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
  27 #ifdef INET_CSK_DEBUG
  28 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
  29 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
  30 #endif
  31
  32 /*
  33  * This struct holds the first and last local port number.
  34  */
  35 struct local_ports sysctl_local_ports __read_mostly = {
  36         .lock = __SEQLOCK_UNLOCKED(sysctl_local_ports.lock),
  37         .range = { 32768, 61000 },
  38 };
  39
  40 unsigned long *sysctl_local_reserved_ports;
  41 EXPORT_SYMBOL(sysctl_local_reserved_ports);
  42
  43 void inet_get_local_port_range(int *low, int *high)
  44 {
  45         unsigned seq;
  46         do {
  47                 seq = read_seqbegin(&sysctl_local_ports.lock);
  48
  49                 *low = sysctl_local_ports.range[0];
  50                 *high = sysctl_local_ports.range[1];
  51         } while (read_seqretry(&sysctl_local_ports.lock, seq));
  52 }
  53 EXPORT_SYMBOL(inet_get_local_port_range);
  54
  55 int inet_csk_bind_conflict(const struct sock *sk,
  56                            const struct inet_bind_bucket *tb)
  57 {
  58         struct sock *sk2;
  59         struct hlist_node *node;
  60         int reuse = sk->sk_reuse;
  61
  62         /*
  63          * Unlike other sk lookup places we do not check
  64          * for sk_net here, since _all_ the socks listed
  65          * in tb->owners list belong to the same net - the
  66          * one this bucket belongs to.
  67          */
  68
  69         sk_for_each_bound(sk2, node, &tb->owners) {
  70                 if (sk != sk2 &&
  71                     !inet_v6_ipv6only(sk2) &&
  72                     (!sk->sk_bound_dev_if ||
  73                      !sk2->sk_bound_dev_if ||
  74                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
  75                         if (!reuse || !sk2->sk_reuse ||
  76                             sk2->sk_state == TCP_LISTEN) {
  77                                 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
  78                                 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
  79                                     sk2_rcv_saddr == sk_rcv_saddr(sk))
  80                                         break;
  81                         }
  82                 }
  83         }
  84         return node != NULL;
  85 }
  86 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
  87
  88 /* Obtain a reference to a local port for the given sock,
  89  * if snum is zero it means select any available local port.
  90  */
  91 int inet_csk_get_port(struct sock *sk, unsigned short snum)
  92 {
  93         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  94         struct inet_bind_hashbucket *head;
  95         struct hlist_node *node;
  96         struct inet_bind_bucket *tb;
  97         int ret, attempts = 5;
  98         struct net *net = sock_net(sk);
  99         int smallest_size = -1, smallest_rover;
 100
 101         local_bh_disable();
 102         if (!snum) {
 103                 int remaining, rover, low, high;
 104
 105 again:
 106                 inet_get_local_port_range(&low, &high);
 107                 remaining = (high - low) + 1;
 108                 smallest_rover = rover = net_random() % remaining + low;
 109
 110                 smallest_size = -1;
 111                 do {
 112                         if (inet_is_reserved_local_port(rover))
 113                                 goto next_nolock;
 114                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 115                                         hashinfo->bhash_size)];
 116                         spin_lock(&head->lock);
 117                         inet_bind_bucket_for_each(tb, node, &head->chain)
 118                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 119                                         if (tb->fastreuse > 0 &&
 120                                             sk->sk_reuse &&
 121                                             sk->sk_state != TCP_LISTEN &&
 122                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 123                                                 smallest_size = tb->num_owners;
 124                                                 smallest_rover = rover;
 125                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
 126                                                         spin_unlock(&head->lock);
 127                                                         snum = smallest_rover;
 128                                                         goto have_snum;
 129                                                 }
 130                                         }
 131                                         if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
 132                                                 spin_unlock(&head->lock);
 133                                                 snum = rover;
 134                                                 goto have_snum;
 135                                         }
 136                                         goto next;
 137                                 }
 138                         break;
 139                 next:
 140                         spin_unlock(&head->lock);
 141                 next_nolock:
 142                         if (++rover > high)
 143                                 rover = low;
 144                 } while (--remaining > 0);
 145
 146                 /* Exhausted local port range during search?  It is not
 147                  * possible for us to be holding one of the bind hash
 148                  * locks if this test triggers, because if 'remaining'
 149                  * drops to zero, we broke out of the do/while loop at
 150                  * the top level, not from the 'break;' statement.
 151                  */
 152                 ret = 1;
 153                 if (remaining <= 0) {
 154                         if (smallest_size != -1) {
 155                                 snum = smallest_rover;
 156                                 goto have_snum;
 157                         }
 158                         goto fail;
 159                 }
 160                 /* OK, here is the one we will use.  HEAD is
 161                  * non-NULL and we hold it's mutex.
 162                  */
 163                 snum = rover;
 164         } else {
 165 have_snum:
 166                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 167                                 hashinfo->bhash_size)];
 168                 spin_lock(&head->lock);
 169                 inet_bind_bucket_for_each(tb, node, &head->chain)
 170                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 171                                 goto tb_found;
 172         }
 173         tb = NULL;
 174         goto tb_not_found;
 175 tb_found:
 176         if (!hlist_empty(&tb->owners)) {
 177                 if (tb->fastreuse > 0 &&
 178                     sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 179                     smallest_size == -1) {
 180                         goto success;
 181                 } else {
 182                         ret = 1;
 183                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
 184                                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 185                                     smallest_size != -1 && --attempts >= 0) {
 186                                         spin_unlock(&head->lock);
 187                                         goto again;
 188                                 }
 189                                 goto fail_unlock;
 190                         }
 191                 }
 192         }
 193 tb_not_found:
 194         ret = 1;
 195         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 196                                         net, head, snum)) == NULL)
 197                 goto fail_unlock;
 198         if (hlist_empty(&tb->owners)) {
 199                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 200                         tb->fastreuse = 1;
 201                 else
 202                         tb->fastreuse = 0;
 203         } else if (tb->fastreuse &&
 204                    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 205                 tb->fastreuse = 0;
 206 success:
 207         if (!inet_csk(sk)->icsk_bind_hash)
 208                 inet_bind_hash(sk, tb, snum);
 209         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 210         ret = 0;
 211
 212 fail_unlock:
 213         spin_unlock(&head->lock);
 214 fail:
 215         local_bh_enable();
 216         return ret;
 217 }
 218 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 219
 220 /*
 221  * Wait for an incoming connection, avoid race conditions. This must be called
 222  * with the socket locked.
 223  */
 224 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 225 {
 226         struct inet_connection_sock *icsk = inet_csk(sk);
 227         DEFINE_WAIT(wait);
 228         int err;
 229
 230         /*
 231          * True wake-one mechanism for incoming connections: only
 232          * one process gets woken up, not the 'whole herd'.
 233          * Since we do not 'race & poll' for established sockets
 234          * anymore, the common case will execute the loop only once.
 235          *
 236          * Subtle issue: "add_wait_queue_exclusive()" will be added
 237          * after any current non-exclusive waiters, and we know that
 238          * it will always _stay_ after any new non-exclusive waiters
 239          * because all non-exclusive waiters are added at the
 240          * beginning of the wait-queue. As such, it's ok to "drop"
 241          * our exclusiveness temporarily when we get woken up without
 242          * having to remove and re-insert us on the wait queue.
 243          */
 244         for (;;) {
 245                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 246                                           TASK_INTERRUPTIBLE);
 247                 release_sock(sk);
 248                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 249                         timeo = schedule_timeout(timeo);
 250                 lock_sock(sk);
 251                 err = 0;
 252                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 253                         break;
 254                 err = -EINVAL;
 255                 if (sk->sk_state != TCP_LISTEN)
 256                         break;
 257                 err = sock_intr_errno(timeo);
 258                 if (signal_pending(current))
 259                         break;
 260                 err = -EAGAIN;
 261                 if (!timeo)
 262                         break;
 263         }
 264         finish_wait(sk_sleep(sk), &wait);
 265         return err;
 266 }
 267
 268 /*
 269  * This will accept the next outstanding connection.
 270  */
 271 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 272 {
 273         struct inet_connection_sock *icsk = inet_csk(sk);
 274         struct sock *newsk;
 275         int error;
 276
 277         lock_sock(sk);
 278
 279         /* We need to make sure that this socket is listening,
 280          * and that it has something pending.
 281          */
 282         error = -EINVAL;
 283         if (sk->sk_state != TCP_LISTEN)
 284                 goto out_err;
 285
 286         /* Find already established connection */
 287         if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
 288                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 289
 290                 /* If this is a non blocking socket don't sleep */
 291                 error = -EAGAIN;
 292                 if (!timeo)
 293                         goto out_err;
 294
 295                 error = inet_csk_wait_for_connect(sk, timeo);
 296                 if (error)
 297                         goto out_err;
 298         }
 299
 300         newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
 301         WARN_ON(newsk->sk_state == TCP_SYN_RECV);
 302 out:
 303         release_sock(sk);
 304         return newsk;
 305 out_err:
 306         newsk = NULL;
 307         *err = error;
 308         goto out;
 309 }
 310 EXPORT_SYMBOL(inet_csk_accept);
 311
 312 /*
 313  * Using different timers for retransmit, delayed acks and probes
 314  * We may wish use just one timer maintaining a list of expire jiffies
 315  * to optimize.
 316  */
 317 void inet_csk_init_xmit_timers(struct sock *sk,
 318                                void (*retransmit_handler)(unsigned long),
 319                                void (*delack_handler)(unsigned long),
 320                                void (*keepalive_handler)(unsigned long))
 321 {
 322         struct inet_connection_sock *icsk = inet_csk(sk);
 323
 324         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 325                         (unsigned long)sk);
 326         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 327                         (unsigned long)sk);
 328         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 329         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 330 }
 331 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 332
 333 void inet_csk_clear_xmit_timers(struct sock *sk)
 334 {
 335         struct inet_connection_sock *icsk = inet_csk(sk);
 336
 337         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 338
 339         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 340         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 341         sk_stop_timer(sk, &sk->sk_timer);
 342 }
 343 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 344
 345 void inet_csk_delete_keepalive_timer(struct sock *sk)
 346 {
 347         sk_stop_timer(sk, &sk->sk_timer);
 348 }
 349 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 350
 351 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 352 {
 353         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 354 }
 355 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 356
 357 struct dst_entry *inet_csk_route_req(struct sock *sk,
 358                                      struct flowi4 *fl4,
 359                                      const struct request_sock *req)
 360 {
 361         struct rtable *rt;
 362         const struct inet_request_sock *ireq = inet_rsk(req);
 363         struct ip_options_rcu *opt = inet_rsk(req)->opt;
 364         struct net *net = sock_net(sk);
 365
 366         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 367                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 368                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 369                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
 370                            ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
 371         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 372         rt = ip_route_output_flow(net, fl4, sk);
 373         if (IS_ERR(rt))
 374                 goto no_route;
 375         if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
 376                 goto route_err;
 377         return &rt->dst;
 378
 379 route_err:
 380         ip_rt_put(rt);
 381 no_route:
 382         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 383         return NULL;
 384 }
 385 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 386
 387 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
 388                                             struct sock *newsk,
 389                                             const struct request_sock *req)
 390 {
 391         const struct inet_request_sock *ireq = inet_rsk(req);
 392         struct inet_sock *newinet = inet_sk(newsk);
 393         struct ip_options_rcu *opt = ireq->opt;
 394         struct net *net = sock_net(sk);
 395         struct flowi4 *fl4;
 396         struct rtable *rt;
 397
 398         fl4 = &newinet->cork.fl.u.ip4;
 399         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 400                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 401                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 402                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
 403                            ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
 404         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 405         rt = ip_route_output_flow(net, fl4, sk);
 406         if (IS_ERR(rt))
 407                 goto no_route;
 408         if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
 409                 goto route_err;
 410         return &rt->dst;
 411
 412 route_err:
 413         ip_rt_put(rt);
 414 no_route:
 415         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 416         return NULL;
 417 }
 418 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 419
 420 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 421                                  const u32 rnd, const u32 synq_hsize)
 422 {
 423         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 424 }
 425
 426 #if IS_ENABLED(CONFIG_IPV6)
 427 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 428 #else
 429 #define AF_INET_FAMILY(fam) 1
 430 #endif
 431
 432 struct request_sock *inet_csk_search_req(const struct sock *sk,
 433                                          struct request_sock ***prevp,
 434                                          const __be16 rport, const __be32 raddr,
 435                                          const __be32 laddr)
 436 {
 437         const struct inet_connection_sock *icsk = inet_csk(sk);
 438         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 439         struct request_sock *req, **prev;
 440
 441         for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
 442                                                     lopt->nr_table_entries)];
 443              (req = *prev) != NULL;
 444              prev = &req->dl_next) {
 445                 const struct inet_request_sock *ireq = inet_rsk(req);
 446
 447                 if (ireq->rmt_port == rport &&
 448                     ireq->rmt_addr == raddr &&
 449                     ireq->loc_addr == laddr &&
 450                     AF_INET_FAMILY(req->rsk_ops->family)) {
 451                         WARN_ON(req->sk);
 452                         *prevp = prev;
 453                         break;
 454                 }
 455         }
 456
 457         return req;
 458 }
 459 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 460
 461 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 462                                    unsigned long timeout)
 463 {
 464         struct inet_connection_sock *icsk = inet_csk(sk);
 465         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 466         const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
 467                                      lopt->hash_rnd, lopt->nr_table_entries);
 468
 469         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 470         inet_csk_reqsk_queue_added(sk, timeout);
 471 }
 472 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 473
 474 /* Only thing we need from tcp.h */
 475 extern int sysctl_tcp_synack_retries;
 476
 477
 478 /* Decide when to expire the request and when to resend SYN-ACK */
 479 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 480                                   const int max_retries,
 481                                   const u8 rskq_defer_accept,
 482                                   int *expire, int *resend)
 483 {
 484         if (!rskq_defer_accept) {
 485                 *expire = req->retrans >= thresh;
 486                 *resend = 1;
 487                 return;
 488         }
 489         *expire = req->retrans >= thresh &&
 490                   (!inet_rsk(req)->acked || req->retrans >= max_retries);
 491         /*
 492          * Do not resend while waiting for data after ACK,
 493          * start to resend on end of deferring period to give
 494          * last chance for data or ACK to create established socket.
 495          */
 496         *resend = !inet_rsk(req)->acked ||
 497                   req->retrans >= rskq_defer_accept - 1;
 498 }
 499
 500 void inet_csk_reqsk_queue_prune(struct sock *parent,
 501                                 const unsigned long interval,
 502                                 const unsigned long timeout,
 503                                 const unsigned long max_rto)
 504 {
 505         struct inet_connection_sock *icsk = inet_csk(parent);
 506         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 507         struct listen_sock *lopt = queue->listen_opt;
 508         int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 509         int thresh = max_retries;
 510         unsigned long now = jiffies;
 511         struct request_sock **reqp, *req;
 512         int i, budget;
 513
 514         if (lopt == NULL || lopt->qlen == 0)
 515                 return;
 516
 517         /* Normally all the openreqs are young and become mature
 518          * (i.e. converted to established socket) for first timeout.
 519          * If synack was not acknowledged for 3 seconds, it means
 520          * one of the following things: synack was lost, ack was lost,
 521          * rtt is high or nobody planned to ack (i.e. synflood).
 522          * When server is a bit loaded, queue is populated with old
 523          * open requests, reducing effective size of queue.
 524          * When server is well loaded, queue size reduces to zero
 525          * after several minutes of work. It is not synflood,
 526          * it is normal operation. The solution is pruning
 527          * too old entries overriding normal timeout, when
 528          * situation becomes dangerous.
 529          *
 530          * Essentially, we reserve half of room for young
 531          * embrions; and abort old ones without pity, if old
 532          * ones are about to clog our table.
 533          */
 534         if (lopt->qlen>>(lopt->max_qlen_log-1)) {
 535                 int young = (lopt->qlen_young<<1);
 536
 537                 while (thresh > 2) {
 538                         if (lopt->qlen < young)
 539                                 break;
 540                         thresh--;
 541                         young <<= 1;
 542                 }
 543         }
 544
 545         if (queue->rskq_defer_accept)
 546                 max_retries = queue->rskq_defer_accept;
 547
 548         budget = 2 * (lopt->nr_table_entries / (timeout / interval));
 549         i = lopt->clock_hand;
 550
 551         do {
 552                 reqp=&lopt->syn_table[i];
 553                 while ((req = *reqp) != NULL) {
 554                         if (time_after_eq(now, req->expires)) {
 555                                 int expire = 0, resend = 0;
 556
 557                                 syn_ack_recalc(req, thresh, max_retries,
 558                                                queue->rskq_defer_accept,
 559                                                &expire, &resend);
 560                                 if (req->rsk_ops->syn_ack_timeout)
 561                                         req->rsk_ops->syn_ack_timeout(parent, req);
 562                                 if (!expire &&
 563                                     (!resend ||
 564                                      !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
 565                                      inet_rsk(req)->acked)) {
 566                                         unsigned long timeo;
 567
 568                                         if (req->retrans++ == 0)
 569                                                 lopt->qlen_young--;
 570                                         timeo = min((timeout << req->retrans), max_rto);
 571                                         req->expires = now + timeo;
 572                                         reqp = &req->dl_next;
 573                                         continue;
 574                                 }
 575
 576                                 /* Drop this request */
 577                                 inet_csk_reqsk_queue_unlink(parent, req, reqp);
 578                                 reqsk_queue_removed(queue, req);
 579                                 reqsk_free(req);
 580                                 continue;
 581                         }
 582                         reqp = &req->dl_next;
 583                 }
 584
 585                 i = (i + 1) & (lopt->nr_table_entries - 1);
 586
 587         } while (--budget > 0);
 588
 589         lopt->clock_hand = i;
 590
 591         if (lopt->qlen)
 592                 inet_csk_reset_keepalive_timer(parent, interval);
 593 }
 594 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
 595
 596 /**
 597  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 598  *      @sk: the socket to clone
 599  *      @req: request_sock
 600  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 601  *
 602  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 603  */
 604 struct sock *inet_csk_clone_lock(const struct sock *sk,
 605                                  const struct request_sock *req,
 606                                  const gfp_t priority)
 607 {
 608         struct sock *newsk = sk_clone_lock(sk, priority);
 609
 610         if (newsk != NULL) {
 611                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 612
 613                 newsk->sk_state = TCP_SYN_RECV;
 614                 newicsk->icsk_bind_hash = NULL;
 615
 616                 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
 617                 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
 618                 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
 619                 newsk->sk_write_space = sk_stream_write_space;
 620
 621                 newicsk->icsk_retransmits = 0;
 622                 newicsk->icsk_backoff     = 0;
 623                 newicsk->icsk_probes_out  = 0;
 624
 625                 /* Deinitialize accept_queue to trap illegal accesses. */
 626                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 627
 628                 security_inet_csk_clone(newsk, req);
 629         }
 630         return newsk;
 631 }
 632 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 633
 634 /*
 635  * At this point, there should be no process reference to this
 636  * socket, and thus no user references at all.  Therefore we
 637  * can assume the socket waitqueue is inactive and nobody will
 638  * try to jump onto it.
 639  */
 640 void inet_csk_destroy_sock(struct sock *sk)
 641 {
 642         WARN_ON(sk->sk_state != TCP_CLOSE);
 643         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 644
 645         /* It cannot be in hash table! */
 646         WARN_ON(!sk_unhashed(sk));
 647
 648         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 649         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 650
 651         sk->sk_prot->destroy(sk);
 652
 653         sk_stream_kill_queues(sk);
 654
 655         xfrm_sk_free_policy(sk);
 656
 657         sk_refcnt_debug_release(sk);
 658
 659         percpu_counter_dec(sk->sk_prot->orphan_count);
 660         sock_put(sk);
 661 }
 662 EXPORT_SYMBOL(inet_csk_destroy_sock);
 663
 664 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 665 {
 666         struct inet_sock *inet = inet_sk(sk);
 667         struct inet_connection_sock *icsk = inet_csk(sk);
 668         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 669
 670         if (rc != 0)
 671                 return rc;
 672
 673         sk->sk_max_ack_backlog = 0;
 674         sk->sk_ack_backlog = 0;
 675         inet_csk_delack_init(sk);
 676
 677         /* There is race window here: we announce ourselves listening,
 678          * but this transition is still not validated by get_port().
 679          * It is OK, because this socket enters to hash table only
 680          * after validation is complete.
 681          */
 682         sk->sk_state = TCP_LISTEN;
 683         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 684                 inet->inet_sport = htons(inet->inet_num);
 685
 686                 sk_dst_reset(sk);
 687                 sk->sk_prot->hash(sk);
 688
 689                 return 0;
 690         }
 691
 692         sk->sk_state = TCP_CLOSE;
 693         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 694         return -EADDRINUSE;
 695 }
 696 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 697
 698 /*
 699  *      This routine closes sockets which have been at least partially
 700  *      opened, but not yet accepted.
 701  */
 702 void inet_csk_listen_stop(struct sock *sk)
 703 {
 704         struct inet_connection_sock *icsk = inet_csk(sk);
 705         struct request_sock *acc_req;
 706         struct request_sock *req;
 707
 708         inet_csk_delete_keepalive_timer(sk);
 709
 710         /* make all the listen_opt local to us */
 711         acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
 712
 713         /* Following specs, it would be better either to send FIN
 714          * (and enter FIN-WAIT-1, it is normal close)
 715          * or to send active reset (abort).
 716          * Certainly, it is pretty dangerous while synflood, but it is
 717          * bad justification for our negligence 8)
 718          * To be honest, we are not able to make either
 719          * of the variants now.                 --ANK
 720          */
 721         reqsk_queue_destroy(&icsk->icsk_accept_queue);
 722
 723         while ((req = acc_req) != NULL) {
 724                 struct sock *child = req->sk;
 725
 726                 acc_req = req->dl_next;
 727
 728                 local_bh_disable();
 729                 bh_lock_sock(child);
 730                 WARN_ON(sock_owned_by_user(child));
 731                 sock_hold(child);
 732
 733                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 734
 735                 sock_orphan(child);
 736
 737                 percpu_counter_inc(sk->sk_prot->orphan_count);
 738
 739                 inet_csk_destroy_sock(child);
 740
 741                 bh_unlock_sock(child);
 742                 local_bh_enable();
 743                 sock_put(child);
 744
 745                 sk_acceptq_removed(sk);
 746                 __reqsk_free(req);
 747         }
 748         WARN_ON(sk->sk_ack_backlog);
 749 }
 750 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 751
 752 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 753 {
 754         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 755         const struct inet_sock *inet = inet_sk(sk);
 756
 757         sin->sin_family         = AF_INET;
 758         sin->sin_addr.s_addr    = inet->inet_daddr;
 759         sin->sin_port           = inet->inet_dport;
 760 }
 761 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 762
 763 #ifdef CONFIG_COMPAT
 764 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 765                                char __user *optval, int __user *optlen)
 766 {
 767         const struct inet_connection_sock *icsk = inet_csk(sk);
 768
 769         if (icsk->icsk_af_ops->compat_getsockopt != NULL)
 770                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 771                                                             optval, optlen);
 772         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 773                                              optval, optlen);
 774 }
 775 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 776
 777 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 778                                char __user *optval, unsigned int optlen)
 779 {
 780         const struct inet_connection_sock *icsk = inet_csk(sk);
 781
 782         if (icsk->icsk_af_ops->compat_setsockopt != NULL)
 783                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 784                                                             optval, optlen);
 785         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 786                                              optval, optlen);
 787 }
 788 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 789 #endif