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 (exact_dif && !dev_match) 152 return -1; 153 if (sk->sk_bound_dev_if && dev_match) 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 return result; 239 } 240 begin: 241 result = NULL; 242 badness = -1; 243 sk_for_each_rcu(sk, &hslot->head) { 244 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif, 245 sdif, exact_dif); 246 if (score > badness) { 247 if (sk->sk_reuseport) { 248 hash = udp6_ehashfn(net, daddr, hnum, 249 saddr, sport); 250 result = reuseport_select_sock(sk, hash, skb, 251 sizeof(struct udphdr)); 252 if (result) 253 return result; 254 } 255 result = sk; 256 badness = score; 257 } 258 } 259 return result; 260 } 261 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 262 263 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 264 __be16 sport, __be16 dport, 265 struct udp_table *udptable) 266 { 267 const struct ipv6hdr *iph = ipv6_hdr(skb); 268 269 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 270 &iph->daddr, dport, inet6_iif(skb), 271 inet6_sdif(skb), udptable, skb); 272 } 273 274 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, 275 __be16 sport, __be16 dport) 276 { 277 const struct ipv6hdr *iph = ipv6_hdr(skb); 278 279 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 280 &iph->daddr, dport, inet6_iif(skb), 281 inet6_sdif(skb), &udp_table, skb); 282 } 283 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); 284 285 /* Must be called under rcu_read_lock(). 286 * Does increment socket refcount. 287 */ 288 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \ 289 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \ 290 IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 291 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 292 const struct in6_addr *daddr, __be16 dport, int dif) 293 { 294 struct sock *sk; 295 296 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 297 dif, 0, &udp_table, NULL); 298 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 299 sk = NULL; 300 return sk; 301 } 302 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 303 #endif 304 305 /* do not use the scratch area len for jumbogram: their length execeeds the 306 * scratch area space; note that the IP6CB flags is still in the first 307 * cacheline, so checking for jumbograms is cheap 308 */ 309 static int udp6_skb_len(struct sk_buff *skb) 310 { 311 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 312 } 313 314 /* 315 * This should be easy, if there is something there we 316 * return it, otherwise we block. 317 */ 318 319 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 320 int noblock, int flags, int *addr_len) 321 { 322 struct ipv6_pinfo *np = inet6_sk(sk); 323 struct inet_sock *inet = inet_sk(sk); 324 struct sk_buff *skb; 325 unsigned int ulen, copied; 326 int peeked, peeking, off; 327 int err; 328 int is_udplite = IS_UDPLITE(sk); 329 bool checksum_valid = false; 330 int is_udp4; 331 332 if (flags & MSG_ERRQUEUE) 333 return ipv6_recv_error(sk, msg, len, addr_len); 334 335 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 336 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 337 338 try_again: 339 peeking = flags & MSG_PEEK; 340 off = sk_peek_offset(sk, flags); 341 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); 342 if (!skb) 343 return err; 344 345 ulen = udp6_skb_len(skb); 346 copied = len; 347 if (copied > ulen - off) 348 copied = ulen - off; 349 else if (copied < ulen) 350 msg->msg_flags |= MSG_TRUNC; 351 352 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 353 354 /* 355 * If checksum is needed at all, try to do it while copying the 356 * data. If the data is truncated, or if we only want a partial 357 * coverage checksum (UDP-Lite), do it before the copy. 358 */ 359 360 if (copied < ulen || peeking || 361 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 362 checksum_valid = udp_skb_csum_unnecessary(skb) || 363 !__udp_lib_checksum_complete(skb); 364 if (!checksum_valid) 365 goto csum_copy_err; 366 } 367 368 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 369 if (udp_skb_is_linear(skb)) 370 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 371 else 372 err = skb_copy_datagram_msg(skb, off, msg, copied); 373 } else { 374 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 375 if (err == -EINVAL) 376 goto csum_copy_err; 377 } 378 if (unlikely(err)) { 379 if (!peeked) { 380 atomic_inc(&sk->sk_drops); 381 if (is_udp4) 382 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 383 is_udplite); 384 else 385 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 386 is_udplite); 387 } 388 kfree_skb(skb); 389 return err; 390 } 391 if (!peeked) { 392 if (is_udp4) 393 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, 394 is_udplite); 395 else 396 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, 397 is_udplite); 398 } 399 400 sock_recv_ts_and_drops(msg, sk, skb); 401 402 /* Copy the address. */ 403 if (msg->msg_name) { 404 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 405 sin6->sin6_family = AF_INET6; 406 sin6->sin6_port = udp_hdr(skb)->source; 407 sin6->sin6_flowinfo = 0; 408 409 if (is_udp4) { 410 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 411 &sin6->sin6_addr); 412 sin6->sin6_scope_id = 0; 413 } else { 414 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 415 sin6->sin6_scope_id = 416 ipv6_iface_scope_id(&sin6->sin6_addr, 417 inet6_iif(skb)); 418 } 419 *addr_len = sizeof(*sin6); 420 } 421 422 if (np->rxopt.all) 423 ip6_datagram_recv_common_ctl(sk, msg, skb); 424 425 if (is_udp4) { 426 if (inet->cmsg_flags) 427 ip_cmsg_recv_offset(msg, sk, skb, 428 sizeof(struct udphdr), off); 429 } else { 430 if (np->rxopt.all) 431 ip6_datagram_recv_specific_ctl(sk, msg, skb); 432 } 433 434 err = copied; 435 if (flags & MSG_TRUNC) 436 err = ulen; 437 438 skb_consume_udp(sk, skb, peeking ? -err : err); 439 return err; 440 441 csum_copy_err: 442 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 443 udp_skb_destructor)) { 444 if (is_udp4) { 445 UDP_INC_STATS(sock_net(sk), 446 UDP_MIB_CSUMERRORS, is_udplite); 447 UDP_INC_STATS(sock_net(sk), 448 UDP_MIB_INERRORS, is_udplite); 449 } else { 450 UDP6_INC_STATS(sock_net(sk), 451 UDP_MIB_CSUMERRORS, is_udplite); 452 UDP6_INC_STATS(sock_net(sk), 453 UDP_MIB_INERRORS, is_udplite); 454 } 455 } 456 kfree_skb(skb); 457 458 /* starting over for a new packet, but check if we need to yield */ 459 cond_resched(); 460 msg->msg_flags &= ~MSG_TRUNC; 461 goto try_again; 462 } 463 464 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 465 u8 type, u8 code, int offset, __be32 info, 466 struct udp_table *udptable) 467 { 468 struct ipv6_pinfo *np; 469 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 470 const struct in6_addr *saddr = &hdr->saddr; 471 const struct in6_addr *daddr = &hdr->daddr; 472 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 473 struct sock *sk; 474 int harderr; 475 int err; 476 struct net *net = dev_net(skb->dev); 477 478 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 479 inet6_iif(skb), 0, udptable, skb); 480 if (!sk) { 481 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 482 ICMP6_MIB_INERRORS); 483 return; 484 } 485 486 harderr = icmpv6_err_convert(type, code, &err); 487 np = inet6_sk(sk); 488 489 if (type == ICMPV6_PKT_TOOBIG) { 490 if (!ip6_sk_accept_pmtu(sk)) 491 goto out; 492 ip6_sk_update_pmtu(skb, sk, info); 493 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 494 harderr = 1; 495 } 496 if (type == NDISC_REDIRECT) { 497 ip6_sk_redirect(skb, sk); 498 goto out; 499 } 500 501 if (!np->recverr) { 502 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 503 goto out; 504 } else { 505 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 506 } 507 508 sk->sk_err = err; 509 sk->sk_error_report(sk); 510 out: 511 return; 512 } 513 514 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 515 { 516 int rc; 517 518 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 519 sock_rps_save_rxhash(sk, skb); 520 sk_mark_napi_id(sk, skb); 521 sk_incoming_cpu_update(sk); 522 } else { 523 sk_mark_napi_id_once(sk, skb); 524 } 525 526 rc = __udp_enqueue_schedule_skb(sk, skb); 527 if (rc < 0) { 528 int is_udplite = IS_UDPLITE(sk); 529 530 /* Note that an ENOMEM error is charged twice */ 531 if (rc == -ENOMEM) 532 UDP6_INC_STATS(sock_net(sk), 533 UDP_MIB_RCVBUFERRORS, is_udplite); 534 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 535 kfree_skb(skb); 536 return -1; 537 } 538 539 return 0; 540 } 541 542 static __inline__ void udpv6_err(struct sk_buff *skb, 543 struct inet6_skb_parm *opt, u8 type, 544 u8 code, int offset, __be32 info) 545 { 546 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 547 } 548 549 static struct static_key udpv6_encap_needed __read_mostly; 550 void udpv6_encap_enable(void) 551 { 552 static_key_enable(&udpv6_encap_needed); 553 } 554 EXPORT_SYMBOL(udpv6_encap_enable); 555 556 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 557 { 558 struct udp_sock *up = udp_sk(sk); 559 int is_udplite = IS_UDPLITE(sk); 560 561 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 562 goto drop; 563 564 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 565 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 566 567 /* 568 * This is an encapsulation socket so pass the skb to 569 * the socket's udp_encap_rcv() hook. Otherwise, just 570 * fall through and pass this up the UDP socket. 571 * up->encap_rcv() returns the following value: 572 * =0 if skb was successfully passed to the encap 573 * handler or was discarded by it. 574 * >0 if skb should be passed on to UDP. 575 * <0 if skb should be resubmitted as proto -N 576 */ 577 578 /* if we're overly short, let UDP handle it */ 579 encap_rcv = READ_ONCE(up->encap_rcv); 580 if (encap_rcv) { 581 int ret; 582 583 /* Verify checksum before giving to encap */ 584 if (udp_lib_checksum_complete(skb)) 585 goto csum_error; 586 587 ret = encap_rcv(sk, skb); 588 if (ret <= 0) { 589 __UDP_INC_STATS(sock_net(sk), 590 UDP_MIB_INDATAGRAMS, 591 is_udplite); 592 return -ret; 593 } 594 } 595 596 /* FALLTHROUGH -- it's a UDP Packet */ 597 } 598 599 /* 600 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 601 */ 602 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 603 604 if (up->pcrlen == 0) { /* full coverage was set */ 605 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 606 UDP_SKB_CB(skb)->cscov, skb->len); 607 goto drop; 608 } 609 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 610 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 611 UDP_SKB_CB(skb)->cscov, up->pcrlen); 612 goto drop; 613 } 614 } 615 616 prefetch(&sk->sk_rmem_alloc); 617 if (rcu_access_pointer(sk->sk_filter) && 618 udp_lib_checksum_complete(skb)) 619 goto csum_error; 620 621 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 622 goto drop; 623 624 udp_csum_pull_header(skb); 625 626 skb_dst_drop(skb); 627 628 return __udpv6_queue_rcv_skb(sk, skb); 629 630 csum_error: 631 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 632 drop: 633 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 634 atomic_inc(&sk->sk_drops); 635 kfree_skb(skb); 636 return -1; 637 } 638 639 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 640 __be16 loc_port, const struct in6_addr *loc_addr, 641 __be16 rmt_port, const struct in6_addr *rmt_addr, 642 int dif, unsigned short hnum) 643 { 644 struct inet_sock *inet = inet_sk(sk); 645 646 if (!net_eq(sock_net(sk), net)) 647 return false; 648 649 if (udp_sk(sk)->udp_port_hash != hnum || 650 sk->sk_family != PF_INET6 || 651 (inet->inet_dport && inet->inet_dport != rmt_port) || 652 (!ipv6_addr_any(&sk->sk_v6_daddr) && 653 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 654 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) || 655 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 656 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 657 return false; 658 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 659 return false; 660 return true; 661 } 662 663 static void udp6_csum_zero_error(struct sk_buff *skb) 664 { 665 /* RFC 2460 section 8.1 says that we SHOULD log 666 * this error. Well, it is reasonable. 667 */ 668 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 669 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 670 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 671 } 672 673 /* 674 * Note: called only from the BH handler context, 675 * so we don't need to lock the hashes. 676 */ 677 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 678 const struct in6_addr *saddr, const struct in6_addr *daddr, 679 struct udp_table *udptable, int proto) 680 { 681 struct sock *sk, *first = NULL; 682 const struct udphdr *uh = udp_hdr(skb); 683 unsigned short hnum = ntohs(uh->dest); 684 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 685 unsigned int offset = offsetof(typeof(*sk), sk_node); 686 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 687 int dif = inet6_iif(skb); 688 struct hlist_node *node; 689 struct sk_buff *nskb; 690 691 if (use_hash2) { 692 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 693 udptable->mask; 694 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 695 start_lookup: 696 hslot = &udptable->hash2[hash2]; 697 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 698 } 699 700 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 701 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 702 uh->source, saddr, dif, hnum)) 703 continue; 704 /* If zero checksum and no_check is not on for 705 * the socket then skip it. 706 */ 707 if (!uh->check && !udp_sk(sk)->no_check6_rx) 708 continue; 709 if (!first) { 710 first = sk; 711 continue; 712 } 713 nskb = skb_clone(skb, GFP_ATOMIC); 714 if (unlikely(!nskb)) { 715 atomic_inc(&sk->sk_drops); 716 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 717 IS_UDPLITE(sk)); 718 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 719 IS_UDPLITE(sk)); 720 continue; 721 } 722 723 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 724 consume_skb(nskb); 725 } 726 727 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 728 if (use_hash2 && hash2 != hash2_any) { 729 hash2 = hash2_any; 730 goto start_lookup; 731 } 732 733 if (first) { 734 if (udpv6_queue_rcv_skb(first, skb) > 0) 735 consume_skb(skb); 736 } else { 737 kfree_skb(skb); 738 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 739 proto == IPPROTO_UDPLITE); 740 } 741 return 0; 742 } 743 744 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 745 { 746 if (udp_sk_rx_dst_set(sk, dst)) { 747 const struct rt6_info *rt = (const struct rt6_info *)dst; 748 749 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 750 } 751 } 752 753 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 754 int proto) 755 { 756 const struct in6_addr *saddr, *daddr; 757 struct net *net = dev_net(skb->dev); 758 struct udphdr *uh; 759 struct sock *sk; 760 u32 ulen = 0; 761 762 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 763 goto discard; 764 765 saddr = &ipv6_hdr(skb)->saddr; 766 daddr = &ipv6_hdr(skb)->daddr; 767 uh = udp_hdr(skb); 768 769 ulen = ntohs(uh->len); 770 if (ulen > skb->len) 771 goto short_packet; 772 773 if (proto == IPPROTO_UDP) { 774 /* UDP validates ulen. */ 775 776 /* Check for jumbo payload */ 777 if (ulen == 0) 778 ulen = skb->len; 779 780 if (ulen < sizeof(*uh)) 781 goto short_packet; 782 783 if (ulen < skb->len) { 784 if (pskb_trim_rcsum(skb, ulen)) 785 goto short_packet; 786 saddr = &ipv6_hdr(skb)->saddr; 787 daddr = &ipv6_hdr(skb)->daddr; 788 uh = udp_hdr(skb); 789 } 790 } 791 792 if (udp6_csum_init(skb, uh, proto)) 793 goto csum_error; 794 795 /* Check if the socket is already available, e.g. due to early demux */ 796 sk = skb_steal_sock(skb); 797 if (sk) { 798 struct dst_entry *dst = skb_dst(skb); 799 int ret; 800 801 if (unlikely(sk->sk_rx_dst != dst)) 802 udp6_sk_rx_dst_set(sk, dst); 803 804 ret = udpv6_queue_rcv_skb(sk, skb); 805 sock_put(sk); 806 807 /* a return value > 0 means to resubmit the input */ 808 if (ret > 0) 809 return ret; 810 return 0; 811 } 812 813 /* 814 * Multicast receive code 815 */ 816 if (ipv6_addr_is_multicast(daddr)) 817 return __udp6_lib_mcast_deliver(net, skb, 818 saddr, daddr, udptable, proto); 819 820 /* Unicast */ 821 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 822 if (sk) { 823 int ret; 824 825 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 826 udp6_csum_zero_error(skb); 827 goto csum_error; 828 } 829 830 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 831 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 832 ip6_compute_pseudo); 833 834 ret = udpv6_queue_rcv_skb(sk, skb); 835 836 /* a return value > 0 means to resubmit the input */ 837 if (ret > 0) 838 return ret; 839 840 return 0; 841 } 842 843 if (!uh->check) { 844 udp6_csum_zero_error(skb); 845 goto csum_error; 846 } 847 848 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 849 goto discard; 850 851 if (udp_lib_checksum_complete(skb)) 852 goto csum_error; 853 854 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 855 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 856 857 kfree_skb(skb); 858 return 0; 859 860 short_packet: 861 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 862 proto == IPPROTO_UDPLITE ? "-Lite" : "", 863 saddr, ntohs(uh->source), 864 ulen, skb->len, 865 daddr, ntohs(uh->dest)); 866 goto discard; 867 csum_error: 868 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 869 discard: 870 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 871 kfree_skb(skb); 872 return 0; 873 } 874 875 876 static struct sock *__udp6_lib_demux_lookup(struct net *net, 877 __be16 loc_port, const struct in6_addr *loc_addr, 878 __be16 rmt_port, const struct in6_addr *rmt_addr, 879 int dif, int sdif) 880 { 881 unsigned short hnum = ntohs(loc_port); 882 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 883 unsigned int slot2 = hash2 & udp_table.mask; 884 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 885 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 886 struct sock *sk; 887 888 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 889 if (sk->sk_state == TCP_ESTABLISHED && 890 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 891 return sk; 892 /* Only check first socket in chain */ 893 break; 894 } 895 return NULL; 896 } 897 898 static void udp_v6_early_demux(struct sk_buff *skb) 899 { 900 struct net *net = dev_net(skb->dev); 901 const struct udphdr *uh; 902 struct sock *sk; 903 struct dst_entry *dst; 904 int dif = skb->dev->ifindex; 905 int sdif = inet6_sdif(skb); 906 907 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 908 sizeof(struct udphdr))) 909 return; 910 911 uh = udp_hdr(skb); 912 913 if (skb->pkt_type == PACKET_HOST) 914 sk = __udp6_lib_demux_lookup(net, uh->dest, 915 &ipv6_hdr(skb)->daddr, 916 uh->source, &ipv6_hdr(skb)->saddr, 917 dif, sdif); 918 else 919 return; 920 921 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 922 return; 923 924 skb->sk = sk; 925 skb->destructor = sock_efree; 926 dst = READ_ONCE(sk->sk_rx_dst); 927 928 if (dst) 929 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 930 if (dst) { 931 /* set noref for now. 932 * any place which wants to hold dst has to call 933 * dst_hold_safe() 934 */ 935 skb_dst_set_noref(skb, dst); 936 } 937 } 938 939 static __inline__ int udpv6_rcv(struct sk_buff *skb) 940 { 941 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 942 } 943 944 /* 945 * Throw away all pending data and cancel the corking. Socket is locked. 946 */ 947 static void udp_v6_flush_pending_frames(struct sock *sk) 948 { 949 struct udp_sock *up = udp_sk(sk); 950 951 if (up->pending == AF_INET) 952 udp_flush_pending_frames(sk); 953 else if (up->pending) { 954 up->len = 0; 955 up->pending = 0; 956 ip6_flush_pending_frames(sk); 957 } 958 } 959 960 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 961 int addr_len) 962 { 963 /* The following checks are replicated from __ip6_datagram_connect() 964 * and intended to prevent BPF program called below from accessing 965 * bytes that are out of the bound specified by user in addr_len. 966 */ 967 if (uaddr->sa_family == AF_INET) { 968 if (__ipv6_only_sock(sk)) 969 return -EAFNOSUPPORT; 970 return udp_pre_connect(sk, uaddr, addr_len); 971 } 972 973 if (addr_len < SIN6_LEN_RFC2133) 974 return -EINVAL; 975 976 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 977 } 978 979 /** 980 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 981 * @sk: socket we are sending on 982 * @skb: sk_buff containing the filled-in UDP header 983 * (checksum field must be zeroed out) 984 */ 985 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 986 const struct in6_addr *saddr, 987 const struct in6_addr *daddr, int len) 988 { 989 unsigned int offset; 990 struct udphdr *uh = udp_hdr(skb); 991 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 992 __wsum csum = 0; 993 994 if (!frags) { 995 /* Only one fragment on the socket. */ 996 skb->csum_start = skb_transport_header(skb) - skb->head; 997 skb->csum_offset = offsetof(struct udphdr, check); 998 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 999 } else { 1000 /* 1001 * HW-checksum won't work as there are two or more 1002 * fragments on the socket so that all csums of sk_buffs 1003 * should be together 1004 */ 1005 offset = skb_transport_offset(skb); 1006 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1007 csum = skb->csum; 1008 1009 skb->ip_summed = CHECKSUM_NONE; 1010 1011 do { 1012 csum = csum_add(csum, frags->csum); 1013 } while ((frags = frags->next)); 1014 1015 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1016 csum); 1017 if (uh->check == 0) 1018 uh->check = CSUM_MANGLED_0; 1019 } 1020 } 1021 1022 /* 1023 * Sending 1024 */ 1025 1026 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6) 1027 { 1028 struct sock *sk = skb->sk; 1029 struct udphdr *uh; 1030 int err = 0; 1031 int is_udplite = IS_UDPLITE(sk); 1032 __wsum csum = 0; 1033 int offset = skb_transport_offset(skb); 1034 int len = skb->len - offset; 1035 1036 /* 1037 * Create a UDP header 1038 */ 1039 uh = udp_hdr(skb); 1040 uh->source = fl6->fl6_sport; 1041 uh->dest = fl6->fl6_dport; 1042 uh->len = htons(len); 1043 uh->check = 0; 1044 1045 if (is_udplite) 1046 csum = udplite_csum(skb); 1047 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1048 skb->ip_summed = CHECKSUM_NONE; 1049 goto send; 1050 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1051 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1052 goto send; 1053 } else 1054 csum = udp_csum(skb); 1055 1056 /* add protocol-dependent pseudo-header */ 1057 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1058 len, fl6->flowi6_proto, csum); 1059 if (uh->check == 0) 1060 uh->check = CSUM_MANGLED_0; 1061 1062 send: 1063 err = ip6_send_skb(skb); 1064 if (err) { 1065 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1066 UDP6_INC_STATS(sock_net(sk), 1067 UDP_MIB_SNDBUFERRORS, is_udplite); 1068 err = 0; 1069 } 1070 } else { 1071 UDP6_INC_STATS(sock_net(sk), 1072 UDP_MIB_OUTDATAGRAMS, is_udplite); 1073 } 1074 return err; 1075 } 1076 1077 static int udp_v6_push_pending_frames(struct sock *sk) 1078 { 1079 struct sk_buff *skb; 1080 struct udp_sock *up = udp_sk(sk); 1081 struct flowi6 fl6; 1082 int err = 0; 1083 1084 if (up->pending == AF_INET) 1085 return udp_push_pending_frames(sk); 1086 1087 /* ip6_finish_skb will release the cork, so make a copy of 1088 * fl6 here. 1089 */ 1090 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1091 1092 skb = ip6_finish_skb(sk); 1093 if (!skb) 1094 goto out; 1095 1096 err = udp_v6_send_skb(skb, &fl6); 1097 1098 out: 1099 up->len = 0; 1100 up->pending = 0; 1101 return err; 1102 } 1103 1104 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1105 { 1106 struct ipv6_txoptions opt_space; 1107 struct udp_sock *up = udp_sk(sk); 1108 struct inet_sock *inet = inet_sk(sk); 1109 struct ipv6_pinfo *np = inet6_sk(sk); 1110 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1111 struct in6_addr *daddr, *final_p, final; 1112 struct ipv6_txoptions *opt = NULL; 1113 struct ipv6_txoptions *opt_to_free = NULL; 1114 struct ip6_flowlabel *flowlabel = NULL; 1115 struct flowi6 fl6; 1116 struct dst_entry *dst; 1117 struct ipcm6_cookie ipc6; 1118 int addr_len = msg->msg_namelen; 1119 bool connected = false; 1120 int ulen = len; 1121 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1122 int err; 1123 int is_udplite = IS_UDPLITE(sk); 1124 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1125 struct sockcm_cookie sockc; 1126 1127 ipc6.hlimit = -1; 1128 ipc6.tclass = -1; 1129 ipc6.dontfrag = -1; 1130 sockc.tsflags = sk->sk_tsflags; 1131 1132 /* destination address check */ 1133 if (sin6) { 1134 if (addr_len < offsetof(struct sockaddr, sa_data)) 1135 return -EINVAL; 1136 1137 switch (sin6->sin6_family) { 1138 case AF_INET6: 1139 if (addr_len < SIN6_LEN_RFC2133) 1140 return -EINVAL; 1141 daddr = &sin6->sin6_addr; 1142 if (ipv6_addr_any(daddr) && 1143 ipv6_addr_v4mapped(&np->saddr)) 1144 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1145 daddr); 1146 break; 1147 case AF_INET: 1148 goto do_udp_sendmsg; 1149 case AF_UNSPEC: 1150 msg->msg_name = sin6 = NULL; 1151 msg->msg_namelen = addr_len = 0; 1152 daddr = NULL; 1153 break; 1154 default: 1155 return -EINVAL; 1156 } 1157 } else if (!up->pending) { 1158 if (sk->sk_state != TCP_ESTABLISHED) 1159 return -EDESTADDRREQ; 1160 daddr = &sk->sk_v6_daddr; 1161 } else 1162 daddr = NULL; 1163 1164 if (daddr) { 1165 if (ipv6_addr_v4mapped(daddr)) { 1166 struct sockaddr_in sin; 1167 sin.sin_family = AF_INET; 1168 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1169 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1170 msg->msg_name = &sin; 1171 msg->msg_namelen = sizeof(sin); 1172 do_udp_sendmsg: 1173 if (__ipv6_only_sock(sk)) 1174 return -ENETUNREACH; 1175 return udp_sendmsg(sk, msg, len); 1176 } 1177 } 1178 1179 if (up->pending == AF_INET) 1180 return udp_sendmsg(sk, msg, len); 1181 1182 /* Rough check on arithmetic overflow, 1183 better check is made in ip6_append_data(). 1184 */ 1185 if (len > INT_MAX - sizeof(struct udphdr)) 1186 return -EMSGSIZE; 1187 1188 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1189 if (up->pending) { 1190 /* 1191 * There are pending frames. 1192 * The socket lock must be held while it's corked. 1193 */ 1194 lock_sock(sk); 1195 if (likely(up->pending)) { 1196 if (unlikely(up->pending != AF_INET6)) { 1197 release_sock(sk); 1198 return -EAFNOSUPPORT; 1199 } 1200 dst = NULL; 1201 goto do_append_data; 1202 } 1203 release_sock(sk); 1204 } 1205 ulen += sizeof(struct udphdr); 1206 1207 memset(&fl6, 0, sizeof(fl6)); 1208 1209 if (sin6) { 1210 if (sin6->sin6_port == 0) 1211 return -EINVAL; 1212 1213 fl6.fl6_dport = sin6->sin6_port; 1214 daddr = &sin6->sin6_addr; 1215 1216 if (np->sndflow) { 1217 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1218 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1219 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1220 if (!flowlabel) 1221 return -EINVAL; 1222 } 1223 } 1224 1225 /* 1226 * Otherwise it will be difficult to maintain 1227 * sk->sk_dst_cache. 1228 */ 1229 if (sk->sk_state == TCP_ESTABLISHED && 1230 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1231 daddr = &sk->sk_v6_daddr; 1232 1233 if (addr_len >= sizeof(struct sockaddr_in6) && 1234 sin6->sin6_scope_id && 1235 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1236 fl6.flowi6_oif = sin6->sin6_scope_id; 1237 } else { 1238 if (sk->sk_state != TCP_ESTABLISHED) 1239 return -EDESTADDRREQ; 1240 1241 fl6.fl6_dport = inet->inet_dport; 1242 daddr = &sk->sk_v6_daddr; 1243 fl6.flowlabel = np->flow_label; 1244 connected = true; 1245 } 1246 1247 if (!fl6.flowi6_oif) 1248 fl6.flowi6_oif = sk->sk_bound_dev_if; 1249 1250 if (!fl6.flowi6_oif) 1251 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1252 1253 fl6.flowi6_mark = sk->sk_mark; 1254 fl6.flowi6_uid = sk->sk_uid; 1255 1256 if (msg->msg_controllen) { 1257 opt = &opt_space; 1258 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1259 opt->tot_len = sizeof(*opt); 1260 ipc6.opt = opt; 1261 1262 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc); 1263 if (err < 0) { 1264 fl6_sock_release(flowlabel); 1265 return err; 1266 } 1267 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1268 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1269 if (!flowlabel) 1270 return -EINVAL; 1271 } 1272 if (!(opt->opt_nflen|opt->opt_flen)) 1273 opt = NULL; 1274 connected = false; 1275 } 1276 if (!opt) { 1277 opt = txopt_get(np); 1278 opt_to_free = opt; 1279 } 1280 if (flowlabel) 1281 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1282 opt = ipv6_fixup_options(&opt_space, opt); 1283 ipc6.opt = opt; 1284 1285 fl6.flowi6_proto = sk->sk_protocol; 1286 if (!ipv6_addr_any(daddr)) 1287 fl6.daddr = *daddr; 1288 else 1289 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1290 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1291 fl6.saddr = np->saddr; 1292 fl6.fl6_sport = inet->inet_sport; 1293 1294 final_p = fl6_update_dst(&fl6, opt, &final); 1295 if (final_p) 1296 connected = false; 1297 1298 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1299 fl6.flowi6_oif = np->mcast_oif; 1300 connected = false; 1301 } else if (!fl6.flowi6_oif) 1302 fl6.flowi6_oif = np->ucast_oif; 1303 1304 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1305 1306 if (ipc6.tclass < 0) 1307 ipc6.tclass = np->tclass; 1308 1309 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1310 1311 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1312 if (IS_ERR(dst)) { 1313 err = PTR_ERR(dst); 1314 dst = NULL; 1315 goto out; 1316 } 1317 1318 if (ipc6.hlimit < 0) 1319 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1320 1321 if (msg->msg_flags&MSG_CONFIRM) 1322 goto do_confirm; 1323 back_from_confirm: 1324 1325 /* Lockless fast path for the non-corking case */ 1326 if (!corkreq) { 1327 struct sk_buff *skb; 1328 1329 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1330 sizeof(struct udphdr), &ipc6, 1331 &fl6, (struct rt6_info *)dst, 1332 msg->msg_flags, &sockc); 1333 err = PTR_ERR(skb); 1334 if (!IS_ERR_OR_NULL(skb)) 1335 err = udp_v6_send_skb(skb, &fl6); 1336 goto out; 1337 } 1338 1339 lock_sock(sk); 1340 if (unlikely(up->pending)) { 1341 /* The socket is already corked while preparing it. */ 1342 /* ... which is an evident application bug. --ANK */ 1343 release_sock(sk); 1344 1345 net_dbg_ratelimited("udp cork app bug 2\n"); 1346 err = -EINVAL; 1347 goto out; 1348 } 1349 1350 up->pending = AF_INET6; 1351 1352 do_append_data: 1353 if (ipc6.dontfrag < 0) 1354 ipc6.dontfrag = np->dontfrag; 1355 up->len += ulen; 1356 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1357 &ipc6, &fl6, (struct rt6_info *)dst, 1358 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc); 1359 if (err) 1360 udp_v6_flush_pending_frames(sk); 1361 else if (!corkreq) 1362 err = udp_v6_push_pending_frames(sk); 1363 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1364 up->pending = 0; 1365 1366 if (err > 0) 1367 err = np->recverr ? net_xmit_errno(err) : 0; 1368 release_sock(sk); 1369 1370 out: 1371 dst_release(dst); 1372 fl6_sock_release(flowlabel); 1373 txopt_put(opt_to_free); 1374 if (!err) 1375 return len; 1376 /* 1377 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1378 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1379 * we don't have a good statistic (IpOutDiscards but it can be too many 1380 * things). We could add another new stat but at least for now that 1381 * seems like overkill. 1382 */ 1383 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1384 UDP6_INC_STATS(sock_net(sk), 1385 UDP_MIB_SNDBUFERRORS, is_udplite); 1386 } 1387 return err; 1388 1389 do_confirm: 1390 if (msg->msg_flags & MSG_PROBE) 1391 dst_confirm_neigh(dst, &fl6.daddr); 1392 if (!(msg->msg_flags&MSG_PROBE) || len) 1393 goto back_from_confirm; 1394 err = 0; 1395 goto out; 1396 } 1397 1398 void udpv6_destroy_sock(struct sock *sk) 1399 { 1400 struct udp_sock *up = udp_sk(sk); 1401 lock_sock(sk); 1402 udp_v6_flush_pending_frames(sk); 1403 release_sock(sk); 1404 1405 if (static_key_false(&udpv6_encap_needed) && up->encap_type) { 1406 void (*encap_destroy)(struct sock *sk); 1407 encap_destroy = READ_ONCE(up->encap_destroy); 1408 if (encap_destroy) 1409 encap_destroy(sk); 1410 } 1411 1412 inet6_destroy_sock(sk); 1413 } 1414 1415 /* 1416 * Socket option code for UDP 1417 */ 1418 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1419 char __user *optval, unsigned int optlen) 1420 { 1421 if (level == SOL_UDP || level == SOL_UDPLITE) 1422 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1423 udp_v6_push_pending_frames); 1424 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1425 } 1426 1427 #ifdef CONFIG_COMPAT 1428 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1429 char __user *optval, unsigned int optlen) 1430 { 1431 if (level == SOL_UDP || level == SOL_UDPLITE) 1432 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1433 udp_v6_push_pending_frames); 1434 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1435 } 1436 #endif 1437 1438 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1439 char __user *optval, int __user *optlen) 1440 { 1441 if (level == SOL_UDP || level == SOL_UDPLITE) 1442 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1443 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1444 } 1445 1446 #ifdef CONFIG_COMPAT 1447 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1448 char __user *optval, int __user *optlen) 1449 { 1450 if (level == SOL_UDP || level == SOL_UDPLITE) 1451 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1452 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1453 } 1454 #endif 1455 1456 /* thinking of making this const? Don't. 1457 * early_demux can change based on sysctl. 1458 */ 1459 static struct inet6_protocol udpv6_protocol = { 1460 .early_demux = udp_v6_early_demux, 1461 .early_demux_handler = udp_v6_early_demux, 1462 .handler = udpv6_rcv, 1463 .err_handler = udpv6_err, 1464 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1465 }; 1466 1467 /* ------------------------------------------------------------------------ */ 1468 #ifdef CONFIG_PROC_FS 1469 int udp6_seq_show(struct seq_file *seq, void *v) 1470 { 1471 if (v == SEQ_START_TOKEN) { 1472 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1473 } else { 1474 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1475 struct inet_sock *inet = inet_sk(v); 1476 __u16 srcp = ntohs(inet->inet_sport); 1477 __u16 destp = ntohs(inet->inet_dport); 1478 ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket); 1479 } 1480 return 0; 1481 } 1482 1483 static const struct file_operations udp6_afinfo_seq_fops = { 1484 .open = udp_seq_open, 1485 .read = seq_read, 1486 .llseek = seq_lseek, 1487 .release = seq_release_net 1488 }; 1489 1490 static struct udp_seq_afinfo udp6_seq_afinfo = { 1491 .name = "udp6", 1492 .family = AF_INET6, 1493 .udp_table = &udp_table, 1494 .seq_fops = &udp6_afinfo_seq_fops, 1495 .seq_ops = { 1496 .show = udp6_seq_show, 1497 }, 1498 }; 1499 1500 int __net_init udp6_proc_init(struct net *net) 1501 { 1502 return udp_proc_register(net, &udp6_seq_afinfo); 1503 } 1504 1505 void udp6_proc_exit(struct net *net) 1506 { 1507 udp_proc_unregister(net, &udp6_seq_afinfo); 1508 } 1509 #endif /* CONFIG_PROC_FS */ 1510 1511 /* ------------------------------------------------------------------------ */ 1512 1513 struct proto udpv6_prot = { 1514 .name = "UDPv6", 1515 .owner = THIS_MODULE, 1516 .close = udp_lib_close, 1517 .pre_connect = udpv6_pre_connect, 1518 .connect = ip6_datagram_connect, 1519 .disconnect = udp_disconnect, 1520 .ioctl = udp_ioctl, 1521 .init = udp_init_sock, 1522 .destroy = udpv6_destroy_sock, 1523 .setsockopt = udpv6_setsockopt, 1524 .getsockopt = udpv6_getsockopt, 1525 .sendmsg = udpv6_sendmsg, 1526 .recvmsg = udpv6_recvmsg, 1527 .release_cb = ip6_datagram_release_cb, 1528 .hash = udp_lib_hash, 1529 .unhash = udp_lib_unhash, 1530 .rehash = udp_v6_rehash, 1531 .get_port = udp_v6_get_port, 1532 .memory_allocated = &udp_memory_allocated, 1533 .sysctl_mem = sysctl_udp_mem, 1534 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1535 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1536 .obj_size = sizeof(struct udp6_sock), 1537 .h.udp_table = &udp_table, 1538 #ifdef CONFIG_COMPAT 1539 .compat_setsockopt = compat_udpv6_setsockopt, 1540 .compat_getsockopt = compat_udpv6_getsockopt, 1541 #endif 1542 .diag_destroy = udp_abort, 1543 }; 1544 1545 static struct inet_protosw udpv6_protosw = { 1546 .type = SOCK_DGRAM, 1547 .protocol = IPPROTO_UDP, 1548 .prot = &udpv6_prot, 1549 .ops = &inet6_dgram_ops, 1550 .flags = INET_PROTOSW_PERMANENT, 1551 }; 1552 1553 int __init udpv6_init(void) 1554 { 1555 int ret; 1556 1557 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1558 if (ret) 1559 goto out; 1560 1561 ret = inet6_register_protosw(&udpv6_protosw); 1562 if (ret) 1563 goto out_udpv6_protocol; 1564 out: 1565 return ret; 1566 1567 out_udpv6_protocol: 1568 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1569 goto out; 1570 } 1571 1572 void udpv6_exit(void) 1573 { 1574 inet6_unregister_protosw(&udpv6_protosw); 1575 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1576 } 1577