1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * The User Datagram Protocol (UDP). 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 12 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 13 * Hirokazu Takahashi, <taka@valinux.co.jp> 14 * 15 * Fixes: 16 * Alan Cox : verify_area() calls 17 * Alan Cox : stopped close while in use off icmp 18 * messages. Not a fix but a botch that 19 * for udp at least is 'valid'. 20 * Alan Cox : Fixed icmp handling properly 21 * Alan Cox : Correct error for oversized datagrams 22 * Alan Cox : Tidied select() semantics. 23 * Alan Cox : udp_err() fixed properly, also now 24 * select and read wake correctly on errors 25 * Alan Cox : udp_send verify_area moved to avoid mem leak 26 * Alan Cox : UDP can count its memory 27 * Alan Cox : send to an unknown connection causes 28 * an ECONNREFUSED off the icmp, but 29 * does NOT close. 30 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 32 * bug no longer crashes it. 33 * Fred Van Kempen : Net2e support for sk->broadcast. 34 * Alan Cox : Uses skb_free_datagram 35 * Alan Cox : Added get/set sockopt support. 36 * Alan Cox : Broadcasting without option set returns EACCES. 37 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 38 * Alan Cox : Use ip_tos and ip_ttl 39 * Alan Cox : SNMP Mibs 40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 41 * Matt Dillon : UDP length checks. 42 * Alan Cox : Smarter af_inet used properly. 43 * Alan Cox : Use new kernel side addressing. 44 * Alan Cox : Incorrect return on truncated datagram receive. 45 * Arnt Gulbrandsen : New udp_send and stuff 46 * Alan Cox : Cache last socket 47 * Alan Cox : Route cache 48 * Jon Peatfield : Minor efficiency fix to sendto(). 49 * Mike Shaver : RFC1122 checks. 50 * Alan Cox : Nonblocking error fix. 51 * Willy Konynenberg : Transparent proxying support. 52 * Mike McLagan : Routing by source 53 * David S. Miller : New socket lookup architecture. 54 * Last socket cache retained as it 55 * does have a high hit rate. 56 * Olaf Kirch : Don't linearise iovec on sendmsg. 57 * Andi Kleen : Some cleanups, cache destination entry 58 * for connect. 59 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 60 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 61 * return ENOTCONN for unconnected sockets (POSIX) 62 * Janos Farkas : don't deliver multi/broadcasts to a different 63 * bound-to-device socket 64 * Hirokazu Takahashi : HW checksumming for outgoing UDP 65 * datagrams. 66 * Hirokazu Takahashi : sendfile() on UDP works now. 67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 70 * a single port at the same time. 71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 72 * James Chapman : Add L2TP encapsulation type. 73 */ 74 75 #define pr_fmt(fmt) "UDP: " fmt 76 77 #include <linux/bpf-cgroup.h> 78 #include <linux/uaccess.h> 79 #include <asm/ioctls.h> 80 #include <linux/memblock.h> 81 #include <linux/highmem.h> 82 #include <linux/types.h> 83 #include <linux/fcntl.h> 84 #include <linux/module.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/igmp.h> 88 #include <linux/inetdevice.h> 89 #include <linux/in.h> 90 #include <linux/errno.h> 91 #include <linux/timer.h> 92 #include <linux/mm.h> 93 #include <linux/inet.h> 94 #include <linux/netdevice.h> 95 #include <linux/slab.h> 96 #include <net/tcp_states.h> 97 #include <linux/skbuff.h> 98 #include <linux/proc_fs.h> 99 #include <linux/seq_file.h> 100 #include <net/net_namespace.h> 101 #include <net/icmp.h> 102 #include <net/inet_hashtables.h> 103 #include <net/ip_tunnels.h> 104 #include <net/route.h> 105 #include <net/checksum.h> 106 #include <net/xfrm.h> 107 #include <trace/events/udp.h> 108 #include <linux/static_key.h> 109 #include <linux/btf_ids.h> 110 #include <trace/events/skb.h> 111 #include <net/busy_poll.h> 112 #include "udp_impl.h" 113 #include <net/sock_reuseport.h> 114 #include <net/addrconf.h> 115 #include <net/udp_tunnel.h> 116 #if IS_ENABLED(CONFIG_IPV6) 117 #include <net/ipv6_stubs.h> 118 #endif 119 120 struct udp_table udp_table __read_mostly; 121 EXPORT_SYMBOL(udp_table); 122 123 long sysctl_udp_mem[3] __read_mostly; 124 EXPORT_SYMBOL(sysctl_udp_mem); 125 126 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp; 127 EXPORT_SYMBOL(udp_memory_allocated); 128 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc); 129 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc); 130 131 #define MAX_UDP_PORTS 65536 132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET) 133 134 static struct udp_table *udp_get_table_prot(struct sock *sk) 135 { 136 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table; 137 } 138 139 static int udp_lib_lport_inuse(struct net *net, __u16 num, 140 const struct udp_hslot *hslot, 141 unsigned long *bitmap, 142 struct sock *sk, unsigned int log) 143 { 144 struct sock *sk2; 145 kuid_t uid = sock_i_uid(sk); 146 147 sk_for_each(sk2, &hslot->head) { 148 if (net_eq(sock_net(sk2), net) && 149 sk2 != sk && 150 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 151 (!sk2->sk_reuse || !sk->sk_reuse) && 152 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 153 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 154 inet_rcv_saddr_equal(sk, sk2, true)) { 155 if (sk2->sk_reuseport && sk->sk_reuseport && 156 !rcu_access_pointer(sk->sk_reuseport_cb) && 157 uid_eq(uid, sock_i_uid(sk2))) { 158 if (!bitmap) 159 return 0; 160 } else { 161 if (!bitmap) 162 return 1; 163 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 164 bitmap); 165 } 166 } 167 } 168 return 0; 169 } 170 171 /* 172 * Note: we still hold spinlock of primary hash chain, so no other writer 173 * can insert/delete a socket with local_port == num 174 */ 175 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 176 struct udp_hslot *hslot2, 177 struct sock *sk) 178 { 179 struct sock *sk2; 180 kuid_t uid = sock_i_uid(sk); 181 int res = 0; 182 183 spin_lock(&hslot2->lock); 184 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 185 if (net_eq(sock_net(sk2), net) && 186 sk2 != sk && 187 (udp_sk(sk2)->udp_port_hash == num) && 188 (!sk2->sk_reuse || !sk->sk_reuse) && 189 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 190 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 191 inet_rcv_saddr_equal(sk, sk2, true)) { 192 if (sk2->sk_reuseport && sk->sk_reuseport && 193 !rcu_access_pointer(sk->sk_reuseport_cb) && 194 uid_eq(uid, sock_i_uid(sk2))) { 195 res = 0; 196 } else { 197 res = 1; 198 } 199 break; 200 } 201 } 202 spin_unlock(&hslot2->lock); 203 return res; 204 } 205 206 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 207 { 208 struct net *net = sock_net(sk); 209 kuid_t uid = sock_i_uid(sk); 210 struct sock *sk2; 211 212 sk_for_each(sk2, &hslot->head) { 213 if (net_eq(sock_net(sk2), net) && 214 sk2 != sk && 215 sk2->sk_family == sk->sk_family && 216 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 217 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 218 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 219 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 220 inet_rcv_saddr_equal(sk, sk2, false)) { 221 return reuseport_add_sock(sk, sk2, 222 inet_rcv_saddr_any(sk)); 223 } 224 } 225 226 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 227 } 228 229 /** 230 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 231 * 232 * @sk: socket struct in question 233 * @snum: port number to look up 234 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 235 * with NULL address 236 */ 237 int udp_lib_get_port(struct sock *sk, unsigned short snum, 238 unsigned int hash2_nulladdr) 239 { 240 struct udp_table *udptable = udp_get_table_prot(sk); 241 struct udp_hslot *hslot, *hslot2; 242 struct net *net = sock_net(sk); 243 int error = -EADDRINUSE; 244 245 if (!snum) { 246 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 247 unsigned short first, last; 248 int low, high, remaining; 249 unsigned int rand; 250 251 inet_sk_get_local_port_range(sk, &low, &high); 252 remaining = (high - low) + 1; 253 254 rand = get_random_u32(); 255 first = reciprocal_scale(rand, remaining) + low; 256 /* 257 * force rand to be an odd multiple of UDP_HTABLE_SIZE 258 */ 259 rand = (rand | 1) * (udptable->mask + 1); 260 last = first + udptable->mask + 1; 261 do { 262 hslot = udp_hashslot(udptable, net, first); 263 bitmap_zero(bitmap, PORTS_PER_CHAIN); 264 spin_lock_bh(&hslot->lock); 265 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 266 udptable->log); 267 268 snum = first; 269 /* 270 * Iterate on all possible values of snum for this hash. 271 * Using steps of an odd multiple of UDP_HTABLE_SIZE 272 * give us randomization and full range coverage. 273 */ 274 do { 275 if (low <= snum && snum <= high && 276 !test_bit(snum >> udptable->log, bitmap) && 277 !inet_is_local_reserved_port(net, snum)) 278 goto found; 279 snum += rand; 280 } while (snum != first); 281 spin_unlock_bh(&hslot->lock); 282 cond_resched(); 283 } while (++first != last); 284 goto fail; 285 } else { 286 hslot = udp_hashslot(udptable, net, snum); 287 spin_lock_bh(&hslot->lock); 288 if (hslot->count > 10) { 289 int exist; 290 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 291 292 slot2 &= udptable->mask; 293 hash2_nulladdr &= udptable->mask; 294 295 hslot2 = udp_hashslot2(udptable, slot2); 296 if (hslot->count < hslot2->count) 297 goto scan_primary_hash; 298 299 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 300 if (!exist && (hash2_nulladdr != slot2)) { 301 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 302 exist = udp_lib_lport_inuse2(net, snum, hslot2, 303 sk); 304 } 305 if (exist) 306 goto fail_unlock; 307 else 308 goto found; 309 } 310 scan_primary_hash: 311 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 312 goto fail_unlock; 313 } 314 found: 315 inet_sk(sk)->inet_num = snum; 316 udp_sk(sk)->udp_port_hash = snum; 317 udp_sk(sk)->udp_portaddr_hash ^= snum; 318 if (sk_unhashed(sk)) { 319 if (sk->sk_reuseport && 320 udp_reuseport_add_sock(sk, hslot)) { 321 inet_sk(sk)->inet_num = 0; 322 udp_sk(sk)->udp_port_hash = 0; 323 udp_sk(sk)->udp_portaddr_hash ^= snum; 324 goto fail_unlock; 325 } 326 327 sk_add_node_rcu(sk, &hslot->head); 328 hslot->count++; 329 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 330 331 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 332 spin_lock(&hslot2->lock); 333 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 334 sk->sk_family == AF_INET6) 335 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 336 &hslot2->head); 337 else 338 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 339 &hslot2->head); 340 hslot2->count++; 341 spin_unlock(&hslot2->lock); 342 } 343 sock_set_flag(sk, SOCK_RCU_FREE); 344 error = 0; 345 fail_unlock: 346 spin_unlock_bh(&hslot->lock); 347 fail: 348 return error; 349 } 350 EXPORT_SYMBOL(udp_lib_get_port); 351 352 int udp_v4_get_port(struct sock *sk, unsigned short snum) 353 { 354 unsigned int hash2_nulladdr = 355 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 356 unsigned int hash2_partial = 357 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 358 359 /* precompute partial secondary hash */ 360 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 361 return udp_lib_get_port(sk, snum, hash2_nulladdr); 362 } 363 364 static int compute_score(struct sock *sk, struct net *net, 365 __be32 saddr, __be16 sport, 366 __be32 daddr, unsigned short hnum, 367 int dif, int sdif) 368 { 369 int score; 370 struct inet_sock *inet; 371 bool dev_match; 372 373 if (!net_eq(sock_net(sk), net) || 374 udp_sk(sk)->udp_port_hash != hnum || 375 ipv6_only_sock(sk)) 376 return -1; 377 378 if (sk->sk_rcv_saddr != daddr) 379 return -1; 380 381 score = (sk->sk_family == PF_INET) ? 2 : 1; 382 383 inet = inet_sk(sk); 384 if (inet->inet_daddr) { 385 if (inet->inet_daddr != saddr) 386 return -1; 387 score += 4; 388 } 389 390 if (inet->inet_dport) { 391 if (inet->inet_dport != sport) 392 return -1; 393 score += 4; 394 } 395 396 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 397 dif, sdif); 398 if (!dev_match) 399 return -1; 400 if (sk->sk_bound_dev_if) 401 score += 4; 402 403 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 404 score++; 405 return score; 406 } 407 408 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 409 const __u16 lport, const __be32 faddr, 410 const __be16 fport) 411 { 412 static u32 udp_ehash_secret __read_mostly; 413 414 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 415 416 return __inet_ehashfn(laddr, lport, faddr, fport, 417 udp_ehash_secret + net_hash_mix(net)); 418 } 419 420 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 421 struct sk_buff *skb, 422 __be32 saddr, __be16 sport, 423 __be32 daddr, unsigned short hnum) 424 { 425 struct sock *reuse_sk = NULL; 426 u32 hash; 427 428 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 429 hash = udp_ehashfn(net, daddr, hnum, saddr, sport); 430 reuse_sk = reuseport_select_sock(sk, hash, skb, 431 sizeof(struct udphdr)); 432 } 433 return reuse_sk; 434 } 435 436 /* called with rcu_read_lock() */ 437 static struct sock *udp4_lib_lookup2(struct net *net, 438 __be32 saddr, __be16 sport, 439 __be32 daddr, unsigned int hnum, 440 int dif, int sdif, 441 struct udp_hslot *hslot2, 442 struct sk_buff *skb) 443 { 444 struct sock *sk, *result; 445 int score, badness; 446 447 result = NULL; 448 badness = 0; 449 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 450 score = compute_score(sk, net, saddr, sport, 451 daddr, hnum, dif, sdif); 452 if (score > badness) { 453 result = lookup_reuseport(net, sk, skb, 454 saddr, sport, daddr, hnum); 455 /* Fall back to scoring if group has connections */ 456 if (result && !reuseport_has_conns(sk)) 457 return result; 458 459 result = result ? : sk; 460 badness = score; 461 } 462 } 463 return result; 464 } 465 466 static struct sock *udp4_lookup_run_bpf(struct net *net, 467 struct udp_table *udptable, 468 struct sk_buff *skb, 469 __be32 saddr, __be16 sport, 470 __be32 daddr, u16 hnum, const int dif) 471 { 472 struct sock *sk, *reuse_sk; 473 bool no_reuseport; 474 475 if (udptable != net->ipv4.udp_table) 476 return NULL; /* only UDP is supported */ 477 478 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport, 479 daddr, hnum, dif, &sk); 480 if (no_reuseport || IS_ERR_OR_NULL(sk)) 481 return sk; 482 483 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 484 if (reuse_sk) 485 sk = reuse_sk; 486 return sk; 487 } 488 489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 490 * harder than this. -DaveM 491 */ 492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 493 __be16 sport, __be32 daddr, __be16 dport, int dif, 494 int sdif, struct udp_table *udptable, struct sk_buff *skb) 495 { 496 unsigned short hnum = ntohs(dport); 497 unsigned int hash2, slot2; 498 struct udp_hslot *hslot2; 499 struct sock *result, *sk; 500 501 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 502 slot2 = hash2 & udptable->mask; 503 hslot2 = &udptable->hash2[slot2]; 504 505 /* Lookup connected or non-wildcard socket */ 506 result = udp4_lib_lookup2(net, saddr, sport, 507 daddr, hnum, dif, sdif, 508 hslot2, skb); 509 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 510 goto done; 511 512 /* Lookup redirect from BPF */ 513 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { 514 sk = udp4_lookup_run_bpf(net, udptable, skb, 515 saddr, sport, daddr, hnum, dif); 516 if (sk) { 517 result = sk; 518 goto done; 519 } 520 } 521 522 /* Got non-wildcard socket or error on first lookup */ 523 if (result) 524 goto done; 525 526 /* Lookup wildcard sockets */ 527 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 528 slot2 = hash2 & udptable->mask; 529 hslot2 = &udptable->hash2[slot2]; 530 531 result = udp4_lib_lookup2(net, saddr, sport, 532 htonl(INADDR_ANY), hnum, dif, sdif, 533 hslot2, skb); 534 done: 535 if (IS_ERR(result)) 536 return NULL; 537 return result; 538 } 539 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 540 541 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 542 __be16 sport, __be16 dport, 543 struct udp_table *udptable) 544 { 545 const struct iphdr *iph = ip_hdr(skb); 546 547 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 548 iph->daddr, dport, inet_iif(skb), 549 inet_sdif(skb), udptable, skb); 550 } 551 552 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 553 __be16 sport, __be16 dport) 554 { 555 const struct iphdr *iph = ip_hdr(skb); 556 struct net *net = dev_net(skb->dev); 557 558 return __udp4_lib_lookup(net, iph->saddr, sport, 559 iph->daddr, dport, inet_iif(skb), 560 inet_sdif(skb), net->ipv4.udp_table, NULL); 561 } 562 563 /* Must be called under rcu_read_lock(). 564 * Does increment socket refcount. 565 */ 566 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 567 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 568 __be32 daddr, __be16 dport, int dif) 569 { 570 struct sock *sk; 571 572 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 573 dif, 0, net->ipv4.udp_table, NULL); 574 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 575 sk = NULL; 576 return sk; 577 } 578 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 579 #endif 580 581 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk, 582 __be16 loc_port, __be32 loc_addr, 583 __be16 rmt_port, __be32 rmt_addr, 584 int dif, int sdif, unsigned short hnum) 585 { 586 const struct inet_sock *inet = inet_sk(sk); 587 588 if (!net_eq(sock_net(sk), net) || 589 udp_sk(sk)->udp_port_hash != hnum || 590 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 591 (inet->inet_dport != rmt_port && inet->inet_dport) || 592 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 593 ipv6_only_sock(sk) || 594 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 595 return false; 596 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 597 return false; 598 return true; 599 } 600 601 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 602 void udp_encap_enable(void) 603 { 604 static_branch_inc(&udp_encap_needed_key); 605 } 606 EXPORT_SYMBOL(udp_encap_enable); 607 608 void udp_encap_disable(void) 609 { 610 static_branch_dec(&udp_encap_needed_key); 611 } 612 EXPORT_SYMBOL(udp_encap_disable); 613 614 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 615 * through error handlers in encapsulations looking for a match. 616 */ 617 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 618 { 619 int i; 620 621 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 622 int (*handler)(struct sk_buff *skb, u32 info); 623 const struct ip_tunnel_encap_ops *encap; 624 625 encap = rcu_dereference(iptun_encaps[i]); 626 if (!encap) 627 continue; 628 handler = encap->err_handler; 629 if (handler && !handler(skb, info)) 630 return 0; 631 } 632 633 return -ENOENT; 634 } 635 636 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 637 * reversing source and destination port: this will match tunnels that force the 638 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 639 * lwtunnels might actually break this assumption by being configured with 640 * different destination ports on endpoints, in this case we won't be able to 641 * trace ICMP messages back to them. 642 * 643 * If this doesn't match any socket, probe tunnels with arbitrary destination 644 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 645 * we've sent packets to won't necessarily match the local destination port. 646 * 647 * Then ask the tunnel implementation to match the error against a valid 648 * association. 649 * 650 * Return an error if we can't find a match, the socket if we need further 651 * processing, zero otherwise. 652 */ 653 static struct sock *__udp4_lib_err_encap(struct net *net, 654 const struct iphdr *iph, 655 struct udphdr *uh, 656 struct udp_table *udptable, 657 struct sock *sk, 658 struct sk_buff *skb, u32 info) 659 { 660 int (*lookup)(struct sock *sk, struct sk_buff *skb); 661 int network_offset, transport_offset; 662 struct udp_sock *up; 663 664 network_offset = skb_network_offset(skb); 665 transport_offset = skb_transport_offset(skb); 666 667 /* Network header needs to point to the outer IPv4 header inside ICMP */ 668 skb_reset_network_header(skb); 669 670 /* Transport header needs to point to the UDP header */ 671 skb_set_transport_header(skb, iph->ihl << 2); 672 673 if (sk) { 674 up = udp_sk(sk); 675 676 lookup = READ_ONCE(up->encap_err_lookup); 677 if (lookup && lookup(sk, skb)) 678 sk = NULL; 679 680 goto out; 681 } 682 683 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 684 iph->saddr, uh->dest, skb->dev->ifindex, 0, 685 udptable, NULL); 686 if (sk) { 687 up = udp_sk(sk); 688 689 lookup = READ_ONCE(up->encap_err_lookup); 690 if (!lookup || lookup(sk, skb)) 691 sk = NULL; 692 } 693 694 out: 695 if (!sk) 696 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 697 698 skb_set_transport_header(skb, transport_offset); 699 skb_set_network_header(skb, network_offset); 700 701 return sk; 702 } 703 704 /* 705 * This routine is called by the ICMP module when it gets some 706 * sort of error condition. If err < 0 then the socket should 707 * be closed and the error returned to the user. If err > 0 708 * it's just the icmp type << 8 | icmp code. 709 * Header points to the ip header of the error packet. We move 710 * on past this. Then (as it used to claim before adjustment) 711 * header points to the first 8 bytes of the udp header. We need 712 * to find the appropriate port. 713 */ 714 715 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 716 { 717 struct inet_sock *inet; 718 const struct iphdr *iph = (const struct iphdr *)skb->data; 719 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 720 const int type = icmp_hdr(skb)->type; 721 const int code = icmp_hdr(skb)->code; 722 bool tunnel = false; 723 struct sock *sk; 724 int harderr; 725 int err; 726 struct net *net = dev_net(skb->dev); 727 728 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 729 iph->saddr, uh->source, skb->dev->ifindex, 730 inet_sdif(skb), udptable, NULL); 731 732 if (!sk || udp_sk(sk)->encap_type) { 733 /* No socket for error: try tunnels before discarding */ 734 if (static_branch_unlikely(&udp_encap_needed_key)) { 735 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 736 info); 737 if (!sk) 738 return 0; 739 } else 740 sk = ERR_PTR(-ENOENT); 741 742 if (IS_ERR(sk)) { 743 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 744 return PTR_ERR(sk); 745 } 746 747 tunnel = true; 748 } 749 750 err = 0; 751 harderr = 0; 752 inet = inet_sk(sk); 753 754 switch (type) { 755 default: 756 case ICMP_TIME_EXCEEDED: 757 err = EHOSTUNREACH; 758 break; 759 case ICMP_SOURCE_QUENCH: 760 goto out; 761 case ICMP_PARAMETERPROB: 762 err = EPROTO; 763 harderr = 1; 764 break; 765 case ICMP_DEST_UNREACH: 766 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 767 ipv4_sk_update_pmtu(skb, sk, info); 768 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 769 err = EMSGSIZE; 770 harderr = 1; 771 break; 772 } 773 goto out; 774 } 775 err = EHOSTUNREACH; 776 if (code <= NR_ICMP_UNREACH) { 777 harderr = icmp_err_convert[code].fatal; 778 err = icmp_err_convert[code].errno; 779 } 780 break; 781 case ICMP_REDIRECT: 782 ipv4_sk_redirect(skb, sk); 783 goto out; 784 } 785 786 /* 787 * RFC1122: OK. Passes ICMP errors back to application, as per 788 * 4.1.3.3. 789 */ 790 if (tunnel) { 791 /* ...not for tunnels though: we don't have a sending socket */ 792 if (udp_sk(sk)->encap_err_rcv) 793 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info, 794 (u8 *)(uh+1)); 795 goto out; 796 } 797 if (!inet->recverr) { 798 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 799 goto out; 800 } else 801 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 802 803 sk->sk_err = err; 804 sk_error_report(sk); 805 out: 806 return 0; 807 } 808 809 int udp_err(struct sk_buff *skb, u32 info) 810 { 811 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table); 812 } 813 814 /* 815 * Throw away all pending data and cancel the corking. Socket is locked. 816 */ 817 void udp_flush_pending_frames(struct sock *sk) 818 { 819 struct udp_sock *up = udp_sk(sk); 820 821 if (up->pending) { 822 up->len = 0; 823 up->pending = 0; 824 ip_flush_pending_frames(sk); 825 } 826 } 827 EXPORT_SYMBOL(udp_flush_pending_frames); 828 829 /** 830 * udp4_hwcsum - handle outgoing HW checksumming 831 * @skb: sk_buff containing the filled-in UDP header 832 * (checksum field must be zeroed out) 833 * @src: source IP address 834 * @dst: destination IP address 835 */ 836 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 837 { 838 struct udphdr *uh = udp_hdr(skb); 839 int offset = skb_transport_offset(skb); 840 int len = skb->len - offset; 841 int hlen = len; 842 __wsum csum = 0; 843 844 if (!skb_has_frag_list(skb)) { 845 /* 846 * Only one fragment on the socket. 847 */ 848 skb->csum_start = skb_transport_header(skb) - skb->head; 849 skb->csum_offset = offsetof(struct udphdr, check); 850 uh->check = ~csum_tcpudp_magic(src, dst, len, 851 IPPROTO_UDP, 0); 852 } else { 853 struct sk_buff *frags; 854 855 /* 856 * HW-checksum won't work as there are two or more 857 * fragments on the socket so that all csums of sk_buffs 858 * should be together 859 */ 860 skb_walk_frags(skb, frags) { 861 csum = csum_add(csum, frags->csum); 862 hlen -= frags->len; 863 } 864 865 csum = skb_checksum(skb, offset, hlen, csum); 866 skb->ip_summed = CHECKSUM_NONE; 867 868 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 869 if (uh->check == 0) 870 uh->check = CSUM_MANGLED_0; 871 } 872 } 873 EXPORT_SYMBOL_GPL(udp4_hwcsum); 874 875 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 876 * for the simple case like when setting the checksum for a UDP tunnel. 877 */ 878 void udp_set_csum(bool nocheck, struct sk_buff *skb, 879 __be32 saddr, __be32 daddr, int len) 880 { 881 struct udphdr *uh = udp_hdr(skb); 882 883 if (nocheck) { 884 uh->check = 0; 885 } else if (skb_is_gso(skb)) { 886 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 887 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 888 uh->check = 0; 889 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 890 if (uh->check == 0) 891 uh->check = CSUM_MANGLED_0; 892 } else { 893 skb->ip_summed = CHECKSUM_PARTIAL; 894 skb->csum_start = skb_transport_header(skb) - skb->head; 895 skb->csum_offset = offsetof(struct udphdr, check); 896 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 897 } 898 } 899 EXPORT_SYMBOL(udp_set_csum); 900 901 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 902 struct inet_cork *cork) 903 { 904 struct sock *sk = skb->sk; 905 struct inet_sock *inet = inet_sk(sk); 906 struct udphdr *uh; 907 int err; 908 int is_udplite = IS_UDPLITE(sk); 909 int offset = skb_transport_offset(skb); 910 int len = skb->len - offset; 911 int datalen = len - sizeof(*uh); 912 __wsum csum = 0; 913 914 /* 915 * Create a UDP header 916 */ 917 uh = udp_hdr(skb); 918 uh->source = inet->inet_sport; 919 uh->dest = fl4->fl4_dport; 920 uh->len = htons(len); 921 uh->check = 0; 922 923 if (cork->gso_size) { 924 const int hlen = skb_network_header_len(skb) + 925 sizeof(struct udphdr); 926 927 if (hlen + cork->gso_size > cork->fragsize) { 928 kfree_skb(skb); 929 return -EINVAL; 930 } 931 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 932 kfree_skb(skb); 933 return -EINVAL; 934 } 935 if (sk->sk_no_check_tx) { 936 kfree_skb(skb); 937 return -EINVAL; 938 } 939 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 940 dst_xfrm(skb_dst(skb))) { 941 kfree_skb(skb); 942 return -EIO; 943 } 944 945 if (datalen > cork->gso_size) { 946 skb_shinfo(skb)->gso_size = cork->gso_size; 947 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 948 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 949 cork->gso_size); 950 } 951 goto csum_partial; 952 } 953 954 if (is_udplite) /* UDP-Lite */ 955 csum = udplite_csum(skb); 956 957 else if (sk->sk_no_check_tx) { /* UDP csum off */ 958 959 skb->ip_summed = CHECKSUM_NONE; 960 goto send; 961 962 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 963 csum_partial: 964 965 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 966 goto send; 967 968 } else 969 csum = udp_csum(skb); 970 971 /* add protocol-dependent pseudo-header */ 972 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 973 sk->sk_protocol, csum); 974 if (uh->check == 0) 975 uh->check = CSUM_MANGLED_0; 976 977 send: 978 err = ip_send_skb(sock_net(sk), skb); 979 if (err) { 980 if (err == -ENOBUFS && !inet->recverr) { 981 UDP_INC_STATS(sock_net(sk), 982 UDP_MIB_SNDBUFERRORS, is_udplite); 983 err = 0; 984 } 985 } else 986 UDP_INC_STATS(sock_net(sk), 987 UDP_MIB_OUTDATAGRAMS, is_udplite); 988 return err; 989 } 990 991 /* 992 * Push out all pending data as one UDP datagram. Socket is locked. 993 */ 994 int udp_push_pending_frames(struct sock *sk) 995 { 996 struct udp_sock *up = udp_sk(sk); 997 struct inet_sock *inet = inet_sk(sk); 998 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 999 struct sk_buff *skb; 1000 int err = 0; 1001 1002 skb = ip_finish_skb(sk, fl4); 1003 if (!skb) 1004 goto out; 1005 1006 err = udp_send_skb(skb, fl4, &inet->cork.base); 1007 1008 out: 1009 up->len = 0; 1010 up->pending = 0; 1011 return err; 1012 } 1013 EXPORT_SYMBOL(udp_push_pending_frames); 1014 1015 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1016 { 1017 switch (cmsg->cmsg_type) { 1018 case UDP_SEGMENT: 1019 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1020 return -EINVAL; 1021 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1022 return 0; 1023 default: 1024 return -EINVAL; 1025 } 1026 } 1027 1028 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1029 { 1030 struct cmsghdr *cmsg; 1031 bool need_ip = false; 1032 int err; 1033 1034 for_each_cmsghdr(cmsg, msg) { 1035 if (!CMSG_OK(msg, cmsg)) 1036 return -EINVAL; 1037 1038 if (cmsg->cmsg_level != SOL_UDP) { 1039 need_ip = true; 1040 continue; 1041 } 1042 1043 err = __udp_cmsg_send(cmsg, gso_size); 1044 if (err) 1045 return err; 1046 } 1047 1048 return need_ip; 1049 } 1050 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1051 1052 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1053 { 1054 struct inet_sock *inet = inet_sk(sk); 1055 struct udp_sock *up = udp_sk(sk); 1056 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1057 struct flowi4 fl4_stack; 1058 struct flowi4 *fl4; 1059 int ulen = len; 1060 struct ipcm_cookie ipc; 1061 struct rtable *rt = NULL; 1062 int free = 0; 1063 int connected = 0; 1064 __be32 daddr, faddr, saddr; 1065 u8 tos, scope; 1066 __be16 dport; 1067 int err, is_udplite = IS_UDPLITE(sk); 1068 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1069 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1070 struct sk_buff *skb; 1071 struct ip_options_data opt_copy; 1072 1073 if (len > 0xFFFF) 1074 return -EMSGSIZE; 1075 1076 /* 1077 * Check the flags. 1078 */ 1079 1080 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1081 return -EOPNOTSUPP; 1082 1083 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1084 1085 fl4 = &inet->cork.fl.u.ip4; 1086 if (up->pending) { 1087 /* 1088 * There are pending frames. 1089 * The socket lock must be held while it's corked. 1090 */ 1091 lock_sock(sk); 1092 if (likely(up->pending)) { 1093 if (unlikely(up->pending != AF_INET)) { 1094 release_sock(sk); 1095 return -EINVAL; 1096 } 1097 goto do_append_data; 1098 } 1099 release_sock(sk); 1100 } 1101 ulen += sizeof(struct udphdr); 1102 1103 /* 1104 * Get and verify the address. 1105 */ 1106 if (usin) { 1107 if (msg->msg_namelen < sizeof(*usin)) 1108 return -EINVAL; 1109 if (usin->sin_family != AF_INET) { 1110 if (usin->sin_family != AF_UNSPEC) 1111 return -EAFNOSUPPORT; 1112 } 1113 1114 daddr = usin->sin_addr.s_addr; 1115 dport = usin->sin_port; 1116 if (dport == 0) 1117 return -EINVAL; 1118 } else { 1119 if (sk->sk_state != TCP_ESTABLISHED) 1120 return -EDESTADDRREQ; 1121 daddr = inet->inet_daddr; 1122 dport = inet->inet_dport; 1123 /* Open fast path for connected socket. 1124 Route will not be used, if at least one option is set. 1125 */ 1126 connected = 1; 1127 } 1128 1129 ipcm_init_sk(&ipc, inet); 1130 ipc.gso_size = READ_ONCE(up->gso_size); 1131 1132 if (msg->msg_controllen) { 1133 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1134 if (err > 0) 1135 err = ip_cmsg_send(sk, msg, &ipc, 1136 sk->sk_family == AF_INET6); 1137 if (unlikely(err < 0)) { 1138 kfree(ipc.opt); 1139 return err; 1140 } 1141 if (ipc.opt) 1142 free = 1; 1143 connected = 0; 1144 } 1145 if (!ipc.opt) { 1146 struct ip_options_rcu *inet_opt; 1147 1148 rcu_read_lock(); 1149 inet_opt = rcu_dereference(inet->inet_opt); 1150 if (inet_opt) { 1151 memcpy(&opt_copy, inet_opt, 1152 sizeof(*inet_opt) + inet_opt->opt.optlen); 1153 ipc.opt = &opt_copy.opt; 1154 } 1155 rcu_read_unlock(); 1156 } 1157 1158 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1159 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1160 (struct sockaddr *)usin, &ipc.addr); 1161 if (err) 1162 goto out_free; 1163 if (usin) { 1164 if (usin->sin_port == 0) { 1165 /* BPF program set invalid port. Reject it. */ 1166 err = -EINVAL; 1167 goto out_free; 1168 } 1169 daddr = usin->sin_addr.s_addr; 1170 dport = usin->sin_port; 1171 } 1172 } 1173 1174 saddr = ipc.addr; 1175 ipc.addr = faddr = daddr; 1176 1177 if (ipc.opt && ipc.opt->opt.srr) { 1178 if (!daddr) { 1179 err = -EINVAL; 1180 goto out_free; 1181 } 1182 faddr = ipc.opt->opt.faddr; 1183 connected = 0; 1184 } 1185 tos = get_rttos(&ipc, inet); 1186 scope = ip_sendmsg_scope(inet, &ipc, msg); 1187 if (scope == RT_SCOPE_LINK) 1188 connected = 0; 1189 1190 if (ipv4_is_multicast(daddr)) { 1191 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1192 ipc.oif = inet->mc_index; 1193 if (!saddr) 1194 saddr = inet->mc_addr; 1195 connected = 0; 1196 } else if (!ipc.oif) { 1197 ipc.oif = inet->uc_index; 1198 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1199 /* oif is set, packet is to local broadcast and 1200 * uc_index is set. oif is most likely set 1201 * by sk_bound_dev_if. If uc_index != oif check if the 1202 * oif is an L3 master and uc_index is an L3 slave. 1203 * If so, we want to allow the send using the uc_index. 1204 */ 1205 if (ipc.oif != inet->uc_index && 1206 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1207 inet->uc_index)) { 1208 ipc.oif = inet->uc_index; 1209 } 1210 } 1211 1212 if (connected) 1213 rt = (struct rtable *)sk_dst_check(sk, 0); 1214 1215 if (!rt) { 1216 struct net *net = sock_net(sk); 1217 __u8 flow_flags = inet_sk_flowi_flags(sk); 1218 1219 fl4 = &fl4_stack; 1220 1221 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope, 1222 sk->sk_protocol, flow_flags, faddr, saddr, 1223 dport, inet->inet_sport, sk->sk_uid); 1224 1225 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1226 rt = ip_route_output_flow(net, fl4, sk); 1227 if (IS_ERR(rt)) { 1228 err = PTR_ERR(rt); 1229 rt = NULL; 1230 if (err == -ENETUNREACH) 1231 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1232 goto out; 1233 } 1234 1235 err = -EACCES; 1236 if ((rt->rt_flags & RTCF_BROADCAST) && 1237 !sock_flag(sk, SOCK_BROADCAST)) 1238 goto out; 1239 if (connected) 1240 sk_dst_set(sk, dst_clone(&rt->dst)); 1241 } 1242 1243 if (msg->msg_flags&MSG_CONFIRM) 1244 goto do_confirm; 1245 back_from_confirm: 1246 1247 saddr = fl4->saddr; 1248 if (!ipc.addr) 1249 daddr = ipc.addr = fl4->daddr; 1250 1251 /* Lockless fast path for the non-corking case. */ 1252 if (!corkreq) { 1253 struct inet_cork cork; 1254 1255 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1256 sizeof(struct udphdr), &ipc, &rt, 1257 &cork, msg->msg_flags); 1258 err = PTR_ERR(skb); 1259 if (!IS_ERR_OR_NULL(skb)) 1260 err = udp_send_skb(skb, fl4, &cork); 1261 goto out; 1262 } 1263 1264 lock_sock(sk); 1265 if (unlikely(up->pending)) { 1266 /* The socket is already corked while preparing it. */ 1267 /* ... which is an evident application bug. --ANK */ 1268 release_sock(sk); 1269 1270 net_dbg_ratelimited("socket already corked\n"); 1271 err = -EINVAL; 1272 goto out; 1273 } 1274 /* 1275 * Now cork the socket to pend data. 1276 */ 1277 fl4 = &inet->cork.fl.u.ip4; 1278 fl4->daddr = daddr; 1279 fl4->saddr = saddr; 1280 fl4->fl4_dport = dport; 1281 fl4->fl4_sport = inet->inet_sport; 1282 up->pending = AF_INET; 1283 1284 do_append_data: 1285 up->len += ulen; 1286 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1287 sizeof(struct udphdr), &ipc, &rt, 1288 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1289 if (err) 1290 udp_flush_pending_frames(sk); 1291 else if (!corkreq) 1292 err = udp_push_pending_frames(sk); 1293 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1294 up->pending = 0; 1295 release_sock(sk); 1296 1297 out: 1298 ip_rt_put(rt); 1299 out_free: 1300 if (free) 1301 kfree(ipc.opt); 1302 if (!err) 1303 return len; 1304 /* 1305 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1306 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1307 * we don't have a good statistic (IpOutDiscards but it can be too many 1308 * things). We could add another new stat but at least for now that 1309 * seems like overkill. 1310 */ 1311 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1312 UDP_INC_STATS(sock_net(sk), 1313 UDP_MIB_SNDBUFERRORS, is_udplite); 1314 } 1315 return err; 1316 1317 do_confirm: 1318 if (msg->msg_flags & MSG_PROBE) 1319 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1320 if (!(msg->msg_flags&MSG_PROBE) || len) 1321 goto back_from_confirm; 1322 err = 0; 1323 goto out; 1324 } 1325 EXPORT_SYMBOL(udp_sendmsg); 1326 1327 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1328 size_t size, int flags) 1329 { 1330 struct bio_vec bvec; 1331 struct msghdr msg = { .msg_flags = flags | MSG_SPLICE_PAGES }; 1332 1333 if (flags & MSG_SENDPAGE_NOTLAST) 1334 msg.msg_flags |= MSG_MORE; 1335 1336 bvec_set_page(&bvec, page, size, offset); 1337 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size); 1338 return udp_sendmsg(sk, &msg, size); 1339 } 1340 1341 #define UDP_SKB_IS_STATELESS 0x80000000 1342 1343 /* all head states (dst, sk, nf conntrack) except skb extensions are 1344 * cleared by udp_rcv(). 1345 * 1346 * We need to preserve secpath, if present, to eventually process 1347 * IP_CMSG_PASSSEC at recvmsg() time. 1348 * 1349 * Other extensions can be cleared. 1350 */ 1351 static bool udp_try_make_stateless(struct sk_buff *skb) 1352 { 1353 if (!skb_has_extensions(skb)) 1354 return true; 1355 1356 if (!secpath_exists(skb)) { 1357 skb_ext_reset(skb); 1358 return true; 1359 } 1360 1361 return false; 1362 } 1363 1364 static void udp_set_dev_scratch(struct sk_buff *skb) 1365 { 1366 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1367 1368 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1369 scratch->_tsize_state = skb->truesize; 1370 #if BITS_PER_LONG == 64 1371 scratch->len = skb->len; 1372 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1373 scratch->is_linear = !skb_is_nonlinear(skb); 1374 #endif 1375 if (udp_try_make_stateless(skb)) 1376 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1377 } 1378 1379 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1380 { 1381 /* We come here after udp_lib_checksum_complete() returned 0. 1382 * This means that __skb_checksum_complete() might have 1383 * set skb->csum_valid to 1. 1384 * On 64bit platforms, we can set csum_unnecessary 1385 * to true, but only if the skb is not shared. 1386 */ 1387 #if BITS_PER_LONG == 64 1388 if (!skb_shared(skb)) 1389 udp_skb_scratch(skb)->csum_unnecessary = true; 1390 #endif 1391 } 1392 1393 static int udp_skb_truesize(struct sk_buff *skb) 1394 { 1395 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1396 } 1397 1398 static bool udp_skb_has_head_state(struct sk_buff *skb) 1399 { 1400 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1401 } 1402 1403 /* fully reclaim rmem/fwd memory allocated for skb */ 1404 static void udp_rmem_release(struct sock *sk, int size, int partial, 1405 bool rx_queue_lock_held) 1406 { 1407 struct udp_sock *up = udp_sk(sk); 1408 struct sk_buff_head *sk_queue; 1409 int amt; 1410 1411 if (likely(partial)) { 1412 up->forward_deficit += size; 1413 size = up->forward_deficit; 1414 if (size < READ_ONCE(up->forward_threshold) && 1415 !skb_queue_empty(&up->reader_queue)) 1416 return; 1417 } else { 1418 size += up->forward_deficit; 1419 } 1420 up->forward_deficit = 0; 1421 1422 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1423 * if the called don't held it already 1424 */ 1425 sk_queue = &sk->sk_receive_queue; 1426 if (!rx_queue_lock_held) 1427 spin_lock(&sk_queue->lock); 1428 1429 1430 sk->sk_forward_alloc += size; 1431 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1432 sk->sk_forward_alloc -= amt; 1433 1434 if (amt) 1435 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 1436 1437 atomic_sub(size, &sk->sk_rmem_alloc); 1438 1439 /* this can save us from acquiring the rx queue lock on next receive */ 1440 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1441 1442 if (!rx_queue_lock_held) 1443 spin_unlock(&sk_queue->lock); 1444 } 1445 1446 /* Note: called with reader_queue.lock held. 1447 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1448 * This avoids a cache line miss while receive_queue lock is held. 1449 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1450 */ 1451 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1452 { 1453 prefetch(&skb->data); 1454 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1455 } 1456 EXPORT_SYMBOL(udp_skb_destructor); 1457 1458 /* as above, but the caller held the rx queue lock, too */ 1459 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1460 { 1461 prefetch(&skb->data); 1462 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1463 } 1464 1465 /* Idea of busylocks is to let producers grab an extra spinlock 1466 * to relieve pressure on the receive_queue spinlock shared by consumer. 1467 * Under flood, this means that only one producer can be in line 1468 * trying to acquire the receive_queue spinlock. 1469 * These busylock can be allocated on a per cpu manner, instead of a 1470 * per socket one (that would consume a cache line per socket) 1471 */ 1472 static int udp_busylocks_log __read_mostly; 1473 static spinlock_t *udp_busylocks __read_mostly; 1474 1475 static spinlock_t *busylock_acquire(void *ptr) 1476 { 1477 spinlock_t *busy; 1478 1479 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1480 spin_lock(busy); 1481 return busy; 1482 } 1483 1484 static void busylock_release(spinlock_t *busy) 1485 { 1486 if (busy) 1487 spin_unlock(busy); 1488 } 1489 1490 static int udp_rmem_schedule(struct sock *sk, int size) 1491 { 1492 int delta; 1493 1494 delta = size - sk->sk_forward_alloc; 1495 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV)) 1496 return -ENOBUFS; 1497 1498 return 0; 1499 } 1500 1501 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1502 { 1503 struct sk_buff_head *list = &sk->sk_receive_queue; 1504 int rmem, err = -ENOMEM; 1505 spinlock_t *busy = NULL; 1506 int size; 1507 1508 /* try to avoid the costly atomic add/sub pair when the receive 1509 * queue is full; always allow at least a packet 1510 */ 1511 rmem = atomic_read(&sk->sk_rmem_alloc); 1512 if (rmem > sk->sk_rcvbuf) 1513 goto drop; 1514 1515 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1516 * having linear skbs : 1517 * - Reduce memory overhead and thus increase receive queue capacity 1518 * - Less cache line misses at copyout() time 1519 * - Less work at consume_skb() (less alien page frag freeing) 1520 */ 1521 if (rmem > (sk->sk_rcvbuf >> 1)) { 1522 skb_condense(skb); 1523 1524 busy = busylock_acquire(sk); 1525 } 1526 size = skb->truesize; 1527 udp_set_dev_scratch(skb); 1528 1529 /* we drop only if the receive buf is full and the receive 1530 * queue contains some other skb 1531 */ 1532 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1533 if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) 1534 goto uncharge_drop; 1535 1536 spin_lock(&list->lock); 1537 err = udp_rmem_schedule(sk, size); 1538 if (err) { 1539 spin_unlock(&list->lock); 1540 goto uncharge_drop; 1541 } 1542 1543 sk->sk_forward_alloc -= size; 1544 1545 /* no need to setup a destructor, we will explicitly release the 1546 * forward allocated memory on dequeue 1547 */ 1548 sock_skb_set_dropcount(sk, skb); 1549 1550 __skb_queue_tail(list, skb); 1551 spin_unlock(&list->lock); 1552 1553 if (!sock_flag(sk, SOCK_DEAD)) 1554 sk->sk_data_ready(sk); 1555 1556 busylock_release(busy); 1557 return 0; 1558 1559 uncharge_drop: 1560 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1561 1562 drop: 1563 atomic_inc(&sk->sk_drops); 1564 busylock_release(busy); 1565 return err; 1566 } 1567 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1568 1569 void udp_destruct_common(struct sock *sk) 1570 { 1571 /* reclaim completely the forward allocated memory */ 1572 struct udp_sock *up = udp_sk(sk); 1573 unsigned int total = 0; 1574 struct sk_buff *skb; 1575 1576 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1577 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1578 total += skb->truesize; 1579 kfree_skb(skb); 1580 } 1581 udp_rmem_release(sk, total, 0, true); 1582 } 1583 EXPORT_SYMBOL_GPL(udp_destruct_common); 1584 1585 static void udp_destruct_sock(struct sock *sk) 1586 { 1587 udp_destruct_common(sk); 1588 inet_sock_destruct(sk); 1589 } 1590 1591 int udp_init_sock(struct sock *sk) 1592 { 1593 udp_lib_init_sock(sk); 1594 sk->sk_destruct = udp_destruct_sock; 1595 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1596 return 0; 1597 } 1598 1599 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1600 { 1601 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1602 bool slow = lock_sock_fast(sk); 1603 1604 sk_peek_offset_bwd(sk, len); 1605 unlock_sock_fast(sk, slow); 1606 } 1607 1608 if (!skb_unref(skb)) 1609 return; 1610 1611 /* In the more common cases we cleared the head states previously, 1612 * see __udp_queue_rcv_skb(). 1613 */ 1614 if (unlikely(udp_skb_has_head_state(skb))) 1615 skb_release_head_state(skb); 1616 __consume_stateless_skb(skb); 1617 } 1618 EXPORT_SYMBOL_GPL(skb_consume_udp); 1619 1620 static struct sk_buff *__first_packet_length(struct sock *sk, 1621 struct sk_buff_head *rcvq, 1622 int *total) 1623 { 1624 struct sk_buff *skb; 1625 1626 while ((skb = skb_peek(rcvq)) != NULL) { 1627 if (udp_lib_checksum_complete(skb)) { 1628 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1629 IS_UDPLITE(sk)); 1630 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1631 IS_UDPLITE(sk)); 1632 atomic_inc(&sk->sk_drops); 1633 __skb_unlink(skb, rcvq); 1634 *total += skb->truesize; 1635 kfree_skb(skb); 1636 } else { 1637 udp_skb_csum_unnecessary_set(skb); 1638 break; 1639 } 1640 } 1641 return skb; 1642 } 1643 1644 /** 1645 * first_packet_length - return length of first packet in receive queue 1646 * @sk: socket 1647 * 1648 * Drops all bad checksum frames, until a valid one is found. 1649 * Returns the length of found skb, or -1 if none is found. 1650 */ 1651 static int first_packet_length(struct sock *sk) 1652 { 1653 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1654 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1655 struct sk_buff *skb; 1656 int total = 0; 1657 int res; 1658 1659 spin_lock_bh(&rcvq->lock); 1660 skb = __first_packet_length(sk, rcvq, &total); 1661 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1662 spin_lock(&sk_queue->lock); 1663 skb_queue_splice_tail_init(sk_queue, rcvq); 1664 spin_unlock(&sk_queue->lock); 1665 1666 skb = __first_packet_length(sk, rcvq, &total); 1667 } 1668 res = skb ? skb->len : -1; 1669 if (total) 1670 udp_rmem_release(sk, total, 1, false); 1671 spin_unlock_bh(&rcvq->lock); 1672 return res; 1673 } 1674 1675 /* 1676 * IOCTL requests applicable to the UDP protocol 1677 */ 1678 1679 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1680 { 1681 switch (cmd) { 1682 case SIOCOUTQ: 1683 { 1684 int amount = sk_wmem_alloc_get(sk); 1685 1686 return put_user(amount, (int __user *)arg); 1687 } 1688 1689 case SIOCINQ: 1690 { 1691 int amount = max_t(int, 0, first_packet_length(sk)); 1692 1693 return put_user(amount, (int __user *)arg); 1694 } 1695 1696 default: 1697 return -ENOIOCTLCMD; 1698 } 1699 1700 return 0; 1701 } 1702 EXPORT_SYMBOL(udp_ioctl); 1703 1704 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1705 int *off, int *err) 1706 { 1707 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1708 struct sk_buff_head *queue; 1709 struct sk_buff *last; 1710 long timeo; 1711 int error; 1712 1713 queue = &udp_sk(sk)->reader_queue; 1714 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1715 do { 1716 struct sk_buff *skb; 1717 1718 error = sock_error(sk); 1719 if (error) 1720 break; 1721 1722 error = -EAGAIN; 1723 do { 1724 spin_lock_bh(&queue->lock); 1725 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1726 err, &last); 1727 if (skb) { 1728 if (!(flags & MSG_PEEK)) 1729 udp_skb_destructor(sk, skb); 1730 spin_unlock_bh(&queue->lock); 1731 return skb; 1732 } 1733 1734 if (skb_queue_empty_lockless(sk_queue)) { 1735 spin_unlock_bh(&queue->lock); 1736 goto busy_check; 1737 } 1738 1739 /* refill the reader queue and walk it again 1740 * keep both queues locked to avoid re-acquiring 1741 * the sk_receive_queue lock if fwd memory scheduling 1742 * is needed. 1743 */ 1744 spin_lock(&sk_queue->lock); 1745 skb_queue_splice_tail_init(sk_queue, queue); 1746 1747 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1748 err, &last); 1749 if (skb && !(flags & MSG_PEEK)) 1750 udp_skb_dtor_locked(sk, skb); 1751 spin_unlock(&sk_queue->lock); 1752 spin_unlock_bh(&queue->lock); 1753 if (skb) 1754 return skb; 1755 1756 busy_check: 1757 if (!sk_can_busy_loop(sk)) 1758 break; 1759 1760 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1761 } while (!skb_queue_empty_lockless(sk_queue)); 1762 1763 /* sk_queue is empty, reader_queue may contain peeked packets */ 1764 } while (timeo && 1765 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1766 &error, &timeo, 1767 (struct sk_buff *)sk_queue)); 1768 1769 *err = error; 1770 return NULL; 1771 } 1772 EXPORT_SYMBOL(__skb_recv_udp); 1773 1774 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) 1775 { 1776 struct sk_buff *skb; 1777 int err; 1778 1779 try_again: 1780 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1781 if (!skb) 1782 return err; 1783 1784 if (udp_lib_checksum_complete(skb)) { 1785 int is_udplite = IS_UDPLITE(sk); 1786 struct net *net = sock_net(sk); 1787 1788 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite); 1789 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite); 1790 atomic_inc(&sk->sk_drops); 1791 kfree_skb(skb); 1792 goto try_again; 1793 } 1794 1795 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); 1796 return recv_actor(sk, skb); 1797 } 1798 EXPORT_SYMBOL(udp_read_skb); 1799 1800 /* 1801 * This should be easy, if there is something there we 1802 * return it, otherwise we block. 1803 */ 1804 1805 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 1806 int *addr_len) 1807 { 1808 struct inet_sock *inet = inet_sk(sk); 1809 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1810 struct sk_buff *skb; 1811 unsigned int ulen, copied; 1812 int off, err, peeking = flags & MSG_PEEK; 1813 int is_udplite = IS_UDPLITE(sk); 1814 bool checksum_valid = false; 1815 1816 if (flags & MSG_ERRQUEUE) 1817 return ip_recv_error(sk, msg, len, addr_len); 1818 1819 try_again: 1820 off = sk_peek_offset(sk, flags); 1821 skb = __skb_recv_udp(sk, flags, &off, &err); 1822 if (!skb) 1823 return err; 1824 1825 ulen = udp_skb_len(skb); 1826 copied = len; 1827 if (copied > ulen - off) 1828 copied = ulen - off; 1829 else if (copied < ulen) 1830 msg->msg_flags |= MSG_TRUNC; 1831 1832 /* 1833 * If checksum is needed at all, try to do it while copying the 1834 * data. If the data is truncated, or if we only want a partial 1835 * coverage checksum (UDP-Lite), do it before the copy. 1836 */ 1837 1838 if (copied < ulen || peeking || 1839 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1840 checksum_valid = udp_skb_csum_unnecessary(skb) || 1841 !__udp_lib_checksum_complete(skb); 1842 if (!checksum_valid) 1843 goto csum_copy_err; 1844 } 1845 1846 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1847 if (udp_skb_is_linear(skb)) 1848 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1849 else 1850 err = skb_copy_datagram_msg(skb, off, msg, copied); 1851 } else { 1852 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1853 1854 if (err == -EINVAL) 1855 goto csum_copy_err; 1856 } 1857 1858 if (unlikely(err)) { 1859 if (!peeking) { 1860 atomic_inc(&sk->sk_drops); 1861 UDP_INC_STATS(sock_net(sk), 1862 UDP_MIB_INERRORS, is_udplite); 1863 } 1864 kfree_skb(skb); 1865 return err; 1866 } 1867 1868 if (!peeking) 1869 UDP_INC_STATS(sock_net(sk), 1870 UDP_MIB_INDATAGRAMS, is_udplite); 1871 1872 sock_recv_cmsgs(msg, sk, skb); 1873 1874 /* Copy the address. */ 1875 if (sin) { 1876 sin->sin_family = AF_INET; 1877 sin->sin_port = udp_hdr(skb)->source; 1878 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1879 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1880 *addr_len = sizeof(*sin); 1881 1882 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1883 (struct sockaddr *)sin); 1884 } 1885 1886 if (udp_sk(sk)->gro_enabled) 1887 udp_cmsg_recv(msg, sk, skb); 1888 1889 if (inet->cmsg_flags) 1890 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1891 1892 err = copied; 1893 if (flags & MSG_TRUNC) 1894 err = ulen; 1895 1896 skb_consume_udp(sk, skb, peeking ? -err : err); 1897 return err; 1898 1899 csum_copy_err: 1900 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1901 udp_skb_destructor)) { 1902 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1903 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1904 } 1905 kfree_skb(skb); 1906 1907 /* starting over for a new packet, but check if we need to yield */ 1908 cond_resched(); 1909 msg->msg_flags &= ~MSG_TRUNC; 1910 goto try_again; 1911 } 1912 1913 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1914 { 1915 /* This check is replicated from __ip4_datagram_connect() and 1916 * intended to prevent BPF program called below from accessing bytes 1917 * that are out of the bound specified by user in addr_len. 1918 */ 1919 if (addr_len < sizeof(struct sockaddr_in)) 1920 return -EINVAL; 1921 1922 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1923 } 1924 EXPORT_SYMBOL(udp_pre_connect); 1925 1926 int __udp_disconnect(struct sock *sk, int flags) 1927 { 1928 struct inet_sock *inet = inet_sk(sk); 1929 /* 1930 * 1003.1g - break association. 1931 */ 1932 1933 sk->sk_state = TCP_CLOSE; 1934 inet->inet_daddr = 0; 1935 inet->inet_dport = 0; 1936 sock_rps_reset_rxhash(sk); 1937 sk->sk_bound_dev_if = 0; 1938 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1939 inet_reset_saddr(sk); 1940 if (sk->sk_prot->rehash && 1941 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1942 sk->sk_prot->rehash(sk); 1943 } 1944 1945 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1946 sk->sk_prot->unhash(sk); 1947 inet->inet_sport = 0; 1948 } 1949 sk_dst_reset(sk); 1950 return 0; 1951 } 1952 EXPORT_SYMBOL(__udp_disconnect); 1953 1954 int udp_disconnect(struct sock *sk, int flags) 1955 { 1956 lock_sock(sk); 1957 __udp_disconnect(sk, flags); 1958 release_sock(sk); 1959 return 0; 1960 } 1961 EXPORT_SYMBOL(udp_disconnect); 1962 1963 void udp_lib_unhash(struct sock *sk) 1964 { 1965 if (sk_hashed(sk)) { 1966 struct udp_table *udptable = udp_get_table_prot(sk); 1967 struct udp_hslot *hslot, *hslot2; 1968 1969 hslot = udp_hashslot(udptable, sock_net(sk), 1970 udp_sk(sk)->udp_port_hash); 1971 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1972 1973 spin_lock_bh(&hslot->lock); 1974 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1975 reuseport_detach_sock(sk); 1976 if (sk_del_node_init_rcu(sk)) { 1977 hslot->count--; 1978 inet_sk(sk)->inet_num = 0; 1979 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1980 1981 spin_lock(&hslot2->lock); 1982 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1983 hslot2->count--; 1984 spin_unlock(&hslot2->lock); 1985 } 1986 spin_unlock_bh(&hslot->lock); 1987 } 1988 } 1989 EXPORT_SYMBOL(udp_lib_unhash); 1990 1991 /* 1992 * inet_rcv_saddr was changed, we must rehash secondary hash 1993 */ 1994 void udp_lib_rehash(struct sock *sk, u16 newhash) 1995 { 1996 if (sk_hashed(sk)) { 1997 struct udp_table *udptable = udp_get_table_prot(sk); 1998 struct udp_hslot *hslot, *hslot2, *nhslot2; 1999 2000 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2001 nhslot2 = udp_hashslot2(udptable, newhash); 2002 udp_sk(sk)->udp_portaddr_hash = newhash; 2003 2004 if (hslot2 != nhslot2 || 2005 rcu_access_pointer(sk->sk_reuseport_cb)) { 2006 hslot = udp_hashslot(udptable, sock_net(sk), 2007 udp_sk(sk)->udp_port_hash); 2008 /* we must lock primary chain too */ 2009 spin_lock_bh(&hslot->lock); 2010 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2011 reuseport_detach_sock(sk); 2012 2013 if (hslot2 != nhslot2) { 2014 spin_lock(&hslot2->lock); 2015 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2016 hslot2->count--; 2017 spin_unlock(&hslot2->lock); 2018 2019 spin_lock(&nhslot2->lock); 2020 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2021 &nhslot2->head); 2022 nhslot2->count++; 2023 spin_unlock(&nhslot2->lock); 2024 } 2025 2026 spin_unlock_bh(&hslot->lock); 2027 } 2028 } 2029 } 2030 EXPORT_SYMBOL(udp_lib_rehash); 2031 2032 void udp_v4_rehash(struct sock *sk) 2033 { 2034 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2035 inet_sk(sk)->inet_rcv_saddr, 2036 inet_sk(sk)->inet_num); 2037 udp_lib_rehash(sk, new_hash); 2038 } 2039 2040 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2041 { 2042 int rc; 2043 2044 if (inet_sk(sk)->inet_daddr) { 2045 sock_rps_save_rxhash(sk, skb); 2046 sk_mark_napi_id(sk, skb); 2047 sk_incoming_cpu_update(sk); 2048 } else { 2049 sk_mark_napi_id_once(sk, skb); 2050 } 2051 2052 rc = __udp_enqueue_schedule_skb(sk, skb); 2053 if (rc < 0) { 2054 int is_udplite = IS_UDPLITE(sk); 2055 int drop_reason; 2056 2057 /* Note that an ENOMEM error is charged twice */ 2058 if (rc == -ENOMEM) { 2059 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2060 is_udplite); 2061 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 2062 } else { 2063 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2064 is_udplite); 2065 drop_reason = SKB_DROP_REASON_PROTO_MEM; 2066 } 2067 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2068 kfree_skb_reason(skb, drop_reason); 2069 trace_udp_fail_queue_rcv_skb(rc, sk); 2070 return -1; 2071 } 2072 2073 return 0; 2074 } 2075 2076 /* returns: 2077 * -1: error 2078 * 0: success 2079 * >0: "udp encap" protocol resubmission 2080 * 2081 * Note that in the success and error cases, the skb is assumed to 2082 * have either been requeued or freed. 2083 */ 2084 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2085 { 2086 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2087 struct udp_sock *up = udp_sk(sk); 2088 int is_udplite = IS_UDPLITE(sk); 2089 2090 /* 2091 * Charge it to the socket, dropping if the queue is full. 2092 */ 2093 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 2094 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2095 goto drop; 2096 } 2097 nf_reset_ct(skb); 2098 2099 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 2100 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2101 2102 /* 2103 * This is an encapsulation socket so pass the skb to 2104 * the socket's udp_encap_rcv() hook. Otherwise, just 2105 * fall through and pass this up the UDP socket. 2106 * up->encap_rcv() returns the following value: 2107 * =0 if skb was successfully passed to the encap 2108 * handler or was discarded by it. 2109 * >0 if skb should be passed on to UDP. 2110 * <0 if skb should be resubmitted as proto -N 2111 */ 2112 2113 /* if we're overly short, let UDP handle it */ 2114 encap_rcv = READ_ONCE(up->encap_rcv); 2115 if (encap_rcv) { 2116 int ret; 2117 2118 /* Verify checksum before giving to encap */ 2119 if (udp_lib_checksum_complete(skb)) 2120 goto csum_error; 2121 2122 ret = encap_rcv(sk, skb); 2123 if (ret <= 0) { 2124 __UDP_INC_STATS(sock_net(sk), 2125 UDP_MIB_INDATAGRAMS, 2126 is_udplite); 2127 return -ret; 2128 } 2129 } 2130 2131 /* FALLTHROUGH -- it's a UDP Packet */ 2132 } 2133 2134 /* 2135 * UDP-Lite specific tests, ignored on UDP sockets 2136 */ 2137 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2138 2139 /* 2140 * MIB statistics other than incrementing the error count are 2141 * disabled for the following two types of errors: these depend 2142 * on the application settings, not on the functioning of the 2143 * protocol stack as such. 2144 * 2145 * RFC 3828 here recommends (sec 3.3): "There should also be a 2146 * way ... to ... at least let the receiving application block 2147 * delivery of packets with coverage values less than a value 2148 * provided by the application." 2149 */ 2150 if (up->pcrlen == 0) { /* full coverage was set */ 2151 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2152 UDP_SKB_CB(skb)->cscov, skb->len); 2153 goto drop; 2154 } 2155 /* The next case involves violating the min. coverage requested 2156 * by the receiver. This is subtle: if receiver wants x and x is 2157 * greater than the buffersize/MTU then receiver will complain 2158 * that it wants x while sender emits packets of smaller size y. 2159 * Therefore the above ...()->partial_cov statement is essential. 2160 */ 2161 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2162 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2163 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2164 goto drop; 2165 } 2166 } 2167 2168 prefetch(&sk->sk_rmem_alloc); 2169 if (rcu_access_pointer(sk->sk_filter) && 2170 udp_lib_checksum_complete(skb)) 2171 goto csum_error; 2172 2173 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 2174 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 2175 goto drop; 2176 } 2177 2178 udp_csum_pull_header(skb); 2179 2180 ipv4_pktinfo_prepare(sk, skb); 2181 return __udp_queue_rcv_skb(sk, skb); 2182 2183 csum_error: 2184 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2185 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2186 drop: 2187 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2188 atomic_inc(&sk->sk_drops); 2189 kfree_skb_reason(skb, drop_reason); 2190 return -1; 2191 } 2192 2193 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2194 { 2195 struct sk_buff *next, *segs; 2196 int ret; 2197 2198 if (likely(!udp_unexpected_gso(sk, skb))) 2199 return udp_queue_rcv_one_skb(sk, skb); 2200 2201 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2202 __skb_push(skb, -skb_mac_offset(skb)); 2203 segs = udp_rcv_segment(sk, skb, true); 2204 skb_list_walk_safe(segs, skb, next) { 2205 __skb_pull(skb, skb_transport_offset(skb)); 2206 2207 udp_post_segment_fix_csum(skb); 2208 ret = udp_queue_rcv_one_skb(sk, skb); 2209 if (ret > 0) 2210 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2211 } 2212 return 0; 2213 } 2214 2215 /* For TCP sockets, sk_rx_dst is protected by socket lock 2216 * For UDP, we use xchg() to guard against concurrent changes. 2217 */ 2218 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2219 { 2220 struct dst_entry *old; 2221 2222 if (dst_hold_safe(dst)) { 2223 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); 2224 dst_release(old); 2225 return old != dst; 2226 } 2227 return false; 2228 } 2229 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2230 2231 /* 2232 * Multicasts and broadcasts go to each listener. 2233 * 2234 * Note: called only from the BH handler context. 2235 */ 2236 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2237 struct udphdr *uh, 2238 __be32 saddr, __be32 daddr, 2239 struct udp_table *udptable, 2240 int proto) 2241 { 2242 struct sock *sk, *first = NULL; 2243 unsigned short hnum = ntohs(uh->dest); 2244 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2245 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2246 unsigned int offset = offsetof(typeof(*sk), sk_node); 2247 int dif = skb->dev->ifindex; 2248 int sdif = inet_sdif(skb); 2249 struct hlist_node *node; 2250 struct sk_buff *nskb; 2251 2252 if (use_hash2) { 2253 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2254 udptable->mask; 2255 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2256 start_lookup: 2257 hslot = &udptable->hash2[hash2]; 2258 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2259 } 2260 2261 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2262 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2263 uh->source, saddr, dif, sdif, hnum)) 2264 continue; 2265 2266 if (!first) { 2267 first = sk; 2268 continue; 2269 } 2270 nskb = skb_clone(skb, GFP_ATOMIC); 2271 2272 if (unlikely(!nskb)) { 2273 atomic_inc(&sk->sk_drops); 2274 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2275 IS_UDPLITE(sk)); 2276 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2277 IS_UDPLITE(sk)); 2278 continue; 2279 } 2280 if (udp_queue_rcv_skb(sk, nskb) > 0) 2281 consume_skb(nskb); 2282 } 2283 2284 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2285 if (use_hash2 && hash2 != hash2_any) { 2286 hash2 = hash2_any; 2287 goto start_lookup; 2288 } 2289 2290 if (first) { 2291 if (udp_queue_rcv_skb(first, skb) > 0) 2292 consume_skb(skb); 2293 } else { 2294 kfree_skb(skb); 2295 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2296 proto == IPPROTO_UDPLITE); 2297 } 2298 return 0; 2299 } 2300 2301 /* Initialize UDP checksum. If exited with zero value (success), 2302 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2303 * Otherwise, csum completion requires checksumming packet body, 2304 * including udp header and folding it to skb->csum. 2305 */ 2306 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2307 int proto) 2308 { 2309 int err; 2310 2311 UDP_SKB_CB(skb)->partial_cov = 0; 2312 UDP_SKB_CB(skb)->cscov = skb->len; 2313 2314 if (proto == IPPROTO_UDPLITE) { 2315 err = udplite_checksum_init(skb, uh); 2316 if (err) 2317 return err; 2318 2319 if (UDP_SKB_CB(skb)->partial_cov) { 2320 skb->csum = inet_compute_pseudo(skb, proto); 2321 return 0; 2322 } 2323 } 2324 2325 /* Note, we are only interested in != 0 or == 0, thus the 2326 * force to int. 2327 */ 2328 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2329 inet_compute_pseudo); 2330 if (err) 2331 return err; 2332 2333 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2334 /* If SW calculated the value, we know it's bad */ 2335 if (skb->csum_complete_sw) 2336 return 1; 2337 2338 /* HW says the value is bad. Let's validate that. 2339 * skb->csum is no longer the full packet checksum, 2340 * so don't treat it as such. 2341 */ 2342 skb_checksum_complete_unset(skb); 2343 } 2344 2345 return 0; 2346 } 2347 2348 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2349 * return code conversion for ip layer consumption 2350 */ 2351 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2352 struct udphdr *uh) 2353 { 2354 int ret; 2355 2356 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2357 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2358 2359 ret = udp_queue_rcv_skb(sk, skb); 2360 2361 /* a return value > 0 means to resubmit the input, but 2362 * it wants the return to be -protocol, or 0 2363 */ 2364 if (ret > 0) 2365 return -ret; 2366 return 0; 2367 } 2368 2369 /* 2370 * All we need to do is get the socket, and then do a checksum. 2371 */ 2372 2373 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2374 int proto) 2375 { 2376 struct sock *sk; 2377 struct udphdr *uh; 2378 unsigned short ulen; 2379 struct rtable *rt = skb_rtable(skb); 2380 __be32 saddr, daddr; 2381 struct net *net = dev_net(skb->dev); 2382 bool refcounted; 2383 int drop_reason; 2384 2385 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2386 2387 /* 2388 * Validate the packet. 2389 */ 2390 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2391 goto drop; /* No space for header. */ 2392 2393 uh = udp_hdr(skb); 2394 ulen = ntohs(uh->len); 2395 saddr = ip_hdr(skb)->saddr; 2396 daddr = ip_hdr(skb)->daddr; 2397 2398 if (ulen > skb->len) 2399 goto short_packet; 2400 2401 if (proto == IPPROTO_UDP) { 2402 /* UDP validates ulen. */ 2403 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2404 goto short_packet; 2405 uh = udp_hdr(skb); 2406 } 2407 2408 if (udp4_csum_init(skb, uh, proto)) 2409 goto csum_error; 2410 2411 sk = skb_steal_sock(skb, &refcounted); 2412 if (sk) { 2413 struct dst_entry *dst = skb_dst(skb); 2414 int ret; 2415 2416 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2417 udp_sk_rx_dst_set(sk, dst); 2418 2419 ret = udp_unicast_rcv_skb(sk, skb, uh); 2420 if (refcounted) 2421 sock_put(sk); 2422 return ret; 2423 } 2424 2425 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2426 return __udp4_lib_mcast_deliver(net, skb, uh, 2427 saddr, daddr, udptable, proto); 2428 2429 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2430 if (sk) 2431 return udp_unicast_rcv_skb(sk, skb, uh); 2432 2433 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2434 goto drop; 2435 nf_reset_ct(skb); 2436 2437 /* No socket. Drop packet silently, if checksum is wrong */ 2438 if (udp_lib_checksum_complete(skb)) 2439 goto csum_error; 2440 2441 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2442 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2443 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2444 2445 /* 2446 * Hmm. We got an UDP packet to a port to which we 2447 * don't wanna listen. Ignore it. 2448 */ 2449 kfree_skb_reason(skb, drop_reason); 2450 return 0; 2451 2452 short_packet: 2453 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2454 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2455 proto == IPPROTO_UDPLITE ? "Lite" : "", 2456 &saddr, ntohs(uh->source), 2457 ulen, skb->len, 2458 &daddr, ntohs(uh->dest)); 2459 goto drop; 2460 2461 csum_error: 2462 /* 2463 * RFC1122: OK. Discards the bad packet silently (as far as 2464 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2465 */ 2466 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2467 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2468 proto == IPPROTO_UDPLITE ? "Lite" : "", 2469 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2470 ulen); 2471 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2472 drop: 2473 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2474 kfree_skb_reason(skb, drop_reason); 2475 return 0; 2476 } 2477 2478 /* We can only early demux multicast if there is a single matching socket. 2479 * If more than one socket found returns NULL 2480 */ 2481 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2482 __be16 loc_port, __be32 loc_addr, 2483 __be16 rmt_port, __be32 rmt_addr, 2484 int dif, int sdif) 2485 { 2486 struct udp_table *udptable = net->ipv4.udp_table; 2487 unsigned short hnum = ntohs(loc_port); 2488 struct sock *sk, *result; 2489 struct udp_hslot *hslot; 2490 unsigned int slot; 2491 2492 slot = udp_hashfn(net, hnum, udptable->mask); 2493 hslot = &udptable->hash[slot]; 2494 2495 /* Do not bother scanning a too big list */ 2496 if (hslot->count > 10) 2497 return NULL; 2498 2499 result = NULL; 2500 sk_for_each_rcu(sk, &hslot->head) { 2501 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2502 rmt_port, rmt_addr, dif, sdif, hnum)) { 2503 if (result) 2504 return NULL; 2505 result = sk; 2506 } 2507 } 2508 2509 return result; 2510 } 2511 2512 /* For unicast we should only early demux connected sockets or we can 2513 * break forwarding setups. The chains here can be long so only check 2514 * if the first socket is an exact match and if not move on. 2515 */ 2516 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2517 __be16 loc_port, __be32 loc_addr, 2518 __be16 rmt_port, __be32 rmt_addr, 2519 int dif, int sdif) 2520 { 2521 struct udp_table *udptable = net->ipv4.udp_table; 2522 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2523 unsigned short hnum = ntohs(loc_port); 2524 unsigned int hash2, slot2; 2525 struct udp_hslot *hslot2; 2526 __portpair ports; 2527 struct sock *sk; 2528 2529 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2530 slot2 = hash2 & udptable->mask; 2531 hslot2 = &udptable->hash2[slot2]; 2532 ports = INET_COMBINED_PORTS(rmt_port, hnum); 2533 2534 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2535 if (inet_match(net, sk, acookie, ports, dif, sdif)) 2536 return sk; 2537 /* Only check first socket in chain */ 2538 break; 2539 } 2540 return NULL; 2541 } 2542 2543 int udp_v4_early_demux(struct sk_buff *skb) 2544 { 2545 struct net *net = dev_net(skb->dev); 2546 struct in_device *in_dev = NULL; 2547 const struct iphdr *iph; 2548 const struct udphdr *uh; 2549 struct sock *sk = NULL; 2550 struct dst_entry *dst; 2551 int dif = skb->dev->ifindex; 2552 int sdif = inet_sdif(skb); 2553 int ours; 2554 2555 /* validate the packet */ 2556 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2557 return 0; 2558 2559 iph = ip_hdr(skb); 2560 uh = udp_hdr(skb); 2561 2562 if (skb->pkt_type == PACKET_MULTICAST) { 2563 in_dev = __in_dev_get_rcu(skb->dev); 2564 2565 if (!in_dev) 2566 return 0; 2567 2568 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2569 iph->protocol); 2570 if (!ours) 2571 return 0; 2572 2573 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2574 uh->source, iph->saddr, 2575 dif, sdif); 2576 } else if (skb->pkt_type == PACKET_HOST) { 2577 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2578 uh->source, iph->saddr, dif, sdif); 2579 } 2580 2581 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2582 return 0; 2583 2584 skb->sk = sk; 2585 skb->destructor = sock_efree; 2586 dst = rcu_dereference(sk->sk_rx_dst); 2587 2588 if (dst) 2589 dst = dst_check(dst, 0); 2590 if (dst) { 2591 u32 itag = 0; 2592 2593 /* set noref for now. 2594 * any place which wants to hold dst has to call 2595 * dst_hold_safe() 2596 */ 2597 skb_dst_set_noref(skb, dst); 2598 2599 /* for unconnected multicast sockets we need to validate 2600 * the source on each packet 2601 */ 2602 if (!inet_sk(sk)->inet_daddr && in_dev) 2603 return ip_mc_validate_source(skb, iph->daddr, 2604 iph->saddr, 2605 iph->tos & IPTOS_RT_MASK, 2606 skb->dev, in_dev, &itag); 2607 } 2608 return 0; 2609 } 2610 2611 int udp_rcv(struct sk_buff *skb) 2612 { 2613 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 2614 } 2615 2616 void udp_destroy_sock(struct sock *sk) 2617 { 2618 struct udp_sock *up = udp_sk(sk); 2619 bool slow = lock_sock_fast(sk); 2620 2621 /* protects from races with udp_abort() */ 2622 sock_set_flag(sk, SOCK_DEAD); 2623 udp_flush_pending_frames(sk); 2624 unlock_sock_fast(sk, slow); 2625 if (static_branch_unlikely(&udp_encap_needed_key)) { 2626 if (up->encap_type) { 2627 void (*encap_destroy)(struct sock *sk); 2628 encap_destroy = READ_ONCE(up->encap_destroy); 2629 if (encap_destroy) 2630 encap_destroy(sk); 2631 } 2632 if (up->encap_enabled) 2633 static_branch_dec(&udp_encap_needed_key); 2634 } 2635 } 2636 2637 /* 2638 * Socket option code for UDP 2639 */ 2640 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2641 sockptr_t optval, unsigned int optlen, 2642 int (*push_pending_frames)(struct sock *)) 2643 { 2644 struct udp_sock *up = udp_sk(sk); 2645 int val, valbool; 2646 int err = 0; 2647 int is_udplite = IS_UDPLITE(sk); 2648 2649 if (level == SOL_SOCKET) { 2650 err = sk_setsockopt(sk, level, optname, optval, optlen); 2651 2652 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) { 2653 sockopt_lock_sock(sk); 2654 /* paired with READ_ONCE in udp_rmem_release() */ 2655 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2); 2656 sockopt_release_sock(sk); 2657 } 2658 return err; 2659 } 2660 2661 if (optlen < sizeof(int)) 2662 return -EINVAL; 2663 2664 if (copy_from_sockptr(&val, optval, sizeof(val))) 2665 return -EFAULT; 2666 2667 valbool = val ? 1 : 0; 2668 2669 switch (optname) { 2670 case UDP_CORK: 2671 if (val != 0) { 2672 WRITE_ONCE(up->corkflag, 1); 2673 } else { 2674 WRITE_ONCE(up->corkflag, 0); 2675 lock_sock(sk); 2676 push_pending_frames(sk); 2677 release_sock(sk); 2678 } 2679 break; 2680 2681 case UDP_ENCAP: 2682 switch (val) { 2683 case 0: 2684 #ifdef CONFIG_XFRM 2685 case UDP_ENCAP_ESPINUDP: 2686 case UDP_ENCAP_ESPINUDP_NON_IKE: 2687 #if IS_ENABLED(CONFIG_IPV6) 2688 if (sk->sk_family == AF_INET6) 2689 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; 2690 else 2691 #endif 2692 up->encap_rcv = xfrm4_udp_encap_rcv; 2693 #endif 2694 fallthrough; 2695 case UDP_ENCAP_L2TPINUDP: 2696 up->encap_type = val; 2697 lock_sock(sk); 2698 udp_tunnel_encap_enable(sk->sk_socket); 2699 release_sock(sk); 2700 break; 2701 default: 2702 err = -ENOPROTOOPT; 2703 break; 2704 } 2705 break; 2706 2707 case UDP_NO_CHECK6_TX: 2708 up->no_check6_tx = valbool; 2709 break; 2710 2711 case UDP_NO_CHECK6_RX: 2712 up->no_check6_rx = valbool; 2713 break; 2714 2715 case UDP_SEGMENT: 2716 if (val < 0 || val > USHRT_MAX) 2717 return -EINVAL; 2718 WRITE_ONCE(up->gso_size, val); 2719 break; 2720 2721 case UDP_GRO: 2722 lock_sock(sk); 2723 2724 /* when enabling GRO, accept the related GSO packet type */ 2725 if (valbool) 2726 udp_tunnel_encap_enable(sk->sk_socket); 2727 up->gro_enabled = valbool; 2728 up->accept_udp_l4 = valbool; 2729 release_sock(sk); 2730 break; 2731 2732 /* 2733 * UDP-Lite's partial checksum coverage (RFC 3828). 2734 */ 2735 /* The sender sets actual checksum coverage length via this option. 2736 * The case coverage > packet length is handled by send module. */ 2737 case UDPLITE_SEND_CSCOV: 2738 if (!is_udplite) /* Disable the option on UDP sockets */ 2739 return -ENOPROTOOPT; 2740 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2741 val = 8; 2742 else if (val > USHRT_MAX) 2743 val = USHRT_MAX; 2744 up->pcslen = val; 2745 up->pcflag |= UDPLITE_SEND_CC; 2746 break; 2747 2748 /* The receiver specifies a minimum checksum coverage value. To make 2749 * sense, this should be set to at least 8 (as done below). If zero is 2750 * used, this again means full checksum coverage. */ 2751 case UDPLITE_RECV_CSCOV: 2752 if (!is_udplite) /* Disable the option on UDP sockets */ 2753 return -ENOPROTOOPT; 2754 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2755 val = 8; 2756 else if (val > USHRT_MAX) 2757 val = USHRT_MAX; 2758 up->pcrlen = val; 2759 up->pcflag |= UDPLITE_RECV_CC; 2760 break; 2761 2762 default: 2763 err = -ENOPROTOOPT; 2764 break; 2765 } 2766 2767 return err; 2768 } 2769 EXPORT_SYMBOL(udp_lib_setsockopt); 2770 2771 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2772 unsigned int optlen) 2773 { 2774 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 2775 return udp_lib_setsockopt(sk, level, optname, 2776 optval, optlen, 2777 udp_push_pending_frames); 2778 return ip_setsockopt(sk, level, optname, optval, optlen); 2779 } 2780 2781 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2782 char __user *optval, int __user *optlen) 2783 { 2784 struct udp_sock *up = udp_sk(sk); 2785 int val, len; 2786 2787 if (get_user(len, optlen)) 2788 return -EFAULT; 2789 2790 len = min_t(unsigned int, len, sizeof(int)); 2791 2792 if (len < 0) 2793 return -EINVAL; 2794 2795 switch (optname) { 2796 case UDP_CORK: 2797 val = READ_ONCE(up->corkflag); 2798 break; 2799 2800 case UDP_ENCAP: 2801 val = up->encap_type; 2802 break; 2803 2804 case UDP_NO_CHECK6_TX: 2805 val = up->no_check6_tx; 2806 break; 2807 2808 case UDP_NO_CHECK6_RX: 2809 val = up->no_check6_rx; 2810 break; 2811 2812 case UDP_SEGMENT: 2813 val = READ_ONCE(up->gso_size); 2814 break; 2815 2816 case UDP_GRO: 2817 val = up->gro_enabled; 2818 break; 2819 2820 /* The following two cannot be changed on UDP sockets, the return is 2821 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2822 case UDPLITE_SEND_CSCOV: 2823 val = up->pcslen; 2824 break; 2825 2826 case UDPLITE_RECV_CSCOV: 2827 val = up->pcrlen; 2828 break; 2829 2830 default: 2831 return -ENOPROTOOPT; 2832 } 2833 2834 if (put_user(len, optlen)) 2835 return -EFAULT; 2836 if (copy_to_user(optval, &val, len)) 2837 return -EFAULT; 2838 return 0; 2839 } 2840 EXPORT_SYMBOL(udp_lib_getsockopt); 2841 2842 int udp_getsockopt(struct sock *sk, int level, int optname, 2843 char __user *optval, int __user *optlen) 2844 { 2845 if (level == SOL_UDP || level == SOL_UDPLITE) 2846 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2847 return ip_getsockopt(sk, level, optname, optval, optlen); 2848 } 2849 2850 /** 2851 * udp_poll - wait for a UDP event. 2852 * @file: - file struct 2853 * @sock: - socket 2854 * @wait: - poll table 2855 * 2856 * This is same as datagram poll, except for the special case of 2857 * blocking sockets. If application is using a blocking fd 2858 * and a packet with checksum error is in the queue; 2859 * then it could get return from select indicating data available 2860 * but then block when reading it. Add special case code 2861 * to work around these arguably broken applications. 2862 */ 2863 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2864 { 2865 __poll_t mask = datagram_poll(file, sock, wait); 2866 struct sock *sk = sock->sk; 2867 2868 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2869 mask |= EPOLLIN | EPOLLRDNORM; 2870 2871 /* Check for false positives due to checksum errors */ 2872 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2873 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2874 mask &= ~(EPOLLIN | EPOLLRDNORM); 2875 2876 /* psock ingress_msg queue should not contain any bad checksum frames */ 2877 if (sk_is_readable(sk)) 2878 mask |= EPOLLIN | EPOLLRDNORM; 2879 return mask; 2880 2881 } 2882 EXPORT_SYMBOL(udp_poll); 2883 2884 int udp_abort(struct sock *sk, int err) 2885 { 2886 lock_sock(sk); 2887 2888 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2889 * with close() 2890 */ 2891 if (sock_flag(sk, SOCK_DEAD)) 2892 goto out; 2893 2894 sk->sk_err = err; 2895 sk_error_report(sk); 2896 __udp_disconnect(sk, 0); 2897 2898 out: 2899 release_sock(sk); 2900 2901 return 0; 2902 } 2903 EXPORT_SYMBOL_GPL(udp_abort); 2904 2905 struct proto udp_prot = { 2906 .name = "UDP", 2907 .owner = THIS_MODULE, 2908 .close = udp_lib_close, 2909 .pre_connect = udp_pre_connect, 2910 .connect = ip4_datagram_connect, 2911 .disconnect = udp_disconnect, 2912 .ioctl = udp_ioctl, 2913 .init = udp_init_sock, 2914 .destroy = udp_destroy_sock, 2915 .setsockopt = udp_setsockopt, 2916 .getsockopt = udp_getsockopt, 2917 .sendmsg = udp_sendmsg, 2918 .recvmsg = udp_recvmsg, 2919 .sendpage = udp_sendpage, 2920 .release_cb = ip4_datagram_release_cb, 2921 .hash = udp_lib_hash, 2922 .unhash = udp_lib_unhash, 2923 .rehash = udp_v4_rehash, 2924 .get_port = udp_v4_get_port, 2925 .put_port = udp_lib_unhash, 2926 #ifdef CONFIG_BPF_SYSCALL 2927 .psock_update_sk_prot = udp_bpf_update_proto, 2928 #endif 2929 .memory_allocated = &udp_memory_allocated, 2930 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 2931 2932 .sysctl_mem = sysctl_udp_mem, 2933 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2934 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2935 .obj_size = sizeof(struct udp_sock), 2936 .h.udp_table = NULL, 2937 .diag_destroy = udp_abort, 2938 }; 2939 EXPORT_SYMBOL(udp_prot); 2940 2941 /* ------------------------------------------------------------------------ */ 2942 #ifdef CONFIG_PROC_FS 2943 2944 static struct udp_table *udp_get_table_afinfo(struct udp_seq_afinfo *afinfo, 2945 struct net *net) 2946 { 2947 return afinfo->udp_table ? : net->ipv4.udp_table; 2948 } 2949 2950 static struct sock *udp_get_first(struct seq_file *seq, int start) 2951 { 2952 struct udp_iter_state *state = seq->private; 2953 struct net *net = seq_file_net(seq); 2954 struct udp_seq_afinfo *afinfo; 2955 struct udp_table *udptable; 2956 struct sock *sk; 2957 2958 if (state->bpf_seq_afinfo) 2959 afinfo = state->bpf_seq_afinfo; 2960 else 2961 afinfo = pde_data(file_inode(seq->file)); 2962 2963 udptable = udp_get_table_afinfo(afinfo, net); 2964 2965 for (state->bucket = start; state->bucket <= udptable->mask; 2966 ++state->bucket) { 2967 struct udp_hslot *hslot = &udptable->hash[state->bucket]; 2968 2969 if (hlist_empty(&hslot->head)) 2970 continue; 2971 2972 spin_lock_bh(&hslot->lock); 2973 sk_for_each(sk, &hslot->head) { 2974 if (!net_eq(sock_net(sk), net)) 2975 continue; 2976 if (afinfo->family == AF_UNSPEC || 2977 sk->sk_family == afinfo->family) 2978 goto found; 2979 } 2980 spin_unlock_bh(&hslot->lock); 2981 } 2982 sk = NULL; 2983 found: 2984 return sk; 2985 } 2986 2987 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2988 { 2989 struct udp_iter_state *state = seq->private; 2990 struct net *net = seq_file_net(seq); 2991 struct udp_seq_afinfo *afinfo; 2992 struct udp_table *udptable; 2993 2994 if (state->bpf_seq_afinfo) 2995 afinfo = state->bpf_seq_afinfo; 2996 else 2997 afinfo = pde_data(file_inode(seq->file)); 2998 2999 do { 3000 sk = sk_next(sk); 3001 } while (sk && (!net_eq(sock_net(sk), net) || 3002 (afinfo->family != AF_UNSPEC && 3003 sk->sk_family != afinfo->family))); 3004 3005 if (!sk) { 3006 udptable = udp_get_table_afinfo(afinfo, net); 3007 3008 if (state->bucket <= udptable->mask) 3009 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3010 3011 return udp_get_first(seq, state->bucket + 1); 3012 } 3013 return sk; 3014 } 3015 3016 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3017 { 3018 struct sock *sk = udp_get_first(seq, 0); 3019 3020 if (sk) 3021 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3022 --pos; 3023 return pos ? NULL : sk; 3024 } 3025 3026 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3027 { 3028 struct udp_iter_state *state = seq->private; 3029 state->bucket = MAX_UDP_PORTS; 3030 3031 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3032 } 3033 EXPORT_SYMBOL(udp_seq_start); 3034 3035 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3036 { 3037 struct sock *sk; 3038 3039 if (v == SEQ_START_TOKEN) 3040 sk = udp_get_idx(seq, 0); 3041 else 3042 sk = udp_get_next(seq, v); 3043 3044 ++*pos; 3045 return sk; 3046 } 3047 EXPORT_SYMBOL(udp_seq_next); 3048 3049 void udp_seq_stop(struct seq_file *seq, void *v) 3050 { 3051 struct udp_iter_state *state = seq->private; 3052 struct udp_seq_afinfo *afinfo; 3053 struct udp_table *udptable; 3054 3055 if (state->bpf_seq_afinfo) 3056 afinfo = state->bpf_seq_afinfo; 3057 else 3058 afinfo = pde_data(file_inode(seq->file)); 3059 3060 udptable = udp_get_table_afinfo(afinfo, seq_file_net(seq)); 3061 3062 if (state->bucket <= udptable->mask) 3063 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3064 } 3065 EXPORT_SYMBOL(udp_seq_stop); 3066 3067 /* ------------------------------------------------------------------------ */ 3068 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3069 int bucket) 3070 { 3071 struct inet_sock *inet = inet_sk(sp); 3072 __be32 dest = inet->inet_daddr; 3073 __be32 src = inet->inet_rcv_saddr; 3074 __u16 destp = ntohs(inet->inet_dport); 3075 __u16 srcp = ntohs(inet->inet_sport); 3076 3077 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3078 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3079 bucket, src, srcp, dest, destp, sp->sk_state, 3080 sk_wmem_alloc_get(sp), 3081 udp_rqueue_get(sp), 3082 0, 0L, 0, 3083 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3084 0, sock_i_ino(sp), 3085 refcount_read(&sp->sk_refcnt), sp, 3086 atomic_read(&sp->sk_drops)); 3087 } 3088 3089 int udp4_seq_show(struct seq_file *seq, void *v) 3090 { 3091 seq_setwidth(seq, 127); 3092 if (v == SEQ_START_TOKEN) 3093 seq_puts(seq, " sl local_address rem_address st tx_queue " 3094 "rx_queue tr tm->when retrnsmt uid timeout " 3095 "inode ref pointer drops"); 3096 else { 3097 struct udp_iter_state *state = seq->private; 3098 3099 udp4_format_sock(v, seq, state->bucket); 3100 } 3101 seq_pad(seq, '\n'); 3102 return 0; 3103 } 3104 3105 #ifdef CONFIG_BPF_SYSCALL 3106 struct bpf_iter__udp { 3107 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3108 __bpf_md_ptr(struct udp_sock *, udp_sk); 3109 uid_t uid __aligned(8); 3110 int bucket __aligned(8); 3111 }; 3112 3113 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3114 struct udp_sock *udp_sk, uid_t uid, int bucket) 3115 { 3116 struct bpf_iter__udp ctx; 3117 3118 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3119 ctx.meta = meta; 3120 ctx.udp_sk = udp_sk; 3121 ctx.uid = uid; 3122 ctx.bucket = bucket; 3123 return bpf_iter_run_prog(prog, &ctx); 3124 } 3125 3126 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3127 { 3128 struct udp_iter_state *state = seq->private; 3129 struct bpf_iter_meta meta; 3130 struct bpf_prog *prog; 3131 struct sock *sk = v; 3132 uid_t uid; 3133 3134 if (v == SEQ_START_TOKEN) 3135 return 0; 3136 3137 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3138 meta.seq = seq; 3139 prog = bpf_iter_get_info(&meta, false); 3140 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3141 } 3142 3143 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3144 { 3145 struct bpf_iter_meta meta; 3146 struct bpf_prog *prog; 3147 3148 if (!v) { 3149 meta.seq = seq; 3150 prog = bpf_iter_get_info(&meta, true); 3151 if (prog) 3152 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3153 } 3154 3155 udp_seq_stop(seq, v); 3156 } 3157 3158 static const struct seq_operations bpf_iter_udp_seq_ops = { 3159 .start = udp_seq_start, 3160 .next = udp_seq_next, 3161 .stop = bpf_iter_udp_seq_stop, 3162 .show = bpf_iter_udp_seq_show, 3163 }; 3164 #endif 3165 3166 const struct seq_operations udp_seq_ops = { 3167 .start = udp_seq_start, 3168 .next = udp_seq_next, 3169 .stop = udp_seq_stop, 3170 .show = udp4_seq_show, 3171 }; 3172 EXPORT_SYMBOL(udp_seq_ops); 3173 3174 static struct udp_seq_afinfo udp4_seq_afinfo = { 3175 .family = AF_INET, 3176 .udp_table = NULL, 3177 }; 3178 3179 static int __net_init udp4_proc_init_net(struct net *net) 3180 { 3181 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3182 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3183 return -ENOMEM; 3184 return 0; 3185 } 3186 3187 static void __net_exit udp4_proc_exit_net(struct net *net) 3188 { 3189 remove_proc_entry("udp", net->proc_net); 3190 } 3191 3192 static struct pernet_operations udp4_net_ops = { 3193 .init = udp4_proc_init_net, 3194 .exit = udp4_proc_exit_net, 3195 }; 3196 3197 int __init udp4_proc_init(void) 3198 { 3199 return register_pernet_subsys(&udp4_net_ops); 3200 } 3201 3202 void udp4_proc_exit(void) 3203 { 3204 unregister_pernet_subsys(&udp4_net_ops); 3205 } 3206 #endif /* CONFIG_PROC_FS */ 3207 3208 static __initdata unsigned long uhash_entries; 3209 static int __init set_uhash_entries(char *str) 3210 { 3211 ssize_t ret; 3212 3213 if (!str) 3214 return 0; 3215 3216 ret = kstrtoul(str, 0, &uhash_entries); 3217 if (ret) 3218 return 0; 3219 3220 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3221 uhash_entries = UDP_HTABLE_SIZE_MIN; 3222 return 1; 3223 } 3224 __setup("uhash_entries=", set_uhash_entries); 3225 3226 void __init udp_table_init(struct udp_table *table, const char *name) 3227 { 3228 unsigned int i; 3229 3230 table->hash = alloc_large_system_hash(name, 3231 2 * sizeof(struct udp_hslot), 3232 uhash_entries, 3233 21, /* one slot per 2 MB */ 3234 0, 3235 &table->log, 3236 &table->mask, 3237 UDP_HTABLE_SIZE_MIN, 3238 UDP_HTABLE_SIZE_MAX); 3239 3240 table->hash2 = table->hash + (table->mask + 1); 3241 for (i = 0; i <= table->mask; i++) { 3242 INIT_HLIST_HEAD(&table->hash[i].head); 3243 table->hash[i].count = 0; 3244 spin_lock_init(&table->hash[i].lock); 3245 } 3246 for (i = 0; i <= table->mask; i++) { 3247 INIT_HLIST_HEAD(&table->hash2[i].head); 3248 table->hash2[i].count = 0; 3249 spin_lock_init(&table->hash2[i].lock); 3250 } 3251 } 3252 3253 u32 udp_flow_hashrnd(void) 3254 { 3255 static u32 hashrnd __read_mostly; 3256 3257 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3258 3259 return hashrnd; 3260 } 3261 EXPORT_SYMBOL(udp_flow_hashrnd); 3262 3263 static void __net_init udp_sysctl_init(struct net *net) 3264 { 3265 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3266 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 3267 3268 #ifdef CONFIG_NET_L3_MASTER_DEV 3269 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3270 #endif 3271 } 3272 3273 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries) 3274 { 3275 struct udp_table *udptable; 3276 int i; 3277 3278 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL); 3279 if (!udptable) 3280 goto out; 3281 3282 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot), 3283 GFP_KERNEL_ACCOUNT); 3284 if (!udptable->hash) 3285 goto free_table; 3286 3287 udptable->hash2 = udptable->hash + hash_entries; 3288 udptable->mask = hash_entries - 1; 3289 udptable->log = ilog2(hash_entries); 3290 3291 for (i = 0; i < hash_entries; i++) { 3292 INIT_HLIST_HEAD(&udptable->hash[i].head); 3293 udptable->hash[i].count = 0; 3294 spin_lock_init(&udptable->hash[i].lock); 3295 3296 INIT_HLIST_HEAD(&udptable->hash2[i].head); 3297 udptable->hash2[i].count = 0; 3298 spin_lock_init(&udptable->hash2[i].lock); 3299 } 3300 3301 return udptable; 3302 3303 free_table: 3304 kfree(udptable); 3305 out: 3306 return NULL; 3307 } 3308 3309 static void __net_exit udp_pernet_table_free(struct net *net) 3310 { 3311 struct udp_table *udptable = net->ipv4.udp_table; 3312 3313 if (udptable == &udp_table) 3314 return; 3315 3316 kvfree(udptable->hash); 3317 kfree(udptable); 3318 } 3319 3320 static void __net_init udp_set_table(struct net *net) 3321 { 3322 struct udp_table *udptable; 3323 unsigned int hash_entries; 3324 struct net *old_net; 3325 3326 if (net_eq(net, &init_net)) 3327 goto fallback; 3328 3329 old_net = current->nsproxy->net_ns; 3330 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries); 3331 if (!hash_entries) 3332 goto fallback; 3333 3334 /* Set min to keep the bitmap on stack in udp_lib_get_port() */ 3335 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET) 3336 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET; 3337 else 3338 hash_entries = roundup_pow_of_two(hash_entries); 3339 3340 udptable = udp_pernet_table_alloc(hash_entries); 3341 if (udptable) { 3342 net->ipv4.udp_table = udptable; 3343 } else { 3344 pr_warn("Failed to allocate UDP hash table (entries: %u) " 3345 "for a netns, fallback to the global one\n", 3346 hash_entries); 3347 fallback: 3348 net->ipv4.udp_table = &udp_table; 3349 } 3350 } 3351 3352 static int __net_init udp_pernet_init(struct net *net) 3353 { 3354 udp_sysctl_init(net); 3355 udp_set_table(net); 3356 3357 return 0; 3358 } 3359 3360 static void __net_exit udp_pernet_exit(struct net *net) 3361 { 3362 udp_pernet_table_free(net); 3363 } 3364 3365 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3366 .init = udp_pernet_init, 3367 .exit = udp_pernet_exit, 3368 }; 3369 3370 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3371 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3372 struct udp_sock *udp_sk, uid_t uid, int bucket) 3373 3374 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3375 { 3376 struct udp_iter_state *st = priv_data; 3377 struct udp_seq_afinfo *afinfo; 3378 int ret; 3379 3380 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN); 3381 if (!afinfo) 3382 return -ENOMEM; 3383 3384 afinfo->family = AF_UNSPEC; 3385 afinfo->udp_table = NULL; 3386 st->bpf_seq_afinfo = afinfo; 3387 ret = bpf_iter_init_seq_net(priv_data, aux); 3388 if (ret) 3389 kfree(afinfo); 3390 return ret; 3391 } 3392 3393 static void bpf_iter_fini_udp(void *priv_data) 3394 { 3395 struct udp_iter_state *st = priv_data; 3396 3397 kfree(st->bpf_seq_afinfo); 3398 bpf_iter_fini_seq_net(priv_data); 3399 } 3400 3401 static const struct bpf_iter_seq_info udp_seq_info = { 3402 .seq_ops = &bpf_iter_udp_seq_ops, 3403 .init_seq_private = bpf_iter_init_udp, 3404 .fini_seq_private = bpf_iter_fini_udp, 3405 .seq_priv_size = sizeof(struct udp_iter_state), 3406 }; 3407 3408 static struct bpf_iter_reg udp_reg_info = { 3409 .target = "udp", 3410 .ctx_arg_info_size = 1, 3411 .ctx_arg_info = { 3412 { offsetof(struct bpf_iter__udp, udp_sk), 3413 PTR_TO_BTF_ID_OR_NULL }, 3414 }, 3415 .seq_info = &udp_seq_info, 3416 }; 3417 3418 static void __init bpf_iter_register(void) 3419 { 3420 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3421 if (bpf_iter_reg_target(&udp_reg_info)) 3422 pr_warn("Warning: could not register bpf iterator udp\n"); 3423 } 3424 #endif 3425 3426 void __init udp_init(void) 3427 { 3428 unsigned long limit; 3429 unsigned int i; 3430 3431 udp_table_init(&udp_table, "UDP"); 3432 limit = nr_free_buffer_pages() / 8; 3433 limit = max(limit, 128UL); 3434 sysctl_udp_mem[0] = limit / 4 * 3; 3435 sysctl_udp_mem[1] = limit; 3436 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3437 3438 /* 16 spinlocks per cpu */ 3439 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3440 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3441 GFP_KERNEL); 3442 if (!udp_busylocks) 3443 panic("UDP: failed to alloc udp_busylocks\n"); 3444 for (i = 0; i < (1U << udp_busylocks_log); i++) 3445 spin_lock_init(udp_busylocks + i); 3446 3447 if (register_pernet_subsys(&udp_sysctl_ops)) 3448 panic("UDP: failed to init sysctl parameters.\n"); 3449 3450 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3451 bpf_iter_register(); 3452 #endif 3453 } 3454