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 #include <net/sock_reuseport.h>
  28
  29 #ifdef INET_CSK_DEBUG
  30 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
  31 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
  32 #endif
  33
  34 void inet_get_local_port_range(struct net *net, int *low, int *high)
  35 {
  36         unsigned int seq;
  37
  38         do {
  39                 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
  40
  41                 *low = net->ipv4.ip_local_ports.range[0];
  42                 *high = net->ipv4.ip_local_ports.range[1];
  43         } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
  44 }
  45 EXPORT_SYMBOL(inet_get_local_port_range);
  46
  47 int inet_csk_bind_conflict(const struct sock *sk,
  48                            const struct inet_bind_bucket *tb, bool relax)
  49 {
  50         struct sock *sk2;
  51         int reuse = sk->sk_reuse;
  52         int reuseport = sk->sk_reuseport;
  53         kuid_t uid = sock_i_uid((struct sock *)sk);
  54
  55         /*
  56          * Unlike other sk lookup places we do not check
  57          * for sk_net here, since _all_ the socks listed
  58          * in tb->owners list belong to the same net - the
  59          * one this bucket belongs to.
  60          */
  61
  62         sk_for_each_bound(sk2, &tb->owners) {
  63                 if (sk != sk2 &&
  64                     !inet_v6_ipv6only(sk2) &&
  65                     (!sk->sk_bound_dev_if ||
  66                      !sk2->sk_bound_dev_if ||
  67                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
  68                         if ((!reuse || !sk2->sk_reuse ||
  69                             sk2->sk_state == TCP_LISTEN) &&
  70                             (!reuseport || !sk2->sk_reuseport ||
  71                              rcu_access_pointer(sk->sk_reuseport_cb) ||
  72                              (sk2->sk_state != TCP_TIME_WAIT &&
  73                              !uid_eq(uid, sock_i_uid(sk2))))) {
  74
  75                                 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
  76                                     sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
  77                                         break;
  78                         }
  79                         if (!relax && reuse && sk2->sk_reuse &&
  80                             sk2->sk_state != TCP_LISTEN) {
  81
  82                                 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
  83                                     sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
  84                                         break;
  85                         }
  86                 }
  87         }
  88         return sk2 != NULL;
  89 }
  90 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
  91
  92 /* Obtain a reference to a local port for the given sock,
  93  * if snum is zero it means select any available local port.
  94  */
  95 int inet_csk_get_port(struct sock *sk, unsigned short snum)
  96 {
  97         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  98         struct inet_bind_hashbucket *head;
  99         struct inet_bind_bucket *tb;
 100         int ret, attempts = 5;
 101         struct net *net = sock_net(sk);
 102         int smallest_size = -1, smallest_rover;
 103         kuid_t uid = sock_i_uid(sk);
 104         int attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
 105
 106         local_bh_disable();
 107         if (!snum) {
 108                 int remaining, rover, low, high;
 109
 110 again:
 111                 inet_get_local_port_range(net, &low, &high);
 112                 if (attempt_half) {
 113                         int half = low + ((high - low) >> 1);
 114
 115                         if (attempt_half == 1)
 116                                 high = half;
 117                         else
 118                                 low = half;
 119                 }
 120                 remaining = (high - low) + 1;
 121                 smallest_rover = rover = prandom_u32() % remaining + low;
 122
 123                 smallest_size = -1;
 124                 do {
 125                         if (inet_is_local_reserved_port(net, rover))
 126                                 goto next_nolock;
 127                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 128                                         hashinfo->bhash_size)];
 129                         spin_lock(&head->lock);
 130                         inet_bind_bucket_for_each(tb, &head->chain)
 131                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 132                                         if (((tb->fastreuse > 0 &&
 133                                               sk->sk_reuse &&
 134                                               sk->sk_state != TCP_LISTEN) ||
 135                                              (tb->fastreuseport > 0 &&
 136                                               sk->sk_reuseport &&
 137                                               !rcu_access_pointer(sk->sk_reuseport_cb) &&
 138                                               uid_eq(tb->fastuid, uid))) &&
 139                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 140                                                 smallest_size = tb->num_owners;
 141                                                 smallest_rover = rover;
 142                                         }
 143                                         if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
 144                                                 snum = rover;
 145                                                 goto tb_found;
 146                                         }
 147                                         goto next;
 148                                 }
 149                         break;
 150                 next:
 151                         spin_unlock(&head->lock);
 152                 next_nolock:
 153                         if (++rover > high)
 154                                 rover = low;
 155                 } while (--remaining > 0);
 156
 157                 /* Exhausted local port range during search?  It is not
 158                  * possible for us to be holding one of the bind hash
 159                  * locks if this test triggers, because if 'remaining'
 160                  * drops to zero, we broke out of the do/while loop at
 161                  * the top level, not from the 'break;' statement.
 162                  */
 163                 ret = 1;
 164                 if (remaining <= 0) {
 165                         if (smallest_size != -1) {
 166                                 snum = smallest_rover;
 167                                 goto have_snum;
 168                         }
 169                         if (attempt_half == 1) {
 170                                 /* OK we now try the upper half of the range */
 171                                 attempt_half = 2;
 172                                 goto again;
 173                         }
 174                         goto fail;
 175                 }
 176                 /* OK, here is the one we will use.  HEAD is
 177                  * non-NULL and we hold it's mutex.
 178                  */
 179                 snum = rover;
 180         } else {
 181 have_snum:
 182                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 183                                 hashinfo->bhash_size)];
 184                 spin_lock(&head->lock);
 185                 inet_bind_bucket_for_each(tb, &head->chain)
 186                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 187                                 goto tb_found;
 188         }
 189         tb = NULL;
 190         goto tb_not_found;
 191 tb_found:
 192         if (!hlist_empty(&tb->owners)) {
 193                 if (sk->sk_reuse == SK_FORCE_REUSE)
 194                         goto success;
 195
 196                 if (((tb->fastreuse > 0 &&
 197                       sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 198                      (tb->fastreuseport > 0 &&
 199                       sk->sk_reuseport &&
 200                       !rcu_access_pointer(sk->sk_reuseport_cb) &&
 201                       uid_eq(tb->fastuid, uid))) && smallest_size == -1) {
 202                         goto success;
 203                 } else {
 204                         ret = 1;
 205                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
 206                                 if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 207                                      (tb->fastreuseport > 0 &&
 208                                       sk->sk_reuseport &&
 209                                       !rcu_access_pointer(sk->sk_reuseport_cb) &&
 210                                       uid_eq(tb->fastuid, uid))) &&
 211                                     smallest_size != -1 && --attempts >= 0) {
 212                                         spin_unlock(&head->lock);
 213                                         goto again;
 214                                 }
 215
 216                                 goto fail_unlock;
 217                         }
 218                 }
 219         }
 220 tb_not_found:
 221         ret = 1;
 222         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 223                                         net, head, snum)) == NULL)
 224                 goto fail_unlock;
 225         if (hlist_empty(&tb->owners)) {
 226                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 227                         tb->fastreuse = 1;
 228                 else
 229                         tb->fastreuse = 0;
 230                 if (sk->sk_reuseport) {
 231                         tb->fastreuseport = 1;
 232                         tb->fastuid = uid;
 233                 } else
 234                         tb->fastreuseport = 0;
 235         } else {
 236                 if (tb->fastreuse &&
 237                     (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 238                         tb->fastreuse = 0;
 239                 if (tb->fastreuseport &&
 240                     (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)))
 241                         tb->fastreuseport = 0;
 242         }
 243 success:
 244         if (!inet_csk(sk)->icsk_bind_hash)
 245                 inet_bind_hash(sk, tb, snum);
 246         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 247         ret = 0;
 248
 249 fail_unlock:
 250         spin_unlock(&head->lock);
 251 fail:
 252         local_bh_enable();
 253         return ret;
 254 }
 255 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 256
 257 /*
 258  * Wait for an incoming connection, avoid race conditions. This must be called
 259  * with the socket locked.
 260  */
 261 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 262 {
 263         struct inet_connection_sock *icsk = inet_csk(sk);
 264         DEFINE_WAIT(wait);
 265         int err;
 266
 267         /*
 268          * True wake-one mechanism for incoming connections: only
 269          * one process gets woken up, not the 'whole herd'.
 270          * Since we do not 'race & poll' for established sockets
 271          * anymore, the common case will execute the loop only once.
 272          *
 273          * Subtle issue: "add_wait_queue_exclusive()" will be added
 274          * after any current non-exclusive waiters, and we know that
 275          * it will always _stay_ after any new non-exclusive waiters
 276          * because all non-exclusive waiters are added at the
 277          * beginning of the wait-queue. As such, it's ok to "drop"
 278          * our exclusiveness temporarily when we get woken up without
 279          * having to remove and re-insert us on the wait queue.
 280          */
 281         for (;;) {
 282                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 283                                           TASK_INTERRUPTIBLE);
 284                 release_sock(sk);
 285                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 286                         timeo = schedule_timeout(timeo);
 287                 sched_annotate_sleep();
 288                 lock_sock(sk);
 289                 err = 0;
 290                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 291                         break;
 292                 err = -EINVAL;
 293                 if (sk->sk_state != TCP_LISTEN)
 294                         break;
 295                 err = sock_intr_errno(timeo);
 296                 if (signal_pending(current))
 297                         break;
 298                 err = -EAGAIN;
 299                 if (!timeo)
 300                         break;
 301         }
 302         finish_wait(sk_sleep(sk), &wait);
 303         return err;
 304 }
 305
 306 /*
 307  * This will accept the next outstanding connection.
 308  */
 309 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 310 {
 311         struct inet_connection_sock *icsk = inet_csk(sk);
 312         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 313         struct request_sock *req;
 314         struct sock *newsk;
 315         int error;
 316
 317         lock_sock(sk);
 318
 319         /* We need to make sure that this socket is listening,
 320          * and that it has something pending.
 321          */
 322         error = -EINVAL;
 323         if (sk->sk_state != TCP_LISTEN)
 324                 goto out_err;
 325
 326         /* Find already established connection */
 327         if (reqsk_queue_empty(queue)) {
 328                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 329
 330                 /* If this is a non blocking socket don't sleep */
 331                 error = -EAGAIN;
 332                 if (!timeo)
 333                         goto out_err;
 334
 335                 error = inet_csk_wait_for_connect(sk, timeo);
 336                 if (error)
 337                         goto out_err;
 338         }
 339         req = reqsk_queue_remove(queue, sk);
 340         newsk = req->sk;
 341
 342         if (sk->sk_protocol == IPPROTO_TCP &&
 343             tcp_rsk(req)->tfo_listener) {
 344                 spin_lock_bh(&queue->fastopenq.lock);
 345                 if (tcp_rsk(req)->tfo_listener) {
 346                         /* We are still waiting for the final ACK from 3WHS
 347                          * so can't free req now. Instead, we set req->sk to
 348                          * NULL to signify that the child socket is taken
 349                          * so reqsk_fastopen_remove() will free the req
 350                          * when 3WHS finishes (or is aborted).
 351                          */
 352                         req->sk = NULL;
 353                         req = NULL;
 354                 }
 355                 spin_unlock_bh(&queue->fastopenq.lock);
 356         }
 357 out:
 358         release_sock(sk);
 359         if (req)
 360                 reqsk_put(req);
 361         return newsk;
 362 out_err:
 363         newsk = NULL;
 364         req = NULL;
 365         *err = error;
 366         goto out;
 367 }
 368 EXPORT_SYMBOL(inet_csk_accept);
 369
 370 /*
 371  * Using different timers for retransmit, delayed acks and probes
 372  * We may wish use just one timer maintaining a list of expire jiffies
 373  * to optimize.
 374  */
 375 void inet_csk_init_xmit_timers(struct sock *sk,
 376                                void (*retransmit_handler)(unsigned long),
 377                                void (*delack_handler)(unsigned long),
 378                                void (*keepalive_handler)(unsigned long))
 379 {
 380         struct inet_connection_sock *icsk = inet_csk(sk);
 381
 382         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 383                         (unsigned long)sk);
 384         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 385                         (unsigned long)sk);
 386         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 387         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 388 }
 389 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 390
 391 void inet_csk_clear_xmit_timers(struct sock *sk)
 392 {
 393         struct inet_connection_sock *icsk = inet_csk(sk);
 394
 395         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 396
 397         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 398         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 399         sk_stop_timer(sk, &sk->sk_timer);
 400 }
 401 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 402
 403 void inet_csk_delete_keepalive_timer(struct sock *sk)
 404 {
 405         sk_stop_timer(sk, &sk->sk_timer);
 406 }
 407 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 408
 409 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 410 {
 411         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 412 }
 413 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 414
 415 struct dst_entry *inet_csk_route_req(const struct sock *sk,
 416                                      struct flowi4 *fl4,
 417                                      const struct request_sock *req)
 418 {
 419         const struct inet_request_sock *ireq = inet_rsk(req);
 420         struct net *net = read_pnet(&ireq->ireq_net);
 421         struct ip_options_rcu *opt = ireq->opt;
 422         struct rtable *rt;
 423
 424         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 425                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 426                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 427                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 428                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 429                            htons(ireq->ir_num));
 430         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 431         rt = ip_route_output_flow(net, fl4, sk);
 432         if (IS_ERR(rt))
 433                 goto no_route;
 434         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 435                 goto route_err;
 436         return &rt->dst;
 437
 438 route_err:
 439         ip_rt_put(rt);
 440 no_route:
 441         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 442         return NULL;
 443 }
 444 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 445
 446 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
 447                                             struct sock *newsk,
 448                                             const struct request_sock *req)
 449 {
 450         const struct inet_request_sock *ireq = inet_rsk(req);
 451         struct net *net = read_pnet(&ireq->ireq_net);
 452         struct inet_sock *newinet = inet_sk(newsk);
 453         struct ip_options_rcu *opt;
 454         struct flowi4 *fl4;
 455         struct rtable *rt;
 456
 457         fl4 = &newinet->cork.fl.u.ip4;
 458
 459         rcu_read_lock();
 460         opt = rcu_dereference(newinet->inet_opt);
 461         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 462                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 463                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 464                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 465                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 466                            htons(ireq->ir_num));
 467         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 468         rt = ip_route_output_flow(net, fl4, sk);
 469         if (IS_ERR(rt))
 470                 goto no_route;
 471         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 472                 goto route_err;
 473         rcu_read_unlock();
 474         return &rt->dst;
 475
 476 route_err:
 477         ip_rt_put(rt);
 478 no_route:
 479         rcu_read_unlock();
 480         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 481         return NULL;
 482 }
 483 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 484
 485 #if IS_ENABLED(CONFIG_IPV6)
 486 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 487 #else
 488 #define AF_INET_FAMILY(fam) true
 489 #endif
 490
 491 /* Decide when to expire the request and when to resend SYN-ACK */
 492 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 493                                   const int max_retries,
 494                                   const u8 rskq_defer_accept,
 495                                   int *expire, int *resend)
 496 {
 497         if (!rskq_defer_accept) {
 498                 *expire = req->num_timeout >= thresh;
 499                 *resend = 1;
 500                 return;
 501         }
 502         *expire = req->num_timeout >= thresh &&
 503                   (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 504         /*
 505          * Do not resend while waiting for data after ACK,
 506          * start to resend on end of deferring period to give
 507          * last chance for data or ACK to create established socket.
 508          */
 509         *resend = !inet_rsk(req)->acked ||
 510                   req->num_timeout >= rskq_defer_accept - 1;
 511 }
 512
 513 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
 514 {
 515         int err = req->rsk_ops->rtx_syn_ack(parent, req);
 516
 517         if (!err)
 518                 req->num_retrans++;
 519         return err;
 520 }
 521 EXPORT_SYMBOL(inet_rtx_syn_ack);
 522
 523 /* return true if req was found in the ehash table */
 524 static bool reqsk_queue_unlink(struct request_sock_queue *queue,
 525                                struct request_sock *req)
 526 {
 527         struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
 528         bool found = false;
 529
 530         if (sk_hashed(req_to_sk(req))) {
 531                 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
 532
 533                 spin_lock(lock);
 534                 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
 535                 spin_unlock(lock);
 536         }
 537         if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
 538                 reqsk_put(req);
 539         return found;
 540 }
 541
 542 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
 543 {
 544         if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
 545                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 546                 reqsk_put(req);
 547         }
 548 }
 549 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
 550
 551 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
 552 {
 553         inet_csk_reqsk_queue_drop(sk, req);
 554         reqsk_put(req);
 555 }
 556 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
 557
 558 static void reqsk_timer_handler(unsigned long data)
 559 {
 560         struct request_sock *req = (struct request_sock *)data;
 561         struct sock *sk_listener = req->rsk_listener;
 562         struct net *net = sock_net(sk_listener);
 563         struct inet_connection_sock *icsk = inet_csk(sk_listener);
 564         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 565         int qlen, expire = 0, resend = 0;
 566         int max_retries, thresh;
 567         u8 defer_accept;
 568
 569         if (sk_state_load(sk_listener) != TCP_LISTEN)
 570                 goto drop;
 571
 572         max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
 573         thresh = max_retries;
 574         /* Normally all the openreqs are young and become mature
 575          * (i.e. converted to established socket) for first timeout.
 576          * If synack was not acknowledged for 1 second, it means
 577          * one of the following things: synack was lost, ack was lost,
 578          * rtt is high or nobody planned to ack (i.e. synflood).
 579          * When server is a bit loaded, queue is populated with old
 580          * open requests, reducing effective size of queue.
 581          * When server is well loaded, queue size reduces to zero
 582          * after several minutes of work. It is not synflood,
 583          * it is normal operation. The solution is pruning
 584          * too old entries overriding normal timeout, when
 585          * situation becomes dangerous.
 586          *
 587          * Essentially, we reserve half of room for young
 588          * embrions; and abort old ones without pity, if old
 589          * ones are about to clog our table.
 590          */
 591         qlen = reqsk_queue_len(queue);
 592         if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
 593                 int young = reqsk_queue_len_young(queue) << 1;
 594
 595                 while (thresh > 2) {
 596                         if (qlen < young)
 597                                 break;
 598                         thresh--;
 599                         young <<= 1;
 600                 }
 601         }
 602         defer_accept = READ_ONCE(queue->rskq_defer_accept);
 603         if (defer_accept)
 604                 max_retries = defer_accept;
 605         syn_ack_recalc(req, thresh, max_retries, defer_accept,
 606                        &expire, &resend);
 607         req->rsk_ops->syn_ack_timeout(req);
 608         if (!expire &&
 609             (!resend ||
 610              !inet_rtx_syn_ack(sk_listener, req) ||
 611              inet_rsk(req)->acked)) {
 612                 unsigned long timeo;
 613
 614                 if (req->num_timeout++ == 0)
 615                         atomic_dec(&queue->young);
 616                 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 617                 mod_timer_pinned(&req->rsk_timer, jiffies + timeo);
 618                 return;
 619         }
 620 drop:
 621         inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
 622 }
 623
 624 static void reqsk_queue_hash_req(struct request_sock *req,
 625                                  unsigned long timeout)
 626 {
 627         req->num_retrans = 0;
 628         req->num_timeout = 0;
 629         req->sk = NULL;
 630
 631         setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req);
 632         mod_timer_pinned(&req->rsk_timer, jiffies + timeout);
 633
 634         inet_ehash_insert(req_to_sk(req), NULL);
 635         /* before letting lookups find us, make sure all req fields
 636          * are committed to memory and refcnt initialized.
 637          */
 638         smp_wmb();
 639         atomic_set(&req->rsk_refcnt, 2 + 1);
 640 }
 641
 642 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 643                                    unsigned long timeout)
 644 {
 645         reqsk_queue_hash_req(req, timeout);
 646         inet_csk_reqsk_queue_added(sk);
 647 }
 648 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 649
 650 /**
 651  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 652  *      @sk: the socket to clone
 653  *      @req: request_sock
 654  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 655  *
 656  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 657  */
 658 struct sock *inet_csk_clone_lock(const struct sock *sk,
 659                                  const struct request_sock *req,
 660                                  const gfp_t priority)
 661 {
 662         struct sock *newsk = sk_clone_lock(sk, priority);
 663
 664         if (newsk) {
 665                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 666
 667                 newsk->sk_state = TCP_SYN_RECV;
 668                 newicsk->icsk_bind_hash = NULL;
 669
 670                 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
 671                 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
 672                 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
 673                 newsk->sk_write_space = sk_stream_write_space;
 674
 675                 newsk->sk_mark = inet_rsk(req)->ir_mark;
 676                 atomic64_set(&newsk->sk_cookie,
 677                              atomic64_read(&inet_rsk(req)->ir_cookie));
 678
 679                 newicsk->icsk_retransmits = 0;
 680                 newicsk->icsk_backoff     = 0;
 681                 newicsk->icsk_probes_out  = 0;
 682
 683                 /* Deinitialize accept_queue to trap illegal accesses. */
 684                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 685
 686                 security_inet_csk_clone(newsk, req);
 687         }
 688         return newsk;
 689 }
 690 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 691
 692 /*
 693  * At this point, there should be no process reference to this
 694  * socket, and thus no user references at all.  Therefore we
 695  * can assume the socket waitqueue is inactive and nobody will
 696  * try to jump onto it.
 697  */
 698 void inet_csk_destroy_sock(struct sock *sk)
 699 {
 700         WARN_ON(sk->sk_state != TCP_CLOSE);
 701         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 702
 703         /* It cannot be in hash table! */
 704         WARN_ON(!sk_unhashed(sk));
 705
 706         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 707         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 708
 709         sk->sk_prot->destroy(sk);
 710
 711         sk_stream_kill_queues(sk);
 712
 713         xfrm_sk_free_policy(sk);
 714
 715         sk_refcnt_debug_release(sk);
 716
 717         percpu_counter_dec(sk->sk_prot->orphan_count);
 718         sock_put(sk);
 719 }
 720 EXPORT_SYMBOL(inet_csk_destroy_sock);
 721
 722 /* This function allows to force a closure of a socket after the call to
 723  * tcp/dccp_create_openreq_child().
 724  */
 725 void inet_csk_prepare_forced_close(struct sock *sk)
 726         __releases(&sk->sk_lock.slock)
 727 {
 728         /* sk_clone_lock locked the socket and set refcnt to 2 */
 729         bh_unlock_sock(sk);
 730         sock_put(sk);
 731
 732         /* The below has to be done to allow calling inet_csk_destroy_sock */
 733         sock_set_flag(sk, SOCK_DEAD);
 734         percpu_counter_inc(sk->sk_prot->orphan_count);
 735         inet_sk(sk)->inet_num = 0;
 736 }
 737 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
 738
 739 int inet_csk_listen_start(struct sock *sk, int backlog)
 740 {
 741         struct inet_connection_sock *icsk = inet_csk(sk);
 742         struct inet_sock *inet = inet_sk(sk);
 743         int err = -EADDRINUSE;
 744
 745         reqsk_queue_alloc(&icsk->icsk_accept_queue);
 746
 747         sk->sk_max_ack_backlog = backlog;
 748         sk->sk_ack_backlog = 0;
 749         inet_csk_delack_init(sk);
 750
 751         /* There is race window here: we announce ourselves listening,
 752          * but this transition is still not validated by get_port().
 753          * It is OK, because this socket enters to hash table only
 754          * after validation is complete.
 755          */
 756         sk_state_store(sk, TCP_LISTEN);
 757         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 758                 inet->inet_sport = htons(inet->inet_num);
 759
 760                 sk_dst_reset(sk);
 761                 err = sk->sk_prot->hash(sk);
 762
 763                 if (likely(!err))
 764                         return 0;
 765         }
 766
 767         sk->sk_state = TCP_CLOSE;
 768         return err;
 769 }
 770 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 771
 772 static void inet_child_forget(struct sock *sk, struct request_sock *req,
 773                               struct sock *child)
 774 {
 775         sk->sk_prot->disconnect(child, O_NONBLOCK);
 776
 777         sock_orphan(child);
 778
 779         percpu_counter_inc(sk->sk_prot->orphan_count);
 780
 781         if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
 782                 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
 783                 BUG_ON(sk != req->rsk_listener);
 784
 785                 /* Paranoid, to prevent race condition if
 786                  * an inbound pkt destined for child is
 787                  * blocked by sock lock in tcp_v4_rcv().
 788                  * Also to satisfy an assertion in
 789                  * tcp_v4_destroy_sock().
 790                  */
 791                 tcp_sk(child)->fastopen_rsk = NULL;
 792         }
 793         inet_csk_destroy_sock(child);
 794         reqsk_put(req);
 795 }
 796
 797 void inet_csk_reqsk_queue_add(struct sock *sk, struct request_sock *req,
 798                               struct sock *child)
 799 {
 800         struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
 801
 802         spin_lock(&queue->rskq_lock);
 803         if (unlikely(sk->sk_state != TCP_LISTEN)) {
 804                 inet_child_forget(sk, req, child);
 805         } else {
 806                 req->sk = child;
 807                 req->dl_next = NULL;
 808                 if (queue->rskq_accept_head == NULL)
 809                         queue->rskq_accept_head = req;
 810                 else
 811                         queue->rskq_accept_tail->dl_next = req;
 812                 queue->rskq_accept_tail = req;
 813                 sk_acceptq_added(sk);
 814         }
 815         spin_unlock(&queue->rskq_lock);
 816 }
 817 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
 818
 819 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
 820                                          struct request_sock *req, bool own_req)
 821 {
 822         if (own_req) {
 823                 inet_csk_reqsk_queue_drop(sk, req);
 824                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 825                 inet_csk_reqsk_queue_add(sk, req, child);
 826                 /* Warning: caller must not call reqsk_put(req);
 827                  * child stole last reference on it.
 828                  */
 829                 return child;
 830         }
 831         /* Too bad, another child took ownership of the request, undo. */
 832         bh_unlock_sock(child);
 833         sock_put(child);
 834         return NULL;
 835 }
 836 EXPORT_SYMBOL(inet_csk_complete_hashdance);
 837
 838 /*
 839  *      This routine closes sockets which have been at least partially
 840  *      opened, but not yet accepted.
 841  */
 842 void inet_csk_listen_stop(struct sock *sk)
 843 {
 844         struct inet_connection_sock *icsk = inet_csk(sk);
 845         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 846         struct request_sock *next, *req;
 847
 848         /* Following specs, it would be better either to send FIN
 849          * (and enter FIN-WAIT-1, it is normal close)
 850          * or to send active reset (abort).
 851          * Certainly, it is pretty dangerous while synflood, but it is
 852          * bad justification for our negligence 8)
 853          * To be honest, we are not able to make either
 854          * of the variants now.                 --ANK
 855          */
 856         while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
 857                 struct sock *child = req->sk;
 858
 859                 local_bh_disable();
 860                 bh_lock_sock(child);
 861                 WARN_ON(sock_owned_by_user(child));
 862                 sock_hold(child);
 863
 864                 inet_child_forget(sk, req, child);
 865                 bh_unlock_sock(child);
 866                 local_bh_enable();
 867                 sock_put(child);
 868
 869                 cond_resched();
 870         }
 871         if (queue->fastopenq.rskq_rst_head) {
 872                 /* Free all the reqs queued in rskq_rst_head. */
 873                 spin_lock_bh(&queue->fastopenq.lock);
 874                 req = queue->fastopenq.rskq_rst_head;
 875                 queue->fastopenq.rskq_rst_head = NULL;
 876                 spin_unlock_bh(&queue->fastopenq.lock);
 877                 while (req != NULL) {
 878                         next = req->dl_next;
 879                         reqsk_put(req);
 880                         req = next;
 881                 }
 882         }
 883         WARN_ON_ONCE(sk->sk_ack_backlog);
 884 }
 885 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 886
 887 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 888 {
 889         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 890         const struct inet_sock *inet = inet_sk(sk);
 891
 892         sin->sin_family         = AF_INET;
 893         sin->sin_addr.s_addr    = inet->inet_daddr;
 894         sin->sin_port           = inet->inet_dport;
 895 }
 896 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 897
 898 #ifdef CONFIG_COMPAT
 899 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 900                                char __user *optval, int __user *optlen)
 901 {
 902         const struct inet_connection_sock *icsk = inet_csk(sk);
 903
 904         if (icsk->icsk_af_ops->compat_getsockopt)
 905                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 906                                                             optval, optlen);
 907         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 908                                              optval, optlen);
 909 }
 910 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 911
 912 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 913                                char __user *optval, unsigned int optlen)
 914 {
 915         const struct inet_connection_sock *icsk = inet_csk(sk);
 916
 917         if (icsk->icsk_af_ops->compat_setsockopt)
 918                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 919                                                             optval, optlen);
 920         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 921                                              optval, optlen);
 922 }
 923 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 924 #endif
 925
 926 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
 927 {
 928         const struct inet_sock *inet = inet_sk(sk);
 929         const struct ip_options_rcu *inet_opt;
 930         __be32 daddr = inet->inet_daddr;
 931         struct flowi4 *fl4;
 932         struct rtable *rt;
 933
 934         rcu_read_lock();
 935         inet_opt = rcu_dereference(inet->inet_opt);
 936         if (inet_opt && inet_opt->opt.srr)
 937                 daddr = inet_opt->opt.faddr;
 938         fl4 = &fl->u.ip4;
 939         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
 940                                    inet->inet_saddr, inet->inet_dport,
 941                                    inet->inet_sport, sk->sk_protocol,
 942                                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
 943         if (IS_ERR(rt))
 944                 rt = NULL;
 945         if (rt)
 946                 sk_setup_caps(sk, &rt->dst);
 947         rcu_read_unlock();
 948
 949         return &rt->dst;
 950 }
 951
 952 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
 953 {
 954         struct dst_entry *dst = __sk_dst_check(sk, 0);
 955         struct inet_sock *inet = inet_sk(sk);
 956
 957         if (!dst) {
 958                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 959                 if (!dst)
 960                         goto out;
 961         }
 962         dst->ops->update_pmtu(dst, sk, NULL, mtu);
 963
 964         dst = __sk_dst_check(sk, 0);
 965         if (!dst)
 966                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 967 out:
 968         return dst;
 969 }
 970 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);