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