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 <linux/uaccess.h> 83 #include <asm/ioctls.h> 84 #include <linux/memblock.h> 85 #include <linux/highmem.h> 86 #include <linux/swap.h> 87 #include <linux/types.h> 88 #include <linux/fcntl.h> 89 #include <linux/module.h> 90 #include <linux/socket.h> 91 #include <linux/sockios.h> 92 #include <linux/igmp.h> 93 #include <linux/inetdevice.h> 94 #include <linux/in.h> 95 #include <linux/errno.h> 96 #include <linux/timer.h> 97 #include <linux/mm.h> 98 #include <linux/inet.h> 99 #include <linux/netdevice.h> 100 #include <linux/slab.h> 101 #include <net/tcp_states.h> 102 #include <linux/skbuff.h> 103 #include <linux/proc_fs.h> 104 #include <linux/seq_file.h> 105 #include <net/net_namespace.h> 106 #include <net/icmp.h> 107 #include <net/inet_hashtables.h> 108 #include <net/ip_tunnels.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 #include <net/sock_reuseport.h> 118 #include <net/addrconf.h> 119 #include <net/udp_tunnel.h> 120 121 struct udp_table udp_table __read_mostly; 122 EXPORT_SYMBOL(udp_table); 123 124 long sysctl_udp_mem[3] __read_mostly; 125 EXPORT_SYMBOL(sysctl_udp_mem); 126 127 atomic_long_t udp_memory_allocated; 128 EXPORT_SYMBOL(udp_memory_allocated); 129 130 #define MAX_UDP_PORTS 65536 131 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 132 133 /* IPCB reference means this can not be used from early demux */ 134 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb) 135 { 136 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 137 if (!net->ipv4.sysctl_udp_l3mdev_accept && 138 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 139 return true; 140 #endif 141 return false; 142 } 143 144 static int udp_lib_lport_inuse(struct net *net, __u16 num, 145 const struct udp_hslot *hslot, 146 unsigned long *bitmap, 147 struct sock *sk, unsigned int log) 148 { 149 struct sock *sk2; 150 kuid_t uid = sock_i_uid(sk); 151 152 sk_for_each(sk2, &hslot->head) { 153 if (net_eq(sock_net(sk2), net) && 154 sk2 != sk && 155 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 156 (!sk2->sk_reuse || !sk->sk_reuse) && 157 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 158 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 159 inet_rcv_saddr_equal(sk, sk2, true)) { 160 if (sk2->sk_reuseport && sk->sk_reuseport && 161 !rcu_access_pointer(sk->sk_reuseport_cb) && 162 uid_eq(uid, sock_i_uid(sk2))) { 163 if (!bitmap) 164 return 0; 165 } else { 166 if (!bitmap) 167 return 1; 168 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 169 bitmap); 170 } 171 } 172 } 173 return 0; 174 } 175 176 /* 177 * Note: we still hold spinlock of primary hash chain, so no other writer 178 * can insert/delete a socket with local_port == num 179 */ 180 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 181 struct udp_hslot *hslot2, 182 struct sock *sk) 183 { 184 struct sock *sk2; 185 kuid_t uid = sock_i_uid(sk); 186 int res = 0; 187 188 spin_lock(&hslot2->lock); 189 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 190 if (net_eq(sock_net(sk2), net) && 191 sk2 != sk && 192 (udp_sk(sk2)->udp_port_hash == num) && 193 (!sk2->sk_reuse || !sk->sk_reuse) && 194 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 195 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 196 inet_rcv_saddr_equal(sk, sk2, true)) { 197 if (sk2->sk_reuseport && sk->sk_reuseport && 198 !rcu_access_pointer(sk->sk_reuseport_cb) && 199 uid_eq(uid, sock_i_uid(sk2))) { 200 res = 0; 201 } else { 202 res = 1; 203 } 204 break; 205 } 206 } 207 spin_unlock(&hslot2->lock); 208 return res; 209 } 210 211 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 212 { 213 struct net *net = sock_net(sk); 214 kuid_t uid = sock_i_uid(sk); 215 struct sock *sk2; 216 217 sk_for_each(sk2, &hslot->head) { 218 if (net_eq(sock_net(sk2), net) && 219 sk2 != sk && 220 sk2->sk_family == sk->sk_family && 221 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 222 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 223 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 224 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 225 inet_rcv_saddr_equal(sk, sk2, false)) { 226 return reuseport_add_sock(sk, sk2, 227 inet_rcv_saddr_any(sk)); 228 } 229 } 230 231 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 232 } 233 234 /** 235 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 236 * 237 * @sk: socket struct in question 238 * @snum: port number to look up 239 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 240 * with NULL address 241 */ 242 int udp_lib_get_port(struct sock *sk, unsigned short snum, 243 unsigned int hash2_nulladdr) 244 { 245 struct udp_hslot *hslot, *hslot2; 246 struct udp_table *udptable = sk->sk_prot->h.udp_table; 247 int error = 1; 248 struct net *net = sock_net(sk); 249 250 if (!snum) { 251 int low, high, remaining; 252 unsigned int rand; 253 unsigned short first, last; 254 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 255 256 inet_get_local_port_range(net, &low, &high); 257 remaining = (high - low) + 1; 258 259 rand = prandom_u32(); 260 first = reciprocal_scale(rand, remaining) + low; 261 /* 262 * force rand to be an odd multiple of UDP_HTABLE_SIZE 263 */ 264 rand = (rand | 1) * (udptable->mask + 1); 265 last = first + udptable->mask + 1; 266 do { 267 hslot = udp_hashslot(udptable, net, first); 268 bitmap_zero(bitmap, PORTS_PER_CHAIN); 269 spin_lock_bh(&hslot->lock); 270 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 271 udptable->log); 272 273 snum = first; 274 /* 275 * Iterate on all possible values of snum for this hash. 276 * Using steps of an odd multiple of UDP_HTABLE_SIZE 277 * give us randomization and full range coverage. 278 */ 279 do { 280 if (low <= snum && snum <= high && 281 !test_bit(snum >> udptable->log, bitmap) && 282 !inet_is_local_reserved_port(net, snum)) 283 goto found; 284 snum += rand; 285 } while (snum != first); 286 spin_unlock_bh(&hslot->lock); 287 cond_resched(); 288 } while (++first != last); 289 goto fail; 290 } else { 291 hslot = udp_hashslot(udptable, net, snum); 292 spin_lock_bh(&hslot->lock); 293 if (hslot->count > 10) { 294 int exist; 295 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 296 297 slot2 &= udptable->mask; 298 hash2_nulladdr &= udptable->mask; 299 300 hslot2 = udp_hashslot2(udptable, slot2); 301 if (hslot->count < hslot2->count) 302 goto scan_primary_hash; 303 304 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 305 if (!exist && (hash2_nulladdr != slot2)) { 306 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 307 exist = udp_lib_lport_inuse2(net, snum, hslot2, 308 sk); 309 } 310 if (exist) 311 goto fail_unlock; 312 else 313 goto found; 314 } 315 scan_primary_hash: 316 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 317 goto fail_unlock; 318 } 319 found: 320 inet_sk(sk)->inet_num = snum; 321 udp_sk(sk)->udp_port_hash = snum; 322 udp_sk(sk)->udp_portaddr_hash ^= snum; 323 if (sk_unhashed(sk)) { 324 if (sk->sk_reuseport && 325 udp_reuseport_add_sock(sk, hslot)) { 326 inet_sk(sk)->inet_num = 0; 327 udp_sk(sk)->udp_port_hash = 0; 328 udp_sk(sk)->udp_portaddr_hash ^= snum; 329 goto fail_unlock; 330 } 331 332 sk_add_node_rcu(sk, &hslot->head); 333 hslot->count++; 334 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 335 336 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 337 spin_lock(&hslot2->lock); 338 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 339 sk->sk_family == AF_INET6) 340 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 341 &hslot2->head); 342 else 343 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 344 &hslot2->head); 345 hslot2->count++; 346 spin_unlock(&hslot2->lock); 347 } 348 sock_set_flag(sk, SOCK_RCU_FREE); 349 error = 0; 350 fail_unlock: 351 spin_unlock_bh(&hslot->lock); 352 fail: 353 return error; 354 } 355 EXPORT_SYMBOL(udp_lib_get_port); 356 357 int udp_v4_get_port(struct sock *sk, unsigned short snum) 358 { 359 unsigned int hash2_nulladdr = 360 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 361 unsigned int hash2_partial = 362 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 363 364 /* precompute partial secondary hash */ 365 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 366 return udp_lib_get_port(sk, snum, hash2_nulladdr); 367 } 368 369 static int compute_score(struct sock *sk, struct net *net, 370 __be32 saddr, __be16 sport, 371 __be32 daddr, unsigned short hnum, 372 int dif, int sdif, bool exact_dif) 373 { 374 int score; 375 struct inet_sock *inet; 376 bool dev_match; 377 378 if (!net_eq(sock_net(sk), net) || 379 udp_sk(sk)->udp_port_hash != hnum || 380 ipv6_only_sock(sk)) 381 return -1; 382 383 score = (sk->sk_family == PF_INET) ? 2 : 1; 384 inet = inet_sk(sk); 385 386 if (inet->inet_rcv_saddr) { 387 if (inet->inet_rcv_saddr != daddr) 388 return -1; 389 score += 4; 390 } 391 392 if (inet->inet_daddr) { 393 if (inet->inet_daddr != saddr) 394 return -1; 395 score += 4; 396 } 397 398 if (inet->inet_dport) { 399 if (inet->inet_dport != sport) 400 return -1; 401 score += 4; 402 } 403 404 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 405 dif, sdif); 406 if (!dev_match) 407 return -1; 408 score += 4; 409 410 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 411 score++; 412 return score; 413 } 414 415 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 416 const __u16 lport, const __be32 faddr, 417 const __be16 fport) 418 { 419 static u32 udp_ehash_secret __read_mostly; 420 421 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 422 423 return __inet_ehashfn(laddr, lport, faddr, fport, 424 udp_ehash_secret + net_hash_mix(net)); 425 } 426 427 /* called with rcu_read_lock() */ 428 static struct sock *udp4_lib_lookup2(struct net *net, 429 __be32 saddr, __be16 sport, 430 __be32 daddr, unsigned int hnum, 431 int dif, int sdif, bool exact_dif, 432 struct udp_hslot *hslot2, 433 struct sk_buff *skb) 434 { 435 struct sock *sk, *result; 436 int score, badness; 437 u32 hash = 0; 438 439 result = NULL; 440 badness = 0; 441 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 442 score = compute_score(sk, net, saddr, sport, 443 daddr, hnum, dif, sdif, exact_dif); 444 if (score > badness) { 445 if (sk->sk_reuseport) { 446 hash = udp_ehashfn(net, daddr, hnum, 447 saddr, sport); 448 result = reuseport_select_sock(sk, hash, skb, 449 sizeof(struct udphdr)); 450 if (result) 451 return result; 452 } 453 badness = score; 454 result = sk; 455 } 456 } 457 return result; 458 } 459 460 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 461 * harder than this. -DaveM 462 */ 463 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 464 __be16 sport, __be32 daddr, __be16 dport, int dif, 465 int sdif, struct udp_table *udptable, struct sk_buff *skb) 466 { 467 struct sock *sk, *result; 468 unsigned short hnum = ntohs(dport); 469 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 470 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 471 bool exact_dif = udp_lib_exact_dif_match(net, skb); 472 int score, badness; 473 u32 hash = 0; 474 475 if (hslot->count > 10) { 476 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 477 slot2 = hash2 & udptable->mask; 478 hslot2 = &udptable->hash2[slot2]; 479 if (hslot->count < hslot2->count) 480 goto begin; 481 482 result = udp4_lib_lookup2(net, saddr, sport, 483 daddr, hnum, dif, sdif, 484 exact_dif, hslot2, skb); 485 if (!result) { 486 unsigned int old_slot2 = slot2; 487 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 488 slot2 = hash2 & udptable->mask; 489 /* avoid searching the same slot again. */ 490 if (unlikely(slot2 == old_slot2)) 491 return result; 492 493 hslot2 = &udptable->hash2[slot2]; 494 if (hslot->count < hslot2->count) 495 goto begin; 496 497 result = udp4_lib_lookup2(net, saddr, sport, 498 daddr, hnum, dif, sdif, 499 exact_dif, hslot2, skb); 500 } 501 if (unlikely(IS_ERR(result))) 502 return NULL; 503 return result; 504 } 505 begin: 506 result = NULL; 507 badness = 0; 508 sk_for_each_rcu(sk, &hslot->head) { 509 score = compute_score(sk, net, saddr, sport, 510 daddr, hnum, dif, sdif, exact_dif); 511 if (score > badness) { 512 if (sk->sk_reuseport) { 513 hash = udp_ehashfn(net, daddr, hnum, 514 saddr, sport); 515 result = reuseport_select_sock(sk, hash, skb, 516 sizeof(struct udphdr)); 517 if (unlikely(IS_ERR(result))) 518 return NULL; 519 if (result) 520 return result; 521 } 522 result = sk; 523 badness = score; 524 } 525 } 526 return result; 527 } 528 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 529 530 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 531 __be16 sport, __be16 dport, 532 struct udp_table *udptable) 533 { 534 const struct iphdr *iph = ip_hdr(skb); 535 536 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 537 iph->daddr, dport, inet_iif(skb), 538 inet_sdif(skb), udptable, skb); 539 } 540 541 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, 542 __be16 sport, __be16 dport) 543 { 544 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); 545 } 546 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); 547 548 /* Must be called under rcu_read_lock(). 549 * Does increment socket refcount. 550 */ 551 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 552 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 553 __be32 daddr, __be16 dport, int dif) 554 { 555 struct sock *sk; 556 557 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 558 dif, 0, &udp_table, NULL); 559 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 560 sk = NULL; 561 return sk; 562 } 563 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 564 #endif 565 566 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 567 __be16 loc_port, __be32 loc_addr, 568 __be16 rmt_port, __be32 rmt_addr, 569 int dif, int sdif, unsigned short hnum) 570 { 571 struct inet_sock *inet = inet_sk(sk); 572 573 if (!net_eq(sock_net(sk), net) || 574 udp_sk(sk)->udp_port_hash != hnum || 575 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 576 (inet->inet_dport != rmt_port && inet->inet_dport) || 577 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 578 ipv6_only_sock(sk) || 579 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif && 580 sk->sk_bound_dev_if != sdif)) 581 return false; 582 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 583 return false; 584 return true; 585 } 586 587 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 588 void udp_encap_enable(void) 589 { 590 static_branch_inc(&udp_encap_needed_key); 591 } 592 EXPORT_SYMBOL(udp_encap_enable); 593 594 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 595 * through error handlers in encapsulations looking for a match. 596 */ 597 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 598 { 599 int i; 600 601 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 602 int (*handler)(struct sk_buff *skb, u32 info); 603 604 if (!iptun_encaps[i]) 605 continue; 606 handler = rcu_dereference(iptun_encaps[i]->err_handler); 607 if (handler && !handler(skb, info)) 608 return 0; 609 } 610 611 return -ENOENT; 612 } 613 614 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 615 * reversing source and destination port: this will match tunnels that force the 616 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 617 * lwtunnels might actually break this assumption by being configured with 618 * different destination ports on endpoints, in this case we won't be able to 619 * trace ICMP messages back to them. 620 * 621 * If this doesn't match any socket, probe tunnels with arbitrary destination 622 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 623 * we've sent packets to won't necessarily match the local destination port. 624 * 625 * Then ask the tunnel implementation to match the error against a valid 626 * association. 627 * 628 * Return an error if we can't find a match, the socket if we need further 629 * processing, zero otherwise. 630 */ 631 static struct sock *__udp4_lib_err_encap(struct net *net, 632 const struct iphdr *iph, 633 struct udphdr *uh, 634 struct udp_table *udptable, 635 struct sk_buff *skb, u32 info) 636 { 637 int network_offset, transport_offset; 638 struct sock *sk; 639 640 network_offset = skb_network_offset(skb); 641 transport_offset = skb_transport_offset(skb); 642 643 /* Network header needs to point to the outer IPv4 header inside ICMP */ 644 skb_reset_network_header(skb); 645 646 /* Transport header needs to point to the UDP header */ 647 skb_set_transport_header(skb, iph->ihl << 2); 648 649 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 650 iph->saddr, uh->dest, skb->dev->ifindex, 0, 651 udptable, NULL); 652 if (sk) { 653 int (*lookup)(struct sock *sk, struct sk_buff *skb); 654 struct udp_sock *up = udp_sk(sk); 655 656 lookup = READ_ONCE(up->encap_err_lookup); 657 if (!lookup || lookup(sk, skb)) 658 sk = NULL; 659 } 660 661 if (!sk) 662 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 663 664 skb_set_transport_header(skb, transport_offset); 665 skb_set_network_header(skb, network_offset); 666 667 return sk; 668 } 669 670 /* 671 * This routine is called by the ICMP module when it gets some 672 * sort of error condition. If err < 0 then the socket should 673 * be closed and the error returned to the user. If err > 0 674 * it's just the icmp type << 8 | icmp code. 675 * Header points to the ip header of the error packet. We move 676 * on past this. Then (as it used to claim before adjustment) 677 * header points to the first 8 bytes of the udp header. We need 678 * to find the appropriate port. 679 */ 680 681 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 682 { 683 struct inet_sock *inet; 684 const struct iphdr *iph = (const struct iphdr *)skb->data; 685 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 686 const int type = icmp_hdr(skb)->type; 687 const int code = icmp_hdr(skb)->code; 688 bool tunnel = false; 689 struct sock *sk; 690 int harderr; 691 int err; 692 struct net *net = dev_net(skb->dev); 693 694 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 695 iph->saddr, uh->source, skb->dev->ifindex, 696 inet_sdif(skb), udptable, NULL); 697 if (!sk) { 698 /* No socket for error: try tunnels before discarding */ 699 sk = ERR_PTR(-ENOENT); 700 if (static_branch_unlikely(&udp_encap_needed_key)) { 701 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb, 702 info); 703 if (!sk) 704 return 0; 705 } 706 707 if (IS_ERR(sk)) { 708 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 709 return PTR_ERR(sk); 710 } 711 712 tunnel = true; 713 } 714 715 err = 0; 716 harderr = 0; 717 inet = inet_sk(sk); 718 719 switch (type) { 720 default: 721 case ICMP_TIME_EXCEEDED: 722 err = EHOSTUNREACH; 723 break; 724 case ICMP_SOURCE_QUENCH: 725 goto out; 726 case ICMP_PARAMETERPROB: 727 err = EPROTO; 728 harderr = 1; 729 break; 730 case ICMP_DEST_UNREACH: 731 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 732 ipv4_sk_update_pmtu(skb, sk, info); 733 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 734 err = EMSGSIZE; 735 harderr = 1; 736 break; 737 } 738 goto out; 739 } 740 err = EHOSTUNREACH; 741 if (code <= NR_ICMP_UNREACH) { 742 harderr = icmp_err_convert[code].fatal; 743 err = icmp_err_convert[code].errno; 744 } 745 break; 746 case ICMP_REDIRECT: 747 ipv4_sk_redirect(skb, sk); 748 goto out; 749 } 750 751 /* 752 * RFC1122: OK. Passes ICMP errors back to application, as per 753 * 4.1.3.3. 754 */ 755 if (tunnel) { 756 /* ...not for tunnels though: we don't have a sending socket */ 757 goto out; 758 } 759 if (!inet->recverr) { 760 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 761 goto out; 762 } else 763 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 764 765 sk->sk_err = err; 766 sk->sk_error_report(sk); 767 out: 768 return 0; 769 } 770 771 int udp_err(struct sk_buff *skb, u32 info) 772 { 773 return __udp4_lib_err(skb, info, &udp_table); 774 } 775 776 /* 777 * Throw away all pending data and cancel the corking. Socket is locked. 778 */ 779 void udp_flush_pending_frames(struct sock *sk) 780 { 781 struct udp_sock *up = udp_sk(sk); 782 783 if (up->pending) { 784 up->len = 0; 785 up->pending = 0; 786 ip_flush_pending_frames(sk); 787 } 788 } 789 EXPORT_SYMBOL(udp_flush_pending_frames); 790 791 /** 792 * udp4_hwcsum - handle outgoing HW checksumming 793 * @skb: sk_buff containing the filled-in UDP header 794 * (checksum field must be zeroed out) 795 * @src: source IP address 796 * @dst: destination IP address 797 */ 798 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 799 { 800 struct udphdr *uh = udp_hdr(skb); 801 int offset = skb_transport_offset(skb); 802 int len = skb->len - offset; 803 int hlen = len; 804 __wsum csum = 0; 805 806 if (!skb_has_frag_list(skb)) { 807 /* 808 * Only one fragment on the socket. 809 */ 810 skb->csum_start = skb_transport_header(skb) - skb->head; 811 skb->csum_offset = offsetof(struct udphdr, check); 812 uh->check = ~csum_tcpudp_magic(src, dst, len, 813 IPPROTO_UDP, 0); 814 } else { 815 struct sk_buff *frags; 816 817 /* 818 * HW-checksum won't work as there are two or more 819 * fragments on the socket so that all csums of sk_buffs 820 * should be together 821 */ 822 skb_walk_frags(skb, frags) { 823 csum = csum_add(csum, frags->csum); 824 hlen -= frags->len; 825 } 826 827 csum = skb_checksum(skb, offset, hlen, csum); 828 skb->ip_summed = CHECKSUM_NONE; 829 830 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 831 if (uh->check == 0) 832 uh->check = CSUM_MANGLED_0; 833 } 834 } 835 EXPORT_SYMBOL_GPL(udp4_hwcsum); 836 837 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 838 * for the simple case like when setting the checksum for a UDP tunnel. 839 */ 840 void udp_set_csum(bool nocheck, struct sk_buff *skb, 841 __be32 saddr, __be32 daddr, int len) 842 { 843 struct udphdr *uh = udp_hdr(skb); 844 845 if (nocheck) { 846 uh->check = 0; 847 } else if (skb_is_gso(skb)) { 848 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 849 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 850 uh->check = 0; 851 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 852 if (uh->check == 0) 853 uh->check = CSUM_MANGLED_0; 854 } else { 855 skb->ip_summed = CHECKSUM_PARTIAL; 856 skb->csum_start = skb_transport_header(skb) - skb->head; 857 skb->csum_offset = offsetof(struct udphdr, check); 858 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 859 } 860 } 861 EXPORT_SYMBOL(udp_set_csum); 862 863 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 864 struct inet_cork *cork) 865 { 866 struct sock *sk = skb->sk; 867 struct inet_sock *inet = inet_sk(sk); 868 struct udphdr *uh; 869 int err = 0; 870 int is_udplite = IS_UDPLITE(sk); 871 int offset = skb_transport_offset(skb); 872 int len = skb->len - offset; 873 __wsum csum = 0; 874 875 /* 876 * Create a UDP header 877 */ 878 uh = udp_hdr(skb); 879 uh->source = inet->inet_sport; 880 uh->dest = fl4->fl4_dport; 881 uh->len = htons(len); 882 uh->check = 0; 883 884 if (cork->gso_size) { 885 const int hlen = skb_network_header_len(skb) + 886 sizeof(struct udphdr); 887 888 if (hlen + cork->gso_size > cork->fragsize) 889 return -EINVAL; 890 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) 891 return -EINVAL; 892 if (sk->sk_no_check_tx) 893 return -EINVAL; 894 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 895 dst_xfrm(skb_dst(skb))) 896 return -EIO; 897 898 skb_shinfo(skb)->gso_size = cork->gso_size; 899 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 900 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh), 901 cork->gso_size); 902 goto csum_partial; 903 } 904 905 if (is_udplite) /* UDP-Lite */ 906 csum = udplite_csum(skb); 907 908 else if (sk->sk_no_check_tx) { /* UDP csum off */ 909 910 skb->ip_summed = CHECKSUM_NONE; 911 goto send; 912 913 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 914 csum_partial: 915 916 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 917 goto send; 918 919 } else 920 csum = udp_csum(skb); 921 922 /* add protocol-dependent pseudo-header */ 923 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 924 sk->sk_protocol, csum); 925 if (uh->check == 0) 926 uh->check = CSUM_MANGLED_0; 927 928 send: 929 err = ip_send_skb(sock_net(sk), skb); 930 if (err) { 931 if (err == -ENOBUFS && !inet->recverr) { 932 UDP_INC_STATS(sock_net(sk), 933 UDP_MIB_SNDBUFERRORS, is_udplite); 934 err = 0; 935 } 936 } else 937 UDP_INC_STATS(sock_net(sk), 938 UDP_MIB_OUTDATAGRAMS, is_udplite); 939 return err; 940 } 941 942 /* 943 * Push out all pending data as one UDP datagram. Socket is locked. 944 */ 945 int udp_push_pending_frames(struct sock *sk) 946 { 947 struct udp_sock *up = udp_sk(sk); 948 struct inet_sock *inet = inet_sk(sk); 949 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 950 struct sk_buff *skb; 951 int err = 0; 952 953 skb = ip_finish_skb(sk, fl4); 954 if (!skb) 955 goto out; 956 957 err = udp_send_skb(skb, fl4, &inet->cork.base); 958 959 out: 960 up->len = 0; 961 up->pending = 0; 962 return err; 963 } 964 EXPORT_SYMBOL(udp_push_pending_frames); 965 966 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 967 { 968 switch (cmsg->cmsg_type) { 969 case UDP_SEGMENT: 970 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 971 return -EINVAL; 972 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 973 return 0; 974 default: 975 return -EINVAL; 976 } 977 } 978 979 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 980 { 981 struct cmsghdr *cmsg; 982 bool need_ip = false; 983 int err; 984 985 for_each_cmsghdr(cmsg, msg) { 986 if (!CMSG_OK(msg, cmsg)) 987 return -EINVAL; 988 989 if (cmsg->cmsg_level != SOL_UDP) { 990 need_ip = true; 991 continue; 992 } 993 994 err = __udp_cmsg_send(cmsg, gso_size); 995 if (err) 996 return err; 997 } 998 999 return need_ip; 1000 } 1001 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1002 1003 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1004 { 1005 struct inet_sock *inet = inet_sk(sk); 1006 struct udp_sock *up = udp_sk(sk); 1007 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1008 struct flowi4 fl4_stack; 1009 struct flowi4 *fl4; 1010 int ulen = len; 1011 struct ipcm_cookie ipc; 1012 struct rtable *rt = NULL; 1013 int free = 0; 1014 int connected = 0; 1015 __be32 daddr, faddr, saddr; 1016 __be16 dport; 1017 u8 tos; 1018 int err, is_udplite = IS_UDPLITE(sk); 1019 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 1020 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1021 struct sk_buff *skb; 1022 struct ip_options_data opt_copy; 1023 1024 if (len > 0xFFFF) 1025 return -EMSGSIZE; 1026 1027 /* 1028 * Check the flags. 1029 */ 1030 1031 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1032 return -EOPNOTSUPP; 1033 1034 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1035 1036 fl4 = &inet->cork.fl.u.ip4; 1037 if (up->pending) { 1038 /* 1039 * There are pending frames. 1040 * The socket lock must be held while it's corked. 1041 */ 1042 lock_sock(sk); 1043 if (likely(up->pending)) { 1044 if (unlikely(up->pending != AF_INET)) { 1045 release_sock(sk); 1046 return -EINVAL; 1047 } 1048 goto do_append_data; 1049 } 1050 release_sock(sk); 1051 } 1052 ulen += sizeof(struct udphdr); 1053 1054 /* 1055 * Get and verify the address. 1056 */ 1057 if (usin) { 1058 if (msg->msg_namelen < sizeof(*usin)) 1059 return -EINVAL; 1060 if (usin->sin_family != AF_INET) { 1061 if (usin->sin_family != AF_UNSPEC) 1062 return -EAFNOSUPPORT; 1063 } 1064 1065 daddr = usin->sin_addr.s_addr; 1066 dport = usin->sin_port; 1067 if (dport == 0) 1068 return -EINVAL; 1069 } else { 1070 if (sk->sk_state != TCP_ESTABLISHED) 1071 return -EDESTADDRREQ; 1072 daddr = inet->inet_daddr; 1073 dport = inet->inet_dport; 1074 /* Open fast path for connected socket. 1075 Route will not be used, if at least one option is set. 1076 */ 1077 connected = 1; 1078 } 1079 1080 ipcm_init_sk(&ipc, inet); 1081 ipc.gso_size = up->gso_size; 1082 1083 if (msg->msg_controllen) { 1084 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1085 if (err > 0) 1086 err = ip_cmsg_send(sk, msg, &ipc, 1087 sk->sk_family == AF_INET6); 1088 if (unlikely(err < 0)) { 1089 kfree(ipc.opt); 1090 return err; 1091 } 1092 if (ipc.opt) 1093 free = 1; 1094 connected = 0; 1095 } 1096 if (!ipc.opt) { 1097 struct ip_options_rcu *inet_opt; 1098 1099 rcu_read_lock(); 1100 inet_opt = rcu_dereference(inet->inet_opt); 1101 if (inet_opt) { 1102 memcpy(&opt_copy, inet_opt, 1103 sizeof(*inet_opt) + inet_opt->opt.optlen); 1104 ipc.opt = &opt_copy.opt; 1105 } 1106 rcu_read_unlock(); 1107 } 1108 1109 if (cgroup_bpf_enabled && !connected) { 1110 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1111 (struct sockaddr *)usin, &ipc.addr); 1112 if (err) 1113 goto out_free; 1114 if (usin) { 1115 if (usin->sin_port == 0) { 1116 /* BPF program set invalid port. Reject it. */ 1117 err = -EINVAL; 1118 goto out_free; 1119 } 1120 daddr = usin->sin_addr.s_addr; 1121 dport = usin->sin_port; 1122 } 1123 } 1124 1125 saddr = ipc.addr; 1126 ipc.addr = faddr = daddr; 1127 1128 if (ipc.opt && ipc.opt->opt.srr) { 1129 if (!daddr) { 1130 err = -EINVAL; 1131 goto out_free; 1132 } 1133 faddr = ipc.opt->opt.faddr; 1134 connected = 0; 1135 } 1136 tos = get_rttos(&ipc, inet); 1137 if (sock_flag(sk, SOCK_LOCALROUTE) || 1138 (msg->msg_flags & MSG_DONTROUTE) || 1139 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1140 tos |= RTO_ONLINK; 1141 connected = 0; 1142 } 1143 1144 if (ipv4_is_multicast(daddr)) { 1145 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1146 ipc.oif = inet->mc_index; 1147 if (!saddr) 1148 saddr = inet->mc_addr; 1149 connected = 0; 1150 } else if (!ipc.oif) { 1151 ipc.oif = inet->uc_index; 1152 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1153 /* oif is set, packet is to local broadcast and 1154 * and uc_index is set. oif is most likely set 1155 * by sk_bound_dev_if. If uc_index != oif check if the 1156 * oif is an L3 master and uc_index is an L3 slave. 1157 * If so, we want to allow the send using the uc_index. 1158 */ 1159 if (ipc.oif != inet->uc_index && 1160 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1161 inet->uc_index)) { 1162 ipc.oif = inet->uc_index; 1163 } 1164 } 1165 1166 if (connected) 1167 rt = (struct rtable *)sk_dst_check(sk, 0); 1168 1169 if (!rt) { 1170 struct net *net = sock_net(sk); 1171 __u8 flow_flags = inet_sk_flowi_flags(sk); 1172 1173 fl4 = &fl4_stack; 1174 1175 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1176 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1177 flow_flags, 1178 faddr, saddr, dport, inet->inet_sport, 1179 sk->sk_uid); 1180 1181 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1182 rt = ip_route_output_flow(net, fl4, sk); 1183 if (IS_ERR(rt)) { 1184 err = PTR_ERR(rt); 1185 rt = NULL; 1186 if (err == -ENETUNREACH) 1187 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1188 goto out; 1189 } 1190 1191 err = -EACCES; 1192 if ((rt->rt_flags & RTCF_BROADCAST) && 1193 !sock_flag(sk, SOCK_BROADCAST)) 1194 goto out; 1195 if (connected) 1196 sk_dst_set(sk, dst_clone(&rt->dst)); 1197 } 1198 1199 if (msg->msg_flags&MSG_CONFIRM) 1200 goto do_confirm; 1201 back_from_confirm: 1202 1203 saddr = fl4->saddr; 1204 if (!ipc.addr) 1205 daddr = ipc.addr = fl4->daddr; 1206 1207 /* Lockless fast path for the non-corking case. */ 1208 if (!corkreq) { 1209 struct inet_cork cork; 1210 1211 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1212 sizeof(struct udphdr), &ipc, &rt, 1213 &cork, msg->msg_flags); 1214 err = PTR_ERR(skb); 1215 if (!IS_ERR_OR_NULL(skb)) 1216 err = udp_send_skb(skb, fl4, &cork); 1217 goto out; 1218 } 1219 1220 lock_sock(sk); 1221 if (unlikely(up->pending)) { 1222 /* The socket is already corked while preparing it. */ 1223 /* ... which is an evident application bug. --ANK */ 1224 release_sock(sk); 1225 1226 net_dbg_ratelimited("socket already corked\n"); 1227 err = -EINVAL; 1228 goto out; 1229 } 1230 /* 1231 * Now cork the socket to pend data. 1232 */ 1233 fl4 = &inet->cork.fl.u.ip4; 1234 fl4->daddr = daddr; 1235 fl4->saddr = saddr; 1236 fl4->fl4_dport = dport; 1237 fl4->fl4_sport = inet->inet_sport; 1238 up->pending = AF_INET; 1239 1240 do_append_data: 1241 up->len += ulen; 1242 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1243 sizeof(struct udphdr), &ipc, &rt, 1244 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1245 if (err) 1246 udp_flush_pending_frames(sk); 1247 else if (!corkreq) 1248 err = udp_push_pending_frames(sk); 1249 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1250 up->pending = 0; 1251 release_sock(sk); 1252 1253 out: 1254 ip_rt_put(rt); 1255 out_free: 1256 if (free) 1257 kfree(ipc.opt); 1258 if (!err) 1259 return len; 1260 /* 1261 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1262 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1263 * we don't have a good statistic (IpOutDiscards but it can be too many 1264 * things). We could add another new stat but at least for now that 1265 * seems like overkill. 1266 */ 1267 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1268 UDP_INC_STATS(sock_net(sk), 1269 UDP_MIB_SNDBUFERRORS, is_udplite); 1270 } 1271 return err; 1272 1273 do_confirm: 1274 if (msg->msg_flags & MSG_PROBE) 1275 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1276 if (!(msg->msg_flags&MSG_PROBE) || len) 1277 goto back_from_confirm; 1278 err = 0; 1279 goto out; 1280 } 1281 EXPORT_SYMBOL(udp_sendmsg); 1282 1283 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1284 size_t size, int flags) 1285 { 1286 struct inet_sock *inet = inet_sk(sk); 1287 struct udp_sock *up = udp_sk(sk); 1288 int ret; 1289 1290 if (flags & MSG_SENDPAGE_NOTLAST) 1291 flags |= MSG_MORE; 1292 1293 if (!up->pending) { 1294 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1295 1296 /* Call udp_sendmsg to specify destination address which 1297 * sendpage interface can't pass. 1298 * This will succeed only when the socket is connected. 1299 */ 1300 ret = udp_sendmsg(sk, &msg, 0); 1301 if (ret < 0) 1302 return ret; 1303 } 1304 1305 lock_sock(sk); 1306 1307 if (unlikely(!up->pending)) { 1308 release_sock(sk); 1309 1310 net_dbg_ratelimited("cork failed\n"); 1311 return -EINVAL; 1312 } 1313 1314 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1315 page, offset, size, flags); 1316 if (ret == -EOPNOTSUPP) { 1317 release_sock(sk); 1318 return sock_no_sendpage(sk->sk_socket, page, offset, 1319 size, flags); 1320 } 1321 if (ret < 0) { 1322 udp_flush_pending_frames(sk); 1323 goto out; 1324 } 1325 1326 up->len += size; 1327 if (!(up->corkflag || (flags&MSG_MORE))) 1328 ret = udp_push_pending_frames(sk); 1329 if (!ret) 1330 ret = size; 1331 out: 1332 release_sock(sk); 1333 return ret; 1334 } 1335 1336 #define UDP_SKB_IS_STATELESS 0x80000000 1337 1338 static void udp_set_dev_scratch(struct sk_buff *skb) 1339 { 1340 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1341 1342 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1343 scratch->_tsize_state = skb->truesize; 1344 #if BITS_PER_LONG == 64 1345 scratch->len = skb->len; 1346 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1347 scratch->is_linear = !skb_is_nonlinear(skb); 1348 #endif 1349 /* all head states execept sp (dst, sk, nf) are always cleared by 1350 * udp_rcv() and we need to preserve secpath, if present, to eventually 1351 * process IP_CMSG_PASSSEC at recvmsg() time 1352 */ 1353 if (likely(!skb_sec_path(skb))) 1354 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1355 } 1356 1357 static int udp_skb_truesize(struct sk_buff *skb) 1358 { 1359 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1360 } 1361 1362 static bool udp_skb_has_head_state(struct sk_buff *skb) 1363 { 1364 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1365 } 1366 1367 /* fully reclaim rmem/fwd memory allocated for skb */ 1368 static void udp_rmem_release(struct sock *sk, int size, int partial, 1369 bool rx_queue_lock_held) 1370 { 1371 struct udp_sock *up = udp_sk(sk); 1372 struct sk_buff_head *sk_queue; 1373 int amt; 1374 1375 if (likely(partial)) { 1376 up->forward_deficit += size; 1377 size = up->forward_deficit; 1378 if (size < (sk->sk_rcvbuf >> 2)) 1379 return; 1380 } else { 1381 size += up->forward_deficit; 1382 } 1383 up->forward_deficit = 0; 1384 1385 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1386 * if the called don't held it already 1387 */ 1388 sk_queue = &sk->sk_receive_queue; 1389 if (!rx_queue_lock_held) 1390 spin_lock(&sk_queue->lock); 1391 1392 1393 sk->sk_forward_alloc += size; 1394 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); 1395 sk->sk_forward_alloc -= amt; 1396 1397 if (amt) 1398 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); 1399 1400 atomic_sub(size, &sk->sk_rmem_alloc); 1401 1402 /* this can save us from acquiring the rx queue lock on next receive */ 1403 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1404 1405 if (!rx_queue_lock_held) 1406 spin_unlock(&sk_queue->lock); 1407 } 1408 1409 /* Note: called with reader_queue.lock held. 1410 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1411 * This avoids a cache line miss while receive_queue lock is held. 1412 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1413 */ 1414 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1415 { 1416 prefetch(&skb->data); 1417 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1418 } 1419 EXPORT_SYMBOL(udp_skb_destructor); 1420 1421 /* as above, but the caller held the rx queue lock, too */ 1422 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1423 { 1424 prefetch(&skb->data); 1425 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1426 } 1427 1428 /* Idea of busylocks is to let producers grab an extra spinlock 1429 * to relieve pressure on the receive_queue spinlock shared by consumer. 1430 * Under flood, this means that only one producer can be in line 1431 * trying to acquire the receive_queue spinlock. 1432 * These busylock can be allocated on a per cpu manner, instead of a 1433 * per socket one (that would consume a cache line per socket) 1434 */ 1435 static int udp_busylocks_log __read_mostly; 1436 static spinlock_t *udp_busylocks __read_mostly; 1437 1438 static spinlock_t *busylock_acquire(void *ptr) 1439 { 1440 spinlock_t *busy; 1441 1442 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1443 spin_lock(busy); 1444 return busy; 1445 } 1446 1447 static void busylock_release(spinlock_t *busy) 1448 { 1449 if (busy) 1450 spin_unlock(busy); 1451 } 1452 1453 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1454 { 1455 struct sk_buff_head *list = &sk->sk_receive_queue; 1456 int rmem, delta, amt, err = -ENOMEM; 1457 spinlock_t *busy = NULL; 1458 int size; 1459 1460 /* try to avoid the costly atomic add/sub pair when the receive 1461 * queue is full; always allow at least a packet 1462 */ 1463 rmem = atomic_read(&sk->sk_rmem_alloc); 1464 if (rmem > sk->sk_rcvbuf) 1465 goto drop; 1466 1467 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1468 * having linear skbs : 1469 * - Reduce memory overhead and thus increase receive queue capacity 1470 * - Less cache line misses at copyout() time 1471 * - Less work at consume_skb() (less alien page frag freeing) 1472 */ 1473 if (rmem > (sk->sk_rcvbuf >> 1)) { 1474 skb_condense(skb); 1475 1476 busy = busylock_acquire(sk); 1477 } 1478 size = skb->truesize; 1479 udp_set_dev_scratch(skb); 1480 1481 /* we drop only if the receive buf is full and the receive 1482 * queue contains some other skb 1483 */ 1484 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1485 if (rmem > (size + sk->sk_rcvbuf)) 1486 goto uncharge_drop; 1487 1488 spin_lock(&list->lock); 1489 if (size >= sk->sk_forward_alloc) { 1490 amt = sk_mem_pages(size); 1491 delta = amt << SK_MEM_QUANTUM_SHIFT; 1492 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1493 err = -ENOBUFS; 1494 spin_unlock(&list->lock); 1495 goto uncharge_drop; 1496 } 1497 1498 sk->sk_forward_alloc += delta; 1499 } 1500 1501 sk->sk_forward_alloc -= size; 1502 1503 /* no need to setup a destructor, we will explicitly release the 1504 * forward allocated memory on dequeue 1505 */ 1506 sock_skb_set_dropcount(sk, skb); 1507 1508 __skb_queue_tail(list, skb); 1509 spin_unlock(&list->lock); 1510 1511 if (!sock_flag(sk, SOCK_DEAD)) 1512 sk->sk_data_ready(sk); 1513 1514 busylock_release(busy); 1515 return 0; 1516 1517 uncharge_drop: 1518 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1519 1520 drop: 1521 atomic_inc(&sk->sk_drops); 1522 busylock_release(busy); 1523 return err; 1524 } 1525 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1526 1527 void udp_destruct_sock(struct sock *sk) 1528 { 1529 /* reclaim completely the forward allocated memory */ 1530 struct udp_sock *up = udp_sk(sk); 1531 unsigned int total = 0; 1532 struct sk_buff *skb; 1533 1534 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1535 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1536 total += skb->truesize; 1537 kfree_skb(skb); 1538 } 1539 udp_rmem_release(sk, total, 0, true); 1540 1541 inet_sock_destruct(sk); 1542 } 1543 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1544 1545 int udp_init_sock(struct sock *sk) 1546 { 1547 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1548 sk->sk_destruct = udp_destruct_sock; 1549 return 0; 1550 } 1551 EXPORT_SYMBOL_GPL(udp_init_sock); 1552 1553 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1554 { 1555 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1556 bool slow = lock_sock_fast(sk); 1557 1558 sk_peek_offset_bwd(sk, len); 1559 unlock_sock_fast(sk, slow); 1560 } 1561 1562 if (!skb_unref(skb)) 1563 return; 1564 1565 /* In the more common cases we cleared the head states previously, 1566 * see __udp_queue_rcv_skb(). 1567 */ 1568 if (unlikely(udp_skb_has_head_state(skb))) 1569 skb_release_head_state(skb); 1570 __consume_stateless_skb(skb); 1571 } 1572 EXPORT_SYMBOL_GPL(skb_consume_udp); 1573 1574 static struct sk_buff *__first_packet_length(struct sock *sk, 1575 struct sk_buff_head *rcvq, 1576 int *total) 1577 { 1578 struct sk_buff *skb; 1579 1580 while ((skb = skb_peek(rcvq)) != NULL) { 1581 if (udp_lib_checksum_complete(skb)) { 1582 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1583 IS_UDPLITE(sk)); 1584 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1585 IS_UDPLITE(sk)); 1586 atomic_inc(&sk->sk_drops); 1587 __skb_unlink(skb, rcvq); 1588 *total += skb->truesize; 1589 kfree_skb(skb); 1590 } else { 1591 /* the csum related bits could be changed, refresh 1592 * the scratch area 1593 */ 1594 udp_set_dev_scratch(skb); 1595 break; 1596 } 1597 } 1598 return skb; 1599 } 1600 1601 /** 1602 * first_packet_length - return length of first packet in receive queue 1603 * @sk: socket 1604 * 1605 * Drops all bad checksum frames, until a valid one is found. 1606 * Returns the length of found skb, or -1 if none is found. 1607 */ 1608 static int first_packet_length(struct sock *sk) 1609 { 1610 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1611 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1612 struct sk_buff *skb; 1613 int total = 0; 1614 int res; 1615 1616 spin_lock_bh(&rcvq->lock); 1617 skb = __first_packet_length(sk, rcvq, &total); 1618 if (!skb && !skb_queue_empty(sk_queue)) { 1619 spin_lock(&sk_queue->lock); 1620 skb_queue_splice_tail_init(sk_queue, rcvq); 1621 spin_unlock(&sk_queue->lock); 1622 1623 skb = __first_packet_length(sk, rcvq, &total); 1624 } 1625 res = skb ? skb->len : -1; 1626 if (total) 1627 udp_rmem_release(sk, total, 1, false); 1628 spin_unlock_bh(&rcvq->lock); 1629 return res; 1630 } 1631 1632 /* 1633 * IOCTL requests applicable to the UDP protocol 1634 */ 1635 1636 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1637 { 1638 switch (cmd) { 1639 case SIOCOUTQ: 1640 { 1641 int amount = sk_wmem_alloc_get(sk); 1642 1643 return put_user(amount, (int __user *)arg); 1644 } 1645 1646 case SIOCINQ: 1647 { 1648 int amount = max_t(int, 0, first_packet_length(sk)); 1649 1650 return put_user(amount, (int __user *)arg); 1651 } 1652 1653 default: 1654 return -ENOIOCTLCMD; 1655 } 1656 1657 return 0; 1658 } 1659 EXPORT_SYMBOL(udp_ioctl); 1660 1661 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1662 int noblock, int *peeked, int *off, int *err) 1663 { 1664 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1665 struct sk_buff_head *queue; 1666 struct sk_buff *last; 1667 long timeo; 1668 int error; 1669 1670 queue = &udp_sk(sk)->reader_queue; 1671 flags |= noblock ? MSG_DONTWAIT : 0; 1672 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1673 do { 1674 struct sk_buff *skb; 1675 1676 error = sock_error(sk); 1677 if (error) 1678 break; 1679 1680 error = -EAGAIN; 1681 *peeked = 0; 1682 do { 1683 spin_lock_bh(&queue->lock); 1684 skb = __skb_try_recv_from_queue(sk, queue, flags, 1685 udp_skb_destructor, 1686 peeked, off, err, 1687 &last); 1688 if (skb) { 1689 spin_unlock_bh(&queue->lock); 1690 return skb; 1691 } 1692 1693 if (skb_queue_empty(sk_queue)) { 1694 spin_unlock_bh(&queue->lock); 1695 goto busy_check; 1696 } 1697 1698 /* refill the reader queue and walk it again 1699 * keep both queues locked to avoid re-acquiring 1700 * the sk_receive_queue lock if fwd memory scheduling 1701 * is needed. 1702 */ 1703 spin_lock(&sk_queue->lock); 1704 skb_queue_splice_tail_init(sk_queue, queue); 1705 1706 skb = __skb_try_recv_from_queue(sk, queue, flags, 1707 udp_skb_dtor_locked, 1708 peeked, off, err, 1709 &last); 1710 spin_unlock(&sk_queue->lock); 1711 spin_unlock_bh(&queue->lock); 1712 if (skb) 1713 return skb; 1714 1715 busy_check: 1716 if (!sk_can_busy_loop(sk)) 1717 break; 1718 1719 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1720 } while (!skb_queue_empty(sk_queue)); 1721 1722 /* sk_queue is empty, reader_queue may contain peeked packets */ 1723 } while (timeo && 1724 !__skb_wait_for_more_packets(sk, &error, &timeo, 1725 (struct sk_buff *)sk_queue)); 1726 1727 *err = error; 1728 return NULL; 1729 } 1730 EXPORT_SYMBOL(__skb_recv_udp); 1731 1732 /* 1733 * This should be easy, if there is something there we 1734 * return it, otherwise we block. 1735 */ 1736 1737 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1738 int flags, int *addr_len) 1739 { 1740 struct inet_sock *inet = inet_sk(sk); 1741 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1742 struct sk_buff *skb; 1743 unsigned int ulen, copied; 1744 int peeked, peeking, off; 1745 int err; 1746 int is_udplite = IS_UDPLITE(sk); 1747 bool checksum_valid = false; 1748 1749 if (flags & MSG_ERRQUEUE) 1750 return ip_recv_error(sk, msg, len, addr_len); 1751 1752 try_again: 1753 peeking = flags & MSG_PEEK; 1754 off = sk_peek_offset(sk, flags); 1755 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); 1756 if (!skb) 1757 return err; 1758 1759 ulen = udp_skb_len(skb); 1760 copied = len; 1761 if (copied > ulen - off) 1762 copied = ulen - off; 1763 else if (copied < ulen) 1764 msg->msg_flags |= MSG_TRUNC; 1765 1766 /* 1767 * If checksum is needed at all, try to do it while copying the 1768 * data. If the data is truncated, or if we only want a partial 1769 * coverage checksum (UDP-Lite), do it before the copy. 1770 */ 1771 1772 if (copied < ulen || peeking || 1773 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1774 checksum_valid = udp_skb_csum_unnecessary(skb) || 1775 !__udp_lib_checksum_complete(skb); 1776 if (!checksum_valid) 1777 goto csum_copy_err; 1778 } 1779 1780 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1781 if (udp_skb_is_linear(skb)) 1782 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1783 else 1784 err = skb_copy_datagram_msg(skb, off, msg, copied); 1785 } else { 1786 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1787 1788 if (err == -EINVAL) 1789 goto csum_copy_err; 1790 } 1791 1792 if (unlikely(err)) { 1793 if (!peeked) { 1794 atomic_inc(&sk->sk_drops); 1795 UDP_INC_STATS(sock_net(sk), 1796 UDP_MIB_INERRORS, is_udplite); 1797 } 1798 kfree_skb(skb); 1799 return err; 1800 } 1801 1802 if (!peeked) 1803 UDP_INC_STATS(sock_net(sk), 1804 UDP_MIB_INDATAGRAMS, is_udplite); 1805 1806 sock_recv_ts_and_drops(msg, sk, skb); 1807 1808 /* Copy the address. */ 1809 if (sin) { 1810 sin->sin_family = AF_INET; 1811 sin->sin_port = udp_hdr(skb)->source; 1812 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1813 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1814 *addr_len = sizeof(*sin); 1815 } 1816 1817 if (udp_sk(sk)->gro_enabled) 1818 udp_cmsg_recv(msg, sk, skb); 1819 1820 if (inet->cmsg_flags) 1821 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1822 1823 err = copied; 1824 if (flags & MSG_TRUNC) 1825 err = ulen; 1826 1827 skb_consume_udp(sk, skb, peeking ? -err : err); 1828 return err; 1829 1830 csum_copy_err: 1831 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1832 udp_skb_destructor)) { 1833 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1834 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1835 } 1836 kfree_skb(skb); 1837 1838 /* starting over for a new packet, but check if we need to yield */ 1839 cond_resched(); 1840 msg->msg_flags &= ~MSG_TRUNC; 1841 goto try_again; 1842 } 1843 1844 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1845 { 1846 /* This check is replicated from __ip4_datagram_connect() and 1847 * intended to prevent BPF program called below from accessing bytes 1848 * that are out of the bound specified by user in addr_len. 1849 */ 1850 if (addr_len < sizeof(struct sockaddr_in)) 1851 return -EINVAL; 1852 1853 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1854 } 1855 EXPORT_SYMBOL(udp_pre_connect); 1856 1857 int __udp_disconnect(struct sock *sk, int flags) 1858 { 1859 struct inet_sock *inet = inet_sk(sk); 1860 /* 1861 * 1003.1g - break association. 1862 */ 1863 1864 sk->sk_state = TCP_CLOSE; 1865 inet->inet_daddr = 0; 1866 inet->inet_dport = 0; 1867 sock_rps_reset_rxhash(sk); 1868 sk->sk_bound_dev_if = 0; 1869 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1870 inet_reset_saddr(sk); 1871 1872 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1873 sk->sk_prot->unhash(sk); 1874 inet->inet_sport = 0; 1875 } 1876 sk_dst_reset(sk); 1877 return 0; 1878 } 1879 EXPORT_SYMBOL(__udp_disconnect); 1880 1881 int udp_disconnect(struct sock *sk, int flags) 1882 { 1883 lock_sock(sk); 1884 __udp_disconnect(sk, flags); 1885 release_sock(sk); 1886 return 0; 1887 } 1888 EXPORT_SYMBOL(udp_disconnect); 1889 1890 void udp_lib_unhash(struct sock *sk) 1891 { 1892 if (sk_hashed(sk)) { 1893 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1894 struct udp_hslot *hslot, *hslot2; 1895 1896 hslot = udp_hashslot(udptable, sock_net(sk), 1897 udp_sk(sk)->udp_port_hash); 1898 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1899 1900 spin_lock_bh(&hslot->lock); 1901 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1902 reuseport_detach_sock(sk); 1903 if (sk_del_node_init_rcu(sk)) { 1904 hslot->count--; 1905 inet_sk(sk)->inet_num = 0; 1906 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1907 1908 spin_lock(&hslot2->lock); 1909 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1910 hslot2->count--; 1911 spin_unlock(&hslot2->lock); 1912 } 1913 spin_unlock_bh(&hslot->lock); 1914 } 1915 } 1916 EXPORT_SYMBOL(udp_lib_unhash); 1917 1918 /* 1919 * inet_rcv_saddr was changed, we must rehash secondary hash 1920 */ 1921 void udp_lib_rehash(struct sock *sk, u16 newhash) 1922 { 1923 if (sk_hashed(sk)) { 1924 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1925 struct udp_hslot *hslot, *hslot2, *nhslot2; 1926 1927 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1928 nhslot2 = udp_hashslot2(udptable, newhash); 1929 udp_sk(sk)->udp_portaddr_hash = newhash; 1930 1931 if (hslot2 != nhslot2 || 1932 rcu_access_pointer(sk->sk_reuseport_cb)) { 1933 hslot = udp_hashslot(udptable, sock_net(sk), 1934 udp_sk(sk)->udp_port_hash); 1935 /* we must lock primary chain too */ 1936 spin_lock_bh(&hslot->lock); 1937 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1938 reuseport_detach_sock(sk); 1939 1940 if (hslot2 != nhslot2) { 1941 spin_lock(&hslot2->lock); 1942 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1943 hslot2->count--; 1944 spin_unlock(&hslot2->lock); 1945 1946 spin_lock(&nhslot2->lock); 1947 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1948 &nhslot2->head); 1949 nhslot2->count++; 1950 spin_unlock(&nhslot2->lock); 1951 } 1952 1953 spin_unlock_bh(&hslot->lock); 1954 } 1955 } 1956 } 1957 EXPORT_SYMBOL(udp_lib_rehash); 1958 1959 static void udp_v4_rehash(struct sock *sk) 1960 { 1961 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 1962 inet_sk(sk)->inet_rcv_saddr, 1963 inet_sk(sk)->inet_num); 1964 udp_lib_rehash(sk, new_hash); 1965 } 1966 1967 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1968 { 1969 int rc; 1970 1971 if (inet_sk(sk)->inet_daddr) { 1972 sock_rps_save_rxhash(sk, skb); 1973 sk_mark_napi_id(sk, skb); 1974 sk_incoming_cpu_update(sk); 1975 } else { 1976 sk_mark_napi_id_once(sk, skb); 1977 } 1978 1979 rc = __udp_enqueue_schedule_skb(sk, skb); 1980 if (rc < 0) { 1981 int is_udplite = IS_UDPLITE(sk); 1982 1983 /* Note that an ENOMEM error is charged twice */ 1984 if (rc == -ENOMEM) 1985 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1986 is_udplite); 1987 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1988 kfree_skb(skb); 1989 trace_udp_fail_queue_rcv_skb(rc, sk); 1990 return -1; 1991 } 1992 1993 return 0; 1994 } 1995 1996 /* returns: 1997 * -1: error 1998 * 0: success 1999 * >0: "udp encap" protocol resubmission 2000 * 2001 * Note that in the success and error cases, the skb is assumed to 2002 * have either been requeued or freed. 2003 */ 2004 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2005 { 2006 struct udp_sock *up = udp_sk(sk); 2007 int is_udplite = IS_UDPLITE(sk); 2008 2009 /* 2010 * Charge it to the socket, dropping if the queue is full. 2011 */ 2012 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 2013 goto drop; 2014 nf_reset(skb); 2015 2016 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 2017 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2018 2019 /* 2020 * This is an encapsulation socket so pass the skb to 2021 * the socket's udp_encap_rcv() hook. Otherwise, just 2022 * fall through and pass this up the UDP socket. 2023 * up->encap_rcv() returns the following value: 2024 * =0 if skb was successfully passed to the encap 2025 * handler or was discarded by it. 2026 * >0 if skb should be passed on to UDP. 2027 * <0 if skb should be resubmitted as proto -N 2028 */ 2029 2030 /* if we're overly short, let UDP handle it */ 2031 encap_rcv = READ_ONCE(up->encap_rcv); 2032 if (encap_rcv) { 2033 int ret; 2034 2035 /* Verify checksum before giving to encap */ 2036 if (udp_lib_checksum_complete(skb)) 2037 goto csum_error; 2038 2039 ret = encap_rcv(sk, skb); 2040 if (ret <= 0) { 2041 __UDP_INC_STATS(sock_net(sk), 2042 UDP_MIB_INDATAGRAMS, 2043 is_udplite); 2044 return -ret; 2045 } 2046 } 2047 2048 /* FALLTHROUGH -- it's a UDP Packet */ 2049 } 2050 2051 /* 2052 * UDP-Lite specific tests, ignored on UDP sockets 2053 */ 2054 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2055 2056 /* 2057 * MIB statistics other than incrementing the error count are 2058 * disabled for the following two types of errors: these depend 2059 * on the application settings, not on the functioning of the 2060 * protocol stack as such. 2061 * 2062 * RFC 3828 here recommends (sec 3.3): "There should also be a 2063 * way ... to ... at least let the receiving application block 2064 * delivery of packets with coverage values less than a value 2065 * provided by the application." 2066 */ 2067 if (up->pcrlen == 0) { /* full coverage was set */ 2068 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2069 UDP_SKB_CB(skb)->cscov, skb->len); 2070 goto drop; 2071 } 2072 /* The next case involves violating the min. coverage requested 2073 * by the receiver. This is subtle: if receiver wants x and x is 2074 * greater than the buffersize/MTU then receiver will complain 2075 * that it wants x while sender emits packets of smaller size y. 2076 * Therefore the above ...()->partial_cov statement is essential. 2077 */ 2078 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2079 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2080 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2081 goto drop; 2082 } 2083 } 2084 2085 prefetch(&sk->sk_rmem_alloc); 2086 if (rcu_access_pointer(sk->sk_filter) && 2087 udp_lib_checksum_complete(skb)) 2088 goto csum_error; 2089 2090 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 2091 goto drop; 2092 2093 udp_csum_pull_header(skb); 2094 2095 ipv4_pktinfo_prepare(sk, skb); 2096 return __udp_queue_rcv_skb(sk, skb); 2097 2098 csum_error: 2099 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2100 drop: 2101 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2102 atomic_inc(&sk->sk_drops); 2103 kfree_skb(skb); 2104 return -1; 2105 } 2106 2107 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2108 { 2109 struct sk_buff *next, *segs; 2110 int ret; 2111 2112 if (likely(!udp_unexpected_gso(sk, skb))) 2113 return udp_queue_rcv_one_skb(sk, skb); 2114 2115 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET); 2116 __skb_push(skb, -skb_mac_offset(skb)); 2117 segs = udp_rcv_segment(sk, skb, true); 2118 for (skb = segs; skb; skb = next) { 2119 next = skb->next; 2120 __skb_pull(skb, skb_transport_offset(skb)); 2121 ret = udp_queue_rcv_one_skb(sk, skb); 2122 if (ret > 0) 2123 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret); 2124 } 2125 return 0; 2126 } 2127 2128 /* For TCP sockets, sk_rx_dst is protected by socket lock 2129 * For UDP, we use xchg() to guard against concurrent changes. 2130 */ 2131 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2132 { 2133 struct dst_entry *old; 2134 2135 if (dst_hold_safe(dst)) { 2136 old = xchg(&sk->sk_rx_dst, dst); 2137 dst_release(old); 2138 return old != dst; 2139 } 2140 return false; 2141 } 2142 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2143 2144 /* 2145 * Multicasts and broadcasts go to each listener. 2146 * 2147 * Note: called only from the BH handler context. 2148 */ 2149 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2150 struct udphdr *uh, 2151 __be32 saddr, __be32 daddr, 2152 struct udp_table *udptable, 2153 int proto) 2154 { 2155 struct sock *sk, *first = NULL; 2156 unsigned short hnum = ntohs(uh->dest); 2157 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2158 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2159 unsigned int offset = offsetof(typeof(*sk), sk_node); 2160 int dif = skb->dev->ifindex; 2161 int sdif = inet_sdif(skb); 2162 struct hlist_node *node; 2163 struct sk_buff *nskb; 2164 2165 if (use_hash2) { 2166 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2167 udptable->mask; 2168 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2169 start_lookup: 2170 hslot = &udptable->hash2[hash2]; 2171 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2172 } 2173 2174 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2175 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2176 uh->source, saddr, dif, sdif, hnum)) 2177 continue; 2178 2179 if (!first) { 2180 first = sk; 2181 continue; 2182 } 2183 nskb = skb_clone(skb, GFP_ATOMIC); 2184 2185 if (unlikely(!nskb)) { 2186 atomic_inc(&sk->sk_drops); 2187 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2188 IS_UDPLITE(sk)); 2189 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2190 IS_UDPLITE(sk)); 2191 continue; 2192 } 2193 if (udp_queue_rcv_skb(sk, nskb) > 0) 2194 consume_skb(nskb); 2195 } 2196 2197 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2198 if (use_hash2 && hash2 != hash2_any) { 2199 hash2 = hash2_any; 2200 goto start_lookup; 2201 } 2202 2203 if (first) { 2204 if (udp_queue_rcv_skb(first, skb) > 0) 2205 consume_skb(skb); 2206 } else { 2207 kfree_skb(skb); 2208 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2209 proto == IPPROTO_UDPLITE); 2210 } 2211 return 0; 2212 } 2213 2214 /* Initialize UDP checksum. If exited with zero value (success), 2215 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2216 * Otherwise, csum completion requires chacksumming packet body, 2217 * including udp header and folding it to skb->csum. 2218 */ 2219 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2220 int proto) 2221 { 2222 int err; 2223 2224 UDP_SKB_CB(skb)->partial_cov = 0; 2225 UDP_SKB_CB(skb)->cscov = skb->len; 2226 2227 if (proto == IPPROTO_UDPLITE) { 2228 err = udplite_checksum_init(skb, uh); 2229 if (err) 2230 return err; 2231 2232 if (UDP_SKB_CB(skb)->partial_cov) { 2233 skb->csum = inet_compute_pseudo(skb, proto); 2234 return 0; 2235 } 2236 } 2237 2238 /* Note, we are only interested in != 0 or == 0, thus the 2239 * force to int. 2240 */ 2241 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2242 inet_compute_pseudo); 2243 if (err) 2244 return err; 2245 2246 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2247 /* If SW calculated the value, we know it's bad */ 2248 if (skb->csum_complete_sw) 2249 return 1; 2250 2251 /* HW says the value is bad. Let's validate that. 2252 * skb->csum is no longer the full packet checksum, 2253 * so don't treat it as such. 2254 */ 2255 skb_checksum_complete_unset(skb); 2256 } 2257 2258 return 0; 2259 } 2260 2261 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2262 * return code conversion for ip layer consumption 2263 */ 2264 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2265 struct udphdr *uh) 2266 { 2267 int ret; 2268 2269 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2270 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 2271 inet_compute_pseudo); 2272 2273 ret = udp_queue_rcv_skb(sk, skb); 2274 2275 /* a return value > 0 means to resubmit the input, but 2276 * it wants the return to be -protocol, or 0 2277 */ 2278 if (ret > 0) 2279 return -ret; 2280 return 0; 2281 } 2282 2283 /* 2284 * All we need to do is get the socket, and then do a checksum. 2285 */ 2286 2287 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2288 int proto) 2289 { 2290 struct sock *sk; 2291 struct udphdr *uh; 2292 unsigned short ulen; 2293 struct rtable *rt = skb_rtable(skb); 2294 __be32 saddr, daddr; 2295 struct net *net = dev_net(skb->dev); 2296 2297 /* 2298 * Validate the packet. 2299 */ 2300 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2301 goto drop; /* No space for header. */ 2302 2303 uh = udp_hdr(skb); 2304 ulen = ntohs(uh->len); 2305 saddr = ip_hdr(skb)->saddr; 2306 daddr = ip_hdr(skb)->daddr; 2307 2308 if (ulen > skb->len) 2309 goto short_packet; 2310 2311 if (proto == IPPROTO_UDP) { 2312 /* UDP validates ulen. */ 2313 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2314 goto short_packet; 2315 uh = udp_hdr(skb); 2316 } 2317 2318 if (udp4_csum_init(skb, uh, proto)) 2319 goto csum_error; 2320 2321 sk = skb_steal_sock(skb); 2322 if (sk) { 2323 struct dst_entry *dst = skb_dst(skb); 2324 int ret; 2325 2326 if (unlikely(sk->sk_rx_dst != dst)) 2327 udp_sk_rx_dst_set(sk, dst); 2328 2329 ret = udp_unicast_rcv_skb(sk, skb, uh); 2330 sock_put(sk); 2331 return ret; 2332 } 2333 2334 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2335 return __udp4_lib_mcast_deliver(net, skb, uh, 2336 saddr, daddr, udptable, proto); 2337 2338 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2339 if (sk) 2340 return udp_unicast_rcv_skb(sk, skb, uh); 2341 2342 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2343 goto drop; 2344 nf_reset(skb); 2345 2346 /* No socket. Drop packet silently, if checksum is wrong */ 2347 if (udp_lib_checksum_complete(skb)) 2348 goto csum_error; 2349 2350 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2351 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2352 2353 /* 2354 * Hmm. We got an UDP packet to a port to which we 2355 * don't wanna listen. Ignore it. 2356 */ 2357 kfree_skb(skb); 2358 return 0; 2359 2360 short_packet: 2361 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2362 proto == IPPROTO_UDPLITE ? "Lite" : "", 2363 &saddr, ntohs(uh->source), 2364 ulen, skb->len, 2365 &daddr, ntohs(uh->dest)); 2366 goto drop; 2367 2368 csum_error: 2369 /* 2370 * RFC1122: OK. Discards the bad packet silently (as far as 2371 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2372 */ 2373 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2374 proto == IPPROTO_UDPLITE ? "Lite" : "", 2375 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2376 ulen); 2377 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2378 drop: 2379 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2380 kfree_skb(skb); 2381 return 0; 2382 } 2383 2384 /* We can only early demux multicast if there is a single matching socket. 2385 * If more than one socket found returns NULL 2386 */ 2387 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2388 __be16 loc_port, __be32 loc_addr, 2389 __be16 rmt_port, __be32 rmt_addr, 2390 int dif, int sdif) 2391 { 2392 struct sock *sk, *result; 2393 unsigned short hnum = ntohs(loc_port); 2394 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2395 struct udp_hslot *hslot = &udp_table.hash[slot]; 2396 2397 /* Do not bother scanning a too big list */ 2398 if (hslot->count > 10) 2399 return NULL; 2400 2401 result = NULL; 2402 sk_for_each_rcu(sk, &hslot->head) { 2403 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2404 rmt_port, rmt_addr, dif, sdif, hnum)) { 2405 if (result) 2406 return NULL; 2407 result = sk; 2408 } 2409 } 2410 2411 return result; 2412 } 2413 2414 /* For unicast we should only early demux connected sockets or we can 2415 * break forwarding setups. The chains here can be long so only check 2416 * if the first socket is an exact match and if not move on. 2417 */ 2418 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2419 __be16 loc_port, __be32 loc_addr, 2420 __be16 rmt_port, __be32 rmt_addr, 2421 int dif, int sdif) 2422 { 2423 unsigned short hnum = ntohs(loc_port); 2424 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2425 unsigned int slot2 = hash2 & udp_table.mask; 2426 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2427 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2428 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2429 struct sock *sk; 2430 2431 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2432 if (INET_MATCH(sk, net, acookie, rmt_addr, 2433 loc_addr, ports, dif, sdif)) 2434 return sk; 2435 /* Only check first socket in chain */ 2436 break; 2437 } 2438 return NULL; 2439 } 2440 2441 int udp_v4_early_demux(struct sk_buff *skb) 2442 { 2443 struct net *net = dev_net(skb->dev); 2444 struct in_device *in_dev = NULL; 2445 const struct iphdr *iph; 2446 const struct udphdr *uh; 2447 struct sock *sk = NULL; 2448 struct dst_entry *dst; 2449 int dif = skb->dev->ifindex; 2450 int sdif = inet_sdif(skb); 2451 int ours; 2452 2453 /* validate the packet */ 2454 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2455 return 0; 2456 2457 iph = ip_hdr(skb); 2458 uh = udp_hdr(skb); 2459 2460 if (skb->pkt_type == PACKET_MULTICAST) { 2461 in_dev = __in_dev_get_rcu(skb->dev); 2462 2463 if (!in_dev) 2464 return 0; 2465 2466 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2467 iph->protocol); 2468 if (!ours) 2469 return 0; 2470 2471 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2472 uh->source, iph->saddr, 2473 dif, sdif); 2474 } else if (skb->pkt_type == PACKET_HOST) { 2475 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2476 uh->source, iph->saddr, dif, sdif); 2477 } 2478 2479 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2480 return 0; 2481 2482 skb->sk = sk; 2483 skb->destructor = sock_efree; 2484 dst = READ_ONCE(sk->sk_rx_dst); 2485 2486 if (dst) 2487 dst = dst_check(dst, 0); 2488 if (dst) { 2489 u32 itag = 0; 2490 2491 /* set noref for now. 2492 * any place which wants to hold dst has to call 2493 * dst_hold_safe() 2494 */ 2495 skb_dst_set_noref(skb, dst); 2496 2497 /* for unconnected multicast sockets we need to validate 2498 * the source on each packet 2499 */ 2500 if (!inet_sk(sk)->inet_daddr && in_dev) 2501 return ip_mc_validate_source(skb, iph->daddr, 2502 iph->saddr, iph->tos, 2503 skb->dev, in_dev, &itag); 2504 } 2505 return 0; 2506 } 2507 2508 int udp_rcv(struct sk_buff *skb) 2509 { 2510 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2511 } 2512 2513 void udp_destroy_sock(struct sock *sk) 2514 { 2515 struct udp_sock *up = udp_sk(sk); 2516 bool slow = lock_sock_fast(sk); 2517 udp_flush_pending_frames(sk); 2518 unlock_sock_fast(sk, slow); 2519 if (static_branch_unlikely(&udp_encap_needed_key)) { 2520 if (up->encap_type) { 2521 void (*encap_destroy)(struct sock *sk); 2522 encap_destroy = READ_ONCE(up->encap_destroy); 2523 if (encap_destroy) 2524 encap_destroy(sk); 2525 } 2526 if (up->encap_enabled) 2527 static_branch_dec(&udp_encap_needed_key); 2528 } 2529 } 2530 2531 /* 2532 * Socket option code for UDP 2533 */ 2534 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2535 char __user *optval, unsigned int optlen, 2536 int (*push_pending_frames)(struct sock *)) 2537 { 2538 struct udp_sock *up = udp_sk(sk); 2539 int val, valbool; 2540 int err = 0; 2541 int is_udplite = IS_UDPLITE(sk); 2542 2543 if (optlen < sizeof(int)) 2544 return -EINVAL; 2545 2546 if (get_user(val, (int __user *)optval)) 2547 return -EFAULT; 2548 2549 valbool = val ? 1 : 0; 2550 2551 switch (optname) { 2552 case UDP_CORK: 2553 if (val != 0) { 2554 up->corkflag = 1; 2555 } else { 2556 up->corkflag = 0; 2557 lock_sock(sk); 2558 push_pending_frames(sk); 2559 release_sock(sk); 2560 } 2561 break; 2562 2563 case UDP_ENCAP: 2564 switch (val) { 2565 case 0: 2566 case UDP_ENCAP_ESPINUDP: 2567 case UDP_ENCAP_ESPINUDP_NON_IKE: 2568 up->encap_rcv = xfrm4_udp_encap_rcv; 2569 /* FALLTHROUGH */ 2570 case UDP_ENCAP_L2TPINUDP: 2571 up->encap_type = val; 2572 lock_sock(sk); 2573 udp_tunnel_encap_enable(sk->sk_socket); 2574 release_sock(sk); 2575 break; 2576 default: 2577 err = -ENOPROTOOPT; 2578 break; 2579 } 2580 break; 2581 2582 case UDP_NO_CHECK6_TX: 2583 up->no_check6_tx = valbool; 2584 break; 2585 2586 case UDP_NO_CHECK6_RX: 2587 up->no_check6_rx = valbool; 2588 break; 2589 2590 case UDP_SEGMENT: 2591 if (val < 0 || val > USHRT_MAX) 2592 return -EINVAL; 2593 up->gso_size = val; 2594 break; 2595 2596 case UDP_GRO: 2597 lock_sock(sk); 2598 if (valbool) 2599 udp_tunnel_encap_enable(sk->sk_socket); 2600 up->gro_enabled = valbool; 2601 release_sock(sk); 2602 break; 2603 2604 /* 2605 * UDP-Lite's partial checksum coverage (RFC 3828). 2606 */ 2607 /* The sender sets actual checksum coverage length via this option. 2608 * The case coverage > packet length is handled by send module. */ 2609 case UDPLITE_SEND_CSCOV: 2610 if (!is_udplite) /* Disable the option on UDP sockets */ 2611 return -ENOPROTOOPT; 2612 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2613 val = 8; 2614 else if (val > USHRT_MAX) 2615 val = USHRT_MAX; 2616 up->pcslen = val; 2617 up->pcflag |= UDPLITE_SEND_CC; 2618 break; 2619 2620 /* The receiver specifies a minimum checksum coverage value. To make 2621 * sense, this should be set to at least 8 (as done below). If zero is 2622 * used, this again means full checksum coverage. */ 2623 case UDPLITE_RECV_CSCOV: 2624 if (!is_udplite) /* Disable the option on UDP sockets */ 2625 return -ENOPROTOOPT; 2626 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2627 val = 8; 2628 else if (val > USHRT_MAX) 2629 val = USHRT_MAX; 2630 up->pcrlen = val; 2631 up->pcflag |= UDPLITE_RECV_CC; 2632 break; 2633 2634 default: 2635 err = -ENOPROTOOPT; 2636 break; 2637 } 2638 2639 return err; 2640 } 2641 EXPORT_SYMBOL(udp_lib_setsockopt); 2642 2643 int udp_setsockopt(struct sock *sk, int level, int optname, 2644 char __user *optval, unsigned int optlen) 2645 { 2646 if (level == SOL_UDP || level == SOL_UDPLITE) 2647 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2648 udp_push_pending_frames); 2649 return ip_setsockopt(sk, level, optname, optval, optlen); 2650 } 2651 2652 #ifdef CONFIG_COMPAT 2653 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2654 char __user *optval, unsigned int optlen) 2655 { 2656 if (level == SOL_UDP || level == SOL_UDPLITE) 2657 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2658 udp_push_pending_frames); 2659 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2660 } 2661 #endif 2662 2663 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2664 char __user *optval, int __user *optlen) 2665 { 2666 struct udp_sock *up = udp_sk(sk); 2667 int val, len; 2668 2669 if (get_user(len, optlen)) 2670 return -EFAULT; 2671 2672 len = min_t(unsigned int, len, sizeof(int)); 2673 2674 if (len < 0) 2675 return -EINVAL; 2676 2677 switch (optname) { 2678 case UDP_CORK: 2679 val = up->corkflag; 2680 break; 2681 2682 case UDP_ENCAP: 2683 val = up->encap_type; 2684 break; 2685 2686 case UDP_NO_CHECK6_TX: 2687 val = up->no_check6_tx; 2688 break; 2689 2690 case UDP_NO_CHECK6_RX: 2691 val = up->no_check6_rx; 2692 break; 2693 2694 case UDP_SEGMENT: 2695 val = up->gso_size; 2696 break; 2697 2698 /* The following two cannot be changed on UDP sockets, the return is 2699 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2700 case UDPLITE_SEND_CSCOV: 2701 val = up->pcslen; 2702 break; 2703 2704 case UDPLITE_RECV_CSCOV: 2705 val = up->pcrlen; 2706 break; 2707 2708 default: 2709 return -ENOPROTOOPT; 2710 } 2711 2712 if (put_user(len, optlen)) 2713 return -EFAULT; 2714 if (copy_to_user(optval, &val, len)) 2715 return -EFAULT; 2716 return 0; 2717 } 2718 EXPORT_SYMBOL(udp_lib_getsockopt); 2719 2720 int udp_getsockopt(struct sock *sk, int level, int optname, 2721 char __user *optval, int __user *optlen) 2722 { 2723 if (level == SOL_UDP || level == SOL_UDPLITE) 2724 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2725 return ip_getsockopt(sk, level, optname, optval, optlen); 2726 } 2727 2728 #ifdef CONFIG_COMPAT 2729 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2730 char __user *optval, int __user *optlen) 2731 { 2732 if (level == SOL_UDP || level == SOL_UDPLITE) 2733 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2734 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2735 } 2736 #endif 2737 /** 2738 * udp_poll - wait for a UDP event. 2739 * @file - file struct 2740 * @sock - socket 2741 * @wait - poll table 2742 * 2743 * This is same as datagram poll, except for the special case of 2744 * blocking sockets. If application is using a blocking fd 2745 * and a packet with checksum error is in the queue; 2746 * then it could get return from select indicating data available 2747 * but then block when reading it. Add special case code 2748 * to work around these arguably broken applications. 2749 */ 2750 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2751 { 2752 __poll_t mask = datagram_poll(file, sock, wait); 2753 struct sock *sk = sock->sk; 2754 2755 if (!skb_queue_empty(&udp_sk(sk)->reader_queue)) 2756 mask |= EPOLLIN | EPOLLRDNORM; 2757 2758 /* Check for false positives due to checksum errors */ 2759 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2760 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2761 mask &= ~(EPOLLIN | EPOLLRDNORM); 2762 2763 return mask; 2764 2765 } 2766 EXPORT_SYMBOL(udp_poll); 2767 2768 int udp_abort(struct sock *sk, int err) 2769 { 2770 lock_sock(sk); 2771 2772 sk->sk_err = err; 2773 sk->sk_error_report(sk); 2774 __udp_disconnect(sk, 0); 2775 2776 release_sock(sk); 2777 2778 return 0; 2779 } 2780 EXPORT_SYMBOL_GPL(udp_abort); 2781 2782 struct proto udp_prot = { 2783 .name = "UDP", 2784 .owner = THIS_MODULE, 2785 .close = udp_lib_close, 2786 .pre_connect = udp_pre_connect, 2787 .connect = ip4_datagram_connect, 2788 .disconnect = udp_disconnect, 2789 .ioctl = udp_ioctl, 2790 .init = udp_init_sock, 2791 .destroy = udp_destroy_sock, 2792 .setsockopt = udp_setsockopt, 2793 .getsockopt = udp_getsockopt, 2794 .sendmsg = udp_sendmsg, 2795 .recvmsg = udp_recvmsg, 2796 .sendpage = udp_sendpage, 2797 .release_cb = ip4_datagram_release_cb, 2798 .hash = udp_lib_hash, 2799 .unhash = udp_lib_unhash, 2800 .rehash = udp_v4_rehash, 2801 .get_port = udp_v4_get_port, 2802 .memory_allocated = &udp_memory_allocated, 2803 .sysctl_mem = sysctl_udp_mem, 2804 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2805 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2806 .obj_size = sizeof(struct udp_sock), 2807 .h.udp_table = &udp_table, 2808 #ifdef CONFIG_COMPAT 2809 .compat_setsockopt = compat_udp_setsockopt, 2810 .compat_getsockopt = compat_udp_getsockopt, 2811 #endif 2812 .diag_destroy = udp_abort, 2813 }; 2814 EXPORT_SYMBOL(udp_prot); 2815 2816 /* ------------------------------------------------------------------------ */ 2817 #ifdef CONFIG_PROC_FS 2818 2819 static struct sock *udp_get_first(struct seq_file *seq, int start) 2820 { 2821 struct sock *sk; 2822 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2823 struct udp_iter_state *state = seq->private; 2824 struct net *net = seq_file_net(seq); 2825 2826 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; 2827 ++state->bucket) { 2828 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; 2829 2830 if (hlist_empty(&hslot->head)) 2831 continue; 2832 2833 spin_lock_bh(&hslot->lock); 2834 sk_for_each(sk, &hslot->head) { 2835 if (!net_eq(sock_net(sk), net)) 2836 continue; 2837 if (sk->sk_family == afinfo->family) 2838 goto found; 2839 } 2840 spin_unlock_bh(&hslot->lock); 2841 } 2842 sk = NULL; 2843 found: 2844 return sk; 2845 } 2846 2847 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2848 { 2849 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2850 struct udp_iter_state *state = seq->private; 2851 struct net *net = seq_file_net(seq); 2852 2853 do { 2854 sk = sk_next(sk); 2855 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family)); 2856 2857 if (!sk) { 2858 if (state->bucket <= afinfo->udp_table->mask) 2859 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 2860 return udp_get_first(seq, state->bucket + 1); 2861 } 2862 return sk; 2863 } 2864 2865 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2866 { 2867 struct sock *sk = udp_get_first(seq, 0); 2868 2869 if (sk) 2870 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2871 --pos; 2872 return pos ? NULL : sk; 2873 } 2874 2875 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2876 { 2877 struct udp_iter_state *state = seq->private; 2878 state->bucket = MAX_UDP_PORTS; 2879 2880 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2881 } 2882 EXPORT_SYMBOL(udp_seq_start); 2883 2884 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2885 { 2886 struct sock *sk; 2887 2888 if (v == SEQ_START_TOKEN) 2889 sk = udp_get_idx(seq, 0); 2890 else 2891 sk = udp_get_next(seq, v); 2892 2893 ++*pos; 2894 return sk; 2895 } 2896 EXPORT_SYMBOL(udp_seq_next); 2897 2898 void udp_seq_stop(struct seq_file *seq, void *v) 2899 { 2900 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2901 struct udp_iter_state *state = seq->private; 2902 2903 if (state->bucket <= afinfo->udp_table->mask) 2904 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 2905 } 2906 EXPORT_SYMBOL(udp_seq_stop); 2907 2908 /* ------------------------------------------------------------------------ */ 2909 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2910 int bucket) 2911 { 2912 struct inet_sock *inet = inet_sk(sp); 2913 __be32 dest = inet->inet_daddr; 2914 __be32 src = inet->inet_rcv_saddr; 2915 __u16 destp = ntohs(inet->inet_dport); 2916 __u16 srcp = ntohs(inet->inet_sport); 2917 2918 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2919 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d", 2920 bucket, src, srcp, dest, destp, sp->sk_state, 2921 sk_wmem_alloc_get(sp), 2922 udp_rqueue_get(sp), 2923 0, 0L, 0, 2924 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2925 0, sock_i_ino(sp), 2926 refcount_read(&sp->sk_refcnt), sp, 2927 atomic_read(&sp->sk_drops)); 2928 } 2929 2930 int udp4_seq_show(struct seq_file *seq, void *v) 2931 { 2932 seq_setwidth(seq, 127); 2933 if (v == SEQ_START_TOKEN) 2934 seq_puts(seq, " sl local_address rem_address st tx_queue " 2935 "rx_queue tr tm->when retrnsmt uid timeout " 2936 "inode ref pointer drops"); 2937 else { 2938 struct udp_iter_state *state = seq->private; 2939 2940 udp4_format_sock(v, seq, state->bucket); 2941 } 2942 seq_pad(seq, '\n'); 2943 return 0; 2944 } 2945 2946 const struct seq_operations udp_seq_ops = { 2947 .start = udp_seq_start, 2948 .next = udp_seq_next, 2949 .stop = udp_seq_stop, 2950 .show = udp4_seq_show, 2951 }; 2952 EXPORT_SYMBOL(udp_seq_ops); 2953 2954 static struct udp_seq_afinfo udp4_seq_afinfo = { 2955 .family = AF_INET, 2956 .udp_table = &udp_table, 2957 }; 2958 2959 static int __net_init udp4_proc_init_net(struct net *net) 2960 { 2961 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 2962 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 2963 return -ENOMEM; 2964 return 0; 2965 } 2966 2967 static void __net_exit udp4_proc_exit_net(struct net *net) 2968 { 2969 remove_proc_entry("udp", net->proc_net); 2970 } 2971 2972 static struct pernet_operations udp4_net_ops = { 2973 .init = udp4_proc_init_net, 2974 .exit = udp4_proc_exit_net, 2975 }; 2976 2977 int __init udp4_proc_init(void) 2978 { 2979 return register_pernet_subsys(&udp4_net_ops); 2980 } 2981 2982 void udp4_proc_exit(void) 2983 { 2984 unregister_pernet_subsys(&udp4_net_ops); 2985 } 2986 #endif /* CONFIG_PROC_FS */ 2987 2988 static __initdata unsigned long uhash_entries; 2989 static int __init set_uhash_entries(char *str) 2990 { 2991 ssize_t ret; 2992 2993 if (!str) 2994 return 0; 2995 2996 ret = kstrtoul(str, 0, &uhash_entries); 2997 if (ret) 2998 return 0; 2999 3000 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3001 uhash_entries = UDP_HTABLE_SIZE_MIN; 3002 return 1; 3003 } 3004 __setup("uhash_entries=", set_uhash_entries); 3005 3006 void __init udp_table_init(struct udp_table *table, const char *name) 3007 { 3008 unsigned int i; 3009 3010 table->hash = alloc_large_system_hash(name, 3011 2 * sizeof(struct udp_hslot), 3012 uhash_entries, 3013 21, /* one slot per 2 MB */ 3014 0, 3015 &table->log, 3016 &table->mask, 3017 UDP_HTABLE_SIZE_MIN, 3018 64 * 1024); 3019 3020 table->hash2 = table->hash + (table->mask + 1); 3021 for (i = 0; i <= table->mask; i++) { 3022 INIT_HLIST_HEAD(&table->hash[i].head); 3023 table->hash[i].count = 0; 3024 spin_lock_init(&table->hash[i].lock); 3025 } 3026 for (i = 0; i <= table->mask; i++) { 3027 INIT_HLIST_HEAD(&table->hash2[i].head); 3028 table->hash2[i].count = 0; 3029 spin_lock_init(&table->hash2[i].lock); 3030 } 3031 } 3032 3033 u32 udp_flow_hashrnd(void) 3034 { 3035 static u32 hashrnd __read_mostly; 3036 3037 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3038 3039 return hashrnd; 3040 } 3041 EXPORT_SYMBOL(udp_flow_hashrnd); 3042 3043 static void __udp_sysctl_init(struct net *net) 3044 { 3045 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; 3046 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; 3047 3048 #ifdef CONFIG_NET_L3_MASTER_DEV 3049 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3050 #endif 3051 } 3052 3053 static int __net_init udp_sysctl_init(struct net *net) 3054 { 3055 __udp_sysctl_init(net); 3056 return 0; 3057 } 3058 3059 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3060 .init = udp_sysctl_init, 3061 }; 3062 3063 void __init udp_init(void) 3064 { 3065 unsigned long limit; 3066 unsigned int i; 3067 3068 udp_table_init(&udp_table, "UDP"); 3069 limit = nr_free_buffer_pages() / 8; 3070 limit = max(limit, 128UL); 3071 sysctl_udp_mem[0] = limit / 4 * 3; 3072 sysctl_udp_mem[1] = limit; 3073 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3074 3075 __udp_sysctl_init(&init_net); 3076 3077 /* 16 spinlocks per cpu */ 3078 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3079 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3080 GFP_KERNEL); 3081 if (!udp_busylocks) 3082 panic("UDP: failed to alloc udp_busylocks\n"); 3083 for (i = 0; i < (1U << udp_busylocks_log); i++) 3084 spin_lock_init(udp_busylocks + i); 3085 3086 if (register_pernet_subsys(&udp_sysctl_ops)) 3087 panic("UDP: failed to init sysctl parameters.\n"); 3088 } 3089