1/* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The User Datagram Protocol (UDP). 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 12 * Hirokazu Takahashi, <taka@valinux.co.jp> 13 * 14 * Fixes: 15 * Alan Cox : verify_area() calls 16 * Alan Cox : stopped close while in use off icmp 17 * messages. Not a fix but a botch that 18 * for udp at least is 'valid'. 19 * Alan Cox : Fixed icmp handling properly 20 * Alan Cox : Correct error for oversized datagrams 21 * Alan Cox : Tidied select() semantics. 22 * Alan Cox : udp_err() fixed properly, also now 23 * select and read wake correctly on errors 24 * Alan Cox : udp_send verify_area moved to avoid mem leak 25 * Alan Cox : UDP can count its memory 26 * Alan Cox : send to an unknown connection causes 27 * an ECONNREFUSED off the icmp, but 28 * does NOT close. 29 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 31 * bug no longer crashes it. 32 * Fred Van Kempen : Net2e support for sk->broadcast. 33 * Alan Cox : Uses skb_free_datagram 34 * Alan Cox : Added get/set sockopt support. 35 * Alan Cox : Broadcasting without option set returns EACCES. 36 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 37 * Alan Cox : Use ip_tos and ip_ttl 38 * Alan Cox : SNMP Mibs 39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 40 * Matt Dillon : UDP length checks. 41 * Alan Cox : Smarter af_inet used properly. 42 * Alan Cox : Use new kernel side addressing. 43 * Alan Cox : Incorrect return on truncated datagram receive. 44 * Arnt Gulbrandsen : New udp_send and stuff 45 * Alan Cox : Cache last socket 46 * Alan Cox : Route cache 47 * Jon Peatfield : Minor efficiency fix to sendto(). 48 * Mike Shaver : RFC1122 checks. 49 * Alan Cox : Nonblocking error fix. 50 * Willy Konynenberg : Transparent proxying support. 51 * Mike McLagan : Routing by source 52 * David S. Miller : New socket lookup architecture. 53 * Last socket cache retained as it 54 * does have a high hit rate. 55 * Olaf Kirch : Don't linearise iovec on sendmsg. 56 * Andi Kleen : Some cleanups, cache destination entry 57 * for connect. 58 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 59 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 60 * return ENOTCONN for unconnected sockets (POSIX) 61 * Janos Farkas : don't deliver multi/broadcasts to a different 62 * bound-to-device socket 63 * Hirokazu Takahashi : HW checksumming for outgoing UDP 64 * datagrams. 65 * Hirokazu Takahashi : sendfile() on UDP works now. 66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 69 * a single port at the same time. 70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71 * James Chapman : Add L2TP encapsulation type. 72 * 73 * 74 * This program is free software; you can redistribute it and/or 75 * modify it under the terms of the GNU General Public License 76 * as published by the Free Software Foundation; either version 77 * 2 of the License, or (at your option) any later version. 78 */ 79 80#define pr_fmt(fmt) "UDP: " fmt 81 82#include <asm/uaccess.h> 83#include <asm/ioctls.h> 84#include <linux/bootmem.h> 85#include <linux/highmem.h> 86#include <linux/swap.h> 87#include <linux/types.h> 88#include <linux/fcntl.h> 89#include <linux/module.h> 90#include <linux/socket.h> 91#include <linux/sockios.h> 92#include <linux/igmp.h> 93#include <linux/inetdevice.h> 94#include <linux/in.h> 95#include <linux/errno.h> 96#include <linux/timer.h> 97#include <linux/mm.h> 98#include <linux/inet.h> 99#include <linux/netdevice.h> 100#include <linux/slab.h> 101#include <net/tcp_states.h> 102#include <linux/skbuff.h> 103#include <linux/netdevice.h> 104#include <linux/proc_fs.h> 105#include <linux/seq_file.h> 106#include <net/net_namespace.h> 107#include <net/icmp.h> 108#include <net/inet_hashtables.h> 109#include <net/route.h> 110#include <net/checksum.h> 111#include <net/xfrm.h> 112#include <trace/events/udp.h> 113#include <linux/static_key.h> 114#include <trace/events/skb.h> 115#include <net/busy_poll.h> 116#include "udp_impl.h" 117 118struct udp_table udp_table __read_mostly; 119EXPORT_SYMBOL(udp_table); 120 121long sysctl_udp_mem[3] __read_mostly; 122EXPORT_SYMBOL(sysctl_udp_mem); 123 124int sysctl_udp_rmem_min __read_mostly; 125EXPORT_SYMBOL(sysctl_udp_rmem_min); 126 127int sysctl_udp_wmem_min __read_mostly; 128EXPORT_SYMBOL(sysctl_udp_wmem_min); 129 130atomic_long_t udp_memory_allocated; 131EXPORT_SYMBOL(udp_memory_allocated); 132 133#define MAX_UDP_PORTS 65536 134#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 135 136static int udp_lib_lport_inuse(struct net *net, __u16 num, 137 const struct udp_hslot *hslot, 138 unsigned long *bitmap, 139 struct sock *sk, 140 int (*saddr_comp)(const struct sock *sk1, 141 const struct sock *sk2), 142 unsigned int log) 143{ 144 struct sock *sk2; 145 struct hlist_nulls_node *node; 146 kuid_t uid = sock_i_uid(sk); 147 148 sk_nulls_for_each(sk2, node, &hslot->head) { 149 if (net_eq(sock_net(sk2), net) && 150 sk2 != sk && 151 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 152 (!sk2->sk_reuse || !sk->sk_reuse) && 153 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 154 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 155 (!sk2->sk_reuseport || !sk->sk_reuseport || 156 !uid_eq(uid, sock_i_uid(sk2))) && 157 saddr_comp(sk, sk2)) { 158 if (!bitmap) 159 return 1; 160 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap); 161 } 162 } 163 return 0; 164} 165 166/* 167 * Note: we still hold spinlock of primary hash chain, so no other writer 168 * can insert/delete a socket with local_port == num 169 */ 170static int udp_lib_lport_inuse2(struct net *net, __u16 num, 171 struct udp_hslot *hslot2, 172 struct sock *sk, 173 int (*saddr_comp)(const struct sock *sk1, 174 const struct sock *sk2)) 175{ 176 struct sock *sk2; 177 struct hlist_nulls_node *node; 178 kuid_t uid = sock_i_uid(sk); 179 int res = 0; 180 181 spin_lock(&hslot2->lock); 182 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) { 183 if (net_eq(sock_net(sk2), net) && 184 sk2 != sk && 185 (udp_sk(sk2)->udp_port_hash == num) && 186 (!sk2->sk_reuse || !sk->sk_reuse) && 187 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 188 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 189 (!sk2->sk_reuseport || !sk->sk_reuseport || 190 !uid_eq(uid, sock_i_uid(sk2))) && 191 saddr_comp(sk, sk2)) { 192 res = 1; 193 break; 194 } 195 } 196 spin_unlock(&hslot2->lock); 197 return res; 198} 199 200/** 201 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 202 * 203 * @sk: socket struct in question 204 * @snum: port number to look up 205 * @saddr_comp: AF-dependent comparison of bound local IP addresses 206 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 207 * with NULL address 208 */ 209int udp_lib_get_port(struct sock *sk, unsigned short snum, 210 int (*saddr_comp)(const struct sock *sk1, 211 const struct sock *sk2), 212 unsigned int hash2_nulladdr) 213{ 214 struct udp_hslot *hslot, *hslot2; 215 struct udp_table *udptable = sk->sk_prot->h.udp_table; 216 int error = 1; 217 struct net *net = sock_net(sk); 218 219 if (!snum) { 220 int low, high, remaining; 221 unsigned int rand; 222 unsigned short first, last; 223 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 224 225 inet_get_local_port_range(net, &low, &high); 226 remaining = (high - low) + 1; 227 228 rand = prandom_u32(); 229 first = reciprocal_scale(rand, remaining) + low; 230 /* 231 * force rand to be an odd multiple of UDP_HTABLE_SIZE 232 */ 233 rand = (rand | 1) * (udptable->mask + 1); 234 last = first + udptable->mask + 1; 235 do { 236 hslot = udp_hashslot(udptable, net, first); 237 bitmap_zero(bitmap, PORTS_PER_CHAIN); 238 spin_lock_bh(&hslot->lock); 239 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 240 saddr_comp, udptable->log); 241 242 snum = first; 243 /* 244 * Iterate on all possible values of snum for this hash. 245 * Using steps of an odd multiple of UDP_HTABLE_SIZE 246 * give us randomization and full range coverage. 247 */ 248 do { 249 if (low <= snum && snum <= high && 250 !test_bit(snum >> udptable->log, bitmap) && 251 !inet_is_local_reserved_port(net, snum)) 252 goto found; 253 snum += rand; 254 } while (snum != first); 255 spin_unlock_bh(&hslot->lock); 256 } while (++first != last); 257 goto fail; 258 } else { 259 hslot = udp_hashslot(udptable, net, snum); 260 spin_lock_bh(&hslot->lock); 261 if (hslot->count > 10) { 262 int exist; 263 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 264 265 slot2 &= udptable->mask; 266 hash2_nulladdr &= udptable->mask; 267 268 hslot2 = udp_hashslot2(udptable, slot2); 269 if (hslot->count < hslot2->count) 270 goto scan_primary_hash; 271 272 exist = udp_lib_lport_inuse2(net, snum, hslot2, 273 sk, saddr_comp); 274 if (!exist && (hash2_nulladdr != slot2)) { 275 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 276 exist = udp_lib_lport_inuse2(net, snum, hslot2, 277 sk, saddr_comp); 278 } 279 if (exist) 280 goto fail_unlock; 281 else 282 goto found; 283 } 284scan_primary_hash: 285 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 286 saddr_comp, 0)) 287 goto fail_unlock; 288 } 289found: 290 inet_sk(sk)->inet_num = snum; 291 udp_sk(sk)->udp_port_hash = snum; 292 udp_sk(sk)->udp_portaddr_hash ^= snum; 293 if (sk_unhashed(sk)) { 294 sk_nulls_add_node_rcu(sk, &hslot->head); 295 hslot->count++; 296 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 297 298 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 299 spin_lock(&hslot2->lock); 300 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 301 &hslot2->head); 302 hslot2->count++; 303 spin_unlock(&hslot2->lock); 304 } 305 error = 0; 306fail_unlock: 307 spin_unlock_bh(&hslot->lock); 308fail: 309 return error; 310} 311EXPORT_SYMBOL(udp_lib_get_port); 312 313static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 314{ 315 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 316 317 return (!ipv6_only_sock(sk2) && 318 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 319 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 320} 321 322static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr, 323 unsigned int port) 324{ 325 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 326} 327 328int udp_v4_get_port(struct sock *sk, unsigned short snum) 329{ 330 unsigned int hash2_nulladdr = 331 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 332 unsigned int hash2_partial = 333 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 334 335 /* precompute partial secondary hash */ 336 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 337 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 338} 339 340static inline int compute_score(struct sock *sk, struct net *net, 341 __be32 saddr, unsigned short hnum, __be16 sport, 342 __be32 daddr, __be16 dport, int dif) 343{ 344 int score; 345 struct inet_sock *inet; 346 347 if (!net_eq(sock_net(sk), net) || 348 udp_sk(sk)->udp_port_hash != hnum || 349 ipv6_only_sock(sk)) 350 return -1; 351 352 score = (sk->sk_family == PF_INET) ? 2 : 1; 353 inet = inet_sk(sk); 354 355 if (inet->inet_rcv_saddr) { 356 if (inet->inet_rcv_saddr != daddr) 357 return -1; 358 score += 4; 359 } 360 361 if (inet->inet_daddr) { 362 if (inet->inet_daddr != saddr) 363 return -1; 364 score += 4; 365 } 366 367 if (inet->inet_dport) { 368 if (inet->inet_dport != sport) 369 return -1; 370 score += 4; 371 } 372 373 if (sk->sk_bound_dev_if) { 374 if (sk->sk_bound_dev_if != dif) 375 return -1; 376 score += 4; 377 } 378 379 return score; 380} 381 382/* 383 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 384 */ 385static inline int compute_score2(struct sock *sk, struct net *net, 386 __be32 saddr, __be16 sport, 387 __be32 daddr, unsigned int hnum, int dif) 388{ 389 int score; 390 struct inet_sock *inet; 391 392 if (!net_eq(sock_net(sk), net) || 393 ipv6_only_sock(sk)) 394 return -1; 395 396 inet = inet_sk(sk); 397 398 if (inet->inet_rcv_saddr != daddr || 399 inet->inet_num != hnum) 400 return -1; 401 402 score = (sk->sk_family == PF_INET) ? 2 : 1; 403 404 if (inet->inet_daddr) { 405 if (inet->inet_daddr != saddr) 406 return -1; 407 score += 4; 408 } 409 410 if (inet->inet_dport) { 411 if (inet->inet_dport != sport) 412 return -1; 413 score += 4; 414 } 415 416 if (sk->sk_bound_dev_if) { 417 if (sk->sk_bound_dev_if != dif) 418 return -1; 419 score += 4; 420 } 421 422 return score; 423} 424 425static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 426 const __u16 lport, const __be32 faddr, 427 const __be16 fport) 428{ 429 static u32 udp_ehash_secret __read_mostly; 430 431 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 432 433 return __inet_ehashfn(laddr, lport, faddr, fport, 434 udp_ehash_secret + net_hash_mix(net)); 435} 436 437/* called with read_rcu_lock() */ 438static struct sock *udp4_lib_lookup2(struct net *net, 439 __be32 saddr, __be16 sport, 440 __be32 daddr, unsigned int hnum, int dif, 441 struct udp_hslot *hslot2, unsigned int slot2) 442{ 443 struct sock *sk, *result; 444 struct hlist_nulls_node *node; 445 int score, badness, matches = 0, reuseport = 0; 446 u32 hash = 0; 447 448begin: 449 result = NULL; 450 badness = 0; 451 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 452 score = compute_score2(sk, net, saddr, sport, 453 daddr, hnum, dif); 454 if (score > badness) { 455 result = sk; 456 badness = score; 457 reuseport = sk->sk_reuseport; 458 if (reuseport) { 459 hash = udp_ehashfn(net, daddr, hnum, 460 saddr, sport); 461 matches = 1; 462 } 463 } else if (score == badness && reuseport) { 464 matches++; 465 if (reciprocal_scale(hash, matches) == 0) 466 result = sk; 467 hash = next_pseudo_random32(hash); 468 } 469 } 470 /* 471 * if the nulls value we got at the end of this lookup is 472 * not the expected one, we must restart lookup. 473 * We probably met an item that was moved to another chain. 474 */ 475 if (get_nulls_value(node) != slot2) 476 goto begin; 477 if (result) { 478 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 479 result = NULL; 480 else if (unlikely(compute_score2(result, net, saddr, sport, 481 daddr, hnum, dif) < badness)) { 482 sock_put(result); 483 goto begin; 484 } 485 } 486 return result; 487} 488 489/* UDP is nearly always wildcards out the wazoo, it makes no sense to try 490 * harder than this. -DaveM 491 */ 492struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 493 __be16 sport, __be32 daddr, __be16 dport, 494 int dif, struct udp_table *udptable) 495{ 496 struct sock *sk, *result; 497 struct hlist_nulls_node *node; 498 unsigned short hnum = ntohs(dport); 499 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 500 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 501 int score, badness, matches = 0, reuseport = 0; 502 u32 hash = 0; 503 504 rcu_read_lock(); 505 if (hslot->count > 10) { 506 hash2 = udp4_portaddr_hash(net, daddr, hnum); 507 slot2 = hash2 & udptable->mask; 508 hslot2 = &udptable->hash2[slot2]; 509 if (hslot->count < hslot2->count) 510 goto begin; 511 512 result = udp4_lib_lookup2(net, saddr, sport, 513 daddr, hnum, dif, 514 hslot2, slot2); 515 if (!result) { 516 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 517 slot2 = hash2 & udptable->mask; 518 hslot2 = &udptable->hash2[slot2]; 519 if (hslot->count < hslot2->count) 520 goto begin; 521 522 result = udp4_lib_lookup2(net, saddr, sport, 523 htonl(INADDR_ANY), hnum, dif, 524 hslot2, slot2); 525 } 526 rcu_read_unlock(); 527 return result; 528 } 529begin: 530 result = NULL; 531 badness = 0; 532 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 533 score = compute_score(sk, net, saddr, hnum, sport, 534 daddr, dport, dif); 535 if (score > badness) { 536 result = sk; 537 badness = score; 538 reuseport = sk->sk_reuseport; 539 if (reuseport) { 540 hash = udp_ehashfn(net, daddr, hnum, 541 saddr, sport); 542 matches = 1; 543 } 544 } else if (score == badness && reuseport) { 545 matches++; 546 if (reciprocal_scale(hash, matches) == 0) 547 result = sk; 548 hash = next_pseudo_random32(hash); 549 } 550 } 551 /* 552 * if the nulls value we got at the end of this lookup is 553 * not the expected one, we must restart lookup. 554 * We probably met an item that was moved to another chain. 555 */ 556 if (get_nulls_value(node) != slot) 557 goto begin; 558 559 if (result) { 560 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 561 result = NULL; 562 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 563 daddr, dport, dif) < badness)) { 564 sock_put(result); 565 goto begin; 566 } 567 } 568 rcu_read_unlock(); 569 return result; 570} 571EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 572 573static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 574 __be16 sport, __be16 dport, 575 struct udp_table *udptable) 576{ 577 const struct iphdr *iph = ip_hdr(skb); 578 579 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 580 iph->daddr, dport, inet_iif(skb), 581 udptable); 582} 583 584struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 585 __be32 daddr, __be16 dport, int dif) 586{ 587 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 588} 589EXPORT_SYMBOL_GPL(udp4_lib_lookup); 590 591static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 592 __be16 loc_port, __be32 loc_addr, 593 __be16 rmt_port, __be32 rmt_addr, 594 int dif, unsigned short hnum) 595{ 596 struct inet_sock *inet = inet_sk(sk); 597 598 if (!net_eq(sock_net(sk), net) || 599 udp_sk(sk)->udp_port_hash != hnum || 600 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 601 (inet->inet_dport != rmt_port && inet->inet_dport) || 602 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 603 ipv6_only_sock(sk) || 604 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)) 605 return false; 606 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif)) 607 return false; 608 return true; 609} 610 611/* 612 * This routine is called by the ICMP module when it gets some 613 * sort of error condition. If err < 0 then the socket should 614 * be closed and the error returned to the user. If err > 0 615 * it's just the icmp type << 8 | icmp code. 616 * Header points to the ip header of the error packet. We move 617 * on past this. Then (as it used to claim before adjustment) 618 * header points to the first 8 bytes of the udp header. We need 619 * to find the appropriate port. 620 */ 621 622void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 623{ 624 struct inet_sock *inet; 625 const struct iphdr *iph = (const struct iphdr *)skb->data; 626 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 627 const int type = icmp_hdr(skb)->type; 628 const int code = icmp_hdr(skb)->code; 629 struct sock *sk; 630 int harderr; 631 int err; 632 struct net *net = dev_net(skb->dev); 633 634 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 635 iph->saddr, uh->source, skb->dev->ifindex, udptable); 636 if (!sk) { 637 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 638 return; /* No socket for error */ 639 } 640 641 err = 0; 642 harderr = 0; 643 inet = inet_sk(sk); 644 645 switch (type) { 646 default: 647 case ICMP_TIME_EXCEEDED: 648 err = EHOSTUNREACH; 649 break; 650 case ICMP_SOURCE_QUENCH: 651 goto out; 652 case ICMP_PARAMETERPROB: 653 err = EPROTO; 654 harderr = 1; 655 break; 656 case ICMP_DEST_UNREACH: 657 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 658 ipv4_sk_update_pmtu(skb, sk, info); 659 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 660 err = EMSGSIZE; 661 harderr = 1; 662 break; 663 } 664 goto out; 665 } 666 err = EHOSTUNREACH; 667 if (code <= NR_ICMP_UNREACH) { 668 harderr = icmp_err_convert[code].fatal; 669 err = icmp_err_convert[code].errno; 670 } 671 break; 672 case ICMP_REDIRECT: 673 ipv4_sk_redirect(skb, sk); 674 goto out; 675 } 676 677 /* 678 * RFC1122: OK. Passes ICMP errors back to application, as per 679 * 4.1.3.3. 680 */ 681 if (!inet->recverr) { 682 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 683 goto out; 684 } else 685 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 686 687 sk->sk_err = err; 688 sk->sk_error_report(sk); 689out: 690 sock_put(sk); 691} 692 693void udp_err(struct sk_buff *skb, u32 info) 694{ 695 __udp4_lib_err(skb, info, &udp_table); 696} 697 698/* 699 * Throw away all pending data and cancel the corking. Socket is locked. 700 */ 701void udp_flush_pending_frames(struct sock *sk) 702{ 703 struct udp_sock *up = udp_sk(sk); 704 705 if (up->pending) { 706 up->len = 0; 707 up->pending = 0; 708 ip_flush_pending_frames(sk); 709 } 710} 711EXPORT_SYMBOL(udp_flush_pending_frames); 712 713/** 714 * udp4_hwcsum - handle outgoing HW checksumming 715 * @skb: sk_buff containing the filled-in UDP header 716 * (checksum field must be zeroed out) 717 * @src: source IP address 718 * @dst: destination IP address 719 */ 720void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 721{ 722 struct udphdr *uh = udp_hdr(skb); 723 int offset = skb_transport_offset(skb); 724 int len = skb->len - offset; 725 int hlen = len; 726 __wsum csum = 0; 727 728 if (!skb_has_frag_list(skb)) { 729 /* 730 * Only one fragment on the socket. 731 */ 732 skb->csum_start = skb_transport_header(skb) - skb->head; 733 skb->csum_offset = offsetof(struct udphdr, check); 734 uh->check = ~csum_tcpudp_magic(src, dst, len, 735 IPPROTO_UDP, 0); 736 } else { 737 struct sk_buff *frags; 738 739 /* 740 * HW-checksum won't work as there are two or more 741 * fragments on the socket so that all csums of sk_buffs 742 * should be together 743 */ 744 skb_walk_frags(skb, frags) { 745 csum = csum_add(csum, frags->csum); 746 hlen -= frags->len; 747 } 748 749 csum = skb_checksum(skb, offset, hlen, csum); 750 skb->ip_summed = CHECKSUM_NONE; 751 752 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 753 if (uh->check == 0) 754 uh->check = CSUM_MANGLED_0; 755 } 756} 757EXPORT_SYMBOL_GPL(udp4_hwcsum); 758 759/* Function to set UDP checksum for an IPv4 UDP packet. This is intended 760 * for the simple case like when setting the checksum for a UDP tunnel. 761 */ 762void udp_set_csum(bool nocheck, struct sk_buff *skb, 763 __be32 saddr, __be32 daddr, int len) 764{ 765 struct udphdr *uh = udp_hdr(skb); 766 767 if (nocheck) 768 uh->check = 0; 769 else if (skb_is_gso(skb)) 770 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 771 else if (skb_dst(skb) && skb_dst(skb)->dev && 772 (skb_dst(skb)->dev->features & NETIF_F_V4_CSUM)) { 773 774 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 775 776 skb->ip_summed = CHECKSUM_PARTIAL; 777 skb->csum_start = skb_transport_header(skb) - skb->head; 778 skb->csum_offset = offsetof(struct udphdr, check); 779 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 780 } else { 781 __wsum csum; 782 783 BUG_ON(skb->ip_summed == CHECKSUM_PARTIAL); 784 785 uh->check = 0; 786 csum = skb_checksum(skb, 0, len, 0); 787 uh->check = udp_v4_check(len, saddr, daddr, csum); 788 if (uh->check == 0) 789 uh->check = CSUM_MANGLED_0; 790 791 skb->ip_summed = CHECKSUM_UNNECESSARY; 792 } 793} 794EXPORT_SYMBOL(udp_set_csum); 795 796static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 797{ 798 struct sock *sk = skb->sk; 799 struct inet_sock *inet = inet_sk(sk); 800 struct udphdr *uh; 801 int err = 0; 802 int is_udplite = IS_UDPLITE(sk); 803 int offset = skb_transport_offset(skb); 804 int len = skb->len - offset; 805 __wsum csum = 0; 806 807 /* 808 * Create a UDP header 809 */ 810 uh = udp_hdr(skb); 811 uh->source = inet->inet_sport; 812 uh->dest = fl4->fl4_dport; 813 uh->len = htons(len); 814 uh->check = 0; 815 816 if (is_udplite) /* UDP-Lite */ 817 csum = udplite_csum(skb); 818 819 else if (sk->sk_no_check_tx) { /* UDP csum disabled */ 820 821 skb->ip_summed = CHECKSUM_NONE; 822 goto send; 823 824 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 825 826 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 827 goto send; 828 829 } else 830 csum = udp_csum(skb); 831 832 /* add protocol-dependent pseudo-header */ 833 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 834 sk->sk_protocol, csum); 835 if (uh->check == 0) 836 uh->check = CSUM_MANGLED_0; 837 838send: 839 err = ip_send_skb(sock_net(sk), skb); 840 if (err) { 841 if (err == -ENOBUFS && !inet->recverr) { 842 UDP_INC_STATS_USER(sock_net(sk), 843 UDP_MIB_SNDBUFERRORS, is_udplite); 844 err = 0; 845 } 846 } else 847 UDP_INC_STATS_USER(sock_net(sk), 848 UDP_MIB_OUTDATAGRAMS, is_udplite); 849 return err; 850} 851 852/* 853 * Push out all pending data as one UDP datagram. Socket is locked. 854 */ 855int udp_push_pending_frames(struct sock *sk) 856{ 857 struct udp_sock *up = udp_sk(sk); 858 struct inet_sock *inet = inet_sk(sk); 859 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 860 struct sk_buff *skb; 861 int err = 0; 862 863 skb = ip_finish_skb(sk, fl4); 864 if (!skb) 865 goto out; 866 867 err = udp_send_skb(skb, fl4); 868 869out: 870 up->len = 0; 871 up->pending = 0; 872 return err; 873} 874EXPORT_SYMBOL(udp_push_pending_frames); 875 876int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 877{ 878 struct inet_sock *inet = inet_sk(sk); 879 struct udp_sock *up = udp_sk(sk); 880 struct flowi4 fl4_stack; 881 struct flowi4 *fl4; 882 int ulen = len; 883 struct ipcm_cookie ipc; 884 struct rtable *rt = NULL; 885 int free = 0; 886 int connected = 0; 887 __be32 daddr, faddr, saddr; 888 __be16 dport; 889 u8 tos; 890 int err, is_udplite = IS_UDPLITE(sk); 891 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 892 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 893 struct sk_buff *skb; 894 struct ip_options_data opt_copy; 895 896 if (len > 0xFFFF) 897 return -EMSGSIZE; 898 899 /* 900 * Check the flags. 901 */ 902 903 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 904 return -EOPNOTSUPP; 905 906 ipc.opt = NULL; 907 ipc.tx_flags = 0; 908 ipc.ttl = 0; 909 ipc.tos = -1; 910 911 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 912 913 fl4 = &inet->cork.fl.u.ip4; 914 if (up->pending) { 915 /* 916 * There are pending frames. 917 * The socket lock must be held while it's corked. 918 */ 919 lock_sock(sk); 920 if (likely(up->pending)) { 921 if (unlikely(up->pending != AF_INET)) { 922 release_sock(sk); 923 return -EINVAL; 924 } 925 goto do_append_data; 926 } 927 release_sock(sk); 928 } 929 ulen += sizeof(struct udphdr); 930 931 /* 932 * Get and verify the address. 933 */ 934 if (msg->msg_name) { 935 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 936 if (msg->msg_namelen < sizeof(*usin)) 937 return -EINVAL; 938 if (usin->sin_family != AF_INET) { 939 if (usin->sin_family != AF_UNSPEC) 940 return -EAFNOSUPPORT; 941 } 942 943 daddr = usin->sin_addr.s_addr; 944 dport = usin->sin_port; 945 if (dport == 0) 946 return -EINVAL; 947 } else { 948 if (sk->sk_state != TCP_ESTABLISHED) 949 return -EDESTADDRREQ; 950 daddr = inet->inet_daddr; 951 dport = inet->inet_dport; 952 /* Open fast path for connected socket. 953 Route will not be used, if at least one option is set. 954 */ 955 connected = 1; 956 } 957 ipc.addr = inet->inet_saddr; 958 959 ipc.oif = sk->sk_bound_dev_if; 960 961 sock_tx_timestamp(sk, &ipc.tx_flags); 962 963 if (msg->msg_controllen) { 964 err = ip_cmsg_send(sock_net(sk), msg, &ipc, 965 sk->sk_family == AF_INET6); 966 if (unlikely(err)) { 967 kfree(ipc.opt); 968 return err; 969 } 970 if (ipc.opt) 971 free = 1; 972 connected = 0; 973 } 974 if (!ipc.opt) { 975 struct ip_options_rcu *inet_opt; 976 977 rcu_read_lock(); 978 inet_opt = rcu_dereference(inet->inet_opt); 979 if (inet_opt) { 980 memcpy(&opt_copy, inet_opt, 981 sizeof(*inet_opt) + inet_opt->opt.optlen); 982 ipc.opt = &opt_copy.opt; 983 } 984 rcu_read_unlock(); 985 } 986 987 saddr = ipc.addr; 988 ipc.addr = faddr = daddr; 989 990 if (ipc.opt && ipc.opt->opt.srr) { 991 if (!daddr) 992 return -EINVAL; 993 faddr = ipc.opt->opt.faddr; 994 connected = 0; 995 } 996 tos = get_rttos(&ipc, inet); 997 if (sock_flag(sk, SOCK_LOCALROUTE) || 998 (msg->msg_flags & MSG_DONTROUTE) || 999 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1000 tos |= RTO_ONLINK; 1001 connected = 0; 1002 } 1003 1004 if (ipv4_is_multicast(daddr)) { 1005 if (!ipc.oif) 1006 ipc.oif = inet->mc_index; 1007 if (!saddr) 1008 saddr = inet->mc_addr; 1009 connected = 0; 1010 } else if (!ipc.oif) 1011 ipc.oif = inet->uc_index; 1012 1013 if (connected) 1014 rt = (struct rtable *)sk_dst_check(sk, 0); 1015 1016 if (!rt) { 1017 struct net *net = sock_net(sk); 1018 1019 fl4 = &fl4_stack; 1020 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1021 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1022 inet_sk_flowi_flags(sk), 1023 faddr, saddr, dport, inet->inet_sport); 1024 1025 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1026 rt = ip_route_output_flow(net, fl4, sk); 1027 if (IS_ERR(rt)) { 1028 err = PTR_ERR(rt); 1029 rt = NULL; 1030 if (err == -ENETUNREACH) 1031 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1032 goto out; 1033 } 1034 1035 err = -EACCES; 1036 if ((rt->rt_flags & RTCF_BROADCAST) && 1037 !sock_flag(sk, SOCK_BROADCAST)) 1038 goto out; 1039 if (connected) 1040 sk_dst_set(sk, dst_clone(&rt->dst)); 1041 } 1042 1043 if (msg->msg_flags&MSG_CONFIRM) 1044 goto do_confirm; 1045back_from_confirm: 1046 1047 saddr = fl4->saddr; 1048 if (!ipc.addr) 1049 daddr = ipc.addr = fl4->daddr; 1050 1051 /* Lockless fast path for the non-corking case. */ 1052 if (!corkreq) { 1053 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1054 sizeof(struct udphdr), &ipc, &rt, 1055 msg->msg_flags); 1056 err = PTR_ERR(skb); 1057 if (!IS_ERR_OR_NULL(skb)) 1058 err = udp_send_skb(skb, fl4); 1059 goto out; 1060 } 1061 1062 lock_sock(sk); 1063 if (unlikely(up->pending)) { 1064 /* The socket is already corked while preparing it. */ 1065 /* ... which is an evident application bug. --ANK */ 1066 release_sock(sk); 1067 1068 net_dbg_ratelimited("cork app bug 2\n"); 1069 err = -EINVAL; 1070 goto out; 1071 } 1072 /* 1073 * Now cork the socket to pend data. 1074 */ 1075 fl4 = &inet->cork.fl.u.ip4; 1076 fl4->daddr = daddr; 1077 fl4->saddr = saddr; 1078 fl4->fl4_dport = dport; 1079 fl4->fl4_sport = inet->inet_sport; 1080 up->pending = AF_INET; 1081 1082do_append_data: 1083 up->len += ulen; 1084 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1085 sizeof(struct udphdr), &ipc, &rt, 1086 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1087 if (err) 1088 udp_flush_pending_frames(sk); 1089 else if (!corkreq) 1090 err = udp_push_pending_frames(sk); 1091 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1092 up->pending = 0; 1093 release_sock(sk); 1094 1095out: 1096 ip_rt_put(rt); 1097 if (free) 1098 kfree(ipc.opt); 1099 if (!err) 1100 return len; 1101 /* 1102 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1103 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1104 * we don't have a good statistic (IpOutDiscards but it can be too many 1105 * things). We could add another new stat but at least for now that 1106 * seems like overkill. 1107 */ 1108 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1109 UDP_INC_STATS_USER(sock_net(sk), 1110 UDP_MIB_SNDBUFERRORS, is_udplite); 1111 } 1112 return err; 1113 1114do_confirm: 1115 dst_confirm(&rt->dst); 1116 if (!(msg->msg_flags&MSG_PROBE) || len) 1117 goto back_from_confirm; 1118 err = 0; 1119 goto out; 1120} 1121EXPORT_SYMBOL(udp_sendmsg); 1122 1123int udp_sendpage(struct sock *sk, struct page *page, int offset, 1124 size_t size, int flags) 1125{ 1126 struct inet_sock *inet = inet_sk(sk); 1127 struct udp_sock *up = udp_sk(sk); 1128 int ret; 1129 1130 if (flags & MSG_SENDPAGE_NOTLAST) 1131 flags |= MSG_MORE; 1132 1133 if (!up->pending) { 1134 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1135 1136 /* Call udp_sendmsg to specify destination address which 1137 * sendpage interface can't pass. 1138 * This will succeed only when the socket is connected. 1139 */ 1140 ret = udp_sendmsg(sk, &msg, 0); 1141 if (ret < 0) 1142 return ret; 1143 } 1144 1145 lock_sock(sk); 1146 1147 if (unlikely(!up->pending)) { 1148 release_sock(sk); 1149 1150 net_dbg_ratelimited("udp cork app bug 3\n"); 1151 return -EINVAL; 1152 } 1153 1154 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1155 page, offset, size, flags); 1156 if (ret == -EOPNOTSUPP) { 1157 release_sock(sk); 1158 return sock_no_sendpage(sk->sk_socket, page, offset, 1159 size, flags); 1160 } 1161 if (ret < 0) { 1162 udp_flush_pending_frames(sk); 1163 goto out; 1164 } 1165 1166 up->len += size; 1167 if (!(up->corkflag || (flags&MSG_MORE))) 1168 ret = udp_push_pending_frames(sk); 1169 if (!ret) 1170 ret = size; 1171out: 1172 release_sock(sk); 1173 return ret; 1174} 1175 1176/** 1177 * first_packet_length - return length of first packet in receive queue 1178 * @sk: socket 1179 * 1180 * Drops all bad checksum frames, until a valid one is found. 1181 * Returns the length of found skb, or 0 if none is found. 1182 */ 1183static unsigned int first_packet_length(struct sock *sk) 1184{ 1185 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1186 struct sk_buff *skb; 1187 unsigned int res; 1188 1189 __skb_queue_head_init(&list_kill); 1190 1191 spin_lock_bh(&rcvq->lock); 1192 while ((skb = skb_peek(rcvq)) != NULL && 1193 udp_lib_checksum_complete(skb)) { 1194 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1195 IS_UDPLITE(sk)); 1196 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1197 IS_UDPLITE(sk)); 1198 atomic_inc(&sk->sk_drops); 1199 __skb_unlink(skb, rcvq); 1200 __skb_queue_tail(&list_kill, skb); 1201 } 1202 res = skb ? skb->len : 0; 1203 spin_unlock_bh(&rcvq->lock); 1204 1205 if (!skb_queue_empty(&list_kill)) { 1206 bool slow = lock_sock_fast(sk); 1207 1208 __skb_queue_purge(&list_kill); 1209 sk_mem_reclaim_partial(sk); 1210 unlock_sock_fast(sk, slow); 1211 } 1212 return res; 1213} 1214 1215/* 1216 * IOCTL requests applicable to the UDP protocol 1217 */ 1218 1219int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1220{ 1221 switch (cmd) { 1222 case SIOCOUTQ: 1223 { 1224 int amount = sk_wmem_alloc_get(sk); 1225 1226 return put_user(amount, (int __user *)arg); 1227 } 1228 1229 case SIOCINQ: 1230 { 1231 unsigned int amount = first_packet_length(sk); 1232 1233 if (amount) 1234 /* 1235 * We will only return the amount 1236 * of this packet since that is all 1237 * that will be read. 1238 */ 1239 amount -= sizeof(struct udphdr); 1240 1241 return put_user(amount, (int __user *)arg); 1242 } 1243 1244 default: 1245 return -ENOIOCTLCMD; 1246 } 1247 1248 return 0; 1249} 1250EXPORT_SYMBOL(udp_ioctl); 1251 1252/* 1253 * This should be easy, if there is something there we 1254 * return it, otherwise we block. 1255 */ 1256 1257int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1258 int flags, int *addr_len) 1259{ 1260 struct inet_sock *inet = inet_sk(sk); 1261 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1262 struct sk_buff *skb; 1263 unsigned int ulen, copied; 1264 int peeked, off = 0; 1265 int err; 1266 int is_udplite = IS_UDPLITE(sk); 1267 bool slow; 1268 1269 if (flags & MSG_ERRQUEUE) 1270 return ip_recv_error(sk, msg, len, addr_len); 1271 1272try_again: 1273 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1274 &peeked, &off, &err); 1275 if (!skb) 1276 goto out; 1277 1278 ulen = skb->len - sizeof(struct udphdr); 1279 copied = len; 1280 if (copied > ulen) 1281 copied = ulen; 1282 else if (copied < ulen) 1283 msg->msg_flags |= MSG_TRUNC; 1284 1285 /* 1286 * If checksum is needed at all, try to do it while copying the 1287 * data. If the data is truncated, or if we only want a partial 1288 * coverage checksum (UDP-Lite), do it before the copy. 1289 */ 1290 1291 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1292 if (udp_lib_checksum_complete(skb)) 1293 goto csum_copy_err; 1294 } 1295 1296 if (skb_csum_unnecessary(skb)) 1297 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr), 1298 msg, copied); 1299 else { 1300 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), 1301 msg); 1302 1303 if (err == -EINVAL) 1304 goto csum_copy_err; 1305 } 1306 1307 if (unlikely(err)) { 1308 trace_kfree_skb(skb, udp_recvmsg); 1309 if (!peeked) { 1310 atomic_inc(&sk->sk_drops); 1311 UDP_INC_STATS_USER(sock_net(sk), 1312 UDP_MIB_INERRORS, is_udplite); 1313 } 1314 goto out_free; 1315 } 1316 1317 if (!peeked) 1318 UDP_INC_STATS_USER(sock_net(sk), 1319 UDP_MIB_INDATAGRAMS, is_udplite); 1320 1321 sock_recv_ts_and_drops(msg, sk, skb); 1322 1323 /* Copy the address. */ 1324 if (sin) { 1325 sin->sin_family = AF_INET; 1326 sin->sin_port = udp_hdr(skb)->source; 1327 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1328 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1329 *addr_len = sizeof(*sin); 1330 } 1331 if (inet->cmsg_flags) 1332 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr)); 1333 1334 err = copied; 1335 if (flags & MSG_TRUNC) 1336 err = ulen; 1337 1338out_free: 1339 skb_free_datagram_locked(sk, skb); 1340out: 1341 return err; 1342 1343csum_copy_err: 1344 slow = lock_sock_fast(sk); 1345 if (!skb_kill_datagram(sk, skb, flags)) { 1346 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1347 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1348 } 1349 unlock_sock_fast(sk, slow); 1350 1351 /* starting over for a new packet, but check if we need to yield */ 1352 cond_resched(); 1353 msg->msg_flags &= ~MSG_TRUNC; 1354 goto try_again; 1355} 1356 1357int udp_disconnect(struct sock *sk, int flags) 1358{ 1359 struct inet_sock *inet = inet_sk(sk); 1360 /* 1361 * 1003.1g - break association. 1362 */ 1363 1364 sk->sk_state = TCP_CLOSE; 1365 inet->inet_daddr = 0; 1366 inet->inet_dport = 0; 1367 sock_rps_reset_rxhash(sk); 1368 sk->sk_bound_dev_if = 0; 1369 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1370 inet_reset_saddr(sk); 1371 1372 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1373 sk->sk_prot->unhash(sk); 1374 inet->inet_sport = 0; 1375 } 1376 sk_dst_reset(sk); 1377 return 0; 1378} 1379EXPORT_SYMBOL(udp_disconnect); 1380 1381void udp_lib_unhash(struct sock *sk) 1382{ 1383 if (sk_hashed(sk)) { 1384 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1385 struct udp_hslot *hslot, *hslot2; 1386 1387 hslot = udp_hashslot(udptable, sock_net(sk), 1388 udp_sk(sk)->udp_port_hash); 1389 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1390 1391 spin_lock_bh(&hslot->lock); 1392 if (sk_nulls_del_node_init_rcu(sk)) { 1393 hslot->count--; 1394 inet_sk(sk)->inet_num = 0; 1395 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1396 1397 spin_lock(&hslot2->lock); 1398 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1399 hslot2->count--; 1400 spin_unlock(&hslot2->lock); 1401 } 1402 spin_unlock_bh(&hslot->lock); 1403 } 1404} 1405EXPORT_SYMBOL(udp_lib_unhash); 1406 1407/* 1408 * inet_rcv_saddr was changed, we must rehash secondary hash 1409 */ 1410void udp_lib_rehash(struct sock *sk, u16 newhash) 1411{ 1412 if (sk_hashed(sk)) { 1413 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1414 struct udp_hslot *hslot, *hslot2, *nhslot2; 1415 1416 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1417 nhslot2 = udp_hashslot2(udptable, newhash); 1418 udp_sk(sk)->udp_portaddr_hash = newhash; 1419 if (hslot2 != nhslot2) { 1420 hslot = udp_hashslot(udptable, sock_net(sk), 1421 udp_sk(sk)->udp_port_hash); 1422 /* we must lock primary chain too */ 1423 spin_lock_bh(&hslot->lock); 1424 1425 spin_lock(&hslot2->lock); 1426 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1427 hslot2->count--; 1428 spin_unlock(&hslot2->lock); 1429 1430 spin_lock(&nhslot2->lock); 1431 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1432 &nhslot2->head); 1433 nhslot2->count++; 1434 spin_unlock(&nhslot2->lock); 1435 1436 spin_unlock_bh(&hslot->lock); 1437 } 1438 } 1439} 1440EXPORT_SYMBOL(udp_lib_rehash); 1441 1442static void udp_v4_rehash(struct sock *sk) 1443{ 1444 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1445 inet_sk(sk)->inet_rcv_saddr, 1446 inet_sk(sk)->inet_num); 1447 udp_lib_rehash(sk, new_hash); 1448} 1449 1450static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1451{ 1452 int rc; 1453 1454 if (inet_sk(sk)->inet_daddr) { 1455 sock_rps_save_rxhash(sk, skb); 1456 sk_mark_napi_id(sk, skb); 1457 sk_incoming_cpu_update(sk); 1458 } 1459 1460 rc = sock_queue_rcv_skb(sk, skb); 1461 if (rc < 0) { 1462 int is_udplite = IS_UDPLITE(sk); 1463 1464 /* Note that an ENOMEM error is charged twice */ 1465 if (rc == -ENOMEM) 1466 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1467 is_udplite); 1468 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1469 kfree_skb(skb); 1470 trace_udp_fail_queue_rcv_skb(rc, sk); 1471 return -1; 1472 } 1473 1474 return 0; 1475 1476} 1477 1478static struct static_key udp_encap_needed __read_mostly; 1479void udp_encap_enable(void) 1480{ 1481 if (!static_key_enabled(&udp_encap_needed)) 1482 static_key_slow_inc(&udp_encap_needed); 1483} 1484EXPORT_SYMBOL(udp_encap_enable); 1485 1486/* returns: 1487 * -1: error 1488 * 0: success 1489 * >0: "udp encap" protocol resubmission 1490 * 1491 * Note that in the success and error cases, the skb is assumed to 1492 * have either been requeued or freed. 1493 */ 1494int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1495{ 1496 struct udp_sock *up = udp_sk(sk); 1497 int rc; 1498 int is_udplite = IS_UDPLITE(sk); 1499 1500 /* 1501 * Charge it to the socket, dropping if the queue is full. 1502 */ 1503 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1504 goto drop; 1505 nf_reset(skb); 1506 1507 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1508 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1509 1510 /* 1511 * This is an encapsulation socket so pass the skb to 1512 * the socket's udp_encap_rcv() hook. Otherwise, just 1513 * fall through and pass this up the UDP socket. 1514 * up->encap_rcv() returns the following value: 1515 * =0 if skb was successfully passed to the encap 1516 * handler or was discarded by it. 1517 * >0 if skb should be passed on to UDP. 1518 * <0 if skb should be resubmitted as proto -N 1519 */ 1520 1521 /* if we're overly short, let UDP handle it */ 1522 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1523 if (skb->len > sizeof(struct udphdr) && encap_rcv) { 1524 int ret; 1525 1526 /* Verify checksum before giving to encap */ 1527 if (udp_lib_checksum_complete(skb)) 1528 goto csum_error; 1529 1530 ret = encap_rcv(sk, skb); 1531 if (ret <= 0) { 1532 UDP_INC_STATS_BH(sock_net(sk), 1533 UDP_MIB_INDATAGRAMS, 1534 is_udplite); 1535 return -ret; 1536 } 1537 } 1538 1539 /* FALLTHROUGH -- it's a UDP Packet */ 1540 } 1541 1542 /* 1543 * UDP-Lite specific tests, ignored on UDP sockets 1544 */ 1545 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1546 1547 /* 1548 * MIB statistics other than incrementing the error count are 1549 * disabled for the following two types of errors: these depend 1550 * on the application settings, not on the functioning of the 1551 * protocol stack as such. 1552 * 1553 * RFC 3828 here recommends (sec 3.3): "There should also be a 1554 * way ... to ... at least let the receiving application block 1555 * delivery of packets with coverage values less than a value 1556 * provided by the application." 1557 */ 1558 if (up->pcrlen == 0) { /* full coverage was set */ 1559 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 1560 UDP_SKB_CB(skb)->cscov, skb->len); 1561 goto drop; 1562 } 1563 /* The next case involves violating the min. coverage requested 1564 * by the receiver. This is subtle: if receiver wants x and x is 1565 * greater than the buffersize/MTU then receiver will complain 1566 * that it wants x while sender emits packets of smaller size y. 1567 * Therefore the above ...()->partial_cov statement is essential. 1568 */ 1569 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1570 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 1571 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1572 goto drop; 1573 } 1574 } 1575 1576 if (rcu_access_pointer(sk->sk_filter) && 1577 udp_lib_checksum_complete(skb)) 1578 goto csum_error; 1579 1580 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 1581 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1582 is_udplite); 1583 goto drop; 1584 } 1585 1586 rc = 0; 1587 1588 ipv4_pktinfo_prepare(sk, skb); 1589 bh_lock_sock(sk); 1590 if (!sock_owned_by_user(sk)) 1591 rc = __udp_queue_rcv_skb(sk, skb); 1592 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1593 bh_unlock_sock(sk); 1594 goto drop; 1595 } 1596 bh_unlock_sock(sk); 1597 1598 return rc; 1599 1600csum_error: 1601 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1602drop: 1603 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1604 atomic_inc(&sk->sk_drops); 1605 kfree_skb(skb); 1606 return -1; 1607} 1608 1609static void flush_stack(struct sock **stack, unsigned int count, 1610 struct sk_buff *skb, unsigned int final) 1611{ 1612 unsigned int i; 1613 struct sk_buff *skb1 = NULL; 1614 struct sock *sk; 1615 1616 for (i = 0; i < count; i++) { 1617 sk = stack[i]; 1618 if (likely(!skb1)) 1619 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1620 1621 if (!skb1) { 1622 atomic_inc(&sk->sk_drops); 1623 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1624 IS_UDPLITE(sk)); 1625 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1626 IS_UDPLITE(sk)); 1627 } 1628 1629 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1630 skb1 = NULL; 1631 1632 sock_put(sk); 1633 } 1634 if (unlikely(skb1)) 1635 kfree_skb(skb1); 1636} 1637 1638/* For TCP sockets, sk_rx_dst is protected by socket lock 1639 * For UDP, we use xchg() to guard against concurrent changes. 1640 */ 1641static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 1642{ 1643 struct dst_entry *old; 1644 1645 dst_hold(dst); 1646 old = xchg(&sk->sk_rx_dst, dst); 1647 dst_release(old); 1648} 1649 1650/* 1651 * Multicasts and broadcasts go to each listener. 1652 * 1653 * Note: called only from the BH handler context. 1654 */ 1655static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1656 struct udphdr *uh, 1657 __be32 saddr, __be32 daddr, 1658 struct udp_table *udptable, 1659 int proto) 1660{ 1661 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1662 struct hlist_nulls_node *node; 1663 unsigned short hnum = ntohs(uh->dest); 1664 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 1665 int dif = skb->dev->ifindex; 1666 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node); 1667 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 1668 bool inner_flushed = false; 1669 1670 if (use_hash2) { 1671 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 1672 udp_table.mask; 1673 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask; 1674start_lookup: 1675 hslot = &udp_table.hash2[hash2]; 1676 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 1677 } 1678 1679 spin_lock(&hslot->lock); 1680 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) { 1681 if (__udp_is_mcast_sock(net, sk, 1682 uh->dest, daddr, 1683 uh->source, saddr, 1684 dif, hnum)) { 1685 if (unlikely(count == ARRAY_SIZE(stack))) { 1686 flush_stack(stack, count, skb, ~0); 1687 inner_flushed = true; 1688 count = 0; 1689 } 1690 stack[count++] = sk; 1691 sock_hold(sk); 1692 } 1693 } 1694 1695 spin_unlock(&hslot->lock); 1696 1697 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 1698 if (use_hash2 && hash2 != hash2_any) { 1699 hash2 = hash2_any; 1700 goto start_lookup; 1701 } 1702 1703 /* 1704 * do the slow work with no lock held 1705 */ 1706 if (count) { 1707 flush_stack(stack, count, skb, count - 1); 1708 } else { 1709 if (!inner_flushed) 1710 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI, 1711 proto == IPPROTO_UDPLITE); 1712 consume_skb(skb); 1713 } 1714 return 0; 1715} 1716 1717/* Initialize UDP checksum. If exited with zero value (success), 1718 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1719 * Otherwise, csum completion requires chacksumming packet body, 1720 * including udp header and folding it to skb->csum. 1721 */ 1722static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1723 int proto) 1724{ 1725 int err; 1726 1727 UDP_SKB_CB(skb)->partial_cov = 0; 1728 UDP_SKB_CB(skb)->cscov = skb->len; 1729 1730 if (proto == IPPROTO_UDPLITE) { 1731 err = udplite_checksum_init(skb, uh); 1732 if (err) 1733 return err; 1734 } 1735 1736 return skb_checksum_init_zero_check(skb, proto, uh->check, 1737 inet_compute_pseudo); 1738} 1739 1740/* 1741 * All we need to do is get the socket, and then do a checksum. 1742 */ 1743 1744int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1745 int proto) 1746{ 1747 struct sock *sk; 1748 struct udphdr *uh; 1749 unsigned short ulen; 1750 struct rtable *rt = skb_rtable(skb); 1751 __be32 saddr, daddr; 1752 struct net *net = dev_net(skb->dev); 1753 1754 /* 1755 * Validate the packet. 1756 */ 1757 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1758 goto drop; /* No space for header. */ 1759 1760 uh = udp_hdr(skb); 1761 ulen = ntohs(uh->len); 1762 saddr = ip_hdr(skb)->saddr; 1763 daddr = ip_hdr(skb)->daddr; 1764 1765 if (ulen > skb->len) 1766 goto short_packet; 1767 1768 if (proto == IPPROTO_UDP) { 1769 /* UDP validates ulen. */ 1770 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1771 goto short_packet; 1772 uh = udp_hdr(skb); 1773 } 1774 1775 if (udp4_csum_init(skb, uh, proto)) 1776 goto csum_error; 1777 1778 sk = skb_steal_sock(skb); 1779 if (sk) { 1780 struct dst_entry *dst = skb_dst(skb); 1781 int ret; 1782 1783 if (unlikely(sk->sk_rx_dst != dst)) 1784 udp_sk_rx_dst_set(sk, dst); 1785 1786 ret = udp_queue_rcv_skb(sk, skb); 1787 sock_put(sk); 1788 /* a return value > 0 means to resubmit the input, but 1789 * it wants the return to be -protocol, or 0 1790 */ 1791 if (ret > 0) 1792 return -ret; 1793 return 0; 1794 } 1795 1796 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1797 return __udp4_lib_mcast_deliver(net, skb, uh, 1798 saddr, daddr, udptable, proto); 1799 1800 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1801 if (sk) { 1802 int ret; 1803 1804 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 1805 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 1806 inet_compute_pseudo); 1807 1808 ret = udp_queue_rcv_skb(sk, skb); 1809 sock_put(sk); 1810 1811 /* a return value > 0 means to resubmit the input, but 1812 * it wants the return to be -protocol, or 0 1813 */ 1814 if (ret > 0) 1815 return -ret; 1816 return 0; 1817 } 1818 1819 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1820 goto drop; 1821 nf_reset(skb); 1822 1823 /* No socket. Drop packet silently, if checksum is wrong */ 1824 if (udp_lib_checksum_complete(skb)) 1825 goto csum_error; 1826 1827 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1828 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1829 1830 /* 1831 * Hmm. We got an UDP packet to a port to which we 1832 * don't wanna listen. Ignore it. 1833 */ 1834 kfree_skb(skb); 1835 return 0; 1836 1837short_packet: 1838 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1839 proto == IPPROTO_UDPLITE ? "Lite" : "", 1840 &saddr, ntohs(uh->source), 1841 ulen, skb->len, 1842 &daddr, ntohs(uh->dest)); 1843 goto drop; 1844 1845csum_error: 1846 /* 1847 * RFC1122: OK. Discards the bad packet silently (as far as 1848 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1849 */ 1850 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1851 proto == IPPROTO_UDPLITE ? "Lite" : "", 1852 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1853 ulen); 1854 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1855drop: 1856 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1857 kfree_skb(skb); 1858 return 0; 1859} 1860 1861/* We can only early demux multicast if there is a single matching socket. 1862 * If more than one socket found returns NULL 1863 */ 1864static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 1865 __be16 loc_port, __be32 loc_addr, 1866 __be16 rmt_port, __be32 rmt_addr, 1867 int dif) 1868{ 1869 struct sock *sk, *result; 1870 struct hlist_nulls_node *node; 1871 unsigned short hnum = ntohs(loc_port); 1872 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask); 1873 struct udp_hslot *hslot = &udp_table.hash[slot]; 1874 1875 /* Do not bother scanning a too big list */ 1876 if (hslot->count > 10) 1877 return NULL; 1878 1879 rcu_read_lock(); 1880begin: 1881 count = 0; 1882 result = NULL; 1883 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 1884 if (__udp_is_mcast_sock(net, sk, 1885 loc_port, loc_addr, 1886 rmt_port, rmt_addr, 1887 dif, hnum)) { 1888 result = sk; 1889 ++count; 1890 } 1891 } 1892 /* 1893 * if the nulls value we got at the end of this lookup is 1894 * not the expected one, we must restart lookup. 1895 * We probably met an item that was moved to another chain. 1896 */ 1897 if (get_nulls_value(node) != slot) 1898 goto begin; 1899 1900 if (result) { 1901 if (count != 1 || 1902 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1903 result = NULL; 1904 else if (unlikely(!__udp_is_mcast_sock(net, result, 1905 loc_port, loc_addr, 1906 rmt_port, rmt_addr, 1907 dif, hnum))) { 1908 sock_put(result); 1909 result = NULL; 1910 } 1911 } 1912 rcu_read_unlock(); 1913 return result; 1914} 1915 1916/* For unicast we should only early demux connected sockets or we can 1917 * break forwarding setups. The chains here can be long so only check 1918 * if the first socket is an exact match and if not move on. 1919 */ 1920static struct sock *__udp4_lib_demux_lookup(struct net *net, 1921 __be16 loc_port, __be32 loc_addr, 1922 __be16 rmt_port, __be32 rmt_addr, 1923 int dif) 1924{ 1925 struct sock *sk, *result; 1926 struct hlist_nulls_node *node; 1927 unsigned short hnum = ntohs(loc_port); 1928 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum); 1929 unsigned int slot2 = hash2 & udp_table.mask; 1930 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 1931 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 1932 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 1933 1934 rcu_read_lock(); 1935 result = NULL; 1936 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 1937 if (INET_MATCH(sk, net, acookie, 1938 rmt_addr, loc_addr, ports, dif)) 1939 result = sk; 1940 /* Only check first socket in chain */ 1941 break; 1942 } 1943 1944 if (result) { 1945 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 1946 result = NULL; 1947 else if (unlikely(!INET_MATCH(sk, net, acookie, 1948 rmt_addr, loc_addr, 1949 ports, dif))) { 1950 sock_put(result); 1951 result = NULL; 1952 } 1953 } 1954 rcu_read_unlock(); 1955 return result; 1956} 1957 1958void udp_v4_early_demux(struct sk_buff *skb) 1959{ 1960 struct net *net = dev_net(skb->dev); 1961 const struct iphdr *iph; 1962 const struct udphdr *uh; 1963 struct sock *sk; 1964 struct dst_entry *dst; 1965 int dif = skb->dev->ifindex; 1966 int ours; 1967 1968 /* validate the packet */ 1969 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 1970 return; 1971 1972 iph = ip_hdr(skb); 1973 uh = udp_hdr(skb); 1974 1975 if (skb->pkt_type == PACKET_BROADCAST || 1976 skb->pkt_type == PACKET_MULTICAST) { 1977 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 1978 1979 if (!in_dev) 1980 return; 1981 1982 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 1983 iph->protocol); 1984 if (!ours) 1985 return; 1986 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 1987 uh->source, iph->saddr, dif); 1988 } else if (skb->pkt_type == PACKET_HOST) { 1989 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 1990 uh->source, iph->saddr, dif); 1991 } else { 1992 return; 1993 } 1994 1995 if (!sk) 1996 return; 1997 1998 skb->sk = sk; 1999 skb->destructor = sock_efree; 2000 dst = READ_ONCE(sk->sk_rx_dst); 2001 2002 if (dst) 2003 dst = dst_check(dst, 0); 2004 if (dst) { 2005 /* DST_NOCACHE can not be used without taking a reference */ 2006 if (dst->flags & DST_NOCACHE) { 2007 if (likely(atomic_inc_not_zero(&dst->__refcnt))) 2008 skb_dst_set(skb, dst); 2009 } else { 2010 skb_dst_set_noref(skb, dst); 2011 } 2012 } 2013} 2014 2015int udp_rcv(struct sk_buff *skb) 2016{ 2017 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2018} 2019 2020void udp_destroy_sock(struct sock *sk) 2021{ 2022 struct udp_sock *up = udp_sk(sk); 2023 bool slow = lock_sock_fast(sk); 2024 udp_flush_pending_frames(sk); 2025 unlock_sock_fast(sk, slow); 2026 if (static_key_false(&udp_encap_needed) && up->encap_type) { 2027 void (*encap_destroy)(struct sock *sk); 2028 encap_destroy = ACCESS_ONCE(up->encap_destroy); 2029 if (encap_destroy) 2030 encap_destroy(sk); 2031 } 2032} 2033 2034/* 2035 * Socket option code for UDP 2036 */ 2037int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2038 char __user *optval, unsigned int optlen, 2039 int (*push_pending_frames)(struct sock *)) 2040{ 2041 struct udp_sock *up = udp_sk(sk); 2042 int val, valbool; 2043 int err = 0; 2044 int is_udplite = IS_UDPLITE(sk); 2045 2046 if (optlen < sizeof(int)) 2047 return -EINVAL; 2048 2049 if (get_user(val, (int __user *)optval)) 2050 return -EFAULT; 2051 2052 valbool = val ? 1 : 0; 2053 2054 switch (optname) { 2055 case UDP_CORK: 2056 if (val != 0) { 2057 up->corkflag = 1; 2058 } else { 2059 up->corkflag = 0; 2060 lock_sock(sk); 2061 push_pending_frames(sk); 2062 release_sock(sk); 2063 } 2064 break; 2065 2066 case UDP_ENCAP: 2067 switch (val) { 2068 case 0: 2069 case UDP_ENCAP_ESPINUDP: 2070 case UDP_ENCAP_ESPINUDP_NON_IKE: 2071 up->encap_rcv = xfrm4_udp_encap_rcv; 2072 /* FALLTHROUGH */ 2073 case UDP_ENCAP_L2TPINUDP: 2074 up->encap_type = val; 2075 udp_encap_enable(); 2076 break; 2077 default: 2078 err = -ENOPROTOOPT; 2079 break; 2080 } 2081 break; 2082 2083 case UDP_NO_CHECK6_TX: 2084 up->no_check6_tx = valbool; 2085 break; 2086 2087 case UDP_NO_CHECK6_RX: 2088 up->no_check6_rx = valbool; 2089 break; 2090 2091 /* 2092 * UDP-Lite's partial checksum coverage (RFC 3828). 2093 */ 2094 /* The sender sets actual checksum coverage length via this option. 2095 * The case coverage > packet length is handled by send module. */ 2096 case UDPLITE_SEND_CSCOV: 2097 if (!is_udplite) /* Disable the option on UDP sockets */ 2098 return -ENOPROTOOPT; 2099 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2100 val = 8; 2101 else if (val > USHRT_MAX) 2102 val = USHRT_MAX; 2103 up->pcslen = val; 2104 up->pcflag |= UDPLITE_SEND_CC; 2105 break; 2106 2107 /* The receiver specifies a minimum checksum coverage value. To make 2108 * sense, this should be set to at least 8 (as done below). If zero is 2109 * used, this again means full checksum coverage. */ 2110 case UDPLITE_RECV_CSCOV: 2111 if (!is_udplite) /* Disable the option on UDP sockets */ 2112 return -ENOPROTOOPT; 2113 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2114 val = 8; 2115 else if (val > USHRT_MAX) 2116 val = USHRT_MAX; 2117 up->pcrlen = val; 2118 up->pcflag |= UDPLITE_RECV_CC; 2119 break; 2120 2121 default: 2122 err = -ENOPROTOOPT; 2123 break; 2124 } 2125 2126 return err; 2127} 2128EXPORT_SYMBOL(udp_lib_setsockopt); 2129 2130int udp_setsockopt(struct sock *sk, int level, int optname, 2131 char __user *optval, unsigned int optlen) 2132{ 2133 if (level == SOL_UDP || level == SOL_UDPLITE) 2134 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2135 udp_push_pending_frames); 2136 return ip_setsockopt(sk, level, optname, optval, optlen); 2137} 2138 2139#ifdef CONFIG_COMPAT 2140int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2141 char __user *optval, unsigned int optlen) 2142{ 2143 if (level == SOL_UDP || level == SOL_UDPLITE) 2144 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2145 udp_push_pending_frames); 2146 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2147} 2148#endif 2149 2150int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2151 char __user *optval, int __user *optlen) 2152{ 2153 struct udp_sock *up = udp_sk(sk); 2154 int val, len; 2155 2156 if (get_user(len, optlen)) 2157 return -EFAULT; 2158 2159 len = min_t(unsigned int, len, sizeof(int)); 2160 2161 if (len < 0) 2162 return -EINVAL; 2163 2164 switch (optname) { 2165 case UDP_CORK: 2166 val = up->corkflag; 2167 break; 2168 2169 case UDP_ENCAP: 2170 val = up->encap_type; 2171 break; 2172 2173 case UDP_NO_CHECK6_TX: 2174 val = up->no_check6_tx; 2175 break; 2176 2177 case UDP_NO_CHECK6_RX: 2178 val = up->no_check6_rx; 2179 break; 2180 2181 /* The following two cannot be changed on UDP sockets, the return is 2182 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2183 case UDPLITE_SEND_CSCOV: 2184 val = up->pcslen; 2185 break; 2186 2187 case UDPLITE_RECV_CSCOV: 2188 val = up->pcrlen; 2189 break; 2190 2191 default: 2192 return -ENOPROTOOPT; 2193 } 2194 2195 if (put_user(len, optlen)) 2196 return -EFAULT; 2197 if (copy_to_user(optval, &val, len)) 2198 return -EFAULT; 2199 return 0; 2200} 2201EXPORT_SYMBOL(udp_lib_getsockopt); 2202 2203int udp_getsockopt(struct sock *sk, int level, int optname, 2204 char __user *optval, int __user *optlen) 2205{ 2206 if (level == SOL_UDP || level == SOL_UDPLITE) 2207 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2208 return ip_getsockopt(sk, level, optname, optval, optlen); 2209} 2210 2211#ifdef CONFIG_COMPAT 2212int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2213 char __user *optval, int __user *optlen) 2214{ 2215 if (level == SOL_UDP || level == SOL_UDPLITE) 2216 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2217 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2218} 2219#endif 2220/** 2221 * udp_poll - wait for a UDP event. 2222 * @file - file struct 2223 * @sock - socket 2224 * @wait - poll table 2225 * 2226 * This is same as datagram poll, except for the special case of 2227 * blocking sockets. If application is using a blocking fd 2228 * and a packet with checksum error is in the queue; 2229 * then it could get return from select indicating data available 2230 * but then block when reading it. Add special case code 2231 * to work around these arguably broken applications. 2232 */ 2233unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2234{ 2235 unsigned int mask = datagram_poll(file, sock, wait); 2236 struct sock *sk = sock->sk; 2237 2238 sock_rps_record_flow(sk); 2239 2240 /* Check for false positives due to checksum errors */ 2241 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2242 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 2243 mask &= ~(POLLIN | POLLRDNORM); 2244 2245 return mask; 2246 2247} 2248EXPORT_SYMBOL(udp_poll); 2249 2250struct proto udp_prot = { 2251 .name = "UDP", 2252 .owner = THIS_MODULE, 2253 .close = udp_lib_close, 2254 .connect = ip4_datagram_connect, 2255 .disconnect = udp_disconnect, 2256 .ioctl = udp_ioctl, 2257 .destroy = udp_destroy_sock, 2258 .setsockopt = udp_setsockopt, 2259 .getsockopt = udp_getsockopt, 2260 .sendmsg = udp_sendmsg, 2261 .recvmsg = udp_recvmsg, 2262 .sendpage = udp_sendpage, 2263 .backlog_rcv = __udp_queue_rcv_skb, 2264 .release_cb = ip4_datagram_release_cb, 2265 .hash = udp_lib_hash, 2266 .unhash = udp_lib_unhash, 2267 .rehash = udp_v4_rehash, 2268 .get_port = udp_v4_get_port, 2269 .memory_allocated = &udp_memory_allocated, 2270 .sysctl_mem = sysctl_udp_mem, 2271 .sysctl_wmem = &sysctl_udp_wmem_min, 2272 .sysctl_rmem = &sysctl_udp_rmem_min, 2273 .obj_size = sizeof(struct udp_sock), 2274 .slab_flags = SLAB_DESTROY_BY_RCU, 2275 .h.udp_table = &udp_table, 2276#ifdef CONFIG_COMPAT 2277 .compat_setsockopt = compat_udp_setsockopt, 2278 .compat_getsockopt = compat_udp_getsockopt, 2279#endif 2280 .clear_sk = sk_prot_clear_portaddr_nulls, 2281}; 2282EXPORT_SYMBOL(udp_prot); 2283 2284/* ------------------------------------------------------------------------ */ 2285#ifdef CONFIG_PROC_FS 2286 2287static struct sock *udp_get_first(struct seq_file *seq, int start) 2288{ 2289 struct sock *sk; 2290 struct udp_iter_state *state = seq->private; 2291 struct net *net = seq_file_net(seq); 2292 2293 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2294 ++state->bucket) { 2295 struct hlist_nulls_node *node; 2296 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2297 2298 if (hlist_nulls_empty(&hslot->head)) 2299 continue; 2300 2301 spin_lock_bh(&hslot->lock); 2302 sk_nulls_for_each(sk, node, &hslot->head) { 2303 if (!net_eq(sock_net(sk), net)) 2304 continue; 2305 if (sk->sk_family == state->family) 2306 goto found; 2307 } 2308 spin_unlock_bh(&hslot->lock); 2309 } 2310 sk = NULL; 2311found: 2312 return sk; 2313} 2314 2315static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2316{ 2317 struct udp_iter_state *state = seq->private; 2318 struct net *net = seq_file_net(seq); 2319 2320 do { 2321 sk = sk_nulls_next(sk); 2322 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2323 2324 if (!sk) { 2325 if (state->bucket <= state->udp_table->mask) 2326 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2327 return udp_get_first(seq, state->bucket + 1); 2328 } 2329 return sk; 2330} 2331 2332static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2333{ 2334 struct sock *sk = udp_get_first(seq, 0); 2335 2336 if (sk) 2337 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2338 --pos; 2339 return pos ? NULL : sk; 2340} 2341 2342static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2343{ 2344 struct udp_iter_state *state = seq->private; 2345 state->bucket = MAX_UDP_PORTS; 2346 2347 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2348} 2349 2350static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2351{ 2352 struct sock *sk; 2353 2354 if (v == SEQ_START_TOKEN) 2355 sk = udp_get_idx(seq, 0); 2356 else 2357 sk = udp_get_next(seq, v); 2358 2359 ++*pos; 2360 return sk; 2361} 2362 2363static void udp_seq_stop(struct seq_file *seq, void *v) 2364{ 2365 struct udp_iter_state *state = seq->private; 2366 2367 if (state->bucket <= state->udp_table->mask) 2368 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2369} 2370 2371int udp_seq_open(struct inode *inode, struct file *file) 2372{ 2373 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2374 struct udp_iter_state *s; 2375 int err; 2376 2377 err = seq_open_net(inode, file, &afinfo->seq_ops, 2378 sizeof(struct udp_iter_state)); 2379 if (err < 0) 2380 return err; 2381 2382 s = ((struct seq_file *)file->private_data)->private; 2383 s->family = afinfo->family; 2384 s->udp_table = afinfo->udp_table; 2385 return err; 2386} 2387EXPORT_SYMBOL(udp_seq_open); 2388 2389/* ------------------------------------------------------------------------ */ 2390int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2391{ 2392 struct proc_dir_entry *p; 2393 int rc = 0; 2394 2395 afinfo->seq_ops.start = udp_seq_start; 2396 afinfo->seq_ops.next = udp_seq_next; 2397 afinfo->seq_ops.stop = udp_seq_stop; 2398 2399 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2400 afinfo->seq_fops, afinfo); 2401 if (!p) 2402 rc = -ENOMEM; 2403 return rc; 2404} 2405EXPORT_SYMBOL(udp_proc_register); 2406 2407void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2408{ 2409 remove_proc_entry(afinfo->name, net->proc_net); 2410} 2411EXPORT_SYMBOL(udp_proc_unregister); 2412 2413/* ------------------------------------------------------------------------ */ 2414static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2415 int bucket) 2416{ 2417 struct inet_sock *inet = inet_sk(sp); 2418 __be32 dest = inet->inet_daddr; 2419 __be32 src = inet->inet_rcv_saddr; 2420 __u16 destp = ntohs(inet->inet_dport); 2421 __u16 srcp = ntohs(inet->inet_sport); 2422 2423 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2424 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2425 bucket, src, srcp, dest, destp, sp->sk_state, 2426 sk_wmem_alloc_get(sp), 2427 sk_rmem_alloc_get(sp), 2428 0, 0L, 0, 2429 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2430 0, sock_i_ino(sp), 2431 atomic_read(&sp->sk_refcnt), sp, 2432 atomic_read(&sp->sk_drops)); 2433} 2434 2435int udp4_seq_show(struct seq_file *seq, void *v) 2436{ 2437 seq_setwidth(seq, 127); 2438 if (v == SEQ_START_TOKEN) 2439 seq_puts(seq, " sl local_address rem_address st tx_queue " 2440 "rx_queue tr tm->when retrnsmt uid timeout " 2441 "inode ref pointer drops"); 2442 else { 2443 struct udp_iter_state *state = seq->private; 2444 2445 udp4_format_sock(v, seq, state->bucket); 2446 } 2447 seq_pad(seq, '\n'); 2448 return 0; 2449} 2450 2451static const struct file_operations udp_afinfo_seq_fops = { 2452 .owner = THIS_MODULE, 2453 .open = udp_seq_open, 2454 .read = seq_read, 2455 .llseek = seq_lseek, 2456 .release = seq_release_net 2457}; 2458 2459/* ------------------------------------------------------------------------ */ 2460static struct udp_seq_afinfo udp4_seq_afinfo = { 2461 .name = "udp", 2462 .family = AF_INET, 2463 .udp_table = &udp_table, 2464 .seq_fops = &udp_afinfo_seq_fops, 2465 .seq_ops = { 2466 .show = udp4_seq_show, 2467 }, 2468}; 2469 2470static int __net_init udp4_proc_init_net(struct net *net) 2471{ 2472 return udp_proc_register(net, &udp4_seq_afinfo); 2473} 2474 2475static void __net_exit udp4_proc_exit_net(struct net *net) 2476{ 2477 udp_proc_unregister(net, &udp4_seq_afinfo); 2478} 2479 2480static struct pernet_operations udp4_net_ops = { 2481 .init = udp4_proc_init_net, 2482 .exit = udp4_proc_exit_net, 2483}; 2484 2485int __init udp4_proc_init(void) 2486{ 2487 return register_pernet_subsys(&udp4_net_ops); 2488} 2489 2490void udp4_proc_exit(void) 2491{ 2492 unregister_pernet_subsys(&udp4_net_ops); 2493} 2494#endif /* CONFIG_PROC_FS */ 2495 2496static __initdata unsigned long uhash_entries; 2497static int __init set_uhash_entries(char *str) 2498{ 2499 ssize_t ret; 2500 2501 if (!str) 2502 return 0; 2503 2504 ret = kstrtoul(str, 0, &uhash_entries); 2505 if (ret) 2506 return 0; 2507 2508 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2509 uhash_entries = UDP_HTABLE_SIZE_MIN; 2510 return 1; 2511} 2512__setup("uhash_entries=", set_uhash_entries); 2513 2514void __init udp_table_init(struct udp_table *table, const char *name) 2515{ 2516 unsigned int i; 2517 2518 table->hash = alloc_large_system_hash(name, 2519 2 * sizeof(struct udp_hslot), 2520 uhash_entries, 2521 21, /* one slot per 2 MB */ 2522 0, 2523 &table->log, 2524 &table->mask, 2525 UDP_HTABLE_SIZE_MIN, 2526 64 * 1024); 2527 2528 table->hash2 = table->hash + (table->mask + 1); 2529 for (i = 0; i <= table->mask; i++) { 2530 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2531 table->hash[i].count = 0; 2532 spin_lock_init(&table->hash[i].lock); 2533 } 2534 for (i = 0; i <= table->mask; i++) { 2535 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2536 table->hash2[i].count = 0; 2537 spin_lock_init(&table->hash2[i].lock); 2538 } 2539} 2540 2541u32 udp_flow_hashrnd(void) 2542{ 2543 static u32 hashrnd __read_mostly; 2544 2545 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2546 2547 return hashrnd; 2548} 2549EXPORT_SYMBOL(udp_flow_hashrnd); 2550 2551void __init udp_init(void) 2552{ 2553 unsigned long limit; 2554 2555 udp_table_init(&udp_table, "UDP"); 2556 limit = nr_free_buffer_pages() / 8; 2557 limit = max(limit, 128UL); 2558 sysctl_udp_mem[0] = limit / 4 * 3; 2559 sysctl_udp_mem[1] = limit; 2560 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2561 2562 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2563 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2564} 2565