struct ipv6_txoptions *opt_to_free = NULL; 1231 struct ip6_flowlabel *flowlabel = NULL; 1232 struct flowi6 fl6; 1233 struct dst_entry *dst; 1234 struct ipcm6_cookie ipc6; 1235 int addr_len = msg->msg_namelen; 1236 bool connected = false; 1237 int ulen = len; 1238 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1239 int err; 1240 int is_udplite = IS_UDPLITE(sk); 1241 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1242 1243 ipcm6_init(&ipc6); 1244 ipc6.gso_size = up->gso_size; 1245 ipc6.sockc.tsflags = sk->sk_tsflags; 1246 1247 /* destination address check */ 1248 if (sin6) { 1249 if (addr_len < offsetof(struct sockaddr, sa_data)) 1250 return -EINVAL; 1251 1252 switch (sin6->sin6_family) { 1253 case AF_INET6: 1254 if (addr_len < SIN6_LEN_RFC2133) 1255 return -EINVAL; 1256 daddr = &sin6->sin6_addr; 1257 if (ipv6_addr_any(daddr) && 1258 ipv6_addr_v4mapped(&np->saddr)) 1259 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1260 daddr); 1261 break; 1262 case AF_INET: 1263 goto do_udp_sendmsg; 1264 case AF_UNSPEC: 1265 msg->msg_name = sin6 = NULL; 1266 msg->msg_namelen = addr_len = 0; 1267 daddr = NULL; 1268 break; 1269 default: 1270 return -EINVAL; 1271 } 1272 } else if (!up->pending) { 1273 if (sk->sk_state != TCP_ESTABLISHED) 1274 return -EDESTADDRREQ; 1275 daddr = &sk->sk_v6_daddr; 1276 } else 1277 daddr = NULL; 1278 1279 if (daddr) { 1280 if (ipv6_addr_v4mapped(daddr)) { 1281 struct sockaddr_in sin; 1282 sin.sin_family = AF_INET; 1283 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1284 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1285 msg->msg_name = &sin; 1286 msg->msg_namelen = sizeof(sin); 1287 do_udp_sendmsg: 1288 if (__ipv6_only_sock(sk)) 1289 return -ENETUNREACH; 1290 return udp_sendmsg(sk, msg, len); 1291 } 1292 } 1293 1294 if (up->pending == AF_INET) 1295 return udp_sendmsg(sk, msg, len); 1296 1297 /* Rough check on arithmetic overflow, 1298 better check is made in ip6_append_data(). 1299 */ 1300 if (len > INT_MAX - sizeof(struct udphdr)) 1301 return -EMSGSIZE; 1302 1303 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1304 if (up->pending) { 1305 /* 1306 * There are pending frames. 1307 * The socket lock must be held while it's corked. 1308 */ 1309 lock_sock(sk); 1310 if (likely(up->pending)) { 1311 if (unlikely(up->pending != AF_INET6)) { 1312 release_sock(sk); 1313 return -EAFNOSUPPORT; 1314 } 1315 dst = NULL; 1316 goto do_append_data; 1317 } 1318 release_sock(sk); 1319 } 1320 ulen += sizeof(struct udphdr); 1321 1322 memset(&fl6, 0, sizeof(fl6)); 1323 1324 if (sin6) { 1325 if (sin6->sin6_port == 0) 1326 return -EINVAL; 1327 1328 fl6.fl6_dport = sin6->sin6_port; 1329 daddr = &sin6->sin6_addr; 1330 1331 if (np->sndflow) { 1332 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1333 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1334 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1335 if (!flowlabel) 1336 return -EINVAL; 1337 } 1338 } 1339 1340 /* 1341 * Otherwise it will be difficult to maintain 1342 * sk->sk_dst_cache. 1343 */ 1344 if (sk->sk_state == TCP_ESTABLISHED && 1345 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1346 daddr = &sk->sk_v6_daddr; 1347 1348 if (addr_len >= sizeof(struct sockaddr_in6) && 1349 sin6->sin6_scope_id && 1350 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1351 fl6.flowi6_oif = sin6->sin6_scope_id; 1352 } else { 1353 if (sk->sk_state != TCP_ESTABLISHED) 1354 return -EDESTADDRREQ; 1355 1356 fl6.fl6_dport = inet->inet_dport; 1357 daddr = &sk->sk_v6_daddr; 1358 fl6.flowlabel = np->flow_label; 1359 connected = true; 1360 } 1361 1362 if (!fl6.flowi6_oif) 1363 fl6.flowi6_oif = sk->sk_bound_dev_if; 1364 1365 if (!fl6.flowi6_oif) 1366 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1367 1368 fl6.flowi6_mark = sk->sk_mark; 1369 fl6.flowi6_uid = sk->sk_uid; 1370 1371 if (msg->msg_controllen) { 1372 opt = &opt_space; 1373 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1374 opt->tot_len = sizeof(*opt); 1375 ipc6.opt = opt; 1376 1377 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1378 if (err > 0) 1379 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1380 &ipc6); 1381 if (err < 0) { 1382 fl6_sock_release(flowlabel); 1383 return err; 1384 } 1385 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1386 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1387 if (!flowlabel) 1388 return -EINVAL; 1389 } 1390 if (!(opt->opt_nflen|opt->opt_flen)) 1391 opt = NULL; 1392 connected = false; 1393 } 1394 if (!opt) { 1395 opt = txopt_get(np); 1396 opt_to_free = opt; 1397 } 1398 if (flowlabel) 1399 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1400 opt = ipv6_fixup_options(&opt_space, opt); 1401 ipc6.opt = opt; 1402 1403 fl6.flowi6_proto = sk->sk_protocol; 1404 fl6.daddr = *daddr; 1405 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1406 fl6.saddr = np->saddr; 1407 fl6.fl6_sport = inet->inet_sport; 1408 1409 if (cgroup_bpf_enabled && !connected) { 1410 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1411 (struct sockaddr *)sin6, &fl6.saddr); 1412 if (err) 1413 goto out_no_dst; 1414 if (sin6) { 1415 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1416 /* BPF program rewrote IPv6-only by IPv4-mapped 1417 * IPv6. It's currently unsupported. 1418 */ 1419 err = -ENOTSUPP; 1420 goto out_no_dst; 1421 } 1422 if (sin6->sin6_port == 0) { 1423 /* BPF program set invalid port. Reject it. */ 1424 err = -EINVAL; 1425 goto out_no_dst; 1426 } 1427 fl6.fl6_dport = sin6->sin6_port; 1428 fl6.daddr = sin6->sin6_addr; 1429 } 1430 } 1431 1432 if (ipv6_addr_any(&fl6.daddr)) 1433 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1434 1435 final_p = fl6_update_dst(&fl6, opt, &final); 1436 if (final_p) 1437 connected = false; 1438 1439 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1440 fl6.flowi6_oif = np->mcast_oif; 1441 connected = false; 1442 } else if (!fl6.flowi6_oif) 1443 fl6.flowi6_oif = np->ucast_oif; 1444 1445 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1446 1447 if (ipc6.tclass < 0) 1448 ipc6.tclass = np->tclass; 1449 1450 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1451 1452 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1453 if (IS_ERR(dst)) { 1454 err = PTR_ERR(dst); 1455 dst = NULL; 1456 goto out; 1457 } 1458 1459 if (ipc6.hlimit < 0) 1460 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1461 1462 if (msg->msg_flags&MSG_CONFIRM) 1463 goto do_confirm; 1464 back_from_confirm: 1465 1466 /* Lockless fast path for the non-corking case */ 1467 if (!corkreq) { 1468 struct inet_cork_full cork; 1469 struct sk_buff *skb; 1470 1471 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1472 sizeof(struct udphdr), &ipc6, 1473 &fl6, (struct rt6_info *)dst, 1474 msg->msg_flags, &cork); 1475 err = PTR_ERR(skb); 1476 if (!IS_ERR_OR_NULL(skb)) 1477 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1478 goto out; 1479 } 1480 1481 lock_sock(sk); 1482 if (unlikely(up->pending)) { 1483 /* The socket is already corked while preparing it. */ 1484 /* ... which is an evident application bug. --ANK */ 1485 release_sock(sk); 1486 1487 net_dbg_ratelimited("udp cork app bug 2\n"); 1488 err = -EINVAL; 1489 goto out; 1490 } 1491 1492 up->pending = AF_INET6; 1493 1494 do_append_data: 1495 if (ipc6.dontfrag < 0) 1496 ipc6.dontfrag = np->dontfrag; 1497 up->len += ulen; 1498 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1499 &ipc6, &fl6, (struct rt6_info *)dst, 1500 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1501 if (err) 1502 udp_v6_flush_pending_frames(sk); 1503 else if (!corkreq) 1504 err = udp_v6_push_pending_frames(sk); 1505 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1506 up->pending = 0; 1507 1508 if (err > 0) 1509 err = np->recverr ? net_xmit_errno(err) : 0; 1510 release_sock(sk); 1511 1512 out: 1513 dst_release(dst); 1514 out_no_dst: 1515 fl6_sock_release(flowlabel); 1516 txopt_put(opt_to_free); 1517 if (!err) 1518 return len; 1519 /* 1520 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1521 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1522 * we don't have a good statistic (IpOutDiscards but it can be too many 1523 * things). We could add another new stat but at least for now that 1524 * seems like overkill. 1525 */ 1526 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1527 UDP6_INC_STATS(sock_net(sk), 1528 UDP_MIB_SNDBUFERRORS, is_udplite); 1529 } 1530 return err; 1531 1532 do_confirm: 1533 if (msg->msg_flags & MSG_PROBE) 1534 dst_confirm_neigh(dst, &fl6.daddr); 1535 if (!(msg->msg_flags&MSG_PROBE) || len) 1536 goto back_from_confirm; 1537 err = 0; 1538 goto out; 1539 } 1540 1541 void udpv6_destroy_sock(struct sock *sk) 1542 { 1543 struct udp_sock *up = udp_sk(sk); 1544 lock_sock(sk); 1545 udp_v6_flush_pending_frames(sk); 1546 release_sock(sk); 1547 1548 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1549 if (up->encap_type) { 1550 void (*encap_destroy)(struct sock *sk); 1551 encap_destroy = READ_ONCE(up->encap_destroy); 1552 if (encap_destroy) 1553 encap_destroy(sk); 1554 } 1555 if (up->encap_enabled) 1556 static_branch_dec(&udpv6_encap_needed_key); 1557 } 1558 1559 inet6_destroy_sock(sk); 1560 } 1561 1562 /* 1563 * Socket option code for UDP 1564 */ 1565 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1566 char __user *optval, unsigned int optlen) 1567 { 1568 if (level == SOL_UDP || level == SOL_UDPLITE) 1569 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1570 udp_v6_push_pending_frames); 1571 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1572 } 1573 1574 #ifdef CONFIG_COMPAT 1575 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1576 char __user *optval, unsigned int optlen) 1577 { 1578 if (level == SOL_UDP || level == SOL_UDPLITE) 1579 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1580 udp_v6_push_pending_frames); 1581 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1582 } 1583 #endif 1584 1585 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1586 char __user *optval, int __user *optlen) 1587 { 1588 if (level == SOL_UDP || level == SOL_UDPLITE) 1589 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1590 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1591 } 1592 1593 #ifdef CONFIG_COMPAT 1594 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1595 char __user *optval, int __user *optlen) 1596 { 1597 if (level == SOL_UDP || level == SOL_UDPLITE) 1598 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1599 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1600 } 1601 #endif 1602 1603 /* thinking of making this const? Don't. 1604 * early_demux can change based on sysctl. 1605 */ 1606 static struct inet6_protocol udpv6_protocol = { 1607 .early_demux = udp_v6_early_demux, 1608 .early_demux_handler = udp_v6_early_demux, 1609 .handler = udpv6_rcv, 1610 .err_handler = udpv6_err, 1611 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1612 }; 1613 1614 /* ------------------------------------------------------------------------ */ 1615 #ifdef CONFIG_PROC_FS 1616 int udp6_seq_show(struct seq_file *seq, void *v) 1617 { 1618 if (v == SEQ_START_TOKEN) { 1619 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1620 } else { 1621 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1622 struct inet_sock *inet = inet_sk(v); 1623 __u16 srcp = ntohs(inet->inet_sport); 1624 __u16 destp = ntohs(inet->inet_dport); 1625 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1626 udp_rqueue_get(v), bucket); 1627 } 1628 return 0; 1629 } 1630 1631 const struct seq_operations udp6_seq_ops = { 1632 .start = udp_seq_start, 1633 .next = udp_seq_next, 1634 .stop = udp_seq_stop, 1635 .show = udp6_seq_show, 1636 }; 1637 EXPORT_SYMBOL(udp6_seq_ops); 1638 1639 static struct udp_seq_afinfo udp6_seq_afinfo = { 1640 .family = AF_INET6, 1641 .udp_table = &udp_table, 1642 }; 1643 1644 int __net_init udp6_proc_init(struct net *net) 1645 { 1646 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1647 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1648 return -ENOMEM; 1649 return 0; 1650 } 1651 1652 void udp6_proc_exit(struct net *net) 1653 { 1654 remove_proc_entry("udp6", net->proc_net); 1655 } 1656 #endif /* CONFIG_PROC_FS */ 1657 1658 /* ------------------------------------------------------------------------ */ 1659 1660 struct proto udpv6_prot = { 1661 .name = "UDPv6", 1662 .owner = THIS_MODULE, 1663 .close = udp_lib_close, 1664 .pre_connect = udpv6_pre_connect, 1665 .connect = ip6_datagram_connect, 1666 .disconnect = udp_disconnect, 1667 .ioctl = udp_ioctl, 1668 .init = udp_init_sock, 1669 .destroy = udpv6_destroy_sock, 1670 .setsockopt = udpv6_setsockopt, 1671 .getsockopt = udpv6_getsockopt, 1672 .sendmsg = udpv6_sendmsg, 1673 .recvmsg = udpv6_recvmsg, 1674 .release_cb = ip6_datagram_release_cb, 1675 .hash = udp_lib_hash, 1676 .unhash = udp_lib_unhash, 1677 .rehash = udp_v6_rehash, 1678 .get_port = udp_v6_get_port, 1679 .memory_allocated = &udp_memory_allocated, 1680 .sysctl_mem = sysctl_udp_mem, 1681 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1682 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1683 .obj_size = sizeof(struct udp6_sock), 1684 .h.udp_table = &udp_table, 1685 #ifdef CONFIG_COMPAT 1686 .compat_setsockopt = compat_udpv6_setsockopt, 1687 .compat_getsockopt = compat_udpv6_getsockopt, 1688 #endif 1689 .diag_destroy = udp_abort, 1690 }; 1691 1692 static struct inet_protosw udpv6_protosw = { 1693 .type = SOCK_DGRAM, 1694 .protocol = IPPROTO_UDP, 1695 .prot = &udpv6_prot, 1696 .ops = &inet6_dgram_ops, 1697 .flags = INET_PROTOSW_PERMANENT, 1698 }; 1699 1700 int __init udpv6_init(void) 1701 { 1702 int ret; 1703 1704 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1705 if (ret) 1706 goto out; 1707 1708 ret = inet6_register_protosw(&udpv6_protosw); 1709 if (ret) 1710 goto out_udpv6_protocol; 1711 out: 1712 return ret; 1713 1714 out_udpv6_protocol: 1715 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1716 goto out; 1717 } 1718 1719 void udpv6_exit(void) 1720 { 1721 inet6_unregister_protosw(&udpv6_protosw); 1722 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1723 } 1724