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 <linux/sock_diag.h> 97 #include <net/tcp_states.h> 98 #include <linux/skbuff.h> 99 #include <linux/proc_fs.h> 100 #include <linux/seq_file.h> 101 #include <net/net_namespace.h> 102 #include <net/icmp.h> 103 #include <net/inet_hashtables.h> 104 #include <net/ip.h> 105 #include <net/ip_tunnels.h> 106 #include <net/route.h> 107 #include <net/checksum.h> 108 #include <net/gso.h> 109 #include <net/xfrm.h> 110 #include <trace/events/udp.h> 111 #include <linux/static_key.h> 112 #include <linux/btf_ids.h> 113 #include <trace/events/skb.h> 114 #include <net/busy_poll.h> 115 #include "udp_impl.h" 116 #include <net/sock_reuseport.h> 117 #include <net/addrconf.h> 118 #include <net/udp_tunnel.h> 119 #include <net/gro.h> 120 #if IS_ENABLED(CONFIG_IPV6) 121 #include <net/ipv6_stubs.h> 122 #endif 123 #include <net/rps.h> 124 125 struct udp_table udp_table __read_mostly; 126 127 long sysctl_udp_mem[3] __read_mostly; 128 EXPORT_IPV6_MOD(sysctl_udp_mem); 129 130 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc); 131 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc); 132 133 #define MAX_UDP_PORTS 65536 134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET) 135 136 static struct udp_table *udp_get_table_prot(struct sock *sk) 137 { 138 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table; 139 } 140 141 static int udp_lib_lport_inuse(struct net *net, __u16 num, 142 const struct udp_hslot *hslot, 143 unsigned long *bitmap, 144 struct sock *sk, unsigned int log) 145 { 146 kuid_t uid = sk_uid(sk); 147 struct sock *sk2; 148 149 sk_for_each(sk2, &hslot->head) { 150 if (net_eq(sock_net(sk2), net) && 151 sk2 != sk && 152 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 153 (!sk2->sk_reuse || !sk->sk_reuse) && 154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 156 inet_rcv_saddr_equal(sk, sk2, true)) { 157 if (sk2->sk_reuseport && sk->sk_reuseport && 158 !rcu_access_pointer(sk->sk_reuseport_cb) && 159 uid_eq(uid, sk_uid(sk2))) { 160 if (!bitmap) 161 return 0; 162 } else { 163 if (!bitmap) 164 return 1; 165 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 166 bitmap); 167 } 168 } 169 } 170 return 0; 171 } 172 173 /* 174 * Note: we still hold spinlock of primary hash chain, so no other writer 175 * can insert/delete a socket with local_port == num 176 */ 177 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 178 struct udp_hslot *hslot2, 179 struct sock *sk) 180 { 181 kuid_t uid = sk_uid(sk); 182 struct sock *sk2; 183 int res = 0; 184 185 spin_lock(&hslot2->lock); 186 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 187 if (net_eq(sock_net(sk2), net) && 188 sk2 != sk && 189 (udp_sk(sk2)->udp_port_hash == num) && 190 (!sk2->sk_reuse || !sk->sk_reuse) && 191 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 192 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 193 inet_rcv_saddr_equal(sk, sk2, true)) { 194 if (sk2->sk_reuseport && sk->sk_reuseport && 195 !rcu_access_pointer(sk->sk_reuseport_cb) && 196 uid_eq(uid, sk_uid(sk2))) { 197 res = 0; 198 } else { 199 res = 1; 200 } 201 break; 202 } 203 } 204 spin_unlock(&hslot2->lock); 205 return res; 206 } 207 208 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 209 { 210 struct net *net = sock_net(sk); 211 kuid_t uid = sk_uid(sk); 212 struct sock *sk2; 213 214 sk_for_each(sk2, &hslot->head) { 215 if (net_eq(sock_net(sk2), net) && 216 sk2 != sk && 217 sk2->sk_family == sk->sk_family && 218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 221 sk2->sk_reuseport && uid_eq(uid, sk_uid(sk2)) && 222 inet_rcv_saddr_equal(sk, sk2, false)) { 223 return reuseport_add_sock(sk, sk2, 224 inet_rcv_saddr_any(sk)); 225 } 226 } 227 228 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 229 } 230 231 /** 232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 233 * 234 * @sk: socket struct in question 235 * @snum: port number to look up 236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 237 * with NULL address 238 */ 239 int udp_lib_get_port(struct sock *sk, unsigned short snum, 240 unsigned int hash2_nulladdr) 241 { 242 struct udp_table *udptable = udp_get_table_prot(sk); 243 struct udp_hslot *hslot, *hslot2; 244 struct net *net = sock_net(sk); 245 int error = -EADDRINUSE; 246 247 if (!snum) { 248 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 249 unsigned short first, last; 250 int low, high, remaining; 251 unsigned int rand; 252 253 inet_sk_get_local_port_range(sk, &low, &high); 254 remaining = (high - low) + 1; 255 256 rand = get_random_u32(); 257 first = reciprocal_scale(rand, remaining) + low; 258 /* 259 * force rand to be an odd multiple of UDP_HTABLE_SIZE 260 */ 261 rand = (rand | 1) * (udptable->mask + 1); 262 last = first + udptable->mask + 1; 263 do { 264 hslot = udp_hashslot(udptable, net, first); 265 bitmap_zero(bitmap, PORTS_PER_CHAIN); 266 spin_lock_bh(&hslot->lock); 267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 268 udptable->log); 269 270 snum = first; 271 /* 272 * Iterate on all possible values of snum for this hash. 273 * Using steps of an odd multiple of UDP_HTABLE_SIZE 274 * give us randomization and full range coverage. 275 */ 276 do { 277 if (low <= snum && snum <= high && 278 !test_bit(snum >> udptable->log, bitmap) && 279 !inet_is_local_reserved_port(net, snum)) 280 goto found; 281 snum += rand; 282 } while (snum != first); 283 spin_unlock_bh(&hslot->lock); 284 cond_resched(); 285 } while (++first != last); 286 goto fail; 287 } else { 288 hslot = udp_hashslot(udptable, net, snum); 289 spin_lock_bh(&hslot->lock); 290 if (hslot->count > 10) { 291 int exist; 292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 293 294 slot2 &= udptable->mask; 295 hash2_nulladdr &= udptable->mask; 296 297 hslot2 = udp_hashslot2(udptable, slot2); 298 if (hslot->count < hslot2->count) 299 goto scan_primary_hash; 300 301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 302 if (!exist && (hash2_nulladdr != slot2)) { 303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 304 exist = udp_lib_lport_inuse2(net, snum, hslot2, 305 sk); 306 } 307 if (exist) 308 goto fail_unlock; 309 else 310 goto found; 311 } 312 scan_primary_hash: 313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 314 goto fail_unlock; 315 } 316 found: 317 inet_sk(sk)->inet_num = snum; 318 udp_sk(sk)->udp_port_hash = snum; 319 udp_sk(sk)->udp_portaddr_hash ^= snum; 320 if (sk_unhashed(sk)) { 321 if (sk->sk_reuseport && 322 udp_reuseport_add_sock(sk, hslot)) { 323 inet_sk(sk)->inet_num = 0; 324 udp_sk(sk)->udp_port_hash = 0; 325 udp_sk(sk)->udp_portaddr_hash ^= snum; 326 goto fail_unlock; 327 } 328 329 sock_set_flag(sk, SOCK_RCU_FREE); 330 331 sk_add_node_rcu(sk, &hslot->head); 332 hslot->count++; 333 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 334 335 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 336 spin_lock(&hslot2->lock); 337 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 338 sk->sk_family == AF_INET6) 339 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 340 &hslot2->head); 341 else 342 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 343 &hslot2->head); 344 hslot2->count++; 345 spin_unlock(&hslot2->lock); 346 } 347 348 error = 0; 349 fail_unlock: 350 spin_unlock_bh(&hslot->lock); 351 fail: 352 return error; 353 } 354 EXPORT_IPV6_MOD(udp_lib_get_port); 355 356 int udp_v4_get_port(struct sock *sk, unsigned short snum) 357 { 358 unsigned int hash2_nulladdr = 359 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 360 unsigned int hash2_partial = 361 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 362 363 /* precompute partial secondary hash */ 364 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 365 return udp_lib_get_port(sk, snum, hash2_nulladdr); 366 } 367 368 static int compute_score(struct sock *sk, const struct net *net, 369 __be32 saddr, __be16 sport, 370 __be32 daddr, unsigned short hnum, 371 int dif, int sdif) 372 { 373 int score; 374 struct inet_sock *inet; 375 bool dev_match; 376 377 if (!net_eq(sock_net(sk), net) || 378 udp_sk(sk)->udp_port_hash != hnum || 379 ipv6_only_sock(sk)) 380 return -1; 381 382 if (sk->sk_rcv_saddr != daddr) 383 return -1; 384 385 score = (sk->sk_family == PF_INET) ? 2 : 1; 386 387 inet = inet_sk(sk); 388 if (inet->inet_daddr) { 389 if (inet->inet_daddr != saddr) 390 return -1; 391 score += 4; 392 } 393 394 if (inet->inet_dport) { 395 if (inet->inet_dport != sport) 396 return -1; 397 score += 4; 398 } 399 400 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 401 dif, sdif); 402 if (!dev_match) 403 return -1; 404 if (sk->sk_bound_dev_if) 405 score += 4; 406 407 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 408 score++; 409 return score; 410 } 411 412 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport, 413 const __be32 faddr, const __be16 fport) 414 { 415 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 416 417 return __inet_ehashfn(laddr, lport, faddr, fport, 418 udp_ehash_secret + net_hash_mix(net)); 419 } 420 EXPORT_IPV6_MOD(udp_ehashfn); 421 422 /** 423 * udp4_lib_lookup1() - Simplified lookup using primary hash (destination port) 424 * @net: Network namespace 425 * @saddr: Source address, network order 426 * @sport: Source port, network order 427 * @daddr: Destination address, network order 428 * @hnum: Destination port, host order 429 * @dif: Destination interface index 430 * @sdif: Destination bridge port index, if relevant 431 * @udptable: Set of UDP hash tables 432 * 433 * Simplified lookup to be used as fallback if no sockets are found due to a 434 * potential race between (receive) address change, and lookup happening before 435 * the rehash operation. This function ignores SO_REUSEPORT groups while scoring 436 * result sockets, because if we have one, we don't need the fallback at all. 437 * 438 * Called under rcu_read_lock(). 439 * 440 * Return: socket with highest matching score if any, NULL if none 441 */ 442 static struct sock *udp4_lib_lookup1(const struct net *net, 443 __be32 saddr, __be16 sport, 444 __be32 daddr, unsigned int hnum, 445 int dif, int sdif, 446 const struct udp_table *udptable) 447 { 448 unsigned int slot = udp_hashfn(net, hnum, udptable->mask); 449 struct udp_hslot *hslot = &udptable->hash[slot]; 450 struct sock *sk, *result = NULL; 451 int score, badness = 0; 452 453 sk_for_each_rcu(sk, &hslot->head) { 454 score = compute_score(sk, net, 455 saddr, sport, daddr, hnum, dif, sdif); 456 if (score > badness) { 457 result = sk; 458 badness = score; 459 } 460 } 461 462 return result; 463 } 464 465 /* called with rcu_read_lock() */ 466 static struct sock *udp4_lib_lookup2(const struct net *net, 467 __be32 saddr, __be16 sport, 468 __be32 daddr, unsigned int hnum, 469 int dif, int sdif, 470 struct udp_hslot *hslot2, 471 struct sk_buff *skb) 472 { 473 struct sock *sk, *result; 474 int score, badness; 475 bool need_rescore; 476 477 result = NULL; 478 badness = 0; 479 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 480 need_rescore = false; 481 rescore: 482 score = compute_score(need_rescore ? result : sk, net, saddr, 483 sport, daddr, hnum, dif, sdif); 484 if (score > badness) { 485 badness = score; 486 487 if (need_rescore) 488 continue; 489 490 if (sk->sk_state == TCP_ESTABLISHED) { 491 result = sk; 492 continue; 493 } 494 495 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 496 saddr, sport, daddr, hnum, udp_ehashfn); 497 if (!result) { 498 result = sk; 499 continue; 500 } 501 502 /* Fall back to scoring if group has connections */ 503 if (!reuseport_has_conns(sk)) 504 return result; 505 506 /* Reuseport logic returned an error, keep original score. */ 507 if (IS_ERR(result)) 508 continue; 509 510 /* compute_score is too long of a function to be 511 * inlined, and calling it again here yields 512 * measureable overhead for some 513 * workloads. Work around it by jumping 514 * backwards to rescore 'result'. 515 */ 516 need_rescore = true; 517 goto rescore; 518 } 519 } 520 return result; 521 } 522 523 #if IS_ENABLED(CONFIG_BASE_SMALL) 524 static struct sock *udp4_lib_lookup4(const struct net *net, 525 __be32 saddr, __be16 sport, 526 __be32 daddr, unsigned int hnum, 527 int dif, int sdif, 528 struct udp_table *udptable) 529 { 530 return NULL; 531 } 532 533 static void udp_rehash4(struct udp_table *udptable, struct sock *sk, 534 u16 newhash4) 535 { 536 } 537 538 static void udp_unhash4(struct udp_table *udptable, struct sock *sk) 539 { 540 } 541 #else /* !CONFIG_BASE_SMALL */ 542 static struct sock *udp4_lib_lookup4(const struct net *net, 543 __be32 saddr, __be16 sport, 544 __be32 daddr, unsigned int hnum, 545 int dif, int sdif, 546 struct udp_table *udptable) 547 { 548 const __portpair ports = INET_COMBINED_PORTS(sport, hnum); 549 const struct hlist_nulls_node *node; 550 struct udp_hslot *hslot4; 551 unsigned int hash4, slot; 552 struct udp_sock *up; 553 struct sock *sk; 554 555 hash4 = udp_ehashfn(net, daddr, hnum, saddr, sport); 556 slot = hash4 & udptable->mask; 557 hslot4 = &udptable->hash4[slot]; 558 INET_ADDR_COOKIE(acookie, saddr, daddr); 559 560 begin: 561 /* SLAB_TYPESAFE_BY_RCU not used, so we don't need to touch sk_refcnt */ 562 udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) { 563 sk = (struct sock *)up; 564 if (inet_match(net, sk, acookie, ports, dif, sdif)) 565 return sk; 566 } 567 568 /* if the nulls value we got at the end of this lookup is not the 569 * expected one, we must restart lookup. We probably met an item that 570 * was moved to another chain due to rehash. 571 */ 572 if (get_nulls_value(node) != slot) 573 goto begin; 574 575 return NULL; 576 } 577 578 /* udp_rehash4() only checks hslot4, and hash4_cnt is not processed. */ 579 static void udp_rehash4(struct udp_table *udptable, struct sock *sk, 580 u16 newhash4) 581 { 582 struct udp_hslot *hslot4, *nhslot4; 583 584 hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash); 585 nhslot4 = udp_hashslot4(udptable, newhash4); 586 udp_sk(sk)->udp_lrpa_hash = newhash4; 587 588 if (hslot4 != nhslot4) { 589 spin_lock_bh(&hslot4->lock); 590 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node); 591 hslot4->count--; 592 spin_unlock_bh(&hslot4->lock); 593 594 spin_lock_bh(&nhslot4->lock); 595 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node, 596 &nhslot4->nulls_head); 597 nhslot4->count++; 598 spin_unlock_bh(&nhslot4->lock); 599 } 600 } 601 602 static void udp_unhash4(struct udp_table *udptable, struct sock *sk) 603 { 604 struct udp_hslot *hslot2, *hslot4; 605 606 if (udp_hashed4(sk)) { 607 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 608 hslot4 = udp_hashslot4(udptable, udp_sk(sk)->udp_lrpa_hash); 609 610 spin_lock(&hslot4->lock); 611 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_lrpa_node); 612 hslot4->count--; 613 spin_unlock(&hslot4->lock); 614 615 spin_lock(&hslot2->lock); 616 udp_hash4_dec(hslot2); 617 spin_unlock(&hslot2->lock); 618 } 619 } 620 621 void udp_lib_hash4(struct sock *sk, u16 hash) 622 { 623 struct udp_hslot *hslot, *hslot2, *hslot4; 624 struct net *net = sock_net(sk); 625 struct udp_table *udptable; 626 627 /* Connected udp socket can re-connect to another remote address, which 628 * will be handled by rehash. Thus no need to redo hash4 here. 629 */ 630 if (udp_hashed4(sk)) 631 return; 632 633 udptable = net->ipv4.udp_table; 634 hslot = udp_hashslot(udptable, net, udp_sk(sk)->udp_port_hash); 635 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 636 hslot4 = udp_hashslot4(udptable, hash); 637 udp_sk(sk)->udp_lrpa_hash = hash; 638 639 spin_lock_bh(&hslot->lock); 640 if (rcu_access_pointer(sk->sk_reuseport_cb)) 641 reuseport_detach_sock(sk); 642 643 spin_lock(&hslot4->lock); 644 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_lrpa_node, 645 &hslot4->nulls_head); 646 hslot4->count++; 647 spin_unlock(&hslot4->lock); 648 649 spin_lock(&hslot2->lock); 650 udp_hash4_inc(hslot2); 651 spin_unlock(&hslot2->lock); 652 653 spin_unlock_bh(&hslot->lock); 654 } 655 EXPORT_IPV6_MOD(udp_lib_hash4); 656 657 /* call with sock lock */ 658 void udp4_hash4(struct sock *sk) 659 { 660 struct net *net = sock_net(sk); 661 unsigned int hash; 662 663 if (sk_unhashed(sk) || sk->sk_rcv_saddr == htonl(INADDR_ANY)) 664 return; 665 666 hash = udp_ehashfn(net, sk->sk_rcv_saddr, sk->sk_num, 667 sk->sk_daddr, sk->sk_dport); 668 669 udp_lib_hash4(sk, hash); 670 } 671 EXPORT_IPV6_MOD(udp4_hash4); 672 #endif /* CONFIG_BASE_SMALL */ 673 674 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 675 * harder than this. -DaveM 676 */ 677 struct sock *__udp4_lib_lookup(const struct net *net, __be32 saddr, 678 __be16 sport, __be32 daddr, __be16 dport, int dif, 679 int sdif, struct udp_table *udptable, struct sk_buff *skb) 680 { 681 unsigned short hnum = ntohs(dport); 682 struct udp_hslot *hslot2; 683 struct sock *result, *sk; 684 unsigned int hash2; 685 686 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 687 hslot2 = udp_hashslot2(udptable, hash2); 688 689 if (udp_has_hash4(hslot2)) { 690 result = udp4_lib_lookup4(net, saddr, sport, daddr, hnum, 691 dif, sdif, udptable); 692 if (result) /* udp4_lib_lookup4 return sk or NULL */ 693 return result; 694 } 695 696 /* Lookup connected or non-wildcard socket */ 697 result = udp4_lib_lookup2(net, saddr, sport, 698 daddr, hnum, dif, sdif, 699 hslot2, skb); 700 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 701 goto done; 702 703 /* Lookup redirect from BPF */ 704 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 705 udptable == net->ipv4.udp_table) { 706 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 707 saddr, sport, daddr, hnum, dif, 708 udp_ehashfn); 709 if (sk) { 710 result = sk; 711 goto done; 712 } 713 } 714 715 /* Got non-wildcard socket or error on first lookup */ 716 if (result) 717 goto done; 718 719 /* Lookup wildcard sockets */ 720 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 721 hslot2 = udp_hashslot2(udptable, hash2); 722 723 result = udp4_lib_lookup2(net, saddr, sport, 724 htonl(INADDR_ANY), hnum, dif, sdif, 725 hslot2, skb); 726 if (!IS_ERR_OR_NULL(result)) 727 goto done; 728 729 /* Primary hash (destination port) lookup as fallback for this race: 730 * 1. __ip4_datagram_connect() sets sk_rcv_saddr 731 * 2. lookup (this function): new sk_rcv_saddr, hashes not updated yet 732 * 3. rehash operation updating _secondary and four-tuple_ hashes 733 * The primary hash doesn't need an update after 1., so, thanks to this 734 * further step, 1. and 3. don't need to be atomic against the lookup. 735 */ 736 result = udp4_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif, 737 udptable); 738 739 done: 740 if (IS_ERR(result)) 741 return NULL; 742 return result; 743 } 744 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 745 746 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 747 __be16 sport, __be16 dport, 748 struct udp_table *udptable) 749 { 750 const struct iphdr *iph = ip_hdr(skb); 751 752 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 753 iph->daddr, dport, inet_iif(skb), 754 inet_sdif(skb), udptable, skb); 755 } 756 757 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 758 __be16 sport, __be16 dport) 759 { 760 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation]; 761 const struct iphdr *iph = (struct iphdr *)(skb->data + offset); 762 struct net *net = dev_net(skb->dev); 763 int iif, sdif; 764 765 inet_get_iif_sdif(skb, &iif, &sdif); 766 767 return __udp4_lib_lookup(net, iph->saddr, sport, 768 iph->daddr, dport, iif, 769 sdif, net->ipv4.udp_table, NULL); 770 } 771 772 /* Must be called under rcu_read_lock(). 773 * Does increment socket refcount. 774 */ 775 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 776 struct sock *udp4_lib_lookup(const struct net *net, __be32 saddr, __be16 sport, 777 __be32 daddr, __be16 dport, int dif) 778 { 779 struct sock *sk; 780 781 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 782 dif, 0, net->ipv4.udp_table, NULL); 783 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 784 sk = NULL; 785 return sk; 786 } 787 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 788 #endif 789 790 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk, 791 __be16 loc_port, __be32 loc_addr, 792 __be16 rmt_port, __be32 rmt_addr, 793 int dif, int sdif, unsigned short hnum) 794 { 795 const struct inet_sock *inet = inet_sk(sk); 796 797 if (!net_eq(sock_net(sk), net) || 798 udp_sk(sk)->udp_port_hash != hnum || 799 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 800 (inet->inet_dport != rmt_port && inet->inet_dport) || 801 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 802 ipv6_only_sock(sk) || 803 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 804 return false; 805 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 806 return false; 807 return true; 808 } 809 810 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 811 EXPORT_IPV6_MOD(udp_encap_needed_key); 812 813 #if IS_ENABLED(CONFIG_IPV6) 814 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 815 EXPORT_IPV6_MOD(udpv6_encap_needed_key); 816 #endif 817 818 void udp_encap_enable(void) 819 { 820 static_branch_inc(&udp_encap_needed_key); 821 } 822 EXPORT_SYMBOL(udp_encap_enable); 823 824 void udp_encap_disable(void) 825 { 826 static_branch_dec(&udp_encap_needed_key); 827 } 828 EXPORT_SYMBOL(udp_encap_disable); 829 830 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 831 * through error handlers in encapsulations looking for a match. 832 */ 833 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 834 { 835 int i; 836 837 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 838 int (*handler)(struct sk_buff *skb, u32 info); 839 const struct ip_tunnel_encap_ops *encap; 840 841 encap = rcu_dereference(iptun_encaps[i]); 842 if (!encap) 843 continue; 844 handler = encap->err_handler; 845 if (handler && !handler(skb, info)) 846 return 0; 847 } 848 849 return -ENOENT; 850 } 851 852 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 853 * reversing source and destination port: this will match tunnels that force the 854 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 855 * lwtunnels might actually break this assumption by being configured with 856 * different destination ports on endpoints, in this case we won't be able to 857 * trace ICMP messages back to them. 858 * 859 * If this doesn't match any socket, probe tunnels with arbitrary destination 860 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 861 * we've sent packets to won't necessarily match the local destination port. 862 * 863 * Then ask the tunnel implementation to match the error against a valid 864 * association. 865 * 866 * Return an error if we can't find a match, the socket if we need further 867 * processing, zero otherwise. 868 */ 869 static struct sock *__udp4_lib_err_encap(struct net *net, 870 const struct iphdr *iph, 871 struct udphdr *uh, 872 struct udp_table *udptable, 873 struct sock *sk, 874 struct sk_buff *skb, u32 info) 875 { 876 int (*lookup)(struct sock *sk, struct sk_buff *skb); 877 int network_offset, transport_offset; 878 struct udp_sock *up; 879 880 network_offset = skb_network_offset(skb); 881 transport_offset = skb_transport_offset(skb); 882 883 /* Network header needs to point to the outer IPv4 header inside ICMP */ 884 skb_reset_network_header(skb); 885 886 /* Transport header needs to point to the UDP header */ 887 skb_set_transport_header(skb, iph->ihl << 2); 888 889 if (sk) { 890 up = udp_sk(sk); 891 892 lookup = READ_ONCE(up->encap_err_lookup); 893 if (lookup && lookup(sk, skb)) 894 sk = NULL; 895 896 goto out; 897 } 898 899 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 900 iph->saddr, uh->dest, skb->dev->ifindex, 0, 901 udptable, NULL); 902 if (sk) { 903 up = udp_sk(sk); 904 905 lookup = READ_ONCE(up->encap_err_lookup); 906 if (!lookup || lookup(sk, skb)) 907 sk = NULL; 908 } 909 910 out: 911 if (!sk) 912 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 913 914 skb_set_transport_header(skb, transport_offset); 915 skb_set_network_header(skb, network_offset); 916 917 return sk; 918 } 919 920 /* 921 * This routine is called by the ICMP module when it gets some 922 * sort of error condition. If err < 0 then the socket should 923 * be closed and the error returned to the user. If err > 0 924 * it's just the icmp type << 8 | icmp code. 925 * Header points to the ip header of the error packet. We move 926 * on past this. Then (as it used to claim before adjustment) 927 * header points to the first 8 bytes of the udp header. We need 928 * to find the appropriate port. 929 */ 930 931 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 932 { 933 struct inet_sock *inet; 934 const struct iphdr *iph = (const struct iphdr *)skb->data; 935 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 936 const int type = icmp_hdr(skb)->type; 937 const int code = icmp_hdr(skb)->code; 938 bool tunnel = false; 939 struct sock *sk; 940 int harderr; 941 int err; 942 struct net *net = dev_net(skb->dev); 943 944 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 945 iph->saddr, uh->source, skb->dev->ifindex, 946 inet_sdif(skb), udptable, NULL); 947 948 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) { 949 /* No socket for error: try tunnels before discarding */ 950 if (static_branch_unlikely(&udp_encap_needed_key)) { 951 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 952 info); 953 if (!sk) 954 return 0; 955 } else 956 sk = ERR_PTR(-ENOENT); 957 958 if (IS_ERR(sk)) { 959 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 960 return PTR_ERR(sk); 961 } 962 963 tunnel = true; 964 } 965 966 err = 0; 967 harderr = 0; 968 inet = inet_sk(sk); 969 970 switch (type) { 971 default: 972 case ICMP_TIME_EXCEEDED: 973 err = EHOSTUNREACH; 974 break; 975 case ICMP_SOURCE_QUENCH: 976 goto out; 977 case ICMP_PARAMETERPROB: 978 err = EPROTO; 979 harderr = 1; 980 break; 981 case ICMP_DEST_UNREACH: 982 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 983 ipv4_sk_update_pmtu(skb, sk, info); 984 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) { 985 err = EMSGSIZE; 986 harderr = 1; 987 break; 988 } 989 goto out; 990 } 991 err = EHOSTUNREACH; 992 if (code <= NR_ICMP_UNREACH) { 993 harderr = icmp_err_convert[code].fatal; 994 err = icmp_err_convert[code].errno; 995 } 996 break; 997 case ICMP_REDIRECT: 998 ipv4_sk_redirect(skb, sk); 999 goto out; 1000 } 1001 1002 /* 1003 * RFC1122: OK. Passes ICMP errors back to application, as per 1004 * 4.1.3.3. 1005 */ 1006 if (tunnel) { 1007 /* ...not for tunnels though: we don't have a sending socket */ 1008 if (udp_sk(sk)->encap_err_rcv) 1009 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info, 1010 (u8 *)(uh+1)); 1011 goto out; 1012 } 1013 if (!inet_test_bit(RECVERR, sk)) { 1014 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 1015 goto out; 1016 } else 1017 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 1018 1019 sk->sk_err = err; 1020 sk_error_report(sk); 1021 out: 1022 return 0; 1023 } 1024 1025 int udp_err(struct sk_buff *skb, u32 info) 1026 { 1027 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table); 1028 } 1029 1030 /* 1031 * Throw away all pending data and cancel the corking. Socket is locked. 1032 */ 1033 void udp_flush_pending_frames(struct sock *sk) 1034 { 1035 struct udp_sock *up = udp_sk(sk); 1036 1037 if (up->pending) { 1038 up->len = 0; 1039 WRITE_ONCE(up->pending, 0); 1040 ip_flush_pending_frames(sk); 1041 } 1042 } 1043 EXPORT_IPV6_MOD(udp_flush_pending_frames); 1044 1045 /** 1046 * udp4_hwcsum - handle outgoing HW checksumming 1047 * @skb: sk_buff containing the filled-in UDP header 1048 * (checksum field must be zeroed out) 1049 * @src: source IP address 1050 * @dst: destination IP address 1051 */ 1052 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 1053 { 1054 struct udphdr *uh = udp_hdr(skb); 1055 int offset = skb_transport_offset(skb); 1056 int len = skb->len - offset; 1057 int hlen = len; 1058 __wsum csum = 0; 1059 1060 if (!skb_has_frag_list(skb)) { 1061 /* 1062 * Only one fragment on the socket. 1063 */ 1064 skb->csum_start = skb_transport_header(skb) - skb->head; 1065 skb->csum_offset = offsetof(struct udphdr, check); 1066 uh->check = ~csum_tcpudp_magic(src, dst, len, 1067 IPPROTO_UDP, 0); 1068 } else { 1069 struct sk_buff *frags; 1070 1071 /* 1072 * HW-checksum won't work as there are two or more 1073 * fragments on the socket so that all csums of sk_buffs 1074 * should be together 1075 */ 1076 skb_walk_frags(skb, frags) { 1077 csum = csum_add(csum, frags->csum); 1078 hlen -= frags->len; 1079 } 1080 1081 csum = skb_checksum(skb, offset, hlen, csum); 1082 skb->ip_summed = CHECKSUM_NONE; 1083 1084 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 1085 if (uh->check == 0) 1086 uh->check = CSUM_MANGLED_0; 1087 } 1088 } 1089 EXPORT_SYMBOL_GPL(udp4_hwcsum); 1090 1091 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 1092 * for the simple case like when setting the checksum for a UDP tunnel. 1093 */ 1094 void udp_set_csum(bool nocheck, struct sk_buff *skb, 1095 __be32 saddr, __be32 daddr, int len) 1096 { 1097 struct udphdr *uh = udp_hdr(skb); 1098 1099 if (nocheck) { 1100 uh->check = 0; 1101 } else if (skb_is_gso(skb)) { 1102 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 1103 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 1104 uh->check = 0; 1105 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 1106 if (uh->check == 0) 1107 uh->check = CSUM_MANGLED_0; 1108 } else { 1109 skb->ip_summed = CHECKSUM_PARTIAL; 1110 skb->csum_start = skb_transport_header(skb) - skb->head; 1111 skb->csum_offset = offsetof(struct udphdr, check); 1112 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 1113 } 1114 } 1115 EXPORT_SYMBOL(udp_set_csum); 1116 1117 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 1118 struct inet_cork *cork) 1119 { 1120 struct sock *sk = skb->sk; 1121 struct inet_sock *inet = inet_sk(sk); 1122 struct udphdr *uh; 1123 int err; 1124 int is_udplite = IS_UDPLITE(sk); 1125 int offset = skb_transport_offset(skb); 1126 int len = skb->len - offset; 1127 int datalen = len - sizeof(*uh); 1128 __wsum csum = 0; 1129 1130 /* 1131 * Create a UDP header 1132 */ 1133 uh = udp_hdr(skb); 1134 uh->source = inet->inet_sport; 1135 uh->dest = fl4->fl4_dport; 1136 uh->len = htons(len); 1137 uh->check = 0; 1138 1139 if (cork->gso_size) { 1140 const int hlen = skb_network_header_len(skb) + 1141 sizeof(struct udphdr); 1142 1143 if (hlen + min(datalen, cork->gso_size) > cork->fragsize) { 1144 kfree_skb(skb); 1145 return -EMSGSIZE; 1146 } 1147 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1148 kfree_skb(skb); 1149 return -EINVAL; 1150 } 1151 if (sk->sk_no_check_tx) { 1152 kfree_skb(skb); 1153 return -EINVAL; 1154 } 1155 if (is_udplite || dst_xfrm(skb_dst(skb))) { 1156 kfree_skb(skb); 1157 return -EIO; 1158 } 1159 1160 if (datalen > cork->gso_size) { 1161 skb_shinfo(skb)->gso_size = cork->gso_size; 1162 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1163 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1164 cork->gso_size); 1165 1166 /* Don't checksum the payload, skb will get segmented */ 1167 goto csum_partial; 1168 } 1169 } 1170 1171 if (is_udplite) /* UDP-Lite */ 1172 csum = udplite_csum(skb); 1173 1174 else if (sk->sk_no_check_tx) { /* UDP csum off */ 1175 1176 skb->ip_summed = CHECKSUM_NONE; 1177 goto send; 1178 1179 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1180 csum_partial: 1181 1182 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 1183 goto send; 1184 1185 } else 1186 csum = udp_csum(skb); 1187 1188 /* add protocol-dependent pseudo-header */ 1189 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 1190 sk->sk_protocol, csum); 1191 if (uh->check == 0) 1192 uh->check = CSUM_MANGLED_0; 1193 1194 send: 1195 err = ip_send_skb(sock_net(sk), skb); 1196 if (err) { 1197 if (err == -ENOBUFS && 1198 !inet_test_bit(RECVERR, sk)) { 1199 UDP_INC_STATS(sock_net(sk), 1200 UDP_MIB_SNDBUFERRORS, is_udplite); 1201 err = 0; 1202 } 1203 } else 1204 UDP_INC_STATS(sock_net(sk), 1205 UDP_MIB_OUTDATAGRAMS, is_udplite); 1206 return err; 1207 } 1208 1209 /* 1210 * Push out all pending data as one UDP datagram. Socket is locked. 1211 */ 1212 int udp_push_pending_frames(struct sock *sk) 1213 { 1214 struct udp_sock *up = udp_sk(sk); 1215 struct inet_sock *inet = inet_sk(sk); 1216 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 1217 struct sk_buff *skb; 1218 int err = 0; 1219 1220 skb = ip_finish_skb(sk, fl4); 1221 if (!skb) 1222 goto out; 1223 1224 err = udp_send_skb(skb, fl4, &inet->cork.base); 1225 1226 out: 1227 up->len = 0; 1228 WRITE_ONCE(up->pending, 0); 1229 return err; 1230 } 1231 EXPORT_IPV6_MOD(udp_push_pending_frames); 1232 1233 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1234 { 1235 switch (cmsg->cmsg_type) { 1236 case UDP_SEGMENT: 1237 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1238 return -EINVAL; 1239 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1240 return 0; 1241 default: 1242 return -EINVAL; 1243 } 1244 } 1245 1246 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1247 { 1248 struct cmsghdr *cmsg; 1249 bool need_ip = false; 1250 int err; 1251 1252 for_each_cmsghdr(cmsg, msg) { 1253 if (!CMSG_OK(msg, cmsg)) 1254 return -EINVAL; 1255 1256 if (cmsg->cmsg_level != SOL_UDP) { 1257 need_ip = true; 1258 continue; 1259 } 1260 1261 err = __udp_cmsg_send(cmsg, gso_size); 1262 if (err) 1263 return err; 1264 } 1265 1266 return need_ip; 1267 } 1268 EXPORT_IPV6_MOD_GPL(udp_cmsg_send); 1269 1270 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1271 { 1272 struct inet_sock *inet = inet_sk(sk); 1273 struct udp_sock *up = udp_sk(sk); 1274 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1275 struct flowi4 fl4_stack; 1276 struct flowi4 *fl4; 1277 int ulen = len; 1278 struct ipcm_cookie ipc; 1279 struct rtable *rt = NULL; 1280 int free = 0; 1281 int connected = 0; 1282 __be32 daddr, faddr, saddr; 1283 u8 scope; 1284 __be16 dport; 1285 int err, is_udplite = IS_UDPLITE(sk); 1286 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE; 1287 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1288 struct sk_buff *skb; 1289 struct ip_options_data opt_copy; 1290 int uc_index; 1291 1292 if (len > 0xFFFF) 1293 return -EMSGSIZE; 1294 1295 /* 1296 * Check the flags. 1297 */ 1298 1299 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1300 return -EOPNOTSUPP; 1301 1302 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1303 1304 fl4 = &inet->cork.fl.u.ip4; 1305 if (READ_ONCE(up->pending)) { 1306 /* 1307 * There are pending frames. 1308 * The socket lock must be held while it's corked. 1309 */ 1310 lock_sock(sk); 1311 if (likely(up->pending)) { 1312 if (unlikely(up->pending != AF_INET)) { 1313 release_sock(sk); 1314 return -EINVAL; 1315 } 1316 goto do_append_data; 1317 } 1318 release_sock(sk); 1319 } 1320 ulen += sizeof(struct udphdr); 1321 1322 /* 1323 * Get and verify the address. 1324 */ 1325 if (usin) { 1326 if (msg->msg_namelen < sizeof(*usin)) 1327 return -EINVAL; 1328 if (usin->sin_family != AF_INET) { 1329 if (usin->sin_family != AF_UNSPEC) 1330 return -EAFNOSUPPORT; 1331 } 1332 1333 daddr = usin->sin_addr.s_addr; 1334 dport = usin->sin_port; 1335 if (dport == 0) 1336 return -EINVAL; 1337 } else { 1338 if (sk->sk_state != TCP_ESTABLISHED) 1339 return -EDESTADDRREQ; 1340 daddr = inet->inet_daddr; 1341 dport = inet->inet_dport; 1342 /* Open fast path for connected socket. 1343 Route will not be used, if at least one option is set. 1344 */ 1345 connected = 1; 1346 } 1347 1348 ipcm_init_sk(&ipc, inet); 1349 ipc.gso_size = READ_ONCE(up->gso_size); 1350 1351 if (msg->msg_controllen) { 1352 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1353 if (err > 0) { 1354 err = ip_cmsg_send(sk, msg, &ipc, 1355 sk->sk_family == AF_INET6); 1356 connected = 0; 1357 } 1358 if (unlikely(err < 0)) { 1359 kfree(ipc.opt); 1360 return err; 1361 } 1362 if (ipc.opt) 1363 free = 1; 1364 } 1365 if (!ipc.opt) { 1366 struct ip_options_rcu *inet_opt; 1367 1368 rcu_read_lock(); 1369 inet_opt = rcu_dereference(inet->inet_opt); 1370 if (inet_opt) { 1371 memcpy(&opt_copy, inet_opt, 1372 sizeof(*inet_opt) + inet_opt->opt.optlen); 1373 ipc.opt = &opt_copy.opt; 1374 } 1375 rcu_read_unlock(); 1376 } 1377 1378 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1379 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1380 (struct sockaddr *)usin, 1381 &msg->msg_namelen, 1382 &ipc.addr); 1383 if (err) 1384 goto out_free; 1385 if (usin) { 1386 if (usin->sin_port == 0) { 1387 /* BPF program set invalid port. Reject it. */ 1388 err = -EINVAL; 1389 goto out_free; 1390 } 1391 daddr = usin->sin_addr.s_addr; 1392 dport = usin->sin_port; 1393 } 1394 } 1395 1396 saddr = ipc.addr; 1397 ipc.addr = faddr = daddr; 1398 1399 if (ipc.opt && ipc.opt->opt.srr) { 1400 if (!daddr) { 1401 err = -EINVAL; 1402 goto out_free; 1403 } 1404 faddr = ipc.opt->opt.faddr; 1405 connected = 0; 1406 } 1407 scope = ip_sendmsg_scope(inet, &ipc, msg); 1408 if (scope == RT_SCOPE_LINK) 1409 connected = 0; 1410 1411 uc_index = READ_ONCE(inet->uc_index); 1412 if (ipv4_is_multicast(daddr)) { 1413 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1414 ipc.oif = READ_ONCE(inet->mc_index); 1415 if (!saddr) 1416 saddr = READ_ONCE(inet->mc_addr); 1417 connected = 0; 1418 } else if (!ipc.oif) { 1419 ipc.oif = uc_index; 1420 } else if (ipv4_is_lbcast(daddr) && uc_index) { 1421 /* oif is set, packet is to local broadcast and 1422 * uc_index is set. oif is most likely set 1423 * by sk_bound_dev_if. If uc_index != oif check if the 1424 * oif is an L3 master and uc_index is an L3 slave. 1425 * If so, we want to allow the send using the uc_index. 1426 */ 1427 if (ipc.oif != uc_index && 1428 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1429 uc_index)) { 1430 ipc.oif = uc_index; 1431 } 1432 } 1433 1434 if (connected) 1435 rt = dst_rtable(sk_dst_check(sk, 0)); 1436 1437 if (!rt) { 1438 struct net *net = sock_net(sk); 1439 __u8 flow_flags = inet_sk_flowi_flags(sk); 1440 1441 fl4 = &fl4_stack; 1442 1443 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, 1444 ipc.tos & INET_DSCP_MASK, scope, 1445 sk->sk_protocol, flow_flags, faddr, saddr, 1446 dport, inet->inet_sport, 1447 sk_uid(sk)); 1448 1449 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1450 rt = ip_route_output_flow(net, fl4, sk); 1451 if (IS_ERR(rt)) { 1452 err = PTR_ERR(rt); 1453 rt = NULL; 1454 if (err == -ENETUNREACH) 1455 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1456 goto out; 1457 } 1458 1459 err = -EACCES; 1460 if ((rt->rt_flags & RTCF_BROADCAST) && 1461 !sock_flag(sk, SOCK_BROADCAST)) 1462 goto out; 1463 if (connected) 1464 sk_dst_set(sk, dst_clone(&rt->dst)); 1465 } 1466 1467 if (msg->msg_flags&MSG_CONFIRM) 1468 goto do_confirm; 1469 back_from_confirm: 1470 1471 saddr = fl4->saddr; 1472 if (!ipc.addr) 1473 daddr = ipc.addr = fl4->daddr; 1474 1475 /* Lockless fast path for the non-corking case. */ 1476 if (!corkreq) { 1477 struct inet_cork cork; 1478 1479 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1480 sizeof(struct udphdr), &ipc, &rt, 1481 &cork, msg->msg_flags); 1482 err = PTR_ERR(skb); 1483 if (!IS_ERR_OR_NULL(skb)) 1484 err = udp_send_skb(skb, fl4, &cork); 1485 goto out; 1486 } 1487 1488 lock_sock(sk); 1489 if (unlikely(up->pending)) { 1490 /* The socket is already corked while preparing it. */ 1491 /* ... which is an evident application bug. --ANK */ 1492 release_sock(sk); 1493 1494 net_dbg_ratelimited("socket already corked\n"); 1495 err = -EINVAL; 1496 goto out; 1497 } 1498 /* 1499 * Now cork the socket to pend data. 1500 */ 1501 fl4 = &inet->cork.fl.u.ip4; 1502 fl4->daddr = daddr; 1503 fl4->saddr = saddr; 1504 fl4->fl4_dport = dport; 1505 fl4->fl4_sport = inet->inet_sport; 1506 WRITE_ONCE(up->pending, AF_INET); 1507 1508 do_append_data: 1509 up->len += ulen; 1510 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1511 sizeof(struct udphdr), &ipc, &rt, 1512 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1513 if (err) 1514 udp_flush_pending_frames(sk); 1515 else if (!corkreq) 1516 err = udp_push_pending_frames(sk); 1517 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1518 WRITE_ONCE(up->pending, 0); 1519 release_sock(sk); 1520 1521 out: 1522 ip_rt_put(rt); 1523 out_free: 1524 if (free) 1525 kfree(ipc.opt); 1526 if (!err) 1527 return len; 1528 /* 1529 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1530 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1531 * we don't have a good statistic (IpOutDiscards but it can be too many 1532 * things). We could add another new stat but at least for now that 1533 * seems like overkill. 1534 */ 1535 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1536 UDP_INC_STATS(sock_net(sk), 1537 UDP_MIB_SNDBUFERRORS, is_udplite); 1538 } 1539 return err; 1540 1541 do_confirm: 1542 if (msg->msg_flags & MSG_PROBE) 1543 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1544 if (!(msg->msg_flags&MSG_PROBE) || len) 1545 goto back_from_confirm; 1546 err = 0; 1547 goto out; 1548 } 1549 EXPORT_SYMBOL(udp_sendmsg); 1550 1551 void udp_splice_eof(struct socket *sock) 1552 { 1553 struct sock *sk = sock->sk; 1554 struct udp_sock *up = udp_sk(sk); 1555 1556 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk)) 1557 return; 1558 1559 lock_sock(sk); 1560 if (up->pending && !udp_test_bit(CORK, sk)) 1561 udp_push_pending_frames(sk); 1562 release_sock(sk); 1563 } 1564 EXPORT_IPV6_MOD_GPL(udp_splice_eof); 1565 1566 #define UDP_SKB_IS_STATELESS 0x80000000 1567 1568 /* all head states (dst, sk, nf conntrack) except skb extensions are 1569 * cleared by udp_rcv(). 1570 * 1571 * We need to preserve secpath, if present, to eventually process 1572 * IP_CMSG_PASSSEC at recvmsg() time. 1573 * 1574 * Other extensions can be cleared. 1575 */ 1576 static bool udp_try_make_stateless(struct sk_buff *skb) 1577 { 1578 if (!skb_has_extensions(skb)) 1579 return true; 1580 1581 if (!secpath_exists(skb)) { 1582 skb_ext_reset(skb); 1583 return true; 1584 } 1585 1586 return false; 1587 } 1588 1589 static void udp_set_dev_scratch(struct sk_buff *skb) 1590 { 1591 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1592 1593 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1594 scratch->_tsize_state = skb->truesize; 1595 #if BITS_PER_LONG == 64 1596 scratch->len = skb->len; 1597 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1598 scratch->is_linear = !skb_is_nonlinear(skb); 1599 #endif 1600 if (udp_try_make_stateless(skb)) 1601 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1602 } 1603 1604 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1605 { 1606 /* We come here after udp_lib_checksum_complete() returned 0. 1607 * This means that __skb_checksum_complete() might have 1608 * set skb->csum_valid to 1. 1609 * On 64bit platforms, we can set csum_unnecessary 1610 * to true, but only if the skb is not shared. 1611 */ 1612 #if BITS_PER_LONG == 64 1613 if (!skb_shared(skb)) 1614 udp_skb_scratch(skb)->csum_unnecessary = true; 1615 #endif 1616 } 1617 1618 static int udp_skb_truesize(struct sk_buff *skb) 1619 { 1620 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1621 } 1622 1623 static bool udp_skb_has_head_state(struct sk_buff *skb) 1624 { 1625 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1626 } 1627 1628 /* fully reclaim rmem/fwd memory allocated for skb */ 1629 static void udp_rmem_release(struct sock *sk, unsigned int size, 1630 int partial, bool rx_queue_lock_held) 1631 { 1632 struct udp_sock *up = udp_sk(sk); 1633 struct sk_buff_head *sk_queue; 1634 unsigned int amt; 1635 1636 if (likely(partial)) { 1637 up->forward_deficit += size; 1638 size = up->forward_deficit; 1639 if (size < READ_ONCE(up->forward_threshold) && 1640 !skb_queue_empty(&up->reader_queue)) 1641 return; 1642 } else { 1643 size += up->forward_deficit; 1644 } 1645 up->forward_deficit = 0; 1646 1647 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1648 * if the called don't held it already 1649 */ 1650 sk_queue = &sk->sk_receive_queue; 1651 if (!rx_queue_lock_held) 1652 spin_lock(&sk_queue->lock); 1653 1654 amt = (size + sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1655 sk_forward_alloc_add(sk, size - amt); 1656 1657 if (amt) 1658 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 1659 1660 atomic_sub(size, &sk->sk_rmem_alloc); 1661 1662 /* this can save us from acquiring the rx queue lock on next receive */ 1663 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1664 1665 if (!rx_queue_lock_held) 1666 spin_unlock(&sk_queue->lock); 1667 } 1668 1669 /* Note: called with reader_queue.lock held. 1670 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1671 * This avoids a cache line miss while receive_queue lock is held. 1672 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1673 */ 1674 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1675 { 1676 prefetch(&skb->data); 1677 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1678 } 1679 EXPORT_IPV6_MOD(udp_skb_destructor); 1680 1681 /* as above, but the caller held the rx queue lock, too */ 1682 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1683 { 1684 prefetch(&skb->data); 1685 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1686 } 1687 1688 /* Idea of busylocks is to let producers grab an extra spinlock 1689 * to relieve pressure on the receive_queue spinlock shared by consumer. 1690 * Under flood, this means that only one producer can be in line 1691 * trying to acquire the receive_queue spinlock. 1692 * These busylock can be allocated on a per cpu manner, instead of a 1693 * per socket one (that would consume a cache line per socket) 1694 */ 1695 static int udp_busylocks_log __read_mostly; 1696 static spinlock_t *udp_busylocks __read_mostly; 1697 1698 static spinlock_t *busylock_acquire(void *ptr) 1699 { 1700 spinlock_t *busy; 1701 1702 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1703 spin_lock(busy); 1704 return busy; 1705 } 1706 1707 static void busylock_release(spinlock_t *busy) 1708 { 1709 if (busy) 1710 spin_unlock(busy); 1711 } 1712 1713 static int udp_rmem_schedule(struct sock *sk, int size) 1714 { 1715 int delta; 1716 1717 delta = size - sk->sk_forward_alloc; 1718 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV)) 1719 return -ENOBUFS; 1720 1721 return 0; 1722 } 1723 1724 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1725 { 1726 struct sk_buff_head *list = &sk->sk_receive_queue; 1727 unsigned int rmem, rcvbuf; 1728 spinlock_t *busy = NULL; 1729 int size, err = -ENOMEM; 1730 1731 rmem = atomic_read(&sk->sk_rmem_alloc); 1732 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1733 size = skb->truesize; 1734 1735 /* Immediately drop when the receive queue is full. 1736 * Cast to unsigned int performs the boundary check for INT_MAX. 1737 */ 1738 if (rmem + size > rcvbuf) { 1739 if (rcvbuf > INT_MAX >> 1) 1740 goto drop; 1741 1742 /* Always allow at least one packet for small buffer. */ 1743 if (rmem > rcvbuf) 1744 goto drop; 1745 } 1746 1747 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1748 * having linear skbs : 1749 * - Reduce memory overhead and thus increase receive queue capacity 1750 * - Less cache line misses at copyout() time 1751 * - Less work at consume_skb() (less alien page frag freeing) 1752 */ 1753 if (rmem > (rcvbuf >> 1)) { 1754 skb_condense(skb); 1755 size = skb->truesize; 1756 busy = busylock_acquire(sk); 1757 } 1758 1759 udp_set_dev_scratch(skb); 1760 1761 atomic_add(size, &sk->sk_rmem_alloc); 1762 1763 spin_lock(&list->lock); 1764 err = udp_rmem_schedule(sk, size); 1765 if (err) { 1766 spin_unlock(&list->lock); 1767 goto uncharge_drop; 1768 } 1769 1770 sk_forward_alloc_add(sk, -size); 1771 1772 /* no need to setup a destructor, we will explicitly release the 1773 * forward allocated memory on dequeue 1774 */ 1775 sock_skb_set_dropcount(sk, skb); 1776 1777 __skb_queue_tail(list, skb); 1778 spin_unlock(&list->lock); 1779 1780 if (!sock_flag(sk, SOCK_DEAD)) 1781 INDIRECT_CALL_1(sk->sk_data_ready, sock_def_readable, sk); 1782 1783 busylock_release(busy); 1784 return 0; 1785 1786 uncharge_drop: 1787 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1788 1789 drop: 1790 atomic_inc(&sk->sk_drops); 1791 busylock_release(busy); 1792 return err; 1793 } 1794 EXPORT_IPV6_MOD_GPL(__udp_enqueue_schedule_skb); 1795 1796 void udp_destruct_common(struct sock *sk) 1797 { 1798 /* reclaim completely the forward allocated memory */ 1799 struct udp_sock *up = udp_sk(sk); 1800 unsigned int total = 0; 1801 struct sk_buff *skb; 1802 1803 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1804 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1805 total += skb->truesize; 1806 kfree_skb(skb); 1807 } 1808 udp_rmem_release(sk, total, 0, true); 1809 } 1810 EXPORT_IPV6_MOD_GPL(udp_destruct_common); 1811 1812 static void udp_destruct_sock(struct sock *sk) 1813 { 1814 udp_destruct_common(sk); 1815 inet_sock_destruct(sk); 1816 } 1817 1818 int udp_init_sock(struct sock *sk) 1819 { 1820 udp_lib_init_sock(sk); 1821 sk->sk_destruct = udp_destruct_sock; 1822 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1823 return 0; 1824 } 1825 1826 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1827 { 1828 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset))) 1829 sk_peek_offset_bwd(sk, len); 1830 1831 if (!skb_unref(skb)) 1832 return; 1833 1834 /* In the more common cases we cleared the head states previously, 1835 * see __udp_queue_rcv_skb(). 1836 */ 1837 if (unlikely(udp_skb_has_head_state(skb))) 1838 skb_release_head_state(skb); 1839 __consume_stateless_skb(skb); 1840 } 1841 EXPORT_IPV6_MOD_GPL(skb_consume_udp); 1842 1843 static struct sk_buff *__first_packet_length(struct sock *sk, 1844 struct sk_buff_head *rcvq, 1845 unsigned int *total) 1846 { 1847 struct sk_buff *skb; 1848 1849 while ((skb = skb_peek(rcvq)) != NULL) { 1850 if (udp_lib_checksum_complete(skb)) { 1851 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1852 IS_UDPLITE(sk)); 1853 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1854 IS_UDPLITE(sk)); 1855 atomic_inc(&sk->sk_drops); 1856 __skb_unlink(skb, rcvq); 1857 *total += skb->truesize; 1858 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM); 1859 } else { 1860 udp_skb_csum_unnecessary_set(skb); 1861 break; 1862 } 1863 } 1864 return skb; 1865 } 1866 1867 /** 1868 * first_packet_length - return length of first packet in receive queue 1869 * @sk: socket 1870 * 1871 * Drops all bad checksum frames, until a valid one is found. 1872 * Returns the length of found skb, or -1 if none is found. 1873 */ 1874 static int first_packet_length(struct sock *sk) 1875 { 1876 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1877 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1878 unsigned int total = 0; 1879 struct sk_buff *skb; 1880 int res; 1881 1882 spin_lock_bh(&rcvq->lock); 1883 skb = __first_packet_length(sk, rcvq, &total); 1884 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1885 spin_lock(&sk_queue->lock); 1886 skb_queue_splice_tail_init(sk_queue, rcvq); 1887 spin_unlock(&sk_queue->lock); 1888 1889 skb = __first_packet_length(sk, rcvq, &total); 1890 } 1891 res = skb ? skb->len : -1; 1892 if (total) 1893 udp_rmem_release(sk, total, 1, false); 1894 spin_unlock_bh(&rcvq->lock); 1895 return res; 1896 } 1897 1898 /* 1899 * IOCTL requests applicable to the UDP protocol 1900 */ 1901 1902 int udp_ioctl(struct sock *sk, int cmd, int *karg) 1903 { 1904 switch (cmd) { 1905 case SIOCOUTQ: 1906 { 1907 *karg = sk_wmem_alloc_get(sk); 1908 return 0; 1909 } 1910 1911 case SIOCINQ: 1912 { 1913 *karg = max_t(int, 0, first_packet_length(sk)); 1914 return 0; 1915 } 1916 1917 default: 1918 return -ENOIOCTLCMD; 1919 } 1920 1921 return 0; 1922 } 1923 EXPORT_IPV6_MOD(udp_ioctl); 1924 1925 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1926 int *off, int *err) 1927 { 1928 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1929 struct sk_buff_head *queue; 1930 struct sk_buff *last; 1931 long timeo; 1932 int error; 1933 1934 queue = &udp_sk(sk)->reader_queue; 1935 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1936 do { 1937 struct sk_buff *skb; 1938 1939 error = sock_error(sk); 1940 if (error) 1941 break; 1942 1943 error = -EAGAIN; 1944 do { 1945 spin_lock_bh(&queue->lock); 1946 skb = __skb_try_recv_from_queue(queue, flags, off, err, 1947 &last); 1948 if (skb) { 1949 if (!(flags & MSG_PEEK)) 1950 udp_skb_destructor(sk, skb); 1951 spin_unlock_bh(&queue->lock); 1952 return skb; 1953 } 1954 1955 if (skb_queue_empty_lockless(sk_queue)) { 1956 spin_unlock_bh(&queue->lock); 1957 goto busy_check; 1958 } 1959 1960 /* refill the reader queue and walk it again 1961 * keep both queues locked to avoid re-acquiring 1962 * the sk_receive_queue lock if fwd memory scheduling 1963 * is needed. 1964 */ 1965 spin_lock(&sk_queue->lock); 1966 skb_queue_splice_tail_init(sk_queue, queue); 1967 1968 skb = __skb_try_recv_from_queue(queue, flags, off, err, 1969 &last); 1970 if (skb && !(flags & MSG_PEEK)) 1971 udp_skb_dtor_locked(sk, skb); 1972 spin_unlock(&sk_queue->lock); 1973 spin_unlock_bh(&queue->lock); 1974 if (skb) 1975 return skb; 1976 1977 busy_check: 1978 if (!sk_can_busy_loop(sk)) 1979 break; 1980 1981 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1982 } while (!skb_queue_empty_lockless(sk_queue)); 1983 1984 /* sk_queue is empty, reader_queue may contain peeked packets */ 1985 } while (timeo && 1986 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1987 &error, &timeo, 1988 (struct sk_buff *)sk_queue)); 1989 1990 *err = error; 1991 return NULL; 1992 } 1993 EXPORT_SYMBOL(__skb_recv_udp); 1994 1995 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) 1996 { 1997 struct sk_buff *skb; 1998 int err; 1999 2000 try_again: 2001 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 2002 if (!skb) 2003 return err; 2004 2005 if (udp_lib_checksum_complete(skb)) { 2006 int is_udplite = IS_UDPLITE(sk); 2007 struct net *net = sock_net(sk); 2008 2009 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite); 2010 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite); 2011 atomic_inc(&sk->sk_drops); 2012 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM); 2013 goto try_again; 2014 } 2015 2016 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); 2017 return recv_actor(sk, skb); 2018 } 2019 EXPORT_IPV6_MOD(udp_read_skb); 2020 2021 /* 2022 * This should be easy, if there is something there we 2023 * return it, otherwise we block. 2024 */ 2025 2026 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 2027 int *addr_len) 2028 { 2029 struct inet_sock *inet = inet_sk(sk); 2030 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 2031 struct sk_buff *skb; 2032 unsigned int ulen, copied; 2033 int off, err, peeking = flags & MSG_PEEK; 2034 int is_udplite = IS_UDPLITE(sk); 2035 bool checksum_valid = false; 2036 2037 if (flags & MSG_ERRQUEUE) 2038 return ip_recv_error(sk, msg, len, addr_len); 2039 2040 try_again: 2041 off = sk_peek_offset(sk, flags); 2042 skb = __skb_recv_udp(sk, flags, &off, &err); 2043 if (!skb) 2044 return err; 2045 2046 ulen = udp_skb_len(skb); 2047 copied = len; 2048 if (copied > ulen - off) 2049 copied = ulen - off; 2050 else if (copied < ulen) 2051 msg->msg_flags |= MSG_TRUNC; 2052 2053 /* 2054 * If checksum is needed at all, try to do it while copying the 2055 * data. If the data is truncated, or if we only want a partial 2056 * coverage checksum (UDP-Lite), do it before the copy. 2057 */ 2058 2059 if (copied < ulen || peeking || 2060 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 2061 checksum_valid = udp_skb_csum_unnecessary(skb) || 2062 !__udp_lib_checksum_complete(skb); 2063 if (!checksum_valid) 2064 goto csum_copy_err; 2065 } 2066 2067 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 2068 if (udp_skb_is_linear(skb)) 2069 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 2070 else 2071 err = skb_copy_datagram_msg(skb, off, msg, copied); 2072 } else { 2073 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 2074 2075 if (err == -EINVAL) 2076 goto csum_copy_err; 2077 } 2078 2079 if (unlikely(err)) { 2080 if (!peeking) { 2081 atomic_inc(&sk->sk_drops); 2082 UDP_INC_STATS(sock_net(sk), 2083 UDP_MIB_INERRORS, is_udplite); 2084 } 2085 kfree_skb(skb); 2086 return err; 2087 } 2088 2089 if (!peeking) 2090 UDP_INC_STATS(sock_net(sk), 2091 UDP_MIB_INDATAGRAMS, is_udplite); 2092 2093 sock_recv_cmsgs(msg, sk, skb); 2094 2095 /* Copy the address. */ 2096 if (sin) { 2097 sin->sin_family = AF_INET; 2098 sin->sin_port = udp_hdr(skb)->source; 2099 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 2100 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 2101 *addr_len = sizeof(*sin); 2102 2103 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 2104 (struct sockaddr *)sin, 2105 addr_len); 2106 } 2107 2108 if (udp_test_bit(GRO_ENABLED, sk)) 2109 udp_cmsg_recv(msg, sk, skb); 2110 2111 if (inet_cmsg_flags(inet)) 2112 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 2113 2114 err = copied; 2115 if (flags & MSG_TRUNC) 2116 err = ulen; 2117 2118 skb_consume_udp(sk, skb, peeking ? -err : err); 2119 return err; 2120 2121 csum_copy_err: 2122 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 2123 udp_skb_destructor)) { 2124 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2125 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2126 } 2127 kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM); 2128 2129 /* starting over for a new packet, but check if we need to yield */ 2130 cond_resched(); 2131 msg->msg_flags &= ~MSG_TRUNC; 2132 goto try_again; 2133 } 2134 2135 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 2136 { 2137 /* This check is replicated from __ip4_datagram_connect() and 2138 * intended to prevent BPF program called below from accessing bytes 2139 * that are out of the bound specified by user in addr_len. 2140 */ 2141 if (addr_len < sizeof(struct sockaddr_in)) 2142 return -EINVAL; 2143 2144 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len); 2145 } 2146 EXPORT_IPV6_MOD(udp_pre_connect); 2147 2148 static int udp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 2149 { 2150 int res; 2151 2152 lock_sock(sk); 2153 res = __ip4_datagram_connect(sk, uaddr, addr_len); 2154 if (!res) 2155 udp4_hash4(sk); 2156 release_sock(sk); 2157 return res; 2158 } 2159 2160 int __udp_disconnect(struct sock *sk, int flags) 2161 { 2162 struct inet_sock *inet = inet_sk(sk); 2163 /* 2164 * 1003.1g - break association. 2165 */ 2166 2167 sk->sk_state = TCP_CLOSE; 2168 inet->inet_daddr = 0; 2169 inet->inet_dport = 0; 2170 sock_rps_reset_rxhash(sk); 2171 sk->sk_bound_dev_if = 0; 2172 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 2173 inet_reset_saddr(sk); 2174 if (sk->sk_prot->rehash && 2175 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 2176 sk->sk_prot->rehash(sk); 2177 } 2178 2179 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 2180 sk->sk_prot->unhash(sk); 2181 inet->inet_sport = 0; 2182 } 2183 sk_dst_reset(sk); 2184 return 0; 2185 } 2186 EXPORT_SYMBOL(__udp_disconnect); 2187 2188 int udp_disconnect(struct sock *sk, int flags) 2189 { 2190 lock_sock(sk); 2191 __udp_disconnect(sk, flags); 2192 release_sock(sk); 2193 return 0; 2194 } 2195 EXPORT_IPV6_MOD(udp_disconnect); 2196 2197 void udp_lib_unhash(struct sock *sk) 2198 { 2199 if (sk_hashed(sk)) { 2200 struct udp_table *udptable = udp_get_table_prot(sk); 2201 struct udp_hslot *hslot, *hslot2; 2202 2203 sock_rps_delete_flow(sk); 2204 hslot = udp_hashslot(udptable, sock_net(sk), 2205 udp_sk(sk)->udp_port_hash); 2206 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2207 2208 spin_lock_bh(&hslot->lock); 2209 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2210 reuseport_detach_sock(sk); 2211 if (sk_del_node_init_rcu(sk)) { 2212 hslot->count--; 2213 inet_sk(sk)->inet_num = 0; 2214 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 2215 2216 spin_lock(&hslot2->lock); 2217 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2218 hslot2->count--; 2219 spin_unlock(&hslot2->lock); 2220 2221 udp_unhash4(udptable, sk); 2222 } 2223 spin_unlock_bh(&hslot->lock); 2224 } 2225 } 2226 EXPORT_IPV6_MOD(udp_lib_unhash); 2227 2228 /* 2229 * inet_rcv_saddr was changed, we must rehash secondary hash 2230 */ 2231 void udp_lib_rehash(struct sock *sk, u16 newhash, u16 newhash4) 2232 { 2233 if (sk_hashed(sk)) { 2234 struct udp_table *udptable = udp_get_table_prot(sk); 2235 struct udp_hslot *hslot, *hslot2, *nhslot2; 2236 2237 hslot = udp_hashslot(udptable, sock_net(sk), 2238 udp_sk(sk)->udp_port_hash); 2239 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2240 nhslot2 = udp_hashslot2(udptable, newhash); 2241 udp_sk(sk)->udp_portaddr_hash = newhash; 2242 2243 if (hslot2 != nhslot2 || 2244 rcu_access_pointer(sk->sk_reuseport_cb)) { 2245 /* we must lock primary chain too */ 2246 spin_lock_bh(&hslot->lock); 2247 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2248 reuseport_detach_sock(sk); 2249 2250 if (hslot2 != nhslot2) { 2251 spin_lock(&hslot2->lock); 2252 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2253 hslot2->count--; 2254 spin_unlock(&hslot2->lock); 2255 2256 spin_lock(&nhslot2->lock); 2257 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2258 &nhslot2->head); 2259 nhslot2->count++; 2260 spin_unlock(&nhslot2->lock); 2261 } 2262 2263 spin_unlock_bh(&hslot->lock); 2264 } 2265 2266 /* Now process hash4 if necessary: 2267 * (1) update hslot4; 2268 * (2) update hslot2->hash4_cnt. 2269 * Note that hslot2/hslot4 should be checked separately, as 2270 * either of them may change with the other unchanged. 2271 */ 2272 if (udp_hashed4(sk)) { 2273 spin_lock_bh(&hslot->lock); 2274 2275 udp_rehash4(udptable, sk, newhash4); 2276 if (hslot2 != nhslot2) { 2277 spin_lock(&hslot2->lock); 2278 udp_hash4_dec(hslot2); 2279 spin_unlock(&hslot2->lock); 2280 2281 spin_lock(&nhslot2->lock); 2282 udp_hash4_inc(nhslot2); 2283 spin_unlock(&nhslot2->lock); 2284 } 2285 2286 spin_unlock_bh(&hslot->lock); 2287 } 2288 } 2289 } 2290 EXPORT_IPV6_MOD(udp_lib_rehash); 2291 2292 void udp_v4_rehash(struct sock *sk) 2293 { 2294 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2295 inet_sk(sk)->inet_rcv_saddr, 2296 inet_sk(sk)->inet_num); 2297 u16 new_hash4 = udp_ehashfn(sock_net(sk), 2298 sk->sk_rcv_saddr, sk->sk_num, 2299 sk->sk_daddr, sk->sk_dport); 2300 2301 udp_lib_rehash(sk, new_hash, new_hash4); 2302 } 2303 2304 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2305 { 2306 int rc; 2307 2308 if (inet_sk(sk)->inet_daddr) { 2309 sock_rps_save_rxhash(sk, skb); 2310 sk_mark_napi_id(sk, skb); 2311 sk_incoming_cpu_update(sk); 2312 } else { 2313 sk_mark_napi_id_once(sk, skb); 2314 } 2315 2316 rc = __udp_enqueue_schedule_skb(sk, skb); 2317 if (rc < 0) { 2318 int is_udplite = IS_UDPLITE(sk); 2319 int drop_reason; 2320 2321 /* Note that an ENOMEM error is charged twice */ 2322 if (rc == -ENOMEM) { 2323 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2324 is_udplite); 2325 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 2326 } else { 2327 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2328 is_udplite); 2329 drop_reason = SKB_DROP_REASON_PROTO_MEM; 2330 } 2331 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2332 trace_udp_fail_queue_rcv_skb(rc, sk, skb); 2333 sk_skb_reason_drop(sk, skb, drop_reason); 2334 return -1; 2335 } 2336 2337 return 0; 2338 } 2339 2340 /* returns: 2341 * -1: error 2342 * 0: success 2343 * >0: "udp encap" protocol resubmission 2344 * 2345 * Note that in the success and error cases, the skb is assumed to 2346 * have either been requeued or freed. 2347 */ 2348 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2349 { 2350 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2351 struct udp_sock *up = udp_sk(sk); 2352 int is_udplite = IS_UDPLITE(sk); 2353 2354 /* 2355 * Charge it to the socket, dropping if the queue is full. 2356 */ 2357 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 2358 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2359 goto drop; 2360 } 2361 nf_reset_ct(skb); 2362 2363 if (static_branch_unlikely(&udp_encap_needed_key) && 2364 READ_ONCE(up->encap_type)) { 2365 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2366 2367 /* 2368 * This is an encapsulation socket so pass the skb to 2369 * the socket's udp_encap_rcv() hook. Otherwise, just 2370 * fall through and pass this up the UDP socket. 2371 * up->encap_rcv() returns the following value: 2372 * =0 if skb was successfully passed to the encap 2373 * handler or was discarded by it. 2374 * >0 if skb should be passed on to UDP. 2375 * <0 if skb should be resubmitted as proto -N 2376 */ 2377 2378 /* if we're overly short, let UDP handle it */ 2379 encap_rcv = READ_ONCE(up->encap_rcv); 2380 if (encap_rcv) { 2381 int ret; 2382 2383 /* Verify checksum before giving to encap */ 2384 if (udp_lib_checksum_complete(skb)) 2385 goto csum_error; 2386 2387 ret = encap_rcv(sk, skb); 2388 if (ret <= 0) { 2389 __UDP_INC_STATS(sock_net(sk), 2390 UDP_MIB_INDATAGRAMS, 2391 is_udplite); 2392 return -ret; 2393 } 2394 } 2395 2396 /* FALLTHROUGH -- it's a UDP Packet */ 2397 } 2398 2399 /* 2400 * UDP-Lite specific tests, ignored on UDP sockets 2401 */ 2402 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) { 2403 u16 pcrlen = READ_ONCE(up->pcrlen); 2404 2405 /* 2406 * MIB statistics other than incrementing the error count are 2407 * disabled for the following two types of errors: these depend 2408 * on the application settings, not on the functioning of the 2409 * protocol stack as such. 2410 * 2411 * RFC 3828 here recommends (sec 3.3): "There should also be a 2412 * way ... to ... at least let the receiving application block 2413 * delivery of packets with coverage values less than a value 2414 * provided by the application." 2415 */ 2416 if (pcrlen == 0) { /* full coverage was set */ 2417 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2418 UDP_SKB_CB(skb)->cscov, skb->len); 2419 goto drop; 2420 } 2421 /* The next case involves violating the min. coverage requested 2422 * by the receiver. This is subtle: if receiver wants x and x is 2423 * greater than the buffersize/MTU then receiver will complain 2424 * that it wants x while sender emits packets of smaller size y. 2425 * Therefore the above ...()->partial_cov statement is essential. 2426 */ 2427 if (UDP_SKB_CB(skb)->cscov < pcrlen) { 2428 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2429 UDP_SKB_CB(skb)->cscov, pcrlen); 2430 goto drop; 2431 } 2432 } 2433 2434 prefetch(&sk->sk_rmem_alloc); 2435 if (rcu_access_pointer(sk->sk_filter) && 2436 udp_lib_checksum_complete(skb)) 2437 goto csum_error; 2438 2439 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr), &drop_reason)) 2440 goto drop; 2441 2442 udp_csum_pull_header(skb); 2443 2444 ipv4_pktinfo_prepare(sk, skb, true); 2445 return __udp_queue_rcv_skb(sk, skb); 2446 2447 csum_error: 2448 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2449 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2450 drop: 2451 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2452 atomic_inc(&sk->sk_drops); 2453 sk_skb_reason_drop(sk, skb, drop_reason); 2454 return -1; 2455 } 2456 2457 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2458 { 2459 struct sk_buff *next, *segs; 2460 int ret; 2461 2462 if (likely(!udp_unexpected_gso(sk, skb))) 2463 return udp_queue_rcv_one_skb(sk, skb); 2464 2465 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2466 __skb_push(skb, -skb_mac_offset(skb)); 2467 segs = udp_rcv_segment(sk, skb, true); 2468 skb_list_walk_safe(segs, skb, next) { 2469 __skb_pull(skb, skb_transport_offset(skb)); 2470 2471 udp_post_segment_fix_csum(skb); 2472 ret = udp_queue_rcv_one_skb(sk, skb); 2473 if (ret > 0) 2474 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2475 } 2476 return 0; 2477 } 2478 2479 /* For TCP sockets, sk_rx_dst is protected by socket lock 2480 * For UDP, we use xchg() to guard against concurrent changes. 2481 */ 2482 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2483 { 2484 struct dst_entry *old; 2485 2486 if (dst_hold_safe(dst)) { 2487 old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst))); 2488 dst_release(old); 2489 return old != dst; 2490 } 2491 return false; 2492 } 2493 EXPORT_IPV6_MOD(udp_sk_rx_dst_set); 2494 2495 /* 2496 * Multicasts and broadcasts go to each listener. 2497 * 2498 * Note: called only from the BH handler context. 2499 */ 2500 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2501 struct udphdr *uh, 2502 __be32 saddr, __be32 daddr, 2503 struct udp_table *udptable, 2504 int proto) 2505 { 2506 struct sock *sk, *first = NULL; 2507 unsigned short hnum = ntohs(uh->dest); 2508 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2509 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2510 unsigned int offset = offsetof(typeof(*sk), sk_node); 2511 int dif = skb->dev->ifindex; 2512 int sdif = inet_sdif(skb); 2513 struct hlist_node *node; 2514 struct sk_buff *nskb; 2515 2516 if (use_hash2) { 2517 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2518 udptable->mask; 2519 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2520 start_lookup: 2521 hslot = &udptable->hash2[hash2].hslot; 2522 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2523 } 2524 2525 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2526 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2527 uh->source, saddr, dif, sdif, hnum)) 2528 continue; 2529 2530 if (!first) { 2531 first = sk; 2532 continue; 2533 } 2534 nskb = skb_clone(skb, GFP_ATOMIC); 2535 2536 if (unlikely(!nskb)) { 2537 atomic_inc(&sk->sk_drops); 2538 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2539 IS_UDPLITE(sk)); 2540 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2541 IS_UDPLITE(sk)); 2542 continue; 2543 } 2544 if (udp_queue_rcv_skb(sk, nskb) > 0) 2545 consume_skb(nskb); 2546 } 2547 2548 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2549 if (use_hash2 && hash2 != hash2_any) { 2550 hash2 = hash2_any; 2551 goto start_lookup; 2552 } 2553 2554 if (first) { 2555 if (udp_queue_rcv_skb(first, skb) > 0) 2556 consume_skb(skb); 2557 } else { 2558 kfree_skb(skb); 2559 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2560 proto == IPPROTO_UDPLITE); 2561 } 2562 return 0; 2563 } 2564 2565 /* Initialize UDP checksum. If exited with zero value (success), 2566 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2567 * Otherwise, csum completion requires checksumming packet body, 2568 * including udp header and folding it to skb->csum. 2569 */ 2570 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2571 int proto) 2572 { 2573 int err; 2574 2575 UDP_SKB_CB(skb)->partial_cov = 0; 2576 UDP_SKB_CB(skb)->cscov = skb->len; 2577 2578 if (proto == IPPROTO_UDPLITE) { 2579 err = udplite_checksum_init(skb, uh); 2580 if (err) 2581 return err; 2582 2583 if (UDP_SKB_CB(skb)->partial_cov) { 2584 skb->csum = inet_compute_pseudo(skb, proto); 2585 return 0; 2586 } 2587 } 2588 2589 /* Note, we are only interested in != 0 or == 0, thus the 2590 * force to int. 2591 */ 2592 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2593 inet_compute_pseudo); 2594 if (err) 2595 return err; 2596 2597 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2598 /* If SW calculated the value, we know it's bad */ 2599 if (skb->csum_complete_sw) 2600 return 1; 2601 2602 /* HW says the value is bad. Let's validate that. 2603 * skb->csum is no longer the full packet checksum, 2604 * so don't treat it as such. 2605 */ 2606 skb_checksum_complete_unset(skb); 2607 } 2608 2609 return 0; 2610 } 2611 2612 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2613 * return code conversion for ip layer consumption 2614 */ 2615 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2616 struct udphdr *uh) 2617 { 2618 int ret; 2619 2620 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2621 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2622 2623 ret = udp_queue_rcv_skb(sk, skb); 2624 2625 /* a return value > 0 means to resubmit the input, but 2626 * it wants the return to be -protocol, or 0 2627 */ 2628 if (ret > 0) 2629 return -ret; 2630 return 0; 2631 } 2632 2633 /* 2634 * All we need to do is get the socket, and then do a checksum. 2635 */ 2636 2637 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2638 int proto) 2639 { 2640 struct sock *sk = NULL; 2641 struct udphdr *uh; 2642 unsigned short ulen; 2643 struct rtable *rt = skb_rtable(skb); 2644 __be32 saddr, daddr; 2645 struct net *net = dev_net(skb->dev); 2646 bool refcounted; 2647 int drop_reason; 2648 2649 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2650 2651 /* 2652 * Validate the packet. 2653 */ 2654 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2655 goto drop; /* No space for header. */ 2656 2657 uh = udp_hdr(skb); 2658 ulen = ntohs(uh->len); 2659 saddr = ip_hdr(skb)->saddr; 2660 daddr = ip_hdr(skb)->daddr; 2661 2662 if (ulen > skb->len) 2663 goto short_packet; 2664 2665 if (proto == IPPROTO_UDP) { 2666 /* UDP validates ulen. */ 2667 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2668 goto short_packet; 2669 uh = udp_hdr(skb); 2670 } 2671 2672 if (udp4_csum_init(skb, uh, proto)) 2673 goto csum_error; 2674 2675 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 2676 &refcounted, udp_ehashfn); 2677 if (IS_ERR(sk)) 2678 goto no_sk; 2679 2680 if (sk) { 2681 struct dst_entry *dst = skb_dst(skb); 2682 int ret; 2683 2684 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2685 udp_sk_rx_dst_set(sk, dst); 2686 2687 ret = udp_unicast_rcv_skb(sk, skb, uh); 2688 if (refcounted) 2689 sock_put(sk); 2690 return ret; 2691 } 2692 2693 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2694 return __udp4_lib_mcast_deliver(net, skb, uh, 2695 saddr, daddr, udptable, proto); 2696 2697 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2698 if (sk) 2699 return udp_unicast_rcv_skb(sk, skb, uh); 2700 no_sk: 2701 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2702 goto drop; 2703 nf_reset_ct(skb); 2704 2705 /* No socket. Drop packet silently, if checksum is wrong */ 2706 if (udp_lib_checksum_complete(skb)) 2707 goto csum_error; 2708 2709 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2710 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2711 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2712 2713 /* 2714 * Hmm. We got an UDP packet to a port to which we 2715 * don't wanna listen. Ignore it. 2716 */ 2717 sk_skb_reason_drop(sk, skb, drop_reason); 2718 return 0; 2719 2720 short_packet: 2721 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2722 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2723 proto == IPPROTO_UDPLITE ? "Lite" : "", 2724 &saddr, ntohs(uh->source), 2725 ulen, skb->len, 2726 &daddr, ntohs(uh->dest)); 2727 goto drop; 2728 2729 csum_error: 2730 /* 2731 * RFC1122: OK. Discards the bad packet silently (as far as 2732 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2733 */ 2734 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2735 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2736 proto == IPPROTO_UDPLITE ? "Lite" : "", 2737 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2738 ulen); 2739 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2740 drop: 2741 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2742 sk_skb_reason_drop(sk, skb, drop_reason); 2743 return 0; 2744 } 2745 2746 /* We can only early demux multicast if there is a single matching socket. 2747 * If more than one socket found returns NULL 2748 */ 2749 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2750 __be16 loc_port, __be32 loc_addr, 2751 __be16 rmt_port, __be32 rmt_addr, 2752 int dif, int sdif) 2753 { 2754 struct udp_table *udptable = net->ipv4.udp_table; 2755 unsigned short hnum = ntohs(loc_port); 2756 struct sock *sk, *result; 2757 struct udp_hslot *hslot; 2758 unsigned int slot; 2759 2760 slot = udp_hashfn(net, hnum, udptable->mask); 2761 hslot = &udptable->hash[slot]; 2762 2763 /* Do not bother scanning a too big list */ 2764 if (hslot->count > 10) 2765 return NULL; 2766 2767 result = NULL; 2768 sk_for_each_rcu(sk, &hslot->head) { 2769 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2770 rmt_port, rmt_addr, dif, sdif, hnum)) { 2771 if (result) 2772 return NULL; 2773 result = sk; 2774 } 2775 } 2776 2777 return result; 2778 } 2779 2780 /* For unicast we should only early demux connected sockets or we can 2781 * break forwarding setups. The chains here can be long so only check 2782 * if the first socket is an exact match and if not move on. 2783 */ 2784 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2785 __be16 loc_port, __be32 loc_addr, 2786 __be16 rmt_port, __be32 rmt_addr, 2787 int dif, int sdif) 2788 { 2789 struct udp_table *udptable = net->ipv4.udp_table; 2790 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2791 unsigned short hnum = ntohs(loc_port); 2792 struct udp_hslot *hslot2; 2793 unsigned int hash2; 2794 __portpair ports; 2795 struct sock *sk; 2796 2797 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2798 hslot2 = udp_hashslot2(udptable, hash2); 2799 ports = INET_COMBINED_PORTS(rmt_port, hnum); 2800 2801 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2802 if (inet_match(net, sk, acookie, ports, dif, sdif)) 2803 return sk; 2804 /* Only check first socket in chain */ 2805 break; 2806 } 2807 return NULL; 2808 } 2809 2810 int udp_v4_early_demux(struct sk_buff *skb) 2811 { 2812 struct net *net = dev_net(skb->dev); 2813 struct in_device *in_dev = NULL; 2814 const struct iphdr *iph; 2815 const struct udphdr *uh; 2816 struct sock *sk = NULL; 2817 struct dst_entry *dst; 2818 int dif = skb->dev->ifindex; 2819 int sdif = inet_sdif(skb); 2820 int ours; 2821 2822 /* validate the packet */ 2823 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2824 return 0; 2825 2826 iph = ip_hdr(skb); 2827 uh = udp_hdr(skb); 2828 2829 if (skb->pkt_type == PACKET_MULTICAST) { 2830 in_dev = __in_dev_get_rcu(skb->dev); 2831 2832 if (!in_dev) 2833 return 0; 2834 2835 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2836 iph->protocol); 2837 if (!ours) 2838 return 0; 2839 2840 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2841 uh->source, iph->saddr, 2842 dif, sdif); 2843 } else if (skb->pkt_type == PACKET_HOST) { 2844 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2845 uh->source, iph->saddr, dif, sdif); 2846 } 2847 2848 if (!sk) 2849 return 0; 2850 2851 skb->sk = sk; 2852 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk)); 2853 skb->destructor = sock_pfree; 2854 dst = rcu_dereference(sk->sk_rx_dst); 2855 2856 if (dst) 2857 dst = dst_check(dst, 0); 2858 if (dst) { 2859 u32 itag = 0; 2860 2861 /* set noref for now. 2862 * any place which wants to hold dst has to call 2863 * dst_hold_safe() 2864 */ 2865 skb_dst_set_noref(skb, dst); 2866 2867 /* for unconnected multicast sockets we need to validate 2868 * the source on each packet 2869 */ 2870 if (!inet_sk(sk)->inet_daddr && in_dev) 2871 return ip_mc_validate_source(skb, iph->daddr, 2872 iph->saddr, 2873 ip4h_dscp(iph), 2874 skb->dev, in_dev, &itag); 2875 } 2876 return 0; 2877 } 2878 2879 int udp_rcv(struct sk_buff *skb) 2880 { 2881 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 2882 } 2883 2884 void udp_destroy_sock(struct sock *sk) 2885 { 2886 struct udp_sock *up = udp_sk(sk); 2887 bool slow = lock_sock_fast(sk); 2888 2889 /* protects from races with udp_abort() */ 2890 sock_set_flag(sk, SOCK_DEAD); 2891 udp_flush_pending_frames(sk); 2892 unlock_sock_fast(sk, slow); 2893 if (static_branch_unlikely(&udp_encap_needed_key)) { 2894 if (up->encap_type) { 2895 void (*encap_destroy)(struct sock *sk); 2896 encap_destroy = READ_ONCE(up->encap_destroy); 2897 if (encap_destroy) 2898 encap_destroy(sk); 2899 } 2900 if (udp_test_bit(ENCAP_ENABLED, sk)) { 2901 static_branch_dec(&udp_encap_needed_key); 2902 udp_tunnel_cleanup_gro(sk); 2903 } 2904 } 2905 } 2906 2907 typedef struct sk_buff *(*udp_gro_receive_t)(struct sock *sk, 2908 struct list_head *head, 2909 struct sk_buff *skb); 2910 2911 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family, 2912 struct sock *sk) 2913 { 2914 #ifdef CONFIG_XFRM 2915 udp_gro_receive_t new_gro_receive; 2916 2917 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) { 2918 if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) 2919 new_gro_receive = ipv6_stub->xfrm6_gro_udp_encap_rcv; 2920 else 2921 new_gro_receive = xfrm4_gro_udp_encap_rcv; 2922 2923 if (udp_sk(sk)->gro_receive != new_gro_receive) { 2924 /* 2925 * With IPV6_ADDRFORM the gro callback could change 2926 * after being set, unregister the old one, if valid. 2927 */ 2928 if (udp_sk(sk)->gro_receive) 2929 udp_tunnel_update_gro_rcv(sk, false); 2930 2931 WRITE_ONCE(udp_sk(sk)->gro_receive, new_gro_receive); 2932 udp_tunnel_update_gro_rcv(sk, true); 2933 } 2934 } 2935 #endif 2936 } 2937 2938 /* 2939 * Socket option code for UDP 2940 */ 2941 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2942 sockptr_t optval, unsigned int optlen, 2943 int (*push_pending_frames)(struct sock *)) 2944 { 2945 struct udp_sock *up = udp_sk(sk); 2946 int val, valbool; 2947 int err = 0; 2948 int is_udplite = IS_UDPLITE(sk); 2949 2950 if (level == SOL_SOCKET) { 2951 err = sk_setsockopt(sk, level, optname, optval, optlen); 2952 2953 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) { 2954 sockopt_lock_sock(sk); 2955 /* paired with READ_ONCE in udp_rmem_release() */ 2956 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2); 2957 sockopt_release_sock(sk); 2958 } 2959 return err; 2960 } 2961 2962 if (optlen < sizeof(int)) 2963 return -EINVAL; 2964 2965 if (copy_from_sockptr(&val, optval, sizeof(val))) 2966 return -EFAULT; 2967 2968 valbool = val ? 1 : 0; 2969 2970 switch (optname) { 2971 case UDP_CORK: 2972 if (val != 0) { 2973 udp_set_bit(CORK, sk); 2974 } else { 2975 udp_clear_bit(CORK, sk); 2976 lock_sock(sk); 2977 push_pending_frames(sk); 2978 release_sock(sk); 2979 } 2980 break; 2981 2982 case UDP_ENCAP: 2983 sockopt_lock_sock(sk); 2984 switch (val) { 2985 case 0: 2986 #ifdef CONFIG_XFRM 2987 case UDP_ENCAP_ESPINUDP: 2988 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk); 2989 #if IS_ENABLED(CONFIG_IPV6) 2990 if (sk->sk_family == AF_INET6) 2991 WRITE_ONCE(up->encap_rcv, 2992 ipv6_stub->xfrm6_udp_encap_rcv); 2993 else 2994 #endif 2995 WRITE_ONCE(up->encap_rcv, 2996 xfrm4_udp_encap_rcv); 2997 #endif 2998 fallthrough; 2999 case UDP_ENCAP_L2TPINUDP: 3000 WRITE_ONCE(up->encap_type, val); 3001 udp_tunnel_encap_enable(sk); 3002 break; 3003 default: 3004 err = -ENOPROTOOPT; 3005 break; 3006 } 3007 sockopt_release_sock(sk); 3008 break; 3009 3010 case UDP_NO_CHECK6_TX: 3011 udp_set_no_check6_tx(sk, valbool); 3012 break; 3013 3014 case UDP_NO_CHECK6_RX: 3015 udp_set_no_check6_rx(sk, valbool); 3016 break; 3017 3018 case UDP_SEGMENT: 3019 if (val < 0 || val > USHRT_MAX) 3020 return -EINVAL; 3021 WRITE_ONCE(up->gso_size, val); 3022 break; 3023 3024 case UDP_GRO: 3025 sockopt_lock_sock(sk); 3026 /* when enabling GRO, accept the related GSO packet type */ 3027 if (valbool) 3028 udp_tunnel_encap_enable(sk); 3029 udp_assign_bit(GRO_ENABLED, sk, valbool); 3030 udp_assign_bit(ACCEPT_L4, sk, valbool); 3031 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk); 3032 sockopt_release_sock(sk); 3033 break; 3034 3035 /* 3036 * UDP-Lite's partial checksum coverage (RFC 3828). 3037 */ 3038 /* The sender sets actual checksum coverage length via this option. 3039 * The case coverage > packet length is handled by send module. */ 3040 case UDPLITE_SEND_CSCOV: 3041 if (!is_udplite) /* Disable the option on UDP sockets */ 3042 return -ENOPROTOOPT; 3043 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 3044 val = 8; 3045 else if (val > USHRT_MAX) 3046 val = USHRT_MAX; 3047 WRITE_ONCE(up->pcslen, val); 3048 udp_set_bit(UDPLITE_SEND_CC, sk); 3049 break; 3050 3051 /* The receiver specifies a minimum checksum coverage value. To make 3052 * sense, this should be set to at least 8 (as done below). If zero is 3053 * used, this again means full checksum coverage. */ 3054 case UDPLITE_RECV_CSCOV: 3055 if (!is_udplite) /* Disable the option on UDP sockets */ 3056 return -ENOPROTOOPT; 3057 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 3058 val = 8; 3059 else if (val > USHRT_MAX) 3060 val = USHRT_MAX; 3061 WRITE_ONCE(up->pcrlen, val); 3062 udp_set_bit(UDPLITE_RECV_CC, sk); 3063 break; 3064 3065 default: 3066 err = -ENOPROTOOPT; 3067 break; 3068 } 3069 3070 return err; 3071 } 3072 EXPORT_IPV6_MOD(udp_lib_setsockopt); 3073 3074 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 3075 unsigned int optlen) 3076 { 3077 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 3078 return udp_lib_setsockopt(sk, level, optname, 3079 optval, optlen, 3080 udp_push_pending_frames); 3081 return ip_setsockopt(sk, level, optname, optval, optlen); 3082 } 3083 3084 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 3085 char __user *optval, int __user *optlen) 3086 { 3087 struct udp_sock *up = udp_sk(sk); 3088 int val, len; 3089 3090 if (get_user(len, optlen)) 3091 return -EFAULT; 3092 3093 if (len < 0) 3094 return -EINVAL; 3095 3096 len = min_t(unsigned int, len, sizeof(int)); 3097 3098 switch (optname) { 3099 case UDP_CORK: 3100 val = udp_test_bit(CORK, sk); 3101 break; 3102 3103 case UDP_ENCAP: 3104 val = READ_ONCE(up->encap_type); 3105 break; 3106 3107 case UDP_NO_CHECK6_TX: 3108 val = udp_get_no_check6_tx(sk); 3109 break; 3110 3111 case UDP_NO_CHECK6_RX: 3112 val = udp_get_no_check6_rx(sk); 3113 break; 3114 3115 case UDP_SEGMENT: 3116 val = READ_ONCE(up->gso_size); 3117 break; 3118 3119 case UDP_GRO: 3120 val = udp_test_bit(GRO_ENABLED, sk); 3121 break; 3122 3123 /* The following two cannot be changed on UDP sockets, the return is 3124 * always 0 (which corresponds to the full checksum coverage of UDP). */ 3125 case UDPLITE_SEND_CSCOV: 3126 val = READ_ONCE(up->pcslen); 3127 break; 3128 3129 case UDPLITE_RECV_CSCOV: 3130 val = READ_ONCE(up->pcrlen); 3131 break; 3132 3133 default: 3134 return -ENOPROTOOPT; 3135 } 3136 3137 if (put_user(len, optlen)) 3138 return -EFAULT; 3139 if (copy_to_user(optval, &val, len)) 3140 return -EFAULT; 3141 return 0; 3142 } 3143 EXPORT_IPV6_MOD(udp_lib_getsockopt); 3144 3145 int udp_getsockopt(struct sock *sk, int level, int optname, 3146 char __user *optval, int __user *optlen) 3147 { 3148 if (level == SOL_UDP || level == SOL_UDPLITE) 3149 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 3150 return ip_getsockopt(sk, level, optname, optval, optlen); 3151 } 3152 3153 /** 3154 * udp_poll - wait for a UDP event. 3155 * @file: - file struct 3156 * @sock: - socket 3157 * @wait: - poll table 3158 * 3159 * This is same as datagram poll, except for the special case of 3160 * blocking sockets. If application is using a blocking fd 3161 * and a packet with checksum error is in the queue; 3162 * then it could get return from select indicating data available 3163 * but then block when reading it. Add special case code 3164 * to work around these arguably broken applications. 3165 */ 3166 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 3167 { 3168 __poll_t mask = datagram_poll(file, sock, wait); 3169 struct sock *sk = sock->sk; 3170 3171 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 3172 mask |= EPOLLIN | EPOLLRDNORM; 3173 3174 /* Check for false positives due to checksum errors */ 3175 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 3176 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 3177 mask &= ~(EPOLLIN | EPOLLRDNORM); 3178 3179 /* psock ingress_msg queue should not contain any bad checksum frames */ 3180 if (sk_is_readable(sk)) 3181 mask |= EPOLLIN | EPOLLRDNORM; 3182 return mask; 3183 3184 } 3185 EXPORT_IPV6_MOD(udp_poll); 3186 3187 int udp_abort(struct sock *sk, int err) 3188 { 3189 if (!has_current_bpf_ctx()) 3190 lock_sock(sk); 3191 3192 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 3193 * with close() 3194 */ 3195 if (sock_flag(sk, SOCK_DEAD)) 3196 goto out; 3197 3198 sk->sk_err = err; 3199 sk_error_report(sk); 3200 __udp_disconnect(sk, 0); 3201 3202 out: 3203 if (!has_current_bpf_ctx()) 3204 release_sock(sk); 3205 3206 return 0; 3207 } 3208 EXPORT_IPV6_MOD_GPL(udp_abort); 3209 3210 struct proto udp_prot = { 3211 .name = "UDP", 3212 .owner = THIS_MODULE, 3213 .close = udp_lib_close, 3214 .pre_connect = udp_pre_connect, 3215 .connect = udp_connect, 3216 .disconnect = udp_disconnect, 3217 .ioctl = udp_ioctl, 3218 .init = udp_init_sock, 3219 .destroy = udp_destroy_sock, 3220 .setsockopt = udp_setsockopt, 3221 .getsockopt = udp_getsockopt, 3222 .sendmsg = udp_sendmsg, 3223 .recvmsg = udp_recvmsg, 3224 .splice_eof = udp_splice_eof, 3225 .release_cb = ip4_datagram_release_cb, 3226 .hash = udp_lib_hash, 3227 .unhash = udp_lib_unhash, 3228 .rehash = udp_v4_rehash, 3229 .get_port = udp_v4_get_port, 3230 .put_port = udp_lib_unhash, 3231 #ifdef CONFIG_BPF_SYSCALL 3232 .psock_update_sk_prot = udp_bpf_update_proto, 3233 #endif 3234 .memory_allocated = &net_aligned_data.udp_memory_allocated, 3235 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 3236 3237 .sysctl_mem = sysctl_udp_mem, 3238 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 3239 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 3240 .obj_size = sizeof(struct udp_sock), 3241 .h.udp_table = NULL, 3242 .diag_destroy = udp_abort, 3243 }; 3244 EXPORT_SYMBOL(udp_prot); 3245 3246 /* ------------------------------------------------------------------------ */ 3247 #ifdef CONFIG_PROC_FS 3248 3249 static unsigned short seq_file_family(const struct seq_file *seq); 3250 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk) 3251 { 3252 unsigned short family = seq_file_family(seq); 3253 3254 /* AF_UNSPEC is used as a match all */ 3255 return ((family == AF_UNSPEC || family == sk->sk_family) && 3256 net_eq(sock_net(sk), seq_file_net(seq))); 3257 } 3258 3259 #ifdef CONFIG_BPF_SYSCALL 3260 static const struct seq_operations bpf_iter_udp_seq_ops; 3261 #endif 3262 static struct udp_table *udp_get_table_seq(struct seq_file *seq, 3263 struct net *net) 3264 { 3265 const struct udp_seq_afinfo *afinfo; 3266 3267 #ifdef CONFIG_BPF_SYSCALL 3268 if (seq->op == &bpf_iter_udp_seq_ops) 3269 return net->ipv4.udp_table; 3270 #endif 3271 3272 afinfo = pde_data(file_inode(seq->file)); 3273 return afinfo->udp_table ? : net->ipv4.udp_table; 3274 } 3275 3276 static struct sock *udp_get_first(struct seq_file *seq, int start) 3277 { 3278 struct udp_iter_state *state = seq->private; 3279 struct net *net = seq_file_net(seq); 3280 struct udp_table *udptable; 3281 struct sock *sk; 3282 3283 udptable = udp_get_table_seq(seq, net); 3284 3285 for (state->bucket = start; state->bucket <= udptable->mask; 3286 ++state->bucket) { 3287 struct udp_hslot *hslot = &udptable->hash[state->bucket]; 3288 3289 if (hlist_empty(&hslot->head)) 3290 continue; 3291 3292 spin_lock_bh(&hslot->lock); 3293 sk_for_each(sk, &hslot->head) { 3294 if (seq_sk_match(seq, sk)) 3295 goto found; 3296 } 3297 spin_unlock_bh(&hslot->lock); 3298 } 3299 sk = NULL; 3300 found: 3301 return sk; 3302 } 3303 3304 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 3305 { 3306 struct udp_iter_state *state = seq->private; 3307 struct net *net = seq_file_net(seq); 3308 struct udp_table *udptable; 3309 3310 do { 3311 sk = sk_next(sk); 3312 } while (sk && !seq_sk_match(seq, sk)); 3313 3314 if (!sk) { 3315 udptable = udp_get_table_seq(seq, net); 3316 3317 if (state->bucket <= udptable->mask) 3318 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3319 3320 return udp_get_first(seq, state->bucket + 1); 3321 } 3322 return sk; 3323 } 3324 3325 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3326 { 3327 struct sock *sk = udp_get_first(seq, 0); 3328 3329 if (sk) 3330 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3331 --pos; 3332 return pos ? NULL : sk; 3333 } 3334 3335 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3336 { 3337 struct udp_iter_state *state = seq->private; 3338 state->bucket = MAX_UDP_PORTS; 3339 3340 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3341 } 3342 EXPORT_IPV6_MOD(udp_seq_start); 3343 3344 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3345 { 3346 struct sock *sk; 3347 3348 if (v == SEQ_START_TOKEN) 3349 sk = udp_get_idx(seq, 0); 3350 else 3351 sk = udp_get_next(seq, v); 3352 3353 ++*pos; 3354 return sk; 3355 } 3356 EXPORT_IPV6_MOD(udp_seq_next); 3357 3358 void udp_seq_stop(struct seq_file *seq, void *v) 3359 { 3360 struct udp_iter_state *state = seq->private; 3361 struct udp_table *udptable; 3362 3363 udptable = udp_get_table_seq(seq, seq_file_net(seq)); 3364 3365 if (state->bucket <= udptable->mask) 3366 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3367 } 3368 EXPORT_IPV6_MOD(udp_seq_stop); 3369 3370 /* ------------------------------------------------------------------------ */ 3371 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3372 int bucket) 3373 { 3374 struct inet_sock *inet = inet_sk(sp); 3375 __be32 dest = inet->inet_daddr; 3376 __be32 src = inet->inet_rcv_saddr; 3377 __u16 destp = ntohs(inet->inet_dport); 3378 __u16 srcp = ntohs(inet->inet_sport); 3379 3380 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3381 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3382 bucket, src, srcp, dest, destp, sp->sk_state, 3383 sk_wmem_alloc_get(sp), 3384 udp_rqueue_get(sp), 3385 0, 0L, 0, 3386 from_kuid_munged(seq_user_ns(f), sk_uid(sp)), 3387 0, sock_i_ino(sp), 3388 refcount_read(&sp->sk_refcnt), sp, 3389 atomic_read(&sp->sk_drops)); 3390 } 3391 3392 int udp4_seq_show(struct seq_file *seq, void *v) 3393 { 3394 seq_setwidth(seq, 127); 3395 if (v == SEQ_START_TOKEN) 3396 seq_puts(seq, " sl local_address rem_address st tx_queue " 3397 "rx_queue tr tm->when retrnsmt uid timeout " 3398 "inode ref pointer drops"); 3399 else { 3400 struct udp_iter_state *state = seq->private; 3401 3402 udp4_format_sock(v, seq, state->bucket); 3403 } 3404 seq_pad(seq, '\n'); 3405 return 0; 3406 } 3407 3408 #ifdef CONFIG_BPF_SYSCALL 3409 struct bpf_iter__udp { 3410 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3411 __bpf_md_ptr(struct udp_sock *, udp_sk); 3412 uid_t uid __aligned(8); 3413 int bucket __aligned(8); 3414 }; 3415 3416 union bpf_udp_iter_batch_item { 3417 struct sock *sk; 3418 __u64 cookie; 3419 }; 3420 3421 struct bpf_udp_iter_state { 3422 struct udp_iter_state state; 3423 unsigned int cur_sk; 3424 unsigned int end_sk; 3425 unsigned int max_sk; 3426 union bpf_udp_iter_batch_item *batch; 3427 }; 3428 3429 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3430 unsigned int new_batch_sz, gfp_t flags); 3431 static struct sock *bpf_iter_udp_resume(struct sock *first_sk, 3432 union bpf_udp_iter_batch_item *cookies, 3433 int n_cookies) 3434 { 3435 struct sock *sk = NULL; 3436 int i; 3437 3438 for (i = 0; i < n_cookies; i++) { 3439 sk = first_sk; 3440 udp_portaddr_for_each_entry_from(sk) 3441 if (cookies[i].cookie == atomic64_read(&sk->sk_cookie)) 3442 goto done; 3443 } 3444 done: 3445 return sk; 3446 } 3447 3448 static struct sock *bpf_iter_udp_batch(struct seq_file *seq) 3449 { 3450 struct bpf_udp_iter_state *iter = seq->private; 3451 struct udp_iter_state *state = &iter->state; 3452 unsigned int find_cookie, end_cookie; 3453 struct net *net = seq_file_net(seq); 3454 struct udp_table *udptable; 3455 unsigned int batch_sks = 0; 3456 int resume_bucket; 3457 int resizes = 0; 3458 struct sock *sk; 3459 int err = 0; 3460 3461 resume_bucket = state->bucket; 3462 3463 /* The current batch is done, so advance the bucket. */ 3464 if (iter->cur_sk == iter->end_sk) 3465 state->bucket++; 3466 3467 udptable = udp_get_table_seq(seq, net); 3468 3469 again: 3470 /* New batch for the next bucket. 3471 * Iterate over the hash table to find a bucket with sockets matching 3472 * the iterator attributes, and return the first matching socket from 3473 * the bucket. The remaining matched sockets from the bucket are batched 3474 * before releasing the bucket lock. This allows BPF programs that are 3475 * called in seq_show to acquire the bucket lock if needed. 3476 */ 3477 find_cookie = iter->cur_sk; 3478 end_cookie = iter->end_sk; 3479 iter->cur_sk = 0; 3480 iter->end_sk = 0; 3481 batch_sks = 0; 3482 3483 for (; state->bucket <= udptable->mask; state->bucket++) { 3484 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket].hslot; 3485 3486 if (hlist_empty(&hslot2->head)) 3487 goto next_bucket; 3488 3489 spin_lock_bh(&hslot2->lock); 3490 sk = hlist_entry_safe(hslot2->head.first, struct sock, 3491 __sk_common.skc_portaddr_node); 3492 /* Resume from the first (in iteration order) unseen socket from 3493 * the last batch that still exists in resume_bucket. Most of 3494 * the time this will just be where the last iteration left off 3495 * in resume_bucket unless that socket disappeared between 3496 * reads. 3497 */ 3498 if (state->bucket == resume_bucket) 3499 sk = bpf_iter_udp_resume(sk, &iter->batch[find_cookie], 3500 end_cookie - find_cookie); 3501 fill_batch: 3502 udp_portaddr_for_each_entry_from(sk) { 3503 if (seq_sk_match(seq, sk)) { 3504 if (iter->end_sk < iter->max_sk) { 3505 sock_hold(sk); 3506 iter->batch[iter->end_sk++].sk = sk; 3507 } 3508 batch_sks++; 3509 } 3510 } 3511 3512 /* Allocate a larger batch and try again. */ 3513 if (unlikely(resizes <= 1 && iter->end_sk && 3514 iter->end_sk != batch_sks)) { 3515 resizes++; 3516 3517 /* First, try with GFP_USER to maximize the chances of 3518 * grabbing more memory. 3519 */ 3520 if (resizes == 1) { 3521 spin_unlock_bh(&hslot2->lock); 3522 err = bpf_iter_udp_realloc_batch(iter, 3523 batch_sks * 3 / 2, 3524 GFP_USER); 3525 if (err) 3526 return ERR_PTR(err); 3527 /* Start over. */ 3528 goto again; 3529 } 3530 3531 /* Next, hold onto the lock, so the bucket doesn't 3532 * change while we get the rest of the sockets. 3533 */ 3534 err = bpf_iter_udp_realloc_batch(iter, batch_sks, 3535 GFP_NOWAIT); 3536 if (err) { 3537 spin_unlock_bh(&hslot2->lock); 3538 return ERR_PTR(err); 3539 } 3540 3541 /* Pick up where we left off. */ 3542 sk = iter->batch[iter->end_sk - 1].sk; 3543 sk = hlist_entry_safe(sk->__sk_common.skc_portaddr_node.next, 3544 struct sock, 3545 __sk_common.skc_portaddr_node); 3546 batch_sks = iter->end_sk; 3547 goto fill_batch; 3548 } 3549 3550 spin_unlock_bh(&hslot2->lock); 3551 3552 if (iter->end_sk) 3553 break; 3554 next_bucket: 3555 resizes = 0; 3556 } 3557 3558 WARN_ON_ONCE(iter->end_sk != batch_sks); 3559 return iter->end_sk ? iter->batch[0].sk : NULL; 3560 } 3561 3562 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3563 { 3564 struct bpf_udp_iter_state *iter = seq->private; 3565 struct sock *sk; 3566 3567 /* Whenever seq_next() is called, the iter->cur_sk is 3568 * done with seq_show(), so unref the iter->cur_sk. 3569 */ 3570 if (iter->cur_sk < iter->end_sk) 3571 sock_put(iter->batch[iter->cur_sk++].sk); 3572 3573 /* After updating iter->cur_sk, check if there are more sockets 3574 * available in the current bucket batch. 3575 */ 3576 if (iter->cur_sk < iter->end_sk) 3577 sk = iter->batch[iter->cur_sk].sk; 3578 else 3579 /* Prepare a new batch. */ 3580 sk = bpf_iter_udp_batch(seq); 3581 3582 ++*pos; 3583 return sk; 3584 } 3585 3586 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos) 3587 { 3588 /* bpf iter does not support lseek, so it always 3589 * continue from where it was stop()-ped. 3590 */ 3591 if (*pos) 3592 return bpf_iter_udp_batch(seq); 3593 3594 return SEQ_START_TOKEN; 3595 } 3596 3597 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3598 struct udp_sock *udp_sk, uid_t uid, int bucket) 3599 { 3600 struct bpf_iter__udp ctx; 3601 3602 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3603 ctx.meta = meta; 3604 ctx.udp_sk = udp_sk; 3605 ctx.uid = uid; 3606 ctx.bucket = bucket; 3607 return bpf_iter_run_prog(prog, &ctx); 3608 } 3609 3610 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3611 { 3612 struct udp_iter_state *state = seq->private; 3613 struct bpf_iter_meta meta; 3614 struct bpf_prog *prog; 3615 struct sock *sk = v; 3616 uid_t uid; 3617 int ret; 3618 3619 if (v == SEQ_START_TOKEN) 3620 return 0; 3621 3622 lock_sock(sk); 3623 3624 if (unlikely(sk_unhashed(sk))) { 3625 ret = SEQ_SKIP; 3626 goto unlock; 3627 } 3628 3629 uid = from_kuid_munged(seq_user_ns(seq), sk_uid(sk)); 3630 meta.seq = seq; 3631 prog = bpf_iter_get_info(&meta, false); 3632 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3633 3634 unlock: 3635 release_sock(sk); 3636 return ret; 3637 } 3638 3639 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter) 3640 { 3641 union bpf_udp_iter_batch_item *item; 3642 unsigned int cur_sk = iter->cur_sk; 3643 __u64 cookie; 3644 3645 /* Remember the cookies of the sockets we haven't seen yet, so we can 3646 * pick up where we left off next time around. 3647 */ 3648 while (cur_sk < iter->end_sk) { 3649 item = &iter->batch[cur_sk++]; 3650 cookie = sock_gen_cookie(item->sk); 3651 sock_put(item->sk); 3652 item->cookie = cookie; 3653 } 3654 } 3655 3656 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3657 { 3658 struct bpf_udp_iter_state *iter = seq->private; 3659 struct bpf_iter_meta meta; 3660 struct bpf_prog *prog; 3661 3662 if (!v) { 3663 meta.seq = seq; 3664 prog = bpf_iter_get_info(&meta, true); 3665 if (prog) 3666 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3667 } 3668 3669 if (iter->cur_sk < iter->end_sk) 3670 bpf_iter_udp_put_batch(iter); 3671 } 3672 3673 static const struct seq_operations bpf_iter_udp_seq_ops = { 3674 .start = bpf_iter_udp_seq_start, 3675 .next = bpf_iter_udp_seq_next, 3676 .stop = bpf_iter_udp_seq_stop, 3677 .show = bpf_iter_udp_seq_show, 3678 }; 3679 #endif 3680 3681 static unsigned short seq_file_family(const struct seq_file *seq) 3682 { 3683 const struct udp_seq_afinfo *afinfo; 3684 3685 #ifdef CONFIG_BPF_SYSCALL 3686 /* BPF iterator: bpf programs to filter sockets. */ 3687 if (seq->op == &bpf_iter_udp_seq_ops) 3688 return AF_UNSPEC; 3689 #endif 3690 3691 /* Proc fs iterator */ 3692 afinfo = pde_data(file_inode(seq->file)); 3693 return afinfo->family; 3694 } 3695 3696 const struct seq_operations udp_seq_ops = { 3697 .start = udp_seq_start, 3698 .next = udp_seq_next, 3699 .stop = udp_seq_stop, 3700 .show = udp4_seq_show, 3701 }; 3702 EXPORT_IPV6_MOD(udp_seq_ops); 3703 3704 static struct udp_seq_afinfo udp4_seq_afinfo = { 3705 .family = AF_INET, 3706 .udp_table = NULL, 3707 }; 3708 3709 static int __net_init udp4_proc_init_net(struct net *net) 3710 { 3711 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3712 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3713 return -ENOMEM; 3714 return 0; 3715 } 3716 3717 static void __net_exit udp4_proc_exit_net(struct net *net) 3718 { 3719 remove_proc_entry("udp", net->proc_net); 3720 } 3721 3722 static struct pernet_operations udp4_net_ops = { 3723 .init = udp4_proc_init_net, 3724 .exit = udp4_proc_exit_net, 3725 }; 3726 3727 int __init udp4_proc_init(void) 3728 { 3729 return register_pernet_subsys(&udp4_net_ops); 3730 } 3731 3732 void udp4_proc_exit(void) 3733 { 3734 unregister_pernet_subsys(&udp4_net_ops); 3735 } 3736 #endif /* CONFIG_PROC_FS */ 3737 3738 static __initdata unsigned long uhash_entries; 3739 static int __init set_uhash_entries(char *str) 3740 { 3741 ssize_t ret; 3742 3743 if (!str) 3744 return 0; 3745 3746 ret = kstrtoul(str, 0, &uhash_entries); 3747 if (ret) 3748 return 0; 3749 3750 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3751 uhash_entries = UDP_HTABLE_SIZE_MIN; 3752 return 1; 3753 } 3754 __setup("uhash_entries=", set_uhash_entries); 3755 3756 void __init udp_table_init(struct udp_table *table, const char *name) 3757 { 3758 unsigned int i, slot_size; 3759 3760 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) + 3761 udp_hash4_slot_size(); 3762 table->hash = alloc_large_system_hash(name, 3763 slot_size, 3764 uhash_entries, 3765 21, /* one slot per 2 MB */ 3766 0, 3767 &table->log, 3768 &table->mask, 3769 UDP_HTABLE_SIZE_MIN, 3770 UDP_HTABLE_SIZE_MAX); 3771 3772 table->hash2 = (void *)(table->hash + (table->mask + 1)); 3773 for (i = 0; i <= table->mask; i++) { 3774 INIT_HLIST_HEAD(&table->hash[i].head); 3775 table->hash[i].count = 0; 3776 spin_lock_init(&table->hash[i].lock); 3777 } 3778 for (i = 0; i <= table->mask; i++) { 3779 INIT_HLIST_HEAD(&table->hash2[i].hslot.head); 3780 table->hash2[i].hslot.count = 0; 3781 spin_lock_init(&table->hash2[i].hslot.lock); 3782 } 3783 udp_table_hash4_init(table); 3784 } 3785 3786 u32 udp_flow_hashrnd(void) 3787 { 3788 static u32 hashrnd __read_mostly; 3789 3790 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3791 3792 return hashrnd; 3793 } 3794 EXPORT_SYMBOL(udp_flow_hashrnd); 3795 3796 static void __net_init udp_sysctl_init(struct net *net) 3797 { 3798 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3799 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 3800 3801 #ifdef CONFIG_NET_L3_MASTER_DEV 3802 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3803 #endif 3804 } 3805 3806 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries) 3807 { 3808 struct udp_table *udptable; 3809 unsigned int slot_size; 3810 int i; 3811 3812 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL); 3813 if (!udptable) 3814 goto out; 3815 3816 slot_size = sizeof(struct udp_hslot) + sizeof(struct udp_hslot_main) + 3817 udp_hash4_slot_size(); 3818 udptable->hash = vmalloc_huge(hash_entries * slot_size, 3819 GFP_KERNEL_ACCOUNT); 3820 if (!udptable->hash) 3821 goto free_table; 3822 3823 udptable->hash2 = (void *)(udptable->hash + hash_entries); 3824 udptable->mask = hash_entries - 1; 3825 udptable->log = ilog2(hash_entries); 3826 3827 for (i = 0; i < hash_entries; i++) { 3828 INIT_HLIST_HEAD(&udptable->hash[i].head); 3829 udptable->hash[i].count = 0; 3830 spin_lock_init(&udptable->hash[i].lock); 3831 3832 INIT_HLIST_HEAD(&udptable->hash2[i].hslot.head); 3833 udptable->hash2[i].hslot.count = 0; 3834 spin_lock_init(&udptable->hash2[i].hslot.lock); 3835 } 3836 udp_table_hash4_init(udptable); 3837 3838 return udptable; 3839 3840 free_table: 3841 kfree(udptable); 3842 out: 3843 return NULL; 3844 } 3845 3846 static void __net_exit udp_pernet_table_free(struct net *net) 3847 { 3848 struct udp_table *udptable = net->ipv4.udp_table; 3849 3850 if (udptable == &udp_table) 3851 return; 3852 3853 kvfree(udptable->hash); 3854 kfree(udptable); 3855 } 3856 3857 static void __net_init udp_set_table(struct net *net) 3858 { 3859 struct udp_table *udptable; 3860 unsigned int hash_entries; 3861 struct net *old_net; 3862 3863 if (net_eq(net, &init_net)) 3864 goto fallback; 3865 3866 old_net = current->nsproxy->net_ns; 3867 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries); 3868 if (!hash_entries) 3869 goto fallback; 3870 3871 /* Set min to keep the bitmap on stack in udp_lib_get_port() */ 3872 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET) 3873 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET; 3874 else 3875 hash_entries = roundup_pow_of_two(hash_entries); 3876 3877 udptable = udp_pernet_table_alloc(hash_entries); 3878 if (udptable) { 3879 net->ipv4.udp_table = udptable; 3880 } else { 3881 pr_warn("Failed to allocate UDP hash table (entries: %u) " 3882 "for a netns, fallback to the global one\n", 3883 hash_entries); 3884 fallback: 3885 net->ipv4.udp_table = &udp_table; 3886 } 3887 } 3888 3889 static int __net_init udp_pernet_init(struct net *net) 3890 { 3891 #if IS_ENABLED(CONFIG_NET_UDP_TUNNEL) 3892 int i; 3893 3894 /* No tunnel is configured */ 3895 for (i = 0; i < ARRAY_SIZE(net->ipv4.udp_tunnel_gro); ++i) { 3896 INIT_HLIST_HEAD(&net->ipv4.udp_tunnel_gro[i].list); 3897 RCU_INIT_POINTER(net->ipv4.udp_tunnel_gro[i].sk, NULL); 3898 } 3899 #endif 3900 udp_sysctl_init(net); 3901 udp_set_table(net); 3902 3903 return 0; 3904 } 3905 3906 static void __net_exit udp_pernet_exit(struct net *net) 3907 { 3908 udp_pernet_table_free(net); 3909 } 3910 3911 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3912 .init = udp_pernet_init, 3913 .exit = udp_pernet_exit, 3914 }; 3915 3916 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3917 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3918 struct udp_sock *udp_sk, uid_t uid, int bucket) 3919 3920 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3921 unsigned int new_batch_sz, gfp_t flags) 3922 { 3923 union bpf_udp_iter_batch_item *new_batch; 3924 3925 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch), 3926 flags | __GFP_NOWARN); 3927 if (!new_batch) 3928 return -ENOMEM; 3929 3930 if (flags != GFP_NOWAIT) 3931 bpf_iter_udp_put_batch(iter); 3932 3933 memcpy(new_batch, iter->batch, sizeof(*iter->batch) * iter->end_sk); 3934 kvfree(iter->batch); 3935 iter->batch = new_batch; 3936 iter->max_sk = new_batch_sz; 3937 3938 return 0; 3939 } 3940 3941 #define INIT_BATCH_SZ 16 3942 3943 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3944 { 3945 struct bpf_udp_iter_state *iter = priv_data; 3946 int ret; 3947 3948 ret = bpf_iter_init_seq_net(priv_data, aux); 3949 if (ret) 3950 return ret; 3951 3952 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ, GFP_USER); 3953 if (ret) 3954 bpf_iter_fini_seq_net(priv_data); 3955 3956 iter->state.bucket = -1; 3957 3958 return ret; 3959 } 3960 3961 static void bpf_iter_fini_udp(void *priv_data) 3962 { 3963 struct bpf_udp_iter_state *iter = priv_data; 3964 3965 bpf_iter_fini_seq_net(priv_data); 3966 kvfree(iter->batch); 3967 } 3968 3969 static const struct bpf_iter_seq_info udp_seq_info = { 3970 .seq_ops = &bpf_iter_udp_seq_ops, 3971 .init_seq_private = bpf_iter_init_udp, 3972 .fini_seq_private = bpf_iter_fini_udp, 3973 .seq_priv_size = sizeof(struct bpf_udp_iter_state), 3974 }; 3975 3976 static struct bpf_iter_reg udp_reg_info = { 3977 .target = "udp", 3978 .ctx_arg_info_size = 1, 3979 .ctx_arg_info = { 3980 { offsetof(struct bpf_iter__udp, udp_sk), 3981 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED }, 3982 }, 3983 .seq_info = &udp_seq_info, 3984 }; 3985 3986 static void __init bpf_iter_register(void) 3987 { 3988 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3989 if (bpf_iter_reg_target(&udp_reg_info)) 3990 pr_warn("Warning: could not register bpf iterator udp\n"); 3991 } 3992 #endif 3993 3994 void __init udp_init(void) 3995 { 3996 unsigned long limit; 3997 unsigned int i; 3998 3999 udp_table_init(&udp_table, "UDP"); 4000 limit = nr_free_buffer_pages() / 8; 4001 limit = max(limit, 128UL); 4002 sysctl_udp_mem[0] = limit / 4 * 3; 4003 sysctl_udp_mem[1] = limit; 4004 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 4005 4006 /* 16 spinlocks per cpu */ 4007 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 4008 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 4009 GFP_KERNEL); 4010 if (!udp_busylocks) 4011 panic("UDP: failed to alloc udp_busylocks\n"); 4012 for (i = 0; i < (1U << udp_busylocks_log); i++) 4013 spin_lock_init(udp_busylocks + i); 4014 4015 if (register_pernet_subsys(&udp_sysctl_ops)) 4016 panic("UDP: failed to init sysctl parameters.\n"); 4017 4018 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 4019 bpf_iter_register(); 4020 #endif 4021 } 4022