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
 103         local_bh_disable();
 104         if (!snum) {
 105                 int remaining, rover, low, high;
 106
 107 again:
 108                 inet_get_local_port_range(net, &low, &high);
 109                 remaining = (high - low) + 1;
 110                 smallest_rover = rover = prandom_u32() % remaining + low;
 111
 112                 smallest_size = -1;
 113                 do {
 114                         if (inet_is_local_reserved_port(net, rover))
 115                                 goto next_nolock;
 116                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 117                                         hashinfo->bhash_size)];
 118                         spin_lock(&head->lock);
 119                         inet_bind_bucket_for_each(tb, &head->chain)
 120                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 121                                         if (((tb->fastreuse > 0 &&
 122                                               sk->sk_reuse &&
 123                                               sk->sk_state != TCP_LISTEN) ||
 124                                              (tb->fastreuseport > 0 &&
 125                                               sk->sk_reuseport &&
 126                                               uid_eq(tb->fastuid, uid))) &&
 127                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 128                                                 smallest_size = tb->num_owners;
 129                                                 smallest_rover = rover;
 130                                         }
 131                                         if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
 132                                                 snum = rover;
 133                                                 goto tb_found;
 134                                         }
 135                                         goto next;
 136                                 }
 137                         break;
 138                 next:
 139                         spin_unlock(&head->lock);
 140                 next_nolock:
 141                         if (++rover > high)
 142                                 rover = low;
 143                 } while (--remaining > 0);
 144
 145                 /* Exhausted local port range during search?  It is not
 146                  * possible for us to be holding one of the bind hash
 147                  * locks if this test triggers, because if 'remaining'
 148                  * drops to zero, we broke out of the do/while loop at
 149                  * the top level, not from the 'break;' statement.
 150                  */
 151                 ret = 1;
 152                 if (remaining <= 0) {
 153                         if (smallest_size != -1) {
 154                                 snum = smallest_rover;
 155                                 goto have_snum;
 156                         }
 157                         goto fail;
 158                 }
 159                 /* OK, here is the one we will use.  HEAD is
 160                  * non-NULL and we hold it's mutex.
 161                  */
 162                 snum = rover;
 163         } else {
 164 have_snum:
 165                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 166                                 hashinfo->bhash_size)];
 167                 spin_lock(&head->lock);
 168                 inet_bind_bucket_for_each(tb, &head->chain)
 169                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 170                                 goto tb_found;
 171         }
 172         tb = NULL;
 173         goto tb_not_found;
 174 tb_found:
 175         if (!hlist_empty(&tb->owners)) {
 176                 if (sk->sk_reuse == SK_FORCE_REUSE)
 177                         goto success;
 178
 179                 if (((tb->fastreuse > 0 &&
 180                       sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 181                      (tb->fastreuseport > 0 &&
 182                       sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
 183                     smallest_size == -1) {
 184                         goto success;
 185                 } else {
 186                         ret = 1;
 187                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
 188                                 if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 189                                      (tb->fastreuseport > 0 &&
 190                                       sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
 191                                     smallest_size != -1 && --attempts >= 0) {
 192                                         spin_unlock(&head->lock);
 193                                         goto again;
 194                                 }
 195
 196                                 goto fail_unlock;
 197                         }
 198                 }
 199         }
 200 tb_not_found:
 201         ret = 1;
 202         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 203                                         net, head, snum)) == NULL)
 204                 goto fail_unlock;
 205         if (hlist_empty(&tb->owners)) {
 206                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 207                         tb->fastreuse = 1;
 208                 else
 209                         tb->fastreuse = 0;
 210                 if (sk->sk_reuseport) {
 211                         tb->fastreuseport = 1;
 212                         tb->fastuid = uid;
 213                 } else
 214                         tb->fastreuseport = 0;
 215         } else {
 216                 if (tb->fastreuse &&
 217                     (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 218                         tb->fastreuse = 0;
 219                 if (tb->fastreuseport &&
 220                     (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)))
 221                         tb->fastreuseport = 0;
 222         }
 223 success:
 224         if (!inet_csk(sk)->icsk_bind_hash)
 225                 inet_bind_hash(sk, tb, snum);
 226         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 227         ret = 0;
 228
 229 fail_unlock:
 230         spin_unlock(&head->lock);
 231 fail:
 232         local_bh_enable();
 233         return ret;
 234 }
 235 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 236
 237 /*
 238  * Wait for an incoming connection, avoid race conditions. This must be called
 239  * with the socket locked.
 240  */
 241 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 242 {
 243         struct inet_connection_sock *icsk = inet_csk(sk);
 244         DEFINE_WAIT(wait);
 245         int err;
 246
 247         /*
 248          * True wake-one mechanism for incoming connections: only
 249          * one process gets woken up, not the 'whole herd'.
 250          * Since we do not 'race & poll' for established sockets
 251          * anymore, the common case will execute the loop only once.
 252          *
 253          * Subtle issue: "add_wait_queue_exclusive()" will be added
 254          * after any current non-exclusive waiters, and we know that
 255          * it will always _stay_ after any new non-exclusive waiters
 256          * because all non-exclusive waiters are added at the
 257          * beginning of the wait-queue. As such, it's ok to "drop"
 258          * our exclusiveness temporarily when we get woken up without
 259          * having to remove and re-insert us on the wait queue.
 260          */
 261         for (;;) {
 262                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 263                                           TASK_INTERRUPTIBLE);
 264                 release_sock(sk);
 265                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 266                         timeo = schedule_timeout(timeo);
 267                 sched_annotate_sleep();
 268                 lock_sock(sk);
 269                 err = 0;
 270                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 271                         break;
 272                 err = -EINVAL;
 273                 if (sk->sk_state != TCP_LISTEN)
 274                         break;
 275                 err = sock_intr_errno(timeo);
 276                 if (signal_pending(current))
 277                         break;
 278                 err = -EAGAIN;
 279                 if (!timeo)
 280                         break;
 281         }
 282         finish_wait(sk_sleep(sk), &wait);
 283         return err;
 284 }
 285
 286 /*
 287  * This will accept the next outstanding connection.
 288  */
 289 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 290 {
 291         struct inet_connection_sock *icsk = inet_csk(sk);
 292         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 293         struct request_sock *req;
 294         struct sock *newsk;
 295         int error;
 296
 297         lock_sock(sk);
 298
 299         /* We need to make sure that this socket is listening,
 300          * and that it has something pending.
 301          */
 302         error = -EINVAL;
 303         if (sk->sk_state != TCP_LISTEN)
 304                 goto out_err;
 305
 306         /* Find already established connection */
 307         if (reqsk_queue_empty(queue)) {
 308                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 309
 310                 /* If this is a non blocking socket don't sleep */
 311                 error = -EAGAIN;
 312                 if (!timeo)
 313                         goto out_err;
 314
 315                 error = inet_csk_wait_for_connect(sk, timeo);
 316                 if (error)
 317                         goto out_err;
 318         }
 319         req = reqsk_queue_remove(queue);
 320         newsk = req->sk;
 321
 322         sk_acceptq_removed(sk);
 323         if (sk->sk_protocol == IPPROTO_TCP &&
 324             tcp_rsk(req)->tfo_listener &&
 325             queue->fastopenq) {
 326                 spin_lock_bh(&queue->fastopenq->lock);
 327                 if (tcp_rsk(req)->tfo_listener) {
 328                         /* We are still waiting for the final ACK from 3WHS
 329                          * so can't free req now. Instead, we set req->sk to
 330                          * NULL to signify that the child socket is taken
 331                          * so reqsk_fastopen_remove() will free the req
 332                          * when 3WHS finishes (or is aborted).
 333                          */
 334                         req->sk = NULL;
 335                         req = NULL;
 336                 }
 337                 spin_unlock_bh(&queue->fastopenq->lock);
 338         }
 339 out:
 340         release_sock(sk);
 341         if (req)
 342                 reqsk_put(req);
 343         return newsk;
 344 out_err:
 345         newsk = NULL;
 346         req = NULL;
 347         *err = error;
 348         goto out;
 349 }
 350 EXPORT_SYMBOL(inet_csk_accept);
 351
 352 /*
 353  * Using different timers for retransmit, delayed acks and probes
 354  * We may wish use just one timer maintaining a list of expire jiffies
 355  * to optimize.
 356  */
 357 void inet_csk_init_xmit_timers(struct sock *sk,
 358                                void (*retransmit_handler)(unsigned long),
 359                                void (*delack_handler)(unsigned long),
 360                                void (*keepalive_handler)(unsigned long))
 361 {
 362         struct inet_connection_sock *icsk = inet_csk(sk);
 363
 364         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 365                         (unsigned long)sk);
 366         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 367                         (unsigned long)sk);
 368         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 369         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 370 }
 371 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 372
 373 void inet_csk_clear_xmit_timers(struct sock *sk)
 374 {
 375         struct inet_connection_sock *icsk = inet_csk(sk);
 376
 377         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 378
 379         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 380         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 381         sk_stop_timer(sk, &sk->sk_timer);
 382 }
 383 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 384
 385 void inet_csk_delete_keepalive_timer(struct sock *sk)
 386 {
 387         sk_stop_timer(sk, &sk->sk_timer);
 388 }
 389 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 390
 391 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 392 {
 393         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 394 }
 395 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 396
 397 struct dst_entry *inet_csk_route_req(struct sock *sk,
 398                                      struct flowi4 *fl4,
 399                                      const struct request_sock *req)
 400 {
 401         const struct inet_request_sock *ireq = inet_rsk(req);
 402         struct net *net = read_pnet(&ireq->ireq_net);
 403         struct ip_options_rcu *opt = ireq->opt;
 404         struct rtable *rt;
 405
 406         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 407                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 408                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 409                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 410                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 411                            htons(ireq->ir_num));
 412         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 413         rt = ip_route_output_flow(net, fl4, sk);
 414         if (IS_ERR(rt))
 415                 goto no_route;
 416         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 417                 goto route_err;
 418         return &rt->dst;
 419
 420 route_err:
 421         ip_rt_put(rt);
 422 no_route:
 423         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 424         return NULL;
 425 }
 426 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 427
 428 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
 429                                             struct sock *newsk,
 430                                             const struct request_sock *req)
 431 {
 432         const struct inet_request_sock *ireq = inet_rsk(req);
 433         struct net *net = read_pnet(&ireq->ireq_net);
 434         struct inet_sock *newinet = inet_sk(newsk);
 435         struct ip_options_rcu *opt;
 436         struct flowi4 *fl4;
 437         struct rtable *rt;
 438
 439         fl4 = &newinet->cork.fl.u.ip4;
 440
 441         rcu_read_lock();
 442         opt = rcu_dereference(newinet->inet_opt);
 443         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 444                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 445                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 446                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 447                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 448                            htons(ireq->ir_num));
 449         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 450         rt = ip_route_output_flow(net, fl4, sk);
 451         if (IS_ERR(rt))
 452                 goto no_route;
 453         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 454                 goto route_err;
 455         rcu_read_unlock();
 456         return &rt->dst;
 457
 458 route_err:
 459         ip_rt_put(rt);
 460 no_route:
 461         rcu_read_unlock();
 462         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 463         return NULL;
 464 }
 465 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 466
 467 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 468                                  const u32 rnd, const u32 synq_hsize)
 469 {
 470         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 471 }
 472
 473 #if IS_ENABLED(CONFIG_IPV6)
 474 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 475 #else
 476 #define AF_INET_FAMILY(fam) true
 477 #endif
 478
 479 /* Note: this is temporary :
 480  * req sock will no longer be in listener hash table
 481 */
 482 struct request_sock *inet_csk_search_req(struct sock *sk,
 483                                          const __be16 rport,
 484                                          const __be32 raddr,
 485                                          const __be32 laddr)
 486 {
 487         struct inet_connection_sock *icsk = inet_csk(sk);
 488         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 489         struct request_sock *req;
 490         u32 hash = inet_synq_hash(raddr, rport, lopt->hash_rnd,
 491                                   lopt->nr_table_entries);
 492
 493         spin_lock(&icsk->icsk_accept_queue.syn_wait_lock);
 494         for (req = lopt->syn_table[hash]; req != NULL; req = req->dl_next) {
 495                 const struct inet_request_sock *ireq = inet_rsk(req);
 496
 497                 if (ireq->ir_rmt_port == rport &&
 498                     ireq->ir_rmt_addr == raddr &&
 499                     ireq->ir_loc_addr == laddr &&
 500                     AF_INET_FAMILY(req->rsk_ops->family)) {
 501                         atomic_inc(&req->rsk_refcnt);
 502                         WARN_ON(req->sk);
 503                         break;
 504                 }
 505         }
 506         spin_unlock(&icsk->icsk_accept_queue.syn_wait_lock);
 507
 508         return req;
 509 }
 510 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 511
 512 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 513                                    unsigned long timeout)
 514 {
 515         struct inet_connection_sock *icsk = inet_csk(sk);
 516         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 517         const u32 h = inet_synq_hash(inet_rsk(req)->ir_rmt_addr,
 518                                      inet_rsk(req)->ir_rmt_port,
 519                                      lopt->hash_rnd, lopt->nr_table_entries);
 520
 521         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 522         inet_csk_reqsk_queue_added(sk, timeout);
 523 }
 524 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 525
 526 /* Only thing we need from tcp.h */
 527 extern int sysctl_tcp_synack_retries;
 528
 529
 530 /* Decide when to expire the request and when to resend SYN-ACK */
 531 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 532                                   const int max_retries,
 533                                   const u8 rskq_defer_accept,
 534                                   int *expire, int *resend)
 535 {
 536         if (!rskq_defer_accept) {
 537                 *expire = req->num_timeout >= thresh;
 538                 *resend = 1;
 539                 return;
 540         }
 541         *expire = req->num_timeout >= thresh &&
 542                   (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 543         /*
 544          * Do not resend while waiting for data after ACK,
 545          * start to resend on end of deferring period to give
 546          * last chance for data or ACK to create established socket.
 547          */
 548         *resend = !inet_rsk(req)->acked ||
 549                   req->num_timeout >= rskq_defer_accept - 1;
 550 }
 551
 552 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req)
 553 {
 554         int err = req->rsk_ops->rtx_syn_ack(parent, req);
 555
 556         if (!err)
 557                 req->num_retrans++;
 558         return err;
 559 }
 560 EXPORT_SYMBOL(inet_rtx_syn_ack);
 561
 562 /* return true if req was found in the syn_table[] */
 563 static bool reqsk_queue_unlink(struct request_sock_queue *queue,
 564                                struct request_sock *req)
 565 {
 566         struct listen_sock *lopt = queue->listen_opt;
 567         struct request_sock **prev;
 568         bool found = false;
 569
 570         spin_lock(&queue->syn_wait_lock);
 571
 572         for (prev = &lopt->syn_table[req->rsk_hash]; *prev != NULL;
 573              prev = &(*prev)->dl_next) {
 574                 if (*prev == req) {
 575                         *prev = req->dl_next;
 576                         found = true;
 577                         break;
 578                 }
 579         }
 580
 581         spin_unlock(&queue->syn_wait_lock);
 582         if (del_timer(&req->rsk_timer))
 583                 reqsk_put(req);
 584         return found;
 585 }
 586
 587 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
 588 {
 589         if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
 590                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 591                 reqsk_put(req);
 592         }
 593 }
 594 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
 595
 596 static void reqsk_timer_handler(unsigned long data)
 597 {
 598         struct request_sock *req = (struct request_sock *)data;
 599         struct sock *sk_listener = req->rsk_listener;
 600         struct inet_connection_sock *icsk = inet_csk(sk_listener);
 601         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 602         struct listen_sock *lopt = queue->listen_opt;
 603         int qlen, expire = 0, resend = 0;
 604         int max_retries, thresh;
 605         u8 defer_accept;
 606
 607         if (sk_listener->sk_state != TCP_LISTEN || !lopt) {
 608                 reqsk_put(req);
 609                 return;
 610         }
 611
 612         max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 613         thresh = max_retries;
 614         /* Normally all the openreqs are young and become mature
 615          * (i.e. converted to established socket) for first timeout.
 616          * If synack was not acknowledged for 1 second, it means
 617          * one of the following things: synack was lost, ack was lost,
 618          * rtt is high or nobody planned to ack (i.e. synflood).
 619          * When server is a bit loaded, queue is populated with old
 620          * open requests, reducing effective size of queue.
 621          * When server is well loaded, queue size reduces to zero
 622          * after several minutes of work. It is not synflood,
 623          * it is normal operation. The solution is pruning
 624          * too old entries overriding normal timeout, when
 625          * situation becomes dangerous.
 626          *
 627          * Essentially, we reserve half of room for young
 628          * embrions; and abort old ones without pity, if old
 629          * ones are about to clog our table.
 630          */
 631         qlen = listen_sock_qlen(lopt);
 632         if (qlen >> (lopt->max_qlen_log - 1)) {
 633                 int young = listen_sock_young(lopt) << 1;
 634
 635                 while (thresh > 2) {
 636                         if (qlen < young)
 637                                 break;
 638                         thresh--;
 639                         young <<= 1;
 640                 }
 641         }
 642         defer_accept = READ_ONCE(queue->rskq_defer_accept);
 643         if (defer_accept)
 644                 max_retries = defer_accept;
 645         syn_ack_recalc(req, thresh, max_retries, defer_accept,
 646                        &expire, &resend);
 647         req->rsk_ops->syn_ack_timeout(req);
 648         if (!expire &&
 649             (!resend ||
 650              !inet_rtx_syn_ack(sk_listener, req) ||
 651              inet_rsk(req)->acked)) {
 652                 unsigned long timeo;
 653
 654                 if (req->num_timeout++ == 0)
 655                         atomic_inc(&lopt->young_dec);
 656                 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 657                 mod_timer_pinned(&req->rsk_timer, jiffies + timeo);
 658                 return;
 659         }
 660         inet_csk_reqsk_queue_drop(sk_listener, req);
 661         reqsk_put(req);
 662 }
 663
 664 void reqsk_queue_hash_req(struct request_sock_queue *queue,
 665                           u32 hash, struct request_sock *req,
 666                           unsigned long timeout)
 667 {
 668         struct listen_sock *lopt = queue->listen_opt;
 669
 670         req->num_retrans = 0;
 671         req->num_timeout = 0;
 672         req->sk = NULL;
 673
 674         /* before letting lookups find us, make sure all req fields
 675          * are committed to memory and refcnt initialized.
 676          */
 677         smp_wmb();
 678         atomic_set(&req->rsk_refcnt, 2);
 679         setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req);
 680         req->rsk_hash = hash;
 681
 682         spin_lock(&queue->syn_wait_lock);
 683         req->dl_next = lopt->syn_table[hash];
 684         lopt->syn_table[hash] = req;
 685         spin_unlock(&queue->syn_wait_lock);
 686
 687         mod_timer_pinned(&req->rsk_timer, jiffies + timeout);
 688 }
 689 EXPORT_SYMBOL(reqsk_queue_hash_req);
 690
 691 /**
 692  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 693  *      @sk: the socket to clone
 694  *      @req: request_sock
 695  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 696  *
 697  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 698  */
 699 struct sock *inet_csk_clone_lock(const struct sock *sk,
 700                                  const struct request_sock *req,
 701                                  const gfp_t priority)
 702 {
 703         struct sock *newsk = sk_clone_lock(sk, priority);
 704
 705         if (newsk) {
 706                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 707
 708                 newsk->sk_state = TCP_SYN_RECV;
 709                 newicsk->icsk_bind_hash = NULL;
 710
 711                 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
 712                 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
 713                 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
 714                 newsk->sk_write_space = sk_stream_write_space;
 715
 716                 newsk->sk_mark = inet_rsk(req)->ir_mark;
 717                 atomic64_set(&newsk->sk_cookie,
 718                              atomic64_read(&inet_rsk(req)->ir_cookie));
 719
 720                 newicsk->icsk_retransmits = 0;
 721                 newicsk->icsk_backoff     = 0;
 722                 newicsk->icsk_probes_out  = 0;
 723
 724                 /* Deinitialize accept_queue to trap illegal accesses. */
 725                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 726
 727                 security_inet_csk_clone(newsk, req);
 728         }
 729         return newsk;
 730 }
 731 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 732
 733 /*
 734  * At this point, there should be no process reference to this
 735  * socket, and thus no user references at all.  Therefore we
 736  * can assume the socket waitqueue is inactive and nobody will
 737  * try to jump onto it.
 738  */
 739 void inet_csk_destroy_sock(struct sock *sk)
 740 {
 741         WARN_ON(sk->sk_state != TCP_CLOSE);
 742         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 743
 744         /* It cannot be in hash table! */
 745         WARN_ON(!sk_unhashed(sk));
 746
 747         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 748         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 749
 750         sk->sk_prot->destroy(sk);
 751
 752         sk_stream_kill_queues(sk);
 753
 754         xfrm_sk_free_policy(sk);
 755
 756         sk_refcnt_debug_release(sk);
 757
 758         percpu_counter_dec(sk->sk_prot->orphan_count);
 759         sock_put(sk);
 760 }
 761 EXPORT_SYMBOL(inet_csk_destroy_sock);
 762
 763 /* This function allows to force a closure of a socket after the call to
 764  * tcp/dccp_create_openreq_child().
 765  */
 766 void inet_csk_prepare_forced_close(struct sock *sk)
 767         __releases(&sk->sk_lock.slock)
 768 {
 769         /* sk_clone_lock locked the socket and set refcnt to 2 */
 770         bh_unlock_sock(sk);
 771         sock_put(sk);
 772
 773         /* The below has to be done to allow calling inet_csk_destroy_sock */
 774         sock_set_flag(sk, SOCK_DEAD);
 775         percpu_counter_inc(sk->sk_prot->orphan_count);
 776         inet_sk(sk)->inet_num = 0;
 777 }
 778 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
 779
 780 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 781 {
 782         struct inet_sock *inet = inet_sk(sk);
 783         struct inet_connection_sock *icsk = inet_csk(sk);
 784         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 785
 786         if (rc != 0)
 787                 return rc;
 788
 789         sk->sk_max_ack_backlog = 0;
 790         sk->sk_ack_backlog = 0;
 791         inet_csk_delack_init(sk);
 792
 793         /* There is race window here: we announce ourselves listening,
 794          * but this transition is still not validated by get_port().
 795          * It is OK, because this socket enters to hash table only
 796          * after validation is complete.
 797          */
 798         sk->sk_state = TCP_LISTEN;
 799         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 800                 inet->inet_sport = htons(inet->inet_num);
 801
 802                 sk_dst_reset(sk);
 803                 sk->sk_prot->hash(sk);
 804
 805                 return 0;
 806         }
 807
 808         sk->sk_state = TCP_CLOSE;
 809         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 810         return -EADDRINUSE;
 811 }
 812 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 813
 814 /*
 815  *      This routine closes sockets which have been at least partially
 816  *      opened, but not yet accepted.
 817  */
 818 void inet_csk_listen_stop(struct sock *sk)
 819 {
 820         struct inet_connection_sock *icsk = inet_csk(sk);
 821         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 822         struct request_sock *acc_req;
 823         struct request_sock *req;
 824
 825         /* make all the listen_opt local to us */
 826         acc_req = reqsk_queue_yank_acceptq(queue);
 827
 828         /* Following specs, it would be better either to send FIN
 829          * (and enter FIN-WAIT-1, it is normal close)
 830          * or to send active reset (abort).
 831          * Certainly, it is pretty dangerous while synflood, but it is
 832          * bad justification for our negligence 8)
 833          * To be honest, we are not able to make either
 834          * of the variants now.                 --ANK
 835          */
 836         reqsk_queue_destroy(queue);
 837
 838         while ((req = acc_req) != NULL) {
 839                 struct sock *child = req->sk;
 840
 841                 acc_req = req->dl_next;
 842
 843                 local_bh_disable();
 844                 bh_lock_sock(child);
 845                 WARN_ON(sock_owned_by_user(child));
 846                 sock_hold(child);
 847
 848                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 849
 850                 sock_orphan(child);
 851
 852                 percpu_counter_inc(sk->sk_prot->orphan_count);
 853
 854                 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
 855                         BUG_ON(tcp_sk(child)->fastopen_rsk != req);
 856                         BUG_ON(sk != req->rsk_listener);
 857
 858                         /* Paranoid, to prevent race condition if
 859                          * an inbound pkt destined for child is
 860                          * blocked by sock lock in tcp_v4_rcv().
 861                          * Also to satisfy an assertion in
 862                          * tcp_v4_destroy_sock().
 863                          */
 864                         tcp_sk(child)->fastopen_rsk = NULL;
 865                 }
 866                 inet_csk_destroy_sock(child);
 867
 868                 bh_unlock_sock(child);
 869                 local_bh_enable();
 870                 sock_put(child);
 871
 872                 sk_acceptq_removed(sk);
 873                 reqsk_put(req);
 874         }
 875         if (queue->fastopenq) {
 876                 /* Free all the reqs queued in rskq_rst_head. */
 877                 spin_lock_bh(&queue->fastopenq->lock);
 878                 acc_req = queue->fastopenq->rskq_rst_head;
 879                 queue->fastopenq->rskq_rst_head = NULL;
 880                 spin_unlock_bh(&queue->fastopenq->lock);
 881                 while ((req = acc_req) != NULL) {
 882                         acc_req = req->dl_next;
 883                         reqsk_put(req);
 884                 }
 885         }
 886         WARN_ON(sk->sk_ack_backlog);
 887 }
 888 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 889
 890 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 891 {
 892         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 893         const struct inet_sock *inet = inet_sk(sk);
 894
 895         sin->sin_family         = AF_INET;
 896         sin->sin_addr.s_addr    = inet->inet_daddr;
 897         sin->sin_port           = inet->inet_dport;
 898 }
 899 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 900
 901 #ifdef CONFIG_COMPAT
 902 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 903                                char __user *optval, int __user *optlen)
 904 {
 905         const struct inet_connection_sock *icsk = inet_csk(sk);
 906
 907         if (icsk->icsk_af_ops->compat_getsockopt)
 908                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 909                                                             optval, optlen);
 910         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 911                                              optval, optlen);
 912 }
 913 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 914
 915 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 916                                char __user *optval, unsigned int optlen)
 917 {
 918         const struct inet_connection_sock *icsk = inet_csk(sk);
 919
 920         if (icsk->icsk_af_ops->compat_setsockopt)
 921                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 922                                                             optval, optlen);
 923         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 924                                              optval, optlen);
 925 }
 926 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 927 #endif
 928
 929 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
 930 {
 931         const struct inet_sock *inet = inet_sk(sk);
 932         const struct ip_options_rcu *inet_opt;
 933         __be32 daddr = inet->inet_daddr;
 934         struct flowi4 *fl4;
 935         struct rtable *rt;
 936
 937         rcu_read_lock();
 938         inet_opt = rcu_dereference(inet->inet_opt);
 939         if (inet_opt && inet_opt->opt.srr)
 940                 daddr = inet_opt->opt.faddr;
 941         fl4 = &fl->u.ip4;
 942         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
 943                                    inet->inet_saddr, inet->inet_dport,
 944                                    inet->inet_sport, sk->sk_protocol,
 945                                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
 946         if (IS_ERR(rt))
 947                 rt = NULL;
 948         if (rt)
 949                 sk_setup_caps(sk, &rt->dst);
 950         rcu_read_unlock();
 951
 952         return &rt->dst;
 953 }
 954
 955 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
 956 {
 957         struct dst_entry *dst = __sk_dst_check(sk, 0);
 958         struct inet_sock *inet = inet_sk(sk);
 959
 960         if (!dst) {
 961                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 962                 if (!dst)
 963                         goto out;
 964         }
 965         dst->ops->update_pmtu(dst, sk, NULL, mtu);
 966
 967         dst = __sk_dst_check(sk, 0);
 968         if (!dst)
 969                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 970 out:
 971         return dst;
 972 }
 973 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);