1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The User Datagram Protocol (UDP). 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 12 * Hirokazu Takahashi, <taka@valinux.co.jp> 13 * 14 * Fixes: 15 * Alan Cox : verify_area() calls 16 * Alan Cox : stopped close while in use off icmp 17 * messages. Not a fix but a botch that 18 * for udp at least is 'valid'. 19 * Alan Cox : Fixed icmp handling properly 20 * Alan Cox : Correct error for oversized datagrams 21 * Alan Cox : Tidied select() semantics. 22 * Alan Cox : udp_err() fixed properly, also now 23 * select and read wake correctly on errors 24 * Alan Cox : udp_send verify_area moved to avoid mem leak 25 * Alan Cox : UDP can count its memory 26 * Alan Cox : send to an unknown connection causes 27 * an ECONNREFUSED off the icmp, but 28 * does NOT close. 29 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 31 * bug no longer crashes it. 32 * Fred Van Kempen : Net2e support for sk->broadcast. 33 * Alan Cox : Uses skb_free_datagram 34 * Alan Cox : Added get/set sockopt support. 35 * Alan Cox : Broadcasting without option set returns EACCES. 36 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 37 * Alan Cox : Use ip_tos and ip_ttl 38 * Alan Cox : SNMP Mibs 39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 40 * Matt Dillon : UDP length checks. 41 * Alan Cox : Smarter af_inet used properly. 42 * Alan Cox : Use new kernel side addressing. 43 * Alan Cox : Incorrect return on truncated datagram receive. 44 * Arnt Gulbrandsen : New udp_send and stuff 45 * Alan Cox : Cache last socket 46 * Alan Cox : Route cache 47 * Jon Peatfield : Minor efficiency fix to sendto(). 48 * Mike Shaver : RFC1122 checks. 49 * Alan Cox : Nonblocking error fix. 50 * Willy Konynenberg : Transparent proxying support. 51 * Mike McLagan : Routing by source 52 * David S. Miller : New socket lookup architecture. 53 * Last socket cache retained as it 54 * does have a high hit rate. 55 * Olaf Kirch : Don't linearise iovec on sendmsg. 56 * Andi Kleen : Some cleanups, cache destination entry 57 * for connect. 58 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 59 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 60 * return ENOTCONN for unconnected sockets (POSIX) 61 * Janos Farkas : don't deliver multi/broadcasts to a different 62 * bound-to-device socket 63 * Hirokazu Takahashi : HW checksumming for outgoing UDP 64 * datagrams. 65 * Hirokazu Takahashi : sendfile() on UDP works now. 66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 69 * a single port at the same time. 70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71 * James Chapman : Add L2TP encapsulation type. 72 * 73 * 74 * This program is free software; you can redistribute it and/or 75 * modify it under the terms of the GNU General Public License 76 * as published by the Free Software Foundation; either version 77 * 2 of the License, or (at your option) any later version. 78 */ 79 80 #define pr_fmt(fmt) "UDP: " fmt 81 82 #include <asm/uaccess.h> 83 #include <asm/ioctls.h> 84 #include <linux/bootmem.h> 85 #include <linux/highmem.h> 86 #include <linux/swap.h> 87 #include <linux/types.h> 88 #include <linux/fcntl.h> 89 #include <linux/module.h> 90 #include <linux/socket.h> 91 #include <linux/sockios.h> 92 #include <linux/igmp.h> 93 #include <linux/inetdevice.h> 94 #include <linux/in.h> 95 #include <linux/errno.h> 96 #include <linux/timer.h> 97 #include <linux/mm.h> 98 #include <linux/inet.h> 99 #include <linux/netdevice.h> 100 #include <linux/slab.h> 101 #include <net/tcp_states.h> 102 #include <linux/skbuff.h> 103 #include <linux/proc_fs.h> 104 #include <linux/seq_file.h> 105 #include <net/net_namespace.h> 106 #include <net/icmp.h> 107 #include <net/inet_hashtables.h> 108 #include <net/route.h> 109 #include <net/checksum.h> 110 #include <net/xfrm.h> 111 #include <trace/events/udp.h> 112 #include <linux/static_key.h> 113 #include <trace/events/skb.h> 114 #include <net/busy_poll.h> 115 #include "udp_impl.h" 116 #include <net/sock_reuseport.h> 117 #include <net/addrconf.h> 118 119 struct udp_table udp_table __read_mostly; 120 EXPORT_SYMBOL(udp_table); 121 122 long sysctl_udp_mem[3] __read_mostly; 123 EXPORT_SYMBOL(sysctl_udp_mem); 124 125 int sysctl_udp_rmem_min __read_mostly; 126 EXPORT_SYMBOL(sysctl_udp_rmem_min); 127 128 int sysctl_udp_wmem_min __read_mostly; 129 EXPORT_SYMBOL(sysctl_udp_wmem_min); 130 131 atomic_long_t udp_memory_allocated; 132 EXPORT_SYMBOL(udp_memory_allocated); 133 134 #define MAX_UDP_PORTS 65536 135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 136 137 static int udp_lib_lport_inuse(struct net *net, __u16 num, 138 const struct udp_hslot *hslot, 139 unsigned long *bitmap, 140 struct sock *sk, 141 int (*saddr_comp)(const struct sock *sk1, 142 const struct sock *sk2, 143 bool match_wildcard), 144 unsigned int log) 145 { 146 struct sock *sk2; 147 kuid_t uid = sock_i_uid(sk); 148 149 sk_for_each(sk2, &hslot->head) { 150 if (net_eq(sock_net(sk2), net) && 151 sk2 != sk && 152 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 153 (!sk2->sk_reuse || !sk->sk_reuse) && 154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 156 (!sk2->sk_reuseport || !sk->sk_reuseport || 157 rcu_access_pointer(sk->sk_reuseport_cb) || 158 !uid_eq(uid, sock_i_uid(sk2))) && 159 saddr_comp(sk, sk2, true)) { 160 if (!bitmap) 161 return 1; 162 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap); 163 } 164 } 165 return 0; 166 } 167 168 /* 169 * Note: we still hold spinlock of primary hash chain, so no other writer 170 * can insert/delete a socket with local_port == num 171 */ 172 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 173 struct udp_hslot *hslot2, 174 struct sock *sk, 175 int (*saddr_comp)(const struct sock *sk1, 176 const struct sock *sk2, 177 bool match_wildcard)) 178 { 179 struct sock *sk2; 180 kuid_t uid = sock_i_uid(sk); 181 int res = 0; 182 183 spin_lock(&hslot2->lock); 184 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 185 if (net_eq(sock_net(sk2), net) && 186 sk2 != sk && 187 (udp_sk(sk2)->udp_port_hash == num) && 188 (!sk2->sk_reuse || !sk->sk_reuse) && 189 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 190 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 191 (!sk2->sk_reuseport || !sk->sk_reuseport || 192 rcu_access_pointer(sk->sk_reuseport_cb) || 193 !uid_eq(uid, sock_i_uid(sk2))) && 194 saddr_comp(sk, sk2, true)) { 195 res = 1; 196 break; 197 } 198 } 199 spin_unlock(&hslot2->lock); 200 return res; 201 } 202 203 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot, 204 int (*saddr_same)(const struct sock *sk1, 205 const struct sock *sk2, 206 bool match_wildcard)) 207 { 208 struct net *net = sock_net(sk); 209 kuid_t uid = sock_i_uid(sk); 210 struct sock *sk2; 211 212 sk_for_each(sk2, &hslot->head) { 213 if (net_eq(sock_net(sk2), net) && 214 sk2 != sk && 215 sk2->sk_family == sk->sk_family && 216 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 217 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 218 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 219 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 220 (*saddr_same)(sk, sk2, false)) { 221 return reuseport_add_sock(sk, sk2); 222 } 223 } 224 225 /* Initial allocation may have already happened via setsockopt */ 226 if (!rcu_access_pointer(sk->sk_reuseport_cb)) 227 return reuseport_alloc(sk); 228 return 0; 229 } 230 231 /** 232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 233 * 234 * @sk: socket struct in question 235 * @snum: port number to look up 236 * @saddr_comp: AF-dependent comparison of bound local IP addresses 237 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 238 * with NULL address 239 */ 240 int udp_lib_get_port(struct sock *sk, unsigned short snum, 241 int (*saddr_comp)(const struct sock *sk1, 242 const struct sock *sk2, 243 bool match_wildcard), 244 unsigned int hash2_nulladdr) 245 { 246 struct udp_hslot *hslot, *hslot2; 247 struct udp_table *udptable = sk->sk_prot->h.udp_table; 248 int error = 1; 249 struct net *net = sock_net(sk); 250 251 if (!snum) { 252 int low, high, remaining; 253 unsigned int rand; 254 unsigned short first, last; 255 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 256 257 inet_get_local_port_range(net, &low, &high); 258 remaining = (high - low) + 1; 259 260 rand = prandom_u32(); 261 first = reciprocal_scale(rand, remaining) + low; 262 /* 263 * force rand to be an odd multiple of UDP_HTABLE_SIZE 264 */ 265 rand = (rand | 1) * (udptable->mask + 1); 266 last = first + udptable->mask + 1; 267 do { 268 hslot = udp_hashslot(udptable, net, first); 269 bitmap_zero(bitmap, PORTS_PER_CHAIN); 270 spin_lock_bh(&hslot->lock); 271 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 272 saddr_comp, udptable->log); 273 274 snum = first; 275 /* 276 * Iterate on all possible values of snum for this hash. 277 * Using steps of an odd multiple of UDP_HTABLE_SIZE 278 * give us randomization and full range coverage. 279 */ 280 do { 281 if (low <= snum && snum <= high && 282 !test_bit(snum >> udptable->log, bitmap) && 283 !inet_is_local_reserved_port(net, snum)) 284 goto found; 285 snum += rand; 286 } while (snum != first); 287 spin_unlock_bh(&hslot->lock); 288 } while (++first != last); 289 goto fail; 290 } else { 291 hslot = udp_hashslot(udptable, net, snum); 292 spin_lock_bh(&hslot->lock); 293 if (hslot->count > 10) { 294 int exist; 295 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 296 297 slot2 &= udptable->mask; 298 hash2_nulladdr &= udptable->mask; 299 300 hslot2 = udp_hashslot2(udptable, slot2); 301 if (hslot->count < hslot2->count) 302 goto scan_primary_hash; 303 304 exist = udp_lib_lport_inuse2(net, snum, hslot2, 305 sk, saddr_comp); 306 if (!exist && (hash2_nulladdr != slot2)) { 307 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 308 exist = udp_lib_lport_inuse2(net, snum, hslot2, 309 sk, saddr_comp); 310 } 311 if (exist) 312 goto fail_unlock; 313 else 314 goto found; 315 } 316 scan_primary_hash: 317 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 318 saddr_comp, 0)) 319 goto fail_unlock; 320 } 321 found: 322 inet_sk(sk)->inet_num = snum; 323 udp_sk(sk)->udp_port_hash = snum; 324 udp_sk(sk)->udp_portaddr_hash ^= snum; 325 if (sk_unhashed(sk)) { 326 if (sk->sk_reuseport && 327 udp_reuseport_add_sock(sk, hslot, saddr_comp)) { 328 inet_sk(sk)->inet_num = 0; 329 udp_sk(sk)->udp_port_hash = 0; 330 udp_sk(sk)->udp_portaddr_hash ^= snum; 331 goto fail_unlock; 332 } 333 334 sk_add_node_rcu(sk, &hslot->head); 335 hslot->count++; 336 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 337 338 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 339 spin_lock(&hslot2->lock); 340 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 341 sk->sk_family == AF_INET6) 342 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 343 &hslot2->head); 344 else 345 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 346 &hslot2->head); 347 hslot2->count++; 348 spin_unlock(&hslot2->lock); 349 } 350 sock_set_flag(sk, SOCK_RCU_FREE); 351 error = 0; 352 fail_unlock: 353 spin_unlock_bh(&hslot->lock); 354 fail: 355 return error; 356 } 357 EXPORT_SYMBOL(udp_lib_get_port); 358 359 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses 360 * match_wildcard == false: addresses must be exactly the same, i.e. 361 * 0.0.0.0 only equals to 0.0.0.0 362 */ 363 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2, 364 bool match_wildcard) 365 { 366 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 367 368 if (!ipv6_only_sock(sk2)) { 369 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr) 370 return 1; 371 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr) 372 return match_wildcard; 373 } 374 return 0; 375 } 376 377 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, 378 unsigned int port) 379 { 380 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 381 } 382 383 int udp_v4_get_port(struct sock *sk, unsigned short snum) 384 { 385 unsigned int hash2_nulladdr = 386 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 387 unsigned int hash2_partial = 388 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 389 390 /* precompute partial secondary hash */ 391 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 392 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 393 } 394 395 static int compute_score(struct sock *sk, struct net *net, 396 __be32 saddr, __be16 sport, 397 __be32 daddr, unsigned short hnum, int dif) 398 { 399 int score; 400 struct inet_sock *inet; 401 402 if (!net_eq(sock_net(sk), net) || 403 udp_sk(sk)->udp_port_hash != hnum || 404 ipv6_only_sock(sk)) 405 return -1; 406 407 score = (sk->sk_family == PF_INET) ? 2 : 1; 408 inet = inet_sk(sk); 409 410 if (inet->inet_rcv_saddr) { 411 if (inet->inet_rcv_saddr != daddr) 412 return -1; 413 score += 4; 414 } 415 416 if (inet->inet_daddr) { 417 if (inet->inet_daddr != saddr) 418 return -1; 419 score += 4; 420 } 421 422 if (inet->inet_dport) { 423 if (inet->inet_dport != sport) 424 return -1; 425 score += 4; 426 } 427 428 if (sk->sk_bound_dev_if) { 429 if (sk->sk_bound_dev_if != dif) 430 return -1; 431 score += 4; 432 } 433 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 434 score++; 435 return score; 436 } 437 438 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 439 const __u16 lport, const __be32 faddr, 440 const __be16 fport) 441 { 442 static u32 udp_ehash_secret __read_mostly; 443 444 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 445 446 return __inet_ehashfn(laddr, lport, faddr, fport, 447 udp_ehash_secret + net_hash_mix(net)); 448 } 449 450 /* called with rcu_read_lock() */ 451 static struct sock *udp4_lib_lookup2(struct net *net, 452 __be32 saddr, __be16 sport, 453 __be32 daddr, unsigned int hnum, int dif, 454 struct udp_hslot *hslot2, 455 struct sk_buff *skb) 456 { 457 struct sock *sk, *result; 458 int score, badness, matches = 0, reuseport = 0; 459 u32 hash = 0; 460 461 result = NULL; 462 badness = 0; 463 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 464 score = compute_score(sk, net, saddr, sport, 465 daddr, hnum, dif); 466 if (score > badness) { 467 reuseport = sk->sk_reuseport; 468 if (reuseport) { 469 hash = udp_ehashfn(net, daddr, hnum, 470 saddr, sport); 471 result = reuseport_select_sock(sk, hash, skb, 472 sizeof(struct udphdr)); 473 if (result) 474 return result; 475 matches = 1; 476 } 477 badness = score; 478 result = sk; 479 } else if (score == badness && reuseport) { 480 matches++; 481 if (reciprocal_scale(hash, matches) == 0) 482 result = sk; 483 hash = next_pseudo_random32(hash); 484 } 485 } 486 return result; 487 } 488 489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 490 * harder than this. -DaveM 491 */ 492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 493 __be16 sport, __be32 daddr, __be16 dport, 494 int dif, struct udp_table *udptable, struct sk_buff *skb) 495 { 496 struct sock *sk, *result; 497 unsigned short hnum = ntohs(dport); 498 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 499 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 500 int score, badness, matches = 0, reuseport = 0; 501 u32 hash = 0; 502 503 if (hslot->count > 10) { 504 hash2 = udp4_portaddr_hash(net, daddr, hnum); 505 slot2 = hash2 & udptable->mask; 506 hslot2 = &udptable->hash2[slot2]; 507 if (hslot->count < hslot2->count) 508 goto begin; 509 510 result = udp4_lib_lookup2(net, saddr, sport, 511 daddr, hnum, dif, 512 hslot2, skb); 513 if (!result) { 514 unsigned int old_slot2 = slot2; 515 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 516 slot2 = hash2 & udptable->mask; 517 /* avoid searching the same slot again. */ 518 if (unlikely(slot2 == old_slot2)) 519 return result; 520 521 hslot2 = &udptable->hash2[slot2]; 522 if (hslot->count < hslot2->count) 523 goto begin; 524 525 result = udp4_lib_lookup2(net, saddr, sport, 526 daddr, hnum, dif, 527 hslot2, skb); 528 } 529 return result; 530 } 531 begin: 532 result = NULL; 533 badness = 0; 534 sk_for_each_rcu(sk, &hslot->head) { 535 score = compute_score(sk, net, saddr, sport, 536 daddr, hnum, dif); 537 if (score > badness) { 538 reuseport = sk->sk_reuseport; 539 if (reuseport) { 540 hash = udp_ehashfn(net, daddr, hnum, 541 saddr, sport); 542 result = reuseport_select_sock(sk, hash, skb, 543 sizeof(struct udphdr)); 544 if (result) 545 return result; 546 matches = 1; 547 } 548 result = sk; 549 badness = score; 550 } else if (score == badness && reuseport) { 551 matches++; 552 if (reciprocal_scale(hash, matches) == 0) 553 result = sk; 554 hash = next_pseudo_random32(hash); 555 } 556 } 557 return result; 558 } 559 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 560 561 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 562 __be16 sport, __be16 dport, 563 struct udp_table *udptable) 564 { 565 const struct iphdr *iph = ip_hdr(skb); 566 567 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 568 iph->daddr, dport, inet_iif(skb), 569 udptable, skb); 570 } 571 572 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, 573 __be16 sport, __be16 dport) 574 { 575 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); 576 } 577 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); 578 579 /* Must be called under rcu_read_lock(). 580 * Does increment socket refcount. 581 */ 582 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \ 583 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) 584 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 585 __be32 daddr, __be16 dport, int dif) 586 { 587 struct sock *sk; 588 589 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 590 dif, &udp_table, NULL); 591 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt)) 592 sk = NULL; 593 return sk; 594 } 595 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 596 #endif 597 598 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 599 __be16 loc_port, __be32 loc_addr, 600 __be16 rmt_port, __be32 rmt_addr, 601 int dif, unsigned short hnum) 602 { 603 struct inet_sock *inet = inet_sk(sk); 604 605 if (!net_eq(sock_net(sk), net) || 606 udp_sk(sk)->udp_port_hash != hnum || 607 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 608 (inet->inet_dport != rmt_port && inet->inet_dport) || 609 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 610 ipv6_only_sock(sk) || 611 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 612 return false; 613 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 614 return false; 615 return true; 616 } 617 618 /* 619 * This routine is called by the ICMP module when it gets some 620 * sort of error condition. If err < 0 then the socket should 621 * be closed and the error returned to the user. If err > 0 622 * it's just the icmp type << 8 | icmp code. 623 * Header points to the ip header of the error packet. We move 624 * on past this. Then (as it used to claim before adjustment) 625 * header points to the first 8 bytes of the udp header. We need 626 * to find the appropriate port. 627 */ 628 629 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 630 { 631 struct inet_sock *inet; 632 const struct iphdr *iph = (const struct iphdr *)skb->data; 633 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 634 const int type = icmp_hdr(skb)->type; 635 const int code = icmp_hdr(skb)->code; 636 struct sock *sk; 637 int harderr; 638 int err; 639 struct net *net = dev_net(skb->dev); 640 641 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 642 iph->saddr, uh->source, skb->dev->ifindex, udptable, 643 NULL); 644 if (!sk) { 645 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 646 return; /* No socket for error */ 647 } 648 649 err = 0; 650 harderr = 0; 651 inet = inet_sk(sk); 652 653 switch (type) { 654 default: 655 case ICMP_TIME_EXCEEDED: 656 err = EHOSTUNREACH; 657 break; 658 case ICMP_SOURCE_QUENCH: 659 goto out; 660 case ICMP_PARAMETERPROB: 661 err = EPROTO; 662 harderr = 1; 663 break; 664 case ICMP_DEST_UNREACH: 665 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 666 ipv4_sk_update_pmtu(skb, sk, info); 667 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 668 err = EMSGSIZE; 669 harderr = 1; 670 break; 671 } 672 goto out; 673 } 674 err = EHOSTUNREACH; 675 if (code <= NR_ICMP_UNREACH) { 676 harderr = icmp_err_convert[code].fatal; 677 err = icmp_err_convert[code].errno; 678 } 679 break; 680 case ICMP_REDIRECT: 681 ipv4_sk_redirect(skb, sk); 682 goto out; 683 } 684 685 /* 686 * RFC1122: OK. Passes ICMP errors back to application, as per 687 * 4.1.3.3. 688 */ 689 if (!inet->recverr) { 690 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 691 goto out; 692 } else 693 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 694 695 sk->sk_err = err; 696 sk->sk_error_report(sk); 697 out: 698 return; 699 } 700 701 void udp_err(struct sk_buff *skb, u32 info) 702 { 703 __udp4_lib_err(skb, info, &udp_table); 704 } 705 706 /* 707 * Throw away all pending data and cancel the corking. Socket is locked. 708 */ 709 void udp_flush_pending_frames(struct sock *sk) 710 { 711 struct udp_sock *up = udp_sk(sk); 712 713 if (up->pending) { 714 up->len = 0; 715 up->pending = 0; 716 ip_flush_pending_frames(sk); 717 } 718 } 719 EXPORT_SYMBOL(udp_flush_pending_frames); 720 721 /** 722 * udp4_hwcsum - handle outgoing HW checksumming 723 * @skb: sk_buff containing the filled-in UDP header 724 * (checksum field must be zeroed out) 725 * @src: source IP address 726 * @dst: destination IP address 727 */ 728 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 729 { 730 struct udphdr *uh = udp_hdr(skb); 731 int offset = skb_transport_offset(skb); 732 int len = skb->len - offset; 733 int hlen = len; 734 __wsum csum = 0; 735 736 if (!skb_has_frag_list(skb)) { 737 /* 738 * Only one fragment on the socket. 739 */ 740 skb->csum_start = skb_transport_header(skb) - skb->head; 741 skb->csum_offset = offsetof(struct udphdr, check); 742 uh->check = ~csum_tcpudp_magic(src, dst, len, 743 IPPROTO_UDP, 0); 744 } else { 745 struct sk_buff *frags; 746 747 /* 748 * HW-checksum won't work as there are two or more 749 * fragments on the socket so that all csums of sk_buffs 750 * should be together 751 */ 752 skb_walk_frags(skb, frags) { 753 csum = csum_add(csum, frags->csum); 754 hlen -= frags->len; 755 } 756 757 csum = skb_checksum(skb, offset, hlen, csum); 758 skb->ip_summed = CHECKSUM_NONE; 759 760 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 761 if (uh->check == 0) 762 uh->check = CSUM_MANGLED_0; 763 } 764 } 765 EXPORT_SYMBOL_GPL(udp4_hwcsum); 766 767 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 768 * for the simple case like when setting the checksum for a UDP tunnel. 769 */ 770 void udp_set_csum(bool nocheck, struct sk_buff *skb, 771 __be32 saddr, __be32 daddr, int len) 772 { 773 struct udphdr *uh = udp_hdr(skb); 774 775 if (nocheck) { 776 uh->check = 0; 777 } else if (skb_is_gso(skb)) { 778 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 779 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 780 uh->check = 0; 781 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 782 if (uh->check == 0) 783 uh->check = CSUM_MANGLED_0; 784 } else { 785 skb->ip_summed = CHECKSUM_PARTIAL; 786 skb->csum_start = skb_transport_header(skb) - skb->head; 787 skb->csum_offset = offsetof(struct udphdr, check); 788 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 789 } 790 } 791 EXPORT_SYMBOL(udp_set_csum); 792 793 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 794 { 795 struct sock *sk = skb->sk; 796 struct inet_sock *inet = inet_sk(sk); 797 struct udphdr *uh; 798 int err = 0; 799 int is_udplite = IS_UDPLITE(sk); 800 int offset = skb_transport_offset(skb); 801 int len = skb->len - offset; 802 __wsum csum = 0; 803 804 /* 805 * Create a UDP header 806 */ 807 uh = udp_hdr(skb); 808 uh->source = inet->inet_sport; 809 uh->dest = fl4->fl4_dport; 810 uh->len = htons(len); 811 uh->check = 0; 812 813 if (is_udplite) /* UDP-Lite */ 814 csum = udplite_csum(skb); 815 816 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 817 818 skb->ip_summed = CHECKSUM_NONE; 819 goto send; 820 821 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 822 823 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 824 goto send; 825 826 } else 827 csum = udp_csum(skb); 828 829 /* add protocol-dependent pseudo-header */ 830 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 831 sk->sk_protocol, csum); 832 if (uh->check == 0) 833 uh->check = CSUM_MANGLED_0; 834 835 send: 836 err = ip_send_skb(sock_net(sk), skb); 837 if (err) { 838 if (err == -ENOBUFS && !inet->recverr) { 839 UDP_INC_STATS(sock_net(sk), 840 UDP_MIB_SNDBUFERRORS, is_udplite); 841 err = 0; 842 } 843 } else 844 UDP_INC_STATS(sock_net(sk), 845 UDP_MIB_OUTDATAGRAMS, is_udplite); 846 return err; 847 } 848 849 /* 850 * Push out all pending data as one UDP datagram. Socket is locked. 851 */ 852 int udp_push_pending_frames(struct sock *sk) 853 { 854 struct udp_sock *up = udp_sk(sk); 855 struct inet_sock *inet = inet_sk(sk); 856 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 857 struct sk_buff *skb; 858 int err = 0; 859 860 skb = ip_finish_skb(sk, fl4); 861 if (!skb) 862 goto out; 863 864 err = udp_send_skb(skb, fl4); 865 866 out: 867 up->len = 0; 868 up->pending = 0; 869 return err; 870 } 871 EXPORT_SYMBOL(udp_push_pending_frames); 872 873 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 874 { 875 struct inet_sock *inet = inet_sk(sk); 876 struct udp_sock *up = udp_sk(sk); 877 struct flowi4 fl4_stack; 878 struct flowi4 *fl4; 879 int ulen = len; 880 struct ipcm_cookie ipc; 881 struct rtable *rt = NULL; 882 int free = 0; 883 int connected = 0; 884 __be32 daddr, faddr, saddr; 885 __be16 dport; 886 u8 tos; 887 int err, is_udplite = IS_UDPLITE(sk); 888 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 889 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 890 struct sk_buff *skb; 891 struct ip_options_data opt_copy; 892 893 if (len > 0xFFFF) 894 return -EMSGSIZE; 895 896 /* 897 * Check the flags. 898 */ 899 900 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 901 return -EOPNOTSUPP; 902 903 ipc.opt = NULL; 904 ipc.tx_flags = 0; 905 ipc.ttl = 0; 906 ipc.tos = -1; 907 908 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 909 910 fl4 = &inet->cork.fl.u.ip4; 911 if (up->pending) { 912 /* 913 * There are pending frames. 914 * The socket lock must be held while it's corked. 915 */ 916 lock_sock(sk); 917 if (likely(up->pending)) { 918 if (unlikely(up->pending != AF_INET)) { 919 release_sock(sk); 920 return -EINVAL; 921 } 922 goto do_append_data; 923 } 924 release_sock(sk); 925 } 926 ulen += sizeof(struct udphdr); 927 928 /* 929 * Get and verify the address. 930 */ 931 if (msg->msg_name) { 932 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 933 if (msg->msg_namelen < sizeof(*usin)) 934 return -EINVAL; 935 if (usin->sin_family != AF_INET) { 936 if (usin->sin_family != AF_UNSPEC) 937 return -EAFNOSUPPORT; 938 } 939 940 daddr = usin->sin_addr.s_addr; 941 dport = usin->sin_port; 942 if (dport == 0) 943 return -EINVAL; 944 } else { 945 if (sk->sk_state != TCP_ESTABLISHED) 946 return -EDESTADDRREQ; 947 daddr = inet->inet_daddr; 948 dport = inet->inet_dport; 949 /* Open fast path for connected socket. 950 Route will not be used, if at least one option is set. 951 */ 952 connected = 1; 953 } 954 955 ipc.sockc.tsflags = sk->sk_tsflags; 956 ipc.addr = inet->inet_saddr; 957 ipc.oif = sk->sk_bound_dev_if; 958 959 if (msg->msg_controllen) { 960 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6); 961 if (unlikely(err)) { 962 kfree(ipc.opt); 963 return err; 964 } 965 if (ipc.opt) 966 free = 1; 967 connected = 0; 968 } 969 if (!ipc.opt) { 970 struct ip_options_rcu *inet_opt; 971 972 rcu_read_lock(); 973 inet_opt = rcu_dereference(inet->inet_opt); 974 if (inet_opt) { 975 memcpy(&opt_copy, inet_opt, 976 sizeof(*inet_opt) + inet_opt->opt.optlen); 977 ipc.opt = &opt_copy.opt; 978 } 979 rcu_read_unlock(); 980 } 981 982 saddr = ipc.addr; 983 ipc.addr = faddr = daddr; 984 985 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags); 986 987 if (ipc.opt && ipc.opt->opt.srr) { 988 if (!daddr) 989 return -EINVAL; 990 faddr = ipc.opt->opt.faddr; 991 connected = 0; 992 } 993 tos = get_rttos(&ipc, inet); 994 if (sock_flag(sk, SOCK_LOCALROUTE) || 995 (msg->msg_flags & MSG_DONTROUTE) || 996 (ipc.opt && ipc.opt->opt.is_strictroute)) { 997 tos |= RTO_ONLINK; 998 connected = 0; 999 } 1000 1001 if (ipv4_is_multicast(daddr)) { 1002 if (!ipc.oif) 1003 ipc.oif = inet->mc_index; 1004 if (!saddr) 1005 saddr = inet->mc_addr; 1006 connected = 0; 1007 } else if (!ipc.oif) 1008 ipc.oif = inet->uc_index; 1009 1010 if (connected) 1011 rt = (struct rtable *)sk_dst_check(sk, 0); 1012 1013 if (!rt) { 1014 struct net *net = sock_net(sk); 1015 __u8 flow_flags = inet_sk_flowi_flags(sk); 1016 1017 fl4 = &fl4_stack; 1018 1019 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1020 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1021 flow_flags, 1022 faddr, saddr, dport, inet->inet_sport); 1023 1024 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1025 rt = ip_route_output_flow(net, fl4, sk); 1026 if (IS_ERR(rt)) { 1027 err = PTR_ERR(rt); 1028 rt = NULL; 1029 if (err == -ENETUNREACH) 1030 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1031 goto out; 1032 } 1033 1034 err = -EACCES; 1035 if ((rt->rt_flags & RTCF_BROADCAST) && 1036 !sock_flag(sk, SOCK_BROADCAST)) 1037 goto out; 1038 if (connected) 1039 sk_dst_set(sk, dst_clone(&rt->dst)); 1040 } 1041 1042 if (msg->msg_flags&MSG_CONFIRM) 1043 goto do_confirm; 1044 back_from_confirm: 1045 1046 saddr = fl4->saddr; 1047 if (!ipc.addr) 1048 daddr = ipc.addr = fl4->daddr; 1049 1050 /* Lockless fast path for the non-corking case. */ 1051 if (!corkreq) { 1052 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1053 sizeof(struct udphdr), &ipc, &rt, 1054 msg->msg_flags); 1055 err = PTR_ERR(skb); 1056 if (!IS_ERR_OR_NULL(skb)) 1057 err = udp_send_skb(skb, fl4); 1058 goto out; 1059 } 1060 1061 lock_sock(sk); 1062 if (unlikely(up->pending)) { 1063 /* The socket is already corked while preparing it. */ 1064 /* ... which is an evident application bug. --ANK */ 1065 release_sock(sk); 1066 1067 net_dbg_ratelimited("cork app bug 2\n"); 1068 err = -EINVAL; 1069 goto out; 1070 } 1071 /* 1072 * Now cork the socket to pend data. 1073 */ 1074 fl4 = &inet->cork.fl.u.ip4; 1075 fl4->daddr = daddr; 1076 fl4->saddr = saddr; 1077 fl4->fl4_dport = dport; 1078 fl4->fl4_sport = inet->inet_sport; 1079 up->pending = AF_INET; 1080 1081 do_append_data: 1082 up->len += ulen; 1083 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1084 sizeof(struct udphdr), &ipc, &rt, 1085 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1086 if (err) 1087 udp_flush_pending_frames(sk); 1088 else if (!corkreq) 1089 err = udp_push_pending_frames(sk); 1090 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1091 up->pending = 0; 1092 release_sock(sk); 1093 1094 out: 1095 ip_rt_put(rt); 1096 if (free) 1097 kfree(ipc.opt); 1098 if (!err) 1099 return len; 1100 /* 1101 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1102 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1103 * we don't have a good statistic (IpOutDiscards but it can be too many 1104 * things). We could add another new stat but at least for now that 1105 * seems like overkill. 1106 */ 1107 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1108 UDP_INC_STATS(sock_net(sk), 1109 UDP_MIB_SNDBUFERRORS, is_udplite); 1110 } 1111 return err; 1112 1113 do_confirm: 1114 dst_confirm(&rt->dst); 1115 if (!(msg->msg_flags&MSG_PROBE) || len) 1116 goto back_from_confirm; 1117 err = 0; 1118 goto out; 1119 } 1120 EXPORT_SYMBOL(udp_sendmsg); 1121 1122 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1123 size_t size, int flags) 1124 { 1125 struct inet_sock *inet = inet_sk(sk); 1126 struct udp_sock *up = udp_sk(sk); 1127 int ret; 1128 1129 if (flags & MSG_SENDPAGE_NOTLAST) 1130 flags |= MSG_MORE; 1131 1132 if (!up->pending) { 1133 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1134 1135 /* Call udp_sendmsg to specify destination address which 1136 * sendpage interface can't pass. 1137 * This will succeed only when the socket is connected. 1138 */ 1139 ret = udp_sendmsg(sk, &msg, 0); 1140 if (ret < 0) 1141 return ret; 1142 } 1143 1144 lock_sock(sk); 1145 1146 if (unlikely(!up->pending)) { 1147 release_sock(sk); 1148 1149 net_dbg_ratelimited("udp cork app bug 3\n"); 1150 return -EINVAL; 1151 } 1152 1153 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1154 page, offset, size, flags); 1155 if (ret == -EOPNOTSUPP) { 1156 release_sock(sk); 1157 return sock_no_sendpage(sk->sk_socket, page, offset, 1158 size, flags); 1159 } 1160 if (ret < 0) { 1161 udp_flush_pending_frames(sk); 1162 goto out; 1163 } 1164 1165 up->len += size; 1166 if (!(up->corkflag || (flags&MSG_MORE))) 1167 ret = udp_push_pending_frames(sk); 1168 if (!ret) 1169 ret = size; 1170 out: 1171 release_sock(sk); 1172 return ret; 1173 } 1174 1175 /** 1176 * first_packet_length - return length of first packet in receive queue 1177 * @sk: socket 1178 * 1179 * Drops all bad checksum frames, until a valid one is found. 1180 * Returns the length of found skb, or -1 if none is found. 1181 */ 1182 static int first_packet_length(struct sock *sk) 1183 { 1184 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1185 struct sk_buff *skb; 1186 int res; 1187 1188 __skb_queue_head_init(&list_kill); 1189 1190 spin_lock_bh(&rcvq->lock); 1191 while ((skb = skb_peek(rcvq)) != NULL && 1192 udp_lib_checksum_complete(skb)) { 1193 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1194 IS_UDPLITE(sk)); 1195 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1196 IS_UDPLITE(sk)); 1197 atomic_inc(&sk->sk_drops); 1198 __skb_unlink(skb, rcvq); 1199 __skb_queue_tail(&list_kill, skb); 1200 } 1201 res = skb ? skb->len : -1; 1202 spin_unlock_bh(&rcvq->lock); 1203 1204 if (!skb_queue_empty(&list_kill)) { 1205 bool slow = lock_sock_fast(sk); 1206 1207 __skb_queue_purge(&list_kill); 1208 sk_mem_reclaim_partial(sk); 1209 unlock_sock_fast(sk, slow); 1210 } 1211 return res; 1212 } 1213 1214 /* 1215 * IOCTL requests applicable to the UDP protocol 1216 */ 1217 1218 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1219 { 1220 switch (cmd) { 1221 case SIOCOUTQ: 1222 { 1223 int amount = sk_wmem_alloc_get(sk); 1224 1225 return put_user(amount, (int __user *)arg); 1226 } 1227 1228 case SIOCINQ: 1229 { 1230 int amount = max_t(int, 0, first_packet_length(sk)); 1231 1232 return put_user(amount, (int __user *)arg); 1233 } 1234 1235 default: 1236 return -ENOIOCTLCMD; 1237 } 1238 1239 return 0; 1240 } 1241 EXPORT_SYMBOL(udp_ioctl); 1242 1243 /* 1244 * This should be easy, if there is something there we 1245 * return it, otherwise we block. 1246 */ 1247 1248 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1249 int flags, int *addr_len) 1250 { 1251 struct inet_sock *inet = inet_sk(sk); 1252 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1253 struct sk_buff *skb; 1254 unsigned int ulen, copied; 1255 int peeked, peeking, off; 1256 int err; 1257 int is_udplite = IS_UDPLITE(sk); 1258 bool checksum_valid = false; 1259 bool slow; 1260 1261 if (flags & MSG_ERRQUEUE) 1262 return ip_recv_error(sk, msg, len, addr_len); 1263 1264 try_again: 1265 peeking = off = sk_peek_offset(sk, flags); 1266 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1267 &peeked, &off, &err); 1268 if (!skb) 1269 return err; 1270 1271 ulen = skb->len; 1272 copied = len; 1273 if (copied > ulen - off) 1274 copied = ulen - off; 1275 else if (copied < ulen) 1276 msg->msg_flags |= MSG_TRUNC; 1277 1278 /* 1279 * If checksum is needed at all, try to do it while copying the 1280 * data. If the data is truncated, or if we only want a partial 1281 * coverage checksum (UDP-Lite), do it before the copy. 1282 */ 1283 1284 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov || peeking) { 1285 checksum_valid = !udp_lib_checksum_complete(skb); 1286 if (!checksum_valid) 1287 goto csum_copy_err; 1288 } 1289 1290 if (checksum_valid || skb_csum_unnecessary(skb)) 1291 err = skb_copy_datagram_msg(skb, off, msg, copied); 1292 else { 1293 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1294 1295 if (err == -EINVAL) 1296 goto csum_copy_err; 1297 } 1298 1299 if (unlikely(err)) { 1300 trace_kfree_skb(skb, udp_recvmsg); 1301 if (!peeked) { 1302 atomic_inc(&sk->sk_drops); 1303 UDP_INC_STATS(sock_net(sk), 1304 UDP_MIB_INERRORS, is_udplite); 1305 } 1306 skb_free_datagram_locked(sk, skb); 1307 return err; 1308 } 1309 1310 if (!peeked) 1311 UDP_INC_STATS(sock_net(sk), 1312 UDP_MIB_INDATAGRAMS, is_udplite); 1313 1314 sock_recv_ts_and_drops(msg, sk, skb); 1315 1316 /* Copy the address. */ 1317 if (sin) { 1318 sin->sin_family = AF_INET; 1319 sin->sin_port = udp_hdr(skb)->source; 1320 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1321 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1322 *addr_len = sizeof(*sin); 1323 } 1324 if (inet->cmsg_flags) 1325 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr), off); 1326 1327 err = copied; 1328 if (flags & MSG_TRUNC) 1329 err = ulen; 1330 1331 __skb_free_datagram_locked(sk, skb, peeking ? -err : err); 1332 return err; 1333 1334 csum_copy_err: 1335 slow = lock_sock_fast(sk); 1336 if (!skb_kill_datagram(sk, skb, flags)) { 1337 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1338 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1339 } 1340 unlock_sock_fast(sk, slow); 1341 1342 /* starting over for a new packet, but check if we need to yield */ 1343 cond_resched(); 1344 msg->msg_flags &= ~MSG_TRUNC; 1345 goto try_again; 1346 } 1347 1348 int __udp_disconnect(struct sock *sk, int flags) 1349 { 1350 struct inet_sock *inet = inet_sk(sk); 1351 /* 1352 * 1003.1g - break association. 1353 */ 1354 1355 sk->sk_state = TCP_CLOSE; 1356 inet->inet_daddr = 0; 1357 inet->inet_dport = 0; 1358 sock_rps_reset_rxhash(sk); 1359 sk->sk_bound_dev_if = 0; 1360 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1361 inet_reset_saddr(sk); 1362 1363 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1364 sk->sk_prot->unhash(sk); 1365 inet->inet_sport = 0; 1366 } 1367 sk_dst_reset(sk); 1368 return 0; 1369 } 1370 EXPORT_SYMBOL(__udp_disconnect); 1371 1372 int udp_disconnect(struct sock *sk, int flags) 1373 { 1374 lock_sock(sk); 1375 __udp_disconnect(sk, flags); 1376 release_sock(sk); 1377 return 0; 1378 } 1379 EXPORT_SYMBOL(udp_disconnect); 1380 1381 void udp_lib_unhash(struct sock *sk) 1382 { 1383 if (sk_hashed(sk)) { 1384 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1385 struct udp_hslot *hslot, *hslot2; 1386 1387 hslot = udp_hashslot(udptable, sock_net(sk), 1388 udp_sk(sk)->udp_port_hash); 1389 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1390 1391 spin_lock_bh(&hslot->lock); 1392 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1393 reuseport_detach_sock(sk); 1394 if (sk_del_node_init_rcu(sk)) { 1395 hslot->count--; 1396 inet_sk(sk)->inet_num = 0; 1397 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1398 1399 spin_lock(&hslot2->lock); 1400 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1401 hslot2->count--; 1402 spin_unlock(&hslot2->lock); 1403 } 1404 spin_unlock_bh(&hslot->lock); 1405 } 1406 } 1407 EXPORT_SYMBOL(udp_lib_unhash); 1408 1409 /* 1410 * inet_rcv_saddr was changed, we must rehash secondary hash 1411 */ 1412 void udp_lib_rehash(struct sock *sk, u16 newhash) 1413 { 1414 if (sk_hashed(sk)) { 1415 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1416 struct udp_hslot *hslot, *hslot2, *nhslot2; 1417 1418 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1419 nhslot2 = udp_hashslot2(udptable, newhash); 1420 udp_sk(sk)->udp_portaddr_hash = newhash; 1421 1422 if (hslot2 != nhslot2 || 1423 rcu_access_pointer(sk->sk_reuseport_cb)) { 1424 hslot = udp_hashslot(udptable, sock_net(sk), 1425 udp_sk(sk)->udp_port_hash); 1426 /* we must lock primary chain too */ 1427 spin_lock_bh(&hslot->lock); 1428 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1429 reuseport_detach_sock(sk); 1430 1431 if (hslot2 != nhslot2) { 1432 spin_lock(&hslot2->lock); 1433 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1434 hslot2->count--; 1435 spin_unlock(&hslot2->lock); 1436 1437 spin_lock(&nhslot2->lock); 1438 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1439 &nhslot2->head); 1440 nhslot2->count++; 1441 spin_unlock(&nhslot2->lock); 1442 } 1443 1444 spin_unlock_bh(&hslot->lock); 1445 } 1446 } 1447 } 1448 EXPORT_SYMBOL(udp_lib_rehash); 1449 1450 static void udp_v4_rehash(struct sock *sk) 1451 { 1452 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1453 inet_sk(sk)->inet_rcv_saddr, 1454 inet_sk(sk)->inet_num); 1455 udp_lib_rehash(sk, new_hash); 1456 } 1457 1458 int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1459 { 1460 int rc; 1461 1462 if (inet_sk(sk)->inet_daddr) { 1463 sock_rps_save_rxhash(sk, skb); 1464 sk_mark_napi_id(sk, skb); 1465 sk_incoming_cpu_update(sk); 1466 } 1467 1468 rc = __sock_queue_rcv_skb(sk, skb); 1469 if (rc < 0) { 1470 int is_udplite = IS_UDPLITE(sk); 1471 1472 /* Note that an ENOMEM error is charged twice */ 1473 if (rc == -ENOMEM) 1474 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1475 is_udplite); 1476 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1477 kfree_skb(skb); 1478 trace_udp_fail_queue_rcv_skb(rc, sk); 1479 return -1; 1480 } 1481 1482 return 0; 1483 1484 } 1485 1486 static struct static_key udp_encap_needed __read_mostly; 1487 void udp_encap_enable(void) 1488 { 1489 if (!static_key_enabled(&udp_encap_needed)) 1490 static_key_slow_inc(&udp_encap_needed); 1491 } 1492 EXPORT_SYMBOL(udp_encap_enable); 1493 1494 /* returns: 1495 * -1: error 1496 * 0: success 1497 * >0: "udp encap" protocol resubmission 1498 * 1499 * Note that in the success and error cases, the skb is assumed to 1500 * have either been requeued or freed. 1501 */ 1502 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1503 { 1504 struct udp_sock *up = udp_sk(sk); 1505 int rc; 1506 int is_udplite = IS_UDPLITE(sk); 1507 1508 /* 1509 * Charge it to the socket, dropping if the queue is full. 1510 */ 1511 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1512 goto drop; 1513 nf_reset(skb); 1514 1515 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1516 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1517 1518 /* 1519 * This is an encapsulation socket so pass the skb to 1520 * the socket's udp_encap_rcv() hook. Otherwise, just 1521 * fall through and pass this up the UDP socket. 1522 * up->encap_rcv() returns the following value: 1523 * =0 if skb was successfully passed to the encap 1524 * handler or was discarded by it. 1525 * >0 if skb should be passed on to UDP. 1526 * <0 if skb should be resubmitted as proto -N 1527 */ 1528 1529 /* if we're overly short, let UDP handle it */ 1530 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1531 if (encap_rcv) { 1532 int ret; 1533 1534 /* Verify checksum before giving to encap */ 1535 if (udp_lib_checksum_complete(skb)) 1536 goto csum_error; 1537 1538 ret = encap_rcv(sk, skb); 1539 if (ret <= 0) { 1540 __UDP_INC_STATS(sock_net(sk), 1541 UDP_MIB_INDATAGRAMS, 1542 is_udplite); 1543 return -ret; 1544 } 1545 } 1546 1547 /* FALLTHROUGH -- it's a UDP Packet */ 1548 } 1549 1550 /* 1551 * UDP-Lite specific tests, ignored on UDP sockets 1552 */ 1553 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1554 1555 /* 1556 * MIB statistics other than incrementing the error count are 1557 * disabled for the following two types of errors: these depend 1558 * on the application settings, not on the functioning of the 1559 * protocol stack as such. 1560 * 1561 * RFC 3828 here recommends (sec 3.3): "There should also be a 1562 * way ... to ... at least let the receiving application block 1563 * delivery of packets with coverage values less than a value 1564 * provided by the application." 1565 */ 1566 if (up->pcrlen == 0) { /* full coverage was set */ 1567 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1568 UDP_SKB_CB(skb)->cscov, skb->len); 1569 goto drop; 1570 } 1571 /* The next case involves violating the min. coverage requested 1572 * by the receiver. This is subtle: if receiver wants x and x is 1573 * greater than the buffersize/MTU then receiver will complain 1574 * that it wants x while sender emits packets of smaller size y. 1575 * Therefore the above ...()->partial_cov statement is essential. 1576 */ 1577 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1578 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1579 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1580 goto drop; 1581 } 1582 } 1583 1584 if (rcu_access_pointer(sk->sk_filter) && 1585 udp_lib_checksum_complete(skb)) 1586 goto csum_error; 1587 1588 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 1589 goto drop; 1590 1591 udp_csum_pull_header(skb); 1592 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1593 __UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1594 is_udplite); 1595 goto drop; 1596 } 1597 1598 rc = 0; 1599 1600 ipv4_pktinfo_prepare(sk, skb); 1601 bh_lock_sock(sk); 1602 if (!sock_owned_by_user(sk)) 1603 rc = __udp_queue_rcv_skb(sk, skb); 1604 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1605 bh_unlock_sock(sk); 1606 goto drop; 1607 } 1608 bh_unlock_sock(sk); 1609 1610 return rc; 1611 1612 csum_error: 1613 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1614 drop: 1615 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1616 atomic_inc(&sk->sk_drops); 1617 kfree_skb(skb); 1618 return -1; 1619 } 1620 1621 /* For TCP sockets, sk_rx_dst is protected by socket lock 1622 * For UDP, we use xchg() to guard against concurrent changes. 1623 */ 1624 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1625 { 1626 struct dst_entry *old; 1627 1628 dst_hold(dst); 1629 old = xchg(&sk->sk_rx_dst, dst); 1630 dst_release(old); 1631 } 1632 1633 /* 1634 * Multicasts and broadcasts go to each listener. 1635 * 1636 * Note: called only from the BH handler context. 1637 */ 1638 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1639 struct udphdr *uh, 1640 __be32 saddr, __be32 daddr, 1641 struct udp_table *udptable, 1642 int proto) 1643 { 1644 struct sock *sk, *first = NULL; 1645 unsigned short hnum = ntohs(uh->dest); 1646 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1647 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1648 unsigned int offset = offsetof(typeof(*sk), sk_node); 1649 int dif = skb->dev->ifindex; 1650 struct hlist_node *node; 1651 struct sk_buff *nskb; 1652 1653 if (use_hash2) { 1654 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1655 udptable->mask; 1656 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask; 1657 start_lookup: 1658 hslot = &udptable->hash2[hash2]; 1659 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1660 } 1661 1662 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 1663 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 1664 uh->source, saddr, dif, hnum)) 1665 continue; 1666 1667 if (!first) { 1668 first = sk; 1669 continue; 1670 } 1671 nskb = skb_clone(skb, GFP_ATOMIC); 1672 1673 if (unlikely(!nskb)) { 1674 atomic_inc(&sk->sk_drops); 1675 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 1676 IS_UDPLITE(sk)); 1677 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 1678 IS_UDPLITE(sk)); 1679 continue; 1680 } 1681 if (udp_queue_rcv_skb(sk, nskb) > 0) 1682 consume_skb(nskb); 1683 } 1684 1685 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1686 if (use_hash2 && hash2 != hash2_any) { 1687 hash2 = hash2_any; 1688 goto start_lookup; 1689 } 1690 1691 if (first) { 1692 if (udp_queue_rcv_skb(first, skb) > 0) 1693 consume_skb(skb); 1694 } else { 1695 kfree_skb(skb); 1696 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 1697 proto == IPPROTO_UDPLITE); 1698 } 1699 return 0; 1700 } 1701 1702 /* Initialize UDP checksum. If exited with zero value (success), 1703 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1704 * Otherwise, csum completion requires chacksumming packet body, 1705 * including udp header and folding it to skb->csum. 1706 */ 1707 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1708 int proto) 1709 { 1710 int err; 1711 1712 UDP_SKB_CB(skb)->partial_cov = 0; 1713 UDP_SKB_CB(skb)->cscov = skb->len; 1714 1715 if (proto == IPPROTO_UDPLITE) { 1716 err = udplite_checksum_init(skb, uh); 1717 if (err) 1718 return err; 1719 } 1720 1721 /* Note, we are only interested in != 0 or == 0, thus the 1722 * force to int. 1723 */ 1724 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 1725 inet_compute_pseudo); 1726 } 1727 1728 /* 1729 * All we need to do is get the socket, and then do a checksum. 1730 */ 1731 1732 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1733 int proto) 1734 { 1735 struct sock *sk; 1736 struct udphdr *uh; 1737 unsigned short ulen; 1738 struct rtable *rt = skb_rtable(skb); 1739 __be32 saddr, daddr; 1740 struct net *net = dev_net(skb->dev); 1741 1742 /* 1743 * Validate the packet. 1744 */ 1745 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1746 goto drop; /* No space for header. */ 1747 1748 uh = udp_hdr(skb); 1749 ulen = ntohs(uh->len); 1750 saddr = ip_hdr(skb)->saddr; 1751 daddr = ip_hdr(skb)->daddr; 1752 1753 if (ulen > skb->len) 1754 goto short_packet; 1755 1756 if (proto == IPPROTO_UDP) { 1757 /* UDP validates ulen. */ 1758 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1759 goto short_packet; 1760 uh = udp_hdr(skb); 1761 } 1762 1763 if (udp4_csum_init(skb, uh, proto)) 1764 goto csum_error; 1765 1766 sk = skb_steal_sock(skb); 1767 if (sk) { 1768 struct dst_entry *dst = skb_dst(skb); 1769 int ret; 1770 1771 if (unlikely(sk->sk_rx_dst != dst)) 1772 udp_sk_rx_dst_set(sk, dst); 1773 1774 ret = udp_queue_rcv_skb(sk, skb); 1775 sock_put(sk); 1776 /* a return value > 0 means to resubmit the input, but 1777 * it wants the return to be -protocol, or 0 1778 */ 1779 if (ret > 0) 1780 return -ret; 1781 return 0; 1782 } 1783 1784 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1785 return __udp4_lib_mcast_deliver(net, skb, uh, 1786 saddr, daddr, udptable, proto); 1787 1788 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1789 if (sk) { 1790 int ret; 1791 1792 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1793 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1794 inet_compute_pseudo); 1795 1796 ret = udp_queue_rcv_skb(sk, skb); 1797 1798 /* a return value > 0 means to resubmit the input, but 1799 * it wants the return to be -protocol, or 0 1800 */ 1801 if (ret > 0) 1802 return -ret; 1803 return 0; 1804 } 1805 1806 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1807 goto drop; 1808 nf_reset(skb); 1809 1810 /* No socket. Drop packet silently, if checksum is wrong */ 1811 if (udp_lib_checksum_complete(skb)) 1812 goto csum_error; 1813 1814 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1815 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1816 1817 /* 1818 * Hmm. We got an UDP packet to a port to which we 1819 * don't wanna listen. Ignore it. 1820 */ 1821 kfree_skb(skb); 1822 return 0; 1823 1824 short_packet: 1825 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1826 proto == IPPROTO_UDPLITE ? "Lite" : "", 1827 &saddr, ntohs(uh->source), 1828 ulen, skb->len, 1829 &daddr, ntohs(uh->dest)); 1830 goto drop; 1831 1832 csum_error: 1833 /* 1834 * RFC1122: OK. Discards the bad packet silently (as far as 1835 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1836 */ 1837 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1838 proto == IPPROTO_UDPLITE ? "Lite" : "", 1839 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1840 ulen); 1841 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1842 drop: 1843 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1844 kfree_skb(skb); 1845 return 0; 1846 } 1847 1848 /* We can only early demux multicast if there is a single matching socket. 1849 * If more than one socket found returns NULL 1850 */ 1851 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1852 __be16 loc_port, __be32 loc_addr, 1853 __be16 rmt_port, __be32 rmt_addr, 1854 int dif) 1855 { 1856 struct sock *sk, *result; 1857 unsigned short hnum = ntohs(loc_port); 1858 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 1859 struct udp_hslot *hslot = &udp_table.hash[slot]; 1860 1861 /* Do not bother scanning a too big list */ 1862 if (hslot->count > 10) 1863 return NULL; 1864 1865 result = NULL; 1866 sk_for_each_rcu(sk, &hslot->head) { 1867 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 1868 rmt_port, rmt_addr, dif, hnum)) { 1869 if (result) 1870 return NULL; 1871 result = sk; 1872 } 1873 } 1874 1875 return result; 1876 } 1877 1878 /* For unicast we should only early demux connected sockets or we can 1879 * break forwarding setups. The chains here can be long so only check 1880 * if the first socket is an exact match and if not move on. 1881 */ 1882 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1883 __be16 loc_port, __be32 loc_addr, 1884 __be16 rmt_port, __be32 rmt_addr, 1885 int dif) 1886 { 1887 unsigned short hnum = ntohs(loc_port); 1888 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1889 unsigned int slot2 = hash2 & udp_table.mask; 1890 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1891 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 1892 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1893 struct sock *sk; 1894 1895 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1896 if (INET_MATCH(sk, net, acookie, rmt_addr, 1897 loc_addr, ports, dif)) 1898 return sk; 1899 /* Only check first socket in chain */ 1900 break; 1901 } 1902 return NULL; 1903 } 1904 1905 void udp_v4_early_demux(struct sk_buff *skb) 1906 { 1907 struct net *net = dev_net(skb->dev); 1908 const struct iphdr *iph; 1909 const struct udphdr *uh; 1910 struct sock *sk = NULL; 1911 struct dst_entry *dst; 1912 int dif = skb->dev->ifindex; 1913 int ours; 1914 1915 /* validate the packet */ 1916 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1917 return; 1918 1919 iph = ip_hdr(skb); 1920 uh = udp_hdr(skb); 1921 1922 if (skb->pkt_type == PACKET_BROADCAST || 1923 skb->pkt_type == PACKET_MULTICAST) { 1924 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 1925 1926 if (!in_dev) 1927 return; 1928 1929 /* we are supposed to accept bcast packets */ 1930 if (skb->pkt_type == PACKET_MULTICAST) { 1931 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 1932 iph->protocol); 1933 if (!ours) 1934 return; 1935 } 1936 1937 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1938 uh->source, iph->saddr, dif); 1939 } else if (skb->pkt_type == PACKET_HOST) { 1940 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1941 uh->source, iph->saddr, dif); 1942 } 1943 1944 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2)) 1945 return; 1946 1947 skb->sk = sk; 1948 skb->destructor = sock_efree; 1949 dst = READ_ONCE(sk->sk_rx_dst); 1950 1951 if (dst) 1952 dst = dst_check(dst, 0); 1953 if (dst) { 1954 /* DST_NOCACHE can not be used without taking a reference */ 1955 if (dst->flags & DST_NOCACHE) { 1956 if (likely(atomic_inc_not_zero(&dst->__refcnt))) 1957 skb_dst_set(skb, dst); 1958 } else { 1959 skb_dst_set_noref(skb, dst); 1960 } 1961 } 1962 } 1963 1964 int udp_rcv(struct sk_buff *skb) 1965 { 1966 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1967 } 1968 1969 void udp_destroy_sock(struct sock *sk) 1970 { 1971 struct udp_sock *up = udp_sk(sk); 1972 bool slow = lock_sock_fast(sk); 1973 udp_flush_pending_frames(sk); 1974 unlock_sock_fast(sk, slow); 1975 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1976 void (*encap_destroy)(struct sock *sk); 1977 encap_destroy = ACCESS_ONCE(up->encap_destroy); 1978 if (encap_destroy) 1979 encap_destroy(sk); 1980 } 1981 } 1982 1983 /* 1984 * Socket option code for UDP 1985 */ 1986 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1987 char __user *optval, unsigned int optlen, 1988 int (*push_pending_frames)(struct sock *)) 1989 { 1990 struct udp_sock *up = udp_sk(sk); 1991 int val, valbool; 1992 int err = 0; 1993 int is_udplite = IS_UDPLITE(sk); 1994 1995 if (optlen < sizeof(int)) 1996 return -EINVAL; 1997 1998 if (get_user(val, (int __user *)optval)) 1999 return -EFAULT; 2000 2001 valbool = val ? 1 : 0; 2002 2003 switch (optname) { 2004 case UDP_CORK: 2005 if (val != 0) { 2006 up->corkflag = 1; 2007 } else { 2008 up->corkflag = 0; 2009 lock_sock(sk); 2010 push_pending_frames(sk); 2011 release_sock(sk); 2012 } 2013 break; 2014 2015 case UDP_ENCAP: 2016 switch (val) { 2017 case 0: 2018 case UDP_ENCAP_ESPINUDP: 2019 case UDP_ENCAP_ESPINUDP_NON_IKE: 2020 up->encap_rcv = xfrm4_udp_encap_rcv; 2021 /* FALLTHROUGH */ 2022 case UDP_ENCAP_L2TPINUDP: 2023 up->encap_type = val; 2024 udp_encap_enable(); 2025 break; 2026 default: 2027 err = -ENOPROTOOPT; 2028 break; 2029 } 2030 break; 2031 2032 case UDP_NO_CHECK6_TX: 2033 up->no_check6_tx = valbool; 2034 break; 2035 2036 case UDP_NO_CHECK6_RX: 2037 up->no_check6_rx = valbool; 2038 break; 2039 2040 /* 2041 * UDP-Lite's partial checksum coverage (RFC 3828). 2042 */ 2043 /* The sender sets actual checksum coverage length via this option. 2044 * The case coverage > packet length is handled by send module. */ 2045 case UDPLITE_SEND_CSCOV: 2046 if (!is_udplite) /* Disable the option on UDP sockets */ 2047 return -ENOPROTOOPT; 2048 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2049 val = 8; 2050 else if (val > USHRT_MAX) 2051 val = USHRT_MAX; 2052 up->pcslen = val; 2053 up->pcflag |= UDPLITE_SEND_CC; 2054 break; 2055 2056 /* The receiver specifies a minimum checksum coverage value. To make 2057 * sense, this should be set to at least 8 (as done below). If zero is 2058 * used, this again means full checksum coverage. */ 2059 case UDPLITE_RECV_CSCOV: 2060 if (!is_udplite) /* Disable the option on UDP sockets */ 2061 return -ENOPROTOOPT; 2062 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2063 val = 8; 2064 else if (val > USHRT_MAX) 2065 val = USHRT_MAX; 2066 up->pcrlen = val; 2067 up->pcflag |= UDPLITE_RECV_CC; 2068 break; 2069 2070 default: 2071 err = -ENOPROTOOPT; 2072 break; 2073 } 2074 2075 return err; 2076 } 2077 EXPORT_SYMBOL(udp_lib_setsockopt); 2078 2079 int udp_setsockopt(struct sock *sk, int level, int optname, 2080 char __user *optval, unsigned int optlen) 2081 { 2082 if (level == SOL_UDP || level == SOL_UDPLITE) 2083 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2084 udp_push_pending_frames); 2085 return ip_setsockopt(sk, level, optname, optval, optlen); 2086 } 2087 2088 #ifdef CONFIG_COMPAT 2089 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2090 char __user *optval, unsigned int optlen) 2091 { 2092 if (level == SOL_UDP || level == SOL_UDPLITE) 2093 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2094 udp_push_pending_frames); 2095 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2096 } 2097 #endif 2098 2099 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2100 char __user *optval, int __user *optlen) 2101 { 2102 struct udp_sock *up = udp_sk(sk); 2103 int val, len; 2104 2105 if (get_user(len, optlen)) 2106 return -EFAULT; 2107 2108 len = min_t(unsigned int, len, sizeof(int)); 2109 2110 if (len < 0) 2111 return -EINVAL; 2112 2113 switch (optname) { 2114 case UDP_CORK: 2115 val = up->corkflag; 2116 break; 2117 2118 case UDP_ENCAP: 2119 val = up->encap_type; 2120 break; 2121 2122 case UDP_NO_CHECK6_TX: 2123 val = up->no_check6_tx; 2124 break; 2125 2126 case UDP_NO_CHECK6_RX: 2127 val = up->no_check6_rx; 2128 break; 2129 2130 /* The following two cannot be changed on UDP sockets, the return is 2131 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2132 case UDPLITE_SEND_CSCOV: 2133 val = up->pcslen; 2134 break; 2135 2136 case UDPLITE_RECV_CSCOV: 2137 val = up->pcrlen; 2138 break; 2139 2140 default: 2141 return -ENOPROTOOPT; 2142 } 2143 2144 if (put_user(len, optlen)) 2145 return -EFAULT; 2146 if (copy_to_user(optval, &val, len)) 2147 return -EFAULT; 2148 return 0; 2149 } 2150 EXPORT_SYMBOL(udp_lib_getsockopt); 2151 2152 int udp_getsockopt(struct sock *sk, int level, int optname, 2153 char __user *optval, int __user *optlen) 2154 { 2155 if (level == SOL_UDP || level == SOL_UDPLITE) 2156 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2157 return ip_getsockopt(sk, level, optname, optval, optlen); 2158 } 2159 2160 #ifdef CONFIG_COMPAT 2161 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2162 char __user *optval, int __user *optlen) 2163 { 2164 if (level == SOL_UDP || level == SOL_UDPLITE) 2165 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2166 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2167 } 2168 #endif 2169 /** 2170 * udp_poll - wait for a UDP event. 2171 * @file - file struct 2172 * @sock - socket 2173 * @wait - poll table 2174 * 2175 * This is same as datagram poll, except for the special case of 2176 * blocking sockets. If application is using a blocking fd 2177 * and a packet with checksum error is in the queue; 2178 * then it could get return from select indicating data available 2179 * but then block when reading it. Add special case code 2180 * to work around these arguably broken applications. 2181 */ 2182 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2183 { 2184 unsigned int mask = datagram_poll(file, sock, wait); 2185 struct sock *sk = sock->sk; 2186 2187 sock_rps_record_flow(sk); 2188 2189 /* Check for false positives due to checksum errors */ 2190 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2191 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2192 mask &= ~(POLLIN | POLLRDNORM); 2193 2194 return mask; 2195 2196 } 2197 EXPORT_SYMBOL(udp_poll); 2198 2199 int udp_abort(struct sock *sk, int err) 2200 { 2201 lock_sock(sk); 2202 2203 sk->sk_err = err; 2204 sk->sk_error_report(sk); 2205 __udp_disconnect(sk, 0); 2206 2207 release_sock(sk); 2208 2209 return 0; 2210 } 2211 EXPORT_SYMBOL_GPL(udp_abort); 2212 2213 struct proto udp_prot = { 2214 .name = "UDP", 2215 .owner = THIS_MODULE, 2216 .close = udp_lib_close, 2217 .connect = ip4_datagram_connect, 2218 .disconnect = udp_disconnect, 2219 .ioctl = udp_ioctl, 2220 .destroy = udp_destroy_sock, 2221 .setsockopt = udp_setsockopt, 2222 .getsockopt = udp_getsockopt, 2223 .sendmsg = udp_sendmsg, 2224 .recvmsg = udp_recvmsg, 2225 .sendpage = udp_sendpage, 2226 .backlog_rcv = __udp_queue_rcv_skb, 2227 .release_cb = ip4_datagram_release_cb, 2228 .hash = udp_lib_hash, 2229 .unhash = udp_lib_unhash, 2230 .rehash = udp_v4_rehash, 2231 .get_port = udp_v4_get_port, 2232 .memory_allocated = &udp_memory_allocated, 2233 .sysctl_mem = sysctl_udp_mem, 2234 .sysctl_wmem = &sysctl_udp_wmem_min, 2235 .sysctl_rmem = &sysctl_udp_rmem_min, 2236 .obj_size = sizeof(struct udp_sock), 2237 .h.udp_table = &udp_table, 2238 #ifdef CONFIG_COMPAT 2239 .compat_setsockopt = compat_udp_setsockopt, 2240 .compat_getsockopt = compat_udp_getsockopt, 2241 #endif 2242 .diag_destroy = udp_abort, 2243 }; 2244 EXPORT_SYMBOL(udp_prot); 2245 2246 /* ------------------------------------------------------------------------ */ 2247 #ifdef CONFIG_PROC_FS 2248 2249 static struct sock *udp_get_first(struct seq_file *seq, int start) 2250 { 2251 struct sock *sk; 2252 struct udp_iter_state *state = seq->private; 2253 struct net *net = seq_file_net(seq); 2254 2255 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2256 ++state->bucket) { 2257 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2258 2259 if (hlist_empty(&hslot->head)) 2260 continue; 2261 2262 spin_lock_bh(&hslot->lock); 2263 sk_for_each(sk, &hslot->head) { 2264 if (!net_eq(sock_net(sk), net)) 2265 continue; 2266 if (sk->sk_family == state->family) 2267 goto found; 2268 } 2269 spin_unlock_bh(&hslot->lock); 2270 } 2271 sk = NULL; 2272 found: 2273 return sk; 2274 } 2275 2276 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2277 { 2278 struct udp_iter_state *state = seq->private; 2279 struct net *net = seq_file_net(seq); 2280 2281 do { 2282 sk = sk_next(sk); 2283 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2284 2285 if (!sk) { 2286 if (state->bucket <= state->udp_table->mask) 2287 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2288 return udp_get_first(seq, state->bucket + 1); 2289 } 2290 return sk; 2291 } 2292 2293 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2294 { 2295 struct sock *sk = udp_get_first(seq, 0); 2296 2297 if (sk) 2298 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2299 --pos; 2300 return pos ? NULL : sk; 2301 } 2302 2303 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2304 { 2305 struct udp_iter_state *state = seq->private; 2306 state->bucket = MAX_UDP_PORTS; 2307 2308 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2309 } 2310 2311 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2312 { 2313 struct sock *sk; 2314 2315 if (v == SEQ_START_TOKEN) 2316 sk = udp_get_idx(seq, 0); 2317 else 2318 sk = udp_get_next(seq, v); 2319 2320 ++*pos; 2321 return sk; 2322 } 2323 2324 static void udp_seq_stop(struct seq_file *seq, void *v) 2325 { 2326 struct udp_iter_state *state = seq->private; 2327 2328 if (state->bucket <= state->udp_table->mask) 2329 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2330 } 2331 2332 int udp_seq_open(struct inode *inode, struct file *file) 2333 { 2334 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2335 struct udp_iter_state *s; 2336 int err; 2337 2338 err = seq_open_net(inode, file, &afinfo->seq_ops, 2339 sizeof(struct udp_iter_state)); 2340 if (err < 0) 2341 return err; 2342 2343 s = ((struct seq_file *)file->private_data)->private; 2344 s->family = afinfo->family; 2345 s->udp_table = afinfo->udp_table; 2346 return err; 2347 } 2348 EXPORT_SYMBOL(udp_seq_open); 2349 2350 /* ------------------------------------------------------------------------ */ 2351 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2352 { 2353 struct proc_dir_entry *p; 2354 int rc = 0; 2355 2356 afinfo->seq_ops.start = udp_seq_start; 2357 afinfo->seq_ops.next = udp_seq_next; 2358 afinfo->seq_ops.stop = udp_seq_stop; 2359 2360 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2361 afinfo->seq_fops, afinfo); 2362 if (!p) 2363 rc = -ENOMEM; 2364 return rc; 2365 } 2366 EXPORT_SYMBOL(udp_proc_register); 2367 2368 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2369 { 2370 remove_proc_entry(afinfo->name, net->proc_net); 2371 } 2372 EXPORT_SYMBOL(udp_proc_unregister); 2373 2374 /* ------------------------------------------------------------------------ */ 2375 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2376 int bucket) 2377 { 2378 struct inet_sock *inet = inet_sk(sp); 2379 __be32 dest = inet->inet_daddr; 2380 __be32 src = inet->inet_rcv_saddr; 2381 __u16 destp = ntohs(inet->inet_dport); 2382 __u16 srcp = ntohs(inet->inet_sport); 2383 2384 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2385 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2386 bucket, src, srcp, dest, destp, sp->sk_state, 2387 sk_wmem_alloc_get(sp), 2388 sk_rmem_alloc_get(sp), 2389 0, 0L, 0, 2390 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2391 0, sock_i_ino(sp), 2392 atomic_read(&sp->sk_refcnt), sp, 2393 atomic_read(&sp->sk_drops)); 2394 } 2395 2396 int udp4_seq_show(struct seq_file *seq, void *v) 2397 { 2398 seq_setwidth(seq, 127); 2399 if (v == SEQ_START_TOKEN) 2400 seq_puts(seq, " sl local_address rem_address st tx_queue " 2401 "rx_queue tr tm->when retrnsmt uid timeout " 2402 "inode ref pointer drops"); 2403 else { 2404 struct udp_iter_state *state = seq->private; 2405 2406 udp4_format_sock(v, seq, state->bucket); 2407 } 2408 seq_pad(seq, '\n'); 2409 return 0; 2410 } 2411 2412 static const struct file_operations udp_afinfo_seq_fops = { 2413 .owner = THIS_MODULE, 2414 .open = udp_seq_open, 2415 .read = seq_read, 2416 .llseek = seq_lseek, 2417 .release = seq_release_net 2418 }; 2419 2420 /* ------------------------------------------------------------------------ */ 2421 static struct udp_seq_afinfo udp4_seq_afinfo = { 2422 .name = "udp", 2423 .family = AF_INET, 2424 .udp_table = &udp_table, 2425 .seq_fops = &udp_afinfo_seq_fops, 2426 .seq_ops = { 2427 .show = udp4_seq_show, 2428 }, 2429 }; 2430 2431 static int __net_init udp4_proc_init_net(struct net *net) 2432 { 2433 return udp_proc_register(net, &udp4_seq_afinfo); 2434 } 2435 2436 static void __net_exit udp4_proc_exit_net(struct net *net) 2437 { 2438 udp_proc_unregister(net, &udp4_seq_afinfo); 2439 } 2440 2441 static struct pernet_operations udp4_net_ops = { 2442 .init = udp4_proc_init_net, 2443 .exit = udp4_proc_exit_net, 2444 }; 2445 2446 int __init udp4_proc_init(void) 2447 { 2448 return register_pernet_subsys(&udp4_net_ops); 2449 } 2450 2451 void udp4_proc_exit(void) 2452 { 2453 unregister_pernet_subsys(&udp4_net_ops); 2454 } 2455 #endif /* CONFIG_PROC_FS */ 2456 2457 static __initdata unsigned long uhash_entries; 2458 static int __init set_uhash_entries(char *str) 2459 { 2460 ssize_t ret; 2461 2462 if (!str) 2463 return 0; 2464 2465 ret = kstrtoul(str, 0, &uhash_entries); 2466 if (ret) 2467 return 0; 2468 2469 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2470 uhash_entries = UDP_HTABLE_SIZE_MIN; 2471 return 1; 2472 } 2473 __setup("uhash_entries=", set_uhash_entries); 2474 2475 void __init udp_table_init(struct udp_table *table, const char *name) 2476 { 2477 unsigned int i; 2478 2479 table->hash = alloc_large_system_hash(name, 2480 2 * sizeof(struct udp_hslot), 2481 uhash_entries, 2482 21, /* one slot per 2 MB */ 2483 0, 2484 &table->log, 2485 &table->mask, 2486 UDP_HTABLE_SIZE_MIN, 2487 64 * 1024); 2488 2489 table->hash2 = table->hash + (table->mask + 1); 2490 for (i = 0; i <= table->mask; i++) { 2491 INIT_HLIST_HEAD(&table->hash[i].head); 2492 table->hash[i].count = 0; 2493 spin_lock_init(&table->hash[i].lock); 2494 } 2495 for (i = 0; i <= table->mask; i++) { 2496 INIT_HLIST_HEAD(&table->hash2[i].head); 2497 table->hash2[i].count = 0; 2498 spin_lock_init(&table->hash2[i].lock); 2499 } 2500 } 2501 2502 u32 udp_flow_hashrnd(void) 2503 { 2504 static u32 hashrnd __read_mostly; 2505 2506 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2507 2508 return hashrnd; 2509 } 2510 EXPORT_SYMBOL(udp_flow_hashrnd); 2511 2512 void __init udp_init(void) 2513 { 2514 unsigned long limit; 2515 2516 udp_table_init(&udp_table, "UDP"); 2517 limit = nr_free_buffer_pages() / 8; 2518 limit = max(limit, 128UL); 2519 sysctl_udp_mem[0] = limit / 4 * 3; 2520 sysctl_udp_mem[1] = limit; 2521 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2522 2523 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2524 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2525 } 2526