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 <linux/uaccess.h> 83 #include <asm/ioctls.h> 84 #include <linux/bootmem.h> 85 #include <linux/highmem.h> 86 #include <linux/swap.h> 87 #include <linux/types.h> 88 #include <linux/fcntl.h> 89 #include <linux/module.h> 90 #include <linux/socket.h> 91 #include <linux/sockios.h> 92 #include <linux/igmp.h> 93 #include <linux/inetdevice.h> 94 #include <linux/in.h> 95 #include <linux/errno.h> 96 #include <linux/timer.h> 97 #include <linux/mm.h> 98 #include <linux/inet.h> 99 #include <linux/netdevice.h> 100 #include <linux/slab.h> 101 #include <net/tcp_states.h> 102 #include <linux/skbuff.h> 103 #include <linux/proc_fs.h> 104 #include <linux/seq_file.h> 105 #include <net/net_namespace.h> 106 #include <net/icmp.h> 107 #include <net/inet_hashtables.h> 108 #include <net/route.h> 109 #include <net/checksum.h> 110 #include <net/xfrm.h> 111 #include <trace/events/udp.h> 112 #include <linux/static_key.h> 113 #include <trace/events/skb.h> 114 #include <net/busy_poll.h> 115 #include "udp_impl.h" 116 #include <net/sock_reuseport.h> 117 #include <net/addrconf.h> 118 119 struct udp_table udp_table __read_mostly; 120 EXPORT_SYMBOL(udp_table); 121 122 long sysctl_udp_mem[3] __read_mostly; 123 EXPORT_SYMBOL(sysctl_udp_mem); 124 125 int sysctl_udp_rmem_min __read_mostly; 126 EXPORT_SYMBOL(sysctl_udp_rmem_min); 127 128 int sysctl_udp_wmem_min __read_mostly; 129 EXPORT_SYMBOL(sysctl_udp_wmem_min); 130 131 atomic_long_t udp_memory_allocated; 132 EXPORT_SYMBOL(udp_memory_allocated); 133 134 #define MAX_UDP_PORTS 65536 135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 136 137 /* IPCB reference means this can not be used from early demux */ 138 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb) 139 { 140 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 141 if (!net->ipv4.sysctl_udp_l3mdev_accept && 142 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 143 return true; 144 #endif 145 return false; 146 } 147 148 static int udp_lib_lport_inuse(struct net *net, __u16 num, 149 const struct udp_hslot *hslot, 150 unsigned long *bitmap, 151 struct sock *sk, unsigned int log) 152 { 153 struct sock *sk2; 154 kuid_t uid = sock_i_uid(sk); 155 156 sk_for_each(sk2, &hslot->head) { 157 if (net_eq(sock_net(sk2), net) && 158 sk2 != sk && 159 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 160 (!sk2->sk_reuse || !sk->sk_reuse) && 161 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 162 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 163 inet_rcv_saddr_equal(sk, sk2, true)) { 164 if (sk2->sk_reuseport && sk->sk_reuseport && 165 !rcu_access_pointer(sk->sk_reuseport_cb) && 166 uid_eq(uid, sock_i_uid(sk2))) { 167 if (!bitmap) 168 return 0; 169 } else { 170 if (!bitmap) 171 return 1; 172 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 173 bitmap); 174 } 175 } 176 } 177 return 0; 178 } 179 180 /* 181 * Note: we still hold spinlock of primary hash chain, so no other writer 182 * can insert/delete a socket with local_port == num 183 */ 184 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 185 struct udp_hslot *hslot2, 186 struct sock *sk) 187 { 188 struct sock *sk2; 189 kuid_t uid = sock_i_uid(sk); 190 int res = 0; 191 192 spin_lock(&hslot2->lock); 193 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 194 if (net_eq(sock_net(sk2), net) && 195 sk2 != sk && 196 (udp_sk(sk2)->udp_port_hash == num) && 197 (!sk2->sk_reuse || !sk->sk_reuse) && 198 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 199 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 200 inet_rcv_saddr_equal(sk, sk2, true)) { 201 if (sk2->sk_reuseport && sk->sk_reuseport && 202 !rcu_access_pointer(sk->sk_reuseport_cb) && 203 uid_eq(uid, sock_i_uid(sk2))) { 204 res = 0; 205 } else { 206 res = 1; 207 } 208 break; 209 } 210 } 211 spin_unlock(&hslot2->lock); 212 return res; 213 } 214 215 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 216 { 217 struct net *net = sock_net(sk); 218 kuid_t uid = sock_i_uid(sk); 219 struct sock *sk2; 220 221 sk_for_each(sk2, &hslot->head) { 222 if (net_eq(sock_net(sk2), net) && 223 sk2 != sk && 224 sk2->sk_family == sk->sk_family && 225 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 226 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 227 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 228 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 229 inet_rcv_saddr_equal(sk, sk2, false)) { 230 return reuseport_add_sock(sk, sk2); 231 } 232 } 233 234 return reuseport_alloc(sk); 235 } 236 237 /** 238 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 239 * 240 * @sk: socket struct in question 241 * @snum: port number to look up 242 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 243 * with NULL address 244 */ 245 int udp_lib_get_port(struct sock *sk, unsigned short snum, 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 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 cond_resched(); 291 } while (++first != last); 292 goto fail; 293 } else { 294 hslot = udp_hashslot(udptable, net, snum); 295 spin_lock_bh(&hslot->lock); 296 if (hslot->count > 10) { 297 int exist; 298 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 299 300 slot2 &= udptable->mask; 301 hash2_nulladdr &= udptable->mask; 302 303 hslot2 = udp_hashslot2(udptable, slot2); 304 if (hslot->count < hslot2->count) 305 goto scan_primary_hash; 306 307 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 308 if (!exist && (hash2_nulladdr != slot2)) { 309 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 310 exist = udp_lib_lport_inuse2(net, snum, hslot2, 311 sk); 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, 0)) 320 goto fail_unlock; 321 } 322 found: 323 inet_sk(sk)->inet_num = snum; 324 udp_sk(sk)->udp_port_hash = snum; 325 udp_sk(sk)->udp_portaddr_hash ^= snum; 326 if (sk_unhashed(sk)) { 327 if (sk->sk_reuseport && 328 udp_reuseport_add_sock(sk, hslot)) { 329 inet_sk(sk)->inet_num = 0; 330 udp_sk(sk)->udp_port_hash = 0; 331 udp_sk(sk)->udp_portaddr_hash ^= snum; 332 goto fail_unlock; 333 } 334 335 sk_add_node_rcu(sk, &hslot->head); 336 hslot->count++; 337 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 338 339 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 340 spin_lock(&hslot2->lock); 341 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 342 sk->sk_family == AF_INET6) 343 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 344 &hslot2->head); 345 else 346 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 347 &hslot2->head); 348 hslot2->count++; 349 spin_unlock(&hslot2->lock); 350 } 351 sock_set_flag(sk, SOCK_RCU_FREE); 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 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, 361 unsigned int port) 362 { 363 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 364 } 365 366 int udp_v4_get_port(struct sock *sk, unsigned short snum) 367 { 368 unsigned int hash2_nulladdr = 369 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 370 unsigned int hash2_partial = 371 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 372 373 /* precompute partial secondary hash */ 374 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 375 return udp_lib_get_port(sk, snum, hash2_nulladdr); 376 } 377 378 static int compute_score(struct sock *sk, struct net *net, 379 __be32 saddr, __be16 sport, 380 __be32 daddr, unsigned short hnum, 381 int dif, int sdif, bool exact_dif) 382 { 383 int score; 384 struct inet_sock *inet; 385 386 if (!net_eq(sock_net(sk), net) || 387 udp_sk(sk)->udp_port_hash != hnum || 388 ipv6_only_sock(sk)) 389 return -1; 390 391 score = (sk->sk_family == PF_INET) ? 2 : 1; 392 inet = inet_sk(sk); 393 394 if (inet->inet_rcv_saddr) { 395 if (inet->inet_rcv_saddr != daddr) 396 return -1; 397 score += 4; 398 } 399 400 if (inet->inet_daddr) { 401 if (inet->inet_daddr != saddr) 402 return -1; 403 score += 4; 404 } 405 406 if (inet->inet_dport) { 407 if (inet->inet_dport != sport) 408 return -1; 409 score += 4; 410 } 411 412 if (sk->sk_bound_dev_if || exact_dif) { 413 bool dev_match = (sk->sk_bound_dev_if == dif || 414 sk->sk_bound_dev_if == sdif); 415 416 if (exact_dif && !dev_match) 417 return -1; 418 if (sk->sk_bound_dev_if && dev_match) 419 score += 4; 420 } 421 422 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 423 score++; 424 return score; 425 } 426 427 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 428 const __u16 lport, const __be32 faddr, 429 const __be16 fport) 430 { 431 static u32 udp_ehash_secret __read_mostly; 432 433 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 434 435 return __inet_ehashfn(laddr, lport, faddr, fport, 436 udp_ehash_secret + net_hash_mix(net)); 437 } 438 439 /* called with rcu_read_lock() */ 440 static struct sock *udp4_lib_lookup2(struct net *net, 441 __be32 saddr, __be16 sport, 442 __be32 daddr, unsigned int hnum, 443 int dif, int sdif, bool exact_dif, 444 struct udp_hslot *hslot2, 445 struct sk_buff *skb) 446 { 447 struct sock *sk, *result; 448 int score, badness, matches = 0, reuseport = 0; 449 u32 hash = 0; 450 451 result = NULL; 452 badness = 0; 453 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 454 score = compute_score(sk, net, saddr, sport, 455 daddr, hnum, dif, sdif, exact_dif); 456 if (score > badness) { 457 reuseport = sk->sk_reuseport; 458 if (reuseport) { 459 hash = udp_ehashfn(net, daddr, hnum, 460 saddr, sport); 461 result = reuseport_select_sock(sk, hash, skb, 462 sizeof(struct udphdr)); 463 if (result) 464 return result; 465 matches = 1; 466 } 467 badness = score; 468 result = sk; 469 } else if (score == badness && reuseport) { 470 matches++; 471 if (reciprocal_scale(hash, matches) == 0) 472 result = sk; 473 hash = next_pseudo_random32(hash); 474 } 475 } 476 return result; 477 } 478 479 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 480 * harder than this. -DaveM 481 */ 482 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 483 __be16 sport, __be32 daddr, __be16 dport, int dif, 484 int sdif, struct udp_table *udptable, struct sk_buff *skb) 485 { 486 struct sock *sk, *result; 487 unsigned short hnum = ntohs(dport); 488 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 489 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 490 bool exact_dif = udp_lib_exact_dif_match(net, skb); 491 int score, badness, matches = 0, reuseport = 0; 492 u32 hash = 0; 493 494 if (hslot->count > 10) { 495 hash2 = udp4_portaddr_hash(net, daddr, hnum); 496 slot2 = hash2 & udptable->mask; 497 hslot2 = &udptable->hash2[slot2]; 498 if (hslot->count < hslot2->count) 499 goto begin; 500 501 result = udp4_lib_lookup2(net, saddr, sport, 502 daddr, hnum, dif, sdif, 503 exact_dif, hslot2, skb); 504 if (!result) { 505 unsigned int old_slot2 = slot2; 506 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 507 slot2 = hash2 & udptable->mask; 508 /* avoid searching the same slot again. */ 509 if (unlikely(slot2 == old_slot2)) 510 return result; 511 512 hslot2 = &udptable->hash2[slot2]; 513 if (hslot->count < hslot2->count) 514 goto begin; 515 516 result = udp4_lib_lookup2(net, saddr, sport, 517 daddr, hnum, dif, sdif, 518 exact_dif, hslot2, skb); 519 } 520 return result; 521 } 522 begin: 523 result = NULL; 524 badness = 0; 525 sk_for_each_rcu(sk, &hslot->head) { 526 score = compute_score(sk, net, saddr, sport, 527 daddr, hnum, dif, sdif, exact_dif); 528 if (score > badness) { 529 reuseport = sk->sk_reuseport; 530 if (reuseport) { 531 hash = udp_ehashfn(net, daddr, hnum, 532 saddr, sport); 533 result = reuseport_select_sock(sk, hash, skb, 534 sizeof(struct udphdr)); 535 if (result) 536 return result; 537 matches = 1; 538 } 539 result = sk; 540 badness = score; 541 } else if (score == badness && reuseport) { 542 matches++; 543 if (reciprocal_scale(hash, matches) == 0) 544 result = sk; 545 hash = next_pseudo_random32(hash); 546 } 547 } 548 return result; 549 } 550 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 551 552 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 553 __be16 sport, __be16 dport, 554 struct udp_table *udptable) 555 { 556 const struct iphdr *iph = ip_hdr(skb); 557 558 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 559 iph->daddr, dport, inet_iif(skb), 560 inet_sdif(skb), udptable, skb); 561 } 562 563 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, 564 __be16 sport, __be16 dport) 565 { 566 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); 567 } 568 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); 569 570 /* Must be called under rcu_read_lock(). 571 * Does increment socket refcount. 572 */ 573 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \ 574 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \ 575 IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 576 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 577 __be32 daddr, __be16 dport, int dif) 578 { 579 struct sock *sk; 580 581 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 582 dif, 0, &udp_table, NULL); 583 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 584 sk = NULL; 585 return sk; 586 } 587 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 588 #endif 589 590 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 591 __be16 loc_port, __be32 loc_addr, 592 __be16 rmt_port, __be32 rmt_addr, 593 int dif, int sdif, unsigned short hnum) 594 { 595 struct inet_sock *inet = inet_sk(sk); 596 597 if (!net_eq(sock_net(sk), net) || 598 udp_sk(sk)->udp_port_hash != hnum || 599 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 600 (inet->inet_dport != rmt_port && inet->inet_dport) || 601 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 602 ipv6_only_sock(sk) || 603 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif && 604 sk->sk_bound_dev_if != sdif)) 605 return false; 606 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 607 return false; 608 return true; 609 } 610 611 /* 612 * This routine is called by the ICMP module when it gets some 613 * sort of error condition. If err < 0 then the socket should 614 * be closed and the error returned to the user. If err > 0 615 * it's just the icmp type << 8 | icmp code. 616 * Header points to the ip header of the error packet. We move 617 * on past this. Then (as it used to claim before adjustment) 618 * header points to the first 8 bytes of the udp header. We need 619 * to find the appropriate port. 620 */ 621 622 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 623 { 624 struct inet_sock *inet; 625 const struct iphdr *iph = (const struct iphdr *)skb->data; 626 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 627 const int type = icmp_hdr(skb)->type; 628 const int code = icmp_hdr(skb)->code; 629 struct sock *sk; 630 int harderr; 631 int err; 632 struct net *net = dev_net(skb->dev); 633 634 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 635 iph->saddr, uh->source, skb->dev->ifindex, 0, 636 udptable, NULL); 637 if (!sk) { 638 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 639 return; /* No socket for error */ 640 } 641 642 err = 0; 643 harderr = 0; 644 inet = inet_sk(sk); 645 646 switch (type) { 647 default: 648 case ICMP_TIME_EXCEEDED: 649 err = EHOSTUNREACH; 650 break; 651 case ICMP_SOURCE_QUENCH: 652 goto out; 653 case ICMP_PARAMETERPROB: 654 err = EPROTO; 655 harderr = 1; 656 break; 657 case ICMP_DEST_UNREACH: 658 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 659 ipv4_sk_update_pmtu(skb, sk, info); 660 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 661 err = EMSGSIZE; 662 harderr = 1; 663 break; 664 } 665 goto out; 666 } 667 err = EHOSTUNREACH; 668 if (code <= NR_ICMP_UNREACH) { 669 harderr = icmp_err_convert[code].fatal; 670 err = icmp_err_convert[code].errno; 671 } 672 break; 673 case ICMP_REDIRECT: 674 ipv4_sk_redirect(skb, sk); 675 goto out; 676 } 677 678 /* 679 * RFC1122: OK. Passes ICMP errors back to application, as per 680 * 4.1.3.3. 681 */ 682 if (!inet->recverr) { 683 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 684 goto out; 685 } else 686 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 687 688 sk->sk_err = err; 689 sk->sk_error_report(sk); 690 out: 691 return; 692 } 693 694 void udp_err(struct sk_buff *skb, u32 info) 695 { 696 __udp4_lib_err(skb, info, &udp_table); 697 } 698 699 /* 700 * Throw away all pending data and cancel the corking. Socket is locked. 701 */ 702 void udp_flush_pending_frames(struct sock *sk) 703 { 704 struct udp_sock *up = udp_sk(sk); 705 706 if (up->pending) { 707 up->len = 0; 708 up->pending = 0; 709 ip_flush_pending_frames(sk); 710 } 711 } 712 EXPORT_SYMBOL(udp_flush_pending_frames); 713 714 /** 715 * udp4_hwcsum - handle outgoing HW checksumming 716 * @skb: sk_buff containing the filled-in UDP header 717 * (checksum field must be zeroed out) 718 * @src: source IP address 719 * @dst: destination IP address 720 */ 721 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 722 { 723 struct udphdr *uh = udp_hdr(skb); 724 int offset = skb_transport_offset(skb); 725 int len = skb->len - offset; 726 int hlen = len; 727 __wsum csum = 0; 728 729 if (!skb_has_frag_list(skb)) { 730 /* 731 * Only one fragment on the socket. 732 */ 733 skb->csum_start = skb_transport_header(skb) - skb->head; 734 skb->csum_offset = offsetof(struct udphdr, check); 735 uh->check = ~csum_tcpudp_magic(src, dst, len, 736 IPPROTO_UDP, 0); 737 } else { 738 struct sk_buff *frags; 739 740 /* 741 * HW-checksum won't work as there are two or more 742 * fragments on the socket so that all csums of sk_buffs 743 * should be together 744 */ 745 skb_walk_frags(skb, frags) { 746 csum = csum_add(csum, frags->csum); 747 hlen -= frags->len; 748 } 749 750 csum = skb_checksum(skb, offset, hlen, csum); 751 skb->ip_summed = CHECKSUM_NONE; 752 753 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 754 if (uh->check == 0) 755 uh->check = CSUM_MANGLED_0; 756 } 757 } 758 EXPORT_SYMBOL_GPL(udp4_hwcsum); 759 760 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 761 * for the simple case like when setting the checksum for a UDP tunnel. 762 */ 763 void udp_set_csum(bool nocheck, struct sk_buff *skb, 764 __be32 saddr, __be32 daddr, int len) 765 { 766 struct udphdr *uh = udp_hdr(skb); 767 768 if (nocheck) { 769 uh->check = 0; 770 } else if (skb_is_gso(skb)) { 771 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 772 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 773 uh->check = 0; 774 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 775 if (uh->check == 0) 776 uh->check = CSUM_MANGLED_0; 777 } else { 778 skb->ip_summed = CHECKSUM_PARTIAL; 779 skb->csum_start = skb_transport_header(skb) - skb->head; 780 skb->csum_offset = offsetof(struct udphdr, check); 781 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 782 } 783 } 784 EXPORT_SYMBOL(udp_set_csum); 785 786 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 787 { 788 struct sock *sk = skb->sk; 789 struct inet_sock *inet = inet_sk(sk); 790 struct udphdr *uh; 791 int err = 0; 792 int is_udplite = IS_UDPLITE(sk); 793 int offset = skb_transport_offset(skb); 794 int len = skb->len - offset; 795 __wsum csum = 0; 796 797 /* 798 * Create a UDP header 799 */ 800 uh = udp_hdr(skb); 801 uh->source = inet->inet_sport; 802 uh->dest = fl4->fl4_dport; 803 uh->len = htons(len); 804 uh->check = 0; 805 806 if (is_udplite) /* UDP-Lite */ 807 csum = udplite_csum(skb); 808 809 else if (sk->sk_no_check_tx) { /* UDP csum off */ 810 811 skb->ip_summed = CHECKSUM_NONE; 812 goto send; 813 814 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 815 816 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 817 goto send; 818 819 } else 820 csum = udp_csum(skb); 821 822 /* add protocol-dependent pseudo-header */ 823 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 824 sk->sk_protocol, csum); 825 if (uh->check == 0) 826 uh->check = CSUM_MANGLED_0; 827 828 send: 829 err = ip_send_skb(sock_net(sk), skb); 830 if (err) { 831 if (err == -ENOBUFS && !inet->recverr) { 832 UDP_INC_STATS(sock_net(sk), 833 UDP_MIB_SNDBUFERRORS, is_udplite); 834 err = 0; 835 } 836 } else 837 UDP_INC_STATS(sock_net(sk), 838 UDP_MIB_OUTDATAGRAMS, is_udplite); 839 return err; 840 } 841 842 /* 843 * Push out all pending data as one UDP datagram. Socket is locked. 844 */ 845 int udp_push_pending_frames(struct sock *sk) 846 { 847 struct udp_sock *up = udp_sk(sk); 848 struct inet_sock *inet = inet_sk(sk); 849 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 850 struct sk_buff *skb; 851 int err = 0; 852 853 skb = ip_finish_skb(sk, fl4); 854 if (!skb) 855 goto out; 856 857 err = udp_send_skb(skb, fl4); 858 859 out: 860 up->len = 0; 861 up->pending = 0; 862 return err; 863 } 864 EXPORT_SYMBOL(udp_push_pending_frames); 865 866 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 867 { 868 struct inet_sock *inet = inet_sk(sk); 869 struct udp_sock *up = udp_sk(sk); 870 struct flowi4 fl4_stack; 871 struct flowi4 *fl4; 872 int ulen = len; 873 struct ipcm_cookie ipc; 874 struct rtable *rt = NULL; 875 int free = 0; 876 int connected = 0; 877 __be32 daddr, faddr, saddr; 878 __be16 dport; 879 u8 tos; 880 int err, is_udplite = IS_UDPLITE(sk); 881 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 882 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 883 struct sk_buff *skb; 884 struct ip_options_data opt_copy; 885 886 if (len > 0xFFFF) 887 return -EMSGSIZE; 888 889 /* 890 * Check the flags. 891 */ 892 893 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 894 return -EOPNOTSUPP; 895 896 ipc.opt = NULL; 897 ipc.tx_flags = 0; 898 ipc.ttl = 0; 899 ipc.tos = -1; 900 901 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 902 903 fl4 = &inet->cork.fl.u.ip4; 904 if (up->pending) { 905 /* 906 * There are pending frames. 907 * The socket lock must be held while it's corked. 908 */ 909 lock_sock(sk); 910 if (likely(up->pending)) { 911 if (unlikely(up->pending != AF_INET)) { 912 release_sock(sk); 913 return -EINVAL; 914 } 915 goto do_append_data; 916 } 917 release_sock(sk); 918 } 919 ulen += sizeof(struct udphdr); 920 921 /* 922 * Get and verify the address. 923 */ 924 if (msg->msg_name) { 925 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 926 if (msg->msg_namelen < sizeof(*usin)) 927 return -EINVAL; 928 if (usin->sin_family != AF_INET) { 929 if (usin->sin_family != AF_UNSPEC) 930 return -EAFNOSUPPORT; 931 } 932 933 daddr = usin->sin_addr.s_addr; 934 dport = usin->sin_port; 935 if (dport == 0) 936 return -EINVAL; 937 } else { 938 if (sk->sk_state != TCP_ESTABLISHED) 939 return -EDESTADDRREQ; 940 daddr = inet->inet_daddr; 941 dport = inet->inet_dport; 942 /* Open fast path for connected socket. 943 Route will not be used, if at least one option is set. 944 */ 945 connected = 1; 946 } 947 948 ipc.sockc.tsflags = sk->sk_tsflags; 949 ipc.addr = inet->inet_saddr; 950 ipc.oif = sk->sk_bound_dev_if; 951 952 if (msg->msg_controllen) { 953 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6); 954 if (unlikely(err)) { 955 kfree(ipc.opt); 956 return err; 957 } 958 if (ipc.opt) 959 free = 1; 960 connected = 0; 961 } 962 if (!ipc.opt) { 963 struct ip_options_rcu *inet_opt; 964 965 rcu_read_lock(); 966 inet_opt = rcu_dereference(inet->inet_opt); 967 if (inet_opt) { 968 memcpy(&opt_copy, inet_opt, 969 sizeof(*inet_opt) + inet_opt->opt.optlen); 970 ipc.opt = &opt_copy.opt; 971 } 972 rcu_read_unlock(); 973 } 974 975 saddr = ipc.addr; 976 ipc.addr = faddr = daddr; 977 978 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags); 979 980 if (ipc.opt && ipc.opt->opt.srr) { 981 if (!daddr) 982 return -EINVAL; 983 faddr = ipc.opt->opt.faddr; 984 connected = 0; 985 } 986 tos = get_rttos(&ipc, inet); 987 if (sock_flag(sk, SOCK_LOCALROUTE) || 988 (msg->msg_flags & MSG_DONTROUTE) || 989 (ipc.opt && ipc.opt->opt.is_strictroute)) { 990 tos |= RTO_ONLINK; 991 connected = 0; 992 } 993 994 if (ipv4_is_multicast(daddr)) { 995 if (!ipc.oif) 996 ipc.oif = inet->mc_index; 997 if (!saddr) 998 saddr = inet->mc_addr; 999 connected = 0; 1000 } else if (!ipc.oif) 1001 ipc.oif = inet->uc_index; 1002 1003 if (connected) 1004 rt = (struct rtable *)sk_dst_check(sk, 0); 1005 1006 if (!rt) { 1007 struct net *net = sock_net(sk); 1008 __u8 flow_flags = inet_sk_flowi_flags(sk); 1009 1010 fl4 = &fl4_stack; 1011 1012 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1013 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1014 flow_flags, 1015 faddr, saddr, dport, inet->inet_sport, 1016 sk->sk_uid); 1017 1018 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1019 rt = ip_route_output_flow(net, fl4, sk); 1020 if (IS_ERR(rt)) { 1021 err = PTR_ERR(rt); 1022 rt = NULL; 1023 if (err == -ENETUNREACH) 1024 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1025 goto out; 1026 } 1027 1028 err = -EACCES; 1029 if ((rt->rt_flags & RTCF_BROADCAST) && 1030 !sock_flag(sk, SOCK_BROADCAST)) 1031 goto out; 1032 if (connected) 1033 sk_dst_set(sk, dst_clone(&rt->dst)); 1034 } 1035 1036 if (msg->msg_flags&MSG_CONFIRM) 1037 goto do_confirm; 1038 back_from_confirm: 1039 1040 saddr = fl4->saddr; 1041 if (!ipc.addr) 1042 daddr = ipc.addr = fl4->daddr; 1043 1044 /* Lockless fast path for the non-corking case. */ 1045 if (!corkreq) { 1046 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1047 sizeof(struct udphdr), &ipc, &rt, 1048 msg->msg_flags); 1049 err = PTR_ERR(skb); 1050 if (!IS_ERR_OR_NULL(skb)) 1051 err = udp_send_skb(skb, fl4); 1052 goto out; 1053 } 1054 1055 lock_sock(sk); 1056 if (unlikely(up->pending)) { 1057 /* The socket is already corked while preparing it. */ 1058 /* ... which is an evident application bug. --ANK */ 1059 release_sock(sk); 1060 1061 net_dbg_ratelimited("socket already corked\n"); 1062 err = -EINVAL; 1063 goto out; 1064 } 1065 /* 1066 * Now cork the socket to pend data. 1067 */ 1068 fl4 = &inet->cork.fl.u.ip4; 1069 fl4->daddr = daddr; 1070 fl4->saddr = saddr; 1071 fl4->fl4_dport = dport; 1072 fl4->fl4_sport = inet->inet_sport; 1073 up->pending = AF_INET; 1074 1075 do_append_data: 1076 up->len += ulen; 1077 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1078 sizeof(struct udphdr), &ipc, &rt, 1079 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1080 if (err) 1081 udp_flush_pending_frames(sk); 1082 else if (!corkreq) 1083 err = udp_push_pending_frames(sk); 1084 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1085 up->pending = 0; 1086 release_sock(sk); 1087 1088 out: 1089 ip_rt_put(rt); 1090 if (free) 1091 kfree(ipc.opt); 1092 if (!err) 1093 return len; 1094 /* 1095 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1096 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1097 * we don't have a good statistic (IpOutDiscards but it can be too many 1098 * things). We could add another new stat but at least for now that 1099 * seems like overkill. 1100 */ 1101 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1102 UDP_INC_STATS(sock_net(sk), 1103 UDP_MIB_SNDBUFERRORS, is_udplite); 1104 } 1105 return err; 1106 1107 do_confirm: 1108 if (msg->msg_flags & MSG_PROBE) 1109 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1110 if (!(msg->msg_flags&MSG_PROBE) || len) 1111 goto back_from_confirm; 1112 err = 0; 1113 goto out; 1114 } 1115 EXPORT_SYMBOL(udp_sendmsg); 1116 1117 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1118 size_t size, int flags) 1119 { 1120 struct inet_sock *inet = inet_sk(sk); 1121 struct udp_sock *up = udp_sk(sk); 1122 int ret; 1123 1124 if (flags & MSG_SENDPAGE_NOTLAST) 1125 flags |= MSG_MORE; 1126 1127 if (!up->pending) { 1128 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1129 1130 /* Call udp_sendmsg to specify destination address which 1131 * sendpage interface can't pass. 1132 * This will succeed only when the socket is connected. 1133 */ 1134 ret = udp_sendmsg(sk, &msg, 0); 1135 if (ret < 0) 1136 return ret; 1137 } 1138 1139 lock_sock(sk); 1140 1141 if (unlikely(!up->pending)) { 1142 release_sock(sk); 1143 1144 net_dbg_ratelimited("cork failed\n"); 1145 return -EINVAL; 1146 } 1147 1148 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1149 page, offset, size, flags); 1150 if (ret == -EOPNOTSUPP) { 1151 release_sock(sk); 1152 return sock_no_sendpage(sk->sk_socket, page, offset, 1153 size, flags); 1154 } 1155 if (ret < 0) { 1156 udp_flush_pending_frames(sk); 1157 goto out; 1158 } 1159 1160 up->len += size; 1161 if (!(up->corkflag || (flags&MSG_MORE))) 1162 ret = udp_push_pending_frames(sk); 1163 if (!ret) 1164 ret = size; 1165 out: 1166 release_sock(sk); 1167 return ret; 1168 } 1169 1170 #define UDP_SKB_IS_STATELESS 0x80000000 1171 1172 static void udp_set_dev_scratch(struct sk_buff *skb) 1173 { 1174 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1175 1176 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1177 scratch->_tsize_state = skb->truesize; 1178 #if BITS_PER_LONG == 64 1179 scratch->len = skb->len; 1180 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1181 scratch->is_linear = !skb_is_nonlinear(skb); 1182 #endif 1183 /* all head states execept sp (dst, sk, nf) are always cleared by 1184 * udp_rcv() and we need to preserve secpath, if present, to eventually 1185 * process IP_CMSG_PASSSEC at recvmsg() time 1186 */ 1187 if (likely(!skb_sec_path(skb))) 1188 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1189 } 1190 1191 static int udp_skb_truesize(struct sk_buff *skb) 1192 { 1193 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1194 } 1195 1196 static bool udp_skb_has_head_state(struct sk_buff *skb) 1197 { 1198 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1199 } 1200 1201 /* fully reclaim rmem/fwd memory allocated for skb */ 1202 static void udp_rmem_release(struct sock *sk, int size, int partial, 1203 bool rx_queue_lock_held) 1204 { 1205 struct udp_sock *up = udp_sk(sk); 1206 struct sk_buff_head *sk_queue; 1207 int amt; 1208 1209 if (likely(partial)) { 1210 up->forward_deficit += size; 1211 size = up->forward_deficit; 1212 if (size < (sk->sk_rcvbuf >> 2)) 1213 return; 1214 } else { 1215 size += up->forward_deficit; 1216 } 1217 up->forward_deficit = 0; 1218 1219 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1220 * if the called don't held it already 1221 */ 1222 sk_queue = &sk->sk_receive_queue; 1223 if (!rx_queue_lock_held) 1224 spin_lock(&sk_queue->lock); 1225 1226 1227 sk->sk_forward_alloc += size; 1228 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); 1229 sk->sk_forward_alloc -= amt; 1230 1231 if (amt) 1232 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); 1233 1234 atomic_sub(size, &sk->sk_rmem_alloc); 1235 1236 /* this can save us from acquiring the rx queue lock on next receive */ 1237 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1238 1239 if (!rx_queue_lock_held) 1240 spin_unlock(&sk_queue->lock); 1241 } 1242 1243 /* Note: called with reader_queue.lock held. 1244 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1245 * This avoids a cache line miss while receive_queue lock is held. 1246 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1247 */ 1248 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1249 { 1250 prefetch(&skb->data); 1251 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1252 } 1253 EXPORT_SYMBOL(udp_skb_destructor); 1254 1255 /* as above, but the caller held the rx queue lock, too */ 1256 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1257 { 1258 prefetch(&skb->data); 1259 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1260 } 1261 1262 /* Idea of busylocks is to let producers grab an extra spinlock 1263 * to relieve pressure on the receive_queue spinlock shared by consumer. 1264 * Under flood, this means that only one producer can be in line 1265 * trying to acquire the receive_queue spinlock. 1266 * These busylock can be allocated on a per cpu manner, instead of a 1267 * per socket one (that would consume a cache line per socket) 1268 */ 1269 static int udp_busylocks_log __read_mostly; 1270 static spinlock_t *udp_busylocks __read_mostly; 1271 1272 static spinlock_t *busylock_acquire(void *ptr) 1273 { 1274 spinlock_t *busy; 1275 1276 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1277 spin_lock(busy); 1278 return busy; 1279 } 1280 1281 static void busylock_release(spinlock_t *busy) 1282 { 1283 if (busy) 1284 spin_unlock(busy); 1285 } 1286 1287 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1288 { 1289 struct sk_buff_head *list = &sk->sk_receive_queue; 1290 int rmem, delta, amt, err = -ENOMEM; 1291 spinlock_t *busy = NULL; 1292 int size; 1293 1294 /* try to avoid the costly atomic add/sub pair when the receive 1295 * queue is full; always allow at least a packet 1296 */ 1297 rmem = atomic_read(&sk->sk_rmem_alloc); 1298 if (rmem > sk->sk_rcvbuf) 1299 goto drop; 1300 1301 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1302 * having linear skbs : 1303 * - Reduce memory overhead and thus increase receive queue capacity 1304 * - Less cache line misses at copyout() time 1305 * - Less work at consume_skb() (less alien page frag freeing) 1306 */ 1307 if (rmem > (sk->sk_rcvbuf >> 1)) { 1308 skb_condense(skb); 1309 1310 busy = busylock_acquire(sk); 1311 } 1312 size = skb->truesize; 1313 udp_set_dev_scratch(skb); 1314 1315 /* we drop only if the receive buf is full and the receive 1316 * queue contains some other skb 1317 */ 1318 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1319 if (rmem > (size + sk->sk_rcvbuf)) 1320 goto uncharge_drop; 1321 1322 spin_lock(&list->lock); 1323 if (size >= sk->sk_forward_alloc) { 1324 amt = sk_mem_pages(size); 1325 delta = amt << SK_MEM_QUANTUM_SHIFT; 1326 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1327 err = -ENOBUFS; 1328 spin_unlock(&list->lock); 1329 goto uncharge_drop; 1330 } 1331 1332 sk->sk_forward_alloc += delta; 1333 } 1334 1335 sk->sk_forward_alloc -= size; 1336 1337 /* no need to setup a destructor, we will explicitly release the 1338 * forward allocated memory on dequeue 1339 */ 1340 sock_skb_set_dropcount(sk, skb); 1341 1342 __skb_queue_tail(list, skb); 1343 spin_unlock(&list->lock); 1344 1345 if (!sock_flag(sk, SOCK_DEAD)) 1346 sk->sk_data_ready(sk); 1347 1348 busylock_release(busy); 1349 return 0; 1350 1351 uncharge_drop: 1352 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1353 1354 drop: 1355 atomic_inc(&sk->sk_drops); 1356 busylock_release(busy); 1357 return err; 1358 } 1359 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1360 1361 void udp_destruct_sock(struct sock *sk) 1362 { 1363 /* reclaim completely the forward allocated memory */ 1364 struct udp_sock *up = udp_sk(sk); 1365 unsigned int total = 0; 1366 struct sk_buff *skb; 1367 1368 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1369 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1370 total += skb->truesize; 1371 kfree_skb(skb); 1372 } 1373 udp_rmem_release(sk, total, 0, true); 1374 1375 inet_sock_destruct(sk); 1376 } 1377 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1378 1379 int udp_init_sock(struct sock *sk) 1380 { 1381 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1382 sk->sk_destruct = udp_destruct_sock; 1383 return 0; 1384 } 1385 EXPORT_SYMBOL_GPL(udp_init_sock); 1386 1387 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1388 { 1389 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1390 bool slow = lock_sock_fast(sk); 1391 1392 sk_peek_offset_bwd(sk, len); 1393 unlock_sock_fast(sk, slow); 1394 } 1395 1396 if (!skb_unref(skb)) 1397 return; 1398 1399 /* In the more common cases we cleared the head states previously, 1400 * see __udp_queue_rcv_skb(). 1401 */ 1402 if (unlikely(udp_skb_has_head_state(skb))) 1403 skb_release_head_state(skb); 1404 __consume_stateless_skb(skb); 1405 } 1406 EXPORT_SYMBOL_GPL(skb_consume_udp); 1407 1408 static struct sk_buff *__first_packet_length(struct sock *sk, 1409 struct sk_buff_head *rcvq, 1410 int *total) 1411 { 1412 struct sk_buff *skb; 1413 1414 while ((skb = skb_peek(rcvq)) != NULL) { 1415 if (udp_lib_checksum_complete(skb)) { 1416 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1417 IS_UDPLITE(sk)); 1418 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1419 IS_UDPLITE(sk)); 1420 atomic_inc(&sk->sk_drops); 1421 __skb_unlink(skb, rcvq); 1422 *total += skb->truesize; 1423 kfree_skb(skb); 1424 } else { 1425 /* the csum related bits could be changed, refresh 1426 * the scratch area 1427 */ 1428 udp_set_dev_scratch(skb); 1429 break; 1430 } 1431 } 1432 return skb; 1433 } 1434 1435 /** 1436 * first_packet_length - return length of first packet in receive queue 1437 * @sk: socket 1438 * 1439 * Drops all bad checksum frames, until a valid one is found. 1440 * Returns the length of found skb, or -1 if none is found. 1441 */ 1442 static int first_packet_length(struct sock *sk) 1443 { 1444 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1445 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1446 struct sk_buff *skb; 1447 int total = 0; 1448 int res; 1449 1450 spin_lock_bh(&rcvq->lock); 1451 skb = __first_packet_length(sk, rcvq, &total); 1452 if (!skb && !skb_queue_empty(sk_queue)) { 1453 spin_lock(&sk_queue->lock); 1454 skb_queue_splice_tail_init(sk_queue, rcvq); 1455 spin_unlock(&sk_queue->lock); 1456 1457 skb = __first_packet_length(sk, rcvq, &total); 1458 } 1459 res = skb ? skb->len : -1; 1460 if (total) 1461 udp_rmem_release(sk, total, 1, false); 1462 spin_unlock_bh(&rcvq->lock); 1463 return res; 1464 } 1465 1466 /* 1467 * IOCTL requests applicable to the UDP protocol 1468 */ 1469 1470 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1471 { 1472 switch (cmd) { 1473 case SIOCOUTQ: 1474 { 1475 int amount = sk_wmem_alloc_get(sk); 1476 1477 return put_user(amount, (int __user *)arg); 1478 } 1479 1480 case SIOCINQ: 1481 { 1482 int amount = max_t(int, 0, first_packet_length(sk)); 1483 1484 return put_user(amount, (int __user *)arg); 1485 } 1486 1487 default: 1488 return -ENOIOCTLCMD; 1489 } 1490 1491 return 0; 1492 } 1493 EXPORT_SYMBOL(udp_ioctl); 1494 1495 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1496 int noblock, int *peeked, int *off, int *err) 1497 { 1498 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1499 struct sk_buff_head *queue; 1500 struct sk_buff *last; 1501 long timeo; 1502 int error; 1503 1504 queue = &udp_sk(sk)->reader_queue; 1505 flags |= noblock ? MSG_DONTWAIT : 0; 1506 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1507 do { 1508 struct sk_buff *skb; 1509 1510 error = sock_error(sk); 1511 if (error) 1512 break; 1513 1514 error = -EAGAIN; 1515 *peeked = 0; 1516 do { 1517 spin_lock_bh(&queue->lock); 1518 skb = __skb_try_recv_from_queue(sk, queue, flags, 1519 udp_skb_destructor, 1520 peeked, off, err, 1521 &last); 1522 if (skb) { 1523 spin_unlock_bh(&queue->lock); 1524 return skb; 1525 } 1526 1527 if (skb_queue_empty(sk_queue)) { 1528 spin_unlock_bh(&queue->lock); 1529 goto busy_check; 1530 } 1531 1532 /* refill the reader queue and walk it again 1533 * keep both queues locked to avoid re-acquiring 1534 * the sk_receive_queue lock if fwd memory scheduling 1535 * is needed. 1536 */ 1537 spin_lock(&sk_queue->lock); 1538 skb_queue_splice_tail_init(sk_queue, queue); 1539 1540 skb = __skb_try_recv_from_queue(sk, queue, flags, 1541 udp_skb_dtor_locked, 1542 peeked, off, err, 1543 &last); 1544 spin_unlock(&sk_queue->lock); 1545 spin_unlock_bh(&queue->lock); 1546 if (skb) 1547 return skb; 1548 1549 busy_check: 1550 if (!sk_can_busy_loop(sk)) 1551 break; 1552 1553 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1554 } while (!skb_queue_empty(sk_queue)); 1555 1556 /* sk_queue is empty, reader_queue may contain peeked packets */ 1557 } while (timeo && 1558 !__skb_wait_for_more_packets(sk, &error, &timeo, 1559 (struct sk_buff *)sk_queue)); 1560 1561 *err = error; 1562 return NULL; 1563 } 1564 EXPORT_SYMBOL_GPL(__skb_recv_udp); 1565 1566 /* 1567 * This should be easy, if there is something there we 1568 * return it, otherwise we block. 1569 */ 1570 1571 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1572 int flags, int *addr_len) 1573 { 1574 struct inet_sock *inet = inet_sk(sk); 1575 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1576 struct sk_buff *skb; 1577 unsigned int ulen, copied; 1578 int peeked, peeking, off; 1579 int err; 1580 int is_udplite = IS_UDPLITE(sk); 1581 bool checksum_valid = false; 1582 1583 if (flags & MSG_ERRQUEUE) 1584 return ip_recv_error(sk, msg, len, addr_len); 1585 1586 try_again: 1587 peeking = flags & MSG_PEEK; 1588 off = sk_peek_offset(sk, flags); 1589 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); 1590 if (!skb) 1591 return err; 1592 1593 ulen = udp_skb_len(skb); 1594 copied = len; 1595 if (copied > ulen - off) 1596 copied = ulen - off; 1597 else if (copied < ulen) 1598 msg->msg_flags |= MSG_TRUNC; 1599 1600 /* 1601 * If checksum is needed at all, try to do it while copying the 1602 * data. If the data is truncated, or if we only want a partial 1603 * coverage checksum (UDP-Lite), do it before the copy. 1604 */ 1605 1606 if (copied < ulen || peeking || 1607 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1608 checksum_valid = udp_skb_csum_unnecessary(skb) || 1609 !__udp_lib_checksum_complete(skb); 1610 if (!checksum_valid) 1611 goto csum_copy_err; 1612 } 1613 1614 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1615 if (udp_skb_is_linear(skb)) 1616 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1617 else 1618 err = skb_copy_datagram_msg(skb, off, msg, copied); 1619 } else { 1620 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1621 1622 if (err == -EINVAL) 1623 goto csum_copy_err; 1624 } 1625 1626 if (unlikely(err)) { 1627 if (!peeked) { 1628 atomic_inc(&sk->sk_drops); 1629 UDP_INC_STATS(sock_net(sk), 1630 UDP_MIB_INERRORS, is_udplite); 1631 } 1632 kfree_skb(skb); 1633 return err; 1634 } 1635 1636 if (!peeked) 1637 UDP_INC_STATS(sock_net(sk), 1638 UDP_MIB_INDATAGRAMS, is_udplite); 1639 1640 sock_recv_ts_and_drops(msg, sk, skb); 1641 1642 /* Copy the address. */ 1643 if (sin) { 1644 sin->sin_family = AF_INET; 1645 sin->sin_port = udp_hdr(skb)->source; 1646 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1647 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1648 *addr_len = sizeof(*sin); 1649 } 1650 if (inet->cmsg_flags) 1651 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1652 1653 err = copied; 1654 if (flags & MSG_TRUNC) 1655 err = ulen; 1656 1657 skb_consume_udp(sk, skb, peeking ? -err : err); 1658 return err; 1659 1660 csum_copy_err: 1661 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1662 udp_skb_destructor)) { 1663 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1664 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1665 } 1666 kfree_skb(skb); 1667 1668 /* starting over for a new packet, but check if we need to yield */ 1669 cond_resched(); 1670 msg->msg_flags &= ~MSG_TRUNC; 1671 goto try_again; 1672 } 1673 1674 int __udp_disconnect(struct sock *sk, int flags) 1675 { 1676 struct inet_sock *inet = inet_sk(sk); 1677 /* 1678 * 1003.1g - break association. 1679 */ 1680 1681 sk->sk_state = TCP_CLOSE; 1682 inet->inet_daddr = 0; 1683 inet->inet_dport = 0; 1684 sock_rps_reset_rxhash(sk); 1685 sk->sk_bound_dev_if = 0; 1686 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1687 inet_reset_saddr(sk); 1688 1689 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1690 sk->sk_prot->unhash(sk); 1691 inet->inet_sport = 0; 1692 } 1693 sk_dst_reset(sk); 1694 return 0; 1695 } 1696 EXPORT_SYMBOL(__udp_disconnect); 1697 1698 int udp_disconnect(struct sock *sk, int flags) 1699 { 1700 lock_sock(sk); 1701 __udp_disconnect(sk, flags); 1702 release_sock(sk); 1703 return 0; 1704 } 1705 EXPORT_SYMBOL(udp_disconnect); 1706 1707 void udp_lib_unhash(struct sock *sk) 1708 { 1709 if (sk_hashed(sk)) { 1710 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1711 struct udp_hslot *hslot, *hslot2; 1712 1713 hslot = udp_hashslot(udptable, sock_net(sk), 1714 udp_sk(sk)->udp_port_hash); 1715 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1716 1717 spin_lock_bh(&hslot->lock); 1718 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1719 reuseport_detach_sock(sk); 1720 if (sk_del_node_init_rcu(sk)) { 1721 hslot->count--; 1722 inet_sk(sk)->inet_num = 0; 1723 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1724 1725 spin_lock(&hslot2->lock); 1726 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1727 hslot2->count--; 1728 spin_unlock(&hslot2->lock); 1729 } 1730 spin_unlock_bh(&hslot->lock); 1731 } 1732 } 1733 EXPORT_SYMBOL(udp_lib_unhash); 1734 1735 /* 1736 * inet_rcv_saddr was changed, we must rehash secondary hash 1737 */ 1738 void udp_lib_rehash(struct sock *sk, u16 newhash) 1739 { 1740 if (sk_hashed(sk)) { 1741 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1742 struct udp_hslot *hslot, *hslot2, *nhslot2; 1743 1744 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1745 nhslot2 = udp_hashslot2(udptable, newhash); 1746 udp_sk(sk)->udp_portaddr_hash = newhash; 1747 1748 if (hslot2 != nhslot2 || 1749 rcu_access_pointer(sk->sk_reuseport_cb)) { 1750 hslot = udp_hashslot(udptable, sock_net(sk), 1751 udp_sk(sk)->udp_port_hash); 1752 /* we must lock primary chain too */ 1753 spin_lock_bh(&hslot->lock); 1754 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1755 reuseport_detach_sock(sk); 1756 1757 if (hslot2 != nhslot2) { 1758 spin_lock(&hslot2->lock); 1759 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1760 hslot2->count--; 1761 spin_unlock(&hslot2->lock); 1762 1763 spin_lock(&nhslot2->lock); 1764 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1765 &nhslot2->head); 1766 nhslot2->count++; 1767 spin_unlock(&nhslot2->lock); 1768 } 1769 1770 spin_unlock_bh(&hslot->lock); 1771 } 1772 } 1773 } 1774 EXPORT_SYMBOL(udp_lib_rehash); 1775 1776 static void udp_v4_rehash(struct sock *sk) 1777 { 1778 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1779 inet_sk(sk)->inet_rcv_saddr, 1780 inet_sk(sk)->inet_num); 1781 udp_lib_rehash(sk, new_hash); 1782 } 1783 1784 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1785 { 1786 int rc; 1787 1788 if (inet_sk(sk)->inet_daddr) { 1789 sock_rps_save_rxhash(sk, skb); 1790 sk_mark_napi_id(sk, skb); 1791 sk_incoming_cpu_update(sk); 1792 } else { 1793 sk_mark_napi_id_once(sk, skb); 1794 } 1795 1796 rc = __udp_enqueue_schedule_skb(sk, skb); 1797 if (rc < 0) { 1798 int is_udplite = IS_UDPLITE(sk); 1799 1800 /* Note that an ENOMEM error is charged twice */ 1801 if (rc == -ENOMEM) 1802 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1803 is_udplite); 1804 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1805 kfree_skb(skb); 1806 trace_udp_fail_queue_rcv_skb(rc, sk); 1807 return -1; 1808 } 1809 1810 return 0; 1811 } 1812 1813 static struct static_key udp_encap_needed __read_mostly; 1814 void udp_encap_enable(void) 1815 { 1816 static_key_enable(&udp_encap_needed); 1817 } 1818 EXPORT_SYMBOL(udp_encap_enable); 1819 1820 /* returns: 1821 * -1: error 1822 * 0: success 1823 * >0: "udp encap" protocol resubmission 1824 * 1825 * Note that in the success and error cases, the skb is assumed to 1826 * have either been requeued or freed. 1827 */ 1828 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1829 { 1830 struct udp_sock *up = udp_sk(sk); 1831 int is_udplite = IS_UDPLITE(sk); 1832 1833 /* 1834 * Charge it to the socket, dropping if the queue is full. 1835 */ 1836 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1837 goto drop; 1838 nf_reset(skb); 1839 1840 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1841 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1842 1843 /* 1844 * This is an encapsulation socket so pass the skb to 1845 * the socket's udp_encap_rcv() hook. Otherwise, just 1846 * fall through and pass this up the UDP socket. 1847 * up->encap_rcv() returns the following value: 1848 * =0 if skb was successfully passed to the encap 1849 * handler or was discarded by it. 1850 * >0 if skb should be passed on to UDP. 1851 * <0 if skb should be resubmitted as proto -N 1852 */ 1853 1854 /* if we're overly short, let UDP handle it */ 1855 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1856 if (encap_rcv) { 1857 int ret; 1858 1859 /* Verify checksum before giving to encap */ 1860 if (udp_lib_checksum_complete(skb)) 1861 goto csum_error; 1862 1863 ret = encap_rcv(sk, skb); 1864 if (ret <= 0) { 1865 __UDP_INC_STATS(sock_net(sk), 1866 UDP_MIB_INDATAGRAMS, 1867 is_udplite); 1868 return -ret; 1869 } 1870 } 1871 1872 /* FALLTHROUGH -- it's a UDP Packet */ 1873 } 1874 1875 /* 1876 * UDP-Lite specific tests, ignored on UDP sockets 1877 */ 1878 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1879 1880 /* 1881 * MIB statistics other than incrementing the error count are 1882 * disabled for the following two types of errors: these depend 1883 * on the application settings, not on the functioning of the 1884 * protocol stack as such. 1885 * 1886 * RFC 3828 here recommends (sec 3.3): "There should also be a 1887 * way ... to ... at least let the receiving application block 1888 * delivery of packets with coverage values less than a value 1889 * provided by the application." 1890 */ 1891 if (up->pcrlen == 0) { /* full coverage was set */ 1892 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1893 UDP_SKB_CB(skb)->cscov, skb->len); 1894 goto drop; 1895 } 1896 /* The next case involves violating the min. coverage requested 1897 * by the receiver. This is subtle: if receiver wants x and x is 1898 * greater than the buffersize/MTU then receiver will complain 1899 * that it wants x while sender emits packets of smaller size y. 1900 * Therefore the above ...()->partial_cov statement is essential. 1901 */ 1902 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1903 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1904 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1905 goto drop; 1906 } 1907 } 1908 1909 prefetch(&sk->sk_rmem_alloc); 1910 if (rcu_access_pointer(sk->sk_filter) && 1911 udp_lib_checksum_complete(skb)) 1912 goto csum_error; 1913 1914 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 1915 goto drop; 1916 1917 udp_csum_pull_header(skb); 1918 1919 ipv4_pktinfo_prepare(sk, skb); 1920 return __udp_queue_rcv_skb(sk, skb); 1921 1922 csum_error: 1923 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1924 drop: 1925 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1926 atomic_inc(&sk->sk_drops); 1927 kfree_skb(skb); 1928 return -1; 1929 } 1930 1931 /* For TCP sockets, sk_rx_dst is protected by socket lock 1932 * For UDP, we use xchg() to guard against concurrent changes. 1933 */ 1934 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1935 { 1936 struct dst_entry *old; 1937 1938 if (dst_hold_safe(dst)) { 1939 old = xchg(&sk->sk_rx_dst, dst); 1940 dst_release(old); 1941 return old != dst; 1942 } 1943 return false; 1944 } 1945 EXPORT_SYMBOL(udp_sk_rx_dst_set); 1946 1947 /* 1948 * Multicasts and broadcasts go to each listener. 1949 * 1950 * Note: called only from the BH handler context. 1951 */ 1952 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1953 struct udphdr *uh, 1954 __be32 saddr, __be32 daddr, 1955 struct udp_table *udptable, 1956 int proto) 1957 { 1958 struct sock *sk, *first = NULL; 1959 unsigned short hnum = ntohs(uh->dest); 1960 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1961 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1962 unsigned int offset = offsetof(typeof(*sk), sk_node); 1963 int dif = skb->dev->ifindex; 1964 int sdif = inet_sdif(skb); 1965 struct hlist_node *node; 1966 struct sk_buff *nskb; 1967 1968 if (use_hash2) { 1969 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1970 udptable->mask; 1971 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask; 1972 start_lookup: 1973 hslot = &udptable->hash2[hash2]; 1974 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1975 } 1976 1977 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 1978 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 1979 uh->source, saddr, dif, sdif, hnum)) 1980 continue; 1981 1982 if (!first) { 1983 first = sk; 1984 continue; 1985 } 1986 nskb = skb_clone(skb, GFP_ATOMIC); 1987 1988 if (unlikely(!nskb)) { 1989 atomic_inc(&sk->sk_drops); 1990 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 1991 IS_UDPLITE(sk)); 1992 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 1993 IS_UDPLITE(sk)); 1994 continue; 1995 } 1996 if (udp_queue_rcv_skb(sk, nskb) > 0) 1997 consume_skb(nskb); 1998 } 1999 2000 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2001 if (use_hash2 && hash2 != hash2_any) { 2002 hash2 = hash2_any; 2003 goto start_lookup; 2004 } 2005 2006 if (first) { 2007 if (udp_queue_rcv_skb(first, skb) > 0) 2008 consume_skb(skb); 2009 } else { 2010 kfree_skb(skb); 2011 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2012 proto == IPPROTO_UDPLITE); 2013 } 2014 return 0; 2015 } 2016 2017 /* Initialize UDP checksum. If exited with zero value (success), 2018 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2019 * Otherwise, csum completion requires chacksumming packet body, 2020 * including udp header and folding it to skb->csum. 2021 */ 2022 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2023 int proto) 2024 { 2025 int err; 2026 2027 UDP_SKB_CB(skb)->partial_cov = 0; 2028 UDP_SKB_CB(skb)->cscov = skb->len; 2029 2030 if (proto == IPPROTO_UDPLITE) { 2031 err = udplite_checksum_init(skb, uh); 2032 if (err) 2033 return err; 2034 } 2035 2036 /* Note, we are only interested in != 0 or == 0, thus the 2037 * force to int. 2038 */ 2039 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2040 inet_compute_pseudo); 2041 } 2042 2043 /* 2044 * All we need to do is get the socket, and then do a checksum. 2045 */ 2046 2047 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2048 int proto) 2049 { 2050 struct sock *sk; 2051 struct udphdr *uh; 2052 unsigned short ulen; 2053 struct rtable *rt = skb_rtable(skb); 2054 __be32 saddr, daddr; 2055 struct net *net = dev_net(skb->dev); 2056 2057 /* 2058 * Validate the packet. 2059 */ 2060 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2061 goto drop; /* No space for header. */ 2062 2063 uh = udp_hdr(skb); 2064 ulen = ntohs(uh->len); 2065 saddr = ip_hdr(skb)->saddr; 2066 daddr = ip_hdr(skb)->daddr; 2067 2068 if (ulen > skb->len) 2069 goto short_packet; 2070 2071 if (proto == IPPROTO_UDP) { 2072 /* UDP validates ulen. */ 2073 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2074 goto short_packet; 2075 uh = udp_hdr(skb); 2076 } 2077 2078 if (udp4_csum_init(skb, uh, proto)) 2079 goto csum_error; 2080 2081 sk = skb_steal_sock(skb); 2082 if (sk) { 2083 struct dst_entry *dst = skb_dst(skb); 2084 int ret; 2085 2086 if (unlikely(sk->sk_rx_dst != dst)) 2087 udp_sk_rx_dst_set(sk, dst); 2088 2089 ret = udp_queue_rcv_skb(sk, skb); 2090 sock_put(sk); 2091 /* a return value > 0 means to resubmit the input, but 2092 * it wants the return to be -protocol, or 0 2093 */ 2094 if (ret > 0) 2095 return -ret; 2096 return 0; 2097 } 2098 2099 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2100 return __udp4_lib_mcast_deliver(net, skb, uh, 2101 saddr, daddr, udptable, proto); 2102 2103 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2104 if (sk) { 2105 int ret; 2106 2107 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2108 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 2109 inet_compute_pseudo); 2110 2111 ret = udp_queue_rcv_skb(sk, skb); 2112 2113 /* a return value > 0 means to resubmit the input, but 2114 * it wants the return to be -protocol, or 0 2115 */ 2116 if (ret > 0) 2117 return -ret; 2118 return 0; 2119 } 2120 2121 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2122 goto drop; 2123 nf_reset(skb); 2124 2125 /* No socket. Drop packet silently, if checksum is wrong */ 2126 if (udp_lib_checksum_complete(skb)) 2127 goto csum_error; 2128 2129 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2130 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2131 2132 /* 2133 * Hmm. We got an UDP packet to a port to which we 2134 * don't wanna listen. Ignore it. 2135 */ 2136 kfree_skb(skb); 2137 return 0; 2138 2139 short_packet: 2140 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2141 proto == IPPROTO_UDPLITE ? "Lite" : "", 2142 &saddr, ntohs(uh->source), 2143 ulen, skb->len, 2144 &daddr, ntohs(uh->dest)); 2145 goto drop; 2146 2147 csum_error: 2148 /* 2149 * RFC1122: OK. Discards the bad packet silently (as far as 2150 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2151 */ 2152 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2153 proto == IPPROTO_UDPLITE ? "Lite" : "", 2154 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2155 ulen); 2156 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2157 drop: 2158 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2159 kfree_skb(skb); 2160 return 0; 2161 } 2162 2163 /* We can only early demux multicast if there is a single matching socket. 2164 * If more than one socket found returns NULL 2165 */ 2166 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2167 __be16 loc_port, __be32 loc_addr, 2168 __be16 rmt_port, __be32 rmt_addr, 2169 int dif, int sdif) 2170 { 2171 struct sock *sk, *result; 2172 unsigned short hnum = ntohs(loc_port); 2173 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2174 struct udp_hslot *hslot = &udp_table.hash[slot]; 2175 2176 /* Do not bother scanning a too big list */ 2177 if (hslot->count > 10) 2178 return NULL; 2179 2180 result = NULL; 2181 sk_for_each_rcu(sk, &hslot->head) { 2182 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2183 rmt_port, rmt_addr, dif, sdif, hnum)) { 2184 if (result) 2185 return NULL; 2186 result = sk; 2187 } 2188 } 2189 2190 return result; 2191 } 2192 2193 /* For unicast we should only early demux connected sockets or we can 2194 * break forwarding setups. The chains here can be long so only check 2195 * if the first socket is an exact match and if not move on. 2196 */ 2197 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2198 __be16 loc_port, __be32 loc_addr, 2199 __be16 rmt_port, __be32 rmt_addr, 2200 int dif, int sdif) 2201 { 2202 unsigned short hnum = ntohs(loc_port); 2203 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 2204 unsigned int slot2 = hash2 & udp_table.mask; 2205 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2206 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2207 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2208 struct sock *sk; 2209 2210 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2211 if (INET_MATCH(sk, net, acookie, rmt_addr, 2212 loc_addr, ports, dif, sdif)) 2213 return sk; 2214 /* Only check first socket in chain */ 2215 break; 2216 } 2217 return NULL; 2218 } 2219 2220 int udp_v4_early_demux(struct sk_buff *skb) 2221 { 2222 struct net *net = dev_net(skb->dev); 2223 struct in_device *in_dev = NULL; 2224 const struct iphdr *iph; 2225 const struct udphdr *uh; 2226 struct sock *sk = NULL; 2227 struct dst_entry *dst; 2228 int dif = skb->dev->ifindex; 2229 int sdif = inet_sdif(skb); 2230 int ours; 2231 2232 /* validate the packet */ 2233 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2234 return 0; 2235 2236 iph = ip_hdr(skb); 2237 uh = udp_hdr(skb); 2238 2239 if (skb->pkt_type == PACKET_MULTICAST) { 2240 in_dev = __in_dev_get_rcu(skb->dev); 2241 2242 if (!in_dev) 2243 return 0; 2244 2245 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2246 iph->protocol); 2247 if (!ours) 2248 return 0; 2249 2250 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2251 uh->source, iph->saddr, 2252 dif, sdif); 2253 } else if (skb->pkt_type == PACKET_HOST) { 2254 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2255 uh->source, iph->saddr, dif, sdif); 2256 } 2257 2258 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2259 return 0; 2260 2261 skb->sk = sk; 2262 skb->destructor = sock_efree; 2263 dst = READ_ONCE(sk->sk_rx_dst); 2264 2265 if (dst) 2266 dst = dst_check(dst, 0); 2267 if (dst) { 2268 u32 itag = 0; 2269 2270 /* set noref for now. 2271 * any place which wants to hold dst has to call 2272 * dst_hold_safe() 2273 */ 2274 skb_dst_set_noref(skb, dst); 2275 2276 /* for unconnected multicast sockets we need to validate 2277 * the source on each packet 2278 */ 2279 if (!inet_sk(sk)->inet_daddr && in_dev) 2280 return ip_mc_validate_source(skb, iph->daddr, 2281 iph->saddr, iph->tos, 2282 skb->dev, in_dev, &itag); 2283 } 2284 return 0; 2285 } 2286 2287 int udp_rcv(struct sk_buff *skb) 2288 { 2289 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2290 } 2291 2292 void udp_destroy_sock(struct sock *sk) 2293 { 2294 struct udp_sock *up = udp_sk(sk); 2295 bool slow = lock_sock_fast(sk); 2296 udp_flush_pending_frames(sk); 2297 unlock_sock_fast(sk, slow); 2298 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2299 void (*encap_destroy)(struct sock *sk); 2300 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2301 if (encap_destroy) 2302 encap_destroy(sk); 2303 } 2304 } 2305 2306 /* 2307 * Socket option code for UDP 2308 */ 2309 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2310 char __user *optval, unsigned int optlen, 2311 int (*push_pending_frames)(struct sock *)) 2312 { 2313 struct udp_sock *up = udp_sk(sk); 2314 int val, valbool; 2315 int err = 0; 2316 int is_udplite = IS_UDPLITE(sk); 2317 2318 if (optlen < sizeof(int)) 2319 return -EINVAL; 2320 2321 if (get_user(val, (int __user *)optval)) 2322 return -EFAULT; 2323 2324 valbool = val ? 1 : 0; 2325 2326 switch (optname) { 2327 case UDP_CORK: 2328 if (val != 0) { 2329 up->corkflag = 1; 2330 } else { 2331 up->corkflag = 0; 2332 lock_sock(sk); 2333 push_pending_frames(sk); 2334 release_sock(sk); 2335 } 2336 break; 2337 2338 case UDP_ENCAP: 2339 switch (val) { 2340 case 0: 2341 case UDP_ENCAP_ESPINUDP: 2342 case UDP_ENCAP_ESPINUDP_NON_IKE: 2343 up->encap_rcv = xfrm4_udp_encap_rcv; 2344 /* FALLTHROUGH */ 2345 case UDP_ENCAP_L2TPINUDP: 2346 up->encap_type = val; 2347 udp_encap_enable(); 2348 break; 2349 default: 2350 err = -ENOPROTOOPT; 2351 break; 2352 } 2353 break; 2354 2355 case UDP_NO_CHECK6_TX: 2356 up->no_check6_tx = valbool; 2357 break; 2358 2359 case UDP_NO_CHECK6_RX: 2360 up->no_check6_rx = valbool; 2361 break; 2362 2363 /* 2364 * UDP-Lite's partial checksum coverage (RFC 3828). 2365 */ 2366 /* The sender sets actual checksum coverage length via this option. 2367 * The case coverage > packet length is handled by send module. */ 2368 case UDPLITE_SEND_CSCOV: 2369 if (!is_udplite) /* Disable the option on UDP sockets */ 2370 return -ENOPROTOOPT; 2371 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2372 val = 8; 2373 else if (val > USHRT_MAX) 2374 val = USHRT_MAX; 2375 up->pcslen = val; 2376 up->pcflag |= UDPLITE_SEND_CC; 2377 break; 2378 2379 /* The receiver specifies a minimum checksum coverage value. To make 2380 * sense, this should be set to at least 8 (as done below). If zero is 2381 * used, this again means full checksum coverage. */ 2382 case UDPLITE_RECV_CSCOV: 2383 if (!is_udplite) /* Disable the option on UDP sockets */ 2384 return -ENOPROTOOPT; 2385 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2386 val = 8; 2387 else if (val > USHRT_MAX) 2388 val = USHRT_MAX; 2389 up->pcrlen = val; 2390 up->pcflag |= UDPLITE_RECV_CC; 2391 break; 2392 2393 default: 2394 err = -ENOPROTOOPT; 2395 break; 2396 } 2397 2398 return err; 2399 } 2400 EXPORT_SYMBOL(udp_lib_setsockopt); 2401 2402 int udp_setsockopt(struct sock *sk, int level, int optname, 2403 char __user *optval, unsigned int optlen) 2404 { 2405 if (level == SOL_UDP || level == SOL_UDPLITE) 2406 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2407 udp_push_pending_frames); 2408 return ip_setsockopt(sk, level, optname, optval, optlen); 2409 } 2410 2411 #ifdef CONFIG_COMPAT 2412 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2413 char __user *optval, unsigned int optlen) 2414 { 2415 if (level == SOL_UDP || level == SOL_UDPLITE) 2416 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2417 udp_push_pending_frames); 2418 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2419 } 2420 #endif 2421 2422 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2423 char __user *optval, int __user *optlen) 2424 { 2425 struct udp_sock *up = udp_sk(sk); 2426 int val, len; 2427 2428 if (get_user(len, optlen)) 2429 return -EFAULT; 2430 2431 len = min_t(unsigned int, len, sizeof(int)); 2432 2433 if (len < 0) 2434 return -EINVAL; 2435 2436 switch (optname) { 2437 case UDP_CORK: 2438 val = up->corkflag; 2439 break; 2440 2441 case UDP_ENCAP: 2442 val = up->encap_type; 2443 break; 2444 2445 case UDP_NO_CHECK6_TX: 2446 val = up->no_check6_tx; 2447 break; 2448 2449 case UDP_NO_CHECK6_RX: 2450 val = up->no_check6_rx; 2451 break; 2452 2453 /* The following two cannot be changed on UDP sockets, the return is 2454 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2455 case UDPLITE_SEND_CSCOV: 2456 val = up->pcslen; 2457 break; 2458 2459 case UDPLITE_RECV_CSCOV: 2460 val = up->pcrlen; 2461 break; 2462 2463 default: 2464 return -ENOPROTOOPT; 2465 } 2466 2467 if (put_user(len, optlen)) 2468 return -EFAULT; 2469 if (copy_to_user(optval, &val, len)) 2470 return -EFAULT; 2471 return 0; 2472 } 2473 EXPORT_SYMBOL(udp_lib_getsockopt); 2474 2475 int udp_getsockopt(struct sock *sk, int level, int optname, 2476 char __user *optval, int __user *optlen) 2477 { 2478 if (level == SOL_UDP || level == SOL_UDPLITE) 2479 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2480 return ip_getsockopt(sk, level, optname, optval, optlen); 2481 } 2482 2483 #ifdef CONFIG_COMPAT 2484 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2485 char __user *optval, int __user *optlen) 2486 { 2487 if (level == SOL_UDP || level == SOL_UDPLITE) 2488 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2489 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2490 } 2491 #endif 2492 /** 2493 * udp_poll - wait for a UDP event. 2494 * @file - file struct 2495 * @sock - socket 2496 * @wait - poll table 2497 * 2498 * This is same as datagram poll, except for the special case of 2499 * blocking sockets. If application is using a blocking fd 2500 * and a packet with checksum error is in the queue; 2501 * then it could get return from select indicating data available 2502 * but then block when reading it. Add special case code 2503 * to work around these arguably broken applications. 2504 */ 2505 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2506 { 2507 unsigned int mask = datagram_poll(file, sock, wait); 2508 struct sock *sk = sock->sk; 2509 2510 if (!skb_queue_empty(&udp_sk(sk)->reader_queue)) 2511 mask |= POLLIN | POLLRDNORM; 2512 2513 sock_rps_record_flow(sk); 2514 2515 /* Check for false positives due to checksum errors */ 2516 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2517 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2518 mask &= ~(POLLIN | POLLRDNORM); 2519 2520 return mask; 2521 2522 } 2523 EXPORT_SYMBOL(udp_poll); 2524 2525 int udp_abort(struct sock *sk, int err) 2526 { 2527 lock_sock(sk); 2528 2529 sk->sk_err = err; 2530 sk->sk_error_report(sk); 2531 __udp_disconnect(sk, 0); 2532 2533 release_sock(sk); 2534 2535 return 0; 2536 } 2537 EXPORT_SYMBOL_GPL(udp_abort); 2538 2539 struct proto udp_prot = { 2540 .name = "UDP", 2541 .owner = THIS_MODULE, 2542 .close = udp_lib_close, 2543 .connect = ip4_datagram_connect, 2544 .disconnect = udp_disconnect, 2545 .ioctl = udp_ioctl, 2546 .init = udp_init_sock, 2547 .destroy = udp_destroy_sock, 2548 .setsockopt = udp_setsockopt, 2549 .getsockopt = udp_getsockopt, 2550 .sendmsg = udp_sendmsg, 2551 .recvmsg = udp_recvmsg, 2552 .sendpage = udp_sendpage, 2553 .release_cb = ip4_datagram_release_cb, 2554 .hash = udp_lib_hash, 2555 .unhash = udp_lib_unhash, 2556 .rehash = udp_v4_rehash, 2557 .get_port = udp_v4_get_port, 2558 .memory_allocated = &udp_memory_allocated, 2559 .sysctl_mem = sysctl_udp_mem, 2560 .sysctl_wmem = &sysctl_udp_wmem_min, 2561 .sysctl_rmem = &sysctl_udp_rmem_min, 2562 .obj_size = sizeof(struct udp_sock), 2563 .h.udp_table = &udp_table, 2564 #ifdef CONFIG_COMPAT 2565 .compat_setsockopt = compat_udp_setsockopt, 2566 .compat_getsockopt = compat_udp_getsockopt, 2567 #endif 2568 .diag_destroy = udp_abort, 2569 }; 2570 EXPORT_SYMBOL(udp_prot); 2571 2572 /* ------------------------------------------------------------------------ */ 2573 #ifdef CONFIG_PROC_FS 2574 2575 static struct sock *udp_get_first(struct seq_file *seq, int start) 2576 { 2577 struct sock *sk; 2578 struct udp_iter_state *state = seq->private; 2579 struct net *net = seq_file_net(seq); 2580 2581 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2582 ++state->bucket) { 2583 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2584 2585 if (hlist_empty(&hslot->head)) 2586 continue; 2587 2588 spin_lock_bh(&hslot->lock); 2589 sk_for_each(sk, &hslot->head) { 2590 if (!net_eq(sock_net(sk), net)) 2591 continue; 2592 if (sk->sk_family == state->family) 2593 goto found; 2594 } 2595 spin_unlock_bh(&hslot->lock); 2596 } 2597 sk = NULL; 2598 found: 2599 return sk; 2600 } 2601 2602 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2603 { 2604 struct udp_iter_state *state = seq->private; 2605 struct net *net = seq_file_net(seq); 2606 2607 do { 2608 sk = sk_next(sk); 2609 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2610 2611 if (!sk) { 2612 if (state->bucket <= state->udp_table->mask) 2613 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2614 return udp_get_first(seq, state->bucket + 1); 2615 } 2616 return sk; 2617 } 2618 2619 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2620 { 2621 struct sock *sk = udp_get_first(seq, 0); 2622 2623 if (sk) 2624 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2625 --pos; 2626 return pos ? NULL : sk; 2627 } 2628 2629 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2630 { 2631 struct udp_iter_state *state = seq->private; 2632 state->bucket = MAX_UDP_PORTS; 2633 2634 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2635 } 2636 2637 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2638 { 2639 struct sock *sk; 2640 2641 if (v == SEQ_START_TOKEN) 2642 sk = udp_get_idx(seq, 0); 2643 else 2644 sk = udp_get_next(seq, v); 2645 2646 ++*pos; 2647 return sk; 2648 } 2649 2650 static void udp_seq_stop(struct seq_file *seq, void *v) 2651 { 2652 struct udp_iter_state *state = seq->private; 2653 2654 if (state->bucket <= state->udp_table->mask) 2655 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2656 } 2657 2658 int udp_seq_open(struct inode *inode, struct file *file) 2659 { 2660 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2661 struct udp_iter_state *s; 2662 int err; 2663 2664 err = seq_open_net(inode, file, &afinfo->seq_ops, 2665 sizeof(struct udp_iter_state)); 2666 if (err < 0) 2667 return err; 2668 2669 s = ((struct seq_file *)file->private_data)->private; 2670 s->family = afinfo->family; 2671 s->udp_table = afinfo->udp_table; 2672 return err; 2673 } 2674 EXPORT_SYMBOL(udp_seq_open); 2675 2676 /* ------------------------------------------------------------------------ */ 2677 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2678 { 2679 struct proc_dir_entry *p; 2680 int rc = 0; 2681 2682 afinfo->seq_ops.start = udp_seq_start; 2683 afinfo->seq_ops.next = udp_seq_next; 2684 afinfo->seq_ops.stop = udp_seq_stop; 2685 2686 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2687 afinfo->seq_fops, afinfo); 2688 if (!p) 2689 rc = -ENOMEM; 2690 return rc; 2691 } 2692 EXPORT_SYMBOL(udp_proc_register); 2693 2694 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2695 { 2696 remove_proc_entry(afinfo->name, net->proc_net); 2697 } 2698 EXPORT_SYMBOL(udp_proc_unregister); 2699 2700 /* ------------------------------------------------------------------------ */ 2701 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2702 int bucket) 2703 { 2704 struct inet_sock *inet = inet_sk(sp); 2705 __be32 dest = inet->inet_daddr; 2706 __be32 src = inet->inet_rcv_saddr; 2707 __u16 destp = ntohs(inet->inet_dport); 2708 __u16 srcp = ntohs(inet->inet_sport); 2709 2710 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2711 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2712 bucket, src, srcp, dest, destp, sp->sk_state, 2713 sk_wmem_alloc_get(sp), 2714 sk_rmem_alloc_get(sp), 2715 0, 0L, 0, 2716 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2717 0, sock_i_ino(sp), 2718 refcount_read(&sp->sk_refcnt), sp, 2719 atomic_read(&sp->sk_drops)); 2720 } 2721 2722 int udp4_seq_show(struct seq_file *seq, void *v) 2723 { 2724 seq_setwidth(seq, 127); 2725 if (v == SEQ_START_TOKEN) 2726 seq_puts(seq, " sl local_address rem_address st tx_queue " 2727 "rx_queue tr tm->when retrnsmt uid timeout " 2728 "inode ref pointer drops"); 2729 else { 2730 struct udp_iter_state *state = seq->private; 2731 2732 udp4_format_sock(v, seq, state->bucket); 2733 } 2734 seq_pad(seq, '\n'); 2735 return 0; 2736 } 2737 2738 static const struct file_operations udp_afinfo_seq_fops = { 2739 .owner = THIS_MODULE, 2740 .open = udp_seq_open, 2741 .read = seq_read, 2742 .llseek = seq_lseek, 2743 .release = seq_release_net 2744 }; 2745 2746 /* ------------------------------------------------------------------------ */ 2747 static struct udp_seq_afinfo udp4_seq_afinfo = { 2748 .name = "udp", 2749 .family = AF_INET, 2750 .udp_table = &udp_table, 2751 .seq_fops = &udp_afinfo_seq_fops, 2752 .seq_ops = { 2753 .show = udp4_seq_show, 2754 }, 2755 }; 2756 2757 static int __net_init udp4_proc_init_net(struct net *net) 2758 { 2759 return udp_proc_register(net, &udp4_seq_afinfo); 2760 } 2761 2762 static void __net_exit udp4_proc_exit_net(struct net *net) 2763 { 2764 udp_proc_unregister(net, &udp4_seq_afinfo); 2765 } 2766 2767 static struct pernet_operations udp4_net_ops = { 2768 .init = udp4_proc_init_net, 2769 .exit = udp4_proc_exit_net, 2770 }; 2771 2772 int __init udp4_proc_init(void) 2773 { 2774 return register_pernet_subsys(&udp4_net_ops); 2775 } 2776 2777 void udp4_proc_exit(void) 2778 { 2779 unregister_pernet_subsys(&udp4_net_ops); 2780 } 2781 #endif /* CONFIG_PROC_FS */ 2782 2783 static __initdata unsigned long uhash_entries; 2784 static int __init set_uhash_entries(char *str) 2785 { 2786 ssize_t ret; 2787 2788 if (!str) 2789 return 0; 2790 2791 ret = kstrtoul(str, 0, &uhash_entries); 2792 if (ret) 2793 return 0; 2794 2795 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2796 uhash_entries = UDP_HTABLE_SIZE_MIN; 2797 return 1; 2798 } 2799 __setup("uhash_entries=", set_uhash_entries); 2800 2801 void __init udp_table_init(struct udp_table *table, const char *name) 2802 { 2803 unsigned int i; 2804 2805 table->hash = alloc_large_system_hash(name, 2806 2 * sizeof(struct udp_hslot), 2807 uhash_entries, 2808 21, /* one slot per 2 MB */ 2809 0, 2810 &table->log, 2811 &table->mask, 2812 UDP_HTABLE_SIZE_MIN, 2813 64 * 1024); 2814 2815 table->hash2 = table->hash + (table->mask + 1); 2816 for (i = 0; i <= table->mask; i++) { 2817 INIT_HLIST_HEAD(&table->hash[i].head); 2818 table->hash[i].count = 0; 2819 spin_lock_init(&table->hash[i].lock); 2820 } 2821 for (i = 0; i <= table->mask; i++) { 2822 INIT_HLIST_HEAD(&table->hash2[i].head); 2823 table->hash2[i].count = 0; 2824 spin_lock_init(&table->hash2[i].lock); 2825 } 2826 } 2827 2828 u32 udp_flow_hashrnd(void) 2829 { 2830 static u32 hashrnd __read_mostly; 2831 2832 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2833 2834 return hashrnd; 2835 } 2836 EXPORT_SYMBOL(udp_flow_hashrnd); 2837 2838 void __init udp_init(void) 2839 { 2840 unsigned long limit; 2841 unsigned int i; 2842 2843 udp_table_init(&udp_table, "UDP"); 2844 limit = nr_free_buffer_pages() / 8; 2845 limit = max(limit, 128UL); 2846 sysctl_udp_mem[0] = limit / 4 * 3; 2847 sysctl_udp_mem[1] = limit; 2848 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2849 2850 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2851 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2852 2853 /* 16 spinlocks per cpu */ 2854 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 2855 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 2856 GFP_KERNEL); 2857 if (!udp_busylocks) 2858 panic("UDP: failed to alloc udp_busylocks\n"); 2859 for (i = 0; i < (1U << udp_busylocks_log); i++) 2860 spin_lock_init(udp_busylocks + i); 2861 } 2862