net/ipv4/inet_connection_sock.c

   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);
 334         newsk = req->sk;
 335
 336         sk_acceptq_removed(sk);
 337         if (sk->sk_protocol == IPPROTO_TCP &&
 338             tcp_rsk(req)->tfo_listener &&
 339             queue->fastopenq) {
 340                 spin_lock_bh(&queue->fastopenq->lock);
 341                 if (tcp_rsk(req)->tfo_listener) {
 342                         /* We are still waiting for the final ACK from 3WHS
 343                          * so can't free req now. Instead, we set req->sk to
 344                          * NULL to signify that the child socket is taken
 345                          * so reqsk_fastopen_remove() will free the req
 346                          * when 3WHS finishes (or is aborted).
 347                          */
 348                         req->sk = NULL;
 349                         req = NULL;
 350                 }
 351                 spin_unlock_bh(&queue->fastopenq->lock);
 352         }
 353 out:
 354         release_sock(sk);
 355         if (req)
 356                 reqsk_put(req);
 357         return newsk;
 358 out_err:
 359         newsk = NULL;
 360         req = NULL;
 361         *err = error;
 362         goto out;
 363 }
 364 EXPORT_SYMBOL(inet_csk_accept);
 365
 366 /*
 367  * Using different timers for retransmit, delayed acks and probes
 368  * We may wish use just one timer maintaining a list of expire jiffies
 369  * to optimize.
 370  */
 371 void inet_csk_init_xmit_timers(struct sock *sk,
 372                                void (*retransmit_handler)(unsigned long),
 373                                void (*delack_handler)(unsigned long),
 374                                void (*keepalive_handler)(unsigned long))
 375 {
 376         struct inet_connection_sock *icsk = inet_csk(sk);
 377
 378         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 379                         (unsigned long)sk);
 380         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 381                         (unsigned long)sk);
 382         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 383         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 384 }
 385 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 386
 387 void inet_csk_clear_xmit_timers(struct sock *sk)
 388 {
 389         struct inet_connection_sock *icsk = inet_csk(sk);
 390
 391         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 392
 393         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 394         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 395         sk_stop_timer(sk, &sk->sk_timer);
 396 }
 397 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 398
 399 void inet_csk_delete_keepalive_timer(struct sock *sk)
 400 {
 401         sk_stop_timer(sk, &sk->sk_timer);
 402 }
 403 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 404
 405 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 406 {
 407         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 408 }
 409 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 410
 411 struct dst_entry *inet_csk_route_req(struct sock *sk,
 412                                      struct flowi4 *fl4,
 413                                      const struct request_sock *req)
 414 {
 415         const struct inet_request_sock *ireq = inet_rsk(req);
 416         struct net *net = read_pnet(&ireq->ireq_net);
 417         struct ip_options_rcu *opt = ireq->opt;
 418         struct rtable *rt;
 419
 420         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 421                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 422                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 423                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 424                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 425                            htons(ireq->ir_num));
 426         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 427         rt = ip_route_output_flow(net, fl4, sk);
 428         if (IS_ERR(rt))
 429                 goto no_route;
 430         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 431                 goto route_err;
 432         return &rt->dst;
 433
 434 route_err:
 435         ip_rt_put(rt);
 436 no_route:
 437         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 438         return NULL;
 439 }
 440 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 441
 442 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
 443                                             struct sock *newsk,
 444                                             const struct request_sock *req)
 445 {
 446         const struct inet_request_sock *ireq = inet_rsk(req);
 447         struct net *net = read_pnet(&ireq->ireq_net);
 448         struct inet_sock *newinet = inet_sk(newsk);
 449         struct ip_options_rcu *opt;
 450         struct flowi4 *fl4;
 451         struct rtable *rt;
 452
 453         fl4 = &newinet->cork.fl.u.ip4;
 454
 455         rcu_read_lock();
 456         opt = rcu_dereference(newinet->inet_opt);
 457         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 458                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 459                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 460                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 461                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 462                            htons(ireq->ir_num));
 463         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 464         rt = ip_route_output_flow(net, fl4, sk);
 465         if (IS_ERR(rt))
 466                 goto no_route;
 467         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 468                 goto route_err;
 469         rcu_read_unlock();
 470         return &rt->dst;
 471
 472 route_err:
 473         ip_rt_put(rt);
 474 no_route:
 475         rcu_read_unlock();
 476         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 477         return NULL;
 478 }
 479 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 480
 481 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 482                                  const u32 rnd, const u32 synq_hsize)
 483 {
 484         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 485 }
 486
 487 #if IS_ENABLED(CONFIG_IPV6)
 488 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 489 #else
 490 #define AF_INET_FAMILY(fam) true
 491 #endif
 492
 493 /* Note: this is temporary :
 494  * req sock will no longer be in listener hash table
 495 */
 496 struct request_sock *inet_csk_search_req(struct sock *sk,
 497                                          const __be16 rport,
 498                                          const __be32 raddr,
 499                                          const __be32 laddr)
 500 {
 501         struct inet_connection_sock *icsk = inet_csk(sk);
 502         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 503         struct request_sock *req;
 504         u32 hash = inet_synq_hash(raddr, rport, lopt->hash_rnd,
 505                                   lopt->nr_table_entries);
 506
 507         spin_lock(&icsk->icsk_accept_queue.syn_wait_lock);
 508         for (req = lopt->syn_table[hash]; req != NULL; req = req->dl_next) {
 509                 const struct inet_request_sock *ireq = inet_rsk(req);
 510
 511                 if (ireq->ir_rmt_port == rport &&
 512                     ireq->ir_rmt_addr == raddr &&
 513                     ireq->ir_loc_addr == laddr &&
 514                     AF_INET_FAMILY(req->rsk_ops->family)) {
 515                         atomic_inc(&req->rsk_refcnt);
 516                         WARN_ON(req->sk);
 517                         break;
 518                 }
 519         }
 520         spin_unlock(&icsk->icsk_accept_queue.syn_wait_lock);
 521
 522         return req;
 523 }
 524 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 525
 526 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 527                                    unsigned long timeout)
 528 {
 529         struct inet_connection_sock *icsk = inet_csk(sk);
 530         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 531         const u32 h = inet_synq_hash(inet_rsk(req)->ir_rmt_addr,
 532                                      inet_rsk(req)->ir_rmt_port,
 533                                      lopt->hash_rnd, lopt->nr_table_entries);
 534
 535         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 536         inet_csk_reqsk_queue_added(sk, timeout);
 537 }
 538 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 539
 540 /* Only thing we need from tcp.h */
 541 extern int sysctl_tcp_synack_retries;
 542
 543
 544 /* Decide when to expire the request and when to resend SYN-ACK */
 545 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 546                                   const int max_retries,
 547                                   const u8 rskq_defer_accept,
 548                                   int *expire, int *resend)
 549 {
 550         if (!rskq_defer_accept) {
 551                 *expire = req->num_timeout >= thresh;
 552                 *resend = 1;
 553                 return;
 554         }
 555         *expire = req->num_timeout >= thresh &&
 556                   (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 557         /*
 558          * Do not resend while waiting for data after ACK,
 559          * start to resend on end of deferring period to give
 560          * last chance for data or ACK to create established socket.
 561          */
 562         *resend = !inet_rsk(req)->acked ||
 563                   req->num_timeout >= rskq_defer_accept - 1;
 564 }
 565
 566 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req)
 567 {
 568         int err = req->rsk_ops->rtx_syn_ack(parent, req);
 569
 570         if (!err)
 571                 req->num_retrans++;
 572         return err;
 573 }
 574 EXPORT_SYMBOL(inet_rtx_syn_ack);
 575
 576 /* return true if req was found in the syn_table[] */
 577 static bool reqsk_queue_unlink(struct request_sock_queue *queue,
 578                                struct request_sock *req)
 579 {
 580         struct listen_sock *lopt = queue->listen_opt;
 581         struct request_sock **prev;
 582         bool found = false;
 583
 584         spin_lock(&queue->syn_wait_lock);
 585
 586         for (prev = &lopt->syn_table[req->rsk_hash]; *prev != NULL;
 587              prev = &(*prev)->dl_next) {
 588                 if (*prev == req) {
 589                         *prev = req->dl_next;
 590                         found = true;
 591                         break;
 592                 }
 593         }
 594
 595         spin_unlock(&queue->syn_wait_lock);
 596         if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
 597                 reqsk_put(req);
 598         return found;
 599 }
 600
 601 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
 602 {
 603         if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
 604                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 605                 reqsk_put(req);
 606         }
 607 }
 608 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
 609
 610 static void reqsk_timer_handler(unsigned long data)
 611 {
 612         struct request_sock *req = (struct request_sock *)data;
 613         struct sock *sk_listener = req->rsk_listener;
 614         struct inet_connection_sock *icsk = inet_csk(sk_listener);
 615         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 616         struct listen_sock *lopt = queue->listen_opt;
 617         int qlen, expire = 0, resend = 0;
 618         int max_retries, thresh;
 619         u8 defer_accept;
 620
 621         if (sk_listener->sk_state != TCP_LISTEN || !lopt) {
 622                 reqsk_put(req);
 623                 return;
 624         }
 625
 626         max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 627         thresh = max_retries;
 628         /* Normally all the openreqs are young and become mature
 629          * (i.e. converted to established socket) for first timeout.
 630          * If synack was not acknowledged for 1 second, it means
 631          * one of the following things: synack was lost, ack was lost,
 632          * rtt is high or nobody planned to ack (i.e. synflood).
 633          * When server is a bit loaded, queue is populated with old
 634          * open requests, reducing effective size of queue.
 635          * When server is well loaded, queue size reduces to zero
 636          * after several minutes of work. It is not synflood,
 637          * it is normal operation. The solution is pruning
 638          * too old entries overriding normal timeout, when
 639          * situation becomes dangerous.
 640          *
 641          * Essentially, we reserve half of room for young
 642          * embrions; and abort old ones without pity, if old
 643          * ones are about to clog our table.
 644          */
 645         qlen = listen_sock_qlen(lopt);
 646         if (qlen >> (lopt->max_qlen_log - 1)) {
 647                 int young = listen_sock_young(lopt) << 1;
 648
 649                 while (thresh > 2) {
 650                         if (qlen < young)
 651                                 break;
 652                         thresh--;
 653                         young <<= 1;
 654                 }
 655         }
 656         defer_accept = READ_ONCE(queue->rskq_defer_accept);
 657         if (defer_accept)
 658                 max_retries = defer_accept;
 659         syn_ack_recalc(req, thresh, max_retries, defer_accept,
 660                        &expire, &resend);
 661         req->rsk_ops->syn_ack_timeout(req);
 662         if (!expire &&
 663             (!resend ||
 664              !inet_rtx_syn_ack(sk_listener, req) ||
 665              inet_rsk(req)->acked)) {
 666                 unsigned long timeo;
 667
 668                 if (req->num_timeout++ == 0)
 669                         atomic_inc(&lopt->young_dec);
 670                 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 671                 mod_timer_pinned(&req->rsk_timer, jiffies + timeo);
 672                 return;
 673         }
 674         inet_csk_reqsk_queue_drop(sk_listener, req);
 675         reqsk_put(req);
 676 }
 677
 678 void reqsk_queue_hash_req(struct request_sock_queue *queue,
 679                           u32 hash, struct request_sock *req,
 680                           unsigned long timeout)
 681 {
 682         struct listen_sock *lopt = queue->listen_opt;
 683
 684         req->num_retrans = 0;
 685         req->num_timeout = 0;
 686         req->sk = NULL;
 687
 688         setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req);
 689         mod_timer_pinned(&req->rsk_timer, jiffies + timeout);
 690         req->rsk_hash = hash;
 691
 692         /* before letting lookups find us, make sure all req fields
 693          * are committed to memory and refcnt initialized.
 694          */
 695         smp_wmb();
 696         atomic_set(&req->rsk_refcnt, 2);
 697
 698         spin_lock(&queue->syn_wait_lock);
 699         req->dl_next = lopt->syn_table[hash];
 700         lopt->syn_table[hash] = req;
 701         spin_unlock(&queue->syn_wait_lock);
 702 }
 703 EXPORT_SYMBOL(reqsk_queue_hash_req);
 704
 705 /**
 706  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 707  *      @sk: the socket to clone
 708  *      @req: request_sock
 709  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 710  *
 711  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 712  */
 713 struct sock *inet_csk_clone_lock(const struct sock *sk,
 714                                  const struct request_sock *req,
 715                                  const gfp_t priority)
 716 {
 717         struct sock *newsk = sk_clone_lock(sk, priority);
 718
 719         if (newsk) {
 720                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 721
 722                 newsk->sk_state = TCP_SYN_RECV;
 723                 newicsk->icsk_bind_hash = NULL;
 724
 725                 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
 726                 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
 727                 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
 728                 newsk->sk_write_space = sk_stream_write_space;
 729
 730                 newsk->sk_mark = inet_rsk(req)->ir_mark;
 731                 atomic64_set(&newsk->sk_cookie,
 732                              atomic64_read(&inet_rsk(req)->ir_cookie));
 733
 734                 newicsk->icsk_retransmits = 0;
 735                 newicsk->icsk_backoff     = 0;
 736                 newicsk->icsk_probes_out  = 0;
 737
 738                 /* Deinitialize accept_queue to trap illegal accesses. */
 739                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 740
 741                 security_inet_csk_clone(newsk, req);
 742         }
 743         return newsk;
 744 }
 745 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 746
 747 /*
 748  * At this point, there should be no process reference to this
 749  * socket, and thus no user references at all.  Therefore we
 750  * can assume the socket waitqueue is inactive and nobody will
 751  * try to jump onto it.
 752  */
 753 void inet_csk_destroy_sock(struct sock *sk)
 754 {
 755         WARN_ON(sk->sk_state != TCP_CLOSE);
 756         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 757
 758         /* It cannot be in hash table! */
 759         WARN_ON(!sk_unhashed(sk));
 760
 761         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 762         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 763
 764         sk->sk_prot->destroy(sk);
 765
 766         sk_stream_kill_queues(sk);
 767
 768         xfrm_sk_free_policy(sk);
 769
 770         sk_refcnt_debug_release(sk);
 771
 772         percpu_counter_dec(sk->sk_prot->orphan_count);
 773         sock_put(sk);
 774 }
 775 EXPORT_SYMBOL(inet_csk_destroy_sock);
 776
 777 /* This function allows to force a closure of a socket after the call to
 778  * tcp/dccp_create_openreq_child().
 779  */
 780 void inet_csk_prepare_forced_close(struct sock *sk)
 781         __releases(&sk->sk_lock.slock)
 782 {
 783         /* sk_clone_lock locked the socket and set refcnt to 2 */
 784         bh_unlock_sock(sk);
 785         sock_put(sk);
 786
 787         /* The below has to be done to allow calling inet_csk_destroy_sock */
 788         sock_set_flag(sk, SOCK_DEAD);
 789         percpu_counter_inc(sk->sk_prot->orphan_count);
 790         inet_sk(sk)->inet_num = 0;
 791 }
 792 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
 793
 794 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 795 {
 796         struct inet_sock *inet = inet_sk(sk);
 797         struct inet_connection_sock *icsk = inet_csk(sk);
 798         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 799
 800         if (rc != 0)
 801                 return rc;
 802
 803         sk->sk_max_ack_backlog = 0;
 804         sk->sk_ack_backlog = 0;
 805         inet_csk_delack_init(sk);
 806
 807         /* There is race window here: we announce ourselves listening,
 808          * but this transition is still not validated by get_port().
 809          * It is OK, because this socket enters to hash table only
 810          * after validation is complete.
 811          */
 812         sk->sk_state = TCP_LISTEN;
 813         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 814                 inet->inet_sport = htons(inet->inet_num);
 815
 816                 sk_dst_reset(sk);
 817                 sk->sk_prot->hash(sk);
 818
 819                 return 0;
 820         }
 821
 822         sk->sk_state = TCP_CLOSE;
 823         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 824         return -EADDRINUSE;
 825 }
 826 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 827
 828 /*
 829  *      This routine closes sockets which have been at least partially
 830  *      opened, but not yet accepted.
 831  */
 832 void inet_csk_listen_stop(struct sock *sk)
 833 {
 834         struct inet_connection_sock *icsk = inet_csk(sk);
 835         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 836         struct request_sock *acc_req;
 837         struct request_sock *req;
 838
 839         /* make all the listen_opt local to us */
 840         acc_req = reqsk_queue_yank_acceptq(queue);
 841
 842         /* Following specs, it would be better either to send FIN
 843          * (and enter FIN-WAIT-1, it is normal close)
 844          * or to send active reset (abort).
 845          * Certainly, it is pretty dangerous while synflood, but it is
 846          * bad justification for our negligence 8)
 847          * To be honest, we are not able to make either
 848          * of the variants now.                 --ANK
 849          */
 850         reqsk_queue_destroy(queue);
 851
 852         while ((req = acc_req) != NULL) {
 853                 struct sock *child = req->sk;
 854
 855                 acc_req = req->dl_next;
 856
 857                 local_bh_disable();
 858                 bh_lock_sock(child);
 859                 WARN_ON(sock_owned_by_user(child));
 860                 sock_hold(child);
 861
 862                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 863
 864                 sock_orphan(child);
 865
 866                 percpu_counter_inc(sk->sk_prot->orphan_count);
 867
 868                 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
 869                         BUG_ON(tcp_sk(child)->fastopen_rsk != req);
 870                         BUG_ON(sk != req->rsk_listener);
 871
 872                         /* Paranoid, to prevent race condition if
 873                          * an inbound pkt destined for child is
 874                          * blocked by sock lock in tcp_v4_rcv().
 875                          * Also to satisfy an assertion in
 876                          * tcp_v4_destroy_sock().
 877                          */
 878                         tcp_sk(child)->fastopen_rsk = NULL;
 879                 }
 880                 inet_csk_destroy_sock(child);
 881
 882                 bh_unlock_sock(child);
 883                 local_bh_enable();
 884                 sock_put(child);
 885
 886                 sk_acceptq_removed(sk);
 887                 reqsk_put(req);
 888         }
 889         if (queue->fastopenq) {
 890                 /* Free all the reqs queued in rskq_rst_head. */
 891                 spin_lock_bh(&queue->fastopenq->lock);
 892                 acc_req = queue->fastopenq->rskq_rst_head;
 893                 queue->fastopenq->rskq_rst_head = NULL;
 894                 spin_unlock_bh(&queue->fastopenq->lock);
 895                 while ((req = acc_req) != NULL) {
 896                         acc_req = req->dl_next;
 897                         reqsk_put(req);
 898                 }
 899         }
 900         WARN_ON(sk->sk_ack_backlog);
 901 }
 902 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 903
 904 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 905 {
 906         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 907         const struct inet_sock *inet = inet_sk(sk);
 908
 909         sin->sin_family         = AF_INET;
 910         sin->sin_addr.s_addr    = inet->inet_daddr;
 911         sin->sin_port           = inet->inet_dport;
 912 }
 913 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 914
 915 #ifdef CONFIG_COMPAT
 916 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 917                                char __user *optval, int __user *optlen)
 918 {
 919         const struct inet_connection_sock *icsk = inet_csk(sk);
 920
 921         if (icsk->icsk_af_ops->compat_getsockopt)
 922                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 923                                                             optval, optlen);
 924         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 925                                              optval, optlen);
 926 }
 927 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 928
 929 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 930                                char __user *optval, unsigned int optlen)
 931 {
 932         const struct inet_connection_sock *icsk = inet_csk(sk);
 933
 934         if (icsk->icsk_af_ops->compat_setsockopt)
 935                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 936                                                             optval, optlen);
 937         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 938                                              optval, optlen);
 939 }
 940 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 941 #endif
 942
 943 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
 944 {
 945         const struct inet_sock *inet = inet_sk(sk);
 946         const struct ip_options_rcu *inet_opt;
 947         __be32 daddr = inet->inet_daddr;
 948         struct flowi4 *fl4;
 949         struct rtable *rt;
 950
 951         rcu_read_lock();
 952         inet_opt = rcu_dereference(inet->inet_opt);
 953         if (inet_opt && inet_opt->opt.srr)
 954                 daddr = inet_opt->opt.faddr;
 955         fl4 = &fl->u.ip4;
 956         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
 957                                    inet->inet_saddr, inet->inet_dport,
 958                                    inet->inet_sport, sk->sk_protocol,
 959                                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
 960         if (IS_ERR(rt))
 961                 rt = NULL;
 962         if (rt)
 963                 sk_setup_caps(sk, &rt->dst);
 964         rcu_read_unlock();
 965
 966         return &rt->dst;
 967 }
 968
 969 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
 970 {
 971         struct dst_entry *dst = __sk_dst_check(sk, 0);
 972         struct inet_sock *inet = inet_sk(sk);
 973
 974         if (!dst) {
 975                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 976                 if (!dst)
 977                         goto out;
 978         }
 979         dst->ops->update_pmtu(dst, sk, NULL, mtu);
 980
 981         dst = __sk_dst_check(sk, 0);
 982         if (!dst)
 983                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 984 out:
 985         return dst;
 986 }
 987 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);