1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * The User Datagram Protocol (UDP). 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 12 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 13 * Hirokazu Takahashi, <taka@valinux.co.jp> 14 * 15 * Fixes: 16 * Alan Cox : verify_area() calls 17 * Alan Cox : stopped close while in use off icmp 18 * messages. Not a fix but a botch that 19 * for udp at least is 'valid'. 20 * Alan Cox : Fixed icmp handling properly 21 * Alan Cox : Correct error for oversized datagrams 22 * Alan Cox : Tidied select() semantics. 23 * Alan Cox : udp_err() fixed properly, also now 24 * select and read wake correctly on errors 25 * Alan Cox : udp_send verify_area moved to avoid mem leak 26 * Alan Cox : UDP can count its memory 27 * Alan Cox : send to an unknown connection causes 28 * an ECONNREFUSED off the icmp, but 29 * does NOT close. 30 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 32 * bug no longer crashes it. 33 * Fred Van Kempen : Net2e support for sk->broadcast. 34 * Alan Cox : Uses skb_free_datagram 35 * Alan Cox : Added get/set sockopt support. 36 * Alan Cox : Broadcasting without option set returns EACCES. 37 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 38 * Alan Cox : Use ip_tos and ip_ttl 39 * Alan Cox : SNMP Mibs 40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 41 * Matt Dillon : UDP length checks. 42 * Alan Cox : Smarter af_inet used properly. 43 * Alan Cox : Use new kernel side addressing. 44 * Alan Cox : Incorrect return on truncated datagram receive. 45 * Arnt Gulbrandsen : New udp_send and stuff 46 * Alan Cox : Cache last socket 47 * Alan Cox : Route cache 48 * Jon Peatfield : Minor efficiency fix to sendto(). 49 * Mike Shaver : RFC1122 checks. 50 * Alan Cox : Nonblocking error fix. 51 * Willy Konynenberg : Transparent proxying support. 52 * Mike McLagan : Routing by source 53 * David S. Miller : New socket lookup architecture. 54 * Last socket cache retained as it 55 * does have a high hit rate. 56 * Olaf Kirch : Don't linearise iovec on sendmsg. 57 * Andi Kleen : Some cleanups, cache destination entry 58 * for connect. 59 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 60 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 61 * return ENOTCONN for unconnected sockets (POSIX) 62 * Janos Farkas : don't deliver multi/broadcasts to a different 63 * bound-to-device socket 64 * Hirokazu Takahashi : HW checksumming for outgoing UDP 65 * datagrams. 66 * Hirokazu Takahashi : sendfile() on UDP works now. 67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 70 * a single port at the same time. 71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 72 * James Chapman : Add L2TP encapsulation type. 73 */ 74 75 #define pr_fmt(fmt) "UDP: " fmt 76 77 #include <linux/bpf-cgroup.h> 78 #include <linux/uaccess.h> 79 #include <asm/ioctls.h> 80 #include <linux/memblock.h> 81 #include <linux/highmem.h> 82 #include <linux/types.h> 83 #include <linux/fcntl.h> 84 #include <linux/module.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/igmp.h> 88 #include <linux/inetdevice.h> 89 #include <linux/in.h> 90 #include <linux/errno.h> 91 #include <linux/timer.h> 92 #include <linux/mm.h> 93 #include <linux/inet.h> 94 #include <linux/netdevice.h> 95 #include <linux/slab.h> 96 #include <net/tcp_states.h> 97 #include <linux/skbuff.h> 98 #include <linux/proc_fs.h> 99 #include <linux/seq_file.h> 100 #include <net/net_namespace.h> 101 #include <net/icmp.h> 102 #include <net/inet_hashtables.h> 103 #include <net/ip_tunnels.h> 104 #include <net/route.h> 105 #include <net/checksum.h> 106 #include <net/gso.h> 107 #include <net/xfrm.h> 108 #include <trace/events/udp.h> 109 #include <linux/static_key.h> 110 #include <linux/btf_ids.h> 111 #include <trace/events/skb.h> 112 #include <net/busy_poll.h> 113 #include "udp_impl.h" 114 #include <net/sock_reuseport.h> 115 #include <net/addrconf.h> 116 #include <net/udp_tunnel.h> 117 #include <net/gro.h> 118 #if IS_ENABLED(CONFIG_IPV6) 119 #include <net/ipv6_stubs.h> 120 #endif 121 122 struct udp_table udp_table __read_mostly; 123 EXPORT_SYMBOL(udp_table); 124 125 long sysctl_udp_mem[3] __read_mostly; 126 EXPORT_SYMBOL(sysctl_udp_mem); 127 128 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp; 129 EXPORT_SYMBOL(udp_memory_allocated); 130 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc); 131 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc); 132 133 #define MAX_UDP_PORTS 65536 134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET) 135 136 static struct udp_table *udp_get_table_prot(struct sock *sk) 137 { 138 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table; 139 } 140 141 static int udp_lib_lport_inuse(struct net *net, __u16 num, 142 const struct udp_hslot *hslot, 143 unsigned long *bitmap, 144 struct sock *sk, unsigned int log) 145 { 146 struct sock *sk2; 147 kuid_t uid = sock_i_uid(sk); 148 149 sk_for_each(sk2, &hslot->head) { 150 if (net_eq(sock_net(sk2), net) && 151 sk2 != sk && 152 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 153 (!sk2->sk_reuse || !sk->sk_reuse) && 154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 156 inet_rcv_saddr_equal(sk, sk2, true)) { 157 if (sk2->sk_reuseport && sk->sk_reuseport && 158 !rcu_access_pointer(sk->sk_reuseport_cb) && 159 uid_eq(uid, sock_i_uid(sk2))) { 160 if (!bitmap) 161 return 0; 162 } else { 163 if (!bitmap) 164 return 1; 165 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 166 bitmap); 167 } 168 } 169 } 170 return 0; 171 } 172 173 /* 174 * Note: we still hold spinlock of primary hash chain, so no other writer 175 * can insert/delete a socket with local_port == num 176 */ 177 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 178 struct udp_hslot *hslot2, 179 struct sock *sk) 180 { 181 struct sock *sk2; 182 kuid_t uid = sock_i_uid(sk); 183 int res = 0; 184 185 spin_lock(&hslot2->lock); 186 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 187 if (net_eq(sock_net(sk2), net) && 188 sk2 != sk && 189 (udp_sk(sk2)->udp_port_hash == num) && 190 (!sk2->sk_reuse || !sk->sk_reuse) && 191 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 192 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 193 inet_rcv_saddr_equal(sk, sk2, true)) { 194 if (sk2->sk_reuseport && sk->sk_reuseport && 195 !rcu_access_pointer(sk->sk_reuseport_cb) && 196 uid_eq(uid, sock_i_uid(sk2))) { 197 res = 0; 198 } else { 199 res = 1; 200 } 201 break; 202 } 203 } 204 spin_unlock(&hslot2->lock); 205 return res; 206 } 207 208 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 209 { 210 struct net *net = sock_net(sk); 211 kuid_t uid = sock_i_uid(sk); 212 struct sock *sk2; 213 214 sk_for_each(sk2, &hslot->head) { 215 if (net_eq(sock_net(sk2), net) && 216 sk2 != sk && 217 sk2->sk_family == sk->sk_family && 218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 221 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 222 inet_rcv_saddr_equal(sk, sk2, false)) { 223 return reuseport_add_sock(sk, sk2, 224 inet_rcv_saddr_any(sk)); 225 } 226 } 227 228 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 229 } 230 231 /** 232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 233 * 234 * @sk: socket struct in question 235 * @snum: port number to look up 236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 237 * with NULL address 238 */ 239 int udp_lib_get_port(struct sock *sk, unsigned short snum, 240 unsigned int hash2_nulladdr) 241 { 242 struct udp_table *udptable = udp_get_table_prot(sk); 243 struct udp_hslot *hslot, *hslot2; 244 struct net *net = sock_net(sk); 245 int error = -EADDRINUSE; 246 247 if (!snum) { 248 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 249 unsigned short first, last; 250 int low, high, remaining; 251 unsigned int rand; 252 253 inet_sk_get_local_port_range(sk, &low, &high); 254 remaining = (high - low) + 1; 255 256 rand = get_random_u32(); 257 first = reciprocal_scale(rand, remaining) + low; 258 /* 259 * force rand to be an odd multiple of UDP_HTABLE_SIZE 260 */ 261 rand = (rand | 1) * (udptable->mask + 1); 262 last = first + udptable->mask + 1; 263 do { 264 hslot = udp_hashslot(udptable, net, first); 265 bitmap_zero(bitmap, PORTS_PER_CHAIN); 266 spin_lock_bh(&hslot->lock); 267 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 268 udptable->log); 269 270 snum = first; 271 /* 272 * Iterate on all possible values of snum for this hash. 273 * Using steps of an odd multiple of UDP_HTABLE_SIZE 274 * give us randomization and full range coverage. 275 */ 276 do { 277 if (low <= snum && snum <= high && 278 !test_bit(snum >> udptable->log, bitmap) && 279 !inet_is_local_reserved_port(net, snum)) 280 goto found; 281 snum += rand; 282 } while (snum != first); 283 spin_unlock_bh(&hslot->lock); 284 cond_resched(); 285 } while (++first != last); 286 goto fail; 287 } else { 288 hslot = udp_hashslot(udptable, net, snum); 289 spin_lock_bh(&hslot->lock); 290 if (hslot->count > 10) { 291 int exist; 292 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 293 294 slot2 &= udptable->mask; 295 hash2_nulladdr &= udptable->mask; 296 297 hslot2 = udp_hashslot2(udptable, slot2); 298 if (hslot->count < hslot2->count) 299 goto scan_primary_hash; 300 301 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 302 if (!exist && (hash2_nulladdr != slot2)) { 303 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 304 exist = udp_lib_lport_inuse2(net, snum, hslot2, 305 sk); 306 } 307 if (exist) 308 goto fail_unlock; 309 else 310 goto found; 311 } 312 scan_primary_hash: 313 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 314 goto fail_unlock; 315 } 316 found: 317 inet_sk(sk)->inet_num = snum; 318 udp_sk(sk)->udp_port_hash = snum; 319 udp_sk(sk)->udp_portaddr_hash ^= snum; 320 if (sk_unhashed(sk)) { 321 if (sk->sk_reuseport && 322 udp_reuseport_add_sock(sk, hslot)) { 323 inet_sk(sk)->inet_num = 0; 324 udp_sk(sk)->udp_port_hash = 0; 325 udp_sk(sk)->udp_portaddr_hash ^= snum; 326 goto fail_unlock; 327 } 328 329 sk_add_node_rcu(sk, &hslot->head); 330 hslot->count++; 331 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 332 333 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 334 spin_lock(&hslot2->lock); 335 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 336 sk->sk_family == AF_INET6) 337 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 338 &hslot2->head); 339 else 340 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 341 &hslot2->head); 342 hslot2->count++; 343 spin_unlock(&hslot2->lock); 344 } 345 sock_set_flag(sk, SOCK_RCU_FREE); 346 error = 0; 347 fail_unlock: 348 spin_unlock_bh(&hslot->lock); 349 fail: 350 return error; 351 } 352 EXPORT_SYMBOL(udp_lib_get_port); 353 354 int udp_v4_get_port(struct sock *sk, unsigned short snum) 355 { 356 unsigned int hash2_nulladdr = 357 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 358 unsigned int hash2_partial = 359 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 360 361 /* precompute partial secondary hash */ 362 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 363 return udp_lib_get_port(sk, snum, hash2_nulladdr); 364 } 365 366 static int compute_score(struct sock *sk, struct net *net, 367 __be32 saddr, __be16 sport, 368 __be32 daddr, unsigned short hnum, 369 int dif, int sdif) 370 { 371 int score; 372 struct inet_sock *inet; 373 bool dev_match; 374 375 if (!net_eq(sock_net(sk), net) || 376 udp_sk(sk)->udp_port_hash != hnum || 377 ipv6_only_sock(sk)) 378 return -1; 379 380 if (sk->sk_rcv_saddr != daddr) 381 return -1; 382 383 score = (sk->sk_family == PF_INET) ? 2 : 1; 384 385 inet = inet_sk(sk); 386 if (inet->inet_daddr) { 387 if (inet->inet_daddr != saddr) 388 return -1; 389 score += 4; 390 } 391 392 if (inet->inet_dport) { 393 if (inet->inet_dport != sport) 394 return -1; 395 score += 4; 396 } 397 398 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 399 dif, sdif); 400 if (!dev_match) 401 return -1; 402 if (sk->sk_bound_dev_if) 403 score += 4; 404 405 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 406 score++; 407 return score; 408 } 409 410 INDIRECT_CALLABLE_SCOPE 411 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport, 412 const __be32 faddr, const __be16 fport) 413 { 414 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 415 416 return __inet_ehashfn(laddr, lport, faddr, fport, 417 udp_ehash_secret + net_hash_mix(net)); 418 } 419 420 /* called with rcu_read_lock() */ 421 static struct sock *udp4_lib_lookup2(struct net *net, 422 __be32 saddr, __be16 sport, 423 __be32 daddr, unsigned int hnum, 424 int dif, int sdif, 425 struct udp_hslot *hslot2, 426 struct sk_buff *skb) 427 { 428 struct sock *sk, *result; 429 int score, badness; 430 bool need_rescore; 431 432 result = NULL; 433 badness = 0; 434 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 435 need_rescore = false; 436 rescore: 437 score = compute_score(need_rescore ? result : sk, net, saddr, 438 sport, daddr, hnum, dif, sdif); 439 if (score > badness) { 440 badness = score; 441 442 if (need_rescore) 443 continue; 444 445 if (sk->sk_state == TCP_ESTABLISHED) { 446 result = sk; 447 continue; 448 } 449 450 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 451 saddr, sport, daddr, hnum, udp_ehashfn); 452 if (!result) { 453 result = sk; 454 continue; 455 } 456 457 /* Fall back to scoring if group has connections */ 458 if (!reuseport_has_conns(sk)) 459 return result; 460 461 /* Reuseport logic returned an error, keep original score. */ 462 if (IS_ERR(result)) 463 continue; 464 465 /* compute_score is too long of a function to be 466 * inlined, and calling it again here yields 467 * measureable overhead for some 468 * workloads. Work around it by jumping 469 * backwards to rescore 'result'. 470 */ 471 need_rescore = true; 472 goto rescore; 473 } 474 } 475 return result; 476 } 477 478 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 479 * harder than this. -DaveM 480 */ 481 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 482 __be16 sport, __be32 daddr, __be16 dport, int dif, 483 int sdif, struct udp_table *udptable, struct sk_buff *skb) 484 { 485 unsigned short hnum = ntohs(dport); 486 unsigned int hash2, slot2; 487 struct udp_hslot *hslot2; 488 struct sock *result, *sk; 489 490 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 491 slot2 = hash2 & udptable->mask; 492 hslot2 = &udptable->hash2[slot2]; 493 494 /* Lookup connected or non-wildcard socket */ 495 result = udp4_lib_lookup2(net, saddr, sport, 496 daddr, hnum, dif, sdif, 497 hslot2, skb); 498 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 499 goto done; 500 501 /* Lookup redirect from BPF */ 502 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 503 udptable == net->ipv4.udp_table) { 504 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 505 saddr, sport, daddr, hnum, dif, 506 udp_ehashfn); 507 if (sk) { 508 result = sk; 509 goto done; 510 } 511 } 512 513 /* Got non-wildcard socket or error on first lookup */ 514 if (result) 515 goto done; 516 517 /* Lookup wildcard sockets */ 518 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 519 slot2 = hash2 & udptable->mask; 520 hslot2 = &udptable->hash2[slot2]; 521 522 result = udp4_lib_lookup2(net, saddr, sport, 523 htonl(INADDR_ANY), hnum, dif, sdif, 524 hslot2, skb); 525 done: 526 if (IS_ERR(result)) 527 return NULL; 528 return result; 529 } 530 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 531 532 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 533 __be16 sport, __be16 dport, 534 struct udp_table *udptable) 535 { 536 const struct iphdr *iph = ip_hdr(skb); 537 538 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 539 iph->daddr, dport, inet_iif(skb), 540 inet_sdif(skb), udptable, skb); 541 } 542 543 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 544 __be16 sport, __be16 dport) 545 { 546 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation]; 547 const struct iphdr *iph = (struct iphdr *)(skb->data + offset); 548 struct net *net = dev_net(skb->dev); 549 int iif, sdif; 550 551 inet_get_iif_sdif(skb, &iif, &sdif); 552 553 return __udp4_lib_lookup(net, iph->saddr, sport, 554 iph->daddr, dport, iif, 555 sdif, net->ipv4.udp_table, NULL); 556 } 557 558 /* Must be called under rcu_read_lock(). 559 * Does increment socket refcount. 560 */ 561 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 562 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 563 __be32 daddr, __be16 dport, int dif) 564 { 565 struct sock *sk; 566 567 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 568 dif, 0, net->ipv4.udp_table, NULL); 569 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 570 sk = NULL; 571 return sk; 572 } 573 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 574 #endif 575 576 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk, 577 __be16 loc_port, __be32 loc_addr, 578 __be16 rmt_port, __be32 rmt_addr, 579 int dif, int sdif, unsigned short hnum) 580 { 581 const struct inet_sock *inet = inet_sk(sk); 582 583 if (!net_eq(sock_net(sk), net) || 584 udp_sk(sk)->udp_port_hash != hnum || 585 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 586 (inet->inet_dport != rmt_port && inet->inet_dport) || 587 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 588 ipv6_only_sock(sk) || 589 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 590 return false; 591 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 592 return false; 593 return true; 594 } 595 596 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 597 EXPORT_SYMBOL(udp_encap_needed_key); 598 599 #if IS_ENABLED(CONFIG_IPV6) 600 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 601 EXPORT_SYMBOL(udpv6_encap_needed_key); 602 #endif 603 604 void udp_encap_enable(void) 605 { 606 static_branch_inc(&udp_encap_needed_key); 607 } 608 EXPORT_SYMBOL(udp_encap_enable); 609 610 void udp_encap_disable(void) 611 { 612 static_branch_dec(&udp_encap_needed_key); 613 } 614 EXPORT_SYMBOL(udp_encap_disable); 615 616 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 617 * through error handlers in encapsulations looking for a match. 618 */ 619 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 620 { 621 int i; 622 623 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 624 int (*handler)(struct sk_buff *skb, u32 info); 625 const struct ip_tunnel_encap_ops *encap; 626 627 encap = rcu_dereference(iptun_encaps[i]); 628 if (!encap) 629 continue; 630 handler = encap->err_handler; 631 if (handler && !handler(skb, info)) 632 return 0; 633 } 634 635 return -ENOENT; 636 } 637 638 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 639 * reversing source and destination port: this will match tunnels that force the 640 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 641 * lwtunnels might actually break this assumption by being configured with 642 * different destination ports on endpoints, in this case we won't be able to 643 * trace ICMP messages back to them. 644 * 645 * If this doesn't match any socket, probe tunnels with arbitrary destination 646 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 647 * we've sent packets to won't necessarily match the local destination port. 648 * 649 * Then ask the tunnel implementation to match the error against a valid 650 * association. 651 * 652 * Return an error if we can't find a match, the socket if we need further 653 * processing, zero otherwise. 654 */ 655 static struct sock *__udp4_lib_err_encap(struct net *net, 656 const struct iphdr *iph, 657 struct udphdr *uh, 658 struct udp_table *udptable, 659 struct sock *sk, 660 struct sk_buff *skb, u32 info) 661 { 662 int (*lookup)(struct sock *sk, struct sk_buff *skb); 663 int network_offset, transport_offset; 664 struct udp_sock *up; 665 666 network_offset = skb_network_offset(skb); 667 transport_offset = skb_transport_offset(skb); 668 669 /* Network header needs to point to the outer IPv4 header inside ICMP */ 670 skb_reset_network_header(skb); 671 672 /* Transport header needs to point to the UDP header */ 673 skb_set_transport_header(skb, iph->ihl << 2); 674 675 if (sk) { 676 up = udp_sk(sk); 677 678 lookup = READ_ONCE(up->encap_err_lookup); 679 if (lookup && lookup(sk, skb)) 680 sk = NULL; 681 682 goto out; 683 } 684 685 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 686 iph->saddr, uh->dest, skb->dev->ifindex, 0, 687 udptable, NULL); 688 if (sk) { 689 up = udp_sk(sk); 690 691 lookup = READ_ONCE(up->encap_err_lookup); 692 if (!lookup || lookup(sk, skb)) 693 sk = NULL; 694 } 695 696 out: 697 if (!sk) 698 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 699 700 skb_set_transport_header(skb, transport_offset); 701 skb_set_network_header(skb, network_offset); 702 703 return sk; 704 } 705 706 /* 707 * This routine is called by the ICMP module when it gets some 708 * sort of error condition. If err < 0 then the socket should 709 * be closed and the error returned to the user. If err > 0 710 * it's just the icmp type << 8 | icmp code. 711 * Header points to the ip header of the error packet. We move 712 * on past this. Then (as it used to claim before adjustment) 713 * header points to the first 8 bytes of the udp header. We need 714 * to find the appropriate port. 715 */ 716 717 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 718 { 719 struct inet_sock *inet; 720 const struct iphdr *iph = (const struct iphdr *)skb->data; 721 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 722 const int type = icmp_hdr(skb)->type; 723 const int code = icmp_hdr(skb)->code; 724 bool tunnel = false; 725 struct sock *sk; 726 int harderr; 727 int err; 728 struct net *net = dev_net(skb->dev); 729 730 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 731 iph->saddr, uh->source, skb->dev->ifindex, 732 inet_sdif(skb), udptable, NULL); 733 734 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) { 735 /* No socket for error: try tunnels before discarding */ 736 if (static_branch_unlikely(&udp_encap_needed_key)) { 737 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 738 info); 739 if (!sk) 740 return 0; 741 } else 742 sk = ERR_PTR(-ENOENT); 743 744 if (IS_ERR(sk)) { 745 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 746 return PTR_ERR(sk); 747 } 748 749 tunnel = true; 750 } 751 752 err = 0; 753 harderr = 0; 754 inet = inet_sk(sk); 755 756 switch (type) { 757 default: 758 case ICMP_TIME_EXCEEDED: 759 err = EHOSTUNREACH; 760 break; 761 case ICMP_SOURCE_QUENCH: 762 goto out; 763 case ICMP_PARAMETERPROB: 764 err = EPROTO; 765 harderr = 1; 766 break; 767 case ICMP_DEST_UNREACH: 768 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 769 ipv4_sk_update_pmtu(skb, sk, info); 770 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) { 771 err = EMSGSIZE; 772 harderr = 1; 773 break; 774 } 775 goto out; 776 } 777 err = EHOSTUNREACH; 778 if (code <= NR_ICMP_UNREACH) { 779 harderr = icmp_err_convert[code].fatal; 780 err = icmp_err_convert[code].errno; 781 } 782 break; 783 case ICMP_REDIRECT: 784 ipv4_sk_redirect(skb, sk); 785 goto out; 786 } 787 788 /* 789 * RFC1122: OK. Passes ICMP errors back to application, as per 790 * 4.1.3.3. 791 */ 792 if (tunnel) { 793 /* ...not for tunnels though: we don't have a sending socket */ 794 if (udp_sk(sk)->encap_err_rcv) 795 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info, 796 (u8 *)(uh+1)); 797 goto out; 798 } 799 if (!inet_test_bit(RECVERR, sk)) { 800 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 801 goto out; 802 } else 803 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 804 805 sk->sk_err = err; 806 sk_error_report(sk); 807 out: 808 return 0; 809 } 810 811 int udp_err(struct sk_buff *skb, u32 info) 812 { 813 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table); 814 } 815 816 /* 817 * Throw away all pending data and cancel the corking. Socket is locked. 818 */ 819 void udp_flush_pending_frames(struct sock *sk) 820 { 821 struct udp_sock *up = udp_sk(sk); 822 823 if (up->pending) { 824 up->len = 0; 825 WRITE_ONCE(up->pending, 0); 826 ip_flush_pending_frames(sk); 827 } 828 } 829 EXPORT_SYMBOL(udp_flush_pending_frames); 830 831 /** 832 * udp4_hwcsum - handle outgoing HW checksumming 833 * @skb: sk_buff containing the filled-in UDP header 834 * (checksum field must be zeroed out) 835 * @src: source IP address 836 * @dst: destination IP address 837 */ 838 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 839 { 840 struct udphdr *uh = udp_hdr(skb); 841 int offset = skb_transport_offset(skb); 842 int len = skb->len - offset; 843 int hlen = len; 844 __wsum csum = 0; 845 846 if (!skb_has_frag_list(skb)) { 847 /* 848 * Only one fragment on the socket. 849 */ 850 skb->csum_start = skb_transport_header(skb) - skb->head; 851 skb->csum_offset = offsetof(struct udphdr, check); 852 uh->check = ~csum_tcpudp_magic(src, dst, len, 853 IPPROTO_UDP, 0); 854 } else { 855 struct sk_buff *frags; 856 857 /* 858 * HW-checksum won't work as there are two or more 859 * fragments on the socket so that all csums of sk_buffs 860 * should be together 861 */ 862 skb_walk_frags(skb, frags) { 863 csum = csum_add(csum, frags->csum); 864 hlen -= frags->len; 865 } 866 867 csum = skb_checksum(skb, offset, hlen, csum); 868 skb->ip_summed = CHECKSUM_NONE; 869 870 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 871 if (uh->check == 0) 872 uh->check = CSUM_MANGLED_0; 873 } 874 } 875 EXPORT_SYMBOL_GPL(udp4_hwcsum); 876 877 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 878 * for the simple case like when setting the checksum for a UDP tunnel. 879 */ 880 void udp_set_csum(bool nocheck, struct sk_buff *skb, 881 __be32 saddr, __be32 daddr, int len) 882 { 883 struct udphdr *uh = udp_hdr(skb); 884 885 if (nocheck) { 886 uh->check = 0; 887 } else if (skb_is_gso(skb)) { 888 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 889 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 890 uh->check = 0; 891 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 892 if (uh->check == 0) 893 uh->check = CSUM_MANGLED_0; 894 } else { 895 skb->ip_summed = CHECKSUM_PARTIAL; 896 skb->csum_start = skb_transport_header(skb) - skb->head; 897 skb->csum_offset = offsetof(struct udphdr, check); 898 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 899 } 900 } 901 EXPORT_SYMBOL(udp_set_csum); 902 903 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 904 struct inet_cork *cork) 905 { 906 struct sock *sk = skb->sk; 907 struct inet_sock *inet = inet_sk(sk); 908 struct udphdr *uh; 909 int err; 910 int is_udplite = IS_UDPLITE(sk); 911 int offset = skb_transport_offset(skb); 912 int len = skb->len - offset; 913 int datalen = len - sizeof(*uh); 914 __wsum csum = 0; 915 916 /* 917 * Create a UDP header 918 */ 919 uh = udp_hdr(skb); 920 uh->source = inet->inet_sport; 921 uh->dest = fl4->fl4_dport; 922 uh->len = htons(len); 923 uh->check = 0; 924 925 if (cork->gso_size) { 926 const int hlen = skb_network_header_len(skb) + 927 sizeof(struct udphdr); 928 929 if (hlen + cork->gso_size > cork->fragsize) { 930 kfree_skb(skb); 931 return -EINVAL; 932 } 933 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 934 kfree_skb(skb); 935 return -EINVAL; 936 } 937 if (sk->sk_no_check_tx) { 938 kfree_skb(skb); 939 return -EINVAL; 940 } 941 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 942 dst_xfrm(skb_dst(skb))) { 943 kfree_skb(skb); 944 return -EIO; 945 } 946 947 if (datalen > cork->gso_size) { 948 skb_shinfo(skb)->gso_size = cork->gso_size; 949 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 950 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 951 cork->gso_size); 952 } 953 goto csum_partial; 954 } 955 956 if (is_udplite) /* UDP-Lite */ 957 csum = udplite_csum(skb); 958 959 else if (sk->sk_no_check_tx) { /* UDP csum off */ 960 961 skb->ip_summed = CHECKSUM_NONE; 962 goto send; 963 964 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 965 csum_partial: 966 967 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 968 goto send; 969 970 } else 971 csum = udp_csum(skb); 972 973 /* add protocol-dependent pseudo-header */ 974 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 975 sk->sk_protocol, csum); 976 if (uh->check == 0) 977 uh->check = CSUM_MANGLED_0; 978 979 send: 980 err = ip_send_skb(sock_net(sk), skb); 981 if (err) { 982 if (err == -ENOBUFS && 983 !inet_test_bit(RECVERR, sk)) { 984 UDP_INC_STATS(sock_net(sk), 985 UDP_MIB_SNDBUFERRORS, is_udplite); 986 err = 0; 987 } 988 } else 989 UDP_INC_STATS(sock_net(sk), 990 UDP_MIB_OUTDATAGRAMS, is_udplite); 991 return err; 992 } 993 994 /* 995 * Push out all pending data as one UDP datagram. Socket is locked. 996 */ 997 int udp_push_pending_frames(struct sock *sk) 998 { 999 struct udp_sock *up = udp_sk(sk); 1000 struct inet_sock *inet = inet_sk(sk); 1001 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 1002 struct sk_buff *skb; 1003 int err = 0; 1004 1005 skb = ip_finish_skb(sk, fl4); 1006 if (!skb) 1007 goto out; 1008 1009 err = udp_send_skb(skb, fl4, &inet->cork.base); 1010 1011 out: 1012 up->len = 0; 1013 WRITE_ONCE(up->pending, 0); 1014 return err; 1015 } 1016 EXPORT_SYMBOL(udp_push_pending_frames); 1017 1018 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1019 { 1020 switch (cmsg->cmsg_type) { 1021 case UDP_SEGMENT: 1022 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1023 return -EINVAL; 1024 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1025 return 0; 1026 default: 1027 return -EINVAL; 1028 } 1029 } 1030 1031 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1032 { 1033 struct cmsghdr *cmsg; 1034 bool need_ip = false; 1035 int err; 1036 1037 for_each_cmsghdr(cmsg, msg) { 1038 if (!CMSG_OK(msg, cmsg)) 1039 return -EINVAL; 1040 1041 if (cmsg->cmsg_level != SOL_UDP) { 1042 need_ip = true; 1043 continue; 1044 } 1045 1046 err = __udp_cmsg_send(cmsg, gso_size); 1047 if (err) 1048 return err; 1049 } 1050 1051 return need_ip; 1052 } 1053 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1054 1055 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1056 { 1057 struct inet_sock *inet = inet_sk(sk); 1058 struct udp_sock *up = udp_sk(sk); 1059 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1060 struct flowi4 fl4_stack; 1061 struct flowi4 *fl4; 1062 int ulen = len; 1063 struct ipcm_cookie ipc; 1064 struct rtable *rt = NULL; 1065 int free = 0; 1066 int connected = 0; 1067 __be32 daddr, faddr, saddr; 1068 u8 tos, scope; 1069 __be16 dport; 1070 int err, is_udplite = IS_UDPLITE(sk); 1071 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE; 1072 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1073 struct sk_buff *skb; 1074 struct ip_options_data opt_copy; 1075 int uc_index; 1076 1077 if (len > 0xFFFF) 1078 return -EMSGSIZE; 1079 1080 /* 1081 * Check the flags. 1082 */ 1083 1084 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1085 return -EOPNOTSUPP; 1086 1087 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1088 1089 fl4 = &inet->cork.fl.u.ip4; 1090 if (READ_ONCE(up->pending)) { 1091 /* 1092 * There are pending frames. 1093 * The socket lock must be held while it's corked. 1094 */ 1095 lock_sock(sk); 1096 if (likely(up->pending)) { 1097 if (unlikely(up->pending != AF_INET)) { 1098 release_sock(sk); 1099 return -EINVAL; 1100 } 1101 goto do_append_data; 1102 } 1103 release_sock(sk); 1104 } 1105 ulen += sizeof(struct udphdr); 1106 1107 /* 1108 * Get and verify the address. 1109 */ 1110 if (usin) { 1111 if (msg->msg_namelen < sizeof(*usin)) 1112 return -EINVAL; 1113 if (usin->sin_family != AF_INET) { 1114 if (usin->sin_family != AF_UNSPEC) 1115 return -EAFNOSUPPORT; 1116 } 1117 1118 daddr = usin->sin_addr.s_addr; 1119 dport = usin->sin_port; 1120 if (dport == 0) 1121 return -EINVAL; 1122 } else { 1123 if (sk->sk_state != TCP_ESTABLISHED) 1124 return -EDESTADDRREQ; 1125 daddr = inet->inet_daddr; 1126 dport = inet->inet_dport; 1127 /* Open fast path for connected socket. 1128 Route will not be used, if at least one option is set. 1129 */ 1130 connected = 1; 1131 } 1132 1133 ipcm_init_sk(&ipc, inet); 1134 ipc.gso_size = READ_ONCE(up->gso_size); 1135 1136 if (msg->msg_controllen) { 1137 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1138 if (err > 0) { 1139 err = ip_cmsg_send(sk, msg, &ipc, 1140 sk->sk_family == AF_INET6); 1141 connected = 0; 1142 } 1143 if (unlikely(err < 0)) { 1144 kfree(ipc.opt); 1145 return err; 1146 } 1147 if (ipc.opt) 1148 free = 1; 1149 } 1150 if (!ipc.opt) { 1151 struct ip_options_rcu *inet_opt; 1152 1153 rcu_read_lock(); 1154 inet_opt = rcu_dereference(inet->inet_opt); 1155 if (inet_opt) { 1156 memcpy(&opt_copy, inet_opt, 1157 sizeof(*inet_opt) + inet_opt->opt.optlen); 1158 ipc.opt = &opt_copy.opt; 1159 } 1160 rcu_read_unlock(); 1161 } 1162 1163 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1164 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1165 (struct sockaddr *)usin, 1166 &msg->msg_namelen, 1167 &ipc.addr); 1168 if (err) 1169 goto out_free; 1170 if (usin) { 1171 if (usin->sin_port == 0) { 1172 /* BPF program set invalid port. Reject it. */ 1173 err = -EINVAL; 1174 goto out_free; 1175 } 1176 daddr = usin->sin_addr.s_addr; 1177 dport = usin->sin_port; 1178 } 1179 } 1180 1181 saddr = ipc.addr; 1182 ipc.addr = faddr = daddr; 1183 1184 if (ipc.opt && ipc.opt->opt.srr) { 1185 if (!daddr) { 1186 err = -EINVAL; 1187 goto out_free; 1188 } 1189 faddr = ipc.opt->opt.faddr; 1190 connected = 0; 1191 } 1192 tos = get_rttos(&ipc, inet); 1193 scope = ip_sendmsg_scope(inet, &ipc, msg); 1194 if (scope == RT_SCOPE_LINK) 1195 connected = 0; 1196 1197 uc_index = READ_ONCE(inet->uc_index); 1198 if (ipv4_is_multicast(daddr)) { 1199 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1200 ipc.oif = READ_ONCE(inet->mc_index); 1201 if (!saddr) 1202 saddr = READ_ONCE(inet->mc_addr); 1203 connected = 0; 1204 } else if (!ipc.oif) { 1205 ipc.oif = uc_index; 1206 } else if (ipv4_is_lbcast(daddr) && uc_index) { 1207 /* oif is set, packet is to local broadcast and 1208 * uc_index is set. oif is most likely set 1209 * by sk_bound_dev_if. If uc_index != oif check if the 1210 * oif is an L3 master and uc_index is an L3 slave. 1211 * If so, we want to allow the send using the uc_index. 1212 */ 1213 if (ipc.oif != uc_index && 1214 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1215 uc_index)) { 1216 ipc.oif = uc_index; 1217 } 1218 } 1219 1220 if (connected) 1221 rt = dst_rtable(sk_dst_check(sk, 0)); 1222 1223 if (!rt) { 1224 struct net *net = sock_net(sk); 1225 __u8 flow_flags = inet_sk_flowi_flags(sk); 1226 1227 fl4 = &fl4_stack; 1228 1229 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope, 1230 sk->sk_protocol, flow_flags, faddr, saddr, 1231 dport, inet->inet_sport, sk->sk_uid); 1232 1233 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1234 rt = ip_route_output_flow(net, fl4, sk); 1235 if (IS_ERR(rt)) { 1236 err = PTR_ERR(rt); 1237 rt = NULL; 1238 if (err == -ENETUNREACH) 1239 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1240 goto out; 1241 } 1242 1243 err = -EACCES; 1244 if ((rt->rt_flags & RTCF_BROADCAST) && 1245 !sock_flag(sk, SOCK_BROADCAST)) 1246 goto out; 1247 if (connected) 1248 sk_dst_set(sk, dst_clone(&rt->dst)); 1249 } 1250 1251 if (msg->msg_flags&MSG_CONFIRM) 1252 goto do_confirm; 1253 back_from_confirm: 1254 1255 saddr = fl4->saddr; 1256 if (!ipc.addr) 1257 daddr = ipc.addr = fl4->daddr; 1258 1259 /* Lockless fast path for the non-corking case. */ 1260 if (!corkreq) { 1261 struct inet_cork cork; 1262 1263 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1264 sizeof(struct udphdr), &ipc, &rt, 1265 &cork, msg->msg_flags); 1266 err = PTR_ERR(skb); 1267 if (!IS_ERR_OR_NULL(skb)) 1268 err = udp_send_skb(skb, fl4, &cork); 1269 goto out; 1270 } 1271 1272 lock_sock(sk); 1273 if (unlikely(up->pending)) { 1274 /* The socket is already corked while preparing it. */ 1275 /* ... which is an evident application bug. --ANK */ 1276 release_sock(sk); 1277 1278 net_dbg_ratelimited("socket already corked\n"); 1279 err = -EINVAL; 1280 goto out; 1281 } 1282 /* 1283 * Now cork the socket to pend data. 1284 */ 1285 fl4 = &inet->cork.fl.u.ip4; 1286 fl4->daddr = daddr; 1287 fl4->saddr = saddr; 1288 fl4->fl4_dport = dport; 1289 fl4->fl4_sport = inet->inet_sport; 1290 WRITE_ONCE(up->pending, AF_INET); 1291 1292 do_append_data: 1293 up->len += ulen; 1294 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1295 sizeof(struct udphdr), &ipc, &rt, 1296 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1297 if (err) 1298 udp_flush_pending_frames(sk); 1299 else if (!corkreq) 1300 err = udp_push_pending_frames(sk); 1301 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1302 WRITE_ONCE(up->pending, 0); 1303 release_sock(sk); 1304 1305 out: 1306 ip_rt_put(rt); 1307 out_free: 1308 if (free) 1309 kfree(ipc.opt); 1310 if (!err) 1311 return len; 1312 /* 1313 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1314 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1315 * we don't have a good statistic (IpOutDiscards but it can be too many 1316 * things). We could add another new stat but at least for now that 1317 * seems like overkill. 1318 */ 1319 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1320 UDP_INC_STATS(sock_net(sk), 1321 UDP_MIB_SNDBUFERRORS, is_udplite); 1322 } 1323 return err; 1324 1325 do_confirm: 1326 if (msg->msg_flags & MSG_PROBE) 1327 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1328 if (!(msg->msg_flags&MSG_PROBE) || len) 1329 goto back_from_confirm; 1330 err = 0; 1331 goto out; 1332 } 1333 EXPORT_SYMBOL(udp_sendmsg); 1334 1335 void udp_splice_eof(struct socket *sock) 1336 { 1337 struct sock *sk = sock->sk; 1338 struct udp_sock *up = udp_sk(sk); 1339 1340 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk)) 1341 return; 1342 1343 lock_sock(sk); 1344 if (up->pending && !udp_test_bit(CORK, sk)) 1345 udp_push_pending_frames(sk); 1346 release_sock(sk); 1347 } 1348 EXPORT_SYMBOL_GPL(udp_splice_eof); 1349 1350 #define UDP_SKB_IS_STATELESS 0x80000000 1351 1352 /* all head states (dst, sk, nf conntrack) except skb extensions are 1353 * cleared by udp_rcv(). 1354 * 1355 * We need to preserve secpath, if present, to eventually process 1356 * IP_CMSG_PASSSEC at recvmsg() time. 1357 * 1358 * Other extensions can be cleared. 1359 */ 1360 static bool udp_try_make_stateless(struct sk_buff *skb) 1361 { 1362 if (!skb_has_extensions(skb)) 1363 return true; 1364 1365 if (!secpath_exists(skb)) { 1366 skb_ext_reset(skb); 1367 return true; 1368 } 1369 1370 return false; 1371 } 1372 1373 static void udp_set_dev_scratch(struct sk_buff *skb) 1374 { 1375 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1376 1377 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1378 scratch->_tsize_state = skb->truesize; 1379 #if BITS_PER_LONG == 64 1380 scratch->len = skb->len; 1381 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1382 scratch->is_linear = !skb_is_nonlinear(skb); 1383 #endif 1384 if (udp_try_make_stateless(skb)) 1385 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1386 } 1387 1388 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1389 { 1390 /* We come here after udp_lib_checksum_complete() returned 0. 1391 * This means that __skb_checksum_complete() might have 1392 * set skb->csum_valid to 1. 1393 * On 64bit platforms, we can set csum_unnecessary 1394 * to true, but only if the skb is not shared. 1395 */ 1396 #if BITS_PER_LONG == 64 1397 if (!skb_shared(skb)) 1398 udp_skb_scratch(skb)->csum_unnecessary = true; 1399 #endif 1400 } 1401 1402 static int udp_skb_truesize(struct sk_buff *skb) 1403 { 1404 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1405 } 1406 1407 static bool udp_skb_has_head_state(struct sk_buff *skb) 1408 { 1409 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1410 } 1411 1412 /* fully reclaim rmem/fwd memory allocated for skb */ 1413 static void udp_rmem_release(struct sock *sk, int size, int partial, 1414 bool rx_queue_lock_held) 1415 { 1416 struct udp_sock *up = udp_sk(sk); 1417 struct sk_buff_head *sk_queue; 1418 int amt; 1419 1420 if (likely(partial)) { 1421 up->forward_deficit += size; 1422 size = up->forward_deficit; 1423 if (size < READ_ONCE(up->forward_threshold) && 1424 !skb_queue_empty(&up->reader_queue)) 1425 return; 1426 } else { 1427 size += up->forward_deficit; 1428 } 1429 up->forward_deficit = 0; 1430 1431 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1432 * if the called don't held it already 1433 */ 1434 sk_queue = &sk->sk_receive_queue; 1435 if (!rx_queue_lock_held) 1436 spin_lock(&sk_queue->lock); 1437 1438 1439 sk_forward_alloc_add(sk, size); 1440 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1441 sk_forward_alloc_add(sk, -amt); 1442 1443 if (amt) 1444 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 1445 1446 atomic_sub(size, &sk->sk_rmem_alloc); 1447 1448 /* this can save us from acquiring the rx queue lock on next receive */ 1449 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1450 1451 if (!rx_queue_lock_held) 1452 spin_unlock(&sk_queue->lock); 1453 } 1454 1455 /* Note: called with reader_queue.lock held. 1456 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1457 * This avoids a cache line miss while receive_queue lock is held. 1458 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1459 */ 1460 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1461 { 1462 prefetch(&skb->data); 1463 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1464 } 1465 EXPORT_SYMBOL(udp_skb_destructor); 1466 1467 /* as above, but the caller held the rx queue lock, too */ 1468 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1469 { 1470 prefetch(&skb->data); 1471 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1472 } 1473 1474 /* Idea of busylocks is to let producers grab an extra spinlock 1475 * to relieve pressure on the receive_queue spinlock shared by consumer. 1476 * Under flood, this means that only one producer can be in line 1477 * trying to acquire the receive_queue spinlock. 1478 * These busylock can be allocated on a per cpu manner, instead of a 1479 * per socket one (that would consume a cache line per socket) 1480 */ 1481 static int udp_busylocks_log __read_mostly; 1482 static spinlock_t *udp_busylocks __read_mostly; 1483 1484 static spinlock_t *busylock_acquire(void *ptr) 1485 { 1486 spinlock_t *busy; 1487 1488 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1489 spin_lock(busy); 1490 return busy; 1491 } 1492 1493 static void busylock_release(spinlock_t *busy) 1494 { 1495 if (busy) 1496 spin_unlock(busy); 1497 } 1498 1499 static int udp_rmem_schedule(struct sock *sk, int size) 1500 { 1501 int delta; 1502 1503 delta = size - sk->sk_forward_alloc; 1504 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV)) 1505 return -ENOBUFS; 1506 1507 return 0; 1508 } 1509 1510 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1511 { 1512 struct sk_buff_head *list = &sk->sk_receive_queue; 1513 int rmem, err = -ENOMEM; 1514 spinlock_t *busy = NULL; 1515 bool becomes_readable; 1516 int size, rcvbuf; 1517 1518 /* Immediately drop when the receive queue is full. 1519 * Always allow at least one packet. 1520 */ 1521 rmem = atomic_read(&sk->sk_rmem_alloc); 1522 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1523 if (rmem > rcvbuf) 1524 goto drop; 1525 1526 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1527 * having linear skbs : 1528 * - Reduce memory overhead and thus increase receive queue capacity 1529 * - Less cache line misses at copyout() time 1530 * - Less work at consume_skb() (less alien page frag freeing) 1531 */ 1532 if (rmem > (rcvbuf >> 1)) { 1533 skb_condense(skb); 1534 1535 busy = busylock_acquire(sk); 1536 } 1537 size = skb->truesize; 1538 udp_set_dev_scratch(skb); 1539 1540 atomic_add(size, &sk->sk_rmem_alloc); 1541 1542 spin_lock(&list->lock); 1543 err = udp_rmem_schedule(sk, size); 1544 if (err) { 1545 spin_unlock(&list->lock); 1546 goto uncharge_drop; 1547 } 1548 1549 sk_forward_alloc_add(sk, -size); 1550 1551 /* no need to setup a destructor, we will explicitly release the 1552 * forward allocated memory on dequeue 1553 */ 1554 sock_skb_set_dropcount(sk, skb); 1555 1556 becomes_readable = skb_queue_empty(list); 1557 __skb_queue_tail(list, skb); 1558 spin_unlock(&list->lock); 1559 1560 if (!sock_flag(sk, SOCK_DEAD)) { 1561 if (becomes_readable || 1562 sk->sk_data_ready != sock_def_readable || 1563 READ_ONCE(sk->sk_peek_off) >= 0) 1564 INDIRECT_CALL_1(sk->sk_data_ready, 1565 sock_def_readable, sk); 1566 else 1567 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 1568 } 1569 busylock_release(busy); 1570 return 0; 1571 1572 uncharge_drop: 1573 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1574 1575 drop: 1576 atomic_inc(&sk->sk_drops); 1577 busylock_release(busy); 1578 return err; 1579 } 1580 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1581 1582 void udp_destruct_common(struct sock *sk) 1583 { 1584 /* reclaim completely the forward allocated memory */ 1585 struct udp_sock *up = udp_sk(sk); 1586 unsigned int total = 0; 1587 struct sk_buff *skb; 1588 1589 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1590 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1591 total += skb->truesize; 1592 kfree_skb(skb); 1593 } 1594 udp_rmem_release(sk, total, 0, true); 1595 } 1596 EXPORT_SYMBOL_GPL(udp_destruct_common); 1597 1598 static void udp_destruct_sock(struct sock *sk) 1599 { 1600 udp_destruct_common(sk); 1601 inet_sock_destruct(sk); 1602 } 1603 1604 int udp_init_sock(struct sock *sk) 1605 { 1606 udp_lib_init_sock(sk); 1607 sk->sk_destruct = udp_destruct_sock; 1608 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1609 return 0; 1610 } 1611 1612 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1613 { 1614 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset))) 1615 sk_peek_offset_bwd(sk, len); 1616 1617 if (!skb_unref(skb)) 1618 return; 1619 1620 /* In the more common cases we cleared the head states previously, 1621 * see __udp_queue_rcv_skb(). 1622 */ 1623 if (unlikely(udp_skb_has_head_state(skb))) 1624 skb_release_head_state(skb); 1625 __consume_stateless_skb(skb); 1626 } 1627 EXPORT_SYMBOL_GPL(skb_consume_udp); 1628 1629 static struct sk_buff *__first_packet_length(struct sock *sk, 1630 struct sk_buff_head *rcvq, 1631 int *total) 1632 { 1633 struct sk_buff *skb; 1634 1635 while ((skb = skb_peek(rcvq)) != NULL) { 1636 if (udp_lib_checksum_complete(skb)) { 1637 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1638 IS_UDPLITE(sk)); 1639 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1640 IS_UDPLITE(sk)); 1641 atomic_inc(&sk->sk_drops); 1642 __skb_unlink(skb, rcvq); 1643 *total += skb->truesize; 1644 kfree_skb(skb); 1645 } else { 1646 udp_skb_csum_unnecessary_set(skb); 1647 break; 1648 } 1649 } 1650 return skb; 1651 } 1652 1653 /** 1654 * first_packet_length - return length of first packet in receive queue 1655 * @sk: socket 1656 * 1657 * Drops all bad checksum frames, until a valid one is found. 1658 * Returns the length of found skb, or -1 if none is found. 1659 */ 1660 static int first_packet_length(struct sock *sk) 1661 { 1662 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1663 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1664 struct sk_buff *skb; 1665 int total = 0; 1666 int res; 1667 1668 spin_lock_bh(&rcvq->lock); 1669 skb = __first_packet_length(sk, rcvq, &total); 1670 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1671 spin_lock(&sk_queue->lock); 1672 skb_queue_splice_tail_init(sk_queue, rcvq); 1673 spin_unlock(&sk_queue->lock); 1674 1675 skb = __first_packet_length(sk, rcvq, &total); 1676 } 1677 res = skb ? skb->len : -1; 1678 if (total) 1679 udp_rmem_release(sk, total, 1, false); 1680 spin_unlock_bh(&rcvq->lock); 1681 return res; 1682 } 1683 1684 /* 1685 * IOCTL requests applicable to the UDP protocol 1686 */ 1687 1688 int udp_ioctl(struct sock *sk, int cmd, int *karg) 1689 { 1690 switch (cmd) { 1691 case SIOCOUTQ: 1692 { 1693 *karg = sk_wmem_alloc_get(sk); 1694 return 0; 1695 } 1696 1697 case SIOCINQ: 1698 { 1699 *karg = max_t(int, 0, first_packet_length(sk)); 1700 return 0; 1701 } 1702 1703 default: 1704 return -ENOIOCTLCMD; 1705 } 1706 1707 return 0; 1708 } 1709 EXPORT_SYMBOL(udp_ioctl); 1710 1711 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1712 int *off, int *err) 1713 { 1714 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1715 struct sk_buff_head *queue; 1716 struct sk_buff *last; 1717 long timeo; 1718 int error; 1719 1720 queue = &udp_sk(sk)->reader_queue; 1721 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1722 do { 1723 struct sk_buff *skb; 1724 1725 error = sock_error(sk); 1726 if (error) 1727 break; 1728 1729 error = -EAGAIN; 1730 do { 1731 spin_lock_bh(&queue->lock); 1732 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1733 err, &last); 1734 if (skb) { 1735 if (!(flags & MSG_PEEK)) 1736 udp_skb_destructor(sk, skb); 1737 spin_unlock_bh(&queue->lock); 1738 return skb; 1739 } 1740 1741 if (skb_queue_empty_lockless(sk_queue)) { 1742 spin_unlock_bh(&queue->lock); 1743 goto busy_check; 1744 } 1745 1746 /* refill the reader queue and walk it again 1747 * keep both queues locked to avoid re-acquiring 1748 * the sk_receive_queue lock if fwd memory scheduling 1749 * is needed. 1750 */ 1751 spin_lock(&sk_queue->lock); 1752 skb_queue_splice_tail_init(sk_queue, queue); 1753 1754 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1755 err, &last); 1756 if (skb && !(flags & MSG_PEEK)) 1757 udp_skb_dtor_locked(sk, skb); 1758 spin_unlock(&sk_queue->lock); 1759 spin_unlock_bh(&queue->lock); 1760 if (skb) 1761 return skb; 1762 1763 busy_check: 1764 if (!sk_can_busy_loop(sk)) 1765 break; 1766 1767 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1768 } while (!skb_queue_empty_lockless(sk_queue)); 1769 1770 /* sk_queue is empty, reader_queue may contain peeked packets */ 1771 } while (timeo && 1772 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1773 &error, &timeo, 1774 (struct sk_buff *)sk_queue)); 1775 1776 *err = error; 1777 return NULL; 1778 } 1779 EXPORT_SYMBOL(__skb_recv_udp); 1780 1781 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) 1782 { 1783 struct sk_buff *skb; 1784 int err; 1785 1786 try_again: 1787 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1788 if (!skb) 1789 return err; 1790 1791 if (udp_lib_checksum_complete(skb)) { 1792 int is_udplite = IS_UDPLITE(sk); 1793 struct net *net = sock_net(sk); 1794 1795 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite); 1796 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite); 1797 atomic_inc(&sk->sk_drops); 1798 kfree_skb(skb); 1799 goto try_again; 1800 } 1801 1802 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); 1803 return recv_actor(sk, skb); 1804 } 1805 EXPORT_SYMBOL(udp_read_skb); 1806 1807 /* 1808 * This should be easy, if there is something there we 1809 * return it, otherwise we block. 1810 */ 1811 1812 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 1813 int *addr_len) 1814 { 1815 struct inet_sock *inet = inet_sk(sk); 1816 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1817 struct sk_buff *skb; 1818 unsigned int ulen, copied; 1819 int off, err, peeking = flags & MSG_PEEK; 1820 int is_udplite = IS_UDPLITE(sk); 1821 bool checksum_valid = false; 1822 1823 if (flags & MSG_ERRQUEUE) 1824 return ip_recv_error(sk, msg, len, addr_len); 1825 1826 try_again: 1827 off = sk_peek_offset(sk, flags); 1828 skb = __skb_recv_udp(sk, flags, &off, &err); 1829 if (!skb) 1830 return err; 1831 1832 ulen = udp_skb_len(skb); 1833 copied = len; 1834 if (copied > ulen - off) 1835 copied = ulen - off; 1836 else if (copied < ulen) 1837 msg->msg_flags |= MSG_TRUNC; 1838 1839 /* 1840 * If checksum is needed at all, try to do it while copying the 1841 * data. If the data is truncated, or if we only want a partial 1842 * coverage checksum (UDP-Lite), do it before the copy. 1843 */ 1844 1845 if (copied < ulen || peeking || 1846 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1847 checksum_valid = udp_skb_csum_unnecessary(skb) || 1848 !__udp_lib_checksum_complete(skb); 1849 if (!checksum_valid) 1850 goto csum_copy_err; 1851 } 1852 1853 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1854 if (udp_skb_is_linear(skb)) 1855 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1856 else 1857 err = skb_copy_datagram_msg(skb, off, msg, copied); 1858 } else { 1859 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1860 1861 if (err == -EINVAL) 1862 goto csum_copy_err; 1863 } 1864 1865 if (unlikely(err)) { 1866 if (!peeking) { 1867 atomic_inc(&sk->sk_drops); 1868 UDP_INC_STATS(sock_net(sk), 1869 UDP_MIB_INERRORS, is_udplite); 1870 } 1871 kfree_skb(skb); 1872 return err; 1873 } 1874 1875 if (!peeking) 1876 UDP_INC_STATS(sock_net(sk), 1877 UDP_MIB_INDATAGRAMS, is_udplite); 1878 1879 sock_recv_cmsgs(msg, sk, skb); 1880 1881 /* Copy the address. */ 1882 if (sin) { 1883 sin->sin_family = AF_INET; 1884 sin->sin_port = udp_hdr(skb)->source; 1885 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1886 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1887 *addr_len = sizeof(*sin); 1888 1889 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1890 (struct sockaddr *)sin, 1891 addr_len); 1892 } 1893 1894 if (udp_test_bit(GRO_ENABLED, sk)) 1895 udp_cmsg_recv(msg, sk, skb); 1896 1897 if (inet_cmsg_flags(inet)) 1898 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1899 1900 err = copied; 1901 if (flags & MSG_TRUNC) 1902 err = ulen; 1903 1904 skb_consume_udp(sk, skb, peeking ? -err : err); 1905 return err; 1906 1907 csum_copy_err: 1908 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1909 udp_skb_destructor)) { 1910 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1911 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1912 } 1913 kfree_skb(skb); 1914 1915 /* starting over for a new packet, but check if we need to yield */ 1916 cond_resched(); 1917 msg->msg_flags &= ~MSG_TRUNC; 1918 goto try_again; 1919 } 1920 1921 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1922 { 1923 /* This check is replicated from __ip4_datagram_connect() and 1924 * intended to prevent BPF program called below from accessing bytes 1925 * that are out of the bound specified by user in addr_len. 1926 */ 1927 if (addr_len < sizeof(struct sockaddr_in)) 1928 return -EINVAL; 1929 1930 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len); 1931 } 1932 EXPORT_SYMBOL(udp_pre_connect); 1933 1934 int __udp_disconnect(struct sock *sk, int flags) 1935 { 1936 struct inet_sock *inet = inet_sk(sk); 1937 /* 1938 * 1003.1g - break association. 1939 */ 1940 1941 sk->sk_state = TCP_CLOSE; 1942 inet->inet_daddr = 0; 1943 inet->inet_dport = 0; 1944 sock_rps_reset_rxhash(sk); 1945 sk->sk_bound_dev_if = 0; 1946 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1947 inet_reset_saddr(sk); 1948 if (sk->sk_prot->rehash && 1949 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1950 sk->sk_prot->rehash(sk); 1951 } 1952 1953 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1954 sk->sk_prot->unhash(sk); 1955 inet->inet_sport = 0; 1956 } 1957 sk_dst_reset(sk); 1958 return 0; 1959 } 1960 EXPORT_SYMBOL(__udp_disconnect); 1961 1962 int udp_disconnect(struct sock *sk, int flags) 1963 { 1964 lock_sock(sk); 1965 __udp_disconnect(sk, flags); 1966 release_sock(sk); 1967 return 0; 1968 } 1969 EXPORT_SYMBOL(udp_disconnect); 1970 1971 void udp_lib_unhash(struct sock *sk) 1972 { 1973 if (sk_hashed(sk)) { 1974 struct udp_table *udptable = udp_get_table_prot(sk); 1975 struct udp_hslot *hslot, *hslot2; 1976 1977 hslot = udp_hashslot(udptable, sock_net(sk), 1978 udp_sk(sk)->udp_port_hash); 1979 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1980 1981 spin_lock_bh(&hslot->lock); 1982 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1983 reuseport_detach_sock(sk); 1984 if (sk_del_node_init_rcu(sk)) { 1985 hslot->count--; 1986 inet_sk(sk)->inet_num = 0; 1987 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1988 1989 spin_lock(&hslot2->lock); 1990 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1991 hslot2->count--; 1992 spin_unlock(&hslot2->lock); 1993 } 1994 spin_unlock_bh(&hslot->lock); 1995 } 1996 } 1997 EXPORT_SYMBOL(udp_lib_unhash); 1998 1999 /* 2000 * inet_rcv_saddr was changed, we must rehash secondary hash 2001 */ 2002 void udp_lib_rehash(struct sock *sk, u16 newhash) 2003 { 2004 if (sk_hashed(sk)) { 2005 struct udp_table *udptable = udp_get_table_prot(sk); 2006 struct udp_hslot *hslot, *hslot2, *nhslot2; 2007 2008 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2009 nhslot2 = udp_hashslot2(udptable, newhash); 2010 udp_sk(sk)->udp_portaddr_hash = newhash; 2011 2012 if (hslot2 != nhslot2 || 2013 rcu_access_pointer(sk->sk_reuseport_cb)) { 2014 hslot = udp_hashslot(udptable, sock_net(sk), 2015 udp_sk(sk)->udp_port_hash); 2016 /* we must lock primary chain too */ 2017 spin_lock_bh(&hslot->lock); 2018 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2019 reuseport_detach_sock(sk); 2020 2021 if (hslot2 != nhslot2) { 2022 spin_lock(&hslot2->lock); 2023 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2024 hslot2->count--; 2025 spin_unlock(&hslot2->lock); 2026 2027 spin_lock(&nhslot2->lock); 2028 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2029 &nhslot2->head); 2030 nhslot2->count++; 2031 spin_unlock(&nhslot2->lock); 2032 } 2033 2034 spin_unlock_bh(&hslot->lock); 2035 } 2036 } 2037 } 2038 EXPORT_SYMBOL(udp_lib_rehash); 2039 2040 void udp_v4_rehash(struct sock *sk) 2041 { 2042 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2043 inet_sk(sk)->inet_rcv_saddr, 2044 inet_sk(sk)->inet_num); 2045 udp_lib_rehash(sk, new_hash); 2046 } 2047 2048 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2049 { 2050 int rc; 2051 2052 if (inet_sk(sk)->inet_daddr) { 2053 sock_rps_save_rxhash(sk, skb); 2054 sk_mark_napi_id(sk, skb); 2055 sk_incoming_cpu_update(sk); 2056 } else { 2057 sk_mark_napi_id_once(sk, skb); 2058 } 2059 2060 rc = __udp_enqueue_schedule_skb(sk, skb); 2061 if (rc < 0) { 2062 int is_udplite = IS_UDPLITE(sk); 2063 int drop_reason; 2064 2065 /* Note that an ENOMEM error is charged twice */ 2066 if (rc == -ENOMEM) { 2067 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2068 is_udplite); 2069 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 2070 } else { 2071 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2072 is_udplite); 2073 drop_reason = SKB_DROP_REASON_PROTO_MEM; 2074 } 2075 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2076 trace_udp_fail_queue_rcv_skb(rc, sk, skb); 2077 kfree_skb_reason(skb, drop_reason); 2078 return -1; 2079 } 2080 2081 return 0; 2082 } 2083 2084 /* returns: 2085 * -1: error 2086 * 0: success 2087 * >0: "udp encap" protocol resubmission 2088 * 2089 * Note that in the success and error cases, the skb is assumed to 2090 * have either been requeued or freed. 2091 */ 2092 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2093 { 2094 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2095 struct udp_sock *up = udp_sk(sk); 2096 int is_udplite = IS_UDPLITE(sk); 2097 2098 /* 2099 * Charge it to the socket, dropping if the queue is full. 2100 */ 2101 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 2102 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2103 goto drop; 2104 } 2105 nf_reset_ct(skb); 2106 2107 if (static_branch_unlikely(&udp_encap_needed_key) && 2108 READ_ONCE(up->encap_type)) { 2109 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2110 2111 /* 2112 * This is an encapsulation socket so pass the skb to 2113 * the socket's udp_encap_rcv() hook. Otherwise, just 2114 * fall through and pass this up the UDP socket. 2115 * up->encap_rcv() returns the following value: 2116 * =0 if skb was successfully passed to the encap 2117 * handler or was discarded by it. 2118 * >0 if skb should be passed on to UDP. 2119 * <0 if skb should be resubmitted as proto -N 2120 */ 2121 2122 /* if we're overly short, let UDP handle it */ 2123 encap_rcv = READ_ONCE(up->encap_rcv); 2124 if (encap_rcv) { 2125 int ret; 2126 2127 /* Verify checksum before giving to encap */ 2128 if (udp_lib_checksum_complete(skb)) 2129 goto csum_error; 2130 2131 ret = encap_rcv(sk, skb); 2132 if (ret <= 0) { 2133 __UDP_INC_STATS(sock_net(sk), 2134 UDP_MIB_INDATAGRAMS, 2135 is_udplite); 2136 return -ret; 2137 } 2138 } 2139 2140 /* FALLTHROUGH -- it's a UDP Packet */ 2141 } 2142 2143 /* 2144 * UDP-Lite specific tests, ignored on UDP sockets 2145 */ 2146 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) { 2147 u16 pcrlen = READ_ONCE(up->pcrlen); 2148 2149 /* 2150 * MIB statistics other than incrementing the error count are 2151 * disabled for the following two types of errors: these depend 2152 * on the application settings, not on the functioning of the 2153 * protocol stack as such. 2154 * 2155 * RFC 3828 here recommends (sec 3.3): "There should also be a 2156 * way ... to ... at least let the receiving application block 2157 * delivery of packets with coverage values less than a value 2158 * provided by the application." 2159 */ 2160 if (pcrlen == 0) { /* full coverage was set */ 2161 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2162 UDP_SKB_CB(skb)->cscov, skb->len); 2163 goto drop; 2164 } 2165 /* The next case involves violating the min. coverage requested 2166 * by the receiver. This is subtle: if receiver wants x and x is 2167 * greater than the buffersize/MTU then receiver will complain 2168 * that it wants x while sender emits packets of smaller size y. 2169 * Therefore the above ...()->partial_cov statement is essential. 2170 */ 2171 if (UDP_SKB_CB(skb)->cscov < pcrlen) { 2172 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2173 UDP_SKB_CB(skb)->cscov, pcrlen); 2174 goto drop; 2175 } 2176 } 2177 2178 prefetch(&sk->sk_rmem_alloc); 2179 if (rcu_access_pointer(sk->sk_filter) && 2180 udp_lib_checksum_complete(skb)) 2181 goto csum_error; 2182 2183 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 2184 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 2185 goto drop; 2186 } 2187 2188 udp_csum_pull_header(skb); 2189 2190 ipv4_pktinfo_prepare(sk, skb, true); 2191 return __udp_queue_rcv_skb(sk, skb); 2192 2193 csum_error: 2194 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2195 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2196 drop: 2197 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2198 atomic_inc(&sk->sk_drops); 2199 kfree_skb_reason(skb, drop_reason); 2200 return -1; 2201 } 2202 2203 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2204 { 2205 struct sk_buff *next, *segs; 2206 int ret; 2207 2208 if (likely(!udp_unexpected_gso(sk, skb))) 2209 return udp_queue_rcv_one_skb(sk, skb); 2210 2211 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2212 __skb_push(skb, -skb_mac_offset(skb)); 2213 segs = udp_rcv_segment(sk, skb, true); 2214 skb_list_walk_safe(segs, skb, next) { 2215 __skb_pull(skb, skb_transport_offset(skb)); 2216 2217 udp_post_segment_fix_csum(skb); 2218 ret = udp_queue_rcv_one_skb(sk, skb); 2219 if (ret > 0) 2220 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2221 } 2222 return 0; 2223 } 2224 2225 /* For TCP sockets, sk_rx_dst is protected by socket lock 2226 * For UDP, we use xchg() to guard against concurrent changes. 2227 */ 2228 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2229 { 2230 struct dst_entry *old; 2231 2232 if (dst_hold_safe(dst)) { 2233 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); 2234 dst_release(old); 2235 return old != dst; 2236 } 2237 return false; 2238 } 2239 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2240 2241 /* 2242 * Multicasts and broadcasts go to each listener. 2243 * 2244 * Note: called only from the BH handler context. 2245 */ 2246 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2247 struct udphdr *uh, 2248 __be32 saddr, __be32 daddr, 2249 struct udp_table *udptable, 2250 int proto) 2251 { 2252 struct sock *sk, *first = NULL; 2253 unsigned short hnum = ntohs(uh->dest); 2254 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2255 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2256 unsigned int offset = offsetof(typeof(*sk), sk_node); 2257 int dif = skb->dev->ifindex; 2258 int sdif = inet_sdif(skb); 2259 struct hlist_node *node; 2260 struct sk_buff *nskb; 2261 2262 if (use_hash2) { 2263 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2264 udptable->mask; 2265 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2266 start_lookup: 2267 hslot = &udptable->hash2[hash2]; 2268 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2269 } 2270 2271 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2272 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2273 uh->source, saddr, dif, sdif, hnum)) 2274 continue; 2275 2276 if (!first) { 2277 first = sk; 2278 continue; 2279 } 2280 nskb = skb_clone(skb, GFP_ATOMIC); 2281 2282 if (unlikely(!nskb)) { 2283 atomic_inc(&sk->sk_drops); 2284 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2285 IS_UDPLITE(sk)); 2286 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2287 IS_UDPLITE(sk)); 2288 continue; 2289 } 2290 if (udp_queue_rcv_skb(sk, nskb) > 0) 2291 consume_skb(nskb); 2292 } 2293 2294 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2295 if (use_hash2 && hash2 != hash2_any) { 2296 hash2 = hash2_any; 2297 goto start_lookup; 2298 } 2299 2300 if (first) { 2301 if (udp_queue_rcv_skb(first, skb) > 0) 2302 consume_skb(skb); 2303 } else { 2304 kfree_skb(skb); 2305 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2306 proto == IPPROTO_UDPLITE); 2307 } 2308 return 0; 2309 } 2310 2311 /* Initialize UDP checksum. If exited with zero value (success), 2312 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2313 * Otherwise, csum completion requires checksumming packet body, 2314 * including udp header and folding it to skb->csum. 2315 */ 2316 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2317 int proto) 2318 { 2319 int err; 2320 2321 UDP_SKB_CB(skb)->partial_cov = 0; 2322 UDP_SKB_CB(skb)->cscov = skb->len; 2323 2324 if (proto == IPPROTO_UDPLITE) { 2325 err = udplite_checksum_init(skb, uh); 2326 if (err) 2327 return err; 2328 2329 if (UDP_SKB_CB(skb)->partial_cov) { 2330 skb->csum = inet_compute_pseudo(skb, proto); 2331 return 0; 2332 } 2333 } 2334 2335 /* Note, we are only interested in != 0 or == 0, thus the 2336 * force to int. 2337 */ 2338 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2339 inet_compute_pseudo); 2340 if (err) 2341 return err; 2342 2343 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2344 /* If SW calculated the value, we know it's bad */ 2345 if (skb->csum_complete_sw) 2346 return 1; 2347 2348 /* HW says the value is bad. Let's validate that. 2349 * skb->csum is no longer the full packet checksum, 2350 * so don't treat it as such. 2351 */ 2352 skb_checksum_complete_unset(skb); 2353 } 2354 2355 return 0; 2356 } 2357 2358 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2359 * return code conversion for ip layer consumption 2360 */ 2361 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2362 struct udphdr *uh) 2363 { 2364 int ret; 2365 2366 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2367 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2368 2369 ret = udp_queue_rcv_skb(sk, skb); 2370 2371 /* a return value > 0 means to resubmit the input, but 2372 * it wants the return to be -protocol, or 0 2373 */ 2374 if (ret > 0) 2375 return -ret; 2376 return 0; 2377 } 2378 2379 /* 2380 * All we need to do is get the socket, and then do a checksum. 2381 */ 2382 2383 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2384 int proto) 2385 { 2386 struct sock *sk; 2387 struct udphdr *uh; 2388 unsigned short ulen; 2389 struct rtable *rt = skb_rtable(skb); 2390 __be32 saddr, daddr; 2391 struct net *net = dev_net(skb->dev); 2392 bool refcounted; 2393 int drop_reason; 2394 2395 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2396 2397 /* 2398 * Validate the packet. 2399 */ 2400 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2401 goto drop; /* No space for header. */ 2402 2403 uh = udp_hdr(skb); 2404 ulen = ntohs(uh->len); 2405 saddr = ip_hdr(skb)->saddr; 2406 daddr = ip_hdr(skb)->daddr; 2407 2408 if (ulen > skb->len) 2409 goto short_packet; 2410 2411 if (proto == IPPROTO_UDP) { 2412 /* UDP validates ulen. */ 2413 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2414 goto short_packet; 2415 uh = udp_hdr(skb); 2416 } 2417 2418 if (udp4_csum_init(skb, uh, proto)) 2419 goto csum_error; 2420 2421 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 2422 &refcounted, udp_ehashfn); 2423 if (IS_ERR(sk)) 2424 goto no_sk; 2425 2426 if (sk) { 2427 struct dst_entry *dst = skb_dst(skb); 2428 int ret; 2429 2430 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2431 udp_sk_rx_dst_set(sk, dst); 2432 2433 ret = udp_unicast_rcv_skb(sk, skb, uh); 2434 if (refcounted) 2435 sock_put(sk); 2436 return ret; 2437 } 2438 2439 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2440 return __udp4_lib_mcast_deliver(net, skb, uh, 2441 saddr, daddr, udptable, proto); 2442 2443 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2444 if (sk) 2445 return udp_unicast_rcv_skb(sk, skb, uh); 2446 no_sk: 2447 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2448 goto drop; 2449 nf_reset_ct(skb); 2450 2451 /* No socket. Drop packet silently, if checksum is wrong */ 2452 if (udp_lib_checksum_complete(skb)) 2453 goto csum_error; 2454 2455 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2456 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2457 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2458 2459 /* 2460 * Hmm. We got an UDP packet to a port to which we 2461 * don't wanna listen. Ignore it. 2462 */ 2463 kfree_skb_reason(skb, drop_reason); 2464 return 0; 2465 2466 short_packet: 2467 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2468 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2469 proto == IPPROTO_UDPLITE ? "Lite" : "", 2470 &saddr, ntohs(uh->source), 2471 ulen, skb->len, 2472 &daddr, ntohs(uh->dest)); 2473 goto drop; 2474 2475 csum_error: 2476 /* 2477 * RFC1122: OK. Discards the bad packet silently (as far as 2478 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2479 */ 2480 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2481 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2482 proto == IPPROTO_UDPLITE ? "Lite" : "", 2483 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2484 ulen); 2485 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2486 drop: 2487 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2488 kfree_skb_reason(skb, drop_reason); 2489 return 0; 2490 } 2491 2492 /* We can only early demux multicast if there is a single matching socket. 2493 * If more than one socket found returns NULL 2494 */ 2495 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2496 __be16 loc_port, __be32 loc_addr, 2497 __be16 rmt_port, __be32 rmt_addr, 2498 int dif, int sdif) 2499 { 2500 struct udp_table *udptable = net->ipv4.udp_table; 2501 unsigned short hnum = ntohs(loc_port); 2502 struct sock *sk, *result; 2503 struct udp_hslot *hslot; 2504 unsigned int slot; 2505 2506 slot = udp_hashfn(net, hnum, udptable->mask); 2507 hslot = &udptable->hash[slot]; 2508 2509 /* Do not bother scanning a too big list */ 2510 if (hslot->count > 10) 2511 return NULL; 2512 2513 result = NULL; 2514 sk_for_each_rcu(sk, &hslot->head) { 2515 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2516 rmt_port, rmt_addr, dif, sdif, hnum)) { 2517 if (result) 2518 return NULL; 2519 result = sk; 2520 } 2521 } 2522 2523 return result; 2524 } 2525 2526 /* For unicast we should only early demux connected sockets or we can 2527 * break forwarding setups. The chains here can be long so only check 2528 * if the first socket is an exact match and if not move on. 2529 */ 2530 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2531 __be16 loc_port, __be32 loc_addr, 2532 __be16 rmt_port, __be32 rmt_addr, 2533 int dif, int sdif) 2534 { 2535 struct udp_table *udptable = net->ipv4.udp_table; 2536 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2537 unsigned short hnum = ntohs(loc_port); 2538 unsigned int hash2, slot2; 2539 struct udp_hslot *hslot2; 2540 __portpair ports; 2541 struct sock *sk; 2542 2543 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2544 slot2 = hash2 & udptable->mask; 2545 hslot2 = &udptable->hash2[slot2]; 2546 ports = INET_COMBINED_PORTS(rmt_port, hnum); 2547 2548 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2549 if (inet_match(net, sk, acookie, ports, dif, sdif)) 2550 return sk; 2551 /* Only check first socket in chain */ 2552 break; 2553 } 2554 return NULL; 2555 } 2556 2557 int udp_v4_early_demux(struct sk_buff *skb) 2558 { 2559 struct net *net = dev_net(skb->dev); 2560 struct in_device *in_dev = NULL; 2561 const struct iphdr *iph; 2562 const struct udphdr *uh; 2563 struct sock *sk = NULL; 2564 struct dst_entry *dst; 2565 int dif = skb->dev->ifindex; 2566 int sdif = inet_sdif(skb); 2567 int ours; 2568 2569 /* validate the packet */ 2570 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2571 return 0; 2572 2573 iph = ip_hdr(skb); 2574 uh = udp_hdr(skb); 2575 2576 if (skb->pkt_type == PACKET_MULTICAST) { 2577 in_dev = __in_dev_get_rcu(skb->dev); 2578 2579 if (!in_dev) 2580 return 0; 2581 2582 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2583 iph->protocol); 2584 if (!ours) 2585 return 0; 2586 2587 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2588 uh->source, iph->saddr, 2589 dif, sdif); 2590 } else if (skb->pkt_type == PACKET_HOST) { 2591 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2592 uh->source, iph->saddr, dif, sdif); 2593 } 2594 2595 if (!sk) 2596 return 0; 2597 2598 skb->sk = sk; 2599 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk)); 2600 skb->destructor = sock_pfree; 2601 dst = rcu_dereference(sk->sk_rx_dst); 2602 2603 if (dst) 2604 dst = dst_check(dst, 0); 2605 if (dst) { 2606 u32 itag = 0; 2607 2608 /* set noref for now. 2609 * any place which wants to hold dst has to call 2610 * dst_hold_safe() 2611 */ 2612 skb_dst_set_noref(skb, dst); 2613 2614 /* for unconnected multicast sockets we need to validate 2615 * the source on each packet 2616 */ 2617 if (!inet_sk(sk)->inet_daddr && in_dev) 2618 return ip_mc_validate_source(skb, iph->daddr, 2619 iph->saddr, 2620 iph->tos & IPTOS_RT_MASK, 2621 skb->dev, in_dev, &itag); 2622 } 2623 return 0; 2624 } 2625 2626 int udp_rcv(struct sk_buff *skb) 2627 { 2628 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 2629 } 2630 2631 void udp_destroy_sock(struct sock *sk) 2632 { 2633 struct udp_sock *up = udp_sk(sk); 2634 bool slow = lock_sock_fast(sk); 2635 2636 /* protects from races with udp_abort() */ 2637 sock_set_flag(sk, SOCK_DEAD); 2638 udp_flush_pending_frames(sk); 2639 unlock_sock_fast(sk, slow); 2640 if (static_branch_unlikely(&udp_encap_needed_key)) { 2641 if (up->encap_type) { 2642 void (*encap_destroy)(struct sock *sk); 2643 encap_destroy = READ_ONCE(up->encap_destroy); 2644 if (encap_destroy) 2645 encap_destroy(sk); 2646 } 2647 if (udp_test_bit(ENCAP_ENABLED, sk)) 2648 static_branch_dec(&udp_encap_needed_key); 2649 } 2650 } 2651 2652 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family, 2653 struct sock *sk) 2654 { 2655 #ifdef CONFIG_XFRM 2656 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) { 2657 if (family == AF_INET) 2658 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv); 2659 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) 2660 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv); 2661 } 2662 #endif 2663 } 2664 2665 /* 2666 * Socket option code for UDP 2667 */ 2668 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2669 sockptr_t optval, unsigned int optlen, 2670 int (*push_pending_frames)(struct sock *)) 2671 { 2672 struct udp_sock *up = udp_sk(sk); 2673 int val, valbool; 2674 int err = 0; 2675 int is_udplite = IS_UDPLITE(sk); 2676 2677 if (level == SOL_SOCKET) { 2678 err = sk_setsockopt(sk, level, optname, optval, optlen); 2679 2680 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) { 2681 sockopt_lock_sock(sk); 2682 /* paired with READ_ONCE in udp_rmem_release() */ 2683 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2); 2684 sockopt_release_sock(sk); 2685 } 2686 return err; 2687 } 2688 2689 if (optlen < sizeof(int)) 2690 return -EINVAL; 2691 2692 if (copy_from_sockptr(&val, optval, sizeof(val))) 2693 return -EFAULT; 2694 2695 valbool = val ? 1 : 0; 2696 2697 switch (optname) { 2698 case UDP_CORK: 2699 if (val != 0) { 2700 udp_set_bit(CORK, sk); 2701 } else { 2702 udp_clear_bit(CORK, sk); 2703 lock_sock(sk); 2704 push_pending_frames(sk); 2705 release_sock(sk); 2706 } 2707 break; 2708 2709 case UDP_ENCAP: 2710 switch (val) { 2711 case 0: 2712 #ifdef CONFIG_XFRM 2713 case UDP_ENCAP_ESPINUDP: 2714 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk); 2715 #if IS_ENABLED(CONFIG_IPV6) 2716 if (sk->sk_family == AF_INET6) 2717 WRITE_ONCE(up->encap_rcv, 2718 ipv6_stub->xfrm6_udp_encap_rcv); 2719 else 2720 #endif 2721 WRITE_ONCE(up->encap_rcv, 2722 xfrm4_udp_encap_rcv); 2723 #endif 2724 fallthrough; 2725 case UDP_ENCAP_L2TPINUDP: 2726 WRITE_ONCE(up->encap_type, val); 2727 udp_tunnel_encap_enable(sk); 2728 break; 2729 default: 2730 err = -ENOPROTOOPT; 2731 break; 2732 } 2733 break; 2734 2735 case UDP_NO_CHECK6_TX: 2736 udp_set_no_check6_tx(sk, valbool); 2737 break; 2738 2739 case UDP_NO_CHECK6_RX: 2740 udp_set_no_check6_rx(sk, valbool); 2741 break; 2742 2743 case UDP_SEGMENT: 2744 if (val < 0 || val > USHRT_MAX) 2745 return -EINVAL; 2746 WRITE_ONCE(up->gso_size, val); 2747 break; 2748 2749 case UDP_GRO: 2750 2751 /* when enabling GRO, accept the related GSO packet type */ 2752 if (valbool) 2753 udp_tunnel_encap_enable(sk); 2754 udp_assign_bit(GRO_ENABLED, sk, valbool); 2755 udp_assign_bit(ACCEPT_L4, sk, valbool); 2756 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk); 2757 break; 2758 2759 /* 2760 * UDP-Lite's partial checksum coverage (RFC 3828). 2761 */ 2762 /* The sender sets actual checksum coverage length via this option. 2763 * The case coverage > packet length is handled by send module. */ 2764 case UDPLITE_SEND_CSCOV: 2765 if (!is_udplite) /* Disable the option on UDP sockets */ 2766 return -ENOPROTOOPT; 2767 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2768 val = 8; 2769 else if (val > USHRT_MAX) 2770 val = USHRT_MAX; 2771 WRITE_ONCE(up->pcslen, val); 2772 udp_set_bit(UDPLITE_SEND_CC, sk); 2773 break; 2774 2775 /* The receiver specifies a minimum checksum coverage value. To make 2776 * sense, this should be set to at least 8 (as done below). If zero is 2777 * used, this again means full checksum coverage. */ 2778 case UDPLITE_RECV_CSCOV: 2779 if (!is_udplite) /* Disable the option on UDP sockets */ 2780 return -ENOPROTOOPT; 2781 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2782 val = 8; 2783 else if (val > USHRT_MAX) 2784 val = USHRT_MAX; 2785 WRITE_ONCE(up->pcrlen, val); 2786 udp_set_bit(UDPLITE_RECV_CC, sk); 2787 break; 2788 2789 default: 2790 err = -ENOPROTOOPT; 2791 break; 2792 } 2793 2794 return err; 2795 } 2796 EXPORT_SYMBOL(udp_lib_setsockopt); 2797 2798 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2799 unsigned int optlen) 2800 { 2801 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 2802 return udp_lib_setsockopt(sk, level, optname, 2803 optval, optlen, 2804 udp_push_pending_frames); 2805 return ip_setsockopt(sk, level, optname, optval, optlen); 2806 } 2807 2808 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2809 char __user *optval, int __user *optlen) 2810 { 2811 struct udp_sock *up = udp_sk(sk); 2812 int val, len; 2813 2814 if (get_user(len, optlen)) 2815 return -EFAULT; 2816 2817 if (len < 0) 2818 return -EINVAL; 2819 2820 len = min_t(unsigned int, len, sizeof(int)); 2821 2822 switch (optname) { 2823 case UDP_CORK: 2824 val = udp_test_bit(CORK, sk); 2825 break; 2826 2827 case UDP_ENCAP: 2828 val = READ_ONCE(up->encap_type); 2829 break; 2830 2831 case UDP_NO_CHECK6_TX: 2832 val = udp_get_no_check6_tx(sk); 2833 break; 2834 2835 case UDP_NO_CHECK6_RX: 2836 val = udp_get_no_check6_rx(sk); 2837 break; 2838 2839 case UDP_SEGMENT: 2840 val = READ_ONCE(up->gso_size); 2841 break; 2842 2843 case UDP_GRO: 2844 val = udp_test_bit(GRO_ENABLED, sk); 2845 break; 2846 2847 /* The following two cannot be changed on UDP sockets, the return is 2848 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2849 case UDPLITE_SEND_CSCOV: 2850 val = READ_ONCE(up->pcslen); 2851 break; 2852 2853 case UDPLITE_RECV_CSCOV: 2854 val = READ_ONCE(up->pcrlen); 2855 break; 2856 2857 default: 2858 return -ENOPROTOOPT; 2859 } 2860 2861 if (put_user(len, optlen)) 2862 return -EFAULT; 2863 if (copy_to_user(optval, &val, len)) 2864 return -EFAULT; 2865 return 0; 2866 } 2867 EXPORT_SYMBOL(udp_lib_getsockopt); 2868 2869 int udp_getsockopt(struct sock *sk, int level, int optname, 2870 char __user *optval, int __user *optlen) 2871 { 2872 if (level == SOL_UDP || level == SOL_UDPLITE) 2873 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2874 return ip_getsockopt(sk, level, optname, optval, optlen); 2875 } 2876 2877 /** 2878 * udp_poll - wait for a UDP event. 2879 * @file: - file struct 2880 * @sock: - socket 2881 * @wait: - poll table 2882 * 2883 * This is same as datagram poll, except for the special case of 2884 * blocking sockets. If application is using a blocking fd 2885 * and a packet with checksum error is in the queue; 2886 * then it could get return from select indicating data available 2887 * but then block when reading it. Add special case code 2888 * to work around these arguably broken applications. 2889 */ 2890 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2891 { 2892 __poll_t mask = datagram_poll(file, sock, wait); 2893 struct sock *sk = sock->sk; 2894 2895 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2896 mask |= EPOLLIN | EPOLLRDNORM; 2897 2898 /* Check for false positives due to checksum errors */ 2899 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2900 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2901 mask &= ~(EPOLLIN | EPOLLRDNORM); 2902 2903 /* psock ingress_msg queue should not contain any bad checksum frames */ 2904 if (sk_is_readable(sk)) 2905 mask |= EPOLLIN | EPOLLRDNORM; 2906 return mask; 2907 2908 } 2909 EXPORT_SYMBOL(udp_poll); 2910 2911 int udp_abort(struct sock *sk, int err) 2912 { 2913 if (!has_current_bpf_ctx()) 2914 lock_sock(sk); 2915 2916 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2917 * with close() 2918 */ 2919 if (sock_flag(sk, SOCK_DEAD)) 2920 goto out; 2921 2922 sk->sk_err = err; 2923 sk_error_report(sk); 2924 __udp_disconnect(sk, 0); 2925 2926 out: 2927 if (!has_current_bpf_ctx()) 2928 release_sock(sk); 2929 2930 return 0; 2931 } 2932 EXPORT_SYMBOL_GPL(udp_abort); 2933 2934 struct proto udp_prot = { 2935 .name = "UDP", 2936 .owner = THIS_MODULE, 2937 .close = udp_lib_close, 2938 .pre_connect = udp_pre_connect, 2939 .connect = ip4_datagram_connect, 2940 .disconnect = udp_disconnect, 2941 .ioctl = udp_ioctl, 2942 .init = udp_init_sock, 2943 .destroy = udp_destroy_sock, 2944 .setsockopt = udp_setsockopt, 2945 .getsockopt = udp_getsockopt, 2946 .sendmsg = udp_sendmsg, 2947 .recvmsg = udp_recvmsg, 2948 .splice_eof = udp_splice_eof, 2949 .release_cb = ip4_datagram_release_cb, 2950 .hash = udp_lib_hash, 2951 .unhash = udp_lib_unhash, 2952 .rehash = udp_v4_rehash, 2953 .get_port = udp_v4_get_port, 2954 .put_port = udp_lib_unhash, 2955 #ifdef CONFIG_BPF_SYSCALL 2956 .psock_update_sk_prot = udp_bpf_update_proto, 2957 #endif 2958 .memory_allocated = &udp_memory_allocated, 2959 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 2960 2961 .sysctl_mem = sysctl_udp_mem, 2962 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2963 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2964 .obj_size = sizeof(struct udp_sock), 2965 .h.udp_table = NULL, 2966 .diag_destroy = udp_abort, 2967 }; 2968 EXPORT_SYMBOL(udp_prot); 2969 2970 /* ------------------------------------------------------------------------ */ 2971 #ifdef CONFIG_PROC_FS 2972 2973 static unsigned short seq_file_family(const struct seq_file *seq); 2974 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk) 2975 { 2976 unsigned short family = seq_file_family(seq); 2977 2978 /* AF_UNSPEC is used as a match all */ 2979 return ((family == AF_UNSPEC || family == sk->sk_family) && 2980 net_eq(sock_net(sk), seq_file_net(seq))); 2981 } 2982 2983 #ifdef CONFIG_BPF_SYSCALL 2984 static const struct seq_operations bpf_iter_udp_seq_ops; 2985 #endif 2986 static struct udp_table *udp_get_table_seq(struct seq_file *seq, 2987 struct net *net) 2988 { 2989 const struct udp_seq_afinfo *afinfo; 2990 2991 #ifdef CONFIG_BPF_SYSCALL 2992 if (seq->op == &bpf_iter_udp_seq_ops) 2993 return net->ipv4.udp_table; 2994 #endif 2995 2996 afinfo = pde_data(file_inode(seq->file)); 2997 return afinfo->udp_table ? : net->ipv4.udp_table; 2998 } 2999 3000 static struct sock *udp_get_first(struct seq_file *seq, int start) 3001 { 3002 struct udp_iter_state *state = seq->private; 3003 struct net *net = seq_file_net(seq); 3004 struct udp_table *udptable; 3005 struct sock *sk; 3006 3007 udptable = udp_get_table_seq(seq, net); 3008 3009 for (state->bucket = start; state->bucket <= udptable->mask; 3010 ++state->bucket) { 3011 struct udp_hslot *hslot = &udptable->hash[state->bucket]; 3012 3013 if (hlist_empty(&hslot->head)) 3014 continue; 3015 3016 spin_lock_bh(&hslot->lock); 3017 sk_for_each(sk, &hslot->head) { 3018 if (seq_sk_match(seq, sk)) 3019 goto found; 3020 } 3021 spin_unlock_bh(&hslot->lock); 3022 } 3023 sk = NULL; 3024 found: 3025 return sk; 3026 } 3027 3028 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 3029 { 3030 struct udp_iter_state *state = seq->private; 3031 struct net *net = seq_file_net(seq); 3032 struct udp_table *udptable; 3033 3034 do { 3035 sk = sk_next(sk); 3036 } while (sk && !seq_sk_match(seq, sk)); 3037 3038 if (!sk) { 3039 udptable = udp_get_table_seq(seq, net); 3040 3041 if (state->bucket <= udptable->mask) 3042 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3043 3044 return udp_get_first(seq, state->bucket + 1); 3045 } 3046 return sk; 3047 } 3048 3049 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3050 { 3051 struct sock *sk = udp_get_first(seq, 0); 3052 3053 if (sk) 3054 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3055 --pos; 3056 return pos ? NULL : sk; 3057 } 3058 3059 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3060 { 3061 struct udp_iter_state *state = seq->private; 3062 state->bucket = MAX_UDP_PORTS; 3063 3064 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3065 } 3066 EXPORT_SYMBOL(udp_seq_start); 3067 3068 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3069 { 3070 struct sock *sk; 3071 3072 if (v == SEQ_START_TOKEN) 3073 sk = udp_get_idx(seq, 0); 3074 else 3075 sk = udp_get_next(seq, v); 3076 3077 ++*pos; 3078 return sk; 3079 } 3080 EXPORT_SYMBOL(udp_seq_next); 3081 3082 void udp_seq_stop(struct seq_file *seq, void *v) 3083 { 3084 struct udp_iter_state *state = seq->private; 3085 struct udp_table *udptable; 3086 3087 udptable = udp_get_table_seq(seq, seq_file_net(seq)); 3088 3089 if (state->bucket <= udptable->mask) 3090 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3091 } 3092 EXPORT_SYMBOL(udp_seq_stop); 3093 3094 /* ------------------------------------------------------------------------ */ 3095 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3096 int bucket) 3097 { 3098 struct inet_sock *inet = inet_sk(sp); 3099 __be32 dest = inet->inet_daddr; 3100 __be32 src = inet->inet_rcv_saddr; 3101 __u16 destp = ntohs(inet->inet_dport); 3102 __u16 srcp = ntohs(inet->inet_sport); 3103 3104 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3105 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3106 bucket, src, srcp, dest, destp, sp->sk_state, 3107 sk_wmem_alloc_get(sp), 3108 udp_rqueue_get(sp), 3109 0, 0L, 0, 3110 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3111 0, sock_i_ino(sp), 3112 refcount_read(&sp->sk_refcnt), sp, 3113 atomic_read(&sp->sk_drops)); 3114 } 3115 3116 int udp4_seq_show(struct seq_file *seq, void *v) 3117 { 3118 seq_setwidth(seq, 127); 3119 if (v == SEQ_START_TOKEN) 3120 seq_puts(seq, " sl local_address rem_address st tx_queue " 3121 "rx_queue tr tm->when retrnsmt uid timeout " 3122 "inode ref pointer drops"); 3123 else { 3124 struct udp_iter_state *state = seq->private; 3125 3126 udp4_format_sock(v, seq, state->bucket); 3127 } 3128 seq_pad(seq, '\n'); 3129 return 0; 3130 } 3131 3132 #ifdef CONFIG_BPF_SYSCALL 3133 struct bpf_iter__udp { 3134 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3135 __bpf_md_ptr(struct udp_sock *, udp_sk); 3136 uid_t uid __aligned(8); 3137 int bucket __aligned(8); 3138 }; 3139 3140 struct bpf_udp_iter_state { 3141 struct udp_iter_state state; 3142 unsigned int cur_sk; 3143 unsigned int end_sk; 3144 unsigned int max_sk; 3145 int offset; 3146 struct sock **batch; 3147 bool st_bucket_done; 3148 }; 3149 3150 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3151 unsigned int new_batch_sz); 3152 static struct sock *bpf_iter_udp_batch(struct seq_file *seq) 3153 { 3154 struct bpf_udp_iter_state *iter = seq->private; 3155 struct udp_iter_state *state = &iter->state; 3156 struct net *net = seq_file_net(seq); 3157 int resume_bucket, resume_offset; 3158 struct udp_table *udptable; 3159 unsigned int batch_sks = 0; 3160 bool resized = false; 3161 struct sock *sk; 3162 3163 resume_bucket = state->bucket; 3164 resume_offset = iter->offset; 3165 3166 /* The current batch is done, so advance the bucket. */ 3167 if (iter->st_bucket_done) 3168 state->bucket++; 3169 3170 udptable = udp_get_table_seq(seq, net); 3171 3172 again: 3173 /* New batch for the next bucket. 3174 * Iterate over the hash table to find a bucket with sockets matching 3175 * the iterator attributes, and return the first matching socket from 3176 * the bucket. The remaining matched sockets from the bucket are batched 3177 * before releasing the bucket lock. This allows BPF programs that are 3178 * called in seq_show to acquire the bucket lock if needed. 3179 */ 3180 iter->cur_sk = 0; 3181 iter->end_sk = 0; 3182 iter->st_bucket_done = false; 3183 batch_sks = 0; 3184 3185 for (; state->bucket <= udptable->mask; state->bucket++) { 3186 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket]; 3187 3188 if (hlist_empty(&hslot2->head)) 3189 continue; 3190 3191 iter->offset = 0; 3192 spin_lock_bh(&hslot2->lock); 3193 udp_portaddr_for_each_entry(sk, &hslot2->head) { 3194 if (seq_sk_match(seq, sk)) { 3195 /* Resume from the last iterated socket at the 3196 * offset in the bucket before iterator was stopped. 3197 */ 3198 if (state->bucket == resume_bucket && 3199 iter->offset < resume_offset) { 3200 ++iter->offset; 3201 continue; 3202 } 3203 if (iter->end_sk < iter->max_sk) { 3204 sock_hold(sk); 3205 iter->batch[iter->end_sk++] = sk; 3206 } 3207 batch_sks++; 3208 } 3209 } 3210 spin_unlock_bh(&hslot2->lock); 3211 3212 if (iter->end_sk) 3213 break; 3214 } 3215 3216 /* All done: no batch made. */ 3217 if (!iter->end_sk) 3218 return NULL; 3219 3220 if (iter->end_sk == batch_sks) { 3221 /* Batching is done for the current bucket; return the first 3222 * socket to be iterated from the batch. 3223 */ 3224 iter->st_bucket_done = true; 3225 goto done; 3226 } 3227 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) { 3228 resized = true; 3229 /* After allocating a larger batch, retry one more time to grab 3230 * the whole bucket. 3231 */ 3232 goto again; 3233 } 3234 done: 3235 return iter->batch[0]; 3236 } 3237 3238 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3239 { 3240 struct bpf_udp_iter_state *iter = seq->private; 3241 struct sock *sk; 3242 3243 /* Whenever seq_next() is called, the iter->cur_sk is 3244 * done with seq_show(), so unref the iter->cur_sk. 3245 */ 3246 if (iter->cur_sk < iter->end_sk) { 3247 sock_put(iter->batch[iter->cur_sk++]); 3248 ++iter->offset; 3249 } 3250 3251 /* After updating iter->cur_sk, check if there are more sockets 3252 * available in the current bucket batch. 3253 */ 3254 if (iter->cur_sk < iter->end_sk) 3255 sk = iter->batch[iter->cur_sk]; 3256 else 3257 /* Prepare a new batch. */ 3258 sk = bpf_iter_udp_batch(seq); 3259 3260 ++*pos; 3261 return sk; 3262 } 3263 3264 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos) 3265 { 3266 /* bpf iter does not support lseek, so it always 3267 * continue from where it was stop()-ped. 3268 */ 3269 if (*pos) 3270 return bpf_iter_udp_batch(seq); 3271 3272 return SEQ_START_TOKEN; 3273 } 3274 3275 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3276 struct udp_sock *udp_sk, uid_t uid, int bucket) 3277 { 3278 struct bpf_iter__udp ctx; 3279 3280 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3281 ctx.meta = meta; 3282 ctx.udp_sk = udp_sk; 3283 ctx.uid = uid; 3284 ctx.bucket = bucket; 3285 return bpf_iter_run_prog(prog, &ctx); 3286 } 3287 3288 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3289 { 3290 struct udp_iter_state *state = seq->private; 3291 struct bpf_iter_meta meta; 3292 struct bpf_prog *prog; 3293 struct sock *sk = v; 3294 uid_t uid; 3295 int ret; 3296 3297 if (v == SEQ_START_TOKEN) 3298 return 0; 3299 3300 lock_sock(sk); 3301 3302 if (unlikely(sk_unhashed(sk))) { 3303 ret = SEQ_SKIP; 3304 goto unlock; 3305 } 3306 3307 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3308 meta.seq = seq; 3309 prog = bpf_iter_get_info(&meta, false); 3310 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3311 3312 unlock: 3313 release_sock(sk); 3314 return ret; 3315 } 3316 3317 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter) 3318 { 3319 while (iter->cur_sk < iter->end_sk) 3320 sock_put(iter->batch[iter->cur_sk++]); 3321 } 3322 3323 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3324 { 3325 struct bpf_udp_iter_state *iter = seq->private; 3326 struct bpf_iter_meta meta; 3327 struct bpf_prog *prog; 3328 3329 if (!v) { 3330 meta.seq = seq; 3331 prog = bpf_iter_get_info(&meta, true); 3332 if (prog) 3333 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3334 } 3335 3336 if (iter->cur_sk < iter->end_sk) { 3337 bpf_iter_udp_put_batch(iter); 3338 iter->st_bucket_done = false; 3339 } 3340 } 3341 3342 static const struct seq_operations bpf_iter_udp_seq_ops = { 3343 .start = bpf_iter_udp_seq_start, 3344 .next = bpf_iter_udp_seq_next, 3345 .stop = bpf_iter_udp_seq_stop, 3346 .show = bpf_iter_udp_seq_show, 3347 }; 3348 #endif 3349 3350 static unsigned short seq_file_family(const struct seq_file *seq) 3351 { 3352 const struct udp_seq_afinfo *afinfo; 3353 3354 #ifdef CONFIG_BPF_SYSCALL 3355 /* BPF iterator: bpf programs to filter sockets. */ 3356 if (seq->op == &bpf_iter_udp_seq_ops) 3357 return AF_UNSPEC; 3358 #endif 3359 3360 /* Proc fs iterator */ 3361 afinfo = pde_data(file_inode(seq->file)); 3362 return afinfo->family; 3363 } 3364 3365 const struct seq_operations udp_seq_ops = { 3366 .start = udp_seq_start, 3367 .next = udp_seq_next, 3368 .stop = udp_seq_stop, 3369 .show = udp4_seq_show, 3370 }; 3371 EXPORT_SYMBOL(udp_seq_ops); 3372 3373 static struct udp_seq_afinfo udp4_seq_afinfo = { 3374 .family = AF_INET, 3375 .udp_table = NULL, 3376 }; 3377 3378 static int __net_init udp4_proc_init_net(struct net *net) 3379 { 3380 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3381 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3382 return -ENOMEM; 3383 return 0; 3384 } 3385 3386 static void __net_exit udp4_proc_exit_net(struct net *net) 3387 { 3388 remove_proc_entry("udp", net->proc_net); 3389 } 3390 3391 static struct pernet_operations udp4_net_ops = { 3392 .init = udp4_proc_init_net, 3393 .exit = udp4_proc_exit_net, 3394 }; 3395 3396 int __init udp4_proc_init(void) 3397 { 3398 return register_pernet_subsys(&udp4_net_ops); 3399 } 3400 3401 void udp4_proc_exit(void) 3402 { 3403 unregister_pernet_subsys(&udp4_net_ops); 3404 } 3405 #endif /* CONFIG_PROC_FS */ 3406 3407 static __initdata unsigned long uhash_entries; 3408 static int __init set_uhash_entries(char *str) 3409 { 3410 ssize_t ret; 3411 3412 if (!str) 3413 return 0; 3414 3415 ret = kstrtoul(str, 0, &uhash_entries); 3416 if (ret) 3417 return 0; 3418 3419 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3420 uhash_entries = UDP_HTABLE_SIZE_MIN; 3421 return 1; 3422 } 3423 __setup("uhash_entries=", set_uhash_entries); 3424 3425 void __init udp_table_init(struct udp_table *table, const char *name) 3426 { 3427 unsigned int i; 3428 3429 table->hash = alloc_large_system_hash(name, 3430 2 * sizeof(struct udp_hslot), 3431 uhash_entries, 3432 21, /* one slot per 2 MB */ 3433 0, 3434 &table->log, 3435 &table->mask, 3436 UDP_HTABLE_SIZE_MIN, 3437 UDP_HTABLE_SIZE_MAX); 3438 3439 table->hash2 = table->hash + (table->mask + 1); 3440 for (i = 0; i <= table->mask; i++) { 3441 INIT_HLIST_HEAD(&table->hash[i].head); 3442 table->hash[i].count = 0; 3443 spin_lock_init(&table->hash[i].lock); 3444 } 3445 for (i = 0; i <= table->mask; i++) { 3446 INIT_HLIST_HEAD(&table->hash2[i].head); 3447 table->hash2[i].count = 0; 3448 spin_lock_init(&table->hash2[i].lock); 3449 } 3450 } 3451 3452 u32 udp_flow_hashrnd(void) 3453 { 3454 static u32 hashrnd __read_mostly; 3455 3456 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3457 3458 return hashrnd; 3459 } 3460 EXPORT_SYMBOL(udp_flow_hashrnd); 3461 3462 static void __net_init udp_sysctl_init(struct net *net) 3463 { 3464 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3465 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 3466 3467 #ifdef CONFIG_NET_L3_MASTER_DEV 3468 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3469 #endif 3470 } 3471 3472 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries) 3473 { 3474 struct udp_table *udptable; 3475 int i; 3476 3477 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL); 3478 if (!udptable) 3479 goto out; 3480 3481 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot), 3482 GFP_KERNEL_ACCOUNT); 3483 if (!udptable->hash) 3484 goto free_table; 3485 3486 udptable->hash2 = udptable->hash + hash_entries; 3487 udptable->mask = hash_entries - 1; 3488 udptable->log = ilog2(hash_entries); 3489 3490 for (i = 0; i < hash_entries; i++) { 3491 INIT_HLIST_HEAD(&udptable->hash[i].head); 3492 udptable->hash[i].count = 0; 3493 spin_lock_init(&udptable->hash[i].lock); 3494 3495 INIT_HLIST_HEAD(&udptable->hash2[i].head); 3496 udptable->hash2[i].count = 0; 3497 spin_lock_init(&udptable->hash2[i].lock); 3498 } 3499 3500 return udptable; 3501 3502 free_table: 3503 kfree(udptable); 3504 out: 3505 return NULL; 3506 } 3507 3508 static void __net_exit udp_pernet_table_free(struct net *net) 3509 { 3510 struct udp_table *udptable = net->ipv4.udp_table; 3511 3512 if (udptable == &udp_table) 3513 return; 3514 3515 kvfree(udptable->hash); 3516 kfree(udptable); 3517 } 3518 3519 static void __net_init udp_set_table(struct net *net) 3520 { 3521 struct udp_table *udptable; 3522 unsigned int hash_entries; 3523 struct net *old_net; 3524 3525 if (net_eq(net, &init_net)) 3526 goto fallback; 3527 3528 old_net = current->nsproxy->net_ns; 3529 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries); 3530 if (!hash_entries) 3531 goto fallback; 3532 3533 /* Set min to keep the bitmap on stack in udp_lib_get_port() */ 3534 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET) 3535 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET; 3536 else 3537 hash_entries = roundup_pow_of_two(hash_entries); 3538 3539 udptable = udp_pernet_table_alloc(hash_entries); 3540 if (udptable) { 3541 net->ipv4.udp_table = udptable; 3542 } else { 3543 pr_warn("Failed to allocate UDP hash table (entries: %u) " 3544 "for a netns, fallback to the global one\n", 3545 hash_entries); 3546 fallback: 3547 net->ipv4.udp_table = &udp_table; 3548 } 3549 } 3550 3551 static int __net_init udp_pernet_init(struct net *net) 3552 { 3553 udp_sysctl_init(net); 3554 udp_set_table(net); 3555 3556 return 0; 3557 } 3558 3559 static void __net_exit udp_pernet_exit(struct net *net) 3560 { 3561 udp_pernet_table_free(net); 3562 } 3563 3564 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3565 .init = udp_pernet_init, 3566 .exit = udp_pernet_exit, 3567 }; 3568 3569 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3570 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3571 struct udp_sock *udp_sk, uid_t uid, int bucket) 3572 3573 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3574 unsigned int new_batch_sz) 3575 { 3576 struct sock **new_batch; 3577 3578 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch), 3579 GFP_USER | __GFP_NOWARN); 3580 if (!new_batch) 3581 return -ENOMEM; 3582 3583 bpf_iter_udp_put_batch(iter); 3584 kvfree(iter->batch); 3585 iter->batch = new_batch; 3586 iter->max_sk = new_batch_sz; 3587 3588 return 0; 3589 } 3590 3591 #define INIT_BATCH_SZ 16 3592 3593 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3594 { 3595 struct bpf_udp_iter_state *iter = priv_data; 3596 int ret; 3597 3598 ret = bpf_iter_init_seq_net(priv_data, aux); 3599 if (ret) 3600 return ret; 3601 3602 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ); 3603 if (ret) 3604 bpf_iter_fini_seq_net(priv_data); 3605 3606 return ret; 3607 } 3608 3609 static void bpf_iter_fini_udp(void *priv_data) 3610 { 3611 struct bpf_udp_iter_state *iter = priv_data; 3612 3613 bpf_iter_fini_seq_net(priv_data); 3614 kvfree(iter->batch); 3615 } 3616 3617 static const struct bpf_iter_seq_info udp_seq_info = { 3618 .seq_ops = &bpf_iter_udp_seq_ops, 3619 .init_seq_private = bpf_iter_init_udp, 3620 .fini_seq_private = bpf_iter_fini_udp, 3621 .seq_priv_size = sizeof(struct bpf_udp_iter_state), 3622 }; 3623 3624 static struct bpf_iter_reg udp_reg_info = { 3625 .target = "udp", 3626 .ctx_arg_info_size = 1, 3627 .ctx_arg_info = { 3628 { offsetof(struct bpf_iter__udp, udp_sk), 3629 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED }, 3630 }, 3631 .seq_info = &udp_seq_info, 3632 }; 3633 3634 static void __init bpf_iter_register(void) 3635 { 3636 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3637 if (bpf_iter_reg_target(&udp_reg_info)) 3638 pr_warn("Warning: could not register bpf iterator udp\n"); 3639 } 3640 #endif 3641 3642 void __init udp_init(void) 3643 { 3644 unsigned long limit; 3645 unsigned int i; 3646 3647 udp_table_init(&udp_table, "UDP"); 3648 limit = nr_free_buffer_pages() / 8; 3649 limit = max(limit, 128UL); 3650 sysctl_udp_mem[0] = limit / 4 * 3; 3651 sysctl_udp_mem[1] = limit; 3652 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3653 3654 /* 16 spinlocks per cpu */ 3655 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3656 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3657 GFP_KERNEL); 3658 if (!udp_busylocks) 3659 panic("UDP: failed to alloc udp_busylocks\n"); 3660 for (i = 0; i < (1U << udp_busylocks_log); i++) 3661 spin_lock_init(udp_busylocks + i); 3662 3663 if (register_pernet_subsys(&udp_sysctl_ops)) 3664 panic("UDP: failed to init sysctl parameters.\n"); 3665 3666 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3667 bpf_iter_register(); 3668 #endif 3669 } 3670