1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * The User Datagram Protocol (UDP). 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 12 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 13 * Hirokazu Takahashi, <taka@valinux.co.jp> 14 * 15 * Fixes: 16 * Alan Cox : verify_area() calls 17 * Alan Cox : stopped close while in use off icmp 18 * messages. Not a fix but a botch that 19 * for udp at least is 'valid'. 20 * Alan Cox : Fixed icmp handling properly 21 * Alan Cox : Correct error for oversized datagrams 22 * Alan Cox : Tidied select() semantics. 23 * Alan Cox : udp_err() fixed properly, also now 24 * select and read wake correctly on errors 25 * Alan Cox : udp_send verify_area moved to avoid mem leak 26 * Alan Cox : UDP can count its memory 27 * Alan Cox : send to an unknown connection causes 28 * an ECONNREFUSED off the icmp, but 29 * does NOT close. 30 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 32 * bug no longer crashes it. 33 * Fred Van Kempen : Net2e support for sk->broadcast. 34 * Alan Cox : Uses skb_free_datagram 35 * Alan Cox : Added get/set sockopt support. 36 * Alan Cox : Broadcasting without option set returns EACCES. 37 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 38 * Alan Cox : Use ip_tos and ip_ttl 39 * Alan Cox : SNMP Mibs 40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 41 * Matt Dillon : UDP length checks. 42 * Alan Cox : Smarter af_inet used properly. 43 * Alan Cox : Use new kernel side addressing. 44 * Alan Cox : Incorrect return on truncated datagram receive. 45 * Arnt Gulbrandsen : New udp_send and stuff 46 * Alan Cox : Cache last socket 47 * Alan Cox : Route cache 48 * Jon Peatfield : Minor efficiency fix to sendto(). 49 * Mike Shaver : RFC1122 checks. 50 * Alan Cox : Nonblocking error fix. 51 * Willy Konynenberg : Transparent proxying support. 52 * Mike McLagan : Routing by source 53 * David S. Miller : New socket lookup architecture. 54 * Last socket cache retained as it 55 * does have a high hit rate. 56 * Olaf Kirch : Don't linearise iovec on sendmsg. 57 * Andi Kleen : Some cleanups, cache destination entry 58 * for connect. 59 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 60 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 61 * return ENOTCONN for unconnected sockets (POSIX) 62 * Janos Farkas : don't deliver multi/broadcasts to a different 63 * bound-to-device socket 64 * Hirokazu Takahashi : HW checksumming for outgoing UDP 65 * datagrams. 66 * Hirokazu Takahashi : sendfile() on UDP works now. 67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 70 * a single port at the same time. 71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 72 * James Chapman : Add L2TP encapsulation type. 73 */ 74 75 #define pr_fmt(fmt) "UDP: " fmt 76 77 #include <linux/bpf-cgroup.h> 78 #include <linux/uaccess.h> 79 #include <asm/ioctls.h> 80 #include <linux/memblock.h> 81 #include <linux/highmem.h> 82 #include <linux/types.h> 83 #include <linux/fcntl.h> 84 #include <linux/module.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/igmp.h> 88 #include <linux/inetdevice.h> 89 #include <linux/in.h> 90 #include <linux/errno.h> 91 #include <linux/timer.h> 92 #include <linux/mm.h> 93 #include <linux/inet.h> 94 #include <linux/netdevice.h> 95 #include <linux/slab.h> 96 #include <net/tcp_states.h> 97 #include <linux/skbuff.h> 98 #include <linux/proc_fs.h> 99 #include <linux/seq_file.h> 100 #include <net/net_namespace.h> 101 #include <net/icmp.h> 102 #include <net/inet_hashtables.h> 103 #include <net/ip_tunnels.h> 104 #include <net/route.h> 105 #include <net/checksum.h> 106 #include <net/gso.h> 107 #include <net/xfrm.h> 108 #include <trace/events/udp.h> 109 #include <linux/static_key.h> 110 #include <linux/btf_ids.h> 111 #include <trace/events/skb.h> 112 #include <net/busy_poll.h> 113 #include "udp_impl.h" 114 #include <net/sock_reuseport.h> 115 #include <net/addrconf.h> 116 #include <net/udp_tunnel.h> 117 #include <net/gro.h> 118 #if IS_ENABLED(CONFIG_IPV6) 119 #include <net/ipv6_stubs.h> 120 #endif 121 122 struct udp_table udp_table __read_mostly; 123 EXPORT_SYMBOL(udp_table); 124 125 long sysctl_udp_mem[3] __read_mostly; 126 EXPORT_SYMBOL(sysctl_udp_mem); 127 128 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp; 129 EXPORT_SYMBOL(udp_memory_allocated); 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 struct sock *sk2; 147 kuid_t uid = sock_i_uid(sk); 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, sock_i_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 struct sock *sk2; 182 kuid_t uid = sock_i_uid(sk); 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, sock_i_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 = sock_i_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, sock_i_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 sk_add_node_rcu(sk, &hslot->head); 330 hslot->count++; 331 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 332 333 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 334 spin_lock(&hslot2->lock); 335 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 336 sk->sk_family == AF_INET6) 337 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 338 &hslot2->head); 339 else 340 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 341 &hslot2->head); 342 hslot2->count++; 343 spin_unlock(&hslot2->lock); 344 } 345 sock_set_flag(sk, SOCK_RCU_FREE); 346 error = 0; 347 fail_unlock: 348 spin_unlock_bh(&hslot->lock); 349 fail: 350 return error; 351 } 352 EXPORT_SYMBOL(udp_lib_get_port); 353 354 int udp_v4_get_port(struct sock *sk, unsigned short snum) 355 { 356 unsigned int hash2_nulladdr = 357 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 358 unsigned int hash2_partial = 359 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 360 361 /* precompute partial secondary hash */ 362 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 363 return udp_lib_get_port(sk, snum, hash2_nulladdr); 364 } 365 366 static int compute_score(struct sock *sk, struct net *net, 367 __be32 saddr, __be16 sport, 368 __be32 daddr, unsigned short hnum, 369 int dif, int sdif) 370 { 371 int score; 372 struct inet_sock *inet; 373 bool dev_match; 374 375 if (!net_eq(sock_net(sk), net) || 376 udp_sk(sk)->udp_port_hash != hnum || 377 ipv6_only_sock(sk)) 378 return -1; 379 380 if (sk->sk_rcv_saddr != daddr) 381 return -1; 382 383 score = (sk->sk_family == PF_INET) ? 2 : 1; 384 385 inet = inet_sk(sk); 386 if (inet->inet_daddr) { 387 if (inet->inet_daddr != saddr) 388 return -1; 389 score += 4; 390 } 391 392 if (inet->inet_dport) { 393 if (inet->inet_dport != sport) 394 return -1; 395 score += 4; 396 } 397 398 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 399 dif, sdif); 400 if (!dev_match) 401 return -1; 402 if (sk->sk_bound_dev_if) 403 score += 4; 404 405 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 406 score++; 407 return score; 408 } 409 410 INDIRECT_CALLABLE_SCOPE 411 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport, 412 const __be32 faddr, const __be16 fport) 413 { 414 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 415 416 return __inet_ehashfn(laddr, lport, faddr, fport, 417 udp_ehash_secret + net_hash_mix(net)); 418 } 419 420 /* called with rcu_read_lock() */ 421 static struct sock *udp4_lib_lookup2(struct net *net, 422 __be32 saddr, __be16 sport, 423 __be32 daddr, unsigned int hnum, 424 int dif, int sdif, 425 struct udp_hslot *hslot2, 426 struct sk_buff *skb) 427 { 428 struct sock *sk, *result; 429 int score, badness; 430 bool need_rescore; 431 432 result = NULL; 433 badness = 0; 434 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 435 need_rescore = false; 436 rescore: 437 score = compute_score(need_rescore ? result : sk, net, saddr, 438 sport, daddr, hnum, dif, sdif); 439 if (score > badness) { 440 badness = score; 441 442 if (need_rescore) 443 continue; 444 445 if (sk->sk_state == TCP_ESTABLISHED) { 446 result = sk; 447 continue; 448 } 449 450 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 451 saddr, sport, daddr, hnum, udp_ehashfn); 452 if (!result) { 453 result = sk; 454 continue; 455 } 456 457 /* Fall back to scoring if group has connections */ 458 if (!reuseport_has_conns(sk)) 459 return result; 460 461 /* Reuseport logic returned an error, keep original score. */ 462 if (IS_ERR(result)) 463 continue; 464 465 /* compute_score is too long of a function to be 466 * inlined, and calling it again here yields 467 * measureable overhead for some 468 * workloads. Work around it by jumping 469 * backwards to rescore 'result'. 470 */ 471 need_rescore = true; 472 goto rescore; 473 } 474 } 475 return result; 476 } 477 478 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 479 * harder than this. -DaveM 480 */ 481 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 482 __be16 sport, __be32 daddr, __be16 dport, int dif, 483 int sdif, struct udp_table *udptable, struct sk_buff *skb) 484 { 485 unsigned short hnum = ntohs(dport); 486 unsigned int hash2, slot2; 487 struct udp_hslot *hslot2; 488 struct sock *result, *sk; 489 490 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 491 slot2 = hash2 & udptable->mask; 492 hslot2 = &udptable->hash2[slot2]; 493 494 /* Lookup connected or non-wildcard socket */ 495 result = udp4_lib_lookup2(net, saddr, sport, 496 daddr, hnum, dif, sdif, 497 hslot2, skb); 498 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 499 goto done; 500 501 /* Lookup redirect from BPF */ 502 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 503 udptable == net->ipv4.udp_table) { 504 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 505 saddr, sport, daddr, hnum, dif, 506 udp_ehashfn); 507 if (sk) { 508 result = sk; 509 goto done; 510 } 511 } 512 513 /* Got non-wildcard socket or error on first lookup */ 514 if (result) 515 goto done; 516 517 /* Lookup wildcard sockets */ 518 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 519 slot2 = hash2 & udptable->mask; 520 hslot2 = &udptable->hash2[slot2]; 521 522 result = udp4_lib_lookup2(net, saddr, sport, 523 htonl(INADDR_ANY), hnum, dif, sdif, 524 hslot2, skb); 525 done: 526 if (IS_ERR(result)) 527 return NULL; 528 return result; 529 } 530 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 531 532 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 533 __be16 sport, __be16 dport, 534 struct udp_table *udptable) 535 { 536 const struct iphdr *iph = ip_hdr(skb); 537 538 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 539 iph->daddr, dport, inet_iif(skb), 540 inet_sdif(skb), udptable, skb); 541 } 542 543 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 544 __be16 sport, __be16 dport) 545 { 546 const struct iphdr *iph = ip_hdr(skb); 547 struct net *net = dev_net(skb->dev); 548 int iif, sdif; 549 550 inet_get_iif_sdif(skb, &iif, &sdif); 551 552 return __udp4_lib_lookup(net, iph->saddr, sport, 553 iph->daddr, dport, iif, 554 sdif, net->ipv4.udp_table, NULL); 555 } 556 557 /* Must be called under rcu_read_lock(). 558 * Does increment socket refcount. 559 */ 560 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 561 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 562 __be32 daddr, __be16 dport, int dif) 563 { 564 struct sock *sk; 565 566 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 567 dif, 0, net->ipv4.udp_table, NULL); 568 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 569 sk = NULL; 570 return sk; 571 } 572 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 573 #endif 574 575 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk, 576 __be16 loc_port, __be32 loc_addr, 577 __be16 rmt_port, __be32 rmt_addr, 578 int dif, int sdif, unsigned short hnum) 579 { 580 const struct inet_sock *inet = inet_sk(sk); 581 582 if (!net_eq(sock_net(sk), net) || 583 udp_sk(sk)->udp_port_hash != hnum || 584 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 585 (inet->inet_dport != rmt_port && inet->inet_dport) || 586 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 587 ipv6_only_sock(sk) || 588 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 589 return false; 590 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 591 return false; 592 return true; 593 } 594 595 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 596 EXPORT_SYMBOL(udp_encap_needed_key); 597 598 #if IS_ENABLED(CONFIG_IPV6) 599 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 600 EXPORT_SYMBOL(udpv6_encap_needed_key); 601 #endif 602 603 void udp_encap_enable(void) 604 { 605 static_branch_inc(&udp_encap_needed_key); 606 } 607 EXPORT_SYMBOL(udp_encap_enable); 608 609 void udp_encap_disable(void) 610 { 611 static_branch_dec(&udp_encap_needed_key); 612 } 613 EXPORT_SYMBOL(udp_encap_disable); 614 615 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 616 * through error handlers in encapsulations looking for a match. 617 */ 618 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 619 { 620 int i; 621 622 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 623 int (*handler)(struct sk_buff *skb, u32 info); 624 const struct ip_tunnel_encap_ops *encap; 625 626 encap = rcu_dereference(iptun_encaps[i]); 627 if (!encap) 628 continue; 629 handler = encap->err_handler; 630 if (handler && !handler(skb, info)) 631 return 0; 632 } 633 634 return -ENOENT; 635 } 636 637 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 638 * reversing source and destination port: this will match tunnels that force the 639 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 640 * lwtunnels might actually break this assumption by being configured with 641 * different destination ports on endpoints, in this case we won't be able to 642 * trace ICMP messages back to them. 643 * 644 * If this doesn't match any socket, probe tunnels with arbitrary destination 645 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 646 * we've sent packets to won't necessarily match the local destination port. 647 * 648 * Then ask the tunnel implementation to match the error against a valid 649 * association. 650 * 651 * Return an error if we can't find a match, the socket if we need further 652 * processing, zero otherwise. 653 */ 654 static struct sock *__udp4_lib_err_encap(struct net *net, 655 const struct iphdr *iph, 656 struct udphdr *uh, 657 struct udp_table *udptable, 658 struct sock *sk, 659 struct sk_buff *skb, u32 info) 660 { 661 int (*lookup)(struct sock *sk, struct sk_buff *skb); 662 int network_offset, transport_offset; 663 struct udp_sock *up; 664 665 network_offset = skb_network_offset(skb); 666 transport_offset = skb_transport_offset(skb); 667 668 /* Network header needs to point to the outer IPv4 header inside ICMP */ 669 skb_reset_network_header(skb); 670 671 /* Transport header needs to point to the UDP header */ 672 skb_set_transport_header(skb, iph->ihl << 2); 673 674 if (sk) { 675 up = udp_sk(sk); 676 677 lookup = READ_ONCE(up->encap_err_lookup); 678 if (lookup && lookup(sk, skb)) 679 sk = NULL; 680 681 goto out; 682 } 683 684 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 685 iph->saddr, uh->dest, skb->dev->ifindex, 0, 686 udptable, NULL); 687 if (sk) { 688 up = udp_sk(sk); 689 690 lookup = READ_ONCE(up->encap_err_lookup); 691 if (!lookup || lookup(sk, skb)) 692 sk = NULL; 693 } 694 695 out: 696 if (!sk) 697 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 698 699 skb_set_transport_header(skb, transport_offset); 700 skb_set_network_header(skb, network_offset); 701 702 return sk; 703 } 704 705 /* 706 * This routine is called by the ICMP module when it gets some 707 * sort of error condition. If err < 0 then the socket should 708 * be closed and the error returned to the user. If err > 0 709 * it's just the icmp type << 8 | icmp code. 710 * Header points to the ip header of the error packet. We move 711 * on past this. Then (as it used to claim before adjustment) 712 * header points to the first 8 bytes of the udp header. We need 713 * to find the appropriate port. 714 */ 715 716 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 717 { 718 struct inet_sock *inet; 719 const struct iphdr *iph = (const struct iphdr *)skb->data; 720 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 721 const int type = icmp_hdr(skb)->type; 722 const int code = icmp_hdr(skb)->code; 723 bool tunnel = false; 724 struct sock *sk; 725 int harderr; 726 int err; 727 struct net *net = dev_net(skb->dev); 728 729 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 730 iph->saddr, uh->source, skb->dev->ifindex, 731 inet_sdif(skb), udptable, NULL); 732 733 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) { 734 /* No socket for error: try tunnels before discarding */ 735 if (static_branch_unlikely(&udp_encap_needed_key)) { 736 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 737 info); 738 if (!sk) 739 return 0; 740 } else 741 sk = ERR_PTR(-ENOENT); 742 743 if (IS_ERR(sk)) { 744 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 745 return PTR_ERR(sk); 746 } 747 748 tunnel = true; 749 } 750 751 err = 0; 752 harderr = 0; 753 inet = inet_sk(sk); 754 755 switch (type) { 756 default: 757 case ICMP_TIME_EXCEEDED: 758 err = EHOSTUNREACH; 759 break; 760 case ICMP_SOURCE_QUENCH: 761 goto out; 762 case ICMP_PARAMETERPROB: 763 err = EPROTO; 764 harderr = 1; 765 break; 766 case ICMP_DEST_UNREACH: 767 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 768 ipv4_sk_update_pmtu(skb, sk, info); 769 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) { 770 err = EMSGSIZE; 771 harderr = 1; 772 break; 773 } 774 goto out; 775 } 776 err = EHOSTUNREACH; 777 if (code <= NR_ICMP_UNREACH) { 778 harderr = icmp_err_convert[code].fatal; 779 err = icmp_err_convert[code].errno; 780 } 781 break; 782 case ICMP_REDIRECT: 783 ipv4_sk_redirect(skb, sk); 784 goto out; 785 } 786 787 /* 788 * RFC1122: OK. Passes ICMP errors back to application, as per 789 * 4.1.3.3. 790 */ 791 if (tunnel) { 792 /* ...not for tunnels though: we don't have a sending socket */ 793 if (udp_sk(sk)->encap_err_rcv) 794 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info, 795 (u8 *)(uh+1)); 796 goto out; 797 } 798 if (!inet_test_bit(RECVERR, sk)) { 799 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 800 goto out; 801 } else 802 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 803 804 sk->sk_err = err; 805 sk_error_report(sk); 806 out: 807 return 0; 808 } 809 810 int udp_err(struct sk_buff *skb, u32 info) 811 { 812 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table); 813 } 814 815 /* 816 * Throw away all pending data and cancel the corking. Socket is locked. 817 */ 818 void udp_flush_pending_frames(struct sock *sk) 819 { 820 struct udp_sock *up = udp_sk(sk); 821 822 if (up->pending) { 823 up->len = 0; 824 WRITE_ONCE(up->pending, 0); 825 ip_flush_pending_frames(sk); 826 } 827 } 828 EXPORT_SYMBOL(udp_flush_pending_frames); 829 830 /** 831 * udp4_hwcsum - handle outgoing HW checksumming 832 * @skb: sk_buff containing the filled-in UDP header 833 * (checksum field must be zeroed out) 834 * @src: source IP address 835 * @dst: destination IP address 836 */ 837 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 838 { 839 struct udphdr *uh = udp_hdr(skb); 840 int offset = skb_transport_offset(skb); 841 int len = skb->len - offset; 842 int hlen = len; 843 __wsum csum = 0; 844 845 if (!skb_has_frag_list(skb)) { 846 /* 847 * Only one fragment on the socket. 848 */ 849 skb->csum_start = skb_transport_header(skb) - skb->head; 850 skb->csum_offset = offsetof(struct udphdr, check); 851 uh->check = ~csum_tcpudp_magic(src, dst, len, 852 IPPROTO_UDP, 0); 853 } else { 854 struct sk_buff *frags; 855 856 /* 857 * HW-checksum won't work as there are two or more 858 * fragments on the socket so that all csums of sk_buffs 859 * should be together 860 */ 861 skb_walk_frags(skb, frags) { 862 csum = csum_add(csum, frags->csum); 863 hlen -= frags->len; 864 } 865 866 csum = skb_checksum(skb, offset, hlen, csum); 867 skb->ip_summed = CHECKSUM_NONE; 868 869 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 870 if (uh->check == 0) 871 uh->check = CSUM_MANGLED_0; 872 } 873 } 874 EXPORT_SYMBOL_GPL(udp4_hwcsum); 875 876 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 877 * for the simple case like when setting the checksum for a UDP tunnel. 878 */ 879 void udp_set_csum(bool nocheck, struct sk_buff *skb, 880 __be32 saddr, __be32 daddr, int len) 881 { 882 struct udphdr *uh = udp_hdr(skb); 883 884 if (nocheck) { 885 uh->check = 0; 886 } else if (skb_is_gso(skb)) { 887 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 888 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 889 uh->check = 0; 890 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 891 if (uh->check == 0) 892 uh->check = CSUM_MANGLED_0; 893 } else { 894 skb->ip_summed = CHECKSUM_PARTIAL; 895 skb->csum_start = skb_transport_header(skb) - skb->head; 896 skb->csum_offset = offsetof(struct udphdr, check); 897 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 898 } 899 } 900 EXPORT_SYMBOL(udp_set_csum); 901 902 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 903 struct inet_cork *cork) 904 { 905 struct sock *sk = skb->sk; 906 struct inet_sock *inet = inet_sk(sk); 907 struct udphdr *uh; 908 int err; 909 int is_udplite = IS_UDPLITE(sk); 910 int offset = skb_transport_offset(skb); 911 int len = skb->len - offset; 912 int datalen = len - sizeof(*uh); 913 __wsum csum = 0; 914 915 /* 916 * Create a UDP header 917 */ 918 uh = udp_hdr(skb); 919 uh->source = inet->inet_sport; 920 uh->dest = fl4->fl4_dport; 921 uh->len = htons(len); 922 uh->check = 0; 923 924 if (cork->gso_size) { 925 const int hlen = skb_network_header_len(skb) + 926 sizeof(struct udphdr); 927 928 if (hlen + cork->gso_size > cork->fragsize) { 929 kfree_skb(skb); 930 return -EINVAL; 931 } 932 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 933 kfree_skb(skb); 934 return -EINVAL; 935 } 936 if (sk->sk_no_check_tx) { 937 kfree_skb(skb); 938 return -EINVAL; 939 } 940 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 941 dst_xfrm(skb_dst(skb))) { 942 kfree_skb(skb); 943 return -EIO; 944 } 945 946 if (datalen > cork->gso_size) { 947 skb_shinfo(skb)->gso_size = cork->gso_size; 948 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 949 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 950 cork->gso_size); 951 } 952 goto csum_partial; 953 } 954 955 if (is_udplite) /* UDP-Lite */ 956 csum = udplite_csum(skb); 957 958 else if (sk->sk_no_check_tx) { /* UDP csum off */ 959 960 skb->ip_summed = CHECKSUM_NONE; 961 goto send; 962 963 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 964 csum_partial: 965 966 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 967 goto send; 968 969 } else 970 csum = udp_csum(skb); 971 972 /* add protocol-dependent pseudo-header */ 973 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 974 sk->sk_protocol, csum); 975 if (uh->check == 0) 976 uh->check = CSUM_MANGLED_0; 977 978 send: 979 err = ip_send_skb(sock_net(sk), skb); 980 if (err) { 981 if (err == -ENOBUFS && 982 !inet_test_bit(RECVERR, sk)) { 983 UDP_INC_STATS(sock_net(sk), 984 UDP_MIB_SNDBUFERRORS, is_udplite); 985 err = 0; 986 } 987 } else 988 UDP_INC_STATS(sock_net(sk), 989 UDP_MIB_OUTDATAGRAMS, is_udplite); 990 return err; 991 } 992 993 /* 994 * Push out all pending data as one UDP datagram. Socket is locked. 995 */ 996 int udp_push_pending_frames(struct sock *sk) 997 { 998 struct udp_sock *up = udp_sk(sk); 999 struct inet_sock *inet = inet_sk(sk); 1000 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 1001 struct sk_buff *skb; 1002 int err = 0; 1003 1004 skb = ip_finish_skb(sk, fl4); 1005 if (!skb) 1006 goto out; 1007 1008 err = udp_send_skb(skb, fl4, &inet->cork.base); 1009 1010 out: 1011 up->len = 0; 1012 WRITE_ONCE(up->pending, 0); 1013 return err; 1014 } 1015 EXPORT_SYMBOL(udp_push_pending_frames); 1016 1017 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1018 { 1019 switch (cmsg->cmsg_type) { 1020 case UDP_SEGMENT: 1021 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1022 return -EINVAL; 1023 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1024 return 0; 1025 default: 1026 return -EINVAL; 1027 } 1028 } 1029 1030 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1031 { 1032 struct cmsghdr *cmsg; 1033 bool need_ip = false; 1034 int err; 1035 1036 for_each_cmsghdr(cmsg, msg) { 1037 if (!CMSG_OK(msg, cmsg)) 1038 return -EINVAL; 1039 1040 if (cmsg->cmsg_level != SOL_UDP) { 1041 need_ip = true; 1042 continue; 1043 } 1044 1045 err = __udp_cmsg_send(cmsg, gso_size); 1046 if (err) 1047 return err; 1048 } 1049 1050 return need_ip; 1051 } 1052 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1053 1054 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1055 { 1056 struct inet_sock *inet = inet_sk(sk); 1057 struct udp_sock *up = udp_sk(sk); 1058 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1059 struct flowi4 fl4_stack; 1060 struct flowi4 *fl4; 1061 int ulen = len; 1062 struct ipcm_cookie ipc; 1063 struct rtable *rt = NULL; 1064 int free = 0; 1065 int connected = 0; 1066 __be32 daddr, faddr, saddr; 1067 u8 tos, scope; 1068 __be16 dport; 1069 int err, is_udplite = IS_UDPLITE(sk); 1070 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE; 1071 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1072 struct sk_buff *skb; 1073 struct ip_options_data opt_copy; 1074 int uc_index; 1075 1076 if (len > 0xFFFF) 1077 return -EMSGSIZE; 1078 1079 /* 1080 * Check the flags. 1081 */ 1082 1083 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1084 return -EOPNOTSUPP; 1085 1086 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1087 1088 fl4 = &inet->cork.fl.u.ip4; 1089 if (READ_ONCE(up->pending)) { 1090 /* 1091 * There are pending frames. 1092 * The socket lock must be held while it's corked. 1093 */ 1094 lock_sock(sk); 1095 if (likely(up->pending)) { 1096 if (unlikely(up->pending != AF_INET)) { 1097 release_sock(sk); 1098 return -EINVAL; 1099 } 1100 goto do_append_data; 1101 } 1102 release_sock(sk); 1103 } 1104 ulen += sizeof(struct udphdr); 1105 1106 /* 1107 * Get and verify the address. 1108 */ 1109 if (usin) { 1110 if (msg->msg_namelen < sizeof(*usin)) 1111 return -EINVAL; 1112 if (usin->sin_family != AF_INET) { 1113 if (usin->sin_family != AF_UNSPEC) 1114 return -EAFNOSUPPORT; 1115 } 1116 1117 daddr = usin->sin_addr.s_addr; 1118 dport = usin->sin_port; 1119 if (dport == 0) 1120 return -EINVAL; 1121 } else { 1122 if (sk->sk_state != TCP_ESTABLISHED) 1123 return -EDESTADDRREQ; 1124 daddr = inet->inet_daddr; 1125 dport = inet->inet_dport; 1126 /* Open fast path for connected socket. 1127 Route will not be used, if at least one option is set. 1128 */ 1129 connected = 1; 1130 } 1131 1132 ipcm_init_sk(&ipc, inet); 1133 ipc.gso_size = READ_ONCE(up->gso_size); 1134 1135 if (msg->msg_controllen) { 1136 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1137 if (err > 0) { 1138 err = ip_cmsg_send(sk, msg, &ipc, 1139 sk->sk_family == AF_INET6); 1140 connected = 0; 1141 } 1142 if (unlikely(err < 0)) { 1143 kfree(ipc.opt); 1144 return err; 1145 } 1146 if (ipc.opt) 1147 free = 1; 1148 } 1149 if (!ipc.opt) { 1150 struct ip_options_rcu *inet_opt; 1151 1152 rcu_read_lock(); 1153 inet_opt = rcu_dereference(inet->inet_opt); 1154 if (inet_opt) { 1155 memcpy(&opt_copy, inet_opt, 1156 sizeof(*inet_opt) + inet_opt->opt.optlen); 1157 ipc.opt = &opt_copy.opt; 1158 } 1159 rcu_read_unlock(); 1160 } 1161 1162 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1163 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1164 (struct sockaddr *)usin, 1165 &msg->msg_namelen, 1166 &ipc.addr); 1167 if (err) 1168 goto out_free; 1169 if (usin) { 1170 if (usin->sin_port == 0) { 1171 /* BPF program set invalid port. Reject it. */ 1172 err = -EINVAL; 1173 goto out_free; 1174 } 1175 daddr = usin->sin_addr.s_addr; 1176 dport = usin->sin_port; 1177 } 1178 } 1179 1180 saddr = ipc.addr; 1181 ipc.addr = faddr = daddr; 1182 1183 if (ipc.opt && ipc.opt->opt.srr) { 1184 if (!daddr) { 1185 err = -EINVAL; 1186 goto out_free; 1187 } 1188 faddr = ipc.opt->opt.faddr; 1189 connected = 0; 1190 } 1191 tos = get_rttos(&ipc, inet); 1192 scope = ip_sendmsg_scope(inet, &ipc, msg); 1193 if (scope == RT_SCOPE_LINK) 1194 connected = 0; 1195 1196 uc_index = READ_ONCE(inet->uc_index); 1197 if (ipv4_is_multicast(daddr)) { 1198 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1199 ipc.oif = READ_ONCE(inet->mc_index); 1200 if (!saddr) 1201 saddr = READ_ONCE(inet->mc_addr); 1202 connected = 0; 1203 } else if (!ipc.oif) { 1204 ipc.oif = uc_index; 1205 } else if (ipv4_is_lbcast(daddr) && uc_index) { 1206 /* oif is set, packet is to local broadcast and 1207 * uc_index is set. oif is most likely set 1208 * by sk_bound_dev_if. If uc_index != oif check if the 1209 * oif is an L3 master and uc_index is an L3 slave. 1210 * If so, we want to allow the send using the uc_index. 1211 */ 1212 if (ipc.oif != uc_index && 1213 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1214 uc_index)) { 1215 ipc.oif = uc_index; 1216 } 1217 } 1218 1219 if (connected) 1220 rt = (struct rtable *)sk_dst_check(sk, 0); 1221 1222 if (!rt) { 1223 struct net *net = sock_net(sk); 1224 __u8 flow_flags = inet_sk_flowi_flags(sk); 1225 1226 fl4 = &fl4_stack; 1227 1228 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope, 1229 sk->sk_protocol, flow_flags, faddr, saddr, 1230 dport, inet->inet_sport, sk->sk_uid); 1231 1232 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1233 rt = ip_route_output_flow(net, fl4, sk); 1234 if (IS_ERR(rt)) { 1235 err = PTR_ERR(rt); 1236 rt = NULL; 1237 if (err == -ENETUNREACH) 1238 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1239 goto out; 1240 } 1241 1242 err = -EACCES; 1243 if ((rt->rt_flags & RTCF_BROADCAST) && 1244 !sock_flag(sk, SOCK_BROADCAST)) 1245 goto out; 1246 if (connected) 1247 sk_dst_set(sk, dst_clone(&rt->dst)); 1248 } 1249 1250 if (msg->msg_flags&MSG_CONFIRM) 1251 goto do_confirm; 1252 back_from_confirm: 1253 1254 saddr = fl4->saddr; 1255 if (!ipc.addr) 1256 daddr = ipc.addr = fl4->daddr; 1257 1258 /* Lockless fast path for the non-corking case. */ 1259 if (!corkreq) { 1260 struct inet_cork cork; 1261 1262 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1263 sizeof(struct udphdr), &ipc, &rt, 1264 &cork, msg->msg_flags); 1265 err = PTR_ERR(skb); 1266 if (!IS_ERR_OR_NULL(skb)) 1267 err = udp_send_skb(skb, fl4, &cork); 1268 goto out; 1269 } 1270 1271 lock_sock(sk); 1272 if (unlikely(up->pending)) { 1273 /* The socket is already corked while preparing it. */ 1274 /* ... which is an evident application bug. --ANK */ 1275 release_sock(sk); 1276 1277 net_dbg_ratelimited("socket already corked\n"); 1278 err = -EINVAL; 1279 goto out; 1280 } 1281 /* 1282 * Now cork the socket to pend data. 1283 */ 1284 fl4 = &inet->cork.fl.u.ip4; 1285 fl4->daddr = daddr; 1286 fl4->saddr = saddr; 1287 fl4->fl4_dport = dport; 1288 fl4->fl4_sport = inet->inet_sport; 1289 WRITE_ONCE(up->pending, AF_INET); 1290 1291 do_append_data: 1292 up->len += ulen; 1293 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1294 sizeof(struct udphdr), &ipc, &rt, 1295 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1296 if (err) 1297 udp_flush_pending_frames(sk); 1298 else if (!corkreq) 1299 err = udp_push_pending_frames(sk); 1300 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1301 WRITE_ONCE(up->pending, 0); 1302 release_sock(sk); 1303 1304 out: 1305 ip_rt_put(rt); 1306 out_free: 1307 if (free) 1308 kfree(ipc.opt); 1309 if (!err) 1310 return len; 1311 /* 1312 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1313 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1314 * we don't have a good statistic (IpOutDiscards but it can be too many 1315 * things). We could add another new stat but at least for now that 1316 * seems like overkill. 1317 */ 1318 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1319 UDP_INC_STATS(sock_net(sk), 1320 UDP_MIB_SNDBUFERRORS, is_udplite); 1321 } 1322 return err; 1323 1324 do_confirm: 1325 if (msg->msg_flags & MSG_PROBE) 1326 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1327 if (!(msg->msg_flags&MSG_PROBE) || len) 1328 goto back_from_confirm; 1329 err = 0; 1330 goto out; 1331 } 1332 EXPORT_SYMBOL(udp_sendmsg); 1333 1334 void udp_splice_eof(struct socket *sock) 1335 { 1336 struct sock *sk = sock->sk; 1337 struct udp_sock *up = udp_sk(sk); 1338 1339 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk)) 1340 return; 1341 1342 lock_sock(sk); 1343 if (up->pending && !udp_test_bit(CORK, sk)) 1344 udp_push_pending_frames(sk); 1345 release_sock(sk); 1346 } 1347 EXPORT_SYMBOL_GPL(udp_splice_eof); 1348 1349 #define UDP_SKB_IS_STATELESS 0x80000000 1350 1351 /* all head states (dst, sk, nf conntrack) except skb extensions are 1352 * cleared by udp_rcv(). 1353 * 1354 * We need to preserve secpath, if present, to eventually process 1355 * IP_CMSG_PASSSEC at recvmsg() time. 1356 * 1357 * Other extensions can be cleared. 1358 */ 1359 static bool udp_try_make_stateless(struct sk_buff *skb) 1360 { 1361 if (!skb_has_extensions(skb)) 1362 return true; 1363 1364 if (!secpath_exists(skb)) { 1365 skb_ext_reset(skb); 1366 return true; 1367 } 1368 1369 return false; 1370 } 1371 1372 static void udp_set_dev_scratch(struct sk_buff *skb) 1373 { 1374 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1375 1376 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1377 scratch->_tsize_state = skb->truesize; 1378 #if BITS_PER_LONG == 64 1379 scratch->len = skb->len; 1380 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1381 scratch->is_linear = !skb_is_nonlinear(skb); 1382 #endif 1383 if (udp_try_make_stateless(skb)) 1384 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1385 } 1386 1387 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1388 { 1389 /* We come here after udp_lib_checksum_complete() returned 0. 1390 * This means that __skb_checksum_complete() might have 1391 * set skb->csum_valid to 1. 1392 * On 64bit platforms, we can set csum_unnecessary 1393 * to true, but only if the skb is not shared. 1394 */ 1395 #if BITS_PER_LONG == 64 1396 if (!skb_shared(skb)) 1397 udp_skb_scratch(skb)->csum_unnecessary = true; 1398 #endif 1399 } 1400 1401 static int udp_skb_truesize(struct sk_buff *skb) 1402 { 1403 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1404 } 1405 1406 static bool udp_skb_has_head_state(struct sk_buff *skb) 1407 { 1408 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1409 } 1410 1411 /* fully reclaim rmem/fwd memory allocated for skb */ 1412 static void udp_rmem_release(struct sock *sk, int size, int partial, 1413 bool rx_queue_lock_held) 1414 { 1415 struct udp_sock *up = udp_sk(sk); 1416 struct sk_buff_head *sk_queue; 1417 int amt; 1418 1419 if (likely(partial)) { 1420 up->forward_deficit += size; 1421 size = up->forward_deficit; 1422 if (size < READ_ONCE(up->forward_threshold) && 1423 !skb_queue_empty(&up->reader_queue)) 1424 return; 1425 } else { 1426 size += up->forward_deficit; 1427 } 1428 up->forward_deficit = 0; 1429 1430 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1431 * if the called don't held it already 1432 */ 1433 sk_queue = &sk->sk_receive_queue; 1434 if (!rx_queue_lock_held) 1435 spin_lock(&sk_queue->lock); 1436 1437 1438 sk_forward_alloc_add(sk, size); 1439 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1440 sk_forward_alloc_add(sk, -amt); 1441 1442 if (amt) 1443 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 1444 1445 atomic_sub(size, &sk->sk_rmem_alloc); 1446 1447 /* this can save us from acquiring the rx queue lock on next receive */ 1448 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1449 1450 if (!rx_queue_lock_held) 1451 spin_unlock(&sk_queue->lock); 1452 } 1453 1454 /* Note: called with reader_queue.lock held. 1455 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1456 * This avoids a cache line miss while receive_queue lock is held. 1457 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1458 */ 1459 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1460 { 1461 prefetch(&skb->data); 1462 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1463 } 1464 EXPORT_SYMBOL(udp_skb_destructor); 1465 1466 /* as above, but the caller held the rx queue lock, too */ 1467 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1468 { 1469 prefetch(&skb->data); 1470 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1471 } 1472 1473 /* Idea of busylocks is to let producers grab an extra spinlock 1474 * to relieve pressure on the receive_queue spinlock shared by consumer. 1475 * Under flood, this means that only one producer can be in line 1476 * trying to acquire the receive_queue spinlock. 1477 * These busylock can be allocated on a per cpu manner, instead of a 1478 * per socket one (that would consume a cache line per socket) 1479 */ 1480 static int udp_busylocks_log __read_mostly; 1481 static spinlock_t *udp_busylocks __read_mostly; 1482 1483 static spinlock_t *busylock_acquire(void *ptr) 1484 { 1485 spinlock_t *busy; 1486 1487 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1488 spin_lock(busy); 1489 return busy; 1490 } 1491 1492 static void busylock_release(spinlock_t *busy) 1493 { 1494 if (busy) 1495 spin_unlock(busy); 1496 } 1497 1498 static int udp_rmem_schedule(struct sock *sk, int size) 1499 { 1500 int delta; 1501 1502 delta = size - sk->sk_forward_alloc; 1503 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV)) 1504 return -ENOBUFS; 1505 1506 return 0; 1507 } 1508 1509 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1510 { 1511 struct sk_buff_head *list = &sk->sk_receive_queue; 1512 int rmem, err = -ENOMEM; 1513 spinlock_t *busy = NULL; 1514 bool becomes_readable; 1515 int size, rcvbuf; 1516 1517 /* Immediately drop when the receive queue is full. 1518 * Always allow at least one packet. 1519 */ 1520 rmem = atomic_read(&sk->sk_rmem_alloc); 1521 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1522 if (rmem > rcvbuf) 1523 goto drop; 1524 1525 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1526 * having linear skbs : 1527 * - Reduce memory overhead and thus increase receive queue capacity 1528 * - Less cache line misses at copyout() time 1529 * - Less work at consume_skb() (less alien page frag freeing) 1530 */ 1531 if (rmem > (rcvbuf >> 1)) { 1532 skb_condense(skb); 1533 1534 busy = busylock_acquire(sk); 1535 } 1536 size = skb->truesize; 1537 udp_set_dev_scratch(skb); 1538 1539 atomic_add(size, &sk->sk_rmem_alloc); 1540 1541 spin_lock(&list->lock); 1542 err = udp_rmem_schedule(sk, size); 1543 if (err) { 1544 spin_unlock(&list->lock); 1545 goto uncharge_drop; 1546 } 1547 1548 sk_forward_alloc_add(sk, -size); 1549 1550 /* no need to setup a destructor, we will explicitly release the 1551 * forward allocated memory on dequeue 1552 */ 1553 sock_skb_set_dropcount(sk, skb); 1554 1555 becomes_readable = skb_queue_empty(list); 1556 __skb_queue_tail(list, skb); 1557 spin_unlock(&list->lock); 1558 1559 if (!sock_flag(sk, SOCK_DEAD)) { 1560 if (becomes_readable || 1561 sk->sk_data_ready != sock_def_readable || 1562 READ_ONCE(sk->sk_peek_off) >= 0) 1563 INDIRECT_CALL_1(sk->sk_data_ready, 1564 sock_def_readable, sk); 1565 else 1566 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 1567 } 1568 busylock_release(busy); 1569 return 0; 1570 1571 uncharge_drop: 1572 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1573 1574 drop: 1575 atomic_inc(&sk->sk_drops); 1576 busylock_release(busy); 1577 return err; 1578 } 1579 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1580 1581 void udp_destruct_common(struct sock *sk) 1582 { 1583 /* reclaim completely the forward allocated memory */ 1584 struct udp_sock *up = udp_sk(sk); 1585 unsigned int total = 0; 1586 struct sk_buff *skb; 1587 1588 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1589 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1590 total += skb->truesize; 1591 kfree_skb(skb); 1592 } 1593 udp_rmem_release(sk, total, 0, true); 1594 } 1595 EXPORT_SYMBOL_GPL(udp_destruct_common); 1596 1597 static void udp_destruct_sock(struct sock *sk) 1598 { 1599 udp_destruct_common(sk); 1600 inet_sock_destruct(sk); 1601 } 1602 1603 int udp_init_sock(struct sock *sk) 1604 { 1605 udp_lib_init_sock(sk); 1606 sk->sk_destruct = udp_destruct_sock; 1607 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1608 return 0; 1609 } 1610 1611 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1612 { 1613 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset))) 1614 sk_peek_offset_bwd(sk, len); 1615 1616 if (!skb_unref(skb)) 1617 return; 1618 1619 /* In the more common cases we cleared the head states previously, 1620 * see __udp_queue_rcv_skb(). 1621 */ 1622 if (unlikely(udp_skb_has_head_state(skb))) 1623 skb_release_head_state(skb); 1624 __consume_stateless_skb(skb); 1625 } 1626 EXPORT_SYMBOL_GPL(skb_consume_udp); 1627 1628 static struct sk_buff *__first_packet_length(struct sock *sk, 1629 struct sk_buff_head *rcvq, 1630 int *total) 1631 { 1632 struct sk_buff *skb; 1633 1634 while ((skb = skb_peek(rcvq)) != NULL) { 1635 if (udp_lib_checksum_complete(skb)) { 1636 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1637 IS_UDPLITE(sk)); 1638 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1639 IS_UDPLITE(sk)); 1640 atomic_inc(&sk->sk_drops); 1641 __skb_unlink(skb, rcvq); 1642 *total += skb->truesize; 1643 kfree_skb(skb); 1644 } else { 1645 udp_skb_csum_unnecessary_set(skb); 1646 break; 1647 } 1648 } 1649 return skb; 1650 } 1651 1652 /** 1653 * first_packet_length - return length of first packet in receive queue 1654 * @sk: socket 1655 * 1656 * Drops all bad checksum frames, until a valid one is found. 1657 * Returns the length of found skb, or -1 if none is found. 1658 */ 1659 static int first_packet_length(struct sock *sk) 1660 { 1661 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1662 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1663 struct sk_buff *skb; 1664 int total = 0; 1665 int res; 1666 1667 spin_lock_bh(&rcvq->lock); 1668 skb = __first_packet_length(sk, rcvq, &total); 1669 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1670 spin_lock(&sk_queue->lock); 1671 skb_queue_splice_tail_init(sk_queue, rcvq); 1672 spin_unlock(&sk_queue->lock); 1673 1674 skb = __first_packet_length(sk, rcvq, &total); 1675 } 1676 res = skb ? skb->len : -1; 1677 if (total) 1678 udp_rmem_release(sk, total, 1, false); 1679 spin_unlock_bh(&rcvq->lock); 1680 return res; 1681 } 1682 1683 /* 1684 * IOCTL requests applicable to the UDP protocol 1685 */ 1686 1687 int udp_ioctl(struct sock *sk, int cmd, int *karg) 1688 { 1689 switch (cmd) { 1690 case SIOCOUTQ: 1691 { 1692 *karg = sk_wmem_alloc_get(sk); 1693 return 0; 1694 } 1695 1696 case SIOCINQ: 1697 { 1698 *karg = max_t(int, 0, first_packet_length(sk)); 1699 return 0; 1700 } 1701 1702 default: 1703 return -ENOIOCTLCMD; 1704 } 1705 1706 return 0; 1707 } 1708 EXPORT_SYMBOL(udp_ioctl); 1709 1710 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1711 int *off, int *err) 1712 { 1713 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1714 struct sk_buff_head *queue; 1715 struct sk_buff *last; 1716 long timeo; 1717 int error; 1718 1719 queue = &udp_sk(sk)->reader_queue; 1720 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1721 do { 1722 struct sk_buff *skb; 1723 1724 error = sock_error(sk); 1725 if (error) 1726 break; 1727 1728 error = -EAGAIN; 1729 do { 1730 spin_lock_bh(&queue->lock); 1731 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1732 err, &last); 1733 if (skb) { 1734 if (!(flags & MSG_PEEK)) 1735 udp_skb_destructor(sk, skb); 1736 spin_unlock_bh(&queue->lock); 1737 return skb; 1738 } 1739 1740 if (skb_queue_empty_lockless(sk_queue)) { 1741 spin_unlock_bh(&queue->lock); 1742 goto busy_check; 1743 } 1744 1745 /* refill the reader queue and walk it again 1746 * keep both queues locked to avoid re-acquiring 1747 * the sk_receive_queue lock if fwd memory scheduling 1748 * is needed. 1749 */ 1750 spin_lock(&sk_queue->lock); 1751 skb_queue_splice_tail_init(sk_queue, queue); 1752 1753 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1754 err, &last); 1755 if (skb && !(flags & MSG_PEEK)) 1756 udp_skb_dtor_locked(sk, skb); 1757 spin_unlock(&sk_queue->lock); 1758 spin_unlock_bh(&queue->lock); 1759 if (skb) 1760 return skb; 1761 1762 busy_check: 1763 if (!sk_can_busy_loop(sk)) 1764 break; 1765 1766 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1767 } while (!skb_queue_empty_lockless(sk_queue)); 1768 1769 /* sk_queue is empty, reader_queue may contain peeked packets */ 1770 } while (timeo && 1771 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1772 &error, &timeo, 1773 (struct sk_buff *)sk_queue)); 1774 1775 *err = error; 1776 return NULL; 1777 } 1778 EXPORT_SYMBOL(__skb_recv_udp); 1779 1780 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) 1781 { 1782 struct sk_buff *skb; 1783 int err; 1784 1785 try_again: 1786 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1787 if (!skb) 1788 return err; 1789 1790 if (udp_lib_checksum_complete(skb)) { 1791 int is_udplite = IS_UDPLITE(sk); 1792 struct net *net = sock_net(sk); 1793 1794 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite); 1795 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite); 1796 atomic_inc(&sk->sk_drops); 1797 kfree_skb(skb); 1798 goto try_again; 1799 } 1800 1801 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); 1802 return recv_actor(sk, skb); 1803 } 1804 EXPORT_SYMBOL(udp_read_skb); 1805 1806 /* 1807 * This should be easy, if there is something there we 1808 * return it, otherwise we block. 1809 */ 1810 1811 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 1812 int *addr_len) 1813 { 1814 struct inet_sock *inet = inet_sk(sk); 1815 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1816 struct sk_buff *skb; 1817 unsigned int ulen, copied; 1818 int off, err, peeking = flags & MSG_PEEK; 1819 int is_udplite = IS_UDPLITE(sk); 1820 bool checksum_valid = false; 1821 1822 if (flags & MSG_ERRQUEUE) 1823 return ip_recv_error(sk, msg, len, addr_len); 1824 1825 try_again: 1826 off = sk_peek_offset(sk, flags); 1827 skb = __skb_recv_udp(sk, flags, &off, &err); 1828 if (!skb) 1829 return err; 1830 1831 ulen = udp_skb_len(skb); 1832 copied = len; 1833 if (copied > ulen - off) 1834 copied = ulen - off; 1835 else if (copied < ulen) 1836 msg->msg_flags |= MSG_TRUNC; 1837 1838 /* 1839 * If checksum is needed at all, try to do it while copying the 1840 * data. If the data is truncated, or if we only want a partial 1841 * coverage checksum (UDP-Lite), do it before the copy. 1842 */ 1843 1844 if (copied < ulen || peeking || 1845 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1846 checksum_valid = udp_skb_csum_unnecessary(skb) || 1847 !__udp_lib_checksum_complete(skb); 1848 if (!checksum_valid) 1849 goto csum_copy_err; 1850 } 1851 1852 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1853 if (udp_skb_is_linear(skb)) 1854 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1855 else 1856 err = skb_copy_datagram_msg(skb, off, msg, copied); 1857 } else { 1858 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1859 1860 if (err == -EINVAL) 1861 goto csum_copy_err; 1862 } 1863 1864 if (unlikely(err)) { 1865 if (!peeking) { 1866 atomic_inc(&sk->sk_drops); 1867 UDP_INC_STATS(sock_net(sk), 1868 UDP_MIB_INERRORS, is_udplite); 1869 } 1870 kfree_skb(skb); 1871 return err; 1872 } 1873 1874 if (!peeking) 1875 UDP_INC_STATS(sock_net(sk), 1876 UDP_MIB_INDATAGRAMS, is_udplite); 1877 1878 sock_recv_cmsgs(msg, sk, skb); 1879 1880 /* Copy the address. */ 1881 if (sin) { 1882 sin->sin_family = AF_INET; 1883 sin->sin_port = udp_hdr(skb)->source; 1884 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1885 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1886 *addr_len = sizeof(*sin); 1887 1888 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1889 (struct sockaddr *)sin, 1890 addr_len); 1891 } 1892 1893 if (udp_test_bit(GRO_ENABLED, sk)) 1894 udp_cmsg_recv(msg, sk, skb); 1895 1896 if (inet_cmsg_flags(inet)) 1897 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1898 1899 err = copied; 1900 if (flags & MSG_TRUNC) 1901 err = ulen; 1902 1903 skb_consume_udp(sk, skb, peeking ? -err : err); 1904 return err; 1905 1906 csum_copy_err: 1907 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1908 udp_skb_destructor)) { 1909 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1910 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1911 } 1912 kfree_skb(skb); 1913 1914 /* starting over for a new packet, but check if we need to yield */ 1915 cond_resched(); 1916 msg->msg_flags &= ~MSG_TRUNC; 1917 goto try_again; 1918 } 1919 1920 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1921 { 1922 /* This check is replicated from __ip4_datagram_connect() and 1923 * intended to prevent BPF program called below from accessing bytes 1924 * that are out of the bound specified by user in addr_len. 1925 */ 1926 if (addr_len < sizeof(struct sockaddr_in)) 1927 return -EINVAL; 1928 1929 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len); 1930 } 1931 EXPORT_SYMBOL(udp_pre_connect); 1932 1933 int __udp_disconnect(struct sock *sk, int flags) 1934 { 1935 struct inet_sock *inet = inet_sk(sk); 1936 /* 1937 * 1003.1g - break association. 1938 */ 1939 1940 sk->sk_state = TCP_CLOSE; 1941 inet->inet_daddr = 0; 1942 inet->inet_dport = 0; 1943 sock_rps_reset_rxhash(sk); 1944 sk->sk_bound_dev_if = 0; 1945 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1946 inet_reset_saddr(sk); 1947 if (sk->sk_prot->rehash && 1948 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1949 sk->sk_prot->rehash(sk); 1950 } 1951 1952 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1953 sk->sk_prot->unhash(sk); 1954 inet->inet_sport = 0; 1955 } 1956 sk_dst_reset(sk); 1957 return 0; 1958 } 1959 EXPORT_SYMBOL(__udp_disconnect); 1960 1961 int udp_disconnect(struct sock *sk, int flags) 1962 { 1963 lock_sock(sk); 1964 __udp_disconnect(sk, flags); 1965 release_sock(sk); 1966 return 0; 1967 } 1968 EXPORT_SYMBOL(udp_disconnect); 1969 1970 void udp_lib_unhash(struct sock *sk) 1971 { 1972 if (sk_hashed(sk)) { 1973 struct udp_table *udptable = udp_get_table_prot(sk); 1974 struct udp_hslot *hslot, *hslot2; 1975 1976 hslot = udp_hashslot(udptable, sock_net(sk), 1977 udp_sk(sk)->udp_port_hash); 1978 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1979 1980 spin_lock_bh(&hslot->lock); 1981 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1982 reuseport_detach_sock(sk); 1983 if (sk_del_node_init_rcu(sk)) { 1984 hslot->count--; 1985 inet_sk(sk)->inet_num = 0; 1986 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1987 1988 spin_lock(&hslot2->lock); 1989 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1990 hslot2->count--; 1991 spin_unlock(&hslot2->lock); 1992 } 1993 spin_unlock_bh(&hslot->lock); 1994 } 1995 } 1996 EXPORT_SYMBOL(udp_lib_unhash); 1997 1998 /* 1999 * inet_rcv_saddr was changed, we must rehash secondary hash 2000 */ 2001 void udp_lib_rehash(struct sock *sk, u16 newhash) 2002 { 2003 if (sk_hashed(sk)) { 2004 struct udp_table *udptable = udp_get_table_prot(sk); 2005 struct udp_hslot *hslot, *hslot2, *nhslot2; 2006 2007 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2008 nhslot2 = udp_hashslot2(udptable, newhash); 2009 udp_sk(sk)->udp_portaddr_hash = newhash; 2010 2011 if (hslot2 != nhslot2 || 2012 rcu_access_pointer(sk->sk_reuseport_cb)) { 2013 hslot = udp_hashslot(udptable, sock_net(sk), 2014 udp_sk(sk)->udp_port_hash); 2015 /* we must lock primary chain too */ 2016 spin_lock_bh(&hslot->lock); 2017 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2018 reuseport_detach_sock(sk); 2019 2020 if (hslot2 != nhslot2) { 2021 spin_lock(&hslot2->lock); 2022 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2023 hslot2->count--; 2024 spin_unlock(&hslot2->lock); 2025 2026 spin_lock(&nhslot2->lock); 2027 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2028 &nhslot2->head); 2029 nhslot2->count++; 2030 spin_unlock(&nhslot2->lock); 2031 } 2032 2033 spin_unlock_bh(&hslot->lock); 2034 } 2035 } 2036 } 2037 EXPORT_SYMBOL(udp_lib_rehash); 2038 2039 void udp_v4_rehash(struct sock *sk) 2040 { 2041 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2042 inet_sk(sk)->inet_rcv_saddr, 2043 inet_sk(sk)->inet_num); 2044 udp_lib_rehash(sk, new_hash); 2045 } 2046 2047 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2048 { 2049 int rc; 2050 2051 if (inet_sk(sk)->inet_daddr) { 2052 sock_rps_save_rxhash(sk, skb); 2053 sk_mark_napi_id(sk, skb); 2054 sk_incoming_cpu_update(sk); 2055 } else { 2056 sk_mark_napi_id_once(sk, skb); 2057 } 2058 2059 rc = __udp_enqueue_schedule_skb(sk, skb); 2060 if (rc < 0) { 2061 int is_udplite = IS_UDPLITE(sk); 2062 int drop_reason; 2063 2064 /* Note that an ENOMEM error is charged twice */ 2065 if (rc == -ENOMEM) { 2066 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2067 is_udplite); 2068 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 2069 } else { 2070 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2071 is_udplite); 2072 drop_reason = SKB_DROP_REASON_PROTO_MEM; 2073 } 2074 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2075 trace_udp_fail_queue_rcv_skb(rc, sk, skb); 2076 kfree_skb_reason(skb, drop_reason); 2077 return -1; 2078 } 2079 2080 return 0; 2081 } 2082 2083 /* returns: 2084 * -1: error 2085 * 0: success 2086 * >0: "udp encap" protocol resubmission 2087 * 2088 * Note that in the success and error cases, the skb is assumed to 2089 * have either been requeued or freed. 2090 */ 2091 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2092 { 2093 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2094 struct udp_sock *up = udp_sk(sk); 2095 int is_udplite = IS_UDPLITE(sk); 2096 2097 /* 2098 * Charge it to the socket, dropping if the queue is full. 2099 */ 2100 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 2101 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2102 goto drop; 2103 } 2104 nf_reset_ct(skb); 2105 2106 if (static_branch_unlikely(&udp_encap_needed_key) && 2107 READ_ONCE(up->encap_type)) { 2108 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2109 2110 /* 2111 * This is an encapsulation socket so pass the skb to 2112 * the socket's udp_encap_rcv() hook. Otherwise, just 2113 * fall through and pass this up the UDP socket. 2114 * up->encap_rcv() returns the following value: 2115 * =0 if skb was successfully passed to the encap 2116 * handler or was discarded by it. 2117 * >0 if skb should be passed on to UDP. 2118 * <0 if skb should be resubmitted as proto -N 2119 */ 2120 2121 /* if we're overly short, let UDP handle it */ 2122 encap_rcv = READ_ONCE(up->encap_rcv); 2123 if (encap_rcv) { 2124 int ret; 2125 2126 /* Verify checksum before giving to encap */ 2127 if (udp_lib_checksum_complete(skb)) 2128 goto csum_error; 2129 2130 ret = encap_rcv(sk, skb); 2131 if (ret <= 0) { 2132 __UDP_INC_STATS(sock_net(sk), 2133 UDP_MIB_INDATAGRAMS, 2134 is_udplite); 2135 return -ret; 2136 } 2137 } 2138 2139 /* FALLTHROUGH -- it's a UDP Packet */ 2140 } 2141 2142 /* 2143 * UDP-Lite specific tests, ignored on UDP sockets 2144 */ 2145 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) { 2146 u16 pcrlen = READ_ONCE(up->pcrlen); 2147 2148 /* 2149 * MIB statistics other than incrementing the error count are 2150 * disabled for the following two types of errors: these depend 2151 * on the application settings, not on the functioning of the 2152 * protocol stack as such. 2153 * 2154 * RFC 3828 here recommends (sec 3.3): "There should also be a 2155 * way ... to ... at least let the receiving application block 2156 * delivery of packets with coverage values less than a value 2157 * provided by the application." 2158 */ 2159 if (pcrlen == 0) { /* full coverage was set */ 2160 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2161 UDP_SKB_CB(skb)->cscov, skb->len); 2162 goto drop; 2163 } 2164 /* The next case involves violating the min. coverage requested 2165 * by the receiver. This is subtle: if receiver wants x and x is 2166 * greater than the buffersize/MTU then receiver will complain 2167 * that it wants x while sender emits packets of smaller size y. 2168 * Therefore the above ...()->partial_cov statement is essential. 2169 */ 2170 if (UDP_SKB_CB(skb)->cscov < pcrlen) { 2171 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2172 UDP_SKB_CB(skb)->cscov, pcrlen); 2173 goto drop; 2174 } 2175 } 2176 2177 prefetch(&sk->sk_rmem_alloc); 2178 if (rcu_access_pointer(sk->sk_filter) && 2179 udp_lib_checksum_complete(skb)) 2180 goto csum_error; 2181 2182 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 2183 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 2184 goto drop; 2185 } 2186 2187 udp_csum_pull_header(skb); 2188 2189 ipv4_pktinfo_prepare(sk, skb, true); 2190 return __udp_queue_rcv_skb(sk, skb); 2191 2192 csum_error: 2193 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2194 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2195 drop: 2196 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2197 atomic_inc(&sk->sk_drops); 2198 kfree_skb_reason(skb, drop_reason); 2199 return -1; 2200 } 2201 2202 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2203 { 2204 struct sk_buff *next, *segs; 2205 int ret; 2206 2207 if (likely(!udp_unexpected_gso(sk, skb))) 2208 return udp_queue_rcv_one_skb(sk, skb); 2209 2210 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2211 __skb_push(skb, -skb_mac_offset(skb)); 2212 segs = udp_rcv_segment(sk, skb, true); 2213 skb_list_walk_safe(segs, skb, next) { 2214 __skb_pull(skb, skb_transport_offset(skb)); 2215 2216 udp_post_segment_fix_csum(skb); 2217 ret = udp_queue_rcv_one_skb(sk, skb); 2218 if (ret > 0) 2219 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2220 } 2221 return 0; 2222 } 2223 2224 /* For TCP sockets, sk_rx_dst is protected by socket lock 2225 * For UDP, we use xchg() to guard against concurrent changes. 2226 */ 2227 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2228 { 2229 struct dst_entry *old; 2230 2231 if (dst_hold_safe(dst)) { 2232 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); 2233 dst_release(old); 2234 return old != dst; 2235 } 2236 return false; 2237 } 2238 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2239 2240 /* 2241 * Multicasts and broadcasts go to each listener. 2242 * 2243 * Note: called only from the BH handler context. 2244 */ 2245 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2246 struct udphdr *uh, 2247 __be32 saddr, __be32 daddr, 2248 struct udp_table *udptable, 2249 int proto) 2250 { 2251 struct sock *sk, *first = NULL; 2252 unsigned short hnum = ntohs(uh->dest); 2253 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2254 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2255 unsigned int offset = offsetof(typeof(*sk), sk_node); 2256 int dif = skb->dev->ifindex; 2257 int sdif = inet_sdif(skb); 2258 struct hlist_node *node; 2259 struct sk_buff *nskb; 2260 2261 if (use_hash2) { 2262 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2263 udptable->mask; 2264 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2265 start_lookup: 2266 hslot = &udptable->hash2[hash2]; 2267 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2268 } 2269 2270 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2271 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2272 uh->source, saddr, dif, sdif, hnum)) 2273 continue; 2274 2275 if (!first) { 2276 first = sk; 2277 continue; 2278 } 2279 nskb = skb_clone(skb, GFP_ATOMIC); 2280 2281 if (unlikely(!nskb)) { 2282 atomic_inc(&sk->sk_drops); 2283 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2284 IS_UDPLITE(sk)); 2285 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2286 IS_UDPLITE(sk)); 2287 continue; 2288 } 2289 if (udp_queue_rcv_skb(sk, nskb) > 0) 2290 consume_skb(nskb); 2291 } 2292 2293 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2294 if (use_hash2 && hash2 != hash2_any) { 2295 hash2 = hash2_any; 2296 goto start_lookup; 2297 } 2298 2299 if (first) { 2300 if (udp_queue_rcv_skb(first, skb) > 0) 2301 consume_skb(skb); 2302 } else { 2303 kfree_skb(skb); 2304 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2305 proto == IPPROTO_UDPLITE); 2306 } 2307 return 0; 2308 } 2309 2310 /* Initialize UDP checksum. If exited with zero value (success), 2311 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2312 * Otherwise, csum completion requires checksumming packet body, 2313 * including udp header and folding it to skb->csum. 2314 */ 2315 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2316 int proto) 2317 { 2318 int err; 2319 2320 UDP_SKB_CB(skb)->partial_cov = 0; 2321 UDP_SKB_CB(skb)->cscov = skb->len; 2322 2323 if (proto == IPPROTO_UDPLITE) { 2324 err = udplite_checksum_init(skb, uh); 2325 if (err) 2326 return err; 2327 2328 if (UDP_SKB_CB(skb)->partial_cov) { 2329 skb->csum = inet_compute_pseudo(skb, proto); 2330 return 0; 2331 } 2332 } 2333 2334 /* Note, we are only interested in != 0 or == 0, thus the 2335 * force to int. 2336 */ 2337 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2338 inet_compute_pseudo); 2339 if (err) 2340 return err; 2341 2342 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2343 /* If SW calculated the value, we know it's bad */ 2344 if (skb->csum_complete_sw) 2345 return 1; 2346 2347 /* HW says the value is bad. Let's validate that. 2348 * skb->csum is no longer the full packet checksum, 2349 * so don't treat it as such. 2350 */ 2351 skb_checksum_complete_unset(skb); 2352 } 2353 2354 return 0; 2355 } 2356 2357 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2358 * return code conversion for ip layer consumption 2359 */ 2360 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2361 struct udphdr *uh) 2362 { 2363 int ret; 2364 2365 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2366 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2367 2368 ret = udp_queue_rcv_skb(sk, skb); 2369 2370 /* a return value > 0 means to resubmit the input, but 2371 * it wants the return to be -protocol, or 0 2372 */ 2373 if (ret > 0) 2374 return -ret; 2375 return 0; 2376 } 2377 2378 /* 2379 * All we need to do is get the socket, and then do a checksum. 2380 */ 2381 2382 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2383 int proto) 2384 { 2385 struct sock *sk; 2386 struct udphdr *uh; 2387 unsigned short ulen; 2388 struct rtable *rt = skb_rtable(skb); 2389 __be32 saddr, daddr; 2390 struct net *net = dev_net(skb->dev); 2391 bool refcounted; 2392 int drop_reason; 2393 2394 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2395 2396 /* 2397 * Validate the packet. 2398 */ 2399 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2400 goto drop; /* No space for header. */ 2401 2402 uh = udp_hdr(skb); 2403 ulen = ntohs(uh->len); 2404 saddr = ip_hdr(skb)->saddr; 2405 daddr = ip_hdr(skb)->daddr; 2406 2407 if (ulen > skb->len) 2408 goto short_packet; 2409 2410 if (proto == IPPROTO_UDP) { 2411 /* UDP validates ulen. */ 2412 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2413 goto short_packet; 2414 uh = udp_hdr(skb); 2415 } 2416 2417 if (udp4_csum_init(skb, uh, proto)) 2418 goto csum_error; 2419 2420 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 2421 &refcounted, udp_ehashfn); 2422 if (IS_ERR(sk)) 2423 goto no_sk; 2424 2425 if (sk) { 2426 struct dst_entry *dst = skb_dst(skb); 2427 int ret; 2428 2429 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2430 udp_sk_rx_dst_set(sk, dst); 2431 2432 ret = udp_unicast_rcv_skb(sk, skb, uh); 2433 if (refcounted) 2434 sock_put(sk); 2435 return ret; 2436 } 2437 2438 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2439 return __udp4_lib_mcast_deliver(net, skb, uh, 2440 saddr, daddr, udptable, proto); 2441 2442 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2443 if (sk) 2444 return udp_unicast_rcv_skb(sk, skb, uh); 2445 no_sk: 2446 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2447 goto drop; 2448 nf_reset_ct(skb); 2449 2450 /* No socket. Drop packet silently, if checksum is wrong */ 2451 if (udp_lib_checksum_complete(skb)) 2452 goto csum_error; 2453 2454 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2455 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2456 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2457 2458 /* 2459 * Hmm. We got an UDP packet to a port to which we 2460 * don't wanna listen. Ignore it. 2461 */ 2462 kfree_skb_reason(skb, drop_reason); 2463 return 0; 2464 2465 short_packet: 2466 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2467 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2468 proto == IPPROTO_UDPLITE ? "Lite" : "", 2469 &saddr, ntohs(uh->source), 2470 ulen, skb->len, 2471 &daddr, ntohs(uh->dest)); 2472 goto drop; 2473 2474 csum_error: 2475 /* 2476 * RFC1122: OK. Discards the bad packet silently (as far as 2477 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2478 */ 2479 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2480 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2481 proto == IPPROTO_UDPLITE ? "Lite" : "", 2482 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2483 ulen); 2484 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2485 drop: 2486 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2487 kfree_skb_reason(skb, drop_reason); 2488 return 0; 2489 } 2490 2491 /* We can only early demux multicast if there is a single matching socket. 2492 * If more than one socket found returns NULL 2493 */ 2494 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2495 __be16 loc_port, __be32 loc_addr, 2496 __be16 rmt_port, __be32 rmt_addr, 2497 int dif, int sdif) 2498 { 2499 struct udp_table *udptable = net->ipv4.udp_table; 2500 unsigned short hnum = ntohs(loc_port); 2501 struct sock *sk, *result; 2502 struct udp_hslot *hslot; 2503 unsigned int slot; 2504 2505 slot = udp_hashfn(net, hnum, udptable->mask); 2506 hslot = &udptable->hash[slot]; 2507 2508 /* Do not bother scanning a too big list */ 2509 if (hslot->count > 10) 2510 return NULL; 2511 2512 result = NULL; 2513 sk_for_each_rcu(sk, &hslot->head) { 2514 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2515 rmt_port, rmt_addr, dif, sdif, hnum)) { 2516 if (result) 2517 return NULL; 2518 result = sk; 2519 } 2520 } 2521 2522 return result; 2523 } 2524 2525 /* For unicast we should only early demux connected sockets or we can 2526 * break forwarding setups. The chains here can be long so only check 2527 * if the first socket is an exact match and if not move on. 2528 */ 2529 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2530 __be16 loc_port, __be32 loc_addr, 2531 __be16 rmt_port, __be32 rmt_addr, 2532 int dif, int sdif) 2533 { 2534 struct udp_table *udptable = net->ipv4.udp_table; 2535 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2536 unsigned short hnum = ntohs(loc_port); 2537 unsigned int hash2, slot2; 2538 struct udp_hslot *hslot2; 2539 __portpair ports; 2540 struct sock *sk; 2541 2542 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2543 slot2 = hash2 & udptable->mask; 2544 hslot2 = &udptable->hash2[slot2]; 2545 ports = INET_COMBINED_PORTS(rmt_port, hnum); 2546 2547 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2548 if (inet_match(net, sk, acookie, ports, dif, sdif)) 2549 return sk; 2550 /* Only check first socket in chain */ 2551 break; 2552 } 2553 return NULL; 2554 } 2555 2556 int udp_v4_early_demux(struct sk_buff *skb) 2557 { 2558 struct net *net = dev_net(skb->dev); 2559 struct in_device *in_dev = NULL; 2560 const struct iphdr *iph; 2561 const struct udphdr *uh; 2562 struct sock *sk = NULL; 2563 struct dst_entry *dst; 2564 int dif = skb->dev->ifindex; 2565 int sdif = inet_sdif(skb); 2566 int ours; 2567 2568 /* validate the packet */ 2569 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2570 return 0; 2571 2572 iph = ip_hdr(skb); 2573 uh = udp_hdr(skb); 2574 2575 if (skb->pkt_type == PACKET_MULTICAST) { 2576 in_dev = __in_dev_get_rcu(skb->dev); 2577 2578 if (!in_dev) 2579 return 0; 2580 2581 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2582 iph->protocol); 2583 if (!ours) 2584 return 0; 2585 2586 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2587 uh->source, iph->saddr, 2588 dif, sdif); 2589 } else if (skb->pkt_type == PACKET_HOST) { 2590 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2591 uh->source, iph->saddr, dif, sdif); 2592 } 2593 2594 if (!sk) 2595 return 0; 2596 2597 skb->sk = sk; 2598 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk)); 2599 skb->destructor = sock_pfree; 2600 dst = rcu_dereference(sk->sk_rx_dst); 2601 2602 if (dst) 2603 dst = dst_check(dst, 0); 2604 if (dst) { 2605 u32 itag = 0; 2606 2607 /* set noref for now. 2608 * any place which wants to hold dst has to call 2609 * dst_hold_safe() 2610 */ 2611 skb_dst_set_noref(skb, dst); 2612 2613 /* for unconnected multicast sockets we need to validate 2614 * the source on each packet 2615 */ 2616 if (!inet_sk(sk)->inet_daddr && in_dev) 2617 return ip_mc_validate_source(skb, iph->daddr, 2618 iph->saddr, 2619 iph->tos & IPTOS_RT_MASK, 2620 skb->dev, in_dev, &itag); 2621 } 2622 return 0; 2623 } 2624 2625 int udp_rcv(struct sk_buff *skb) 2626 { 2627 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 2628 } 2629 2630 void udp_destroy_sock(struct sock *sk) 2631 { 2632 struct udp_sock *up = udp_sk(sk); 2633 bool slow = lock_sock_fast(sk); 2634 2635 /* protects from races with udp_abort() */ 2636 sock_set_flag(sk, SOCK_DEAD); 2637 udp_flush_pending_frames(sk); 2638 unlock_sock_fast(sk, slow); 2639 if (static_branch_unlikely(&udp_encap_needed_key)) { 2640 if (up->encap_type) { 2641 void (*encap_destroy)(struct sock *sk); 2642 encap_destroy = READ_ONCE(up->encap_destroy); 2643 if (encap_destroy) 2644 encap_destroy(sk); 2645 } 2646 if (udp_test_bit(ENCAP_ENABLED, sk)) 2647 static_branch_dec(&udp_encap_needed_key); 2648 } 2649 } 2650 2651 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family, 2652 struct sock *sk) 2653 { 2654 #ifdef CONFIG_XFRM 2655 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) { 2656 if (family == AF_INET) 2657 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv); 2658 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) 2659 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv); 2660 } 2661 #endif 2662 } 2663 2664 /* 2665 * Socket option code for UDP 2666 */ 2667 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2668 sockptr_t optval, unsigned int optlen, 2669 int (*push_pending_frames)(struct sock *)) 2670 { 2671 struct udp_sock *up = udp_sk(sk); 2672 int val, valbool; 2673 int err = 0; 2674 int is_udplite = IS_UDPLITE(sk); 2675 2676 if (level == SOL_SOCKET) { 2677 err = sk_setsockopt(sk, level, optname, optval, optlen); 2678 2679 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) { 2680 sockopt_lock_sock(sk); 2681 /* paired with READ_ONCE in udp_rmem_release() */ 2682 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2); 2683 sockopt_release_sock(sk); 2684 } 2685 return err; 2686 } 2687 2688 if (optlen < sizeof(int)) 2689 return -EINVAL; 2690 2691 if (copy_from_sockptr(&val, optval, sizeof(val))) 2692 return -EFAULT; 2693 2694 valbool = val ? 1 : 0; 2695 2696 switch (optname) { 2697 case UDP_CORK: 2698 if (val != 0) { 2699 udp_set_bit(CORK, sk); 2700 } else { 2701 udp_clear_bit(CORK, sk); 2702 lock_sock(sk); 2703 push_pending_frames(sk); 2704 release_sock(sk); 2705 } 2706 break; 2707 2708 case UDP_ENCAP: 2709 switch (val) { 2710 case 0: 2711 #ifdef CONFIG_XFRM 2712 case UDP_ENCAP_ESPINUDP: 2713 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk); 2714 fallthrough; 2715 case UDP_ENCAP_ESPINUDP_NON_IKE: 2716 #if IS_ENABLED(CONFIG_IPV6) 2717 if (sk->sk_family == AF_INET6) 2718 WRITE_ONCE(up->encap_rcv, 2719 ipv6_stub->xfrm6_udp_encap_rcv); 2720 else 2721 #endif 2722 WRITE_ONCE(up->encap_rcv, 2723 xfrm4_udp_encap_rcv); 2724 #endif 2725 fallthrough; 2726 case UDP_ENCAP_L2TPINUDP: 2727 WRITE_ONCE(up->encap_type, val); 2728 udp_tunnel_encap_enable(sk); 2729 break; 2730 default: 2731 err = -ENOPROTOOPT; 2732 break; 2733 } 2734 break; 2735 2736 case UDP_NO_CHECK6_TX: 2737 udp_set_no_check6_tx(sk, valbool); 2738 break; 2739 2740 case UDP_NO_CHECK6_RX: 2741 udp_set_no_check6_rx(sk, valbool); 2742 break; 2743 2744 case UDP_SEGMENT: 2745 if (val < 0 || val > USHRT_MAX) 2746 return -EINVAL; 2747 WRITE_ONCE(up->gso_size, val); 2748 break; 2749 2750 case UDP_GRO: 2751 2752 /* when enabling GRO, accept the related GSO packet type */ 2753 if (valbool) 2754 udp_tunnel_encap_enable(sk); 2755 udp_assign_bit(GRO_ENABLED, sk, valbool); 2756 udp_assign_bit(ACCEPT_L4, sk, valbool); 2757 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk); 2758 break; 2759 2760 /* 2761 * UDP-Lite's partial checksum coverage (RFC 3828). 2762 */ 2763 /* The sender sets actual checksum coverage length via this option. 2764 * The case coverage > packet length is handled by send module. */ 2765 case UDPLITE_SEND_CSCOV: 2766 if (!is_udplite) /* Disable the option on UDP sockets */ 2767 return -ENOPROTOOPT; 2768 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2769 val = 8; 2770 else if (val > USHRT_MAX) 2771 val = USHRT_MAX; 2772 WRITE_ONCE(up->pcslen, val); 2773 udp_set_bit(UDPLITE_SEND_CC, sk); 2774 break; 2775 2776 /* The receiver specifies a minimum checksum coverage value. To make 2777 * sense, this should be set to at least 8 (as done below). If zero is 2778 * used, this again means full checksum coverage. */ 2779 case UDPLITE_RECV_CSCOV: 2780 if (!is_udplite) /* Disable the option on UDP sockets */ 2781 return -ENOPROTOOPT; 2782 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2783 val = 8; 2784 else if (val > USHRT_MAX) 2785 val = USHRT_MAX; 2786 WRITE_ONCE(up->pcrlen, val); 2787 udp_set_bit(UDPLITE_RECV_CC, sk); 2788 break; 2789 2790 default: 2791 err = -ENOPROTOOPT; 2792 break; 2793 } 2794 2795 return err; 2796 } 2797 EXPORT_SYMBOL(udp_lib_setsockopt); 2798 2799 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2800 unsigned int optlen) 2801 { 2802 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 2803 return udp_lib_setsockopt(sk, level, optname, 2804 optval, optlen, 2805 udp_push_pending_frames); 2806 return ip_setsockopt(sk, level, optname, optval, optlen); 2807 } 2808 2809 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2810 char __user *optval, int __user *optlen) 2811 { 2812 struct udp_sock *up = udp_sk(sk); 2813 int val, len; 2814 2815 if (get_user(len, optlen)) 2816 return -EFAULT; 2817 2818 if (len < 0) 2819 return -EINVAL; 2820 2821 len = min_t(unsigned int, len, sizeof(int)); 2822 2823 switch (optname) { 2824 case UDP_CORK: 2825 val = udp_test_bit(CORK, sk); 2826 break; 2827 2828 case UDP_ENCAP: 2829 val = READ_ONCE(up->encap_type); 2830 break; 2831 2832 case UDP_NO_CHECK6_TX: 2833 val = udp_get_no_check6_tx(sk); 2834 break; 2835 2836 case UDP_NO_CHECK6_RX: 2837 val = udp_get_no_check6_rx(sk); 2838 break; 2839 2840 case UDP_SEGMENT: 2841 val = READ_ONCE(up->gso_size); 2842 break; 2843 2844 case UDP_GRO: 2845 val = udp_test_bit(GRO_ENABLED, sk); 2846 break; 2847 2848 /* The following two cannot be changed on UDP sockets, the return is 2849 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2850 case UDPLITE_SEND_CSCOV: 2851 val = READ_ONCE(up->pcslen); 2852 break; 2853 2854 case UDPLITE_RECV_CSCOV: 2855 val = READ_ONCE(up->pcrlen); 2856 break; 2857 2858 default: 2859 return -ENOPROTOOPT; 2860 } 2861 2862 if (put_user(len, optlen)) 2863 return -EFAULT; 2864 if (copy_to_user(optval, &val, len)) 2865 return -EFAULT; 2866 return 0; 2867 } 2868 EXPORT_SYMBOL(udp_lib_getsockopt); 2869 2870 int udp_getsockopt(struct sock *sk, int level, int optname, 2871 char __user *optval, int __user *optlen) 2872 { 2873 if (level == SOL_UDP || level == SOL_UDPLITE) 2874 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2875 return ip_getsockopt(sk, level, optname, optval, optlen); 2876 } 2877 2878 /** 2879 * udp_poll - wait for a UDP event. 2880 * @file: - file struct 2881 * @sock: - socket 2882 * @wait: - poll table 2883 * 2884 * This is same as datagram poll, except for the special case of 2885 * blocking sockets. If application is using a blocking fd 2886 * and a packet with checksum error is in the queue; 2887 * then it could get return from select indicating data available 2888 * but then block when reading it. Add special case code 2889 * to work around these arguably broken applications. 2890 */ 2891 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2892 { 2893 __poll_t mask = datagram_poll(file, sock, wait); 2894 struct sock *sk = sock->sk; 2895 2896 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2897 mask |= EPOLLIN | EPOLLRDNORM; 2898 2899 /* Check for false positives due to checksum errors */ 2900 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2901 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2902 mask &= ~(EPOLLIN | EPOLLRDNORM); 2903 2904 /* psock ingress_msg queue should not contain any bad checksum frames */ 2905 if (sk_is_readable(sk)) 2906 mask |= EPOLLIN | EPOLLRDNORM; 2907 return mask; 2908 2909 } 2910 EXPORT_SYMBOL(udp_poll); 2911 2912 int udp_abort(struct sock *sk, int err) 2913 { 2914 if (!has_current_bpf_ctx()) 2915 lock_sock(sk); 2916 2917 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2918 * with close() 2919 */ 2920 if (sock_flag(sk, SOCK_DEAD)) 2921 goto out; 2922 2923 sk->sk_err = err; 2924 sk_error_report(sk); 2925 __udp_disconnect(sk, 0); 2926 2927 out: 2928 if (!has_current_bpf_ctx()) 2929 release_sock(sk); 2930 2931 return 0; 2932 } 2933 EXPORT_SYMBOL_GPL(udp_abort); 2934 2935 struct proto udp_prot = { 2936 .name = "UDP", 2937 .owner = THIS_MODULE, 2938 .close = udp_lib_close, 2939 .pre_connect = udp_pre_connect, 2940 .connect = ip4_datagram_connect, 2941 .disconnect = udp_disconnect, 2942 .ioctl = udp_ioctl, 2943 .init = udp_init_sock, 2944 .destroy = udp_destroy_sock, 2945 .setsockopt = udp_setsockopt, 2946 .getsockopt = udp_getsockopt, 2947 .sendmsg = udp_sendmsg, 2948 .recvmsg = udp_recvmsg, 2949 .splice_eof = udp_splice_eof, 2950 .release_cb = ip4_datagram_release_cb, 2951 .hash = udp_lib_hash, 2952 .unhash = udp_lib_unhash, 2953 .rehash = udp_v4_rehash, 2954 .get_port = udp_v4_get_port, 2955 .put_port = udp_lib_unhash, 2956 #ifdef CONFIG_BPF_SYSCALL 2957 .psock_update_sk_prot = udp_bpf_update_proto, 2958 #endif 2959 .memory_allocated = &udp_memory_allocated, 2960 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 2961 2962 .sysctl_mem = sysctl_udp_mem, 2963 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2964 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2965 .obj_size = sizeof(struct udp_sock), 2966 .h.udp_table = NULL, 2967 .diag_destroy = udp_abort, 2968 }; 2969 EXPORT_SYMBOL(udp_prot); 2970 2971 /* ------------------------------------------------------------------------ */ 2972 #ifdef CONFIG_PROC_FS 2973 2974 static unsigned short seq_file_family(const struct seq_file *seq); 2975 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk) 2976 { 2977 unsigned short family = seq_file_family(seq); 2978 2979 /* AF_UNSPEC is used as a match all */ 2980 return ((family == AF_UNSPEC || family == sk->sk_family) && 2981 net_eq(sock_net(sk), seq_file_net(seq))); 2982 } 2983 2984 #ifdef CONFIG_BPF_SYSCALL 2985 static const struct seq_operations bpf_iter_udp_seq_ops; 2986 #endif 2987 static struct udp_table *udp_get_table_seq(struct seq_file *seq, 2988 struct net *net) 2989 { 2990 const struct udp_seq_afinfo *afinfo; 2991 2992 #ifdef CONFIG_BPF_SYSCALL 2993 if (seq->op == &bpf_iter_udp_seq_ops) 2994 return net->ipv4.udp_table; 2995 #endif 2996 2997 afinfo = pde_data(file_inode(seq->file)); 2998 return afinfo->udp_table ? : net->ipv4.udp_table; 2999 } 3000 3001 static struct sock *udp_get_first(struct seq_file *seq, int start) 3002 { 3003 struct udp_iter_state *state = seq->private; 3004 struct net *net = seq_file_net(seq); 3005 struct udp_table *udptable; 3006 struct sock *sk; 3007 3008 udptable = udp_get_table_seq(seq, net); 3009 3010 for (state->bucket = start; state->bucket <= udptable->mask; 3011 ++state->bucket) { 3012 struct udp_hslot *hslot = &udptable->hash[state->bucket]; 3013 3014 if (hlist_empty(&hslot->head)) 3015 continue; 3016 3017 spin_lock_bh(&hslot->lock); 3018 sk_for_each(sk, &hslot->head) { 3019 if (seq_sk_match(seq, sk)) 3020 goto found; 3021 } 3022 spin_unlock_bh(&hslot->lock); 3023 } 3024 sk = NULL; 3025 found: 3026 return sk; 3027 } 3028 3029 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 3030 { 3031 struct udp_iter_state *state = seq->private; 3032 struct net *net = seq_file_net(seq); 3033 struct udp_table *udptable; 3034 3035 do { 3036 sk = sk_next(sk); 3037 } while (sk && !seq_sk_match(seq, sk)); 3038 3039 if (!sk) { 3040 udptable = udp_get_table_seq(seq, net); 3041 3042 if (state->bucket <= udptable->mask) 3043 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3044 3045 return udp_get_first(seq, state->bucket + 1); 3046 } 3047 return sk; 3048 } 3049 3050 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3051 { 3052 struct sock *sk = udp_get_first(seq, 0); 3053 3054 if (sk) 3055 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3056 --pos; 3057 return pos ? NULL : sk; 3058 } 3059 3060 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3061 { 3062 struct udp_iter_state *state = seq->private; 3063 state->bucket = MAX_UDP_PORTS; 3064 3065 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3066 } 3067 EXPORT_SYMBOL(udp_seq_start); 3068 3069 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3070 { 3071 struct sock *sk; 3072 3073 if (v == SEQ_START_TOKEN) 3074 sk = udp_get_idx(seq, 0); 3075 else 3076 sk = udp_get_next(seq, v); 3077 3078 ++*pos; 3079 return sk; 3080 } 3081 EXPORT_SYMBOL(udp_seq_next); 3082 3083 void udp_seq_stop(struct seq_file *seq, void *v) 3084 { 3085 struct udp_iter_state *state = seq->private; 3086 struct udp_table *udptable; 3087 3088 udptable = udp_get_table_seq(seq, seq_file_net(seq)); 3089 3090 if (state->bucket <= udptable->mask) 3091 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3092 } 3093 EXPORT_SYMBOL(udp_seq_stop); 3094 3095 /* ------------------------------------------------------------------------ */ 3096 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3097 int bucket) 3098 { 3099 struct inet_sock *inet = inet_sk(sp); 3100 __be32 dest = inet->inet_daddr; 3101 __be32 src = inet->inet_rcv_saddr; 3102 __u16 destp = ntohs(inet->inet_dport); 3103 __u16 srcp = ntohs(inet->inet_sport); 3104 3105 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3106 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3107 bucket, src, srcp, dest, destp, sp->sk_state, 3108 sk_wmem_alloc_get(sp), 3109 udp_rqueue_get(sp), 3110 0, 0L, 0, 3111 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3112 0, sock_i_ino(sp), 3113 refcount_read(&sp->sk_refcnt), sp, 3114 atomic_read(&sp->sk_drops)); 3115 } 3116 3117 int udp4_seq_show(struct seq_file *seq, void *v) 3118 { 3119 seq_setwidth(seq, 127); 3120 if (v == SEQ_START_TOKEN) 3121 seq_puts(seq, " sl local_address rem_address st tx_queue " 3122 "rx_queue tr tm->when retrnsmt uid timeout " 3123 "inode ref pointer drops"); 3124 else { 3125 struct udp_iter_state *state = seq->private; 3126 3127 udp4_format_sock(v, seq, state->bucket); 3128 } 3129 seq_pad(seq, '\n'); 3130 return 0; 3131 } 3132 3133 #ifdef CONFIG_BPF_SYSCALL 3134 struct bpf_iter__udp { 3135 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3136 __bpf_md_ptr(struct udp_sock *, udp_sk); 3137 uid_t uid __aligned(8); 3138 int bucket __aligned(8); 3139 }; 3140 3141 struct bpf_udp_iter_state { 3142 struct udp_iter_state state; 3143 unsigned int cur_sk; 3144 unsigned int end_sk; 3145 unsigned int max_sk; 3146 int offset; 3147 struct sock **batch; 3148 bool st_bucket_done; 3149 }; 3150 3151 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3152 unsigned int new_batch_sz); 3153 static struct sock *bpf_iter_udp_batch(struct seq_file *seq) 3154 { 3155 struct bpf_udp_iter_state *iter = seq->private; 3156 struct udp_iter_state *state = &iter->state; 3157 struct net *net = seq_file_net(seq); 3158 int resume_bucket, resume_offset; 3159 struct udp_table *udptable; 3160 unsigned int batch_sks = 0; 3161 bool resized = false; 3162 struct sock *sk; 3163 3164 resume_bucket = state->bucket; 3165 resume_offset = iter->offset; 3166 3167 /* The current batch is done, so advance the bucket. */ 3168 if (iter->st_bucket_done) 3169 state->bucket++; 3170 3171 udptable = udp_get_table_seq(seq, net); 3172 3173 again: 3174 /* New batch for the next bucket. 3175 * Iterate over the hash table to find a bucket with sockets matching 3176 * the iterator attributes, and return the first matching socket from 3177 * the bucket. The remaining matched sockets from the bucket are batched 3178 * before releasing the bucket lock. This allows BPF programs that are 3179 * called in seq_show to acquire the bucket lock if needed. 3180 */ 3181 iter->cur_sk = 0; 3182 iter->end_sk = 0; 3183 iter->st_bucket_done = false; 3184 batch_sks = 0; 3185 3186 for (; state->bucket <= udptable->mask; state->bucket++) { 3187 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket]; 3188 3189 if (hlist_empty(&hslot2->head)) 3190 continue; 3191 3192 iter->offset = 0; 3193 spin_lock_bh(&hslot2->lock); 3194 udp_portaddr_for_each_entry(sk, &hslot2->head) { 3195 if (seq_sk_match(seq, sk)) { 3196 /* Resume from the last iterated socket at the 3197 * offset in the bucket before iterator was stopped. 3198 */ 3199 if (state->bucket == resume_bucket && 3200 iter->offset < resume_offset) { 3201 ++iter->offset; 3202 continue; 3203 } 3204 if (iter->end_sk < iter->max_sk) { 3205 sock_hold(sk); 3206 iter->batch[iter->end_sk++] = sk; 3207 } 3208 batch_sks++; 3209 } 3210 } 3211 spin_unlock_bh(&hslot2->lock); 3212 3213 if (iter->end_sk) 3214 break; 3215 } 3216 3217 /* All done: no batch made. */ 3218 if (!iter->end_sk) 3219 return NULL; 3220 3221 if (iter->end_sk == batch_sks) { 3222 /* Batching is done for the current bucket; return the first 3223 * socket to be iterated from the batch. 3224 */ 3225 iter->st_bucket_done = true; 3226 goto done; 3227 } 3228 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) { 3229 resized = true; 3230 /* After allocating a larger batch, retry one more time to grab 3231 * the whole bucket. 3232 */ 3233 goto again; 3234 } 3235 done: 3236 return iter->batch[0]; 3237 } 3238 3239 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3240 { 3241 struct bpf_udp_iter_state *iter = seq->private; 3242 struct sock *sk; 3243 3244 /* Whenever seq_next() is called, the iter->cur_sk is 3245 * done with seq_show(), so unref the iter->cur_sk. 3246 */ 3247 if (iter->cur_sk < iter->end_sk) { 3248 sock_put(iter->batch[iter->cur_sk++]); 3249 ++iter->offset; 3250 } 3251 3252 /* After updating iter->cur_sk, check if there are more sockets 3253 * available in the current bucket batch. 3254 */ 3255 if (iter->cur_sk < iter->end_sk) 3256 sk = iter->batch[iter->cur_sk]; 3257 else 3258 /* Prepare a new batch. */ 3259 sk = bpf_iter_udp_batch(seq); 3260 3261 ++*pos; 3262 return sk; 3263 } 3264 3265 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos) 3266 { 3267 /* bpf iter does not support lseek, so it always 3268 * continue from where it was stop()-ped. 3269 */ 3270 if (*pos) 3271 return bpf_iter_udp_batch(seq); 3272 3273 return SEQ_START_TOKEN; 3274 } 3275 3276 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3277 struct udp_sock *udp_sk, uid_t uid, int bucket) 3278 { 3279 struct bpf_iter__udp ctx; 3280 3281 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3282 ctx.meta = meta; 3283 ctx.udp_sk = udp_sk; 3284 ctx.uid = uid; 3285 ctx.bucket = bucket; 3286 return bpf_iter_run_prog(prog, &ctx); 3287 } 3288 3289 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3290 { 3291 struct udp_iter_state *state = seq->private; 3292 struct bpf_iter_meta meta; 3293 struct bpf_prog *prog; 3294 struct sock *sk = v; 3295 uid_t uid; 3296 int ret; 3297 3298 if (v == SEQ_START_TOKEN) 3299 return 0; 3300 3301 lock_sock(sk); 3302 3303 if (unlikely(sk_unhashed(sk))) { 3304 ret = SEQ_SKIP; 3305 goto unlock; 3306 } 3307 3308 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3309 meta.seq = seq; 3310 prog = bpf_iter_get_info(&meta, false); 3311 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3312 3313 unlock: 3314 release_sock(sk); 3315 return ret; 3316 } 3317 3318 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter) 3319 { 3320 while (iter->cur_sk < iter->end_sk) 3321 sock_put(iter->batch[iter->cur_sk++]); 3322 } 3323 3324 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3325 { 3326 struct bpf_udp_iter_state *iter = seq->private; 3327 struct bpf_iter_meta meta; 3328 struct bpf_prog *prog; 3329 3330 if (!v) { 3331 meta.seq = seq; 3332 prog = bpf_iter_get_info(&meta, true); 3333 if (prog) 3334 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3335 } 3336 3337 if (iter->cur_sk < iter->end_sk) { 3338 bpf_iter_udp_put_batch(iter); 3339 iter->st_bucket_done = false; 3340 } 3341 } 3342 3343 static const struct seq_operations bpf_iter_udp_seq_ops = { 3344 .start = bpf_iter_udp_seq_start, 3345 .next = bpf_iter_udp_seq_next, 3346 .stop = bpf_iter_udp_seq_stop, 3347 .show = bpf_iter_udp_seq_show, 3348 }; 3349 #endif 3350 3351 static unsigned short seq_file_family(const struct seq_file *seq) 3352 { 3353 const struct udp_seq_afinfo *afinfo; 3354 3355 #ifdef CONFIG_BPF_SYSCALL 3356 /* BPF iterator: bpf programs to filter sockets. */ 3357 if (seq->op == &bpf_iter_udp_seq_ops) 3358 return AF_UNSPEC; 3359 #endif 3360 3361 /* Proc fs iterator */ 3362 afinfo = pde_data(file_inode(seq->file)); 3363 return afinfo->family; 3364 } 3365 3366 const struct seq_operations udp_seq_ops = { 3367 .start = udp_seq_start, 3368 .next = udp_seq_next, 3369 .stop = udp_seq_stop, 3370 .show = udp4_seq_show, 3371 }; 3372 EXPORT_SYMBOL(udp_seq_ops); 3373 3374 static struct udp_seq_afinfo udp4_seq_afinfo = { 3375 .family = AF_INET, 3376 .udp_table = NULL, 3377 }; 3378 3379 static int __net_init udp4_proc_init_net(struct net *net) 3380 { 3381 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3382 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3383 return -ENOMEM; 3384 return 0; 3385 } 3386 3387 static void __net_exit udp4_proc_exit_net(struct net *net) 3388 { 3389 remove_proc_entry("udp", net->proc_net); 3390 } 3391 3392 static struct pernet_operations udp4_net_ops = { 3393 .init = udp4_proc_init_net, 3394 .exit = udp4_proc_exit_net, 3395 }; 3396 3397 int __init udp4_proc_init(void) 3398 { 3399 return register_pernet_subsys(&udp4_net_ops); 3400 } 3401 3402 void udp4_proc_exit(void) 3403 { 3404 unregister_pernet_subsys(&udp4_net_ops); 3405 } 3406 #endif /* CONFIG_PROC_FS */ 3407 3408 static __initdata unsigned long uhash_entries; 3409 static int __init set_uhash_entries(char *str) 3410 { 3411 ssize_t ret; 3412 3413 if (!str) 3414 return 0; 3415 3416 ret = kstrtoul(str, 0, &uhash_entries); 3417 if (ret) 3418 return 0; 3419 3420 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3421 uhash_entries = UDP_HTABLE_SIZE_MIN; 3422 return 1; 3423 } 3424 __setup("uhash_entries=", set_uhash_entries); 3425 3426 void __init udp_table_init(struct udp_table *table, const char *name) 3427 { 3428 unsigned int i; 3429 3430 table->hash = alloc_large_system_hash(name, 3431 2 * sizeof(struct udp_hslot), 3432 uhash_entries, 3433 21, /* one slot per 2 MB */ 3434 0, 3435 &table->log, 3436 &table->mask, 3437 UDP_HTABLE_SIZE_MIN, 3438 UDP_HTABLE_SIZE_MAX); 3439 3440 table->hash2 = table->hash + (table->mask + 1); 3441 for (i = 0; i <= table->mask; i++) { 3442 INIT_HLIST_HEAD(&table->hash[i].head); 3443 table->hash[i].count = 0; 3444 spin_lock_init(&table->hash[i].lock); 3445 } 3446 for (i = 0; i <= table->mask; i++) { 3447 INIT_HLIST_HEAD(&table->hash2[i].head); 3448 table->hash2[i].count = 0; 3449 spin_lock_init(&table->hash2[i].lock); 3450 } 3451 } 3452 3453 u32 udp_flow_hashrnd(void) 3454 { 3455 static u32 hashrnd __read_mostly; 3456 3457 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3458 3459 return hashrnd; 3460 } 3461 EXPORT_SYMBOL(udp_flow_hashrnd); 3462 3463 static void __net_init udp_sysctl_init(struct net *net) 3464 { 3465 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3466 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 3467 3468 #ifdef CONFIG_NET_L3_MASTER_DEV 3469 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3470 #endif 3471 } 3472 3473 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries) 3474 { 3475 struct udp_table *udptable; 3476 int i; 3477 3478 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL); 3479 if (!udptable) 3480 goto out; 3481 3482 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot), 3483 GFP_KERNEL_ACCOUNT); 3484 if (!udptable->hash) 3485 goto free_table; 3486 3487 udptable->hash2 = udptable->hash + hash_entries; 3488 udptable->mask = hash_entries - 1; 3489 udptable->log = ilog2(hash_entries); 3490 3491 for (i = 0; i < hash_entries; i++) { 3492 INIT_HLIST_HEAD(&udptable->hash[i].head); 3493 udptable->hash[i].count = 0; 3494 spin_lock_init(&udptable->hash[i].lock); 3495 3496 INIT_HLIST_HEAD(&udptable->hash2[i].head); 3497 udptable->hash2[i].count = 0; 3498 spin_lock_init(&udptable->hash2[i].lock); 3499 } 3500 3501 return udptable; 3502 3503 free_table: 3504 kfree(udptable); 3505 out: 3506 return NULL; 3507 } 3508 3509 static void __net_exit udp_pernet_table_free(struct net *net) 3510 { 3511 struct udp_table *udptable = net->ipv4.udp_table; 3512 3513 if (udptable == &udp_table) 3514 return; 3515 3516 kvfree(udptable->hash); 3517 kfree(udptable); 3518 } 3519 3520 static void __net_init udp_set_table(struct net *net) 3521 { 3522 struct udp_table *udptable; 3523 unsigned int hash_entries; 3524 struct net *old_net; 3525 3526 if (net_eq(net, &init_net)) 3527 goto fallback; 3528 3529 old_net = current->nsproxy->net_ns; 3530 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries); 3531 if (!hash_entries) 3532 goto fallback; 3533 3534 /* Set min to keep the bitmap on stack in udp_lib_get_port() */ 3535 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET) 3536 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET; 3537 else 3538 hash_entries = roundup_pow_of_two(hash_entries); 3539 3540 udptable = udp_pernet_table_alloc(hash_entries); 3541 if (udptable) { 3542 net->ipv4.udp_table = udptable; 3543 } else { 3544 pr_warn("Failed to allocate UDP hash table (entries: %u) " 3545 "for a netns, fallback to the global one\n", 3546 hash_entries); 3547 fallback: 3548 net->ipv4.udp_table = &udp_table; 3549 } 3550 } 3551 3552 static int __net_init udp_pernet_init(struct net *net) 3553 { 3554 udp_sysctl_init(net); 3555 udp_set_table(net); 3556 3557 return 0; 3558 } 3559 3560 static void __net_exit udp_pernet_exit(struct net *net) 3561 { 3562 udp_pernet_table_free(net); 3563 } 3564 3565 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3566 .init = udp_pernet_init, 3567 .exit = udp_pernet_exit, 3568 }; 3569 3570 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3571 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3572 struct udp_sock *udp_sk, uid_t uid, int bucket) 3573 3574 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3575 unsigned int new_batch_sz) 3576 { 3577 struct sock **new_batch; 3578 3579 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch), 3580 GFP_USER | __GFP_NOWARN); 3581 if (!new_batch) 3582 return -ENOMEM; 3583 3584 bpf_iter_udp_put_batch(iter); 3585 kvfree(iter->batch); 3586 iter->batch = new_batch; 3587 iter->max_sk = new_batch_sz; 3588 3589 return 0; 3590 } 3591 3592 #define INIT_BATCH_SZ 16 3593 3594 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3595 { 3596 struct bpf_udp_iter_state *iter = priv_data; 3597 int ret; 3598 3599 ret = bpf_iter_init_seq_net(priv_data, aux); 3600 if (ret) 3601 return ret; 3602 3603 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ); 3604 if (ret) 3605 bpf_iter_fini_seq_net(priv_data); 3606 3607 return ret; 3608 } 3609 3610 static void bpf_iter_fini_udp(void *priv_data) 3611 { 3612 struct bpf_udp_iter_state *iter = priv_data; 3613 3614 bpf_iter_fini_seq_net(priv_data); 3615 kvfree(iter->batch); 3616 } 3617 3618 static const struct bpf_iter_seq_info udp_seq_info = { 3619 .seq_ops = &bpf_iter_udp_seq_ops, 3620 .init_seq_private = bpf_iter_init_udp, 3621 .fini_seq_private = bpf_iter_fini_udp, 3622 .seq_priv_size = sizeof(struct bpf_udp_iter_state), 3623 }; 3624 3625 static struct bpf_iter_reg udp_reg_info = { 3626 .target = "udp", 3627 .ctx_arg_info_size = 1, 3628 .ctx_arg_info = { 3629 { offsetof(struct bpf_iter__udp, udp_sk), 3630 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED }, 3631 }, 3632 .seq_info = &udp_seq_info, 3633 }; 3634 3635 static void __init bpf_iter_register(void) 3636 { 3637 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3638 if (bpf_iter_reg_target(&udp_reg_info)) 3639 pr_warn("Warning: could not register bpf iterator udp\n"); 3640 } 3641 #endif 3642 3643 void __init udp_init(void) 3644 { 3645 unsigned long limit; 3646 unsigned int i; 3647 3648 udp_table_init(&udp_table, "UDP"); 3649 limit = nr_free_buffer_pages() / 8; 3650 limit = max(limit, 128UL); 3651 sysctl_udp_mem[0] = limit / 4 * 3; 3652 sysctl_udp_mem[1] = limit; 3653 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3654 3655 /* 16 spinlocks per cpu */ 3656 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3657 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3658 GFP_KERNEL); 3659 if (!udp_busylocks) 3660 panic("UDP: failed to alloc udp_busylocks\n"); 3661 for (i = 0; i < (1U << udp_busylocks_log); i++) 3662 spin_lock_init(udp_busylocks + i); 3663 3664 if (register_pernet_subsys(&udp_sysctl_ops)) 3665 panic("UDP: failed to init sysctl parameters.\n"); 3666 3667 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3668 bpf_iter_register(); 3669 #endif 3670 } 3671