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/inetdevice.h> 94 #include <linux/in.h> 95 #include <linux/errno.h> 96 #include <linux/timer.h> 97 #include <linux/mm.h> 98 #include <linux/inet.h> 99 #include <linux/netdevice.h> 100 #include <linux/slab.h> 101 #include <net/tcp_states.h> 102 #include <linux/skbuff.h> 103 #include <linux/netdevice.h> 104 #include <linux/proc_fs.h> 105 #include <linux/seq_file.h> 106 #include <net/net_namespace.h> 107 #include <net/icmp.h> 108 #include <net/inet_hashtables.h> 109 #include <net/route.h> 110 #include <net/checksum.h> 111 #include <net/xfrm.h> 112 #include <trace/events/udp.h> 113 #include <linux/static_key.h> 114 #include <trace/events/skb.h> 115 #include <net/busy_poll.h> 116 #include "udp_impl.h" 117 118 struct udp_table udp_table __read_mostly; 119 EXPORT_SYMBOL(udp_table); 120 121 long sysctl_udp_mem[3] __read_mostly; 122 EXPORT_SYMBOL(sysctl_udp_mem); 123 124 int sysctl_udp_rmem_min __read_mostly; 125 EXPORT_SYMBOL(sysctl_udp_rmem_min); 126 127 int sysctl_udp_wmem_min __read_mostly; 128 EXPORT_SYMBOL(sysctl_udp_wmem_min); 129 130 atomic_long_t udp_memory_allocated; 131 EXPORT_SYMBOL(udp_memory_allocated); 132 133 #define MAX_UDP_PORTS 65536 134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 135 136 static int udp_lib_lport_inuse(struct net *net, __u16 num, 137 const struct udp_hslot *hslot, 138 unsigned long *bitmap, 139 struct sock *sk, 140 int (*saddr_comp)(const struct sock *sk1, 141 const struct sock *sk2), 142 unsigned int log) 143 { 144 struct sock *sk2; 145 struct hlist_nulls_node *node; 146 kuid_t uid = sock_i_uid(sk); 147 148 sk_nulls_for_each(sk2, node, &hslot->head) { 149 if (net_eq(sock_net(sk2), net) && 150 sk2 != sk && 151 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 152 (!sk2->sk_reuse || !sk->sk_reuse) && 153 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 154 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 155 (!sk2->sk_reuseport || !sk->sk_reuseport || 156 !uid_eq(uid, sock_i_uid(sk2))) && 157 saddr_comp(sk, sk2)) { 158 if (!bitmap) 159 return 1; 160 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap); 161 } 162 } 163 return 0; 164 } 165 166 /* 167 * Note: we still hold spinlock of primary hash chain, so no other writer 168 * can insert/delete a socket with local_port == num 169 */ 170 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 171 struct udp_hslot *hslot2, 172 struct sock *sk, 173 int (*saddr_comp)(const struct sock *sk1, 174 const struct sock *sk2)) 175 { 176 struct sock *sk2; 177 struct hlist_nulls_node *node; 178 kuid_t uid = sock_i_uid(sk); 179 int res = 0; 180 181 spin_lock(&hslot2->lock); 182 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) { 183 if (net_eq(sock_net(sk2), net) && 184 sk2 != sk && 185 (udp_sk(sk2)->udp_port_hash == num) && 186 (!sk2->sk_reuse || !sk->sk_reuse) && 187 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 188 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 189 (!sk2->sk_reuseport || !sk->sk_reuseport || 190 !uid_eq(uid, sock_i_uid(sk2))) && 191 saddr_comp(sk, sk2)) { 192 res = 1; 193 break; 194 } 195 } 196 spin_unlock(&hslot2->lock); 197 return res; 198 } 199 200 /** 201 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 202 * 203 * @sk: socket struct in question 204 * @snum: port number to look up 205 * @saddr_comp: AF-dependent comparison of bound local IP addresses 206 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 207 * with NULL address 208 */ 209 int udp_lib_get_port(struct sock *sk, unsigned short snum, 210 int (*saddr_comp)(const struct sock *sk1, 211 const struct sock *sk2), 212 unsigned int hash2_nulladdr) 213 { 214 struct udp_hslot *hslot, *hslot2; 215 struct udp_table *udptable = sk->sk_prot->h.udp_table; 216 int error = 1; 217 struct net *net = sock_net(sk); 218 219 if (!snum) { 220 int low, high, remaining; 221 unsigned int rand; 222 unsigned short first, last; 223 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 224 225 inet_get_local_port_range(net, &low, &high); 226 remaining = (high - low) + 1; 227 228 rand = prandom_u32(); 229 first = reciprocal_scale(rand, remaining) + low; 230 /* 231 * force rand to be an odd multiple of UDP_HTABLE_SIZE 232 */ 233 rand = (rand | 1) * (udptable->mask + 1); 234 last = first + udptable->mask + 1; 235 do { 236 hslot = udp_hashslot(udptable, net, first); 237 bitmap_zero(bitmap, PORTS_PER_CHAIN); 238 spin_lock_bh(&hslot->lock); 239 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 240 saddr_comp, udptable->log); 241 242 snum = first; 243 /* 244 * Iterate on all possible values of snum for this hash. 245 * Using steps of an odd multiple of UDP_HTABLE_SIZE 246 * give us randomization and full range coverage. 247 */ 248 do { 249 if (low <= snum && snum <= high && 250 !test_bit(snum >> udptable->log, bitmap) && 251 !inet_is_local_reserved_port(net, snum)) 252 goto found; 253 snum += rand; 254 } while (snum != first); 255 spin_unlock_bh(&hslot->lock); 256 } while (++first != last); 257 goto fail; 258 } else { 259 hslot = udp_hashslot(udptable, net, snum); 260 spin_lock_bh(&hslot->lock); 261 if (hslot->count > 10) { 262 int exist; 263 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 264 265 slot2 &= udptable->mask; 266 hash2_nulladdr &= udptable->mask; 267 268 hslot2 = udp_hashslot2(udptable, slot2); 269 if (hslot->count < hslot2->count) 270 goto scan_primary_hash; 271 272 exist = udp_lib_lport_inuse2(net, snum, hslot2, 273 sk, saddr_comp); 274 if (!exist && (hash2_nulladdr != slot2)) { 275 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 276 exist = udp_lib_lport_inuse2(net, snum, hslot2, 277 sk, saddr_comp); 278 } 279 if (exist) 280 goto fail_unlock; 281 else 282 goto found; 283 } 284 scan_primary_hash: 285 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 286 saddr_comp, 0)) 287 goto fail_unlock; 288 } 289 found: 290 inet_sk(sk)->inet_num = snum; 291 udp_sk(sk)->udp_port_hash = snum; 292 udp_sk(sk)->udp_portaddr_hash ^= snum; 293 if (sk_unhashed(sk)) { 294 sk_nulls_add_node_rcu(sk, &hslot->head); 295 hslot->count++; 296 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 297 298 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 299 spin_lock(&hslot2->lock); 300 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 301 &hslot2->head); 302 hslot2->count++; 303 spin_unlock(&hslot2->lock); 304 } 305 error = 0; 306 fail_unlock: 307 spin_unlock_bh(&hslot->lock); 308 fail: 309 return error; 310 } 311 EXPORT_SYMBOL(udp_lib_get_port); 312 313 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 314 { 315 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 316 317 return (!ipv6_only_sock(sk2) && 318 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 319 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 320 } 321 322 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, 323 unsigned int port) 324 { 325 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 326 } 327 328 int udp_v4_get_port(struct sock *sk, unsigned short snum) 329 { 330 unsigned int hash2_nulladdr = 331 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 332 unsigned int hash2_partial = 333 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 334 335 /* precompute partial secondary hash */ 336 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 337 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 338 } 339 340 static inline int compute_score(struct sock *sk, struct net *net, 341 __be32 saddr, unsigned short hnum, __be16 sport, 342 __be32 daddr, __be16 dport, int dif) 343 { 344 int score; 345 struct inet_sock *inet; 346 347 if (!net_eq(sock_net(sk), net) || 348 udp_sk(sk)->udp_port_hash != hnum || 349 ipv6_only_sock(sk)) 350 return -1; 351 352 score = (sk->sk_family == PF_INET) ? 2 : 1; 353 inet = inet_sk(sk); 354 355 if (inet->inet_rcv_saddr) { 356 if (inet->inet_rcv_saddr != daddr) 357 return -1; 358 score += 4; 359 } 360 361 if (inet->inet_daddr) { 362 if (inet->inet_daddr != saddr) 363 return -1; 364 score += 4; 365 } 366 367 if (inet->inet_dport) { 368 if (inet->inet_dport != sport) 369 return -1; 370 score += 4; 371 } 372 373 if (sk->sk_bound_dev_if) { 374 if (sk->sk_bound_dev_if != dif) 375 return -1; 376 score += 4; 377 } 378 379 return score; 380 } 381 382 /* 383 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 384 */ 385 static inline int compute_score2(struct sock *sk, struct net *net, 386 __be32 saddr, __be16 sport, 387 __be32 daddr, unsigned int hnum, int dif) 388 { 389 int score; 390 struct inet_sock *inet; 391 392 if (!net_eq(sock_net(sk), net) || 393 ipv6_only_sock(sk)) 394 return -1; 395 396 inet = inet_sk(sk); 397 398 if (inet->inet_rcv_saddr != daddr || 399 inet->inet_num != hnum) 400 return -1; 401 402 score = (sk->sk_family == PF_INET) ? 2 : 1; 403 404 if (inet->inet_daddr) { 405 if (inet->inet_daddr != saddr) 406 return -1; 407 score += 4; 408 } 409 410 if (inet->inet_dport) { 411 if (inet->inet_dport != sport) 412 return -1; 413 score += 4; 414 } 415 416 if (sk->sk_bound_dev_if) { 417 if (sk->sk_bound_dev_if != dif) 418 return -1; 419 score += 4; 420 } 421 422 return score; 423 } 424 425 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 426 const __u16 lport, const __be32 faddr, 427 const __be16 fport) 428 { 429 static u32 udp_ehash_secret __read_mostly; 430 431 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 432 433 return __inet_ehashfn(laddr, lport, faddr, fport, 434 udp_ehash_secret + net_hash_mix(net)); 435 } 436 437 /* called with read_rcu_lock() */ 438 static struct sock *udp4_lib_lookup2(struct net *net, 439 __be32 saddr, __be16 sport, 440 __be32 daddr, unsigned int hnum, int dif, 441 struct udp_hslot *hslot2, unsigned int slot2) 442 { 443 struct sock *sk, *result; 444 struct hlist_nulls_node *node; 445 int score, badness, matches = 0, reuseport = 0; 446 u32 hash = 0; 447 448 begin: 449 result = NULL; 450 badness = 0; 451 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 452 score = compute_score2(sk, net, saddr, sport, 453 daddr, hnum, dif); 454 if (score > badness) { 455 result = sk; 456 badness = score; 457 reuseport = sk->sk_reuseport; 458 if (reuseport) { 459 hash = udp_ehashfn(net, daddr, hnum, 460 saddr, sport); 461 matches = 1; 462 } 463 } else if (score == badness && reuseport) { 464 matches++; 465 if (reciprocal_scale(hash, matches) == 0) 466 result = sk; 467 hash = next_pseudo_random32(hash); 468 } 469 } 470 /* 471 * if the nulls value we got at the end of this lookup is 472 * not the expected one, we must restart lookup. 473 * We probably met an item that was moved to another chain. 474 */ 475 if (get_nulls_value(node) != slot2) 476 goto begin; 477 if (result) { 478 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 479 result = NULL; 480 else if (unlikely(compute_score2(result, net, saddr, sport, 481 daddr, hnum, dif) < badness)) { 482 sock_put(result); 483 goto begin; 484 } 485 } 486 return result; 487 } 488 489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 490 * harder than this. -DaveM 491 */ 492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 493 __be16 sport, __be32 daddr, __be16 dport, 494 int dif, struct udp_table *udptable) 495 { 496 struct sock *sk, *result; 497 struct hlist_nulls_node *node; 498 unsigned short hnum = ntohs(dport); 499 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 500 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 501 int score, badness, matches = 0, reuseport = 0; 502 u32 hash = 0; 503 504 rcu_read_lock(); 505 if (hslot->count > 10) { 506 hash2 = udp4_portaddr_hash(net, daddr, hnum); 507 slot2 = hash2 & udptable->mask; 508 hslot2 = &udptable->hash2[slot2]; 509 if (hslot->count < hslot2->count) 510 goto begin; 511 512 result = udp4_lib_lookup2(net, saddr, sport, 513 daddr, hnum, dif, 514 hslot2, slot2); 515 if (!result) { 516 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 517 slot2 = hash2 & udptable->mask; 518 hslot2 = &udptable->hash2[slot2]; 519 if (hslot->count < hslot2->count) 520 goto begin; 521 522 result = udp4_lib_lookup2(net, saddr, sport, 523 htonl(INADDR_ANY), hnum, dif, 524 hslot2, slot2); 525 } 526 rcu_read_unlock(); 527 return result; 528 } 529 begin: 530 result = NULL; 531 badness = 0; 532 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 533 score = compute_score(sk, net, saddr, hnum, sport, 534 daddr, dport, dif); 535 if (score > badness) { 536 result = sk; 537 badness = score; 538 reuseport = sk->sk_reuseport; 539 if (reuseport) { 540 hash = udp_ehashfn(net, daddr, hnum, 541 saddr, sport); 542 matches = 1; 543 } 544 } else if (score == badness && reuseport) { 545 matches++; 546 if (reciprocal_scale(hash, matches) == 0) 547 result = sk; 548 hash = next_pseudo_random32(hash); 549 } 550 } 551 /* 552 * if the nulls value we got at the end of this lookup is 553 * not the expected one, we must restart lookup. 554 * We probably met an item that was moved to another chain. 555 */ 556 if (get_nulls_value(node) != slot) 557 goto begin; 558 559 if (result) { 560 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 561 result = NULL; 562 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 563 daddr, dport, dif) < badness)) { 564 sock_put(result); 565 goto begin; 566 } 567 } 568 rcu_read_unlock(); 569 return result; 570 } 571 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 572 573 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 574 __be16 sport, __be16 dport, 575 struct udp_table *udptable) 576 { 577 const struct iphdr *iph = ip_hdr(skb); 578 579 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 580 iph->daddr, dport, inet_iif(skb), 581 udptable); 582 } 583 584 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 585 __be32 daddr, __be16 dport, int dif) 586 { 587 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 588 } 589 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 590 591 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 592 __be16 loc_port, __be32 loc_addr, 593 __be16 rmt_port, __be32 rmt_addr, 594 int dif, unsigned short hnum) 595 { 596 struct inet_sock *inet = inet_sk(sk); 597 598 if (!net_eq(sock_net(sk), net) || 599 udp_sk(sk)->udp_port_hash != hnum || 600 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 601 (inet->inet_dport != rmt_port && inet->inet_dport) || 602 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 603 ipv6_only_sock(sk) || 604 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 605 return false; 606 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 607 return false; 608 return true; 609 } 610 611 /* 612 * This routine is called by the ICMP module when it gets some 613 * sort of error condition. If err < 0 then the socket should 614 * be closed and the error returned to the user. If err > 0 615 * it's just the icmp type << 8 | icmp code. 616 * Header points to the ip header of the error packet. We move 617 * on past this. Then (as it used to claim before adjustment) 618 * header points to the first 8 bytes of the udp header. We need 619 * to find the appropriate port. 620 */ 621 622 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 623 { 624 struct inet_sock *inet; 625 const struct iphdr *iph = (const struct iphdr *)skb->data; 626 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 627 const int type = icmp_hdr(skb)->type; 628 const int code = icmp_hdr(skb)->code; 629 struct sock *sk; 630 int harderr; 631 int err; 632 struct net *net = dev_net(skb->dev); 633 634 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 635 iph->saddr, uh->source, skb->dev->ifindex, udptable); 636 if (!sk) { 637 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 638 return; /* No socket for error */ 639 } 640 641 err = 0; 642 harderr = 0; 643 inet = inet_sk(sk); 644 645 switch (type) { 646 default: 647 case ICMP_TIME_EXCEEDED: 648 err = EHOSTUNREACH; 649 break; 650 case ICMP_SOURCE_QUENCH: 651 goto out; 652 case ICMP_PARAMETERPROB: 653 err = EPROTO; 654 harderr = 1; 655 break; 656 case ICMP_DEST_UNREACH: 657 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 658 ipv4_sk_update_pmtu(skb, sk, info); 659 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 660 err = EMSGSIZE; 661 harderr = 1; 662 break; 663 } 664 goto out; 665 } 666 err = EHOSTUNREACH; 667 if (code <= NR_ICMP_UNREACH) { 668 harderr = icmp_err_convert[code].fatal; 669 err = icmp_err_convert[code].errno; 670 } 671 break; 672 case ICMP_REDIRECT: 673 ipv4_sk_redirect(skb, sk); 674 goto out; 675 } 676 677 /* 678 * RFC1122: OK. Passes ICMP errors back to application, as per 679 * 4.1.3.3. 680 */ 681 if (!inet->recverr) { 682 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 683 goto out; 684 } else 685 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 686 687 sk->sk_err = err; 688 sk->sk_error_report(sk); 689 out: 690 sock_put(sk); 691 } 692 693 void udp_err(struct sk_buff *skb, u32 info) 694 { 695 __udp4_lib_err(skb, info, &udp_table); 696 } 697 698 /* 699 * Throw away all pending data and cancel the corking. Socket is locked. 700 */ 701 void udp_flush_pending_frames(struct sock *sk) 702 { 703 struct udp_sock *up = udp_sk(sk); 704 705 if (up->pending) { 706 up->len = 0; 707 up->pending = 0; 708 ip_flush_pending_frames(sk); 709 } 710 } 711 EXPORT_SYMBOL(udp_flush_pending_frames); 712 713 /** 714 * udp4_hwcsum - handle outgoing HW checksumming 715 * @skb: sk_buff containing the filled-in UDP header 716 * (checksum field must be zeroed out) 717 * @src: source IP address 718 * @dst: destination IP address 719 */ 720 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 721 { 722 struct udphdr *uh = udp_hdr(skb); 723 int offset = skb_transport_offset(skb); 724 int len = skb->len - offset; 725 int hlen = len; 726 __wsum csum = 0; 727 728 if (!skb_has_frag_list(skb)) { 729 /* 730 * Only one fragment on the socket. 731 */ 732 skb->csum_start = skb_transport_header(skb) - skb->head; 733 skb->csum_offset = offsetof(struct udphdr, check); 734 uh->check = ~csum_tcpudp_magic(src, dst, len, 735 IPPROTO_UDP, 0); 736 } else { 737 struct sk_buff *frags; 738 739 /* 740 * HW-checksum won't work as there are two or more 741 * fragments on the socket so that all csums of sk_buffs 742 * should be together 743 */ 744 skb_walk_frags(skb, frags) { 745 csum = csum_add(csum, frags->csum); 746 hlen -= frags->len; 747 } 748 749 csum = skb_checksum(skb, offset, hlen, csum); 750 skb->ip_summed = CHECKSUM_NONE; 751 752 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 753 if (uh->check == 0) 754 uh->check = CSUM_MANGLED_0; 755 } 756 } 757 EXPORT_SYMBOL_GPL(udp4_hwcsum); 758 759 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 760 * for the simple case like when setting the checksum for a UDP tunnel. 761 */ 762 void udp_set_csum(bool nocheck, struct sk_buff *skb, 763 __be32 saddr, __be32 daddr, int len) 764 { 765 struct udphdr *uh = udp_hdr(skb); 766 767 if (nocheck) 768 uh->check = 0; 769 else if (skb_is_gso(skb)) 770 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 771 else if (skb_dst(skb) && skb_dst(skb)->dev && 772 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) { 773 774 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 775 776 skb->ip_summed = CHECKSUM_PARTIAL; 777 skb->csum_start = skb_transport_header(skb) - skb->head; 778 skb->csum_offset = offsetof(struct udphdr, check); 779 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 780 } else { 781 __wsum csum; 782 783 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 784 785 uh->check = 0; 786 csum = skb_checksum(skb, 0, len, 0); 787 uh->check = udp_v4_check(len, saddr, daddr, csum); 788 if (uh->check == 0) 789 uh->check = CSUM_MANGLED_0; 790 791 skb->ip_summed = CHECKSUM_UNNECESSARY; 792 } 793 } 794 EXPORT_SYMBOL(udp_set_csum); 795 796 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 797 { 798 struct sock *sk = skb->sk; 799 struct inet_sock *inet = inet_sk(sk); 800 struct udphdr *uh; 801 int err = 0; 802 int is_udplite = IS_UDPLITE(sk); 803 int offset = skb_transport_offset(skb); 804 int len = skb->len - offset; 805 __wsum csum = 0; 806 807 /* 808 * Create a UDP header 809 */ 810 uh = udp_hdr(skb); 811 uh->source = inet->inet_sport; 812 uh->dest = fl4->fl4_dport; 813 uh->len = htons(len); 814 uh->check = 0; 815 816 if (is_udplite) /* UDP-Lite */ 817 csum = udplite_csum(skb); 818 819 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 820 821 skb->ip_summed = CHECKSUM_NONE; 822 goto send; 823 824 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 825 826 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 827 goto send; 828 829 } else 830 csum = udp_csum(skb); 831 832 /* add protocol-dependent pseudo-header */ 833 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 834 sk->sk_protocol, csum); 835 if (uh->check == 0) 836 uh->check = CSUM_MANGLED_0; 837 838 send: 839 err = ip_send_skb(sock_net(sk), skb); 840 if (err) { 841 if (err == -ENOBUFS && !inet->recverr) { 842 UDP_INC_STATS_USER(sock_net(sk), 843 UDP_MIB_SNDBUFERRORS, is_udplite); 844 err = 0; 845 } 846 } else 847 UDP_INC_STATS_USER(sock_net(sk), 848 UDP_MIB_OUTDATAGRAMS, is_udplite); 849 return err; 850 } 851 852 /* 853 * Push out all pending data as one UDP datagram. Socket is locked. 854 */ 855 int udp_push_pending_frames(struct sock *sk) 856 { 857 struct udp_sock *up = udp_sk(sk); 858 struct inet_sock *inet = inet_sk(sk); 859 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 860 struct sk_buff *skb; 861 int err = 0; 862 863 skb = ip_finish_skb(sk, fl4); 864 if (!skb) 865 goto out; 866 867 err = udp_send_skb(skb, fl4); 868 869 out: 870 up->len = 0; 871 up->pending = 0; 872 return err; 873 } 874 EXPORT_SYMBOL(udp_push_pending_frames); 875 876 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 877 { 878 struct inet_sock *inet = inet_sk(sk); 879 struct udp_sock *up = udp_sk(sk); 880 struct flowi4 fl4_stack; 881 struct flowi4 *fl4; 882 int ulen = len; 883 struct ipcm_cookie ipc; 884 struct rtable *rt = NULL; 885 int free = 0; 886 int connected = 0; 887 __be32 daddr, faddr, saddr; 888 __be16 dport; 889 u8 tos; 890 int err, is_udplite = IS_UDPLITE(sk); 891 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 892 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 893 struct sk_buff *skb; 894 struct ip_options_data opt_copy; 895 896 if (len > 0xFFFF) 897 return -EMSGSIZE; 898 899 /* 900 * Check the flags. 901 */ 902 903 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 904 return -EOPNOTSUPP; 905 906 ipc.opt = NULL; 907 ipc.tx_flags = 0; 908 ipc.ttl = 0; 909 ipc.tos = -1; 910 911 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 912 913 fl4 = &inet->cork.fl.u.ip4; 914 if (up->pending) { 915 /* 916 * There are pending frames. 917 * The socket lock must be held while it's corked. 918 */ 919 lock_sock(sk); 920 if (likely(up->pending)) { 921 if (unlikely(up->pending != AF_INET)) { 922 release_sock(sk); 923 return -EINVAL; 924 } 925 goto do_append_data; 926 } 927 release_sock(sk); 928 } 929 ulen += sizeof(struct udphdr); 930 931 /* 932 * Get and verify the address. 933 */ 934 if (msg->msg_name) { 935 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 936 if (msg->msg_namelen < sizeof(*usin)) 937 return -EINVAL; 938 if (usin->sin_family != AF_INET) { 939 if (usin->sin_family != AF_UNSPEC) 940 return -EAFNOSUPPORT; 941 } 942 943 daddr = usin->sin_addr.s_addr; 944 dport = usin->sin_port; 945 if (dport == 0) 946 return -EINVAL; 947 } else { 948 if (sk->sk_state != TCP_ESTABLISHED) 949 return -EDESTADDRREQ; 950 daddr = inet->inet_daddr; 951 dport = inet->inet_dport; 952 /* Open fast path for connected socket. 953 Route will not be used, if at least one option is set. 954 */ 955 connected = 1; 956 } 957 ipc.addr = inet->inet_saddr; 958 959 ipc.oif = sk->sk_bound_dev_if; 960 961 sock_tx_timestamp(sk, &ipc.tx_flags); 962 963 if (msg->msg_controllen) { 964 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 965 sk->sk_family == AF_INET6); 966 if (err) 967 return err; 968 if (ipc.opt) 969 free = 1; 970 connected = 0; 971 } 972 if (!ipc.opt) { 973 struct ip_options_rcu *inet_opt; 974 975 rcu_read_lock(); 976 inet_opt = rcu_dereference(inet->inet_opt); 977 if (inet_opt) { 978 memcpy(&opt_copy, inet_opt, 979 sizeof(*inet_opt) + inet_opt->opt.optlen); 980 ipc.opt = &opt_copy.opt; 981 } 982 rcu_read_unlock(); 983 } 984 985 saddr = ipc.addr; 986 ipc.addr = faddr = daddr; 987 988 if (ipc.opt && ipc.opt->opt.srr) { 989 if (!daddr) 990 return -EINVAL; 991 faddr = ipc.opt->opt.faddr; 992 connected = 0; 993 } 994 tos = get_rttos(&ipc, inet); 995 if (sock_flag(sk, SOCK_LOCALROUTE) || 996 (msg->msg_flags & MSG_DONTROUTE) || 997 (ipc.opt && ipc.opt->opt.is_strictroute)) { 998 tos |= RTO_ONLINK; 999 connected = 0; 1000 } 1001 1002 if (ipv4_is_multicast(daddr)) { 1003 if (!ipc.oif) 1004 ipc.oif = inet->mc_index; 1005 if (!saddr) 1006 saddr = inet->mc_addr; 1007 connected = 0; 1008 } else if (!ipc.oif) 1009 ipc.oif = inet->uc_index; 1010 1011 if (connected) 1012 rt = (struct rtable *)sk_dst_check(sk, 0); 1013 1014 if (!rt) { 1015 struct net *net = sock_net(sk); 1016 1017 fl4 = &fl4_stack; 1018 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1019 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1020 inet_sk_flowi_flags(sk), 1021 faddr, saddr, dport, inet->inet_sport); 1022 1023 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1024 rt = ip_route_output_flow(net, fl4, sk); 1025 if (IS_ERR(rt)) { 1026 err = PTR_ERR(rt); 1027 rt = NULL; 1028 if (err == -ENETUNREACH) 1029 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1030 goto out; 1031 } 1032 1033 err = -EACCES; 1034 if ((rt->rt_flags & RTCF_BROADCAST) && 1035 !sock_flag(sk, SOCK_BROADCAST)) 1036 goto out; 1037 if (connected) 1038 sk_dst_set(sk, dst_clone(&rt->dst)); 1039 } 1040 1041 if (msg->msg_flags&MSG_CONFIRM) 1042 goto do_confirm; 1043 back_from_confirm: 1044 1045 saddr = fl4->saddr; 1046 if (!ipc.addr) 1047 daddr = ipc.addr = fl4->daddr; 1048 1049 /* Lockless fast path for the non-corking case. */ 1050 if (!corkreq) { 1051 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1052 sizeof(struct udphdr), &ipc, &rt, 1053 msg->msg_flags); 1054 err = PTR_ERR(skb); 1055 if (!IS_ERR_OR_NULL(skb)) 1056 err = udp_send_skb(skb, fl4); 1057 goto out; 1058 } 1059 1060 lock_sock(sk); 1061 if (unlikely(up->pending)) { 1062 /* The socket is already corked while preparing it. */ 1063 /* ... which is an evident application bug. --ANK */ 1064 release_sock(sk); 1065 1066 net_dbg_ratelimited("cork app bug 2\n"); 1067 err = -EINVAL; 1068 goto out; 1069 } 1070 /* 1071 * Now cork the socket to pend data. 1072 */ 1073 fl4 = &inet->cork.fl.u.ip4; 1074 fl4->daddr = daddr; 1075 fl4->saddr = saddr; 1076 fl4->fl4_dport = dport; 1077 fl4->fl4_sport = inet->inet_sport; 1078 up->pending = AF_INET; 1079 1080 do_append_data: 1081 up->len += ulen; 1082 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1083 sizeof(struct udphdr), &ipc, &rt, 1084 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1085 if (err) 1086 udp_flush_pending_frames(sk); 1087 else if (!corkreq) 1088 err = udp_push_pending_frames(sk); 1089 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1090 up->pending = 0; 1091 release_sock(sk); 1092 1093 out: 1094 ip_rt_put(rt); 1095 if (free) 1096 kfree(ipc.opt); 1097 if (!err) 1098 return len; 1099 /* 1100 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1101 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1102 * we don't have a good statistic (IpOutDiscards but it can be too many 1103 * things). We could add another new stat but at least for now that 1104 * seems like overkill. 1105 */ 1106 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1107 UDP_INC_STATS_USER(sock_net(sk), 1108 UDP_MIB_SNDBUFERRORS, is_udplite); 1109 } 1110 return err; 1111 1112 do_confirm: 1113 dst_confirm(&rt->dst); 1114 if (!(msg->msg_flags&MSG_PROBE) || len) 1115 goto back_from_confirm; 1116 err = 0; 1117 goto out; 1118 } 1119 EXPORT_SYMBOL(udp_sendmsg); 1120 1121 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1122 size_t size, int flags) 1123 { 1124 struct inet_sock *inet = inet_sk(sk); 1125 struct udp_sock *up = udp_sk(sk); 1126 int ret; 1127 1128 if (flags & MSG_SENDPAGE_NOTLAST) 1129 flags |= MSG_MORE; 1130 1131 if (!up->pending) { 1132 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1133 1134 /* Call udp_sendmsg to specify destination address which 1135 * sendpage interface can't pass. 1136 * This will succeed only when the socket is connected. 1137 */ 1138 ret = udp_sendmsg(sk, &msg, 0); 1139 if (ret < 0) 1140 return ret; 1141 } 1142 1143 lock_sock(sk); 1144 1145 if (unlikely(!up->pending)) { 1146 release_sock(sk); 1147 1148 net_dbg_ratelimited("udp cork app bug 3\n"); 1149 return -EINVAL; 1150 } 1151 1152 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1153 page, offset, size, flags); 1154 if (ret == -EOPNOTSUPP) { 1155 release_sock(sk); 1156 return sock_no_sendpage(sk->sk_socket, page, offset, 1157 size, flags); 1158 } 1159 if (ret < 0) { 1160 udp_flush_pending_frames(sk); 1161 goto out; 1162 } 1163 1164 up->len += size; 1165 if (!(up->corkflag || (flags&MSG_MORE))) 1166 ret = udp_push_pending_frames(sk); 1167 if (!ret) 1168 ret = size; 1169 out: 1170 release_sock(sk); 1171 return ret; 1172 } 1173 1174 /** 1175 * first_packet_length - return length of first packet in receive queue 1176 * @sk: socket 1177 * 1178 * Drops all bad checksum frames, until a valid one is found. 1179 * Returns the length of found skb, or 0 if none is found. 1180 */ 1181 static unsigned int first_packet_length(struct sock *sk) 1182 { 1183 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1184 struct sk_buff *skb; 1185 unsigned int res; 1186 1187 __skb_queue_head_init(&list_kill); 1188 1189 spin_lock_bh(&rcvq->lock); 1190 while ((skb = skb_peek(rcvq)) != NULL && 1191 udp_lib_checksum_complete(skb)) { 1192 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1193 IS_UDPLITE(sk)); 1194 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1195 IS_UDPLITE(sk)); 1196 atomic_inc(&sk->sk_drops); 1197 __skb_unlink(skb, rcvq); 1198 __skb_queue_tail(&list_kill, skb); 1199 } 1200 res = skb ? skb->len : 0; 1201 spin_unlock_bh(&rcvq->lock); 1202 1203 if (!skb_queue_empty(&list_kill)) { 1204 bool slow = lock_sock_fast(sk); 1205 1206 __skb_queue_purge(&list_kill); 1207 sk_mem_reclaim_partial(sk); 1208 unlock_sock_fast(sk, slow); 1209 } 1210 return res; 1211 } 1212 1213 /* 1214 * IOCTL requests applicable to the UDP protocol 1215 */ 1216 1217 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1218 { 1219 switch (cmd) { 1220 case SIOCOUTQ: 1221 { 1222 int amount = sk_wmem_alloc_get(sk); 1223 1224 return put_user(amount, (int __user *)arg); 1225 } 1226 1227 case SIOCINQ: 1228 { 1229 unsigned int amount = first_packet_length(sk); 1230 1231 if (amount) 1232 /* 1233 * We will only return the amount 1234 * of this packet since that is all 1235 * that will be read. 1236 */ 1237 amount -= sizeof(struct udphdr); 1238 1239 return put_user(amount, (int __user *)arg); 1240 } 1241 1242 default: 1243 return -ENOIOCTLCMD; 1244 } 1245 1246 return 0; 1247 } 1248 EXPORT_SYMBOL(udp_ioctl); 1249 1250 /* 1251 * This should be easy, if there is something there we 1252 * return it, otherwise we block. 1253 */ 1254 1255 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1256 int flags, int *addr_len) 1257 { 1258 struct inet_sock *inet = inet_sk(sk); 1259 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1260 struct sk_buff *skb; 1261 unsigned int ulen, copied; 1262 int peeked, off = 0; 1263 int err; 1264 int is_udplite = IS_UDPLITE(sk); 1265 bool slow; 1266 1267 if (flags & MSG_ERRQUEUE) 1268 return ip_recv_error(sk, msg, len, addr_len); 1269 1270 try_again: 1271 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1272 &peeked, &off, &err); 1273 if (!skb) 1274 goto out; 1275 1276 ulen = skb->len - sizeof(struct udphdr); 1277 copied = len; 1278 if (copied > ulen) 1279 copied = ulen; 1280 else if (copied < ulen) 1281 msg->msg_flags |= MSG_TRUNC; 1282 1283 /* 1284 * If checksum is needed at all, try to do it while copying the 1285 * data. If the data is truncated, or if we only want a partial 1286 * coverage checksum (UDP-Lite), do it before the copy. 1287 */ 1288 1289 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1290 if (udp_lib_checksum_complete(skb)) 1291 goto csum_copy_err; 1292 } 1293 1294 if (skb_csum_unnecessary(skb)) 1295 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 1296 msg, copied); 1297 else { 1298 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), 1299 msg); 1300 1301 if (err == -EINVAL) 1302 goto csum_copy_err; 1303 } 1304 1305 if (unlikely(err)) { 1306 trace_kfree_skb(skb, udp_recvmsg); 1307 if (!peeked) { 1308 atomic_inc(&sk->sk_drops); 1309 UDP_INC_STATS_USER(sock_net(sk), 1310 UDP_MIB_INERRORS, is_udplite); 1311 } 1312 goto out_free; 1313 } 1314 1315 if (!peeked) 1316 UDP_INC_STATS_USER(sock_net(sk), 1317 UDP_MIB_INDATAGRAMS, is_udplite); 1318 1319 sock_recv_ts_and_drops(msg, sk, skb); 1320 1321 /* Copy the address. */ 1322 if (sin) { 1323 sin->sin_family = AF_INET; 1324 sin->sin_port = udp_hdr(skb)->source; 1325 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1326 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1327 *addr_len = sizeof(*sin); 1328 } 1329 if (inet->cmsg_flags) 1330 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr)); 1331 1332 err = copied; 1333 if (flags & MSG_TRUNC) 1334 err = ulen; 1335 1336 out_free: 1337 skb_free_datagram_locked(sk, skb); 1338 out: 1339 return err; 1340 1341 csum_copy_err: 1342 slow = lock_sock_fast(sk); 1343 if (!skb_kill_datagram(sk, skb, flags)) { 1344 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1345 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1346 } 1347 unlock_sock_fast(sk, slow); 1348 1349 /* starting over for a new packet, but check if we need to yield */ 1350 cond_resched(); 1351 msg->msg_flags &= ~MSG_TRUNC; 1352 goto try_again; 1353 } 1354 1355 int udp_disconnect(struct sock *sk, int flags) 1356 { 1357 struct inet_sock *inet = inet_sk(sk); 1358 /* 1359 * 1003.1g - break association. 1360 */ 1361 1362 sk->sk_state = TCP_CLOSE; 1363 inet->inet_daddr = 0; 1364 inet->inet_dport = 0; 1365 sock_rps_reset_rxhash(sk); 1366 sk->sk_bound_dev_if = 0; 1367 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1368 inet_reset_saddr(sk); 1369 1370 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1371 sk->sk_prot->unhash(sk); 1372 inet->inet_sport = 0; 1373 } 1374 sk_dst_reset(sk); 1375 return 0; 1376 } 1377 EXPORT_SYMBOL(udp_disconnect); 1378 1379 void udp_lib_unhash(struct sock *sk) 1380 { 1381 if (sk_hashed(sk)) { 1382 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1383 struct udp_hslot *hslot, *hslot2; 1384 1385 hslot = udp_hashslot(udptable, sock_net(sk), 1386 udp_sk(sk)->udp_port_hash); 1387 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1388 1389 spin_lock_bh(&hslot->lock); 1390 if (sk_nulls_del_node_init_rcu(sk)) { 1391 hslot->count--; 1392 inet_sk(sk)->inet_num = 0; 1393 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1394 1395 spin_lock(&hslot2->lock); 1396 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1397 hslot2->count--; 1398 spin_unlock(&hslot2->lock); 1399 } 1400 spin_unlock_bh(&hslot->lock); 1401 } 1402 } 1403 EXPORT_SYMBOL(udp_lib_unhash); 1404 1405 /* 1406 * inet_rcv_saddr was changed, we must rehash secondary hash 1407 */ 1408 void udp_lib_rehash(struct sock *sk, u16 newhash) 1409 { 1410 if (sk_hashed(sk)) { 1411 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1412 struct udp_hslot *hslot, *hslot2, *nhslot2; 1413 1414 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1415 nhslot2 = udp_hashslot2(udptable, newhash); 1416 udp_sk(sk)->udp_portaddr_hash = newhash; 1417 if (hslot2 != nhslot2) { 1418 hslot = udp_hashslot(udptable, sock_net(sk), 1419 udp_sk(sk)->udp_port_hash); 1420 /* we must lock primary chain too */ 1421 spin_lock_bh(&hslot->lock); 1422 1423 spin_lock(&hslot2->lock); 1424 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1425 hslot2->count--; 1426 spin_unlock(&hslot2->lock); 1427 1428 spin_lock(&nhslot2->lock); 1429 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1430 &nhslot2->head); 1431 nhslot2->count++; 1432 spin_unlock(&nhslot2->lock); 1433 1434 spin_unlock_bh(&hslot->lock); 1435 } 1436 } 1437 } 1438 EXPORT_SYMBOL(udp_lib_rehash); 1439 1440 static void udp_v4_rehash(struct sock *sk) 1441 { 1442 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1443 inet_sk(sk)->inet_rcv_saddr, 1444 inet_sk(sk)->inet_num); 1445 udp_lib_rehash(sk, new_hash); 1446 } 1447 1448 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1449 { 1450 int rc; 1451 1452 if (inet_sk(sk)->inet_daddr) { 1453 sock_rps_save_rxhash(sk, skb); 1454 sk_mark_napi_id(sk, skb); 1455 sk_incoming_cpu_update(sk); 1456 } 1457 1458 rc = sock_queue_rcv_skb(sk, skb); 1459 if (rc < 0) { 1460 int is_udplite = IS_UDPLITE(sk); 1461 1462 /* Note that an ENOMEM error is charged twice */ 1463 if (rc == -ENOMEM) 1464 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1465 is_udplite); 1466 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1467 kfree_skb(skb); 1468 trace_udp_fail_queue_rcv_skb(rc, sk); 1469 return -1; 1470 } 1471 1472 return 0; 1473 1474 } 1475 1476 static struct static_key udp_encap_needed __read_mostly; 1477 void udp_encap_enable(void) 1478 { 1479 if (!static_key_enabled(&udp_encap_needed)) 1480 static_key_slow_inc(&udp_encap_needed); 1481 } 1482 EXPORT_SYMBOL(udp_encap_enable); 1483 1484 /* returns: 1485 * -1: error 1486 * 0: success 1487 * >0: "udp encap" protocol resubmission 1488 * 1489 * Note that in the success and error cases, the skb is assumed to 1490 * have either been requeued or freed. 1491 */ 1492 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1493 { 1494 struct udp_sock *up = udp_sk(sk); 1495 int rc; 1496 int is_udplite = IS_UDPLITE(sk); 1497 1498 /* 1499 * Charge it to the socket, dropping if the queue is full. 1500 */ 1501 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1502 goto drop; 1503 nf_reset(skb); 1504 1505 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1506 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1507 1508 /* 1509 * This is an encapsulation socket so pass the skb to 1510 * the socket's udp_encap_rcv() hook. Otherwise, just 1511 * fall through and pass this up the UDP socket. 1512 * up->encap_rcv() returns the following value: 1513 * =0 if skb was successfully passed to the encap 1514 * handler or was discarded by it. 1515 * >0 if skb should be passed on to UDP. 1516 * <0 if skb should be resubmitted as proto -N 1517 */ 1518 1519 /* if we're overly short, let UDP handle it */ 1520 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1521 if (skb->len > sizeof(struct udphdr) && encap_rcv) { 1522 int ret; 1523 1524 /* Verify checksum before giving to encap */ 1525 if (udp_lib_checksum_complete(skb)) 1526 goto csum_error; 1527 1528 ret = encap_rcv(sk, skb); 1529 if (ret <= 0) { 1530 UDP_INC_STATS_BH(sock_net(sk), 1531 UDP_MIB_INDATAGRAMS, 1532 is_udplite); 1533 return -ret; 1534 } 1535 } 1536 1537 /* FALLTHROUGH -- it's a UDP Packet */ 1538 } 1539 1540 /* 1541 * UDP-Lite specific tests, ignored on UDP sockets 1542 */ 1543 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1544 1545 /* 1546 * MIB statistics other than incrementing the error count are 1547 * disabled for the following two types of errors: these depend 1548 * on the application settings, not on the functioning of the 1549 * protocol stack as such. 1550 * 1551 * RFC 3828 here recommends (sec 3.3): "There should also be a 1552 * way ... to ... at least let the receiving application block 1553 * delivery of packets with coverage values less than a value 1554 * provided by the application." 1555 */ 1556 if (up->pcrlen == 0) { /* full coverage was set */ 1557 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1558 UDP_SKB_CB(skb)->cscov, skb->len); 1559 goto drop; 1560 } 1561 /* The next case involves violating the min. coverage requested 1562 * by the receiver. This is subtle: if receiver wants x and x is 1563 * greater than the buffersize/MTU then receiver will complain 1564 * that it wants x while sender emits packets of smaller size y. 1565 * Therefore the above ...()->partial_cov statement is essential. 1566 */ 1567 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1568 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1569 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1570 goto drop; 1571 } 1572 } 1573 1574 if (rcu_access_pointer(sk->sk_filter) && 1575 udp_lib_checksum_complete(skb)) 1576 goto csum_error; 1577 1578 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1579 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1580 is_udplite); 1581 goto drop; 1582 } 1583 1584 rc = 0; 1585 1586 ipv4_pktinfo_prepare(sk, skb); 1587 bh_lock_sock(sk); 1588 if (!sock_owned_by_user(sk)) 1589 rc = __udp_queue_rcv_skb(sk, skb); 1590 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1591 bh_unlock_sock(sk); 1592 goto drop; 1593 } 1594 bh_unlock_sock(sk); 1595 1596 return rc; 1597 1598 csum_error: 1599 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1600 drop: 1601 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1602 atomic_inc(&sk->sk_drops); 1603 kfree_skb(skb); 1604 return -1; 1605 } 1606 1607 static void flush_stack(struct sock **stack, unsigned int count, 1608 struct sk_buff *skb, unsigned int final) 1609 { 1610 unsigned int i; 1611 struct sk_buff *skb1 = NULL; 1612 struct sock *sk; 1613 1614 for (i = 0; i < count; i++) { 1615 sk = stack[i]; 1616 if (likely(!skb1)) 1617 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1618 1619 if (!skb1) { 1620 atomic_inc(&sk->sk_drops); 1621 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1622 IS_UDPLITE(sk)); 1623 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1624 IS_UDPLITE(sk)); 1625 } 1626 1627 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1628 skb1 = NULL; 1629 1630 sock_put(sk); 1631 } 1632 if (unlikely(skb1)) 1633 kfree_skb(skb1); 1634 } 1635 1636 /* For TCP sockets, sk_rx_dst is protected by socket lock 1637 * For UDP, we use xchg() to guard against concurrent changes. 1638 */ 1639 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1640 { 1641 struct dst_entry *old; 1642 1643 dst_hold(dst); 1644 old = xchg(&sk->sk_rx_dst, dst); 1645 dst_release(old); 1646 } 1647 1648 /* 1649 * Multicasts and broadcasts go to each listener. 1650 * 1651 * Note: called only from the BH handler context. 1652 */ 1653 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1654 struct udphdr *uh, 1655 __be32 saddr, __be32 daddr, 1656 struct udp_table *udptable, 1657 int proto) 1658 { 1659 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1660 struct hlist_nulls_node *node; 1661 unsigned short hnum = ntohs(uh->dest); 1662 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1663 int dif = skb->dev->ifindex; 1664 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 1665 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1666 bool inner_flushed = false; 1667 1668 if (use_hash2) { 1669 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1670 udp_table.mask; 1671 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask; 1672 start_lookup: 1673 hslot = &udp_table.hash2[hash2]; 1674 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1675 } 1676 1677 spin_lock(&hslot->lock); 1678 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 1679 if (__udp_is_mcast_sock(net, sk, 1680 uh->dest, daddr, 1681 uh->source, saddr, 1682 dif, hnum)) { 1683 if (unlikely(count == ARRAY_SIZE(stack))) { 1684 flush_stack(stack, count, skb, ~0); 1685 inner_flushed = true; 1686 count = 0; 1687 } 1688 stack[count++] = sk; 1689 sock_hold(sk); 1690 } 1691 } 1692 1693 spin_unlock(&hslot->lock); 1694 1695 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1696 if (use_hash2 && hash2 != hash2_any) { 1697 hash2 = hash2_any; 1698 goto start_lookup; 1699 } 1700 1701 /* 1702 * do the slow work with no lock held 1703 */ 1704 if (count) { 1705 flush_stack(stack, count, skb, count - 1); 1706 } else { 1707 if (!inner_flushed) 1708 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 1709 proto == IPPROTO_UDPLITE); 1710 consume_skb(skb); 1711 } 1712 return 0; 1713 } 1714 1715 /* Initialize UDP checksum. If exited with zero value (success), 1716 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1717 * Otherwise, csum completion requires chacksumming packet body, 1718 * including udp header and folding it to skb->csum. 1719 */ 1720 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1721 int proto) 1722 { 1723 int err; 1724 1725 UDP_SKB_CB(skb)->partial_cov = 0; 1726 UDP_SKB_CB(skb)->cscov = skb->len; 1727 1728 if (proto == IPPROTO_UDPLITE) { 1729 err = udplite_checksum_init(skb, uh); 1730 if (err) 1731 return err; 1732 } 1733 1734 return skb_checksum_init_zero_check(skb, proto, uh->check, 1735 inet_compute_pseudo); 1736 } 1737 1738 /* 1739 * All we need to do is get the socket, and then do a checksum. 1740 */ 1741 1742 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1743 int proto) 1744 { 1745 struct sock *sk; 1746 struct udphdr *uh; 1747 unsigned short ulen; 1748 struct rtable *rt = skb_rtable(skb); 1749 __be32 saddr, daddr; 1750 struct net *net = dev_net(skb->dev); 1751 1752 /* 1753 * Validate the packet. 1754 */ 1755 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1756 goto drop; /* No space for header. */ 1757 1758 uh = udp_hdr(skb); 1759 ulen = ntohs(uh->len); 1760 saddr = ip_hdr(skb)->saddr; 1761 daddr = ip_hdr(skb)->daddr; 1762 1763 if (ulen > skb->len) 1764 goto short_packet; 1765 1766 if (proto == IPPROTO_UDP) { 1767 /* UDP validates ulen. */ 1768 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1769 goto short_packet; 1770 uh = udp_hdr(skb); 1771 } 1772 1773 if (udp4_csum_init(skb, uh, proto)) 1774 goto csum_error; 1775 1776 sk = skb_steal_sock(skb); 1777 if (sk) { 1778 struct dst_entry *dst = skb_dst(skb); 1779 int ret; 1780 1781 if (unlikely(sk->sk_rx_dst != dst)) 1782 udp_sk_rx_dst_set(sk, dst); 1783 1784 ret = udp_queue_rcv_skb(sk, skb); 1785 sock_put(sk); 1786 /* a return value > 0 means to resubmit the input, but 1787 * it wants the return to be -protocol, or 0 1788 */ 1789 if (ret > 0) 1790 return -ret; 1791 return 0; 1792 } 1793 1794 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1795 return __udp4_lib_mcast_deliver(net, skb, uh, 1796 saddr, daddr, udptable, proto); 1797 1798 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1799 if (sk) { 1800 int ret; 1801 1802 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1803 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1804 inet_compute_pseudo); 1805 1806 ret = udp_queue_rcv_skb(sk, skb); 1807 sock_put(sk); 1808 1809 /* a return value > 0 means to resubmit the input, but 1810 * it wants the return to be -protocol, or 0 1811 */ 1812 if (ret > 0) 1813 return -ret; 1814 return 0; 1815 } 1816 1817 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1818 goto drop; 1819 nf_reset(skb); 1820 1821 /* No socket. Drop packet silently, if checksum is wrong */ 1822 if (udp_lib_checksum_complete(skb)) 1823 goto csum_error; 1824 1825 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1826 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1827 1828 /* 1829 * Hmm. We got an UDP packet to a port to which we 1830 * don't wanna listen. Ignore it. 1831 */ 1832 kfree_skb(skb); 1833 return 0; 1834 1835 short_packet: 1836 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1837 proto == IPPROTO_UDPLITE ? "Lite" : "", 1838 &saddr, ntohs(uh->source), 1839 ulen, skb->len, 1840 &daddr, ntohs(uh->dest)); 1841 goto drop; 1842 1843 csum_error: 1844 /* 1845 * RFC1122: OK. Discards the bad packet silently (as far as 1846 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1847 */ 1848 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1849 proto == IPPROTO_UDPLITE ? "Lite" : "", 1850 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1851 ulen); 1852 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1853 drop: 1854 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1855 kfree_skb(skb); 1856 return 0; 1857 } 1858 1859 /* We can only early demux multicast if there is a single matching socket. 1860 * If more than one socket found returns NULL 1861 */ 1862 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1863 __be16 loc_port, __be32 loc_addr, 1864 __be16 rmt_port, __be32 rmt_addr, 1865 int dif) 1866 { 1867 struct sock *sk, *result; 1868 struct hlist_nulls_node *node; 1869 unsigned short hnum = ntohs(loc_port); 1870 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1871 struct udp_hslot *hslot = &udp_table.hash[slot]; 1872 1873 /* Do not bother scanning a too big list */ 1874 if (hslot->count > 10) 1875 return NULL; 1876 1877 rcu_read_lock(); 1878 begin: 1879 count = 0; 1880 result = NULL; 1881 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1882 if (__udp_is_mcast_sock(net, sk, 1883 loc_port, loc_addr, 1884 rmt_port, rmt_addr, 1885 dif, hnum)) { 1886 result = sk; 1887 ++count; 1888 } 1889 } 1890 /* 1891 * if the nulls value we got at the end of this lookup is 1892 * not the expected one, we must restart lookup. 1893 * We probably met an item that was moved to another chain. 1894 */ 1895 if (get_nulls_value(node) != slot) 1896 goto begin; 1897 1898 if (result) { 1899 if (count != 1 || 1900 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1901 result = NULL; 1902 else if (unlikely(!__udp_is_mcast_sock(net, result, 1903 loc_port, loc_addr, 1904 rmt_port, rmt_addr, 1905 dif, hnum))) { 1906 sock_put(result); 1907 result = NULL; 1908 } 1909 } 1910 rcu_read_unlock(); 1911 return result; 1912 } 1913 1914 /* For unicast we should only early demux connected sockets or we can 1915 * break forwarding setups. The chains here can be long so only check 1916 * if the first socket is an exact match and if not move on. 1917 */ 1918 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1919 __be16 loc_port, __be32 loc_addr, 1920 __be16 rmt_port, __be32 rmt_addr, 1921 int dif) 1922 { 1923 struct sock *sk, *result; 1924 struct hlist_nulls_node *node; 1925 unsigned short hnum = ntohs(loc_port); 1926 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1927 unsigned int slot2 = hash2 & udp_table.mask; 1928 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1929 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 1930 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1931 1932 rcu_read_lock(); 1933 result = NULL; 1934 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 1935 if (INET_MATCH(sk, net, acookie, 1936 rmt_addr, loc_addr, ports, dif)) 1937 result = sk; 1938 /* Only check first socket in chain */ 1939 break; 1940 } 1941 1942 if (result) { 1943 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1944 result = NULL; 1945 else if (unlikely(!INET_MATCH(sk, net, acookie, 1946 rmt_addr, loc_addr, 1947 ports, dif))) { 1948 sock_put(result); 1949 result = NULL; 1950 } 1951 } 1952 rcu_read_unlock(); 1953 return result; 1954 } 1955 1956 void udp_v4_early_demux(struct sk_buff *skb) 1957 { 1958 struct net *net = dev_net(skb->dev); 1959 const struct iphdr *iph; 1960 const struct udphdr *uh; 1961 struct sock *sk; 1962 struct dst_entry *dst; 1963 int dif = skb->dev->ifindex; 1964 int ours; 1965 1966 /* validate the packet */ 1967 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1968 return; 1969 1970 iph = ip_hdr(skb); 1971 uh = udp_hdr(skb); 1972 1973 if (skb->pkt_type == PACKET_BROADCAST || 1974 skb->pkt_type == PACKET_MULTICAST) { 1975 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 1976 1977 if (!in_dev) 1978 return; 1979 1980 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 1981 iph->protocol); 1982 if (!ours) 1983 return; 1984 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1985 uh->source, iph->saddr, dif); 1986 } else if (skb->pkt_type == PACKET_HOST) { 1987 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1988 uh->source, iph->saddr, dif); 1989 } else { 1990 return; 1991 } 1992 1993 if (!sk) 1994 return; 1995 1996 skb->sk = sk; 1997 skb->destructor = sock_efree; 1998 dst = sk->sk_rx_dst; 1999 2000 if (dst) 2001 dst = dst_check(dst, 0); 2002 if (dst) 2003 skb_dst_set_noref(skb, dst); 2004 } 2005 2006 int udp_rcv(struct sk_buff *skb) 2007 { 2008 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2009 } 2010 2011 void udp_destroy_sock(struct sock *sk) 2012 { 2013 struct udp_sock *up = udp_sk(sk); 2014 bool slow = lock_sock_fast(sk); 2015 udp_flush_pending_frames(sk); 2016 unlock_sock_fast(sk, slow); 2017 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2018 void (*encap_destroy)(struct sock *sk); 2019 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2020 if (encap_destroy) 2021 encap_destroy(sk); 2022 } 2023 } 2024 2025 /* 2026 * Socket option code for UDP 2027 */ 2028 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2029 char __user *optval, unsigned int optlen, 2030 int (*push_pending_frames)(struct sock *)) 2031 { 2032 struct udp_sock *up = udp_sk(sk); 2033 int val, valbool; 2034 int err = 0; 2035 int is_udplite = IS_UDPLITE(sk); 2036 2037 if (optlen < sizeof(int)) 2038 return -EINVAL; 2039 2040 if (get_user(val, (int __user *)optval)) 2041 return -EFAULT; 2042 2043 valbool = val ? 1 : 0; 2044 2045 switch (optname) { 2046 case UDP_CORK: 2047 if (val != 0) { 2048 up->corkflag = 1; 2049 } else { 2050 up->corkflag = 0; 2051 lock_sock(sk); 2052 push_pending_frames(sk); 2053 release_sock(sk); 2054 } 2055 break; 2056 2057 case UDP_ENCAP: 2058 switch (val) { 2059 case 0: 2060 case UDP_ENCAP_ESPINUDP: 2061 case UDP_ENCAP_ESPINUDP_NON_IKE: 2062 up->encap_rcv = xfrm4_udp_encap_rcv; 2063 /* FALLTHROUGH */ 2064 case UDP_ENCAP_L2TPINUDP: 2065 up->encap_type = val; 2066 udp_encap_enable(); 2067 break; 2068 default: 2069 err = -ENOPROTOOPT; 2070 break; 2071 } 2072 break; 2073 2074 case UDP_NO_CHECK6_TX: 2075 up->no_check6_tx = valbool; 2076 break; 2077 2078 case UDP_NO_CHECK6_RX: 2079 up->no_check6_rx = valbool; 2080 break; 2081 2082 /* 2083 * UDP-Lite's partial checksum coverage (RFC 3828). 2084 */ 2085 /* The sender sets actual checksum coverage length via this option. 2086 * The case coverage > packet length is handled by send module. */ 2087 case UDPLITE_SEND_CSCOV: 2088 if (!is_udplite) /* Disable the option on UDP sockets */ 2089 return -ENOPROTOOPT; 2090 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2091 val = 8; 2092 else if (val > USHRT_MAX) 2093 val = USHRT_MAX; 2094 up->pcslen = val; 2095 up->pcflag |= UDPLITE_SEND_CC; 2096 break; 2097 2098 /* The receiver specifies a minimum checksum coverage value. To make 2099 * sense, this should be set to at least 8 (as done below). If zero is 2100 * used, this again means full checksum coverage. */ 2101 case UDPLITE_RECV_CSCOV: 2102 if (!is_udplite) /* Disable the option on UDP sockets */ 2103 return -ENOPROTOOPT; 2104 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2105 val = 8; 2106 else if (val > USHRT_MAX) 2107 val = USHRT_MAX; 2108 up->pcrlen = val; 2109 up->pcflag |= UDPLITE_RECV_CC; 2110 break; 2111 2112 default: 2113 err = -ENOPROTOOPT; 2114 break; 2115 } 2116 2117 return err; 2118 } 2119 EXPORT_SYMBOL(udp_lib_setsockopt); 2120 2121 int udp_setsockopt(struct sock *sk, int level, int optname, 2122 char __user *optval, unsigned int optlen) 2123 { 2124 if (level == SOL_UDP || level == SOL_UDPLITE) 2125 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2126 udp_push_pending_frames); 2127 return ip_setsockopt(sk, level, optname, optval, optlen); 2128 } 2129 2130 #ifdef CONFIG_COMPAT 2131 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2132 char __user *optval, unsigned int optlen) 2133 { 2134 if (level == SOL_UDP || level == SOL_UDPLITE) 2135 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2136 udp_push_pending_frames); 2137 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2138 } 2139 #endif 2140 2141 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2142 char __user *optval, int __user *optlen) 2143 { 2144 struct udp_sock *up = udp_sk(sk); 2145 int val, len; 2146 2147 if (get_user(len, optlen)) 2148 return -EFAULT; 2149 2150 len = min_t(unsigned int, len, sizeof(int)); 2151 2152 if (len < 0) 2153 return -EINVAL; 2154 2155 switch (optname) { 2156 case UDP_CORK: 2157 val = up->corkflag; 2158 break; 2159 2160 case UDP_ENCAP: 2161 val = up->encap_type; 2162 break; 2163 2164 case UDP_NO_CHECK6_TX: 2165 val = up->no_check6_tx; 2166 break; 2167 2168 case UDP_NO_CHECK6_RX: 2169 val = up->no_check6_rx; 2170 break; 2171 2172 /* The following two cannot be changed on UDP sockets, the return is 2173 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2174 case UDPLITE_SEND_CSCOV: 2175 val = up->pcslen; 2176 break; 2177 2178 case UDPLITE_RECV_CSCOV: 2179 val = up->pcrlen; 2180 break; 2181 2182 default: 2183 return -ENOPROTOOPT; 2184 } 2185 2186 if (put_user(len, optlen)) 2187 return -EFAULT; 2188 if (copy_to_user(optval, &val, len)) 2189 return -EFAULT; 2190 return 0; 2191 } 2192 EXPORT_SYMBOL(udp_lib_getsockopt); 2193 2194 int udp_getsockopt(struct sock *sk, int level, int optname, 2195 char __user *optval, int __user *optlen) 2196 { 2197 if (level == SOL_UDP || level == SOL_UDPLITE) 2198 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2199 return ip_getsockopt(sk, level, optname, optval, optlen); 2200 } 2201 2202 #ifdef CONFIG_COMPAT 2203 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2204 char __user *optval, int __user *optlen) 2205 { 2206 if (level == SOL_UDP || level == SOL_UDPLITE) 2207 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2208 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2209 } 2210 #endif 2211 /** 2212 * udp_poll - wait for a UDP event. 2213 * @file - file struct 2214 * @sock - socket 2215 * @wait - poll table 2216 * 2217 * This is same as datagram poll, except for the special case of 2218 * blocking sockets. If application is using a blocking fd 2219 * and a packet with checksum error is in the queue; 2220 * then it could get return from select indicating data available 2221 * but then block when reading it. Add special case code 2222 * to work around these arguably broken applications. 2223 */ 2224 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2225 { 2226 unsigned int mask = datagram_poll(file, sock, wait); 2227 struct sock *sk = sock->sk; 2228 2229 sock_rps_record_flow(sk); 2230 2231 /* Check for false positives due to checksum errors */ 2232 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2233 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2234 mask &= ~(POLLIN | POLLRDNORM); 2235 2236 return mask; 2237 2238 } 2239 EXPORT_SYMBOL(udp_poll); 2240 2241 struct proto udp_prot = { 2242 .name = "UDP", 2243 .owner = THIS_MODULE, 2244 .close = udp_lib_close, 2245 .connect = ip4_datagram_connect, 2246 .disconnect = udp_disconnect, 2247 .ioctl = udp_ioctl, 2248 .destroy = udp_destroy_sock, 2249 .setsockopt = udp_setsockopt, 2250 .getsockopt = udp_getsockopt, 2251 .sendmsg = udp_sendmsg, 2252 .recvmsg = udp_recvmsg, 2253 .sendpage = udp_sendpage, 2254 .backlog_rcv = __udp_queue_rcv_skb, 2255 .release_cb = ip4_datagram_release_cb, 2256 .hash = udp_lib_hash, 2257 .unhash = udp_lib_unhash, 2258 .rehash = udp_v4_rehash, 2259 .get_port = udp_v4_get_port, 2260 .memory_allocated = &udp_memory_allocated, 2261 .sysctl_mem = sysctl_udp_mem, 2262 .sysctl_wmem = &sysctl_udp_wmem_min, 2263 .sysctl_rmem = &sysctl_udp_rmem_min, 2264 .obj_size = sizeof(struct udp_sock), 2265 .slab_flags = SLAB_DESTROY_BY_RCU, 2266 .h.udp_table = &udp_table, 2267 #ifdef CONFIG_COMPAT 2268 .compat_setsockopt = compat_udp_setsockopt, 2269 .compat_getsockopt = compat_udp_getsockopt, 2270 #endif 2271 .clear_sk = sk_prot_clear_portaddr_nulls, 2272 }; 2273 EXPORT_SYMBOL(udp_prot); 2274 2275 /* ------------------------------------------------------------------------ */ 2276 #ifdef CONFIG_PROC_FS 2277 2278 static struct sock *udp_get_first(struct seq_file *seq, int start) 2279 { 2280 struct sock *sk; 2281 struct udp_iter_state *state = seq->private; 2282 struct net *net = seq_file_net(seq); 2283 2284 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2285 ++state->bucket) { 2286 struct hlist_nulls_node *node; 2287 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2288 2289 if (hlist_nulls_empty(&hslot->head)) 2290 continue; 2291 2292 spin_lock_bh(&hslot->lock); 2293 sk_nulls_for_each(sk, node, &hslot->head) { 2294 if (!net_eq(sock_net(sk), net)) 2295 continue; 2296 if (sk->sk_family == state->family) 2297 goto found; 2298 } 2299 spin_unlock_bh(&hslot->lock); 2300 } 2301 sk = NULL; 2302 found: 2303 return sk; 2304 } 2305 2306 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2307 { 2308 struct udp_iter_state *state = seq->private; 2309 struct net *net = seq_file_net(seq); 2310 2311 do { 2312 sk = sk_nulls_next(sk); 2313 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2314 2315 if (!sk) { 2316 if (state->bucket <= state->udp_table->mask) 2317 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2318 return udp_get_first(seq, state->bucket + 1); 2319 } 2320 return sk; 2321 } 2322 2323 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2324 { 2325 struct sock *sk = udp_get_first(seq, 0); 2326 2327 if (sk) 2328 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2329 --pos; 2330 return pos ? NULL : sk; 2331 } 2332 2333 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2334 { 2335 struct udp_iter_state *state = seq->private; 2336 state->bucket = MAX_UDP_PORTS; 2337 2338 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2339 } 2340 2341 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2342 { 2343 struct sock *sk; 2344 2345 if (v == SEQ_START_TOKEN) 2346 sk = udp_get_idx(seq, 0); 2347 else 2348 sk = udp_get_next(seq, v); 2349 2350 ++*pos; 2351 return sk; 2352 } 2353 2354 static void udp_seq_stop(struct seq_file *seq, void *v) 2355 { 2356 struct udp_iter_state *state = seq->private; 2357 2358 if (state->bucket <= state->udp_table->mask) 2359 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2360 } 2361 2362 int udp_seq_open(struct inode *inode, struct file *file) 2363 { 2364 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2365 struct udp_iter_state *s; 2366 int err; 2367 2368 err = seq_open_net(inode, file, &afinfo->seq_ops, 2369 sizeof(struct udp_iter_state)); 2370 if (err < 0) 2371 return err; 2372 2373 s = ((struct seq_file *)file->private_data)->private; 2374 s->family = afinfo->family; 2375 s->udp_table = afinfo->udp_table; 2376 return err; 2377 } 2378 EXPORT_SYMBOL(udp_seq_open); 2379 2380 /* ------------------------------------------------------------------------ */ 2381 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2382 { 2383 struct proc_dir_entry *p; 2384 int rc = 0; 2385 2386 afinfo->seq_ops.start = udp_seq_start; 2387 afinfo->seq_ops.next = udp_seq_next; 2388 afinfo->seq_ops.stop = udp_seq_stop; 2389 2390 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2391 afinfo->seq_fops, afinfo); 2392 if (!p) 2393 rc = -ENOMEM; 2394 return rc; 2395 } 2396 EXPORT_SYMBOL(udp_proc_register); 2397 2398 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2399 { 2400 remove_proc_entry(afinfo->name, net->proc_net); 2401 } 2402 EXPORT_SYMBOL(udp_proc_unregister); 2403 2404 /* ------------------------------------------------------------------------ */ 2405 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2406 int bucket) 2407 { 2408 struct inet_sock *inet = inet_sk(sp); 2409 __be32 dest = inet->inet_daddr; 2410 __be32 src = inet->inet_rcv_saddr; 2411 __u16 destp = ntohs(inet->inet_dport); 2412 __u16 srcp = ntohs(inet->inet_sport); 2413 2414 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2415 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2416 bucket, src, srcp, dest, destp, sp->sk_state, 2417 sk_wmem_alloc_get(sp), 2418 sk_rmem_alloc_get(sp), 2419 0, 0L, 0, 2420 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2421 0, sock_i_ino(sp), 2422 atomic_read(&sp->sk_refcnt), sp, 2423 atomic_read(&sp->sk_drops)); 2424 } 2425 2426 int udp4_seq_show(struct seq_file *seq, void *v) 2427 { 2428 seq_setwidth(seq, 127); 2429 if (v == SEQ_START_TOKEN) 2430 seq_puts(seq, " sl local_address rem_address st tx_queue " 2431 "rx_queue tr tm->when retrnsmt uid timeout " 2432 "inode ref pointer drops"); 2433 else { 2434 struct udp_iter_state *state = seq->private; 2435 2436 udp4_format_sock(v, seq, state->bucket); 2437 } 2438 seq_pad(seq, '\n'); 2439 return 0; 2440 } 2441 2442 static const struct file_operations udp_afinfo_seq_fops = { 2443 .owner = THIS_MODULE, 2444 .open = udp_seq_open, 2445 .read = seq_read, 2446 .llseek = seq_lseek, 2447 .release = seq_release_net 2448 }; 2449 2450 /* ------------------------------------------------------------------------ */ 2451 static struct udp_seq_afinfo udp4_seq_afinfo = { 2452 .name = "udp", 2453 .family = AF_INET, 2454 .udp_table = &udp_table, 2455 .seq_fops = &udp_afinfo_seq_fops, 2456 .seq_ops = { 2457 .show = udp4_seq_show, 2458 }, 2459 }; 2460 2461 static int __net_init udp4_proc_init_net(struct net *net) 2462 { 2463 return udp_proc_register(net, &udp4_seq_afinfo); 2464 } 2465 2466 static void __net_exit udp4_proc_exit_net(struct net *net) 2467 { 2468 udp_proc_unregister(net, &udp4_seq_afinfo); 2469 } 2470 2471 static struct pernet_operations udp4_net_ops = { 2472 .init = udp4_proc_init_net, 2473 .exit = udp4_proc_exit_net, 2474 }; 2475 2476 int __init udp4_proc_init(void) 2477 { 2478 return register_pernet_subsys(&udp4_net_ops); 2479 } 2480 2481 void udp4_proc_exit(void) 2482 { 2483 unregister_pernet_subsys(&udp4_net_ops); 2484 } 2485 #endif /* CONFIG_PROC_FS */ 2486 2487 static __initdata unsigned long uhash_entries; 2488 static int __init set_uhash_entries(char *str) 2489 { 2490 ssize_t ret; 2491 2492 if (!str) 2493 return 0; 2494 2495 ret = kstrtoul(str, 0, &uhash_entries); 2496 if (ret) 2497 return 0; 2498 2499 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2500 uhash_entries = UDP_HTABLE_SIZE_MIN; 2501 return 1; 2502 } 2503 __setup("uhash_entries=", set_uhash_entries); 2504 2505 void __init udp_table_init(struct udp_table *table, const char *name) 2506 { 2507 unsigned int i; 2508 2509 table->hash = alloc_large_system_hash(name, 2510 2 * sizeof(struct udp_hslot), 2511 uhash_entries, 2512 21, /* one slot per 2 MB */ 2513 0, 2514 &table->log, 2515 &table->mask, 2516 UDP_HTABLE_SIZE_MIN, 2517 64 * 1024); 2518 2519 table->hash2 = table->hash + (table->mask + 1); 2520 for (i = 0; i <= table->mask; i++) { 2521 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2522 table->hash[i].count = 0; 2523 spin_lock_init(&table->hash[i].lock); 2524 } 2525 for (i = 0; i <= table->mask; i++) { 2526 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2527 table->hash2[i].count = 0; 2528 spin_lock_init(&table->hash2[i].lock); 2529 } 2530 } 2531 2532 u32 udp_flow_hashrnd(void) 2533 { 2534 static u32 hashrnd __read_mostly; 2535 2536 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2537 2538 return hashrnd; 2539 } 2540 EXPORT_SYMBOL(udp_flow_hashrnd); 2541 2542 void __init udp_init(void) 2543 { 2544 unsigned long limit; 2545 2546 udp_table_init(&udp_table, "UDP"); 2547 limit = nr_free_buffer_pages() / 8; 2548 limit = max(limit, 128UL); 2549 sysctl_udp_mem[0] = limit / 4 * 3; 2550 sysctl_udp_mem[1] = limit; 2551 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2552 2553 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2554 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2555 } 2556