kernel/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                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
 131                                                     !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
 132                                                         snum = smallest_rover;
 133                                                         goto tb_found;
 134                                                 }
 135                                         }
 136                                         if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
 137                                                 snum = rover;
 138                                                 goto tb_found;
 139                                         }
 140                                         goto next;
 141                                 }
 142                         break;
 143                 next:
 144                         spin_unlock(&head->lock);
 145                 next_nolock:
 146                         if (++rover > high)
 147                                 rover = low;
 148                 } while (--remaining > 0);
 149
 150                 /* Exhausted local port range during search?  It is not
 151                  * possible for us to be holding one of the bind hash
 152                  * locks if this test triggers, because if 'remaining'
 153                  * drops to zero, we broke out of the do/while loop at
 154                  * the top level, not from the 'break;' statement.
 155                  */
 156                 ret = 1;
 157                 if (remaining <= 0) {
 158                         if (smallest_size != -1) {
 159                                 snum = smallest_rover;
 160                                 goto have_snum;
 161                         }
 162                         goto fail;
 163                 }
 164                 /* OK, here is the one we will use.  HEAD is
 165                  * non-NULL and we hold it's mutex.
 166                  */
 167                 snum = rover;
 168         } else {
 169 have_snum:
 170                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 171                                 hashinfo->bhash_size)];
 172                 spin_lock(&head->lock);
 173                 inet_bind_bucket_for_each(tb, &head->chain)
 174                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 175                                 goto tb_found;
 176         }
 177         tb = NULL;
 178         goto tb_not_found;
 179 tb_found:
 180         if (!hlist_empty(&tb->owners)) {
 181                 if (sk->sk_reuse == SK_FORCE_REUSE)
 182                         goto success;
 183
 184                 if (((tb->fastreuse > 0 &&
 185                       sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 186                      (tb->fastreuseport > 0 &&
 187                       sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
 188                     smallest_size == -1) {
 189                         goto success;
 190                 } else {
 191                         ret = 1;
 192                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
 193                                 if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 194                                      (tb->fastreuseport > 0 &&
 195                                       sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
 196                                     smallest_size != -1 && --attempts >= 0) {
 197                                         spin_unlock(&head->lock);
 198                                         goto again;
 199                                 }
 200
 201                                 goto fail_unlock;
 202                         }
 203                 }
 204         }
 205 tb_not_found:
 206         ret = 1;
 207         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 208                                         net, head, snum)) == NULL)
 209                 goto fail_unlock;
 210         if (hlist_empty(&tb->owners)) {
 211                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 212                         tb->fastreuse = 1;
 213                 else
 214                         tb->fastreuse = 0;
 215                 if (sk->sk_reuseport) {
 216                         tb->fastreuseport = 1;
 217                         tb->fastuid = uid;
 218                 } else
 219                         tb->fastreuseport = 0;
 220         } else {
 221                 if (tb->fastreuse &&
 222                     (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 223                         tb->fastreuse = 0;
 224                 if (tb->fastreuseport &&
 225                     (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)))
 226                         tb->fastreuseport = 0;
 227         }
 228 success:
 229         if (!inet_csk(sk)->icsk_bind_hash)
 230                 inet_bind_hash(sk, tb, snum);
 231         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 232         ret = 0;
 233
 234 fail_unlock:
 235         spin_unlock(&head->lock);
 236 fail:
 237         local_bh_enable();
 238         return ret;
 239 }
 240 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 241
 242 /*
 243  * Wait for an incoming connection, avoid race conditions. This must be called
 244  * with the socket locked.
 245  */
 246 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 247 {
 248         struct inet_connection_sock *icsk = inet_csk(sk);
 249         DEFINE_WAIT(wait);
 250         int err;
 251
 252         /*
 253          * True wake-one mechanism for incoming connections: only
 254          * one process gets woken up, not the 'whole herd'.
 255          * Since we do not 'race & poll' for established sockets
 256          * anymore, the common case will execute the loop only once.
 257          *
 258          * Subtle issue: "add_wait_queue_exclusive()" will be added
 259          * after any current non-exclusive waiters, and we know that
 260          * it will always _stay_ after any new non-exclusive waiters
 261          * because all non-exclusive waiters are added at the
 262          * beginning of the wait-queue. As such, it's ok to "drop"
 263          * our exclusiveness temporarily when we get woken up without
 264          * having to remove and re-insert us on the wait queue.
 265          */
 266         for (;;) {
 267                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 268                                           TASK_INTERRUPTIBLE);
 269                 release_sock(sk);
 270                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 271                         timeo = schedule_timeout(timeo);
 272                 sched_annotate_sleep();
 273                 lock_sock(sk);
 274                 err = 0;
 275                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 276                         break;
 277                 err = -EINVAL;
 278                 if (sk->sk_state != TCP_LISTEN)
 279                         break;
 280                 err = sock_intr_errno(timeo);
 281                 if (signal_pending(current))
 282                         break;
 283                 err = -EAGAIN;
 284                 if (!timeo)
 285                         break;
 286         }
 287         finish_wait(sk_sleep(sk), &wait);
 288         return err;
 289 }
 290
 291 /*
 292  * This will accept the next outstanding connection.
 293  */
 294 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 295 {
 296         struct inet_connection_sock *icsk = inet_csk(sk);
 297         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 298         struct request_sock *req;
 299         struct sock *newsk;
 300         int error;
 301
 302         lock_sock(sk);
 303
 304         /* We need to make sure that this socket is listening,
 305          * and that it has something pending.
 306          */
 307         error = -EINVAL;
 308         if (sk->sk_state != TCP_LISTEN)
 309                 goto out_err;
 310
 311         /* Find already established connection */
 312         if (reqsk_queue_empty(queue)) {
 313                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 314
 315                 /* If this is a non blocking socket don't sleep */
 316                 error = -EAGAIN;
 317                 if (!timeo)
 318                         goto out_err;
 319
 320                 error = inet_csk_wait_for_connect(sk, timeo);
 321                 if (error)
 322                         goto out_err;
 323         }
 324         req = reqsk_queue_remove(queue);
 325         newsk = req->sk;
 326
 327         sk_acceptq_removed(sk);
 328         if (sk->sk_protocol == IPPROTO_TCP &&
 329             tcp_rsk(req)->tfo_listener &&
 330             queue->fastopenq) {
 331                 spin_lock_bh(&queue->fastopenq->lock);
 332                 if (tcp_rsk(req)->tfo_listener) {
 333                         /* We are still waiting for the final ACK from 3WHS
 334                          * so can't free req now. Instead, we set req->sk to
 335                          * NULL to signify that the child socket is taken
 336                          * so reqsk_fastopen_remove() will free the req
 337                          * when 3WHS finishes (or is aborted).
 338                          */
 339                         req->sk = NULL;
 340                         req = NULL;
 341                 }
 342                 spin_unlock_bh(&queue->fastopenq->lock);
 343         }
 344 out:
 345         release_sock(sk);
 346         if (req)
 347                 reqsk_put(req);
 348         return newsk;
 349 out_err:
 350         newsk = NULL;
 351         req = NULL;
 352         *err = error;
 353         goto out;
 354 }
 355 EXPORT_SYMBOL(inet_csk_accept);
 356
 357 /*
 358  * Using different timers for retransmit, delayed acks and probes
 359  * We may wish use just one timer maintaining a list of expire jiffies
 360  * to optimize.
 361  */
 362 void inet_csk_init_xmit_timers(struct sock *sk,
 363                                void (*retransmit_handler)(unsigned long),
 364                                void (*delack_handler)(unsigned long),
 365                                void (*keepalive_handler)(unsigned long))
 366 {
 367         struct inet_connection_sock *icsk = inet_csk(sk);
 368
 369         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 370                         (unsigned long)sk);
 371         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 372                         (unsigned long)sk);
 373         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 374         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 375 }
 376 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 377
 378 void inet_csk_clear_xmit_timers(struct sock *sk)
 379 {
 380         struct inet_connection_sock *icsk = inet_csk(sk);
 381
 382         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 383
 384         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 385         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 386         sk_stop_timer(sk, &sk->sk_timer);
 387 }
 388 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 389
 390 void inet_csk_delete_keepalive_timer(struct sock *sk)
 391 {
 392         sk_stop_timer(sk, &sk->sk_timer);
 393 }
 394 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 395
 396 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 397 {
 398         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 399 }
 400 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 401
 402 struct dst_entry *inet_csk_route_req(struct sock *sk,
 403                                      struct flowi4 *fl4,
 404                                      const struct request_sock *req)
 405 {
 406         const struct inet_request_sock *ireq = inet_rsk(req);
 407         struct net *net = read_pnet(&ireq->ireq_net);
 408         struct ip_options_rcu *opt = ireq->opt;
 409         struct rtable *rt;
 410
 411         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 412                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 413                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 414                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 415                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 416                            htons(ireq->ir_num));
 417         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 418         rt = ip_route_output_flow(net, fl4, sk);
 419         if (IS_ERR(rt))
 420                 goto no_route;
 421         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 422                 goto route_err;
 423         return &rt->dst;
 424
 425 route_err:
 426         ip_rt_put(rt);
 427 no_route:
 428         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 429         return NULL;
 430 }
 431 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 432
 433 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
 434                                             struct sock *newsk,
 435                                             const struct request_sock *req)
 436 {
 437         const struct inet_request_sock *ireq = inet_rsk(req);
 438         struct net *net = read_pnet(&ireq->ireq_net);
 439         struct inet_sock *newinet = inet_sk(newsk);
 440         struct ip_options_rcu *opt;
 441         struct flowi4 *fl4;
 442         struct rtable *rt;
 443
 444         fl4 = &newinet->cork.fl.u.ip4;
 445
 446         rcu_read_lock();
 447         opt = rcu_dereference(newinet->inet_opt);
 448         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 449                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 450                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 451                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 452                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 453                            htons(ireq->ir_num));
 454         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 455         rt = ip_route_output_flow(net, fl4, sk);
 456         if (IS_ERR(rt))
 457                 goto no_route;
 458         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 459                 goto route_err;
 460         rcu_read_unlock();
 461         return &rt->dst;
 462
 463 route_err:
 464         ip_rt_put(rt);
 465 no_route:
 466         rcu_read_unlock();
 467         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 468         return NULL;
 469 }
 470 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 471
 472 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 473                                  const u32 rnd, const u32 synq_hsize)
 474 {
 475         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 476 }
 477
 478 #if IS_ENABLED(CONFIG_IPV6)
 479 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 480 #else
 481 #define AF_INET_FAMILY(fam) true
 482 #endif
 483
 484 /* Note: this is temporary :
 485  * req sock will no longer be in listener hash table
 486 */
 487 struct request_sock *inet_csk_search_req(struct sock *sk,
 488                                          const __be16 rport,
 489                                          const __be32 raddr,
 490                                          const __be32 laddr)
 491 {
 492         struct inet_connection_sock *icsk = inet_csk(sk);
 493         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 494         struct request_sock *req;
 495         u32 hash = inet_synq_hash(raddr, rport, lopt->hash_rnd,
 496                                   lopt->nr_table_entries);
 497
 498         spin_lock(&icsk->icsk_accept_queue.syn_wait_lock);
 499         for (req = lopt->syn_table[hash]; req != NULL; req = req->dl_next) {
 500                 const struct inet_request_sock *ireq = inet_rsk(req);
 501
 502                 if (ireq->ir_rmt_port == rport &&
 503                     ireq->ir_rmt_addr == raddr &&
 504                     ireq->ir_loc_addr == laddr &&
 505                     AF_INET_FAMILY(req->rsk_ops->family)) {
 506                         atomic_inc(&req->rsk_refcnt);
 507                         WARN_ON(req->sk);
 508                         break;
 509                 }
 510         }
 511         spin_unlock(&icsk->icsk_accept_queue.syn_wait_lock);
 512
 513         return req;
 514 }
 515 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 516
 517 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 518                                    unsigned long timeout)
 519 {
 520         struct inet_connection_sock *icsk = inet_csk(sk);
 521         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 522         const u32 h = inet_synq_hash(inet_rsk(req)->ir_rmt_addr,
 523                                      inet_rsk(req)->ir_rmt_port,
 524                                      lopt->hash_rnd, lopt->nr_table_entries);
 525
 526         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 527         inet_csk_reqsk_queue_added(sk, timeout);
 528 }
 529 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 530
 531 /* Only thing we need from tcp.h */
 532 extern int sysctl_tcp_synack_retries;
 533
 534
 535 /* Decide when to expire the request and when to resend SYN-ACK */
 536 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 537                                   const int max_retries,
 538                                   const u8 rskq_defer_accept,
 539                                   int *expire, int *resend)
 540 {
 541         if (!rskq_defer_accept) {
 542                 *expire = req->num_timeout >= thresh;
 543                 *resend = 1;
 544                 return;
 545         }
 546         *expire = req->num_timeout >= thresh &&
 547                   (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 548         /*
 549          * Do not resend while waiting for data after ACK,
 550          * start to resend on end of deferring period to give
 551          * last chance for data or ACK to create established socket.
 552          */
 553         *resend = !inet_rsk(req)->acked ||
 554                   req->num_timeout >= rskq_defer_accept - 1;
 555 }
 556
 557 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req)
 558 {
 559         int err = req->rsk_ops->rtx_syn_ack(parent, req);
 560
 561         if (!err)
 562                 req->num_retrans++;
 563         return err;
 564 }
 565 EXPORT_SYMBOL(inet_rtx_syn_ack);
 566
 567 /* return true if req was found in the syn_table[] */
 568 static bool reqsk_queue_unlink(struct request_sock_queue *queue,
 569                                struct request_sock *req)
 570 {
 571         struct listen_sock *lopt = queue->listen_opt;
 572         struct request_sock **prev;
 573         bool found = false;
 574
 575         spin_lock(&queue->syn_wait_lock);
 576
 577         for (prev = &lopt->syn_table[req->rsk_hash]; *prev != NULL;
 578              prev = &(*prev)->dl_next) {
 579                 if (*prev == req) {
 580                         *prev = req->dl_next;
 581                         found = true;
 582                         break;
 583                 }
 584         }
 585
 586         spin_unlock(&queue->syn_wait_lock);
 587         if (del_timer(&req->rsk_timer))
 588                 reqsk_put(req);
 589         return found;
 590 }
 591
 592 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
 593 {
 594         if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
 595                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 596                 reqsk_put(req);
 597         }
 598 }
 599 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
 600
 601 static void reqsk_timer_handler(unsigned long data)
 602 {
 603         struct request_sock *req = (struct request_sock *)data;
 604         struct sock *sk_listener = req->rsk_listener;
 605         struct inet_connection_sock *icsk = inet_csk(sk_listener);
 606         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 607         struct listen_sock *lopt = queue->listen_opt;
 608         int qlen, expire = 0, resend = 0;
 609         int max_retries, thresh;
 610         u8 defer_accept;
 611
 612         if (sk_listener->sk_state != TCP_LISTEN || !lopt) {
 613                 reqsk_put(req);
 614                 return;
 615         }
 616
 617         max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 618         thresh = max_retries;
 619         /* Normally all the openreqs are young and become mature
 620          * (i.e. converted to established socket) for first timeout.
 621          * If synack was not acknowledged for 1 second, it means
 622          * one of the following things: synack was lost, ack was lost,
 623          * rtt is high or nobody planned to ack (i.e. synflood).
 624          * When server is a bit loaded, queue is populated with old
 625          * open requests, reducing effective size of queue.
 626          * When server is well loaded, queue size reduces to zero
 627          * after several minutes of work. It is not synflood,
 628          * it is normal operation. The solution is pruning
 629          * too old entries overriding normal timeout, when
 630          * situation becomes dangerous.
 631          *
 632          * Essentially, we reserve half of room for young
 633          * embrions; and abort old ones without pity, if old
 634          * ones are about to clog our table.
 635          */
 636         qlen = listen_sock_qlen(lopt);
 637         if (qlen >> (lopt->max_qlen_log - 1)) {
 638                 int young = listen_sock_young(lopt) << 1;
 639
 640                 while (thresh > 2) {
 641                         if (qlen < young)
 642                                 break;
 643                         thresh--;
 644                         young <<= 1;
 645                 }
 646         }
 647         defer_accept = READ_ONCE(queue->rskq_defer_accept);
 648         if (defer_accept)
 649                 max_retries = defer_accept;
 650         syn_ack_recalc(req, thresh, max_retries, defer_accept,
 651                        &expire, &resend);
 652         req->rsk_ops->syn_ack_timeout(req);
 653         if (!expire &&
 654             (!resend ||
 655              !inet_rtx_syn_ack(sk_listener, req) ||
 656              inet_rsk(req)->acked)) {
 657                 unsigned long timeo;
 658
 659                 if (req->num_timeout++ == 0)
 660                         atomic_inc(&lopt->young_dec);
 661                 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 662                 mod_timer_pinned(&req->rsk_timer, jiffies + timeo);
 663                 return;
 664         }
 665         inet_csk_reqsk_queue_drop(sk_listener, req);
 666         reqsk_put(req);
 667 }
 668
 669 void reqsk_queue_hash_req(struct request_sock_queue *queue,
 670                           u32 hash, struct request_sock *req,
 671                           unsigned long timeout)
 672 {
 673         struct listen_sock *lopt = queue->listen_opt;
 674
 675         req->num_retrans = 0;
 676         req->num_timeout = 0;
 677         req->sk = NULL;
 678
 679         /* before letting lookups find us, make sure all req fields
 680          * are committed to memory and refcnt initialized.
 681          */
 682         smp_wmb();
 683         atomic_set(&req->rsk_refcnt, 2);
 684         setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req);
 685         req->rsk_hash = hash;
 686
 687         spin_lock(&queue->syn_wait_lock);
 688         req->dl_next = lopt->syn_table[hash];
 689         lopt->syn_table[hash] = req;
 690         spin_unlock(&queue->syn_wait_lock);
 691
 692         mod_timer_pinned(&req->rsk_timer, jiffies + timeout);
 693 }
 694 EXPORT_SYMBOL(reqsk_queue_hash_req);
 695
 696 /**
 697  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 698  *      @sk: the socket to clone
 699  *      @req: request_sock
 700  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 701  *
 702  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 703  */
 704 struct sock *inet_csk_clone_lock(const struct sock *sk,
 705                                  const struct request_sock *req,
 706                                  const gfp_t priority)
 707 {
 708         struct sock *newsk = sk_clone_lock(sk, priority);
 709
 710         if (newsk) {
 711                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 712
 713                 newsk->sk_state = TCP_SYN_RECV;
 714                 newicsk->icsk_bind_hash = NULL;
 715
 716                 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
 717                 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
 718                 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
 719                 newsk->sk_write_space = sk_stream_write_space;
 720
 721                 newsk->sk_mark = inet_rsk(req)->ir_mark;
 722                 atomic64_set(&newsk->sk_cookie,
 723                              atomic64_read(&inet_rsk(req)->ir_cookie));
 724
 725                 newicsk->icsk_retransmits = 0;
 726                 newicsk->icsk_backoff     = 0;
 727                 newicsk->icsk_probes_out  = 0;
 728
 729                 /* Deinitialize accept_queue to trap illegal accesses. */
 730                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 731
 732                 security_inet_csk_clone(newsk, req);
 733         }
 734         return newsk;
 735 }
 736 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 737
 738 /*
 739  * At this point, there should be no process reference to this
 740  * socket, and thus no user references at all.  Therefore we
 741  * can assume the socket waitqueue is inactive and nobody will
 742  * try to jump onto it.
 743  */
 744 void inet_csk_destroy_sock(struct sock *sk)
 745 {
 746         WARN_ON(sk->sk_state != TCP_CLOSE);
 747         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 748
 749         /* It cannot be in hash table! */
 750         WARN_ON(!sk_unhashed(sk));
 751
 752         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 753         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 754
 755         sk->sk_prot->destroy(sk);
 756
 757         sk_stream_kill_queues(sk);
 758
 759         xfrm_sk_free_policy(sk);
 760
 761         sk_refcnt_debug_release(sk);
 762
 763         percpu_counter_dec(sk->sk_prot->orphan_count);
 764         sock_put(sk);
 765 }
 766 EXPORT_SYMBOL(inet_csk_destroy_sock);
 767
 768 /* This function allows to force a closure of a socket after the call to
 769  * tcp/dccp_create_openreq_child().
 770  */
 771 void inet_csk_prepare_forced_close(struct sock *sk)
 772         __releases(&sk->sk_lock.slock)
 773 {
 774         /* sk_clone_lock locked the socket and set refcnt to 2 */
 775         bh_unlock_sock(sk);
 776         sock_put(sk);
 777
 778         /* The below has to be done to allow calling inet_csk_destroy_sock */
 779         sock_set_flag(sk, SOCK_DEAD);
 780         percpu_counter_inc(sk->sk_prot->orphan_count);
 781         inet_sk(sk)->inet_num = 0;
 782 }
 783 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
 784
 785 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 786 {
 787         struct inet_sock *inet = inet_sk(sk);
 788         struct inet_connection_sock *icsk = inet_csk(sk);
 789         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 790
 791         if (rc != 0)
 792                 return rc;
 793
 794         sk->sk_max_ack_backlog = 0;
 795         sk->sk_ack_backlog = 0;
 796         inet_csk_delack_init(sk);
 797
 798         /* There is race window here: we announce ourselves listening,
 799          * but this transition is still not validated by get_port().
 800          * It is OK, because this socket enters to hash table only
 801          * after validation is complete.
 802          */
 803         sk->sk_state = TCP_LISTEN;
 804         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 805                 inet->inet_sport = htons(inet->inet_num);
 806
 807                 sk_dst_reset(sk);
 808                 sk->sk_prot->hash(sk);
 809
 810                 return 0;
 811         }
 812
 813         sk->sk_state = TCP_CLOSE;
 814         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 815         return -EADDRINUSE;
 816 }
 817 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 818
 819 /*
 820  *      This routine closes sockets which have been at least partially
 821  *      opened, but not yet accepted.
 822  */
 823 void inet_csk_listen_stop(struct sock *sk)
 824 {
 825         struct inet_connection_sock *icsk = inet_csk(sk);
 826         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 827         struct request_sock *acc_req;
 828         struct request_sock *req;
 829
 830         /* make all the listen_opt local to us */
 831         acc_req = reqsk_queue_yank_acceptq(queue);
 832
 833         /* Following specs, it would be better either to send FIN
 834          * (and enter FIN-WAIT-1, it is normal close)
 835          * or to send active reset (abort).
 836          * Certainly, it is pretty dangerous while synflood, but it is
 837          * bad justification for our negligence 8)
 838          * To be honest, we are not able to make either
 839          * of the variants now.                 --ANK
 840          */
 841         reqsk_queue_destroy(queue);
 842
 843         while ((req = acc_req) != NULL) {
 844                 struct sock *child = req->sk;
 845
 846                 acc_req = req->dl_next;
 847
 848                 local_bh_disable();
 849                 bh_lock_sock(child);
 850                 WARN_ON(sock_owned_by_user(child));
 851                 sock_hold(child);
 852
 853                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 854
 855                 sock_orphan(child);
 856
 857                 percpu_counter_inc(sk->sk_prot->orphan_count);
 858
 859                 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
 860                         BUG_ON(tcp_sk(child)->fastopen_rsk != req);
 861                         BUG_ON(sk != req->rsk_listener);
 862
 863                         /* Paranoid, to prevent race condition if
 864                          * an inbound pkt destined for child is
 865                          * blocked by sock lock in tcp_v4_rcv().
 866                          * Also to satisfy an assertion in
 867                          * tcp_v4_destroy_sock().
 868                          */
 869                         tcp_sk(child)->fastopen_rsk = NULL;
 870                 }
 871                 inet_csk_destroy_sock(child);
 872
 873                 bh_unlock_sock(child);
 874                 local_bh_enable();
 875                 sock_put(child);
 876
 877                 sk_acceptq_removed(sk);
 878                 reqsk_put(req);
 879         }
 880         if (queue->fastopenq) {
 881                 /* Free all the reqs queued in rskq_rst_head. */
 882                 spin_lock_bh(&queue->fastopenq->lock);
 883                 acc_req = queue->fastopenq->rskq_rst_head;
 884                 queue->fastopenq->rskq_rst_head = NULL;
 885                 spin_unlock_bh(&queue->fastopenq->lock);
 886                 while ((req = acc_req) != NULL) {
 887                         acc_req = req->dl_next;
 888                         reqsk_put(req);
 889                 }
 890         }
 891         WARN_ON(sk->sk_ack_backlog);
 892 }
 893 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 894
 895 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 896 {
 897         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 898         const struct inet_sock *inet = inet_sk(sk);
 899
 900         sin->sin_family         = AF_INET;
 901         sin->sin_addr.s_addr    = inet->inet_daddr;
 902         sin->sin_port           = inet->inet_dport;
 903 }
 904 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 905
 906 #ifdef CONFIG_COMPAT
 907 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 908                                char __user *optval, int __user *optlen)
 909 {
 910         const struct inet_connection_sock *icsk = inet_csk(sk);
 911
 912         if (icsk->icsk_af_ops->compat_getsockopt)
 913                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 914                                                             optval, optlen);
 915         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 916                                              optval, optlen);
 917 }
 918 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 919
 920 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 921                                char __user *optval, unsigned int optlen)
 922 {
 923         const struct inet_connection_sock *icsk = inet_csk(sk);
 924
 925         if (icsk->icsk_af_ops->compat_setsockopt)
 926                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 927                                                             optval, optlen);
 928         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 929                                              optval, optlen);
 930 }
 931 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 932 #endif
 933
 934 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
 935 {
 936         const struct inet_sock *inet = inet_sk(sk);
 937         const struct ip_options_rcu *inet_opt;
 938         __be32 daddr = inet->inet_daddr;
 939         struct flowi4 *fl4;
 940         struct rtable *rt;
 941
 942         rcu_read_lock();
 943         inet_opt = rcu_dereference(inet->inet_opt);
 944         if (inet_opt && inet_opt->opt.srr)
 945                 daddr = inet_opt->opt.faddr;
 946         fl4 = &fl->u.ip4;
 947         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
 948                                    inet->inet_saddr, inet->inet_dport,
 949                                    inet->inet_sport, sk->sk_protocol,
 950                                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
 951         if (IS_ERR(rt))
 952                 rt = NULL;
 953         if (rt)
 954                 sk_setup_caps(sk, &rt->dst);
 955         rcu_read_unlock();
 956
 957         return &rt->dst;
 958 }
 959
 960 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
 961 {
 962         struct dst_entry *dst = __sk_dst_check(sk, 0);
 963         struct inet_sock *inet = inet_sk(sk);
 964
 965         if (!dst) {
 966                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 967                 if (!dst)
 968                         goto out;
 969         }
 970         dst->ops->update_pmtu(dst, sk, NULL, mtu);
 971
 972         dst = __sk_dst_check(sk, 0);
 973         if (!dst)
 974                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 975 out:
 976         return dst;
 977 }
 978 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);