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 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_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 289 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 290 const struct in6_addr *daddr, __be16 dport, int dif) 291 { 292 struct sock *sk; 293 294 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 295 dif, 0, &udp_table, NULL); 296 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 297 sk = NULL; 298 return sk; 299 } 300 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 301 #endif 302 303 /* do not use the scratch area len for jumbogram: their length execeeds the 304 * scratch area space; note that the IP6CB flags is still in the first 305 * cacheline, so checking for jumbograms is cheap 306 */ 307 static int udp6_skb_len(struct sk_buff *skb) 308 { 309 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 310 } 311 312 /* 313 * This should be easy, if there is something there we 314 * return it, otherwise we block. 315 */ 316 317 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 318 int noblock, int flags, int *addr_len) 319 { 320 struct ipv6_pinfo *np = inet6_sk(sk); 321 struct inet_sock *inet = inet_sk(sk); 322 struct sk_buff *skb; 323 unsigned int ulen, copied; 324 int peeked, peeking, off; 325 int err; 326 int is_udplite = IS_UDPLITE(sk); 327 bool checksum_valid = false; 328 int is_udp4; 329 330 if (flags & MSG_ERRQUEUE) 331 return ipv6_recv_error(sk, msg, len, addr_len); 332 333 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 334 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 335 336 try_again: 337 peeking = flags & MSG_PEEK; 338 off = sk_peek_offset(sk, flags); 339 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); 340 if (!skb) 341 return err; 342 343 ulen = udp6_skb_len(skb); 344 copied = len; 345 if (copied > ulen - off) 346 copied = ulen - off; 347 else if (copied < ulen) 348 msg->msg_flags |= MSG_TRUNC; 349 350 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 351 352 /* 353 * If checksum is needed at all, try to do it while copying the 354 * data. If the data is truncated, or if we only want a partial 355 * coverage checksum (UDP-Lite), do it before the copy. 356 */ 357 358 if (copied < ulen || peeking || 359 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 360 checksum_valid = udp_skb_csum_unnecessary(skb) || 361 !__udp_lib_checksum_complete(skb); 362 if (!checksum_valid) 363 goto csum_copy_err; 364 } 365 366 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 367 if (udp_skb_is_linear(skb)) 368 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 369 else 370 err = skb_copy_datagram_msg(skb, off, msg, copied); 371 } else { 372 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 373 if (err == -EINVAL) 374 goto csum_copy_err; 375 } 376 if (unlikely(err)) { 377 if (!peeked) { 378 atomic_inc(&sk->sk_drops); 379 if (is_udp4) 380 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 381 is_udplite); 382 else 383 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 384 is_udplite); 385 } 386 kfree_skb(skb); 387 return err; 388 } 389 if (!peeked) { 390 if (is_udp4) 391 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, 392 is_udplite); 393 else 394 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, 395 is_udplite); 396 } 397 398 sock_recv_ts_and_drops(msg, sk, skb); 399 400 /* Copy the address. */ 401 if (msg->msg_name) { 402 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 403 sin6->sin6_family = AF_INET6; 404 sin6->sin6_port = udp_hdr(skb)->source; 405 sin6->sin6_flowinfo = 0; 406 407 if (is_udp4) { 408 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 409 &sin6->sin6_addr); 410 sin6->sin6_scope_id = 0; 411 } else { 412 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 413 sin6->sin6_scope_id = 414 ipv6_iface_scope_id(&sin6->sin6_addr, 415 inet6_iif(skb)); 416 } 417 *addr_len = sizeof(*sin6); 418 } 419 420 if (np->rxopt.all) 421 ip6_datagram_recv_common_ctl(sk, msg, skb); 422 423 if (is_udp4) { 424 if (inet->cmsg_flags) 425 ip_cmsg_recv_offset(msg, sk, skb, 426 sizeof(struct udphdr), off); 427 } else { 428 if (np->rxopt.all) 429 ip6_datagram_recv_specific_ctl(sk, msg, skb); 430 } 431 432 err = copied; 433 if (flags & MSG_TRUNC) 434 err = ulen; 435 436 skb_consume_udp(sk, skb, peeking ? -err : err); 437 return err; 438 439 csum_copy_err: 440 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 441 udp_skb_destructor)) { 442 if (is_udp4) { 443 UDP_INC_STATS(sock_net(sk), 444 UDP_MIB_CSUMERRORS, is_udplite); 445 UDP_INC_STATS(sock_net(sk), 446 UDP_MIB_INERRORS, is_udplite); 447 } else { 448 UDP6_INC_STATS(sock_net(sk), 449 UDP_MIB_CSUMERRORS, is_udplite); 450 UDP6_INC_STATS(sock_net(sk), 451 UDP_MIB_INERRORS, is_udplite); 452 } 453 } 454 kfree_skb(skb); 455 456 /* starting over for a new packet, but check if we need to yield */ 457 cond_resched(); 458 msg->msg_flags &= ~MSG_TRUNC; 459 goto try_again; 460 } 461 462 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 463 u8 type, u8 code, int offset, __be32 info, 464 struct udp_table *udptable) 465 { 466 struct ipv6_pinfo *np; 467 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 468 const struct in6_addr *saddr = &hdr->saddr; 469 const struct in6_addr *daddr = &hdr->daddr; 470 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 471 struct sock *sk; 472 int harderr; 473 int err; 474 struct net *net = dev_net(skb->dev); 475 476 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 477 inet6_iif(skb), 0, udptable, skb); 478 if (!sk) { 479 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 480 ICMP6_MIB_INERRORS); 481 return; 482 } 483 484 harderr = icmpv6_err_convert(type, code, &err); 485 np = inet6_sk(sk); 486 487 if (type == ICMPV6_PKT_TOOBIG) { 488 if (!ip6_sk_accept_pmtu(sk)) 489 goto out; 490 ip6_sk_update_pmtu(skb, sk, info); 491 if (np->pmtudisc != IPV6_PMTUDISC_DONT) 492 harderr = 1; 493 } 494 if (type == NDISC_REDIRECT) { 495 ip6_sk_redirect(skb, sk); 496 goto out; 497 } 498 499 if (!np->recverr) { 500 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 501 goto out; 502 } else { 503 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 504 } 505 506 sk->sk_err = err; 507 sk->sk_error_report(sk); 508 out: 509 return; 510 } 511 512 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 513 { 514 int rc; 515 516 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 517 sock_rps_save_rxhash(sk, skb); 518 sk_mark_napi_id(sk, skb); 519 sk_incoming_cpu_update(sk); 520 } else { 521 sk_mark_napi_id_once(sk, skb); 522 } 523 524 rc = __udp_enqueue_schedule_skb(sk, skb); 525 if (rc < 0) { 526 int is_udplite = IS_UDPLITE(sk); 527 528 /* Note that an ENOMEM error is charged twice */ 529 if (rc == -ENOMEM) 530 UDP6_INC_STATS(sock_net(sk), 531 UDP_MIB_RCVBUFERRORS, is_udplite); 532 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 533 kfree_skb(skb); 534 return -1; 535 } 536 537 return 0; 538 } 539 540 static __inline__ void udpv6_err(struct sk_buff *skb, 541 struct inet6_skb_parm *opt, u8 type, 542 u8 code, int offset, __be32 info) 543 { 544 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); 545 } 546 547 static DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 548 void udpv6_encap_enable(void) 549 { 550 static_branch_enable(&udpv6_encap_needed_key); 551 } 552 EXPORT_SYMBOL(udpv6_encap_enable); 553 554 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 555 { 556 struct udp_sock *up = udp_sk(sk); 557 int is_udplite = IS_UDPLITE(sk); 558 559 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) 560 goto drop; 561 562 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 563 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 564 565 /* 566 * This is an encapsulation socket so pass the skb to 567 * the socket's udp_encap_rcv() hook. Otherwise, just 568 * fall through and pass this up the UDP socket. 569 * up->encap_rcv() returns the following value: 570 * =0 if skb was successfully passed to the encap 571 * handler or was discarded by it. 572 * >0 if skb should be passed on to UDP. 573 * <0 if skb should be resubmitted as proto -N 574 */ 575 576 /* if we're overly short, let UDP handle it */ 577 encap_rcv = READ_ONCE(up->encap_rcv); 578 if (encap_rcv) { 579 int ret; 580 581 /* Verify checksum before giving to encap */ 582 if (udp_lib_checksum_complete(skb)) 583 goto csum_error; 584 585 ret = encap_rcv(sk, skb); 586 if (ret <= 0) { 587 __UDP_INC_STATS(sock_net(sk), 588 UDP_MIB_INDATAGRAMS, 589 is_udplite); 590 return -ret; 591 } 592 } 593 594 /* FALLTHROUGH -- it's a UDP Packet */ 595 } 596 597 /* 598 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 599 */ 600 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 601 602 if (up->pcrlen == 0) { /* full coverage was set */ 603 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 604 UDP_SKB_CB(skb)->cscov, skb->len); 605 goto drop; 606 } 607 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 608 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 609 UDP_SKB_CB(skb)->cscov, up->pcrlen); 610 goto drop; 611 } 612 } 613 614 prefetch(&sk->sk_rmem_alloc); 615 if (rcu_access_pointer(sk->sk_filter) && 616 udp_lib_checksum_complete(skb)) 617 goto csum_error; 618 619 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 620 goto drop; 621 622 udp_csum_pull_header(skb); 623 624 skb_dst_drop(skb); 625 626 return __udpv6_queue_rcv_skb(sk, skb); 627 628 csum_error: 629 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 630 drop: 631 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 632 atomic_inc(&sk->sk_drops); 633 kfree_skb(skb); 634 return -1; 635 } 636 637 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, 638 __be16 loc_port, const struct in6_addr *loc_addr, 639 __be16 rmt_port, const struct in6_addr *rmt_addr, 640 int dif, unsigned short hnum) 641 { 642 struct inet_sock *inet = inet_sk(sk); 643 644 if (!net_eq(sock_net(sk), net)) 645 return false; 646 647 if (udp_sk(sk)->udp_port_hash != hnum || 648 sk->sk_family != PF_INET6 || 649 (inet->inet_dport && inet->inet_dport != rmt_port) || 650 (!ipv6_addr_any(&sk->sk_v6_daddr) && 651 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 652 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) || 653 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 654 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 655 return false; 656 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 657 return false; 658 return true; 659 } 660 661 static void udp6_csum_zero_error(struct sk_buff *skb) 662 { 663 /* RFC 2460 section 8.1 says that we SHOULD log 664 * this error. Well, it is reasonable. 665 */ 666 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 667 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 668 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 669 } 670 671 /* 672 * Note: called only from the BH handler context, 673 * so we don't need to lock the hashes. 674 */ 675 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 676 const struct in6_addr *saddr, const struct in6_addr *daddr, 677 struct udp_table *udptable, int proto) 678 { 679 struct sock *sk, *first = NULL; 680 const struct udphdr *uh = udp_hdr(skb); 681 unsigned short hnum = ntohs(uh->dest); 682 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 683 unsigned int offset = offsetof(typeof(*sk), sk_node); 684 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 685 int dif = inet6_iif(skb); 686 struct hlist_node *node; 687 struct sk_buff *nskb; 688 689 if (use_hash2) { 690 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 691 udptable->mask; 692 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 693 start_lookup: 694 hslot = &udptable->hash2[hash2]; 695 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 696 } 697 698 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 699 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 700 uh->source, saddr, dif, hnum)) 701 continue; 702 /* If zero checksum and no_check is not on for 703 * the socket then skip it. 704 */ 705 if (!uh->check && !udp_sk(sk)->no_check6_rx) 706 continue; 707 if (!first) { 708 first = sk; 709 continue; 710 } 711 nskb = skb_clone(skb, GFP_ATOMIC); 712 if (unlikely(!nskb)) { 713 atomic_inc(&sk->sk_drops); 714 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 715 IS_UDPLITE(sk)); 716 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 717 IS_UDPLITE(sk)); 718 continue; 719 } 720 721 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 722 consume_skb(nskb); 723 } 724 725 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 726 if (use_hash2 && hash2 != hash2_any) { 727 hash2 = hash2_any; 728 goto start_lookup; 729 } 730 731 if (first) { 732 if (udpv6_queue_rcv_skb(first, skb) > 0) 733 consume_skb(skb); 734 } else { 735 kfree_skb(skb); 736 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 737 proto == IPPROTO_UDPLITE); 738 } 739 return 0; 740 } 741 742 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 743 { 744 if (udp_sk_rx_dst_set(sk, dst)) { 745 const struct rt6_info *rt = (const struct rt6_info *)dst; 746 747 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); 748 } 749 } 750 751 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 752 int proto) 753 { 754 const struct in6_addr *saddr, *daddr; 755 struct net *net = dev_net(skb->dev); 756 struct udphdr *uh; 757 struct sock *sk; 758 u32 ulen = 0; 759 760 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 761 goto discard; 762 763 saddr = &ipv6_hdr(skb)->saddr; 764 daddr = &ipv6_hdr(skb)->daddr; 765 uh = udp_hdr(skb); 766 767 ulen = ntohs(uh->len); 768 if (ulen > skb->len) 769 goto short_packet; 770 771 if (proto == IPPROTO_UDP) { 772 /* UDP validates ulen. */ 773 774 /* Check for jumbo payload */ 775 if (ulen == 0) 776 ulen = skb->len; 777 778 if (ulen < sizeof(*uh)) 779 goto short_packet; 780 781 if (ulen < skb->len) { 782 if (pskb_trim_rcsum(skb, ulen)) 783 goto short_packet; 784 saddr = &ipv6_hdr(skb)->saddr; 785 daddr = &ipv6_hdr(skb)->daddr; 786 uh = udp_hdr(skb); 787 } 788 } 789 790 if (udp6_csum_init(skb, uh, proto)) 791 goto csum_error; 792 793 /* Check if the socket is already available, e.g. due to early demux */ 794 sk = skb_steal_sock(skb); 795 if (sk) { 796 struct dst_entry *dst = skb_dst(skb); 797 int ret; 798 799 if (unlikely(sk->sk_rx_dst != dst)) 800 udp6_sk_rx_dst_set(sk, dst); 801 802 ret = udpv6_queue_rcv_skb(sk, skb); 803 sock_put(sk); 804 805 /* a return value > 0 means to resubmit the input */ 806 if (ret > 0) 807 return ret; 808 return 0; 809 } 810 811 /* 812 * Multicast receive code 813 */ 814 if (ipv6_addr_is_multicast(daddr)) 815 return __udp6_lib_mcast_deliver(net, skb, 816 saddr, daddr, udptable, proto); 817 818 /* Unicast */ 819 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 820 if (sk) { 821 int ret; 822 823 if (!uh->check && !udp_sk(sk)->no_check6_rx) { 824 udp6_csum_zero_error(skb); 825 goto csum_error; 826 } 827 828 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 829 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 830 ip6_compute_pseudo); 831 832 ret = udpv6_queue_rcv_skb(sk, skb); 833 834 /* a return value > 0 means to resubmit the input */ 835 if (ret > 0) 836 return ret; 837 838 return 0; 839 } 840 841 if (!uh->check) { 842 udp6_csum_zero_error(skb); 843 goto csum_error; 844 } 845 846 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 847 goto discard; 848 849 if (udp_lib_checksum_complete(skb)) 850 goto csum_error; 851 852 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 853 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 854 855 kfree_skb(skb); 856 return 0; 857 858 short_packet: 859 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 860 proto == IPPROTO_UDPLITE ? "-Lite" : "", 861 saddr, ntohs(uh->source), 862 ulen, skb->len, 863 daddr, ntohs(uh->dest)); 864 goto discard; 865 csum_error: 866 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 867 discard: 868 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 869 kfree_skb(skb); 870 return 0; 871 } 872 873 874 static struct sock *__udp6_lib_demux_lookup(struct net *net, 875 __be16 loc_port, const struct in6_addr *loc_addr, 876 __be16 rmt_port, const struct in6_addr *rmt_addr, 877 int dif, int sdif) 878 { 879 unsigned short hnum = ntohs(loc_port); 880 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 881 unsigned int slot2 = hash2 & udp_table.mask; 882 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 883 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 884 struct sock *sk; 885 886 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 887 if (sk->sk_state == TCP_ESTABLISHED && 888 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif)) 889 return sk; 890 /* Only check first socket in chain */ 891 break; 892 } 893 return NULL; 894 } 895 896 static void udp_v6_early_demux(struct sk_buff *skb) 897 { 898 struct net *net = dev_net(skb->dev); 899 const struct udphdr *uh; 900 struct sock *sk; 901 struct dst_entry *dst; 902 int dif = skb->dev->ifindex; 903 int sdif = inet6_sdif(skb); 904 905 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 906 sizeof(struct udphdr))) 907 return; 908 909 uh = udp_hdr(skb); 910 911 if (skb->pkt_type == PACKET_HOST) 912 sk = __udp6_lib_demux_lookup(net, uh->dest, 913 &ipv6_hdr(skb)->daddr, 914 uh->source, &ipv6_hdr(skb)->saddr, 915 dif, sdif); 916 else 917 return; 918 919 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 920 return; 921 922 skb->sk = sk; 923 skb->destructor = sock_efree; 924 dst = READ_ONCE(sk->sk_rx_dst); 925 926 if (dst) 927 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); 928 if (dst) { 929 /* set noref for now. 930 * any place which wants to hold dst has to call 931 * dst_hold_safe() 932 */ 933 skb_dst_set_noref(skb, dst); 934 } 935 } 936 937 static __inline__ int udpv6_rcv(struct sk_buff *skb) 938 { 939 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); 940 } 941 942 /* 943 * Throw away all pending data and cancel the corking. Socket is locked. 944 */ 945 static void udp_v6_flush_pending_frames(struct sock *sk) 946 { 947 struct udp_sock *up = udp_sk(sk); 948 949 if (up->pending == AF_INET) 950 udp_flush_pending_frames(sk); 951 else if (up->pending) { 952 up->len = 0; 953 up->pending = 0; 954 ip6_flush_pending_frames(sk); 955 } 956 } 957 958 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 959 int addr_len) 960 { 961 /* The following checks are replicated from __ip6_datagram_connect() 962 * and intended to prevent BPF program called below from accessing 963 * bytes that are out of the bound specified by user in addr_len. 964 */ 965 if (uaddr->sa_family == AF_INET) { 966 if (__ipv6_only_sock(sk)) 967 return -EAFNOSUPPORT; 968 return udp_pre_connect(sk, uaddr, addr_len); 969 } 970 971 if (addr_len < SIN6_LEN_RFC2133) 972 return -EINVAL; 973 974 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr); 975 } 976 977 /** 978 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 979 * @sk: socket we are sending on 980 * @skb: sk_buff containing the filled-in UDP header 981 * (checksum field must be zeroed out) 982 */ 983 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 984 const struct in6_addr *saddr, 985 const struct in6_addr *daddr, int len) 986 { 987 unsigned int offset; 988 struct udphdr *uh = udp_hdr(skb); 989 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 990 __wsum csum = 0; 991 992 if (!frags) { 993 /* Only one fragment on the socket. */ 994 skb->csum_start = skb_transport_header(skb) - skb->head; 995 skb->csum_offset = offsetof(struct udphdr, check); 996 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 997 } else { 998 /* 999 * HW-checksum won't work as there are two or more 1000 * fragments on the socket so that all csums of sk_buffs 1001 * should be together 1002 */ 1003 offset = skb_transport_offset(skb); 1004 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1005 csum = skb->csum; 1006 1007 skb->ip_summed = CHECKSUM_NONE; 1008 1009 do { 1010 csum = csum_add(csum, frags->csum); 1011 } while ((frags = frags->next)); 1012 1013 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1014 csum); 1015 if (uh->check == 0) 1016 uh->check = CSUM_MANGLED_0; 1017 } 1018 } 1019 1020 /* 1021 * Sending 1022 */ 1023 1024 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1025 struct inet_cork *cork) 1026 { 1027 struct sock *sk = skb->sk; 1028 struct udphdr *uh; 1029 int err = 0; 1030 int is_udplite = IS_UDPLITE(sk); 1031 __wsum csum = 0; 1032 int offset = skb_transport_offset(skb); 1033 int len = skb->len - offset; 1034 1035 /* 1036 * Create a UDP header 1037 */ 1038 uh = udp_hdr(skb); 1039 uh->source = fl6->fl6_sport; 1040 uh->dest = fl6->fl6_dport; 1041 uh->len = htons(len); 1042 uh->check = 0; 1043 1044 if (cork->gso_size) { 1045 const int hlen = skb_network_header_len(skb) + 1046 sizeof(struct udphdr); 1047 1048 if (hlen + cork->gso_size > cork->fragsize) 1049 return -EINVAL; 1050 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) 1051 return -EINVAL; 1052 if (udp_sk(sk)->no_check6_tx) 1053 return -EINVAL; 1054 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1055 dst_xfrm(skb_dst(skb))) 1056 return -EIO; 1057 1058 skb_shinfo(skb)->gso_size = cork->gso_size; 1059 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1060 goto csum_partial; 1061 } 1062 1063 if (is_udplite) 1064 csum = udplite_csum(skb); 1065 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ 1066 skb->ip_summed = CHECKSUM_NONE; 1067 goto send; 1068 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1069 csum_partial: 1070 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1071 goto send; 1072 } else 1073 csum = udp_csum(skb); 1074 1075 /* add protocol-dependent pseudo-header */ 1076 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1077 len, fl6->flowi6_proto, csum); 1078 if (uh->check == 0) 1079 uh->check = CSUM_MANGLED_0; 1080 1081 send: 1082 err = ip6_send_skb(skb); 1083 if (err) { 1084 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { 1085 UDP6_INC_STATS(sock_net(sk), 1086 UDP_MIB_SNDBUFERRORS, is_udplite); 1087 err = 0; 1088 } 1089 } else { 1090 UDP6_INC_STATS(sock_net(sk), 1091 UDP_MIB_OUTDATAGRAMS, is_udplite); 1092 } 1093 return err; 1094 } 1095 1096 static int udp_v6_push_pending_frames(struct sock *sk) 1097 { 1098 struct sk_buff *skb; 1099 struct udp_sock *up = udp_sk(sk); 1100 struct flowi6 fl6; 1101 int err = 0; 1102 1103 if (up->pending == AF_INET) 1104 return udp_push_pending_frames(sk); 1105 1106 /* ip6_finish_skb will release the cork, so make a copy of 1107 * fl6 here. 1108 */ 1109 fl6 = inet_sk(sk)->cork.fl.u.ip6; 1110 1111 skb = ip6_finish_skb(sk); 1112 if (!skb) 1113 goto out; 1114 1115 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base); 1116 1117 out: 1118 up->len = 0; 1119 up->pending = 0; 1120 return err; 1121 } 1122 1123 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1124 { 1125 struct ipv6_txoptions opt_space; 1126 struct udp_sock *up = udp_sk(sk); 1127 struct inet_sock *inet = inet_sk(sk); 1128 struct ipv6_pinfo *np = inet6_sk(sk); 1129 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1130 struct in6_addr *daddr, *final_p, final; 1131 struct ipv6_txoptions *opt = NULL; 1132 struct ipv6_txoptions *opt_to_free = NULL; 1133 struct ip6_flowlabel *flowlabel = NULL; 1134 struct flowi6 fl6; 1135 struct dst_entry *dst; 1136 struct ipcm6_cookie ipc6; 1137 int addr_len = msg->msg_namelen; 1138 bool connected = false; 1139 int ulen = len; 1140 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1141 int err; 1142 int is_udplite = IS_UDPLITE(sk); 1143 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1144 struct sockcm_cookie sockc; 1145 1146 ipc6.hlimit = -1; 1147 ipc6.tclass = -1; 1148 ipc6.dontfrag = -1; 1149 ipc6.gso_size = up->gso_size; 1150 sockc.tsflags = sk->sk_tsflags; 1151 1152 /* destination address check */ 1153 if (sin6) { 1154 if (addr_len < offsetof(struct sockaddr, sa_data)) 1155 return -EINVAL; 1156 1157 switch (sin6->sin6_family) { 1158 case AF_INET6: 1159 if (addr_len < SIN6_LEN_RFC2133) 1160 return -EINVAL; 1161 daddr = &sin6->sin6_addr; 1162 if (ipv6_addr_any(daddr) && 1163 ipv6_addr_v4mapped(&np->saddr)) 1164 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1165 daddr); 1166 break; 1167 case AF_INET: 1168 goto do_udp_sendmsg; 1169 case AF_UNSPEC: 1170 msg->msg_name = sin6 = NULL; 1171 msg->msg_namelen = addr_len = 0; 1172 daddr = NULL; 1173 break; 1174 default: 1175 return -EINVAL; 1176 } 1177 } else if (!up->pending) { 1178 if (sk->sk_state != TCP_ESTABLISHED) 1179 return -EDESTADDRREQ; 1180 daddr = &sk->sk_v6_daddr; 1181 } else 1182 daddr = NULL; 1183 1184 if (daddr) { 1185 if (ipv6_addr_v4mapped(daddr)) { 1186 struct sockaddr_in sin; 1187 sin.sin_family = AF_INET; 1188 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1189 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1190 msg->msg_name = &sin; 1191 msg->msg_namelen = sizeof(sin); 1192 do_udp_sendmsg: 1193 if (__ipv6_only_sock(sk)) 1194 return -ENETUNREACH; 1195 return udp_sendmsg(sk, msg, len); 1196 } 1197 } 1198 1199 if (up->pending == AF_INET) 1200 return udp_sendmsg(sk, msg, len); 1201 1202 /* Rough check on arithmetic overflow, 1203 better check is made in ip6_append_data(). 1204 */ 1205 if (len > INT_MAX - sizeof(struct udphdr)) 1206 return -EMSGSIZE; 1207 1208 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1209 if (up->pending) { 1210 /* 1211 * There are pending frames. 1212 * The socket lock must be held while it's corked. 1213 */ 1214 lock_sock(sk); 1215 if (likely(up->pending)) { 1216 if (unlikely(up->pending != AF_INET6)) { 1217 release_sock(sk); 1218 return -EAFNOSUPPORT; 1219 } 1220 dst = NULL; 1221 goto do_append_data; 1222 } 1223 release_sock(sk); 1224 } 1225 ulen += sizeof(struct udphdr); 1226 1227 memset(&fl6, 0, sizeof(fl6)); 1228 1229 if (sin6) { 1230 if (sin6->sin6_port == 0) 1231 return -EINVAL; 1232 1233 fl6.fl6_dport = sin6->sin6_port; 1234 daddr = &sin6->sin6_addr; 1235 1236 if (np->sndflow) { 1237 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1238 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { 1239 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1240 if (!flowlabel) 1241 return -EINVAL; 1242 } 1243 } 1244 1245 /* 1246 * Otherwise it will be difficult to maintain 1247 * sk->sk_dst_cache. 1248 */ 1249 if (sk->sk_state == TCP_ESTABLISHED && 1250 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1251 daddr = &sk->sk_v6_daddr; 1252 1253 if (addr_len >= sizeof(struct sockaddr_in6) && 1254 sin6->sin6_scope_id && 1255 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1256 fl6.flowi6_oif = sin6->sin6_scope_id; 1257 } else { 1258 if (sk->sk_state != TCP_ESTABLISHED) 1259 return -EDESTADDRREQ; 1260 1261 fl6.fl6_dport = inet->inet_dport; 1262 daddr = &sk->sk_v6_daddr; 1263 fl6.flowlabel = np->flow_label; 1264 connected = true; 1265 } 1266 1267 if (!fl6.flowi6_oif) 1268 fl6.flowi6_oif = sk->sk_bound_dev_if; 1269 1270 if (!fl6.flowi6_oif) 1271 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1272 1273 fl6.flowi6_mark = sk->sk_mark; 1274 fl6.flowi6_uid = sk->sk_uid; 1275 1276 if (msg->msg_controllen) { 1277 opt = &opt_space; 1278 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1279 opt->tot_len = sizeof(*opt); 1280 ipc6.opt = opt; 1281 1282 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1283 if (err > 0) 1284 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, 1285 &ipc6, &sockc); 1286 if (err < 0) { 1287 fl6_sock_release(flowlabel); 1288 return err; 1289 } 1290 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1291 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); 1292 if (!flowlabel) 1293 return -EINVAL; 1294 } 1295 if (!(opt->opt_nflen|opt->opt_flen)) 1296 opt = NULL; 1297 connected = false; 1298 } 1299 if (!opt) { 1300 opt = txopt_get(np); 1301 opt_to_free = opt; 1302 } 1303 if (flowlabel) 1304 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1305 opt = ipv6_fixup_options(&opt_space, opt); 1306 ipc6.opt = opt; 1307 1308 fl6.flowi6_proto = sk->sk_protocol; 1309 if (!ipv6_addr_any(daddr)) 1310 fl6.daddr = *daddr; 1311 else 1312 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1313 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) 1314 fl6.saddr = np->saddr; 1315 fl6.fl6_sport = inet->inet_sport; 1316 1317 if (cgroup_bpf_enabled && !connected) { 1318 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1319 (struct sockaddr *)sin6, &fl6.saddr); 1320 if (err) 1321 goto out_no_dst; 1322 if (sin6) { 1323 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1324 /* BPF program rewrote IPv6-only by IPv4-mapped 1325 * IPv6. It's currently unsupported. 1326 */ 1327 err = -ENOTSUPP; 1328 goto out_no_dst; 1329 } 1330 if (sin6->sin6_port == 0) { 1331 /* BPF program set invalid port. Reject it. */ 1332 err = -EINVAL; 1333 goto out_no_dst; 1334 } 1335 fl6.fl6_dport = sin6->sin6_port; 1336 fl6.daddr = sin6->sin6_addr; 1337 } 1338 } 1339 1340 final_p = fl6_update_dst(&fl6, opt, &final); 1341 if (final_p) 1342 connected = false; 1343 1344 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { 1345 fl6.flowi6_oif = np->mcast_oif; 1346 connected = false; 1347 } else if (!fl6.flowi6_oif) 1348 fl6.flowi6_oif = np->ucast_oif; 1349 1350 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); 1351 1352 if (ipc6.tclass < 0) 1353 ipc6.tclass = np->tclass; 1354 1355 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); 1356 1357 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected); 1358 if (IS_ERR(dst)) { 1359 err = PTR_ERR(dst); 1360 dst = NULL; 1361 goto out; 1362 } 1363 1364 if (ipc6.hlimit < 0) 1365 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); 1366 1367 if (msg->msg_flags&MSG_CONFIRM) 1368 goto do_confirm; 1369 back_from_confirm: 1370 1371 /* Lockless fast path for the non-corking case */ 1372 if (!corkreq) { 1373 struct inet_cork_full cork; 1374 struct sk_buff *skb; 1375 1376 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1377 sizeof(struct udphdr), &ipc6, 1378 &fl6, (struct rt6_info *)dst, 1379 msg->msg_flags, &cork, &sockc); 1380 err = PTR_ERR(skb); 1381 if (!IS_ERR_OR_NULL(skb)) 1382 err = udp_v6_send_skb(skb, &fl6, &cork.base); 1383 goto out; 1384 } 1385 1386 lock_sock(sk); 1387 if (unlikely(up->pending)) { 1388 /* The socket is already corked while preparing it. */ 1389 /* ... which is an evident application bug. --ANK */ 1390 release_sock(sk); 1391 1392 net_dbg_ratelimited("udp cork app bug 2\n"); 1393 err = -EINVAL; 1394 goto out; 1395 } 1396 1397 up->pending = AF_INET6; 1398 1399 do_append_data: 1400 if (ipc6.dontfrag < 0) 1401 ipc6.dontfrag = np->dontfrag; 1402 up->len += ulen; 1403 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1404 &ipc6, &fl6, (struct rt6_info *)dst, 1405 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc); 1406 if (err) 1407 udp_v6_flush_pending_frames(sk); 1408 else if (!corkreq) 1409 err = udp_v6_push_pending_frames(sk); 1410 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1411 up->pending = 0; 1412 1413 if (err > 0) 1414 err = np->recverr ? net_xmit_errno(err) : 0; 1415 release_sock(sk); 1416 1417 out: 1418 dst_release(dst); 1419 out_no_dst: 1420 fl6_sock_release(flowlabel); 1421 txopt_put(opt_to_free); 1422 if (!err) 1423 return len; 1424 /* 1425 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1426 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1427 * we don't have a good statistic (IpOutDiscards but it can be too many 1428 * things). We could add another new stat but at least for now that 1429 * seems like overkill. 1430 */ 1431 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1432 UDP6_INC_STATS(sock_net(sk), 1433 UDP_MIB_SNDBUFERRORS, is_udplite); 1434 } 1435 return err; 1436 1437 do_confirm: 1438 if (msg->msg_flags & MSG_PROBE) 1439 dst_confirm_neigh(dst, &fl6.daddr); 1440 if (!(msg->msg_flags&MSG_PROBE) || len) 1441 goto back_from_confirm; 1442 err = 0; 1443 goto out; 1444 } 1445 1446 void udpv6_destroy_sock(struct sock *sk) 1447 { 1448 struct udp_sock *up = udp_sk(sk); 1449 lock_sock(sk); 1450 udp_v6_flush_pending_frames(sk); 1451 release_sock(sk); 1452 1453 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) { 1454 void (*encap_destroy)(struct sock *sk); 1455 encap_destroy = READ_ONCE(up->encap_destroy); 1456 if (encap_destroy) 1457 encap_destroy(sk); 1458 } 1459 1460 inet6_destroy_sock(sk); 1461 } 1462 1463 /* 1464 * Socket option code for UDP 1465 */ 1466 int udpv6_setsockopt(struct sock *sk, int level, int optname, 1467 char __user *optval, unsigned int optlen) 1468 { 1469 if (level == SOL_UDP || level == SOL_UDPLITE) 1470 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1471 udp_v6_push_pending_frames); 1472 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1473 } 1474 1475 #ifdef CONFIG_COMPAT 1476 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, 1477 char __user *optval, unsigned int optlen) 1478 { 1479 if (level == SOL_UDP || level == SOL_UDPLITE) 1480 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1481 udp_v6_push_pending_frames); 1482 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); 1483 } 1484 #endif 1485 1486 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1487 char __user *optval, int __user *optlen) 1488 { 1489 if (level == SOL_UDP || level == SOL_UDPLITE) 1490 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1491 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1492 } 1493 1494 #ifdef CONFIG_COMPAT 1495 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, 1496 char __user *optval, int __user *optlen) 1497 { 1498 if (level == SOL_UDP || level == SOL_UDPLITE) 1499 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1500 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); 1501 } 1502 #endif 1503 1504 /* thinking of making this const? Don't. 1505 * early_demux can change based on sysctl. 1506 */ 1507 static struct inet6_protocol udpv6_protocol = { 1508 .early_demux = udp_v6_early_demux, 1509 .early_demux_handler = udp_v6_early_demux, 1510 .handler = udpv6_rcv, 1511 .err_handler = udpv6_err, 1512 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, 1513 }; 1514 1515 /* ------------------------------------------------------------------------ */ 1516 #ifdef CONFIG_PROC_FS 1517 int udp6_seq_show(struct seq_file *seq, void *v) 1518 { 1519 if (v == SEQ_START_TOKEN) { 1520 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1521 } else { 1522 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1523 struct inet_sock *inet = inet_sk(v); 1524 __u16 srcp = ntohs(inet->inet_sport); 1525 __u16 destp = ntohs(inet->inet_dport); 1526 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1527 udp_rqueue_get(v), bucket); 1528 } 1529 return 0; 1530 } 1531 1532 const struct seq_operations udp6_seq_ops = { 1533 .start = udp_seq_start, 1534 .next = udp_seq_next, 1535 .stop = udp_seq_stop, 1536 .show = udp6_seq_show, 1537 }; 1538 EXPORT_SYMBOL(udp6_seq_ops); 1539 1540 static struct udp_seq_afinfo udp6_seq_afinfo = { 1541 .family = AF_INET6, 1542 .udp_table = &udp_table, 1543 }; 1544 1545 int __net_init udp6_proc_init(struct net *net) 1546 { 1547 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1548 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1549 return -ENOMEM; 1550 return 0; 1551 } 1552 1553 void udp6_proc_exit(struct net *net) 1554 { 1555 remove_proc_entry("udp6", net->proc_net); 1556 } 1557 #endif /* CONFIG_PROC_FS */ 1558 1559 /* ------------------------------------------------------------------------ */ 1560 1561 struct proto udpv6_prot = { 1562 .name = "UDPv6", 1563 .owner = THIS_MODULE, 1564 .close = udp_lib_close, 1565 .pre_connect = udpv6_pre_connect, 1566 .connect = ip6_datagram_connect, 1567 .disconnect = udp_disconnect, 1568 .ioctl = udp_ioctl, 1569 .init = udp_init_sock, 1570 .destroy = udpv6_destroy_sock, 1571 .setsockopt = udpv6_setsockopt, 1572 .getsockopt = udpv6_getsockopt, 1573 .sendmsg = udpv6_sendmsg, 1574 .recvmsg = udpv6_recvmsg, 1575 .release_cb = ip6_datagram_release_cb, 1576 .hash = udp_lib_hash, 1577 .unhash = udp_lib_unhash, 1578 .rehash = udp_v6_rehash, 1579 .get_port = udp_v6_get_port, 1580 .memory_allocated = &udp_memory_allocated, 1581 .sysctl_mem = sysctl_udp_mem, 1582 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1583 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1584 .obj_size = sizeof(struct udp6_sock), 1585 .h.udp_table = &udp_table, 1586 #ifdef CONFIG_COMPAT 1587 .compat_setsockopt = compat_udpv6_setsockopt, 1588 .compat_getsockopt = compat_udpv6_getsockopt, 1589 #endif 1590 .diag_destroy = udp_abort, 1591 }; 1592 1593 static struct inet_protosw udpv6_protosw = { 1594 .type = SOCK_DGRAM, 1595 .protocol = IPPROTO_UDP, 1596 .prot = &udpv6_prot, 1597 .ops = &inet6_dgram_ops, 1598 .flags = INET_PROTOSW_PERMANENT, 1599 }; 1600 1601 int __init udpv6_init(void) 1602 { 1603 int ret; 1604 1605 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); 1606 if (ret) 1607 goto out; 1608 1609 ret = inet6_register_protosw(&udpv6_protosw); 1610 if (ret) 1611 goto out_udpv6_protocol; 1612 out: 1613 return ret; 1614 1615 out_udpv6_protocol: 1616 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1617 goto out; 1618 } 1619 1620 void udpv6_exit(void) 1621 { 1622 inet6_unregister_protosw(&udpv6_protosw); 1623 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); 1624 } 1625