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> 117 118 struct udp_table udp_table __read_mostly; 119 EXPORT_SYMBOL(udp_table); 120 121 long sysctl_udp_mem[3] __read_mostly; 122 EXPORT_SYMBOL(sysctl_udp_mem); 123 124 int sysctl_udp_rmem_min __read_mostly; 125 EXPORT_SYMBOL(sysctl_udp_rmem_min); 126 127 int sysctl_udp_wmem_min __read_mostly; 128 EXPORT_SYMBOL(sysctl_udp_wmem_min); 129 130 atomic_long_t udp_memory_allocated; 131 EXPORT_SYMBOL(udp_memory_allocated); 132 133 #define MAX_UDP_PORTS 65536 134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 135 136 static int udp_lib_lport_inuse(struct net *net, __u16 num, 137 const struct udp_hslot *hslot, 138 unsigned long *bitmap, 139 struct sock *sk, 140 int (*saddr_comp)(const struct sock *sk1, 141 const struct sock *sk2, 142 bool match_wildcard), 143 unsigned int log) 144 { 145 struct sock *sk2; 146 kuid_t uid = sock_i_uid(sk); 147 148 sk_for_each(sk2, &hslot->head) { 149 if (net_eq(sock_net(sk2), net) && 150 sk2 != sk && 151 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 152 (!sk2->sk_reuse || !sk->sk_reuse) && 153 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 154 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 155 (!sk2->sk_reuseport || !sk->sk_reuseport || 156 rcu_access_pointer(sk->sk_reuseport_cb) || 157 !uid_eq(uid, sock_i_uid(sk2))) && 158 saddr_comp(sk, sk2, true)) { 159 if (!bitmap) 160 return 1; 161 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap); 162 } 163 } 164 return 0; 165 } 166 167 /* 168 * Note: we still hold spinlock of primary hash chain, so no other writer 169 * can insert/delete a socket with local_port == num 170 */ 171 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 172 struct udp_hslot *hslot2, 173 struct sock *sk, 174 int (*saddr_comp)(const struct sock *sk1, 175 const struct sock *sk2, 176 bool match_wildcard)) 177 { 178 struct sock *sk2; 179 kuid_t uid = sock_i_uid(sk); 180 int res = 0; 181 182 spin_lock(&hslot2->lock); 183 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 184 if (net_eq(sock_net(sk2), net) && 185 sk2 != sk && 186 (udp_sk(sk2)->udp_port_hash == num) && 187 (!sk2->sk_reuse || !sk->sk_reuse) && 188 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 189 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 190 (!sk2->sk_reuseport || !sk->sk_reuseport || 191 rcu_access_pointer(sk->sk_reuseport_cb) || 192 !uid_eq(uid, sock_i_uid(sk2))) && 193 saddr_comp(sk, sk2, true)) { 194 res = 1; 195 break; 196 } 197 } 198 spin_unlock(&hslot2->lock); 199 return res; 200 } 201 202 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot, 203 int (*saddr_same)(const struct sock *sk1, 204 const struct sock *sk2, 205 bool match_wildcard)) 206 { 207 struct net *net = sock_net(sk); 208 kuid_t uid = sock_i_uid(sk); 209 struct sock *sk2; 210 211 sk_for_each(sk2, &hslot->head) { 212 if (net_eq(sock_net(sk2), net) && 213 sk2 != sk && 214 sk2->sk_family == sk->sk_family && 215 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 216 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 217 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 218 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 219 (*saddr_same)(sk, sk2, false)) { 220 return reuseport_add_sock(sk, sk2); 221 } 222 } 223 224 /* Initial allocation may have already happened via setsockopt */ 225 if (!rcu_access_pointer(sk->sk_reuseport_cb)) 226 return reuseport_alloc(sk); 227 return 0; 228 } 229 230 /** 231 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 232 * 233 * @sk: socket struct in question 234 * @snum: port number to look up 235 * @saddr_comp: AF-dependent comparison of bound local IP addresses 236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 237 * with NULL address 238 */ 239 int udp_lib_get_port(struct sock *sk, unsigned short snum, 240 int (*saddr_comp)(const struct sock *sk1, 241 const struct sock *sk2, 242 bool match_wildcard), 243 unsigned int hash2_nulladdr) 244 { 245 struct udp_hslot *hslot, *hslot2; 246 struct udp_table *udptable = sk->sk_prot->h.udp_table; 247 int error = 1; 248 struct net *net = sock_net(sk); 249 250 if (!snum) { 251 int low, high, remaining; 252 unsigned int rand; 253 unsigned short first, last; 254 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 255 256 inet_get_local_port_range(net, &low, &high); 257 remaining = (high - low) + 1; 258 259 rand = prandom_u32(); 260 first = reciprocal_scale(rand, remaining) + low; 261 /* 262 * force rand to be an odd multiple of UDP_HTABLE_SIZE 263 */ 264 rand = (rand | 1) * (udptable->mask + 1); 265 last = first + udptable->mask + 1; 266 do { 267 hslot = udp_hashslot(udptable, net, first); 268 bitmap_zero(bitmap, PORTS_PER_CHAIN); 269 spin_lock_bh(&hslot->lock); 270 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 271 saddr_comp, udptable->log); 272 273 snum = first; 274 /* 275 * Iterate on all possible values of snum for this hash. 276 * Using steps of an odd multiple of UDP_HTABLE_SIZE 277 * give us randomization and full range coverage. 278 */ 279 do { 280 if (low <= snum && snum <= high && 281 !test_bit(snum >> udptable->log, bitmap) && 282 !inet_is_local_reserved_port(net, snum)) 283 goto found; 284 snum += rand; 285 } while (snum != first); 286 spin_unlock_bh(&hslot->lock); 287 } while (++first != last); 288 goto fail; 289 } else { 290 hslot = udp_hashslot(udptable, net, snum); 291 spin_lock_bh(&hslot->lock); 292 if (hslot->count > 10) { 293 int exist; 294 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 295 296 slot2 &= udptable->mask; 297 hash2_nulladdr &= udptable->mask; 298 299 hslot2 = udp_hashslot2(udptable, slot2); 300 if (hslot->count < hslot2->count) 301 goto scan_primary_hash; 302 303 exist = udp_lib_lport_inuse2(net, snum, hslot2, 304 sk, saddr_comp); 305 if (!exist && (hash2_nulladdr != slot2)) { 306 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 307 exist = udp_lib_lport_inuse2(net, snum, hslot2, 308 sk, saddr_comp); 309 } 310 if (exist) 311 goto fail_unlock; 312 else 313 goto found; 314 } 315 scan_primary_hash: 316 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 317 saddr_comp, 0)) 318 goto fail_unlock; 319 } 320 found: 321 inet_sk(sk)->inet_num = snum; 322 udp_sk(sk)->udp_port_hash = snum; 323 udp_sk(sk)->udp_portaddr_hash ^= snum; 324 if (sk_unhashed(sk)) { 325 if (sk->sk_reuseport && 326 udp_reuseport_add_sock(sk, hslot, saddr_comp)) { 327 inet_sk(sk)->inet_num = 0; 328 udp_sk(sk)->udp_port_hash = 0; 329 udp_sk(sk)->udp_portaddr_hash ^= snum; 330 goto fail_unlock; 331 } 332 333 sk_add_node_rcu(sk, &hslot->head); 334 hslot->count++; 335 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 336 337 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 338 spin_lock(&hslot2->lock); 339 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 340 sk->sk_family == AF_INET6) 341 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 342 &hslot2->head); 343 else 344 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 345 &hslot2->head); 346 hslot2->count++; 347 spin_unlock(&hslot2->lock); 348 } 349 sock_set_flag(sk, SOCK_RCU_FREE); 350 error = 0; 351 fail_unlock: 352 spin_unlock_bh(&hslot->lock); 353 fail: 354 return error; 355 } 356 EXPORT_SYMBOL(udp_lib_get_port); 357 358 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses 359 * match_wildcard == false: addresses must be exactly the same, i.e. 360 * 0.0.0.0 only equals to 0.0.0.0 361 */ 362 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2, 363 bool match_wildcard) 364 { 365 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 366 367 if (!ipv6_only_sock(sk2)) { 368 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr) 369 return 1; 370 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr) 371 return match_wildcard; 372 } 373 return 0; 374 } 375 376 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, 377 unsigned int port) 378 { 379 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 380 } 381 382 int udp_v4_get_port(struct sock *sk, unsigned short snum) 383 { 384 unsigned int hash2_nulladdr = 385 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 386 unsigned int hash2_partial = 387 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 388 389 /* precompute partial secondary hash */ 390 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 391 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 392 } 393 394 static inline int compute_score(struct sock *sk, struct net *net, 395 __be32 saddr, unsigned short hnum, __be16 sport, 396 __be32 daddr, __be16 dport, int dif) 397 { 398 int score; 399 struct inet_sock *inet; 400 401 if (!net_eq(sock_net(sk), net) || 402 udp_sk(sk)->udp_port_hash != hnum || 403 ipv6_only_sock(sk)) 404 return -1; 405 406 score = (sk->sk_family == PF_INET) ? 2 : 1; 407 inet = inet_sk(sk); 408 409 if (inet->inet_rcv_saddr) { 410 if (inet->inet_rcv_saddr != daddr) 411 return -1; 412 score += 4; 413 } 414 415 if (inet->inet_daddr) { 416 if (inet->inet_daddr != saddr) 417 return -1; 418 score += 4; 419 } 420 421 if (inet->inet_dport) { 422 if (inet->inet_dport != sport) 423 return -1; 424 score += 4; 425 } 426 427 if (sk->sk_bound_dev_if) { 428 if (sk->sk_bound_dev_if != dif) 429 return -1; 430 score += 4; 431 } 432 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 433 score++; 434 return score; 435 } 436 437 /* 438 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 439 */ 440 static inline int compute_score2(struct sock *sk, struct net *net, 441 __be32 saddr, __be16 sport, 442 __be32 daddr, unsigned int hnum, int dif) 443 { 444 int score; 445 struct inet_sock *inet; 446 447 if (!net_eq(sock_net(sk), net) || 448 ipv6_only_sock(sk)) 449 return -1; 450 451 inet = inet_sk(sk); 452 453 if (inet->inet_rcv_saddr != daddr || 454 inet->inet_num != hnum) 455 return -1; 456 457 score = (sk->sk_family == PF_INET) ? 2 : 1; 458 459 if (inet->inet_daddr) { 460 if (inet->inet_daddr != saddr) 461 return -1; 462 score += 4; 463 } 464 465 if (inet->inet_dport) { 466 if (inet->inet_dport != sport) 467 return -1; 468 score += 4; 469 } 470 471 if (sk->sk_bound_dev_if) { 472 if (sk->sk_bound_dev_if != dif) 473 return -1; 474 score += 4; 475 } 476 477 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 478 score++; 479 480 return score; 481 } 482 483 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 484 const __u16 lport, const __be32 faddr, 485 const __be16 fport) 486 { 487 static u32 udp_ehash_secret __read_mostly; 488 489 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 490 491 return __inet_ehashfn(laddr, lport, faddr, fport, 492 udp_ehash_secret + net_hash_mix(net)); 493 } 494 495 /* called with read_rcu_lock() */ 496 static struct sock *udp4_lib_lookup2(struct net *net, 497 __be32 saddr, __be16 sport, 498 __be32 daddr, unsigned int hnum, int dif, 499 struct udp_hslot *hslot2, unsigned int slot2, 500 struct sk_buff *skb) 501 { 502 struct sock *sk, *result; 503 int score, badness, matches = 0, reuseport = 0; 504 u32 hash = 0; 505 506 result = NULL; 507 badness = 0; 508 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 509 score = compute_score2(sk, net, saddr, sport, 510 daddr, hnum, dif); 511 if (score > badness) { 512 reuseport = sk->sk_reuseport; 513 if (reuseport) { 514 hash = udp_ehashfn(net, daddr, hnum, 515 saddr, sport); 516 result = reuseport_select_sock(sk, hash, skb, 517 sizeof(struct udphdr)); 518 if (result) 519 return result; 520 matches = 1; 521 } 522 badness = score; 523 result = sk; 524 } else if (score == badness && reuseport) { 525 matches++; 526 if (reciprocal_scale(hash, matches) == 0) 527 result = sk; 528 hash = next_pseudo_random32(hash); 529 } 530 } 531 return result; 532 } 533 534 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 535 * harder than this. -DaveM 536 */ 537 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 538 __be16 sport, __be32 daddr, __be16 dport, 539 int dif, struct udp_table *udptable, struct sk_buff *skb) 540 { 541 struct sock *sk, *result; 542 unsigned short hnum = ntohs(dport); 543 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 544 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 545 int score, badness, matches = 0, reuseport = 0; 546 u32 hash = 0; 547 548 if (hslot->count > 10) { 549 hash2 = udp4_portaddr_hash(net, daddr, hnum); 550 slot2 = hash2 & udptable->mask; 551 hslot2 = &udptable->hash2[slot2]; 552 if (hslot->count < hslot2->count) 553 goto begin; 554 555 result = udp4_lib_lookup2(net, saddr, sport, 556 daddr, hnum, dif, 557 hslot2, slot2, skb); 558 if (!result) { 559 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 560 slot2 = hash2 & udptable->mask; 561 hslot2 = &udptable->hash2[slot2]; 562 if (hslot->count < hslot2->count) 563 goto begin; 564 565 result = udp4_lib_lookup2(net, saddr, sport, 566 htonl(INADDR_ANY), hnum, dif, 567 hslot2, slot2, skb); 568 } 569 return result; 570 } 571 begin: 572 result = NULL; 573 badness = 0; 574 sk_for_each_rcu(sk, &hslot->head) { 575 score = compute_score(sk, net, saddr, hnum, sport, 576 daddr, dport, dif); 577 if (score > badness) { 578 reuseport = sk->sk_reuseport; 579 if (reuseport) { 580 hash = udp_ehashfn(net, daddr, hnum, 581 saddr, sport); 582 result = reuseport_select_sock(sk, hash, skb, 583 sizeof(struct udphdr)); 584 if (result) 585 return result; 586 matches = 1; 587 } 588 result = sk; 589 badness = score; 590 } else if (score == badness && reuseport) { 591 matches++; 592 if (reciprocal_scale(hash, matches) == 0) 593 result = sk; 594 hash = next_pseudo_random32(hash); 595 } 596 } 597 return result; 598 } 599 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 600 601 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 602 __be16 sport, __be16 dport, 603 struct udp_table *udptable) 604 { 605 const struct iphdr *iph = ip_hdr(skb); 606 607 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 608 iph->daddr, dport, inet_iif(skb), 609 udptable, skb); 610 } 611 612 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, 613 __be16 sport, __be16 dport) 614 { 615 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); 616 } 617 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); 618 619 /* Must be called under rcu_read_lock(). 620 * Does increment socket refcount. 621 */ 622 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \ 623 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) 624 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 625 __be32 daddr, __be16 dport, int dif) 626 { 627 struct sock *sk; 628 629 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 630 dif, &udp_table, NULL); 631 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt)) 632 sk = NULL; 633 return sk; 634 } 635 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 636 #endif 637 638 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 639 __be16 loc_port, __be32 loc_addr, 640 __be16 rmt_port, __be32 rmt_addr, 641 int dif, unsigned short hnum) 642 { 643 struct inet_sock *inet = inet_sk(sk); 644 645 if (!net_eq(sock_net(sk), net) || 646 udp_sk(sk)->udp_port_hash != hnum || 647 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 648 (inet->inet_dport != rmt_port && inet->inet_dport) || 649 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 650 ipv6_only_sock(sk) || 651 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 652 return false; 653 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 654 return false; 655 return true; 656 } 657 658 /* 659 * This routine is called by the ICMP module when it gets some 660 * sort of error condition. If err < 0 then the socket should 661 * be closed and the error returned to the user. If err > 0 662 * it's just the icmp type << 8 | icmp code. 663 * Header points to the ip header of the error packet. We move 664 * on past this. Then (as it used to claim before adjustment) 665 * header points to the first 8 bytes of the udp header. We need 666 * to find the appropriate port. 667 */ 668 669 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 670 { 671 struct inet_sock *inet; 672 const struct iphdr *iph = (const struct iphdr *)skb->data; 673 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 674 const int type = icmp_hdr(skb)->type; 675 const int code = icmp_hdr(skb)->code; 676 struct sock *sk; 677 int harderr; 678 int err; 679 struct net *net = dev_net(skb->dev); 680 681 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 682 iph->saddr, uh->source, skb->dev->ifindex, udptable, 683 NULL); 684 if (!sk) { 685 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 686 return; /* No socket for error */ 687 } 688 689 err = 0; 690 harderr = 0; 691 inet = inet_sk(sk); 692 693 switch (type) { 694 default: 695 case ICMP_TIME_EXCEEDED: 696 err = EHOSTUNREACH; 697 break; 698 case ICMP_SOURCE_QUENCH: 699 goto out; 700 case ICMP_PARAMETERPROB: 701 err = EPROTO; 702 harderr = 1; 703 break; 704 case ICMP_DEST_UNREACH: 705 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 706 ipv4_sk_update_pmtu(skb, sk, info); 707 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 708 err = EMSGSIZE; 709 harderr = 1; 710 break; 711 } 712 goto out; 713 } 714 err = EHOSTUNREACH; 715 if (code <= NR_ICMP_UNREACH) { 716 harderr = icmp_err_convert[code].fatal; 717 err = icmp_err_convert[code].errno; 718 } 719 break; 720 case ICMP_REDIRECT: 721 ipv4_sk_redirect(skb, sk); 722 goto out; 723 } 724 725 /* 726 * RFC1122: OK. Passes ICMP errors back to application, as per 727 * 4.1.3.3. 728 */ 729 if (!inet->recverr) { 730 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 731 goto out; 732 } else 733 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 734 735 sk->sk_err = err; 736 sk->sk_error_report(sk); 737 out: 738 return; 739 } 740 741 void udp_err(struct sk_buff *skb, u32 info) 742 { 743 __udp4_lib_err(skb, info, &udp_table); 744 } 745 746 /* 747 * Throw away all pending data and cancel the corking. Socket is locked. 748 */ 749 void udp_flush_pending_frames(struct sock *sk) 750 { 751 struct udp_sock *up = udp_sk(sk); 752 753 if (up->pending) { 754 up->len = 0; 755 up->pending = 0; 756 ip_flush_pending_frames(sk); 757 } 758 } 759 EXPORT_SYMBOL(udp_flush_pending_frames); 760 761 /** 762 * udp4_hwcsum - handle outgoing HW checksumming 763 * @skb: sk_buff containing the filled-in UDP header 764 * (checksum field must be zeroed out) 765 * @src: source IP address 766 * @dst: destination IP address 767 */ 768 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 769 { 770 struct udphdr *uh = udp_hdr(skb); 771 int offset = skb_transport_offset(skb); 772 int len = skb->len - offset; 773 int hlen = len; 774 __wsum csum = 0; 775 776 if (!skb_has_frag_list(skb)) { 777 /* 778 * Only one fragment on the socket. 779 */ 780 skb->csum_start = skb_transport_header(skb) - skb->head; 781 skb->csum_offset = offsetof(struct udphdr, check); 782 uh->check = ~csum_tcpudp_magic(src, dst, len, 783 IPPROTO_UDP, 0); 784 } else { 785 struct sk_buff *frags; 786 787 /* 788 * HW-checksum won't work as there are two or more 789 * fragments on the socket so that all csums of sk_buffs 790 * should be together 791 */ 792 skb_walk_frags(skb, frags) { 793 csum = csum_add(csum, frags->csum); 794 hlen -= frags->len; 795 } 796 797 csum = skb_checksum(skb, offset, hlen, csum); 798 skb->ip_summed = CHECKSUM_NONE; 799 800 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 801 if (uh->check == 0) 802 uh->check = CSUM_MANGLED_0; 803 } 804 } 805 EXPORT_SYMBOL_GPL(udp4_hwcsum); 806 807 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 808 * for the simple case like when setting the checksum for a UDP tunnel. 809 */ 810 void udp_set_csum(bool nocheck, struct sk_buff *skb, 811 __be32 saddr, __be32 daddr, int len) 812 { 813 struct udphdr *uh = udp_hdr(skb); 814 815 if (nocheck) { 816 uh->check = 0; 817 } else if (skb_is_gso(skb)) { 818 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 819 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 820 uh->check = 0; 821 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 822 if (uh->check == 0) 823 uh->check = CSUM_MANGLED_0; 824 } else { 825 skb->ip_summed = CHECKSUM_PARTIAL; 826 skb->csum_start = skb_transport_header(skb) - skb->head; 827 skb->csum_offset = offsetof(struct udphdr, check); 828 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 829 } 830 } 831 EXPORT_SYMBOL(udp_set_csum); 832 833 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 834 { 835 struct sock *sk = skb->sk; 836 struct inet_sock *inet = inet_sk(sk); 837 struct udphdr *uh; 838 int err = 0; 839 int is_udplite = IS_UDPLITE(sk); 840 int offset = skb_transport_offset(skb); 841 int len = skb->len - offset; 842 __wsum csum = 0; 843 844 /* 845 * Create a UDP header 846 */ 847 uh = udp_hdr(skb); 848 uh->source = inet->inet_sport; 849 uh->dest = fl4->fl4_dport; 850 uh->len = htons(len); 851 uh->check = 0; 852 853 if (is_udplite) /* UDP-Lite */ 854 csum = udplite_csum(skb); 855 856 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 857 858 skb->ip_summed = CHECKSUM_NONE; 859 goto send; 860 861 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 862 863 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 864 goto send; 865 866 } else 867 csum = udp_csum(skb); 868 869 /* add protocol-dependent pseudo-header */ 870 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 871 sk->sk_protocol, csum); 872 if (uh->check == 0) 873 uh->check = CSUM_MANGLED_0; 874 875 send: 876 err = ip_send_skb(sock_net(sk), skb); 877 if (err) { 878 if (err == -ENOBUFS && !inet->recverr) { 879 UDP_INC_STATS(sock_net(sk), 880 UDP_MIB_SNDBUFERRORS, is_udplite); 881 err = 0; 882 } 883 } else 884 UDP_INC_STATS(sock_net(sk), 885 UDP_MIB_OUTDATAGRAMS, is_udplite); 886 return err; 887 } 888 889 /* 890 * Push out all pending data as one UDP datagram. Socket is locked. 891 */ 892 int udp_push_pending_frames(struct sock *sk) 893 { 894 struct udp_sock *up = udp_sk(sk); 895 struct inet_sock *inet = inet_sk(sk); 896 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 897 struct sk_buff *skb; 898 int err = 0; 899 900 skb = ip_finish_skb(sk, fl4); 901 if (!skb) 902 goto out; 903 904 err = udp_send_skb(skb, fl4); 905 906 out: 907 up->len = 0; 908 up->pending = 0; 909 return err; 910 } 911 EXPORT_SYMBOL(udp_push_pending_frames); 912 913 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 914 { 915 struct inet_sock *inet = inet_sk(sk); 916 struct udp_sock *up = udp_sk(sk); 917 struct flowi4 fl4_stack; 918 struct flowi4 *fl4; 919 int ulen = len; 920 struct ipcm_cookie ipc; 921 struct rtable *rt = NULL; 922 int free = 0; 923 int connected = 0; 924 __be32 daddr, faddr, saddr; 925 __be16 dport; 926 u8 tos; 927 int err, is_udplite = IS_UDPLITE(sk); 928 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 929 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 930 struct sk_buff *skb; 931 struct ip_options_data opt_copy; 932 933 if (len > 0xFFFF) 934 return -EMSGSIZE; 935 936 /* 937 * Check the flags. 938 */ 939 940 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 941 return -EOPNOTSUPP; 942 943 ipc.opt = NULL; 944 ipc.tx_flags = 0; 945 ipc.ttl = 0; 946 ipc.tos = -1; 947 948 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 949 950 fl4 = &inet->cork.fl.u.ip4; 951 if (up->pending) { 952 /* 953 * There are pending frames. 954 * The socket lock must be held while it's corked. 955 */ 956 lock_sock(sk); 957 if (likely(up->pending)) { 958 if (unlikely(up->pending != AF_INET)) { 959 release_sock(sk); 960 return -EINVAL; 961 } 962 goto do_append_data; 963 } 964 release_sock(sk); 965 } 966 ulen += sizeof(struct udphdr); 967 968 /* 969 * Get and verify the address. 970 */ 971 if (msg->msg_name) { 972 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 973 if (msg->msg_namelen < sizeof(*usin)) 974 return -EINVAL; 975 if (usin->sin_family != AF_INET) { 976 if (usin->sin_family != AF_UNSPEC) 977 return -EAFNOSUPPORT; 978 } 979 980 daddr = usin->sin_addr.s_addr; 981 dport = usin->sin_port; 982 if (dport == 0) 983 return -EINVAL; 984 } else { 985 if (sk->sk_state != TCP_ESTABLISHED) 986 return -EDESTADDRREQ; 987 daddr = inet->inet_daddr; 988 dport = inet->inet_dport; 989 /* Open fast path for connected socket. 990 Route will not be used, if at least one option is set. 991 */ 992 connected = 1; 993 } 994 995 ipc.sockc.tsflags = sk->sk_tsflags; 996 ipc.addr = inet->inet_saddr; 997 ipc.oif = sk->sk_bound_dev_if; 998 999 if (msg->msg_controllen) { 1000 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6); 1001 if (unlikely(err)) { 1002 kfree(ipc.opt); 1003 return err; 1004 } 1005 if (ipc.opt) 1006 free = 1; 1007 connected = 0; 1008 } 1009 if (!ipc.opt) { 1010 struct ip_options_rcu *inet_opt; 1011 1012 rcu_read_lock(); 1013 inet_opt = rcu_dereference(inet->inet_opt); 1014 if (inet_opt) { 1015 memcpy(&opt_copy, inet_opt, 1016 sizeof(*inet_opt) + inet_opt->opt.optlen); 1017 ipc.opt = &opt_copy.opt; 1018 } 1019 rcu_read_unlock(); 1020 } 1021 1022 saddr = ipc.addr; 1023 ipc.addr = faddr = daddr; 1024 1025 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags); 1026 1027 if (ipc.opt && ipc.opt->opt.srr) { 1028 if (!daddr) 1029 return -EINVAL; 1030 faddr = ipc.opt->opt.faddr; 1031 connected = 0; 1032 } 1033 tos = get_rttos(&ipc, inet); 1034 if (sock_flag(sk, SOCK_LOCALROUTE) || 1035 (msg->msg_flags & MSG_DONTROUTE) || 1036 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1037 tos |= RTO_ONLINK; 1038 connected = 0; 1039 } 1040 1041 if (ipv4_is_multicast(daddr)) { 1042 if (!ipc.oif) 1043 ipc.oif = inet->mc_index; 1044 if (!saddr) 1045 saddr = inet->mc_addr; 1046 connected = 0; 1047 } else if (!ipc.oif) 1048 ipc.oif = inet->uc_index; 1049 1050 if (connected) 1051 rt = (struct rtable *)sk_dst_check(sk, 0); 1052 1053 if (!rt) { 1054 struct net *net = sock_net(sk); 1055 __u8 flow_flags = inet_sk_flowi_flags(sk); 1056 1057 fl4 = &fl4_stack; 1058 1059 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1060 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1061 flow_flags, 1062 faddr, saddr, dport, inet->inet_sport); 1063 1064 if (!saddr && ipc.oif) { 1065 err = l3mdev_get_saddr(net, ipc.oif, fl4); 1066 if (err < 0) 1067 goto out; 1068 } 1069 1070 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1071 rt = ip_route_output_flow(net, fl4, sk); 1072 if (IS_ERR(rt)) { 1073 err = PTR_ERR(rt); 1074 rt = NULL; 1075 if (err == -ENETUNREACH) 1076 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1077 goto out; 1078 } 1079 1080 err = -EACCES; 1081 if ((rt->rt_flags & RTCF_BROADCAST) && 1082 !sock_flag(sk, SOCK_BROADCAST)) 1083 goto out; 1084 if (connected) 1085 sk_dst_set(sk, dst_clone(&rt->dst)); 1086 } 1087 1088 if (msg->msg_flags&MSG_CONFIRM) 1089 goto do_confirm; 1090 back_from_confirm: 1091 1092 saddr = fl4->saddr; 1093 if (!ipc.addr) 1094 daddr = ipc.addr = fl4->daddr; 1095 1096 /* Lockless fast path for the non-corking case. */ 1097 if (!corkreq) { 1098 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1099 sizeof(struct udphdr), &ipc, &rt, 1100 msg->msg_flags); 1101 err = PTR_ERR(skb); 1102 if (!IS_ERR_OR_NULL(skb)) 1103 err = udp_send_skb(skb, fl4); 1104 goto out; 1105 } 1106 1107 lock_sock(sk); 1108 if (unlikely(up->pending)) { 1109 /* The socket is already corked while preparing it. */ 1110 /* ... which is an evident application bug. --ANK */ 1111 release_sock(sk); 1112 1113 net_dbg_ratelimited("cork app bug 2\n"); 1114 err = -EINVAL; 1115 goto out; 1116 } 1117 /* 1118 * Now cork the socket to pend data. 1119 */ 1120 fl4 = &inet->cork.fl.u.ip4; 1121 fl4->daddr = daddr; 1122 fl4->saddr = saddr; 1123 fl4->fl4_dport = dport; 1124 fl4->fl4_sport = inet->inet_sport; 1125 up->pending = AF_INET; 1126 1127 do_append_data: 1128 up->len += ulen; 1129 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1130 sizeof(struct udphdr), &ipc, &rt, 1131 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1132 if (err) 1133 udp_flush_pending_frames(sk); 1134 else if (!corkreq) 1135 err = udp_push_pending_frames(sk); 1136 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1137 up->pending = 0; 1138 release_sock(sk); 1139 1140 out: 1141 ip_rt_put(rt); 1142 if (free) 1143 kfree(ipc.opt); 1144 if (!err) 1145 return len; 1146 /* 1147 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1148 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1149 * we don't have a good statistic (IpOutDiscards but it can be too many 1150 * things). We could add another new stat but at least for now that 1151 * seems like overkill. 1152 */ 1153 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1154 UDP_INC_STATS(sock_net(sk), 1155 UDP_MIB_SNDBUFERRORS, is_udplite); 1156 } 1157 return err; 1158 1159 do_confirm: 1160 dst_confirm(&rt->dst); 1161 if (!(msg->msg_flags&MSG_PROBE) || len) 1162 goto back_from_confirm; 1163 err = 0; 1164 goto out; 1165 } 1166 EXPORT_SYMBOL(udp_sendmsg); 1167 1168 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1169 size_t size, int flags) 1170 { 1171 struct inet_sock *inet = inet_sk(sk); 1172 struct udp_sock *up = udp_sk(sk); 1173 int ret; 1174 1175 if (flags & MSG_SENDPAGE_NOTLAST) 1176 flags |= MSG_MORE; 1177 1178 if (!up->pending) { 1179 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1180 1181 /* Call udp_sendmsg to specify destination address which 1182 * sendpage interface can't pass. 1183 * This will succeed only when the socket is connected. 1184 */ 1185 ret = udp_sendmsg(sk, &msg, 0); 1186 if (ret < 0) 1187 return ret; 1188 } 1189 1190 lock_sock(sk); 1191 1192 if (unlikely(!up->pending)) { 1193 release_sock(sk); 1194 1195 net_dbg_ratelimited("udp cork app bug 3\n"); 1196 return -EINVAL; 1197 } 1198 1199 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1200 page, offset, size, flags); 1201 if (ret == -EOPNOTSUPP) { 1202 release_sock(sk); 1203 return sock_no_sendpage(sk->sk_socket, page, offset, 1204 size, flags); 1205 } 1206 if (ret < 0) { 1207 udp_flush_pending_frames(sk); 1208 goto out; 1209 } 1210 1211 up->len += size; 1212 if (!(up->corkflag || (flags&MSG_MORE))) 1213 ret = udp_push_pending_frames(sk); 1214 if (!ret) 1215 ret = size; 1216 out: 1217 release_sock(sk); 1218 return ret; 1219 } 1220 1221 /** 1222 * first_packet_length - return length of first packet in receive queue 1223 * @sk: socket 1224 * 1225 * Drops all bad checksum frames, until a valid one is found. 1226 * Returns the length of found skb, or 0 if none is found. 1227 */ 1228 static unsigned int first_packet_length(struct sock *sk) 1229 { 1230 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1231 struct sk_buff *skb; 1232 unsigned int res; 1233 1234 __skb_queue_head_init(&list_kill); 1235 1236 spin_lock_bh(&rcvq->lock); 1237 while ((skb = skb_peek(rcvq)) != NULL && 1238 udp_lib_checksum_complete(skb)) { 1239 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1240 IS_UDPLITE(sk)); 1241 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1242 IS_UDPLITE(sk)); 1243 atomic_inc(&sk->sk_drops); 1244 __skb_unlink(skb, rcvq); 1245 __skb_queue_tail(&list_kill, skb); 1246 } 1247 res = skb ? skb->len : 0; 1248 spin_unlock_bh(&rcvq->lock); 1249 1250 if (!skb_queue_empty(&list_kill)) { 1251 bool slow = lock_sock_fast(sk); 1252 1253 __skb_queue_purge(&list_kill); 1254 sk_mem_reclaim_partial(sk); 1255 unlock_sock_fast(sk, slow); 1256 } 1257 return res; 1258 } 1259 1260 /* 1261 * IOCTL requests applicable to the UDP protocol 1262 */ 1263 1264 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1265 { 1266 switch (cmd) { 1267 case SIOCOUTQ: 1268 { 1269 int amount = sk_wmem_alloc_get(sk); 1270 1271 return put_user(amount, (int __user *)arg); 1272 } 1273 1274 case SIOCINQ: 1275 { 1276 unsigned int amount = first_packet_length(sk); 1277 1278 return put_user(amount, (int __user *)arg); 1279 } 1280 1281 default: 1282 return -ENOIOCTLCMD; 1283 } 1284 1285 return 0; 1286 } 1287 EXPORT_SYMBOL(udp_ioctl); 1288 1289 /* 1290 * This should be easy, if there is something there we 1291 * return it, otherwise we block. 1292 */ 1293 1294 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1295 int flags, int *addr_len) 1296 { 1297 struct inet_sock *inet = inet_sk(sk); 1298 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1299 struct sk_buff *skb; 1300 unsigned int ulen, copied; 1301 int peeked, peeking, off; 1302 int err; 1303 int is_udplite = IS_UDPLITE(sk); 1304 bool checksum_valid = false; 1305 bool slow; 1306 1307 if (flags & MSG_ERRQUEUE) 1308 return ip_recv_error(sk, msg, len, addr_len); 1309 1310 try_again: 1311 peeking = off = sk_peek_offset(sk, flags); 1312 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1313 &peeked, &off, &err); 1314 if (!skb) 1315 return err; 1316 1317 ulen = skb->len; 1318 copied = len; 1319 if (copied > ulen - off) 1320 copied = ulen - off; 1321 else if (copied < ulen) 1322 msg->msg_flags |= MSG_TRUNC; 1323 1324 /* 1325 * If checksum is needed at all, try to do it while copying the 1326 * data. If the data is truncated, or if we only want a partial 1327 * coverage checksum (UDP-Lite), do it before the copy. 1328 */ 1329 1330 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov || peeking) { 1331 checksum_valid = !udp_lib_checksum_complete(skb); 1332 if (!checksum_valid) 1333 goto csum_copy_err; 1334 } 1335 1336 if (checksum_valid || skb_csum_unnecessary(skb)) 1337 err = skb_copy_datagram_msg(skb, off, msg, copied); 1338 else { 1339 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1340 1341 if (err == -EINVAL) 1342 goto csum_copy_err; 1343 } 1344 1345 if (unlikely(err)) { 1346 trace_kfree_skb(skb, udp_recvmsg); 1347 if (!peeked) { 1348 atomic_inc(&sk->sk_drops); 1349 UDP_INC_STATS(sock_net(sk), 1350 UDP_MIB_INERRORS, is_udplite); 1351 } 1352 skb_free_datagram_locked(sk, skb); 1353 return err; 1354 } 1355 1356 if (!peeked) 1357 UDP_INC_STATS(sock_net(sk), 1358 UDP_MIB_INDATAGRAMS, is_udplite); 1359 1360 sock_recv_ts_and_drops(msg, sk, skb); 1361 1362 /* Copy the address. */ 1363 if (sin) { 1364 sin->sin_family = AF_INET; 1365 sin->sin_port = udp_hdr(skb)->source; 1366 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1367 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1368 *addr_len = sizeof(*sin); 1369 } 1370 if (inet->cmsg_flags) 1371 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr) + off); 1372 1373 err = copied; 1374 if (flags & MSG_TRUNC) 1375 err = ulen; 1376 1377 __skb_free_datagram_locked(sk, skb, peeking ? -err : err); 1378 return err; 1379 1380 csum_copy_err: 1381 slow = lock_sock_fast(sk); 1382 if (!skb_kill_datagram(sk, skb, flags)) { 1383 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1384 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1385 } 1386 unlock_sock_fast(sk, slow); 1387 1388 /* starting over for a new packet, but check if we need to yield */ 1389 cond_resched(); 1390 msg->msg_flags &= ~MSG_TRUNC; 1391 goto try_again; 1392 } 1393 1394 int udp_disconnect(struct sock *sk, int flags) 1395 { 1396 struct inet_sock *inet = inet_sk(sk); 1397 /* 1398 * 1003.1g - break association. 1399 */ 1400 1401 sk->sk_state = TCP_CLOSE; 1402 inet->inet_daddr = 0; 1403 inet->inet_dport = 0; 1404 sock_rps_reset_rxhash(sk); 1405 sk->sk_bound_dev_if = 0; 1406 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1407 inet_reset_saddr(sk); 1408 1409 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1410 sk->sk_prot->unhash(sk); 1411 inet->inet_sport = 0; 1412 } 1413 sk_dst_reset(sk); 1414 return 0; 1415 } 1416 EXPORT_SYMBOL(udp_disconnect); 1417 1418 void udp_lib_unhash(struct sock *sk) 1419 { 1420 if (sk_hashed(sk)) { 1421 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1422 struct udp_hslot *hslot, *hslot2; 1423 1424 hslot = udp_hashslot(udptable, sock_net(sk), 1425 udp_sk(sk)->udp_port_hash); 1426 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1427 1428 spin_lock_bh(&hslot->lock); 1429 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1430 reuseport_detach_sock(sk); 1431 if (sk_del_node_init_rcu(sk)) { 1432 hslot->count--; 1433 inet_sk(sk)->inet_num = 0; 1434 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1435 1436 spin_lock(&hslot2->lock); 1437 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1438 hslot2->count--; 1439 spin_unlock(&hslot2->lock); 1440 } 1441 spin_unlock_bh(&hslot->lock); 1442 } 1443 } 1444 EXPORT_SYMBOL(udp_lib_unhash); 1445 1446 /* 1447 * inet_rcv_saddr was changed, we must rehash secondary hash 1448 */ 1449 void udp_lib_rehash(struct sock *sk, u16 newhash) 1450 { 1451 if (sk_hashed(sk)) { 1452 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1453 struct udp_hslot *hslot, *hslot2, *nhslot2; 1454 1455 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1456 nhslot2 = udp_hashslot2(udptable, newhash); 1457 udp_sk(sk)->udp_portaddr_hash = newhash; 1458 1459 if (hslot2 != nhslot2 || 1460 rcu_access_pointer(sk->sk_reuseport_cb)) { 1461 hslot = udp_hashslot(udptable, sock_net(sk), 1462 udp_sk(sk)->udp_port_hash); 1463 /* we must lock primary chain too */ 1464 spin_lock_bh(&hslot->lock); 1465 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1466 reuseport_detach_sock(sk); 1467 1468 if (hslot2 != nhslot2) { 1469 spin_lock(&hslot2->lock); 1470 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1471 hslot2->count--; 1472 spin_unlock(&hslot2->lock); 1473 1474 spin_lock(&nhslot2->lock); 1475 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1476 &nhslot2->head); 1477 nhslot2->count++; 1478 spin_unlock(&nhslot2->lock); 1479 } 1480 1481 spin_unlock_bh(&hslot->lock); 1482 } 1483 } 1484 } 1485 EXPORT_SYMBOL(udp_lib_rehash); 1486 1487 static void udp_v4_rehash(struct sock *sk) 1488 { 1489 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1490 inet_sk(sk)->inet_rcv_saddr, 1491 inet_sk(sk)->inet_num); 1492 udp_lib_rehash(sk, new_hash); 1493 } 1494 1495 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1496 { 1497 int rc; 1498 1499 if (inet_sk(sk)->inet_daddr) { 1500 sock_rps_save_rxhash(sk, skb); 1501 sk_mark_napi_id(sk, skb); 1502 sk_incoming_cpu_update(sk); 1503 } 1504 1505 rc = __sock_queue_rcv_skb(sk, skb); 1506 if (rc < 0) { 1507 int is_udplite = IS_UDPLITE(sk); 1508 1509 /* Note that an ENOMEM error is charged twice */ 1510 if (rc == -ENOMEM) 1511 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1512 is_udplite); 1513 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1514 kfree_skb(skb); 1515 trace_udp_fail_queue_rcv_skb(rc, sk); 1516 return -1; 1517 } 1518 1519 return 0; 1520 1521 } 1522 1523 static struct static_key udp_encap_needed __read_mostly; 1524 void udp_encap_enable(void) 1525 { 1526 if (!static_key_enabled(&udp_encap_needed)) 1527 static_key_slow_inc(&udp_encap_needed); 1528 } 1529 EXPORT_SYMBOL(udp_encap_enable); 1530 1531 /* returns: 1532 * -1: error 1533 * 0: success 1534 * >0: "udp encap" protocol resubmission 1535 * 1536 * Note that in the success and error cases, the skb is assumed to 1537 * have either been requeued or freed. 1538 */ 1539 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1540 { 1541 struct udp_sock *up = udp_sk(sk); 1542 int rc; 1543 int is_udplite = IS_UDPLITE(sk); 1544 1545 /* 1546 * Charge it to the socket, dropping if the queue is full. 1547 */ 1548 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1549 goto drop; 1550 nf_reset(skb); 1551 1552 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1553 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1554 1555 /* 1556 * This is an encapsulation socket so pass the skb to 1557 * the socket's udp_encap_rcv() hook. Otherwise, just 1558 * fall through and pass this up the UDP socket. 1559 * up->encap_rcv() returns the following value: 1560 * =0 if skb was successfully passed to the encap 1561 * handler or was discarded by it. 1562 * >0 if skb should be passed on to UDP. 1563 * <0 if skb should be resubmitted as proto -N 1564 */ 1565 1566 /* if we're overly short, let UDP handle it */ 1567 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1568 if (encap_rcv) { 1569 int ret; 1570 1571 /* Verify checksum before giving to encap */ 1572 if (udp_lib_checksum_complete(skb)) 1573 goto csum_error; 1574 1575 ret = encap_rcv(sk, skb); 1576 if (ret <= 0) { 1577 __UDP_INC_STATS(sock_net(sk), 1578 UDP_MIB_INDATAGRAMS, 1579 is_udplite); 1580 return -ret; 1581 } 1582 } 1583 1584 /* FALLTHROUGH -- it's a UDP Packet */ 1585 } 1586 1587 /* 1588 * UDP-Lite specific tests, ignored on UDP sockets 1589 */ 1590 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1591 1592 /* 1593 * MIB statistics other than incrementing the error count are 1594 * disabled for the following two types of errors: these depend 1595 * on the application settings, not on the functioning of the 1596 * protocol stack as such. 1597 * 1598 * RFC 3828 here recommends (sec 3.3): "There should also be a 1599 * way ... to ... at least let the receiving application block 1600 * delivery of packets with coverage values less than a value 1601 * provided by the application." 1602 */ 1603 if (up->pcrlen == 0) { /* full coverage was set */ 1604 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1605 UDP_SKB_CB(skb)->cscov, skb->len); 1606 goto drop; 1607 } 1608 /* The next case involves violating the min. coverage requested 1609 * by the receiver. This is subtle: if receiver wants x and x is 1610 * greater than the buffersize/MTU then receiver will complain 1611 * that it wants x while sender emits packets of smaller size y. 1612 * Therefore the above ...()->partial_cov statement is essential. 1613 */ 1614 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1615 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1616 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1617 goto drop; 1618 } 1619 } 1620 1621 if (rcu_access_pointer(sk->sk_filter)) { 1622 if (udp_lib_checksum_complete(skb)) 1623 goto csum_error; 1624 if (sk_filter(sk, skb)) 1625 goto drop; 1626 } 1627 1628 udp_csum_pull_header(skb); 1629 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1630 __UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1631 is_udplite); 1632 goto drop; 1633 } 1634 1635 rc = 0; 1636 1637 ipv4_pktinfo_prepare(sk, skb); 1638 bh_lock_sock(sk); 1639 if (!sock_owned_by_user(sk)) 1640 rc = __udp_queue_rcv_skb(sk, skb); 1641 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1642 bh_unlock_sock(sk); 1643 goto drop; 1644 } 1645 bh_unlock_sock(sk); 1646 1647 return rc; 1648 1649 csum_error: 1650 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1651 drop: 1652 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1653 atomic_inc(&sk->sk_drops); 1654 kfree_skb(skb); 1655 return -1; 1656 } 1657 1658 /* For TCP sockets, sk_rx_dst is protected by socket lock 1659 * For UDP, we use xchg() to guard against concurrent changes. 1660 */ 1661 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1662 { 1663 struct dst_entry *old; 1664 1665 dst_hold(dst); 1666 old = xchg(&sk->sk_rx_dst, dst); 1667 dst_release(old); 1668 } 1669 1670 /* 1671 * Multicasts and broadcasts go to each listener. 1672 * 1673 * Note: called only from the BH handler context. 1674 */ 1675 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1676 struct udphdr *uh, 1677 __be32 saddr, __be32 daddr, 1678 struct udp_table *udptable, 1679 int proto) 1680 { 1681 struct sock *sk, *first = NULL; 1682 unsigned short hnum = ntohs(uh->dest); 1683 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1684 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1685 unsigned int offset = offsetof(typeof(*sk), sk_node); 1686 int dif = skb->dev->ifindex; 1687 struct hlist_node *node; 1688 struct sk_buff *nskb; 1689 1690 if (use_hash2) { 1691 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1692 udp_table.mask; 1693 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask; 1694 start_lookup: 1695 hslot = &udp_table.hash2[hash2]; 1696 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1697 } 1698 1699 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 1700 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 1701 uh->source, saddr, dif, hnum)) 1702 continue; 1703 1704 if (!first) { 1705 first = sk; 1706 continue; 1707 } 1708 nskb = skb_clone(skb, GFP_ATOMIC); 1709 1710 if (unlikely(!nskb)) { 1711 atomic_inc(&sk->sk_drops); 1712 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 1713 IS_UDPLITE(sk)); 1714 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 1715 IS_UDPLITE(sk)); 1716 continue; 1717 } 1718 if (udp_queue_rcv_skb(sk, nskb) > 0) 1719 consume_skb(nskb); 1720 } 1721 1722 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1723 if (use_hash2 && hash2 != hash2_any) { 1724 hash2 = hash2_any; 1725 goto start_lookup; 1726 } 1727 1728 if (first) { 1729 if (udp_queue_rcv_skb(first, skb) > 0) 1730 consume_skb(skb); 1731 } else { 1732 kfree_skb(skb); 1733 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 1734 proto == IPPROTO_UDPLITE); 1735 } 1736 return 0; 1737 } 1738 1739 /* Initialize UDP checksum. If exited with zero value (success), 1740 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1741 * Otherwise, csum completion requires chacksumming packet body, 1742 * including udp header and folding it to skb->csum. 1743 */ 1744 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1745 int proto) 1746 { 1747 int err; 1748 1749 UDP_SKB_CB(skb)->partial_cov = 0; 1750 UDP_SKB_CB(skb)->cscov = skb->len; 1751 1752 if (proto == IPPROTO_UDPLITE) { 1753 err = udplite_checksum_init(skb, uh); 1754 if (err) 1755 return err; 1756 } 1757 1758 return skb_checksum_init_zero_check(skb, proto, uh->check, 1759 inet_compute_pseudo); 1760 } 1761 1762 /* 1763 * All we need to do is get the socket, and then do a checksum. 1764 */ 1765 1766 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1767 int proto) 1768 { 1769 struct sock *sk; 1770 struct udphdr *uh; 1771 unsigned short ulen; 1772 struct rtable *rt = skb_rtable(skb); 1773 __be32 saddr, daddr; 1774 struct net *net = dev_net(skb->dev); 1775 1776 /* 1777 * Validate the packet. 1778 */ 1779 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1780 goto drop; /* No space for header. */ 1781 1782 uh = udp_hdr(skb); 1783 ulen = ntohs(uh->len); 1784 saddr = ip_hdr(skb)->saddr; 1785 daddr = ip_hdr(skb)->daddr; 1786 1787 if (ulen > skb->len) 1788 goto short_packet; 1789 1790 if (proto == IPPROTO_UDP) { 1791 /* UDP validates ulen. */ 1792 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1793 goto short_packet; 1794 uh = udp_hdr(skb); 1795 } 1796 1797 if (udp4_csum_init(skb, uh, proto)) 1798 goto csum_error; 1799 1800 sk = skb_steal_sock(skb); 1801 if (sk) { 1802 struct dst_entry *dst = skb_dst(skb); 1803 int ret; 1804 1805 if (unlikely(sk->sk_rx_dst != dst)) 1806 udp_sk_rx_dst_set(sk, dst); 1807 1808 ret = udp_queue_rcv_skb(sk, skb); 1809 sock_put(sk); 1810 /* a return value > 0 means to resubmit the input, but 1811 * it wants the return to be -protocol, or 0 1812 */ 1813 if (ret > 0) 1814 return -ret; 1815 return 0; 1816 } 1817 1818 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1819 return __udp4_lib_mcast_deliver(net, skb, uh, 1820 saddr, daddr, udptable, proto); 1821 1822 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1823 if (sk) { 1824 int ret; 1825 1826 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1827 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1828 inet_compute_pseudo); 1829 1830 ret = udp_queue_rcv_skb(sk, skb); 1831 1832 /* a return value > 0 means to resubmit the input, but 1833 * it wants the return to be -protocol, or 0 1834 */ 1835 if (ret > 0) 1836 return -ret; 1837 return 0; 1838 } 1839 1840 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1841 goto drop; 1842 nf_reset(skb); 1843 1844 /* No socket. Drop packet silently, if checksum is wrong */ 1845 if (udp_lib_checksum_complete(skb)) 1846 goto csum_error; 1847 1848 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1849 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1850 1851 /* 1852 * Hmm. We got an UDP packet to a port to which we 1853 * don't wanna listen. Ignore it. 1854 */ 1855 kfree_skb(skb); 1856 return 0; 1857 1858 short_packet: 1859 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1860 proto == IPPROTO_UDPLITE ? "Lite" : "", 1861 &saddr, ntohs(uh->source), 1862 ulen, skb->len, 1863 &daddr, ntohs(uh->dest)); 1864 goto drop; 1865 1866 csum_error: 1867 /* 1868 * RFC1122: OK. Discards the bad packet silently (as far as 1869 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1870 */ 1871 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1872 proto == IPPROTO_UDPLITE ? "Lite" : "", 1873 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1874 ulen); 1875 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1876 drop: 1877 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1878 kfree_skb(skb); 1879 return 0; 1880 } 1881 1882 /* We can only early demux multicast if there is a single matching socket. 1883 * If more than one socket found returns NULL 1884 */ 1885 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1886 __be16 loc_port, __be32 loc_addr, 1887 __be16 rmt_port, __be32 rmt_addr, 1888 int dif) 1889 { 1890 struct sock *sk, *result; 1891 unsigned short hnum = ntohs(loc_port); 1892 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 1893 struct udp_hslot *hslot = &udp_table.hash[slot]; 1894 1895 /* Do not bother scanning a too big list */ 1896 if (hslot->count > 10) 1897 return NULL; 1898 1899 result = NULL; 1900 sk_for_each_rcu(sk, &hslot->head) { 1901 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 1902 rmt_port, rmt_addr, dif, hnum)) { 1903 if (result) 1904 return NULL; 1905 result = sk; 1906 } 1907 } 1908 1909 return result; 1910 } 1911 1912 /* For unicast we should only early demux connected sockets or we can 1913 * break forwarding setups. The chains here can be long so only check 1914 * if the first socket is an exact match and if not move on. 1915 */ 1916 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1917 __be16 loc_port, __be32 loc_addr, 1918 __be16 rmt_port, __be32 rmt_addr, 1919 int dif) 1920 { 1921 unsigned short hnum = ntohs(loc_port); 1922 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1923 unsigned int slot2 = hash2 & udp_table.mask; 1924 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1925 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 1926 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1927 struct sock *sk; 1928 1929 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1930 if (INET_MATCH(sk, net, acookie, rmt_addr, 1931 loc_addr, ports, dif)) 1932 return sk; 1933 /* Only check first socket in chain */ 1934 break; 1935 } 1936 return NULL; 1937 } 1938 1939 void udp_v4_early_demux(struct sk_buff *skb) 1940 { 1941 struct net *net = dev_net(skb->dev); 1942 const struct iphdr *iph; 1943 const struct udphdr *uh; 1944 struct sock *sk = NULL; 1945 struct dst_entry *dst; 1946 int dif = skb->dev->ifindex; 1947 int ours; 1948 1949 /* validate the packet */ 1950 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1951 return; 1952 1953 iph = ip_hdr(skb); 1954 uh = udp_hdr(skb); 1955 1956 if (skb->pkt_type == PACKET_BROADCAST || 1957 skb->pkt_type == PACKET_MULTICAST) { 1958 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 1959 1960 if (!in_dev) 1961 return; 1962 1963 /* we are supposed to accept bcast packets */ 1964 if (skb->pkt_type == PACKET_MULTICAST) { 1965 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 1966 iph->protocol); 1967 if (!ours) 1968 return; 1969 } 1970 1971 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1972 uh->source, iph->saddr, dif); 1973 } else if (skb->pkt_type == PACKET_HOST) { 1974 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1975 uh->source, iph->saddr, dif); 1976 } 1977 1978 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2)) 1979 return; 1980 1981 skb->sk = sk; 1982 skb->destructor = sock_efree; 1983 dst = READ_ONCE(sk->sk_rx_dst); 1984 1985 if (dst) 1986 dst = dst_check(dst, 0); 1987 if (dst) { 1988 /* DST_NOCACHE can not be used without taking a reference */ 1989 if (dst->flags & DST_NOCACHE) { 1990 if (likely(atomic_inc_not_zero(&dst->__refcnt))) 1991 skb_dst_set(skb, dst); 1992 } else { 1993 skb_dst_set_noref(skb, dst); 1994 } 1995 } 1996 } 1997 1998 int udp_rcv(struct sk_buff *skb) 1999 { 2000 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2001 } 2002 2003 void udp_destroy_sock(struct sock *sk) 2004 { 2005 struct udp_sock *up = udp_sk(sk); 2006 bool slow = lock_sock_fast(sk); 2007 udp_flush_pending_frames(sk); 2008 unlock_sock_fast(sk, slow); 2009 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2010 void (*encap_destroy)(struct sock *sk); 2011 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2012 if (encap_destroy) 2013 encap_destroy(sk); 2014 } 2015 } 2016 2017 /* 2018 * Socket option code for UDP 2019 */ 2020 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2021 char __user *optval, unsigned int optlen, 2022 int (*push_pending_frames)(struct sock *)) 2023 { 2024 struct udp_sock *up = udp_sk(sk); 2025 int val, valbool; 2026 int err = 0; 2027 int is_udplite = IS_UDPLITE(sk); 2028 2029 if (optlen < sizeof(int)) 2030 return -EINVAL; 2031 2032 if (get_user(val, (int __user *)optval)) 2033 return -EFAULT; 2034 2035 valbool = val ? 1 : 0; 2036 2037 switch (optname) { 2038 case UDP_CORK: 2039 if (val != 0) { 2040 up->corkflag = 1; 2041 } else { 2042 up->corkflag = 0; 2043 lock_sock(sk); 2044 push_pending_frames(sk); 2045 release_sock(sk); 2046 } 2047 break; 2048 2049 case UDP_ENCAP: 2050 switch (val) { 2051 case 0: 2052 case UDP_ENCAP_ESPINUDP: 2053 case UDP_ENCAP_ESPINUDP_NON_IKE: 2054 up->encap_rcv = xfrm4_udp_encap_rcv; 2055 /* FALLTHROUGH */ 2056 case UDP_ENCAP_L2TPINUDP: 2057 up->encap_type = val; 2058 udp_encap_enable(); 2059 break; 2060 default: 2061 err = -ENOPROTOOPT; 2062 break; 2063 } 2064 break; 2065 2066 case UDP_NO_CHECK6_TX: 2067 up->no_check6_tx = valbool; 2068 break; 2069 2070 case UDP_NO_CHECK6_RX: 2071 up->no_check6_rx = valbool; 2072 break; 2073 2074 /* 2075 * UDP-Lite's partial checksum coverage (RFC 3828). 2076 */ 2077 /* The sender sets actual checksum coverage length via this option. 2078 * The case coverage > packet length is handled by send module. */ 2079 case UDPLITE_SEND_CSCOV: 2080 if (!is_udplite) /* Disable the option on UDP sockets */ 2081 return -ENOPROTOOPT; 2082 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2083 val = 8; 2084 else if (val > USHRT_MAX) 2085 val = USHRT_MAX; 2086 up->pcslen = val; 2087 up->pcflag |= UDPLITE_SEND_CC; 2088 break; 2089 2090 /* The receiver specifies a minimum checksum coverage value. To make 2091 * sense, this should be set to at least 8 (as done below). If zero is 2092 * used, this again means full checksum coverage. */ 2093 case UDPLITE_RECV_CSCOV: 2094 if (!is_udplite) /* Disable the option on UDP sockets */ 2095 return -ENOPROTOOPT; 2096 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2097 val = 8; 2098 else if (val > USHRT_MAX) 2099 val = USHRT_MAX; 2100 up->pcrlen = val; 2101 up->pcflag |= UDPLITE_RECV_CC; 2102 break; 2103 2104 default: 2105 err = -ENOPROTOOPT; 2106 break; 2107 } 2108 2109 return err; 2110 } 2111 EXPORT_SYMBOL(udp_lib_setsockopt); 2112 2113 int udp_setsockopt(struct sock *sk, int level, int optname, 2114 char __user *optval, unsigned int optlen) 2115 { 2116 if (level == SOL_UDP || level == SOL_UDPLITE) 2117 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2118 udp_push_pending_frames); 2119 return ip_setsockopt(sk, level, optname, optval, optlen); 2120 } 2121 2122 #ifdef CONFIG_COMPAT 2123 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2124 char __user *optval, unsigned int optlen) 2125 { 2126 if (level == SOL_UDP || level == SOL_UDPLITE) 2127 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2128 udp_push_pending_frames); 2129 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2130 } 2131 #endif 2132 2133 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2134 char __user *optval, int __user *optlen) 2135 { 2136 struct udp_sock *up = udp_sk(sk); 2137 int val, len; 2138 2139 if (get_user(len, optlen)) 2140 return -EFAULT; 2141 2142 len = min_t(unsigned int, len, sizeof(int)); 2143 2144 if (len < 0) 2145 return -EINVAL; 2146 2147 switch (optname) { 2148 case UDP_CORK: 2149 val = up->corkflag; 2150 break; 2151 2152 case UDP_ENCAP: 2153 val = up->encap_type; 2154 break; 2155 2156 case UDP_NO_CHECK6_TX: 2157 val = up->no_check6_tx; 2158 break; 2159 2160 case UDP_NO_CHECK6_RX: 2161 val = up->no_check6_rx; 2162 break; 2163 2164 /* The following two cannot be changed on UDP sockets, the return is 2165 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2166 case UDPLITE_SEND_CSCOV: 2167 val = up->pcslen; 2168 break; 2169 2170 case UDPLITE_RECV_CSCOV: 2171 val = up->pcrlen; 2172 break; 2173 2174 default: 2175 return -ENOPROTOOPT; 2176 } 2177 2178 if (put_user(len, optlen)) 2179 return -EFAULT; 2180 if (copy_to_user(optval, &val, len)) 2181 return -EFAULT; 2182 return 0; 2183 } 2184 EXPORT_SYMBOL(udp_lib_getsockopt); 2185 2186 int udp_getsockopt(struct sock *sk, int level, int optname, 2187 char __user *optval, int __user *optlen) 2188 { 2189 if (level == SOL_UDP || level == SOL_UDPLITE) 2190 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2191 return ip_getsockopt(sk, level, optname, optval, optlen); 2192 } 2193 2194 #ifdef CONFIG_COMPAT 2195 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2196 char __user *optval, int __user *optlen) 2197 { 2198 if (level == SOL_UDP || level == SOL_UDPLITE) 2199 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2200 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2201 } 2202 #endif 2203 /** 2204 * udp_poll - wait for a UDP event. 2205 * @file - file struct 2206 * @sock - socket 2207 * @wait - poll table 2208 * 2209 * This is same as datagram poll, except for the special case of 2210 * blocking sockets. If application is using a blocking fd 2211 * and a packet with checksum error is in the queue; 2212 * then it could get return from select indicating data available 2213 * but then block when reading it. Add special case code 2214 * to work around these arguably broken applications. 2215 */ 2216 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2217 { 2218 unsigned int mask = datagram_poll(file, sock, wait); 2219 struct sock *sk = sock->sk; 2220 2221 sock_rps_record_flow(sk); 2222 2223 /* Check for false positives due to checksum errors */ 2224 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2225 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2226 mask &= ~(POLLIN | POLLRDNORM); 2227 2228 return mask; 2229 2230 } 2231 EXPORT_SYMBOL(udp_poll); 2232 2233 struct proto udp_prot = { 2234 .name = "UDP", 2235 .owner = THIS_MODULE, 2236 .close = udp_lib_close, 2237 .connect = ip4_datagram_connect, 2238 .disconnect = udp_disconnect, 2239 .ioctl = udp_ioctl, 2240 .destroy = udp_destroy_sock, 2241 .setsockopt = udp_setsockopt, 2242 .getsockopt = udp_getsockopt, 2243 .sendmsg = udp_sendmsg, 2244 .recvmsg = udp_recvmsg, 2245 .sendpage = udp_sendpage, 2246 .backlog_rcv = __udp_queue_rcv_skb, 2247 .release_cb = ip4_datagram_release_cb, 2248 .hash = udp_lib_hash, 2249 .unhash = udp_lib_unhash, 2250 .rehash = udp_v4_rehash, 2251 .get_port = udp_v4_get_port, 2252 .memory_allocated = &udp_memory_allocated, 2253 .sysctl_mem = sysctl_udp_mem, 2254 .sysctl_wmem = &sysctl_udp_wmem_min, 2255 .sysctl_rmem = &sysctl_udp_rmem_min, 2256 .obj_size = sizeof(struct udp_sock), 2257 .slab_flags = SLAB_DESTROY_BY_RCU, 2258 .h.udp_table = &udp_table, 2259 #ifdef CONFIG_COMPAT 2260 .compat_setsockopt = compat_udp_setsockopt, 2261 .compat_getsockopt = compat_udp_getsockopt, 2262 #endif 2263 .clear_sk = sk_prot_clear_portaddr_nulls, 2264 }; 2265 EXPORT_SYMBOL(udp_prot); 2266 2267 /* ------------------------------------------------------------------------ */ 2268 #ifdef CONFIG_PROC_FS 2269 2270 static struct sock *udp_get_first(struct seq_file *seq, int start) 2271 { 2272 struct sock *sk; 2273 struct udp_iter_state *state = seq->private; 2274 struct net *net = seq_file_net(seq); 2275 2276 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2277 ++state->bucket) { 2278 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2279 2280 if (hlist_empty(&hslot->head)) 2281 continue; 2282 2283 spin_lock_bh(&hslot->lock); 2284 sk_for_each(sk, &hslot->head) { 2285 if (!net_eq(sock_net(sk), net)) 2286 continue; 2287 if (sk->sk_family == state->family) 2288 goto found; 2289 } 2290 spin_unlock_bh(&hslot->lock); 2291 } 2292 sk = NULL; 2293 found: 2294 return sk; 2295 } 2296 2297 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2298 { 2299 struct udp_iter_state *state = seq->private; 2300 struct net *net = seq_file_net(seq); 2301 2302 do { 2303 sk = sk_next(sk); 2304 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2305 2306 if (!sk) { 2307 if (state->bucket <= state->udp_table->mask) 2308 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2309 return udp_get_first(seq, state->bucket + 1); 2310 } 2311 return sk; 2312 } 2313 2314 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2315 { 2316 struct sock *sk = udp_get_first(seq, 0); 2317 2318 if (sk) 2319 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2320 --pos; 2321 return pos ? NULL : sk; 2322 } 2323 2324 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2325 { 2326 struct udp_iter_state *state = seq->private; 2327 state->bucket = MAX_UDP_PORTS; 2328 2329 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2330 } 2331 2332 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2333 { 2334 struct sock *sk; 2335 2336 if (v == SEQ_START_TOKEN) 2337 sk = udp_get_idx(seq, 0); 2338 else 2339 sk = udp_get_next(seq, v); 2340 2341 ++*pos; 2342 return sk; 2343 } 2344 2345 static void udp_seq_stop(struct seq_file *seq, void *v) 2346 { 2347 struct udp_iter_state *state = seq->private; 2348 2349 if (state->bucket <= state->udp_table->mask) 2350 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2351 } 2352 2353 int udp_seq_open(struct inode *inode, struct file *file) 2354 { 2355 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2356 struct udp_iter_state *s; 2357 int err; 2358 2359 err = seq_open_net(inode, file, &afinfo->seq_ops, 2360 sizeof(struct udp_iter_state)); 2361 if (err < 0) 2362 return err; 2363 2364 s = ((struct seq_file *)file->private_data)->private; 2365 s->family = afinfo->family; 2366 s->udp_table = afinfo->udp_table; 2367 return err; 2368 } 2369 EXPORT_SYMBOL(udp_seq_open); 2370 2371 /* ------------------------------------------------------------------------ */ 2372 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2373 { 2374 struct proc_dir_entry *p; 2375 int rc = 0; 2376 2377 afinfo->seq_ops.start = udp_seq_start; 2378 afinfo->seq_ops.next = udp_seq_next; 2379 afinfo->seq_ops.stop = udp_seq_stop; 2380 2381 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2382 afinfo->seq_fops, afinfo); 2383 if (!p) 2384 rc = -ENOMEM; 2385 return rc; 2386 } 2387 EXPORT_SYMBOL(udp_proc_register); 2388 2389 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2390 { 2391 remove_proc_entry(afinfo->name, net->proc_net); 2392 } 2393 EXPORT_SYMBOL(udp_proc_unregister); 2394 2395 /* ------------------------------------------------------------------------ */ 2396 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2397 int bucket) 2398 { 2399 struct inet_sock *inet = inet_sk(sp); 2400 __be32 dest = inet->inet_daddr; 2401 __be32 src = inet->inet_rcv_saddr; 2402 __u16 destp = ntohs(inet->inet_dport); 2403 __u16 srcp = ntohs(inet->inet_sport); 2404 2405 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2406 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2407 bucket, src, srcp, dest, destp, sp->sk_state, 2408 sk_wmem_alloc_get(sp), 2409 sk_rmem_alloc_get(sp), 2410 0, 0L, 0, 2411 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2412 0, sock_i_ino(sp), 2413 atomic_read(&sp->sk_refcnt), sp, 2414 atomic_read(&sp->sk_drops)); 2415 } 2416 2417 int udp4_seq_show(struct seq_file *seq, void *v) 2418 { 2419 seq_setwidth(seq, 127); 2420 if (v == SEQ_START_TOKEN) 2421 seq_puts(seq, " sl local_address rem_address st tx_queue " 2422 "rx_queue tr tm->when retrnsmt uid timeout " 2423 "inode ref pointer drops"); 2424 else { 2425 struct udp_iter_state *state = seq->private; 2426 2427 udp4_format_sock(v, seq, state->bucket); 2428 } 2429 seq_pad(seq, '\n'); 2430 return 0; 2431 } 2432 2433 static const struct file_operations udp_afinfo_seq_fops = { 2434 .owner = THIS_MODULE, 2435 .open = udp_seq_open, 2436 .read = seq_read, 2437 .llseek = seq_lseek, 2438 .release = seq_release_net 2439 }; 2440 2441 /* ------------------------------------------------------------------------ */ 2442 static struct udp_seq_afinfo udp4_seq_afinfo = { 2443 .name = "udp", 2444 .family = AF_INET, 2445 .udp_table = &udp_table, 2446 .seq_fops = &udp_afinfo_seq_fops, 2447 .seq_ops = { 2448 .show = udp4_seq_show, 2449 }, 2450 }; 2451 2452 static int __net_init udp4_proc_init_net(struct net *net) 2453 { 2454 return udp_proc_register(net, &udp4_seq_afinfo); 2455 } 2456 2457 static void __net_exit udp4_proc_exit_net(struct net *net) 2458 { 2459 udp_proc_unregister(net, &udp4_seq_afinfo); 2460 } 2461 2462 static struct pernet_operations udp4_net_ops = { 2463 .init = udp4_proc_init_net, 2464 .exit = udp4_proc_exit_net, 2465 }; 2466 2467 int __init udp4_proc_init(void) 2468 { 2469 return register_pernet_subsys(&udp4_net_ops); 2470 } 2471 2472 void udp4_proc_exit(void) 2473 { 2474 unregister_pernet_subsys(&udp4_net_ops); 2475 } 2476 #endif /* CONFIG_PROC_FS */ 2477 2478 static __initdata unsigned long uhash_entries; 2479 static int __init set_uhash_entries(char *str) 2480 { 2481 ssize_t ret; 2482 2483 if (!str) 2484 return 0; 2485 2486 ret = kstrtoul(str, 0, &uhash_entries); 2487 if (ret) 2488 return 0; 2489 2490 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2491 uhash_entries = UDP_HTABLE_SIZE_MIN; 2492 return 1; 2493 } 2494 __setup("uhash_entries=", set_uhash_entries); 2495 2496 void __init udp_table_init(struct udp_table *table, const char *name) 2497 { 2498 unsigned int i; 2499 2500 table->hash = alloc_large_system_hash(name, 2501 2 * sizeof(struct udp_hslot), 2502 uhash_entries, 2503 21, /* one slot per 2 MB */ 2504 0, 2505 &table->log, 2506 &table->mask, 2507 UDP_HTABLE_SIZE_MIN, 2508 64 * 1024); 2509 2510 table->hash2 = table->hash + (table->mask + 1); 2511 for (i = 0; i <= table->mask; i++) { 2512 INIT_HLIST_HEAD(&table->hash[i].head); 2513 table->hash[i].count = 0; 2514 spin_lock_init(&table->hash[i].lock); 2515 } 2516 for (i = 0; i <= table->mask; i++) { 2517 INIT_HLIST_HEAD(&table->hash2[i].head); 2518 table->hash2[i].count = 0; 2519 spin_lock_init(&table->hash2[i].lock); 2520 } 2521 } 2522 2523 u32 udp_flow_hashrnd(void) 2524 { 2525 static u32 hashrnd __read_mostly; 2526 2527 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2528 2529 return hashrnd; 2530 } 2531 EXPORT_SYMBOL(udp_flow_hashrnd); 2532 2533 void __init udp_init(void) 2534 { 2535 unsigned long limit; 2536 2537 udp_table_init(&udp_table, "UDP"); 2538 limit = nr_free_buffer_pages() / 8; 2539 limit = max(limit, 128UL); 2540 sysctl_udp_mem[0] = limit / 4 * 3; 2541 sysctl_udp_mem[1] = limit; 2542 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2543 2544 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2545 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2546 } 2547