1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The User Datagram Protocol (UDP). 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 12 * Hirokazu Takahashi, <taka@valinux.co.jp> 13 * 14 * Fixes: 15 * Alan Cox : verify_area() calls 16 * Alan Cox : stopped close while in use off icmp 17 * messages. Not a fix but a botch that 18 * for udp at least is 'valid'. 19 * Alan Cox : Fixed icmp handling properly 20 * Alan Cox : Correct error for oversized datagrams 21 * Alan Cox : Tidied select() semantics. 22 * Alan Cox : udp_err() fixed properly, also now 23 * select and read wake correctly on errors 24 * Alan Cox : udp_send verify_area moved to avoid mem leak 25 * Alan Cox : UDP can count its memory 26 * Alan Cox : send to an unknown connection causes 27 * an ECONNREFUSED off the icmp, but 28 * does NOT close. 29 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 31 * bug no longer crashes it. 32 * Fred Van Kempen : Net2e support for sk->broadcast. 33 * Alan Cox : Uses skb_free_datagram 34 * Alan Cox : Added get/set sockopt support. 35 * Alan Cox : Broadcasting without option set returns EACCES. 36 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 37 * Alan Cox : Use ip_tos and ip_ttl 38 * Alan Cox : SNMP Mibs 39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 40 * Matt Dillon : UDP length checks. 41 * Alan Cox : Smarter af_inet used properly. 42 * Alan Cox : Use new kernel side addressing. 43 * Alan Cox : Incorrect return on truncated datagram receive. 44 * Arnt Gulbrandsen : New udp_send and stuff 45 * Alan Cox : Cache last socket 46 * Alan Cox : Route cache 47 * Jon Peatfield : Minor efficiency fix to sendto(). 48 * Mike Shaver : RFC1122 checks. 49 * Alan Cox : Nonblocking error fix. 50 * Willy Konynenberg : Transparent proxying support. 51 * Mike McLagan : Routing by source 52 * David S. Miller : New socket lookup architecture. 53 * Last socket cache retained as it 54 * does have a high hit rate. 55 * Olaf Kirch : Don't linearise iovec on sendmsg. 56 * Andi Kleen : Some cleanups, cache destination entry 57 * for connect. 58 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 59 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 60 * return ENOTCONN for unconnected sockets (POSIX) 61 * Janos Farkas : don't deliver multi/broadcasts to a different 62 * bound-to-device socket 63 * Hirokazu Takahashi : HW checksumming for outgoing UDP 64 * datagrams. 65 * Hirokazu Takahashi : sendfile() on UDP works now. 66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 69 * a single port at the same time. 70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71 * James Chapman : Add L2TP encapsulation type. 72 * 73 * 74 * This program is free software; you can redistribute it and/or 75 * modify it under the terms of the GNU General Public License 76 * as published by the Free Software Foundation; either version 77 * 2 of the License, or (at your option) any later version. 78 */ 79 80 #define pr_fmt(fmt) "UDP: " fmt 81 82 #include <asm/uaccess.h> 83 #include <asm/ioctls.h> 84 #include <linux/bootmem.h> 85 #include <linux/highmem.h> 86 #include <linux/swap.h> 87 #include <linux/types.h> 88 #include <linux/fcntl.h> 89 #include <linux/module.h> 90 #include <linux/socket.h> 91 #include <linux/sockios.h> 92 #include <linux/igmp.h> 93 #include <linux/in.h> 94 #include <linux/errno.h> 95 #include <linux/timer.h> 96 #include <linux/mm.h> 97 #include <linux/inet.h> 98 #include <linux/netdevice.h> 99 #include <linux/slab.h> 100 #include <net/tcp_states.h> 101 #include <linux/skbuff.h> 102 #include <linux/proc_fs.h> 103 #include <linux/seq_file.h> 104 #include <net/net_namespace.h> 105 #include <net/icmp.h> 106 #include <net/route.h> 107 #include <net/checksum.h> 108 #include <net/xfrm.h> 109 #include <trace/events/udp.h> 110 #include <linux/static_key.h> 111 #include <trace/events/skb.h> 112 #include "udp_impl.h" 113 114 struct udp_table udp_table __read_mostly; 115 EXPORT_SYMBOL(udp_table); 116 117 long sysctl_udp_mem[3] __read_mostly; 118 EXPORT_SYMBOL(sysctl_udp_mem); 119 120 int sysctl_udp_rmem_min __read_mostly; 121 EXPORT_SYMBOL(sysctl_udp_rmem_min); 122 123 int sysctl_udp_wmem_min __read_mostly; 124 EXPORT_SYMBOL(sysctl_udp_wmem_min); 125 126 atomic_long_t udp_memory_allocated; 127 EXPORT_SYMBOL(udp_memory_allocated); 128 129 #define MAX_UDP_PORTS 65536 130 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) 131 132 static int udp_lib_lport_inuse(struct net *net, __u16 num, 133 const struct udp_hslot *hslot, 134 unsigned long *bitmap, 135 struct sock *sk, 136 int (*saddr_comp)(const struct sock *sk1, 137 const struct sock *sk2), 138 unsigned int log) 139 { 140 struct sock *sk2; 141 struct hlist_nulls_node *node; 142 kuid_t uid = sock_i_uid(sk); 143 144 sk_nulls_for_each(sk2, node, &hslot->head) 145 if (net_eq(sock_net(sk2), net) && 146 sk2 != sk && 147 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 148 (!sk2->sk_reuse || !sk->sk_reuse) && 149 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 150 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 151 (!sk2->sk_reuseport || !sk->sk_reuseport || 152 !uid_eq(uid, sock_i_uid(sk2))) && 153 (*saddr_comp)(sk, sk2)) { 154 if (bitmap) 155 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 156 bitmap); 157 else 158 return 1; 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 int (*saddr_comp)(const struct sock *sk1, 171 const struct sock *sk2)) 172 { 173 struct sock *sk2; 174 struct hlist_nulls_node *node; 175 kuid_t uid = sock_i_uid(sk); 176 int res = 0; 177 178 spin_lock(&hslot2->lock); 179 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) 180 if (net_eq(sock_net(sk2), net) && 181 sk2 != sk && 182 (udp_sk(sk2)->udp_port_hash == num) && 183 (!sk2->sk_reuse || !sk->sk_reuse) && 184 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 185 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 186 (!sk2->sk_reuseport || !sk->sk_reuseport || 187 !uid_eq(uid, sock_i_uid(sk2))) && 188 (*saddr_comp)(sk, sk2)) { 189 res = 1; 190 break; 191 } 192 spin_unlock(&hslot2->lock); 193 return res; 194 } 195 196 /** 197 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 198 * 199 * @sk: socket struct in question 200 * @snum: port number to look up 201 * @saddr_comp: AF-dependent comparison of bound local IP addresses 202 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 203 * with NULL address 204 */ 205 int udp_lib_get_port(struct sock *sk, unsigned short snum, 206 int (*saddr_comp)(const struct sock *sk1, 207 const struct sock *sk2), 208 unsigned int hash2_nulladdr) 209 { 210 struct udp_hslot *hslot, *hslot2; 211 struct udp_table *udptable = sk->sk_prot->h.udp_table; 212 int error = 1; 213 struct net *net = sock_net(sk); 214 215 if (!snum) { 216 int low, high, remaining; 217 unsigned int rand; 218 unsigned short first, last; 219 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 220 221 inet_get_local_port_range(&low, &high); 222 remaining = (high - low) + 1; 223 224 rand = net_random(); 225 first = (((u64)rand * remaining) >> 32) + low; 226 /* 227 * force rand to be an odd multiple of UDP_HTABLE_SIZE 228 */ 229 rand = (rand | 1) * (udptable->mask + 1); 230 last = first + udptable->mask + 1; 231 do { 232 hslot = udp_hashslot(udptable, net, first); 233 bitmap_zero(bitmap, PORTS_PER_CHAIN); 234 spin_lock_bh(&hslot->lock); 235 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 236 saddr_comp, udptable->log); 237 238 snum = first; 239 /* 240 * Iterate on all possible values of snum for this hash. 241 * Using steps of an odd multiple of UDP_HTABLE_SIZE 242 * give us randomization and full range coverage. 243 */ 244 do { 245 if (low <= snum && snum <= high && 246 !test_bit(snum >> udptable->log, bitmap) && 247 !inet_is_reserved_local_port(snum)) 248 goto found; 249 snum += rand; 250 } while (snum != first); 251 spin_unlock_bh(&hslot->lock); 252 } while (++first != last); 253 goto fail; 254 } else { 255 hslot = udp_hashslot(udptable, net, snum); 256 spin_lock_bh(&hslot->lock); 257 if (hslot->count > 10) { 258 int exist; 259 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 260 261 slot2 &= udptable->mask; 262 hash2_nulladdr &= udptable->mask; 263 264 hslot2 = udp_hashslot2(udptable, slot2); 265 if (hslot->count < hslot2->count) 266 goto scan_primary_hash; 267 268 exist = udp_lib_lport_inuse2(net, snum, hslot2, 269 sk, saddr_comp); 270 if (!exist && (hash2_nulladdr != slot2)) { 271 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 272 exist = udp_lib_lport_inuse2(net, snum, hslot2, 273 sk, saddr_comp); 274 } 275 if (exist) 276 goto fail_unlock; 277 else 278 goto found; 279 } 280 scan_primary_hash: 281 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 282 saddr_comp, 0)) 283 goto fail_unlock; 284 } 285 found: 286 inet_sk(sk)->inet_num = snum; 287 udp_sk(sk)->udp_port_hash = snum; 288 udp_sk(sk)->udp_portaddr_hash ^= snum; 289 if (sk_unhashed(sk)) { 290 sk_nulls_add_node_rcu(sk, &hslot->head); 291 hslot->count++; 292 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 293 294 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 295 spin_lock(&hslot2->lock); 296 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 297 &hslot2->head); 298 hslot2->count++; 299 spin_unlock(&hslot2->lock); 300 } 301 error = 0; 302 fail_unlock: 303 spin_unlock_bh(&hslot->lock); 304 fail: 305 return error; 306 } 307 EXPORT_SYMBOL(udp_lib_get_port); 308 309 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 310 { 311 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 312 313 return (!ipv6_only_sock(sk2) && 314 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 315 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 316 } 317 318 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 319 unsigned int port) 320 { 321 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 322 } 323 324 int udp_v4_get_port(struct sock *sk, unsigned short snum) 325 { 326 unsigned int hash2_nulladdr = 327 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 328 unsigned int hash2_partial = 329 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 330 331 /* precompute partial secondary hash */ 332 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 333 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 334 } 335 336 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 337 unsigned short hnum, 338 __be16 sport, __be32 daddr, __be16 dport, int dif) 339 { 340 int score = -1; 341 342 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 343 !ipv6_only_sock(sk)) { 344 struct inet_sock *inet = inet_sk(sk); 345 346 score = (sk->sk_family == PF_INET ? 2 : 1); 347 if (inet->inet_rcv_saddr) { 348 if (inet->inet_rcv_saddr != daddr) 349 return -1; 350 score += 4; 351 } 352 if (inet->inet_daddr) { 353 if (inet->inet_daddr != saddr) 354 return -1; 355 score += 4; 356 } 357 if (inet->inet_dport) { 358 if (inet->inet_dport != sport) 359 return -1; 360 score += 4; 361 } 362 if (sk->sk_bound_dev_if) { 363 if (sk->sk_bound_dev_if != dif) 364 return -1; 365 score += 4; 366 } 367 } 368 return score; 369 } 370 371 /* 372 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 373 */ 374 static inline int compute_score2(struct sock *sk, struct net *net, 375 __be32 saddr, __be16 sport, 376 __be32 daddr, unsigned int hnum, int dif) 377 { 378 int score = -1; 379 380 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) { 381 struct inet_sock *inet = inet_sk(sk); 382 383 if (inet->inet_rcv_saddr != daddr) 384 return -1; 385 if (inet->inet_num != hnum) 386 return -1; 387 388 score = (sk->sk_family == PF_INET ? 2 : 1); 389 if (inet->inet_daddr) { 390 if (inet->inet_daddr != saddr) 391 return -1; 392 score += 4; 393 } 394 if (inet->inet_dport) { 395 if (inet->inet_dport != sport) 396 return -1; 397 score += 4; 398 } 399 if (sk->sk_bound_dev_if) { 400 if (sk->sk_bound_dev_if != dif) 401 return -1; 402 score += 4; 403 } 404 } 405 return score; 406 } 407 408 409 /* called with read_rcu_lock() */ 410 static struct sock *udp4_lib_lookup2(struct net *net, 411 __be32 saddr, __be16 sport, 412 __be32 daddr, unsigned int hnum, int dif, 413 struct udp_hslot *hslot2, unsigned int slot2) 414 { 415 struct sock *sk, *result; 416 struct hlist_nulls_node *node; 417 int score, badness, matches = 0, reuseport = 0; 418 u32 hash = 0; 419 420 begin: 421 result = NULL; 422 badness = 0; 423 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 424 score = compute_score2(sk, net, saddr, sport, 425 daddr, hnum, dif); 426 if (score > badness) { 427 result = sk; 428 badness = score; 429 reuseport = sk->sk_reuseport; 430 if (reuseport) { 431 hash = inet_ehashfn(net, daddr, hnum, 432 saddr, htons(sport)); 433 matches = 1; 434 } 435 } else if (score == badness && reuseport) { 436 matches++; 437 if (((u64)hash * matches) >> 32 == 0) 438 result = sk; 439 hash = next_pseudo_random32(hash); 440 } 441 } 442 /* 443 * if the nulls value we got at the end of this lookup is 444 * not the expected one, we must restart lookup. 445 * We probably met an item that was moved to another chain. 446 */ 447 if (get_nulls_value(node) != slot2) 448 goto begin; 449 if (result) { 450 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 451 result = NULL; 452 else if (unlikely(compute_score2(result, net, saddr, sport, 453 daddr, hnum, dif) < badness)) { 454 sock_put(result); 455 goto begin; 456 } 457 } 458 return result; 459 } 460 461 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 462 * harder than this. -DaveM 463 */ 464 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 465 __be16 sport, __be32 daddr, __be16 dport, 466 int dif, struct udp_table *udptable) 467 { 468 struct sock *sk, *result; 469 struct hlist_nulls_node *node; 470 unsigned short hnum = ntohs(dport); 471 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 472 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 473 int score, badness, matches = 0, reuseport = 0; 474 u32 hash = 0; 475 476 rcu_read_lock(); 477 if (hslot->count > 10) { 478 hash2 = udp4_portaddr_hash(net, daddr, hnum); 479 slot2 = hash2 & udptable->mask; 480 hslot2 = &udptable->hash2[slot2]; 481 if (hslot->count < hslot2->count) 482 goto begin; 483 484 result = udp4_lib_lookup2(net, saddr, sport, 485 daddr, hnum, dif, 486 hslot2, slot2); 487 if (!result) { 488 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 489 slot2 = hash2 & udptable->mask; 490 hslot2 = &udptable->hash2[slot2]; 491 if (hslot->count < hslot2->count) 492 goto begin; 493 494 result = udp4_lib_lookup2(net, saddr, sport, 495 htonl(INADDR_ANY), hnum, dif, 496 hslot2, slot2); 497 } 498 rcu_read_unlock(); 499 return result; 500 } 501 begin: 502 result = NULL; 503 badness = 0; 504 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 505 score = compute_score(sk, net, saddr, hnum, sport, 506 daddr, dport, dif); 507 if (score > badness) { 508 result = sk; 509 badness = score; 510 reuseport = sk->sk_reuseport; 511 if (reuseport) { 512 hash = inet_ehashfn(net, daddr, hnum, 513 saddr, htons(sport)); 514 matches = 1; 515 } 516 } else if (score == badness && reuseport) { 517 matches++; 518 if (((u64)hash * matches) >> 32 == 0) 519 result = sk; 520 hash = next_pseudo_random32(hash); 521 } 522 } 523 /* 524 * if the nulls value we got at the end of this lookup is 525 * not the expected one, we must restart lookup. 526 * We probably met an item that was moved to another chain. 527 */ 528 if (get_nulls_value(node) != slot) 529 goto begin; 530 531 if (result) { 532 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 533 result = NULL; 534 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 535 daddr, dport, dif) < badness)) { 536 sock_put(result); 537 goto begin; 538 } 539 } 540 rcu_read_unlock(); 541 return result; 542 } 543 EXPORT_SYMBOL_GPL(__udp4_lib_lookup); 544 545 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 546 __be16 sport, __be16 dport, 547 struct udp_table *udptable) 548 { 549 struct sock *sk; 550 const struct iphdr *iph = ip_hdr(skb); 551 552 if (unlikely(sk = skb_steal_sock(skb))) 553 return sk; 554 else 555 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 556 iph->daddr, dport, inet_iif(skb), 557 udptable); 558 } 559 560 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 561 __be32 daddr, __be16 dport, int dif) 562 { 563 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 564 } 565 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 566 567 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 568 __be16 loc_port, __be32 loc_addr, 569 __be16 rmt_port, __be32 rmt_addr, 570 int dif) 571 { 572 struct hlist_nulls_node *node; 573 struct sock *s = sk; 574 unsigned short hnum = ntohs(loc_port); 575 576 sk_nulls_for_each_from(s, node) { 577 struct inet_sock *inet = inet_sk(s); 578 579 if (!net_eq(sock_net(s), net) || 580 udp_sk(s)->udp_port_hash != hnum || 581 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 582 (inet->inet_dport != rmt_port && inet->inet_dport) || 583 (inet->inet_rcv_saddr && 584 inet->inet_rcv_saddr != loc_addr) || 585 ipv6_only_sock(s) || 586 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 587 continue; 588 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 589 continue; 590 goto found; 591 } 592 s = NULL; 593 found: 594 return s; 595 } 596 597 /* 598 * This routine is called by the ICMP module when it gets some 599 * sort of error condition. If err < 0 then the socket should 600 * be closed and the error returned to the user. If err > 0 601 * it's just the icmp type << 8 | icmp code. 602 * Header points to the ip header of the error packet. We move 603 * on past this. Then (as it used to claim before adjustment) 604 * header points to the first 8 bytes of the udp header. We need 605 * to find the appropriate port. 606 */ 607 608 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 609 { 610 struct inet_sock *inet; 611 const struct iphdr *iph = (const struct iphdr *)skb->data; 612 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 613 const int type = icmp_hdr(skb)->type; 614 const int code = icmp_hdr(skb)->code; 615 struct sock *sk; 616 int harderr; 617 int err; 618 struct net *net = dev_net(skb->dev); 619 620 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 621 iph->saddr, uh->source, skb->dev->ifindex, udptable); 622 if (sk == NULL) { 623 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 624 return; /* No socket for error */ 625 } 626 627 err = 0; 628 harderr = 0; 629 inet = inet_sk(sk); 630 631 switch (type) { 632 default: 633 case ICMP_TIME_EXCEEDED: 634 err = EHOSTUNREACH; 635 break; 636 case ICMP_SOURCE_QUENCH: 637 goto out; 638 case ICMP_PARAMETERPROB: 639 err = EPROTO; 640 harderr = 1; 641 break; 642 case ICMP_DEST_UNREACH: 643 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 644 ipv4_sk_update_pmtu(skb, sk, info); 645 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 646 err = EMSGSIZE; 647 harderr = 1; 648 break; 649 } 650 goto out; 651 } 652 err = EHOSTUNREACH; 653 if (code <= NR_ICMP_UNREACH) { 654 harderr = icmp_err_convert[code].fatal; 655 err = icmp_err_convert[code].errno; 656 } 657 break; 658 case ICMP_REDIRECT: 659 ipv4_sk_redirect(skb, sk); 660 break; 661 } 662 663 /* 664 * RFC1122: OK. Passes ICMP errors back to application, as per 665 * 4.1.3.3. 666 */ 667 if (!inet->recverr) { 668 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 669 goto out; 670 } else 671 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 672 673 sk->sk_err = err; 674 sk->sk_error_report(sk); 675 out: 676 sock_put(sk); 677 } 678 679 void udp_err(struct sk_buff *skb, u32 info) 680 { 681 __udp4_lib_err(skb, info, &udp_table); 682 } 683 684 /* 685 * Throw away all pending data and cancel the corking. Socket is locked. 686 */ 687 void udp_flush_pending_frames(struct sock *sk) 688 { 689 struct udp_sock *up = udp_sk(sk); 690 691 if (up->pending) { 692 up->len = 0; 693 up->pending = 0; 694 ip_flush_pending_frames(sk); 695 } 696 } 697 EXPORT_SYMBOL(udp_flush_pending_frames); 698 699 /** 700 * udp4_hwcsum - handle outgoing HW checksumming 701 * @skb: sk_buff containing the filled-in UDP header 702 * (checksum field must be zeroed out) 703 * @src: source IP address 704 * @dst: destination IP address 705 */ 706 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 707 { 708 struct udphdr *uh = udp_hdr(skb); 709 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 710 int offset = skb_transport_offset(skb); 711 int len = skb->len - offset; 712 int hlen = len; 713 __wsum csum = 0; 714 715 if (!frags) { 716 /* 717 * Only one fragment on the socket. 718 */ 719 skb->csum_start = skb_transport_header(skb) - skb->head; 720 skb->csum_offset = offsetof(struct udphdr, check); 721 uh->check = ~csum_tcpudp_magic(src, dst, len, 722 IPPROTO_UDP, 0); 723 } else { 724 /* 725 * HW-checksum won't work as there are two or more 726 * fragments on the socket so that all csums of sk_buffs 727 * should be together 728 */ 729 do { 730 csum = csum_add(csum, frags->csum); 731 hlen -= frags->len; 732 } while ((frags = frags->next)); 733 734 csum = skb_checksum(skb, offset, hlen, csum); 735 skb->ip_summed = CHECKSUM_NONE; 736 737 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 738 if (uh->check == 0) 739 uh->check = CSUM_MANGLED_0; 740 } 741 } 742 743 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 744 { 745 struct sock *sk = skb->sk; 746 struct inet_sock *inet = inet_sk(sk); 747 struct udphdr *uh; 748 int err = 0; 749 int is_udplite = IS_UDPLITE(sk); 750 int offset = skb_transport_offset(skb); 751 int len = skb->len - offset; 752 __wsum csum = 0; 753 754 /* 755 * Create a UDP header 756 */ 757 uh = udp_hdr(skb); 758 uh->source = inet->inet_sport; 759 uh->dest = fl4->fl4_dport; 760 uh->len = htons(len); 761 uh->check = 0; 762 763 if (is_udplite) /* UDP-Lite */ 764 csum = udplite_csum(skb); 765 766 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 767 768 skb->ip_summed = CHECKSUM_NONE; 769 goto send; 770 771 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 772 773 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 774 goto send; 775 776 } else 777 csum = udp_csum(skb); 778 779 /* add protocol-dependent pseudo-header */ 780 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 781 sk->sk_protocol, csum); 782 if (uh->check == 0) 783 uh->check = CSUM_MANGLED_0; 784 785 send: 786 err = ip_send_skb(sock_net(sk), skb); 787 if (err) { 788 if (err == -ENOBUFS && !inet->recverr) { 789 UDP_INC_STATS_USER(sock_net(sk), 790 UDP_MIB_SNDBUFERRORS, is_udplite); 791 err = 0; 792 } 793 } else 794 UDP_INC_STATS_USER(sock_net(sk), 795 UDP_MIB_OUTDATAGRAMS, is_udplite); 796 return err; 797 } 798 799 /* 800 * Push out all pending data as one UDP datagram. Socket is locked. 801 */ 802 static int udp_push_pending_frames(struct sock *sk) 803 { 804 struct udp_sock *up = udp_sk(sk); 805 struct inet_sock *inet = inet_sk(sk); 806 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 807 struct sk_buff *skb; 808 int err = 0; 809 810 skb = ip_finish_skb(sk, fl4); 811 if (!skb) 812 goto out; 813 814 err = udp_send_skb(skb, fl4); 815 816 out: 817 up->len = 0; 818 up->pending = 0; 819 return err; 820 } 821 822 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 823 size_t len) 824 { 825 struct inet_sock *inet = inet_sk(sk); 826 struct udp_sock *up = udp_sk(sk); 827 struct flowi4 fl4_stack; 828 struct flowi4 *fl4; 829 int ulen = len; 830 struct ipcm_cookie ipc; 831 struct rtable *rt = NULL; 832 int free = 0; 833 int connected = 0; 834 __be32 daddr, faddr, saddr; 835 __be16 dport; 836 u8 tos; 837 int err, is_udplite = IS_UDPLITE(sk); 838 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 839 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 840 struct sk_buff *skb; 841 struct ip_options_data opt_copy; 842 843 if (len > 0xFFFF) 844 return -EMSGSIZE; 845 846 /* 847 * Check the flags. 848 */ 849 850 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 851 return -EOPNOTSUPP; 852 853 ipc.opt = NULL; 854 ipc.tx_flags = 0; 855 856 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 857 858 fl4 = &inet->cork.fl.u.ip4; 859 if (up->pending) { 860 /* 861 * There are pending frames. 862 * The socket lock must be held while it's corked. 863 */ 864 lock_sock(sk); 865 if (likely(up->pending)) { 866 if (unlikely(up->pending != AF_INET)) { 867 release_sock(sk); 868 return -EINVAL; 869 } 870 goto do_append_data; 871 } 872 release_sock(sk); 873 } 874 ulen += sizeof(struct udphdr); 875 876 /* 877 * Get and verify the address. 878 */ 879 if (msg->msg_name) { 880 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name; 881 if (msg->msg_namelen < sizeof(*usin)) 882 return -EINVAL; 883 if (usin->sin_family != AF_INET) { 884 if (usin->sin_family != AF_UNSPEC) 885 return -EAFNOSUPPORT; 886 } 887 888 daddr = usin->sin_addr.s_addr; 889 dport = usin->sin_port; 890 if (dport == 0) 891 return -EINVAL; 892 } else { 893 if (sk->sk_state != TCP_ESTABLISHED) 894 return -EDESTADDRREQ; 895 daddr = inet->inet_daddr; 896 dport = inet->inet_dport; 897 /* Open fast path for connected socket. 898 Route will not be used, if at least one option is set. 899 */ 900 connected = 1; 901 } 902 ipc.addr = inet->inet_saddr; 903 904 ipc.oif = sk->sk_bound_dev_if; 905 906 sock_tx_timestamp(sk, &ipc.tx_flags); 907 908 if (msg->msg_controllen) { 909 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 910 if (err) 911 return err; 912 if (ipc.opt) 913 free = 1; 914 connected = 0; 915 } 916 if (!ipc.opt) { 917 struct ip_options_rcu *inet_opt; 918 919 rcu_read_lock(); 920 inet_opt = rcu_dereference(inet->inet_opt); 921 if (inet_opt) { 922 memcpy(&opt_copy, inet_opt, 923 sizeof(*inet_opt) + inet_opt->opt.optlen); 924 ipc.opt = &opt_copy.opt; 925 } 926 rcu_read_unlock(); 927 } 928 929 saddr = ipc.addr; 930 ipc.addr = faddr = daddr; 931 932 if (ipc.opt && ipc.opt->opt.srr) { 933 if (!daddr) 934 return -EINVAL; 935 faddr = ipc.opt->opt.faddr; 936 connected = 0; 937 } 938 tos = RT_TOS(inet->tos); 939 if (sock_flag(sk, SOCK_LOCALROUTE) || 940 (msg->msg_flags & MSG_DONTROUTE) || 941 (ipc.opt && ipc.opt->opt.is_strictroute)) { 942 tos |= RTO_ONLINK; 943 connected = 0; 944 } 945 946 if (ipv4_is_multicast(daddr)) { 947 if (!ipc.oif) 948 ipc.oif = inet->mc_index; 949 if (!saddr) 950 saddr = inet->mc_addr; 951 connected = 0; 952 } else if (!ipc.oif) 953 ipc.oif = inet->uc_index; 954 955 if (connected) 956 rt = (struct rtable *)sk_dst_check(sk, 0); 957 958 if (rt == NULL) { 959 struct net *net = sock_net(sk); 960 961 fl4 = &fl4_stack; 962 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 963 RT_SCOPE_UNIVERSE, sk->sk_protocol, 964 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP, 965 faddr, saddr, dport, inet->inet_sport); 966 967 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 968 rt = ip_route_output_flow(net, fl4, sk); 969 if (IS_ERR(rt)) { 970 err = PTR_ERR(rt); 971 rt = NULL; 972 if (err == -ENETUNREACH) 973 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 974 goto out; 975 } 976 977 err = -EACCES; 978 if ((rt->rt_flags & RTCF_BROADCAST) && 979 !sock_flag(sk, SOCK_BROADCAST)) 980 goto out; 981 if (connected) 982 sk_dst_set(sk, dst_clone(&rt->dst)); 983 } 984 985 if (msg->msg_flags&MSG_CONFIRM) 986 goto do_confirm; 987 back_from_confirm: 988 989 saddr = fl4->saddr; 990 if (!ipc.addr) 991 daddr = ipc.addr = fl4->daddr; 992 993 /* Lockless fast path for the non-corking case. */ 994 if (!corkreq) { 995 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen, 996 sizeof(struct udphdr), &ipc, &rt, 997 msg->msg_flags); 998 err = PTR_ERR(skb); 999 if (!IS_ERR_OR_NULL(skb)) 1000 err = udp_send_skb(skb, fl4); 1001 goto out; 1002 } 1003 1004 lock_sock(sk); 1005 if (unlikely(up->pending)) { 1006 /* The socket is already corked while preparing it. */ 1007 /* ... which is an evident application bug. --ANK */ 1008 release_sock(sk); 1009 1010 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n")); 1011 err = -EINVAL; 1012 goto out; 1013 } 1014 /* 1015 * Now cork the socket to pend data. 1016 */ 1017 fl4 = &inet->cork.fl.u.ip4; 1018 fl4->daddr = daddr; 1019 fl4->saddr = saddr; 1020 fl4->fl4_dport = dport; 1021 fl4->fl4_sport = inet->inet_sport; 1022 up->pending = AF_INET; 1023 1024 do_append_data: 1025 up->len += ulen; 1026 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen, 1027 sizeof(struct udphdr), &ipc, &rt, 1028 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1029 if (err) 1030 udp_flush_pending_frames(sk); 1031 else if (!corkreq) 1032 err = udp_push_pending_frames(sk); 1033 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1034 up->pending = 0; 1035 release_sock(sk); 1036 1037 out: 1038 ip_rt_put(rt); 1039 if (free) 1040 kfree(ipc.opt); 1041 if (!err) 1042 return len; 1043 /* 1044 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1045 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1046 * we don't have a good statistic (IpOutDiscards but it can be too many 1047 * things). We could add another new stat but at least for now that 1048 * seems like overkill. 1049 */ 1050 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1051 UDP_INC_STATS_USER(sock_net(sk), 1052 UDP_MIB_SNDBUFERRORS, is_udplite); 1053 } 1054 return err; 1055 1056 do_confirm: 1057 dst_confirm(&rt->dst); 1058 if (!(msg->msg_flags&MSG_PROBE) || len) 1059 goto back_from_confirm; 1060 err = 0; 1061 goto out; 1062 } 1063 EXPORT_SYMBOL(udp_sendmsg); 1064 1065 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1066 size_t size, int flags) 1067 { 1068 struct inet_sock *inet = inet_sk(sk); 1069 struct udp_sock *up = udp_sk(sk); 1070 int ret; 1071 1072 if (!up->pending) { 1073 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1074 1075 /* Call udp_sendmsg to specify destination address which 1076 * sendpage interface can't pass. 1077 * This will succeed only when the socket is connected. 1078 */ 1079 ret = udp_sendmsg(NULL, sk, &msg, 0); 1080 if (ret < 0) 1081 return ret; 1082 } 1083 1084 lock_sock(sk); 1085 1086 if (unlikely(!up->pending)) { 1087 release_sock(sk); 1088 1089 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n")); 1090 return -EINVAL; 1091 } 1092 1093 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1094 page, offset, size, flags); 1095 if (ret == -EOPNOTSUPP) { 1096 release_sock(sk); 1097 return sock_no_sendpage(sk->sk_socket, page, offset, 1098 size, flags); 1099 } 1100 if (ret < 0) { 1101 udp_flush_pending_frames(sk); 1102 goto out; 1103 } 1104 1105 up->len += size; 1106 if (!(up->corkflag || (flags&MSG_MORE))) 1107 ret = udp_push_pending_frames(sk); 1108 if (!ret) 1109 ret = size; 1110 out: 1111 release_sock(sk); 1112 return ret; 1113 } 1114 1115 1116 /** 1117 * first_packet_length - return length of first packet in receive queue 1118 * @sk: socket 1119 * 1120 * Drops all bad checksum frames, until a valid one is found. 1121 * Returns the length of found skb, or 0 if none is found. 1122 */ 1123 static unsigned int first_packet_length(struct sock *sk) 1124 { 1125 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1126 struct sk_buff *skb; 1127 unsigned int res; 1128 1129 __skb_queue_head_init(&list_kill); 1130 1131 spin_lock_bh(&rcvq->lock); 1132 while ((skb = skb_peek(rcvq)) != NULL && 1133 udp_lib_checksum_complete(skb)) { 1134 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, 1135 IS_UDPLITE(sk)); 1136 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1137 IS_UDPLITE(sk)); 1138 atomic_inc(&sk->sk_drops); 1139 __skb_unlink(skb, rcvq); 1140 __skb_queue_tail(&list_kill, skb); 1141 } 1142 res = skb ? skb->len : 0; 1143 spin_unlock_bh(&rcvq->lock); 1144 1145 if (!skb_queue_empty(&list_kill)) { 1146 bool slow = lock_sock_fast(sk); 1147 1148 __skb_queue_purge(&list_kill); 1149 sk_mem_reclaim_partial(sk); 1150 unlock_sock_fast(sk, slow); 1151 } 1152 return res; 1153 } 1154 1155 /* 1156 * IOCTL requests applicable to the UDP protocol 1157 */ 1158 1159 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1160 { 1161 switch (cmd) { 1162 case SIOCOUTQ: 1163 { 1164 int amount = sk_wmem_alloc_get(sk); 1165 1166 return put_user(amount, (int __user *)arg); 1167 } 1168 1169 case SIOCINQ: 1170 { 1171 unsigned int amount = first_packet_length(sk); 1172 1173 if (amount) 1174 /* 1175 * We will only return the amount 1176 * of this packet since that is all 1177 * that will be read. 1178 */ 1179 amount -= sizeof(struct udphdr); 1180 1181 return put_user(amount, (int __user *)arg); 1182 } 1183 1184 default: 1185 return -ENOIOCTLCMD; 1186 } 1187 1188 return 0; 1189 } 1190 EXPORT_SYMBOL(udp_ioctl); 1191 1192 /* 1193 * This should be easy, if there is something there we 1194 * return it, otherwise we block. 1195 */ 1196 1197 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1198 size_t len, int noblock, int flags, int *addr_len) 1199 { 1200 struct inet_sock *inet = inet_sk(sk); 1201 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 1202 struct sk_buff *skb; 1203 unsigned int ulen, copied; 1204 int peeked, off = 0; 1205 int err; 1206 int is_udplite = IS_UDPLITE(sk); 1207 bool slow; 1208 1209 /* 1210 * Check any passed addresses 1211 */ 1212 if (addr_len) 1213 *addr_len = sizeof(*sin); 1214 1215 if (flags & MSG_ERRQUEUE) 1216 return ip_recv_error(sk, msg, len); 1217 1218 try_again: 1219 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1220 &peeked, &off, &err); 1221 if (!skb) 1222 goto out; 1223 1224 ulen = skb->len - sizeof(struct udphdr); 1225 copied = len; 1226 if (copied > ulen) 1227 copied = ulen; 1228 else if (copied < ulen) 1229 msg->msg_flags |= MSG_TRUNC; 1230 1231 /* 1232 * If checksum is needed at all, try to do it while copying the 1233 * data. If the data is truncated, or if we only want a partial 1234 * coverage checksum (UDP-Lite), do it before the copy. 1235 */ 1236 1237 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 1238 if (udp_lib_checksum_complete(skb)) 1239 goto csum_copy_err; 1240 } 1241 1242 if (skb_csum_unnecessary(skb)) 1243 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1244 msg->msg_iov, copied); 1245 else { 1246 err = skb_copy_and_csum_datagram_iovec(skb, 1247 sizeof(struct udphdr), 1248 msg->msg_iov); 1249 1250 if (err == -EINVAL) 1251 goto csum_copy_err; 1252 } 1253 1254 if (unlikely(err)) { 1255 trace_kfree_skb(skb, udp_recvmsg); 1256 if (!peeked) { 1257 atomic_inc(&sk->sk_drops); 1258 UDP_INC_STATS_USER(sock_net(sk), 1259 UDP_MIB_INERRORS, is_udplite); 1260 } 1261 goto out_free; 1262 } 1263 1264 if (!peeked) 1265 UDP_INC_STATS_USER(sock_net(sk), 1266 UDP_MIB_INDATAGRAMS, is_udplite); 1267 1268 sock_recv_ts_and_drops(msg, sk, skb); 1269 1270 /* Copy the address. */ 1271 if (sin) { 1272 sin->sin_family = AF_INET; 1273 sin->sin_port = udp_hdr(skb)->source; 1274 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1275 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1276 } 1277 if (inet->cmsg_flags) 1278 ip_cmsg_recv(msg, skb); 1279 1280 err = copied; 1281 if (flags & MSG_TRUNC) 1282 err = ulen; 1283 1284 out_free: 1285 skb_free_datagram_locked(sk, skb); 1286 out: 1287 return err; 1288 1289 csum_copy_err: 1290 slow = lock_sock_fast(sk); 1291 if (!skb_kill_datagram(sk, skb, flags)) { 1292 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1293 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1294 } 1295 unlock_sock_fast(sk, slow); 1296 1297 if (noblock) 1298 return -EAGAIN; 1299 1300 /* starting over for a new packet */ 1301 msg->msg_flags &= ~MSG_TRUNC; 1302 goto try_again; 1303 } 1304 1305 1306 int udp_disconnect(struct sock *sk, int flags) 1307 { 1308 struct inet_sock *inet = inet_sk(sk); 1309 /* 1310 * 1003.1g - break association. 1311 */ 1312 1313 sk->sk_state = TCP_CLOSE; 1314 inet->inet_daddr = 0; 1315 inet->inet_dport = 0; 1316 sock_rps_reset_rxhash(sk); 1317 sk->sk_bound_dev_if = 0; 1318 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1319 inet_reset_saddr(sk); 1320 1321 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1322 sk->sk_prot->unhash(sk); 1323 inet->inet_sport = 0; 1324 } 1325 sk_dst_reset(sk); 1326 return 0; 1327 } 1328 EXPORT_SYMBOL(udp_disconnect); 1329 1330 void udp_lib_unhash(struct sock *sk) 1331 { 1332 if (sk_hashed(sk)) { 1333 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1334 struct udp_hslot *hslot, *hslot2; 1335 1336 hslot = udp_hashslot(udptable, sock_net(sk), 1337 udp_sk(sk)->udp_port_hash); 1338 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1339 1340 spin_lock_bh(&hslot->lock); 1341 if (sk_nulls_del_node_init_rcu(sk)) { 1342 hslot->count--; 1343 inet_sk(sk)->inet_num = 0; 1344 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1345 1346 spin_lock(&hslot2->lock); 1347 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1348 hslot2->count--; 1349 spin_unlock(&hslot2->lock); 1350 } 1351 spin_unlock_bh(&hslot->lock); 1352 } 1353 } 1354 EXPORT_SYMBOL(udp_lib_unhash); 1355 1356 /* 1357 * inet_rcv_saddr was changed, we must rehash secondary hash 1358 */ 1359 void udp_lib_rehash(struct sock *sk, u16 newhash) 1360 { 1361 if (sk_hashed(sk)) { 1362 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1363 struct udp_hslot *hslot, *hslot2, *nhslot2; 1364 1365 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1366 nhslot2 = udp_hashslot2(udptable, newhash); 1367 udp_sk(sk)->udp_portaddr_hash = newhash; 1368 if (hslot2 != nhslot2) { 1369 hslot = udp_hashslot(udptable, sock_net(sk), 1370 udp_sk(sk)->udp_port_hash); 1371 /* we must lock primary chain too */ 1372 spin_lock_bh(&hslot->lock); 1373 1374 spin_lock(&hslot2->lock); 1375 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1376 hslot2->count--; 1377 spin_unlock(&hslot2->lock); 1378 1379 spin_lock(&nhslot2->lock); 1380 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1381 &nhslot2->head); 1382 nhslot2->count++; 1383 spin_unlock(&nhslot2->lock); 1384 1385 spin_unlock_bh(&hslot->lock); 1386 } 1387 } 1388 } 1389 EXPORT_SYMBOL(udp_lib_rehash); 1390 1391 static void udp_v4_rehash(struct sock *sk) 1392 { 1393 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1394 inet_sk(sk)->inet_rcv_saddr, 1395 inet_sk(sk)->inet_num); 1396 udp_lib_rehash(sk, new_hash); 1397 } 1398 1399 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1400 { 1401 int rc; 1402 1403 if (inet_sk(sk)->inet_daddr) 1404 sock_rps_save_rxhash(sk, skb); 1405 1406 rc = sock_queue_rcv_skb(sk, skb); 1407 if (rc < 0) { 1408 int is_udplite = IS_UDPLITE(sk); 1409 1410 /* Note that an ENOMEM error is charged twice */ 1411 if (rc == -ENOMEM) 1412 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1413 is_udplite); 1414 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1415 kfree_skb(skb); 1416 trace_udp_fail_queue_rcv_skb(rc, sk); 1417 return -1; 1418 } 1419 1420 return 0; 1421 1422 } 1423 1424 static struct static_key udp_encap_needed __read_mostly; 1425 void udp_encap_enable(void) 1426 { 1427 if (!static_key_enabled(&udp_encap_needed)) 1428 static_key_slow_inc(&udp_encap_needed); 1429 } 1430 EXPORT_SYMBOL(udp_encap_enable); 1431 1432 /* returns: 1433 * -1: error 1434 * 0: success 1435 * >0: "udp encap" protocol resubmission 1436 * 1437 * Note that in the success and error cases, the skb is assumed to 1438 * have either been requeued or freed. 1439 */ 1440 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1441 { 1442 struct udp_sock *up = udp_sk(sk); 1443 int rc; 1444 int is_udplite = IS_UDPLITE(sk); 1445 1446 /* 1447 * Charge it to the socket, dropping if the queue is full. 1448 */ 1449 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1450 goto drop; 1451 nf_reset(skb); 1452 1453 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1454 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 1455 1456 /* 1457 * This is an encapsulation socket so pass the skb to 1458 * the socket's udp_encap_rcv() hook. Otherwise, just 1459 * fall through and pass this up the UDP socket. 1460 * up->encap_rcv() returns the following value: 1461 * =0 if skb was successfully passed to the encap 1462 * handler or was discarded by it. 1463 * >0 if skb should be passed on to UDP. 1464 * <0 if skb should be resubmitted as proto -N 1465 */ 1466 1467 /* if we're overly short, let UDP handle it */ 1468 encap_rcv = ACCESS_ONCE(up->encap_rcv); 1469 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { 1470 int ret; 1471 1472 ret = encap_rcv(sk, skb); 1473 if (ret <= 0) { 1474 UDP_INC_STATS_BH(sock_net(sk), 1475 UDP_MIB_INDATAGRAMS, 1476 is_udplite); 1477 return -ret; 1478 } 1479 } 1480 1481 /* FALLTHROUGH -- it's a UDP Packet */ 1482 } 1483 1484 /* 1485 * UDP-Lite specific tests, ignored on UDP sockets 1486 */ 1487 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1488 1489 /* 1490 * MIB statistics other than incrementing the error count are 1491 * disabled for the following two types of errors: these depend 1492 * on the application settings, not on the functioning of the 1493 * protocol stack as such. 1494 * 1495 * RFC 3828 here recommends (sec 3.3): "There should also be a 1496 * way ... to ... at least let the receiving application block 1497 * delivery of packets with coverage values less than a value 1498 * provided by the application." 1499 */ 1500 if (up->pcrlen == 0) { /* full coverage was set */ 1501 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n", 1502 UDP_SKB_CB(skb)->cscov, skb->len); 1503 goto drop; 1504 } 1505 /* The next case involves violating the min. coverage requested 1506 * by the receiver. This is subtle: if receiver wants x and x is 1507 * greater than the buffersize/MTU then receiver will complain 1508 * that it wants x while sender emits packets of smaller size y. 1509 * Therefore the above ...()->partial_cov statement is essential. 1510 */ 1511 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1512 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n", 1513 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1514 goto drop; 1515 } 1516 } 1517 1518 if (rcu_access_pointer(sk->sk_filter) && 1519 udp_lib_checksum_complete(skb)) 1520 goto csum_error; 1521 1522 1523 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) 1524 goto drop; 1525 1526 rc = 0; 1527 1528 ipv4_pktinfo_prepare(skb); 1529 bh_lock_sock(sk); 1530 if (!sock_owned_by_user(sk)) 1531 rc = __udp_queue_rcv_skb(sk, skb); 1532 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 1533 bh_unlock_sock(sk); 1534 goto drop; 1535 } 1536 bh_unlock_sock(sk); 1537 1538 return rc; 1539 1540 csum_error: 1541 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 1542 drop: 1543 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1544 atomic_inc(&sk->sk_drops); 1545 kfree_skb(skb); 1546 return -1; 1547 } 1548 1549 1550 static void flush_stack(struct sock **stack, unsigned int count, 1551 struct sk_buff *skb, unsigned int final) 1552 { 1553 unsigned int i; 1554 struct sk_buff *skb1 = NULL; 1555 struct sock *sk; 1556 1557 for (i = 0; i < count; i++) { 1558 sk = stack[i]; 1559 if (likely(skb1 == NULL)) 1560 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1561 1562 if (!skb1) { 1563 atomic_inc(&sk->sk_drops); 1564 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1565 IS_UDPLITE(sk)); 1566 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1567 IS_UDPLITE(sk)); 1568 } 1569 1570 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1571 skb1 = NULL; 1572 } 1573 if (unlikely(skb1)) 1574 kfree_skb(skb1); 1575 } 1576 1577 /* 1578 * Multicasts and broadcasts go to each listener. 1579 * 1580 * Note: called only from the BH handler context. 1581 */ 1582 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1583 struct udphdr *uh, 1584 __be32 saddr, __be32 daddr, 1585 struct udp_table *udptable) 1586 { 1587 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1588 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1589 int dif; 1590 unsigned int i, count = 0; 1591 1592 spin_lock(&hslot->lock); 1593 sk = sk_nulls_head(&hslot->head); 1594 dif = skb->dev->ifindex; 1595 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1596 while (sk) { 1597 stack[count++] = sk; 1598 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1599 daddr, uh->source, saddr, dif); 1600 if (unlikely(count == ARRAY_SIZE(stack))) { 1601 if (!sk) 1602 break; 1603 flush_stack(stack, count, skb, ~0); 1604 count = 0; 1605 } 1606 } 1607 /* 1608 * before releasing chain lock, we must take a reference on sockets 1609 */ 1610 for (i = 0; i < count; i++) 1611 sock_hold(stack[i]); 1612 1613 spin_unlock(&hslot->lock); 1614 1615 /* 1616 * do the slow work with no lock held 1617 */ 1618 if (count) { 1619 flush_stack(stack, count, skb, count - 1); 1620 1621 for (i = 0; i < count; i++) 1622 sock_put(stack[i]); 1623 } else { 1624 kfree_skb(skb); 1625 } 1626 return 0; 1627 } 1628 1629 /* Initialize UDP checksum. If exited with zero value (success), 1630 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1631 * Otherwise, csum completion requires chacksumming packet body, 1632 * including udp header and folding it to skb->csum. 1633 */ 1634 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1635 int proto) 1636 { 1637 const struct iphdr *iph; 1638 int err; 1639 1640 UDP_SKB_CB(skb)->partial_cov = 0; 1641 UDP_SKB_CB(skb)->cscov = skb->len; 1642 1643 if (proto == IPPROTO_UDPLITE) { 1644 err = udplite_checksum_init(skb, uh); 1645 if (err) 1646 return err; 1647 } 1648 1649 iph = ip_hdr(skb); 1650 if (uh->check == 0) { 1651 skb->ip_summed = CHECKSUM_UNNECESSARY; 1652 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1653 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1654 proto, skb->csum)) 1655 skb->ip_summed = CHECKSUM_UNNECESSARY; 1656 } 1657 if (!skb_csum_unnecessary(skb)) 1658 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1659 skb->len, proto, 0); 1660 /* Probably, we should checksum udp header (it should be in cache 1661 * in any case) and data in tiny packets (< rx copybreak). 1662 */ 1663 1664 return 0; 1665 } 1666 1667 /* 1668 * All we need to do is get the socket, and then do a checksum. 1669 */ 1670 1671 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1672 int proto) 1673 { 1674 struct sock *sk; 1675 struct udphdr *uh; 1676 unsigned short ulen; 1677 struct rtable *rt = skb_rtable(skb); 1678 __be32 saddr, daddr; 1679 struct net *net = dev_net(skb->dev); 1680 1681 /* 1682 * Validate the packet. 1683 */ 1684 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1685 goto drop; /* No space for header. */ 1686 1687 uh = udp_hdr(skb); 1688 ulen = ntohs(uh->len); 1689 saddr = ip_hdr(skb)->saddr; 1690 daddr = ip_hdr(skb)->daddr; 1691 1692 if (ulen > skb->len) 1693 goto short_packet; 1694 1695 if (proto == IPPROTO_UDP) { 1696 /* UDP validates ulen. */ 1697 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1698 goto short_packet; 1699 uh = udp_hdr(skb); 1700 } 1701 1702 if (udp4_csum_init(skb, uh, proto)) 1703 goto csum_error; 1704 1705 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1706 return __udp4_lib_mcast_deliver(net, skb, uh, 1707 saddr, daddr, udptable); 1708 1709 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1710 1711 if (sk != NULL) { 1712 int ret = udp_queue_rcv_skb(sk, skb); 1713 sock_put(sk); 1714 1715 /* a return value > 0 means to resubmit the input, but 1716 * it wants the return to be -protocol, or 0 1717 */ 1718 if (ret > 0) 1719 return -ret; 1720 return 0; 1721 } 1722 1723 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1724 goto drop; 1725 nf_reset(skb); 1726 1727 /* No socket. Drop packet silently, if checksum is wrong */ 1728 if (udp_lib_checksum_complete(skb)) 1729 goto csum_error; 1730 1731 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1732 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1733 1734 /* 1735 * Hmm. We got an UDP packet to a port to which we 1736 * don't wanna listen. Ignore it. 1737 */ 1738 kfree_skb(skb); 1739 return 0; 1740 1741 short_packet: 1742 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1743 proto == IPPROTO_UDPLITE ? "Lite" : "", 1744 &saddr, ntohs(uh->source), 1745 ulen, skb->len, 1746 &daddr, ntohs(uh->dest)); 1747 goto drop; 1748 1749 csum_error: 1750 /* 1751 * RFC1122: OK. Discards the bad packet silently (as far as 1752 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1753 */ 1754 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1755 proto == IPPROTO_UDPLITE ? "Lite" : "", 1756 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), 1757 ulen); 1758 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1759 drop: 1760 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1761 kfree_skb(skb); 1762 return 0; 1763 } 1764 1765 int udp_rcv(struct sk_buff *skb) 1766 { 1767 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1768 } 1769 1770 void udp_destroy_sock(struct sock *sk) 1771 { 1772 struct udp_sock *up = udp_sk(sk); 1773 bool slow = lock_sock_fast(sk); 1774 udp_flush_pending_frames(sk); 1775 unlock_sock_fast(sk, slow); 1776 if (static_key_false(&udp_encap_needed) && up->encap_type) { 1777 void (*encap_destroy)(struct sock *sk); 1778 encap_destroy = ACCESS_ONCE(up->encap_destroy); 1779 if (encap_destroy) 1780 encap_destroy(sk); 1781 } 1782 } 1783 1784 /* 1785 * Socket option code for UDP 1786 */ 1787 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1788 char __user *optval, unsigned int optlen, 1789 int (*push_pending_frames)(struct sock *)) 1790 { 1791 struct udp_sock *up = udp_sk(sk); 1792 int val; 1793 int err = 0; 1794 int is_udplite = IS_UDPLITE(sk); 1795 1796 if (optlen < sizeof(int)) 1797 return -EINVAL; 1798 1799 if (get_user(val, (int __user *)optval)) 1800 return -EFAULT; 1801 1802 switch (optname) { 1803 case UDP_CORK: 1804 if (val != 0) { 1805 up->corkflag = 1; 1806 } else { 1807 up->corkflag = 0; 1808 lock_sock(sk); 1809 (*push_pending_frames)(sk); 1810 release_sock(sk); 1811 } 1812 break; 1813 1814 case UDP_ENCAP: 1815 switch (val) { 1816 case 0: 1817 case UDP_ENCAP_ESPINUDP: 1818 case UDP_ENCAP_ESPINUDP_NON_IKE: 1819 up->encap_rcv = xfrm4_udp_encap_rcv; 1820 /* FALLTHROUGH */ 1821 case UDP_ENCAP_L2TPINUDP: 1822 up->encap_type = val; 1823 udp_encap_enable(); 1824 break; 1825 default: 1826 err = -ENOPROTOOPT; 1827 break; 1828 } 1829 break; 1830 1831 /* 1832 * UDP-Lite's partial checksum coverage (RFC 3828). 1833 */ 1834 /* The sender sets actual checksum coverage length via this option. 1835 * The case coverage > packet length is handled by send module. */ 1836 case UDPLITE_SEND_CSCOV: 1837 if (!is_udplite) /* Disable the option on UDP sockets */ 1838 return -ENOPROTOOPT; 1839 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1840 val = 8; 1841 else if (val > USHRT_MAX) 1842 val = USHRT_MAX; 1843 up->pcslen = val; 1844 up->pcflag |= UDPLITE_SEND_CC; 1845 break; 1846 1847 /* The receiver specifies a minimum checksum coverage value. To make 1848 * sense, this should be set to at least 8 (as done below). If zero is 1849 * used, this again means full checksum coverage. */ 1850 case UDPLITE_RECV_CSCOV: 1851 if (!is_udplite) /* Disable the option on UDP sockets */ 1852 return -ENOPROTOOPT; 1853 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1854 val = 8; 1855 else if (val > USHRT_MAX) 1856 val = USHRT_MAX; 1857 up->pcrlen = val; 1858 up->pcflag |= UDPLITE_RECV_CC; 1859 break; 1860 1861 default: 1862 err = -ENOPROTOOPT; 1863 break; 1864 } 1865 1866 return err; 1867 } 1868 EXPORT_SYMBOL(udp_lib_setsockopt); 1869 1870 int udp_setsockopt(struct sock *sk, int level, int optname, 1871 char __user *optval, unsigned int optlen) 1872 { 1873 if (level == SOL_UDP || level == SOL_UDPLITE) 1874 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1875 udp_push_pending_frames); 1876 return ip_setsockopt(sk, level, optname, optval, optlen); 1877 } 1878 1879 #ifdef CONFIG_COMPAT 1880 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1881 char __user *optval, unsigned int optlen) 1882 { 1883 if (level == SOL_UDP || level == SOL_UDPLITE) 1884 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1885 udp_push_pending_frames); 1886 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1887 } 1888 #endif 1889 1890 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1891 char __user *optval, int __user *optlen) 1892 { 1893 struct udp_sock *up = udp_sk(sk); 1894 int val, len; 1895 1896 if (get_user(len, optlen)) 1897 return -EFAULT; 1898 1899 len = min_t(unsigned int, len, sizeof(int)); 1900 1901 if (len < 0) 1902 return -EINVAL; 1903 1904 switch (optname) { 1905 case UDP_CORK: 1906 val = up->corkflag; 1907 break; 1908 1909 case UDP_ENCAP: 1910 val = up->encap_type; 1911 break; 1912 1913 /* The following two cannot be changed on UDP sockets, the return is 1914 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1915 case UDPLITE_SEND_CSCOV: 1916 val = up->pcslen; 1917 break; 1918 1919 case UDPLITE_RECV_CSCOV: 1920 val = up->pcrlen; 1921 break; 1922 1923 default: 1924 return -ENOPROTOOPT; 1925 } 1926 1927 if (put_user(len, optlen)) 1928 return -EFAULT; 1929 if (copy_to_user(optval, &val, len)) 1930 return -EFAULT; 1931 return 0; 1932 } 1933 EXPORT_SYMBOL(udp_lib_getsockopt); 1934 1935 int udp_getsockopt(struct sock *sk, int level, int optname, 1936 char __user *optval, int __user *optlen) 1937 { 1938 if (level == SOL_UDP || level == SOL_UDPLITE) 1939 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1940 return ip_getsockopt(sk, level, optname, optval, optlen); 1941 } 1942 1943 #ifdef CONFIG_COMPAT 1944 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1945 char __user *optval, int __user *optlen) 1946 { 1947 if (level == SOL_UDP || level == SOL_UDPLITE) 1948 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1949 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1950 } 1951 #endif 1952 /** 1953 * udp_poll - wait for a UDP event. 1954 * @file - file struct 1955 * @sock - socket 1956 * @wait - poll table 1957 * 1958 * This is same as datagram poll, except for the special case of 1959 * blocking sockets. If application is using a blocking fd 1960 * and a packet with checksum error is in the queue; 1961 * then it could get return from select indicating data available 1962 * but then block when reading it. Add special case code 1963 * to work around these arguably broken applications. 1964 */ 1965 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1966 { 1967 unsigned int mask = datagram_poll(file, sock, wait); 1968 struct sock *sk = sock->sk; 1969 1970 /* Check for false positives due to checksum errors */ 1971 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 1972 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 1973 mask &= ~(POLLIN | POLLRDNORM); 1974 1975 return mask; 1976 1977 } 1978 EXPORT_SYMBOL(udp_poll); 1979 1980 struct proto udp_prot = { 1981 .name = "UDP", 1982 .owner = THIS_MODULE, 1983 .close = udp_lib_close, 1984 .connect = ip4_datagram_connect, 1985 .disconnect = udp_disconnect, 1986 .ioctl = udp_ioctl, 1987 .destroy = udp_destroy_sock, 1988 .setsockopt = udp_setsockopt, 1989 .getsockopt = udp_getsockopt, 1990 .sendmsg = udp_sendmsg, 1991 .recvmsg = udp_recvmsg, 1992 .sendpage = udp_sendpage, 1993 .backlog_rcv = __udp_queue_rcv_skb, 1994 .release_cb = ip4_datagram_release_cb, 1995 .hash = udp_lib_hash, 1996 .unhash = udp_lib_unhash, 1997 .rehash = udp_v4_rehash, 1998 .get_port = udp_v4_get_port, 1999 .memory_allocated = &udp_memory_allocated, 2000 .sysctl_mem = sysctl_udp_mem, 2001 .sysctl_wmem = &sysctl_udp_wmem_min, 2002 .sysctl_rmem = &sysctl_udp_rmem_min, 2003 .obj_size = sizeof(struct udp_sock), 2004 .slab_flags = SLAB_DESTROY_BY_RCU, 2005 .h.udp_table = &udp_table, 2006 #ifdef CONFIG_COMPAT 2007 .compat_setsockopt = compat_udp_setsockopt, 2008 .compat_getsockopt = compat_udp_getsockopt, 2009 #endif 2010 .clear_sk = sk_prot_clear_portaddr_nulls, 2011 }; 2012 EXPORT_SYMBOL(udp_prot); 2013 2014 /* ------------------------------------------------------------------------ */ 2015 #ifdef CONFIG_PROC_FS 2016 2017 static struct sock *udp_get_first(struct seq_file *seq, int start) 2018 { 2019 struct sock *sk; 2020 struct udp_iter_state *state = seq->private; 2021 struct net *net = seq_file_net(seq); 2022 2023 for (state->bucket = start; state->bucket <= state->udp_table->mask; 2024 ++state->bucket) { 2025 struct hlist_nulls_node *node; 2026 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 2027 2028 if (hlist_nulls_empty(&hslot->head)) 2029 continue; 2030 2031 spin_lock_bh(&hslot->lock); 2032 sk_nulls_for_each(sk, node, &hslot->head) { 2033 if (!net_eq(sock_net(sk), net)) 2034 continue; 2035 if (sk->sk_family == state->family) 2036 goto found; 2037 } 2038 spin_unlock_bh(&hslot->lock); 2039 } 2040 sk = NULL; 2041 found: 2042 return sk; 2043 } 2044 2045 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 2046 { 2047 struct udp_iter_state *state = seq->private; 2048 struct net *net = seq_file_net(seq); 2049 2050 do { 2051 sk = sk_nulls_next(sk); 2052 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 2053 2054 if (!sk) { 2055 if (state->bucket <= state->udp_table->mask) 2056 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2057 return udp_get_first(seq, state->bucket + 1); 2058 } 2059 return sk; 2060 } 2061 2062 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 2063 { 2064 struct sock *sk = udp_get_first(seq, 0); 2065 2066 if (sk) 2067 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2068 --pos; 2069 return pos ? NULL : sk; 2070 } 2071 2072 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2073 { 2074 struct udp_iter_state *state = seq->private; 2075 state->bucket = MAX_UDP_PORTS; 2076 2077 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2078 } 2079 2080 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2081 { 2082 struct sock *sk; 2083 2084 if (v == SEQ_START_TOKEN) 2085 sk = udp_get_idx(seq, 0); 2086 else 2087 sk = udp_get_next(seq, v); 2088 2089 ++*pos; 2090 return sk; 2091 } 2092 2093 static void udp_seq_stop(struct seq_file *seq, void *v) 2094 { 2095 struct udp_iter_state *state = seq->private; 2096 2097 if (state->bucket <= state->udp_table->mask) 2098 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2099 } 2100 2101 int udp_seq_open(struct inode *inode, struct file *file) 2102 { 2103 struct udp_seq_afinfo *afinfo = PDE_DATA(inode); 2104 struct udp_iter_state *s; 2105 int err; 2106 2107 err = seq_open_net(inode, file, &afinfo->seq_ops, 2108 sizeof(struct udp_iter_state)); 2109 if (err < 0) 2110 return err; 2111 2112 s = ((struct seq_file *)file->private_data)->private; 2113 s->family = afinfo->family; 2114 s->udp_table = afinfo->udp_table; 2115 return err; 2116 } 2117 EXPORT_SYMBOL(udp_seq_open); 2118 2119 /* ------------------------------------------------------------------------ */ 2120 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2121 { 2122 struct proc_dir_entry *p; 2123 int rc = 0; 2124 2125 afinfo->seq_ops.start = udp_seq_start; 2126 afinfo->seq_ops.next = udp_seq_next; 2127 afinfo->seq_ops.stop = udp_seq_stop; 2128 2129 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2130 afinfo->seq_fops, afinfo); 2131 if (!p) 2132 rc = -ENOMEM; 2133 return rc; 2134 } 2135 EXPORT_SYMBOL(udp_proc_register); 2136 2137 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2138 { 2139 remove_proc_entry(afinfo->name, net->proc_net); 2140 } 2141 EXPORT_SYMBOL(udp_proc_unregister); 2142 2143 /* ------------------------------------------------------------------------ */ 2144 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2145 int bucket, int *len) 2146 { 2147 struct inet_sock *inet = inet_sk(sp); 2148 __be32 dest = inet->inet_daddr; 2149 __be32 src = inet->inet_rcv_saddr; 2150 __u16 destp = ntohs(inet->inet_dport); 2151 __u16 srcp = ntohs(inet->inet_sport); 2152 2153 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2154 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n", 2155 bucket, src, srcp, dest, destp, sp->sk_state, 2156 sk_wmem_alloc_get(sp), 2157 sk_rmem_alloc_get(sp), 2158 0, 0L, 0, 2159 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), 2160 0, sock_i_ino(sp), 2161 atomic_read(&sp->sk_refcnt), sp, 2162 atomic_read(&sp->sk_drops), len); 2163 } 2164 2165 int udp4_seq_show(struct seq_file *seq, void *v) 2166 { 2167 if (v == SEQ_START_TOKEN) 2168 seq_printf(seq, "%-127s\n", 2169 " sl local_address rem_address st tx_queue " 2170 "rx_queue tr tm->when retrnsmt uid timeout " 2171 "inode ref pointer drops"); 2172 else { 2173 struct udp_iter_state *state = seq->private; 2174 int len; 2175 2176 udp4_format_sock(v, seq, state->bucket, &len); 2177 seq_printf(seq, "%*s\n", 127 - len, ""); 2178 } 2179 return 0; 2180 } 2181 2182 static const struct file_operations udp_afinfo_seq_fops = { 2183 .owner = THIS_MODULE, 2184 .open = udp_seq_open, 2185 .read = seq_read, 2186 .llseek = seq_lseek, 2187 .release = seq_release_net 2188 }; 2189 2190 /* ------------------------------------------------------------------------ */ 2191 static struct udp_seq_afinfo udp4_seq_afinfo = { 2192 .name = "udp", 2193 .family = AF_INET, 2194 .udp_table = &udp_table, 2195 .seq_fops = &udp_afinfo_seq_fops, 2196 .seq_ops = { 2197 .show = udp4_seq_show, 2198 }, 2199 }; 2200 2201 static int __net_init udp4_proc_init_net(struct net *net) 2202 { 2203 return udp_proc_register(net, &udp4_seq_afinfo); 2204 } 2205 2206 static void __net_exit udp4_proc_exit_net(struct net *net) 2207 { 2208 udp_proc_unregister(net, &udp4_seq_afinfo); 2209 } 2210 2211 static struct pernet_operations udp4_net_ops = { 2212 .init = udp4_proc_init_net, 2213 .exit = udp4_proc_exit_net, 2214 }; 2215 2216 int __init udp4_proc_init(void) 2217 { 2218 return register_pernet_subsys(&udp4_net_ops); 2219 } 2220 2221 void udp4_proc_exit(void) 2222 { 2223 unregister_pernet_subsys(&udp4_net_ops); 2224 } 2225 #endif /* CONFIG_PROC_FS */ 2226 2227 static __initdata unsigned long uhash_entries; 2228 static int __init set_uhash_entries(char *str) 2229 { 2230 ssize_t ret; 2231 2232 if (!str) 2233 return 0; 2234 2235 ret = kstrtoul(str, 0, &uhash_entries); 2236 if (ret) 2237 return 0; 2238 2239 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2240 uhash_entries = UDP_HTABLE_SIZE_MIN; 2241 return 1; 2242 } 2243 __setup("uhash_entries=", set_uhash_entries); 2244 2245 void __init udp_table_init(struct udp_table *table, const char *name) 2246 { 2247 unsigned int i; 2248 2249 table->hash = alloc_large_system_hash(name, 2250 2 * sizeof(struct udp_hslot), 2251 uhash_entries, 2252 21, /* one slot per 2 MB */ 2253 0, 2254 &table->log, 2255 &table->mask, 2256 UDP_HTABLE_SIZE_MIN, 2257 64 * 1024); 2258 2259 table->hash2 = table->hash + (table->mask + 1); 2260 for (i = 0; i <= table->mask; i++) { 2261 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2262 table->hash[i].count = 0; 2263 spin_lock_init(&table->hash[i].lock); 2264 } 2265 for (i = 0; i <= table->mask; i++) { 2266 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2267 table->hash2[i].count = 0; 2268 spin_lock_init(&table->hash2[i].lock); 2269 } 2270 } 2271 2272 void __init udp_init(void) 2273 { 2274 unsigned long limit; 2275 2276 udp_table_init(&udp_table, "UDP"); 2277 limit = nr_free_buffer_pages() / 8; 2278 limit = max(limit, 128UL); 2279 sysctl_udp_mem[0] = limit / 4 * 3; 2280 sysctl_udp_mem[1] = limit; 2281 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2282 2283 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2284 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2285 } 2286 2287 int udp4_ufo_send_check(struct sk_buff *skb) 2288 { 2289 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 2290 return -EINVAL; 2291 2292 if (likely(!skb->encapsulation)) { 2293 const struct iphdr *iph; 2294 struct udphdr *uh; 2295 2296 iph = ip_hdr(skb); 2297 uh = udp_hdr(skb); 2298 2299 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 2300 IPPROTO_UDP, 0); 2301 skb->csum_start = skb_transport_header(skb) - skb->head; 2302 skb->csum_offset = offsetof(struct udphdr, check); 2303 skb->ip_summed = CHECKSUM_PARTIAL; 2304 } 2305 return 0; 2306 } 2307 2308 static struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb, 2309 netdev_features_t features) 2310 { 2311 struct sk_buff *segs = ERR_PTR(-EINVAL); 2312 int mac_len = skb->mac_len; 2313 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb); 2314 __be16 protocol = skb->protocol; 2315 netdev_features_t enc_features; 2316 int outer_hlen; 2317 2318 if (unlikely(!pskb_may_pull(skb, tnl_hlen))) 2319 goto out; 2320 2321 skb->encapsulation = 0; 2322 __skb_pull(skb, tnl_hlen); 2323 skb_reset_mac_header(skb); 2324 skb_set_network_header(skb, skb_inner_network_offset(skb)); 2325 skb->mac_len = skb_inner_network_offset(skb); 2326 skb->protocol = htons(ETH_P_TEB); 2327 2328 /* segment inner packet. */ 2329 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb); 2330 segs = skb_mac_gso_segment(skb, enc_features); 2331 if (!segs || IS_ERR(segs)) 2332 goto out; 2333 2334 outer_hlen = skb_tnl_header_len(skb); 2335 skb = segs; 2336 do { 2337 struct udphdr *uh; 2338 int udp_offset = outer_hlen - tnl_hlen; 2339 2340 skb->mac_len = mac_len; 2341 2342 skb_push(skb, outer_hlen); 2343 skb_reset_mac_header(skb); 2344 skb_set_network_header(skb, mac_len); 2345 skb_set_transport_header(skb, udp_offset); 2346 uh = udp_hdr(skb); 2347 uh->len = htons(skb->len - udp_offset); 2348 2349 /* csum segment if tunnel sets skb with csum. */ 2350 if (unlikely(uh->check)) { 2351 struct iphdr *iph = ip_hdr(skb); 2352 2353 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 2354 skb->len - udp_offset, 2355 IPPROTO_UDP, 0); 2356 uh->check = csum_fold(skb_checksum(skb, udp_offset, 2357 skb->len - udp_offset, 0)); 2358 if (uh->check == 0) 2359 uh->check = CSUM_MANGLED_0; 2360 2361 } 2362 skb->ip_summed = CHECKSUM_NONE; 2363 skb->protocol = protocol; 2364 } while ((skb = skb->next)); 2365 out: 2366 return segs; 2367 } 2368 2369 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, 2370 netdev_features_t features) 2371 { 2372 struct sk_buff *segs = ERR_PTR(-EINVAL); 2373 unsigned int mss; 2374 mss = skb_shinfo(skb)->gso_size; 2375 if (unlikely(skb->len <= mss)) 2376 goto out; 2377 2378 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 2379 /* Packet is from an untrusted source, reset gso_segs. */ 2380 int type = skb_shinfo(skb)->gso_type; 2381 2382 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY | 2383 SKB_GSO_UDP_TUNNEL | 2384 SKB_GSO_GRE) || 2385 !(type & (SKB_GSO_UDP)))) 2386 goto out; 2387 2388 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 2389 2390 segs = NULL; 2391 goto out; 2392 } 2393 2394 /* Fragment the skb. IP headers of the fragments are updated in 2395 * inet_gso_segment() 2396 */ 2397 if (skb->encapsulation && skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL) 2398 segs = skb_udp_tunnel_segment(skb, features); 2399 else { 2400 int offset; 2401 __wsum csum; 2402 2403 /* Do software UFO. Complete and fill in the UDP checksum as 2404 * HW cannot do checksum of UDP packets sent as multiple 2405 * IP fragments. 2406 */ 2407 offset = skb_checksum_start_offset(skb); 2408 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2409 offset += skb->csum_offset; 2410 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2411 skb->ip_summed = CHECKSUM_NONE; 2412 2413 segs = skb_segment(skb, features); 2414 } 2415 out: 2416 return segs; 2417 } 2418