1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * The User Datagram Protocol (UDP). 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 12 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 13 * Hirokazu Takahashi, <taka@valinux.co.jp> 14 * 15 * Fixes: 16 * Alan Cox : verify_area() calls 17 * Alan Cox : stopped close while in use off icmp 18 * messages. Not a fix but a botch that 19 * for udp at least is 'valid'. 20 * Alan Cox : Fixed icmp handling properly 21 * Alan Cox : Correct error for oversized datagrams 22 * Alan Cox : Tidied select() semantics. 23 * Alan Cox : udp_err() fixed properly, also now 24 * select and read wake correctly on errors 25 * Alan Cox : udp_send verify_area moved to avoid mem leak 26 * Alan Cox : UDP can count its memory 27 * Alan Cox : send to an unknown connection causes 28 * an ECONNREFUSED off the icmp, but 29 * does NOT close. 30 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 32 * bug no longer crashes it. 33 * Fred Van Kempen : Net2e support for sk->broadcast. 34 * Alan Cox : Uses skb_free_datagram 35 * Alan Cox : Added get/set sockopt support. 36 * Alan Cox : Broadcasting without option set returns EACCES. 37 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 38 * Alan Cox : Use ip_tos and ip_ttl 39 * Alan Cox : SNMP Mibs 40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 41 * Matt Dillon : UDP length checks. 42 * Alan Cox : Smarter af_inet used properly. 43 * Alan Cox : Use new kernel side addressing. 44 * Alan Cox : Incorrect return on truncated datagram receive. 45 * Arnt Gulbrandsen : New udp_send and stuff 46 * Alan Cox : Cache last socket 47 * Alan Cox : Route cache 48 * Jon Peatfield : Minor efficiency fix to sendto(). 49 * Mike Shaver : RFC1122 checks. 50 * Alan Cox : Nonblocking error fix. 51 * Willy Konynenberg : Transparent proxying support. 52 * Mike McLagan : Routing by source 53 * David S. Miller : New socket lookup architecture. 54 * Last socket cache retained as it 55 * does have a high hit rate. 56 * Olaf Kirch : Don't linearise iovec on sendmsg. 57 * Andi Kleen : Some cleanups, cache destination entry 58 * for connect. 59 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 60 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 61 * return ENOTCONN for unconnected sockets (POSIX) 62 * Janos Farkas : don't deliver multi/broadcasts to a different 63 * bound-to-device socket 64 * Hirokazu Takahashi : HW checksumming for outgoing UDP 65 * datagrams. 66 * Hirokazu Takahashi : sendfile() on UDP works now. 67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 70 * a single port at the same time. 71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 72 * James Chapman : Add L2TP encapsulation type. 73 */ 74 75 #define pr_fmt(fmt) "UDP: " fmt 76 77 #include <linux/bpf-cgroup.h> 78 #include <linux/uaccess.h> 79 #include <asm/ioctls.h> 80 #include <linux/memblock.h> 81 #include <linux/highmem.h> 82 #include <linux/types.h> 83 #include <linux/fcntl.h> 84 #include <linux/module.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/igmp.h> 88 #include <linux/inetdevice.h> 89 #include <linux/in.h> 90 #include <linux/errno.h> 91 #include <linux/timer.h> 92 #include <linux/mm.h> 93 #include <linux/inet.h> 94 #include <linux/netdevice.h> 95 #include <linux/slab.h> 96 #include <net/tcp_states.h> 97 #include <linux/skbuff.h> 98 #include <linux/proc_fs.h> 99 #include <linux/seq_file.h> 100 #include <net/net_namespace.h> 101 #include <net/icmp.h> 102 #include <net/inet_hashtables.h> 103 #include <net/ip_tunnels.h> 104 #include <net/route.h> 105 #include <net/checksum.h> 106 #include <net/xfrm.h> 107 #include <trace/events/udp.h> 108 #include <linux/static_key.h> 109 #include <linux/btf_ids.h> 110 #include <trace/events/skb.h> 111 #include <net/busy_poll.h> 112 #include "udp_impl.h" 113 #include <net/sock_reuseport.h> 114 #include <net/addrconf.h> 115 #include <net/udp_tunnel.h> 116 #if IS_ENABLED(CONFIG_IPV6) 117 #include <net/ipv6_stubs.h> 118 #endif 119 120 struct udp_table udp_table __read_mostly; 121 EXPORT_SYMBOL(udp_table); 122 123 long sysctl_udp_mem[3] __read_mostly; 124 EXPORT_SYMBOL(sysctl_udp_mem); 125 126 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp; 127 EXPORT_SYMBOL(udp_memory_allocated); 128 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc); 129 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc); 130 131 #define MAX_UDP_PORTS 65536 132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 133 134 static int udp_lib_lport_inuse(struct net *net, __u16 num, 135 const struct udp_hslot *hslot, 136 unsigned long *bitmap, 137 struct sock *sk, unsigned int log) 138 { 139 struct sock *sk2; 140 kuid_t uid = sock_i_uid(sk); 141 142 sk_for_each(sk2, &hslot->head) { 143 if (net_eq(sock_net(sk2), net) && 144 sk2 != sk && 145 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 146 (!sk2->sk_reuse || !sk->sk_reuse) && 147 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 148 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 149 inet_rcv_saddr_equal(sk, sk2, true)) { 150 if (sk2->sk_reuseport && sk->sk_reuseport && 151 !rcu_access_pointer(sk->sk_reuseport_cb) && 152 uid_eq(uid, sock_i_uid(sk2))) { 153 if (!bitmap) 154 return 0; 155 } else { 156 if (!bitmap) 157 return 1; 158 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 159 bitmap); 160 } 161 } 162 } 163 return 0; 164 } 165 166 /* 167 * Note: we still hold spinlock of primary hash chain, so no other writer 168 * can insert/delete a socket with local_port == num 169 */ 170 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 171 struct udp_hslot *hslot2, 172 struct sock *sk) 173 { 174 struct sock *sk2; 175 kuid_t uid = sock_i_uid(sk); 176 int res = 0; 177 178 spin_lock(&hslot2->lock); 179 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 180 if (net_eq(sock_net(sk2), net) && 181 sk2 != sk && 182 (udp_sk(sk2)->udp_port_hash == num) && 183 (!sk2->sk_reuse || !sk->sk_reuse) && 184 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 185 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 186 inet_rcv_saddr_equal(sk, sk2, true)) { 187 if (sk2->sk_reuseport && sk->sk_reuseport && 188 !rcu_access_pointer(sk->sk_reuseport_cb) && 189 uid_eq(uid, sock_i_uid(sk2))) { 190 res = 0; 191 } else { 192 res = 1; 193 } 194 break; 195 } 196 } 197 spin_unlock(&hslot2->lock); 198 return res; 199 } 200 201 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 202 { 203 struct net *net = sock_net(sk); 204 kuid_t uid = sock_i_uid(sk); 205 struct sock *sk2; 206 207 sk_for_each(sk2, &hslot->head) { 208 if (net_eq(sock_net(sk2), net) && 209 sk2 != sk && 210 sk2->sk_family == sk->sk_family && 211 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 212 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 213 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 214 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 215 inet_rcv_saddr_equal(sk, sk2, false)) { 216 return reuseport_add_sock(sk, sk2, 217 inet_rcv_saddr_any(sk)); 218 } 219 } 220 221 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 222 } 223 224 /** 225 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 226 * 227 * @sk: socket struct in question 228 * @snum: port number to look up 229 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 230 * with NULL address 231 */ 232 int udp_lib_get_port(struct sock *sk, unsigned short snum, 233 unsigned int hash2_nulladdr) 234 { 235 struct udp_hslot *hslot, *hslot2; 236 struct udp_table *udptable = sk->sk_prot->h.udp_table; 237 int error = 1; 238 struct net *net = sock_net(sk); 239 240 if (!snum) { 241 int low, high, remaining; 242 unsigned int rand; 243 unsigned short first, last; 244 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 245 246 inet_get_local_port_range(net, &low, &high); 247 remaining = (high - low) + 1; 248 249 rand = prandom_u32(); 250 first = reciprocal_scale(rand, remaining) + low; 251 /* 252 * force rand to be an odd multiple of UDP_HTABLE_SIZE 253 */ 254 rand = (rand | 1) * (udptable->mask + 1); 255 last = first + udptable->mask + 1; 256 do { 257 hslot = udp_hashslot(udptable, net, first); 258 bitmap_zero(bitmap, PORTS_PER_CHAIN); 259 spin_lock_bh(&hslot->lock); 260 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 261 udptable->log); 262 263 snum = first; 264 /* 265 * Iterate on all possible values of snum for this hash. 266 * Using steps of an odd multiple of UDP_HTABLE_SIZE 267 * give us randomization and full range coverage. 268 */ 269 do { 270 if (low <= snum && snum <= high && 271 !test_bit(snum >> udptable->log, bitmap) && 272 !inet_is_local_reserved_port(net, snum)) 273 goto found; 274 snum += rand; 275 } while (snum != first); 276 spin_unlock_bh(&hslot->lock); 277 cond_resched(); 278 } while (++first != last); 279 goto fail; 280 } else { 281 hslot = udp_hashslot(udptable, net, snum); 282 spin_lock_bh(&hslot->lock); 283 if (hslot->count > 10) { 284 int exist; 285 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 286 287 slot2 &= udptable->mask; 288 hash2_nulladdr &= udptable->mask; 289 290 hslot2 = udp_hashslot2(udptable, slot2); 291 if (hslot->count < hslot2->count) 292 goto scan_primary_hash; 293 294 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 295 if (!exist && (hash2_nulladdr != slot2)) { 296 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 297 exist = udp_lib_lport_inuse2(net, snum, hslot2, 298 sk); 299 } 300 if (exist) 301 goto fail_unlock; 302 else 303 goto found; 304 } 305 scan_primary_hash: 306 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 307 goto fail_unlock; 308 } 309 found: 310 inet_sk(sk)->inet_num = snum; 311 udp_sk(sk)->udp_port_hash = snum; 312 udp_sk(sk)->udp_portaddr_hash ^= snum; 313 if (sk_unhashed(sk)) { 314 if (sk->sk_reuseport && 315 udp_reuseport_add_sock(sk, hslot)) { 316 inet_sk(sk)->inet_num = 0; 317 udp_sk(sk)->udp_port_hash = 0; 318 udp_sk(sk)->udp_portaddr_hash ^= snum; 319 goto fail_unlock; 320 } 321 322 sk_add_node_rcu(sk, &hslot->head); 323 hslot->count++; 324 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 325 326 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 327 spin_lock(&hslot2->lock); 328 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 329 sk->sk_family == AF_INET6) 330 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 331 &hslot2->head); 332 else 333 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 334 &hslot2->head); 335 hslot2->count++; 336 spin_unlock(&hslot2->lock); 337 } 338 sock_set_flag(sk, SOCK_RCU_FREE); 339 error = 0; 340 fail_unlock: 341 spin_unlock_bh(&hslot->lock); 342 fail: 343 return error; 344 } 345 EXPORT_SYMBOL(udp_lib_get_port); 346 347 int udp_v4_get_port(struct sock *sk, unsigned short snum) 348 { 349 unsigned int hash2_nulladdr = 350 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 351 unsigned int hash2_partial = 352 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 353 354 /* precompute partial secondary hash */ 355 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 356 return udp_lib_get_port(sk, snum, hash2_nulladdr); 357 } 358 359 static int compute_score(struct sock *sk, struct net *net, 360 __be32 saddr, __be16 sport, 361 __be32 daddr, unsigned short hnum, 362 int dif, int sdif) 363 { 364 int score; 365 struct inet_sock *inet; 366 bool dev_match; 367 368 if (!net_eq(sock_net(sk), net) || 369 udp_sk(sk)->udp_port_hash != hnum || 370 ipv6_only_sock(sk)) 371 return -1; 372 373 if (sk->sk_rcv_saddr != daddr) 374 return -1; 375 376 score = (sk->sk_family == PF_INET) ? 2 : 1; 377 378 inet = inet_sk(sk); 379 if (inet->inet_daddr) { 380 if (inet->inet_daddr != saddr) 381 return -1; 382 score += 4; 383 } 384 385 if (inet->inet_dport) { 386 if (inet->inet_dport != sport) 387 return -1; 388 score += 4; 389 } 390 391 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 392 dif, sdif); 393 if (!dev_match) 394 return -1; 395 if (sk->sk_bound_dev_if) 396 score += 4; 397 398 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 399 score++; 400 return score; 401 } 402 403 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 404 const __u16 lport, const __be32 faddr, 405 const __be16 fport) 406 { 407 static u32 udp_ehash_secret __read_mostly; 408 409 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 410 411 return __inet_ehashfn(laddr, lport, faddr, fport, 412 udp_ehash_secret + net_hash_mix(net)); 413 } 414 415 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 416 struct sk_buff *skb, 417 __be32 saddr, __be16 sport, 418 __be32 daddr, unsigned short hnum) 419 { 420 struct sock *reuse_sk = NULL; 421 u32 hash; 422 423 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 424 hash = udp_ehashfn(net, daddr, hnum, saddr, sport); 425 reuse_sk = reuseport_select_sock(sk, hash, skb, 426 sizeof(struct udphdr)); 427 } 428 return reuse_sk; 429 } 430 431 /* called with rcu_read_lock() */ 432 static struct sock *udp4_lib_lookup2(struct net *net, 433 __be32 saddr, __be16 sport, 434 __be32 daddr, unsigned int hnum, 435 int dif, int sdif, 436 struct udp_hslot *hslot2, 437 struct sk_buff *skb) 438 { 439 struct sock *sk, *result; 440 int score, badness; 441 442 result = NULL; 443 badness = 0; 444 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 445 score = compute_score(sk, net, saddr, sport, 446 daddr, hnum, dif, sdif); 447 if (score > badness) { 448 result = lookup_reuseport(net, sk, skb, 449 saddr, sport, daddr, hnum); 450 /* Fall back to scoring if group has connections */ 451 if (result && !reuseport_has_conns(sk, false)) 452 return result; 453 454 result = result ? : sk; 455 badness = score; 456 } 457 } 458 return result; 459 } 460 461 static struct sock *udp4_lookup_run_bpf(struct net *net, 462 struct udp_table *udptable, 463 struct sk_buff *skb, 464 __be32 saddr, __be16 sport, 465 __be32 daddr, u16 hnum, const int dif) 466 { 467 struct sock *sk, *reuse_sk; 468 bool no_reuseport; 469 470 if (udptable != &udp_table) 471 return NULL; /* only UDP is supported */ 472 473 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport, 474 daddr, hnum, dif, &sk); 475 if (no_reuseport || IS_ERR_OR_NULL(sk)) 476 return sk; 477 478 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 479 if (reuse_sk) 480 sk = reuse_sk; 481 return sk; 482 } 483 484 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 485 * harder than this. -DaveM 486 */ 487 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 488 __be16 sport, __be32 daddr, __be16 dport, int dif, 489 int sdif, struct udp_table *udptable, struct sk_buff *skb) 490 { 491 unsigned short hnum = ntohs(dport); 492 unsigned int hash2, slot2; 493 struct udp_hslot *hslot2; 494 struct sock *result, *sk; 495 496 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 497 slot2 = hash2 & udptable->mask; 498 hslot2 = &udptable->hash2[slot2]; 499 500 /* Lookup connected or non-wildcard socket */ 501 result = udp4_lib_lookup2(net, saddr, sport, 502 daddr, hnum, dif, sdif, 503 hslot2, skb); 504 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 505 goto done; 506 507 /* Lookup redirect from BPF */ 508 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { 509 sk = udp4_lookup_run_bpf(net, udptable, skb, 510 saddr, sport, daddr, hnum, dif); 511 if (sk) { 512 result = sk; 513 goto done; 514 } 515 } 516 517 /* Got non-wildcard socket or error on first lookup */ 518 if (result) 519 goto done; 520 521 /* Lookup wildcard sockets */ 522 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 523 slot2 = hash2 & udptable->mask; 524 hslot2 = &udptable->hash2[slot2]; 525 526 result = udp4_lib_lookup2(net, saddr, sport, 527 htonl(INADDR_ANY), hnum, dif, sdif, 528 hslot2, skb); 529 done: 530 if (IS_ERR(result)) 531 return NULL; 532 return result; 533 } 534 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 535 536 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 537 __be16 sport, __be16 dport, 538 struct udp_table *udptable) 539 { 540 const struct iphdr *iph = ip_hdr(skb); 541 542 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 543 iph->daddr, dport, inet_iif(skb), 544 inet_sdif(skb), udptable, skb); 545 } 546 547 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 548 __be16 sport, __be16 dport) 549 { 550 const struct iphdr *iph = ip_hdr(skb); 551 552 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 553 iph->daddr, dport, inet_iif(skb), 554 inet_sdif(skb), &udp_table, NULL); 555 } 556 557 /* Must be called under rcu_read_lock(). 558 * Does increment socket refcount. 559 */ 560 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 561 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 562 __be32 daddr, __be16 dport, int dif) 563 { 564 struct sock *sk; 565 566 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 567 dif, 0, &udp_table, NULL); 568 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 569 sk = NULL; 570 return sk; 571 } 572 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 573 #endif 574 575 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 576 __be16 loc_port, __be32 loc_addr, 577 __be16 rmt_port, __be32 rmt_addr, 578 int dif, int sdif, unsigned short hnum) 579 { 580 struct inet_sock *inet = inet_sk(sk); 581 582 if (!net_eq(sock_net(sk), net) || 583 udp_sk(sk)->udp_port_hash != hnum || 584 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 585 (inet->inet_dport != rmt_port && inet->inet_dport) || 586 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 587 ipv6_only_sock(sk) || 588 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 589 return false; 590 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 591 return false; 592 return true; 593 } 594 595 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 596 void udp_encap_enable(void) 597 { 598 static_branch_inc(&udp_encap_needed_key); 599 } 600 EXPORT_SYMBOL(udp_encap_enable); 601 602 void udp_encap_disable(void) 603 { 604 static_branch_dec(&udp_encap_needed_key); 605 } 606 EXPORT_SYMBOL(udp_encap_disable); 607 608 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 609 * through error handlers in encapsulations looking for a match. 610 */ 611 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 612 { 613 int i; 614 615 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 616 int (*handler)(struct sk_buff *skb, u32 info); 617 const struct ip_tunnel_encap_ops *encap; 618 619 encap = rcu_dereference(iptun_encaps[i]); 620 if (!encap) 621 continue; 622 handler = encap->err_handler; 623 if (handler && !handler(skb, info)) 624 return 0; 625 } 626 627 return -ENOENT; 628 } 629 630 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 631 * reversing source and destination port: this will match tunnels that force the 632 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 633 * lwtunnels might actually break this assumption by being configured with 634 * different destination ports on endpoints, in this case we won't be able to 635 * trace ICMP messages back to them. 636 * 637 * If this doesn't match any socket, probe tunnels with arbitrary destination 638 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 639 * we've sent packets to won't necessarily match the local destination port. 640 * 641 * Then ask the tunnel implementation to match the error against a valid 642 * association. 643 * 644 * Return an error if we can't find a match, the socket if we need further 645 * processing, zero otherwise. 646 */ 647 static struct sock *__udp4_lib_err_encap(struct net *net, 648 const struct iphdr *iph, 649 struct udphdr *uh, 650 struct udp_table *udptable, 651 struct sock *sk, 652 struct sk_buff *skb, u32 info) 653 { 654 int (*lookup)(struct sock *sk, struct sk_buff *skb); 655 int network_offset, transport_offset; 656 struct udp_sock *up; 657 658 network_offset = skb_network_offset(skb); 659 transport_offset = skb_transport_offset(skb); 660 661 /* Network header needs to point to the outer IPv4 header inside ICMP */ 662 skb_reset_network_header(skb); 663 664 /* Transport header needs to point to the UDP header */ 665 skb_set_transport_header(skb, iph->ihl << 2); 666 667 if (sk) { 668 up = udp_sk(sk); 669 670 lookup = READ_ONCE(up->encap_err_lookup); 671 if (lookup && lookup(sk, skb)) 672 sk = NULL; 673 674 goto out; 675 } 676 677 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 678 iph->saddr, uh->dest, skb->dev->ifindex, 0, 679 udptable, NULL); 680 if (sk) { 681 up = udp_sk(sk); 682 683 lookup = READ_ONCE(up->encap_err_lookup); 684 if (!lookup || lookup(sk, skb)) 685 sk = NULL; 686 } 687 688 out: 689 if (!sk) 690 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 691 692 skb_set_transport_header(skb, transport_offset); 693 skb_set_network_header(skb, network_offset); 694 695 return sk; 696 } 697 698 /* 699 * This routine is called by the ICMP module when it gets some 700 * sort of error condition. If err < 0 then the socket should 701 * be closed and the error returned to the user. If err > 0 702 * it's just the icmp type << 8 | icmp code. 703 * Header points to the ip header of the error packet. We move 704 * on past this. Then (as it used to claim before adjustment) 705 * header points to the first 8 bytes of the udp header. We need 706 * to find the appropriate port. 707 */ 708 709 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 710 { 711 struct inet_sock *inet; 712 const struct iphdr *iph = (const struct iphdr *)skb->data; 713 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 714 const int type = icmp_hdr(skb)->type; 715 const int code = icmp_hdr(skb)->code; 716 bool tunnel = false; 717 struct sock *sk; 718 int harderr; 719 int err; 720 struct net *net = dev_net(skb->dev); 721 722 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 723 iph->saddr, uh->source, skb->dev->ifindex, 724 inet_sdif(skb), udptable, NULL); 725 726 if (!sk || udp_sk(sk)->encap_type) { 727 /* No socket for error: try tunnels before discarding */ 728 if (static_branch_unlikely(&udp_encap_needed_key)) { 729 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 730 info); 731 if (!sk) 732 return 0; 733 } else 734 sk = ERR_PTR(-ENOENT); 735 736 if (IS_ERR(sk)) { 737 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 738 return PTR_ERR(sk); 739 } 740 741 tunnel = true; 742 } 743 744 err = 0; 745 harderr = 0; 746 inet = inet_sk(sk); 747 748 switch (type) { 749 default: 750 case ICMP_TIME_EXCEEDED: 751 err = EHOSTUNREACH; 752 break; 753 case ICMP_SOURCE_QUENCH: 754 goto out; 755 case ICMP_PARAMETERPROB: 756 err = EPROTO; 757 harderr = 1; 758 break; 759 case ICMP_DEST_UNREACH: 760 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 761 ipv4_sk_update_pmtu(skb, sk, info); 762 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 763 err = EMSGSIZE; 764 harderr = 1; 765 break; 766 } 767 goto out; 768 } 769 err = EHOSTUNREACH; 770 if (code <= NR_ICMP_UNREACH) { 771 harderr = icmp_err_convert[code].fatal; 772 err = icmp_err_convert[code].errno; 773 } 774 break; 775 case ICMP_REDIRECT: 776 ipv4_sk_redirect(skb, sk); 777 goto out; 778 } 779 780 /* 781 * RFC1122: OK. Passes ICMP errors back to application, as per 782 * 4.1.3.3. 783 */ 784 if (tunnel) { 785 /* ...not for tunnels though: we don't have a sending socket */ 786 goto out; 787 } 788 if (!inet->recverr) { 789 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 790 goto out; 791 } else 792 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 793 794 sk->sk_err = err; 795 sk_error_report(sk); 796 out: 797 return 0; 798 } 799 800 int udp_err(struct sk_buff *skb, u32 info) 801 { 802 return __udp4_lib_err(skb, info, &udp_table); 803 } 804 805 /* 806 * Throw away all pending data and cancel the corking. Socket is locked. 807 */ 808 void udp_flush_pending_frames(struct sock *sk) 809 { 810 struct udp_sock *up = udp_sk(sk); 811 812 if (up->pending) { 813 up->len = 0; 814 up->pending = 0; 815 ip_flush_pending_frames(sk); 816 } 817 } 818 EXPORT_SYMBOL(udp_flush_pending_frames); 819 820 /** 821 * udp4_hwcsum - handle outgoing HW checksumming 822 * @skb: sk_buff containing the filled-in UDP header 823 * (checksum field must be zeroed out) 824 * @src: source IP address 825 * @dst: destination IP address 826 */ 827 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 828 { 829 struct udphdr *uh = udp_hdr(skb); 830 int offset = skb_transport_offset(skb); 831 int len = skb->len - offset; 832 int hlen = len; 833 __wsum csum = 0; 834 835 if (!skb_has_frag_list(skb)) { 836 /* 837 * Only one fragment on the socket. 838 */ 839 skb->csum_start = skb_transport_header(skb) - skb->head; 840 skb->csum_offset = offsetof(struct udphdr, check); 841 uh->check = ~csum_tcpudp_magic(src, dst, len, 842 IPPROTO_UDP, 0); 843 } else { 844 struct sk_buff *frags; 845 846 /* 847 * HW-checksum won't work as there are two or more 848 * fragments on the socket so that all csums of sk_buffs 849 * should be together 850 */ 851 skb_walk_frags(skb, frags) { 852 csum = csum_add(csum, frags->csum); 853 hlen -= frags->len; 854 } 855 856 csum = skb_checksum(skb, offset, hlen, csum); 857 skb->ip_summed = CHECKSUM_NONE; 858 859 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 860 if (uh->check == 0) 861 uh->check = CSUM_MANGLED_0; 862 } 863 } 864 EXPORT_SYMBOL_GPL(udp4_hwcsum); 865 866 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 867 * for the simple case like when setting the checksum for a UDP tunnel. 868 */ 869 void udp_set_csum(bool nocheck, struct sk_buff *skb, 870 __be32 saddr, __be32 daddr, int len) 871 { 872 struct udphdr *uh = udp_hdr(skb); 873 874 if (nocheck) { 875 uh->check = 0; 876 } else if (skb_is_gso(skb)) { 877 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 878 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 879 uh->check = 0; 880 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 881 if (uh->check == 0) 882 uh->check = CSUM_MANGLED_0; 883 } else { 884 skb->ip_summed = CHECKSUM_PARTIAL; 885 skb->csum_start = skb_transport_header(skb) - skb->head; 886 skb->csum_offset = offsetof(struct udphdr, check); 887 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 888 } 889 } 890 EXPORT_SYMBOL(udp_set_csum); 891 892 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 893 struct inet_cork *cork) 894 { 895 struct sock *sk = skb->sk; 896 struct inet_sock *inet = inet_sk(sk); 897 struct udphdr *uh; 898 int err; 899 int is_udplite = IS_UDPLITE(sk); 900 int offset = skb_transport_offset(skb); 901 int len = skb->len - offset; 902 int datalen = len - sizeof(*uh); 903 __wsum csum = 0; 904 905 /* 906 * Create a UDP header 907 */ 908 uh = udp_hdr(skb); 909 uh->source = inet->inet_sport; 910 uh->dest = fl4->fl4_dport; 911 uh->len = htons(len); 912 uh->check = 0; 913 914 if (cork->gso_size) { 915 const int hlen = skb_network_header_len(skb) + 916 sizeof(struct udphdr); 917 918 if (hlen + cork->gso_size > cork->fragsize) { 919 kfree_skb(skb); 920 return -EINVAL; 921 } 922 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 923 kfree_skb(skb); 924 return -EINVAL; 925 } 926 if (sk->sk_no_check_tx) { 927 kfree_skb(skb); 928 return -EINVAL; 929 } 930 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 931 dst_xfrm(skb_dst(skb))) { 932 kfree_skb(skb); 933 return -EIO; 934 } 935 936 if (datalen > cork->gso_size) { 937 skb_shinfo(skb)->gso_size = cork->gso_size; 938 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 939 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 940 cork->gso_size); 941 } 942 goto csum_partial; 943 } 944 945 if (is_udplite) /* UDP-Lite */ 946 csum = udplite_csum(skb); 947 948 else if (sk->sk_no_check_tx) { /* UDP csum off */ 949 950 skb->ip_summed = CHECKSUM_NONE; 951 goto send; 952 953 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 954 csum_partial: 955 956 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 957 goto send; 958 959 } else 960 csum = udp_csum(skb); 961 962 /* add protocol-dependent pseudo-header */ 963 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 964 sk->sk_protocol, csum); 965 if (uh->check == 0) 966 uh->check = CSUM_MANGLED_0; 967 968 send: 969 err = ip_send_skb(sock_net(sk), skb); 970 if (err) { 971 if (err == -ENOBUFS && !inet->recverr) { 972 UDP_INC_STATS(sock_net(sk), 973 UDP_MIB_SNDBUFERRORS, is_udplite); 974 err = 0; 975 } 976 } else 977 UDP_INC_STATS(sock_net(sk), 978 UDP_MIB_OUTDATAGRAMS, is_udplite); 979 return err; 980 } 981 982 /* 983 * Push out all pending data as one UDP datagram. Socket is locked. 984 */ 985 int udp_push_pending_frames(struct sock *sk) 986 { 987 struct udp_sock *up = udp_sk(sk); 988 struct inet_sock *inet = inet_sk(sk); 989 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 990 struct sk_buff *skb; 991 int err = 0; 992 993 skb = ip_finish_skb(sk, fl4); 994 if (!skb) 995 goto out; 996 997 err = udp_send_skb(skb, fl4, &inet->cork.base); 998 999 out: 1000 up->len = 0; 1001 up->pending = 0; 1002 return err; 1003 } 1004 EXPORT_SYMBOL(udp_push_pending_frames); 1005 1006 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1007 { 1008 switch (cmsg->cmsg_type) { 1009 case UDP_SEGMENT: 1010 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1011 return -EINVAL; 1012 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1013 return 0; 1014 default: 1015 return -EINVAL; 1016 } 1017 } 1018 1019 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1020 { 1021 struct cmsghdr *cmsg; 1022 bool need_ip = false; 1023 int err; 1024 1025 for_each_cmsghdr(cmsg, msg) { 1026 if (!CMSG_OK(msg, cmsg)) 1027 return -EINVAL; 1028 1029 if (cmsg->cmsg_level != SOL_UDP) { 1030 need_ip = true; 1031 continue; 1032 } 1033 1034 err = __udp_cmsg_send(cmsg, gso_size); 1035 if (err) 1036 return err; 1037 } 1038 1039 return need_ip; 1040 } 1041 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1042 1043 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1044 { 1045 struct inet_sock *inet = inet_sk(sk); 1046 struct udp_sock *up = udp_sk(sk); 1047 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1048 struct flowi4 fl4_stack; 1049 struct flowi4 *fl4; 1050 int ulen = len; 1051 struct ipcm_cookie ipc; 1052 struct rtable *rt = NULL; 1053 int free = 0; 1054 int connected = 0; 1055 __be32 daddr, faddr, saddr; 1056 __be16 dport; 1057 u8 tos; 1058 int err, is_udplite = IS_UDPLITE(sk); 1059 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1060 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1061 struct sk_buff *skb; 1062 struct ip_options_data opt_copy; 1063 1064 if (len > 0xFFFF) 1065 return -EMSGSIZE; 1066 1067 /* 1068 * Check the flags. 1069 */ 1070 1071 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1072 return -EOPNOTSUPP; 1073 1074 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1075 1076 fl4 = &inet->cork.fl.u.ip4; 1077 if (up->pending) { 1078 /* 1079 * There are pending frames. 1080 * The socket lock must be held while it's corked. 1081 */ 1082 lock_sock(sk); 1083 if (likely(up->pending)) { 1084 if (unlikely(up->pending != AF_INET)) { 1085 release_sock(sk); 1086 return -EINVAL; 1087 } 1088 goto do_append_data; 1089 } 1090 release_sock(sk); 1091 } 1092 ulen += sizeof(struct udphdr); 1093 1094 /* 1095 * Get and verify the address. 1096 */ 1097 if (usin) { 1098 if (msg->msg_namelen < sizeof(*usin)) 1099 return -EINVAL; 1100 if (usin->sin_family != AF_INET) { 1101 if (usin->sin_family != AF_UNSPEC) 1102 return -EAFNOSUPPORT; 1103 } 1104 1105 daddr = usin->sin_addr.s_addr; 1106 dport = usin->sin_port; 1107 if (dport == 0) 1108 return -EINVAL; 1109 } else { 1110 if (sk->sk_state != TCP_ESTABLISHED) 1111 return -EDESTADDRREQ; 1112 daddr = inet->inet_daddr; 1113 dport = inet->inet_dport; 1114 /* Open fast path for connected socket. 1115 Route will not be used, if at least one option is set. 1116 */ 1117 connected = 1; 1118 } 1119 1120 ipcm_init_sk(&ipc, inet); 1121 ipc.gso_size = READ_ONCE(up->gso_size); 1122 1123 if (msg->msg_controllen) { 1124 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1125 if (err > 0) 1126 err = ip_cmsg_send(sk, msg, &ipc, 1127 sk->sk_family == AF_INET6); 1128 if (unlikely(err < 0)) { 1129 kfree(ipc.opt); 1130 return err; 1131 } 1132 if (ipc.opt) 1133 free = 1; 1134 connected = 0; 1135 } 1136 if (!ipc.opt) { 1137 struct ip_options_rcu *inet_opt; 1138 1139 rcu_read_lock(); 1140 inet_opt = rcu_dereference(inet->inet_opt); 1141 if (inet_opt) { 1142 memcpy(&opt_copy, inet_opt, 1143 sizeof(*inet_opt) + inet_opt->opt.optlen); 1144 ipc.opt = &opt_copy.opt; 1145 } 1146 rcu_read_unlock(); 1147 } 1148 1149 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1150 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1151 (struct sockaddr *)usin, &ipc.addr); 1152 if (err) 1153 goto out_free; 1154 if (usin) { 1155 if (usin->sin_port == 0) { 1156 /* BPF program set invalid port. Reject it. */ 1157 err = -EINVAL; 1158 goto out_free; 1159 } 1160 daddr = usin->sin_addr.s_addr; 1161 dport = usin->sin_port; 1162 } 1163 } 1164 1165 saddr = ipc.addr; 1166 ipc.addr = faddr = daddr; 1167 1168 if (ipc.opt && ipc.opt->opt.srr) { 1169 if (!daddr) { 1170 err = -EINVAL; 1171 goto out_free; 1172 } 1173 faddr = ipc.opt->opt.faddr; 1174 connected = 0; 1175 } 1176 tos = get_rttos(&ipc, inet); 1177 if (sock_flag(sk, SOCK_LOCALROUTE) || 1178 (msg->msg_flags & MSG_DONTROUTE) || 1179 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1180 tos |= RTO_ONLINK; 1181 connected = 0; 1182 } 1183 1184 if (ipv4_is_multicast(daddr)) { 1185 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1186 ipc.oif = inet->mc_index; 1187 if (!saddr) 1188 saddr = inet->mc_addr; 1189 connected = 0; 1190 } else if (!ipc.oif) { 1191 ipc.oif = inet->uc_index; 1192 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1193 /* oif is set, packet is to local broadcast and 1194 * uc_index is set. oif is most likely set 1195 * by sk_bound_dev_if. If uc_index != oif check if the 1196 * oif is an L3 master and uc_index is an L3 slave. 1197 * If so, we want to allow the send using the uc_index. 1198 */ 1199 if (ipc.oif != inet->uc_index && 1200 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1201 inet->uc_index)) { 1202 ipc.oif = inet->uc_index; 1203 } 1204 } 1205 1206 if (connected) 1207 rt = (struct rtable *)sk_dst_check(sk, 0); 1208 1209 if (!rt) { 1210 struct net *net = sock_net(sk); 1211 __u8 flow_flags = inet_sk_flowi_flags(sk); 1212 1213 fl4 = &fl4_stack; 1214 1215 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, 1216 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1217 flow_flags, 1218 faddr, saddr, dport, inet->inet_sport, 1219 sk->sk_uid); 1220 1221 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1222 rt = ip_route_output_flow(net, fl4, sk); 1223 if (IS_ERR(rt)) { 1224 err = PTR_ERR(rt); 1225 rt = NULL; 1226 if (err == -ENETUNREACH) 1227 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1228 goto out; 1229 } 1230 1231 err = -EACCES; 1232 if ((rt->rt_flags & RTCF_BROADCAST) && 1233 !sock_flag(sk, SOCK_BROADCAST)) 1234 goto out; 1235 if (connected) 1236 sk_dst_set(sk, dst_clone(&rt->dst)); 1237 } 1238 1239 if (msg->msg_flags&MSG_CONFIRM) 1240 goto do_confirm; 1241 back_from_confirm: 1242 1243 saddr = fl4->saddr; 1244 if (!ipc.addr) 1245 daddr = ipc.addr = fl4->daddr; 1246 1247 /* Lockless fast path for the non-corking case. */ 1248 if (!corkreq) { 1249 struct inet_cork cork; 1250 1251 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1252 sizeof(struct udphdr), &ipc, &rt, 1253 &cork, msg->msg_flags); 1254 err = PTR_ERR(skb); 1255 if (!IS_ERR_OR_NULL(skb)) 1256 err = udp_send_skb(skb, fl4, &cork); 1257 goto out; 1258 } 1259 1260 lock_sock(sk); 1261 if (unlikely(up->pending)) { 1262 /* The socket is already corked while preparing it. */ 1263 /* ... which is an evident application bug. --ANK */ 1264 release_sock(sk); 1265 1266 net_dbg_ratelimited("socket already corked\n"); 1267 err = -EINVAL; 1268 goto out; 1269 } 1270 /* 1271 * Now cork the socket to pend data. 1272 */ 1273 fl4 = &inet->cork.fl.u.ip4; 1274 fl4->daddr = daddr; 1275 fl4->saddr = saddr; 1276 fl4->fl4_dport = dport; 1277 fl4->fl4_sport = inet->inet_sport; 1278 up->pending = AF_INET; 1279 1280 do_append_data: 1281 up->len += ulen; 1282 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1283 sizeof(struct udphdr), &ipc, &rt, 1284 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1285 if (err) 1286 udp_flush_pending_frames(sk); 1287 else if (!corkreq) 1288 err = udp_push_pending_frames(sk); 1289 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1290 up->pending = 0; 1291 release_sock(sk); 1292 1293 out: 1294 ip_rt_put(rt); 1295 out_free: 1296 if (free) 1297 kfree(ipc.opt); 1298 if (!err) 1299 return len; 1300 /* 1301 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1302 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1303 * we don't have a good statistic (IpOutDiscards but it can be too many 1304 * things). We could add another new stat but at least for now that 1305 * seems like overkill. 1306 */ 1307 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1308 UDP_INC_STATS(sock_net(sk), 1309 UDP_MIB_SNDBUFERRORS, is_udplite); 1310 } 1311 return err; 1312 1313 do_confirm: 1314 if (msg->msg_flags & MSG_PROBE) 1315 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1316 if (!(msg->msg_flags&MSG_PROBE) || len) 1317 goto back_from_confirm; 1318 err = 0; 1319 goto out; 1320 } 1321 EXPORT_SYMBOL(udp_sendmsg); 1322 1323 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1324 size_t size, int flags) 1325 { 1326 struct inet_sock *inet = inet_sk(sk); 1327 struct udp_sock *up = udp_sk(sk); 1328 int ret; 1329 1330 if (flags & MSG_SENDPAGE_NOTLAST) 1331 flags |= MSG_MORE; 1332 1333 if (!up->pending) { 1334 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1335 1336 /* Call udp_sendmsg to specify destination address which 1337 * sendpage interface can't pass. 1338 * This will succeed only when the socket is connected. 1339 */ 1340 ret = udp_sendmsg(sk, &msg, 0); 1341 if (ret < 0) 1342 return ret; 1343 } 1344 1345 lock_sock(sk); 1346 1347 if (unlikely(!up->pending)) { 1348 release_sock(sk); 1349 1350 net_dbg_ratelimited("cork failed\n"); 1351 return -EINVAL; 1352 } 1353 1354 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1355 page, offset, size, flags); 1356 if (ret == -EOPNOTSUPP) { 1357 release_sock(sk); 1358 return sock_no_sendpage(sk->sk_socket, page, offset, 1359 size, flags); 1360 } 1361 if (ret < 0) { 1362 udp_flush_pending_frames(sk); 1363 goto out; 1364 } 1365 1366 up->len += size; 1367 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE))) 1368 ret = udp_push_pending_frames(sk); 1369 if (!ret) 1370 ret = size; 1371 out: 1372 release_sock(sk); 1373 return ret; 1374 } 1375 1376 #define UDP_SKB_IS_STATELESS 0x80000000 1377 1378 /* all head states (dst, sk, nf conntrack) except skb extensions are 1379 * cleared by udp_rcv(). 1380 * 1381 * We need to preserve secpath, if present, to eventually process 1382 * IP_CMSG_PASSSEC at recvmsg() time. 1383 * 1384 * Other extensions can be cleared. 1385 */ 1386 static bool udp_try_make_stateless(struct sk_buff *skb) 1387 { 1388 if (!skb_has_extensions(skb)) 1389 return true; 1390 1391 if (!secpath_exists(skb)) { 1392 skb_ext_reset(skb); 1393 return true; 1394 } 1395 1396 return false; 1397 } 1398 1399 static void udp_set_dev_scratch(struct sk_buff *skb) 1400 { 1401 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1402 1403 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1404 scratch->_tsize_state = skb->truesize; 1405 #if BITS_PER_LONG == 64 1406 scratch->len = skb->len; 1407 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1408 scratch->is_linear = !skb_is_nonlinear(skb); 1409 #endif 1410 if (udp_try_make_stateless(skb)) 1411 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1412 } 1413 1414 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1415 { 1416 /* We come here after udp_lib_checksum_complete() returned 0. 1417 * This means that __skb_checksum_complete() might have 1418 * set skb->csum_valid to 1. 1419 * On 64bit platforms, we can set csum_unnecessary 1420 * to true, but only if the skb is not shared. 1421 */ 1422 #if BITS_PER_LONG == 64 1423 if (!skb_shared(skb)) 1424 udp_skb_scratch(skb)->csum_unnecessary = true; 1425 #endif 1426 } 1427 1428 static int udp_skb_truesize(struct sk_buff *skb) 1429 { 1430 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1431 } 1432 1433 static bool udp_skb_has_head_state(struct sk_buff *skb) 1434 { 1435 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1436 } 1437 1438 /* fully reclaim rmem/fwd memory allocated for skb */ 1439 static void udp_rmem_release(struct sock *sk, int size, int partial, 1440 bool rx_queue_lock_held) 1441 { 1442 struct udp_sock *up = udp_sk(sk); 1443 struct sk_buff_head *sk_queue; 1444 int amt; 1445 1446 if (likely(partial)) { 1447 up->forward_deficit += size; 1448 size = up->forward_deficit; 1449 if (size < (sk->sk_rcvbuf >> 2) && 1450 !skb_queue_empty(&up->reader_queue)) 1451 return; 1452 } else { 1453 size += up->forward_deficit; 1454 } 1455 up->forward_deficit = 0; 1456 1457 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1458 * if the called don't held it already 1459 */ 1460 sk_queue = &sk->sk_receive_queue; 1461 if (!rx_queue_lock_held) 1462 spin_lock(&sk_queue->lock); 1463 1464 1465 sk->sk_forward_alloc += size; 1466 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1467 sk->sk_forward_alloc -= amt; 1468 1469 if (amt) 1470 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 1471 1472 atomic_sub(size, &sk->sk_rmem_alloc); 1473 1474 /* this can save us from acquiring the rx queue lock on next receive */ 1475 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1476 1477 if (!rx_queue_lock_held) 1478 spin_unlock(&sk_queue->lock); 1479 } 1480 1481 /* Note: called with reader_queue.lock held. 1482 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1483 * This avoids a cache line miss while receive_queue lock is held. 1484 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1485 */ 1486 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1487 { 1488 prefetch(&skb->data); 1489 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1490 } 1491 EXPORT_SYMBOL(udp_skb_destructor); 1492 1493 /* as above, but the caller held the rx queue lock, too */ 1494 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1495 { 1496 prefetch(&skb->data); 1497 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1498 } 1499 1500 /* Idea of busylocks is to let producers grab an extra spinlock 1501 * to relieve pressure on the receive_queue spinlock shared by consumer. 1502 * Under flood, this means that only one producer can be in line 1503 * trying to acquire the receive_queue spinlock. 1504 * These busylock can be allocated on a per cpu manner, instead of a 1505 * per socket one (that would consume a cache line per socket) 1506 */ 1507 static int udp_busylocks_log __read_mostly; 1508 static spinlock_t *udp_busylocks __read_mostly; 1509 1510 static spinlock_t *busylock_acquire(void *ptr) 1511 { 1512 spinlock_t *busy; 1513 1514 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1515 spin_lock(busy); 1516 return busy; 1517 } 1518 1519 static void busylock_release(spinlock_t *busy) 1520 { 1521 if (busy) 1522 spin_unlock(busy); 1523 } 1524 1525 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1526 { 1527 struct sk_buff_head *list = &sk->sk_receive_queue; 1528 int rmem, delta, amt, err = -ENOMEM; 1529 spinlock_t *busy = NULL; 1530 int size; 1531 1532 /* try to avoid the costly atomic add/sub pair when the receive 1533 * queue is full; always allow at least a packet 1534 */ 1535 rmem = atomic_read(&sk->sk_rmem_alloc); 1536 if (rmem > sk->sk_rcvbuf) 1537 goto drop; 1538 1539 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1540 * having linear skbs : 1541 * - Reduce memory overhead and thus increase receive queue capacity 1542 * - Less cache line misses at copyout() time 1543 * - Less work at consume_skb() (less alien page frag freeing) 1544 */ 1545 if (rmem > (sk->sk_rcvbuf >> 1)) { 1546 skb_condense(skb); 1547 1548 busy = busylock_acquire(sk); 1549 } 1550 size = skb->truesize; 1551 udp_set_dev_scratch(skb); 1552 1553 /* we drop only if the receive buf is full and the receive 1554 * queue contains some other skb 1555 */ 1556 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1557 if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) 1558 goto uncharge_drop; 1559 1560 spin_lock(&list->lock); 1561 if (size >= sk->sk_forward_alloc) { 1562 amt = sk_mem_pages(size); 1563 delta = amt << PAGE_SHIFT; 1564 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1565 err = -ENOBUFS; 1566 spin_unlock(&list->lock); 1567 goto uncharge_drop; 1568 } 1569 1570 sk->sk_forward_alloc += delta; 1571 } 1572 1573 sk->sk_forward_alloc -= size; 1574 1575 /* no need to setup a destructor, we will explicitly release the 1576 * forward allocated memory on dequeue 1577 */ 1578 sock_skb_set_dropcount(sk, skb); 1579 1580 __skb_queue_tail(list, skb); 1581 spin_unlock(&list->lock); 1582 1583 if (!sock_flag(sk, SOCK_DEAD)) 1584 sk->sk_data_ready(sk); 1585 1586 busylock_release(busy); 1587 return 0; 1588 1589 uncharge_drop: 1590 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1591 1592 drop: 1593 atomic_inc(&sk->sk_drops); 1594 busylock_release(busy); 1595 return err; 1596 } 1597 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1598 1599 void udp_destruct_sock(struct sock *sk) 1600 { 1601 /* reclaim completely the forward allocated memory */ 1602 struct udp_sock *up = udp_sk(sk); 1603 unsigned int total = 0; 1604 struct sk_buff *skb; 1605 1606 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1607 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1608 total += skb->truesize; 1609 kfree_skb(skb); 1610 } 1611 udp_rmem_release(sk, total, 0, true); 1612 1613 inet_sock_destruct(sk); 1614 } 1615 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1616 1617 int udp_init_sock(struct sock *sk) 1618 { 1619 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1620 sk->sk_destruct = udp_destruct_sock; 1621 return 0; 1622 } 1623 EXPORT_SYMBOL_GPL(udp_init_sock); 1624 1625 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1626 { 1627 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1628 bool slow = lock_sock_fast(sk); 1629 1630 sk_peek_offset_bwd(sk, len); 1631 unlock_sock_fast(sk, slow); 1632 } 1633 1634 if (!skb_unref(skb)) 1635 return; 1636 1637 /* In the more common cases we cleared the head states previously, 1638 * see __udp_queue_rcv_skb(). 1639 */ 1640 if (unlikely(udp_skb_has_head_state(skb))) 1641 skb_release_head_state(skb); 1642 __consume_stateless_skb(skb); 1643 } 1644 EXPORT_SYMBOL_GPL(skb_consume_udp); 1645 1646 static struct sk_buff *__first_packet_length(struct sock *sk, 1647 struct sk_buff_head *rcvq, 1648 int *total) 1649 { 1650 struct sk_buff *skb; 1651 1652 while ((skb = skb_peek(rcvq)) != NULL) { 1653 if (udp_lib_checksum_complete(skb)) { 1654 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1655 IS_UDPLITE(sk)); 1656 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1657 IS_UDPLITE(sk)); 1658 atomic_inc(&sk->sk_drops); 1659 __skb_unlink(skb, rcvq); 1660 *total += skb->truesize; 1661 kfree_skb(skb); 1662 } else { 1663 udp_skb_csum_unnecessary_set(skb); 1664 break; 1665 } 1666 } 1667 return skb; 1668 } 1669 1670 /** 1671 * first_packet_length - return length of first packet in receive queue 1672 * @sk: socket 1673 * 1674 * Drops all bad checksum frames, until a valid one is found. 1675 * Returns the length of found skb, or -1 if none is found. 1676 */ 1677 static int first_packet_length(struct sock *sk) 1678 { 1679 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1680 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1681 struct sk_buff *skb; 1682 int total = 0; 1683 int res; 1684 1685 spin_lock_bh(&rcvq->lock); 1686 skb = __first_packet_length(sk, rcvq, &total); 1687 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1688 spin_lock(&sk_queue->lock); 1689 skb_queue_splice_tail_init(sk_queue, rcvq); 1690 spin_unlock(&sk_queue->lock); 1691 1692 skb = __first_packet_length(sk, rcvq, &total); 1693 } 1694 res = skb ? skb->len : -1; 1695 if (total) 1696 udp_rmem_release(sk, total, 1, false); 1697 spin_unlock_bh(&rcvq->lock); 1698 return res; 1699 } 1700 1701 /* 1702 * IOCTL requests applicable to the UDP protocol 1703 */ 1704 1705 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1706 { 1707 switch (cmd) { 1708 case SIOCOUTQ: 1709 { 1710 int amount = sk_wmem_alloc_get(sk); 1711 1712 return put_user(amount, (int __user *)arg); 1713 } 1714 1715 case SIOCINQ: 1716 { 1717 int amount = max_t(int, 0, first_packet_length(sk)); 1718 1719 return put_user(amount, (int __user *)arg); 1720 } 1721 1722 default: 1723 return -ENOIOCTLCMD; 1724 } 1725 1726 return 0; 1727 } 1728 EXPORT_SYMBOL(udp_ioctl); 1729 1730 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1731 int *off, int *err) 1732 { 1733 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1734 struct sk_buff_head *queue; 1735 struct sk_buff *last; 1736 long timeo; 1737 int error; 1738 1739 queue = &udp_sk(sk)->reader_queue; 1740 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1741 do { 1742 struct sk_buff *skb; 1743 1744 error = sock_error(sk); 1745 if (error) 1746 break; 1747 1748 error = -EAGAIN; 1749 do { 1750 spin_lock_bh(&queue->lock); 1751 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1752 err, &last); 1753 if (skb) { 1754 if (!(flags & MSG_PEEK)) 1755 udp_skb_destructor(sk, skb); 1756 spin_unlock_bh(&queue->lock); 1757 return skb; 1758 } 1759 1760 if (skb_queue_empty_lockless(sk_queue)) { 1761 spin_unlock_bh(&queue->lock); 1762 goto busy_check; 1763 } 1764 1765 /* refill the reader queue and walk it again 1766 * keep both queues locked to avoid re-acquiring 1767 * the sk_receive_queue lock if fwd memory scheduling 1768 * is needed. 1769 */ 1770 spin_lock(&sk_queue->lock); 1771 skb_queue_splice_tail_init(sk_queue, queue); 1772 1773 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1774 err, &last); 1775 if (skb && !(flags & MSG_PEEK)) 1776 udp_skb_dtor_locked(sk, skb); 1777 spin_unlock(&sk_queue->lock); 1778 spin_unlock_bh(&queue->lock); 1779 if (skb) 1780 return skb; 1781 1782 busy_check: 1783 if (!sk_can_busy_loop(sk)) 1784 break; 1785 1786 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1787 } while (!skb_queue_empty_lockless(sk_queue)); 1788 1789 /* sk_queue is empty, reader_queue may contain peeked packets */ 1790 } while (timeo && 1791 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1792 &error, &timeo, 1793 (struct sk_buff *)sk_queue)); 1794 1795 *err = error; 1796 return NULL; 1797 } 1798 EXPORT_SYMBOL(__skb_recv_udp); 1799 1800 int udp_read_sock(struct sock *sk, read_descriptor_t *desc, 1801 sk_read_actor_t recv_actor) 1802 { 1803 int copied = 0; 1804 1805 while (1) { 1806 struct sk_buff *skb; 1807 int err, used; 1808 1809 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1810 if (!skb) 1811 return err; 1812 1813 if (udp_lib_checksum_complete(skb)) { 1814 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1815 IS_UDPLITE(sk)); 1816 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1817 IS_UDPLITE(sk)); 1818 atomic_inc(&sk->sk_drops); 1819 kfree_skb(skb); 1820 continue; 1821 } 1822 1823 used = recv_actor(desc, skb, 0, skb->len); 1824 if (used <= 0) { 1825 if (!copied) 1826 copied = used; 1827 kfree_skb(skb); 1828 break; 1829 } else if (used <= skb->len) { 1830 copied += used; 1831 } 1832 1833 kfree_skb(skb); 1834 if (!desc->count) 1835 break; 1836 } 1837 1838 return copied; 1839 } 1840 EXPORT_SYMBOL(udp_read_sock); 1841 1842 /* 1843 * This should be easy, if there is something there we 1844 * return it, otherwise we block. 1845 */ 1846 1847 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 1848 int *addr_len) 1849 { 1850 struct inet_sock *inet = inet_sk(sk); 1851 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1852 struct sk_buff *skb; 1853 unsigned int ulen, copied; 1854 int off, err, peeking = flags & MSG_PEEK; 1855 int is_udplite = IS_UDPLITE(sk); 1856 bool checksum_valid = false; 1857 1858 if (flags & MSG_ERRQUEUE) 1859 return ip_recv_error(sk, msg, len, addr_len); 1860 1861 try_again: 1862 off = sk_peek_offset(sk, flags); 1863 skb = __skb_recv_udp(sk, flags, &off, &err); 1864 if (!skb) 1865 return err; 1866 1867 ulen = udp_skb_len(skb); 1868 copied = len; 1869 if (copied > ulen - off) 1870 copied = ulen - off; 1871 else if (copied < ulen) 1872 msg->msg_flags |= MSG_TRUNC; 1873 1874 /* 1875 * If checksum is needed at all, try to do it while copying the 1876 * data. If the data is truncated, or if we only want a partial 1877 * coverage checksum (UDP-Lite), do it before the copy. 1878 */ 1879 1880 if (copied < ulen || peeking || 1881 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1882 checksum_valid = udp_skb_csum_unnecessary(skb) || 1883 !__udp_lib_checksum_complete(skb); 1884 if (!checksum_valid) 1885 goto csum_copy_err; 1886 } 1887 1888 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1889 if (udp_skb_is_linear(skb)) 1890 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1891 else 1892 err = skb_copy_datagram_msg(skb, off, msg, copied); 1893 } else { 1894 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1895 1896 if (err == -EINVAL) 1897 goto csum_copy_err; 1898 } 1899 1900 if (unlikely(err)) { 1901 if (!peeking) { 1902 atomic_inc(&sk->sk_drops); 1903 UDP_INC_STATS(sock_net(sk), 1904 UDP_MIB_INERRORS, is_udplite); 1905 } 1906 kfree_skb(skb); 1907 return err; 1908 } 1909 1910 if (!peeking) 1911 UDP_INC_STATS(sock_net(sk), 1912 UDP_MIB_INDATAGRAMS, is_udplite); 1913 1914 sock_recv_cmsgs(msg, sk, skb); 1915 1916 /* Copy the address. */ 1917 if (sin) { 1918 sin->sin_family = AF_INET; 1919 sin->sin_port = udp_hdr(skb)->source; 1920 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1921 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1922 *addr_len = sizeof(*sin); 1923 1924 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1925 (struct sockaddr *)sin); 1926 } 1927 1928 if (udp_sk(sk)->gro_enabled) 1929 udp_cmsg_recv(msg, sk, skb); 1930 1931 if (inet->cmsg_flags) 1932 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1933 1934 err = copied; 1935 if (flags & MSG_TRUNC) 1936 err = ulen; 1937 1938 skb_consume_udp(sk, skb, peeking ? -err : err); 1939 return err; 1940 1941 csum_copy_err: 1942 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1943 udp_skb_destructor)) { 1944 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1945 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1946 } 1947 kfree_skb(skb); 1948 1949 /* starting over for a new packet, but check if we need to yield */ 1950 cond_resched(); 1951 msg->msg_flags &= ~MSG_TRUNC; 1952 goto try_again; 1953 } 1954 1955 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1956 { 1957 /* This check is replicated from __ip4_datagram_connect() and 1958 * intended to prevent BPF program called below from accessing bytes 1959 * that are out of the bound specified by user in addr_len. 1960 */ 1961 if (addr_len < sizeof(struct sockaddr_in)) 1962 return -EINVAL; 1963 1964 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1965 } 1966 EXPORT_SYMBOL(udp_pre_connect); 1967 1968 int __udp_disconnect(struct sock *sk, int flags) 1969 { 1970 struct inet_sock *inet = inet_sk(sk); 1971 /* 1972 * 1003.1g - break association. 1973 */ 1974 1975 sk->sk_state = TCP_CLOSE; 1976 inet->inet_daddr = 0; 1977 inet->inet_dport = 0; 1978 sock_rps_reset_rxhash(sk); 1979 sk->sk_bound_dev_if = 0; 1980 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1981 inet_reset_saddr(sk); 1982 if (sk->sk_prot->rehash && 1983 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1984 sk->sk_prot->rehash(sk); 1985 } 1986 1987 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1988 sk->sk_prot->unhash(sk); 1989 inet->inet_sport = 0; 1990 } 1991 sk_dst_reset(sk); 1992 return 0; 1993 } 1994 EXPORT_SYMBOL(__udp_disconnect); 1995 1996 int udp_disconnect(struct sock *sk, int flags) 1997 { 1998 lock_sock(sk); 1999 __udp_disconnect(sk, flags); 2000 release_sock(sk); 2001 return 0; 2002 } 2003 EXPORT_SYMBOL(udp_disconnect); 2004 2005 void udp_lib_unhash(struct sock *sk) 2006 { 2007 if (sk_hashed(sk)) { 2008 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2009 struct udp_hslot *hslot, *hslot2; 2010 2011 hslot = udp_hashslot(udptable, sock_net(sk), 2012 udp_sk(sk)->udp_port_hash); 2013 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2014 2015 spin_lock_bh(&hslot->lock); 2016 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2017 reuseport_detach_sock(sk); 2018 if (sk_del_node_init_rcu(sk)) { 2019 hslot->count--; 2020 inet_sk(sk)->inet_num = 0; 2021 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 2022 2023 spin_lock(&hslot2->lock); 2024 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2025 hslot2->count--; 2026 spin_unlock(&hslot2->lock); 2027 } 2028 spin_unlock_bh(&hslot->lock); 2029 } 2030 } 2031 EXPORT_SYMBOL(udp_lib_unhash); 2032 2033 /* 2034 * inet_rcv_saddr was changed, we must rehash secondary hash 2035 */ 2036 void udp_lib_rehash(struct sock *sk, u16 newhash) 2037 { 2038 if (sk_hashed(sk)) { 2039 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2040 struct udp_hslot *hslot, *hslot2, *nhslot2; 2041 2042 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2043 nhslot2 = udp_hashslot2(udptable, newhash); 2044 udp_sk(sk)->udp_portaddr_hash = newhash; 2045 2046 if (hslot2 != nhslot2 || 2047 rcu_access_pointer(sk->sk_reuseport_cb)) { 2048 hslot = udp_hashslot(udptable, sock_net(sk), 2049 udp_sk(sk)->udp_port_hash); 2050 /* we must lock primary chain too */ 2051 spin_lock_bh(&hslot->lock); 2052 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2053 reuseport_detach_sock(sk); 2054 2055 if (hslot2 != nhslot2) { 2056 spin_lock(&hslot2->lock); 2057 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2058 hslot2->count--; 2059 spin_unlock(&hslot2->lock); 2060 2061 spin_lock(&nhslot2->lock); 2062 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2063 &nhslot2->head); 2064 nhslot2->count++; 2065 spin_unlock(&nhslot2->lock); 2066 } 2067 2068 spin_unlock_bh(&hslot->lock); 2069 } 2070 } 2071 } 2072 EXPORT_SYMBOL(udp_lib_rehash); 2073 2074 void udp_v4_rehash(struct sock *sk) 2075 { 2076 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2077 inet_sk(sk)->inet_rcv_saddr, 2078 inet_sk(sk)->inet_num); 2079 udp_lib_rehash(sk, new_hash); 2080 } 2081 2082 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2083 { 2084 int rc; 2085 2086 if (inet_sk(sk)->inet_daddr) { 2087 sock_rps_save_rxhash(sk, skb); 2088 sk_mark_napi_id(sk, skb); 2089 sk_incoming_cpu_update(sk); 2090 } else { 2091 sk_mark_napi_id_once(sk, skb); 2092 } 2093 2094 rc = __udp_enqueue_schedule_skb(sk, skb); 2095 if (rc < 0) { 2096 int is_udplite = IS_UDPLITE(sk); 2097 int drop_reason; 2098 2099 /* Note that an ENOMEM error is charged twice */ 2100 if (rc == -ENOMEM) { 2101 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2102 is_udplite); 2103 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 2104 } else { 2105 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2106 is_udplite); 2107 drop_reason = SKB_DROP_REASON_PROTO_MEM; 2108 } 2109 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2110 kfree_skb_reason(skb, drop_reason); 2111 trace_udp_fail_queue_rcv_skb(rc, sk); 2112 return -1; 2113 } 2114 2115 return 0; 2116 } 2117 2118 /* returns: 2119 * -1: error 2120 * 0: success 2121 * >0: "udp encap" protocol resubmission 2122 * 2123 * Note that in the success and error cases, the skb is assumed to 2124 * have either been requeued or freed. 2125 */ 2126 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2127 { 2128 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2129 struct udp_sock *up = udp_sk(sk); 2130 int is_udplite = IS_UDPLITE(sk); 2131 2132 /* 2133 * Charge it to the socket, dropping if the queue is full. 2134 */ 2135 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 2136 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2137 goto drop; 2138 } 2139 nf_reset_ct(skb); 2140 2141 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 2142 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2143 2144 /* 2145 * This is an encapsulation socket so pass the skb to 2146 * the socket's udp_encap_rcv() hook. Otherwise, just 2147 * fall through and pass this up the UDP socket. 2148 * up->encap_rcv() returns the following value: 2149 * =0 if skb was successfully passed to the encap 2150 * handler or was discarded by it. 2151 * >0 if skb should be passed on to UDP. 2152 * <0 if skb should be resubmitted as proto -N 2153 */ 2154 2155 /* if we're overly short, let UDP handle it */ 2156 encap_rcv = READ_ONCE(up->encap_rcv); 2157 if (encap_rcv) { 2158 int ret; 2159 2160 /* Verify checksum before giving to encap */ 2161 if (udp_lib_checksum_complete(skb)) 2162 goto csum_error; 2163 2164 ret = encap_rcv(sk, skb); 2165 if (ret <= 0) { 2166 __UDP_INC_STATS(sock_net(sk), 2167 UDP_MIB_INDATAGRAMS, 2168 is_udplite); 2169 return -ret; 2170 } 2171 } 2172 2173 /* FALLTHROUGH -- it's a UDP Packet */ 2174 } 2175 2176 /* 2177 * UDP-Lite specific tests, ignored on UDP sockets 2178 */ 2179 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2180 2181 /* 2182 * MIB statistics other than incrementing the error count are 2183 * disabled for the following two types of errors: these depend 2184 * on the application settings, not on the functioning of the 2185 * protocol stack as such. 2186 * 2187 * RFC 3828 here recommends (sec 3.3): "There should also be a 2188 * way ... to ... at least let the receiving application block 2189 * delivery of packets with coverage values less than a value 2190 * provided by the application." 2191 */ 2192 if (up->pcrlen == 0) { /* full coverage was set */ 2193 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2194 UDP_SKB_CB(skb)->cscov, skb->len); 2195 goto drop; 2196 } 2197 /* The next case involves violating the min. coverage requested 2198 * by the receiver. This is subtle: if receiver wants x and x is 2199 * greater than the buffersize/MTU then receiver will complain 2200 * that it wants x while sender emits packets of smaller size y. 2201 * Therefore the above ...()->partial_cov statement is essential. 2202 */ 2203 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2204 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2205 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2206 goto drop; 2207 } 2208 } 2209 2210 prefetch(&sk->sk_rmem_alloc); 2211 if (rcu_access_pointer(sk->sk_filter) && 2212 udp_lib_checksum_complete(skb)) 2213 goto csum_error; 2214 2215 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 2216 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 2217 goto drop; 2218 } 2219 2220 udp_csum_pull_header(skb); 2221 2222 ipv4_pktinfo_prepare(sk, skb); 2223 return __udp_queue_rcv_skb(sk, skb); 2224 2225 csum_error: 2226 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2227 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2228 drop: 2229 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2230 atomic_inc(&sk->sk_drops); 2231 kfree_skb_reason(skb, drop_reason); 2232 return -1; 2233 } 2234 2235 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2236 { 2237 struct sk_buff *next, *segs; 2238 int ret; 2239 2240 if (likely(!udp_unexpected_gso(sk, skb))) 2241 return udp_queue_rcv_one_skb(sk, skb); 2242 2243 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2244 __skb_push(skb, -skb_mac_offset(skb)); 2245 segs = udp_rcv_segment(sk, skb, true); 2246 skb_list_walk_safe(segs, skb, next) { 2247 __skb_pull(skb, skb_transport_offset(skb)); 2248 2249 udp_post_segment_fix_csum(skb); 2250 ret = udp_queue_rcv_one_skb(sk, skb); 2251 if (ret > 0) 2252 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2253 } 2254 return 0; 2255 } 2256 2257 /* For TCP sockets, sk_rx_dst is protected by socket lock 2258 * For UDP, we use xchg() to guard against concurrent changes. 2259 */ 2260 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2261 { 2262 struct dst_entry *old; 2263 2264 if (dst_hold_safe(dst)) { 2265 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); 2266 dst_release(old); 2267 return old != dst; 2268 } 2269 return false; 2270 } 2271 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2272 2273 /* 2274 * Multicasts and broadcasts go to each listener. 2275 * 2276 * Note: called only from the BH handler context. 2277 */ 2278 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2279 struct udphdr *uh, 2280 __be32 saddr, __be32 daddr, 2281 struct udp_table *udptable, 2282 int proto) 2283 { 2284 struct sock *sk, *first = NULL; 2285 unsigned short hnum = ntohs(uh->dest); 2286 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2287 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2288 unsigned int offset = offsetof(typeof(*sk), sk_node); 2289 int dif = skb->dev->ifindex; 2290 int sdif = inet_sdif(skb); 2291 struct hlist_node *node; 2292 struct sk_buff *nskb; 2293 2294 if (use_hash2) { 2295 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2296 udptable->mask; 2297 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2298 start_lookup: 2299 hslot = &udptable->hash2[hash2]; 2300 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2301 } 2302 2303 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2304 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2305 uh->source, saddr, dif, sdif, hnum)) 2306 continue; 2307 2308 if (!first) { 2309 first = sk; 2310 continue; 2311 } 2312 nskb = skb_clone(skb, GFP_ATOMIC); 2313 2314 if (unlikely(!nskb)) { 2315 atomic_inc(&sk->sk_drops); 2316 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2317 IS_UDPLITE(sk)); 2318 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2319 IS_UDPLITE(sk)); 2320 continue; 2321 } 2322 if (udp_queue_rcv_skb(sk, nskb) > 0) 2323 consume_skb(nskb); 2324 } 2325 2326 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2327 if (use_hash2 && hash2 != hash2_any) { 2328 hash2 = hash2_any; 2329 goto start_lookup; 2330 } 2331 2332 if (first) { 2333 if (udp_queue_rcv_skb(first, skb) > 0) 2334 consume_skb(skb); 2335 } else { 2336 kfree_skb(skb); 2337 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2338 proto == IPPROTO_UDPLITE); 2339 } 2340 return 0; 2341 } 2342 2343 /* Initialize UDP checksum. If exited with zero value (success), 2344 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2345 * Otherwise, csum completion requires checksumming packet body, 2346 * including udp header and folding it to skb->csum. 2347 */ 2348 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2349 int proto) 2350 { 2351 int err; 2352 2353 UDP_SKB_CB(skb)->partial_cov = 0; 2354 UDP_SKB_CB(skb)->cscov = skb->len; 2355 2356 if (proto == IPPROTO_UDPLITE) { 2357 err = udplite_checksum_init(skb, uh); 2358 if (err) 2359 return err; 2360 2361 if (UDP_SKB_CB(skb)->partial_cov) { 2362 skb->csum = inet_compute_pseudo(skb, proto); 2363 return 0; 2364 } 2365 } 2366 2367 /* Note, we are only interested in != 0 or == 0, thus the 2368 * force to int. 2369 */ 2370 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2371 inet_compute_pseudo); 2372 if (err) 2373 return err; 2374 2375 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2376 /* If SW calculated the value, we know it's bad */ 2377 if (skb->csum_complete_sw) 2378 return 1; 2379 2380 /* HW says the value is bad. Let's validate that. 2381 * skb->csum is no longer the full packet checksum, 2382 * so don't treat it as such. 2383 */ 2384 skb_checksum_complete_unset(skb); 2385 } 2386 2387 return 0; 2388 } 2389 2390 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2391 * return code conversion for ip layer consumption 2392 */ 2393 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2394 struct udphdr *uh) 2395 { 2396 int ret; 2397 2398 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2399 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2400 2401 ret = udp_queue_rcv_skb(sk, skb); 2402 2403 /* a return value > 0 means to resubmit the input, but 2404 * it wants the return to be -protocol, or 0 2405 */ 2406 if (ret > 0) 2407 return -ret; 2408 return 0; 2409 } 2410 2411 /* 2412 * All we need to do is get the socket, and then do a checksum. 2413 */ 2414 2415 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2416 int proto) 2417 { 2418 struct sock *sk; 2419 struct udphdr *uh; 2420 unsigned short ulen; 2421 struct rtable *rt = skb_rtable(skb); 2422 __be32 saddr, daddr; 2423 struct net *net = dev_net(skb->dev); 2424 bool refcounted; 2425 int drop_reason; 2426 2427 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2428 2429 /* 2430 * Validate the packet. 2431 */ 2432 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2433 goto drop; /* No space for header. */ 2434 2435 uh = udp_hdr(skb); 2436 ulen = ntohs(uh->len); 2437 saddr = ip_hdr(skb)->saddr; 2438 daddr = ip_hdr(skb)->daddr; 2439 2440 if (ulen > skb->len) 2441 goto short_packet; 2442 2443 if (proto == IPPROTO_UDP) { 2444 /* UDP validates ulen. */ 2445 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2446 goto short_packet; 2447 uh = udp_hdr(skb); 2448 } 2449 2450 if (udp4_csum_init(skb, uh, proto)) 2451 goto csum_error; 2452 2453 sk = skb_steal_sock(skb, &refcounted); 2454 if (sk) { 2455 struct dst_entry *dst = skb_dst(skb); 2456 int ret; 2457 2458 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2459 udp_sk_rx_dst_set(sk, dst); 2460 2461 ret = udp_unicast_rcv_skb(sk, skb, uh); 2462 if (refcounted) 2463 sock_put(sk); 2464 return ret; 2465 } 2466 2467 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2468 return __udp4_lib_mcast_deliver(net, skb, uh, 2469 saddr, daddr, udptable, proto); 2470 2471 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2472 if (sk) 2473 return udp_unicast_rcv_skb(sk, skb, uh); 2474 2475 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2476 goto drop; 2477 nf_reset_ct(skb); 2478 2479 /* No socket. Drop packet silently, if checksum is wrong */ 2480 if (udp_lib_checksum_complete(skb)) 2481 goto csum_error; 2482 2483 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2484 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2485 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2486 2487 /* 2488 * Hmm. We got an UDP packet to a port to which we 2489 * don't wanna listen. Ignore it. 2490 */ 2491 kfree_skb_reason(skb, drop_reason); 2492 return 0; 2493 2494 short_packet: 2495 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2496 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2497 proto == IPPROTO_UDPLITE ? "Lite" : "", 2498 &saddr, ntohs(uh->source), 2499 ulen, skb->len, 2500 &daddr, ntohs(uh->dest)); 2501 goto drop; 2502 2503 csum_error: 2504 /* 2505 * RFC1122: OK. Discards the bad packet silently (as far as 2506 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2507 */ 2508 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2509 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2510 proto == IPPROTO_UDPLITE ? "Lite" : "", 2511 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2512 ulen); 2513 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2514 drop: 2515 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2516 kfree_skb_reason(skb, drop_reason); 2517 return 0; 2518 } 2519 2520 /* We can only early demux multicast if there is a single matching socket. 2521 * If more than one socket found returns NULL 2522 */ 2523 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2524 __be16 loc_port, __be32 loc_addr, 2525 __be16 rmt_port, __be32 rmt_addr, 2526 int dif, int sdif) 2527 { 2528 struct sock *sk, *result; 2529 unsigned short hnum = ntohs(loc_port); 2530 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2531 struct udp_hslot *hslot = &udp_table.hash[slot]; 2532 2533 /* Do not bother scanning a too big list */ 2534 if (hslot->count > 10) 2535 return NULL; 2536 2537 result = NULL; 2538 sk_for_each_rcu(sk, &hslot->head) { 2539 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2540 rmt_port, rmt_addr, dif, sdif, hnum)) { 2541 if (result) 2542 return NULL; 2543 result = sk; 2544 } 2545 } 2546 2547 return result; 2548 } 2549 2550 /* For unicast we should only early demux connected sockets or we can 2551 * break forwarding setups. The chains here can be long so only check 2552 * if the first socket is an exact match and if not move on. 2553 */ 2554 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2555 __be16 loc_port, __be32 loc_addr, 2556 __be16 rmt_port, __be32 rmt_addr, 2557 int dif, int sdif) 2558 { 2559 unsigned short hnum = ntohs(loc_port); 2560 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2561 unsigned int slot2 = hash2 & udp_table.mask; 2562 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2563 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2564 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2565 struct sock *sk; 2566 2567 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2568 if (inet_match(net, sk, acookie, ports, dif, sdif)) 2569 return sk; 2570 /* Only check first socket in chain */ 2571 break; 2572 } 2573 return NULL; 2574 } 2575 2576 int udp_v4_early_demux(struct sk_buff *skb) 2577 { 2578 struct net *net = dev_net(skb->dev); 2579 struct in_device *in_dev = NULL; 2580 const struct iphdr *iph; 2581 const struct udphdr *uh; 2582 struct sock *sk = NULL; 2583 struct dst_entry *dst; 2584 int dif = skb->dev->ifindex; 2585 int sdif = inet_sdif(skb); 2586 int ours; 2587 2588 /* validate the packet */ 2589 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2590 return 0; 2591 2592 iph = ip_hdr(skb); 2593 uh = udp_hdr(skb); 2594 2595 if (skb->pkt_type == PACKET_MULTICAST) { 2596 in_dev = __in_dev_get_rcu(skb->dev); 2597 2598 if (!in_dev) 2599 return 0; 2600 2601 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2602 iph->protocol); 2603 if (!ours) 2604 return 0; 2605 2606 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2607 uh->source, iph->saddr, 2608 dif, sdif); 2609 } else if (skb->pkt_type == PACKET_HOST) { 2610 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2611 uh->source, iph->saddr, dif, sdif); 2612 } 2613 2614 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2615 return 0; 2616 2617 skb->sk = sk; 2618 skb->destructor = sock_efree; 2619 dst = rcu_dereference(sk->sk_rx_dst); 2620 2621 if (dst) 2622 dst = dst_check(dst, 0); 2623 if (dst) { 2624 u32 itag = 0; 2625 2626 /* set noref for now. 2627 * any place which wants to hold dst has to call 2628 * dst_hold_safe() 2629 */ 2630 skb_dst_set_noref(skb, dst); 2631 2632 /* for unconnected multicast sockets we need to validate 2633 * the source on each packet 2634 */ 2635 if (!inet_sk(sk)->inet_daddr && in_dev) 2636 return ip_mc_validate_source(skb, iph->daddr, 2637 iph->saddr, 2638 iph->tos & IPTOS_RT_MASK, 2639 skb->dev, in_dev, &itag); 2640 } 2641 return 0; 2642 } 2643 2644 int udp_rcv(struct sk_buff *skb) 2645 { 2646 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2647 } 2648 2649 void udp_destroy_sock(struct sock *sk) 2650 { 2651 struct udp_sock *up = udp_sk(sk); 2652 bool slow = lock_sock_fast(sk); 2653 2654 /* protects from races with udp_abort() */ 2655 sock_set_flag(sk, SOCK_DEAD); 2656 udp_flush_pending_frames(sk); 2657 unlock_sock_fast(sk, slow); 2658 if (static_branch_unlikely(&udp_encap_needed_key)) { 2659 if (up->encap_type) { 2660 void (*encap_destroy)(struct sock *sk); 2661 encap_destroy = READ_ONCE(up->encap_destroy); 2662 if (encap_destroy) 2663 encap_destroy(sk); 2664 } 2665 if (up->encap_enabled) 2666 static_branch_dec(&udp_encap_needed_key); 2667 } 2668 } 2669 2670 /* 2671 * Socket option code for UDP 2672 */ 2673 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2674 sockptr_t optval, unsigned int optlen, 2675 int (*push_pending_frames)(struct sock *)) 2676 { 2677 struct udp_sock *up = udp_sk(sk); 2678 int val, valbool; 2679 int err = 0; 2680 int is_udplite = IS_UDPLITE(sk); 2681 2682 if (optlen < sizeof(int)) 2683 return -EINVAL; 2684 2685 if (copy_from_sockptr(&val, optval, sizeof(val))) 2686 return -EFAULT; 2687 2688 valbool = val ? 1 : 0; 2689 2690 switch (optname) { 2691 case UDP_CORK: 2692 if (val != 0) { 2693 WRITE_ONCE(up->corkflag, 1); 2694 } else { 2695 WRITE_ONCE(up->corkflag, 0); 2696 lock_sock(sk); 2697 push_pending_frames(sk); 2698 release_sock(sk); 2699 } 2700 break; 2701 2702 case UDP_ENCAP: 2703 switch (val) { 2704 case 0: 2705 #ifdef CONFIG_XFRM 2706 case UDP_ENCAP_ESPINUDP: 2707 case UDP_ENCAP_ESPINUDP_NON_IKE: 2708 #if IS_ENABLED(CONFIG_IPV6) 2709 if (sk->sk_family == AF_INET6) 2710 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; 2711 else 2712 #endif 2713 up->encap_rcv = xfrm4_udp_encap_rcv; 2714 #endif 2715 fallthrough; 2716 case UDP_ENCAP_L2TPINUDP: 2717 up->encap_type = val; 2718 lock_sock(sk); 2719 udp_tunnel_encap_enable(sk->sk_socket); 2720 release_sock(sk); 2721 break; 2722 default: 2723 err = -ENOPROTOOPT; 2724 break; 2725 } 2726 break; 2727 2728 case UDP_NO_CHECK6_TX: 2729 up->no_check6_tx = valbool; 2730 break; 2731 2732 case UDP_NO_CHECK6_RX: 2733 up->no_check6_rx = valbool; 2734 break; 2735 2736 case UDP_SEGMENT: 2737 if (val < 0 || val > USHRT_MAX) 2738 return -EINVAL; 2739 WRITE_ONCE(up->gso_size, val); 2740 break; 2741 2742 case UDP_GRO: 2743 lock_sock(sk); 2744 2745 /* when enabling GRO, accept the related GSO packet type */ 2746 if (valbool) 2747 udp_tunnel_encap_enable(sk->sk_socket); 2748 up->gro_enabled = valbool; 2749 up->accept_udp_l4 = valbool; 2750 release_sock(sk); 2751 break; 2752 2753 /* 2754 * UDP-Lite's partial checksum coverage (RFC 3828). 2755 */ 2756 /* The sender sets actual checksum coverage length via this option. 2757 * The case coverage > packet length is handled by send module. */ 2758 case UDPLITE_SEND_CSCOV: 2759 if (!is_udplite) /* Disable the option on UDP sockets */ 2760 return -ENOPROTOOPT; 2761 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2762 val = 8; 2763 else if (val > USHRT_MAX) 2764 val = USHRT_MAX; 2765 up->pcslen = val; 2766 up->pcflag |= UDPLITE_SEND_CC; 2767 break; 2768 2769 /* The receiver specifies a minimum checksum coverage value. To make 2770 * sense, this should be set to at least 8 (as done below). If zero is 2771 * used, this again means full checksum coverage. */ 2772 case UDPLITE_RECV_CSCOV: 2773 if (!is_udplite) /* Disable the option on UDP sockets */ 2774 return -ENOPROTOOPT; 2775 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2776 val = 8; 2777 else if (val > USHRT_MAX) 2778 val = USHRT_MAX; 2779 up->pcrlen = val; 2780 up->pcflag |= UDPLITE_RECV_CC; 2781 break; 2782 2783 default: 2784 err = -ENOPROTOOPT; 2785 break; 2786 } 2787 2788 return err; 2789 } 2790 EXPORT_SYMBOL(udp_lib_setsockopt); 2791 2792 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2793 unsigned int optlen) 2794 { 2795 if (level == SOL_UDP || level == SOL_UDPLITE) 2796 return udp_lib_setsockopt(sk, level, optname, 2797 optval, optlen, 2798 udp_push_pending_frames); 2799 return ip_setsockopt(sk, level, optname, optval, optlen); 2800 } 2801 2802 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2803 char __user *optval, int __user *optlen) 2804 { 2805 struct udp_sock *up = udp_sk(sk); 2806 int val, len; 2807 2808 if (get_user(len, optlen)) 2809 return -EFAULT; 2810 2811 len = min_t(unsigned int, len, sizeof(int)); 2812 2813 if (len < 0) 2814 return -EINVAL; 2815 2816 switch (optname) { 2817 case UDP_CORK: 2818 val = READ_ONCE(up->corkflag); 2819 break; 2820 2821 case UDP_ENCAP: 2822 val = up->encap_type; 2823 break; 2824 2825 case UDP_NO_CHECK6_TX: 2826 val = up->no_check6_tx; 2827 break; 2828 2829 case UDP_NO_CHECK6_RX: 2830 val = up->no_check6_rx; 2831 break; 2832 2833 case UDP_SEGMENT: 2834 val = READ_ONCE(up->gso_size); 2835 break; 2836 2837 case UDP_GRO: 2838 val = up->gro_enabled; 2839 break; 2840 2841 /* The following two cannot be changed on UDP sockets, the return is 2842 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2843 case UDPLITE_SEND_CSCOV: 2844 val = up->pcslen; 2845 break; 2846 2847 case UDPLITE_RECV_CSCOV: 2848 val = up->pcrlen; 2849 break; 2850 2851 default: 2852 return -ENOPROTOOPT; 2853 } 2854 2855 if (put_user(len, optlen)) 2856 return -EFAULT; 2857 if (copy_to_user(optval, &val, len)) 2858 return -EFAULT; 2859 return 0; 2860 } 2861 EXPORT_SYMBOL(udp_lib_getsockopt); 2862 2863 int udp_getsockopt(struct sock *sk, int level, int optname, 2864 char __user *optval, int __user *optlen) 2865 { 2866 if (level == SOL_UDP || level == SOL_UDPLITE) 2867 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2868 return ip_getsockopt(sk, level, optname, optval, optlen); 2869 } 2870 2871 /** 2872 * udp_poll - wait for a UDP event. 2873 * @file: - file struct 2874 * @sock: - socket 2875 * @wait: - poll table 2876 * 2877 * This is same as datagram poll, except for the special case of 2878 * blocking sockets. If application is using a blocking fd 2879 * and a packet with checksum error is in the queue; 2880 * then it could get return from select indicating data available 2881 * but then block when reading it. Add special case code 2882 * to work around these arguably broken applications. 2883 */ 2884 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2885 { 2886 __poll_t mask = datagram_poll(file, sock, wait); 2887 struct sock *sk = sock->sk; 2888 2889 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2890 mask |= EPOLLIN | EPOLLRDNORM; 2891 2892 /* Check for false positives due to checksum errors */ 2893 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2894 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2895 mask &= ~(EPOLLIN | EPOLLRDNORM); 2896 2897 /* psock ingress_msg queue should not contain any bad checksum frames */ 2898 if (sk_is_readable(sk)) 2899 mask |= EPOLLIN | EPOLLRDNORM; 2900 return mask; 2901 2902 } 2903 EXPORT_SYMBOL(udp_poll); 2904 2905 int udp_abort(struct sock *sk, int err) 2906 { 2907 lock_sock(sk); 2908 2909 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2910 * with close() 2911 */ 2912 if (sock_flag(sk, SOCK_DEAD)) 2913 goto out; 2914 2915 sk->sk_err = err; 2916 sk_error_report(sk); 2917 __udp_disconnect(sk, 0); 2918 2919 out: 2920 release_sock(sk); 2921 2922 return 0; 2923 } 2924 EXPORT_SYMBOL_GPL(udp_abort); 2925 2926 struct proto udp_prot = { 2927 .name = "UDP", 2928 .owner = THIS_MODULE, 2929 .close = udp_lib_close, 2930 .pre_connect = udp_pre_connect, 2931 .connect = ip4_datagram_connect, 2932 .disconnect = udp_disconnect, 2933 .ioctl = udp_ioctl, 2934 .init = udp_init_sock, 2935 .destroy = udp_destroy_sock, 2936 .setsockopt = udp_setsockopt, 2937 .getsockopt = udp_getsockopt, 2938 .sendmsg = udp_sendmsg, 2939 .recvmsg = udp_recvmsg, 2940 .sendpage = udp_sendpage, 2941 .release_cb = ip4_datagram_release_cb, 2942 .hash = udp_lib_hash, 2943 .unhash = udp_lib_unhash, 2944 .rehash = udp_v4_rehash, 2945 .get_port = udp_v4_get_port, 2946 .put_port = udp_lib_unhash, 2947 #ifdef CONFIG_BPF_SYSCALL 2948 .psock_update_sk_prot = udp_bpf_update_proto, 2949 #endif 2950 .memory_allocated = &udp_memory_allocated, 2951 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 2952 2953 .sysctl_mem = sysctl_udp_mem, 2954 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2955 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2956 .obj_size = sizeof(struct udp_sock), 2957 .h.udp_table = &udp_table, 2958 .diag_destroy = udp_abort, 2959 }; 2960 EXPORT_SYMBOL(udp_prot); 2961 2962 /* ------------------------------------------------------------------------ */ 2963 #ifdef CONFIG_PROC_FS 2964 2965 static struct sock *udp_get_first(struct seq_file *seq, int start) 2966 { 2967 struct sock *sk; 2968 struct udp_seq_afinfo *afinfo; 2969 struct udp_iter_state *state = seq->private; 2970 struct net *net = seq_file_net(seq); 2971 2972 if (state->bpf_seq_afinfo) 2973 afinfo = state->bpf_seq_afinfo; 2974 else 2975 afinfo = pde_data(file_inode(seq->file)); 2976 2977 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; 2978 ++state->bucket) { 2979 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; 2980 2981 if (hlist_empty(&hslot->head)) 2982 continue; 2983 2984 spin_lock_bh(&hslot->lock); 2985 sk_for_each(sk, &hslot->head) { 2986 if (!net_eq(sock_net(sk), net)) 2987 continue; 2988 if (afinfo->family == AF_UNSPEC || 2989 sk->sk_family == afinfo->family) 2990 goto found; 2991 } 2992 spin_unlock_bh(&hslot->lock); 2993 } 2994 sk = NULL; 2995 found: 2996 return sk; 2997 } 2998 2999 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 3000 { 3001 struct udp_seq_afinfo *afinfo; 3002 struct udp_iter_state *state = seq->private; 3003 struct net *net = seq_file_net(seq); 3004 3005 if (state->bpf_seq_afinfo) 3006 afinfo = state->bpf_seq_afinfo; 3007 else 3008 afinfo = pde_data(file_inode(seq->file)); 3009 3010 do { 3011 sk = sk_next(sk); 3012 } while (sk && (!net_eq(sock_net(sk), net) || 3013 (afinfo->family != AF_UNSPEC && 3014 sk->sk_family != afinfo->family))); 3015 3016 if (!sk) { 3017 if (state->bucket <= afinfo->udp_table->mask) 3018 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 3019 return udp_get_first(seq, state->bucket + 1); 3020 } 3021 return sk; 3022 } 3023 3024 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3025 { 3026 struct sock *sk = udp_get_first(seq, 0); 3027 3028 if (sk) 3029 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3030 --pos; 3031 return pos ? NULL : sk; 3032 } 3033 3034 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3035 { 3036 struct udp_iter_state *state = seq->private; 3037 state->bucket = MAX_UDP_PORTS; 3038 3039 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3040 } 3041 EXPORT_SYMBOL(udp_seq_start); 3042 3043 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3044 { 3045 struct sock *sk; 3046 3047 if (v == SEQ_START_TOKEN) 3048 sk = udp_get_idx(seq, 0); 3049 else 3050 sk = udp_get_next(seq, v); 3051 3052 ++*pos; 3053 return sk; 3054 } 3055 EXPORT_SYMBOL(udp_seq_next); 3056 3057 void udp_seq_stop(struct seq_file *seq, void *v) 3058 { 3059 struct udp_seq_afinfo *afinfo; 3060 struct udp_iter_state *state = seq->private; 3061 3062 if (state->bpf_seq_afinfo) 3063 afinfo = state->bpf_seq_afinfo; 3064 else 3065 afinfo = pde_data(file_inode(seq->file)); 3066 3067 if (state->bucket <= afinfo->udp_table->mask) 3068 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 3069 } 3070 EXPORT_SYMBOL(udp_seq_stop); 3071 3072 /* ------------------------------------------------------------------------ */ 3073 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3074 int bucket) 3075 { 3076 struct inet_sock *inet = inet_sk(sp); 3077 __be32 dest = inet->inet_daddr; 3078 __be32 src = inet->inet_rcv_saddr; 3079 __u16 destp = ntohs(inet->inet_dport); 3080 __u16 srcp = ntohs(inet->inet_sport); 3081 3082 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3083 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3084 bucket, src, srcp, dest, destp, sp->sk_state, 3085 sk_wmem_alloc_get(sp), 3086 udp_rqueue_get(sp), 3087 0, 0L, 0, 3088 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3089 0, sock_i_ino(sp), 3090 refcount_read(&sp->sk_refcnt), sp, 3091 atomic_read(&sp->sk_drops)); 3092 } 3093 3094 int udp4_seq_show(struct seq_file *seq, void *v) 3095 { 3096 seq_setwidth(seq, 127); 3097 if (v == SEQ_START_TOKEN) 3098 seq_puts(seq, " sl local_address rem_address st tx_queue " 3099 "rx_queue tr tm->when retrnsmt uid timeout " 3100 "inode ref pointer drops"); 3101 else { 3102 struct udp_iter_state *state = seq->private; 3103 3104 udp4_format_sock(v, seq, state->bucket); 3105 } 3106 seq_pad(seq, '\n'); 3107 return 0; 3108 } 3109 3110 #ifdef CONFIG_BPF_SYSCALL 3111 struct bpf_iter__udp { 3112 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3113 __bpf_md_ptr(struct udp_sock *, udp_sk); 3114 uid_t uid __aligned(8); 3115 int bucket __aligned(8); 3116 }; 3117 3118 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3119 struct udp_sock *udp_sk, uid_t uid, int bucket) 3120 { 3121 struct bpf_iter__udp ctx; 3122 3123 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3124 ctx.meta = meta; 3125 ctx.udp_sk = udp_sk; 3126 ctx.uid = uid; 3127 ctx.bucket = bucket; 3128 return bpf_iter_run_prog(prog, &ctx); 3129 } 3130 3131 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3132 { 3133 struct udp_iter_state *state = seq->private; 3134 struct bpf_iter_meta meta; 3135 struct bpf_prog *prog; 3136 struct sock *sk = v; 3137 uid_t uid; 3138 3139 if (v == SEQ_START_TOKEN) 3140 return 0; 3141 3142 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3143 meta.seq = seq; 3144 prog = bpf_iter_get_info(&meta, false); 3145 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3146 } 3147 3148 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3149 { 3150 struct bpf_iter_meta meta; 3151 struct bpf_prog *prog; 3152 3153 if (!v) { 3154 meta.seq = seq; 3155 prog = bpf_iter_get_info(&meta, true); 3156 if (prog) 3157 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3158 } 3159 3160 udp_seq_stop(seq, v); 3161 } 3162 3163 static const struct seq_operations bpf_iter_udp_seq_ops = { 3164 .start = udp_seq_start, 3165 .next = udp_seq_next, 3166 .stop = bpf_iter_udp_seq_stop, 3167 .show = bpf_iter_udp_seq_show, 3168 }; 3169 #endif 3170 3171 const struct seq_operations udp_seq_ops = { 3172 .start = udp_seq_start, 3173 .next = udp_seq_next, 3174 .stop = udp_seq_stop, 3175 .show = udp4_seq_show, 3176 }; 3177 EXPORT_SYMBOL(udp_seq_ops); 3178 3179 static struct udp_seq_afinfo udp4_seq_afinfo = { 3180 .family = AF_INET, 3181 .udp_table = &udp_table, 3182 }; 3183 3184 static int __net_init udp4_proc_init_net(struct net *net) 3185 { 3186 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3187 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3188 return -ENOMEM; 3189 return 0; 3190 } 3191 3192 static void __net_exit udp4_proc_exit_net(struct net *net) 3193 { 3194 remove_proc_entry("udp", net->proc_net); 3195 } 3196 3197 static struct pernet_operations udp4_net_ops = { 3198 .init = udp4_proc_init_net, 3199 .exit = udp4_proc_exit_net, 3200 }; 3201 3202 int __init udp4_proc_init(void) 3203 { 3204 return register_pernet_subsys(&udp4_net_ops); 3205 } 3206 3207 void udp4_proc_exit(void) 3208 { 3209 unregister_pernet_subsys(&udp4_net_ops); 3210 } 3211 #endif /* CONFIG_PROC_FS */ 3212 3213 static __initdata unsigned long uhash_entries; 3214 static int __init set_uhash_entries(char *str) 3215 { 3216 ssize_t ret; 3217 3218 if (!str) 3219 return 0; 3220 3221 ret = kstrtoul(str, 0, &uhash_entries); 3222 if (ret) 3223 return 0; 3224 3225 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3226 uhash_entries = UDP_HTABLE_SIZE_MIN; 3227 return 1; 3228 } 3229 __setup("uhash_entries=", set_uhash_entries); 3230 3231 void __init udp_table_init(struct udp_table *table, const char *name) 3232 { 3233 unsigned int i; 3234 3235 table->hash = alloc_large_system_hash(name, 3236 2 * sizeof(struct udp_hslot), 3237 uhash_entries, 3238 21, /* one slot per 2 MB */ 3239 0, 3240 &table->log, 3241 &table->mask, 3242 UDP_HTABLE_SIZE_MIN, 3243 64 * 1024); 3244 3245 table->hash2 = table->hash + (table->mask + 1); 3246 for (i = 0; i <= table->mask; i++) { 3247 INIT_HLIST_HEAD(&table->hash[i].head); 3248 table->hash[i].count = 0; 3249 spin_lock_init(&table->hash[i].lock); 3250 } 3251 for (i = 0; i <= table->mask; i++) { 3252 INIT_HLIST_HEAD(&table->hash2[i].head); 3253 table->hash2[i].count = 0; 3254 spin_lock_init(&table->hash2[i].lock); 3255 } 3256 } 3257 3258 u32 udp_flow_hashrnd(void) 3259 { 3260 static u32 hashrnd __read_mostly; 3261 3262 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3263 3264 return hashrnd; 3265 } 3266 EXPORT_SYMBOL(udp_flow_hashrnd); 3267 3268 static void __udp_sysctl_init(struct net *net) 3269 { 3270 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3271 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 3272 3273 #ifdef CONFIG_NET_L3_MASTER_DEV 3274 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3275 #endif 3276 } 3277 3278 static int __net_init udp_sysctl_init(struct net *net) 3279 { 3280 __udp_sysctl_init(net); 3281 return 0; 3282 } 3283 3284 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3285 .init = udp_sysctl_init, 3286 }; 3287 3288 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3289 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3290 struct udp_sock *udp_sk, uid_t uid, int bucket) 3291 3292 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3293 { 3294 struct udp_iter_state *st = priv_data; 3295 struct udp_seq_afinfo *afinfo; 3296 int ret; 3297 3298 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN); 3299 if (!afinfo) 3300 return -ENOMEM; 3301 3302 afinfo->family = AF_UNSPEC; 3303 afinfo->udp_table = &udp_table; 3304 st->bpf_seq_afinfo = afinfo; 3305 ret = bpf_iter_init_seq_net(priv_data, aux); 3306 if (ret) 3307 kfree(afinfo); 3308 return ret; 3309 } 3310 3311 static void bpf_iter_fini_udp(void *priv_data) 3312 { 3313 struct udp_iter_state *st = priv_data; 3314 3315 kfree(st->bpf_seq_afinfo); 3316 bpf_iter_fini_seq_net(priv_data); 3317 } 3318 3319 static const struct bpf_iter_seq_info udp_seq_info = { 3320 .seq_ops = &bpf_iter_udp_seq_ops, 3321 .init_seq_private = bpf_iter_init_udp, 3322 .fini_seq_private = bpf_iter_fini_udp, 3323 .seq_priv_size = sizeof(struct udp_iter_state), 3324 }; 3325 3326 static struct bpf_iter_reg udp_reg_info = { 3327 .target = "udp", 3328 .ctx_arg_info_size = 1, 3329 .ctx_arg_info = { 3330 { offsetof(struct bpf_iter__udp, udp_sk), 3331 PTR_TO_BTF_ID_OR_NULL }, 3332 }, 3333 .seq_info = &udp_seq_info, 3334 }; 3335 3336 static void __init bpf_iter_register(void) 3337 { 3338 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3339 if (bpf_iter_reg_target(&udp_reg_info)) 3340 pr_warn("Warning: could not register bpf iterator udp\n"); 3341 } 3342 #endif 3343 3344 void __init udp_init(void) 3345 { 3346 unsigned long limit; 3347 unsigned int i; 3348 3349 udp_table_init(&udp_table, "UDP"); 3350 limit = nr_free_buffer_pages() / 8; 3351 limit = max(limit, 128UL); 3352 sysctl_udp_mem[0] = limit / 4 * 3; 3353 sysctl_udp_mem[1] = limit; 3354 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3355 3356 __udp_sysctl_init(&init_net); 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