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 struct hlist_nulls_node *node; 147 kuid_t uid = sock_i_uid(sk); 148 149 sk_nulls_for_each(sk2, node, &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 struct hlist_nulls_node *node; 181 kuid_t uid = sock_i_uid(sk); 182 int res = 0; 183 184 spin_lock(&hslot2->lock); 185 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) { 186 if (net_eq(sock_net(sk2), net) && 187 sk2 != sk && 188 (udp_sk(sk2)->udp_port_hash == num) && 189 (!sk2->sk_reuse || !sk->sk_reuse) && 190 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 191 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 192 (!sk2->sk_reuseport || !sk->sk_reuseport || 193 rcu_access_pointer(sk->sk_reuseport_cb) || 194 !uid_eq(uid, sock_i_uid(sk2))) && 195 saddr_comp(sk, sk2, true)) { 196 res = 1; 197 break; 198 } 199 } 200 spin_unlock(&hslot2->lock); 201 return res; 202 } 203 204 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot, 205 int (*saddr_same)(const struct sock *sk1, 206 const struct sock *sk2, 207 bool match_wildcard)) 208 { 209 struct net *net = sock_net(sk); 210 struct hlist_nulls_node *node; 211 kuid_t uid = sock_i_uid(sk); 212 struct sock *sk2; 213 214 sk_nulls_for_each(sk2, node, &hslot->head) { 215 if (net_eq(sock_net(sk2), net) && 216 sk2 != sk && 217 sk2->sk_family == sk->sk_family && 218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 221 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 222 (*saddr_same)(sk, sk2, false)) { 223 return reuseport_add_sock(sk, sk2); 224 } 225 } 226 227 /* Initial allocation may have already happened via setsockopt */ 228 if (!rcu_access_pointer(sk->sk_reuseport_cb)) 229 return reuseport_alloc(sk); 230 return 0; 231 } 232 233 /** 234 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 235 * 236 * @sk: socket struct in question 237 * @snum: port number to look up 238 * @saddr_comp: AF-dependent comparison of bound local IP addresses 239 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 240 * with NULL address 241 */ 242 int udp_lib_get_port(struct sock *sk, unsigned short snum, 243 int (*saddr_comp)(const struct sock *sk1, 244 const struct sock *sk2, 245 bool match_wildcard), 246 unsigned int hash2_nulladdr) 247 { 248 struct udp_hslot *hslot, *hslot2; 249 struct udp_table *udptable = sk->sk_prot->h.udp_table; 250 int error = 1; 251 struct net *net = sock_net(sk); 252 253 if (!snum) { 254 int low, high, remaining; 255 unsigned int rand; 256 unsigned short first, last; 257 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 258 259 inet_get_local_port_range(net, &low, &high); 260 remaining = (high - low) + 1; 261 262 rand = prandom_u32(); 263 first = reciprocal_scale(rand, remaining) + low; 264 /* 265 * force rand to be an odd multiple of UDP_HTABLE_SIZE 266 */ 267 rand = (rand | 1) * (udptable->mask + 1); 268 last = first + udptable->mask + 1; 269 do { 270 hslot = udp_hashslot(udptable, net, first); 271 bitmap_zero(bitmap, PORTS_PER_CHAIN); 272 spin_lock_bh(&hslot->lock); 273 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 274 saddr_comp, udptable->log); 275 276 snum = first; 277 /* 278 * Iterate on all possible values of snum for this hash. 279 * Using steps of an odd multiple of UDP_HTABLE_SIZE 280 * give us randomization and full range coverage. 281 */ 282 do { 283 if (low <= snum && snum <= high && 284 !test_bit(snum >> udptable->log, bitmap) && 285 !inet_is_local_reserved_port(net, snum)) 286 goto found; 287 snum += rand; 288 } while (snum != first); 289 spin_unlock_bh(&hslot->lock); 290 } while (++first != last); 291 goto fail; 292 } else { 293 hslot = udp_hashslot(udptable, net, snum); 294 spin_lock_bh(&hslot->lock); 295 if (hslot->count > 10) { 296 int exist; 297 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 298 299 slot2 &= udptable->mask; 300 hash2_nulladdr &= udptable->mask; 301 302 hslot2 = udp_hashslot2(udptable, slot2); 303 if (hslot->count < hslot2->count) 304 goto scan_primary_hash; 305 306 exist = udp_lib_lport_inuse2(net, snum, hslot2, 307 sk, saddr_comp); 308 if (!exist && (hash2_nulladdr != slot2)) { 309 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 310 exist = udp_lib_lport_inuse2(net, snum, hslot2, 311 sk, saddr_comp); 312 } 313 if (exist) 314 goto fail_unlock; 315 else 316 goto found; 317 } 318 scan_primary_hash: 319 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 320 saddr_comp, 0)) 321 goto fail_unlock; 322 } 323 found: 324 inet_sk(sk)->inet_num = snum; 325 udp_sk(sk)->udp_port_hash = snum; 326 udp_sk(sk)->udp_portaddr_hash ^= snum; 327 if (sk_unhashed(sk)) { 328 if (sk->sk_reuseport && 329 udp_reuseport_add_sock(sk, hslot, saddr_comp)) { 330 inet_sk(sk)->inet_num = 0; 331 udp_sk(sk)->udp_port_hash = 0; 332 udp_sk(sk)->udp_portaddr_hash ^= snum; 333 goto fail_unlock; 334 } 335 336 sk_nulls_add_node_rcu(sk, &hslot->head); 337 hslot->count++; 338 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 339 340 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 341 spin_lock(&hslot2->lock); 342 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 343 sk->sk_family == AF_INET6) 344 hlist_nulls_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 345 &hslot2->head); 346 else 347 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 348 &hslot2->head); 349 hslot2->count++; 350 spin_unlock(&hslot2->lock); 351 } 352 error = 0; 353 fail_unlock: 354 spin_unlock_bh(&hslot->lock); 355 fail: 356 return error; 357 } 358 EXPORT_SYMBOL(udp_lib_get_port); 359 360 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses 361 * match_wildcard == false: addresses must be exactly the same, i.e. 362 * 0.0.0.0 only equals to 0.0.0.0 363 */ 364 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2, 365 bool match_wildcard) 366 { 367 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 368 369 if (!ipv6_only_sock(sk2)) { 370 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr) 371 return 1; 372 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr) 373 return match_wildcard; 374 } 375 return 0; 376 } 377 378 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, 379 unsigned int port) 380 { 381 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 382 } 383 384 int udp_v4_get_port(struct sock *sk, unsigned short snum) 385 { 386 unsigned int hash2_nulladdr = 387 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 388 unsigned int hash2_partial = 389 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 390 391 /* precompute partial secondary hash */ 392 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 393 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 394 } 395 396 static inline int compute_score(struct sock *sk, struct net *net, 397 __be32 saddr, unsigned short hnum, __be16 sport, 398 __be32 daddr, __be16 dport, int dif) 399 { 400 int score; 401 struct inet_sock *inet; 402 403 if (!net_eq(sock_net(sk), net) || 404 udp_sk(sk)->udp_port_hash != hnum || 405 ipv6_only_sock(sk)) 406 return -1; 407 408 score = (sk->sk_family == PF_INET) ? 2 : 1; 409 inet = inet_sk(sk); 410 411 if (inet->inet_rcv_saddr) { 412 if (inet->inet_rcv_saddr != daddr) 413 return -1; 414 score += 4; 415 } 416 417 if (inet->inet_daddr) { 418 if (inet->inet_daddr != saddr) 419 return -1; 420 score += 4; 421 } 422 423 if (inet->inet_dport) { 424 if (inet->inet_dport != sport) 425 return -1; 426 score += 4; 427 } 428 429 if (sk->sk_bound_dev_if) { 430 if (sk->sk_bound_dev_if != dif) 431 return -1; 432 score += 4; 433 } 434 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 435 score++; 436 return score; 437 } 438 439 /* 440 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 441 */ 442 static inline int compute_score2(struct sock *sk, struct net *net, 443 __be32 saddr, __be16 sport, 444 __be32 daddr, unsigned int hnum, int dif) 445 { 446 int score; 447 struct inet_sock *inet; 448 449 if (!net_eq(sock_net(sk), net) || 450 ipv6_only_sock(sk)) 451 return -1; 452 453 inet = inet_sk(sk); 454 455 if (inet->inet_rcv_saddr != daddr || 456 inet->inet_num != hnum) 457 return -1; 458 459 score = (sk->sk_family == PF_INET) ? 2 : 1; 460 461 if (inet->inet_daddr) { 462 if (inet->inet_daddr != saddr) 463 return -1; 464 score += 4; 465 } 466 467 if (inet->inet_dport) { 468 if (inet->inet_dport != sport) 469 return -1; 470 score += 4; 471 } 472 473 if (sk->sk_bound_dev_if) { 474 if (sk->sk_bound_dev_if != dif) 475 return -1; 476 score += 4; 477 } 478 479 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 480 score++; 481 482 return score; 483 } 484 485 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 486 const __u16 lport, const __be32 faddr, 487 const __be16 fport) 488 { 489 static u32 udp_ehash_secret __read_mostly; 490 491 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 492 493 return __inet_ehashfn(laddr, lport, faddr, fport, 494 udp_ehash_secret + net_hash_mix(net)); 495 } 496 497 /* called with read_rcu_lock() */ 498 static struct sock *udp4_lib_lookup2(struct net *net, 499 __be32 saddr, __be16 sport, 500 __be32 daddr, unsigned int hnum, int dif, 501 struct udp_hslot *hslot2, unsigned int slot2, 502 struct sk_buff *skb) 503 { 504 struct sock *sk, *result; 505 struct hlist_nulls_node *node; 506 int score, badness, matches = 0, reuseport = 0; 507 bool select_ok = true; 508 u32 hash = 0; 509 510 begin: 511 result = NULL; 512 badness = 0; 513 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 514 score = compute_score2(sk, net, saddr, sport, 515 daddr, hnum, dif); 516 if (score > badness) { 517 result = sk; 518 badness = score; 519 reuseport = sk->sk_reuseport; 520 if (reuseport) { 521 hash = udp_ehashfn(net, daddr, hnum, 522 saddr, sport); 523 if (select_ok) { 524 struct sock *sk2; 525 526 sk2 = reuseport_select_sock(sk, hash, skb, 527 sizeof(struct udphdr)); 528 if (sk2) { 529 result = sk2; 530 select_ok = false; 531 goto found; 532 } 533 } 534 matches = 1; 535 } 536 } else if (score == badness && reuseport) { 537 matches++; 538 if (reciprocal_scale(hash, matches) == 0) 539 result = sk; 540 hash = next_pseudo_random32(hash); 541 } 542 } 543 /* 544 * if the nulls value we got at the end of this lookup is 545 * not the expected one, we must restart lookup. 546 * We probably met an item that was moved to another chain. 547 */ 548 if (get_nulls_value(node) != slot2) 549 goto begin; 550 if (result) { 551 found: 552 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 553 result = NULL; 554 else if (unlikely(compute_score2(result, net, saddr, sport, 555 daddr, hnum, dif) < badness)) { 556 sock_put(result); 557 goto begin; 558 } 559 } 560 return result; 561 } 562 563 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 564 * harder than this. -DaveM 565 */ 566 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 567 __be16 sport, __be32 daddr, __be16 dport, 568 int dif, struct udp_table *udptable, struct sk_buff *skb) 569 { 570 struct sock *sk, *result; 571 struct hlist_nulls_node *node; 572 unsigned short hnum = ntohs(dport); 573 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 574 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 575 int score, badness, matches = 0, reuseport = 0; 576 bool select_ok = true; 577 u32 hash = 0; 578 579 rcu_read_lock(); 580 if (hslot->count > 10) { 581 hash2 = udp4_portaddr_hash(net, daddr, hnum); 582 slot2 = hash2 & udptable->mask; 583 hslot2 = &udptable->hash2[slot2]; 584 if (hslot->count < hslot2->count) 585 goto begin; 586 587 result = udp4_lib_lookup2(net, saddr, sport, 588 daddr, hnum, dif, 589 hslot2, slot2, skb); 590 if (!result) { 591 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 592 slot2 = hash2 & udptable->mask; 593 hslot2 = &udptable->hash2[slot2]; 594 if (hslot->count < hslot2->count) 595 goto begin; 596 597 result = udp4_lib_lookup2(net, saddr, sport, 598 htonl(INADDR_ANY), hnum, dif, 599 hslot2, slot2, skb); 600 } 601 rcu_read_unlock(); 602 return result; 603 } 604 begin: 605 result = NULL; 606 badness = 0; 607 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 608 score = compute_score(sk, net, saddr, hnum, sport, 609 daddr, dport, dif); 610 if (score > badness) { 611 result = sk; 612 badness = score; 613 reuseport = sk->sk_reuseport; 614 if (reuseport) { 615 hash = udp_ehashfn(net, daddr, hnum, 616 saddr, sport); 617 if (select_ok) { 618 struct sock *sk2; 619 620 sk2 = reuseport_select_sock(sk, hash, skb, 621 sizeof(struct udphdr)); 622 if (sk2) { 623 result = sk2; 624 select_ok = false; 625 goto found; 626 } 627 } 628 matches = 1; 629 } 630 } else if (score == badness && reuseport) { 631 matches++; 632 if (reciprocal_scale(hash, matches) == 0) 633 result = sk; 634 hash = next_pseudo_random32(hash); 635 } 636 } 637 /* 638 * if the nulls value we got at the end of this lookup is 639 * not the expected one, we must restart lookup. 640 * We probably met an item that was moved to another chain. 641 */ 642 if (get_nulls_value(node) != slot) 643 goto begin; 644 645 if (result) { 646 found: 647 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 648 result = NULL; 649 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 650 daddr, dport, dif) < badness)) { 651 sock_put(result); 652 goto begin; 653 } 654 } 655 rcu_read_unlock(); 656 return result; 657 } 658 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 659 660 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 661 __be16 sport, __be16 dport, 662 struct udp_table *udptable) 663 { 664 const struct iphdr *iph = ip_hdr(skb); 665 666 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 667 iph->daddr, dport, inet_iif(skb), 668 udptable, skb); 669 } 670 671 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 672 __be32 daddr, __be16 dport, int dif) 673 { 674 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, 675 &udp_table, NULL); 676 } 677 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 678 679 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 680 __be16 loc_port, __be32 loc_addr, 681 __be16 rmt_port, __be32 rmt_addr, 682 int dif, unsigned short hnum) 683 { 684 struct inet_sock *inet = inet_sk(sk); 685 686 if (!net_eq(sock_net(sk), net) || 687 udp_sk(sk)->udp_port_hash != hnum || 688 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 689 (inet->inet_dport != rmt_port && inet->inet_dport) || 690 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 691 ipv6_only_sock(sk) || 692 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 693 return false; 694 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 695 return false; 696 return true; 697 } 698 699 /* 700 * This routine is called by the ICMP module when it gets some 701 * sort of error condition. If err < 0 then the socket should 702 * be closed and the error returned to the user. If err > 0 703 * it's just the icmp type << 8 | icmp code. 704 * Header points to the ip header of the error packet. We move 705 * on past this. Then (as it used to claim before adjustment) 706 * header points to the first 8 bytes of the udp header. We need 707 * to find the appropriate port. 708 */ 709 710 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 711 { 712 struct inet_sock *inet; 713 const struct iphdr *iph = (const struct iphdr *)skb->data; 714 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 715 const int type = icmp_hdr(skb)->type; 716 const int code = icmp_hdr(skb)->code; 717 struct sock *sk; 718 int harderr; 719 int err; 720 struct net *net = dev_net(skb->dev); 721 722 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 723 iph->saddr, uh->source, skb->dev->ifindex, udptable, 724 NULL); 725 if (!sk) { 726 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 727 return; /* No socket for error */ 728 } 729 730 err = 0; 731 harderr = 0; 732 inet = inet_sk(sk); 733 734 switch (type) { 735 default: 736 case ICMP_TIME_EXCEEDED: 737 err = EHOSTUNREACH; 738 break; 739 case ICMP_SOURCE_QUENCH: 740 goto out; 741 case ICMP_PARAMETERPROB: 742 err = EPROTO; 743 harderr = 1; 744 break; 745 case ICMP_DEST_UNREACH: 746 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 747 ipv4_sk_update_pmtu(skb, sk, info); 748 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 749 err = EMSGSIZE; 750 harderr = 1; 751 break; 752 } 753 goto out; 754 } 755 err = EHOSTUNREACH; 756 if (code <= NR_ICMP_UNREACH) { 757 harderr = icmp_err_convert[code].fatal; 758 err = icmp_err_convert[code].errno; 759 } 760 break; 761 case ICMP_REDIRECT: 762 ipv4_sk_redirect(skb, sk); 763 goto out; 764 } 765 766 /* 767 * RFC1122: OK. Passes ICMP errors back to application, as per 768 * 4.1.3.3. 769 */ 770 if (!inet->recverr) { 771 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 772 goto out; 773 } else 774 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 775 776 sk->sk_err = err; 777 sk->sk_error_report(sk); 778 out: 779 sock_put(sk); 780 } 781 782 void udp_err(struct sk_buff *skb, u32 info) 783 { 784 __udp4_lib_err(skb, info, &udp_table); 785 } 786 787 /* 788 * Throw away all pending data and cancel the corking. Socket is locked. 789 */ 790 void udp_flush_pending_frames(struct sock *sk) 791 { 792 struct udp_sock *up = udp_sk(sk); 793 794 if (up->pending) { 795 up->len = 0; 796 up->pending = 0; 797 ip_flush_pending_frames(sk); 798 } 799 } 800 EXPORT_SYMBOL(udp_flush_pending_frames); 801 802 /** 803 * udp4_hwcsum - handle outgoing HW checksumming 804 * @skb: sk_buff containing the filled-in UDP header 805 * (checksum field must be zeroed out) 806 * @src: source IP address 807 * @dst: destination IP address 808 */ 809 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 810 { 811 struct udphdr *uh = udp_hdr(skb); 812 int offset = skb_transport_offset(skb); 813 int len = skb->len - offset; 814 int hlen = len; 815 __wsum csum = 0; 816 817 if (!skb_has_frag_list(skb)) { 818 /* 819 * Only one fragment on the socket. 820 */ 821 skb->csum_start = skb_transport_header(skb) - skb->head; 822 skb->csum_offset = offsetof(struct udphdr, check); 823 uh->check = ~csum_tcpudp_magic(src, dst, len, 824 IPPROTO_UDP, 0); 825 } else { 826 struct sk_buff *frags; 827 828 /* 829 * HW-checksum won't work as there are two or more 830 * fragments on the socket so that all csums of sk_buffs 831 * should be together 832 */ 833 skb_walk_frags(skb, frags) { 834 csum = csum_add(csum, frags->csum); 835 hlen -= frags->len; 836 } 837 838 csum = skb_checksum(skb, offset, hlen, csum); 839 skb->ip_summed = CHECKSUM_NONE; 840 841 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 842 if (uh->check == 0) 843 uh->check = CSUM_MANGLED_0; 844 } 845 } 846 EXPORT_SYMBOL_GPL(udp4_hwcsum); 847 848 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 849 * for the simple case like when setting the checksum for a UDP tunnel. 850 */ 851 void udp_set_csum(bool nocheck, struct sk_buff *skb, 852 __be32 saddr, __be32 daddr, int len) 853 { 854 struct udphdr *uh = udp_hdr(skb); 855 856 if (nocheck) { 857 uh->check = 0; 858 } else if (skb_is_gso(skb)) { 859 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 860 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 861 uh->check = 0; 862 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 863 if (uh->check == 0) 864 uh->check = CSUM_MANGLED_0; 865 } else { 866 skb->ip_summed = CHECKSUM_PARTIAL; 867 skb->csum_start = skb_transport_header(skb) - skb->head; 868 skb->csum_offset = offsetof(struct udphdr, check); 869 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 870 } 871 } 872 EXPORT_SYMBOL(udp_set_csum); 873 874 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 875 { 876 struct sock *sk = skb->sk; 877 struct inet_sock *inet = inet_sk(sk); 878 struct udphdr *uh; 879 int err = 0; 880 int is_udplite = IS_UDPLITE(sk); 881 int offset = skb_transport_offset(skb); 882 int len = skb->len - offset; 883 __wsum csum = 0; 884 885 /* 886 * Create a UDP header 887 */ 888 uh = udp_hdr(skb); 889 uh->source = inet->inet_sport; 890 uh->dest = fl4->fl4_dport; 891 uh->len = htons(len); 892 uh->check = 0; 893 894 if (is_udplite) /* UDP-Lite */ 895 csum = udplite_csum(skb); 896 897 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 898 899 skb->ip_summed = CHECKSUM_NONE; 900 goto send; 901 902 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 903 904 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 905 goto send; 906 907 } else 908 csum = udp_csum(skb); 909 910 /* add protocol-dependent pseudo-header */ 911 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 912 sk->sk_protocol, csum); 913 if (uh->check == 0) 914 uh->check = CSUM_MANGLED_0; 915 916 send: 917 err = ip_send_skb(sock_net(sk), skb); 918 if (err) { 919 if (err == -ENOBUFS && !inet->recverr) { 920 UDP_INC_STATS_USER(sock_net(sk), 921 UDP_MIB_SNDBUFERRORS, is_udplite); 922 err = 0; 923 } 924 } else 925 UDP_INC_STATS_USER(sock_net(sk), 926 UDP_MIB_OUTDATAGRAMS, is_udplite); 927 return err; 928 } 929 930 /* 931 * Push out all pending data as one UDP datagram. Socket is locked. 932 */ 933 int udp_push_pending_frames(struct sock *sk) 934 { 935 struct udp_sock *up = udp_sk(sk); 936 struct inet_sock *inet = inet_sk(sk); 937 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 938 struct sk_buff *skb; 939 int err = 0; 940 941 skb = ip_finish_skb(sk, fl4); 942 if (!skb) 943 goto out; 944 945 err = udp_send_skb(skb, fl4); 946 947 out: 948 up->len = 0; 949 up->pending = 0; 950 return err; 951 } 952 EXPORT_SYMBOL(udp_push_pending_frames); 953 954 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 955 { 956 struct inet_sock *inet = inet_sk(sk); 957 struct udp_sock *up = udp_sk(sk); 958 struct flowi4 fl4_stack; 959 struct flowi4 *fl4; 960 int ulen = len; 961 struct ipcm_cookie ipc; 962 struct rtable *rt = NULL; 963 int free = 0; 964 int connected = 0; 965 __be32 daddr, faddr, saddr; 966 __be16 dport; 967 u8 tos; 968 int err, is_udplite = IS_UDPLITE(sk); 969 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 970 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 971 struct sk_buff *skb; 972 struct ip_options_data opt_copy; 973 974 if (len > 0xFFFF) 975 return -EMSGSIZE; 976 977 /* 978 * Check the flags. 979 */ 980 981 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 982 return -EOPNOTSUPP; 983 984 ipc.opt = NULL; 985 ipc.tx_flags = 0; 986 ipc.ttl = 0; 987 ipc.tos = -1; 988 989 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 990 991 fl4 = &inet->cork.fl.u.ip4; 992 if (up->pending) { 993 /* 994 * There are pending frames. 995 * The socket lock must be held while it's corked. 996 */ 997 lock_sock(sk); 998 if (likely(up->pending)) { 999 if (unlikely(up->pending != AF_INET)) { 1000 release_sock(sk); 1001 return -EINVAL; 1002 } 1003 goto do_append_data; 1004 } 1005 release_sock(sk); 1006 } 1007 ulen += sizeof(struct udphdr); 1008 1009 /* 1010 * Get and verify the address. 1011 */ 1012 if (msg->msg_name) { 1013 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1014 if (msg->msg_namelen < sizeof(*usin)) 1015 return -EINVAL; 1016 if (usin->sin_family != AF_INET) { 1017 if (usin->sin_family != AF_UNSPEC) 1018 return -EAFNOSUPPORT; 1019 } 1020 1021 daddr = usin->sin_addr.s_addr; 1022 dport = usin->sin_port; 1023 if (dport == 0) 1024 return -EINVAL; 1025 } else { 1026 if (sk->sk_state != TCP_ESTABLISHED) 1027 return -EDESTADDRREQ; 1028 daddr = inet->inet_daddr; 1029 dport = inet->inet_dport; 1030 /* Open fast path for connected socket. 1031 Route will not be used, if at least one option is set. 1032 */ 1033 connected = 1; 1034 } 1035 ipc.addr = inet->inet_saddr; 1036 1037 ipc.oif = sk->sk_bound_dev_if; 1038 1039 sock_tx_timestamp(sk, &ipc.tx_flags); 1040 1041 if (msg->msg_controllen) { 1042 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 1043 sk->sk_family == AF_INET6); 1044 if (unlikely(err)) { 1045 kfree(ipc.opt); 1046 return err; 1047 } 1048 if (ipc.opt) 1049 free = 1; 1050 connected = 0; 1051 } 1052 if (!ipc.opt) { 1053 struct ip_options_rcu *inet_opt; 1054 1055 rcu_read_lock(); 1056 inet_opt = rcu_dereference(inet->inet_opt); 1057 if (inet_opt) { 1058 memcpy(&opt_copy, inet_opt, 1059 sizeof(*inet_opt) + inet_opt->opt.optlen); 1060 ipc.opt = &opt_copy.opt; 1061 } 1062 rcu_read_unlock(); 1063 } 1064 1065 saddr = ipc.addr; 1066 ipc.addr = faddr = daddr; 1067 1068 if (ipc.opt && ipc.opt->opt.srr) { 1069 if (!daddr) 1070 return -EINVAL; 1071 faddr = ipc.opt->opt.faddr; 1072 connected = 0; 1073 } 1074 tos = get_rttos(&ipc, inet); 1075 if (sock_flag(sk, SOCK_LOCALROUTE) || 1076 (msg->msg_flags & MSG_DONTROUTE) || 1077 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1078 tos |= RTO_ONLINK; 1079 connected = 0; 1080 } 1081 1082 if (ipv4_is_multicast(daddr)) { 1083 if (!ipc.oif) 1084 ipc.oif = inet->mc_index; 1085 if (!saddr) 1086 saddr = inet->mc_addr; 1087 connected = 0; 1088 } else if (!ipc.oif) 1089 ipc.oif = inet->uc_index; 1090 1091 if (connected) 1092 rt = (struct rtable *)sk_dst_check(sk, 0); 1093 1094 if (!rt) { 1095 struct net *net = sock_net(sk); 1096 __u8 flow_flags = inet_sk_flowi_flags(sk); 1097 1098 fl4 = &fl4_stack; 1099 1100 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1101 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1102 flow_flags, 1103 faddr, saddr, dport, inet->inet_sport); 1104 1105 if (!saddr && ipc.oif) { 1106 err = l3mdev_get_saddr(net, ipc.oif, fl4); 1107 if (err < 0) 1108 goto out; 1109 } 1110 1111 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1112 rt = ip_route_output_flow(net, fl4, sk); 1113 if (IS_ERR(rt)) { 1114 err = PTR_ERR(rt); 1115 rt = NULL; 1116 if (err == -ENETUNREACH) 1117 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1118 goto out; 1119 } 1120 1121 err = -EACCES; 1122 if ((rt->rt_flags & RTCF_BROADCAST) && 1123 !sock_flag(sk, SOCK_BROADCAST)) 1124 goto out; 1125 if (connected) 1126 sk_dst_set(sk, dst_clone(&rt->dst)); 1127 } 1128 1129 if (msg->msg_flags&MSG_CONFIRM) 1130 goto do_confirm; 1131 back_from_confirm: 1132 1133 saddr = fl4->saddr; 1134 if (!ipc.addr) 1135 daddr = ipc.addr = fl4->daddr; 1136 1137 /* Lockless fast path for the non-corking case. */ 1138 if (!corkreq) { 1139 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1140 sizeof(struct udphdr), &ipc, &rt, 1141 msg->msg_flags); 1142 err = PTR_ERR(skb); 1143 if (!IS_ERR_OR_NULL(skb)) 1144 err = udp_send_skb(skb, fl4); 1145 goto out; 1146 } 1147 1148 lock_sock(sk); 1149 if (unlikely(up->pending)) { 1150 /* The socket is already corked while preparing it. */ 1151 /* ... which is an evident application bug. --ANK */ 1152 release_sock(sk); 1153 1154 net_dbg_ratelimited("cork app bug 2\n"); 1155 err = -EINVAL; 1156 goto out; 1157 } 1158 /* 1159 * Now cork the socket to pend data. 1160 */ 1161 fl4 = &inet->cork.fl.u.ip4; 1162 fl4->daddr = daddr; 1163 fl4->saddr = saddr; 1164 fl4->fl4_dport = dport; 1165 fl4->fl4_sport = inet->inet_sport; 1166 up->pending = AF_INET; 1167 1168 do_append_data: 1169 up->len += ulen; 1170 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1171 sizeof(struct udphdr), &ipc, &rt, 1172 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1173 if (err) 1174 udp_flush_pending_frames(sk); 1175 else if (!corkreq) 1176 err = udp_push_pending_frames(sk); 1177 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1178 up->pending = 0; 1179 release_sock(sk); 1180 1181 out: 1182 ip_rt_put(rt); 1183 if (free) 1184 kfree(ipc.opt); 1185 if (!err) 1186 return len; 1187 /* 1188 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1189 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1190 * we don't have a good statistic (IpOutDiscards but it can be too many 1191 * things). We could add another new stat but at least for now that 1192 * seems like overkill. 1193 */ 1194 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1195 UDP_INC_STATS_USER(sock_net(sk), 1196 UDP_MIB_SNDBUFERRORS, is_udplite); 1197 } 1198 return err; 1199 1200 do_confirm: 1201 dst_confirm(&rt->dst); 1202 if (!(msg->msg_flags&MSG_PROBE) || len) 1203 goto back_from_confirm; 1204 err = 0; 1205 goto out; 1206 } 1207 EXPORT_SYMBOL(udp_sendmsg); 1208 1209 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1210 size_t size, int flags) 1211 { 1212 struct inet_sock *inet = inet_sk(sk); 1213 struct udp_sock *up = udp_sk(sk); 1214 int ret; 1215 1216 if (flags & MSG_SENDPAGE_NOTLAST) 1217 flags |= MSG_MORE; 1218 1219 if (!up->pending) { 1220 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1221 1222 /* Call udp_sendmsg to specify destination address which 1223 * sendpage interface can't pass. 1224 * This will succeed only when the socket is connected. 1225 */ 1226 ret = udp_sendmsg(sk, &msg, 0); 1227 if (ret < 0) 1228 return ret; 1229 } 1230 1231 lock_sock(sk); 1232 1233 if (unlikely(!up->pending)) { 1234 release_sock(sk); 1235 1236 net_dbg_ratelimited("udp cork app bug 3\n"); 1237 return -EINVAL; 1238 } 1239 1240 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1241 page, offset, size, flags); 1242 if (ret == -EOPNOTSUPP) { 1243 release_sock(sk); 1244 return sock_no_sendpage(sk->sk_socket, page, offset, 1245 size, flags); 1246 } 1247 if (ret < 0) { 1248 udp_flush_pending_frames(sk); 1249 goto out; 1250 } 1251 1252 up->len += size; 1253 if (!(up->corkflag || (flags&MSG_MORE))) 1254 ret = udp_push_pending_frames(sk); 1255 if (!ret) 1256 ret = size; 1257 out: 1258 release_sock(sk); 1259 return ret; 1260 } 1261 1262 /** 1263 * first_packet_length - return length of first packet in receive queue 1264 * @sk: socket 1265 * 1266 * Drops all bad checksum frames, until a valid one is found. 1267 * Returns the length of found skb, or 0 if none is found. 1268 */ 1269 static unsigned int first_packet_length(struct sock *sk) 1270 { 1271 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1272 struct sk_buff *skb; 1273 unsigned int res; 1274 1275 __skb_queue_head_init(&list_kill); 1276 1277 spin_lock_bh(&rcvq->lock); 1278 while ((skb = skb_peek(rcvq)) != NULL && 1279 udp_lib_checksum_complete(skb)) { 1280 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1281 IS_UDPLITE(sk)); 1282 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1283 IS_UDPLITE(sk)); 1284 atomic_inc(&sk->sk_drops); 1285 __skb_unlink(skb, rcvq); 1286 __skb_queue_tail(&list_kill, skb); 1287 } 1288 res = skb ? skb->len : 0; 1289 spin_unlock_bh(&rcvq->lock); 1290 1291 if (!skb_queue_empty(&list_kill)) { 1292 bool slow = lock_sock_fast(sk); 1293 1294 __skb_queue_purge(&list_kill); 1295 sk_mem_reclaim_partial(sk); 1296 unlock_sock_fast(sk, slow); 1297 } 1298 return res; 1299 } 1300 1301 /* 1302 * IOCTL requests applicable to the UDP protocol 1303 */ 1304 1305 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1306 { 1307 switch (cmd) { 1308 case SIOCOUTQ: 1309 { 1310 int amount = sk_wmem_alloc_get(sk); 1311 1312 return put_user(amount, (int __user *)arg); 1313 } 1314 1315 case SIOCINQ: 1316 { 1317 unsigned int amount = first_packet_length(sk); 1318 1319 if (amount) 1320 /* 1321 * We will only return the amount 1322 * of this packet since that is all 1323 * that will be read. 1324 */ 1325 amount -= sizeof(struct udphdr); 1326 1327 return put_user(amount, (int __user *)arg); 1328 } 1329 1330 default: 1331 return -ENOIOCTLCMD; 1332 } 1333 1334 return 0; 1335 } 1336 EXPORT_SYMBOL(udp_ioctl); 1337 1338 /* 1339 * This should be easy, if there is something there we 1340 * return it, otherwise we block. 1341 */ 1342 1343 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1344 int flags, int *addr_len) 1345 { 1346 struct inet_sock *inet = inet_sk(sk); 1347 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1348 struct sk_buff *skb; 1349 unsigned int ulen, copied; 1350 int peeked, off = 0; 1351 int err; 1352 int is_udplite = IS_UDPLITE(sk); 1353 bool checksum_valid = false; 1354 bool slow; 1355 1356 if (flags & MSG_ERRQUEUE) 1357 return ip_recv_error(sk, msg, len, addr_len); 1358 1359 try_again: 1360 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1361 &peeked, &off, &err); 1362 if (!skb) 1363 goto out; 1364 1365 ulen = skb->len - sizeof(struct udphdr); 1366 copied = len; 1367 if (copied > ulen) 1368 copied = ulen; 1369 else if (copied < ulen) 1370 msg->msg_flags |= MSG_TRUNC; 1371 1372 /* 1373 * If checksum is needed at all, try to do it while copying the 1374 * data. If the data is truncated, or if we only want a partial 1375 * coverage checksum (UDP-Lite), do it before the copy. 1376 */ 1377 1378 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1379 checksum_valid = !udp_lib_checksum_complete(skb); 1380 if (!checksum_valid) 1381 goto csum_copy_err; 1382 } 1383 1384 if (checksum_valid || skb_csum_unnecessary(skb)) 1385 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 1386 msg, copied); 1387 else { 1388 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), 1389 msg); 1390 1391 if (err == -EINVAL) 1392 goto csum_copy_err; 1393 } 1394 1395 if (unlikely(err)) { 1396 trace_kfree_skb(skb, udp_recvmsg); 1397 if (!peeked) { 1398 atomic_inc(&sk->sk_drops); 1399 UDP_INC_STATS_USER(sock_net(sk), 1400 UDP_MIB_INERRORS, is_udplite); 1401 } 1402 goto out_free; 1403 } 1404 1405 if (!peeked) 1406 UDP_INC_STATS_USER(sock_net(sk), 1407 UDP_MIB_INDATAGRAMS, is_udplite); 1408 1409 sock_recv_ts_and_drops(msg, sk, skb); 1410 1411 /* Copy the address. */ 1412 if (sin) { 1413 sin->sin_family = AF_INET; 1414 sin->sin_port = udp_hdr(skb)->source; 1415 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1416 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1417 *addr_len = sizeof(*sin); 1418 } 1419 if (inet->cmsg_flags) 1420 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr)); 1421 1422 err = copied; 1423 if (flags & MSG_TRUNC) 1424 err = ulen; 1425 1426 out_free: 1427 skb_free_datagram_locked(sk, skb); 1428 out: 1429 return err; 1430 1431 csum_copy_err: 1432 slow = lock_sock_fast(sk); 1433 if (!skb_kill_datagram(sk, skb, flags)) { 1434 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1435 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1436 } 1437 unlock_sock_fast(sk, slow); 1438 1439 /* starting over for a new packet, but check if we need to yield */ 1440 cond_resched(); 1441 msg->msg_flags &= ~MSG_TRUNC; 1442 goto try_again; 1443 } 1444 1445 int udp_disconnect(struct sock *sk, int flags) 1446 { 1447 struct inet_sock *inet = inet_sk(sk); 1448 /* 1449 * 1003.1g - break association. 1450 */ 1451 1452 sk->sk_state = TCP_CLOSE; 1453 inet->inet_daddr = 0; 1454 inet->inet_dport = 0; 1455 sock_rps_reset_rxhash(sk); 1456 sk->sk_bound_dev_if = 0; 1457 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1458 inet_reset_saddr(sk); 1459 1460 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1461 sk->sk_prot->unhash(sk); 1462 inet->inet_sport = 0; 1463 } 1464 sk_dst_reset(sk); 1465 return 0; 1466 } 1467 EXPORT_SYMBOL(udp_disconnect); 1468 1469 void udp_lib_unhash(struct sock *sk) 1470 { 1471 if (sk_hashed(sk)) { 1472 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1473 struct udp_hslot *hslot, *hslot2; 1474 1475 hslot = udp_hashslot(udptable, sock_net(sk), 1476 udp_sk(sk)->udp_port_hash); 1477 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1478 1479 spin_lock_bh(&hslot->lock); 1480 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1481 reuseport_detach_sock(sk); 1482 if (sk_nulls_del_node_init_rcu(sk)) { 1483 hslot->count--; 1484 inet_sk(sk)->inet_num = 0; 1485 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1486 1487 spin_lock(&hslot2->lock); 1488 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1489 hslot2->count--; 1490 spin_unlock(&hslot2->lock); 1491 } 1492 spin_unlock_bh(&hslot->lock); 1493 } 1494 } 1495 EXPORT_SYMBOL(udp_lib_unhash); 1496 1497 /* 1498 * inet_rcv_saddr was changed, we must rehash secondary hash 1499 */ 1500 void udp_lib_rehash(struct sock *sk, u16 newhash) 1501 { 1502 if (sk_hashed(sk)) { 1503 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1504 struct udp_hslot *hslot, *hslot2, *nhslot2; 1505 1506 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1507 nhslot2 = udp_hashslot2(udptable, newhash); 1508 udp_sk(sk)->udp_portaddr_hash = newhash; 1509 1510 if (hslot2 != nhslot2 || 1511 rcu_access_pointer(sk->sk_reuseport_cb)) { 1512 hslot = udp_hashslot(udptable, sock_net(sk), 1513 udp_sk(sk)->udp_port_hash); 1514 /* we must lock primary chain too */ 1515 spin_lock_bh(&hslot->lock); 1516 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1517 reuseport_detach_sock(sk); 1518 1519 if (hslot2 != nhslot2) { 1520 spin_lock(&hslot2->lock); 1521 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1522 hslot2->count--; 1523 spin_unlock(&hslot2->lock); 1524 1525 spin_lock(&nhslot2->lock); 1526 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1527 &nhslot2->head); 1528 nhslot2->count++; 1529 spin_unlock(&nhslot2->lock); 1530 } 1531 1532 spin_unlock_bh(&hslot->lock); 1533 } 1534 } 1535 } 1536 EXPORT_SYMBOL(udp_lib_rehash); 1537 1538 static void udp_v4_rehash(struct sock *sk) 1539 { 1540 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1541 inet_sk(sk)->inet_rcv_saddr, 1542 inet_sk(sk)->inet_num); 1543 udp_lib_rehash(sk, new_hash); 1544 } 1545 1546 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1547 { 1548 int rc; 1549 1550 if (inet_sk(sk)->inet_daddr) { 1551 sock_rps_save_rxhash(sk, skb); 1552 sk_mark_napi_id(sk, skb); 1553 sk_incoming_cpu_update(sk); 1554 } 1555 1556 rc = sock_queue_rcv_skb(sk, skb); 1557 if (rc < 0) { 1558 int is_udplite = IS_UDPLITE(sk); 1559 1560 /* Note that an ENOMEM error is charged twice */ 1561 if (rc == -ENOMEM) 1562 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1563 is_udplite); 1564 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1565 kfree_skb(skb); 1566 trace_udp_fail_queue_rcv_skb(rc, sk); 1567 return -1; 1568 } 1569 1570 return 0; 1571 1572 } 1573 1574 static struct static_key udp_encap_needed __read_mostly; 1575 void udp_encap_enable(void) 1576 { 1577 if (!static_key_enabled(&udp_encap_needed)) 1578 static_key_slow_inc(&udp_encap_needed); 1579 } 1580 EXPORT_SYMBOL(udp_encap_enable); 1581 1582 /* returns: 1583 * -1: error 1584 * 0: success 1585 * >0: "udp encap" protocol resubmission 1586 * 1587 * Note that in the success and error cases, the skb is assumed to 1588 * have either been requeued or freed. 1589 */ 1590 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1591 { 1592 struct udp_sock *up = udp_sk(sk); 1593 int rc; 1594 int is_udplite = IS_UDPLITE(sk); 1595 1596 /* 1597 * Charge it to the socket, dropping if the queue is full. 1598 */ 1599 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1600 goto drop; 1601 nf_reset(skb); 1602 1603 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1604 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1605 1606 /* 1607 * This is an encapsulation socket so pass the skb to 1608 * the socket's udp_encap_rcv() hook. Otherwise, just 1609 * fall through and pass this up the UDP socket. 1610 * up->encap_rcv() returns the following value: 1611 * =0 if skb was successfully passed to the encap 1612 * handler or was discarded by it. 1613 * >0 if skb should be passed on to UDP. 1614 * <0 if skb should be resubmitted as proto -N 1615 */ 1616 1617 /* if we're overly short, let UDP handle it */ 1618 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1619 if (skb->len > sizeof(struct udphdr) && encap_rcv) { 1620 int ret; 1621 1622 /* Verify checksum before giving to encap */ 1623 if (udp_lib_checksum_complete(skb)) 1624 goto csum_error; 1625 1626 ret = encap_rcv(sk, skb); 1627 if (ret <= 0) { 1628 UDP_INC_STATS_BH(sock_net(sk), 1629 UDP_MIB_INDATAGRAMS, 1630 is_udplite); 1631 return -ret; 1632 } 1633 } 1634 1635 /* FALLTHROUGH -- it's a UDP Packet */ 1636 } 1637 1638 /* 1639 * UDP-Lite specific tests, ignored on UDP sockets 1640 */ 1641 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1642 1643 /* 1644 * MIB statistics other than incrementing the error count are 1645 * disabled for the following two types of errors: these depend 1646 * on the application settings, not on the functioning of the 1647 * protocol stack as such. 1648 * 1649 * RFC 3828 here recommends (sec 3.3): "There should also be a 1650 * way ... to ... at least let the receiving application block 1651 * delivery of packets with coverage values less than a value 1652 * provided by the application." 1653 */ 1654 if (up->pcrlen == 0) { /* full coverage was set */ 1655 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1656 UDP_SKB_CB(skb)->cscov, skb->len); 1657 goto drop; 1658 } 1659 /* The next case involves violating the min. coverage requested 1660 * by the receiver. This is subtle: if receiver wants x and x is 1661 * greater than the buffersize/MTU then receiver will complain 1662 * that it wants x while sender emits packets of smaller size y. 1663 * Therefore the above ...()->partial_cov statement is essential. 1664 */ 1665 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1666 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1667 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1668 goto drop; 1669 } 1670 } 1671 1672 if (rcu_access_pointer(sk->sk_filter) && 1673 udp_lib_checksum_complete(skb)) 1674 goto csum_error; 1675 1676 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1677 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1678 is_udplite); 1679 goto drop; 1680 } 1681 1682 rc = 0; 1683 1684 ipv4_pktinfo_prepare(sk, skb); 1685 bh_lock_sock(sk); 1686 if (!sock_owned_by_user(sk)) 1687 rc = __udp_queue_rcv_skb(sk, skb); 1688 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1689 bh_unlock_sock(sk); 1690 goto drop; 1691 } 1692 bh_unlock_sock(sk); 1693 1694 return rc; 1695 1696 csum_error: 1697 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1698 drop: 1699 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1700 atomic_inc(&sk->sk_drops); 1701 kfree_skb(skb); 1702 return -1; 1703 } 1704 1705 static void flush_stack(struct sock **stack, unsigned int count, 1706 struct sk_buff *skb, unsigned int final) 1707 { 1708 unsigned int i; 1709 struct sk_buff *skb1 = NULL; 1710 struct sock *sk; 1711 1712 for (i = 0; i < count; i++) { 1713 sk = stack[i]; 1714 if (likely(!skb1)) 1715 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1716 1717 if (!skb1) { 1718 atomic_inc(&sk->sk_drops); 1719 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1720 IS_UDPLITE(sk)); 1721 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1722 IS_UDPLITE(sk)); 1723 } 1724 1725 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1726 skb1 = NULL; 1727 1728 sock_put(sk); 1729 } 1730 if (unlikely(skb1)) 1731 kfree_skb(skb1); 1732 } 1733 1734 /* For TCP sockets, sk_rx_dst is protected by socket lock 1735 * For UDP, we use xchg() to guard against concurrent changes. 1736 */ 1737 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1738 { 1739 struct dst_entry *old; 1740 1741 dst_hold(dst); 1742 old = xchg(&sk->sk_rx_dst, dst); 1743 dst_release(old); 1744 } 1745 1746 /* 1747 * Multicasts and broadcasts go to each listener. 1748 * 1749 * Note: called only from the BH handler context. 1750 */ 1751 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1752 struct udphdr *uh, 1753 __be32 saddr, __be32 daddr, 1754 struct udp_table *udptable, 1755 int proto) 1756 { 1757 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1758 struct hlist_nulls_node *node; 1759 unsigned short hnum = ntohs(uh->dest); 1760 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1761 int dif = skb->dev->ifindex; 1762 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 1763 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1764 bool inner_flushed = false; 1765 1766 if (use_hash2) { 1767 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1768 udp_table.mask; 1769 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask; 1770 start_lookup: 1771 hslot = &udp_table.hash2[hash2]; 1772 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1773 } 1774 1775 spin_lock(&hslot->lock); 1776 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 1777 if (__udp_is_mcast_sock(net, sk, 1778 uh->dest, daddr, 1779 uh->source, saddr, 1780 dif, hnum)) { 1781 if (unlikely(count == ARRAY_SIZE(stack))) { 1782 flush_stack(stack, count, skb, ~0); 1783 inner_flushed = true; 1784 count = 0; 1785 } 1786 stack[count++] = sk; 1787 sock_hold(sk); 1788 } 1789 } 1790 1791 spin_unlock(&hslot->lock); 1792 1793 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1794 if (use_hash2 && hash2 != hash2_any) { 1795 hash2 = hash2_any; 1796 goto start_lookup; 1797 } 1798 1799 /* 1800 * do the slow work with no lock held 1801 */ 1802 if (count) { 1803 flush_stack(stack, count, skb, count - 1); 1804 } else { 1805 if (!inner_flushed) 1806 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 1807 proto == IPPROTO_UDPLITE); 1808 consume_skb(skb); 1809 } 1810 return 0; 1811 } 1812 1813 /* Initialize UDP checksum. If exited with zero value (success), 1814 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1815 * Otherwise, csum completion requires chacksumming packet body, 1816 * including udp header and folding it to skb->csum. 1817 */ 1818 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1819 int proto) 1820 { 1821 int err; 1822 1823 UDP_SKB_CB(skb)->partial_cov = 0; 1824 UDP_SKB_CB(skb)->cscov = skb->len; 1825 1826 if (proto == IPPROTO_UDPLITE) { 1827 err = udplite_checksum_init(skb, uh); 1828 if (err) 1829 return err; 1830 } 1831 1832 return skb_checksum_init_zero_check(skb, proto, uh->check, 1833 inet_compute_pseudo); 1834 } 1835 1836 /* 1837 * All we need to do is get the socket, and then do a checksum. 1838 */ 1839 1840 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1841 int proto) 1842 { 1843 struct sock *sk; 1844 struct udphdr *uh; 1845 unsigned short ulen; 1846 struct rtable *rt = skb_rtable(skb); 1847 __be32 saddr, daddr; 1848 struct net *net = dev_net(skb->dev); 1849 1850 /* 1851 * Validate the packet. 1852 */ 1853 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1854 goto drop; /* No space for header. */ 1855 1856 uh = udp_hdr(skb); 1857 ulen = ntohs(uh->len); 1858 saddr = ip_hdr(skb)->saddr; 1859 daddr = ip_hdr(skb)->daddr; 1860 1861 if (ulen > skb->len) 1862 goto short_packet; 1863 1864 if (proto == IPPROTO_UDP) { 1865 /* UDP validates ulen. */ 1866 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1867 goto short_packet; 1868 uh = udp_hdr(skb); 1869 } 1870 1871 if (udp4_csum_init(skb, uh, proto)) 1872 goto csum_error; 1873 1874 sk = skb_steal_sock(skb); 1875 if (sk) { 1876 struct dst_entry *dst = skb_dst(skb); 1877 int ret; 1878 1879 if (unlikely(sk->sk_rx_dst != dst)) 1880 udp_sk_rx_dst_set(sk, dst); 1881 1882 ret = udp_queue_rcv_skb(sk, skb); 1883 sock_put(sk); 1884 /* a return value > 0 means to resubmit the input, but 1885 * it wants the return to be -protocol, or 0 1886 */ 1887 if (ret > 0) 1888 return -ret; 1889 return 0; 1890 } 1891 1892 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1893 return __udp4_lib_mcast_deliver(net, skb, uh, 1894 saddr, daddr, udptable, proto); 1895 1896 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1897 if (sk) { 1898 int ret; 1899 1900 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1901 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1902 inet_compute_pseudo); 1903 1904 ret = udp_queue_rcv_skb(sk, skb); 1905 sock_put(sk); 1906 1907 /* a return value > 0 means to resubmit the input, but 1908 * it wants the return to be -protocol, or 0 1909 */ 1910 if (ret > 0) 1911 return -ret; 1912 return 0; 1913 } 1914 1915 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1916 goto drop; 1917 nf_reset(skb); 1918 1919 /* No socket. Drop packet silently, if checksum is wrong */ 1920 if (udp_lib_checksum_complete(skb)) 1921 goto csum_error; 1922 1923 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1924 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1925 1926 /* 1927 * Hmm. We got an UDP packet to a port to which we 1928 * don't wanna listen. Ignore it. 1929 */ 1930 kfree_skb(skb); 1931 return 0; 1932 1933 short_packet: 1934 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1935 proto == IPPROTO_UDPLITE ? "Lite" : "", 1936 &saddr, ntohs(uh->source), 1937 ulen, skb->len, 1938 &daddr, ntohs(uh->dest)); 1939 goto drop; 1940 1941 csum_error: 1942 /* 1943 * RFC1122: OK. Discards the bad packet silently (as far as 1944 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1945 */ 1946 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1947 proto == IPPROTO_UDPLITE ? "Lite" : "", 1948 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1949 ulen); 1950 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1951 drop: 1952 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1953 kfree_skb(skb); 1954 return 0; 1955 } 1956 1957 /* We can only early demux multicast if there is a single matching socket. 1958 * If more than one socket found returns NULL 1959 */ 1960 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1961 __be16 loc_port, __be32 loc_addr, 1962 __be16 rmt_port, __be32 rmt_addr, 1963 int dif) 1964 { 1965 struct sock *sk, *result; 1966 struct hlist_nulls_node *node; 1967 unsigned short hnum = ntohs(loc_port); 1968 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1969 struct udp_hslot *hslot = &udp_table.hash[slot]; 1970 1971 /* Do not bother scanning a too big list */ 1972 if (hslot->count > 10) 1973 return NULL; 1974 1975 rcu_read_lock(); 1976 begin: 1977 count = 0; 1978 result = NULL; 1979 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1980 if (__udp_is_mcast_sock(net, sk, 1981 loc_port, loc_addr, 1982 rmt_port, rmt_addr, 1983 dif, hnum)) { 1984 result = sk; 1985 ++count; 1986 } 1987 } 1988 /* 1989 * if the nulls value we got at the end of this lookup is 1990 * not the expected one, we must restart lookup. 1991 * We probably met an item that was moved to another chain. 1992 */ 1993 if (get_nulls_value(node) != slot) 1994 goto begin; 1995 1996 if (result) { 1997 if (count != 1 || 1998 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1999 result = NULL; 2000 else if (unlikely(!__udp_is_mcast_sock(net, result, 2001 loc_port, loc_addr, 2002 rmt_port, rmt_addr, 2003 dif, hnum))) { 2004 sock_put(result); 2005 result = NULL; 2006 } 2007 } 2008 rcu_read_unlock(); 2009 return result; 2010 } 2011 2012 /* For unicast we should only early demux connected sockets or we can 2013 * break forwarding setups. The chains here can be long so only check 2014 * if the first socket is an exact match and if not move on. 2015 */ 2016 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2017 __be16 loc_port, __be32 loc_addr, 2018 __be16 rmt_port, __be32 rmt_addr, 2019 int dif) 2020 { 2021 struct sock *sk, *result; 2022 struct hlist_nulls_node *node; 2023 unsigned short hnum = ntohs(loc_port); 2024 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 2025 unsigned int slot2 = hash2 & udp_table.mask; 2026 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2027 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2028 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2029 2030 rcu_read_lock(); 2031 result = NULL; 2032 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 2033 if (INET_MATCH(sk, net, acookie, 2034 rmt_addr, loc_addr, ports, dif)) 2035 result = sk; 2036 /* Only check first socket in chain */ 2037 break; 2038 } 2039 2040 if (result) { 2041 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 2042 result = NULL; 2043 else if (unlikely(!INET_MATCH(sk, net, acookie, 2044 rmt_addr, loc_addr, 2045 ports, dif))) { 2046 sock_put(result); 2047 result = NULL; 2048 } 2049 } 2050 rcu_read_unlock(); 2051 return result; 2052 } 2053 2054 void udp_v4_early_demux(struct sk_buff *skb) 2055 { 2056 struct net *net = dev_net(skb->dev); 2057 const struct iphdr *iph; 2058 const struct udphdr *uh; 2059 struct sock *sk; 2060 struct dst_entry *dst; 2061 int dif = skb->dev->ifindex; 2062 int ours; 2063 2064 /* validate the packet */ 2065 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2066 return; 2067 2068 iph = ip_hdr(skb); 2069 uh = udp_hdr(skb); 2070 2071 if (skb->pkt_type == PACKET_BROADCAST || 2072 skb->pkt_type == PACKET_MULTICAST) { 2073 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 2074 2075 if (!in_dev) 2076 return; 2077 2078 /* we are supposed to accept bcast packets */ 2079 if (skb->pkt_type == PACKET_MULTICAST) { 2080 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2081 iph->protocol); 2082 if (!ours) 2083 return; 2084 } 2085 2086 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2087 uh->source, iph->saddr, dif); 2088 } else if (skb->pkt_type == PACKET_HOST) { 2089 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2090 uh->source, iph->saddr, dif); 2091 } else { 2092 return; 2093 } 2094 2095 if (!sk) 2096 return; 2097 2098 skb->sk = sk; 2099 skb->destructor = sock_efree; 2100 dst = READ_ONCE(sk->sk_rx_dst); 2101 2102 if (dst) 2103 dst = dst_check(dst, 0); 2104 if (dst) { 2105 /* DST_NOCACHE can not be used without taking a reference */ 2106 if (dst->flags & DST_NOCACHE) { 2107 if (likely(atomic_inc_not_zero(&dst->__refcnt))) 2108 skb_dst_set(skb, dst); 2109 } else { 2110 skb_dst_set_noref(skb, dst); 2111 } 2112 } 2113 } 2114 2115 int udp_rcv(struct sk_buff *skb) 2116 { 2117 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2118 } 2119 2120 void udp_destroy_sock(struct sock *sk) 2121 { 2122 struct udp_sock *up = udp_sk(sk); 2123 bool slow = lock_sock_fast(sk); 2124 udp_flush_pending_frames(sk); 2125 unlock_sock_fast(sk, slow); 2126 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2127 void (*encap_destroy)(struct sock *sk); 2128 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2129 if (encap_destroy) 2130 encap_destroy(sk); 2131 } 2132 } 2133 2134 /* 2135 * Socket option code for UDP 2136 */ 2137 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2138 char __user *optval, unsigned int optlen, 2139 int (*push_pending_frames)(struct sock *)) 2140 { 2141 struct udp_sock *up = udp_sk(sk); 2142 int val, valbool; 2143 int err = 0; 2144 int is_udplite = IS_UDPLITE(sk); 2145 2146 if (optlen < sizeof(int)) 2147 return -EINVAL; 2148 2149 if (get_user(val, (int __user *)optval)) 2150 return -EFAULT; 2151 2152 valbool = val ? 1 : 0; 2153 2154 switch (optname) { 2155 case UDP_CORK: 2156 if (val != 0) { 2157 up->corkflag = 1; 2158 } else { 2159 up->corkflag = 0; 2160 lock_sock(sk); 2161 push_pending_frames(sk); 2162 release_sock(sk); 2163 } 2164 break; 2165 2166 case UDP_ENCAP: 2167 switch (val) { 2168 case 0: 2169 case UDP_ENCAP_ESPINUDP: 2170 case UDP_ENCAP_ESPINUDP_NON_IKE: 2171 up->encap_rcv = xfrm4_udp_encap_rcv; 2172 /* FALLTHROUGH */ 2173 case UDP_ENCAP_L2TPINUDP: 2174 up->encap_type = val; 2175 udp_encap_enable(); 2176 break; 2177 default: 2178 err = -ENOPROTOOPT; 2179 break; 2180 } 2181 break; 2182 2183 case UDP_NO_CHECK6_TX: 2184 up->no_check6_tx = valbool; 2185 break; 2186 2187 case UDP_NO_CHECK6_RX: 2188 up->no_check6_rx = valbool; 2189 break; 2190 2191 /* 2192 * UDP-Lite's partial checksum coverage (RFC 3828). 2193 */ 2194 /* The sender sets actual checksum coverage length via this option. 2195 * The case coverage > packet length is handled by send module. */ 2196 case UDPLITE_SEND_CSCOV: 2197 if (!is_udplite) /* Disable the option on UDP sockets */ 2198 return -ENOPROTOOPT; 2199 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2200 val = 8; 2201 else if (val > USHRT_MAX) 2202 val = USHRT_MAX; 2203 up->pcslen = val; 2204 up->pcflag |= UDPLITE_SEND_CC; 2205 break; 2206 2207 /* The receiver specifies a minimum checksum coverage value. To make 2208 * sense, this should be set to at least 8 (as done below). If zero is 2209 * used, this again means full checksum coverage. */ 2210 case UDPLITE_RECV_CSCOV: 2211 if (!is_udplite) /* Disable the option on UDP sockets */ 2212 return -ENOPROTOOPT; 2213 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2214 val = 8; 2215 else if (val > USHRT_MAX) 2216 val = USHRT_MAX; 2217 up->pcrlen = val; 2218 up->pcflag |= UDPLITE_RECV_CC; 2219 break; 2220 2221 default: 2222 err = -ENOPROTOOPT; 2223 break; 2224 } 2225 2226 return err; 2227 } 2228 EXPORT_SYMBOL(udp_lib_setsockopt); 2229 2230 int udp_setsockopt(struct sock *sk, int level, int optname, 2231 char __user *optval, unsigned int optlen) 2232 { 2233 if (level == SOL_UDP || level == SOL_UDPLITE) 2234 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2235 udp_push_pending_frames); 2236 return ip_setsockopt(sk, level, optname, optval, optlen); 2237 } 2238 2239 #ifdef CONFIG_COMPAT 2240 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2241 char __user *optval, unsigned int optlen) 2242 { 2243 if (level == SOL_UDP || level == SOL_UDPLITE) 2244 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2245 udp_push_pending_frames); 2246 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2247 } 2248 #endif 2249 2250 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2251 char __user *optval, int __user *optlen) 2252 { 2253 struct udp_sock *up = udp_sk(sk); 2254 int val, len; 2255 2256 if (get_user(len, optlen)) 2257 return -EFAULT; 2258 2259 len = min_t(unsigned int, len, sizeof(int)); 2260 2261 if (len < 0) 2262 return -EINVAL; 2263 2264 switch (optname) { 2265 case UDP_CORK: 2266 val = up->corkflag; 2267 break; 2268 2269 case UDP_ENCAP: 2270 val = up->encap_type; 2271 break; 2272 2273 case UDP_NO_CHECK6_TX: 2274 val = up->no_check6_tx; 2275 break; 2276 2277 case UDP_NO_CHECK6_RX: 2278 val = up->no_check6_rx; 2279 break; 2280 2281 /* The following two cannot be changed on UDP sockets, the return is 2282 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2283 case UDPLITE_SEND_CSCOV: 2284 val = up->pcslen; 2285 break; 2286 2287 case UDPLITE_RECV_CSCOV: 2288 val = up->pcrlen; 2289 break; 2290 2291 default: 2292 return -ENOPROTOOPT; 2293 } 2294 2295 if (put_user(len, optlen)) 2296 return -EFAULT; 2297 if (copy_to_user(optval, &val, len)) 2298 return -EFAULT; 2299 return 0; 2300 } 2301 EXPORT_SYMBOL(udp_lib_getsockopt); 2302 2303 int udp_getsockopt(struct sock *sk, int level, int optname, 2304 char __user *optval, int __user *optlen) 2305 { 2306 if (level == SOL_UDP || level == SOL_UDPLITE) 2307 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2308 return ip_getsockopt(sk, level, optname, optval, optlen); 2309 } 2310 2311 #ifdef CONFIG_COMPAT 2312 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2313 char __user *optval, int __user *optlen) 2314 { 2315 if (level == SOL_UDP || level == SOL_UDPLITE) 2316 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2317 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2318 } 2319 #endif 2320 /** 2321 * udp_poll - wait for a UDP event. 2322 * @file - file struct 2323 * @sock - socket 2324 * @wait - poll table 2325 * 2326 * This is same as datagram poll, except for the special case of 2327 * blocking sockets. If application is using a blocking fd 2328 * and a packet with checksum error is in the queue; 2329 * then it could get return from select indicating data available 2330 * but then block when reading it. Add special case code 2331 * to work around these arguably broken applications. 2332 */ 2333 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2334 { 2335 unsigned int mask = datagram_poll(file, sock, wait); 2336 struct sock *sk = sock->sk; 2337 2338 sock_rps_record_flow(sk); 2339 2340 /* Check for false positives due to checksum errors */ 2341 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2342 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2343 mask &= ~(POLLIN | POLLRDNORM); 2344 2345 return mask; 2346 2347 } 2348 EXPORT_SYMBOL(udp_poll); 2349 2350 struct proto udp_prot = { 2351 .name = "UDP", 2352 .owner = THIS_MODULE, 2353 .close = udp_lib_close, 2354 .connect = ip4_datagram_connect, 2355 .disconnect = udp_disconnect, 2356 .ioctl = udp_ioctl, 2357 .destroy = udp_destroy_sock, 2358 .setsockopt = udp_setsockopt, 2359 .getsockopt = udp_getsockopt, 2360 .sendmsg = udp_sendmsg, 2361 .recvmsg = udp_recvmsg, 2362 .sendpage = udp_sendpage, 2363 .backlog_rcv = __udp_queue_rcv_skb, 2364 .release_cb = ip4_datagram_release_cb, 2365 .hash = udp_lib_hash, 2366 .unhash = udp_lib_unhash, 2367 .rehash = udp_v4_rehash, 2368 .get_port = udp_v4_get_port, 2369 .memory_allocated = &udp_memory_allocated, 2370 .sysctl_mem = sysctl_udp_mem, 2371 .sysctl_wmem = &sysctl_udp_wmem_min, 2372 .sysctl_rmem = &sysctl_udp_rmem_min, 2373 .obj_size = sizeof(struct udp_sock), 2374 .slab_flags = SLAB_DESTROY_BY_RCU, 2375 .h.udp_table = &udp_table, 2376 #ifdef CONFIG_COMPAT 2377 .compat_setsockopt = compat_udp_setsockopt, 2378 .compat_getsockopt = compat_udp_getsockopt, 2379 #endif 2380 .clear_sk = sk_prot_clear_portaddr_nulls, 2381 }; 2382 EXPORT_SYMBOL(udp_prot); 2383 2384 /* ------------------------------------------------------------------------ */ 2385 #ifdef CONFIG_PROC_FS 2386 2387 static struct sock *udp_get_first(struct seq_file *seq, int start) 2388 { 2389 struct sock *sk; 2390 struct udp_iter_state *state = seq->private; 2391 struct net *net = seq_file_net(seq); 2392 2393 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2394 ++state->bucket) { 2395 struct hlist_nulls_node *node; 2396 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2397 2398 if (hlist_nulls_empty(&hslot->head)) 2399 continue; 2400 2401 spin_lock_bh(&hslot->lock); 2402 sk_nulls_for_each(sk, node, &hslot->head) { 2403 if (!net_eq(sock_net(sk), net)) 2404 continue; 2405 if (sk->sk_family == state->family) 2406 goto found; 2407 } 2408 spin_unlock_bh(&hslot->lock); 2409 } 2410 sk = NULL; 2411 found: 2412 return sk; 2413 } 2414 2415 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2416 { 2417 struct udp_iter_state *state = seq->private; 2418 struct net *net = seq_file_net(seq); 2419 2420 do { 2421 sk = sk_nulls_next(sk); 2422 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2423 2424 if (!sk) { 2425 if (state->bucket <= state->udp_table->mask) 2426 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2427 return udp_get_first(seq, state->bucket + 1); 2428 } 2429 return sk; 2430 } 2431 2432 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2433 { 2434 struct sock *sk = udp_get_first(seq, 0); 2435 2436 if (sk) 2437 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2438 --pos; 2439 return pos ? NULL : sk; 2440 } 2441 2442 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2443 { 2444 struct udp_iter_state *state = seq->private; 2445 state->bucket = MAX_UDP_PORTS; 2446 2447 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2448 } 2449 2450 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2451 { 2452 struct sock *sk; 2453 2454 if (v == SEQ_START_TOKEN) 2455 sk = udp_get_idx(seq, 0); 2456 else 2457 sk = udp_get_next(seq, v); 2458 2459 ++*pos; 2460 return sk; 2461 } 2462 2463 static void udp_seq_stop(struct seq_file *seq, void *v) 2464 { 2465 struct udp_iter_state *state = seq->private; 2466 2467 if (state->bucket <= state->udp_table->mask) 2468 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2469 } 2470 2471 int udp_seq_open(struct inode *inode, struct file *file) 2472 { 2473 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2474 struct udp_iter_state *s; 2475 int err; 2476 2477 err = seq_open_net(inode, file, &afinfo->seq_ops, 2478 sizeof(struct udp_iter_state)); 2479 if (err < 0) 2480 return err; 2481 2482 s = ((struct seq_file *)file->private_data)->private; 2483 s->family = afinfo->family; 2484 s->udp_table = afinfo->udp_table; 2485 return err; 2486 } 2487 EXPORT_SYMBOL(udp_seq_open); 2488 2489 /* ------------------------------------------------------------------------ */ 2490 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2491 { 2492 struct proc_dir_entry *p; 2493 int rc = 0; 2494 2495 afinfo->seq_ops.start = udp_seq_start; 2496 afinfo->seq_ops.next = udp_seq_next; 2497 afinfo->seq_ops.stop = udp_seq_stop; 2498 2499 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2500 afinfo->seq_fops, afinfo); 2501 if (!p) 2502 rc = -ENOMEM; 2503 return rc; 2504 } 2505 EXPORT_SYMBOL(udp_proc_register); 2506 2507 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2508 { 2509 remove_proc_entry(afinfo->name, net->proc_net); 2510 } 2511 EXPORT_SYMBOL(udp_proc_unregister); 2512 2513 /* ------------------------------------------------------------------------ */ 2514 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2515 int bucket) 2516 { 2517 struct inet_sock *inet = inet_sk(sp); 2518 __be32 dest = inet->inet_daddr; 2519 __be32 src = inet->inet_rcv_saddr; 2520 __u16 destp = ntohs(inet->inet_dport); 2521 __u16 srcp = ntohs(inet->inet_sport); 2522 2523 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2524 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2525 bucket, src, srcp, dest, destp, sp->sk_state, 2526 sk_wmem_alloc_get(sp), 2527 sk_rmem_alloc_get(sp), 2528 0, 0L, 0, 2529 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2530 0, sock_i_ino(sp), 2531 atomic_read(&sp->sk_refcnt), sp, 2532 atomic_read(&sp->sk_drops)); 2533 } 2534 2535 int udp4_seq_show(struct seq_file *seq, void *v) 2536 { 2537 seq_setwidth(seq, 127); 2538 if (v == SEQ_START_TOKEN) 2539 seq_puts(seq, " sl local_address rem_address st tx_queue " 2540 "rx_queue tr tm->when retrnsmt uid timeout " 2541 "inode ref pointer drops"); 2542 else { 2543 struct udp_iter_state *state = seq->private; 2544 2545 udp4_format_sock(v, seq, state->bucket); 2546 } 2547 seq_pad(seq, '\n'); 2548 return 0; 2549 } 2550 2551 static const struct file_operations udp_afinfo_seq_fops = { 2552 .owner = THIS_MODULE, 2553 .open = udp_seq_open, 2554 .read = seq_read, 2555 .llseek = seq_lseek, 2556 .release = seq_release_net 2557 }; 2558 2559 /* ------------------------------------------------------------------------ */ 2560 static struct udp_seq_afinfo udp4_seq_afinfo = { 2561 .name = "udp", 2562 .family = AF_INET, 2563 .udp_table = &udp_table, 2564 .seq_fops = &udp_afinfo_seq_fops, 2565 .seq_ops = { 2566 .show = udp4_seq_show, 2567 }, 2568 }; 2569 2570 static int __net_init udp4_proc_init_net(struct net *net) 2571 { 2572 return udp_proc_register(net, &udp4_seq_afinfo); 2573 } 2574 2575 static void __net_exit udp4_proc_exit_net(struct net *net) 2576 { 2577 udp_proc_unregister(net, &udp4_seq_afinfo); 2578 } 2579 2580 static struct pernet_operations udp4_net_ops = { 2581 .init = udp4_proc_init_net, 2582 .exit = udp4_proc_exit_net, 2583 }; 2584 2585 int __init udp4_proc_init(void) 2586 { 2587 return register_pernet_subsys(&udp4_net_ops); 2588 } 2589 2590 void udp4_proc_exit(void) 2591 { 2592 unregister_pernet_subsys(&udp4_net_ops); 2593 } 2594 #endif /* CONFIG_PROC_FS */ 2595 2596 static __initdata unsigned long uhash_entries; 2597 static int __init set_uhash_entries(char *str) 2598 { 2599 ssize_t ret; 2600 2601 if (!str) 2602 return 0; 2603 2604 ret = kstrtoul(str, 0, &uhash_entries); 2605 if (ret) 2606 return 0; 2607 2608 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2609 uhash_entries = UDP_HTABLE_SIZE_MIN; 2610 return 1; 2611 } 2612 __setup("uhash_entries=", set_uhash_entries); 2613 2614 void __init udp_table_init(struct udp_table *table, const char *name) 2615 { 2616 unsigned int i; 2617 2618 table->hash = alloc_large_system_hash(name, 2619 2 * sizeof(struct udp_hslot), 2620 uhash_entries, 2621 21, /* one slot per 2 MB */ 2622 0, 2623 &table->log, 2624 &table->mask, 2625 UDP_HTABLE_SIZE_MIN, 2626 64 * 1024); 2627 2628 table->hash2 = table->hash + (table->mask + 1); 2629 for (i = 0; i <= table->mask; i++) { 2630 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2631 table->hash[i].count = 0; 2632 spin_lock_init(&table->hash[i].lock); 2633 } 2634 for (i = 0; i <= table->mask; i++) { 2635 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2636 table->hash2[i].count = 0; 2637 spin_lock_init(&table->hash2[i].lock); 2638 } 2639 } 2640 2641 u32 udp_flow_hashrnd(void) 2642 { 2643 static u32 hashrnd __read_mostly; 2644 2645 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2646 2647 return hashrnd; 2648 } 2649 EXPORT_SYMBOL(udp_flow_hashrnd); 2650 2651 void __init udp_init(void) 2652 { 2653 unsigned long limit; 2654 2655 udp_table_init(&udp_table, "UDP"); 2656 limit = nr_free_buffer_pages() / 8; 2657 limit = max(limit, 128UL); 2658 sysctl_udp_mem[0] = limit / 4 * 3; 2659 sysctl_udp_mem[1] = limit; 2660 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2661 2662 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2663 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2664 } 2665