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 #include <asm/system.h> 81 #include <asm/uaccess.h> 82 #include <asm/ioctls.h> 83 #include <linux/bootmem.h> 84 #include <linux/highmem.h> 85 #include <linux/swap.h> 86 #include <linux/types.h> 87 #include <linux/fcntl.h> 88 #include <linux/module.h> 89 #include <linux/socket.h> 90 #include <linux/sockios.h> 91 #include <linux/igmp.h> 92 #include <linux/in.h> 93 #include <linux/errno.h> 94 #include <linux/timer.h> 95 #include <linux/mm.h> 96 #include <linux/inet.h> 97 #include <linux/netdevice.h> 98 #include <linux/slab.h> 99 #include <net/tcp_states.h> 100 #include <linux/skbuff.h> 101 #include <linux/proc_fs.h> 102 #include <linux/seq_file.h> 103 #include <net/net_namespace.h> 104 #include <net/icmp.h> 105 #include <net/route.h> 106 #include <net/checksum.h> 107 #include <net/xfrm.h> 108 #include "udp_impl.h" 109 110 struct udp_table udp_table __read_mostly; 111 EXPORT_SYMBOL(udp_table); 112 113 long sysctl_udp_mem[3] __read_mostly; 114 EXPORT_SYMBOL(sysctl_udp_mem); 115 116 int sysctl_udp_rmem_min __read_mostly; 117 EXPORT_SYMBOL(sysctl_udp_rmem_min); 118 119 int sysctl_udp_wmem_min __read_mostly; 120 EXPORT_SYMBOL(sysctl_udp_wmem_min); 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 static int udp_lib_lport_inuse(struct net *net, __u16 num, 129 const struct udp_hslot *hslot, 130 unsigned long *bitmap, 131 struct sock *sk, 132 int (*saddr_comp)(const struct sock *sk1, 133 const struct sock *sk2), 134 unsigned int log) 135 { 136 struct sock *sk2; 137 struct hlist_nulls_node *node; 138 139 sk_nulls_for_each(sk2, node, &hslot->head) 140 if (net_eq(sock_net(sk2), net) && 141 sk2 != sk && 142 (bitmap || udp_sk(sk2)->udp_port_hash == num) && 143 (!sk2->sk_reuse || !sk->sk_reuse) && 144 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 145 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 146 (*saddr_comp)(sk, sk2)) { 147 if (bitmap) 148 __set_bit(udp_sk(sk2)->udp_port_hash >> log, 149 bitmap); 150 else 151 return 1; 152 } 153 return 0; 154 } 155 156 /* 157 * Note: we still hold spinlock of primary hash chain, so no other writer 158 * can insert/delete a socket with local_port == num 159 */ 160 static int udp_lib_lport_inuse2(struct net *net, __u16 num, 161 struct udp_hslot *hslot2, 162 struct sock *sk, 163 int (*saddr_comp)(const struct sock *sk1, 164 const struct sock *sk2)) 165 { 166 struct sock *sk2; 167 struct hlist_nulls_node *node; 168 int res = 0; 169 170 spin_lock(&hslot2->lock); 171 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) 172 if (net_eq(sock_net(sk2), net) && 173 sk2 != sk && 174 (udp_sk(sk2)->udp_port_hash == num) && 175 (!sk2->sk_reuse || !sk->sk_reuse) && 176 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 177 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 178 (*saddr_comp)(sk, sk2)) { 179 res = 1; 180 break; 181 } 182 spin_unlock(&hslot2->lock); 183 return res; 184 } 185 186 /** 187 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 188 * 189 * @sk: socket struct in question 190 * @snum: port number to look up 191 * @saddr_comp: AF-dependent comparison of bound local IP addresses 192 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, 193 * with NULL address 194 */ 195 int udp_lib_get_port(struct sock *sk, unsigned short snum, 196 int (*saddr_comp)(const struct sock *sk1, 197 const struct sock *sk2), 198 unsigned int hash2_nulladdr) 199 { 200 struct udp_hslot *hslot, *hslot2; 201 struct udp_table *udptable = sk->sk_prot->h.udp_table; 202 int error = 1; 203 struct net *net = sock_net(sk); 204 205 if (!snum) { 206 int low, high, remaining; 207 unsigned rand; 208 unsigned short first, last; 209 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); 210 211 inet_get_local_port_range(&low, &high); 212 remaining = (high - low) + 1; 213 214 rand = net_random(); 215 first = (((u64)rand * remaining) >> 32) + low; 216 /* 217 * force rand to be an odd multiple of UDP_HTABLE_SIZE 218 */ 219 rand = (rand | 1) * (udptable->mask + 1); 220 last = first + udptable->mask + 1; 221 do { 222 hslot = udp_hashslot(udptable, net, first); 223 bitmap_zero(bitmap, PORTS_PER_CHAIN); 224 spin_lock_bh(&hslot->lock); 225 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, 226 saddr_comp, udptable->log); 227 228 snum = first; 229 /* 230 * Iterate on all possible values of snum for this hash. 231 * Using steps of an odd multiple of UDP_HTABLE_SIZE 232 * give us randomization and full range coverage. 233 */ 234 do { 235 if (low <= snum && snum <= high && 236 !test_bit(snum >> udptable->log, bitmap) && 237 !inet_is_reserved_local_port(snum)) 238 goto found; 239 snum += rand; 240 } while (snum != first); 241 spin_unlock_bh(&hslot->lock); 242 } while (++first != last); 243 goto fail; 244 } else { 245 hslot = udp_hashslot(udptable, net, snum); 246 spin_lock_bh(&hslot->lock); 247 if (hslot->count > 10) { 248 int exist; 249 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; 250 251 slot2 &= udptable->mask; 252 hash2_nulladdr &= udptable->mask; 253 254 hslot2 = udp_hashslot2(udptable, slot2); 255 if (hslot->count < hslot2->count) 256 goto scan_primary_hash; 257 258 exist = udp_lib_lport_inuse2(net, snum, hslot2, 259 sk, saddr_comp); 260 if (!exist && (hash2_nulladdr != slot2)) { 261 hslot2 = udp_hashslot2(udptable, hash2_nulladdr); 262 exist = udp_lib_lport_inuse2(net, snum, hslot2, 263 sk, saddr_comp); 264 } 265 if (exist) 266 goto fail_unlock; 267 else 268 goto found; 269 } 270 scan_primary_hash: 271 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 272 saddr_comp, 0)) 273 goto fail_unlock; 274 } 275 found: 276 inet_sk(sk)->inet_num = snum; 277 udp_sk(sk)->udp_port_hash = snum; 278 udp_sk(sk)->udp_portaddr_hash ^= snum; 279 if (sk_unhashed(sk)) { 280 sk_nulls_add_node_rcu(sk, &hslot->head); 281 hslot->count++; 282 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 283 284 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 285 spin_lock(&hslot2->lock); 286 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 287 &hslot2->head); 288 hslot2->count++; 289 spin_unlock(&hslot2->lock); 290 } 291 error = 0; 292 fail_unlock: 293 spin_unlock_bh(&hslot->lock); 294 fail: 295 return error; 296 } 297 EXPORT_SYMBOL(udp_lib_get_port); 298 299 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 300 { 301 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 302 303 return (!ipv6_only_sock(sk2) && 304 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr || 305 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)); 306 } 307 308 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr, 309 unsigned int port) 310 { 311 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port; 312 } 313 314 int udp_v4_get_port(struct sock *sk, unsigned short snum) 315 { 316 unsigned int hash2_nulladdr = 317 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); 318 unsigned int hash2_partial = 319 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); 320 321 /* precompute partial secondary hash */ 322 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 323 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr); 324 } 325 326 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr, 327 unsigned short hnum, 328 __be16 sport, __be32 daddr, __be16 dport, int dif) 329 { 330 int score = -1; 331 332 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && 333 !ipv6_only_sock(sk)) { 334 struct inet_sock *inet = inet_sk(sk); 335 336 score = (sk->sk_family == PF_INET ? 1 : 0); 337 if (inet->inet_rcv_saddr) { 338 if (inet->inet_rcv_saddr != daddr) 339 return -1; 340 score += 2; 341 } 342 if (inet->inet_daddr) { 343 if (inet->inet_daddr != saddr) 344 return -1; 345 score += 2; 346 } 347 if (inet->inet_dport) { 348 if (inet->inet_dport != sport) 349 return -1; 350 score += 2; 351 } 352 if (sk->sk_bound_dev_if) { 353 if (sk->sk_bound_dev_if != dif) 354 return -1; 355 score += 2; 356 } 357 } 358 return score; 359 } 360 361 /* 362 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num) 363 */ 364 #define SCORE2_MAX (1 + 2 + 2 + 2) 365 static inline int compute_score2(struct sock *sk, struct net *net, 366 __be32 saddr, __be16 sport, 367 __be32 daddr, unsigned int hnum, int dif) 368 { 369 int score = -1; 370 371 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) { 372 struct inet_sock *inet = inet_sk(sk); 373 374 if (inet->inet_rcv_saddr != daddr) 375 return -1; 376 if (inet->inet_num != hnum) 377 return -1; 378 379 score = (sk->sk_family == PF_INET ? 1 : 0); 380 if (inet->inet_daddr) { 381 if (inet->inet_daddr != saddr) 382 return -1; 383 score += 2; 384 } 385 if (inet->inet_dport) { 386 if (inet->inet_dport != sport) 387 return -1; 388 score += 2; 389 } 390 if (sk->sk_bound_dev_if) { 391 if (sk->sk_bound_dev_if != dif) 392 return -1; 393 score += 2; 394 } 395 } 396 return score; 397 } 398 399 400 /* called with read_rcu_lock() */ 401 static struct sock *udp4_lib_lookup2(struct net *net, 402 __be32 saddr, __be16 sport, 403 __be32 daddr, unsigned int hnum, int dif, 404 struct udp_hslot *hslot2, unsigned int slot2) 405 { 406 struct sock *sk, *result; 407 struct hlist_nulls_node *node; 408 int score, badness; 409 410 begin: 411 result = NULL; 412 badness = -1; 413 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { 414 score = compute_score2(sk, net, saddr, sport, 415 daddr, hnum, dif); 416 if (score > badness) { 417 result = sk; 418 badness = score; 419 if (score == SCORE2_MAX) 420 goto exact_match; 421 } 422 } 423 /* 424 * if the nulls value we got at the end of this lookup is 425 * not the expected one, we must restart lookup. 426 * We probably met an item that was moved to another chain. 427 */ 428 if (get_nulls_value(node) != slot2) 429 goto begin; 430 431 if (result) { 432 exact_match: 433 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 434 result = NULL; 435 else if (unlikely(compute_score2(result, net, saddr, sport, 436 daddr, hnum, dif) < badness)) { 437 sock_put(result); 438 goto begin; 439 } 440 } 441 return result; 442 } 443 444 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 445 * harder than this. -DaveM 446 */ 447 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 448 __be16 sport, __be32 daddr, __be16 dport, 449 int dif, struct udp_table *udptable) 450 { 451 struct sock *sk, *result; 452 struct hlist_nulls_node *node; 453 unsigned short hnum = ntohs(dport); 454 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); 455 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; 456 int score, badness; 457 458 rcu_read_lock(); 459 if (hslot->count > 10) { 460 hash2 = udp4_portaddr_hash(net, daddr, hnum); 461 slot2 = hash2 & udptable->mask; 462 hslot2 = &udptable->hash2[slot2]; 463 if (hslot->count < hslot2->count) 464 goto begin; 465 466 result = udp4_lib_lookup2(net, saddr, sport, 467 daddr, hnum, dif, 468 hslot2, slot2); 469 if (!result) { 470 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum); 471 slot2 = hash2 & udptable->mask; 472 hslot2 = &udptable->hash2[slot2]; 473 if (hslot->count < hslot2->count) 474 goto begin; 475 476 result = udp4_lib_lookup2(net, saddr, sport, 477 htonl(INADDR_ANY), hnum, dif, 478 hslot2, slot2); 479 } 480 rcu_read_unlock(); 481 return result; 482 } 483 begin: 484 result = NULL; 485 badness = -1; 486 sk_nulls_for_each_rcu(sk, node, &hslot->head) { 487 score = compute_score(sk, net, saddr, hnum, sport, 488 daddr, dport, dif); 489 if (score > badness) { 490 result = sk; 491 badness = score; 492 } 493 } 494 /* 495 * if the nulls value we got at the end of this lookup is 496 * not the expected one, we must restart lookup. 497 * We probably met an item that was moved to another chain. 498 */ 499 if (get_nulls_value(node) != slot) 500 goto begin; 501 502 if (result) { 503 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) 504 result = NULL; 505 else if (unlikely(compute_score(result, net, saddr, hnum, sport, 506 daddr, dport, dif) < badness)) { 507 sock_put(result); 508 goto begin; 509 } 510 } 511 rcu_read_unlock(); 512 return result; 513 } 514 515 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 516 __be16 sport, __be16 dport, 517 struct udp_table *udptable) 518 { 519 struct sock *sk; 520 const struct iphdr *iph = ip_hdr(skb); 521 522 if (unlikely(sk = skb_steal_sock(skb))) 523 return sk; 524 else 525 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport, 526 iph->daddr, dport, inet_iif(skb), 527 udptable); 528 } 529 530 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 531 __be32 daddr, __be16 dport, int dif) 532 { 533 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); 534 } 535 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 536 537 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 538 __be16 loc_port, __be32 loc_addr, 539 __be16 rmt_port, __be32 rmt_addr, 540 int dif) 541 { 542 struct hlist_nulls_node *node; 543 struct sock *s = sk; 544 unsigned short hnum = ntohs(loc_port); 545 546 sk_nulls_for_each_from(s, node) { 547 struct inet_sock *inet = inet_sk(s); 548 549 if (!net_eq(sock_net(s), net) || 550 udp_sk(s)->udp_port_hash != hnum || 551 (inet->inet_daddr && inet->inet_daddr != rmt_addr) || 552 (inet->inet_dport != rmt_port && inet->inet_dport) || 553 (inet->inet_rcv_saddr && 554 inet->inet_rcv_saddr != loc_addr) || 555 ipv6_only_sock(s) || 556 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 557 continue; 558 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 559 continue; 560 goto found; 561 } 562 s = NULL; 563 found: 564 return s; 565 } 566 567 /* 568 * This routine is called by the ICMP module when it gets some 569 * sort of error condition. If err < 0 then the socket should 570 * be closed and the error returned to the user. If err > 0 571 * it's just the icmp type << 8 | icmp code. 572 * Header points to the ip header of the error packet. We move 573 * on past this. Then (as it used to claim before adjustment) 574 * header points to the first 8 bytes of the udp header. We need 575 * to find the appropriate port. 576 */ 577 578 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) 579 { 580 struct inet_sock *inet; 581 const struct iphdr *iph = (const struct iphdr *)skb->data; 582 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); 583 const int type = icmp_hdr(skb)->type; 584 const int code = icmp_hdr(skb)->code; 585 struct sock *sk; 586 int harderr; 587 int err; 588 struct net *net = dev_net(skb->dev); 589 590 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 591 iph->saddr, uh->source, skb->dev->ifindex, udptable); 592 if (sk == NULL) { 593 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 594 return; /* No socket for error */ 595 } 596 597 err = 0; 598 harderr = 0; 599 inet = inet_sk(sk); 600 601 switch (type) { 602 default: 603 case ICMP_TIME_EXCEEDED: 604 err = EHOSTUNREACH; 605 break; 606 case ICMP_SOURCE_QUENCH: 607 goto out; 608 case ICMP_PARAMETERPROB: 609 err = EPROTO; 610 harderr = 1; 611 break; 612 case ICMP_DEST_UNREACH: 613 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 614 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 615 err = EMSGSIZE; 616 harderr = 1; 617 break; 618 } 619 goto out; 620 } 621 err = EHOSTUNREACH; 622 if (code <= NR_ICMP_UNREACH) { 623 harderr = icmp_err_convert[code].fatal; 624 err = icmp_err_convert[code].errno; 625 } 626 break; 627 } 628 629 /* 630 * RFC1122: OK. Passes ICMP errors back to application, as per 631 * 4.1.3.3. 632 */ 633 if (!inet->recverr) { 634 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 635 goto out; 636 } else 637 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); 638 639 sk->sk_err = err; 640 sk->sk_error_report(sk); 641 out: 642 sock_put(sk); 643 } 644 645 void udp_err(struct sk_buff *skb, u32 info) 646 { 647 __udp4_lib_err(skb, info, &udp_table); 648 } 649 650 /* 651 * Throw away all pending data and cancel the corking. Socket is locked. 652 */ 653 void udp_flush_pending_frames(struct sock *sk) 654 { 655 struct udp_sock *up = udp_sk(sk); 656 657 if (up->pending) { 658 up->len = 0; 659 up->pending = 0; 660 ip_flush_pending_frames(sk); 661 } 662 } 663 EXPORT_SYMBOL(udp_flush_pending_frames); 664 665 /** 666 * udp4_hwcsum - handle outgoing HW checksumming 667 * @skb: sk_buff containing the filled-in UDP header 668 * (checksum field must be zeroed out) 669 * @src: source IP address 670 * @dst: destination IP address 671 */ 672 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) 673 { 674 struct udphdr *uh = udp_hdr(skb); 675 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 676 int offset = skb_transport_offset(skb); 677 int len = skb->len - offset; 678 int hlen = len; 679 __wsum csum = 0; 680 681 if (!frags) { 682 /* 683 * Only one fragment on the socket. 684 */ 685 skb->csum_start = skb_transport_header(skb) - skb->head; 686 skb->csum_offset = offsetof(struct udphdr, check); 687 uh->check = ~csum_tcpudp_magic(src, dst, len, 688 IPPROTO_UDP, 0); 689 } else { 690 /* 691 * HW-checksum won't work as there are two or more 692 * fragments on the socket so that all csums of sk_buffs 693 * should be together 694 */ 695 do { 696 csum = csum_add(csum, frags->csum); 697 hlen -= frags->len; 698 } while ((frags = frags->next)); 699 700 csum = skb_checksum(skb, offset, hlen, csum); 701 skb->ip_summed = CHECKSUM_NONE; 702 703 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 704 if (uh->check == 0) 705 uh->check = CSUM_MANGLED_0; 706 } 707 } 708 709 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4) 710 { 711 struct sock *sk = skb->sk; 712 struct inet_sock *inet = inet_sk(sk); 713 struct udphdr *uh; 714 int err = 0; 715 int is_udplite = IS_UDPLITE(sk); 716 int offset = skb_transport_offset(skb); 717 int len = skb->len - offset; 718 __wsum csum = 0; 719 720 /* 721 * Create a UDP header 722 */ 723 uh = udp_hdr(skb); 724 uh->source = inet->inet_sport; 725 uh->dest = fl4->fl4_dport; 726 uh->len = htons(len); 727 uh->check = 0; 728 729 if (is_udplite) /* UDP-Lite */ 730 csum = udplite_csum(skb); 731 732 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 733 734 skb->ip_summed = CHECKSUM_NONE; 735 goto send; 736 737 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 738 739 udp4_hwcsum(skb, fl4->saddr, fl4->daddr); 740 goto send; 741 742 } else 743 csum = udp_csum(skb); 744 745 /* add protocol-dependent pseudo-header */ 746 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, 747 sk->sk_protocol, csum); 748 if (uh->check == 0) 749 uh->check = CSUM_MANGLED_0; 750 751 send: 752 err = ip_send_skb(skb); 753 if (err) { 754 if (err == -ENOBUFS && !inet->recverr) { 755 UDP_INC_STATS_USER(sock_net(sk), 756 UDP_MIB_SNDBUFERRORS, is_udplite); 757 err = 0; 758 } 759 } else 760 UDP_INC_STATS_USER(sock_net(sk), 761 UDP_MIB_OUTDATAGRAMS, is_udplite); 762 return err; 763 } 764 765 /* 766 * Push out all pending data as one UDP datagram. Socket is locked. 767 */ 768 static int udp_push_pending_frames(struct sock *sk) 769 { 770 struct udp_sock *up = udp_sk(sk); 771 struct inet_sock *inet = inet_sk(sk); 772 struct flowi4 *fl4 = &inet->cork.fl.u.ip4; 773 struct sk_buff *skb; 774 int err = 0; 775 776 skb = ip_finish_skb(sk, fl4); 777 if (!skb) 778 goto out; 779 780 err = udp_send_skb(skb, fl4); 781 782 out: 783 up->len = 0; 784 up->pending = 0; 785 return err; 786 } 787 788 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 789 size_t len) 790 { 791 struct inet_sock *inet = inet_sk(sk); 792 struct udp_sock *up = udp_sk(sk); 793 struct flowi4 fl4_stack; 794 struct flowi4 *fl4; 795 int ulen = len; 796 struct ipcm_cookie ipc; 797 struct rtable *rt = NULL; 798 int free = 0; 799 int connected = 0; 800 __be32 daddr, faddr, saddr; 801 __be16 dport; 802 u8 tos; 803 int err, is_udplite = IS_UDPLITE(sk); 804 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 805 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 806 struct sk_buff *skb; 807 struct ip_options_data opt_copy; 808 809 if (len > 0xFFFF) 810 return -EMSGSIZE; 811 812 /* 813 * Check the flags. 814 */ 815 816 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ 817 return -EOPNOTSUPP; 818 819 ipc.opt = NULL; 820 ipc.tx_flags = 0; 821 822 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 823 824 fl4 = &inet->cork.fl.u.ip4; 825 if (up->pending) { 826 /* 827 * There are pending frames. 828 * The socket lock must be held while it's corked. 829 */ 830 lock_sock(sk); 831 if (likely(up->pending)) { 832 if (unlikely(up->pending != AF_INET)) { 833 release_sock(sk); 834 return -EINVAL; 835 } 836 goto do_append_data; 837 } 838 release_sock(sk); 839 } 840 ulen += sizeof(struct udphdr); 841 842 /* 843 * Get and verify the address. 844 */ 845 if (msg->msg_name) { 846 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name; 847 if (msg->msg_namelen < sizeof(*usin)) 848 return -EINVAL; 849 if (usin->sin_family != AF_INET) { 850 if (usin->sin_family != AF_UNSPEC) 851 return -EAFNOSUPPORT; 852 } 853 854 daddr = usin->sin_addr.s_addr; 855 dport = usin->sin_port; 856 if (dport == 0) 857 return -EINVAL; 858 } else { 859 if (sk->sk_state != TCP_ESTABLISHED) 860 return -EDESTADDRREQ; 861 daddr = inet->inet_daddr; 862 dport = inet->inet_dport; 863 /* Open fast path for connected socket. 864 Route will not be used, if at least one option is set. 865 */ 866 connected = 1; 867 } 868 ipc.addr = inet->inet_saddr; 869 870 ipc.oif = sk->sk_bound_dev_if; 871 err = sock_tx_timestamp(sk, &ipc.tx_flags); 872 if (err) 873 return err; 874 if (msg->msg_controllen) { 875 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 876 if (err) 877 return err; 878 if (ipc.opt) 879 free = 1; 880 connected = 0; 881 } 882 if (!ipc.opt) { 883 struct ip_options_rcu *inet_opt; 884 885 rcu_read_lock(); 886 inet_opt = rcu_dereference(inet->inet_opt); 887 if (inet_opt) { 888 memcpy(&opt_copy, inet_opt, 889 sizeof(*inet_opt) + inet_opt->opt.optlen); 890 ipc.opt = &opt_copy.opt; 891 } 892 rcu_read_unlock(); 893 } 894 895 saddr = ipc.addr; 896 ipc.addr = faddr = daddr; 897 898 if (ipc.opt && ipc.opt->opt.srr) { 899 if (!daddr) 900 return -EINVAL; 901 faddr = ipc.opt->opt.faddr; 902 connected = 0; 903 } 904 tos = RT_TOS(inet->tos); 905 if (sock_flag(sk, SOCK_LOCALROUTE) || 906 (msg->msg_flags & MSG_DONTROUTE) || 907 (ipc.opt && ipc.opt->opt.is_strictroute)) { 908 tos |= RTO_ONLINK; 909 connected = 0; 910 } 911 912 if (ipv4_is_multicast(daddr)) { 913 if (!ipc.oif) 914 ipc.oif = inet->mc_index; 915 if (!saddr) 916 saddr = inet->mc_addr; 917 connected = 0; 918 } 919 920 if (connected) 921 rt = (struct rtable *)sk_dst_check(sk, 0); 922 923 if (rt == NULL) { 924 struct net *net = sock_net(sk); 925 926 fl4 = &fl4_stack; 927 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, 928 RT_SCOPE_UNIVERSE, sk->sk_protocol, 929 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP, 930 faddr, saddr, dport, inet->inet_sport); 931 932 security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); 933 rt = ip_route_output_flow(net, fl4, sk); 934 if (IS_ERR(rt)) { 935 err = PTR_ERR(rt); 936 rt = NULL; 937 if (err == -ENETUNREACH) 938 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 939 goto out; 940 } 941 942 err = -EACCES; 943 if ((rt->rt_flags & RTCF_BROADCAST) && 944 !sock_flag(sk, SOCK_BROADCAST)) 945 goto out; 946 if (connected) 947 sk_dst_set(sk, dst_clone(&rt->dst)); 948 } 949 950 if (msg->msg_flags&MSG_CONFIRM) 951 goto do_confirm; 952 back_from_confirm: 953 954 saddr = fl4->saddr; 955 if (!ipc.addr) 956 daddr = ipc.addr = fl4->daddr; 957 958 /* Lockless fast path for the non-corking case. */ 959 if (!corkreq) { 960 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen, 961 sizeof(struct udphdr), &ipc, &rt, 962 msg->msg_flags); 963 err = PTR_ERR(skb); 964 if (skb && !IS_ERR(skb)) 965 err = udp_send_skb(skb, fl4); 966 goto out; 967 } 968 969 lock_sock(sk); 970 if (unlikely(up->pending)) { 971 /* The socket is already corked while preparing it. */ 972 /* ... which is an evident application bug. --ANK */ 973 release_sock(sk); 974 975 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 976 err = -EINVAL; 977 goto out; 978 } 979 /* 980 * Now cork the socket to pend data. 981 */ 982 fl4 = &inet->cork.fl.u.ip4; 983 fl4->daddr = daddr; 984 fl4->saddr = saddr; 985 fl4->fl4_dport = dport; 986 fl4->fl4_sport = inet->inet_sport; 987 up->pending = AF_INET; 988 989 do_append_data: 990 up->len += ulen; 991 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen, 992 sizeof(struct udphdr), &ipc, &rt, 993 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 994 if (err) 995 udp_flush_pending_frames(sk); 996 else if (!corkreq) 997 err = udp_push_pending_frames(sk); 998 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 999 up->pending = 0; 1000 release_sock(sk); 1001 1002 out: 1003 ip_rt_put(rt); 1004 if (free) 1005 kfree(ipc.opt); 1006 if (!err) 1007 return len; 1008 /* 1009 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1010 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1011 * we don't have a good statistic (IpOutDiscards but it can be too many 1012 * things). We could add another new stat but at least for now that 1013 * seems like overkill. 1014 */ 1015 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1016 UDP_INC_STATS_USER(sock_net(sk), 1017 UDP_MIB_SNDBUFERRORS, is_udplite); 1018 } 1019 return err; 1020 1021 do_confirm: 1022 dst_confirm(&rt->dst); 1023 if (!(msg->msg_flags&MSG_PROBE) || len) 1024 goto back_from_confirm; 1025 err = 0; 1026 goto out; 1027 } 1028 EXPORT_SYMBOL(udp_sendmsg); 1029 1030 int udp_sendpage(struct sock *sk, struct page *page, int offset, 1031 size_t size, int flags) 1032 { 1033 struct inet_sock *inet = inet_sk(sk); 1034 struct udp_sock *up = udp_sk(sk); 1035 int ret; 1036 1037 if (!up->pending) { 1038 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 1039 1040 /* Call udp_sendmsg to specify destination address which 1041 * sendpage interface can't pass. 1042 * This will succeed only when the socket is connected. 1043 */ 1044 ret = udp_sendmsg(NULL, sk, &msg, 0); 1045 if (ret < 0) 1046 return ret; 1047 } 1048 1049 lock_sock(sk); 1050 1051 if (unlikely(!up->pending)) { 1052 release_sock(sk); 1053 1054 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 1055 return -EINVAL; 1056 } 1057 1058 ret = ip_append_page(sk, &inet->cork.fl.u.ip4, 1059 page, offset, size, flags); 1060 if (ret == -EOPNOTSUPP) { 1061 release_sock(sk); 1062 return sock_no_sendpage(sk->sk_socket, page, offset, 1063 size, flags); 1064 } 1065 if (ret < 0) { 1066 udp_flush_pending_frames(sk); 1067 goto out; 1068 } 1069 1070 up->len += size; 1071 if (!(up->corkflag || (flags&MSG_MORE))) 1072 ret = udp_push_pending_frames(sk); 1073 if (!ret) 1074 ret = size; 1075 out: 1076 release_sock(sk); 1077 return ret; 1078 } 1079 1080 1081 /** 1082 * first_packet_length - return length of first packet in receive queue 1083 * @sk: socket 1084 * 1085 * Drops all bad checksum frames, until a valid one is found. 1086 * Returns the length of found skb, or 0 if none is found. 1087 */ 1088 static unsigned int first_packet_length(struct sock *sk) 1089 { 1090 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue; 1091 struct sk_buff *skb; 1092 unsigned int res; 1093 1094 __skb_queue_head_init(&list_kill); 1095 1096 spin_lock_bh(&rcvq->lock); 1097 while ((skb = skb_peek(rcvq)) != NULL && 1098 udp_lib_checksum_complete(skb)) { 1099 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1100 IS_UDPLITE(sk)); 1101 atomic_inc(&sk->sk_drops); 1102 __skb_unlink(skb, rcvq); 1103 __skb_queue_tail(&list_kill, skb); 1104 } 1105 res = skb ? skb->len : 0; 1106 spin_unlock_bh(&rcvq->lock); 1107 1108 if (!skb_queue_empty(&list_kill)) { 1109 bool slow = lock_sock_fast(sk); 1110 1111 __skb_queue_purge(&list_kill); 1112 sk_mem_reclaim_partial(sk); 1113 unlock_sock_fast(sk, slow); 1114 } 1115 return res; 1116 } 1117 1118 /* 1119 * IOCTL requests applicable to the UDP protocol 1120 */ 1121 1122 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 1123 { 1124 switch (cmd) { 1125 case SIOCOUTQ: 1126 { 1127 int amount = sk_wmem_alloc_get(sk); 1128 1129 return put_user(amount, (int __user *)arg); 1130 } 1131 1132 case SIOCINQ: 1133 { 1134 unsigned int amount = first_packet_length(sk); 1135 1136 if (amount) 1137 /* 1138 * We will only return the amount 1139 * of this packet since that is all 1140 * that will be read. 1141 */ 1142 amount -= sizeof(struct udphdr); 1143 1144 return put_user(amount, (int __user *)arg); 1145 } 1146 1147 default: 1148 return -ENOIOCTLCMD; 1149 } 1150 1151 return 0; 1152 } 1153 EXPORT_SYMBOL(udp_ioctl); 1154 1155 /* 1156 * This should be easy, if there is something there we 1157 * return it, otherwise we block. 1158 */ 1159 1160 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 1161 size_t len, int noblock, int flags, int *addr_len) 1162 { 1163 struct inet_sock *inet = inet_sk(sk); 1164 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 1165 struct sk_buff *skb; 1166 unsigned int ulen; 1167 int peeked; 1168 int err; 1169 int is_udplite = IS_UDPLITE(sk); 1170 bool slow; 1171 1172 /* 1173 * Check any passed addresses 1174 */ 1175 if (addr_len) 1176 *addr_len = sizeof(*sin); 1177 1178 if (flags & MSG_ERRQUEUE) 1179 return ip_recv_error(sk, msg, len); 1180 1181 try_again: 1182 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 1183 &peeked, &err); 1184 if (!skb) 1185 goto out; 1186 1187 ulen = skb->len - sizeof(struct udphdr); 1188 if (len > ulen) 1189 len = ulen; 1190 else if (len < ulen) 1191 msg->msg_flags |= MSG_TRUNC; 1192 1193 /* 1194 * If checksum is needed at all, try to do it while copying the 1195 * data. If the data is truncated, or if we only want a partial 1196 * coverage checksum (UDP-Lite), do it before the copy. 1197 */ 1198 1199 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) { 1200 if (udp_lib_checksum_complete(skb)) 1201 goto csum_copy_err; 1202 } 1203 1204 if (skb_csum_unnecessary(skb)) 1205 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 1206 msg->msg_iov, len); 1207 else { 1208 err = skb_copy_and_csum_datagram_iovec(skb, 1209 sizeof(struct udphdr), 1210 msg->msg_iov); 1211 1212 if (err == -EINVAL) 1213 goto csum_copy_err; 1214 } 1215 1216 if (err) 1217 goto out_free; 1218 1219 if (!peeked) 1220 UDP_INC_STATS_USER(sock_net(sk), 1221 UDP_MIB_INDATAGRAMS, is_udplite); 1222 1223 sock_recv_ts_and_drops(msg, sk, skb); 1224 1225 /* Copy the address. */ 1226 if (sin) { 1227 sin->sin_family = AF_INET; 1228 sin->sin_port = udp_hdr(skb)->source; 1229 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 1230 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 1231 } 1232 if (inet->cmsg_flags) 1233 ip_cmsg_recv(msg, skb); 1234 1235 err = len; 1236 if (flags & MSG_TRUNC) 1237 err = ulen; 1238 1239 out_free: 1240 skb_free_datagram_locked(sk, skb); 1241 out: 1242 return err; 1243 1244 csum_copy_err: 1245 slow = lock_sock_fast(sk); 1246 if (!skb_kill_datagram(sk, skb, flags)) 1247 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1248 unlock_sock_fast(sk, slow); 1249 1250 if (noblock) 1251 return -EAGAIN; 1252 goto try_again; 1253 } 1254 1255 1256 int udp_disconnect(struct sock *sk, int flags) 1257 { 1258 struct inet_sock *inet = inet_sk(sk); 1259 /* 1260 * 1003.1g - break association. 1261 */ 1262 1263 sk->sk_state = TCP_CLOSE; 1264 inet->inet_daddr = 0; 1265 inet->inet_dport = 0; 1266 sock_rps_save_rxhash(sk, 0); 1267 sk->sk_bound_dev_if = 0; 1268 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 1269 inet_reset_saddr(sk); 1270 1271 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 1272 sk->sk_prot->unhash(sk); 1273 inet->inet_sport = 0; 1274 } 1275 sk_dst_reset(sk); 1276 return 0; 1277 } 1278 EXPORT_SYMBOL(udp_disconnect); 1279 1280 void udp_lib_unhash(struct sock *sk) 1281 { 1282 if (sk_hashed(sk)) { 1283 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1284 struct udp_hslot *hslot, *hslot2; 1285 1286 hslot = udp_hashslot(udptable, sock_net(sk), 1287 udp_sk(sk)->udp_port_hash); 1288 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1289 1290 spin_lock_bh(&hslot->lock); 1291 if (sk_nulls_del_node_init_rcu(sk)) { 1292 hslot->count--; 1293 inet_sk(sk)->inet_num = 0; 1294 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 1295 1296 spin_lock(&hslot2->lock); 1297 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1298 hslot2->count--; 1299 spin_unlock(&hslot2->lock); 1300 } 1301 spin_unlock_bh(&hslot->lock); 1302 } 1303 } 1304 EXPORT_SYMBOL(udp_lib_unhash); 1305 1306 /* 1307 * inet_rcv_saddr was changed, we must rehash secondary hash 1308 */ 1309 void udp_lib_rehash(struct sock *sk, u16 newhash) 1310 { 1311 if (sk_hashed(sk)) { 1312 struct udp_table *udptable = sk->sk_prot->h.udp_table; 1313 struct udp_hslot *hslot, *hslot2, *nhslot2; 1314 1315 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); 1316 nhslot2 = udp_hashslot2(udptable, newhash); 1317 udp_sk(sk)->udp_portaddr_hash = newhash; 1318 if (hslot2 != nhslot2) { 1319 hslot = udp_hashslot(udptable, sock_net(sk), 1320 udp_sk(sk)->udp_port_hash); 1321 /* we must lock primary chain too */ 1322 spin_lock_bh(&hslot->lock); 1323 1324 spin_lock(&hslot2->lock); 1325 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); 1326 hslot2->count--; 1327 spin_unlock(&hslot2->lock); 1328 1329 spin_lock(&nhslot2->lock); 1330 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, 1331 &nhslot2->head); 1332 nhslot2->count++; 1333 spin_unlock(&nhslot2->lock); 1334 1335 spin_unlock_bh(&hslot->lock); 1336 } 1337 } 1338 } 1339 EXPORT_SYMBOL(udp_lib_rehash); 1340 1341 static void udp_v4_rehash(struct sock *sk) 1342 { 1343 u16 new_hash = udp4_portaddr_hash(sock_net(sk), 1344 inet_sk(sk)->inet_rcv_saddr, 1345 inet_sk(sk)->inet_num); 1346 udp_lib_rehash(sk, new_hash); 1347 } 1348 1349 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1350 { 1351 int rc; 1352 1353 if (inet_sk(sk)->inet_daddr) 1354 sock_rps_save_rxhash(sk, skb->rxhash); 1355 1356 rc = ip_queue_rcv_skb(sk, skb); 1357 if (rc < 0) { 1358 int is_udplite = IS_UDPLITE(sk); 1359 1360 /* Note that an ENOMEM error is charged twice */ 1361 if (rc == -ENOMEM) 1362 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1363 is_udplite); 1364 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1365 kfree_skb(skb); 1366 return -1; 1367 } 1368 1369 return 0; 1370 1371 } 1372 1373 /* returns: 1374 * -1: error 1375 * 0: success 1376 * >0: "udp encap" protocol resubmission 1377 * 1378 * Note that in the success and error cases, the skb is assumed to 1379 * have either been requeued or freed. 1380 */ 1381 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 1382 { 1383 struct udp_sock *up = udp_sk(sk); 1384 int rc; 1385 int is_udplite = IS_UDPLITE(sk); 1386 1387 /* 1388 * Charge it to the socket, dropping if the queue is full. 1389 */ 1390 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1391 goto drop; 1392 nf_reset(skb); 1393 1394 if (up->encap_type) { 1395 /* 1396 * This is an encapsulation socket so pass the skb to 1397 * the socket's udp_encap_rcv() hook. Otherwise, just 1398 * fall through and pass this up the UDP socket. 1399 * up->encap_rcv() returns the following value: 1400 * =0 if skb was successfully passed to the encap 1401 * handler or was discarded by it. 1402 * >0 if skb should be passed on to UDP. 1403 * <0 if skb should be resubmitted as proto -N 1404 */ 1405 1406 /* if we're overly short, let UDP handle it */ 1407 if (skb->len > sizeof(struct udphdr) && 1408 up->encap_rcv != NULL) { 1409 int ret; 1410 1411 ret = (*up->encap_rcv)(sk, skb); 1412 if (ret <= 0) { 1413 UDP_INC_STATS_BH(sock_net(sk), 1414 UDP_MIB_INDATAGRAMS, 1415 is_udplite); 1416 return -ret; 1417 } 1418 } 1419 1420 /* FALLTHROUGH -- it's a UDP Packet */ 1421 } 1422 1423 /* 1424 * UDP-Lite specific tests, ignored on UDP sockets 1425 */ 1426 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1427 1428 /* 1429 * MIB statistics other than incrementing the error count are 1430 * disabled for the following two types of errors: these depend 1431 * on the application settings, not on the functioning of the 1432 * protocol stack as such. 1433 * 1434 * RFC 3828 here recommends (sec 3.3): "There should also be a 1435 * way ... to ... at least let the receiving application block 1436 * delivery of packets with coverage values less than a value 1437 * provided by the application." 1438 */ 1439 if (up->pcrlen == 0) { /* full coverage was set */ 1440 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1441 "%d while full coverage %d requested\n", 1442 UDP_SKB_CB(skb)->cscov, skb->len); 1443 goto drop; 1444 } 1445 /* The next case involves violating the min. coverage requested 1446 * by the receiver. This is subtle: if receiver wants x and x is 1447 * greater than the buffersize/MTU then receiver will complain 1448 * that it wants x while sender emits packets of smaller size y. 1449 * Therefore the above ...()->partial_cov statement is essential. 1450 */ 1451 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1452 LIMIT_NETDEBUG(KERN_WARNING 1453 "UDPLITE: coverage %d too small, need min %d\n", 1454 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1455 goto drop; 1456 } 1457 } 1458 1459 if (rcu_dereference_raw(sk->sk_filter)) { 1460 if (udp_lib_checksum_complete(skb)) 1461 goto drop; 1462 } 1463 1464 1465 if (sk_rcvqueues_full(sk, skb)) 1466 goto drop; 1467 1468 rc = 0; 1469 1470 bh_lock_sock(sk); 1471 if (!sock_owned_by_user(sk)) 1472 rc = __udp_queue_rcv_skb(sk, skb); 1473 else if (sk_add_backlog(sk, skb)) { 1474 bh_unlock_sock(sk); 1475 goto drop; 1476 } 1477 bh_unlock_sock(sk); 1478 1479 return rc; 1480 1481 drop: 1482 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1483 atomic_inc(&sk->sk_drops); 1484 kfree_skb(skb); 1485 return -1; 1486 } 1487 1488 1489 static void flush_stack(struct sock **stack, unsigned int count, 1490 struct sk_buff *skb, unsigned int final) 1491 { 1492 unsigned int i; 1493 struct sk_buff *skb1 = NULL; 1494 struct sock *sk; 1495 1496 for (i = 0; i < count; i++) { 1497 sk = stack[i]; 1498 if (likely(skb1 == NULL)) 1499 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); 1500 1501 if (!skb1) { 1502 atomic_inc(&sk->sk_drops); 1503 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 1504 IS_UDPLITE(sk)); 1505 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, 1506 IS_UDPLITE(sk)); 1507 } 1508 1509 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0) 1510 skb1 = NULL; 1511 } 1512 if (unlikely(skb1)) 1513 kfree_skb(skb1); 1514 } 1515 1516 /* 1517 * Multicasts and broadcasts go to each listener. 1518 * 1519 * Note: called only from the BH handler context. 1520 */ 1521 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1522 struct udphdr *uh, 1523 __be32 saddr, __be32 daddr, 1524 struct udp_table *udptable) 1525 { 1526 struct sock *sk, *stack[256 / sizeof(struct sock *)]; 1527 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); 1528 int dif; 1529 unsigned int i, count = 0; 1530 1531 spin_lock(&hslot->lock); 1532 sk = sk_nulls_head(&hslot->head); 1533 dif = skb->dev->ifindex; 1534 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1535 while (sk) { 1536 stack[count++] = sk; 1537 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest, 1538 daddr, uh->source, saddr, dif); 1539 if (unlikely(count == ARRAY_SIZE(stack))) { 1540 if (!sk) 1541 break; 1542 flush_stack(stack, count, skb, ~0); 1543 count = 0; 1544 } 1545 } 1546 /* 1547 * before releasing chain lock, we must take a reference on sockets 1548 */ 1549 for (i = 0; i < count; i++) 1550 sock_hold(stack[i]); 1551 1552 spin_unlock(&hslot->lock); 1553 1554 /* 1555 * do the slow work with no lock held 1556 */ 1557 if (count) { 1558 flush_stack(stack, count, skb, count - 1); 1559 1560 for (i = 0; i < count; i++) 1561 sock_put(stack[i]); 1562 } else { 1563 kfree_skb(skb); 1564 } 1565 return 0; 1566 } 1567 1568 /* Initialize UDP checksum. If exited with zero value (success), 1569 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1570 * Otherwise, csum completion requires chacksumming packet body, 1571 * including udp header and folding it to skb->csum. 1572 */ 1573 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1574 int proto) 1575 { 1576 const struct iphdr *iph; 1577 int err; 1578 1579 UDP_SKB_CB(skb)->partial_cov = 0; 1580 UDP_SKB_CB(skb)->cscov = skb->len; 1581 1582 if (proto == IPPROTO_UDPLITE) { 1583 err = udplite_checksum_init(skb, uh); 1584 if (err) 1585 return err; 1586 } 1587 1588 iph = ip_hdr(skb); 1589 if (uh->check == 0) { 1590 skb->ip_summed = CHECKSUM_UNNECESSARY; 1591 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1592 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1593 proto, skb->csum)) 1594 skb->ip_summed = CHECKSUM_UNNECESSARY; 1595 } 1596 if (!skb_csum_unnecessary(skb)) 1597 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1598 skb->len, proto, 0); 1599 /* Probably, we should checksum udp header (it should be in cache 1600 * in any case) and data in tiny packets (< rx copybreak). 1601 */ 1602 1603 return 0; 1604 } 1605 1606 /* 1607 * All we need to do is get the socket, and then do a checksum. 1608 */ 1609 1610 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 1611 int proto) 1612 { 1613 struct sock *sk; 1614 struct udphdr *uh; 1615 unsigned short ulen; 1616 struct rtable *rt = skb_rtable(skb); 1617 __be32 saddr, daddr; 1618 struct net *net = dev_net(skb->dev); 1619 1620 /* 1621 * Validate the packet. 1622 */ 1623 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1624 goto drop; /* No space for header. */ 1625 1626 uh = udp_hdr(skb); 1627 ulen = ntohs(uh->len); 1628 saddr = ip_hdr(skb)->saddr; 1629 daddr = ip_hdr(skb)->daddr; 1630 1631 if (ulen > skb->len) 1632 goto short_packet; 1633 1634 if (proto == IPPROTO_UDP) { 1635 /* UDP validates ulen. */ 1636 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1637 goto short_packet; 1638 uh = udp_hdr(skb); 1639 } 1640 1641 if (udp4_csum_init(skb, uh, proto)) 1642 goto csum_error; 1643 1644 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1645 return __udp4_lib_mcast_deliver(net, skb, uh, 1646 saddr, daddr, udptable); 1647 1648 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1649 1650 if (sk != NULL) { 1651 int ret = udp_queue_rcv_skb(sk, skb); 1652 sock_put(sk); 1653 1654 /* a return value > 0 means to resubmit the input, but 1655 * it wants the return to be -protocol, or 0 1656 */ 1657 if (ret > 0) 1658 return -ret; 1659 return 0; 1660 } 1661 1662 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1663 goto drop; 1664 nf_reset(skb); 1665 1666 /* No socket. Drop packet silently, if checksum is wrong */ 1667 if (udp_lib_checksum_complete(skb)) 1668 goto csum_error; 1669 1670 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1671 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1672 1673 /* 1674 * Hmm. We got an UDP packet to a port to which we 1675 * don't wanna listen. Ignore it. 1676 */ 1677 kfree_skb(skb); 1678 return 0; 1679 1680 short_packet: 1681 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", 1682 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1683 &saddr, 1684 ntohs(uh->source), 1685 ulen, 1686 skb->len, 1687 &daddr, 1688 ntohs(uh->dest)); 1689 goto drop; 1690 1691 csum_error: 1692 /* 1693 * RFC1122: OK. Discards the bad packet silently (as far as 1694 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1695 */ 1696 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", 1697 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1698 &saddr, 1699 ntohs(uh->source), 1700 &daddr, 1701 ntohs(uh->dest), 1702 ulen); 1703 drop: 1704 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1705 kfree_skb(skb); 1706 return 0; 1707 } 1708 1709 int udp_rcv(struct sk_buff *skb) 1710 { 1711 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); 1712 } 1713 1714 void udp_destroy_sock(struct sock *sk) 1715 { 1716 bool slow = lock_sock_fast(sk); 1717 udp_flush_pending_frames(sk); 1718 unlock_sock_fast(sk, slow); 1719 } 1720 1721 /* 1722 * Socket option code for UDP 1723 */ 1724 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1725 char __user *optval, unsigned int optlen, 1726 int (*push_pending_frames)(struct sock *)) 1727 { 1728 struct udp_sock *up = udp_sk(sk); 1729 int val; 1730 int err = 0; 1731 int is_udplite = IS_UDPLITE(sk); 1732 1733 if (optlen < sizeof(int)) 1734 return -EINVAL; 1735 1736 if (get_user(val, (int __user *)optval)) 1737 return -EFAULT; 1738 1739 switch (optname) { 1740 case UDP_CORK: 1741 if (val != 0) { 1742 up->corkflag = 1; 1743 } else { 1744 up->corkflag = 0; 1745 lock_sock(sk); 1746 (*push_pending_frames)(sk); 1747 release_sock(sk); 1748 } 1749 break; 1750 1751 case UDP_ENCAP: 1752 switch (val) { 1753 case 0: 1754 case UDP_ENCAP_ESPINUDP: 1755 case UDP_ENCAP_ESPINUDP_NON_IKE: 1756 up->encap_rcv = xfrm4_udp_encap_rcv; 1757 /* FALLTHROUGH */ 1758 case UDP_ENCAP_L2TPINUDP: 1759 up->encap_type = val; 1760 break; 1761 default: 1762 err = -ENOPROTOOPT; 1763 break; 1764 } 1765 break; 1766 1767 /* 1768 * UDP-Lite's partial checksum coverage (RFC 3828). 1769 */ 1770 /* The sender sets actual checksum coverage length via this option. 1771 * The case coverage > packet length is handled by send module. */ 1772 case UDPLITE_SEND_CSCOV: 1773 if (!is_udplite) /* Disable the option on UDP sockets */ 1774 return -ENOPROTOOPT; 1775 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1776 val = 8; 1777 else if (val > USHRT_MAX) 1778 val = USHRT_MAX; 1779 up->pcslen = val; 1780 up->pcflag |= UDPLITE_SEND_CC; 1781 break; 1782 1783 /* The receiver specifies a minimum checksum coverage value. To make 1784 * sense, this should be set to at least 8 (as done below). If zero is 1785 * used, this again means full checksum coverage. */ 1786 case UDPLITE_RECV_CSCOV: 1787 if (!is_udplite) /* Disable the option on UDP sockets */ 1788 return -ENOPROTOOPT; 1789 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1790 val = 8; 1791 else if (val > USHRT_MAX) 1792 val = USHRT_MAX; 1793 up->pcrlen = val; 1794 up->pcflag |= UDPLITE_RECV_CC; 1795 break; 1796 1797 default: 1798 err = -ENOPROTOOPT; 1799 break; 1800 } 1801 1802 return err; 1803 } 1804 EXPORT_SYMBOL(udp_lib_setsockopt); 1805 1806 int udp_setsockopt(struct sock *sk, int level, int optname, 1807 char __user *optval, unsigned int optlen) 1808 { 1809 if (level == SOL_UDP || level == SOL_UDPLITE) 1810 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1811 udp_push_pending_frames); 1812 return ip_setsockopt(sk, level, optname, optval, optlen); 1813 } 1814 1815 #ifdef CONFIG_COMPAT 1816 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1817 char __user *optval, unsigned int optlen) 1818 { 1819 if (level == SOL_UDP || level == SOL_UDPLITE) 1820 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1821 udp_push_pending_frames); 1822 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1823 } 1824 #endif 1825 1826 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1827 char __user *optval, int __user *optlen) 1828 { 1829 struct udp_sock *up = udp_sk(sk); 1830 int val, len; 1831 1832 if (get_user(len, optlen)) 1833 return -EFAULT; 1834 1835 len = min_t(unsigned int, len, sizeof(int)); 1836 1837 if (len < 0) 1838 return -EINVAL; 1839 1840 switch (optname) { 1841 case UDP_CORK: 1842 val = up->corkflag; 1843 break; 1844 1845 case UDP_ENCAP: 1846 val = up->encap_type; 1847 break; 1848 1849 /* The following two cannot be changed on UDP sockets, the return is 1850 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1851 case UDPLITE_SEND_CSCOV: 1852 val = up->pcslen; 1853 break; 1854 1855 case UDPLITE_RECV_CSCOV: 1856 val = up->pcrlen; 1857 break; 1858 1859 default: 1860 return -ENOPROTOOPT; 1861 } 1862 1863 if (put_user(len, optlen)) 1864 return -EFAULT; 1865 if (copy_to_user(optval, &val, len)) 1866 return -EFAULT; 1867 return 0; 1868 } 1869 EXPORT_SYMBOL(udp_lib_getsockopt); 1870 1871 int udp_getsockopt(struct sock *sk, int level, int optname, 1872 char __user *optval, int __user *optlen) 1873 { 1874 if (level == SOL_UDP || level == SOL_UDPLITE) 1875 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1876 return ip_getsockopt(sk, level, optname, optval, optlen); 1877 } 1878 1879 #ifdef CONFIG_COMPAT 1880 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1881 char __user *optval, int __user *optlen) 1882 { 1883 if (level == SOL_UDP || level == SOL_UDPLITE) 1884 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1885 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1886 } 1887 #endif 1888 /** 1889 * udp_poll - wait for a UDP event. 1890 * @file - file struct 1891 * @sock - socket 1892 * @wait - poll table 1893 * 1894 * This is same as datagram poll, except for the special case of 1895 * blocking sockets. If application is using a blocking fd 1896 * and a packet with checksum error is in the queue; 1897 * then it could get return from select indicating data available 1898 * but then block when reading it. Add special case code 1899 * to work around these arguably broken applications. 1900 */ 1901 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1902 { 1903 unsigned int mask = datagram_poll(file, sock, wait); 1904 struct sock *sk = sock->sk; 1905 1906 /* Check for false positives due to checksum errors */ 1907 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) && 1908 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk)) 1909 mask &= ~(POLLIN | POLLRDNORM); 1910 1911 return mask; 1912 1913 } 1914 EXPORT_SYMBOL(udp_poll); 1915 1916 struct proto udp_prot = { 1917 .name = "UDP", 1918 .owner = THIS_MODULE, 1919 .close = udp_lib_close, 1920 .connect = ip4_datagram_connect, 1921 .disconnect = udp_disconnect, 1922 .ioctl = udp_ioctl, 1923 .destroy = udp_destroy_sock, 1924 .setsockopt = udp_setsockopt, 1925 .getsockopt = udp_getsockopt, 1926 .sendmsg = udp_sendmsg, 1927 .recvmsg = udp_recvmsg, 1928 .sendpage = udp_sendpage, 1929 .backlog_rcv = __udp_queue_rcv_skb, 1930 .hash = udp_lib_hash, 1931 .unhash = udp_lib_unhash, 1932 .rehash = udp_v4_rehash, 1933 .get_port = udp_v4_get_port, 1934 .memory_allocated = &udp_memory_allocated, 1935 .sysctl_mem = sysctl_udp_mem, 1936 .sysctl_wmem = &sysctl_udp_wmem_min, 1937 .sysctl_rmem = &sysctl_udp_rmem_min, 1938 .obj_size = sizeof(struct udp_sock), 1939 .slab_flags = SLAB_DESTROY_BY_RCU, 1940 .h.udp_table = &udp_table, 1941 #ifdef CONFIG_COMPAT 1942 .compat_setsockopt = compat_udp_setsockopt, 1943 .compat_getsockopt = compat_udp_getsockopt, 1944 #endif 1945 .clear_sk = sk_prot_clear_portaddr_nulls, 1946 }; 1947 EXPORT_SYMBOL(udp_prot); 1948 1949 /* ------------------------------------------------------------------------ */ 1950 #ifdef CONFIG_PROC_FS 1951 1952 static struct sock *udp_get_first(struct seq_file *seq, int start) 1953 { 1954 struct sock *sk; 1955 struct udp_iter_state *state = seq->private; 1956 struct net *net = seq_file_net(seq); 1957 1958 for (state->bucket = start; state->bucket <= state->udp_table->mask; 1959 ++state->bucket) { 1960 struct hlist_nulls_node *node; 1961 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket]; 1962 1963 if (hlist_nulls_empty(&hslot->head)) 1964 continue; 1965 1966 spin_lock_bh(&hslot->lock); 1967 sk_nulls_for_each(sk, node, &hslot->head) { 1968 if (!net_eq(sock_net(sk), net)) 1969 continue; 1970 if (sk->sk_family == state->family) 1971 goto found; 1972 } 1973 spin_unlock_bh(&hslot->lock); 1974 } 1975 sk = NULL; 1976 found: 1977 return sk; 1978 } 1979 1980 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1981 { 1982 struct udp_iter_state *state = seq->private; 1983 struct net *net = seq_file_net(seq); 1984 1985 do { 1986 sk = sk_nulls_next(sk); 1987 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1988 1989 if (!sk) { 1990 if (state->bucket <= state->udp_table->mask) 1991 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 1992 return udp_get_first(seq, state->bucket + 1); 1993 } 1994 return sk; 1995 } 1996 1997 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1998 { 1999 struct sock *sk = udp_get_first(seq, 0); 2000 2001 if (sk) 2002 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 2003 --pos; 2004 return pos ? NULL : sk; 2005 } 2006 2007 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 2008 { 2009 struct udp_iter_state *state = seq->private; 2010 state->bucket = MAX_UDP_PORTS; 2011 2012 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 2013 } 2014 2015 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2016 { 2017 struct sock *sk; 2018 2019 if (v == SEQ_START_TOKEN) 2020 sk = udp_get_idx(seq, 0); 2021 else 2022 sk = udp_get_next(seq, v); 2023 2024 ++*pos; 2025 return sk; 2026 } 2027 2028 static void udp_seq_stop(struct seq_file *seq, void *v) 2029 { 2030 struct udp_iter_state *state = seq->private; 2031 2032 if (state->bucket <= state->udp_table->mask) 2033 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock); 2034 } 2035 2036 static int udp_seq_open(struct inode *inode, struct file *file) 2037 { 2038 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 2039 struct udp_iter_state *s; 2040 int err; 2041 2042 err = seq_open_net(inode, file, &afinfo->seq_ops, 2043 sizeof(struct udp_iter_state)); 2044 if (err < 0) 2045 return err; 2046 2047 s = ((struct seq_file *)file->private_data)->private; 2048 s->family = afinfo->family; 2049 s->udp_table = afinfo->udp_table; 2050 return err; 2051 } 2052 2053 /* ------------------------------------------------------------------------ */ 2054 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 2055 { 2056 struct proc_dir_entry *p; 2057 int rc = 0; 2058 2059 afinfo->seq_fops.open = udp_seq_open; 2060 afinfo->seq_fops.read = seq_read; 2061 afinfo->seq_fops.llseek = seq_lseek; 2062 afinfo->seq_fops.release = seq_release_net; 2063 2064 afinfo->seq_ops.start = udp_seq_start; 2065 afinfo->seq_ops.next = udp_seq_next; 2066 afinfo->seq_ops.stop = udp_seq_stop; 2067 2068 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 2069 &afinfo->seq_fops, afinfo); 2070 if (!p) 2071 rc = -ENOMEM; 2072 return rc; 2073 } 2074 EXPORT_SYMBOL(udp_proc_register); 2075 2076 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 2077 { 2078 proc_net_remove(net, afinfo->name); 2079 } 2080 EXPORT_SYMBOL(udp_proc_unregister); 2081 2082 /* ------------------------------------------------------------------------ */ 2083 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 2084 int bucket, int *len) 2085 { 2086 struct inet_sock *inet = inet_sk(sp); 2087 __be32 dest = inet->inet_daddr; 2088 __be32 src = inet->inet_rcv_saddr; 2089 __u16 destp = ntohs(inet->inet_dport); 2090 __u16 srcp = ntohs(inet->inet_sport); 2091 2092 seq_printf(f, "%5d: %08X:%04X %08X:%04X" 2093 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 2094 bucket, src, srcp, dest, destp, sp->sk_state, 2095 sk_wmem_alloc_get(sp), 2096 sk_rmem_alloc_get(sp), 2097 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 2098 atomic_read(&sp->sk_refcnt), sp, 2099 atomic_read(&sp->sk_drops), len); 2100 } 2101 2102 int udp4_seq_show(struct seq_file *seq, void *v) 2103 { 2104 if (v == SEQ_START_TOKEN) 2105 seq_printf(seq, "%-127s\n", 2106 " sl local_address rem_address st tx_queue " 2107 "rx_queue tr tm->when retrnsmt uid timeout " 2108 "inode ref pointer drops"); 2109 else { 2110 struct udp_iter_state *state = seq->private; 2111 int len; 2112 2113 udp4_format_sock(v, seq, state->bucket, &len); 2114 seq_printf(seq, "%*s\n", 127 - len, ""); 2115 } 2116 return 0; 2117 } 2118 2119 /* ------------------------------------------------------------------------ */ 2120 static struct udp_seq_afinfo udp4_seq_afinfo = { 2121 .name = "udp", 2122 .family = AF_INET, 2123 .udp_table = &udp_table, 2124 .seq_fops = { 2125 .owner = THIS_MODULE, 2126 }, 2127 .seq_ops = { 2128 .show = udp4_seq_show, 2129 }, 2130 }; 2131 2132 static int __net_init udp4_proc_init_net(struct net *net) 2133 { 2134 return udp_proc_register(net, &udp4_seq_afinfo); 2135 } 2136 2137 static void __net_exit udp4_proc_exit_net(struct net *net) 2138 { 2139 udp_proc_unregister(net, &udp4_seq_afinfo); 2140 } 2141 2142 static struct pernet_operations udp4_net_ops = { 2143 .init = udp4_proc_init_net, 2144 .exit = udp4_proc_exit_net, 2145 }; 2146 2147 int __init udp4_proc_init(void) 2148 { 2149 return register_pernet_subsys(&udp4_net_ops); 2150 } 2151 2152 void udp4_proc_exit(void) 2153 { 2154 unregister_pernet_subsys(&udp4_net_ops); 2155 } 2156 #endif /* CONFIG_PROC_FS */ 2157 2158 static __initdata unsigned long uhash_entries; 2159 static int __init set_uhash_entries(char *str) 2160 { 2161 if (!str) 2162 return 0; 2163 uhash_entries = simple_strtoul(str, &str, 0); 2164 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) 2165 uhash_entries = UDP_HTABLE_SIZE_MIN; 2166 return 1; 2167 } 2168 __setup("uhash_entries=", set_uhash_entries); 2169 2170 void __init udp_table_init(struct udp_table *table, const char *name) 2171 { 2172 unsigned int i; 2173 2174 if (!CONFIG_BASE_SMALL) 2175 table->hash = alloc_large_system_hash(name, 2176 2 * sizeof(struct udp_hslot), 2177 uhash_entries, 2178 21, /* one slot per 2 MB */ 2179 0, 2180 &table->log, 2181 &table->mask, 2182 64 * 1024); 2183 /* 2184 * Make sure hash table has the minimum size 2185 */ 2186 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) { 2187 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN * 2188 2 * sizeof(struct udp_hslot), GFP_KERNEL); 2189 if (!table->hash) 2190 panic(name); 2191 table->log = ilog2(UDP_HTABLE_SIZE_MIN); 2192 table->mask = UDP_HTABLE_SIZE_MIN - 1; 2193 } 2194 table->hash2 = table->hash + (table->mask + 1); 2195 for (i = 0; i <= table->mask; i++) { 2196 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i); 2197 table->hash[i].count = 0; 2198 spin_lock_init(&table->hash[i].lock); 2199 } 2200 for (i = 0; i <= table->mask; i++) { 2201 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i); 2202 table->hash2[i].count = 0; 2203 spin_lock_init(&table->hash2[i].lock); 2204 } 2205 } 2206 2207 void __init udp_init(void) 2208 { 2209 unsigned long nr_pages, limit; 2210 2211 udp_table_init(&udp_table, "UDP"); 2212 /* Set the pressure threshold up by the same strategy of TCP. It is a 2213 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 2214 * toward zero with the amount of memory, with a floor of 128 pages. 2215 */ 2216 nr_pages = totalram_pages - totalhigh_pages; 2217 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 2218 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 2219 limit = max(limit, 128UL); 2220 sysctl_udp_mem[0] = limit / 4 * 3; 2221 sysctl_udp_mem[1] = limit; 2222 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 2223 2224 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 2225 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 2226 } 2227 2228 int udp4_ufo_send_check(struct sk_buff *skb) 2229 { 2230 const struct iphdr *iph; 2231 struct udphdr *uh; 2232 2233 if (!pskb_may_pull(skb, sizeof(*uh))) 2234 return -EINVAL; 2235 2236 iph = ip_hdr(skb); 2237 uh = udp_hdr(skb); 2238 2239 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 2240 IPPROTO_UDP, 0); 2241 skb->csum_start = skb_transport_header(skb) - skb->head; 2242 skb->csum_offset = offsetof(struct udphdr, check); 2243 skb->ip_summed = CHECKSUM_PARTIAL; 2244 return 0; 2245 } 2246 2247 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, u32 features) 2248 { 2249 struct sk_buff *segs = ERR_PTR(-EINVAL); 2250 unsigned int mss; 2251 int offset; 2252 __wsum csum; 2253 2254 mss = skb_shinfo(skb)->gso_size; 2255 if (unlikely(skb->len <= mss)) 2256 goto out; 2257 2258 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 2259 /* Packet is from an untrusted source, reset gso_segs. */ 2260 int type = skb_shinfo(skb)->gso_type; 2261 2262 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || 2263 !(type & (SKB_GSO_UDP)))) 2264 goto out; 2265 2266 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 2267 2268 segs = NULL; 2269 goto out; 2270 } 2271 2272 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot 2273 * do checksum of UDP packets sent as multiple IP fragments. 2274 */ 2275 offset = skb_checksum_start_offset(skb); 2276 csum = skb_checksum(skb, offset, skb->len - offset, 0); 2277 offset += skb->csum_offset; 2278 *(__sum16 *)(skb->data + offset) = csum_fold(csum); 2279 skb->ip_summed = CHECKSUM_NONE; 2280 2281 /* Fragment the skb. IP headers of the fragments are updated in 2282 * inet_gso_segment() 2283 */ 2284 segs = skb_segment(skb, features); 2285 out: 2286 return segs; 2287 } 2288 2289