1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The User Datagram Protocol (UDP). 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 12 * Hirokazu Takahashi, <taka@valinux.co.jp> 13 * 14 * Fixes: 15 * Alan Cox : verify_area() calls 16 * Alan Cox : stopped close while in use off icmp 17 * messages. Not a fix but a botch that 18 * for udp at least is 'valid'. 19 * Alan Cox : Fixed icmp handling properly 20 * Alan Cox : Correct error for oversized datagrams 21 * Alan Cox : Tidied select() semantics. 22 * Alan Cox : udp_err() fixed properly, also now 23 * select and read wake correctly on errors 24 * Alan Cox : udp_send verify_area moved to avoid mem leak 25 * Alan Cox : UDP can count its memory 26 * Alan Cox : send to an unknown connection causes 27 * an ECONNREFUSED off the icmp, but 28 * does NOT close. 29 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 31 * bug no longer crashes it. 32 * Fred Van Kempen : Net2e support for sk->broadcast. 33 * Alan Cox : Uses skb_free_datagram 34 * Alan Cox : Added get/set sockopt support. 35 * Alan Cox : Broadcasting without option set returns EACCES. 36 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 37 * Alan Cox : Use ip_tos and ip_ttl 38 * Alan Cox : SNMP Mibs 39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 40 * Matt Dillon : UDP length checks. 41 * Alan Cox : Smarter af_inet used properly. 42 * Alan Cox : Use new kernel side addressing. 43 * Alan Cox : Incorrect return on truncated datagram receive. 44 * Arnt Gulbrandsen : New udp_send and stuff 45 * Alan Cox : Cache last socket 46 * Alan Cox : Route cache 47 * Jon Peatfield : Minor efficiency fix to sendto(). 48 * Mike Shaver : RFC1122 checks. 49 * Alan Cox : Nonblocking error fix. 50 * Willy Konynenberg : Transparent proxying support. 51 * Mike McLagan : Routing by source 52 * David S. Miller : New socket lookup architecture. 53 * Last socket cache retained as it 54 * does have a high hit rate. 55 * Olaf Kirch : Don't linearise iovec on sendmsg. 56 * Andi Kleen : Some cleanups, cache destination entry 57 * for connect. 58 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 59 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 60 * return ENOTCONN for unconnected sockets (POSIX) 61 * Janos Farkas : don't deliver multi/broadcasts to a different 62 * bound-to-device socket 63 * Hirokazu Takahashi : HW checksumming for outgoing UDP 64 * datagrams. 65 * Hirokazu Takahashi : sendfile() on UDP works now. 66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 69 * a single port at the same time. 70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71 * James Chapman : Add L2TP encapsulation type. 72 * 73 * 74 * This program is free software; you can redistribute it and/or 75 * modify it under the terms of the GNU General Public License 76 * as published by the Free Software Foundation; either version 77 * 2 of the License, or (at your option) any later version. 78 */ 79 80 #define pr_fmt(fmt) "UDP: " fmt 81 82 #include <asm/uaccess.h> 83 #include <asm/ioctls.h> 84 #include <linux/bootmem.h> 85 #include <linux/highmem.h> 86 #include <linux/swap.h> 87 #include <linux/types.h> 88 #include <linux/fcntl.h> 89 #include <linux/module.h> 90 #include <linux/socket.h> 91 #include <linux/sockios.h> 92 #include <linux/igmp.h> 93 #include <linux/in.h> 94 #include <linux/errno.h> 95 #include <linux/timer.h> 96 #include <linux/mm.h> 97 #include <linux/inet.h> 98 #include <linux/netdevice.h> 99 #include <linux/slab.h> 100 #include <net/tcp_states.h> 101 #include <linux/skbuff.h> 102 #include <linux/netdevice.h> 103 #include <linux/proc_fs.h> 104 #include <linux/seq_file.h> 105 #include <net/net_namespace.h> 106 #include <net/icmp.h> 107 #include <net/inet_hashtables.h> 108 #include <net/route.h> 109 #include <net/checksum.h> 110 #include <net/xfrm.h> 111 #include <trace/events/udp.h> 112 #include <linux/static_key.h> 113 #include <trace/events/skb.h> 114 #include <net/busy_poll.h> 115 #include "udp_impl.h" 116 117 struct udp_table udp_table __read_mostly; 118 EXPORT_SYMBOL(udp_table); 119 120 long sysctl_udp_mem[3] __read_mostly; 121 EXPORT_SYMBOL(sysctl_udp_mem); 122 123 int sysctl_udp_rmem_min __read_mostly; 124 EXPORT_SYMBOL(sysctl_udp_rmem_min); 125 126 int sysctl_udp_wmem_min __read_mostly; 127 EXPORT_SYMBOL(sysctl_udp_wmem_min); 128 129 atomic_long_t udp_memory_allocated; 130 EXPORT_SYMBOL(udp_memory_allocated); 131 132 #define MAX_UDP_PORTS 65536 133 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 134 135 static int udp_lib_lport_inuse(struct net *net, __u16 num, 136 const struct udp_hslot *hslot, 137 unsigned long *bitmap, 138 struct sock *sk, 139 int (*saddr_comp)(const struct sock *sk1, 140 const struct sock *sk2), 141 unsigned int log) 142 { 143 struct sock *sk2; 144 struct hlist_nulls_node *node; 145 kuid_t uid = sock_i_uid(sk); 146 147 sk_nulls_for_each(sk2, node, &hslot->head) { 148 if (net_eq(sock_net(sk2), net) && 149 sk2 != sk && 150 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 151 (!sk2->sk_reuse || !sk->sk_reuse) && 152 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 153 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 154 (!sk2->sk_reuseport || !sk->sk_reuseport || 155 !uid_eq(uid, sock_i_uid(sk2))) && 156 saddr_comp(sk, sk2)) { 157 if (!bitmap) 158 return 1; 159 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap); 160 } 161 } 162 return 0; 163 } 164 165 /* 166 * Note: we still hold spinlock of primary hash chain, so no other writer 167 * can insert/delete a socket with local_port == num 168 */ 169 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 170 struct udp_hslot *hslot2, 171 struct sock *sk, 172 int (*saddr_comp)(const struct sock *sk1, 173 const struct sock *sk2)) 174 { 175 struct sock *sk2; 176 struct hlist_nulls_node *node; 177 kuid_t uid = sock_i_uid(sk); 178 int res = 0; 179 180 spin_lock(&hslot2->lock); 181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) { 182 if (net_eq(sock_net(sk2), net) && 183 sk2 != sk && 184 (udp_sk(sk2)->udp_port_hash == num) && 185 (!sk2->sk_reuse || !sk->sk_reuse) && 186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 188 (!sk2->sk_reuseport || !sk->sk_reuseport || 189 !uid_eq(uid, sock_i_uid(sk2))) && 190 saddr_comp(sk, sk2)) { 191 res = 1; 192 break; 193 } 194 } 195 spin_unlock(&hslot2->lock); 196 return res; 197 } 198 199 /** 200 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 201 * 202 * @sk: socket struct in question 203 * @snum: port number to look up 204 * @saddr_comp: AF-dependent comparison of bound local IP addresses 205 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 206 * with NULL address 207 */ 208 int udp_lib_get_port(struct sock *sk, unsigned short snum, 209 int (*saddr_comp)(const struct sock *sk1, 210 const struct sock *sk2), 211 unsigned int hash2_nulladdr) 212 { 213 struct udp_hslot *hslot, *hslot2; 214 struct udp_table *udptable = sk->sk_prot->h.udp_table; 215 int error = 1; 216 struct net *net = sock_net(sk); 217 218 if (!snum) { 219 int low, high, remaining; 220 unsigned int rand; 221 unsigned short first, last; 222 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 223 224 inet_get_local_port_range(net, &low, &high); 225 remaining = (high - low) + 1; 226 227 rand = prandom_u32(); 228 first = reciprocal_scale(rand, remaining) + low; 229 /* 230 * force rand to be an odd multiple of UDP_HTABLE_SIZE 231 */ 232 rand = (rand | 1) * (udptable->mask + 1); 233 last = first + udptable->mask + 1; 234 do { 235 hslot = udp_hashslot(udptable, net, first); 236 bitmap_zero(bitmap, PORTS_PER_CHAIN); 237 spin_lock_bh(&hslot->lock); 238 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 239 saddr_comp, udptable->log); 240 241 snum = first; 242 /* 243 * Iterate on all possible values of snum for this hash. 244 * Using steps of an odd multiple of UDP_HTABLE_SIZE 245 * give us randomization and full range coverage. 246 */ 247 do { 248 if (low <= snum && snum <= high && 249 !test_bit(snum >> udptable->log, bitmap) && 250 !inet_is_local_reserved_port(net, snum)) 251 goto found; 252 snum += rand; 253 } while (snum != first); 254 spin_unlock_bh(&hslot->lock); 255 } while (++first != last); 256 goto fail; 257 } else { 258 hslot = udp_hashslot(udptable, net, snum); 259 spin_lock_bh(&hslot->lock); 260 if (hslot->count > 10) { 261 int exist; 262 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 263 264 slot2 &= udptable->mask; 265 hash2_nulladdr &= udptable->mask; 266 267 hslot2 = udp_hashslot2(udptable, slot2); 268 if (hslot->count < hslot2->count) 269 goto scan_primary_hash; 270 271 exist = udp_lib_lport_inuse2(net, snum, hslot2, 272 sk, saddr_comp); 273 if (!exist && (hash2_nulladdr != slot2)) { 274 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 275 exist = udp_lib_lport_inuse2(net, snum, hslot2, 276 sk, saddr_comp); 277 } 278 if (exist) 279 goto fail_unlock; 280 else 281 goto found; 282 } 283 scan_primary_hash: 284 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 285 saddr_comp, 0)) 286 goto fail_unlock; 287 } 288 found: 289 inet_sk(sk)->inet_num = snum; 290 udp_sk(sk)->udp_port_hash = snum; 291 udp_sk(sk)->udp_portaddr_hash ^= snum; 292 if (sk_unhashed(sk)) { 293 sk_nulls_add_node_rcu(sk, &hslot->head); 294 hslot->count++; 295 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 296 297 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 298 spin_lock(&hslot2->lock); 299 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 300 &hslot2->head); 301 hslot2->count++; 302 spin_unlock(&hslot2->lock); 303 } 304 error = 0; 305 fail_unlock: 306 spin_unlock_bh(&hslot->lock); 307 fail: 308 return error; 309 } 310 EXPORT_SYMBOL(udp_lib_get_port); 311 312 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 313 { 314 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 315 316 return (!ipv6_only_sock(sk2) && 317 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 318 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 319 } 320 321 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, 322 unsigned int port) 323 { 324 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 325 } 326 327 int udp_v4_get_port(struct sock *sk, unsigned short snum) 328 { 329 unsigned int hash2_nulladdr = 330 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 331 unsigned int hash2_partial = 332 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 333 334 /* precompute partial secondary hash */ 335 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 336 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 337 } 338 339 static inline int compute_score(struct sock *sk, struct net *net, 340 __be32 saddr, unsigned short hnum, __be16 sport, 341 __be32 daddr, __be16 dport, int dif) 342 { 343 int score; 344 struct inet_sock *inet; 345 346 if (!net_eq(sock_net(sk), net) || 347 udp_sk(sk)->udp_port_hash != hnum || 348 ipv6_only_sock(sk)) 349 return -1; 350 351 score = (sk->sk_family == PF_INET) ? 2 : 1; 352 inet = inet_sk(sk); 353 354 if (inet->inet_rcv_saddr) { 355 if (inet->inet_rcv_saddr != daddr) 356 return -1; 357 score += 4; 358 } 359 360 if (inet->inet_daddr) { 361 if (inet->inet_daddr != saddr) 362 return -1; 363 score += 4; 364 } 365 366 if (inet->inet_dport) { 367 if (inet->inet_dport != sport) 368 return -1; 369 score += 4; 370 } 371 372 if (sk->sk_bound_dev_if) { 373 if (sk->sk_bound_dev_if != dif) 374 return -1; 375 score += 4; 376 } 377 378 return score; 379 } 380 381 /* 382 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 383 */ 384 static inline int compute_score2(struct sock *sk, struct net *net, 385 __be32 saddr, __be16 sport, 386 __be32 daddr, unsigned int hnum, int dif) 387 { 388 int score; 389 struct inet_sock *inet; 390 391 if (!net_eq(sock_net(sk), net) || 392 ipv6_only_sock(sk)) 393 return -1; 394 395 inet = inet_sk(sk); 396 397 if (inet->inet_rcv_saddr != daddr || 398 inet->inet_num != hnum) 399 return -1; 400 401 score = (sk->sk_family == PF_INET) ? 2 : 1; 402 403 if (inet->inet_daddr) { 404 if (inet->inet_daddr != saddr) 405 return -1; 406 score += 4; 407 } 408 409 if (inet->inet_dport) { 410 if (inet->inet_dport != sport) 411 return -1; 412 score += 4; 413 } 414 415 if (sk->sk_bound_dev_if) { 416 if (sk->sk_bound_dev_if != dif) 417 return -1; 418 score += 4; 419 } 420 421 return score; 422 } 423 424 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 425 const __u16 lport, const __be32 faddr, 426 const __be16 fport) 427 { 428 static u32 udp_ehash_secret __read_mostly; 429 430 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 431 432 return __inet_ehashfn(laddr, lport, faddr, fport, 433 udp_ehash_secret + net_hash_mix(net)); 434 } 435 436 /* called with read_rcu_lock() */ 437 static struct sock *udp4_lib_lookup2(struct net *net, 438 __be32 saddr, __be16 sport, 439 __be32 daddr, unsigned int hnum, int dif, 440 struct udp_hslot *hslot2, unsigned int slot2) 441 { 442 struct sock *sk, *result; 443 struct hlist_nulls_node *node; 444 int score, badness, matches = 0, reuseport = 0; 445 u32 hash = 0; 446 447 begin: 448 result = NULL; 449 badness = 0; 450 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 451 score = compute_score2(sk, net, saddr, sport, 452 daddr, hnum, dif); 453 if (score > badness) { 454 result = sk; 455 badness = score; 456 reuseport = sk->sk_reuseport; 457 if (reuseport) { 458 hash = udp_ehashfn(net, daddr, hnum, 459 saddr, sport); 460 matches = 1; 461 } 462 } else if (score == badness && reuseport) { 463 matches++; 464 if (reciprocal_scale(hash, matches) == 0) 465 result = sk; 466 hash = next_pseudo_random32(hash); 467 } 468 } 469 /* 470 * if the nulls value we got at the end of this lookup is 471 * not the expected one, we must restart lookup. 472 * We probably met an item that was moved to another chain. 473 */ 474 if (get_nulls_value(node) != slot2) 475 goto begin; 476 if (result) { 477 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 478 result = NULL; 479 else if (unlikely(compute_score2(result, net, saddr, sport, 480 daddr, hnum, dif) < badness)) { 481 sock_put(result); 482 goto begin; 483 } 484 } 485 return result; 486 } 487 488 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 489 * harder than this. -DaveM 490 */ 491 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 492 __be16 sport, __be32 daddr, __be16 dport, 493 int dif, struct udp_table *udptable) 494 { 495 struct sock *sk, *result; 496 struct hlist_nulls_node *node; 497 unsigned short hnum = ntohs(dport); 498 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 499 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 500 int score, badness, matches = 0, reuseport = 0; 501 u32 hash = 0; 502 503 rcu_read_lock(); 504 if (hslot->count > 10) { 505 hash2 = udp4_portaddr_hash(net, daddr, hnum); 506 slot2 = hash2 & udptable->mask; 507 hslot2 = &udptable->hash2[slot2]; 508 if (hslot->count < hslot2->count) 509 goto begin; 510 511 result = udp4_lib_lookup2(net, saddr, sport, 512 daddr, hnum, dif, 513 hslot2, slot2); 514 if (!result) { 515 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 516 slot2 = hash2 & udptable->mask; 517 hslot2 = &udptable->hash2[slot2]; 518 if (hslot->count < hslot2->count) 519 goto begin; 520 521 result = udp4_lib_lookup2(net, saddr, sport, 522 htonl(INADDR_ANY), hnum, dif, 523 hslot2, slot2); 524 } 525 rcu_read_unlock(); 526 return result; 527 } 528 begin: 529 result = NULL; 530 badness = 0; 531 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 532 score = compute_score(sk, net, saddr, hnum, sport, 533 daddr, dport, dif); 534 if (score > badness) { 535 result = sk; 536 badness = score; 537 reuseport = sk->sk_reuseport; 538 if (reuseport) { 539 hash = udp_ehashfn(net, daddr, hnum, 540 saddr, sport); 541 matches = 1; 542 } 543 } else if (score == badness && reuseport) { 544 matches++; 545 if (reciprocal_scale(hash, matches) == 0) 546 result = sk; 547 hash = next_pseudo_random32(hash); 548 } 549 } 550 /* 551 * if the nulls value we got at the end of this lookup is 552 * not the expected one, we must restart lookup. 553 * We probably met an item that was moved to another chain. 554 */ 555 if (get_nulls_value(node) != slot) 556 goto begin; 557 558 if (result) { 559 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 560 result = NULL; 561 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 562 daddr, dport, dif) < badness)) { 563 sock_put(result); 564 goto begin; 565 } 566 } 567 rcu_read_unlock(); 568 return result; 569 } 570 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 571 572 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 573 __be16 sport, __be16 dport, 574 struct udp_table *udptable) 575 { 576 const struct iphdr *iph = ip_hdr(skb); 577 578 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 579 iph->daddr, dport, inet_iif(skb), 580 udptable); 581 } 582 583 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 584 __be32 daddr, __be16 dport, int dif) 585 { 586 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 587 } 588 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 589 590 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 591 __be16 loc_port, __be32 loc_addr, 592 __be16 rmt_port, __be32 rmt_addr, 593 int dif, unsigned short hnum) 594 { 595 struct inet_sock *inet = inet_sk(sk); 596 597 if (!net_eq(sock_net(sk), net) || 598 udp_sk(sk)->udp_port_hash != hnum || 599 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 600 (inet->inet_dport != rmt_port && inet->inet_dport) || 601 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 602 ipv6_only_sock(sk) || 603 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 604 return false; 605 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 606 return false; 607 return true; 608 } 609 610 /* 611 * This routine is called by the ICMP module when it gets some 612 * sort of error condition. If err < 0 then the socket should 613 * be closed and the error returned to the user. If err > 0 614 * it's just the icmp type << 8 | icmp code. 615 * Header points to the ip header of the error packet. We move 616 * on past this. Then (as it used to claim before adjustment) 617 * header points to the first 8 bytes of the udp header. We need 618 * to find the appropriate port. 619 */ 620 621 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 622 { 623 struct inet_sock *inet; 624 const struct iphdr *iph = (const struct iphdr *)skb->data; 625 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 626 const int type = icmp_hdr(skb)->type; 627 const int code = icmp_hdr(skb)->code; 628 struct sock *sk; 629 int harderr; 630 int err; 631 struct net *net = dev_net(skb->dev); 632 633 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 634 iph->saddr, uh->source, skb->dev->ifindex, udptable); 635 if (!sk) { 636 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 637 return; /* No socket for error */ 638 } 639 640 err = 0; 641 harderr = 0; 642 inet = inet_sk(sk); 643 644 switch (type) { 645 default: 646 case ICMP_TIME_EXCEEDED: 647 err = EHOSTUNREACH; 648 break; 649 case ICMP_SOURCE_QUENCH: 650 goto out; 651 case ICMP_PARAMETERPROB: 652 err = EPROTO; 653 harderr = 1; 654 break; 655 case ICMP_DEST_UNREACH: 656 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 657 ipv4_sk_update_pmtu(skb, sk, info); 658 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 659 err = EMSGSIZE; 660 harderr = 1; 661 break; 662 } 663 goto out; 664 } 665 err = EHOSTUNREACH; 666 if (code <= NR_ICMP_UNREACH) { 667 harderr = icmp_err_convert[code].fatal; 668 err = icmp_err_convert[code].errno; 669 } 670 break; 671 case ICMP_REDIRECT: 672 ipv4_sk_redirect(skb, sk); 673 goto out; 674 } 675 676 /* 677 * RFC1122: OK. Passes ICMP errors back to application, as per 678 * 4.1.3.3. 679 */ 680 if (!inet->recverr) { 681 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 682 goto out; 683 } else 684 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 685 686 sk->sk_err = err; 687 sk->sk_error_report(sk); 688 out: 689 sock_put(sk); 690 } 691 692 void udp_err(struct sk_buff *skb, u32 info) 693 { 694 __udp4_lib_err(skb, info, &udp_table); 695 } 696 697 /* 698 * Throw away all pending data and cancel the corking. Socket is locked. 699 */ 700 void udp_flush_pending_frames(struct sock *sk) 701 { 702 struct udp_sock *up = udp_sk(sk); 703 704 if (up->pending) { 705 up->len = 0; 706 up->pending = 0; 707 ip_flush_pending_frames(sk); 708 } 709 } 710 EXPORT_SYMBOL(udp_flush_pending_frames); 711 712 /** 713 * udp4_hwcsum - handle outgoing HW checksumming 714 * @skb: sk_buff containing the filled-in UDP header 715 * (checksum field must be zeroed out) 716 * @src: source IP address 717 * @dst: destination IP address 718 */ 719 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 720 { 721 struct udphdr *uh = udp_hdr(skb); 722 int offset = skb_transport_offset(skb); 723 int len = skb->len - offset; 724 int hlen = len; 725 __wsum csum = 0; 726 727 if (!skb_has_frag_list(skb)) { 728 /* 729 * Only one fragment on the socket. 730 */ 731 skb->csum_start = skb_transport_header(skb) - skb->head; 732 skb->csum_offset = offsetof(struct udphdr, check); 733 uh->check = ~csum_tcpudp_magic(src, dst, len, 734 IPPROTO_UDP, 0); 735 } else { 736 struct sk_buff *frags; 737 738 /* 739 * HW-checksum won't work as there are two or more 740 * fragments on the socket so that all csums of sk_buffs 741 * should be together 742 */ 743 skb_walk_frags(skb, frags) { 744 csum = csum_add(csum, frags->csum); 745 hlen -= frags->len; 746 } 747 748 csum = skb_checksum(skb, offset, hlen, csum); 749 skb->ip_summed = CHECKSUM_NONE; 750 751 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 752 if (uh->check == 0) 753 uh->check = CSUM_MANGLED_0; 754 } 755 } 756 EXPORT_SYMBOL_GPL(udp4_hwcsum); 757 758 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 759 * for the simple case like when setting the checksum for a UDP tunnel. 760 */ 761 void udp_set_csum(bool nocheck, struct sk_buff *skb, 762 __be32 saddr, __be32 daddr, int len) 763 { 764 struct udphdr *uh = udp_hdr(skb); 765 766 if (nocheck) 767 uh->check = 0; 768 else if (skb_is_gso(skb)) 769 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 770 else if (skb_dst(skb) && skb_dst(skb)->dev && 771 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) { 772 773 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 774 775 skb->ip_summed = CHECKSUM_PARTIAL; 776 skb->csum_start = skb_transport_header(skb) - skb->head; 777 skb->csum_offset = offsetof(struct udphdr, check); 778 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 779 } else { 780 __wsum csum; 781 782 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 783 784 uh->check = 0; 785 csum = skb_checksum(skb, 0, len, 0); 786 uh->check = udp_v4_check(len, saddr, daddr, csum); 787 if (uh->check == 0) 788 uh->check = CSUM_MANGLED_0; 789 790 skb->ip_summed = CHECKSUM_UNNECESSARY; 791 } 792 } 793 EXPORT_SYMBOL(udp_set_csum); 794 795 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 796 { 797 struct sock *sk = skb->sk; 798 struct inet_sock *inet = inet_sk(sk); 799 struct udphdr *uh; 800 int err = 0; 801 int is_udplite = IS_UDPLITE(sk); 802 int offset = skb_transport_offset(skb); 803 int len = skb->len - offset; 804 __wsum csum = 0; 805 806 /* 807 * Create a UDP header 808 */ 809 uh = udp_hdr(skb); 810 uh->source = inet->inet_sport; 811 uh->dest = fl4->fl4_dport; 812 uh->len = htons(len); 813 uh->check = 0; 814 815 if (is_udplite) /* UDP-Lite */ 816 csum = udplite_csum(skb); 817 818 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 819 820 skb->ip_summed = CHECKSUM_NONE; 821 goto send; 822 823 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 824 825 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 826 goto send; 827 828 } else 829 csum = udp_csum(skb); 830 831 /* add protocol-dependent pseudo-header */ 832 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 833 sk->sk_protocol, csum); 834 if (uh->check == 0) 835 uh->check = CSUM_MANGLED_0; 836 837 send: 838 err = ip_send_skb(sock_net(sk), skb); 839 if (err) { 840 if (err == -ENOBUFS && !inet->recverr) { 841 UDP_INC_STATS_USER(sock_net(sk), 842 UDP_MIB_SNDBUFERRORS, is_udplite); 843 err = 0; 844 } 845 } else 846 UDP_INC_STATS_USER(sock_net(sk), 847 UDP_MIB_OUTDATAGRAMS, is_udplite); 848 return err; 849 } 850 851 /* 852 * Push out all pending data as one UDP datagram. Socket is locked. 853 */ 854 int udp_push_pending_frames(struct sock *sk) 855 { 856 struct udp_sock *up = udp_sk(sk); 857 struct inet_sock *inet = inet_sk(sk); 858 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 859 struct sk_buff *skb; 860 int err = 0; 861 862 skb = ip_finish_skb(sk, fl4); 863 if (!skb) 864 goto out; 865 866 err = udp_send_skb(skb, fl4); 867 868 out: 869 up->len = 0; 870 up->pending = 0; 871 return err; 872 } 873 EXPORT_SYMBOL(udp_push_pending_frames); 874 875 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 876 { 877 struct inet_sock *inet = inet_sk(sk); 878 struct udp_sock *up = udp_sk(sk); 879 struct flowi4 fl4_stack; 880 struct flowi4 *fl4; 881 int ulen = len; 882 struct ipcm_cookie ipc; 883 struct rtable *rt = NULL; 884 int free = 0; 885 int connected = 0; 886 __be32 daddr, faddr, saddr; 887 __be16 dport; 888 u8 tos; 889 int err, is_udplite = IS_UDPLITE(sk); 890 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 891 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 892 struct sk_buff *skb; 893 struct ip_options_data opt_copy; 894 895 if (len > 0xFFFF) 896 return -EMSGSIZE; 897 898 /* 899 * Check the flags. 900 */ 901 902 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 903 return -EOPNOTSUPP; 904 905 ipc.opt = NULL; 906 ipc.tx_flags = 0; 907 ipc.ttl = 0; 908 ipc.tos = -1; 909 910 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 911 912 fl4 = &inet->cork.fl.u.ip4; 913 if (up->pending) { 914 /* 915 * There are pending frames. 916 * The socket lock must be held while it's corked. 917 */ 918 lock_sock(sk); 919 if (likely(up->pending)) { 920 if (unlikely(up->pending != AF_INET)) { 921 release_sock(sk); 922 return -EINVAL; 923 } 924 goto do_append_data; 925 } 926 release_sock(sk); 927 } 928 ulen += sizeof(struct udphdr); 929 930 /* 931 * Get and verify the address. 932 */ 933 if (msg->msg_name) { 934 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 935 if (msg->msg_namelen < sizeof(*usin)) 936 return -EINVAL; 937 if (usin->sin_family != AF_INET) { 938 if (usin->sin_family != AF_UNSPEC) 939 return -EAFNOSUPPORT; 940 } 941 942 daddr = usin->sin_addr.s_addr; 943 dport = usin->sin_port; 944 if (dport == 0) 945 return -EINVAL; 946 } else { 947 if (sk->sk_state != TCP_ESTABLISHED) 948 return -EDESTADDRREQ; 949 daddr = inet->inet_daddr; 950 dport = inet->inet_dport; 951 /* Open fast path for connected socket. 952 Route will not be used, if at least one option is set. 953 */ 954 connected = 1; 955 } 956 ipc.addr = inet->inet_saddr; 957 958 ipc.oif = sk->sk_bound_dev_if; 959 960 sock_tx_timestamp(sk, &ipc.tx_flags); 961 962 if (msg->msg_controllen) { 963 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 964 sk->sk_family == AF_INET6); 965 if (err) 966 return err; 967 if (ipc.opt) 968 free = 1; 969 connected = 0; 970 } 971 if (!ipc.opt) { 972 struct ip_options_rcu *inet_opt; 973 974 rcu_read_lock(); 975 inet_opt = rcu_dereference(inet->inet_opt); 976 if (inet_opt) { 977 memcpy(&opt_copy, inet_opt, 978 sizeof(*inet_opt) + inet_opt->opt.optlen); 979 ipc.opt = &opt_copy.opt; 980 } 981 rcu_read_unlock(); 982 } 983 984 saddr = ipc.addr; 985 ipc.addr = faddr = daddr; 986 987 if (ipc.opt && ipc.opt->opt.srr) { 988 if (!daddr) 989 return -EINVAL; 990 faddr = ipc.opt->opt.faddr; 991 connected = 0; 992 } 993 tos = get_rttos(&ipc, inet); 994 if (sock_flag(sk, SOCK_LOCALROUTE) || 995 (msg->msg_flags & MSG_DONTROUTE) || 996 (ipc.opt && ipc.opt->opt.is_strictroute)) { 997 tos |= RTO_ONLINK; 998 connected = 0; 999 } 1000 1001 if (ipv4_is_multicast(daddr)) { 1002 if (!ipc.oif) 1003 ipc.oif = inet->mc_index; 1004 if (!saddr) 1005 saddr = inet->mc_addr; 1006 connected = 0; 1007 } else if (!ipc.oif) 1008 ipc.oif = inet->uc_index; 1009 1010 if (connected) 1011 rt = (struct rtable *)sk_dst_check(sk, 0); 1012 1013 if (!rt) { 1014 struct net *net = sock_net(sk); 1015 1016 fl4 = &fl4_stack; 1017 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1018 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1019 inet_sk_flowi_flags(sk), 1020 faddr, saddr, dport, inet->inet_sport); 1021 1022 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1023 rt = ip_route_output_flow(net, fl4, sk); 1024 if (IS_ERR(rt)) { 1025 err = PTR_ERR(rt); 1026 rt = NULL; 1027 if (err == -ENETUNREACH) 1028 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1029 goto out; 1030 } 1031 1032 err = -EACCES; 1033 if ((rt->rt_flags & RTCF_BROADCAST) && 1034 !sock_flag(sk, SOCK_BROADCAST)) 1035 goto out; 1036 if (connected) 1037 sk_dst_set(sk, dst_clone(&rt->dst)); 1038 } 1039 1040 if (msg->msg_flags&MSG_CONFIRM) 1041 goto do_confirm; 1042 back_from_confirm: 1043 1044 saddr = fl4->saddr; 1045 if (!ipc.addr) 1046 daddr = ipc.addr = fl4->daddr; 1047 1048 /* Lockless fast path for the non-corking case. */ 1049 if (!corkreq) { 1050 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1051 sizeof(struct udphdr), &ipc, &rt, 1052 msg->msg_flags); 1053 err = PTR_ERR(skb); 1054 if (!IS_ERR_OR_NULL(skb)) 1055 err = udp_send_skb(skb, fl4); 1056 goto out; 1057 } 1058 1059 lock_sock(sk); 1060 if (unlikely(up->pending)) { 1061 /* The socket is already corked while preparing it. */ 1062 /* ... which is an evident application bug. --ANK */ 1063 release_sock(sk); 1064 1065 net_dbg_ratelimited("cork app bug 2\n"); 1066 err = -EINVAL; 1067 goto out; 1068 } 1069 /* 1070 * Now cork the socket to pend data. 1071 */ 1072 fl4 = &inet->cork.fl.u.ip4; 1073 fl4->daddr = daddr; 1074 fl4->saddr = saddr; 1075 fl4->fl4_dport = dport; 1076 fl4->fl4_sport = inet->inet_sport; 1077 up->pending = AF_INET; 1078 1079 do_append_data: 1080 up->len += ulen; 1081 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1082 sizeof(struct udphdr), &ipc, &rt, 1083 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1084 if (err) 1085 udp_flush_pending_frames(sk); 1086 else if (!corkreq) 1087 err = udp_push_pending_frames(sk); 1088 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1089 up->pending = 0; 1090 release_sock(sk); 1091 1092 out: 1093 ip_rt_put(rt); 1094 if (free) 1095 kfree(ipc.opt); 1096 if (!err) 1097 return len; 1098 /* 1099 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1100 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1101 * we don't have a good statistic (IpOutDiscards but it can be too many 1102 * things). We could add another new stat but at least for now that 1103 * seems like overkill. 1104 */ 1105 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1106 UDP_INC_STATS_USER(sock_net(sk), 1107 UDP_MIB_SNDBUFERRORS, is_udplite); 1108 } 1109 return err; 1110 1111 do_confirm: 1112 dst_confirm(&rt->dst); 1113 if (!(msg->msg_flags&MSG_PROBE) || len) 1114 goto back_from_confirm; 1115 err = 0; 1116 goto out; 1117 } 1118 EXPORT_SYMBOL(udp_sendmsg); 1119 1120 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1121 size_t size, int flags) 1122 { 1123 struct inet_sock *inet = inet_sk(sk); 1124 struct udp_sock *up = udp_sk(sk); 1125 int ret; 1126 1127 if (flags & MSG_SENDPAGE_NOTLAST) 1128 flags |= MSG_MORE; 1129 1130 if (!up->pending) { 1131 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1132 1133 /* Call udp_sendmsg to specify destination address which 1134 * sendpage interface can't pass. 1135 * This will succeed only when the socket is connected. 1136 */ 1137 ret = udp_sendmsg(sk, &msg, 0); 1138 if (ret < 0) 1139 return ret; 1140 } 1141 1142 lock_sock(sk); 1143 1144 if (unlikely(!up->pending)) { 1145 release_sock(sk); 1146 1147 net_dbg_ratelimited("udp cork app bug 3\n"); 1148 return -EINVAL; 1149 } 1150 1151 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1152 page, offset, size, flags); 1153 if (ret == -EOPNOTSUPP) { 1154 release_sock(sk); 1155 return sock_no_sendpage(sk->sk_socket, page, offset, 1156 size, flags); 1157 } 1158 if (ret < 0) { 1159 udp_flush_pending_frames(sk); 1160 goto out; 1161 } 1162 1163 up->len += size; 1164 if (!(up->corkflag || (flags&MSG_MORE))) 1165 ret = udp_push_pending_frames(sk); 1166 if (!ret) 1167 ret = size; 1168 out: 1169 release_sock(sk); 1170 return ret; 1171 } 1172 1173 /** 1174 * first_packet_length - return length of first packet in receive queue 1175 * @sk: socket 1176 * 1177 * Drops all bad checksum frames, until a valid one is found. 1178 * Returns the length of found skb, or 0 if none is found. 1179 */ 1180 static unsigned int first_packet_length(struct sock *sk) 1181 { 1182 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1183 struct sk_buff *skb; 1184 unsigned int res; 1185 1186 __skb_queue_head_init(&list_kill); 1187 1188 spin_lock_bh(&rcvq->lock); 1189 while ((skb = skb_peek(rcvq)) != NULL && 1190 udp_lib_checksum_complete(skb)) { 1191 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1192 IS_UDPLITE(sk)); 1193 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1194 IS_UDPLITE(sk)); 1195 atomic_inc(&sk->sk_drops); 1196 __skb_unlink(skb, rcvq); 1197 __skb_queue_tail(&list_kill, skb); 1198 } 1199 res = skb ? skb->len : 0; 1200 spin_unlock_bh(&rcvq->lock); 1201 1202 if (!skb_queue_empty(&list_kill)) { 1203 bool slow = lock_sock_fast(sk); 1204 1205 __skb_queue_purge(&list_kill); 1206 sk_mem_reclaim_partial(sk); 1207 unlock_sock_fast(sk, slow); 1208 } 1209 return res; 1210 } 1211 1212 /* 1213 * IOCTL requests applicable to the UDP protocol 1214 */ 1215 1216 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1217 { 1218 switch (cmd) { 1219 case SIOCOUTQ: 1220 { 1221 int amount = sk_wmem_alloc_get(sk); 1222 1223 return put_user(amount, (int __user *)arg); 1224 } 1225 1226 case SIOCINQ: 1227 { 1228 unsigned int amount = first_packet_length(sk); 1229 1230 if (amount) 1231 /* 1232 * We will only return the amount 1233 * of this packet since that is all 1234 * that will be read. 1235 */ 1236 amount -= sizeof(struct udphdr); 1237 1238 return put_user(amount, (int __user *)arg); 1239 } 1240 1241 default: 1242 return -ENOIOCTLCMD; 1243 } 1244 1245 return 0; 1246 } 1247 EXPORT_SYMBOL(udp_ioctl); 1248 1249 /* 1250 * This should be easy, if there is something there we 1251 * return it, otherwise we block. 1252 */ 1253 1254 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1255 int flags, int *addr_len) 1256 { 1257 struct inet_sock *inet = inet_sk(sk); 1258 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1259 struct sk_buff *skb; 1260 unsigned int ulen, copied; 1261 int peeked, off = 0; 1262 int err; 1263 int is_udplite = IS_UDPLITE(sk); 1264 bool slow; 1265 1266 if (flags & MSG_ERRQUEUE) 1267 return ip_recv_error(sk, msg, len, addr_len); 1268 1269 try_again: 1270 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1271 &peeked, &off, &err); 1272 if (!skb) 1273 goto out; 1274 1275 ulen = skb->len - sizeof(struct udphdr); 1276 copied = len; 1277 if (copied > ulen) 1278 copied = ulen; 1279 else if (copied < ulen) 1280 msg->msg_flags |= MSG_TRUNC; 1281 1282 /* 1283 * If checksum is needed at all, try to do it while copying the 1284 * data. If the data is truncated, or if we only want a partial 1285 * coverage checksum (UDP-Lite), do it before the copy. 1286 */ 1287 1288 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1289 if (udp_lib_checksum_complete(skb)) 1290 goto csum_copy_err; 1291 } 1292 1293 if (skb_csum_unnecessary(skb)) 1294 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 1295 msg, copied); 1296 else { 1297 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), 1298 msg); 1299 1300 if (err == -EINVAL) 1301 goto csum_copy_err; 1302 } 1303 1304 if (unlikely(err)) { 1305 trace_kfree_skb(skb, udp_recvmsg); 1306 if (!peeked) { 1307 atomic_inc(&sk->sk_drops); 1308 UDP_INC_STATS_USER(sock_net(sk), 1309 UDP_MIB_INERRORS, is_udplite); 1310 } 1311 goto out_free; 1312 } 1313 1314 if (!peeked) 1315 UDP_INC_STATS_USER(sock_net(sk), 1316 UDP_MIB_INDATAGRAMS, is_udplite); 1317 1318 sock_recv_ts_and_drops(msg, sk, skb); 1319 1320 /* Copy the address. */ 1321 if (sin) { 1322 sin->sin_family = AF_INET; 1323 sin->sin_port = udp_hdr(skb)->source; 1324 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1325 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1326 *addr_len = sizeof(*sin); 1327 } 1328 if (inet->cmsg_flags) 1329 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr)); 1330 1331 err = copied; 1332 if (flags & MSG_TRUNC) 1333 err = ulen; 1334 1335 out_free: 1336 skb_free_datagram_locked(sk, skb); 1337 out: 1338 return err; 1339 1340 csum_copy_err: 1341 slow = lock_sock_fast(sk); 1342 if (!skb_kill_datagram(sk, skb, flags)) { 1343 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1344 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1345 } 1346 unlock_sock_fast(sk, slow); 1347 1348 /* starting over for a new packet, but check if we need to yield */ 1349 cond_resched(); 1350 msg->msg_flags &= ~MSG_TRUNC; 1351 goto try_again; 1352 } 1353 1354 int udp_disconnect(struct sock *sk, int flags) 1355 { 1356 struct inet_sock *inet = inet_sk(sk); 1357 /* 1358 * 1003.1g - break association. 1359 */ 1360 1361 sk->sk_state = TCP_CLOSE; 1362 inet->inet_daddr = 0; 1363 inet->inet_dport = 0; 1364 sock_rps_reset_rxhash(sk); 1365 sk->sk_bound_dev_if = 0; 1366 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1367 inet_reset_saddr(sk); 1368 1369 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1370 sk->sk_prot->unhash(sk); 1371 inet->inet_sport = 0; 1372 } 1373 sk_dst_reset(sk); 1374 return 0; 1375 } 1376 EXPORT_SYMBOL(udp_disconnect); 1377 1378 void udp_lib_unhash(struct sock *sk) 1379 { 1380 if (sk_hashed(sk)) { 1381 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1382 struct udp_hslot *hslot, *hslot2; 1383 1384 hslot = udp_hashslot(udptable, sock_net(sk), 1385 udp_sk(sk)->udp_port_hash); 1386 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1387 1388 spin_lock_bh(&hslot->lock); 1389 if (sk_nulls_del_node_init_rcu(sk)) { 1390 hslot->count--; 1391 inet_sk(sk)->inet_num = 0; 1392 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1393 1394 spin_lock(&hslot2->lock); 1395 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1396 hslot2->count--; 1397 spin_unlock(&hslot2->lock); 1398 } 1399 spin_unlock_bh(&hslot->lock); 1400 } 1401 } 1402 EXPORT_SYMBOL(udp_lib_unhash); 1403 1404 /* 1405 * inet_rcv_saddr was changed, we must rehash secondary hash 1406 */ 1407 void udp_lib_rehash(struct sock *sk, u16 newhash) 1408 { 1409 if (sk_hashed(sk)) { 1410 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1411 struct udp_hslot *hslot, *hslot2, *nhslot2; 1412 1413 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1414 nhslot2 = udp_hashslot2(udptable, newhash); 1415 udp_sk(sk)->udp_portaddr_hash = newhash; 1416 if (hslot2 != nhslot2) { 1417 hslot = udp_hashslot(udptable, sock_net(sk), 1418 udp_sk(sk)->udp_port_hash); 1419 /* we must lock primary chain too */ 1420 spin_lock_bh(&hslot->lock); 1421 1422 spin_lock(&hslot2->lock); 1423 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1424 hslot2->count--; 1425 spin_unlock(&hslot2->lock); 1426 1427 spin_lock(&nhslot2->lock); 1428 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1429 &nhslot2->head); 1430 nhslot2->count++; 1431 spin_unlock(&nhslot2->lock); 1432 1433 spin_unlock_bh(&hslot->lock); 1434 } 1435 } 1436 } 1437 EXPORT_SYMBOL(udp_lib_rehash); 1438 1439 static void udp_v4_rehash(struct sock *sk) 1440 { 1441 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1442 inet_sk(sk)->inet_rcv_saddr, 1443 inet_sk(sk)->inet_num); 1444 udp_lib_rehash(sk, new_hash); 1445 } 1446 1447 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1448 { 1449 int rc; 1450 1451 if (inet_sk(sk)->inet_daddr) { 1452 sock_rps_save_rxhash(sk, skb); 1453 sk_mark_napi_id(sk, skb); 1454 sk_incoming_cpu_update(sk); 1455 } 1456 1457 rc = sock_queue_rcv_skb(sk, skb); 1458 if (rc < 0) { 1459 int is_udplite = IS_UDPLITE(sk); 1460 1461 /* Note that an ENOMEM error is charged twice */ 1462 if (rc == -ENOMEM) 1463 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1464 is_udplite); 1465 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1466 kfree_skb(skb); 1467 trace_udp_fail_queue_rcv_skb(rc, sk); 1468 return -1; 1469 } 1470 1471 return 0; 1472 1473 } 1474 1475 static struct static_key udp_encap_needed __read_mostly; 1476 void udp_encap_enable(void) 1477 { 1478 if (!static_key_enabled(&udp_encap_needed)) 1479 static_key_slow_inc(&udp_encap_needed); 1480 } 1481 EXPORT_SYMBOL(udp_encap_enable); 1482 1483 /* returns: 1484 * -1: error 1485 * 0: success 1486 * >0: "udp encap" protocol resubmission 1487 * 1488 * Note that in the success and error cases, the skb is assumed to 1489 * have either been requeued or freed. 1490 */ 1491 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1492 { 1493 struct udp_sock *up = udp_sk(sk); 1494 int rc; 1495 int is_udplite = IS_UDPLITE(sk); 1496 1497 /* 1498 * Charge it to the socket, dropping if the queue is full. 1499 */ 1500 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1501 goto drop; 1502 nf_reset(skb); 1503 1504 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1505 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1506 1507 /* 1508 * This is an encapsulation socket so pass the skb to 1509 * the socket's udp_encap_rcv() hook. Otherwise, just 1510 * fall through and pass this up the UDP socket. 1511 * up->encap_rcv() returns the following value: 1512 * =0 if skb was successfully passed to the encap 1513 * handler or was discarded by it. 1514 * >0 if skb should be passed on to UDP. 1515 * <0 if skb should be resubmitted as proto -N 1516 */ 1517 1518 /* if we're overly short, let UDP handle it */ 1519 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1520 if (skb->len > sizeof(struct udphdr) && encap_rcv) { 1521 int ret; 1522 1523 /* Verify checksum before giving to encap */ 1524 if (udp_lib_checksum_complete(skb)) 1525 goto csum_error; 1526 1527 ret = encap_rcv(sk, skb); 1528 if (ret <= 0) { 1529 UDP_INC_STATS_BH(sock_net(sk), 1530 UDP_MIB_INDATAGRAMS, 1531 is_udplite); 1532 return -ret; 1533 } 1534 } 1535 1536 /* FALLTHROUGH -- it's a UDP Packet */ 1537 } 1538 1539 /* 1540 * UDP-Lite specific tests, ignored on UDP sockets 1541 */ 1542 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1543 1544 /* 1545 * MIB statistics other than incrementing the error count are 1546 * disabled for the following two types of errors: these depend 1547 * on the application settings, not on the functioning of the 1548 * protocol stack as such. 1549 * 1550 * RFC 3828 here recommends (sec 3.3): "There should also be a 1551 * way ... to ... at least let the receiving application block 1552 * delivery of packets with coverage values less than a value 1553 * provided by the application." 1554 */ 1555 if (up->pcrlen == 0) { /* full coverage was set */ 1556 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1557 UDP_SKB_CB(skb)->cscov, skb->len); 1558 goto drop; 1559 } 1560 /* The next case involves violating the min. coverage requested 1561 * by the receiver. This is subtle: if receiver wants x and x is 1562 * greater than the buffersize/MTU then receiver will complain 1563 * that it wants x while sender emits packets of smaller size y. 1564 * Therefore the above ...()->partial_cov statement is essential. 1565 */ 1566 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1567 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1568 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1569 goto drop; 1570 } 1571 } 1572 1573 if (rcu_access_pointer(sk->sk_filter) && 1574 udp_lib_checksum_complete(skb)) 1575 goto csum_error; 1576 1577 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1578 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1579 is_udplite); 1580 goto drop; 1581 } 1582 1583 rc = 0; 1584 1585 ipv4_pktinfo_prepare(sk, skb); 1586 bh_lock_sock(sk); 1587 if (!sock_owned_by_user(sk)) 1588 rc = __udp_queue_rcv_skb(sk, skb); 1589 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1590 bh_unlock_sock(sk); 1591 goto drop; 1592 } 1593 bh_unlock_sock(sk); 1594 1595 return rc; 1596 1597 csum_error: 1598 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1599 drop: 1600 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1601 atomic_inc(&sk->sk_drops); 1602 kfree_skb(skb); 1603 return -1; 1604 } 1605 1606 static void flush_stack(struct sock **stack, unsigned int count, 1607 struct sk_buff *skb, unsigned int final) 1608 { 1609 unsigned int i; 1610 struct sk_buff *skb1 = NULL; 1611 struct sock *sk; 1612 1613 for (i = 0; i < count; i++) { 1614 sk = stack[i]; 1615 if (likely(!skb1)) 1616 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1617 1618 if (!skb1) { 1619 atomic_inc(&sk->sk_drops); 1620 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1621 IS_UDPLITE(sk)); 1622 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1623 IS_UDPLITE(sk)); 1624 } 1625 1626 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1627 skb1 = NULL; 1628 1629 sock_put(sk); 1630 } 1631 if (unlikely(skb1)) 1632 kfree_skb(skb1); 1633 } 1634 1635 /* For TCP sockets, sk_rx_dst is protected by socket lock 1636 * For UDP, we use xchg() to guard against concurrent changes. 1637 */ 1638 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1639 { 1640 struct dst_entry *old; 1641 1642 dst_hold(dst); 1643 old = xchg(&sk->sk_rx_dst, dst); 1644 dst_release(old); 1645 } 1646 1647 /* 1648 * Multicasts and broadcasts go to each listener. 1649 * 1650 * Note: called only from the BH handler context. 1651 */ 1652 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1653 struct udphdr *uh, 1654 __be32 saddr, __be32 daddr, 1655 struct udp_table *udptable, 1656 int proto) 1657 { 1658 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1659 struct hlist_nulls_node *node; 1660 unsigned short hnum = ntohs(uh->dest); 1661 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1662 int dif = skb->dev->ifindex; 1663 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 1664 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1665 bool inner_flushed = false; 1666 1667 if (use_hash2) { 1668 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1669 udp_table.mask; 1670 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask; 1671 start_lookup: 1672 hslot = &udp_table.hash2[hash2]; 1673 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1674 } 1675 1676 spin_lock(&hslot->lock); 1677 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 1678 if (__udp_is_mcast_sock(net, sk, 1679 uh->dest, daddr, 1680 uh->source, saddr, 1681 dif, hnum)) { 1682 if (unlikely(count == ARRAY_SIZE(stack))) { 1683 flush_stack(stack, count, skb, ~0); 1684 inner_flushed = true; 1685 count = 0; 1686 } 1687 stack[count++] = sk; 1688 sock_hold(sk); 1689 } 1690 } 1691 1692 spin_unlock(&hslot->lock); 1693 1694 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1695 if (use_hash2 && hash2 != hash2_any) { 1696 hash2 = hash2_any; 1697 goto start_lookup; 1698 } 1699 1700 /* 1701 * do the slow work with no lock held 1702 */ 1703 if (count) { 1704 flush_stack(stack, count, skb, count - 1); 1705 } else { 1706 if (!inner_flushed) 1707 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 1708 proto == IPPROTO_UDPLITE); 1709 consume_skb(skb); 1710 } 1711 return 0; 1712 } 1713 1714 /* Initialize UDP checksum. If exited with zero value (success), 1715 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1716 * Otherwise, csum completion requires chacksumming packet body, 1717 * including udp header and folding it to skb->csum. 1718 */ 1719 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1720 int proto) 1721 { 1722 int err; 1723 1724 UDP_SKB_CB(skb)->partial_cov = 0; 1725 UDP_SKB_CB(skb)->cscov = skb->len; 1726 1727 if (proto == IPPROTO_UDPLITE) { 1728 err = udplite_checksum_init(skb, uh); 1729 if (err) 1730 return err; 1731 } 1732 1733 return skb_checksum_init_zero_check(skb, proto, uh->check, 1734 inet_compute_pseudo); 1735 } 1736 1737 /* 1738 * All we need to do is get the socket, and then do a checksum. 1739 */ 1740 1741 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1742 int proto) 1743 { 1744 struct sock *sk; 1745 struct udphdr *uh; 1746 unsigned short ulen; 1747 struct rtable *rt = skb_rtable(skb); 1748 __be32 saddr, daddr; 1749 struct net *net = dev_net(skb->dev); 1750 1751 /* 1752 * Validate the packet. 1753 */ 1754 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1755 goto drop; /* No space for header. */ 1756 1757 uh = udp_hdr(skb); 1758 ulen = ntohs(uh->len); 1759 saddr = ip_hdr(skb)->saddr; 1760 daddr = ip_hdr(skb)->daddr; 1761 1762 if (ulen > skb->len) 1763 goto short_packet; 1764 1765 if (proto == IPPROTO_UDP) { 1766 /* UDP validates ulen. */ 1767 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1768 goto short_packet; 1769 uh = udp_hdr(skb); 1770 } 1771 1772 if (udp4_csum_init(skb, uh, proto)) 1773 goto csum_error; 1774 1775 sk = skb_steal_sock(skb); 1776 if (sk) { 1777 struct dst_entry *dst = skb_dst(skb); 1778 int ret; 1779 1780 if (unlikely(sk->sk_rx_dst != dst)) 1781 udp_sk_rx_dst_set(sk, dst); 1782 1783 ret = udp_queue_rcv_skb(sk, skb); 1784 sock_put(sk); 1785 /* a return value > 0 means to resubmit the input, but 1786 * it wants the return to be -protocol, or 0 1787 */ 1788 if (ret > 0) 1789 return -ret; 1790 return 0; 1791 } 1792 1793 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1794 return __udp4_lib_mcast_deliver(net, skb, uh, 1795 saddr, daddr, udptable, proto); 1796 1797 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1798 if (sk) { 1799 int ret; 1800 1801 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1802 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1803 inet_compute_pseudo); 1804 1805 ret = udp_queue_rcv_skb(sk, skb); 1806 sock_put(sk); 1807 1808 /* a return value > 0 means to resubmit the input, but 1809 * it wants the return to be -protocol, or 0 1810 */ 1811 if (ret > 0) 1812 return -ret; 1813 return 0; 1814 } 1815 1816 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1817 goto drop; 1818 nf_reset(skb); 1819 1820 /* No socket. Drop packet silently, if checksum is wrong */ 1821 if (udp_lib_checksum_complete(skb)) 1822 goto csum_error; 1823 1824 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1825 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1826 1827 /* 1828 * Hmm. We got an UDP packet to a port to which we 1829 * don't wanna listen. Ignore it. 1830 */ 1831 kfree_skb(skb); 1832 return 0; 1833 1834 short_packet: 1835 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1836 proto == IPPROTO_UDPLITE ? "Lite" : "", 1837 &saddr, ntohs(uh->source), 1838 ulen, skb->len, 1839 &daddr, ntohs(uh->dest)); 1840 goto drop; 1841 1842 csum_error: 1843 /* 1844 * RFC1122: OK. Discards the bad packet silently (as far as 1845 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1846 */ 1847 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1848 proto == IPPROTO_UDPLITE ? "Lite" : "", 1849 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1850 ulen); 1851 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1852 drop: 1853 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1854 kfree_skb(skb); 1855 return 0; 1856 } 1857 1858 /* We can only early demux multicast if there is a single matching socket. 1859 * If more than one socket found returns NULL 1860 */ 1861 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1862 __be16 loc_port, __be32 loc_addr, 1863 __be16 rmt_port, __be32 rmt_addr, 1864 int dif) 1865 { 1866 struct sock *sk, *result; 1867 struct hlist_nulls_node *node; 1868 unsigned short hnum = ntohs(loc_port); 1869 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1870 struct udp_hslot *hslot = &udp_table.hash[slot]; 1871 1872 /* Do not bother scanning a too big list */ 1873 if (hslot->count > 10) 1874 return NULL; 1875 1876 rcu_read_lock(); 1877 begin: 1878 count = 0; 1879 result = NULL; 1880 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1881 if (__udp_is_mcast_sock(net, sk, 1882 loc_port, loc_addr, 1883 rmt_port, rmt_addr, 1884 dif, hnum)) { 1885 result = sk; 1886 ++count; 1887 } 1888 } 1889 /* 1890 * if the nulls value we got at the end of this lookup is 1891 * not the expected one, we must restart lookup. 1892 * We probably met an item that was moved to another chain. 1893 */ 1894 if (get_nulls_value(node) != slot) 1895 goto begin; 1896 1897 if (result) { 1898 if (count != 1 || 1899 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1900 result = NULL; 1901 else if (unlikely(!__udp_is_mcast_sock(net, result, 1902 loc_port, loc_addr, 1903 rmt_port, rmt_addr, 1904 dif, hnum))) { 1905 sock_put(result); 1906 result = NULL; 1907 } 1908 } 1909 rcu_read_unlock(); 1910 return result; 1911 } 1912 1913 /* For unicast we should only early demux connected sockets or we can 1914 * break forwarding setups. The chains here can be long so only check 1915 * if the first socket is an exact match and if not move on. 1916 */ 1917 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1918 __be16 loc_port, __be32 loc_addr, 1919 __be16 rmt_port, __be32 rmt_addr, 1920 int dif) 1921 { 1922 struct sock *sk, *result; 1923 struct hlist_nulls_node *node; 1924 unsigned short hnum = ntohs(loc_port); 1925 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1926 unsigned int slot2 = hash2 & udp_table.mask; 1927 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1928 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 1929 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1930 1931 rcu_read_lock(); 1932 result = NULL; 1933 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 1934 if (INET_MATCH(sk, net, acookie, 1935 rmt_addr, loc_addr, ports, dif)) 1936 result = sk; 1937 /* Only check first socket in chain */ 1938 break; 1939 } 1940 1941 if (result) { 1942 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1943 result = NULL; 1944 else if (unlikely(!INET_MATCH(sk, net, acookie, 1945 rmt_addr, loc_addr, 1946 ports, dif))) { 1947 sock_put(result); 1948 result = NULL; 1949 } 1950 } 1951 rcu_read_unlock(); 1952 return result; 1953 } 1954 1955 void udp_v4_early_demux(struct sk_buff *skb) 1956 { 1957 struct net *net = dev_net(skb->dev); 1958 const struct iphdr *iph; 1959 const struct udphdr *uh; 1960 struct sock *sk; 1961 struct dst_entry *dst; 1962 int dif = skb->dev->ifindex; 1963 1964 /* validate the packet */ 1965 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1966 return; 1967 1968 iph = ip_hdr(skb); 1969 uh = udp_hdr(skb); 1970 1971 if (skb->pkt_type == PACKET_BROADCAST || 1972 skb->pkt_type == PACKET_MULTICAST) 1973 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1974 uh->source, iph->saddr, dif); 1975 else if (skb->pkt_type == PACKET_HOST) 1976 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1977 uh->source, iph->saddr, dif); 1978 else 1979 return; 1980 1981 if (!sk) 1982 return; 1983 1984 skb->sk = sk; 1985 skb->destructor = sock_efree; 1986 dst = sk->sk_rx_dst; 1987 1988 if (dst) 1989 dst = dst_check(dst, 0); 1990 if (dst) 1991 skb_dst_set_noref(skb, dst); 1992 } 1993 1994 int udp_rcv(struct sk_buff *skb) 1995 { 1996 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1997 } 1998 1999 void udp_destroy_sock(struct sock *sk) 2000 { 2001 struct udp_sock *up = udp_sk(sk); 2002 bool slow = lock_sock_fast(sk); 2003 udp_flush_pending_frames(sk); 2004 unlock_sock_fast(sk, slow); 2005 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2006 void (*encap_destroy)(struct sock *sk); 2007 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2008 if (encap_destroy) 2009 encap_destroy(sk); 2010 } 2011 } 2012 2013 /* 2014 * Socket option code for UDP 2015 */ 2016 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2017 char __user *optval, unsigned int optlen, 2018 int (*push_pending_frames)(struct sock *)) 2019 { 2020 struct udp_sock *up = udp_sk(sk); 2021 int val, valbool; 2022 int err = 0; 2023 int is_udplite = IS_UDPLITE(sk); 2024 2025 if (optlen < sizeof(int)) 2026 return -EINVAL; 2027 2028 if (get_user(val, (int __user *)optval)) 2029 return -EFAULT; 2030 2031 valbool = val ? 1 : 0; 2032 2033 switch (optname) { 2034 case UDP_CORK: 2035 if (val != 0) { 2036 up->corkflag = 1; 2037 } else { 2038 up->corkflag = 0; 2039 lock_sock(sk); 2040 push_pending_frames(sk); 2041 release_sock(sk); 2042 } 2043 break; 2044 2045 case UDP_ENCAP: 2046 switch (val) { 2047 case 0: 2048 case UDP_ENCAP_ESPINUDP: 2049 case UDP_ENCAP_ESPINUDP_NON_IKE: 2050 up->encap_rcv = xfrm4_udp_encap_rcv; 2051 /* FALLTHROUGH */ 2052 case UDP_ENCAP_L2TPINUDP: 2053 up->encap_type = val; 2054 udp_encap_enable(); 2055 break; 2056 default: 2057 err = -ENOPROTOOPT; 2058 break; 2059 } 2060 break; 2061 2062 case UDP_NO_CHECK6_TX: 2063 up->no_check6_tx = valbool; 2064 break; 2065 2066 case UDP_NO_CHECK6_RX: 2067 up->no_check6_rx = valbool; 2068 break; 2069 2070 /* 2071 * UDP-Lite's partial checksum coverage (RFC 3828). 2072 */ 2073 /* The sender sets actual checksum coverage length via this option. 2074 * The case coverage > packet length is handled by send module. */ 2075 case UDPLITE_SEND_CSCOV: 2076 if (!is_udplite) /* Disable the option on UDP sockets */ 2077 return -ENOPROTOOPT; 2078 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2079 val = 8; 2080 else if (val > USHRT_MAX) 2081 val = USHRT_MAX; 2082 up->pcslen = val; 2083 up->pcflag |= UDPLITE_SEND_CC; 2084 break; 2085 2086 /* The receiver specifies a minimum checksum coverage value. To make 2087 * sense, this should be set to at least 8 (as done below). If zero is 2088 * used, this again means full checksum coverage. */ 2089 case UDPLITE_RECV_CSCOV: 2090 if (!is_udplite) /* Disable the option on UDP sockets */ 2091 return -ENOPROTOOPT; 2092 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2093 val = 8; 2094 else if (val > USHRT_MAX) 2095 val = USHRT_MAX; 2096 up->pcrlen = val; 2097 up->pcflag |= UDPLITE_RECV_CC; 2098 break; 2099 2100 default: 2101 err = -ENOPROTOOPT; 2102 break; 2103 } 2104 2105 return err; 2106 } 2107 EXPORT_SYMBOL(udp_lib_setsockopt); 2108 2109 int udp_setsockopt(struct sock *sk, int level, int optname, 2110 char __user *optval, unsigned int optlen) 2111 { 2112 if (level == SOL_UDP || level == SOL_UDPLITE) 2113 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2114 udp_push_pending_frames); 2115 return ip_setsockopt(sk, level, optname, optval, optlen); 2116 } 2117 2118 #ifdef CONFIG_COMPAT 2119 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2120 char __user *optval, unsigned int optlen) 2121 { 2122 if (level == SOL_UDP || level == SOL_UDPLITE) 2123 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2124 udp_push_pending_frames); 2125 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2126 } 2127 #endif 2128 2129 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2130 char __user *optval, int __user *optlen) 2131 { 2132 struct udp_sock *up = udp_sk(sk); 2133 int val, len; 2134 2135 if (get_user(len, optlen)) 2136 return -EFAULT; 2137 2138 len = min_t(unsigned int, len, sizeof(int)); 2139 2140 if (len < 0) 2141 return -EINVAL; 2142 2143 switch (optname) { 2144 case UDP_CORK: 2145 val = up->corkflag; 2146 break; 2147 2148 case UDP_ENCAP: 2149 val = up->encap_type; 2150 break; 2151 2152 case UDP_NO_CHECK6_TX: 2153 val = up->no_check6_tx; 2154 break; 2155 2156 case UDP_NO_CHECK6_RX: 2157 val = up->no_check6_rx; 2158 break; 2159 2160 /* The following two cannot be changed on UDP sockets, the return is 2161 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2162 case UDPLITE_SEND_CSCOV: 2163 val = up->pcslen; 2164 break; 2165 2166 case UDPLITE_RECV_CSCOV: 2167 val = up->pcrlen; 2168 break; 2169 2170 default: 2171 return -ENOPROTOOPT; 2172 } 2173 2174 if (put_user(len, optlen)) 2175 return -EFAULT; 2176 if (copy_to_user(optval, &val, len)) 2177 return -EFAULT; 2178 return 0; 2179 } 2180 EXPORT_SYMBOL(udp_lib_getsockopt); 2181 2182 int udp_getsockopt(struct sock *sk, int level, int optname, 2183 char __user *optval, int __user *optlen) 2184 { 2185 if (level == SOL_UDP || level == SOL_UDPLITE) 2186 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2187 return ip_getsockopt(sk, level, optname, optval, optlen); 2188 } 2189 2190 #ifdef CONFIG_COMPAT 2191 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2192 char __user *optval, int __user *optlen) 2193 { 2194 if (level == SOL_UDP || level == SOL_UDPLITE) 2195 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2196 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2197 } 2198 #endif 2199 /** 2200 * udp_poll - wait for a UDP event. 2201 * @file - file struct 2202 * @sock - socket 2203 * @wait - poll table 2204 * 2205 * This is same as datagram poll, except for the special case of 2206 * blocking sockets. If application is using a blocking fd 2207 * and a packet with checksum error is in the queue; 2208 * then it could get return from select indicating data available 2209 * but then block when reading it. Add special case code 2210 * to work around these arguably broken applications. 2211 */ 2212 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2213 { 2214 unsigned int mask = datagram_poll(file, sock, wait); 2215 struct sock *sk = sock->sk; 2216 2217 sock_rps_record_flow(sk); 2218 2219 /* Check for false positives due to checksum errors */ 2220 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2221 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2222 mask &= ~(POLLIN | POLLRDNORM); 2223 2224 return mask; 2225 2226 } 2227 EXPORT_SYMBOL(udp_poll); 2228 2229 struct proto udp_prot = { 2230 .name = "UDP", 2231 .owner = THIS_MODULE, 2232 .close = udp_lib_close, 2233 .connect = ip4_datagram_connect, 2234 .disconnect = udp_disconnect, 2235 .ioctl = udp_ioctl, 2236 .destroy = udp_destroy_sock, 2237 .setsockopt = udp_setsockopt, 2238 .getsockopt = udp_getsockopt, 2239 .sendmsg = udp_sendmsg, 2240 .recvmsg = udp_recvmsg, 2241 .sendpage = udp_sendpage, 2242 .backlog_rcv = __udp_queue_rcv_skb, 2243 .release_cb = ip4_datagram_release_cb, 2244 .hash = udp_lib_hash, 2245 .unhash = udp_lib_unhash, 2246 .rehash = udp_v4_rehash, 2247 .get_port = udp_v4_get_port, 2248 .memory_allocated = &udp_memory_allocated, 2249 .sysctl_mem = sysctl_udp_mem, 2250 .sysctl_wmem = &sysctl_udp_wmem_min, 2251 .sysctl_rmem = &sysctl_udp_rmem_min, 2252 .obj_size = sizeof(struct udp_sock), 2253 .slab_flags = SLAB_DESTROY_BY_RCU, 2254 .h.udp_table = &udp_table, 2255 #ifdef CONFIG_COMPAT 2256 .compat_setsockopt = compat_udp_setsockopt, 2257 .compat_getsockopt = compat_udp_getsockopt, 2258 #endif 2259 .clear_sk = sk_prot_clear_portaddr_nulls, 2260 }; 2261 EXPORT_SYMBOL(udp_prot); 2262 2263 /* ------------------------------------------------------------------------ */ 2264 #ifdef CONFIG_PROC_FS 2265 2266 static struct sock *udp_get_first(struct seq_file *seq, int start) 2267 { 2268 struct sock *sk; 2269 struct udp_iter_state *state = seq->private; 2270 struct net *net = seq_file_net(seq); 2271 2272 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2273 ++state->bucket) { 2274 struct hlist_nulls_node *node; 2275 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2276 2277 if (hlist_nulls_empty(&hslot->head)) 2278 continue; 2279 2280 spin_lock_bh(&hslot->lock); 2281 sk_nulls_for_each(sk, node, &hslot->head) { 2282 if (!net_eq(sock_net(sk), net)) 2283 continue; 2284 if (sk->sk_family == state->family) 2285 goto found; 2286 } 2287 spin_unlock_bh(&hslot->lock); 2288 } 2289 sk = NULL; 2290 found: 2291 return sk; 2292 } 2293 2294 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2295 { 2296 struct udp_iter_state *state = seq->private; 2297 struct net *net = seq_file_net(seq); 2298 2299 do { 2300 sk = sk_nulls_next(sk); 2301 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2302 2303 if (!sk) { 2304 if (state->bucket <= state->udp_table->mask) 2305 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2306 return udp_get_first(seq, state->bucket + 1); 2307 } 2308 return sk; 2309 } 2310 2311 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2312 { 2313 struct sock *sk = udp_get_first(seq, 0); 2314 2315 if (sk) 2316 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2317 --pos; 2318 return pos ? NULL : sk; 2319 } 2320 2321 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2322 { 2323 struct udp_iter_state *state = seq->private; 2324 state->bucket = MAX_UDP_PORTS; 2325 2326 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2327 } 2328 2329 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2330 { 2331 struct sock *sk; 2332 2333 if (v == SEQ_START_TOKEN) 2334 sk = udp_get_idx(seq, 0); 2335 else 2336 sk = udp_get_next(seq, v); 2337 2338 ++*pos; 2339 return sk; 2340 } 2341 2342 static void udp_seq_stop(struct seq_file *seq, void *v) 2343 { 2344 struct udp_iter_state *state = seq->private; 2345 2346 if (state->bucket <= state->udp_table->mask) 2347 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2348 } 2349 2350 int udp_seq_open(struct inode *inode, struct file *file) 2351 { 2352 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2353 struct udp_iter_state *s; 2354 int err; 2355 2356 err = seq_open_net(inode, file, &afinfo->seq_ops, 2357 sizeof(struct udp_iter_state)); 2358 if (err < 0) 2359 return err; 2360 2361 s = ((struct seq_file *)file->private_data)->private; 2362 s->family = afinfo->family; 2363 s->udp_table = afinfo->udp_table; 2364 return err; 2365 } 2366 EXPORT_SYMBOL(udp_seq_open); 2367 2368 /* ------------------------------------------------------------------------ */ 2369 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2370 { 2371 struct proc_dir_entry *p; 2372 int rc = 0; 2373 2374 afinfo->seq_ops.start = udp_seq_start; 2375 afinfo->seq_ops.next = udp_seq_next; 2376 afinfo->seq_ops.stop = udp_seq_stop; 2377 2378 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2379 afinfo->seq_fops, afinfo); 2380 if (!p) 2381 rc = -ENOMEM; 2382 return rc; 2383 } 2384 EXPORT_SYMBOL(udp_proc_register); 2385 2386 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2387 { 2388 remove_proc_entry(afinfo->name, net->proc_net); 2389 } 2390 EXPORT_SYMBOL(udp_proc_unregister); 2391 2392 /* ------------------------------------------------------------------------ */ 2393 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2394 int bucket) 2395 { 2396 struct inet_sock *inet = inet_sk(sp); 2397 __be32 dest = inet->inet_daddr; 2398 __be32 src = inet->inet_rcv_saddr; 2399 __u16 destp = ntohs(inet->inet_dport); 2400 __u16 srcp = ntohs(inet->inet_sport); 2401 2402 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2403 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2404 bucket, src, srcp, dest, destp, sp->sk_state, 2405 sk_wmem_alloc_get(sp), 2406 sk_rmem_alloc_get(sp), 2407 0, 0L, 0, 2408 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2409 0, sock_i_ino(sp), 2410 atomic_read(&sp->sk_refcnt), sp, 2411 atomic_read(&sp->sk_drops)); 2412 } 2413 2414 int udp4_seq_show(struct seq_file *seq, void *v) 2415 { 2416 seq_setwidth(seq, 127); 2417 if (v == SEQ_START_TOKEN) 2418 seq_puts(seq, " sl local_address rem_address st tx_queue " 2419 "rx_queue tr tm->when retrnsmt uid timeout " 2420 "inode ref pointer drops"); 2421 else { 2422 struct udp_iter_state *state = seq->private; 2423 2424 udp4_format_sock(v, seq, state->bucket); 2425 } 2426 seq_pad(seq, '\n'); 2427 return 0; 2428 } 2429 2430 static const struct file_operations udp_afinfo_seq_fops = { 2431 .owner = THIS_MODULE, 2432 .open = udp_seq_open, 2433 .read = seq_read, 2434 .llseek = seq_lseek, 2435 .release = seq_release_net 2436 }; 2437 2438 /* ------------------------------------------------------------------------ */ 2439 static struct udp_seq_afinfo udp4_seq_afinfo = { 2440 .name = "udp", 2441 .family = AF_INET, 2442 .udp_table = &udp_table, 2443 .seq_fops = &udp_afinfo_seq_fops, 2444 .seq_ops = { 2445 .show = udp4_seq_show, 2446 }, 2447 }; 2448 2449 static int __net_init udp4_proc_init_net(struct net *net) 2450 { 2451 return udp_proc_register(net, &udp4_seq_afinfo); 2452 } 2453 2454 static void __net_exit udp4_proc_exit_net(struct net *net) 2455 { 2456 udp_proc_unregister(net, &udp4_seq_afinfo); 2457 } 2458 2459 static struct pernet_operations udp4_net_ops = { 2460 .init = udp4_proc_init_net, 2461 .exit = udp4_proc_exit_net, 2462 }; 2463 2464 int __init udp4_proc_init(void) 2465 { 2466 return register_pernet_subsys(&udp4_net_ops); 2467 } 2468 2469 void udp4_proc_exit(void) 2470 { 2471 unregister_pernet_subsys(&udp4_net_ops); 2472 } 2473 #endif /* CONFIG_PROC_FS */ 2474 2475 static __initdata unsigned long uhash_entries; 2476 static int __init set_uhash_entries(char *str) 2477 { 2478 ssize_t ret; 2479 2480 if (!str) 2481 return 0; 2482 2483 ret = kstrtoul(str, 0, &uhash_entries); 2484 if (ret) 2485 return 0; 2486 2487 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2488 uhash_entries = UDP_HTABLE_SIZE_MIN; 2489 return 1; 2490 } 2491 __setup("uhash_entries=", set_uhash_entries); 2492 2493 void __init udp_table_init(struct udp_table *table, const char *name) 2494 { 2495 unsigned int i; 2496 2497 table->hash = alloc_large_system_hash(name, 2498 2 * sizeof(struct udp_hslot), 2499 uhash_entries, 2500 21, /* one slot per 2 MB */ 2501 0, 2502 &table->log, 2503 &table->mask, 2504 UDP_HTABLE_SIZE_MIN, 2505 64 * 1024); 2506 2507 table->hash2 = table->hash + (table->mask + 1); 2508 for (i = 0; i <= table->mask; i++) { 2509 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2510 table->hash[i].count = 0; 2511 spin_lock_init(&table->hash[i].lock); 2512 } 2513 for (i = 0; i <= table->mask; i++) { 2514 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2515 table->hash2[i].count = 0; 2516 spin_lock_init(&table->hash2[i].lock); 2517 } 2518 } 2519 2520 u32 udp_flow_hashrnd(void) 2521 { 2522 static u32 hashrnd __read_mostly; 2523 2524 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2525 2526 return hashrnd; 2527 } 2528 EXPORT_SYMBOL(udp_flow_hashrnd); 2529 2530 void __init udp_init(void) 2531 { 2532 unsigned long limit; 2533 2534 udp_table_init(&udp_table, "UDP"); 2535 limit = nr_free_buffer_pages() / 8; 2536 limit = max(limit, 128UL); 2537 sysctl_udp_mem[0] = limit / 4 * 3; 2538 sysctl_udp_mem[1] = limit; 2539 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2540 2541 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2542 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2543 } 2544