1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The User Datagram Protocol (UDP). 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 12 * Hirokazu Takahashi, <taka@valinux.co.jp> 13 * 14 * Fixes: 15 * Alan Cox : verify_area() calls 16 * Alan Cox : stopped close while in use off icmp 17 * messages. Not a fix but a botch that 18 * for udp at least is 'valid'. 19 * Alan Cox : Fixed icmp handling properly 20 * Alan Cox : Correct error for oversized datagrams 21 * Alan Cox : Tidied select() semantics. 22 * Alan Cox : udp_err() fixed properly, also now 23 * select and read wake correctly on errors 24 * Alan Cox : udp_send verify_area moved to avoid mem leak 25 * Alan Cox : UDP can count its memory 26 * Alan Cox : send to an unknown connection causes 27 * an ECONNREFUSED off the icmp, but 28 * does NOT close. 29 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 31 * bug no longer crashes it. 32 * Fred Van Kempen : Net2e support for sk->broadcast. 33 * Alan Cox : Uses skb_free_datagram 34 * Alan Cox : Added get/set sockopt support. 35 * Alan Cox : Broadcasting without option set returns EACCES. 36 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 37 * Alan Cox : Use ip_tos and ip_ttl 38 * Alan Cox : SNMP Mibs 39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 40 * Matt Dillon : UDP length checks. 41 * Alan Cox : Smarter af_inet used properly. 42 * Alan Cox : Use new kernel side addressing. 43 * Alan Cox : Incorrect return on truncated datagram receive. 44 * Arnt Gulbrandsen : New udp_send and stuff 45 * Alan Cox : Cache last socket 46 * Alan Cox : Route cache 47 * Jon Peatfield : Minor efficiency fix to sendto(). 48 * Mike Shaver : RFC1122 checks. 49 * Alan Cox : Nonblocking error fix. 50 * Willy Konynenberg : Transparent proxying support. 51 * Mike McLagan : Routing by source 52 * David S. Miller : New socket lookup architecture. 53 * Last socket cache retained as it 54 * does have a high hit rate. 55 * Olaf Kirch : Don't linearise iovec on sendmsg. 56 * Andi Kleen : Some cleanups, cache destination entry 57 * for connect. 58 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 59 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 60 * return ENOTCONN for unconnected sockets (POSIX) 61 * Janos Farkas : don't deliver multi/broadcasts to a different 62 * bound-to-device socket 63 * Hirokazu Takahashi : HW checksumming for outgoing UDP 64 * datagrams. 65 * Hirokazu Takahashi : sendfile() on UDP works now. 66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 69 * a single port at the same time. 70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71 * James Chapman : Add L2TP encapsulation type. 72 * 73 * 74 * This program is free software; you can redistribute it and/or 75 * modify it under the terms of the GNU General Public License 76 * as published by the Free Software Foundation; either version 77 * 2 of the License, or (at your option) any later version. 78 */ 79 80 #define pr_fmt(fmt) "UDP: " fmt 81 82 #include <asm/uaccess.h> 83 #include <asm/ioctls.h> 84 #include <linux/bootmem.h> 85 #include <linux/highmem.h> 86 #include <linux/swap.h> 87 #include <linux/types.h> 88 #include <linux/fcntl.h> 89 #include <linux/module.h> 90 #include <linux/socket.h> 91 #include <linux/sockios.h> 92 #include <linux/igmp.h> 93 #include <linux/in.h> 94 #include <linux/errno.h> 95 #include <linux/timer.h> 96 #include <linux/mm.h> 97 #include <linux/inet.h> 98 #include <linux/netdevice.h> 99 #include <linux/slab.h> 100 #include <net/tcp_states.h> 101 #include <linux/skbuff.h> 102 #include <linux/proc_fs.h> 103 #include <linux/seq_file.h> 104 #include <net/net_namespace.h> 105 #include <net/icmp.h> 106 #include <net/inet_hashtables.h> 107 #include <net/route.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <trace/events/udp.h> 111 #include <linux/static_key.h> 112 #include <trace/events/skb.h> 113 #include <net/busy_poll.h> 114 #include "udp_impl.h" 115 116 struct udp_table udp_table __read_mostly; 117 EXPORT_SYMBOL(udp_table); 118 119 long sysctl_udp_mem[3] __read_mostly; 120 EXPORT_SYMBOL(sysctl_udp_mem); 121 122 int sysctl_udp_rmem_min __read_mostly; 123 EXPORT_SYMBOL(sysctl_udp_rmem_min); 124 125 int sysctl_udp_wmem_min __read_mostly; 126 EXPORT_SYMBOL(sysctl_udp_wmem_min); 127 128 atomic_long_t udp_memory_allocated; 129 EXPORT_SYMBOL(udp_memory_allocated); 130 131 #define MAX_UDP_PORTS 65536 132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 133 134 static int udp_lib_lport_inuse(struct net *net, __u16 num, 135 const struct udp_hslot *hslot, 136 unsigned long *bitmap, 137 struct sock *sk, 138 int (*saddr_comp)(const struct sock *sk1, 139 const struct sock *sk2), 140 unsigned int log) 141 { 142 struct sock *sk2; 143 struct hlist_nulls_node *node; 144 kuid_t uid = sock_i_uid(sk); 145 146 sk_nulls_for_each(sk2, node, &hslot->head) 147 if (net_eq(sock_net(sk2), net) && 148 sk2 != sk && 149 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 150 (!sk2->sk_reuse || !sk->sk_reuse) && 151 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 152 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 153 (!sk2->sk_reuseport || !sk->sk_reuseport || 154 !uid_eq(uid, sock_i_uid(sk2))) && 155 (*saddr_comp)(sk, sk2)) { 156 if (bitmap) 157 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 158 bitmap); 159 else 160 return 1; 161 } 162 return 0; 163 } 164 165 /* 166 * Note: we still hold spinlock of primary hash chain, so no other writer 167 * can insert/delete a socket with local_port == num 168 */ 169 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 170 struct udp_hslot *hslot2, 171 struct sock *sk, 172 int (*saddr_comp)(const struct sock *sk1, 173 const struct sock *sk2)) 174 { 175 struct sock *sk2; 176 struct hlist_nulls_node *node; 177 kuid_t uid = sock_i_uid(sk); 178 int res = 0; 179 180 spin_lock(&hslot2->lock); 181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) 182 if (net_eq(sock_net(sk2), net) && 183 sk2 != sk && 184 (udp_sk(sk2)->udp_port_hash == num) && 185 (!sk2->sk_reuse || !sk->sk_reuse) && 186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 188 (!sk2->sk_reuseport || !sk->sk_reuseport || 189 !uid_eq(uid, sock_i_uid(sk2))) && 190 (*saddr_comp)(sk, sk2)) { 191 res = 1; 192 break; 193 } 194 spin_unlock(&hslot2->lock); 195 return res; 196 } 197 198 /** 199 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 200 * 201 * @sk: socket struct in question 202 * @snum: port number to look up 203 * @saddr_comp: AF-dependent comparison of bound local IP addresses 204 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 205 * with NULL address 206 */ 207 int udp_lib_get_port(struct sock *sk, unsigned short snum, 208 int (*saddr_comp)(const struct sock *sk1, 209 const struct sock *sk2), 210 unsigned int hash2_nulladdr) 211 { 212 struct udp_hslot *hslot, *hslot2; 213 struct udp_table *udptable = sk->sk_prot->h.udp_table; 214 int error = 1; 215 struct net *net = sock_net(sk); 216 217 if (!snum) { 218 int low, high, remaining; 219 unsigned int rand; 220 unsigned short first, last; 221 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 222 223 inet_get_local_port_range(net, &low, &high); 224 remaining = (high - low) + 1; 225 226 rand = net_random(); 227 first = (((u64)rand * remaining) >> 32) + low; 228 /* 229 * force rand to be an odd multiple of UDP_HTABLE_SIZE 230 */ 231 rand = (rand | 1) * (udptable->mask + 1); 232 last = first + udptable->mask + 1; 233 do { 234 hslot = udp_hashslot(udptable, net, first); 235 bitmap_zero(bitmap, PORTS_PER_CHAIN); 236 spin_lock_bh(&hslot->lock); 237 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 238 saddr_comp, udptable->log); 239 240 snum = first; 241 /* 242 * Iterate on all possible values of snum for this hash. 243 * Using steps of an odd multiple of UDP_HTABLE_SIZE 244 * give us randomization and full range coverage. 245 */ 246 do { 247 if (low <= snum && snum <= high && 248 !test_bit(snum >> udptable->log, bitmap) && 249 !inet_is_reserved_local_port(snum)) 250 goto found; 251 snum += rand; 252 } while (snum != first); 253 spin_unlock_bh(&hslot->lock); 254 } while (++first != last); 255 goto fail; 256 } else { 257 hslot = udp_hashslot(udptable, net, snum); 258 spin_lock_bh(&hslot->lock); 259 if (hslot->count > 10) { 260 int exist; 261 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 262 263 slot2 &= udptable->mask; 264 hash2_nulladdr &= udptable->mask; 265 266 hslot2 = udp_hashslot2(udptable, slot2); 267 if (hslot->count < hslot2->count) 268 goto scan_primary_hash; 269 270 exist = udp_lib_lport_inuse2(net, snum, hslot2, 271 sk, saddr_comp); 272 if (!exist && (hash2_nulladdr != slot2)) { 273 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 274 exist = udp_lib_lport_inuse2(net, snum, hslot2, 275 sk, saddr_comp); 276 } 277 if (exist) 278 goto fail_unlock; 279 else 280 goto found; 281 } 282 scan_primary_hash: 283 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 284 saddr_comp, 0)) 285 goto fail_unlock; 286 } 287 found: 288 inet_sk(sk)->inet_num = snum; 289 udp_sk(sk)->udp_port_hash = snum; 290 udp_sk(sk)->udp_portaddr_hash ^= snum; 291 if (sk_unhashed(sk)) { 292 sk_nulls_add_node_rcu(sk, &hslot->head); 293 hslot->count++; 294 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 295 296 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 297 spin_lock(&hslot2->lock); 298 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 299 &hslot2->head); 300 hslot2->count++; 301 spin_unlock(&hslot2->lock); 302 } 303 error = 0; 304 fail_unlock: 305 spin_unlock_bh(&hslot->lock); 306 fail: 307 return error; 308 } 309 EXPORT_SYMBOL(udp_lib_get_port); 310 311 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 312 { 313 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 314 315 return (!ipv6_only_sock(sk2) && 316 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 317 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 318 } 319 320 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 321 unsigned int port) 322 { 323 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 324 } 325 326 int udp_v4_get_port(struct sock *sk, unsigned short snum) 327 { 328 unsigned int hash2_nulladdr = 329 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 330 unsigned int hash2_partial = 331 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 332 333 /* precompute partial secondary hash */ 334 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 335 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 336 } 337 338 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 339 unsigned short hnum, 340 __be16 sport, __be32 daddr, __be16 dport, int dif) 341 { 342 int score = -1; 343 344 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 345 !ipv6_only_sock(sk)) { 346 struct inet_sock *inet = inet_sk(sk); 347 348 score = (sk->sk_family == PF_INET ? 2 : 1); 349 if (inet->inet_rcv_saddr) { 350 if (inet->inet_rcv_saddr != daddr) 351 return -1; 352 score += 4; 353 } 354 if (inet->inet_daddr) { 355 if (inet->inet_daddr != saddr) 356 return -1; 357 score += 4; 358 } 359 if (inet->inet_dport) { 360 if (inet->inet_dport != sport) 361 return -1; 362 score += 4; 363 } 364 if (sk->sk_bound_dev_if) { 365 if (sk->sk_bound_dev_if != dif) 366 return -1; 367 score += 4; 368 } 369 } 370 return score; 371 } 372 373 /* 374 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 375 */ 376 static inline int compute_score2(struct sock *sk, struct net *net, 377 __be32 saddr, __be16 sport, 378 __be32 daddr, unsigned int hnum, int dif) 379 { 380 int score = -1; 381 382 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) { 383 struct inet_sock *inet = inet_sk(sk); 384 385 if (inet->inet_rcv_saddr != daddr) 386 return -1; 387 if (inet->inet_num != hnum) 388 return -1; 389 390 score = (sk->sk_family == PF_INET ? 2 : 1); 391 if (inet->inet_daddr) { 392 if (inet->inet_daddr != saddr) 393 return -1; 394 score += 4; 395 } 396 if (inet->inet_dport) { 397 if (inet->inet_dport != sport) 398 return -1; 399 score += 4; 400 } 401 if (sk->sk_bound_dev_if) { 402 if (sk->sk_bound_dev_if != dif) 403 return -1; 404 score += 4; 405 } 406 } 407 return score; 408 } 409 410 static unsigned int udp_ehashfn(struct net *net, const __be32 laddr, 411 const __u16 lport, const __be32 faddr, 412 const __be16 fport) 413 { 414 static u32 udp_ehash_secret __read_mostly; 415 416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 417 418 return __inet_ehashfn(laddr, lport, faddr, fport, 419 udp_ehash_secret + net_hash_mix(net)); 420 } 421 422 423 /* called with read_rcu_lock() */ 424 static struct sock *udp4_lib_lookup2(struct net *net, 425 __be32 saddr, __be16 sport, 426 __be32 daddr, unsigned int hnum, int dif, 427 struct udp_hslot *hslot2, unsigned int slot2) 428 { 429 struct sock *sk, *result; 430 struct hlist_nulls_node *node; 431 int score, badness, matches = 0, reuseport = 0; 432 u32 hash = 0; 433 434 begin: 435 result = NULL; 436 badness = 0; 437 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 438 score = compute_score2(sk, net, saddr, sport, 439 daddr, hnum, dif); 440 if (score > badness) { 441 result = sk; 442 badness = score; 443 reuseport = sk->sk_reuseport; 444 if (reuseport) { 445 hash = udp_ehashfn(net, daddr, hnum, 446 saddr, sport); 447 matches = 1; 448 } 449 } else if (score == badness && reuseport) { 450 matches++; 451 if (((u64)hash * matches) >> 32 == 0) 452 result = sk; 453 hash = next_pseudo_random32(hash); 454 } 455 } 456 /* 457 * if the nulls value we got at the end of this lookup is 458 * not the expected one, we must restart lookup. 459 * We probably met an item that was moved to another chain. 460 */ 461 if (get_nulls_value(node) != slot2) 462 goto begin; 463 if (result) { 464 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 465 result = NULL; 466 else if (unlikely(compute_score2(result, net, saddr, sport, 467 daddr, hnum, dif) < badness)) { 468 sock_put(result); 469 goto begin; 470 } 471 } 472 return result; 473 } 474 475 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 476 * harder than this. -DaveM 477 */ 478 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 479 __be16 sport, __be32 daddr, __be16 dport, 480 int dif, struct udp_table *udptable) 481 { 482 struct sock *sk, *result; 483 struct hlist_nulls_node *node; 484 unsigned short hnum = ntohs(dport); 485 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 486 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 487 int score, badness, matches = 0, reuseport = 0; 488 u32 hash = 0; 489 490 rcu_read_lock(); 491 if (hslot->count > 10) { 492 hash2 = udp4_portaddr_hash(net, daddr, hnum); 493 slot2 = hash2 & udptable->mask; 494 hslot2 = &udptable->hash2[slot2]; 495 if (hslot->count < hslot2->count) 496 goto begin; 497 498 result = udp4_lib_lookup2(net, saddr, sport, 499 daddr, hnum, dif, 500 hslot2, slot2); 501 if (!result) { 502 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 503 slot2 = hash2 & udptable->mask; 504 hslot2 = &udptable->hash2[slot2]; 505 if (hslot->count < hslot2->count) 506 goto begin; 507 508 result = udp4_lib_lookup2(net, saddr, sport, 509 htonl(INADDR_ANY), hnum, dif, 510 hslot2, slot2); 511 } 512 rcu_read_unlock(); 513 return result; 514 } 515 begin: 516 result = NULL; 517 badness = 0; 518 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 519 score = compute_score(sk, net, saddr, hnum, sport, 520 daddr, dport, dif); 521 if (score > badness) { 522 result = sk; 523 badness = score; 524 reuseport = sk->sk_reuseport; 525 if (reuseport) { 526 hash = udp_ehashfn(net, daddr, hnum, 527 saddr, sport); 528 matches = 1; 529 } 530 } else if (score == badness && reuseport) { 531 matches++; 532 if (((u64)hash * matches) >> 32 == 0) 533 result = sk; 534 hash = next_pseudo_random32(hash); 535 } 536 } 537 /* 538 * if the nulls value we got at the end of this lookup is 539 * not the expected one, we must restart lookup. 540 * We probably met an item that was moved to another chain. 541 */ 542 if (get_nulls_value(node) != slot) 543 goto begin; 544 545 if (result) { 546 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 547 result = NULL; 548 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 549 daddr, dport, dif) < badness)) { 550 sock_put(result); 551 goto begin; 552 } 553 } 554 rcu_read_unlock(); 555 return result; 556 } 557 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 558 559 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 560 __be16 sport, __be16 dport, 561 struct udp_table *udptable) 562 { 563 struct sock *sk; 564 const struct iphdr *iph = ip_hdr(skb); 565 566 if (unlikely(sk = skb_steal_sock(skb))) 567 return sk; 568 else 569 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 570 iph->daddr, dport, inet_iif(skb), 571 udptable); 572 } 573 574 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 575 __be32 daddr, __be16 dport, int dif) 576 { 577 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 578 } 579 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 580 581 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 582 __be16 loc_port, __be32 loc_addr, 583 __be16 rmt_port, __be32 rmt_addr, 584 int dif, unsigned short hnum) 585 { 586 struct inet_sock *inet = inet_sk(sk); 587 588 if (!net_eq(sock_net(sk), net) || 589 udp_sk(sk)->udp_port_hash != hnum || 590 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 591 (inet->inet_dport != rmt_port && inet->inet_dport) || 592 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 593 ipv6_only_sock(sk) || 594 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 595 return false; 596 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 597 return false; 598 return true; 599 } 600 601 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 602 __be16 loc_port, __be32 loc_addr, 603 __be16 rmt_port, __be32 rmt_addr, 604 int dif) 605 { 606 struct hlist_nulls_node *node; 607 struct sock *s = sk; 608 unsigned short hnum = ntohs(loc_port); 609 610 sk_nulls_for_each_from(s, node) { 611 if (__udp_is_mcast_sock(net, s, 612 loc_port, loc_addr, 613 rmt_port, rmt_addr, 614 dif, hnum)) 615 goto found; 616 } 617 s = NULL; 618 found: 619 return s; 620 } 621 622 /* 623 * This routine is called by the ICMP module when it gets some 624 * sort of error condition. If err < 0 then the socket should 625 * be closed and the error returned to the user. If err > 0 626 * it's just the icmp type << 8 | icmp code. 627 * Header points to the ip header of the error packet. We move 628 * on past this. Then (as it used to claim before adjustment) 629 * header points to the first 8 bytes of the udp header. We need 630 * to find the appropriate port. 631 */ 632 633 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 634 { 635 struct inet_sock *inet; 636 const struct iphdr *iph = (const struct iphdr *)skb->data; 637 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 638 const int type = icmp_hdr(skb)->type; 639 const int code = icmp_hdr(skb)->code; 640 struct sock *sk; 641 int harderr; 642 int err; 643 struct net *net = dev_net(skb->dev); 644 645 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 646 iph->saddr, uh->source, skb->dev->ifindex, udptable); 647 if (sk == NULL) { 648 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 649 return; /* No socket for error */ 650 } 651 652 err = 0; 653 harderr = 0; 654 inet = inet_sk(sk); 655 656 switch (type) { 657 default: 658 case ICMP_TIME_EXCEEDED: 659 err = EHOSTUNREACH; 660 break; 661 case ICMP_SOURCE_QUENCH: 662 goto out; 663 case ICMP_PARAMETERPROB: 664 err = EPROTO; 665 harderr = 1; 666 break; 667 case ICMP_DEST_UNREACH: 668 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 669 ipv4_sk_update_pmtu(skb, sk, info); 670 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 671 err = EMSGSIZE; 672 harderr = 1; 673 break; 674 } 675 goto out; 676 } 677 err = EHOSTUNREACH; 678 if (code <= NR_ICMP_UNREACH) { 679 harderr = icmp_err_convert[code].fatal; 680 err = icmp_err_convert[code].errno; 681 } 682 break; 683 case ICMP_REDIRECT: 684 ipv4_sk_redirect(skb, sk); 685 goto out; 686 } 687 688 /* 689 * RFC1122: OK. Passes ICMP errors back to application, as per 690 * 4.1.3.3. 691 */ 692 if (!inet->recverr) { 693 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 694 goto out; 695 } else 696 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 697 698 sk->sk_err = err; 699 sk->sk_error_report(sk); 700 out: 701 sock_put(sk); 702 } 703 704 void udp_err(struct sk_buff *skb, u32 info) 705 { 706 __udp4_lib_err(skb, info, &udp_table); 707 } 708 709 /* 710 * Throw away all pending data and cancel the corking. Socket is locked. 711 */ 712 void udp_flush_pending_frames(struct sock *sk) 713 { 714 struct udp_sock *up = udp_sk(sk); 715 716 if (up->pending) { 717 up->len = 0; 718 up->pending = 0; 719 ip_flush_pending_frames(sk); 720 } 721 } 722 EXPORT_SYMBOL(udp_flush_pending_frames); 723 724 /** 725 * udp4_hwcsum - handle outgoing HW checksumming 726 * @skb: sk_buff containing the filled-in UDP header 727 * (checksum field must be zeroed out) 728 * @src: source IP address 729 * @dst: destination IP address 730 */ 731 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 732 { 733 struct udphdr *uh = udp_hdr(skb); 734 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 735 int offset = skb_transport_offset(skb); 736 int len = skb->len - offset; 737 int hlen = len; 738 __wsum csum = 0; 739 740 if (!frags) { 741 /* 742 * Only one fragment on the socket. 743 */ 744 skb->csum_start = skb_transport_header(skb) - skb->head; 745 skb->csum_offset = offsetof(struct udphdr, check); 746 uh->check = ~csum_tcpudp_magic(src, dst, len, 747 IPPROTO_UDP, 0); 748 } else { 749 /* 750 * HW-checksum won't work as there are two or more 751 * fragments on the socket so that all csums of sk_buffs 752 * should be together 753 */ 754 do { 755 csum = csum_add(csum, frags->csum); 756 hlen -= frags->len; 757 } while ((frags = frags->next)); 758 759 csum = skb_checksum(skb, offset, hlen, csum); 760 skb->ip_summed = CHECKSUM_NONE; 761 762 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 763 if (uh->check == 0) 764 uh->check = CSUM_MANGLED_0; 765 } 766 } 767 EXPORT_SYMBOL_GPL(udp4_hwcsum); 768 769 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 770 { 771 struct sock *sk = skb->sk; 772 struct inet_sock *inet = inet_sk(sk); 773 struct udphdr *uh; 774 int err = 0; 775 int is_udplite = IS_UDPLITE(sk); 776 int offset = skb_transport_offset(skb); 777 int len = skb->len - offset; 778 __wsum csum = 0; 779 780 /* 781 * Create a UDP header 782 */ 783 uh = udp_hdr(skb); 784 uh->source = inet->inet_sport; 785 uh->dest = fl4->fl4_dport; 786 uh->len = htons(len); 787 uh->check = 0; 788 789 if (is_udplite) /* UDP-Lite */ 790 csum = udplite_csum(skb); 791 792 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 793 794 skb->ip_summed = CHECKSUM_NONE; 795 goto send; 796 797 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 798 799 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 800 goto send; 801 802 } else 803 csum = udp_csum(skb); 804 805 /* add protocol-dependent pseudo-header */ 806 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 807 sk->sk_protocol, csum); 808 if (uh->check == 0) 809 uh->check = CSUM_MANGLED_0; 810 811 send: 812 err = ip_send_skb(sock_net(sk), skb); 813 if (err) { 814 if (err == -ENOBUFS && !inet->recverr) { 815 UDP_INC_STATS_USER(sock_net(sk), 816 UDP_MIB_SNDBUFERRORS, is_udplite); 817 err = 0; 818 } 819 } else 820 UDP_INC_STATS_USER(sock_net(sk), 821 UDP_MIB_OUTDATAGRAMS, is_udplite); 822 return err; 823 } 824 825 /* 826 * Push out all pending data as one UDP datagram. Socket is locked. 827 */ 828 int udp_push_pending_frames(struct sock *sk) 829 { 830 struct udp_sock *up = udp_sk(sk); 831 struct inet_sock *inet = inet_sk(sk); 832 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 833 struct sk_buff *skb; 834 int err = 0; 835 836 skb = ip_finish_skb(sk, fl4); 837 if (!skb) 838 goto out; 839 840 err = udp_send_skb(skb, fl4); 841 842 out: 843 up->len = 0; 844 up->pending = 0; 845 return err; 846 } 847 EXPORT_SYMBOL(udp_push_pending_frames); 848 849 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 850 size_t len) 851 { 852 struct inet_sock *inet = inet_sk(sk); 853 struct udp_sock *up = udp_sk(sk); 854 struct flowi4 fl4_stack; 855 struct flowi4 *fl4; 856 int ulen = len; 857 struct ipcm_cookie ipc; 858 struct rtable *rt = NULL; 859 int free = 0; 860 int connected = 0; 861 __be32 daddr, faddr, saddr; 862 __be16 dport; 863 u8 tos; 864 int err, is_udplite = IS_UDPLITE(sk); 865 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 866 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 867 struct sk_buff *skb; 868 struct ip_options_data opt_copy; 869 870 if (len > 0xFFFF) 871 return -EMSGSIZE; 872 873 /* 874 * Check the flags. 875 */ 876 877 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 878 return -EOPNOTSUPP; 879 880 ipc.opt = NULL; 881 ipc.tx_flags = 0; 882 ipc.ttl = 0; 883 ipc.tos = -1; 884 885 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 886 887 fl4 = &inet->cork.fl.u.ip4; 888 if (up->pending) { 889 /* 890 * There are pending frames. 891 * The socket lock must be held while it's corked. 892 */ 893 lock_sock(sk); 894 if (likely(up->pending)) { 895 if (unlikely(up->pending != AF_INET)) { 896 release_sock(sk); 897 return -EINVAL; 898 } 899 goto do_append_data; 900 } 901 release_sock(sk); 902 } 903 ulen += sizeof(struct udphdr); 904 905 /* 906 * Get and verify the address. 907 */ 908 if (msg->msg_name) { 909 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name; 910 if (msg->msg_namelen < sizeof(*usin)) 911 return -EINVAL; 912 if (usin->sin_family != AF_INET) { 913 if (usin->sin_family != AF_UNSPEC) 914 return -EAFNOSUPPORT; 915 } 916 917 daddr = usin->sin_addr.s_addr; 918 dport = usin->sin_port; 919 if (dport == 0) 920 return -EINVAL; 921 } else { 922 if (sk->sk_state != TCP_ESTABLISHED) 923 return -EDESTADDRREQ; 924 daddr = inet->inet_daddr; 925 dport = inet->inet_dport; 926 /* Open fast path for connected socket. 927 Route will not be used, if at least one option is set. 928 */ 929 connected = 1; 930 } 931 ipc.addr = inet->inet_saddr; 932 933 ipc.oif = sk->sk_bound_dev_if; 934 935 sock_tx_timestamp(sk, &ipc.tx_flags); 936 937 if (msg->msg_controllen) { 938 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 939 if (err) 940 return err; 941 if (ipc.opt) 942 free = 1; 943 connected = 0; 944 } 945 if (!ipc.opt) { 946 struct ip_options_rcu *inet_opt; 947 948 rcu_read_lock(); 949 inet_opt = rcu_dereference(inet->inet_opt); 950 if (inet_opt) { 951 memcpy(&opt_copy, inet_opt, 952 sizeof(*inet_opt) + inet_opt->opt.optlen); 953 ipc.opt = &opt_copy.opt; 954 } 955 rcu_read_unlock(); 956 } 957 958 saddr = ipc.addr; 959 ipc.addr = faddr = daddr; 960 961 if (ipc.opt && ipc.opt->opt.srr) { 962 if (!daddr) 963 return -EINVAL; 964 faddr = ipc.opt->opt.faddr; 965 connected = 0; 966 } 967 tos = get_rttos(&ipc, inet); 968 if (sock_flag(sk, SOCK_LOCALROUTE) || 969 (msg->msg_flags & MSG_DONTROUTE) || 970 (ipc.opt && ipc.opt->opt.is_strictroute)) { 971 tos |= RTO_ONLINK; 972 connected = 0; 973 } 974 975 if (ipv4_is_multicast(daddr)) { 976 if (!ipc.oif) 977 ipc.oif = inet->mc_index; 978 if (!saddr) 979 saddr = inet->mc_addr; 980 connected = 0; 981 } else if (!ipc.oif) 982 ipc.oif = inet->uc_index; 983 984 if (connected) 985 rt = (struct rtable *)sk_dst_check(sk, 0); 986 987 if (rt == NULL) { 988 struct net *net = sock_net(sk); 989 990 fl4 = &fl4_stack; 991 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 992 RT_SCOPE_UNIVERSE, sk->sk_protocol, 993 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP, 994 faddr, saddr, dport, inet->inet_sport); 995 996 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 997 rt = ip_route_output_flow(net, fl4, sk); 998 if (IS_ERR(rt)) { 999 err = PTR_ERR(rt); 1000 rt = NULL; 1001 if (err == -ENETUNREACH) 1002 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1003 goto out; 1004 } 1005 1006 err = -EACCES; 1007 if ((rt->rt_flags & RTCF_BROADCAST) && 1008 !sock_flag(sk, SOCK_BROADCAST)) 1009 goto out; 1010 if (connected) 1011 sk_dst_set(sk, dst_clone(&rt->dst)); 1012 } 1013 1014 if (msg->msg_flags&MSG_CONFIRM) 1015 goto do_confirm; 1016 back_from_confirm: 1017 1018 saddr = fl4->saddr; 1019 if (!ipc.addr) 1020 daddr = ipc.addr = fl4->daddr; 1021 1022 /* Lockless fast path for the non-corking case. */ 1023 if (!corkreq) { 1024 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen, 1025 sizeof(struct udphdr), &ipc, &rt, 1026 msg->msg_flags); 1027 err = PTR_ERR(skb); 1028 if (!IS_ERR_OR_NULL(skb)) 1029 err = udp_send_skb(skb, fl4); 1030 goto out; 1031 } 1032 1033 lock_sock(sk); 1034 if (unlikely(up->pending)) { 1035 /* The socket is already corked while preparing it. */ 1036 /* ... which is an evident application bug. --ANK */ 1037 release_sock(sk); 1038 1039 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n")); 1040 err = -EINVAL; 1041 goto out; 1042 } 1043 /* 1044 * Now cork the socket to pend data. 1045 */ 1046 fl4 = &inet->cork.fl.u.ip4; 1047 fl4->daddr = daddr; 1048 fl4->saddr = saddr; 1049 fl4->fl4_dport = dport; 1050 fl4->fl4_sport = inet->inet_sport; 1051 up->pending = AF_INET; 1052 1053 do_append_data: 1054 up->len += ulen; 1055 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen, 1056 sizeof(struct udphdr), &ipc, &rt, 1057 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1058 if (err) 1059 udp_flush_pending_frames(sk); 1060 else if (!corkreq) 1061 err = udp_push_pending_frames(sk); 1062 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1063 up->pending = 0; 1064 release_sock(sk); 1065 1066 out: 1067 ip_rt_put(rt); 1068 if (free) 1069 kfree(ipc.opt); 1070 if (!err) 1071 return len; 1072 /* 1073 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1074 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1075 * we don't have a good statistic (IpOutDiscards but it can be too many 1076 * things). We could add another new stat but at least for now that 1077 * seems like overkill. 1078 */ 1079 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1080 UDP_INC_STATS_USER(sock_net(sk), 1081 UDP_MIB_SNDBUFERRORS, is_udplite); 1082 } 1083 return err; 1084 1085 do_confirm: 1086 dst_confirm(&rt->dst); 1087 if (!(msg->msg_flags&MSG_PROBE) || len) 1088 goto back_from_confirm; 1089 err = 0; 1090 goto out; 1091 } 1092 EXPORT_SYMBOL(udp_sendmsg); 1093 1094 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1095 size_t size, int flags) 1096 { 1097 struct inet_sock *inet = inet_sk(sk); 1098 struct udp_sock *up = udp_sk(sk); 1099 int ret; 1100 1101 if (flags & MSG_SENDPAGE_NOTLAST) 1102 flags |= MSG_MORE; 1103 1104 if (!up->pending) { 1105 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1106 1107 /* Call udp_sendmsg to specify destination address which 1108 * sendpage interface can't pass. 1109 * This will succeed only when the socket is connected. 1110 */ 1111 ret = udp_sendmsg(NULL, sk, &msg, 0); 1112 if (ret < 0) 1113 return ret; 1114 } 1115 1116 lock_sock(sk); 1117 1118 if (unlikely(!up->pending)) { 1119 release_sock(sk); 1120 1121 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n")); 1122 return -EINVAL; 1123 } 1124 1125 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1126 page, offset, size, flags); 1127 if (ret == -EOPNOTSUPP) { 1128 release_sock(sk); 1129 return sock_no_sendpage(sk->sk_socket, page, offset, 1130 size, flags); 1131 } 1132 if (ret < 0) { 1133 udp_flush_pending_frames(sk); 1134 goto out; 1135 } 1136 1137 up->len += size; 1138 if (!(up->corkflag || (flags&MSG_MORE))) 1139 ret = udp_push_pending_frames(sk); 1140 if (!ret) 1141 ret = size; 1142 out: 1143 release_sock(sk); 1144 return ret; 1145 } 1146 1147 1148 /** 1149 * first_packet_length - return length of first packet in receive queue 1150 * @sk: socket 1151 * 1152 * Drops all bad checksum frames, until a valid one is found. 1153 * Returns the length of found skb, or 0 if none is found. 1154 */ 1155 static unsigned int first_packet_length(struct sock *sk) 1156 { 1157 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1158 struct sk_buff *skb; 1159 unsigned int res; 1160 1161 __skb_queue_head_init(&list_kill); 1162 1163 spin_lock_bh(&rcvq->lock); 1164 while ((skb = skb_peek(rcvq)) != NULL && 1165 udp_lib_checksum_complete(skb)) { 1166 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1167 IS_UDPLITE(sk)); 1168 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1169 IS_UDPLITE(sk)); 1170 atomic_inc(&sk->sk_drops); 1171 __skb_unlink(skb, rcvq); 1172 __skb_queue_tail(&list_kill, skb); 1173 } 1174 res = skb ? skb->len : 0; 1175 spin_unlock_bh(&rcvq->lock); 1176 1177 if (!skb_queue_empty(&list_kill)) { 1178 bool slow = lock_sock_fast(sk); 1179 1180 __skb_queue_purge(&list_kill); 1181 sk_mem_reclaim_partial(sk); 1182 unlock_sock_fast(sk, slow); 1183 } 1184 return res; 1185 } 1186 1187 /* 1188 * IOCTL requests applicable to the UDP protocol 1189 */ 1190 1191 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1192 { 1193 switch (cmd) { 1194 case SIOCOUTQ: 1195 { 1196 int amount = sk_wmem_alloc_get(sk); 1197 1198 return put_user(amount, (int __user *)arg); 1199 } 1200 1201 case SIOCINQ: 1202 { 1203 unsigned int amount = first_packet_length(sk); 1204 1205 if (amount) 1206 /* 1207 * We will only return the amount 1208 * of this packet since that is all 1209 * that will be read. 1210 */ 1211 amount -= sizeof(struct udphdr); 1212 1213 return put_user(amount, (int __user *)arg); 1214 } 1215 1216 default: 1217 return -ENOIOCTLCMD; 1218 } 1219 1220 return 0; 1221 } 1222 EXPORT_SYMBOL(udp_ioctl); 1223 1224 /* 1225 * This should be easy, if there is something there we 1226 * return it, otherwise we block. 1227 */ 1228 1229 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1230 size_t len, int noblock, int flags, int *addr_len) 1231 { 1232 struct inet_sock *inet = inet_sk(sk); 1233 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 1234 struct sk_buff *skb; 1235 unsigned int ulen, copied; 1236 int peeked, off = 0; 1237 int err; 1238 int is_udplite = IS_UDPLITE(sk); 1239 bool slow; 1240 1241 if (flags & MSG_ERRQUEUE) 1242 return ip_recv_error(sk, msg, len, addr_len); 1243 1244 try_again: 1245 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1246 &peeked, &off, &err); 1247 if (!skb) 1248 goto out; 1249 1250 ulen = skb->len - sizeof(struct udphdr); 1251 copied = len; 1252 if (copied > ulen) 1253 copied = ulen; 1254 else if (copied < ulen) 1255 msg->msg_flags |= MSG_TRUNC; 1256 1257 /* 1258 * If checksum is needed at all, try to do it while copying the 1259 * data. If the data is truncated, or if we only want a partial 1260 * coverage checksum (UDP-Lite), do it before the copy. 1261 */ 1262 1263 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1264 if (udp_lib_checksum_complete(skb)) 1265 goto csum_copy_err; 1266 } 1267 1268 if (skb_csum_unnecessary(skb)) 1269 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1270 msg->msg_iov, copied); 1271 else { 1272 err = skb_copy_and_csum_datagram_iovec(skb, 1273 sizeof(struct udphdr), 1274 msg->msg_iov); 1275 1276 if (err == -EINVAL) 1277 goto csum_copy_err; 1278 } 1279 1280 if (unlikely(err)) { 1281 trace_kfree_skb(skb, udp_recvmsg); 1282 if (!peeked) { 1283 atomic_inc(&sk->sk_drops); 1284 UDP_INC_STATS_USER(sock_net(sk), 1285 UDP_MIB_INERRORS, is_udplite); 1286 } 1287 goto out_free; 1288 } 1289 1290 if (!peeked) 1291 UDP_INC_STATS_USER(sock_net(sk), 1292 UDP_MIB_INDATAGRAMS, is_udplite); 1293 1294 sock_recv_ts_and_drops(msg, sk, skb); 1295 1296 /* Copy the address. */ 1297 if (sin) { 1298 sin->sin_family = AF_INET; 1299 sin->sin_port = udp_hdr(skb)->source; 1300 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1301 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1302 *addr_len = sizeof(*sin); 1303 } 1304 if (inet->cmsg_flags) 1305 ip_cmsg_recv(msg, skb); 1306 1307 err = copied; 1308 if (flags & MSG_TRUNC) 1309 err = ulen; 1310 1311 out_free: 1312 skb_free_datagram_locked(sk, skb); 1313 out: 1314 return err; 1315 1316 csum_copy_err: 1317 slow = lock_sock_fast(sk); 1318 if (!skb_kill_datagram(sk, skb, flags)) { 1319 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1320 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1321 } 1322 unlock_sock_fast(sk, slow); 1323 1324 if (noblock) 1325 return -EAGAIN; 1326 1327 /* starting over for a new packet */ 1328 msg->msg_flags &= ~MSG_TRUNC; 1329 goto try_again; 1330 } 1331 1332 1333 int udp_disconnect(struct sock *sk, int flags) 1334 { 1335 struct inet_sock *inet = inet_sk(sk); 1336 /* 1337 * 1003.1g - break association. 1338 */ 1339 1340 sk->sk_state = TCP_CLOSE; 1341 inet->inet_daddr = 0; 1342 inet->inet_dport = 0; 1343 sock_rps_reset_rxhash(sk); 1344 sk->sk_bound_dev_if = 0; 1345 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1346 inet_reset_saddr(sk); 1347 1348 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1349 sk->sk_prot->unhash(sk); 1350 inet->inet_sport = 0; 1351 } 1352 sk_dst_reset(sk); 1353 return 0; 1354 } 1355 EXPORT_SYMBOL(udp_disconnect); 1356 1357 void udp_lib_unhash(struct sock *sk) 1358 { 1359 if (sk_hashed(sk)) { 1360 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1361 struct udp_hslot *hslot, *hslot2; 1362 1363 hslot = udp_hashslot(udptable, sock_net(sk), 1364 udp_sk(sk)->udp_port_hash); 1365 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1366 1367 spin_lock_bh(&hslot->lock); 1368 if (sk_nulls_del_node_init_rcu(sk)) { 1369 hslot->count--; 1370 inet_sk(sk)->inet_num = 0; 1371 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1372 1373 spin_lock(&hslot2->lock); 1374 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1375 hslot2->count--; 1376 spin_unlock(&hslot2->lock); 1377 } 1378 spin_unlock_bh(&hslot->lock); 1379 } 1380 } 1381 EXPORT_SYMBOL(udp_lib_unhash); 1382 1383 /* 1384 * inet_rcv_saddr was changed, we must rehash secondary hash 1385 */ 1386 void udp_lib_rehash(struct sock *sk, u16 newhash) 1387 { 1388 if (sk_hashed(sk)) { 1389 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1390 struct udp_hslot *hslot, *hslot2, *nhslot2; 1391 1392 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1393 nhslot2 = udp_hashslot2(udptable, newhash); 1394 udp_sk(sk)->udp_portaddr_hash = newhash; 1395 if (hslot2 != nhslot2) { 1396 hslot = udp_hashslot(udptable, sock_net(sk), 1397 udp_sk(sk)->udp_port_hash); 1398 /* we must lock primary chain too */ 1399 spin_lock_bh(&hslot->lock); 1400 1401 spin_lock(&hslot2->lock); 1402 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1403 hslot2->count--; 1404 spin_unlock(&hslot2->lock); 1405 1406 spin_lock(&nhslot2->lock); 1407 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1408 &nhslot2->head); 1409 nhslot2->count++; 1410 spin_unlock(&nhslot2->lock); 1411 1412 spin_unlock_bh(&hslot->lock); 1413 } 1414 } 1415 } 1416 EXPORT_SYMBOL(udp_lib_rehash); 1417 1418 static void udp_v4_rehash(struct sock *sk) 1419 { 1420 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1421 inet_sk(sk)->inet_rcv_saddr, 1422 inet_sk(sk)->inet_num); 1423 udp_lib_rehash(sk, new_hash); 1424 } 1425 1426 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1427 { 1428 int rc; 1429 1430 if (inet_sk(sk)->inet_daddr) { 1431 sock_rps_save_rxhash(sk, skb); 1432 sk_mark_napi_id(sk, skb); 1433 } 1434 1435 rc = sock_queue_rcv_skb(sk, skb); 1436 if (rc < 0) { 1437 int is_udplite = IS_UDPLITE(sk); 1438 1439 /* Note that an ENOMEM error is charged twice */ 1440 if (rc == -ENOMEM) 1441 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1442 is_udplite); 1443 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1444 kfree_skb(skb); 1445 trace_udp_fail_queue_rcv_skb(rc, sk); 1446 return -1; 1447 } 1448 1449 return 0; 1450 1451 } 1452 1453 static struct static_key udp_encap_needed __read_mostly; 1454 void udp_encap_enable(void) 1455 { 1456 if (!static_key_enabled(&udp_encap_needed)) 1457 static_key_slow_inc(&udp_encap_needed); 1458 } 1459 EXPORT_SYMBOL(udp_encap_enable); 1460 1461 /* returns: 1462 * -1: error 1463 * 0: success 1464 * >0: "udp encap" protocol resubmission 1465 * 1466 * Note that in the success and error cases, the skb is assumed to 1467 * have either been requeued or freed. 1468 */ 1469 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1470 { 1471 struct udp_sock *up = udp_sk(sk); 1472 int rc; 1473 int is_udplite = IS_UDPLITE(sk); 1474 1475 /* 1476 * Charge it to the socket, dropping if the queue is full. 1477 */ 1478 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1479 goto drop; 1480 nf_reset(skb); 1481 1482 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1483 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1484 1485 /* 1486 * This is an encapsulation socket so pass the skb to 1487 * the socket's udp_encap_rcv() hook. Otherwise, just 1488 * fall through and pass this up the UDP socket. 1489 * up->encap_rcv() returns the following value: 1490 * =0 if skb was successfully passed to the encap 1491 * handler or was discarded by it. 1492 * >0 if skb should be passed on to UDP. 1493 * <0 if skb should be resubmitted as proto -N 1494 */ 1495 1496 /* if we're overly short, let UDP handle it */ 1497 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1498 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { 1499 int ret; 1500 1501 ret = encap_rcv(sk, skb); 1502 if (ret <= 0) { 1503 UDP_INC_STATS_BH(sock_net(sk), 1504 UDP_MIB_INDATAGRAMS, 1505 is_udplite); 1506 return -ret; 1507 } 1508 } 1509 1510 /* FALLTHROUGH -- it's a UDP Packet */ 1511 } 1512 1513 /* 1514 * UDP-Lite specific tests, ignored on UDP sockets 1515 */ 1516 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1517 1518 /* 1519 * MIB statistics other than incrementing the error count are 1520 * disabled for the following two types of errors: these depend 1521 * on the application settings, not on the functioning of the 1522 * protocol stack as such. 1523 * 1524 * RFC 3828 here recommends (sec 3.3): "There should also be a 1525 * way ... to ... at least let the receiving application block 1526 * delivery of packets with coverage values less than a value 1527 * provided by the application." 1528 */ 1529 if (up->pcrlen == 0) { /* full coverage was set */ 1530 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n", 1531 UDP_SKB_CB(skb)->cscov, skb->len); 1532 goto drop; 1533 } 1534 /* The next case involves violating the min. coverage requested 1535 * by the receiver. This is subtle: if receiver wants x and x is 1536 * greater than the buffersize/MTU then receiver will complain 1537 * that it wants x while sender emits packets of smaller size y. 1538 * Therefore the above ...()->partial_cov statement is essential. 1539 */ 1540 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1541 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n", 1542 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1543 goto drop; 1544 } 1545 } 1546 1547 if (rcu_access_pointer(sk->sk_filter) && 1548 udp_lib_checksum_complete(skb)) 1549 goto csum_error; 1550 1551 1552 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) 1553 goto drop; 1554 1555 rc = 0; 1556 1557 ipv4_pktinfo_prepare(sk, skb); 1558 bh_lock_sock(sk); 1559 if (!sock_owned_by_user(sk)) 1560 rc = __udp_queue_rcv_skb(sk, skb); 1561 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1562 bh_unlock_sock(sk); 1563 goto drop; 1564 } 1565 bh_unlock_sock(sk); 1566 1567 return rc; 1568 1569 csum_error: 1570 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1571 drop: 1572 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1573 atomic_inc(&sk->sk_drops); 1574 kfree_skb(skb); 1575 return -1; 1576 } 1577 1578 1579 static void flush_stack(struct sock **stack, unsigned int count, 1580 struct sk_buff *skb, unsigned int final) 1581 { 1582 unsigned int i; 1583 struct sk_buff *skb1 = NULL; 1584 struct sock *sk; 1585 1586 for (i = 0; i < count; i++) { 1587 sk = stack[i]; 1588 if (likely(skb1 == NULL)) 1589 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1590 1591 if (!skb1) { 1592 atomic_inc(&sk->sk_drops); 1593 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1594 IS_UDPLITE(sk)); 1595 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1596 IS_UDPLITE(sk)); 1597 } 1598 1599 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1600 skb1 = NULL; 1601 } 1602 if (unlikely(skb1)) 1603 kfree_skb(skb1); 1604 } 1605 1606 static void udp_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) 1607 { 1608 struct dst_entry *dst = skb_dst(skb); 1609 1610 dst_hold(dst); 1611 sk->sk_rx_dst = dst; 1612 } 1613 1614 /* 1615 * Multicasts and broadcasts go to each listener. 1616 * 1617 * Note: called only from the BH handler context. 1618 */ 1619 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1620 struct udphdr *uh, 1621 __be32 saddr, __be32 daddr, 1622 struct udp_table *udptable) 1623 { 1624 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1625 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1626 int dif; 1627 unsigned int i, count = 0; 1628 1629 spin_lock(&hslot->lock); 1630 sk = sk_nulls_head(&hslot->head); 1631 dif = skb->dev->ifindex; 1632 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1633 while (sk) { 1634 stack[count++] = sk; 1635 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1636 daddr, uh->source, saddr, dif); 1637 if (unlikely(count == ARRAY_SIZE(stack))) { 1638 if (!sk) 1639 break; 1640 flush_stack(stack, count, skb, ~0); 1641 count = 0; 1642 } 1643 } 1644 /* 1645 * before releasing chain lock, we must take a reference on sockets 1646 */ 1647 for (i = 0; i < count; i++) 1648 sock_hold(stack[i]); 1649 1650 spin_unlock(&hslot->lock); 1651 1652 /* 1653 * do the slow work with no lock held 1654 */ 1655 if (count) { 1656 flush_stack(stack, count, skb, count - 1); 1657 1658 for (i = 0; i < count; i++) 1659 sock_put(stack[i]); 1660 } else { 1661 kfree_skb(skb); 1662 } 1663 return 0; 1664 } 1665 1666 /* Initialize UDP checksum. If exited with zero value (success), 1667 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1668 * Otherwise, csum completion requires chacksumming packet body, 1669 * including udp header and folding it to skb->csum. 1670 */ 1671 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1672 int proto) 1673 { 1674 const struct iphdr *iph; 1675 int err; 1676 1677 UDP_SKB_CB(skb)->partial_cov = 0; 1678 UDP_SKB_CB(skb)->cscov = skb->len; 1679 1680 if (proto == IPPROTO_UDPLITE) { 1681 err = udplite_checksum_init(skb, uh); 1682 if (err) 1683 return err; 1684 } 1685 1686 iph = ip_hdr(skb); 1687 if (uh->check == 0) { 1688 skb->ip_summed = CHECKSUM_UNNECESSARY; 1689 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1690 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1691 proto, skb->csum)) 1692 skb->ip_summed = CHECKSUM_UNNECESSARY; 1693 } 1694 if (!skb_csum_unnecessary(skb)) 1695 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1696 skb->len, proto, 0); 1697 /* Probably, we should checksum udp header (it should be in cache 1698 * in any case) and data in tiny packets (< rx copybreak). 1699 */ 1700 1701 return 0; 1702 } 1703 1704 /* 1705 * All we need to do is get the socket, and then do a checksum. 1706 */ 1707 1708 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1709 int proto) 1710 { 1711 struct sock *sk; 1712 struct udphdr *uh; 1713 unsigned short ulen; 1714 struct rtable *rt = skb_rtable(skb); 1715 __be32 saddr, daddr; 1716 struct net *net = dev_net(skb->dev); 1717 1718 /* 1719 * Validate the packet. 1720 */ 1721 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1722 goto drop; /* No space for header. */ 1723 1724 uh = udp_hdr(skb); 1725 ulen = ntohs(uh->len); 1726 saddr = ip_hdr(skb)->saddr; 1727 daddr = ip_hdr(skb)->daddr; 1728 1729 if (ulen > skb->len) 1730 goto short_packet; 1731 1732 if (proto == IPPROTO_UDP) { 1733 /* UDP validates ulen. */ 1734 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1735 goto short_packet; 1736 uh = udp_hdr(skb); 1737 } 1738 1739 if (udp4_csum_init(skb, uh, proto)) 1740 goto csum_error; 1741 1742 if (skb->sk) { 1743 int ret; 1744 sk = skb->sk; 1745 1746 if (unlikely(sk->sk_rx_dst == NULL)) 1747 udp_sk_rx_dst_set(sk, skb); 1748 1749 ret = udp_queue_rcv_skb(sk, skb); 1750 1751 /* a return value > 0 means to resubmit the input, but 1752 * it wants the return to be -protocol, or 0 1753 */ 1754 if (ret > 0) 1755 return -ret; 1756 return 0; 1757 } else { 1758 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1759 return __udp4_lib_mcast_deliver(net, skb, uh, 1760 saddr, daddr, udptable); 1761 1762 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1763 } 1764 1765 if (sk != NULL) { 1766 int ret; 1767 1768 ret = udp_queue_rcv_skb(sk, skb); 1769 sock_put(sk); 1770 1771 /* a return value > 0 means to resubmit the input, but 1772 * it wants the return to be -protocol, or 0 1773 */ 1774 if (ret > 0) 1775 return -ret; 1776 return 0; 1777 } 1778 1779 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1780 goto drop; 1781 nf_reset(skb); 1782 1783 /* No socket. Drop packet silently, if checksum is wrong */ 1784 if (udp_lib_checksum_complete(skb)) 1785 goto csum_error; 1786 1787 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1788 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1789 1790 /* 1791 * Hmm. We got an UDP packet to a port to which we 1792 * don't wanna listen. Ignore it. 1793 */ 1794 kfree_skb(skb); 1795 return 0; 1796 1797 short_packet: 1798 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1799 proto == IPPROTO_UDPLITE ? "Lite" : "", 1800 &saddr, ntohs(uh->source), 1801 ulen, skb->len, 1802 &daddr, ntohs(uh->dest)); 1803 goto drop; 1804 1805 csum_error: 1806 /* 1807 * RFC1122: OK. Discards the bad packet silently (as far as 1808 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1809 */ 1810 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1811 proto == IPPROTO_UDPLITE ? "Lite" : "", 1812 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1813 ulen); 1814 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1815 drop: 1816 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1817 kfree_skb(skb); 1818 return 0; 1819 } 1820 1821 /* We can only early demux multicast if there is a single matching socket. 1822 * If more than one socket found returns NULL 1823 */ 1824 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1825 __be16 loc_port, __be32 loc_addr, 1826 __be16 rmt_port, __be32 rmt_addr, 1827 int dif) 1828 { 1829 struct sock *sk, *result; 1830 struct hlist_nulls_node *node; 1831 unsigned short hnum = ntohs(loc_port); 1832 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1833 struct udp_hslot *hslot = &udp_table.hash[slot]; 1834 1835 rcu_read_lock(); 1836 begin: 1837 count = 0; 1838 result = NULL; 1839 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1840 if (__udp_is_mcast_sock(net, sk, 1841 loc_port, loc_addr, 1842 rmt_port, rmt_addr, 1843 dif, hnum)) { 1844 result = sk; 1845 ++count; 1846 } 1847 } 1848 /* 1849 * if the nulls value we got at the end of this lookup is 1850 * not the expected one, we must restart lookup. 1851 * We probably met an item that was moved to another chain. 1852 */ 1853 if (get_nulls_value(node) != slot) 1854 goto begin; 1855 1856 if (result) { 1857 if (count != 1 || 1858 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1859 result = NULL; 1860 else if (unlikely(!__udp_is_mcast_sock(net, result, 1861 loc_port, loc_addr, 1862 rmt_port, rmt_addr, 1863 dif, hnum))) { 1864 sock_put(result); 1865 result = NULL; 1866 } 1867 } 1868 rcu_read_unlock(); 1869 return result; 1870 } 1871 1872 /* For unicast we should only early demux connected sockets or we can 1873 * break forwarding setups. The chains here can be long so only check 1874 * if the first socket is an exact match and if not move on. 1875 */ 1876 static struct sock *__udp4_lib_demux_lookup(struct net *net, 1877 __be16 loc_port, __be32 loc_addr, 1878 __be16 rmt_port, __be32 rmt_addr, 1879 int dif) 1880 { 1881 struct sock *sk, *result; 1882 struct hlist_nulls_node *node; 1883 unsigned short hnum = ntohs(loc_port); 1884 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1885 unsigned int slot2 = hash2 & udp_table.mask; 1886 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1887 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr) 1888 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1889 1890 rcu_read_lock(); 1891 result = NULL; 1892 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 1893 if (INET_MATCH(sk, net, acookie, 1894 rmt_addr, loc_addr, ports, dif)) 1895 result = sk; 1896 /* Only check first socket in chain */ 1897 break; 1898 } 1899 1900 if (result) { 1901 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1902 result = NULL; 1903 else if (unlikely(!INET_MATCH(sk, net, acookie, 1904 rmt_addr, loc_addr, 1905 ports, dif))) { 1906 sock_put(result); 1907 result = NULL; 1908 } 1909 } 1910 rcu_read_unlock(); 1911 return result; 1912 } 1913 1914 void udp_v4_early_demux(struct sk_buff *skb) 1915 { 1916 const struct iphdr *iph = ip_hdr(skb); 1917 const struct udphdr *uh = udp_hdr(skb); 1918 struct sock *sk; 1919 struct dst_entry *dst; 1920 struct net *net = dev_net(skb->dev); 1921 int dif = skb->dev->ifindex; 1922 1923 /* validate the packet */ 1924 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1925 return; 1926 1927 if (skb->pkt_type == PACKET_BROADCAST || 1928 skb->pkt_type == PACKET_MULTICAST) 1929 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1930 uh->source, iph->saddr, dif); 1931 else if (skb->pkt_type == PACKET_HOST) 1932 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1933 uh->source, iph->saddr, dif); 1934 else 1935 return; 1936 1937 if (!sk) 1938 return; 1939 1940 skb->sk = sk; 1941 skb->destructor = sock_edemux; 1942 dst = sk->sk_rx_dst; 1943 1944 if (dst) 1945 dst = dst_check(dst, 0); 1946 if (dst) 1947 skb_dst_set_noref(skb, dst); 1948 } 1949 1950 int udp_rcv(struct sk_buff *skb) 1951 { 1952 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1953 } 1954 1955 void udp_destroy_sock(struct sock *sk) 1956 { 1957 struct udp_sock *up = udp_sk(sk); 1958 bool slow = lock_sock_fast(sk); 1959 udp_flush_pending_frames(sk); 1960 unlock_sock_fast(sk, slow); 1961 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1962 void (*encap_destroy)(struct sock *sk); 1963 encap_destroy = ACCESS_ONCE(up->encap_destroy); 1964 if (encap_destroy) 1965 encap_destroy(sk); 1966 } 1967 } 1968 1969 /* 1970 * Socket option code for UDP 1971 */ 1972 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1973 char __user *optval, unsigned int optlen, 1974 int (*push_pending_frames)(struct sock *)) 1975 { 1976 struct udp_sock *up = udp_sk(sk); 1977 int val; 1978 int err = 0; 1979 int is_udplite = IS_UDPLITE(sk); 1980 1981 if (optlen < sizeof(int)) 1982 return -EINVAL; 1983 1984 if (get_user(val, (int __user *)optval)) 1985 return -EFAULT; 1986 1987 switch (optname) { 1988 case UDP_CORK: 1989 if (val != 0) { 1990 up->corkflag = 1; 1991 } else { 1992 up->corkflag = 0; 1993 lock_sock(sk); 1994 (*push_pending_frames)(sk); 1995 release_sock(sk); 1996 } 1997 break; 1998 1999 case UDP_ENCAP: 2000 switch (val) { 2001 case 0: 2002 case UDP_ENCAP_ESPINUDP: 2003 case UDP_ENCAP_ESPINUDP_NON_IKE: 2004 up->encap_rcv = xfrm4_udp_encap_rcv; 2005 /* FALLTHROUGH */ 2006 case UDP_ENCAP_L2TPINUDP: 2007 up->encap_type = val; 2008 udp_encap_enable(); 2009 break; 2010 default: 2011 err = -ENOPROTOOPT; 2012 break; 2013 } 2014 break; 2015 2016 /* 2017 * UDP-Lite's partial checksum coverage (RFC 3828). 2018 */ 2019 /* The sender sets actual checksum coverage length via this option. 2020 * The case coverage > packet length is handled by send module. */ 2021 case UDPLITE_SEND_CSCOV: 2022 if (!is_udplite) /* Disable the option on UDP sockets */ 2023 return -ENOPROTOOPT; 2024 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2025 val = 8; 2026 else if (val > USHRT_MAX) 2027 val = USHRT_MAX; 2028 up->pcslen = val; 2029 up->pcflag |= UDPLITE_SEND_CC; 2030 break; 2031 2032 /* The receiver specifies a minimum checksum coverage value. To make 2033 * sense, this should be set to at least 8 (as done below). If zero is 2034 * used, this again means full checksum coverage. */ 2035 case UDPLITE_RECV_CSCOV: 2036 if (!is_udplite) /* Disable the option on UDP sockets */ 2037 return -ENOPROTOOPT; 2038 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2039 val = 8; 2040 else if (val > USHRT_MAX) 2041 val = USHRT_MAX; 2042 up->pcrlen = val; 2043 up->pcflag |= UDPLITE_RECV_CC; 2044 break; 2045 2046 default: 2047 err = -ENOPROTOOPT; 2048 break; 2049 } 2050 2051 return err; 2052 } 2053 EXPORT_SYMBOL(udp_lib_setsockopt); 2054 2055 int udp_setsockopt(struct sock *sk, int level, int optname, 2056 char __user *optval, unsigned int optlen) 2057 { 2058 if (level == SOL_UDP || level == SOL_UDPLITE) 2059 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2060 udp_push_pending_frames); 2061 return ip_setsockopt(sk, level, optname, optval, optlen); 2062 } 2063 2064 #ifdef CONFIG_COMPAT 2065 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2066 char __user *optval, unsigned int optlen) 2067 { 2068 if (level == SOL_UDP || level == SOL_UDPLITE) 2069 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2070 udp_push_pending_frames); 2071 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2072 } 2073 #endif 2074 2075 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2076 char __user *optval, int __user *optlen) 2077 { 2078 struct udp_sock *up = udp_sk(sk); 2079 int val, len; 2080 2081 if (get_user(len, optlen)) 2082 return -EFAULT; 2083 2084 len = min_t(unsigned int, len, sizeof(int)); 2085 2086 if (len < 0) 2087 return -EINVAL; 2088 2089 switch (optname) { 2090 case UDP_CORK: 2091 val = up->corkflag; 2092 break; 2093 2094 case UDP_ENCAP: 2095 val = up->encap_type; 2096 break; 2097 2098 /* The following two cannot be changed on UDP sockets, the return is 2099 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2100 case UDPLITE_SEND_CSCOV: 2101 val = up->pcslen; 2102 break; 2103 2104 case UDPLITE_RECV_CSCOV: 2105 val = up->pcrlen; 2106 break; 2107 2108 default: 2109 return -ENOPROTOOPT; 2110 } 2111 2112 if (put_user(len, optlen)) 2113 return -EFAULT; 2114 if (copy_to_user(optval, &val, len)) 2115 return -EFAULT; 2116 return 0; 2117 } 2118 EXPORT_SYMBOL(udp_lib_getsockopt); 2119 2120 int udp_getsockopt(struct sock *sk, int level, int optname, 2121 char __user *optval, int __user *optlen) 2122 { 2123 if (level == SOL_UDP || level == SOL_UDPLITE) 2124 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2125 return ip_getsockopt(sk, level, optname, optval, optlen); 2126 } 2127 2128 #ifdef CONFIG_COMPAT 2129 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2130 char __user *optval, int __user *optlen) 2131 { 2132 if (level == SOL_UDP || level == SOL_UDPLITE) 2133 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2134 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2135 } 2136 #endif 2137 /** 2138 * udp_poll - wait for a UDP event. 2139 * @file - file struct 2140 * @sock - socket 2141 * @wait - poll table 2142 * 2143 * This is same as datagram poll, except for the special case of 2144 * blocking sockets. If application is using a blocking fd 2145 * and a packet with checksum error is in the queue; 2146 * then it could get return from select indicating data available 2147 * but then block when reading it. Add special case code 2148 * to work around these arguably broken applications. 2149 */ 2150 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2151 { 2152 unsigned int mask = datagram_poll(file, sock, wait); 2153 struct sock *sk = sock->sk; 2154 2155 sock_rps_record_flow(sk); 2156 2157 /* Check for false positives due to checksum errors */ 2158 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2159 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2160 mask &= ~(POLLIN | POLLRDNORM); 2161 2162 return mask; 2163 2164 } 2165 EXPORT_SYMBOL(udp_poll); 2166 2167 struct proto udp_prot = { 2168 .name = "UDP", 2169 .owner = THIS_MODULE, 2170 .close = udp_lib_close, 2171 .connect = ip4_datagram_connect, 2172 .disconnect = udp_disconnect, 2173 .ioctl = udp_ioctl, 2174 .destroy = udp_destroy_sock, 2175 .setsockopt = udp_setsockopt, 2176 .getsockopt = udp_getsockopt, 2177 .sendmsg = udp_sendmsg, 2178 .recvmsg = udp_recvmsg, 2179 .sendpage = udp_sendpage, 2180 .backlog_rcv = __udp_queue_rcv_skb, 2181 .release_cb = ip4_datagram_release_cb, 2182 .hash = udp_lib_hash, 2183 .unhash = udp_lib_unhash, 2184 .rehash = udp_v4_rehash, 2185 .get_port = udp_v4_get_port, 2186 .memory_allocated = &udp_memory_allocated, 2187 .sysctl_mem = sysctl_udp_mem, 2188 .sysctl_wmem = &sysctl_udp_wmem_min, 2189 .sysctl_rmem = &sysctl_udp_rmem_min, 2190 .obj_size = sizeof(struct udp_sock), 2191 .slab_flags = SLAB_DESTROY_BY_RCU, 2192 .h.udp_table = &udp_table, 2193 #ifdef CONFIG_COMPAT 2194 .compat_setsockopt = compat_udp_setsockopt, 2195 .compat_getsockopt = compat_udp_getsockopt, 2196 #endif 2197 .clear_sk = sk_prot_clear_portaddr_nulls, 2198 }; 2199 EXPORT_SYMBOL(udp_prot); 2200 2201 /* ------------------------------------------------------------------------ */ 2202 #ifdef CONFIG_PROC_FS 2203 2204 static struct sock *udp_get_first(struct seq_file *seq, int start) 2205 { 2206 struct sock *sk; 2207 struct udp_iter_state *state = seq->private; 2208 struct net *net = seq_file_net(seq); 2209 2210 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2211 ++state->bucket) { 2212 struct hlist_nulls_node *node; 2213 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2214 2215 if (hlist_nulls_empty(&hslot->head)) 2216 continue; 2217 2218 spin_lock_bh(&hslot->lock); 2219 sk_nulls_for_each(sk, node, &hslot->head) { 2220 if (!net_eq(sock_net(sk), net)) 2221 continue; 2222 if (sk->sk_family == state->family) 2223 goto found; 2224 } 2225 spin_unlock_bh(&hslot->lock); 2226 } 2227 sk = NULL; 2228 found: 2229 return sk; 2230 } 2231 2232 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2233 { 2234 struct udp_iter_state *state = seq->private; 2235 struct net *net = seq_file_net(seq); 2236 2237 do { 2238 sk = sk_nulls_next(sk); 2239 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2240 2241 if (!sk) { 2242 if (state->bucket <= state->udp_table->mask) 2243 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2244 return udp_get_first(seq, state->bucket + 1); 2245 } 2246 return sk; 2247 } 2248 2249 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2250 { 2251 struct sock *sk = udp_get_first(seq, 0); 2252 2253 if (sk) 2254 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2255 --pos; 2256 return pos ? NULL : sk; 2257 } 2258 2259 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2260 { 2261 struct udp_iter_state *state = seq->private; 2262 state->bucket = MAX_UDP_PORTS; 2263 2264 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2265 } 2266 2267 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2268 { 2269 struct sock *sk; 2270 2271 if (v == SEQ_START_TOKEN) 2272 sk = udp_get_idx(seq, 0); 2273 else 2274 sk = udp_get_next(seq, v); 2275 2276 ++*pos; 2277 return sk; 2278 } 2279 2280 static void udp_seq_stop(struct seq_file *seq, void *v) 2281 { 2282 struct udp_iter_state *state = seq->private; 2283 2284 if (state->bucket <= state->udp_table->mask) 2285 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2286 } 2287 2288 int udp_seq_open(struct inode *inode, struct file *file) 2289 { 2290 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2291 struct udp_iter_state *s; 2292 int err; 2293 2294 err = seq_open_net(inode, file, &afinfo->seq_ops, 2295 sizeof(struct udp_iter_state)); 2296 if (err < 0) 2297 return err; 2298 2299 s = ((struct seq_file *)file->private_data)->private; 2300 s->family = afinfo->family; 2301 s->udp_table = afinfo->udp_table; 2302 return err; 2303 } 2304 EXPORT_SYMBOL(udp_seq_open); 2305 2306 /* ------------------------------------------------------------------------ */ 2307 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2308 { 2309 struct proc_dir_entry *p; 2310 int rc = 0; 2311 2312 afinfo->seq_ops.start = udp_seq_start; 2313 afinfo->seq_ops.next = udp_seq_next; 2314 afinfo->seq_ops.stop = udp_seq_stop; 2315 2316 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2317 afinfo->seq_fops, afinfo); 2318 if (!p) 2319 rc = -ENOMEM; 2320 return rc; 2321 } 2322 EXPORT_SYMBOL(udp_proc_register); 2323 2324 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2325 { 2326 remove_proc_entry(afinfo->name, net->proc_net); 2327 } 2328 EXPORT_SYMBOL(udp_proc_unregister); 2329 2330 /* ------------------------------------------------------------------------ */ 2331 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2332 int bucket) 2333 { 2334 struct inet_sock *inet = inet_sk(sp); 2335 __be32 dest = inet->inet_daddr; 2336 __be32 src = inet->inet_rcv_saddr; 2337 __u16 destp = ntohs(inet->inet_dport); 2338 __u16 srcp = ntohs(inet->inet_sport); 2339 2340 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2341 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2342 bucket, src, srcp, dest, destp, sp->sk_state, 2343 sk_wmem_alloc_get(sp), 2344 sk_rmem_alloc_get(sp), 2345 0, 0L, 0, 2346 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2347 0, sock_i_ino(sp), 2348 atomic_read(&sp->sk_refcnt), sp, 2349 atomic_read(&sp->sk_drops)); 2350 } 2351 2352 int udp4_seq_show(struct seq_file *seq, void *v) 2353 { 2354 seq_setwidth(seq, 127); 2355 if (v == SEQ_START_TOKEN) 2356 seq_puts(seq, " sl local_address rem_address st tx_queue " 2357 "rx_queue tr tm->when retrnsmt uid timeout " 2358 "inode ref pointer drops"); 2359 else { 2360 struct udp_iter_state *state = seq->private; 2361 2362 udp4_format_sock(v, seq, state->bucket); 2363 } 2364 seq_pad(seq, '\n'); 2365 return 0; 2366 } 2367 2368 static const struct file_operations udp_afinfo_seq_fops = { 2369 .owner = THIS_MODULE, 2370 .open = udp_seq_open, 2371 .read = seq_read, 2372 .llseek = seq_lseek, 2373 .release = seq_release_net 2374 }; 2375 2376 /* ------------------------------------------------------------------------ */ 2377 static struct udp_seq_afinfo udp4_seq_afinfo = { 2378 .name = "udp", 2379 .family = AF_INET, 2380 .udp_table = &udp_table, 2381 .seq_fops = &udp_afinfo_seq_fops, 2382 .seq_ops = { 2383 .show = udp4_seq_show, 2384 }, 2385 }; 2386 2387 static int __net_init udp4_proc_init_net(struct net *net) 2388 { 2389 return udp_proc_register(net, &udp4_seq_afinfo); 2390 } 2391 2392 static void __net_exit udp4_proc_exit_net(struct net *net) 2393 { 2394 udp_proc_unregister(net, &udp4_seq_afinfo); 2395 } 2396 2397 static struct pernet_operations udp4_net_ops = { 2398 .init = udp4_proc_init_net, 2399 .exit = udp4_proc_exit_net, 2400 }; 2401 2402 int __init udp4_proc_init(void) 2403 { 2404 return register_pernet_subsys(&udp4_net_ops); 2405 } 2406 2407 void udp4_proc_exit(void) 2408 { 2409 unregister_pernet_subsys(&udp4_net_ops); 2410 } 2411 #endif /* CONFIG_PROC_FS */ 2412 2413 static __initdata unsigned long uhash_entries; 2414 static int __init set_uhash_entries(char *str) 2415 { 2416 ssize_t ret; 2417 2418 if (!str) 2419 return 0; 2420 2421 ret = kstrtoul(str, 0, &uhash_entries); 2422 if (ret) 2423 return 0; 2424 2425 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2426 uhash_entries = UDP_HTABLE_SIZE_MIN; 2427 return 1; 2428 } 2429 __setup("uhash_entries=", set_uhash_entries); 2430 2431 void __init udp_table_init(struct udp_table *table, const char *name) 2432 { 2433 unsigned int i; 2434 2435 table->hash = alloc_large_system_hash(name, 2436 2 * sizeof(struct udp_hslot), 2437 uhash_entries, 2438 21, /* one slot per 2 MB */ 2439 0, 2440 &table->log, 2441 &table->mask, 2442 UDP_HTABLE_SIZE_MIN, 2443 64 * 1024); 2444 2445 table->hash2 = table->hash + (table->mask + 1); 2446 for (i = 0; i <= table->mask; i++) { 2447 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2448 table->hash[i].count = 0; 2449 spin_lock_init(&table->hash[i].lock); 2450 } 2451 for (i = 0; i <= table->mask; i++) { 2452 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2453 table->hash2[i].count = 0; 2454 spin_lock_init(&table->hash2[i].lock); 2455 } 2456 } 2457 2458 void __init udp_init(void) 2459 { 2460 unsigned long limit; 2461 2462 udp_table_init(&udp_table, "UDP"); 2463 limit = nr_free_buffer_pages() / 8; 2464 limit = max(limit, 128UL); 2465 sysctl_udp_mem[0] = limit / 4 * 3; 2466 sysctl_udp_mem[1] = limit; 2467 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2468 2469 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2470 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2471 } 2472 2473 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb, 2474 netdev_features_t features) 2475 { 2476 struct sk_buff *segs = ERR_PTR(-EINVAL); 2477 int mac_len = skb->mac_len; 2478 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb); 2479 __be16 protocol = skb->protocol; 2480 netdev_features_t enc_features; 2481 int outer_hlen; 2482 2483 if (unlikely(!pskb_may_pull(skb, tnl_hlen))) 2484 goto out; 2485 2486 skb->encapsulation = 0; 2487 __skb_pull(skb, tnl_hlen); 2488 skb_reset_mac_header(skb); 2489 skb_set_network_header(skb, skb_inner_network_offset(skb)); 2490 skb->mac_len = skb_inner_network_offset(skb); 2491 skb->protocol = htons(ETH_P_TEB); 2492 2493 /* segment inner packet. */ 2494 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb); 2495 segs = skb_mac_gso_segment(skb, enc_features); 2496 if (!segs || IS_ERR(segs)) 2497 goto out; 2498 2499 outer_hlen = skb_tnl_header_len(skb); 2500 skb = segs; 2501 do { 2502 struct udphdr *uh; 2503 int udp_offset = outer_hlen - tnl_hlen; 2504 2505 skb_reset_inner_headers(skb); 2506 skb->encapsulation = 1; 2507 2508 skb->mac_len = mac_len; 2509 2510 skb_push(skb, outer_hlen); 2511 skb_reset_mac_header(skb); 2512 skb_set_network_header(skb, mac_len); 2513 skb_set_transport_header(skb, udp_offset); 2514 uh = udp_hdr(skb); 2515 uh->len = htons(skb->len - udp_offset); 2516 2517 /* csum segment if tunnel sets skb with csum. */ 2518 if (protocol == htons(ETH_P_IP) && unlikely(uh->check)) { 2519 struct iphdr *iph = ip_hdr(skb); 2520 2521 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 2522 skb->len - udp_offset, 2523 IPPROTO_UDP, 0); 2524 uh->check = csum_fold(skb_checksum(skb, udp_offset, 2525 skb->len - udp_offset, 0)); 2526 if (uh->check == 0) 2527 uh->check = CSUM_MANGLED_0; 2528 2529 } else if (protocol == htons(ETH_P_IPV6)) { 2530 struct ipv6hdr *ipv6h = ipv6_hdr(skb); 2531 u32 len = skb->len - udp_offset; 2532 2533 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, 2534 len, IPPROTO_UDP, 0); 2535 uh->check = csum_fold(skb_checksum(skb, udp_offset, len, 0)); 2536 if (uh->check == 0) 2537 uh->check = CSUM_MANGLED_0; 2538 skb->ip_summed = CHECKSUM_NONE; 2539 } 2540 2541 skb->protocol = protocol; 2542 } while ((skb = skb->next)); 2543 out: 2544 return segs; 2545 } 2546