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 atomic_long_t udp_memory_allocated; 126 EXPORT_SYMBOL(udp_memory_allocated); 127 128 #define MAX_UDP_PORTS 65536 129 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 130 131 /* IPCB reference means this can not be used from early demux */ 132 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb) 133 { 134 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 135 if (!net->ipv4.sysctl_udp_l3mdev_accept && 136 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 137 return true; 138 #endif 139 return false; 140 } 141 142 static int udp_lib_lport_inuse(struct net *net, __u16 num, 143 const struct udp_hslot *hslot, 144 unsigned long *bitmap, 145 struct sock *sk, unsigned int log) 146 { 147 struct sock *sk2; 148 kuid_t uid = sock_i_uid(sk); 149 150 sk_for_each(sk2, &hslot->head) { 151 if (net_eq(sock_net(sk2), net) && 152 sk2 != sk && 153 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 154 (!sk2->sk_reuse || !sk->sk_reuse) && 155 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 156 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 157 inet_rcv_saddr_equal(sk, sk2, true)) { 158 if (sk2->sk_reuseport && sk->sk_reuseport && 159 !rcu_access_pointer(sk->sk_reuseport_cb) && 160 uid_eq(uid, sock_i_uid(sk2))) { 161 if (!bitmap) 162 return 0; 163 } else { 164 if (!bitmap) 165 return 1; 166 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 167 bitmap); 168 } 169 } 170 } 171 return 0; 172 } 173 174 /* 175 * Note: we still hold spinlock of primary hash chain, so no other writer 176 * can insert/delete a socket with local_port == num 177 */ 178 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 179 struct udp_hslot *hslot2, 180 struct sock *sk) 181 { 182 struct sock *sk2; 183 kuid_t uid = sock_i_uid(sk); 184 int res = 0; 185 186 spin_lock(&hslot2->lock); 187 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 188 if (net_eq(sock_net(sk2), net) && 189 sk2 != sk && 190 (udp_sk(sk2)->udp_port_hash == num) && 191 (!sk2->sk_reuse || !sk->sk_reuse) && 192 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 193 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 194 inet_rcv_saddr_equal(sk, sk2, true)) { 195 if (sk2->sk_reuseport && sk->sk_reuseport && 196 !rcu_access_pointer(sk->sk_reuseport_cb) && 197 uid_eq(uid, sock_i_uid(sk2))) { 198 res = 0; 199 } else { 200 res = 1; 201 } 202 break; 203 } 204 } 205 spin_unlock(&hslot2->lock); 206 return res; 207 } 208 209 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 210 { 211 struct net *net = sock_net(sk); 212 kuid_t uid = sock_i_uid(sk); 213 struct sock *sk2; 214 215 sk_for_each(sk2, &hslot->head) { 216 if (net_eq(sock_net(sk2), net) && 217 sk2 != sk && 218 sk2->sk_family == sk->sk_family && 219 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 220 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 221 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 222 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 223 inet_rcv_saddr_equal(sk, sk2, false)) { 224 return reuseport_add_sock(sk, sk2); 225 } 226 } 227 228 return reuseport_alloc(sk); 229 } 230 231 /** 232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 233 * 234 * @sk: socket struct in question 235 * @snum: port number to look up 236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 237 * with NULL address 238 */ 239 int udp_lib_get_port(struct sock *sk, unsigned short snum, 240 unsigned int hash2_nulladdr) 241 { 242 struct udp_hslot *hslot, *hslot2; 243 struct udp_table *udptable = sk->sk_prot->h.udp_table; 244 int error = 1; 245 struct net *net = sock_net(sk); 246 247 if (!snum) { 248 int low, high, remaining; 249 unsigned int rand; 250 unsigned short first, last; 251 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 252 253 inet_get_local_port_range(net, &low, &high); 254 remaining = (high - low) + 1; 255 256 rand = prandom_u32(); 257 first = reciprocal_scale(rand, remaining) + low; 258 /* 259 * force rand to be an odd multiple of UDP_HTABLE_SIZE 260 */ 261 rand = (rand | 1) * (udptable->mask + 1); 262 last = first + udptable->mask + 1; 263 do { 264 hslot = udp_hashslot(udptable, net, first); 265 bitmap_zero(bitmap, PORTS_PER_CHAIN); 266 spin_lock_bh(&hslot->lock); 267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 268 udptable->log); 269 270 snum = first; 271 /* 272 * Iterate on all possible values of snum for this hash. 273 * Using steps of an odd multiple of UDP_HTABLE_SIZE 274 * give us randomization and full range coverage. 275 */ 276 do { 277 if (low <= snum && snum <= high && 278 !test_bit(snum >> udptable->log, bitmap) && 279 !inet_is_local_reserved_port(net, snum)) 280 goto found; 281 snum += rand; 282 } while (snum != first); 283 spin_unlock_bh(&hslot->lock); 284 cond_resched(); 285 } while (++first != last); 286 goto fail; 287 } else { 288 hslot = udp_hashslot(udptable, net, snum); 289 spin_lock_bh(&hslot->lock); 290 if (hslot->count > 10) { 291 int exist; 292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 293 294 slot2 &= udptable->mask; 295 hash2_nulladdr &= udptable->mask; 296 297 hslot2 = udp_hashslot2(udptable, slot2); 298 if (hslot->count < hslot2->count) 299 goto scan_primary_hash; 300 301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 302 if (!exist && (hash2_nulladdr != slot2)) { 303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 304 exist = udp_lib_lport_inuse2(net, snum, hslot2, 305 sk); 306 } 307 if (exist) 308 goto fail_unlock; 309 else 310 goto found; 311 } 312 scan_primary_hash: 313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 314 goto fail_unlock; 315 } 316 found: 317 inet_sk(sk)->inet_num = snum; 318 udp_sk(sk)->udp_port_hash = snum; 319 udp_sk(sk)->udp_portaddr_hash ^= snum; 320 if (sk_unhashed(sk)) { 321 if (sk->sk_reuseport && 322 udp_reuseport_add_sock(sk, hslot)) { 323 inet_sk(sk)->inet_num = 0; 324 udp_sk(sk)->udp_port_hash = 0; 325 udp_sk(sk)->udp_portaddr_hash ^= snum; 326 goto fail_unlock; 327 } 328 329 sk_add_node_rcu(sk, &hslot->head); 330 hslot->count++; 331 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 332 333 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 334 spin_lock(&hslot2->lock); 335 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 336 sk->sk_family == AF_INET6) 337 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 338 &hslot2->head); 339 else 340 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 341 &hslot2->head); 342 hslot2->count++; 343 spin_unlock(&hslot2->lock); 344 } 345 sock_set_flag(sk, SOCK_RCU_FREE); 346 error = 0; 347 fail_unlock: 348 spin_unlock_bh(&hslot->lock); 349 fail: 350 return error; 351 } 352 EXPORT_SYMBOL(udp_lib_get_port); 353 354 int udp_v4_get_port(struct sock *sk, unsigned short snum) 355 { 356 unsigned int hash2_nulladdr = 357 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 358 unsigned int hash2_partial = 359 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 360 361 /* precompute partial secondary hash */ 362 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 363 return udp_lib_get_port(sk, snum, hash2_nulladdr); 364 } 365 366 static int compute_score(struct sock *sk, struct net *net, 367 __be32 saddr, __be16 sport, 368 __be32 daddr, unsigned short hnum, 369 int dif, int sdif, bool exact_dif) 370 { 371 int score; 372 struct inet_sock *inet; 373 374 if (!net_eq(sock_net(sk), net) || 375 udp_sk(sk)->udp_port_hash != hnum || 376 ipv6_only_sock(sk)) 377 return -1; 378 379 score = (sk->sk_family == PF_INET) ? 2 : 1; 380 inet = inet_sk(sk); 381 382 if (inet->inet_rcv_saddr) { 383 if (inet->inet_rcv_saddr != daddr) 384 return -1; 385 score += 4; 386 } 387 388 if (inet->inet_daddr) { 389 if (inet->inet_daddr != saddr) 390 return -1; 391 score += 4; 392 } 393 394 if (inet->inet_dport) { 395 if (inet->inet_dport != sport) 396 return -1; 397 score += 4; 398 } 399 400 if (sk->sk_bound_dev_if || exact_dif) { 401 bool dev_match = (sk->sk_bound_dev_if == dif || 402 sk->sk_bound_dev_if == sdif); 403 404 if (exact_dif && !dev_match) 405 return -1; 406 if (sk->sk_bound_dev_if && dev_match) 407 score += 4; 408 } 409 410 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 411 score++; 412 return score; 413 } 414 415 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 416 const __u16 lport, const __be32 faddr, 417 const __be16 fport) 418 { 419 static u32 udp_ehash_secret __read_mostly; 420 421 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 422 423 return __inet_ehashfn(laddr, lport, faddr, fport, 424 udp_ehash_secret + net_hash_mix(net)); 425 } 426 427 /* called with rcu_read_lock() */ 428 static struct sock *udp4_lib_lookup2(struct net *net, 429 __be32 saddr, __be16 sport, 430 __be32 daddr, unsigned int hnum, 431 int dif, int sdif, bool exact_dif, 432 struct udp_hslot *hslot2, 433 struct sk_buff *skb) 434 { 435 struct sock *sk, *result; 436 int score, badness; 437 u32 hash = 0; 438 439 result = NULL; 440 badness = 0; 441 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 442 score = compute_score(sk, net, saddr, sport, 443 daddr, hnum, dif, sdif, exact_dif); 444 if (score > badness) { 445 if (sk->sk_reuseport) { 446 hash = udp_ehashfn(net, daddr, hnum, 447 saddr, sport); 448 result = reuseport_select_sock(sk, hash, skb, 449 sizeof(struct udphdr)); 450 if (result) 451 return result; 452 } 453 badness = score; 454 result = sk; 455 } 456 } 457 return result; 458 } 459 460 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 461 * harder than this. -DaveM 462 */ 463 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 464 __be16 sport, __be32 daddr, __be16 dport, int dif, 465 int sdif, struct udp_table *udptable, struct sk_buff *skb) 466 { 467 struct sock *sk, *result; 468 unsigned short hnum = ntohs(dport); 469 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 470 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 471 bool exact_dif = udp_lib_exact_dif_match(net, skb); 472 int score, badness; 473 u32 hash = 0; 474 475 if (hslot->count > 10) { 476 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 477 slot2 = hash2 & udptable->mask; 478 hslot2 = &udptable->hash2[slot2]; 479 if (hslot->count < hslot2->count) 480 goto begin; 481 482 result = udp4_lib_lookup2(net, saddr, sport, 483 daddr, hnum, dif, sdif, 484 exact_dif, hslot2, skb); 485 if (!result) { 486 unsigned int old_slot2 = slot2; 487 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 488 slot2 = hash2 & udptable->mask; 489 /* avoid searching the same slot again. */ 490 if (unlikely(slot2 == old_slot2)) 491 return result; 492 493 hslot2 = &udptable->hash2[slot2]; 494 if (hslot->count < hslot2->count) 495 goto begin; 496 497 result = udp4_lib_lookup2(net, saddr, sport, 498 daddr, hnum, dif, sdif, 499 exact_dif, hslot2, skb); 500 } 501 return result; 502 } 503 begin: 504 result = NULL; 505 badness = 0; 506 sk_for_each_rcu(sk, &hslot->head) { 507 score = compute_score(sk, net, saddr, sport, 508 daddr, hnum, dif, sdif, exact_dif); 509 if (score > badness) { 510 if (sk->sk_reuseport) { 511 hash = udp_ehashfn(net, daddr, hnum, 512 saddr, sport); 513 result = reuseport_select_sock(sk, hash, skb, 514 sizeof(struct udphdr)); 515 if (result) 516 return result; 517 } 518 result = sk; 519 badness = score; 520 } 521 } 522 return result; 523 } 524 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 525 526 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 527 __be16 sport, __be16 dport, 528 struct udp_table *udptable) 529 { 530 const struct iphdr *iph = ip_hdr(skb); 531 532 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 533 iph->daddr, dport, inet_iif(skb), 534 inet_sdif(skb), udptable, skb); 535 } 536 537 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, 538 __be16 sport, __be16 dport) 539 { 540 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); 541 } 542 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); 543 544 /* Must be called under rcu_read_lock(). 545 * Does increment socket refcount. 546 */ 547 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \ 548 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \ 549 IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 550 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 551 __be32 daddr, __be16 dport, int dif) 552 { 553 struct sock *sk; 554 555 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 556 dif, 0, &udp_table, NULL); 557 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 558 sk = NULL; 559 return sk; 560 } 561 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 562 #endif 563 564 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 565 __be16 loc_port, __be32 loc_addr, 566 __be16 rmt_port, __be32 rmt_addr, 567 int dif, int sdif, unsigned short hnum) 568 { 569 struct inet_sock *inet = inet_sk(sk); 570 571 if (!net_eq(sock_net(sk), net) || 572 udp_sk(sk)->udp_port_hash != hnum || 573 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 574 (inet->inet_dport != rmt_port && inet->inet_dport) || 575 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 576 ipv6_only_sock(sk) || 577 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif && 578 sk->sk_bound_dev_if != sdif)) 579 return false; 580 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 581 return false; 582 return true; 583 } 584 585 /* 586 * This routine is called by the ICMP module when it gets some 587 * sort of error condition. If err < 0 then the socket should 588 * be closed and the error returned to the user. If err > 0 589 * it's just the icmp type << 8 | icmp code. 590 * Header points to the ip header of the error packet. We move 591 * on past this. Then (as it used to claim before adjustment) 592 * header points to the first 8 bytes of the udp header. We need 593 * to find the appropriate port. 594 */ 595 596 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 597 { 598 struct inet_sock *inet; 599 const struct iphdr *iph = (const struct iphdr *)skb->data; 600 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 601 const int type = icmp_hdr(skb)->type; 602 const int code = icmp_hdr(skb)->code; 603 struct sock *sk; 604 int harderr; 605 int err; 606 struct net *net = dev_net(skb->dev); 607 608 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 609 iph->saddr, uh->source, skb->dev->ifindex, 0, 610 udptable, NULL); 611 if (!sk) { 612 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 613 return; /* No socket for error */ 614 } 615 616 err = 0; 617 harderr = 0; 618 inet = inet_sk(sk); 619 620 switch (type) { 621 default: 622 case ICMP_TIME_EXCEEDED: 623 err = EHOSTUNREACH; 624 break; 625 case ICMP_SOURCE_QUENCH: 626 goto out; 627 case ICMP_PARAMETERPROB: 628 err = EPROTO; 629 harderr = 1; 630 break; 631 case ICMP_DEST_UNREACH: 632 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 633 ipv4_sk_update_pmtu(skb, sk, info); 634 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 635 err = EMSGSIZE; 636 harderr = 1; 637 break; 638 } 639 goto out; 640 } 641 err = EHOSTUNREACH; 642 if (code <= NR_ICMP_UNREACH) { 643 harderr = icmp_err_convert[code].fatal; 644 err = icmp_err_convert[code].errno; 645 } 646 break; 647 case ICMP_REDIRECT: 648 ipv4_sk_redirect(skb, sk); 649 goto out; 650 } 651 652 /* 653 * RFC1122: OK. Passes ICMP errors back to application, as per 654 * 4.1.3.3. 655 */ 656 if (!inet->recverr) { 657 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 658 goto out; 659 } else 660 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 661 662 sk->sk_err = err; 663 sk->sk_error_report(sk); 664 out: 665 return; 666 } 667 668 void udp_err(struct sk_buff *skb, u32 info) 669 { 670 __udp4_lib_err(skb, info, &udp_table); 671 } 672 673 /* 674 * Throw away all pending data and cancel the corking. Socket is locked. 675 */ 676 void udp_flush_pending_frames(struct sock *sk) 677 { 678 struct udp_sock *up = udp_sk(sk); 679 680 if (up->pending) { 681 up->len = 0; 682 up->pending = 0; 683 ip_flush_pending_frames(sk); 684 } 685 } 686 EXPORT_SYMBOL(udp_flush_pending_frames); 687 688 /** 689 * udp4_hwcsum - handle outgoing HW checksumming 690 * @skb: sk_buff containing the filled-in UDP header 691 * (checksum field must be zeroed out) 692 * @src: source IP address 693 * @dst: destination IP address 694 */ 695 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 696 { 697 struct udphdr *uh = udp_hdr(skb); 698 int offset = skb_transport_offset(skb); 699 int len = skb->len - offset; 700 int hlen = len; 701 __wsum csum = 0; 702 703 if (!skb_has_frag_list(skb)) { 704 /* 705 * Only one fragment on the socket. 706 */ 707 skb->csum_start = skb_transport_header(skb) - skb->head; 708 skb->csum_offset = offsetof(struct udphdr, check); 709 uh->check = ~csum_tcpudp_magic(src, dst, len, 710 IPPROTO_UDP, 0); 711 } else { 712 struct sk_buff *frags; 713 714 /* 715 * HW-checksum won't work as there are two or more 716 * fragments on the socket so that all csums of sk_buffs 717 * should be together 718 */ 719 skb_walk_frags(skb, frags) { 720 csum = csum_add(csum, frags->csum); 721 hlen -= frags->len; 722 } 723 724 csum = skb_checksum(skb, offset, hlen, csum); 725 skb->ip_summed = CHECKSUM_NONE; 726 727 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 728 if (uh->check == 0) 729 uh->check = CSUM_MANGLED_0; 730 } 731 } 732 EXPORT_SYMBOL_GPL(udp4_hwcsum); 733 734 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 735 * for the simple case like when setting the checksum for a UDP tunnel. 736 */ 737 void udp_set_csum(bool nocheck, struct sk_buff *skb, 738 __be32 saddr, __be32 daddr, int len) 739 { 740 struct udphdr *uh = udp_hdr(skb); 741 742 if (nocheck) { 743 uh->check = 0; 744 } else if (skb_is_gso(skb)) { 745 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 746 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 747 uh->check = 0; 748 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 749 if (uh->check == 0) 750 uh->check = CSUM_MANGLED_0; 751 } else { 752 skb->ip_summed = CHECKSUM_PARTIAL; 753 skb->csum_start = skb_transport_header(skb) - skb->head; 754 skb->csum_offset = offsetof(struct udphdr, check); 755 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 756 } 757 } 758 EXPORT_SYMBOL(udp_set_csum); 759 760 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 761 struct inet_cork *cork) 762 { 763 struct sock *sk = skb->sk; 764 struct inet_sock *inet = inet_sk(sk); 765 struct udphdr *uh; 766 int err = 0; 767 int is_udplite = IS_UDPLITE(sk); 768 int offset = skb_transport_offset(skb); 769 int len = skb->len - offset; 770 __wsum csum = 0; 771 772 /* 773 * Create a UDP header 774 */ 775 uh = udp_hdr(skb); 776 uh->source = inet->inet_sport; 777 uh->dest = fl4->fl4_dport; 778 uh->len = htons(len); 779 uh->check = 0; 780 781 if (cork->gso_size) { 782 const int hlen = skb_network_header_len(skb) + 783 sizeof(struct udphdr); 784 785 if (hlen + cork->gso_size > cork->fragsize) 786 return -EINVAL; 787 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) 788 return -EINVAL; 789 if (sk->sk_no_check_tx) 790 return -EINVAL; 791 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite) 792 return -EIO; 793 794 skb_shinfo(skb)->gso_size = cork->gso_size; 795 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 796 goto csum_partial; 797 } 798 799 if (is_udplite) /* UDP-Lite */ 800 csum = udplite_csum(skb); 801 802 else if (sk->sk_no_check_tx) { /* UDP csum off */ 803 804 skb->ip_summed = CHECKSUM_NONE; 805 goto send; 806 807 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 808 csum_partial: 809 810 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 811 goto send; 812 813 } else 814 csum = udp_csum(skb); 815 816 /* add protocol-dependent pseudo-header */ 817 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 818 sk->sk_protocol, csum); 819 if (uh->check == 0) 820 uh->check = CSUM_MANGLED_0; 821 822 send: 823 err = ip_send_skb(sock_net(sk), skb); 824 if (err) { 825 if (err == -ENOBUFS && !inet->recverr) { 826 UDP_INC_STATS(sock_net(sk), 827 UDP_MIB_SNDBUFERRORS, is_udplite); 828 err = 0; 829 } 830 } else 831 UDP_INC_STATS(sock_net(sk), 832 UDP_MIB_OUTDATAGRAMS, is_udplite); 833 return err; 834 } 835 836 /* 837 * Push out all pending data as one UDP datagram. Socket is locked. 838 */ 839 int udp_push_pending_frames(struct sock *sk) 840 { 841 struct udp_sock *up = udp_sk(sk); 842 struct inet_sock *inet = inet_sk(sk); 843 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 844 struct sk_buff *skb; 845 int err = 0; 846 847 skb = ip_finish_skb(sk, fl4); 848 if (!skb) 849 goto out; 850 851 err = udp_send_skb(skb, fl4, &inet->cork.base); 852 853 out: 854 up->len = 0; 855 up->pending = 0; 856 return err; 857 } 858 EXPORT_SYMBOL(udp_push_pending_frames); 859 860 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 861 { 862 switch (cmsg->cmsg_type) { 863 case UDP_SEGMENT: 864 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 865 return -EINVAL; 866 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 867 return 0; 868 default: 869 return -EINVAL; 870 } 871 } 872 873 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 874 { 875 struct cmsghdr *cmsg; 876 bool need_ip = false; 877 int err; 878 879 for_each_cmsghdr(cmsg, msg) { 880 if (!CMSG_OK(msg, cmsg)) 881 return -EINVAL; 882 883 if (cmsg->cmsg_level != SOL_UDP) { 884 need_ip = true; 885 continue; 886 } 887 888 err = __udp_cmsg_send(cmsg, gso_size); 889 if (err) 890 return err; 891 } 892 893 return need_ip; 894 } 895 EXPORT_SYMBOL_GPL(udp_cmsg_send); 896 897 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 898 { 899 struct inet_sock *inet = inet_sk(sk); 900 struct udp_sock *up = udp_sk(sk); 901 struct flowi4 fl4_stack; 902 struct flowi4 *fl4; 903 int ulen = len; 904 struct ipcm_cookie ipc; 905 struct rtable *rt = NULL; 906 int free = 0; 907 int connected = 0; 908 __be32 daddr, faddr, saddr; 909 __be16 dport; 910 u8 tos; 911 int err, is_udplite = IS_UDPLITE(sk); 912 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 913 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 914 struct sk_buff *skb; 915 struct ip_options_data opt_copy; 916 917 if (len > 0xFFFF) 918 return -EMSGSIZE; 919 920 /* 921 * Check the flags. 922 */ 923 924 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 925 return -EOPNOTSUPP; 926 927 ipc.opt = NULL; 928 ipc.tx_flags = 0; 929 ipc.ttl = 0; 930 ipc.tos = -1; 931 932 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 933 934 fl4 = &inet->cork.fl.u.ip4; 935 if (up->pending) { 936 /* 937 * There are pending frames. 938 * The socket lock must be held while it's corked. 939 */ 940 lock_sock(sk); 941 if (likely(up->pending)) { 942 if (unlikely(up->pending != AF_INET)) { 943 release_sock(sk); 944 return -EINVAL; 945 } 946 goto do_append_data; 947 } 948 release_sock(sk); 949 } 950 ulen += sizeof(struct udphdr); 951 952 /* 953 * Get and verify the address. 954 */ 955 if (msg->msg_name) { 956 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 957 if (msg->msg_namelen < sizeof(*usin)) 958 return -EINVAL; 959 if (usin->sin_family != AF_INET) { 960 if (usin->sin_family != AF_UNSPEC) 961 return -EAFNOSUPPORT; 962 } 963 964 daddr = usin->sin_addr.s_addr; 965 dport = usin->sin_port; 966 if (dport == 0) 967 return -EINVAL; 968 } else { 969 if (sk->sk_state != TCP_ESTABLISHED) 970 return -EDESTADDRREQ; 971 daddr = inet->inet_daddr; 972 dport = inet->inet_dport; 973 /* Open fast path for connected socket. 974 Route will not be used, if at least one option is set. 975 */ 976 connected = 1; 977 } 978 979 ipc.sockc.tsflags = sk->sk_tsflags; 980 ipc.addr = inet->inet_saddr; 981 ipc.oif = sk->sk_bound_dev_if; 982 ipc.gso_size = up->gso_size; 983 984 if (msg->msg_controllen) { 985 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 986 if (err > 0) 987 err = ip_cmsg_send(sk, msg, &ipc, 988 sk->sk_family == AF_INET6); 989 if (unlikely(err < 0)) { 990 kfree(ipc.opt); 991 return err; 992 } 993 if (ipc.opt) 994 free = 1; 995 connected = 0; 996 } 997 if (!ipc.opt) { 998 struct ip_options_rcu *inet_opt; 999 1000 rcu_read_lock(); 1001 inet_opt = rcu_dereference(inet->inet_opt); 1002 if (inet_opt) { 1003 memcpy(&opt_copy, inet_opt, 1004 sizeof(*inet_opt) + inet_opt->opt.optlen); 1005 ipc.opt = &opt_copy.opt; 1006 } 1007 rcu_read_unlock(); 1008 } 1009 1010 saddr = ipc.addr; 1011 ipc.addr = faddr = daddr; 1012 1013 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags); 1014 1015 if (ipc.opt && ipc.opt->opt.srr) { 1016 if (!daddr) 1017 return -EINVAL; 1018 faddr = ipc.opt->opt.faddr; 1019 connected = 0; 1020 } 1021 tos = get_rttos(&ipc, inet); 1022 if (sock_flag(sk, SOCK_LOCALROUTE) || 1023 (msg->msg_flags & MSG_DONTROUTE) || 1024 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1025 tos |= RTO_ONLINK; 1026 connected = 0; 1027 } 1028 1029 if (ipv4_is_multicast(daddr)) { 1030 if (!ipc.oif) 1031 ipc.oif = inet->mc_index; 1032 if (!saddr) 1033 saddr = inet->mc_addr; 1034 connected = 0; 1035 } else if (!ipc.oif) { 1036 ipc.oif = inet->uc_index; 1037 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1038 /* oif is set, packet is to local broadcast and 1039 * and uc_index is set. oif is most likely set 1040 * by sk_bound_dev_if. If uc_index != oif check if the 1041 * oif is an L3 master and uc_index is an L3 slave. 1042 * If so, we want to allow the send using the uc_index. 1043 */ 1044 if (ipc.oif != inet->uc_index && 1045 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1046 inet->uc_index)) { 1047 ipc.oif = inet->uc_index; 1048 } 1049 } 1050 1051 if (connected) 1052 rt = (struct rtable *)sk_dst_check(sk, 0); 1053 1054 if (!rt) { 1055 struct net *net = sock_net(sk); 1056 __u8 flow_flags = inet_sk_flowi_flags(sk); 1057 1058 fl4 = &fl4_stack; 1059 1060 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1061 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1062 flow_flags, 1063 faddr, saddr, dport, inet->inet_sport, 1064 sk->sk_uid); 1065 1066 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1067 rt = ip_route_output_flow(net, fl4, sk); 1068 if (IS_ERR(rt)) { 1069 err = PTR_ERR(rt); 1070 rt = NULL; 1071 if (err == -ENETUNREACH) 1072 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1073 goto out; 1074 } 1075 1076 err = -EACCES; 1077 if ((rt->rt_flags & RTCF_BROADCAST) && 1078 !sock_flag(sk, SOCK_BROADCAST)) 1079 goto out; 1080 if (connected) 1081 sk_dst_set(sk, dst_clone(&rt->dst)); 1082 } 1083 1084 if (msg->msg_flags&MSG_CONFIRM) 1085 goto do_confirm; 1086 back_from_confirm: 1087 1088 saddr = fl4->saddr; 1089 if (!ipc.addr) 1090 daddr = ipc.addr = fl4->daddr; 1091 1092 /* Lockless fast path for the non-corking case. */ 1093 if (!corkreq) { 1094 struct inet_cork cork; 1095 1096 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1097 sizeof(struct udphdr), &ipc, &rt, 1098 &cork, msg->msg_flags); 1099 err = PTR_ERR(skb); 1100 if (!IS_ERR_OR_NULL(skb)) 1101 err = udp_send_skb(skb, fl4, &cork); 1102 goto out; 1103 } 1104 1105 lock_sock(sk); 1106 if (unlikely(up->pending)) { 1107 /* The socket is already corked while preparing it. */ 1108 /* ... which is an evident application bug. --ANK */ 1109 release_sock(sk); 1110 1111 net_dbg_ratelimited("socket already corked\n"); 1112 err = -EINVAL; 1113 goto out; 1114 } 1115 /* 1116 * Now cork the socket to pend data. 1117 */ 1118 fl4 = &inet->cork.fl.u.ip4; 1119 fl4->daddr = daddr; 1120 fl4->saddr = saddr; 1121 fl4->fl4_dport = dport; 1122 fl4->fl4_sport = inet->inet_sport; 1123 up->pending = AF_INET; 1124 1125 do_append_data: 1126 up->len += ulen; 1127 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1128 sizeof(struct udphdr), &ipc, &rt, 1129 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1130 if (err) 1131 udp_flush_pending_frames(sk); 1132 else if (!corkreq) 1133 err = udp_push_pending_frames(sk); 1134 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1135 up->pending = 0; 1136 release_sock(sk); 1137 1138 out: 1139 ip_rt_put(rt); 1140 if (free) 1141 kfree(ipc.opt); 1142 if (!err) 1143 return len; 1144 /* 1145 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1146 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1147 * we don't have a good statistic (IpOutDiscards but it can be too many 1148 * things). We could add another new stat but at least for now that 1149 * seems like overkill. 1150 */ 1151 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1152 UDP_INC_STATS(sock_net(sk), 1153 UDP_MIB_SNDBUFERRORS, is_udplite); 1154 } 1155 return err; 1156 1157 do_confirm: 1158 if (msg->msg_flags & MSG_PROBE) 1159 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1160 if (!(msg->msg_flags&MSG_PROBE) || len) 1161 goto back_from_confirm; 1162 err = 0; 1163 goto out; 1164 } 1165 EXPORT_SYMBOL(udp_sendmsg); 1166 1167 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1168 size_t size, int flags) 1169 { 1170 struct inet_sock *inet = inet_sk(sk); 1171 struct udp_sock *up = udp_sk(sk); 1172 int ret; 1173 1174 if (flags & MSG_SENDPAGE_NOTLAST) 1175 flags |= MSG_MORE; 1176 1177 if (!up->pending) { 1178 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1179 1180 /* Call udp_sendmsg to specify destination address which 1181 * sendpage interface can't pass. 1182 * This will succeed only when the socket is connected. 1183 */ 1184 ret = udp_sendmsg(sk, &msg, 0); 1185 if (ret < 0) 1186 return ret; 1187 } 1188 1189 lock_sock(sk); 1190 1191 if (unlikely(!up->pending)) { 1192 release_sock(sk); 1193 1194 net_dbg_ratelimited("cork failed\n"); 1195 return -EINVAL; 1196 } 1197 1198 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1199 page, offset, size, flags); 1200 if (ret == -EOPNOTSUPP) { 1201 release_sock(sk); 1202 return sock_no_sendpage(sk->sk_socket, page, offset, 1203 size, flags); 1204 } 1205 if (ret < 0) { 1206 udp_flush_pending_frames(sk); 1207 goto out; 1208 } 1209 1210 up->len += size; 1211 if (!(up->corkflag || (flags&MSG_MORE))) 1212 ret = udp_push_pending_frames(sk); 1213 if (!ret) 1214 ret = size; 1215 out: 1216 release_sock(sk); 1217 return ret; 1218 } 1219 1220 #define UDP_SKB_IS_STATELESS 0x80000000 1221 1222 static void udp_set_dev_scratch(struct sk_buff *skb) 1223 { 1224 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1225 1226 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1227 scratch->_tsize_state = skb->truesize; 1228 #if BITS_PER_LONG == 64 1229 scratch->len = skb->len; 1230 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1231 scratch->is_linear = !skb_is_nonlinear(skb); 1232 #endif 1233 /* all head states execept sp (dst, sk, nf) are always cleared by 1234 * udp_rcv() and we need to preserve secpath, if present, to eventually 1235 * process IP_CMSG_PASSSEC at recvmsg() time 1236 */ 1237 if (likely(!skb_sec_path(skb))) 1238 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1239 } 1240 1241 static int udp_skb_truesize(struct sk_buff *skb) 1242 { 1243 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1244 } 1245 1246 static bool udp_skb_has_head_state(struct sk_buff *skb) 1247 { 1248 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1249 } 1250 1251 /* fully reclaim rmem/fwd memory allocated for skb */ 1252 static void udp_rmem_release(struct sock *sk, int size, int partial, 1253 bool rx_queue_lock_held) 1254 { 1255 struct udp_sock *up = udp_sk(sk); 1256 struct sk_buff_head *sk_queue; 1257 int amt; 1258 1259 if (likely(partial)) { 1260 up->forward_deficit += size; 1261 size = up->forward_deficit; 1262 if (size < (sk->sk_rcvbuf >> 2)) 1263 return; 1264 } else { 1265 size += up->forward_deficit; 1266 } 1267 up->forward_deficit = 0; 1268 1269 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1270 * if the called don't held it already 1271 */ 1272 sk_queue = &sk->sk_receive_queue; 1273 if (!rx_queue_lock_held) 1274 spin_lock(&sk_queue->lock); 1275 1276 1277 sk->sk_forward_alloc += size; 1278 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); 1279 sk->sk_forward_alloc -= amt; 1280 1281 if (amt) 1282 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); 1283 1284 atomic_sub(size, &sk->sk_rmem_alloc); 1285 1286 /* this can save us from acquiring the rx queue lock on next receive */ 1287 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1288 1289 if (!rx_queue_lock_held) 1290 spin_unlock(&sk_queue->lock); 1291 } 1292 1293 /* Note: called with reader_queue.lock held. 1294 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1295 * This avoids a cache line miss while receive_queue lock is held. 1296 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1297 */ 1298 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1299 { 1300 prefetch(&skb->data); 1301 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1302 } 1303 EXPORT_SYMBOL(udp_skb_destructor); 1304 1305 /* as above, but the caller held the rx queue lock, too */ 1306 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1307 { 1308 prefetch(&skb->data); 1309 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1310 } 1311 1312 /* Idea of busylocks is to let producers grab an extra spinlock 1313 * to relieve pressure on the receive_queue spinlock shared by consumer. 1314 * Under flood, this means that only one producer can be in line 1315 * trying to acquire the receive_queue spinlock. 1316 * These busylock can be allocated on a per cpu manner, instead of a 1317 * per socket one (that would consume a cache line per socket) 1318 */ 1319 static int udp_busylocks_log __read_mostly; 1320 static spinlock_t *udp_busylocks __read_mostly; 1321 1322 static spinlock_t *busylock_acquire(void *ptr) 1323 { 1324 spinlock_t *busy; 1325 1326 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1327 spin_lock(busy); 1328 return busy; 1329 } 1330 1331 static void busylock_release(spinlock_t *busy) 1332 { 1333 if (busy) 1334 spin_unlock(busy); 1335 } 1336 1337 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1338 { 1339 struct sk_buff_head *list = &sk->sk_receive_queue; 1340 int rmem, delta, amt, err = -ENOMEM; 1341 spinlock_t *busy = NULL; 1342 int size; 1343 1344 /* try to avoid the costly atomic add/sub pair when the receive 1345 * queue is full; always allow at least a packet 1346 */ 1347 rmem = atomic_read(&sk->sk_rmem_alloc); 1348 if (rmem > sk->sk_rcvbuf) 1349 goto drop; 1350 1351 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1352 * having linear skbs : 1353 * - Reduce memory overhead and thus increase receive queue capacity 1354 * - Less cache line misses at copyout() time 1355 * - Less work at consume_skb() (less alien page frag freeing) 1356 */ 1357 if (rmem > (sk->sk_rcvbuf >> 1)) { 1358 skb_condense(skb); 1359 1360 busy = busylock_acquire(sk); 1361 } 1362 size = skb->truesize; 1363 udp_set_dev_scratch(skb); 1364 1365 /* we drop only if the receive buf is full and the receive 1366 * queue contains some other skb 1367 */ 1368 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1369 if (rmem > (size + sk->sk_rcvbuf)) 1370 goto uncharge_drop; 1371 1372 spin_lock(&list->lock); 1373 if (size >= sk->sk_forward_alloc) { 1374 amt = sk_mem_pages(size); 1375 delta = amt << SK_MEM_QUANTUM_SHIFT; 1376 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1377 err = -ENOBUFS; 1378 spin_unlock(&list->lock); 1379 goto uncharge_drop; 1380 } 1381 1382 sk->sk_forward_alloc += delta; 1383 } 1384 1385 sk->sk_forward_alloc -= size; 1386 1387 /* no need to setup a destructor, we will explicitly release the 1388 * forward allocated memory on dequeue 1389 */ 1390 sock_skb_set_dropcount(sk, skb); 1391 1392 __skb_queue_tail(list, skb); 1393 spin_unlock(&list->lock); 1394 1395 if (!sock_flag(sk, SOCK_DEAD)) 1396 sk->sk_data_ready(sk); 1397 1398 busylock_release(busy); 1399 return 0; 1400 1401 uncharge_drop: 1402 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1403 1404 drop: 1405 atomic_inc(&sk->sk_drops); 1406 busylock_release(busy); 1407 return err; 1408 } 1409 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1410 1411 void udp_destruct_sock(struct sock *sk) 1412 { 1413 /* reclaim completely the forward allocated memory */ 1414 struct udp_sock *up = udp_sk(sk); 1415 unsigned int total = 0; 1416 struct sk_buff *skb; 1417 1418 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1419 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1420 total += skb->truesize; 1421 kfree_skb(skb); 1422 } 1423 udp_rmem_release(sk, total, 0, true); 1424 1425 inet_sock_destruct(sk); 1426 } 1427 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1428 1429 int udp_init_sock(struct sock *sk) 1430 { 1431 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1432 sk->sk_destruct = udp_destruct_sock; 1433 return 0; 1434 } 1435 EXPORT_SYMBOL_GPL(udp_init_sock); 1436 1437 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1438 { 1439 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1440 bool slow = lock_sock_fast(sk); 1441 1442 sk_peek_offset_bwd(sk, len); 1443 unlock_sock_fast(sk, slow); 1444 } 1445 1446 if (!skb_unref(skb)) 1447 return; 1448 1449 /* In the more common cases we cleared the head states previously, 1450 * see __udp_queue_rcv_skb(). 1451 */ 1452 if (unlikely(udp_skb_has_head_state(skb))) 1453 skb_release_head_state(skb); 1454 __consume_stateless_skb(skb); 1455 } 1456 EXPORT_SYMBOL_GPL(skb_consume_udp); 1457 1458 static struct sk_buff *__first_packet_length(struct sock *sk, 1459 struct sk_buff_head *rcvq, 1460 int *total) 1461 { 1462 struct sk_buff *skb; 1463 1464 while ((skb = skb_peek(rcvq)) != NULL) { 1465 if (udp_lib_checksum_complete(skb)) { 1466 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1467 IS_UDPLITE(sk)); 1468 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1469 IS_UDPLITE(sk)); 1470 atomic_inc(&sk->sk_drops); 1471 __skb_unlink(skb, rcvq); 1472 *total += skb->truesize; 1473 kfree_skb(skb); 1474 } else { 1475 /* the csum related bits could be changed, refresh 1476 * the scratch area 1477 */ 1478 udp_set_dev_scratch(skb); 1479 break; 1480 } 1481 } 1482 return skb; 1483 } 1484 1485 /** 1486 * first_packet_length - return length of first packet in receive queue 1487 * @sk: socket 1488 * 1489 * Drops all bad checksum frames, until a valid one is found. 1490 * Returns the length of found skb, or -1 if none is found. 1491 */ 1492 static int first_packet_length(struct sock *sk) 1493 { 1494 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1495 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1496 struct sk_buff *skb; 1497 int total = 0; 1498 int res; 1499 1500 spin_lock_bh(&rcvq->lock); 1501 skb = __first_packet_length(sk, rcvq, &total); 1502 if (!skb && !skb_queue_empty(sk_queue)) { 1503 spin_lock(&sk_queue->lock); 1504 skb_queue_splice_tail_init(sk_queue, rcvq); 1505 spin_unlock(&sk_queue->lock); 1506 1507 skb = __first_packet_length(sk, rcvq, &total); 1508 } 1509 res = skb ? skb->len : -1; 1510 if (total) 1511 udp_rmem_release(sk, total, 1, false); 1512 spin_unlock_bh(&rcvq->lock); 1513 return res; 1514 } 1515 1516 /* 1517 * IOCTL requests applicable to the UDP protocol 1518 */ 1519 1520 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1521 { 1522 switch (cmd) { 1523 case SIOCOUTQ: 1524 { 1525 int amount = sk_wmem_alloc_get(sk); 1526 1527 return put_user(amount, (int __user *)arg); 1528 } 1529 1530 case SIOCINQ: 1531 { 1532 int amount = max_t(int, 0, first_packet_length(sk)); 1533 1534 return put_user(amount, (int __user *)arg); 1535 } 1536 1537 default: 1538 return -ENOIOCTLCMD; 1539 } 1540 1541 return 0; 1542 } 1543 EXPORT_SYMBOL(udp_ioctl); 1544 1545 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1546 int noblock, int *peeked, int *off, int *err) 1547 { 1548 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1549 struct sk_buff_head *queue; 1550 struct sk_buff *last; 1551 long timeo; 1552 int error; 1553 1554 queue = &udp_sk(sk)->reader_queue; 1555 flags |= noblock ? MSG_DONTWAIT : 0; 1556 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1557 do { 1558 struct sk_buff *skb; 1559 1560 error = sock_error(sk); 1561 if (error) 1562 break; 1563 1564 error = -EAGAIN; 1565 *peeked = 0; 1566 do { 1567 spin_lock_bh(&queue->lock); 1568 skb = __skb_try_recv_from_queue(sk, queue, flags, 1569 udp_skb_destructor, 1570 peeked, off, err, 1571 &last); 1572 if (skb) { 1573 spin_unlock_bh(&queue->lock); 1574 return skb; 1575 } 1576 1577 if (skb_queue_empty(sk_queue)) { 1578 spin_unlock_bh(&queue->lock); 1579 goto busy_check; 1580 } 1581 1582 /* refill the reader queue and walk it again 1583 * keep both queues locked to avoid re-acquiring 1584 * the sk_receive_queue lock if fwd memory scheduling 1585 * is needed. 1586 */ 1587 spin_lock(&sk_queue->lock); 1588 skb_queue_splice_tail_init(sk_queue, queue); 1589 1590 skb = __skb_try_recv_from_queue(sk, queue, flags, 1591 udp_skb_dtor_locked, 1592 peeked, off, err, 1593 &last); 1594 spin_unlock(&sk_queue->lock); 1595 spin_unlock_bh(&queue->lock); 1596 if (skb) 1597 return skb; 1598 1599 busy_check: 1600 if (!sk_can_busy_loop(sk)) 1601 break; 1602 1603 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1604 } while (!skb_queue_empty(sk_queue)); 1605 1606 /* sk_queue is empty, reader_queue may contain peeked packets */ 1607 } while (timeo && 1608 !__skb_wait_for_more_packets(sk, &error, &timeo, 1609 (struct sk_buff *)sk_queue)); 1610 1611 *err = error; 1612 return NULL; 1613 } 1614 EXPORT_SYMBOL_GPL(__skb_recv_udp); 1615 1616 /* 1617 * This should be easy, if there is something there we 1618 * return it, otherwise we block. 1619 */ 1620 1621 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1622 int flags, int *addr_len) 1623 { 1624 struct inet_sock *inet = inet_sk(sk); 1625 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1626 struct sk_buff *skb; 1627 unsigned int ulen, copied; 1628 int peeked, peeking, off; 1629 int err; 1630 int is_udplite = IS_UDPLITE(sk); 1631 bool checksum_valid = false; 1632 1633 if (flags & MSG_ERRQUEUE) 1634 return ip_recv_error(sk, msg, len, addr_len); 1635 1636 try_again: 1637 peeking = flags & MSG_PEEK; 1638 off = sk_peek_offset(sk, flags); 1639 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); 1640 if (!skb) 1641 return err; 1642 1643 ulen = udp_skb_len(skb); 1644 copied = len; 1645 if (copied > ulen - off) 1646 copied = ulen - off; 1647 else if (copied < ulen) 1648 msg->msg_flags |= MSG_TRUNC; 1649 1650 /* 1651 * If checksum is needed at all, try to do it while copying the 1652 * data. If the data is truncated, or if we only want a partial 1653 * coverage checksum (UDP-Lite), do it before the copy. 1654 */ 1655 1656 if (copied < ulen || peeking || 1657 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1658 checksum_valid = udp_skb_csum_unnecessary(skb) || 1659 !__udp_lib_checksum_complete(skb); 1660 if (!checksum_valid) 1661 goto csum_copy_err; 1662 } 1663 1664 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1665 if (udp_skb_is_linear(skb)) 1666 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1667 else 1668 err = skb_copy_datagram_msg(skb, off, msg, copied); 1669 } else { 1670 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1671 1672 if (err == -EINVAL) 1673 goto csum_copy_err; 1674 } 1675 1676 if (unlikely(err)) { 1677 if (!peeked) { 1678 atomic_inc(&sk->sk_drops); 1679 UDP_INC_STATS(sock_net(sk), 1680 UDP_MIB_INERRORS, is_udplite); 1681 } 1682 kfree_skb(skb); 1683 return err; 1684 } 1685 1686 if (!peeked) 1687 UDP_INC_STATS(sock_net(sk), 1688 UDP_MIB_INDATAGRAMS, is_udplite); 1689 1690 sock_recv_ts_and_drops(msg, sk, skb); 1691 1692 /* Copy the address. */ 1693 if (sin) { 1694 sin->sin_family = AF_INET; 1695 sin->sin_port = udp_hdr(skb)->source; 1696 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1697 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1698 *addr_len = sizeof(*sin); 1699 } 1700 if (inet->cmsg_flags) 1701 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1702 1703 err = copied; 1704 if (flags & MSG_TRUNC) 1705 err = ulen; 1706 1707 skb_consume_udp(sk, skb, peeking ? -err : err); 1708 return err; 1709 1710 csum_copy_err: 1711 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1712 udp_skb_destructor)) { 1713 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1714 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1715 } 1716 kfree_skb(skb); 1717 1718 /* starting over for a new packet, but check if we need to yield */ 1719 cond_resched(); 1720 msg->msg_flags &= ~MSG_TRUNC; 1721 goto try_again; 1722 } 1723 1724 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1725 { 1726 /* This check is replicated from __ip4_datagram_connect() and 1727 * intended to prevent BPF program called below from accessing bytes 1728 * that are out of the bound specified by user in addr_len. 1729 */ 1730 if (addr_len < sizeof(struct sockaddr_in)) 1731 return -EINVAL; 1732 1733 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1734 } 1735 EXPORT_SYMBOL(udp_pre_connect); 1736 1737 int __udp_disconnect(struct sock *sk, int flags) 1738 { 1739 struct inet_sock *inet = inet_sk(sk); 1740 /* 1741 * 1003.1g - break association. 1742 */ 1743 1744 sk->sk_state = TCP_CLOSE; 1745 inet->inet_daddr = 0; 1746 inet->inet_dport = 0; 1747 sock_rps_reset_rxhash(sk); 1748 sk->sk_bound_dev_if = 0; 1749 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1750 inet_reset_saddr(sk); 1751 1752 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1753 sk->sk_prot->unhash(sk); 1754 inet->inet_sport = 0; 1755 } 1756 sk_dst_reset(sk); 1757 return 0; 1758 } 1759 EXPORT_SYMBOL(__udp_disconnect); 1760 1761 int udp_disconnect(struct sock *sk, int flags) 1762 { 1763 lock_sock(sk); 1764 __udp_disconnect(sk, flags); 1765 release_sock(sk); 1766 return 0; 1767 } 1768 EXPORT_SYMBOL(udp_disconnect); 1769 1770 void udp_lib_unhash(struct sock *sk) 1771 { 1772 if (sk_hashed(sk)) { 1773 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1774 struct udp_hslot *hslot, *hslot2; 1775 1776 hslot = udp_hashslot(udptable, sock_net(sk), 1777 udp_sk(sk)->udp_port_hash); 1778 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1779 1780 spin_lock_bh(&hslot->lock); 1781 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1782 reuseport_detach_sock(sk); 1783 if (sk_del_node_init_rcu(sk)) { 1784 hslot->count--; 1785 inet_sk(sk)->inet_num = 0; 1786 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1787 1788 spin_lock(&hslot2->lock); 1789 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1790 hslot2->count--; 1791 spin_unlock(&hslot2->lock); 1792 } 1793 spin_unlock_bh(&hslot->lock); 1794 } 1795 } 1796 EXPORT_SYMBOL(udp_lib_unhash); 1797 1798 /* 1799 * inet_rcv_saddr was changed, we must rehash secondary hash 1800 */ 1801 void udp_lib_rehash(struct sock *sk, u16 newhash) 1802 { 1803 if (sk_hashed(sk)) { 1804 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1805 struct udp_hslot *hslot, *hslot2, *nhslot2; 1806 1807 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1808 nhslot2 = udp_hashslot2(udptable, newhash); 1809 udp_sk(sk)->udp_portaddr_hash = newhash; 1810 1811 if (hslot2 != nhslot2 || 1812 rcu_access_pointer(sk->sk_reuseport_cb)) { 1813 hslot = udp_hashslot(udptable, sock_net(sk), 1814 udp_sk(sk)->udp_port_hash); 1815 /* we must lock primary chain too */ 1816 spin_lock_bh(&hslot->lock); 1817 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1818 reuseport_detach_sock(sk); 1819 1820 if (hslot2 != nhslot2) { 1821 spin_lock(&hslot2->lock); 1822 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1823 hslot2->count--; 1824 spin_unlock(&hslot2->lock); 1825 1826 spin_lock(&nhslot2->lock); 1827 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1828 &nhslot2->head); 1829 nhslot2->count++; 1830 spin_unlock(&nhslot2->lock); 1831 } 1832 1833 spin_unlock_bh(&hslot->lock); 1834 } 1835 } 1836 } 1837 EXPORT_SYMBOL(udp_lib_rehash); 1838 1839 static void udp_v4_rehash(struct sock *sk) 1840 { 1841 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 1842 inet_sk(sk)->inet_rcv_saddr, 1843 inet_sk(sk)->inet_num); 1844 udp_lib_rehash(sk, new_hash); 1845 } 1846 1847 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1848 { 1849 int rc; 1850 1851 if (inet_sk(sk)->inet_daddr) { 1852 sock_rps_save_rxhash(sk, skb); 1853 sk_mark_napi_id(sk, skb); 1854 sk_incoming_cpu_update(sk); 1855 } else { 1856 sk_mark_napi_id_once(sk, skb); 1857 } 1858 1859 rc = __udp_enqueue_schedule_skb(sk, skb); 1860 if (rc < 0) { 1861 int is_udplite = IS_UDPLITE(sk); 1862 1863 /* Note that an ENOMEM error is charged twice */ 1864 if (rc == -ENOMEM) 1865 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1866 is_udplite); 1867 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1868 kfree_skb(skb); 1869 trace_udp_fail_queue_rcv_skb(rc, sk); 1870 return -1; 1871 } 1872 1873 return 0; 1874 } 1875 1876 static struct static_key udp_encap_needed __read_mostly; 1877 void udp_encap_enable(void) 1878 { 1879 static_key_enable(&udp_encap_needed); 1880 } 1881 EXPORT_SYMBOL(udp_encap_enable); 1882 1883 /* returns: 1884 * -1: error 1885 * 0: success 1886 * >0: "udp encap" protocol resubmission 1887 * 1888 * Note that in the success and error cases, the skb is assumed to 1889 * have either been requeued or freed. 1890 */ 1891 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1892 { 1893 struct udp_sock *up = udp_sk(sk); 1894 int is_udplite = IS_UDPLITE(sk); 1895 1896 /* 1897 * Charge it to the socket, dropping if the queue is full. 1898 */ 1899 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1900 goto drop; 1901 nf_reset(skb); 1902 1903 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1904 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1905 1906 /* 1907 * This is an encapsulation socket so pass the skb to 1908 * the socket's udp_encap_rcv() hook. Otherwise, just 1909 * fall through and pass this up the UDP socket. 1910 * up->encap_rcv() returns the following value: 1911 * =0 if skb was successfully passed to the encap 1912 * handler or was discarded by it. 1913 * >0 if skb should be passed on to UDP. 1914 * <0 if skb should be resubmitted as proto -N 1915 */ 1916 1917 /* if we're overly short, let UDP handle it */ 1918 encap_rcv = READ_ONCE(up->encap_rcv); 1919 if (encap_rcv) { 1920 int ret; 1921 1922 /* Verify checksum before giving to encap */ 1923 if (udp_lib_checksum_complete(skb)) 1924 goto csum_error; 1925 1926 ret = encap_rcv(sk, skb); 1927 if (ret <= 0) { 1928 __UDP_INC_STATS(sock_net(sk), 1929 UDP_MIB_INDATAGRAMS, 1930 is_udplite); 1931 return -ret; 1932 } 1933 } 1934 1935 /* FALLTHROUGH -- it's a UDP Packet */ 1936 } 1937 1938 /* 1939 * UDP-Lite specific tests, ignored on UDP sockets 1940 */ 1941 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1942 1943 /* 1944 * MIB statistics other than incrementing the error count are 1945 * disabled for the following two types of errors: these depend 1946 * on the application settings, not on the functioning of the 1947 * protocol stack as such. 1948 * 1949 * RFC 3828 here recommends (sec 3.3): "There should also be a 1950 * way ... to ... at least let the receiving application block 1951 * delivery of packets with coverage values less than a value 1952 * provided by the application." 1953 */ 1954 if (up->pcrlen == 0) { /* full coverage was set */ 1955 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1956 UDP_SKB_CB(skb)->cscov, skb->len); 1957 goto drop; 1958 } 1959 /* The next case involves violating the min. coverage requested 1960 * by the receiver. This is subtle: if receiver wants x and x is 1961 * greater than the buffersize/MTU then receiver will complain 1962 * that it wants x while sender emits packets of smaller size y. 1963 * Therefore the above ...()->partial_cov statement is essential. 1964 */ 1965 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1966 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1967 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1968 goto drop; 1969 } 1970 } 1971 1972 prefetch(&sk->sk_rmem_alloc); 1973 if (rcu_access_pointer(sk->sk_filter) && 1974 udp_lib_checksum_complete(skb)) 1975 goto csum_error; 1976 1977 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 1978 goto drop; 1979 1980 udp_csum_pull_header(skb); 1981 1982 ipv4_pktinfo_prepare(sk, skb); 1983 return __udp_queue_rcv_skb(sk, skb); 1984 1985 csum_error: 1986 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1987 drop: 1988 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1989 atomic_inc(&sk->sk_drops); 1990 kfree_skb(skb); 1991 return -1; 1992 } 1993 1994 /* For TCP sockets, sk_rx_dst is protected by socket lock 1995 * For UDP, we use xchg() to guard against concurrent changes. 1996 */ 1997 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1998 { 1999 struct dst_entry *old; 2000 2001 if (dst_hold_safe(dst)) { 2002 old = xchg(&sk->sk_rx_dst, dst); 2003 dst_release(old); 2004 return old != dst; 2005 } 2006 return false; 2007 } 2008 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2009 2010 /* 2011 * Multicasts and broadcasts go to each listener. 2012 * 2013 * Note: called only from the BH handler context. 2014 */ 2015 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2016 struct udphdr *uh, 2017 __be32 saddr, __be32 daddr, 2018 struct udp_table *udptable, 2019 int proto) 2020 { 2021 struct sock *sk, *first = NULL; 2022 unsigned short hnum = ntohs(uh->dest); 2023 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2024 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2025 unsigned int offset = offsetof(typeof(*sk), sk_node); 2026 int dif = skb->dev->ifindex; 2027 int sdif = inet_sdif(skb); 2028 struct hlist_node *node; 2029 struct sk_buff *nskb; 2030 2031 if (use_hash2) { 2032 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2033 udptable->mask; 2034 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2035 start_lookup: 2036 hslot = &udptable->hash2[hash2]; 2037 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2038 } 2039 2040 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2041 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2042 uh->source, saddr, dif, sdif, hnum)) 2043 continue; 2044 2045 if (!first) { 2046 first = sk; 2047 continue; 2048 } 2049 nskb = skb_clone(skb, GFP_ATOMIC); 2050 2051 if (unlikely(!nskb)) { 2052 atomic_inc(&sk->sk_drops); 2053 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2054 IS_UDPLITE(sk)); 2055 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2056 IS_UDPLITE(sk)); 2057 continue; 2058 } 2059 if (udp_queue_rcv_skb(sk, nskb) > 0) 2060 consume_skb(nskb); 2061 } 2062 2063 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2064 if (use_hash2 && hash2 != hash2_any) { 2065 hash2 = hash2_any; 2066 goto start_lookup; 2067 } 2068 2069 if (first) { 2070 if (udp_queue_rcv_skb(first, skb) > 0) 2071 consume_skb(skb); 2072 } else { 2073 kfree_skb(skb); 2074 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2075 proto == IPPROTO_UDPLITE); 2076 } 2077 return 0; 2078 } 2079 2080 /* Initialize UDP checksum. If exited with zero value (success), 2081 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2082 * Otherwise, csum completion requires chacksumming packet body, 2083 * including udp header and folding it to skb->csum. 2084 */ 2085 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2086 int proto) 2087 { 2088 int err; 2089 2090 UDP_SKB_CB(skb)->partial_cov = 0; 2091 UDP_SKB_CB(skb)->cscov = skb->len; 2092 2093 if (proto == IPPROTO_UDPLITE) { 2094 err = udplite_checksum_init(skb, uh); 2095 if (err) 2096 return err; 2097 2098 if (UDP_SKB_CB(skb)->partial_cov) { 2099 skb->csum = inet_compute_pseudo(skb, proto); 2100 return 0; 2101 } 2102 } 2103 2104 /* Note, we are only interested in != 0 or == 0, thus the 2105 * force to int. 2106 */ 2107 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2108 inet_compute_pseudo); 2109 } 2110 2111 /* 2112 * All we need to do is get the socket, and then do a checksum. 2113 */ 2114 2115 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2116 int proto) 2117 { 2118 struct sock *sk; 2119 struct udphdr *uh; 2120 unsigned short ulen; 2121 struct rtable *rt = skb_rtable(skb); 2122 __be32 saddr, daddr; 2123 struct net *net = dev_net(skb->dev); 2124 2125 /* 2126 * Validate the packet. 2127 */ 2128 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2129 goto drop; /* No space for header. */ 2130 2131 uh = udp_hdr(skb); 2132 ulen = ntohs(uh->len); 2133 saddr = ip_hdr(skb)->saddr; 2134 daddr = ip_hdr(skb)->daddr; 2135 2136 if (ulen > skb->len) 2137 goto short_packet; 2138 2139 if (proto == IPPROTO_UDP) { 2140 /* UDP validates ulen. */ 2141 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2142 goto short_packet; 2143 uh = udp_hdr(skb); 2144 } 2145 2146 if (udp4_csum_init(skb, uh, proto)) 2147 goto csum_error; 2148 2149 sk = skb_steal_sock(skb); 2150 if (sk) { 2151 struct dst_entry *dst = skb_dst(skb); 2152 int ret; 2153 2154 if (unlikely(sk->sk_rx_dst != dst)) 2155 udp_sk_rx_dst_set(sk, dst); 2156 2157 ret = udp_queue_rcv_skb(sk, skb); 2158 sock_put(sk); 2159 /* a return value > 0 means to resubmit the input, but 2160 * it wants the return to be -protocol, or 0 2161 */ 2162 if (ret > 0) 2163 return -ret; 2164 return 0; 2165 } 2166 2167 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2168 return __udp4_lib_mcast_deliver(net, skb, uh, 2169 saddr, daddr, udptable, proto); 2170 2171 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2172 if (sk) { 2173 int ret; 2174 2175 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2176 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 2177 inet_compute_pseudo); 2178 2179 ret = udp_queue_rcv_skb(sk, skb); 2180 2181 /* a return value > 0 means to resubmit the input, but 2182 * it wants the return to be -protocol, or 0 2183 */ 2184 if (ret > 0) 2185 return -ret; 2186 return 0; 2187 } 2188 2189 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2190 goto drop; 2191 nf_reset(skb); 2192 2193 /* No socket. Drop packet silently, if checksum is wrong */ 2194 if (udp_lib_checksum_complete(skb)) 2195 goto csum_error; 2196 2197 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2198 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2199 2200 /* 2201 * Hmm. We got an UDP packet to a port to which we 2202 * don't wanna listen. Ignore it. 2203 */ 2204 kfree_skb(skb); 2205 return 0; 2206 2207 short_packet: 2208 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2209 proto == IPPROTO_UDPLITE ? "Lite" : "", 2210 &saddr, ntohs(uh->source), 2211 ulen, skb->len, 2212 &daddr, ntohs(uh->dest)); 2213 goto drop; 2214 2215 csum_error: 2216 /* 2217 * RFC1122: OK. Discards the bad packet silently (as far as 2218 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2219 */ 2220 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2221 proto == IPPROTO_UDPLITE ? "Lite" : "", 2222 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2223 ulen); 2224 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2225 drop: 2226 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2227 kfree_skb(skb); 2228 return 0; 2229 } 2230 2231 /* We can only early demux multicast if there is a single matching socket. 2232 * If more than one socket found returns NULL 2233 */ 2234 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2235 __be16 loc_port, __be32 loc_addr, 2236 __be16 rmt_port, __be32 rmt_addr, 2237 int dif, int sdif) 2238 { 2239 struct sock *sk, *result; 2240 unsigned short hnum = ntohs(loc_port); 2241 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2242 struct udp_hslot *hslot = &udp_table.hash[slot]; 2243 2244 /* Do not bother scanning a too big list */ 2245 if (hslot->count > 10) 2246 return NULL; 2247 2248 result = NULL; 2249 sk_for_each_rcu(sk, &hslot->head) { 2250 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2251 rmt_port, rmt_addr, dif, sdif, hnum)) { 2252 if (result) 2253 return NULL; 2254 result = sk; 2255 } 2256 } 2257 2258 return result; 2259 } 2260 2261 /* For unicast we should only early demux connected sockets or we can 2262 * break forwarding setups. The chains here can be long so only check 2263 * if the first socket is an exact match and if not move on. 2264 */ 2265 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2266 __be16 loc_port, __be32 loc_addr, 2267 __be16 rmt_port, __be32 rmt_addr, 2268 int dif, int sdif) 2269 { 2270 unsigned short hnum = ntohs(loc_port); 2271 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2272 unsigned int slot2 = hash2 & udp_table.mask; 2273 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2274 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2275 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2276 struct sock *sk; 2277 2278 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2279 if (INET_MATCH(sk, net, acookie, rmt_addr, 2280 loc_addr, ports, dif, sdif)) 2281 return sk; 2282 /* Only check first socket in chain */ 2283 break; 2284 } 2285 return NULL; 2286 } 2287 2288 int udp_v4_early_demux(struct sk_buff *skb) 2289 { 2290 struct net *net = dev_net(skb->dev); 2291 struct in_device *in_dev = NULL; 2292 const struct iphdr *iph; 2293 const struct udphdr *uh; 2294 struct sock *sk = NULL; 2295 struct dst_entry *dst; 2296 int dif = skb->dev->ifindex; 2297 int sdif = inet_sdif(skb); 2298 int ours; 2299 2300 /* validate the packet */ 2301 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2302 return 0; 2303 2304 iph = ip_hdr(skb); 2305 uh = udp_hdr(skb); 2306 2307 if (skb->pkt_type == PACKET_MULTICAST) { 2308 in_dev = __in_dev_get_rcu(skb->dev); 2309 2310 if (!in_dev) 2311 return 0; 2312 2313 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2314 iph->protocol); 2315 if (!ours) 2316 return 0; 2317 2318 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2319 uh->source, iph->saddr, 2320 dif, sdif); 2321 } else if (skb->pkt_type == PACKET_HOST) { 2322 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2323 uh->source, iph->saddr, dif, sdif); 2324 } 2325 2326 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2327 return 0; 2328 2329 skb->sk = sk; 2330 skb->destructor = sock_efree; 2331 dst = READ_ONCE(sk->sk_rx_dst); 2332 2333 if (dst) 2334 dst = dst_check(dst, 0); 2335 if (dst) { 2336 u32 itag = 0; 2337 2338 /* set noref for now. 2339 * any place which wants to hold dst has to call 2340 * dst_hold_safe() 2341 */ 2342 skb_dst_set_noref(skb, dst); 2343 2344 /* for unconnected multicast sockets we need to validate 2345 * the source on each packet 2346 */ 2347 if (!inet_sk(sk)->inet_daddr && in_dev) 2348 return ip_mc_validate_source(skb, iph->daddr, 2349 iph->saddr, iph->tos, 2350 skb->dev, in_dev, &itag); 2351 } 2352 return 0; 2353 } 2354 2355 int udp_rcv(struct sk_buff *skb) 2356 { 2357 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2358 } 2359 2360 void udp_destroy_sock(struct sock *sk) 2361 { 2362 struct udp_sock *up = udp_sk(sk); 2363 bool slow = lock_sock_fast(sk); 2364 udp_flush_pending_frames(sk); 2365 unlock_sock_fast(sk, slow); 2366 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2367 void (*encap_destroy)(struct sock *sk); 2368 encap_destroy = READ_ONCE(up->encap_destroy); 2369 if (encap_destroy) 2370 encap_destroy(sk); 2371 } 2372 } 2373 2374 /* 2375 * Socket option code for UDP 2376 */ 2377 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2378 char __user *optval, unsigned int optlen, 2379 int (*push_pending_frames)(struct sock *)) 2380 { 2381 struct udp_sock *up = udp_sk(sk); 2382 int val, valbool; 2383 int err = 0; 2384 int is_udplite = IS_UDPLITE(sk); 2385 2386 if (optlen < sizeof(int)) 2387 return -EINVAL; 2388 2389 if (get_user(val, (int __user *)optval)) 2390 return -EFAULT; 2391 2392 valbool = val ? 1 : 0; 2393 2394 switch (optname) { 2395 case UDP_CORK: 2396 if (val != 0) { 2397 up->corkflag = 1; 2398 } else { 2399 up->corkflag = 0; 2400 lock_sock(sk); 2401 push_pending_frames(sk); 2402 release_sock(sk); 2403 } 2404 break; 2405 2406 case UDP_ENCAP: 2407 switch (val) { 2408 case 0: 2409 case UDP_ENCAP_ESPINUDP: 2410 case UDP_ENCAP_ESPINUDP_NON_IKE: 2411 up->encap_rcv = xfrm4_udp_encap_rcv; 2412 /* FALLTHROUGH */ 2413 case UDP_ENCAP_L2TPINUDP: 2414 up->encap_type = val; 2415 udp_encap_enable(); 2416 break; 2417 default: 2418 err = -ENOPROTOOPT; 2419 break; 2420 } 2421 break; 2422 2423 case UDP_NO_CHECK6_TX: 2424 up->no_check6_tx = valbool; 2425 break; 2426 2427 case UDP_NO_CHECK6_RX: 2428 up->no_check6_rx = valbool; 2429 break; 2430 2431 case UDP_SEGMENT: 2432 if (val < 0 || val > USHRT_MAX) 2433 return -EINVAL; 2434 up->gso_size = val; 2435 break; 2436 2437 /* 2438 * UDP-Lite's partial checksum coverage (RFC 3828). 2439 */ 2440 /* The sender sets actual checksum coverage length via this option. 2441 * The case coverage > packet length is handled by send module. */ 2442 case UDPLITE_SEND_CSCOV: 2443 if (!is_udplite) /* Disable the option on UDP sockets */ 2444 return -ENOPROTOOPT; 2445 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2446 val = 8; 2447 else if (val > USHRT_MAX) 2448 val = USHRT_MAX; 2449 up->pcslen = val; 2450 up->pcflag |= UDPLITE_SEND_CC; 2451 break; 2452 2453 /* The receiver specifies a minimum checksum coverage value. To make 2454 * sense, this should be set to at least 8 (as done below). If zero is 2455 * used, this again means full checksum coverage. */ 2456 case UDPLITE_RECV_CSCOV: 2457 if (!is_udplite) /* Disable the option on UDP sockets */ 2458 return -ENOPROTOOPT; 2459 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2460 val = 8; 2461 else if (val > USHRT_MAX) 2462 val = USHRT_MAX; 2463 up->pcrlen = val; 2464 up->pcflag |= UDPLITE_RECV_CC; 2465 break; 2466 2467 default: 2468 err = -ENOPROTOOPT; 2469 break; 2470 } 2471 2472 return err; 2473 } 2474 EXPORT_SYMBOL(udp_lib_setsockopt); 2475 2476 int udp_setsockopt(struct sock *sk, int level, int optname, 2477 char __user *optval, unsigned int optlen) 2478 { 2479 if (level == SOL_UDP || level == SOL_UDPLITE) 2480 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2481 udp_push_pending_frames); 2482 return ip_setsockopt(sk, level, optname, optval, optlen); 2483 } 2484 2485 #ifdef CONFIG_COMPAT 2486 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2487 char __user *optval, unsigned int optlen) 2488 { 2489 if (level == SOL_UDP || level == SOL_UDPLITE) 2490 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2491 udp_push_pending_frames); 2492 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2493 } 2494 #endif 2495 2496 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2497 char __user *optval, int __user *optlen) 2498 { 2499 struct udp_sock *up = udp_sk(sk); 2500 int val, len; 2501 2502 if (get_user(len, optlen)) 2503 return -EFAULT; 2504 2505 len = min_t(unsigned int, len, sizeof(int)); 2506 2507 if (len < 0) 2508 return -EINVAL; 2509 2510 switch (optname) { 2511 case UDP_CORK: 2512 val = up->corkflag; 2513 break; 2514 2515 case UDP_ENCAP: 2516 val = up->encap_type; 2517 break; 2518 2519 case UDP_NO_CHECK6_TX: 2520 val = up->no_check6_tx; 2521 break; 2522 2523 case UDP_NO_CHECK6_RX: 2524 val = up->no_check6_rx; 2525 break; 2526 2527 case UDP_SEGMENT: 2528 val = up->gso_size; 2529 break; 2530 2531 /* The following two cannot be changed on UDP sockets, the return is 2532 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2533 case UDPLITE_SEND_CSCOV: 2534 val = up->pcslen; 2535 break; 2536 2537 case UDPLITE_RECV_CSCOV: 2538 val = up->pcrlen; 2539 break; 2540 2541 default: 2542 return -ENOPROTOOPT; 2543 } 2544 2545 if (put_user(len, optlen)) 2546 return -EFAULT; 2547 if (copy_to_user(optval, &val, len)) 2548 return -EFAULT; 2549 return 0; 2550 } 2551 EXPORT_SYMBOL(udp_lib_getsockopt); 2552 2553 int udp_getsockopt(struct sock *sk, int level, int optname, 2554 char __user *optval, int __user *optlen) 2555 { 2556 if (level == SOL_UDP || level == SOL_UDPLITE) 2557 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2558 return ip_getsockopt(sk, level, optname, optval, optlen); 2559 } 2560 2561 #ifdef CONFIG_COMPAT 2562 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2563 char __user *optval, int __user *optlen) 2564 { 2565 if (level == SOL_UDP || level == SOL_UDPLITE) 2566 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2567 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2568 } 2569 #endif 2570 /** 2571 * udp_poll - wait for a UDP event. 2572 * @file - file struct 2573 * @sock - socket 2574 * @wait - poll table 2575 * 2576 * This is same as datagram poll, except for the special case of 2577 * blocking sockets. If application is using a blocking fd 2578 * and a packet with checksum error is in the queue; 2579 * then it could get return from select indicating data available 2580 * but then block when reading it. Add special case code 2581 * to work around these arguably broken applications. 2582 */ 2583 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2584 { 2585 __poll_t mask = datagram_poll(file, sock, wait); 2586 struct sock *sk = sock->sk; 2587 2588 if (!skb_queue_empty(&udp_sk(sk)->reader_queue)) 2589 mask |= EPOLLIN | EPOLLRDNORM; 2590 2591 /* Check for false positives due to checksum errors */ 2592 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2593 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2594 mask &= ~(EPOLLIN | EPOLLRDNORM); 2595 2596 return mask; 2597 2598 } 2599 EXPORT_SYMBOL(udp_poll); 2600 2601 int udp_abort(struct sock *sk, int err) 2602 { 2603 lock_sock(sk); 2604 2605 sk->sk_err = err; 2606 sk->sk_error_report(sk); 2607 __udp_disconnect(sk, 0); 2608 2609 release_sock(sk); 2610 2611 return 0; 2612 } 2613 EXPORT_SYMBOL_GPL(udp_abort); 2614 2615 struct proto udp_prot = { 2616 .name = "UDP", 2617 .owner = THIS_MODULE, 2618 .close = udp_lib_close, 2619 .pre_connect = udp_pre_connect, 2620 .connect = ip4_datagram_connect, 2621 .disconnect = udp_disconnect, 2622 .ioctl = udp_ioctl, 2623 .init = udp_init_sock, 2624 .destroy = udp_destroy_sock, 2625 .setsockopt = udp_setsockopt, 2626 .getsockopt = udp_getsockopt, 2627 .sendmsg = udp_sendmsg, 2628 .recvmsg = udp_recvmsg, 2629 .sendpage = udp_sendpage, 2630 .release_cb = ip4_datagram_release_cb, 2631 .hash = udp_lib_hash, 2632 .unhash = udp_lib_unhash, 2633 .rehash = udp_v4_rehash, 2634 .get_port = udp_v4_get_port, 2635 .memory_allocated = &udp_memory_allocated, 2636 .sysctl_mem = sysctl_udp_mem, 2637 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2638 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2639 .obj_size = sizeof(struct udp_sock), 2640 .h.udp_table = &udp_table, 2641 #ifdef CONFIG_COMPAT 2642 .compat_setsockopt = compat_udp_setsockopt, 2643 .compat_getsockopt = compat_udp_getsockopt, 2644 #endif 2645 .diag_destroy = udp_abort, 2646 }; 2647 EXPORT_SYMBOL(udp_prot); 2648 2649 /* ------------------------------------------------------------------------ */ 2650 #ifdef CONFIG_PROC_FS 2651 2652 static struct sock *udp_get_first(struct seq_file *seq, int start) 2653 { 2654 struct sock *sk; 2655 struct udp_iter_state *state = seq->private; 2656 struct net *net = seq_file_net(seq); 2657 2658 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2659 ++state->bucket) { 2660 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2661 2662 if (hlist_empty(&hslot->head)) 2663 continue; 2664 2665 spin_lock_bh(&hslot->lock); 2666 sk_for_each(sk, &hslot->head) { 2667 if (!net_eq(sock_net(sk), net)) 2668 continue; 2669 if (sk->sk_family == state->family) 2670 goto found; 2671 } 2672 spin_unlock_bh(&hslot->lock); 2673 } 2674 sk = NULL; 2675 found: 2676 return sk; 2677 } 2678 2679 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2680 { 2681 struct udp_iter_state *state = seq->private; 2682 struct net *net = seq_file_net(seq); 2683 2684 do { 2685 sk = sk_next(sk); 2686 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2687 2688 if (!sk) { 2689 if (state->bucket <= state->udp_table->mask) 2690 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2691 return udp_get_first(seq, state->bucket + 1); 2692 } 2693 return sk; 2694 } 2695 2696 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2697 { 2698 struct sock *sk = udp_get_first(seq, 0); 2699 2700 if (sk) 2701 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2702 --pos; 2703 return pos ? NULL : sk; 2704 } 2705 2706 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2707 { 2708 struct udp_iter_state *state = seq->private; 2709 state->bucket = MAX_UDP_PORTS; 2710 2711 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2712 } 2713 2714 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2715 { 2716 struct sock *sk; 2717 2718 if (v == SEQ_START_TOKEN) 2719 sk = udp_get_idx(seq, 0); 2720 else 2721 sk = udp_get_next(seq, v); 2722 2723 ++*pos; 2724 return sk; 2725 } 2726 2727 static void udp_seq_stop(struct seq_file *seq, void *v) 2728 { 2729 struct udp_iter_state *state = seq->private; 2730 2731 if (state->bucket <= state->udp_table->mask) 2732 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2733 } 2734 2735 int udp_seq_open(struct inode *inode, struct file *file) 2736 { 2737 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2738 struct udp_iter_state *s; 2739 int err; 2740 2741 err = seq_open_net(inode, file, &afinfo->seq_ops, 2742 sizeof(struct udp_iter_state)); 2743 if (err < 0) 2744 return err; 2745 2746 s = ((struct seq_file *)file->private_data)->private; 2747 s->family = afinfo->family; 2748 s->udp_table = afinfo->udp_table; 2749 return err; 2750 } 2751 EXPORT_SYMBOL(udp_seq_open); 2752 2753 /* ------------------------------------------------------------------------ */ 2754 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2755 { 2756 struct proc_dir_entry *p; 2757 int rc = 0; 2758 2759 afinfo->seq_ops.start = udp_seq_start; 2760 afinfo->seq_ops.next = udp_seq_next; 2761 afinfo->seq_ops.stop = udp_seq_stop; 2762 2763 p = proc_create_data(afinfo->name, 0444, net->proc_net, 2764 afinfo->seq_fops, afinfo); 2765 if (!p) 2766 rc = -ENOMEM; 2767 return rc; 2768 } 2769 EXPORT_SYMBOL(udp_proc_register); 2770 2771 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2772 { 2773 remove_proc_entry(afinfo->name, net->proc_net); 2774 } 2775 EXPORT_SYMBOL(udp_proc_unregister); 2776 2777 /* ------------------------------------------------------------------------ */ 2778 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2779 int bucket) 2780 { 2781 struct inet_sock *inet = inet_sk(sp); 2782 __be32 dest = inet->inet_daddr; 2783 __be32 src = inet->inet_rcv_saddr; 2784 __u16 destp = ntohs(inet->inet_dport); 2785 __u16 srcp = ntohs(inet->inet_sport); 2786 2787 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2788 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2789 bucket, src, srcp, dest, destp, sp->sk_state, 2790 sk_wmem_alloc_get(sp), 2791 sk_rmem_alloc_get(sp), 2792 0, 0L, 0, 2793 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2794 0, sock_i_ino(sp), 2795 refcount_read(&sp->sk_refcnt), sp, 2796 atomic_read(&sp->sk_drops)); 2797 } 2798 2799 int udp4_seq_show(struct seq_file *seq, void *v) 2800 { 2801 seq_setwidth(seq, 127); 2802 if (v == SEQ_START_TOKEN) 2803 seq_puts(seq, " sl local_address rem_address st tx_queue " 2804 "rx_queue tr tm->when retrnsmt uid timeout " 2805 "inode ref pointer drops"); 2806 else { 2807 struct udp_iter_state *state = seq->private; 2808 2809 udp4_format_sock(v, seq, state->bucket); 2810 } 2811 seq_pad(seq, '\n'); 2812 return 0; 2813 } 2814 2815 static const struct file_operations udp_afinfo_seq_fops = { 2816 .open = udp_seq_open, 2817 .read = seq_read, 2818 .llseek = seq_lseek, 2819 .release = seq_release_net 2820 }; 2821 2822 /* ------------------------------------------------------------------------ */ 2823 static struct udp_seq_afinfo udp4_seq_afinfo = { 2824 .name = "udp", 2825 .family = AF_INET, 2826 .udp_table = &udp_table, 2827 .seq_fops = &udp_afinfo_seq_fops, 2828 .seq_ops = { 2829 .show = udp4_seq_show, 2830 }, 2831 }; 2832 2833 static int __net_init udp4_proc_init_net(struct net *net) 2834 { 2835 return udp_proc_register(net, &udp4_seq_afinfo); 2836 } 2837 2838 static void __net_exit udp4_proc_exit_net(struct net *net) 2839 { 2840 udp_proc_unregister(net, &udp4_seq_afinfo); 2841 } 2842 2843 static struct pernet_operations udp4_net_ops = { 2844 .init = udp4_proc_init_net, 2845 .exit = udp4_proc_exit_net, 2846 }; 2847 2848 int __init udp4_proc_init(void) 2849 { 2850 return register_pernet_subsys(&udp4_net_ops); 2851 } 2852 2853 void udp4_proc_exit(void) 2854 { 2855 unregister_pernet_subsys(&udp4_net_ops); 2856 } 2857 #endif /* CONFIG_PROC_FS */ 2858 2859 static __initdata unsigned long uhash_entries; 2860 static int __init set_uhash_entries(char *str) 2861 { 2862 ssize_t ret; 2863 2864 if (!str) 2865 return 0; 2866 2867 ret = kstrtoul(str, 0, &uhash_entries); 2868 if (ret) 2869 return 0; 2870 2871 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2872 uhash_entries = UDP_HTABLE_SIZE_MIN; 2873 return 1; 2874 } 2875 __setup("uhash_entries=", set_uhash_entries); 2876 2877 void __init udp_table_init(struct udp_table *table, const char *name) 2878 { 2879 unsigned int i; 2880 2881 table->hash = alloc_large_system_hash(name, 2882 2 * sizeof(struct udp_hslot), 2883 uhash_entries, 2884 21, /* one slot per 2 MB */ 2885 0, 2886 &table->log, 2887 &table->mask, 2888 UDP_HTABLE_SIZE_MIN, 2889 64 * 1024); 2890 2891 table->hash2 = table->hash + (table->mask + 1); 2892 for (i = 0; i <= table->mask; i++) { 2893 INIT_HLIST_HEAD(&table->hash[i].head); 2894 table->hash[i].count = 0; 2895 spin_lock_init(&table->hash[i].lock); 2896 } 2897 for (i = 0; i <= table->mask; i++) { 2898 INIT_HLIST_HEAD(&table->hash2[i].head); 2899 table->hash2[i].count = 0; 2900 spin_lock_init(&table->hash2[i].lock); 2901 } 2902 } 2903 2904 u32 udp_flow_hashrnd(void) 2905 { 2906 static u32 hashrnd __read_mostly; 2907 2908 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2909 2910 return hashrnd; 2911 } 2912 EXPORT_SYMBOL(udp_flow_hashrnd); 2913 2914 static void __udp_sysctl_init(struct net *net) 2915 { 2916 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2917 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2918 2919 #ifdef CONFIG_NET_L3_MASTER_DEV 2920 net->ipv4.sysctl_udp_l3mdev_accept = 0; 2921 #endif 2922 } 2923 2924 static int __net_init udp_sysctl_init(struct net *net) 2925 { 2926 __udp_sysctl_init(net); 2927 return 0; 2928 } 2929 2930 static struct pernet_operations __net_initdata udp_sysctl_ops = { 2931 .init = udp_sysctl_init, 2932 }; 2933 2934 void __init udp_init(void) 2935 { 2936 unsigned long limit; 2937 unsigned int i; 2938 2939 udp_table_init(&udp_table, "UDP"); 2940 limit = nr_free_buffer_pages() / 8; 2941 limit = max(limit, 128UL); 2942 sysctl_udp_mem[0] = limit / 4 * 3; 2943 sysctl_udp_mem[1] = limit; 2944 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2945 2946 __udp_sysctl_init(&init_net); 2947 2948 /* 16 spinlocks per cpu */ 2949 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 2950 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 2951 GFP_KERNEL); 2952 if (!udp_busylocks) 2953 panic("UDP: failed to alloc udp_busylocks\n"); 2954 for (i = 0; i < (1U << udp_busylocks_log); i++) 2955 spin_lock_init(udp_busylocks + i); 2956 2957 if (register_pernet_subsys(&udp_sysctl_ops)) 2958 panic("UDP: failed to init sysctl parameters.\n"); 2959 } 2960