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