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 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
  91                                struct sock *sk)
  92 {
  93         kuid_t uid = sock_i_uid(sk);
  94
  95         if (hlist_empty(&tb->owners)) {
  96                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
  97                         tb->fastreuse = 1;
  98                 else
  99                         tb->fastreuse = 0;
 100                 if (sk->sk_reuseport) {
 101                         tb->fastreuseport = 1;
 102                         tb->fastuid = uid;
 103                 } else
 104                         tb->fastreuseport = 0;
 105         } else {
 106                 if (tb->fastreuse &&
 107                     (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 108                         tb->fastreuse = 0;
 109                 if (tb->fastreuseport &&
 110                     (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)))
 111                         tb->fastreuseport = 0;
 112         }
 113 }
 114
 115 /* Obtain a reference to a local port for the given sock,
 116  * if snum is zero it means select any available local port.
 117  */
 118 int inet_csk_get_port(struct sock *sk, unsigned short snum)
 119 {
 120         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 121         struct inet_bind_hashbucket *head;
 122         struct inet_bind_bucket *tb;
 123         int ret, attempts = 5;
 124         struct net *net = sock_net(sk);
 125         int smallest_size = -1, smallest_rover;
 126         kuid_t uid = sock_i_uid(sk);
 127         int attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
 128
 129         local_bh_disable();
 130         if (!snum) {
 131                 int remaining, rover, low, high;
 132
 133 again:
 134                 inet_get_local_port_range(net, &low, &high);
 135                 if (attempt_half) {
 136                         int half = low + ((high - low) >> 1);
 137
 138                         if (attempt_half == 1)
 139                                 high = half;
 140                         else
 141                                 low = half;
 142                 }
 143                 remaining = (high - low) + 1;
 144                 smallest_rover = rover = prandom_u32() % remaining + low;
 145
 146                 smallest_size = -1;
 147                 do {
 148                         if (inet_is_local_reserved_port(net, rover))
 149                                 goto next_nolock;
 150                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 151                                         hashinfo->bhash_size)];
 152                         spin_lock(&head->lock);
 153                         inet_bind_bucket_for_each(tb, &head->chain)
 154                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 155                                         if (((tb->fastreuse > 0 &&
 156                                               sk->sk_reuse &&
 157                                               sk->sk_state != TCP_LISTEN) ||
 158                                              (tb->fastreuseport > 0 &&
 159                                               sk->sk_reuseport &&
 160                                               uid_eq(tb->fastuid, uid))) &&
 161                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 162                                                 smallest_size = tb->num_owners;
 163                                                 smallest_rover = rover;
 164                                         }
 165                                         if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
 166                                                 snum = rover;
 167                                                 goto tb_found;
 168                                         }
 169                                         goto next;
 170                                 }
 171                         break;
 172                 next:
 173                         spin_unlock(&head->lock);
 174                 next_nolock:
 175                         if (++rover > high)
 176                                 rover = low;
 177                 } while (--remaining > 0);
 178
 179                 /* Exhausted local port range during search?  It is not
 180                  * possible for us to be holding one of the bind hash
 181                  * locks if this test triggers, because if 'remaining'
 182                  * drops to zero, we broke out of the do/while loop at
 183                  * the top level, not from the 'break;' statement.
 184                  */
 185                 ret = 1;
 186                 if (remaining <= 0) {
 187                         if (smallest_size != -1) {
 188                                 snum = smallest_rover;
 189                                 goto have_snum;
 190                         }
 191                         if (attempt_half == 1) {
 192                                 /* OK we now try the upper half of the range */
 193                                 attempt_half = 2;
 194                                 goto again;
 195                         }
 196                         goto fail;
 197                 }
 198                 /* OK, here is the one we will use.  HEAD is
 199                  * non-NULL and we hold it's mutex.
 200                  */
 201                 snum = rover;
 202         } else {
 203 have_snum:
 204                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 205                                 hashinfo->bhash_size)];
 206                 spin_lock(&head->lock);
 207                 inet_bind_bucket_for_each(tb, &head->chain)
 208                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 209                                 goto tb_found;
 210         }
 211         tb = NULL;
 212         goto tb_not_found;
 213 tb_found:
 214         if (!hlist_empty(&tb->owners)) {
 215                 if (sk->sk_reuse == SK_FORCE_REUSE)
 216                         goto success;
 217
 218                 if (((tb->fastreuse > 0 &&
 219                       sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 220                      (tb->fastreuseport > 0 &&
 221                       sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
 222                     smallest_size == -1) {
 223                         goto success;
 224                 } else {
 225                         ret = 1;
 226                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
 227                                 if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 228                                      (tb->fastreuseport > 0 &&
 229                                       sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
 230                                     smallest_size != -1 && --attempts >= 0) {
 231                                         spin_unlock(&head->lock);
 232                                         goto again;
 233                                 }
 234
 235                                 goto fail_unlock;
 236                         }
 237                 }
 238         }
 239 tb_not_found:
 240         ret = 1;
 241         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 242                                         net, head, snum)) == NULL)
 243                 goto fail_unlock;
 244
 245         inet_csk_update_fastreuse(tb, sk);
 246
 247 success:
 248         if (!inet_csk(sk)->icsk_bind_hash)
 249                 inet_bind_hash(sk, tb, snum);
 250         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 251         ret = 0;
 252
 253 fail_unlock:
 254         spin_unlock(&head->lock);
 255 fail:
 256         local_bh_enable();
 257         return ret;
 258 }
 259 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 260
 261 /*
 262  * Wait for an incoming connection, avoid race conditions. This must be called
 263  * with the socket locked.
 264  */
 265 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 266 {
 267         struct inet_connection_sock *icsk = inet_csk(sk);
 268         DEFINE_WAIT(wait);
 269         int err;
 270
 271         /*
 272          * True wake-one mechanism for incoming connections: only
 273          * one process gets woken up, not the 'whole herd'.
 274          * Since we do not 'race & poll' for established sockets
 275          * anymore, the common case will execute the loop only once.
 276          *
 277          * Subtle issue: "add_wait_queue_exclusive()" will be added
 278          * after any current non-exclusive waiters, and we know that
 279          * it will always _stay_ after any new non-exclusive waiters
 280          * because all non-exclusive waiters are added at the
 281          * beginning of the wait-queue. As such, it's ok to "drop"
 282          * our exclusiveness temporarily when we get woken up without
 283          * having to remove and re-insert us on the wait queue.
 284          */
 285         for (;;) {
 286                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 287                                           TASK_INTERRUPTIBLE);
 288                 release_sock(sk);
 289                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 290                         timeo = schedule_timeout(timeo);
 291                 sched_annotate_sleep();
 292                 lock_sock(sk);
 293                 err = 0;
 294                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 295                         break;
 296                 err = -EINVAL;
 297                 if (sk->sk_state != TCP_LISTEN)
 298                         break;
 299                 err = sock_intr_errno(timeo);
 300                 if (signal_pending(current))
 301                         break;
 302                 err = -EAGAIN;
 303                 if (!timeo)
 304                         break;
 305         }
 306         finish_wait(sk_sleep(sk), &wait);
 307         return err;
 308 }
 309
 310 /*
 311  * This will accept the next outstanding connection.
 312  */
 313 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 314 {
 315         struct inet_connection_sock *icsk = inet_csk(sk);
 316         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 317         struct request_sock *req;
 318         struct sock *newsk;
 319         int error;
 320
 321         lock_sock(sk);
 322
 323         /* We need to make sure that this socket is listening,
 324          * and that it has something pending.
 325          */
 326         error = -EINVAL;
 327         if (sk->sk_state != TCP_LISTEN)
 328                 goto out_err;
 329
 330         /* Find already established connection */
 331         if (reqsk_queue_empty(queue)) {
 332                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 333
 334                 /* If this is a non blocking socket don't sleep */
 335                 error = -EAGAIN;
 336                 if (!timeo)
 337                         goto out_err;
 338
 339                 error = inet_csk_wait_for_connect(sk, timeo);
 340                 if (error)
 341                         goto out_err;
 342         }
 343         req = reqsk_queue_remove(queue, sk);
 344         newsk = req->sk;
 345
 346         if (sk->sk_protocol == IPPROTO_TCP &&
 347             tcp_rsk(req)->tfo_listener) {
 348                 spin_lock_bh(&queue->fastopenq.lock);
 349                 if (tcp_rsk(req)->tfo_listener) {
 350                         /* We are still waiting for the final ACK from 3WHS
 351                          * so can't free req now. Instead, we set req->sk to
 352                          * NULL to signify that the child socket is taken
 353                          * so reqsk_fastopen_remove() will free the req
 354                          * when 3WHS finishes (or is aborted).
 355                          */
 356                         req->sk = NULL;
 357                         req = NULL;
 358                 }
 359                 spin_unlock_bh(&queue->fastopenq.lock);
 360         }
 361 out:
 362         release_sock(sk);
 363         if (req)
 364                 reqsk_put(req);
 365         return newsk;
 366 out_err:
 367         newsk = NULL;
 368         req = NULL;
 369         *err = error;
 370         goto out;
 371 }
 372 EXPORT_SYMBOL(inet_csk_accept);
 373
 374 /*
 375  * Using different timers for retransmit, delayed acks and probes
 376  * We may wish use just one timer maintaining a list of expire jiffies
 377  * to optimize.
 378  */
 379 void inet_csk_init_xmit_timers(struct sock *sk,
 380                                void (*retransmit_handler)(unsigned long),
 381                                void (*delack_handler)(unsigned long),
 382                                void (*keepalive_handler)(unsigned long))
 383 {
 384         struct inet_connection_sock *icsk = inet_csk(sk);
 385
 386         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 387                         (unsigned long)sk);
 388         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 389                         (unsigned long)sk);
 390         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 391         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 392 }
 393 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 394
 395 void inet_csk_clear_xmit_timers(struct sock *sk)
 396 {
 397         struct inet_connection_sock *icsk = inet_csk(sk);
 398
 399         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 400
 401         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 402         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 403         sk_stop_timer(sk, &sk->sk_timer);
 404 }
 405 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 406
 407 void inet_csk_delete_keepalive_timer(struct sock *sk)
 408 {
 409         sk_stop_timer(sk, &sk->sk_timer);
 410 }
 411 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 412
 413 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 414 {
 415         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 416 }
 417 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 418
 419 struct dst_entry *inet_csk_route_req(const struct sock *sk,
 420                                      struct flowi4 *fl4,
 421                                      const struct request_sock *req)
 422 {
 423         const struct inet_request_sock *ireq = inet_rsk(req);
 424         struct net *net = read_pnet(&ireq->ireq_net);
 425         struct ip_options_rcu *opt;
 426         struct rtable *rt;
 427
 428         opt = ireq_opt_deref(ireq);
 429
 430         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 431                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 432                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 433                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 434                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 435                            htons(ireq->ir_num));
 436         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 437         rt = ip_route_output_flow(net, fl4, sk);
 438         if (IS_ERR(rt))
 439                 goto no_route;
 440         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 441                 goto route_err;
 442         return &rt->dst;
 443
 444 route_err:
 445         ip_rt_put(rt);
 446 no_route:
 447         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 448         return NULL;
 449 }
 450 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 451
 452 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
 453                                             struct sock *newsk,
 454                                             const struct request_sock *req)
 455 {
 456         const struct inet_request_sock *ireq = inet_rsk(req);
 457         struct net *net = read_pnet(&ireq->ireq_net);
 458         struct inet_sock *newinet = inet_sk(newsk);
 459         struct ip_options_rcu *opt;
 460         struct flowi4 *fl4;
 461         struct rtable *rt;
 462
 463         opt = rcu_dereference(ireq->ireq_opt);
 464         fl4 = &newinet->cork.fl.u.ip4;
 465
 466         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 467                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 468                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 469                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 470                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 471                            htons(ireq->ir_num));
 472         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 473         rt = ip_route_output_flow(net, fl4, sk);
 474         if (IS_ERR(rt))
 475                 goto no_route;
 476         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 477                 goto route_err;
 478         return &rt->dst;
 479
 480 route_err:
 481         ip_rt_put(rt);
 482 no_route:
 483         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 484         return NULL;
 485 }
 486 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 487
 488 #if IS_ENABLED(CONFIG_IPV6)
 489 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 490 #else
 491 #define AF_INET_FAMILY(fam) true
 492 #endif
 493
 494 /* Only thing we need from tcp.h */
 495 extern int sysctl_tcp_synack_retries;
 496
 497
 498 /* Decide when to expire the request and when to resend SYN-ACK */
 499 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 500                                   const int max_retries,
 501                                   const u8 rskq_defer_accept,
 502                                   int *expire, int *resend)
 503 {
 504         if (!rskq_defer_accept) {
 505                 *expire = req->num_timeout >= thresh;
 506                 *resend = 1;
 507                 return;
 508         }
 509         *expire = req->num_timeout >= thresh &&
 510                   (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 511         /*
 512          * Do not resend while waiting for data after ACK,
 513          * start to resend on end of deferring period to give
 514          * last chance for data or ACK to create established socket.
 515          */
 516         *resend = !inet_rsk(req)->acked ||
 517                   req->num_timeout >= rskq_defer_accept - 1;
 518 }
 519
 520 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
 521 {
 522         int err = req->rsk_ops->rtx_syn_ack(parent, req);
 523
 524         if (!err)
 525                 req->num_retrans++;
 526         return err;
 527 }
 528 EXPORT_SYMBOL(inet_rtx_syn_ack);
 529
 530 /* return true if req was found in the ehash table */
 531 static bool reqsk_queue_unlink(struct request_sock_queue *queue,
 532                                struct request_sock *req)
 533 {
 534         struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
 535         bool found = false;
 536
 537         if (sk_hashed(req_to_sk(req))) {
 538                 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
 539
 540                 spin_lock(lock);
 541                 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
 542                 spin_unlock(lock);
 543         }
 544         if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
 545                 reqsk_put(req);
 546         return found;
 547 }
 548
 549 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
 550 {
 551         if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
 552                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 553                 reqsk_put(req);
 554         }
 555 }
 556 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
 557
 558 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
 559 {
 560         inet_csk_reqsk_queue_drop(sk, req);
 561         reqsk_put(req);
 562 }
 563 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
 564
 565 static void reqsk_timer_handler(unsigned long data)
 566 {
 567         struct request_sock *req = (struct request_sock *)data;
 568         struct sock *sk_listener = req->rsk_listener;
 569         struct inet_connection_sock *icsk = inet_csk(sk_listener);
 570         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 571         int qlen, expire = 0, resend = 0;
 572         int max_retries, thresh;
 573         u8 defer_accept;
 574
 575         if (sk_state_load(sk_listener) != TCP_LISTEN)
 576                 goto drop;
 577
 578         max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 579         thresh = max_retries;
 580         /* Normally all the openreqs are young and become mature
 581          * (i.e. converted to established socket) for first timeout.
 582          * If synack was not acknowledged for 1 second, it means
 583          * one of the following things: synack was lost, ack was lost,
 584          * rtt is high or nobody planned to ack (i.e. synflood).
 585          * When server is a bit loaded, queue is populated with old
 586          * open requests, reducing effective size of queue.
 587          * When server is well loaded, queue size reduces to zero
 588          * after several minutes of work. It is not synflood,
 589          * it is normal operation. The solution is pruning
 590          * too old entries overriding normal timeout, when
 591          * situation becomes dangerous.
 592          *
 593          * Essentially, we reserve half of room for young
 594          * embrions; and abort old ones without pity, if old
 595          * ones are about to clog our table.
 596          */
 597         qlen = reqsk_queue_len(queue);
 598         if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
 599                 int young = reqsk_queue_len_young(queue) << 1;
 600
 601                 while (thresh > 2) {
 602                         if (qlen < young)
 603                                 break;
 604                         thresh--;
 605                         young <<= 1;
 606                 }
 607         }
 608         defer_accept = READ_ONCE(queue->rskq_defer_accept);
 609         if (defer_accept)
 610                 max_retries = defer_accept;
 611         syn_ack_recalc(req, thresh, max_retries, defer_accept,
 612                        &expire, &resend);
 613         req->rsk_ops->syn_ack_timeout(req);
 614         if (!expire &&
 615             (!resend ||
 616              !inet_rtx_syn_ack(sk_listener, req) ||
 617              inet_rsk(req)->acked)) {
 618                 unsigned long timeo;
 619
 620                 if (req->num_timeout++ == 0)
 621                         atomic_dec(&queue->young);
 622                 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 623                 mod_timer_pinned(&req->rsk_timer, jiffies + timeo);
 624                 return;
 625         }
 626 drop:
 627         inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
 628 }
 629
 630 static void reqsk_queue_hash_req(struct request_sock *req,
 631                                  unsigned long timeout)
 632 {
 633         req->num_retrans = 0;
 634         req->num_timeout = 0;
 635         req->sk = NULL;
 636
 637         setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req);
 638         mod_timer_pinned(&req->rsk_timer, jiffies + timeout);
 639
 640         inet_ehash_insert(req_to_sk(req), NULL);
 641         /* before letting lookups find us, make sure all req fields
 642          * are committed to memory and refcnt initialized.
 643          */
 644         smp_wmb();
 645         atomic_set(&req->rsk_refcnt, 2 + 1);
 646 }
 647
 648 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 649                                    unsigned long timeout)
 650 {
 651         reqsk_queue_hash_req(req, timeout);
 652         inet_csk_reqsk_queue_added(sk);
 653 }
 654 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 655
 656 /**
 657  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 658  *      @sk: the socket to clone
 659  *      @req: request_sock
 660  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 661  *
 662  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 663  */
 664 struct sock *inet_csk_clone_lock(const struct sock *sk,
 665                                  const struct request_sock *req,
 666                                  const gfp_t priority)
 667 {
 668         struct sock *newsk = sk_clone_lock(sk, priority);
 669
 670         if (newsk) {
 671                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 672
 673                 newsk->sk_state = TCP_SYN_RECV;
 674                 newicsk->icsk_bind_hash = NULL;
 675
 676                 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
 677                 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
 678                 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
 679                 newsk->sk_write_space = sk_stream_write_space;
 680
 681                 inet_sk(newsk)->mc_list = NULL;
 682
 683                 newsk->sk_mark = inet_rsk(req)->ir_mark;
 684                 atomic64_set(&newsk->sk_cookie,
 685                              atomic64_read(&inet_rsk(req)->ir_cookie));
 686
 687                 newicsk->icsk_retransmits = 0;
 688                 newicsk->icsk_backoff     = 0;
 689                 newicsk->icsk_probes_out  = 0;
 690
 691                 /* Deinitialize accept_queue to trap illegal accesses. */
 692                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 693
 694                 security_inet_csk_clone(newsk, req);
 695         }
 696         return newsk;
 697 }
 698 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 699
 700 /*
 701  * At this point, there should be no process reference to this
 702  * socket, and thus no user references at all.  Therefore we
 703  * can assume the socket waitqueue is inactive and nobody will
 704  * try to jump onto it.
 705  */
 706 void inet_csk_destroy_sock(struct sock *sk)
 707 {
 708         WARN_ON(sk->sk_state != TCP_CLOSE);
 709         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 710
 711         /* It cannot be in hash table! */
 712         WARN_ON(!sk_unhashed(sk));
 713
 714         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 715         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 716
 717         sk->sk_prot->destroy(sk);
 718
 719         sk_stream_kill_queues(sk);
 720
 721         xfrm_sk_free_policy(sk);
 722
 723         sk_refcnt_debug_release(sk);
 724
 725         percpu_counter_dec(sk->sk_prot->orphan_count);
 726         sock_put(sk);
 727 }
 728 EXPORT_SYMBOL(inet_csk_destroy_sock);
 729
 730 /* This function allows to force a closure of a socket after the call to
 731  * tcp/dccp_create_openreq_child().
 732  */
 733 void inet_csk_prepare_forced_close(struct sock *sk)
 734         __releases(&sk->sk_lock.slock)
 735 {
 736         /* sk_clone_lock locked the socket and set refcnt to 2 */
 737         bh_unlock_sock(sk);
 738         sock_put(sk);
 739
 740         /* The below has to be done to allow calling inet_csk_destroy_sock */
 741         sock_set_flag(sk, SOCK_DEAD);
 742         percpu_counter_inc(sk->sk_prot->orphan_count);
 743         inet_sk(sk)->inet_num = 0;
 744 }
 745 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
 746
 747 int inet_csk_listen_start(struct sock *sk, int backlog)
 748 {
 749         struct inet_connection_sock *icsk = inet_csk(sk);
 750         struct inet_sock *inet = inet_sk(sk);
 751
 752         reqsk_queue_alloc(&icsk->icsk_accept_queue);
 753
 754         sk->sk_max_ack_backlog = backlog;
 755         sk->sk_ack_backlog = 0;
 756         inet_csk_delack_init(sk);
 757
 758         /* There is race window here: we announce ourselves listening,
 759          * but this transition is still not validated by get_port().
 760          * It is OK, because this socket enters to hash table only
 761          * after validation is complete.
 762          */
 763         sk_state_store(sk, TCP_LISTEN);
 764         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 765                 inet->inet_sport = htons(inet->inet_num);
 766
 767                 sk_dst_reset(sk);
 768                 sk->sk_prot->hash(sk);
 769
 770                 return 0;
 771         }
 772
 773         sk->sk_state = TCP_CLOSE;
 774         return -EADDRINUSE;
 775 }
 776 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 777
 778 static void inet_child_forget(struct sock *sk, struct request_sock *req,
 779                               struct sock *child)
 780 {
 781         sk->sk_prot->disconnect(child, O_NONBLOCK);
 782
 783         sock_orphan(child);
 784
 785         percpu_counter_inc(sk->sk_prot->orphan_count);
 786
 787         if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
 788                 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
 789                 BUG_ON(sk != req->rsk_listener);
 790
 791                 /* Paranoid, to prevent race condition if
 792                  * an inbound pkt destined for child is
 793                  * blocked by sock lock in tcp_v4_rcv().
 794                  * Also to satisfy an assertion in
 795                  * tcp_v4_destroy_sock().
 796                  */
 797                 tcp_sk(child)->fastopen_rsk = NULL;
 798         }
 799         inet_csk_destroy_sock(child);
 800 }
 801
 802 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
 803                                       struct request_sock *req,
 804                                       struct sock *child)
 805 {
 806         struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
 807
 808         spin_lock(&queue->rskq_lock);
 809         if (unlikely(sk->sk_state != TCP_LISTEN)) {
 810                 inet_child_forget(sk, req, child);
 811                 child = NULL;
 812         } else {
 813                 req->sk = child;
 814                 req->dl_next = NULL;
 815                 if (queue->rskq_accept_head == NULL)
 816                         queue->rskq_accept_head = req;
 817                 else
 818                         queue->rskq_accept_tail->dl_next = req;
 819                 queue->rskq_accept_tail = req;
 820                 sk_acceptq_added(sk);
 821         }
 822         spin_unlock(&queue->rskq_lock);
 823         return child;
 824 }
 825 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
 826
 827 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
 828                                          struct request_sock *req, bool own_req)
 829 {
 830         if (own_req) {
 831                 inet_csk_reqsk_queue_drop(sk, req);
 832                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 833                 if (inet_csk_reqsk_queue_add(sk, req, child))
 834                         return child;
 835         }
 836         /* Too bad, another child took ownership of the request, undo. */
 837         bh_unlock_sock(child);
 838         sock_put(child);
 839         return NULL;
 840 }
 841 EXPORT_SYMBOL(inet_csk_complete_hashdance);
 842
 843 /*
 844  *      This routine closes sockets which have been at least partially
 845  *      opened, but not yet accepted.
 846  */
 847 void inet_csk_listen_stop(struct sock *sk)
 848 {
 849         struct inet_connection_sock *icsk = inet_csk(sk);
 850         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 851         struct request_sock *next, *req;
 852
 853         /* Following specs, it would be better either to send FIN
 854          * (and enter FIN-WAIT-1, it is normal close)
 855          * or to send active reset (abort).
 856          * Certainly, it is pretty dangerous while synflood, but it is
 857          * bad justification for our negligence 8)
 858          * To be honest, we are not able to make either
 859          * of the variants now.                 --ANK
 860          */
 861         while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
 862                 struct sock *child = req->sk;
 863
 864                 local_bh_disable();
 865                 bh_lock_sock(child);
 866                 WARN_ON(sock_owned_by_user(child));
 867                 sock_hold(child);
 868
 869                 inet_child_forget(sk, req, child);
 870                 reqsk_put(req);
 871                 bh_unlock_sock(child);
 872                 local_bh_enable();
 873                 sock_put(child);
 874
 875                 cond_resched();
 876         }
 877         if (queue->fastopenq.rskq_rst_head) {
 878                 /* Free all the reqs queued in rskq_rst_head. */
 879                 spin_lock_bh(&queue->fastopenq.lock);
 880                 req = queue->fastopenq.rskq_rst_head;
 881                 queue->fastopenq.rskq_rst_head = NULL;
 882                 spin_unlock_bh(&queue->fastopenq.lock);
 883                 while (req != NULL) {
 884                         next = req->dl_next;
 885                         reqsk_put(req);
 886                         req = next;
 887                 }
 888         }
 889         WARN_ON_ONCE(sk->sk_ack_backlog);
 890 }
 891 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 892
 893 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 894 {
 895         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 896         const struct inet_sock *inet = inet_sk(sk);
 897
 898         sin->sin_family         = AF_INET;
 899         sin->sin_addr.s_addr    = inet->inet_daddr;
 900         sin->sin_port           = inet->inet_dport;
 901 }
 902 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 903
 904 #ifdef CONFIG_COMPAT
 905 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 906                                char __user *optval, int __user *optlen)
 907 {
 908         const struct inet_connection_sock *icsk = inet_csk(sk);
 909
 910         if (icsk->icsk_af_ops->compat_getsockopt)
 911                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 912                                                             optval, optlen);
 913         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 914                                              optval, optlen);
 915 }
 916 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 917
 918 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 919                                char __user *optval, unsigned int optlen)
 920 {
 921         const struct inet_connection_sock *icsk = inet_csk(sk);
 922
 923         if (icsk->icsk_af_ops->compat_setsockopt)
 924                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 925                                                             optval, optlen);
 926         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 927                                              optval, optlen);
 928 }
 929 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 930 #endif
 931
 932 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
 933 {
 934         const struct inet_sock *inet = inet_sk(sk);
 935         const struct ip_options_rcu *inet_opt;
 936         __be32 daddr = inet->inet_daddr;
 937         struct flowi4 *fl4;
 938         struct rtable *rt;
 939
 940         rcu_read_lock();
 941         inet_opt = rcu_dereference(inet->inet_opt);
 942         if (inet_opt && inet_opt->opt.srr)
 943                 daddr = inet_opt->opt.faddr;
 944         fl4 = &fl->u.ip4;
 945         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
 946                                    inet->inet_saddr, inet->inet_dport,
 947                                    inet->inet_sport, sk->sk_protocol,
 948                                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
 949         if (IS_ERR(rt))
 950                 rt = NULL;
 951         if (rt)
 952                 sk_setup_caps(sk, &rt->dst);
 953         rcu_read_unlock();
 954
 955         return &rt->dst;
 956 }
 957
 958 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
 959 {
 960         struct dst_entry *dst = __sk_dst_check(sk, 0);
 961         struct inet_sock *inet = inet_sk(sk);
 962
 963         if (!dst) {
 964                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 965                 if (!dst)
 966                         goto out;
 967         }
 968         dst->ops->update_pmtu(dst, sk, NULL, mtu);
 969
 970         dst = __sk_dst_check(sk, 0);
 971         if (!dst)
 972                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 973 out:
 974         return dst;
 975 }
 976 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);