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 * Support for INET connection oriented protocols. 7 * 8 * Authors: See the TCP sources 9 * 10 * This program is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU General Public License 12 * as published by the Free Software Foundation; either version 13 * 2 of the License, or(at your option) any later version. 14 */ 15 16 #include <linux/module.h> 17 #include <linux/jhash.h> 18 19 #include <net/inet_connection_sock.h> 20 #include <net/inet_hashtables.h> 21 #include <net/inet_timewait_sock.h> 22 #include <net/ip.h> 23 #include <net/route.h> 24 #include <net/tcp_states.h> 25 #include <net/xfrm.h> 26 #include <net/tcp.h> 27 28 #ifdef INET_CSK_DEBUG 29 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n"; 30 EXPORT_SYMBOL(inet_csk_timer_bug_msg); 31 #endif 32 33 void inet_get_local_port_range(struct net *net, int *low, int *high) 34 { 35 unsigned int seq; 36 37 do { 38 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock); 39 40 *low = net->ipv4.ip_local_ports.range[0]; 41 *high = net->ipv4.ip_local_ports.range[1]; 42 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq)); 43 } 44 EXPORT_SYMBOL(inet_get_local_port_range); 45 46 int inet_csk_bind_conflict(const struct sock *sk, 47 const struct inet_bind_bucket *tb, bool relax) 48 { 49 struct sock *sk2; 50 int reuse = sk->sk_reuse; 51 int reuseport = sk->sk_reuseport; 52 kuid_t uid = sock_i_uid((struct sock *)sk); 53 54 /* 55 * Unlike other sk lookup places we do not check 56 * for sk_net here, since _all_ the socks listed 57 * in tb->owners list belong to the same net - the 58 * one this bucket belongs to. 59 */ 60 61 sk_for_each_bound(sk2, &tb->owners) { 62 if (sk != sk2 && 63 !inet_v6_ipv6only(sk2) && 64 (!sk->sk_bound_dev_if || 65 !sk2->sk_bound_dev_if || 66 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { 67 if ((!reuse || !sk2->sk_reuse || 68 sk2->sk_state == TCP_LISTEN) && 69 (!reuseport || !sk2->sk_reuseport || 70 (sk2->sk_state != TCP_TIME_WAIT && 71 !uid_eq(uid, sock_i_uid(sk2))))) { 72 73 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr || 74 sk2->sk_rcv_saddr == sk->sk_rcv_saddr) 75 break; 76 } 77 if (!relax && reuse && sk2->sk_reuse && 78 sk2->sk_state != TCP_LISTEN) { 79 80 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr || 81 sk2->sk_rcv_saddr == sk->sk_rcv_saddr) 82 break; 83 } 84 } 85 } 86 return sk2 != NULL; 87 } 88 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict); 89 90 /* Obtain a reference to a local port for the given sock, 91 * if snum is zero it means select any available local port. 92 */ 93 int inet_csk_get_port(struct sock *sk, unsigned short snum) 94 { 95 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; 96 struct inet_bind_hashbucket *head; 97 struct inet_bind_bucket *tb; 98 int ret, attempts = 5; 99 struct net *net = sock_net(sk); 100 int smallest_size = -1, smallest_rover; 101 kuid_t uid = sock_i_uid(sk); 102 int attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0; 103 104 local_bh_disable(); 105 if (!snum) { 106 int remaining, rover, low, high; 107 108 again: 109 inet_get_local_port_range(net, &low, &high); 110 if (attempt_half) { 111 int half = low + ((high - low) >> 1); 112 113 if (attempt_half == 1) 114 high = half; 115 else 116 low = half; 117 } 118 remaining = (high - low) + 1; 119 smallest_rover = rover = prandom_u32() % remaining + low; 120 121 smallest_size = -1; 122 do { 123 if (inet_is_local_reserved_port(net, rover)) 124 goto next_nolock; 125 head = &hashinfo->bhash[inet_bhashfn(net, rover, 126 hashinfo->bhash_size)]; 127 spin_lock(&head->lock); 128 inet_bind_bucket_for_each(tb, &head->chain) 129 if (net_eq(ib_net(tb), net) && tb->port == rover) { 130 if (((tb->fastreuse > 0 && 131 sk->sk_reuse && 132 sk->sk_state != TCP_LISTEN) || 133 (tb->fastreuseport > 0 && 134 sk->sk_reuseport && 135 uid_eq(tb->fastuid, uid))) && 136 (tb->num_owners < smallest_size || smallest_size == -1)) { 137 smallest_size = tb->num_owners; 138 smallest_rover = rover; 139 } 140 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) { 141 snum = rover; 142 goto tb_found; 143 } 144 goto next; 145 } 146 break; 147 next: 148 spin_unlock(&head->lock); 149 next_nolock: 150 if (++rover > high) 151 rover = low; 152 } while (--remaining > 0); 153 154 /* Exhausted local port range during search? It is not 155 * possible for us to be holding one of the bind hash 156 * locks if this test triggers, because if 'remaining' 157 * drops to zero, we broke out of the do/while loop at 158 * the top level, not from the 'break;' statement. 159 */ 160 ret = 1; 161 if (remaining <= 0) { 162 if (smallest_size != -1) { 163 snum = smallest_rover; 164 goto have_snum; 165 } 166 if (attempt_half == 1) { 167 /* OK we now try the upper half of the range */ 168 attempt_half = 2; 169 goto again; 170 } 171 goto fail; 172 } 173 /* OK, here is the one we will use. HEAD is 174 * non-NULL and we hold it's mutex. 175 */ 176 snum = rover; 177 } else { 178 have_snum: 179 head = &hashinfo->bhash[inet_bhashfn(net, snum, 180 hashinfo->bhash_size)]; 181 spin_lock(&head->lock); 182 inet_bind_bucket_for_each(tb, &head->chain) 183 if (net_eq(ib_net(tb), net) && tb->port == snum) 184 goto tb_found; 185 } 186 tb = NULL; 187 goto tb_not_found; 188 tb_found: 189 if (!hlist_empty(&tb->owners)) { 190 if (sk->sk_reuse == SK_FORCE_REUSE) 191 goto success; 192 193 if (((tb->fastreuse > 0 && 194 sk->sk_reuse && sk->sk_state != TCP_LISTEN) || 195 (tb->fastreuseport > 0 && 196 sk->sk_reuseport && uid_eq(tb->fastuid, uid))) && 197 smallest_size == -1) { 198 goto success; 199 } else { 200 ret = 1; 201 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) { 202 if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) || 203 (tb->fastreuseport > 0 && 204 sk->sk_reuseport && uid_eq(tb->fastuid, uid))) && 205 smallest_size != -1 && --attempts >= 0) { 206 spin_unlock(&head->lock); 207 goto again; 208 } 209 210 goto fail_unlock; 211 } 212 } 213 } 214 tb_not_found: 215 ret = 1; 216 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep, 217 net, head, snum)) == NULL) 218 goto fail_unlock; 219 if (hlist_empty(&tb->owners)) { 220 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN) 221 tb->fastreuse = 1; 222 else 223 tb->fastreuse = 0; 224 if (sk->sk_reuseport) { 225 tb->fastreuseport = 1; 226 tb->fastuid = uid; 227 } else 228 tb->fastreuseport = 0; 229 } else { 230 if (tb->fastreuse && 231 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN)) 232 tb->fastreuse = 0; 233 if (tb->fastreuseport && 234 (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid))) 235 tb->fastreuseport = 0; 236 } 237 success: 238 if (!inet_csk(sk)->icsk_bind_hash) 239 inet_bind_hash(sk, tb, snum); 240 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); 241 ret = 0; 242 243 fail_unlock: 244 spin_unlock(&head->lock); 245 fail: 246 local_bh_enable(); 247 return ret; 248 } 249 EXPORT_SYMBOL_GPL(inet_csk_get_port); 250 251 /* 252 * Wait for an incoming connection, avoid race conditions. This must be called 253 * with the socket locked. 254 */ 255 static int inet_csk_wait_for_connect(struct sock *sk, long timeo) 256 { 257 struct inet_connection_sock *icsk = inet_csk(sk); 258 DEFINE_WAIT(wait); 259 int err; 260 261 /* 262 * True wake-one mechanism for incoming connections: only 263 * one process gets woken up, not the 'whole herd'. 264 * Since we do not 'race & poll' for established sockets 265 * anymore, the common case will execute the loop only once. 266 * 267 * Subtle issue: "add_wait_queue_exclusive()" will be added 268 * after any current non-exclusive waiters, and we know that 269 * it will always _stay_ after any new non-exclusive waiters 270 * because all non-exclusive waiters are added at the 271 * beginning of the wait-queue. As such, it's ok to "drop" 272 * our exclusiveness temporarily when we get woken up without 273 * having to remove and re-insert us on the wait queue. 274 */ 275 for (;;) { 276 prepare_to_wait_exclusive(sk_sleep(sk), &wait, 277 TASK_INTERRUPTIBLE); 278 release_sock(sk); 279 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) 280 timeo = schedule_timeout(timeo); 281 sched_annotate_sleep(); 282 lock_sock(sk); 283 err = 0; 284 if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) 285 break; 286 err = -EINVAL; 287 if (sk->sk_state != TCP_LISTEN) 288 break; 289 err = sock_intr_errno(timeo); 290 if (signal_pending(current)) 291 break; 292 err = -EAGAIN; 293 if (!timeo) 294 break; 295 } 296 finish_wait(sk_sleep(sk), &wait); 297 return err; 298 } 299 300 /* 301 * This will accept the next outstanding connection. 302 */ 303 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err) 304 { 305 struct inet_connection_sock *icsk = inet_csk(sk); 306 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 307 struct request_sock *req; 308 struct sock *newsk; 309 int error; 310 311 lock_sock(sk); 312 313 /* We need to make sure that this socket is listening, 314 * and that it has something pending. 315 */ 316 error = -EINVAL; 317 if (sk->sk_state != TCP_LISTEN) 318 goto out_err; 319 320 /* Find already established connection */ 321 if (reqsk_queue_empty(queue)) { 322 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 323 324 /* If this is a non blocking socket don't sleep */ 325 error = -EAGAIN; 326 if (!timeo) 327 goto out_err; 328 329 error = inet_csk_wait_for_connect(sk, timeo); 330 if (error) 331 goto out_err; 332 } 333 req = reqsk_queue_remove(queue, sk); 334 newsk = req->sk; 335 336 if (sk->sk_protocol == IPPROTO_TCP && 337 tcp_rsk(req)->tfo_listener) { 338 spin_lock_bh(&queue->fastopenq.lock); 339 if (tcp_rsk(req)->tfo_listener) { 340 /* We are still waiting for the final ACK from 3WHS 341 * so can't free req now. Instead, we set req->sk to 342 * NULL to signify that the child socket is taken 343 * so reqsk_fastopen_remove() will free the req 344 * when 3WHS finishes (or is aborted). 345 */ 346 req->sk = NULL; 347 req = NULL; 348 } 349 spin_unlock_bh(&queue->fastopenq.lock); 350 } 351 out: 352 release_sock(sk); 353 if (req) 354 reqsk_put(req); 355 return newsk; 356 out_err: 357 newsk = NULL; 358 req = NULL; 359 *err = error; 360 goto out; 361 } 362 EXPORT_SYMBOL(inet_csk_accept); 363 364 /* 365 * Using different timers for retransmit, delayed acks and probes 366 * We may wish use just one timer maintaining a list of expire jiffies 367 * to optimize. 368 */ 369 void inet_csk_init_xmit_timers(struct sock *sk, 370 void (*retransmit_handler)(unsigned long), 371 void (*delack_handler)(unsigned long), 372 void (*keepalive_handler)(unsigned long)) 373 { 374 struct inet_connection_sock *icsk = inet_csk(sk); 375 376 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler, 377 (unsigned long)sk); 378 setup_timer(&icsk->icsk_delack_timer, delack_handler, 379 (unsigned long)sk); 380 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk); 381 icsk->icsk_pending = icsk->icsk_ack.pending = 0; 382 } 383 EXPORT_SYMBOL(inet_csk_init_xmit_timers); 384 385 void inet_csk_clear_xmit_timers(struct sock *sk) 386 { 387 struct inet_connection_sock *icsk = inet_csk(sk); 388 389 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0; 390 391 sk_stop_timer(sk, &icsk->icsk_retransmit_timer); 392 sk_stop_timer(sk, &icsk->icsk_delack_timer); 393 sk_stop_timer(sk, &sk->sk_timer); 394 } 395 EXPORT_SYMBOL(inet_csk_clear_xmit_timers); 396 397 void inet_csk_delete_keepalive_timer(struct sock *sk) 398 { 399 sk_stop_timer(sk, &sk->sk_timer); 400 } 401 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); 402 403 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) 404 { 405 sk_reset_timer(sk, &sk->sk_timer, jiffies + len); 406 } 407 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); 408 409 struct dst_entry *inet_csk_route_req(const struct sock *sk, 410 struct flowi4 *fl4, 411 const struct request_sock *req) 412 { 413 const struct inet_request_sock *ireq = inet_rsk(req); 414 struct net *net = read_pnet(&ireq->ireq_net); 415 struct ip_options_rcu *opt = ireq->opt; 416 struct rtable *rt; 417 418 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, 419 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 420 sk->sk_protocol, inet_sk_flowi_flags(sk), 421 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, 422 ireq->ir_loc_addr, ireq->ir_rmt_port, 423 htons(ireq->ir_num)); 424 security_req_classify_flow(req, flowi4_to_flowi(fl4)); 425 rt = ip_route_output_flow(net, fl4, sk); 426 if (IS_ERR(rt)) 427 goto no_route; 428 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) 429 goto route_err; 430 return &rt->dst; 431 432 route_err: 433 ip_rt_put(rt); 434 no_route: 435 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 436 return NULL; 437 } 438 EXPORT_SYMBOL_GPL(inet_csk_route_req); 439 440 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk, 441 struct sock *newsk, 442 const struct request_sock *req) 443 { 444 const struct inet_request_sock *ireq = inet_rsk(req); 445 struct net *net = read_pnet(&ireq->ireq_net); 446 struct inet_sock *newinet = inet_sk(newsk); 447 struct ip_options_rcu *opt; 448 struct flowi4 *fl4; 449 struct rtable *rt; 450 451 fl4 = &newinet->cork.fl.u.ip4; 452 453 rcu_read_lock(); 454 opt = rcu_dereference(newinet->inet_opt); 455 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, 456 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 457 sk->sk_protocol, inet_sk_flowi_flags(sk), 458 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, 459 ireq->ir_loc_addr, ireq->ir_rmt_port, 460 htons(ireq->ir_num)); 461 security_req_classify_flow(req, flowi4_to_flowi(fl4)); 462 rt = ip_route_output_flow(net, fl4, sk); 463 if (IS_ERR(rt)) 464 goto no_route; 465 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) 466 goto route_err; 467 rcu_read_unlock(); 468 return &rt->dst; 469 470 route_err: 471 ip_rt_put(rt); 472 no_route: 473 rcu_read_unlock(); 474 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 475 return NULL; 476 } 477 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); 478 479 #if IS_ENABLED(CONFIG_IPV6) 480 #define AF_INET_FAMILY(fam) ((fam) == AF_INET) 481 #else 482 #define AF_INET_FAMILY(fam) true 483 #endif 484 485 /* Only thing we need from tcp.h */ 486 extern int sysctl_tcp_synack_retries; 487 488 489 /* Decide when to expire the request and when to resend SYN-ACK */ 490 static inline void syn_ack_recalc(struct request_sock *req, const int thresh, 491 const int max_retries, 492 const u8 rskq_defer_accept, 493 int *expire, int *resend) 494 { 495 if (!rskq_defer_accept) { 496 *expire = req->num_timeout >= thresh; 497 *resend = 1; 498 return; 499 } 500 *expire = req->num_timeout >= thresh && 501 (!inet_rsk(req)->acked || req->num_timeout >= max_retries); 502 /* 503 * Do not resend while waiting for data after ACK, 504 * start to resend on end of deferring period to give 505 * last chance for data or ACK to create established socket. 506 */ 507 *resend = !inet_rsk(req)->acked || 508 req->num_timeout >= rskq_defer_accept - 1; 509 } 510 511 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req) 512 { 513 int err = req->rsk_ops->rtx_syn_ack(parent, req); 514 515 if (!err) 516 req->num_retrans++; 517 return err; 518 } 519 EXPORT_SYMBOL(inet_rtx_syn_ack); 520 521 /* return true if req was found in the ehash table */ 522 static bool reqsk_queue_unlink(struct request_sock_queue *queue, 523 struct request_sock *req) 524 { 525 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo; 526 bool found = false; 527 528 if (sk_hashed(req_to_sk(req))) { 529 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash); 530 531 spin_lock(lock); 532 found = __sk_nulls_del_node_init_rcu(req_to_sk(req)); 533 spin_unlock(lock); 534 } 535 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer)) 536 reqsk_put(req); 537 return found; 538 } 539 540 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req) 541 { 542 if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) { 543 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); 544 reqsk_put(req); 545 } 546 } 547 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop); 548 549 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req) 550 { 551 inet_csk_reqsk_queue_drop(sk, req); 552 reqsk_put(req); 553 } 554 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put); 555 556 static void reqsk_timer_handler(unsigned long data) 557 { 558 struct request_sock *req = (struct request_sock *)data; 559 struct sock *sk_listener = req->rsk_listener; 560 struct inet_connection_sock *icsk = inet_csk(sk_listener); 561 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 562 int qlen, expire = 0, resend = 0; 563 int max_retries, thresh; 564 u8 defer_accept; 565 566 if (sk_state_load(sk_listener) != TCP_LISTEN) 567 goto drop; 568 569 max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries; 570 thresh = max_retries; 571 /* Normally all the openreqs are young and become mature 572 * (i.e. converted to established socket) for first timeout. 573 * If synack was not acknowledged for 1 second, it means 574 * one of the following things: synack was lost, ack was lost, 575 * rtt is high or nobody planned to ack (i.e. synflood). 576 * When server is a bit loaded, queue is populated with old 577 * open requests, reducing effective size of queue. 578 * When server is well loaded, queue size reduces to zero 579 * after several minutes of work. It is not synflood, 580 * it is normal operation. The solution is pruning 581 * too old entries overriding normal timeout, when 582 * situation becomes dangerous. 583 * 584 * Essentially, we reserve half of room for young 585 * embrions; and abort old ones without pity, if old 586 * ones are about to clog our table. 587 */ 588 qlen = reqsk_queue_len(queue); 589 if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) { 590 int young = reqsk_queue_len_young(queue) << 1; 591 592 while (thresh > 2) { 593 if (qlen < young) 594 break; 595 thresh--; 596 young <<= 1; 597 } 598 } 599 defer_accept = READ_ONCE(queue->rskq_defer_accept); 600 if (defer_accept) 601 max_retries = defer_accept; 602 syn_ack_recalc(req, thresh, max_retries, defer_accept, 603 &expire, &resend); 604 req->rsk_ops->syn_ack_timeout(req); 605 if (!expire && 606 (!resend || 607 !inet_rtx_syn_ack(sk_listener, req) || 608 inet_rsk(req)->acked)) { 609 unsigned long timeo; 610 611 if (req->num_timeout++ == 0) 612 atomic_dec(&queue->young); 613 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); 614 mod_timer_pinned(&req->rsk_timer, jiffies + timeo); 615 return; 616 } 617 drop: 618 inet_csk_reqsk_queue_drop_and_put(sk_listener, req); 619 } 620 621 static void reqsk_queue_hash_req(struct request_sock *req, 622 unsigned long timeout) 623 { 624 req->num_retrans = 0; 625 req->num_timeout = 0; 626 req->sk = NULL; 627 628 setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req); 629 mod_timer_pinned(&req->rsk_timer, jiffies + timeout); 630 631 inet_ehash_insert(req_to_sk(req), NULL); 632 /* before letting lookups find us, make sure all req fields 633 * are committed to memory and refcnt initialized. 634 */ 635 smp_wmb(); 636 atomic_set(&req->rsk_refcnt, 2 + 1); 637 } 638 639 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, 640 unsigned long timeout) 641 { 642 reqsk_queue_hash_req(req, timeout); 643 inet_csk_reqsk_queue_added(sk); 644 } 645 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); 646 647 /** 648 * inet_csk_clone_lock - clone an inet socket, and lock its clone 649 * @sk: the socket to clone 650 * @req: request_sock 651 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 652 * 653 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 654 */ 655 struct sock *inet_csk_clone_lock(const struct sock *sk, 656 const struct request_sock *req, 657 const gfp_t priority) 658 { 659 struct sock *newsk = sk_clone_lock(sk, priority); 660 661 if (newsk) { 662 struct inet_connection_sock *newicsk = inet_csk(newsk); 663 664 newsk->sk_state = TCP_SYN_RECV; 665 newicsk->icsk_bind_hash = NULL; 666 667 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port; 668 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num; 669 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num); 670 newsk->sk_write_space = sk_stream_write_space; 671 672 newsk->sk_mark = inet_rsk(req)->ir_mark; 673 atomic64_set(&newsk->sk_cookie, 674 atomic64_read(&inet_rsk(req)->ir_cookie)); 675 676 newicsk->icsk_retransmits = 0; 677 newicsk->icsk_backoff = 0; 678 newicsk->icsk_probes_out = 0; 679 680 /* Deinitialize accept_queue to trap illegal accesses. */ 681 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); 682 683 security_inet_csk_clone(newsk, req); 684 } 685 return newsk; 686 } 687 EXPORT_SYMBOL_GPL(inet_csk_clone_lock); 688 689 /* 690 * At this point, there should be no process reference to this 691 * socket, and thus no user references at all. Therefore we 692 * can assume the socket waitqueue is inactive and nobody will 693 * try to jump onto it. 694 */ 695 void inet_csk_destroy_sock(struct sock *sk) 696 { 697 WARN_ON(sk->sk_state != TCP_CLOSE); 698 WARN_ON(!sock_flag(sk, SOCK_DEAD)); 699 700 /* It cannot be in hash table! */ 701 WARN_ON(!sk_unhashed(sk)); 702 703 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ 704 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); 705 706 sk->sk_prot->destroy(sk); 707 708 sk_stream_kill_queues(sk); 709 710 xfrm_sk_free_policy(sk); 711 712 sk_refcnt_debug_release(sk); 713 714 percpu_counter_dec(sk->sk_prot->orphan_count); 715 sock_put(sk); 716 } 717 EXPORT_SYMBOL(inet_csk_destroy_sock); 718 719 /* This function allows to force a closure of a socket after the call to 720 * tcp/dccp_create_openreq_child(). 721 */ 722 void inet_csk_prepare_forced_close(struct sock *sk) 723 __releases(&sk->sk_lock.slock) 724 { 725 /* sk_clone_lock locked the socket and set refcnt to 2 */ 726 bh_unlock_sock(sk); 727 sock_put(sk); 728 729 /* The below has to be done to allow calling inet_csk_destroy_sock */ 730 sock_set_flag(sk, SOCK_DEAD); 731 percpu_counter_inc(sk->sk_prot->orphan_count); 732 inet_sk(sk)->inet_num = 0; 733 } 734 EXPORT_SYMBOL(inet_csk_prepare_forced_close); 735 736 int inet_csk_listen_start(struct sock *sk, int backlog) 737 { 738 struct inet_connection_sock *icsk = inet_csk(sk); 739 struct inet_sock *inet = inet_sk(sk); 740 741 reqsk_queue_alloc(&icsk->icsk_accept_queue); 742 743 sk->sk_max_ack_backlog = backlog; 744 sk->sk_ack_backlog = 0; 745 inet_csk_delack_init(sk); 746 747 /* There is race window here: we announce ourselves listening, 748 * but this transition is still not validated by get_port(). 749 * It is OK, because this socket enters to hash table only 750 * after validation is complete. 751 */ 752 sk_state_store(sk, TCP_LISTEN); 753 if (!sk->sk_prot->get_port(sk, inet->inet_num)) { 754 inet->inet_sport = htons(inet->inet_num); 755 756 sk_dst_reset(sk); 757 sk->sk_prot->hash(sk); 758 759 return 0; 760 } 761 762 sk->sk_state = TCP_CLOSE; 763 return -EADDRINUSE; 764 } 765 EXPORT_SYMBOL_GPL(inet_csk_listen_start); 766 767 static void inet_child_forget(struct sock *sk, struct request_sock *req, 768 struct sock *child) 769 { 770 sk->sk_prot->disconnect(child, O_NONBLOCK); 771 772 sock_orphan(child); 773 774 percpu_counter_inc(sk->sk_prot->orphan_count); 775 776 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) { 777 BUG_ON(tcp_sk(child)->fastopen_rsk != req); 778 BUG_ON(sk != req->rsk_listener); 779 780 /* Paranoid, to prevent race condition if 781 * an inbound pkt destined for child is 782 * blocked by sock lock in tcp_v4_rcv(). 783 * Also to satisfy an assertion in 784 * tcp_v4_destroy_sock(). 785 */ 786 tcp_sk(child)->fastopen_rsk = NULL; 787 } 788 inet_csk_destroy_sock(child); 789 reqsk_put(req); 790 } 791 792 void inet_csk_reqsk_queue_add(struct sock *sk, struct request_sock *req, 793 struct sock *child) 794 { 795 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; 796 797 spin_lock(&queue->rskq_lock); 798 if (unlikely(sk->sk_state != TCP_LISTEN)) { 799 inet_child_forget(sk, req, child); 800 } else { 801 req->sk = child; 802 req->dl_next = NULL; 803 if (queue->rskq_accept_head == NULL) 804 queue->rskq_accept_head = req; 805 else 806 queue->rskq_accept_tail->dl_next = req; 807 queue->rskq_accept_tail = req; 808 sk_acceptq_added(sk); 809 } 810 spin_unlock(&queue->rskq_lock); 811 } 812 EXPORT_SYMBOL(inet_csk_reqsk_queue_add); 813 814 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child, 815 struct request_sock *req, bool own_req) 816 { 817 if (own_req) { 818 inet_csk_reqsk_queue_drop(sk, req); 819 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); 820 inet_csk_reqsk_queue_add(sk, req, child); 821 /* Warning: caller must not call reqsk_put(req); 822 * child stole last reference on it. 823 */ 824 return child; 825 } 826 /* Too bad, another child took ownership of the request, undo. */ 827 bh_unlock_sock(child); 828 sock_put(child); 829 return NULL; 830 } 831 EXPORT_SYMBOL(inet_csk_complete_hashdance); 832 833 /* 834 * This routine closes sockets which have been at least partially 835 * opened, but not yet accepted. 836 */ 837 void inet_csk_listen_stop(struct sock *sk) 838 { 839 struct inet_connection_sock *icsk = inet_csk(sk); 840 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 841 struct request_sock *next, *req; 842 843 /* Following specs, it would be better either to send FIN 844 * (and enter FIN-WAIT-1, it is normal close) 845 * or to send active reset (abort). 846 * Certainly, it is pretty dangerous while synflood, but it is 847 * bad justification for our negligence 8) 848 * To be honest, we are not able to make either 849 * of the variants now. --ANK 850 */ 851 while ((req = reqsk_queue_remove(queue, sk)) != NULL) { 852 struct sock *child = req->sk; 853 854 local_bh_disable(); 855 bh_lock_sock(child); 856 WARN_ON(sock_owned_by_user(child)); 857 sock_hold(child); 858 859 inet_child_forget(sk, req, child); 860 bh_unlock_sock(child); 861 local_bh_enable(); 862 sock_put(child); 863 864 cond_resched(); 865 } 866 if (queue->fastopenq.rskq_rst_head) { 867 /* Free all the reqs queued in rskq_rst_head. */ 868 spin_lock_bh(&queue->fastopenq.lock); 869 req = queue->fastopenq.rskq_rst_head; 870 queue->fastopenq.rskq_rst_head = NULL; 871 spin_unlock_bh(&queue->fastopenq.lock); 872 while (req != NULL) { 873 next = req->dl_next; 874 reqsk_put(req); 875 req = next; 876 } 877 } 878 WARN_ON_ONCE(sk->sk_ack_backlog); 879 } 880 EXPORT_SYMBOL_GPL(inet_csk_listen_stop); 881 882 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) 883 { 884 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 885 const struct inet_sock *inet = inet_sk(sk); 886 887 sin->sin_family = AF_INET; 888 sin->sin_addr.s_addr = inet->inet_daddr; 889 sin->sin_port = inet->inet_dport; 890 } 891 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); 892 893 #ifdef CONFIG_COMPAT 894 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname, 895 char __user *optval, int __user *optlen) 896 { 897 const struct inet_connection_sock *icsk = inet_csk(sk); 898 899 if (icsk->icsk_af_ops->compat_getsockopt) 900 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname, 901 optval, optlen); 902 return icsk->icsk_af_ops->getsockopt(sk, level, optname, 903 optval, optlen); 904 } 905 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt); 906 907 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname, 908 char __user *optval, unsigned int optlen) 909 { 910 const struct inet_connection_sock *icsk = inet_csk(sk); 911 912 if (icsk->icsk_af_ops->compat_setsockopt) 913 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname, 914 optval, optlen); 915 return icsk->icsk_af_ops->setsockopt(sk, level, optname, 916 optval, optlen); 917 } 918 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt); 919 #endif 920 921 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl) 922 { 923 const struct inet_sock *inet = inet_sk(sk); 924 const struct ip_options_rcu *inet_opt; 925 __be32 daddr = inet->inet_daddr; 926 struct flowi4 *fl4; 927 struct rtable *rt; 928 929 rcu_read_lock(); 930 inet_opt = rcu_dereference(inet->inet_opt); 931 if (inet_opt && inet_opt->opt.srr) 932 daddr = inet_opt->opt.faddr; 933 fl4 = &fl->u.ip4; 934 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, 935 inet->inet_saddr, inet->inet_dport, 936 inet->inet_sport, sk->sk_protocol, 937 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if); 938 if (IS_ERR(rt)) 939 rt = NULL; 940 if (rt) 941 sk_setup_caps(sk, &rt->dst); 942 rcu_read_unlock(); 943 944 return &rt->dst; 945 } 946 947 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu) 948 { 949 struct dst_entry *dst = __sk_dst_check(sk, 0); 950 struct inet_sock *inet = inet_sk(sk); 951 952 if (!dst) { 953 dst = inet_csk_rebuild_route(sk, &inet->cork.fl); 954 if (!dst) 955 goto out; 956 } 957 dst->ops->update_pmtu(dst, sk, NULL, mtu); 958 959 dst = __sk_dst_check(sk, 0); 960 if (!dst) 961 dst = inet_csk_rebuild_route(sk, &inet->cork.fl); 962 out: 963 return dst; 964 } 965 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu); 966