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 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 379 score++; 380 return score; 381 } 382 383 /* 384 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 385 */ 386 static inline int compute_score2(struct sock *sk, struct net *net, 387 __be32 saddr, __be16 sport, 388 __be32 daddr, unsigned int hnum, int dif) 389 { 390 int score; 391 struct inet_sock *inet; 392 393 if (!net_eq(sock_net(sk), net) || 394 ipv6_only_sock(sk)) 395 return -1; 396 397 inet = inet_sk(sk); 398 399 if (inet->inet_rcv_saddr != daddr || 400 inet->inet_num != hnum) 401 return -1; 402 403 score = (sk->sk_family == PF_INET) ? 2 : 1; 404 405 if (inet->inet_daddr) { 406 if (inet->inet_daddr != saddr) 407 return -1; 408 score += 4; 409 } 410 411 if (inet->inet_dport) { 412 if (inet->inet_dport != sport) 413 return -1; 414 score += 4; 415 } 416 417 if (sk->sk_bound_dev_if) { 418 if (sk->sk_bound_dev_if != dif) 419 return -1; 420 score += 4; 421 } 422 423 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 424 score++; 425 426 return score; 427 } 428 429 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 430 const __u16 lport, const __be32 faddr, 431 const __be16 fport) 432 { 433 static u32 udp_ehash_secret __read_mostly; 434 435 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 436 437 return __inet_ehashfn(laddr, lport, faddr, fport, 438 udp_ehash_secret + net_hash_mix(net)); 439 } 440 441 /* called with read_rcu_lock() */ 442 static struct sock *udp4_lib_lookup2(struct net *net, 443 __be32 saddr, __be16 sport, 444 __be32 daddr, unsigned int hnum, int dif, 445 struct udp_hslot *hslot2, unsigned int slot2) 446 { 447 struct sock *sk, *result; 448 struct hlist_nulls_node *node; 449 int score, badness, matches = 0, reuseport = 0; 450 u32 hash = 0; 451 452 begin: 453 result = NULL; 454 badness = 0; 455 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 456 score = compute_score2(sk, net, saddr, sport, 457 daddr, hnum, dif); 458 if (score > badness) { 459 result = sk; 460 badness = score; 461 reuseport = sk->sk_reuseport; 462 if (reuseport) { 463 hash = udp_ehashfn(net, daddr, hnum, 464 saddr, sport); 465 matches = 1; 466 } 467 } else if (score == badness && reuseport) { 468 matches++; 469 if (reciprocal_scale(hash, matches) == 0) 470 result = sk; 471 hash = next_pseudo_random32(hash); 472 } 473 } 474 /* 475 * if the nulls value we got at the end of this lookup is 476 * not the expected one, we must restart lookup. 477 * We probably met an item that was moved to another chain. 478 */ 479 if (get_nulls_value(node) != slot2) 480 goto begin; 481 if (result) { 482 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 483 result = NULL; 484 else if (unlikely(compute_score2(result, net, saddr, sport, 485 daddr, hnum, dif) < badness)) { 486 sock_put(result); 487 goto begin; 488 } 489 } 490 return result; 491 } 492 493 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 494 * harder than this. -DaveM 495 */ 496 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 497 __be16 sport, __be32 daddr, __be16 dport, 498 int dif, struct udp_table *udptable) 499 { 500 struct sock *sk, *result; 501 struct hlist_nulls_node *node; 502 unsigned short hnum = ntohs(dport); 503 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 504 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 505 int score, badness, matches = 0, reuseport = 0; 506 u32 hash = 0; 507 508 rcu_read_lock(); 509 if (hslot->count > 10) { 510 hash2 = udp4_portaddr_hash(net, daddr, hnum); 511 slot2 = hash2 & udptable->mask; 512 hslot2 = &udptable->hash2[slot2]; 513 if (hslot->count < hslot2->count) 514 goto begin; 515 516 result = udp4_lib_lookup2(net, saddr, sport, 517 daddr, hnum, dif, 518 hslot2, slot2); 519 if (!result) { 520 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 521 slot2 = hash2 & udptable->mask; 522 hslot2 = &udptable->hash2[slot2]; 523 if (hslot->count < hslot2->count) 524 goto begin; 525 526 result = udp4_lib_lookup2(net, saddr, sport, 527 htonl(INADDR_ANY), hnum, dif, 528 hslot2, slot2); 529 } 530 rcu_read_unlock(); 531 return result; 532 } 533 begin: 534 result = NULL; 535 badness = 0; 536 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 537 score = compute_score(sk, net, saddr, hnum, sport, 538 daddr, dport, dif); 539 if (score > badness) { 540 result = sk; 541 badness = score; 542 reuseport = sk->sk_reuseport; 543 if (reuseport) { 544 hash = udp_ehashfn(net, daddr, hnum, 545 saddr, sport); 546 matches = 1; 547 } 548 } else if (score == badness && reuseport) { 549 matches++; 550 if (reciprocal_scale(hash, matches) == 0) 551 result = sk; 552 hash = next_pseudo_random32(hash); 553 } 554 } 555 /* 556 * if the nulls value we got at the end of this lookup is 557 * not the expected one, we must restart lookup. 558 * We probably met an item that was moved to another chain. 559 */ 560 if (get_nulls_value(node) != slot) 561 goto begin; 562 563 if (result) { 564 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 565 result = NULL; 566 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 567 daddr, dport, dif) < badness)) { 568 sock_put(result); 569 goto begin; 570 } 571 } 572 rcu_read_unlock(); 573 return result; 574 } 575 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 576 577 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 578 __be16 sport, __be16 dport, 579 struct udp_table *udptable) 580 { 581 const struct iphdr *iph = ip_hdr(skb); 582 583 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 584 iph->daddr, dport, inet_iif(skb), 585 udptable); 586 } 587 588 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 589 __be32 daddr, __be16 dport, int dif) 590 { 591 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 592 } 593 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 594 595 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 596 __be16 loc_port, __be32 loc_addr, 597 __be16 rmt_port, __be32 rmt_addr, 598 int dif, unsigned short hnum) 599 { 600 struct inet_sock *inet = inet_sk(sk); 601 602 if (!net_eq(sock_net(sk), net) || 603 udp_sk(sk)->udp_port_hash != hnum || 604 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 605 (inet->inet_dport != rmt_port && inet->inet_dport) || 606 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 607 ipv6_only_sock(sk) || 608 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 609 return false; 610 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 611 return false; 612 return true; 613 } 614 615 /* 616 * This routine is called by the ICMP module when it gets some 617 * sort of error condition. If err < 0 then the socket should 618 * be closed and the error returned to the user. If err > 0 619 * it's just the icmp type << 8 | icmp code. 620 * Header points to the ip header of the error packet. We move 621 * on past this. Then (as it used to claim before adjustment) 622 * header points to the first 8 bytes of the udp header. We need 623 * to find the appropriate port. 624 */ 625 626 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 627 { 628 struct inet_sock *inet; 629 const struct iphdr *iph = (const struct iphdr *)skb->data; 630 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 631 const int type = icmp_hdr(skb)->type; 632 const int code = icmp_hdr(skb)->code; 633 struct sock *sk; 634 int harderr; 635 int err; 636 struct net *net = dev_net(skb->dev); 637 638 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 639 iph->saddr, uh->source, skb->dev->ifindex, udptable); 640 if (!sk) { 641 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 642 return; /* No socket for error */ 643 } 644 645 err = 0; 646 harderr = 0; 647 inet = inet_sk(sk); 648 649 switch (type) { 650 default: 651 case ICMP_TIME_EXCEEDED: 652 err = EHOSTUNREACH; 653 break; 654 case ICMP_SOURCE_QUENCH: 655 goto out; 656 case ICMP_PARAMETERPROB: 657 err = EPROTO; 658 harderr = 1; 659 break; 660 case ICMP_DEST_UNREACH: 661 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 662 ipv4_sk_update_pmtu(skb, sk, info); 663 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 664 err = EMSGSIZE; 665 harderr = 1; 666 break; 667 } 668 goto out; 669 } 670 err = EHOSTUNREACH; 671 if (code <= NR_ICMP_UNREACH) { 672 harderr = icmp_err_convert[code].fatal; 673 err = icmp_err_convert[code].errno; 674 } 675 break; 676 case ICMP_REDIRECT: 677 ipv4_sk_redirect(skb, sk); 678 goto out; 679 } 680 681 /* 682 * RFC1122: OK. Passes ICMP errors back to application, as per 683 * 4.1.3.3. 684 */ 685 if (!inet->recverr) { 686 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 687 goto out; 688 } else 689 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 690 691 sk->sk_err = err; 692 sk->sk_error_report(sk); 693 out: 694 sock_put(sk); 695 } 696 697 void udp_err(struct sk_buff *skb, u32 info) 698 { 699 __udp4_lib_err(skb, info, &udp_table); 700 } 701 702 /* 703 * Throw away all pending data and cancel the corking. Socket is locked. 704 */ 705 void udp_flush_pending_frames(struct sock *sk) 706 { 707 struct udp_sock *up = udp_sk(sk); 708 709 if (up->pending) { 710 up->len = 0; 711 up->pending = 0; 712 ip_flush_pending_frames(sk); 713 } 714 } 715 EXPORT_SYMBOL(udp_flush_pending_frames); 716 717 /** 718 * udp4_hwcsum - handle outgoing HW checksumming 719 * @skb: sk_buff containing the filled-in UDP header 720 * (checksum field must be zeroed out) 721 * @src: source IP address 722 * @dst: destination IP address 723 */ 724 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 725 { 726 struct udphdr *uh = udp_hdr(skb); 727 int offset = skb_transport_offset(skb); 728 int len = skb->len - offset; 729 int hlen = len; 730 __wsum csum = 0; 731 732 if (!skb_has_frag_list(skb)) { 733 /* 734 * Only one fragment on the socket. 735 */ 736 skb->csum_start = skb_transport_header(skb) - skb->head; 737 skb->csum_offset = offsetof(struct udphdr, check); 738 uh->check = ~csum_tcpudp_magic(src, dst, len, 739 IPPROTO_UDP, 0); 740 } else { 741 struct sk_buff *frags; 742 743 /* 744 * HW-checksum won't work as there are two or more 745 * fragments on the socket so that all csums of sk_buffs 746 * should be together 747 */ 748 skb_walk_frags(skb, frags) { 749 csum = csum_add(csum, frags->csum); 750 hlen -= frags->len; 751 } 752 753 csum = skb_checksum(skb, offset, hlen, csum); 754 skb->ip_summed = CHECKSUM_NONE; 755 756 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 757 if (uh->check == 0) 758 uh->check = CSUM_MANGLED_0; 759 } 760 } 761 EXPORT_SYMBOL_GPL(udp4_hwcsum); 762 763 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 764 * for the simple case like when setting the checksum for a UDP tunnel. 765 */ 766 void udp_set_csum(bool nocheck, struct sk_buff *skb, 767 __be32 saddr, __be32 daddr, int len) 768 { 769 struct udphdr *uh = udp_hdr(skb); 770 771 if (nocheck) 772 uh->check = 0; 773 else if (skb_is_gso(skb)) 774 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 775 else if (skb_dst(skb) && skb_dst(skb)->dev && 776 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) { 777 778 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 779 780 skb->ip_summed = CHECKSUM_PARTIAL; 781 skb->csum_start = skb_transport_header(skb) - skb->head; 782 skb->csum_offset = offsetof(struct udphdr, check); 783 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 784 } else { 785 __wsum csum; 786 787 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 788 789 uh->check = 0; 790 csum = skb_checksum(skb, 0, len, 0); 791 uh->check = udp_v4_check(len, saddr, daddr, csum); 792 if (uh->check == 0) 793 uh->check = CSUM_MANGLED_0; 794 795 skb->ip_summed = CHECKSUM_UNNECESSARY; 796 } 797 } 798 EXPORT_SYMBOL(udp_set_csum); 799 800 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 801 { 802 struct sock *sk = skb->sk; 803 struct inet_sock *inet = inet_sk(sk); 804 struct udphdr *uh; 805 int err = 0; 806 int is_udplite = IS_UDPLITE(sk); 807 int offset = skb_transport_offset(skb); 808 int len = skb->len - offset; 809 __wsum csum = 0; 810 811 /* 812 * Create a UDP header 813 */ 814 uh = udp_hdr(skb); 815 uh->source = inet->inet_sport; 816 uh->dest = fl4->fl4_dport; 817 uh->len = htons(len); 818 uh->check = 0; 819 820 if (is_udplite) /* UDP-Lite */ 821 csum = udplite_csum(skb); 822 823 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 824 825 skb->ip_summed = CHECKSUM_NONE; 826 goto send; 827 828 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 829 830 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 831 goto send; 832 833 } else 834 csum = udp_csum(skb); 835 836 /* add protocol-dependent pseudo-header */ 837 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 838 sk->sk_protocol, csum); 839 if (uh->check == 0) 840 uh->check = CSUM_MANGLED_0; 841 842 send: 843 err = ip_send_skb(sock_net(sk), skb); 844 if (err) { 845 if (err == -ENOBUFS && !inet->recverr) { 846 UDP_INC_STATS_USER(sock_net(sk), 847 UDP_MIB_SNDBUFERRORS, is_udplite); 848 err = 0; 849 } 850 } else 851 UDP_INC_STATS_USER(sock_net(sk), 852 UDP_MIB_OUTDATAGRAMS, is_udplite); 853 return err; 854 } 855 856 /* 857 * Push out all pending data as one UDP datagram. Socket is locked. 858 */ 859 int udp_push_pending_frames(struct sock *sk) 860 { 861 struct udp_sock *up = udp_sk(sk); 862 struct inet_sock *inet = inet_sk(sk); 863 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 864 struct sk_buff *skb; 865 int err = 0; 866 867 skb = ip_finish_skb(sk, fl4); 868 if (!skb) 869 goto out; 870 871 err = udp_send_skb(skb, fl4); 872 873 out: 874 up->len = 0; 875 up->pending = 0; 876 return err; 877 } 878 EXPORT_SYMBOL(udp_push_pending_frames); 879 880 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 881 { 882 struct inet_sock *inet = inet_sk(sk); 883 struct udp_sock *up = udp_sk(sk); 884 struct flowi4 fl4_stack; 885 struct flowi4 *fl4; 886 int ulen = len; 887 struct ipcm_cookie ipc; 888 struct rtable *rt = NULL; 889 int free = 0; 890 int connected = 0; 891 __be32 daddr, faddr, saddr; 892 __be16 dport; 893 u8 tos; 894 int err, is_udplite = IS_UDPLITE(sk); 895 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 896 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 897 struct sk_buff *skb; 898 struct ip_options_data opt_copy; 899 900 if (len > 0xFFFF) 901 return -EMSGSIZE; 902 903 /* 904 * Check the flags. 905 */ 906 907 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 908 return -EOPNOTSUPP; 909 910 ipc.opt = NULL; 911 ipc.tx_flags = 0; 912 ipc.ttl = 0; 913 ipc.tos = -1; 914 915 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 916 917 fl4 = &inet->cork.fl.u.ip4; 918 if (up->pending) { 919 /* 920 * There are pending frames. 921 * The socket lock must be held while it's corked. 922 */ 923 lock_sock(sk); 924 if (likely(up->pending)) { 925 if (unlikely(up->pending != AF_INET)) { 926 release_sock(sk); 927 return -EINVAL; 928 } 929 goto do_append_data; 930 } 931 release_sock(sk); 932 } 933 ulen += sizeof(struct udphdr); 934 935 /* 936 * Get and verify the address. 937 */ 938 if (msg->msg_name) { 939 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 940 if (msg->msg_namelen < sizeof(*usin)) 941 return -EINVAL; 942 if (usin->sin_family != AF_INET) { 943 if (usin->sin_family != AF_UNSPEC) 944 return -EAFNOSUPPORT; 945 } 946 947 daddr = usin->sin_addr.s_addr; 948 dport = usin->sin_port; 949 if (dport == 0) 950 return -EINVAL; 951 } else { 952 if (sk->sk_state != TCP_ESTABLISHED) 953 return -EDESTADDRREQ; 954 daddr = inet->inet_daddr; 955 dport = inet->inet_dport; 956 /* Open fast path for connected socket. 957 Route will not be used, if at least one option is set. 958 */ 959 connected = 1; 960 } 961 ipc.addr = inet->inet_saddr; 962 963 ipc.oif = sk->sk_bound_dev_if; 964 965 sock_tx_timestamp(sk, &ipc.tx_flags); 966 967 if (msg->msg_controllen) { 968 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 969 sk->sk_family == AF_INET6); 970 if (err) 971 return err; 972 if (ipc.opt) 973 free = 1; 974 connected = 0; 975 } 976 if (!ipc.opt) { 977 struct ip_options_rcu *inet_opt; 978 979 rcu_read_lock(); 980 inet_opt = rcu_dereference(inet->inet_opt); 981 if (inet_opt) { 982 memcpy(&opt_copy, inet_opt, 983 sizeof(*inet_opt) + inet_opt->opt.optlen); 984 ipc.opt = &opt_copy.opt; 985 } 986 rcu_read_unlock(); 987 } 988 989 saddr = ipc.addr; 990 ipc.addr = faddr = daddr; 991 992 if (ipc.opt && ipc.opt->opt.srr) { 993 if (!daddr) 994 return -EINVAL; 995 faddr = ipc.opt->opt.faddr; 996 connected = 0; 997 } 998 tos = get_rttos(&ipc, inet); 999 if (sock_flag(sk, SOCK_LOCALROUTE) || 1000 (msg->msg_flags & MSG_DONTROUTE) || 1001 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1002 tos |= RTO_ONLINK; 1003 connected = 0; 1004 } 1005 1006 if (ipv4_is_multicast(daddr)) { 1007 if (!ipc.oif) 1008 ipc.oif = inet->mc_index; 1009 if (!saddr) 1010 saddr = inet->mc_addr; 1011 connected = 0; 1012 } else if (!ipc.oif) 1013 ipc.oif = inet->uc_index; 1014 1015 if (connected) 1016 rt = (struct rtable *)sk_dst_check(sk, 0); 1017 1018 if (!rt) { 1019 struct net *net = sock_net(sk); 1020 __u8 flow_flags = inet_sk_flowi_flags(sk); 1021 1022 fl4 = &fl4_stack; 1023 1024 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1025 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1026 flow_flags, 1027 faddr, saddr, dport, inet->inet_sport); 1028 1029 if (!saddr && ipc.oif) 1030 l3mdev_get_saddr(net, ipc.oif, fl4); 1031 1032 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1033 rt = ip_route_output_flow(net, fl4, sk); 1034 if (IS_ERR(rt)) { 1035 err = PTR_ERR(rt); 1036 rt = NULL; 1037 if (err == -ENETUNREACH) 1038 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1039 goto out; 1040 } 1041 1042 err = -EACCES; 1043 if ((rt->rt_flags & RTCF_BROADCAST) && 1044 !sock_flag(sk, SOCK_BROADCAST)) 1045 goto out; 1046 if (connected) 1047 sk_dst_set(sk, dst_clone(&rt->dst)); 1048 } 1049 1050 if (msg->msg_flags&MSG_CONFIRM) 1051 goto do_confirm; 1052 back_from_confirm: 1053 1054 saddr = fl4->saddr; 1055 if (!ipc.addr) 1056 daddr = ipc.addr = fl4->daddr; 1057 1058 /* Lockless fast path for the non-corking case. */ 1059 if (!corkreq) { 1060 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1061 sizeof(struct udphdr), &ipc, &rt, 1062 msg->msg_flags); 1063 err = PTR_ERR(skb); 1064 if (!IS_ERR_OR_NULL(skb)) 1065 err = udp_send_skb(skb, fl4); 1066 goto out; 1067 } 1068 1069 lock_sock(sk); 1070 if (unlikely(up->pending)) { 1071 /* The socket is already corked while preparing it. */ 1072 /* ... which is an evident application bug. --ANK */ 1073 release_sock(sk); 1074 1075 net_dbg_ratelimited("cork app bug 2\n"); 1076 err = -EINVAL; 1077 goto out; 1078 } 1079 /* 1080 * Now cork the socket to pend data. 1081 */ 1082 fl4 = &inet->cork.fl.u.ip4; 1083 fl4->daddr = daddr; 1084 fl4->saddr = saddr; 1085 fl4->fl4_dport = dport; 1086 fl4->fl4_sport = inet->inet_sport; 1087 up->pending = AF_INET; 1088 1089 do_append_data: 1090 up->len += ulen; 1091 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1092 sizeof(struct udphdr), &ipc, &rt, 1093 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1094 if (err) 1095 udp_flush_pending_frames(sk); 1096 else if (!corkreq) 1097 err = udp_push_pending_frames(sk); 1098 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1099 up->pending = 0; 1100 release_sock(sk); 1101 1102 out: 1103 ip_rt_put(rt); 1104 if (free) 1105 kfree(ipc.opt); 1106 if (!err) 1107 return len; 1108 /* 1109 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1110 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1111 * we don't have a good statistic (IpOutDiscards but it can be too many 1112 * things). We could add another new stat but at least for now that 1113 * seems like overkill. 1114 */ 1115 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1116 UDP_INC_STATS_USER(sock_net(sk), 1117 UDP_MIB_SNDBUFERRORS, is_udplite); 1118 } 1119 return err; 1120 1121 do_confirm: 1122 dst_confirm(&rt->dst); 1123 if (!(msg->msg_flags&MSG_PROBE) || len) 1124 goto back_from_confirm; 1125 err = 0; 1126 goto out; 1127 } 1128 EXPORT_SYMBOL(udp_sendmsg); 1129 1130 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1131 size_t size, int flags) 1132 { 1133 struct inet_sock *inet = inet_sk(sk); 1134 struct udp_sock *up = udp_sk(sk); 1135 int ret; 1136 1137 if (flags & MSG_SENDPAGE_NOTLAST) 1138 flags |= MSG_MORE; 1139 1140 if (!up->pending) { 1141 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1142 1143 /* Call udp_sendmsg to specify destination address which 1144 * sendpage interface can't pass. 1145 * This will succeed only when the socket is connected. 1146 */ 1147 ret = udp_sendmsg(sk, &msg, 0); 1148 if (ret < 0) 1149 return ret; 1150 } 1151 1152 lock_sock(sk); 1153 1154 if (unlikely(!up->pending)) { 1155 release_sock(sk); 1156 1157 net_dbg_ratelimited("udp cork app bug 3\n"); 1158 return -EINVAL; 1159 } 1160 1161 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1162 page, offset, size, flags); 1163 if (ret == -EOPNOTSUPP) { 1164 release_sock(sk); 1165 return sock_no_sendpage(sk->sk_socket, page, offset, 1166 size, flags); 1167 } 1168 if (ret < 0) { 1169 udp_flush_pending_frames(sk); 1170 goto out; 1171 } 1172 1173 up->len += size; 1174 if (!(up->corkflag || (flags&MSG_MORE))) 1175 ret = udp_push_pending_frames(sk); 1176 if (!ret) 1177 ret = size; 1178 out: 1179 release_sock(sk); 1180 return ret; 1181 } 1182 1183 /** 1184 * first_packet_length - return length of first packet in receive queue 1185 * @sk: socket 1186 * 1187 * Drops all bad checksum frames, until a valid one is found. 1188 * Returns the length of found skb, or 0 if none is found. 1189 */ 1190 static unsigned int first_packet_length(struct sock *sk) 1191 { 1192 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1193 struct sk_buff *skb; 1194 unsigned int res; 1195 1196 __skb_queue_head_init(&list_kill); 1197 1198 spin_lock_bh(&rcvq->lock); 1199 while ((skb = skb_peek(rcvq)) != NULL && 1200 udp_lib_checksum_complete(skb)) { 1201 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1202 IS_UDPLITE(sk)); 1203 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1204 IS_UDPLITE(sk)); 1205 atomic_inc(&sk->sk_drops); 1206 __skb_unlink(skb, rcvq); 1207 __skb_queue_tail(&list_kill, skb); 1208 } 1209 res = skb ? skb->len : 0; 1210 spin_unlock_bh(&rcvq->lock); 1211 1212 if (!skb_queue_empty(&list_kill)) { 1213 bool slow = lock_sock_fast(sk); 1214 1215 __skb_queue_purge(&list_kill); 1216 sk_mem_reclaim_partial(sk); 1217 unlock_sock_fast(sk, slow); 1218 } 1219 return res; 1220 } 1221 1222 /* 1223 * IOCTL requests applicable to the UDP protocol 1224 */ 1225 1226 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1227 { 1228 switch (cmd) { 1229 case SIOCOUTQ: 1230 { 1231 int amount = sk_wmem_alloc_get(sk); 1232 1233 return put_user(amount, (int __user *)arg); 1234 } 1235 1236 case SIOCINQ: 1237 { 1238 unsigned int amount = first_packet_length(sk); 1239 1240 if (amount) 1241 /* 1242 * We will only return the amount 1243 * of this packet since that is all 1244 * that will be read. 1245 */ 1246 amount -= sizeof(struct udphdr); 1247 1248 return put_user(amount, (int __user *)arg); 1249 } 1250 1251 default: 1252 return -ENOIOCTLCMD; 1253 } 1254 1255 return 0; 1256 } 1257 EXPORT_SYMBOL(udp_ioctl); 1258 1259 /* 1260 * This should be easy, if there is something there we 1261 * return it, otherwise we block. 1262 */ 1263 1264 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1265 int flags, int *addr_len) 1266 { 1267 struct inet_sock *inet = inet_sk(sk); 1268 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1269 struct sk_buff *skb; 1270 unsigned int ulen, copied; 1271 int peeked, off = 0; 1272 int err; 1273 int is_udplite = IS_UDPLITE(sk); 1274 bool slow; 1275 1276 if (flags & MSG_ERRQUEUE) 1277 return ip_recv_error(sk, msg, len, addr_len); 1278 1279 try_again: 1280 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1281 &peeked, &off, &err); 1282 if (!skb) 1283 goto out; 1284 1285 ulen = skb->len - sizeof(struct udphdr); 1286 copied = len; 1287 if (copied > ulen) 1288 copied = ulen; 1289 else if (copied < ulen) 1290 msg->msg_flags |= MSG_TRUNC; 1291 1292 /* 1293 * If checksum is needed at all, try to do it while copying the 1294 * data. If the data is truncated, or if we only want a partial 1295 * coverage checksum (UDP-Lite), do it before the copy. 1296 */ 1297 1298 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1299 if (udp_lib_checksum_complete(skb)) 1300 goto csum_copy_err; 1301 } 1302 1303 if (skb_csum_unnecessary(skb)) 1304 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 1305 msg, copied); 1306 else { 1307 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), 1308 msg); 1309 1310 if (err == -EINVAL) 1311 goto csum_copy_err; 1312 } 1313 1314 if (unlikely(err)) { 1315 trace_kfree_skb(skb, udp_recvmsg); 1316 if (!peeked) { 1317 atomic_inc(&sk->sk_drops); 1318 UDP_INC_STATS_USER(sock_net(sk), 1319 UDP_MIB_INERRORS, is_udplite); 1320 } 1321 goto out_free; 1322 } 1323 1324 if (!peeked) 1325 UDP_INC_STATS_USER(sock_net(sk), 1326 UDP_MIB_INDATAGRAMS, is_udplite); 1327 1328 sock_recv_ts_and_drops(msg, sk, skb); 1329 1330 /* Copy the address. */ 1331 if (sin) { 1332 sin->sin_family = AF_INET; 1333 sin->sin_port = udp_hdr(skb)->source; 1334 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1335 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1336 *addr_len = sizeof(*sin); 1337 } 1338 if (inet->cmsg_flags) 1339 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr)); 1340 1341 err = copied; 1342 if (flags & MSG_TRUNC) 1343 err = ulen; 1344 1345 out_free: 1346 skb_free_datagram_locked(sk, skb); 1347 out: 1348 return err; 1349 1350 csum_copy_err: 1351 slow = lock_sock_fast(sk); 1352 if (!skb_kill_datagram(sk, skb, flags)) { 1353 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1354 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1355 } 1356 unlock_sock_fast(sk, slow); 1357 1358 /* starting over for a new packet, but check if we need to yield */ 1359 cond_resched(); 1360 msg->msg_flags &= ~MSG_TRUNC; 1361 goto try_again; 1362 } 1363 1364 int udp_disconnect(struct sock *sk, int flags) 1365 { 1366 struct inet_sock *inet = inet_sk(sk); 1367 /* 1368 * 1003.1g - break association. 1369 */ 1370 1371 sk->sk_state = TCP_CLOSE; 1372 inet->inet_daddr = 0; 1373 inet->inet_dport = 0; 1374 sock_rps_reset_rxhash(sk); 1375 sk->sk_bound_dev_if = 0; 1376 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1377 inet_reset_saddr(sk); 1378 1379 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1380 sk->sk_prot->unhash(sk); 1381 inet->inet_sport = 0; 1382 } 1383 sk_dst_reset(sk); 1384 return 0; 1385 } 1386 EXPORT_SYMBOL(udp_disconnect); 1387 1388 void udp_lib_unhash(struct sock *sk) 1389 { 1390 if (sk_hashed(sk)) { 1391 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1392 struct udp_hslot *hslot, *hslot2; 1393 1394 hslot = udp_hashslot(udptable, sock_net(sk), 1395 udp_sk(sk)->udp_port_hash); 1396 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1397 1398 spin_lock_bh(&hslot->lock); 1399 if (sk_nulls_del_node_init_rcu(sk)) { 1400 hslot->count--; 1401 inet_sk(sk)->inet_num = 0; 1402 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1403 1404 spin_lock(&hslot2->lock); 1405 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1406 hslot2->count--; 1407 spin_unlock(&hslot2->lock); 1408 } 1409 spin_unlock_bh(&hslot->lock); 1410 } 1411 } 1412 EXPORT_SYMBOL(udp_lib_unhash); 1413 1414 /* 1415 * inet_rcv_saddr was changed, we must rehash secondary hash 1416 */ 1417 void udp_lib_rehash(struct sock *sk, u16 newhash) 1418 { 1419 if (sk_hashed(sk)) { 1420 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1421 struct udp_hslot *hslot, *hslot2, *nhslot2; 1422 1423 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1424 nhslot2 = udp_hashslot2(udptable, newhash); 1425 udp_sk(sk)->udp_portaddr_hash = newhash; 1426 if (hslot2 != nhslot2) { 1427 hslot = udp_hashslot(udptable, sock_net(sk), 1428 udp_sk(sk)->udp_port_hash); 1429 /* we must lock primary chain too */ 1430 spin_lock_bh(&hslot->lock); 1431 1432 spin_lock(&hslot2->lock); 1433 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1434 hslot2->count--; 1435 spin_unlock(&hslot2->lock); 1436 1437 spin_lock(&nhslot2->lock); 1438 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1439 &nhslot2->head); 1440 nhslot2->count++; 1441 spin_unlock(&nhslot2->lock); 1442 1443 spin_unlock_bh(&hslot->lock); 1444 } 1445 } 1446 } 1447 EXPORT_SYMBOL(udp_lib_rehash); 1448 1449 static void udp_v4_rehash(struct sock *sk) 1450 { 1451 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1452 inet_sk(sk)->inet_rcv_saddr, 1453 inet_sk(sk)->inet_num); 1454 udp_lib_rehash(sk, new_hash); 1455 } 1456 1457 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1458 { 1459 int rc; 1460 1461 if (inet_sk(sk)->inet_daddr) { 1462 sock_rps_save_rxhash(sk, skb); 1463 sk_mark_napi_id(sk, skb); 1464 sk_incoming_cpu_update(sk); 1465 } 1466 1467 rc = sock_queue_rcv_skb(sk, skb); 1468 if (rc < 0) { 1469 int is_udplite = IS_UDPLITE(sk); 1470 1471 /* Note that an ENOMEM error is charged twice */ 1472 if (rc == -ENOMEM) 1473 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1474 is_udplite); 1475 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1476 kfree_skb(skb); 1477 trace_udp_fail_queue_rcv_skb(rc, sk); 1478 return -1; 1479 } 1480 1481 return 0; 1482 1483 } 1484 1485 static struct static_key udp_encap_needed __read_mostly; 1486 void udp_encap_enable(void) 1487 { 1488 if (!static_key_enabled(&udp_encap_needed)) 1489 static_key_slow_inc(&udp_encap_needed); 1490 } 1491 EXPORT_SYMBOL(udp_encap_enable); 1492 1493 /* returns: 1494 * -1: error 1495 * 0: success 1496 * >0: "udp encap" protocol resubmission 1497 * 1498 * Note that in the success and error cases, the skb is assumed to 1499 * have either been requeued or freed. 1500 */ 1501 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1502 { 1503 struct udp_sock *up = udp_sk(sk); 1504 int rc; 1505 int is_udplite = IS_UDPLITE(sk); 1506 1507 /* 1508 * Charge it to the socket, dropping if the queue is full. 1509 */ 1510 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1511 goto drop; 1512 nf_reset(skb); 1513 1514 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1515 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1516 1517 /* 1518 * This is an encapsulation socket so pass the skb to 1519 * the socket's udp_encap_rcv() hook. Otherwise, just 1520 * fall through and pass this up the UDP socket. 1521 * up->encap_rcv() returns the following value: 1522 * =0 if skb was successfully passed to the encap 1523 * handler or was discarded by it. 1524 * >0 if skb should be passed on to UDP. 1525 * <0 if skb should be resubmitted as proto -N 1526 */ 1527 1528 /* if we're overly short, let UDP handle it */ 1529 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1530 if (skb->len > sizeof(struct udphdr) && encap_rcv) { 1531 int ret; 1532 1533 /* Verify checksum before giving to encap */ 1534 if (udp_lib_checksum_complete(skb)) 1535 goto csum_error; 1536 1537 ret = encap_rcv(sk, skb); 1538 if (ret <= 0) { 1539 UDP_INC_STATS_BH(sock_net(sk), 1540 UDP_MIB_INDATAGRAMS, 1541 is_udplite); 1542 return -ret; 1543 } 1544 } 1545 1546 /* FALLTHROUGH -- it's a UDP Packet */ 1547 } 1548 1549 /* 1550 * UDP-Lite specific tests, ignored on UDP sockets 1551 */ 1552 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1553 1554 /* 1555 * MIB statistics other than incrementing the error count are 1556 * disabled for the following two types of errors: these depend 1557 * on the application settings, not on the functioning of the 1558 * protocol stack as such. 1559 * 1560 * RFC 3828 here recommends (sec 3.3): "There should also be a 1561 * way ... to ... at least let the receiving application block 1562 * delivery of packets with coverage values less than a value 1563 * provided by the application." 1564 */ 1565 if (up->pcrlen == 0) { /* full coverage was set */ 1566 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1567 UDP_SKB_CB(skb)->cscov, skb->len); 1568 goto drop; 1569 } 1570 /* The next case involves violating the min. coverage requested 1571 * by the receiver. This is subtle: if receiver wants x and x is 1572 * greater than the buffersize/MTU then receiver will complain 1573 * that it wants x while sender emits packets of smaller size y. 1574 * Therefore the above ...()->partial_cov statement is essential. 1575 */ 1576 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1577 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1578 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1579 goto drop; 1580 } 1581 } 1582 1583 if (rcu_access_pointer(sk->sk_filter) && 1584 udp_lib_checksum_complete(skb)) 1585 goto csum_error; 1586 1587 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1588 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1589 is_udplite); 1590 goto drop; 1591 } 1592 1593 rc = 0; 1594 1595 ipv4_pktinfo_prepare(sk, skb); 1596 bh_lock_sock(sk); 1597 if (!sock_owned_by_user(sk)) 1598 rc = __udp_queue_rcv_skb(sk, skb); 1599 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1600 bh_unlock_sock(sk); 1601 goto drop; 1602 } 1603 bh_unlock_sock(sk); 1604 1605 return rc; 1606 1607 csum_error: 1608 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1609 drop: 1610 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1611 atomic_inc(&sk->sk_drops); 1612 kfree_skb(skb); 1613 return -1; 1614 } 1615 1616 static void flush_stack(struct sock **stack, unsigned int count, 1617 struct sk_buff *skb, unsigned int final) 1618 { 1619 unsigned int i; 1620 struct sk_buff *skb1 = NULL; 1621 struct sock *sk; 1622 1623 for (i = 0; i < count; i++) { 1624 sk = stack[i]; 1625 if (likely(!skb1)) 1626 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1627 1628 if (!skb1) { 1629 atomic_inc(&sk->sk_drops); 1630 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1631 IS_UDPLITE(sk)); 1632 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1633 IS_UDPLITE(sk)); 1634 } 1635 1636 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1637 skb1 = NULL; 1638 1639 sock_put(sk); 1640 } 1641 if (unlikely(skb1)) 1642 kfree_skb(skb1); 1643 } 1644 1645 /* For TCP sockets, sk_rx_dst is protected by socket lock 1646 * For UDP, we use xchg() to guard against concurrent changes. 1647 */ 1648 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1649 { 1650 struct dst_entry *old; 1651 1652 dst_hold(dst); 1653 old = xchg(&sk->sk_rx_dst, dst); 1654 dst_release(old); 1655 } 1656 1657 /* 1658 * Multicasts and broadcasts go to each listener. 1659 * 1660 * Note: called only from the BH handler context. 1661 */ 1662 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1663 struct udphdr *uh, 1664 __be32 saddr, __be32 daddr, 1665 struct udp_table *udptable, 1666 int proto) 1667 { 1668 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1669 struct hlist_nulls_node *node; 1670 unsigned short hnum = ntohs(uh->dest); 1671 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1672 int dif = skb->dev->ifindex; 1673 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 1674 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1675 bool inner_flushed = false; 1676 1677 if (use_hash2) { 1678 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1679 udp_table.mask; 1680 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask; 1681 start_lookup: 1682 hslot = &udp_table.hash2[hash2]; 1683 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1684 } 1685 1686 spin_lock(&hslot->lock); 1687 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 1688 if (__udp_is_mcast_sock(net, sk, 1689 uh->dest, daddr, 1690 uh->source, saddr, 1691 dif, hnum)) { 1692 if (unlikely(count == ARRAY_SIZE(stack))) { 1693 flush_stack(stack, count, skb, ~0); 1694 inner_flushed = true; 1695 count = 0; 1696 } 1697 stack[count++] = sk; 1698 sock_hold(sk); 1699 } 1700 } 1701 1702 spin_unlock(&hslot->lock); 1703 1704 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1705 if (use_hash2 && hash2 != hash2_any) { 1706 hash2 = hash2_any; 1707 goto start_lookup; 1708 } 1709 1710 /* 1711 * do the slow work with no lock held 1712 */ 1713 if (count) { 1714 flush_stack(stack, count, skb, count - 1); 1715 } else { 1716 if (!inner_flushed) 1717 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 1718 proto == IPPROTO_UDPLITE); 1719 consume_skb(skb); 1720 } 1721 return 0; 1722 } 1723 1724 /* Initialize UDP checksum. If exited with zero value (success), 1725 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1726 * Otherwise, csum completion requires chacksumming packet body, 1727 * including udp header and folding it to skb->csum. 1728 */ 1729 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1730 int proto) 1731 { 1732 int err; 1733 1734 UDP_SKB_CB(skb)->partial_cov = 0; 1735 UDP_SKB_CB(skb)->cscov = skb->len; 1736 1737 if (proto == IPPROTO_UDPLITE) { 1738 err = udplite_checksum_init(skb, uh); 1739 if (err) 1740 return err; 1741 } 1742 1743 return skb_checksum_init_zero_check(skb, proto, uh->check, 1744 inet_compute_pseudo); 1745 } 1746 1747 /* 1748 * All we need to do is get the socket, and then do a checksum. 1749 */ 1750 1751 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1752 int proto) 1753 { 1754 struct sock *sk; 1755 struct udphdr *uh; 1756 unsigned short ulen; 1757 struct rtable *rt = skb_rtable(skb); 1758 __be32 saddr, daddr; 1759 struct net *net = dev_net(skb->dev); 1760 1761 /* 1762 * Validate the packet. 1763 */ 1764 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1765 goto drop; /* No space for header. */ 1766 1767 uh = udp_hdr(skb); 1768 ulen = ntohs(uh->len); 1769 saddr = ip_hdr(skb)->saddr; 1770 daddr = ip_hdr(skb)->daddr; 1771 1772 if (ulen > skb->len) 1773 goto short_packet; 1774 1775 if (proto == IPPROTO_UDP) { 1776 /* UDP validates ulen. */ 1777 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1778 goto short_packet; 1779 uh = udp_hdr(skb); 1780 } 1781 1782 if (udp4_csum_init(skb, uh, proto)) 1783 goto csum_error; 1784 1785 sk = skb_steal_sock(skb); 1786 if (sk) { 1787 struct dst_entry *dst = skb_dst(skb); 1788 int ret; 1789 1790 if (unlikely(sk->sk_rx_dst != dst)) 1791 udp_sk_rx_dst_set(sk, dst); 1792 1793 ret = udp_queue_rcv_skb(sk, skb); 1794 sock_put(sk); 1795 /* a return value > 0 means to resubmit the input, but 1796 * it wants the return to be -protocol, or 0 1797 */ 1798 if (ret > 0) 1799 return -ret; 1800 return 0; 1801 } 1802 1803 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1804 return __udp4_lib_mcast_deliver(net, skb, uh, 1805 saddr, daddr, udptable, proto); 1806 1807 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1808 if (sk) { 1809 int ret; 1810 1811 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1812 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1813 inet_compute_pseudo); 1814 1815 ret = udp_queue_rcv_skb(sk, skb); 1816 sock_put(sk); 1817 1818 /* a return value > 0 means to resubmit the input, but 1819 * it wants the return to be -protocol, or 0 1820 */ 1821 if (ret > 0) 1822 return -ret; 1823 return 0; 1824 } 1825 1826 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1827 goto drop; 1828 nf_reset(skb); 1829 1830 /* No socket. Drop packet silently, if checksum is wrong */ 1831 if (udp_lib_checksum_complete(skb)) 1832 goto csum_error; 1833 1834 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1835 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1836 1837 /* 1838 * Hmm. We got an UDP packet to a port to which we 1839 * don't wanna listen. Ignore it. 1840 */ 1841 kfree_skb(skb); 1842 return 0; 1843 1844 short_packet: 1845 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1846 proto == IPPROTO_UDPLITE ? "Lite" : "", 1847 &saddr, ntohs(uh->source), 1848 ulen, skb->len, 1849 &daddr, ntohs(uh->dest)); 1850 goto drop; 1851 1852 csum_error: 1853 /* 1854 * RFC1122: OK. Discards the bad packet silently (as far as 1855 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1856 */ 1857 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1858 proto == IPPROTO_UDPLITE ? "Lite" : "", 1859 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1860 ulen); 1861 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1862 drop: 1863 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1864 kfree_skb(skb); 1865 return 0; 1866 } 1867 1868 /* We can only early demux multicast if there is a single matching socket. 1869 * If more than one socket found returns NULL 1870 */ 1871 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1872 __be16 loc_port, __be32 loc_addr, 1873 __be16 rmt_port, __be32 rmt_addr, 1874 int dif) 1875 { 1876 struct sock *sk, *result; 1877 struct hlist_nulls_node *node; 1878 unsigned short hnum = ntohs(loc_port); 1879 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1880 struct udp_hslot *hslot = &udp_table.hash[slot]; 1881 1882 /* Do not bother scanning a too big list */ 1883 if (hslot->count > 10) 1884 return NULL; 1885 1886 rcu_read_lock(); 1887 begin: 1888 count = 0; 1889 result = NULL; 1890 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1891 if (__udp_is_mcast_sock(net, sk, 1892 loc_port, loc_addr, 1893 rmt_port, rmt_addr, 1894 dif, hnum)) { 1895 result = sk; 1896 ++count; 1897 } 1898 } 1899 /* 1900 * if the nulls value we got at the end of this lookup is 1901 * not the expected one, we must restart lookup. 1902 * We probably met an item that was moved to another chain. 1903 */ 1904 if (get_nulls_value(node) != slot) 1905 goto begin; 1906 1907 if (result) { 1908 if (count != 1 || 1909 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1910 result = NULL; 1911 else if (unlikely(!__udp_is_mcast_sock(net, result, 1912 loc_port, loc_addr, 1913 rmt_port, rmt_addr, 1914 dif, hnum))) { 1915 sock_put(result); 1916 result = NULL; 1917 } 1918 } 1919 rcu_read_unlock(); 1920 return result; 1921 } 1922 1923 /* For unicast we should only early demux connected sockets or we can 1924 * break forwarding setups. The chains here can be long so only check 1925 * if the first socket is an exact match and if not move on. 1926 */ 1927 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1928 __be16 loc_port, __be32 loc_addr, 1929 __be16 rmt_port, __be32 rmt_addr, 1930 int dif) 1931 { 1932 struct sock *sk, *result; 1933 struct hlist_nulls_node *node; 1934 unsigned short hnum = ntohs(loc_port); 1935 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1936 unsigned int slot2 = hash2 & udp_table.mask; 1937 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1938 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 1939 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1940 1941 rcu_read_lock(); 1942 result = NULL; 1943 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 1944 if (INET_MATCH(sk, net, acookie, 1945 rmt_addr, loc_addr, ports, dif)) 1946 result = sk; 1947 /* Only check first socket in chain */ 1948 break; 1949 } 1950 1951 if (result) { 1952 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1953 result = NULL; 1954 else if (unlikely(!INET_MATCH(sk, net, acookie, 1955 rmt_addr, loc_addr, 1956 ports, dif))) { 1957 sock_put(result); 1958 result = NULL; 1959 } 1960 } 1961 rcu_read_unlock(); 1962 return result; 1963 } 1964 1965 void udp_v4_early_demux(struct sk_buff *skb) 1966 { 1967 struct net *net = dev_net(skb->dev); 1968 const struct iphdr *iph; 1969 const struct udphdr *uh; 1970 struct sock *sk; 1971 struct dst_entry *dst; 1972 int dif = skb->dev->ifindex; 1973 int ours; 1974 1975 /* validate the packet */ 1976 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1977 return; 1978 1979 iph = ip_hdr(skb); 1980 uh = udp_hdr(skb); 1981 1982 if (skb->pkt_type == PACKET_BROADCAST || 1983 skb->pkt_type == PACKET_MULTICAST) { 1984 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 1985 1986 if (!in_dev) 1987 return; 1988 1989 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 1990 iph->protocol); 1991 if (!ours) 1992 return; 1993 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1994 uh->source, iph->saddr, dif); 1995 } else if (skb->pkt_type == PACKET_HOST) { 1996 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1997 uh->source, iph->saddr, dif); 1998 } else { 1999 return; 2000 } 2001 2002 if (!sk) 2003 return; 2004 2005 skb->sk = sk; 2006 skb->destructor = sock_efree; 2007 dst = READ_ONCE(sk->sk_rx_dst); 2008 2009 if (dst) 2010 dst = dst_check(dst, 0); 2011 if (dst) { 2012 /* DST_NOCACHE can not be used without taking a reference */ 2013 if (dst->flags & DST_NOCACHE) { 2014 if (likely(atomic_inc_not_zero(&dst->__refcnt))) 2015 skb_dst_set(skb, dst); 2016 } else { 2017 skb_dst_set_noref(skb, dst); 2018 } 2019 } 2020 } 2021 2022 int udp_rcv(struct sk_buff *skb) 2023 { 2024 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2025 } 2026 2027 void udp_destroy_sock(struct sock *sk) 2028 { 2029 struct udp_sock *up = udp_sk(sk); 2030 bool slow = lock_sock_fast(sk); 2031 udp_flush_pending_frames(sk); 2032 unlock_sock_fast(sk, slow); 2033 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2034 void (*encap_destroy)(struct sock *sk); 2035 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2036 if (encap_destroy) 2037 encap_destroy(sk); 2038 } 2039 } 2040 2041 /* 2042 * Socket option code for UDP 2043 */ 2044 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2045 char __user *optval, unsigned int optlen, 2046 int (*push_pending_frames)(struct sock *)) 2047 { 2048 struct udp_sock *up = udp_sk(sk); 2049 int val, valbool; 2050 int err = 0; 2051 int is_udplite = IS_UDPLITE(sk); 2052 2053 if (optlen < sizeof(int)) 2054 return -EINVAL; 2055 2056 if (get_user(val, (int __user *)optval)) 2057 return -EFAULT; 2058 2059 valbool = val ? 1 : 0; 2060 2061 switch (optname) { 2062 case UDP_CORK: 2063 if (val != 0) { 2064 up->corkflag = 1; 2065 } else { 2066 up->corkflag = 0; 2067 lock_sock(sk); 2068 push_pending_frames(sk); 2069 release_sock(sk); 2070 } 2071 break; 2072 2073 case UDP_ENCAP: 2074 switch (val) { 2075 case 0: 2076 case UDP_ENCAP_ESPINUDP: 2077 case UDP_ENCAP_ESPINUDP_NON_IKE: 2078 up->encap_rcv = xfrm4_udp_encap_rcv; 2079 /* FALLTHROUGH */ 2080 case UDP_ENCAP_L2TPINUDP: 2081 up->encap_type = val; 2082 udp_encap_enable(); 2083 break; 2084 default: 2085 err = -ENOPROTOOPT; 2086 break; 2087 } 2088 break; 2089 2090 case UDP_NO_CHECK6_TX: 2091 up->no_check6_tx = valbool; 2092 break; 2093 2094 case UDP_NO_CHECK6_RX: 2095 up->no_check6_rx = valbool; 2096 break; 2097 2098 /* 2099 * UDP-Lite's partial checksum coverage (RFC 3828). 2100 */ 2101 /* The sender sets actual checksum coverage length via this option. 2102 * The case coverage > packet length is handled by send module. */ 2103 case UDPLITE_SEND_CSCOV: 2104 if (!is_udplite) /* Disable the option on UDP sockets */ 2105 return -ENOPROTOOPT; 2106 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2107 val = 8; 2108 else if (val > USHRT_MAX) 2109 val = USHRT_MAX; 2110 up->pcslen = val; 2111 up->pcflag |= UDPLITE_SEND_CC; 2112 break; 2113 2114 /* The receiver specifies a minimum checksum coverage value. To make 2115 * sense, this should be set to at least 8 (as done below). If zero is 2116 * used, this again means full checksum coverage. */ 2117 case UDPLITE_RECV_CSCOV: 2118 if (!is_udplite) /* Disable the option on UDP sockets */ 2119 return -ENOPROTOOPT; 2120 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2121 val = 8; 2122 else if (val > USHRT_MAX) 2123 val = USHRT_MAX; 2124 up->pcrlen = val; 2125 up->pcflag |= UDPLITE_RECV_CC; 2126 break; 2127 2128 default: 2129 err = -ENOPROTOOPT; 2130 break; 2131 } 2132 2133 return err; 2134 } 2135 EXPORT_SYMBOL(udp_lib_setsockopt); 2136 2137 int udp_setsockopt(struct sock *sk, int level, int optname, 2138 char __user *optval, unsigned int optlen) 2139 { 2140 if (level == SOL_UDP || level == SOL_UDPLITE) 2141 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2142 udp_push_pending_frames); 2143 return ip_setsockopt(sk, level, optname, optval, optlen); 2144 } 2145 2146 #ifdef CONFIG_COMPAT 2147 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2148 char __user *optval, unsigned int optlen) 2149 { 2150 if (level == SOL_UDP || level == SOL_UDPLITE) 2151 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2152 udp_push_pending_frames); 2153 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2154 } 2155 #endif 2156 2157 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2158 char __user *optval, int __user *optlen) 2159 { 2160 struct udp_sock *up = udp_sk(sk); 2161 int val, len; 2162 2163 if (get_user(len, optlen)) 2164 return -EFAULT; 2165 2166 len = min_t(unsigned int, len, sizeof(int)); 2167 2168 if (len < 0) 2169 return -EINVAL; 2170 2171 switch (optname) { 2172 case UDP_CORK: 2173 val = up->corkflag; 2174 break; 2175 2176 case UDP_ENCAP: 2177 val = up->encap_type; 2178 break; 2179 2180 case UDP_NO_CHECK6_TX: 2181 val = up->no_check6_tx; 2182 break; 2183 2184 case UDP_NO_CHECK6_RX: 2185 val = up->no_check6_rx; 2186 break; 2187 2188 /* The following two cannot be changed on UDP sockets, the return is 2189 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2190 case UDPLITE_SEND_CSCOV: 2191 val = up->pcslen; 2192 break; 2193 2194 case UDPLITE_RECV_CSCOV: 2195 val = up->pcrlen; 2196 break; 2197 2198 default: 2199 return -ENOPROTOOPT; 2200 } 2201 2202 if (put_user(len, optlen)) 2203 return -EFAULT; 2204 if (copy_to_user(optval, &val, len)) 2205 return -EFAULT; 2206 return 0; 2207 } 2208 EXPORT_SYMBOL(udp_lib_getsockopt); 2209 2210 int udp_getsockopt(struct sock *sk, int level, int optname, 2211 char __user *optval, int __user *optlen) 2212 { 2213 if (level == SOL_UDP || level == SOL_UDPLITE) 2214 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2215 return ip_getsockopt(sk, level, optname, optval, optlen); 2216 } 2217 2218 #ifdef CONFIG_COMPAT 2219 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2220 char __user *optval, int __user *optlen) 2221 { 2222 if (level == SOL_UDP || level == SOL_UDPLITE) 2223 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2224 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2225 } 2226 #endif 2227 /** 2228 * udp_poll - wait for a UDP event. 2229 * @file - file struct 2230 * @sock - socket 2231 * @wait - poll table 2232 * 2233 * This is same as datagram poll, except for the special case of 2234 * blocking sockets. If application is using a blocking fd 2235 * and a packet with checksum error is in the queue; 2236 * then it could get return from select indicating data available 2237 * but then block when reading it. Add special case code 2238 * to work around these arguably broken applications. 2239 */ 2240 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2241 { 2242 unsigned int mask = datagram_poll(file, sock, wait); 2243 struct sock *sk = sock->sk; 2244 2245 sock_rps_record_flow(sk); 2246 2247 /* Check for false positives due to checksum errors */ 2248 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2249 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2250 mask &= ~(POLLIN | POLLRDNORM); 2251 2252 return mask; 2253 2254 } 2255 EXPORT_SYMBOL(udp_poll); 2256 2257 struct proto udp_prot = { 2258 .name = "UDP", 2259 .owner = THIS_MODULE, 2260 .close = udp_lib_close, 2261 .connect = ip4_datagram_connect, 2262 .disconnect = udp_disconnect, 2263 .ioctl = udp_ioctl, 2264 .destroy = udp_destroy_sock, 2265 .setsockopt = udp_setsockopt, 2266 .getsockopt = udp_getsockopt, 2267 .sendmsg = udp_sendmsg, 2268 .recvmsg = udp_recvmsg, 2269 .sendpage = udp_sendpage, 2270 .backlog_rcv = __udp_queue_rcv_skb, 2271 .release_cb = ip4_datagram_release_cb, 2272 .hash = udp_lib_hash, 2273 .unhash = udp_lib_unhash, 2274 .rehash = udp_v4_rehash, 2275 .get_port = udp_v4_get_port, 2276 .memory_allocated = &udp_memory_allocated, 2277 .sysctl_mem = sysctl_udp_mem, 2278 .sysctl_wmem = &sysctl_udp_wmem_min, 2279 .sysctl_rmem = &sysctl_udp_rmem_min, 2280 .obj_size = sizeof(struct udp_sock), 2281 .slab_flags = SLAB_DESTROY_BY_RCU, 2282 .h.udp_table = &udp_table, 2283 #ifdef CONFIG_COMPAT 2284 .compat_setsockopt = compat_udp_setsockopt, 2285 .compat_getsockopt = compat_udp_getsockopt, 2286 #endif 2287 .clear_sk = sk_prot_clear_portaddr_nulls, 2288 }; 2289 EXPORT_SYMBOL(udp_prot); 2290 2291 /* ------------------------------------------------------------------------ */ 2292 #ifdef CONFIG_PROC_FS 2293 2294 static struct sock *udp_get_first(struct seq_file *seq, int start) 2295 { 2296 struct sock *sk; 2297 struct udp_iter_state *state = seq->private; 2298 struct net *net = seq_file_net(seq); 2299 2300 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2301 ++state->bucket) { 2302 struct hlist_nulls_node *node; 2303 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2304 2305 if (hlist_nulls_empty(&hslot->head)) 2306 continue; 2307 2308 spin_lock_bh(&hslot->lock); 2309 sk_nulls_for_each(sk, node, &hslot->head) { 2310 if (!net_eq(sock_net(sk), net)) 2311 continue; 2312 if (sk->sk_family == state->family) 2313 goto found; 2314 } 2315 spin_unlock_bh(&hslot->lock); 2316 } 2317 sk = NULL; 2318 found: 2319 return sk; 2320 } 2321 2322 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2323 { 2324 struct udp_iter_state *state = seq->private; 2325 struct net *net = seq_file_net(seq); 2326 2327 do { 2328 sk = sk_nulls_next(sk); 2329 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2330 2331 if (!sk) { 2332 if (state->bucket <= state->udp_table->mask) 2333 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2334 return udp_get_first(seq, state->bucket + 1); 2335 } 2336 return sk; 2337 } 2338 2339 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2340 { 2341 struct sock *sk = udp_get_first(seq, 0); 2342 2343 if (sk) 2344 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2345 --pos; 2346 return pos ? NULL : sk; 2347 } 2348 2349 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2350 { 2351 struct udp_iter_state *state = seq->private; 2352 state->bucket = MAX_UDP_PORTS; 2353 2354 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2355 } 2356 2357 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2358 { 2359 struct sock *sk; 2360 2361 if (v == SEQ_START_TOKEN) 2362 sk = udp_get_idx(seq, 0); 2363 else 2364 sk = udp_get_next(seq, v); 2365 2366 ++*pos; 2367 return sk; 2368 } 2369 2370 static void udp_seq_stop(struct seq_file *seq, void *v) 2371 { 2372 struct udp_iter_state *state = seq->private; 2373 2374 if (state->bucket <= state->udp_table->mask) 2375 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2376 } 2377 2378 int udp_seq_open(struct inode *inode, struct file *file) 2379 { 2380 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2381 struct udp_iter_state *s; 2382 int err; 2383 2384 err = seq_open_net(inode, file, &afinfo->seq_ops, 2385 sizeof(struct udp_iter_state)); 2386 if (err < 0) 2387 return err; 2388 2389 s = ((struct seq_file *)file->private_data)->private; 2390 s->family = afinfo->family; 2391 s->udp_table = afinfo->udp_table; 2392 return err; 2393 } 2394 EXPORT_SYMBOL(udp_seq_open); 2395 2396 /* ------------------------------------------------------------------------ */ 2397 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2398 { 2399 struct proc_dir_entry *p; 2400 int rc = 0; 2401 2402 afinfo->seq_ops.start = udp_seq_start; 2403 afinfo->seq_ops.next = udp_seq_next; 2404 afinfo->seq_ops.stop = udp_seq_stop; 2405 2406 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2407 afinfo->seq_fops, afinfo); 2408 if (!p) 2409 rc = -ENOMEM; 2410 return rc; 2411 } 2412 EXPORT_SYMBOL(udp_proc_register); 2413 2414 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2415 { 2416 remove_proc_entry(afinfo->name, net->proc_net); 2417 } 2418 EXPORT_SYMBOL(udp_proc_unregister); 2419 2420 /* ------------------------------------------------------------------------ */ 2421 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2422 int bucket) 2423 { 2424 struct inet_sock *inet = inet_sk(sp); 2425 __be32 dest = inet->inet_daddr; 2426 __be32 src = inet->inet_rcv_saddr; 2427 __u16 destp = ntohs(inet->inet_dport); 2428 __u16 srcp = ntohs(inet->inet_sport); 2429 2430 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2431 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2432 bucket, src, srcp, dest, destp, sp->sk_state, 2433 sk_wmem_alloc_get(sp), 2434 sk_rmem_alloc_get(sp), 2435 0, 0L, 0, 2436 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2437 0, sock_i_ino(sp), 2438 atomic_read(&sp->sk_refcnt), sp, 2439 atomic_read(&sp->sk_drops)); 2440 } 2441 2442 int udp4_seq_show(struct seq_file *seq, void *v) 2443 { 2444 seq_setwidth(seq, 127); 2445 if (v == SEQ_START_TOKEN) 2446 seq_puts(seq, " sl local_address rem_address st tx_queue " 2447 "rx_queue tr tm->when retrnsmt uid timeout " 2448 "inode ref pointer drops"); 2449 else { 2450 struct udp_iter_state *state = seq->private; 2451 2452 udp4_format_sock(v, seq, state->bucket); 2453 } 2454 seq_pad(seq, '\n'); 2455 return 0; 2456 } 2457 2458 static const struct file_operations udp_afinfo_seq_fops = { 2459 .owner = THIS_MODULE, 2460 .open = udp_seq_open, 2461 .read = seq_read, 2462 .llseek = seq_lseek, 2463 .release = seq_release_net 2464 }; 2465 2466 /* ------------------------------------------------------------------------ */ 2467 static struct udp_seq_afinfo udp4_seq_afinfo = { 2468 .name = "udp", 2469 .family = AF_INET, 2470 .udp_table = &udp_table, 2471 .seq_fops = &udp_afinfo_seq_fops, 2472 .seq_ops = { 2473 .show = udp4_seq_show, 2474 }, 2475 }; 2476 2477 static int __net_init udp4_proc_init_net(struct net *net) 2478 { 2479 return udp_proc_register(net, &udp4_seq_afinfo); 2480 } 2481 2482 static void __net_exit udp4_proc_exit_net(struct net *net) 2483 { 2484 udp_proc_unregister(net, &udp4_seq_afinfo); 2485 } 2486 2487 static struct pernet_operations udp4_net_ops = { 2488 .init = udp4_proc_init_net, 2489 .exit = udp4_proc_exit_net, 2490 }; 2491 2492 int __init udp4_proc_init(void) 2493 { 2494 return register_pernet_subsys(&udp4_net_ops); 2495 } 2496 2497 void udp4_proc_exit(void) 2498 { 2499 unregister_pernet_subsys(&udp4_net_ops); 2500 } 2501 #endif /* CONFIG_PROC_FS */ 2502 2503 static __initdata unsigned long uhash_entries; 2504 static int __init set_uhash_entries(char *str) 2505 { 2506 ssize_t ret; 2507 2508 if (!str) 2509 return 0; 2510 2511 ret = kstrtoul(str, 0, &uhash_entries); 2512 if (ret) 2513 return 0; 2514 2515 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2516 uhash_entries = UDP_HTABLE_SIZE_MIN; 2517 return 1; 2518 } 2519 __setup("uhash_entries=", set_uhash_entries); 2520 2521 void __init udp_table_init(struct udp_table *table, const char *name) 2522 { 2523 unsigned int i; 2524 2525 table->hash = alloc_large_system_hash(name, 2526 2 * sizeof(struct udp_hslot), 2527 uhash_entries, 2528 21, /* one slot per 2 MB */ 2529 0, 2530 &table->log, 2531 &table->mask, 2532 UDP_HTABLE_SIZE_MIN, 2533 64 * 1024); 2534 2535 table->hash2 = table->hash + (table->mask + 1); 2536 for (i = 0; i <= table->mask; i++) { 2537 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2538 table->hash[i].count = 0; 2539 spin_lock_init(&table->hash[i].lock); 2540 } 2541 for (i = 0; i <= table->mask; i++) { 2542 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2543 table->hash2[i].count = 0; 2544 spin_lock_init(&table->hash2[i].lock); 2545 } 2546 } 2547 2548 u32 udp_flow_hashrnd(void) 2549 { 2550 static u32 hashrnd __read_mostly; 2551 2552 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2553 2554 return hashrnd; 2555 } 2556 EXPORT_SYMBOL(udp_flow_hashrnd); 2557 2558 void __init udp_init(void) 2559 { 2560 unsigned long limit; 2561 2562 udp_table_init(&udp_table, "UDP"); 2563 limit = nr_free_buffer_pages() / 8; 2564 limit = max(limit, 128UL); 2565 sysctl_udp_mem[0] = limit / 4 * 3; 2566 sysctl_udp_mem[1] = limit; 2567 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2568 2569 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2570 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2571 } 2572