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) 133 134 static int udp_lib_lport_inuse(struct net *net, __u16 num, 135 const struct udp_hslot *hslot, 136 unsigned long *bitmap, 137 struct sock *sk, unsigned int log) 138 { 139 struct sock *sk2; 140 kuid_t uid = sock_i_uid(sk); 141 142 sk_for_each(sk2, &hslot->head) { 143 if (net_eq(sock_net(sk2), net) && 144 sk2 != sk && 145 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 146 (!sk2->sk_reuse || !sk->sk_reuse) && 147 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 148 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 149 inet_rcv_saddr_equal(sk, sk2, true)) { 150 if (sk2->sk_reuseport && sk->sk_reuseport && 151 !rcu_access_pointer(sk->sk_reuseport_cb) && 152 uid_eq(uid, sock_i_uid(sk2))) { 153 if (!bitmap) 154 return 0; 155 } else { 156 if (!bitmap) 157 return 1; 158 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 159 bitmap); 160 } 161 } 162 } 163 return 0; 164 } 165 166 /* 167 * Note: we still hold spinlock of primary hash chain, so no other writer 168 * can insert/delete a socket with local_port == num 169 */ 170 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 171 struct udp_hslot *hslot2, 172 struct sock *sk) 173 { 174 struct sock *sk2; 175 kuid_t uid = sock_i_uid(sk); 176 int res = 0; 177 178 spin_lock(&hslot2->lock); 179 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 180 if (net_eq(sock_net(sk2), net) && 181 sk2 != sk && 182 (udp_sk(sk2)->udp_port_hash == num) && 183 (!sk2->sk_reuse || !sk->sk_reuse) && 184 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 185 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 186 inet_rcv_saddr_equal(sk, sk2, true)) { 187 if (sk2->sk_reuseport && sk->sk_reuseport && 188 !rcu_access_pointer(sk->sk_reuseport_cb) && 189 uid_eq(uid, sock_i_uid(sk2))) { 190 res = 0; 191 } else { 192 res = 1; 193 } 194 break; 195 } 196 } 197 spin_unlock(&hslot2->lock); 198 return res; 199 } 200 201 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 202 { 203 struct net *net = sock_net(sk); 204 kuid_t uid = sock_i_uid(sk); 205 struct sock *sk2; 206 207 sk_for_each(sk2, &hslot->head) { 208 if (net_eq(sock_net(sk2), net) && 209 sk2 != sk && 210 sk2->sk_family == sk->sk_family && 211 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 212 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 213 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 214 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 215 inet_rcv_saddr_equal(sk, sk2, false)) { 216 return reuseport_add_sock(sk, sk2, 217 inet_rcv_saddr_any(sk)); 218 } 219 } 220 221 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 222 } 223 224 /** 225 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 226 * 227 * @sk: socket struct in question 228 * @snum: port number to look up 229 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 230 * with NULL address 231 */ 232 int udp_lib_get_port(struct sock *sk, unsigned short snum, 233 unsigned int hash2_nulladdr) 234 { 235 struct udp_hslot *hslot, *hslot2; 236 struct udp_table *udptable = sk->sk_prot->h.udp_table; 237 int error = 1; 238 struct net *net = sock_net(sk); 239 240 if (!snum) { 241 int low, high, remaining; 242 unsigned int rand; 243 unsigned short first, last; 244 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 245 246 inet_get_local_port_range(net, &low, &high); 247 remaining = (high - low) + 1; 248 249 rand = prandom_u32(); 250 first = reciprocal_scale(rand, remaining) + low; 251 /* 252 * force rand to be an odd multiple of UDP_HTABLE_SIZE 253 */ 254 rand = (rand | 1) * (udptable->mask + 1); 255 last = first + udptable->mask + 1; 256 do { 257 hslot = udp_hashslot(udptable, net, first); 258 bitmap_zero(bitmap, PORTS_PER_CHAIN); 259 spin_lock_bh(&hslot->lock); 260 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 261 udptable->log); 262 263 snum = first; 264 /* 265 * Iterate on all possible values of snum for this hash. 266 * Using steps of an odd multiple of UDP_HTABLE_SIZE 267 * give us randomization and full range coverage. 268 */ 269 do { 270 if (low <= snum && snum <= high && 271 !test_bit(snum >> udptable->log, bitmap) && 272 !inet_is_local_reserved_port(net, snum)) 273 goto found; 274 snum += rand; 275 } while (snum != first); 276 spin_unlock_bh(&hslot->lock); 277 cond_resched(); 278 } while (++first != last); 279 goto fail; 280 } else { 281 hslot = udp_hashslot(udptable, net, snum); 282 spin_lock_bh(&hslot->lock); 283 if (hslot->count > 10) { 284 int exist; 285 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 286 287 slot2 &= udptable->mask; 288 hash2_nulladdr &= udptable->mask; 289 290 hslot2 = udp_hashslot2(udptable, slot2); 291 if (hslot->count < hslot2->count) 292 goto scan_primary_hash; 293 294 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 295 if (!exist && (hash2_nulladdr != slot2)) { 296 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 297 exist = udp_lib_lport_inuse2(net, snum, hslot2, 298 sk); 299 } 300 if (exist) 301 goto fail_unlock; 302 else 303 goto found; 304 } 305 scan_primary_hash: 306 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 307 goto fail_unlock; 308 } 309 found: 310 inet_sk(sk)->inet_num = snum; 311 udp_sk(sk)->udp_port_hash = snum; 312 udp_sk(sk)->udp_portaddr_hash ^= snum; 313 if (sk_unhashed(sk)) { 314 if (sk->sk_reuseport && 315 udp_reuseport_add_sock(sk, hslot)) { 316 inet_sk(sk)->inet_num = 0; 317 udp_sk(sk)->udp_port_hash = 0; 318 udp_sk(sk)->udp_portaddr_hash ^= snum; 319 goto fail_unlock; 320 } 321 322 sk_add_node_rcu(sk, &hslot->head); 323 hslot->count++; 324 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 325 326 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 327 spin_lock(&hslot2->lock); 328 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 329 sk->sk_family == AF_INET6) 330 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 331 &hslot2->head); 332 else 333 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 334 &hslot2->head); 335 hslot2->count++; 336 spin_unlock(&hslot2->lock); 337 } 338 sock_set_flag(sk, SOCK_RCU_FREE); 339 error = 0; 340 fail_unlock: 341 spin_unlock_bh(&hslot->lock); 342 fail: 343 return error; 344 } 345 EXPORT_SYMBOL(udp_lib_get_port); 346 347 int udp_v4_get_port(struct sock *sk, unsigned short snum) 348 { 349 unsigned int hash2_nulladdr = 350 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 351 unsigned int hash2_partial = 352 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 353 354 /* precompute partial secondary hash */ 355 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 356 return udp_lib_get_port(sk, snum, hash2_nulladdr); 357 } 358 359 static int compute_score(struct sock *sk, struct net *net, 360 __be32 saddr, __be16 sport, 361 __be32 daddr, unsigned short hnum, 362 int dif, int sdif) 363 { 364 int score; 365 struct inet_sock *inet; 366 bool dev_match; 367 368 if (!net_eq(sock_net(sk), net) || 369 udp_sk(sk)->udp_port_hash != hnum || 370 ipv6_only_sock(sk)) 371 return -1; 372 373 if (sk->sk_rcv_saddr != daddr) 374 return -1; 375 376 score = (sk->sk_family == PF_INET) ? 2 : 1; 377 378 inet = inet_sk(sk); 379 if (inet->inet_daddr) { 380 if (inet->inet_daddr != saddr) 381 return -1; 382 score += 4; 383 } 384 385 if (inet->inet_dport) { 386 if (inet->inet_dport != sport) 387 return -1; 388 score += 4; 389 } 390 391 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 392 dif, sdif); 393 if (!dev_match) 394 return -1; 395 if (sk->sk_bound_dev_if) 396 score += 4; 397 398 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 399 score++; 400 return score; 401 } 402 403 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 404 const __u16 lport, const __be32 faddr, 405 const __be16 fport) 406 { 407 static u32 udp_ehash_secret __read_mostly; 408 409 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 410 411 return __inet_ehashfn(laddr, lport, faddr, fport, 412 udp_ehash_secret + net_hash_mix(net)); 413 } 414 415 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 416 struct sk_buff *skb, 417 __be32 saddr, __be16 sport, 418 __be32 daddr, unsigned short hnum) 419 { 420 struct sock *reuse_sk = NULL; 421 u32 hash; 422 423 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 424 hash = udp_ehashfn(net, daddr, hnum, saddr, sport); 425 reuse_sk = reuseport_select_sock(sk, hash, skb, 426 sizeof(struct udphdr)); 427 } 428 return reuse_sk; 429 } 430 431 /* called with rcu_read_lock() */ 432 static struct sock *udp4_lib_lookup2(struct net *net, 433 __be32 saddr, __be16 sport, 434 __be32 daddr, unsigned int hnum, 435 int dif, int sdif, 436 struct udp_hslot *hslot2, 437 struct sk_buff *skb) 438 { 439 struct sock *sk, *result; 440 int score, badness; 441 442 result = NULL; 443 badness = 0; 444 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 445 score = compute_score(sk, net, saddr, sport, 446 daddr, hnum, dif, sdif); 447 if (score > badness) { 448 result = lookup_reuseport(net, sk, skb, 449 saddr, sport, daddr, hnum); 450 /* Fall back to scoring if group has connections */ 451 if (result && !reuseport_has_conns(sk, false)) 452 return result; 453 454 result = result ? : sk; 455 badness = score; 456 } 457 } 458 return result; 459 } 460 461 static struct sock *udp4_lookup_run_bpf(struct net *net, 462 struct udp_table *udptable, 463 struct sk_buff *skb, 464 __be32 saddr, __be16 sport, 465 __be32 daddr, u16 hnum, const int dif) 466 { 467 struct sock *sk, *reuse_sk; 468 bool no_reuseport; 469 470 if (udptable != &udp_table) 471 return NULL; /* only UDP is supported */ 472 473 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport, 474 daddr, hnum, dif, &sk); 475 if (no_reuseport || IS_ERR_OR_NULL(sk)) 476 return sk; 477 478 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 479 if (reuse_sk) 480 sk = reuse_sk; 481 return sk; 482 } 483 484 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 485 * harder than this. -DaveM 486 */ 487 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 488 __be16 sport, __be32 daddr, __be16 dport, int dif, 489 int sdif, struct udp_table *udptable, struct sk_buff *skb) 490 { 491 unsigned short hnum = ntohs(dport); 492 unsigned int hash2, slot2; 493 struct udp_hslot *hslot2; 494 struct sock *result, *sk; 495 496 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 497 slot2 = hash2 & udptable->mask; 498 hslot2 = &udptable->hash2[slot2]; 499 500 /* Lookup connected or non-wildcard socket */ 501 result = udp4_lib_lookup2(net, saddr, sport, 502 daddr, hnum, dif, sdif, 503 hslot2, skb); 504 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 505 goto done; 506 507 /* Lookup redirect from BPF */ 508 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { 509 sk = udp4_lookup_run_bpf(net, udptable, skb, 510 saddr, sport, daddr, hnum, dif); 511 if (sk) { 512 result = sk; 513 goto done; 514 } 515 } 516 517 /* Got non-wildcard socket or error on first lookup */ 518 if (result) 519 goto done; 520 521 /* Lookup wildcard sockets */ 522 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 523 slot2 = hash2 & udptable->mask; 524 hslot2 = &udptable->hash2[slot2]; 525 526 result = udp4_lib_lookup2(net, saddr, sport, 527 htonl(INADDR_ANY), hnum, dif, sdif, 528 hslot2, skb); 529 done: 530 if (IS_ERR(result)) 531 return NULL; 532 return result; 533 } 534 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 535 536 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 537 __be16 sport, __be16 dport, 538 struct udp_table *udptable) 539 { 540 const struct iphdr *iph = ip_hdr(skb); 541 542 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 543 iph->daddr, dport, inet_iif(skb), 544 inet_sdif(skb), udptable, skb); 545 } 546 547 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 548 __be16 sport, __be16 dport) 549 { 550 const struct iphdr *iph = ip_hdr(skb); 551 552 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 553 iph->daddr, dport, inet_iif(skb), 554 inet_sdif(skb), &udp_table, NULL); 555 } 556 557 /* Must be called under rcu_read_lock(). 558 * Does increment socket refcount. 559 */ 560 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 561 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 562 __be32 daddr, __be16 dport, int dif) 563 { 564 struct sock *sk; 565 566 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 567 dif, 0, &udp_table, NULL); 568 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 569 sk = NULL; 570 return sk; 571 } 572 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 573 #endif 574 575 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 576 __be16 loc_port, __be32 loc_addr, 577 __be16 rmt_port, __be32 rmt_addr, 578 int dif, int sdif, unsigned short hnum) 579 { 580 struct inet_sock *inet = inet_sk(sk); 581 582 if (!net_eq(sock_net(sk), net) || 583 udp_sk(sk)->udp_port_hash != hnum || 584 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 585 (inet->inet_dport != rmt_port && inet->inet_dport) || 586 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 587 ipv6_only_sock(sk) || 588 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 589 return false; 590 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 591 return false; 592 return true; 593 } 594 595 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 596 void udp_encap_enable(void) 597 { 598 static_branch_inc(&udp_encap_needed_key); 599 } 600 EXPORT_SYMBOL(udp_encap_enable); 601 602 void udp_encap_disable(void) 603 { 604 static_branch_dec(&udp_encap_needed_key); 605 } 606 EXPORT_SYMBOL(udp_encap_disable); 607 608 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 609 * through error handlers in encapsulations looking for a match. 610 */ 611 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 612 { 613 int i; 614 615 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 616 int (*handler)(struct sk_buff *skb, u32 info); 617 const struct ip_tunnel_encap_ops *encap; 618 619 encap = rcu_dereference(iptun_encaps[i]); 620 if (!encap) 621 continue; 622 handler = encap->err_handler; 623 if (handler && !handler(skb, info)) 624 return 0; 625 } 626 627 return -ENOENT; 628 } 629 630 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 631 * reversing source and destination port: this will match tunnels that force the 632 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 633 * lwtunnels might actually break this assumption by being configured with 634 * different destination ports on endpoints, in this case we won't be able to 635 * trace ICMP messages back to them. 636 * 637 * If this doesn't match any socket, probe tunnels with arbitrary destination 638 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 639 * we've sent packets to won't necessarily match the local destination port. 640 * 641 * Then ask the tunnel implementation to match the error against a valid 642 * association. 643 * 644 * Return an error if we can't find a match, the socket if we need further 645 * processing, zero otherwise. 646 */ 647 static struct sock *__udp4_lib_err_encap(struct net *net, 648 const struct iphdr *iph, 649 struct udphdr *uh, 650 struct udp_table *udptable, 651 struct sock *sk, 652 struct sk_buff *skb, u32 info) 653 { 654 int (*lookup)(struct sock *sk, struct sk_buff *skb); 655 int network_offset, transport_offset; 656 struct udp_sock *up; 657 658 network_offset = skb_network_offset(skb); 659 transport_offset = skb_transport_offset(skb); 660 661 /* Network header needs to point to the outer IPv4 header inside ICMP */ 662 skb_reset_network_header(skb); 663 664 /* Transport header needs to point to the UDP header */ 665 skb_set_transport_header(skb, iph->ihl << 2); 666 667 if (sk) { 668 up = udp_sk(sk); 669 670 lookup = READ_ONCE(up->encap_err_lookup); 671 if (lookup && lookup(sk, skb)) 672 sk = NULL; 673 674 goto out; 675 } 676 677 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 678 iph->saddr, uh->dest, skb->dev->ifindex, 0, 679 udptable, NULL); 680 if (sk) { 681 up = udp_sk(sk); 682 683 lookup = READ_ONCE(up->encap_err_lookup); 684 if (!lookup || lookup(sk, skb)) 685 sk = NULL; 686 } 687 688 out: 689 if (!sk) 690 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 691 692 skb_set_transport_header(skb, transport_offset); 693 skb_set_network_header(skb, network_offset); 694 695 return sk; 696 } 697 698 /* 699 * This routine is called by the ICMP module when it gets some 700 * sort of error condition. If err < 0 then the socket should 701 * be closed and the error returned to the user. If err > 0 702 * it's just the icmp type << 8 | icmp code. 703 * Header points to the ip header of the error packet. We move 704 * on past this. Then (as it used to claim before adjustment) 705 * header points to the first 8 bytes of the udp header. We need 706 * to find the appropriate port. 707 */ 708 709 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 710 { 711 struct inet_sock *inet; 712 const struct iphdr *iph = (const struct iphdr *)skb->data; 713 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 714 const int type = icmp_hdr(skb)->type; 715 const int code = icmp_hdr(skb)->code; 716 bool tunnel = false; 717 struct sock *sk; 718 int harderr; 719 int err; 720 struct net *net = dev_net(skb->dev); 721 722 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 723 iph->saddr, uh->source, skb->dev->ifindex, 724 inet_sdif(skb), udptable, NULL); 725 726 if (!sk || udp_sk(sk)->encap_type) { 727 /* No socket for error: try tunnels before discarding */ 728 if (static_branch_unlikely(&udp_encap_needed_key)) { 729 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 730 info); 731 if (!sk) 732 return 0; 733 } else 734 sk = ERR_PTR(-ENOENT); 735 736 if (IS_ERR(sk)) { 737 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 738 return PTR_ERR(sk); 739 } 740 741 tunnel = true; 742 } 743 744 err = 0; 745 harderr = 0; 746 inet = inet_sk(sk); 747 748 switch (type) { 749 default: 750 case ICMP_TIME_EXCEEDED: 751 err = EHOSTUNREACH; 752 break; 753 case ICMP_SOURCE_QUENCH: 754 goto out; 755 case ICMP_PARAMETERPROB: 756 err = EPROTO; 757 harderr = 1; 758 break; 759 case ICMP_DEST_UNREACH: 760 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 761 ipv4_sk_update_pmtu(skb, sk, info); 762 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 763 err = EMSGSIZE; 764 harderr = 1; 765 break; 766 } 767 goto out; 768 } 769 err = EHOSTUNREACH; 770 if (code <= NR_ICMP_UNREACH) { 771 harderr = icmp_err_convert[code].fatal; 772 err = icmp_err_convert[code].errno; 773 } 774 break; 775 case ICMP_REDIRECT: 776 ipv4_sk_redirect(skb, sk); 777 goto out; 778 } 779 780 /* 781 * RFC1122: OK. Passes ICMP errors back to application, as per 782 * 4.1.3.3. 783 */ 784 if (tunnel) { 785 /* ...not for tunnels though: we don't have a sending socket */ 786 if (udp_sk(sk)->encap_err_rcv) 787 udp_sk(sk)->encap_err_rcv(sk, skb, iph->ihl << 2); 788 goto out; 789 } 790 if (!inet->recverr) { 791 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 792 goto out; 793 } else 794 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 795 796 sk->sk_err = err; 797 sk_error_report(sk); 798 out: 799 return 0; 800 } 801 802 int udp_err(struct sk_buff *skb, u32 info) 803 { 804 return __udp4_lib_err(skb, info, &udp_table); 805 } 806 807 /* 808 * Throw away all pending data and cancel the corking. Socket is locked. 809 */ 810 void udp_flush_pending_frames(struct sock *sk) 811 { 812 struct udp_sock *up = udp_sk(sk); 813 814 if (up->pending) { 815 up->len = 0; 816 up->pending = 0; 817 ip_flush_pending_frames(sk); 818 } 819 } 820 EXPORT_SYMBOL(udp_flush_pending_frames); 821 822 /** 823 * udp4_hwcsum - handle outgoing HW checksumming 824 * @skb: sk_buff containing the filled-in UDP header 825 * (checksum field must be zeroed out) 826 * @src: source IP address 827 * @dst: destination IP address 828 */ 829 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 830 { 831 struct udphdr *uh = udp_hdr(skb); 832 int offset = skb_transport_offset(skb); 833 int len = skb->len - offset; 834 int hlen = len; 835 __wsum csum = 0; 836 837 if (!skb_has_frag_list(skb)) { 838 /* 839 * Only one fragment on the socket. 840 */ 841 skb->csum_start = skb_transport_header(skb) - skb->head; 842 skb->csum_offset = offsetof(struct udphdr, check); 843 uh->check = ~csum_tcpudp_magic(src, dst, len, 844 IPPROTO_UDP, 0); 845 } else { 846 struct sk_buff *frags; 847 848 /* 849 * HW-checksum won't work as there are two or more 850 * fragments on the socket so that all csums of sk_buffs 851 * should be together 852 */ 853 skb_walk_frags(skb, frags) { 854 csum = csum_add(csum, frags->csum); 855 hlen -= frags->len; 856 } 857 858 csum = skb_checksum(skb, offset, hlen, csum); 859 skb->ip_summed = CHECKSUM_NONE; 860 861 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 862 if (uh->check == 0) 863 uh->check = CSUM_MANGLED_0; 864 } 865 } 866 EXPORT_SYMBOL_GPL(udp4_hwcsum); 867 868 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 869 * for the simple case like when setting the checksum for a UDP tunnel. 870 */ 871 void udp_set_csum(bool nocheck, struct sk_buff *skb, 872 __be32 saddr, __be32 daddr, int len) 873 { 874 struct udphdr *uh = udp_hdr(skb); 875 876 if (nocheck) { 877 uh->check = 0; 878 } else if (skb_is_gso(skb)) { 879 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 880 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 881 uh->check = 0; 882 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 883 if (uh->check == 0) 884 uh->check = CSUM_MANGLED_0; 885 } else { 886 skb->ip_summed = CHECKSUM_PARTIAL; 887 skb->csum_start = skb_transport_header(skb) - skb->head; 888 skb->csum_offset = offsetof(struct udphdr, check); 889 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 890 } 891 } 892 EXPORT_SYMBOL(udp_set_csum); 893 894 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 895 struct inet_cork *cork) 896 { 897 struct sock *sk = skb->sk; 898 struct inet_sock *inet = inet_sk(sk); 899 struct udphdr *uh; 900 int err; 901 int is_udplite = IS_UDPLITE(sk); 902 int offset = skb_transport_offset(skb); 903 int len = skb->len - offset; 904 int datalen = len - sizeof(*uh); 905 __wsum csum = 0; 906 907 /* 908 * Create a UDP header 909 */ 910 uh = udp_hdr(skb); 911 uh->source = inet->inet_sport; 912 uh->dest = fl4->fl4_dport; 913 uh->len = htons(len); 914 uh->check = 0; 915 916 if (cork->gso_size) { 917 const int hlen = skb_network_header_len(skb) + 918 sizeof(struct udphdr); 919 920 if (hlen + cork->gso_size > cork->fragsize) { 921 kfree_skb(skb); 922 return -EINVAL; 923 } 924 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 925 kfree_skb(skb); 926 return -EINVAL; 927 } 928 if (sk->sk_no_check_tx) { 929 kfree_skb(skb); 930 return -EINVAL; 931 } 932 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 933 dst_xfrm(skb_dst(skb))) { 934 kfree_skb(skb); 935 return -EIO; 936 } 937 938 if (datalen > cork->gso_size) { 939 skb_shinfo(skb)->gso_size = cork->gso_size; 940 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 941 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 942 cork->gso_size); 943 } 944 goto csum_partial; 945 } 946 947 if (is_udplite) /* UDP-Lite */ 948 csum = udplite_csum(skb); 949 950 else if (sk->sk_no_check_tx) { /* UDP csum off */ 951 952 skb->ip_summed = CHECKSUM_NONE; 953 goto send; 954 955 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 956 csum_partial: 957 958 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 959 goto send; 960 961 } else 962 csum = udp_csum(skb); 963 964 /* add protocol-dependent pseudo-header */ 965 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 966 sk->sk_protocol, csum); 967 if (uh->check == 0) 968 uh->check = CSUM_MANGLED_0; 969 970 send: 971 err = ip_send_skb(sock_net(sk), skb); 972 if (err) { 973 if (err == -ENOBUFS && !inet->recverr) { 974 UDP_INC_STATS(sock_net(sk), 975 UDP_MIB_SNDBUFERRORS, is_udplite); 976 err = 0; 977 } 978 } else 979 UDP_INC_STATS(sock_net(sk), 980 UDP_MIB_OUTDATAGRAMS, is_udplite); 981 return err; 982 } 983 984 /* 985 * Push out all pending data as one UDP datagram. Socket is locked. 986 */ 987 int udp_push_pending_frames(struct sock *sk) 988 { 989 struct udp_sock *up = udp_sk(sk); 990 struct inet_sock *inet = inet_sk(sk); 991 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 992 struct sk_buff *skb; 993 int err = 0; 994 995 skb = ip_finish_skb(sk, fl4); 996 if (!skb) 997 goto out; 998 999 err = udp_send_skb(skb, fl4, &inet->cork.base); 1000 1001 out: 1002 up->len = 0; 1003 up->pending = 0; 1004 return err; 1005 } 1006 EXPORT_SYMBOL(udp_push_pending_frames); 1007 1008 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1009 { 1010 switch (cmsg->cmsg_type) { 1011 case UDP_SEGMENT: 1012 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1013 return -EINVAL; 1014 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1015 return 0; 1016 default: 1017 return -EINVAL; 1018 } 1019 } 1020 1021 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1022 { 1023 struct cmsghdr *cmsg; 1024 bool need_ip = false; 1025 int err; 1026 1027 for_each_cmsghdr(cmsg, msg) { 1028 if (!CMSG_OK(msg, cmsg)) 1029 return -EINVAL; 1030 1031 if (cmsg->cmsg_level != SOL_UDP) { 1032 need_ip = true; 1033 continue; 1034 } 1035 1036 err = __udp_cmsg_send(cmsg, gso_size); 1037 if (err) 1038 return err; 1039 } 1040 1041 return need_ip; 1042 } 1043 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1044 1045 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1046 { 1047 struct inet_sock *inet = inet_sk(sk); 1048 struct udp_sock *up = udp_sk(sk); 1049 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1050 struct flowi4 fl4_stack; 1051 struct flowi4 *fl4; 1052 int ulen = len; 1053 struct ipcm_cookie ipc; 1054 struct rtable *rt = NULL; 1055 int free = 0; 1056 int connected = 0; 1057 __be32 daddr, faddr, saddr; 1058 __be16 dport; 1059 u8 tos; 1060 int err, is_udplite = IS_UDPLITE(sk); 1061 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1062 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1063 struct sk_buff *skb; 1064 struct ip_options_data opt_copy; 1065 1066 if (len > 0xFFFF) 1067 return -EMSGSIZE; 1068 1069 /* 1070 * Check the flags. 1071 */ 1072 1073 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1074 return -EOPNOTSUPP; 1075 1076 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1077 1078 fl4 = &inet->cork.fl.u.ip4; 1079 if (up->pending) { 1080 /* 1081 * There are pending frames. 1082 * The socket lock must be held while it's corked. 1083 */ 1084 lock_sock(sk); 1085 if (likely(up->pending)) { 1086 if (unlikely(up->pending != AF_INET)) { 1087 release_sock(sk); 1088 return -EINVAL; 1089 } 1090 goto do_append_data; 1091 } 1092 release_sock(sk); 1093 } 1094 ulen += sizeof(struct udphdr); 1095 1096 /* 1097 * Get and verify the address. 1098 */ 1099 if (usin) { 1100 if (msg->msg_namelen < sizeof(*usin)) 1101 return -EINVAL; 1102 if (usin->sin_family != AF_INET) { 1103 if (usin->sin_family != AF_UNSPEC) 1104 return -EAFNOSUPPORT; 1105 } 1106 1107 daddr = usin->sin_addr.s_addr; 1108 dport = usin->sin_port; 1109 if (dport == 0) 1110 return -EINVAL; 1111 } else { 1112 if (sk->sk_state != TCP_ESTABLISHED) 1113 return -EDESTADDRREQ; 1114 daddr = inet->inet_daddr; 1115 dport = inet->inet_dport; 1116 /* Open fast path for connected socket. 1117 Route will not be used, if at least one option is set. 1118 */ 1119 connected = 1; 1120 } 1121 1122 ipcm_init_sk(&ipc, inet); 1123 ipc.gso_size = READ_ONCE(up->gso_size); 1124 1125 if (msg->msg_controllen) { 1126 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1127 if (err > 0) 1128 err = ip_cmsg_send(sk, msg, &ipc, 1129 sk->sk_family == AF_INET6); 1130 if (unlikely(err < 0)) { 1131 kfree(ipc.opt); 1132 return err; 1133 } 1134 if (ipc.opt) 1135 free = 1; 1136 connected = 0; 1137 } 1138 if (!ipc.opt) { 1139 struct ip_options_rcu *inet_opt; 1140 1141 rcu_read_lock(); 1142 inet_opt = rcu_dereference(inet->inet_opt); 1143 if (inet_opt) { 1144 memcpy(&opt_copy, inet_opt, 1145 sizeof(*inet_opt) + inet_opt->opt.optlen); 1146 ipc.opt = &opt_copy.opt; 1147 } 1148 rcu_read_unlock(); 1149 } 1150 1151 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1152 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1153 (struct sockaddr *)usin, &ipc.addr); 1154 if (err) 1155 goto out_free; 1156 if (usin) { 1157 if (usin->sin_port == 0) { 1158 /* BPF program set invalid port. Reject it. */ 1159 err = -EINVAL; 1160 goto out_free; 1161 } 1162 daddr = usin->sin_addr.s_addr; 1163 dport = usin->sin_port; 1164 } 1165 } 1166 1167 saddr = ipc.addr; 1168 ipc.addr = faddr = daddr; 1169 1170 if (ipc.opt && ipc.opt->opt.srr) { 1171 if (!daddr) { 1172 err = -EINVAL; 1173 goto out_free; 1174 } 1175 faddr = ipc.opt->opt.faddr; 1176 connected = 0; 1177 } 1178 tos = get_rttos(&ipc, inet); 1179 if (sock_flag(sk, SOCK_LOCALROUTE) || 1180 (msg->msg_flags & MSG_DONTROUTE) || 1181 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1182 tos |= RTO_ONLINK; 1183 connected = 0; 1184 } 1185 1186 if (ipv4_is_multicast(daddr)) { 1187 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1188 ipc.oif = inet->mc_index; 1189 if (!saddr) 1190 saddr = inet->mc_addr; 1191 connected = 0; 1192 } else if (!ipc.oif) { 1193 ipc.oif = inet->uc_index; 1194 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1195 /* oif is set, packet is to local broadcast and 1196 * uc_index is set. oif is most likely set 1197 * by sk_bound_dev_if. If uc_index != oif check if the 1198 * oif is an L3 master and uc_index is an L3 slave. 1199 * If so, we want to allow the send using the uc_index. 1200 */ 1201 if (ipc.oif != inet->uc_index && 1202 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1203 inet->uc_index)) { 1204 ipc.oif = inet->uc_index; 1205 } 1206 } 1207 1208 if (connected) 1209 rt = (struct rtable *)sk_dst_check(sk, 0); 1210 1211 if (!rt) { 1212 struct net *net = sock_net(sk); 1213 __u8 flow_flags = inet_sk_flowi_flags(sk); 1214 1215 fl4 = &fl4_stack; 1216 1217 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, 1218 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1219 flow_flags, 1220 faddr, saddr, dport, inet->inet_sport, 1221 sk->sk_uid); 1222 1223 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1224 rt = ip_route_output_flow(net, fl4, sk); 1225 if (IS_ERR(rt)) { 1226 err = PTR_ERR(rt); 1227 rt = NULL; 1228 if (err == -ENETUNREACH) 1229 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1230 goto out; 1231 } 1232 1233 err = -EACCES; 1234 if ((rt->rt_flags & RTCF_BROADCAST) && 1235 !sock_flag(sk, SOCK_BROADCAST)) 1236 goto out; 1237 if (connected) 1238 sk_dst_set(sk, dst_clone(&rt->dst)); 1239 } 1240 1241 if (msg->msg_flags&MSG_CONFIRM) 1242 goto do_confirm; 1243 back_from_confirm: 1244 1245 saddr = fl4->saddr; 1246 if (!ipc.addr) 1247 daddr = ipc.addr = fl4->daddr; 1248 1249 /* Lockless fast path for the non-corking case. */ 1250 if (!corkreq) { 1251 struct inet_cork cork; 1252 1253 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1254 sizeof(struct udphdr), &ipc, &rt, 1255 &cork, msg->msg_flags); 1256 err = PTR_ERR(skb); 1257 if (!IS_ERR_OR_NULL(skb)) 1258 err = udp_send_skb(skb, fl4, &cork); 1259 goto out; 1260 } 1261 1262 lock_sock(sk); 1263 if (unlikely(up->pending)) { 1264 /* The socket is already corked while preparing it. */ 1265 /* ... which is an evident application bug. --ANK */ 1266 release_sock(sk); 1267 1268 net_dbg_ratelimited("socket already corked\n"); 1269 err = -EINVAL; 1270 goto out; 1271 } 1272 /* 1273 * Now cork the socket to pend data. 1274 */ 1275 fl4 = &inet->cork.fl.u.ip4; 1276 fl4->daddr = daddr; 1277 fl4->saddr = saddr; 1278 fl4->fl4_dport = dport; 1279 fl4->fl4_sport = inet->inet_sport; 1280 up->pending = AF_INET; 1281 1282 do_append_data: 1283 up->len += ulen; 1284 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1285 sizeof(struct udphdr), &ipc, &rt, 1286 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1287 if (err) 1288 udp_flush_pending_frames(sk); 1289 else if (!corkreq) 1290 err = udp_push_pending_frames(sk); 1291 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1292 up->pending = 0; 1293 release_sock(sk); 1294 1295 out: 1296 ip_rt_put(rt); 1297 out_free: 1298 if (free) 1299 kfree(ipc.opt); 1300 if (!err) 1301 return len; 1302 /* 1303 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1304 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1305 * we don't have a good statistic (IpOutDiscards but it can be too many 1306 * things). We could add another new stat but at least for now that 1307 * seems like overkill. 1308 */ 1309 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1310 UDP_INC_STATS(sock_net(sk), 1311 UDP_MIB_SNDBUFERRORS, is_udplite); 1312 } 1313 return err; 1314 1315 do_confirm: 1316 if (msg->msg_flags & MSG_PROBE) 1317 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1318 if (!(msg->msg_flags&MSG_PROBE) || len) 1319 goto back_from_confirm; 1320 err = 0; 1321 goto out; 1322 } 1323 EXPORT_SYMBOL(udp_sendmsg); 1324 1325 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1326 size_t size, int flags) 1327 { 1328 struct inet_sock *inet = inet_sk(sk); 1329 struct udp_sock *up = udp_sk(sk); 1330 int ret; 1331 1332 if (flags & MSG_SENDPAGE_NOTLAST) 1333 flags |= MSG_MORE; 1334 1335 if (!up->pending) { 1336 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1337 1338 /* Call udp_sendmsg to specify destination address which 1339 * sendpage interface can't pass. 1340 * This will succeed only when the socket is connected. 1341 */ 1342 ret = udp_sendmsg(sk, &msg, 0); 1343 if (ret < 0) 1344 return ret; 1345 } 1346 1347 lock_sock(sk); 1348 1349 if (unlikely(!up->pending)) { 1350 release_sock(sk); 1351 1352 net_dbg_ratelimited("cork failed\n"); 1353 return -EINVAL; 1354 } 1355 1356 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1357 page, offset, size, flags); 1358 if (ret == -EOPNOTSUPP) { 1359 release_sock(sk); 1360 return sock_no_sendpage(sk->sk_socket, page, offset, 1361 size, flags); 1362 } 1363 if (ret < 0) { 1364 udp_flush_pending_frames(sk); 1365 goto out; 1366 } 1367 1368 up->len += size; 1369 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE))) 1370 ret = udp_push_pending_frames(sk); 1371 if (!ret) 1372 ret = size; 1373 out: 1374 release_sock(sk); 1375 return ret; 1376 } 1377 1378 #define UDP_SKB_IS_STATELESS 0x80000000 1379 1380 /* all head states (dst, sk, nf conntrack) except skb extensions are 1381 * cleared by udp_rcv(). 1382 * 1383 * We need to preserve secpath, if present, to eventually process 1384 * IP_CMSG_PASSSEC at recvmsg() time. 1385 * 1386 * Other extensions can be cleared. 1387 */ 1388 static bool udp_try_make_stateless(struct sk_buff *skb) 1389 { 1390 if (!skb_has_extensions(skb)) 1391 return true; 1392 1393 if (!secpath_exists(skb)) { 1394 skb_ext_reset(skb); 1395 return true; 1396 } 1397 1398 return false; 1399 } 1400 1401 static void udp_set_dev_scratch(struct sk_buff *skb) 1402 { 1403 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1404 1405 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1406 scratch->_tsize_state = skb->truesize; 1407 #if BITS_PER_LONG == 64 1408 scratch->len = skb->len; 1409 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1410 scratch->is_linear = !skb_is_nonlinear(skb); 1411 #endif 1412 if (udp_try_make_stateless(skb)) 1413 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1414 } 1415 1416 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1417 { 1418 /* We come here after udp_lib_checksum_complete() returned 0. 1419 * This means that __skb_checksum_complete() might have 1420 * set skb->csum_valid to 1. 1421 * On 64bit platforms, we can set csum_unnecessary 1422 * to true, but only if the skb is not shared. 1423 */ 1424 #if BITS_PER_LONG == 64 1425 if (!skb_shared(skb)) 1426 udp_skb_scratch(skb)->csum_unnecessary = true; 1427 #endif 1428 } 1429 1430 static int udp_skb_truesize(struct sk_buff *skb) 1431 { 1432 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1433 } 1434 1435 static bool udp_skb_has_head_state(struct sk_buff *skb) 1436 { 1437 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1438 } 1439 1440 /* fully reclaim rmem/fwd memory allocated for skb */ 1441 static void udp_rmem_release(struct sock *sk, int size, int partial, 1442 bool rx_queue_lock_held) 1443 { 1444 struct udp_sock *up = udp_sk(sk); 1445 struct sk_buff_head *sk_queue; 1446 int amt; 1447 1448 if (likely(partial)) { 1449 up->forward_deficit += size; 1450 size = up->forward_deficit; 1451 if (size < (sk->sk_rcvbuf >> 2) && 1452 !skb_queue_empty(&up->reader_queue)) 1453 return; 1454 } else { 1455 size += up->forward_deficit; 1456 } 1457 up->forward_deficit = 0; 1458 1459 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1460 * if the called don't held it already 1461 */ 1462 sk_queue = &sk->sk_receive_queue; 1463 if (!rx_queue_lock_held) 1464 spin_lock(&sk_queue->lock); 1465 1466 1467 sk->sk_forward_alloc += size; 1468 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1469 sk->sk_forward_alloc -= amt; 1470 1471 if (amt) 1472 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 1473 1474 atomic_sub(size, &sk->sk_rmem_alloc); 1475 1476 /* this can save us from acquiring the rx queue lock on next receive */ 1477 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1478 1479 if (!rx_queue_lock_held) 1480 spin_unlock(&sk_queue->lock); 1481 } 1482 1483 /* Note: called with reader_queue.lock held. 1484 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1485 * This avoids a cache line miss while receive_queue lock is held. 1486 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1487 */ 1488 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1489 { 1490 prefetch(&skb->data); 1491 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1492 } 1493 EXPORT_SYMBOL(udp_skb_destructor); 1494 1495 /* as above, but the caller held the rx queue lock, too */ 1496 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1497 { 1498 prefetch(&skb->data); 1499 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1500 } 1501 1502 /* Idea of busylocks is to let producers grab an extra spinlock 1503 * to relieve pressure on the receive_queue spinlock shared by consumer. 1504 * Under flood, this means that only one producer can be in line 1505 * trying to acquire the receive_queue spinlock. 1506 * These busylock can be allocated on a per cpu manner, instead of a 1507 * per socket one (that would consume a cache line per socket) 1508 */ 1509 static int udp_busylocks_log __read_mostly; 1510 static spinlock_t *udp_busylocks __read_mostly; 1511 1512 static spinlock_t *busylock_acquire(void *ptr) 1513 { 1514 spinlock_t *busy; 1515 1516 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1517 spin_lock(busy); 1518 return busy; 1519 } 1520 1521 static void busylock_release(spinlock_t *busy) 1522 { 1523 if (busy) 1524 spin_unlock(busy); 1525 } 1526 1527 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1528 { 1529 struct sk_buff_head *list = &sk->sk_receive_queue; 1530 int rmem, delta, amt, err = -ENOMEM; 1531 spinlock_t *busy = NULL; 1532 int size; 1533 1534 /* try to avoid the costly atomic add/sub pair when the receive 1535 * queue is full; always allow at least a packet 1536 */ 1537 rmem = atomic_read(&sk->sk_rmem_alloc); 1538 if (rmem > sk->sk_rcvbuf) 1539 goto drop; 1540 1541 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1542 * having linear skbs : 1543 * - Reduce memory overhead and thus increase receive queue capacity 1544 * - Less cache line misses at copyout() time 1545 * - Less work at consume_skb() (less alien page frag freeing) 1546 */ 1547 if (rmem > (sk->sk_rcvbuf >> 1)) { 1548 skb_condense(skb); 1549 1550 busy = busylock_acquire(sk); 1551 } 1552 size = skb->truesize; 1553 udp_set_dev_scratch(skb); 1554 1555 /* we drop only if the receive buf is full and the receive 1556 * queue contains some other skb 1557 */ 1558 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1559 if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) 1560 goto uncharge_drop; 1561 1562 spin_lock(&list->lock); 1563 if (size >= sk->sk_forward_alloc) { 1564 amt = sk_mem_pages(size); 1565 delta = amt << PAGE_SHIFT; 1566 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1567 err = -ENOBUFS; 1568 spin_unlock(&list->lock); 1569 goto uncharge_drop; 1570 } 1571 1572 sk->sk_forward_alloc += delta; 1573 } 1574 1575 sk->sk_forward_alloc -= size; 1576 1577 /* no need to setup a destructor, we will explicitly release the 1578 * forward allocated memory on dequeue 1579 */ 1580 sock_skb_set_dropcount(sk, skb); 1581 1582 __skb_queue_tail(list, skb); 1583 spin_unlock(&list->lock); 1584 1585 if (!sock_flag(sk, SOCK_DEAD)) 1586 sk->sk_data_ready(sk); 1587 1588 busylock_release(busy); 1589 return 0; 1590 1591 uncharge_drop: 1592 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1593 1594 drop: 1595 atomic_inc(&sk->sk_drops); 1596 busylock_release(busy); 1597 return err; 1598 } 1599 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1600 1601 void udp_destruct_sock(struct sock *sk) 1602 { 1603 /* reclaim completely the forward allocated memory */ 1604 struct udp_sock *up = udp_sk(sk); 1605 unsigned int total = 0; 1606 struct sk_buff *skb; 1607 1608 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1609 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1610 total += skb->truesize; 1611 kfree_skb(skb); 1612 } 1613 udp_rmem_release(sk, total, 0, true); 1614 1615 inet_sock_destruct(sk); 1616 } 1617 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1618 1619 int udp_init_sock(struct sock *sk) 1620 { 1621 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1622 sk->sk_destruct = udp_destruct_sock; 1623 return 0; 1624 } 1625 EXPORT_SYMBOL_GPL(udp_init_sock); 1626 1627 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1628 { 1629 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1630 bool slow = lock_sock_fast(sk); 1631 1632 sk_peek_offset_bwd(sk, len); 1633 unlock_sock_fast(sk, slow); 1634 } 1635 1636 if (!skb_unref(skb)) 1637 return; 1638 1639 /* In the more common cases we cleared the head states previously, 1640 * see __udp_queue_rcv_skb(). 1641 */ 1642 if (unlikely(udp_skb_has_head_state(skb))) 1643 skb_release_head_state(skb); 1644 __consume_stateless_skb(skb); 1645 } 1646 EXPORT_SYMBOL_GPL(skb_consume_udp); 1647 1648 static struct sk_buff *__first_packet_length(struct sock *sk, 1649 struct sk_buff_head *rcvq, 1650 int *total) 1651 { 1652 struct sk_buff *skb; 1653 1654 while ((skb = skb_peek(rcvq)) != NULL) { 1655 if (udp_lib_checksum_complete(skb)) { 1656 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1657 IS_UDPLITE(sk)); 1658 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1659 IS_UDPLITE(sk)); 1660 atomic_inc(&sk->sk_drops); 1661 __skb_unlink(skb, rcvq); 1662 *total += skb->truesize; 1663 kfree_skb(skb); 1664 } else { 1665 udp_skb_csum_unnecessary_set(skb); 1666 break; 1667 } 1668 } 1669 return skb; 1670 } 1671 1672 /** 1673 * first_packet_length - return length of first packet in receive queue 1674 * @sk: socket 1675 * 1676 * Drops all bad checksum frames, until a valid one is found. 1677 * Returns the length of found skb, or -1 if none is found. 1678 */ 1679 static int first_packet_length(struct sock *sk) 1680 { 1681 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1682 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1683 struct sk_buff *skb; 1684 int total = 0; 1685 int res; 1686 1687 spin_lock_bh(&rcvq->lock); 1688 skb = __first_packet_length(sk, rcvq, &total); 1689 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1690 spin_lock(&sk_queue->lock); 1691 skb_queue_splice_tail_init(sk_queue, rcvq); 1692 spin_unlock(&sk_queue->lock); 1693 1694 skb = __first_packet_length(sk, rcvq, &total); 1695 } 1696 res = skb ? skb->len : -1; 1697 if (total) 1698 udp_rmem_release(sk, total, 1, false); 1699 spin_unlock_bh(&rcvq->lock); 1700 return res; 1701 } 1702 1703 /* 1704 * IOCTL requests applicable to the UDP protocol 1705 */ 1706 1707 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1708 { 1709 switch (cmd) { 1710 case SIOCOUTQ: 1711 { 1712 int amount = sk_wmem_alloc_get(sk); 1713 1714 return put_user(amount, (int __user *)arg); 1715 } 1716 1717 case SIOCINQ: 1718 { 1719 int amount = max_t(int, 0, first_packet_length(sk)); 1720 1721 return put_user(amount, (int __user *)arg); 1722 } 1723 1724 default: 1725 return -ENOIOCTLCMD; 1726 } 1727 1728 return 0; 1729 } 1730 EXPORT_SYMBOL(udp_ioctl); 1731 1732 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1733 int *off, int *err) 1734 { 1735 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1736 struct sk_buff_head *queue; 1737 struct sk_buff *last; 1738 long timeo; 1739 int error; 1740 1741 queue = &udp_sk(sk)->reader_queue; 1742 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1743 do { 1744 struct sk_buff *skb; 1745 1746 error = sock_error(sk); 1747 if (error) 1748 break; 1749 1750 error = -EAGAIN; 1751 do { 1752 spin_lock_bh(&queue->lock); 1753 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1754 err, &last); 1755 if (skb) { 1756 if (!(flags & MSG_PEEK)) 1757 udp_skb_destructor(sk, skb); 1758 spin_unlock_bh(&queue->lock); 1759 return skb; 1760 } 1761 1762 if (skb_queue_empty_lockless(sk_queue)) { 1763 spin_unlock_bh(&queue->lock); 1764 goto busy_check; 1765 } 1766 1767 /* refill the reader queue and walk it again 1768 * keep both queues locked to avoid re-acquiring 1769 * the sk_receive_queue lock if fwd memory scheduling 1770 * is needed. 1771 */ 1772 spin_lock(&sk_queue->lock); 1773 skb_queue_splice_tail_init(sk_queue, queue); 1774 1775 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1776 err, &last); 1777 if (skb && !(flags & MSG_PEEK)) 1778 udp_skb_dtor_locked(sk, skb); 1779 spin_unlock(&sk_queue->lock); 1780 spin_unlock_bh(&queue->lock); 1781 if (skb) 1782 return skb; 1783 1784 busy_check: 1785 if (!sk_can_busy_loop(sk)) 1786 break; 1787 1788 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1789 } while (!skb_queue_empty_lockless(sk_queue)); 1790 1791 /* sk_queue is empty, reader_queue may contain peeked packets */ 1792 } while (timeo && 1793 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1794 &error, &timeo, 1795 (struct sk_buff *)sk_queue)); 1796 1797 *err = error; 1798 return NULL; 1799 } 1800 EXPORT_SYMBOL(__skb_recv_udp); 1801 1802 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) 1803 { 1804 int copied = 0; 1805 1806 while (1) { 1807 struct sk_buff *skb; 1808 int err, used; 1809 1810 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1811 if (!skb) 1812 return err; 1813 1814 if (udp_lib_checksum_complete(skb)) { 1815 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1816 IS_UDPLITE(sk)); 1817 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1818 IS_UDPLITE(sk)); 1819 atomic_inc(&sk->sk_drops); 1820 kfree_skb(skb); 1821 continue; 1822 } 1823 1824 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); 1825 used = recv_actor(sk, skb); 1826 if (used <= 0) { 1827 if (!copied) 1828 copied = used; 1829 kfree_skb(skb); 1830 break; 1831 } else if (used <= skb->len) { 1832 copied += used; 1833 } 1834 1835 kfree_skb(skb); 1836 break; 1837 } 1838 1839 return copied; 1840 } 1841 EXPORT_SYMBOL(udp_read_skb); 1842 1843 /* 1844 * This should be easy, if there is something there we 1845 * return it, otherwise we block. 1846 */ 1847 1848 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 1849 int *addr_len) 1850 { 1851 struct inet_sock *inet = inet_sk(sk); 1852 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1853 struct sk_buff *skb; 1854 unsigned int ulen, copied; 1855 int off, err, peeking = flags & MSG_PEEK; 1856 int is_udplite = IS_UDPLITE(sk); 1857 bool checksum_valid = false; 1858 1859 if (flags & MSG_ERRQUEUE) 1860 return ip_recv_error(sk, msg, len, addr_len); 1861 1862 try_again: 1863 off = sk_peek_offset(sk, flags); 1864 skb = __skb_recv_udp(sk, flags, &off, &err); 1865 if (!skb) 1866 return err; 1867 1868 ulen = udp_skb_len(skb); 1869 copied = len; 1870 if (copied > ulen - off) 1871 copied = ulen - off; 1872 else if (copied < ulen) 1873 msg->msg_flags |= MSG_TRUNC; 1874 1875 /* 1876 * If checksum is needed at all, try to do it while copying the 1877 * data. If the data is truncated, or if we only want a partial 1878 * coverage checksum (UDP-Lite), do it before the copy. 1879 */ 1880 1881 if (copied < ulen || peeking || 1882 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1883 checksum_valid = udp_skb_csum_unnecessary(skb) || 1884 !__udp_lib_checksum_complete(skb); 1885 if (!checksum_valid) 1886 goto csum_copy_err; 1887 } 1888 1889 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1890 if (udp_skb_is_linear(skb)) 1891 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1892 else 1893 err = skb_copy_datagram_msg(skb, off, msg, copied); 1894 } else { 1895 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1896 1897 if (err == -EINVAL) 1898 goto csum_copy_err; 1899 } 1900 1901 if (unlikely(err)) { 1902 if (!peeking) { 1903 atomic_inc(&sk->sk_drops); 1904 UDP_INC_STATS(sock_net(sk), 1905 UDP_MIB_INERRORS, is_udplite); 1906 } 1907 kfree_skb(skb); 1908 return err; 1909 } 1910 1911 if (!peeking) 1912 UDP_INC_STATS(sock_net(sk), 1913 UDP_MIB_INDATAGRAMS, is_udplite); 1914 1915 sock_recv_cmsgs(msg, sk, skb); 1916 1917 /* Copy the address. */ 1918 if (sin) { 1919 sin->sin_family = AF_INET; 1920 sin->sin_port = udp_hdr(skb)->source; 1921 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1922 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1923 *addr_len = sizeof(*sin); 1924 1925 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1926 (struct sockaddr *)sin); 1927 } 1928 1929 if (udp_sk(sk)->gro_enabled) 1930 udp_cmsg_recv(msg, sk, skb); 1931 1932 if (inet->cmsg_flags) 1933 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1934 1935 err = copied; 1936 if (flags & MSG_TRUNC) 1937 err = ulen; 1938 1939 skb_consume_udp(sk, skb, peeking ? -err : err); 1940 return err; 1941 1942 csum_copy_err: 1943 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1944 udp_skb_destructor)) { 1945 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1946 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1947 } 1948 kfree_skb(skb); 1949 1950 /* starting over for a new packet, but check if we need to yield */ 1951 cond_resched(); 1952 msg->msg_flags &= ~MSG_TRUNC; 1953 goto try_again; 1954 } 1955 1956 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1957 { 1958 /* This check is replicated from __ip4_datagram_connect() and 1959 * intended to prevent BPF program called below from accessing bytes 1960 * that are out of the bound specified by user in addr_len. 1961 */ 1962 if (addr_len < sizeof(struct sockaddr_in)) 1963 return -EINVAL; 1964 1965 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1966 } 1967 EXPORT_SYMBOL(udp_pre_connect); 1968 1969 int __udp_disconnect(struct sock *sk, int flags) 1970 { 1971 struct inet_sock *inet = inet_sk(sk); 1972 /* 1973 * 1003.1g - break association. 1974 */ 1975 1976 sk->sk_state = TCP_CLOSE; 1977 inet->inet_daddr = 0; 1978 inet->inet_dport = 0; 1979 sock_rps_reset_rxhash(sk); 1980 sk->sk_bound_dev_if = 0; 1981 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1982 inet_reset_saddr(sk); 1983 if (sk->sk_prot->rehash && 1984 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1985 sk->sk_prot->rehash(sk); 1986 } 1987 1988 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1989 sk->sk_prot->unhash(sk); 1990 inet->inet_sport = 0; 1991 } 1992 sk_dst_reset(sk); 1993 return 0; 1994 } 1995 EXPORT_SYMBOL(__udp_disconnect); 1996 1997 int udp_disconnect(struct sock *sk, int flags) 1998 { 1999 lock_sock(sk); 2000 __udp_disconnect(sk, flags); 2001 release_sock(sk); 2002 return 0; 2003 } 2004 EXPORT_SYMBOL(udp_disconnect); 2005 2006 void udp_lib_unhash(struct sock *sk) 2007 { 2008 if (sk_hashed(sk)) { 2009 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2010 struct udp_hslot *hslot, *hslot2; 2011 2012 hslot = udp_hashslot(udptable, sock_net(sk), 2013 udp_sk(sk)->udp_port_hash); 2014 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2015 2016 spin_lock_bh(&hslot->lock); 2017 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2018 reuseport_detach_sock(sk); 2019 if (sk_del_node_init_rcu(sk)) { 2020 hslot->count--; 2021 inet_sk(sk)->inet_num = 0; 2022 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 2023 2024 spin_lock(&hslot2->lock); 2025 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2026 hslot2->count--; 2027 spin_unlock(&hslot2->lock); 2028 } 2029 spin_unlock_bh(&hslot->lock); 2030 } 2031 } 2032 EXPORT_SYMBOL(udp_lib_unhash); 2033 2034 /* 2035 * inet_rcv_saddr was changed, we must rehash secondary hash 2036 */ 2037 void udp_lib_rehash(struct sock *sk, u16 newhash) 2038 { 2039 if (sk_hashed(sk)) { 2040 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2041 struct udp_hslot *hslot, *hslot2, *nhslot2; 2042 2043 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2044 nhslot2 = udp_hashslot2(udptable, newhash); 2045 udp_sk(sk)->udp_portaddr_hash = newhash; 2046 2047 if (hslot2 != nhslot2 || 2048 rcu_access_pointer(sk->sk_reuseport_cb)) { 2049 hslot = udp_hashslot(udptable, sock_net(sk), 2050 udp_sk(sk)->udp_port_hash); 2051 /* we must lock primary chain too */ 2052 spin_lock_bh(&hslot->lock); 2053 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2054 reuseport_detach_sock(sk); 2055 2056 if (hslot2 != nhslot2) { 2057 spin_lock(&hslot2->lock); 2058 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2059 hslot2->count--; 2060 spin_unlock(&hslot2->lock); 2061 2062 spin_lock(&nhslot2->lock); 2063 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2064 &nhslot2->head); 2065 nhslot2->count++; 2066 spin_unlock(&nhslot2->lock); 2067 } 2068 2069 spin_unlock_bh(&hslot->lock); 2070 } 2071 } 2072 } 2073 EXPORT_SYMBOL(udp_lib_rehash); 2074 2075 void udp_v4_rehash(struct sock *sk) 2076 { 2077 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2078 inet_sk(sk)->inet_rcv_saddr, 2079 inet_sk(sk)->inet_num); 2080 udp_lib_rehash(sk, new_hash); 2081 } 2082 2083 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2084 { 2085 int rc; 2086 2087 if (inet_sk(sk)->inet_daddr) { 2088 sock_rps_save_rxhash(sk, skb); 2089 sk_mark_napi_id(sk, skb); 2090 sk_incoming_cpu_update(sk); 2091 } else { 2092 sk_mark_napi_id_once(sk, skb); 2093 } 2094 2095 rc = __udp_enqueue_schedule_skb(sk, skb); 2096 if (rc < 0) { 2097 int is_udplite = IS_UDPLITE(sk); 2098 int drop_reason; 2099 2100 /* Note that an ENOMEM error is charged twice */ 2101 if (rc == -ENOMEM) { 2102 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2103 is_udplite); 2104 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 2105 } else { 2106 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2107 is_udplite); 2108 drop_reason = SKB_DROP_REASON_PROTO_MEM; 2109 } 2110 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2111 kfree_skb_reason(skb, drop_reason); 2112 trace_udp_fail_queue_rcv_skb(rc, sk); 2113 return -1; 2114 } 2115 2116 return 0; 2117 } 2118 2119 /* returns: 2120 * -1: error 2121 * 0: success 2122 * >0: "udp encap" protocol resubmission 2123 * 2124 * Note that in the success and error cases, the skb is assumed to 2125 * have either been requeued or freed. 2126 */ 2127 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2128 { 2129 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2130 struct udp_sock *up = udp_sk(sk); 2131 int is_udplite = IS_UDPLITE(sk); 2132 2133 /* 2134 * Charge it to the socket, dropping if the queue is full. 2135 */ 2136 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 2137 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2138 goto drop; 2139 } 2140 nf_reset_ct(skb); 2141 2142 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 2143 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2144 2145 /* 2146 * This is an encapsulation socket so pass the skb to 2147 * the socket's udp_encap_rcv() hook. Otherwise, just 2148 * fall through and pass this up the UDP socket. 2149 * up->encap_rcv() returns the following value: 2150 * =0 if skb was successfully passed to the encap 2151 * handler or was discarded by it. 2152 * >0 if skb should be passed on to UDP. 2153 * <0 if skb should be resubmitted as proto -N 2154 */ 2155 2156 /* if we're overly short, let UDP handle it */ 2157 encap_rcv = READ_ONCE(up->encap_rcv); 2158 if (encap_rcv) { 2159 int ret; 2160 2161 /* Verify checksum before giving to encap */ 2162 if (udp_lib_checksum_complete(skb)) 2163 goto csum_error; 2164 2165 ret = encap_rcv(sk, skb); 2166 if (ret <= 0) { 2167 __UDP_INC_STATS(sock_net(sk), 2168 UDP_MIB_INDATAGRAMS, 2169 is_udplite); 2170 return -ret; 2171 } 2172 } 2173 2174 /* FALLTHROUGH -- it's a UDP Packet */ 2175 } 2176 2177 /* 2178 * UDP-Lite specific tests, ignored on UDP sockets 2179 */ 2180 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2181 2182 /* 2183 * MIB statistics other than incrementing the error count are 2184 * disabled for the following two types of errors: these depend 2185 * on the application settings, not on the functioning of the 2186 * protocol stack as such. 2187 * 2188 * RFC 3828 here recommends (sec 3.3): "There should also be a 2189 * way ... to ... at least let the receiving application block 2190 * delivery of packets with coverage values less than a value 2191 * provided by the application." 2192 */ 2193 if (up->pcrlen == 0) { /* full coverage was set */ 2194 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2195 UDP_SKB_CB(skb)->cscov, skb->len); 2196 goto drop; 2197 } 2198 /* The next case involves violating the min. coverage requested 2199 * by the receiver. This is subtle: if receiver wants x and x is 2200 * greater than the buffersize/MTU then receiver will complain 2201 * that it wants x while sender emits packets of smaller size y. 2202 * Therefore the above ...()->partial_cov statement is essential. 2203 */ 2204 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2205 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2206 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2207 goto drop; 2208 } 2209 } 2210 2211 prefetch(&sk->sk_rmem_alloc); 2212 if (rcu_access_pointer(sk->sk_filter) && 2213 udp_lib_checksum_complete(skb)) 2214 goto csum_error; 2215 2216 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 2217 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 2218 goto drop; 2219 } 2220 2221 udp_csum_pull_header(skb); 2222 2223 ipv4_pktinfo_prepare(sk, skb); 2224 return __udp_queue_rcv_skb(sk, skb); 2225 2226 csum_error: 2227 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2228 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2229 drop: 2230 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2231 atomic_inc(&sk->sk_drops); 2232 kfree_skb_reason(skb, drop_reason); 2233 return -1; 2234 } 2235 2236 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2237 { 2238 struct sk_buff *next, *segs; 2239 int ret; 2240 2241 if (likely(!udp_unexpected_gso(sk, skb))) 2242 return udp_queue_rcv_one_skb(sk, skb); 2243 2244 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2245 __skb_push(skb, -skb_mac_offset(skb)); 2246 segs = udp_rcv_segment(sk, skb, true); 2247 skb_list_walk_safe(segs, skb, next) { 2248 __skb_pull(skb, skb_transport_offset(skb)); 2249 2250 udp_post_segment_fix_csum(skb); 2251 ret = udp_queue_rcv_one_skb(sk, skb); 2252 if (ret > 0) 2253 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2254 } 2255 return 0; 2256 } 2257 2258 /* For TCP sockets, sk_rx_dst is protected by socket lock 2259 * For UDP, we use xchg() to guard against concurrent changes. 2260 */ 2261 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2262 { 2263 struct dst_entry *old; 2264 2265 if (dst_hold_safe(dst)) { 2266 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); 2267 dst_release(old); 2268 return old != dst; 2269 } 2270 return false; 2271 } 2272 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2273 2274 /* 2275 * Multicasts and broadcasts go to each listener. 2276 * 2277 * Note: called only from the BH handler context. 2278 */ 2279 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2280 struct udphdr *uh, 2281 __be32 saddr, __be32 daddr, 2282 struct udp_table *udptable, 2283 int proto) 2284 { 2285 struct sock *sk, *first = NULL; 2286 unsigned short hnum = ntohs(uh->dest); 2287 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2288 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2289 unsigned int offset = offsetof(typeof(*sk), sk_node); 2290 int dif = skb->dev->ifindex; 2291 int sdif = inet_sdif(skb); 2292 struct hlist_node *node; 2293 struct sk_buff *nskb; 2294 2295 if (use_hash2) { 2296 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2297 udptable->mask; 2298 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2299 start_lookup: 2300 hslot = &udptable->hash2[hash2]; 2301 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2302 } 2303 2304 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2305 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2306 uh->source, saddr, dif, sdif, hnum)) 2307 continue; 2308 2309 if (!first) { 2310 first = sk; 2311 continue; 2312 } 2313 nskb = skb_clone(skb, GFP_ATOMIC); 2314 2315 if (unlikely(!nskb)) { 2316 atomic_inc(&sk->sk_drops); 2317 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2318 IS_UDPLITE(sk)); 2319 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2320 IS_UDPLITE(sk)); 2321 continue; 2322 } 2323 if (udp_queue_rcv_skb(sk, nskb) > 0) 2324 consume_skb(nskb); 2325 } 2326 2327 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2328 if (use_hash2 && hash2 != hash2_any) { 2329 hash2 = hash2_any; 2330 goto start_lookup; 2331 } 2332 2333 if (first) { 2334 if (udp_queue_rcv_skb(first, skb) > 0) 2335 consume_skb(skb); 2336 } else { 2337 kfree_skb(skb); 2338 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2339 proto == IPPROTO_UDPLITE); 2340 } 2341 return 0; 2342 } 2343 2344 /* Initialize UDP checksum. If exited with zero value (success), 2345 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2346 * Otherwise, csum completion requires checksumming packet body, 2347 * including udp header and folding it to skb->csum. 2348 */ 2349 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2350 int proto) 2351 { 2352 int err; 2353 2354 UDP_SKB_CB(skb)->partial_cov = 0; 2355 UDP_SKB_CB(skb)->cscov = skb->len; 2356 2357 if (proto == IPPROTO_UDPLITE) { 2358 err = udplite_checksum_init(skb, uh); 2359 if (err) 2360 return err; 2361 2362 if (UDP_SKB_CB(skb)->partial_cov) { 2363 skb->csum = inet_compute_pseudo(skb, proto); 2364 return 0; 2365 } 2366 } 2367 2368 /* Note, we are only interested in != 0 or == 0, thus the 2369 * force to int. 2370 */ 2371 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2372 inet_compute_pseudo); 2373 if (err) 2374 return err; 2375 2376 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2377 /* If SW calculated the value, we know it's bad */ 2378 if (skb->csum_complete_sw) 2379 return 1; 2380 2381 /* HW says the value is bad. Let's validate that. 2382 * skb->csum is no longer the full packet checksum, 2383 * so don't treat it as such. 2384 */ 2385 skb_checksum_complete_unset(skb); 2386 } 2387 2388 return 0; 2389 } 2390 2391 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2392 * return code conversion for ip layer consumption 2393 */ 2394 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2395 struct udphdr *uh) 2396 { 2397 int ret; 2398 2399 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2400 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2401 2402 ret = udp_queue_rcv_skb(sk, skb); 2403 2404 /* a return value > 0 means to resubmit the input, but 2405 * it wants the return to be -protocol, or 0 2406 */ 2407 if (ret > 0) 2408 return -ret; 2409 return 0; 2410 } 2411 2412 /* 2413 * All we need to do is get the socket, and then do a checksum. 2414 */ 2415 2416 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2417 int proto) 2418 { 2419 struct sock *sk; 2420 struct udphdr *uh; 2421 unsigned short ulen; 2422 struct rtable *rt = skb_rtable(skb); 2423 __be32 saddr, daddr; 2424 struct net *net = dev_net(skb->dev); 2425 bool refcounted; 2426 int drop_reason; 2427 2428 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2429 2430 /* 2431 * Validate the packet. 2432 */ 2433 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2434 goto drop; /* No space for header. */ 2435 2436 uh = udp_hdr(skb); 2437 ulen = ntohs(uh->len); 2438 saddr = ip_hdr(skb)->saddr; 2439 daddr = ip_hdr(skb)->daddr; 2440 2441 if (ulen > skb->len) 2442 goto short_packet; 2443 2444 if (proto == IPPROTO_UDP) { 2445 /* UDP validates ulen. */ 2446 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2447 goto short_packet; 2448 uh = udp_hdr(skb); 2449 } 2450 2451 if (udp4_csum_init(skb, uh, proto)) 2452 goto csum_error; 2453 2454 sk = skb_steal_sock(skb, &refcounted); 2455 if (sk) { 2456 struct dst_entry *dst = skb_dst(skb); 2457 int ret; 2458 2459 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2460 udp_sk_rx_dst_set(sk, dst); 2461 2462 ret = udp_unicast_rcv_skb(sk, skb, uh); 2463 if (refcounted) 2464 sock_put(sk); 2465 return ret; 2466 } 2467 2468 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2469 return __udp4_lib_mcast_deliver(net, skb, uh, 2470 saddr, daddr, udptable, proto); 2471 2472 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2473 if (sk) 2474 return udp_unicast_rcv_skb(sk, skb, uh); 2475 2476 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2477 goto drop; 2478 nf_reset_ct(skb); 2479 2480 /* No socket. Drop packet silently, if checksum is wrong */ 2481 if (udp_lib_checksum_complete(skb)) 2482 goto csum_error; 2483 2484 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2485 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2486 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2487 2488 /* 2489 * Hmm. We got an UDP packet to a port to which we 2490 * don't wanna listen. Ignore it. 2491 */ 2492 kfree_skb_reason(skb, drop_reason); 2493 return 0; 2494 2495 short_packet: 2496 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2497 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2498 proto == IPPROTO_UDPLITE ? "Lite" : "", 2499 &saddr, ntohs(uh->source), 2500 ulen, skb->len, 2501 &daddr, ntohs(uh->dest)); 2502 goto drop; 2503 2504 csum_error: 2505 /* 2506 * RFC1122: OK. Discards the bad packet silently (as far as 2507 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2508 */ 2509 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2510 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2511 proto == IPPROTO_UDPLITE ? "Lite" : "", 2512 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2513 ulen); 2514 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2515 drop: 2516 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2517 kfree_skb_reason(skb, drop_reason); 2518 return 0; 2519 } 2520 2521 /* We can only early demux multicast if there is a single matching socket. 2522 * If more than one socket found returns NULL 2523 */ 2524 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2525 __be16 loc_port, __be32 loc_addr, 2526 __be16 rmt_port, __be32 rmt_addr, 2527 int dif, int sdif) 2528 { 2529 struct sock *sk, *result; 2530 unsigned short hnum = ntohs(loc_port); 2531 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2532 struct udp_hslot *hslot = &udp_table.hash[slot]; 2533 2534 /* Do not bother scanning a too big list */ 2535 if (hslot->count > 10) 2536 return NULL; 2537 2538 result = NULL; 2539 sk_for_each_rcu(sk, &hslot->head) { 2540 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2541 rmt_port, rmt_addr, dif, sdif, hnum)) { 2542 if (result) 2543 return NULL; 2544 result = sk; 2545 } 2546 } 2547 2548 return result; 2549 } 2550 2551 /* For unicast we should only early demux connected sockets or we can 2552 * break forwarding setups. The chains here can be long so only check 2553 * if the first socket is an exact match and if not move on. 2554 */ 2555 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2556 __be16 loc_port, __be32 loc_addr, 2557 __be16 rmt_port, __be32 rmt_addr, 2558 int dif, int sdif) 2559 { 2560 unsigned short hnum = ntohs(loc_port); 2561 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2562 unsigned int slot2 = hash2 & udp_table.mask; 2563 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2564 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2565 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2566 struct sock *sk; 2567 2568 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2569 if (inet_match(net, sk, acookie, ports, dif, sdif)) 2570 return sk; 2571 /* Only check first socket in chain */ 2572 break; 2573 } 2574 return NULL; 2575 } 2576 2577 int udp_v4_early_demux(struct sk_buff *skb) 2578 { 2579 struct net *net = dev_net(skb->dev); 2580 struct in_device *in_dev = NULL; 2581 const struct iphdr *iph; 2582 const struct udphdr *uh; 2583 struct sock *sk = NULL; 2584 struct dst_entry *dst; 2585 int dif = skb->dev->ifindex; 2586 int sdif = inet_sdif(skb); 2587 int ours; 2588 2589 /* validate the packet */ 2590 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2591 return 0; 2592 2593 iph = ip_hdr(skb); 2594 uh = udp_hdr(skb); 2595 2596 if (skb->pkt_type == PACKET_MULTICAST) { 2597 in_dev = __in_dev_get_rcu(skb->dev); 2598 2599 if (!in_dev) 2600 return 0; 2601 2602 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2603 iph->protocol); 2604 if (!ours) 2605 return 0; 2606 2607 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2608 uh->source, iph->saddr, 2609 dif, sdif); 2610 } else if (skb->pkt_type == PACKET_HOST) { 2611 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2612 uh->source, iph->saddr, dif, sdif); 2613 } 2614 2615 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2616 return 0; 2617 2618 skb->sk = sk; 2619 skb->destructor = sock_efree; 2620 dst = rcu_dereference(sk->sk_rx_dst); 2621 2622 if (dst) 2623 dst = dst_check(dst, 0); 2624 if (dst) { 2625 u32 itag = 0; 2626 2627 /* set noref for now. 2628 * any place which wants to hold dst has to call 2629 * dst_hold_safe() 2630 */ 2631 skb_dst_set_noref(skb, dst); 2632 2633 /* for unconnected multicast sockets we need to validate 2634 * the source on each packet 2635 */ 2636 if (!inet_sk(sk)->inet_daddr && in_dev) 2637 return ip_mc_validate_source(skb, iph->daddr, 2638 iph->saddr, 2639 iph->tos & IPTOS_RT_MASK, 2640 skb->dev, in_dev, &itag); 2641 } 2642 return 0; 2643 } 2644 2645 int udp_rcv(struct sk_buff *skb) 2646 { 2647 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2648 } 2649 2650 void udp_destroy_sock(struct sock *sk) 2651 { 2652 struct udp_sock *up = udp_sk(sk); 2653 bool slow = lock_sock_fast(sk); 2654 2655 /* protects from races with udp_abort() */ 2656 sock_set_flag(sk, SOCK_DEAD); 2657 udp_flush_pending_frames(sk); 2658 unlock_sock_fast(sk, slow); 2659 if (static_branch_unlikely(&udp_encap_needed_key)) { 2660 if (up->encap_type) { 2661 void (*encap_destroy)(struct sock *sk); 2662 encap_destroy = READ_ONCE(up->encap_destroy); 2663 if (encap_destroy) 2664 encap_destroy(sk); 2665 } 2666 if (up->encap_enabled) 2667 static_branch_dec(&udp_encap_needed_key); 2668 } 2669 } 2670 2671 /* 2672 * Socket option code for UDP 2673 */ 2674 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2675 sockptr_t optval, unsigned int optlen, 2676 int (*push_pending_frames)(struct sock *)) 2677 { 2678 struct udp_sock *up = udp_sk(sk); 2679 int val, valbool; 2680 int err = 0; 2681 int is_udplite = IS_UDPLITE(sk); 2682 2683 if (optlen < sizeof(int)) 2684 return -EINVAL; 2685 2686 if (copy_from_sockptr(&val, optval, sizeof(val))) 2687 return -EFAULT; 2688 2689 valbool = val ? 1 : 0; 2690 2691 switch (optname) { 2692 case UDP_CORK: 2693 if (val != 0) { 2694 WRITE_ONCE(up->corkflag, 1); 2695 } else { 2696 WRITE_ONCE(up->corkflag, 0); 2697 lock_sock(sk); 2698 push_pending_frames(sk); 2699 release_sock(sk); 2700 } 2701 break; 2702 2703 case UDP_ENCAP: 2704 switch (val) { 2705 case 0: 2706 #ifdef CONFIG_XFRM 2707 case UDP_ENCAP_ESPINUDP: 2708 case UDP_ENCAP_ESPINUDP_NON_IKE: 2709 #if IS_ENABLED(CONFIG_IPV6) 2710 if (sk->sk_family == AF_INET6) 2711 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; 2712 else 2713 #endif 2714 up->encap_rcv = xfrm4_udp_encap_rcv; 2715 #endif 2716 fallthrough; 2717 case UDP_ENCAP_L2TPINUDP: 2718 up->encap_type = val; 2719 lock_sock(sk); 2720 udp_tunnel_encap_enable(sk->sk_socket); 2721 release_sock(sk); 2722 break; 2723 default: 2724 err = -ENOPROTOOPT; 2725 break; 2726 } 2727 break; 2728 2729 case UDP_NO_CHECK6_TX: 2730 up->no_check6_tx = valbool; 2731 break; 2732 2733 case UDP_NO_CHECK6_RX: 2734 up->no_check6_rx = valbool; 2735 break; 2736 2737 case UDP_SEGMENT: 2738 if (val < 0 || val > USHRT_MAX) 2739 return -EINVAL; 2740 WRITE_ONCE(up->gso_size, val); 2741 break; 2742 2743 case UDP_GRO: 2744 lock_sock(sk); 2745 2746 /* when enabling GRO, accept the related GSO packet type */ 2747 if (valbool) 2748 udp_tunnel_encap_enable(sk->sk_socket); 2749 up->gro_enabled = valbool; 2750 up->accept_udp_l4 = valbool; 2751 release_sock(sk); 2752 break; 2753 2754 /* 2755 * UDP-Lite's partial checksum coverage (RFC 3828). 2756 */ 2757 /* The sender sets actual checksum coverage length via this option. 2758 * The case coverage > packet length is handled by send module. */ 2759 case UDPLITE_SEND_CSCOV: 2760 if (!is_udplite) /* Disable the option on UDP sockets */ 2761 return -ENOPROTOOPT; 2762 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2763 val = 8; 2764 else if (val > USHRT_MAX) 2765 val = USHRT_MAX; 2766 up->pcslen = val; 2767 up->pcflag |= UDPLITE_SEND_CC; 2768 break; 2769 2770 /* The receiver specifies a minimum checksum coverage value. To make 2771 * sense, this should be set to at least 8 (as done below). If zero is 2772 * used, this again means full checksum coverage. */ 2773 case UDPLITE_RECV_CSCOV: 2774 if (!is_udplite) /* Disable the option on UDP sockets */ 2775 return -ENOPROTOOPT; 2776 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2777 val = 8; 2778 else if (val > USHRT_MAX) 2779 val = USHRT_MAX; 2780 up->pcrlen = val; 2781 up->pcflag |= UDPLITE_RECV_CC; 2782 break; 2783 2784 default: 2785 err = -ENOPROTOOPT; 2786 break; 2787 } 2788 2789 return err; 2790 } 2791 EXPORT_SYMBOL(udp_lib_setsockopt); 2792 2793 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2794 unsigned int optlen) 2795 { 2796 if (level == SOL_UDP || level == SOL_UDPLITE) 2797 return udp_lib_setsockopt(sk, level, optname, 2798 optval, optlen, 2799 udp_push_pending_frames); 2800 return ip_setsockopt(sk, level, optname, optval, optlen); 2801 } 2802 2803 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2804 char __user *optval, int __user *optlen) 2805 { 2806 struct udp_sock *up = udp_sk(sk); 2807 int val, len; 2808 2809 if (get_user(len, optlen)) 2810 return -EFAULT; 2811 2812 len = min_t(unsigned int, len, sizeof(int)); 2813 2814 if (len < 0) 2815 return -EINVAL; 2816 2817 switch (optname) { 2818 case UDP_CORK: 2819 val = READ_ONCE(up->corkflag); 2820 break; 2821 2822 case UDP_ENCAP: 2823 val = up->encap_type; 2824 break; 2825 2826 case UDP_NO_CHECK6_TX: 2827 val = up->no_check6_tx; 2828 break; 2829 2830 case UDP_NO_CHECK6_RX: 2831 val = up->no_check6_rx; 2832 break; 2833 2834 case UDP_SEGMENT: 2835 val = READ_ONCE(up->gso_size); 2836 break; 2837 2838 case UDP_GRO: 2839 val = up->gro_enabled; 2840 break; 2841 2842 /* The following two cannot be changed on UDP sockets, the return is 2843 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2844 case UDPLITE_SEND_CSCOV: 2845 val = up->pcslen; 2846 break; 2847 2848 case UDPLITE_RECV_CSCOV: 2849 val = up->pcrlen; 2850 break; 2851 2852 default: 2853 return -ENOPROTOOPT; 2854 } 2855 2856 if (put_user(len, optlen)) 2857 return -EFAULT; 2858 if (copy_to_user(optval, &val, len)) 2859 return -EFAULT; 2860 return 0; 2861 } 2862 EXPORT_SYMBOL(udp_lib_getsockopt); 2863 2864 int udp_getsockopt(struct sock *sk, int level, int optname, 2865 char __user *optval, int __user *optlen) 2866 { 2867 if (level == SOL_UDP || level == SOL_UDPLITE) 2868 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2869 return ip_getsockopt(sk, level, optname, optval, optlen); 2870 } 2871 2872 /** 2873 * udp_poll - wait for a UDP event. 2874 * @file: - file struct 2875 * @sock: - socket 2876 * @wait: - poll table 2877 * 2878 * This is same as datagram poll, except for the special case of 2879 * blocking sockets. If application is using a blocking fd 2880 * and a packet with checksum error is in the queue; 2881 * then it could get return from select indicating data available 2882 * but then block when reading it. Add special case code 2883 * to work around these arguably broken applications. 2884 */ 2885 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2886 { 2887 __poll_t mask = datagram_poll(file, sock, wait); 2888 struct sock *sk = sock->sk; 2889 2890 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2891 mask |= EPOLLIN | EPOLLRDNORM; 2892 2893 /* Check for false positives due to checksum errors */ 2894 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2895 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2896 mask &= ~(EPOLLIN | EPOLLRDNORM); 2897 2898 /* psock ingress_msg queue should not contain any bad checksum frames */ 2899 if (sk_is_readable(sk)) 2900 mask |= EPOLLIN | EPOLLRDNORM; 2901 return mask; 2902 2903 } 2904 EXPORT_SYMBOL(udp_poll); 2905 2906 int udp_abort(struct sock *sk, int err) 2907 { 2908 lock_sock(sk); 2909 2910 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2911 * with close() 2912 */ 2913 if (sock_flag(sk, SOCK_DEAD)) 2914 goto out; 2915 2916 sk->sk_err = err; 2917 sk_error_report(sk); 2918 __udp_disconnect(sk, 0); 2919 2920 out: 2921 release_sock(sk); 2922 2923 return 0; 2924 } 2925 EXPORT_SYMBOL_GPL(udp_abort); 2926 2927 struct proto udp_prot = { 2928 .name = "UDP", 2929 .owner = THIS_MODULE, 2930 .close = udp_lib_close, 2931 .pre_connect = udp_pre_connect, 2932 .connect = ip4_datagram_connect, 2933 .disconnect = udp_disconnect, 2934 .ioctl = udp_ioctl, 2935 .init = udp_init_sock, 2936 .destroy = udp_destroy_sock, 2937 .setsockopt = udp_setsockopt, 2938 .getsockopt = udp_getsockopt, 2939 .sendmsg = udp_sendmsg, 2940 .recvmsg = udp_recvmsg, 2941 .sendpage = udp_sendpage, 2942 .release_cb = ip4_datagram_release_cb, 2943 .hash = udp_lib_hash, 2944 .unhash = udp_lib_unhash, 2945 .rehash = udp_v4_rehash, 2946 .get_port = udp_v4_get_port, 2947 .put_port = udp_lib_unhash, 2948 #ifdef CONFIG_BPF_SYSCALL 2949 .psock_update_sk_prot = udp_bpf_update_proto, 2950 #endif 2951 .memory_allocated = &udp_memory_allocated, 2952 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 2953 2954 .sysctl_mem = sysctl_udp_mem, 2955 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2956 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2957 .obj_size = sizeof(struct udp_sock), 2958 .h.udp_table = &udp_table, 2959 .diag_destroy = udp_abort, 2960 }; 2961 EXPORT_SYMBOL(udp_prot); 2962 2963 /* ------------------------------------------------------------------------ */ 2964 #ifdef CONFIG_PROC_FS 2965 2966 static struct sock *udp_get_first(struct seq_file *seq, int start) 2967 { 2968 struct sock *sk; 2969 struct udp_seq_afinfo *afinfo; 2970 struct udp_iter_state *state = seq->private; 2971 struct net *net = seq_file_net(seq); 2972 2973 if (state->bpf_seq_afinfo) 2974 afinfo = state->bpf_seq_afinfo; 2975 else 2976 afinfo = pde_data(file_inode(seq->file)); 2977 2978 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; 2979 ++state->bucket) { 2980 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; 2981 2982 if (hlist_empty(&hslot->head)) 2983 continue; 2984 2985 spin_lock_bh(&hslot->lock); 2986 sk_for_each(sk, &hslot->head) { 2987 if (!net_eq(sock_net(sk), net)) 2988 continue; 2989 if (afinfo->family == AF_UNSPEC || 2990 sk->sk_family == afinfo->family) 2991 goto found; 2992 } 2993 spin_unlock_bh(&hslot->lock); 2994 } 2995 sk = NULL; 2996 found: 2997 return sk; 2998 } 2999 3000 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 3001 { 3002 struct udp_seq_afinfo *afinfo; 3003 struct udp_iter_state *state = seq->private; 3004 struct net *net = seq_file_net(seq); 3005 3006 if (state->bpf_seq_afinfo) 3007 afinfo = state->bpf_seq_afinfo; 3008 else 3009 afinfo = pde_data(file_inode(seq->file)); 3010 3011 do { 3012 sk = sk_next(sk); 3013 } while (sk && (!net_eq(sock_net(sk), net) || 3014 (afinfo->family != AF_UNSPEC && 3015 sk->sk_family != afinfo->family))); 3016 3017 if (!sk) { 3018 if (state->bucket <= afinfo->udp_table->mask) 3019 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 3020 return udp_get_first(seq, state->bucket + 1); 3021 } 3022 return sk; 3023 } 3024 3025 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3026 { 3027 struct sock *sk = udp_get_first(seq, 0); 3028 3029 if (sk) 3030 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3031 --pos; 3032 return pos ? NULL : sk; 3033 } 3034 3035 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3036 { 3037 struct udp_iter_state *state = seq->private; 3038 state->bucket = MAX_UDP_PORTS; 3039 3040 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3041 } 3042 EXPORT_SYMBOL(udp_seq_start); 3043 3044 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3045 { 3046 struct sock *sk; 3047 3048 if (v == SEQ_START_TOKEN) 3049 sk = udp_get_idx(seq, 0); 3050 else 3051 sk = udp_get_next(seq, v); 3052 3053 ++*pos; 3054 return sk; 3055 } 3056 EXPORT_SYMBOL(udp_seq_next); 3057 3058 void udp_seq_stop(struct seq_file *seq, void *v) 3059 { 3060 struct udp_seq_afinfo *afinfo; 3061 struct udp_iter_state *state = seq->private; 3062 3063 if (state->bpf_seq_afinfo) 3064 afinfo = state->bpf_seq_afinfo; 3065 else 3066 afinfo = pde_data(file_inode(seq->file)); 3067 3068 if (state->bucket <= afinfo->udp_table->mask) 3069 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 3070 } 3071 EXPORT_SYMBOL(udp_seq_stop); 3072 3073 /* ------------------------------------------------------------------------ */ 3074 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3075 int bucket) 3076 { 3077 struct inet_sock *inet = inet_sk(sp); 3078 __be32 dest = inet->inet_daddr; 3079 __be32 src = inet->inet_rcv_saddr; 3080 __u16 destp = ntohs(inet->inet_dport); 3081 __u16 srcp = ntohs(inet->inet_sport); 3082 3083 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3084 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3085 bucket, src, srcp, dest, destp, sp->sk_state, 3086 sk_wmem_alloc_get(sp), 3087 udp_rqueue_get(sp), 3088 0, 0L, 0, 3089 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3090 0, sock_i_ino(sp), 3091 refcount_read(&sp->sk_refcnt), sp, 3092 atomic_read(&sp->sk_drops)); 3093 } 3094 3095 int udp4_seq_show(struct seq_file *seq, void *v) 3096 { 3097 seq_setwidth(seq, 127); 3098 if (v == SEQ_START_TOKEN) 3099 seq_puts(seq, " sl local_address rem_address st tx_queue " 3100 "rx_queue tr tm->when retrnsmt uid timeout " 3101 "inode ref pointer drops"); 3102 else { 3103 struct udp_iter_state *state = seq->private; 3104 3105 udp4_format_sock(v, seq, state->bucket); 3106 } 3107 seq_pad(seq, '\n'); 3108 return 0; 3109 } 3110 3111 #ifdef CONFIG_BPF_SYSCALL 3112 struct bpf_iter__udp { 3113 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3114 __bpf_md_ptr(struct udp_sock *, udp_sk); 3115 uid_t uid __aligned(8); 3116 int bucket __aligned(8); 3117 }; 3118 3119 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3120 struct udp_sock *udp_sk, uid_t uid, int bucket) 3121 { 3122 struct bpf_iter__udp ctx; 3123 3124 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3125 ctx.meta = meta; 3126 ctx.udp_sk = udp_sk; 3127 ctx.uid = uid; 3128 ctx.bucket = bucket; 3129 return bpf_iter_run_prog(prog, &ctx); 3130 } 3131 3132 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3133 { 3134 struct udp_iter_state *state = seq->private; 3135 struct bpf_iter_meta meta; 3136 struct bpf_prog *prog; 3137 struct sock *sk = v; 3138 uid_t uid; 3139 3140 if (v == SEQ_START_TOKEN) 3141 return 0; 3142 3143 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3144 meta.seq = seq; 3145 prog = bpf_iter_get_info(&meta, false); 3146 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3147 } 3148 3149 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3150 { 3151 struct bpf_iter_meta meta; 3152 struct bpf_prog *prog; 3153 3154 if (!v) { 3155 meta.seq = seq; 3156 prog = bpf_iter_get_info(&meta, true); 3157 if (prog) 3158 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3159 } 3160 3161 udp_seq_stop(seq, v); 3162 } 3163 3164 static const struct seq_operations bpf_iter_udp_seq_ops = { 3165 .start = udp_seq_start, 3166 .next = udp_seq_next, 3167 .stop = bpf_iter_udp_seq_stop, 3168 .show = bpf_iter_udp_seq_show, 3169 }; 3170 #endif 3171 3172 const struct seq_operations udp_seq_ops = { 3173 .start = udp_seq_start, 3174 .next = udp_seq_next, 3175 .stop = udp_seq_stop, 3176 .show = udp4_seq_show, 3177 }; 3178 EXPORT_SYMBOL(udp_seq_ops); 3179 3180 static struct udp_seq_afinfo udp4_seq_afinfo = { 3181 .family = AF_INET, 3182 .udp_table = &udp_table, 3183 }; 3184 3185 static int __net_init udp4_proc_init_net(struct net *net) 3186 { 3187 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3188 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3189 return -ENOMEM; 3190 return 0; 3191 } 3192 3193 static void __net_exit udp4_proc_exit_net(struct net *net) 3194 { 3195 remove_proc_entry("udp", net->proc_net); 3196 } 3197 3198 static struct pernet_operations udp4_net_ops = { 3199 .init = udp4_proc_init_net, 3200 .exit = udp4_proc_exit_net, 3201 }; 3202 3203 int __init udp4_proc_init(void) 3204 { 3205 return register_pernet_subsys(&udp4_net_ops); 3206 } 3207 3208 void udp4_proc_exit(void) 3209 { 3210 unregister_pernet_subsys(&udp4_net_ops); 3211 } 3212 #endif /* CONFIG_PROC_FS */ 3213 3214 static __initdata unsigned long uhash_entries; 3215 static int __init set_uhash_entries(char *str) 3216 { 3217 ssize_t ret; 3218 3219 if (!str) 3220 return 0; 3221 3222 ret = kstrtoul(str, 0, &uhash_entries); 3223 if (ret) 3224 return 0; 3225 3226 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3227 uhash_entries = UDP_HTABLE_SIZE_MIN; 3228 return 1; 3229 } 3230 __setup("uhash_entries=", set_uhash_entries); 3231 3232 void __init udp_table_init(struct udp_table *table, const char *name) 3233 { 3234 unsigned int i; 3235 3236 table->hash = alloc_large_system_hash(name, 3237 2 * sizeof(struct udp_hslot), 3238 uhash_entries, 3239 21, /* one slot per 2 MB */ 3240 0, 3241 &table->log, 3242 &table->mask, 3243 UDP_HTABLE_SIZE_MIN, 3244 64 * 1024); 3245 3246 table->hash2 = table->hash + (table->mask + 1); 3247 for (i = 0; i <= table->mask; i++) { 3248 INIT_HLIST_HEAD(&table->hash[i].head); 3249 table->hash[i].count = 0; 3250 spin_lock_init(&table->hash[i].lock); 3251 } 3252 for (i = 0; i <= table->mask; i++) { 3253 INIT_HLIST_HEAD(&table->hash2[i].head); 3254 table->hash2[i].count = 0; 3255 spin_lock_init(&table->hash2[i].lock); 3256 } 3257 } 3258 3259 u32 udp_flow_hashrnd(void) 3260 { 3261 static u32 hashrnd __read_mostly; 3262 3263 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3264 3265 return hashrnd; 3266 } 3267 EXPORT_SYMBOL(udp_flow_hashrnd); 3268 3269 static int __net_init udp_sysctl_init(struct net *net) 3270 { 3271 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3272 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 3273 3274 #ifdef CONFIG_NET_L3_MASTER_DEV 3275 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3276 #endif 3277 3278 return 0; 3279 } 3280 3281 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3282 .init = udp_sysctl_init, 3283 }; 3284 3285 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3286 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3287 struct udp_sock *udp_sk, uid_t uid, int bucket) 3288 3289 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3290 { 3291 struct udp_iter_state *st = priv_data; 3292 struct udp_seq_afinfo *afinfo; 3293 int ret; 3294 3295 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN); 3296 if (!afinfo) 3297 return -ENOMEM; 3298 3299 afinfo->family = AF_UNSPEC; 3300 afinfo->udp_table = &udp_table; 3301 st->bpf_seq_afinfo = afinfo; 3302 ret = bpf_iter_init_seq_net(priv_data, aux); 3303 if (ret) 3304 kfree(afinfo); 3305 return ret; 3306 } 3307 3308 static void bpf_iter_fini_udp(void *priv_data) 3309 { 3310 struct udp_iter_state *st = priv_data; 3311 3312 kfree(st->bpf_seq_afinfo); 3313 bpf_iter_fini_seq_net(priv_data); 3314 } 3315 3316 static const struct bpf_iter_seq_info udp_seq_info = { 3317 .seq_ops = &bpf_iter_udp_seq_ops, 3318 .init_seq_private = bpf_iter_init_udp, 3319 .fini_seq_private = bpf_iter_fini_udp, 3320 .seq_priv_size = sizeof(struct udp_iter_state), 3321 }; 3322 3323 static struct bpf_iter_reg udp_reg_info = { 3324 .target = "udp", 3325 .ctx_arg_info_size = 1, 3326 .ctx_arg_info = { 3327 { offsetof(struct bpf_iter__udp, udp_sk), 3328 PTR_TO_BTF_ID_OR_NULL }, 3329 }, 3330 .seq_info = &udp_seq_info, 3331 }; 3332 3333 static void __init bpf_iter_register(void) 3334 { 3335 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3336 if (bpf_iter_reg_target(&udp_reg_info)) 3337 pr_warn("Warning: could not register bpf iterator udp\n"); 3338 } 3339 #endif 3340 3341 void __init udp_init(void) 3342 { 3343 unsigned long limit; 3344 unsigned int i; 3345 3346 udp_table_init(&udp_table, "UDP"); 3347 limit = nr_free_buffer_pages() / 8; 3348 limit = max(limit, 128UL); 3349 sysctl_udp_mem[0] = limit / 4 * 3; 3350 sysctl_udp_mem[1] = limit; 3351 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3352 3353 /* 16 spinlocks per cpu */ 3354 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3355 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3356 GFP_KERNEL); 3357 if (!udp_busylocks) 3358 panic("UDP: failed to alloc udp_busylocks\n"); 3359 for (i = 0; i < (1U << udp_busylocks_log); i++) 3360 spin_lock_init(udp_busylocks + i); 3361 3362 if (register_pernet_subsys(&udp_sysctl_ops)) 3363 panic("UDP: failed to init sysctl parameters.\n"); 3364 3365 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3366 bpf_iter_register(); 3367 #endif 3368 } 3369