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 = get_random_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)) 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, err, uh->dest, info, 788 (u8 *)(uh+1)); 789 goto out; 790 } 791 if (!inet->recverr) { 792 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 793 goto out; 794 } else 795 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 796 797 sk->sk_err = err; 798 sk_error_report(sk); 799 out: 800 return 0; 801 } 802 803 int udp_err(struct sk_buff *skb, u32 info) 804 { 805 return __udp4_lib_err(skb, info, &udp_table); 806 } 807 808 /* 809 * Throw away all pending data and cancel the corking. Socket is locked. 810 */ 811 void udp_flush_pending_frames(struct sock *sk) 812 { 813 struct udp_sock *up = udp_sk(sk); 814 815 if (up->pending) { 816 up->len = 0; 817 up->pending = 0; 818 ip_flush_pending_frames(sk); 819 } 820 } 821 EXPORT_SYMBOL(udp_flush_pending_frames); 822 823 /** 824 * udp4_hwcsum - handle outgoing HW checksumming 825 * @skb: sk_buff containing the filled-in UDP header 826 * (checksum field must be zeroed out) 827 * @src: source IP address 828 * @dst: destination IP address 829 */ 830 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 831 { 832 struct udphdr *uh = udp_hdr(skb); 833 int offset = skb_transport_offset(skb); 834 int len = skb->len - offset; 835 int hlen = len; 836 __wsum csum = 0; 837 838 if (!skb_has_frag_list(skb)) { 839 /* 840 * Only one fragment on the socket. 841 */ 842 skb->csum_start = skb_transport_header(skb) - skb->head; 843 skb->csum_offset = offsetof(struct udphdr, check); 844 uh->check = ~csum_tcpudp_magic(src, dst, len, 845 IPPROTO_UDP, 0); 846 } else { 847 struct sk_buff *frags; 848 849 /* 850 * HW-checksum won't work as there are two or more 851 * fragments on the socket so that all csums of sk_buffs 852 * should be together 853 */ 854 skb_walk_frags(skb, frags) { 855 csum = csum_add(csum, frags->csum); 856 hlen -= frags->len; 857 } 858 859 csum = skb_checksum(skb, offset, hlen, csum); 860 skb->ip_summed = CHECKSUM_NONE; 861 862 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 863 if (uh->check == 0) 864 uh->check = CSUM_MANGLED_0; 865 } 866 } 867 EXPORT_SYMBOL_GPL(udp4_hwcsum); 868 869 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 870 * for the simple case like when setting the checksum for a UDP tunnel. 871 */ 872 void udp_set_csum(bool nocheck, struct sk_buff *skb, 873 __be32 saddr, __be32 daddr, int len) 874 { 875 struct udphdr *uh = udp_hdr(skb); 876 877 if (nocheck) { 878 uh->check = 0; 879 } else if (skb_is_gso(skb)) { 880 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 881 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 882 uh->check = 0; 883 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 884 if (uh->check == 0) 885 uh->check = CSUM_MANGLED_0; 886 } else { 887 skb->ip_summed = CHECKSUM_PARTIAL; 888 skb->csum_start = skb_transport_header(skb) - skb->head; 889 skb->csum_offset = offsetof(struct udphdr, check); 890 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 891 } 892 } 893 EXPORT_SYMBOL(udp_set_csum); 894 895 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 896 struct inet_cork *cork) 897 { 898 struct sock *sk = skb->sk; 899 struct inet_sock *inet = inet_sk(sk); 900 struct udphdr *uh; 901 int err; 902 int is_udplite = IS_UDPLITE(sk); 903 int offset = skb_transport_offset(skb); 904 int len = skb->len - offset; 905 int datalen = len - sizeof(*uh); 906 __wsum csum = 0; 907 908 /* 909 * Create a UDP header 910 */ 911 uh = udp_hdr(skb); 912 uh->source = inet->inet_sport; 913 uh->dest = fl4->fl4_dport; 914 uh->len = htons(len); 915 uh->check = 0; 916 917 if (cork->gso_size) { 918 const int hlen = skb_network_header_len(skb) + 919 sizeof(struct udphdr); 920 921 if (hlen + cork->gso_size > cork->fragsize) { 922 kfree_skb(skb); 923 return -EINVAL; 924 } 925 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 926 kfree_skb(skb); 927 return -EINVAL; 928 } 929 if (sk->sk_no_check_tx) { 930 kfree_skb(skb); 931 return -EINVAL; 932 } 933 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 934 dst_xfrm(skb_dst(skb))) { 935 kfree_skb(skb); 936 return -EIO; 937 } 938 939 if (datalen > cork->gso_size) { 940 skb_shinfo(skb)->gso_size = cork->gso_size; 941 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 942 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 943 cork->gso_size); 944 } 945 goto csum_partial; 946 } 947 948 if (is_udplite) /* UDP-Lite */ 949 csum = udplite_csum(skb); 950 951 else if (sk->sk_no_check_tx) { /* UDP csum off */ 952 953 skb->ip_summed = CHECKSUM_NONE; 954 goto send; 955 956 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 957 csum_partial: 958 959 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 960 goto send; 961 962 } else 963 csum = udp_csum(skb); 964 965 /* add protocol-dependent pseudo-header */ 966 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 967 sk->sk_protocol, csum); 968 if (uh->check == 0) 969 uh->check = CSUM_MANGLED_0; 970 971 send: 972 err = ip_send_skb(sock_net(sk), skb); 973 if (err) { 974 if (err == -ENOBUFS && !inet->recverr) { 975 UDP_INC_STATS(sock_net(sk), 976 UDP_MIB_SNDBUFERRORS, is_udplite); 977 err = 0; 978 } 979 } else 980 UDP_INC_STATS(sock_net(sk), 981 UDP_MIB_OUTDATAGRAMS, is_udplite); 982 return err; 983 } 984 985 /* 986 * Push out all pending data as one UDP datagram. Socket is locked. 987 */ 988 int udp_push_pending_frames(struct sock *sk) 989 { 990 struct udp_sock *up = udp_sk(sk); 991 struct inet_sock *inet = inet_sk(sk); 992 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 993 struct sk_buff *skb; 994 int err = 0; 995 996 skb = ip_finish_skb(sk, fl4); 997 if (!skb) 998 goto out; 999 1000 err = udp_send_skb(skb, fl4, &inet->cork.base); 1001 1002 out: 1003 up->len = 0; 1004 up->pending = 0; 1005 return err; 1006 } 1007 EXPORT_SYMBOL(udp_push_pending_frames); 1008 1009 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1010 { 1011 switch (cmsg->cmsg_type) { 1012 case UDP_SEGMENT: 1013 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1014 return -EINVAL; 1015 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1016 return 0; 1017 default: 1018 return -EINVAL; 1019 } 1020 } 1021 1022 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1023 { 1024 struct cmsghdr *cmsg; 1025 bool need_ip = false; 1026 int err; 1027 1028 for_each_cmsghdr(cmsg, msg) { 1029 if (!CMSG_OK(msg, cmsg)) 1030 return -EINVAL; 1031 1032 if (cmsg->cmsg_level != SOL_UDP) { 1033 need_ip = true; 1034 continue; 1035 } 1036 1037 err = __udp_cmsg_send(cmsg, gso_size); 1038 if (err) 1039 return err; 1040 } 1041 1042 return need_ip; 1043 } 1044 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1045 1046 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1047 { 1048 struct inet_sock *inet = inet_sk(sk); 1049 struct udp_sock *up = udp_sk(sk); 1050 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1051 struct flowi4 fl4_stack; 1052 struct flowi4 *fl4; 1053 int ulen = len; 1054 struct ipcm_cookie ipc; 1055 struct rtable *rt = NULL; 1056 int free = 0; 1057 int connected = 0; 1058 __be32 daddr, faddr, saddr; 1059 __be16 dport; 1060 u8 tos; 1061 int err, is_udplite = IS_UDPLITE(sk); 1062 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1063 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1064 struct sk_buff *skb; 1065 struct ip_options_data opt_copy; 1066 1067 if (len > 0xFFFF) 1068 return -EMSGSIZE; 1069 1070 /* 1071 * Check the flags. 1072 */ 1073 1074 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1075 return -EOPNOTSUPP; 1076 1077 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1078 1079 fl4 = &inet->cork.fl.u.ip4; 1080 if (up->pending) { 1081 /* 1082 * There are pending frames. 1083 * The socket lock must be held while it's corked. 1084 */ 1085 lock_sock(sk); 1086 if (likely(up->pending)) { 1087 if (unlikely(up->pending != AF_INET)) { 1088 release_sock(sk); 1089 return -EINVAL; 1090 } 1091 goto do_append_data; 1092 } 1093 release_sock(sk); 1094 } 1095 ulen += sizeof(struct udphdr); 1096 1097 /* 1098 * Get and verify the address. 1099 */ 1100 if (usin) { 1101 if (msg->msg_namelen < sizeof(*usin)) 1102 return -EINVAL; 1103 if (usin->sin_family != AF_INET) { 1104 if (usin->sin_family != AF_UNSPEC) 1105 return -EAFNOSUPPORT; 1106 } 1107 1108 daddr = usin->sin_addr.s_addr; 1109 dport = usin->sin_port; 1110 if (dport == 0) 1111 return -EINVAL; 1112 } else { 1113 if (sk->sk_state != TCP_ESTABLISHED) 1114 return -EDESTADDRREQ; 1115 daddr = inet->inet_daddr; 1116 dport = inet->inet_dport; 1117 /* Open fast path for connected socket. 1118 Route will not be used, if at least one option is set. 1119 */ 1120 connected = 1; 1121 } 1122 1123 ipcm_init_sk(&ipc, inet); 1124 ipc.gso_size = READ_ONCE(up->gso_size); 1125 1126 if (msg->msg_controllen) { 1127 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1128 if (err > 0) 1129 err = ip_cmsg_send(sk, msg, &ipc, 1130 sk->sk_family == AF_INET6); 1131 if (unlikely(err < 0)) { 1132 kfree(ipc.opt); 1133 return err; 1134 } 1135 if (ipc.opt) 1136 free = 1; 1137 connected = 0; 1138 } 1139 if (!ipc.opt) { 1140 struct ip_options_rcu *inet_opt; 1141 1142 rcu_read_lock(); 1143 inet_opt = rcu_dereference(inet->inet_opt); 1144 if (inet_opt) { 1145 memcpy(&opt_copy, inet_opt, 1146 sizeof(*inet_opt) + inet_opt->opt.optlen); 1147 ipc.opt = &opt_copy.opt; 1148 } 1149 rcu_read_unlock(); 1150 } 1151 1152 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1153 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1154 (struct sockaddr *)usin, &ipc.addr); 1155 if (err) 1156 goto out_free; 1157 if (usin) { 1158 if (usin->sin_port == 0) { 1159 /* BPF program set invalid port. Reject it. */ 1160 err = -EINVAL; 1161 goto out_free; 1162 } 1163 daddr = usin->sin_addr.s_addr; 1164 dport = usin->sin_port; 1165 } 1166 } 1167 1168 saddr = ipc.addr; 1169 ipc.addr = faddr = daddr; 1170 1171 if (ipc.opt && ipc.opt->opt.srr) { 1172 if (!daddr) { 1173 err = -EINVAL; 1174 goto out_free; 1175 } 1176 faddr = ipc.opt->opt.faddr; 1177 connected = 0; 1178 } 1179 tos = get_rttos(&ipc, inet); 1180 if (sock_flag(sk, SOCK_LOCALROUTE) || 1181 (msg->msg_flags & MSG_DONTROUTE) || 1182 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1183 tos |= RTO_ONLINK; 1184 connected = 0; 1185 } 1186 1187 if (ipv4_is_multicast(daddr)) { 1188 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1189 ipc.oif = inet->mc_index; 1190 if (!saddr) 1191 saddr = inet->mc_addr; 1192 connected = 0; 1193 } else if (!ipc.oif) { 1194 ipc.oif = inet->uc_index; 1195 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1196 /* oif is set, packet is to local broadcast and 1197 * uc_index is set. oif is most likely set 1198 * by sk_bound_dev_if. If uc_index != oif check if the 1199 * oif is an L3 master and uc_index is an L3 slave. 1200 * If so, we want to allow the send using the uc_index. 1201 */ 1202 if (ipc.oif != inet->uc_index && 1203 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1204 inet->uc_index)) { 1205 ipc.oif = inet->uc_index; 1206 } 1207 } 1208 1209 if (connected) 1210 rt = (struct rtable *)sk_dst_check(sk, 0); 1211 1212 if (!rt) { 1213 struct net *net = sock_net(sk); 1214 __u8 flow_flags = inet_sk_flowi_flags(sk); 1215 1216 fl4 = &fl4_stack; 1217 1218 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, 1219 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1220 flow_flags, 1221 faddr, saddr, dport, inet->inet_sport, 1222 sk->sk_uid); 1223 1224 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1225 rt = ip_route_output_flow(net, fl4, sk); 1226 if (IS_ERR(rt)) { 1227 err = PTR_ERR(rt); 1228 rt = NULL; 1229 if (err == -ENETUNREACH) 1230 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1231 goto out; 1232 } 1233 1234 err = -EACCES; 1235 if ((rt->rt_flags & RTCF_BROADCAST) && 1236 !sock_flag(sk, SOCK_BROADCAST)) 1237 goto out; 1238 if (connected) 1239 sk_dst_set(sk, dst_clone(&rt->dst)); 1240 } 1241 1242 if (msg->msg_flags&MSG_CONFIRM) 1243 goto do_confirm; 1244 back_from_confirm: 1245 1246 saddr = fl4->saddr; 1247 if (!ipc.addr) 1248 daddr = ipc.addr = fl4->daddr; 1249 1250 /* Lockless fast path for the non-corking case. */ 1251 if (!corkreq) { 1252 struct inet_cork cork; 1253 1254 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1255 sizeof(struct udphdr), &ipc, &rt, 1256 &cork, msg->msg_flags); 1257 err = PTR_ERR(skb); 1258 if (!IS_ERR_OR_NULL(skb)) 1259 err = udp_send_skb(skb, fl4, &cork); 1260 goto out; 1261 } 1262 1263 lock_sock(sk); 1264 if (unlikely(up->pending)) { 1265 /* The socket is already corked while preparing it. */ 1266 /* ... which is an evident application bug. --ANK */ 1267 release_sock(sk); 1268 1269 net_dbg_ratelimited("socket already corked\n"); 1270 err = -EINVAL; 1271 goto out; 1272 } 1273 /* 1274 * Now cork the socket to pend data. 1275 */ 1276 fl4 = &inet->cork.fl.u.ip4; 1277 fl4->daddr = daddr; 1278 fl4->saddr = saddr; 1279 fl4->fl4_dport = dport; 1280 fl4->fl4_sport = inet->inet_sport; 1281 up->pending = AF_INET; 1282 1283 do_append_data: 1284 up->len += ulen; 1285 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1286 sizeof(struct udphdr), &ipc, &rt, 1287 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1288 if (err) 1289 udp_flush_pending_frames(sk); 1290 else if (!corkreq) 1291 err = udp_push_pending_frames(sk); 1292 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1293 up->pending = 0; 1294 release_sock(sk); 1295 1296 out: 1297 ip_rt_put(rt); 1298 out_free: 1299 if (free) 1300 kfree(ipc.opt); 1301 if (!err) 1302 return len; 1303 /* 1304 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1305 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1306 * we don't have a good statistic (IpOutDiscards but it can be too many 1307 * things). We could add another new stat but at least for now that 1308 * seems like overkill. 1309 */ 1310 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1311 UDP_INC_STATS(sock_net(sk), 1312 UDP_MIB_SNDBUFERRORS, is_udplite); 1313 } 1314 return err; 1315 1316 do_confirm: 1317 if (msg->msg_flags & MSG_PROBE) 1318 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1319 if (!(msg->msg_flags&MSG_PROBE) || len) 1320 goto back_from_confirm; 1321 err = 0; 1322 goto out; 1323 } 1324 EXPORT_SYMBOL(udp_sendmsg); 1325 1326 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1327 size_t size, int flags) 1328 { 1329 struct inet_sock *inet = inet_sk(sk); 1330 struct udp_sock *up = udp_sk(sk); 1331 int ret; 1332 1333 if (flags & MSG_SENDPAGE_NOTLAST) 1334 flags |= MSG_MORE; 1335 1336 if (!up->pending) { 1337 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1338 1339 /* Call udp_sendmsg to specify destination address which 1340 * sendpage interface can't pass. 1341 * This will succeed only when the socket is connected. 1342 */ 1343 ret = udp_sendmsg(sk, &msg, 0); 1344 if (ret < 0) 1345 return ret; 1346 } 1347 1348 lock_sock(sk); 1349 1350 if (unlikely(!up->pending)) { 1351 release_sock(sk); 1352 1353 net_dbg_ratelimited("cork failed\n"); 1354 return -EINVAL; 1355 } 1356 1357 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1358 page, offset, size, flags); 1359 if (ret == -EOPNOTSUPP) { 1360 release_sock(sk); 1361 return sock_no_sendpage(sk->sk_socket, page, offset, 1362 size, flags); 1363 } 1364 if (ret < 0) { 1365 udp_flush_pending_frames(sk); 1366 goto out; 1367 } 1368 1369 up->len += size; 1370 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE))) 1371 ret = udp_push_pending_frames(sk); 1372 if (!ret) 1373 ret = size; 1374 out: 1375 release_sock(sk); 1376 return ret; 1377 } 1378 1379 #define UDP_SKB_IS_STATELESS 0x80000000 1380 1381 /* all head states (dst, sk, nf conntrack) except skb extensions are 1382 * cleared by udp_rcv(). 1383 * 1384 * We need to preserve secpath, if present, to eventually process 1385 * IP_CMSG_PASSSEC at recvmsg() time. 1386 * 1387 * Other extensions can be cleared. 1388 */ 1389 static bool udp_try_make_stateless(struct sk_buff *skb) 1390 { 1391 if (!skb_has_extensions(skb)) 1392 return true; 1393 1394 if (!secpath_exists(skb)) { 1395 skb_ext_reset(skb); 1396 return true; 1397 } 1398 1399 return false; 1400 } 1401 1402 static void udp_set_dev_scratch(struct sk_buff *skb) 1403 { 1404 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1405 1406 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1407 scratch->_tsize_state = skb->truesize; 1408 #if BITS_PER_LONG == 64 1409 scratch->len = skb->len; 1410 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1411 scratch->is_linear = !skb_is_nonlinear(skb); 1412 #endif 1413 if (udp_try_make_stateless(skb)) 1414 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1415 } 1416 1417 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1418 { 1419 /* We come here after udp_lib_checksum_complete() returned 0. 1420 * This means that __skb_checksum_complete() might have 1421 * set skb->csum_valid to 1. 1422 * On 64bit platforms, we can set csum_unnecessary 1423 * to true, but only if the skb is not shared. 1424 */ 1425 #if BITS_PER_LONG == 64 1426 if (!skb_shared(skb)) 1427 udp_skb_scratch(skb)->csum_unnecessary = true; 1428 #endif 1429 } 1430 1431 static int udp_skb_truesize(struct sk_buff *skb) 1432 { 1433 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1434 } 1435 1436 static bool udp_skb_has_head_state(struct sk_buff *skb) 1437 { 1438 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1439 } 1440 1441 /* fully reclaim rmem/fwd memory allocated for skb */ 1442 static void udp_rmem_release(struct sock *sk, int size, int partial, 1443 bool rx_queue_lock_held) 1444 { 1445 struct udp_sock *up = udp_sk(sk); 1446 struct sk_buff_head *sk_queue; 1447 int amt; 1448 1449 if (likely(partial)) { 1450 up->forward_deficit += size; 1451 size = up->forward_deficit; 1452 if (size < READ_ONCE(up->forward_threshold) && 1453 !skb_queue_empty(&up->reader_queue)) 1454 return; 1455 } else { 1456 size += up->forward_deficit; 1457 } 1458 up->forward_deficit = 0; 1459 1460 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1461 * if the called don't held it already 1462 */ 1463 sk_queue = &sk->sk_receive_queue; 1464 if (!rx_queue_lock_held) 1465 spin_lock(&sk_queue->lock); 1466 1467 1468 sk->sk_forward_alloc += size; 1469 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1470 sk->sk_forward_alloc -= amt; 1471 1472 if (amt) 1473 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 1474 1475 atomic_sub(size, &sk->sk_rmem_alloc); 1476 1477 /* this can save us from acquiring the rx queue lock on next receive */ 1478 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1479 1480 if (!rx_queue_lock_held) 1481 spin_unlock(&sk_queue->lock); 1482 } 1483 1484 /* Note: called with reader_queue.lock held. 1485 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1486 * This avoids a cache line miss while receive_queue lock is held. 1487 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1488 */ 1489 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1490 { 1491 prefetch(&skb->data); 1492 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1493 } 1494 EXPORT_SYMBOL(udp_skb_destructor); 1495 1496 /* as above, but the caller held the rx queue lock, too */ 1497 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1498 { 1499 prefetch(&skb->data); 1500 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1501 } 1502 1503 /* Idea of busylocks is to let producers grab an extra spinlock 1504 * to relieve pressure on the receive_queue spinlock shared by consumer. 1505 * Under flood, this means that only one producer can be in line 1506 * trying to acquire the receive_queue spinlock. 1507 * These busylock can be allocated on a per cpu manner, instead of a 1508 * per socket one (that would consume a cache line per socket) 1509 */ 1510 static int udp_busylocks_log __read_mostly; 1511 static spinlock_t *udp_busylocks __read_mostly; 1512 1513 static spinlock_t *busylock_acquire(void *ptr) 1514 { 1515 spinlock_t *busy; 1516 1517 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1518 spin_lock(busy); 1519 return busy; 1520 } 1521 1522 static void busylock_release(spinlock_t *busy) 1523 { 1524 if (busy) 1525 spin_unlock(busy); 1526 } 1527 1528 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1529 { 1530 struct sk_buff_head *list = &sk->sk_receive_queue; 1531 int rmem, delta, amt, err = -ENOMEM; 1532 spinlock_t *busy = NULL; 1533 int size; 1534 1535 /* try to avoid the costly atomic add/sub pair when the receive 1536 * queue is full; always allow at least a packet 1537 */ 1538 rmem = atomic_read(&sk->sk_rmem_alloc); 1539 if (rmem > sk->sk_rcvbuf) 1540 goto drop; 1541 1542 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1543 * having linear skbs : 1544 * - Reduce memory overhead and thus increase receive queue capacity 1545 * - Less cache line misses at copyout() time 1546 * - Less work at consume_skb() (less alien page frag freeing) 1547 */ 1548 if (rmem > (sk->sk_rcvbuf >> 1)) { 1549 skb_condense(skb); 1550 1551 busy = busylock_acquire(sk); 1552 } 1553 size = skb->truesize; 1554 udp_set_dev_scratch(skb); 1555 1556 /* we drop only if the receive buf is full and the receive 1557 * queue contains some other skb 1558 */ 1559 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1560 if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) 1561 goto uncharge_drop; 1562 1563 spin_lock(&list->lock); 1564 if (size >= sk->sk_forward_alloc) { 1565 amt = sk_mem_pages(size); 1566 delta = amt << PAGE_SHIFT; 1567 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1568 err = -ENOBUFS; 1569 spin_unlock(&list->lock); 1570 goto uncharge_drop; 1571 } 1572 1573 sk->sk_forward_alloc += delta; 1574 } 1575 1576 sk->sk_forward_alloc -= size; 1577 1578 /* no need to setup a destructor, we will explicitly release the 1579 * forward allocated memory on dequeue 1580 */ 1581 sock_skb_set_dropcount(sk, skb); 1582 1583 __skb_queue_tail(list, skb); 1584 spin_unlock(&list->lock); 1585 1586 if (!sock_flag(sk, SOCK_DEAD)) 1587 sk->sk_data_ready(sk); 1588 1589 busylock_release(busy); 1590 return 0; 1591 1592 uncharge_drop: 1593 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1594 1595 drop: 1596 atomic_inc(&sk->sk_drops); 1597 busylock_release(busy); 1598 return err; 1599 } 1600 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1601 1602 void udp_destruct_common(struct sock *sk) 1603 { 1604 /* reclaim completely the forward allocated memory */ 1605 struct udp_sock *up = udp_sk(sk); 1606 unsigned int total = 0; 1607 struct sk_buff *skb; 1608 1609 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1610 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1611 total += skb->truesize; 1612 kfree_skb(skb); 1613 } 1614 udp_rmem_release(sk, total, 0, true); 1615 } 1616 EXPORT_SYMBOL_GPL(udp_destruct_common); 1617 1618 static void udp_destruct_sock(struct sock *sk) 1619 { 1620 udp_destruct_common(sk); 1621 inet_sock_destruct(sk); 1622 } 1623 1624 int udp_init_sock(struct sock *sk) 1625 { 1626 udp_lib_init_sock(sk); 1627 sk->sk_destruct = udp_destruct_sock; 1628 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1629 return 0; 1630 } 1631 1632 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1633 { 1634 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1635 bool slow = lock_sock_fast(sk); 1636 1637 sk_peek_offset_bwd(sk, len); 1638 unlock_sock_fast(sk, slow); 1639 } 1640 1641 if (!skb_unref(skb)) 1642 return; 1643 1644 /* In the more common cases we cleared the head states previously, 1645 * see __udp_queue_rcv_skb(). 1646 */ 1647 if (unlikely(udp_skb_has_head_state(skb))) 1648 skb_release_head_state(skb); 1649 __consume_stateless_skb(skb); 1650 } 1651 EXPORT_SYMBOL_GPL(skb_consume_udp); 1652 1653 static struct sk_buff *__first_packet_length(struct sock *sk, 1654 struct sk_buff_head *rcvq, 1655 int *total) 1656 { 1657 struct sk_buff *skb; 1658 1659 while ((skb = skb_peek(rcvq)) != NULL) { 1660 if (udp_lib_checksum_complete(skb)) { 1661 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1662 IS_UDPLITE(sk)); 1663 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1664 IS_UDPLITE(sk)); 1665 atomic_inc(&sk->sk_drops); 1666 __skb_unlink(skb, rcvq); 1667 *total += skb->truesize; 1668 kfree_skb(skb); 1669 } else { 1670 udp_skb_csum_unnecessary_set(skb); 1671 break; 1672 } 1673 } 1674 return skb; 1675 } 1676 1677 /** 1678 * first_packet_length - return length of first packet in receive queue 1679 * @sk: socket 1680 * 1681 * Drops all bad checksum frames, until a valid one is found. 1682 * Returns the length of found skb, or -1 if none is found. 1683 */ 1684 static int first_packet_length(struct sock *sk) 1685 { 1686 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1687 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1688 struct sk_buff *skb; 1689 int total = 0; 1690 int res; 1691 1692 spin_lock_bh(&rcvq->lock); 1693 skb = __first_packet_length(sk, rcvq, &total); 1694 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1695 spin_lock(&sk_queue->lock); 1696 skb_queue_splice_tail_init(sk_queue, rcvq); 1697 spin_unlock(&sk_queue->lock); 1698 1699 skb = __first_packet_length(sk, rcvq, &total); 1700 } 1701 res = skb ? skb->len : -1; 1702 if (total) 1703 udp_rmem_release(sk, total, 1, false); 1704 spin_unlock_bh(&rcvq->lock); 1705 return res; 1706 } 1707 1708 /* 1709 * IOCTL requests applicable to the UDP protocol 1710 */ 1711 1712 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1713 { 1714 switch (cmd) { 1715 case SIOCOUTQ: 1716 { 1717 int amount = sk_wmem_alloc_get(sk); 1718 1719 return put_user(amount, (int __user *)arg); 1720 } 1721 1722 case SIOCINQ: 1723 { 1724 int amount = max_t(int, 0, first_packet_length(sk)); 1725 1726 return put_user(amount, (int __user *)arg); 1727 } 1728 1729 default: 1730 return -ENOIOCTLCMD; 1731 } 1732 1733 return 0; 1734 } 1735 EXPORT_SYMBOL(udp_ioctl); 1736 1737 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1738 int *off, int *err) 1739 { 1740 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1741 struct sk_buff_head *queue; 1742 struct sk_buff *last; 1743 long timeo; 1744 int error; 1745 1746 queue = &udp_sk(sk)->reader_queue; 1747 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1748 do { 1749 struct sk_buff *skb; 1750 1751 error = sock_error(sk); 1752 if (error) 1753 break; 1754 1755 error = -EAGAIN; 1756 do { 1757 spin_lock_bh(&queue->lock); 1758 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1759 err, &last); 1760 if (skb) { 1761 if (!(flags & MSG_PEEK)) 1762 udp_skb_destructor(sk, skb); 1763 spin_unlock_bh(&queue->lock); 1764 return skb; 1765 } 1766 1767 if (skb_queue_empty_lockless(sk_queue)) { 1768 spin_unlock_bh(&queue->lock); 1769 goto busy_check; 1770 } 1771 1772 /* refill the reader queue and walk it again 1773 * keep both queues locked to avoid re-acquiring 1774 * the sk_receive_queue lock if fwd memory scheduling 1775 * is needed. 1776 */ 1777 spin_lock(&sk_queue->lock); 1778 skb_queue_splice_tail_init(sk_queue, queue); 1779 1780 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1781 err, &last); 1782 if (skb && !(flags & MSG_PEEK)) 1783 udp_skb_dtor_locked(sk, skb); 1784 spin_unlock(&sk_queue->lock); 1785 spin_unlock_bh(&queue->lock); 1786 if (skb) 1787 return skb; 1788 1789 busy_check: 1790 if (!sk_can_busy_loop(sk)) 1791 break; 1792 1793 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1794 } while (!skb_queue_empty_lockless(sk_queue)); 1795 1796 /* sk_queue is empty, reader_queue may contain peeked packets */ 1797 } while (timeo && 1798 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1799 &error, &timeo, 1800 (struct sk_buff *)sk_queue)); 1801 1802 *err = error; 1803 return NULL; 1804 } 1805 EXPORT_SYMBOL(__skb_recv_udp); 1806 1807 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) 1808 { 1809 struct sk_buff *skb; 1810 int err, copied; 1811 1812 try_again: 1813 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1814 if (!skb) 1815 return err; 1816 1817 if (udp_lib_checksum_complete(skb)) { 1818 int is_udplite = IS_UDPLITE(sk); 1819 struct net *net = sock_net(sk); 1820 1821 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite); 1822 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite); 1823 atomic_inc(&sk->sk_drops); 1824 kfree_skb(skb); 1825 goto try_again; 1826 } 1827 1828 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); 1829 copied = recv_actor(sk, skb); 1830 kfree_skb(skb); 1831 1832 return copied; 1833 } 1834 EXPORT_SYMBOL(udp_read_skb); 1835 1836 /* 1837 * This should be easy, if there is something there we 1838 * return it, otherwise we block. 1839 */ 1840 1841 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 1842 int *addr_len) 1843 { 1844 struct inet_sock *inet = inet_sk(sk); 1845 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1846 struct sk_buff *skb; 1847 unsigned int ulen, copied; 1848 int off, err, peeking = flags & MSG_PEEK; 1849 int is_udplite = IS_UDPLITE(sk); 1850 bool checksum_valid = false; 1851 1852 if (flags & MSG_ERRQUEUE) 1853 return ip_recv_error(sk, msg, len, addr_len); 1854 1855 try_again: 1856 off = sk_peek_offset(sk, flags); 1857 skb = __skb_recv_udp(sk, flags, &off, &err); 1858 if (!skb) 1859 return err; 1860 1861 ulen = udp_skb_len(skb); 1862 copied = len; 1863 if (copied > ulen - off) 1864 copied = ulen - off; 1865 else if (copied < ulen) 1866 msg->msg_flags |= MSG_TRUNC; 1867 1868 /* 1869 * If checksum is needed at all, try to do it while copying the 1870 * data. If the data is truncated, or if we only want a partial 1871 * coverage checksum (UDP-Lite), do it before the copy. 1872 */ 1873 1874 if (copied < ulen || peeking || 1875 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1876 checksum_valid = udp_skb_csum_unnecessary(skb) || 1877 !__udp_lib_checksum_complete(skb); 1878 if (!checksum_valid) 1879 goto csum_copy_err; 1880 } 1881 1882 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1883 if (udp_skb_is_linear(skb)) 1884 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1885 else 1886 err = skb_copy_datagram_msg(skb, off, msg, copied); 1887 } else { 1888 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1889 1890 if (err == -EINVAL) 1891 goto csum_copy_err; 1892 } 1893 1894 if (unlikely(err)) { 1895 if (!peeking) { 1896 atomic_inc(&sk->sk_drops); 1897 UDP_INC_STATS(sock_net(sk), 1898 UDP_MIB_INERRORS, is_udplite); 1899 } 1900 kfree_skb(skb); 1901 return err; 1902 } 1903 1904 if (!peeking) 1905 UDP_INC_STATS(sock_net(sk), 1906 UDP_MIB_INDATAGRAMS, is_udplite); 1907 1908 sock_recv_cmsgs(msg, sk, skb); 1909 1910 /* Copy the address. */ 1911 if (sin) { 1912 sin->sin_family = AF_INET; 1913 sin->sin_port = udp_hdr(skb)->source; 1914 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1915 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1916 *addr_len = sizeof(*sin); 1917 1918 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1919 (struct sockaddr *)sin); 1920 } 1921 1922 if (udp_sk(sk)->gro_enabled) 1923 udp_cmsg_recv(msg, sk, skb); 1924 1925 if (inet->cmsg_flags) 1926 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1927 1928 err = copied; 1929 if (flags & MSG_TRUNC) 1930 err = ulen; 1931 1932 skb_consume_udp(sk, skb, peeking ? -err : err); 1933 return err; 1934 1935 csum_copy_err: 1936 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1937 udp_skb_destructor)) { 1938 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1939 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1940 } 1941 kfree_skb(skb); 1942 1943 /* starting over for a new packet, but check if we need to yield */ 1944 cond_resched(); 1945 msg->msg_flags &= ~MSG_TRUNC; 1946 goto try_again; 1947 } 1948 1949 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1950 { 1951 /* This check is replicated from __ip4_datagram_connect() and 1952 * intended to prevent BPF program called below from accessing bytes 1953 * that are out of the bound specified by user in addr_len. 1954 */ 1955 if (addr_len < sizeof(struct sockaddr_in)) 1956 return -EINVAL; 1957 1958 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1959 } 1960 EXPORT_SYMBOL(udp_pre_connect); 1961 1962 int __udp_disconnect(struct sock *sk, int flags) 1963 { 1964 struct inet_sock *inet = inet_sk(sk); 1965 /* 1966 * 1003.1g - break association. 1967 */ 1968 1969 sk->sk_state = TCP_CLOSE; 1970 inet->inet_daddr = 0; 1971 inet->inet_dport = 0; 1972 sock_rps_reset_rxhash(sk); 1973 sk->sk_bound_dev_if = 0; 1974 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1975 inet_reset_saddr(sk); 1976 if (sk->sk_prot->rehash && 1977 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1978 sk->sk_prot->rehash(sk); 1979 } 1980 1981 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1982 sk->sk_prot->unhash(sk); 1983 inet->inet_sport = 0; 1984 } 1985 sk_dst_reset(sk); 1986 return 0; 1987 } 1988 EXPORT_SYMBOL(__udp_disconnect); 1989 1990 int udp_disconnect(struct sock *sk, int flags) 1991 { 1992 lock_sock(sk); 1993 __udp_disconnect(sk, flags); 1994 release_sock(sk); 1995 return 0; 1996 } 1997 EXPORT_SYMBOL(udp_disconnect); 1998 1999 void udp_lib_unhash(struct sock *sk) 2000 { 2001 if (sk_hashed(sk)) { 2002 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2003 struct udp_hslot *hslot, *hslot2; 2004 2005 hslot = udp_hashslot(udptable, sock_net(sk), 2006 udp_sk(sk)->udp_port_hash); 2007 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2008 2009 spin_lock_bh(&hslot->lock); 2010 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2011 reuseport_detach_sock(sk); 2012 if (sk_del_node_init_rcu(sk)) { 2013 hslot->count--; 2014 inet_sk(sk)->inet_num = 0; 2015 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 2016 2017 spin_lock(&hslot2->lock); 2018 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2019 hslot2->count--; 2020 spin_unlock(&hslot2->lock); 2021 } 2022 spin_unlock_bh(&hslot->lock); 2023 } 2024 } 2025 EXPORT_SYMBOL(udp_lib_unhash); 2026 2027 /* 2028 * inet_rcv_saddr was changed, we must rehash secondary hash 2029 */ 2030 void udp_lib_rehash(struct sock *sk, u16 newhash) 2031 { 2032 if (sk_hashed(sk)) { 2033 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2034 struct udp_hslot *hslot, *hslot2, *nhslot2; 2035 2036 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2037 nhslot2 = udp_hashslot2(udptable, newhash); 2038 udp_sk(sk)->udp_portaddr_hash = newhash; 2039 2040 if (hslot2 != nhslot2 || 2041 rcu_access_pointer(sk->sk_reuseport_cb)) { 2042 hslot = udp_hashslot(udptable, sock_net(sk), 2043 udp_sk(sk)->udp_port_hash); 2044 /* we must lock primary chain too */ 2045 spin_lock_bh(&hslot->lock); 2046 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2047 reuseport_detach_sock(sk); 2048 2049 if (hslot2 != nhslot2) { 2050 spin_lock(&hslot2->lock); 2051 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2052 hslot2->count--; 2053 spin_unlock(&hslot2->lock); 2054 2055 spin_lock(&nhslot2->lock); 2056 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2057 &nhslot2->head); 2058 nhslot2->count++; 2059 spin_unlock(&nhslot2->lock); 2060 } 2061 2062 spin_unlock_bh(&hslot->lock); 2063 } 2064 } 2065 } 2066 EXPORT_SYMBOL(udp_lib_rehash); 2067 2068 void udp_v4_rehash(struct sock *sk) 2069 { 2070 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2071 inet_sk(sk)->inet_rcv_saddr, 2072 inet_sk(sk)->inet_num); 2073 udp_lib_rehash(sk, new_hash); 2074 } 2075 2076 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2077 { 2078 int rc; 2079 2080 if (inet_sk(sk)->inet_daddr) { 2081 sock_rps_save_rxhash(sk, skb); 2082 sk_mark_napi_id(sk, skb); 2083 sk_incoming_cpu_update(sk); 2084 } else { 2085 sk_mark_napi_id_once(sk, skb); 2086 } 2087 2088 rc = __udp_enqueue_schedule_skb(sk, skb); 2089 if (rc < 0) { 2090 int is_udplite = IS_UDPLITE(sk); 2091 int drop_reason; 2092 2093 /* Note that an ENOMEM error is charged twice */ 2094 if (rc == -ENOMEM) { 2095 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2096 is_udplite); 2097 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 2098 } else { 2099 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2100 is_udplite); 2101 drop_reason = SKB_DROP_REASON_PROTO_MEM; 2102 } 2103 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2104 kfree_skb_reason(skb, drop_reason); 2105 trace_udp_fail_queue_rcv_skb(rc, sk); 2106 return -1; 2107 } 2108 2109 return 0; 2110 } 2111 2112 /* returns: 2113 * -1: error 2114 * 0: success 2115 * >0: "udp encap" protocol resubmission 2116 * 2117 * Note that in the success and error cases, the skb is assumed to 2118 * have either been requeued or freed. 2119 */ 2120 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2121 { 2122 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2123 struct udp_sock *up = udp_sk(sk); 2124 int is_udplite = IS_UDPLITE(sk); 2125 2126 /* 2127 * Charge it to the socket, dropping if the queue is full. 2128 */ 2129 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 2130 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2131 goto drop; 2132 } 2133 nf_reset_ct(skb); 2134 2135 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 2136 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2137 2138 /* 2139 * This is an encapsulation socket so pass the skb to 2140 * the socket's udp_encap_rcv() hook. Otherwise, just 2141 * fall through and pass this up the UDP socket. 2142 * up->encap_rcv() returns the following value: 2143 * =0 if skb was successfully passed to the encap 2144 * handler or was discarded by it. 2145 * >0 if skb should be passed on to UDP. 2146 * <0 if skb should be resubmitted as proto -N 2147 */ 2148 2149 /* if we're overly short, let UDP handle it */ 2150 encap_rcv = READ_ONCE(up->encap_rcv); 2151 if (encap_rcv) { 2152 int ret; 2153 2154 /* Verify checksum before giving to encap */ 2155 if (udp_lib_checksum_complete(skb)) 2156 goto csum_error; 2157 2158 ret = encap_rcv(sk, skb); 2159 if (ret <= 0) { 2160 __UDP_INC_STATS(sock_net(sk), 2161 UDP_MIB_INDATAGRAMS, 2162 is_udplite); 2163 return -ret; 2164 } 2165 } 2166 2167 /* FALLTHROUGH -- it's a UDP Packet */ 2168 } 2169 2170 /* 2171 * UDP-Lite specific tests, ignored on UDP sockets 2172 */ 2173 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2174 2175 /* 2176 * MIB statistics other than incrementing the error count are 2177 * disabled for the following two types of errors: these depend 2178 * on the application settings, not on the functioning of the 2179 * protocol stack as such. 2180 * 2181 * RFC 3828 here recommends (sec 3.3): "There should also be a 2182 * way ... to ... at least let the receiving application block 2183 * delivery of packets with coverage values less than a value 2184 * provided by the application." 2185 */ 2186 if (up->pcrlen == 0) { /* full coverage was set */ 2187 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2188 UDP_SKB_CB(skb)->cscov, skb->len); 2189 goto drop; 2190 } 2191 /* The next case involves violating the min. coverage requested 2192 * by the receiver. This is subtle: if receiver wants x and x is 2193 * greater than the buffersize/MTU then receiver will complain 2194 * that it wants x while sender emits packets of smaller size y. 2195 * Therefore the above ...()->partial_cov statement is essential. 2196 */ 2197 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2198 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2199 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2200 goto drop; 2201 } 2202 } 2203 2204 prefetch(&sk->sk_rmem_alloc); 2205 if (rcu_access_pointer(sk->sk_filter) && 2206 udp_lib_checksum_complete(skb)) 2207 goto csum_error; 2208 2209 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 2210 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 2211 goto drop; 2212 } 2213 2214 udp_csum_pull_header(skb); 2215 2216 ipv4_pktinfo_prepare(sk, skb); 2217 return __udp_queue_rcv_skb(sk, skb); 2218 2219 csum_error: 2220 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2221 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2222 drop: 2223 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2224 atomic_inc(&sk->sk_drops); 2225 kfree_skb_reason(skb, drop_reason); 2226 return -1; 2227 } 2228 2229 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2230 { 2231 struct sk_buff *next, *segs; 2232 int ret; 2233 2234 if (likely(!udp_unexpected_gso(sk, skb))) 2235 return udp_queue_rcv_one_skb(sk, skb); 2236 2237 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2238 __skb_push(skb, -skb_mac_offset(skb)); 2239 segs = udp_rcv_segment(sk, skb, true); 2240 skb_list_walk_safe(segs, skb, next) { 2241 __skb_pull(skb, skb_transport_offset(skb)); 2242 2243 udp_post_segment_fix_csum(skb); 2244 ret = udp_queue_rcv_one_skb(sk, skb); 2245 if (ret > 0) 2246 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2247 } 2248 return 0; 2249 } 2250 2251 /* For TCP sockets, sk_rx_dst is protected by socket lock 2252 * For UDP, we use xchg() to guard against concurrent changes. 2253 */ 2254 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2255 { 2256 struct dst_entry *old; 2257 2258 if (dst_hold_safe(dst)) { 2259 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); 2260 dst_release(old); 2261 return old != dst; 2262 } 2263 return false; 2264 } 2265 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2266 2267 /* 2268 * Multicasts and broadcasts go to each listener. 2269 * 2270 * Note: called only from the BH handler context. 2271 */ 2272 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2273 struct udphdr *uh, 2274 __be32 saddr, __be32 daddr, 2275 struct udp_table *udptable, 2276 int proto) 2277 { 2278 struct sock *sk, *first = NULL; 2279 unsigned short hnum = ntohs(uh->dest); 2280 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2281 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2282 unsigned int offset = offsetof(typeof(*sk), sk_node); 2283 int dif = skb->dev->ifindex; 2284 int sdif = inet_sdif(skb); 2285 struct hlist_node *node; 2286 struct sk_buff *nskb; 2287 2288 if (use_hash2) { 2289 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2290 udptable->mask; 2291 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2292 start_lookup: 2293 hslot = &udptable->hash2[hash2]; 2294 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2295 } 2296 2297 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2298 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2299 uh->source, saddr, dif, sdif, hnum)) 2300 continue; 2301 2302 if (!first) { 2303 first = sk; 2304 continue; 2305 } 2306 nskb = skb_clone(skb, GFP_ATOMIC); 2307 2308 if (unlikely(!nskb)) { 2309 atomic_inc(&sk->sk_drops); 2310 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2311 IS_UDPLITE(sk)); 2312 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2313 IS_UDPLITE(sk)); 2314 continue; 2315 } 2316 if (udp_queue_rcv_skb(sk, nskb) > 0) 2317 consume_skb(nskb); 2318 } 2319 2320 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2321 if (use_hash2 && hash2 != hash2_any) { 2322 hash2 = hash2_any; 2323 goto start_lookup; 2324 } 2325 2326 if (first) { 2327 if (udp_queue_rcv_skb(first, skb) > 0) 2328 consume_skb(skb); 2329 } else { 2330 kfree_skb(skb); 2331 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2332 proto == IPPROTO_UDPLITE); 2333 } 2334 return 0; 2335 } 2336 2337 /* Initialize UDP checksum. If exited with zero value (success), 2338 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2339 * Otherwise, csum completion requires checksumming packet body, 2340 * including udp header and folding it to skb->csum. 2341 */ 2342 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2343 int proto) 2344 { 2345 int err; 2346 2347 UDP_SKB_CB(skb)->partial_cov = 0; 2348 UDP_SKB_CB(skb)->cscov = skb->len; 2349 2350 if (proto == IPPROTO_UDPLITE) { 2351 err = udplite_checksum_init(skb, uh); 2352 if (err) 2353 return err; 2354 2355 if (UDP_SKB_CB(skb)->partial_cov) { 2356 skb->csum = inet_compute_pseudo(skb, proto); 2357 return 0; 2358 } 2359 } 2360 2361 /* Note, we are only interested in != 0 or == 0, thus the 2362 * force to int. 2363 */ 2364 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2365 inet_compute_pseudo); 2366 if (err) 2367 return err; 2368 2369 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2370 /* If SW calculated the value, we know it's bad */ 2371 if (skb->csum_complete_sw) 2372 return 1; 2373 2374 /* HW says the value is bad. Let's validate that. 2375 * skb->csum is no longer the full packet checksum, 2376 * so don't treat it as such. 2377 */ 2378 skb_checksum_complete_unset(skb); 2379 } 2380 2381 return 0; 2382 } 2383 2384 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2385 * return code conversion for ip layer consumption 2386 */ 2387 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2388 struct udphdr *uh) 2389 { 2390 int ret; 2391 2392 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2393 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2394 2395 ret = udp_queue_rcv_skb(sk, skb); 2396 2397 /* a return value > 0 means to resubmit the input, but 2398 * it wants the return to be -protocol, or 0 2399 */ 2400 if (ret > 0) 2401 return -ret; 2402 return 0; 2403 } 2404 2405 /* 2406 * All we need to do is get the socket, and then do a checksum. 2407 */ 2408 2409 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2410 int proto) 2411 { 2412 struct sock *sk; 2413 struct udphdr *uh; 2414 unsigned short ulen; 2415 struct rtable *rt = skb_rtable(skb); 2416 __be32 saddr, daddr; 2417 struct net *net = dev_net(skb->dev); 2418 bool refcounted; 2419 int drop_reason; 2420 2421 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2422 2423 /* 2424 * Validate the packet. 2425 */ 2426 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2427 goto drop; /* No space for header. */ 2428 2429 uh = udp_hdr(skb); 2430 ulen = ntohs(uh->len); 2431 saddr = ip_hdr(skb)->saddr; 2432 daddr = ip_hdr(skb)->daddr; 2433 2434 if (ulen > skb->len) 2435 goto short_packet; 2436 2437 if (proto == IPPROTO_UDP) { 2438 /* UDP validates ulen. */ 2439 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2440 goto short_packet; 2441 uh = udp_hdr(skb); 2442 } 2443 2444 if (udp4_csum_init(skb, uh, proto)) 2445 goto csum_error; 2446 2447 sk = skb_steal_sock(skb, &refcounted); 2448 if (sk) { 2449 struct dst_entry *dst = skb_dst(skb); 2450 int ret; 2451 2452 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2453 udp_sk_rx_dst_set(sk, dst); 2454 2455 ret = udp_unicast_rcv_skb(sk, skb, uh); 2456 if (refcounted) 2457 sock_put(sk); 2458 return ret; 2459 } 2460 2461 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2462 return __udp4_lib_mcast_deliver(net, skb, uh, 2463 saddr, daddr, udptable, proto); 2464 2465 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2466 if (sk) 2467 return udp_unicast_rcv_skb(sk, skb, uh); 2468 2469 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2470 goto drop; 2471 nf_reset_ct(skb); 2472 2473 /* No socket. Drop packet silently, if checksum is wrong */ 2474 if (udp_lib_checksum_complete(skb)) 2475 goto csum_error; 2476 2477 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2478 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2479 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2480 2481 /* 2482 * Hmm. We got an UDP packet to a port to which we 2483 * don't wanna listen. Ignore it. 2484 */ 2485 kfree_skb_reason(skb, drop_reason); 2486 return 0; 2487 2488 short_packet: 2489 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2490 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2491 proto == IPPROTO_UDPLITE ? "Lite" : "", 2492 &saddr, ntohs(uh->source), 2493 ulen, skb->len, 2494 &daddr, ntohs(uh->dest)); 2495 goto drop; 2496 2497 csum_error: 2498 /* 2499 * RFC1122: OK. Discards the bad packet silently (as far as 2500 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2501 */ 2502 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2503 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2504 proto == IPPROTO_UDPLITE ? "Lite" : "", 2505 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2506 ulen); 2507 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2508 drop: 2509 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2510 kfree_skb_reason(skb, drop_reason); 2511 return 0; 2512 } 2513 2514 /* We can only early demux multicast if there is a single matching socket. 2515 * If more than one socket found returns NULL 2516 */ 2517 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2518 __be16 loc_port, __be32 loc_addr, 2519 __be16 rmt_port, __be32 rmt_addr, 2520 int dif, int sdif) 2521 { 2522 struct sock *sk, *result; 2523 unsigned short hnum = ntohs(loc_port); 2524 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2525 struct udp_hslot *hslot = &udp_table.hash[slot]; 2526 2527 /* Do not bother scanning a too big list */ 2528 if (hslot->count > 10) 2529 return NULL; 2530 2531 result = NULL; 2532 sk_for_each_rcu(sk, &hslot->head) { 2533 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2534 rmt_port, rmt_addr, dif, sdif, hnum)) { 2535 if (result) 2536 return NULL; 2537 result = sk; 2538 } 2539 } 2540 2541 return result; 2542 } 2543 2544 /* For unicast we should only early demux connected sockets or we can 2545 * break forwarding setups. The chains here can be long so only check 2546 * if the first socket is an exact match and if not move on. 2547 */ 2548 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2549 __be16 loc_port, __be32 loc_addr, 2550 __be16 rmt_port, __be32 rmt_addr, 2551 int dif, int sdif) 2552 { 2553 unsigned short hnum = ntohs(loc_port); 2554 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2555 unsigned int slot2 = hash2 & udp_table.mask; 2556 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2557 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2558 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2559 struct sock *sk; 2560 2561 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2562 if (inet_match(net, sk, acookie, ports, dif, sdif)) 2563 return sk; 2564 /* Only check first socket in chain */ 2565 break; 2566 } 2567 return NULL; 2568 } 2569 2570 int udp_v4_early_demux(struct sk_buff *skb) 2571 { 2572 struct net *net = dev_net(skb->dev); 2573 struct in_device *in_dev = NULL; 2574 const struct iphdr *iph; 2575 const struct udphdr *uh; 2576 struct sock *sk = NULL; 2577 struct dst_entry *dst; 2578 int dif = skb->dev->ifindex; 2579 int sdif = inet_sdif(skb); 2580 int ours; 2581 2582 /* validate the packet */ 2583 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2584 return 0; 2585 2586 iph = ip_hdr(skb); 2587 uh = udp_hdr(skb); 2588 2589 if (skb->pkt_type == PACKET_MULTICAST) { 2590 in_dev = __in_dev_get_rcu(skb->dev); 2591 2592 if (!in_dev) 2593 return 0; 2594 2595 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2596 iph->protocol); 2597 if (!ours) 2598 return 0; 2599 2600 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2601 uh->source, iph->saddr, 2602 dif, sdif); 2603 } else if (skb->pkt_type == PACKET_HOST) { 2604 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2605 uh->source, iph->saddr, dif, sdif); 2606 } 2607 2608 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2609 return 0; 2610 2611 skb->sk = sk; 2612 skb->destructor = sock_efree; 2613 dst = rcu_dereference(sk->sk_rx_dst); 2614 2615 if (dst) 2616 dst = dst_check(dst, 0); 2617 if (dst) { 2618 u32 itag = 0; 2619 2620 /* set noref for now. 2621 * any place which wants to hold dst has to call 2622 * dst_hold_safe() 2623 */ 2624 skb_dst_set_noref(skb, dst); 2625 2626 /* for unconnected multicast sockets we need to validate 2627 * the source on each packet 2628 */ 2629 if (!inet_sk(sk)->inet_daddr && in_dev) 2630 return ip_mc_validate_source(skb, iph->daddr, 2631 iph->saddr, 2632 iph->tos & IPTOS_RT_MASK, 2633 skb->dev, in_dev, &itag); 2634 } 2635 return 0; 2636 } 2637 2638 int udp_rcv(struct sk_buff *skb) 2639 { 2640 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2641 } 2642 2643 void udp_destroy_sock(struct sock *sk) 2644 { 2645 struct udp_sock *up = udp_sk(sk); 2646 bool slow = lock_sock_fast(sk); 2647 2648 /* protects from races with udp_abort() */ 2649 sock_set_flag(sk, SOCK_DEAD); 2650 udp_flush_pending_frames(sk); 2651 unlock_sock_fast(sk, slow); 2652 if (static_branch_unlikely(&udp_encap_needed_key)) { 2653 if (up->encap_type) { 2654 void (*encap_destroy)(struct sock *sk); 2655 encap_destroy = READ_ONCE(up->encap_destroy); 2656 if (encap_destroy) 2657 encap_destroy(sk); 2658 } 2659 if (up->encap_enabled) 2660 static_branch_dec(&udp_encap_needed_key); 2661 } 2662 } 2663 2664 /* 2665 * Socket option code for UDP 2666 */ 2667 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2668 sockptr_t optval, unsigned int optlen, 2669 int (*push_pending_frames)(struct sock *)) 2670 { 2671 struct udp_sock *up = udp_sk(sk); 2672 int val, valbool; 2673 int err = 0; 2674 int is_udplite = IS_UDPLITE(sk); 2675 2676 if (level == SOL_SOCKET) { 2677 err = sk_setsockopt(sk, level, optname, optval, optlen); 2678 2679 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) { 2680 sockopt_lock_sock(sk); 2681 /* paired with READ_ONCE in udp_rmem_release() */ 2682 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2); 2683 sockopt_release_sock(sk); 2684 } 2685 return err; 2686 } 2687 2688 if (optlen < sizeof(int)) 2689 return -EINVAL; 2690 2691 if (copy_from_sockptr(&val, optval, sizeof(val))) 2692 return -EFAULT; 2693 2694 valbool = val ? 1 : 0; 2695 2696 switch (optname) { 2697 case UDP_CORK: 2698 if (val != 0) { 2699 WRITE_ONCE(up->corkflag, 1); 2700 } else { 2701 WRITE_ONCE(up->corkflag, 0); 2702 lock_sock(sk); 2703 push_pending_frames(sk); 2704 release_sock(sk); 2705 } 2706 break; 2707 2708 case UDP_ENCAP: 2709 switch (val) { 2710 case 0: 2711 #ifdef CONFIG_XFRM 2712 case UDP_ENCAP_ESPINUDP: 2713 case UDP_ENCAP_ESPINUDP_NON_IKE: 2714 #if IS_ENABLED(CONFIG_IPV6) 2715 if (sk->sk_family == AF_INET6) 2716 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; 2717 else 2718 #endif 2719 up->encap_rcv = xfrm4_udp_encap_rcv; 2720 #endif 2721 fallthrough; 2722 case UDP_ENCAP_L2TPINUDP: 2723 up->encap_type = val; 2724 lock_sock(sk); 2725 udp_tunnel_encap_enable(sk->sk_socket); 2726 release_sock(sk); 2727 break; 2728 default: 2729 err = -ENOPROTOOPT; 2730 break; 2731 } 2732 break; 2733 2734 case UDP_NO_CHECK6_TX: 2735 up->no_check6_tx = valbool; 2736 break; 2737 2738 case UDP_NO_CHECK6_RX: 2739 up->no_check6_rx = valbool; 2740 break; 2741 2742 case UDP_SEGMENT: 2743 if (val < 0 || val > USHRT_MAX) 2744 return -EINVAL; 2745 WRITE_ONCE(up->gso_size, val); 2746 break; 2747 2748 case UDP_GRO: 2749 lock_sock(sk); 2750 2751 /* when enabling GRO, accept the related GSO packet type */ 2752 if (valbool) 2753 udp_tunnel_encap_enable(sk->sk_socket); 2754 up->gro_enabled = valbool; 2755 up->accept_udp_l4 = valbool; 2756 release_sock(sk); 2757 break; 2758 2759 /* 2760 * UDP-Lite's partial checksum coverage (RFC 3828). 2761 */ 2762 /* The sender sets actual checksum coverage length via this option. 2763 * The case coverage > packet length is handled by send module. */ 2764 case UDPLITE_SEND_CSCOV: 2765 if (!is_udplite) /* Disable the option on UDP sockets */ 2766 return -ENOPROTOOPT; 2767 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2768 val = 8; 2769 else if (val > USHRT_MAX) 2770 val = USHRT_MAX; 2771 up->pcslen = val; 2772 up->pcflag |= UDPLITE_SEND_CC; 2773 break; 2774 2775 /* The receiver specifies a minimum checksum coverage value. To make 2776 * sense, this should be set to at least 8 (as done below). If zero is 2777 * used, this again means full checksum coverage. */ 2778 case UDPLITE_RECV_CSCOV: 2779 if (!is_udplite) /* Disable the option on UDP sockets */ 2780 return -ENOPROTOOPT; 2781 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2782 val = 8; 2783 else if (val > USHRT_MAX) 2784 val = USHRT_MAX; 2785 up->pcrlen = val; 2786 up->pcflag |= UDPLITE_RECV_CC; 2787 break; 2788 2789 default: 2790 err = -ENOPROTOOPT; 2791 break; 2792 } 2793 2794 return err; 2795 } 2796 EXPORT_SYMBOL(udp_lib_setsockopt); 2797 2798 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2799 unsigned int optlen) 2800 { 2801 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 2802 return udp_lib_setsockopt(sk, level, optname, 2803 optval, optlen, 2804 udp_push_pending_frames); 2805 return ip_setsockopt(sk, level, optname, optval, optlen); 2806 } 2807 2808 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2809 char __user *optval, int __user *optlen) 2810 { 2811 struct udp_sock *up = udp_sk(sk); 2812 int val, len; 2813 2814 if (get_user(len, optlen)) 2815 return -EFAULT; 2816 2817 len = min_t(unsigned int, len, sizeof(int)); 2818 2819 if (len < 0) 2820 return -EINVAL; 2821 2822 switch (optname) { 2823 case UDP_CORK: 2824 val = READ_ONCE(up->corkflag); 2825 break; 2826 2827 case UDP_ENCAP: 2828 val = up->encap_type; 2829 break; 2830 2831 case UDP_NO_CHECK6_TX: 2832 val = up->no_check6_tx; 2833 break; 2834 2835 case UDP_NO_CHECK6_RX: 2836 val = up->no_check6_rx; 2837 break; 2838 2839 case UDP_SEGMENT: 2840 val = READ_ONCE(up->gso_size); 2841 break; 2842 2843 case UDP_GRO: 2844 val = up->gro_enabled; 2845 break; 2846 2847 /* The following two cannot be changed on UDP sockets, the return is 2848 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2849 case UDPLITE_SEND_CSCOV: 2850 val = up->pcslen; 2851 break; 2852 2853 case UDPLITE_RECV_CSCOV: 2854 val = up->pcrlen; 2855 break; 2856 2857 default: 2858 return -ENOPROTOOPT; 2859 } 2860 2861 if (put_user(len, optlen)) 2862 return -EFAULT; 2863 if (copy_to_user(optval, &val, len)) 2864 return -EFAULT; 2865 return 0; 2866 } 2867 EXPORT_SYMBOL(udp_lib_getsockopt); 2868 2869 int udp_getsockopt(struct sock *sk, int level, int optname, 2870 char __user *optval, int __user *optlen) 2871 { 2872 if (level == SOL_UDP || level == SOL_UDPLITE) 2873 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2874 return ip_getsockopt(sk, level, optname, optval, optlen); 2875 } 2876 2877 /** 2878 * udp_poll - wait for a UDP event. 2879 * @file: - file struct 2880 * @sock: - socket 2881 * @wait: - poll table 2882 * 2883 * This is same as datagram poll, except for the special case of 2884 * blocking sockets. If application is using a blocking fd 2885 * and a packet with checksum error is in the queue; 2886 * then it could get return from select indicating data available 2887 * but then block when reading it. Add special case code 2888 * to work around these arguably broken applications. 2889 */ 2890 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2891 { 2892 __poll_t mask = datagram_poll(file, sock, wait); 2893 struct sock *sk = sock->sk; 2894 2895 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2896 mask |= EPOLLIN | EPOLLRDNORM; 2897 2898 /* Check for false positives due to checksum errors */ 2899 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2900 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2901 mask &= ~(EPOLLIN | EPOLLRDNORM); 2902 2903 /* psock ingress_msg queue should not contain any bad checksum frames */ 2904 if (sk_is_readable(sk)) 2905 mask |= EPOLLIN | EPOLLRDNORM; 2906 return mask; 2907 2908 } 2909 EXPORT_SYMBOL(udp_poll); 2910 2911 int udp_abort(struct sock *sk, int err) 2912 { 2913 lock_sock(sk); 2914 2915 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2916 * with close() 2917 */ 2918 if (sock_flag(sk, SOCK_DEAD)) 2919 goto out; 2920 2921 sk->sk_err = err; 2922 sk_error_report(sk); 2923 __udp_disconnect(sk, 0); 2924 2925 out: 2926 release_sock(sk); 2927 2928 return 0; 2929 } 2930 EXPORT_SYMBOL_GPL(udp_abort); 2931 2932 struct proto udp_prot = { 2933 .name = "UDP", 2934 .owner = THIS_MODULE, 2935 .close = udp_lib_close, 2936 .pre_connect = udp_pre_connect, 2937 .connect = ip4_datagram_connect, 2938 .disconnect = udp_disconnect, 2939 .ioctl = udp_ioctl, 2940 .init = udp_init_sock, 2941 .destroy = udp_destroy_sock, 2942 .setsockopt = udp_setsockopt, 2943 .getsockopt = udp_getsockopt, 2944 .sendmsg = udp_sendmsg, 2945 .recvmsg = udp_recvmsg, 2946 .sendpage = udp_sendpage, 2947 .release_cb = ip4_datagram_release_cb, 2948 .hash = udp_lib_hash, 2949 .unhash = udp_lib_unhash, 2950 .rehash = udp_v4_rehash, 2951 .get_port = udp_v4_get_port, 2952 .put_port = udp_lib_unhash, 2953 #ifdef CONFIG_BPF_SYSCALL 2954 .psock_update_sk_prot = udp_bpf_update_proto, 2955 #endif 2956 .memory_allocated = &udp_memory_allocated, 2957 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 2958 2959 .sysctl_mem = sysctl_udp_mem, 2960 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2961 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2962 .obj_size = sizeof(struct udp_sock), 2963 .h.udp_table = &udp_table, 2964 .diag_destroy = udp_abort, 2965 }; 2966 EXPORT_SYMBOL(udp_prot); 2967 2968 /* ------------------------------------------------------------------------ */ 2969 #ifdef CONFIG_PROC_FS 2970 2971 static struct sock *udp_get_first(struct seq_file *seq, int start) 2972 { 2973 struct sock *sk; 2974 struct udp_seq_afinfo *afinfo; 2975 struct udp_iter_state *state = seq->private; 2976 struct net *net = seq_file_net(seq); 2977 2978 if (state->bpf_seq_afinfo) 2979 afinfo = state->bpf_seq_afinfo; 2980 else 2981 afinfo = pde_data(file_inode(seq->file)); 2982 2983 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; 2984 ++state->bucket) { 2985 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; 2986 2987 if (hlist_empty(&hslot->head)) 2988 continue; 2989 2990 spin_lock_bh(&hslot->lock); 2991 sk_for_each(sk, &hslot->head) { 2992 if (!net_eq(sock_net(sk), net)) 2993 continue; 2994 if (afinfo->family == AF_UNSPEC || 2995 sk->sk_family == afinfo->family) 2996 goto found; 2997 } 2998 spin_unlock_bh(&hslot->lock); 2999 } 3000 sk = NULL; 3001 found: 3002 return sk; 3003 } 3004 3005 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 3006 { 3007 struct udp_seq_afinfo *afinfo; 3008 struct udp_iter_state *state = seq->private; 3009 struct net *net = seq_file_net(seq); 3010 3011 if (state->bpf_seq_afinfo) 3012 afinfo = state->bpf_seq_afinfo; 3013 else 3014 afinfo = pde_data(file_inode(seq->file)); 3015 3016 do { 3017 sk = sk_next(sk); 3018 } while (sk && (!net_eq(sock_net(sk), net) || 3019 (afinfo->family != AF_UNSPEC && 3020 sk->sk_family != afinfo->family))); 3021 3022 if (!sk) { 3023 if (state->bucket <= afinfo->udp_table->mask) 3024 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 3025 return udp_get_first(seq, state->bucket + 1); 3026 } 3027 return sk; 3028 } 3029 3030 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3031 { 3032 struct sock *sk = udp_get_first(seq, 0); 3033 3034 if (sk) 3035 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3036 --pos; 3037 return pos ? NULL : sk; 3038 } 3039 3040 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3041 { 3042 struct udp_iter_state *state = seq->private; 3043 state->bucket = MAX_UDP_PORTS; 3044 3045 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3046 } 3047 EXPORT_SYMBOL(udp_seq_start); 3048 3049 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3050 { 3051 struct sock *sk; 3052 3053 if (v == SEQ_START_TOKEN) 3054 sk = udp_get_idx(seq, 0); 3055 else 3056 sk = udp_get_next(seq, v); 3057 3058 ++*pos; 3059 return sk; 3060 } 3061 EXPORT_SYMBOL(udp_seq_next); 3062 3063 void udp_seq_stop(struct seq_file *seq, void *v) 3064 { 3065 struct udp_seq_afinfo *afinfo; 3066 struct udp_iter_state *state = seq->private; 3067 3068 if (state->bpf_seq_afinfo) 3069 afinfo = state->bpf_seq_afinfo; 3070 else 3071 afinfo = pde_data(file_inode(seq->file)); 3072 3073 if (state->bucket <= afinfo->udp_table->mask) 3074 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 3075 } 3076 EXPORT_SYMBOL(udp_seq_stop); 3077 3078 /* ------------------------------------------------------------------------ */ 3079 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3080 int bucket) 3081 { 3082 struct inet_sock *inet = inet_sk(sp); 3083 __be32 dest = inet->inet_daddr; 3084 __be32 src = inet->inet_rcv_saddr; 3085 __u16 destp = ntohs(inet->inet_dport); 3086 __u16 srcp = ntohs(inet->inet_sport); 3087 3088 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3089 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3090 bucket, src, srcp, dest, destp, sp->sk_state, 3091 sk_wmem_alloc_get(sp), 3092 udp_rqueue_get(sp), 3093 0, 0L, 0, 3094 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3095 0, sock_i_ino(sp), 3096 refcount_read(&sp->sk_refcnt), sp, 3097 atomic_read(&sp->sk_drops)); 3098 } 3099 3100 int udp4_seq_show(struct seq_file *seq, void *v) 3101 { 3102 seq_setwidth(seq, 127); 3103 if (v == SEQ_START_TOKEN) 3104 seq_puts(seq, " sl local_address rem_address st tx_queue " 3105 "rx_queue tr tm->when retrnsmt uid timeout " 3106 "inode ref pointer drops"); 3107 else { 3108 struct udp_iter_state *state = seq->private; 3109 3110 udp4_format_sock(v, seq, state->bucket); 3111 } 3112 seq_pad(seq, '\n'); 3113 return 0; 3114 } 3115 3116 #ifdef CONFIG_BPF_SYSCALL 3117 struct bpf_iter__udp { 3118 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3119 __bpf_md_ptr(struct udp_sock *, udp_sk); 3120 uid_t uid __aligned(8); 3121 int bucket __aligned(8); 3122 }; 3123 3124 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3125 struct udp_sock *udp_sk, uid_t uid, int bucket) 3126 { 3127 struct bpf_iter__udp ctx; 3128 3129 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3130 ctx.meta = meta; 3131 ctx.udp_sk = udp_sk; 3132 ctx.uid = uid; 3133 ctx.bucket = bucket; 3134 return bpf_iter_run_prog(prog, &ctx); 3135 } 3136 3137 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3138 { 3139 struct udp_iter_state *state = seq->private; 3140 struct bpf_iter_meta meta; 3141 struct bpf_prog *prog; 3142 struct sock *sk = v; 3143 uid_t uid; 3144 3145 if (v == SEQ_START_TOKEN) 3146 return 0; 3147 3148 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3149 meta.seq = seq; 3150 prog = bpf_iter_get_info(&meta, false); 3151 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3152 } 3153 3154 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3155 { 3156 struct bpf_iter_meta meta; 3157 struct bpf_prog *prog; 3158 3159 if (!v) { 3160 meta.seq = seq; 3161 prog = bpf_iter_get_info(&meta, true); 3162 if (prog) 3163 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3164 } 3165 3166 udp_seq_stop(seq, v); 3167 } 3168 3169 static const struct seq_operations bpf_iter_udp_seq_ops = { 3170 .start = udp_seq_start, 3171 .next = udp_seq_next, 3172 .stop = bpf_iter_udp_seq_stop, 3173 .show = bpf_iter_udp_seq_show, 3174 }; 3175 #endif 3176 3177 const struct seq_operations udp_seq_ops = { 3178 .start = udp_seq_start, 3179 .next = udp_seq_next, 3180 .stop = udp_seq_stop, 3181 .show = udp4_seq_show, 3182 }; 3183 EXPORT_SYMBOL(udp_seq_ops); 3184 3185 static struct udp_seq_afinfo udp4_seq_afinfo = { 3186 .family = AF_INET, 3187 .udp_table = &udp_table, 3188 }; 3189 3190 static int __net_init udp4_proc_init_net(struct net *net) 3191 { 3192 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3193 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3194 return -ENOMEM; 3195 return 0; 3196 } 3197 3198 static void __net_exit udp4_proc_exit_net(struct net *net) 3199 { 3200 remove_proc_entry("udp", net->proc_net); 3201 } 3202 3203 static struct pernet_operations udp4_net_ops = { 3204 .init = udp4_proc_init_net, 3205 .exit = udp4_proc_exit_net, 3206 }; 3207 3208 int __init udp4_proc_init(void) 3209 { 3210 return register_pernet_subsys(&udp4_net_ops); 3211 } 3212 3213 void udp4_proc_exit(void) 3214 { 3215 unregister_pernet_subsys(&udp4_net_ops); 3216 } 3217 #endif /* CONFIG_PROC_FS */ 3218 3219 static __initdata unsigned long uhash_entries; 3220 static int __init set_uhash_entries(char *str) 3221 { 3222 ssize_t ret; 3223 3224 if (!str) 3225 return 0; 3226 3227 ret = kstrtoul(str, 0, &uhash_entries); 3228 if (ret) 3229 return 0; 3230 3231 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3232 uhash_entries = UDP_HTABLE_SIZE_MIN; 3233 return 1; 3234 } 3235 __setup("uhash_entries=", set_uhash_entries); 3236 3237 void __init udp_table_init(struct udp_table *table, const char *name) 3238 { 3239 unsigned int i; 3240 3241 table->hash = alloc_large_system_hash(name, 3242 2 * sizeof(struct udp_hslot), 3243 uhash_entries, 3244 21, /* one slot per 2 MB */ 3245 0, 3246 &table->log, 3247 &table->mask, 3248 UDP_HTABLE_SIZE_MIN, 3249 64 * 1024); 3250 3251 table->hash2 = table->hash + (table->mask + 1); 3252 for (i = 0; i <= table->mask; i++) { 3253 INIT_HLIST_HEAD(&table->hash[i].head); 3254 table->hash[i].count = 0; 3255 spin_lock_init(&table->hash[i].lock); 3256 } 3257 for (i = 0; i <= table->mask; i++) { 3258 INIT_HLIST_HEAD(&table->hash2[i].head); 3259 table->hash2[i].count = 0; 3260 spin_lock_init(&table->hash2[i].lock); 3261 } 3262 } 3263 3264 u32 udp_flow_hashrnd(void) 3265 { 3266 static u32 hashrnd __read_mostly; 3267 3268 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3269 3270 return hashrnd; 3271 } 3272 EXPORT_SYMBOL(udp_flow_hashrnd); 3273 3274 static int __net_init udp_sysctl_init(struct net *net) 3275 { 3276 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3277 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 3278 3279 #ifdef CONFIG_NET_L3_MASTER_DEV 3280 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3281 #endif 3282 3283 return 0; 3284 } 3285 3286 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3287 .init = udp_sysctl_init, 3288 }; 3289 3290 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3291 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3292 struct udp_sock *udp_sk, uid_t uid, int bucket) 3293 3294 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3295 { 3296 struct udp_iter_state *st = priv_data; 3297 struct udp_seq_afinfo *afinfo; 3298 int ret; 3299 3300 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN); 3301 if (!afinfo) 3302 return -ENOMEM; 3303 3304 afinfo->family = AF_UNSPEC; 3305 afinfo->udp_table = &udp_table; 3306 st->bpf_seq_afinfo = afinfo; 3307 ret = bpf_iter_init_seq_net(priv_data, aux); 3308 if (ret) 3309 kfree(afinfo); 3310 return ret; 3311 } 3312 3313 static void bpf_iter_fini_udp(void *priv_data) 3314 { 3315 struct udp_iter_state *st = priv_data; 3316 3317 kfree(st->bpf_seq_afinfo); 3318 bpf_iter_fini_seq_net(priv_data); 3319 } 3320 3321 static const struct bpf_iter_seq_info udp_seq_info = { 3322 .seq_ops = &bpf_iter_udp_seq_ops, 3323 .init_seq_private = bpf_iter_init_udp, 3324 .fini_seq_private = bpf_iter_fini_udp, 3325 .seq_priv_size = sizeof(struct udp_iter_state), 3326 }; 3327 3328 static struct bpf_iter_reg udp_reg_info = { 3329 .target = "udp", 3330 .ctx_arg_info_size = 1, 3331 .ctx_arg_info = { 3332 { offsetof(struct bpf_iter__udp, udp_sk), 3333 PTR_TO_BTF_ID_OR_NULL }, 3334 }, 3335 .seq_info = &udp_seq_info, 3336 }; 3337 3338 static void __init bpf_iter_register(void) 3339 { 3340 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3341 if (bpf_iter_reg_target(&udp_reg_info)) 3342 pr_warn("Warning: could not register bpf iterator udp\n"); 3343 } 3344 #endif 3345 3346 void __init udp_init(void) 3347 { 3348 unsigned long limit; 3349 unsigned int i; 3350 3351 udp_table_init(&udp_table, "UDP"); 3352 limit = nr_free_buffer_pages() / 8; 3353 limit = max(limit, 128UL); 3354 sysctl_udp_mem[0] = limit / 4 * 3; 3355 sysctl_udp_mem[1] = limit; 3356 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3357 3358 /* 16 spinlocks per cpu */ 3359 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3360 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3361 GFP_KERNEL); 3362 if (!udp_busylocks) 3363 panic("UDP: failed to alloc udp_busylocks\n"); 3364 for (i = 0; i < (1U << udp_busylocks_log); i++) 3365 spin_lock_init(udp_busylocks + i); 3366 3367 if (register_pernet_subsys(&udp_sysctl_ops)) 3368 panic("UDP: failed to init sysctl parameters.\n"); 3369 3370 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3371 bpf_iter_register(); 3372 #endif 3373 } 3374