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/uaccess.h> 78 #include <asm/ioctls.h> 79 #include <linux/memblock.h> 80 #include <linux/highmem.h> 81 #include <linux/types.h> 82 #include <linux/fcntl.h> 83 #include <linux/module.h> 84 #include <linux/socket.h> 85 #include <linux/sockios.h> 86 #include <linux/igmp.h> 87 #include <linux/inetdevice.h> 88 #include <linux/in.h> 89 #include <linux/errno.h> 90 #include <linux/timer.h> 91 #include <linux/mm.h> 92 #include <linux/inet.h> 93 #include <linux/netdevice.h> 94 #include <linux/slab.h> 95 #include <net/tcp_states.h> 96 #include <linux/skbuff.h> 97 #include <linux/proc_fs.h> 98 #include <linux/seq_file.h> 99 #include <net/net_namespace.h> 100 #include <net/icmp.h> 101 #include <net/inet_hashtables.h> 102 #include <net/ip_tunnels.h> 103 #include <net/route.h> 104 #include <net/checksum.h> 105 #include <net/xfrm.h> 106 #include <trace/events/udp.h> 107 #include <linux/static_key.h> 108 #include <linux/btf_ids.h> 109 #include <trace/events/skb.h> 110 #include <net/busy_poll.h> 111 #include "udp_impl.h" 112 #include <net/sock_reuseport.h> 113 #include <net/addrconf.h> 114 #include <net/udp_tunnel.h> 115 #if IS_ENABLED(CONFIG_IPV6) 116 #include <net/ipv6_stubs.h> 117 #endif 118 119 struct udp_table udp_table __read_mostly; 120 EXPORT_SYMBOL(udp_table); 121 122 long sysctl_udp_mem[3] __read_mostly; 123 EXPORT_SYMBOL(sysctl_udp_mem); 124 125 atomic_long_t udp_memory_allocated; 126 EXPORT_SYMBOL(udp_memory_allocated); 127 128 #define MAX_UDP_PORTS 65536 129 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 130 131 static int udp_lib_lport_inuse(struct net *net, __u16 num, 132 const struct udp_hslot *hslot, 133 unsigned long *bitmap, 134 struct sock *sk, unsigned int log) 135 { 136 struct sock *sk2; 137 kuid_t uid = sock_i_uid(sk); 138 139 sk_for_each(sk2, &hslot->head) { 140 if (net_eq(sock_net(sk2), net) && 141 sk2 != sk && 142 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 143 (!sk2->sk_reuse || !sk->sk_reuse) && 144 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 145 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 146 inet_rcv_saddr_equal(sk, sk2, true)) { 147 if (sk2->sk_reuseport && sk->sk_reuseport && 148 !rcu_access_pointer(sk->sk_reuseport_cb) && 149 uid_eq(uid, sock_i_uid(sk2))) { 150 if (!bitmap) 151 return 0; 152 } else { 153 if (!bitmap) 154 return 1; 155 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 156 bitmap); 157 } 158 } 159 } 160 return 0; 161 } 162 163 /* 164 * Note: we still hold spinlock of primary hash chain, so no other writer 165 * can insert/delete a socket with local_port == num 166 */ 167 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 168 struct udp_hslot *hslot2, 169 struct sock *sk) 170 { 171 struct sock *sk2; 172 kuid_t uid = sock_i_uid(sk); 173 int res = 0; 174 175 spin_lock(&hslot2->lock); 176 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 177 if (net_eq(sock_net(sk2), net) && 178 sk2 != sk && 179 (udp_sk(sk2)->udp_port_hash == num) && 180 (!sk2->sk_reuse || !sk->sk_reuse) && 181 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 182 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 183 inet_rcv_saddr_equal(sk, sk2, true)) { 184 if (sk2->sk_reuseport && sk->sk_reuseport && 185 !rcu_access_pointer(sk->sk_reuseport_cb) && 186 uid_eq(uid, sock_i_uid(sk2))) { 187 res = 0; 188 } else { 189 res = 1; 190 } 191 break; 192 } 193 } 194 spin_unlock(&hslot2->lock); 195 return res; 196 } 197 198 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 199 { 200 struct net *net = sock_net(sk); 201 kuid_t uid = sock_i_uid(sk); 202 struct sock *sk2; 203 204 sk_for_each(sk2, &hslot->head) { 205 if (net_eq(sock_net(sk2), net) && 206 sk2 != sk && 207 sk2->sk_family == sk->sk_family && 208 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 209 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 210 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 211 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 212 inet_rcv_saddr_equal(sk, sk2, false)) { 213 return reuseport_add_sock(sk, sk2, 214 inet_rcv_saddr_any(sk)); 215 } 216 } 217 218 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 219 } 220 221 /** 222 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 223 * 224 * @sk: socket struct in question 225 * @snum: port number to look up 226 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 227 * with NULL address 228 */ 229 int udp_lib_get_port(struct sock *sk, unsigned short snum, 230 unsigned int hash2_nulladdr) 231 { 232 struct udp_hslot *hslot, *hslot2; 233 struct udp_table *udptable = sk->sk_prot->h.udp_table; 234 int error = 1; 235 struct net *net = sock_net(sk); 236 237 if (!snum) { 238 int low, high, remaining; 239 unsigned int rand; 240 unsigned short first, last; 241 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 242 243 inet_get_local_port_range(net, &low, &high); 244 remaining = (high - low) + 1; 245 246 rand = prandom_u32(); 247 first = reciprocal_scale(rand, remaining) + low; 248 /* 249 * force rand to be an odd multiple of UDP_HTABLE_SIZE 250 */ 251 rand = (rand | 1) * (udptable->mask + 1); 252 last = first + udptable->mask + 1; 253 do { 254 hslot = udp_hashslot(udptable, net, first); 255 bitmap_zero(bitmap, PORTS_PER_CHAIN); 256 spin_lock_bh(&hslot->lock); 257 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 258 udptable->log); 259 260 snum = first; 261 /* 262 * Iterate on all possible values of snum for this hash. 263 * Using steps of an odd multiple of UDP_HTABLE_SIZE 264 * give us randomization and full range coverage. 265 */ 266 do { 267 if (low <= snum && snum <= high && 268 !test_bit(snum >> udptable->log, bitmap) && 269 !inet_is_local_reserved_port(net, snum)) 270 goto found; 271 snum += rand; 272 } while (snum != first); 273 spin_unlock_bh(&hslot->lock); 274 cond_resched(); 275 } while (++first != last); 276 goto fail; 277 } else { 278 hslot = udp_hashslot(udptable, net, snum); 279 spin_lock_bh(&hslot->lock); 280 if (hslot->count > 10) { 281 int exist; 282 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 283 284 slot2 &= udptable->mask; 285 hash2_nulladdr &= udptable->mask; 286 287 hslot2 = udp_hashslot2(udptable, slot2); 288 if (hslot->count < hslot2->count) 289 goto scan_primary_hash; 290 291 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 292 if (!exist && (hash2_nulladdr != slot2)) { 293 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 294 exist = udp_lib_lport_inuse2(net, snum, hslot2, 295 sk); 296 } 297 if (exist) 298 goto fail_unlock; 299 else 300 goto found; 301 } 302 scan_primary_hash: 303 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 304 goto fail_unlock; 305 } 306 found: 307 inet_sk(sk)->inet_num = snum; 308 udp_sk(sk)->udp_port_hash = snum; 309 udp_sk(sk)->udp_portaddr_hash ^= snum; 310 if (sk_unhashed(sk)) { 311 if (sk->sk_reuseport && 312 udp_reuseport_add_sock(sk, hslot)) { 313 inet_sk(sk)->inet_num = 0; 314 udp_sk(sk)->udp_port_hash = 0; 315 udp_sk(sk)->udp_portaddr_hash ^= snum; 316 goto fail_unlock; 317 } 318 319 sk_add_node_rcu(sk, &hslot->head); 320 hslot->count++; 321 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 322 323 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 324 spin_lock(&hslot2->lock); 325 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 326 sk->sk_family == AF_INET6) 327 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 328 &hslot2->head); 329 else 330 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 331 &hslot2->head); 332 hslot2->count++; 333 spin_unlock(&hslot2->lock); 334 } 335 sock_set_flag(sk, SOCK_RCU_FREE); 336 error = 0; 337 fail_unlock: 338 spin_unlock_bh(&hslot->lock); 339 fail: 340 return error; 341 } 342 EXPORT_SYMBOL(udp_lib_get_port); 343 344 int udp_v4_get_port(struct sock *sk, unsigned short snum) 345 { 346 unsigned int hash2_nulladdr = 347 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 348 unsigned int hash2_partial = 349 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 350 351 /* precompute partial secondary hash */ 352 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 353 return udp_lib_get_port(sk, snum, hash2_nulladdr); 354 } 355 356 static int compute_score(struct sock *sk, struct net *net, 357 __be32 saddr, __be16 sport, 358 __be32 daddr, unsigned short hnum, 359 int dif, int sdif) 360 { 361 int score; 362 struct inet_sock *inet; 363 bool dev_match; 364 365 if (!net_eq(sock_net(sk), net) || 366 udp_sk(sk)->udp_port_hash != hnum || 367 ipv6_only_sock(sk)) 368 return -1; 369 370 if (sk->sk_rcv_saddr != daddr) 371 return -1; 372 373 score = (sk->sk_family == PF_INET) ? 2 : 1; 374 375 inet = inet_sk(sk); 376 if (inet->inet_daddr) { 377 if (inet->inet_daddr != saddr) 378 return -1; 379 score += 4; 380 } 381 382 if (inet->inet_dport) { 383 if (inet->inet_dport != sport) 384 return -1; 385 score += 4; 386 } 387 388 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 389 dif, sdif); 390 if (!dev_match) 391 return -1; 392 if (sk->sk_bound_dev_if) 393 score += 4; 394 395 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 396 score++; 397 return score; 398 } 399 400 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 401 const __u16 lport, const __be32 faddr, 402 const __be16 fport) 403 { 404 static u32 udp_ehash_secret __read_mostly; 405 406 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 407 408 return __inet_ehashfn(laddr, lport, faddr, fport, 409 udp_ehash_secret + net_hash_mix(net)); 410 } 411 412 static struct sock *lookup_reuseport(struct net *net, struct sock *sk, 413 struct sk_buff *skb, 414 __be32 saddr, __be16 sport, 415 __be32 daddr, unsigned short hnum) 416 { 417 struct sock *reuse_sk = NULL; 418 u32 hash; 419 420 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) { 421 hash = udp_ehashfn(net, daddr, hnum, saddr, sport); 422 reuse_sk = reuseport_select_sock(sk, hash, skb, 423 sizeof(struct udphdr)); 424 } 425 return reuse_sk; 426 } 427 428 /* called with rcu_read_lock() */ 429 static struct sock *udp4_lib_lookup2(struct net *net, 430 __be32 saddr, __be16 sport, 431 __be32 daddr, unsigned int hnum, 432 int dif, int sdif, 433 struct udp_hslot *hslot2, 434 struct sk_buff *skb) 435 { 436 struct sock *sk, *result; 437 int score, badness; 438 439 result = NULL; 440 badness = 0; 441 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 442 score = compute_score(sk, net, saddr, sport, 443 daddr, hnum, dif, sdif); 444 if (score > badness) { 445 result = lookup_reuseport(net, sk, skb, 446 saddr, sport, daddr, hnum); 447 /* Fall back to scoring if group has connections */ 448 if (result && !reuseport_has_conns(sk, false)) 449 return result; 450 451 result = result ? : sk; 452 badness = score; 453 } 454 } 455 return result; 456 } 457 458 static struct sock *udp4_lookup_run_bpf(struct net *net, 459 struct udp_table *udptable, 460 struct sk_buff *skb, 461 __be32 saddr, __be16 sport, 462 __be32 daddr, u16 hnum) 463 { 464 struct sock *sk, *reuse_sk; 465 bool no_reuseport; 466 467 if (udptable != &udp_table) 468 return NULL; /* only UDP is supported */ 469 470 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_UDP, 471 saddr, sport, daddr, hnum, &sk); 472 if (no_reuseport || IS_ERR_OR_NULL(sk)) 473 return sk; 474 475 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum); 476 if (reuse_sk) 477 sk = reuse_sk; 478 return sk; 479 } 480 481 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 482 * harder than this. -DaveM 483 */ 484 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 485 __be16 sport, __be32 daddr, __be16 dport, int dif, 486 int sdif, struct udp_table *udptable, struct sk_buff *skb) 487 { 488 unsigned short hnum = ntohs(dport); 489 unsigned int hash2, slot2; 490 struct udp_hslot *hslot2; 491 struct sock *result, *sk; 492 493 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 494 slot2 = hash2 & udptable->mask; 495 hslot2 = &udptable->hash2[slot2]; 496 497 /* Lookup connected or non-wildcard socket */ 498 result = udp4_lib_lookup2(net, saddr, sport, 499 daddr, hnum, dif, sdif, 500 hslot2, skb); 501 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 502 goto done; 503 504 /* Lookup redirect from BPF */ 505 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) { 506 sk = udp4_lookup_run_bpf(net, udptable, skb, 507 saddr, sport, daddr, hnum); 508 if (sk) { 509 result = sk; 510 goto done; 511 } 512 } 513 514 /* Got non-wildcard socket or error on first lookup */ 515 if (result) 516 goto done; 517 518 /* Lookup wildcard sockets */ 519 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 520 slot2 = hash2 & udptable->mask; 521 hslot2 = &udptable->hash2[slot2]; 522 523 result = udp4_lib_lookup2(net, saddr, sport, 524 htonl(INADDR_ANY), hnum, dif, sdif, 525 hslot2, skb); 526 done: 527 if (IS_ERR(result)) 528 return NULL; 529 return result; 530 } 531 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 532 533 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 534 __be16 sport, __be16 dport, 535 struct udp_table *udptable) 536 { 537 const struct iphdr *iph = ip_hdr(skb); 538 539 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 540 iph->daddr, dport, inet_iif(skb), 541 inet_sdif(skb), udptable, skb); 542 } 543 544 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb, 545 __be16 sport, __be16 dport) 546 { 547 const struct iphdr *iph = ip_hdr(skb); 548 549 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 550 iph->daddr, dport, inet_iif(skb), 551 inet_sdif(skb), &udp_table, NULL); 552 } 553 554 /* Must be called under rcu_read_lock(). 555 * Does increment socket refcount. 556 */ 557 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 558 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 559 __be32 daddr, __be16 dport, int dif) 560 { 561 struct sock *sk; 562 563 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 564 dif, 0, &udp_table, NULL); 565 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 566 sk = NULL; 567 return sk; 568 } 569 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 570 #endif 571 572 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 573 __be16 loc_port, __be32 loc_addr, 574 __be16 rmt_port, __be32 rmt_addr, 575 int dif, int sdif, unsigned short hnum) 576 { 577 struct inet_sock *inet = inet_sk(sk); 578 579 if (!net_eq(sock_net(sk), net) || 580 udp_sk(sk)->udp_port_hash != hnum || 581 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 582 (inet->inet_dport != rmt_port && inet->inet_dport) || 583 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 584 ipv6_only_sock(sk) || 585 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 586 return false; 587 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 588 return false; 589 return true; 590 } 591 592 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 593 void udp_encap_enable(void) 594 { 595 static_branch_inc(&udp_encap_needed_key); 596 } 597 EXPORT_SYMBOL(udp_encap_enable); 598 599 void udp_encap_disable(void) 600 { 601 static_branch_dec(&udp_encap_needed_key); 602 } 603 EXPORT_SYMBOL(udp_encap_disable); 604 605 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 606 * through error handlers in encapsulations looking for a match. 607 */ 608 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 609 { 610 int i; 611 612 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 613 int (*handler)(struct sk_buff *skb, u32 info); 614 const struct ip_tunnel_encap_ops *encap; 615 616 encap = rcu_dereference(iptun_encaps[i]); 617 if (!encap) 618 continue; 619 handler = encap->err_handler; 620 if (handler && !handler(skb, info)) 621 return 0; 622 } 623 624 return -ENOENT; 625 } 626 627 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 628 * reversing source and destination port: this will match tunnels that force the 629 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 630 * lwtunnels might actually break this assumption by being configured with 631 * different destination ports on endpoints, in this case we won't be able to 632 * trace ICMP messages back to them. 633 * 634 * If this doesn't match any socket, probe tunnels with arbitrary destination 635 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 636 * we've sent packets to won't necessarily match the local destination port. 637 * 638 * Then ask the tunnel implementation to match the error against a valid 639 * association. 640 * 641 * Return an error if we can't find a match, the socket if we need further 642 * processing, zero otherwise. 643 */ 644 static struct sock *__udp4_lib_err_encap(struct net *net, 645 const struct iphdr *iph, 646 struct udphdr *uh, 647 struct udp_table *udptable, 648 struct sock *sk, 649 struct sk_buff *skb, u32 info) 650 { 651 int (*lookup)(struct sock *sk, struct sk_buff *skb); 652 int network_offset, transport_offset; 653 struct udp_sock *up; 654 655 network_offset = skb_network_offset(skb); 656 transport_offset = skb_transport_offset(skb); 657 658 /* Network header needs to point to the outer IPv4 header inside ICMP */ 659 skb_reset_network_header(skb); 660 661 /* Transport header needs to point to the UDP header */ 662 skb_set_transport_header(skb, iph->ihl << 2); 663 664 if (sk) { 665 up = udp_sk(sk); 666 667 lookup = READ_ONCE(up->encap_err_lookup); 668 if (lookup && lookup(sk, skb)) 669 sk = NULL; 670 671 goto out; 672 } 673 674 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 675 iph->saddr, uh->dest, skb->dev->ifindex, 0, 676 udptable, NULL); 677 if (sk) { 678 up = udp_sk(sk); 679 680 lookup = READ_ONCE(up->encap_err_lookup); 681 if (!lookup || lookup(sk, skb)) 682 sk = NULL; 683 } 684 685 out: 686 if (!sk) 687 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 688 689 skb_set_transport_header(skb, transport_offset); 690 skb_set_network_header(skb, network_offset); 691 692 return sk; 693 } 694 695 /* 696 * This routine is called by the ICMP module when it gets some 697 * sort of error condition. If err < 0 then the socket should 698 * be closed and the error returned to the user. If err > 0 699 * it's just the icmp type << 8 | icmp code. 700 * Header points to the ip header of the error packet. We move 701 * on past this. Then (as it used to claim before adjustment) 702 * header points to the first 8 bytes of the udp header. We need 703 * to find the appropriate port. 704 */ 705 706 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 707 { 708 struct inet_sock *inet; 709 const struct iphdr *iph = (const struct iphdr *)skb->data; 710 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 711 const int type = icmp_hdr(skb)->type; 712 const int code = icmp_hdr(skb)->code; 713 bool tunnel = false; 714 struct sock *sk; 715 int harderr; 716 int err; 717 struct net *net = dev_net(skb->dev); 718 719 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 720 iph->saddr, uh->source, skb->dev->ifindex, 721 inet_sdif(skb), udptable, NULL); 722 723 if (!sk || udp_sk(sk)->encap_type) { 724 /* No socket for error: try tunnels before discarding */ 725 if (static_branch_unlikely(&udp_encap_needed_key)) { 726 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb, 727 info); 728 if (!sk) 729 return 0; 730 } else 731 sk = ERR_PTR(-ENOENT); 732 733 if (IS_ERR(sk)) { 734 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 735 return PTR_ERR(sk); 736 } 737 738 tunnel = true; 739 } 740 741 err = 0; 742 harderr = 0; 743 inet = inet_sk(sk); 744 745 switch (type) { 746 default: 747 case ICMP_TIME_EXCEEDED: 748 err = EHOSTUNREACH; 749 break; 750 case ICMP_SOURCE_QUENCH: 751 goto out; 752 case ICMP_PARAMETERPROB: 753 err = EPROTO; 754 harderr = 1; 755 break; 756 case ICMP_DEST_UNREACH: 757 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 758 ipv4_sk_update_pmtu(skb, sk, info); 759 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 760 err = EMSGSIZE; 761 harderr = 1; 762 break; 763 } 764 goto out; 765 } 766 err = EHOSTUNREACH; 767 if (code <= NR_ICMP_UNREACH) { 768 harderr = icmp_err_convert[code].fatal; 769 err = icmp_err_convert[code].errno; 770 } 771 break; 772 case ICMP_REDIRECT: 773 ipv4_sk_redirect(skb, sk); 774 goto out; 775 } 776 777 /* 778 * RFC1122: OK. Passes ICMP errors back to application, as per 779 * 4.1.3.3. 780 */ 781 if (tunnel) { 782 /* ...not for tunnels though: we don't have a sending socket */ 783 goto out; 784 } 785 if (!inet->recverr) { 786 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 787 goto out; 788 } else 789 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 790 791 sk->sk_err = err; 792 sk_error_report(sk); 793 out: 794 return 0; 795 } 796 797 int udp_err(struct sk_buff *skb, u32 info) 798 { 799 return __udp4_lib_err(skb, info, &udp_table); 800 } 801 802 /* 803 * Throw away all pending data and cancel the corking. Socket is locked. 804 */ 805 void udp_flush_pending_frames(struct sock *sk) 806 { 807 struct udp_sock *up = udp_sk(sk); 808 809 if (up->pending) { 810 up->len = 0; 811 up->pending = 0; 812 ip_flush_pending_frames(sk); 813 } 814 } 815 EXPORT_SYMBOL(udp_flush_pending_frames); 816 817 /** 818 * udp4_hwcsum - handle outgoing HW checksumming 819 * @skb: sk_buff containing the filled-in UDP header 820 * (checksum field must be zeroed out) 821 * @src: source IP address 822 * @dst: destination IP address 823 */ 824 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 825 { 826 struct udphdr *uh = udp_hdr(skb); 827 int offset = skb_transport_offset(skb); 828 int len = skb->len - offset; 829 int hlen = len; 830 __wsum csum = 0; 831 832 if (!skb_has_frag_list(skb)) { 833 /* 834 * Only one fragment on the socket. 835 */ 836 skb->csum_start = skb_transport_header(skb) - skb->head; 837 skb->csum_offset = offsetof(struct udphdr, check); 838 uh->check = ~csum_tcpudp_magic(src, dst, len, 839 IPPROTO_UDP, 0); 840 } else { 841 struct sk_buff *frags; 842 843 /* 844 * HW-checksum won't work as there are two or more 845 * fragments on the socket so that all csums of sk_buffs 846 * should be together 847 */ 848 skb_walk_frags(skb, frags) { 849 csum = csum_add(csum, frags->csum); 850 hlen -= frags->len; 851 } 852 853 csum = skb_checksum(skb, offset, hlen, csum); 854 skb->ip_summed = CHECKSUM_NONE; 855 856 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 857 if (uh->check == 0) 858 uh->check = CSUM_MANGLED_0; 859 } 860 } 861 EXPORT_SYMBOL_GPL(udp4_hwcsum); 862 863 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 864 * for the simple case like when setting the checksum for a UDP tunnel. 865 */ 866 void udp_set_csum(bool nocheck, struct sk_buff *skb, 867 __be32 saddr, __be32 daddr, int len) 868 { 869 struct udphdr *uh = udp_hdr(skb); 870 871 if (nocheck) { 872 uh->check = 0; 873 } else if (skb_is_gso(skb)) { 874 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 875 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 876 uh->check = 0; 877 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 878 if (uh->check == 0) 879 uh->check = CSUM_MANGLED_0; 880 } else { 881 skb->ip_summed = CHECKSUM_PARTIAL; 882 skb->csum_start = skb_transport_header(skb) - skb->head; 883 skb->csum_offset = offsetof(struct udphdr, check); 884 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 885 } 886 } 887 EXPORT_SYMBOL(udp_set_csum); 888 889 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 890 struct inet_cork *cork) 891 { 892 struct sock *sk = skb->sk; 893 struct inet_sock *inet = inet_sk(sk); 894 struct udphdr *uh; 895 int err; 896 int is_udplite = IS_UDPLITE(sk); 897 int offset = skb_transport_offset(skb); 898 int len = skb->len - offset; 899 int datalen = len - sizeof(*uh); 900 __wsum csum = 0; 901 902 /* 903 * Create a UDP header 904 */ 905 uh = udp_hdr(skb); 906 uh->source = inet->inet_sport; 907 uh->dest = fl4->fl4_dport; 908 uh->len = htons(len); 909 uh->check = 0; 910 911 if (cork->gso_size) { 912 const int hlen = skb_network_header_len(skb) + 913 sizeof(struct udphdr); 914 915 if (hlen + cork->gso_size > cork->fragsize) { 916 kfree_skb(skb); 917 return -EINVAL; 918 } 919 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 920 kfree_skb(skb); 921 return -EINVAL; 922 } 923 if (sk->sk_no_check_tx) { 924 kfree_skb(skb); 925 return -EINVAL; 926 } 927 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 928 dst_xfrm(skb_dst(skb))) { 929 kfree_skb(skb); 930 return -EIO; 931 } 932 933 if (datalen > cork->gso_size) { 934 skb_shinfo(skb)->gso_size = cork->gso_size; 935 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 936 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 937 cork->gso_size); 938 } 939 goto csum_partial; 940 } 941 942 if (is_udplite) /* UDP-Lite */ 943 csum = udplite_csum(skb); 944 945 else if (sk->sk_no_check_tx) { /* UDP csum off */ 946 947 skb->ip_summed = CHECKSUM_NONE; 948 goto send; 949 950 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 951 csum_partial: 952 953 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 954 goto send; 955 956 } else 957 csum = udp_csum(skb); 958 959 /* add protocol-dependent pseudo-header */ 960 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 961 sk->sk_protocol, csum); 962 if (uh->check == 0) 963 uh->check = CSUM_MANGLED_0; 964 965 send: 966 err = ip_send_skb(sock_net(sk), skb); 967 if (err) { 968 if (err == -ENOBUFS && !inet->recverr) { 969 UDP_INC_STATS(sock_net(sk), 970 UDP_MIB_SNDBUFERRORS, is_udplite); 971 err = 0; 972 } 973 } else 974 UDP_INC_STATS(sock_net(sk), 975 UDP_MIB_OUTDATAGRAMS, is_udplite); 976 return err; 977 } 978 979 /* 980 * Push out all pending data as one UDP datagram. Socket is locked. 981 */ 982 int udp_push_pending_frames(struct sock *sk) 983 { 984 struct udp_sock *up = udp_sk(sk); 985 struct inet_sock *inet = inet_sk(sk); 986 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 987 struct sk_buff *skb; 988 int err = 0; 989 990 skb = ip_finish_skb(sk, fl4); 991 if (!skb) 992 goto out; 993 994 err = udp_send_skb(skb, fl4, &inet->cork.base); 995 996 out: 997 up->len = 0; 998 up->pending = 0; 999 return err; 1000 } 1001 EXPORT_SYMBOL(udp_push_pending_frames); 1002 1003 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 1004 { 1005 switch (cmsg->cmsg_type) { 1006 case UDP_SEGMENT: 1007 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 1008 return -EINVAL; 1009 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 1010 return 0; 1011 default: 1012 return -EINVAL; 1013 } 1014 } 1015 1016 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 1017 { 1018 struct cmsghdr *cmsg; 1019 bool need_ip = false; 1020 int err; 1021 1022 for_each_cmsghdr(cmsg, msg) { 1023 if (!CMSG_OK(msg, cmsg)) 1024 return -EINVAL; 1025 1026 if (cmsg->cmsg_level != SOL_UDP) { 1027 need_ip = true; 1028 continue; 1029 } 1030 1031 err = __udp_cmsg_send(cmsg, gso_size); 1032 if (err) 1033 return err; 1034 } 1035 1036 return need_ip; 1037 } 1038 EXPORT_SYMBOL_GPL(udp_cmsg_send); 1039 1040 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1041 { 1042 struct inet_sock *inet = inet_sk(sk); 1043 struct udp_sock *up = udp_sk(sk); 1044 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 1045 struct flowi4 fl4_stack; 1046 struct flowi4 *fl4; 1047 int ulen = len; 1048 struct ipcm_cookie ipc; 1049 struct rtable *rt = NULL; 1050 int free = 0; 1051 int connected = 0; 1052 __be32 daddr, faddr, saddr; 1053 __be16 dport; 1054 u8 tos; 1055 int err, is_udplite = IS_UDPLITE(sk); 1056 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE; 1057 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1058 struct sk_buff *skb; 1059 struct ip_options_data opt_copy; 1060 1061 if (len > 0xFFFF) 1062 return -EMSGSIZE; 1063 1064 /* 1065 * Check the flags. 1066 */ 1067 1068 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 1069 return -EOPNOTSUPP; 1070 1071 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1072 1073 fl4 = &inet->cork.fl.u.ip4; 1074 if (up->pending) { 1075 /* 1076 * There are pending frames. 1077 * The socket lock must be held while it's corked. 1078 */ 1079 lock_sock(sk); 1080 if (likely(up->pending)) { 1081 if (unlikely(up->pending != AF_INET)) { 1082 release_sock(sk); 1083 return -EINVAL; 1084 } 1085 goto do_append_data; 1086 } 1087 release_sock(sk); 1088 } 1089 ulen += sizeof(struct udphdr); 1090 1091 /* 1092 * Get and verify the address. 1093 */ 1094 if (usin) { 1095 if (msg->msg_namelen < sizeof(*usin)) 1096 return -EINVAL; 1097 if (usin->sin_family != AF_INET) { 1098 if (usin->sin_family != AF_UNSPEC) 1099 return -EAFNOSUPPORT; 1100 } 1101 1102 daddr = usin->sin_addr.s_addr; 1103 dport = usin->sin_port; 1104 if (dport == 0) 1105 return -EINVAL; 1106 } else { 1107 if (sk->sk_state != TCP_ESTABLISHED) 1108 return -EDESTADDRREQ; 1109 daddr = inet->inet_daddr; 1110 dport = inet->inet_dport; 1111 /* Open fast path for connected socket. 1112 Route will not be used, if at least one option is set. 1113 */ 1114 connected = 1; 1115 } 1116 1117 ipcm_init_sk(&ipc, inet); 1118 ipc.gso_size = READ_ONCE(up->gso_size); 1119 1120 if (msg->msg_controllen) { 1121 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1122 if (err > 0) 1123 err = ip_cmsg_send(sk, msg, &ipc, 1124 sk->sk_family == AF_INET6); 1125 if (unlikely(err < 0)) { 1126 kfree(ipc.opt); 1127 return err; 1128 } 1129 if (ipc.opt) 1130 free = 1; 1131 connected = 0; 1132 } 1133 if (!ipc.opt) { 1134 struct ip_options_rcu *inet_opt; 1135 1136 rcu_read_lock(); 1137 inet_opt = rcu_dereference(inet->inet_opt); 1138 if (inet_opt) { 1139 memcpy(&opt_copy, inet_opt, 1140 sizeof(*inet_opt) + inet_opt->opt.optlen); 1141 ipc.opt = &opt_copy.opt; 1142 } 1143 rcu_read_unlock(); 1144 } 1145 1146 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) { 1147 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1148 (struct sockaddr *)usin, &ipc.addr); 1149 if (err) 1150 goto out_free; 1151 if (usin) { 1152 if (usin->sin_port == 0) { 1153 /* BPF program set invalid port. Reject it. */ 1154 err = -EINVAL; 1155 goto out_free; 1156 } 1157 daddr = usin->sin_addr.s_addr; 1158 dport = usin->sin_port; 1159 } 1160 } 1161 1162 saddr = ipc.addr; 1163 ipc.addr = faddr = daddr; 1164 1165 if (ipc.opt && ipc.opt->opt.srr) { 1166 if (!daddr) { 1167 err = -EINVAL; 1168 goto out_free; 1169 } 1170 faddr = ipc.opt->opt.faddr; 1171 connected = 0; 1172 } 1173 tos = get_rttos(&ipc, inet); 1174 if (sock_flag(sk, SOCK_LOCALROUTE) || 1175 (msg->msg_flags & MSG_DONTROUTE) || 1176 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1177 tos |= RTO_ONLINK; 1178 connected = 0; 1179 } 1180 1181 if (ipv4_is_multicast(daddr)) { 1182 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1183 ipc.oif = inet->mc_index; 1184 if (!saddr) 1185 saddr = inet->mc_addr; 1186 connected = 0; 1187 } else if (!ipc.oif) { 1188 ipc.oif = inet->uc_index; 1189 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1190 /* oif is set, packet is to local broadcast and 1191 * uc_index is set. oif is most likely set 1192 * by sk_bound_dev_if. If uc_index != oif check if the 1193 * oif is an L3 master and uc_index is an L3 slave. 1194 * If so, we want to allow the send using the uc_index. 1195 */ 1196 if (ipc.oif != inet->uc_index && 1197 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1198 inet->uc_index)) { 1199 ipc.oif = inet->uc_index; 1200 } 1201 } 1202 1203 if (connected) 1204 rt = (struct rtable *)sk_dst_check(sk, 0); 1205 1206 if (!rt) { 1207 struct net *net = sock_net(sk); 1208 __u8 flow_flags = inet_sk_flowi_flags(sk); 1209 1210 fl4 = &fl4_stack; 1211 1212 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, 1213 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1214 flow_flags, 1215 faddr, saddr, dport, inet->inet_sport, 1216 sk->sk_uid); 1217 1218 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4)); 1219 rt = ip_route_output_flow(net, fl4, sk); 1220 if (IS_ERR(rt)) { 1221 err = PTR_ERR(rt); 1222 rt = NULL; 1223 if (err == -ENETUNREACH) 1224 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1225 goto out; 1226 } 1227 1228 err = -EACCES; 1229 if ((rt->rt_flags & RTCF_BROADCAST) && 1230 !sock_flag(sk, SOCK_BROADCAST)) 1231 goto out; 1232 if (connected) 1233 sk_dst_set(sk, dst_clone(&rt->dst)); 1234 } 1235 1236 if (msg->msg_flags&MSG_CONFIRM) 1237 goto do_confirm; 1238 back_from_confirm: 1239 1240 saddr = fl4->saddr; 1241 if (!ipc.addr) 1242 daddr = ipc.addr = fl4->daddr; 1243 1244 /* Lockless fast path for the non-corking case. */ 1245 if (!corkreq) { 1246 struct inet_cork cork; 1247 1248 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1249 sizeof(struct udphdr), &ipc, &rt, 1250 &cork, msg->msg_flags); 1251 err = PTR_ERR(skb); 1252 if (!IS_ERR_OR_NULL(skb)) 1253 err = udp_send_skb(skb, fl4, &cork); 1254 goto out; 1255 } 1256 1257 lock_sock(sk); 1258 if (unlikely(up->pending)) { 1259 /* The socket is already corked while preparing it. */ 1260 /* ... which is an evident application bug. --ANK */ 1261 release_sock(sk); 1262 1263 net_dbg_ratelimited("socket already corked\n"); 1264 err = -EINVAL; 1265 goto out; 1266 } 1267 /* 1268 * Now cork the socket to pend data. 1269 */ 1270 fl4 = &inet->cork.fl.u.ip4; 1271 fl4->daddr = daddr; 1272 fl4->saddr = saddr; 1273 fl4->fl4_dport = dport; 1274 fl4->fl4_sport = inet->inet_sport; 1275 up->pending = AF_INET; 1276 1277 do_append_data: 1278 up->len += ulen; 1279 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1280 sizeof(struct udphdr), &ipc, &rt, 1281 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1282 if (err) 1283 udp_flush_pending_frames(sk); 1284 else if (!corkreq) 1285 err = udp_push_pending_frames(sk); 1286 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1287 up->pending = 0; 1288 release_sock(sk); 1289 1290 out: 1291 ip_rt_put(rt); 1292 out_free: 1293 if (free) 1294 kfree(ipc.opt); 1295 if (!err) 1296 return len; 1297 /* 1298 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1299 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1300 * we don't have a good statistic (IpOutDiscards but it can be too many 1301 * things). We could add another new stat but at least for now that 1302 * seems like overkill. 1303 */ 1304 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1305 UDP_INC_STATS(sock_net(sk), 1306 UDP_MIB_SNDBUFERRORS, is_udplite); 1307 } 1308 return err; 1309 1310 do_confirm: 1311 if (msg->msg_flags & MSG_PROBE) 1312 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1313 if (!(msg->msg_flags&MSG_PROBE) || len) 1314 goto back_from_confirm; 1315 err = 0; 1316 goto out; 1317 } 1318 EXPORT_SYMBOL(udp_sendmsg); 1319 1320 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1321 size_t size, int flags) 1322 { 1323 struct inet_sock *inet = inet_sk(sk); 1324 struct udp_sock *up = udp_sk(sk); 1325 int ret; 1326 1327 if (flags & MSG_SENDPAGE_NOTLAST) 1328 flags |= MSG_MORE; 1329 1330 if (!up->pending) { 1331 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1332 1333 /* Call udp_sendmsg to specify destination address which 1334 * sendpage interface can't pass. 1335 * This will succeed only when the socket is connected. 1336 */ 1337 ret = udp_sendmsg(sk, &msg, 0); 1338 if (ret < 0) 1339 return ret; 1340 } 1341 1342 lock_sock(sk); 1343 1344 if (unlikely(!up->pending)) { 1345 release_sock(sk); 1346 1347 net_dbg_ratelimited("cork failed\n"); 1348 return -EINVAL; 1349 } 1350 1351 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1352 page, offset, size, flags); 1353 if (ret == -EOPNOTSUPP) { 1354 release_sock(sk); 1355 return sock_no_sendpage(sk->sk_socket, page, offset, 1356 size, flags); 1357 } 1358 if (ret < 0) { 1359 udp_flush_pending_frames(sk); 1360 goto out; 1361 } 1362 1363 up->len += size; 1364 if (!(READ_ONCE(up->corkflag) || (flags&MSG_MORE))) 1365 ret = udp_push_pending_frames(sk); 1366 if (!ret) 1367 ret = size; 1368 out: 1369 release_sock(sk); 1370 return ret; 1371 } 1372 1373 #define UDP_SKB_IS_STATELESS 0x80000000 1374 1375 /* all head states (dst, sk, nf conntrack) except skb extensions are 1376 * cleared by udp_rcv(). 1377 * 1378 * We need to preserve secpath, if present, to eventually process 1379 * IP_CMSG_PASSSEC at recvmsg() time. 1380 * 1381 * Other extensions can be cleared. 1382 */ 1383 static bool udp_try_make_stateless(struct sk_buff *skb) 1384 { 1385 if (!skb_has_extensions(skb)) 1386 return true; 1387 1388 if (!secpath_exists(skb)) { 1389 skb_ext_reset(skb); 1390 return true; 1391 } 1392 1393 return false; 1394 } 1395 1396 static void udp_set_dev_scratch(struct sk_buff *skb) 1397 { 1398 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1399 1400 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1401 scratch->_tsize_state = skb->truesize; 1402 #if BITS_PER_LONG == 64 1403 scratch->len = skb->len; 1404 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1405 scratch->is_linear = !skb_is_nonlinear(skb); 1406 #endif 1407 if (udp_try_make_stateless(skb)) 1408 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1409 } 1410 1411 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1412 { 1413 /* We come here after udp_lib_checksum_complete() returned 0. 1414 * This means that __skb_checksum_complete() might have 1415 * set skb->csum_valid to 1. 1416 * On 64bit platforms, we can set csum_unnecessary 1417 * to true, but only if the skb is not shared. 1418 */ 1419 #if BITS_PER_LONG == 64 1420 if (!skb_shared(skb)) 1421 udp_skb_scratch(skb)->csum_unnecessary = true; 1422 #endif 1423 } 1424 1425 static int udp_skb_truesize(struct sk_buff *skb) 1426 { 1427 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1428 } 1429 1430 static bool udp_skb_has_head_state(struct sk_buff *skb) 1431 { 1432 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1433 } 1434 1435 /* fully reclaim rmem/fwd memory allocated for skb */ 1436 static void udp_rmem_release(struct sock *sk, int size, int partial, 1437 bool rx_queue_lock_held) 1438 { 1439 struct udp_sock *up = udp_sk(sk); 1440 struct sk_buff_head *sk_queue; 1441 int amt; 1442 1443 if (likely(partial)) { 1444 up->forward_deficit += size; 1445 size = up->forward_deficit; 1446 if (size < (sk->sk_rcvbuf >> 2) && 1447 !skb_queue_empty(&up->reader_queue)) 1448 return; 1449 } else { 1450 size += up->forward_deficit; 1451 } 1452 up->forward_deficit = 0; 1453 1454 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1455 * if the called don't held it already 1456 */ 1457 sk_queue = &sk->sk_receive_queue; 1458 if (!rx_queue_lock_held) 1459 spin_lock(&sk_queue->lock); 1460 1461 1462 sk->sk_forward_alloc += size; 1463 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); 1464 sk->sk_forward_alloc -= amt; 1465 1466 if (amt) 1467 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); 1468 1469 atomic_sub(size, &sk->sk_rmem_alloc); 1470 1471 /* this can save us from acquiring the rx queue lock on next receive */ 1472 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1473 1474 if (!rx_queue_lock_held) 1475 spin_unlock(&sk_queue->lock); 1476 } 1477 1478 /* Note: called with reader_queue.lock held. 1479 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1480 * This avoids a cache line miss while receive_queue lock is held. 1481 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1482 */ 1483 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1484 { 1485 prefetch(&skb->data); 1486 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1487 } 1488 EXPORT_SYMBOL(udp_skb_destructor); 1489 1490 /* as above, but the caller held the rx queue lock, too */ 1491 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1492 { 1493 prefetch(&skb->data); 1494 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1495 } 1496 1497 /* Idea of busylocks is to let producers grab an extra spinlock 1498 * to relieve pressure on the receive_queue spinlock shared by consumer. 1499 * Under flood, this means that only one producer can be in line 1500 * trying to acquire the receive_queue spinlock. 1501 * These busylock can be allocated on a per cpu manner, instead of a 1502 * per socket one (that would consume a cache line per socket) 1503 */ 1504 static int udp_busylocks_log __read_mostly; 1505 static spinlock_t *udp_busylocks __read_mostly; 1506 1507 static spinlock_t *busylock_acquire(void *ptr) 1508 { 1509 spinlock_t *busy; 1510 1511 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1512 spin_lock(busy); 1513 return busy; 1514 } 1515 1516 static void busylock_release(spinlock_t *busy) 1517 { 1518 if (busy) 1519 spin_unlock(busy); 1520 } 1521 1522 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1523 { 1524 struct sk_buff_head *list = &sk->sk_receive_queue; 1525 int rmem, delta, amt, err = -ENOMEM; 1526 spinlock_t *busy = NULL; 1527 int size; 1528 1529 /* try to avoid the costly atomic add/sub pair when the receive 1530 * queue is full; always allow at least a packet 1531 */ 1532 rmem = atomic_read(&sk->sk_rmem_alloc); 1533 if (rmem > sk->sk_rcvbuf) 1534 goto drop; 1535 1536 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1537 * having linear skbs : 1538 * - Reduce memory overhead and thus increase receive queue capacity 1539 * - Less cache line misses at copyout() time 1540 * - Less work at consume_skb() (less alien page frag freeing) 1541 */ 1542 if (rmem > (sk->sk_rcvbuf >> 1)) { 1543 skb_condense(skb); 1544 1545 busy = busylock_acquire(sk); 1546 } 1547 size = skb->truesize; 1548 udp_set_dev_scratch(skb); 1549 1550 /* we drop only if the receive buf is full and the receive 1551 * queue contains some other skb 1552 */ 1553 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1554 if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) 1555 goto uncharge_drop; 1556 1557 spin_lock(&list->lock); 1558 if (size >= sk->sk_forward_alloc) { 1559 amt = sk_mem_pages(size); 1560 delta = amt << SK_MEM_QUANTUM_SHIFT; 1561 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1562 err = -ENOBUFS; 1563 spin_unlock(&list->lock); 1564 goto uncharge_drop; 1565 } 1566 1567 sk->sk_forward_alloc += delta; 1568 } 1569 1570 sk->sk_forward_alloc -= size; 1571 1572 /* no need to setup a destructor, we will explicitly release the 1573 * forward allocated memory on dequeue 1574 */ 1575 sock_skb_set_dropcount(sk, skb); 1576 1577 __skb_queue_tail(list, skb); 1578 spin_unlock(&list->lock); 1579 1580 if (!sock_flag(sk, SOCK_DEAD)) 1581 sk->sk_data_ready(sk); 1582 1583 busylock_release(busy); 1584 return 0; 1585 1586 uncharge_drop: 1587 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1588 1589 drop: 1590 atomic_inc(&sk->sk_drops); 1591 busylock_release(busy); 1592 return err; 1593 } 1594 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1595 1596 void udp_destruct_sock(struct sock *sk) 1597 { 1598 /* reclaim completely the forward allocated memory */ 1599 struct udp_sock *up = udp_sk(sk); 1600 unsigned int total = 0; 1601 struct sk_buff *skb; 1602 1603 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1604 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1605 total += skb->truesize; 1606 kfree_skb(skb); 1607 } 1608 udp_rmem_release(sk, total, 0, true); 1609 1610 inet_sock_destruct(sk); 1611 } 1612 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1613 1614 int udp_init_sock(struct sock *sk) 1615 { 1616 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1617 sk->sk_destruct = udp_destruct_sock; 1618 return 0; 1619 } 1620 EXPORT_SYMBOL_GPL(udp_init_sock); 1621 1622 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1623 { 1624 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1625 bool slow = lock_sock_fast(sk); 1626 1627 sk_peek_offset_bwd(sk, len); 1628 unlock_sock_fast(sk, slow); 1629 } 1630 1631 if (!skb_unref(skb)) 1632 return; 1633 1634 /* In the more common cases we cleared the head states previously, 1635 * see __udp_queue_rcv_skb(). 1636 */ 1637 if (unlikely(udp_skb_has_head_state(skb))) 1638 skb_release_head_state(skb); 1639 __consume_stateless_skb(skb); 1640 } 1641 EXPORT_SYMBOL_GPL(skb_consume_udp); 1642 1643 static struct sk_buff *__first_packet_length(struct sock *sk, 1644 struct sk_buff_head *rcvq, 1645 int *total) 1646 { 1647 struct sk_buff *skb; 1648 1649 while ((skb = skb_peek(rcvq)) != NULL) { 1650 if (udp_lib_checksum_complete(skb)) { 1651 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1652 IS_UDPLITE(sk)); 1653 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1654 IS_UDPLITE(sk)); 1655 atomic_inc(&sk->sk_drops); 1656 __skb_unlink(skb, rcvq); 1657 *total += skb->truesize; 1658 kfree_skb(skb); 1659 } else { 1660 udp_skb_csum_unnecessary_set(skb); 1661 break; 1662 } 1663 } 1664 return skb; 1665 } 1666 1667 /** 1668 * first_packet_length - return length of first packet in receive queue 1669 * @sk: socket 1670 * 1671 * Drops all bad checksum frames, until a valid one is found. 1672 * Returns the length of found skb, or -1 if none is found. 1673 */ 1674 static int first_packet_length(struct sock *sk) 1675 { 1676 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1677 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1678 struct sk_buff *skb; 1679 int total = 0; 1680 int res; 1681 1682 spin_lock_bh(&rcvq->lock); 1683 skb = __first_packet_length(sk, rcvq, &total); 1684 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1685 spin_lock(&sk_queue->lock); 1686 skb_queue_splice_tail_init(sk_queue, rcvq); 1687 spin_unlock(&sk_queue->lock); 1688 1689 skb = __first_packet_length(sk, rcvq, &total); 1690 } 1691 res = skb ? skb->len : -1; 1692 if (total) 1693 udp_rmem_release(sk, total, 1, false); 1694 spin_unlock_bh(&rcvq->lock); 1695 return res; 1696 } 1697 1698 /* 1699 * IOCTL requests applicable to the UDP protocol 1700 */ 1701 1702 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1703 { 1704 switch (cmd) { 1705 case SIOCOUTQ: 1706 { 1707 int amount = sk_wmem_alloc_get(sk); 1708 1709 return put_user(amount, (int __user *)arg); 1710 } 1711 1712 case SIOCINQ: 1713 { 1714 int amount = max_t(int, 0, first_packet_length(sk)); 1715 1716 return put_user(amount, (int __user *)arg); 1717 } 1718 1719 default: 1720 return -ENOIOCTLCMD; 1721 } 1722 1723 return 0; 1724 } 1725 EXPORT_SYMBOL(udp_ioctl); 1726 1727 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1728 int noblock, int *off, int *err) 1729 { 1730 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1731 struct sk_buff_head *queue; 1732 struct sk_buff *last; 1733 long timeo; 1734 int error; 1735 1736 queue = &udp_sk(sk)->reader_queue; 1737 flags |= noblock ? MSG_DONTWAIT : 0; 1738 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1739 do { 1740 struct sk_buff *skb; 1741 1742 error = sock_error(sk); 1743 if (error) 1744 break; 1745 1746 error = -EAGAIN; 1747 do { 1748 spin_lock_bh(&queue->lock); 1749 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1750 err, &last); 1751 if (skb) { 1752 if (!(flags & MSG_PEEK)) 1753 udp_skb_destructor(sk, skb); 1754 spin_unlock_bh(&queue->lock); 1755 return skb; 1756 } 1757 1758 if (skb_queue_empty_lockless(sk_queue)) { 1759 spin_unlock_bh(&queue->lock); 1760 goto busy_check; 1761 } 1762 1763 /* refill the reader queue and walk it again 1764 * keep both queues locked to avoid re-acquiring 1765 * the sk_receive_queue lock if fwd memory scheduling 1766 * is needed. 1767 */ 1768 spin_lock(&sk_queue->lock); 1769 skb_queue_splice_tail_init(sk_queue, queue); 1770 1771 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1772 err, &last); 1773 if (skb && !(flags & MSG_PEEK)) 1774 udp_skb_dtor_locked(sk, skb); 1775 spin_unlock(&sk_queue->lock); 1776 spin_unlock_bh(&queue->lock); 1777 if (skb) 1778 return skb; 1779 1780 busy_check: 1781 if (!sk_can_busy_loop(sk)) 1782 break; 1783 1784 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1785 } while (!skb_queue_empty_lockless(sk_queue)); 1786 1787 /* sk_queue is empty, reader_queue may contain peeked packets */ 1788 } while (timeo && 1789 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1790 &error, &timeo, 1791 (struct sk_buff *)sk_queue)); 1792 1793 *err = error; 1794 return NULL; 1795 } 1796 EXPORT_SYMBOL(__skb_recv_udp); 1797 1798 int udp_read_sock(struct sock *sk, read_descriptor_t *desc, 1799 sk_read_actor_t recv_actor) 1800 { 1801 int copied = 0; 1802 1803 while (1) { 1804 struct sk_buff *skb; 1805 int err, used; 1806 1807 skb = skb_recv_udp(sk, 0, 1, &err); 1808 if (!skb) 1809 return err; 1810 1811 if (udp_lib_checksum_complete(skb)) { 1812 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1813 IS_UDPLITE(sk)); 1814 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1815 IS_UDPLITE(sk)); 1816 atomic_inc(&sk->sk_drops); 1817 kfree_skb(skb); 1818 continue; 1819 } 1820 1821 used = recv_actor(desc, skb, 0, skb->len); 1822 if (used <= 0) { 1823 if (!copied) 1824 copied = used; 1825 kfree_skb(skb); 1826 break; 1827 } else if (used <= skb->len) { 1828 copied += used; 1829 } 1830 1831 kfree_skb(skb); 1832 if (!desc->count) 1833 break; 1834 } 1835 1836 return copied; 1837 } 1838 EXPORT_SYMBOL(udp_read_sock); 1839 1840 /* 1841 * This should be easy, if there is something there we 1842 * return it, otherwise we block. 1843 */ 1844 1845 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1846 int flags, int *addr_len) 1847 { 1848 struct inet_sock *inet = inet_sk(sk); 1849 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1850 struct sk_buff *skb; 1851 unsigned int ulen, copied; 1852 int off, err, peeking = flags & MSG_PEEK; 1853 int is_udplite = IS_UDPLITE(sk); 1854 bool checksum_valid = false; 1855 1856 if (flags & MSG_ERRQUEUE) 1857 return ip_recv_error(sk, msg, len, addr_len); 1858 1859 try_again: 1860 off = sk_peek_offset(sk, flags); 1861 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 1862 if (!skb) 1863 return err; 1864 1865 ulen = udp_skb_len(skb); 1866 copied = len; 1867 if (copied > ulen - off) 1868 copied = ulen - off; 1869 else if (copied < ulen) 1870 msg->msg_flags |= MSG_TRUNC; 1871 1872 /* 1873 * If checksum is needed at all, try to do it while copying the 1874 * data. If the data is truncated, or if we only want a partial 1875 * coverage checksum (UDP-Lite), do it before the copy. 1876 */ 1877 1878 if (copied < ulen || peeking || 1879 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1880 checksum_valid = udp_skb_csum_unnecessary(skb) || 1881 !__udp_lib_checksum_complete(skb); 1882 if (!checksum_valid) 1883 goto csum_copy_err; 1884 } 1885 1886 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1887 if (udp_skb_is_linear(skb)) 1888 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1889 else 1890 err = skb_copy_datagram_msg(skb, off, msg, copied); 1891 } else { 1892 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1893 1894 if (err == -EINVAL) 1895 goto csum_copy_err; 1896 } 1897 1898 if (unlikely(err)) { 1899 if (!peeking) { 1900 atomic_inc(&sk->sk_drops); 1901 UDP_INC_STATS(sock_net(sk), 1902 UDP_MIB_INERRORS, is_udplite); 1903 } 1904 kfree_skb(skb); 1905 return err; 1906 } 1907 1908 if (!peeking) 1909 UDP_INC_STATS(sock_net(sk), 1910 UDP_MIB_INDATAGRAMS, is_udplite); 1911 1912 sock_recv_ts_and_drops(msg, sk, skb); 1913 1914 /* Copy the address. */ 1915 if (sin) { 1916 sin->sin_family = AF_INET; 1917 sin->sin_port = udp_hdr(skb)->source; 1918 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1919 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1920 *addr_len = sizeof(*sin); 1921 1922 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1923 (struct sockaddr *)sin); 1924 } 1925 1926 if (udp_sk(sk)->gro_enabled) 1927 udp_cmsg_recv(msg, sk, skb); 1928 1929 if (inet->cmsg_flags) 1930 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1931 1932 err = copied; 1933 if (flags & MSG_TRUNC) 1934 err = ulen; 1935 1936 skb_consume_udp(sk, skb, peeking ? -err : err); 1937 return err; 1938 1939 csum_copy_err: 1940 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1941 udp_skb_destructor)) { 1942 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1943 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1944 } 1945 kfree_skb(skb); 1946 1947 /* starting over for a new packet, but check if we need to yield */ 1948 cond_resched(); 1949 msg->msg_flags &= ~MSG_TRUNC; 1950 goto try_again; 1951 } 1952 1953 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1954 { 1955 /* This check is replicated from __ip4_datagram_connect() and 1956 * intended to prevent BPF program called below from accessing bytes 1957 * that are out of the bound specified by user in addr_len. 1958 */ 1959 if (addr_len < sizeof(struct sockaddr_in)) 1960 return -EINVAL; 1961 1962 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1963 } 1964 EXPORT_SYMBOL(udp_pre_connect); 1965 1966 int __udp_disconnect(struct sock *sk, int flags) 1967 { 1968 struct inet_sock *inet = inet_sk(sk); 1969 /* 1970 * 1003.1g - break association. 1971 */ 1972 1973 sk->sk_state = TCP_CLOSE; 1974 inet->inet_daddr = 0; 1975 inet->inet_dport = 0; 1976 sock_rps_reset_rxhash(sk); 1977 sk->sk_bound_dev_if = 0; 1978 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1979 inet_reset_saddr(sk); 1980 if (sk->sk_prot->rehash && 1981 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1982 sk->sk_prot->rehash(sk); 1983 } 1984 1985 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1986 sk->sk_prot->unhash(sk); 1987 inet->inet_sport = 0; 1988 } 1989 sk_dst_reset(sk); 1990 return 0; 1991 } 1992 EXPORT_SYMBOL(__udp_disconnect); 1993 1994 int udp_disconnect(struct sock *sk, int flags) 1995 { 1996 lock_sock(sk); 1997 __udp_disconnect(sk, flags); 1998 release_sock(sk); 1999 return 0; 2000 } 2001 EXPORT_SYMBOL(udp_disconnect); 2002 2003 void udp_lib_unhash(struct sock *sk) 2004 { 2005 if (sk_hashed(sk)) { 2006 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2007 struct udp_hslot *hslot, *hslot2; 2008 2009 hslot = udp_hashslot(udptable, sock_net(sk), 2010 udp_sk(sk)->udp_port_hash); 2011 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2012 2013 spin_lock_bh(&hslot->lock); 2014 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2015 reuseport_detach_sock(sk); 2016 if (sk_del_node_init_rcu(sk)) { 2017 hslot->count--; 2018 inet_sk(sk)->inet_num = 0; 2019 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 2020 2021 spin_lock(&hslot2->lock); 2022 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2023 hslot2->count--; 2024 spin_unlock(&hslot2->lock); 2025 } 2026 spin_unlock_bh(&hslot->lock); 2027 } 2028 } 2029 EXPORT_SYMBOL(udp_lib_unhash); 2030 2031 /* 2032 * inet_rcv_saddr was changed, we must rehash secondary hash 2033 */ 2034 void udp_lib_rehash(struct sock *sk, u16 newhash) 2035 { 2036 if (sk_hashed(sk)) { 2037 struct udp_table *udptable = sk->sk_prot->h.udp_table; 2038 struct udp_hslot *hslot, *hslot2, *nhslot2; 2039 2040 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 2041 nhslot2 = udp_hashslot2(udptable, newhash); 2042 udp_sk(sk)->udp_portaddr_hash = newhash; 2043 2044 if (hslot2 != nhslot2 || 2045 rcu_access_pointer(sk->sk_reuseport_cb)) { 2046 hslot = udp_hashslot(udptable, sock_net(sk), 2047 udp_sk(sk)->udp_port_hash); 2048 /* we must lock primary chain too */ 2049 spin_lock_bh(&hslot->lock); 2050 if (rcu_access_pointer(sk->sk_reuseport_cb)) 2051 reuseport_detach_sock(sk); 2052 2053 if (hslot2 != nhslot2) { 2054 spin_lock(&hslot2->lock); 2055 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 2056 hslot2->count--; 2057 spin_unlock(&hslot2->lock); 2058 2059 spin_lock(&nhslot2->lock); 2060 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 2061 &nhslot2->head); 2062 nhslot2->count++; 2063 spin_unlock(&nhslot2->lock); 2064 } 2065 2066 spin_unlock_bh(&hslot->lock); 2067 } 2068 } 2069 } 2070 EXPORT_SYMBOL(udp_lib_rehash); 2071 2072 void udp_v4_rehash(struct sock *sk) 2073 { 2074 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 2075 inet_sk(sk)->inet_rcv_saddr, 2076 inet_sk(sk)->inet_num); 2077 udp_lib_rehash(sk, new_hash); 2078 } 2079 2080 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2081 { 2082 int rc; 2083 2084 if (inet_sk(sk)->inet_daddr) { 2085 sock_rps_save_rxhash(sk, skb); 2086 sk_mark_napi_id(sk, skb); 2087 sk_incoming_cpu_update(sk); 2088 } else { 2089 sk_mark_napi_id_once(sk, skb); 2090 } 2091 2092 rc = __udp_enqueue_schedule_skb(sk, skb); 2093 if (rc < 0) { 2094 int is_udplite = IS_UDPLITE(sk); 2095 2096 /* Note that an ENOMEM error is charged twice */ 2097 if (rc == -ENOMEM) 2098 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 2099 is_udplite); 2100 else 2101 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS, 2102 is_udplite); 2103 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2104 kfree_skb(skb); 2105 trace_udp_fail_queue_rcv_skb(rc, sk); 2106 return -1; 2107 } 2108 2109 return 0; 2110 } 2111 2112 /* returns: 2113 * -1: error 2114 * 0: success 2115 * >0: "udp encap" protocol resubmission 2116 * 2117 * Note that in the success and error cases, the skb is assumed to 2118 * have either been requeued or freed. 2119 */ 2120 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2121 { 2122 struct udp_sock *up = udp_sk(sk); 2123 int is_udplite = IS_UDPLITE(sk); 2124 2125 /* 2126 * Charge it to the socket, dropping if the queue is full. 2127 */ 2128 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 2129 goto drop; 2130 nf_reset_ct(skb); 2131 2132 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 2133 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2134 2135 /* 2136 * This is an encapsulation socket so pass the skb to 2137 * the socket's udp_encap_rcv() hook. Otherwise, just 2138 * fall through and pass this up the UDP socket. 2139 * up->encap_rcv() returns the following value: 2140 * =0 if skb was successfully passed to the encap 2141 * handler or was discarded by it. 2142 * >0 if skb should be passed on to UDP. 2143 * <0 if skb should be resubmitted as proto -N 2144 */ 2145 2146 /* if we're overly short, let UDP handle it */ 2147 encap_rcv = READ_ONCE(up->encap_rcv); 2148 if (encap_rcv) { 2149 int ret; 2150 2151 /* Verify checksum before giving to encap */ 2152 if (udp_lib_checksum_complete(skb)) 2153 goto csum_error; 2154 2155 ret = encap_rcv(sk, skb); 2156 if (ret <= 0) { 2157 __UDP_INC_STATS(sock_net(sk), 2158 UDP_MIB_INDATAGRAMS, 2159 is_udplite); 2160 return -ret; 2161 } 2162 } 2163 2164 /* FALLTHROUGH -- it's a UDP Packet */ 2165 } 2166 2167 /* 2168 * UDP-Lite specific tests, ignored on UDP sockets 2169 */ 2170 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2171 2172 /* 2173 * MIB statistics other than incrementing the error count are 2174 * disabled for the following two types of errors: these depend 2175 * on the application settings, not on the functioning of the 2176 * protocol stack as such. 2177 * 2178 * RFC 3828 here recommends (sec 3.3): "There should also be a 2179 * way ... to ... at least let the receiving application block 2180 * delivery of packets with coverage values less than a value 2181 * provided by the application." 2182 */ 2183 if (up->pcrlen == 0) { /* full coverage was set */ 2184 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2185 UDP_SKB_CB(skb)->cscov, skb->len); 2186 goto drop; 2187 } 2188 /* The next case involves violating the min. coverage requested 2189 * by the receiver. This is subtle: if receiver wants x and x is 2190 * greater than the buffersize/MTU then receiver will complain 2191 * that it wants x while sender emits packets of smaller size y. 2192 * Therefore the above ...()->partial_cov statement is essential. 2193 */ 2194 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2195 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2196 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2197 goto drop; 2198 } 2199 } 2200 2201 prefetch(&sk->sk_rmem_alloc); 2202 if (rcu_access_pointer(sk->sk_filter) && 2203 udp_lib_checksum_complete(skb)) 2204 goto csum_error; 2205 2206 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 2207 goto drop; 2208 2209 udp_csum_pull_header(skb); 2210 2211 ipv4_pktinfo_prepare(sk, skb); 2212 return __udp_queue_rcv_skb(sk, skb); 2213 2214 csum_error: 2215 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2216 drop: 2217 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2218 atomic_inc(&sk->sk_drops); 2219 kfree_skb(skb); 2220 return -1; 2221 } 2222 2223 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2224 { 2225 struct sk_buff *next, *segs; 2226 int ret; 2227 2228 if (likely(!udp_unexpected_gso(sk, skb))) 2229 return udp_queue_rcv_one_skb(sk, skb); 2230 2231 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2232 __skb_push(skb, -skb_mac_offset(skb)); 2233 segs = udp_rcv_segment(sk, skb, true); 2234 skb_list_walk_safe(segs, skb, next) { 2235 __skb_pull(skb, skb_transport_offset(skb)); 2236 2237 udp_post_segment_fix_csum(skb); 2238 ret = udp_queue_rcv_one_skb(sk, skb); 2239 if (ret > 0) 2240 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret); 2241 } 2242 return 0; 2243 } 2244 2245 /* For TCP sockets, sk_rx_dst is protected by socket lock 2246 * For UDP, we use xchg() to guard against concurrent changes. 2247 */ 2248 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2249 { 2250 struct dst_entry *old; 2251 2252 if (dst_hold_safe(dst)) { 2253 old = xchg(&sk->sk_rx_dst, dst); 2254 dst_release(old); 2255 return old != dst; 2256 } 2257 return false; 2258 } 2259 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2260 2261 /* 2262 * Multicasts and broadcasts go to each listener. 2263 * 2264 * Note: called only from the BH handler context. 2265 */ 2266 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2267 struct udphdr *uh, 2268 __be32 saddr, __be32 daddr, 2269 struct udp_table *udptable, 2270 int proto) 2271 { 2272 struct sock *sk, *first = NULL; 2273 unsigned short hnum = ntohs(uh->dest); 2274 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2275 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2276 unsigned int offset = offsetof(typeof(*sk), sk_node); 2277 int dif = skb->dev->ifindex; 2278 int sdif = inet_sdif(skb); 2279 struct hlist_node *node; 2280 struct sk_buff *nskb; 2281 2282 if (use_hash2) { 2283 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2284 udptable->mask; 2285 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2286 start_lookup: 2287 hslot = &udptable->hash2[hash2]; 2288 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2289 } 2290 2291 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2292 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2293 uh->source, saddr, dif, sdif, hnum)) 2294 continue; 2295 2296 if (!first) { 2297 first = sk; 2298 continue; 2299 } 2300 nskb = skb_clone(skb, GFP_ATOMIC); 2301 2302 if (unlikely(!nskb)) { 2303 atomic_inc(&sk->sk_drops); 2304 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2305 IS_UDPLITE(sk)); 2306 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2307 IS_UDPLITE(sk)); 2308 continue; 2309 } 2310 if (udp_queue_rcv_skb(sk, nskb) > 0) 2311 consume_skb(nskb); 2312 } 2313 2314 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2315 if (use_hash2 && hash2 != hash2_any) { 2316 hash2 = hash2_any; 2317 goto start_lookup; 2318 } 2319 2320 if (first) { 2321 if (udp_queue_rcv_skb(first, skb) > 0) 2322 consume_skb(skb); 2323 } else { 2324 kfree_skb(skb); 2325 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2326 proto == IPPROTO_UDPLITE); 2327 } 2328 return 0; 2329 } 2330 2331 /* Initialize UDP checksum. If exited with zero value (success), 2332 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2333 * Otherwise, csum completion requires checksumming packet body, 2334 * including udp header and folding it to skb->csum. 2335 */ 2336 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2337 int proto) 2338 { 2339 int err; 2340 2341 UDP_SKB_CB(skb)->partial_cov = 0; 2342 UDP_SKB_CB(skb)->cscov = skb->len; 2343 2344 if (proto == IPPROTO_UDPLITE) { 2345 err = udplite_checksum_init(skb, uh); 2346 if (err) 2347 return err; 2348 2349 if (UDP_SKB_CB(skb)->partial_cov) { 2350 skb->csum = inet_compute_pseudo(skb, proto); 2351 return 0; 2352 } 2353 } 2354 2355 /* Note, we are only interested in != 0 or == 0, thus the 2356 * force to int. 2357 */ 2358 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2359 inet_compute_pseudo); 2360 if (err) 2361 return err; 2362 2363 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2364 /* If SW calculated the value, we know it's bad */ 2365 if (skb->csum_complete_sw) 2366 return 1; 2367 2368 /* HW says the value is bad. Let's validate that. 2369 * skb->csum is no longer the full packet checksum, 2370 * so don't treat it as such. 2371 */ 2372 skb_checksum_complete_unset(skb); 2373 } 2374 2375 return 0; 2376 } 2377 2378 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2379 * return code conversion for ip layer consumption 2380 */ 2381 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2382 struct udphdr *uh) 2383 { 2384 int ret; 2385 2386 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2387 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2388 2389 ret = udp_queue_rcv_skb(sk, skb); 2390 2391 /* a return value > 0 means to resubmit the input, but 2392 * it wants the return to be -protocol, or 0 2393 */ 2394 if (ret > 0) 2395 return -ret; 2396 return 0; 2397 } 2398 2399 /* 2400 * All we need to do is get the socket, and then do a checksum. 2401 */ 2402 2403 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2404 int proto) 2405 { 2406 struct sock *sk; 2407 struct udphdr *uh; 2408 unsigned short ulen; 2409 struct rtable *rt = skb_rtable(skb); 2410 __be32 saddr, daddr; 2411 struct net *net = dev_net(skb->dev); 2412 bool refcounted; 2413 2414 /* 2415 * Validate the packet. 2416 */ 2417 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2418 goto drop; /* No space for header. */ 2419 2420 uh = udp_hdr(skb); 2421 ulen = ntohs(uh->len); 2422 saddr = ip_hdr(skb)->saddr; 2423 daddr = ip_hdr(skb)->daddr; 2424 2425 if (ulen > skb->len) 2426 goto short_packet; 2427 2428 if (proto == IPPROTO_UDP) { 2429 /* UDP validates ulen. */ 2430 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2431 goto short_packet; 2432 uh = udp_hdr(skb); 2433 } 2434 2435 if (udp4_csum_init(skb, uh, proto)) 2436 goto csum_error; 2437 2438 sk = skb_steal_sock(skb, &refcounted); 2439 if (sk) { 2440 struct dst_entry *dst = skb_dst(skb); 2441 int ret; 2442 2443 if (unlikely(sk->sk_rx_dst != dst)) 2444 udp_sk_rx_dst_set(sk, dst); 2445 2446 ret = udp_unicast_rcv_skb(sk, skb, uh); 2447 if (refcounted) 2448 sock_put(sk); 2449 return ret; 2450 } 2451 2452 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2453 return __udp4_lib_mcast_deliver(net, skb, uh, 2454 saddr, daddr, udptable, proto); 2455 2456 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2457 if (sk) 2458 return udp_unicast_rcv_skb(sk, skb, uh); 2459 2460 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2461 goto drop; 2462 nf_reset_ct(skb); 2463 2464 /* No socket. Drop packet silently, if checksum is wrong */ 2465 if (udp_lib_checksum_complete(skb)) 2466 goto csum_error; 2467 2468 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2469 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2470 2471 /* 2472 * Hmm. We got an UDP packet to a port to which we 2473 * don't wanna listen. Ignore it. 2474 */ 2475 kfree_skb(skb); 2476 return 0; 2477 2478 short_packet: 2479 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2480 proto == IPPROTO_UDPLITE ? "Lite" : "", 2481 &saddr, ntohs(uh->source), 2482 ulen, skb->len, 2483 &daddr, ntohs(uh->dest)); 2484 goto drop; 2485 2486 csum_error: 2487 /* 2488 * RFC1122: OK. Discards the bad packet silently (as far as 2489 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2490 */ 2491 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2492 proto == IPPROTO_UDPLITE ? "Lite" : "", 2493 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2494 ulen); 2495 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2496 drop: 2497 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2498 kfree_skb(skb); 2499 return 0; 2500 } 2501 2502 /* We can only early demux multicast if there is a single matching socket. 2503 * If more than one socket found returns NULL 2504 */ 2505 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2506 __be16 loc_port, __be32 loc_addr, 2507 __be16 rmt_port, __be32 rmt_addr, 2508 int dif, int sdif) 2509 { 2510 struct sock *sk, *result; 2511 unsigned short hnum = ntohs(loc_port); 2512 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2513 struct udp_hslot *hslot = &udp_table.hash[slot]; 2514 2515 /* Do not bother scanning a too big list */ 2516 if (hslot->count > 10) 2517 return NULL; 2518 2519 result = NULL; 2520 sk_for_each_rcu(sk, &hslot->head) { 2521 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2522 rmt_port, rmt_addr, dif, sdif, hnum)) { 2523 if (result) 2524 return NULL; 2525 result = sk; 2526 } 2527 } 2528 2529 return result; 2530 } 2531 2532 /* For unicast we should only early demux connected sockets or we can 2533 * break forwarding setups. The chains here can be long so only check 2534 * if the first socket is an exact match and if not move on. 2535 */ 2536 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2537 __be16 loc_port, __be32 loc_addr, 2538 __be16 rmt_port, __be32 rmt_addr, 2539 int dif, int sdif) 2540 { 2541 unsigned short hnum = ntohs(loc_port); 2542 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2543 unsigned int slot2 = hash2 & udp_table.mask; 2544 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2545 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2546 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2547 struct sock *sk; 2548 2549 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2550 if (INET_MATCH(sk, net, acookie, rmt_addr, 2551 loc_addr, ports, dif, sdif)) 2552 return sk; 2553 /* Only check first socket in chain */ 2554 break; 2555 } 2556 return NULL; 2557 } 2558 2559 int udp_v4_early_demux(struct sk_buff *skb) 2560 { 2561 struct net *net = dev_net(skb->dev); 2562 struct in_device *in_dev = NULL; 2563 const struct iphdr *iph; 2564 const struct udphdr *uh; 2565 struct sock *sk = NULL; 2566 struct dst_entry *dst; 2567 int dif = skb->dev->ifindex; 2568 int sdif = inet_sdif(skb); 2569 int ours; 2570 2571 /* validate the packet */ 2572 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2573 return 0; 2574 2575 iph = ip_hdr(skb); 2576 uh = udp_hdr(skb); 2577 2578 if (skb->pkt_type == PACKET_MULTICAST) { 2579 in_dev = __in_dev_get_rcu(skb->dev); 2580 2581 if (!in_dev) 2582 return 0; 2583 2584 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2585 iph->protocol); 2586 if (!ours) 2587 return 0; 2588 2589 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2590 uh->source, iph->saddr, 2591 dif, sdif); 2592 } else if (skb->pkt_type == PACKET_HOST) { 2593 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2594 uh->source, iph->saddr, dif, sdif); 2595 } 2596 2597 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2598 return 0; 2599 2600 skb->sk = sk; 2601 skb->destructor = sock_efree; 2602 dst = READ_ONCE(sk->sk_rx_dst); 2603 2604 if (dst) 2605 dst = dst_check(dst, 0); 2606 if (dst) { 2607 u32 itag = 0; 2608 2609 /* set noref for now. 2610 * any place which wants to hold dst has to call 2611 * dst_hold_safe() 2612 */ 2613 skb_dst_set_noref(skb, dst); 2614 2615 /* for unconnected multicast sockets we need to validate 2616 * the source on each packet 2617 */ 2618 if (!inet_sk(sk)->inet_daddr && in_dev) 2619 return ip_mc_validate_source(skb, iph->daddr, 2620 iph->saddr, 2621 iph->tos & IPTOS_RT_MASK, 2622 skb->dev, in_dev, &itag); 2623 } 2624 return 0; 2625 } 2626 2627 int udp_rcv(struct sk_buff *skb) 2628 { 2629 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2630 } 2631 2632 void udp_destroy_sock(struct sock *sk) 2633 { 2634 struct udp_sock *up = udp_sk(sk); 2635 bool slow = lock_sock_fast(sk); 2636 2637 /* protects from races with udp_abort() */ 2638 sock_set_flag(sk, SOCK_DEAD); 2639 udp_flush_pending_frames(sk); 2640 unlock_sock_fast(sk, slow); 2641 if (static_branch_unlikely(&udp_encap_needed_key)) { 2642 if (up->encap_type) { 2643 void (*encap_destroy)(struct sock *sk); 2644 encap_destroy = READ_ONCE(up->encap_destroy); 2645 if (encap_destroy) 2646 encap_destroy(sk); 2647 } 2648 if (up->encap_enabled) 2649 static_branch_dec(&udp_encap_needed_key); 2650 } 2651 } 2652 2653 /* 2654 * Socket option code for UDP 2655 */ 2656 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2657 sockptr_t optval, unsigned int optlen, 2658 int (*push_pending_frames)(struct sock *)) 2659 { 2660 struct udp_sock *up = udp_sk(sk); 2661 int val, valbool; 2662 int err = 0; 2663 int is_udplite = IS_UDPLITE(sk); 2664 2665 if (optlen < sizeof(int)) 2666 return -EINVAL; 2667 2668 if (copy_from_sockptr(&val, optval, sizeof(val))) 2669 return -EFAULT; 2670 2671 valbool = val ? 1 : 0; 2672 2673 switch (optname) { 2674 case UDP_CORK: 2675 if (val != 0) { 2676 WRITE_ONCE(up->corkflag, 1); 2677 } else { 2678 WRITE_ONCE(up->corkflag, 0); 2679 lock_sock(sk); 2680 push_pending_frames(sk); 2681 release_sock(sk); 2682 } 2683 break; 2684 2685 case UDP_ENCAP: 2686 switch (val) { 2687 case 0: 2688 #ifdef CONFIG_XFRM 2689 case UDP_ENCAP_ESPINUDP: 2690 case UDP_ENCAP_ESPINUDP_NON_IKE: 2691 #if IS_ENABLED(CONFIG_IPV6) 2692 if (sk->sk_family == AF_INET6) 2693 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; 2694 else 2695 #endif 2696 up->encap_rcv = xfrm4_udp_encap_rcv; 2697 #endif 2698 fallthrough; 2699 case UDP_ENCAP_L2TPINUDP: 2700 up->encap_type = val; 2701 lock_sock(sk); 2702 udp_tunnel_encap_enable(sk->sk_socket); 2703 release_sock(sk); 2704 break; 2705 default: 2706 err = -ENOPROTOOPT; 2707 break; 2708 } 2709 break; 2710 2711 case UDP_NO_CHECK6_TX: 2712 up->no_check6_tx = valbool; 2713 break; 2714 2715 case UDP_NO_CHECK6_RX: 2716 up->no_check6_rx = valbool; 2717 break; 2718 2719 case UDP_SEGMENT: 2720 if (val < 0 || val > USHRT_MAX) 2721 return -EINVAL; 2722 WRITE_ONCE(up->gso_size, val); 2723 break; 2724 2725 case UDP_GRO: 2726 lock_sock(sk); 2727 2728 /* when enabling GRO, accept the related GSO packet type */ 2729 if (valbool) 2730 udp_tunnel_encap_enable(sk->sk_socket); 2731 up->gro_enabled = valbool; 2732 up->accept_udp_l4 = valbool; 2733 release_sock(sk); 2734 break; 2735 2736 /* 2737 * UDP-Lite's partial checksum coverage (RFC 3828). 2738 */ 2739 /* The sender sets actual checksum coverage length via this option. 2740 * The case coverage > packet length is handled by send module. */ 2741 case UDPLITE_SEND_CSCOV: 2742 if (!is_udplite) /* Disable the option on UDP sockets */ 2743 return -ENOPROTOOPT; 2744 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2745 val = 8; 2746 else if (val > USHRT_MAX) 2747 val = USHRT_MAX; 2748 up->pcslen = val; 2749 up->pcflag |= UDPLITE_SEND_CC; 2750 break; 2751 2752 /* The receiver specifies a minimum checksum coverage value. To make 2753 * sense, this should be set to at least 8 (as done below). If zero is 2754 * used, this again means full checksum coverage. */ 2755 case UDPLITE_RECV_CSCOV: 2756 if (!is_udplite) /* Disable the option on UDP sockets */ 2757 return -ENOPROTOOPT; 2758 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2759 val = 8; 2760 else if (val > USHRT_MAX) 2761 val = USHRT_MAX; 2762 up->pcrlen = val; 2763 up->pcflag |= UDPLITE_RECV_CC; 2764 break; 2765 2766 default: 2767 err = -ENOPROTOOPT; 2768 break; 2769 } 2770 2771 return err; 2772 } 2773 EXPORT_SYMBOL(udp_lib_setsockopt); 2774 2775 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 2776 unsigned int optlen) 2777 { 2778 if (level == SOL_UDP || level == SOL_UDPLITE) 2779 return udp_lib_setsockopt(sk, level, optname, 2780 optval, optlen, 2781 udp_push_pending_frames); 2782 return ip_setsockopt(sk, level, optname, optval, optlen); 2783 } 2784 2785 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2786 char __user *optval, int __user *optlen) 2787 { 2788 struct udp_sock *up = udp_sk(sk); 2789 int val, len; 2790 2791 if (get_user(len, optlen)) 2792 return -EFAULT; 2793 2794 len = min_t(unsigned int, len, sizeof(int)); 2795 2796 if (len < 0) 2797 return -EINVAL; 2798 2799 switch (optname) { 2800 case UDP_CORK: 2801 val = READ_ONCE(up->corkflag); 2802 break; 2803 2804 case UDP_ENCAP: 2805 val = up->encap_type; 2806 break; 2807 2808 case UDP_NO_CHECK6_TX: 2809 val = up->no_check6_tx; 2810 break; 2811 2812 case UDP_NO_CHECK6_RX: 2813 val = up->no_check6_rx; 2814 break; 2815 2816 case UDP_SEGMENT: 2817 val = READ_ONCE(up->gso_size); 2818 break; 2819 2820 case UDP_GRO: 2821 val = up->gro_enabled; 2822 break; 2823 2824 /* The following two cannot be changed on UDP sockets, the return is 2825 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2826 case UDPLITE_SEND_CSCOV: 2827 val = up->pcslen; 2828 break; 2829 2830 case UDPLITE_RECV_CSCOV: 2831 val = up->pcrlen; 2832 break; 2833 2834 default: 2835 return -ENOPROTOOPT; 2836 } 2837 2838 if (put_user(len, optlen)) 2839 return -EFAULT; 2840 if (copy_to_user(optval, &val, len)) 2841 return -EFAULT; 2842 return 0; 2843 } 2844 EXPORT_SYMBOL(udp_lib_getsockopt); 2845 2846 int udp_getsockopt(struct sock *sk, int level, int optname, 2847 char __user *optval, int __user *optlen) 2848 { 2849 if (level == SOL_UDP || level == SOL_UDPLITE) 2850 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2851 return ip_getsockopt(sk, level, optname, optval, optlen); 2852 } 2853 2854 /** 2855 * udp_poll - wait for a UDP event. 2856 * @file: - file struct 2857 * @sock: - socket 2858 * @wait: - poll table 2859 * 2860 * This is same as datagram poll, except for the special case of 2861 * blocking sockets. If application is using a blocking fd 2862 * and a packet with checksum error is in the queue; 2863 * then it could get return from select indicating data available 2864 * but then block when reading it. Add special case code 2865 * to work around these arguably broken applications. 2866 */ 2867 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2868 { 2869 __poll_t mask = datagram_poll(file, sock, wait); 2870 struct sock *sk = sock->sk; 2871 2872 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2873 mask |= EPOLLIN | EPOLLRDNORM; 2874 2875 /* Check for false positives due to checksum errors */ 2876 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2877 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2878 mask &= ~(EPOLLIN | EPOLLRDNORM); 2879 2880 /* psock ingress_msg queue should not contain any bad checksum frames */ 2881 if (sk_is_readable(sk)) 2882 mask |= EPOLLIN | EPOLLRDNORM; 2883 return mask; 2884 2885 } 2886 EXPORT_SYMBOL(udp_poll); 2887 2888 int udp_abort(struct sock *sk, int err) 2889 { 2890 lock_sock(sk); 2891 2892 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing 2893 * with close() 2894 */ 2895 if (sock_flag(sk, SOCK_DEAD)) 2896 goto out; 2897 2898 sk->sk_err = err; 2899 sk_error_report(sk); 2900 __udp_disconnect(sk, 0); 2901 2902 out: 2903 release_sock(sk); 2904 2905 return 0; 2906 } 2907 EXPORT_SYMBOL_GPL(udp_abort); 2908 2909 struct proto udp_prot = { 2910 .name = "UDP", 2911 .owner = THIS_MODULE, 2912 .close = udp_lib_close, 2913 .pre_connect = udp_pre_connect, 2914 .connect = ip4_datagram_connect, 2915 .disconnect = udp_disconnect, 2916 .ioctl = udp_ioctl, 2917 .init = udp_init_sock, 2918 .destroy = udp_destroy_sock, 2919 .setsockopt = udp_setsockopt, 2920 .getsockopt = udp_getsockopt, 2921 .sendmsg = udp_sendmsg, 2922 .recvmsg = udp_recvmsg, 2923 .sendpage = udp_sendpage, 2924 .release_cb = ip4_datagram_release_cb, 2925 .hash = udp_lib_hash, 2926 .unhash = udp_lib_unhash, 2927 .rehash = udp_v4_rehash, 2928 .get_port = udp_v4_get_port, 2929 #ifdef CONFIG_BPF_SYSCALL 2930 .psock_update_sk_prot = udp_bpf_update_proto, 2931 #endif 2932 .memory_allocated = &udp_memory_allocated, 2933 .sysctl_mem = sysctl_udp_mem, 2934 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2935 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2936 .obj_size = sizeof(struct udp_sock), 2937 .h.udp_table = &udp_table, 2938 .diag_destroy = udp_abort, 2939 }; 2940 EXPORT_SYMBOL(udp_prot); 2941 2942 /* ------------------------------------------------------------------------ */ 2943 #ifdef CONFIG_PROC_FS 2944 2945 static struct sock *udp_get_first(struct seq_file *seq, int start) 2946 { 2947 struct sock *sk; 2948 struct udp_seq_afinfo *afinfo; 2949 struct udp_iter_state *state = seq->private; 2950 struct net *net = seq_file_net(seq); 2951 2952 if (state->bpf_seq_afinfo) 2953 afinfo = state->bpf_seq_afinfo; 2954 else 2955 afinfo = PDE_DATA(file_inode(seq->file)); 2956 2957 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; 2958 ++state->bucket) { 2959 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; 2960 2961 if (hlist_empty(&hslot->head)) 2962 continue; 2963 2964 spin_lock_bh(&hslot->lock); 2965 sk_for_each(sk, &hslot->head) { 2966 if (!net_eq(sock_net(sk), net)) 2967 continue; 2968 if (afinfo->family == AF_UNSPEC || 2969 sk->sk_family == afinfo->family) 2970 goto found; 2971 } 2972 spin_unlock_bh(&hslot->lock); 2973 } 2974 sk = NULL; 2975 found: 2976 return sk; 2977 } 2978 2979 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2980 { 2981 struct udp_seq_afinfo *afinfo; 2982 struct udp_iter_state *state = seq->private; 2983 struct net *net = seq_file_net(seq); 2984 2985 if (state->bpf_seq_afinfo) 2986 afinfo = state->bpf_seq_afinfo; 2987 else 2988 afinfo = PDE_DATA(file_inode(seq->file)); 2989 2990 do { 2991 sk = sk_next(sk); 2992 } while (sk && (!net_eq(sock_net(sk), net) || 2993 (afinfo->family != AF_UNSPEC && 2994 sk->sk_family != afinfo->family))); 2995 2996 if (!sk) { 2997 if (state->bucket <= afinfo->udp_table->mask) 2998 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 2999 return udp_get_first(seq, state->bucket + 1); 3000 } 3001 return sk; 3002 } 3003 3004 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 3005 { 3006 struct sock *sk = udp_get_first(seq, 0); 3007 3008 if (sk) 3009 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 3010 --pos; 3011 return pos ? NULL : sk; 3012 } 3013 3014 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 3015 { 3016 struct udp_iter_state *state = seq->private; 3017 state->bucket = MAX_UDP_PORTS; 3018 3019 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 3020 } 3021 EXPORT_SYMBOL(udp_seq_start); 3022 3023 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3024 { 3025 struct sock *sk; 3026 3027 if (v == SEQ_START_TOKEN) 3028 sk = udp_get_idx(seq, 0); 3029 else 3030 sk = udp_get_next(seq, v); 3031 3032 ++*pos; 3033 return sk; 3034 } 3035 EXPORT_SYMBOL(udp_seq_next); 3036 3037 void udp_seq_stop(struct seq_file *seq, void *v) 3038 { 3039 struct udp_seq_afinfo *afinfo; 3040 struct udp_iter_state *state = seq->private; 3041 3042 if (state->bpf_seq_afinfo) 3043 afinfo = state->bpf_seq_afinfo; 3044 else 3045 afinfo = PDE_DATA(file_inode(seq->file)); 3046 3047 if (state->bucket <= afinfo->udp_table->mask) 3048 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 3049 } 3050 EXPORT_SYMBOL(udp_seq_stop); 3051 3052 /* ------------------------------------------------------------------------ */ 3053 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 3054 int bucket) 3055 { 3056 struct inet_sock *inet = inet_sk(sp); 3057 __be32 dest = inet->inet_daddr; 3058 __be32 src = inet->inet_rcv_saddr; 3059 __u16 destp = ntohs(inet->inet_dport); 3060 __u16 srcp = ntohs(inet->inet_sport); 3061 3062 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 3063 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 3064 bucket, src, srcp, dest, destp, sp->sk_state, 3065 sk_wmem_alloc_get(sp), 3066 udp_rqueue_get(sp), 3067 0, 0L, 0, 3068 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 3069 0, sock_i_ino(sp), 3070 refcount_read(&sp->sk_refcnt), sp, 3071 atomic_read(&sp->sk_drops)); 3072 } 3073 3074 int udp4_seq_show(struct seq_file *seq, void *v) 3075 { 3076 seq_setwidth(seq, 127); 3077 if (v == SEQ_START_TOKEN) 3078 seq_puts(seq, " sl local_address rem_address st tx_queue " 3079 "rx_queue tr tm->when retrnsmt uid timeout " 3080 "inode ref pointer drops"); 3081 else { 3082 struct udp_iter_state *state = seq->private; 3083 3084 udp4_format_sock(v, seq, state->bucket); 3085 } 3086 seq_pad(seq, '\n'); 3087 return 0; 3088 } 3089 3090 #ifdef CONFIG_BPF_SYSCALL 3091 struct bpf_iter__udp { 3092 __bpf_md_ptr(struct bpf_iter_meta *, meta); 3093 __bpf_md_ptr(struct udp_sock *, udp_sk); 3094 uid_t uid __aligned(8); 3095 int bucket __aligned(8); 3096 }; 3097 3098 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta, 3099 struct udp_sock *udp_sk, uid_t uid, int bucket) 3100 { 3101 struct bpf_iter__udp ctx; 3102 3103 meta->seq_num--; /* skip SEQ_START_TOKEN */ 3104 ctx.meta = meta; 3105 ctx.udp_sk = udp_sk; 3106 ctx.uid = uid; 3107 ctx.bucket = bucket; 3108 return bpf_iter_run_prog(prog, &ctx); 3109 } 3110 3111 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v) 3112 { 3113 struct udp_iter_state *state = seq->private; 3114 struct bpf_iter_meta meta; 3115 struct bpf_prog *prog; 3116 struct sock *sk = v; 3117 uid_t uid; 3118 3119 if (v == SEQ_START_TOKEN) 3120 return 0; 3121 3122 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk)); 3123 meta.seq = seq; 3124 prog = bpf_iter_get_info(&meta, false); 3125 return udp_prog_seq_show(prog, &meta, v, uid, state->bucket); 3126 } 3127 3128 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v) 3129 { 3130 struct bpf_iter_meta meta; 3131 struct bpf_prog *prog; 3132 3133 if (!v) { 3134 meta.seq = seq; 3135 prog = bpf_iter_get_info(&meta, true); 3136 if (prog) 3137 (void)udp_prog_seq_show(prog, &meta, v, 0, 0); 3138 } 3139 3140 udp_seq_stop(seq, v); 3141 } 3142 3143 static const struct seq_operations bpf_iter_udp_seq_ops = { 3144 .start = udp_seq_start, 3145 .next = udp_seq_next, 3146 .stop = bpf_iter_udp_seq_stop, 3147 .show = bpf_iter_udp_seq_show, 3148 }; 3149 #endif 3150 3151 const struct seq_operations udp_seq_ops = { 3152 .start = udp_seq_start, 3153 .next = udp_seq_next, 3154 .stop = udp_seq_stop, 3155 .show = udp4_seq_show, 3156 }; 3157 EXPORT_SYMBOL(udp_seq_ops); 3158 3159 static struct udp_seq_afinfo udp4_seq_afinfo = { 3160 .family = AF_INET, 3161 .udp_table = &udp_table, 3162 }; 3163 3164 static int __net_init udp4_proc_init_net(struct net *net) 3165 { 3166 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 3167 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 3168 return -ENOMEM; 3169 return 0; 3170 } 3171 3172 static void __net_exit udp4_proc_exit_net(struct net *net) 3173 { 3174 remove_proc_entry("udp", net->proc_net); 3175 } 3176 3177 static struct pernet_operations udp4_net_ops = { 3178 .init = udp4_proc_init_net, 3179 .exit = udp4_proc_exit_net, 3180 }; 3181 3182 int __init udp4_proc_init(void) 3183 { 3184 return register_pernet_subsys(&udp4_net_ops); 3185 } 3186 3187 void udp4_proc_exit(void) 3188 { 3189 unregister_pernet_subsys(&udp4_net_ops); 3190 } 3191 #endif /* CONFIG_PROC_FS */ 3192 3193 static __initdata unsigned long uhash_entries; 3194 static int __init set_uhash_entries(char *str) 3195 { 3196 ssize_t ret; 3197 3198 if (!str) 3199 return 0; 3200 3201 ret = kstrtoul(str, 0, &uhash_entries); 3202 if (ret) 3203 return 0; 3204 3205 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3206 uhash_entries = UDP_HTABLE_SIZE_MIN; 3207 return 1; 3208 } 3209 __setup("uhash_entries=", set_uhash_entries); 3210 3211 void __init udp_table_init(struct udp_table *table, const char *name) 3212 { 3213 unsigned int i; 3214 3215 table->hash = alloc_large_system_hash(name, 3216 2 * sizeof(struct udp_hslot), 3217 uhash_entries, 3218 21, /* one slot per 2 MB */ 3219 0, 3220 &table->log, 3221 &table->mask, 3222 UDP_HTABLE_SIZE_MIN, 3223 64 * 1024); 3224 3225 table->hash2 = table->hash + (table->mask + 1); 3226 for (i = 0; i <= table->mask; i++) { 3227 INIT_HLIST_HEAD(&table->hash[i].head); 3228 table->hash[i].count = 0; 3229 spin_lock_init(&table->hash[i].lock); 3230 } 3231 for (i = 0; i <= table->mask; i++) { 3232 INIT_HLIST_HEAD(&table->hash2[i].head); 3233 table->hash2[i].count = 0; 3234 spin_lock_init(&table->hash2[i].lock); 3235 } 3236 } 3237 3238 u32 udp_flow_hashrnd(void) 3239 { 3240 static u32 hashrnd __read_mostly; 3241 3242 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3243 3244 return hashrnd; 3245 } 3246 EXPORT_SYMBOL(udp_flow_hashrnd); 3247 3248 static void __udp_sysctl_init(struct net *net) 3249 { 3250 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; 3251 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; 3252 3253 #ifdef CONFIG_NET_L3_MASTER_DEV 3254 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3255 #endif 3256 } 3257 3258 static int __net_init udp_sysctl_init(struct net *net) 3259 { 3260 __udp_sysctl_init(net); 3261 return 0; 3262 } 3263 3264 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3265 .init = udp_sysctl_init, 3266 }; 3267 3268 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3269 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta, 3270 struct udp_sock *udp_sk, uid_t uid, int bucket) 3271 3272 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux) 3273 { 3274 struct udp_iter_state *st = priv_data; 3275 struct udp_seq_afinfo *afinfo; 3276 int ret; 3277 3278 afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN); 3279 if (!afinfo) 3280 return -ENOMEM; 3281 3282 afinfo->family = AF_UNSPEC; 3283 afinfo->udp_table = &udp_table; 3284 st->bpf_seq_afinfo = afinfo; 3285 ret = bpf_iter_init_seq_net(priv_data, aux); 3286 if (ret) 3287 kfree(afinfo); 3288 return ret; 3289 } 3290 3291 static void bpf_iter_fini_udp(void *priv_data) 3292 { 3293 struct udp_iter_state *st = priv_data; 3294 3295 kfree(st->bpf_seq_afinfo); 3296 bpf_iter_fini_seq_net(priv_data); 3297 } 3298 3299 static const struct bpf_iter_seq_info udp_seq_info = { 3300 .seq_ops = &bpf_iter_udp_seq_ops, 3301 .init_seq_private = bpf_iter_init_udp, 3302 .fini_seq_private = bpf_iter_fini_udp, 3303 .seq_priv_size = sizeof(struct udp_iter_state), 3304 }; 3305 3306 static struct bpf_iter_reg udp_reg_info = { 3307 .target = "udp", 3308 .ctx_arg_info_size = 1, 3309 .ctx_arg_info = { 3310 { offsetof(struct bpf_iter__udp, udp_sk), 3311 PTR_TO_BTF_ID_OR_NULL }, 3312 }, 3313 .seq_info = &udp_seq_info, 3314 }; 3315 3316 static void __init bpf_iter_register(void) 3317 { 3318 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP]; 3319 if (bpf_iter_reg_target(&udp_reg_info)) 3320 pr_warn("Warning: could not register bpf iterator udp\n"); 3321 } 3322 #endif 3323 3324 void __init udp_init(void) 3325 { 3326 unsigned long limit; 3327 unsigned int i; 3328 3329 udp_table_init(&udp_table, "UDP"); 3330 limit = nr_free_buffer_pages() / 8; 3331 limit = max(limit, 128UL); 3332 sysctl_udp_mem[0] = limit / 4 * 3; 3333 sysctl_udp_mem[1] = limit; 3334 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3335 3336 __udp_sysctl_init(&init_net); 3337 3338 /* 16 spinlocks per cpu */ 3339 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3340 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3341 GFP_KERNEL); 3342 if (!udp_busylocks) 3343 panic("UDP: failed to alloc udp_busylocks\n"); 3344 for (i = 0; i < (1U << udp_busylocks_log); i++) 3345 spin_lock_init(udp_busylocks + i); 3346 3347 if (register_pernet_subsys(&udp_sysctl_ops)) 3348 panic("UDP: failed to init sysctl parameters.\n"); 3349 3350 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS) 3351 bpf_iter_register(); 3352 #endif 3353 } 3354