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_reserved_local_port(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 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 731 int offset = skb_transport_offset(skb); 732 int len = skb->len - offset; 733 int hlen = len; 734 __wsum csum = 0; 735 736 if (!frags) { 737 /* 738 * Only one fragment on the socket. 739 */ 740 skb->csum_start = skb_transport_header(skb) - skb->head; 741 skb->csum_offset = offsetof(struct udphdr, check); 742 uh->check = ~csum_tcpudp_magic(src, dst, len, 743 IPPROTO_UDP, 0); 744 } else { 745 /* 746 * HW-checksum won't work as there are two or more 747 * fragments on the socket so that all csums of sk_buffs 748 * should be together 749 */ 750 do { 751 csum = csum_add(csum, frags->csum); 752 hlen -= frags->len; 753 } while ((frags = frags->next)); 754 755 csum = skb_checksum(skb, offset, hlen, csum); 756 skb->ip_summed = CHECKSUM_NONE; 757 758 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 759 if (uh->check == 0) 760 uh->check = CSUM_MANGLED_0; 761 } 762 } 763 EXPORT_SYMBOL_GPL(udp4_hwcsum); 764 765 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 766 { 767 struct sock *sk = skb->sk; 768 struct inet_sock *inet = inet_sk(sk); 769 struct udphdr *uh; 770 int err = 0; 771 int is_udplite = IS_UDPLITE(sk); 772 int offset = skb_transport_offset(skb); 773 int len = skb->len - offset; 774 __wsum csum = 0; 775 776 /* 777 * Create a UDP header 778 */ 779 uh = udp_hdr(skb); 780 uh->source = inet->inet_sport; 781 uh->dest = fl4->fl4_dport; 782 uh->len = htons(len); 783 uh->check = 0; 784 785 if (is_udplite) /* UDP-Lite */ 786 csum = udplite_csum(skb); 787 788 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 789 790 skb->ip_summed = CHECKSUM_NONE; 791 goto send; 792 793 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 794 795 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 796 goto send; 797 798 } else 799 csum = udp_csum(skb); 800 801 /* add protocol-dependent pseudo-header */ 802 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 803 sk->sk_protocol, csum); 804 if (uh->check == 0) 805 uh->check = CSUM_MANGLED_0; 806 807 send: 808 err = ip_send_skb(sock_net(sk), skb); 809 if (err) { 810 if (err == -ENOBUFS && !inet->recverr) { 811 UDP_INC_STATS_USER(sock_net(sk), 812 UDP_MIB_SNDBUFERRORS, is_udplite); 813 err = 0; 814 } 815 } else 816 UDP_INC_STATS_USER(sock_net(sk), 817 UDP_MIB_OUTDATAGRAMS, is_udplite); 818 return err; 819 } 820 821 /* 822 * Push out all pending data as one UDP datagram. Socket is locked. 823 */ 824 int udp_push_pending_frames(struct sock *sk) 825 { 826 struct udp_sock *up = udp_sk(sk); 827 struct inet_sock *inet = inet_sk(sk); 828 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 829 struct sk_buff *skb; 830 int err = 0; 831 832 skb = ip_finish_skb(sk, fl4); 833 if (!skb) 834 goto out; 835 836 err = udp_send_skb(skb, fl4); 837 838 out: 839 up->len = 0; 840 up->pending = 0; 841 return err; 842 } 843 EXPORT_SYMBOL(udp_push_pending_frames); 844 845 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 846 size_t len) 847 { 848 struct inet_sock *inet = inet_sk(sk); 849 struct udp_sock *up = udp_sk(sk); 850 struct flowi4 fl4_stack; 851 struct flowi4 *fl4; 852 int ulen = len; 853 struct ipcm_cookie ipc; 854 struct rtable *rt = NULL; 855 int free = 0; 856 int connected = 0; 857 __be32 daddr, faddr, saddr; 858 __be16 dport; 859 u8 tos; 860 int err, is_udplite = IS_UDPLITE(sk); 861 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 862 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 863 struct sk_buff *skb; 864 struct ip_options_data opt_copy; 865 866 if (len > 0xFFFF) 867 return -EMSGSIZE; 868 869 /* 870 * Check the flags. 871 */ 872 873 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 874 return -EOPNOTSUPP; 875 876 ipc.opt = NULL; 877 ipc.tx_flags = 0; 878 ipc.ttl = 0; 879 ipc.tos = -1; 880 881 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 882 883 fl4 = &inet->cork.fl.u.ip4; 884 if (up->pending) { 885 /* 886 * There are pending frames. 887 * The socket lock must be held while it's corked. 888 */ 889 lock_sock(sk); 890 if (likely(up->pending)) { 891 if (unlikely(up->pending != AF_INET)) { 892 release_sock(sk); 893 return -EINVAL; 894 } 895 goto do_append_data; 896 } 897 release_sock(sk); 898 } 899 ulen += sizeof(struct udphdr); 900 901 /* 902 * Get and verify the address. 903 */ 904 if (msg->msg_name) { 905 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 906 if (msg->msg_namelen < sizeof(*usin)) 907 return -EINVAL; 908 if (usin->sin_family != AF_INET) { 909 if (usin->sin_family != AF_UNSPEC) 910 return -EAFNOSUPPORT; 911 } 912 913 daddr = usin->sin_addr.s_addr; 914 dport = usin->sin_port; 915 if (dport == 0) 916 return -EINVAL; 917 } else { 918 if (sk->sk_state != TCP_ESTABLISHED) 919 return -EDESTADDRREQ; 920 daddr = inet->inet_daddr; 921 dport = inet->inet_dport; 922 /* Open fast path for connected socket. 923 Route will not be used, if at least one option is set. 924 */ 925 connected = 1; 926 } 927 ipc.addr = inet->inet_saddr; 928 929 ipc.oif = sk->sk_bound_dev_if; 930 931 sock_tx_timestamp(sk, &ipc.tx_flags); 932 933 if (msg->msg_controllen) { 934 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 935 sk->sk_family == AF_INET6); 936 if (err) 937 return err; 938 if (ipc.opt) 939 free = 1; 940 connected = 0; 941 } 942 if (!ipc.opt) { 943 struct ip_options_rcu *inet_opt; 944 945 rcu_read_lock(); 946 inet_opt = rcu_dereference(inet->inet_opt); 947 if (inet_opt) { 948 memcpy(&opt_copy, inet_opt, 949 sizeof(*inet_opt) + inet_opt->opt.optlen); 950 ipc.opt = &opt_copy.opt; 951 } 952 rcu_read_unlock(); 953 } 954 955 saddr = ipc.addr; 956 ipc.addr = faddr = daddr; 957 958 if (ipc.opt && ipc.opt->opt.srr) { 959 if (!daddr) 960 return -EINVAL; 961 faddr = ipc.opt->opt.faddr; 962 connected = 0; 963 } 964 tos = get_rttos(&ipc, inet); 965 if (sock_flag(sk, SOCK_LOCALROUTE) || 966 (msg->msg_flags & MSG_DONTROUTE) || 967 (ipc.opt && ipc.opt->opt.is_strictroute)) { 968 tos |= RTO_ONLINK; 969 connected = 0; 970 } 971 972 if (ipv4_is_multicast(daddr)) { 973 if (!ipc.oif) 974 ipc.oif = inet->mc_index; 975 if (!saddr) 976 saddr = inet->mc_addr; 977 connected = 0; 978 } else if (!ipc.oif) 979 ipc.oif = inet->uc_index; 980 981 if (connected) 982 rt = (struct rtable *)sk_dst_check(sk, 0); 983 984 if (rt == NULL) { 985 struct net *net = sock_net(sk); 986 987 fl4 = &fl4_stack; 988 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 989 RT_SCOPE_UNIVERSE, sk->sk_protocol, 990 inet_sk_flowi_flags(sk), 991 faddr, saddr, dport, inet->inet_sport); 992 993 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 994 rt = ip_route_output_flow(net, fl4, sk); 995 if (IS_ERR(rt)) { 996 err = PTR_ERR(rt); 997 rt = NULL; 998 if (err == -ENETUNREACH) 999 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1000 goto out; 1001 } 1002 1003 err = -EACCES; 1004 if ((rt->rt_flags & RTCF_BROADCAST) && 1005 !sock_flag(sk, SOCK_BROADCAST)) 1006 goto out; 1007 if (connected) 1008 sk_dst_set(sk, dst_clone(&rt->dst)); 1009 } 1010 1011 if (msg->msg_flags&MSG_CONFIRM) 1012 goto do_confirm; 1013 back_from_confirm: 1014 1015 saddr = fl4->saddr; 1016 if (!ipc.addr) 1017 daddr = ipc.addr = fl4->daddr; 1018 1019 /* Lockless fast path for the non-corking case. */ 1020 if (!corkreq) { 1021 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen, 1022 sizeof(struct udphdr), &ipc, &rt, 1023 msg->msg_flags); 1024 err = PTR_ERR(skb); 1025 if (!IS_ERR_OR_NULL(skb)) 1026 err = udp_send_skb(skb, fl4); 1027 goto out; 1028 } 1029 1030 lock_sock(sk); 1031 if (unlikely(up->pending)) { 1032 /* The socket is already corked while preparing it. */ 1033 /* ... which is an evident application bug. --ANK */ 1034 release_sock(sk); 1035 1036 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n")); 1037 err = -EINVAL; 1038 goto out; 1039 } 1040 /* 1041 * Now cork the socket to pend data. 1042 */ 1043 fl4 = &inet->cork.fl.u.ip4; 1044 fl4->daddr = daddr; 1045 fl4->saddr = saddr; 1046 fl4->fl4_dport = dport; 1047 fl4->fl4_sport = inet->inet_sport; 1048 up->pending = AF_INET; 1049 1050 do_append_data: 1051 up->len += ulen; 1052 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen, 1053 sizeof(struct udphdr), &ipc, &rt, 1054 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1055 if (err) 1056 udp_flush_pending_frames(sk); 1057 else if (!corkreq) 1058 err = udp_push_pending_frames(sk); 1059 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1060 up->pending = 0; 1061 release_sock(sk); 1062 1063 out: 1064 ip_rt_put(rt); 1065 if (free) 1066 kfree(ipc.opt); 1067 if (!err) 1068 return len; 1069 /* 1070 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1071 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1072 * we don't have a good statistic (IpOutDiscards but it can be too many 1073 * things). We could add another new stat but at least for now that 1074 * seems like overkill. 1075 */ 1076 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1077 UDP_INC_STATS_USER(sock_net(sk), 1078 UDP_MIB_SNDBUFERRORS, is_udplite); 1079 } 1080 return err; 1081 1082 do_confirm: 1083 dst_confirm(&rt->dst); 1084 if (!(msg->msg_flags&MSG_PROBE) || len) 1085 goto back_from_confirm; 1086 err = 0; 1087 goto out; 1088 } 1089 EXPORT_SYMBOL(udp_sendmsg); 1090 1091 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1092 size_t size, int flags) 1093 { 1094 struct inet_sock *inet = inet_sk(sk); 1095 struct udp_sock *up = udp_sk(sk); 1096 int ret; 1097 1098 if (flags & MSG_SENDPAGE_NOTLAST) 1099 flags |= MSG_MORE; 1100 1101 if (!up->pending) { 1102 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1103 1104 /* Call udp_sendmsg to specify destination address which 1105 * sendpage interface can't pass. 1106 * This will succeed only when the socket is connected. 1107 */ 1108 ret = udp_sendmsg(NULL, sk, &msg, 0); 1109 if (ret < 0) 1110 return ret; 1111 } 1112 1113 lock_sock(sk); 1114 1115 if (unlikely(!up->pending)) { 1116 release_sock(sk); 1117 1118 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n")); 1119 return -EINVAL; 1120 } 1121 1122 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1123 page, offset, size, flags); 1124 if (ret == -EOPNOTSUPP) { 1125 release_sock(sk); 1126 return sock_no_sendpage(sk->sk_socket, page, offset, 1127 size, flags); 1128 } 1129 if (ret < 0) { 1130 udp_flush_pending_frames(sk); 1131 goto out; 1132 } 1133 1134 up->len += size; 1135 if (!(up->corkflag || (flags&MSG_MORE))) 1136 ret = udp_push_pending_frames(sk); 1137 if (!ret) 1138 ret = size; 1139 out: 1140 release_sock(sk); 1141 return ret; 1142 } 1143 1144 1145 /** 1146 * first_packet_length - return length of first packet in receive queue 1147 * @sk: socket 1148 * 1149 * Drops all bad checksum frames, until a valid one is found. 1150 * Returns the length of found skb, or 0 if none is found. 1151 */ 1152 static unsigned int first_packet_length(struct sock *sk) 1153 { 1154 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1155 struct sk_buff *skb; 1156 unsigned int res; 1157 1158 __skb_queue_head_init(&list_kill); 1159 1160 spin_lock_bh(&rcvq->lock); 1161 while ((skb = skb_peek(rcvq)) != NULL && 1162 udp_lib_checksum_complete(skb)) { 1163 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1164 IS_UDPLITE(sk)); 1165 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1166 IS_UDPLITE(sk)); 1167 atomic_inc(&sk->sk_drops); 1168 __skb_unlink(skb, rcvq); 1169 __skb_queue_tail(&list_kill, skb); 1170 } 1171 res = skb ? skb->len : 0; 1172 spin_unlock_bh(&rcvq->lock); 1173 1174 if (!skb_queue_empty(&list_kill)) { 1175 bool slow = lock_sock_fast(sk); 1176 1177 __skb_queue_purge(&list_kill); 1178 sk_mem_reclaim_partial(sk); 1179 unlock_sock_fast(sk, slow); 1180 } 1181 return res; 1182 } 1183 1184 /* 1185 * IOCTL requests applicable to the UDP protocol 1186 */ 1187 1188 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1189 { 1190 switch (cmd) { 1191 case SIOCOUTQ: 1192 { 1193 int amount = sk_wmem_alloc_get(sk); 1194 1195 return put_user(amount, (int __user *)arg); 1196 } 1197 1198 case SIOCINQ: 1199 { 1200 unsigned int amount = first_packet_length(sk); 1201 1202 if (amount) 1203 /* 1204 * We will only return the amount 1205 * of this packet since that is all 1206 * that will be read. 1207 */ 1208 amount -= sizeof(struct udphdr); 1209 1210 return put_user(amount, (int __user *)arg); 1211 } 1212 1213 default: 1214 return -ENOIOCTLCMD; 1215 } 1216 1217 return 0; 1218 } 1219 EXPORT_SYMBOL(udp_ioctl); 1220 1221 /* 1222 * This should be easy, if there is something there we 1223 * return it, otherwise we block. 1224 */ 1225 1226 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1227 size_t len, int noblock, int flags, int *addr_len) 1228 { 1229 struct inet_sock *inet = inet_sk(sk); 1230 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1231 struct sk_buff *skb; 1232 unsigned int ulen, copied; 1233 int peeked, off = 0; 1234 int err; 1235 int is_udplite = IS_UDPLITE(sk); 1236 bool slow; 1237 1238 if (flags & MSG_ERRQUEUE) 1239 return ip_recv_error(sk, msg, len, addr_len); 1240 1241 try_again: 1242 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1243 &peeked, &off, &err); 1244 if (!skb) 1245 goto out; 1246 1247 ulen = skb->len - sizeof(struct udphdr); 1248 copied = len; 1249 if (copied > ulen) 1250 copied = ulen; 1251 else if (copied < ulen) 1252 msg->msg_flags |= MSG_TRUNC; 1253 1254 /* 1255 * If checksum is needed at all, try to do it while copying the 1256 * data. If the data is truncated, or if we only want a partial 1257 * coverage checksum (UDP-Lite), do it before the copy. 1258 */ 1259 1260 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1261 if (udp_lib_checksum_complete(skb)) 1262 goto csum_copy_err; 1263 } 1264 1265 if (skb_csum_unnecessary(skb)) 1266 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1267 msg->msg_iov, copied); 1268 else { 1269 err = skb_copy_and_csum_datagram_iovec(skb, 1270 sizeof(struct udphdr), 1271 msg->msg_iov); 1272 1273 if (err == -EINVAL) 1274 goto csum_copy_err; 1275 } 1276 1277 if (unlikely(err)) { 1278 trace_kfree_skb(skb, udp_recvmsg); 1279 if (!peeked) { 1280 atomic_inc(&sk->sk_drops); 1281 UDP_INC_STATS_USER(sock_net(sk), 1282 UDP_MIB_INERRORS, is_udplite); 1283 } 1284 goto out_free; 1285 } 1286 1287 if (!peeked) 1288 UDP_INC_STATS_USER(sock_net(sk), 1289 UDP_MIB_INDATAGRAMS, is_udplite); 1290 1291 sock_recv_ts_and_drops(msg, sk, skb); 1292 1293 /* Copy the address. */ 1294 if (sin) { 1295 sin->sin_family = AF_INET; 1296 sin->sin_port = udp_hdr(skb)->source; 1297 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1298 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1299 *addr_len = sizeof(*sin); 1300 } 1301 if (inet->cmsg_flags) 1302 ip_cmsg_recv(msg, skb); 1303 1304 err = copied; 1305 if (flags & MSG_TRUNC) 1306 err = ulen; 1307 1308 out_free: 1309 skb_free_datagram_locked(sk, skb); 1310 out: 1311 return err; 1312 1313 csum_copy_err: 1314 slow = lock_sock_fast(sk); 1315 if (!skb_kill_datagram(sk, skb, flags)) { 1316 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1317 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1318 } 1319 unlock_sock_fast(sk, slow); 1320 1321 if (noblock) 1322 return -EAGAIN; 1323 1324 /* starting over for a new packet */ 1325 msg->msg_flags &= ~MSG_TRUNC; 1326 goto try_again; 1327 } 1328 1329 1330 int udp_disconnect(struct sock *sk, int flags) 1331 { 1332 struct inet_sock *inet = inet_sk(sk); 1333 /* 1334 * 1003.1g - break association. 1335 */ 1336 1337 sk->sk_state = TCP_CLOSE; 1338 inet->inet_daddr = 0; 1339 inet->inet_dport = 0; 1340 sock_rps_reset_rxhash(sk); 1341 sk->sk_bound_dev_if = 0; 1342 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1343 inet_reset_saddr(sk); 1344 1345 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1346 sk->sk_prot->unhash(sk); 1347 inet->inet_sport = 0; 1348 } 1349 sk_dst_reset(sk); 1350 return 0; 1351 } 1352 EXPORT_SYMBOL(udp_disconnect); 1353 1354 void udp_lib_unhash(struct sock *sk) 1355 { 1356 if (sk_hashed(sk)) { 1357 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1358 struct udp_hslot *hslot, *hslot2; 1359 1360 hslot = udp_hashslot(udptable, sock_net(sk), 1361 udp_sk(sk)->udp_port_hash); 1362 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1363 1364 spin_lock_bh(&hslot->lock); 1365 if (sk_nulls_del_node_init_rcu(sk)) { 1366 hslot->count--; 1367 inet_sk(sk)->inet_num = 0; 1368 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1369 1370 spin_lock(&hslot2->lock); 1371 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1372 hslot2->count--; 1373 spin_unlock(&hslot2->lock); 1374 } 1375 spin_unlock_bh(&hslot->lock); 1376 } 1377 } 1378 EXPORT_SYMBOL(udp_lib_unhash); 1379 1380 /* 1381 * inet_rcv_saddr was changed, we must rehash secondary hash 1382 */ 1383 void udp_lib_rehash(struct sock *sk, u16 newhash) 1384 { 1385 if (sk_hashed(sk)) { 1386 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1387 struct udp_hslot *hslot, *hslot2, *nhslot2; 1388 1389 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1390 nhslot2 = udp_hashslot2(udptable, newhash); 1391 udp_sk(sk)->udp_portaddr_hash = newhash; 1392 if (hslot2 != nhslot2) { 1393 hslot = udp_hashslot(udptable, sock_net(sk), 1394 udp_sk(sk)->udp_port_hash); 1395 /* we must lock primary chain too */ 1396 spin_lock_bh(&hslot->lock); 1397 1398 spin_lock(&hslot2->lock); 1399 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1400 hslot2->count--; 1401 spin_unlock(&hslot2->lock); 1402 1403 spin_lock(&nhslot2->lock); 1404 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1405 &nhslot2->head); 1406 nhslot2->count++; 1407 spin_unlock(&nhslot2->lock); 1408 1409 spin_unlock_bh(&hslot->lock); 1410 } 1411 } 1412 } 1413 EXPORT_SYMBOL(udp_lib_rehash); 1414 1415 static void udp_v4_rehash(struct sock *sk) 1416 { 1417 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1418 inet_sk(sk)->inet_rcv_saddr, 1419 inet_sk(sk)->inet_num); 1420 udp_lib_rehash(sk, new_hash); 1421 } 1422 1423 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1424 { 1425 int rc; 1426 1427 if (inet_sk(sk)->inet_daddr) { 1428 sock_rps_save_rxhash(sk, skb); 1429 sk_mark_napi_id(sk, skb); 1430 } 1431 1432 rc = sock_queue_rcv_skb(sk, skb); 1433 if (rc < 0) { 1434 int is_udplite = IS_UDPLITE(sk); 1435 1436 /* Note that an ENOMEM error is charged twice */ 1437 if (rc == -ENOMEM) 1438 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1439 is_udplite); 1440 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1441 kfree_skb(skb); 1442 trace_udp_fail_queue_rcv_skb(rc, sk); 1443 return -1; 1444 } 1445 1446 return 0; 1447 1448 } 1449 1450 static struct static_key udp_encap_needed __read_mostly; 1451 void udp_encap_enable(void) 1452 { 1453 if (!static_key_enabled(&udp_encap_needed)) 1454 static_key_slow_inc(&udp_encap_needed); 1455 } 1456 EXPORT_SYMBOL(udp_encap_enable); 1457 1458 /* returns: 1459 * -1: error 1460 * 0: success 1461 * >0: "udp encap" protocol resubmission 1462 * 1463 * Note that in the success and error cases, the skb is assumed to 1464 * have either been requeued or freed. 1465 */ 1466 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1467 { 1468 struct udp_sock *up = udp_sk(sk); 1469 int rc; 1470 int is_udplite = IS_UDPLITE(sk); 1471 1472 /* 1473 * Charge it to the socket, dropping if the queue is full. 1474 */ 1475 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1476 goto drop; 1477 nf_reset(skb); 1478 1479 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1480 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1481 1482 /* 1483 * This is an encapsulation socket so pass the skb to 1484 * the socket's udp_encap_rcv() hook. Otherwise, just 1485 * fall through and pass this up the UDP socket. 1486 * up->encap_rcv() returns the following value: 1487 * =0 if skb was successfully passed to the encap 1488 * handler or was discarded by it. 1489 * >0 if skb should be passed on to UDP. 1490 * <0 if skb should be resubmitted as proto -N 1491 */ 1492 1493 /* if we're overly short, let UDP handle it */ 1494 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1495 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { 1496 int ret; 1497 1498 ret = encap_rcv(sk, skb); 1499 if (ret <= 0) { 1500 UDP_INC_STATS_BH(sock_net(sk), 1501 UDP_MIB_INDATAGRAMS, 1502 is_udplite); 1503 return -ret; 1504 } 1505 } 1506 1507 /* FALLTHROUGH -- it's a UDP Packet */ 1508 } 1509 1510 /* 1511 * UDP-Lite specific tests, ignored on UDP sockets 1512 */ 1513 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1514 1515 /* 1516 * MIB statistics other than incrementing the error count are 1517 * disabled for the following two types of errors: these depend 1518 * on the application settings, not on the functioning of the 1519 * protocol stack as such. 1520 * 1521 * RFC 3828 here recommends (sec 3.3): "There should also be a 1522 * way ... to ... at least let the receiving application block 1523 * delivery of packets with coverage values less than a value 1524 * provided by the application." 1525 */ 1526 if (up->pcrlen == 0) { /* full coverage was set */ 1527 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n", 1528 UDP_SKB_CB(skb)->cscov, skb->len); 1529 goto drop; 1530 } 1531 /* The next case involves violating the min. coverage requested 1532 * by the receiver. This is subtle: if receiver wants x and x is 1533 * greater than the buffersize/MTU then receiver will complain 1534 * that it wants x while sender emits packets of smaller size y. 1535 * Therefore the above ...()->partial_cov statement is essential. 1536 */ 1537 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1538 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n", 1539 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1540 goto drop; 1541 } 1542 } 1543 1544 if (rcu_access_pointer(sk->sk_filter) && 1545 udp_lib_checksum_complete(skb)) 1546 goto csum_error; 1547 1548 1549 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) 1550 goto drop; 1551 1552 rc = 0; 1553 1554 ipv4_pktinfo_prepare(sk, skb); 1555 bh_lock_sock(sk); 1556 if (!sock_owned_by_user(sk)) 1557 rc = __udp_queue_rcv_skb(sk, skb); 1558 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1559 bh_unlock_sock(sk); 1560 goto drop; 1561 } 1562 bh_unlock_sock(sk); 1563 1564 return rc; 1565 1566 csum_error: 1567 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1568 drop: 1569 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1570 atomic_inc(&sk->sk_drops); 1571 kfree_skb(skb); 1572 return -1; 1573 } 1574 1575 1576 static void flush_stack(struct sock **stack, unsigned int count, 1577 struct sk_buff *skb, unsigned int final) 1578 { 1579 unsigned int i; 1580 struct sk_buff *skb1 = NULL; 1581 struct sock *sk; 1582 1583 for (i = 0; i < count; i++) { 1584 sk = stack[i]; 1585 if (likely(skb1 == NULL)) 1586 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1587 1588 if (!skb1) { 1589 atomic_inc(&sk->sk_drops); 1590 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1591 IS_UDPLITE(sk)); 1592 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1593 IS_UDPLITE(sk)); 1594 } 1595 1596 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1597 skb1 = NULL; 1598 } 1599 if (unlikely(skb1)) 1600 kfree_skb(skb1); 1601 } 1602 1603 /* For TCP sockets, sk_rx_dst is protected by socket lock 1604 * For UDP, we use xchg() to guard against concurrent changes. 1605 */ 1606 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1607 { 1608 struct dst_entry *old; 1609 1610 dst_hold(dst); 1611 old = xchg(&sk->sk_rx_dst, dst); 1612 dst_release(old); 1613 } 1614 1615 /* 1616 * Multicasts and broadcasts go to each listener. 1617 * 1618 * Note: called only from the BH handler context. 1619 */ 1620 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1621 struct udphdr *uh, 1622 __be32 saddr, __be32 daddr, 1623 struct udp_table *udptable) 1624 { 1625 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1626 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1627 int dif; 1628 unsigned int i, count = 0; 1629 1630 spin_lock(&hslot->lock); 1631 sk = sk_nulls_head(&hslot->head); 1632 dif = skb->dev->ifindex; 1633 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1634 while (sk) { 1635 stack[count++] = sk; 1636 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1637 daddr, uh->source, saddr, dif); 1638 if (unlikely(count == ARRAY_SIZE(stack))) { 1639 if (!sk) 1640 break; 1641 flush_stack(stack, count, skb, ~0); 1642 count = 0; 1643 } 1644 } 1645 /* 1646 * before releasing chain lock, we must take a reference on sockets 1647 */ 1648 for (i = 0; i < count; i++) 1649 sock_hold(stack[i]); 1650 1651 spin_unlock(&hslot->lock); 1652 1653 /* 1654 * do the slow work with no lock held 1655 */ 1656 if (count) { 1657 flush_stack(stack, count, skb, count - 1); 1658 1659 for (i = 0; i < count; i++) 1660 sock_put(stack[i]); 1661 } else { 1662 kfree_skb(skb); 1663 } 1664 return 0; 1665 } 1666 1667 /* Initialize UDP checksum. If exited with zero value (success), 1668 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1669 * Otherwise, csum completion requires chacksumming packet body, 1670 * including udp header and folding it to skb->csum. 1671 */ 1672 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1673 int proto) 1674 { 1675 const struct iphdr *iph; 1676 int err; 1677 1678 UDP_SKB_CB(skb)->partial_cov = 0; 1679 UDP_SKB_CB(skb)->cscov = skb->len; 1680 1681 if (proto == IPPROTO_UDPLITE) { 1682 err = udplite_checksum_init(skb, uh); 1683 if (err) 1684 return err; 1685 } 1686 1687 iph = ip_hdr(skb); 1688 if (uh->check == 0) { 1689 skb->ip_summed = CHECKSUM_UNNECESSARY; 1690 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1691 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1692 proto, skb->csum)) 1693 skb->ip_summed = CHECKSUM_UNNECESSARY; 1694 } 1695 if (!skb_csum_unnecessary(skb)) 1696 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1697 skb->len, proto, 0); 1698 /* Probably, we should checksum udp header (it should be in cache 1699 * in any case) and data in tiny packets (< rx copybreak). 1700 */ 1701 1702 return 0; 1703 } 1704 1705 /* 1706 * All we need to do is get the socket, and then do a checksum. 1707 */ 1708 1709 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1710 int proto) 1711 { 1712 struct sock *sk; 1713 struct udphdr *uh; 1714 unsigned short ulen; 1715 struct rtable *rt = skb_rtable(skb); 1716 __be32 saddr, daddr; 1717 struct net *net = dev_net(skb->dev); 1718 1719 /* 1720 * Validate the packet. 1721 */ 1722 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1723 goto drop; /* No space for header. */ 1724 1725 uh = udp_hdr(skb); 1726 ulen = ntohs(uh->len); 1727 saddr = ip_hdr(skb)->saddr; 1728 daddr = ip_hdr(skb)->daddr; 1729 1730 if (ulen > skb->len) 1731 goto short_packet; 1732 1733 if (proto == IPPROTO_UDP) { 1734 /* UDP validates ulen. */ 1735 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1736 goto short_packet; 1737 uh = udp_hdr(skb); 1738 } 1739 1740 if (udp4_csum_init(skb, uh, proto)) 1741 goto csum_error; 1742 1743 sk = skb_steal_sock(skb); 1744 if (sk) { 1745 struct dst_entry *dst = skb_dst(skb); 1746 int ret; 1747 1748 if (unlikely(sk->sk_rx_dst != dst)) 1749 udp_sk_rx_dst_set(sk, dst); 1750 1751 ret = udp_queue_rcv_skb(sk, skb); 1752 sock_put(sk); 1753 /* a return value > 0 means to resubmit the input, but 1754 * it wants the return to be -protocol, or 0 1755 */ 1756 if (ret > 0) 1757 return -ret; 1758 return 0; 1759 } else { 1760 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1761 return __udp4_lib_mcast_deliver(net, skb, uh, 1762 saddr, daddr, udptable); 1763 1764 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1765 } 1766 1767 if (sk != NULL) { 1768 int ret; 1769 1770 ret = udp_queue_rcv_skb(sk, skb); 1771 sock_put(sk); 1772 1773 /* a return value > 0 means to resubmit the input, but 1774 * it wants the return to be -protocol, or 0 1775 */ 1776 if (ret > 0) 1777 return -ret; 1778 return 0; 1779 } 1780 1781 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1782 goto drop; 1783 nf_reset(skb); 1784 1785 /* No socket. Drop packet silently, if checksum is wrong */ 1786 if (udp_lib_checksum_complete(skb)) 1787 goto csum_error; 1788 1789 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1790 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1791 1792 /* 1793 * Hmm. We got an UDP packet to a port to which we 1794 * don't wanna listen. Ignore it. 1795 */ 1796 kfree_skb(skb); 1797 return 0; 1798 1799 short_packet: 1800 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1801 proto == IPPROTO_UDPLITE ? "Lite" : "", 1802 &saddr, ntohs(uh->source), 1803 ulen, skb->len, 1804 &daddr, ntohs(uh->dest)); 1805 goto drop; 1806 1807 csum_error: 1808 /* 1809 * RFC1122: OK. Discards the bad packet silently (as far as 1810 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1811 */ 1812 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1813 proto == IPPROTO_UDPLITE ? "Lite" : "", 1814 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1815 ulen); 1816 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1817 drop: 1818 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1819 kfree_skb(skb); 1820 return 0; 1821 } 1822 1823 /* We can only early demux multicast if there is a single matching socket. 1824 * If more than one socket found returns NULL 1825 */ 1826 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1827 __be16 loc_port, __be32 loc_addr, 1828 __be16 rmt_port, __be32 rmt_addr, 1829 int dif) 1830 { 1831 struct sock *sk, *result; 1832 struct hlist_nulls_node *node; 1833 unsigned short hnum = ntohs(loc_port); 1834 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1835 struct udp_hslot *hslot = &udp_table.hash[slot]; 1836 1837 rcu_read_lock(); 1838 begin: 1839 count = 0; 1840 result = NULL; 1841 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1842 if (__udp_is_mcast_sock(net, sk, 1843 loc_port, loc_addr, 1844 rmt_port, rmt_addr, 1845 dif, hnum)) { 1846 result = sk; 1847 ++count; 1848 } 1849 } 1850 /* 1851 * if the nulls value we got at the end of this lookup is 1852 * not the expected one, we must restart lookup. 1853 * We probably met an item that was moved to another chain. 1854 */ 1855 if (get_nulls_value(node) != slot) 1856 goto begin; 1857 1858 if (result) { 1859 if (count != 1 || 1860 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1861 result = NULL; 1862 else if (unlikely(!__udp_is_mcast_sock(net, result, 1863 loc_port, loc_addr, 1864 rmt_port, rmt_addr, 1865 dif, hnum))) { 1866 sock_put(result); 1867 result = NULL; 1868 } 1869 } 1870 rcu_read_unlock(); 1871 return result; 1872 } 1873 1874 /* For unicast we should only early demux connected sockets or we can 1875 * break forwarding setups. The chains here can be long so only check 1876 * if the first socket is an exact match and if not move on. 1877 */ 1878 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1879 __be16 loc_port, __be32 loc_addr, 1880 __be16 rmt_port, __be32 rmt_addr, 1881 int dif) 1882 { 1883 struct sock *sk, *result; 1884 struct hlist_nulls_node *node; 1885 unsigned short hnum = ntohs(loc_port); 1886 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1887 unsigned int slot2 = hash2 & udp_table.mask; 1888 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1889 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr) 1890 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1891 1892 rcu_read_lock(); 1893 result = NULL; 1894 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 1895 if (INET_MATCH(sk, net, acookie, 1896 rmt_addr, loc_addr, ports, dif)) 1897 result = sk; 1898 /* Only check first socket in chain */ 1899 break; 1900 } 1901 1902 if (result) { 1903 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1904 result = NULL; 1905 else if (unlikely(!INET_MATCH(sk, net, acookie, 1906 rmt_addr, loc_addr, 1907 ports, dif))) { 1908 sock_put(result); 1909 result = NULL; 1910 } 1911 } 1912 rcu_read_unlock(); 1913 return result; 1914 } 1915 1916 void udp_v4_early_demux(struct sk_buff *skb) 1917 { 1918 struct net *net = dev_net(skb->dev); 1919 const struct iphdr *iph; 1920 const struct udphdr *uh; 1921 struct sock *sk; 1922 struct dst_entry *dst; 1923 int dif = skb->dev->ifindex; 1924 1925 /* validate the packet */ 1926 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1927 return; 1928 1929 iph = ip_hdr(skb); 1930 uh = udp_hdr(skb); 1931 1932 if (skb->pkt_type == PACKET_BROADCAST || 1933 skb->pkt_type == PACKET_MULTICAST) 1934 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1935 uh->source, iph->saddr, dif); 1936 else if (skb->pkt_type == PACKET_HOST) 1937 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1938 uh->source, iph->saddr, dif); 1939 else 1940 return; 1941 1942 if (!sk) 1943 return; 1944 1945 skb->sk = sk; 1946 skb->destructor = sock_edemux; 1947 dst = sk->sk_rx_dst; 1948 1949 if (dst) 1950 dst = dst_check(dst, 0); 1951 if (dst) 1952 skb_dst_set_noref(skb, dst); 1953 } 1954 1955 int udp_rcv(struct sk_buff *skb) 1956 { 1957 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1958 } 1959 1960 void udp_destroy_sock(struct sock *sk) 1961 { 1962 struct udp_sock *up = udp_sk(sk); 1963 bool slow = lock_sock_fast(sk); 1964 udp_flush_pending_frames(sk); 1965 unlock_sock_fast(sk, slow); 1966 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1967 void (*encap_destroy)(struct sock *sk); 1968 encap_destroy = ACCESS_ONCE(up->encap_destroy); 1969 if (encap_destroy) 1970 encap_destroy(sk); 1971 } 1972 } 1973 1974 /* 1975 * Socket option code for UDP 1976 */ 1977 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1978 char __user *optval, unsigned int optlen, 1979 int (*push_pending_frames)(struct sock *)) 1980 { 1981 struct udp_sock *up = udp_sk(sk); 1982 int val; 1983 int err = 0; 1984 int is_udplite = IS_UDPLITE(sk); 1985 1986 if (optlen < sizeof(int)) 1987 return -EINVAL; 1988 1989 if (get_user(val, (int __user *)optval)) 1990 return -EFAULT; 1991 1992 switch (optname) { 1993 case UDP_CORK: 1994 if (val != 0) { 1995 up->corkflag = 1; 1996 } else { 1997 up->corkflag = 0; 1998 lock_sock(sk); 1999 (*push_pending_frames)(sk); 2000 release_sock(sk); 2001 } 2002 break; 2003 2004 case UDP_ENCAP: 2005 switch (val) { 2006 case 0: 2007 case UDP_ENCAP_ESPINUDP: 2008 case UDP_ENCAP_ESPINUDP_NON_IKE: 2009 up->encap_rcv = xfrm4_udp_encap_rcv; 2010 /* FALLTHROUGH */ 2011 case UDP_ENCAP_L2TPINUDP: 2012 up->encap_type = val; 2013 udp_encap_enable(); 2014 break; 2015 default: 2016 err = -ENOPROTOOPT; 2017 break; 2018 } 2019 break; 2020 2021 /* 2022 * UDP-Lite's partial checksum coverage (RFC 3828). 2023 */ 2024 /* The sender sets actual checksum coverage length via this option. 2025 * The case coverage > packet length is handled by send module. */ 2026 case UDPLITE_SEND_CSCOV: 2027 if (!is_udplite) /* Disable the option on UDP sockets */ 2028 return -ENOPROTOOPT; 2029 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2030 val = 8; 2031 else if (val > USHRT_MAX) 2032 val = USHRT_MAX; 2033 up->pcslen = val; 2034 up->pcflag |= UDPLITE_SEND_CC; 2035 break; 2036 2037 /* The receiver specifies a minimum checksum coverage value. To make 2038 * sense, this should be set to at least 8 (as done below). If zero is 2039 * used, this again means full checksum coverage. */ 2040 case UDPLITE_RECV_CSCOV: 2041 if (!is_udplite) /* Disable the option on UDP sockets */ 2042 return -ENOPROTOOPT; 2043 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2044 val = 8; 2045 else if (val > USHRT_MAX) 2046 val = USHRT_MAX; 2047 up->pcrlen = val; 2048 up->pcflag |= UDPLITE_RECV_CC; 2049 break; 2050 2051 default: 2052 err = -ENOPROTOOPT; 2053 break; 2054 } 2055 2056 return err; 2057 } 2058 EXPORT_SYMBOL(udp_lib_setsockopt); 2059 2060 int udp_setsockopt(struct sock *sk, int level, int optname, 2061 char __user *optval, unsigned int optlen) 2062 { 2063 if (level == SOL_UDP || level == SOL_UDPLITE) 2064 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2065 udp_push_pending_frames); 2066 return ip_setsockopt(sk, level, optname, optval, optlen); 2067 } 2068 2069 #ifdef CONFIG_COMPAT 2070 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2071 char __user *optval, unsigned int optlen) 2072 { 2073 if (level == SOL_UDP || level == SOL_UDPLITE) 2074 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2075 udp_push_pending_frames); 2076 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2077 } 2078 #endif 2079 2080 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2081 char __user *optval, int __user *optlen) 2082 { 2083 struct udp_sock *up = udp_sk(sk); 2084 int val, len; 2085 2086 if (get_user(len, optlen)) 2087 return -EFAULT; 2088 2089 len = min_t(unsigned int, len, sizeof(int)); 2090 2091 if (len < 0) 2092 return -EINVAL; 2093 2094 switch (optname) { 2095 case UDP_CORK: 2096 val = up->corkflag; 2097 break; 2098 2099 case UDP_ENCAP: 2100 val = up->encap_type; 2101 break; 2102 2103 /* The following two cannot be changed on UDP sockets, the return is 2104 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2105 case UDPLITE_SEND_CSCOV: 2106 val = up->pcslen; 2107 break; 2108 2109 case UDPLITE_RECV_CSCOV: 2110 val = up->pcrlen; 2111 break; 2112 2113 default: 2114 return -ENOPROTOOPT; 2115 } 2116 2117 if (put_user(len, optlen)) 2118 return -EFAULT; 2119 if (copy_to_user(optval, &val, len)) 2120 return -EFAULT; 2121 return 0; 2122 } 2123 EXPORT_SYMBOL(udp_lib_getsockopt); 2124 2125 int udp_getsockopt(struct sock *sk, int level, int optname, 2126 char __user *optval, int __user *optlen) 2127 { 2128 if (level == SOL_UDP || level == SOL_UDPLITE) 2129 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2130 return ip_getsockopt(sk, level, optname, optval, optlen); 2131 } 2132 2133 #ifdef CONFIG_COMPAT 2134 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2135 char __user *optval, int __user *optlen) 2136 { 2137 if (level == SOL_UDP || level == SOL_UDPLITE) 2138 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2139 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2140 } 2141 #endif 2142 /** 2143 * udp_poll - wait for a UDP event. 2144 * @file - file struct 2145 * @sock - socket 2146 * @wait - poll table 2147 * 2148 * This is same as datagram poll, except for the special case of 2149 * blocking sockets. If application is using a blocking fd 2150 * and a packet with checksum error is in the queue; 2151 * then it could get return from select indicating data available 2152 * but then block when reading it. Add special case code 2153 * to work around these arguably broken applications. 2154 */ 2155 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2156 { 2157 unsigned int mask = datagram_poll(file, sock, wait); 2158 struct sock *sk = sock->sk; 2159 2160 sock_rps_record_flow(sk); 2161 2162 /* Check for false positives due to checksum errors */ 2163 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2164 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2165 mask &= ~(POLLIN | POLLRDNORM); 2166 2167 return mask; 2168 2169 } 2170 EXPORT_SYMBOL(udp_poll); 2171 2172 struct proto udp_prot = { 2173 .name = "UDP", 2174 .owner = THIS_MODULE, 2175 .close = udp_lib_close, 2176 .connect = ip4_datagram_connect, 2177 .disconnect = udp_disconnect, 2178 .ioctl = udp_ioctl, 2179 .destroy = udp_destroy_sock, 2180 .setsockopt = udp_setsockopt, 2181 .getsockopt = udp_getsockopt, 2182 .sendmsg = udp_sendmsg, 2183 .recvmsg = udp_recvmsg, 2184 .sendpage = udp_sendpage, 2185 .backlog_rcv = __udp_queue_rcv_skb, 2186 .release_cb = ip4_datagram_release_cb, 2187 .hash = udp_lib_hash, 2188 .unhash = udp_lib_unhash, 2189 .rehash = udp_v4_rehash, 2190 .get_port = udp_v4_get_port, 2191 .memory_allocated = &udp_memory_allocated, 2192 .sysctl_mem = sysctl_udp_mem, 2193 .sysctl_wmem = &sysctl_udp_wmem_min, 2194 .sysctl_rmem = &sysctl_udp_rmem_min, 2195 .obj_size = sizeof(struct udp_sock), 2196 .slab_flags = SLAB_DESTROY_BY_RCU, 2197 .h.udp_table = &udp_table, 2198 #ifdef CONFIG_COMPAT 2199 .compat_setsockopt = compat_udp_setsockopt, 2200 .compat_getsockopt = compat_udp_getsockopt, 2201 #endif 2202 .clear_sk = sk_prot_clear_portaddr_nulls, 2203 }; 2204 EXPORT_SYMBOL(udp_prot); 2205 2206 /* ------------------------------------------------------------------------ */ 2207 #ifdef CONFIG_PROC_FS 2208 2209 static struct sock *udp_get_first(struct seq_file *seq, int start) 2210 { 2211 struct sock *sk; 2212 struct udp_iter_state *state = seq->private; 2213 struct net *net = seq_file_net(seq); 2214 2215 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2216 ++state->bucket) { 2217 struct hlist_nulls_node *node; 2218 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2219 2220 if (hlist_nulls_empty(&hslot->head)) 2221 continue; 2222 2223 spin_lock_bh(&hslot->lock); 2224 sk_nulls_for_each(sk, node, &hslot->head) { 2225 if (!net_eq(sock_net(sk), net)) 2226 continue; 2227 if (sk->sk_family == state->family) 2228 goto found; 2229 } 2230 spin_unlock_bh(&hslot->lock); 2231 } 2232 sk = NULL; 2233 found: 2234 return sk; 2235 } 2236 2237 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2238 { 2239 struct udp_iter_state *state = seq->private; 2240 struct net *net = seq_file_net(seq); 2241 2242 do { 2243 sk = sk_nulls_next(sk); 2244 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2245 2246 if (!sk) { 2247 if (state->bucket <= state->udp_table->mask) 2248 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2249 return udp_get_first(seq, state->bucket + 1); 2250 } 2251 return sk; 2252 } 2253 2254 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2255 { 2256 struct sock *sk = udp_get_first(seq, 0); 2257 2258 if (sk) 2259 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2260 --pos; 2261 return pos ? NULL : sk; 2262 } 2263 2264 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2265 { 2266 struct udp_iter_state *state = seq->private; 2267 state->bucket = MAX_UDP_PORTS; 2268 2269 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2270 } 2271 2272 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2273 { 2274 struct sock *sk; 2275 2276 if (v == SEQ_START_TOKEN) 2277 sk = udp_get_idx(seq, 0); 2278 else 2279 sk = udp_get_next(seq, v); 2280 2281 ++*pos; 2282 return sk; 2283 } 2284 2285 static void udp_seq_stop(struct seq_file *seq, void *v) 2286 { 2287 struct udp_iter_state *state = seq->private; 2288 2289 if (state->bucket <= state->udp_table->mask) 2290 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2291 } 2292 2293 int udp_seq_open(struct inode *inode, struct file *file) 2294 { 2295 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2296 struct udp_iter_state *s; 2297 int err; 2298 2299 err = seq_open_net(inode, file, &afinfo->seq_ops, 2300 sizeof(struct udp_iter_state)); 2301 if (err < 0) 2302 return err; 2303 2304 s = ((struct seq_file *)file->private_data)->private; 2305 s->family = afinfo->family; 2306 s->udp_table = afinfo->udp_table; 2307 return err; 2308 } 2309 EXPORT_SYMBOL(udp_seq_open); 2310 2311 /* ------------------------------------------------------------------------ */ 2312 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2313 { 2314 struct proc_dir_entry *p; 2315 int rc = 0; 2316 2317 afinfo->seq_ops.start = udp_seq_start; 2318 afinfo->seq_ops.next = udp_seq_next; 2319 afinfo->seq_ops.stop = udp_seq_stop; 2320 2321 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2322 afinfo->seq_fops, afinfo); 2323 if (!p) 2324 rc = -ENOMEM; 2325 return rc; 2326 } 2327 EXPORT_SYMBOL(udp_proc_register); 2328 2329 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2330 { 2331 remove_proc_entry(afinfo->name, net->proc_net); 2332 } 2333 EXPORT_SYMBOL(udp_proc_unregister); 2334 2335 /* ------------------------------------------------------------------------ */ 2336 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2337 int bucket) 2338 { 2339 struct inet_sock *inet = inet_sk(sp); 2340 __be32 dest = inet->inet_daddr; 2341 __be32 src = inet->inet_rcv_saddr; 2342 __u16 destp = ntohs(inet->inet_dport); 2343 __u16 srcp = ntohs(inet->inet_sport); 2344 2345 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2346 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2347 bucket, src, srcp, dest, destp, sp->sk_state, 2348 sk_wmem_alloc_get(sp), 2349 sk_rmem_alloc_get(sp), 2350 0, 0L, 0, 2351 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2352 0, sock_i_ino(sp), 2353 atomic_read(&sp->sk_refcnt), sp, 2354 atomic_read(&sp->sk_drops)); 2355 } 2356 2357 int udp4_seq_show(struct seq_file *seq, void *v) 2358 { 2359 seq_setwidth(seq, 127); 2360 if (v == SEQ_START_TOKEN) 2361 seq_puts(seq, " sl local_address rem_address st tx_queue " 2362 "rx_queue tr tm->when retrnsmt uid timeout " 2363 "inode ref pointer drops"); 2364 else { 2365 struct udp_iter_state *state = seq->private; 2366 2367 udp4_format_sock(v, seq, state->bucket); 2368 } 2369 seq_pad(seq, '\n'); 2370 return 0; 2371 } 2372 2373 static const struct file_operations udp_afinfo_seq_fops = { 2374 .owner = THIS_MODULE, 2375 .open = udp_seq_open, 2376 .read = seq_read, 2377 .llseek = seq_lseek, 2378 .release = seq_release_net 2379 }; 2380 2381 /* ------------------------------------------------------------------------ */ 2382 static struct udp_seq_afinfo udp4_seq_afinfo = { 2383 .name = "udp", 2384 .family = AF_INET, 2385 .udp_table = &udp_table, 2386 .seq_fops = &udp_afinfo_seq_fops, 2387 .seq_ops = { 2388 .show = udp4_seq_show, 2389 }, 2390 }; 2391 2392 static int __net_init udp4_proc_init_net(struct net *net) 2393 { 2394 return udp_proc_register(net, &udp4_seq_afinfo); 2395 } 2396 2397 static void __net_exit udp4_proc_exit_net(struct net *net) 2398 { 2399 udp_proc_unregister(net, &udp4_seq_afinfo); 2400 } 2401 2402 static struct pernet_operations udp4_net_ops = { 2403 .init = udp4_proc_init_net, 2404 .exit = udp4_proc_exit_net, 2405 }; 2406 2407 int __init udp4_proc_init(void) 2408 { 2409 return register_pernet_subsys(&udp4_net_ops); 2410 } 2411 2412 void udp4_proc_exit(void) 2413 { 2414 unregister_pernet_subsys(&udp4_net_ops); 2415 } 2416 #endif /* CONFIG_PROC_FS */ 2417 2418 static __initdata unsigned long uhash_entries; 2419 static int __init set_uhash_entries(char *str) 2420 { 2421 ssize_t ret; 2422 2423 if (!str) 2424 return 0; 2425 2426 ret = kstrtoul(str, 0, &uhash_entries); 2427 if (ret) 2428 return 0; 2429 2430 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2431 uhash_entries = UDP_HTABLE_SIZE_MIN; 2432 return 1; 2433 } 2434 __setup("uhash_entries=", set_uhash_entries); 2435 2436 void __init udp_table_init(struct udp_table *table, const char *name) 2437 { 2438 unsigned int i; 2439 2440 table->hash = alloc_large_system_hash(name, 2441 2 * sizeof(struct udp_hslot), 2442 uhash_entries, 2443 21, /* one slot per 2 MB */ 2444 0, 2445 &table->log, 2446 &table->mask, 2447 UDP_HTABLE_SIZE_MIN, 2448 64 * 1024); 2449 2450 table->hash2 = table->hash + (table->mask + 1); 2451 for (i = 0; i <= table->mask; i++) { 2452 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2453 table->hash[i].count = 0; 2454 spin_lock_init(&table->hash[i].lock); 2455 } 2456 for (i = 0; i <= table->mask; i++) { 2457 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2458 table->hash2[i].count = 0; 2459 spin_lock_init(&table->hash2[i].lock); 2460 } 2461 } 2462 2463 void __init udp_init(void) 2464 { 2465 unsigned long limit; 2466 2467 udp_table_init(&udp_table, "UDP"); 2468 limit = nr_free_buffer_pages() / 8; 2469 limit = max(limit, 128UL); 2470 sysctl_udp_mem[0] = limit / 4 * 3; 2471 sysctl_udp_mem[1] = limit; 2472 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2473 2474 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2475 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2476 } 2477 2478 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb, 2479 netdev_features_t features) 2480 { 2481 struct sk_buff *segs = ERR_PTR(-EINVAL); 2482 u16 mac_offset = skb->mac_header; 2483 int mac_len = skb->mac_len; 2484 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb); 2485 __be16 protocol = skb->protocol; 2486 netdev_features_t enc_features; 2487 int outer_hlen; 2488 2489 if (unlikely(!pskb_may_pull(skb, tnl_hlen))) 2490 goto out; 2491 2492 skb->encapsulation = 0; 2493 __skb_pull(skb, tnl_hlen); 2494 skb_reset_mac_header(skb); 2495 skb_set_network_header(skb, skb_inner_network_offset(skb)); 2496 skb->mac_len = skb_inner_network_offset(skb); 2497 skb->protocol = htons(ETH_P_TEB); 2498 2499 /* segment inner packet. */ 2500 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb); 2501 segs = skb_mac_gso_segment(skb, enc_features); 2502 if (!segs || IS_ERR(segs)) { 2503 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset, 2504 mac_len); 2505 goto out; 2506 } 2507 2508 outer_hlen = skb_tnl_header_len(skb); 2509 skb = segs; 2510 do { 2511 struct udphdr *uh; 2512 int udp_offset = outer_hlen - tnl_hlen; 2513 2514 skb_reset_inner_headers(skb); 2515 skb->encapsulation = 1; 2516 2517 skb->mac_len = mac_len; 2518 2519 skb_push(skb, outer_hlen); 2520 skb_reset_mac_header(skb); 2521 skb_set_network_header(skb, mac_len); 2522 skb_set_transport_header(skb, udp_offset); 2523 uh = udp_hdr(skb); 2524 uh->len = htons(skb->len - udp_offset); 2525 2526 /* csum segment if tunnel sets skb with csum. */ 2527 if (protocol == htons(ETH_P_IP) && unlikely(uh->check)) { 2528 struct iphdr *iph = ip_hdr(skb); 2529 2530 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 2531 skb->len - udp_offset, 2532 IPPROTO_UDP, 0); 2533 uh->check = csum_fold(skb_checksum(skb, udp_offset, 2534 skb->len - udp_offset, 0)); 2535 if (uh->check == 0) 2536 uh->check = CSUM_MANGLED_0; 2537 2538 } else if (protocol == htons(ETH_P_IPV6)) { 2539 struct ipv6hdr *ipv6h = ipv6_hdr(skb); 2540 u32 len = skb->len - udp_offset; 2541 2542 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, 2543 len, IPPROTO_UDP, 0); 2544 uh->check = csum_fold(skb_checksum(skb, udp_offset, len, 0)); 2545 if (uh->check == 0) 2546 uh->check = CSUM_MANGLED_0; 2547 skb->ip_summed = CHECKSUM_NONE; 2548 } 2549 2550 skb->protocol = protocol; 2551 } while ((skb = skb->next)); 2552 out: 2553 return segs; 2554 } 2555