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 sock_set_flag(sk, SOCK_RCU_FREE); 330 331 sk_add_node_rcu(sk, &hslot->head); 332 hslot->count++; 333 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 334 335 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 336 spin_lock(&hslot2->lock); 337 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 338 sk->sk_family == AF_INET6) 339 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 340 &hslot2->head); 341 else 342 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 343 &hslot2->head); 344 hslot2->count++; 345 spin_unlock(&hslot2->lock); 346 } 347 348 error = 0; 349 fail_unlock: 350 spin_unlock_bh(&hslot->lock); 351 fail: 352 return error; 353 } 354 EXPORT_SYMBOL(udp_lib_get_port); 355 356 int udp_v4_get_port(struct sock *sk, unsigned short snum) 357 { 358 unsigned int hash2_nulladdr = 359 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 360 unsigned int hash2_partial = 361 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 362 363 /* precompute partial secondary hash */ 364 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 365 return udp_lib_get_port(sk, snum, hash2_nulladdr); 366 } 367 368 static int compute_score(struct sock *sk, struct net *net, 369 __be32 saddr, __be16 sport, 370 __be32 daddr, unsigned short hnum, 371 int dif, int sdif) 372 { 373 int score; 374 struct inet_sock *inet; 375 bool dev_match; 376 377 if (!net_eq(sock_net(sk), net) || 378 udp_sk(sk)->udp_port_hash != hnum || 379 ipv6_only_sock(sk)) 380 return -1; 381 382 if (sk->sk_rcv_saddr != daddr) 383 return -1; 384 385 score = (sk->sk_family == PF_INET) ? 2 : 1; 386 387 inet = inet_sk(sk); 388 if (inet->inet_daddr) { 389 if (inet->inet_daddr != saddr) 390 return -1; 391 score += 4; 392 } 393 394 if (inet->inet_dport) { 395 if (inet->inet_dport != sport) 396 return -1; 397 score += 4; 398 } 399 400 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 401 dif, sdif); 402 if (!dev_match) 403 return -1; 404 if (sk->sk_bound_dev_if) 405 score += 4; 406 407 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 408 score++; 409 return score; 410 } 411 412 INDIRECT_CALLABLE_SCOPE 413 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport, 414 const __be32 faddr, const __be16 fport) 415 { 416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 417 418 return __inet_ehashfn(laddr, lport, faddr, fport, 419 udp_ehash_secret + net_hash_mix(net)); 420 } 421 422 /* called with rcu_read_lock() */ 423 static struct sock *udp4_lib_lookup2(struct net *net, 424 __be32 saddr, __be16 sport, 425 __be32 daddr, unsigned int hnum, 426 int dif, int sdif, 427 struct udp_hslot *hslot2, 428 struct sk_buff *skb) 429 { 430 struct sock *sk, *result; 431 int score, badness; 432 bool need_rescore; 433 434 result = NULL; 435 badness = 0; 436 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 437 need_rescore = false; 438 rescore: 439 score = compute_score(need_rescore ? result : sk, net, saddr, 440 sport, daddr, hnum, dif, sdif); 441 if (score > badness) { 442 badness = score; 443 444 if (need_rescore) 445 continue; 446 447 if (sk->sk_state == TCP_ESTABLISHED) { 448 result = sk; 449 continue; 450 } 451 452 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 453 saddr, sport, daddr, hnum, udp_ehashfn); 454 if (!result) { 455 result = sk; 456 continue; 457 } 458 459 /* Fall back to scoring if group has connections */ 460 if (!reuseport_has_conns(sk)) 461 return result; 462 463 /* Reuseport logic returned an error, keep original score. */ 464 if (IS_ERR(result)) 465 continue; 466 467 /* compute_score is too long of a function to be 468 * inlined, and calling it again here yields 469 * measureable overhead for some 470 * workloads. Work around it by jumping 471 * backwards to rescore 'result'. 472 */ 473 need_rescore = true; 474 goto rescore; 475 } 476 } 477 return result; 478 } 479 480 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 481 * harder than this. -DaveM 482 */ 483 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 484 __be16 sport, __be32 daddr, __be16 dport, int dif, 485 int sdif, struct udp_table *udptable, struct sk_buff *skb) 486 { 487 unsigned short hnum = ntohs(dport); 488 unsigned int hash2, slot2; 489 struct udp_hslot *hslot2; 490 struct sock *result, *sk; 491 492 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 493 slot2 = hash2 & udptable->mask; 494 hslot2 = &udptable->hash2[slot2]; 495 496 /* Lookup connected or non-wildcard socket */ 497 result = udp4_lib_lookup2(net, saddr, sport, 498 daddr, hnum, dif, sdif, 499 hslot2, skb); 500 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 501 goto done; 502 503 /* Lookup redirect from BPF */ 504 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 505 udptable == net->ipv4.udp_table) { 506 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 507 saddr, sport, daddr, hnum, dif, 508 udp_ehashfn); 509 if (sk) { 510 result = sk; 511 goto done; 512 } 513 } 514 515 /* Got non-wildcard socket or error on first lookup */ 516 if (result) 517 goto done; 518 519 /* Lookup wildcard sockets */ 520 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 521 slot2 = hash2 & udptable->mask; 522 hslot2 = &udptable->hash2[slot2]; 523 524 result = udp4_lib_lookup2(net, saddr, sport, 525 htonl(INADDR_ANY), hnum, dif, sdif, 526 hslot2, skb); 527 done: 528 if (IS_ERR(result)) 529 return NULL; 530 return result; 531 } 532 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 533 534 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 535 __be16 sport, __be16 dport, 536 struct udp_table *udptable) 537 { 538 const struct iphdr *iph = ip_hdr(skb); 539 540 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 541 iph->daddr, dport, inet_iif(skb), 542 inet_sdif(skb), udptable, skb); 543 } 544 545 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 546 __be16 sport, __be16 dport) 547 { 548 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation]; 549 const struct iphdr *iph = (struct iphdr *)(skb->data + offset); 550 struct net *net = dev_net(skb->dev); 551 int iif, sdif; 552 553 inet_get_iif_sdif(skb, &iif, &sdif); 554 555 return __udp4_lib_lookup(net, iph->saddr, sport, 556 iph->daddr, dport, iif, 557 sdif, net->ipv4.udp_table, NULL); 558 } 559 560 /* Must be called under rcu_read_lock(). 561 * Does increment socket refcount. 562 */ 563 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 564 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 565 __be32 daddr, __be16 dport, int dif) 566 { 567 struct sock *sk; 568 569 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 570 dif, 0, net->ipv4.udp_table, NULL); 571 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 572 sk = NULL; 573 return sk; 574 } 575 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 576 #endif 577 578 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk, 579 __be16 loc_port, __be32 loc_addr, 580 __be16 rmt_port, __be32 rmt_addr, 581 int dif, int sdif, unsigned short hnum) 582 { 583 const struct inet_sock *inet = inet_sk(sk); 584 585 if (!net_eq(sock_net(sk), net) || 586 udp_sk(sk)->udp_port_hash != hnum || 587 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 588 (inet->inet_dport != rmt_port && inet->inet_dport) || 589 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 590 ipv6_only_sock(sk) || 591 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 592 return false; 593 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 594 return false; 595 return true; 596 } 597 598 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 599 EXPORT_SYMBOL(udp_encap_needed_key); 600 601 #if IS_ENABLED(CONFIG_IPV6) 602 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 603 EXPORT_SYMBOL(udpv6_encap_needed_key); 604 #endif 605 606 void udp_encap_enable(void) 607 { 608 static_branch_inc(&udp_encap_needed_key); 609 } 610 EXPORT_SYMBOL(udp_encap_enable); 611 612 void udp_encap_disable(void) 613 { 614 static_branch_dec(&udp_encap_needed_key); 615 } 616 EXPORT_SYMBOL(udp_encap_disable); 617 618 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 619 * through error handlers in encapsulations looking for a match. 620 */ 621 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 622 { 623 int i; 624 625 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 626 int (*handler)(struct sk_buff *skb, u32 info); 627 const struct ip_tunnel_encap_ops *encap; 628 629 encap = rcu_dereference(iptun_encaps[i]); 630 if (!encap) 631 continue; 632 handler = encap->err_handler; 633 if (handler && !handler(skb, info)) 634 return 0; 635 } 636 637 return -ENOENT; 638 } 639 640 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 641 * reversing source and destination port: this will match tunnels that force the 642 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 643 * lwtunnels might actually break this assumption by being configured with 644 * different destination ports on endpoints, in this case we won't be able to 645 * trace ICMP messages back to them. 646 * 647 * If this doesn't match any socket, probe tunnels with arbitrary destination 648 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 649 * we've sent packets to won't necessarily match the local destination port. 650 * 651 * Then ask the tunnel implementation to match the error against a valid 652 * association. 653 * 654 * Return an error if we can't find a match, the socket if we need further 655 * processing, zero otherwise. 656 */ 657 static struct sock *__udp4_lib_err_encap(struct net *net, 658 const struct iphdr *iph, 659 struct udphdr *uh, 660 struct udp_table *udptable, 661 struct sock *sk, 662 struct sk_buff *skb, u32 info) 663 { 664 int (*lookup)(struct sock *sk, struct sk_buff *skb); 665 int network_offset, transport_offset; 666 struct udp_sock *up; 667 668 network_offset = skb_network_offset(skb); 669 transport_offset = skb_transport_offset(skb); 670 671 /* Network header needs to point to the outer IPv4 header inside ICMP */ 672 skb_reset_network_header(skb); 673 674 /* Transport header needs to point to the UDP header */ 675 skb_set_transport_header(skb, iph->ihl << 2); 676 677 if (sk) { 678 up = udp_sk(sk); 679 680 lookup = READ_ONCE(up->encap_err_lookup); 681 if (lookup && lookup(sk, skb)) 682 sk = NULL; 683 684 goto out; 685 } 686 687 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 688 iph->saddr, uh->dest, skb->dev->ifindex, 0, 689 udptable, NULL); 690 if (sk) { 691 up = udp_sk(sk); 692 693 lookup = READ_ONCE(up->encap_err_lookup); 694 if (!lookup || lookup(sk, skb)) 695 sk = NULL; 696 } 697 698 out: 699 if (!sk) 700 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 701 702 skb_set_transport_header(skb, transport_offset); 703 skb_set_network_header(skb, network_offset); 704 705 return sk; 706 } 707 708 /* 709 * This routine is called by the ICMP module when it gets some 710 * sort of error condition. If err < 0 then the socket should 711 * be closed and the error returned to the user. If err > 0 712 * it's just the icmp type << 8 | icmp code. 713 * Header points to the ip header of the error packet. We move 714 * on past this. Then (as it used to claim before adjustment) 715 * header points to the first 8 bytes of the udp header. We need 716 * to find the appropriate port. 717 */ 718 719 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 720 { 721 struct inet_sock *inet; 722 const struct iphdr *iph = (const struct iphdr *)skb->data; 723 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 724 const int type = icmp_hdr(skb)->type; 725 const int code = icmp_hdr(skb)->code; 726 bool tunnel = false; 727 struct sock *sk; 728 int harderr; 729 int err; 730 struct net *net = dev_net(skb->dev); 731 732 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 733 iph->saddr, uh->source, skb->dev->ifindex, 734 inet_sdif(skb), udptable, NULL); 735 736 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) { 737 /* No socket for error: try tunnels before discarding */ 738 if (static_branch_unlikely(&udp_encap_needed_key)) { 739 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 740 info); 741 if (!sk) 742 return 0; 743 } else 744 sk = ERR_PTR(-ENOENT); 745 746 if (IS_ERR(sk)) { 747 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 748 return PTR_ERR(sk); 749 } 750 751 tunnel = true; 752 } 753 754 err = 0; 755 harderr = 0; 756 inet = inet_sk(sk); 757 758 switch (type) { 759 default: 760 case ICMP_TIME_EXCEEDED: 761 err = EHOSTUNREACH; 762 break; 763 case ICMP_SOURCE_QUENCH: 764 goto out; 765 case ICMP_PARAMETERPROB: 766 err = EPROTO; 767 harderr = 1; 768 break; 769 case ICMP_DEST_UNREACH: 770 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 771 ipv4_sk_update_pmtu(skb, sk, info); 772 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) { 773 err = EMSGSIZE; 774 harderr = 1; 775 break; 776 } 777 goto out; 778 } 779 err = EHOSTUNREACH; 780 if (code <= NR_ICMP_UNREACH) { 781 harderr = icmp_err_convert[code].fatal; 782 err = icmp_err_convert[code].errno; 783 } 784 break; 785 case ICMP_REDIRECT: 786 ipv4_sk_redirect(skb, sk); 787 goto out; 788 } 789 790 /* 791 * RFC1122: OK. Passes ICMP errors back to application, as per 792 * 4.1.3.3. 793 */ 794 if (tunnel) { 795 /* ...not for tunnels though: we don't have a sending socket */ 796 if (udp_sk(sk)->encap_err_rcv) 797 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info, 798 (u8 *)(uh+1)); 799 goto out; 800 } 801 if (!inet_test_bit(RECVERR, sk)) { 802 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 803 goto out; 804 } else 805 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 806 807 sk->sk_err = err; 808 sk_error_report(sk); 809 out: 810 return 0; 811 } 812 813 int udp_err(struct sk_buff *skb, u32 info) 814 { 815 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table); 816 } 817 818 /* 819 * Throw away all pending data and cancel the corking. Socket is locked. 820 */ 821 void udp_flush_pending_frames(struct sock *sk) 822 { 823 struct udp_sock *up = udp_sk(sk); 824 825 if (up->pending) { 826 up->len = 0; 827 WRITE_ONCE(up->pending, 0); 828 ip_flush_pending_frames(sk); 829 } 830 } 831 EXPORT_SYMBOL(udp_flush_pending_frames); 832 833 /** 834 * udp4_hwcsum - handle outgoing HW checksumming 835 * @skb: sk_buff containing the filled-in UDP header 836 * (checksum field must be zeroed out) 837 * @src: source IP address 838 * @dst: destination IP address 839 */ 840 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 841 { 842 struct udphdr *uh = udp_hdr(skb); 843 int offset = skb_transport_offset(skb); 844 int len = skb->len - offset; 845 int hlen = len; 846 __wsum csum = 0; 847 848 if (!skb_has_frag_list(skb)) { 849 /* 850 * Only one fragment on the socket. 851 */ 852 skb->csum_start = skb_transport_header(skb) - skb->head; 853 skb->csum_offset = offsetof(struct udphdr, check); 854 uh->check = ~csum_tcpudp_magic(src, dst, len, 855 IPPROTO_UDP, 0); 856 } else { 857 struct sk_buff *frags; 858 859 /* 860 * HW-checksum won't work as there are two or more 861 * fragments on the socket so that all csums of sk_buffs 862 * should be together 863 */ 864 skb_walk_frags(skb, frags) { 865 csum = csum_add(csum, frags->csum); 866 hlen -= frags->len; 867 } 868 869 csum = skb_checksum(skb, offset, hlen, csum); 870 skb->ip_summed = CHECKSUM_NONE; 871 872 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 873 if (uh->check == 0) 874 uh->check = CSUM_MANGLED_0; 875 } 876 } 877 EXPORT_SYMBOL_GPL(udp4_hwcsum); 878 879 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 880 * for the simple case like when setting the checksum for a UDP tunnel. 881 */ 882 void udp_set_csum(bool nocheck, struct sk_buff *skb, 883 __be32 saddr, __be32 daddr, int len) 884 { 885 struct udphdr *uh = udp_hdr(skb); 886 887 if (nocheck) { 888 uh->check = 0; 889 } else if (skb_is_gso(skb)) { 890 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 891 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 892 uh->check = 0; 893 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 894 if (uh->check == 0) 895 uh->check = CSUM_MANGLED_0; 896 } else { 897 skb->ip_summed = CHECKSUM_PARTIAL; 898 skb->csum_start = skb_transport_header(skb) - skb->head; 899 skb->csum_offset = offsetof(struct udphdr, check); 900 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 901 } 902 } 903 EXPORT_SYMBOL(udp_set_csum); 904 905 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 906 struct inet_cork *cork) 907 { 908 struct sock *sk = skb->sk; 909 struct inet_sock *inet = inet_sk(sk); 910 struct udphdr *uh; 911 int err; 912 int is_udplite = IS_UDPLITE(sk); 913 int offset = skb_transport_offset(skb); 914 int len = skb->len - offset; 915 int datalen = len - sizeof(*uh); 916 __wsum csum = 0; 917 918 /* 919 * Create a UDP header 920 */ 921 uh = udp_hdr(skb); 922 uh->source = inet->inet_sport; 923 uh->dest = fl4->fl4_dport; 924 uh->len = htons(len); 925 uh->check = 0; 926 927 if (cork->gso_size) { 928 const int hlen = skb_network_header_len(skb) + 929 sizeof(struct udphdr); 930 931 if (hlen + cork->gso_size > cork->fragsize) { 932 kfree_skb(skb); 933 return -EINVAL; 934 } 935 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 936 kfree_skb(skb); 937 return -EINVAL; 938 } 939 if (sk->sk_no_check_tx) { 940 kfree_skb(skb); 941 return -EINVAL; 942 } 943 if (is_udplite || dst_xfrm(skb_dst(skb))) { 944 kfree_skb(skb); 945 return -EIO; 946 } 947 948 if (datalen > cork->gso_size) { 949 skb_shinfo(skb)->gso_size = cork->gso_size; 950 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 951 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 952 cork->gso_size); 953 } 954 goto csum_partial; 955 } 956 957 if (is_udplite) /* UDP-Lite */ 958 csum = udplite_csum(skb); 959 960 else if (sk->sk_no_check_tx) { /* UDP csum off */ 961 962 skb->ip_summed = CHECKSUM_NONE; 963 goto send; 964 965 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 966 csum_partial: 967 968 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 969 goto send; 970 971 } else 972 csum = udp_csum(skb); 973 974 /* add protocol-dependent pseudo-header */ 975 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 976 sk->sk_protocol, csum); 977 if (uh->check == 0) 978 uh->check = CSUM_MANGLED_0; 979 980 send: 981 err = ip_send_skb(sock_net(sk), skb); 982 if (err) { 983 if (err == -ENOBUFS && 984 !inet_test_bit(RECVERR, sk)) { 985 UDP_INC_STATS(sock_net(sk), 986 UDP_MIB_SNDBUFERRORS, is_udplite); 987 err = 0; 988 } 989 } else 990 UDP_INC_STATS(sock_net(sk), 991 UDP_MIB_OUTDATAGRAMS, is_udplite); 992 return err; 993 } 994 995 /* 996 * Push out all pending data as one UDP datagram. Socket is locked. 997 */ 998 int udp_push_pending_frames(struct sock *sk) 999 { 1000 struct udp_sock *up = udp_sk(sk); 1001 struct inet_sock *inet = inet_sk(sk); 1002 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 1003 struct sk_buff *skb; 1004 int err = 0; 1005 1006 skb = ip_finish_skb(sk, fl4); 1007 if (!skb) 1008 goto out; 1009 1010 err = udp_send_skb(skb, fl4, &inet->cork.base); 1011 1012 out: 1013 up->len = 0; 1014 WRITE_ONCE(up->pending, 0); 1015 return err; 1016 } 1017 EXPORT_SYMBOL(udp_push_pending_frames); 1018 1019 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1020 { 1021 switch (cmsg->cmsg_type) { 1022 case UDP_SEGMENT: 1023 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1024 return -EINVAL; 1025 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1026 return 0; 1027 default: 1028 return -EINVAL; 1029 } 1030 } 1031 1032 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1033 { 1034 struct cmsghdr *cmsg; 1035 bool need_ip = false; 1036 int err; 1037 1038 for_each_cmsghdr(cmsg, msg) { 1039 if (!CMSG_OK(msg, cmsg)) 1040 return -EINVAL; 1041 1042 if (cmsg->cmsg_level != SOL_UDP) { 1043 need_ip = true; 1044 continue; 1045 } 1046 1047 err = __udp_cmsg_send(cmsg, gso_size); 1048 if (err) 1049 return err; 1050 } 1051 1052 return need_ip; 1053 } 1054 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1055 1056 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1057 { 1058 struct inet_sock *inet = inet_sk(sk); 1059 struct udp_sock *up = udp_sk(sk); 1060 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1061 struct flowi4 fl4_stack; 1062 struct flowi4 *fl4; 1063 int ulen = len; 1064 struct ipcm_cookie ipc; 1065 struct rtable *rt = NULL; 1066 int free = 0; 1067 int connected = 0; 1068 __be32 daddr, faddr, saddr; 1069 u8 tos, scope; 1070 __be16 dport; 1071 int err, is_udplite = IS_UDPLITE(sk); 1072 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE; 1073 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1074 struct sk_buff *skb; 1075 struct ip_options_data opt_copy; 1076 int uc_index; 1077 1078 if (len > 0xFFFF) 1079 return -EMSGSIZE; 1080 1081 /* 1082 * Check the flags. 1083 */ 1084 1085 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1086 return -EOPNOTSUPP; 1087 1088 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1089 1090 fl4 = &inet->cork.fl.u.ip4; 1091 if (READ_ONCE(up->pending)) { 1092 /* 1093 * There are pending frames. 1094 * The socket lock must be held while it's corked. 1095 */ 1096 lock_sock(sk); 1097 if (likely(up->pending)) { 1098 if (unlikely(up->pending != AF_INET)) { 1099 release_sock(sk); 1100 return -EINVAL; 1101 } 1102 goto do_append_data; 1103 } 1104 release_sock(sk); 1105 } 1106 ulen += sizeof(struct udphdr); 1107 1108 /* 1109 * Get and verify the address. 1110 */ 1111 if (usin) { 1112 if (msg->msg_namelen < sizeof(*usin)) 1113 return -EINVAL; 1114 if (usin->sin_family != AF_INET) { 1115 if (usin->sin_family != AF_UNSPEC) 1116 return -EAFNOSUPPORT; 1117 } 1118 1119 daddr = usin->sin_addr.s_addr; 1120 dport = usin->sin_port; 1121 if (dport == 0) 1122 return -EINVAL; 1123 } else { 1124 if (sk->sk_state != TCP_ESTABLISHED) 1125 return -EDESTADDRREQ; 1126 daddr = inet->inet_daddr; 1127 dport = inet->inet_dport; 1128 /* Open fast path for connected socket. 1129 Route will not be used, if at least one option is set. 1130 */ 1131 connected = 1; 1132 } 1133 1134 ipcm_init_sk(&ipc, inet); 1135 ipc.gso_size = READ_ONCE(up->gso_size); 1136 1137 if (msg->msg_controllen) { 1138 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1139 if (err > 0) { 1140 err = ip_cmsg_send(sk, msg, &ipc, 1141 sk->sk_family == AF_INET6); 1142 connected = 0; 1143 } 1144 if (unlikely(err < 0)) { 1145 kfree(ipc.opt); 1146 return err; 1147 } 1148 if (ipc.opt) 1149 free = 1; 1150 } 1151 if (!ipc.opt) { 1152 struct ip_options_rcu *inet_opt; 1153 1154 rcu_read_lock(); 1155 inet_opt = rcu_dereference(inet->inet_opt); 1156 if (inet_opt) { 1157 memcpy(&opt_copy, inet_opt, 1158 sizeof(*inet_opt) + inet_opt->opt.optlen); 1159 ipc.opt = &opt_copy.opt; 1160 } 1161 rcu_read_unlock(); 1162 } 1163 1164 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1165 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1166 (struct sockaddr *)usin, 1167 &msg->msg_namelen, 1168 &ipc.addr); 1169 if (err) 1170 goto out_free; 1171 if (usin) { 1172 if (usin->sin_port == 0) { 1173 /* BPF program set invalid port. Reject it. */ 1174 err = -EINVAL; 1175 goto out_free; 1176 } 1177 daddr = usin->sin_addr.s_addr; 1178 dport = usin->sin_port; 1179 } 1180 } 1181 1182 saddr = ipc.addr; 1183 ipc.addr = faddr = daddr; 1184 1185 if (ipc.opt && ipc.opt->opt.srr) { 1186 if (!daddr) { 1187 err = -EINVAL; 1188 goto out_free; 1189 } 1190 faddr = ipc.opt->opt.faddr; 1191 connected = 0; 1192 } 1193 tos = get_rttos(&ipc, inet); 1194 scope = ip_sendmsg_scope(inet, &ipc, msg); 1195 if (scope == RT_SCOPE_LINK) 1196 connected = 0; 1197 1198 uc_index = READ_ONCE(inet->uc_index); 1199 if (ipv4_is_multicast(daddr)) { 1200 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1201 ipc.oif = READ_ONCE(inet->mc_index); 1202 if (!saddr) 1203 saddr = READ_ONCE(inet->mc_addr); 1204 connected = 0; 1205 } else if (!ipc.oif) { 1206 ipc.oif = uc_index; 1207 } else if (ipv4_is_lbcast(daddr) && uc_index) { 1208 /* oif is set, packet is to local broadcast and 1209 * uc_index is set. oif is most likely set 1210 * by sk_bound_dev_if. If uc_index != oif check if the 1211 * oif is an L3 master and uc_index is an L3 slave. 1212 * If so, we want to allow the send using the uc_index. 1213 */ 1214 if (ipc.oif != uc_index && 1215 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1216 uc_index)) { 1217 ipc.oif = uc_index; 1218 } 1219 } 1220 1221 if (connected) 1222 rt = dst_rtable(sk_dst_check(sk, 0)); 1223 1224 if (!rt) { 1225 struct net *net = sock_net(sk); 1226 __u8 flow_flags = inet_sk_flowi_flags(sk); 1227 1228 fl4 = &fl4_stack; 1229 1230 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope, 1231 sk->sk_protocol, flow_flags, faddr, saddr, 1232 dport, inet->inet_sport, sk->sk_uid); 1233 1234 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1235 rt = ip_route_output_flow(net, fl4, sk); 1236 if (IS_ERR(rt)) { 1237 err = PTR_ERR(rt); 1238 rt = NULL; 1239 if (err == -ENETUNREACH) 1240 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1241 goto out; 1242 } 1243 1244 err = -EACCES; 1245 if ((rt->rt_flags & RTCF_BROADCAST) && 1246 !sock_flag(sk, SOCK_BROADCAST)) 1247 goto out; 1248 if (connected) 1249 sk_dst_set(sk, dst_clone(&rt->dst)); 1250 } 1251 1252 if (msg->msg_flags&MSG_CONFIRM) 1253 goto do_confirm; 1254 back_from_confirm: 1255 1256 saddr = fl4->saddr; 1257 if (!ipc.addr) 1258 daddr = ipc.addr = fl4->daddr; 1259 1260 /* Lockless fast path for the non-corking case. */ 1261 if (!corkreq) { 1262 struct inet_cork cork; 1263 1264 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1265 sizeof(struct udphdr), &ipc, &rt, 1266 &cork, msg->msg_flags); 1267 err = PTR_ERR(skb); 1268 if (!IS_ERR_OR_NULL(skb)) 1269 err = udp_send_skb(skb, fl4, &cork); 1270 goto out; 1271 } 1272 1273 lock_sock(sk); 1274 if (unlikely(up->pending)) { 1275 /* The socket is already corked while preparing it. */ 1276 /* ... which is an evident application bug. --ANK */ 1277 release_sock(sk); 1278 1279 net_dbg_ratelimited("socket already corked\n"); 1280 err = -EINVAL; 1281 goto out; 1282 } 1283 /* 1284 * Now cork the socket to pend data. 1285 */ 1286 fl4 = &inet->cork.fl.u.ip4; 1287 fl4->daddr = daddr; 1288 fl4->saddr = saddr; 1289 fl4->fl4_dport = dport; 1290 fl4->fl4_sport = inet->inet_sport; 1291 WRITE_ONCE(up->pending, AF_INET); 1292 1293 do_append_data: 1294 up->len += ulen; 1295 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1296 sizeof(struct udphdr), &ipc, &rt, 1297 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1298 if (err) 1299 udp_flush_pending_frames(sk); 1300 else if (!corkreq) 1301 err = udp_push_pending_frames(sk); 1302 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1303 WRITE_ONCE(up->pending, 0); 1304 release_sock(sk); 1305 1306 out: 1307 ip_rt_put(rt); 1308 out_free: 1309 if (free) 1310 kfree(ipc.opt); 1311 if (!err) 1312 return len; 1313 /* 1314 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1315 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1316 * we don't have a good statistic (IpOutDiscards but it can be too many 1317 * things). We could add another new stat but at least for now that 1318 * seems like overkill. 1319 */ 1320 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1321 UDP_INC_STATS(sock_net(sk), 1322 UDP_MIB_SNDBUFERRORS, is_udplite); 1323 } 1324 return err; 1325 1326 do_confirm: 1327 if (msg->msg_flags & MSG_PROBE) 1328 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1329 if (!(msg->msg_flags&MSG_PROBE) || len) 1330 goto back_from_confirm; 1331 err = 0; 1332 goto out; 1333 } 1334 EXPORT_SYMBOL(udp_sendmsg); 1335 1336 void udp_splice_eof(struct socket *sock) 1337 { 1338 struct sock *sk = sock->sk; 1339 struct udp_sock *up = udp_sk(sk); 1340 1341 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk)) 1342 return; 1343 1344 lock_sock(sk); 1345 if (up->pending && !udp_test_bit(CORK, sk)) 1346 udp_push_pending_frames(sk); 1347 release_sock(sk); 1348 } 1349 EXPORT_SYMBOL_GPL(udp_splice_eof); 1350 1351 #define UDP_SKB_IS_STATELESS 0x80000000 1352 1353 /* all head states (dst, sk, nf conntrack) except skb extensions are 1354 * cleared by udp_rcv(). 1355 * 1356 * We need to preserve secpath, if present, to eventually process 1357 * IP_CMSG_PASSSEC at recvmsg() time. 1358 * 1359 * Other extensions can be cleared. 1360 */ 1361 static bool udp_try_make_stateless(struct sk_buff *skb) 1362 { 1363 if (!skb_has_extensions(skb)) 1364 return true; 1365 1366 if (!secpath_exists(skb)) { 1367 skb_ext_reset(skb); 1368 return true; 1369 } 1370 1371 return false; 1372 } 1373 1374 static void udp_set_dev_scratch(struct sk_buff *skb) 1375 { 1376 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1377 1378 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1379 scratch->_tsize_state = skb->truesize; 1380 #if BITS_PER_LONG == 64 1381 scratch->len = skb->len; 1382 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1383 scratch->is_linear = !skb_is_nonlinear(skb); 1384 #endif 1385 if (udp_try_make_stateless(skb)) 1386 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1387 } 1388 1389 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1390 { 1391 /* We come here after udp_lib_checksum_complete() returned 0. 1392 * This means that __skb_checksum_complete() might have 1393 * set skb->csum_valid to 1. 1394 * On 64bit platforms, we can set csum_unnecessary 1395 * to true, but only if the skb is not shared. 1396 */ 1397 #if BITS_PER_LONG == 64 1398 if (!skb_shared(skb)) 1399 udp_skb_scratch(skb)->csum_unnecessary = true; 1400 #endif 1401 } 1402 1403 static int udp_skb_truesize(struct sk_buff *skb) 1404 { 1405 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1406 } 1407 1408 static bool udp_skb_has_head_state(struct sk_buff *skb) 1409 { 1410 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1411 } 1412 1413 /* fully reclaim rmem/fwd memory allocated for skb */ 1414 static void udp_rmem_release(struct sock *sk, int size, int partial, 1415 bool rx_queue_lock_held) 1416 { 1417 struct udp_sock *up = udp_sk(sk); 1418 struct sk_buff_head *sk_queue; 1419 int amt; 1420 1421 if (likely(partial)) { 1422 up->forward_deficit += size; 1423 size = up->forward_deficit; 1424 if (size < READ_ONCE(up->forward_threshold) && 1425 !skb_queue_empty(&up->reader_queue)) 1426 return; 1427 } else { 1428 size += up->forward_deficit; 1429 } 1430 up->forward_deficit = 0; 1431 1432 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1433 * if the called don't held it already 1434 */ 1435 sk_queue = &sk->sk_receive_queue; 1436 if (!rx_queue_lock_held) 1437 spin_lock(&sk_queue->lock); 1438 1439 1440 sk_forward_alloc_add(sk, size); 1441 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1442 sk_forward_alloc_add(sk, -amt); 1443 1444 if (amt) 1445 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 1446 1447 atomic_sub(size, &sk->sk_rmem_alloc); 1448 1449 /* this can save us from acquiring the rx queue lock on next receive */ 1450 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1451 1452 if (!rx_queue_lock_held) 1453 spin_unlock(&sk_queue->lock); 1454 } 1455 1456 /* Note: called with reader_queue.lock held. 1457 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1458 * This avoids a cache line miss while receive_queue lock is held. 1459 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1460 */ 1461 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1462 { 1463 prefetch(&skb->data); 1464 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1465 } 1466 EXPORT_SYMBOL(udp_skb_destructor); 1467 1468 /* as above, but the caller held the rx queue lock, too */ 1469 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1470 { 1471 prefetch(&skb->data); 1472 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1473 } 1474 1475 /* Idea of busylocks is to let producers grab an extra spinlock 1476 * to relieve pressure on the receive_queue spinlock shared by consumer. 1477 * Under flood, this means that only one producer can be in line 1478 * trying to acquire the receive_queue spinlock. 1479 * These busylock can be allocated on a per cpu manner, instead of a 1480 * per socket one (that would consume a cache line per socket) 1481 */ 1482 static int udp_busylocks_log __read_mostly; 1483 static spinlock_t *udp_busylocks __read_mostly; 1484 1485 static spinlock_t *busylock_acquire(void *ptr) 1486 { 1487 spinlock_t *busy; 1488 1489 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1490 spin_lock(busy); 1491 return busy; 1492 } 1493 1494 static void busylock_release(spinlock_t *busy) 1495 { 1496 if (busy) 1497 spin_unlock(busy); 1498 } 1499 1500 static int udp_rmem_schedule(struct sock *sk, int size) 1501 { 1502 int delta; 1503 1504 delta = size - sk->sk_forward_alloc; 1505 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV)) 1506 return -ENOBUFS; 1507 1508 return 0; 1509 } 1510 1511 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1512 { 1513 struct sk_buff_head *list = &sk->sk_receive_queue; 1514 int rmem, err = -ENOMEM; 1515 spinlock_t *busy = NULL; 1516 bool becomes_readable; 1517 int size, rcvbuf; 1518 1519 /* Immediately drop when the receive queue is full. 1520 * Always allow at least one packet. 1521 */ 1522 rmem = atomic_read(&sk->sk_rmem_alloc); 1523 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1524 if (rmem > rcvbuf) 1525 goto drop; 1526 1527 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1528 * having linear skbs : 1529 * - Reduce memory overhead and thus increase receive queue capacity 1530 * - Less cache line misses at copyout() time 1531 * - Less work at consume_skb() (less alien page frag freeing) 1532 */ 1533 if (rmem > (rcvbuf >> 1)) { 1534 skb_condense(skb); 1535 1536 busy = busylock_acquire(sk); 1537 } 1538 size = skb->truesize; 1539 udp_set_dev_scratch(skb); 1540 1541 atomic_add(size, &sk->sk_rmem_alloc); 1542 1543 spin_lock(&list->lock); 1544 err = udp_rmem_schedule(sk, size); 1545 if (err) { 1546 spin_unlock(&list->lock); 1547 goto uncharge_drop; 1548 } 1549 1550 sk_forward_alloc_add(sk, -size); 1551 1552 /* no need to setup a destructor, we will explicitly release the 1553 * forward allocated memory on dequeue 1554 */ 1555 sock_skb_set_dropcount(sk, skb); 1556 1557 becomes_readable = skb_queue_empty(list); 1558 __skb_queue_tail(list, skb); 1559 spin_unlock(&list->lock); 1560 1561 if (!sock_flag(sk, SOCK_DEAD)) { 1562 if (becomes_readable || 1563 sk->sk_data_ready != sock_def_readable || 1564 READ_ONCE(sk->sk_peek_off) >= 0) 1565 INDIRECT_CALL_1(sk->sk_data_ready, 1566 sock_def_readable, sk); 1567 else 1568 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 1569 } 1570 busylock_release(busy); 1571 return 0; 1572 1573 uncharge_drop: 1574 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1575 1576 drop: 1577 atomic_inc(&sk->sk_drops); 1578 busylock_release(busy); 1579 return err; 1580 } 1581 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1582 1583 void udp_destruct_common(struct sock *sk) 1584 { 1585 /* reclaim completely the forward allocated memory */ 1586 struct udp_sock *up = udp_sk(sk); 1587 unsigned int total = 0; 1588 struct sk_buff *skb; 1589 1590 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1591 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1592 total += skb->truesize; 1593 kfree_skb(skb); 1594 } 1595 udp_rmem_release(sk, total, 0, true); 1596 } 1597 EXPORT_SYMBOL_GPL(udp_destruct_common); 1598 1599 static void udp_destruct_sock(struct sock *sk) 1600 { 1601 udp_destruct_common(sk); 1602 inet_sock_destruct(sk); 1603 } 1604 1605 int udp_init_sock(struct sock *sk) 1606 { 1607 udp_lib_init_sock(sk); 1608 sk->sk_destruct = udp_destruct_sock; 1609 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1610 return 0; 1611 } 1612 1613 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1614 { 1615 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset))) 1616 sk_peek_offset_bwd(sk, len); 1617 1618 if (!skb_unref(skb)) 1619 return; 1620 1621 /* In the more common cases we cleared the head states previously, 1622 * see __udp_queue_rcv_skb(). 1623 */ 1624 if (unlikely(udp_skb_has_head_state(skb))) 1625 skb_release_head_state(skb); 1626 __consume_stateless_skb(skb); 1627 } 1628 EXPORT_SYMBOL_GPL(skb_consume_udp); 1629 1630 static struct sk_buff *__first_packet_length(struct sock *sk, 1631 struct sk_buff_head *rcvq, 1632 int *total) 1633 { 1634 struct sk_buff *skb; 1635 1636 while ((skb = skb_peek(rcvq)) != NULL) { 1637 if (udp_lib_checksum_complete(skb)) { 1638 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1639 IS_UDPLITE(sk)); 1640 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1641 IS_UDPLITE(sk)); 1642 atomic_inc(&sk->sk_drops); 1643 __skb_unlink(skb, rcvq); 1644 *total += skb->truesize; 1645 kfree_skb(skb); 1646 } else { 1647 udp_skb_csum_unnecessary_set(skb); 1648 break; 1649 } 1650 } 1651 return skb; 1652 } 1653 1654 /** 1655 * first_packet_length - return length of first packet in receive queue 1656 * @sk: socket 1657 * 1658 * Drops all bad checksum frames, until a valid one is found. 1659 * Returns the length of found skb, or -1 if none is found. 1660 */ 1661 static int first_packet_length(struct sock *sk) 1662 { 1663 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1664 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1665 struct sk_buff *skb; 1666 int total = 0; 1667 int res; 1668 1669 spin_lock_bh(&rcvq->lock); 1670 skb = __first_packet_length(sk, rcvq, &total); 1671 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1672 spin_lock(&sk_queue->lock); 1673 skb_queue_splice_tail_init(sk_queue, rcvq); 1674 spin_unlock(&sk_queue->lock); 1675 1676 skb = __first_packet_length(sk, rcvq, &total); 1677 } 1678 res = skb ? skb->len : -1; 1679 if (total) 1680 udp_rmem_release(sk, total, 1, false); 1681 spin_unlock_bh(&rcvq->lock); 1682 return res; 1683 } 1684 1685 /* 1686 * IOCTL requests applicable to the UDP protocol 1687 */ 1688 1689 int udp_ioctl(struct sock *sk, int cmd, int *karg) 1690 { 1691 switch (cmd) { 1692 case SIOCOUTQ: 1693 { 1694 *karg = sk_wmem_alloc_get(sk); 1695 return 0; 1696 } 1697 1698 case SIOCINQ: 1699 { 1700 *karg = max_t(int, 0, first_packet_length(sk)); 1701 return 0; 1702 } 1703 1704 default: 1705 return -ENOIOCTLCMD; 1706 } 1707 1708 return 0; 1709 } 1710 EXPORT_SYMBOL(udp_ioctl); 1711 1712 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1713 int *off, int *err) 1714 { 1715 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1716 struct sk_buff_head *queue; 1717 struct sk_buff *last; 1718 long timeo; 1719 int error; 1720 1721 queue = &udp_sk(sk)->reader_queue; 1722 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1723 do { 1724 struct sk_buff *skb; 1725 1726 error = sock_error(sk); 1727 if (error) 1728 break; 1729 1730 error = -EAGAIN; 1731 do { 1732 spin_lock_bh(&queue->lock); 1733 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1734 err, &last); 1735 if (skb) { 1736 if (!(flags & MSG_PEEK)) 1737 udp_skb_destructor(sk, skb); 1738 spin_unlock_bh(&queue->lock); 1739 return skb; 1740 } 1741 1742 if (skb_queue_empty_lockless(sk_queue)) { 1743 spin_unlock_bh(&queue->lock); 1744 goto busy_check; 1745 } 1746 1747 /* refill the reader queue and walk it again 1748 * keep both queues locked to avoid re-acquiring 1749 * the sk_receive_queue lock if fwd memory scheduling 1750 * is needed. 1751 */ 1752 spin_lock(&sk_queue->lock); 1753 skb_queue_splice_tail_init(sk_queue, queue); 1754 1755 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1756 err, &last); 1757 if (skb && !(flags & MSG_PEEK)) 1758 udp_skb_dtor_locked(sk, skb); 1759 spin_unlock(&sk_queue->lock); 1760 spin_unlock_bh(&queue->lock); 1761 if (skb) 1762 return skb; 1763 1764 busy_check: 1765 if (!sk_can_busy_loop(sk)) 1766 break; 1767 1768 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1769 } while (!skb_queue_empty_lockless(sk_queue)); 1770 1771 /* sk_queue is empty, reader_queue may contain peeked packets */ 1772 } while (timeo && 1773 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1774 &error, &timeo, 1775 (struct sk_buff *)sk_queue)); 1776 1777 *err = error; 1778 return NULL; 1779 } 1780 EXPORT_SYMBOL(__skb_recv_udp); 1781 1782 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) 1783 { 1784 struct sk_buff *skb; 1785 int err; 1786 1787 try_again: 1788 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1789 if (!skb) 1790 return err; 1791 1792 if (udp_lib_checksum_complete(skb)) { 1793 int is_udplite = IS_UDPLITE(sk); 1794 struct net *net = sock_net(sk); 1795 1796 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite); 1797 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite); 1798 atomic_inc(&sk->sk_drops); 1799 kfree_skb(skb); 1800 goto try_again; 1801 } 1802 1803 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); 1804 return recv_actor(sk, skb); 1805 } 1806 EXPORT_SYMBOL(udp_read_skb); 1807 1808 /* 1809 * This should be easy, if there is something there we 1810 * return it, otherwise we block. 1811 */ 1812 1813 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 1814 int *addr_len) 1815 { 1816 struct inet_sock *inet = inet_sk(sk); 1817 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1818 struct sk_buff *skb; 1819 unsigned int ulen, copied; 1820 int off, err, peeking = flags & MSG_PEEK; 1821 int is_udplite = IS_UDPLITE(sk); 1822 bool checksum_valid = false; 1823 1824 if (flags & MSG_ERRQUEUE) 1825 return ip_recv_error(sk, msg, len, addr_len); 1826 1827 try_again: 1828 off = sk_peek_offset(sk, flags); 1829 skb = __skb_recv_udp(sk, flags, &off, &err); 1830 if (!skb) 1831 return err; 1832 if (ccs_socket_post_recvmsg_permission(sk, skb, flags)) 1833 return -EAGAIN; /* Hope less harmful than -EPERM. */ 1834 1835 ulen = udp_skb_len(skb); 1836 copied = len; 1837 if (copied > ulen - off) 1838 copied = ulen - off; 1839 else if (copied < ulen) 1840 msg->msg_flags |= MSG_TRUNC; 1841 1842 /* 1843 * If checksum is needed at all, try to do it while copying the 1844 * data. If the data is truncated, or if we only want a partial 1845 * coverage checksum (UDP-Lite), do it before the copy. 1846 */ 1847 1848 if (copied < ulen || peeking || 1849 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1850 checksum_valid = udp_skb_csum_unnecessary(skb) || 1851 !__udp_lib_checksum_complete(skb); 1852 if (!checksum_valid) 1853 goto csum_copy_err; 1854 } 1855 1856 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1857 if (udp_skb_is_linear(skb)) 1858 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1859 else 1860 err = skb_copy_datagram_msg(skb, off, msg, copied); 1861 } else { 1862 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1863 1864 if (err == -EINVAL) 1865 goto csum_copy_err; 1866 } 1867 1868 if (unlikely(err)) { 1869 if (!peeking) { 1870 atomic_inc(&sk->sk_drops); 1871 UDP_INC_STATS(sock_net(sk), 1872 UDP_MIB_INERRORS, is_udplite); 1873 } 1874 kfree_skb(skb); 1875 return err; 1876 } 1877 1878 if (!peeking) 1879 UDP_INC_STATS(sock_net(sk), 1880 UDP_MIB_INDATAGRAMS, is_udplite); 1881 1882 sock_recv_cmsgs(msg, sk, skb); 1883 1884 /* Copy the address. */ 1885 if (sin) { 1886 sin->sin_family = AF_INET; 1887 sin->sin_port = udp_hdr(skb)->source; 1888 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1889 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1890 *addr_len = sizeof(*sin); 1891 1892 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1893 (struct sockaddr *)sin, 1894 addr_len); 1895 } 1896 1897 if (udp_test_bit(GRO_ENABLED, sk)) 1898 udp_cmsg_recv(msg, sk, skb); 1899 1900 if (inet_cmsg_flags(inet)) 1901 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1902 1903 err = copied; 1904 if (flags & MSG_TRUNC) 1905 err = ulen; 1906 1907 skb_consume_udp(sk, skb, peeking ? -err : err); 1908 return err; 1909 1910 csum_copy_err: 1911 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1912 udp_skb_destructor)) { 1913 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1914 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1915 } 1916 kfree_skb(skb); 1917 1918 /* starting over for a new packet, but check if we need to yield */ 1919 cond_resched(); 1920 msg->msg_flags &= ~MSG_TRUNC; 1921 goto try_again; 1922 } 1923 1924 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1925 { 1926 /* This check is replicated from __ip4_datagram_connect() and 1927 * intended to prevent BPF program called below from accessing bytes 1928 * that are out of the bound specified by user in addr_len. 1929 */ 1930 if (addr_len < sizeof(struct sockaddr_in)) 1931 return -EINVAL; 1932 1933 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len); 1934 } 1935 EXPORT_SYMBOL(udp_pre_connect); 1936 1937 int __udp_disconnect(struct sock *sk, int flags) 1938 { 1939 struct inet_sock *inet = inet_sk(sk); 1940 /* 1941 * 1003.1g - break association. 1942 */ 1943 1944 sk->sk_state = TCP_CLOSE; 1945 inet->inet_daddr = 0; 1946 inet->inet_dport = 0; 1947 sock_rps_reset_rxhash(sk); 1948 sk->sk_bound_dev_if = 0; 1949 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1950 inet_reset_saddr(sk); 1951 if (sk->sk_prot->rehash && 1952 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1953 sk->sk_prot->rehash(sk); 1954 } 1955 1956 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1957 sk->sk_prot->unhash(sk); 1958 inet->inet_sport = 0; 1959 } 1960 sk_dst_reset(sk); 1961 return 0; 1962 } 1963 EXPORT_SYMBOL(__udp_disconnect); 1964 1965 int udp_disconnect(struct sock *sk, int flags) 1966 { 1967 lock_sock(sk); 1968 __udp_disconnect(sk, flags); 1969 release_sock(sk); 1970 return 0; 1971 } 1972 EXPORT_SYMBOL(udp_disconnect); 1973 1974 void udp_lib_unhash(struct sock *sk) 1975 { 1976 if (sk_hashed(sk)) { 1977 struct udp_table *udptable = udp_get_table_prot(sk); 1978 struct udp_hslot *hslot, *hslot2; 1979 1980 hslot = udp_hashslot(udptable, sock_net(sk), 1981 udp_sk(sk)->udp_port_hash); 1982 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1983 1984 spin_lock_bh(&hslot->lock); 1985 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1986 reuseport_detach_sock(sk); 1987 if (sk_del_node_init_rcu(sk)) { 1988 hslot->count--; 1989 inet_sk(sk)->inet_num = 0; 1990 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1991 1992 spin_lock(&hslot2->lock); 1993 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1994 hslot2->count--; 1995 spin_unlock(&hslot2->lock); 1996 } 1997 spin_unlock_bh(&hslot->lock); 1998 } 1999 } 2000 EXPORT_SYMBOL(udp_lib_unhash); 2001 2002 /* 2003 * inet_rcv_saddr was changed, we must rehash secondary hash 2004 */ 2005 void udp_lib_rehash(struct sock *sk, u16 newhash) 2006 { 2007 if (sk_hashed(sk)) { 2008 struct udp_table *udptable = udp_get_table_prot(sk); 2009 struct udp_hslot *hslot, *hslot2, *nhslot2; 2010 2011 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2012 nhslot2 = udp_hashslot2(udptable, newhash); 2013 udp_sk(sk)->udp_portaddr_hash = newhash; 2014 2015 if (hslot2 != nhslot2 || 2016 rcu_access_pointer(sk->sk_reuseport_cb)) { 2017 hslot = udp_hashslot(udptable, sock_net(sk), 2018 udp_sk(sk)->udp_port_hash); 2019 /* we must lock primary chain too */ 2020 spin_lock_bh(&hslot->lock); 2021 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2022 reuseport_detach_sock(sk); 2023 2024 if (hslot2 != nhslot2) { 2025 spin_lock(&hslot2->lock); 2026 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2027 hslot2->count--; 2028 spin_unlock(&hslot2->lock); 2029 2030 spin_lock(&nhslot2->lock); 2031 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2032 &nhslot2->head); 2033 nhslot2->count++; 2034 spin_unlock(&nhslot2->lock); 2035 } 2036 2037 spin_unlock_bh(&hslot->lock); 2038 } 2039 } 2040 } 2041 EXPORT_SYMBOL(udp_lib_rehash); 2042 2043 void udp_v4_rehash(struct sock *sk) 2044 { 2045 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2046 inet_sk(sk)->inet_rcv_saddr, 2047 inet_sk(sk)->inet_num); 2048 udp_lib_rehash(sk, new_hash); 2049 } 2050 2051 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2052 { 2053 int rc; 2054 2055 if (inet_sk(sk)->inet_daddr) { 2056 sock_rps_save_rxhash(sk, skb); 2057 sk_mark_napi_id(sk, skb); 2058 sk_incoming_cpu_update(sk); 2059 } else { 2060 sk_mark_napi_id_once(sk, skb); 2061 } 2062 2063 rc = __udp_enqueue_schedule_skb(sk, skb); 2064 if (rc < 0) { 2065 int is_udplite = IS_UDPLITE(sk); 2066 int drop_reason; 2067 2068 /* Note that an ENOMEM error is charged twice */ 2069 if (rc == -ENOMEM) { 2070 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2071 is_udplite); 2072 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 2073 } else { 2074 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2075 is_udplite); 2076 drop_reason = SKB_DROP_REASON_PROTO_MEM; 2077 } 2078 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2079 trace_udp_fail_queue_rcv_skb(rc, sk, skb); 2080 sk_skb_reason_drop(sk, skb, drop_reason); 2081 return -1; 2082 } 2083 2084 return 0; 2085 } 2086 2087 /* returns: 2088 * -1: error 2089 * 0: success 2090 * >0: "udp encap" protocol resubmission 2091 * 2092 * Note that in the success and error cases, the skb is assumed to 2093 * have either been requeued or freed. 2094 */ 2095 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2096 { 2097 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2098 struct udp_sock *up = udp_sk(sk); 2099 int is_udplite = IS_UDPLITE(sk); 2100 2101 /* 2102 * Charge it to the socket, dropping if the queue is full. 2103 */ 2104 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 2105 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2106 goto drop; 2107 } 2108 nf_reset_ct(skb); 2109 2110 if (static_branch_unlikely(&udp_encap_needed_key) && 2111 READ_ONCE(up->encap_type)) { 2112 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2113 2114 /* 2115 * This is an encapsulation socket so pass the skb to 2116 * the socket's udp_encap_rcv() hook. Otherwise, just 2117 * fall through and pass this up the UDP socket. 2118 * up->encap_rcv() returns the following value: 2119 * =0 if skb was successfully passed to the encap 2120 * handler or was discarded by it. 2121 * >0 if skb should be passed on to UDP. 2122 * <0 if skb should be resubmitted as proto -N 2123 */ 2124 2125 /* if we're overly short, let UDP handle it */ 2126 encap_rcv = READ_ONCE(up->encap_rcv); 2127 if (encap_rcv) { 2128 int ret; 2129 2130 /* Verify checksum before giving to encap */ 2131 if (udp_lib_checksum_complete(skb)) 2132 goto csum_error; 2133 2134 ret = encap_rcv(sk, skb); 2135 if (ret <= 0) { 2136 __UDP_INC_STATS(sock_net(sk), 2137 UDP_MIB_INDATAGRAMS, 2138 is_udplite); 2139 return -ret; 2140 } 2141 } 2142 2143 /* FALLTHROUGH -- it's a UDP Packet */ 2144 } 2145 2146 /* 2147 * UDP-Lite specific tests, ignored on UDP sockets 2148 */ 2149 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) { 2150 u16 pcrlen = READ_ONCE(up->pcrlen); 2151 2152 /* 2153 * MIB statistics other than incrementing the error count are 2154 * disabled for the following two types of errors: these depend 2155 * on the application settings, not on the functioning of the 2156 * protocol stack as such. 2157 * 2158 * RFC 3828 here recommends (sec 3.3): "There should also be a 2159 * way ... to ... at least let the receiving application block 2160 * delivery of packets with coverage values less than a value 2161 * provided by the application." 2162 */ 2163 if (pcrlen == 0) { /* full coverage was set */ 2164 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2165 UDP_SKB_CB(skb)->cscov, skb->len); 2166 goto drop; 2167 } 2168 /* The next case involves violating the min. coverage requested 2169 * by the receiver. This is subtle: if receiver wants x and x is 2170 * greater than the buffersize/MTU then receiver will complain 2171 * that it wants x while sender emits packets of smaller size y. 2172 * Therefore the above ...()->partial_cov statement is essential. 2173 */ 2174 if (UDP_SKB_CB(skb)->cscov < pcrlen) { 2175 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2176 UDP_SKB_CB(skb)->cscov, pcrlen); 2177 goto drop; 2178 } 2179 } 2180 2181 prefetch(&sk->sk_rmem_alloc); 2182 if (rcu_access_pointer(sk->sk_filter) && 2183 udp_lib_checksum_complete(skb)) 2184 goto csum_error; 2185 2186 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 2187 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 2188 goto drop; 2189 } 2190 2191 udp_csum_pull_header(skb); 2192 2193 ipv4_pktinfo_prepare(sk, skb, true); 2194 return __udp_queue_rcv_skb(sk, skb); 2195 2196 csum_error: 2197 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2198 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2199 drop: 2200 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2201 atomic_inc(&sk->sk_drops); 2202 sk_skb_reason_drop(sk, skb, drop_reason); 2203 return -1; 2204 } 2205 2206 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2207 { 2208 struct sk_buff *next, *segs; 2209 int ret; 2210 2211 if (likely(!udp_unexpected_gso(sk, skb))) 2212 return udp_queue_rcv_one_skb(sk, skb); 2213 2214 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2215 __skb_push(skb, -skb_mac_offset(skb)); 2216 segs = udp_rcv_segment(sk, skb, true); 2217 skb_list_walk_safe(segs, skb, next) { 2218 __skb_pull(skb, skb_transport_offset(skb)); 2219 2220 udp_post_segment_fix_csum(skb); 2221 ret = udp_queue_rcv_one_skb(sk, skb); 2222 if (ret > 0) 2223 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2224 } 2225 return 0; 2226 } 2227 2228 /* For TCP sockets, sk_rx_dst is protected by socket lock 2229 * For UDP, we use xchg() to guard against concurrent changes. 2230 */ 2231 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2232 { 2233 struct dst_entry *old; 2234 2235 if (dst_hold_safe(dst)) { 2236 old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst))); 2237 dst_release(old); 2238 return old != dst; 2239 } 2240 return false; 2241 } 2242 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2243 2244 /* 2245 * Multicasts and broadcasts go to each listener. 2246 * 2247 * Note: called only from the BH handler context. 2248 */ 2249 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2250 struct udphdr *uh, 2251 __be32 saddr, __be32 daddr, 2252 struct udp_table *udptable, 2253 int proto) 2254 { 2255 struct sock *sk, *first = NULL; 2256 unsigned short hnum = ntohs(uh->dest); 2257 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2258 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2259 unsigned int offset = offsetof(typeof(*sk), sk_node); 2260 int dif = skb->dev->ifindex; 2261 int sdif = inet_sdif(skb); 2262 struct hlist_node *node; 2263 struct sk_buff *nskb; 2264 2265 if (use_hash2) { 2266 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2267 udptable->mask; 2268 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2269 start_lookup: 2270 hslot = &udptable->hash2[hash2]; 2271 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2272 } 2273 2274 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2275 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2276 uh->source, saddr, dif, sdif, hnum)) 2277 continue; 2278 2279 if (!first) { 2280 first = sk; 2281 continue; 2282 } 2283 nskb = skb_clone(skb, GFP_ATOMIC); 2284 2285 if (unlikely(!nskb)) { 2286 atomic_inc(&sk->sk_drops); 2287 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2288 IS_UDPLITE(sk)); 2289 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2290 IS_UDPLITE(sk)); 2291 continue; 2292 } 2293 if (udp_queue_rcv_skb(sk, nskb) > 0) 2294 consume_skb(nskb); 2295 } 2296 2297 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2298 if (use_hash2 && hash2 != hash2_any) { 2299 hash2 = hash2_any; 2300 goto start_lookup; 2301 } 2302 2303 if (first) { 2304 if (udp_queue_rcv_skb(first, skb) > 0) 2305 consume_skb(skb); 2306 } else { 2307 kfree_skb(skb); 2308 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2309 proto == IPPROTO_UDPLITE); 2310 } 2311 return 0; 2312 } 2313 2314 /* Initialize UDP checksum. If exited with zero value (success), 2315 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2316 * Otherwise, csum completion requires checksumming packet body, 2317 * including udp header and folding it to skb->csum. 2318 */ 2319 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2320 int proto) 2321 { 2322 int err; 2323 2324 UDP_SKB_CB(skb)->partial_cov = 0; 2325 UDP_SKB_CB(skb)->cscov = skb->len; 2326 2327 if (proto == IPPROTO_UDPLITE) { 2328 err = udplite_checksum_init(skb, uh); 2329 if (err) 2330 return err; 2331 2332 if (UDP_SKB_CB(skb)->partial_cov) { 2333 skb->csum = inet_compute_pseudo(skb, proto); 2334 return 0; 2335 } 2336 } 2337 2338 /* Note, we are only interested in != 0 or == 0, thus the 2339 * force to int. 2340 */ 2341 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2342 inet_compute_pseudo); 2343 if (err) 2344 return err; 2345 2346 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2347 /* If SW calculated the value, we know it's bad */ 2348 if (skb->csum_complete_sw) 2349 return 1; 2350 2351 /* HW says the value is bad. Let's validate that. 2352 * skb->csum is no longer the full packet checksum, 2353 * so don't treat it as such. 2354 */ 2355 skb_checksum_complete_unset(skb); 2356 } 2357 2358 return 0; 2359 } 2360 2361 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2362 * return code conversion for ip layer consumption 2363 */ 2364 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2365 struct udphdr *uh) 2366 { 2367 int ret; 2368 2369 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2370 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2371 2372 ret = udp_queue_rcv_skb(sk, skb); 2373 2374 /* a return value > 0 means to resubmit the input, but 2375 * it wants the return to be -protocol, or 0 2376 */ 2377 if (ret > 0) 2378 return -ret; 2379 return 0; 2380 } 2381 2382 /* 2383 * All we need to do is get the socket, and then do a checksum. 2384 */ 2385 2386 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2387 int proto) 2388 { 2389 struct sock *sk = NULL; 2390 struct udphdr *uh; 2391 unsigned short ulen; 2392 struct rtable *rt = skb_rtable(skb); 2393 __be32 saddr, daddr; 2394 struct net *net = dev_net(skb->dev); 2395 bool refcounted; 2396 int drop_reason; 2397 2398 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2399 2400 /* 2401 * Validate the packet. 2402 */ 2403 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2404 goto drop; /* No space for header. */ 2405 2406 uh = udp_hdr(skb); 2407 ulen = ntohs(uh->len); 2408 saddr = ip_hdr(skb)->saddr; 2409 daddr = ip_hdr(skb)->daddr; 2410 2411 if (ulen > skb->len) 2412 goto short_packet; 2413 2414 if (proto == IPPROTO_UDP) { 2415 /* UDP validates ulen. */ 2416 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2417 goto short_packet; 2418 uh = udp_hdr(skb); 2419 } 2420 2421 if (udp4_csum_init(skb, uh, proto)) 2422 goto csum_error; 2423 2424 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 2425 &refcounted, udp_ehashfn); 2426 if (IS_ERR(sk)) 2427 goto no_sk; 2428 2429 if (sk) { 2430 struct dst_entry *dst = skb_dst(skb); 2431 int ret; 2432 2433 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2434 udp_sk_rx_dst_set(sk, dst); 2435 2436 ret = udp_unicast_rcv_skb(sk, skb, uh); 2437 if (refcounted) 2438 sock_put(sk); 2439 return ret; 2440 } 2441 2442 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2443 return __udp4_lib_mcast_deliver(net, skb, uh, 2444 saddr, daddr, udptable, proto); 2445 2446 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2447 if (sk) 2448 return udp_unicast_rcv_skb(sk, skb, uh); 2449 no_sk: 2450 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2451 goto drop; 2452 nf_reset_ct(skb); 2453 2454 /* No socket. Drop packet silently, if checksum is wrong */ 2455 if (udp_lib_checksum_complete(skb)) 2456 goto csum_error; 2457 2458 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2459 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2460 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2461 2462 /* 2463 * Hmm. We got an UDP packet to a port to which we 2464 * don't wanna listen. Ignore it. 2465 */ 2466 sk_skb_reason_drop(sk, skb, drop_reason); 2467 return 0; 2468 2469 short_packet: 2470 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2471 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2472 proto == IPPROTO_UDPLITE ? "Lite" : "", 2473 &saddr, ntohs(uh->source), 2474 ulen, skb->len, 2475 &daddr, ntohs(uh->dest)); 2476 goto drop; 2477 2478 csum_error: 2479 /* 2480 * RFC1122: OK. Discards the bad packet silently (as far as 2481 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2482 */ 2483 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2484 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2485 proto == IPPROTO_UDPLITE ? "Lite" : "", 2486 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2487 ulen); 2488 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2489 drop: 2490 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2491 sk_skb_reason_drop(sk, skb, drop_reason); 2492 return 0; 2493 } 2494 2495 /* We can only early demux multicast if there is a single matching socket. 2496 * If more than one socket found returns NULL 2497 */ 2498 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2499 __be16 loc_port, __be32 loc_addr, 2500 __be16 rmt_port, __be32 rmt_addr, 2501 int dif, int sdif) 2502 { 2503 struct udp_table *udptable = net->ipv4.udp_table; 2504 unsigned short hnum = ntohs(loc_port); 2505 struct sock *sk, *result; 2506 struct udp_hslot *hslot; 2507 unsigned int slot; 2508 2509 slot = udp_hashfn(net, hnum, udptable->mask); 2510 hslot = &udptable->hash[slot]; 2511 2512 /* Do not bother scanning a too big list */ 2513 if (hslot->count > 10) 2514 return NULL; 2515 2516 result = NULL; 2517 sk_for_each_rcu(sk, &hslot->head) { 2518 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2519 rmt_port, rmt_addr, dif, sdif, hnum)) { 2520 if (result) 2521 return NULL; 2522 result = sk; 2523 } 2524 } 2525 2526 return result; 2527 } 2528 2529 /* For unicast we should only early demux connected sockets or we can 2530 * break forwarding setups. The chains here can be long so only check 2531 * if the first socket is an exact match and if not move on. 2532 */ 2533 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2534 __be16 loc_port, __be32 loc_addr, 2535 __be16 rmt_port, __be32 rmt_addr, 2536 int dif, int sdif) 2537 { 2538 struct udp_table *udptable = net->ipv4.udp_table; 2539 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2540 unsigned short hnum = ntohs(loc_port); 2541 unsigned int hash2, slot2; 2542 struct udp_hslot *hslot2; 2543 __portpair ports; 2544 struct sock *sk; 2545 2546 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2547 slot2 = hash2 & udptable->mask; 2548 hslot2 = &udptable->hash2[slot2]; 2549 ports = INET_COMBINED_PORTS(rmt_port, hnum); 2550 2551 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2552 if (inet_match(net, sk, acookie, ports, dif, sdif)) 2553 return sk; 2554 /* Only check first socket in chain */ 2555 break; 2556 } 2557 return NULL; 2558 } 2559 2560 int udp_v4_early_demux(struct sk_buff *skb) 2561 { 2562 struct net *net = dev_net(skb->dev); 2563 struct in_device *in_dev = NULL; 2564 const struct iphdr *iph; 2565 const struct udphdr *uh; 2566 struct sock *sk = NULL; 2567 struct dst_entry *dst; 2568 int dif = skb->dev->ifindex; 2569 int sdif = inet_sdif(skb); 2570 int ours; 2571 2572 /* validate the packet */ 2573 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2574 return 0; 2575 2576 iph = ip_hdr(skb); 2577 uh = udp_hdr(skb); 2578 2579 if (skb->pkt_type == PACKET_MULTICAST) { 2580 in_dev = __in_dev_get_rcu(skb->dev); 2581 2582 if (!in_dev) 2583 return 0; 2584 2585 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2586 iph->protocol); 2587 if (!ours) 2588 return 0; 2589 2590 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2591 uh->source, iph->saddr, 2592 dif, sdif); 2593 } else if (skb->pkt_type == PACKET_HOST) { 2594 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2595 uh->source, iph->saddr, dif, sdif); 2596 } 2597 2598 if (!sk) 2599 return 0; 2600 2601 skb->sk = sk; 2602 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk)); 2603 skb->destructor = sock_pfree; 2604 dst = rcu_dereference(sk->sk_rx_dst); 2605 2606 if (dst) 2607 dst = dst_check(dst, 0); 2608 if (dst) { 2609 u32 itag = 0; 2610 2611 /* set noref for now. 2612 * any place which wants to hold dst has to call 2613 * dst_hold_safe() 2614 */ 2615 skb_dst_set_noref(skb, dst); 2616 2617 /* for unconnected multicast sockets we need to validate 2618 * the source on each packet 2619 */ 2620 if (!inet_sk(sk)->inet_daddr && in_dev) 2621 return ip_mc_validate_source(skb, iph->daddr, 2622 iph->saddr, 2623 iph->tos & IPTOS_RT_MASK, 2624 skb->dev, in_dev, &itag); 2625 } 2626 return 0; 2627 } 2628 2629 int udp_rcv(struct sk_buff *skb) 2630 { 2631 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 2632 } 2633 2634 void udp_destroy_sock(struct sock *sk) 2635 { 2636 struct udp_sock *up = udp_sk(sk); 2637 bool slow = lock_sock_fast(sk); 2638 2639 /* protects from races with udp_abort() */ 2640 sock_set_flag(sk, SOCK_DEAD); 2641 udp_flush_pending_frames(sk); 2642 unlock_sock_fast(sk, slow); 2643 if (static_branch_unlikely(&udp_encap_needed_key)) { 2644 if (up->encap_type) { 2645 void (*encap_destroy)(struct sock *sk); 2646 encap_destroy = READ_ONCE(up->encap_destroy); 2647 if (encap_destroy) 2648 encap_destroy(sk); 2649 } 2650 if (udp_test_bit(ENCAP_ENABLED, sk)) 2651 static_branch_dec(&udp_encap_needed_key); 2652 } 2653 } 2654 2655 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family, 2656 struct sock *sk) 2657 { 2658 #ifdef CONFIG_XFRM 2659 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) { 2660 if (family == AF_INET) 2661 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv); 2662 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) 2663 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv); 2664 } 2665 #endif 2666 } 2667 2668 /* 2669 * Socket option code for UDP 2670 */ 2671 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2672 sockptr_t optval, unsigned int optlen, 2673 int (*push_pending_frames)(struct sock *)) 2674 { 2675 struct udp_sock *up = udp_sk(sk); 2676 int val, valbool; 2677 int err = 0; 2678 int is_udplite = IS_UDPLITE(sk); 2679 2680 if (level == SOL_SOCKET) { 2681 err = sk_setsockopt(sk, level, optname, optval, optlen); 2682 2683 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) { 2684 sockopt_lock_sock(sk); 2685 /* paired with READ_ONCE in udp_rmem_release() */ 2686 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2); 2687 sockopt_release_sock(sk); 2688 } 2689 return err; 2690 } 2691 2692 if (optlen < sizeof(int)) 2693 return -EINVAL; 2694 2695 if (copy_from_sockptr(&val, optval, sizeof(val))) 2696 return -EFAULT; 2697 2698 valbool = val ? 1 : 0; 2699 2700 switch (optname) { 2701 case UDP_CORK: 2702 if (val != 0) { 2703 udp_set_bit(CORK, sk); 2704 } else { 2705 udp_clear_bit(CORK, sk); 2706 lock_sock(sk); 2707 push_pending_frames(sk); 2708 release_sock(sk); 2709 } 2710 break; 2711 2712 case UDP_ENCAP: 2713 switch (val) { 2714 case 0: 2715 #ifdef CONFIG_XFRM 2716 case UDP_ENCAP_ESPINUDP: 2717 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk); 2718 #if IS_ENABLED(CONFIG_IPV6) 2719 if (sk->sk_family == AF_INET6) 2720 WRITE_ONCE(up->encap_rcv, 2721 ipv6_stub->xfrm6_udp_encap_rcv); 2722 else 2723 #endif 2724 WRITE_ONCE(up->encap_rcv, 2725 xfrm4_udp_encap_rcv); 2726 #endif 2727 fallthrough; 2728 case UDP_ENCAP_L2TPINUDP: 2729 WRITE_ONCE(up->encap_type, val); 2730 udp_tunnel_encap_enable(sk); 2731 break; 2732 default: 2733 err = -ENOPROTOOPT; 2734 break; 2735 } 2736 break; 2737 2738 case UDP_NO_CHECK6_TX: 2739 udp_set_no_check6_tx(sk, valbool); 2740 break; 2741 2742 case UDP_NO_CHECK6_RX: 2743 udp_set_no_check6_rx(sk, valbool); 2744 break; 2745 2746 case UDP_SEGMENT: 2747 if (val < 0 || val > USHRT_MAX) 2748 return -EINVAL; 2749 WRITE_ONCE(up->gso_size, val); 2750 break; 2751 2752 case UDP_GRO: 2753 2754 /* when enabling GRO, accept the related GSO packet type */ 2755 if (valbool) 2756 udp_tunnel_encap_enable(sk); 2757 udp_assign_bit(GRO_ENABLED, sk, valbool); 2758 udp_assign_bit(ACCEPT_L4, sk, valbool); 2759 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk); 2760 break; 2761 2762 /* 2763 * UDP-Lite's partial checksum coverage (RFC 3828). 2764 */ 2765 /* The sender sets actual checksum coverage length via this option. 2766 * The case coverage > packet length is handled by send module. */ 2767 case UDPLITE_SEND_CSCOV: 2768 if (!is_udplite) /* Disable the option on UDP sockets */ 2769 return -ENOPROTOOPT; 2770 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2771 val = 8; 2772 else if (val > USHRT_MAX) 2773 val = USHRT_MAX; 2774 WRITE_ONCE(up->pcslen, val); 2775 udp_set_bit(UDPLITE_SEND_CC, sk); 2776 break; 2777 2778 /* The receiver specifies a minimum checksum coverage value. To make 2779 * sense, this should be set to at least 8 (as done below). If zero is 2780 * used, this again means full checksum coverage. */ 2781 case UDPLITE_RECV_CSCOV: 2782 if (!is_udplite) /* Disable the option on UDP sockets */ 2783 return -ENOPROTOOPT; 2784 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2785 val = 8; 2786 else if (val > USHRT_MAX) 2787 val = USHRT_MAX; 2788 WRITE_ONCE(up->pcrlen, val); 2789 udp_set_bit(UDPLITE_RECV_CC, sk); 2790 break; 2791 2792 default: 2793 err = -ENOPROTOOPT; 2794 break; 2795 } 2796 2797 return err; 2798 } 2799 EXPORT_SYMBOL(udp_lib_setsockopt); 2800 2801 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2802 unsigned int optlen) 2803 { 2804 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 2805 return udp_lib_setsockopt(sk, level, optname, 2806 optval, optlen, 2807 udp_push_pending_frames); 2808 return ip_setsockopt(sk, level, optname, optval, optlen); 2809 } 2810 2811 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2812 char __user *optval, int __user *optlen) 2813 { 2814 struct udp_sock *up = udp_sk(sk); 2815 int val, len; 2816 2817 if (get_user(len, optlen)) 2818 return -EFAULT; 2819 2820 if (len < 0) 2821 return -EINVAL; 2822 2823 len = min_t(unsigned int, len, sizeof(int)); 2824 2825 switch (optname) { 2826 case UDP_CORK: 2827 val = udp_test_bit(CORK, sk); 2828 break; 2829 2830 case UDP_ENCAP: 2831 val = READ_ONCE(up->encap_type); 2832 break; 2833 2834 case UDP_NO_CHECK6_TX: 2835 val = udp_get_no_check6_tx(sk); 2836 break; 2837 2838 case UDP_NO_CHECK6_RX: 2839 val = udp_get_no_check6_rx(sk); 2840 break; 2841 2842 case UDP_SEGMENT: 2843 val = READ_ONCE(up->gso_size); 2844 break; 2845 2846 case UDP_GRO: 2847 val = udp_test_bit(GRO_ENABLED, sk); 2848 break; 2849 2850 /* The following two cannot be changed on UDP sockets, the return is 2851 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2852 case UDPLITE_SEND_CSCOV: 2853 val = READ_ONCE(up->pcslen); 2854 break; 2855 2856 case UDPLITE_RECV_CSCOV: 2857 val = READ_ONCE(up->pcrlen); 2858 break; 2859 2860 default: 2861 return -ENOPROTOOPT; 2862 } 2863 2864 if (put_user(len, optlen)) 2865 return -EFAULT; 2866 if (copy_to_user(optval, &val, len)) 2867 return -EFAULT; 2868 return 0; 2869 } 2870 EXPORT_SYMBOL(udp_lib_getsockopt); 2871 2872 int udp_getsockopt(struct sock *sk, int level, int optname, 2873 char __user *optval, int __user *optlen) 2874 { 2875 if (level == SOL_UDP || level == SOL_UDPLITE) 2876 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2877 return ip_getsockopt(sk, level, optname, optval, optlen); 2878 } 2879 2880 /** 2881 * udp_poll - wait for a UDP event. 2882 * @file: - file struct 2883 * @sock: - socket 2884 * @wait: - poll table 2885 * 2886 * This is same as datagram poll, except for the special case of 2887 * blocking sockets. If application is using a blocking fd 2888 * and a packet with checksum error is in the queue; 2889 * then it could get return from select indicating data available 2890 * but then block when reading it. Add special case code 2891 * to work around these arguably broken applications. 2892 */ 2893 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2894 { 2895 __poll_t mask = datagram_poll(file, sock, wait); 2896 struct sock *sk = sock->sk; 2897 2898 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2899 mask |= EPOLLIN | EPOLLRDNORM; 2900 2901 /* Check for false positives due to checksum errors */ 2902 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2903 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2904 mask &= ~(EPOLLIN | EPOLLRDNORM); 2905 2906 /* psock ingress_msg queue should not contain any bad checksum frames */ 2907 if (sk_is_readable(sk)) 2908 mask |= EPOLLIN | EPOLLRDNORM; 2909 return mask; 2910 2911 } 2912 EXPORT_SYMBOL(udp_poll); 2913 2914 int udp_abort(struct sock *sk, int err) 2915 { 2916 if (!has_current_bpf_ctx()) 2917 lock_sock(sk); 2918 2919 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2920 * with close() 2921 */ 2922 if (sock_flag(sk, SOCK_DEAD)) 2923 goto out; 2924 2925 sk->sk_err = err; 2926 sk_error_report(sk); 2927 __udp_disconnect(sk, 0); 2928 2929 out: 2930 if (!has_current_bpf_ctx()) 2931 release_sock(sk); 2932 2933 return 0; 2934 } 2935 EXPORT_SYMBOL_GPL(udp_abort); 2936 2937 struct proto udp_prot = { 2938 .name = "UDP", 2939 .owner = THIS_MODULE, 2940 .close = udp_lib_close, 2941 .pre_connect = udp_pre_connect, 2942 .connect = ip4_datagram_connect, 2943 .disconnect = udp_disconnect, 2944 .ioctl = udp_ioctl, 2945 .init = udp_init_sock, 2946 .destroy = udp_destroy_sock, 2947 .setsockopt = udp_setsockopt, 2948 .getsockopt = udp_getsockopt, 2949 .sendmsg = udp_sendmsg, 2950 .recvmsg = udp_recvmsg, 2951 .splice_eof = udp_splice_eof, 2952 .release_cb = ip4_datagram_release_cb, 2953 .hash = udp_lib_hash, 2954 .unhash = udp_lib_unhash, 2955 .rehash = udp_v4_rehash, 2956 .get_port = udp_v4_get_port, 2957 .put_port = udp_lib_unhash, 2958 #ifdef CONFIG_BPF_SYSCALL 2959 .psock_update_sk_prot = udp_bpf_update_proto, 2960 #endif 2961 .memory_allocated = &udp_memory_allocated, 2962 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 2963 2964 .sysctl_mem = sysctl_udp_mem, 2965 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2966 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2967 .obj_size = sizeof(struct udp_sock), 2968 .h.udp_table = NULL, 2969 .diag_destroy = udp_abort, 2970 }; 2971 EXPORT_SYMBOL(udp_prot); 2972 2973 /* ------------------------------------------------------------------------ */ 2974 #ifdef CONFIG_PROC_FS 2975 2976 static unsigned short seq_file_family(const struct seq_file *seq); 2977 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk) 2978 { 2979 unsigned short family = seq_file_family(seq); 2980 2981 /* AF_UNSPEC is used as a match all */ 2982 return ((family == AF_UNSPEC || family == sk->sk_family) && 2983 net_eq(sock_net(sk), seq_file_net(seq))); 2984 } 2985 2986 #ifdef CONFIG_BPF_SYSCALL 2987 static const struct seq_operations bpf_iter_udp_seq_ops; 2988 #endif 2989 static struct udp_table *udp_get_table_seq(struct seq_file *seq, 2990 struct net *net) 2991 { 2992 const struct udp_seq_afinfo *afinfo; 2993 2994 #ifdef CONFIG_BPF_SYSCALL 2995 if (seq->op == &bpf_iter_udp_seq_ops) 2996 return net->ipv4.udp_table; 2997 #endif 2998 2999 afinfo = pde_data(file_inode(seq->file)); 3000 return afinfo->udp_table ? : net->ipv4.udp_table; 3001 } 3002 3003 static struct sock *udp_get_first(struct seq_file *seq, int start) 3004 { 3005 struct udp_iter_state *state = seq->private; 3006 struct net *net = seq_file_net(seq); 3007 struct udp_table *udptable; 3008 struct sock *sk; 3009 3010 udptable = udp_get_table_seq(seq, net); 3011 3012 for (state->bucket = start; state->bucket <= udptable->mask; 3013 ++state->bucket) { 3014 struct udp_hslot *hslot = &udptable->hash[state->bucket]; 3015 3016 if (hlist_empty(&hslot->head)) 3017 continue; 3018 3019 spin_lock_bh(&hslot->lock); 3020 sk_for_each(sk, &hslot->head) { 3021 if (seq_sk_match(seq, sk)) 3022 goto found; 3023 } 3024 spin_unlock_bh(&hslot->lock); 3025 } 3026 sk = NULL; 3027 found: 3028 return sk; 3029 } 3030 3031 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 3032 { 3033 struct udp_iter_state *state = seq->private; 3034 struct net *net = seq_file_net(seq); 3035 struct udp_table *udptable; 3036 3037 do { 3038 sk = sk_next(sk); 3039 } while (sk && !seq_sk_match(seq, sk)); 3040 3041 if (!sk) { 3042 udptable = udp_get_table_seq(seq, net); 3043 3044 if (state->bucket <= udptable->mask) 3045 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3046 3047 return udp_get_first(seq, state->bucket + 1); 3048 } 3049 return sk; 3050 } 3051 3052 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3053 { 3054 struct sock *sk = udp_get_first(seq, 0); 3055 3056 if (sk) 3057 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3058 --pos; 3059 return pos ? NULL : sk; 3060 } 3061 3062 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3063 { 3064 struct udp_iter_state *state = seq->private; 3065 state->bucket = MAX_UDP_PORTS; 3066 3067 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3068 } 3069 EXPORT_SYMBOL(udp_seq_start); 3070 3071 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3072 { 3073 struct sock *sk; 3074 3075 if (v == SEQ_START_TOKEN) 3076 sk = udp_get_idx(seq, 0); 3077 else 3078 sk = udp_get_next(seq, v); 3079 3080 ++*pos; 3081 return sk; 3082 } 3083 EXPORT_SYMBOL(udp_seq_next); 3084 3085 void udp_seq_stop(struct seq_file *seq, void *v) 3086 { 3087 struct udp_iter_state *state = seq->private; 3088 struct udp_table *udptable; 3089 3090 udptable = udp_get_table_seq(seq, seq_file_net(seq)); 3091 3092 if (state->bucket <= udptable->mask) 3093 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3094 } 3095 EXPORT_SYMBOL(udp_seq_stop); 3096 3097 /* ------------------------------------------------------------------------ */ 3098 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3099 int bucket) 3100 { 3101 struct inet_sock *inet = inet_sk(sp); 3102 __be32 dest = inet->inet_daddr; 3103 __be32 src = inet->inet_rcv_saddr; 3104 __u16 destp = ntohs(inet->inet_dport); 3105 __u16 srcp = ntohs(inet->inet_sport); 3106 3107 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3108 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3109 bucket, src, srcp, dest, destp, sp->sk_state, 3110 sk_wmem_alloc_get(sp), 3111 udp_rqueue_get(sp), 3112 0, 0L, 0, 3113 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3114 0, sock_i_ino(sp), 3115 refcount_read(&sp->sk_refcnt), sp, 3116 atomic_read(&sp->sk_drops)); 3117 } 3118 3119 int udp4_seq_show(struct seq_file *seq, void *v) 3120 { 3121 seq_setwidth(seq, 127); 3122 if (v == SEQ_START_TOKEN) 3123 seq_puts(seq, " sl local_address rem_address st tx_queue " 3124 "rx_queue tr tm->when retrnsmt uid timeout " 3125 "inode ref pointer drops"); 3126 else { 3127 struct udp_iter_state *state = seq->private; 3128 3129 udp4_format_sock(v, seq, state->bucket); 3130 } 3131 seq_pad(seq, '\n'); 3132 return 0; 3133 } 3134 3135 #ifdef CONFIG_BPF_SYSCALL 3136 struct bpf_iter__udp { 3137 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3138 __bpf_md_ptr(struct udp_sock *, udp_sk); 3139 uid_t uid __aligned(8); 3140 int bucket __aligned(8); 3141 }; 3142 3143 struct bpf_udp_iter_state { 3144 struct udp_iter_state state; 3145 unsigned int cur_sk; 3146 unsigned int end_sk; 3147 unsigned int max_sk; 3148 int offset; 3149 struct sock **batch; 3150 bool st_bucket_done; 3151 }; 3152 3153 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3154 unsigned int new_batch_sz); 3155 static struct sock *bpf_iter_udp_batch(struct seq_file *seq) 3156 { 3157 struct bpf_udp_iter_state *iter = seq->private; 3158 struct udp_iter_state *state = &iter->state; 3159 struct net *net = seq_file_net(seq); 3160 int resume_bucket, resume_offset; 3161 struct udp_table *udptable; 3162 unsigned int batch_sks = 0; 3163 bool resized = false; 3164 struct sock *sk; 3165 3166 resume_bucket = state->bucket; 3167 resume_offset = iter->offset; 3168 3169 /* The current batch is done, so advance the bucket. */ 3170 if (iter->st_bucket_done) 3171 state->bucket++; 3172 3173 udptable = udp_get_table_seq(seq, net); 3174 3175 again: 3176 /* New batch for the next bucket. 3177 * Iterate over the hash table to find a bucket with sockets matching 3178 * the iterator attributes, and return the first matching socket from 3179 * the bucket. The remaining matched sockets from the bucket are batched 3180 * before releasing the bucket lock. This allows BPF programs that are 3181 * called in seq_show to acquire the bucket lock if needed. 3182 */ 3183 iter->cur_sk = 0; 3184 iter->end_sk = 0; 3185 iter->st_bucket_done = false; 3186 batch_sks = 0; 3187 3188 for (; state->bucket <= udptable->mask; state->bucket++) { 3189 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket]; 3190 3191 if (hlist_empty(&hslot2->head)) 3192 continue; 3193 3194 iter->offset = 0; 3195 spin_lock_bh(&hslot2->lock); 3196 udp_portaddr_for_each_entry(sk, &hslot2->head) { 3197 if (seq_sk_match(seq, sk)) { 3198 /* Resume from the last iterated socket at the 3199 * offset in the bucket before iterator was stopped. 3200 */ 3201 if (state->bucket == resume_bucket && 3202 iter->offset < resume_offset) { 3203 ++iter->offset; 3204 continue; 3205 } 3206 if (iter->end_sk < iter->max_sk) { 3207 sock_hold(sk); 3208 iter->batch[iter->end_sk++] = sk; 3209 } 3210 batch_sks++; 3211 } 3212 } 3213 spin_unlock_bh(&hslot2->lock); 3214 3215 if (iter->end_sk) 3216 break; 3217 } 3218 3219 /* All done: no batch made. */ 3220 if (!iter->end_sk) 3221 return NULL; 3222 3223 if (iter->end_sk == batch_sks) { 3224 /* Batching is done for the current bucket; return the first 3225 * socket to be iterated from the batch. 3226 */ 3227 iter->st_bucket_done = true; 3228 goto done; 3229 } 3230 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) { 3231 resized = true; 3232 /* After allocating a larger batch, retry one more time to grab 3233 * the whole bucket. 3234 */ 3235 goto again; 3236 } 3237 done: 3238 return iter->batch[0]; 3239 } 3240 3241 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3242 { 3243 struct bpf_udp_iter_state *iter = seq->private; 3244 struct sock *sk; 3245 3246 /* Whenever seq_next() is called, the iter->cur_sk is 3247 * done with seq_show(), so unref the iter->cur_sk. 3248 */ 3249 if (iter->cur_sk < iter->end_sk) { 3250 sock_put(iter->batch[iter->cur_sk++]); 3251 ++iter->offset; 3252 } 3253 3254 /* After updating iter->cur_sk, check if there are more sockets 3255 * available in the current bucket batch. 3256 */ 3257 if (iter->cur_sk < iter->end_sk) 3258 sk = iter->batch[iter->cur_sk]; 3259 else 3260 /* Prepare a new batch. */ 3261 sk = bpf_iter_udp_batch(seq); 3262 3263 ++*pos; 3264 return sk; 3265 } 3266 3267 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos) 3268 { 3269 /* bpf iter does not support lseek, so it always 3270 * continue from where it was stop()-ped. 3271 */ 3272 if (*pos) 3273 return bpf_iter_udp_batch(seq); 3274 3275 return SEQ_START_TOKEN; 3276 } 3277 3278 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3279 struct udp_sock *udp_sk, uid_t uid, int bucket) 3280 { 3281 struct bpf_iter__udp ctx; 3282 3283 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3284 ctx.meta = meta; 3285 ctx.udp_sk = udp_sk; 3286 ctx.uid = uid; 3287 ctx.bucket = bucket; 3288 return bpf_iter_run_prog(prog, &ctx); 3289 } 3290 3291 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3292 { 3293 struct udp_iter_state *state = seq->private; 3294 struct bpf_iter_meta meta; 3295 struct bpf_prog *prog; 3296 struct sock *sk = v; 3297 uid_t uid; 3298 int ret; 3299 3300 if (v == SEQ_START_TOKEN) 3301 return 0; 3302 3303 lock_sock(sk); 3304 3305 if (unlikely(sk_unhashed(sk))) { 3306 ret = SEQ_SKIP; 3307 goto unlock; 3308 } 3309 3310 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3311 meta.seq = seq; 3312 prog = bpf_iter_get_info(&meta, false); 3313 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3314 3315 unlock: 3316 release_sock(sk); 3317 return ret; 3318 } 3319 3320 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter) 3321 { 3322 while (iter->cur_sk < iter->end_sk) 3323 sock_put(iter->batch[iter->cur_sk++]); 3324 } 3325 3326 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3327 { 3328 struct bpf_udp_iter_state *iter = seq->private; 3329 struct bpf_iter_meta meta; 3330 struct bpf_prog *prog; 3331 3332 if (!v) { 3333 meta.seq = seq; 3334 prog = bpf_iter_get_info(&meta, true); 3335 if (prog) 3336 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3337 } 3338 3339 if (iter->cur_sk < iter->end_sk) { 3340 bpf_iter_udp_put_batch(iter); 3341 iter->st_bucket_done = false; 3342 } 3343 } 3344 3345 static const struct seq_operations bpf_iter_udp_seq_ops = { 3346 .start = bpf_iter_udp_seq_start, 3347 .next = bpf_iter_udp_seq_next, 3348 .stop = bpf_iter_udp_seq_stop, 3349 .show = bpf_iter_udp_seq_show, 3350 }; 3351 #endif 3352 3353 static unsigned short seq_file_family(const struct seq_file *seq) 3354 { 3355 const struct udp_seq_afinfo *afinfo; 3356 3357 #ifdef CONFIG_BPF_SYSCALL 3358 /* BPF iterator: bpf programs to filter sockets. */ 3359 if (seq->op == &bpf_iter_udp_seq_ops) 3360 return AF_UNSPEC; 3361 #endif 3362 3363 /* Proc fs iterator */ 3364 afinfo = pde_data(file_inode(seq->file)); 3365 return afinfo->family; 3366 } 3367 3368 const struct seq_operations udp_seq_ops = { 3369 .start = udp_seq_start, 3370 .next = udp_seq_next, 3371 .stop = udp_seq_stop, 3372 .show = udp4_seq_show, 3373 }; 3374 EXPORT_SYMBOL(udp_seq_ops); 3375 3376 static struct udp_seq_afinfo udp4_seq_afinfo = { 3377 .family = AF_INET, 3378 .udp_table = NULL, 3379 }; 3380 3381 static int __net_init udp4_proc_init_net(struct net *net) 3382 { 3383 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3384 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3385 return -ENOMEM; 3386 return 0; 3387 } 3388 3389 static void __net_exit udp4_proc_exit_net(struct net *net) 3390 { 3391 remove_proc_entry("udp", net->proc_net); 3392 } 3393 3394 static struct pernet_operations udp4_net_ops = { 3395 .init = udp4_proc_init_net, 3396 .exit = udp4_proc_exit_net, 3397 }; 3398 3399 int __init udp4_proc_init(void) 3400 { 3401 return register_pernet_subsys(&udp4_net_ops); 3402 } 3403 3404 void udp4_proc_exit(void) 3405 { 3406 unregister_pernet_subsys(&udp4_net_ops); 3407 } 3408 #endif /* CONFIG_PROC_FS */ 3409 3410 static __initdata unsigned long uhash_entries; 3411 static int __init set_uhash_entries(char *str) 3412 { 3413 ssize_t ret; 3414 3415 if (!str) 3416 return 0; 3417 3418 ret = kstrtoul(str, 0, &uhash_entries); 3419 if (ret) 3420 return 0; 3421 3422 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3423 uhash_entries = UDP_HTABLE_SIZE_MIN; 3424 return 1; 3425 } 3426 __setup("uhash_entries=", set_uhash_entries); 3427 3428 void __init udp_table_init(struct udp_table *table, const char *name) 3429 { 3430 unsigned int i; 3431 3432 table->hash = alloc_large_system_hash(name, 3433 2 * sizeof(struct udp_hslot), 3434 uhash_entries, 3435 21, /* one slot per 2 MB */ 3436 0, 3437 &table->log, 3438 &table->mask, 3439 UDP_HTABLE_SIZE_MIN, 3440 UDP_HTABLE_SIZE_MAX); 3441 3442 table->hash2 = table->hash + (table->mask + 1); 3443 for (i = 0; i <= table->mask; i++) { 3444 INIT_HLIST_HEAD(&table->hash[i].head); 3445 table->hash[i].count = 0; 3446 spin_lock_init(&table->hash[i].lock); 3447 } 3448 for (i = 0; i <= table->mask; i++) { 3449 INIT_HLIST_HEAD(&table->hash2[i].head); 3450 table->hash2[i].count = 0; 3451 spin_lock_init(&table->hash2[i].lock); 3452 } 3453 } 3454 3455 u32 udp_flow_hashrnd(void) 3456 { 3457 static u32 hashrnd __read_mostly; 3458 3459 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3460 3461 return hashrnd; 3462 } 3463 EXPORT_SYMBOL(udp_flow_hashrnd); 3464 3465 static void __net_init udp_sysctl_init(struct net *net) 3466 { 3467 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3468 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 3469 3470 #ifdef CONFIG_NET_L3_MASTER_DEV 3471 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3472 #endif 3473 } 3474 3475 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries) 3476 { 3477 struct udp_table *udptable; 3478 int i; 3479 3480 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL); 3481 if (!udptable) 3482 goto out; 3483 3484 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot), 3485 GFP_KERNEL_ACCOUNT); 3486 if (!udptable->hash) 3487 goto free_table; 3488 3489 udptable->hash2 = udptable->hash + hash_entries; 3490 udptable->mask = hash_entries - 1; 3491 udptable->log = ilog2(hash_entries); 3492 3493 for (i = 0; i < hash_entries; i++) { 3494 INIT_HLIST_HEAD(&udptable->hash[i].head); 3495 udptable->hash[i].count = 0; 3496 spin_lock_init(&udptable->hash[i].lock); 3497 3498 INIT_HLIST_HEAD(&udptable->hash2[i].head); 3499 udptable->hash2[i].count = 0; 3500 spin_lock_init(&udptable->hash2[i].lock); 3501 } 3502 3503 return udptable; 3504 3505 free_table: 3506 kfree(udptable); 3507 out: 3508 return NULL; 3509 } 3510 3511 static void __net_exit udp_pernet_table_free(struct net *net) 3512 { 3513 struct udp_table *udptable = net->ipv4.udp_table; 3514 3515 if (udptable == &udp_table) 3516 return; 3517 3518 kvfree(udptable->hash); 3519 kfree(udptable); 3520 } 3521 3522 static void __net_init udp_set_table(struct net *net) 3523 { 3524 struct udp_table *udptable; 3525 unsigned int hash_entries; 3526 struct net *old_net; 3527 3528 if (net_eq(net, &init_net)) 3529 goto fallback; 3530 3531 old_net = current->nsproxy->net_ns; 3532 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries); 3533 if (!hash_entries) 3534 goto fallback; 3535 3536 /* Set min to keep the bitmap on stack in udp_lib_get_port() */ 3537 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET) 3538 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET; 3539 else 3540 hash_entries = roundup_pow_of_two(hash_entries); 3541 3542 udptable = udp_pernet_table_alloc(hash_entries); 3543 if (udptable) { 3544 net->ipv4.udp_table = udptable; 3545 } else { 3546 pr_warn("Failed to allocate UDP hash table (entries: %u) " 3547 "for a netns, fallback to the global one\n", 3548 hash_entries); 3549 fallback: 3550 net->ipv4.udp_table = &udp_table; 3551 } 3552 } 3553 3554 static int __net_init udp_pernet_init(struct net *net) 3555 { 3556 udp_sysctl_init(net); 3557 udp_set_table(net); 3558 3559 return 0; 3560 } 3561 3562 static void __net_exit udp_pernet_exit(struct net *net) 3563 { 3564 udp_pernet_table_free(net); 3565 } 3566 3567 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3568 .init = udp_pernet_init, 3569 .exit = udp_pernet_exit, 3570 }; 3571 3572 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3573 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3574 struct udp_sock *udp_sk, uid_t uid, int bucket) 3575 3576 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3577 unsigned int new_batch_sz) 3578 { 3579 struct sock **new_batch; 3580 3581 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch), 3582 GFP_USER | __GFP_NOWARN); 3583 if (!new_batch) 3584 return -ENOMEM; 3585 3586 bpf_iter_udp_put_batch(iter); 3587 kvfree(iter->batch); 3588 iter->batch = new_batch; 3589 iter->max_sk = new_batch_sz; 3590 3591 return 0; 3592 } 3593 3594 #define INIT_BATCH_SZ 16 3595 3596 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3597 { 3598 struct bpf_udp_iter_state *iter = priv_data; 3599 int ret; 3600 3601 ret = bpf_iter_init_seq_net(priv_data, aux); 3602 if (ret) 3603 return ret; 3604 3605 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ); 3606 if (ret) 3607 bpf_iter_fini_seq_net(priv_data); 3608 3609 return ret; 3610 } 3611 3612 static void bpf_iter_fini_udp(void *priv_data) 3613 { 3614 struct bpf_udp_iter_state *iter = priv_data; 3615 3616 bpf_iter_fini_seq_net(priv_data); 3617 kvfree(iter->batch); 3618 } 3619 3620 static const struct bpf_iter_seq_info udp_seq_info = { 3621 .seq_ops = &bpf_iter_udp_seq_ops, 3622 .init_seq_private = bpf_iter_init_udp, 3623 .fini_seq_private = bpf_iter_fini_udp, 3624 .seq_priv_size = sizeof(struct bpf_udp_iter_state), 3625 }; 3626 3627 static struct bpf_iter_reg udp_reg_info = { 3628 .target = "udp", 3629 .ctx_arg_info_size = 1, 3630 .ctx_arg_info = { 3631 { offsetof(struct bpf_iter__udp, udp_sk), 3632 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED }, 3633 }, 3634 .seq_info = &udp_seq_info, 3635 }; 3636 3637 static void __init bpf_iter_register(void) 3638 { 3639 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3640 if (bpf_iter_reg_target(&udp_reg_info)) 3641 pr_warn("Warning: could not register bpf iterator udp\n"); 3642 } 3643 #endif 3644 3645 void __init udp_init(void) 3646 { 3647 unsigned long limit; 3648 unsigned int i; 3649 3650 udp_table_init(&udp_table, "UDP"); 3651 limit = nr_free_buffer_pages() / 8; 3652 limit = max(limit, 128UL); 3653 sysctl_udp_mem[0] = limit / 4 * 3; 3654 sysctl_udp_mem[1] = limit; 3655 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3656 3657 /* 16 spinlocks per cpu */ 3658 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3659 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3660 GFP_KERNEL); 3661 if (!udp_busylocks) 3662 panic("UDP: failed to alloc udp_busylocks\n"); 3663 for (i = 0; i < (1U << udp_busylocks_log); i++) 3664 spin_lock_init(udp_busylocks + i); 3665 3666 if (register_pernet_subsys(&udp_sysctl_ops)) 3667 panic("UDP: failed to init sysctl parameters.\n"); 3668 3669 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3670 bpf_iter_register(); 3671 #endif 3672 } 3673
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.