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 if (sk->sk_rcv_saddr != daddr) 384 return -1; 385 386 score = (sk->sk_family == PF_INET) ? 2 : 1; 387 388 inet = inet_sk(sk); 389 if (inet->inet_daddr) { 390 if (inet->inet_daddr != saddr) 391 return -1; 392 score += 4; 393 } 394 395 if (inet->inet_dport) { 396 if (inet->inet_dport != sport) 397 return -1; 398 score += 4; 399 } 400 401 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 402 dif, sdif); 403 if (!dev_match) 404 return -1; 405 score += 4; 406 407 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 408 score++; 409 return score; 410 } 411 412 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 413 const __u16 lport, const __be32 faddr, 414 const __be16 fport) 415 { 416 static u32 udp_ehash_secret __read_mostly; 417 418 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 419 420 return __inet_ehashfn(laddr, lport, faddr, fport, 421 udp_ehash_secret + net_hash_mix(net)); 422 } 423 424 /* called with rcu_read_lock() */ 425 static struct sock *udp4_lib_lookup2(struct net *net, 426 __be32 saddr, __be16 sport, 427 __be32 daddr, unsigned int hnum, 428 int dif, int sdif, bool exact_dif, 429 struct udp_hslot *hslot2, 430 struct sk_buff *skb) 431 { 432 struct sock *sk, *result; 433 int score, badness; 434 u32 hash = 0; 435 436 result = NULL; 437 badness = 0; 438 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 439 score = compute_score(sk, net, saddr, sport, 440 daddr, hnum, dif, sdif, exact_dif); 441 if (score > badness) { 442 if (sk->sk_reuseport) { 443 hash = udp_ehashfn(net, daddr, hnum, 444 saddr, sport); 445 result = reuseport_select_sock(sk, hash, skb, 446 sizeof(struct udphdr)); 447 if (result) 448 return result; 449 } 450 badness = score; 451 result = sk; 452 } 453 } 454 return result; 455 } 456 457 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 458 * harder than this. -DaveM 459 */ 460 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 461 __be16 sport, __be32 daddr, __be16 dport, int dif, 462 int sdif, struct udp_table *udptable, struct sk_buff *skb) 463 { 464 struct sock *result; 465 unsigned short hnum = ntohs(dport); 466 unsigned int hash2, slot2; 467 struct udp_hslot *hslot2; 468 bool exact_dif = udp_lib_exact_dif_match(net, skb); 469 470 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 471 slot2 = hash2 & udptable->mask; 472 hslot2 = &udptable->hash2[slot2]; 473 474 result = udp4_lib_lookup2(net, saddr, sport, 475 daddr, hnum, dif, sdif, 476 exact_dif, hslot2, skb); 477 if (!result) { 478 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 479 slot2 = hash2 & udptable->mask; 480 hslot2 = &udptable->hash2[slot2]; 481 482 result = udp4_lib_lookup2(net, saddr, sport, 483 htonl(INADDR_ANY), hnum, dif, sdif, 484 exact_dif, hslot2, skb); 485 } 486 if (unlikely(IS_ERR(result))) 487 return NULL; 488 return result; 489 } 490 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 491 492 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 493 __be16 sport, __be16 dport, 494 struct udp_table *udptable) 495 { 496 const struct iphdr *iph = ip_hdr(skb); 497 498 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 499 iph->daddr, dport, inet_iif(skb), 500 inet_sdif(skb), udptable, skb); 501 } 502 503 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, 504 __be16 sport, __be16 dport) 505 { 506 const struct iphdr *iph = ip_hdr(skb); 507 508 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 509 iph->daddr, dport, inet_iif(skb), 510 inet_sdif(skb), &udp_table, NULL); 511 } 512 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); 513 514 /* Must be called under rcu_read_lock(). 515 * Does increment socket refcount. 516 */ 517 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 518 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 519 __be32 daddr, __be16 dport, int dif) 520 { 521 struct sock *sk; 522 523 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 524 dif, 0, &udp_table, NULL); 525 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 526 sk = NULL; 527 return sk; 528 } 529 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 530 #endif 531 532 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 533 __be16 loc_port, __be32 loc_addr, 534 __be16 rmt_port, __be32 rmt_addr, 535 int dif, int sdif, unsigned short hnum) 536 { 537 struct inet_sock *inet = inet_sk(sk); 538 539 if (!net_eq(sock_net(sk), net) || 540 udp_sk(sk)->udp_port_hash != hnum || 541 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 542 (inet->inet_dport != rmt_port && inet->inet_dport) || 543 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 544 ipv6_only_sock(sk) || 545 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif && 546 sk->sk_bound_dev_if != sdif)) 547 return false; 548 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 549 return false; 550 return true; 551 } 552 553 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 554 void udp_encap_enable(void) 555 { 556 static_branch_inc(&udp_encap_needed_key); 557 } 558 EXPORT_SYMBOL(udp_encap_enable); 559 560 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 561 * through error handlers in encapsulations looking for a match. 562 */ 563 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 564 { 565 int i; 566 567 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 568 int (*handler)(struct sk_buff *skb, u32 info); 569 const struct ip_tunnel_encap_ops *encap; 570 571 encap = rcu_dereference(iptun_encaps[i]); 572 if (!encap) 573 continue; 574 handler = encap->err_handler; 575 if (handler && !handler(skb, info)) 576 return 0; 577 } 578 579 return -ENOENT; 580 } 581 582 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 583 * reversing source and destination port: this will match tunnels that force the 584 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 585 * lwtunnels might actually break this assumption by being configured with 586 * different destination ports on endpoints, in this case we won't be able to 587 * trace ICMP messages back to them. 588 * 589 * If this doesn't match any socket, probe tunnels with arbitrary destination 590 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 591 * we've sent packets to won't necessarily match the local destination port. 592 * 593 * Then ask the tunnel implementation to match the error against a valid 594 * association. 595 * 596 * Return an error if we can't find a match, the socket if we need further 597 * processing, zero otherwise. 598 */ 599 static struct sock *__udp4_lib_err_encap(struct net *net, 600 const struct iphdr *iph, 601 struct udphdr *uh, 602 struct udp_table *udptable, 603 struct sk_buff *skb, u32 info) 604 { 605 int network_offset, transport_offset; 606 struct sock *sk; 607 608 network_offset = skb_network_offset(skb); 609 transport_offset = skb_transport_offset(skb); 610 611 /* Network header needs to point to the outer IPv4 header inside ICMP */ 612 skb_reset_network_header(skb); 613 614 /* Transport header needs to point to the UDP header */ 615 skb_set_transport_header(skb, iph->ihl << 2); 616 617 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 618 iph->saddr, uh->dest, skb->dev->ifindex, 0, 619 udptable, NULL); 620 if (sk) { 621 int (*lookup)(struct sock *sk, struct sk_buff *skb); 622 struct udp_sock *up = udp_sk(sk); 623 624 lookup = READ_ONCE(up->encap_err_lookup); 625 if (!lookup || lookup(sk, skb)) 626 sk = NULL; 627 } 628 629 if (!sk) 630 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 631 632 skb_set_transport_header(skb, transport_offset); 633 skb_set_network_header(skb, network_offset); 634 635 return sk; 636 } 637 638 /* 639 * This routine is called by the ICMP module when it gets some 640 * sort of error condition. If err < 0 then the socket should 641 * be closed and the error returned to the user. If err > 0 642 * it's just the icmp type << 8 | icmp code. 643 * Header points to the ip header of the error packet. We move 644 * on past this. Then (as it used to claim before adjustment) 645 * header points to the first 8 bytes of the udp header. We need 646 * to find the appropriate port. 647 */ 648 649 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 650 { 651 struct inet_sock *inet; 652 const struct iphdr *iph = (const struct iphdr *)skb->data; 653 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 654 const int type = icmp_hdr(skb)->type; 655 const int code = icmp_hdr(skb)->code; 656 bool tunnel = false; 657 struct sock *sk; 658 int harderr; 659 int err; 660 struct net *net = dev_net(skb->dev); 661 662 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 663 iph->saddr, uh->source, skb->dev->ifindex, 664 inet_sdif(skb), udptable, NULL); 665 if (!sk) { 666 /* No socket for error: try tunnels before discarding */ 667 sk = ERR_PTR(-ENOENT); 668 if (static_branch_unlikely(&udp_encap_needed_key)) { 669 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb, 670 info); 671 if (!sk) 672 return 0; 673 } 674 675 if (IS_ERR(sk)) { 676 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 677 return PTR_ERR(sk); 678 } 679 680 tunnel = true; 681 } 682 683 err = 0; 684 harderr = 0; 685 inet = inet_sk(sk); 686 687 switch (type) { 688 default: 689 case ICMP_TIME_EXCEEDED: 690 err = EHOSTUNREACH; 691 break; 692 case ICMP_SOURCE_QUENCH: 693 goto out; 694 case ICMP_PARAMETERPROB: 695 err = EPROTO; 696 harderr = 1; 697 break; 698 case ICMP_DEST_UNREACH: 699 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 700 ipv4_sk_update_pmtu(skb, sk, info); 701 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 702 err = EMSGSIZE; 703 harderr = 1; 704 break; 705 } 706 goto out; 707 } 708 err = EHOSTUNREACH; 709 if (code <= NR_ICMP_UNREACH) { 710 harderr = icmp_err_convert[code].fatal; 711 err = icmp_err_convert[code].errno; 712 } 713 break; 714 case ICMP_REDIRECT: 715 ipv4_sk_redirect(skb, sk); 716 goto out; 717 } 718 719 /* 720 * RFC1122: OK. Passes ICMP errors back to application, as per 721 * 4.1.3.3. 722 */ 723 if (tunnel) { 724 /* ...not for tunnels though: we don't have a sending socket */ 725 goto out; 726 } 727 if (!inet->recverr) { 728 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 729 goto out; 730 } else 731 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 732 733 sk->sk_err = err; 734 sk->sk_error_report(sk); 735 out: 736 return 0; 737 } 738 739 int udp_err(struct sk_buff *skb, u32 info) 740 { 741 return __udp4_lib_err(skb, info, &udp_table); 742 } 743 744 /* 745 * Throw away all pending data and cancel the corking. Socket is locked. 746 */ 747 void udp_flush_pending_frames(struct sock *sk) 748 { 749 struct udp_sock *up = udp_sk(sk); 750 751 if (up->pending) { 752 up->len = 0; 753 up->pending = 0; 754 ip_flush_pending_frames(sk); 755 } 756 } 757 EXPORT_SYMBOL(udp_flush_pending_frames); 758 759 /** 760 * udp4_hwcsum - handle outgoing HW checksumming 761 * @skb: sk_buff containing the filled-in UDP header 762 * (checksum field must be zeroed out) 763 * @src: source IP address 764 * @dst: destination IP address 765 */ 766 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 767 { 768 struct udphdr *uh = udp_hdr(skb); 769 int offset = skb_transport_offset(skb); 770 int len = skb->len - offset; 771 int hlen = len; 772 __wsum csum = 0; 773 774 if (!skb_has_frag_list(skb)) { 775 /* 776 * Only one fragment on the socket. 777 */ 778 skb->csum_start = skb_transport_header(skb) - skb->head; 779 skb->csum_offset = offsetof(struct udphdr, check); 780 uh->check = ~csum_tcpudp_magic(src, dst, len, 781 IPPROTO_UDP, 0); 782 } else { 783 struct sk_buff *frags; 784 785 /* 786 * HW-checksum won't work as there are two or more 787 * fragments on the socket so that all csums of sk_buffs 788 * should be together 789 */ 790 skb_walk_frags(skb, frags) { 791 csum = csum_add(csum, frags->csum); 792 hlen -= frags->len; 793 } 794 795 csum = skb_checksum(skb, offset, hlen, csum); 796 skb->ip_summed = CHECKSUM_NONE; 797 798 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 799 if (uh->check == 0) 800 uh->check = CSUM_MANGLED_0; 801 } 802 } 803 EXPORT_SYMBOL_GPL(udp4_hwcsum); 804 805 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 806 * for the simple case like when setting the checksum for a UDP tunnel. 807 */ 808 void udp_set_csum(bool nocheck, struct sk_buff *skb, 809 __be32 saddr, __be32 daddr, int len) 810 { 811 struct udphdr *uh = udp_hdr(skb); 812 813 if (nocheck) { 814 uh->check = 0; 815 } else if (skb_is_gso(skb)) { 816 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 817 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 818 uh->check = 0; 819 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 820 if (uh->check == 0) 821 uh->check = CSUM_MANGLED_0; 822 } else { 823 skb->ip_summed = CHECKSUM_PARTIAL; 824 skb->csum_start = skb_transport_header(skb) - skb->head; 825 skb->csum_offset = offsetof(struct udphdr, check); 826 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 827 } 828 } 829 EXPORT_SYMBOL(udp_set_csum); 830 831 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 832 struct inet_cork *cork) 833 { 834 struct sock *sk = skb->sk; 835 struct inet_sock *inet = inet_sk(sk); 836 struct udphdr *uh; 837 int err = 0; 838 int is_udplite = IS_UDPLITE(sk); 839 int offset = skb_transport_offset(skb); 840 int len = skb->len - offset; 841 __wsum csum = 0; 842 843 /* 844 * Create a UDP header 845 */ 846 uh = udp_hdr(skb); 847 uh->source = inet->inet_sport; 848 uh->dest = fl4->fl4_dport; 849 uh->len = htons(len); 850 uh->check = 0; 851 852 if (cork->gso_size) { 853 const int hlen = skb_network_header_len(skb) + 854 sizeof(struct udphdr); 855 856 if (hlen + cork->gso_size > cork->fragsize) { 857 kfree_skb(skb); 858 return -EINVAL; 859 } 860 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 861 kfree_skb(skb); 862 return -EINVAL; 863 } 864 if (sk->sk_no_check_tx) { 865 kfree_skb(skb); 866 return -EINVAL; 867 } 868 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 869 dst_xfrm(skb_dst(skb))) { 870 kfree_skb(skb); 871 return -EIO; 872 } 873 874 skb_shinfo(skb)->gso_size = cork->gso_size; 875 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 876 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh), 877 cork->gso_size); 878 goto csum_partial; 879 } 880 881 if (is_udplite) /* UDP-Lite */ 882 csum = udplite_csum(skb); 883 884 else if (sk->sk_no_check_tx) { /* UDP csum off */ 885 886 skb->ip_summed = CHECKSUM_NONE; 887 goto send; 888 889 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 890 csum_partial: 891 892 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 893 goto send; 894 895 } else 896 csum = udp_csum(skb); 897 898 /* add protocol-dependent pseudo-header */ 899 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 900 sk->sk_protocol, csum); 901 if (uh->check == 0) 902 uh->check = CSUM_MANGLED_0; 903 904 send: 905 err = ip_send_skb(sock_net(sk), skb); 906 if (err) { 907 if (err == -ENOBUFS && !inet->recverr) { 908 UDP_INC_STATS(sock_net(sk), 909 UDP_MIB_SNDBUFERRORS, is_udplite); 910 err = 0; 911 } 912 } else 913 UDP_INC_STATS(sock_net(sk), 914 UDP_MIB_OUTDATAGRAMS, is_udplite); 915 return err; 916 } 917 918 /* 919 * Push out all pending data as one UDP datagram. Socket is locked. 920 */ 921 int udp_push_pending_frames(struct sock *sk) 922 { 923 struct udp_sock *up = udp_sk(sk); 924 struct inet_sock *inet = inet_sk(sk); 925 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 926 struct sk_buff *skb; 927 int err = 0; 928 929 skb = ip_finish_skb(sk, fl4); 930 if (!skb) 931 goto out; 932 933 err = udp_send_skb(skb, fl4, &inet->cork.base); 934 935 out: 936 up->len = 0; 937 up->pending = 0; 938 return err; 939 } 940 EXPORT_SYMBOL(udp_push_pending_frames); 941 942 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 943 { 944 switch (cmsg->cmsg_type) { 945 case UDP_SEGMENT: 946 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 947 return -EINVAL; 948 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 949 return 0; 950 default: 951 return -EINVAL; 952 } 953 } 954 955 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 956 { 957 struct cmsghdr *cmsg; 958 bool need_ip = false; 959 int err; 960 961 for_each_cmsghdr(cmsg, msg) { 962 if (!CMSG_OK(msg, cmsg)) 963 return -EINVAL; 964 965 if (cmsg->cmsg_level != SOL_UDP) { 966 need_ip = true; 967 continue; 968 } 969 970 err = __udp_cmsg_send(cmsg, gso_size); 971 if (err) 972 return err; 973 } 974 975 return need_ip; 976 } 977 EXPORT_SYMBOL_GPL(udp_cmsg_send); 978 979 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 980 { 981 struct inet_sock *inet = inet_sk(sk); 982 struct udp_sock *up = udp_sk(sk); 983 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 984 struct flowi4 fl4_stack; 985 struct flowi4 *fl4; 986 int ulen = len; 987 struct ipcm_cookie ipc; 988 struct rtable *rt = NULL; 989 int free = 0; 990 int connected = 0; 991 __be32 daddr, faddr, saddr; 992 __be16 dport; 993 u8 tos; 994 int err, is_udplite = IS_UDPLITE(sk); 995 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 996 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 997 struct sk_buff *skb; 998 struct ip_options_data opt_copy; 999 1000 if (len > 0xFFFF) 1001 return -EMSGSIZE; 1002 1003 /* 1004 * Check the flags. 1005 */ 1006 1007 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1008 return -EOPNOTSUPP; 1009 1010 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1011 1012 fl4 = &inet->cork.fl.u.ip4; 1013 if (up->pending) { 1014 /* 1015 * There are pending frames. 1016 * The socket lock must be held while it's corked. 1017 */ 1018 lock_sock(sk); 1019 if (likely(up->pending)) { 1020 if (unlikely(up->pending != AF_INET)) { 1021 release_sock(sk); 1022 return -EINVAL; 1023 } 1024 goto do_append_data; 1025 } 1026 release_sock(sk); 1027 } 1028 ulen += sizeof(struct udphdr); 1029 1030 /* 1031 * Get and verify the address. 1032 */ 1033 if (usin) { 1034 if (msg->msg_namelen < sizeof(*usin)) 1035 return -EINVAL; 1036 if (usin->sin_family != AF_INET) { 1037 if (usin->sin_family != AF_UNSPEC) 1038 return -EAFNOSUPPORT; 1039 } 1040 1041 daddr = usin->sin_addr.s_addr; 1042 dport = usin->sin_port; 1043 if (dport == 0) 1044 return -EINVAL; 1045 } else { 1046 if (sk->sk_state != TCP_ESTABLISHED) 1047 return -EDESTADDRREQ; 1048 daddr = inet->inet_daddr; 1049 dport = inet->inet_dport; 1050 /* Open fast path for connected socket. 1051 Route will not be used, if at least one option is set. 1052 */ 1053 connected = 1; 1054 } 1055 1056 ipcm_init_sk(&ipc, inet); 1057 ipc.gso_size = up->gso_size; 1058 1059 if (msg->msg_controllen) { 1060 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1061 if (err > 0) 1062 err = ip_cmsg_send(sk, msg, &ipc, 1063 sk->sk_family == AF_INET6); 1064 if (unlikely(err < 0)) { 1065 kfree(ipc.opt); 1066 return err; 1067 } 1068 if (ipc.opt) 1069 free = 1; 1070 connected = 0; 1071 } 1072 if (!ipc.opt) { 1073 struct ip_options_rcu *inet_opt; 1074 1075 rcu_read_lock(); 1076 inet_opt = rcu_dereference(inet->inet_opt); 1077 if (inet_opt) { 1078 memcpy(&opt_copy, inet_opt, 1079 sizeof(*inet_opt) + inet_opt->opt.optlen); 1080 ipc.opt = &opt_copy.opt; 1081 } 1082 rcu_read_unlock(); 1083 } 1084 1085 if (cgroup_bpf_enabled && !connected) { 1086 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1087 (struct sockaddr *)usin, &ipc.addr); 1088 if (err) 1089 goto out_free; 1090 if (usin) { 1091 if (usin->sin_port == 0) { 1092 /* BPF program set invalid port. Reject it. */ 1093 err = -EINVAL; 1094 goto out_free; 1095 } 1096 daddr = usin->sin_addr.s_addr; 1097 dport = usin->sin_port; 1098 } 1099 } 1100 1101 saddr = ipc.addr; 1102 ipc.addr = faddr = daddr; 1103 1104 if (ipc.opt && ipc.opt->opt.srr) { 1105 if (!daddr) { 1106 err = -EINVAL; 1107 goto out_free; 1108 } 1109 faddr = ipc.opt->opt.faddr; 1110 connected = 0; 1111 } 1112 tos = get_rttos(&ipc, inet); 1113 if (sock_flag(sk, SOCK_LOCALROUTE) || 1114 (msg->msg_flags & MSG_DONTROUTE) || 1115 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1116 tos |= RTO_ONLINK; 1117 connected = 0; 1118 } 1119 1120 if (ipv4_is_multicast(daddr)) { 1121 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1122 ipc.oif = inet->mc_index; 1123 if (!saddr) 1124 saddr = inet->mc_addr; 1125 connected = 0; 1126 } else if (!ipc.oif) { 1127 ipc.oif = inet->uc_index; 1128 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1129 /* oif is set, packet is to local broadcast and 1130 * and uc_index is set. oif is most likely set 1131 * by sk_bound_dev_if. If uc_index != oif check if the 1132 * oif is an L3 master and uc_index is an L3 slave. 1133 * If so, we want to allow the send using the uc_index. 1134 */ 1135 if (ipc.oif != inet->uc_index && 1136 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1137 inet->uc_index)) { 1138 ipc.oif = inet->uc_index; 1139 } 1140 } 1141 1142 if (connected) 1143 rt = (struct rtable *)sk_dst_check(sk, 0); 1144 1145 if (!rt) { 1146 struct net *net = sock_net(sk); 1147 __u8 flow_flags = inet_sk_flowi_flags(sk); 1148 1149 fl4 = &fl4_stack; 1150 1151 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1152 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1153 flow_flags, 1154 faddr, saddr, dport, inet->inet_sport, 1155 sk->sk_uid); 1156 1157 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1158 rt = ip_route_output_flow(net, fl4, sk); 1159 if (IS_ERR(rt)) { 1160 err = PTR_ERR(rt); 1161 rt = NULL; 1162 if (err == -ENETUNREACH) 1163 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1164 goto out; 1165 } 1166 1167 err = -EACCES; 1168 if ((rt->rt_flags & RTCF_BROADCAST) && 1169 !sock_flag(sk, SOCK_BROADCAST)) 1170 goto out; 1171 if (connected) 1172 sk_dst_set(sk, dst_clone(&rt->dst)); 1173 } 1174 1175 if (msg->msg_flags&MSG_CONFIRM) 1176 goto do_confirm; 1177 back_from_confirm: 1178 1179 saddr = fl4->saddr; 1180 if (!ipc.addr) 1181 daddr = ipc.addr = fl4->daddr; 1182 1183 /* Lockless fast path for the non-corking case. */ 1184 if (!corkreq) { 1185 struct inet_cork cork; 1186 1187 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1188 sizeof(struct udphdr), &ipc, &rt, 1189 &cork, msg->msg_flags); 1190 err = PTR_ERR(skb); 1191 if (!IS_ERR_OR_NULL(skb)) 1192 err = udp_send_skb(skb, fl4, &cork); 1193 goto out; 1194 } 1195 1196 lock_sock(sk); 1197 if (unlikely(up->pending)) { 1198 /* The socket is already corked while preparing it. */ 1199 /* ... which is an evident application bug. --ANK */ 1200 release_sock(sk); 1201 1202 net_dbg_ratelimited("socket already corked\n"); 1203 err = -EINVAL; 1204 goto out; 1205 } 1206 /* 1207 * Now cork the socket to pend data. 1208 */ 1209 fl4 = &inet->cork.fl.u.ip4; 1210 fl4->daddr = daddr; 1211 fl4->saddr = saddr; 1212 fl4->fl4_dport = dport; 1213 fl4->fl4_sport = inet->inet_sport; 1214 up->pending = AF_INET; 1215 1216 do_append_data: 1217 up->len += ulen; 1218 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1219 sizeof(struct udphdr), &ipc, &rt, 1220 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1221 if (err) 1222 udp_flush_pending_frames(sk); 1223 else if (!corkreq) 1224 err = udp_push_pending_frames(sk); 1225 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1226 up->pending = 0; 1227 release_sock(sk); 1228 1229 out: 1230 ip_rt_put(rt); 1231 out_free: 1232 if (free) 1233 kfree(ipc.opt); 1234 if (!err) 1235 return len; 1236 /* 1237 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1238 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1239 * we don't have a good statistic (IpOutDiscards but it can be too many 1240 * things). We could add another new stat but at least for now that 1241 * seems like overkill. 1242 */ 1243 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1244 UDP_INC_STATS(sock_net(sk), 1245 UDP_MIB_SNDBUFERRORS, is_udplite); 1246 } 1247 return err; 1248 1249 do_confirm: 1250 if (msg->msg_flags & MSG_PROBE) 1251 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1252 if (!(msg->msg_flags&MSG_PROBE) || len) 1253 goto back_from_confirm; 1254 err = 0; 1255 goto out; 1256 } 1257 EXPORT_SYMBOL(udp_sendmsg); 1258 1259 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1260 size_t size, int flags) 1261 { 1262 struct inet_sock *inet = inet_sk(sk); 1263 struct udp_sock *up = udp_sk(sk); 1264 int ret; 1265 1266 if (flags & MSG_SENDPAGE_NOTLAST) 1267 flags |= MSG_MORE; 1268 1269 if (!up->pending) { 1270 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1271 1272 /* Call udp_sendmsg to specify destination address which 1273 * sendpage interface can't pass. 1274 * This will succeed only when the socket is connected. 1275 */ 1276 ret = udp_sendmsg(sk, &msg, 0); 1277 if (ret < 0) 1278 return ret; 1279 } 1280 1281 lock_sock(sk); 1282 1283 if (unlikely(!up->pending)) { 1284 release_sock(sk); 1285 1286 net_dbg_ratelimited("cork failed\n"); 1287 return -EINVAL; 1288 } 1289 1290 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1291 page, offset, size, flags); 1292 if (ret == -EOPNOTSUPP) { 1293 release_sock(sk); 1294 return sock_no_sendpage(sk->sk_socket, page, offset, 1295 size, flags); 1296 } 1297 if (ret < 0) { 1298 udp_flush_pending_frames(sk); 1299 goto out; 1300 } 1301 1302 up->len += size; 1303 if (!(up->corkflag || (flags&MSG_MORE))) 1304 ret = udp_push_pending_frames(sk); 1305 if (!ret) 1306 ret = size; 1307 out: 1308 release_sock(sk); 1309 return ret; 1310 } 1311 1312 #define UDP_SKB_IS_STATELESS 0x80000000 1313 1314 static void udp_set_dev_scratch(struct sk_buff *skb) 1315 { 1316 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1317 1318 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1319 scratch->_tsize_state = skb->truesize; 1320 #if BITS_PER_LONG == 64 1321 scratch->len = skb->len; 1322 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1323 scratch->is_linear = !skb_is_nonlinear(skb); 1324 #endif 1325 /* all head states execept sp (dst, sk, nf) are always cleared by 1326 * udp_rcv() and we need to preserve secpath, if present, to eventually 1327 * process IP_CMSG_PASSSEC at recvmsg() time 1328 */ 1329 if (likely(!skb_sec_path(skb))) 1330 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1331 } 1332 1333 static int udp_skb_truesize(struct sk_buff *skb) 1334 { 1335 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1336 } 1337 1338 static bool udp_skb_has_head_state(struct sk_buff *skb) 1339 { 1340 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1341 } 1342 1343 /* fully reclaim rmem/fwd memory allocated for skb */ 1344 static void udp_rmem_release(struct sock *sk, int size, int partial, 1345 bool rx_queue_lock_held) 1346 { 1347 struct udp_sock *up = udp_sk(sk); 1348 struct sk_buff_head *sk_queue; 1349 int amt; 1350 1351 if (likely(partial)) { 1352 up->forward_deficit += size; 1353 size = up->forward_deficit; 1354 if (size < (sk->sk_rcvbuf >> 2)) 1355 return; 1356 } else { 1357 size += up->forward_deficit; 1358 } 1359 up->forward_deficit = 0; 1360 1361 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1362 * if the called don't held it already 1363 */ 1364 sk_queue = &sk->sk_receive_queue; 1365 if (!rx_queue_lock_held) 1366 spin_lock(&sk_queue->lock); 1367 1368 1369 sk->sk_forward_alloc += size; 1370 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); 1371 sk->sk_forward_alloc -= amt; 1372 1373 if (amt) 1374 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); 1375 1376 atomic_sub(size, &sk->sk_rmem_alloc); 1377 1378 /* this can save us from acquiring the rx queue lock on next receive */ 1379 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1380 1381 if (!rx_queue_lock_held) 1382 spin_unlock(&sk_queue->lock); 1383 } 1384 1385 /* Note: called with reader_queue.lock held. 1386 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1387 * This avoids a cache line miss while receive_queue lock is held. 1388 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1389 */ 1390 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1391 { 1392 prefetch(&skb->data); 1393 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1394 } 1395 EXPORT_SYMBOL(udp_skb_destructor); 1396 1397 /* as above, but the caller held the rx queue lock, too */ 1398 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1399 { 1400 prefetch(&skb->data); 1401 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1402 } 1403 1404 /* Idea of busylocks is to let producers grab an extra spinlock 1405 * to relieve pressure on the receive_queue spinlock shared by consumer. 1406 * Under flood, this means that only one producer can be in line 1407 * trying to acquire the receive_queue spinlock. 1408 * These busylock can be allocated on a per cpu manner, instead of a 1409 * per socket one (that would consume a cache line per socket) 1410 */ 1411 static int udp_busylocks_log __read_mostly; 1412 static spinlock_t *udp_busylocks __read_mostly; 1413 1414 static spinlock_t *busylock_acquire(void *ptr) 1415 { 1416 spinlock_t *busy; 1417 1418 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1419 spin_lock(busy); 1420 return busy; 1421 } 1422 1423 static void busylock_release(spinlock_t *busy) 1424 { 1425 if (busy) 1426 spin_unlock(busy); 1427 } 1428 1429 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1430 { 1431 struct sk_buff_head *list = &sk->sk_receive_queue; 1432 int rmem, delta, amt, err = -ENOMEM; 1433 spinlock_t *busy = NULL; 1434 int size; 1435 1436 /* try to avoid the costly atomic add/sub pair when the receive 1437 * queue is full; always allow at least a packet 1438 */ 1439 rmem = atomic_read(&sk->sk_rmem_alloc); 1440 if (rmem > sk->sk_rcvbuf) 1441 goto drop; 1442 1443 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1444 * having linear skbs : 1445 * - Reduce memory overhead and thus increase receive queue capacity 1446 * - Less cache line misses at copyout() time 1447 * - Less work at consume_skb() (less alien page frag freeing) 1448 */ 1449 if (rmem > (sk->sk_rcvbuf >> 1)) { 1450 skb_condense(skb); 1451 1452 busy = busylock_acquire(sk); 1453 } 1454 size = skb->truesize; 1455 udp_set_dev_scratch(skb); 1456 1457 /* we drop only if the receive buf is full and the receive 1458 * queue contains some other skb 1459 */ 1460 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1461 if (rmem > (size + sk->sk_rcvbuf)) 1462 goto uncharge_drop; 1463 1464 spin_lock(&list->lock); 1465 if (size >= sk->sk_forward_alloc) { 1466 amt = sk_mem_pages(size); 1467 delta = amt << SK_MEM_QUANTUM_SHIFT; 1468 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1469 err = -ENOBUFS; 1470 spin_unlock(&list->lock); 1471 goto uncharge_drop; 1472 } 1473 1474 sk->sk_forward_alloc += delta; 1475 } 1476 1477 sk->sk_forward_alloc -= size; 1478 1479 /* no need to setup a destructor, we will explicitly release the 1480 * forward allocated memory on dequeue 1481 */ 1482 sock_skb_set_dropcount(sk, skb); 1483 1484 __skb_queue_tail(list, skb); 1485 spin_unlock(&list->lock); 1486 1487 if (!sock_flag(sk, SOCK_DEAD)) 1488 sk->sk_data_ready(sk); 1489 1490 busylock_release(busy); 1491 return 0; 1492 1493 uncharge_drop: 1494 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1495 1496 drop: 1497 atomic_inc(&sk->sk_drops); 1498 busylock_release(busy); 1499 return err; 1500 } 1501 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1502 1503 void udp_destruct_sock(struct sock *sk) 1504 { 1505 /* reclaim completely the forward allocated memory */ 1506 struct udp_sock *up = udp_sk(sk); 1507 unsigned int total = 0; 1508 struct sk_buff *skb; 1509 1510 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1511 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1512 total += skb->truesize; 1513 kfree_skb(skb); 1514 } 1515 udp_rmem_release(sk, total, 0, true); 1516 1517 inet_sock_destruct(sk); 1518 } 1519 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1520 1521 int udp_init_sock(struct sock *sk) 1522 { 1523 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1524 sk->sk_destruct = udp_destruct_sock; 1525 return 0; 1526 } 1527 EXPORT_SYMBOL_GPL(udp_init_sock); 1528 1529 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1530 { 1531 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1532 bool slow = lock_sock_fast(sk); 1533 1534 sk_peek_offset_bwd(sk, len); 1535 unlock_sock_fast(sk, slow); 1536 } 1537 1538 if (!skb_unref(skb)) 1539 return; 1540 1541 /* In the more common cases we cleared the head states previously, 1542 * see __udp_queue_rcv_skb(). 1543 */ 1544 if (unlikely(udp_skb_has_head_state(skb))) 1545 skb_release_head_state(skb); 1546 __consume_stateless_skb(skb); 1547 } 1548 EXPORT_SYMBOL_GPL(skb_consume_udp); 1549 1550 static struct sk_buff *__first_packet_length(struct sock *sk, 1551 struct sk_buff_head *rcvq, 1552 int *total) 1553 { 1554 struct sk_buff *skb; 1555 1556 while ((skb = skb_peek(rcvq)) != NULL) { 1557 if (udp_lib_checksum_complete(skb)) { 1558 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1559 IS_UDPLITE(sk)); 1560 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1561 IS_UDPLITE(sk)); 1562 atomic_inc(&sk->sk_drops); 1563 __skb_unlink(skb, rcvq); 1564 *total += skb->truesize; 1565 kfree_skb(skb); 1566 } else { 1567 /* the csum related bits could be changed, refresh 1568 * the scratch area 1569 */ 1570 udp_set_dev_scratch(skb); 1571 break; 1572 } 1573 } 1574 return skb; 1575 } 1576 1577 /** 1578 * first_packet_length - return length of first packet in receive queue 1579 * @sk: socket 1580 * 1581 * Drops all bad checksum frames, until a valid one is found. 1582 * Returns the length of found skb, or -1 if none is found. 1583 */ 1584 static int first_packet_length(struct sock *sk) 1585 { 1586 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1587 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1588 struct sk_buff *skb; 1589 int total = 0; 1590 int res; 1591 1592 spin_lock_bh(&rcvq->lock); 1593 skb = __first_packet_length(sk, rcvq, &total); 1594 if (!skb && !skb_queue_empty(sk_queue)) { 1595 spin_lock(&sk_queue->lock); 1596 skb_queue_splice_tail_init(sk_queue, rcvq); 1597 spin_unlock(&sk_queue->lock); 1598 1599 skb = __first_packet_length(sk, rcvq, &total); 1600 } 1601 res = skb ? skb->len : -1; 1602 if (total) 1603 udp_rmem_release(sk, total, 1, false); 1604 spin_unlock_bh(&rcvq->lock); 1605 return res; 1606 } 1607 1608 /* 1609 * IOCTL requests applicable to the UDP protocol 1610 */ 1611 1612 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1613 { 1614 switch (cmd) { 1615 case SIOCOUTQ: 1616 { 1617 int amount = sk_wmem_alloc_get(sk); 1618 1619 return put_user(amount, (int __user *)arg); 1620 } 1621 1622 case SIOCINQ: 1623 { 1624 int amount = max_t(int, 0, first_packet_length(sk)); 1625 1626 return put_user(amount, (int __user *)arg); 1627 } 1628 1629 default: 1630 return -ENOIOCTLCMD; 1631 } 1632 1633 return 0; 1634 } 1635 EXPORT_SYMBOL(udp_ioctl); 1636 1637 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1638 int noblock, int *off, int *err) 1639 { 1640 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1641 struct sk_buff_head *queue; 1642 struct sk_buff *last; 1643 long timeo; 1644 int error; 1645 1646 queue = &udp_sk(sk)->reader_queue; 1647 flags |= noblock ? MSG_DONTWAIT : 0; 1648 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1649 do { 1650 struct sk_buff *skb; 1651 1652 error = sock_error(sk); 1653 if (error) 1654 break; 1655 1656 error = -EAGAIN; 1657 do { 1658 spin_lock_bh(&queue->lock); 1659 skb = __skb_try_recv_from_queue(sk, queue, flags, 1660 udp_skb_destructor, 1661 off, err, &last); 1662 if (skb) { 1663 spin_unlock_bh(&queue->lock); 1664 return skb; 1665 } 1666 1667 if (skb_queue_empty(sk_queue)) { 1668 spin_unlock_bh(&queue->lock); 1669 goto busy_check; 1670 } 1671 1672 /* refill the reader queue and walk it again 1673 * keep both queues locked to avoid re-acquiring 1674 * the sk_receive_queue lock if fwd memory scheduling 1675 * is needed. 1676 */ 1677 spin_lock(&sk_queue->lock); 1678 skb_queue_splice_tail_init(sk_queue, queue); 1679 1680 skb = __skb_try_recv_from_queue(sk, queue, flags, 1681 udp_skb_dtor_locked, 1682 off, err, &last); 1683 spin_unlock(&sk_queue->lock); 1684 spin_unlock_bh(&queue->lock); 1685 if (skb) 1686 return skb; 1687 1688 busy_check: 1689 if (!sk_can_busy_loop(sk)) 1690 break; 1691 1692 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1693 } while (!skb_queue_empty(sk_queue)); 1694 1695 /* sk_queue is empty, reader_queue may contain peeked packets */ 1696 } while (timeo && 1697 !__skb_wait_for_more_packets(sk, &error, &timeo, 1698 (struct sk_buff *)sk_queue)); 1699 1700 *err = error; 1701 return NULL; 1702 } 1703 EXPORT_SYMBOL(__skb_recv_udp); 1704 1705 /* 1706 * This should be easy, if there is something there we 1707 * return it, otherwise we block. 1708 */ 1709 1710 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1711 int flags, int *addr_len) 1712 { 1713 struct inet_sock *inet = inet_sk(sk); 1714 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1715 struct sk_buff *skb; 1716 unsigned int ulen, copied; 1717 int off, err, peeking = flags & MSG_PEEK; 1718 int is_udplite = IS_UDPLITE(sk); 1719 bool checksum_valid = false; 1720 1721 if (flags & MSG_ERRQUEUE) 1722 return ip_recv_error(sk, msg, len, addr_len); 1723 1724 try_again: 1725 off = sk_peek_offset(sk, flags); 1726 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 1727 if (!skb) 1728 return err; 1729 1730 ulen = udp_skb_len(skb); 1731 copied = len; 1732 if (copied > ulen - off) 1733 copied = ulen - off; 1734 else if (copied < ulen) 1735 msg->msg_flags |= MSG_TRUNC; 1736 1737 /* 1738 * If checksum is needed at all, try to do it while copying the 1739 * data. If the data is truncated, or if we only want a partial 1740 * coverage checksum (UDP-Lite), do it before the copy. 1741 */ 1742 1743 if (copied < ulen || peeking || 1744 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1745 checksum_valid = udp_skb_csum_unnecessary(skb) || 1746 !__udp_lib_checksum_complete(skb); 1747 if (!checksum_valid) 1748 goto csum_copy_err; 1749 } 1750 1751 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1752 if (udp_skb_is_linear(skb)) 1753 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1754 else 1755 err = skb_copy_datagram_msg(skb, off, msg, copied); 1756 } else { 1757 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1758 1759 if (err == -EINVAL) 1760 goto csum_copy_err; 1761 } 1762 1763 if (unlikely(err)) { 1764 if (!peeking) { 1765 atomic_inc(&sk->sk_drops); 1766 UDP_INC_STATS(sock_net(sk), 1767 UDP_MIB_INERRORS, is_udplite); 1768 } 1769 kfree_skb(skb); 1770 return err; 1771 } 1772 1773 if (!peeking) 1774 UDP_INC_STATS(sock_net(sk), 1775 UDP_MIB_INDATAGRAMS, is_udplite); 1776 1777 sock_recv_ts_and_drops(msg, sk, skb); 1778 1779 /* Copy the address. */ 1780 if (sin) { 1781 sin->sin_family = AF_INET; 1782 sin->sin_port = udp_hdr(skb)->source; 1783 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1784 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1785 *addr_len = sizeof(*sin); 1786 1787 if (cgroup_bpf_enabled) 1788 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1789 (struct sockaddr *)sin); 1790 } 1791 1792 if (udp_sk(sk)->gro_enabled) 1793 udp_cmsg_recv(msg, sk, skb); 1794 1795 if (inet->cmsg_flags) 1796 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1797 1798 err = copied; 1799 if (flags & MSG_TRUNC) 1800 err = ulen; 1801 1802 skb_consume_udp(sk, skb, peeking ? -err : err); 1803 return err; 1804 1805 csum_copy_err: 1806 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1807 udp_skb_destructor)) { 1808 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1809 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1810 } 1811 kfree_skb(skb); 1812 1813 /* starting over for a new packet, but check if we need to yield */ 1814 cond_resched(); 1815 msg->msg_flags &= ~MSG_TRUNC; 1816 goto try_again; 1817 } 1818 1819 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1820 { 1821 /* This check is replicated from __ip4_datagram_connect() and 1822 * intended to prevent BPF program called below from accessing bytes 1823 * that are out of the bound specified by user in addr_len. 1824 */ 1825 if (addr_len < sizeof(struct sockaddr_in)) 1826 return -EINVAL; 1827 1828 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1829 } 1830 EXPORT_SYMBOL(udp_pre_connect); 1831 1832 int __udp_disconnect(struct sock *sk, int flags) 1833 { 1834 struct inet_sock *inet = inet_sk(sk); 1835 /* 1836 * 1003.1g - break association. 1837 */ 1838 1839 sk->sk_state = TCP_CLOSE; 1840 inet->inet_daddr = 0; 1841 inet->inet_dport = 0; 1842 sock_rps_reset_rxhash(sk); 1843 sk->sk_bound_dev_if = 0; 1844 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1845 inet_reset_saddr(sk); 1846 1847 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1848 sk->sk_prot->unhash(sk); 1849 inet->inet_sport = 0; 1850 } 1851 sk_dst_reset(sk); 1852 return 0; 1853 } 1854 EXPORT_SYMBOL(__udp_disconnect); 1855 1856 int udp_disconnect(struct sock *sk, int flags) 1857 { 1858 lock_sock(sk); 1859 __udp_disconnect(sk, flags); 1860 release_sock(sk); 1861 return 0; 1862 } 1863 EXPORT_SYMBOL(udp_disconnect); 1864 1865 void udp_lib_unhash(struct sock *sk) 1866 { 1867 if (sk_hashed(sk)) { 1868 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1869 struct udp_hslot *hslot, *hslot2; 1870 1871 hslot = udp_hashslot(udptable, sock_net(sk), 1872 udp_sk(sk)->udp_port_hash); 1873 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1874 1875 spin_lock_bh(&hslot->lock); 1876 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1877 reuseport_detach_sock(sk); 1878 if (sk_del_node_init_rcu(sk)) { 1879 hslot->count--; 1880 inet_sk(sk)->inet_num = 0; 1881 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1882 1883 spin_lock(&hslot2->lock); 1884 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1885 hslot2->count--; 1886 spin_unlock(&hslot2->lock); 1887 } 1888 spin_unlock_bh(&hslot->lock); 1889 } 1890 } 1891 EXPORT_SYMBOL(udp_lib_unhash); 1892 1893 /* 1894 * inet_rcv_saddr was changed, we must rehash secondary hash 1895 */ 1896 void udp_lib_rehash(struct sock *sk, u16 newhash) 1897 { 1898 if (sk_hashed(sk)) { 1899 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1900 struct udp_hslot *hslot, *hslot2, *nhslot2; 1901 1902 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1903 nhslot2 = udp_hashslot2(udptable, newhash); 1904 udp_sk(sk)->udp_portaddr_hash = newhash; 1905 1906 if (hslot2 != nhslot2 || 1907 rcu_access_pointer(sk->sk_reuseport_cb)) { 1908 hslot = udp_hashslot(udptable, sock_net(sk), 1909 udp_sk(sk)->udp_port_hash); 1910 /* we must lock primary chain too */ 1911 spin_lock_bh(&hslot->lock); 1912 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1913 reuseport_detach_sock(sk); 1914 1915 if (hslot2 != nhslot2) { 1916 spin_lock(&hslot2->lock); 1917 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1918 hslot2->count--; 1919 spin_unlock(&hslot2->lock); 1920 1921 spin_lock(&nhslot2->lock); 1922 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1923 &nhslot2->head); 1924 nhslot2->count++; 1925 spin_unlock(&nhslot2->lock); 1926 } 1927 1928 spin_unlock_bh(&hslot->lock); 1929 } 1930 } 1931 } 1932 EXPORT_SYMBOL(udp_lib_rehash); 1933 1934 void udp_v4_rehash(struct sock *sk) 1935 { 1936 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 1937 inet_sk(sk)->inet_rcv_saddr, 1938 inet_sk(sk)->inet_num); 1939 udp_lib_rehash(sk, new_hash); 1940 } 1941 1942 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1943 { 1944 int rc; 1945 1946 if (inet_sk(sk)->inet_daddr) { 1947 sock_rps_save_rxhash(sk, skb); 1948 sk_mark_napi_id(sk, skb); 1949 sk_incoming_cpu_update(sk); 1950 } else { 1951 sk_mark_napi_id_once(sk, skb); 1952 } 1953 1954 rc = __udp_enqueue_schedule_skb(sk, skb); 1955 if (rc < 0) { 1956 int is_udplite = IS_UDPLITE(sk); 1957 1958 /* Note that an ENOMEM error is charged twice */ 1959 if (rc == -ENOMEM) 1960 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1961 is_udplite); 1962 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1963 kfree_skb(skb); 1964 trace_udp_fail_queue_rcv_skb(rc, sk); 1965 return -1; 1966 } 1967 1968 return 0; 1969 } 1970 1971 /* returns: 1972 * -1: error 1973 * 0: success 1974 * >0: "udp encap" protocol resubmission 1975 * 1976 * Note that in the success and error cases, the skb is assumed to 1977 * have either been requeued or freed. 1978 */ 1979 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 1980 { 1981 struct udp_sock *up = udp_sk(sk); 1982 int is_udplite = IS_UDPLITE(sk); 1983 1984 /* 1985 * Charge it to the socket, dropping if the queue is full. 1986 */ 1987 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1988 goto drop; 1989 nf_reset(skb); 1990 1991 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 1992 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1993 1994 /* 1995 * This is an encapsulation socket so pass the skb to 1996 * the socket's udp_encap_rcv() hook. Otherwise, just 1997 * fall through and pass this up the UDP socket. 1998 * up->encap_rcv() returns the following value: 1999 * =0 if skb was successfully passed to the encap 2000 * handler or was discarded by it. 2001 * >0 if skb should be passed on to UDP. 2002 * <0 if skb should be resubmitted as proto -N 2003 */ 2004 2005 /* if we're overly short, let UDP handle it */ 2006 encap_rcv = READ_ONCE(up->encap_rcv); 2007 if (encap_rcv) { 2008 int ret; 2009 2010 /* Verify checksum before giving to encap */ 2011 if (udp_lib_checksum_complete(skb)) 2012 goto csum_error; 2013 2014 ret = encap_rcv(sk, skb); 2015 if (ret <= 0) { 2016 __UDP_INC_STATS(sock_net(sk), 2017 UDP_MIB_INDATAGRAMS, 2018 is_udplite); 2019 return -ret; 2020 } 2021 } 2022 2023 /* FALLTHROUGH -- it's a UDP Packet */ 2024 } 2025 2026 /* 2027 * UDP-Lite specific tests, ignored on UDP sockets 2028 */ 2029 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2030 2031 /* 2032 * MIB statistics other than incrementing the error count are 2033 * disabled for the following two types of errors: these depend 2034 * on the application settings, not on the functioning of the 2035 * protocol stack as such. 2036 * 2037 * RFC 3828 here recommends (sec 3.3): "There should also be a 2038 * way ... to ... at least let the receiving application block 2039 * delivery of packets with coverage values less than a value 2040 * provided by the application." 2041 */ 2042 if (up->pcrlen == 0) { /* full coverage was set */ 2043 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2044 UDP_SKB_CB(skb)->cscov, skb->len); 2045 goto drop; 2046 } 2047 /* The next case involves violating the min. coverage requested 2048 * by the receiver. This is subtle: if receiver wants x and x is 2049 * greater than the buffersize/MTU then receiver will complain 2050 * that it wants x while sender emits packets of smaller size y. 2051 * Therefore the above ...()->partial_cov statement is essential. 2052 */ 2053 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2054 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2055 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2056 goto drop; 2057 } 2058 } 2059 2060 prefetch(&sk->sk_rmem_alloc); 2061 if (rcu_access_pointer(sk->sk_filter) && 2062 udp_lib_checksum_complete(skb)) 2063 goto csum_error; 2064 2065 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 2066 goto drop; 2067 2068 udp_csum_pull_header(skb); 2069 2070 ipv4_pktinfo_prepare(sk, skb); 2071 return __udp_queue_rcv_skb(sk, skb); 2072 2073 csum_error: 2074 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2075 drop: 2076 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2077 atomic_inc(&sk->sk_drops); 2078 kfree_skb(skb); 2079 return -1; 2080 } 2081 2082 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2083 { 2084 struct sk_buff *next, *segs; 2085 int ret; 2086 2087 if (likely(!udp_unexpected_gso(sk, skb))) 2088 return udp_queue_rcv_one_skb(sk, skb); 2089 2090 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET); 2091 __skb_push(skb, -skb_mac_offset(skb)); 2092 segs = udp_rcv_segment(sk, skb, true); 2093 for (skb = segs; skb; skb = next) { 2094 next = skb->next; 2095 __skb_pull(skb, skb_transport_offset(skb)); 2096 ret = udp_queue_rcv_one_skb(sk, skb); 2097 if (ret > 0) 2098 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret); 2099 } 2100 return 0; 2101 } 2102 2103 /* For TCP sockets, sk_rx_dst is protected by socket lock 2104 * For UDP, we use xchg() to guard against concurrent changes. 2105 */ 2106 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2107 { 2108 struct dst_entry *old; 2109 2110 if (dst_hold_safe(dst)) { 2111 old = xchg(&sk->sk_rx_dst, dst); 2112 dst_release(old); 2113 return old != dst; 2114 } 2115 return false; 2116 } 2117 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2118 2119 /* 2120 * Multicasts and broadcasts go to each listener. 2121 * 2122 * Note: called only from the BH handler context. 2123 */ 2124 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2125 struct udphdr *uh, 2126 __be32 saddr, __be32 daddr, 2127 struct udp_table *udptable, 2128 int proto) 2129 { 2130 struct sock *sk, *first = NULL; 2131 unsigned short hnum = ntohs(uh->dest); 2132 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2133 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2134 unsigned int offset = offsetof(typeof(*sk), sk_node); 2135 int dif = skb->dev->ifindex; 2136 int sdif = inet_sdif(skb); 2137 struct hlist_node *node; 2138 struct sk_buff *nskb; 2139 2140 if (use_hash2) { 2141 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2142 udptable->mask; 2143 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2144 start_lookup: 2145 hslot = &udptable->hash2[hash2]; 2146 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2147 } 2148 2149 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2150 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2151 uh->source, saddr, dif, sdif, hnum)) 2152 continue; 2153 2154 if (!first) { 2155 first = sk; 2156 continue; 2157 } 2158 nskb = skb_clone(skb, GFP_ATOMIC); 2159 2160 if (unlikely(!nskb)) { 2161 atomic_inc(&sk->sk_drops); 2162 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2163 IS_UDPLITE(sk)); 2164 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2165 IS_UDPLITE(sk)); 2166 continue; 2167 } 2168 if (udp_queue_rcv_skb(sk, nskb) > 0) 2169 consume_skb(nskb); 2170 } 2171 2172 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2173 if (use_hash2 && hash2 != hash2_any) { 2174 hash2 = hash2_any; 2175 goto start_lookup; 2176 } 2177 2178 if (first) { 2179 if (udp_queue_rcv_skb(first, skb) > 0) 2180 consume_skb(skb); 2181 } else { 2182 kfree_skb(skb); 2183 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2184 proto == IPPROTO_UDPLITE); 2185 } 2186 return 0; 2187 } 2188 2189 /* Initialize UDP checksum. If exited with zero value (success), 2190 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2191 * Otherwise, csum completion requires chacksumming packet body, 2192 * including udp header and folding it to skb->csum. 2193 */ 2194 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2195 int proto) 2196 { 2197 int err; 2198 2199 UDP_SKB_CB(skb)->partial_cov = 0; 2200 UDP_SKB_CB(skb)->cscov = skb->len; 2201 2202 if (proto == IPPROTO_UDPLITE) { 2203 err = udplite_checksum_init(skb, uh); 2204 if (err) 2205 return err; 2206 2207 if (UDP_SKB_CB(skb)->partial_cov) { 2208 skb->csum = inet_compute_pseudo(skb, proto); 2209 return 0; 2210 } 2211 } 2212 2213 /* Note, we are only interested in != 0 or == 0, thus the 2214 * force to int. 2215 */ 2216 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2217 inet_compute_pseudo); 2218 if (err) 2219 return err; 2220 2221 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2222 /* If SW calculated the value, we know it's bad */ 2223 if (skb->csum_complete_sw) 2224 return 1; 2225 2226 /* HW says the value is bad. Let's validate that. 2227 * skb->csum is no longer the full packet checksum, 2228 * so don't treat it as such. 2229 */ 2230 skb_checksum_complete_unset(skb); 2231 } 2232 2233 return 0; 2234 } 2235 2236 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2237 * return code conversion for ip layer consumption 2238 */ 2239 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2240 struct udphdr *uh) 2241 { 2242 int ret; 2243 2244 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2245 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 2246 inet_compute_pseudo); 2247 2248 ret = udp_queue_rcv_skb(sk, skb); 2249 2250 /* a return value > 0 means to resubmit the input, but 2251 * it wants the return to be -protocol, or 0 2252 */ 2253 if (ret > 0) 2254 return -ret; 2255 return 0; 2256 } 2257 2258 /* 2259 * All we need to do is get the socket, and then do a checksum. 2260 */ 2261 2262 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2263 int proto) 2264 { 2265 struct sock *sk; 2266 struct udphdr *uh; 2267 unsigned short ulen; 2268 struct rtable *rt = skb_rtable(skb); 2269 __be32 saddr, daddr; 2270 struct net *net = dev_net(skb->dev); 2271 2272 /* 2273 * Validate the packet. 2274 */ 2275 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2276 goto drop; /* No space for header. */ 2277 2278 uh = udp_hdr(skb); 2279 ulen = ntohs(uh->len); 2280 saddr = ip_hdr(skb)->saddr; 2281 daddr = ip_hdr(skb)->daddr; 2282 2283 if (ulen > skb->len) 2284 goto short_packet; 2285 2286 if (proto == IPPROTO_UDP) { 2287 /* UDP validates ulen. */ 2288 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2289 goto short_packet; 2290 uh = udp_hdr(skb); 2291 } 2292 2293 if (udp4_csum_init(skb, uh, proto)) 2294 goto csum_error; 2295 2296 sk = skb_steal_sock(skb); 2297 if (sk) { 2298 struct dst_entry *dst = skb_dst(skb); 2299 int ret; 2300 2301 if (unlikely(sk->sk_rx_dst != dst)) 2302 udp_sk_rx_dst_set(sk, dst); 2303 2304 ret = udp_unicast_rcv_skb(sk, skb, uh); 2305 sock_put(sk); 2306 return ret; 2307 } 2308 2309 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2310 return __udp4_lib_mcast_deliver(net, skb, uh, 2311 saddr, daddr, udptable, proto); 2312 2313 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2314 if (sk) 2315 return udp_unicast_rcv_skb(sk, skb, uh); 2316 2317 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2318 goto drop; 2319 nf_reset(skb); 2320 2321 /* No socket. Drop packet silently, if checksum is wrong */ 2322 if (udp_lib_checksum_complete(skb)) 2323 goto csum_error; 2324 2325 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2326 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2327 2328 /* 2329 * Hmm. We got an UDP packet to a port to which we 2330 * don't wanna listen. Ignore it. 2331 */ 2332 kfree_skb(skb); 2333 return 0; 2334 2335 short_packet: 2336 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2337 proto == IPPROTO_UDPLITE ? "Lite" : "", 2338 &saddr, ntohs(uh->source), 2339 ulen, skb->len, 2340 &daddr, ntohs(uh->dest)); 2341 goto drop; 2342 2343 csum_error: 2344 /* 2345 * RFC1122: OK. Discards the bad packet silently (as far as 2346 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2347 */ 2348 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2349 proto == IPPROTO_UDPLITE ? "Lite" : "", 2350 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2351 ulen); 2352 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2353 drop: 2354 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2355 kfree_skb(skb); 2356 return 0; 2357 } 2358 2359 /* We can only early demux multicast if there is a single matching socket. 2360 * If more than one socket found returns NULL 2361 */ 2362 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2363 __be16 loc_port, __be32 loc_addr, 2364 __be16 rmt_port, __be32 rmt_addr, 2365 int dif, int sdif) 2366 { 2367 struct sock *sk, *result; 2368 unsigned short hnum = ntohs(loc_port); 2369 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2370 struct udp_hslot *hslot = &udp_table.hash[slot]; 2371 2372 /* Do not bother scanning a too big list */ 2373 if (hslot->count > 10) 2374 return NULL; 2375 2376 result = NULL; 2377 sk_for_each_rcu(sk, &hslot->head) { 2378 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2379 rmt_port, rmt_addr, dif, sdif, hnum)) { 2380 if (result) 2381 return NULL; 2382 result = sk; 2383 } 2384 } 2385 2386 return result; 2387 } 2388 2389 /* For unicast we should only early demux connected sockets or we can 2390 * break forwarding setups. The chains here can be long so only check 2391 * if the first socket is an exact match and if not move on. 2392 */ 2393 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2394 __be16 loc_port, __be32 loc_addr, 2395 __be16 rmt_port, __be32 rmt_addr, 2396 int dif, int sdif) 2397 { 2398 unsigned short hnum = ntohs(loc_port); 2399 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2400 unsigned int slot2 = hash2 & udp_table.mask; 2401 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2402 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2403 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2404 struct sock *sk; 2405 2406 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2407 if (INET_MATCH(sk, net, acookie, rmt_addr, 2408 loc_addr, ports, dif, sdif)) 2409 return sk; 2410 /* Only check first socket in chain */ 2411 break; 2412 } 2413 return NULL; 2414 } 2415 2416 int udp_v4_early_demux(struct sk_buff *skb) 2417 { 2418 struct net *net = dev_net(skb->dev); 2419 struct in_device *in_dev = NULL; 2420 const struct iphdr *iph; 2421 const struct udphdr *uh; 2422 struct sock *sk = NULL; 2423 struct dst_entry *dst; 2424 int dif = skb->dev->ifindex; 2425 int sdif = inet_sdif(skb); 2426 int ours; 2427 2428 /* validate the packet */ 2429 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2430 return 0; 2431 2432 iph = ip_hdr(skb); 2433 uh = udp_hdr(skb); 2434 2435 if (skb->pkt_type == PACKET_MULTICAST) { 2436 in_dev = __in_dev_get_rcu(skb->dev); 2437 2438 if (!in_dev) 2439 return 0; 2440 2441 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2442 iph->protocol); 2443 if (!ours) 2444 return 0; 2445 2446 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2447 uh->source, iph->saddr, 2448 dif, sdif); 2449 } else if (skb->pkt_type == PACKET_HOST) { 2450 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2451 uh->source, iph->saddr, dif, sdif); 2452 } 2453 2454 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2455 return 0; 2456 2457 skb->sk = sk; 2458 skb->destructor = sock_efree; 2459 dst = READ_ONCE(sk->sk_rx_dst); 2460 2461 if (dst) 2462 dst = dst_check(dst, 0); 2463 if (dst) { 2464 u32 itag = 0; 2465 2466 /* set noref for now. 2467 * any place which wants to hold dst has to call 2468 * dst_hold_safe() 2469 */ 2470 skb_dst_set_noref(skb, dst); 2471 2472 /* for unconnected multicast sockets we need to validate 2473 * the source on each packet 2474 */ 2475 if (!inet_sk(sk)->inet_daddr && in_dev) 2476 return ip_mc_validate_source(skb, iph->daddr, 2477 iph->saddr, iph->tos, 2478 skb->dev, in_dev, &itag); 2479 } 2480 return 0; 2481 } 2482 2483 int udp_rcv(struct sk_buff *skb) 2484 { 2485 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2486 } 2487 2488 void udp_destroy_sock(struct sock *sk) 2489 { 2490 struct udp_sock *up = udp_sk(sk); 2491 bool slow = lock_sock_fast(sk); 2492 udp_flush_pending_frames(sk); 2493 unlock_sock_fast(sk, slow); 2494 if (static_branch_unlikely(&udp_encap_needed_key)) { 2495 if (up->encap_type) { 2496 void (*encap_destroy)(struct sock *sk); 2497 encap_destroy = READ_ONCE(up->encap_destroy); 2498 if (encap_destroy) 2499 encap_destroy(sk); 2500 } 2501 if (up->encap_enabled) 2502 static_branch_dec(&udp_encap_needed_key); 2503 } 2504 } 2505 2506 /* 2507 * Socket option code for UDP 2508 */ 2509 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2510 char __user *optval, unsigned int optlen, 2511 int (*push_pending_frames)(struct sock *)) 2512 { 2513 struct udp_sock *up = udp_sk(sk); 2514 int val, valbool; 2515 int err = 0; 2516 int is_udplite = IS_UDPLITE(sk); 2517 2518 if (optlen < sizeof(int)) 2519 return -EINVAL; 2520 2521 if (get_user(val, (int __user *)optval)) 2522 return -EFAULT; 2523 2524 valbool = val ? 1 : 0; 2525 2526 switch (optname) { 2527 case UDP_CORK: 2528 if (val != 0) { 2529 up->corkflag = 1; 2530 } else { 2531 up->corkflag = 0; 2532 lock_sock(sk); 2533 push_pending_frames(sk); 2534 release_sock(sk); 2535 } 2536 break; 2537 2538 case UDP_ENCAP: 2539 switch (val) { 2540 case 0: 2541 case UDP_ENCAP_ESPINUDP: 2542 case UDP_ENCAP_ESPINUDP_NON_IKE: 2543 up->encap_rcv = xfrm4_udp_encap_rcv; 2544 /* FALLTHROUGH */ 2545 case UDP_ENCAP_L2TPINUDP: 2546 up->encap_type = val; 2547 lock_sock(sk); 2548 udp_tunnel_encap_enable(sk->sk_socket); 2549 release_sock(sk); 2550 break; 2551 default: 2552 err = -ENOPROTOOPT; 2553 break; 2554 } 2555 break; 2556 2557 case UDP_NO_CHECK6_TX: 2558 up->no_check6_tx = valbool; 2559 break; 2560 2561 case UDP_NO_CHECK6_RX: 2562 up->no_check6_rx = valbool; 2563 break; 2564 2565 case UDP_SEGMENT: 2566 if (val < 0 || val > USHRT_MAX) 2567 return -EINVAL; 2568 up->gso_size = val; 2569 break; 2570 2571 case UDP_GRO: 2572 lock_sock(sk); 2573 if (valbool) 2574 udp_tunnel_encap_enable(sk->sk_socket); 2575 up->gro_enabled = valbool; 2576 release_sock(sk); 2577 break; 2578 2579 /* 2580 * UDP-Lite's partial checksum coverage (RFC 3828). 2581 */ 2582 /* The sender sets actual checksum coverage length via this option. 2583 * The case coverage > packet length is handled by send module. */ 2584 case UDPLITE_SEND_CSCOV: 2585 if (!is_udplite) /* Disable the option on UDP sockets */ 2586 return -ENOPROTOOPT; 2587 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2588 val = 8; 2589 else if (val > USHRT_MAX) 2590 val = USHRT_MAX; 2591 up->pcslen = val; 2592 up->pcflag |= UDPLITE_SEND_CC; 2593 break; 2594 2595 /* The receiver specifies a minimum checksum coverage value. To make 2596 * sense, this should be set to at least 8 (as done below). If zero is 2597 * used, this again means full checksum coverage. */ 2598 case UDPLITE_RECV_CSCOV: 2599 if (!is_udplite) /* Disable the option on UDP sockets */ 2600 return -ENOPROTOOPT; 2601 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2602 val = 8; 2603 else if (val > USHRT_MAX) 2604 val = USHRT_MAX; 2605 up->pcrlen = val; 2606 up->pcflag |= UDPLITE_RECV_CC; 2607 break; 2608 2609 default: 2610 err = -ENOPROTOOPT; 2611 break; 2612 } 2613 2614 return err; 2615 } 2616 EXPORT_SYMBOL(udp_lib_setsockopt); 2617 2618 int udp_setsockopt(struct sock *sk, int level, int optname, 2619 char __user *optval, unsigned int optlen) 2620 { 2621 if (level == SOL_UDP || level == SOL_UDPLITE) 2622 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2623 udp_push_pending_frames); 2624 return ip_setsockopt(sk, level, optname, optval, optlen); 2625 } 2626 2627 #ifdef CONFIG_COMPAT 2628 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2629 char __user *optval, unsigned int optlen) 2630 { 2631 if (level == SOL_UDP || level == SOL_UDPLITE) 2632 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2633 udp_push_pending_frames); 2634 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2635 } 2636 #endif 2637 2638 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2639 char __user *optval, int __user *optlen) 2640 { 2641 struct udp_sock *up = udp_sk(sk); 2642 int val, len; 2643 2644 if (get_user(len, optlen)) 2645 return -EFAULT; 2646 2647 len = min_t(unsigned int, len, sizeof(int)); 2648 2649 if (len < 0) 2650 return -EINVAL; 2651 2652 switch (optname) { 2653 case UDP_CORK: 2654 val = up->corkflag; 2655 break; 2656 2657 case UDP_ENCAP: 2658 val = up->encap_type; 2659 break; 2660 2661 case UDP_NO_CHECK6_TX: 2662 val = up->no_check6_tx; 2663 break; 2664 2665 case UDP_NO_CHECK6_RX: 2666 val = up->no_check6_rx; 2667 break; 2668 2669 case UDP_SEGMENT: 2670 val = up->gso_size; 2671 break; 2672 2673 /* The following two cannot be changed on UDP sockets, the return is 2674 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2675 case UDPLITE_SEND_CSCOV: 2676 val = up->pcslen; 2677 break; 2678 2679 case UDPLITE_RECV_CSCOV: 2680 val = up->pcrlen; 2681 break; 2682 2683 default: 2684 return -ENOPROTOOPT; 2685 } 2686 2687 if (put_user(len, optlen)) 2688 return -EFAULT; 2689 if (copy_to_user(optval, &val, len)) 2690 return -EFAULT; 2691 return 0; 2692 } 2693 EXPORT_SYMBOL(udp_lib_getsockopt); 2694 2695 int udp_getsockopt(struct sock *sk, int level, int optname, 2696 char __user *optval, int __user *optlen) 2697 { 2698 if (level == SOL_UDP || level == SOL_UDPLITE) 2699 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2700 return ip_getsockopt(sk, level, optname, optval, optlen); 2701 } 2702 2703 #ifdef CONFIG_COMPAT 2704 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2705 char __user *optval, int __user *optlen) 2706 { 2707 if (level == SOL_UDP || level == SOL_UDPLITE) 2708 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2709 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2710 } 2711 #endif 2712 /** 2713 * udp_poll - wait for a UDP event. 2714 * @file - file struct 2715 * @sock - socket 2716 * @wait - poll table 2717 * 2718 * This is same as datagram poll, except for the special case of 2719 * blocking sockets. If application is using a blocking fd 2720 * and a packet with checksum error is in the queue; 2721 * then it could get return from select indicating data available 2722 * but then block when reading it. Add special case code 2723 * to work around these arguably broken applications. 2724 */ 2725 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2726 { 2727 __poll_t mask = datagram_poll(file, sock, wait); 2728 struct sock *sk = sock->sk; 2729 2730 if (!skb_queue_empty(&udp_sk(sk)->reader_queue)) 2731 mask |= EPOLLIN | EPOLLRDNORM; 2732 2733 /* Check for false positives due to checksum errors */ 2734 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2735 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2736 mask &= ~(EPOLLIN | EPOLLRDNORM); 2737 2738 return mask; 2739 2740 } 2741 EXPORT_SYMBOL(udp_poll); 2742 2743 int udp_abort(struct sock *sk, int err) 2744 { 2745 lock_sock(sk); 2746 2747 sk->sk_err = err; 2748 sk->sk_error_report(sk); 2749 __udp_disconnect(sk, 0); 2750 2751 release_sock(sk); 2752 2753 return 0; 2754 } 2755 EXPORT_SYMBOL_GPL(udp_abort); 2756 2757 struct proto udp_prot = { 2758 .name = "UDP", 2759 .owner = THIS_MODULE, 2760 .close = udp_lib_close, 2761 .pre_connect = udp_pre_connect, 2762 .connect = ip4_datagram_connect, 2763 .disconnect = udp_disconnect, 2764 .ioctl = udp_ioctl, 2765 .init = udp_init_sock, 2766 .destroy = udp_destroy_sock, 2767 .setsockopt = udp_setsockopt, 2768 .getsockopt = udp_getsockopt, 2769 .sendmsg = udp_sendmsg, 2770 .recvmsg = udp_recvmsg, 2771 .sendpage = udp_sendpage, 2772 .release_cb = ip4_datagram_release_cb, 2773 .hash = udp_lib_hash, 2774 .unhash = udp_lib_unhash, 2775 .rehash = udp_v4_rehash, 2776 .get_port = udp_v4_get_port, 2777 .memory_allocated = &udp_memory_allocated, 2778 .sysctl_mem = sysctl_udp_mem, 2779 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2780 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2781 .obj_size = sizeof(struct udp_sock), 2782 .h.udp_table = &udp_table, 2783 #ifdef CONFIG_COMPAT 2784 .compat_setsockopt = compat_udp_setsockopt, 2785 .compat_getsockopt = compat_udp_getsockopt, 2786 #endif 2787 .diag_destroy = udp_abort, 2788 }; 2789 EXPORT_SYMBOL(udp_prot); 2790 2791 /* ------------------------------------------------------------------------ */ 2792 #ifdef CONFIG_PROC_FS 2793 2794 static struct sock *udp_get_first(struct seq_file *seq, int start) 2795 { 2796 struct sock *sk; 2797 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2798 struct udp_iter_state *state = seq->private; 2799 struct net *net = seq_file_net(seq); 2800 2801 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; 2802 ++state->bucket) { 2803 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; 2804 2805 if (hlist_empty(&hslot->head)) 2806 continue; 2807 2808 spin_lock_bh(&hslot->lock); 2809 sk_for_each(sk, &hslot->head) { 2810 if (!net_eq(sock_net(sk), net)) 2811 continue; 2812 if (sk->sk_family == afinfo->family) 2813 goto found; 2814 } 2815 spin_unlock_bh(&hslot->lock); 2816 } 2817 sk = NULL; 2818 found: 2819 return sk; 2820 } 2821 2822 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2823 { 2824 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2825 struct udp_iter_state *state = seq->private; 2826 struct net *net = seq_file_net(seq); 2827 2828 do { 2829 sk = sk_next(sk); 2830 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family)); 2831 2832 if (!sk) { 2833 if (state->bucket <= afinfo->udp_table->mask) 2834 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 2835 return udp_get_first(seq, state->bucket + 1); 2836 } 2837 return sk; 2838 } 2839 2840 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2841 { 2842 struct sock *sk = udp_get_first(seq, 0); 2843 2844 if (sk) 2845 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2846 --pos; 2847 return pos ? NULL : sk; 2848 } 2849 2850 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2851 { 2852 struct udp_iter_state *state = seq->private; 2853 state->bucket = MAX_UDP_PORTS; 2854 2855 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2856 } 2857 EXPORT_SYMBOL(udp_seq_start); 2858 2859 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2860 { 2861 struct sock *sk; 2862 2863 if (v == SEQ_START_TOKEN) 2864 sk = udp_get_idx(seq, 0); 2865 else 2866 sk = udp_get_next(seq, v); 2867 2868 ++*pos; 2869 return sk; 2870 } 2871 EXPORT_SYMBOL(udp_seq_next); 2872 2873 void udp_seq_stop(struct seq_file *seq, void *v) 2874 { 2875 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2876 struct udp_iter_state *state = seq->private; 2877 2878 if (state->bucket <= afinfo->udp_table->mask) 2879 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 2880 } 2881 EXPORT_SYMBOL(udp_seq_stop); 2882 2883 /* ------------------------------------------------------------------------ */ 2884 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2885 int bucket) 2886 { 2887 struct inet_sock *inet = inet_sk(sp); 2888 __be32 dest = inet->inet_daddr; 2889 __be32 src = inet->inet_rcv_saddr; 2890 __u16 destp = ntohs(inet->inet_dport); 2891 __u16 srcp = ntohs(inet->inet_sport); 2892 2893 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2894 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 2895 bucket, src, srcp, dest, destp, sp->sk_state, 2896 sk_wmem_alloc_get(sp), 2897 udp_rqueue_get(sp), 2898 0, 0L, 0, 2899 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2900 0, sock_i_ino(sp), 2901 refcount_read(&sp->sk_refcnt), sp, 2902 atomic_read(&sp->sk_drops)); 2903 } 2904 2905 int udp4_seq_show(struct seq_file *seq, void *v) 2906 { 2907 seq_setwidth(seq, 127); 2908 if (v == SEQ_START_TOKEN) 2909 seq_puts(seq, " sl local_address rem_address st tx_queue " 2910 "rx_queue tr tm->when retrnsmt uid timeout " 2911 "inode ref pointer drops"); 2912 else { 2913 struct udp_iter_state *state = seq->private; 2914 2915 udp4_format_sock(v, seq, state->bucket); 2916 } 2917 seq_pad(seq, '\n'); 2918 return 0; 2919 } 2920 2921 const struct seq_operations udp_seq_ops = { 2922 .start = udp_seq_start, 2923 .next = udp_seq_next, 2924 .stop = udp_seq_stop, 2925 .show = udp4_seq_show, 2926 }; 2927 EXPORT_SYMBOL(udp_seq_ops); 2928 2929 static struct udp_seq_afinfo udp4_seq_afinfo = { 2930 .family = AF_INET, 2931 .udp_table = &udp_table, 2932 }; 2933 2934 static int __net_init udp4_proc_init_net(struct net *net) 2935 { 2936 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 2937 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 2938 return -ENOMEM; 2939 return 0; 2940 } 2941 2942 static void __net_exit udp4_proc_exit_net(struct net *net) 2943 { 2944 remove_proc_entry("udp", net->proc_net); 2945 } 2946 2947 static struct pernet_operations udp4_net_ops = { 2948 .init = udp4_proc_init_net, 2949 .exit = udp4_proc_exit_net, 2950 }; 2951 2952 int __init udp4_proc_init(void) 2953 { 2954 return register_pernet_subsys(&udp4_net_ops); 2955 } 2956 2957 void udp4_proc_exit(void) 2958 { 2959 unregister_pernet_subsys(&udp4_net_ops); 2960 } 2961 #endif /* CONFIG_PROC_FS */ 2962 2963 static __initdata unsigned long uhash_entries; 2964 static int __init set_uhash_entries(char *str) 2965 { 2966 ssize_t ret; 2967 2968 if (!str) 2969 return 0; 2970 2971 ret = kstrtoul(str, 0, &uhash_entries); 2972 if (ret) 2973 return 0; 2974 2975 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2976 uhash_entries = UDP_HTABLE_SIZE_MIN; 2977 return 1; 2978 } 2979 __setup("uhash_entries=", set_uhash_entries); 2980 2981 void __init udp_table_init(struct udp_table *table, const char *name) 2982 { 2983 unsigned int i; 2984 2985 table->hash = alloc_large_system_hash(name, 2986 2 * sizeof(struct udp_hslot), 2987 uhash_entries, 2988 21, /* one slot per 2 MB */ 2989 0, 2990 &table->log, 2991 &table->mask, 2992 UDP_HTABLE_SIZE_MIN, 2993 64 * 1024); 2994 2995 table->hash2 = table->hash + (table->mask + 1); 2996 for (i = 0; i <= table->mask; i++) { 2997 INIT_HLIST_HEAD(&table->hash[i].head); 2998 table->hash[i].count = 0; 2999 spin_lock_init(&table->hash[i].lock); 3000 } 3001 for (i = 0; i <= table->mask; i++) { 3002 INIT_HLIST_HEAD(&table->hash2[i].head); 3003 table->hash2[i].count = 0; 3004 spin_lock_init(&table->hash2[i].lock); 3005 } 3006 } 3007 3008 u32 udp_flow_hashrnd(void) 3009 { 3010 static u32 hashrnd __read_mostly; 3011 3012 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3013 3014 return hashrnd; 3015 } 3016 EXPORT_SYMBOL(udp_flow_hashrnd); 3017 3018 static void __udp_sysctl_init(struct net *net) 3019 { 3020 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; 3021 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; 3022 3023 #ifdef CONFIG_NET_L3_MASTER_DEV 3024 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3025 #endif 3026 } 3027 3028 static int __net_init udp_sysctl_init(struct net *net) 3029 { 3030 __udp_sysctl_init(net); 3031 return 0; 3032 } 3033 3034 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3035 .init = udp_sysctl_init, 3036 }; 3037 3038 void __init udp_init(void) 3039 { 3040 unsigned long limit; 3041 unsigned int i; 3042 3043 udp_table_init(&udp_table, "UDP"); 3044 limit = nr_free_buffer_pages() / 8; 3045 limit = max(limit, 128UL); 3046 sysctl_udp_mem[0] = limit / 4 * 3; 3047 sysctl_udp_mem[1] = limit; 3048 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3049 3050 __udp_sysctl_init(&init_net); 3051 3052 /* 16 spinlocks per cpu */ 3053 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3054 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3055 GFP_KERNEL); 3056 if (!udp_busylocks) 3057 panic("UDP: failed to alloc udp_busylocks\n"); 3058 for (i = 0; i < (1U << udp_busylocks_log); i++) 3059 spin_lock_init(udp_busylocks + i); 3060 3061 if (register_pernet_subsys(&udp_sysctl_ops)) 3062 panic("UDP: failed to init sysctl parameters.\n"); 3063 } 3064