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 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 343 &hslot2->head); 344 hslot2->count++; 345 spin_unlock(&hslot2->lock); 346 } 347 error = 0; 348 fail_unlock: 349 spin_unlock_bh(&hslot->lock); 350 fail: 351 return error; 352 } 353 EXPORT_SYMBOL(udp_lib_get_port); 354 355 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses 356 * match_wildcard == false: addresses must be exactly the same, i.e. 357 * 0.0.0.0 only equals to 0.0.0.0 358 */ 359 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2, 360 bool match_wildcard) 361 { 362 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 363 364 if (!ipv6_only_sock(sk2)) { 365 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr) 366 return 1; 367 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr) 368 return match_wildcard; 369 } 370 return 0; 371 } 372 373 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, 374 unsigned int port) 375 { 376 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 377 } 378 379 int udp_v4_get_port(struct sock *sk, unsigned short snum) 380 { 381 unsigned int hash2_nulladdr = 382 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 383 unsigned int hash2_partial = 384 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 385 386 /* precompute partial secondary hash */ 387 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 388 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 389 } 390 391 static inline int compute_score(struct sock *sk, struct net *net, 392 __be32 saddr, unsigned short hnum, __be16 sport, 393 __be32 daddr, __be16 dport, int dif) 394 { 395 int score; 396 struct inet_sock *inet; 397 398 if (!net_eq(sock_net(sk), net) || 399 udp_sk(sk)->udp_port_hash != hnum || 400 ipv6_only_sock(sk)) 401 return -1; 402 403 score = (sk->sk_family == PF_INET) ? 2 : 1; 404 inet = inet_sk(sk); 405 406 if (inet->inet_rcv_saddr) { 407 if (inet->inet_rcv_saddr != daddr) 408 return -1; 409 score += 4; 410 } 411 412 if (inet->inet_daddr) { 413 if (inet->inet_daddr != saddr) 414 return -1; 415 score += 4; 416 } 417 418 if (inet->inet_dport) { 419 if (inet->inet_dport != sport) 420 return -1; 421 score += 4; 422 } 423 424 if (sk->sk_bound_dev_if) { 425 if (sk->sk_bound_dev_if != dif) 426 return -1; 427 score += 4; 428 } 429 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 430 score++; 431 return score; 432 } 433 434 /* 435 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 436 */ 437 static inline int compute_score2(struct sock *sk, struct net *net, 438 __be32 saddr, __be16 sport, 439 __be32 daddr, unsigned int hnum, int dif) 440 { 441 int score; 442 struct inet_sock *inet; 443 444 if (!net_eq(sock_net(sk), net) || 445 ipv6_only_sock(sk)) 446 return -1; 447 448 inet = inet_sk(sk); 449 450 if (inet->inet_rcv_saddr != daddr || 451 inet->inet_num != hnum) 452 return -1; 453 454 score = (sk->sk_family == PF_INET) ? 2 : 1; 455 456 if (inet->inet_daddr) { 457 if (inet->inet_daddr != saddr) 458 return -1; 459 score += 4; 460 } 461 462 if (inet->inet_dport) { 463 if (inet->inet_dport != sport) 464 return -1; 465 score += 4; 466 } 467 468 if (sk->sk_bound_dev_if) { 469 if (sk->sk_bound_dev_if != dif) 470 return -1; 471 score += 4; 472 } 473 474 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 475 score++; 476 477 return score; 478 } 479 480 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 481 const __u16 lport, const __be32 faddr, 482 const __be16 fport) 483 { 484 static u32 udp_ehash_secret __read_mostly; 485 486 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 487 488 return __inet_ehashfn(laddr, lport, faddr, fport, 489 udp_ehash_secret + net_hash_mix(net)); 490 } 491 492 /* called with read_rcu_lock() */ 493 static struct sock *udp4_lib_lookup2(struct net *net, 494 __be32 saddr, __be16 sport, 495 __be32 daddr, unsigned int hnum, int dif, 496 struct udp_hslot *hslot2, unsigned int slot2, 497 struct sk_buff *skb) 498 { 499 struct sock *sk, *result; 500 struct hlist_nulls_node *node; 501 int score, badness, matches = 0, reuseport = 0; 502 bool select_ok = true; 503 u32 hash = 0; 504 505 begin: 506 result = NULL; 507 badness = 0; 508 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 509 score = compute_score2(sk, net, saddr, sport, 510 daddr, hnum, dif); 511 if (score > badness) { 512 result = sk; 513 badness = score; 514 reuseport = sk->sk_reuseport; 515 if (reuseport) { 516 hash = udp_ehashfn(net, daddr, hnum, 517 saddr, sport); 518 if (select_ok) { 519 struct sock *sk2; 520 521 sk2 = reuseport_select_sock(sk, hash, skb, 522 sizeof(struct udphdr)); 523 if (sk2) { 524 result = sk2; 525 select_ok = false; 526 goto found; 527 } 528 } 529 matches = 1; 530 } 531 } else if (score == badness && reuseport) { 532 matches++; 533 if (reciprocal_scale(hash, matches) == 0) 534 result = sk; 535 hash = next_pseudo_random32(hash); 536 } 537 } 538 /* 539 * if the nulls value we got at the end of this lookup is 540 * not the expected one, we must restart lookup. 541 * We probably met an item that was moved to another chain. 542 */ 543 if (get_nulls_value(node) != slot2) 544 goto begin; 545 if (result) { 546 found: 547 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 548 result = NULL; 549 else if (unlikely(compute_score2(result, net, saddr, sport, 550 daddr, hnum, dif) < badness)) { 551 sock_put(result); 552 goto begin; 553 } 554 } 555 return result; 556 } 557 558 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 559 * harder than this. -DaveM 560 */ 561 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 562 __be16 sport, __be32 daddr, __be16 dport, 563 int dif, struct udp_table *udptable, struct sk_buff *skb) 564 { 565 struct sock *sk, *result; 566 struct hlist_nulls_node *node; 567 unsigned short hnum = ntohs(dport); 568 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 569 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 570 int score, badness, matches = 0, reuseport = 0; 571 bool select_ok = true; 572 u32 hash = 0; 573 574 rcu_read_lock(); 575 if (hslot->count > 10) { 576 hash2 = udp4_portaddr_hash(net, daddr, hnum); 577 slot2 = hash2 & udptable->mask; 578 hslot2 = &udptable->hash2[slot2]; 579 if (hslot->count < hslot2->count) 580 goto begin; 581 582 result = udp4_lib_lookup2(net, saddr, sport, 583 daddr, hnum, dif, 584 hslot2, slot2, skb); 585 if (!result) { 586 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 587 slot2 = hash2 & udptable->mask; 588 hslot2 = &udptable->hash2[slot2]; 589 if (hslot->count < hslot2->count) 590 goto begin; 591 592 result = udp4_lib_lookup2(net, saddr, sport, 593 htonl(INADDR_ANY), hnum, dif, 594 hslot2, slot2, skb); 595 } 596 rcu_read_unlock(); 597 return result; 598 } 599 begin: 600 result = NULL; 601 badness = 0; 602 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 603 score = compute_score(sk, net, saddr, hnum, sport, 604 daddr, dport, dif); 605 if (score > badness) { 606 result = sk; 607 badness = score; 608 reuseport = sk->sk_reuseport; 609 if (reuseport) { 610 hash = udp_ehashfn(net, daddr, hnum, 611 saddr, sport); 612 if (select_ok) { 613 struct sock *sk2; 614 615 sk2 = reuseport_select_sock(sk, hash, skb, 616 sizeof(struct udphdr)); 617 if (sk2) { 618 result = sk2; 619 select_ok = false; 620 goto found; 621 } 622 } 623 matches = 1; 624 } 625 } else if (score == badness && reuseport) { 626 matches++; 627 if (reciprocal_scale(hash, matches) == 0) 628 result = sk; 629 hash = next_pseudo_random32(hash); 630 } 631 } 632 /* 633 * if the nulls value we got at the end of this lookup is 634 * not the expected one, we must restart lookup. 635 * We probably met an item that was moved to another chain. 636 */ 637 if (get_nulls_value(node) != slot) 638 goto begin; 639 640 if (result) { 641 found: 642 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 643 result = NULL; 644 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 645 daddr, dport, dif) < badness)) { 646 sock_put(result); 647 goto begin; 648 } 649 } 650 rcu_read_unlock(); 651 return result; 652 } 653 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 654 655 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 656 __be16 sport, __be16 dport, 657 struct udp_table *udptable) 658 { 659 const struct iphdr *iph = ip_hdr(skb); 660 661 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 662 iph->daddr, dport, inet_iif(skb), 663 udptable, skb); 664 } 665 666 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 667 __be32 daddr, __be16 dport, int dif) 668 { 669 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, 670 &udp_table, NULL); 671 } 672 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 673 674 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 675 __be16 loc_port, __be32 loc_addr, 676 __be16 rmt_port, __be32 rmt_addr, 677 int dif, unsigned short hnum) 678 { 679 struct inet_sock *inet = inet_sk(sk); 680 681 if (!net_eq(sock_net(sk), net) || 682 udp_sk(sk)->udp_port_hash != hnum || 683 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 684 (inet->inet_dport != rmt_port && inet->inet_dport) || 685 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 686 ipv6_only_sock(sk) || 687 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 688 return false; 689 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 690 return false; 691 return true; 692 } 693 694 /* 695 * This routine is called by the ICMP module when it gets some 696 * sort of error condition. If err < 0 then the socket should 697 * be closed and the error returned to the user. If err > 0 698 * it's just the icmp type << 8 | icmp code. 699 * Header points to the ip header of the error packet. We move 700 * on past this. Then (as it used to claim before adjustment) 701 * header points to the first 8 bytes of the udp header. We need 702 * to find the appropriate port. 703 */ 704 705 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 706 { 707 struct inet_sock *inet; 708 const struct iphdr *iph = (const struct iphdr *)skb->data; 709 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 710 const int type = icmp_hdr(skb)->type; 711 const int code = icmp_hdr(skb)->code; 712 struct sock *sk; 713 int harderr; 714 int err; 715 struct net *net = dev_net(skb->dev); 716 717 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 718 iph->saddr, uh->source, skb->dev->ifindex, udptable, 719 NULL); 720 if (!sk) { 721 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 722 return; /* No socket for error */ 723 } 724 725 err = 0; 726 harderr = 0; 727 inet = inet_sk(sk); 728 729 switch (type) { 730 default: 731 case ICMP_TIME_EXCEEDED: 732 err = EHOSTUNREACH; 733 break; 734 case ICMP_SOURCE_QUENCH: 735 goto out; 736 case ICMP_PARAMETERPROB: 737 err = EPROTO; 738 harderr = 1; 739 break; 740 case ICMP_DEST_UNREACH: 741 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 742 ipv4_sk_update_pmtu(skb, sk, info); 743 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 744 err = EMSGSIZE; 745 harderr = 1; 746 break; 747 } 748 goto out; 749 } 750 err = EHOSTUNREACH; 751 if (code <= NR_ICMP_UNREACH) { 752 harderr = icmp_err_convert[code].fatal; 753 err = icmp_err_convert[code].errno; 754 } 755 break; 756 case ICMP_REDIRECT: 757 ipv4_sk_redirect(skb, sk); 758 goto out; 759 } 760 761 /* 762 * RFC1122: OK. Passes ICMP errors back to application, as per 763 * 4.1.3.3. 764 */ 765 if (!inet->recverr) { 766 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 767 goto out; 768 } else 769 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 770 771 sk->sk_err = err; 772 sk->sk_error_report(sk); 773 out: 774 sock_put(sk); 775 } 776 777 void udp_err(struct sk_buff *skb, u32 info) 778 { 779 __udp4_lib_err(skb, info, &udp_table); 780 } 781 782 /* 783 * Throw away all pending data and cancel the corking. Socket is locked. 784 */ 785 void udp_flush_pending_frames(struct sock *sk) 786 { 787 struct udp_sock *up = udp_sk(sk); 788 789 if (up->pending) { 790 up->len = 0; 791 up->pending = 0; 792 ip_flush_pending_frames(sk); 793 } 794 } 795 EXPORT_SYMBOL(udp_flush_pending_frames); 796 797 /** 798 * udp4_hwcsum - handle outgoing HW checksumming 799 * @skb: sk_buff containing the filled-in UDP header 800 * (checksum field must be zeroed out) 801 * @src: source IP address 802 * @dst: destination IP address 803 */ 804 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 805 { 806 struct udphdr *uh = udp_hdr(skb); 807 int offset = skb_transport_offset(skb); 808 int len = skb->len - offset; 809 int hlen = len; 810 __wsum csum = 0; 811 812 if (!skb_has_frag_list(skb)) { 813 /* 814 * Only one fragment on the socket. 815 */ 816 skb->csum_start = skb_transport_header(skb) - skb->head; 817 skb->csum_offset = offsetof(struct udphdr, check); 818 uh->check = ~csum_tcpudp_magic(src, dst, len, 819 IPPROTO_UDP, 0); 820 } else { 821 struct sk_buff *frags; 822 823 /* 824 * HW-checksum won't work as there are two or more 825 * fragments on the socket so that all csums of sk_buffs 826 * should be together 827 */ 828 skb_walk_frags(skb, frags) { 829 csum = csum_add(csum, frags->csum); 830 hlen -= frags->len; 831 } 832 833 csum = skb_checksum(skb, offset, hlen, csum); 834 skb->ip_summed = CHECKSUM_NONE; 835 836 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 837 if (uh->check == 0) 838 uh->check = CSUM_MANGLED_0; 839 } 840 } 841 EXPORT_SYMBOL_GPL(udp4_hwcsum); 842 843 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 844 * for the simple case like when setting the checksum for a UDP tunnel. 845 */ 846 void udp_set_csum(bool nocheck, struct sk_buff *skb, 847 __be32 saddr, __be32 daddr, int len) 848 { 849 struct udphdr *uh = udp_hdr(skb); 850 851 if (nocheck) 852 uh->check = 0; 853 else if (skb_is_gso(skb)) 854 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 855 else if (skb_dst(skb) && skb_dst(skb)->dev && 856 (skb_dst(skb)->dev->features & 857 (NETIF_F_IP_CSUM | NETIF_F_HW_CSUM))) { 858 859 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 860 861 skb->ip_summed = CHECKSUM_PARTIAL; 862 skb->csum_start = skb_transport_header(skb) - skb->head; 863 skb->csum_offset = offsetof(struct udphdr, check); 864 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 865 } else { 866 __wsum csum; 867 868 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 869 870 uh->check = 0; 871 csum = skb_checksum(skb, 0, len, 0); 872 uh->check = udp_v4_check(len, saddr, daddr, csum); 873 if (uh->check == 0) 874 uh->check = CSUM_MANGLED_0; 875 876 skb->ip_summed = CHECKSUM_UNNECESSARY; 877 } 878 } 879 EXPORT_SYMBOL(udp_set_csum); 880 881 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 882 { 883 struct sock *sk = skb->sk; 884 struct inet_sock *inet = inet_sk(sk); 885 struct udphdr *uh; 886 int err = 0; 887 int is_udplite = IS_UDPLITE(sk); 888 int offset = skb_transport_offset(skb); 889 int len = skb->len - offset; 890 __wsum csum = 0; 891 892 /* 893 * Create a UDP header 894 */ 895 uh = udp_hdr(skb); 896 uh->source = inet->inet_sport; 897 uh->dest = fl4->fl4_dport; 898 uh->len = htons(len); 899 uh->check = 0; 900 901 if (is_udplite) /* UDP-Lite */ 902 csum = udplite_csum(skb); 903 904 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 905 906 skb->ip_summed = CHECKSUM_NONE; 907 goto send; 908 909 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 910 911 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 912 goto send; 913 914 } else 915 csum = udp_csum(skb); 916 917 /* add protocol-dependent pseudo-header */ 918 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 919 sk->sk_protocol, csum); 920 if (uh->check == 0) 921 uh->check = CSUM_MANGLED_0; 922 923 send: 924 err = ip_send_skb(sock_net(sk), skb); 925 if (err) { 926 if (err == -ENOBUFS && !inet->recverr) { 927 UDP_INC_STATS_USER(sock_net(sk), 928 UDP_MIB_SNDBUFERRORS, is_udplite); 929 err = 0; 930 } 931 } else 932 UDP_INC_STATS_USER(sock_net(sk), 933 UDP_MIB_OUTDATAGRAMS, is_udplite); 934 return err; 935 } 936 937 /* 938 * Push out all pending data as one UDP datagram. Socket is locked. 939 */ 940 int udp_push_pending_frames(struct sock *sk) 941 { 942 struct udp_sock *up = udp_sk(sk); 943 struct inet_sock *inet = inet_sk(sk); 944 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 945 struct sk_buff *skb; 946 int err = 0; 947 948 skb = ip_finish_skb(sk, fl4); 949 if (!skb) 950 goto out; 951 952 err = udp_send_skb(skb, fl4); 953 954 out: 955 up->len = 0; 956 up->pending = 0; 957 return err; 958 } 959 EXPORT_SYMBOL(udp_push_pending_frames); 960 961 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 962 { 963 struct inet_sock *inet = inet_sk(sk); 964 struct udp_sock *up = udp_sk(sk); 965 struct flowi4 fl4_stack; 966 struct flowi4 *fl4; 967 int ulen = len; 968 struct ipcm_cookie ipc; 969 struct rtable *rt = NULL; 970 int free = 0; 971 int connected = 0; 972 __be32 daddr, faddr, saddr; 973 __be16 dport; 974 u8 tos; 975 int err, is_udplite = IS_UDPLITE(sk); 976 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 977 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 978 struct sk_buff *skb; 979 struct ip_options_data opt_copy; 980 981 if (len > 0xFFFF) 982 return -EMSGSIZE; 983 984 /* 985 * Check the flags. 986 */ 987 988 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 989 return -EOPNOTSUPP; 990 991 ipc.opt = NULL; 992 ipc.tx_flags = 0; 993 ipc.ttl = 0; 994 ipc.tos = -1; 995 996 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 997 998 fl4 = &inet->cork.fl.u.ip4; 999 if (up->pending) { 1000 /* 1001 * There are pending frames. 1002 * The socket lock must be held while it's corked. 1003 */ 1004 lock_sock(sk); 1005 if (likely(up->pending)) { 1006 if (unlikely(up->pending != AF_INET)) { 1007 release_sock(sk); 1008 return -EINVAL; 1009 } 1010 goto do_append_data; 1011 } 1012 release_sock(sk); 1013 } 1014 ulen += sizeof(struct udphdr); 1015 1016 /* 1017 * Get and verify the address. 1018 */ 1019 if (msg->msg_name) { 1020 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1021 if (msg->msg_namelen < sizeof(*usin)) 1022 return -EINVAL; 1023 if (usin->sin_family != AF_INET) { 1024 if (usin->sin_family != AF_UNSPEC) 1025 return -EAFNOSUPPORT; 1026 } 1027 1028 daddr = usin->sin_addr.s_addr; 1029 dport = usin->sin_port; 1030 if (dport == 0) 1031 return -EINVAL; 1032 } else { 1033 if (sk->sk_state != TCP_ESTABLISHED) 1034 return -EDESTADDRREQ; 1035 daddr = inet->inet_daddr; 1036 dport = inet->inet_dport; 1037 /* Open fast path for connected socket. 1038 Route will not be used, if at least one option is set. 1039 */ 1040 connected = 1; 1041 } 1042 ipc.addr = inet->inet_saddr; 1043 1044 ipc.oif = sk->sk_bound_dev_if; 1045 1046 sock_tx_timestamp(sk, &ipc.tx_flags); 1047 1048 if (msg->msg_controllen) { 1049 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 1050 sk->sk_family == AF_INET6); 1051 if (err) 1052 return err; 1053 if (ipc.opt) 1054 free = 1; 1055 connected = 0; 1056 } 1057 if (!ipc.opt) { 1058 struct ip_options_rcu *inet_opt; 1059 1060 rcu_read_lock(); 1061 inet_opt = rcu_dereference(inet->inet_opt); 1062 if (inet_opt) { 1063 memcpy(&opt_copy, inet_opt, 1064 sizeof(*inet_opt) + inet_opt->opt.optlen); 1065 ipc.opt = &opt_copy.opt; 1066 } 1067 rcu_read_unlock(); 1068 } 1069 1070 saddr = ipc.addr; 1071 ipc.addr = faddr = daddr; 1072 1073 if (ipc.opt && ipc.opt->opt.srr) { 1074 if (!daddr) 1075 return -EINVAL; 1076 faddr = ipc.opt->opt.faddr; 1077 connected = 0; 1078 } 1079 tos = get_rttos(&ipc, inet); 1080 if (sock_flag(sk, SOCK_LOCALROUTE) || 1081 (msg->msg_flags & MSG_DONTROUTE) || 1082 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1083 tos |= RTO_ONLINK; 1084 connected = 0; 1085 } 1086 1087 if (ipv4_is_multicast(daddr)) { 1088 if (!ipc.oif) 1089 ipc.oif = inet->mc_index; 1090 if (!saddr) 1091 saddr = inet->mc_addr; 1092 connected = 0; 1093 } else if (!ipc.oif) 1094 ipc.oif = inet->uc_index; 1095 1096 if (connected) 1097 rt = (struct rtable *)sk_dst_check(sk, 0); 1098 1099 if (!rt) { 1100 struct net *net = sock_net(sk); 1101 __u8 flow_flags = inet_sk_flowi_flags(sk); 1102 1103 fl4 = &fl4_stack; 1104 1105 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1106 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1107 flow_flags, 1108 faddr, saddr, dport, inet->inet_sport); 1109 1110 if (!saddr && ipc.oif) { 1111 err = l3mdev_get_saddr(net, ipc.oif, fl4); 1112 if (err < 0) 1113 goto out; 1114 } 1115 1116 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1117 rt = ip_route_output_flow(net, fl4, sk); 1118 if (IS_ERR(rt)) { 1119 err = PTR_ERR(rt); 1120 rt = NULL; 1121 if (err == -ENETUNREACH) 1122 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1123 goto out; 1124 } 1125 1126 err = -EACCES; 1127 if ((rt->rt_flags & RTCF_BROADCAST) && 1128 !sock_flag(sk, SOCK_BROADCAST)) 1129 goto out; 1130 if (connected) 1131 sk_dst_set(sk, dst_clone(&rt->dst)); 1132 } 1133 1134 if (msg->msg_flags&MSG_CONFIRM) 1135 goto do_confirm; 1136 back_from_confirm: 1137 1138 saddr = fl4->saddr; 1139 if (!ipc.addr) 1140 daddr = ipc.addr = fl4->daddr; 1141 1142 /* Lockless fast path for the non-corking case. */ 1143 if (!corkreq) { 1144 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1145 sizeof(struct udphdr), &ipc, &rt, 1146 msg->msg_flags); 1147 err = PTR_ERR(skb); 1148 if (!IS_ERR_OR_NULL(skb)) 1149 err = udp_send_skb(skb, fl4); 1150 goto out; 1151 } 1152 1153 lock_sock(sk); 1154 if (unlikely(up->pending)) { 1155 /* The socket is already corked while preparing it. */ 1156 /* ... which is an evident application bug. --ANK */ 1157 release_sock(sk); 1158 1159 net_dbg_ratelimited("cork app bug 2\n"); 1160 err = -EINVAL; 1161 goto out; 1162 } 1163 /* 1164 * Now cork the socket to pend data. 1165 */ 1166 fl4 = &inet->cork.fl.u.ip4; 1167 fl4->daddr = daddr; 1168 fl4->saddr = saddr; 1169 fl4->fl4_dport = dport; 1170 fl4->fl4_sport = inet->inet_sport; 1171 up->pending = AF_INET; 1172 1173 do_append_data: 1174 up->len += ulen; 1175 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1176 sizeof(struct udphdr), &ipc, &rt, 1177 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1178 if (err) 1179 udp_flush_pending_frames(sk); 1180 else if (!corkreq) 1181 err = udp_push_pending_frames(sk); 1182 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1183 up->pending = 0; 1184 release_sock(sk); 1185 1186 out: 1187 ip_rt_put(rt); 1188 if (free) 1189 kfree(ipc.opt); 1190 if (!err) 1191 return len; 1192 /* 1193 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1194 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1195 * we don't have a good statistic (IpOutDiscards but it can be too many 1196 * things). We could add another new stat but at least for now that 1197 * seems like overkill. 1198 */ 1199 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1200 UDP_INC_STATS_USER(sock_net(sk), 1201 UDP_MIB_SNDBUFERRORS, is_udplite); 1202 } 1203 return err; 1204 1205 do_confirm: 1206 dst_confirm(&rt->dst); 1207 if (!(msg->msg_flags&MSG_PROBE) || len) 1208 goto back_from_confirm; 1209 err = 0; 1210 goto out; 1211 } 1212 EXPORT_SYMBOL(udp_sendmsg); 1213 1214 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1215 size_t size, int flags) 1216 { 1217 struct inet_sock *inet = inet_sk(sk); 1218 struct udp_sock *up = udp_sk(sk); 1219 int ret; 1220 1221 if (flags & MSG_SENDPAGE_NOTLAST) 1222 flags |= MSG_MORE; 1223 1224 if (!up->pending) { 1225 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1226 1227 /* Call udp_sendmsg to specify destination address which 1228 * sendpage interface can't pass. 1229 * This will succeed only when the socket is connected. 1230 */ 1231 ret = udp_sendmsg(sk, &msg, 0); 1232 if (ret < 0) 1233 return ret; 1234 } 1235 1236 lock_sock(sk); 1237 1238 if (unlikely(!up->pending)) { 1239 release_sock(sk); 1240 1241 net_dbg_ratelimited("udp cork app bug 3\n"); 1242 return -EINVAL; 1243 } 1244 1245 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1246 page, offset, size, flags); 1247 if (ret == -EOPNOTSUPP) { 1248 release_sock(sk); 1249 return sock_no_sendpage(sk->sk_socket, page, offset, 1250 size, flags); 1251 } 1252 if (ret < 0) { 1253 udp_flush_pending_frames(sk); 1254 goto out; 1255 } 1256 1257 up->len += size; 1258 if (!(up->corkflag || (flags&MSG_MORE))) 1259 ret = udp_push_pending_frames(sk); 1260 if (!ret) 1261 ret = size; 1262 out: 1263 release_sock(sk); 1264 return ret; 1265 } 1266 1267 /** 1268 * first_packet_length - return length of first packet in receive queue 1269 * @sk: socket 1270 * 1271 * Drops all bad checksum frames, until a valid one is found. 1272 * Returns the length of found skb, or 0 if none is found. 1273 */ 1274 static unsigned int first_packet_length(struct sock *sk) 1275 { 1276 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1277 struct sk_buff *skb; 1278 unsigned int res; 1279 1280 __skb_queue_head_init(&list_kill); 1281 1282 spin_lock_bh(&rcvq->lock); 1283 while ((skb = skb_peek(rcvq)) != NULL && 1284 udp_lib_checksum_complete(skb)) { 1285 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1286 IS_UDPLITE(sk)); 1287 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1288 IS_UDPLITE(sk)); 1289 atomic_inc(&sk->sk_drops); 1290 __skb_unlink(skb, rcvq); 1291 __skb_queue_tail(&list_kill, skb); 1292 } 1293 res = skb ? skb->len : 0; 1294 spin_unlock_bh(&rcvq->lock); 1295 1296 if (!skb_queue_empty(&list_kill)) { 1297 bool slow = lock_sock_fast(sk); 1298 1299 __skb_queue_purge(&list_kill); 1300 sk_mem_reclaim_partial(sk); 1301 unlock_sock_fast(sk, slow); 1302 } 1303 return res; 1304 } 1305 1306 /* 1307 * IOCTL requests applicable to the UDP protocol 1308 */ 1309 1310 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1311 { 1312 switch (cmd) { 1313 case SIOCOUTQ: 1314 { 1315 int amount = sk_wmem_alloc_get(sk); 1316 1317 return put_user(amount, (int __user *)arg); 1318 } 1319 1320 case SIOCINQ: 1321 { 1322 unsigned int amount = first_packet_length(sk); 1323 1324 if (amount) 1325 /* 1326 * We will only return the amount 1327 * of this packet since that is all 1328 * that will be read. 1329 */ 1330 amount -= sizeof(struct udphdr); 1331 1332 return put_user(amount, (int __user *)arg); 1333 } 1334 1335 default: 1336 return -ENOIOCTLCMD; 1337 } 1338 1339 return 0; 1340 } 1341 EXPORT_SYMBOL(udp_ioctl); 1342 1343 /* 1344 * This should be easy, if there is something there we 1345 * return it, otherwise we block. 1346 */ 1347 1348 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1349 int flags, int *addr_len) 1350 { 1351 struct inet_sock *inet = inet_sk(sk); 1352 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1353 struct sk_buff *skb; 1354 unsigned int ulen, copied; 1355 int peeked, off = 0; 1356 int err; 1357 int is_udplite = IS_UDPLITE(sk); 1358 bool checksum_valid = false; 1359 bool slow; 1360 1361 if (flags & MSG_ERRQUEUE) 1362 return ip_recv_error(sk, msg, len, addr_len); 1363 1364 try_again: 1365 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1366 &peeked, &off, &err); 1367 if (!skb) 1368 goto out; 1369 1370 ulen = skb->len - sizeof(struct udphdr); 1371 copied = len; 1372 if (copied > ulen) 1373 copied = ulen; 1374 else if (copied < ulen) 1375 msg->msg_flags |= MSG_TRUNC; 1376 1377 /* 1378 * If checksum is needed at all, try to do it while copying the 1379 * data. If the data is truncated, or if we only want a partial 1380 * coverage checksum (UDP-Lite), do it before the copy. 1381 */ 1382 1383 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1384 checksum_valid = !udp_lib_checksum_complete(skb); 1385 if (!checksum_valid) 1386 goto csum_copy_err; 1387 } 1388 1389 if (checksum_valid || skb_csum_unnecessary(skb)) 1390 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 1391 msg, copied); 1392 else { 1393 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), 1394 msg); 1395 1396 if (err == -EINVAL) 1397 goto csum_copy_err; 1398 } 1399 1400 if (unlikely(err)) { 1401 trace_kfree_skb(skb, udp_recvmsg); 1402 if (!peeked) { 1403 atomic_inc(&sk->sk_drops); 1404 UDP_INC_STATS_USER(sock_net(sk), 1405 UDP_MIB_INERRORS, is_udplite); 1406 } 1407 goto out_free; 1408 } 1409 1410 if (!peeked) 1411 UDP_INC_STATS_USER(sock_net(sk), 1412 UDP_MIB_INDATAGRAMS, is_udplite); 1413 1414 sock_recv_ts_and_drops(msg, sk, skb); 1415 1416 /* Copy the address. */ 1417 if (sin) { 1418 sin->sin_family = AF_INET; 1419 sin->sin_port = udp_hdr(skb)->source; 1420 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1421 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1422 *addr_len = sizeof(*sin); 1423 } 1424 if (inet->cmsg_flags) 1425 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr)); 1426 1427 err = copied; 1428 if (flags & MSG_TRUNC) 1429 err = ulen; 1430 1431 out_free: 1432 skb_free_datagram_locked(sk, skb); 1433 out: 1434 return err; 1435 1436 csum_copy_err: 1437 slow = lock_sock_fast(sk); 1438 if (!skb_kill_datagram(sk, skb, flags)) { 1439 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1440 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1441 } 1442 unlock_sock_fast(sk, slow); 1443 1444 /* starting over for a new packet, but check if we need to yield */ 1445 cond_resched(); 1446 msg->msg_flags &= ~MSG_TRUNC; 1447 goto try_again; 1448 } 1449 1450 int udp_disconnect(struct sock *sk, int flags) 1451 { 1452 struct inet_sock *inet = inet_sk(sk); 1453 /* 1454 * 1003.1g - break association. 1455 */ 1456 1457 sk->sk_state = TCP_CLOSE; 1458 inet->inet_daddr = 0; 1459 inet->inet_dport = 0; 1460 sock_rps_reset_rxhash(sk); 1461 sk->sk_bound_dev_if = 0; 1462 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1463 inet_reset_saddr(sk); 1464 1465 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1466 sk->sk_prot->unhash(sk); 1467 inet->inet_sport = 0; 1468 } 1469 sk_dst_reset(sk); 1470 return 0; 1471 } 1472 EXPORT_SYMBOL(udp_disconnect); 1473 1474 void udp_lib_unhash(struct sock *sk) 1475 { 1476 if (sk_hashed(sk)) { 1477 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1478 struct udp_hslot *hslot, *hslot2; 1479 1480 hslot = udp_hashslot(udptable, sock_net(sk), 1481 udp_sk(sk)->udp_port_hash); 1482 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1483 1484 spin_lock_bh(&hslot->lock); 1485 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1486 reuseport_detach_sock(sk); 1487 if (sk_nulls_del_node_init_rcu(sk)) { 1488 hslot->count--; 1489 inet_sk(sk)->inet_num = 0; 1490 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1491 1492 spin_lock(&hslot2->lock); 1493 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1494 hslot2->count--; 1495 spin_unlock(&hslot2->lock); 1496 } 1497 spin_unlock_bh(&hslot->lock); 1498 } 1499 } 1500 EXPORT_SYMBOL(udp_lib_unhash); 1501 1502 /* 1503 * inet_rcv_saddr was changed, we must rehash secondary hash 1504 */ 1505 void udp_lib_rehash(struct sock *sk, u16 newhash) 1506 { 1507 if (sk_hashed(sk)) { 1508 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1509 struct udp_hslot *hslot, *hslot2, *nhslot2; 1510 1511 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1512 nhslot2 = udp_hashslot2(udptable, newhash); 1513 udp_sk(sk)->udp_portaddr_hash = newhash; 1514 1515 if (hslot2 != nhslot2 || 1516 rcu_access_pointer(sk->sk_reuseport_cb)) { 1517 hslot = udp_hashslot(udptable, sock_net(sk), 1518 udp_sk(sk)->udp_port_hash); 1519 /* we must lock primary chain too */ 1520 spin_lock_bh(&hslot->lock); 1521 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1522 reuseport_detach_sock(sk); 1523 1524 if (hslot2 != nhslot2) { 1525 spin_lock(&hslot2->lock); 1526 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1527 hslot2->count--; 1528 spin_unlock(&hslot2->lock); 1529 1530 spin_lock(&nhslot2->lock); 1531 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1532 &nhslot2->head); 1533 nhslot2->count++; 1534 spin_unlock(&nhslot2->lock); 1535 } 1536 1537 spin_unlock_bh(&hslot->lock); 1538 } 1539 } 1540 } 1541 EXPORT_SYMBOL(udp_lib_rehash); 1542 1543 static void udp_v4_rehash(struct sock *sk) 1544 { 1545 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1546 inet_sk(sk)->inet_rcv_saddr, 1547 inet_sk(sk)->inet_num); 1548 udp_lib_rehash(sk, new_hash); 1549 } 1550 1551 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1552 { 1553 int rc; 1554 1555 if (inet_sk(sk)->inet_daddr) { 1556 sock_rps_save_rxhash(sk, skb); 1557 sk_mark_napi_id(sk, skb); 1558 sk_incoming_cpu_update(sk); 1559 } 1560 1561 rc = sock_queue_rcv_skb(sk, skb); 1562 if (rc < 0) { 1563 int is_udplite = IS_UDPLITE(sk); 1564 1565 /* Note that an ENOMEM error is charged twice */ 1566 if (rc == -ENOMEM) 1567 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1568 is_udplite); 1569 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1570 kfree_skb(skb); 1571 trace_udp_fail_queue_rcv_skb(rc, sk); 1572 return -1; 1573 } 1574 1575 return 0; 1576 1577 } 1578 1579 static struct static_key udp_encap_needed __read_mostly; 1580 void udp_encap_enable(void) 1581 { 1582 if (!static_key_enabled(&udp_encap_needed)) 1583 static_key_slow_inc(&udp_encap_needed); 1584 } 1585 EXPORT_SYMBOL(udp_encap_enable); 1586 1587 /* returns: 1588 * -1: error 1589 * 0: success 1590 * >0: "udp encap" protocol resubmission 1591 * 1592 * Note that in the success and error cases, the skb is assumed to 1593 * have either been requeued or freed. 1594 */ 1595 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1596 { 1597 struct udp_sock *up = udp_sk(sk); 1598 int rc; 1599 int is_udplite = IS_UDPLITE(sk); 1600 1601 /* 1602 * Charge it to the socket, dropping if the queue is full. 1603 */ 1604 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1605 goto drop; 1606 nf_reset(skb); 1607 1608 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1609 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1610 1611 /* 1612 * This is an encapsulation socket so pass the skb to 1613 * the socket's udp_encap_rcv() hook. Otherwise, just 1614 * fall through and pass this up the UDP socket. 1615 * up->encap_rcv() returns the following value: 1616 * =0 if skb was successfully passed to the encap 1617 * handler or was discarded by it. 1618 * >0 if skb should be passed on to UDP. 1619 * <0 if skb should be resubmitted as proto -N 1620 */ 1621 1622 /* if we're overly short, let UDP handle it */ 1623 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1624 if (skb->len > sizeof(struct udphdr) && encap_rcv) { 1625 int ret; 1626 1627 /* Verify checksum before giving to encap */ 1628 if (udp_lib_checksum_complete(skb)) 1629 goto csum_error; 1630 1631 ret = encap_rcv(sk, skb); 1632 if (ret <= 0) { 1633 UDP_INC_STATS_BH(sock_net(sk), 1634 UDP_MIB_INDATAGRAMS, 1635 is_udplite); 1636 return -ret; 1637 } 1638 } 1639 1640 /* FALLTHROUGH -- it's a UDP Packet */ 1641 } 1642 1643 /* 1644 * UDP-Lite specific tests, ignored on UDP sockets 1645 */ 1646 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1647 1648 /* 1649 * MIB statistics other than incrementing the error count are 1650 * disabled for the following two types of errors: these depend 1651 * on the application settings, not on the functioning of the 1652 * protocol stack as such. 1653 * 1654 * RFC 3828 here recommends (sec 3.3): "There should also be a 1655 * way ... to ... at least let the receiving application block 1656 * delivery of packets with coverage values less than a value 1657 * provided by the application." 1658 */ 1659 if (up->pcrlen == 0) { /* full coverage was set */ 1660 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1661 UDP_SKB_CB(skb)->cscov, skb->len); 1662 goto drop; 1663 } 1664 /* The next case involves violating the min. coverage requested 1665 * by the receiver. This is subtle: if receiver wants x and x is 1666 * greater than the buffersize/MTU then receiver will complain 1667 * that it wants x while sender emits packets of smaller size y. 1668 * Therefore the above ...()->partial_cov statement is essential. 1669 */ 1670 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1671 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1672 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1673 goto drop; 1674 } 1675 } 1676 1677 if (rcu_access_pointer(sk->sk_filter) && 1678 udp_lib_checksum_complete(skb)) 1679 goto csum_error; 1680 1681 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1682 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1683 is_udplite); 1684 goto drop; 1685 } 1686 1687 rc = 0; 1688 1689 ipv4_pktinfo_prepare(sk, skb); 1690 bh_lock_sock(sk); 1691 if (!sock_owned_by_user(sk)) 1692 rc = __udp_queue_rcv_skb(sk, skb); 1693 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1694 bh_unlock_sock(sk); 1695 goto drop; 1696 } 1697 bh_unlock_sock(sk); 1698 1699 return rc; 1700 1701 csum_error: 1702 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1703 drop: 1704 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1705 atomic_inc(&sk->sk_drops); 1706 kfree_skb(skb); 1707 return -1; 1708 } 1709 1710 static void flush_stack(struct sock **stack, unsigned int count, 1711 struct sk_buff *skb, unsigned int final) 1712 { 1713 unsigned int i; 1714 struct sk_buff *skb1 = NULL; 1715 struct sock *sk; 1716 1717 for (i = 0; i < count; i++) { 1718 sk = stack[i]; 1719 if (likely(!skb1)) 1720 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1721 1722 if (!skb1) { 1723 atomic_inc(&sk->sk_drops); 1724 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1725 IS_UDPLITE(sk)); 1726 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1727 IS_UDPLITE(sk)); 1728 } 1729 1730 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1731 skb1 = NULL; 1732 1733 sock_put(sk); 1734 } 1735 if (unlikely(skb1)) 1736 kfree_skb(skb1); 1737 } 1738 1739 /* For TCP sockets, sk_rx_dst is protected by socket lock 1740 * For UDP, we use xchg() to guard against concurrent changes. 1741 */ 1742 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1743 { 1744 struct dst_entry *old; 1745 1746 dst_hold(dst); 1747 old = xchg(&sk->sk_rx_dst, dst); 1748 dst_release(old); 1749 } 1750 1751 /* 1752 * Multicasts and broadcasts go to each listener. 1753 * 1754 * Note: called only from the BH handler context. 1755 */ 1756 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1757 struct udphdr *uh, 1758 __be32 saddr, __be32 daddr, 1759 struct udp_table *udptable, 1760 int proto) 1761 { 1762 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1763 struct hlist_nulls_node *node; 1764 unsigned short hnum = ntohs(uh->dest); 1765 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1766 int dif = skb->dev->ifindex; 1767 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 1768 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1769 bool inner_flushed = false; 1770 1771 if (use_hash2) { 1772 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1773 udp_table.mask; 1774 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask; 1775 start_lookup: 1776 hslot = &udp_table.hash2[hash2]; 1777 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1778 } 1779 1780 spin_lock(&hslot->lock); 1781 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 1782 if (__udp_is_mcast_sock(net, sk, 1783 uh->dest, daddr, 1784 uh->source, saddr, 1785 dif, hnum)) { 1786 if (unlikely(count == ARRAY_SIZE(stack))) { 1787 flush_stack(stack, count, skb, ~0); 1788 inner_flushed = true; 1789 count = 0; 1790 } 1791 stack[count++] = sk; 1792 sock_hold(sk); 1793 } 1794 } 1795 1796 spin_unlock(&hslot->lock); 1797 1798 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1799 if (use_hash2 && hash2 != hash2_any) { 1800 hash2 = hash2_any; 1801 goto start_lookup; 1802 } 1803 1804 /* 1805 * do the slow work with no lock held 1806 */ 1807 if (count) { 1808 flush_stack(stack, count, skb, count - 1); 1809 } else { 1810 if (!inner_flushed) 1811 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 1812 proto == IPPROTO_UDPLITE); 1813 consume_skb(skb); 1814 } 1815 return 0; 1816 } 1817 1818 /* Initialize UDP checksum. If exited with zero value (success), 1819 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1820 * Otherwise, csum completion requires chacksumming packet body, 1821 * including udp header and folding it to skb->csum. 1822 */ 1823 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1824 int proto) 1825 { 1826 int err; 1827 1828 UDP_SKB_CB(skb)->partial_cov = 0; 1829 UDP_SKB_CB(skb)->cscov = skb->len; 1830 1831 if (proto == IPPROTO_UDPLITE) { 1832 err = udplite_checksum_init(skb, uh); 1833 if (err) 1834 return err; 1835 } 1836 1837 return skb_checksum_init_zero_check(skb, proto, uh->check, 1838 inet_compute_pseudo); 1839 } 1840 1841 /* 1842 * All we need to do is get the socket, and then do a checksum. 1843 */ 1844 1845 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1846 int proto) 1847 { 1848 struct sock *sk; 1849 struct udphdr *uh; 1850 unsigned short ulen; 1851 struct rtable *rt = skb_rtable(skb); 1852 __be32 saddr, daddr; 1853 struct net *net = dev_net(skb->dev); 1854 1855 /* 1856 * Validate the packet. 1857 */ 1858 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1859 goto drop; /* No space for header. */ 1860 1861 uh = udp_hdr(skb); 1862 ulen = ntohs(uh->len); 1863 saddr = ip_hdr(skb)->saddr; 1864 daddr = ip_hdr(skb)->daddr; 1865 1866 if (ulen > skb->len) 1867 goto short_packet; 1868 1869 if (proto == IPPROTO_UDP) { 1870 /* UDP validates ulen. */ 1871 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1872 goto short_packet; 1873 uh = udp_hdr(skb); 1874 } 1875 1876 if (udp4_csum_init(skb, uh, proto)) 1877 goto csum_error; 1878 1879 sk = skb_steal_sock(skb); 1880 if (sk) { 1881 struct dst_entry *dst = skb_dst(skb); 1882 int ret; 1883 1884 if (unlikely(sk->sk_rx_dst != dst)) 1885 udp_sk_rx_dst_set(sk, dst); 1886 1887 ret = udp_queue_rcv_skb(sk, skb); 1888 sock_put(sk); 1889 /* a return value > 0 means to resubmit the input, but 1890 * it wants the return to be -protocol, or 0 1891 */ 1892 if (ret > 0) 1893 return -ret; 1894 return 0; 1895 } 1896 1897 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1898 return __udp4_lib_mcast_deliver(net, skb, uh, 1899 saddr, daddr, udptable, proto); 1900 1901 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1902 if (sk) { 1903 int ret; 1904 1905 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1906 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1907 inet_compute_pseudo); 1908 1909 ret = udp_queue_rcv_skb(sk, skb); 1910 sock_put(sk); 1911 1912 /* a return value > 0 means to resubmit the input, but 1913 * it wants the return to be -protocol, or 0 1914 */ 1915 if (ret > 0) 1916 return -ret; 1917 return 0; 1918 } 1919 1920 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1921 goto drop; 1922 nf_reset(skb); 1923 1924 /* No socket. Drop packet silently, if checksum is wrong */ 1925 if (udp_lib_checksum_complete(skb)) 1926 goto csum_error; 1927 1928 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1929 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1930 1931 /* 1932 * Hmm. We got an UDP packet to a port to which we 1933 * don't wanna listen. Ignore it. 1934 */ 1935 kfree_skb(skb); 1936 return 0; 1937 1938 short_packet: 1939 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1940 proto == IPPROTO_UDPLITE ? "Lite" : "", 1941 &saddr, ntohs(uh->source), 1942 ulen, skb->len, 1943 &daddr, ntohs(uh->dest)); 1944 goto drop; 1945 1946 csum_error: 1947 /* 1948 * RFC1122: OK. Discards the bad packet silently (as far as 1949 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1950 */ 1951 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1952 proto == IPPROTO_UDPLITE ? "Lite" : "", 1953 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1954 ulen); 1955 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1956 drop: 1957 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1958 kfree_skb(skb); 1959 return 0; 1960 } 1961 1962 /* We can only early demux multicast if there is a single matching socket. 1963 * If more than one socket found returns NULL 1964 */ 1965 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1966 __be16 loc_port, __be32 loc_addr, 1967 __be16 rmt_port, __be32 rmt_addr, 1968 int dif) 1969 { 1970 struct sock *sk, *result; 1971 struct hlist_nulls_node *node; 1972 unsigned short hnum = ntohs(loc_port); 1973 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1974 struct udp_hslot *hslot = &udp_table.hash[slot]; 1975 1976 /* Do not bother scanning a too big list */ 1977 if (hslot->count > 10) 1978 return NULL; 1979 1980 rcu_read_lock(); 1981 begin: 1982 count = 0; 1983 result = NULL; 1984 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1985 if (__udp_is_mcast_sock(net, sk, 1986 loc_port, loc_addr, 1987 rmt_port, rmt_addr, 1988 dif, hnum)) { 1989 result = sk; 1990 ++count; 1991 } 1992 } 1993 /* 1994 * if the nulls value we got at the end of this lookup is 1995 * not the expected one, we must restart lookup. 1996 * We probably met an item that was moved to another chain. 1997 */ 1998 if (get_nulls_value(node) != slot) 1999 goto begin; 2000 2001 if (result) { 2002 if (count != 1 || 2003 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 2004 result = NULL; 2005 else if (unlikely(!__udp_is_mcast_sock(net, result, 2006 loc_port, loc_addr, 2007 rmt_port, rmt_addr, 2008 dif, hnum))) { 2009 sock_put(result); 2010 result = NULL; 2011 } 2012 } 2013 rcu_read_unlock(); 2014 return result; 2015 } 2016 2017 /* For unicast we should only early demux connected sockets or we can 2018 * break forwarding setups. The chains here can be long so only check 2019 * if the first socket is an exact match and if not move on. 2020 */ 2021 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2022 __be16 loc_port, __be32 loc_addr, 2023 __be16 rmt_port, __be32 rmt_addr, 2024 int dif) 2025 { 2026 struct sock *sk, *result; 2027 struct hlist_nulls_node *node; 2028 unsigned short hnum = ntohs(loc_port); 2029 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 2030 unsigned int slot2 = hash2 & udp_table.mask; 2031 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2032 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2033 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2034 2035 rcu_read_lock(); 2036 result = NULL; 2037 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 2038 if (INET_MATCH(sk, net, acookie, 2039 rmt_addr, loc_addr, ports, dif)) 2040 result = sk; 2041 /* Only check first socket in chain */ 2042 break; 2043 } 2044 2045 if (result) { 2046 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 2047 result = NULL; 2048 else if (unlikely(!INET_MATCH(sk, net, acookie, 2049 rmt_addr, loc_addr, 2050 ports, dif))) { 2051 sock_put(result); 2052 result = NULL; 2053 } 2054 } 2055 rcu_read_unlock(); 2056 return result; 2057 } 2058 2059 void udp_v4_early_demux(struct sk_buff *skb) 2060 { 2061 struct net *net = dev_net(skb->dev); 2062 const struct iphdr *iph; 2063 const struct udphdr *uh; 2064 struct sock *sk; 2065 struct dst_entry *dst; 2066 int dif = skb->dev->ifindex; 2067 int ours; 2068 2069 /* validate the packet */ 2070 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2071 return; 2072 2073 iph = ip_hdr(skb); 2074 uh = udp_hdr(skb); 2075 2076 if (skb->pkt_type == PACKET_BROADCAST || 2077 skb->pkt_type == PACKET_MULTICAST) { 2078 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 2079 2080 if (!in_dev) 2081 return; 2082 2083 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2084 iph->protocol); 2085 if (!ours) 2086 return; 2087 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2088 uh->source, iph->saddr, dif); 2089 } else if (skb->pkt_type == PACKET_HOST) { 2090 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2091 uh->source, iph->saddr, dif); 2092 } else { 2093 return; 2094 } 2095 2096 if (!sk) 2097 return; 2098 2099 skb->sk = sk; 2100 skb->destructor = sock_efree; 2101 dst = READ_ONCE(sk->sk_rx_dst); 2102 2103 if (dst) 2104 dst = dst_check(dst, 0); 2105 if (dst) { 2106 /* DST_NOCACHE can not be used without taking a reference */ 2107 if (dst->flags & DST_NOCACHE) { 2108 if (likely(atomic_inc_not_zero(&dst->__refcnt))) 2109 skb_dst_set(skb, dst); 2110 } else { 2111 skb_dst_set_noref(skb, dst); 2112 } 2113 } 2114 } 2115 2116 int udp_rcv(struct sk_buff *skb) 2117 { 2118 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2119 } 2120 2121 void udp_destroy_sock(struct sock *sk) 2122 { 2123 struct udp_sock *up = udp_sk(sk); 2124 bool slow = lock_sock_fast(sk); 2125 udp_flush_pending_frames(sk); 2126 unlock_sock_fast(sk, slow); 2127 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2128 void (*encap_destroy)(struct sock *sk); 2129 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2130 if (encap_destroy) 2131 encap_destroy(sk); 2132 } 2133 } 2134 2135 /* 2136 * Socket option code for UDP 2137 */ 2138 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2139 char __user *optval, unsigned int optlen, 2140 int (*push_pending_frames)(struct sock *)) 2141 { 2142 struct udp_sock *up = udp_sk(sk); 2143 int val, valbool; 2144 int err = 0; 2145 int is_udplite = IS_UDPLITE(sk); 2146 2147 if (optlen < sizeof(int)) 2148 return -EINVAL; 2149 2150 if (get_user(val, (int __user *)optval)) 2151 return -EFAULT; 2152 2153 valbool = val ? 1 : 0; 2154 2155 switch (optname) { 2156 case UDP_CORK: 2157 if (val != 0) { 2158 up->corkflag = 1; 2159 } else { 2160 up->corkflag = 0; 2161 lock_sock(sk); 2162 push_pending_frames(sk); 2163 release_sock(sk); 2164 } 2165 break; 2166 2167 case UDP_ENCAP: 2168 switch (val) { 2169 case 0: 2170 case UDP_ENCAP_ESPINUDP: 2171 case UDP_ENCAP_ESPINUDP_NON_IKE: 2172 up->encap_rcv = xfrm4_udp_encap_rcv; 2173 /* FALLTHROUGH */ 2174 case UDP_ENCAP_L2TPINUDP: 2175 up->encap_type = val; 2176 udp_encap_enable(); 2177 break; 2178 default: 2179 err = -ENOPROTOOPT; 2180 break; 2181 } 2182 break; 2183 2184 case UDP_NO_CHECK6_TX: 2185 up->no_check6_tx = valbool; 2186 break; 2187 2188 case UDP_NO_CHECK6_RX: 2189 up->no_check6_rx = valbool; 2190 break; 2191 2192 /* 2193 * UDP-Lite's partial checksum coverage (RFC 3828). 2194 */ 2195 /* The sender sets actual checksum coverage length via this option. 2196 * The case coverage > packet length is handled by send module. */ 2197 case UDPLITE_SEND_CSCOV: 2198 if (!is_udplite) /* Disable the option on UDP sockets */ 2199 return -ENOPROTOOPT; 2200 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2201 val = 8; 2202 else if (val > USHRT_MAX) 2203 val = USHRT_MAX; 2204 up->pcslen = val; 2205 up->pcflag |= UDPLITE_SEND_CC; 2206 break; 2207 2208 /* The receiver specifies a minimum checksum coverage value. To make 2209 * sense, this should be set to at least 8 (as done below). If zero is 2210 * used, this again means full checksum coverage. */ 2211 case UDPLITE_RECV_CSCOV: 2212 if (!is_udplite) /* Disable the option on UDP sockets */ 2213 return -ENOPROTOOPT; 2214 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2215 val = 8; 2216 else if (val > USHRT_MAX) 2217 val = USHRT_MAX; 2218 up->pcrlen = val; 2219 up->pcflag |= UDPLITE_RECV_CC; 2220 break; 2221 2222 default: 2223 err = -ENOPROTOOPT; 2224 break; 2225 } 2226 2227 return err; 2228 } 2229 EXPORT_SYMBOL(udp_lib_setsockopt); 2230 2231 int udp_setsockopt(struct sock *sk, int level, int optname, 2232 char __user *optval, unsigned int optlen) 2233 { 2234 if (level == SOL_UDP || level == SOL_UDPLITE) 2235 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2236 udp_push_pending_frames); 2237 return ip_setsockopt(sk, level, optname, optval, optlen); 2238 } 2239 2240 #ifdef CONFIG_COMPAT 2241 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2242 char __user *optval, unsigned int optlen) 2243 { 2244 if (level == SOL_UDP || level == SOL_UDPLITE) 2245 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2246 udp_push_pending_frames); 2247 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2248 } 2249 #endif 2250 2251 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2252 char __user *optval, int __user *optlen) 2253 { 2254 struct udp_sock *up = udp_sk(sk); 2255 int val, len; 2256 2257 if (get_user(len, optlen)) 2258 return -EFAULT; 2259 2260 len = min_t(unsigned int, len, sizeof(int)); 2261 2262 if (len < 0) 2263 return -EINVAL; 2264 2265 switch (optname) { 2266 case UDP_CORK: 2267 val = up->corkflag; 2268 break; 2269 2270 case UDP_ENCAP: 2271 val = up->encap_type; 2272 break; 2273 2274 case UDP_NO_CHECK6_TX: 2275 val = up->no_check6_tx; 2276 break; 2277 2278 case UDP_NO_CHECK6_RX: 2279 val = up->no_check6_rx; 2280 break; 2281 2282 /* The following two cannot be changed on UDP sockets, the return is 2283 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2284 case UDPLITE_SEND_CSCOV: 2285 val = up->pcslen; 2286 break; 2287 2288 case UDPLITE_RECV_CSCOV: 2289 val = up->pcrlen; 2290 break; 2291 2292 default: 2293 return -ENOPROTOOPT; 2294 } 2295 2296 if (put_user(len, optlen)) 2297 return -EFAULT; 2298 if (copy_to_user(optval, &val, len)) 2299 return -EFAULT; 2300 return 0; 2301 } 2302 EXPORT_SYMBOL(udp_lib_getsockopt); 2303 2304 int udp_getsockopt(struct sock *sk, int level, int optname, 2305 char __user *optval, int __user *optlen) 2306 { 2307 if (level == SOL_UDP || level == SOL_UDPLITE) 2308 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2309 return ip_getsockopt(sk, level, optname, optval, optlen); 2310 } 2311 2312 #ifdef CONFIG_COMPAT 2313 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2314 char __user *optval, int __user *optlen) 2315 { 2316 if (level == SOL_UDP || level == SOL_UDPLITE) 2317 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2318 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2319 } 2320 #endif 2321 /** 2322 * udp_poll - wait for a UDP event. 2323 * @file - file struct 2324 * @sock - socket 2325 * @wait - poll table 2326 * 2327 * This is same as datagram poll, except for the special case of 2328 * blocking sockets. If application is using a blocking fd 2329 * and a packet with checksum error is in the queue; 2330 * then it could get return from select indicating data available 2331 * but then block when reading it. Add special case code 2332 * to work around these arguably broken applications. 2333 */ 2334 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2335 { 2336 unsigned int mask = datagram_poll(file, sock, wait); 2337 struct sock *sk = sock->sk; 2338 2339 sock_rps_record_flow(sk); 2340 2341 /* Check for false positives due to checksum errors */ 2342 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2343 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2344 mask &= ~(POLLIN | POLLRDNORM); 2345 2346 return mask; 2347 2348 } 2349 EXPORT_SYMBOL(udp_poll); 2350 2351 struct proto udp_prot = { 2352 .name = "UDP", 2353 .owner = THIS_MODULE, 2354 .close = udp_lib_close, 2355 .connect = ip4_datagram_connect, 2356 .disconnect = udp_disconnect, 2357 .ioctl = udp_ioctl, 2358 .destroy = udp_destroy_sock, 2359 .setsockopt = udp_setsockopt, 2360 .getsockopt = udp_getsockopt, 2361 .sendmsg = udp_sendmsg, 2362 .recvmsg = udp_recvmsg, 2363 .sendpage = udp_sendpage, 2364 .backlog_rcv = __udp_queue_rcv_skb, 2365 .release_cb = ip4_datagram_release_cb, 2366 .hash = udp_lib_hash, 2367 .unhash = udp_lib_unhash, 2368 .rehash = udp_v4_rehash, 2369 .get_port = udp_v4_get_port, 2370 .memory_allocated = &udp_memory_allocated, 2371 .sysctl_mem = sysctl_udp_mem, 2372 .sysctl_wmem = &sysctl_udp_wmem_min, 2373 .sysctl_rmem = &sysctl_udp_rmem_min, 2374 .obj_size = sizeof(struct udp_sock), 2375 .slab_flags = SLAB_DESTROY_BY_RCU, 2376 .h.udp_table = &udp_table, 2377 #ifdef CONFIG_COMPAT 2378 .compat_setsockopt = compat_udp_setsockopt, 2379 .compat_getsockopt = compat_udp_getsockopt, 2380 #endif 2381 .clear_sk = sk_prot_clear_portaddr_nulls, 2382 }; 2383 EXPORT_SYMBOL(udp_prot); 2384 2385 /* ------------------------------------------------------------------------ */ 2386 #ifdef CONFIG_PROC_FS 2387 2388 static struct sock *udp_get_first(struct seq_file *seq, int start) 2389 { 2390 struct sock *sk; 2391 struct udp_iter_state *state = seq->private; 2392 struct net *net = seq_file_net(seq); 2393 2394 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2395 ++state->bucket) { 2396 struct hlist_nulls_node *node; 2397 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2398 2399 if (hlist_nulls_empty(&hslot->head)) 2400 continue; 2401 2402 spin_lock_bh(&hslot->lock); 2403 sk_nulls_for_each(sk, node, &hslot->head) { 2404 if (!net_eq(sock_net(sk), net)) 2405 continue; 2406 if (sk->sk_family == state->family) 2407 goto found; 2408 } 2409 spin_unlock_bh(&hslot->lock); 2410 } 2411 sk = NULL; 2412 found: 2413 return sk; 2414 } 2415 2416 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2417 { 2418 struct udp_iter_state *state = seq->private; 2419 struct net *net = seq_file_net(seq); 2420 2421 do { 2422 sk = sk_nulls_next(sk); 2423 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2424 2425 if (!sk) { 2426 if (state->bucket <= state->udp_table->mask) 2427 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2428 return udp_get_first(seq, state->bucket + 1); 2429 } 2430 return sk; 2431 } 2432 2433 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2434 { 2435 struct sock *sk = udp_get_first(seq, 0); 2436 2437 if (sk) 2438 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2439 --pos; 2440 return pos ? NULL : sk; 2441 } 2442 2443 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2444 { 2445 struct udp_iter_state *state = seq->private; 2446 state->bucket = MAX_UDP_PORTS; 2447 2448 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2449 } 2450 2451 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2452 { 2453 struct sock *sk; 2454 2455 if (v == SEQ_START_TOKEN) 2456 sk = udp_get_idx(seq, 0); 2457 else 2458 sk = udp_get_next(seq, v); 2459 2460 ++*pos; 2461 return sk; 2462 } 2463 2464 static void udp_seq_stop(struct seq_file *seq, void *v) 2465 { 2466 struct udp_iter_state *state = seq->private; 2467 2468 if (state->bucket <= state->udp_table->mask) 2469 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2470 } 2471 2472 int udp_seq_open(struct inode *inode, struct file *file) 2473 { 2474 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2475 struct udp_iter_state *s; 2476 int err; 2477 2478 err = seq_open_net(inode, file, &afinfo->seq_ops, 2479 sizeof(struct udp_iter_state)); 2480 if (err < 0) 2481 return err; 2482 2483 s = ((struct seq_file *)file->private_data)->private; 2484 s->family = afinfo->family; 2485 s->udp_table = afinfo->udp_table; 2486 return err; 2487 } 2488 EXPORT_SYMBOL(udp_seq_open); 2489 2490 /* ------------------------------------------------------------------------ */ 2491 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2492 { 2493 struct proc_dir_entry *p; 2494 int rc = 0; 2495 2496 afinfo->seq_ops.start = udp_seq_start; 2497 afinfo->seq_ops.next = udp_seq_next; 2498 afinfo->seq_ops.stop = udp_seq_stop; 2499 2500 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2501 afinfo->seq_fops, afinfo); 2502 if (!p) 2503 rc = -ENOMEM; 2504 return rc; 2505 } 2506 EXPORT_SYMBOL(udp_proc_register); 2507 2508 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2509 { 2510 remove_proc_entry(afinfo->name, net->proc_net); 2511 } 2512 EXPORT_SYMBOL(udp_proc_unregister); 2513 2514 /* ------------------------------------------------------------------------ */ 2515 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2516 int bucket) 2517 { 2518 struct inet_sock *inet = inet_sk(sp); 2519 __be32 dest = inet->inet_daddr; 2520 __be32 src = inet->inet_rcv_saddr; 2521 __u16 destp = ntohs(inet->inet_dport); 2522 __u16 srcp = ntohs(inet->inet_sport); 2523 2524 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2525 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2526 bucket, src, srcp, dest, destp, sp->sk_state, 2527 sk_wmem_alloc_get(sp), 2528 sk_rmem_alloc_get(sp), 2529 0, 0L, 0, 2530 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2531 0, sock_i_ino(sp), 2532 atomic_read(&sp->sk_refcnt), sp, 2533 atomic_read(&sp->sk_drops)); 2534 } 2535 2536 int udp4_seq_show(struct seq_file *seq, void *v) 2537 { 2538 seq_setwidth(seq, 127); 2539 if (v == SEQ_START_TOKEN) 2540 seq_puts(seq, " sl local_address rem_address st tx_queue " 2541 "rx_queue tr tm->when retrnsmt uid timeout " 2542 "inode ref pointer drops"); 2543 else { 2544 struct udp_iter_state *state = seq->private; 2545 2546 udp4_format_sock(v, seq, state->bucket); 2547 } 2548 seq_pad(seq, '\n'); 2549 return 0; 2550 } 2551 2552 static const struct file_operations udp_afinfo_seq_fops = { 2553 .owner = THIS_MODULE, 2554 .open = udp_seq_open, 2555 .read = seq_read, 2556 .llseek = seq_lseek, 2557 .release = seq_release_net 2558 }; 2559 2560 /* ------------------------------------------------------------------------ */ 2561 static struct udp_seq_afinfo udp4_seq_afinfo = { 2562 .name = "udp", 2563 .family = AF_INET, 2564 .udp_table = &udp_table, 2565 .seq_fops = &udp_afinfo_seq_fops, 2566 .seq_ops = { 2567 .show = udp4_seq_show, 2568 }, 2569 }; 2570 2571 static int __net_init udp4_proc_init_net(struct net *net) 2572 { 2573 return udp_proc_register(net, &udp4_seq_afinfo); 2574 } 2575 2576 static void __net_exit udp4_proc_exit_net(struct net *net) 2577 { 2578 udp_proc_unregister(net, &udp4_seq_afinfo); 2579 } 2580 2581 static struct pernet_operations udp4_net_ops = { 2582 .init = udp4_proc_init_net, 2583 .exit = udp4_proc_exit_net, 2584 }; 2585 2586 int __init udp4_proc_init(void) 2587 { 2588 return register_pernet_subsys(&udp4_net_ops); 2589 } 2590 2591 void udp4_proc_exit(void) 2592 { 2593 unregister_pernet_subsys(&udp4_net_ops); 2594 } 2595 #endif /* CONFIG_PROC_FS */ 2596 2597 static __initdata unsigned long uhash_entries; 2598 static int __init set_uhash_entries(char *str) 2599 { 2600 ssize_t ret; 2601 2602 if (!str) 2603 return 0; 2604 2605 ret = kstrtoul(str, 0, &uhash_entries); 2606 if (ret) 2607 return 0; 2608 2609 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2610 uhash_entries = UDP_HTABLE_SIZE_MIN; 2611 return 1; 2612 } 2613 __setup("uhash_entries=", set_uhash_entries); 2614 2615 void __init udp_table_init(struct udp_table *table, const char *name) 2616 { 2617 unsigned int i; 2618 2619 table->hash = alloc_large_system_hash(name, 2620 2 * sizeof(struct udp_hslot), 2621 uhash_entries, 2622 21, /* one slot per 2 MB */ 2623 0, 2624 &table->log, 2625 &table->mask, 2626 UDP_HTABLE_SIZE_MIN, 2627 64 * 1024); 2628 2629 table->hash2 = table->hash + (table->mask + 1); 2630 for (i = 0; i <= table->mask; i++) { 2631 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2632 table->hash[i].count = 0; 2633 spin_lock_init(&table->hash[i].lock); 2634 } 2635 for (i = 0; i <= table->mask; i++) { 2636 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2637 table->hash2[i].count = 0; 2638 spin_lock_init(&table->hash2[i].lock); 2639 } 2640 } 2641 2642 u32 udp_flow_hashrnd(void) 2643 { 2644 static u32 hashrnd __read_mostly; 2645 2646 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2647 2648 return hashrnd; 2649 } 2650 EXPORT_SYMBOL(udp_flow_hashrnd); 2651 2652 void __init udp_init(void) 2653 { 2654 unsigned long limit; 2655 2656 udp_table_init(&udp_table, "UDP"); 2657 limit = nr_free_buffer_pages() / 8; 2658 limit = max(limit, 128UL); 2659 sysctl_udp_mem[0] = limit / 4 * 3; 2660 sysctl_udp_mem[1] = limit; 2661 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2662 2663 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2664 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2665 } 2666