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/proc_fs.h> 103 #include <linux/seq_file.h> 104 #include <net/net_namespace.h> 105 #include <net/icmp.h> 106 #include <net/inet_hashtables.h> 107 #include <net/route.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <trace/events/udp.h> 111 #include <linux/static_key.h> 112 #include <trace/events/skb.h> 113 #include <net/busy_poll.h> 114 #include "udp_impl.h" 115 116 struct udp_table udp_table __read_mostly; 117 EXPORT_SYMBOL(udp_table); 118 119 long sysctl_udp_mem[3] __read_mostly; 120 EXPORT_SYMBOL(sysctl_udp_mem); 121 122 int sysctl_udp_rmem_min __read_mostly; 123 EXPORT_SYMBOL(sysctl_udp_rmem_min); 124 125 int sysctl_udp_wmem_min __read_mostly; 126 EXPORT_SYMBOL(sysctl_udp_wmem_min); 127 128 atomic_long_t udp_memory_allocated; 129 EXPORT_SYMBOL(udp_memory_allocated); 130 131 #define MAX_UDP_PORTS 65536 132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 133 134 static int udp_lib_lport_inuse(struct net *net, __u16 num, 135 const struct udp_hslot *hslot, 136 unsigned long *bitmap, 137 struct sock *sk, 138 int (*saddr_comp)(const struct sock *sk1, 139 const struct sock *sk2), 140 unsigned int log) 141 { 142 struct sock *sk2; 143 struct hlist_nulls_node *node; 144 kuid_t uid = sock_i_uid(sk); 145 146 sk_nulls_for_each(sk2, node, &hslot->head) 147 if (net_eq(sock_net(sk2), net) && 148 sk2 != sk && 149 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 150 (!sk2->sk_reuse || !sk->sk_reuse) && 151 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 152 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 153 (!sk2->sk_reuseport || !sk->sk_reuseport || 154 !uid_eq(uid, sock_i_uid(sk2))) && 155 (*saddr_comp)(sk, sk2)) { 156 if (bitmap) 157 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 158 bitmap); 159 else 160 return 1; 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 spin_unlock(&hslot2->lock); 195 return res; 196 } 197 198 /** 199 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 200 * 201 * @sk: socket struct in question 202 * @snum: port number to look up 203 * @saddr_comp: AF-dependent comparison of bound local IP addresses 204 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 205 * with NULL address 206 */ 207 int udp_lib_get_port(struct sock *sk, unsigned short snum, 208 int (*saddr_comp)(const struct sock *sk1, 209 const struct sock *sk2), 210 unsigned int hash2_nulladdr) 211 { 212 struct udp_hslot *hslot, *hslot2; 213 struct udp_table *udptable = sk->sk_prot->h.udp_table; 214 int error = 1; 215 struct net *net = sock_net(sk); 216 217 if (!snum) { 218 int low, high, remaining; 219 unsigned int rand; 220 unsigned short first, last; 221 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 222 223 inet_get_local_port_range(net, &low, &high); 224 remaining = (high - low) + 1; 225 226 rand = prandom_u32(); 227 first = (((u64)rand * remaining) >> 32) + low; 228 /* 229 * force rand to be an odd multiple of UDP_HTABLE_SIZE 230 */ 231 rand = (rand | 1) * (udptable->mask + 1); 232 last = first + udptable->mask + 1; 233 do { 234 hslot = udp_hashslot(udptable, net, first); 235 bitmap_zero(bitmap, PORTS_PER_CHAIN); 236 spin_lock_bh(&hslot->lock); 237 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 238 saddr_comp, udptable->log); 239 240 snum = first; 241 /* 242 * Iterate on all possible values of snum for this hash. 243 * Using steps of an odd multiple of UDP_HTABLE_SIZE 244 * give us randomization and full range coverage. 245 */ 246 do { 247 if (low <= snum && snum <= high && 248 !test_bit(snum >> udptable->log, bitmap) && 249 !inet_is_local_reserved_port(net, snum)) 250 goto found; 251 snum += rand; 252 } while (snum != first); 253 spin_unlock_bh(&hslot->lock); 254 } while (++first != last); 255 goto fail; 256 } else { 257 hslot = udp_hashslot(udptable, net, snum); 258 spin_lock_bh(&hslot->lock); 259 if (hslot->count > 10) { 260 int exist; 261 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 262 263 slot2 &= udptable->mask; 264 hash2_nulladdr &= udptable->mask; 265 266 hslot2 = udp_hashslot2(udptable, slot2); 267 if (hslot->count < hslot2->count) 268 goto scan_primary_hash; 269 270 exist = udp_lib_lport_inuse2(net, snum, hslot2, 271 sk, saddr_comp); 272 if (!exist && (hash2_nulladdr != slot2)) { 273 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 274 exist = udp_lib_lport_inuse2(net, snum, hslot2, 275 sk, saddr_comp); 276 } 277 if (exist) 278 goto fail_unlock; 279 else 280 goto found; 281 } 282 scan_primary_hash: 283 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 284 saddr_comp, 0)) 285 goto fail_unlock; 286 } 287 found: 288 inet_sk(sk)->inet_num = snum; 289 udp_sk(sk)->udp_port_hash = snum; 290 udp_sk(sk)->udp_portaddr_hash ^= snum; 291 if (sk_unhashed(sk)) { 292 sk_nulls_add_node_rcu(sk, &hslot->head); 293 hslot->count++; 294 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 295 296 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 297 spin_lock(&hslot2->lock); 298 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 299 &hslot2->head); 300 hslot2->count++; 301 spin_unlock(&hslot2->lock); 302 } 303 error = 0; 304 fail_unlock: 305 spin_unlock_bh(&hslot->lock); 306 fail: 307 return error; 308 } 309 EXPORT_SYMBOL(udp_lib_get_port); 310 311 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 312 { 313 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 314 315 return (!ipv6_only_sock(sk2) && 316 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 317 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 318 } 319 320 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 321 unsigned int port) 322 { 323 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 324 } 325 326 int udp_v4_get_port(struct sock *sk, unsigned short snum) 327 { 328 unsigned int hash2_nulladdr = 329 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 330 unsigned int hash2_partial = 331 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 332 333 /* precompute partial secondary hash */ 334 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 335 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 336 } 337 338 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 339 unsigned short hnum, 340 __be16 sport, __be32 daddr, __be16 dport, int dif) 341 { 342 int score = -1; 343 344 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 345 !ipv6_only_sock(sk)) { 346 struct inet_sock *inet = inet_sk(sk); 347 348 score = (sk->sk_family == PF_INET ? 2 : 1); 349 if (inet->inet_rcv_saddr) { 350 if (inet->inet_rcv_saddr != daddr) 351 return -1; 352 score += 4; 353 } 354 if (inet->inet_daddr) { 355 if (inet->inet_daddr != saddr) 356 return -1; 357 score += 4; 358 } 359 if (inet->inet_dport) { 360 if (inet->inet_dport != sport) 361 return -1; 362 score += 4; 363 } 364 if (sk->sk_bound_dev_if) { 365 if (sk->sk_bound_dev_if != dif) 366 return -1; 367 score += 4; 368 } 369 } 370 return score; 371 } 372 373 /* 374 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 375 */ 376 static inline int compute_score2(struct sock *sk, struct net *net, 377 __be32 saddr, __be16 sport, 378 __be32 daddr, unsigned int hnum, int dif) 379 { 380 int score = -1; 381 382 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) { 383 struct inet_sock *inet = inet_sk(sk); 384 385 if (inet->inet_rcv_saddr != daddr) 386 return -1; 387 if (inet->inet_num != hnum) 388 return -1; 389 390 score = (sk->sk_family == PF_INET ? 2 : 1); 391 if (inet->inet_daddr) { 392 if (inet->inet_daddr != saddr) 393 return -1; 394 score += 4; 395 } 396 if (inet->inet_dport) { 397 if (inet->inet_dport != sport) 398 return -1; 399 score += 4; 400 } 401 if (sk->sk_bound_dev_if) { 402 if (sk->sk_bound_dev_if != dif) 403 return -1; 404 score += 4; 405 } 406 } 407 return score; 408 } 409 410 static unsigned int udp_ehashfn(struct net *net, const __be32 laddr, 411 const __u16 lport, const __be32 faddr, 412 const __be16 fport) 413 { 414 static u32 udp_ehash_secret __read_mostly; 415 416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 417 418 return __inet_ehashfn(laddr, lport, faddr, fport, 419 udp_ehash_secret + net_hash_mix(net)); 420 } 421 422 423 /* called with read_rcu_lock() */ 424 static struct sock *udp4_lib_lookup2(struct net *net, 425 __be32 saddr, __be16 sport, 426 __be32 daddr, unsigned int hnum, int dif, 427 struct udp_hslot *hslot2, unsigned int slot2) 428 { 429 struct sock *sk, *result; 430 struct hlist_nulls_node *node; 431 int score, badness, matches = 0, reuseport = 0; 432 u32 hash = 0; 433 434 begin: 435 result = NULL; 436 badness = 0; 437 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 438 score = compute_score2(sk, net, saddr, sport, 439 daddr, hnum, dif); 440 if (score > badness) { 441 result = sk; 442 badness = score; 443 reuseport = sk->sk_reuseport; 444 if (reuseport) { 445 hash = udp_ehashfn(net, daddr, hnum, 446 saddr, sport); 447 matches = 1; 448 } 449 } else if (score == badness && reuseport) { 450 matches++; 451 if (((u64)hash * matches) >> 32 == 0) 452 result = sk; 453 hash = next_pseudo_random32(hash); 454 } 455 } 456 /* 457 * if the nulls value we got at the end of this lookup is 458 * not the expected one, we must restart lookup. 459 * We probably met an item that was moved to another chain. 460 */ 461 if (get_nulls_value(node) != slot2) 462 goto begin; 463 if (result) { 464 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 465 result = NULL; 466 else if (unlikely(compute_score2(result, net, saddr, sport, 467 daddr, hnum, dif) < badness)) { 468 sock_put(result); 469 goto begin; 470 } 471 } 472 return result; 473 } 474 475 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 476 * harder than this. -DaveM 477 */ 478 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 479 __be16 sport, __be32 daddr, __be16 dport, 480 int dif, struct udp_table *udptable) 481 { 482 struct sock *sk, *result; 483 struct hlist_nulls_node *node; 484 unsigned short hnum = ntohs(dport); 485 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 486 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 487 int score, badness, matches = 0, reuseport = 0; 488 u32 hash = 0; 489 490 rcu_read_lock(); 491 if (hslot->count > 10) { 492 hash2 = udp4_portaddr_hash(net, daddr, hnum); 493 slot2 = hash2 & udptable->mask; 494 hslot2 = &udptable->hash2[slot2]; 495 if (hslot->count < hslot2->count) 496 goto begin; 497 498 result = udp4_lib_lookup2(net, saddr, sport, 499 daddr, hnum, dif, 500 hslot2, slot2); 501 if (!result) { 502 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 503 slot2 = hash2 & udptable->mask; 504 hslot2 = &udptable->hash2[slot2]; 505 if (hslot->count < hslot2->count) 506 goto begin; 507 508 result = udp4_lib_lookup2(net, saddr, sport, 509 htonl(INADDR_ANY), hnum, dif, 510 hslot2, slot2); 511 } 512 rcu_read_unlock(); 513 return result; 514 } 515 begin: 516 result = NULL; 517 badness = 0; 518 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 519 score = compute_score(sk, net, saddr, hnum, sport, 520 daddr, dport, dif); 521 if (score > badness) { 522 result = sk; 523 badness = score; 524 reuseport = sk->sk_reuseport; 525 if (reuseport) { 526 hash = udp_ehashfn(net, daddr, hnum, 527 saddr, sport); 528 matches = 1; 529 } 530 } else if (score == badness && reuseport) { 531 matches++; 532 if (((u64)hash * matches) >> 32 == 0) 533 result = sk; 534 hash = next_pseudo_random32(hash); 535 } 536 } 537 /* 538 * if the nulls value we got at the end of this lookup is 539 * not the expected one, we must restart lookup. 540 * We probably met an item that was moved to another chain. 541 */ 542 if (get_nulls_value(node) != slot) 543 goto begin; 544 545 if (result) { 546 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 547 result = NULL; 548 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 549 daddr, dport, dif) < badness)) { 550 sock_put(result); 551 goto begin; 552 } 553 } 554 rcu_read_unlock(); 555 return result; 556 } 557 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 558 559 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 560 __be16 sport, __be16 dport, 561 struct udp_table *udptable) 562 { 563 const struct iphdr *iph = ip_hdr(skb); 564 565 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 566 iph->daddr, dport, inet_iif(skb), 567 udptable); 568 } 569 570 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 571 __be32 daddr, __be16 dport, int dif) 572 { 573 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 574 } 575 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 576 577 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 578 __be16 loc_port, __be32 loc_addr, 579 __be16 rmt_port, __be32 rmt_addr, 580 int dif, unsigned short hnum) 581 { 582 struct inet_sock *inet = inet_sk(sk); 583 584 if (!net_eq(sock_net(sk), net) || 585 udp_sk(sk)->udp_port_hash != hnum || 586 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 587 (inet->inet_dport != rmt_port && inet->inet_dport) || 588 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 589 ipv6_only_sock(sk) || 590 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 591 return false; 592 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 593 return false; 594 return true; 595 } 596 597 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 598 __be16 loc_port, __be32 loc_addr, 599 __be16 rmt_port, __be32 rmt_addr, 600 int dif) 601 { 602 struct hlist_nulls_node *node; 603 struct sock *s = sk; 604 unsigned short hnum = ntohs(loc_port); 605 606 sk_nulls_for_each_from(s, node) { 607 if (__udp_is_mcast_sock(net, s, 608 loc_port, loc_addr, 609 rmt_port, rmt_addr, 610 dif, hnum)) 611 goto found; 612 } 613 s = NULL; 614 found: 615 return s; 616 } 617 618 /* 619 * This routine is called by the ICMP module when it gets some 620 * sort of error condition. If err < 0 then the socket should 621 * be closed and the error returned to the user. If err > 0 622 * it's just the icmp type << 8 | icmp code. 623 * Header points to the ip header of the error packet. We move 624 * on past this. Then (as it used to claim before adjustment) 625 * header points to the first 8 bytes of the udp header. We need 626 * to find the appropriate port. 627 */ 628 629 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 630 { 631 struct inet_sock *inet; 632 const struct iphdr *iph = (const struct iphdr *)skb->data; 633 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 634 const int type = icmp_hdr(skb)->type; 635 const int code = icmp_hdr(skb)->code; 636 struct sock *sk; 637 int harderr; 638 int err; 639 struct net *net = dev_net(skb->dev); 640 641 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 642 iph->saddr, uh->source, skb->dev->ifindex, udptable); 643 if (sk == NULL) { 644 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 645 return; /* No socket for error */ 646 } 647 648 err = 0; 649 harderr = 0; 650 inet = inet_sk(sk); 651 652 switch (type) { 653 default: 654 case ICMP_TIME_EXCEEDED: 655 err = EHOSTUNREACH; 656 break; 657 case ICMP_SOURCE_QUENCH: 658 goto out; 659 case ICMP_PARAMETERPROB: 660 err = EPROTO; 661 harderr = 1; 662 break; 663 case ICMP_DEST_UNREACH: 664 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 665 ipv4_sk_update_pmtu(skb, sk, info); 666 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 667 err = EMSGSIZE; 668 harderr = 1; 669 break; 670 } 671 goto out; 672 } 673 err = EHOSTUNREACH; 674 if (code <= NR_ICMP_UNREACH) { 675 harderr = icmp_err_convert[code].fatal; 676 err = icmp_err_convert[code].errno; 677 } 678 break; 679 case ICMP_REDIRECT: 680 ipv4_sk_redirect(skb, sk); 681 goto out; 682 } 683 684 /* 685 * RFC1122: OK. Passes ICMP errors back to application, as per 686 * 4.1.3.3. 687 */ 688 if (!inet->recverr) { 689 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 690 goto out; 691 } else 692 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 693 694 sk->sk_err = err; 695 sk->sk_error_report(sk); 696 out: 697 sock_put(sk); 698 } 699 700 void udp_err(struct sk_buff *skb, u32 info) 701 { 702 __udp4_lib_err(skb, info, &udp_table); 703 } 704 705 /* 706 * Throw away all pending data and cancel the corking. Socket is locked. 707 */ 708 void udp_flush_pending_frames(struct sock *sk) 709 { 710 struct udp_sock *up = udp_sk(sk); 711 712 if (up->pending) { 713 up->len = 0; 714 up->pending = 0; 715 ip_flush_pending_frames(sk); 716 } 717 } 718 EXPORT_SYMBOL(udp_flush_pending_frames); 719 720 /** 721 * udp4_hwcsum - handle outgoing HW checksumming 722 * @skb: sk_buff containing the filled-in UDP header 723 * (checksum field must be zeroed out) 724 * @src: source IP address 725 * @dst: destination IP address 726 */ 727 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 728 { 729 struct udphdr *uh = udp_hdr(skb); 730 int offset = skb_transport_offset(skb); 731 int len = skb->len - offset; 732 int hlen = len; 733 __wsum csum = 0; 734 735 if (!skb_has_frag_list(skb)) { 736 /* 737 * Only one fragment on the socket. 738 */ 739 skb->csum_start = skb_transport_header(skb) - skb->head; 740 skb->csum_offset = offsetof(struct udphdr, check); 741 uh->check = ~csum_tcpudp_magic(src, dst, len, 742 IPPROTO_UDP, 0); 743 } else { 744 struct sk_buff *frags; 745 746 /* 747 * HW-checksum won't work as there are two or more 748 * fragments on the socket so that all csums of sk_buffs 749 * should be together 750 */ 751 skb_walk_frags(skb, frags) { 752 csum = csum_add(csum, frags->csum); 753 hlen -= frags->len; 754 } 755 756 csum = skb_checksum(skb, offset, hlen, csum); 757 skb->ip_summed = CHECKSUM_NONE; 758 759 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 760 if (uh->check == 0) 761 uh->check = CSUM_MANGLED_0; 762 } 763 } 764 EXPORT_SYMBOL_GPL(udp4_hwcsum); 765 766 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 767 * for the simple case like when setting the checksum for a UDP tunnel. 768 */ 769 void udp_set_csum(bool nocheck, struct sk_buff *skb, 770 __be32 saddr, __be32 daddr, int len) 771 { 772 struct udphdr *uh = udp_hdr(skb); 773 774 if (nocheck) 775 uh->check = 0; 776 else if (skb_is_gso(skb)) 777 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 778 else if (skb_dst(skb) && skb_dst(skb)->dev && 779 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) { 780 781 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 782 783 skb->ip_summed = CHECKSUM_PARTIAL; 784 skb->csum_start = skb_transport_header(skb) - skb->head; 785 skb->csum_offset = offsetof(struct udphdr, check); 786 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 787 } else { 788 __wsum csum; 789 790 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 791 792 uh->check = 0; 793 csum = skb_checksum(skb, 0, len, 0); 794 uh->check = udp_v4_check(len, saddr, daddr, csum); 795 if (uh->check == 0) 796 uh->check = CSUM_MANGLED_0; 797 798 skb->ip_summed = CHECKSUM_UNNECESSARY; 799 } 800 } 801 EXPORT_SYMBOL(udp_set_csum); 802 803 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 804 { 805 struct sock *sk = skb->sk; 806 struct inet_sock *inet = inet_sk(sk); 807 struct udphdr *uh; 808 int err = 0; 809 int is_udplite = IS_UDPLITE(sk); 810 int offset = skb_transport_offset(skb); 811 int len = skb->len - offset; 812 __wsum csum = 0; 813 814 /* 815 * Create a UDP header 816 */ 817 uh = udp_hdr(skb); 818 uh->source = inet->inet_sport; 819 uh->dest = fl4->fl4_dport; 820 uh->len = htons(len); 821 uh->check = 0; 822 823 if (is_udplite) /* UDP-Lite */ 824 csum = udplite_csum(skb); 825 826 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 827 828 skb->ip_summed = CHECKSUM_NONE; 829 goto send; 830 831 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 832 833 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 834 goto send; 835 836 } else 837 csum = udp_csum(skb); 838 839 /* add protocol-dependent pseudo-header */ 840 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 841 sk->sk_protocol, csum); 842 if (uh->check == 0) 843 uh->check = CSUM_MANGLED_0; 844 845 send: 846 err = ip_send_skb(sock_net(sk), skb); 847 if (err) { 848 if (err == -ENOBUFS && !inet->recverr) { 849 UDP_INC_STATS_USER(sock_net(sk), 850 UDP_MIB_SNDBUFERRORS, is_udplite); 851 err = 0; 852 } 853 } else 854 UDP_INC_STATS_USER(sock_net(sk), 855 UDP_MIB_OUTDATAGRAMS, is_udplite); 856 return err; 857 } 858 859 /* 860 * Push out all pending data as one UDP datagram. Socket is locked. 861 */ 862 int udp_push_pending_frames(struct sock *sk) 863 { 864 struct udp_sock *up = udp_sk(sk); 865 struct inet_sock *inet = inet_sk(sk); 866 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 867 struct sk_buff *skb; 868 int err = 0; 869 870 skb = ip_finish_skb(sk, fl4); 871 if (!skb) 872 goto out; 873 874 err = udp_send_skb(skb, fl4); 875 876 out: 877 up->len = 0; 878 up->pending = 0; 879 return err; 880 } 881 EXPORT_SYMBOL(udp_push_pending_frames); 882 883 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 884 size_t len) 885 { 886 struct inet_sock *inet = inet_sk(sk); 887 struct udp_sock *up = udp_sk(sk); 888 struct flowi4 fl4_stack; 889 struct flowi4 *fl4; 890 int ulen = len; 891 struct ipcm_cookie ipc; 892 struct rtable *rt = NULL; 893 int free = 0; 894 int connected = 0; 895 __be32 daddr, faddr, saddr; 896 __be16 dport; 897 u8 tos; 898 int err, is_udplite = IS_UDPLITE(sk); 899 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 900 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 901 struct sk_buff *skb; 902 struct ip_options_data opt_copy; 903 904 if (len > 0xFFFF) 905 return -EMSGSIZE; 906 907 /* 908 * Check the flags. 909 */ 910 911 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 912 return -EOPNOTSUPP; 913 914 ipc.opt = NULL; 915 ipc.tx_flags = 0; 916 ipc.ttl = 0; 917 ipc.tos = -1; 918 919 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 920 921 fl4 = &inet->cork.fl.u.ip4; 922 if (up->pending) { 923 /* 924 * There are pending frames. 925 * The socket lock must be held while it's corked. 926 */ 927 lock_sock(sk); 928 if (likely(up->pending)) { 929 if (unlikely(up->pending != AF_INET)) { 930 release_sock(sk); 931 return -EINVAL; 932 } 933 goto do_append_data; 934 } 935 release_sock(sk); 936 } 937 ulen += sizeof(struct udphdr); 938 939 /* 940 * Get and verify the address. 941 */ 942 if (msg->msg_name) { 943 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 944 if (msg->msg_namelen < sizeof(*usin)) 945 return -EINVAL; 946 if (usin->sin_family != AF_INET) { 947 if (usin->sin_family != AF_UNSPEC) 948 return -EAFNOSUPPORT; 949 } 950 951 daddr = usin->sin_addr.s_addr; 952 dport = usin->sin_port; 953 if (dport == 0) 954 return -EINVAL; 955 } else { 956 if (sk->sk_state != TCP_ESTABLISHED) 957 return -EDESTADDRREQ; 958 daddr = inet->inet_daddr; 959 dport = inet->inet_dport; 960 /* Open fast path for connected socket. 961 Route will not be used, if at least one option is set. 962 */ 963 connected = 1; 964 } 965 ipc.addr = inet->inet_saddr; 966 967 ipc.oif = sk->sk_bound_dev_if; 968 969 sock_tx_timestamp(sk, &ipc.tx_flags); 970 971 if (msg->msg_controllen) { 972 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 973 sk->sk_family == AF_INET6); 974 if (err) 975 return err; 976 if (ipc.opt) 977 free = 1; 978 connected = 0; 979 } 980 if (!ipc.opt) { 981 struct ip_options_rcu *inet_opt; 982 983 rcu_read_lock(); 984 inet_opt = rcu_dereference(inet->inet_opt); 985 if (inet_opt) { 986 memcpy(&opt_copy, inet_opt, 987 sizeof(*inet_opt) + inet_opt->opt.optlen); 988 ipc.opt = &opt_copy.opt; 989 } 990 rcu_read_unlock(); 991 } 992 993 saddr = ipc.addr; 994 ipc.addr = faddr = daddr; 995 996 if (ipc.opt && ipc.opt->opt.srr) { 997 if (!daddr) 998 return -EINVAL; 999 faddr = ipc.opt->opt.faddr; 1000 connected = 0; 1001 } 1002 tos = get_rttos(&ipc, inet); 1003 if (sock_flag(sk, SOCK_LOCALROUTE) || 1004 (msg->msg_flags & MSG_DONTROUTE) || 1005 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1006 tos |= RTO_ONLINK; 1007 connected = 0; 1008 } 1009 1010 if (ipv4_is_multicast(daddr)) { 1011 if (!ipc.oif) 1012 ipc.oif = inet->mc_index; 1013 if (!saddr) 1014 saddr = inet->mc_addr; 1015 connected = 0; 1016 } else if (!ipc.oif) 1017 ipc.oif = inet->uc_index; 1018 1019 if (connected) 1020 rt = (struct rtable *)sk_dst_check(sk, 0); 1021 1022 if (rt == NULL) { 1023 struct net *net = sock_net(sk); 1024 1025 fl4 = &fl4_stack; 1026 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1027 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1028 inet_sk_flowi_flags(sk), 1029 faddr, saddr, dport, inet->inet_sport); 1030 1031 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1032 rt = ip_route_output_flow(net, fl4, sk); 1033 if (IS_ERR(rt)) { 1034 err = PTR_ERR(rt); 1035 rt = NULL; 1036 if (err == -ENETUNREACH) 1037 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1038 goto out; 1039 } 1040 1041 err = -EACCES; 1042 if ((rt->rt_flags & RTCF_BROADCAST) && 1043 !sock_flag(sk, SOCK_BROADCAST)) 1044 goto out; 1045 if (connected) 1046 sk_dst_set(sk, dst_clone(&rt->dst)); 1047 } 1048 1049 if (msg->msg_flags&MSG_CONFIRM) 1050 goto do_confirm; 1051 back_from_confirm: 1052 1053 saddr = fl4->saddr; 1054 if (!ipc.addr) 1055 daddr = ipc.addr = fl4->daddr; 1056 1057 /* Lockless fast path for the non-corking case. */ 1058 if (!corkreq) { 1059 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen, 1060 sizeof(struct udphdr), &ipc, &rt, 1061 msg->msg_flags); 1062 err = PTR_ERR(skb); 1063 if (!IS_ERR_OR_NULL(skb)) 1064 err = udp_send_skb(skb, fl4); 1065 goto out; 1066 } 1067 1068 lock_sock(sk); 1069 if (unlikely(up->pending)) { 1070 /* The socket is already corked while preparing it. */ 1071 /* ... which is an evident application bug. --ANK */ 1072 release_sock(sk); 1073 1074 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n")); 1075 err = -EINVAL; 1076 goto out; 1077 } 1078 /* 1079 * Now cork the socket to pend data. 1080 */ 1081 fl4 = &inet->cork.fl.u.ip4; 1082 fl4->daddr = daddr; 1083 fl4->saddr = saddr; 1084 fl4->fl4_dport = dport; 1085 fl4->fl4_sport = inet->inet_sport; 1086 up->pending = AF_INET; 1087 1088 do_append_data: 1089 up->len += ulen; 1090 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen, 1091 sizeof(struct udphdr), &ipc, &rt, 1092 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1093 if (err) 1094 udp_flush_pending_frames(sk); 1095 else if (!corkreq) 1096 err = udp_push_pending_frames(sk); 1097 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1098 up->pending = 0; 1099 release_sock(sk); 1100 1101 out: 1102 ip_rt_put(rt); 1103 if (free) 1104 kfree(ipc.opt); 1105 if (!err) 1106 return len; 1107 /* 1108 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1109 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1110 * we don't have a good statistic (IpOutDiscards but it can be too many 1111 * things). We could add another new stat but at least for now that 1112 * seems like overkill. 1113 */ 1114 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1115 UDP_INC_STATS_USER(sock_net(sk), 1116 UDP_MIB_SNDBUFERRORS, is_udplite); 1117 } 1118 return err; 1119 1120 do_confirm: 1121 dst_confirm(&rt->dst); 1122 if (!(msg->msg_flags&MSG_PROBE) || len) 1123 goto back_from_confirm; 1124 err = 0; 1125 goto out; 1126 } 1127 EXPORT_SYMBOL(udp_sendmsg); 1128 1129 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1130 size_t size, int flags) 1131 { 1132 struct inet_sock *inet = inet_sk(sk); 1133 struct udp_sock *up = udp_sk(sk); 1134 int ret; 1135 1136 if (flags & MSG_SENDPAGE_NOTLAST) 1137 flags |= MSG_MORE; 1138 1139 if (!up->pending) { 1140 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1141 1142 /* Call udp_sendmsg to specify destination address which 1143 * sendpage interface can't pass. 1144 * This will succeed only when the socket is connected. 1145 */ 1146 ret = udp_sendmsg(NULL, sk, &msg, 0); 1147 if (ret < 0) 1148 return ret; 1149 } 1150 1151 lock_sock(sk); 1152 1153 if (unlikely(!up->pending)) { 1154 release_sock(sk); 1155 1156 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n")); 1157 return -EINVAL; 1158 } 1159 1160 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1161 page, offset, size, flags); 1162 if (ret == -EOPNOTSUPP) { 1163 release_sock(sk); 1164 return sock_no_sendpage(sk->sk_socket, page, offset, 1165 size, flags); 1166 } 1167 if (ret < 0) { 1168 udp_flush_pending_frames(sk); 1169 goto out; 1170 } 1171 1172 up->len += size; 1173 if (!(up->corkflag || (flags&MSG_MORE))) 1174 ret = udp_push_pending_frames(sk); 1175 if (!ret) 1176 ret = size; 1177 out: 1178 release_sock(sk); 1179 return ret; 1180 } 1181 1182 1183 /** 1184 * first_packet_length - return length of first packet in receive queue 1185 * @sk: socket 1186 * 1187 * Drops all bad checksum frames, until a valid one is found. 1188 * Returns the length of found skb, or 0 if none is found. 1189 */ 1190 static unsigned int first_packet_length(struct sock *sk) 1191 { 1192 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1193 struct sk_buff *skb; 1194 unsigned int res; 1195 1196 __skb_queue_head_init(&list_kill); 1197 1198 spin_lock_bh(&rcvq->lock); 1199 while ((skb = skb_peek(rcvq)) != NULL && 1200 udp_lib_checksum_complete(skb)) { 1201 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1202 IS_UDPLITE(sk)); 1203 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1204 IS_UDPLITE(sk)); 1205 atomic_inc(&sk->sk_drops); 1206 __skb_unlink(skb, rcvq); 1207 __skb_queue_tail(&list_kill, skb); 1208 } 1209 res = skb ? skb->len : 0; 1210 spin_unlock_bh(&rcvq->lock); 1211 1212 if (!skb_queue_empty(&list_kill)) { 1213 bool slow = lock_sock_fast(sk); 1214 1215 __skb_queue_purge(&list_kill); 1216 sk_mem_reclaim_partial(sk); 1217 unlock_sock_fast(sk, slow); 1218 } 1219 return res; 1220 } 1221 1222 /* 1223 * IOCTL requests applicable to the UDP protocol 1224 */ 1225 1226 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1227 { 1228 switch (cmd) { 1229 case SIOCOUTQ: 1230 { 1231 int amount = sk_wmem_alloc_get(sk); 1232 1233 return put_user(amount, (int __user *)arg); 1234 } 1235 1236 case SIOCINQ: 1237 { 1238 unsigned int amount = first_packet_length(sk); 1239 1240 if (amount) 1241 /* 1242 * We will only return the amount 1243 * of this packet since that is all 1244 * that will be read. 1245 */ 1246 amount -= sizeof(struct udphdr); 1247 1248 return put_user(amount, (int __user *)arg); 1249 } 1250 1251 default: 1252 return -ENOIOCTLCMD; 1253 } 1254 1255 return 0; 1256 } 1257 EXPORT_SYMBOL(udp_ioctl); 1258 1259 /* 1260 * This should be easy, if there is something there we 1261 * return it, otherwise we block. 1262 */ 1263 1264 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1265 size_t len, int noblock, int flags, int *addr_len) 1266 { 1267 struct inet_sock *inet = inet_sk(sk); 1268 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1269 struct sk_buff *skb; 1270 unsigned int ulen, copied; 1271 int peeked, off = 0; 1272 int err; 1273 int is_udplite = IS_UDPLITE(sk); 1274 bool slow; 1275 1276 if (flags & MSG_ERRQUEUE) 1277 return ip_recv_error(sk, msg, len, addr_len); 1278 1279 try_again: 1280 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1281 &peeked, &off, &err); 1282 if (!skb) 1283 goto out; 1284 1285 ulen = skb->len - sizeof(struct udphdr); 1286 copied = len; 1287 if (copied > ulen) 1288 copied = ulen; 1289 else if (copied < ulen) 1290 msg->msg_flags |= MSG_TRUNC; 1291 1292 /* 1293 * If checksum is needed at all, try to do it while copying the 1294 * data. If the data is truncated, or if we only want a partial 1295 * coverage checksum (UDP-Lite), do it before the copy. 1296 */ 1297 1298 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1299 if (udp_lib_checksum_complete(skb)) 1300 goto csum_copy_err; 1301 } 1302 1303 if (skb_csum_unnecessary(skb)) 1304 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1305 msg->msg_iov, copied); 1306 else { 1307 err = skb_copy_and_csum_datagram_iovec(skb, 1308 sizeof(struct udphdr), 1309 msg->msg_iov); 1310 1311 if (err == -EINVAL) 1312 goto csum_copy_err; 1313 } 1314 1315 if (unlikely(err)) { 1316 trace_kfree_skb(skb, udp_recvmsg); 1317 if (!peeked) { 1318 atomic_inc(&sk->sk_drops); 1319 UDP_INC_STATS_USER(sock_net(sk), 1320 UDP_MIB_INERRORS, is_udplite); 1321 } 1322 goto out_free; 1323 } 1324 1325 if (!peeked) 1326 UDP_INC_STATS_USER(sock_net(sk), 1327 UDP_MIB_INDATAGRAMS, is_udplite); 1328 1329 sock_recv_ts_and_drops(msg, sk, skb); 1330 1331 /* Copy the address. */ 1332 if (sin) { 1333 sin->sin_family = AF_INET; 1334 sin->sin_port = udp_hdr(skb)->source; 1335 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1336 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1337 *addr_len = sizeof(*sin); 1338 } 1339 if (inet->cmsg_flags) 1340 ip_cmsg_recv(msg, skb); 1341 1342 err = copied; 1343 if (flags & MSG_TRUNC) 1344 err = ulen; 1345 1346 out_free: 1347 skb_free_datagram_locked(sk, skb); 1348 out: 1349 return err; 1350 1351 csum_copy_err: 1352 slow = lock_sock_fast(sk); 1353 if (!skb_kill_datagram(sk, skb, flags)) { 1354 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1355 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1356 } 1357 unlock_sock_fast(sk, slow); 1358 1359 if (noblock) 1360 return -EAGAIN; 1361 1362 /* starting over for a new packet */ 1363 msg->msg_flags &= ~MSG_TRUNC; 1364 goto try_again; 1365 } 1366 1367 1368 int udp_disconnect(struct sock *sk, int flags) 1369 { 1370 struct inet_sock *inet = inet_sk(sk); 1371 /* 1372 * 1003.1g - break association. 1373 */ 1374 1375 sk->sk_state = TCP_CLOSE; 1376 inet->inet_daddr = 0; 1377 inet->inet_dport = 0; 1378 sock_rps_reset_rxhash(sk); 1379 sk->sk_bound_dev_if = 0; 1380 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1381 inet_reset_saddr(sk); 1382 1383 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1384 sk->sk_prot->unhash(sk); 1385 inet->inet_sport = 0; 1386 } 1387 sk_dst_reset(sk); 1388 return 0; 1389 } 1390 EXPORT_SYMBOL(udp_disconnect); 1391 1392 void udp_lib_unhash(struct sock *sk) 1393 { 1394 if (sk_hashed(sk)) { 1395 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1396 struct udp_hslot *hslot, *hslot2; 1397 1398 hslot = udp_hashslot(udptable, sock_net(sk), 1399 udp_sk(sk)->udp_port_hash); 1400 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1401 1402 spin_lock_bh(&hslot->lock); 1403 if (sk_nulls_del_node_init_rcu(sk)) { 1404 hslot->count--; 1405 inet_sk(sk)->inet_num = 0; 1406 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1407 1408 spin_lock(&hslot2->lock); 1409 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1410 hslot2->count--; 1411 spin_unlock(&hslot2->lock); 1412 } 1413 spin_unlock_bh(&hslot->lock); 1414 } 1415 } 1416 EXPORT_SYMBOL(udp_lib_unhash); 1417 1418 /* 1419 * inet_rcv_saddr was changed, we must rehash secondary hash 1420 */ 1421 void udp_lib_rehash(struct sock *sk, u16 newhash) 1422 { 1423 if (sk_hashed(sk)) { 1424 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1425 struct udp_hslot *hslot, *hslot2, *nhslot2; 1426 1427 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1428 nhslot2 = udp_hashslot2(udptable, newhash); 1429 udp_sk(sk)->udp_portaddr_hash = newhash; 1430 if (hslot2 != nhslot2) { 1431 hslot = udp_hashslot(udptable, sock_net(sk), 1432 udp_sk(sk)->udp_port_hash); 1433 /* we must lock primary chain too */ 1434 spin_lock_bh(&hslot->lock); 1435 1436 spin_lock(&hslot2->lock); 1437 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1438 hslot2->count--; 1439 spin_unlock(&hslot2->lock); 1440 1441 spin_lock(&nhslot2->lock); 1442 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1443 &nhslot2->head); 1444 nhslot2->count++; 1445 spin_unlock(&nhslot2->lock); 1446 1447 spin_unlock_bh(&hslot->lock); 1448 } 1449 } 1450 } 1451 EXPORT_SYMBOL(udp_lib_rehash); 1452 1453 static void udp_v4_rehash(struct sock *sk) 1454 { 1455 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1456 inet_sk(sk)->inet_rcv_saddr, 1457 inet_sk(sk)->inet_num); 1458 udp_lib_rehash(sk, new_hash); 1459 } 1460 1461 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1462 { 1463 int rc; 1464 1465 if (inet_sk(sk)->inet_daddr) { 1466 sock_rps_save_rxhash(sk, skb); 1467 sk_mark_napi_id(sk, skb); 1468 } 1469 1470 rc = sock_queue_rcv_skb(sk, skb); 1471 if (rc < 0) { 1472 int is_udplite = IS_UDPLITE(sk); 1473 1474 /* Note that an ENOMEM error is charged twice */ 1475 if (rc == -ENOMEM) 1476 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1477 is_udplite); 1478 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1479 kfree_skb(skb); 1480 trace_udp_fail_queue_rcv_skb(rc, sk); 1481 return -1; 1482 } 1483 1484 return 0; 1485 1486 } 1487 1488 static struct static_key udp_encap_needed __read_mostly; 1489 void udp_encap_enable(void) 1490 { 1491 if (!static_key_enabled(&udp_encap_needed)) 1492 static_key_slow_inc(&udp_encap_needed); 1493 } 1494 EXPORT_SYMBOL(udp_encap_enable); 1495 1496 /* returns: 1497 * -1: error 1498 * 0: success 1499 * >0: "udp encap" protocol resubmission 1500 * 1501 * Note that in the success and error cases, the skb is assumed to 1502 * have either been requeued or freed. 1503 */ 1504 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1505 { 1506 struct udp_sock *up = udp_sk(sk); 1507 int rc; 1508 int is_udplite = IS_UDPLITE(sk); 1509 1510 /* 1511 * Charge it to the socket, dropping if the queue is full. 1512 */ 1513 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1514 goto drop; 1515 nf_reset(skb); 1516 1517 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1518 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1519 1520 /* 1521 * This is an encapsulation socket so pass the skb to 1522 * the socket's udp_encap_rcv() hook. Otherwise, just 1523 * fall through and pass this up the UDP socket. 1524 * up->encap_rcv() returns the following value: 1525 * =0 if skb was successfully passed to the encap 1526 * handler or was discarded by it. 1527 * >0 if skb should be passed on to UDP. 1528 * <0 if skb should be resubmitted as proto -N 1529 */ 1530 1531 /* if we're overly short, let UDP handle it */ 1532 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1533 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { 1534 int ret; 1535 1536 /* Verify checksum before giving to encap */ 1537 if (udp_lib_checksum_complete(skb)) 1538 goto csum_error; 1539 1540 ret = encap_rcv(sk, skb); 1541 if (ret <= 0) { 1542 UDP_INC_STATS_BH(sock_net(sk), 1543 UDP_MIB_INDATAGRAMS, 1544 is_udplite); 1545 return -ret; 1546 } 1547 } 1548 1549 /* FALLTHROUGH -- it's a UDP Packet */ 1550 } 1551 1552 /* 1553 * UDP-Lite specific tests, ignored on UDP sockets 1554 */ 1555 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1556 1557 /* 1558 * MIB statistics other than incrementing the error count are 1559 * disabled for the following two types of errors: these depend 1560 * on the application settings, not on the functioning of the 1561 * protocol stack as such. 1562 * 1563 * RFC 3828 here recommends (sec 3.3): "There should also be a 1564 * way ... to ... at least let the receiving application block 1565 * delivery of packets with coverage values less than a value 1566 * provided by the application." 1567 */ 1568 if (up->pcrlen == 0) { /* full coverage was set */ 1569 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n", 1570 UDP_SKB_CB(skb)->cscov, skb->len); 1571 goto drop; 1572 } 1573 /* The next case involves violating the min. coverage requested 1574 * by the receiver. This is subtle: if receiver wants x and x is 1575 * greater than the buffersize/MTU then receiver will complain 1576 * that it wants x while sender emits packets of smaller size y. 1577 * Therefore the above ...()->partial_cov statement is essential. 1578 */ 1579 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1580 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n", 1581 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1582 goto drop; 1583 } 1584 } 1585 1586 if (rcu_access_pointer(sk->sk_filter) && 1587 udp_lib_checksum_complete(skb)) 1588 goto csum_error; 1589 1590 1591 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) 1592 goto drop; 1593 1594 rc = 0; 1595 1596 ipv4_pktinfo_prepare(sk, skb); 1597 bh_lock_sock(sk); 1598 if (!sock_owned_by_user(sk)) 1599 rc = __udp_queue_rcv_skb(sk, skb); 1600 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1601 bh_unlock_sock(sk); 1602 goto drop; 1603 } 1604 bh_unlock_sock(sk); 1605 1606 return rc; 1607 1608 csum_error: 1609 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1610 drop: 1611 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1612 atomic_inc(&sk->sk_drops); 1613 kfree_skb(skb); 1614 return -1; 1615 } 1616 1617 1618 static void flush_stack(struct sock **stack, unsigned int count, 1619 struct sk_buff *skb, unsigned int final) 1620 { 1621 unsigned int i; 1622 struct sk_buff *skb1 = NULL; 1623 struct sock *sk; 1624 1625 for (i = 0; i < count; i++) { 1626 sk = stack[i]; 1627 if (likely(skb1 == NULL)) 1628 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1629 1630 if (!skb1) { 1631 atomic_inc(&sk->sk_drops); 1632 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1633 IS_UDPLITE(sk)); 1634 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1635 IS_UDPLITE(sk)); 1636 } 1637 1638 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1639 skb1 = NULL; 1640 } 1641 if (unlikely(skb1)) 1642 kfree_skb(skb1); 1643 } 1644 1645 /* For TCP sockets, sk_rx_dst is protected by socket lock 1646 * For UDP, we use xchg() to guard against concurrent changes. 1647 */ 1648 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1649 { 1650 struct dst_entry *old; 1651 1652 dst_hold(dst); 1653 old = xchg(&sk->sk_rx_dst, dst); 1654 dst_release(old); 1655 } 1656 1657 /* 1658 * Multicasts and broadcasts go to each listener. 1659 * 1660 * Note: called only from the BH handler context. 1661 */ 1662 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1663 struct udphdr *uh, 1664 __be32 saddr, __be32 daddr, 1665 struct udp_table *udptable) 1666 { 1667 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1668 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1669 int dif; 1670 unsigned int i, count = 0; 1671 1672 spin_lock(&hslot->lock); 1673 sk = sk_nulls_head(&hslot->head); 1674 dif = skb->dev->ifindex; 1675 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1676 while (sk) { 1677 stack[count++] = sk; 1678 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1679 daddr, uh->source, saddr, dif); 1680 if (unlikely(count == ARRAY_SIZE(stack))) { 1681 if (!sk) 1682 break; 1683 flush_stack(stack, count, skb, ~0); 1684 count = 0; 1685 } 1686 } 1687 /* 1688 * before releasing chain lock, we must take a reference on sockets 1689 */ 1690 for (i = 0; i < count; i++) 1691 sock_hold(stack[i]); 1692 1693 spin_unlock(&hslot->lock); 1694 1695 /* 1696 * do the slow work with no lock held 1697 */ 1698 if (count) { 1699 flush_stack(stack, count, skb, count - 1); 1700 1701 for (i = 0; i < count; i++) 1702 sock_put(stack[i]); 1703 } else { 1704 kfree_skb(skb); 1705 } 1706 return 0; 1707 } 1708 1709 /* Initialize UDP checksum. If exited with zero value (success), 1710 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1711 * Otherwise, csum completion requires chacksumming packet body, 1712 * including udp header and folding it to skb->csum. 1713 */ 1714 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1715 int proto) 1716 { 1717 int err; 1718 1719 UDP_SKB_CB(skb)->partial_cov = 0; 1720 UDP_SKB_CB(skb)->cscov = skb->len; 1721 1722 if (proto == IPPROTO_UDPLITE) { 1723 err = udplite_checksum_init(skb, uh); 1724 if (err) 1725 return err; 1726 } 1727 1728 return skb_checksum_init_zero_check(skb, proto, uh->check, 1729 inet_compute_pseudo); 1730 } 1731 1732 /* 1733 * All we need to do is get the socket, and then do a checksum. 1734 */ 1735 1736 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1737 int proto) 1738 { 1739 struct sock *sk; 1740 struct udphdr *uh; 1741 unsigned short ulen; 1742 struct rtable *rt = skb_rtable(skb); 1743 __be32 saddr, daddr; 1744 struct net *net = dev_net(skb->dev); 1745 1746 /* 1747 * Validate the packet. 1748 */ 1749 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1750 goto drop; /* No space for header. */ 1751 1752 uh = udp_hdr(skb); 1753 ulen = ntohs(uh->len); 1754 saddr = ip_hdr(skb)->saddr; 1755 daddr = ip_hdr(skb)->daddr; 1756 1757 if (ulen > skb->len) 1758 goto short_packet; 1759 1760 if (proto == IPPROTO_UDP) { 1761 /* UDP validates ulen. */ 1762 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1763 goto short_packet; 1764 uh = udp_hdr(skb); 1765 } 1766 1767 if (udp4_csum_init(skb, uh, proto)) 1768 goto csum_error; 1769 1770 sk = skb_steal_sock(skb); 1771 if (sk) { 1772 struct dst_entry *dst = skb_dst(skb); 1773 int ret; 1774 1775 if (unlikely(sk->sk_rx_dst != dst)) 1776 udp_sk_rx_dst_set(sk, dst); 1777 1778 ret = udp_queue_rcv_skb(sk, skb); 1779 sock_put(sk); 1780 /* a return value > 0 means to resubmit the input, but 1781 * it wants the return to be -protocol, or 0 1782 */ 1783 if (ret > 0) 1784 return -ret; 1785 return 0; 1786 } else { 1787 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1788 return __udp4_lib_mcast_deliver(net, skb, uh, 1789 saddr, daddr, udptable); 1790 1791 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1792 } 1793 1794 if (sk != NULL) { 1795 int ret; 1796 1797 ret = udp_queue_rcv_skb(sk, skb); 1798 sock_put(sk); 1799 1800 /* a return value > 0 means to resubmit the input, but 1801 * it wants the return to be -protocol, or 0 1802 */ 1803 if (ret > 0) 1804 return -ret; 1805 return 0; 1806 } 1807 1808 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1809 goto drop; 1810 nf_reset(skb); 1811 1812 /* No socket. Drop packet silently, if checksum is wrong */ 1813 if (udp_lib_checksum_complete(skb)) 1814 goto csum_error; 1815 1816 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1817 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1818 1819 /* 1820 * Hmm. We got an UDP packet to a port to which we 1821 * don't wanna listen. Ignore it. 1822 */ 1823 kfree_skb(skb); 1824 return 0; 1825 1826 short_packet: 1827 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1828 proto == IPPROTO_UDPLITE ? "Lite" : "", 1829 &saddr, ntohs(uh->source), 1830 ulen, skb->len, 1831 &daddr, ntohs(uh->dest)); 1832 goto drop; 1833 1834 csum_error: 1835 /* 1836 * RFC1122: OK. Discards the bad packet silently (as far as 1837 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1838 */ 1839 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1840 proto == IPPROTO_UDPLITE ? "Lite" : "", 1841 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1842 ulen); 1843 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1844 drop: 1845 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1846 kfree_skb(skb); 1847 return 0; 1848 } 1849 1850 /* We can only early demux multicast if there is a single matching socket. 1851 * If more than one socket found returns NULL 1852 */ 1853 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1854 __be16 loc_port, __be32 loc_addr, 1855 __be16 rmt_port, __be32 rmt_addr, 1856 int dif) 1857 { 1858 struct sock *sk, *result; 1859 struct hlist_nulls_node *node; 1860 unsigned short hnum = ntohs(loc_port); 1861 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1862 struct udp_hslot *hslot = &udp_table.hash[slot]; 1863 1864 /* Do not bother scanning a too big list */ 1865 if (hslot->count > 10) 1866 return NULL; 1867 1868 rcu_read_lock(); 1869 begin: 1870 count = 0; 1871 result = NULL; 1872 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1873 if (__udp_is_mcast_sock(net, sk, 1874 loc_port, loc_addr, 1875 rmt_port, rmt_addr, 1876 dif, hnum)) { 1877 result = sk; 1878 ++count; 1879 } 1880 } 1881 /* 1882 * if the nulls value we got at the end of this lookup is 1883 * not the expected one, we must restart lookup. 1884 * We probably met an item that was moved to another chain. 1885 */ 1886 if (get_nulls_value(node) != slot) 1887 goto begin; 1888 1889 if (result) { 1890 if (count != 1 || 1891 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1892 result = NULL; 1893 else if (unlikely(!__udp_is_mcast_sock(net, result, 1894 loc_port, loc_addr, 1895 rmt_port, rmt_addr, 1896 dif, hnum))) { 1897 sock_put(result); 1898 result = NULL; 1899 } 1900 } 1901 rcu_read_unlock(); 1902 return result; 1903 } 1904 1905 /* For unicast we should only early demux connected sockets or we can 1906 * break forwarding setups. The chains here can be long so only check 1907 * if the first socket is an exact match and if not move on. 1908 */ 1909 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1910 __be16 loc_port, __be32 loc_addr, 1911 __be16 rmt_port, __be32 rmt_addr, 1912 int dif) 1913 { 1914 struct sock *sk, *result; 1915 struct hlist_nulls_node *node; 1916 unsigned short hnum = ntohs(loc_port); 1917 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1918 unsigned int slot2 = hash2 & udp_table.mask; 1919 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1920 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 1921 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1922 1923 rcu_read_lock(); 1924 result = NULL; 1925 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 1926 if (INET_MATCH(sk, net, acookie, 1927 rmt_addr, loc_addr, ports, dif)) 1928 result = sk; 1929 /* Only check first socket in chain */ 1930 break; 1931 } 1932 1933 if (result) { 1934 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1935 result = NULL; 1936 else if (unlikely(!INET_MATCH(sk, net, acookie, 1937 rmt_addr, loc_addr, 1938 ports, dif))) { 1939 sock_put(result); 1940 result = NULL; 1941 } 1942 } 1943 rcu_read_unlock(); 1944 return result; 1945 } 1946 1947 void udp_v4_early_demux(struct sk_buff *skb) 1948 { 1949 struct net *net = dev_net(skb->dev); 1950 const struct iphdr *iph; 1951 const struct udphdr *uh; 1952 struct sock *sk; 1953 struct dst_entry *dst; 1954 int dif = skb->dev->ifindex; 1955 1956 /* validate the packet */ 1957 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1958 return; 1959 1960 iph = ip_hdr(skb); 1961 uh = udp_hdr(skb); 1962 1963 if (skb->pkt_type == PACKET_BROADCAST || 1964 skb->pkt_type == PACKET_MULTICAST) 1965 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1966 uh->source, iph->saddr, dif); 1967 else if (skb->pkt_type == PACKET_HOST) 1968 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1969 uh->source, iph->saddr, dif); 1970 else 1971 return; 1972 1973 if (!sk) 1974 return; 1975 1976 skb->sk = sk; 1977 skb->destructor = sock_edemux; 1978 dst = sk->sk_rx_dst; 1979 1980 if (dst) 1981 dst = dst_check(dst, 0); 1982 if (dst) 1983 skb_dst_set_noref(skb, dst); 1984 } 1985 1986 int udp_rcv(struct sk_buff *skb) 1987 { 1988 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1989 } 1990 1991 void udp_destroy_sock(struct sock *sk) 1992 { 1993 struct udp_sock *up = udp_sk(sk); 1994 bool slow = lock_sock_fast(sk); 1995 udp_flush_pending_frames(sk); 1996 unlock_sock_fast(sk, slow); 1997 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1998 void (*encap_destroy)(struct sock *sk); 1999 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2000 if (encap_destroy) 2001 encap_destroy(sk); 2002 } 2003 } 2004 2005 /* 2006 * Socket option code for UDP 2007 */ 2008 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2009 char __user *optval, unsigned int optlen, 2010 int (*push_pending_frames)(struct sock *)) 2011 { 2012 struct udp_sock *up = udp_sk(sk); 2013 int val, valbool; 2014 int err = 0; 2015 int is_udplite = IS_UDPLITE(sk); 2016 2017 if (optlen < sizeof(int)) 2018 return -EINVAL; 2019 2020 if (get_user(val, (int __user *)optval)) 2021 return -EFAULT; 2022 2023 valbool = val ? 1 : 0; 2024 2025 switch (optname) { 2026 case UDP_CORK: 2027 if (val != 0) { 2028 up->corkflag = 1; 2029 } else { 2030 up->corkflag = 0; 2031 lock_sock(sk); 2032 (*push_pending_frames)(sk); 2033 release_sock(sk); 2034 } 2035 break; 2036 2037 case UDP_ENCAP: 2038 switch (val) { 2039 case 0: 2040 case UDP_ENCAP_ESPINUDP: 2041 case UDP_ENCAP_ESPINUDP_NON_IKE: 2042 up->encap_rcv = xfrm4_udp_encap_rcv; 2043 /* FALLTHROUGH */ 2044 case UDP_ENCAP_L2TPINUDP: 2045 up->encap_type = val; 2046 udp_encap_enable(); 2047 break; 2048 default: 2049 err = -ENOPROTOOPT; 2050 break; 2051 } 2052 break; 2053 2054 case UDP_NO_CHECK6_TX: 2055 up->no_check6_tx = valbool; 2056 break; 2057 2058 case UDP_NO_CHECK6_RX: 2059 up->no_check6_rx = valbool; 2060 break; 2061 2062 /* 2063 * UDP-Lite's partial checksum coverage (RFC 3828). 2064 */ 2065 /* The sender sets actual checksum coverage length via this option. 2066 * The case coverage > packet length is handled by send module. */ 2067 case UDPLITE_SEND_CSCOV: 2068 if (!is_udplite) /* Disable the option on UDP sockets */ 2069 return -ENOPROTOOPT; 2070 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2071 val = 8; 2072 else if (val > USHRT_MAX) 2073 val = USHRT_MAX; 2074 up->pcslen = val; 2075 up->pcflag |= UDPLITE_SEND_CC; 2076 break; 2077 2078 /* The receiver specifies a minimum checksum coverage value. To make 2079 * sense, this should be set to at least 8 (as done below). If zero is 2080 * used, this again means full checksum coverage. */ 2081 case UDPLITE_RECV_CSCOV: 2082 if (!is_udplite) /* Disable the option on UDP sockets */ 2083 return -ENOPROTOOPT; 2084 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2085 val = 8; 2086 else if (val > USHRT_MAX) 2087 val = USHRT_MAX; 2088 up->pcrlen = val; 2089 up->pcflag |= UDPLITE_RECV_CC; 2090 break; 2091 2092 default: 2093 err = -ENOPROTOOPT; 2094 break; 2095 } 2096 2097 return err; 2098 } 2099 EXPORT_SYMBOL(udp_lib_setsockopt); 2100 2101 int udp_setsockopt(struct sock *sk, int level, int optname, 2102 char __user *optval, unsigned int optlen) 2103 { 2104 if (level == SOL_UDP || level == SOL_UDPLITE) 2105 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2106 udp_push_pending_frames); 2107 return ip_setsockopt(sk, level, optname, optval, optlen); 2108 } 2109 2110 #ifdef CONFIG_COMPAT 2111 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2112 char __user *optval, unsigned int optlen) 2113 { 2114 if (level == SOL_UDP || level == SOL_UDPLITE) 2115 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2116 udp_push_pending_frames); 2117 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2118 } 2119 #endif 2120 2121 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2122 char __user *optval, int __user *optlen) 2123 { 2124 struct udp_sock *up = udp_sk(sk); 2125 int val, len; 2126 2127 if (get_user(len, optlen)) 2128 return -EFAULT; 2129 2130 len = min_t(unsigned int, len, sizeof(int)); 2131 2132 if (len < 0) 2133 return -EINVAL; 2134 2135 switch (optname) { 2136 case UDP_CORK: 2137 val = up->corkflag; 2138 break; 2139 2140 case UDP_ENCAP: 2141 val = up->encap_type; 2142 break; 2143 2144 case UDP_NO_CHECK6_TX: 2145 val = up->no_check6_tx; 2146 break; 2147 2148 case UDP_NO_CHECK6_RX: 2149 val = up->no_check6_rx; 2150 break; 2151 2152 /* The following two cannot be changed on UDP sockets, the return is 2153 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2154 case UDPLITE_SEND_CSCOV: 2155 val = up->pcslen; 2156 break; 2157 2158 case UDPLITE_RECV_CSCOV: 2159 val = up->pcrlen; 2160 break; 2161 2162 default: 2163 return -ENOPROTOOPT; 2164 } 2165 2166 if (put_user(len, optlen)) 2167 return -EFAULT; 2168 if (copy_to_user(optval, &val, len)) 2169 return -EFAULT; 2170 return 0; 2171 } 2172 EXPORT_SYMBOL(udp_lib_getsockopt); 2173 2174 int udp_getsockopt(struct sock *sk, int level, int optname, 2175 char __user *optval, int __user *optlen) 2176 { 2177 if (level == SOL_UDP || level == SOL_UDPLITE) 2178 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2179 return ip_getsockopt(sk, level, optname, optval, optlen); 2180 } 2181 2182 #ifdef CONFIG_COMPAT 2183 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2184 char __user *optval, int __user *optlen) 2185 { 2186 if (level == SOL_UDP || level == SOL_UDPLITE) 2187 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2188 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2189 } 2190 #endif 2191 /** 2192 * udp_poll - wait for a UDP event. 2193 * @file - file struct 2194 * @sock - socket 2195 * @wait - poll table 2196 * 2197 * This is same as datagram poll, except for the special case of 2198 * blocking sockets. If application is using a blocking fd 2199 * and a packet with checksum error is in the queue; 2200 * then it could get return from select indicating data available 2201 * but then block when reading it. Add special case code 2202 * to work around these arguably broken applications. 2203 */ 2204 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2205 { 2206 unsigned int mask = datagram_poll(file, sock, wait); 2207 struct sock *sk = sock->sk; 2208 2209 sock_rps_record_flow(sk); 2210 2211 /* Check for false positives due to checksum errors */ 2212 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2213 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2214 mask &= ~(POLLIN | POLLRDNORM); 2215 2216 return mask; 2217 2218 } 2219 EXPORT_SYMBOL(udp_poll); 2220 2221 struct proto udp_prot = { 2222 .name = "UDP", 2223 .owner = THIS_MODULE, 2224 .close = udp_lib_close, 2225 .connect = ip4_datagram_connect, 2226 .disconnect = udp_disconnect, 2227 .ioctl = udp_ioctl, 2228 .destroy = udp_destroy_sock, 2229 .setsockopt = udp_setsockopt, 2230 .getsockopt = udp_getsockopt, 2231 .sendmsg = udp_sendmsg, 2232 .recvmsg = udp_recvmsg, 2233 .sendpage = udp_sendpage, 2234 .backlog_rcv = __udp_queue_rcv_skb, 2235 .release_cb = ip4_datagram_release_cb, 2236 .hash = udp_lib_hash, 2237 .unhash = udp_lib_unhash, 2238 .rehash = udp_v4_rehash, 2239 .get_port = udp_v4_get_port, 2240 .memory_allocated = &udp_memory_allocated, 2241 .sysctl_mem = sysctl_udp_mem, 2242 .sysctl_wmem = &sysctl_udp_wmem_min, 2243 .sysctl_rmem = &sysctl_udp_rmem_min, 2244 .obj_size = sizeof(struct udp_sock), 2245 .slab_flags = SLAB_DESTROY_BY_RCU, 2246 .h.udp_table = &udp_table, 2247 #ifdef CONFIG_COMPAT 2248 .compat_setsockopt = compat_udp_setsockopt, 2249 .compat_getsockopt = compat_udp_getsockopt, 2250 #endif 2251 .clear_sk = sk_prot_clear_portaddr_nulls, 2252 }; 2253 EXPORT_SYMBOL(udp_prot); 2254 2255 /* ------------------------------------------------------------------------ */ 2256 #ifdef CONFIG_PROC_FS 2257 2258 static struct sock *udp_get_first(struct seq_file *seq, int start) 2259 { 2260 struct sock *sk; 2261 struct udp_iter_state *state = seq->private; 2262 struct net *net = seq_file_net(seq); 2263 2264 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2265 ++state->bucket) { 2266 struct hlist_nulls_node *node; 2267 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2268 2269 if (hlist_nulls_empty(&hslot->head)) 2270 continue; 2271 2272 spin_lock_bh(&hslot->lock); 2273 sk_nulls_for_each(sk, node, &hslot->head) { 2274 if (!net_eq(sock_net(sk), net)) 2275 continue; 2276 if (sk->sk_family == state->family) 2277 goto found; 2278 } 2279 spin_unlock_bh(&hslot->lock); 2280 } 2281 sk = NULL; 2282 found: 2283 return sk; 2284 } 2285 2286 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2287 { 2288 struct udp_iter_state *state = seq->private; 2289 struct net *net = seq_file_net(seq); 2290 2291 do { 2292 sk = sk_nulls_next(sk); 2293 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2294 2295 if (!sk) { 2296 if (state->bucket <= state->udp_table->mask) 2297 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2298 return udp_get_first(seq, state->bucket + 1); 2299 } 2300 return sk; 2301 } 2302 2303 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2304 { 2305 struct sock *sk = udp_get_first(seq, 0); 2306 2307 if (sk) 2308 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2309 --pos; 2310 return pos ? NULL : sk; 2311 } 2312 2313 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2314 { 2315 struct udp_iter_state *state = seq->private; 2316 state->bucket = MAX_UDP_PORTS; 2317 2318 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2319 } 2320 2321 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2322 { 2323 struct sock *sk; 2324 2325 if (v == SEQ_START_TOKEN) 2326 sk = udp_get_idx(seq, 0); 2327 else 2328 sk = udp_get_next(seq, v); 2329 2330 ++*pos; 2331 return sk; 2332 } 2333 2334 static void udp_seq_stop(struct seq_file *seq, void *v) 2335 { 2336 struct udp_iter_state *state = seq->private; 2337 2338 if (state->bucket <= state->udp_table->mask) 2339 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2340 } 2341 2342 int udp_seq_open(struct inode *inode, struct file *file) 2343 { 2344 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2345 struct udp_iter_state *s; 2346 int err; 2347 2348 err = seq_open_net(inode, file, &afinfo->seq_ops, 2349 sizeof(struct udp_iter_state)); 2350 if (err < 0) 2351 return err; 2352 2353 s = ((struct seq_file *)file->private_data)->private; 2354 s->family = afinfo->family; 2355 s->udp_table = afinfo->udp_table; 2356 return err; 2357 } 2358 EXPORT_SYMBOL(udp_seq_open); 2359 2360 /* ------------------------------------------------------------------------ */ 2361 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2362 { 2363 struct proc_dir_entry *p; 2364 int rc = 0; 2365 2366 afinfo->seq_ops.start = udp_seq_start; 2367 afinfo->seq_ops.next = udp_seq_next; 2368 afinfo->seq_ops.stop = udp_seq_stop; 2369 2370 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2371 afinfo->seq_fops, afinfo); 2372 if (!p) 2373 rc = -ENOMEM; 2374 return rc; 2375 } 2376 EXPORT_SYMBOL(udp_proc_register); 2377 2378 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2379 { 2380 remove_proc_entry(afinfo->name, net->proc_net); 2381 } 2382 EXPORT_SYMBOL(udp_proc_unregister); 2383 2384 /* ------------------------------------------------------------------------ */ 2385 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2386 int bucket) 2387 { 2388 struct inet_sock *inet = inet_sk(sp); 2389 __be32 dest = inet->inet_daddr; 2390 __be32 src = inet->inet_rcv_saddr; 2391 __u16 destp = ntohs(inet->inet_dport); 2392 __u16 srcp = ntohs(inet->inet_sport); 2393 2394 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2395 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2396 bucket, src, srcp, dest, destp, sp->sk_state, 2397 sk_wmem_alloc_get(sp), 2398 sk_rmem_alloc_get(sp), 2399 0, 0L, 0, 2400 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2401 0, sock_i_ino(sp), 2402 atomic_read(&sp->sk_refcnt), sp, 2403 atomic_read(&sp->sk_drops)); 2404 } 2405 2406 int udp4_seq_show(struct seq_file *seq, void *v) 2407 { 2408 seq_setwidth(seq, 127); 2409 if (v == SEQ_START_TOKEN) 2410 seq_puts(seq, " sl local_address rem_address st tx_queue " 2411 "rx_queue tr tm->when retrnsmt uid timeout " 2412 "inode ref pointer drops"); 2413 else { 2414 struct udp_iter_state *state = seq->private; 2415 2416 udp4_format_sock(v, seq, state->bucket); 2417 } 2418 seq_pad(seq, '\n'); 2419 return 0; 2420 } 2421 2422 static const struct file_operations udp_afinfo_seq_fops = { 2423 .owner = THIS_MODULE, 2424 .open = udp_seq_open, 2425 .read = seq_read, 2426 .llseek = seq_lseek, 2427 .release = seq_release_net 2428 }; 2429 2430 /* ------------------------------------------------------------------------ */ 2431 static struct udp_seq_afinfo udp4_seq_afinfo = { 2432 .name = "udp", 2433 .family = AF_INET, 2434 .udp_table = &udp_table, 2435 .seq_fops = &udp_afinfo_seq_fops, 2436 .seq_ops = { 2437 .show = udp4_seq_show, 2438 }, 2439 }; 2440 2441 static int __net_init udp4_proc_init_net(struct net *net) 2442 { 2443 return udp_proc_register(net, &udp4_seq_afinfo); 2444 } 2445 2446 static void __net_exit udp4_proc_exit_net(struct net *net) 2447 { 2448 udp_proc_unregister(net, &udp4_seq_afinfo); 2449 } 2450 2451 static struct pernet_operations udp4_net_ops = { 2452 .init = udp4_proc_init_net, 2453 .exit = udp4_proc_exit_net, 2454 }; 2455 2456 int __init udp4_proc_init(void) 2457 { 2458 return register_pernet_subsys(&udp4_net_ops); 2459 } 2460 2461 void udp4_proc_exit(void) 2462 { 2463 unregister_pernet_subsys(&udp4_net_ops); 2464 } 2465 #endif /* CONFIG_PROC_FS */ 2466 2467 static __initdata unsigned long uhash_entries; 2468 static int __init set_uhash_entries(char *str) 2469 { 2470 ssize_t ret; 2471 2472 if (!str) 2473 return 0; 2474 2475 ret = kstrtoul(str, 0, &uhash_entries); 2476 if (ret) 2477 return 0; 2478 2479 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2480 uhash_entries = UDP_HTABLE_SIZE_MIN; 2481 return 1; 2482 } 2483 __setup("uhash_entries=", set_uhash_entries); 2484 2485 void __init udp_table_init(struct udp_table *table, const char *name) 2486 { 2487 unsigned int i; 2488 2489 table->hash = alloc_large_system_hash(name, 2490 2 * sizeof(struct udp_hslot), 2491 uhash_entries, 2492 21, /* one slot per 2 MB */ 2493 0, 2494 &table->log, 2495 &table->mask, 2496 UDP_HTABLE_SIZE_MIN, 2497 64 * 1024); 2498 2499 table->hash2 = table->hash + (table->mask + 1); 2500 for (i = 0; i <= table->mask; i++) { 2501 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2502 table->hash[i].count = 0; 2503 spin_lock_init(&table->hash[i].lock); 2504 } 2505 for (i = 0; i <= table->mask; i++) { 2506 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2507 table->hash2[i].count = 0; 2508 spin_lock_init(&table->hash2[i].lock); 2509 } 2510 } 2511 2512 void __init udp_init(void) 2513 { 2514 unsigned long limit; 2515 2516 udp_table_init(&udp_table, "UDP"); 2517 limit = nr_free_buffer_pages() / 8; 2518 limit = max(limit, 128UL); 2519 sysctl_udp_mem[0] = limit / 4 * 3; 2520 sysctl_udp_mem[1] = limit; 2521 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2522 2523 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2524 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2525 } 2526 2527 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb, 2528 netdev_features_t features) 2529 { 2530 struct sk_buff *segs = ERR_PTR(-EINVAL); 2531 u16 mac_offset = skb->mac_header; 2532 int mac_len = skb->mac_len; 2533 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb); 2534 __be16 protocol = skb->protocol; 2535 netdev_features_t enc_features; 2536 int udp_offset, outer_hlen; 2537 unsigned int oldlen; 2538 bool need_csum; 2539 2540 oldlen = (u16)~skb->len; 2541 2542 if (unlikely(!pskb_may_pull(skb, tnl_hlen))) 2543 goto out; 2544 2545 skb->encapsulation = 0; 2546 __skb_pull(skb, tnl_hlen); 2547 skb_reset_mac_header(skb); 2548 skb_set_network_header(skb, skb_inner_network_offset(skb)); 2549 skb->mac_len = skb_inner_network_offset(skb); 2550 skb->protocol = htons(ETH_P_TEB); 2551 2552 need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM); 2553 if (need_csum) 2554 skb->encap_hdr_csum = 1; 2555 2556 /* segment inner packet. */ 2557 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb); 2558 segs = skb_mac_gso_segment(skb, enc_features); 2559 if (!segs || IS_ERR(segs)) { 2560 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset, 2561 mac_len); 2562 goto out; 2563 } 2564 2565 outer_hlen = skb_tnl_header_len(skb); 2566 udp_offset = outer_hlen - tnl_hlen; 2567 skb = segs; 2568 do { 2569 struct udphdr *uh; 2570 int len; 2571 2572 skb_reset_inner_headers(skb); 2573 skb->encapsulation = 1; 2574 2575 skb->mac_len = mac_len; 2576 2577 skb_push(skb, outer_hlen); 2578 skb_reset_mac_header(skb); 2579 skb_set_network_header(skb, mac_len); 2580 skb_set_transport_header(skb, udp_offset); 2581 len = skb->len - udp_offset; 2582 uh = udp_hdr(skb); 2583 uh->len = htons(len); 2584 2585 if (need_csum) { 2586 __be32 delta = htonl(oldlen + len); 2587 2588 uh->check = ~csum_fold((__force __wsum) 2589 ((__force u32)uh->check + 2590 (__force u32)delta)); 2591 uh->check = gso_make_checksum(skb, ~uh->check); 2592 2593 if (uh->check == 0) 2594 uh->check = CSUM_MANGLED_0; 2595 } 2596 2597 skb->protocol = protocol; 2598 } while ((skb = skb->next)); 2599 out: 2600 return segs; 2601 } 2602