Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | * Implementation of the Transmission Control Protocol(TCP). |
| 7 | * |
| 8 | * Version: $Id: tcp_ipv4.c,v 1.240 2002/02/01 22:01:04 davem Exp $ |
| 9 | * |
| 10 | * IPv4 specific functions |
| 11 | * |
| 12 | * |
| 13 | * code split from: |
| 14 | * linux/ipv4/tcp.c |
| 15 | * linux/ipv4/tcp_input.c |
| 16 | * linux/ipv4/tcp_output.c |
| 17 | * |
| 18 | * See tcp.c for author information |
| 19 | * |
| 20 | * This program is free software; you can redistribute it and/or |
| 21 | * modify it under the terms of the GNU General Public License |
| 22 | * as published by the Free Software Foundation; either version |
| 23 | * 2 of the License, or (at your option) any later version. |
| 24 | */ |
| 25 | |
| 26 | /* |
| 27 | * Changes: |
| 28 | * David S. Miller : New socket lookup architecture. |
| 29 | * This code is dedicated to John Dyson. |
| 30 | * David S. Miller : Change semantics of established hash, |
| 31 | * half is devoted to TIME_WAIT sockets |
| 32 | * and the rest go in the other half. |
| 33 | * Andi Kleen : Add support for syncookies and fixed |
| 34 | * some bugs: ip options weren't passed to |
| 35 | * the TCP layer, missed a check for an |
| 36 | * ACK bit. |
| 37 | * Andi Kleen : Implemented fast path mtu discovery. |
| 38 | * Fixed many serious bugs in the |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 39 | * request_sock handling and moved |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | * most of it into the af independent code. |
| 41 | * Added tail drop and some other bugfixes. |
| 42 | * Added new listen sematics. |
| 43 | * Mike McLagan : Routing by source |
| 44 | * Juan Jose Ciarlante: ip_dynaddr bits |
| 45 | * Andi Kleen: various fixes. |
| 46 | * Vitaly E. Lavrov : Transparent proxy revived after year |
| 47 | * coma. |
| 48 | * Andi Kleen : Fix new listen. |
| 49 | * Andi Kleen : Fix accept error reporting. |
| 50 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which |
| 51 | * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind |
| 52 | * a single port at the same time. |
| 53 | */ |
| 54 | |
| 55 | #include <linux/config.h> |
| 56 | |
| 57 | #include <linux/types.h> |
| 58 | #include <linux/fcntl.h> |
| 59 | #include <linux/module.h> |
| 60 | #include <linux/random.h> |
| 61 | #include <linux/cache.h> |
| 62 | #include <linux/jhash.h> |
| 63 | #include <linux/init.h> |
| 64 | #include <linux/times.h> |
| 65 | |
| 66 | #include <net/icmp.h> |
| 67 | #include <net/tcp.h> |
| 68 | #include <net/ipv6.h> |
| 69 | #include <net/inet_common.h> |
| 70 | #include <net/xfrm.h> |
| 71 | |
| 72 | #include <linux/inet.h> |
| 73 | #include <linux/ipv6.h> |
| 74 | #include <linux/stddef.h> |
| 75 | #include <linux/proc_fs.h> |
| 76 | #include <linux/seq_file.h> |
| 77 | |
| 78 | extern int sysctl_ip_dynaddr; |
| 79 | int sysctl_tcp_tw_reuse; |
| 80 | int sysctl_tcp_low_latency; |
| 81 | |
| 82 | /* Check TCP sequence numbers in ICMP packets. */ |
| 83 | #define ICMP_MIN_LENGTH 8 |
| 84 | |
| 85 | /* Socket used for sending RSTs */ |
| 86 | static struct socket *tcp_socket; |
| 87 | |
| 88 | void tcp_v4_send_check(struct sock *sk, struct tcphdr *th, int len, |
| 89 | struct sk_buff *skb); |
| 90 | |
| 91 | struct tcp_hashinfo __cacheline_aligned tcp_hashinfo = { |
| 92 | .__tcp_lhash_lock = RW_LOCK_UNLOCKED, |
| 93 | .__tcp_lhash_users = ATOMIC_INIT(0), |
| 94 | .__tcp_lhash_wait |
| 95 | = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.__tcp_lhash_wait), |
| 96 | .__tcp_portalloc_lock = SPIN_LOCK_UNLOCKED |
| 97 | }; |
| 98 | |
| 99 | /* |
| 100 | * This array holds the first and last local port number. |
| 101 | * For high-usage systems, use sysctl to change this to |
| 102 | * 32768-61000 |
| 103 | */ |
| 104 | int sysctl_local_port_range[2] = { 1024, 4999 }; |
| 105 | int tcp_port_rover = 1024 - 1; |
| 106 | |
| 107 | static __inline__ int tcp_hashfn(__u32 laddr, __u16 lport, |
| 108 | __u32 faddr, __u16 fport) |
| 109 | { |
| 110 | int h = (laddr ^ lport) ^ (faddr ^ fport); |
| 111 | h ^= h >> 16; |
| 112 | h ^= h >> 8; |
| 113 | return h & (tcp_ehash_size - 1); |
| 114 | } |
| 115 | |
| 116 | static __inline__ int tcp_sk_hashfn(struct sock *sk) |
| 117 | { |
| 118 | struct inet_sock *inet = inet_sk(sk); |
| 119 | __u32 laddr = inet->rcv_saddr; |
| 120 | __u16 lport = inet->num; |
| 121 | __u32 faddr = inet->daddr; |
| 122 | __u16 fport = inet->dport; |
| 123 | |
| 124 | return tcp_hashfn(laddr, lport, faddr, fport); |
| 125 | } |
| 126 | |
| 127 | /* Allocate and initialize a new TCP local port bind bucket. |
| 128 | * The bindhash mutex for snum's hash chain must be held here. |
| 129 | */ |
| 130 | struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head, |
| 131 | unsigned short snum) |
| 132 | { |
| 133 | struct tcp_bind_bucket *tb = kmem_cache_alloc(tcp_bucket_cachep, |
| 134 | SLAB_ATOMIC); |
| 135 | if (tb) { |
| 136 | tb->port = snum; |
| 137 | tb->fastreuse = 0; |
| 138 | INIT_HLIST_HEAD(&tb->owners); |
| 139 | hlist_add_head(&tb->node, &head->chain); |
| 140 | } |
| 141 | return tb; |
| 142 | } |
| 143 | |
| 144 | /* Caller must hold hashbucket lock for this tb with local BH disabled */ |
| 145 | void tcp_bucket_destroy(struct tcp_bind_bucket *tb) |
| 146 | { |
| 147 | if (hlist_empty(&tb->owners)) { |
| 148 | __hlist_del(&tb->node); |
| 149 | kmem_cache_free(tcp_bucket_cachep, tb); |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | /* Caller must disable local BH processing. */ |
| 154 | static __inline__ void __tcp_inherit_port(struct sock *sk, struct sock *child) |
| 155 | { |
| 156 | struct tcp_bind_hashbucket *head = |
| 157 | &tcp_bhash[tcp_bhashfn(inet_sk(child)->num)]; |
| 158 | struct tcp_bind_bucket *tb; |
| 159 | |
| 160 | spin_lock(&head->lock); |
| 161 | tb = tcp_sk(sk)->bind_hash; |
| 162 | sk_add_bind_node(child, &tb->owners); |
| 163 | tcp_sk(child)->bind_hash = tb; |
| 164 | spin_unlock(&head->lock); |
| 165 | } |
| 166 | |
| 167 | inline void tcp_inherit_port(struct sock *sk, struct sock *child) |
| 168 | { |
| 169 | local_bh_disable(); |
| 170 | __tcp_inherit_port(sk, child); |
| 171 | local_bh_enable(); |
| 172 | } |
| 173 | |
| 174 | void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb, |
| 175 | unsigned short snum) |
| 176 | { |
| 177 | inet_sk(sk)->num = snum; |
| 178 | sk_add_bind_node(sk, &tb->owners); |
| 179 | tcp_sk(sk)->bind_hash = tb; |
| 180 | } |
| 181 | |
| 182 | static inline int tcp_bind_conflict(struct sock *sk, struct tcp_bind_bucket *tb) |
| 183 | { |
| 184 | const u32 sk_rcv_saddr = tcp_v4_rcv_saddr(sk); |
| 185 | struct sock *sk2; |
| 186 | struct hlist_node *node; |
| 187 | int reuse = sk->sk_reuse; |
| 188 | |
| 189 | sk_for_each_bound(sk2, node, &tb->owners) { |
| 190 | if (sk != sk2 && |
| 191 | !tcp_v6_ipv6only(sk2) && |
| 192 | (!sk->sk_bound_dev_if || |
| 193 | !sk2->sk_bound_dev_if || |
| 194 | sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { |
| 195 | if (!reuse || !sk2->sk_reuse || |
| 196 | sk2->sk_state == TCP_LISTEN) { |
| 197 | const u32 sk2_rcv_saddr = tcp_v4_rcv_saddr(sk2); |
| 198 | if (!sk2_rcv_saddr || !sk_rcv_saddr || |
| 199 | sk2_rcv_saddr == sk_rcv_saddr) |
| 200 | break; |
| 201 | } |
| 202 | } |
| 203 | } |
| 204 | return node != NULL; |
| 205 | } |
| 206 | |
| 207 | /* Obtain a reference to a local port for the given sock, |
| 208 | * if snum is zero it means select any available local port. |
| 209 | */ |
| 210 | static int tcp_v4_get_port(struct sock *sk, unsigned short snum) |
| 211 | { |
| 212 | struct tcp_bind_hashbucket *head; |
| 213 | struct hlist_node *node; |
| 214 | struct tcp_bind_bucket *tb; |
| 215 | int ret; |
| 216 | |
| 217 | local_bh_disable(); |
| 218 | if (!snum) { |
| 219 | int low = sysctl_local_port_range[0]; |
| 220 | int high = sysctl_local_port_range[1]; |
| 221 | int remaining = (high - low) + 1; |
| 222 | int rover; |
| 223 | |
| 224 | spin_lock(&tcp_portalloc_lock); |
Folkert van Heusden | 0b2531b | 2005-05-03 14:36:08 -0700 | [diff] [blame] | 225 | if (tcp_port_rover < low) |
| 226 | rover = low; |
| 227 | else |
| 228 | rover = tcp_port_rover; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 229 | do { |
| 230 | rover++; |
Folkert van Heusden | 0b2531b | 2005-05-03 14:36:08 -0700 | [diff] [blame] | 231 | if (rover > high) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | rover = low; |
| 233 | head = &tcp_bhash[tcp_bhashfn(rover)]; |
| 234 | spin_lock(&head->lock); |
| 235 | tb_for_each(tb, node, &head->chain) |
| 236 | if (tb->port == rover) |
| 237 | goto next; |
| 238 | break; |
| 239 | next: |
| 240 | spin_unlock(&head->lock); |
| 241 | } while (--remaining > 0); |
| 242 | tcp_port_rover = rover; |
| 243 | spin_unlock(&tcp_portalloc_lock); |
| 244 | |
| 245 | /* Exhausted local port range during search? */ |
| 246 | ret = 1; |
| 247 | if (remaining <= 0) |
| 248 | goto fail; |
| 249 | |
| 250 | /* OK, here is the one we will use. HEAD is |
| 251 | * non-NULL and we hold it's mutex. |
| 252 | */ |
| 253 | snum = rover; |
| 254 | } else { |
| 255 | head = &tcp_bhash[tcp_bhashfn(snum)]; |
| 256 | spin_lock(&head->lock); |
| 257 | tb_for_each(tb, node, &head->chain) |
| 258 | if (tb->port == snum) |
| 259 | goto tb_found; |
| 260 | } |
| 261 | tb = NULL; |
| 262 | goto tb_not_found; |
| 263 | tb_found: |
| 264 | if (!hlist_empty(&tb->owners)) { |
| 265 | if (sk->sk_reuse > 1) |
| 266 | goto success; |
| 267 | if (tb->fastreuse > 0 && |
| 268 | sk->sk_reuse && sk->sk_state != TCP_LISTEN) { |
| 269 | goto success; |
| 270 | } else { |
| 271 | ret = 1; |
| 272 | if (tcp_bind_conflict(sk, tb)) |
| 273 | goto fail_unlock; |
| 274 | } |
| 275 | } |
| 276 | tb_not_found: |
| 277 | ret = 1; |
| 278 | if (!tb && (tb = tcp_bucket_create(head, snum)) == NULL) |
| 279 | goto fail_unlock; |
| 280 | if (hlist_empty(&tb->owners)) { |
| 281 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN) |
| 282 | tb->fastreuse = 1; |
| 283 | else |
| 284 | tb->fastreuse = 0; |
| 285 | } else if (tb->fastreuse && |
| 286 | (!sk->sk_reuse || sk->sk_state == TCP_LISTEN)) |
| 287 | tb->fastreuse = 0; |
| 288 | success: |
| 289 | if (!tcp_sk(sk)->bind_hash) |
| 290 | tcp_bind_hash(sk, tb, snum); |
| 291 | BUG_TRAP(tcp_sk(sk)->bind_hash == tb); |
| 292 | ret = 0; |
| 293 | |
| 294 | fail_unlock: |
| 295 | spin_unlock(&head->lock); |
| 296 | fail: |
| 297 | local_bh_enable(); |
| 298 | return ret; |
| 299 | } |
| 300 | |
| 301 | /* Get rid of any references to a local port held by the |
| 302 | * given sock. |
| 303 | */ |
| 304 | static void __tcp_put_port(struct sock *sk) |
| 305 | { |
| 306 | struct inet_sock *inet = inet_sk(sk); |
| 307 | struct tcp_bind_hashbucket *head = &tcp_bhash[tcp_bhashfn(inet->num)]; |
| 308 | struct tcp_bind_bucket *tb; |
| 309 | |
| 310 | spin_lock(&head->lock); |
| 311 | tb = tcp_sk(sk)->bind_hash; |
| 312 | __sk_del_bind_node(sk); |
| 313 | tcp_sk(sk)->bind_hash = NULL; |
| 314 | inet->num = 0; |
| 315 | tcp_bucket_destroy(tb); |
| 316 | spin_unlock(&head->lock); |
| 317 | } |
| 318 | |
| 319 | void tcp_put_port(struct sock *sk) |
| 320 | { |
| 321 | local_bh_disable(); |
| 322 | __tcp_put_port(sk); |
| 323 | local_bh_enable(); |
| 324 | } |
| 325 | |
| 326 | /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it can be very bad on SMP. |
| 327 | * Look, when several writers sleep and reader wakes them up, all but one |
| 328 | * immediately hit write lock and grab all the cpus. Exclusive sleep solves |
| 329 | * this, _but_ remember, it adds useless work on UP machines (wake up each |
| 330 | * exclusive lock release). It should be ifdefed really. |
| 331 | */ |
| 332 | |
| 333 | void tcp_listen_wlock(void) |
| 334 | { |
| 335 | write_lock(&tcp_lhash_lock); |
| 336 | |
| 337 | if (atomic_read(&tcp_lhash_users)) { |
| 338 | DEFINE_WAIT(wait); |
| 339 | |
| 340 | for (;;) { |
| 341 | prepare_to_wait_exclusive(&tcp_lhash_wait, |
| 342 | &wait, TASK_UNINTERRUPTIBLE); |
| 343 | if (!atomic_read(&tcp_lhash_users)) |
| 344 | break; |
| 345 | write_unlock_bh(&tcp_lhash_lock); |
| 346 | schedule(); |
| 347 | write_lock_bh(&tcp_lhash_lock); |
| 348 | } |
| 349 | |
| 350 | finish_wait(&tcp_lhash_wait, &wait); |
| 351 | } |
| 352 | } |
| 353 | |
| 354 | static __inline__ void __tcp_v4_hash(struct sock *sk, const int listen_possible) |
| 355 | { |
| 356 | struct hlist_head *list; |
| 357 | rwlock_t *lock; |
| 358 | |
| 359 | BUG_TRAP(sk_unhashed(sk)); |
| 360 | if (listen_possible && sk->sk_state == TCP_LISTEN) { |
| 361 | list = &tcp_listening_hash[tcp_sk_listen_hashfn(sk)]; |
| 362 | lock = &tcp_lhash_lock; |
| 363 | tcp_listen_wlock(); |
| 364 | } else { |
| 365 | list = &tcp_ehash[(sk->sk_hashent = tcp_sk_hashfn(sk))].chain; |
| 366 | lock = &tcp_ehash[sk->sk_hashent].lock; |
| 367 | write_lock(lock); |
| 368 | } |
| 369 | __sk_add_node(sk, list); |
| 370 | sock_prot_inc_use(sk->sk_prot); |
| 371 | write_unlock(lock); |
| 372 | if (listen_possible && sk->sk_state == TCP_LISTEN) |
| 373 | wake_up(&tcp_lhash_wait); |
| 374 | } |
| 375 | |
| 376 | static void tcp_v4_hash(struct sock *sk) |
| 377 | { |
| 378 | if (sk->sk_state != TCP_CLOSE) { |
| 379 | local_bh_disable(); |
| 380 | __tcp_v4_hash(sk, 1); |
| 381 | local_bh_enable(); |
| 382 | } |
| 383 | } |
| 384 | |
| 385 | void tcp_unhash(struct sock *sk) |
| 386 | { |
| 387 | rwlock_t *lock; |
| 388 | |
| 389 | if (sk_unhashed(sk)) |
| 390 | goto ende; |
| 391 | |
| 392 | if (sk->sk_state == TCP_LISTEN) { |
| 393 | local_bh_disable(); |
| 394 | tcp_listen_wlock(); |
| 395 | lock = &tcp_lhash_lock; |
| 396 | } else { |
| 397 | struct tcp_ehash_bucket *head = &tcp_ehash[sk->sk_hashent]; |
| 398 | lock = &head->lock; |
| 399 | write_lock_bh(&head->lock); |
| 400 | } |
| 401 | |
| 402 | if (__sk_del_node_init(sk)) |
| 403 | sock_prot_dec_use(sk->sk_prot); |
| 404 | write_unlock_bh(lock); |
| 405 | |
| 406 | ende: |
| 407 | if (sk->sk_state == TCP_LISTEN) |
| 408 | wake_up(&tcp_lhash_wait); |
| 409 | } |
| 410 | |
| 411 | /* Don't inline this cruft. Here are some nice properties to |
| 412 | * exploit here. The BSD API does not allow a listening TCP |
| 413 | * to specify the remote port nor the remote address for the |
| 414 | * connection. So always assume those are both wildcarded |
| 415 | * during the search since they can never be otherwise. |
| 416 | */ |
| 417 | static struct sock *__tcp_v4_lookup_listener(struct hlist_head *head, u32 daddr, |
| 418 | unsigned short hnum, int dif) |
| 419 | { |
| 420 | struct sock *result = NULL, *sk; |
| 421 | struct hlist_node *node; |
| 422 | int score, hiscore; |
| 423 | |
| 424 | hiscore=-1; |
| 425 | sk_for_each(sk, node, head) { |
| 426 | struct inet_sock *inet = inet_sk(sk); |
| 427 | |
| 428 | if (inet->num == hnum && !ipv6_only_sock(sk)) { |
| 429 | __u32 rcv_saddr = inet->rcv_saddr; |
| 430 | |
| 431 | score = (sk->sk_family == PF_INET ? 1 : 0); |
| 432 | if (rcv_saddr) { |
| 433 | if (rcv_saddr != daddr) |
| 434 | continue; |
| 435 | score+=2; |
| 436 | } |
| 437 | if (sk->sk_bound_dev_if) { |
| 438 | if (sk->sk_bound_dev_if != dif) |
| 439 | continue; |
| 440 | score+=2; |
| 441 | } |
| 442 | if (score == 5) |
| 443 | return sk; |
| 444 | if (score > hiscore) { |
| 445 | hiscore = score; |
| 446 | result = sk; |
| 447 | } |
| 448 | } |
| 449 | } |
| 450 | return result; |
| 451 | } |
| 452 | |
| 453 | /* Optimize the common listener case. */ |
| 454 | static inline struct sock *tcp_v4_lookup_listener(u32 daddr, |
| 455 | unsigned short hnum, int dif) |
| 456 | { |
| 457 | struct sock *sk = NULL; |
| 458 | struct hlist_head *head; |
| 459 | |
| 460 | read_lock(&tcp_lhash_lock); |
| 461 | head = &tcp_listening_hash[tcp_lhashfn(hnum)]; |
| 462 | if (!hlist_empty(head)) { |
| 463 | struct inet_sock *inet = inet_sk((sk = __sk_head(head))); |
| 464 | |
| 465 | if (inet->num == hnum && !sk->sk_node.next && |
| 466 | (!inet->rcv_saddr || inet->rcv_saddr == daddr) && |
| 467 | (sk->sk_family == PF_INET || !ipv6_only_sock(sk)) && |
| 468 | !sk->sk_bound_dev_if) |
| 469 | goto sherry_cache; |
| 470 | sk = __tcp_v4_lookup_listener(head, daddr, hnum, dif); |
| 471 | } |
| 472 | if (sk) { |
| 473 | sherry_cache: |
| 474 | sock_hold(sk); |
| 475 | } |
| 476 | read_unlock(&tcp_lhash_lock); |
| 477 | return sk; |
| 478 | } |
| 479 | |
| 480 | /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so |
| 481 | * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM |
| 482 | * |
| 483 | * Local BH must be disabled here. |
| 484 | */ |
| 485 | |
| 486 | static inline struct sock *__tcp_v4_lookup_established(u32 saddr, u16 sport, |
| 487 | u32 daddr, u16 hnum, |
| 488 | int dif) |
| 489 | { |
| 490 | struct tcp_ehash_bucket *head; |
| 491 | TCP_V4_ADDR_COOKIE(acookie, saddr, daddr) |
| 492 | __u32 ports = TCP_COMBINED_PORTS(sport, hnum); |
| 493 | struct sock *sk; |
| 494 | struct hlist_node *node; |
| 495 | /* Optimize here for direct hit, only listening connections can |
| 496 | * have wildcards anyways. |
| 497 | */ |
| 498 | int hash = tcp_hashfn(daddr, hnum, saddr, sport); |
| 499 | head = &tcp_ehash[hash]; |
| 500 | read_lock(&head->lock); |
| 501 | sk_for_each(sk, node, &head->chain) { |
| 502 | if (TCP_IPV4_MATCH(sk, acookie, saddr, daddr, ports, dif)) |
| 503 | goto hit; /* You sunk my battleship! */ |
| 504 | } |
| 505 | |
| 506 | /* Must check for a TIME_WAIT'er before going to listener hash. */ |
| 507 | sk_for_each(sk, node, &(head + tcp_ehash_size)->chain) { |
| 508 | if (TCP_IPV4_TW_MATCH(sk, acookie, saddr, daddr, ports, dif)) |
| 509 | goto hit; |
| 510 | } |
| 511 | sk = NULL; |
| 512 | out: |
| 513 | read_unlock(&head->lock); |
| 514 | return sk; |
| 515 | hit: |
| 516 | sock_hold(sk); |
| 517 | goto out; |
| 518 | } |
| 519 | |
| 520 | static inline struct sock *__tcp_v4_lookup(u32 saddr, u16 sport, |
| 521 | u32 daddr, u16 hnum, int dif) |
| 522 | { |
| 523 | struct sock *sk = __tcp_v4_lookup_established(saddr, sport, |
| 524 | daddr, hnum, dif); |
| 525 | |
| 526 | return sk ? : tcp_v4_lookup_listener(daddr, hnum, dif); |
| 527 | } |
| 528 | |
| 529 | inline struct sock *tcp_v4_lookup(u32 saddr, u16 sport, u32 daddr, |
| 530 | u16 dport, int dif) |
| 531 | { |
| 532 | struct sock *sk; |
| 533 | |
| 534 | local_bh_disable(); |
| 535 | sk = __tcp_v4_lookup(saddr, sport, daddr, ntohs(dport), dif); |
| 536 | local_bh_enable(); |
| 537 | |
| 538 | return sk; |
| 539 | } |
| 540 | |
| 541 | EXPORT_SYMBOL_GPL(tcp_v4_lookup); |
| 542 | |
| 543 | static inline __u32 tcp_v4_init_sequence(struct sock *sk, struct sk_buff *skb) |
| 544 | { |
| 545 | return secure_tcp_sequence_number(skb->nh.iph->daddr, |
| 546 | skb->nh.iph->saddr, |
| 547 | skb->h.th->dest, |
| 548 | skb->h.th->source); |
| 549 | } |
| 550 | |
| 551 | /* called with local bh disabled */ |
| 552 | static int __tcp_v4_check_established(struct sock *sk, __u16 lport, |
| 553 | struct tcp_tw_bucket **twp) |
| 554 | { |
| 555 | struct inet_sock *inet = inet_sk(sk); |
| 556 | u32 daddr = inet->rcv_saddr; |
| 557 | u32 saddr = inet->daddr; |
| 558 | int dif = sk->sk_bound_dev_if; |
| 559 | TCP_V4_ADDR_COOKIE(acookie, saddr, daddr) |
| 560 | __u32 ports = TCP_COMBINED_PORTS(inet->dport, lport); |
| 561 | int hash = tcp_hashfn(daddr, lport, saddr, inet->dport); |
| 562 | struct tcp_ehash_bucket *head = &tcp_ehash[hash]; |
| 563 | struct sock *sk2; |
| 564 | struct hlist_node *node; |
| 565 | struct tcp_tw_bucket *tw; |
| 566 | |
| 567 | write_lock(&head->lock); |
| 568 | |
| 569 | /* Check TIME-WAIT sockets first. */ |
| 570 | sk_for_each(sk2, node, &(head + tcp_ehash_size)->chain) { |
| 571 | tw = (struct tcp_tw_bucket *)sk2; |
| 572 | |
| 573 | if (TCP_IPV4_TW_MATCH(sk2, acookie, saddr, daddr, ports, dif)) { |
| 574 | struct tcp_sock *tp = tcp_sk(sk); |
| 575 | |
| 576 | /* With PAWS, it is safe from the viewpoint |
| 577 | of data integrity. Even without PAWS it |
| 578 | is safe provided sequence spaces do not |
| 579 | overlap i.e. at data rates <= 80Mbit/sec. |
| 580 | |
| 581 | Actually, the idea is close to VJ's one, |
| 582 | only timestamp cache is held not per host, |
| 583 | but per port pair and TW bucket is used |
| 584 | as state holder. |
| 585 | |
| 586 | If TW bucket has been already destroyed we |
| 587 | fall back to VJ's scheme and use initial |
| 588 | timestamp retrieved from peer table. |
| 589 | */ |
| 590 | if (tw->tw_ts_recent_stamp && |
| 591 | (!twp || (sysctl_tcp_tw_reuse && |
| 592 | xtime.tv_sec - |
| 593 | tw->tw_ts_recent_stamp > 1))) { |
| 594 | if ((tp->write_seq = |
| 595 | tw->tw_snd_nxt + 65535 + 2) == 0) |
| 596 | tp->write_seq = 1; |
| 597 | tp->rx_opt.ts_recent = tw->tw_ts_recent; |
| 598 | tp->rx_opt.ts_recent_stamp = tw->tw_ts_recent_stamp; |
| 599 | sock_hold(sk2); |
| 600 | goto unique; |
| 601 | } else |
| 602 | goto not_unique; |
| 603 | } |
| 604 | } |
| 605 | tw = NULL; |
| 606 | |
| 607 | /* And established part... */ |
| 608 | sk_for_each(sk2, node, &head->chain) { |
| 609 | if (TCP_IPV4_MATCH(sk2, acookie, saddr, daddr, ports, dif)) |
| 610 | goto not_unique; |
| 611 | } |
| 612 | |
| 613 | unique: |
| 614 | /* Must record num and sport now. Otherwise we will see |
| 615 | * in hash table socket with a funny identity. */ |
| 616 | inet->num = lport; |
| 617 | inet->sport = htons(lport); |
| 618 | sk->sk_hashent = hash; |
| 619 | BUG_TRAP(sk_unhashed(sk)); |
| 620 | __sk_add_node(sk, &head->chain); |
| 621 | sock_prot_inc_use(sk->sk_prot); |
| 622 | write_unlock(&head->lock); |
| 623 | |
| 624 | if (twp) { |
| 625 | *twp = tw; |
| 626 | NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED); |
| 627 | } else if (tw) { |
| 628 | /* Silly. Should hash-dance instead... */ |
| 629 | tcp_tw_deschedule(tw); |
| 630 | NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED); |
| 631 | |
| 632 | tcp_tw_put(tw); |
| 633 | } |
| 634 | |
| 635 | return 0; |
| 636 | |
| 637 | not_unique: |
| 638 | write_unlock(&head->lock); |
| 639 | return -EADDRNOTAVAIL; |
| 640 | } |
| 641 | |
| 642 | static inline u32 connect_port_offset(const struct sock *sk) |
| 643 | { |
| 644 | const struct inet_sock *inet = inet_sk(sk); |
| 645 | |
| 646 | return secure_tcp_port_ephemeral(inet->rcv_saddr, inet->daddr, |
| 647 | inet->dport); |
| 648 | } |
| 649 | |
| 650 | /* |
| 651 | * Bind a port for a connect operation and hash it. |
| 652 | */ |
| 653 | static inline int tcp_v4_hash_connect(struct sock *sk) |
| 654 | { |
| 655 | unsigned short snum = inet_sk(sk)->num; |
| 656 | struct tcp_bind_hashbucket *head; |
| 657 | struct tcp_bind_bucket *tb; |
| 658 | int ret; |
| 659 | |
| 660 | if (!snum) { |
| 661 | int low = sysctl_local_port_range[0]; |
| 662 | int high = sysctl_local_port_range[1]; |
| 663 | int range = high - low; |
| 664 | int i; |
| 665 | int port; |
| 666 | static u32 hint; |
| 667 | u32 offset = hint + connect_port_offset(sk); |
| 668 | struct hlist_node *node; |
| 669 | struct tcp_tw_bucket *tw = NULL; |
| 670 | |
| 671 | local_bh_disable(); |
| 672 | for (i = 1; i <= range; i++) { |
| 673 | port = low + (i + offset) % range; |
| 674 | head = &tcp_bhash[tcp_bhashfn(port)]; |
| 675 | spin_lock(&head->lock); |
| 676 | |
| 677 | /* Does not bother with rcv_saddr checks, |
| 678 | * because the established check is already |
| 679 | * unique enough. |
| 680 | */ |
| 681 | tb_for_each(tb, node, &head->chain) { |
| 682 | if (tb->port == port) { |
| 683 | BUG_TRAP(!hlist_empty(&tb->owners)); |
| 684 | if (tb->fastreuse >= 0) |
| 685 | goto next_port; |
| 686 | if (!__tcp_v4_check_established(sk, |
| 687 | port, |
| 688 | &tw)) |
| 689 | goto ok; |
| 690 | goto next_port; |
| 691 | } |
| 692 | } |
| 693 | |
| 694 | tb = tcp_bucket_create(head, port); |
| 695 | if (!tb) { |
| 696 | spin_unlock(&head->lock); |
| 697 | break; |
| 698 | } |
| 699 | tb->fastreuse = -1; |
| 700 | goto ok; |
| 701 | |
| 702 | next_port: |
| 703 | spin_unlock(&head->lock); |
| 704 | } |
| 705 | local_bh_enable(); |
| 706 | |
| 707 | return -EADDRNOTAVAIL; |
| 708 | |
| 709 | ok: |
| 710 | hint += i; |
| 711 | |
| 712 | /* Head lock still held and bh's disabled */ |
| 713 | tcp_bind_hash(sk, tb, port); |
| 714 | if (sk_unhashed(sk)) { |
| 715 | inet_sk(sk)->sport = htons(port); |
| 716 | __tcp_v4_hash(sk, 0); |
| 717 | } |
| 718 | spin_unlock(&head->lock); |
| 719 | |
| 720 | if (tw) { |
| 721 | tcp_tw_deschedule(tw); |
| 722 | tcp_tw_put(tw); |
| 723 | } |
| 724 | |
| 725 | ret = 0; |
| 726 | goto out; |
| 727 | } |
| 728 | |
| 729 | head = &tcp_bhash[tcp_bhashfn(snum)]; |
| 730 | tb = tcp_sk(sk)->bind_hash; |
| 731 | spin_lock_bh(&head->lock); |
| 732 | if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) { |
| 733 | __tcp_v4_hash(sk, 0); |
| 734 | spin_unlock_bh(&head->lock); |
| 735 | return 0; |
| 736 | } else { |
| 737 | spin_unlock(&head->lock); |
| 738 | /* No definite answer... Walk to established hash table */ |
| 739 | ret = __tcp_v4_check_established(sk, snum, NULL); |
| 740 | out: |
| 741 | local_bh_enable(); |
| 742 | return ret; |
| 743 | } |
| 744 | } |
| 745 | |
| 746 | /* This will initiate an outgoing connection. */ |
| 747 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) |
| 748 | { |
| 749 | struct inet_sock *inet = inet_sk(sk); |
| 750 | struct tcp_sock *tp = tcp_sk(sk); |
| 751 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; |
| 752 | struct rtable *rt; |
| 753 | u32 daddr, nexthop; |
| 754 | int tmp; |
| 755 | int err; |
| 756 | |
| 757 | if (addr_len < sizeof(struct sockaddr_in)) |
| 758 | return -EINVAL; |
| 759 | |
| 760 | if (usin->sin_family != AF_INET) |
| 761 | return -EAFNOSUPPORT; |
| 762 | |
| 763 | nexthop = daddr = usin->sin_addr.s_addr; |
| 764 | if (inet->opt && inet->opt->srr) { |
| 765 | if (!daddr) |
| 766 | return -EINVAL; |
| 767 | nexthop = inet->opt->faddr; |
| 768 | } |
| 769 | |
| 770 | tmp = ip_route_connect(&rt, nexthop, inet->saddr, |
| 771 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, |
| 772 | IPPROTO_TCP, |
| 773 | inet->sport, usin->sin_port, sk); |
| 774 | if (tmp < 0) |
| 775 | return tmp; |
| 776 | |
| 777 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { |
| 778 | ip_rt_put(rt); |
| 779 | return -ENETUNREACH; |
| 780 | } |
| 781 | |
| 782 | if (!inet->opt || !inet->opt->srr) |
| 783 | daddr = rt->rt_dst; |
| 784 | |
| 785 | if (!inet->saddr) |
| 786 | inet->saddr = rt->rt_src; |
| 787 | inet->rcv_saddr = inet->saddr; |
| 788 | |
| 789 | if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) { |
| 790 | /* Reset inherited state */ |
| 791 | tp->rx_opt.ts_recent = 0; |
| 792 | tp->rx_opt.ts_recent_stamp = 0; |
| 793 | tp->write_seq = 0; |
| 794 | } |
| 795 | |
| 796 | if (sysctl_tcp_tw_recycle && |
| 797 | !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) { |
| 798 | struct inet_peer *peer = rt_get_peer(rt); |
| 799 | |
| 800 | /* VJ's idea. We save last timestamp seen from |
| 801 | * the destination in peer table, when entering state TIME-WAIT |
| 802 | * and initialize rx_opt.ts_recent from it, when trying new connection. |
| 803 | */ |
| 804 | |
| 805 | if (peer && peer->tcp_ts_stamp + TCP_PAWS_MSL >= xtime.tv_sec) { |
| 806 | tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp; |
| 807 | tp->rx_opt.ts_recent = peer->tcp_ts; |
| 808 | } |
| 809 | } |
| 810 | |
| 811 | inet->dport = usin->sin_port; |
| 812 | inet->daddr = daddr; |
| 813 | |
| 814 | tp->ext_header_len = 0; |
| 815 | if (inet->opt) |
| 816 | tp->ext_header_len = inet->opt->optlen; |
| 817 | |
| 818 | tp->rx_opt.mss_clamp = 536; |
| 819 | |
| 820 | /* Socket identity is still unknown (sport may be zero). |
| 821 | * However we set state to SYN-SENT and not releasing socket |
| 822 | * lock select source port, enter ourselves into the hash tables and |
| 823 | * complete initialization after this. |
| 824 | */ |
| 825 | tcp_set_state(sk, TCP_SYN_SENT); |
| 826 | err = tcp_v4_hash_connect(sk); |
| 827 | if (err) |
| 828 | goto failure; |
| 829 | |
| 830 | err = ip_route_newports(&rt, inet->sport, inet->dport, sk); |
| 831 | if (err) |
| 832 | goto failure; |
| 833 | |
| 834 | /* OK, now commit destination to socket. */ |
| 835 | __sk_dst_set(sk, &rt->u.dst); |
| 836 | tcp_v4_setup_caps(sk, &rt->u.dst); |
| 837 | |
| 838 | if (!tp->write_seq) |
| 839 | tp->write_seq = secure_tcp_sequence_number(inet->saddr, |
| 840 | inet->daddr, |
| 841 | inet->sport, |
| 842 | usin->sin_port); |
| 843 | |
| 844 | inet->id = tp->write_seq ^ jiffies; |
| 845 | |
| 846 | err = tcp_connect(sk); |
| 847 | rt = NULL; |
| 848 | if (err) |
| 849 | goto failure; |
| 850 | |
| 851 | return 0; |
| 852 | |
| 853 | failure: |
| 854 | /* This unhashes the socket and releases the local port, if necessary. */ |
| 855 | tcp_set_state(sk, TCP_CLOSE); |
| 856 | ip_rt_put(rt); |
| 857 | sk->sk_route_caps = 0; |
| 858 | inet->dport = 0; |
| 859 | return err; |
| 860 | } |
| 861 | |
| 862 | static __inline__ int tcp_v4_iif(struct sk_buff *skb) |
| 863 | { |
| 864 | return ((struct rtable *)skb->dst)->rt_iif; |
| 865 | } |
| 866 | |
| 867 | static __inline__ u32 tcp_v4_synq_hash(u32 raddr, u16 rport, u32 rnd) |
| 868 | { |
| 869 | return (jhash_2words(raddr, (u32) rport, rnd) & (TCP_SYNQ_HSIZE - 1)); |
| 870 | } |
| 871 | |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 872 | static struct request_sock *tcp_v4_search_req(struct tcp_sock *tp, |
| 873 | struct request_sock ***prevp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 874 | __u16 rport, |
| 875 | __u32 raddr, __u32 laddr) |
| 876 | { |
Arnaldo Carvalho de Melo | 2ad69c5 | 2005-06-18 22:48:55 -0700 | [diff] [blame] | 877 | struct listen_sock *lopt = tp->accept_queue.listen_opt; |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 878 | struct request_sock *req, **prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 879 | |
| 880 | for (prev = &lopt->syn_table[tcp_v4_synq_hash(raddr, rport, lopt->hash_rnd)]; |
| 881 | (req = *prev) != NULL; |
| 882 | prev = &req->dl_next) { |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 883 | const struct inet_request_sock *ireq = inet_rsk(req); |
| 884 | |
| 885 | if (ireq->rmt_port == rport && |
| 886 | ireq->rmt_addr == raddr && |
| 887 | ireq->loc_addr == laddr && |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 888 | TCP_INET_FAMILY(req->rsk_ops->family)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 889 | BUG_TRAP(!req->sk); |
| 890 | *prevp = prev; |
| 891 | break; |
| 892 | } |
| 893 | } |
| 894 | |
| 895 | return req; |
| 896 | } |
| 897 | |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 898 | static void tcp_v4_synq_add(struct sock *sk, struct request_sock *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 899 | { |
| 900 | struct tcp_sock *tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 2ad69c5 | 2005-06-18 22:48:55 -0700 | [diff] [blame] | 901 | struct listen_sock *lopt = tp->accept_queue.listen_opt; |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 902 | u32 h = tcp_v4_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port, lopt->hash_rnd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 903 | |
Arnaldo Carvalho de Melo | 0e87506 | 2005-06-18 22:47:59 -0700 | [diff] [blame] | 904 | reqsk_queue_hash_req(&tp->accept_queue, h, req, TCP_TIMEOUT_INIT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 905 | tcp_synq_added(sk); |
| 906 | } |
| 907 | |
| 908 | |
| 909 | /* |
| 910 | * This routine does path mtu discovery as defined in RFC1191. |
| 911 | */ |
| 912 | static inline void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, |
| 913 | u32 mtu) |
| 914 | { |
| 915 | struct dst_entry *dst; |
| 916 | struct inet_sock *inet = inet_sk(sk); |
| 917 | struct tcp_sock *tp = tcp_sk(sk); |
| 918 | |
| 919 | /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs |
| 920 | * send out by Linux are always <576bytes so they should go through |
| 921 | * unfragmented). |
| 922 | */ |
| 923 | if (sk->sk_state == TCP_LISTEN) |
| 924 | return; |
| 925 | |
| 926 | /* We don't check in the destentry if pmtu discovery is forbidden |
| 927 | * on this route. We just assume that no packet_to_big packets |
| 928 | * are send back when pmtu discovery is not active. |
| 929 | * There is a small race when the user changes this flag in the |
| 930 | * route, but I think that's acceptable. |
| 931 | */ |
| 932 | if ((dst = __sk_dst_check(sk, 0)) == NULL) |
| 933 | return; |
| 934 | |
| 935 | dst->ops->update_pmtu(dst, mtu); |
| 936 | |
| 937 | /* Something is about to be wrong... Remember soft error |
| 938 | * for the case, if this connection will not able to recover. |
| 939 | */ |
| 940 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) |
| 941 | sk->sk_err_soft = EMSGSIZE; |
| 942 | |
| 943 | mtu = dst_mtu(dst); |
| 944 | |
| 945 | if (inet->pmtudisc != IP_PMTUDISC_DONT && |
| 946 | tp->pmtu_cookie > mtu) { |
| 947 | tcp_sync_mss(sk, mtu); |
| 948 | |
| 949 | /* Resend the TCP packet because it's |
| 950 | * clear that the old packet has been |
| 951 | * dropped. This is the new "fast" path mtu |
| 952 | * discovery. |
| 953 | */ |
| 954 | tcp_simple_retransmit(sk); |
| 955 | } /* else let the usual retransmit timer handle it */ |
| 956 | } |
| 957 | |
| 958 | /* |
| 959 | * This routine is called by the ICMP module when it gets some |
| 960 | * sort of error condition. If err < 0 then the socket should |
| 961 | * be closed and the error returned to the user. If err > 0 |
| 962 | * it's just the icmp type << 8 | icmp code. After adjustment |
| 963 | * header points to the first 8 bytes of the tcp header. We need |
| 964 | * to find the appropriate port. |
| 965 | * |
| 966 | * The locking strategy used here is very "optimistic". When |
| 967 | * someone else accesses the socket the ICMP is just dropped |
| 968 | * and for some paths there is no check at all. |
| 969 | * A more general error queue to queue errors for later handling |
| 970 | * is probably better. |
| 971 | * |
| 972 | */ |
| 973 | |
| 974 | void tcp_v4_err(struct sk_buff *skb, u32 info) |
| 975 | { |
| 976 | struct iphdr *iph = (struct iphdr *)skb->data; |
| 977 | struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2)); |
| 978 | struct tcp_sock *tp; |
| 979 | struct inet_sock *inet; |
| 980 | int type = skb->h.icmph->type; |
| 981 | int code = skb->h.icmph->code; |
| 982 | struct sock *sk; |
| 983 | __u32 seq; |
| 984 | int err; |
| 985 | |
| 986 | if (skb->len < (iph->ihl << 2) + 8) { |
| 987 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); |
| 988 | return; |
| 989 | } |
| 990 | |
| 991 | sk = tcp_v4_lookup(iph->daddr, th->dest, iph->saddr, |
| 992 | th->source, tcp_v4_iif(skb)); |
| 993 | if (!sk) { |
| 994 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); |
| 995 | return; |
| 996 | } |
| 997 | if (sk->sk_state == TCP_TIME_WAIT) { |
| 998 | tcp_tw_put((struct tcp_tw_bucket *)sk); |
| 999 | return; |
| 1000 | } |
| 1001 | |
| 1002 | bh_lock_sock(sk); |
| 1003 | /* If too many ICMPs get dropped on busy |
| 1004 | * servers this needs to be solved differently. |
| 1005 | */ |
| 1006 | if (sock_owned_by_user(sk)) |
| 1007 | NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS); |
| 1008 | |
| 1009 | if (sk->sk_state == TCP_CLOSE) |
| 1010 | goto out; |
| 1011 | |
| 1012 | tp = tcp_sk(sk); |
| 1013 | seq = ntohl(th->seq); |
| 1014 | if (sk->sk_state != TCP_LISTEN && |
| 1015 | !between(seq, tp->snd_una, tp->snd_nxt)) { |
| 1016 | NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS); |
| 1017 | goto out; |
| 1018 | } |
| 1019 | |
| 1020 | switch (type) { |
| 1021 | case ICMP_SOURCE_QUENCH: |
| 1022 | /* Just silently ignore these. */ |
| 1023 | goto out; |
| 1024 | case ICMP_PARAMETERPROB: |
| 1025 | err = EPROTO; |
| 1026 | break; |
| 1027 | case ICMP_DEST_UNREACH: |
| 1028 | if (code > NR_ICMP_UNREACH) |
| 1029 | goto out; |
| 1030 | |
| 1031 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ |
| 1032 | if (!sock_owned_by_user(sk)) |
| 1033 | do_pmtu_discovery(sk, iph, info); |
| 1034 | goto out; |
| 1035 | } |
| 1036 | |
| 1037 | err = icmp_err_convert[code].errno; |
| 1038 | break; |
| 1039 | case ICMP_TIME_EXCEEDED: |
| 1040 | err = EHOSTUNREACH; |
| 1041 | break; |
| 1042 | default: |
| 1043 | goto out; |
| 1044 | } |
| 1045 | |
| 1046 | switch (sk->sk_state) { |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1047 | struct request_sock *req, **prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1048 | case TCP_LISTEN: |
| 1049 | if (sock_owned_by_user(sk)) |
| 1050 | goto out; |
| 1051 | |
| 1052 | req = tcp_v4_search_req(tp, &prev, th->dest, |
| 1053 | iph->daddr, iph->saddr); |
| 1054 | if (!req) |
| 1055 | goto out; |
| 1056 | |
| 1057 | /* ICMPs are not backlogged, hence we cannot get |
| 1058 | an established socket here. |
| 1059 | */ |
| 1060 | BUG_TRAP(!req->sk); |
| 1061 | |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1062 | if (seq != tcp_rsk(req)->snt_isn) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1063 | NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS); |
| 1064 | goto out; |
| 1065 | } |
| 1066 | |
| 1067 | /* |
| 1068 | * Still in SYN_RECV, just remove it silently. |
| 1069 | * There is no good way to pass the error to the newly |
| 1070 | * created socket, and POSIX does not want network |
| 1071 | * errors returned from accept(). |
| 1072 | */ |
| 1073 | tcp_synq_drop(sk, req, prev); |
| 1074 | goto out; |
| 1075 | |
| 1076 | case TCP_SYN_SENT: |
| 1077 | case TCP_SYN_RECV: /* Cannot happen. |
| 1078 | It can f.e. if SYNs crossed. |
| 1079 | */ |
| 1080 | if (!sock_owned_by_user(sk)) { |
| 1081 | TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS); |
| 1082 | sk->sk_err = err; |
| 1083 | |
| 1084 | sk->sk_error_report(sk); |
| 1085 | |
| 1086 | tcp_done(sk); |
| 1087 | } else { |
| 1088 | sk->sk_err_soft = err; |
| 1089 | } |
| 1090 | goto out; |
| 1091 | } |
| 1092 | |
| 1093 | /* If we've already connected we will keep trying |
| 1094 | * until we time out, or the user gives up. |
| 1095 | * |
| 1096 | * rfc1122 4.2.3.9 allows to consider as hard errors |
| 1097 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, |
| 1098 | * but it is obsoleted by pmtu discovery). |
| 1099 | * |
| 1100 | * Note, that in modern internet, where routing is unreliable |
| 1101 | * and in each dark corner broken firewalls sit, sending random |
| 1102 | * errors ordered by their masters even this two messages finally lose |
| 1103 | * their original sense (even Linux sends invalid PORT_UNREACHs) |
| 1104 | * |
| 1105 | * Now we are in compliance with RFCs. |
| 1106 | * --ANK (980905) |
| 1107 | */ |
| 1108 | |
| 1109 | inet = inet_sk(sk); |
| 1110 | if (!sock_owned_by_user(sk) && inet->recverr) { |
| 1111 | sk->sk_err = err; |
| 1112 | sk->sk_error_report(sk); |
| 1113 | } else { /* Only an error on timeout */ |
| 1114 | sk->sk_err_soft = err; |
| 1115 | } |
| 1116 | |
| 1117 | out: |
| 1118 | bh_unlock_sock(sk); |
| 1119 | sock_put(sk); |
| 1120 | } |
| 1121 | |
| 1122 | /* This routine computes an IPv4 TCP checksum. */ |
| 1123 | void tcp_v4_send_check(struct sock *sk, struct tcphdr *th, int len, |
| 1124 | struct sk_buff *skb) |
| 1125 | { |
| 1126 | struct inet_sock *inet = inet_sk(sk); |
| 1127 | |
| 1128 | if (skb->ip_summed == CHECKSUM_HW) { |
| 1129 | th->check = ~tcp_v4_check(th, len, inet->saddr, inet->daddr, 0); |
| 1130 | skb->csum = offsetof(struct tcphdr, check); |
| 1131 | } else { |
| 1132 | th->check = tcp_v4_check(th, len, inet->saddr, inet->daddr, |
| 1133 | csum_partial((char *)th, |
| 1134 | th->doff << 2, |
| 1135 | skb->csum)); |
| 1136 | } |
| 1137 | } |
| 1138 | |
| 1139 | /* |
| 1140 | * This routine will send an RST to the other tcp. |
| 1141 | * |
| 1142 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) |
| 1143 | * for reset. |
| 1144 | * Answer: if a packet caused RST, it is not for a socket |
| 1145 | * existing in our system, if it is matched to a socket, |
| 1146 | * it is just duplicate segment or bug in other side's TCP. |
| 1147 | * So that we build reply only basing on parameters |
| 1148 | * arrived with segment. |
| 1149 | * Exception: precedence violation. We do not implement it in any case. |
| 1150 | */ |
| 1151 | |
| 1152 | static void tcp_v4_send_reset(struct sk_buff *skb) |
| 1153 | { |
| 1154 | struct tcphdr *th = skb->h.th; |
| 1155 | struct tcphdr rth; |
| 1156 | struct ip_reply_arg arg; |
| 1157 | |
| 1158 | /* Never send a reset in response to a reset. */ |
| 1159 | if (th->rst) |
| 1160 | return; |
| 1161 | |
| 1162 | if (((struct rtable *)skb->dst)->rt_type != RTN_LOCAL) |
| 1163 | return; |
| 1164 | |
| 1165 | /* Swap the send and the receive. */ |
| 1166 | memset(&rth, 0, sizeof(struct tcphdr)); |
| 1167 | rth.dest = th->source; |
| 1168 | rth.source = th->dest; |
| 1169 | rth.doff = sizeof(struct tcphdr) / 4; |
| 1170 | rth.rst = 1; |
| 1171 | |
| 1172 | if (th->ack) { |
| 1173 | rth.seq = th->ack_seq; |
| 1174 | } else { |
| 1175 | rth.ack = 1; |
| 1176 | rth.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + |
| 1177 | skb->len - (th->doff << 2)); |
| 1178 | } |
| 1179 | |
| 1180 | memset(&arg, 0, sizeof arg); |
| 1181 | arg.iov[0].iov_base = (unsigned char *)&rth; |
| 1182 | arg.iov[0].iov_len = sizeof rth; |
| 1183 | arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, |
| 1184 | skb->nh.iph->saddr, /*XXX*/ |
| 1185 | sizeof(struct tcphdr), IPPROTO_TCP, 0); |
| 1186 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; |
| 1187 | |
| 1188 | ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth); |
| 1189 | |
| 1190 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); |
| 1191 | TCP_INC_STATS_BH(TCP_MIB_OUTRSTS); |
| 1192 | } |
| 1193 | |
| 1194 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states |
| 1195 | outside socket context is ugly, certainly. What can I do? |
| 1196 | */ |
| 1197 | |
| 1198 | static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, |
| 1199 | u32 win, u32 ts) |
| 1200 | { |
| 1201 | struct tcphdr *th = skb->h.th; |
| 1202 | struct { |
| 1203 | struct tcphdr th; |
| 1204 | u32 tsopt[3]; |
| 1205 | } rep; |
| 1206 | struct ip_reply_arg arg; |
| 1207 | |
| 1208 | memset(&rep.th, 0, sizeof(struct tcphdr)); |
| 1209 | memset(&arg, 0, sizeof arg); |
| 1210 | |
| 1211 | arg.iov[0].iov_base = (unsigned char *)&rep; |
| 1212 | arg.iov[0].iov_len = sizeof(rep.th); |
| 1213 | if (ts) { |
| 1214 | rep.tsopt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
| 1215 | (TCPOPT_TIMESTAMP << 8) | |
| 1216 | TCPOLEN_TIMESTAMP); |
| 1217 | rep.tsopt[1] = htonl(tcp_time_stamp); |
| 1218 | rep.tsopt[2] = htonl(ts); |
| 1219 | arg.iov[0].iov_len = sizeof(rep); |
| 1220 | } |
| 1221 | |
| 1222 | /* Swap the send and the receive. */ |
| 1223 | rep.th.dest = th->source; |
| 1224 | rep.th.source = th->dest; |
| 1225 | rep.th.doff = arg.iov[0].iov_len / 4; |
| 1226 | rep.th.seq = htonl(seq); |
| 1227 | rep.th.ack_seq = htonl(ack); |
| 1228 | rep.th.ack = 1; |
| 1229 | rep.th.window = htons(win); |
| 1230 | |
| 1231 | arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, |
| 1232 | skb->nh.iph->saddr, /*XXX*/ |
| 1233 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
| 1234 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; |
| 1235 | |
| 1236 | ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len); |
| 1237 | |
| 1238 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); |
| 1239 | } |
| 1240 | |
| 1241 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) |
| 1242 | { |
| 1243 | struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk; |
| 1244 | |
| 1245 | tcp_v4_send_ack(skb, tw->tw_snd_nxt, tw->tw_rcv_nxt, |
| 1246 | tw->tw_rcv_wnd >> tw->tw_rcv_wscale, tw->tw_ts_recent); |
| 1247 | |
| 1248 | tcp_tw_put(tw); |
| 1249 | } |
| 1250 | |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1251 | static void tcp_v4_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1252 | { |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1253 | tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1254 | req->ts_recent); |
| 1255 | } |
| 1256 | |
| 1257 | static struct dst_entry* tcp_v4_route_req(struct sock *sk, |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1258 | struct request_sock *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1259 | { |
| 1260 | struct rtable *rt; |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1261 | const struct inet_request_sock *ireq = inet_rsk(req); |
| 1262 | struct ip_options *opt = inet_rsk(req)->opt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1263 | struct flowi fl = { .oif = sk->sk_bound_dev_if, |
| 1264 | .nl_u = { .ip4_u = |
| 1265 | { .daddr = ((opt && opt->srr) ? |
| 1266 | opt->faddr : |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1267 | ireq->rmt_addr), |
| 1268 | .saddr = ireq->loc_addr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1269 | .tos = RT_CONN_FLAGS(sk) } }, |
| 1270 | .proto = IPPROTO_TCP, |
| 1271 | .uli_u = { .ports = |
| 1272 | { .sport = inet_sk(sk)->sport, |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1273 | .dport = ireq->rmt_port } } }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1274 | |
| 1275 | if (ip_route_output_flow(&rt, &fl, sk, 0)) { |
| 1276 | IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); |
| 1277 | return NULL; |
| 1278 | } |
| 1279 | if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) { |
| 1280 | ip_rt_put(rt); |
| 1281 | IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); |
| 1282 | return NULL; |
| 1283 | } |
| 1284 | return &rt->u.dst; |
| 1285 | } |
| 1286 | |
| 1287 | /* |
| 1288 | * Send a SYN-ACK after having received an ACK. |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1289 | * This still operates on a request_sock only, not on a big |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1290 | * socket. |
| 1291 | */ |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1292 | static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | struct dst_entry *dst) |
| 1294 | { |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1295 | const struct inet_request_sock *ireq = inet_rsk(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1296 | int err = -1; |
| 1297 | struct sk_buff * skb; |
| 1298 | |
| 1299 | /* First, grab a route. */ |
| 1300 | if (!dst && (dst = tcp_v4_route_req(sk, req)) == NULL) |
| 1301 | goto out; |
| 1302 | |
| 1303 | skb = tcp_make_synack(sk, dst, req); |
| 1304 | |
| 1305 | if (skb) { |
| 1306 | struct tcphdr *th = skb->h.th; |
| 1307 | |
| 1308 | th->check = tcp_v4_check(th, skb->len, |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1309 | ireq->loc_addr, |
| 1310 | ireq->rmt_addr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1311 | csum_partial((char *)th, skb->len, |
| 1312 | skb->csum)); |
| 1313 | |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1314 | err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr, |
| 1315 | ireq->rmt_addr, |
| 1316 | ireq->opt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1317 | if (err == NET_XMIT_CN) |
| 1318 | err = 0; |
| 1319 | } |
| 1320 | |
| 1321 | out: |
| 1322 | dst_release(dst); |
| 1323 | return err; |
| 1324 | } |
| 1325 | |
| 1326 | /* |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1327 | * IPv4 request_sock destructor. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1328 | */ |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1329 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1330 | { |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1331 | if (inet_rsk(req)->opt) |
| 1332 | kfree(inet_rsk(req)->opt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | } |
| 1334 | |
| 1335 | static inline void syn_flood_warning(struct sk_buff *skb) |
| 1336 | { |
| 1337 | static unsigned long warntime; |
| 1338 | |
| 1339 | if (time_after(jiffies, (warntime + HZ * 60))) { |
| 1340 | warntime = jiffies; |
| 1341 | printk(KERN_INFO |
| 1342 | "possible SYN flooding on port %d. Sending cookies.\n", |
| 1343 | ntohs(skb->h.th->dest)); |
| 1344 | } |
| 1345 | } |
| 1346 | |
| 1347 | /* |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1348 | * Save and compile IPv4 options into the request_sock if needed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1349 | */ |
| 1350 | static inline struct ip_options *tcp_v4_save_options(struct sock *sk, |
| 1351 | struct sk_buff *skb) |
| 1352 | { |
| 1353 | struct ip_options *opt = &(IPCB(skb)->opt); |
| 1354 | struct ip_options *dopt = NULL; |
| 1355 | |
| 1356 | if (opt && opt->optlen) { |
| 1357 | int opt_size = optlength(opt); |
| 1358 | dopt = kmalloc(opt_size, GFP_ATOMIC); |
| 1359 | if (dopt) { |
| 1360 | if (ip_options_echo(dopt, skb)) { |
| 1361 | kfree(dopt); |
| 1362 | dopt = NULL; |
| 1363 | } |
| 1364 | } |
| 1365 | } |
| 1366 | return dopt; |
| 1367 | } |
| 1368 | |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1369 | struct request_sock_ops tcp_request_sock_ops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | .family = PF_INET, |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1371 | .obj_size = sizeof(struct tcp_request_sock), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | .rtx_syn_ack = tcp_v4_send_synack, |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1373 | .send_ack = tcp_v4_reqsk_send_ack, |
| 1374 | .destructor = tcp_v4_reqsk_destructor, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1375 | .send_reset = tcp_v4_send_reset, |
| 1376 | }; |
| 1377 | |
| 1378 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) |
| 1379 | { |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1380 | struct inet_request_sock *ireq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1381 | struct tcp_options_received tmp_opt; |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1382 | struct request_sock *req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1383 | __u32 saddr = skb->nh.iph->saddr; |
| 1384 | __u32 daddr = skb->nh.iph->daddr; |
| 1385 | __u32 isn = TCP_SKB_CB(skb)->when; |
| 1386 | struct dst_entry *dst = NULL; |
| 1387 | #ifdef CONFIG_SYN_COOKIES |
| 1388 | int want_cookie = 0; |
| 1389 | #else |
| 1390 | #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */ |
| 1391 | #endif |
| 1392 | |
| 1393 | /* Never answer to SYNs send to broadcast or multicast */ |
| 1394 | if (((struct rtable *)skb->dst)->rt_flags & |
| 1395 | (RTCF_BROADCAST | RTCF_MULTICAST)) |
| 1396 | goto drop; |
| 1397 | |
| 1398 | /* TW buckets are converted to open requests without |
| 1399 | * limitations, they conserve resources and peer is |
| 1400 | * evidently real one. |
| 1401 | */ |
| 1402 | if (tcp_synq_is_full(sk) && !isn) { |
| 1403 | #ifdef CONFIG_SYN_COOKIES |
| 1404 | if (sysctl_tcp_syncookies) { |
| 1405 | want_cookie = 1; |
| 1406 | } else |
| 1407 | #endif |
| 1408 | goto drop; |
| 1409 | } |
| 1410 | |
| 1411 | /* Accept backlog is full. If we have already queued enough |
| 1412 | * of warm entries in syn queue, drop request. It is better than |
| 1413 | * clogging syn queue with openreqs with exponentially increasing |
| 1414 | * timeout. |
| 1415 | */ |
| 1416 | if (sk_acceptq_is_full(sk) && tcp_synq_young(sk) > 1) |
| 1417 | goto drop; |
| 1418 | |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1419 | req = reqsk_alloc(&tcp_request_sock_ops); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1420 | if (!req) |
| 1421 | goto drop; |
| 1422 | |
| 1423 | tcp_clear_options(&tmp_opt); |
| 1424 | tmp_opt.mss_clamp = 536; |
| 1425 | tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss; |
| 1426 | |
| 1427 | tcp_parse_options(skb, &tmp_opt, 0); |
| 1428 | |
| 1429 | if (want_cookie) { |
| 1430 | tcp_clear_options(&tmp_opt); |
| 1431 | tmp_opt.saw_tstamp = 0; |
| 1432 | } |
| 1433 | |
| 1434 | if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) { |
| 1435 | /* Some OSes (unknown ones, but I see them on web server, which |
| 1436 | * contains information interesting only for windows' |
| 1437 | * users) do not send their stamp in SYN. It is easy case. |
| 1438 | * We simply do not advertise TS support. |
| 1439 | */ |
| 1440 | tmp_opt.saw_tstamp = 0; |
| 1441 | tmp_opt.tstamp_ok = 0; |
| 1442 | } |
| 1443 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; |
| 1444 | |
| 1445 | tcp_openreq_init(req, &tmp_opt, skb); |
| 1446 | |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1447 | ireq = inet_rsk(req); |
| 1448 | ireq->loc_addr = daddr; |
| 1449 | ireq->rmt_addr = saddr; |
| 1450 | ireq->opt = tcp_v4_save_options(sk, skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1451 | if (!want_cookie) |
| 1452 | TCP_ECN_create_request(req, skb->h.th); |
| 1453 | |
| 1454 | if (want_cookie) { |
| 1455 | #ifdef CONFIG_SYN_COOKIES |
| 1456 | syn_flood_warning(skb); |
| 1457 | #endif |
| 1458 | isn = cookie_v4_init_sequence(sk, skb, &req->mss); |
| 1459 | } else if (!isn) { |
| 1460 | struct inet_peer *peer = NULL; |
| 1461 | |
| 1462 | /* VJ's idea. We save last timestamp seen |
| 1463 | * from the destination in peer table, when entering |
| 1464 | * state TIME-WAIT, and check against it before |
| 1465 | * accepting new connection request. |
| 1466 | * |
| 1467 | * If "isn" is not zero, this request hit alive |
| 1468 | * timewait bucket, so that all the necessary checks |
| 1469 | * are made in the function processing timewait state. |
| 1470 | */ |
| 1471 | if (tmp_opt.saw_tstamp && |
| 1472 | sysctl_tcp_tw_recycle && |
| 1473 | (dst = tcp_v4_route_req(sk, req)) != NULL && |
| 1474 | (peer = rt_get_peer((struct rtable *)dst)) != NULL && |
| 1475 | peer->v4daddr == saddr) { |
| 1476 | if (xtime.tv_sec < peer->tcp_ts_stamp + TCP_PAWS_MSL && |
| 1477 | (s32)(peer->tcp_ts - req->ts_recent) > |
| 1478 | TCP_PAWS_WINDOW) { |
| 1479 | NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED); |
| 1480 | dst_release(dst); |
| 1481 | goto drop_and_free; |
| 1482 | } |
| 1483 | } |
| 1484 | /* Kill the following clause, if you dislike this way. */ |
| 1485 | else if (!sysctl_tcp_syncookies && |
| 1486 | (sysctl_max_syn_backlog - tcp_synq_len(sk) < |
| 1487 | (sysctl_max_syn_backlog >> 2)) && |
| 1488 | (!peer || !peer->tcp_ts_stamp) && |
| 1489 | (!dst || !dst_metric(dst, RTAX_RTT))) { |
| 1490 | /* Without syncookies last quarter of |
| 1491 | * backlog is filled with destinations, |
| 1492 | * proven to be alive. |
| 1493 | * It means that we continue to communicate |
| 1494 | * to destinations, already remembered |
| 1495 | * to the moment of synflood. |
| 1496 | */ |
Heikki Orsila | ca93345 | 2005-08-08 14:26:52 -0700 | [diff] [blame] | 1497 | LIMIT_NETDEBUG(printk(KERN_DEBUG "TCP: drop open " |
| 1498 | "request from %u.%u." |
| 1499 | "%u.%u/%u\n", |
| 1500 | NIPQUAD(saddr), |
| 1501 | ntohs(skb->h.th->source))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | dst_release(dst); |
| 1503 | goto drop_and_free; |
| 1504 | } |
| 1505 | |
| 1506 | isn = tcp_v4_init_sequence(sk, skb); |
| 1507 | } |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1508 | tcp_rsk(req)->snt_isn = isn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1509 | |
| 1510 | if (tcp_v4_send_synack(sk, req, dst)) |
| 1511 | goto drop_and_free; |
| 1512 | |
| 1513 | if (want_cookie) { |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1514 | reqsk_free(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1515 | } else { |
| 1516 | tcp_v4_synq_add(sk, req); |
| 1517 | } |
| 1518 | return 0; |
| 1519 | |
| 1520 | drop_and_free: |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1521 | reqsk_free(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1522 | drop: |
| 1523 | TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS); |
| 1524 | return 0; |
| 1525 | } |
| 1526 | |
| 1527 | |
| 1528 | /* |
| 1529 | * The three way handshake has completed - we got a valid synack - |
| 1530 | * now create the new socket. |
| 1531 | */ |
| 1532 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1533 | struct request_sock *req, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1534 | struct dst_entry *dst) |
| 1535 | { |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1536 | struct inet_request_sock *ireq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1537 | struct inet_sock *newinet; |
| 1538 | struct tcp_sock *newtp; |
| 1539 | struct sock *newsk; |
| 1540 | |
| 1541 | if (sk_acceptq_is_full(sk)) |
| 1542 | goto exit_overflow; |
| 1543 | |
| 1544 | if (!dst && (dst = tcp_v4_route_req(sk, req)) == NULL) |
| 1545 | goto exit; |
| 1546 | |
| 1547 | newsk = tcp_create_openreq_child(sk, req, skb); |
| 1548 | if (!newsk) |
| 1549 | goto exit; |
| 1550 | |
| 1551 | newsk->sk_dst_cache = dst; |
| 1552 | tcp_v4_setup_caps(newsk, dst); |
| 1553 | |
| 1554 | newtp = tcp_sk(newsk); |
| 1555 | newinet = inet_sk(newsk); |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 1556 | ireq = inet_rsk(req); |
| 1557 | newinet->daddr = ireq->rmt_addr; |
| 1558 | newinet->rcv_saddr = ireq->loc_addr; |
| 1559 | newinet->saddr = ireq->loc_addr; |
| 1560 | newinet->opt = ireq->opt; |
| 1561 | ireq->opt = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1562 | newinet->mc_index = tcp_v4_iif(skb); |
| 1563 | newinet->mc_ttl = skb->nh.iph->ttl; |
| 1564 | newtp->ext_header_len = 0; |
| 1565 | if (newinet->opt) |
| 1566 | newtp->ext_header_len = newinet->opt->optlen; |
| 1567 | newinet->id = newtp->write_seq ^ jiffies; |
| 1568 | |
| 1569 | tcp_sync_mss(newsk, dst_mtu(dst)); |
| 1570 | newtp->advmss = dst_metric(dst, RTAX_ADVMSS); |
| 1571 | tcp_initialize_rcv_mss(newsk); |
| 1572 | |
| 1573 | __tcp_v4_hash(newsk, 0); |
| 1574 | __tcp_inherit_port(sk, newsk); |
| 1575 | |
| 1576 | return newsk; |
| 1577 | |
| 1578 | exit_overflow: |
| 1579 | NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS); |
| 1580 | exit: |
| 1581 | NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS); |
| 1582 | dst_release(dst); |
| 1583 | return NULL; |
| 1584 | } |
| 1585 | |
| 1586 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) |
| 1587 | { |
| 1588 | struct tcphdr *th = skb->h.th; |
| 1589 | struct iphdr *iph = skb->nh.iph; |
| 1590 | struct tcp_sock *tp = tcp_sk(sk); |
| 1591 | struct sock *nsk; |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1592 | struct request_sock **prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1593 | /* Find possible connection requests. */ |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 1594 | struct request_sock *req = tcp_v4_search_req(tp, &prev, th->source, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | iph->saddr, iph->daddr); |
| 1596 | if (req) |
| 1597 | return tcp_check_req(sk, skb, req, prev); |
| 1598 | |
| 1599 | nsk = __tcp_v4_lookup_established(skb->nh.iph->saddr, |
| 1600 | th->source, |
| 1601 | skb->nh.iph->daddr, |
| 1602 | ntohs(th->dest), |
| 1603 | tcp_v4_iif(skb)); |
| 1604 | |
| 1605 | if (nsk) { |
| 1606 | if (nsk->sk_state != TCP_TIME_WAIT) { |
| 1607 | bh_lock_sock(nsk); |
| 1608 | return nsk; |
| 1609 | } |
| 1610 | tcp_tw_put((struct tcp_tw_bucket *)nsk); |
| 1611 | return NULL; |
| 1612 | } |
| 1613 | |
| 1614 | #ifdef CONFIG_SYN_COOKIES |
| 1615 | if (!th->rst && !th->syn && th->ack) |
| 1616 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); |
| 1617 | #endif |
| 1618 | return sk; |
| 1619 | } |
| 1620 | |
| 1621 | static int tcp_v4_checksum_init(struct sk_buff *skb) |
| 1622 | { |
| 1623 | if (skb->ip_summed == CHECKSUM_HW) { |
| 1624 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
| 1625 | if (!tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr, |
| 1626 | skb->nh.iph->daddr, skb->csum)) |
| 1627 | return 0; |
| 1628 | |
Heikki Orsila | ca93345 | 2005-08-08 14:26:52 -0700 | [diff] [blame] | 1629 | LIMIT_NETDEBUG(printk(KERN_DEBUG "hw tcp v4 csum failed\n")); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1630 | skb->ip_summed = CHECKSUM_NONE; |
| 1631 | } |
| 1632 | if (skb->len <= 76) { |
| 1633 | if (tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr, |
| 1634 | skb->nh.iph->daddr, |
| 1635 | skb_checksum(skb, 0, skb->len, 0))) |
| 1636 | return -1; |
| 1637 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
| 1638 | } else { |
| 1639 | skb->csum = ~tcp_v4_check(skb->h.th, skb->len, |
| 1640 | skb->nh.iph->saddr, |
| 1641 | skb->nh.iph->daddr, 0); |
| 1642 | } |
| 1643 | return 0; |
| 1644 | } |
| 1645 | |
| 1646 | |
| 1647 | /* The socket must have it's spinlock held when we get |
| 1648 | * here. |
| 1649 | * |
| 1650 | * We have a potential double-lock case here, so even when |
| 1651 | * doing backlog processing we use the BH locking scheme. |
| 1652 | * This is because we cannot sleep with the original spinlock |
| 1653 | * held. |
| 1654 | */ |
| 1655 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) |
| 1656 | { |
| 1657 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
| 1658 | TCP_CHECK_TIMER(sk); |
| 1659 | if (tcp_rcv_established(sk, skb, skb->h.th, skb->len)) |
| 1660 | goto reset; |
| 1661 | TCP_CHECK_TIMER(sk); |
| 1662 | return 0; |
| 1663 | } |
| 1664 | |
| 1665 | if (skb->len < (skb->h.th->doff << 2) || tcp_checksum_complete(skb)) |
| 1666 | goto csum_err; |
| 1667 | |
| 1668 | if (sk->sk_state == TCP_LISTEN) { |
| 1669 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); |
| 1670 | if (!nsk) |
| 1671 | goto discard; |
| 1672 | |
| 1673 | if (nsk != sk) { |
| 1674 | if (tcp_child_process(sk, nsk, skb)) |
| 1675 | goto reset; |
| 1676 | return 0; |
| 1677 | } |
| 1678 | } |
| 1679 | |
| 1680 | TCP_CHECK_TIMER(sk); |
| 1681 | if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len)) |
| 1682 | goto reset; |
| 1683 | TCP_CHECK_TIMER(sk); |
| 1684 | return 0; |
| 1685 | |
| 1686 | reset: |
| 1687 | tcp_v4_send_reset(skb); |
| 1688 | discard: |
| 1689 | kfree_skb(skb); |
| 1690 | /* Be careful here. If this function gets more complicated and |
| 1691 | * gcc suffers from register pressure on the x86, sk (in %ebx) |
| 1692 | * might be destroyed here. This current version compiles correctly, |
| 1693 | * but you have been warned. |
| 1694 | */ |
| 1695 | return 0; |
| 1696 | |
| 1697 | csum_err: |
| 1698 | TCP_INC_STATS_BH(TCP_MIB_INERRS); |
| 1699 | goto discard; |
| 1700 | } |
| 1701 | |
| 1702 | /* |
| 1703 | * From tcp_input.c |
| 1704 | */ |
| 1705 | |
| 1706 | int tcp_v4_rcv(struct sk_buff *skb) |
| 1707 | { |
| 1708 | struct tcphdr *th; |
| 1709 | struct sock *sk; |
| 1710 | int ret; |
| 1711 | |
| 1712 | if (skb->pkt_type != PACKET_HOST) |
| 1713 | goto discard_it; |
| 1714 | |
| 1715 | /* Count it even if it's bad */ |
| 1716 | TCP_INC_STATS_BH(TCP_MIB_INSEGS); |
| 1717 | |
| 1718 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) |
| 1719 | goto discard_it; |
| 1720 | |
| 1721 | th = skb->h.th; |
| 1722 | |
| 1723 | if (th->doff < sizeof(struct tcphdr) / 4) |
| 1724 | goto bad_packet; |
| 1725 | if (!pskb_may_pull(skb, th->doff * 4)) |
| 1726 | goto discard_it; |
| 1727 | |
| 1728 | /* An explanation is required here, I think. |
| 1729 | * Packet length and doff are validated by header prediction, |
| 1730 | * provided case of th->doff==0 is elimineted. |
| 1731 | * So, we defer the checks. */ |
| 1732 | if ((skb->ip_summed != CHECKSUM_UNNECESSARY && |
| 1733 | tcp_v4_checksum_init(skb) < 0)) |
| 1734 | goto bad_packet; |
| 1735 | |
| 1736 | th = skb->h.th; |
| 1737 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); |
| 1738 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + |
| 1739 | skb->len - th->doff * 4); |
| 1740 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); |
| 1741 | TCP_SKB_CB(skb)->when = 0; |
| 1742 | TCP_SKB_CB(skb)->flags = skb->nh.iph->tos; |
| 1743 | TCP_SKB_CB(skb)->sacked = 0; |
| 1744 | |
| 1745 | sk = __tcp_v4_lookup(skb->nh.iph->saddr, th->source, |
| 1746 | skb->nh.iph->daddr, ntohs(th->dest), |
| 1747 | tcp_v4_iif(skb)); |
| 1748 | |
| 1749 | if (!sk) |
| 1750 | goto no_tcp_socket; |
| 1751 | |
| 1752 | process: |
| 1753 | if (sk->sk_state == TCP_TIME_WAIT) |
| 1754 | goto do_time_wait; |
| 1755 | |
| 1756 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) |
| 1757 | goto discard_and_relse; |
| 1758 | |
| 1759 | if (sk_filter(sk, skb, 0)) |
| 1760 | goto discard_and_relse; |
| 1761 | |
| 1762 | skb->dev = NULL; |
| 1763 | |
| 1764 | bh_lock_sock(sk); |
| 1765 | ret = 0; |
| 1766 | if (!sock_owned_by_user(sk)) { |
| 1767 | if (!tcp_prequeue(sk, skb)) |
| 1768 | ret = tcp_v4_do_rcv(sk, skb); |
| 1769 | } else |
| 1770 | sk_add_backlog(sk, skb); |
| 1771 | bh_unlock_sock(sk); |
| 1772 | |
| 1773 | sock_put(sk); |
| 1774 | |
| 1775 | return ret; |
| 1776 | |
| 1777 | no_tcp_socket: |
| 1778 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) |
| 1779 | goto discard_it; |
| 1780 | |
| 1781 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { |
| 1782 | bad_packet: |
| 1783 | TCP_INC_STATS_BH(TCP_MIB_INERRS); |
| 1784 | } else { |
| 1785 | tcp_v4_send_reset(skb); |
| 1786 | } |
| 1787 | |
| 1788 | discard_it: |
| 1789 | /* Discard frame. */ |
| 1790 | kfree_skb(skb); |
| 1791 | return 0; |
| 1792 | |
| 1793 | discard_and_relse: |
| 1794 | sock_put(sk); |
| 1795 | goto discard_it; |
| 1796 | |
| 1797 | do_time_wait: |
| 1798 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { |
| 1799 | tcp_tw_put((struct tcp_tw_bucket *) sk); |
| 1800 | goto discard_it; |
| 1801 | } |
| 1802 | |
| 1803 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { |
| 1804 | TCP_INC_STATS_BH(TCP_MIB_INERRS); |
| 1805 | tcp_tw_put((struct tcp_tw_bucket *) sk); |
| 1806 | goto discard_it; |
| 1807 | } |
| 1808 | switch (tcp_timewait_state_process((struct tcp_tw_bucket *)sk, |
| 1809 | skb, th, skb->len)) { |
| 1810 | case TCP_TW_SYN: { |
| 1811 | struct sock *sk2 = tcp_v4_lookup_listener(skb->nh.iph->daddr, |
| 1812 | ntohs(th->dest), |
| 1813 | tcp_v4_iif(skb)); |
| 1814 | if (sk2) { |
| 1815 | tcp_tw_deschedule((struct tcp_tw_bucket *)sk); |
| 1816 | tcp_tw_put((struct tcp_tw_bucket *)sk); |
| 1817 | sk = sk2; |
| 1818 | goto process; |
| 1819 | } |
| 1820 | /* Fall through to ACK */ |
| 1821 | } |
| 1822 | case TCP_TW_ACK: |
| 1823 | tcp_v4_timewait_ack(sk, skb); |
| 1824 | break; |
| 1825 | case TCP_TW_RST: |
| 1826 | goto no_tcp_socket; |
| 1827 | case TCP_TW_SUCCESS:; |
| 1828 | } |
| 1829 | goto discard_it; |
| 1830 | } |
| 1831 | |
| 1832 | /* With per-bucket locks this operation is not-atomic, so that |
| 1833 | * this version is not worse. |
| 1834 | */ |
| 1835 | static void __tcp_v4_rehash(struct sock *sk) |
| 1836 | { |
| 1837 | sk->sk_prot->unhash(sk); |
| 1838 | sk->sk_prot->hash(sk); |
| 1839 | } |
| 1840 | |
| 1841 | static int tcp_v4_reselect_saddr(struct sock *sk) |
| 1842 | { |
| 1843 | struct inet_sock *inet = inet_sk(sk); |
| 1844 | int err; |
| 1845 | struct rtable *rt; |
| 1846 | __u32 old_saddr = inet->saddr; |
| 1847 | __u32 new_saddr; |
| 1848 | __u32 daddr = inet->daddr; |
| 1849 | |
| 1850 | if (inet->opt && inet->opt->srr) |
| 1851 | daddr = inet->opt->faddr; |
| 1852 | |
| 1853 | /* Query new route. */ |
| 1854 | err = ip_route_connect(&rt, daddr, 0, |
| 1855 | RT_CONN_FLAGS(sk), |
| 1856 | sk->sk_bound_dev_if, |
| 1857 | IPPROTO_TCP, |
| 1858 | inet->sport, inet->dport, sk); |
| 1859 | if (err) |
| 1860 | return err; |
| 1861 | |
| 1862 | __sk_dst_set(sk, &rt->u.dst); |
| 1863 | tcp_v4_setup_caps(sk, &rt->u.dst); |
| 1864 | |
| 1865 | new_saddr = rt->rt_src; |
| 1866 | |
| 1867 | if (new_saddr == old_saddr) |
| 1868 | return 0; |
| 1869 | |
| 1870 | if (sysctl_ip_dynaddr > 1) { |
| 1871 | printk(KERN_INFO "tcp_v4_rebuild_header(): shifting inet->" |
| 1872 | "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n", |
| 1873 | NIPQUAD(old_saddr), |
| 1874 | NIPQUAD(new_saddr)); |
| 1875 | } |
| 1876 | |
| 1877 | inet->saddr = new_saddr; |
| 1878 | inet->rcv_saddr = new_saddr; |
| 1879 | |
| 1880 | /* XXX The only one ugly spot where we need to |
| 1881 | * XXX really change the sockets identity after |
| 1882 | * XXX it has entered the hashes. -DaveM |
| 1883 | * |
| 1884 | * Besides that, it does not check for connection |
| 1885 | * uniqueness. Wait for troubles. |
| 1886 | */ |
| 1887 | __tcp_v4_rehash(sk); |
| 1888 | return 0; |
| 1889 | } |
| 1890 | |
| 1891 | int tcp_v4_rebuild_header(struct sock *sk) |
| 1892 | { |
| 1893 | struct inet_sock *inet = inet_sk(sk); |
| 1894 | struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); |
| 1895 | u32 daddr; |
| 1896 | int err; |
| 1897 | |
| 1898 | /* Route is OK, nothing to do. */ |
| 1899 | if (rt) |
| 1900 | return 0; |
| 1901 | |
| 1902 | /* Reroute. */ |
| 1903 | daddr = inet->daddr; |
| 1904 | if (inet->opt && inet->opt->srr) |
| 1905 | daddr = inet->opt->faddr; |
| 1906 | |
| 1907 | { |
| 1908 | struct flowi fl = { .oif = sk->sk_bound_dev_if, |
| 1909 | .nl_u = { .ip4_u = |
| 1910 | { .daddr = daddr, |
| 1911 | .saddr = inet->saddr, |
| 1912 | .tos = RT_CONN_FLAGS(sk) } }, |
| 1913 | .proto = IPPROTO_TCP, |
| 1914 | .uli_u = { .ports = |
| 1915 | { .sport = inet->sport, |
| 1916 | .dport = inet->dport } } }; |
| 1917 | |
| 1918 | err = ip_route_output_flow(&rt, &fl, sk, 0); |
| 1919 | } |
| 1920 | if (!err) { |
| 1921 | __sk_dst_set(sk, &rt->u.dst); |
| 1922 | tcp_v4_setup_caps(sk, &rt->u.dst); |
| 1923 | return 0; |
| 1924 | } |
| 1925 | |
| 1926 | /* Routing failed... */ |
| 1927 | sk->sk_route_caps = 0; |
| 1928 | |
| 1929 | if (!sysctl_ip_dynaddr || |
| 1930 | sk->sk_state != TCP_SYN_SENT || |
| 1931 | (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || |
| 1932 | (err = tcp_v4_reselect_saddr(sk)) != 0) |
| 1933 | sk->sk_err_soft = -err; |
| 1934 | |
| 1935 | return err; |
| 1936 | } |
| 1937 | |
| 1938 | static void v4_addr2sockaddr(struct sock *sk, struct sockaddr * uaddr) |
| 1939 | { |
| 1940 | struct sockaddr_in *sin = (struct sockaddr_in *) uaddr; |
| 1941 | struct inet_sock *inet = inet_sk(sk); |
| 1942 | |
| 1943 | sin->sin_family = AF_INET; |
| 1944 | sin->sin_addr.s_addr = inet->daddr; |
| 1945 | sin->sin_port = inet->dport; |
| 1946 | } |
| 1947 | |
| 1948 | /* VJ's idea. Save last timestamp seen from this destination |
| 1949 | * and hold it at least for normal timewait interval to use for duplicate |
| 1950 | * segment detection in subsequent connections, before they enter synchronized |
| 1951 | * state. |
| 1952 | */ |
| 1953 | |
| 1954 | int tcp_v4_remember_stamp(struct sock *sk) |
| 1955 | { |
| 1956 | struct inet_sock *inet = inet_sk(sk); |
| 1957 | struct tcp_sock *tp = tcp_sk(sk); |
| 1958 | struct rtable *rt = (struct rtable *)__sk_dst_get(sk); |
| 1959 | struct inet_peer *peer = NULL; |
| 1960 | int release_it = 0; |
| 1961 | |
| 1962 | if (!rt || rt->rt_dst != inet->daddr) { |
| 1963 | peer = inet_getpeer(inet->daddr, 1); |
| 1964 | release_it = 1; |
| 1965 | } else { |
| 1966 | if (!rt->peer) |
| 1967 | rt_bind_peer(rt, 1); |
| 1968 | peer = rt->peer; |
| 1969 | } |
| 1970 | |
| 1971 | if (peer) { |
| 1972 | if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 || |
| 1973 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && |
| 1974 | peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) { |
| 1975 | peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp; |
| 1976 | peer->tcp_ts = tp->rx_opt.ts_recent; |
| 1977 | } |
| 1978 | if (release_it) |
| 1979 | inet_putpeer(peer); |
| 1980 | return 1; |
| 1981 | } |
| 1982 | |
| 1983 | return 0; |
| 1984 | } |
| 1985 | |
| 1986 | int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw) |
| 1987 | { |
| 1988 | struct inet_peer *peer = NULL; |
| 1989 | |
| 1990 | peer = inet_getpeer(tw->tw_daddr, 1); |
| 1991 | |
| 1992 | if (peer) { |
| 1993 | if ((s32)(peer->tcp_ts - tw->tw_ts_recent) <= 0 || |
| 1994 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && |
| 1995 | peer->tcp_ts_stamp <= tw->tw_ts_recent_stamp)) { |
| 1996 | peer->tcp_ts_stamp = tw->tw_ts_recent_stamp; |
| 1997 | peer->tcp_ts = tw->tw_ts_recent; |
| 1998 | } |
| 1999 | inet_putpeer(peer); |
| 2000 | return 1; |
| 2001 | } |
| 2002 | |
| 2003 | return 0; |
| 2004 | } |
| 2005 | |
| 2006 | struct tcp_func ipv4_specific = { |
| 2007 | .queue_xmit = ip_queue_xmit, |
| 2008 | .send_check = tcp_v4_send_check, |
| 2009 | .rebuild_header = tcp_v4_rebuild_header, |
| 2010 | .conn_request = tcp_v4_conn_request, |
| 2011 | .syn_recv_sock = tcp_v4_syn_recv_sock, |
| 2012 | .remember_stamp = tcp_v4_remember_stamp, |
| 2013 | .net_header_len = sizeof(struct iphdr), |
| 2014 | .setsockopt = ip_setsockopt, |
| 2015 | .getsockopt = ip_getsockopt, |
| 2016 | .addr2sockaddr = v4_addr2sockaddr, |
| 2017 | .sockaddr_len = sizeof(struct sockaddr_in), |
| 2018 | }; |
| 2019 | |
| 2020 | /* NOTE: A lot of things set to zero explicitly by call to |
| 2021 | * sk_alloc() so need not be done here. |
| 2022 | */ |
| 2023 | static int tcp_v4_init_sock(struct sock *sk) |
| 2024 | { |
| 2025 | struct tcp_sock *tp = tcp_sk(sk); |
| 2026 | |
| 2027 | skb_queue_head_init(&tp->out_of_order_queue); |
| 2028 | tcp_init_xmit_timers(sk); |
| 2029 | tcp_prequeue_init(tp); |
| 2030 | |
| 2031 | tp->rto = TCP_TIMEOUT_INIT; |
| 2032 | tp->mdev = TCP_TIMEOUT_INIT; |
| 2033 | |
| 2034 | /* So many TCP implementations out there (incorrectly) count the |
| 2035 | * initial SYN frame in their delayed-ACK and congestion control |
| 2036 | * algorithms that we must have the following bandaid to talk |
| 2037 | * efficiently to them. -DaveM |
| 2038 | */ |
| 2039 | tp->snd_cwnd = 2; |
| 2040 | |
| 2041 | /* See draft-stevens-tcpca-spec-01 for discussion of the |
| 2042 | * initialization of these values. |
| 2043 | */ |
| 2044 | tp->snd_ssthresh = 0x7fffffff; /* Infinity */ |
| 2045 | tp->snd_cwnd_clamp = ~0; |
David S. Miller | c1b4a7e | 2005-07-05 15:24:38 -0700 | [diff] [blame] | 2046 | tp->mss_cache = 536; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2047 | |
| 2048 | tp->reordering = sysctl_tcp_reordering; |
Stephen Hemminger | 5f8ef48 | 2005-06-23 20:37:36 -0700 | [diff] [blame] | 2049 | tp->ca_ops = &tcp_init_congestion_ops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2050 | |
| 2051 | sk->sk_state = TCP_CLOSE; |
| 2052 | |
| 2053 | sk->sk_write_space = sk_stream_write_space; |
| 2054 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); |
| 2055 | |
| 2056 | tp->af_specific = &ipv4_specific; |
| 2057 | |
| 2058 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; |
| 2059 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; |
| 2060 | |
| 2061 | atomic_inc(&tcp_sockets_allocated); |
| 2062 | |
| 2063 | return 0; |
| 2064 | } |
| 2065 | |
| 2066 | int tcp_v4_destroy_sock(struct sock *sk) |
| 2067 | { |
| 2068 | struct tcp_sock *tp = tcp_sk(sk); |
| 2069 | |
| 2070 | tcp_clear_xmit_timers(sk); |
| 2071 | |
Stephen Hemminger | 317a76f | 2005-06-23 12:19:55 -0700 | [diff] [blame] | 2072 | tcp_cleanup_congestion_control(tp); |
| 2073 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2074 | /* Cleanup up the write buffer. */ |
| 2075 | sk_stream_writequeue_purge(sk); |
| 2076 | |
| 2077 | /* Cleans up our, hopefully empty, out_of_order_queue. */ |
| 2078 | __skb_queue_purge(&tp->out_of_order_queue); |
| 2079 | |
| 2080 | /* Clean prequeue, it must be empty really */ |
| 2081 | __skb_queue_purge(&tp->ucopy.prequeue); |
| 2082 | |
| 2083 | /* Clean up a referenced TCP bind bucket. */ |
| 2084 | if (tp->bind_hash) |
| 2085 | tcp_put_port(sk); |
| 2086 | |
| 2087 | /* |
| 2088 | * If sendmsg cached page exists, toss it. |
| 2089 | */ |
| 2090 | if (sk->sk_sndmsg_page) { |
| 2091 | __free_page(sk->sk_sndmsg_page); |
| 2092 | sk->sk_sndmsg_page = NULL; |
| 2093 | } |
| 2094 | |
| 2095 | atomic_dec(&tcp_sockets_allocated); |
| 2096 | |
| 2097 | return 0; |
| 2098 | } |
| 2099 | |
| 2100 | EXPORT_SYMBOL(tcp_v4_destroy_sock); |
| 2101 | |
| 2102 | #ifdef CONFIG_PROC_FS |
| 2103 | /* Proc filesystem TCP sock list dumping. */ |
| 2104 | |
| 2105 | static inline struct tcp_tw_bucket *tw_head(struct hlist_head *head) |
| 2106 | { |
| 2107 | return hlist_empty(head) ? NULL : |
| 2108 | list_entry(head->first, struct tcp_tw_bucket, tw_node); |
| 2109 | } |
| 2110 | |
| 2111 | static inline struct tcp_tw_bucket *tw_next(struct tcp_tw_bucket *tw) |
| 2112 | { |
| 2113 | return tw->tw_node.next ? |
| 2114 | hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL; |
| 2115 | } |
| 2116 | |
| 2117 | static void *listening_get_next(struct seq_file *seq, void *cur) |
| 2118 | { |
| 2119 | struct tcp_sock *tp; |
| 2120 | struct hlist_node *node; |
| 2121 | struct sock *sk = cur; |
| 2122 | struct tcp_iter_state* st = seq->private; |
| 2123 | |
| 2124 | if (!sk) { |
| 2125 | st->bucket = 0; |
| 2126 | sk = sk_head(&tcp_listening_hash[0]); |
| 2127 | goto get_sk; |
| 2128 | } |
| 2129 | |
| 2130 | ++st->num; |
| 2131 | |
| 2132 | if (st->state == TCP_SEQ_STATE_OPENREQ) { |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 2133 | struct request_sock *req = cur; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2134 | |
| 2135 | tp = tcp_sk(st->syn_wait_sk); |
| 2136 | req = req->dl_next; |
| 2137 | while (1) { |
| 2138 | while (req) { |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 2139 | if (req->rsk_ops->family == st->family) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2140 | cur = req; |
| 2141 | goto out; |
| 2142 | } |
| 2143 | req = req->dl_next; |
| 2144 | } |
| 2145 | if (++st->sbucket >= TCP_SYNQ_HSIZE) |
| 2146 | break; |
| 2147 | get_req: |
Arnaldo Carvalho de Melo | 0e87506 | 2005-06-18 22:47:59 -0700 | [diff] [blame] | 2148 | req = tp->accept_queue.listen_opt->syn_table[st->sbucket]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2149 | } |
| 2150 | sk = sk_next(st->syn_wait_sk); |
| 2151 | st->state = TCP_SEQ_STATE_LISTENING; |
Arnaldo Carvalho de Melo | 0e87506 | 2005-06-18 22:47:59 -0700 | [diff] [blame] | 2152 | read_unlock_bh(&tp->accept_queue.syn_wait_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2153 | } else { |
| 2154 | tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 0e87506 | 2005-06-18 22:47:59 -0700 | [diff] [blame] | 2155 | read_lock_bh(&tp->accept_queue.syn_wait_lock); |
| 2156 | if (reqsk_queue_len(&tp->accept_queue)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2157 | goto start_req; |
Arnaldo Carvalho de Melo | 0e87506 | 2005-06-18 22:47:59 -0700 | [diff] [blame] | 2158 | read_unlock_bh(&tp->accept_queue.syn_wait_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2159 | sk = sk_next(sk); |
| 2160 | } |
| 2161 | get_sk: |
| 2162 | sk_for_each_from(sk, node) { |
| 2163 | if (sk->sk_family == st->family) { |
| 2164 | cur = sk; |
| 2165 | goto out; |
| 2166 | } |
| 2167 | tp = tcp_sk(sk); |
Arnaldo Carvalho de Melo | 0e87506 | 2005-06-18 22:47:59 -0700 | [diff] [blame] | 2168 | read_lock_bh(&tp->accept_queue.syn_wait_lock); |
| 2169 | if (reqsk_queue_len(&tp->accept_queue)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2170 | start_req: |
| 2171 | st->uid = sock_i_uid(sk); |
| 2172 | st->syn_wait_sk = sk; |
| 2173 | st->state = TCP_SEQ_STATE_OPENREQ; |
| 2174 | st->sbucket = 0; |
| 2175 | goto get_req; |
| 2176 | } |
Arnaldo Carvalho de Melo | 0e87506 | 2005-06-18 22:47:59 -0700 | [diff] [blame] | 2177 | read_unlock_bh(&tp->accept_queue.syn_wait_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2178 | } |
| 2179 | if (++st->bucket < TCP_LHTABLE_SIZE) { |
| 2180 | sk = sk_head(&tcp_listening_hash[st->bucket]); |
| 2181 | goto get_sk; |
| 2182 | } |
| 2183 | cur = NULL; |
| 2184 | out: |
| 2185 | return cur; |
| 2186 | } |
| 2187 | |
| 2188 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) |
| 2189 | { |
| 2190 | void *rc = listening_get_next(seq, NULL); |
| 2191 | |
| 2192 | while (rc && *pos) { |
| 2193 | rc = listening_get_next(seq, rc); |
| 2194 | --*pos; |
| 2195 | } |
| 2196 | return rc; |
| 2197 | } |
| 2198 | |
| 2199 | static void *established_get_first(struct seq_file *seq) |
| 2200 | { |
| 2201 | struct tcp_iter_state* st = seq->private; |
| 2202 | void *rc = NULL; |
| 2203 | |
| 2204 | for (st->bucket = 0; st->bucket < tcp_ehash_size; ++st->bucket) { |
| 2205 | struct sock *sk; |
| 2206 | struct hlist_node *node; |
| 2207 | struct tcp_tw_bucket *tw; |
| 2208 | |
| 2209 | /* We can reschedule _before_ having picked the target: */ |
| 2210 | cond_resched_softirq(); |
| 2211 | |
| 2212 | read_lock(&tcp_ehash[st->bucket].lock); |
| 2213 | sk_for_each(sk, node, &tcp_ehash[st->bucket].chain) { |
| 2214 | if (sk->sk_family != st->family) { |
| 2215 | continue; |
| 2216 | } |
| 2217 | rc = sk; |
| 2218 | goto out; |
| 2219 | } |
| 2220 | st->state = TCP_SEQ_STATE_TIME_WAIT; |
| 2221 | tw_for_each(tw, node, |
| 2222 | &tcp_ehash[st->bucket + tcp_ehash_size].chain) { |
| 2223 | if (tw->tw_family != st->family) { |
| 2224 | continue; |
| 2225 | } |
| 2226 | rc = tw; |
| 2227 | goto out; |
| 2228 | } |
| 2229 | read_unlock(&tcp_ehash[st->bucket].lock); |
| 2230 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
| 2231 | } |
| 2232 | out: |
| 2233 | return rc; |
| 2234 | } |
| 2235 | |
| 2236 | static void *established_get_next(struct seq_file *seq, void *cur) |
| 2237 | { |
| 2238 | struct sock *sk = cur; |
| 2239 | struct tcp_tw_bucket *tw; |
| 2240 | struct hlist_node *node; |
| 2241 | struct tcp_iter_state* st = seq->private; |
| 2242 | |
| 2243 | ++st->num; |
| 2244 | |
| 2245 | if (st->state == TCP_SEQ_STATE_TIME_WAIT) { |
| 2246 | tw = cur; |
| 2247 | tw = tw_next(tw); |
| 2248 | get_tw: |
| 2249 | while (tw && tw->tw_family != st->family) { |
| 2250 | tw = tw_next(tw); |
| 2251 | } |
| 2252 | if (tw) { |
| 2253 | cur = tw; |
| 2254 | goto out; |
| 2255 | } |
| 2256 | read_unlock(&tcp_ehash[st->bucket].lock); |
| 2257 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
| 2258 | |
| 2259 | /* We can reschedule between buckets: */ |
| 2260 | cond_resched_softirq(); |
| 2261 | |
| 2262 | if (++st->bucket < tcp_ehash_size) { |
| 2263 | read_lock(&tcp_ehash[st->bucket].lock); |
| 2264 | sk = sk_head(&tcp_ehash[st->bucket].chain); |
| 2265 | } else { |
| 2266 | cur = NULL; |
| 2267 | goto out; |
| 2268 | } |
| 2269 | } else |
| 2270 | sk = sk_next(sk); |
| 2271 | |
| 2272 | sk_for_each_from(sk, node) { |
| 2273 | if (sk->sk_family == st->family) |
| 2274 | goto found; |
| 2275 | } |
| 2276 | |
| 2277 | st->state = TCP_SEQ_STATE_TIME_WAIT; |
| 2278 | tw = tw_head(&tcp_ehash[st->bucket + tcp_ehash_size].chain); |
| 2279 | goto get_tw; |
| 2280 | found: |
| 2281 | cur = sk; |
| 2282 | out: |
| 2283 | return cur; |
| 2284 | } |
| 2285 | |
| 2286 | static void *established_get_idx(struct seq_file *seq, loff_t pos) |
| 2287 | { |
| 2288 | void *rc = established_get_first(seq); |
| 2289 | |
| 2290 | while (rc && pos) { |
| 2291 | rc = established_get_next(seq, rc); |
| 2292 | --pos; |
| 2293 | } |
| 2294 | return rc; |
| 2295 | } |
| 2296 | |
| 2297 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) |
| 2298 | { |
| 2299 | void *rc; |
| 2300 | struct tcp_iter_state* st = seq->private; |
| 2301 | |
| 2302 | tcp_listen_lock(); |
| 2303 | st->state = TCP_SEQ_STATE_LISTENING; |
| 2304 | rc = listening_get_idx(seq, &pos); |
| 2305 | |
| 2306 | if (!rc) { |
| 2307 | tcp_listen_unlock(); |
| 2308 | local_bh_disable(); |
| 2309 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
| 2310 | rc = established_get_idx(seq, pos); |
| 2311 | } |
| 2312 | |
| 2313 | return rc; |
| 2314 | } |
| 2315 | |
| 2316 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) |
| 2317 | { |
| 2318 | struct tcp_iter_state* st = seq->private; |
| 2319 | st->state = TCP_SEQ_STATE_LISTENING; |
| 2320 | st->num = 0; |
| 2321 | return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; |
| 2322 | } |
| 2323 | |
| 2324 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
| 2325 | { |
| 2326 | void *rc = NULL; |
| 2327 | struct tcp_iter_state* st; |
| 2328 | |
| 2329 | if (v == SEQ_START_TOKEN) { |
| 2330 | rc = tcp_get_idx(seq, 0); |
| 2331 | goto out; |
| 2332 | } |
| 2333 | st = seq->private; |
| 2334 | |
| 2335 | switch (st->state) { |
| 2336 | case TCP_SEQ_STATE_OPENREQ: |
| 2337 | case TCP_SEQ_STATE_LISTENING: |
| 2338 | rc = listening_get_next(seq, v); |
| 2339 | if (!rc) { |
| 2340 | tcp_listen_unlock(); |
| 2341 | local_bh_disable(); |
| 2342 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
| 2343 | rc = established_get_first(seq); |
| 2344 | } |
| 2345 | break; |
| 2346 | case TCP_SEQ_STATE_ESTABLISHED: |
| 2347 | case TCP_SEQ_STATE_TIME_WAIT: |
| 2348 | rc = established_get_next(seq, v); |
| 2349 | break; |
| 2350 | } |
| 2351 | out: |
| 2352 | ++*pos; |
| 2353 | return rc; |
| 2354 | } |
| 2355 | |
| 2356 | static void tcp_seq_stop(struct seq_file *seq, void *v) |
| 2357 | { |
| 2358 | struct tcp_iter_state* st = seq->private; |
| 2359 | |
| 2360 | switch (st->state) { |
| 2361 | case TCP_SEQ_STATE_OPENREQ: |
| 2362 | if (v) { |
| 2363 | struct tcp_sock *tp = tcp_sk(st->syn_wait_sk); |
Arnaldo Carvalho de Melo | 0e87506 | 2005-06-18 22:47:59 -0700 | [diff] [blame] | 2364 | read_unlock_bh(&tp->accept_queue.syn_wait_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2365 | } |
| 2366 | case TCP_SEQ_STATE_LISTENING: |
| 2367 | if (v != SEQ_START_TOKEN) |
| 2368 | tcp_listen_unlock(); |
| 2369 | break; |
| 2370 | case TCP_SEQ_STATE_TIME_WAIT: |
| 2371 | case TCP_SEQ_STATE_ESTABLISHED: |
| 2372 | if (v) |
| 2373 | read_unlock(&tcp_ehash[st->bucket].lock); |
| 2374 | local_bh_enable(); |
| 2375 | break; |
| 2376 | } |
| 2377 | } |
| 2378 | |
| 2379 | static int tcp_seq_open(struct inode *inode, struct file *file) |
| 2380 | { |
| 2381 | struct tcp_seq_afinfo *afinfo = PDE(inode)->data; |
| 2382 | struct seq_file *seq; |
| 2383 | struct tcp_iter_state *s; |
| 2384 | int rc; |
| 2385 | |
| 2386 | if (unlikely(afinfo == NULL)) |
| 2387 | return -EINVAL; |
| 2388 | |
| 2389 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
| 2390 | if (!s) |
| 2391 | return -ENOMEM; |
| 2392 | memset(s, 0, sizeof(*s)); |
| 2393 | s->family = afinfo->family; |
| 2394 | s->seq_ops.start = tcp_seq_start; |
| 2395 | s->seq_ops.next = tcp_seq_next; |
| 2396 | s->seq_ops.show = afinfo->seq_show; |
| 2397 | s->seq_ops.stop = tcp_seq_stop; |
| 2398 | |
| 2399 | rc = seq_open(file, &s->seq_ops); |
| 2400 | if (rc) |
| 2401 | goto out_kfree; |
| 2402 | seq = file->private_data; |
| 2403 | seq->private = s; |
| 2404 | out: |
| 2405 | return rc; |
| 2406 | out_kfree: |
| 2407 | kfree(s); |
| 2408 | goto out; |
| 2409 | } |
| 2410 | |
| 2411 | int tcp_proc_register(struct tcp_seq_afinfo *afinfo) |
| 2412 | { |
| 2413 | int rc = 0; |
| 2414 | struct proc_dir_entry *p; |
| 2415 | |
| 2416 | if (!afinfo) |
| 2417 | return -EINVAL; |
| 2418 | afinfo->seq_fops->owner = afinfo->owner; |
| 2419 | afinfo->seq_fops->open = tcp_seq_open; |
| 2420 | afinfo->seq_fops->read = seq_read; |
| 2421 | afinfo->seq_fops->llseek = seq_lseek; |
| 2422 | afinfo->seq_fops->release = seq_release_private; |
| 2423 | |
| 2424 | p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops); |
| 2425 | if (p) |
| 2426 | p->data = afinfo; |
| 2427 | else |
| 2428 | rc = -ENOMEM; |
| 2429 | return rc; |
| 2430 | } |
| 2431 | |
| 2432 | void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo) |
| 2433 | { |
| 2434 | if (!afinfo) |
| 2435 | return; |
| 2436 | proc_net_remove(afinfo->name); |
| 2437 | memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops)); |
| 2438 | } |
| 2439 | |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 2440 | static void get_openreq4(struct sock *sk, struct request_sock *req, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2441 | char *tmpbuf, int i, int uid) |
| 2442 | { |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 2443 | const struct inet_request_sock *ireq = inet_rsk(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2444 | int ttd = req->expires - jiffies; |
| 2445 | |
| 2446 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" |
| 2447 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p", |
| 2448 | i, |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 2449 | ireq->loc_addr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2450 | ntohs(inet_sk(sk)->sport), |
Arnaldo Carvalho de Melo | 2e6599c | 2005-06-18 22:46:52 -0700 | [diff] [blame] | 2451 | ireq->rmt_addr, |
| 2452 | ntohs(ireq->rmt_port), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2453 | TCP_SYN_RECV, |
| 2454 | 0, 0, /* could print option size, but that is af dependent. */ |
| 2455 | 1, /* timers active (only the expire timer) */ |
| 2456 | jiffies_to_clock_t(ttd), |
| 2457 | req->retrans, |
| 2458 | uid, |
| 2459 | 0, /* non standard timer */ |
| 2460 | 0, /* open_requests have no inode */ |
| 2461 | atomic_read(&sk->sk_refcnt), |
| 2462 | req); |
| 2463 | } |
| 2464 | |
| 2465 | static void get_tcp4_sock(struct sock *sp, char *tmpbuf, int i) |
| 2466 | { |
| 2467 | int timer_active; |
| 2468 | unsigned long timer_expires; |
| 2469 | struct tcp_sock *tp = tcp_sk(sp); |
| 2470 | struct inet_sock *inet = inet_sk(sp); |
| 2471 | unsigned int dest = inet->daddr; |
| 2472 | unsigned int src = inet->rcv_saddr; |
| 2473 | __u16 destp = ntohs(inet->dport); |
| 2474 | __u16 srcp = ntohs(inet->sport); |
| 2475 | |
| 2476 | if (tp->pending == TCP_TIME_RETRANS) { |
| 2477 | timer_active = 1; |
| 2478 | timer_expires = tp->timeout; |
| 2479 | } else if (tp->pending == TCP_TIME_PROBE0) { |
| 2480 | timer_active = 4; |
| 2481 | timer_expires = tp->timeout; |
| 2482 | } else if (timer_pending(&sp->sk_timer)) { |
| 2483 | timer_active = 2; |
| 2484 | timer_expires = sp->sk_timer.expires; |
| 2485 | } else { |
| 2486 | timer_active = 0; |
| 2487 | timer_expires = jiffies; |
| 2488 | } |
| 2489 | |
| 2490 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " |
| 2491 | "%08X %5d %8d %lu %d %p %u %u %u %u %d", |
| 2492 | i, src, srcp, dest, destp, sp->sk_state, |
| 2493 | tp->write_seq - tp->snd_una, tp->rcv_nxt - tp->copied_seq, |
| 2494 | timer_active, |
| 2495 | jiffies_to_clock_t(timer_expires - jiffies), |
| 2496 | tp->retransmits, |
| 2497 | sock_i_uid(sp), |
| 2498 | tp->probes_out, |
| 2499 | sock_i_ino(sp), |
| 2500 | atomic_read(&sp->sk_refcnt), sp, |
| 2501 | tp->rto, tp->ack.ato, (tp->ack.quick << 1) | tp->ack.pingpong, |
| 2502 | tp->snd_cwnd, |
| 2503 | tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh); |
| 2504 | } |
| 2505 | |
| 2506 | static void get_timewait4_sock(struct tcp_tw_bucket *tw, char *tmpbuf, int i) |
| 2507 | { |
| 2508 | unsigned int dest, src; |
| 2509 | __u16 destp, srcp; |
| 2510 | int ttd = tw->tw_ttd - jiffies; |
| 2511 | |
| 2512 | if (ttd < 0) |
| 2513 | ttd = 0; |
| 2514 | |
| 2515 | dest = tw->tw_daddr; |
| 2516 | src = tw->tw_rcv_saddr; |
| 2517 | destp = ntohs(tw->tw_dport); |
| 2518 | srcp = ntohs(tw->tw_sport); |
| 2519 | |
| 2520 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" |
| 2521 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p", |
| 2522 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, |
| 2523 | 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0, |
| 2524 | atomic_read(&tw->tw_refcnt), tw); |
| 2525 | } |
| 2526 | |
| 2527 | #define TMPSZ 150 |
| 2528 | |
| 2529 | static int tcp4_seq_show(struct seq_file *seq, void *v) |
| 2530 | { |
| 2531 | struct tcp_iter_state* st; |
| 2532 | char tmpbuf[TMPSZ + 1]; |
| 2533 | |
| 2534 | if (v == SEQ_START_TOKEN) { |
| 2535 | seq_printf(seq, "%-*s\n", TMPSZ - 1, |
| 2536 | " sl local_address rem_address st tx_queue " |
| 2537 | "rx_queue tr tm->when retrnsmt uid timeout " |
| 2538 | "inode"); |
| 2539 | goto out; |
| 2540 | } |
| 2541 | st = seq->private; |
| 2542 | |
| 2543 | switch (st->state) { |
| 2544 | case TCP_SEQ_STATE_LISTENING: |
| 2545 | case TCP_SEQ_STATE_ESTABLISHED: |
| 2546 | get_tcp4_sock(v, tmpbuf, st->num); |
| 2547 | break; |
| 2548 | case TCP_SEQ_STATE_OPENREQ: |
| 2549 | get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid); |
| 2550 | break; |
| 2551 | case TCP_SEQ_STATE_TIME_WAIT: |
| 2552 | get_timewait4_sock(v, tmpbuf, st->num); |
| 2553 | break; |
| 2554 | } |
| 2555 | seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf); |
| 2556 | out: |
| 2557 | return 0; |
| 2558 | } |
| 2559 | |
| 2560 | static struct file_operations tcp4_seq_fops; |
| 2561 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { |
| 2562 | .owner = THIS_MODULE, |
| 2563 | .name = "tcp", |
| 2564 | .family = AF_INET, |
| 2565 | .seq_show = tcp4_seq_show, |
| 2566 | .seq_fops = &tcp4_seq_fops, |
| 2567 | }; |
| 2568 | |
| 2569 | int __init tcp4_proc_init(void) |
| 2570 | { |
| 2571 | return tcp_proc_register(&tcp4_seq_afinfo); |
| 2572 | } |
| 2573 | |
| 2574 | void tcp4_proc_exit(void) |
| 2575 | { |
| 2576 | tcp_proc_unregister(&tcp4_seq_afinfo); |
| 2577 | } |
| 2578 | #endif /* CONFIG_PROC_FS */ |
| 2579 | |
| 2580 | struct proto tcp_prot = { |
| 2581 | .name = "TCP", |
| 2582 | .owner = THIS_MODULE, |
| 2583 | .close = tcp_close, |
| 2584 | .connect = tcp_v4_connect, |
| 2585 | .disconnect = tcp_disconnect, |
| 2586 | .accept = tcp_accept, |
| 2587 | .ioctl = tcp_ioctl, |
| 2588 | .init = tcp_v4_init_sock, |
| 2589 | .destroy = tcp_v4_destroy_sock, |
| 2590 | .shutdown = tcp_shutdown, |
| 2591 | .setsockopt = tcp_setsockopt, |
| 2592 | .getsockopt = tcp_getsockopt, |
| 2593 | .sendmsg = tcp_sendmsg, |
| 2594 | .recvmsg = tcp_recvmsg, |
| 2595 | .backlog_rcv = tcp_v4_do_rcv, |
| 2596 | .hash = tcp_v4_hash, |
| 2597 | .unhash = tcp_unhash, |
| 2598 | .get_port = tcp_v4_get_port, |
| 2599 | .enter_memory_pressure = tcp_enter_memory_pressure, |
| 2600 | .sockets_allocated = &tcp_sockets_allocated, |
| 2601 | .memory_allocated = &tcp_memory_allocated, |
| 2602 | .memory_pressure = &tcp_memory_pressure, |
| 2603 | .sysctl_mem = sysctl_tcp_mem, |
| 2604 | .sysctl_wmem = sysctl_tcp_wmem, |
| 2605 | .sysctl_rmem = sysctl_tcp_rmem, |
| 2606 | .max_header = MAX_TCP_HEADER, |
| 2607 | .obj_size = sizeof(struct tcp_sock), |
Arnaldo Carvalho de Melo | 60236fd | 2005-06-18 22:47:21 -0700 | [diff] [blame] | 2608 | .rsk_prot = &tcp_request_sock_ops, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2609 | }; |
| 2610 | |
| 2611 | |
| 2612 | |
| 2613 | void __init tcp_v4_init(struct net_proto_family *ops) |
| 2614 | { |
| 2615 | int err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_TCP, &tcp_socket); |
| 2616 | if (err < 0) |
| 2617 | panic("Failed to create the TCP control socket.\n"); |
| 2618 | tcp_socket->sk->sk_allocation = GFP_ATOMIC; |
| 2619 | inet_sk(tcp_socket->sk)->uc_ttl = -1; |
| 2620 | |
| 2621 | /* Unhash it so that IP input processing does not even |
| 2622 | * see it, we do not wish this socket to see incoming |
| 2623 | * packets. |
| 2624 | */ |
| 2625 | tcp_socket->sk->sk_prot->unhash(tcp_socket->sk); |
| 2626 | } |
| 2627 | |
| 2628 | EXPORT_SYMBOL(ipv4_specific); |
| 2629 | EXPORT_SYMBOL(tcp_bind_hash); |
| 2630 | EXPORT_SYMBOL(tcp_bucket_create); |
| 2631 | EXPORT_SYMBOL(tcp_hashinfo); |
| 2632 | EXPORT_SYMBOL(tcp_inherit_port); |
| 2633 | EXPORT_SYMBOL(tcp_listen_wlock); |
| 2634 | EXPORT_SYMBOL(tcp_port_rover); |
| 2635 | EXPORT_SYMBOL(tcp_prot); |
| 2636 | EXPORT_SYMBOL(tcp_put_port); |
| 2637 | EXPORT_SYMBOL(tcp_unhash); |
| 2638 | EXPORT_SYMBOL(tcp_v4_conn_request); |
| 2639 | EXPORT_SYMBOL(tcp_v4_connect); |
| 2640 | EXPORT_SYMBOL(tcp_v4_do_rcv); |
| 2641 | EXPORT_SYMBOL(tcp_v4_rebuild_header); |
| 2642 | EXPORT_SYMBOL(tcp_v4_remember_stamp); |
| 2643 | EXPORT_SYMBOL(tcp_v4_send_check); |
| 2644 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); |
| 2645 | |
| 2646 | #ifdef CONFIG_PROC_FS |
| 2647 | EXPORT_SYMBOL(tcp_proc_register); |
| 2648 | EXPORT_SYMBOL(tcp_proc_unregister); |
| 2649 | #endif |
| 2650 | EXPORT_SYMBOL(sysctl_local_port_range); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2651 | EXPORT_SYMBOL(sysctl_tcp_low_latency); |
| 2652 | EXPORT_SYMBOL(sysctl_tcp_tw_reuse); |
| 2653 | |