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/swap.h> 82 #include <linux/types.h> 83 #include <linux/fcntl.h> 84 #include <linux/module.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/igmp.h> 88 #include <linux/inetdevice.h> 89 #include <linux/in.h> 90 #include <linux/errno.h> 91 #include <linux/timer.h> 92 #include <linux/mm.h> 93 #include <linux/inet.h> 94 #include <linux/netdevice.h> 95 #include <linux/slab.h> 96 #include <net/tcp_states.h> 97 #include <linux/skbuff.h> 98 #include <linux/proc_fs.h> 99 #include <linux/seq_file.h> 100 #include <net/net_namespace.h> 101 #include <net/icmp.h> 102 #include <net/inet_hashtables.h> 103 #include <net/ip_tunnels.h> 104 #include <net/route.h> 105 #include <net/checksum.h> 106 #include <net/xfrm.h> 107 #include <trace/events/udp.h> 108 #include <linux/static_key.h> 109 #include <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 score += 4; 393 394 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 395 score++; 396 return score; 397 } 398 399 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 400 const __u16 lport, const __be32 faddr, 401 const __be16 fport) 402 { 403 static u32 udp_ehash_secret __read_mostly; 404 405 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 406 407 return __inet_ehashfn(laddr, lport, faddr, fport, 408 udp_ehash_secret + net_hash_mix(net)); 409 } 410 411 /* called with rcu_read_lock() */ 412 static struct sock *udp4_lib_lookup2(struct net *net, 413 __be32 saddr, __be16 sport, 414 __be32 daddr, unsigned int hnum, 415 int dif, int sdif, 416 struct udp_hslot *hslot2, 417 struct sk_buff *skb) 418 { 419 struct sock *sk, *result; 420 int score, badness; 421 u32 hash = 0; 422 423 result = NULL; 424 badness = 0; 425 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 426 score = compute_score(sk, net, saddr, sport, 427 daddr, hnum, dif, sdif); 428 if (score > badness) { 429 if (sk->sk_reuseport && 430 sk->sk_state != TCP_ESTABLISHED) { 431 hash = udp_ehashfn(net, daddr, hnum, 432 saddr, sport); 433 result = reuseport_select_sock(sk, hash, skb, 434 sizeof(struct udphdr)); 435 if (result && !reuseport_has_conns(sk, false)) 436 return result; 437 } 438 badness = score; 439 result = sk; 440 } 441 } 442 return result; 443 } 444 445 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 446 * harder than this. -DaveM 447 */ 448 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 449 __be16 sport, __be32 daddr, __be16 dport, int dif, 450 int sdif, struct udp_table *udptable, struct sk_buff *skb) 451 { 452 struct sock *result; 453 unsigned short hnum = ntohs(dport); 454 unsigned int hash2, slot2; 455 struct udp_hslot *hslot2; 456 457 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 458 slot2 = hash2 & udptable->mask; 459 hslot2 = &udptable->hash2[slot2]; 460 461 result = udp4_lib_lookup2(net, saddr, sport, 462 daddr, hnum, dif, sdif, 463 hslot2, skb); 464 if (!result) { 465 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 466 slot2 = hash2 & udptable->mask; 467 hslot2 = &udptable->hash2[slot2]; 468 469 result = udp4_lib_lookup2(net, saddr, sport, 470 htonl(INADDR_ANY), hnum, dif, sdif, 471 hslot2, skb); 472 } 473 if (IS_ERR(result)) 474 return NULL; 475 return result; 476 } 477 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 478 479 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 480 __be16 sport, __be16 dport, 481 struct udp_table *udptable) 482 { 483 const struct iphdr *iph = ip_hdr(skb); 484 485 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 486 iph->daddr, dport, inet_iif(skb), 487 inet_sdif(skb), udptable, skb); 488 } 489 490 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, 491 __be16 sport, __be16 dport) 492 { 493 const struct iphdr *iph = ip_hdr(skb); 494 495 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 496 iph->daddr, dport, inet_iif(skb), 497 inet_sdif(skb), &udp_table, NULL); 498 } 499 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); 500 501 /* Must be called under rcu_read_lock(). 502 * Does increment socket refcount. 503 */ 504 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 505 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 506 __be32 daddr, __be16 dport, int dif) 507 { 508 struct sock *sk; 509 510 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 511 dif, 0, &udp_table, NULL); 512 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 513 sk = NULL; 514 return sk; 515 } 516 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 517 #endif 518 519 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 520 __be16 loc_port, __be32 loc_addr, 521 __be16 rmt_port, __be32 rmt_addr, 522 int dif, int sdif, unsigned short hnum) 523 { 524 struct inet_sock *inet = inet_sk(sk); 525 526 if (!net_eq(sock_net(sk), net) || 527 udp_sk(sk)->udp_port_hash != hnum || 528 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 529 (inet->inet_dport != rmt_port && inet->inet_dport) || 530 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 531 ipv6_only_sock(sk) || 532 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 533 return false; 534 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 535 return false; 536 return true; 537 } 538 539 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 540 void udp_encap_enable(void) 541 { 542 static_branch_inc(&udp_encap_needed_key); 543 } 544 EXPORT_SYMBOL(udp_encap_enable); 545 546 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 547 * through error handlers in encapsulations looking for a match. 548 */ 549 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 550 { 551 int i; 552 553 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 554 int (*handler)(struct sk_buff *skb, u32 info); 555 const struct ip_tunnel_encap_ops *encap; 556 557 encap = rcu_dereference(iptun_encaps[i]); 558 if (!encap) 559 continue; 560 handler = encap->err_handler; 561 if (handler && !handler(skb, info)) 562 return 0; 563 } 564 565 return -ENOENT; 566 } 567 568 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 569 * reversing source and destination port: this will match tunnels that force the 570 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 571 * lwtunnels might actually break this assumption by being configured with 572 * different destination ports on endpoints, in this case we won't be able to 573 * trace ICMP messages back to them. 574 * 575 * If this doesn't match any socket, probe tunnels with arbitrary destination 576 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 577 * we've sent packets to won't necessarily match the local destination port. 578 * 579 * Then ask the tunnel implementation to match the error against a valid 580 * association. 581 * 582 * Return an error if we can't find a match, the socket if we need further 583 * processing, zero otherwise. 584 */ 585 static struct sock *__udp4_lib_err_encap(struct net *net, 586 const struct iphdr *iph, 587 struct udphdr *uh, 588 struct udp_table *udptable, 589 struct sk_buff *skb, u32 info) 590 { 591 int network_offset, transport_offset; 592 struct sock *sk; 593 594 network_offset = skb_network_offset(skb); 595 transport_offset = skb_transport_offset(skb); 596 597 /* Network header needs to point to the outer IPv4 header inside ICMP */ 598 skb_reset_network_header(skb); 599 600 /* Transport header needs to point to the UDP header */ 601 skb_set_transport_header(skb, iph->ihl << 2); 602 603 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 604 iph->saddr, uh->dest, skb->dev->ifindex, 0, 605 udptable, NULL); 606 if (sk) { 607 int (*lookup)(struct sock *sk, struct sk_buff *skb); 608 struct udp_sock *up = udp_sk(sk); 609 610 lookup = READ_ONCE(up->encap_err_lookup); 611 if (!lookup || lookup(sk, skb)) 612 sk = NULL; 613 } 614 615 if (!sk) 616 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 617 618 skb_set_transport_header(skb, transport_offset); 619 skb_set_network_header(skb, network_offset); 620 621 return sk; 622 } 623 624 /* 625 * This routine is called by the ICMP module when it gets some 626 * sort of error condition. If err < 0 then the socket should 627 * be closed and the error returned to the user. If err > 0 628 * it's just the icmp type << 8 | icmp code. 629 * Header points to the ip header of the error packet. We move 630 * on past this. Then (as it used to claim before adjustment) 631 * header points to the first 8 bytes of the udp header. We need 632 * to find the appropriate port. 633 */ 634 635 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 636 { 637 struct inet_sock *inet; 638 const struct iphdr *iph = (const struct iphdr *)skb->data; 639 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 640 const int type = icmp_hdr(skb)->type; 641 const int code = icmp_hdr(skb)->code; 642 bool tunnel = false; 643 struct sock *sk; 644 int harderr; 645 int err; 646 struct net *net = dev_net(skb->dev); 647 648 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 649 iph->saddr, uh->source, skb->dev->ifindex, 650 inet_sdif(skb), udptable, NULL); 651 if (!sk) { 652 /* No socket for error: try tunnels before discarding */ 653 sk = ERR_PTR(-ENOENT); 654 if (static_branch_unlikely(&udp_encap_needed_key)) { 655 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb, 656 info); 657 if (!sk) 658 return 0; 659 } 660 661 if (IS_ERR(sk)) { 662 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 663 return PTR_ERR(sk); 664 } 665 666 tunnel = true; 667 } 668 669 err = 0; 670 harderr = 0; 671 inet = inet_sk(sk); 672 673 switch (type) { 674 default: 675 case ICMP_TIME_EXCEEDED: 676 err = EHOSTUNREACH; 677 break; 678 case ICMP_SOURCE_QUENCH: 679 goto out; 680 case ICMP_PARAMETERPROB: 681 err = EPROTO; 682 harderr = 1; 683 break; 684 case ICMP_DEST_UNREACH: 685 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 686 ipv4_sk_update_pmtu(skb, sk, info); 687 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 688 err = EMSGSIZE; 689 harderr = 1; 690 break; 691 } 692 goto out; 693 } 694 err = EHOSTUNREACH; 695 if (code <= NR_ICMP_UNREACH) { 696 harderr = icmp_err_convert[code].fatal; 697 err = icmp_err_convert[code].errno; 698 } 699 break; 700 case ICMP_REDIRECT: 701 ipv4_sk_redirect(skb, sk); 702 goto out; 703 } 704 705 /* 706 * RFC1122: OK. Passes ICMP errors back to application, as per 707 * 4.1.3.3. 708 */ 709 if (tunnel) { 710 /* ...not for tunnels though: we don't have a sending socket */ 711 goto out; 712 } 713 if (!inet->recverr) { 714 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 715 goto out; 716 } else 717 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 718 719 sk->sk_err = err; 720 sk->sk_error_report(sk); 721 out: 722 return 0; 723 } 724 725 int udp_err(struct sk_buff *skb, u32 info) 726 { 727 return __udp4_lib_err(skb, info, &udp_table); 728 } 729 730 /* 731 * Throw away all pending data and cancel the corking. Socket is locked. 732 */ 733 void udp_flush_pending_frames(struct sock *sk) 734 { 735 struct udp_sock *up = udp_sk(sk); 736 737 if (up->pending) { 738 up->len = 0; 739 up->pending = 0; 740 ip_flush_pending_frames(sk); 741 } 742 } 743 EXPORT_SYMBOL(udp_flush_pending_frames); 744 745 /** 746 * udp4_hwcsum - handle outgoing HW checksumming 747 * @skb: sk_buff containing the filled-in UDP header 748 * (checksum field must be zeroed out) 749 * @src: source IP address 750 * @dst: destination IP address 751 */ 752 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 753 { 754 struct udphdr *uh = udp_hdr(skb); 755 int offset = skb_transport_offset(skb); 756 int len = skb->len - offset; 757 int hlen = len; 758 __wsum csum = 0; 759 760 if (!skb_has_frag_list(skb)) { 761 /* 762 * Only one fragment on the socket. 763 */ 764 skb->csum_start = skb_transport_header(skb) - skb->head; 765 skb->csum_offset = offsetof(struct udphdr, check); 766 uh->check = ~csum_tcpudp_magic(src, dst, len, 767 IPPROTO_UDP, 0); 768 } else { 769 struct sk_buff *frags; 770 771 /* 772 * HW-checksum won't work as there are two or more 773 * fragments on the socket so that all csums of sk_buffs 774 * should be together 775 */ 776 skb_walk_frags(skb, frags) { 777 csum = csum_add(csum, frags->csum); 778 hlen -= frags->len; 779 } 780 781 csum = skb_checksum(skb, offset, hlen, csum); 782 skb->ip_summed = CHECKSUM_NONE; 783 784 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 785 if (uh->check == 0) 786 uh->check = CSUM_MANGLED_0; 787 } 788 } 789 EXPORT_SYMBOL_GPL(udp4_hwcsum); 790 791 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 792 * for the simple case like when setting the checksum for a UDP tunnel. 793 */ 794 void udp_set_csum(bool nocheck, struct sk_buff *skb, 795 __be32 saddr, __be32 daddr, int len) 796 { 797 struct udphdr *uh = udp_hdr(skb); 798 799 if (nocheck) { 800 uh->check = 0; 801 } else if (skb_is_gso(skb)) { 802 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 803 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 804 uh->check = 0; 805 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 806 if (uh->check == 0) 807 uh->check = CSUM_MANGLED_0; 808 } else { 809 skb->ip_summed = CHECKSUM_PARTIAL; 810 skb->csum_start = skb_transport_header(skb) - skb->head; 811 skb->csum_offset = offsetof(struct udphdr, check); 812 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 813 } 814 } 815 EXPORT_SYMBOL(udp_set_csum); 816 817 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 818 struct inet_cork *cork) 819 { 820 struct sock *sk = skb->sk; 821 struct inet_sock *inet = inet_sk(sk); 822 struct udphdr *uh; 823 int err = 0; 824 int is_udplite = IS_UDPLITE(sk); 825 int offset = skb_transport_offset(skb); 826 int len = skb->len - offset; 827 int datalen = len - sizeof(*uh); 828 __wsum csum = 0; 829 830 /* 831 * Create a UDP header 832 */ 833 uh = udp_hdr(skb); 834 uh->source = inet->inet_sport; 835 uh->dest = fl4->fl4_dport; 836 uh->len = htons(len); 837 uh->check = 0; 838 839 if (cork->gso_size) { 840 const int hlen = skb_network_header_len(skb) + 841 sizeof(struct udphdr); 842 843 if (hlen + cork->gso_size > cork->fragsize) { 844 kfree_skb(skb); 845 return -EINVAL; 846 } 847 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 848 kfree_skb(skb); 849 return -EINVAL; 850 } 851 if (sk->sk_no_check_tx) { 852 kfree_skb(skb); 853 return -EINVAL; 854 } 855 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 856 dst_xfrm(skb_dst(skb))) { 857 kfree_skb(skb); 858 return -EIO; 859 } 860 861 if (datalen > cork->gso_size) { 862 skb_shinfo(skb)->gso_size = cork->gso_size; 863 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 864 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 865 cork->gso_size); 866 } 867 goto csum_partial; 868 } 869 870 if (is_udplite) /* UDP-Lite */ 871 csum = udplite_csum(skb); 872 873 else if (sk->sk_no_check_tx) { /* UDP csum off */ 874 875 skb->ip_summed = CHECKSUM_NONE; 876 goto send; 877 878 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 879 csum_partial: 880 881 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 882 goto send; 883 884 } else 885 csum = udp_csum(skb); 886 887 /* add protocol-dependent pseudo-header */ 888 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 889 sk->sk_protocol, csum); 890 if (uh->check == 0) 891 uh->check = CSUM_MANGLED_0; 892 893 send: 894 err = ip_send_skb(sock_net(sk), skb); 895 if (err) { 896 if (err == -ENOBUFS && !inet->recverr) { 897 UDP_INC_STATS(sock_net(sk), 898 UDP_MIB_SNDBUFERRORS, is_udplite); 899 err = 0; 900 } 901 } else 902 UDP_INC_STATS(sock_net(sk), 903 UDP_MIB_OUTDATAGRAMS, is_udplite); 904 return err; 905 } 906 907 /* 908 * Push out all pending data as one UDP datagram. Socket is locked. 909 */ 910 int udp_push_pending_frames(struct sock *sk) 911 { 912 struct udp_sock *up = udp_sk(sk); 913 struct inet_sock *inet = inet_sk(sk); 914 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 915 struct sk_buff *skb; 916 int err = 0; 917 918 skb = ip_finish_skb(sk, fl4); 919 if (!skb) 920 goto out; 921 922 err = udp_send_skb(skb, fl4, &inet->cork.base); 923 924 out: 925 up->len = 0; 926 up->pending = 0; 927 return err; 928 } 929 EXPORT_SYMBOL(udp_push_pending_frames); 930 931 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 932 { 933 switch (cmsg->cmsg_type) { 934 case UDP_SEGMENT: 935 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 936 return -EINVAL; 937 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 938 return 0; 939 default: 940 return -EINVAL; 941 } 942 } 943 944 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 945 { 946 struct cmsghdr *cmsg; 947 bool need_ip = false; 948 int err; 949 950 for_each_cmsghdr(cmsg, msg) { 951 if (!CMSG_OK(msg, cmsg)) 952 return -EINVAL; 953 954 if (cmsg->cmsg_level != SOL_UDP) { 955 need_ip = true; 956 continue; 957 } 958 959 err = __udp_cmsg_send(cmsg, gso_size); 960 if (err) 961 return err; 962 } 963 964 return need_ip; 965 } 966 EXPORT_SYMBOL_GPL(udp_cmsg_send); 967 968 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 969 { 970 struct inet_sock *inet = inet_sk(sk); 971 struct udp_sock *up = udp_sk(sk); 972 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 973 struct flowi4 fl4_stack; 974 struct flowi4 *fl4; 975 int ulen = len; 976 struct ipcm_cookie ipc; 977 struct rtable *rt = NULL; 978 int free = 0; 979 int connected = 0; 980 __be32 daddr, faddr, saddr; 981 __be16 dport; 982 u8 tos; 983 int err, is_udplite = IS_UDPLITE(sk); 984 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 985 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 986 struct sk_buff *skb; 987 struct ip_options_data opt_copy; 988 989 if (len > 0xFFFF) 990 return -EMSGSIZE; 991 992 /* 993 * Check the flags. 994 */ 995 996 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 997 return -EOPNOTSUPP; 998 999 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1000 1001 fl4 = &inet->cork.fl.u.ip4; 1002 if (up->pending) { 1003 /* 1004 * There are pending frames. 1005 * The socket lock must be held while it's corked. 1006 */ 1007 lock_sock(sk); 1008 if (likely(up->pending)) { 1009 if (unlikely(up->pending != AF_INET)) { 1010 release_sock(sk); 1011 return -EINVAL; 1012 } 1013 goto do_append_data; 1014 } 1015 release_sock(sk); 1016 } 1017 ulen += sizeof(struct udphdr); 1018 1019 /* 1020 * Get and verify the address. 1021 */ 1022 if (usin) { 1023 if (msg->msg_namelen < sizeof(*usin)) 1024 return -EINVAL; 1025 if (usin->sin_family != AF_INET) { 1026 if (usin->sin_family != AF_UNSPEC) 1027 return -EAFNOSUPPORT; 1028 } 1029 1030 daddr = usin->sin_addr.s_addr; 1031 dport = usin->sin_port; 1032 if (dport == 0) 1033 return -EINVAL; 1034 } else { 1035 if (sk->sk_state != TCP_ESTABLISHED) 1036 return -EDESTADDRREQ; 1037 daddr = inet->inet_daddr; 1038 dport = inet->inet_dport; 1039 /* Open fast path for connected socket. 1040 Route will not be used, if at least one option is set. 1041 */ 1042 connected = 1; 1043 } 1044 1045 ipcm_init_sk(&ipc, inet); 1046 ipc.gso_size = up->gso_size; 1047 1048 if (msg->msg_controllen) { 1049 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1050 if (err > 0) 1051 err = ip_cmsg_send(sk, msg, &ipc, 1052 sk->sk_family == AF_INET6); 1053 if (unlikely(err < 0)) { 1054 kfree(ipc.opt); 1055 return err; 1056 } 1057 if (ipc.opt) 1058 free = 1; 1059 connected = 0; 1060 } 1061 if (!ipc.opt) { 1062 struct ip_options_rcu *inet_opt; 1063 1064 rcu_read_lock(); 1065 inet_opt = rcu_dereference(inet->inet_opt); 1066 if (inet_opt) { 1067 memcpy(&opt_copy, inet_opt, 1068 sizeof(*inet_opt) + inet_opt->opt.optlen); 1069 ipc.opt = &opt_copy.opt; 1070 } 1071 rcu_read_unlock(); 1072 } 1073 1074 if (cgroup_bpf_enabled && !connected) { 1075 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1076 (struct sockaddr *)usin, &ipc.addr); 1077 if (err) 1078 goto out_free; 1079 if (usin) { 1080 if (usin->sin_port == 0) { 1081 /* BPF program set invalid port. Reject it. */ 1082 err = -EINVAL; 1083 goto out_free; 1084 } 1085 daddr = usin->sin_addr.s_addr; 1086 dport = usin->sin_port; 1087 } 1088 } 1089 1090 saddr = ipc.addr; 1091 ipc.addr = faddr = daddr; 1092 1093 if (ipc.opt && ipc.opt->opt.srr) { 1094 if (!daddr) { 1095 err = -EINVAL; 1096 goto out_free; 1097 } 1098 faddr = ipc.opt->opt.faddr; 1099 connected = 0; 1100 } 1101 tos = get_rttos(&ipc, inet); 1102 if (sock_flag(sk, SOCK_LOCALROUTE) || 1103 (msg->msg_flags & MSG_DONTROUTE) || 1104 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1105 tos |= RTO_ONLINK; 1106 connected = 0; 1107 } 1108 1109 if (ipv4_is_multicast(daddr)) { 1110 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1111 ipc.oif = inet->mc_index; 1112 if (!saddr) 1113 saddr = inet->mc_addr; 1114 connected = 0; 1115 } else if (!ipc.oif) { 1116 ipc.oif = inet->uc_index; 1117 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1118 /* oif is set, packet is to local broadcast and 1119 * and uc_index is set. oif is most likely set 1120 * by sk_bound_dev_if. If uc_index != oif check if the 1121 * oif is an L3 master and uc_index is an L3 slave. 1122 * If so, we want to allow the send using the uc_index. 1123 */ 1124 if (ipc.oif != inet->uc_index && 1125 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1126 inet->uc_index)) { 1127 ipc.oif = inet->uc_index; 1128 } 1129 } 1130 1131 if (connected) 1132 rt = (struct rtable *)sk_dst_check(sk, 0); 1133 1134 if (!rt) { 1135 struct net *net = sock_net(sk); 1136 __u8 flow_flags = inet_sk_flowi_flags(sk); 1137 1138 fl4 = &fl4_stack; 1139 1140 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, 1141 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1142 flow_flags, 1143 faddr, saddr, dport, inet->inet_sport, 1144 sk->sk_uid); 1145 1146 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1147 rt = ip_route_output_flow(net, fl4, sk); 1148 if (IS_ERR(rt)) { 1149 err = PTR_ERR(rt); 1150 rt = NULL; 1151 if (err == -ENETUNREACH) 1152 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1153 goto out; 1154 } 1155 1156 err = -EACCES; 1157 if ((rt->rt_flags & RTCF_BROADCAST) && 1158 !sock_flag(sk, SOCK_BROADCAST)) 1159 goto out; 1160 if (connected) 1161 sk_dst_set(sk, dst_clone(&rt->dst)); 1162 } 1163 1164 if (msg->msg_flags&MSG_CONFIRM) 1165 goto do_confirm; 1166 back_from_confirm: 1167 1168 saddr = fl4->saddr; 1169 if (!ipc.addr) 1170 daddr = ipc.addr = fl4->daddr; 1171 1172 /* Lockless fast path for the non-corking case. */ 1173 if (!corkreq) { 1174 struct inet_cork cork; 1175 1176 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1177 sizeof(struct udphdr), &ipc, &rt, 1178 &cork, msg->msg_flags); 1179 err = PTR_ERR(skb); 1180 if (!IS_ERR_OR_NULL(skb)) 1181 err = udp_send_skb(skb, fl4, &cork); 1182 goto out; 1183 } 1184 1185 lock_sock(sk); 1186 if (unlikely(up->pending)) { 1187 /* The socket is already corked while preparing it. */ 1188 /* ... which is an evident application bug. --ANK */ 1189 release_sock(sk); 1190 1191 net_dbg_ratelimited("socket already corked\n"); 1192 err = -EINVAL; 1193 goto out; 1194 } 1195 /* 1196 * Now cork the socket to pend data. 1197 */ 1198 fl4 = &inet->cork.fl.u.ip4; 1199 fl4->daddr = daddr; 1200 fl4->saddr = saddr; 1201 fl4->fl4_dport = dport; 1202 fl4->fl4_sport = inet->inet_sport; 1203 up->pending = AF_INET; 1204 1205 do_append_data: 1206 up->len += ulen; 1207 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1208 sizeof(struct udphdr), &ipc, &rt, 1209 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1210 if (err) 1211 udp_flush_pending_frames(sk); 1212 else if (!corkreq) 1213 err = udp_push_pending_frames(sk); 1214 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1215 up->pending = 0; 1216 release_sock(sk); 1217 1218 out: 1219 ip_rt_put(rt); 1220 out_free: 1221 if (free) 1222 kfree(ipc.opt); 1223 if (!err) 1224 return len; 1225 /* 1226 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1227 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1228 * we don't have a good statistic (IpOutDiscards but it can be too many 1229 * things). We could add another new stat but at least for now that 1230 * seems like overkill. 1231 */ 1232 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1233 UDP_INC_STATS(sock_net(sk), 1234 UDP_MIB_SNDBUFERRORS, is_udplite); 1235 } 1236 return err; 1237 1238 do_confirm: 1239 if (msg->msg_flags & MSG_PROBE) 1240 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1241 if (!(msg->msg_flags&MSG_PROBE) || len) 1242 goto back_from_confirm; 1243 err = 0; 1244 goto out; 1245 } 1246 EXPORT_SYMBOL(udp_sendmsg); 1247 1248 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1249 size_t size, int flags) 1250 { 1251 struct inet_sock *inet = inet_sk(sk); 1252 struct udp_sock *up = udp_sk(sk); 1253 int ret; 1254 1255 if (flags & MSG_SENDPAGE_NOTLAST) 1256 flags |= MSG_MORE; 1257 1258 if (!up->pending) { 1259 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1260 1261 /* Call udp_sendmsg to specify destination address which 1262 * sendpage interface can't pass. 1263 * This will succeed only when the socket is connected. 1264 */ 1265 ret = udp_sendmsg(sk, &msg, 0); 1266 if (ret < 0) 1267 return ret; 1268 } 1269 1270 lock_sock(sk); 1271 1272 if (unlikely(!up->pending)) { 1273 release_sock(sk); 1274 1275 net_dbg_ratelimited("cork failed\n"); 1276 return -EINVAL; 1277 } 1278 1279 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1280 page, offset, size, flags); 1281 if (ret == -EOPNOTSUPP) { 1282 release_sock(sk); 1283 return sock_no_sendpage(sk->sk_socket, page, offset, 1284 size, flags); 1285 } 1286 if (ret < 0) { 1287 udp_flush_pending_frames(sk); 1288 goto out; 1289 } 1290 1291 up->len += size; 1292 if (!(up->corkflag || (flags&MSG_MORE))) 1293 ret = udp_push_pending_frames(sk); 1294 if (!ret) 1295 ret = size; 1296 out: 1297 release_sock(sk); 1298 return ret; 1299 } 1300 1301 #define UDP_SKB_IS_STATELESS 0x80000000 1302 1303 /* all head states (dst, sk, nf conntrack) except skb extensions are 1304 * cleared by udp_rcv(). 1305 * 1306 * We need to preserve secpath, if present, to eventually process 1307 * IP_CMSG_PASSSEC at recvmsg() time. 1308 * 1309 * Other extensions can be cleared. 1310 */ 1311 static bool udp_try_make_stateless(struct sk_buff *skb) 1312 { 1313 if (!skb_has_extensions(skb)) 1314 return true; 1315 1316 if (!secpath_exists(skb)) { 1317 skb_ext_reset(skb); 1318 return true; 1319 } 1320 1321 return false; 1322 } 1323 1324 static void udp_set_dev_scratch(struct sk_buff *skb) 1325 { 1326 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1327 1328 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1329 scratch->_tsize_state = skb->truesize; 1330 #if BITS_PER_LONG == 64 1331 scratch->len = skb->len; 1332 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1333 scratch->is_linear = !skb_is_nonlinear(skb); 1334 #endif 1335 if (udp_try_make_stateless(skb)) 1336 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1337 } 1338 1339 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb) 1340 { 1341 /* We come here after udp_lib_checksum_complete() returned 0. 1342 * This means that __skb_checksum_complete() might have 1343 * set skb->csum_valid to 1. 1344 * On 64bit platforms, we can set csum_unnecessary 1345 * to true, but only if the skb is not shared. 1346 */ 1347 #if BITS_PER_LONG == 64 1348 if (!skb_shared(skb)) 1349 udp_skb_scratch(skb)->csum_unnecessary = true; 1350 #endif 1351 } 1352 1353 static int udp_skb_truesize(struct sk_buff *skb) 1354 { 1355 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1356 } 1357 1358 static bool udp_skb_has_head_state(struct sk_buff *skb) 1359 { 1360 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1361 } 1362 1363 /* fully reclaim rmem/fwd memory allocated for skb */ 1364 static void udp_rmem_release(struct sock *sk, int size, int partial, 1365 bool rx_queue_lock_held) 1366 { 1367 struct udp_sock *up = udp_sk(sk); 1368 struct sk_buff_head *sk_queue; 1369 int amt; 1370 1371 if (likely(partial)) { 1372 up->forward_deficit += size; 1373 size = up->forward_deficit; 1374 if (size < (sk->sk_rcvbuf >> 2) && 1375 !skb_queue_empty(&up->reader_queue)) 1376 return; 1377 } else { 1378 size += up->forward_deficit; 1379 } 1380 up->forward_deficit = 0; 1381 1382 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1383 * if the called don't held it already 1384 */ 1385 sk_queue = &sk->sk_receive_queue; 1386 if (!rx_queue_lock_held) 1387 spin_lock(&sk_queue->lock); 1388 1389 1390 sk->sk_forward_alloc += size; 1391 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); 1392 sk->sk_forward_alloc -= amt; 1393 1394 if (amt) 1395 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); 1396 1397 atomic_sub(size, &sk->sk_rmem_alloc); 1398 1399 /* this can save us from acquiring the rx queue lock on next receive */ 1400 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1401 1402 if (!rx_queue_lock_held) 1403 spin_unlock(&sk_queue->lock); 1404 } 1405 1406 /* Note: called with reader_queue.lock held. 1407 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1408 * This avoids a cache line miss while receive_queue lock is held. 1409 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1410 */ 1411 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1412 { 1413 prefetch(&skb->data); 1414 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1415 } 1416 EXPORT_SYMBOL(udp_skb_destructor); 1417 1418 /* as above, but the caller held the rx queue lock, too */ 1419 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1420 { 1421 prefetch(&skb->data); 1422 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1423 } 1424 1425 /* Idea of busylocks is to let producers grab an extra spinlock 1426 * to relieve pressure on the receive_queue spinlock shared by consumer. 1427 * Under flood, this means that only one producer can be in line 1428 * trying to acquire the receive_queue spinlock. 1429 * These busylock can be allocated on a per cpu manner, instead of a 1430 * per socket one (that would consume a cache line per socket) 1431 */ 1432 static int udp_busylocks_log __read_mostly; 1433 static spinlock_t *udp_busylocks __read_mostly; 1434 1435 static spinlock_t *busylock_acquire(void *ptr) 1436 { 1437 spinlock_t *busy; 1438 1439 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1440 spin_lock(busy); 1441 return busy; 1442 } 1443 1444 static void busylock_release(spinlock_t *busy) 1445 { 1446 if (busy) 1447 spin_unlock(busy); 1448 } 1449 1450 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1451 { 1452 struct sk_buff_head *list = &sk->sk_receive_queue; 1453 int rmem, delta, amt, err = -ENOMEM; 1454 spinlock_t *busy = NULL; 1455 int size; 1456 1457 /* try to avoid the costly atomic add/sub pair when the receive 1458 * queue is full; always allow at least a packet 1459 */ 1460 rmem = atomic_read(&sk->sk_rmem_alloc); 1461 if (rmem > sk->sk_rcvbuf) 1462 goto drop; 1463 1464 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1465 * having linear skbs : 1466 * - Reduce memory overhead and thus increase receive queue capacity 1467 * - Less cache line misses at copyout() time 1468 * - Less work at consume_skb() (less alien page frag freeing) 1469 */ 1470 if (rmem > (sk->sk_rcvbuf >> 1)) { 1471 skb_condense(skb); 1472 1473 busy = busylock_acquire(sk); 1474 } 1475 size = skb->truesize; 1476 udp_set_dev_scratch(skb); 1477 1478 /* we drop only if the receive buf is full and the receive 1479 * queue contains some other skb 1480 */ 1481 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1482 if (rmem > (size + (unsigned int)sk->sk_rcvbuf)) 1483 goto uncharge_drop; 1484 1485 spin_lock(&list->lock); 1486 if (size >= sk->sk_forward_alloc) { 1487 amt = sk_mem_pages(size); 1488 delta = amt << SK_MEM_QUANTUM_SHIFT; 1489 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1490 err = -ENOBUFS; 1491 spin_unlock(&list->lock); 1492 goto uncharge_drop; 1493 } 1494 1495 sk->sk_forward_alloc += delta; 1496 } 1497 1498 sk->sk_forward_alloc -= size; 1499 1500 /* no need to setup a destructor, we will explicitly release the 1501 * forward allocated memory on dequeue 1502 */ 1503 sock_skb_set_dropcount(sk, skb); 1504 1505 __skb_queue_tail(list, skb); 1506 spin_unlock(&list->lock); 1507 1508 if (!sock_flag(sk, SOCK_DEAD)) 1509 sk->sk_data_ready(sk); 1510 1511 busylock_release(busy); 1512 return 0; 1513 1514 uncharge_drop: 1515 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1516 1517 drop: 1518 atomic_inc(&sk->sk_drops); 1519 busylock_release(busy); 1520 return err; 1521 } 1522 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1523 1524 void udp_destruct_sock(struct sock *sk) 1525 { 1526 /* reclaim completely the forward allocated memory */ 1527 struct udp_sock *up = udp_sk(sk); 1528 unsigned int total = 0; 1529 struct sk_buff *skb; 1530 1531 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1532 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1533 total += skb->truesize; 1534 kfree_skb(skb); 1535 } 1536 udp_rmem_release(sk, total, 0, true); 1537 1538 inet_sock_destruct(sk); 1539 } 1540 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1541 1542 int udp_init_sock(struct sock *sk) 1543 { 1544 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1545 sk->sk_destruct = udp_destruct_sock; 1546 return 0; 1547 } 1548 EXPORT_SYMBOL_GPL(udp_init_sock); 1549 1550 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1551 { 1552 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1553 bool slow = lock_sock_fast(sk); 1554 1555 sk_peek_offset_bwd(sk, len); 1556 unlock_sock_fast(sk, slow); 1557 } 1558 1559 if (!skb_unref(skb)) 1560 return; 1561 1562 /* In the more common cases we cleared the head states previously, 1563 * see __udp_queue_rcv_skb(). 1564 */ 1565 if (unlikely(udp_skb_has_head_state(skb))) 1566 skb_release_head_state(skb); 1567 __consume_stateless_skb(skb); 1568 } 1569 EXPORT_SYMBOL_GPL(skb_consume_udp); 1570 1571 static struct sk_buff *__first_packet_length(struct sock *sk, 1572 struct sk_buff_head *rcvq, 1573 int *total) 1574 { 1575 struct sk_buff *skb; 1576 1577 while ((skb = skb_peek(rcvq)) != NULL) { 1578 if (udp_lib_checksum_complete(skb)) { 1579 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1580 IS_UDPLITE(sk)); 1581 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1582 IS_UDPLITE(sk)); 1583 atomic_inc(&sk->sk_drops); 1584 __skb_unlink(skb, rcvq); 1585 *total += skb->truesize; 1586 kfree_skb(skb); 1587 } else { 1588 udp_skb_csum_unnecessary_set(skb); 1589 break; 1590 } 1591 } 1592 return skb; 1593 } 1594 1595 /** 1596 * first_packet_length - return length of first packet in receive queue 1597 * @sk: socket 1598 * 1599 * Drops all bad checksum frames, until a valid one is found. 1600 * Returns the length of found skb, or -1 if none is found. 1601 */ 1602 static int first_packet_length(struct sock *sk) 1603 { 1604 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1605 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1606 struct sk_buff *skb; 1607 int total = 0; 1608 int res; 1609 1610 spin_lock_bh(&rcvq->lock); 1611 skb = __first_packet_length(sk, rcvq, &total); 1612 if (!skb && !skb_queue_empty_lockless(sk_queue)) { 1613 spin_lock(&sk_queue->lock); 1614 skb_queue_splice_tail_init(sk_queue, rcvq); 1615 spin_unlock(&sk_queue->lock); 1616 1617 skb = __first_packet_length(sk, rcvq, &total); 1618 } 1619 res = skb ? skb->len : -1; 1620 if (total) 1621 udp_rmem_release(sk, total, 1, false); 1622 spin_unlock_bh(&rcvq->lock); 1623 return res; 1624 } 1625 1626 /* 1627 * IOCTL requests applicable to the UDP protocol 1628 */ 1629 1630 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1631 { 1632 switch (cmd) { 1633 case SIOCOUTQ: 1634 { 1635 int amount = sk_wmem_alloc_get(sk); 1636 1637 return put_user(amount, (int __user *)arg); 1638 } 1639 1640 case SIOCINQ: 1641 { 1642 int amount = max_t(int, 0, first_packet_length(sk)); 1643 1644 return put_user(amount, (int __user *)arg); 1645 } 1646 1647 default: 1648 return -ENOIOCTLCMD; 1649 } 1650 1651 return 0; 1652 } 1653 EXPORT_SYMBOL(udp_ioctl); 1654 1655 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1656 int noblock, int *off, int *err) 1657 { 1658 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1659 struct sk_buff_head *queue; 1660 struct sk_buff *last; 1661 long timeo; 1662 int error; 1663 1664 queue = &udp_sk(sk)->reader_queue; 1665 flags |= noblock ? MSG_DONTWAIT : 0; 1666 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1667 do { 1668 struct sk_buff *skb; 1669 1670 error = sock_error(sk); 1671 if (error) 1672 break; 1673 1674 error = -EAGAIN; 1675 do { 1676 spin_lock_bh(&queue->lock); 1677 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1678 err, &last); 1679 if (skb) { 1680 if (!(flags & MSG_PEEK)) 1681 udp_skb_destructor(sk, skb); 1682 spin_unlock_bh(&queue->lock); 1683 return skb; 1684 } 1685 1686 if (skb_queue_empty_lockless(sk_queue)) { 1687 spin_unlock_bh(&queue->lock); 1688 goto busy_check; 1689 } 1690 1691 /* refill the reader queue and walk it again 1692 * keep both queues locked to avoid re-acquiring 1693 * the sk_receive_queue lock if fwd memory scheduling 1694 * is needed. 1695 */ 1696 spin_lock(&sk_queue->lock); 1697 skb_queue_splice_tail_init(sk_queue, queue); 1698 1699 skb = __skb_try_recv_from_queue(sk, queue, flags, off, 1700 err, &last); 1701 if (skb && !(flags & MSG_PEEK)) 1702 udp_skb_dtor_locked(sk, skb); 1703 spin_unlock(&sk_queue->lock); 1704 spin_unlock_bh(&queue->lock); 1705 if (skb) 1706 return skb; 1707 1708 busy_check: 1709 if (!sk_can_busy_loop(sk)) 1710 break; 1711 1712 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1713 } while (!skb_queue_empty_lockless(sk_queue)); 1714 1715 /* sk_queue is empty, reader_queue may contain peeked packets */ 1716 } while (timeo && 1717 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue, 1718 &error, &timeo, 1719 (struct sk_buff *)sk_queue)); 1720 1721 *err = error; 1722 return NULL; 1723 } 1724 EXPORT_SYMBOL(__skb_recv_udp); 1725 1726 /* 1727 * This should be easy, if there is something there we 1728 * return it, otherwise we block. 1729 */ 1730 1731 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1732 int flags, int *addr_len) 1733 { 1734 struct inet_sock *inet = inet_sk(sk); 1735 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1736 struct sk_buff *skb; 1737 unsigned int ulen, copied; 1738 int off, err, peeking = flags & MSG_PEEK; 1739 int is_udplite = IS_UDPLITE(sk); 1740 bool checksum_valid = false; 1741 1742 if (flags & MSG_ERRQUEUE) 1743 return ip_recv_error(sk, msg, len, addr_len); 1744 1745 try_again: 1746 off = sk_peek_offset(sk, flags); 1747 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 1748 if (!skb) 1749 return err; 1750 1751 ulen = udp_skb_len(skb); 1752 copied = len; 1753 if (copied > ulen - off) 1754 copied = ulen - off; 1755 else if (copied < ulen) 1756 msg->msg_flags |= MSG_TRUNC; 1757 1758 /* 1759 * If checksum is needed at all, try to do it while copying the 1760 * data. If the data is truncated, or if we only want a partial 1761 * coverage checksum (UDP-Lite), do it before the copy. 1762 */ 1763 1764 if (copied < ulen || peeking || 1765 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1766 checksum_valid = udp_skb_csum_unnecessary(skb) || 1767 !__udp_lib_checksum_complete(skb); 1768 if (!checksum_valid) 1769 goto csum_copy_err; 1770 } 1771 1772 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1773 if (udp_skb_is_linear(skb)) 1774 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1775 else 1776 err = skb_copy_datagram_msg(skb, off, msg, copied); 1777 } else { 1778 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1779 1780 if (err == -EINVAL) 1781 goto csum_copy_err; 1782 } 1783 1784 if (unlikely(err)) { 1785 if (!peeking) { 1786 atomic_inc(&sk->sk_drops); 1787 UDP_INC_STATS(sock_net(sk), 1788 UDP_MIB_INERRORS, is_udplite); 1789 } 1790 kfree_skb(skb); 1791 return err; 1792 } 1793 1794 if (!peeking) 1795 UDP_INC_STATS(sock_net(sk), 1796 UDP_MIB_INDATAGRAMS, is_udplite); 1797 1798 sock_recv_ts_and_drops(msg, sk, skb); 1799 1800 /* Copy the address. */ 1801 if (sin) { 1802 sin->sin_family = AF_INET; 1803 sin->sin_port = udp_hdr(skb)->source; 1804 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1805 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1806 *addr_len = sizeof(*sin); 1807 1808 if (cgroup_bpf_enabled) 1809 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, 1810 (struct sockaddr *)sin); 1811 } 1812 1813 if (udp_sk(sk)->gro_enabled) 1814 udp_cmsg_recv(msg, sk, skb); 1815 1816 if (inet->cmsg_flags) 1817 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1818 1819 err = copied; 1820 if (flags & MSG_TRUNC) 1821 err = ulen; 1822 1823 skb_consume_udp(sk, skb, peeking ? -err : err); 1824 return err; 1825 1826 csum_copy_err: 1827 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1828 udp_skb_destructor)) { 1829 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1830 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1831 } 1832 kfree_skb(skb); 1833 1834 /* starting over for a new packet, but check if we need to yield */ 1835 cond_resched(); 1836 msg->msg_flags &= ~MSG_TRUNC; 1837 goto try_again; 1838 } 1839 1840 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1841 { 1842 /* This check is replicated from __ip4_datagram_connect() and 1843 * intended to prevent BPF program called below from accessing bytes 1844 * that are out of the bound specified by user in addr_len. 1845 */ 1846 if (addr_len < sizeof(struct sockaddr_in)) 1847 return -EINVAL; 1848 1849 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1850 } 1851 EXPORT_SYMBOL(udp_pre_connect); 1852 1853 int __udp_disconnect(struct sock *sk, int flags) 1854 { 1855 struct inet_sock *inet = inet_sk(sk); 1856 /* 1857 * 1003.1g - break association. 1858 */ 1859 1860 sk->sk_state = TCP_CLOSE; 1861 inet->inet_daddr = 0; 1862 inet->inet_dport = 0; 1863 sock_rps_reset_rxhash(sk); 1864 sk->sk_bound_dev_if = 0; 1865 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) { 1866 inet_reset_saddr(sk); 1867 if (sk->sk_prot->rehash && 1868 (sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 1869 sk->sk_prot->rehash(sk); 1870 } 1871 1872 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1873 sk->sk_prot->unhash(sk); 1874 inet->inet_sport = 0; 1875 } 1876 sk_dst_reset(sk); 1877 return 0; 1878 } 1879 EXPORT_SYMBOL(__udp_disconnect); 1880 1881 int udp_disconnect(struct sock *sk, int flags) 1882 { 1883 lock_sock(sk); 1884 __udp_disconnect(sk, flags); 1885 release_sock(sk); 1886 return 0; 1887 } 1888 EXPORT_SYMBOL(udp_disconnect); 1889 1890 void udp_lib_unhash(struct sock *sk) 1891 { 1892 if (sk_hashed(sk)) { 1893 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1894 struct udp_hslot *hslot, *hslot2; 1895 1896 hslot = udp_hashslot(udptable, sock_net(sk), 1897 udp_sk(sk)->udp_port_hash); 1898 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1899 1900 spin_lock_bh(&hslot->lock); 1901 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1902 reuseport_detach_sock(sk); 1903 if (sk_del_node_init_rcu(sk)) { 1904 hslot->count--; 1905 inet_sk(sk)->inet_num = 0; 1906 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1907 1908 spin_lock(&hslot2->lock); 1909 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1910 hslot2->count--; 1911 spin_unlock(&hslot2->lock); 1912 } 1913 spin_unlock_bh(&hslot->lock); 1914 } 1915 } 1916 EXPORT_SYMBOL(udp_lib_unhash); 1917 1918 /* 1919 * inet_rcv_saddr was changed, we must rehash secondary hash 1920 */ 1921 void udp_lib_rehash(struct sock *sk, u16 newhash) 1922 { 1923 if (sk_hashed(sk)) { 1924 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1925 struct udp_hslot *hslot, *hslot2, *nhslot2; 1926 1927 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1928 nhslot2 = udp_hashslot2(udptable, newhash); 1929 udp_sk(sk)->udp_portaddr_hash = newhash; 1930 1931 if (hslot2 != nhslot2 || 1932 rcu_access_pointer(sk->sk_reuseport_cb)) { 1933 hslot = udp_hashslot(udptable, sock_net(sk), 1934 udp_sk(sk)->udp_port_hash); 1935 /* we must lock primary chain too */ 1936 spin_lock_bh(&hslot->lock); 1937 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1938 reuseport_detach_sock(sk); 1939 1940 if (hslot2 != nhslot2) { 1941 spin_lock(&hslot2->lock); 1942 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1943 hslot2->count--; 1944 spin_unlock(&hslot2->lock); 1945 1946 spin_lock(&nhslot2->lock); 1947 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1948 &nhslot2->head); 1949 nhslot2->count++; 1950 spin_unlock(&nhslot2->lock); 1951 } 1952 1953 spin_unlock_bh(&hslot->lock); 1954 } 1955 } 1956 } 1957 EXPORT_SYMBOL(udp_lib_rehash); 1958 1959 void udp_v4_rehash(struct sock *sk) 1960 { 1961 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 1962 inet_sk(sk)->inet_rcv_saddr, 1963 inet_sk(sk)->inet_num); 1964 udp_lib_rehash(sk, new_hash); 1965 } 1966 1967 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1968 { 1969 int rc; 1970 1971 if (inet_sk(sk)->inet_daddr) { 1972 sock_rps_save_rxhash(sk, skb); 1973 sk_mark_napi_id(sk, skb); 1974 sk_incoming_cpu_update(sk); 1975 } else { 1976 sk_mark_napi_id_once(sk, skb); 1977 } 1978 1979 rc = __udp_enqueue_schedule_skb(sk, skb); 1980 if (rc < 0) { 1981 int is_udplite = IS_UDPLITE(sk); 1982 1983 /* Note that an ENOMEM error is charged twice */ 1984 if (rc == -ENOMEM) 1985 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1986 is_udplite); 1987 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1988 kfree_skb(skb); 1989 trace_udp_fail_queue_rcv_skb(rc, sk); 1990 return -1; 1991 } 1992 1993 return 0; 1994 } 1995 1996 /* returns: 1997 * -1: error 1998 * 0: success 1999 * >0: "udp encap" protocol resubmission 2000 * 2001 * Note that in the success and error cases, the skb is assumed to 2002 * have either been requeued or freed. 2003 */ 2004 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 2005 { 2006 struct udp_sock *up = udp_sk(sk); 2007 int is_udplite = IS_UDPLITE(sk); 2008 2009 /* 2010 * Charge it to the socket, dropping if the queue is full. 2011 */ 2012 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 2013 goto drop; 2014 nf_reset_ct(skb); 2015 2016 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 2017 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 2018 2019 /* 2020 * This is an encapsulation socket so pass the skb to 2021 * the socket's udp_encap_rcv() hook. Otherwise, just 2022 * fall through and pass this up the UDP socket. 2023 * up->encap_rcv() returns the following value: 2024 * =0 if skb was successfully passed to the encap 2025 * handler or was discarded by it. 2026 * >0 if skb should be passed on to UDP. 2027 * <0 if skb should be resubmitted as proto -N 2028 */ 2029 2030 /* if we're overly short, let UDP handle it */ 2031 encap_rcv = READ_ONCE(up->encap_rcv); 2032 if (encap_rcv) { 2033 int ret; 2034 2035 /* Verify checksum before giving to encap */ 2036 if (udp_lib_checksum_complete(skb)) 2037 goto csum_error; 2038 2039 ret = encap_rcv(sk, skb); 2040 if (ret <= 0) { 2041 __UDP_INC_STATS(sock_net(sk), 2042 UDP_MIB_INDATAGRAMS, 2043 is_udplite); 2044 return -ret; 2045 } 2046 } 2047 2048 /* FALLTHROUGH -- it's a UDP Packet */ 2049 } 2050 2051 /* 2052 * UDP-Lite specific tests, ignored on UDP sockets 2053 */ 2054 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2055 2056 /* 2057 * MIB statistics other than incrementing the error count are 2058 * disabled for the following two types of errors: these depend 2059 * on the application settings, not on the functioning of the 2060 * protocol stack as such. 2061 * 2062 * RFC 3828 here recommends (sec 3.3): "There should also be a 2063 * way ... to ... at least let the receiving application block 2064 * delivery of packets with coverage values less than a value 2065 * provided by the application." 2066 */ 2067 if (up->pcrlen == 0) { /* full coverage was set */ 2068 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2069 UDP_SKB_CB(skb)->cscov, skb->len); 2070 goto drop; 2071 } 2072 /* The next case involves violating the min. coverage requested 2073 * by the receiver. This is subtle: if receiver wants x and x is 2074 * greater than the buffersize/MTU then receiver will complain 2075 * that it wants x while sender emits packets of smaller size y. 2076 * Therefore the above ...()->partial_cov statement is essential. 2077 */ 2078 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2079 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2080 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2081 goto drop; 2082 } 2083 } 2084 2085 prefetch(&sk->sk_rmem_alloc); 2086 if (rcu_access_pointer(sk->sk_filter) && 2087 udp_lib_checksum_complete(skb)) 2088 goto csum_error; 2089 2090 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 2091 goto drop; 2092 2093 udp_csum_pull_header(skb); 2094 2095 ipv4_pktinfo_prepare(sk, skb); 2096 return __udp_queue_rcv_skb(sk, skb); 2097 2098 csum_error: 2099 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2100 drop: 2101 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2102 atomic_inc(&sk->sk_drops); 2103 kfree_skb(skb); 2104 return -1; 2105 } 2106 2107 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2108 { 2109 struct sk_buff *next, *segs; 2110 int ret; 2111 2112 if (likely(!udp_unexpected_gso(sk, skb))) 2113 return udp_queue_rcv_one_skb(sk, skb); 2114 2115 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET); 2116 __skb_push(skb, -skb_mac_offset(skb)); 2117 segs = udp_rcv_segment(sk, skb, true); 2118 skb_list_walk_safe(segs, skb, next) { 2119 __skb_pull(skb, skb_transport_offset(skb)); 2120 ret = udp_queue_rcv_one_skb(sk, skb); 2121 if (ret > 0) 2122 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret); 2123 } 2124 return 0; 2125 } 2126 2127 /* For TCP sockets, sk_rx_dst is protected by socket lock 2128 * For UDP, we use xchg() to guard against concurrent changes. 2129 */ 2130 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2131 { 2132 struct dst_entry *old; 2133 2134 if (dst_hold_safe(dst)) { 2135 old = xchg(&sk->sk_rx_dst, dst); 2136 dst_release(old); 2137 return old != dst; 2138 } 2139 return false; 2140 } 2141 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2142 2143 /* 2144 * Multicasts and broadcasts go to each listener. 2145 * 2146 * Note: called only from the BH handler context. 2147 */ 2148 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2149 struct udphdr *uh, 2150 __be32 saddr, __be32 daddr, 2151 struct udp_table *udptable, 2152 int proto) 2153 { 2154 struct sock *sk, *first = NULL; 2155 unsigned short hnum = ntohs(uh->dest); 2156 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2157 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2158 unsigned int offset = offsetof(typeof(*sk), sk_node); 2159 int dif = skb->dev->ifindex; 2160 int sdif = inet_sdif(skb); 2161 struct hlist_node *node; 2162 struct sk_buff *nskb; 2163 2164 if (use_hash2) { 2165 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2166 udptable->mask; 2167 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2168 start_lookup: 2169 hslot = &udptable->hash2[hash2]; 2170 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2171 } 2172 2173 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2174 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2175 uh->source, saddr, dif, sdif, hnum)) 2176 continue; 2177 2178 if (!first) { 2179 first = sk; 2180 continue; 2181 } 2182 nskb = skb_clone(skb, GFP_ATOMIC); 2183 2184 if (unlikely(!nskb)) { 2185 atomic_inc(&sk->sk_drops); 2186 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2187 IS_UDPLITE(sk)); 2188 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2189 IS_UDPLITE(sk)); 2190 continue; 2191 } 2192 if (udp_queue_rcv_skb(sk, nskb) > 0) 2193 consume_skb(nskb); 2194 } 2195 2196 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2197 if (use_hash2 && hash2 != hash2_any) { 2198 hash2 = hash2_any; 2199 goto start_lookup; 2200 } 2201 2202 if (first) { 2203 if (udp_queue_rcv_skb(first, skb) > 0) 2204 consume_skb(skb); 2205 } else { 2206 kfree_skb(skb); 2207 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2208 proto == IPPROTO_UDPLITE); 2209 } 2210 return 0; 2211 } 2212 2213 /* Initialize UDP checksum. If exited with zero value (success), 2214 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2215 * Otherwise, csum completion requires checksumming packet body, 2216 * including udp header and folding it to skb->csum. 2217 */ 2218 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2219 int proto) 2220 { 2221 int err; 2222 2223 UDP_SKB_CB(skb)->partial_cov = 0; 2224 UDP_SKB_CB(skb)->cscov = skb->len; 2225 2226 if (proto == IPPROTO_UDPLITE) { 2227 err = udplite_checksum_init(skb, uh); 2228 if (err) 2229 return err; 2230 2231 if (UDP_SKB_CB(skb)->partial_cov) { 2232 skb->csum = inet_compute_pseudo(skb, proto); 2233 return 0; 2234 } 2235 } 2236 2237 /* Note, we are only interested in != 0 or == 0, thus the 2238 * force to int. 2239 */ 2240 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2241 inet_compute_pseudo); 2242 if (err) 2243 return err; 2244 2245 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2246 /* If SW calculated the value, we know it's bad */ 2247 if (skb->csum_complete_sw) 2248 return 1; 2249 2250 /* HW says the value is bad. Let's validate that. 2251 * skb->csum is no longer the full packet checksum, 2252 * so don't treat it as such. 2253 */ 2254 skb_checksum_complete_unset(skb); 2255 } 2256 2257 return 0; 2258 } 2259 2260 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2261 * return code conversion for ip layer consumption 2262 */ 2263 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2264 struct udphdr *uh) 2265 { 2266 int ret; 2267 2268 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2269 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo); 2270 2271 ret = udp_queue_rcv_skb(sk, skb); 2272 2273 /* a return value > 0 means to resubmit the input, but 2274 * it wants the return to be -protocol, or 0 2275 */ 2276 if (ret > 0) 2277 return -ret; 2278 return 0; 2279 } 2280 2281 /* 2282 * All we need to do is get the socket, and then do a checksum. 2283 */ 2284 2285 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2286 int proto) 2287 { 2288 struct sock *sk; 2289 struct udphdr *uh; 2290 unsigned short ulen; 2291 struct rtable *rt = skb_rtable(skb); 2292 __be32 saddr, daddr; 2293 struct net *net = dev_net(skb->dev); 2294 bool refcounted; 2295 2296 /* 2297 * Validate the packet. 2298 */ 2299 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2300 goto drop; /* No space for header. */ 2301 2302 uh = udp_hdr(skb); 2303 ulen = ntohs(uh->len); 2304 saddr = ip_hdr(skb)->saddr; 2305 daddr = ip_hdr(skb)->daddr; 2306 2307 if (ulen > skb->len) 2308 goto short_packet; 2309 2310 if (proto == IPPROTO_UDP) { 2311 /* UDP validates ulen. */ 2312 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2313 goto short_packet; 2314 uh = udp_hdr(skb); 2315 } 2316 2317 if (udp4_csum_init(skb, uh, proto)) 2318 goto csum_error; 2319 2320 sk = skb_steal_sock(skb, &refcounted); 2321 if (sk) { 2322 struct dst_entry *dst = skb_dst(skb); 2323 int ret; 2324 2325 if (unlikely(sk->sk_rx_dst != dst)) 2326 udp_sk_rx_dst_set(sk, dst); 2327 2328 ret = udp_unicast_rcv_skb(sk, skb, uh); 2329 if (refcounted) 2330 sock_put(sk); 2331 return ret; 2332 } 2333 2334 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2335 return __udp4_lib_mcast_deliver(net, skb, uh, 2336 saddr, daddr, udptable, proto); 2337 2338 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2339 if (sk) 2340 return udp_unicast_rcv_skb(sk, skb, uh); 2341 2342 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2343 goto drop; 2344 nf_reset_ct(skb); 2345 2346 /* No socket. Drop packet silently, if checksum is wrong */ 2347 if (udp_lib_checksum_complete(skb)) 2348 goto csum_error; 2349 2350 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2351 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2352 2353 /* 2354 * Hmm. We got an UDP packet to a port to which we 2355 * don't wanna listen. Ignore it. 2356 */ 2357 kfree_skb(skb); 2358 return 0; 2359 2360 short_packet: 2361 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2362 proto == IPPROTO_UDPLITE ? "Lite" : "", 2363 &saddr, ntohs(uh->source), 2364 ulen, skb->len, 2365 &daddr, ntohs(uh->dest)); 2366 goto drop; 2367 2368 csum_error: 2369 /* 2370 * RFC1122: OK. Discards the bad packet silently (as far as 2371 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2372 */ 2373 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2374 proto == IPPROTO_UDPLITE ? "Lite" : "", 2375 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2376 ulen); 2377 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2378 drop: 2379 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2380 kfree_skb(skb); 2381 return 0; 2382 } 2383 2384 /* We can only early demux multicast if there is a single matching socket. 2385 * If more than one socket found returns NULL 2386 */ 2387 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2388 __be16 loc_port, __be32 loc_addr, 2389 __be16 rmt_port, __be32 rmt_addr, 2390 int dif, int sdif) 2391 { 2392 struct sock *sk, *result; 2393 unsigned short hnum = ntohs(loc_port); 2394 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2395 struct udp_hslot *hslot = &udp_table.hash[slot]; 2396 2397 /* Do not bother scanning a too big list */ 2398 if (hslot->count > 10) 2399 return NULL; 2400 2401 result = NULL; 2402 sk_for_each_rcu(sk, &hslot->head) { 2403 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2404 rmt_port, rmt_addr, dif, sdif, hnum)) { 2405 if (result) 2406 return NULL; 2407 result = sk; 2408 } 2409 } 2410 2411 return result; 2412 } 2413 2414 /* For unicast we should only early demux connected sockets or we can 2415 * break forwarding setups. The chains here can be long so only check 2416 * if the first socket is an exact match and if not move on. 2417 */ 2418 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2419 __be16 loc_port, __be32 loc_addr, 2420 __be16 rmt_port, __be32 rmt_addr, 2421 int dif, int sdif) 2422 { 2423 unsigned short hnum = ntohs(loc_port); 2424 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2425 unsigned int slot2 = hash2 & udp_table.mask; 2426 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2427 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2428 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2429 struct sock *sk; 2430 2431 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2432 if (INET_MATCH(sk, net, acookie, rmt_addr, 2433 loc_addr, ports, dif, sdif)) 2434 return sk; 2435 /* Only check first socket in chain */ 2436 break; 2437 } 2438 return NULL; 2439 } 2440 2441 int udp_v4_early_demux(struct sk_buff *skb) 2442 { 2443 struct net *net = dev_net(skb->dev); 2444 struct in_device *in_dev = NULL; 2445 const struct iphdr *iph; 2446 const struct udphdr *uh; 2447 struct sock *sk = NULL; 2448 struct dst_entry *dst; 2449 int dif = skb->dev->ifindex; 2450 int sdif = inet_sdif(skb); 2451 int ours; 2452 2453 /* validate the packet */ 2454 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2455 return 0; 2456 2457 iph = ip_hdr(skb); 2458 uh = udp_hdr(skb); 2459 2460 if (skb->pkt_type == PACKET_MULTICAST) { 2461 in_dev = __in_dev_get_rcu(skb->dev); 2462 2463 if (!in_dev) 2464 return 0; 2465 2466 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2467 iph->protocol); 2468 if (!ours) 2469 return 0; 2470 2471 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2472 uh->source, iph->saddr, 2473 dif, sdif); 2474 } else if (skb->pkt_type == PACKET_HOST) { 2475 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2476 uh->source, iph->saddr, dif, sdif); 2477 } 2478 2479 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2480 return 0; 2481 2482 skb->sk = sk; 2483 skb->destructor = sock_efree; 2484 dst = READ_ONCE(sk->sk_rx_dst); 2485 2486 if (dst) 2487 dst = dst_check(dst, 0); 2488 if (dst) { 2489 u32 itag = 0; 2490 2491 /* set noref for now. 2492 * any place which wants to hold dst has to call 2493 * dst_hold_safe() 2494 */ 2495 skb_dst_set_noref(skb, dst); 2496 2497 /* for unconnected multicast sockets we need to validate 2498 * the source on each packet 2499 */ 2500 if (!inet_sk(sk)->inet_daddr && in_dev) 2501 return ip_mc_validate_source(skb, iph->daddr, 2502 iph->saddr, iph->tos, 2503 skb->dev, in_dev, &itag); 2504 } 2505 return 0; 2506 } 2507 2508 int udp_rcv(struct sk_buff *skb) 2509 { 2510 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2511 } 2512 2513 void udp_destroy_sock(struct sock *sk) 2514 { 2515 struct udp_sock *up = udp_sk(sk); 2516 bool slow = lock_sock_fast(sk); 2517 udp_flush_pending_frames(sk); 2518 unlock_sock_fast(sk, slow); 2519 if (static_branch_unlikely(&udp_encap_needed_key)) { 2520 if (up->encap_type) { 2521 void (*encap_destroy)(struct sock *sk); 2522 encap_destroy = READ_ONCE(up->encap_destroy); 2523 if (encap_destroy) 2524 encap_destroy(sk); 2525 } 2526 if (up->encap_enabled) 2527 static_branch_dec(&udp_encap_needed_key); 2528 } 2529 } 2530 2531 /* 2532 * Socket option code for UDP 2533 */ 2534 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2535 char __user *optval, unsigned int optlen, 2536 int (*push_pending_frames)(struct sock *)) 2537 { 2538 struct udp_sock *up = udp_sk(sk); 2539 int val, valbool; 2540 int err = 0; 2541 int is_udplite = IS_UDPLITE(sk); 2542 2543 if (optlen < sizeof(int)) 2544 return -EINVAL; 2545 2546 if (get_user(val, (int __user *)optval)) 2547 return -EFAULT; 2548 2549 valbool = val ? 1 : 0; 2550 2551 switch (optname) { 2552 case UDP_CORK: 2553 if (val != 0) { 2554 up->corkflag = 1; 2555 } else { 2556 up->corkflag = 0; 2557 lock_sock(sk); 2558 push_pending_frames(sk); 2559 release_sock(sk); 2560 } 2561 break; 2562 2563 case UDP_ENCAP: 2564 switch (val) { 2565 case 0: 2566 #ifdef CONFIG_XFRM 2567 case UDP_ENCAP_ESPINUDP: 2568 case UDP_ENCAP_ESPINUDP_NON_IKE: 2569 #if IS_ENABLED(CONFIG_IPV6) 2570 if (sk->sk_family == AF_INET6) 2571 up->encap_rcv = ipv6_stub->xfrm6_udp_encap_rcv; 2572 else 2573 #endif 2574 up->encap_rcv = xfrm4_udp_encap_rcv; 2575 #endif 2576 fallthrough; 2577 case UDP_ENCAP_L2TPINUDP: 2578 up->encap_type = val; 2579 lock_sock(sk); 2580 udp_tunnel_encap_enable(sk->sk_socket); 2581 release_sock(sk); 2582 break; 2583 default: 2584 err = -ENOPROTOOPT; 2585 break; 2586 } 2587 break; 2588 2589 case UDP_NO_CHECK6_TX: 2590 up->no_check6_tx = valbool; 2591 break; 2592 2593 case UDP_NO_CHECK6_RX: 2594 up->no_check6_rx = valbool; 2595 break; 2596 2597 case UDP_SEGMENT: 2598 if (val < 0 || val > USHRT_MAX) 2599 return -EINVAL; 2600 up->gso_size = val; 2601 break; 2602 2603 case UDP_GRO: 2604 lock_sock(sk); 2605 if (valbool) 2606 udp_tunnel_encap_enable(sk->sk_socket); 2607 up->gro_enabled = valbool; 2608 release_sock(sk); 2609 break; 2610 2611 /* 2612 * UDP-Lite's partial checksum coverage (RFC 3828). 2613 */ 2614 /* The sender sets actual checksum coverage length via this option. 2615 * The case coverage > packet length is handled by send module. */ 2616 case UDPLITE_SEND_CSCOV: 2617 if (!is_udplite) /* Disable the option on UDP sockets */ 2618 return -ENOPROTOOPT; 2619 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2620 val = 8; 2621 else if (val > USHRT_MAX) 2622 val = USHRT_MAX; 2623 up->pcslen = val; 2624 up->pcflag |= UDPLITE_SEND_CC; 2625 break; 2626 2627 /* The receiver specifies a minimum checksum coverage value. To make 2628 * sense, this should be set to at least 8 (as done below). If zero is 2629 * used, this again means full checksum coverage. */ 2630 case UDPLITE_RECV_CSCOV: 2631 if (!is_udplite) /* Disable the option on UDP sockets */ 2632 return -ENOPROTOOPT; 2633 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2634 val = 8; 2635 else if (val > USHRT_MAX) 2636 val = USHRT_MAX; 2637 up->pcrlen = val; 2638 up->pcflag |= UDPLITE_RECV_CC; 2639 break; 2640 2641 default: 2642 err = -ENOPROTOOPT; 2643 break; 2644 } 2645 2646 return err; 2647 } 2648 EXPORT_SYMBOL(udp_lib_setsockopt); 2649 2650 int udp_setsockopt(struct sock *sk, int level, int optname, 2651 char __user *optval, unsigned int optlen) 2652 { 2653 if (level == SOL_UDP || level == SOL_UDPLITE) 2654 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2655 udp_push_pending_frames); 2656 return ip_setsockopt(sk, level, optname, optval, optlen); 2657 } 2658 2659 #ifdef CONFIG_COMPAT 2660 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2661 char __user *optval, unsigned int optlen) 2662 { 2663 if (level == SOL_UDP || level == SOL_UDPLITE) 2664 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2665 udp_push_pending_frames); 2666 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2667 } 2668 #endif 2669 2670 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2671 char __user *optval, int __user *optlen) 2672 { 2673 struct udp_sock *up = udp_sk(sk); 2674 int val, len; 2675 2676 if (get_user(len, optlen)) 2677 return -EFAULT; 2678 2679 len = min_t(unsigned int, len, sizeof(int)); 2680 2681 if (len < 0) 2682 return -EINVAL; 2683 2684 switch (optname) { 2685 case UDP_CORK: 2686 val = up->corkflag; 2687 break; 2688 2689 case UDP_ENCAP: 2690 val = up->encap_type; 2691 break; 2692 2693 case UDP_NO_CHECK6_TX: 2694 val = up->no_check6_tx; 2695 break; 2696 2697 case UDP_NO_CHECK6_RX: 2698 val = up->no_check6_rx; 2699 break; 2700 2701 case UDP_SEGMENT: 2702 val = up->gso_size; 2703 break; 2704 2705 /* The following two cannot be changed on UDP sockets, the return is 2706 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2707 case UDPLITE_SEND_CSCOV: 2708 val = up->pcslen; 2709 break; 2710 2711 case UDPLITE_RECV_CSCOV: 2712 val = up->pcrlen; 2713 break; 2714 2715 default: 2716 return -ENOPROTOOPT; 2717 } 2718 2719 if (put_user(len, optlen)) 2720 return -EFAULT; 2721 if (copy_to_user(optval, &val, len)) 2722 return -EFAULT; 2723 return 0; 2724 } 2725 EXPORT_SYMBOL(udp_lib_getsockopt); 2726 2727 int udp_getsockopt(struct sock *sk, int level, int optname, 2728 char __user *optval, int __user *optlen) 2729 { 2730 if (level == SOL_UDP || level == SOL_UDPLITE) 2731 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2732 return ip_getsockopt(sk, level, optname, optval, optlen); 2733 } 2734 2735 #ifdef CONFIG_COMPAT 2736 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2737 char __user *optval, int __user *optlen) 2738 { 2739 if (level == SOL_UDP || level == SOL_UDPLITE) 2740 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2741 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2742 } 2743 #endif 2744 /** 2745 * udp_poll - wait for a UDP event. 2746 * @file - file struct 2747 * @sock - socket 2748 * @wait - poll table 2749 * 2750 * This is same as datagram poll, except for the special case of 2751 * blocking sockets. If application is using a blocking fd 2752 * and a packet with checksum error is in the queue; 2753 * then it could get return from select indicating data available 2754 * but then block when reading it. Add special case code 2755 * to work around these arguably broken applications. 2756 */ 2757 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2758 { 2759 __poll_t mask = datagram_poll(file, sock, wait); 2760 struct sock *sk = sock->sk; 2761 2762 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 2763 mask |= EPOLLIN | EPOLLRDNORM; 2764 2765 /* Check for false positives due to checksum errors */ 2766 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2767 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2768 mask &= ~(EPOLLIN | EPOLLRDNORM); 2769 2770 return mask; 2771 2772 } 2773 EXPORT_SYMBOL(udp_poll); 2774 2775 int udp_abort(struct sock *sk, int err) 2776 { 2777 lock_sock(sk); 2778 2779 sk->sk_err = err; 2780 sk->sk_error_report(sk); 2781 __udp_disconnect(sk, 0); 2782 2783 release_sock(sk); 2784 2785 return 0; 2786 } 2787 EXPORT_SYMBOL_GPL(udp_abort); 2788 2789 struct proto udp_prot = { 2790 .name = "UDP", 2791 .owner = THIS_MODULE, 2792 .close = udp_lib_close, 2793 .pre_connect = udp_pre_connect, 2794 .connect = ip4_datagram_connect, 2795 .disconnect = udp_disconnect, 2796 .ioctl = udp_ioctl, 2797 .init = udp_init_sock, 2798 .destroy = udp_destroy_sock, 2799 .setsockopt = udp_setsockopt, 2800 .getsockopt = udp_getsockopt, 2801 .sendmsg = udp_sendmsg, 2802 .recvmsg = udp_recvmsg, 2803 .sendpage = udp_sendpage, 2804 .release_cb = ip4_datagram_release_cb, 2805 .hash = udp_lib_hash, 2806 .unhash = udp_lib_unhash, 2807 .rehash = udp_v4_rehash, 2808 .get_port = udp_v4_get_port, 2809 .memory_allocated = &udp_memory_allocated, 2810 .sysctl_mem = sysctl_udp_mem, 2811 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2812 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2813 .obj_size = sizeof(struct udp_sock), 2814 .h.udp_table = &udp_table, 2815 #ifdef CONFIG_COMPAT 2816 .compat_setsockopt = compat_udp_setsockopt, 2817 .compat_getsockopt = compat_udp_getsockopt, 2818 #endif 2819 .diag_destroy = udp_abort, 2820 }; 2821 EXPORT_SYMBOL(udp_prot); 2822 2823 /* ------------------------------------------------------------------------ */ 2824 #ifdef CONFIG_PROC_FS 2825 2826 static struct sock *udp_get_first(struct seq_file *seq, int start) 2827 { 2828 struct sock *sk; 2829 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2830 struct udp_iter_state *state = seq->private; 2831 struct net *net = seq_file_net(seq); 2832 2833 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; 2834 ++state->bucket) { 2835 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; 2836 2837 if (hlist_empty(&hslot->head)) 2838 continue; 2839 2840 spin_lock_bh(&hslot->lock); 2841 sk_for_each(sk, &hslot->head) { 2842 if (!net_eq(sock_net(sk), net)) 2843 continue; 2844 if (sk->sk_family == afinfo->family) 2845 goto found; 2846 } 2847 spin_unlock_bh(&hslot->lock); 2848 } 2849 sk = NULL; 2850 found: 2851 return sk; 2852 } 2853 2854 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2855 { 2856 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2857 struct udp_iter_state *state = seq->private; 2858 struct net *net = seq_file_net(seq); 2859 2860 do { 2861 sk = sk_next(sk); 2862 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family)); 2863 2864 if (!sk) { 2865 if (state->bucket <= afinfo->udp_table->mask) 2866 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 2867 return udp_get_first(seq, state->bucket + 1); 2868 } 2869 return sk; 2870 } 2871 2872 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2873 { 2874 struct sock *sk = udp_get_first(seq, 0); 2875 2876 if (sk) 2877 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2878 --pos; 2879 return pos ? NULL : sk; 2880 } 2881 2882 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2883 { 2884 struct udp_iter_state *state = seq->private; 2885 state->bucket = MAX_UDP_PORTS; 2886 2887 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2888 } 2889 EXPORT_SYMBOL(udp_seq_start); 2890 2891 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2892 { 2893 struct sock *sk; 2894 2895 if (v == SEQ_START_TOKEN) 2896 sk = udp_get_idx(seq, 0); 2897 else 2898 sk = udp_get_next(seq, v); 2899 2900 ++*pos; 2901 return sk; 2902 } 2903 EXPORT_SYMBOL(udp_seq_next); 2904 2905 void udp_seq_stop(struct seq_file *seq, void *v) 2906 { 2907 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2908 struct udp_iter_state *state = seq->private; 2909 2910 if (state->bucket <= afinfo->udp_table->mask) 2911 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 2912 } 2913 EXPORT_SYMBOL(udp_seq_stop); 2914 2915 /* ------------------------------------------------------------------------ */ 2916 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2917 int bucket) 2918 { 2919 struct inet_sock *inet = inet_sk(sp); 2920 __be32 dest = inet->inet_daddr; 2921 __be32 src = inet->inet_rcv_saddr; 2922 __u16 destp = ntohs(inet->inet_dport); 2923 __u16 srcp = ntohs(inet->inet_sport); 2924 2925 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2926 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 2927 bucket, src, srcp, dest, destp, sp->sk_state, 2928 sk_wmem_alloc_get(sp), 2929 udp_rqueue_get(sp), 2930 0, 0L, 0, 2931 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2932 0, sock_i_ino(sp), 2933 refcount_read(&sp->sk_refcnt), sp, 2934 atomic_read(&sp->sk_drops)); 2935 } 2936 2937 int udp4_seq_show(struct seq_file *seq, void *v) 2938 { 2939 seq_setwidth(seq, 127); 2940 if (v == SEQ_START_TOKEN) 2941 seq_puts(seq, " sl local_address rem_address st tx_queue " 2942 "rx_queue tr tm->when retrnsmt uid timeout " 2943 "inode ref pointer drops"); 2944 else { 2945 struct udp_iter_state *state = seq->private; 2946 2947 udp4_format_sock(v, seq, state->bucket); 2948 } 2949 seq_pad(seq, '\n'); 2950 return 0; 2951 } 2952 2953 const struct seq_operations udp_seq_ops = { 2954 .start = udp_seq_start, 2955 .next = udp_seq_next, 2956 .stop = udp_seq_stop, 2957 .show = udp4_seq_show, 2958 }; 2959 EXPORT_SYMBOL(udp_seq_ops); 2960 2961 static struct udp_seq_afinfo udp4_seq_afinfo = { 2962 .family = AF_INET, 2963 .udp_table = &udp_table, 2964 }; 2965 2966 static int __net_init udp4_proc_init_net(struct net *net) 2967 { 2968 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 2969 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 2970 return -ENOMEM; 2971 return 0; 2972 } 2973 2974 static void __net_exit udp4_proc_exit_net(struct net *net) 2975 { 2976 remove_proc_entry("udp", net->proc_net); 2977 } 2978 2979 static struct pernet_operations udp4_net_ops = { 2980 .init = udp4_proc_init_net, 2981 .exit = udp4_proc_exit_net, 2982 }; 2983 2984 int __init udp4_proc_init(void) 2985 { 2986 return register_pernet_subsys(&udp4_net_ops); 2987 } 2988 2989 void udp4_proc_exit(void) 2990 { 2991 unregister_pernet_subsys(&udp4_net_ops); 2992 } 2993 #endif /* CONFIG_PROC_FS */ 2994 2995 static __initdata unsigned long uhash_entries; 2996 static int __init set_uhash_entries(char *str) 2997 { 2998 ssize_t ret; 2999 3000 if (!str) 3001 return 0; 3002 3003 ret = kstrtoul(str, 0, &uhash_entries); 3004 if (ret) 3005 return 0; 3006 3007 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 3008 uhash_entries = UDP_HTABLE_SIZE_MIN; 3009 return 1; 3010 } 3011 __setup("uhash_entries=", set_uhash_entries); 3012 3013 void __init udp_table_init(struct udp_table *table, const char *name) 3014 { 3015 unsigned int i; 3016 3017 table->hash = alloc_large_system_hash(name, 3018 2 * sizeof(struct udp_hslot), 3019 uhash_entries, 3020 21, /* one slot per 2 MB */ 3021 0, 3022 &table->log, 3023 &table->mask, 3024 UDP_HTABLE_SIZE_MIN, 3025 64 * 1024); 3026 3027 table->hash2 = table->hash + (table->mask + 1); 3028 for (i = 0; i <= table->mask; i++) { 3029 INIT_HLIST_HEAD(&table->hash[i].head); 3030 table->hash[i].count = 0; 3031 spin_lock_init(&table->hash[i].lock); 3032 } 3033 for (i = 0; i <= table->mask; i++) { 3034 INIT_HLIST_HEAD(&table->hash2[i].head); 3035 table->hash2[i].count = 0; 3036 spin_lock_init(&table->hash2[i].lock); 3037 } 3038 } 3039 3040 u32 udp_flow_hashrnd(void) 3041 { 3042 static u32 hashrnd __read_mostly; 3043 3044 net_get_random_once(&hashrnd, sizeof(hashrnd)); 3045 3046 return hashrnd; 3047 } 3048 EXPORT_SYMBOL(udp_flow_hashrnd); 3049 3050 static void __udp_sysctl_init(struct net *net) 3051 { 3052 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; 3053 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; 3054 3055 #ifdef CONFIG_NET_L3_MASTER_DEV 3056 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3057 #endif 3058 } 3059 3060 static int __net_init udp_sysctl_init(struct net *net) 3061 { 3062 __udp_sysctl_init(net); 3063 return 0; 3064 } 3065 3066 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3067 .init = udp_sysctl_init, 3068 }; 3069 3070 void __init udp_init(void) 3071 { 3072 unsigned long limit; 3073 unsigned int i; 3074 3075 udp_table_init(&udp_table, "UDP"); 3076 limit = nr_free_buffer_pages() / 8; 3077 limit = max(limit, 128UL); 3078 sysctl_udp_mem[0] = limit / 4 * 3; 3079 sysctl_udp_mem[1] = limit; 3080 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3081 3082 __udp_sysctl_init(&init_net); 3083 3084 /* 16 spinlocks per cpu */ 3085 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3086 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3087 GFP_KERNEL); 3088 if (!udp_busylocks) 3089 panic("UDP: failed to alloc udp_busylocks\n"); 3090 for (i = 0; i < (1U << udp_busylocks_log); i++) 3091 spin_lock_init(udp_busylocks + i); 3092 3093 if (register_pernet_subsys(&udp_sysctl_ops)) 3094 panic("UDP: failed to init sysctl parameters.\n"); 3095 } 3096