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/proc_fs.h> 104 #include <linux/seq_file.h> 105 #include <net/net_namespace.h> 106 #include <net/icmp.h> 107 #include <net/inet_hashtables.h> 108 #include <net/route.h> 109 #include <net/checksum.h> 110 #include <net/xfrm.h> 111 #include <trace/events/udp.h> 112 #include <linux/static_key.h> 113 #include <trace/events/skb.h> 114 #include <net/busy_poll.h> 115 #include "udp_impl.h" 116 117 struct udp_table udp_table __read_mostly; 118 EXPORT_SYMBOL(udp_table); 119 120 long sysctl_udp_mem[3] __read_mostly; 121 EXPORT_SYMBOL(sysctl_udp_mem); 122 123 int sysctl_udp_rmem_min __read_mostly; 124 EXPORT_SYMBOL(sysctl_udp_rmem_min); 125 126 int sysctl_udp_wmem_min __read_mostly; 127 EXPORT_SYMBOL(sysctl_udp_wmem_min); 128 129 atomic_long_t udp_memory_allocated; 130 EXPORT_SYMBOL(udp_memory_allocated); 131 132 #define MAX_UDP_PORTS 65536 133 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 134 135 static int udp_lib_lport_inuse(struct net *net, __u16 num, 136 const struct udp_hslot *hslot, 137 unsigned long *bitmap, 138 struct sock *sk, 139 int (*saddr_comp)(const struct sock *sk1, 140 const struct sock *sk2), 141 unsigned int log) 142 { 143 struct sock *sk2; 144 struct hlist_nulls_node *node; 145 kuid_t uid = sock_i_uid(sk); 146 147 sk_nulls_for_each(sk2, node, &hslot->head) { 148 if (net_eq(sock_net(sk2), net) && 149 sk2 != sk && 150 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 151 (!sk2->sk_reuse || !sk->sk_reuse) && 152 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 153 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 154 (!sk2->sk_reuseport || !sk->sk_reuseport || 155 !uid_eq(uid, sock_i_uid(sk2))) && 156 saddr_comp(sk, sk2)) { 157 if (!bitmap) 158 return 1; 159 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap); 160 } 161 } 162 return 0; 163 } 164 165 /* 166 * Note: we still hold spinlock of primary hash chain, so no other writer 167 * can insert/delete a socket with local_port == num 168 */ 169 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 170 struct udp_hslot *hslot2, 171 struct sock *sk, 172 int (*saddr_comp)(const struct sock *sk1, 173 const struct sock *sk2)) 174 { 175 struct sock *sk2; 176 struct hlist_nulls_node *node; 177 kuid_t uid = sock_i_uid(sk); 178 int res = 0; 179 180 spin_lock(&hslot2->lock); 181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) { 182 if (net_eq(sock_net(sk2), net) && 183 sk2 != sk && 184 (udp_sk(sk2)->udp_port_hash == num) && 185 (!sk2->sk_reuse || !sk->sk_reuse) && 186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 188 (!sk2->sk_reuseport || !sk->sk_reuseport || 189 !uid_eq(uid, sock_i_uid(sk2))) && 190 saddr_comp(sk, sk2)) { 191 res = 1; 192 break; 193 } 194 } 195 spin_unlock(&hslot2->lock); 196 return res; 197 } 198 199 /** 200 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 201 * 202 * @sk: socket struct in question 203 * @snum: port number to look up 204 * @saddr_comp: AF-dependent comparison of bound local IP addresses 205 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 206 * with NULL address 207 */ 208 int udp_lib_get_port(struct sock *sk, unsigned short snum, 209 int (*saddr_comp)(const struct sock *sk1, 210 const struct sock *sk2), 211 unsigned int hash2_nulladdr) 212 { 213 struct udp_hslot *hslot, *hslot2; 214 struct udp_table *udptable = sk->sk_prot->h.udp_table; 215 int error = 1; 216 struct net *net = sock_net(sk); 217 218 if (!snum) { 219 int low, high, remaining; 220 unsigned int rand; 221 unsigned short first, last; 222 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 223 224 inet_get_local_port_range(net, &low, &high); 225 remaining = (high - low) + 1; 226 227 rand = prandom_u32(); 228 first = reciprocal_scale(rand, remaining) + low; 229 /* 230 * force rand to be an odd multiple of UDP_HTABLE_SIZE 231 */ 232 rand = (rand | 1) * (udptable->mask + 1); 233 last = first + udptable->mask + 1; 234 do { 235 hslot = udp_hashslot(udptable, net, first); 236 bitmap_zero(bitmap, PORTS_PER_CHAIN); 237 spin_lock_bh(&hslot->lock); 238 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 239 saddr_comp, udptable->log); 240 241 snum = first; 242 /* 243 * Iterate on all possible values of snum for this hash. 244 * Using steps of an odd multiple of UDP_HTABLE_SIZE 245 * give us randomization and full range coverage. 246 */ 247 do { 248 if (low <= snum && snum <= high && 249 !test_bit(snum >> udptable->log, bitmap) && 250 !inet_is_local_reserved_port(net, snum)) 251 goto found; 252 snum += rand; 253 } while (snum != first); 254 spin_unlock_bh(&hslot->lock); 255 } while (++first != last); 256 goto fail; 257 } else { 258 hslot = udp_hashslot(udptable, net, snum); 259 spin_lock_bh(&hslot->lock); 260 if (hslot->count > 10) { 261 int exist; 262 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 263 264 slot2 &= udptable->mask; 265 hash2_nulladdr &= udptable->mask; 266 267 hslot2 = udp_hashslot2(udptable, slot2); 268 if (hslot->count < hslot2->count) 269 goto scan_primary_hash; 270 271 exist = udp_lib_lport_inuse2(net, snum, hslot2, 272 sk, saddr_comp); 273 if (!exist && (hash2_nulladdr != slot2)) { 274 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 275 exist = udp_lib_lport_inuse2(net, snum, hslot2, 276 sk, saddr_comp); 277 } 278 if (exist) 279 goto fail_unlock; 280 else 281 goto found; 282 } 283 scan_primary_hash: 284 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 285 saddr_comp, 0)) 286 goto fail_unlock; 287 } 288 found: 289 inet_sk(sk)->inet_num = snum; 290 udp_sk(sk)->udp_port_hash = snum; 291 udp_sk(sk)->udp_portaddr_hash ^= snum; 292 if (sk_unhashed(sk)) { 293 sk_nulls_add_node_rcu(sk, &hslot->head); 294 hslot->count++; 295 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 296 297 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 298 spin_lock(&hslot2->lock); 299 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 300 &hslot2->head); 301 hslot2->count++; 302 spin_unlock(&hslot2->lock); 303 } 304 error = 0; 305 fail_unlock: 306 spin_unlock_bh(&hslot->lock); 307 fail: 308 return error; 309 } 310 EXPORT_SYMBOL(udp_lib_get_port); 311 312 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 313 { 314 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 315 316 return (!ipv6_only_sock(sk2) && 317 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 318 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 319 } 320 321 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, 322 unsigned int port) 323 { 324 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 325 } 326 327 int udp_v4_get_port(struct sock *sk, unsigned short snum) 328 { 329 unsigned int hash2_nulladdr = 330 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 331 unsigned int hash2_partial = 332 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 333 334 /* precompute partial secondary hash */ 335 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 336 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 337 } 338 339 static inline int compute_score(struct sock *sk, struct net *net, 340 __be32 saddr, unsigned short hnum, __be16 sport, 341 __be32 daddr, __be16 dport, int dif) 342 { 343 int score; 344 struct inet_sock *inet; 345 346 if (!net_eq(sock_net(sk), net) || 347 udp_sk(sk)->udp_port_hash != hnum || 348 ipv6_only_sock(sk)) 349 return -1; 350 351 score = (sk->sk_family == PF_INET) ? 2 : 1; 352 inet = inet_sk(sk); 353 354 if (inet->inet_rcv_saddr) { 355 if (inet->inet_rcv_saddr != daddr) 356 return -1; 357 score += 4; 358 } 359 360 if (inet->inet_daddr) { 361 if (inet->inet_daddr != saddr) 362 return -1; 363 score += 4; 364 } 365 366 if (inet->inet_dport) { 367 if (inet->inet_dport != sport) 368 return -1; 369 score += 4; 370 } 371 372 if (sk->sk_bound_dev_if) { 373 if (sk->sk_bound_dev_if != dif) 374 return -1; 375 score += 4; 376 } 377 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 378 score++; 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 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 423 score++; 424 425 return score; 426 } 427 428 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 429 const __u16 lport, const __be32 faddr, 430 const __be16 fport) 431 { 432 static u32 udp_ehash_secret __read_mostly; 433 434 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 435 436 return __inet_ehashfn(laddr, lport, faddr, fport, 437 udp_ehash_secret + net_hash_mix(net)); 438 } 439 440 /* called with read_rcu_lock() */ 441 static struct sock *udp4_lib_lookup2(struct net *net, 442 __be32 saddr, __be16 sport, 443 __be32 daddr, unsigned int hnum, int dif, 444 struct udp_hslot *hslot2, unsigned int slot2) 445 { 446 struct sock *sk, *result; 447 struct hlist_nulls_node *node; 448 int score, badness, matches = 0, reuseport = 0; 449 u32 hash = 0; 450 451 begin: 452 result = NULL; 453 badness = 0; 454 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 455 score = compute_score2(sk, net, saddr, sport, 456 daddr, hnum, dif); 457 if (score > badness) { 458 result = sk; 459 badness = score; 460 reuseport = sk->sk_reuseport; 461 if (reuseport) { 462 hash = udp_ehashfn(net, daddr, hnum, 463 saddr, sport); 464 matches = 1; 465 } 466 } else if (score == badness && reuseport) { 467 matches++; 468 if (reciprocal_scale(hash, matches) == 0) 469 result = sk; 470 hash = next_pseudo_random32(hash); 471 } 472 } 473 /* 474 * if the nulls value we got at the end of this lookup is 475 * not the expected one, we must restart lookup. 476 * We probably met an item that was moved to another chain. 477 */ 478 if (get_nulls_value(node) != slot2) 479 goto begin; 480 if (result) { 481 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 482 result = NULL; 483 else if (unlikely(compute_score2(result, net, saddr, sport, 484 daddr, hnum, dif) < badness)) { 485 sock_put(result); 486 goto begin; 487 } 488 } 489 return result; 490 } 491 492 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 493 * harder than this. -DaveM 494 */ 495 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 496 __be16 sport, __be32 daddr, __be16 dport, 497 int dif, struct udp_table *udptable) 498 { 499 struct sock *sk, *result; 500 struct hlist_nulls_node *node; 501 unsigned short hnum = ntohs(dport); 502 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 503 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 504 int score, badness, matches = 0, reuseport = 0; 505 u32 hash = 0; 506 507 rcu_read_lock(); 508 if (hslot->count > 10) { 509 hash2 = udp4_portaddr_hash(net, daddr, hnum); 510 slot2 = hash2 & udptable->mask; 511 hslot2 = &udptable->hash2[slot2]; 512 if (hslot->count < hslot2->count) 513 goto begin; 514 515 result = udp4_lib_lookup2(net, saddr, sport, 516 daddr, hnum, dif, 517 hslot2, slot2); 518 if (!result) { 519 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 520 slot2 = hash2 & udptable->mask; 521 hslot2 = &udptable->hash2[slot2]; 522 if (hslot->count < hslot2->count) 523 goto begin; 524 525 result = udp4_lib_lookup2(net, saddr, sport, 526 htonl(INADDR_ANY), hnum, dif, 527 hslot2, slot2); 528 } 529 rcu_read_unlock(); 530 return result; 531 } 532 begin: 533 result = NULL; 534 badness = 0; 535 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 536 score = compute_score(sk, net, saddr, hnum, sport, 537 daddr, dport, dif); 538 if (score > badness) { 539 result = sk; 540 badness = score; 541 reuseport = sk->sk_reuseport; 542 if (reuseport) { 543 hash = udp_ehashfn(net, daddr, hnum, 544 saddr, sport); 545 matches = 1; 546 } 547 } else if (score == badness && reuseport) { 548 matches++; 549 if (reciprocal_scale(hash, matches) == 0) 550 result = sk; 551 hash = next_pseudo_random32(hash); 552 } 553 } 554 /* 555 * if the nulls value we got at the end of this lookup is 556 * not the expected one, we must restart lookup. 557 * We probably met an item that was moved to another chain. 558 */ 559 if (get_nulls_value(node) != slot) 560 goto begin; 561 562 if (result) { 563 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 564 result = NULL; 565 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 566 daddr, dport, dif) < badness)) { 567 sock_put(result); 568 goto begin; 569 } 570 } 571 rcu_read_unlock(); 572 return result; 573 } 574 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 575 576 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 577 __be16 sport, __be16 dport, 578 struct udp_table *udptable) 579 { 580 const struct iphdr *iph = ip_hdr(skb); 581 582 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 583 iph->daddr, dport, inet_iif(skb), 584 udptable); 585 } 586 587 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 588 __be32 daddr, __be16 dport, int dif) 589 { 590 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 591 } 592 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 593 594 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 595 __be16 loc_port, __be32 loc_addr, 596 __be16 rmt_port, __be32 rmt_addr, 597 int dif, unsigned short hnum) 598 { 599 struct inet_sock *inet = inet_sk(sk); 600 601 if (!net_eq(sock_net(sk), net) || 602 udp_sk(sk)->udp_port_hash != hnum || 603 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 604 (inet->inet_dport != rmt_port && inet->inet_dport) || 605 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 606 ipv6_only_sock(sk) || 607 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 608 return false; 609 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 610 return false; 611 return true; 612 } 613 614 /* 615 * This routine is called by the ICMP module when it gets some 616 * sort of error condition. If err < 0 then the socket should 617 * be closed and the error returned to the user. If err > 0 618 * it's just the icmp type << 8 | icmp code. 619 * Header points to the ip header of the error packet. We move 620 * on past this. Then (as it used to claim before adjustment) 621 * header points to the first 8 bytes of the udp header. We need 622 * to find the appropriate port. 623 */ 624 625 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 626 { 627 struct inet_sock *inet; 628 const struct iphdr *iph = (const struct iphdr *)skb->data; 629 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 630 const int type = icmp_hdr(skb)->type; 631 const int code = icmp_hdr(skb)->code; 632 struct sock *sk; 633 int harderr; 634 int err; 635 struct net *net = dev_net(skb->dev); 636 637 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 638 iph->saddr, uh->source, skb->dev->ifindex, udptable); 639 if (!sk) { 640 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 641 return; /* No socket for error */ 642 } 643 644 err = 0; 645 harderr = 0; 646 inet = inet_sk(sk); 647 648 switch (type) { 649 default: 650 case ICMP_TIME_EXCEEDED: 651 err = EHOSTUNREACH; 652 break; 653 case ICMP_SOURCE_QUENCH: 654 goto out; 655 case ICMP_PARAMETERPROB: 656 err = EPROTO; 657 harderr = 1; 658 break; 659 case ICMP_DEST_UNREACH: 660 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 661 ipv4_sk_update_pmtu(skb, sk, info); 662 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 663 err = EMSGSIZE; 664 harderr = 1; 665 break; 666 } 667 goto out; 668 } 669 err = EHOSTUNREACH; 670 if (code <= NR_ICMP_UNREACH) { 671 harderr = icmp_err_convert[code].fatal; 672 err = icmp_err_convert[code].errno; 673 } 674 break; 675 case ICMP_REDIRECT: 676 ipv4_sk_redirect(skb, sk); 677 goto out; 678 } 679 680 /* 681 * RFC1122: OK. Passes ICMP errors back to application, as per 682 * 4.1.3.3. 683 */ 684 if (!inet->recverr) { 685 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 686 goto out; 687 } else 688 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 689 690 sk->sk_err = err; 691 sk->sk_error_report(sk); 692 out: 693 sock_put(sk); 694 } 695 696 void udp_err(struct sk_buff *skb, u32 info) 697 { 698 __udp4_lib_err(skb, info, &udp_table); 699 } 700 701 /* 702 * Throw away all pending data and cancel the corking. Socket is locked. 703 */ 704 void udp_flush_pending_frames(struct sock *sk) 705 { 706 struct udp_sock *up = udp_sk(sk); 707 708 if (up->pending) { 709 up->len = 0; 710 up->pending = 0; 711 ip_flush_pending_frames(sk); 712 } 713 } 714 EXPORT_SYMBOL(udp_flush_pending_frames); 715 716 /** 717 * udp4_hwcsum - handle outgoing HW checksumming 718 * @skb: sk_buff containing the filled-in UDP header 719 * (checksum field must be zeroed out) 720 * @src: source IP address 721 * @dst: destination IP address 722 */ 723 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 724 { 725 struct udphdr *uh = udp_hdr(skb); 726 int offset = skb_transport_offset(skb); 727 int len = skb->len - offset; 728 int hlen = len; 729 __wsum csum = 0; 730 731 if (!skb_has_frag_list(skb)) { 732 /* 733 * Only one fragment on the socket. 734 */ 735 skb->csum_start = skb_transport_header(skb) - skb->head; 736 skb->csum_offset = offsetof(struct udphdr, check); 737 uh->check = ~csum_tcpudp_magic(src, dst, len, 738 IPPROTO_UDP, 0); 739 } else { 740 struct sk_buff *frags; 741 742 /* 743 * HW-checksum won't work as there are two or more 744 * fragments on the socket so that all csums of sk_buffs 745 * should be together 746 */ 747 skb_walk_frags(skb, frags) { 748 csum = csum_add(csum, frags->csum); 749 hlen -= frags->len; 750 } 751 752 csum = skb_checksum(skb, offset, hlen, csum); 753 skb->ip_summed = CHECKSUM_NONE; 754 755 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 756 if (uh->check == 0) 757 uh->check = CSUM_MANGLED_0; 758 } 759 } 760 EXPORT_SYMBOL_GPL(udp4_hwcsum); 761 762 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 763 * for the simple case like when setting the checksum for a UDP tunnel. 764 */ 765 void udp_set_csum(bool nocheck, struct sk_buff *skb, 766 __be32 saddr, __be32 daddr, int len) 767 { 768 struct udphdr *uh = udp_hdr(skb); 769 770 if (nocheck) 771 uh->check = 0; 772 else if (skb_is_gso(skb)) 773 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 774 else if (skb_dst(skb) && skb_dst(skb)->dev && 775 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) { 776 777 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 778 779 skb->ip_summed = CHECKSUM_PARTIAL; 780 skb->csum_start = skb_transport_header(skb) - skb->head; 781 skb->csum_offset = offsetof(struct udphdr, check); 782 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 783 } else { 784 __wsum csum; 785 786 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 787 788 uh->check = 0; 789 csum = skb_checksum(skb, 0, len, 0); 790 uh->check = udp_v4_check(len, saddr, daddr, csum); 791 if (uh->check == 0) 792 uh->check = CSUM_MANGLED_0; 793 794 skb->ip_summed = CHECKSUM_UNNECESSARY; 795 } 796 } 797 EXPORT_SYMBOL(udp_set_csum); 798 799 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 800 { 801 struct sock *sk = skb->sk; 802 struct inet_sock *inet = inet_sk(sk); 803 struct udphdr *uh; 804 int err = 0; 805 int is_udplite = IS_UDPLITE(sk); 806 int offset = skb_transport_offset(skb); 807 int len = skb->len - offset; 808 __wsum csum = 0; 809 810 /* 811 * Create a UDP header 812 */ 813 uh = udp_hdr(skb); 814 uh->source = inet->inet_sport; 815 uh->dest = fl4->fl4_dport; 816 uh->len = htons(len); 817 uh->check = 0; 818 819 if (is_udplite) /* UDP-Lite */ 820 csum = udplite_csum(skb); 821 822 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 823 824 skb->ip_summed = CHECKSUM_NONE; 825 goto send; 826 827 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 828 829 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 830 goto send; 831 832 } else 833 csum = udp_csum(skb); 834 835 /* add protocol-dependent pseudo-header */ 836 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 837 sk->sk_protocol, csum); 838 if (uh->check == 0) 839 uh->check = CSUM_MANGLED_0; 840 841 send: 842 err = ip_send_skb(sock_net(sk), skb); 843 if (err) { 844 if (err == -ENOBUFS && !inet->recverr) { 845 UDP_INC_STATS_USER(sock_net(sk), 846 UDP_MIB_SNDBUFERRORS, is_udplite); 847 err = 0; 848 } 849 } else 850 UDP_INC_STATS_USER(sock_net(sk), 851 UDP_MIB_OUTDATAGRAMS, is_udplite); 852 return err; 853 } 854 855 /* 856 * Push out all pending data as one UDP datagram. Socket is locked. 857 */ 858 int udp_push_pending_frames(struct sock *sk) 859 { 860 struct udp_sock *up = udp_sk(sk); 861 struct inet_sock *inet = inet_sk(sk); 862 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 863 struct sk_buff *skb; 864 int err = 0; 865 866 skb = ip_finish_skb(sk, fl4); 867 if (!skb) 868 goto out; 869 870 err = udp_send_skb(skb, fl4); 871 872 out: 873 up->len = 0; 874 up->pending = 0; 875 return err; 876 } 877 EXPORT_SYMBOL(udp_push_pending_frames); 878 879 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 880 { 881 struct inet_sock *inet = inet_sk(sk); 882 struct udp_sock *up = udp_sk(sk); 883 struct flowi4 fl4_stack; 884 struct flowi4 *fl4; 885 int ulen = len; 886 struct ipcm_cookie ipc; 887 struct rtable *rt = NULL; 888 int free = 0; 889 int connected = 0; 890 __be32 daddr, faddr, saddr; 891 __be16 dport; 892 u8 tos; 893 int err, is_udplite = IS_UDPLITE(sk); 894 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 895 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 896 struct sk_buff *skb; 897 struct ip_options_data opt_copy; 898 899 if (len > 0xFFFF) 900 return -EMSGSIZE; 901 902 /* 903 * Check the flags. 904 */ 905 906 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 907 return -EOPNOTSUPP; 908 909 ipc.opt = NULL; 910 ipc.tx_flags = 0; 911 ipc.ttl = 0; 912 ipc.tos = -1; 913 914 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 915 916 fl4 = &inet->cork.fl.u.ip4; 917 if (up->pending) { 918 /* 919 * There are pending frames. 920 * The socket lock must be held while it's corked. 921 */ 922 lock_sock(sk); 923 if (likely(up->pending)) { 924 if (unlikely(up->pending != AF_INET)) { 925 release_sock(sk); 926 return -EINVAL; 927 } 928 goto do_append_data; 929 } 930 release_sock(sk); 931 } 932 ulen += sizeof(struct udphdr); 933 934 /* 935 * Get and verify the address. 936 */ 937 if (msg->msg_name) { 938 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 939 if (msg->msg_namelen < sizeof(*usin)) 940 return -EINVAL; 941 if (usin->sin_family != AF_INET) { 942 if (usin->sin_family != AF_UNSPEC) 943 return -EAFNOSUPPORT; 944 } 945 946 daddr = usin->sin_addr.s_addr; 947 dport = usin->sin_port; 948 if (dport == 0) 949 return -EINVAL; 950 } else { 951 if (sk->sk_state != TCP_ESTABLISHED) 952 return -EDESTADDRREQ; 953 daddr = inet->inet_daddr; 954 dport = inet->inet_dport; 955 /* Open fast path for connected socket. 956 Route will not be used, if at least one option is set. 957 */ 958 connected = 1; 959 } 960 ipc.addr = inet->inet_saddr; 961 962 ipc.oif = sk->sk_bound_dev_if; 963 964 sock_tx_timestamp(sk, &ipc.tx_flags); 965 966 if (msg->msg_controllen) { 967 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 968 sk->sk_family == AF_INET6); 969 if (err) 970 return err; 971 if (ipc.opt) 972 free = 1; 973 connected = 0; 974 } 975 if (!ipc.opt) { 976 struct ip_options_rcu *inet_opt; 977 978 rcu_read_lock(); 979 inet_opt = rcu_dereference(inet->inet_opt); 980 if (inet_opt) { 981 memcpy(&opt_copy, inet_opt, 982 sizeof(*inet_opt) + inet_opt->opt.optlen); 983 ipc.opt = &opt_copy.opt; 984 } 985 rcu_read_unlock(); 986 } 987 988 saddr = ipc.addr; 989 ipc.addr = faddr = daddr; 990 991 if (ipc.opt && ipc.opt->opt.srr) { 992 if (!daddr) 993 return -EINVAL; 994 faddr = ipc.opt->opt.faddr; 995 connected = 0; 996 } 997 tos = get_rttos(&ipc, inet); 998 if (sock_flag(sk, SOCK_LOCALROUTE) || 999 (msg->msg_flags & MSG_DONTROUTE) || 1000 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1001 tos |= RTO_ONLINK; 1002 connected = 0; 1003 } 1004 1005 if (ipv4_is_multicast(daddr)) { 1006 if (!ipc.oif) 1007 ipc.oif = inet->mc_index; 1008 if (!saddr) 1009 saddr = inet->mc_addr; 1010 connected = 0; 1011 } else if (!ipc.oif) 1012 ipc.oif = inet->uc_index; 1013 1014 if (connected) 1015 rt = (struct rtable *)sk_dst_check(sk, 0); 1016 1017 if (!rt) { 1018 struct net *net = sock_net(sk); 1019 __u8 flow_flags = inet_sk_flowi_flags(sk); 1020 1021 fl4 = &fl4_stack; 1022 1023 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1024 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1025 flow_flags, 1026 faddr, saddr, dport, inet->inet_sport); 1027 1028 if (!saddr && ipc.oif) { 1029 err = l3mdev_get_saddr(net, ipc.oif, fl4); 1030 if (err < 0) 1031 goto out; 1032 } 1033 1034 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1035 rt = ip_route_output_flow(net, fl4, sk); 1036 if (IS_ERR(rt)) { 1037 err = PTR_ERR(rt); 1038 rt = NULL; 1039 if (err == -ENETUNREACH) 1040 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1041 goto out; 1042 } 1043 1044 err = -EACCES; 1045 if ((rt->rt_flags & RTCF_BROADCAST) && 1046 !sock_flag(sk, SOCK_BROADCAST)) 1047 goto out; 1048 if (connected) 1049 sk_dst_set(sk, dst_clone(&rt->dst)); 1050 } 1051 1052 if (msg->msg_flags&MSG_CONFIRM) 1053 goto do_confirm; 1054 back_from_confirm: 1055 1056 saddr = fl4->saddr; 1057 if (!ipc.addr) 1058 daddr = ipc.addr = fl4->daddr; 1059 1060 /* Lockless fast path for the non-corking case. */ 1061 if (!corkreq) { 1062 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1063 sizeof(struct udphdr), &ipc, &rt, 1064 msg->msg_flags); 1065 err = PTR_ERR(skb); 1066 if (!IS_ERR_OR_NULL(skb)) 1067 err = udp_send_skb(skb, fl4); 1068 goto out; 1069 } 1070 1071 lock_sock(sk); 1072 if (unlikely(up->pending)) { 1073 /* The socket is already corked while preparing it. */ 1074 /* ... which is an evident application bug. --ANK */ 1075 release_sock(sk); 1076 1077 net_dbg_ratelimited("cork app bug 2\n"); 1078 err = -EINVAL; 1079 goto out; 1080 } 1081 /* 1082 * Now cork the socket to pend data. 1083 */ 1084 fl4 = &inet->cork.fl.u.ip4; 1085 fl4->daddr = daddr; 1086 fl4->saddr = saddr; 1087 fl4->fl4_dport = dport; 1088 fl4->fl4_sport = inet->inet_sport; 1089 up->pending = AF_INET; 1090 1091 do_append_data: 1092 up->len += ulen; 1093 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1094 sizeof(struct udphdr), &ipc, &rt, 1095 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1096 if (err) 1097 udp_flush_pending_frames(sk); 1098 else if (!corkreq) 1099 err = udp_push_pending_frames(sk); 1100 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1101 up->pending = 0; 1102 release_sock(sk); 1103 1104 out: 1105 ip_rt_put(rt); 1106 if (free) 1107 kfree(ipc.opt); 1108 if (!err) 1109 return len; 1110 /* 1111 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1112 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1113 * we don't have a good statistic (IpOutDiscards but it can be too many 1114 * things). We could add another new stat but at least for now that 1115 * seems like overkill. 1116 */ 1117 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1118 UDP_INC_STATS_USER(sock_net(sk), 1119 UDP_MIB_SNDBUFERRORS, is_udplite); 1120 } 1121 return err; 1122 1123 do_confirm: 1124 dst_confirm(&rt->dst); 1125 if (!(msg->msg_flags&MSG_PROBE) || len) 1126 goto back_from_confirm; 1127 err = 0; 1128 goto out; 1129 } 1130 EXPORT_SYMBOL(udp_sendmsg); 1131 1132 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1133 size_t size, int flags) 1134 { 1135 struct inet_sock *inet = inet_sk(sk); 1136 struct udp_sock *up = udp_sk(sk); 1137 int ret; 1138 1139 if (flags & MSG_SENDPAGE_NOTLAST) 1140 flags |= MSG_MORE; 1141 1142 if (!up->pending) { 1143 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1144 1145 /* Call udp_sendmsg to specify destination address which 1146 * sendpage interface can't pass. 1147 * This will succeed only when the socket is connected. 1148 */ 1149 ret = udp_sendmsg(sk, &msg, 0); 1150 if (ret < 0) 1151 return ret; 1152 } 1153 1154 lock_sock(sk); 1155 1156 if (unlikely(!up->pending)) { 1157 release_sock(sk); 1158 1159 net_dbg_ratelimited("udp cork app bug 3\n"); 1160 return -EINVAL; 1161 } 1162 1163 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1164 page, offset, size, flags); 1165 if (ret == -EOPNOTSUPP) { 1166 release_sock(sk); 1167 return sock_no_sendpage(sk->sk_socket, page, offset, 1168 size, flags); 1169 } 1170 if (ret < 0) { 1171 udp_flush_pending_frames(sk); 1172 goto out; 1173 } 1174 1175 up->len += size; 1176 if (!(up->corkflag || (flags&MSG_MORE))) 1177 ret = udp_push_pending_frames(sk); 1178 if (!ret) 1179 ret = size; 1180 out: 1181 release_sock(sk); 1182 return ret; 1183 } 1184 1185 /** 1186 * first_packet_length - return length of first packet in receive queue 1187 * @sk: socket 1188 * 1189 * Drops all bad checksum frames, until a valid one is found. 1190 * Returns the length of found skb, or 0 if none is found. 1191 */ 1192 static unsigned int first_packet_length(struct sock *sk) 1193 { 1194 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1195 struct sk_buff *skb; 1196 unsigned int res; 1197 1198 __skb_queue_head_init(&list_kill); 1199 1200 spin_lock_bh(&rcvq->lock); 1201 while ((skb = skb_peek(rcvq)) != NULL && 1202 udp_lib_checksum_complete(skb)) { 1203 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1204 IS_UDPLITE(sk)); 1205 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1206 IS_UDPLITE(sk)); 1207 atomic_inc(&sk->sk_drops); 1208 __skb_unlink(skb, rcvq); 1209 __skb_queue_tail(&list_kill, skb); 1210 } 1211 res = skb ? skb->len : 0; 1212 spin_unlock_bh(&rcvq->lock); 1213 1214 if (!skb_queue_empty(&list_kill)) { 1215 bool slow = lock_sock_fast(sk); 1216 1217 __skb_queue_purge(&list_kill); 1218 sk_mem_reclaim_partial(sk); 1219 unlock_sock_fast(sk, slow); 1220 } 1221 return res; 1222 } 1223 1224 /* 1225 * IOCTL requests applicable to the UDP protocol 1226 */ 1227 1228 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1229 { 1230 switch (cmd) { 1231 case SIOCOUTQ: 1232 { 1233 int amount = sk_wmem_alloc_get(sk); 1234 1235 return put_user(amount, (int __user *)arg); 1236 } 1237 1238 case SIOCINQ: 1239 { 1240 unsigned int amount = first_packet_length(sk); 1241 1242 if (amount) 1243 /* 1244 * We will only return the amount 1245 * of this packet since that is all 1246 * that will be read. 1247 */ 1248 amount -= sizeof(struct udphdr); 1249 1250 return put_user(amount, (int __user *)arg); 1251 } 1252 1253 default: 1254 return -ENOIOCTLCMD; 1255 } 1256 1257 return 0; 1258 } 1259 EXPORT_SYMBOL(udp_ioctl); 1260 1261 /* 1262 * This should be easy, if there is something there we 1263 * return it, otherwise we block. 1264 */ 1265 1266 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1267 int flags, int *addr_len) 1268 { 1269 struct inet_sock *inet = inet_sk(sk); 1270 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1271 struct sk_buff *skb; 1272 unsigned int ulen, copied; 1273 int peeked, off = 0; 1274 int err; 1275 int is_udplite = IS_UDPLITE(sk); 1276 bool slow; 1277 1278 if (flags & MSG_ERRQUEUE) 1279 return ip_recv_error(sk, msg, len, addr_len); 1280 1281 try_again: 1282 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1283 &peeked, &off, &err); 1284 if (!skb) 1285 goto out; 1286 1287 ulen = skb->len - sizeof(struct udphdr); 1288 copied = len; 1289 if (copied > ulen) 1290 copied = ulen; 1291 else if (copied < ulen) 1292 msg->msg_flags |= MSG_TRUNC; 1293 1294 /* 1295 * If checksum is needed at all, try to do it while copying the 1296 * data. If the data is truncated, or if we only want a partial 1297 * coverage checksum (UDP-Lite), do it before the copy. 1298 */ 1299 1300 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1301 if (udp_lib_checksum_complete(skb)) 1302 goto csum_copy_err; 1303 } 1304 1305 if (skb_csum_unnecessary(skb)) 1306 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 1307 msg, copied); 1308 else { 1309 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), 1310 msg); 1311 1312 if (err == -EINVAL) 1313 goto csum_copy_err; 1314 } 1315 1316 if (unlikely(err)) { 1317 trace_kfree_skb(skb, udp_recvmsg); 1318 if (!peeked) { 1319 atomic_inc(&sk->sk_drops); 1320 UDP_INC_STATS_USER(sock_net(sk), 1321 UDP_MIB_INERRORS, is_udplite); 1322 } 1323 goto out_free; 1324 } 1325 1326 if (!peeked) 1327 UDP_INC_STATS_USER(sock_net(sk), 1328 UDP_MIB_INDATAGRAMS, is_udplite); 1329 1330 sock_recv_ts_and_drops(msg, sk, skb); 1331 1332 /* Copy the address. */ 1333 if (sin) { 1334 sin->sin_family = AF_INET; 1335 sin->sin_port = udp_hdr(skb)->source; 1336 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1337 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1338 *addr_len = sizeof(*sin); 1339 } 1340 if (inet->cmsg_flags) 1341 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr)); 1342 1343 err = copied; 1344 if (flags & MSG_TRUNC) 1345 err = ulen; 1346 1347 out_free: 1348 skb_free_datagram_locked(sk, skb); 1349 out: 1350 return err; 1351 1352 csum_copy_err: 1353 slow = lock_sock_fast(sk); 1354 if (!skb_kill_datagram(sk, skb, flags)) { 1355 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1356 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1357 } 1358 unlock_sock_fast(sk, slow); 1359 1360 /* starting over for a new packet, but check if we need to yield */ 1361 cond_resched(); 1362 msg->msg_flags &= ~MSG_TRUNC; 1363 goto try_again; 1364 } 1365 1366 int udp_disconnect(struct sock *sk, int flags) 1367 { 1368 struct inet_sock *inet = inet_sk(sk); 1369 /* 1370 * 1003.1g - break association. 1371 */ 1372 1373 sk->sk_state = TCP_CLOSE; 1374 inet->inet_daddr = 0; 1375 inet->inet_dport = 0; 1376 sock_rps_reset_rxhash(sk); 1377 sk->sk_bound_dev_if = 0; 1378 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1379 inet_reset_saddr(sk); 1380 1381 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1382 sk->sk_prot->unhash(sk); 1383 inet->inet_sport = 0; 1384 } 1385 sk_dst_reset(sk); 1386 return 0; 1387 } 1388 EXPORT_SYMBOL(udp_disconnect); 1389 1390 void udp_lib_unhash(struct sock *sk) 1391 { 1392 if (sk_hashed(sk)) { 1393 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1394 struct udp_hslot *hslot, *hslot2; 1395 1396 hslot = udp_hashslot(udptable, sock_net(sk), 1397 udp_sk(sk)->udp_port_hash); 1398 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1399 1400 spin_lock_bh(&hslot->lock); 1401 if (sk_nulls_del_node_init_rcu(sk)) { 1402 hslot->count--; 1403 inet_sk(sk)->inet_num = 0; 1404 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1405 1406 spin_lock(&hslot2->lock); 1407 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1408 hslot2->count--; 1409 spin_unlock(&hslot2->lock); 1410 } 1411 spin_unlock_bh(&hslot->lock); 1412 } 1413 } 1414 EXPORT_SYMBOL(udp_lib_unhash); 1415 1416 /* 1417 * inet_rcv_saddr was changed, we must rehash secondary hash 1418 */ 1419 void udp_lib_rehash(struct sock *sk, u16 newhash) 1420 { 1421 if (sk_hashed(sk)) { 1422 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1423 struct udp_hslot *hslot, *hslot2, *nhslot2; 1424 1425 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1426 nhslot2 = udp_hashslot2(udptable, newhash); 1427 udp_sk(sk)->udp_portaddr_hash = newhash; 1428 if (hslot2 != nhslot2) { 1429 hslot = udp_hashslot(udptable, sock_net(sk), 1430 udp_sk(sk)->udp_port_hash); 1431 /* we must lock primary chain too */ 1432 spin_lock_bh(&hslot->lock); 1433 1434 spin_lock(&hslot2->lock); 1435 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1436 hslot2->count--; 1437 spin_unlock(&hslot2->lock); 1438 1439 spin_lock(&nhslot2->lock); 1440 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1441 &nhslot2->head); 1442 nhslot2->count++; 1443 spin_unlock(&nhslot2->lock); 1444 1445 spin_unlock_bh(&hslot->lock); 1446 } 1447 } 1448 } 1449 EXPORT_SYMBOL(udp_lib_rehash); 1450 1451 static void udp_v4_rehash(struct sock *sk) 1452 { 1453 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1454 inet_sk(sk)->inet_rcv_saddr, 1455 inet_sk(sk)->inet_num); 1456 udp_lib_rehash(sk, new_hash); 1457 } 1458 1459 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1460 { 1461 int rc; 1462 1463 if (inet_sk(sk)->inet_daddr) { 1464 sock_rps_save_rxhash(sk, skb); 1465 sk_mark_napi_id(sk, skb); 1466 sk_incoming_cpu_update(sk); 1467 } 1468 1469 rc = sock_queue_rcv_skb(sk, skb); 1470 if (rc < 0) { 1471 int is_udplite = IS_UDPLITE(sk); 1472 1473 /* Note that an ENOMEM error is charged twice */ 1474 if (rc == -ENOMEM) 1475 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1476 is_udplite); 1477 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1478 kfree_skb(skb); 1479 trace_udp_fail_queue_rcv_skb(rc, sk); 1480 return -1; 1481 } 1482 1483 return 0; 1484 1485 } 1486 1487 static struct static_key udp_encap_needed __read_mostly; 1488 void udp_encap_enable(void) 1489 { 1490 if (!static_key_enabled(&udp_encap_needed)) 1491 static_key_slow_inc(&udp_encap_needed); 1492 } 1493 EXPORT_SYMBOL(udp_encap_enable); 1494 1495 /* returns: 1496 * -1: error 1497 * 0: success 1498 * >0: "udp encap" protocol resubmission 1499 * 1500 * Note that in the success and error cases, the skb is assumed to 1501 * have either been requeued or freed. 1502 */ 1503 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1504 { 1505 struct udp_sock *up = udp_sk(sk); 1506 int rc; 1507 int is_udplite = IS_UDPLITE(sk); 1508 1509 /* 1510 * Charge it to the socket, dropping if the queue is full. 1511 */ 1512 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1513 goto drop; 1514 nf_reset(skb); 1515 1516 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1517 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1518 1519 /* 1520 * This is an encapsulation socket so pass the skb to 1521 * the socket's udp_encap_rcv() hook. Otherwise, just 1522 * fall through and pass this up the UDP socket. 1523 * up->encap_rcv() returns the following value: 1524 * =0 if skb was successfully passed to the encap 1525 * handler or was discarded by it. 1526 * >0 if skb should be passed on to UDP. 1527 * <0 if skb should be resubmitted as proto -N 1528 */ 1529 1530 /* if we're overly short, let UDP handle it */ 1531 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1532 if (skb->len > sizeof(struct udphdr) && encap_rcv) { 1533 int ret; 1534 1535 /* Verify checksum before giving to encap */ 1536 if (udp_lib_checksum_complete(skb)) 1537 goto csum_error; 1538 1539 ret = encap_rcv(sk, skb); 1540 if (ret <= 0) { 1541 UDP_INC_STATS_BH(sock_net(sk), 1542 UDP_MIB_INDATAGRAMS, 1543 is_udplite); 1544 return -ret; 1545 } 1546 } 1547 1548 /* FALLTHROUGH -- it's a UDP Packet */ 1549 } 1550 1551 /* 1552 * UDP-Lite specific tests, ignored on UDP sockets 1553 */ 1554 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1555 1556 /* 1557 * MIB statistics other than incrementing the error count are 1558 * disabled for the following two types of errors: these depend 1559 * on the application settings, not on the functioning of the 1560 * protocol stack as such. 1561 * 1562 * RFC 3828 here recommends (sec 3.3): "There should also be a 1563 * way ... to ... at least let the receiving application block 1564 * delivery of packets with coverage values less than a value 1565 * provided by the application." 1566 */ 1567 if (up->pcrlen == 0) { /* full coverage was set */ 1568 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1569 UDP_SKB_CB(skb)->cscov, skb->len); 1570 goto drop; 1571 } 1572 /* The next case involves violating the min. coverage requested 1573 * by the receiver. This is subtle: if receiver wants x and x is 1574 * greater than the buffersize/MTU then receiver will complain 1575 * that it wants x while sender emits packets of smaller size y. 1576 * Therefore the above ...()->partial_cov statement is essential. 1577 */ 1578 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1579 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1580 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1581 goto drop; 1582 } 1583 } 1584 1585 if (rcu_access_pointer(sk->sk_filter) && 1586 udp_lib_checksum_complete(skb)) 1587 goto csum_error; 1588 1589 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1590 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1591 is_udplite); 1592 goto drop; 1593 } 1594 1595 rc = 0; 1596 1597 ipv4_pktinfo_prepare(sk, skb); 1598 bh_lock_sock(sk); 1599 if (!sock_owned_by_user(sk)) 1600 rc = __udp_queue_rcv_skb(sk, skb); 1601 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1602 bh_unlock_sock(sk); 1603 goto drop; 1604 } 1605 bh_unlock_sock(sk); 1606 1607 return rc; 1608 1609 csum_error: 1610 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1611 drop: 1612 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1613 atomic_inc(&sk->sk_drops); 1614 kfree_skb(skb); 1615 return -1; 1616 } 1617 1618 static void flush_stack(struct sock **stack, unsigned int count, 1619 struct sk_buff *skb, unsigned int final) 1620 { 1621 unsigned int i; 1622 struct sk_buff *skb1 = NULL; 1623 struct sock *sk; 1624 1625 for (i = 0; i < count; i++) { 1626 sk = stack[i]; 1627 if (likely(!skb1)) 1628 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1629 1630 if (!skb1) { 1631 atomic_inc(&sk->sk_drops); 1632 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1633 IS_UDPLITE(sk)); 1634 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1635 IS_UDPLITE(sk)); 1636 } 1637 1638 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1639 skb1 = NULL; 1640 1641 sock_put(sk); 1642 } 1643 if (unlikely(skb1)) 1644 kfree_skb(skb1); 1645 } 1646 1647 /* For TCP sockets, sk_rx_dst is protected by socket lock 1648 * For UDP, we use xchg() to guard against concurrent changes. 1649 */ 1650 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1651 { 1652 struct dst_entry *old; 1653 1654 dst_hold(dst); 1655 old = xchg(&sk->sk_rx_dst, dst); 1656 dst_release(old); 1657 } 1658 1659 /* 1660 * Multicasts and broadcasts go to each listener. 1661 * 1662 * Note: called only from the BH handler context. 1663 */ 1664 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1665 struct udphdr *uh, 1666 __be32 saddr, __be32 daddr, 1667 struct udp_table *udptable, 1668 int proto) 1669 { 1670 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1671 struct hlist_nulls_node *node; 1672 unsigned short hnum = ntohs(uh->dest); 1673 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1674 int dif = skb->dev->ifindex; 1675 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 1676 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1677 bool inner_flushed = false; 1678 1679 if (use_hash2) { 1680 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1681 udp_table.mask; 1682 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask; 1683 start_lookup: 1684 hslot = &udp_table.hash2[hash2]; 1685 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1686 } 1687 1688 spin_lock(&hslot->lock); 1689 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 1690 if (__udp_is_mcast_sock(net, sk, 1691 uh->dest, daddr, 1692 uh->source, saddr, 1693 dif, hnum)) { 1694 if (unlikely(count == ARRAY_SIZE(stack))) { 1695 flush_stack(stack, count, skb, ~0); 1696 inner_flushed = true; 1697 count = 0; 1698 } 1699 stack[count++] = sk; 1700 sock_hold(sk); 1701 } 1702 } 1703 1704 spin_unlock(&hslot->lock); 1705 1706 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1707 if (use_hash2 && hash2 != hash2_any) { 1708 hash2 = hash2_any; 1709 goto start_lookup; 1710 } 1711 1712 /* 1713 * do the slow work with no lock held 1714 */ 1715 if (count) { 1716 flush_stack(stack, count, skb, count - 1); 1717 } else { 1718 if (!inner_flushed) 1719 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 1720 proto == IPPROTO_UDPLITE); 1721 consume_skb(skb); 1722 } 1723 return 0; 1724 } 1725 1726 /* Initialize UDP checksum. If exited with zero value (success), 1727 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1728 * Otherwise, csum completion requires chacksumming packet body, 1729 * including udp header and folding it to skb->csum. 1730 */ 1731 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1732 int proto) 1733 { 1734 int err; 1735 1736 UDP_SKB_CB(skb)->partial_cov = 0; 1737 UDP_SKB_CB(skb)->cscov = skb->len; 1738 1739 if (proto == IPPROTO_UDPLITE) { 1740 err = udplite_checksum_init(skb, uh); 1741 if (err) 1742 return err; 1743 } 1744 1745 return skb_checksum_init_zero_check(skb, proto, uh->check, 1746 inet_compute_pseudo); 1747 } 1748 1749 /* 1750 * All we need to do is get the socket, and then do a checksum. 1751 */ 1752 1753 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1754 int proto) 1755 { 1756 struct sock *sk; 1757 struct udphdr *uh; 1758 unsigned short ulen; 1759 struct rtable *rt = skb_rtable(skb); 1760 __be32 saddr, daddr; 1761 struct net *net = dev_net(skb->dev); 1762 1763 /* 1764 * Validate the packet. 1765 */ 1766 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1767 goto drop; /* No space for header. */ 1768 1769 uh = udp_hdr(skb); 1770 ulen = ntohs(uh->len); 1771 saddr = ip_hdr(skb)->saddr; 1772 daddr = ip_hdr(skb)->daddr; 1773 1774 if (ulen > skb->len) 1775 goto short_packet; 1776 1777 if (proto == IPPROTO_UDP) { 1778 /* UDP validates ulen. */ 1779 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1780 goto short_packet; 1781 uh = udp_hdr(skb); 1782 } 1783 1784 if (udp4_csum_init(skb, uh, proto)) 1785 goto csum_error; 1786 1787 sk = skb_steal_sock(skb); 1788 if (sk) { 1789 struct dst_entry *dst = skb_dst(skb); 1790 int ret; 1791 1792 if (unlikely(sk->sk_rx_dst != dst)) 1793 udp_sk_rx_dst_set(sk, dst); 1794 1795 ret = udp_queue_rcv_skb(sk, skb); 1796 sock_put(sk); 1797 /* a return value > 0 means to resubmit the input, but 1798 * it wants the return to be -protocol, or 0 1799 */ 1800 if (ret > 0) 1801 return -ret; 1802 return 0; 1803 } 1804 1805 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1806 return __udp4_lib_mcast_deliver(net, skb, uh, 1807 saddr, daddr, udptable, proto); 1808 1809 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1810 if (sk) { 1811 int ret; 1812 1813 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1814 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1815 inet_compute_pseudo); 1816 1817 ret = udp_queue_rcv_skb(sk, skb); 1818 sock_put(sk); 1819 1820 /* a return value > 0 means to resubmit the input, but 1821 * it wants the return to be -protocol, or 0 1822 */ 1823 if (ret > 0) 1824 return -ret; 1825 return 0; 1826 } 1827 1828 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1829 goto drop; 1830 nf_reset(skb); 1831 1832 /* No socket. Drop packet silently, if checksum is wrong */ 1833 if (udp_lib_checksum_complete(skb)) 1834 goto csum_error; 1835 1836 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1837 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1838 1839 /* 1840 * Hmm. We got an UDP packet to a port to which we 1841 * don't wanna listen. Ignore it. 1842 */ 1843 kfree_skb(skb); 1844 return 0; 1845 1846 short_packet: 1847 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1848 proto == IPPROTO_UDPLITE ? "Lite" : "", 1849 &saddr, ntohs(uh->source), 1850 ulen, skb->len, 1851 &daddr, ntohs(uh->dest)); 1852 goto drop; 1853 1854 csum_error: 1855 /* 1856 * RFC1122: OK. Discards the bad packet silently (as far as 1857 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1858 */ 1859 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1860 proto == IPPROTO_UDPLITE ? "Lite" : "", 1861 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1862 ulen); 1863 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1864 drop: 1865 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1866 kfree_skb(skb); 1867 return 0; 1868 } 1869 1870 /* We can only early demux multicast if there is a single matching socket. 1871 * If more than one socket found returns NULL 1872 */ 1873 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1874 __be16 loc_port, __be32 loc_addr, 1875 __be16 rmt_port, __be32 rmt_addr, 1876 int dif) 1877 { 1878 struct sock *sk, *result; 1879 struct hlist_nulls_node *node; 1880 unsigned short hnum = ntohs(loc_port); 1881 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1882 struct udp_hslot *hslot = &udp_table.hash[slot]; 1883 1884 /* Do not bother scanning a too big list */ 1885 if (hslot->count > 10) 1886 return NULL; 1887 1888 rcu_read_lock(); 1889 begin: 1890 count = 0; 1891 result = NULL; 1892 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1893 if (__udp_is_mcast_sock(net, sk, 1894 loc_port, loc_addr, 1895 rmt_port, rmt_addr, 1896 dif, hnum)) { 1897 result = sk; 1898 ++count; 1899 } 1900 } 1901 /* 1902 * if the nulls value we got at the end of this lookup is 1903 * not the expected one, we must restart lookup. 1904 * We probably met an item that was moved to another chain. 1905 */ 1906 if (get_nulls_value(node) != slot) 1907 goto begin; 1908 1909 if (result) { 1910 if (count != 1 || 1911 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1912 result = NULL; 1913 else if (unlikely(!__udp_is_mcast_sock(net, result, 1914 loc_port, loc_addr, 1915 rmt_port, rmt_addr, 1916 dif, hnum))) { 1917 sock_put(result); 1918 result = NULL; 1919 } 1920 } 1921 rcu_read_unlock(); 1922 return result; 1923 } 1924 1925 /* For unicast we should only early demux connected sockets or we can 1926 * break forwarding setups. The chains here can be long so only check 1927 * if the first socket is an exact match and if not move on. 1928 */ 1929 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1930 __be16 loc_port, __be32 loc_addr, 1931 __be16 rmt_port, __be32 rmt_addr, 1932 int dif) 1933 { 1934 struct sock *sk, *result; 1935 struct hlist_nulls_node *node; 1936 unsigned short hnum = ntohs(loc_port); 1937 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1938 unsigned int slot2 = hash2 & udp_table.mask; 1939 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1940 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 1941 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1942 1943 rcu_read_lock(); 1944 result = NULL; 1945 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 1946 if (INET_MATCH(sk, net, acookie, 1947 rmt_addr, loc_addr, ports, dif)) 1948 result = sk; 1949 /* Only check first socket in chain */ 1950 break; 1951 } 1952 1953 if (result) { 1954 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1955 result = NULL; 1956 else if (unlikely(!INET_MATCH(sk, net, acookie, 1957 rmt_addr, loc_addr, 1958 ports, dif))) { 1959 sock_put(result); 1960 result = NULL; 1961 } 1962 } 1963 rcu_read_unlock(); 1964 return result; 1965 } 1966 1967 void udp_v4_early_demux(struct sk_buff *skb) 1968 { 1969 struct net *net = dev_net(skb->dev); 1970 const struct iphdr *iph; 1971 const struct udphdr *uh; 1972 struct sock *sk; 1973 struct dst_entry *dst; 1974 int dif = skb->dev->ifindex; 1975 int ours; 1976 1977 /* validate the packet */ 1978 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1979 return; 1980 1981 iph = ip_hdr(skb); 1982 uh = udp_hdr(skb); 1983 1984 if (skb->pkt_type == PACKET_BROADCAST || 1985 skb->pkt_type == PACKET_MULTICAST) { 1986 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 1987 1988 if (!in_dev) 1989 return; 1990 1991 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 1992 iph->protocol); 1993 if (!ours) 1994 return; 1995 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1996 uh->source, iph->saddr, dif); 1997 } else if (skb->pkt_type == PACKET_HOST) { 1998 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1999 uh->source, iph->saddr, dif); 2000 } else { 2001 return; 2002 } 2003 2004 if (!sk) 2005 return; 2006 2007 skb->sk = sk; 2008 skb->destructor = sock_efree; 2009 dst = READ_ONCE(sk->sk_rx_dst); 2010 2011 if (dst) 2012 dst = dst_check(dst, 0); 2013 if (dst) { 2014 /* DST_NOCACHE can not be used without taking a reference */ 2015 if (dst->flags & DST_NOCACHE) { 2016 if (likely(atomic_inc_not_zero(&dst->__refcnt))) 2017 skb_dst_set(skb, dst); 2018 } else { 2019 skb_dst_set_noref(skb, dst); 2020 } 2021 } 2022 } 2023 2024 int udp_rcv(struct sk_buff *skb) 2025 { 2026 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2027 } 2028 2029 void udp_destroy_sock(struct sock *sk) 2030 { 2031 struct udp_sock *up = udp_sk(sk); 2032 bool slow = lock_sock_fast(sk); 2033 udp_flush_pending_frames(sk); 2034 unlock_sock_fast(sk, slow); 2035 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2036 void (*encap_destroy)(struct sock *sk); 2037 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2038 if (encap_destroy) 2039 encap_destroy(sk); 2040 } 2041 } 2042 2043 /* 2044 * Socket option code for UDP 2045 */ 2046 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2047 char __user *optval, unsigned int optlen, 2048 int (*push_pending_frames)(struct sock *)) 2049 { 2050 struct udp_sock *up = udp_sk(sk); 2051 int val, valbool; 2052 int err = 0; 2053 int is_udplite = IS_UDPLITE(sk); 2054 2055 if (optlen < sizeof(int)) 2056 return -EINVAL; 2057 2058 if (get_user(val, (int __user *)optval)) 2059 return -EFAULT; 2060 2061 valbool = val ? 1 : 0; 2062 2063 switch (optname) { 2064 case UDP_CORK: 2065 if (val != 0) { 2066 up->corkflag = 1; 2067 } else { 2068 up->corkflag = 0; 2069 lock_sock(sk); 2070 push_pending_frames(sk); 2071 release_sock(sk); 2072 } 2073 break; 2074 2075 case UDP_ENCAP: 2076 switch (val) { 2077 case 0: 2078 case UDP_ENCAP_ESPINUDP: 2079 case UDP_ENCAP_ESPINUDP_NON_IKE: 2080 up->encap_rcv = xfrm4_udp_encap_rcv; 2081 /* FALLTHROUGH */ 2082 case UDP_ENCAP_L2TPINUDP: 2083 up->encap_type = val; 2084 udp_encap_enable(); 2085 break; 2086 default: 2087 err = -ENOPROTOOPT; 2088 break; 2089 } 2090 break; 2091 2092 case UDP_NO_CHECK6_TX: 2093 up->no_check6_tx = valbool; 2094 break; 2095 2096 case UDP_NO_CHECK6_RX: 2097 up->no_check6_rx = valbool; 2098 break; 2099 2100 /* 2101 * UDP-Lite's partial checksum coverage (RFC 3828). 2102 */ 2103 /* The sender sets actual checksum coverage length via this option. 2104 * The case coverage > packet length is handled by send module. */ 2105 case UDPLITE_SEND_CSCOV: 2106 if (!is_udplite) /* Disable the option on UDP sockets */ 2107 return -ENOPROTOOPT; 2108 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2109 val = 8; 2110 else if (val > USHRT_MAX) 2111 val = USHRT_MAX; 2112 up->pcslen = val; 2113 up->pcflag |= UDPLITE_SEND_CC; 2114 break; 2115 2116 /* The receiver specifies a minimum checksum coverage value. To make 2117 * sense, this should be set to at least 8 (as done below). If zero is 2118 * used, this again means full checksum coverage. */ 2119 case UDPLITE_RECV_CSCOV: 2120 if (!is_udplite) /* Disable the option on UDP sockets */ 2121 return -ENOPROTOOPT; 2122 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2123 val = 8; 2124 else if (val > USHRT_MAX) 2125 val = USHRT_MAX; 2126 up->pcrlen = val; 2127 up->pcflag |= UDPLITE_RECV_CC; 2128 break; 2129 2130 default: 2131 err = -ENOPROTOOPT; 2132 break; 2133 } 2134 2135 return err; 2136 } 2137 EXPORT_SYMBOL(udp_lib_setsockopt); 2138 2139 int udp_setsockopt(struct sock *sk, int level, int optname, 2140 char __user *optval, unsigned int optlen) 2141 { 2142 if (level == SOL_UDP || level == SOL_UDPLITE) 2143 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2144 udp_push_pending_frames); 2145 return ip_setsockopt(sk, level, optname, optval, optlen); 2146 } 2147 2148 #ifdef CONFIG_COMPAT 2149 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2150 char __user *optval, unsigned int optlen) 2151 { 2152 if (level == SOL_UDP || level == SOL_UDPLITE) 2153 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2154 udp_push_pending_frames); 2155 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2156 } 2157 #endif 2158 2159 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2160 char __user *optval, int __user *optlen) 2161 { 2162 struct udp_sock *up = udp_sk(sk); 2163 int val, len; 2164 2165 if (get_user(len, optlen)) 2166 return -EFAULT; 2167 2168 len = min_t(unsigned int, len, sizeof(int)); 2169 2170 if (len < 0) 2171 return -EINVAL; 2172 2173 switch (optname) { 2174 case UDP_CORK: 2175 val = up->corkflag; 2176 break; 2177 2178 case UDP_ENCAP: 2179 val = up->encap_type; 2180 break; 2181 2182 case UDP_NO_CHECK6_TX: 2183 val = up->no_check6_tx; 2184 break; 2185 2186 case UDP_NO_CHECK6_RX: 2187 val = up->no_check6_rx; 2188 break; 2189 2190 /* The following two cannot be changed on UDP sockets, the return is 2191 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2192 case UDPLITE_SEND_CSCOV: 2193 val = up->pcslen; 2194 break; 2195 2196 case UDPLITE_RECV_CSCOV: 2197 val = up->pcrlen; 2198 break; 2199 2200 default: 2201 return -ENOPROTOOPT; 2202 } 2203 2204 if (put_user(len, optlen)) 2205 return -EFAULT; 2206 if (copy_to_user(optval, &val, len)) 2207 return -EFAULT; 2208 return 0; 2209 } 2210 EXPORT_SYMBOL(udp_lib_getsockopt); 2211 2212 int udp_getsockopt(struct sock *sk, int level, int optname, 2213 char __user *optval, int __user *optlen) 2214 { 2215 if (level == SOL_UDP || level == SOL_UDPLITE) 2216 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2217 return ip_getsockopt(sk, level, optname, optval, optlen); 2218 } 2219 2220 #ifdef CONFIG_COMPAT 2221 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2222 char __user *optval, int __user *optlen) 2223 { 2224 if (level == SOL_UDP || level == SOL_UDPLITE) 2225 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2226 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2227 } 2228 #endif 2229 /** 2230 * udp_poll - wait for a UDP event. 2231 * @file - file struct 2232 * @sock - socket 2233 * @wait - poll table 2234 * 2235 * This is same as datagram poll, except for the special case of 2236 * blocking sockets. If application is using a blocking fd 2237 * and a packet with checksum error is in the queue; 2238 * then it could get return from select indicating data available 2239 * but then block when reading it. Add special case code 2240 * to work around these arguably broken applications. 2241 */ 2242 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2243 { 2244 unsigned int mask = datagram_poll(file, sock, wait); 2245 struct sock *sk = sock->sk; 2246 2247 sock_rps_record_flow(sk); 2248 2249 /* Check for false positives due to checksum errors */ 2250 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2251 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2252 mask &= ~(POLLIN | POLLRDNORM); 2253 2254 return mask; 2255 2256 } 2257 EXPORT_SYMBOL(udp_poll); 2258 2259 struct proto udp_prot = { 2260 .name = "UDP", 2261 .owner = THIS_MODULE, 2262 .close = udp_lib_close, 2263 .connect = ip4_datagram_connect, 2264 .disconnect = udp_disconnect, 2265 .ioctl = udp_ioctl, 2266 .destroy = udp_destroy_sock, 2267 .setsockopt = udp_setsockopt, 2268 .getsockopt = udp_getsockopt, 2269 .sendmsg = udp_sendmsg, 2270 .recvmsg = udp_recvmsg, 2271 .sendpage = udp_sendpage, 2272 .backlog_rcv = __udp_queue_rcv_skb, 2273 .release_cb = ip4_datagram_release_cb, 2274 .hash = udp_lib_hash, 2275 .unhash = udp_lib_unhash, 2276 .rehash = udp_v4_rehash, 2277 .get_port = udp_v4_get_port, 2278 .memory_allocated = &udp_memory_allocated, 2279 .sysctl_mem = sysctl_udp_mem, 2280 .sysctl_wmem = &sysctl_udp_wmem_min, 2281 .sysctl_rmem = &sysctl_udp_rmem_min, 2282 .obj_size = sizeof(struct udp_sock), 2283 .slab_flags = SLAB_DESTROY_BY_RCU, 2284 .h.udp_table = &udp_table, 2285 #ifdef CONFIG_COMPAT 2286 .compat_setsockopt = compat_udp_setsockopt, 2287 .compat_getsockopt = compat_udp_getsockopt, 2288 #endif 2289 .clear_sk = sk_prot_clear_portaddr_nulls, 2290 }; 2291 EXPORT_SYMBOL(udp_prot); 2292 2293 /* ------------------------------------------------------------------------ */ 2294 #ifdef CONFIG_PROC_FS 2295 2296 static struct sock *udp_get_first(struct seq_file *seq, int start) 2297 { 2298 struct sock *sk; 2299 struct udp_iter_state *state = seq->private; 2300 struct net *net = seq_file_net(seq); 2301 2302 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2303 ++state->bucket) { 2304 struct hlist_nulls_node *node; 2305 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2306 2307 if (hlist_nulls_empty(&hslot->head)) 2308 continue; 2309 2310 spin_lock_bh(&hslot->lock); 2311 sk_nulls_for_each(sk, node, &hslot->head) { 2312 if (!net_eq(sock_net(sk), net)) 2313 continue; 2314 if (sk->sk_family == state->family) 2315 goto found; 2316 } 2317 spin_unlock_bh(&hslot->lock); 2318 } 2319 sk = NULL; 2320 found: 2321 return sk; 2322 } 2323 2324 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2325 { 2326 struct udp_iter_state *state = seq->private; 2327 struct net *net = seq_file_net(seq); 2328 2329 do { 2330 sk = sk_nulls_next(sk); 2331 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2332 2333 if (!sk) { 2334 if (state->bucket <= state->udp_table->mask) 2335 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2336 return udp_get_first(seq, state->bucket + 1); 2337 } 2338 return sk; 2339 } 2340 2341 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2342 { 2343 struct sock *sk = udp_get_first(seq, 0); 2344 2345 if (sk) 2346 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2347 --pos; 2348 return pos ? NULL : sk; 2349 } 2350 2351 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2352 { 2353 struct udp_iter_state *state = seq->private; 2354 state->bucket = MAX_UDP_PORTS; 2355 2356 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2357 } 2358 2359 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2360 { 2361 struct sock *sk; 2362 2363 if (v == SEQ_START_TOKEN) 2364 sk = udp_get_idx(seq, 0); 2365 else 2366 sk = udp_get_next(seq, v); 2367 2368 ++*pos; 2369 return sk; 2370 } 2371 2372 static void udp_seq_stop(struct seq_file *seq, void *v) 2373 { 2374 struct udp_iter_state *state = seq->private; 2375 2376 if (state->bucket <= state->udp_table->mask) 2377 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2378 } 2379 2380 int udp_seq_open(struct inode *inode, struct file *file) 2381 { 2382 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2383 struct udp_iter_state *s; 2384 int err; 2385 2386 err = seq_open_net(inode, file, &afinfo->seq_ops, 2387 sizeof(struct udp_iter_state)); 2388 if (err < 0) 2389 return err; 2390 2391 s = ((struct seq_file *)file->private_data)->private; 2392 s->family = afinfo->family; 2393 s->udp_table = afinfo->udp_table; 2394 return err; 2395 } 2396 EXPORT_SYMBOL(udp_seq_open); 2397 2398 /* ------------------------------------------------------------------------ */ 2399 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2400 { 2401 struct proc_dir_entry *p; 2402 int rc = 0; 2403 2404 afinfo->seq_ops.start = udp_seq_start; 2405 afinfo->seq_ops.next = udp_seq_next; 2406 afinfo->seq_ops.stop = udp_seq_stop; 2407 2408 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2409 afinfo->seq_fops, afinfo); 2410 if (!p) 2411 rc = -ENOMEM; 2412 return rc; 2413 } 2414 EXPORT_SYMBOL(udp_proc_register); 2415 2416 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2417 { 2418 remove_proc_entry(afinfo->name, net->proc_net); 2419 } 2420 EXPORT_SYMBOL(udp_proc_unregister); 2421 2422 /* ------------------------------------------------------------------------ */ 2423 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2424 int bucket) 2425 { 2426 struct inet_sock *inet = inet_sk(sp); 2427 __be32 dest = inet->inet_daddr; 2428 __be32 src = inet->inet_rcv_saddr; 2429 __u16 destp = ntohs(inet->inet_dport); 2430 __u16 srcp = ntohs(inet->inet_sport); 2431 2432 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2433 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2434 bucket, src, srcp, dest, destp, sp->sk_state, 2435 sk_wmem_alloc_get(sp), 2436 sk_rmem_alloc_get(sp), 2437 0, 0L, 0, 2438 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2439 0, sock_i_ino(sp), 2440 atomic_read(&sp->sk_refcnt), sp, 2441 atomic_read(&sp->sk_drops)); 2442 } 2443 2444 int udp4_seq_show(struct seq_file *seq, void *v) 2445 { 2446 seq_setwidth(seq, 127); 2447 if (v == SEQ_START_TOKEN) 2448 seq_puts(seq, " sl local_address rem_address st tx_queue " 2449 "rx_queue tr tm->when retrnsmt uid timeout " 2450 "inode ref pointer drops"); 2451 else { 2452 struct udp_iter_state *state = seq->private; 2453 2454 udp4_format_sock(v, seq, state->bucket); 2455 } 2456 seq_pad(seq, '\n'); 2457 return 0; 2458 } 2459 2460 static const struct file_operations udp_afinfo_seq_fops = { 2461 .owner = THIS_MODULE, 2462 .open = udp_seq_open, 2463 .read = seq_read, 2464 .llseek = seq_lseek, 2465 .release = seq_release_net 2466 }; 2467 2468 /* ------------------------------------------------------------------------ */ 2469 static struct udp_seq_afinfo udp4_seq_afinfo = { 2470 .name = "udp", 2471 .family = AF_INET, 2472 .udp_table = &udp_table, 2473 .seq_fops = &udp_afinfo_seq_fops, 2474 .seq_ops = { 2475 .show = udp4_seq_show, 2476 }, 2477 }; 2478 2479 static int __net_init udp4_proc_init_net(struct net *net) 2480 { 2481 return udp_proc_register(net, &udp4_seq_afinfo); 2482 } 2483 2484 static void __net_exit udp4_proc_exit_net(struct net *net) 2485 { 2486 udp_proc_unregister(net, &udp4_seq_afinfo); 2487 } 2488 2489 static struct pernet_operations udp4_net_ops = { 2490 .init = udp4_proc_init_net, 2491 .exit = udp4_proc_exit_net, 2492 }; 2493 2494 int __init udp4_proc_init(void) 2495 { 2496 return register_pernet_subsys(&udp4_net_ops); 2497 } 2498 2499 void udp4_proc_exit(void) 2500 { 2501 unregister_pernet_subsys(&udp4_net_ops); 2502 } 2503 #endif /* CONFIG_PROC_FS */ 2504 2505 static __initdata unsigned long uhash_entries; 2506 static int __init set_uhash_entries(char *str) 2507 { 2508 ssize_t ret; 2509 2510 if (!str) 2511 return 0; 2512 2513 ret = kstrtoul(str, 0, &uhash_entries); 2514 if (ret) 2515 return 0; 2516 2517 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2518 uhash_entries = UDP_HTABLE_SIZE_MIN; 2519 return 1; 2520 } 2521 __setup("uhash_entries=", set_uhash_entries); 2522 2523 void __init udp_table_init(struct udp_table *table, const char *name) 2524 { 2525 unsigned int i; 2526 2527 table->hash = alloc_large_system_hash(name, 2528 2 * sizeof(struct udp_hslot), 2529 uhash_entries, 2530 21, /* one slot per 2 MB */ 2531 0, 2532 &table->log, 2533 &table->mask, 2534 UDP_HTABLE_SIZE_MIN, 2535 64 * 1024); 2536 2537 table->hash2 = table->hash + (table->mask + 1); 2538 for (i = 0; i <= table->mask; i++) { 2539 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2540 table->hash[i].count = 0; 2541 spin_lock_init(&table->hash[i].lock); 2542 } 2543 for (i = 0; i <= table->mask; i++) { 2544 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2545 table->hash2[i].count = 0; 2546 spin_lock_init(&table->hash2[i].lock); 2547 } 2548 } 2549 2550 u32 udp_flow_hashrnd(void) 2551 { 2552 static u32 hashrnd __read_mostly; 2553 2554 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2555 2556 return hashrnd; 2557 } 2558 EXPORT_SYMBOL(udp_flow_hashrnd); 2559 2560 void __init udp_init(void) 2561 { 2562 unsigned long limit; 2563 2564 udp_table_init(&udp_table, "UDP"); 2565 limit = nr_free_buffer_pages() / 8; 2566 limit = max(limit, 128UL); 2567 sysctl_udp_mem[0] = limit / 4 * 3; 2568 sysctl_udp_mem[1] = limit; 2569 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2570 2571 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2572 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2573 } 2574