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> 115 116 struct udp_table udp_table __read_mostly; 117 EXPORT_SYMBOL(udp_table); 118 119 long sysctl_udp_mem[3] __read_mostly; 120 EXPORT_SYMBOL(sysctl_udp_mem); 121 122 atomic_long_t udp_memory_allocated; 123 EXPORT_SYMBOL(udp_memory_allocated); 124 125 #define MAX_UDP_PORTS 65536 126 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 127 128 /* IPCB reference means this can not be used from early demux */ 129 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb) 130 { 131 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 132 if (!net->ipv4.sysctl_udp_l3mdev_accept && 133 skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) 134 return true; 135 #endif 136 return false; 137 } 138 139 static int udp_lib_lport_inuse(struct net *net, __u16 num, 140 const struct udp_hslot *hslot, 141 unsigned long *bitmap, 142 struct sock *sk, unsigned int log) 143 { 144 struct sock *sk2; 145 kuid_t uid = sock_i_uid(sk); 146 147 sk_for_each(sk2, &hslot->head) { 148 if (net_eq(sock_net(sk2), net) && 149 sk2 != sk && 150 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 151 (!sk2->sk_reuse || !sk->sk_reuse) && 152 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 153 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 154 inet_rcv_saddr_equal(sk, sk2, true)) { 155 if (sk2->sk_reuseport && sk->sk_reuseport && 156 !rcu_access_pointer(sk->sk_reuseport_cb) && 157 uid_eq(uid, sock_i_uid(sk2))) { 158 if (!bitmap) 159 return 0; 160 } else { 161 if (!bitmap) 162 return 1; 163 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 164 bitmap); 165 } 166 } 167 } 168 return 0; 169 } 170 171 /* 172 * Note: we still hold spinlock of primary hash chain, so no other writer 173 * can insert/delete a socket with local_port == num 174 */ 175 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 176 struct udp_hslot *hslot2, 177 struct sock *sk) 178 { 179 struct sock *sk2; 180 kuid_t uid = sock_i_uid(sk); 181 int res = 0; 182 183 spin_lock(&hslot2->lock); 184 udp_portaddr_for_each_entry(sk2, &hslot2->head) { 185 if (net_eq(sock_net(sk2), net) && 186 sk2 != sk && 187 (udp_sk(sk2)->udp_port_hash == num) && 188 (!sk2->sk_reuse || !sk->sk_reuse) && 189 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 190 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 191 inet_rcv_saddr_equal(sk, sk2, true)) { 192 if (sk2->sk_reuseport && sk->sk_reuseport && 193 !rcu_access_pointer(sk->sk_reuseport_cb) && 194 uid_eq(uid, sock_i_uid(sk2))) { 195 res = 0; 196 } else { 197 res = 1; 198 } 199 break; 200 } 201 } 202 spin_unlock(&hslot2->lock); 203 return res; 204 } 205 206 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) 207 { 208 struct net *net = sock_net(sk); 209 kuid_t uid = sock_i_uid(sk); 210 struct sock *sk2; 211 212 sk_for_each(sk2, &hslot->head) { 213 if (net_eq(sock_net(sk2), net) && 214 sk2 != sk && 215 sk2->sk_family == sk->sk_family && 216 ipv6_only_sock(sk2) == ipv6_only_sock(sk) && 217 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && 218 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 219 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && 220 inet_rcv_saddr_equal(sk, sk2, false)) { 221 return reuseport_add_sock(sk, sk2, 222 inet_rcv_saddr_any(sk)); 223 } 224 } 225 226 return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); 227 } 228 229 /** 230 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 231 * 232 * @sk: socket struct in question 233 * @snum: port number to look up 234 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 235 * with NULL address 236 */ 237 int udp_lib_get_port(struct sock *sk, unsigned short snum, 238 unsigned int hash2_nulladdr) 239 { 240 struct udp_hslot *hslot, *hslot2; 241 struct udp_table *udptable = sk->sk_prot->h.udp_table; 242 int error = 1; 243 struct net *net = sock_net(sk); 244 245 if (!snum) { 246 int low, high, remaining; 247 unsigned int rand; 248 unsigned short first, last; 249 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 250 251 inet_get_local_port_range(net, &low, &high); 252 remaining = (high - low) + 1; 253 254 rand = prandom_u32(); 255 first = reciprocal_scale(rand, remaining) + low; 256 /* 257 * force rand to be an odd multiple of UDP_HTABLE_SIZE 258 */ 259 rand = (rand | 1) * (udptable->mask + 1); 260 last = first + udptable->mask + 1; 261 do { 262 hslot = udp_hashslot(udptable, net, first); 263 bitmap_zero(bitmap, PORTS_PER_CHAIN); 264 spin_lock_bh(&hslot->lock); 265 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 266 udptable->log); 267 268 snum = first; 269 /* 270 * Iterate on all possible values of snum for this hash. 271 * Using steps of an odd multiple of UDP_HTABLE_SIZE 272 * give us randomization and full range coverage. 273 */ 274 do { 275 if (low <= snum && snum <= high && 276 !test_bit(snum >> udptable->log, bitmap) && 277 !inet_is_local_reserved_port(net, snum)) 278 goto found; 279 snum += rand; 280 } while (snum != first); 281 spin_unlock_bh(&hslot->lock); 282 cond_resched(); 283 } while (++first != last); 284 goto fail; 285 } else { 286 hslot = udp_hashslot(udptable, net, snum); 287 spin_lock_bh(&hslot->lock); 288 if (hslot->count > 10) { 289 int exist; 290 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 291 292 slot2 &= udptable->mask; 293 hash2_nulladdr &= udptable->mask; 294 295 hslot2 = udp_hashslot2(udptable, slot2); 296 if (hslot->count < hslot2->count) 297 goto scan_primary_hash; 298 299 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); 300 if (!exist && (hash2_nulladdr != slot2)) { 301 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 302 exist = udp_lib_lport_inuse2(net, snum, hslot2, 303 sk); 304 } 305 if (exist) 306 goto fail_unlock; 307 else 308 goto found; 309 } 310 scan_primary_hash: 311 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) 312 goto fail_unlock; 313 } 314 found: 315 inet_sk(sk)->inet_num = snum; 316 udp_sk(sk)->udp_port_hash = snum; 317 udp_sk(sk)->udp_portaddr_hash ^= snum; 318 if (sk_unhashed(sk)) { 319 if (sk->sk_reuseport && 320 udp_reuseport_add_sock(sk, hslot)) { 321 inet_sk(sk)->inet_num = 0; 322 udp_sk(sk)->udp_port_hash = 0; 323 udp_sk(sk)->udp_portaddr_hash ^= snum; 324 goto fail_unlock; 325 } 326 327 sk_add_node_rcu(sk, &hslot->head); 328 hslot->count++; 329 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 330 331 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 332 spin_lock(&hslot2->lock); 333 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 334 sk->sk_family == AF_INET6) 335 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, 336 &hslot2->head); 337 else 338 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 339 &hslot2->head); 340 hslot2->count++; 341 spin_unlock(&hslot2->lock); 342 } 343 sock_set_flag(sk, SOCK_RCU_FREE); 344 error = 0; 345 fail_unlock: 346 spin_unlock_bh(&hslot->lock); 347 fail: 348 return error; 349 } 350 EXPORT_SYMBOL(udp_lib_get_port); 351 352 int udp_v4_get_port(struct sock *sk, unsigned short snum) 353 { 354 unsigned int hash2_nulladdr = 355 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 356 unsigned int hash2_partial = 357 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 358 359 /* precompute partial secondary hash */ 360 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 361 return udp_lib_get_port(sk, snum, hash2_nulladdr); 362 } 363 364 static int compute_score(struct sock *sk, struct net *net, 365 __be32 saddr, __be16 sport, 366 __be32 daddr, unsigned short hnum, 367 int dif, int sdif, bool exact_dif) 368 { 369 int score; 370 struct inet_sock *inet; 371 bool dev_match; 372 373 if (!net_eq(sock_net(sk), net) || 374 udp_sk(sk)->udp_port_hash != hnum || 375 ipv6_only_sock(sk)) 376 return -1; 377 378 if (sk->sk_rcv_saddr != daddr) 379 return -1; 380 381 score = (sk->sk_family == PF_INET) ? 2 : 1; 382 383 inet = inet_sk(sk); 384 if (inet->inet_daddr) { 385 if (inet->inet_daddr != saddr) 386 return -1; 387 score += 4; 388 } 389 390 if (inet->inet_dport) { 391 if (inet->inet_dport != sport) 392 return -1; 393 score += 4; 394 } 395 396 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, 397 dif, sdif); 398 if (!dev_match) 399 return -1; 400 score += 4; 401 402 if (sk->sk_incoming_cpu == raw_smp_processor_id()) 403 score++; 404 return score; 405 } 406 407 static u32 udp_ehashfn(const struct net *net, const __be32 laddr, 408 const __u16 lport, const __be32 faddr, 409 const __be16 fport) 410 { 411 static u32 udp_ehash_secret __read_mostly; 412 413 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); 414 415 return __inet_ehashfn(laddr, lport, faddr, fport, 416 udp_ehash_secret + net_hash_mix(net)); 417 } 418 419 /* called with rcu_read_lock() */ 420 static struct sock *udp4_lib_lookup2(struct net *net, 421 __be32 saddr, __be16 sport, 422 __be32 daddr, unsigned int hnum, 423 int dif, int sdif, bool exact_dif, 424 struct udp_hslot *hslot2, 425 struct sk_buff *skb) 426 { 427 struct sock *sk, *result; 428 int score, badness; 429 u32 hash = 0; 430 431 result = NULL; 432 badness = 0; 433 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 434 score = compute_score(sk, net, saddr, sport, 435 daddr, hnum, dif, sdif, exact_dif); 436 if (score > badness) { 437 if (sk->sk_reuseport) { 438 hash = udp_ehashfn(net, daddr, hnum, 439 saddr, sport); 440 result = reuseport_select_sock(sk, hash, skb, 441 sizeof(struct udphdr)); 442 if (result) 443 return result; 444 } 445 badness = score; 446 result = sk; 447 } 448 } 449 return result; 450 } 451 452 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 453 * harder than this. -DaveM 454 */ 455 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 456 __be16 sport, __be32 daddr, __be16 dport, int dif, 457 int sdif, struct udp_table *udptable, struct sk_buff *skb) 458 { 459 struct sock *result; 460 unsigned short hnum = ntohs(dport); 461 unsigned int hash2, slot2; 462 struct udp_hslot *hslot2; 463 bool exact_dif = udp_lib_exact_dif_match(net, skb); 464 465 hash2 = ipv4_portaddr_hash(net, daddr, hnum); 466 slot2 = hash2 & udptable->mask; 467 hslot2 = &udptable->hash2[slot2]; 468 469 result = udp4_lib_lookup2(net, saddr, sport, 470 daddr, hnum, dif, sdif, 471 exact_dif, hslot2, skb); 472 if (!result) { 473 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 474 slot2 = hash2 & udptable->mask; 475 hslot2 = &udptable->hash2[slot2]; 476 477 result = udp4_lib_lookup2(net, saddr, sport, 478 htonl(INADDR_ANY), hnum, dif, sdif, 479 exact_dif, hslot2, skb); 480 } 481 if (unlikely(IS_ERR(result))) 482 return NULL; 483 return result; 484 } 485 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 486 487 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 488 __be16 sport, __be16 dport, 489 struct udp_table *udptable) 490 { 491 const struct iphdr *iph = ip_hdr(skb); 492 493 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, 494 iph->daddr, dport, inet_iif(skb), 495 inet_sdif(skb), udptable, skb); 496 } 497 498 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, 499 __be16 sport, __be16 dport) 500 { 501 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); 502 } 503 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); 504 505 /* Must be called under rcu_read_lock(). 506 * Does increment socket refcount. 507 */ 508 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) 509 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 510 __be32 daddr, __be16 dport, int dif) 511 { 512 struct sock *sk; 513 514 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, 515 dif, 0, &udp_table, NULL); 516 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 517 sk = NULL; 518 return sk; 519 } 520 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 521 #endif 522 523 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, 524 __be16 loc_port, __be32 loc_addr, 525 __be16 rmt_port, __be32 rmt_addr, 526 int dif, int sdif, unsigned short hnum) 527 { 528 struct inet_sock *inet = inet_sk(sk); 529 530 if (!net_eq(sock_net(sk), net) || 531 udp_sk(sk)->udp_port_hash != hnum || 532 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 533 (inet->inet_dport != rmt_port && inet->inet_dport) || 534 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || 535 ipv6_only_sock(sk) || 536 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) 537 return false; 538 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) 539 return false; 540 return true; 541 } 542 543 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); 544 void udp_encap_enable(void) 545 { 546 static_branch_inc(&udp_encap_needed_key); 547 } 548 EXPORT_SYMBOL(udp_encap_enable); 549 550 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 551 * through error handlers in encapsulations looking for a match. 552 */ 553 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) 554 { 555 int i; 556 557 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 558 int (*handler)(struct sk_buff *skb, u32 info); 559 const struct ip_tunnel_encap_ops *encap; 560 561 encap = rcu_dereference(iptun_encaps[i]); 562 if (!encap) 563 continue; 564 handler = encap->err_handler; 565 if (handler && !handler(skb, info)) 566 return 0; 567 } 568 569 return -ENOENT; 570 } 571 572 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 573 * reversing source and destination port: this will match tunnels that force the 574 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 575 * lwtunnels might actually break this assumption by being configured with 576 * different destination ports on endpoints, in this case we won't be able to 577 * trace ICMP messages back to them. 578 * 579 * If this doesn't match any socket, probe tunnels with arbitrary destination 580 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 581 * we've sent packets to won't necessarily match the local destination port. 582 * 583 * Then ask the tunnel implementation to match the error against a valid 584 * association. 585 * 586 * Return an error if we can't find a match, the socket if we need further 587 * processing, zero otherwise. 588 */ 589 static struct sock *__udp4_lib_err_encap(struct net *net, 590 const struct iphdr *iph, 591 struct udphdr *uh, 592 struct udp_table *udptable, 593 struct sk_buff *skb, u32 info) 594 { 595 int network_offset, transport_offset; 596 struct sock *sk; 597 598 network_offset = skb_network_offset(skb); 599 transport_offset = skb_transport_offset(skb); 600 601 /* Network header needs to point to the outer IPv4 header inside ICMP */ 602 skb_reset_network_header(skb); 603 604 /* Transport header needs to point to the UDP header */ 605 skb_set_transport_header(skb, iph->ihl << 2); 606 607 sk = __udp4_lib_lookup(net, iph->daddr, uh->source, 608 iph->saddr, uh->dest, skb->dev->ifindex, 0, 609 udptable, NULL); 610 if (sk) { 611 int (*lookup)(struct sock *sk, struct sk_buff *skb); 612 struct udp_sock *up = udp_sk(sk); 613 614 lookup = READ_ONCE(up->encap_err_lookup); 615 if (!lookup || lookup(sk, skb)) 616 sk = NULL; 617 } 618 619 if (!sk) 620 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); 621 622 skb_set_transport_header(skb, transport_offset); 623 skb_set_network_header(skb, network_offset); 624 625 return sk; 626 } 627 628 /* 629 * This routine is called by the ICMP module when it gets some 630 * sort of error condition. If err < 0 then the socket should 631 * be closed and the error returned to the user. If err > 0 632 * it's just the icmp type << 8 | icmp code. 633 * Header points to the ip header of the error packet. We move 634 * on past this. Then (as it used to claim before adjustment) 635 * header points to the first 8 bytes of the udp header. We need 636 * to find the appropriate port. 637 */ 638 639 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 640 { 641 struct inet_sock *inet; 642 const struct iphdr *iph = (const struct iphdr *)skb->data; 643 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 644 const int type = icmp_hdr(skb)->type; 645 const int code = icmp_hdr(skb)->code; 646 bool tunnel = false; 647 struct sock *sk; 648 int harderr; 649 int err; 650 struct net *net = dev_net(skb->dev); 651 652 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 653 iph->saddr, uh->source, skb->dev->ifindex, 654 inet_sdif(skb), udptable, NULL); 655 if (!sk) { 656 /* No socket for error: try tunnels before discarding */ 657 sk = ERR_PTR(-ENOENT); 658 if (static_branch_unlikely(&udp_encap_needed_key)) { 659 sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb, 660 info); 661 if (!sk) 662 return 0; 663 } 664 665 if (IS_ERR(sk)) { 666 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 667 return PTR_ERR(sk); 668 } 669 670 tunnel = true; 671 } 672 673 err = 0; 674 harderr = 0; 675 inet = inet_sk(sk); 676 677 switch (type) { 678 default: 679 case ICMP_TIME_EXCEEDED: 680 err = EHOSTUNREACH; 681 break; 682 case ICMP_SOURCE_QUENCH: 683 goto out; 684 case ICMP_PARAMETERPROB: 685 err = EPROTO; 686 harderr = 1; 687 break; 688 case ICMP_DEST_UNREACH: 689 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 690 ipv4_sk_update_pmtu(skb, sk, info); 691 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 692 err = EMSGSIZE; 693 harderr = 1; 694 break; 695 } 696 goto out; 697 } 698 err = EHOSTUNREACH; 699 if (code <= NR_ICMP_UNREACH) { 700 harderr = icmp_err_convert[code].fatal; 701 err = icmp_err_convert[code].errno; 702 } 703 break; 704 case ICMP_REDIRECT: 705 ipv4_sk_redirect(skb, sk); 706 goto out; 707 } 708 709 /* 710 * RFC1122: OK. Passes ICMP errors back to application, as per 711 * 4.1.3.3. 712 */ 713 if (tunnel) { 714 /* ...not for tunnels though: we don't have a sending socket */ 715 goto out; 716 } 717 if (!inet->recverr) { 718 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 719 goto out; 720 } else 721 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 722 723 sk->sk_err = err; 724 sk->sk_error_report(sk); 725 out: 726 return 0; 727 } 728 729 int udp_err(struct sk_buff *skb, u32 info) 730 { 731 return __udp4_lib_err(skb, info, &udp_table); 732 } 733 734 /* 735 * Throw away all pending data and cancel the corking. Socket is locked. 736 */ 737 void udp_flush_pending_frames(struct sock *sk) 738 { 739 struct udp_sock *up = udp_sk(sk); 740 741 if (up->pending) { 742 up->len = 0; 743 up->pending = 0; 744 ip_flush_pending_frames(sk); 745 } 746 } 747 EXPORT_SYMBOL(udp_flush_pending_frames); 748 749 /** 750 * udp4_hwcsum - handle outgoing HW checksumming 751 * @skb: sk_buff containing the filled-in UDP header 752 * (checksum field must be zeroed out) 753 * @src: source IP address 754 * @dst: destination IP address 755 */ 756 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 757 { 758 struct udphdr *uh = udp_hdr(skb); 759 int offset = skb_transport_offset(skb); 760 int len = skb->len - offset; 761 int hlen = len; 762 __wsum csum = 0; 763 764 if (!skb_has_frag_list(skb)) { 765 /* 766 * Only one fragment on the socket. 767 */ 768 skb->csum_start = skb_transport_header(skb) - skb->head; 769 skb->csum_offset = offsetof(struct udphdr, check); 770 uh->check = ~csum_tcpudp_magic(src, dst, len, 771 IPPROTO_UDP, 0); 772 } else { 773 struct sk_buff *frags; 774 775 /* 776 * HW-checksum won't work as there are two or more 777 * fragments on the socket so that all csums of sk_buffs 778 * should be together 779 */ 780 skb_walk_frags(skb, frags) { 781 csum = csum_add(csum, frags->csum); 782 hlen -= frags->len; 783 } 784 785 csum = skb_checksum(skb, offset, hlen, csum); 786 skb->ip_summed = CHECKSUM_NONE; 787 788 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 789 if (uh->check == 0) 790 uh->check = CSUM_MANGLED_0; 791 } 792 } 793 EXPORT_SYMBOL_GPL(udp4_hwcsum); 794 795 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended 796 * for the simple case like when setting the checksum for a UDP tunnel. 797 */ 798 void udp_set_csum(bool nocheck, struct sk_buff *skb, 799 __be32 saddr, __be32 daddr, int len) 800 { 801 struct udphdr *uh = udp_hdr(skb); 802 803 if (nocheck) { 804 uh->check = 0; 805 } else if (skb_is_gso(skb)) { 806 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 807 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 808 uh->check = 0; 809 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); 810 if (uh->check == 0) 811 uh->check = CSUM_MANGLED_0; 812 } else { 813 skb->ip_summed = CHECKSUM_PARTIAL; 814 skb->csum_start = skb_transport_header(skb) - skb->head; 815 skb->csum_offset = offsetof(struct udphdr, check); 816 uh->check = ~udp_v4_check(len, saddr, daddr, 0); 817 } 818 } 819 EXPORT_SYMBOL(udp_set_csum); 820 821 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, 822 struct inet_cork *cork) 823 { 824 struct sock *sk = skb->sk; 825 struct inet_sock *inet = inet_sk(sk); 826 struct udphdr *uh; 827 int err = 0; 828 int is_udplite = IS_UDPLITE(sk); 829 int offset = skb_transport_offset(skb); 830 int len = skb->len - offset; 831 __wsum csum = 0; 832 833 /* 834 * Create a UDP header 835 */ 836 uh = udp_hdr(skb); 837 uh->source = inet->inet_sport; 838 uh->dest = fl4->fl4_dport; 839 uh->len = htons(len); 840 uh->check = 0; 841 842 if (cork->gso_size) { 843 const int hlen = skb_network_header_len(skb) + 844 sizeof(struct udphdr); 845 846 if (hlen + cork->gso_size > cork->fragsize) { 847 kfree_skb(skb); 848 return -EINVAL; 849 } 850 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { 851 kfree_skb(skb); 852 return -EINVAL; 853 } 854 if (sk->sk_no_check_tx) { 855 kfree_skb(skb); 856 return -EINVAL; 857 } 858 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 859 dst_xfrm(skb_dst(skb))) { 860 kfree_skb(skb); 861 return -EIO; 862 } 863 864 skb_shinfo(skb)->gso_size = cork->gso_size; 865 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 866 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh), 867 cork->gso_size); 868 goto csum_partial; 869 } 870 871 if (is_udplite) /* UDP-Lite */ 872 csum = udplite_csum(skb); 873 874 else if (sk->sk_no_check_tx) { /* UDP csum off */ 875 876 skb->ip_summed = CHECKSUM_NONE; 877 goto send; 878 879 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 880 csum_partial: 881 882 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 883 goto send; 884 885 } else 886 csum = udp_csum(skb); 887 888 /* add protocol-dependent pseudo-header */ 889 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 890 sk->sk_protocol, csum); 891 if (uh->check == 0) 892 uh->check = CSUM_MANGLED_0; 893 894 send: 895 err = ip_send_skb(sock_net(sk), skb); 896 if (err) { 897 if (err == -ENOBUFS && !inet->recverr) { 898 UDP_INC_STATS(sock_net(sk), 899 UDP_MIB_SNDBUFERRORS, is_udplite); 900 err = 0; 901 } 902 } else 903 UDP_INC_STATS(sock_net(sk), 904 UDP_MIB_OUTDATAGRAMS, is_udplite); 905 return err; 906 } 907 908 /* 909 * Push out all pending data as one UDP datagram. Socket is locked. 910 */ 911 int udp_push_pending_frames(struct sock *sk) 912 { 913 struct udp_sock *up = udp_sk(sk); 914 struct inet_sock *inet = inet_sk(sk); 915 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 916 struct sk_buff *skb; 917 int err = 0; 918 919 skb = ip_finish_skb(sk, fl4); 920 if (!skb) 921 goto out; 922 923 err = udp_send_skb(skb, fl4, &inet->cork.base); 924 925 out: 926 up->len = 0; 927 up->pending = 0; 928 return err; 929 } 930 EXPORT_SYMBOL(udp_push_pending_frames); 931 932 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) 933 { 934 switch (cmsg->cmsg_type) { 935 case UDP_SEGMENT: 936 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) 937 return -EINVAL; 938 *gso_size = *(__u16 *)CMSG_DATA(cmsg); 939 return 0; 940 default: 941 return -EINVAL; 942 } 943 } 944 945 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) 946 { 947 struct cmsghdr *cmsg; 948 bool need_ip = false; 949 int err; 950 951 for_each_cmsghdr(cmsg, msg) { 952 if (!CMSG_OK(msg, cmsg)) 953 return -EINVAL; 954 955 if (cmsg->cmsg_level != SOL_UDP) { 956 need_ip = true; 957 continue; 958 } 959 960 err = __udp_cmsg_send(cmsg, gso_size); 961 if (err) 962 return err; 963 } 964 965 return need_ip; 966 } 967 EXPORT_SYMBOL_GPL(udp_cmsg_send); 968 969 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 970 { 971 struct inet_sock *inet = inet_sk(sk); 972 struct udp_sock *up = udp_sk(sk); 973 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); 974 struct flowi4 fl4_stack; 975 struct flowi4 *fl4; 976 int ulen = len; 977 struct ipcm_cookie ipc; 978 struct rtable *rt = NULL; 979 int free = 0; 980 int connected = 0; 981 __be32 daddr, faddr, saddr; 982 __be16 dport; 983 u8 tos; 984 int err, is_udplite = IS_UDPLITE(sk); 985 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 986 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 987 struct sk_buff *skb; 988 struct ip_options_data opt_copy; 989 990 if (len > 0xFFFF) 991 return -EMSGSIZE; 992 993 /* 994 * Check the flags. 995 */ 996 997 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 998 return -EOPNOTSUPP; 999 1000 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1001 1002 fl4 = &inet->cork.fl.u.ip4; 1003 if (up->pending) { 1004 /* 1005 * There are pending frames. 1006 * The socket lock must be held while it's corked. 1007 */ 1008 lock_sock(sk); 1009 if (likely(up->pending)) { 1010 if (unlikely(up->pending != AF_INET)) { 1011 release_sock(sk); 1012 return -EINVAL; 1013 } 1014 goto do_append_data; 1015 } 1016 release_sock(sk); 1017 } 1018 ulen += sizeof(struct udphdr); 1019 1020 /* 1021 * Get and verify the address. 1022 */ 1023 if (usin) { 1024 if (msg->msg_namelen < sizeof(*usin)) 1025 return -EINVAL; 1026 if (usin->sin_family != AF_INET) { 1027 if (usin->sin_family != AF_UNSPEC) 1028 return -EAFNOSUPPORT; 1029 } 1030 1031 daddr = usin->sin_addr.s_addr; 1032 dport = usin->sin_port; 1033 if (dport == 0) 1034 return -EINVAL; 1035 } else { 1036 if (sk->sk_state != TCP_ESTABLISHED) 1037 return -EDESTADDRREQ; 1038 daddr = inet->inet_daddr; 1039 dport = inet->inet_dport; 1040 /* Open fast path for connected socket. 1041 Route will not be used, if at least one option is set. 1042 */ 1043 connected = 1; 1044 } 1045 1046 ipcm_init_sk(&ipc, inet); 1047 ipc.gso_size = up->gso_size; 1048 1049 if (msg->msg_controllen) { 1050 err = udp_cmsg_send(sk, msg, &ipc.gso_size); 1051 if (err > 0) 1052 err = ip_cmsg_send(sk, msg, &ipc, 1053 sk->sk_family == AF_INET6); 1054 if (unlikely(err < 0)) { 1055 kfree(ipc.opt); 1056 return err; 1057 } 1058 if (ipc.opt) 1059 free = 1; 1060 connected = 0; 1061 } 1062 if (!ipc.opt) { 1063 struct ip_options_rcu *inet_opt; 1064 1065 rcu_read_lock(); 1066 inet_opt = rcu_dereference(inet->inet_opt); 1067 if (inet_opt) { 1068 memcpy(&opt_copy, inet_opt, 1069 sizeof(*inet_opt) + inet_opt->opt.optlen); 1070 ipc.opt = &opt_copy.opt; 1071 } 1072 rcu_read_unlock(); 1073 } 1074 1075 if (cgroup_bpf_enabled && !connected) { 1076 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, 1077 (struct sockaddr *)usin, &ipc.addr); 1078 if (err) 1079 goto out_free; 1080 if (usin) { 1081 if (usin->sin_port == 0) { 1082 /* BPF program set invalid port. Reject it. */ 1083 err = -EINVAL; 1084 goto out_free; 1085 } 1086 daddr = usin->sin_addr.s_addr; 1087 dport = usin->sin_port; 1088 } 1089 } 1090 1091 saddr = ipc.addr; 1092 ipc.addr = faddr = daddr; 1093 1094 if (ipc.opt && ipc.opt->opt.srr) { 1095 if (!daddr) { 1096 err = -EINVAL; 1097 goto out_free; 1098 } 1099 faddr = ipc.opt->opt.faddr; 1100 connected = 0; 1101 } 1102 tos = get_rttos(&ipc, inet); 1103 if (sock_flag(sk, SOCK_LOCALROUTE) || 1104 (msg->msg_flags & MSG_DONTROUTE) || 1105 (ipc.opt && ipc.opt->opt.is_strictroute)) { 1106 tos |= RTO_ONLINK; 1107 connected = 0; 1108 } 1109 1110 if (ipv4_is_multicast(daddr)) { 1111 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) 1112 ipc.oif = inet->mc_index; 1113 if (!saddr) 1114 saddr = inet->mc_addr; 1115 connected = 0; 1116 } else if (!ipc.oif) { 1117 ipc.oif = inet->uc_index; 1118 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { 1119 /* oif is set, packet is to local broadcast and 1120 * and uc_index is set. oif is most likely set 1121 * by sk_bound_dev_if. If uc_index != oif check if the 1122 * oif is an L3 master and uc_index is an L3 slave. 1123 * If so, we want to allow the send using the uc_index. 1124 */ 1125 if (ipc.oif != inet->uc_index && 1126 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), 1127 inet->uc_index)) { 1128 ipc.oif = inet->uc_index; 1129 } 1130 } 1131 1132 if (connected) 1133 rt = (struct rtable *)sk_dst_check(sk, 0); 1134 1135 if (!rt) { 1136 struct net *net = sock_net(sk); 1137 __u8 flow_flags = inet_sk_flowi_flags(sk); 1138 1139 fl4 = &fl4_stack; 1140 1141 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 1142 RT_SCOPE_UNIVERSE, sk->sk_protocol, 1143 flow_flags, 1144 faddr, saddr, dport, inet->inet_sport, 1145 sk->sk_uid); 1146 1147 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 1148 rt = ip_route_output_flow(net, fl4, sk); 1149 if (IS_ERR(rt)) { 1150 err = PTR_ERR(rt); 1151 rt = NULL; 1152 if (err == -ENETUNREACH) 1153 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 1154 goto out; 1155 } 1156 1157 err = -EACCES; 1158 if ((rt->rt_flags & RTCF_BROADCAST) && 1159 !sock_flag(sk, SOCK_BROADCAST)) 1160 goto out; 1161 if (connected) 1162 sk_dst_set(sk, dst_clone(&rt->dst)); 1163 } 1164 1165 if (msg->msg_flags&MSG_CONFIRM) 1166 goto do_confirm; 1167 back_from_confirm: 1168 1169 saddr = fl4->saddr; 1170 if (!ipc.addr) 1171 daddr = ipc.addr = fl4->daddr; 1172 1173 /* Lockless fast path for the non-corking case. */ 1174 if (!corkreq) { 1175 struct inet_cork cork; 1176 1177 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, 1178 sizeof(struct udphdr), &ipc, &rt, 1179 &cork, msg->msg_flags); 1180 err = PTR_ERR(skb); 1181 if (!IS_ERR_OR_NULL(skb)) 1182 err = udp_send_skb(skb, fl4, &cork); 1183 goto out; 1184 } 1185 1186 lock_sock(sk); 1187 if (unlikely(up->pending)) { 1188 /* The socket is already corked while preparing it. */ 1189 /* ... which is an evident application bug. --ANK */ 1190 release_sock(sk); 1191 1192 net_dbg_ratelimited("socket already corked\n"); 1193 err = -EINVAL; 1194 goto out; 1195 } 1196 /* 1197 * Now cork the socket to pend data. 1198 */ 1199 fl4 = &inet->cork.fl.u.ip4; 1200 fl4->daddr = daddr; 1201 fl4->saddr = saddr; 1202 fl4->fl4_dport = dport; 1203 fl4->fl4_sport = inet->inet_sport; 1204 up->pending = AF_INET; 1205 1206 do_append_data: 1207 up->len += ulen; 1208 err = ip_append_data(sk, fl4, getfrag, msg, ulen, 1209 sizeof(struct udphdr), &ipc, &rt, 1210 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1211 if (err) 1212 udp_flush_pending_frames(sk); 1213 else if (!corkreq) 1214 err = udp_push_pending_frames(sk); 1215 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1216 up->pending = 0; 1217 release_sock(sk); 1218 1219 out: 1220 ip_rt_put(rt); 1221 out_free: 1222 if (free) 1223 kfree(ipc.opt); 1224 if (!err) 1225 return len; 1226 /* 1227 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1228 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1229 * we don't have a good statistic (IpOutDiscards but it can be too many 1230 * things). We could add another new stat but at least for now that 1231 * seems like overkill. 1232 */ 1233 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1234 UDP_INC_STATS(sock_net(sk), 1235 UDP_MIB_SNDBUFERRORS, is_udplite); 1236 } 1237 return err; 1238 1239 do_confirm: 1240 if (msg->msg_flags & MSG_PROBE) 1241 dst_confirm_neigh(&rt->dst, &fl4->daddr); 1242 if (!(msg->msg_flags&MSG_PROBE) || len) 1243 goto back_from_confirm; 1244 err = 0; 1245 goto out; 1246 } 1247 EXPORT_SYMBOL(udp_sendmsg); 1248 1249 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1250 size_t size, int flags) 1251 { 1252 struct inet_sock *inet = inet_sk(sk); 1253 struct udp_sock *up = udp_sk(sk); 1254 int ret; 1255 1256 if (flags & MSG_SENDPAGE_NOTLAST) 1257 flags |= MSG_MORE; 1258 1259 if (!up->pending) { 1260 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1261 1262 /* Call udp_sendmsg to specify destination address which 1263 * sendpage interface can't pass. 1264 * This will succeed only when the socket is connected. 1265 */ 1266 ret = udp_sendmsg(sk, &msg, 0); 1267 if (ret < 0) 1268 return ret; 1269 } 1270 1271 lock_sock(sk); 1272 1273 if (unlikely(!up->pending)) { 1274 release_sock(sk); 1275 1276 net_dbg_ratelimited("cork failed\n"); 1277 return -EINVAL; 1278 } 1279 1280 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1281 page, offset, size, flags); 1282 if (ret == -EOPNOTSUPP) { 1283 release_sock(sk); 1284 return sock_no_sendpage(sk->sk_socket, page, offset, 1285 size, flags); 1286 } 1287 if (ret < 0) { 1288 udp_flush_pending_frames(sk); 1289 goto out; 1290 } 1291 1292 up->len += size; 1293 if (!(up->corkflag || (flags&MSG_MORE))) 1294 ret = udp_push_pending_frames(sk); 1295 if (!ret) 1296 ret = size; 1297 out: 1298 release_sock(sk); 1299 return ret; 1300 } 1301 1302 #define UDP_SKB_IS_STATELESS 0x80000000 1303 1304 static void udp_set_dev_scratch(struct sk_buff *skb) 1305 { 1306 struct udp_dev_scratch *scratch = udp_skb_scratch(skb); 1307 1308 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); 1309 scratch->_tsize_state = skb->truesize; 1310 #if BITS_PER_LONG == 64 1311 scratch->len = skb->len; 1312 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); 1313 scratch->is_linear = !skb_is_nonlinear(skb); 1314 #endif 1315 /* all head states execept sp (dst, sk, nf) are always cleared by 1316 * udp_rcv() and we need to preserve secpath, if present, to eventually 1317 * process IP_CMSG_PASSSEC at recvmsg() time 1318 */ 1319 if (likely(!skb_sec_path(skb))) 1320 scratch->_tsize_state |= UDP_SKB_IS_STATELESS; 1321 } 1322 1323 static int udp_skb_truesize(struct sk_buff *skb) 1324 { 1325 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; 1326 } 1327 1328 static bool udp_skb_has_head_state(struct sk_buff *skb) 1329 { 1330 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); 1331 } 1332 1333 /* fully reclaim rmem/fwd memory allocated for skb */ 1334 static void udp_rmem_release(struct sock *sk, int size, int partial, 1335 bool rx_queue_lock_held) 1336 { 1337 struct udp_sock *up = udp_sk(sk); 1338 struct sk_buff_head *sk_queue; 1339 int amt; 1340 1341 if (likely(partial)) { 1342 up->forward_deficit += size; 1343 size = up->forward_deficit; 1344 if (size < (sk->sk_rcvbuf >> 2)) 1345 return; 1346 } else { 1347 size += up->forward_deficit; 1348 } 1349 up->forward_deficit = 0; 1350 1351 /* acquire the sk_receive_queue for fwd allocated memory scheduling, 1352 * if the called don't held it already 1353 */ 1354 sk_queue = &sk->sk_receive_queue; 1355 if (!rx_queue_lock_held) 1356 spin_lock(&sk_queue->lock); 1357 1358 1359 sk->sk_forward_alloc += size; 1360 amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); 1361 sk->sk_forward_alloc -= amt; 1362 1363 if (amt) 1364 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); 1365 1366 atomic_sub(size, &sk->sk_rmem_alloc); 1367 1368 /* this can save us from acquiring the rx queue lock on next receive */ 1369 skb_queue_splice_tail_init(sk_queue, &up->reader_queue); 1370 1371 if (!rx_queue_lock_held) 1372 spin_unlock(&sk_queue->lock); 1373 } 1374 1375 /* Note: called with reader_queue.lock held. 1376 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch 1377 * This avoids a cache line miss while receive_queue lock is held. 1378 * Look at __udp_enqueue_schedule_skb() to find where this copy is done. 1379 */ 1380 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) 1381 { 1382 prefetch(&skb->data); 1383 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); 1384 } 1385 EXPORT_SYMBOL(udp_skb_destructor); 1386 1387 /* as above, but the caller held the rx queue lock, too */ 1388 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) 1389 { 1390 prefetch(&skb->data); 1391 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); 1392 } 1393 1394 /* Idea of busylocks is to let producers grab an extra spinlock 1395 * to relieve pressure on the receive_queue spinlock shared by consumer. 1396 * Under flood, this means that only one producer can be in line 1397 * trying to acquire the receive_queue spinlock. 1398 * These busylock can be allocated on a per cpu manner, instead of a 1399 * per socket one (that would consume a cache line per socket) 1400 */ 1401 static int udp_busylocks_log __read_mostly; 1402 static spinlock_t *udp_busylocks __read_mostly; 1403 1404 static spinlock_t *busylock_acquire(void *ptr) 1405 { 1406 spinlock_t *busy; 1407 1408 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); 1409 spin_lock(busy); 1410 return busy; 1411 } 1412 1413 static void busylock_release(spinlock_t *busy) 1414 { 1415 if (busy) 1416 spin_unlock(busy); 1417 } 1418 1419 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) 1420 { 1421 struct sk_buff_head *list = &sk->sk_receive_queue; 1422 int rmem, delta, amt, err = -ENOMEM; 1423 spinlock_t *busy = NULL; 1424 int size; 1425 1426 /* try to avoid the costly atomic add/sub pair when the receive 1427 * queue is full; always allow at least a packet 1428 */ 1429 rmem = atomic_read(&sk->sk_rmem_alloc); 1430 if (rmem > sk->sk_rcvbuf) 1431 goto drop; 1432 1433 /* Under mem pressure, it might be helpful to help udp_recvmsg() 1434 * having linear skbs : 1435 * - Reduce memory overhead and thus increase receive queue capacity 1436 * - Less cache line misses at copyout() time 1437 * - Less work at consume_skb() (less alien page frag freeing) 1438 */ 1439 if (rmem > (sk->sk_rcvbuf >> 1)) { 1440 skb_condense(skb); 1441 1442 busy = busylock_acquire(sk); 1443 } 1444 size = skb->truesize; 1445 udp_set_dev_scratch(skb); 1446 1447 /* we drop only if the receive buf is full and the receive 1448 * queue contains some other skb 1449 */ 1450 rmem = atomic_add_return(size, &sk->sk_rmem_alloc); 1451 if (rmem > (size + sk->sk_rcvbuf)) 1452 goto uncharge_drop; 1453 1454 spin_lock(&list->lock); 1455 if (size >= sk->sk_forward_alloc) { 1456 amt = sk_mem_pages(size); 1457 delta = amt << SK_MEM_QUANTUM_SHIFT; 1458 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { 1459 err = -ENOBUFS; 1460 spin_unlock(&list->lock); 1461 goto uncharge_drop; 1462 } 1463 1464 sk->sk_forward_alloc += delta; 1465 } 1466 1467 sk->sk_forward_alloc -= size; 1468 1469 /* no need to setup a destructor, we will explicitly release the 1470 * forward allocated memory on dequeue 1471 */ 1472 sock_skb_set_dropcount(sk, skb); 1473 1474 __skb_queue_tail(list, skb); 1475 spin_unlock(&list->lock); 1476 1477 if (!sock_flag(sk, SOCK_DEAD)) 1478 sk->sk_data_ready(sk); 1479 1480 busylock_release(busy); 1481 return 0; 1482 1483 uncharge_drop: 1484 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1485 1486 drop: 1487 atomic_inc(&sk->sk_drops); 1488 busylock_release(busy); 1489 return err; 1490 } 1491 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); 1492 1493 void udp_destruct_sock(struct sock *sk) 1494 { 1495 /* reclaim completely the forward allocated memory */ 1496 struct udp_sock *up = udp_sk(sk); 1497 unsigned int total = 0; 1498 struct sk_buff *skb; 1499 1500 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); 1501 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { 1502 total += skb->truesize; 1503 kfree_skb(skb); 1504 } 1505 udp_rmem_release(sk, total, 0, true); 1506 1507 inet_sock_destruct(sk); 1508 } 1509 EXPORT_SYMBOL_GPL(udp_destruct_sock); 1510 1511 int udp_init_sock(struct sock *sk) 1512 { 1513 skb_queue_head_init(&udp_sk(sk)->reader_queue); 1514 sk->sk_destruct = udp_destruct_sock; 1515 return 0; 1516 } 1517 EXPORT_SYMBOL_GPL(udp_init_sock); 1518 1519 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) 1520 { 1521 if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { 1522 bool slow = lock_sock_fast(sk); 1523 1524 sk_peek_offset_bwd(sk, len); 1525 unlock_sock_fast(sk, slow); 1526 } 1527 1528 if (!skb_unref(skb)) 1529 return; 1530 1531 /* In the more common cases we cleared the head states previously, 1532 * see __udp_queue_rcv_skb(). 1533 */ 1534 if (unlikely(udp_skb_has_head_state(skb))) 1535 skb_release_head_state(skb); 1536 __consume_stateless_skb(skb); 1537 } 1538 EXPORT_SYMBOL_GPL(skb_consume_udp); 1539 1540 static struct sk_buff *__first_packet_length(struct sock *sk, 1541 struct sk_buff_head *rcvq, 1542 int *total) 1543 { 1544 struct sk_buff *skb; 1545 1546 while ((skb = skb_peek(rcvq)) != NULL) { 1547 if (udp_lib_checksum_complete(skb)) { 1548 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, 1549 IS_UDPLITE(sk)); 1550 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, 1551 IS_UDPLITE(sk)); 1552 atomic_inc(&sk->sk_drops); 1553 __skb_unlink(skb, rcvq); 1554 *total += skb->truesize; 1555 kfree_skb(skb); 1556 } else { 1557 /* the csum related bits could be changed, refresh 1558 * the scratch area 1559 */ 1560 udp_set_dev_scratch(skb); 1561 break; 1562 } 1563 } 1564 return skb; 1565 } 1566 1567 /** 1568 * first_packet_length - return length of first packet in receive queue 1569 * @sk: socket 1570 * 1571 * Drops all bad checksum frames, until a valid one is found. 1572 * Returns the length of found skb, or -1 if none is found. 1573 */ 1574 static int first_packet_length(struct sock *sk) 1575 { 1576 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; 1577 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1578 struct sk_buff *skb; 1579 int total = 0; 1580 int res; 1581 1582 spin_lock_bh(&rcvq->lock); 1583 skb = __first_packet_length(sk, rcvq, &total); 1584 if (!skb && !skb_queue_empty(sk_queue)) { 1585 spin_lock(&sk_queue->lock); 1586 skb_queue_splice_tail_init(sk_queue, rcvq); 1587 spin_unlock(&sk_queue->lock); 1588 1589 skb = __first_packet_length(sk, rcvq, &total); 1590 } 1591 res = skb ? skb->len : -1; 1592 if (total) 1593 udp_rmem_release(sk, total, 1, false); 1594 spin_unlock_bh(&rcvq->lock); 1595 return res; 1596 } 1597 1598 /* 1599 * IOCTL requests applicable to the UDP protocol 1600 */ 1601 1602 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1603 { 1604 switch (cmd) { 1605 case SIOCOUTQ: 1606 { 1607 int amount = sk_wmem_alloc_get(sk); 1608 1609 return put_user(amount, (int __user *)arg); 1610 } 1611 1612 case SIOCINQ: 1613 { 1614 int amount = max_t(int, 0, first_packet_length(sk)); 1615 1616 return put_user(amount, (int __user *)arg); 1617 } 1618 1619 default: 1620 return -ENOIOCTLCMD; 1621 } 1622 1623 return 0; 1624 } 1625 EXPORT_SYMBOL(udp_ioctl); 1626 1627 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, 1628 int noblock, int *off, int *err) 1629 { 1630 struct sk_buff_head *sk_queue = &sk->sk_receive_queue; 1631 struct sk_buff_head *queue; 1632 struct sk_buff *last; 1633 long timeo; 1634 int error; 1635 1636 queue = &udp_sk(sk)->reader_queue; 1637 flags |= noblock ? MSG_DONTWAIT : 0; 1638 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 1639 do { 1640 struct sk_buff *skb; 1641 1642 error = sock_error(sk); 1643 if (error) 1644 break; 1645 1646 error = -EAGAIN; 1647 do { 1648 spin_lock_bh(&queue->lock); 1649 skb = __skb_try_recv_from_queue(sk, queue, flags, 1650 udp_skb_destructor, 1651 off, err, &last); 1652 if (skb) { 1653 spin_unlock_bh(&queue->lock); 1654 return skb; 1655 } 1656 1657 if (skb_queue_empty(sk_queue)) { 1658 spin_unlock_bh(&queue->lock); 1659 goto busy_check; 1660 } 1661 1662 /* refill the reader queue and walk it again 1663 * keep both queues locked to avoid re-acquiring 1664 * the sk_receive_queue lock if fwd memory scheduling 1665 * is needed. 1666 */ 1667 spin_lock(&sk_queue->lock); 1668 skb_queue_splice_tail_init(sk_queue, queue); 1669 1670 skb = __skb_try_recv_from_queue(sk, queue, flags, 1671 udp_skb_dtor_locked, 1672 off, err, &last); 1673 spin_unlock(&sk_queue->lock); 1674 spin_unlock_bh(&queue->lock); 1675 if (skb) 1676 return skb; 1677 1678 busy_check: 1679 if (!sk_can_busy_loop(sk)) 1680 break; 1681 1682 sk_busy_loop(sk, flags & MSG_DONTWAIT); 1683 } while (!skb_queue_empty(sk_queue)); 1684 1685 /* sk_queue is empty, reader_queue may contain peeked packets */ 1686 } while (timeo && 1687 !__skb_wait_for_more_packets(sk, &error, &timeo, 1688 (struct sk_buff *)sk_queue)); 1689 1690 *err = error; 1691 return NULL; 1692 } 1693 EXPORT_SYMBOL(__skb_recv_udp); 1694 1695 /* 1696 * This should be easy, if there is something there we 1697 * return it, otherwise we block. 1698 */ 1699 1700 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, 1701 int flags, int *addr_len) 1702 { 1703 struct inet_sock *inet = inet_sk(sk); 1704 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); 1705 struct sk_buff *skb; 1706 unsigned int ulen, copied; 1707 int off, err, peeking = flags & MSG_PEEK; 1708 int is_udplite = IS_UDPLITE(sk); 1709 bool checksum_valid = false; 1710 1711 if (flags & MSG_ERRQUEUE) 1712 return ip_recv_error(sk, msg, len, addr_len); 1713 1714 try_again: 1715 off = sk_peek_offset(sk, flags); 1716 skb = __skb_recv_udp(sk, flags, noblock, &off, &err); 1717 if (!skb) 1718 return err; 1719 1720 ulen = udp_skb_len(skb); 1721 copied = len; 1722 if (copied > ulen - off) 1723 copied = ulen - off; 1724 else if (copied < ulen) 1725 msg->msg_flags |= MSG_TRUNC; 1726 1727 /* 1728 * If checksum is needed at all, try to do it while copying the 1729 * data. If the data is truncated, or if we only want a partial 1730 * coverage checksum (UDP-Lite), do it before the copy. 1731 */ 1732 1733 if (copied < ulen || peeking || 1734 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 1735 checksum_valid = udp_skb_csum_unnecessary(skb) || 1736 !__udp_lib_checksum_complete(skb); 1737 if (!checksum_valid) 1738 goto csum_copy_err; 1739 } 1740 1741 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 1742 if (udp_skb_is_linear(skb)) 1743 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 1744 else 1745 err = skb_copy_datagram_msg(skb, off, msg, copied); 1746 } else { 1747 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 1748 1749 if (err == -EINVAL) 1750 goto csum_copy_err; 1751 } 1752 1753 if (unlikely(err)) { 1754 if (!peeking) { 1755 atomic_inc(&sk->sk_drops); 1756 UDP_INC_STATS(sock_net(sk), 1757 UDP_MIB_INERRORS, is_udplite); 1758 } 1759 kfree_skb(skb); 1760 return err; 1761 } 1762 1763 if (!peeking) 1764 UDP_INC_STATS(sock_net(sk), 1765 UDP_MIB_INDATAGRAMS, is_udplite); 1766 1767 sock_recv_ts_and_drops(msg, sk, skb); 1768 1769 /* Copy the address. */ 1770 if (sin) { 1771 sin->sin_family = AF_INET; 1772 sin->sin_port = udp_hdr(skb)->source; 1773 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1774 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1775 *addr_len = sizeof(*sin); 1776 } 1777 1778 if (udp_sk(sk)->gro_enabled) 1779 udp_cmsg_recv(msg, sk, skb); 1780 1781 if (inet->cmsg_flags) 1782 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); 1783 1784 err = copied; 1785 if (flags & MSG_TRUNC) 1786 err = ulen; 1787 1788 skb_consume_udp(sk, skb, peeking ? -err : err); 1789 return err; 1790 1791 csum_copy_err: 1792 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 1793 udp_skb_destructor)) { 1794 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1795 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1796 } 1797 kfree_skb(skb); 1798 1799 /* starting over for a new packet, but check if we need to yield */ 1800 cond_resched(); 1801 msg->msg_flags &= ~MSG_TRUNC; 1802 goto try_again; 1803 } 1804 1805 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) 1806 { 1807 /* This check is replicated from __ip4_datagram_connect() and 1808 * intended to prevent BPF program called below from accessing bytes 1809 * that are out of the bound specified by user in addr_len. 1810 */ 1811 if (addr_len < sizeof(struct sockaddr_in)) 1812 return -EINVAL; 1813 1814 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); 1815 } 1816 EXPORT_SYMBOL(udp_pre_connect); 1817 1818 int __udp_disconnect(struct sock *sk, int flags) 1819 { 1820 struct inet_sock *inet = inet_sk(sk); 1821 /* 1822 * 1003.1g - break association. 1823 */ 1824 1825 sk->sk_state = TCP_CLOSE; 1826 inet->inet_daddr = 0; 1827 inet->inet_dport = 0; 1828 sock_rps_reset_rxhash(sk); 1829 sk->sk_bound_dev_if = 0; 1830 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1831 inet_reset_saddr(sk); 1832 1833 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1834 sk->sk_prot->unhash(sk); 1835 inet->inet_sport = 0; 1836 } 1837 sk_dst_reset(sk); 1838 return 0; 1839 } 1840 EXPORT_SYMBOL(__udp_disconnect); 1841 1842 int udp_disconnect(struct sock *sk, int flags) 1843 { 1844 lock_sock(sk); 1845 __udp_disconnect(sk, flags); 1846 release_sock(sk); 1847 return 0; 1848 } 1849 EXPORT_SYMBOL(udp_disconnect); 1850 1851 void udp_lib_unhash(struct sock *sk) 1852 { 1853 if (sk_hashed(sk)) { 1854 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1855 struct udp_hslot *hslot, *hslot2; 1856 1857 hslot = udp_hashslot(udptable, sock_net(sk), 1858 udp_sk(sk)->udp_port_hash); 1859 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1860 1861 spin_lock_bh(&hslot->lock); 1862 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1863 reuseport_detach_sock(sk); 1864 if (sk_del_node_init_rcu(sk)) { 1865 hslot->count--; 1866 inet_sk(sk)->inet_num = 0; 1867 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1868 1869 spin_lock(&hslot2->lock); 1870 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1871 hslot2->count--; 1872 spin_unlock(&hslot2->lock); 1873 } 1874 spin_unlock_bh(&hslot->lock); 1875 } 1876 } 1877 EXPORT_SYMBOL(udp_lib_unhash); 1878 1879 /* 1880 * inet_rcv_saddr was changed, we must rehash secondary hash 1881 */ 1882 void udp_lib_rehash(struct sock *sk, u16 newhash) 1883 { 1884 if (sk_hashed(sk)) { 1885 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1886 struct udp_hslot *hslot, *hslot2, *nhslot2; 1887 1888 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1889 nhslot2 = udp_hashslot2(udptable, newhash); 1890 udp_sk(sk)->udp_portaddr_hash = newhash; 1891 1892 if (hslot2 != nhslot2 || 1893 rcu_access_pointer(sk->sk_reuseport_cb)) { 1894 hslot = udp_hashslot(udptable, sock_net(sk), 1895 udp_sk(sk)->udp_port_hash); 1896 /* we must lock primary chain too */ 1897 spin_lock_bh(&hslot->lock); 1898 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1899 reuseport_detach_sock(sk); 1900 1901 if (hslot2 != nhslot2) { 1902 spin_lock(&hslot2->lock); 1903 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1904 hslot2->count--; 1905 spin_unlock(&hslot2->lock); 1906 1907 spin_lock(&nhslot2->lock); 1908 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1909 &nhslot2->head); 1910 nhslot2->count++; 1911 spin_unlock(&nhslot2->lock); 1912 } 1913 1914 spin_unlock_bh(&hslot->lock); 1915 } 1916 } 1917 } 1918 EXPORT_SYMBOL(udp_lib_rehash); 1919 1920 void udp_v4_rehash(struct sock *sk) 1921 { 1922 u16 new_hash = ipv4_portaddr_hash(sock_net(sk), 1923 inet_sk(sk)->inet_rcv_saddr, 1924 inet_sk(sk)->inet_num); 1925 udp_lib_rehash(sk, new_hash); 1926 } 1927 1928 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1929 { 1930 int rc; 1931 1932 if (inet_sk(sk)->inet_daddr) { 1933 sock_rps_save_rxhash(sk, skb); 1934 sk_mark_napi_id(sk, skb); 1935 sk_incoming_cpu_update(sk); 1936 } else { 1937 sk_mark_napi_id_once(sk, skb); 1938 } 1939 1940 rc = __udp_enqueue_schedule_skb(sk, skb); 1941 if (rc < 0) { 1942 int is_udplite = IS_UDPLITE(sk); 1943 1944 /* Note that an ENOMEM error is charged twice */ 1945 if (rc == -ENOMEM) 1946 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1947 is_udplite); 1948 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1949 kfree_skb(skb); 1950 trace_udp_fail_queue_rcv_skb(rc, sk); 1951 return -1; 1952 } 1953 1954 return 0; 1955 } 1956 1957 /* returns: 1958 * -1: error 1959 * 0: success 1960 * >0: "udp encap" protocol resubmission 1961 * 1962 * Note that in the success and error cases, the skb is assumed to 1963 * have either been requeued or freed. 1964 */ 1965 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 1966 { 1967 struct udp_sock *up = udp_sk(sk); 1968 int is_udplite = IS_UDPLITE(sk); 1969 1970 /* 1971 * Charge it to the socket, dropping if the queue is full. 1972 */ 1973 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1974 goto drop; 1975 nf_reset(skb); 1976 1977 if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { 1978 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1979 1980 /* 1981 * This is an encapsulation socket so pass the skb to 1982 * the socket's udp_encap_rcv() hook. Otherwise, just 1983 * fall through and pass this up the UDP socket. 1984 * up->encap_rcv() returns the following value: 1985 * =0 if skb was successfully passed to the encap 1986 * handler or was discarded by it. 1987 * >0 if skb should be passed on to UDP. 1988 * <0 if skb should be resubmitted as proto -N 1989 */ 1990 1991 /* if we're overly short, let UDP handle it */ 1992 encap_rcv = READ_ONCE(up->encap_rcv); 1993 if (encap_rcv) { 1994 int ret; 1995 1996 /* Verify checksum before giving to encap */ 1997 if (udp_lib_checksum_complete(skb)) 1998 goto csum_error; 1999 2000 ret = encap_rcv(sk, skb); 2001 if (ret <= 0) { 2002 __UDP_INC_STATS(sock_net(sk), 2003 UDP_MIB_INDATAGRAMS, 2004 is_udplite); 2005 return -ret; 2006 } 2007 } 2008 2009 /* FALLTHROUGH -- it's a UDP Packet */ 2010 } 2011 2012 /* 2013 * UDP-Lite specific tests, ignored on UDP sockets 2014 */ 2015 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 2016 2017 /* 2018 * MIB statistics other than incrementing the error count are 2019 * disabled for the following two types of errors: these depend 2020 * on the application settings, not on the functioning of the 2021 * protocol stack as such. 2022 * 2023 * RFC 3828 here recommends (sec 3.3): "There should also be a 2024 * way ... to ... at least let the receiving application block 2025 * delivery of packets with coverage values less than a value 2026 * provided by the application." 2027 */ 2028 if (up->pcrlen == 0) { /* full coverage was set */ 2029 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", 2030 UDP_SKB_CB(skb)->cscov, skb->len); 2031 goto drop; 2032 } 2033 /* The next case involves violating the min. coverage requested 2034 * by the receiver. This is subtle: if receiver wants x and x is 2035 * greater than the buffersize/MTU then receiver will complain 2036 * that it wants x while sender emits packets of smaller size y. 2037 * Therefore the above ...()->partial_cov statement is essential. 2038 */ 2039 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 2040 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", 2041 UDP_SKB_CB(skb)->cscov, up->pcrlen); 2042 goto drop; 2043 } 2044 } 2045 2046 prefetch(&sk->sk_rmem_alloc); 2047 if (rcu_access_pointer(sk->sk_filter) && 2048 udp_lib_checksum_complete(skb)) 2049 goto csum_error; 2050 2051 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) 2052 goto drop; 2053 2054 udp_csum_pull_header(skb); 2055 2056 ipv4_pktinfo_prepare(sk, skb); 2057 return __udp_queue_rcv_skb(sk, skb); 2058 2059 csum_error: 2060 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 2061 drop: 2062 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 2063 atomic_inc(&sk->sk_drops); 2064 kfree_skb(skb); 2065 return -1; 2066 } 2067 2068 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2069 { 2070 struct sk_buff *next, *segs; 2071 int ret; 2072 2073 if (likely(!udp_unexpected_gso(sk, skb))) 2074 return udp_queue_rcv_one_skb(sk, skb); 2075 2076 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET); 2077 __skb_push(skb, -skb_mac_offset(skb)); 2078 segs = udp_rcv_segment(sk, skb, true); 2079 for (skb = segs; skb; skb = next) { 2080 next = skb->next; 2081 __skb_pull(skb, skb_transport_offset(skb)); 2082 ret = udp_queue_rcv_one_skb(sk, skb); 2083 if (ret > 0) 2084 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret); 2085 } 2086 return 0; 2087 } 2088 2089 /* For TCP sockets, sk_rx_dst is protected by socket lock 2090 * For UDP, we use xchg() to guard against concurrent changes. 2091 */ 2092 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 2093 { 2094 struct dst_entry *old; 2095 2096 if (dst_hold_safe(dst)) { 2097 old = xchg(&sk->sk_rx_dst, dst); 2098 dst_release(old); 2099 return old != dst; 2100 } 2101 return false; 2102 } 2103 EXPORT_SYMBOL(udp_sk_rx_dst_set); 2104 2105 /* 2106 * Multicasts and broadcasts go to each listener. 2107 * 2108 * Note: called only from the BH handler context. 2109 */ 2110 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 2111 struct udphdr *uh, 2112 __be32 saddr, __be32 daddr, 2113 struct udp_table *udptable, 2114 int proto) 2115 { 2116 struct sock *sk, *first = NULL; 2117 unsigned short hnum = ntohs(uh->dest); 2118 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 2119 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 2120 unsigned int offset = offsetof(typeof(*sk), sk_node); 2121 int dif = skb->dev->ifindex; 2122 int sdif = inet_sdif(skb); 2123 struct hlist_node *node; 2124 struct sk_buff *nskb; 2125 2126 if (use_hash2) { 2127 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & 2128 udptable->mask; 2129 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; 2130 start_lookup: 2131 hslot = &udptable->hash2[hash2]; 2132 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 2133 } 2134 2135 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 2136 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, 2137 uh->source, saddr, dif, sdif, hnum)) 2138 continue; 2139 2140 if (!first) { 2141 first = sk; 2142 continue; 2143 } 2144 nskb = skb_clone(skb, GFP_ATOMIC); 2145 2146 if (unlikely(!nskb)) { 2147 atomic_inc(&sk->sk_drops); 2148 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 2149 IS_UDPLITE(sk)); 2150 __UDP_INC_STATS(net, UDP_MIB_INERRORS, 2151 IS_UDPLITE(sk)); 2152 continue; 2153 } 2154 if (udp_queue_rcv_skb(sk, nskb) > 0) 2155 consume_skb(nskb); 2156 } 2157 2158 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 2159 if (use_hash2 && hash2 != hash2_any) { 2160 hash2 = hash2_any; 2161 goto start_lookup; 2162 } 2163 2164 if (first) { 2165 if (udp_queue_rcv_skb(first, skb) > 0) 2166 consume_skb(skb); 2167 } else { 2168 kfree_skb(skb); 2169 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 2170 proto == IPPROTO_UDPLITE); 2171 } 2172 return 0; 2173 } 2174 2175 /* Initialize UDP checksum. If exited with zero value (success), 2176 * CHECKSUM_UNNECESSARY means, that no more checks are required. 2177 * Otherwise, csum completion requires chacksumming packet body, 2178 * including udp header and folding it to skb->csum. 2179 */ 2180 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 2181 int proto) 2182 { 2183 int err; 2184 2185 UDP_SKB_CB(skb)->partial_cov = 0; 2186 UDP_SKB_CB(skb)->cscov = skb->len; 2187 2188 if (proto == IPPROTO_UDPLITE) { 2189 err = udplite_checksum_init(skb, uh); 2190 if (err) 2191 return err; 2192 2193 if (UDP_SKB_CB(skb)->partial_cov) { 2194 skb->csum = inet_compute_pseudo(skb, proto); 2195 return 0; 2196 } 2197 } 2198 2199 /* Note, we are only interested in != 0 or == 0, thus the 2200 * force to int. 2201 */ 2202 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, 2203 inet_compute_pseudo); 2204 if (err) 2205 return err; 2206 2207 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { 2208 /* If SW calculated the value, we know it's bad */ 2209 if (skb->csum_complete_sw) 2210 return 1; 2211 2212 /* HW says the value is bad. Let's validate that. 2213 * skb->csum is no longer the full packet checksum, 2214 * so don't treat it as such. 2215 */ 2216 skb_checksum_complete_unset(skb); 2217 } 2218 2219 return 0; 2220 } 2221 2222 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 2223 * return code conversion for ip layer consumption 2224 */ 2225 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 2226 struct udphdr *uh) 2227 { 2228 int ret; 2229 2230 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 2231 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, 2232 inet_compute_pseudo); 2233 2234 ret = udp_queue_rcv_skb(sk, skb); 2235 2236 /* a return value > 0 means to resubmit the input, but 2237 * it wants the return to be -protocol, or 0 2238 */ 2239 if (ret > 0) 2240 return -ret; 2241 return 0; 2242 } 2243 2244 /* 2245 * All we need to do is get the socket, and then do a checksum. 2246 */ 2247 2248 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 2249 int proto) 2250 { 2251 struct sock *sk; 2252 struct udphdr *uh; 2253 unsigned short ulen; 2254 struct rtable *rt = skb_rtable(skb); 2255 __be32 saddr, daddr; 2256 struct net *net = dev_net(skb->dev); 2257 2258 /* 2259 * Validate the packet. 2260 */ 2261 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2262 goto drop; /* No space for header. */ 2263 2264 uh = udp_hdr(skb); 2265 ulen = ntohs(uh->len); 2266 saddr = ip_hdr(skb)->saddr; 2267 daddr = ip_hdr(skb)->daddr; 2268 2269 if (ulen > skb->len) 2270 goto short_packet; 2271 2272 if (proto == IPPROTO_UDP) { 2273 /* UDP validates ulen. */ 2274 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 2275 goto short_packet; 2276 uh = udp_hdr(skb); 2277 } 2278 2279 if (udp4_csum_init(skb, uh, proto)) 2280 goto csum_error; 2281 2282 sk = skb_steal_sock(skb); 2283 if (sk) { 2284 struct dst_entry *dst = skb_dst(skb); 2285 int ret; 2286 2287 if (unlikely(sk->sk_rx_dst != dst)) 2288 udp_sk_rx_dst_set(sk, dst); 2289 2290 ret = udp_unicast_rcv_skb(sk, skb, uh); 2291 sock_put(sk); 2292 return ret; 2293 } 2294 2295 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 2296 return __udp4_lib_mcast_deliver(net, skb, uh, 2297 saddr, daddr, udptable, proto); 2298 2299 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 2300 if (sk) 2301 return udp_unicast_rcv_skb(sk, skb, uh); 2302 2303 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 2304 goto drop; 2305 nf_reset(skb); 2306 2307 /* No socket. Drop packet silently, if checksum is wrong */ 2308 if (udp_lib_checksum_complete(skb)) 2309 goto csum_error; 2310 2311 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 2312 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 2313 2314 /* 2315 * Hmm. We got an UDP packet to a port to which we 2316 * don't wanna listen. Ignore it. 2317 */ 2318 kfree_skb(skb); 2319 return 0; 2320 2321 short_packet: 2322 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 2323 proto == IPPROTO_UDPLITE ? "Lite" : "", 2324 &saddr, ntohs(uh->source), 2325 ulen, skb->len, 2326 &daddr, ntohs(uh->dest)); 2327 goto drop; 2328 2329 csum_error: 2330 /* 2331 * RFC1122: OK. Discards the bad packet silently (as far as 2332 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 2333 */ 2334 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 2335 proto == IPPROTO_UDPLITE ? "Lite" : "", 2336 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 2337 ulen); 2338 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 2339 drop: 2340 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 2341 kfree_skb(skb); 2342 return 0; 2343 } 2344 2345 /* We can only early demux multicast if there is a single matching socket. 2346 * If more than one socket found returns NULL 2347 */ 2348 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, 2349 __be16 loc_port, __be32 loc_addr, 2350 __be16 rmt_port, __be32 rmt_addr, 2351 int dif, int sdif) 2352 { 2353 struct sock *sk, *result; 2354 unsigned short hnum = ntohs(loc_port); 2355 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); 2356 struct udp_hslot *hslot = &udp_table.hash[slot]; 2357 2358 /* Do not bother scanning a too big list */ 2359 if (hslot->count > 10) 2360 return NULL; 2361 2362 result = NULL; 2363 sk_for_each_rcu(sk, &hslot->head) { 2364 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, 2365 rmt_port, rmt_addr, dif, sdif, hnum)) { 2366 if (result) 2367 return NULL; 2368 result = sk; 2369 } 2370 } 2371 2372 return result; 2373 } 2374 2375 /* For unicast we should only early demux connected sockets or we can 2376 * break forwarding setups. The chains here can be long so only check 2377 * if the first socket is an exact match and if not move on. 2378 */ 2379 static struct sock *__udp4_lib_demux_lookup(struct net *net, 2380 __be16 loc_port, __be32 loc_addr, 2381 __be16 rmt_port, __be32 rmt_addr, 2382 int dif, int sdif) 2383 { 2384 unsigned short hnum = ntohs(loc_port); 2385 unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); 2386 unsigned int slot2 = hash2 & udp_table.mask; 2387 struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; 2388 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); 2389 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); 2390 struct sock *sk; 2391 2392 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 2393 if (INET_MATCH(sk, net, acookie, rmt_addr, 2394 loc_addr, ports, dif, sdif)) 2395 return sk; 2396 /* Only check first socket in chain */ 2397 break; 2398 } 2399 return NULL; 2400 } 2401 2402 int udp_v4_early_demux(struct sk_buff *skb) 2403 { 2404 struct net *net = dev_net(skb->dev); 2405 struct in_device *in_dev = NULL; 2406 const struct iphdr *iph; 2407 const struct udphdr *uh; 2408 struct sock *sk = NULL; 2409 struct dst_entry *dst; 2410 int dif = skb->dev->ifindex; 2411 int sdif = inet_sdif(skb); 2412 int ours; 2413 2414 /* validate the packet */ 2415 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) 2416 return 0; 2417 2418 iph = ip_hdr(skb); 2419 uh = udp_hdr(skb); 2420 2421 if (skb->pkt_type == PACKET_MULTICAST) { 2422 in_dev = __in_dev_get_rcu(skb->dev); 2423 2424 if (!in_dev) 2425 return 0; 2426 2427 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, 2428 iph->protocol); 2429 if (!ours) 2430 return 0; 2431 2432 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, 2433 uh->source, iph->saddr, 2434 dif, sdif); 2435 } else if (skb->pkt_type == PACKET_HOST) { 2436 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, 2437 uh->source, iph->saddr, dif, sdif); 2438 } 2439 2440 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) 2441 return 0; 2442 2443 skb->sk = sk; 2444 skb->destructor = sock_efree; 2445 dst = READ_ONCE(sk->sk_rx_dst); 2446 2447 if (dst) 2448 dst = dst_check(dst, 0); 2449 if (dst) { 2450 u32 itag = 0; 2451 2452 /* set noref for now. 2453 * any place which wants to hold dst has to call 2454 * dst_hold_safe() 2455 */ 2456 skb_dst_set_noref(skb, dst); 2457 2458 /* for unconnected multicast sockets we need to validate 2459 * the source on each packet 2460 */ 2461 if (!inet_sk(sk)->inet_daddr && in_dev) 2462 return ip_mc_validate_source(skb, iph->daddr, 2463 iph->saddr, iph->tos, 2464 skb->dev, in_dev, &itag); 2465 } 2466 return 0; 2467 } 2468 2469 int udp_rcv(struct sk_buff *skb) 2470 { 2471 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 2472 } 2473 2474 void udp_destroy_sock(struct sock *sk) 2475 { 2476 struct udp_sock *up = udp_sk(sk); 2477 bool slow = lock_sock_fast(sk); 2478 udp_flush_pending_frames(sk); 2479 unlock_sock_fast(sk, slow); 2480 if (static_branch_unlikely(&udp_encap_needed_key)) { 2481 if (up->encap_type) { 2482 void (*encap_destroy)(struct sock *sk); 2483 encap_destroy = READ_ONCE(up->encap_destroy); 2484 if (encap_destroy) 2485 encap_destroy(sk); 2486 } 2487 if (up->encap_enabled) 2488 static_branch_dec(&udp_encap_needed_key); 2489 } 2490 } 2491 2492 /* 2493 * Socket option code for UDP 2494 */ 2495 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 2496 char __user *optval, unsigned int optlen, 2497 int (*push_pending_frames)(struct sock *)) 2498 { 2499 struct udp_sock *up = udp_sk(sk); 2500 int val, valbool; 2501 int err = 0; 2502 int is_udplite = IS_UDPLITE(sk); 2503 2504 if (optlen < sizeof(int)) 2505 return -EINVAL; 2506 2507 if (get_user(val, (int __user *)optval)) 2508 return -EFAULT; 2509 2510 valbool = val ? 1 : 0; 2511 2512 switch (optname) { 2513 case UDP_CORK: 2514 if (val != 0) { 2515 up->corkflag = 1; 2516 } else { 2517 up->corkflag = 0; 2518 lock_sock(sk); 2519 push_pending_frames(sk); 2520 release_sock(sk); 2521 } 2522 break; 2523 2524 case UDP_ENCAP: 2525 switch (val) { 2526 case 0: 2527 case UDP_ENCAP_ESPINUDP: 2528 case UDP_ENCAP_ESPINUDP_NON_IKE: 2529 up->encap_rcv = xfrm4_udp_encap_rcv; 2530 /* FALLTHROUGH */ 2531 case UDP_ENCAP_L2TPINUDP: 2532 up->encap_type = val; 2533 lock_sock(sk); 2534 udp_tunnel_encap_enable(sk->sk_socket); 2535 release_sock(sk); 2536 break; 2537 default: 2538 err = -ENOPROTOOPT; 2539 break; 2540 } 2541 break; 2542 2543 case UDP_NO_CHECK6_TX: 2544 up->no_check6_tx = valbool; 2545 break; 2546 2547 case UDP_NO_CHECK6_RX: 2548 up->no_check6_rx = valbool; 2549 break; 2550 2551 case UDP_SEGMENT: 2552 if (val < 0 || val > USHRT_MAX) 2553 return -EINVAL; 2554 up->gso_size = val; 2555 break; 2556 2557 case UDP_GRO: 2558 lock_sock(sk); 2559 if (valbool) 2560 udp_tunnel_encap_enable(sk->sk_socket); 2561 up->gro_enabled = valbool; 2562 release_sock(sk); 2563 break; 2564 2565 /* 2566 * UDP-Lite's partial checksum coverage (RFC 3828). 2567 */ 2568 /* The sender sets actual checksum coverage length via this option. 2569 * The case coverage > packet length is handled by send module. */ 2570 case UDPLITE_SEND_CSCOV: 2571 if (!is_udplite) /* Disable the option on UDP sockets */ 2572 return -ENOPROTOOPT; 2573 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 2574 val = 8; 2575 else if (val > USHRT_MAX) 2576 val = USHRT_MAX; 2577 up->pcslen = val; 2578 up->pcflag |= UDPLITE_SEND_CC; 2579 break; 2580 2581 /* The receiver specifies a minimum checksum coverage value. To make 2582 * sense, this should be set to at least 8 (as done below). If zero is 2583 * used, this again means full checksum coverage. */ 2584 case UDPLITE_RECV_CSCOV: 2585 if (!is_udplite) /* Disable the option on UDP sockets */ 2586 return -ENOPROTOOPT; 2587 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 2588 val = 8; 2589 else if (val > USHRT_MAX) 2590 val = USHRT_MAX; 2591 up->pcrlen = val; 2592 up->pcflag |= UDPLITE_RECV_CC; 2593 break; 2594 2595 default: 2596 err = -ENOPROTOOPT; 2597 break; 2598 } 2599 2600 return err; 2601 } 2602 EXPORT_SYMBOL(udp_lib_setsockopt); 2603 2604 int udp_setsockopt(struct sock *sk, int level, int optname, 2605 char __user *optval, unsigned int optlen) 2606 { 2607 if (level == SOL_UDP || level == SOL_UDPLITE) 2608 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2609 udp_push_pending_frames); 2610 return ip_setsockopt(sk, level, optname, optval, optlen); 2611 } 2612 2613 #ifdef CONFIG_COMPAT 2614 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 2615 char __user *optval, unsigned int optlen) 2616 { 2617 if (level == SOL_UDP || level == SOL_UDPLITE) 2618 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 2619 udp_push_pending_frames); 2620 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 2621 } 2622 #endif 2623 2624 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 2625 char __user *optval, int __user *optlen) 2626 { 2627 struct udp_sock *up = udp_sk(sk); 2628 int val, len; 2629 2630 if (get_user(len, optlen)) 2631 return -EFAULT; 2632 2633 len = min_t(unsigned int, len, sizeof(int)); 2634 2635 if (len < 0) 2636 return -EINVAL; 2637 2638 switch (optname) { 2639 case UDP_CORK: 2640 val = up->corkflag; 2641 break; 2642 2643 case UDP_ENCAP: 2644 val = up->encap_type; 2645 break; 2646 2647 case UDP_NO_CHECK6_TX: 2648 val = up->no_check6_tx; 2649 break; 2650 2651 case UDP_NO_CHECK6_RX: 2652 val = up->no_check6_rx; 2653 break; 2654 2655 case UDP_SEGMENT: 2656 val = up->gso_size; 2657 break; 2658 2659 /* The following two cannot be changed on UDP sockets, the return is 2660 * always 0 (which corresponds to the full checksum coverage of UDP). */ 2661 case UDPLITE_SEND_CSCOV: 2662 val = up->pcslen; 2663 break; 2664 2665 case UDPLITE_RECV_CSCOV: 2666 val = up->pcrlen; 2667 break; 2668 2669 default: 2670 return -ENOPROTOOPT; 2671 } 2672 2673 if (put_user(len, optlen)) 2674 return -EFAULT; 2675 if (copy_to_user(optval, &val, len)) 2676 return -EFAULT; 2677 return 0; 2678 } 2679 EXPORT_SYMBOL(udp_lib_getsockopt); 2680 2681 int udp_getsockopt(struct sock *sk, int level, int optname, 2682 char __user *optval, int __user *optlen) 2683 { 2684 if (level == SOL_UDP || level == SOL_UDPLITE) 2685 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2686 return ip_getsockopt(sk, level, optname, optval, optlen); 2687 } 2688 2689 #ifdef CONFIG_COMPAT 2690 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 2691 char __user *optval, int __user *optlen) 2692 { 2693 if (level == SOL_UDP || level == SOL_UDPLITE) 2694 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 2695 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 2696 } 2697 #endif 2698 /** 2699 * udp_poll - wait for a UDP event. 2700 * @file - file struct 2701 * @sock - socket 2702 * @wait - poll table 2703 * 2704 * This is same as datagram poll, except for the special case of 2705 * blocking sockets. If application is using a blocking fd 2706 * and a packet with checksum error is in the queue; 2707 * then it could get return from select indicating data available 2708 * but then block when reading it. Add special case code 2709 * to work around these arguably broken applications. 2710 */ 2711 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) 2712 { 2713 __poll_t mask = datagram_poll(file, sock, wait); 2714 struct sock *sk = sock->sk; 2715 2716 if (!skb_queue_empty(&udp_sk(sk)->reader_queue)) 2717 mask |= EPOLLIN | EPOLLRDNORM; 2718 2719 /* Check for false positives due to checksum errors */ 2720 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 2721 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) 2722 mask &= ~(EPOLLIN | EPOLLRDNORM); 2723 2724 return mask; 2725 2726 } 2727 EXPORT_SYMBOL(udp_poll); 2728 2729 int udp_abort(struct sock *sk, int err) 2730 { 2731 lock_sock(sk); 2732 2733 sk->sk_err = err; 2734 sk->sk_error_report(sk); 2735 __udp_disconnect(sk, 0); 2736 2737 release_sock(sk); 2738 2739 return 0; 2740 } 2741 EXPORT_SYMBOL_GPL(udp_abort); 2742 2743 struct proto udp_prot = { 2744 .name = "UDP", 2745 .owner = THIS_MODULE, 2746 .close = udp_lib_close, 2747 .pre_connect = udp_pre_connect, 2748 .connect = ip4_datagram_connect, 2749 .disconnect = udp_disconnect, 2750 .ioctl = udp_ioctl, 2751 .init = udp_init_sock, 2752 .destroy = udp_destroy_sock, 2753 .setsockopt = udp_setsockopt, 2754 .getsockopt = udp_getsockopt, 2755 .sendmsg = udp_sendmsg, 2756 .recvmsg = udp_recvmsg, 2757 .sendpage = udp_sendpage, 2758 .release_cb = ip4_datagram_release_cb, 2759 .hash = udp_lib_hash, 2760 .unhash = udp_lib_unhash, 2761 .rehash = udp_v4_rehash, 2762 .get_port = udp_v4_get_port, 2763 .memory_allocated = &udp_memory_allocated, 2764 .sysctl_mem = sysctl_udp_mem, 2765 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 2766 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 2767 .obj_size = sizeof(struct udp_sock), 2768 .h.udp_table = &udp_table, 2769 #ifdef CONFIG_COMPAT 2770 .compat_setsockopt = compat_udp_setsockopt, 2771 .compat_getsockopt = compat_udp_getsockopt, 2772 #endif 2773 .diag_destroy = udp_abort, 2774 }; 2775 EXPORT_SYMBOL(udp_prot); 2776 2777 /* ------------------------------------------------------------------------ */ 2778 #ifdef CONFIG_PROC_FS 2779 2780 static struct sock *udp_get_first(struct seq_file *seq, int start) 2781 { 2782 struct sock *sk; 2783 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2784 struct udp_iter_state *state = seq->private; 2785 struct net *net = seq_file_net(seq); 2786 2787 for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; 2788 ++state->bucket) { 2789 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; 2790 2791 if (hlist_empty(&hslot->head)) 2792 continue; 2793 2794 spin_lock_bh(&hslot->lock); 2795 sk_for_each(sk, &hslot->head) { 2796 if (!net_eq(sock_net(sk), net)) 2797 continue; 2798 if (sk->sk_family == afinfo->family) 2799 goto found; 2800 } 2801 spin_unlock_bh(&hslot->lock); 2802 } 2803 sk = NULL; 2804 found: 2805 return sk; 2806 } 2807 2808 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2809 { 2810 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2811 struct udp_iter_state *state = seq->private; 2812 struct net *net = seq_file_net(seq); 2813 2814 do { 2815 sk = sk_next(sk); 2816 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family)); 2817 2818 if (!sk) { 2819 if (state->bucket <= afinfo->udp_table->mask) 2820 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 2821 return udp_get_first(seq, state->bucket + 1); 2822 } 2823 return sk; 2824 } 2825 2826 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2827 { 2828 struct sock *sk = udp_get_first(seq, 0); 2829 2830 if (sk) 2831 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2832 --pos; 2833 return pos ? NULL : sk; 2834 } 2835 2836 void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2837 { 2838 struct udp_iter_state *state = seq->private; 2839 state->bucket = MAX_UDP_PORTS; 2840 2841 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2842 } 2843 EXPORT_SYMBOL(udp_seq_start); 2844 2845 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2846 { 2847 struct sock *sk; 2848 2849 if (v == SEQ_START_TOKEN) 2850 sk = udp_get_idx(seq, 0); 2851 else 2852 sk = udp_get_next(seq, v); 2853 2854 ++*pos; 2855 return sk; 2856 } 2857 EXPORT_SYMBOL(udp_seq_next); 2858 2859 void udp_seq_stop(struct seq_file *seq, void *v) 2860 { 2861 struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); 2862 struct udp_iter_state *state = seq->private; 2863 2864 if (state->bucket <= afinfo->udp_table->mask) 2865 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); 2866 } 2867 EXPORT_SYMBOL(udp_seq_stop); 2868 2869 /* ------------------------------------------------------------------------ */ 2870 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2871 int bucket) 2872 { 2873 struct inet_sock *inet = inet_sk(sp); 2874 __be32 dest = inet->inet_daddr; 2875 __be32 src = inet->inet_rcv_saddr; 2876 __u16 destp = ntohs(inet->inet_dport); 2877 __u16 srcp = ntohs(inet->inet_sport); 2878 2879 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2880 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", 2881 bucket, src, srcp, dest, destp, sp->sk_state, 2882 sk_wmem_alloc_get(sp), 2883 udp_rqueue_get(sp), 2884 0, 0L, 0, 2885 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2886 0, sock_i_ino(sp), 2887 refcount_read(&sp->sk_refcnt), sp, 2888 atomic_read(&sp->sk_drops)); 2889 } 2890 2891 int udp4_seq_show(struct seq_file *seq, void *v) 2892 { 2893 seq_setwidth(seq, 127); 2894 if (v == SEQ_START_TOKEN) 2895 seq_puts(seq, " sl local_address rem_address st tx_queue " 2896 "rx_queue tr tm->when retrnsmt uid timeout " 2897 "inode ref pointer drops"); 2898 else { 2899 struct udp_iter_state *state = seq->private; 2900 2901 udp4_format_sock(v, seq, state->bucket); 2902 } 2903 seq_pad(seq, '\n'); 2904 return 0; 2905 } 2906 2907 const struct seq_operations udp_seq_ops = { 2908 .start = udp_seq_start, 2909 .next = udp_seq_next, 2910 .stop = udp_seq_stop, 2911 .show = udp4_seq_show, 2912 }; 2913 EXPORT_SYMBOL(udp_seq_ops); 2914 2915 static struct udp_seq_afinfo udp4_seq_afinfo = { 2916 .family = AF_INET, 2917 .udp_table = &udp_table, 2918 }; 2919 2920 static int __net_init udp4_proc_init_net(struct net *net) 2921 { 2922 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, 2923 sizeof(struct udp_iter_state), &udp4_seq_afinfo)) 2924 return -ENOMEM; 2925 return 0; 2926 } 2927 2928 static void __net_exit udp4_proc_exit_net(struct net *net) 2929 { 2930 remove_proc_entry("udp", net->proc_net); 2931 } 2932 2933 static struct pernet_operations udp4_net_ops = { 2934 .init = udp4_proc_init_net, 2935 .exit = udp4_proc_exit_net, 2936 }; 2937 2938 int __init udp4_proc_init(void) 2939 { 2940 return register_pernet_subsys(&udp4_net_ops); 2941 } 2942 2943 void udp4_proc_exit(void) 2944 { 2945 unregister_pernet_subsys(&udp4_net_ops); 2946 } 2947 #endif /* CONFIG_PROC_FS */ 2948 2949 static __initdata unsigned long uhash_entries; 2950 static int __init set_uhash_entries(char *str) 2951 { 2952 ssize_t ret; 2953 2954 if (!str) 2955 return 0; 2956 2957 ret = kstrtoul(str, 0, &uhash_entries); 2958 if (ret) 2959 return 0; 2960 2961 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2962 uhash_entries = UDP_HTABLE_SIZE_MIN; 2963 return 1; 2964 } 2965 __setup("uhash_entries=", set_uhash_entries); 2966 2967 void __init udp_table_init(struct udp_table *table, const char *name) 2968 { 2969 unsigned int i; 2970 2971 table->hash = alloc_large_system_hash(name, 2972 2 * sizeof(struct udp_hslot), 2973 uhash_entries, 2974 21, /* one slot per 2 MB */ 2975 0, 2976 &table->log, 2977 &table->mask, 2978 UDP_HTABLE_SIZE_MIN, 2979 64 * 1024); 2980 2981 table->hash2 = table->hash + (table->mask + 1); 2982 for (i = 0; i <= table->mask; i++) { 2983 INIT_HLIST_HEAD(&table->hash[i].head); 2984 table->hash[i].count = 0; 2985 spin_lock_init(&table->hash[i].lock); 2986 } 2987 for (i = 0; i <= table->mask; i++) { 2988 INIT_HLIST_HEAD(&table->hash2[i].head); 2989 table->hash2[i].count = 0; 2990 spin_lock_init(&table->hash2[i].lock); 2991 } 2992 } 2993 2994 u32 udp_flow_hashrnd(void) 2995 { 2996 static u32 hashrnd __read_mostly; 2997 2998 net_get_random_once(&hashrnd, sizeof(hashrnd)); 2999 3000 return hashrnd; 3001 } 3002 EXPORT_SYMBOL(udp_flow_hashrnd); 3003 3004 static void __udp_sysctl_init(struct net *net) 3005 { 3006 net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; 3007 net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; 3008 3009 #ifdef CONFIG_NET_L3_MASTER_DEV 3010 net->ipv4.sysctl_udp_l3mdev_accept = 0; 3011 #endif 3012 } 3013 3014 static int __net_init udp_sysctl_init(struct net *net) 3015 { 3016 __udp_sysctl_init(net); 3017 return 0; 3018 } 3019 3020 static struct pernet_operations __net_initdata udp_sysctl_ops = { 3021 .init = udp_sysctl_init, 3022 }; 3023 3024 void __init udp_init(void) 3025 { 3026 unsigned long limit; 3027 unsigned int i; 3028 3029 udp_table_init(&udp_table, "UDP"); 3030 limit = nr_free_buffer_pages() / 8; 3031 limit = max(limit, 128UL); 3032 sysctl_udp_mem[0] = limit / 4 * 3; 3033 sysctl_udp_mem[1] = limit; 3034 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 3035 3036 __udp_sysctl_init(&init_net); 3037 3038 /* 16 spinlocks per cpu */ 3039 udp_busylocks_log = ilog2(nr_cpu_ids) + 4; 3040 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, 3041 GFP_KERNEL); 3042 if (!udp_busylocks) 3043 panic("UDP: failed to alloc udp_busylocks\n"); 3044 for (i = 0; i < (1U << udp_busylocks_log); i++) 3045 spin_lock_init(udp_busylocks + i); 3046 3047 if (register_pernet_subsys(&udp_sysctl_ops)) 3048 panic("UDP: failed to init sysctl parameters.\n"); 3049 } 3050