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/xfrm.h> 49 #include <net/inet_hashtables.h> 50 #include <net/inet6_hashtables.h> 51 #include <net/busy_poll.h> 52 #include <net/sock_reuseport.h> 53 54 #include <linux/proc_fs.h> 55 #include <linux/seq_file.h> 56 #include <trace/events/skb.h> 57 #include "udp_impl.h" 58 59 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb) 60 { 61 #if defined(CONFIG_NET_L3_MASTER_DEV) 62 if (!net->ipv4.sysctl_udp_l3mdev_accept && 63 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags)) 64 return true; 65 #endif 66 return false; 67 } 68 69 static u32 udp6_ehashfn(const struct net *net, 70 const struct in6_addr *laddr, 71 const u16 lport, 72 const struct in6_addr *faddr, 73 const __be16 fport) 74 { 75 static u32 udp6_ehash_secret __read_mostly; 76 static u32 udp_ipv6_hash_secret __read_mostly; 77 78 u32 lhash, fhash; 79 80 net_get_random_once(&udp6_ehash_secret, 81 sizeof(udp6_ehash_secret)); 82 net_get_random_once(&udp_ipv6_hash_secret, 83 sizeof(udp_ipv6_hash_secret)); 84 85 lhash = (__force u32)laddr->s6_addr32[3]; 86 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 87 88 return __inet6_ehashfn(lhash, lport, fhash, fport, 89 udp_ipv6_hash_secret + net_hash_mix(net)); 90 } 91 92 int udp_v6_get_port(struct sock *sk, unsigned short snum) 93 { 94 unsigned int hash2_nulladdr = 95 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 96 unsigned int hash2_partial = 97 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 98 99 /* precompute partial secondary hash */ 100 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 101 return udp_lib_get_port(sk, snum, hash2_nulladdr); 102 } 103 104 static void udp_v6_rehash(struct sock *sk) 105 { 106 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 107 &sk->sk_v6_rcv_saddr, 108 inet_sk(sk)->inet_num); 109 110 udp_lib_rehash(sk, new_hash); 111 } 112 113 static int compute_score(struct sock *sk, struct net *net, 114 const struct in6_addr *saddr, __be16 sport, 115 const struct in6_addr *daddr, unsigned short hnum, 116 int dif, int sdif, bool exact_dif) 117 { 118 int score; 119 struct inet_sock *inet; 120 121 if (!net_eq(sock_net(sk), net) || 122 udp_sk(sk)->udp_port_hash != hnum || 123 sk->sk_family != PF_INET6) 124 return -1; 125 126 score = 0; 127 inet = inet_sk(sk); 128 129 if (inet->inet_dport) { 130 if (inet->inet_dport != sport) 131 return -1; 132 score++; 133 } 134 135 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { 136 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 137 return -1; 138 score++; 139 } 140 141 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 142 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 143 return -1; 144 score++; 145 } 146 147 if (sk->sk_bound_dev_if || exact_dif) { 148 bool dev_match = (sk->sk_bound_dev_if == dif || 149 sk->sk_bound_dev_if == sdif); 150 151 if (!dev_match) 152 return -1; 153 if (sk->sk_bound_dev_if) 154 score++; 155 } 156 157 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 158 score++; 159 160 return score; 161 } 162 163 /* called with rcu_read_lock() */ 164 static struct sock *udp6_lib_lookup2(struct net *net, 165 const struct in6_addr *saddr, __be16 sport, 166 const struct in6_addr *daddr, unsigned int hnum, 167 int dif, int sdif, bool exact_dif, 168 struct udp_hslot *hslot2, struct sk_buff *skb) 169 { 170 struct sock *sk, *result; 171 int score, badness; 172 u32 hash = 0; 173 174 result = NULL; 175 badness = -1; 176 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 177 score = compute_score(sk, net, saddr, sport, 178 daddr, hnum, dif, sdif, exact_dif); 179 if (score > badness) { 180 if (sk->sk_reuseport) { 181 hash = udp6_ehashfn(net, daddr, hnum, 182 saddr, sport); 183 184 result = reuseport_select_sock(sk, hash, skb, 185 sizeof(struct udphdr)); 186 if (result) 187 return result; 188 } 189 result = sk; 190 badness = score; 191 } 192 } 193 return result; 194 } 195 196 /* rcu_read_lock() must be held */ 197 struct sock *__udp6_lib_lookup(struct net *net, 198 const struct in6_addr *saddr, __be16 sport, 199 const struct in6_addr *daddr, __be16 dport, 200 int dif, int sdif, struct udp_table *udptable, 201 struct sk_buff *skb) 202 { 203 struct sock *sk, *result; 204 unsigned short hnum = ntohs(dport); 205 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 206 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 207 bool exact_dif = udp6_lib_exact_dif_match(net, skb); 208 int score, badness; 209 u32 hash = 0; 210 211 if (hslot->count > 10) { 212 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 213 slot2 = hash2 & udptable->mask; 214 hslot2 = &udptable->hash2[slot2]; 215 if (hslot->count < hslot2->count) 216 goto begin; 217 218 result = udp6_lib_lookup2(net, saddr, sport, 219 daddr, hnum, dif, sdif, exact_dif, 220 hslot2, skb); 221 if (!result) { 222 unsigned int old_slot2 = slot2; 223 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 224 slot2 = hash2 & udptable->mask; 225 /* avoid searching the same slot again. */ 226 if (unlikely(slot2 == old_slot2)) 227 return result; 228 229 hslot2 = &udptable->hash2[slot2]; 230 if (hslot->count < hslot2->count) 231 goto begin; 232 233 result = udp6_lib_lookup2(net, saddr, sport, 234 daddr, hnum, dif, sdif, 235 exact_dif, hslot2, 236 skb); 237 } 238 if (unlikely(IS_ERR(result))) 239 return NULL; 240 return result; 241 } 242 begin: 243 result = NULL; 244 badness = -1; 245 sk_for_each_rcu(sk, &hslot->head) { 246 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif, 247 sdif, exact_dif); 248 if (score > badness) { 249 if (sk->sk_reuseport) { 250 hash = udp6_ehashfn(net, daddr, hnum, 251 saddr, sport); 252 result = reuseport_select_sock(sk, hash, skb, 253 sizeof(struct udphdr)); 254 if (unlikely(IS_ERR(result))) 255 return NULL; 256 if (result) 257 return result; 258 } 259 result = sk; 260 badness = score; 261 } 262 } 263 return result; 264 } 265 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 266 267 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 268 __be16 sport, __be16 dport, 269 struct udp_table *udptable) 270 { 271 const struct ipv6hdr *iph = ipv6_hdr(skb); 272 273 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 274 &iph->daddr, dport, inet6_iif(skb), 275 inet6_sdif(skb), udptable, skb); 276 } 277 278 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, 279 __be16 sport, __be16 dport) 280 { 281 const struct ipv6hdr *iph = ipv6_hdr(skb); 282 283 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 284 &iph->daddr, dport, inet6_iif(skb), 285 inet6_sdif(skb), &udp_table, skb); 286 } 287 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); 288 289 /* Must be called under rcu_read_lock(). 290 * Does increment socket refcount. 291 */ 292 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 293 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 294 const struct in6_addr *daddr, __be16 dport, int dif) 295 { 296 struct sock *sk; 297 298 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 299 dif, 0, &udp_table, NULL); 300 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 301 sk = NULL; 302 return sk; 303 } 304 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 305 #endif 306 307 /* do not use the scratch area len for jumbogram: their length execeeds the 308 * scratch area space; note that the IP6CB flags is still in the first 309 * cacheline, so checking for jumbograms is cheap 310 */ 311 static int udp6_skb_len(struct sk_buff *skb) 312 { 313 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 314 } 315 316 /* 317 * This should be easy, if there is something there we 318 * return it, otherwise we block. 319 */ 320 321 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 322 int noblock, int flags, int *addr_len) 323 { 324 struct ipv6_pinfo *np = inet6_sk(sk); 325 struct inet_sock *inet = inet_sk(sk); 326 struct sk_buff *skb; 327 unsigned int ulen, copied; 328 int peeked, peeking, off; 329 int err; 330 int is_udplite = IS_UDPLITE(sk); 331 bool checksum_valid = false; 332 int is_udp4; 333 334 if (flags & MSG_ERRQUEUE) 335 return ipv6_recv_error(sk, msg, len, addr_len); 336 337 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 338 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 339 340 try_again: 341 peeking = flags & MSG_PEEK; 342 off = sk_peek_offset(sk, flags); 343 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); 344 if (!skb) 345 return err; 346 347 ulen = udp6_skb_len(skb); 348 copied = len; 349 if (copied > ulen - off) 350 copied = ulen - off; 351 else if (copied < ulen) 352 msg->msg_flags |= MSG_TRUNC; 353 354 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 355 356 /* 357 * If checksum is needed at all, try to do it while copying the 358 * data. If the data is truncated, or if we only want a partial 359 * coverage checksum (UDP-Lite), do it before the copy. 360 */ 361 362 if (copied < ulen || peeking || 363 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 364 checksum_valid = udp_skb_csum_unnecessary(skb) || 365 !__udp_lib_checksum_complete(skb); 366 if (!checksum_valid) 367 goto csum_copy_err; 368 } 369 370 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 371 if (udp_skb_is_linear(skb)) 372 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 373 else 374 err = skb_copy_datagram_msg(skb, off, msg, copied); 375 } else { 376 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 377 if (err == -EINVAL) 378 goto csum_copy_err; 379 } 380 if (unlikely(err)) { 381 if (!peeked) { 382 atomic_inc(&sk->sk_drops); 383 if (is_udp4) 384 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 385 is_udplite); 386 else 387 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 388 is_udplite); 389 } 390 kfree_skb(skb); 391 return err; 392 } 393 if (!peeked) { 394 if (is_udp4) 395 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, 396 is_udplite); 397 else 398 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, 399 is_udplite); 400 } 401 402 sock_recv_ts_and_drops(msg, sk, skb); 403 404 /* Copy the address. */ 405 if (msg->msg_name) { 406 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 407 sin6->sin6_family = AF_INET6; 408 sin6->sin6_port = udp_hdr(skb)->source; 409 sin6->sin6_flowinfo = 0; 410 411 if (is_udp4) { 412 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 413 &sin6->sin6_addr); 414 sin6->sin6_scope_id = 0; 415 } else { 416 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 417 sin6->sin6_scope_id = 418 ipv6_iface_scope_id(&sin6->sin6_addr, 419 inet6_iif(skb)); 420 } 421 *addr_len = sizeof(*sin6); 422 } 423 424 if (np->rxopt.all) 425 ip6_datagram_recv_common_ctl(sk, msg, skb); 426 427 if (is_udp4) { 428 if (inet->cmsg_flags) 429 ip_cmsg_recv_offset(msg, sk, skb, 430 sizeof(struct udphdr), off); 431 } else { 432 if (np->rxopt.all) 433 ip6_datagram_recv_specific_ctl(sk, msg, skb); 434 } 435 436 err = copied; 437 if (flags & MSG_TRUNC) 438 err = ulen; 439 440 skb_consume_udp(sk, skb, peeking ? -err : err); 441 return err; 442 443 csum_copy_err: 444 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 445 udp_skb_destructor)) { 446 if (is_udp4) { 447 UDP_INC_STATS(sock_net(sk), 448 UDP_MIB_CSUMERRORS, is_udplite); 449 UDP_INC_STATS(sock_net(sk), 450 UDP_MIB_INERRORS, is_udplite); 451 } else { 452 UDP6_INC_STATS(sock_net(sk), 453 UDP_MIB_CSUMERRORS, is_udplite); 454 UDP6_INC_STATS(sock_net(sk), 455 UDP_MIB_INERRORS, is_udplite); 456 } 457 } 458 kfree_skb(skb); 459 460 /* starting over for a new packet, but check if we need to yield */ 461 cond_resched(); 462 msg->msg_flags &= ~MSG_TRUNC; 463 goto try_again; 464 } 465 466 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 467 u8 type, u8 code, int offset, __be32 info, 468 struct udp_table *udptable) 469 { 470 struct ipv6_pinfo *np; 471 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 472 const struct in6_addr *saddr = &hdr->saddr; 473 const struct in6_addr *daddr = &hdr->daddr; 474 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 475 struct sock *sk; 476 int harderr; 477 int err; 478 struct net *net = dev_net(skb->dev); 479 480 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 481 inet6_iif(skb), inet6_sdif(skb), udptable, skb); 482 if (!sk) { 483 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 484 ICMP6_MIB_INERRORS); 485 return; 486 } 487 488 harderr = icmpv6_err_convert(type, code, &err); 489 np = inet6_sk(sk); 490 491 if (type == ICMPV6_PKT_TOOBIG) { 492 if (!ip6_sk_accept_pmtu(sk)) 493 goto out; 494 ip6_sk_update_pmtu(skb, sk, info); 495 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 496 harderr = 1; 497 } 498 if (type == NDISC_REDIRECT) { 499 ip6_sk_redirect(skb, sk); 500 goto out; 501 } 502 503 if (!np->recverr) { 504 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 505 goto out; 506 } else { 507 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 508 } 509 510 sk->sk_err = err; 511 sk->sk_error_report(sk); 512 out: 513 return; 514 } 515 516 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 517 { 518 int rc; 519 520 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 521 sock_rps_save_rxhash(sk, skb); 522 sk_mark_napi_id(sk, skb); 523 sk_incoming_cpu_update(sk); 524 } else { 525 sk_mark_napi_id_once(sk, skb); 526 } 527 528 rc = __udp_enqueue_schedule_skb(sk, skb); 529 if (rc < 0) { 530 int is_udplite = IS_UDPLITE(sk); 531 532 /* Note that an ENOMEM error is charged twice */ 533 if (rc == -ENOMEM) 534 UDP6_INC_STATS(sock_net(sk), 535 UDP_MIB_RCVBUFERRORS, is_udplite); 536 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 537 kfree_skb(skb); 538 return -1; 539 } 540 541 return 0; 542 } 543 544 static __inline__ void udpv6_err(struct sk_buff *skb, 545 struct inet6_skb_parm *opt, u8 type, 546 u8 code, int offset, __be32 info) 547 { 548 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 549 } 550 551 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 552 void udpv6_encap_enable(void) 553 { 554 static_branch_enable(&udpv6_encap_needed_key); 555 } 556 EXPORT_SYMBOL(udpv6_encap_enable); 557 558 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 559 { 560 struct udp_sock *up = udp_sk(sk); 561 int is_udplite = IS_UDPLITE(sk); 562 563 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 564 goto drop; 565 566 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 567 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 568 569 /* 570 * This is an encapsulation socket so pass the skb to 571 * the socket's udp_encap_rcv() hook. Otherwise, just 572 * fall through and pass this up the UDP socket. 573 * up->encap_rcv() returns the following value: 574 * =0 if skb was successfully passed to the encap 575 * handler or was discarded by it. 576 * >0 if skb should be passed on to UDP. 577 * <0 if skb should be resubmitted as proto -N 578 */ 579 580 /* if we're overly short, let UDP handle it */ 581 encap_rcv = READ_ONCE(up->encap_rcv); 582 if (encap_rcv) { 583 int ret; 584 585 /* Verify checksum before giving to encap */ 586 if (udp_lib_checksum_complete(skb)) 587 goto csum_error; 588 589 ret = encap_rcv(sk, skb); 590 if (ret <= 0) { 591 __UDP_INC_STATS(sock_net(sk), 592 UDP_MIB_INDATAGRAMS, 593 is_udplite); 594 return -ret; 595 } 596 } 597 598 /* FALLTHROUGH -- it's a UDP Packet */ 599 } 600 601 /* 602 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 603 */ 604 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 605 606 if (up->pcrlen == 0) { /* full coverage was set */ 607 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 608 UDP_SKB_CB(skb)->cscov, skb->len); 609 goto drop; 610 } 611 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 612 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 613 UDP_SKB_CB(skb)->cscov, up->pcrlen); 614 goto drop; 615 } 616 } 617 618 prefetch(&sk->sk_rmem_alloc); 619 if (rcu_access_pointer(sk->sk_filter) && 620 udp_lib_checksum_complete(skb)) 621 goto csum_error; 622 623 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 624 goto drop; 625 626 udp_csum_pull_header(skb); 627 628 skb_dst_drop(skb); 629 630 return __udpv6_queue_rcv_skb(sk, skb); 631 632 csum_error: 633 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 634 drop: 635 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 636 atomic_inc(&sk->sk_drops); 637 kfree_skb(skb); 638 return -1; 639 } 640 641 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 642 __be16 loc_port, const struct in6_addr *loc_addr, 643 __be16 rmt_port, const struct in6_addr *rmt_addr, 644 int dif, unsigned short hnum) 645 { 646 struct inet_sock *inet = inet_sk(sk); 647 648 if (!net_eq(sock_net(sk), net)) 649 return false; 650 651 if (udp_sk(sk)->udp_port_hash != hnum || 652 sk->sk_family != PF_INET6 || 653 (inet->inet_dport && inet->inet_dport != rmt_port) || 654 (!ipv6_addr_any(&sk->sk_v6_daddr) && 655 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 656 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) || 657 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 658 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 659 return false; 660 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 661 return false; 662 return true; 663 } 664 665 static void udp6_csum_zero_error(struct sk_buff *skb) 666 { 667 /* RFC 2460 section 8.1 says that we SHOULD log 668 * this error. Well, it is reasonable. 669 */ 670 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 671 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 672 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 673 } 674 675 /* 676 * Note: called only from the BH handler context, 677 * so we don't need to lock the hashes. 678 */ 679 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 680 const struct in6_addr *saddr, const struct in6_addr *daddr, 681 struct udp_table *udptable, int proto) 682 { 683 struct sock *sk, *first = NULL; 684 const struct udphdr *uh = udp_hdr(skb); 685 unsigned short hnum = ntohs(uh->dest); 686 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 687 unsigned int offset = offsetof(typeof(*sk), sk_node); 688 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 689 int dif = inet6_iif(skb); 690 struct hlist_node *node; 691 struct sk_buff *nskb; 692 693 if (use_hash2) { 694 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 695 udptable->mask; 696 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 697 start_lookup: 698 hslot = &udptable->hash2[hash2]; 699 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 700 } 701 702 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 703 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 704 uh->source, saddr, dif, hnum)) 705 continue; 706 /* If zero checksum and no_check is not on for 707 * the socket then skip it. 708 */ 709 if (!uh->check && !udp_sk(sk)->no_check6_rx) 710 continue; 711 if (!first) { 712 first = sk; 713 continue; 714 } 715 nskb = skb_clone(skb, GFP_ATOMIC); 716 if (unlikely(!nskb)) { 717 atomic_inc(&sk->sk_drops); 718 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 719 IS_UDPLITE(sk)); 720 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 721 IS_UDPLITE(sk)); 722 continue; 723 } 724 725 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 726 consume_skb(nskb); 727 } 728 729 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 730 if (use_hash2 && hash2 != hash2_any) { 731 hash2 = hash2_any; 732 goto start_lookup; 733 } 734 735 if (first) { 736 if (udpv6_queue_rcv_skb(first, skb) > 0) 737 consume_skb(skb); 738 } else { 739 kfree_skb(skb); 740 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 741 proto == IPPROTO_UDPLITE); 742 } 743 return 0; 744 } 745 746 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 747 { 748 if (udp_sk_rx_dst_set(sk, dst)) { 749 const struct rt6_info *rt = (const struct rt6_info *)dst; 750 751 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 752 } 753 } 754 755 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 756 * return code conversion for ip layer consumption 757 */ 758 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 759 struct udphdr *uh) 760 { 761 int ret; 762 763 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 764 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 765 ip6_compute_pseudo); 766 767 ret = udpv6_queue_rcv_skb(sk, skb); 768 769 /* a return value > 0 means to resubmit the input */ 770 if (ret > 0) 771 return ret; 772 return 0; 773 } 774 775 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 776 int proto) 777 { 778 const struct in6_addr *saddr, *daddr; 779 struct net *net = dev_net(skb->dev); 780 struct udphdr *uh; 781 struct sock *sk; 782 u32 ulen = 0; 783 784 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 785 goto discard; 786 787 saddr = &ipv6_hdr(skb)->saddr; 788 daddr = &ipv6_hdr(skb)->daddr; 789 uh = udp_hdr(skb); 790 791 ulen = ntohs(uh->len); 792 if (ulen > skb->len) 793 goto short_packet; 794 795 if (proto == IPPROTO_UDP) { 796 /* UDP validates ulen. */ 797 798 /* Check for jumbo payload */ 799 if (ulen == 0) 800 ulen = skb->len; 801 802 if (ulen < sizeof(*uh)) 803 goto short_packet; 804 805 if (ulen < skb->len) { 806 if (pskb_trim_rcsum(skb, ulen)) 807 goto short_packet; 808 saddr = &ipv6_hdr(skb)->saddr; 809 daddr = &ipv6_hdr(skb)->daddr; 810 uh = udp_hdr(skb); 811 } 812 } 813 814 if (udp6_csum_init(skb, uh, proto)) 815 goto csum_error; 816 817 /* Check if the socket is already available, e.g. due to early demux */ 818 sk = skb_steal_sock(skb); 819 if (sk) { 820 struct dst_entry *dst = skb_dst(skb); 821 int ret; 822 823 if (unlikely(sk->sk_rx_dst != dst)) 824 udp6_sk_rx_dst_set(sk, dst); 825 826 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 827 sock_put(sk); 828 goto report_csum_error; 829 } 830 831 ret = udp6_unicast_rcv_skb(sk, skb, uh); 832 sock_put(sk); 833 return ret; 834 } 835 836 /* 837 * Multicast receive code 838 */ 839 if (ipv6_addr_is_multicast(daddr)) 840 return __udp6_lib_mcast_deliver(net, skb, 841 saddr, daddr, udptable, proto); 842 843 /* Unicast */ 844 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 845 if (sk) { 846 if (!uh->check && !udp_sk(sk)->no_check6_rx) 847 goto report_csum_error; 848 return udp6_unicast_rcv_skb(sk, skb, uh); 849 } 850 851 if (!uh->check) 852 goto report_csum_error; 853 854 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 855 goto discard; 856 857 if (udp_lib_checksum_complete(skb)) 858 goto csum_error; 859 860 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 861 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 862 863 kfree_skb(skb); 864 return 0; 865 866 short_packet: 867 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 868 proto == IPPROTO_UDPLITE ? "-Lite" : "", 869 saddr, ntohs(uh->source), 870 ulen, skb->len, 871 daddr, ntohs(uh->dest)); 872 goto discard; 873 874 report_csum_error: 875 udp6_csum_zero_error(skb); 876 csum_error: 877 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 878 discard: 879 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 880 kfree_skb(skb); 881 return 0; 882 } 883 884 885 static struct sock *__udp6_lib_demux_lookup(struct net *net, 886 __be16 loc_port, const struct in6_addr *loc_addr, 887 __be16 rmt_port, const struct in6_addr *rmt_addr, 888 int dif, int sdif) 889 { 890 unsigned short hnum = ntohs(loc_port); 891 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 892 unsigned int slot2 = hash2 & udp_table.mask; 893 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 894 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 895 struct sock *sk; 896 897 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 898 if (sk->sk_state == TCP_ESTABLISHED && 899 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 900 return sk; 901 /* Only check first socket in chain */ 902 break; 903 } 904 return NULL; 905 } 906 907 static void udp_v6_early_demux(struct sk_buff *skb) 908 { 909 struct net *net = dev_net(skb->dev); 910 const struct udphdr *uh; 911 struct sock *sk; 912 struct dst_entry *dst; 913 int dif = skb->dev->ifindex; 914 int sdif = inet6_sdif(skb); 915 916 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 917 sizeof(struct udphdr))) 918 return; 919 920 uh = udp_hdr(skb); 921 922 if (skb->pkt_type == PACKET_HOST) 923 sk = __udp6_lib_demux_lookup(net, uh->dest, 924 &ipv6_hdr(skb)->daddr, 925 uh->source, &ipv6_hdr(skb)->saddr, 926 dif, sdif); 927 else 928 return; 929 930 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 931 return; 932 933 skb->sk = sk; 934 skb->destructor = sock_efree; 935 dst = READ_ONCE(sk->sk_rx_dst); 936 937 if (dst) 938 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 939 if (dst) { 940 /* set noref for now. 941 * any place which wants to hold dst has to call 942 * dst_hold_safe() 943 */ 944 skb_dst_set_noref(skb, dst); 945 } 946 } 947 948 static __inline__ int udpv6_rcv(struct sk_buff *skb) 949 { 950 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 951 } 952 953 /* 954 * Throw away all pending data and cancel the corking. Socket is locked. 955 */ 956 static void udp_v6_flush_pending_frames(struct sock *sk) 957 { 958 struct udp_sock *up = udp_sk(sk); 959 960 if (up->pending == AF_INET) 961 udp_flush_pending_frames(sk); 962 else if (up->pending) { 963 up->len = 0; 964 up->pending = 0; 965 ip6_flush_pending_frames(sk); 966 } 967 } 968 969 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 970 int addr_len) 971 { 972 /* The following checks are replicated from __ip6_datagram_connect() 973 * and intended to prevent BPF program called below from accessing 974 * bytes that are out of the bound specified by user in addr_len. 975 */ 976 if (uaddr->sa_family == AF_INET) { 977 if (__ipv6_only_sock(sk)) 978 return -EAFNOSUPPORT; 979 return udp_pre_connect(sk, uaddr, addr_len); 980 } 981 982 if (addr_len < SIN6_LEN_RFC2133) 983 return -EINVAL; 984 985 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 986 } 987 988 /** 989 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 990 * @sk: socket we are sending on 991 * @skb: sk_buff containing the filled-in UDP header 992 * (checksum field must be zeroed out) 993 */ 994 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 995 const struct in6_addr *saddr, 996 const struct in6_addr *daddr, int len) 997 { 998 unsigned int offset; 999 struct udphdr *uh = udp_hdr(skb); 1000 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1001 __wsum csum = 0; 1002 1003 if (!frags) { 1004 /* Only one fragment on the socket. */ 1005 skb->csum_start = skb_transport_header(skb) - skb->head; 1006 skb->csum_offset = offsetof(struct udphdr, check); 1007 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1008 } else { 1009 /* 1010 * HW-checksum won't work as there are two or more 1011 * fragments on the socket so that all csums of sk_buffs 1012 * should be together 1013 */ 1014 offset = skb_transport_offset(skb); 1015 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1016 csum = skb->csum; 1017 1018 skb->ip_summed = CHECKSUM_NONE; 1019 1020 do { 1021 csum = csum_add(csum, frags->csum); 1022 } while ((frags = frags->next)); 1023 1024 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1025 csum); 1026 if (uh->check == 0) 1027 uh->check = CSUM_MANGLED_0; 1028 } 1029 } 1030 1031 /* 1032 * Sending 1033 */ 1034 1035 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1036 struct inet_cork *cork) 1037 { 1038 struct sock *sk = skb->sk; 1039 struct udphdr *uh; 1040 int err = 0; 1041 int is_udplite = IS_UDPLITE(sk); 1042 __wsum csum = 0; 1043 int offset = skb_transport_offset(skb); 1044 int len = skb->len - offset; 1045 1046 /* 1047 * Create a UDP header 1048 */ 1049 uh = udp_hdr(skb); 1050 uh->source = fl6->fl6_sport; 1051 uh->dest = fl6->fl6_dport; 1052 uh->len = htons(len); 1053 uh->check = 0; 1054 1055 if (cork->gso_size) { 1056 const int hlen = skb_network_header_len(skb) + 1057 sizeof(struct udphdr); 1058 1059 if (hlen + cork->gso_size > cork->fragsize) 1060 return -EINVAL; 1061 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) 1062 return -EINVAL; 1063 if (udp_sk(sk)->no_check6_tx) 1064 return -EINVAL; 1065 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1066 dst_xfrm(skb_dst(skb))) 1067 return -EIO; 1068 1069 skb_shinfo(skb)->gso_size = cork->gso_size; 1070 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1071 goto csum_partial; 1072 } 1073 1074 if (is_udplite) 1075 csum = udplite_csum(skb); 1076 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1077 skb->ip_summed = CHECKSUM_NONE; 1078 goto send; 1079 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1080 csum_partial: 1081 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1082 goto send; 1083 } else 1084 csum = udp_csum(skb); 1085 1086 /* add protocol-dependent pseudo-header */ 1087 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1088 len, fl6->flowi6_proto, csum); 1089 if (uh->check == 0) 1090 uh->check = CSUM_MANGLED_0; 1091 1092 send: 1093 err = ip6_send_skb(skb); 1094 if (err) { 1095 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1096 UDP6_INC_STATS(sock_net(sk), 1097 UDP_MIB_SNDBUFERRORS, is_udplite); 1098 err = 0; 1099 } 1100 } else { 1101 UDP6_INC_STATS(sock_net(sk), 1102 UDP_MIB_OUTDATAGRAMS, is_udplite); 1103 } 1104 return err; 1105 } 1106 1107 static int udp_v6_push_pending_frames(struct sock *sk) 1108 { 1109 struct sk_buff *skb; 1110 struct udp_sock *up = udp_sk(sk); 1111 struct flowi6 fl6; 1112 int err = 0; 1113 1114 if (up->pending == AF_INET) 1115 return udp_push_pending_frames(sk); 1116 1117 /* ip6_finish_skb will release the cork, so make a copy of 1118 * fl6 here. 1119 */ 1120 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1121 1122 skb = ip6_finish_skb(sk); 1123 if (!skb) 1124 goto out; 1125 1126 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1127 1128 out: 1129 up->len = 0; 1130 up->pending = 0; 1131 return err; 1132 } 1133 1134 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1135 { 1136 struct ipv6_txoptions opt_space; 1137 struct udp_sock *up = udp_sk(sk); 1138 struct inet_sock *inet = inet_sk(sk); 1139 struct ipv6_pinfo *np = inet6_sk(sk); 1140 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1141 struct in6_addr *daddr, *final_p, final; 1142 struct ipv6_txoptions *opt = NULL; 1143 struct ipv6_txoptions *opt_to_free = NULL; 1144 struct ip6_flowlabel *flowlabel = NULL; 1145 struct flowi6 fl6; 1146 struct dst_entry *dst; 1147 struct ipcm6_cookie ipc6; 1148 int addr_len = msg->msg_namelen; 1149 bool connected = false; 1150 int ulen = len; 1151 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1152 int err; 1153 int is_udplite = IS_UDPLITE(sk); 1154 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1155 1156 ipcm6_init(&ipc6); 1157 ipc6.gso_size = up->gso_size; 1158 ipc6.sockc.tsflags = sk->sk_tsflags; 1159 1160 /* destination address check */ 1161 if (sin6) { 1162 if (addr_len < offsetof(struct sockaddr, sa_data)) 1163 return -EINVAL; 1164 1165 switch (sin6->sin6_family) { 1166 case AF_INET6: 1167 if (addr_len < SIN6_LEN_RFC2133) 1168 return -EINVAL; 1169 daddr = &sin6->sin6_addr; 1170 if (ipv6_addr_any(daddr) && 1171 ipv6_addr_v4mapped(&np->saddr)) 1172 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1173 daddr); 1174 break; 1175 case AF_INET: 1176 goto do_udp_sendmsg; 1177 case AF_UNSPEC: 1178 msg->msg_name = sin6 = NULL; 1179 msg->msg_namelen = addr_len = 0; 1180 daddr = NULL; 1181 break; 1182 default: 1183 return -EINVAL; 1184 } 1185 } else if (!up->pending) { 1186 if (sk->sk_state != TCP_ESTABLISHED) 1187 return -EDESTADDRREQ; 1188 daddr = &sk->sk_v6_daddr; 1189 } else 1190 daddr = NULL; 1191 1192 if (daddr) { 1193 if (ipv6_addr_v4mapped(daddr)) { 1194 struct sockaddr_in sin; 1195 sin.sin_family = AF_INET; 1196 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1197 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1198 msg->msg_name = &sin; 1199 msg->msg_namelen = sizeof(sin); 1200 do_udp_sendmsg: 1201 if (__ipv6_only_sock(sk)) 1202 return -ENETUNREACH; 1203 return udp_sendmsg(sk, msg, len); 1204 } 1205 } 1206 1207 if (up->pending == AF_INET) 1208 return udp_sendmsg(sk, msg, len); 1209 1210 /* Rough check on arithmetic overflow, 1211 better check is made in ip6_append_data(). 1212 */ 1213 if (len > INT_MAX - sizeof(struct udphdr)) 1214 return -EMSGSIZE; 1215 1216 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1217 if (up->pending) { 1218 /* 1219 * There are pending frames. 1220 * The socket lock must be held while it's corked. 1221 */ 1222 lock_sock(sk); 1223 if (likely(up->pending)) { 1224 if (unlikely(up->pending != AF_INET6)) { 1225 release_sock(sk); 1226 return -EAFNOSUPPORT; 1227 } 1228 dst = NULL; 1229 goto do_append_data; 1230 } 1231 release_sock(sk); 1232 } 1233 ulen += sizeof(struct udphdr); 1234 1235 memset(&fl6, 0, sizeof(fl6)); 1236 1237 if (sin6) { 1238 if (sin6->sin6_port == 0) 1239 return -EINVAL; 1240 1241 fl6.fl6_dport = sin6->sin6_port; 1242 daddr = &sin6->sin6_addr; 1243 1244 if (np->sndflow) { 1245 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1246 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1247 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1248 if (!flowlabel) 1249 return -EINVAL; 1250 } 1251 } 1252 1253 /* 1254 * Otherwise it will be difficult to maintain 1255 * sk->sk_dst_cache. 1256 */ 1257 if (sk->sk_state == TCP_ESTABLISHED && 1258 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1259 daddr = &sk->sk_v6_daddr; 1260 1261 if (addr_len >= sizeof(struct sockaddr_in6) && 1262 sin6->sin6_scope_id && 1263 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1264 fl6.flowi6_oif = sin6->sin6_scope_id; 1265 } else { 1266 if (sk->sk_state != TCP_ESTABLISHED) 1267 return -EDESTADDRREQ; 1268 1269 fl6.fl6_dport = inet->inet_dport; 1270 daddr = &sk->sk_v6_daddr; 1271 fl6.flowlabel = np->flow_label; 1272 connected = true; 1273 } 1274 1275 if (!fl6.flowi6_oif) 1276 fl6.flowi6_oif = sk->sk_bound_dev_if; 1277 1278 if (!fl6.flowi6_oif) 1279 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1280 1281 fl6.flowi6_mark = sk->sk_mark; 1282 fl6.flowi6_uid = sk->sk_uid; 1283 1284 if (msg->msg_controllen) { 1285 opt = &opt_space; 1286 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1287 opt->tot_len = sizeof(*opt); 1288 ipc6.opt = opt; 1289 1290 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1291 if (err > 0) 1292 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1293 &ipc6); 1294 if (err < 0) { 1295 fl6_sock_release(flowlabel); 1296 return err; 1297 } 1298 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1299 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1300 if (!flowlabel) 1301 return -EINVAL; 1302 } 1303 if (!(opt->opt_nflen|opt->opt_flen)) 1304 opt = NULL; 1305 connected = false; 1306 } 1307 if (!opt) { 1308 opt = txopt_get(np); 1309 opt_to_free = opt; 1310 } 1311 if (flowlabel) 1312 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1313 opt = ipv6_fixup_options(&opt_space, opt); 1314 ipc6.opt = opt; 1315 1316 fl6.flowi6_proto = sk->sk_protocol; 1317 if (!ipv6_addr_any(daddr)) 1318 fl6.daddr = *daddr; 1319 else 1320 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1321 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1322 fl6.saddr = np->saddr; 1323 fl6.fl6_sport = inet->inet_sport; 1324 1325 if (cgroup_bpf_enabled && !connected) { 1326 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1327 (struct sockaddr *)sin6, &fl6.saddr); 1328 if (err) 1329 goto out_no_dst; 1330 if (sin6) { 1331 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1332 /* BPF program rewrote IPv6-only by IPv4-mapped 1333 * IPv6. It's currently unsupported. 1334 */ 1335 err = -ENOTSUPP; 1336 goto out_no_dst; 1337 } 1338 if (sin6->sin6_port == 0) { 1339 /* BPF program set invalid port. Reject it. */ 1340 err = -EINVAL; 1341 goto out_no_dst; 1342 } 1343 fl6.fl6_dport = sin6->sin6_port; 1344 fl6.daddr = sin6->sin6_addr; 1345 } 1346 } 1347 1348 final_p = fl6_update_dst(&fl6, opt, &final); 1349 if (final_p) 1350 connected = false; 1351 1352 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1353 fl6.flowi6_oif = np->mcast_oif; 1354 connected = false; 1355 } else if (!fl6.flowi6_oif) 1356 fl6.flowi6_oif = np->ucast_oif; 1357 1358 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1359 1360 if (ipc6.tclass < 0) 1361 ipc6.tclass = np->tclass; 1362 1363 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1364 1365 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1366 if (IS_ERR(dst)) { 1367 err = PTR_ERR(dst); 1368 dst = NULL; 1369 goto out; 1370 } 1371 1372 if (ipc6.hlimit < 0) 1373 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1374 1375 if (msg->msg_flags&MSG_CONFIRM) 1376 goto do_confirm; 1377 back_from_confirm: 1378 1379 /* Lockless fast path for the non-corking case */ 1380 if (!corkreq) { 1381 struct inet_cork_full cork; 1382 struct sk_buff *skb; 1383 1384 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1385 sizeof(struct udphdr), &ipc6, 1386 &fl6, (struct rt6_info *)dst, 1387 msg->msg_flags, &cork); 1388 err = PTR_ERR(skb); 1389 if (!IS_ERR_OR_NULL(skb)) 1390 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1391 goto out; 1392 } 1393 1394 lock_sock(sk); 1395 if (unlikely(up->pending)) { 1396 /* The socket is already corked while preparing it. */ 1397 /* ... which is an evident application bug. --ANK */ 1398 release_sock(sk); 1399 1400 net_dbg_ratelimited("udp cork app bug 2\n"); 1401 err = -EINVAL; 1402 goto out; 1403 } 1404 1405 up->pending = AF_INET6; 1406 1407 do_append_data: 1408 if (ipc6.dontfrag < 0) 1409 ipc6.dontfrag = np->dontfrag; 1410 up->len += ulen; 1411 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1412 &ipc6, &fl6, (struct rt6_info *)dst, 1413 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1414 if (err) 1415 udp_v6_flush_pending_frames(sk); 1416 else if (!corkreq) 1417 err = udp_v6_push_pending_frames(sk); 1418 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1419 up->pending = 0; 1420 1421 if (err > 0) 1422 err = np->recverr ? net_xmit_errno(err) : 0; 1423 release_sock(sk); 1424 1425 out: 1426 dst_release(dst); 1427 out_no_dst: 1428 fl6_sock_release(flowlabel); 1429 txopt_put(opt_to_free); 1430 if (!err) 1431 return len; 1432 /* 1433 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1434 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1435 * we don't have a good statistic (IpOutDiscards but it can be too many 1436 * things). We could add another new stat but at least for now that 1437 * seems like overkill. 1438 */ 1439 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1440 UDP6_INC_STATS(sock_net(sk), 1441 UDP_MIB_SNDBUFERRORS, is_udplite); 1442 } 1443 return err; 1444 1445 do_confirm: 1446 if (msg->msg_flags & MSG_PROBE) 1447 dst_confirm_neigh(dst, &fl6.daddr); 1448 if (!(msg->msg_flags&MSG_PROBE) || len) 1449 goto back_from_confirm; 1450 err = 0; 1451 goto out; 1452 } 1453 1454 void udpv6_destroy_sock(struct sock *sk) 1455 { 1456 struct udp_sock *up = udp_sk(sk); 1457 lock_sock(sk); 1458 udp_v6_flush_pending_frames(sk); 1459 release_sock(sk); 1460 1461 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 1462 void (*encap_destroy)(struct sock *sk); 1463 encap_destroy = READ_ONCE(up->encap_destroy); 1464 if (encap_destroy) 1465 encap_destroy(sk); 1466 } 1467 1468 inet6_destroy_sock(sk); 1469 } 1470 1471 /* 1472 * Socket option code for UDP 1473 */ 1474 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1475 char __user *optval, unsigned int optlen) 1476 { 1477 if (level == SOL_UDP || level == SOL_UDPLITE) 1478 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1479 udp_v6_push_pending_frames); 1480 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1481 } 1482 1483 #ifdef CONFIG_COMPAT 1484 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1485 char __user *optval, unsigned int optlen) 1486 { 1487 if (level == SOL_UDP || level == SOL_UDPLITE) 1488 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1489 udp_v6_push_pending_frames); 1490 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1491 } 1492 #endif 1493 1494 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1495 char __user *optval, int __user *optlen) 1496 { 1497 if (level == SOL_UDP || level == SOL_UDPLITE) 1498 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1499 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1500 } 1501 1502 #ifdef CONFIG_COMPAT 1503 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1504 char __user *optval, int __user *optlen) 1505 { 1506 if (level == SOL_UDP || level == SOL_UDPLITE) 1507 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1508 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1509 } 1510 #endif 1511 1512 /* thinking of making this const? Don't. 1513 * early_demux can change based on sysctl. 1514 */ 1515 static struct inet6_protocol udpv6_protocol = { 1516 .early_demux = udp_v6_early_demux, 1517 .early_demux_handler = udp_v6_early_demux, 1518 .handler = udpv6_rcv, 1519 .err_handler = udpv6_err, 1520 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1521 }; 1522 1523 /* ------------------------------------------------------------------------ */ 1524 #ifdef CONFIG_PROC_FS 1525 int udp6_seq_show(struct seq_file *seq, void *v) 1526 { 1527 if (v == SEQ_START_TOKEN) { 1528 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1529 } else { 1530 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1531 struct inet_sock *inet = inet_sk(v); 1532 __u16 srcp = ntohs(inet->inet_sport); 1533 __u16 destp = ntohs(inet->inet_dport); 1534 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1535 udp_rqueue_get(v), bucket); 1536 } 1537 return 0; 1538 } 1539 1540 const struct seq_operations udp6_seq_ops = { 1541 .start = udp_seq_start, 1542 .next = udp_seq_next, 1543 .stop = udp_seq_stop, 1544 .show = udp6_seq_show, 1545 }; 1546 EXPORT_SYMBOL(udp6_seq_ops); 1547 1548 static struct udp_seq_afinfo udp6_seq_afinfo = { 1549 .family = AF_INET6, 1550 .udp_table = &udp_table, 1551 }; 1552 1553 int __net_init udp6_proc_init(struct net *net) 1554 { 1555 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1556 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1557 return -ENOMEM; 1558 return 0; 1559 } 1560 1561 void udp6_proc_exit(struct net *net) 1562 { 1563 remove_proc_entry("udp6", net->proc_net); 1564 } 1565 #endif /* CONFIG_PROC_FS */ 1566 1567 /* ------------------------------------------------------------------------ */ 1568 1569 struct proto udpv6_prot = { 1570 .name = "UDPv6", 1571 .owner = THIS_MODULE, 1572 .close = udp_lib_close, 1573 .pre_connect = udpv6_pre_connect, 1574 .connect = ip6_datagram_connect, 1575 .disconnect = udp_disconnect, 1576 .ioctl = udp_ioctl, 1577 .init = udp_init_sock, 1578 .destroy = udpv6_destroy_sock, 1579 .setsockopt = udpv6_setsockopt, 1580 .getsockopt = udpv6_getsockopt, 1581 .sendmsg = udpv6_sendmsg, 1582 .recvmsg = udpv6_recvmsg, 1583 .release_cb = ip6_datagram_release_cb, 1584 .hash = udp_lib_hash, 1585 .unhash = udp_lib_unhash, 1586 .rehash = udp_v6_rehash, 1587 .get_port = udp_v6_get_port, 1588 .memory_allocated = &udp_memory_allocated, 1589 .sysctl_mem = sysctl_udp_mem, 1590 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1591 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1592 .obj_size = sizeof(struct udp6_sock), 1593 .h.udp_table = &udp_table, 1594 #ifdef CONFIG_COMPAT 1595 .compat_setsockopt = compat_udpv6_setsockopt, 1596 .compat_getsockopt = compat_udpv6_getsockopt, 1597 #endif 1598 .diag_destroy = udp_abort, 1599 }; 1600 1601 static struct inet_protosw udpv6_protosw = { 1602 .type = SOCK_DGRAM, 1603 .protocol = IPPROTO_UDP, 1604 .prot = &udpv6_prot, 1605 .ops = &inet6_dgram_ops, 1606 .flags = INET_PROTOSW_PERMANENT, 1607 }; 1608 1609 int __init udpv6_init(void) 1610 { 1611 int ret; 1612 1613 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1614 if (ret) 1615 goto out; 1616 1617 ret = inet6_register_protosw(&udpv6_protosw); 1618 if (ret) 1619 goto out_udpv6_protocol; 1620 out: 1621 return ret; 1622 1623 out_udpv6_protocol: 1624 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1625 goto out; 1626 } 1627 1628 void udpv6_exit(void) 1629 { 1630 inet6_unregister_protosw(&udpv6_protosw); 1631 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1632 } 1633