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/xfrm.h> 107 #include <trace/events/udp.h> 108 #include <linux/static_key.h> 109 #include <linux/btf_ids.h> 110 #include <trace/events/skb.h> 111 #include <net/busy_poll.h> 112 #include "udp_impl.h" 113 #include <net/sock_reuseport.h> 114 #include <net/addrconf.h> 115 #include <net/udp_tunnel.h> 116 #if IS_ENABLED(CONFIG_IPV6) 117 #include <net/ipv6_stubs.h> 118 #endif 119 120 struct udp_table udp_table __read_mostly; 121 EXPORT_SYMBOL(udp_table); 122 123 long sysctl_udp_mem[3] __read_mostly; 124 EXPORT_SYMBOL(sysctl_udp_mem); 125 126 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp; 127 EXPORT_SYMBOL(udp_memory_allocated); 128 129 #define MAX_UDP_PORTS 65536 130 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 131 132 static int udp_lib_lport_inuse(struct net *net, __u16 num, 133 const struct udp_hslot *hslot, 134 unsigned long *bitmap, 135 struct sock *sk, unsigned int log) 136 { 137 struct sock *sk2; 138 kuid_t uid = sock_i_uid(sk); 139 140 sk_for_each(sk2, &hslot->head) { 141 if (net_eq(sock_net(sk2), net) && 142 sk2 != sk && 143 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 144 (!sk2->sk_reuse || !sk->sk_reuse) && 145 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 146 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 147 inet_rcv_saddr_equal(sk, sk2, true)) { 148 if (sk2->sk_reuseport && sk->sk_reuseport && 149 !rcu_access_pointer(sk->sk_reuseport_cb) && 150 uid_eq(uid, sock_i_uid(sk2))) { 151 if (!bitmap) 152 return 0; 153 } else { 154 if (!bitmap) 155 return 1; 156 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 157 bitmap); 158 } 159 } 160 } 161 return 0; 162 } 163 164 /* 165 * Note: we still hold spinlock of primary hash chain, so no other writer 166 * can insert/delete a socket with local_port == num 167 */ 168 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 169 struct udp_hslot *hslot2, 170 struct sock *sk) 171 { 172 struct sock *sk2; 173 kuid_t uid = sock_i_uid(sk); 174 int res = 0; 175 176 spin_lock(&hslot2->lock); 177 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 178 if (net_eq(sock_net(sk2), net) && 179 sk2 != sk && 180 (udp_sk(sk2)->udp_port_hash == num) && 181 (!sk2->sk_reuse || !sk->sk_reuse) && 182 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 183 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 184 inet_rcv_saddr_equal(sk, sk2, true)) { 185 if (sk2->sk_reuseport && sk->sk_reuseport && 186 !rcu_access_pointer(sk->sk_reuseport_cb) && 187 uid_eq(uid, sock_i_uid(sk2))) { 188 res = 0; 189 } else { 190 res = 1; 191 } 192 break; 193 } 194 } 195 spin_unlock(&hslot2->lock); 196 return res; 197 } 198 199 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 200 { 201 struct net *net = sock_net(sk); 202 kuid_t uid = sock_i_uid(sk); 203 struct sock *sk2; 204 205 sk_for_each(sk2, &hslot->head) { 206 if (net_eq(sock_net(sk2), net) && 207 sk2 != sk && 208 sk2->sk_family == sk->sk_family && 209 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 210 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 211 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 212 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 213 inet_rcv_saddr_equal(sk, sk2, false)) { 214 return reuseport_add_sock(sk, sk2, 215 inet_rcv_saddr_any(sk)); 216 } 217 } 218 219 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 220 } 221 222 /** 223 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 224 * 225 * @sk: socket struct in question 226 * @snum: port number to look up 227 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 228 * with NULL address 229 */ 230 int udp_lib_get_port(struct sock *sk, unsigned short snum, 231 unsigned int hash2_nulladdr) 232 { 233 struct udp_hslot *hslot, *hslot2; 234 struct udp_table *udptable = sk->sk_prot->h.udp_table; 235 int error = 1; 236 struct net *net = sock_net(sk); 237 238 if (!snum) { 239 int low, high, remaining; 240 unsigned int rand; 241 unsigned short first, last; 242 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 243 244 inet_get_local_port_range(net, &low, &high); 245 remaining = (high - low) + 1; 246 247 rand = prandom_u32(); 248 first = reciprocal_scale(rand, remaining) + low; 249 /* 250 * force rand to be an odd multiple of UDP_HTABLE_SIZE 251 */ 252 rand = (rand | 1) * (udptable->mask + 1); 253 last = first + udptable->mask + 1; 254 do { 255 hslot = udp_hashslot(udptable, net, first); 256 bitmap_zero(bitmap, PORTS_PER_CHAIN); 257 spin_lock_bh(&hslot->lock); 258 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 259 udptable->log); 260 261 snum = first; 262 /* 263 * Iterate on all possible values of snum for this hash. 264 * Using steps of an odd multiple of UDP_HTABLE_SIZE 265 * give us randomization and full range coverage. 266 */ 267 do { 268 if (low <= snum && snum <= high && 269 !test_bit(snum >> udptable->log, bitmap) && 270 !inet_is_local_reserved_port(net, snum)) 271 goto found; 272 snum += rand; 273 } while (snum != first); 274 spin_unlock_bh(&hslot->lock); 275 cond_resched(); 276 } while (++first != last); 277 goto fail; 278 } else { 279 hslot = udp_hashslot(udptable, net, snum); 280 spin_lock_bh(&hslot->lock); 281 if (hslot->count > 10) { 282 int exist; 283 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 284 285 slot2 &= udptable->mask; 286 hash2_nulladdr &= udptable->mask; 287 288 hslot2 = udp_hashslot2(udptable, slot2); 289 if (hslot->count < hslot2->count) 290 goto scan_primary_hash; 291 292 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 293 if (!exist && (hash2_nulladdr != slot2)) { 294 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 295 exist = udp_lib_lport_inuse2(net, snum, hslot2, 296 sk); 297 } 298 if (exist) 299 goto fail_unlock; 300 else 301 goto found; 302 } 303 scan_primary_hash: 304 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 305 goto fail_unlock; 306 } 307 found: 308 inet_sk(sk)->inet_num = snum; 309 udp_sk(sk)->udp_port_hash = snum; 310 udp_sk(sk)->udp_portaddr_hash ^= snum; 311 if (sk_unhashed(sk)) { 312 if (sk->sk_reuseport && 313 udp_reuseport_add_sock(sk, hslot)) { 314 inet_sk(sk)->inet_num = 0; 315 udp_sk(sk)->udp_port_hash = 0; 316 udp_sk(sk)->udp_portaddr_hash ^= snum; 317 goto fail_unlock; 318 } 319 320 sk_add_node_rcu(sk, &hslot->head); 321 hslot->count++; 322 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 323 324 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 325 spin_lock(&hslot2->lock); 326 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 327 sk->sk_family == AF_INET6) 328 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 329 &hslot2->head); 330 else 331 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 332 &hslot2->head); 333 hslot2->count++; 334 spin_unlock(&hslot2->lock); 335 } 336 sock_set_flag(sk, SOCK_RCU_FREE); 337 error = 0; 338 fail_unlock: 339 spin_unlock_bh(&hslot->lock); 340 fail: 341 return error; 342 } 343 EXPORT_SYMBOL(udp_lib_get_port); 344 345 int udp_v4_get_port(struct sock *sk, unsigned short snum) 346 { 347 unsigned int hash2_nulladdr = 348 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 349 unsigned int hash2_partial = 350 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 351 352 /* precompute partial secondary hash */ 353 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 354 return udp_lib_get_port(sk, snum, hash2_nulladdr); 355 } 356 357 static int compute_score(struct sock *sk, struct net *net, 358 __be32 saddr, __be16 sport, 359 __be32 daddr, unsigned short hnum, 360 int dif, int sdif) 361 { 362 int score; 363 struct inet_sock *inet; 364 bool dev_match; 365 366 if (!net_eq(sock_net(sk), net) || 367 udp_sk(sk)->udp_port_hash != hnum || 368 ipv6_only_sock(sk)) 369 return -1; 370 371 if (sk->sk_rcv_saddr != daddr) 372 return -1; 373 374 score = (sk->sk_family == PF_INET) ? 2 : 1; 375 376 inet = inet_sk(sk); 377 if (inet->inet_daddr) { 378 if (inet->inet_daddr != saddr) 379 return -1; 380 score += 4; 381 } 382 383 if (inet->inet_dport) { 384 if (inet->inet_dport != sport) 385 return -1; 386 score += 4; 387 } 388 389 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 390 dif, sdif); 391 if (!dev_match) 392 return -1; 393 if (sk->sk_bound_dev_if) 394 score += 4; 395 396 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 397 score++; 398 return score; 399 } 400 401 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 402 const __u16 lport, const __be32 faddr, 403 const __be16 fport) 404 { 405 static u32 udp_ehash_secret __read_mostly; 406 407 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 408 409 return __inet_ehashfn(laddr, lport, faddr, fport, 410 udp_ehash_secret + net_hash_mix(net)); 411 } 412 413 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 414 struct sk_buff *skb, 415 __be32 saddr, __be16 sport, 416 __be32 daddr, unsigned short hnum) 417 { 418 struct sock *reuse_sk = NULL; 419 u32 hash; 420 421 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 422 hash = udp_ehashfn(net, daddr, hnum, saddr, sport); 423 reuse_sk = reuseport_select_sock(sk, hash, skb, 424 sizeof(struct udphdr)); 425 } 426 return reuse_sk; 427 } 428 429 /* called with rcu_read_lock() */ 430 static struct sock *udp4_lib_lookup2(struct net *net, 431 __be32 saddr, __be16 sport, 432 __be32 daddr, unsigned int hnum, 433 int dif, int sdif, 434 struct udp_hslot *hslot2, 435 struct sk_buff *skb) 436 { 437 struct sock *sk, *result; 438 int score, badness; 439 440 result = NULL; 441 badness = 0; 442 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 443 score = compute_score(sk, net, saddr, sport, 444 daddr, hnum, dif, sdif); 445 if (score > badness) { 446 result = lookup_reuseport(net, sk, skb, 447 saddr, sport, daddr, hnum); 448 /* Fall back to scoring if group has connections */ 449 if (result && !reuseport_has_conns(sk, false)) 450 return result; 451 452 result = result ? : sk; 453 badness = score; 454 } 455 } 456 return result; 457 } 458 459 static struct sock *udp4_lookup_run_bpf(struct net *net, 460 struct udp_table *udptable, 461 struct sk_buff *skb, 462 __be32 saddr, __be16 sport, 463 __be32 daddr, u16 hnum, const int dif) 464 { 465 struct sock *sk, *reuse_sk; 466 bool no_reuseport; 467 468 if (udptable != &udp_table) 469 return NULL; /* only UDP is supported */ 470 471 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, saddr, sport, 472 daddr, hnum, dif, &sk); 473 if (no_reuseport || IS_ERR_OR_NULL(sk)) 474 return sk; 475 476 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 477 if (reuse_sk) 478 sk = reuse_sk; 479 return sk; 480 } 481 482 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 483 * harder than this. -DaveM 484 */ 485 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 486 __be16 sport, __be32 daddr, __be16 dport, int dif, 487 int sdif, struct udp_table *udptable, struct sk_buff *skb) 488 { 489 unsigned short hnum = ntohs(dport); 490 unsigned int hash2, slot2; 491 struct udp_hslot *hslot2; 492 struct sock *result, *sk; 493 494 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 495 slot2 = hash2 & udptable->mask; 496 hslot2 = &udptable->hash2[slot2]; 497 498 /* Lookup connected or non-wildcard socket */ 499 result = udp4_lib_lookup2(net, saddr, sport, 500 daddr, hnum, dif, sdif, 501 hslot2, skb); 502 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 503 goto done; 504 505 /* Lookup redirect from BPF */ 506 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { 507 sk = udp4_lookup_run_bpf(net, udptable, skb, 508 saddr, sport, daddr, hnum, dif); 509 if (sk) { 510 result = sk; 511 goto done; 512 } 513 } 514 515 /* Got non-wildcard socket or error on first lookup */ 516 if (result) 517 goto done; 518 519 /* Lookup wildcard sockets */ 520 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 521 slot2 = hash2 & udptable->mask; 522 hslot2 = &udptable->hash2[slot2]; 523 524 result = udp4_lib_lookup2(net, saddr, sport, 525 htonl(INADDR_ANY), hnum, dif, sdif, 526 hslot2, skb); 527 done: 528 if (IS_ERR(result)) 529 return NULL; 530 return result; 531 } 532 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 533 534 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 535 __be16 sport, __be16 dport, 536 struct udp_table *udptable) 537 { 538 const struct iphdr *iph = ip_hdr(skb); 539 540 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 541 iph->daddr, dport, inet_iif(skb), 542 inet_sdif(skb), udptable, skb); 543 } 544 545 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 546 __be16 sport, __be16 dport) 547 { 548 const struct iphdr *iph = ip_hdr(skb); 549 550 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 551 iph->daddr, dport, inet_iif(skb), 552 inet_sdif(skb), &udp_table, NULL); 553 } 554 555 /* Must be called under rcu_read_lock(). 556 * Does increment socket refcount. 557 */ 558 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 559 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 560 __be32 daddr, __be16 dport, int dif) 561 { 562 struct sock *sk; 563 564 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 565 dif, 0, &udp_table, NULL); 566 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 567 sk = NULL; 568 return sk; 569 } 570 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 571 #endif 572 573 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 574 __be16 loc_port, __be32 loc_addr, 575 __be16 rmt_port, __be32 rmt_addr, 576 int dif, int sdif, unsigned short hnum) 577 { 578 struct inet_sock *inet = inet_sk(sk); 579 580 if (!net_eq(sock_net(sk), net) || 581 udp_sk(sk)->udp_port_hash != hnum || 582 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 583 (inet->inet_dport != rmt_port && inet->inet_dport) || 584 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 585 ipv6_only_sock(sk) || 586 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 587 return false; 588 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 589 return false; 590 return true; 591 } 592 593 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 594 void udp_encap_enable(void) 595 { 596 static_branch_inc(&udp_encap_needed_key); 597 } 598 EXPORT_SYMBOL(udp_encap_enable); 599 600 void udp_encap_disable(void) 601 { 602 static_branch_dec(&udp_encap_needed_key); 603 } 604 EXPORT_SYMBOL(udp_encap_disable); 605 606 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 607 * through error handlers in encapsulations looking for a match. 608 */ 609 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 610 { 611 int i; 612 613 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 614 int (*handler)(struct sk_buff *skb, u32 info); 615 const struct ip_tunnel_encap_ops *encap; 616 617 encap = rcu_dereference(iptun_encaps[i]); 618 if (!encap) 619 continue; 620 handler = encap->err_handler; 621 if (handler && !handler(skb, info)) 622 return 0; 623 } 624 625 return -ENOENT; 626 } 627 628 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 629 * reversing source and destination port: this will match tunnels that force the 630 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 631 * lwtunnels might actually break this assumption by being configured with 632 * different destination ports on endpoints, in this case we won't be able to 633 * trace ICMP messages back to them. 634 * 635 * If this doesn't match any socket, probe tunnels with arbitrary destination 636 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 637 * we've sent packets to won't necessarily match the local destination port. 638 * 639 * Then ask the tunnel implementation to match the error against a valid 640 * association. 641 * 642 * Return an error if we can't find a match, the socket if we need further 643 * processing, zero otherwise. 644 */ 645 static struct sock *__udp4_lib_err_encap(struct net *net, 646 const struct iphdr *iph, 647 struct udphdr *uh, 648 struct udp_table *udptable, 649 struct sock *sk, 650 struct sk_buff *skb, u32 info) 651 { 652 int (*lookup)(struct sock *sk, struct sk_buff *skb); 653 int network_offset, transport_offset; 654 struct udp_sock *up; 655 656 network_offset = skb_network_offset(skb); 657 transport_offset = skb_transport_offset(skb); 658 659 /* Network header needs to point to the outer IPv4 header inside ICMP */ 660 skb_reset_network_header(skb); 661 662 /* Transport header needs to point to the UDP header */ 663 skb_set_transport_header(skb, iph->ihl << 2); 664 665 if (sk) { 666 up = udp_sk(sk); 667 668 lookup = READ_ONCE(up->encap_err_lookup); 669 if (lookup && lookup(sk, skb)) 670 sk = NULL; 671 672 goto out; 673 } 674 675 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 676 iph->saddr, uh->dest, skb->dev->ifindex, 0, 677 udptable, NULL); 678 if (sk) { 679 up = udp_sk(sk); 680 681 lookup = READ_ONCE(up->encap_err_lookup); 682 if (!lookup || lookup(sk, skb)) 683 sk = NULL; 684 } 685 686 out: 687 if (!sk) 688 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 689 690 skb_set_transport_header(skb, transport_offset); 691 skb_set_network_header(skb, network_offset); 692 693 return sk; 694 } 695 696 /* 697 * This routine is called by the ICMP module when it gets some 698 * sort of error condition. If err < 0 then the socket should 699 * be closed and the error returned to the user. If err > 0 700 * it's just the icmp type << 8 | icmp code. 701 * Header points to the ip header of the error packet. We move 702 * on past this. Then (as it used to claim before adjustment) 703 * header points to the first 8 bytes of the udp header. We need 704 * to find the appropriate port. 705 */ 706 707 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 708 { 709 struct inet_sock *inet; 710 const struct iphdr *iph = (const struct iphdr *)skb->data; 711 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 712 const int type = icmp_hdr(skb)->type; 713 const int code = icmp_hdr(skb)->code; 714 bool tunnel = false; 715 struct sock *sk; 716 int harderr; 717 int err; 718 struct net *net = dev_net(skb->dev); 719 720 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 721 iph->saddr, uh->source, skb->dev->ifindex, 722 inet_sdif(skb), udptable, NULL); 723 724 if (!sk || udp_sk(sk)->encap_type) { 725 /* No socket for error: try tunnels before discarding */ 726 if (static_branch_unlikely(&udp_encap_needed_key)) { 727 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 728 info); 729 if (!sk) 730 return 0; 731 } else 732 sk = ERR_PTR(-ENOENT); 733 734 if (IS_ERR(sk)) { 735 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 736 return PTR_ERR(sk); 737 } 738 739 tunnel = true; 740 } 741 742 err = 0; 743 harderr = 0; 744 inet = inet_sk(sk); 745 746 switch (type) { 747 default: 748 case ICMP_TIME_EXCEEDED: 749 err = EHOSTUNREACH; 750 break; 751 case ICMP_SOURCE_QUENCH: 752 goto out; 753 case ICMP_PARAMETERPROB: 754 err = EPROTO; 755 harderr = 1; 756 break; 757 case ICMP_DEST_UNREACH: 758 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 759 ipv4_sk_update_pmtu(skb, sk, info); 760 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 761 err = EMSGSIZE; 762 harderr = 1; 763 break; 764 } 765 goto out; 766 } 767 err = EHOSTUNREACH; 768 if (code <= NR_ICMP_UNREACH) { 769 harderr = icmp_err_convert[code].fatal; 770 err = icmp_err_convert[code].errno; 771 } 772 break; 773 case ICMP_REDIRECT: 774 ipv4_sk_redirect(skb, sk); 775 goto out; 776 } 777 778 /* 779 * RFC1122: OK. Passes ICMP errors back to application, as per 780 * 4.1.3.3. 781 */ 782 if (tunnel) { 783 /* ...not for tunnels though: we don't have a sending socket */ 784 goto out; 785 } 786 if (!inet->recverr) { 787 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 788 goto out; 789 } else 790 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 791 792 sk->sk_err = err; 793 sk_error_report(sk); 794 out: 795 return 0; 796 } 797 798 int udp_err(struct sk_buff *skb, u32 info) 799 { 800 return __udp4_lib_err(skb, info, &udp_table); 801 } 802 803 /* 804 * Throw away all pending data and cancel the corking. Socket is locked. 805 */ 806 void udp_flush_pending_frames(struct sock *sk) 807 { 808 struct udp_sock *up = udp_sk(sk); 809 810 if (up->pending) { 811 up->len = 0; 812 up->pending = 0; 813 ip_flush_pending_frames(sk); 814 } 815 } 816 EXPORT_SYMBOL(udp_flush_pending_frames); 817 818 /** 819 * udp4_hwcsum - handle outgoing HW checksumming 820 * @skb: sk_buff containing the filled-in UDP header 821 * (checksum field must be zeroed out) 822 * @src: source IP address 823 * @dst: destination IP address 824 */ 825 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 826 { 827 struct udphdr *uh = udp_hdr(skb); 828 int offset = skb_transport_offset(skb); 829 int len = skb->len - offset; 830 int hlen = len; 831 __wsum csum = 0; 832 833 if (!skb_has_frag_list(skb)) { 834 /* 835 * Only one fragment on the socket. 836 */ 837 skb->csum_start = skb_transport_header(skb) - skb->head; 838 skb->csum_offset = offsetof(struct udphdr, check); 839 uh->check = ~csum_tcpudp_magic(src, dst, len, 840 IPPROTO_UDP, 0); 841 } else { 842 struct sk_buff *frags; 843 844 /* 845 * HW-checksum won't work as there are two or more 846 * fragments on the socket so that all csums of sk_buffs 847 * should be together 848 */ 849 skb_walk_frags(skb, frags) { 850 csum = csum_add(csum, frags->csum); 851 hlen -= frags->len; 852 } 853 854 csum = skb_checksum(skb, offset, hlen, csum); 855 skb->ip_summed = CHECKSUM_NONE; 856 857 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 858 if (uh->check == 0) 859 uh->check = CSUM_MANGLED_0; 860 } 861 } 862 EXPORT_SYMBOL_GPL(udp4_hwcsum); 863 864 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 865 * for the simple case like when setting the checksum for a UDP tunnel. 866 */ 867 void udp_set_csum(bool nocheck, struct sk_buff *skb, 868 __be32 saddr, __be32 daddr, int len) 869 { 870 struct udphdr *uh = udp_hdr(skb); 871 872 if (nocheck) { 873 uh->check = 0; 874 } else if (skb_is_gso(skb)) { 875 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 876 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 877 uh->check = 0; 878 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 879 if (uh->check == 0) 880 uh->check = CSUM_MANGLED_0; 881 } else { 882 skb->ip_summed = CHECKSUM_PARTIAL; 883 skb->csum_start = skb_transport_header(skb) - skb->head; 884 skb->csum_offset = offsetof(struct udphdr, check); 885 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 886 } 887 } 888 EXPORT_SYMBOL(udp_set_csum); 889 890 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 891 struct inet_cork *cork) 892 { 893 struct sock *sk = skb->sk; 894 struct inet_sock *inet = inet_sk(sk); 895 struct udphdr *uh; 896 int err; 897 int is_udplite = IS_UDPLITE(sk); 898 int offset = skb_transport_offset(skb); 899 int len = skb->len - offset; 900 int datalen = len - sizeof(*uh); 901 __wsum csum = 0; 902 903 /* 904 * Create a UDP header 905 */ 906 uh = udp_hdr(skb); 907 uh->source = inet->inet_sport; 908 uh->dest = fl4->fl4_dport; 909 uh->len = htons(len); 910 uh->check = 0; 911 912 if (cork->gso_size) { 913 const int hlen = skb_network_header_len(skb) + 914 sizeof(struct udphdr); 915 916 if (hlen + cork->gso_size > cork->fragsize) { 917 kfree_skb(skb); 918 return -EINVAL; 919 } 920 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 921 kfree_skb(skb); 922 return -EINVAL; 923 } 924 if (sk->sk_no_check_tx) { 925 kfree_skb(skb); 926 return -EINVAL; 927 } 928 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 929 dst_xfrm(skb_dst(skb))) { 930 kfree_skb(skb); 931 return -EIO; 932 } 933 934 if (datalen > cork->gso_size) { 935 skb_shinfo(skb)->gso_size = cork->gso_size; 936 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 937 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 938 cork->gso_size); 939 } 940 goto csum_partial; 941 } 942 943 if (is_udplite) /* UDP-Lite */ 944 csum = udplite_csum(skb); 945 946 else if (sk->sk_no_check_tx) { /* UDP csum off */ 947 948 skb->ip_summed = CHECKSUM_NONE; 949 goto send; 950 951 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 952 csum_partial: 953 954 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 955 goto send; 956 957 } else 958 csum = udp_csum(skb); 959 960 /* add protocol-dependent pseudo-header */ 961 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 962 sk->sk_protocol, csum); 963 if (uh->check == 0) 964 uh->check = CSUM_MANGLED_0; 965 966 send: 967 err = ip_send_skb(sock_net(sk), skb); 968 if (err) { 969 if (err == -ENOBUFS && !inet->recverr) { 970 UDP_INC_STATS(sock_net(sk), 971 UDP_MIB_SNDBUFERRORS, is_udplite); 972 err = 0; 973 } 974 } else 975 UDP_INC_STATS(sock_net(sk), 976 UDP_MIB_OUTDATAGRAMS, is_udplite); 977 return err; 978 } 979 980 /* 981 * Push out all pending data as one UDP datagram. Socket is locked. 982 */ 983 int udp_push_pending_frames(struct sock *sk) 984 { 985 struct udp_sock *up = udp_sk(sk); 986 struct inet_sock *inet = inet_sk(sk); 987 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 988 struct sk_buff *skb; 989 int err = 0; 990 991 skb = ip_finish_skb(sk, fl4); 992 if (!skb) 993 goto out; 994 995 err = udp_send_skb(skb, fl4, &inet->cork.base); 996 997 out: 998 up->len = 0; 999 up->pending = 0; 1000 return err; 1001 } 1002 EXPORT_SYMBOL(udp_push_pending_frames); 1003 1004 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1005 { 1006 switch (cmsg->cmsg_type) { 1007 case UDP_SEGMENT: 1008 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1009 return -EINVAL; 1010 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1011 return 0; 1012 default: 1013 return -EINVAL; 1014 } 1015 } 1016 1017 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1018 { 1019 struct cmsghdr *cmsg; 1020 bool need_ip = false; 1021 int err; 1022 1023 for_each_cmsghdr(cmsg, msg) { 1024 if (!CMSG_OK(msg, cmsg)) 1025 return -EINVAL; 1026 1027 if (cmsg->cmsg_level != SOL_UDP) { 1028 need_ip = true; 1029 continue; 1030 } 1031 1032 err = __udp_cmsg_send(cmsg, gso_size); 1033 if (err) 1034 return err; 1035 } 1036 1037 return need_ip; 1038 } 1039 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1040 1041 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1042 { 1043 struct inet_sock *inet = inet_sk(sk); 1044 struct udp_sock *up = udp_sk(sk); 1045 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1046 struct flowi4 fl4_stack; 1047 struct flowi4 *fl4; 1048 int ulen = len; 1049 struct ipcm_cookie ipc; 1050 struct rtable *rt = NULL; 1051 int free = 0; 1052 int connected = 0; 1053 __be32 daddr, faddr, saddr; 1054 __be16 dport; 1055 u8 tos; 1056 int err, is_udplite = IS_UDPLITE(sk); 1057 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1058 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1059 struct sk_buff *skb; 1060 struct ip_options_data opt_copy; 1061 1062 if (len > 0xFFFF) 1063 return -EMSGSIZE; 1064 1065 /* 1066 * Check the flags. 1067 */ 1068 1069 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1070 return -EOPNOTSUPP; 1071 1072 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1073 1074 fl4 = &inet->cork.fl.u.ip4; 1075 if (up->pending) { 1076 /* 1077 * There are pending frames. 1078 * The socket lock must be held while it's corked. 1079 */ 1080 lock_sock(sk); 1081 if (likely(up->pending)) { 1082 if (unlikely(up->pending != AF_INET)) { 1083 release_sock(sk); 1084 return -EINVAL; 1085 } 1086 goto do_append_data; 1087 } 1088 release_sock(sk); 1089 } 1090 ulen += sizeof(struct udphdr); 1091 1092 /* 1093 * Get and verify the address. 1094 */ 1095 if (usin) { 1096 if (msg->msg_namelen < sizeof(*usin)) 1097 return -EINVAL; 1098 if (usin->sin_family != AF_INET) { 1099 if (usin->sin_family != AF_UNSPEC) 1100 return -EAFNOSUPPORT; 1101 } 1102 1103 daddr = usin->sin_addr.s_addr; 1104 dport = usin->sin_port; 1105 if (dport == 0) 1106 return -EINVAL; 1107 } else { 1108 if (sk->sk_state != TCP_ESTABLISHED) 1109 return -EDESTADDRREQ; 1110 daddr = inet->inet_daddr; 1111 dport = inet->inet_dport; 1112 /* Open fast path for connected socket. 1113 Route will not be used, if at least one option is set. 1114 */ 1115 connected = 1; 1116 } 1117 1118 ipcm_init_sk(&ipc, inet); 1119 ipc.gso_size = READ_ONCE(up->gso_size); 1120 1121 if (msg->msg_controllen) { 1122 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1123 if (err > 0) 1124 err = ip_cmsg_send(sk, msg, &ipc, 1125 sk->sk_family == AF_INET6); 1126 if (unlikely(err < 0)) { 1127 kfree(ipc.opt); 1128 return err; 1129 } 1130 if (ipc.opt) 1131 free = 1; 1132 connected = 0; 1133 } 1134 if (!ipc.opt) { 1135 struct ip_options_rcu *inet_opt; 1136 1137 rcu_read_lock(); 1138 inet_opt = rcu_dereference(inet->inet_opt); 1139 if (inet_opt) { 1140 memcpy(&opt_copy, inet_opt, 1141 sizeof(*inet_opt) + inet_opt->opt.optlen); 1142 ipc.opt = &opt_copy.opt; 1143 } 1144 rcu_read_unlock(); 1145 } 1146 1147 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1148 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1149 (struct sockaddr *)usin, &ipc.addr); 1150 if (err) 1151 goto out_free; 1152 if (usin) { 1153 if (usin->sin_port == 0) { 1154 /* BPF program set invalid port. Reject it. */ 1155 err = -EINVAL; 1156 goto out_free; 1157 } 1158 daddr = usin->sin_addr.s_addr; 1159 dport = usin->sin_port; 1160 } 1161 } 1162 1163 saddr = ipc.addr; 1164 ipc.addr = faddr = daddr; 1165 1166 if (ipc.opt && ipc.opt->opt.srr) { 1167 if (!daddr) { 1168 err = -EINVAL; 1169 goto out_free; 1170 } 1171 faddr = ipc.opt->opt.faddr; 1172 connected = 0; 1173 } 1174 tos = get_rttos(&ipc, inet); 1175 if (sock_flag(sk, SOCK_LOCALROUTE) || 1176 (msg->msg_flags & MSG_DONTROUTE) || 1177 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1178 tos |= RTO_ONLINK; 1179 connected = 0; 1180 } 1181 1182 if (ipv4_is_multicast(daddr)) { 1183 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1184 ipc.oif = inet->mc_index; 1185 if (!saddr) 1186 saddr = inet->mc_addr; 1187 connected = 0; 1188 } else if (!ipc.oif) { 1189 ipc.oif = inet->uc_index; 1190 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1191 /* oif is set, packet is to local broadcast and 1192 * uc_index is set. oif is most likely set 1193 * by sk_bound_dev_if. If uc_index != oif check if the 1194 * oif is an L3 master and uc_index is an L3 slave. 1195 * If so, we want to allow the send using the uc_index. 1196 */ 1197 if (ipc.oif != inet->uc_index && 1198 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1199 inet->uc_index)) { 1200 ipc.oif = inet->uc_index; 1201 } 1202 } 1203 1204 if (connected) 1205 rt = (struct rtable *)sk_dst_check(sk, 0); 1206 1207 if (!rt) { 1208 struct net *net = sock_net(sk); 1209 __u8 flow_flags = inet_sk_flowi_flags(sk); 1210 1211 fl4 = &fl4_stack; 1212 1213 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, 1214 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1215 flow_flags, 1216 faddr, saddr, dport, inet->inet_sport, 1217 sk->sk_uid); 1218 1219 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1220 rt = ip_route_output_flow(net, fl4, sk); 1221 if (IS_ERR(rt)) { 1222 err = PTR_ERR(rt); 1223 rt = NULL; 1224 if (err == -ENETUNREACH) 1225 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1226 goto out; 1227 } 1228 1229 err = -EACCES; 1230 if ((rt->rt_flags & RTCF_BROADCAST) && 1231 !sock_flag(sk, SOCK_BROADCAST)) 1232 goto out; 1233 if (connected) 1234 sk_dst_set(sk, dst_clone(&rt->dst)); 1235 } 1236 1237 if (msg->msg_flags&MSG_CONFIRM) 1238 goto do_confirm; 1239 back_from_confirm: 1240 1241 saddr = fl4->saddr; 1242 if (!ipc.addr) 1243 daddr = ipc.addr = fl4->daddr; 1244 1245 /* Lockless fast path for the non-corking case. */ 1246 if (!corkreq) { 1247 struct inet_cork cork; 1248 1249 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1250 sizeof(struct udphdr), &ipc, &rt, 1251 &cork, msg->msg_flags); 1252 err = PTR_ERR(skb); 1253 if (!IS_ERR_OR_NULL(skb)) 1254 err = udp_send_skb(skb, fl4, &cork); 1255 goto out; 1256 } 1257 1258 lock_sock(sk); 1259 if (unlikely(up->pending)) { 1260 /* The socket is already corked while preparing it. */ 1261 /* ... which is an evident application bug. --ANK */ 1262 release_sock(sk); 1263 1264 net_dbg_ratelimited("socket already corked\n"); 1265 err = -EINVAL; 1266 goto out; 1267 } 1268 /* 1269 * Now cork the socket to pend data. 1270 */ 1271 fl4 = &inet->cork.fl.u.ip4; 1272 fl4->daddr = daddr; 1273 fl4->saddr = saddr; 1274 fl4->fl4_dport = dport; 1275 fl4->fl4_sport = inet->inet_sport; 1276 up->pending = AF_INET; 1277 1278 do_append_data: 1279 up->len += ulen; 1280 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1281 sizeof(struct udphdr), &ipc, &rt, 1282 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1283 if (err) 1284 udp_flush_pending_frames(sk); 1285 else if (!corkreq) 1286 err = udp_push_pending_frames(sk); 1287 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1288 up->pending = 0; 1289 release_sock(sk); 1290 1291 out: 1292 ip_rt_put(rt); 1293 out_free: 1294 if (free) 1295 kfree(ipc.opt); 1296 if (!err) 1297 return len; 1298 /* 1299 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1300 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1301 * we don't have a good statistic (IpOutDiscards but it can be too many 1302 * things). We could add another new stat but at least for now that 1303 * seems like overkill. 1304 */ 1305 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1306 UDP_INC_STATS(sock_net(sk), 1307 UDP_MIB_SNDBUFERRORS, is_udplite); 1308 } 1309 return err; 1310 1311 do_confirm: 1312 if (msg->msg_flags & MSG_PROBE) 1313 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1314 if (!(msg->msg_flags&MSG_PROBE) || len) 1315 goto back_from_confirm; 1316 err = 0; 1317 goto out; 1318 } 1319 EXPORT_SYMBOL(udp_sendmsg); 1320 1321 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1322 size_t size, int flags) 1323 { 1324 struct inet_sock *inet = inet_sk(sk); 1325 struct udp_sock *up = udp_sk(sk); 1326 int ret; 1327 1328 if (flags & MSG_SENDPAGE_NOTLAST) 1329 flags |= MSG_MORE; 1330 1331 if (!up->pending) { 1332 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1333 1334 /* Call udp_sendmsg to specify destination address which 1335 * sendpage interface can't pass. 1336 * This will succeed only when the socket is connected. 1337 */ 1338 ret = udp_sendmsg(sk, &msg, 0); 1339 if (ret < 0) 1340 return ret; 1341 } 1342 1343 lock_sock(sk); 1344 1345 if (unlikely(!up->pending)) { 1346 release_sock(sk); 1347 1348 net_dbg_ratelimited("cork failed\n"); 1349 return -EINVAL; 1350 } 1351 1352 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1353 page, offset, size, flags); 1354 if (ret == -EOPNOTSUPP) { 1355 release_sock(sk); 1356 return sock_no_sendpage(sk->sk_socket, page, offset, 1357 size, flags); 1358 } 1359 if (ret < 0) { 1360 udp_flush_pending_frames(sk); 1361 goto out; 1362 } 1363 1364 up->len += size; 1365 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE))) 1366 ret = udp_push_pending_frames(sk); 1367 if (!ret) 1368 ret = size; 1369 out: 1370 release_sock(sk); 1371 return ret; 1372 } 1373 1374 #define UDP_SKB_IS_STATELESS 0x80000000 1375 1376 /* all head states (dst, sk, nf conntrack) except skb extensions are 1377 * cleared by udp_rcv(). 1378 * 1379 * We need to preserve secpath, if present, to eventually process 1380 * IP_CMSG_PASSSEC at recvmsg() time. 1381 * 1382 * Other extensions can be cleared. 1383 */ 1384 static bool udp_try_make_stateless(struct sk_buff *skb) 1385 { 1386 if (!skb_has_extensions(skb)) 1387 return true; 1388 1389 if (!secpath_exists(skb)) { 1390 skb_ext_reset(skb); 1391 return true; 1392 } 1393 1394 return false; 1395 } 1396 1397 static void udp_set_dev_scratch(struct sk_buff *skb) 1398 { 1399 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1400 1401 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1402 scratch->_tsize_state = skb->truesize; 1403 #if BITS_PER_LONG == 64 1404 scratch->len = skb->len; 1405 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1406 scratch->is_linear = !skb_is_nonlinear(skb); 1407 #endif 1408 if (udp_try_make_stateless(skb)) 1409 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1410 } 1411 1412 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1413 { 1414 /* We come here after udp_lib_checksum_complete() returned 0. 1415 * This means that __skb_checksum_complete() might have 1416 * set skb->csum_valid to 1. 1417 * On 64bit platforms, we can set csum_unnecessary 1418 * to true, but only if the skb is not shared. 1419 */ 1420 #if BITS_PER_LONG == 64 1421 if (!skb_shared(skb)) 1422 udp_skb_scratch(skb)->csum_unnecessary = true; 1423 #endif 1424 } 1425 1426 static int udp_skb_truesize(struct sk_buff *skb) 1427 { 1428 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1429 } 1430 1431 static bool udp_skb_has_head_state(struct sk_buff *skb) 1432 { 1433 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1434 } 1435 1436 /* fully reclaim rmem/fwd memory allocated for skb */ 1437 static void udp_rmem_release(struct sock *sk, int size, int partial, 1438 bool rx_queue_lock_held) 1439 { 1440 struct udp_sock *up = udp_sk(sk); 1441 struct sk_buff_head *sk_queue; 1442 int amt; 1443 1444 if (likely(partial)) { 1445 up->forward_deficit += size; 1446 size = up->forward_deficit; 1447 if (size < (sk->sk_rcvbuf >> 2) && 1448 !skb_queue_empty(&up->reader_queue)) 1449 return; 1450 } else { 1451 size += up->forward_deficit; 1452 } 1453 up->forward_deficit = 0; 1454 1455 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1456 * if the called don't held it already 1457 */ 1458 sk_queue = &sk->sk_receive_queue; 1459 if (!rx_queue_lock_held) 1460 spin_lock(&sk_queue->lock); 1461 1462 1463 sk->sk_forward_alloc += size; 1464 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); 1465 sk->sk_forward_alloc -= amt; 1466 1467 if (amt) 1468 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); 1469 1470 atomic_sub(size, &sk->sk_rmem_alloc); 1471 1472 /* this can save us from acquiring the rx queue lock on next receive */ 1473 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1474 1475 if (!rx_queue_lock_held) 1476 spin_unlock(&sk_queue->lock); 1477 } 1478 1479 /* Note: called with reader_queue.lock held. 1480 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1481 * This avoids a cache line miss while receive_queue lock is held. 1482 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1483 */ 1484 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1485 { 1486 prefetch(&skb->data); 1487 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1488 } 1489 EXPORT_SYMBOL(udp_skb_destructor); 1490 1491 /* as above, but the caller held the rx queue lock, too */ 1492 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1493 { 1494 prefetch(&skb->data); 1495 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1496 } 1497 1498 /* Idea of busylocks is to let producers grab an extra spinlock 1499 * to relieve pressure on the receive_queue spinlock shared by consumer. 1500 * Under flood, this means that only one producer can be in line 1501 * trying to acquire the receive_queue spinlock. 1502 * These busylock can be allocated on a per cpu manner, instead of a 1503 * per socket one (that would consume a cache line per socket) 1504 */ 1505 static int udp_busylocks_log __read_mostly; 1506 static spinlock_t *udp_busylocks __read_mostly; 1507 1508 static spinlock_t *busylock_acquire(void *ptr) 1509 { 1510 spinlock_t *busy; 1511 1512 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1513 spin_lock(busy); 1514 return busy; 1515 } 1516 1517 static void busylock_release(spinlock_t *busy) 1518 { 1519 if (busy) 1520 spin_unlock(busy); 1521 } 1522 1523 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1524 { 1525 struct sk_buff_head *list = &sk->sk_receive_queue; 1526 int rmem, delta, amt, err = -ENOMEM; 1527 spinlock_t *busy = NULL; 1528 int size; 1529 1530 /* try to avoid the costly atomic add/sub pair when the receive 1531 * queue is full; always allow at least a packet 1532 */ 1533 rmem = atomic_read(&sk->sk_rmem_alloc); 1534 if (rmem > sk->sk_rcvbuf) 1535 goto drop; 1536 1537 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1538 * having linear skbs : 1539 * - Reduce memory overhead and thus increase receive queue capacity 1540 * - Less cache line misses at copyout() time 1541 * - Less work at consume_skb() (less alien page frag freeing) 1542 */ 1543 if (rmem > (sk->sk_rcvbuf >> 1)) { 1544 skb_condense(skb); 1545 1546 busy = busylock_acquire(sk); 1547 } 1548 size = skb->truesize; 1549 udp_set_dev_scratch(skb); 1550 1551 /* we drop only if the receive buf is full and the receive 1552 * queue contains some other skb 1553 */ 1554 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1555 if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) 1556 goto uncharge_drop; 1557 1558 spin_lock(&list->lock); 1559 if (size >= sk->sk_forward_alloc) { 1560 amt = sk_mem_pages(size); 1561 delta = amt << SK_MEM_QUANTUM_SHIFT; 1562 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1563 err = -ENOBUFS; 1564 spin_unlock(&list->lock); 1565 goto uncharge_drop; 1566 } 1567 1568 sk->sk_forward_alloc += delta; 1569 } 1570 1571 sk->sk_forward_alloc -= size; 1572 1573 /* no need to setup a destructor, we will explicitly release the 1574 * forward allocated memory on dequeue 1575 */ 1576 sock_skb_set_dropcount(sk, skb); 1577 1578 __skb_queue_tail(list, skb); 1579 spin_unlock(&list->lock); 1580 1581 if (!sock_flag(sk, SOCK_DEAD)) 1582 sk->sk_data_ready(sk); 1583 1584 busylock_release(busy); 1585 return 0; 1586 1587 uncharge_drop: 1588 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1589 1590 drop: 1591 atomic_inc(&sk->sk_drops); 1592 busylock_release(busy); 1593 return err; 1594 } 1595 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1596 1597 void udp_destruct_sock(struct sock *sk) 1598 { 1599 /* reclaim completely the forward allocated memory */ 1600 struct udp_sock *up = udp_sk(sk); 1601 unsigned int total = 0; 1602 struct sk_buff *skb; 1603 1604 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1605 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1606 total += skb->truesize; 1607 kfree_skb(skb); 1608 } 1609 udp_rmem_release(sk, total, 0, true); 1610 1611 inet_sock_destruct(sk); 1612 } 1613 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1614 1615 int udp_init_sock(struct sock *sk) 1616 { 1617 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1618 sk->sk_destruct = udp_destruct_sock; 1619 return 0; 1620 } 1621 EXPORT_SYMBOL_GPL(udp_init_sock); 1622 1623 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1624 { 1625 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1626 bool slow = lock_sock_fast(sk); 1627 1628 sk_peek_offset_bwd(sk, len); 1629 unlock_sock_fast(sk, slow); 1630 } 1631 1632 if (!skb_unref(skb)) 1633 return; 1634 1635 /* In the more common cases we cleared the head states previously, 1636 * see __udp_queue_rcv_skb(). 1637 */ 1638 if (unlikely(udp_skb_has_head_state(skb))) 1639 skb_release_head_state(skb); 1640 __consume_stateless_skb(skb); 1641 } 1642 EXPORT_SYMBOL_GPL(skb_consume_udp); 1643 1644 static struct sk_buff *__first_packet_length(struct sock *sk, 1645 struct sk_buff_head *rcvq, 1646 int *total) 1647 { 1648 struct sk_buff *skb; 1649 1650 while ((skb = skb_peek(rcvq)) != NULL) { 1651 if (udp_lib_checksum_complete(skb)) { 1652 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1653 IS_UDPLITE(sk)); 1654 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1655 IS_UDPLITE(sk)); 1656 atomic_inc(&sk->sk_drops); 1657 __skb_unlink(skb, rcvq); 1658 *total += skb->truesize; 1659 kfree_skb(skb); 1660 } else { 1661 udp_skb_csum_unnecessary_set(skb); 1662 break; 1663 } 1664 } 1665 return skb; 1666 } 1667 1668 /** 1669 * first_packet_length - return length of first packet in receive queue 1670 * @sk: socket 1671 * 1672 * Drops all bad checksum frames, until a valid one is found. 1673 * Returns the length of found skb, or -1 if none is found. 1674 */ 1675 static int first_packet_length(struct sock *sk) 1676 { 1677 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1678 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1679 struct sk_buff *skb; 1680 int total = 0; 1681 int res; 1682 1683 spin_lock_bh(&rcvq->lock); 1684 skb = __first_packet_length(sk, rcvq, &total); 1685 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1686 spin_lock(&sk_queue->lock); 1687 skb_queue_splice_tail_init(sk_queue, rcvq); 1688 spin_unlock(&sk_queue->lock); 1689 1690 skb = __first_packet_length(sk, rcvq, &total); 1691 } 1692 res = skb ? skb->len : -1; 1693 if (total) 1694 udp_rmem_release(sk, total, 1, false); 1695 spin_unlock_bh(&rcvq->lock); 1696 return res; 1697 } 1698 1699 /* 1700 * IOCTL requests applicable to the UDP protocol 1701 */ 1702 1703 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1704 { 1705 switch (cmd) { 1706 case SIOCOUTQ: 1707 { 1708 int amount = sk_wmem_alloc_get(sk); 1709 1710 return put_user(amount, (int __user *)arg); 1711 } 1712 1713 case SIOCINQ: 1714 { 1715 int amount = max_t(int, 0, first_packet_length(sk)); 1716 1717 return put_user(amount, (int __user *)arg); 1718 } 1719 1720 default: 1721 return -ENOIOCTLCMD; 1722 } 1723 1724 return 0; 1725 } 1726 EXPORT_SYMBOL(udp_ioctl); 1727 1728 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1729 int noblock, int *off, int *err) 1730 { 1731 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1732 struct sk_buff_head *queue; 1733 struct sk_buff *last; 1734 long timeo; 1735 int error; 1736 1737 queue = &udp_sk(sk)->reader_queue; 1738 flags |= noblock ? MSG_DONTWAIT : 0; 1739 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1740 do { 1741 struct sk_buff *skb; 1742 1743 error = sock_error(sk); 1744 if (error) 1745 break; 1746 1747 error = -EAGAIN; 1748 do { 1749 spin_lock_bh(&queue->lock); 1750 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1751 err, &last); 1752 if (skb) { 1753 if (!(flags & MSG_PEEK)) 1754 udp_skb_destructor(sk, skb); 1755 spin_unlock_bh(&queue->lock); 1756 return skb; 1757 } 1758 1759 if (skb_queue_empty_lockless(sk_queue)) { 1760 spin_unlock_bh(&queue->lock); 1761 goto busy_check; 1762 } 1763 1764 /* refill the reader queue and walk it again 1765 * keep both queues locked to avoid re-acquiring 1766 * the sk_receive_queue lock if fwd memory scheduling 1767 * is needed. 1768 */ 1769 spin_lock(&sk_queue->lock); 1770 skb_queue_splice_tail_init(sk_queue, queue); 1771 1772 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1773 err, &last); 1774 if (skb && !(flags & MSG_PEEK)) 1775 udp_skb_dtor_locked(sk, skb); 1776 spin_unlock(&sk_queue->lock); 1777 spin_unlock_bh(&queue->lock); 1778 if (skb) 1779 return skb; 1780 1781 busy_check: 1782 if (!sk_can_busy_loop(sk)) 1783 break; 1784 1785 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1786 } while (!skb_queue_empty_lockless(sk_queue)); 1787 1788 /* sk_queue is empty, reader_queue may contain peeked packets */ 1789 } while (timeo && 1790 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1791 &error, &timeo, 1792 (struct sk_buff *)sk_queue)); 1793 1794 *err = error; 1795 return NULL; 1796 } 1797 EXPORT_SYMBOL(__skb_recv_udp); 1798 1799 int udp_read_sock(struct sock *sk, read_descriptor_t *desc, 1800 sk_read_actor_t recv_actor) 1801 { 1802 int copied = 0; 1803 1804 while (1) { 1805 struct sk_buff *skb; 1806 int err, used; 1807 1808 skb = skb_recv_udp(sk, 0, 1, &err); 1809 if (!skb) 1810 return err; 1811 1812 if (udp_lib_checksum_complete(skb)) { 1813 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1814 IS_UDPLITE(sk)); 1815 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1816 IS_UDPLITE(sk)); 1817 atomic_inc(&sk->sk_drops); 1818 kfree_skb(skb); 1819 continue; 1820 } 1821 1822 used = recv_actor(desc, skb, 0, skb->len); 1823 if (used <= 0) { 1824 if (!copied) 1825 copied = used; 1826 kfree_skb(skb); 1827 break; 1828 } else if (used <= skb->len) { 1829 copied += used; 1830 } 1831 1832 kfree_skb(skb); 1833 if (!desc->count) 1834 break; 1835 } 1836 1837 return copied; 1838 } 1839 EXPORT_SYMBOL(udp_read_sock); 1840 1841 /* 1842 * This should be easy, if there is something there we 1843 * return it, otherwise we block. 1844 */ 1845 1846 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1847 int flags, int *addr_len) 1848 { 1849 struct inet_sock *inet = inet_sk(sk); 1850 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1851 struct sk_buff *skb; 1852 unsigned int ulen, copied; 1853 int off, err, peeking = flags & MSG_PEEK; 1854 int is_udplite = IS_UDPLITE(sk); 1855 bool checksum_valid = false; 1856 1857 if (flags & MSG_ERRQUEUE) 1858 return ip_recv_error(sk, msg, len, addr_len); 1859 1860 try_again: 1861 off = sk_peek_offset(sk, flags); 1862 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 1863 if (!skb) 1864 return err; 1865 1866 ulen = udp_skb_len(skb); 1867 copied = len; 1868 if (copied > ulen - off) 1869 copied = ulen - off; 1870 else if (copied < ulen) 1871 msg->msg_flags |= MSG_TRUNC; 1872 1873 /* 1874 * If checksum is needed at all, try to do it while copying the 1875 * data. If the data is truncated, or if we only want a partial 1876 * coverage checksum (UDP-Lite), do it before the copy. 1877 */ 1878 1879 if (copied < ulen || peeking || 1880 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1881 checksum_valid = udp_skb_csum_unnecessary(skb) || 1882 !__udp_lib_checksum_complete(skb); 1883 if (!checksum_valid) 1884 goto csum_copy_err; 1885 } 1886 1887 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1888 if (udp_skb_is_linear(skb)) 1889 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1890 else 1891 err = skb_copy_datagram_msg(skb, off, msg, copied); 1892 } else { 1893 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1894 1895 if (err == -EINVAL) 1896 goto csum_copy_err; 1897 } 1898 1899 if (unlikely(err)) { 1900 if (!peeking) { 1901 atomic_inc(&sk->sk_drops); 1902 UDP_INC_STATS(sock_net(sk), 1903 UDP_MIB_INERRORS, is_udplite); 1904 } 1905 kfree_skb(skb); 1906 return err; 1907 } 1908 1909 if (!peeking) 1910 UDP_INC_STATS(sock_net(sk), 1911 UDP_MIB_INDATAGRAMS, is_udplite); 1912 1913 sock_recv_ts_and_drops(msg, sk, skb); 1914 1915 /* Copy the address. */ 1916 if (sin) { 1917 sin->sin_family = AF_INET; 1918 sin->sin_port = udp_hdr(skb)->source; 1919 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1920 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1921 *addr_len = sizeof(*sin); 1922 1923 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1924 (struct sockaddr *)sin); 1925 } 1926 1927 if (udp_sk(sk)->gro_enabled) 1928 udp_cmsg_recv(msg, sk, skb); 1929 1930 if (inet->cmsg_flags) 1931 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1932 1933 err = copied; 1934 if (flags & MSG_TRUNC) 1935 err = ulen; 1936 1937 skb_consume_udp(sk, skb, peeking ? -err : err); 1938 return err; 1939 1940 csum_copy_err: 1941 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1942 udp_skb_destructor)) { 1943 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1944 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1945 } 1946 kfree_skb(skb); 1947 1948 /* starting over for a new packet, but check if we need to yield */ 1949 cond_resched(); 1950 msg->msg_flags &= ~MSG_TRUNC; 1951 goto try_again; 1952 } 1953 1954 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1955 { 1956 /* This check is replicated from __ip4_datagram_connect() and 1957 * intended to prevent BPF program called below from accessing bytes 1958 * that are out of the bound specified by user in addr_len. 1959 */ 1960 if (addr_len < sizeof(struct sockaddr_in)) 1961 return -EINVAL; 1962 1963 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1964 } 1965 EXPORT_SYMBOL(udp_pre_connect); 1966 1967 int __udp_disconnect(struct sock *sk, int flags) 1968 { 1969 struct inet_sock *inet = inet_sk(sk); 1970 /* 1971 * 1003.1g - break association. 1972 */ 1973 1974 sk->sk_state = TCP_CLOSE; 1975 inet->inet_daddr = 0; 1976 inet->inet_dport = 0; 1977 sock_rps_reset_rxhash(sk); 1978 sk->sk_bound_dev_if = 0; 1979 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1980 inet_reset_saddr(sk); 1981 if (sk->sk_prot->rehash && 1982 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1983 sk->sk_prot->rehash(sk); 1984 } 1985 1986 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1987 sk->sk_prot->unhash(sk); 1988 inet->inet_sport = 0; 1989 } 1990 sk_dst_reset(sk); 1991 return 0; 1992 } 1993 EXPORT_SYMBOL(__udp_disconnect); 1994 1995 int udp_disconnect(struct sock *sk, int flags) 1996 { 1997 lock_sock(sk); 1998 __udp_disconnect(sk, flags); 1999 release_sock(sk); 2000 return 0; 2001 } 2002 EXPORT_SYMBOL(udp_disconnect); 2003 2004 void udp_lib_unhash(struct sock *sk) 2005 { 2006 if (sk_hashed(sk)) { 2007 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2008 struct udp_hslot *hslot, *hslot2; 2009 2010 hslot = udp_hashslot(udptable, sock_net(sk), 2011 udp_sk(sk)->udp_port_hash); 2012 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2013 2014 spin_lock_bh(&hslot->lock); 2015 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2016 reuseport_detach_sock(sk); 2017 if (sk_del_node_init_rcu(sk)) { 2018 hslot->count--; 2019 inet_sk(sk)->inet_num = 0; 2020 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 2021 2022 spin_lock(&hslot2->lock); 2023 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2024 hslot2->count--; 2025 spin_unlock(&hslot2->lock); 2026 } 2027 spin_unlock_bh(&hslot->lock); 2028 } 2029 } 2030 EXPORT_SYMBOL(udp_lib_unhash); 2031 2032 /* 2033 * inet_rcv_saddr was changed, we must rehash secondary hash 2034 */ 2035 void udp_lib_rehash(struct sock *sk, u16 newhash) 2036 { 2037 if (sk_hashed(sk)) { 2038 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2039 struct udp_hslot *hslot, *hslot2, *nhslot2; 2040 2041 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2042 nhslot2 = udp_hashslot2(udptable, newhash); 2043 udp_sk(sk)->udp_portaddr_hash = newhash; 2044 2045 if (hslot2 != nhslot2 || 2046 rcu_access_pointer(sk->sk_reuseport_cb)) { 2047 hslot = udp_hashslot(udptable, sock_net(sk), 2048 udp_sk(sk)->udp_port_hash); 2049 /* we must lock primary chain too */ 2050 spin_lock_bh(&hslot->lock); 2051 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2052 reuseport_detach_sock(sk); 2053 2054 if (hslot2 != nhslot2) { 2055 spin_lock(&hslot2->lock); 2056 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2057 hslot2->count--; 2058 spin_unlock(&hslot2->lock); 2059 2060 spin_lock(&nhslot2->lock); 2061 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2062 &nhslot2->head); 2063 nhslot2->count++; 2064 spin_unlock(&nhslot2->lock); 2065 } 2066 2067 spin_unlock_bh(&hslot->lock); 2068 } 2069 } 2070 } 2071 EXPORT_SYMBOL(udp_lib_rehash); 2072 2073 void udp_v4_rehash(struct sock *sk) 2074 { 2075 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2076 inet_sk(sk)->inet_rcv_saddr, 2077 inet_sk(sk)->inet_num); 2078 udp_lib_rehash(sk, new_hash); 2079 } 2080 2081 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2082 { 2083 int rc; 2084 2085 if (inet_sk(sk)->inet_daddr) { 2086 sock_rps_save_rxhash(sk, skb); 2087 sk_mark_napi_id(sk, skb); 2088 sk_incoming_cpu_update(sk); 2089 } else { 2090 sk_mark_napi_id_once(sk, skb); 2091 } 2092 2093 rc = __udp_enqueue_schedule_skb(sk, skb); 2094 if (rc < 0) { 2095 int is_udplite = IS_UDPLITE(sk); 2096 2097 /* Note that an ENOMEM error is charged twice */ 2098 if (rc == -ENOMEM) 2099 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2100 is_udplite); 2101 else 2102 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2103 is_udplite); 2104 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2105 kfree_skb(skb); 2106 trace_udp_fail_queue_rcv_skb(rc, sk); 2107 return -1; 2108 } 2109 2110 return 0; 2111 } 2112 2113 /* returns: 2114 * -1: error 2115 * 0: success 2116 * >0: "udp encap" protocol resubmission 2117 * 2118 * Note that in the success and error cases, the skb is assumed to 2119 * have either been requeued or freed. 2120 */ 2121 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2122 { 2123 struct udp_sock *up = udp_sk(sk); 2124 int is_udplite = IS_UDPLITE(sk); 2125 2126 /* 2127 * Charge it to the socket, dropping if the queue is full. 2128 */ 2129 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 2130 goto drop; 2131 nf_reset_ct(skb); 2132 2133 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 2134 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2135 2136 /* 2137 * This is an encapsulation socket so pass the skb to 2138 * the socket's udp_encap_rcv() hook. Otherwise, just 2139 * fall through and pass this up the UDP socket. 2140 * up->encap_rcv() returns the following value: 2141 * =0 if skb was successfully passed to the encap 2142 * handler or was discarded by it. 2143 * >0 if skb should be passed on to UDP. 2144 * <0 if skb should be resubmitted as proto -N 2145 */ 2146 2147 /* if we're overly short, let UDP handle it */ 2148 encap_rcv = READ_ONCE(up->encap_rcv); 2149 if (encap_rcv) { 2150 int ret; 2151 2152 /* Verify checksum before giving to encap */ 2153 if (udp_lib_checksum_complete(skb)) 2154 goto csum_error; 2155 2156 ret = encap_rcv(sk, skb); 2157 if (ret <= 0) { 2158 __UDP_INC_STATS(sock_net(sk), 2159 UDP_MIB_INDATAGRAMS, 2160 is_udplite); 2161 return -ret; 2162 } 2163 } 2164 2165 /* FALLTHROUGH -- it's a UDP Packet */ 2166 } 2167 2168 /* 2169 * UDP-Lite specific tests, ignored on UDP sockets 2170 */ 2171 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2172 2173 /* 2174 * MIB statistics other than incrementing the error count are 2175 * disabled for the following two types of errors: these depend 2176 * on the application settings, not on the functioning of the 2177 * protocol stack as such. 2178 * 2179 * RFC 3828 here recommends (sec 3.3): "There should also be a 2180 * way ... to ... at least let the receiving application block 2181 * delivery of packets with coverage values less than a value 2182 * provided by the application." 2183 */ 2184 if (up->pcrlen == 0) { /* full coverage was set */ 2185 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2186 UDP_SKB_CB(skb)->cscov, skb->len); 2187 goto drop; 2188 } 2189 /* The next case involves violating the min. coverage requested 2190 * by the receiver. This is subtle: if receiver wants x and x is 2191 * greater than the buffersize/MTU then receiver will complain 2192 * that it wants x while sender emits packets of smaller size y. 2193 * Therefore the above ...()->partial_cov statement is essential. 2194 */ 2195 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2196 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2197 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2198 goto drop; 2199 } 2200 } 2201 2202 prefetch(&sk->sk_rmem_alloc); 2203 if (rcu_access_pointer(sk->sk_filter) && 2204 udp_lib_checksum_complete(skb)) 2205 goto csum_error; 2206 2207 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 2208 goto drop; 2209 2210 udp_csum_pull_header(skb); 2211 2212 ipv4_pktinfo_prepare(sk, skb); 2213 return __udp_queue_rcv_skb(sk, skb); 2214 2215 csum_error: 2216 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2217 drop: 2218 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2219 atomic_inc(&sk->sk_drops); 2220 kfree_skb(skb); 2221 return -1; 2222 } 2223 2224 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2225 { 2226 struct sk_buff *next, *segs; 2227 int ret; 2228 2229 if (likely(!udp_unexpected_gso(sk, skb))) 2230 return udp_queue_rcv_one_skb(sk, skb); 2231 2232 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2233 __skb_push(skb, -skb_mac_offset(skb)); 2234 segs = udp_rcv_segment(sk, skb, true); 2235 skb_list_walk_safe(segs, skb, next) { 2236 __skb_pull(skb, skb_transport_offset(skb)); 2237 2238 udp_post_segment_fix_csum(skb); 2239 ret = udp_queue_rcv_one_skb(sk, skb); 2240 if (ret > 0) 2241 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2242 } 2243 return 0; 2244 } 2245 2246 /* For TCP sockets, sk_rx_dst is protected by socket lock 2247 * For UDP, we use xchg() to guard against concurrent changes. 2248 */ 2249 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2250 { 2251 struct dst_entry *old; 2252 2253 if (dst_hold_safe(dst)) { 2254 old = xchg((__force struct dst_entry **)&sk->sk_rx_dst, dst); 2255 dst_release(old); 2256 return old != dst; 2257 } 2258 return false; 2259 } 2260 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2261 2262 /* 2263 * Multicasts and broadcasts go to each listener. 2264 * 2265 * Note: called only from the BH handler context. 2266 */ 2267 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2268 struct udphdr *uh, 2269 __be32 saddr, __be32 daddr, 2270 struct udp_table *udptable, 2271 int proto) 2272 { 2273 struct sock *sk, *first = NULL; 2274 unsigned short hnum = ntohs(uh->dest); 2275 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2276 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2277 unsigned int offset = offsetof(typeof(*sk), sk_node); 2278 int dif = skb->dev->ifindex; 2279 int sdif = inet_sdif(skb); 2280 struct hlist_node *node; 2281 struct sk_buff *nskb; 2282 2283 if (use_hash2) { 2284 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2285 udptable->mask; 2286 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2287 start_lookup: 2288 hslot = &udptable->hash2[hash2]; 2289 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2290 } 2291 2292 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2293 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2294 uh->source, saddr, dif, sdif, hnum)) 2295 continue; 2296 2297 if (!first) { 2298 first = sk; 2299 continue; 2300 } 2301 nskb = skb_clone(skb, GFP_ATOMIC); 2302 2303 if (unlikely(!nskb)) { 2304 atomic_inc(&sk->sk_drops); 2305 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2306 IS_UDPLITE(sk)); 2307 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2308 IS_UDPLITE(sk)); 2309 continue; 2310 } 2311 if (udp_queue_rcv_skb(sk, nskb) > 0) 2312 consume_skb(nskb); 2313 } 2314 2315 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2316 if (use_hash2 && hash2 != hash2_any) { 2317 hash2 = hash2_any; 2318 goto start_lookup; 2319 } 2320 2321 if (first) { 2322 if (udp_queue_rcv_skb(first, skb) > 0) 2323 consume_skb(skb); 2324 } else { 2325 kfree_skb(skb); 2326 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2327 proto == IPPROTO_UDPLITE); 2328 } 2329 return 0; 2330 } 2331 2332 /* Initialize UDP checksum. If exited with zero value (success), 2333 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2334 * Otherwise, csum completion requires checksumming packet body, 2335 * including udp header and folding it to skb->csum. 2336 */ 2337 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2338 int proto) 2339 { 2340 int err; 2341 2342 UDP_SKB_CB(skb)->partial_cov = 0; 2343 UDP_SKB_CB(skb)->cscov = skb->len; 2344 2345 if (proto == IPPROTO_UDPLITE) { 2346 err = udplite_checksum_init(skb, uh); 2347 if (err) 2348 return err; 2349 2350 if (UDP_SKB_CB(skb)->partial_cov) { 2351 skb->csum = inet_compute_pseudo(skb, proto); 2352 return 0; 2353 } 2354 } 2355 2356 /* Note, we are only interested in != 0 or == 0, thus the 2357 * force to int. 2358 */ 2359 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2360 inet_compute_pseudo); 2361 if (err) 2362 return err; 2363 2364 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2365 /* If SW calculated the value, we know it's bad */ 2366 if (skb->csum_complete_sw) 2367 return 1; 2368 2369 /* HW says the value is bad. Let's validate that. 2370 * skb->csum is no longer the full packet checksum, 2371 * so don't treat it as such. 2372 */ 2373 skb_checksum_complete_unset(skb); 2374 } 2375 2376 return 0; 2377 } 2378 2379 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2380 * return code conversion for ip layer consumption 2381 */ 2382 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2383 struct udphdr *uh) 2384 { 2385 int ret; 2386 2387 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2388 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2389 2390 ret = udp_queue_rcv_skb(sk, skb); 2391 2392 /* a return value > 0 means to resubmit the input, but 2393 * it wants the return to be -protocol, or 0 2394 */ 2395 if (ret > 0) 2396 return -ret; 2397 return 0; 2398 } 2399 2400 /* 2401 * All we need to do is get the socket, and then do a checksum. 2402 */ 2403 2404 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2405 int proto) 2406 { 2407 struct sock *sk; 2408 struct udphdr *uh; 2409 unsigned short ulen; 2410 struct rtable *rt = skb_rtable(skb); 2411 __be32 saddr, daddr; 2412 struct net *net = dev_net(skb->dev); 2413 bool refcounted; 2414 int drop_reason; 2415 2416 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 2417 2418 /* 2419 * Validate the packet. 2420 */ 2421 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2422 goto drop; /* No space for header. */ 2423 2424 uh = udp_hdr(skb); 2425 ulen = ntohs(uh->len); 2426 saddr = ip_hdr(skb)->saddr; 2427 daddr = ip_hdr(skb)->daddr; 2428 2429 if (ulen > skb->len) 2430 goto short_packet; 2431 2432 if (proto == IPPROTO_UDP) { 2433 /* UDP validates ulen. */ 2434 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2435 goto short_packet; 2436 uh = udp_hdr(skb); 2437 } 2438 2439 if (udp4_csum_init(skb, uh, proto)) 2440 goto csum_error; 2441 2442 sk = skb_steal_sock(skb, &refcounted); 2443 if (sk) { 2444 struct dst_entry *dst = skb_dst(skb); 2445 int ret; 2446 2447 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 2448 udp_sk_rx_dst_set(sk, dst); 2449 2450 ret = udp_unicast_rcv_skb(sk, skb, uh); 2451 if (refcounted) 2452 sock_put(sk); 2453 return ret; 2454 } 2455 2456 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2457 return __udp4_lib_mcast_deliver(net, skb, uh, 2458 saddr, daddr, udptable, proto); 2459 2460 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2461 if (sk) 2462 return udp_unicast_rcv_skb(sk, skb, uh); 2463 2464 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2465 goto drop; 2466 nf_reset_ct(skb); 2467 2468 /* No socket. Drop packet silently, if checksum is wrong */ 2469 if (udp_lib_checksum_complete(skb)) 2470 goto csum_error; 2471 2472 drop_reason = SKB_DROP_REASON_NO_SOCKET; 2473 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2474 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2475 2476 /* 2477 * Hmm. We got an UDP packet to a port to which we 2478 * don't wanna listen. Ignore it. 2479 */ 2480 kfree_skb_reason(skb, drop_reason); 2481 return 0; 2482 2483 short_packet: 2484 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; 2485 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2486 proto == IPPROTO_UDPLITE ? "Lite" : "", 2487 &saddr, ntohs(uh->source), 2488 ulen, skb->len, 2489 &daddr, ntohs(uh->dest)); 2490 goto drop; 2491 2492 csum_error: 2493 /* 2494 * RFC1122: OK. Discards the bad packet silently (as far as 2495 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2496 */ 2497 drop_reason = SKB_DROP_REASON_UDP_CSUM; 2498 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2499 proto == IPPROTO_UDPLITE ? "Lite" : "", 2500 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2501 ulen); 2502 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2503 drop: 2504 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2505 kfree_skb_reason(skb, drop_reason); 2506 return 0; 2507 } 2508 2509 /* We can only early demux multicast if there is a single matching socket. 2510 * If more than one socket found returns NULL 2511 */ 2512 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2513 __be16 loc_port, __be32 loc_addr, 2514 __be16 rmt_port, __be32 rmt_addr, 2515 int dif, int sdif) 2516 { 2517 struct sock *sk, *result; 2518 unsigned short hnum = ntohs(loc_port); 2519 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2520 struct udp_hslot *hslot = &udp_table.hash[slot]; 2521 2522 /* Do not bother scanning a too big list */ 2523 if (hslot->count > 10) 2524 return NULL; 2525 2526 result = NULL; 2527 sk_for_each_rcu(sk, &hslot->head) { 2528 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2529 rmt_port, rmt_addr, dif, sdif, hnum)) { 2530 if (result) 2531 return NULL; 2532 result = sk; 2533 } 2534 } 2535 2536 return result; 2537 } 2538 2539 /* For unicast we should only early demux connected sockets or we can 2540 * break forwarding setups. The chains here can be long so only check 2541 * if the first socket is an exact match and if not move on. 2542 */ 2543 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2544 __be16 loc_port, __be32 loc_addr, 2545 __be16 rmt_port, __be32 rmt_addr, 2546 int dif, int sdif) 2547 { 2548 unsigned short hnum = ntohs(loc_port); 2549 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2550 unsigned int slot2 = hash2 & udp_table.mask; 2551 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2552 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2553 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2554 struct sock *sk; 2555 2556 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2557 if (INET_MATCH(sk, net, acookie, rmt_addr, 2558 loc_addr, ports, dif, sdif)) 2559 return sk; 2560 /* Only check first socket in chain */ 2561 break; 2562 } 2563 return NULL; 2564 } 2565 2566 int udp_v4_early_demux(struct sk_buff *skb) 2567 { 2568 struct net *net = dev_net(skb->dev); 2569 struct in_device *in_dev = NULL; 2570 const struct iphdr *iph; 2571 const struct udphdr *uh; 2572 struct sock *sk = NULL; 2573 struct dst_entry *dst; 2574 int dif = skb->dev->ifindex; 2575 int sdif = inet_sdif(skb); 2576 int ours; 2577 2578 /* validate the packet */ 2579 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2580 return 0; 2581 2582 iph = ip_hdr(skb); 2583 uh = udp_hdr(skb); 2584 2585 if (skb->pkt_type == PACKET_MULTICAST) { 2586 in_dev = __in_dev_get_rcu(skb->dev); 2587 2588 if (!in_dev) 2589 return 0; 2590 2591 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2592 iph->protocol); 2593 if (!ours) 2594 return 0; 2595 2596 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2597 uh->source, iph->saddr, 2598 dif, sdif); 2599 } else if (skb->pkt_type == PACKET_HOST) { 2600 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2601 uh->source, iph->saddr, dif, sdif); 2602 } 2603 2604 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2605 return 0; 2606 2607 skb->sk = sk; 2608 skb->destructor = sock_efree; 2609 dst = rcu_dereference(sk->sk_rx_dst); 2610 2611 if (dst) 2612 dst = dst_check(dst, 0); 2613 if (dst) { 2614 u32 itag = 0; 2615 2616 /* set noref for now. 2617 * any place which wants to hold dst has to call 2618 * dst_hold_safe() 2619 */ 2620 skb_dst_set_noref(skb, dst); 2621 2622 /* for unconnected multicast sockets we need to validate 2623 * the source on each packet 2624 */ 2625 if (!inet_sk(sk)->inet_daddr && in_dev) 2626 return ip_mc_validate_source(skb, iph->daddr, 2627 iph->saddr, 2628 iph->tos & IPTOS_RT_MASK, 2629 skb->dev, in_dev, &itag); 2630 } 2631 return 0; 2632 } 2633 2634 int udp_rcv(struct sk_buff *skb) 2635 { 2636 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2637 } 2638 2639 void udp_destroy_sock(struct sock *sk) 2640 { 2641 struct udp_sock *up = udp_sk(sk); 2642 bool slow = lock_sock_fast(sk); 2643 2644 /* protects from races with udp_abort() */ 2645 sock_set_flag(sk, SOCK_DEAD); 2646 udp_flush_pending_frames(sk); 2647 unlock_sock_fast(sk, slow); 2648 if (static_branch_unlikely(&udp_encap_needed_key)) { 2649 if (up->encap_type) { 2650 void (*encap_destroy)(struct sock *sk); 2651 encap_destroy = READ_ONCE(up->encap_destroy); 2652 if (encap_destroy) 2653 encap_destroy(sk); 2654 } 2655 if (up->encap_enabled) 2656 static_branch_dec(&udp_encap_needed_key); 2657 } 2658 } 2659 2660 /* 2661 * Socket option code for UDP 2662 */ 2663 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2664 sockptr_t optval, unsigned int optlen, 2665 int (*push_pending_frames)(struct sock *)) 2666 { 2667 struct udp_sock *up = udp_sk(sk); 2668 int val, valbool; 2669 int err = 0; 2670 int is_udplite = IS_UDPLITE(sk); 2671 2672 if (optlen < sizeof(int)) 2673 return -EINVAL; 2674 2675 if (copy_from_sockptr(&val, optval, sizeof(val))) 2676 return -EFAULT; 2677 2678 valbool = val ? 1 : 0; 2679 2680 switch (optname) { 2681 case UDP_CORK: 2682 if (val != 0) { 2683 WRITE_ONCE(up->corkflag, 1); 2684 } else { 2685 WRITE_ONCE(up->corkflag, 0); 2686 lock_sock(sk); 2687 push_pending_frames(sk); 2688 release_sock(sk); 2689 } 2690 break; 2691 2692 case UDP_ENCAP: 2693 switch (val) { 2694 case 0: 2695 #ifdef CONFIG_XFRM 2696 case UDP_ENCAP_ESPINUDP: 2697 case UDP_ENCAP_ESPINUDP_NON_IKE: 2698 #if IS_ENABLED(CONFIG_IPV6) 2699 if (sk->sk_family == AF_INET6) 2700 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; 2701 else 2702 #endif 2703 up->encap_rcv = xfrm4_udp_encap_rcv; 2704 #endif 2705 fallthrough; 2706 case UDP_ENCAP_L2TPINUDP: 2707 up->encap_type = val; 2708 lock_sock(sk); 2709 udp_tunnel_encap_enable(sk->sk_socket); 2710 release_sock(sk); 2711 break; 2712 default: 2713 err = -ENOPROTOOPT; 2714 break; 2715 } 2716 break; 2717 2718 case UDP_NO_CHECK6_TX: 2719 up->no_check6_tx = valbool; 2720 break; 2721 2722 case UDP_NO_CHECK6_RX: 2723 up->no_check6_rx = valbool; 2724 break; 2725 2726 case UDP_SEGMENT: 2727 if (val < 0 || val > USHRT_MAX) 2728 return -EINVAL; 2729 WRITE_ONCE(up->gso_size, val); 2730 break; 2731 2732 case UDP_GRO: 2733 lock_sock(sk); 2734 2735 /* when enabling GRO, accept the related GSO packet type */ 2736 if (valbool) 2737 udp_tunnel_encap_enable(sk->sk_socket); 2738 up->gro_enabled = valbool; 2739 up->accept_udp_l4 = valbool; 2740 release_sock(sk); 2741 break; 2742 2743 /* 2744 * UDP-Lite's partial checksum coverage (RFC 3828). 2745 */ 2746 /* The sender sets actual checksum coverage length via this option. 2747 * The case coverage > packet length is handled by send module. */ 2748 case UDPLITE_SEND_CSCOV: 2749 if (!is_udplite) /* Disable the option on UDP sockets */ 2750 return -ENOPROTOOPT; 2751 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2752 val = 8; 2753 else if (val > USHRT_MAX) 2754 val = USHRT_MAX; 2755 up->pcslen = val; 2756 up->pcflag |= UDPLITE_SEND_CC; 2757 break; 2758 2759 /* The receiver specifies a minimum checksum coverage value. To make 2760 * sense, this should be set to at least 8 (as done below). If zero is 2761 * used, this again means full checksum coverage. */ 2762 case UDPLITE_RECV_CSCOV: 2763 if (!is_udplite) /* Disable the option on UDP sockets */ 2764 return -ENOPROTOOPT; 2765 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2766 val = 8; 2767 else if (val > USHRT_MAX) 2768 val = USHRT_MAX; 2769 up->pcrlen = val; 2770 up->pcflag |= UDPLITE_RECV_CC; 2771 break; 2772 2773 default: 2774 err = -ENOPROTOOPT; 2775 break; 2776 } 2777 2778 return err; 2779 } 2780 EXPORT_SYMBOL(udp_lib_setsockopt); 2781 2782 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2783 unsigned int optlen) 2784 { 2785 if (level == SOL_UDP || level == SOL_UDPLITE) 2786 return udp_lib_setsockopt(sk, level, optname, 2787 optval, optlen, 2788 udp_push_pending_frames); 2789 return ip_setsockopt(sk, level, optname, optval, optlen); 2790 } 2791 2792 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2793 char __user *optval, int __user *optlen) 2794 { 2795 struct udp_sock *up = udp_sk(sk); 2796 int val, len; 2797 2798 if (get_user(len, optlen)) 2799 return -EFAULT; 2800 2801 len = min_t(unsigned int, len, sizeof(int)); 2802 2803 if (len < 0) 2804 return -EINVAL; 2805 2806 switch (optname) { 2807 case UDP_CORK: 2808 val = READ_ONCE(up->corkflag); 2809 break; 2810 2811 case UDP_ENCAP: 2812 val = up->encap_type; 2813 break; 2814 2815 case UDP_NO_CHECK6_TX: 2816 val = up->no_check6_tx; 2817 break; 2818 2819 case UDP_NO_CHECK6_RX: 2820 val = up->no_check6_rx; 2821 break; 2822 2823 case UDP_SEGMENT: 2824 val = READ_ONCE(up->gso_size); 2825 break; 2826 2827 case UDP_GRO: 2828 val = up->gro_enabled; 2829 break; 2830 2831 /* The following two cannot be changed on UDP sockets, the return is 2832 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2833 case UDPLITE_SEND_CSCOV: 2834 val = up->pcslen; 2835 break; 2836 2837 case UDPLITE_RECV_CSCOV: 2838 val = up->pcrlen; 2839 break; 2840 2841 default: 2842 return -ENOPROTOOPT; 2843 } 2844 2845 if (put_user(len, optlen)) 2846 return -EFAULT; 2847 if (copy_to_user(optval, &val, len)) 2848 return -EFAULT; 2849 return 0; 2850 } 2851 EXPORT_SYMBOL(udp_lib_getsockopt); 2852 2853 int udp_getsockopt(struct sock *sk, int level, int optname, 2854 char __user *optval, int __user *optlen) 2855 { 2856 if (level == SOL_UDP || level == SOL_UDPLITE) 2857 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2858 return ip_getsockopt(sk, level, optname, optval, optlen); 2859 } 2860 2861 /** 2862 * udp_poll - wait for a UDP event. 2863 * @file: - file struct 2864 * @sock: - socket 2865 * @wait: - poll table 2866 * 2867 * This is same as datagram poll, except for the special case of 2868 * blocking sockets. If application is using a blocking fd 2869 * and a packet with checksum error is in the queue; 2870 * then it could get return from select indicating data available 2871 * but then block when reading it. Add special case code 2872 * to work around these arguably broken applications. 2873 */ 2874 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2875 { 2876 __poll_t mask = datagram_poll(file, sock, wait); 2877 struct sock *sk = sock->sk; 2878 2879 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2880 mask |= EPOLLIN | EPOLLRDNORM; 2881 2882 /* Check for false positives due to checksum errors */ 2883 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2884 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2885 mask &= ~(EPOLLIN | EPOLLRDNORM); 2886 2887 /* psock ingress_msg queue should not contain any bad checksum frames */ 2888 if (sk_is_readable(sk)) 2889 mask |= EPOLLIN | EPOLLRDNORM; 2890 return mask; 2891 2892 } 2893 EXPORT_SYMBOL(udp_poll); 2894 2895 int udp_abort(struct sock *sk, int err) 2896 { 2897 lock_sock(sk); 2898 2899 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2900 * with close() 2901 */ 2902 if (sock_flag(sk, SOCK_DEAD)) 2903 goto out; 2904 2905 sk->sk_err = err; 2906 sk_error_report(sk); 2907 __udp_disconnect(sk, 0); 2908 2909 out: 2910 release_sock(sk); 2911 2912 return 0; 2913 } 2914 EXPORT_SYMBOL_GPL(udp_abort); 2915 2916 struct proto udp_prot = { 2917 .name = "UDP", 2918 .owner = THIS_MODULE, 2919 .close = udp_lib_close, 2920 .pre_connect = udp_pre_connect, 2921 .connect = ip4_datagram_connect, 2922 .disconnect = udp_disconnect, 2923 .ioctl = udp_ioctl, 2924 .init = udp_init_sock, 2925 .destroy = udp_destroy_sock, 2926 .setsockopt = udp_setsockopt, 2927 .getsockopt = udp_getsockopt, 2928 .sendmsg = udp_sendmsg, 2929 .recvmsg = udp_recvmsg, 2930 .sendpage = udp_sendpage, 2931 .release_cb = ip4_datagram_release_cb, 2932 .hash = udp_lib_hash, 2933 .unhash = udp_lib_unhash, 2934 .rehash = udp_v4_rehash, 2935 .get_port = udp_v4_get_port, 2936 .put_port = udp_lib_unhash, 2937 #ifdef CONFIG_BPF_SYSCALL 2938 .psock_update_sk_prot = udp_bpf_update_proto, 2939 #endif 2940 .memory_allocated = &udp_memory_allocated, 2941 .sysctl_mem = sysctl_udp_mem, 2942 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2943 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2944 .obj_size = sizeof(struct udp_sock), 2945 .h.udp_table = &udp_table, 2946 .diag_destroy = udp_abort, 2947 }; 2948 EXPORT_SYMBOL(udp_prot); 2949 2950 /* ------------------------------------------------------------------------ */ 2951 #ifdef CONFIG_PROC_FS 2952 2953 static struct sock *udp_get_first(struct seq_file *seq, int start) 2954 { 2955 struct sock *sk; 2956 struct udp_seq_afinfo *afinfo; 2957 struct udp_iter_state *state = seq->private; 2958 struct net *net = seq_file_net(seq); 2959 2960 if (state->bpf_seq_afinfo) 2961 afinfo = state->bpf_seq_afinfo; 2962 else 2963 afinfo = pde_data(file_inode(seq->file)); 2964 2965 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; 2966 ++state->bucket) { 2967 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; 2968 2969 if (hlist_empty(&hslot->head)) 2970 continue; 2971 2972 spin_lock_bh(&hslot->lock); 2973 sk_for_each(sk, &hslot->head) { 2974 if (!net_eq(sock_net(sk), net)) 2975 continue; 2976 if (afinfo->family == AF_UNSPEC || 2977 sk->sk_family == afinfo->family) 2978 goto found; 2979 } 2980 spin_unlock_bh(&hslot->lock); 2981 } 2982 sk = NULL; 2983 found: 2984 return sk; 2985 } 2986 2987 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2988 { 2989 struct udp_seq_afinfo *afinfo; 2990 struct udp_iter_state *state = seq->private; 2991 struct net *net = seq_file_net(seq); 2992 2993 if (state->bpf_seq_afinfo) 2994 afinfo = state->bpf_seq_afinfo; 2995 else 2996 afinfo = pde_data(file_inode(seq->file)); 2997 2998 do { 2999 sk = sk_next(sk); 3000 } while (sk && (!net_eq(sock_net(sk), net) || 3001 (afinfo->family != AF_UNSPEC && 3002 sk->sk_family != afinfo->family))); 3003 3004 if (!sk) { 3005 if (state->bucket <= afinfo->udp_table->mask) 3006 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 3007 return udp_get_first(seq, state->bucket + 1); 3008 } 3009 return sk; 3010 } 3011 3012 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3013 { 3014 struct sock *sk = udp_get_first(seq, 0); 3015 3016 if (sk) 3017 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3018 --pos; 3019 return pos ? NULL : sk; 3020 } 3021 3022 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3023 { 3024 struct udp_iter_state *state = seq->private; 3025 state->bucket = MAX_UDP_PORTS; 3026 3027 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3028 } 3029 EXPORT_SYMBOL(udp_seq_start); 3030 3031 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3032 { 3033 struct sock *sk; 3034 3035 if (v == SEQ_START_TOKEN) 3036 sk = udp_get_idx(seq, 0); 3037 else 3038 sk = udp_get_next(seq, v); 3039 3040 ++*pos; 3041 return sk; 3042 } 3043 EXPORT_SYMBOL(udp_seq_next); 3044 3045 void udp_seq_stop(struct seq_file *seq, void *v) 3046 { 3047 struct udp_seq_afinfo *afinfo; 3048 struct udp_iter_state *state = seq->private; 3049 3050 if (state->bpf_seq_afinfo) 3051 afinfo = state->bpf_seq_afinfo; 3052 else 3053 afinfo = pde_data(file_inode(seq->file)); 3054 3055 if (state->bucket <= afinfo->udp_table->mask) 3056 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 3057 } 3058 EXPORT_SYMBOL(udp_seq_stop); 3059 3060 /* ------------------------------------------------------------------------ */ 3061 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3062 int bucket) 3063 { 3064 struct inet_sock *inet = inet_sk(sp); 3065 __be32 dest = inet->inet_daddr; 3066 __be32 src = inet->inet_rcv_saddr; 3067 __u16 destp = ntohs(inet->inet_dport); 3068 __u16 srcp = ntohs(inet->inet_sport); 3069 3070 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3071 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3072 bucket, src, srcp, dest, destp, sp->sk_state, 3073 sk_wmem_alloc_get(sp), 3074 udp_rqueue_get(sp), 3075 0, 0L, 0, 3076 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3077 0, sock_i_ino(sp), 3078 refcount_read(&sp->sk_refcnt), sp, 3079 atomic_read(&sp->sk_drops)); 3080 } 3081 3082 int udp4_seq_show(struct seq_file *seq, void *v) 3083 { 3084 seq_setwidth(seq, 127); 3085 if (v == SEQ_START_TOKEN) 3086 seq_puts(seq, " sl local_address rem_address st tx_queue " 3087 "rx_queue tr tm->when retrnsmt uid timeout " 3088 "inode ref pointer drops"); 3089 else { 3090 struct udp_iter_state *state = seq->private; 3091 3092 udp4_format_sock(v, seq, state->bucket); 3093 } 3094 seq_pad(seq, '\n'); 3095 return 0; 3096 } 3097 3098 #ifdef CONFIG_BPF_SYSCALL 3099 struct bpf_iter__udp { 3100 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3101 __bpf_md_ptr(struct udp_sock *, udp_sk); 3102 uid_t uid __aligned(8); 3103 int bucket __aligned(8); 3104 }; 3105 3106 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3107 struct udp_sock *udp_sk, uid_t uid, int bucket) 3108 { 3109 struct bpf_iter__udp ctx; 3110 3111 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3112 ctx.meta = meta; 3113 ctx.udp_sk = udp_sk; 3114 ctx.uid = uid; 3115 ctx.bucket = bucket; 3116 return bpf_iter_run_prog(prog, &ctx); 3117 } 3118 3119 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3120 { 3121 struct udp_iter_state *state = seq->private; 3122 struct bpf_iter_meta meta; 3123 struct bpf_prog *prog; 3124 struct sock *sk = v; 3125 uid_t uid; 3126 3127 if (v == SEQ_START_TOKEN) 3128 return 0; 3129 3130 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3131 meta.seq = seq; 3132 prog = bpf_iter_get_info(&meta, false); 3133 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3134 } 3135 3136 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3137 { 3138 struct bpf_iter_meta meta; 3139 struct bpf_prog *prog; 3140 3141 if (!v) { 3142 meta.seq = seq; 3143 prog = bpf_iter_get_info(&meta, true); 3144 if (prog) 3145 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3146 } 3147 3148 udp_seq_stop(seq, v); 3149 } 3150 3151 static const struct seq_operations bpf_iter_udp_seq_ops = { 3152 .start = udp_seq_start, 3153 .next = udp_seq_next, 3154 .stop = bpf_iter_udp_seq_stop, 3155 .show = bpf_iter_udp_seq_show, 3156 }; 3157 #endif 3158 3159 const struct seq_operations udp_seq_ops = { 3160 .start = udp_seq_start, 3161 .next = udp_seq_next, 3162 .stop = udp_seq_stop, 3163 .show = udp4_seq_show, 3164 }; 3165 EXPORT_SYMBOL(udp_seq_ops); 3166 3167 static struct udp_seq_afinfo udp4_seq_afinfo = { 3168 .family = AF_INET, 3169 .udp_table = &udp_table, 3170 }; 3171 3172 static int __net_init udp4_proc_init_net(struct net *net) 3173 { 3174 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3175 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3176 return -ENOMEM; 3177 return 0; 3178 } 3179 3180 static void __net_exit udp4_proc_exit_net(struct net *net) 3181 { 3182 remove_proc_entry("udp", net->proc_net); 3183 } 3184 3185 static struct pernet_operations udp4_net_ops = { 3186 .init = udp4_proc_init_net, 3187 .exit = udp4_proc_exit_net, 3188 }; 3189 3190 int __init udp4_proc_init(void) 3191 { 3192 return register_pernet_subsys(&udp4_net_ops); 3193 } 3194 3195 void udp4_proc_exit(void) 3196 { 3197 unregister_pernet_subsys(&udp4_net_ops); 3198 } 3199 #endif /* CONFIG_PROC_FS */ 3200 3201 static __initdata unsigned long uhash_entries; 3202 static int __init set_uhash_entries(char *str) 3203 { 3204 ssize_t ret; 3205 3206 if (!str) 3207 return 0; 3208 3209 ret = kstrtoul(str, 0, &uhash_entries); 3210 if (ret) 3211 return 0; 3212 3213 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3214 uhash_entries = UDP_HTABLE_SIZE_MIN; 3215 return 1; 3216 } 3217 __setup("uhash_entries=", set_uhash_entries); 3218 3219 void __init udp_table_init(struct udp_table *table, const char *name) 3220 { 3221 unsigned int i; 3222 3223 table->hash = alloc_large_system_hash(name, 3224 2 * sizeof(struct udp_hslot), 3225 uhash_entries, 3226 21, /* one slot per 2 MB */ 3227 0, 3228 &table->log, 3229 &table->mask, 3230 UDP_HTABLE_SIZE_MIN, 3231 64 * 1024); 3232 3233 table->hash2 = table->hash + (table->mask + 1); 3234 for (i = 0; i <= table->mask; i++) { 3235 INIT_HLIST_HEAD(&table->hash[i].head); 3236 table->hash[i].count = 0; 3237 spin_lock_init(&table->hash[i].lock); 3238 } 3239 for (i = 0; i <= table->mask; i++) { 3240 INIT_HLIST_HEAD(&table->hash2[i].head); 3241 table->hash2[i].count = 0; 3242 spin_lock_init(&table->hash2[i].lock); 3243 } 3244 } 3245 3246 u32 udp_flow_hashrnd(void) 3247 { 3248 static u32 hashrnd __read_mostly; 3249 3250 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3251 3252 return hashrnd; 3253 } 3254 EXPORT_SYMBOL(udp_flow_hashrnd); 3255 3256 static void __udp_sysctl_init(struct net *net) 3257 { 3258 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; 3259 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; 3260 3261 #ifdef CONFIG_NET_L3_MASTER_DEV 3262 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3263 #endif 3264 } 3265 3266 static int __net_init udp_sysctl_init(struct net *net) 3267 { 3268 __udp_sysctl_init(net); 3269 return 0; 3270 } 3271 3272 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3273 .init = udp_sysctl_init, 3274 }; 3275 3276 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3277 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3278 struct udp_sock *udp_sk, uid_t uid, int bucket) 3279 3280 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3281 { 3282 struct udp_iter_state *st = priv_data; 3283 struct udp_seq_afinfo *afinfo; 3284 int ret; 3285 3286 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN); 3287 if (!afinfo) 3288 return -ENOMEM; 3289 3290 afinfo->family = AF_UNSPEC; 3291 afinfo->udp_table = &udp_table; 3292 st->bpf_seq_afinfo = afinfo; 3293 ret = bpf_iter_init_seq_net(priv_data, aux); 3294 if (ret) 3295 kfree(afinfo); 3296 return ret; 3297 } 3298 3299 static void bpf_iter_fini_udp(void *priv_data) 3300 { 3301 struct udp_iter_state *st = priv_data; 3302 3303 kfree(st->bpf_seq_afinfo); 3304 bpf_iter_fini_seq_net(priv_data); 3305 } 3306 3307 static const struct bpf_iter_seq_info udp_seq_info = { 3308 .seq_ops = &bpf_iter_udp_seq_ops, 3309 .init_seq_private = bpf_iter_init_udp, 3310 .fini_seq_private = bpf_iter_fini_udp, 3311 .seq_priv_size = sizeof(struct udp_iter_state), 3312 }; 3313 3314 static struct bpf_iter_reg udp_reg_info = { 3315 .target = "udp", 3316 .ctx_arg_info_size = 1, 3317 .ctx_arg_info = { 3318 { offsetof(struct bpf_iter__udp, udp_sk), 3319 PTR_TO_BTF_ID_OR_NULL }, 3320 }, 3321 .seq_info = &udp_seq_info, 3322 }; 3323 3324 static void __init bpf_iter_register(void) 3325 { 3326 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3327 if (bpf_iter_reg_target(&udp_reg_info)) 3328 pr_warn("Warning: could not register bpf iterator udp\n"); 3329 } 3330 #endif 3331 3332 void __init udp_init(void) 3333 { 3334 unsigned long limit; 3335 unsigned int i; 3336 3337 udp_table_init(&udp_table, "UDP"); 3338 limit = nr_free_buffer_pages() / 8; 3339 limit = max(limit, 128UL); 3340 sysctl_udp_mem[0] = limit / 4 * 3; 3341 sysctl_udp_mem[1] = limit; 3342 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3343 3344 __udp_sysctl_init(&init_net); 3345 3346 /* 16 spinlocks per cpu */ 3347 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3348 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3349 GFP_KERNEL); 3350 if (!udp_busylocks) 3351 panic("UDP: failed to alloc udp_busylocks\n"); 3352 for (i = 0; i < (1U << udp_busylocks_log); i++) 3353 spin_lock_init(udp_busylocks + i); 3354 3355 if (register_pernet_subsys(&udp_sysctl_ops)) 3356 panic("UDP: failed to init sysctl parameters.\n"); 3357 3358 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3359 bpf_iter_register(); 3360 #endif 3361 } 3362