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 #if IS_ENABLED(CONFIG_IPV6) 118 #include <net/ipv6_stubs.h> 119 #endif 120 121 struct udp_table udp_table __read_mostly; 122 EXPORT_SYMBOL(udp_table); 123 124 long sysctl_udp_mem[3] __read_mostly; 125 EXPORT_SYMBOL(sysctl_udp_mem); 126 127 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp; 128 EXPORT_SYMBOL(udp_memory_allocated); 129 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc); 130 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc); 131 132 #define MAX_UDP_PORTS 65536 133 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET) 134 135 static struct udp_table *udp_get_table_prot(struct sock *sk) 136 { 137 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table; 138 } 139 140 static int udp_lib_lport_inuse(struct net *net, __u16 num, 141 const struct udp_hslot *hslot, 142 unsigned long *bitmap, 143 struct sock *sk, unsigned int log) 144 { 145 struct sock *sk2; 146 kuid_t uid = sock_i_uid(sk); 147 148 sk_for_each(sk2, &hslot->head) { 149 if (net_eq(sock_net(sk2), net) && 150 sk2 != sk && 151 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 152 (!sk2->sk_reuse || !sk->sk_reuse) && 153 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 154 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 155 inet_rcv_saddr_equal(sk, sk2, true)) { 156 if (sk2->sk_reuseport && sk->sk_reuseport && 157 !rcu_access_pointer(sk->sk_reuseport_cb) && 158 uid_eq(uid, sock_i_uid(sk2))) { 159 if (!bitmap) 160 return 0; 161 } else { 162 if (!bitmap) 163 return 1; 164 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 165 bitmap); 166 } 167 } 168 } 169 return 0; 170 } 171 172 /* 173 * Note: we still hold spinlock of primary hash chain, so no other writer 174 * can insert/delete a socket with local_port == num 175 */ 176 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 177 struct udp_hslot *hslot2, 178 struct sock *sk) 179 { 180 struct sock *sk2; 181 kuid_t uid = sock_i_uid(sk); 182 int res = 0; 183 184 spin_lock(&hslot2->lock); 185 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 186 if (net_eq(sock_net(sk2), net) && 187 sk2 != sk && 188 (udp_sk(sk2)->udp_port_hash == num) && 189 (!sk2->sk_reuse || !sk->sk_reuse) && 190 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 191 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 192 inet_rcv_saddr_equal(sk, sk2, true)) { 193 if (sk2->sk_reuseport && sk->sk_reuseport && 194 !rcu_access_pointer(sk->sk_reuseport_cb) && 195 uid_eq(uid, sock_i_uid(sk2))) { 196 res = 0; 197 } else { 198 res = 1; 199 } 200 break; 201 } 202 } 203 spin_unlock(&hslot2->lock); 204 return res; 205 } 206 207 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 208 { 209 struct net *net = sock_net(sk); 210 kuid_t uid = sock_i_uid(sk); 211 struct sock *sk2; 212 213 sk_for_each(sk2, &hslot->head) { 214 if (net_eq(sock_net(sk2), net) && 215 sk2 != sk && 216 sk2->sk_family == sk->sk_family && 217 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 218 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 219 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 220 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 221 inet_rcv_saddr_equal(sk, sk2, false)) { 222 return reuseport_add_sock(sk, sk2, 223 inet_rcv_saddr_any(sk)); 224 } 225 } 226 227 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 228 } 229 230 /** 231 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 232 * 233 * @sk: socket struct in question 234 * @snum: port number to look up 235 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 236 * with NULL address 237 */ 238 int udp_lib_get_port(struct sock *sk, unsigned short snum, 239 unsigned int hash2_nulladdr) 240 { 241 struct udp_table *udptable = udp_get_table_prot(sk); 242 struct udp_hslot *hslot, *hslot2; 243 struct net *net = sock_net(sk); 244 int error = -EADDRINUSE; 245 246 if (!snum) { 247 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 248 unsigned short first, last; 249 int low, high, remaining; 250 unsigned int rand; 251 252 inet_sk_get_local_port_range(sk, &low, &high); 253 remaining = (high - low) + 1; 254 255 rand = get_random_u32(); 256 first = reciprocal_scale(rand, remaining) + low; 257 /* 258 * force rand to be an odd multiple of UDP_HTABLE_SIZE 259 */ 260 rand = (rand | 1) * (udptable->mask + 1); 261 last = first + udptable->mask + 1; 262 do { 263 hslot = udp_hashslot(udptable, net, first); 264 bitmap_zero(bitmap, PORTS_PER_CHAIN); 265 spin_lock_bh(&hslot->lock); 266 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 267 udptable->log); 268 269 snum = first; 270 /* 271 * Iterate on all possible values of snum for this hash. 272 * Using steps of an odd multiple of UDP_HTABLE_SIZE 273 * give us randomization and full range coverage. 274 */ 275 do { 276 if (low <= snum && snum <= high && 277 !test_bit(snum >> udptable->log, bitmap) && 278 !inet_is_local_reserved_port(net, snum)) 279 goto found; 280 snum += rand; 281 } while (snum != first); 282 spin_unlock_bh(&hslot->lock); 283 cond_resched(); 284 } while (++first != last); 285 goto fail; 286 } else { 287 hslot = udp_hashslot(udptable, net, snum); 288 spin_lock_bh(&hslot->lock); 289 if (hslot->count > 10) { 290 int exist; 291 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 292 293 slot2 &= udptable->mask; 294 hash2_nulladdr &= udptable->mask; 295 296 hslot2 = udp_hashslot2(udptable, slot2); 297 if (hslot->count < hslot2->count) 298 goto scan_primary_hash; 299 300 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 301 if (!exist && (hash2_nulladdr != slot2)) { 302 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 303 exist = udp_lib_lport_inuse2(net, snum, hslot2, 304 sk); 305 } 306 if (exist) 307 goto fail_unlock; 308 else 309 goto found; 310 } 311 scan_primary_hash: 312 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 313 goto fail_unlock; 314 } 315 found: 316 inet_sk(sk)->inet_num = snum; 317 udp_sk(sk)->udp_port_hash = snum; 318 udp_sk(sk)->udp_portaddr_hash ^= snum; 319 if (sk_unhashed(sk)) { 320 if (sk->sk_reuseport && 321 udp_reuseport_add_sock(sk, hslot)) { 322 inet_sk(sk)->inet_num = 0; 323 udp_sk(sk)->udp_port_hash = 0; 324 udp_sk(sk)->udp_portaddr_hash ^= snum; 325 goto fail_unlock; 326 } 327 328 sk_add_node_rcu(sk, &hslot->head); 329 hslot->count++; 330 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 331 332 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 333 spin_lock(&hslot2->lock); 334 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 335 sk->sk_family == AF_INET6) 336 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 337 &hslot2->head); 338 else 339 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 340 &hslot2->head); 341 hslot2->count++; 342 spin_unlock(&hslot2->lock); 343 } 344 sock_set_flag(sk, SOCK_RCU_FREE); 345 error = 0; 346 fail_unlock: 347 spin_unlock_bh(&hslot->lock); 348 fail: 349 return error; 350 } 351 EXPORT_SYMBOL(udp_lib_get_port); 352 353 int udp_v4_get_port(struct sock *sk, unsigned short snum) 354 { 355 unsigned int hash2_nulladdr = 356 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 357 unsigned int hash2_partial = 358 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 359 360 /* precompute partial secondary hash */ 361 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 362 return udp_lib_get_port(sk, snum, hash2_nulladdr); 363 } 364 365 static int compute_score(struct sock *sk, struct net *net, 366 __be32 saddr, __be16 sport, 367 __be32 daddr, unsigned short hnum, 368 int dif, int sdif) 369 { 370 int score; 371 struct inet_sock *inet; 372 bool dev_match; 373 374 if (!net_eq(sock_net(sk), net) || 375 udp_sk(sk)->udp_port_hash != hnum || 376 ipv6_only_sock(sk)) 377 return -1; 378 379 if (sk->sk_rcv_saddr != daddr) 380 return -1; 381 382 score = (sk->sk_family == PF_INET) ? 2 : 1; 383 384 inet = inet_sk(sk); 385 if (inet->inet_daddr) { 386 if (inet->inet_daddr != saddr) 387 return -1; 388 score += 4; 389 } 390 391 if (inet->inet_dport) { 392 if (inet->inet_dport != sport) 393 return -1; 394 score += 4; 395 } 396 397 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 398 dif, sdif); 399 if (!dev_match) 400 return -1; 401 if (sk->sk_bound_dev_if) 402 score += 4; 403 404 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 405 score++; 406 return score; 407 } 408 409 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 410 const __u16 lport, const __be32 faddr, 411 const __be16 fport) 412 { 413 static u32 udp_ehash_secret __read_mostly; 414 415 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 416 417 return __inet_ehashfn(laddr, lport, faddr, fport, 418 udp_ehash_secret + net_hash_mix(net)); 419 } 420 421 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 422 struct sk_buff *skb, 423 __be32 saddr, __be16 sport, 424 __be32 daddr, unsigned short hnum) 425 { 426 struct sock *reuse_sk = NULL; 427 u32 hash; 428 429 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 430 hash = udp_ehashfn(net, daddr, hnum, saddr, sport); 431 reuse_sk = reuseport_select_sock(sk, hash, skb, 432 sizeof(struct udphdr)); 433 } 434 return reuse_sk; 435 } 436 437 /* called with rcu_read_lock() */ 438 static struct sock *udp4_lib_lookup2(struct net *net, 439 __be32 saddr, __be16 sport, 440 __be32 daddr, unsigned int hnum, 441 int dif, int sdif, 442 struct udp_hslot *hslot2, 443 struct sk_buff *skb) 444 { 445 struct sock *sk, *result; 446 int score, badness; 447 448 result = NULL; 449 badness = 0; 450 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 451 score = compute_score(sk, net, saddr, sport, 452 daddr, hnum, dif, sdif); 453 if (score > badness) { 454 result = lookup_reuseport(net, sk, skb, 455 saddr, sport, daddr, hnum); 456 /* Fall back to scoring if group has connections */ 457 if (result && !reuseport_has_conns(sk)) 458 return result; 459 460 result = result ? : sk; 461 badness = score; 462 } 463 } 464 return result; 465 } 466 467 static struct sock *udp4_lookup_run_bpf(struct net *net, 468 struct udp_table *udptable, 469 struct sk_buff *skb, 470 __be32 saddr, __be16 sport, 471 __be32 daddr, u16 hnum, const int dif) 472 { 473 struct sock *sk, *reuse_sk; 474 bool no_reuseport; 475 476 if (udptable != net->ipv4.udp_table) 477 return NULL; /* only UDP is supported */ 478 479 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport, 480 daddr, hnum, dif, &sk); 481 if (no_reuseport || IS_ERR_OR_NULL(sk)) 482 return sk; 483 484 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 485 if (reuse_sk) 486 sk = reuse_sk; 487 return sk; 488 } 489 490 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 491 * harder than this. -DaveM 492 */ 493 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 494 __be16 sport, __be32 daddr, __be16 dport, int dif, 495 int sdif, struct udp_table *udptable, struct sk_buff *skb) 496 { 497 unsigned short hnum = ntohs(dport); 498 unsigned int hash2, slot2; 499 struct udp_hslot *hslot2; 500 struct sock *result, *sk; 501 502 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 503 slot2 = hash2 & udptable->mask; 504 hslot2 = &udptable->hash2[slot2]; 505 506 /* Lookup connected or non-wildcard socket */ 507 result = udp4_lib_lookup2(net, saddr, sport, 508 daddr, hnum, dif, sdif, 509 hslot2, skb); 510 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 511 goto done; 512 513 /* Lookup redirect from BPF */ 514 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { 515 sk = udp4_lookup_run_bpf(net, udptable, skb, 516 saddr, sport, daddr, hnum, dif); 517 if (sk) { 518 result = sk; 519 goto done; 520 } 521 } 522 523 /* Got non-wildcard socket or error on first lookup */ 524 if (result) 525 goto done; 526 527 /* Lookup wildcard sockets */ 528 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 529 slot2 = hash2 & udptable->mask; 530 hslot2 = &udptable->hash2[slot2]; 531 532 result = udp4_lib_lookup2(net, saddr, sport, 533 htonl(INADDR_ANY), hnum, dif, sdif, 534 hslot2, skb); 535 done: 536 if (IS_ERR(result)) 537 return NULL; 538 return result; 539 } 540 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 541 542 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 543 __be16 sport, __be16 dport, 544 struct udp_table *udptable) 545 { 546 const struct iphdr *iph = ip_hdr(skb); 547 548 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 549 iph->daddr, dport, inet_iif(skb), 550 inet_sdif(skb), udptable, skb); 551 } 552 553 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 554 __be16 sport, __be16 dport) 555 { 556 const struct iphdr *iph = ip_hdr(skb); 557 struct net *net = dev_net(skb->dev); 558 559 return __udp4_lib_lookup(net, iph->saddr, sport, 560 iph->daddr, dport, inet_iif(skb), 561 inet_sdif(skb), net->ipv4.udp_table, NULL); 562 } 563 564 /* Must be called under rcu_read_lock(). 565 * Does increment socket refcount. 566 */ 567 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 568 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 569 __be32 daddr, __be16 dport, int dif) 570 { 571 struct sock *sk; 572 573 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 574 dif, 0, net->ipv4.udp_table, NULL); 575 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 576 sk = NULL; 577 return sk; 578 } 579 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 580 #endif 581 582 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk, 583 __be16 loc_port, __be32 loc_addr, 584 __be16 rmt_port, __be32 rmt_addr, 585 int dif, int sdif, unsigned short hnum) 586 { 587 const struct inet_sock *inet = inet_sk(sk); 588 589 if (!net_eq(sock_net(sk), net) || 590 udp_sk(sk)->udp_port_hash != hnum || 591 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 592 (inet->inet_dport != rmt_port && inet->inet_dport) || 593 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 594 ipv6_only_sock(sk) || 595 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 596 return false; 597 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 598 return false; 599 return true; 600 } 601 602 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 603 void udp_encap_enable(void) 604 { 605 static_branch_inc(&udp_encap_needed_key); 606 } 607 EXPORT_SYMBOL(udp_encap_enable); 608 609 void udp_encap_disable(void) 610 { 611 static_branch_dec(&udp_encap_needed_key); 612 } 613 EXPORT_SYMBOL(udp_encap_disable); 614 615 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 616 * through error handlers in encapsulations looking for a match. 617 */ 618 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 619 { 620 int i; 621 622 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 623 int (*handler)(struct sk_buff *skb, u32 info); 624 const struct ip_tunnel_encap_ops *encap; 625 626 encap = rcu_dereference(iptun_encaps[i]); 627 if (!encap) 628 continue; 629 handler = encap->err_handler; 630 if (handler && !handler(skb, info)) 631 return 0; 632 } 633 634 return -ENOENT; 635 } 636 637 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 638 * reversing source and destination port: this will match tunnels that force the 639 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 640 * lwtunnels might actually break this assumption by being configured with 641 * different destination ports on endpoints, in this case we won't be able to 642 * trace ICMP messages back to them. 643 * 644 * If this doesn't match any socket, probe tunnels with arbitrary destination 645 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 646 * we've sent packets to won't necessarily match the local destination port. 647 * 648 * Then ask the tunnel implementation to match the error against a valid 649 * association. 650 * 651 * Return an error if we can't find a match, the socket if we need further 652 * processing, zero otherwise. 653 */ 654 static struct sock *__udp4_lib_err_encap(struct net *net, 655 const struct iphdr *iph, 656 struct udphdr *uh, 657 struct udp_table *udptable, 658 struct sock *sk, 659 struct sk_buff *skb, u32 info) 660 { 661 int (*lookup)(struct sock *sk, struct sk_buff *skb); 662 int network_offset, transport_offset; 663 struct udp_sock *up; 664 665 network_offset = skb_network_offset(skb); 666 transport_offset = skb_transport_offset(skb); 667 668 /* Network header needs to point to the outer IPv4 header inside ICMP */ 669 skb_reset_network_header(skb); 670 671 /* Transport header needs to point to the UDP header */ 672 skb_set_transport_header(skb, iph->ihl << 2); 673 674 if (sk) { 675 up = udp_sk(sk); 676 677 lookup = READ_ONCE(up->encap_err_lookup); 678 if (lookup && lookup(sk, skb)) 679 sk = NULL; 680 681 goto out; 682 } 683 684 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 685 iph->saddr, uh->dest, skb->dev->ifindex, 0, 686 udptable, NULL); 687 if (sk) { 688 up = udp_sk(sk); 689 690 lookup = READ_ONCE(up->encap_err_lookup); 691 if (!lookup || lookup(sk, skb)) 692 sk = NULL; 693 } 694 695 out: 696 if (!sk) 697 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 698 699 skb_set_transport_header(skb, transport_offset); 700 skb_set_network_header(skb, network_offset); 701 702 return sk; 703 } 704 705 /* 706 * This routine is called by the ICMP module when it gets some 707 * sort of error condition. If err < 0 then the socket should 708 * be closed and the error returned to the user. If err > 0 709 * it's just the icmp type << 8 | icmp code. 710 * Header points to the ip header of the error packet. We move 711 * on past this. Then (as it used to claim before adjustment) 712 * header points to the first 8 bytes of the udp header. We need 713 * to find the appropriate port. 714 */ 715 716 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 717 { 718 struct inet_sock *inet; 719 const struct iphdr *iph = (const struct iphdr *)skb->data; 720 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 721 const int type = icmp_hdr(skb)->type; 722 const int code = icmp_hdr(skb)->code; 723 bool tunnel = false; 724 struct sock *sk; 725 int harderr; 726 int err; 727 struct net *net = dev_net(skb->dev); 728 729 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 730 iph->saddr, uh->source, skb->dev->ifindex, 731 inet_sdif(skb), udptable, NULL); 732 733 if (!sk || udp_sk(sk)->encap_type) { 734 /* No socket for error: try tunnels before discarding */ 735 if (static_branch_unlikely(&udp_encap_needed_key)) { 736 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 737 info); 738 if (!sk) 739 return 0; 740 } else 741 sk = ERR_PTR(-ENOENT); 742 743 if (IS_ERR(sk)) { 744 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 745 return PTR_ERR(sk); 746 } 747 748 tunnel = true; 749 } 750 751 err = 0; 752 harderr = 0; 753 inet = inet_sk(sk); 754 755 switch (type) { 756 default: 757 case ICMP_TIME_EXCEEDED: 758 err = EHOSTUNREACH; 759 break; 760 case ICMP_SOURCE_QUENCH: 761 goto out; 762 case ICMP_PARAMETERPROB: 763 err = EPROTO; 764 harderr = 1; 765 break; 766 case ICMP_DEST_UNREACH: 767 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 768 ipv4_sk_update_pmtu(skb, sk, info); 769 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 770 err = EMSGSIZE; 771 harderr = 1; 772 break; 773 } 774 goto out; 775 } 776 err = EHOSTUNREACH; 777 if (code <= NR_ICMP_UNREACH) { 778 harderr = icmp_err_convert[code].fatal; 779 err = icmp_err_convert[code].errno; 780 } 781 break; 782 case ICMP_REDIRECT: 783 ipv4_sk_redirect(skb, sk); 784 goto out; 785 } 786 787 /* 788 * RFC1122: OK. Passes ICMP errors back to application, as per 789 * 4.1.3.3. 790 */ 791 if (tunnel) { 792 /* ...not for tunnels though: we don't have a sending socket */ 793 if (udp_sk(sk)->encap_err_rcv) 794 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info, 795 (u8 *)(uh+1)); 796 goto out; 797 } 798 if (!inet->recverr) { 799 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 800 goto out; 801 } else 802 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 803 804 sk->sk_err = err; 805 sk_error_report(sk); 806 out: 807 return 0; 808 } 809 810 int udp_err(struct sk_buff *skb, u32 info) 811 { 812 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table); 813 } 814 815 /* 816 * Throw away all pending data and cancel the corking. Socket is locked. 817 */ 818 void udp_flush_pending_frames(struct sock *sk) 819 { 820 struct udp_sock *up = udp_sk(sk); 821 822 if (up->pending) { 823 up->len = 0; 824 up->pending = 0; 825 ip_flush_pending_frames(sk); 826 } 827 } 828 EXPORT_SYMBOL(udp_flush_pending_frames); 829 830 /** 831 * udp4_hwcsum - handle outgoing HW checksumming 832 * @skb: sk_buff containing the filled-in UDP header 833 * (checksum field must be zeroed out) 834 * @src: source IP address 835 * @dst: destination IP address 836 */ 837 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 838 { 839 struct udphdr *uh = udp_hdr(skb); 840 int offset = skb_transport_offset(skb); 841 int len = skb->len - offset; 842 int hlen = len; 843 __wsum csum = 0; 844 845 if (!skb_has_frag_list(skb)) { 846 /* 847 * Only one fragment on the socket. 848 */ 849 skb->csum_start = skb_transport_header(skb) - skb->head; 850 skb->csum_offset = offsetof(struct udphdr, check); 851 uh->check = ~csum_tcpudp_magic(src, dst, len, 852 IPPROTO_UDP, 0); 853 } else { 854 struct sk_buff *frags; 855 856 /* 857 * HW-checksum won't work as there are two or more 858 * fragments on the socket so that all csums of sk_buffs 859 * should be together 860 */ 861 skb_walk_frags(skb, frags) { 862 csum = csum_add(csum, frags->csum); 863 hlen -= frags->len; 864 } 865 866 csum = skb_checksum(skb, offset, hlen, csum); 867 skb->ip_summed = CHECKSUM_NONE; 868 869 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 870 if (uh->check == 0) 871 uh->check = CSUM_MANGLED_0; 872 } 873 } 874 EXPORT_SYMBOL_GPL(udp4_hwcsum); 875 876 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 877 * for the simple case like when setting the checksum for a UDP tunnel. 878 */ 879 void udp_set_csum(bool nocheck, struct sk_buff *skb, 880 __be32 saddr, __be32 daddr, int len) 881 { 882 struct udphdr *uh = udp_hdr(skb); 883 884 if (nocheck) { 885 uh->check = 0; 886 } else if (skb_is_gso(skb)) { 887 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 888 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 889 uh->check = 0; 890 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 891 if (uh->check == 0) 892 uh->check = CSUM_MANGLED_0; 893 } else { 894 skb->ip_summed = CHECKSUM_PARTIAL; 895 skb->csum_start = skb_transport_header(skb) - skb->head; 896 skb->csum_offset = offsetof(struct udphdr, check); 897 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 898 } 899 } 900 EXPORT_SYMBOL(udp_set_csum); 901 902 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 903 struct inet_cork *cork) 904 { 905 struct sock *sk = skb->sk; 906 struct inet_sock *inet = inet_sk(sk); 907 struct udphdr *uh; 908 int err; 909 int is_udplite = IS_UDPLITE(sk); 910 int offset = skb_transport_offset(skb); 911 int len = skb->len - offset; 912 int datalen = len - sizeof(*uh); 913 __wsum csum = 0; 914 915 /* 916 * Create a UDP header 917 */ 918 uh = udp_hdr(skb); 919 uh->source = inet->inet_sport; 920 uh->dest = fl4->fl4_dport; 921 uh->len = htons(len); 922 uh->check = 0; 923 924 if (cork->gso_size) { 925 const int hlen = skb_network_header_len(skb) + 926 sizeof(struct udphdr); 927 928 if (hlen + cork->gso_size > cork->fragsize) { 929 kfree_skb(skb); 930 return -EINVAL; 931 } 932 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 933 kfree_skb(skb); 934 return -EINVAL; 935 } 936 if (sk->sk_no_check_tx) { 937 kfree_skb(skb); 938 return -EINVAL; 939 } 940 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 941 dst_xfrm(skb_dst(skb))) { 942 kfree_skb(skb); 943 return -EIO; 944 } 945 946 if (datalen > cork->gso_size) { 947 skb_shinfo(skb)->gso_size = cork->gso_size; 948 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 949 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 950 cork->gso_size); 951 } 952 goto csum_partial; 953 } 954 955 if (is_udplite) /* UDP-Lite */ 956 csum = udplite_csum(skb); 957 958 else if (sk->sk_no_check_tx) { /* UDP csum off */ 959 960 skb->ip_summed = CHECKSUM_NONE; 961 goto send; 962 963 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 964 csum_partial: 965 966 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 967 goto send; 968 969 } else 970 csum = udp_csum(skb); 971 972 /* add protocol-dependent pseudo-header */ 973 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 974 sk->sk_protocol, csum); 975 if (uh->check == 0) 976 uh->check = CSUM_MANGLED_0; 977 978 send: 979 err = ip_send_skb(sock_net(sk), skb); 980 if (err) { 981 if (err == -ENOBUFS && !inet->recverr) { 982 UDP_INC_STATS(sock_net(sk), 983 UDP_MIB_SNDBUFERRORS, is_udplite); 984 err = 0; 985 } 986 } else 987 UDP_INC_STATS(sock_net(sk), 988 UDP_MIB_OUTDATAGRAMS, is_udplite); 989 return err; 990 } 991 992 /* 993 * Push out all pending data as one UDP datagram. Socket is locked. 994 */ 995 int udp_push_pending_frames(struct sock *sk) 996 { 997 struct udp_sock *up = udp_sk(sk); 998 struct inet_sock *inet = inet_sk(sk); 999 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 1000 struct sk_buff *skb; 1001 int err = 0; 1002 1003 skb = ip_finish_skb(sk, fl4); 1004 if (!skb) 1005 goto out; 1006 1007 err = udp_send_skb(skb, fl4, &inet->cork.base); 1008 1009 out: 1010 up->len = 0; 1011 up->pending = 0; 1012 return err; 1013 } 1014 EXPORT_SYMBOL(udp_push_pending_frames); 1015 1016 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1017 { 1018 switch (cmsg->cmsg_type) { 1019 case UDP_SEGMENT: 1020 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1021 return -EINVAL; 1022 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1023 return 0; 1024 default: 1025 return -EINVAL; 1026 } 1027 } 1028 1029 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1030 { 1031 struct cmsghdr *cmsg; 1032 bool need_ip = false; 1033 int err; 1034 1035 for_each_cmsghdr(cmsg, msg) { 1036 if (!CMSG_OK(msg, cmsg)) 1037 return -EINVAL; 1038 1039 if (cmsg->cmsg_level != SOL_UDP) { 1040 need_ip = true; 1041 continue; 1042 } 1043 1044 err = __udp_cmsg_send(cmsg, gso_size); 1045 if (err) 1046 return err; 1047 } 1048 1049 return need_ip; 1050 } 1051 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1052 1053 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1054 { 1055 struct inet_sock *inet = inet_sk(sk); 1056 struct udp_sock *up = udp_sk(sk); 1057 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1058 struct flowi4 fl4_stack; 1059 struct flowi4 *fl4; 1060 int ulen = len; 1061 struct ipcm_cookie ipc; 1062 struct rtable *rt = NULL; 1063 int free = 0; 1064 int connected = 0; 1065 __be32 daddr, faddr, saddr; 1066 u8 tos, scope; 1067 __be16 dport; 1068 int err, is_udplite = IS_UDPLITE(sk); 1069 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1070 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1071 struct sk_buff *skb; 1072 struct ip_options_data opt_copy; 1073 1074 if (len > 0xFFFF) 1075 return -EMSGSIZE; 1076 1077 /* 1078 * Check the flags. 1079 */ 1080 1081 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1082 return -EOPNOTSUPP; 1083 1084 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1085 1086 fl4 = &inet->cork.fl.u.ip4; 1087 if (up->pending) { 1088 /* 1089 * There are pending frames. 1090 * The socket lock must be held while it's corked. 1091 */ 1092 lock_sock(sk); 1093 if (likely(up->pending)) { 1094 if (unlikely(up->pending != AF_INET)) { 1095 release_sock(sk); 1096 return -EINVAL; 1097 } 1098 goto do_append_data; 1099 } 1100 release_sock(sk); 1101 } 1102 ulen += sizeof(struct udphdr); 1103 1104 /* 1105 * Get and verify the address. 1106 */ 1107 if (usin) { 1108 if (msg->msg_namelen < sizeof(*usin)) 1109 return -EINVAL; 1110 if (usin->sin_family != AF_INET) { 1111 if (usin->sin_family != AF_UNSPEC) 1112 return -EAFNOSUPPORT; 1113 } 1114 1115 daddr = usin->sin_addr.s_addr; 1116 dport = usin->sin_port; 1117 if (dport == 0) 1118 return -EINVAL; 1119 } else { 1120 if (sk->sk_state != TCP_ESTABLISHED) 1121 return -EDESTADDRREQ; 1122 daddr = inet->inet_daddr; 1123 dport = inet->inet_dport; 1124 /* Open fast path for connected socket. 1125 Route will not be used, if at least one option is set. 1126 */ 1127 connected = 1; 1128 } 1129 1130 ipcm_init_sk(&ipc, inet); 1131 ipc.gso_size = READ_ONCE(up->gso_size); 1132 1133 if (msg->msg_controllen) { 1134 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1135 if (err > 0) 1136 err = ip_cmsg_send(sk, msg, &ipc, 1137 sk->sk_family == AF_INET6); 1138 if (unlikely(err < 0)) { 1139 kfree(ipc.opt); 1140 return err; 1141 } 1142 if (ipc.opt) 1143 free = 1; 1144 connected = 0; 1145 } 1146 if (!ipc.opt) { 1147 struct ip_options_rcu *inet_opt; 1148 1149 rcu_read_lock(); 1150 inet_opt = rcu_dereference(inet->inet_opt); 1151 if (inet_opt) { 1152 memcpy(&opt_copy, inet_opt, 1153 sizeof(*inet_opt) + inet_opt->opt.optlen); 1154 ipc.opt = &opt_copy.opt; 1155 } 1156 rcu_read_unlock(); 1157 } 1158 1159 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1160 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1161 (struct sockaddr *)usin, &ipc.addr); 1162 if (err) 1163 goto out_free; 1164 if (usin) { 1165 if (usin->sin_port == 0) { 1166 /* BPF program set invalid port. Reject it. */ 1167 err = -EINVAL; 1168 goto out_free; 1169 } 1170 daddr = usin->sin_addr.s_addr; 1171 dport = usin->sin_port; 1172 } 1173 } 1174 1175 saddr = ipc.addr; 1176 ipc.addr = faddr = daddr; 1177 1178 if (ipc.opt && ipc.opt->opt.srr) { 1179 if (!daddr) { 1180 err = -EINVAL; 1181 goto out_free; 1182 } 1183 faddr = ipc.opt->opt.faddr; 1184 connected = 0; 1185 } 1186 tos = get_rttos(&ipc, inet); 1187 scope = ip_sendmsg_scope(inet, &ipc, msg); 1188 if (scope == RT_SCOPE_LINK) 1189 connected = 0; 1190 1191 if (ipv4_is_multicast(daddr)) { 1192 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1193 ipc.oif = inet->mc_index; 1194 if (!saddr) 1195 saddr = inet->mc_addr; 1196 connected = 0; 1197 } else if (!ipc.oif) { 1198 ipc.oif = inet->uc_index; 1199 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1200 /* oif is set, packet is to local broadcast and 1201 * uc_index is set. oif is most likely set 1202 * by sk_bound_dev_if. If uc_index != oif check if the 1203 * oif is an L3 master and uc_index is an L3 slave. 1204 * If so, we want to allow the send using the uc_index. 1205 */ 1206 if (ipc.oif != inet->uc_index && 1207 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1208 inet->uc_index)) { 1209 ipc.oif = inet->uc_index; 1210 } 1211 } 1212 1213 if (connected) 1214 rt = (struct rtable *)sk_dst_check(sk, 0); 1215 1216 if (!rt) { 1217 struct net *net = sock_net(sk); 1218 __u8 flow_flags = inet_sk_flowi_flags(sk); 1219 1220 fl4 = &fl4_stack; 1221 1222 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope, 1223 sk->sk_protocol, flow_flags, faddr, saddr, 1224 dport, inet->inet_sport, sk->sk_uid); 1225 1226 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1227 rt = ip_route_output_flow(net, fl4, sk); 1228 if (IS_ERR(rt)) { 1229 err = PTR_ERR(rt); 1230 rt = NULL; 1231 if (err == -ENETUNREACH) 1232 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1233 goto out; 1234 } 1235 1236 err = -EACCES; 1237 if ((rt->rt_flags & RTCF_BROADCAST) && 1238 !sock_flag(sk, SOCK_BROADCAST)) 1239 goto out; 1240 if (connected) 1241 sk_dst_set(sk, dst_clone(&rt->dst)); 1242 } 1243 1244 if (msg->msg_flags&MSG_CONFIRM) 1245 goto do_confirm; 1246 back_from_confirm: 1247 1248 saddr = fl4->saddr; 1249 if (!ipc.addr) 1250 daddr = ipc.addr = fl4->daddr; 1251 1252 /* Lockless fast path for the non-corking case. */ 1253 if (!corkreq) { 1254 struct inet_cork cork; 1255 1256 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1257 sizeof(struct udphdr), &ipc, &rt, 1258 &cork, msg->msg_flags); 1259 err = PTR_ERR(skb); 1260 if (!IS_ERR_OR_NULL(skb)) 1261 err = udp_send_skb(skb, fl4, &cork); 1262 goto out; 1263 } 1264 1265 lock_sock(sk); 1266 if (unlikely(up->pending)) { 1267 /* The socket is already corked while preparing it. */ 1268 /* ... which is an evident application bug. --ANK */ 1269 release_sock(sk); 1270 1271 net_dbg_ratelimited("socket already corked\n"); 1272 err = -EINVAL; 1273 goto out; 1274 } 1275 /* 1276 * Now cork the socket to pend data. 1277 */ 1278 fl4 = &inet->cork.fl.u.ip4; 1279 fl4->daddr = daddr; 1280 fl4->saddr = saddr; 1281 fl4->fl4_dport = dport; 1282 fl4->fl4_sport = inet->inet_sport; 1283 up->pending = AF_INET; 1284 1285 do_append_data: 1286 up->len += ulen; 1287 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1288 sizeof(struct udphdr), &ipc, &rt, 1289 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1290 if (err) 1291 udp_flush_pending_frames(sk); 1292 else if (!corkreq) 1293 err = udp_push_pending_frames(sk); 1294 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1295 up->pending = 0; 1296 release_sock(sk); 1297 1298 out: 1299 ip_rt_put(rt); 1300 out_free: 1301 if (free) 1302 kfree(ipc.opt); 1303 if (!err) 1304 return len; 1305 /* 1306 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1307 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1308 * we don't have a good statistic (IpOutDiscards but it can be too many 1309 * things). We could add another new stat but at least for now that 1310 * seems like overkill. 1311 */ 1312 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1313 UDP_INC_STATS(sock_net(sk), 1314 UDP_MIB_SNDBUFERRORS, is_udplite); 1315 } 1316 return err; 1317 1318 do_confirm: 1319 if (msg->msg_flags & MSG_PROBE) 1320 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1321 if (!(msg->msg_flags&MSG_PROBE) || len) 1322 goto back_from_confirm; 1323 err = 0; 1324 goto out; 1325 } 1326 EXPORT_SYMBOL(udp_sendmsg); 1327 1328 void udp_splice_eof(struct socket *sock) 1329 { 1330 struct sock *sk = sock->sk; 1331 struct udp_sock *up = udp_sk(sk); 1332 1333 if (!up->pending || READ_ONCE(up->corkflag)) 1334 return; 1335 1336 lock_sock(sk); 1337 if (up->pending && !READ_ONCE(up->corkflag)) 1338 udp_push_pending_frames(sk); 1339 release_sock(sk); 1340 } 1341 EXPORT_SYMBOL_GPL(udp_splice_eof); 1342 1343 #define UDP_SKB_IS_STATELESS 0x80000000 1344 1345 /* all head states (dst, sk, nf conntrack) except skb extensions are 1346 * cleared by udp_rcv(). 1347 * 1348 * We need to preserve secpath, if present, to eventually process 1349 * IP_CMSG_PASSSEC at recvmsg() time. 1350 * 1351 * Other extensions can be cleared. 1352 */ 1353 static bool udp_try_make_stateless(struct sk_buff *skb) 1354 { 1355 if (!skb_has_extensions(skb)) 1356 return true; 1357 1358 if (!secpath_exists(skb)) { 1359 skb_ext_reset(skb); 1360 return true; 1361 } 1362 1363 return false; 1364 } 1365 1366 static void udp_set_dev_scratch(struct sk_buff *skb) 1367 { 1368 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1369 1370 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1371 scratch->_tsize_state = skb->truesize; 1372 #if BITS_PER_LONG == 64 1373 scratch->len = skb->len; 1374 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1375 scratch->is_linear = !skb_is_nonlinear(skb); 1376 #endif 1377 if (udp_try_make_stateless(skb)) 1378 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1379 } 1380 1381 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1382 { 1383 /* We come here after udp_lib_checksum_complete() returned 0. 1384 * This means that __skb_checksum_complete() might have 1385 * set skb->csum_valid to 1. 1386 * On 64bit platforms, we can set csum_unnecessary 1387 * to true, but only if the skb is not shared. 1388 */ 1389 #if BITS_PER_LONG == 64 1390 if (!skb_shared(skb)) 1391 udp_skb_scratch(skb)->csum_unnecessary = true; 1392 #endif 1393 } 1394 1395 static int udp_skb_truesize(struct sk_buff *skb) 1396 { 1397 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1398 } 1399 1400 static bool udp_skb_has_head_state(struct sk_buff *skb) 1401 { 1402 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1403 } 1404 1405 /* fully reclaim rmem/fwd memory allocated for skb */ 1406 static void udp_rmem_release(struct sock *sk, int size, int partial, 1407 bool rx_queue_lock_held) 1408 { 1409 struct udp_sock *up = udp_sk(sk); 1410 struct sk_buff_head *sk_queue; 1411 int amt; 1412 1413 if (likely(partial)) { 1414 up->forward_deficit += size; 1415 size = up->forward_deficit; 1416 if (size < READ_ONCE(up->forward_threshold) && 1417 !skb_queue_empty(&up->reader_queue)) 1418 return; 1419 } else { 1420 size += up->forward_deficit; 1421 } 1422 up->forward_deficit = 0; 1423 1424 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1425 * if the called don't held it already 1426 */ 1427 sk_queue = &sk->sk_receive_queue; 1428 if (!rx_queue_lock_held) 1429 spin_lock(&sk_queue->lock); 1430 1431 1432 sk->sk_forward_alloc += size; 1433 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1); 1434 sk->sk_forward_alloc -= amt; 1435 1436 if (amt) 1437 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT); 1438 1439 atomic_sub(size, &sk->sk_rmem_alloc); 1440 1441 /* this can save us from acquiring the rx queue lock on next receive */ 1442 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1443 1444 if (!rx_queue_lock_held) 1445 spin_unlock(&sk_queue->lock); 1446 } 1447 1448 /* Note: called with reader_queue.lock held. 1449 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1450 * This avoids a cache line miss while receive_queue lock is held. 1451 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1452 */ 1453 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1454 { 1455 prefetch(&skb->data); 1456 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1457 } 1458 EXPORT_SYMBOL(udp_skb_destructor); 1459 1460 /* as above, but the caller held the rx queue lock, too */ 1461 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1462 { 1463 prefetch(&skb->data); 1464 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1465 } 1466 1467 /* Idea of busylocks is to let producers grab an extra spinlock 1468 * to relieve pressure on the receive_queue spinlock shared by consumer. 1469 * Under flood, this means that only one producer can be in line 1470 * trying to acquire the receive_queue spinlock. 1471 * These busylock can be allocated on a per cpu manner, instead of a 1472 * per socket one (that would consume a cache line per socket) 1473 */ 1474 static int udp_busylocks_log __read_mostly; 1475 static spinlock_t *udp_busylocks __read_mostly; 1476 1477 static spinlock_t *busylock_acquire(void *ptr) 1478 { 1479 spinlock_t *busy; 1480 1481 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1482 spin_lock(busy); 1483 return busy; 1484 } 1485 1486 static void busylock_release(spinlock_t *busy) 1487 { 1488 if (busy) 1489 spin_unlock(busy); 1490 } 1491 1492 static int udp_rmem_schedule(struct sock *sk, int size) 1493 { 1494 int delta; 1495 1496 delta = size - sk->sk_forward_alloc; 1497 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV)) 1498 return -ENOBUFS; 1499 1500 return 0; 1501 } 1502 1503 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1504 { 1505 struct sk_buff_head *list = &sk->sk_receive_queue; 1506 int rmem, err = -ENOMEM; 1507 spinlock_t *busy = NULL; 1508 int size; 1509 1510 /* try to avoid the costly atomic add/sub pair when the receive 1511 * queue is full; always allow at least a packet 1512 */ 1513 rmem = atomic_read(&sk->sk_rmem_alloc); 1514 if (rmem > sk->sk_rcvbuf) 1515 goto drop; 1516 1517 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1518 * having linear skbs : 1519 * - Reduce memory overhead and thus increase receive queue capacity 1520 * - Less cache line misses at copyout() time 1521 * - Less work at consume_skb() (less alien page frag freeing) 1522 */ 1523 if (rmem > (sk->sk_rcvbuf >> 1)) { 1524 skb_condense(skb); 1525 1526 busy = busylock_acquire(sk); 1527 } 1528 size = skb->truesize; 1529 udp_set_dev_scratch(skb); 1530 1531 /* we drop only if the receive buf is full and the receive 1532 * queue contains some other skb 1533 */ 1534 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1535 if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) 1536 goto uncharge_drop; 1537 1538 spin_lock(&list->lock); 1539 err = udp_rmem_schedule(sk, size); 1540 if (err) { 1541 spin_unlock(&list->lock); 1542 goto uncharge_drop; 1543 } 1544 1545 sk->sk_forward_alloc -= size; 1546 1547 /* no need to setup a destructor, we will explicitly release the 1548 * forward allocated memory on dequeue 1549 */ 1550 sock_skb_set_dropcount(sk, skb); 1551 1552 __skb_queue_tail(list, skb); 1553 spin_unlock(&list->lock); 1554 1555 if (!sock_flag(sk, SOCK_DEAD)) 1556 sk->sk_data_ready(sk); 1557 1558 busylock_release(busy); 1559 return 0; 1560 1561 uncharge_drop: 1562 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1563 1564 drop: 1565 atomic_inc(&sk->sk_drops); 1566 busylock_release(busy); 1567 return err; 1568 } 1569 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1570 1571 void udp_destruct_common(struct sock *sk) 1572 { 1573 /* reclaim completely the forward allocated memory */ 1574 struct udp_sock *up = udp_sk(sk); 1575 unsigned int total = 0; 1576 struct sk_buff *skb; 1577 1578 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1579 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1580 total += skb->truesize; 1581 kfree_skb(skb); 1582 } 1583 udp_rmem_release(sk, total, 0, true); 1584 } 1585 EXPORT_SYMBOL_GPL(udp_destruct_common); 1586 1587 static void udp_destruct_sock(struct sock *sk) 1588 { 1589 udp_destruct_common(sk); 1590 inet_sock_destruct(sk); 1591 } 1592 1593 int udp_init_sock(struct sock *sk) 1594 { 1595 udp_lib_init_sock(sk); 1596 sk->sk_destruct = udp_destruct_sock; 1597 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1598 return 0; 1599 } 1600 1601 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1602 { 1603 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1604 bool slow = lock_sock_fast(sk); 1605 1606 sk_peek_offset_bwd(sk, len); 1607 unlock_sock_fast(sk, slow); 1608 } 1609 1610 if (!skb_unref(skb)) 1611 return; 1612 1613 /* In the more common cases we cleared the head states previously, 1614 * see __udp_queue_rcv_skb(). 1615 */ 1616 if (unlikely(udp_skb_has_head_state(skb))) 1617 skb_release_head_state(skb); 1618 __consume_stateless_skb(skb); 1619 } 1620 EXPORT_SYMBOL_GPL(skb_consume_udp); 1621 1622 static struct sk_buff *__first_packet_length(struct sock *sk, 1623 struct sk_buff_head *rcvq, 1624 int *total) 1625 { 1626 struct sk_buff *skb; 1627 1628 while ((skb = skb_peek(rcvq)) != NULL) { 1629 if (udp_lib_checksum_complete(skb)) { 1630 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1631 IS_UDPLITE(sk)); 1632 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1633 IS_UDPLITE(sk)); 1634 atomic_inc(&sk->sk_drops); 1635 __skb_unlink(skb, rcvq); 1636 *total += skb->truesize; 1637 kfree_skb(skb); 1638 } else { 1639 udp_skb_csum_unnecessary_set(skb); 1640 break; 1641 } 1642 } 1643 return skb; 1644 } 1645 1646 /** 1647 * first_packet_length - return length of first packet in receive queue 1648 * @sk: socket 1649 * 1650 * Drops all bad checksum frames, until a valid one is found. 1651 * Returns the length of found skb, or -1 if none is found. 1652 */ 1653 static int first_packet_length(struct sock *sk) 1654 { 1655 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1656 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1657 struct sk_buff *skb; 1658 int total = 0; 1659 int res; 1660 1661 spin_lock_bh(&rcvq->lock); 1662 skb = __first_packet_length(sk, rcvq, &total); 1663 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1664 spin_lock(&sk_queue->lock); 1665 skb_queue_splice_tail_init(sk_queue, rcvq); 1666 spin_unlock(&sk_queue->lock); 1667 1668 skb = __first_packet_length(sk, rcvq, &total); 1669 } 1670 res = skb ? skb->len : -1; 1671 if (total) 1672 udp_rmem_release(sk, total, 1, false); 1673 spin_unlock_bh(&rcvq->lock); 1674 return res; 1675 } 1676 1677 /* 1678 * IOCTL requests applicable to the UDP protocol 1679 */ 1680 1681 int udp_ioctl(struct sock *sk, int cmd, int *karg) 1682 { 1683 switch (cmd) { 1684 case SIOCOUTQ: 1685 { 1686 *karg = sk_wmem_alloc_get(sk); 1687 return 0; 1688 } 1689 1690 case SIOCINQ: 1691 { 1692 *karg = max_t(int, 0, first_packet_length(sk)); 1693 return 0; 1694 } 1695 1696 default: 1697 return -ENOIOCTLCMD; 1698 } 1699 1700 return 0; 1701 } 1702 EXPORT_SYMBOL(udp_ioctl); 1703 1704 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1705 int *off, int *err) 1706 { 1707 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1708 struct sk_buff_head *queue; 1709 struct sk_buff *last; 1710 long timeo; 1711 int error; 1712 1713 queue = &udp_sk(sk)->reader_queue; 1714 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1715 do { 1716 struct sk_buff *skb; 1717 1718 error = sock_error(sk); 1719 if (error) 1720 break; 1721 1722 error = -EAGAIN; 1723 do { 1724 spin_lock_bh(&queue->lock); 1725 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1726 err, &last); 1727 if (skb) { 1728 if (!(flags & MSG_PEEK)) 1729 udp_skb_destructor(sk, skb); 1730 spin_unlock_bh(&queue->lock); 1731 return skb; 1732 } 1733 1734 if (skb_queue_empty_lockless(sk_queue)) { 1735 spin_unlock_bh(&queue->lock); 1736 goto busy_check; 1737 } 1738 1739 /* refill the reader queue and walk it again 1740 * keep both queues locked to avoid re-acquiring 1741 * the sk_receive_queue lock if fwd memory scheduling 1742 * is needed. 1743 */ 1744 spin_lock(&sk_queue->lock); 1745 skb_queue_splice_tail_init(sk_queue, queue); 1746 1747 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1748 err, &last); 1749 if (skb && !(flags & MSG_PEEK)) 1750 udp_skb_dtor_locked(sk, skb); 1751 spin_unlock(&sk_queue->lock); 1752 spin_unlock_bh(&queue->lock); 1753 if (skb) 1754 return skb; 1755 1756 busy_check: 1757 if (!sk_can_busy_loop(sk)) 1758 break; 1759 1760 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1761 } while (!skb_queue_empty_lockless(sk_queue)); 1762 1763 /* sk_queue is empty, reader_queue may contain peeked packets */ 1764 } while (timeo && 1765 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1766 &error, &timeo, 1767 (struct sk_buff *)sk_queue)); 1768 1769 *err = error; 1770 return NULL; 1771 } 1772 EXPORT_SYMBOL(__skb_recv_udp); 1773 1774 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor) 1775 { 1776 struct sk_buff *skb; 1777 int err; 1778 1779 try_again: 1780 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err); 1781 if (!skb) 1782 return err; 1783 1784 if (udp_lib_checksum_complete(skb)) { 1785 int is_udplite = IS_UDPLITE(sk); 1786 struct net *net = sock_net(sk); 1787 1788 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite); 1789 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite); 1790 atomic_inc(&sk->sk_drops); 1791 kfree_skb(skb); 1792 goto try_again; 1793 } 1794 1795 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk)); 1796 return recv_actor(sk, skb); 1797 } 1798 EXPORT_SYMBOL(udp_read_skb); 1799 1800 /* 1801 * This should be easy, if there is something there we 1802 * return it, otherwise we block. 1803 */ 1804 1805 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 1806 int *addr_len) 1807 { 1808 struct inet_sock *inet = inet_sk(sk); 1809 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1810 struct sk_buff *skb; 1811 unsigned int ulen, copied; 1812 int off, err, peeking = flags & MSG_PEEK; 1813 int is_udplite = IS_UDPLITE(sk); 1814 bool checksum_valid = false; 1815 1816 if (flags & MSG_ERRQUEUE) 1817 return ip_recv_error(sk, msg, len, addr_len); 1818 1819 try_again: 1820 off = sk_peek_offset(sk, flags); 1821 skb = __skb_recv_udp(sk, flags, &off, &err); 1822 if (!skb) 1823 return err; 1824 1825 ulen = udp_skb_len(skb); 1826 copied = len; 1827 if (copied > ulen - off) 1828 copied = ulen - off; 1829 else if (copied < ulen) 1830 msg->msg_flags |= MSG_TRUNC; 1831 1832 /* 1833 * If checksum is needed at all, try to do it while copying the 1834 * data. If the data is truncated, or if we only want a partial 1835 * coverage checksum (UDP-Lite), do it before the copy. 1836 */ 1837 1838 if (copied < ulen || peeking || 1839 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1840 checksum_valid = udp_skb_csum_unnecessary(skb) || 1841 !__udp_lib_checksum_complete(skb); 1842 if (!checksum_valid) 1843 goto csum_copy_err; 1844 } 1845 1846 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1847 if (udp_skb_is_linear(skb)) 1848 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1849 else 1850 err = skb_copy_datagram_msg(skb, off, msg, copied); 1851 } else { 1852 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1853 1854 if (err == -EINVAL) 1855 goto csum_copy_err; 1856 } 1857 1858 if (unlikely(err)) { 1859 if (!peeking) { 1860 atomic_inc(&sk->sk_drops); 1861 UDP_INC_STATS(sock_net(sk), 1862 UDP_MIB_INERRORS, is_udplite); 1863 } 1864 kfree_skb(skb); 1865 return err; 1866 } 1867 1868 if (!peeking) 1869 UDP_INC_STATS(sock_net(sk), 1870 UDP_MIB_INDATAGRAMS, is_udplite); 1871 1872 sock_recv_cmsgs(msg, sk, skb); 1873 1874 /* Copy the address. */ 1875 if (sin) { 1876 sin->sin_family = AF_INET; 1877 sin->sin_port = udp_hdr(skb)->source; 1878 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1879 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1880 *addr_len = sizeof(*sin); 1881 1882 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1883 (struct sockaddr *)sin); 1884 } 1885 1886 if (udp_sk(sk)->gro_enabled) 1887 udp_cmsg_recv(msg, sk, skb); 1888 1889 if (inet->cmsg_flags) 1890 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1891 1892 err = copied; 1893 if (flags & MSG_TRUNC) 1894 err = ulen; 1895 1896 skb_consume_udp(sk, skb, peeking ? -err : err); 1897 return err; 1898 1899 csum_copy_err: 1900 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1901 udp_skb_destructor)) { 1902 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1903 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1904 } 1905 kfree_skb(skb); 1906 1907 /* starting over for a new packet, but check if we need to yield */ 1908 cond_resched(); 1909 msg->msg_flags &= ~MSG_TRUNC; 1910 goto try_again; 1911 } 1912 1913 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1914 { 1915 /* This check is replicated from __ip4_datagram_connect() and 1916 * intended to prevent BPF program called below from accessing bytes 1917 * that are out of the bound specified by user in addr_len. 1918 */ 1919 if (addr_len < sizeof(struct sockaddr_in)) 1920 return -EINVAL; 1921 1922 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1923 } 1924 EXPORT_SYMBOL(udp_pre_connect); 1925 1926 int __udp_disconnect(struct sock *sk, int flags) 1927 { 1928 struct inet_sock *inet = inet_sk(sk); 1929 /* 1930 * 1003.1g - break association. 1931 */ 1932 1933 sk->sk_state = TCP_CLOSE; 1934 inet->inet_daddr = 0; 1935 inet->inet_dport = 0; 1936 sock_rps_reset_rxhash(sk); 1937 sk->sk_bound_dev_if = 0; 1938 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1939 inet_reset_saddr(sk); 1940 if (sk->sk_prot->rehash && 1941 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1942 sk->sk_prot->rehash(sk); 1943 } 1944 1945 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1946 sk->sk_prot->unhash(sk); 1947 inet->inet_sport = 0; 1948 } 1949 sk_dst_reset(sk); 1950 return 0; 1951 } 1952 EXPORT_SYMBOL(__udp_disconnect); 1953 1954 int udp_disconnect(struct sock *sk, int flags) 1955 { 1956 lock_sock(sk); 1957 __udp_disconnect(sk, flags); 1958 release_sock(sk); 1959 return 0; 1960 } 1961 EXPORT_SYMBOL(udp_disconnect); 1962 1963 void udp_lib_unhash(struct sock *sk) 1964 { 1965 if (sk_hashed(sk)) { 1966 struct udp_table *udptable = udp_get_table_prot(sk); 1967 struct udp_hslot *hslot, *hslot2; 1968 1969 hslot = udp_hashslot(udptable, sock_net(sk), 1970 udp_sk(sk)->udp_port_hash); 1971 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1972 1973 spin_lock_bh(&hslot->lock); 1974 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1975 reuseport_detach_sock(sk); 1976 if (sk_del_node_init_rcu(sk)) { 1977 hslot->count--; 1978 inet_sk(sk)->inet_num = 0; 1979 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1980 1981 spin_lock(&hslot2->lock); 1982 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1983 hslot2->count--; 1984 spin_unlock(&hslot2->lock); 1985 } 1986 spin_unlock_bh(&hslot->lock); 1987 } 1988 } 1989 EXPORT_SYMBOL(udp_lib_unhash); 1990 1991 /* 1992 * inet_rcv_saddr was changed, we must rehash secondary hash 1993 */ 1994 void udp_lib_rehash(struct sock *sk, u16 newhash) 1995 { 1996 if (sk_hashed(sk)) { 1997 struct udp_table *udptable = udp_get_table_prot(sk); 1998 struct udp_hslot *hslot, *hslot2, *nhslot2; 1999 2000 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2001 nhslot2 = udp_hashslot2(udptable, newhash); 2002 udp_sk(sk)->udp_portaddr_hash = newhash; 2003 2004 if (hslot2 != nhslot2 || 2005 rcu_access_pointer(sk->sk_reuseport_cb)) { 2006 hslot = udp_hashslot(udptable, sock_net(sk), 2007 udp_sk(sk)->udp_port_hash); 2008 /* we must lock primary chain too */ 2009 spin_lock_bh(&hslot->lock); 2010 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2011 reuseport_detach_sock(sk); 2012 2013 if (hslot2 != nhslot2) { 2014 spin_lock(&hslot2->lock); 2015 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2016 hslot2->count--; 2017 spin_unlock(&hslot2->lock); 2018 2019 spin_lock(&nhslot2->lock); 2020 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2021 &nhslot2->head); 2022 nhslot2->count++; 2023 spin_unlock(&nhslot2->lock); 2024 } 2025 2026 spin_unlock_bh(&hslot->lock); 2027 } 2028 } 2029 } 2030 EXPORT_SYMBOL(udp_lib_rehash); 2031 2032 void udp_v4_rehash(struct sock *sk) 2033 { 2034 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2035 inet_sk(sk)->inet_rcv_saddr, 2036 inet_sk(sk)->inet_num); 2037 udp_lib_rehash(sk, new_hash); 2038 } 2039 2040 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2041 { 2042 int rc; 2043 2044 if (inet_sk(sk)->inet_daddr) { 2045 sock_rps_save_rxhash(sk, skb); 2046 sk_mark_napi_id(sk, skb); 2047 sk_incoming_cpu_update(sk); 2048 } else { 2049 sk_mark_napi_id_once(sk, skb); 2050 } 2051 2052 rc = __udp_enqueue_schedule_skb(sk, skb); 2053 if (rc < 0) { 2054 int is_udplite = IS_UDPLITE(sk); 2055 int drop_reason; 2056 2057 /* Note that an ENOMEM error is charged twice */ 2058 if (rc == -ENOMEM) { 2059 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2060 is_udplite); 2061 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 2062 } else { 2063 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2064 is_udplite); 2065 drop_reason = SKB_DROP_REASON_PROTO_MEM; 2066 } 2067 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2068 kfree_skb_reason(skb, drop_reason); 2069 trace_udp_fail_queue_rcv_skb(rc, sk); 2070 return -1; 2071 } 2072 2073 return 0; 2074 } 2075 2076 /* returns: 2077 * -1: error 2078 * 0: success 2079 * >0: "udp encap" protocol resubmission 2080 * 2081 * Note that in the success and error cases, the skb is assumed to 2082 * have either been requeued or freed. 2083 */ 2084 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2085 { 2086 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2087 struct udp_sock *up = udp_sk(sk); 2088 int is_udplite = IS_UDPLITE(sk); 2089 2090 /* 2091 * Charge it to the socket, dropping if the queue is full. 2092 */ 2093 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) { 2094 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 2095 goto drop; 2096 } 2097 nf_reset_ct(skb); 2098 2099 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 2100 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2101 2102 /* 2103 * This is an encapsulation socket so pass the skb to 2104 * the socket's udp_encap_rcv() hook. Otherwise, just 2105 * fall through and pass this up the UDP socket. 2106 * up->encap_rcv() returns the following value: 2107 * =0 if skb was successfully passed to the encap 2108 * handler or was discarded by it. 2109 * >0 if skb should be passed on to UDP. 2110 * <0 if skb should be resubmitted as proto -N 2111 */ 2112 2113 /* if we're overly short, let UDP handle it */ 2114 encap_rcv = READ_ONCE(up->encap_rcv); 2115 if (encap_rcv) { 2116 int ret; 2117 2118 /* Verify checksum before giving to encap */ 2119 if (udp_lib_checksum_complete(skb)) 2120 goto csum_error; 2121 2122 ret = encap_rcv(sk, skb); 2123 if (ret <= 0) { 2124 __UDP_INC_STATS(sock_net(sk), 2125 UDP_MIB_INDATAGRAMS, 2126 is_udplite); 2127 return -ret; 2128 } 2129 } 2130 2131 /* FALLTHROUGH -- it's a UDP Packet */ 2132 } 2133 2134 /* 2135 * UDP-Lite specific tests, ignored on UDP sockets 2136 */ 2137 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2138 2139 /* 2140 * MIB statistics other than incrementing the error count are 2141 * disabled for the following two types of errors: these depend 2142 * on the application settings, not on the functioning of the 2143 * protocol stack as such. 2144 * 2145 * RFC 3828 here recommends (sec 3.3): "There should also be a 2146 * way ... to ... at least let the receiving application block 2147 * delivery of packets with coverage values less than a value 2148 * provided by the application." 2149 */ 2150 if (up->pcrlen == 0) { /* full coverage was set */ 2151 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2152 UDP_SKB_CB(skb)->cscov, skb->len); 2153 goto drop; 2154 } 2155 /* The next case involves violating the min. coverage requested 2156 * by the receiver. This is subtle: if receiver wants x and x is 2157 * greater than the buffersize/MTU then receiver will complain 2158 * that it wants x while sender emits packets of smaller size y. 2159 * Therefore the above ...()->partial_cov statement is essential. 2160 */ 2161 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2162 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2163 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2164 goto drop; 2165 } 2166 } 2167 2168 prefetch(&sk->sk_rmem_alloc); 2169 if (rcu_access_pointer(sk->sk_filter) && 2170 udp_lib_checksum_complete(skb)) 2171 goto csum_error; 2172 2173 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 2174 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 2175 goto drop; 2176 } 2177 2178 udp_csum_pull_header(skb); 2179 2180 ipv4_pktinfo_prepare(sk, skb); 2181 return __udp_queue_rcv_skb(sk, skb); 2182 2183 csum_error: 2184 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2185 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2186 drop: 2187 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2188 atomic_inc(&sk->sk_drops); 2189 kfree_skb_reason(skb, drop_reason); 2190 return -1; 2191 } 2192 2193 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2194 { 2195 struct sk_buff *next, *segs; 2196 int ret; 2197 2198 if (likely(!udp_unexpected_gso(sk, skb))) 2199 return udp_queue_rcv_one_skb(sk, skb); 2200 2201 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2202 __skb_push(skb, -skb_mac_offset(skb)); 2203 segs = udp_rcv_segment(sk, skb, true); 2204 skb_list_walk_safe(segs, skb, next) { 2205 __skb_pull(skb, skb_transport_offset(skb)); 2206 2207 udp_post_segment_fix_csum(skb); 2208 ret = udp_queue_rcv_one_skb(sk, skb); 2209 if (ret > 0) 2210 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2211 } 2212 return 0; 2213 } 2214 2215 /* For TCP sockets, sk_rx_dst is protected by socket lock 2216 * For UDP, we use xchg() to guard against concurrent changes. 2217 */ 2218 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2219 { 2220 struct dst_entry *old; 2221 2222 if (dst_hold_safe(dst)) { 2223 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); 2224 dst_release(old); 2225 return old != dst; 2226 } 2227 return false; 2228 } 2229 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2230 2231 /* 2232 * Multicasts and broadcasts go to each listener. 2233 * 2234 * Note: called only from the BH handler context. 2235 */ 2236 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2237 struct udphdr *uh, 2238 __be32 saddr, __be32 daddr, 2239 struct udp_table *udptable, 2240 int proto) 2241 { 2242 struct sock *sk, *first = NULL; 2243 unsigned short hnum = ntohs(uh->dest); 2244 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2245 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2246 unsigned int offset = offsetof(typeof(*sk), sk_node); 2247 int dif = skb->dev->ifindex; 2248 int sdif = inet_sdif(skb); 2249 struct hlist_node *node; 2250 struct sk_buff *nskb; 2251 2252 if (use_hash2) { 2253 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2254 udptable->mask; 2255 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2256 start_lookup: 2257 hslot = &udptable->hash2[hash2]; 2258 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2259 } 2260 2261 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2262 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2263 uh->source, saddr, dif, sdif, hnum)) 2264 continue; 2265 2266 if (!first) { 2267 first = sk; 2268 continue; 2269 } 2270 nskb = skb_clone(skb, GFP_ATOMIC); 2271 2272 if (unlikely(!nskb)) { 2273 atomic_inc(&sk->sk_drops); 2274 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2275 IS_UDPLITE(sk)); 2276 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2277 IS_UDPLITE(sk)); 2278 continue; 2279 } 2280 if (udp_queue_rcv_skb(sk, nskb) > 0) 2281 consume_skb(nskb); 2282 } 2283 2284 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2285 if (use_hash2 && hash2 != hash2_any) { 2286 hash2 = hash2_any; 2287 goto start_lookup; 2288 } 2289 2290 if (first) { 2291 if (udp_queue_rcv_skb(first, skb) > 0) 2292 consume_skb(skb); 2293 } else { 2294 kfree_skb(skb); 2295 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2296 proto == IPPROTO_UDPLITE); 2297 } 2298 return 0; 2299 } 2300 2301 /* Initialize UDP checksum. If exited with zero value (success), 2302 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2303 * Otherwise, csum completion requires checksumming packet body, 2304 * including udp header and folding it to skb->csum. 2305 */ 2306 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2307 int proto) 2308 { 2309 int err; 2310 2311 UDP_SKB_CB(skb)->partial_cov = 0; 2312 UDP_SKB_CB(skb)->cscov = skb->len; 2313 2314 if (proto == IPPROTO_UDPLITE) { 2315 err = udplite_checksum_init(skb, uh); 2316 if (err) 2317 return err; 2318 2319 if (UDP_SKB_CB(skb)->partial_cov) { 2320 skb->csum = inet_compute_pseudo(skb, proto); 2321 return 0; 2322 } 2323 } 2324 2325 /* Note, we are only interested in != 0 or == 0, thus the 2326 * force to int. 2327 */ 2328 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2329 inet_compute_pseudo); 2330 if (err) 2331 return err; 2332 2333 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2334 /* If SW calculated the value, we know it's bad */ 2335 if (skb->csum_complete_sw) 2336 return 1; 2337 2338 /* HW says the value is bad. Let's validate that. 2339 * skb->csum is no longer the full packet checksum, 2340 * so don't treat it as such. 2341 */ 2342 skb_checksum_complete_unset(skb); 2343 } 2344 2345 return 0; 2346 } 2347 2348 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2349 * return code conversion for ip layer consumption 2350 */ 2351 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2352 struct udphdr *uh) 2353 { 2354 int ret; 2355 2356 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2357 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2358 2359 ret = udp_queue_rcv_skb(sk, skb); 2360 2361 /* a return value > 0 means to resubmit the input, but 2362 * it wants the return to be -protocol, or 0 2363 */ 2364 if (ret > 0) 2365 return -ret; 2366 return 0; 2367 } 2368 2369 /* 2370 * All we need to do is get the socket, and then do a checksum. 2371 */ 2372 2373 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2374 int proto) 2375 { 2376 struct sock *sk; 2377 struct udphdr *uh; 2378 unsigned short ulen; 2379 struct rtable *rt = skb_rtable(skb); 2380 __be32 saddr, daddr; 2381 struct net *net = dev_net(skb->dev); 2382 bool refcounted; 2383 int drop_reason; 2384 2385 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2386 2387 /* 2388 * Validate the packet. 2389 */ 2390 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2391 goto drop; /* No space for header. */ 2392 2393 uh = udp_hdr(skb); 2394 ulen = ntohs(uh->len); 2395 saddr = ip_hdr(skb)->saddr; 2396 daddr = ip_hdr(skb)->daddr; 2397 2398 if (ulen > skb->len) 2399 goto short_packet; 2400 2401 if (proto == IPPROTO_UDP) { 2402 /* UDP validates ulen. */ 2403 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2404 goto short_packet; 2405 uh = udp_hdr(skb); 2406 } 2407 2408 if (udp4_csum_init(skb, uh, proto)) 2409 goto csum_error; 2410 2411 sk = skb_steal_sock(skb, &refcounted); 2412 if (sk) { 2413 struct dst_entry *dst = skb_dst(skb); 2414 int ret; 2415 2416 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2417 udp_sk_rx_dst_set(sk, dst); 2418 2419 ret = udp_unicast_rcv_skb(sk, skb, uh); 2420 if (refcounted) 2421 sock_put(sk); 2422 return ret; 2423 } 2424 2425 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2426 return __udp4_lib_mcast_deliver(net, skb, uh, 2427 saddr, daddr, udptable, proto); 2428 2429 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2430 if (sk) 2431 return udp_unicast_rcv_skb(sk, skb, uh); 2432 2433 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2434 goto drop; 2435 nf_reset_ct(skb); 2436 2437 /* No socket. Drop packet silently, if checksum is wrong */ 2438 if (udp_lib_checksum_complete(skb)) 2439 goto csum_error; 2440 2441 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2442 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2443 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2444 2445 /* 2446 * Hmm. We got an UDP packet to a port to which we 2447 * don't wanna listen. Ignore it. 2448 */ 2449 kfree_skb_reason(skb, drop_reason); 2450 return 0; 2451 2452 short_packet: 2453 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2454 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2455 proto == IPPROTO_UDPLITE ? "Lite" : "", 2456 &saddr, ntohs(uh->source), 2457 ulen, skb->len, 2458 &daddr, ntohs(uh->dest)); 2459 goto drop; 2460 2461 csum_error: 2462 /* 2463 * RFC1122: OK. Discards the bad packet silently (as far as 2464 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2465 */ 2466 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2467 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2468 proto == IPPROTO_UDPLITE ? "Lite" : "", 2469 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2470 ulen); 2471 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2472 drop: 2473 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2474 kfree_skb_reason(skb, drop_reason); 2475 return 0; 2476 } 2477 2478 /* We can only early demux multicast if there is a single matching socket. 2479 * If more than one socket found returns NULL 2480 */ 2481 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2482 __be16 loc_port, __be32 loc_addr, 2483 __be16 rmt_port, __be32 rmt_addr, 2484 int dif, int sdif) 2485 { 2486 struct udp_table *udptable = net->ipv4.udp_table; 2487 unsigned short hnum = ntohs(loc_port); 2488 struct sock *sk, *result; 2489 struct udp_hslot *hslot; 2490 unsigned int slot; 2491 2492 slot = udp_hashfn(net, hnum, udptable->mask); 2493 hslot = &udptable->hash[slot]; 2494 2495 /* Do not bother scanning a too big list */ 2496 if (hslot->count > 10) 2497 return NULL; 2498 2499 result = NULL; 2500 sk_for_each_rcu(sk, &hslot->head) { 2501 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2502 rmt_port, rmt_addr, dif, sdif, hnum)) { 2503 if (result) 2504 return NULL; 2505 result = sk; 2506 } 2507 } 2508 2509 return result; 2510 } 2511 2512 /* For unicast we should only early demux connected sockets or we can 2513 * break forwarding setups. The chains here can be long so only check 2514 * if the first socket is an exact match and if not move on. 2515 */ 2516 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2517 __be16 loc_port, __be32 loc_addr, 2518 __be16 rmt_port, __be32 rmt_addr, 2519 int dif, int sdif) 2520 { 2521 struct udp_table *udptable = net->ipv4.udp_table; 2522 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2523 unsigned short hnum = ntohs(loc_port); 2524 unsigned int hash2, slot2; 2525 struct udp_hslot *hslot2; 2526 __portpair ports; 2527 struct sock *sk; 2528 2529 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2530 slot2 = hash2 & udptable->mask; 2531 hslot2 = &udptable->hash2[slot2]; 2532 ports = INET_COMBINED_PORTS(rmt_port, hnum); 2533 2534 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2535 if (inet_match(net, sk, acookie, ports, dif, sdif)) 2536 return sk; 2537 /* Only check first socket in chain */ 2538 break; 2539 } 2540 return NULL; 2541 } 2542 2543 int udp_v4_early_demux(struct sk_buff *skb) 2544 { 2545 struct net *net = dev_net(skb->dev); 2546 struct in_device *in_dev = NULL; 2547 const struct iphdr *iph; 2548 const struct udphdr *uh; 2549 struct sock *sk = NULL; 2550 struct dst_entry *dst; 2551 int dif = skb->dev->ifindex; 2552 int sdif = inet_sdif(skb); 2553 int ours; 2554 2555 /* validate the packet */ 2556 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2557 return 0; 2558 2559 iph = ip_hdr(skb); 2560 uh = udp_hdr(skb); 2561 2562 if (skb->pkt_type == PACKET_MULTICAST) { 2563 in_dev = __in_dev_get_rcu(skb->dev); 2564 2565 if (!in_dev) 2566 return 0; 2567 2568 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2569 iph->protocol); 2570 if (!ours) 2571 return 0; 2572 2573 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2574 uh->source, iph->saddr, 2575 dif, sdif); 2576 } else if (skb->pkt_type == PACKET_HOST) { 2577 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2578 uh->source, iph->saddr, dif, sdif); 2579 } 2580 2581 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2582 return 0; 2583 2584 skb->sk = sk; 2585 skb->destructor = sock_efree; 2586 dst = rcu_dereference(sk->sk_rx_dst); 2587 2588 if (dst) 2589 dst = dst_check(dst, 0); 2590 if (dst) { 2591 u32 itag = 0; 2592 2593 /* set noref for now. 2594 * any place which wants to hold dst has to call 2595 * dst_hold_safe() 2596 */ 2597 skb_dst_set_noref(skb, dst); 2598 2599 /* for unconnected multicast sockets we need to validate 2600 * the source on each packet 2601 */ 2602 if (!inet_sk(sk)->inet_daddr && in_dev) 2603 return ip_mc_validate_source(skb, iph->daddr, 2604 iph->saddr, 2605 iph->tos & IPTOS_RT_MASK, 2606 skb->dev, in_dev, &itag); 2607 } 2608 return 0; 2609 } 2610 2611 int udp_rcv(struct sk_buff *skb) 2612 { 2613 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 2614 } 2615 2616 void udp_destroy_sock(struct sock *sk) 2617 { 2618 struct udp_sock *up = udp_sk(sk); 2619 bool slow = lock_sock_fast(sk); 2620 2621 /* protects from races with udp_abort() */ 2622 sock_set_flag(sk, SOCK_DEAD); 2623 udp_flush_pending_frames(sk); 2624 unlock_sock_fast(sk, slow); 2625 if (static_branch_unlikely(&udp_encap_needed_key)) { 2626 if (up->encap_type) { 2627 void (*encap_destroy)(struct sock *sk); 2628 encap_destroy = READ_ONCE(up->encap_destroy); 2629 if (encap_destroy) 2630 encap_destroy(sk); 2631 } 2632 if (up->encap_enabled) 2633 static_branch_dec(&udp_encap_needed_key); 2634 } 2635 } 2636 2637 /* 2638 * Socket option code for UDP 2639 */ 2640 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2641 sockptr_t optval, unsigned int optlen, 2642 int (*push_pending_frames)(struct sock *)) 2643 { 2644 struct udp_sock *up = udp_sk(sk); 2645 int val, valbool; 2646 int err = 0; 2647 int is_udplite = IS_UDPLITE(sk); 2648 2649 if (level == SOL_SOCKET) { 2650 err = sk_setsockopt(sk, level, optname, optval, optlen); 2651 2652 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) { 2653 sockopt_lock_sock(sk); 2654 /* paired with READ_ONCE in udp_rmem_release() */ 2655 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2); 2656 sockopt_release_sock(sk); 2657 } 2658 return err; 2659 } 2660 2661 if (optlen < sizeof(int)) 2662 return -EINVAL; 2663 2664 if (copy_from_sockptr(&val, optval, sizeof(val))) 2665 return -EFAULT; 2666 2667 valbool = val ? 1 : 0; 2668 2669 switch (optname) { 2670 case UDP_CORK: 2671 if (val != 0) { 2672 WRITE_ONCE(up->corkflag, 1); 2673 } else { 2674 WRITE_ONCE(up->corkflag, 0); 2675 lock_sock(sk); 2676 push_pending_frames(sk); 2677 release_sock(sk); 2678 } 2679 break; 2680 2681 case UDP_ENCAP: 2682 switch (val) { 2683 case 0: 2684 #ifdef CONFIG_XFRM 2685 case UDP_ENCAP_ESPINUDP: 2686 case UDP_ENCAP_ESPINUDP_NON_IKE: 2687 #if IS_ENABLED(CONFIG_IPV6) 2688 if (sk->sk_family == AF_INET6) 2689 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; 2690 else 2691 #endif 2692 up->encap_rcv = xfrm4_udp_encap_rcv; 2693 #endif 2694 fallthrough; 2695 case UDP_ENCAP_L2TPINUDP: 2696 up->encap_type = val; 2697 lock_sock(sk); 2698 udp_tunnel_encap_enable(sk->sk_socket); 2699 release_sock(sk); 2700 break; 2701 default: 2702 err = -ENOPROTOOPT; 2703 break; 2704 } 2705 break; 2706 2707 case UDP_NO_CHECK6_TX: 2708 up->no_check6_tx = valbool; 2709 break; 2710 2711 case UDP_NO_CHECK6_RX: 2712 up->no_check6_rx = valbool; 2713 break; 2714 2715 case UDP_SEGMENT: 2716 if (val < 0 || val > USHRT_MAX) 2717 return -EINVAL; 2718 WRITE_ONCE(up->gso_size, val); 2719 break; 2720 2721 case UDP_GRO: 2722 lock_sock(sk); 2723 2724 /* when enabling GRO, accept the related GSO packet type */ 2725 if (valbool) 2726 udp_tunnel_encap_enable(sk->sk_socket); 2727 up->gro_enabled = valbool; 2728 up->accept_udp_l4 = valbool; 2729 release_sock(sk); 2730 break; 2731 2732 /* 2733 * UDP-Lite's partial checksum coverage (RFC 3828). 2734 */ 2735 /* The sender sets actual checksum coverage length via this option. 2736 * The case coverage > packet length is handled by send module. */ 2737 case UDPLITE_SEND_CSCOV: 2738 if (!is_udplite) /* Disable the option on UDP sockets */ 2739 return -ENOPROTOOPT; 2740 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2741 val = 8; 2742 else if (val > USHRT_MAX) 2743 val = USHRT_MAX; 2744 up->pcslen = val; 2745 up->pcflag |= UDPLITE_SEND_CC; 2746 break; 2747 2748 /* The receiver specifies a minimum checksum coverage value. To make 2749 * sense, this should be set to at least 8 (as done below). If zero is 2750 * used, this again means full checksum coverage. */ 2751 case UDPLITE_RECV_CSCOV: 2752 if (!is_udplite) /* Disable the option on UDP sockets */ 2753 return -ENOPROTOOPT; 2754 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2755 val = 8; 2756 else if (val > USHRT_MAX) 2757 val = USHRT_MAX; 2758 up->pcrlen = val; 2759 up->pcflag |= UDPLITE_RECV_CC; 2760 break; 2761 2762 default: 2763 err = -ENOPROTOOPT; 2764 break; 2765 } 2766 2767 return err; 2768 } 2769 EXPORT_SYMBOL(udp_lib_setsockopt); 2770 2771 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2772 unsigned int optlen) 2773 { 2774 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 2775 return udp_lib_setsockopt(sk, level, optname, 2776 optval, optlen, 2777 udp_push_pending_frames); 2778 return ip_setsockopt(sk, level, optname, optval, optlen); 2779 } 2780 2781 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2782 char __user *optval, int __user *optlen) 2783 { 2784 struct udp_sock *up = udp_sk(sk); 2785 int val, len; 2786 2787 if (get_user(len, optlen)) 2788 return -EFAULT; 2789 2790 len = min_t(unsigned int, len, sizeof(int)); 2791 2792 if (len < 0) 2793 return -EINVAL; 2794 2795 switch (optname) { 2796 case UDP_CORK: 2797 val = READ_ONCE(up->corkflag); 2798 break; 2799 2800 case UDP_ENCAP: 2801 val = up->encap_type; 2802 break; 2803 2804 case UDP_NO_CHECK6_TX: 2805 val = up->no_check6_tx; 2806 break; 2807 2808 case UDP_NO_CHECK6_RX: 2809 val = up->no_check6_rx; 2810 break; 2811 2812 case UDP_SEGMENT: 2813 val = READ_ONCE(up->gso_size); 2814 break; 2815 2816 case UDP_GRO: 2817 val = up->gro_enabled; 2818 break; 2819 2820 /* The following two cannot be changed on UDP sockets, the return is 2821 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2822 case UDPLITE_SEND_CSCOV: 2823 val = up->pcslen; 2824 break; 2825 2826 case UDPLITE_RECV_CSCOV: 2827 val = up->pcrlen; 2828 break; 2829 2830 default: 2831 return -ENOPROTOOPT; 2832 } 2833 2834 if (put_user(len, optlen)) 2835 return -EFAULT; 2836 if (copy_to_user(optval, &val, len)) 2837 return -EFAULT; 2838 return 0; 2839 } 2840 EXPORT_SYMBOL(udp_lib_getsockopt); 2841 2842 int udp_getsockopt(struct sock *sk, int level, int optname, 2843 char __user *optval, int __user *optlen) 2844 { 2845 if (level == SOL_UDP || level == SOL_UDPLITE) 2846 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2847 return ip_getsockopt(sk, level, optname, optval, optlen); 2848 } 2849 2850 /** 2851 * udp_poll - wait for a UDP event. 2852 * @file: - file struct 2853 * @sock: - socket 2854 * @wait: - poll table 2855 * 2856 * This is same as datagram poll, except for the special case of 2857 * blocking sockets. If application is using a blocking fd 2858 * and a packet with checksum error is in the queue; 2859 * then it could get return from select indicating data available 2860 * but then block when reading it. Add special case code 2861 * to work around these arguably broken applications. 2862 */ 2863 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2864 { 2865 __poll_t mask = datagram_poll(file, sock, wait); 2866 struct sock *sk = sock->sk; 2867 2868 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2869 mask |= EPOLLIN | EPOLLRDNORM; 2870 2871 /* Check for false positives due to checksum errors */ 2872 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2873 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2874 mask &= ~(EPOLLIN | EPOLLRDNORM); 2875 2876 /* psock ingress_msg queue should not contain any bad checksum frames */ 2877 if (sk_is_readable(sk)) 2878 mask |= EPOLLIN | EPOLLRDNORM; 2879 return mask; 2880 2881 } 2882 EXPORT_SYMBOL(udp_poll); 2883 2884 int udp_abort(struct sock *sk, int err) 2885 { 2886 if (!has_current_bpf_ctx()) 2887 lock_sock(sk); 2888 2889 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2890 * with close() 2891 */ 2892 if (sock_flag(sk, SOCK_DEAD)) 2893 goto out; 2894 2895 sk->sk_err = err; 2896 sk_error_report(sk); 2897 __udp_disconnect(sk, 0); 2898 2899 out: 2900 if (!has_current_bpf_ctx()) 2901 release_sock(sk); 2902 2903 return 0; 2904 } 2905 EXPORT_SYMBOL_GPL(udp_abort); 2906 2907 struct proto udp_prot = { 2908 .name = "UDP", 2909 .owner = THIS_MODULE, 2910 .close = udp_lib_close, 2911 .pre_connect = udp_pre_connect, 2912 .connect = ip4_datagram_connect, 2913 .disconnect = udp_disconnect, 2914 .ioctl = udp_ioctl, 2915 .init = udp_init_sock, 2916 .destroy = udp_destroy_sock, 2917 .setsockopt = udp_setsockopt, 2918 .getsockopt = udp_getsockopt, 2919 .sendmsg = udp_sendmsg, 2920 .recvmsg = udp_recvmsg, 2921 .splice_eof = udp_splice_eof, 2922 .release_cb = ip4_datagram_release_cb, 2923 .hash = udp_lib_hash, 2924 .unhash = udp_lib_unhash, 2925 .rehash = udp_v4_rehash, 2926 .get_port = udp_v4_get_port, 2927 .put_port = udp_lib_unhash, 2928 #ifdef CONFIG_BPF_SYSCALL 2929 .psock_update_sk_prot = udp_bpf_update_proto, 2930 #endif 2931 .memory_allocated = &udp_memory_allocated, 2932 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 2933 2934 .sysctl_mem = sysctl_udp_mem, 2935 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2936 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2937 .obj_size = sizeof(struct udp_sock), 2938 .h.udp_table = NULL, 2939 .diag_destroy = udp_abort, 2940 }; 2941 EXPORT_SYMBOL(udp_prot); 2942 2943 /* ------------------------------------------------------------------------ */ 2944 #ifdef CONFIG_PROC_FS 2945 2946 static unsigned short seq_file_family(const struct seq_file *seq); 2947 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk) 2948 { 2949 unsigned short family = seq_file_family(seq); 2950 2951 /* AF_UNSPEC is used as a match all */ 2952 return ((family == AF_UNSPEC || family == sk->sk_family) && 2953 net_eq(sock_net(sk), seq_file_net(seq))); 2954 } 2955 2956 #ifdef CONFIG_BPF_SYSCALL 2957 static const struct seq_operations bpf_iter_udp_seq_ops; 2958 #endif 2959 static struct udp_table *udp_get_table_seq(struct seq_file *seq, 2960 struct net *net) 2961 { 2962 const struct udp_seq_afinfo *afinfo; 2963 2964 #ifdef CONFIG_BPF_SYSCALL 2965 if (seq->op == &bpf_iter_udp_seq_ops) 2966 return net->ipv4.udp_table; 2967 #endif 2968 2969 afinfo = pde_data(file_inode(seq->file)); 2970 return afinfo->udp_table ? : net->ipv4.udp_table; 2971 } 2972 2973 static struct sock *udp_get_first(struct seq_file *seq, int start) 2974 { 2975 struct udp_iter_state *state = seq->private; 2976 struct net *net = seq_file_net(seq); 2977 struct udp_table *udptable; 2978 struct sock *sk; 2979 2980 udptable = udp_get_table_seq(seq, net); 2981 2982 for (state->bucket = start; state->bucket <= udptable->mask; 2983 ++state->bucket) { 2984 struct udp_hslot *hslot = &udptable->hash[state->bucket]; 2985 2986 if (hlist_empty(&hslot->head)) 2987 continue; 2988 2989 spin_lock_bh(&hslot->lock); 2990 sk_for_each(sk, &hslot->head) { 2991 if (seq_sk_match(seq, sk)) 2992 goto found; 2993 } 2994 spin_unlock_bh(&hslot->lock); 2995 } 2996 sk = NULL; 2997 found: 2998 return sk; 2999 } 3000 3001 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 3002 { 3003 struct udp_iter_state *state = seq->private; 3004 struct net *net = seq_file_net(seq); 3005 struct udp_table *udptable; 3006 3007 do { 3008 sk = sk_next(sk); 3009 } while (sk && !seq_sk_match(seq, sk)); 3010 3011 if (!sk) { 3012 udptable = udp_get_table_seq(seq, net); 3013 3014 if (state->bucket <= udptable->mask) 3015 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3016 3017 return udp_get_first(seq, state->bucket + 1); 3018 } 3019 return sk; 3020 } 3021 3022 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3023 { 3024 struct sock *sk = udp_get_first(seq, 0); 3025 3026 if (sk) 3027 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3028 --pos; 3029 return pos ? NULL : sk; 3030 } 3031 3032 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3033 { 3034 struct udp_iter_state *state = seq->private; 3035 state->bucket = MAX_UDP_PORTS; 3036 3037 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3038 } 3039 EXPORT_SYMBOL(udp_seq_start); 3040 3041 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3042 { 3043 struct sock *sk; 3044 3045 if (v == SEQ_START_TOKEN) 3046 sk = udp_get_idx(seq, 0); 3047 else 3048 sk = udp_get_next(seq, v); 3049 3050 ++*pos; 3051 return sk; 3052 } 3053 EXPORT_SYMBOL(udp_seq_next); 3054 3055 void udp_seq_stop(struct seq_file *seq, void *v) 3056 { 3057 struct udp_iter_state *state = seq->private; 3058 struct udp_table *udptable; 3059 3060 udptable = udp_get_table_seq(seq, seq_file_net(seq)); 3061 3062 if (state->bucket <= udptable->mask) 3063 spin_unlock_bh(&udptable->hash[state->bucket].lock); 3064 } 3065 EXPORT_SYMBOL(udp_seq_stop); 3066 3067 /* ------------------------------------------------------------------------ */ 3068 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3069 int bucket) 3070 { 3071 struct inet_sock *inet = inet_sk(sp); 3072 __be32 dest = inet->inet_daddr; 3073 __be32 src = inet->inet_rcv_saddr; 3074 __u16 destp = ntohs(inet->inet_dport); 3075 __u16 srcp = ntohs(inet->inet_sport); 3076 3077 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3078 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3079 bucket, src, srcp, dest, destp, sp->sk_state, 3080 sk_wmem_alloc_get(sp), 3081 udp_rqueue_get(sp), 3082 0, 0L, 0, 3083 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3084 0, sock_i_ino(sp), 3085 refcount_read(&sp->sk_refcnt), sp, 3086 atomic_read(&sp->sk_drops)); 3087 } 3088 3089 int udp4_seq_show(struct seq_file *seq, void *v) 3090 { 3091 seq_setwidth(seq, 127); 3092 if (v == SEQ_START_TOKEN) 3093 seq_puts(seq, " sl local_address rem_address st tx_queue " 3094 "rx_queue tr tm->when retrnsmt uid timeout " 3095 "inode ref pointer drops"); 3096 else { 3097 struct udp_iter_state *state = seq->private; 3098 3099 udp4_format_sock(v, seq, state->bucket); 3100 } 3101 seq_pad(seq, '\n'); 3102 return 0; 3103 } 3104 3105 #ifdef CONFIG_BPF_SYSCALL 3106 struct bpf_iter__udp { 3107 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3108 __bpf_md_ptr(struct udp_sock *, udp_sk); 3109 uid_t uid __aligned(8); 3110 int bucket __aligned(8); 3111 }; 3112 3113 struct bpf_udp_iter_state { 3114 struct udp_iter_state state; 3115 unsigned int cur_sk; 3116 unsigned int end_sk; 3117 unsigned int max_sk; 3118 int offset; 3119 struct sock **batch; 3120 bool st_bucket_done; 3121 }; 3122 3123 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3124 unsigned int new_batch_sz); 3125 static struct sock *bpf_iter_udp_batch(struct seq_file *seq) 3126 { 3127 struct bpf_udp_iter_state *iter = seq->private; 3128 struct udp_iter_state *state = &iter->state; 3129 struct net *net = seq_file_net(seq); 3130 struct udp_table *udptable; 3131 unsigned int batch_sks = 0; 3132 bool resized = false; 3133 struct sock *sk; 3134 3135 /* The current batch is done, so advance the bucket. */ 3136 if (iter->st_bucket_done) { 3137 state->bucket++; 3138 iter->offset = 0; 3139 } 3140 3141 udptable = udp_get_table_seq(seq, net); 3142 3143 again: 3144 /* New batch for the next bucket. 3145 * Iterate over the hash table to find a bucket with sockets matching 3146 * the iterator attributes, and return the first matching socket from 3147 * the bucket. The remaining matched sockets from the bucket are batched 3148 * before releasing the bucket lock. This allows BPF programs that are 3149 * called in seq_show to acquire the bucket lock if needed. 3150 */ 3151 iter->cur_sk = 0; 3152 iter->end_sk = 0; 3153 iter->st_bucket_done = false; 3154 batch_sks = 0; 3155 3156 for (; state->bucket <= udptable->mask; state->bucket++) { 3157 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket]; 3158 3159 if (hlist_empty(&hslot2->head)) { 3160 iter->offset = 0; 3161 continue; 3162 } 3163 3164 spin_lock_bh(&hslot2->lock); 3165 udp_portaddr_for_each_entry(sk, &hslot2->head) { 3166 if (seq_sk_match(seq, sk)) { 3167 /* Resume from the last iterated socket at the 3168 * offset in the bucket before iterator was stopped. 3169 */ 3170 if (iter->offset) { 3171 --iter->offset; 3172 continue; 3173 } 3174 if (iter->end_sk < iter->max_sk) { 3175 sock_hold(sk); 3176 iter->batch[iter->end_sk++] = sk; 3177 } 3178 batch_sks++; 3179 } 3180 } 3181 spin_unlock_bh(&hslot2->lock); 3182 3183 if (iter->end_sk) 3184 break; 3185 3186 /* Reset the current bucket's offset before moving to the next bucket. */ 3187 iter->offset = 0; 3188 } 3189 3190 /* All done: no batch made. */ 3191 if (!iter->end_sk) 3192 return NULL; 3193 3194 if (iter->end_sk == batch_sks) { 3195 /* Batching is done for the current bucket; return the first 3196 * socket to be iterated from the batch. 3197 */ 3198 iter->st_bucket_done = true; 3199 goto done; 3200 } 3201 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) { 3202 resized = true; 3203 /* After allocating a larger batch, retry one more time to grab 3204 * the whole bucket. 3205 */ 3206 state->bucket--; 3207 goto again; 3208 } 3209 done: 3210 return iter->batch[0]; 3211 } 3212 3213 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3214 { 3215 struct bpf_udp_iter_state *iter = seq->private; 3216 struct sock *sk; 3217 3218 /* Whenever seq_next() is called, the iter->cur_sk is 3219 * done with seq_show(), so unref the iter->cur_sk. 3220 */ 3221 if (iter->cur_sk < iter->end_sk) { 3222 sock_put(iter->batch[iter->cur_sk++]); 3223 ++iter->offset; 3224 } 3225 3226 /* After updating iter->cur_sk, check if there are more sockets 3227 * available in the current bucket batch. 3228 */ 3229 if (iter->cur_sk < iter->end_sk) 3230 sk = iter->batch[iter->cur_sk]; 3231 else 3232 /* Prepare a new batch. */ 3233 sk = bpf_iter_udp_batch(seq); 3234 3235 ++*pos; 3236 return sk; 3237 } 3238 3239 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos) 3240 { 3241 /* bpf iter does not support lseek, so it always 3242 * continue from where it was stop()-ped. 3243 */ 3244 if (*pos) 3245 return bpf_iter_udp_batch(seq); 3246 3247 return SEQ_START_TOKEN; 3248 } 3249 3250 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3251 struct udp_sock *udp_sk, uid_t uid, int bucket) 3252 { 3253 struct bpf_iter__udp ctx; 3254 3255 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3256 ctx.meta = meta; 3257 ctx.udp_sk = udp_sk; 3258 ctx.uid = uid; 3259 ctx.bucket = bucket; 3260 return bpf_iter_run_prog(prog, &ctx); 3261 } 3262 3263 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3264 { 3265 struct udp_iter_state *state = seq->private; 3266 struct bpf_iter_meta meta; 3267 struct bpf_prog *prog; 3268 struct sock *sk = v; 3269 uid_t uid; 3270 int ret; 3271 3272 if (v == SEQ_START_TOKEN) 3273 return 0; 3274 3275 lock_sock(sk); 3276 3277 if (unlikely(sk_unhashed(sk))) { 3278 ret = SEQ_SKIP; 3279 goto unlock; 3280 } 3281 3282 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3283 meta.seq = seq; 3284 prog = bpf_iter_get_info(&meta, false); 3285 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3286 3287 unlock: 3288 release_sock(sk); 3289 return ret; 3290 } 3291 3292 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter) 3293 { 3294 while (iter->cur_sk < iter->end_sk) 3295 sock_put(iter->batch[iter->cur_sk++]); 3296 } 3297 3298 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3299 { 3300 struct bpf_udp_iter_state *iter = seq->private; 3301 struct bpf_iter_meta meta; 3302 struct bpf_prog *prog; 3303 3304 if (!v) { 3305 meta.seq = seq; 3306 prog = bpf_iter_get_info(&meta, true); 3307 if (prog) 3308 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3309 } 3310 3311 if (iter->cur_sk < iter->end_sk) { 3312 bpf_iter_udp_put_batch(iter); 3313 iter->st_bucket_done = false; 3314 } 3315 } 3316 3317 static const struct seq_operations bpf_iter_udp_seq_ops = { 3318 .start = bpf_iter_udp_seq_start, 3319 .next = bpf_iter_udp_seq_next, 3320 .stop = bpf_iter_udp_seq_stop, 3321 .show = bpf_iter_udp_seq_show, 3322 }; 3323 #endif 3324 3325 static unsigned short seq_file_family(const struct seq_file *seq) 3326 { 3327 const struct udp_seq_afinfo *afinfo; 3328 3329 #ifdef CONFIG_BPF_SYSCALL 3330 /* BPF iterator: bpf programs to filter sockets. */ 3331 if (seq->op == &bpf_iter_udp_seq_ops) 3332 return AF_UNSPEC; 3333 #endif 3334 3335 /* Proc fs iterator */ 3336 afinfo = pde_data(file_inode(seq->file)); 3337 return afinfo->family; 3338 } 3339 3340 const struct seq_operations udp_seq_ops = { 3341 .start = udp_seq_start, 3342 .next = udp_seq_next, 3343 .stop = udp_seq_stop, 3344 .show = udp4_seq_show, 3345 }; 3346 EXPORT_SYMBOL(udp_seq_ops); 3347 3348 static struct udp_seq_afinfo udp4_seq_afinfo = { 3349 .family = AF_INET, 3350 .udp_table = NULL, 3351 }; 3352 3353 static int __net_init udp4_proc_init_net(struct net *net) 3354 { 3355 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3356 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3357 return -ENOMEM; 3358 return 0; 3359 } 3360 3361 static void __net_exit udp4_proc_exit_net(struct net *net) 3362 { 3363 remove_proc_entry("udp", net->proc_net); 3364 } 3365 3366 static struct pernet_operations udp4_net_ops = { 3367 .init = udp4_proc_init_net, 3368 .exit = udp4_proc_exit_net, 3369 }; 3370 3371 int __init udp4_proc_init(void) 3372 { 3373 return register_pernet_subsys(&udp4_net_ops); 3374 } 3375 3376 void udp4_proc_exit(void) 3377 { 3378 unregister_pernet_subsys(&udp4_net_ops); 3379 } 3380 #endif /* CONFIG_PROC_FS */ 3381 3382 static __initdata unsigned long uhash_entries; 3383 static int __init set_uhash_entries(char *str) 3384 { 3385 ssize_t ret; 3386 3387 if (!str) 3388 return 0; 3389 3390 ret = kstrtoul(str, 0, &uhash_entries); 3391 if (ret) 3392 return 0; 3393 3394 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3395 uhash_entries = UDP_HTABLE_SIZE_MIN; 3396 return 1; 3397 } 3398 __setup("uhash_entries=", set_uhash_entries); 3399 3400 void __init udp_table_init(struct udp_table *table, const char *name) 3401 { 3402 unsigned int i; 3403 3404 table->hash = alloc_large_system_hash(name, 3405 2 * sizeof(struct udp_hslot), 3406 uhash_entries, 3407 21, /* one slot per 2 MB */ 3408 0, 3409 &table->log, 3410 &table->mask, 3411 UDP_HTABLE_SIZE_MIN, 3412 UDP_HTABLE_SIZE_MAX); 3413 3414 table->hash2 = table->hash + (table->mask + 1); 3415 for (i = 0; i <= table->mask; i++) { 3416 INIT_HLIST_HEAD(&table->hash[i].head); 3417 table->hash[i].count = 0; 3418 spin_lock_init(&table->hash[i].lock); 3419 } 3420 for (i = 0; i <= table->mask; i++) { 3421 INIT_HLIST_HEAD(&table->hash2[i].head); 3422 table->hash2[i].count = 0; 3423 spin_lock_init(&table->hash2[i].lock); 3424 } 3425 } 3426 3427 u32 udp_flow_hashrnd(void) 3428 { 3429 static u32 hashrnd __read_mostly; 3430 3431 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3432 3433 return hashrnd; 3434 } 3435 EXPORT_SYMBOL(udp_flow_hashrnd); 3436 3437 static void __net_init udp_sysctl_init(struct net *net) 3438 { 3439 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE; 3440 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE; 3441 3442 #ifdef CONFIG_NET_L3_MASTER_DEV 3443 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3444 #endif 3445 } 3446 3447 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries) 3448 { 3449 struct udp_table *udptable; 3450 int i; 3451 3452 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL); 3453 if (!udptable) 3454 goto out; 3455 3456 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot), 3457 GFP_KERNEL_ACCOUNT); 3458 if (!udptable->hash) 3459 goto free_table; 3460 3461 udptable->hash2 = udptable->hash + hash_entries; 3462 udptable->mask = hash_entries - 1; 3463 udptable->log = ilog2(hash_entries); 3464 3465 for (i = 0; i < hash_entries; i++) { 3466 INIT_HLIST_HEAD(&udptable->hash[i].head); 3467 udptable->hash[i].count = 0; 3468 spin_lock_init(&udptable->hash[i].lock); 3469 3470 INIT_HLIST_HEAD(&udptable->hash2[i].head); 3471 udptable->hash2[i].count = 0; 3472 spin_lock_init(&udptable->hash2[i].lock); 3473 } 3474 3475 return udptable; 3476 3477 free_table: 3478 kfree(udptable); 3479 out: 3480 return NULL; 3481 } 3482 3483 static void __net_exit udp_pernet_table_free(struct net *net) 3484 { 3485 struct udp_table *udptable = net->ipv4.udp_table; 3486 3487 if (udptable == &udp_table) 3488 return; 3489 3490 kvfree(udptable->hash); 3491 kfree(udptable); 3492 } 3493 3494 static void __net_init udp_set_table(struct net *net) 3495 { 3496 struct udp_table *udptable; 3497 unsigned int hash_entries; 3498 struct net *old_net; 3499 3500 if (net_eq(net, &init_net)) 3501 goto fallback; 3502 3503 old_net = current->nsproxy->net_ns; 3504 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries); 3505 if (!hash_entries) 3506 goto fallback; 3507 3508 /* Set min to keep the bitmap on stack in udp_lib_get_port() */ 3509 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET) 3510 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET; 3511 else 3512 hash_entries = roundup_pow_of_two(hash_entries); 3513 3514 udptable = udp_pernet_table_alloc(hash_entries); 3515 if (udptable) { 3516 net->ipv4.udp_table = udptable; 3517 } else { 3518 pr_warn("Failed to allocate UDP hash table (entries: %u) " 3519 "for a netns, fallback to the global one\n", 3520 hash_entries); 3521 fallback: 3522 net->ipv4.udp_table = &udp_table; 3523 } 3524 } 3525 3526 static int __net_init udp_pernet_init(struct net *net) 3527 { 3528 udp_sysctl_init(net); 3529 udp_set_table(net); 3530 3531 return 0; 3532 } 3533 3534 static void __net_exit udp_pernet_exit(struct net *net) 3535 { 3536 udp_pernet_table_free(net); 3537 } 3538 3539 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3540 .init = udp_pernet_init, 3541 .exit = udp_pernet_exit, 3542 }; 3543 3544 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3545 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3546 struct udp_sock *udp_sk, uid_t uid, int bucket) 3547 3548 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter, 3549 unsigned int new_batch_sz) 3550 { 3551 struct sock **new_batch; 3552 3553 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch), 3554 GFP_USER | __GFP_NOWARN); 3555 if (!new_batch) 3556 return -ENOMEM; 3557 3558 bpf_iter_udp_put_batch(iter); 3559 kvfree(iter->batch); 3560 iter->batch = new_batch; 3561 iter->max_sk = new_batch_sz; 3562 3563 return 0; 3564 } 3565 3566 #define INIT_BATCH_SZ 16 3567 3568 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3569 { 3570 struct bpf_udp_iter_state *iter = priv_data; 3571 int ret; 3572 3573 ret = bpf_iter_init_seq_net(priv_data, aux); 3574 if (ret) 3575 return ret; 3576 3577 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ); 3578 if (ret) 3579 bpf_iter_fini_seq_net(priv_data); 3580 3581 return ret; 3582 } 3583 3584 static void bpf_iter_fini_udp(void *priv_data) 3585 { 3586 struct bpf_udp_iter_state *iter = priv_data; 3587 3588 bpf_iter_fini_seq_net(priv_data); 3589 kvfree(iter->batch); 3590 } 3591 3592 static const struct bpf_iter_seq_info udp_seq_info = { 3593 .seq_ops = &bpf_iter_udp_seq_ops, 3594 .init_seq_private = bpf_iter_init_udp, 3595 .fini_seq_private = bpf_iter_fini_udp, 3596 .seq_priv_size = sizeof(struct bpf_udp_iter_state), 3597 }; 3598 3599 static struct bpf_iter_reg udp_reg_info = { 3600 .target = "udp", 3601 .ctx_arg_info_size = 1, 3602 .ctx_arg_info = { 3603 { offsetof(struct bpf_iter__udp, udp_sk), 3604 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED }, 3605 }, 3606 .seq_info = &udp_seq_info, 3607 }; 3608 3609 static void __init bpf_iter_register(void) 3610 { 3611 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3612 if (bpf_iter_reg_target(&udp_reg_info)) 3613 pr_warn("Warning: could not register bpf iterator udp\n"); 3614 } 3615 #endif 3616 3617 void __init udp_init(void) 3618 { 3619 unsigned long limit; 3620 unsigned int i; 3621 3622 udp_table_init(&udp_table, "UDP"); 3623 limit = nr_free_buffer_pages() / 8; 3624 limit = max(limit, 128UL); 3625 sysctl_udp_mem[0] = limit / 4 * 3; 3626 sysctl_udp_mem[1] = limit; 3627 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3628 3629 /* 16 spinlocks per cpu */ 3630 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3631 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3632 GFP_KERNEL); 3633 if (!udp_busylocks) 3634 panic("UDP: failed to alloc udp_busylocks\n"); 3635 for (i = 0; i < (1U << udp_busylocks_log); i++) 3636 spin_lock_init(udp_busylocks + i); 3637 3638 if (register_pernet_subsys(&udp_sysctl_ops)) 3639 panic("UDP: failed to init sysctl parameters.\n"); 3640 3641 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3642 bpf_iter_register(); 3643 #endif 3644 } 3645