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