1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The User Datagram Protocol (UDP). 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 12 * Hirokazu Takahashi, <taka@valinux.co.jp> 13 * 14 * Fixes: 15 * Alan Cox : verify_area() calls 16 * Alan Cox : stopped close while in use off icmp 17 * messages. Not a fix but a botch that 18 * for udp at least is 'valid'. 19 * Alan Cox : Fixed icmp handling properly 20 * Alan Cox : Correct error for oversized datagrams 21 * Alan Cox : Tidied select() semantics. 22 * Alan Cox : udp_err() fixed properly, also now 23 * select and read wake correctly on errors 24 * Alan Cox : udp_send verify_area moved to avoid mem leak 25 * Alan Cox : UDP can count its memory 26 * Alan Cox : send to an unknown connection causes 27 * an ECONNREFUSED off the icmp, but 28 * does NOT close. 29 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 31 * bug no longer crashes it. 32 * Fred Van Kempen : Net2e support for sk->broadcast. 33 * Alan Cox : Uses skb_free_datagram 34 * Alan Cox : Added get/set sockopt support. 35 * Alan Cox : Broadcasting without option set returns EACCES. 36 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 37 * Alan Cox : Use ip_tos and ip_ttl 38 * Alan Cox : SNMP Mibs 39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 40 * Matt Dillon : UDP length checks. 41 * Alan Cox : Smarter af_inet used properly. 42 * Alan Cox : Use new kernel side addressing. 43 * Alan Cox : Incorrect return on truncated datagram receive. 44 * Arnt Gulbrandsen : New udp_send and stuff 45 * Alan Cox : Cache last socket 46 * Alan Cox : Route cache 47 * Jon Peatfield : Minor efficiency fix to sendto(). 48 * Mike Shaver : RFC1122 checks. 49 * Alan Cox : Nonblocking error fix. 50 * Willy Konynenberg : Transparent proxying support. 51 * Mike McLagan : Routing by source 52 * David S. Miller : New socket lookup architecture. 53 * Last socket cache retained as it 54 * does have a high hit rate. 55 * Olaf Kirch : Don't linearise iovec on sendmsg. 56 * Andi Kleen : Some cleanups, cache destination entry 57 * for connect. 58 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 59 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 60 * return ENOTCONN for unconnected sockets (POSIX) 61 * Janos Farkas : don't deliver multi/broadcasts to a different 62 * bound-to-device socket 63 * Hirokazu Takahashi : HW checksumming for outgoing UDP 64 * datagrams. 65 * Hirokazu Takahashi : sendfile() on UDP works now. 66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 69 * a single port at the same time. 70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71 * James Chapman : Add L2TP encapsulation type. 72 * 73 * 74 * This program is free software; you can redistribute it and/or 75 * modify it under the terms of the GNU General Public License 76 * as published by the Free Software Foundation; either version 77 * 2 of the License, or (at your option) any later version. 78 */ 79 80 #define pr_fmt(fmt) "UDP: " fmt 81 82 #include <asm/uaccess.h> 83 #include <asm/ioctls.h> 84 #include <linux/bootmem.h> 85 #include <linux/highmem.h> 86 #include <linux/swap.h> 87 #include <linux/types.h> 88 #include <linux/fcntl.h> 89 #include <linux/module.h> 90 #include <linux/socket.h> 91 #include <linux/sockios.h> 92 #include <linux/igmp.h> 93 #include <linux/in.h> 94 #include <linux/errno.h> 95 #include <linux/timer.h> 96 #include <linux/mm.h> 97 #include <linux/inet.h> 98 #include <linux/netdevice.h> 99 #include <linux/slab.h> 100 #include <net/tcp_states.h> 101 #include <linux/skbuff.h> 102 #include <linux/netdevice.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" 116 117 struct udp_table udp_table __read_mostly; 118 EXPORT_SYMBOL(udp_table); 119 120 long sysctl_udp_mem[3] __read_mostly; 121 EXPORT_SYMBOL(sysctl_udp_mem); 122 123 int sysctl_udp_rmem_min __read_mostly; 124 EXPORT_SYMBOL(sysctl_udp_rmem_min); 125 126 int sysctl_udp_wmem_min __read_mostly; 127 EXPORT_SYMBOL(sysctl_udp_wmem_min); 128 129 atomic_long_t udp_memory_allocated; 130 EXPORT_SYMBOL(udp_memory_allocated); 131 132 #define MAX_UDP_PORTS 65536 133 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 134 135 static int udp_lib_lport_inuse(struct net *net, __u16 num, 136 const struct udp_hslot *hslot, 137 unsigned long *bitmap, 138 struct sock *sk, 139 int (*saddr_comp)(const struct sock *sk1, 140 const struct sock *sk2), 141 unsigned int log) 142 { 143 struct sock *sk2; 144 struct hlist_nulls_node *node; 145 kuid_t uid = sock_i_uid(sk); 146 147 sk_nulls_for_each(sk2, node, &hslot->head) { 148 if (net_eq(sock_net(sk2), net) && 149 sk2 != sk && 150 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 151 (!sk2->sk_reuse || !sk->sk_reuse) && 152 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 153 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 154 (!sk2->sk_reuseport || !sk->sk_reuseport || 155 !uid_eq(uid, sock_i_uid(sk2))) && 156 saddr_comp(sk, sk2)) { 157 if (!bitmap) 158 return 1; 159 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap); 160 } 161 } 162 return 0; 163 } 164 165 /* 166 * Note: we still hold spinlock of primary hash chain, so no other writer 167 * can insert/delete a socket with local_port == num 168 */ 169 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 170 struct udp_hslot *hslot2, 171 struct sock *sk, 172 int (*saddr_comp)(const struct sock *sk1, 173 const struct sock *sk2)) 174 { 175 struct sock *sk2; 176 struct hlist_nulls_node *node; 177 kuid_t uid = sock_i_uid(sk); 178 int res = 0; 179 180 spin_lock(&hslot2->lock); 181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) { 182 if (net_eq(sock_net(sk2), net) && 183 sk2 != sk && 184 (udp_sk(sk2)->udp_port_hash == num) && 185 (!sk2->sk_reuse || !sk->sk_reuse) && 186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 188 (!sk2->sk_reuseport || !sk->sk_reuseport || 189 !uid_eq(uid, sock_i_uid(sk2))) && 190 saddr_comp(sk, sk2)) { 191 res = 1; 192 break; 193 } 194 } 195 spin_unlock(&hslot2->lock); 196 return res; 197 } 198 199 /** 200 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 201 * 202 * @sk: socket struct in question 203 * @snum: port number to look up 204 * @saddr_comp: AF-dependent comparison of bound local IP addresses 205 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 206 * with NULL address 207 */ 208 int udp_lib_get_port(struct sock *sk, unsigned short snum, 209 int (*saddr_comp)(const struct sock *sk1, 210 const struct sock *sk2), 211 unsigned int hash2_nulladdr) 212 { 213 struct udp_hslot *hslot, *hslot2; 214 struct udp_table *udptable = sk->sk_prot->h.udp_table; 215 int error = 1; 216 struct net *net = sock_net(sk); 217 218 if (!snum) { 219 int low, high, remaining; 220 unsigned int rand; 221 unsigned short first, last; 222 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 223 224 inet_get_local_port_range(net, &low, &high); 225 remaining = (high - low) + 1; 226 227 rand = prandom_u32(); 228 first = reciprocal_scale(rand, remaining) + low; 229 /* 230 * force rand to be an odd multiple of UDP_HTABLE_SIZE 231 */ 232 rand = (rand | 1) * (udptable->mask + 1); 233 last = first + udptable->mask + 1; 234 do { 235 hslot = udp_hashslot(udptable, net, first); 236 bitmap_zero(bitmap, PORTS_PER_CHAIN); 237 spin_lock_bh(&hslot->lock); 238 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 239 saddr_comp, udptable->log); 240 241 snum = first; 242 /* 243 * Iterate on all possible values of snum for this hash. 244 * Using steps of an odd multiple of UDP_HTABLE_SIZE 245 * give us randomization and full range coverage. 246 */ 247 do { 248 if (low <= snum && snum <= high && 249 !test_bit(snum >> udptable->log, bitmap) && 250 !inet_is_local_reserved_port(net, snum)) 251 goto found; 252 snum += rand; 253 } while (snum != first); 254 spin_unlock_bh(&hslot->lock); 255 } while (++first != last); 256 goto fail; 257 } else { 258 hslot = udp_hashslot(udptable, net, snum); 259 spin_lock_bh(&hslot->lock); 260 if (hslot->count > 10) { 261 int exist; 262 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 263 264 slot2 &= udptable->mask; 265 hash2_nulladdr &= udptable->mask; 266 267 hslot2 = udp_hashslot2(udptable, slot2); 268 if (hslot->count < hslot2->count) 269 goto scan_primary_hash; 270 271 exist = udp_lib_lport_inuse2(net, snum, hslot2, 272 sk, saddr_comp); 273 if (!exist && (hash2_nulladdr != slot2)) { 274 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 275 exist = udp_lib_lport_inuse2(net, snum, hslot2, 276 sk, saddr_comp); 277 } 278 if (exist) 279 goto fail_unlock; 280 else 281 goto found; 282 } 283 scan_primary_hash: 284 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 285 saddr_comp, 0)) 286 goto fail_unlock; 287 } 288 found: 289 inet_sk(sk)->inet_num = snum; 290 udp_sk(sk)->udp_port_hash = snum; 291 udp_sk(sk)->udp_portaddr_hash ^= snum; 292 if (sk_unhashed(sk)) { 293 sk_nulls_add_node_rcu(sk, &hslot->head); 294 hslot->count++; 295 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 296 297 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 298 spin_lock(&hslot2->lock); 299 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 300 &hslot2->head); 301 hslot2->count++; 302 spin_unlock(&hslot2->lock); 303 } 304 error = 0; 305 fail_unlock: 306 spin_unlock_bh(&hslot->lock); 307 fail: 308 return error; 309 } 310 EXPORT_SYMBOL(udp_lib_get_port); 311 312 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 313 { 314 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 315 316 return (!ipv6_only_sock(sk2) && 317 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 318 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 319 } 320 321 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 322 unsigned int port) 323 { 324 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 325 } 326 327 int udp_v4_get_port(struct sock *sk, unsigned short snum) 328 { 329 unsigned int hash2_nulladdr = 330 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 331 unsigned int hash2_partial = 332 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 333 334 /* precompute partial secondary hash */ 335 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 336 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 337 } 338 339 static inline int compute_score(struct sock *sk, struct net *net, 340 __be32 saddr, unsigned short hnum, __be16 sport, 341 __be32 daddr, __be16 dport, int dif) 342 { 343 int score; 344 struct inet_sock *inet; 345 346 if (!net_eq(sock_net(sk), net) || 347 udp_sk(sk)->udp_port_hash != hnum || 348 ipv6_only_sock(sk)) 349 return -1; 350 351 score = (sk->sk_family == PF_INET) ? 2 : 1; 352 inet = inet_sk(sk); 353 354 if (inet->inet_rcv_saddr) { 355 if (inet->inet_rcv_saddr != daddr) 356 return -1; 357 score += 4; 358 } 359 360 if (inet->inet_daddr) { 361 if (inet->inet_daddr != saddr) 362 return -1; 363 score += 4; 364 } 365 366 if (inet->inet_dport) { 367 if (inet->inet_dport != sport) 368 return -1; 369 score += 4; 370 } 371 372 if (sk->sk_bound_dev_if) { 373 if (sk->sk_bound_dev_if != dif) 374 return -1; 375 score += 4; 376 } 377 378 return score; 379 } 380 381 /* 382 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 383 */ 384 static inline int compute_score2(struct sock *sk, struct net *net, 385 __be32 saddr, __be16 sport, 386 __be32 daddr, unsigned int hnum, int dif) 387 { 388 int score; 389 struct inet_sock *inet; 390 391 if (!net_eq(sock_net(sk), net) || 392 ipv6_only_sock(sk)) 393 return -1; 394 395 inet = inet_sk(sk); 396 397 if (inet->inet_rcv_saddr != daddr || 398 inet->inet_num != hnum) 399 return -1; 400 401 score = (sk->sk_family == PF_INET) ? 2 : 1; 402 403 if (inet->inet_daddr) { 404 if (inet->inet_daddr != saddr) 405 return -1; 406 score += 4; 407 } 408 409 if (inet->inet_dport) { 410 if (inet->inet_dport != sport) 411 return -1; 412 score += 4; 413 } 414 415 if (sk->sk_bound_dev_if) { 416 if (sk->sk_bound_dev_if != dif) 417 return -1; 418 score += 4; 419 } 420 421 return score; 422 } 423 424 static unsigned int udp_ehashfn(struct net *net, const __be32 laddr, 425 const __u16 lport, const __be32 faddr, 426 const __be16 fport) 427 { 428 static u32 udp_ehash_secret __read_mostly; 429 430 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 431 432 return __inet_ehashfn(laddr, lport, faddr, fport, 433 udp_ehash_secret + net_hash_mix(net)); 434 } 435 436 437 /* called with read_rcu_lock() */ 438 static struct sock *udp4_lib_lookup2(struct net *net, 439 __be32 saddr, __be16 sport, 440 __be32 daddr, unsigned int hnum, int dif, 441 struct udp_hslot *hslot2, unsigned int slot2) 442 { 443 struct sock *sk, *result; 444 struct hlist_nulls_node *node; 445 int score, badness, matches = 0, reuseport = 0; 446 u32 hash = 0; 447 448 begin: 449 result = NULL; 450 badness = 0; 451 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 452 score = compute_score2(sk, net, saddr, sport, 453 daddr, hnum, dif); 454 if (score > badness) { 455 result = sk; 456 badness = score; 457 reuseport = sk->sk_reuseport; 458 if (reuseport) { 459 hash = udp_ehashfn(net, daddr, hnum, 460 saddr, sport); 461 matches = 1; 462 } 463 } else if (score == badness && reuseport) { 464 matches++; 465 if (reciprocal_scale(hash, matches) == 0) 466 result = sk; 467 hash = next_pseudo_random32(hash); 468 } 469 } 470 /* 471 * if the nulls value we got at the end of this lookup is 472 * not the expected one, we must restart lookup. 473 * We probably met an item that was moved to another chain. 474 */ 475 if (get_nulls_value(node) != slot2) 476 goto begin; 477 if (result) { 478 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 479 result = NULL; 480 else if (unlikely(compute_score2(result, net, saddr, sport, 481 daddr, hnum, dif) < badness)) { 482 sock_put(result); 483 goto begin; 484 } 485 } 486 return result; 487 } 488 489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 490 * harder than this. -DaveM 491 */ 492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 493 __be16 sport, __be32 daddr, __be16 dport, 494 int dif, struct udp_table *udptable) 495 { 496 struct sock *sk, *result; 497 struct hlist_nulls_node *node; 498 unsigned short hnum = ntohs(dport); 499 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 500 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 501 int score, badness, matches = 0, reuseport = 0; 502 u32 hash = 0; 503 504 rcu_read_lock(); 505 if (hslot->count > 10) { 506 hash2 = udp4_portaddr_hash(net, daddr, hnum); 507 slot2 = hash2 & udptable->mask; 508 hslot2 = &udptable->hash2[slot2]; 509 if (hslot->count < hslot2->count) 510 goto begin; 511 512 result = udp4_lib_lookup2(net, saddr, sport, 513 daddr, hnum, dif, 514 hslot2, slot2); 515 if (!result) { 516 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 517 slot2 = hash2 & udptable->mask; 518 hslot2 = &udptable->hash2[slot2]; 519 if (hslot->count < hslot2->count) 520 goto begin; 521 522 result = udp4_lib_lookup2(net, saddr, sport, 523 htonl(INADDR_ANY), hnum, dif, 524 hslot2, slot2); 525 } 526 rcu_read_unlock(); 527 return result; 528 } 529 begin: 530 result = NULL; 531 badness = 0; 532 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 533 score = compute_score(sk, net, saddr, hnum, sport, 534 daddr, dport, dif); 535 if (score > badness) { 536 result = sk; 537 badness = score; 538 reuseport = sk->sk_reuseport; 539 if (reuseport) { 540 hash = udp_ehashfn(net, daddr, hnum, 541 saddr, sport); 542 matches = 1; 543 } 544 } else if (score == badness && reuseport) { 545 matches++; 546 if (reciprocal_scale(hash, matches) == 0) 547 result = sk; 548 hash = next_pseudo_random32(hash); 549 } 550 } 551 /* 552 * if the nulls value we got at the end of this lookup is 553 * not the expected one, we must restart lookup. 554 * We probably met an item that was moved to another chain. 555 */ 556 if (get_nulls_value(node) != slot) 557 goto begin; 558 559 if (result) { 560 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 561 result = NULL; 562 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 563 daddr, dport, dif) < badness)) { 564 sock_put(result); 565 goto begin; 566 } 567 } 568 rcu_read_unlock(); 569 return result; 570 } 571 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 572 573 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 574 __be16 sport, __be16 dport, 575 struct udp_table *udptable) 576 { 577 const struct iphdr *iph = ip_hdr(skb); 578 579 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 580 iph->daddr, dport, inet_iif(skb), 581 udptable); 582 } 583 584 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 585 __be32 daddr, __be16 dport, int dif) 586 { 587 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 588 } 589 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 590 591 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 592 __be16 loc_port, __be32 loc_addr, 593 __be16 rmt_port, __be32 rmt_addr, 594 int dif, unsigned short hnum) 595 { 596 struct inet_sock *inet = inet_sk(sk); 597 598 if (!net_eq(sock_net(sk), net) || 599 udp_sk(sk)->udp_port_hash != hnum || 600 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 601 (inet->inet_dport != rmt_port && inet->inet_dport) || 602 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 603 ipv6_only_sock(sk) || 604 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 605 return false; 606 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 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, udptable); 636 if (sk == NULL) { 637 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 638 return; /* No socket for error */ 639 } 640 641 err = 0; 642 harderr = 0; 643 inet = inet_sk(sk); 644 645 switch (type) { 646 default: 647 case ICMP_TIME_EXCEEDED: 648 err = EHOSTUNREACH; 649 break; 650 case ICMP_SOURCE_QUENCH: 651 goto out; 652 case ICMP_PARAMETERPROB: 653 err = EPROTO; 654 harderr = 1; 655 break; 656 case ICMP_DEST_UNREACH: 657 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 658 ipv4_sk_update_pmtu(skb, sk, info); 659 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 660 err = EMSGSIZE; 661 harderr = 1; 662 break; 663 } 664 goto out; 665 } 666 err = EHOSTUNREACH; 667 if (code <= NR_ICMP_UNREACH) { 668 harderr = icmp_err_convert[code].fatal; 669 err = icmp_err_convert[code].errno; 670 } 671 break; 672 case ICMP_REDIRECT: 673 ipv4_sk_redirect(skb, sk); 674 goto out; 675 } 676 677 /* 678 * RFC1122: OK. Passes ICMP errors back to application, as per 679 * 4.1.3.3. 680 */ 681 if (!inet->recverr) { 682 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 683 goto out; 684 } else 685 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 686 687 sk->sk_err = err; 688 sk->sk_error_report(sk); 689 out: 690 sock_put(sk); 691 } 692 693 void udp_err(struct sk_buff *skb, u32 info) 694 { 695 __udp4_lib_err(skb, info, &udp_table); 696 } 697 698 /* 699 * Throw away all pending data and cancel the corking. Socket is locked. 700 */ 701 void udp_flush_pending_frames(struct sock *sk) 702 { 703 struct udp_sock *up = udp_sk(sk); 704 705 if (up->pending) { 706 up->len = 0; 707 up->pending = 0; 708 ip_flush_pending_frames(sk); 709 } 710 } 711 EXPORT_SYMBOL(udp_flush_pending_frames); 712 713 /** 714 * udp4_hwcsum - handle outgoing HW checksumming 715 * @skb: sk_buff containing the filled-in UDP header 716 * (checksum field must be zeroed out) 717 * @src: source IP address 718 * @dst: destination IP address 719 */ 720 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 721 { 722 struct udphdr *uh = udp_hdr(skb); 723 int offset = skb_transport_offset(skb); 724 int len = skb->len - offset; 725 int hlen = len; 726 __wsum csum = 0; 727 728 if (!skb_has_frag_list(skb)) { 729 /* 730 * Only one fragment on the socket. 731 */ 732 skb->csum_start = skb_transport_header(skb) - skb->head; 733 skb->csum_offset = offsetof(struct udphdr, check); 734 uh->check = ~csum_tcpudp_magic(src, dst, len, 735 IPPROTO_UDP, 0); 736 } else { 737 struct sk_buff *frags; 738 739 /* 740 * HW-checksum won't work as there are two or more 741 * fragments on the socket so that all csums of sk_buffs 742 * should be together 743 */ 744 skb_walk_frags(skb, frags) { 745 csum = csum_add(csum, frags->csum); 746 hlen -= frags->len; 747 } 748 749 csum = skb_checksum(skb, offset, hlen, csum); 750 skb->ip_summed = CHECKSUM_NONE; 751 752 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 753 if (uh->check == 0) 754 uh->check = CSUM_MANGLED_0; 755 } 756 } 757 EXPORT_SYMBOL_GPL(udp4_hwcsum); 758 759 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 760 * for the simple case like when setting the checksum for a UDP tunnel. 761 */ 762 void udp_set_csum(bool nocheck, struct sk_buff *skb, 763 __be32 saddr, __be32 daddr, int len) 764 { 765 struct udphdr *uh = udp_hdr(skb); 766 767 if (nocheck) 768 uh->check = 0; 769 else if (skb_is_gso(skb)) 770 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 771 else if (skb_dst(skb) && skb_dst(skb)->dev && 772 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) { 773 774 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 775 776 skb->ip_summed = CHECKSUM_PARTIAL; 777 skb->csum_start = skb_transport_header(skb) - skb->head; 778 skb->csum_offset = offsetof(struct udphdr, check); 779 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 780 } else { 781 __wsum csum; 782 783 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 784 785 uh->check = 0; 786 csum = skb_checksum(skb, 0, len, 0); 787 uh->check = udp_v4_check(len, saddr, daddr, csum); 788 if (uh->check == 0) 789 uh->check = CSUM_MANGLED_0; 790 791 skb->ip_summed = CHECKSUM_UNNECESSARY; 792 } 793 } 794 EXPORT_SYMBOL(udp_set_csum); 795 796 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 797 { 798 struct sock *sk = skb->sk; 799 struct inet_sock *inet = inet_sk(sk); 800 struct udphdr *uh; 801 int err = 0; 802 int is_udplite = IS_UDPLITE(sk); 803 int offset = skb_transport_offset(skb); 804 int len = skb->len - offset; 805 __wsum csum = 0; 806 807 /* 808 * Create a UDP header 809 */ 810 uh = udp_hdr(skb); 811 uh->source = inet->inet_sport; 812 uh->dest = fl4->fl4_dport; 813 uh->len = htons(len); 814 uh->check = 0; 815 816 if (is_udplite) /* UDP-Lite */ 817 csum = udplite_csum(skb); 818 819 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 820 821 skb->ip_summed = CHECKSUM_NONE; 822 goto send; 823 824 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 825 826 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 827 goto send; 828 829 } else 830 csum = udp_csum(skb); 831 832 /* add protocol-dependent pseudo-header */ 833 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 834 sk->sk_protocol, csum); 835 if (uh->check == 0) 836 uh->check = CSUM_MANGLED_0; 837 838 send: 839 err = ip_send_skb(sock_net(sk), skb); 840 if (err) { 841 if (err == -ENOBUFS && !inet->recverr) { 842 UDP_INC_STATS_USER(sock_net(sk), 843 UDP_MIB_SNDBUFERRORS, is_udplite); 844 err = 0; 845 } 846 } else 847 UDP_INC_STATS_USER(sock_net(sk), 848 UDP_MIB_OUTDATAGRAMS, is_udplite); 849 return err; 850 } 851 852 /* 853 * Push out all pending data as one UDP datagram. Socket is locked. 854 */ 855 int udp_push_pending_frames(struct sock *sk) 856 { 857 struct udp_sock *up = udp_sk(sk); 858 struct inet_sock *inet = inet_sk(sk); 859 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 860 struct sk_buff *skb; 861 int err = 0; 862 863 skb = ip_finish_skb(sk, fl4); 864 if (!skb) 865 goto out; 866 867 err = udp_send_skb(skb, fl4); 868 869 out: 870 up->len = 0; 871 up->pending = 0; 872 return err; 873 } 874 EXPORT_SYMBOL(udp_push_pending_frames); 875 876 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 877 size_t len) 878 { 879 struct inet_sock *inet = inet_sk(sk); 880 struct udp_sock *up = udp_sk(sk); 881 struct flowi4 fl4_stack; 882 struct flowi4 *fl4; 883 int ulen = len; 884 struct ipcm_cookie ipc; 885 struct rtable *rt = NULL; 886 int free = 0; 887 int connected = 0; 888 __be32 daddr, faddr, saddr; 889 __be16 dport; 890 u8 tos; 891 int err, is_udplite = IS_UDPLITE(sk); 892 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 893 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 894 struct sk_buff *skb; 895 struct ip_options_data opt_copy; 896 897 if (len > 0xFFFF) 898 return -EMSGSIZE; 899 900 /* 901 * Check the flags. 902 */ 903 904 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 905 return -EOPNOTSUPP; 906 907 ipc.opt = NULL; 908 ipc.tx_flags = 0; 909 ipc.ttl = 0; 910 ipc.tos = -1; 911 912 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 913 914 fl4 = &inet->cork.fl.u.ip4; 915 if (up->pending) { 916 /* 917 * There are pending frames. 918 * The socket lock must be held while it's corked. 919 */ 920 lock_sock(sk); 921 if (likely(up->pending)) { 922 if (unlikely(up->pending != AF_INET)) { 923 release_sock(sk); 924 return -EINVAL; 925 } 926 goto do_append_data; 927 } 928 release_sock(sk); 929 } 930 ulen += sizeof(struct udphdr); 931 932 /* 933 * Get and verify the address. 934 */ 935 if (msg->msg_name) { 936 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 937 if (msg->msg_namelen < sizeof(*usin)) 938 return -EINVAL; 939 if (usin->sin_family != AF_INET) { 940 if (usin->sin_family != AF_UNSPEC) 941 return -EAFNOSUPPORT; 942 } 943 944 daddr = usin->sin_addr.s_addr; 945 dport = usin->sin_port; 946 if (dport == 0) 947 return -EINVAL; 948 } else { 949 if (sk->sk_state != TCP_ESTABLISHED) 950 return -EDESTADDRREQ; 951 daddr = inet->inet_daddr; 952 dport = inet->inet_dport; 953 /* Open fast path for connected socket. 954 Route will not be used, if at least one option is set. 955 */ 956 connected = 1; 957 } 958 ipc.addr = inet->inet_saddr; 959 960 ipc.oif = sk->sk_bound_dev_if; 961 962 sock_tx_timestamp(sk, &ipc.tx_flags); 963 964 if (msg->msg_controllen) { 965 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 966 sk->sk_family == AF_INET6); 967 if (err) 968 return err; 969 if (ipc.opt) 970 free = 1; 971 connected = 0; 972 } 973 if (!ipc.opt) { 974 struct ip_options_rcu *inet_opt; 975 976 rcu_read_lock(); 977 inet_opt = rcu_dereference(inet->inet_opt); 978 if (inet_opt) { 979 memcpy(&opt_copy, inet_opt, 980 sizeof(*inet_opt) + inet_opt->opt.optlen); 981 ipc.opt = &opt_copy.opt; 982 } 983 rcu_read_unlock(); 984 } 985 986 saddr = ipc.addr; 987 ipc.addr = faddr = daddr; 988 989 if (ipc.opt && ipc.opt->opt.srr) { 990 if (!daddr) 991 return -EINVAL; 992 faddr = ipc.opt->opt.faddr; 993 connected = 0; 994 } 995 tos = get_rttos(&ipc, inet); 996 if (sock_flag(sk, SOCK_LOCALROUTE) || 997 (msg->msg_flags & MSG_DONTROUTE) || 998 (ipc.opt && ipc.opt->opt.is_strictroute)) { 999 tos |= RTO_ONLINK; 1000 connected = 0; 1001 } 1002 1003 if (ipv4_is_multicast(daddr)) { 1004 if (!ipc.oif) 1005 ipc.oif = inet->mc_index; 1006 if (!saddr) 1007 saddr = inet->mc_addr; 1008 connected = 0; 1009 } else if (!ipc.oif) 1010 ipc.oif = inet->uc_index; 1011 1012 if (connected) 1013 rt = (struct rtable *)sk_dst_check(sk, 0); 1014 1015 if (rt == NULL) { 1016 struct net *net = sock_net(sk); 1017 1018 fl4 = &fl4_stack; 1019 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1020 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1021 inet_sk_flowi_flags(sk), 1022 faddr, saddr, dport, inet->inet_sport); 1023 1024 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1025 rt = ip_route_output_flow(net, fl4, sk); 1026 if (IS_ERR(rt)) { 1027 err = PTR_ERR(rt); 1028 rt = NULL; 1029 if (err == -ENETUNREACH) 1030 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1031 goto out; 1032 } 1033 1034 err = -EACCES; 1035 if ((rt->rt_flags & RTCF_BROADCAST) && 1036 !sock_flag(sk, SOCK_BROADCAST)) 1037 goto out; 1038 if (connected) 1039 sk_dst_set(sk, dst_clone(&rt->dst)); 1040 } 1041 1042 if (msg->msg_flags&MSG_CONFIRM) 1043 goto do_confirm; 1044 back_from_confirm: 1045 1046 saddr = fl4->saddr; 1047 if (!ipc.addr) 1048 daddr = ipc.addr = fl4->daddr; 1049 1050 /* Lockless fast path for the non-corking case. */ 1051 if (!corkreq) { 1052 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1053 sizeof(struct udphdr), &ipc, &rt, 1054 msg->msg_flags); 1055 err = PTR_ERR(skb); 1056 if (!IS_ERR_OR_NULL(skb)) 1057 err = udp_send_skb(skb, fl4); 1058 goto out; 1059 } 1060 1061 lock_sock(sk); 1062 if (unlikely(up->pending)) { 1063 /* The socket is already corked while preparing it. */ 1064 /* ... which is an evident application bug. --ANK */ 1065 release_sock(sk); 1066 1067 net_dbg_ratelimited("cork app bug 2\n"); 1068 err = -EINVAL; 1069 goto out; 1070 } 1071 /* 1072 * Now cork the socket to pend data. 1073 */ 1074 fl4 = &inet->cork.fl.u.ip4; 1075 fl4->daddr = daddr; 1076 fl4->saddr = saddr; 1077 fl4->fl4_dport = dport; 1078 fl4->fl4_sport = inet->inet_sport; 1079 up->pending = AF_INET; 1080 1081 do_append_data: 1082 up->len += ulen; 1083 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1084 sizeof(struct udphdr), &ipc, &rt, 1085 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1086 if (err) 1087 udp_flush_pending_frames(sk); 1088 else if (!corkreq) 1089 err = udp_push_pending_frames(sk); 1090 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1091 up->pending = 0; 1092 release_sock(sk); 1093 1094 out: 1095 ip_rt_put(rt); 1096 if (free) 1097 kfree(ipc.opt); 1098 if (!err) 1099 return len; 1100 /* 1101 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1102 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1103 * we don't have a good statistic (IpOutDiscards but it can be too many 1104 * things). We could add another new stat but at least for now that 1105 * seems like overkill. 1106 */ 1107 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1108 UDP_INC_STATS_USER(sock_net(sk), 1109 UDP_MIB_SNDBUFERRORS, is_udplite); 1110 } 1111 return err; 1112 1113 do_confirm: 1114 dst_confirm(&rt->dst); 1115 if (!(msg->msg_flags&MSG_PROBE) || len) 1116 goto back_from_confirm; 1117 err = 0; 1118 goto out; 1119 } 1120 EXPORT_SYMBOL(udp_sendmsg); 1121 1122 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1123 size_t size, int flags) 1124 { 1125 struct inet_sock *inet = inet_sk(sk); 1126 struct udp_sock *up = udp_sk(sk); 1127 int ret; 1128 1129 if (flags & MSG_SENDPAGE_NOTLAST) 1130 flags |= MSG_MORE; 1131 1132 if (!up->pending) { 1133 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1134 1135 /* Call udp_sendmsg to specify destination address which 1136 * sendpage interface can't pass. 1137 * This will succeed only when the socket is connected. 1138 */ 1139 ret = udp_sendmsg(NULL, sk, &msg, 0); 1140 if (ret < 0) 1141 return ret; 1142 } 1143 1144 lock_sock(sk); 1145 1146 if (unlikely(!up->pending)) { 1147 release_sock(sk); 1148 1149 net_dbg_ratelimited("udp cork app bug 3\n"); 1150 return -EINVAL; 1151 } 1152 1153 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1154 page, offset, size, flags); 1155 if (ret == -EOPNOTSUPP) { 1156 release_sock(sk); 1157 return sock_no_sendpage(sk->sk_socket, page, offset, 1158 size, flags); 1159 } 1160 if (ret < 0) { 1161 udp_flush_pending_frames(sk); 1162 goto out; 1163 } 1164 1165 up->len += size; 1166 if (!(up->corkflag || (flags&MSG_MORE))) 1167 ret = udp_push_pending_frames(sk); 1168 if (!ret) 1169 ret = size; 1170 out: 1171 release_sock(sk); 1172 return ret; 1173 } 1174 1175 1176 /** 1177 * first_packet_length - return length of first packet in receive queue 1178 * @sk: socket 1179 * 1180 * Drops all bad checksum frames, until a valid one is found. 1181 * Returns the length of found skb, or 0 if none is found. 1182 */ 1183 static unsigned int first_packet_length(struct sock *sk) 1184 { 1185 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1186 struct sk_buff *skb; 1187 unsigned int res; 1188 1189 __skb_queue_head_init(&list_kill); 1190 1191 spin_lock_bh(&rcvq->lock); 1192 while ((skb = skb_peek(rcvq)) != NULL && 1193 udp_lib_checksum_complete(skb)) { 1194 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1195 IS_UDPLITE(sk)); 1196 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1197 IS_UDPLITE(sk)); 1198 atomic_inc(&sk->sk_drops); 1199 __skb_unlink(skb, rcvq); 1200 __skb_queue_tail(&list_kill, skb); 1201 } 1202 res = skb ? skb->len : 0; 1203 spin_unlock_bh(&rcvq->lock); 1204 1205 if (!skb_queue_empty(&list_kill)) { 1206 bool slow = lock_sock_fast(sk); 1207 1208 __skb_queue_purge(&list_kill); 1209 sk_mem_reclaim_partial(sk); 1210 unlock_sock_fast(sk, slow); 1211 } 1212 return res; 1213 } 1214 1215 /* 1216 * IOCTL requests applicable to the UDP protocol 1217 */ 1218 1219 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1220 { 1221 switch (cmd) { 1222 case SIOCOUTQ: 1223 { 1224 int amount = sk_wmem_alloc_get(sk); 1225 1226 return put_user(amount, (int __user *)arg); 1227 } 1228 1229 case SIOCINQ: 1230 { 1231 unsigned int amount = first_packet_length(sk); 1232 1233 if (amount) 1234 /* 1235 * We will only return the amount 1236 * of this packet since that is all 1237 * that will be read. 1238 */ 1239 amount -= sizeof(struct udphdr); 1240 1241 return put_user(amount, (int __user *)arg); 1242 } 1243 1244 default: 1245 return -ENOIOCTLCMD; 1246 } 1247 1248 return 0; 1249 } 1250 EXPORT_SYMBOL(udp_ioctl); 1251 1252 /* 1253 * This should be easy, if there is something there we 1254 * return it, otherwise we block. 1255 */ 1256 1257 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1258 size_t len, int noblock, int flags, int *addr_len) 1259 { 1260 struct inet_sock *inet = inet_sk(sk); 1261 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1262 struct sk_buff *skb; 1263 unsigned int ulen, copied; 1264 int peeked, off = 0; 1265 int err; 1266 int is_udplite = IS_UDPLITE(sk); 1267 bool slow; 1268 1269 if (flags & MSG_ERRQUEUE) 1270 return ip_recv_error(sk, msg, len, addr_len); 1271 1272 try_again: 1273 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1274 &peeked, &off, &err); 1275 if (!skb) 1276 goto out; 1277 1278 ulen = skb->len - sizeof(struct udphdr); 1279 copied = len; 1280 if (copied > ulen) 1281 copied = ulen; 1282 else if (copied < ulen) 1283 msg->msg_flags |= MSG_TRUNC; 1284 1285 /* 1286 * If checksum is needed at all, try to do it while copying the 1287 * data. If the data is truncated, or if we only want a partial 1288 * coverage checksum (UDP-Lite), do it before the copy. 1289 */ 1290 1291 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1292 if (udp_lib_checksum_complete(skb)) 1293 goto csum_copy_err; 1294 } 1295 1296 if (skb_csum_unnecessary(skb)) 1297 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 1298 msg, copied); 1299 else { 1300 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), 1301 msg); 1302 1303 if (err == -EINVAL) 1304 goto csum_copy_err; 1305 } 1306 1307 if (unlikely(err)) { 1308 trace_kfree_skb(skb, udp_recvmsg); 1309 if (!peeked) { 1310 atomic_inc(&sk->sk_drops); 1311 UDP_INC_STATS_USER(sock_net(sk), 1312 UDP_MIB_INERRORS, is_udplite); 1313 } 1314 goto out_free; 1315 } 1316 1317 if (!peeked) 1318 UDP_INC_STATS_USER(sock_net(sk), 1319 UDP_MIB_INDATAGRAMS, is_udplite); 1320 1321 sock_recv_ts_and_drops(msg, sk, skb); 1322 1323 /* Copy the address. */ 1324 if (sin) { 1325 sin->sin_family = AF_INET; 1326 sin->sin_port = udp_hdr(skb)->source; 1327 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1328 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1329 *addr_len = sizeof(*sin); 1330 } 1331 if (inet->cmsg_flags) 1332 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr)); 1333 1334 err = copied; 1335 if (flags & MSG_TRUNC) 1336 err = ulen; 1337 1338 out_free: 1339 skb_free_datagram_locked(sk, skb); 1340 out: 1341 return err; 1342 1343 csum_copy_err: 1344 slow = lock_sock_fast(sk); 1345 if (!skb_kill_datagram(sk, skb, flags)) { 1346 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1347 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1348 } 1349 unlock_sock_fast(sk, slow); 1350 1351 if (noblock) 1352 return -EAGAIN; 1353 1354 /* starting over for a new packet */ 1355 msg->msg_flags &= ~MSG_TRUNC; 1356 goto try_again; 1357 } 1358 1359 1360 int udp_disconnect(struct sock *sk, int flags) 1361 { 1362 struct inet_sock *inet = inet_sk(sk); 1363 /* 1364 * 1003.1g - break association. 1365 */ 1366 1367 sk->sk_state = TCP_CLOSE; 1368 inet->inet_daddr = 0; 1369 inet->inet_dport = 0; 1370 sock_rps_reset_rxhash(sk); 1371 sk->sk_bound_dev_if = 0; 1372 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1373 inet_reset_saddr(sk); 1374 1375 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1376 sk->sk_prot->unhash(sk); 1377 inet->inet_sport = 0; 1378 } 1379 sk_dst_reset(sk); 1380 return 0; 1381 } 1382 EXPORT_SYMBOL(udp_disconnect); 1383 1384 void udp_lib_unhash(struct sock *sk) 1385 { 1386 if (sk_hashed(sk)) { 1387 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1388 struct udp_hslot *hslot, *hslot2; 1389 1390 hslot = udp_hashslot(udptable, sock_net(sk), 1391 udp_sk(sk)->udp_port_hash); 1392 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1393 1394 spin_lock_bh(&hslot->lock); 1395 if (sk_nulls_del_node_init_rcu(sk)) { 1396 hslot->count--; 1397 inet_sk(sk)->inet_num = 0; 1398 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1399 1400 spin_lock(&hslot2->lock); 1401 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1402 hslot2->count--; 1403 spin_unlock(&hslot2->lock); 1404 } 1405 spin_unlock_bh(&hslot->lock); 1406 } 1407 } 1408 EXPORT_SYMBOL(udp_lib_unhash); 1409 1410 /* 1411 * inet_rcv_saddr was changed, we must rehash secondary hash 1412 */ 1413 void udp_lib_rehash(struct sock *sk, u16 newhash) 1414 { 1415 if (sk_hashed(sk)) { 1416 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1417 struct udp_hslot *hslot, *hslot2, *nhslot2; 1418 1419 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1420 nhslot2 = udp_hashslot2(udptable, newhash); 1421 udp_sk(sk)->udp_portaddr_hash = newhash; 1422 if (hslot2 != nhslot2) { 1423 hslot = udp_hashslot(udptable, sock_net(sk), 1424 udp_sk(sk)->udp_port_hash); 1425 /* we must lock primary chain too */ 1426 spin_lock_bh(&hslot->lock); 1427 1428 spin_lock(&hslot2->lock); 1429 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1430 hslot2->count--; 1431 spin_unlock(&hslot2->lock); 1432 1433 spin_lock(&nhslot2->lock); 1434 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1435 &nhslot2->head); 1436 nhslot2->count++; 1437 spin_unlock(&nhslot2->lock); 1438 1439 spin_unlock_bh(&hslot->lock); 1440 } 1441 } 1442 } 1443 EXPORT_SYMBOL(udp_lib_rehash); 1444 1445 static void udp_v4_rehash(struct sock *sk) 1446 { 1447 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1448 inet_sk(sk)->inet_rcv_saddr, 1449 inet_sk(sk)->inet_num); 1450 udp_lib_rehash(sk, new_hash); 1451 } 1452 1453 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1454 { 1455 int rc; 1456 1457 if (inet_sk(sk)->inet_daddr) { 1458 sock_rps_save_rxhash(sk, skb); 1459 sk_mark_napi_id(sk, skb); 1460 sk_incoming_cpu_update(sk); 1461 } 1462 1463 rc = sock_queue_rcv_skb(sk, skb); 1464 if (rc < 0) { 1465 int is_udplite = IS_UDPLITE(sk); 1466 1467 /* Note that an ENOMEM error is charged twice */ 1468 if (rc == -ENOMEM) 1469 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1470 is_udplite); 1471 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1472 kfree_skb(skb); 1473 trace_udp_fail_queue_rcv_skb(rc, sk); 1474 return -1; 1475 } 1476 1477 return 0; 1478 1479 } 1480 1481 static struct static_key udp_encap_needed __read_mostly; 1482 void udp_encap_enable(void) 1483 { 1484 if (!static_key_enabled(&udp_encap_needed)) 1485 static_key_slow_inc(&udp_encap_needed); 1486 } 1487 EXPORT_SYMBOL(udp_encap_enable); 1488 1489 /* returns: 1490 * -1: error 1491 * 0: success 1492 * >0: "udp encap" protocol resubmission 1493 * 1494 * Note that in the success and error cases, the skb is assumed to 1495 * have either been requeued or freed. 1496 */ 1497 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1498 { 1499 struct udp_sock *up = udp_sk(sk); 1500 int rc; 1501 int is_udplite = IS_UDPLITE(sk); 1502 1503 /* 1504 * Charge it to the socket, dropping if the queue is full. 1505 */ 1506 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1507 goto drop; 1508 nf_reset(skb); 1509 1510 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1511 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1512 1513 /* 1514 * This is an encapsulation socket so pass the skb to 1515 * the socket's udp_encap_rcv() hook. Otherwise, just 1516 * fall through and pass this up the UDP socket. 1517 * up->encap_rcv() returns the following value: 1518 * =0 if skb was successfully passed to the encap 1519 * handler or was discarded by it. 1520 * >0 if skb should be passed on to UDP. 1521 * <0 if skb should be resubmitted as proto -N 1522 */ 1523 1524 /* if we're overly short, let UDP handle it */ 1525 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1526 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { 1527 int ret; 1528 1529 /* Verify checksum before giving to encap */ 1530 if (udp_lib_checksum_complete(skb)) 1531 goto csum_error; 1532 1533 ret = encap_rcv(sk, skb); 1534 if (ret <= 0) { 1535 UDP_INC_STATS_BH(sock_net(sk), 1536 UDP_MIB_INDATAGRAMS, 1537 is_udplite); 1538 return -ret; 1539 } 1540 } 1541 1542 /* FALLTHROUGH -- it's a UDP Packet */ 1543 } 1544 1545 /* 1546 * UDP-Lite specific tests, ignored on UDP sockets 1547 */ 1548 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1549 1550 /* 1551 * MIB statistics other than incrementing the error count are 1552 * disabled for the following two types of errors: these depend 1553 * on the application settings, not on the functioning of the 1554 * protocol stack as such. 1555 * 1556 * RFC 3828 here recommends (sec 3.3): "There should also be a 1557 * way ... to ... at least let the receiving application block 1558 * delivery of packets with coverage values less than a value 1559 * provided by the application." 1560 */ 1561 if (up->pcrlen == 0) { /* full coverage was set */ 1562 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1563 UDP_SKB_CB(skb)->cscov, skb->len); 1564 goto drop; 1565 } 1566 /* The next case involves violating the min. coverage requested 1567 * by the receiver. This is subtle: if receiver wants x and x is 1568 * greater than the buffersize/MTU then receiver will complain 1569 * that it wants x while sender emits packets of smaller size y. 1570 * Therefore the above ...()->partial_cov statement is essential. 1571 */ 1572 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1573 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1574 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1575 goto drop; 1576 } 1577 } 1578 1579 if (rcu_access_pointer(sk->sk_filter) && 1580 udp_lib_checksum_complete(skb)) 1581 goto csum_error; 1582 1583 1584 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1585 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1586 is_udplite); 1587 goto drop; 1588 } 1589 1590 rc = 0; 1591 1592 ipv4_pktinfo_prepare(sk, skb); 1593 bh_lock_sock(sk); 1594 if (!sock_owned_by_user(sk)) 1595 rc = __udp_queue_rcv_skb(sk, skb); 1596 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1597 bh_unlock_sock(sk); 1598 goto drop; 1599 } 1600 bh_unlock_sock(sk); 1601 1602 return rc; 1603 1604 csum_error: 1605 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1606 drop: 1607 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1608 atomic_inc(&sk->sk_drops); 1609 kfree_skb(skb); 1610 return -1; 1611 } 1612 1613 1614 static void flush_stack(struct sock **stack, unsigned int count, 1615 struct sk_buff *skb, unsigned int final) 1616 { 1617 unsigned int i; 1618 struct sk_buff *skb1 = NULL; 1619 struct sock *sk; 1620 1621 for (i = 0; i < count; i++) { 1622 sk = stack[i]; 1623 if (likely(skb1 == NULL)) 1624 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1625 1626 if (!skb1) { 1627 atomic_inc(&sk->sk_drops); 1628 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1629 IS_UDPLITE(sk)); 1630 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1631 IS_UDPLITE(sk)); 1632 } 1633 1634 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1635 skb1 = NULL; 1636 1637 sock_put(sk); 1638 } 1639 if (unlikely(skb1)) 1640 kfree_skb(skb1); 1641 } 1642 1643 /* For TCP sockets, sk_rx_dst is protected by socket lock 1644 * For UDP, we use xchg() to guard against concurrent changes. 1645 */ 1646 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1647 { 1648 struct dst_entry *old; 1649 1650 dst_hold(dst); 1651 old = xchg(&sk->sk_rx_dst, dst); 1652 dst_release(old); 1653 } 1654 1655 /* 1656 * Multicasts and broadcasts go to each listener. 1657 * 1658 * Note: called only from the BH handler context. 1659 */ 1660 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1661 struct udphdr *uh, 1662 __be32 saddr, __be32 daddr, 1663 struct udp_table *udptable, 1664 int proto) 1665 { 1666 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1667 struct hlist_nulls_node *node; 1668 unsigned short hnum = ntohs(uh->dest); 1669 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1670 int dif = skb->dev->ifindex; 1671 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 1672 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1673 bool inner_flushed = false; 1674 1675 if (use_hash2) { 1676 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1677 udp_table.mask; 1678 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask; 1679 start_lookup: 1680 hslot = &udp_table.hash2[hash2]; 1681 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1682 } 1683 1684 spin_lock(&hslot->lock); 1685 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 1686 if (__udp_is_mcast_sock(net, sk, 1687 uh->dest, daddr, 1688 uh->source, saddr, 1689 dif, hnum)) { 1690 if (unlikely(count == ARRAY_SIZE(stack))) { 1691 flush_stack(stack, count, skb, ~0); 1692 inner_flushed = true; 1693 count = 0; 1694 } 1695 stack[count++] = sk; 1696 sock_hold(sk); 1697 } 1698 } 1699 1700 spin_unlock(&hslot->lock); 1701 1702 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1703 if (use_hash2 && hash2 != hash2_any) { 1704 hash2 = hash2_any; 1705 goto start_lookup; 1706 } 1707 1708 /* 1709 * do the slow work with no lock held 1710 */ 1711 if (count) { 1712 flush_stack(stack, count, skb, count - 1); 1713 } else { 1714 if (!inner_flushed) 1715 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 1716 proto == IPPROTO_UDPLITE); 1717 consume_skb(skb); 1718 } 1719 return 0; 1720 } 1721 1722 /* Initialize UDP checksum. If exited with zero value (success), 1723 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1724 * Otherwise, csum completion requires chacksumming packet body, 1725 * including udp header and folding it to skb->csum. 1726 */ 1727 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1728 int proto) 1729 { 1730 int err; 1731 1732 UDP_SKB_CB(skb)->partial_cov = 0; 1733 UDP_SKB_CB(skb)->cscov = skb->len; 1734 1735 if (proto == IPPROTO_UDPLITE) { 1736 err = udplite_checksum_init(skb, uh); 1737 if (err) 1738 return err; 1739 } 1740 1741 return skb_checksum_init_zero_check(skb, proto, uh->check, 1742 inet_compute_pseudo); 1743 } 1744 1745 /* 1746 * All we need to do is get the socket, and then do a checksum. 1747 */ 1748 1749 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1750 int proto) 1751 { 1752 struct sock *sk; 1753 struct udphdr *uh; 1754 unsigned short ulen; 1755 struct rtable *rt = skb_rtable(skb); 1756 __be32 saddr, daddr; 1757 struct net *net = dev_net(skb->dev); 1758 1759 /* 1760 * Validate the packet. 1761 */ 1762 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1763 goto drop; /* No space for header. */ 1764 1765 uh = udp_hdr(skb); 1766 ulen = ntohs(uh->len); 1767 saddr = ip_hdr(skb)->saddr; 1768 daddr = ip_hdr(skb)->daddr; 1769 1770 if (ulen > skb->len) 1771 goto short_packet; 1772 1773 if (proto == IPPROTO_UDP) { 1774 /* UDP validates ulen. */ 1775 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1776 goto short_packet; 1777 uh = udp_hdr(skb); 1778 } 1779 1780 if (udp4_csum_init(skb, uh, proto)) 1781 goto csum_error; 1782 1783 sk = skb_steal_sock(skb); 1784 if (sk) { 1785 struct dst_entry *dst = skb_dst(skb); 1786 int ret; 1787 1788 if (unlikely(sk->sk_rx_dst != dst)) 1789 udp_sk_rx_dst_set(sk, dst); 1790 1791 ret = udp_queue_rcv_skb(sk, skb); 1792 sock_put(sk); 1793 /* a return value > 0 means to resubmit the input, but 1794 * it wants the return to be -protocol, or 0 1795 */ 1796 if (ret > 0) 1797 return -ret; 1798 return 0; 1799 } 1800 1801 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1802 return __udp4_lib_mcast_deliver(net, skb, uh, 1803 saddr, daddr, udptable, proto); 1804 1805 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1806 if (sk != NULL) { 1807 int ret; 1808 1809 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1810 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1811 inet_compute_pseudo); 1812 1813 ret = udp_queue_rcv_skb(sk, skb); 1814 sock_put(sk); 1815 1816 /* a return value > 0 means to resubmit the input, but 1817 * it wants the return to be -protocol, or 0 1818 */ 1819 if (ret > 0) 1820 return -ret; 1821 return 0; 1822 } 1823 1824 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1825 goto drop; 1826 nf_reset(skb); 1827 1828 /* No socket. Drop packet silently, if checksum is wrong */ 1829 if (udp_lib_checksum_complete(skb)) 1830 goto csum_error; 1831 1832 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1833 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1834 1835 /* 1836 * Hmm. We got an UDP packet to a port to which we 1837 * don't wanna listen. Ignore it. 1838 */ 1839 kfree_skb(skb); 1840 return 0; 1841 1842 short_packet: 1843 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1844 proto == IPPROTO_UDPLITE ? "Lite" : "", 1845 &saddr, ntohs(uh->source), 1846 ulen, skb->len, 1847 &daddr, ntohs(uh->dest)); 1848 goto drop; 1849 1850 csum_error: 1851 /* 1852 * RFC1122: OK. Discards the bad packet silently (as far as 1853 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1854 */ 1855 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1856 proto == IPPROTO_UDPLITE ? "Lite" : "", 1857 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1858 ulen); 1859 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1860 drop: 1861 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1862 kfree_skb(skb); 1863 return 0; 1864 } 1865 1866 /* We can only early demux multicast if there is a single matching socket. 1867 * If more than one socket found returns NULL 1868 */ 1869 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1870 __be16 loc_port, __be32 loc_addr, 1871 __be16 rmt_port, __be32 rmt_addr, 1872 int dif) 1873 { 1874 struct sock *sk, *result; 1875 struct hlist_nulls_node *node; 1876 unsigned short hnum = ntohs(loc_port); 1877 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1878 struct udp_hslot *hslot = &udp_table.hash[slot]; 1879 1880 /* Do not bother scanning a too big list */ 1881 if (hslot->count > 10) 1882 return NULL; 1883 1884 rcu_read_lock(); 1885 begin: 1886 count = 0; 1887 result = NULL; 1888 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1889 if (__udp_is_mcast_sock(net, sk, 1890 loc_port, loc_addr, 1891 rmt_port, rmt_addr, 1892 dif, hnum)) { 1893 result = sk; 1894 ++count; 1895 } 1896 } 1897 /* 1898 * if the nulls value we got at the end of this lookup is 1899 * not the expected one, we must restart lookup. 1900 * We probably met an item that was moved to another chain. 1901 */ 1902 if (get_nulls_value(node) != slot) 1903 goto begin; 1904 1905 if (result) { 1906 if (count != 1 || 1907 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1908 result = NULL; 1909 else if (unlikely(!__udp_is_mcast_sock(net, result, 1910 loc_port, loc_addr, 1911 rmt_port, rmt_addr, 1912 dif, hnum))) { 1913 sock_put(result); 1914 result = NULL; 1915 } 1916 } 1917 rcu_read_unlock(); 1918 return result; 1919 } 1920 1921 /* For unicast we should only early demux connected sockets or we can 1922 * break forwarding setups. The chains here can be long so only check 1923 * if the first socket is an exact match and if not move on. 1924 */ 1925 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1926 __be16 loc_port, __be32 loc_addr, 1927 __be16 rmt_port, __be32 rmt_addr, 1928 int dif) 1929 { 1930 struct sock *sk, *result; 1931 struct hlist_nulls_node *node; 1932 unsigned short hnum = ntohs(loc_port); 1933 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1934 unsigned int slot2 = hash2 & udp_table.mask; 1935 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1936 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 1937 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1938 1939 rcu_read_lock(); 1940 result = NULL; 1941 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 1942 if (INET_MATCH(sk, net, acookie, 1943 rmt_addr, loc_addr, ports, dif)) 1944 result = sk; 1945 /* Only check first socket in chain */ 1946 break; 1947 } 1948 1949 if (result) { 1950 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1951 result = NULL; 1952 else if (unlikely(!INET_MATCH(sk, net, acookie, 1953 rmt_addr, loc_addr, 1954 ports, dif))) { 1955 sock_put(result); 1956 result = NULL; 1957 } 1958 } 1959 rcu_read_unlock(); 1960 return result; 1961 } 1962 1963 void udp_v4_early_demux(struct sk_buff *skb) 1964 { 1965 struct net *net = dev_net(skb->dev); 1966 const struct iphdr *iph; 1967 const struct udphdr *uh; 1968 struct sock *sk; 1969 struct dst_entry *dst; 1970 int dif = skb->dev->ifindex; 1971 1972 /* validate the packet */ 1973 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1974 return; 1975 1976 iph = ip_hdr(skb); 1977 uh = udp_hdr(skb); 1978 1979 if (skb->pkt_type == PACKET_BROADCAST || 1980 skb->pkt_type == PACKET_MULTICAST) 1981 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1982 uh->source, iph->saddr, dif); 1983 else if (skb->pkt_type == PACKET_HOST) 1984 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1985 uh->source, iph->saddr, dif); 1986 else 1987 return; 1988 1989 if (!sk) 1990 return; 1991 1992 skb->sk = sk; 1993 skb->destructor = sock_efree; 1994 dst = sk->sk_rx_dst; 1995 1996 if (dst) 1997 dst = dst_check(dst, 0); 1998 if (dst) 1999 skb_dst_set_noref(skb, dst); 2000 } 2001 2002 int udp_rcv(struct sk_buff *skb) 2003 { 2004 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2005 } 2006 2007 void udp_destroy_sock(struct sock *sk) 2008 { 2009 struct udp_sock *up = udp_sk(sk); 2010 bool slow = lock_sock_fast(sk); 2011 udp_flush_pending_frames(sk); 2012 unlock_sock_fast(sk, slow); 2013 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2014 void (*encap_destroy)(struct sock *sk); 2015 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2016 if (encap_destroy) 2017 encap_destroy(sk); 2018 } 2019 } 2020 2021 /* 2022 * Socket option code for UDP 2023 */ 2024 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2025 char __user *optval, unsigned int optlen, 2026 int (*push_pending_frames)(struct sock *)) 2027 { 2028 struct udp_sock *up = udp_sk(sk); 2029 int val, valbool; 2030 int err = 0; 2031 int is_udplite = IS_UDPLITE(sk); 2032 2033 if (optlen < sizeof(int)) 2034 return -EINVAL; 2035 2036 if (get_user(val, (int __user *)optval)) 2037 return -EFAULT; 2038 2039 valbool = val ? 1 : 0; 2040 2041 switch (optname) { 2042 case UDP_CORK: 2043 if (val != 0) { 2044 up->corkflag = 1; 2045 } else { 2046 up->corkflag = 0; 2047 lock_sock(sk); 2048 push_pending_frames(sk); 2049 release_sock(sk); 2050 } 2051 break; 2052 2053 case UDP_ENCAP: 2054 switch (val) { 2055 case 0: 2056 case UDP_ENCAP_ESPINUDP: 2057 case UDP_ENCAP_ESPINUDP_NON_IKE: 2058 up->encap_rcv = xfrm4_udp_encap_rcv; 2059 /* FALLTHROUGH */ 2060 case UDP_ENCAP_L2TPINUDP: 2061 up->encap_type = val; 2062 udp_encap_enable(); 2063 break; 2064 default: 2065 err = -ENOPROTOOPT; 2066 break; 2067 } 2068 break; 2069 2070 case UDP_NO_CHECK6_TX: 2071 up->no_check6_tx = valbool; 2072 break; 2073 2074 case UDP_NO_CHECK6_RX: 2075 up->no_check6_rx = valbool; 2076 break; 2077 2078 /* 2079 * UDP-Lite's partial checksum coverage (RFC 3828). 2080 */ 2081 /* The sender sets actual checksum coverage length via this option. 2082 * The case coverage > packet length is handled by send module. */ 2083 case UDPLITE_SEND_CSCOV: 2084 if (!is_udplite) /* Disable the option on UDP sockets */ 2085 return -ENOPROTOOPT; 2086 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2087 val = 8; 2088 else if (val > USHRT_MAX) 2089 val = USHRT_MAX; 2090 up->pcslen = val; 2091 up->pcflag |= UDPLITE_SEND_CC; 2092 break; 2093 2094 /* The receiver specifies a minimum checksum coverage value. To make 2095 * sense, this should be set to at least 8 (as done below). If zero is 2096 * used, this again means full checksum coverage. */ 2097 case UDPLITE_RECV_CSCOV: 2098 if (!is_udplite) /* Disable the option on UDP sockets */ 2099 return -ENOPROTOOPT; 2100 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2101 val = 8; 2102 else if (val > USHRT_MAX) 2103 val = USHRT_MAX; 2104 up->pcrlen = val; 2105 up->pcflag |= UDPLITE_RECV_CC; 2106 break; 2107 2108 default: 2109 err = -ENOPROTOOPT; 2110 break; 2111 } 2112 2113 return err; 2114 } 2115 EXPORT_SYMBOL(udp_lib_setsockopt); 2116 2117 int udp_setsockopt(struct sock *sk, int level, int optname, 2118 char __user *optval, unsigned int optlen) 2119 { 2120 if (level == SOL_UDP || level == SOL_UDPLITE) 2121 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2122 udp_push_pending_frames); 2123 return ip_setsockopt(sk, level, optname, optval, optlen); 2124 } 2125 2126 #ifdef CONFIG_COMPAT 2127 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2128 char __user *optval, unsigned int optlen) 2129 { 2130 if (level == SOL_UDP || level == SOL_UDPLITE) 2131 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2132 udp_push_pending_frames); 2133 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2134 } 2135 #endif 2136 2137 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2138 char __user *optval, int __user *optlen) 2139 { 2140 struct udp_sock *up = udp_sk(sk); 2141 int val, len; 2142 2143 if (get_user(len, optlen)) 2144 return -EFAULT; 2145 2146 len = min_t(unsigned int, len, sizeof(int)); 2147 2148 if (len < 0) 2149 return -EINVAL; 2150 2151 switch (optname) { 2152 case UDP_CORK: 2153 val = up->corkflag; 2154 break; 2155 2156 case UDP_ENCAP: 2157 val = up->encap_type; 2158 break; 2159 2160 case UDP_NO_CHECK6_TX: 2161 val = up->no_check6_tx; 2162 break; 2163 2164 case UDP_NO_CHECK6_RX: 2165 val = up->no_check6_rx; 2166 break; 2167 2168 /* The following two cannot be changed on UDP sockets, the return is 2169 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2170 case UDPLITE_SEND_CSCOV: 2171 val = up->pcslen; 2172 break; 2173 2174 case UDPLITE_RECV_CSCOV: 2175 val = up->pcrlen; 2176 break; 2177 2178 default: 2179 return -ENOPROTOOPT; 2180 } 2181 2182 if (put_user(len, optlen)) 2183 return -EFAULT; 2184 if (copy_to_user(optval, &val, len)) 2185 return -EFAULT; 2186 return 0; 2187 } 2188 EXPORT_SYMBOL(udp_lib_getsockopt); 2189 2190 int udp_getsockopt(struct sock *sk, int level, int optname, 2191 char __user *optval, int __user *optlen) 2192 { 2193 if (level == SOL_UDP || level == SOL_UDPLITE) 2194 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2195 return ip_getsockopt(sk, level, optname, optval, optlen); 2196 } 2197 2198 #ifdef CONFIG_COMPAT 2199 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2200 char __user *optval, int __user *optlen) 2201 { 2202 if (level == SOL_UDP || level == SOL_UDPLITE) 2203 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2204 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2205 } 2206 #endif 2207 /** 2208 * udp_poll - wait for a UDP event. 2209 * @file - file struct 2210 * @sock - socket 2211 * @wait - poll table 2212 * 2213 * This is same as datagram poll, except for the special case of 2214 * blocking sockets. If application is using a blocking fd 2215 * and a packet with checksum error is in the queue; 2216 * then it could get return from select indicating data available 2217 * but then block when reading it. Add special case code 2218 * to work around these arguably broken applications. 2219 */ 2220 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2221 { 2222 unsigned int mask = datagram_poll(file, sock, wait); 2223 struct sock *sk = sock->sk; 2224 2225 sock_rps_record_flow(sk); 2226 2227 /* Check for false positives due to checksum errors */ 2228 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2229 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2230 mask &= ~(POLLIN | POLLRDNORM); 2231 2232 return mask; 2233 2234 } 2235 EXPORT_SYMBOL(udp_poll); 2236 2237 struct proto udp_prot = { 2238 .name = "UDP", 2239 .owner = THIS_MODULE, 2240 .close = udp_lib_close, 2241 .connect = ip4_datagram_connect, 2242 .disconnect = udp_disconnect, 2243 .ioctl = udp_ioctl, 2244 .destroy = udp_destroy_sock, 2245 .setsockopt = udp_setsockopt, 2246 .getsockopt = udp_getsockopt, 2247 .sendmsg = udp_sendmsg, 2248 .recvmsg = udp_recvmsg, 2249 .sendpage = udp_sendpage, 2250 .backlog_rcv = __udp_queue_rcv_skb, 2251 .release_cb = ip4_datagram_release_cb, 2252 .hash = udp_lib_hash, 2253 .unhash = udp_lib_unhash, 2254 .rehash = udp_v4_rehash, 2255 .get_port = udp_v4_get_port, 2256 .memory_allocated = &udp_memory_allocated, 2257 .sysctl_mem = sysctl_udp_mem, 2258 .sysctl_wmem = &sysctl_udp_wmem_min, 2259 .sysctl_rmem = &sysctl_udp_rmem_min, 2260 .obj_size = sizeof(struct udp_sock), 2261 .slab_flags = SLAB_DESTROY_BY_RCU, 2262 .h.udp_table = &udp_table, 2263 #ifdef CONFIG_COMPAT 2264 .compat_setsockopt = compat_udp_setsockopt, 2265 .compat_getsockopt = compat_udp_getsockopt, 2266 #endif 2267 .clear_sk = sk_prot_clear_portaddr_nulls, 2268 }; 2269 EXPORT_SYMBOL(udp_prot); 2270 2271 /* ------------------------------------------------------------------------ */ 2272 #ifdef CONFIG_PROC_FS 2273 2274 static struct sock *udp_get_first(struct seq_file *seq, int start) 2275 { 2276 struct sock *sk; 2277 struct udp_iter_state *state = seq->private; 2278 struct net *net = seq_file_net(seq); 2279 2280 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2281 ++state->bucket) { 2282 struct hlist_nulls_node *node; 2283 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2284 2285 if (hlist_nulls_empty(&hslot->head)) 2286 continue; 2287 2288 spin_lock_bh(&hslot->lock); 2289 sk_nulls_for_each(sk, node, &hslot->head) { 2290 if (!net_eq(sock_net(sk), net)) 2291 continue; 2292 if (sk->sk_family == state->family) 2293 goto found; 2294 } 2295 spin_unlock_bh(&hslot->lock); 2296 } 2297 sk = NULL; 2298 found: 2299 return sk; 2300 } 2301 2302 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2303 { 2304 struct udp_iter_state *state = seq->private; 2305 struct net *net = seq_file_net(seq); 2306 2307 do { 2308 sk = sk_nulls_next(sk); 2309 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2310 2311 if (!sk) { 2312 if (state->bucket <= state->udp_table->mask) 2313 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2314 return udp_get_first(seq, state->bucket + 1); 2315 } 2316 return sk; 2317 } 2318 2319 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2320 { 2321 struct sock *sk = udp_get_first(seq, 0); 2322 2323 if (sk) 2324 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2325 --pos; 2326 return pos ? NULL : sk; 2327 } 2328 2329 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2330 { 2331 struct udp_iter_state *state = seq->private; 2332 state->bucket = MAX_UDP_PORTS; 2333 2334 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2335 } 2336 2337 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2338 { 2339 struct sock *sk; 2340 2341 if (v == SEQ_START_TOKEN) 2342 sk = udp_get_idx(seq, 0); 2343 else 2344 sk = udp_get_next(seq, v); 2345 2346 ++*pos; 2347 return sk; 2348 } 2349 2350 static void udp_seq_stop(struct seq_file *seq, void *v) 2351 { 2352 struct udp_iter_state *state = seq->private; 2353 2354 if (state->bucket <= state->udp_table->mask) 2355 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2356 } 2357 2358 int udp_seq_open(struct inode *inode, struct file *file) 2359 { 2360 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2361 struct udp_iter_state *s; 2362 int err; 2363 2364 err = seq_open_net(inode, file, &afinfo->seq_ops, 2365 sizeof(struct udp_iter_state)); 2366 if (err < 0) 2367 return err; 2368 2369 s = ((struct seq_file *)file->private_data)->private; 2370 s->family = afinfo->family; 2371 s->udp_table = afinfo->udp_table; 2372 return err; 2373 } 2374 EXPORT_SYMBOL(udp_seq_open); 2375 2376 /* ------------------------------------------------------------------------ */ 2377 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2378 { 2379 struct proc_dir_entry *p; 2380 int rc = 0; 2381 2382 afinfo->seq_ops.start = udp_seq_start; 2383 afinfo->seq_ops.next = udp_seq_next; 2384 afinfo->seq_ops.stop = udp_seq_stop; 2385 2386 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2387 afinfo->seq_fops, afinfo); 2388 if (!p) 2389 rc = -ENOMEM; 2390 return rc; 2391 } 2392 EXPORT_SYMBOL(udp_proc_register); 2393 2394 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2395 { 2396 remove_proc_entry(afinfo->name, net->proc_net); 2397 } 2398 EXPORT_SYMBOL(udp_proc_unregister); 2399 2400 /* ------------------------------------------------------------------------ */ 2401 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2402 int bucket) 2403 { 2404 struct inet_sock *inet = inet_sk(sp); 2405 __be32 dest = inet->inet_daddr; 2406 __be32 src = inet->inet_rcv_saddr; 2407 __u16 destp = ntohs(inet->inet_dport); 2408 __u16 srcp = ntohs(inet->inet_sport); 2409 2410 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2411 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2412 bucket, src, srcp, dest, destp, sp->sk_state, 2413 sk_wmem_alloc_get(sp), 2414 sk_rmem_alloc_get(sp), 2415 0, 0L, 0, 2416 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2417 0, sock_i_ino(sp), 2418 atomic_read(&sp->sk_refcnt), sp, 2419 atomic_read(&sp->sk_drops)); 2420 } 2421 2422 int udp4_seq_show(struct seq_file *seq, void *v) 2423 { 2424 seq_setwidth(seq, 127); 2425 if (v == SEQ_START_TOKEN) 2426 seq_puts(seq, " sl local_address rem_address st tx_queue " 2427 "rx_queue tr tm->when retrnsmt uid timeout " 2428 "inode ref pointer drops"); 2429 else { 2430 struct udp_iter_state *state = seq->private; 2431 2432 udp4_format_sock(v, seq, state->bucket); 2433 } 2434 seq_pad(seq, '\n'); 2435 return 0; 2436 } 2437 2438 static const struct file_operations udp_afinfo_seq_fops = { 2439 .owner = THIS_MODULE, 2440 .open = udp_seq_open, 2441 .read = seq_read, 2442 .llseek = seq_lseek, 2443 .release = seq_release_net 2444 }; 2445 2446 /* ------------------------------------------------------------------------ */ 2447 static struct udp_seq_afinfo udp4_seq_afinfo = { 2448 .name = "udp", 2449 .family = AF_INET, 2450 .udp_table = &udp_table, 2451 .seq_fops = &udp_afinfo_seq_fops, 2452 .seq_ops = { 2453 .show = udp4_seq_show, 2454 }, 2455 }; 2456 2457 static int __net_init udp4_proc_init_net(struct net *net) 2458 { 2459 return udp_proc_register(net, &udp4_seq_afinfo); 2460 } 2461 2462 static void __net_exit udp4_proc_exit_net(struct net *net) 2463 { 2464 udp_proc_unregister(net, &udp4_seq_afinfo); 2465 } 2466 2467 static struct pernet_operations udp4_net_ops = { 2468 .init = udp4_proc_init_net, 2469 .exit = udp4_proc_exit_net, 2470 }; 2471 2472 int __init udp4_proc_init(void) 2473 { 2474 return register_pernet_subsys(&udp4_net_ops); 2475 } 2476 2477 void udp4_proc_exit(void) 2478 { 2479 unregister_pernet_subsys(&udp4_net_ops); 2480 } 2481 #endif /* CONFIG_PROC_FS */ 2482 2483 static __initdata unsigned long uhash_entries; 2484 static int __init set_uhash_entries(char *str) 2485 { 2486 ssize_t ret; 2487 2488 if (!str) 2489 return 0; 2490 2491 ret = kstrtoul(str, 0, &uhash_entries); 2492 if (ret) 2493 return 0; 2494 2495 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2496 uhash_entries = UDP_HTABLE_SIZE_MIN; 2497 return 1; 2498 } 2499 __setup("uhash_entries=", set_uhash_entries); 2500 2501 void __init udp_table_init(struct udp_table *table, const char *name) 2502 { 2503 unsigned int i; 2504 2505 table->hash = alloc_large_system_hash(name, 2506 2 * sizeof(struct udp_hslot), 2507 uhash_entries, 2508 21, /* one slot per 2 MB */ 2509 0, 2510 &table->log, 2511 &table->mask, 2512 UDP_HTABLE_SIZE_MIN, 2513 64 * 1024); 2514 2515 table->hash2 = table->hash + (table->mask + 1); 2516 for (i = 0; i <= table->mask; i++) { 2517 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2518 table->hash[i].count = 0; 2519 spin_lock_init(&table->hash[i].lock); 2520 } 2521 for (i = 0; i <= table->mask; i++) { 2522 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2523 table->hash2[i].count = 0; 2524 spin_lock_init(&table->hash2[i].lock); 2525 } 2526 } 2527 2528 void __init udp_init(void) 2529 { 2530 unsigned long limit; 2531 2532 udp_table_init(&udp_table, "UDP"); 2533 limit = nr_free_buffer_pages() / 8; 2534 limit = max(limit, 128UL); 2535 sysctl_udp_mem[0] = limit / 4 * 3; 2536 sysctl_udp_mem[1] = limit; 2537 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2538 2539 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2540 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2541 } 2542