net/rxrpc/conn_client.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* Client connection-specific management code.
   3  *
   4  * Copyright (C) 2016, 2020 Red Hat, Inc. All Rights Reserved.
   5  * Written by David Howells (dhowells@redhat.com)
   6  *
   7  * Client connections need to be cached for a little while after they've made a
   8  * call so as to handle retransmitted DATA packets in case the server didn't
   9  * receive the final ACK or terminating ABORT we sent it.
  10  *
  11  * There are flags of relevance to the cache:
  12  *
  13  *  (2) DONT_REUSE - The connection should be discarded as soon as possible and
  14  *      should not be reused.  This is set when an exclusive connection is used
  15  *      or a call ID counter overflows.
  16  *
  17  * The caching state may only be changed if the cache lock is held.
  18  *
  19  * There are two idle client connection expiry durations.  If the total number
  20  * of connections is below the reap threshold, we use the normal duration; if
  21  * it's above, we use the fast duration.
  22  */
  23
  24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  25
  26 #include <linux/slab.h>
  27 #include <linux/idr.h>
  28 #include <linux/timer.h>
  29 #include <linux/sched/signal.h>
  30
  31 #include "ar-internal.h"
  32
  33 __read_mostly unsigned int rxrpc_reap_client_connections = 900;
  34 __read_mostly unsigned long rxrpc_conn_idle_client_expiry = 2 * 60 * HZ;
  35 __read_mostly unsigned long rxrpc_conn_idle_client_fast_expiry = 2 * HZ;
  36
  37 /*
  38  * We use machine-unique IDs for our client connections.
  39  */
  40 DEFINE_IDR(rxrpc_client_conn_ids);
  41 static DEFINE_SPINLOCK(rxrpc_conn_id_lock);
  42
  43 /*
  44  * Get a connection ID and epoch for a client connection from the global pool.
  45  * The connection struct pointer is then recorded in the idr radix tree.  The
  46  * epoch doesn't change until the client is rebooted (or, at least, unless the
  47  * module is unloaded).
  48  */
  49 static int rxrpc_get_client_connection_id(struct rxrpc_connection *conn,
  50                                           gfp_t gfp)
  51 {
  52         struct rxrpc_net *rxnet = conn->params.local->rxnet;
  53         int id;
  54
  55         _enter("");
  56
  57         idr_preload(gfp);
  58         spin_lock(&rxrpc_conn_id_lock);
  59
  60         id = idr_alloc_cyclic(&rxrpc_client_conn_ids, conn,
  61                               1, 0x40000000, GFP_NOWAIT);
  62         if (id < 0)
  63                 goto error;
  64
  65         spin_unlock(&rxrpc_conn_id_lock);
  66         idr_preload_end();
  67
  68         conn->proto.epoch = rxnet->epoch;
  69         conn->proto.cid = id << RXRPC_CIDSHIFT;
  70         set_bit(RXRPC_CONN_HAS_IDR, &conn->flags);
  71         _leave(" [CID %x]", conn->proto.cid);
  72         return 0;
  73
  74 error:
  75         spin_unlock(&rxrpc_conn_id_lock);
  76         idr_preload_end();
  77         _leave(" = %d", id);
  78         return id;
  79 }
  80
  81 /*
  82  * Release a connection ID for a client connection from the global pool.
  83  */
  84 static void rxrpc_put_client_connection_id(struct rxrpc_connection *conn)
  85 {
  86         if (test_bit(RXRPC_CONN_HAS_IDR, &conn->flags)) {
  87                 spin_lock(&rxrpc_conn_id_lock);
  88                 idr_remove(&rxrpc_client_conn_ids,
  89                            conn->proto.cid >> RXRPC_CIDSHIFT);
  90                 spin_unlock(&rxrpc_conn_id_lock);
  91         }
  92 }
  93
  94 /*
  95  * Destroy the client connection ID tree.
  96  */
  97 void rxrpc_destroy_client_conn_ids(void)
  98 {
  99         struct rxrpc_connection *conn;
 100         int id;
 101
 102         if (!idr_is_empty(&rxrpc_client_conn_ids)) {
 103                 idr_for_each_entry(&rxrpc_client_conn_ids, conn, id) {
 104                         pr_err("AF_RXRPC: Leaked client conn %p {%d}\n",
 105                                conn, atomic_read(&conn->usage));
 106                 }
 107                 BUG();
 108         }
 109
 110         idr_destroy(&rxrpc_client_conn_ids);
 111 }
 112
 113 /*
 114  * Allocate a connection bundle.
 115  */
 116 static struct rxrpc_bundle *rxrpc_alloc_bundle(struct rxrpc_conn_parameters *cp,
 117                                                gfp_t gfp)
 118 {
 119         struct rxrpc_bundle *bundle;
 120
 121         bundle = kzalloc(sizeof(*bundle), gfp);
 122         if (bundle) {
 123                 bundle->params = *cp;
 124                 rxrpc_get_peer(bundle->params.peer);
 125                 atomic_set(&bundle->usage, 1);
 126                 spin_lock_init(&bundle->channel_lock);
 127                 INIT_LIST_HEAD(&bundle->waiting_calls);
 128         }
 129         return bundle;
 130 }
 131
 132 struct rxrpc_bundle *rxrpc_get_bundle(struct rxrpc_bundle *bundle)
 133 {
 134         atomic_inc(&bundle->usage);
 135         return bundle;
 136 }
 137
 138 void rxrpc_put_bundle(struct rxrpc_bundle *bundle)
 139 {
 140         unsigned int d = bundle->debug_id;
 141         unsigned int u = atomic_dec_return(&bundle->usage);
 142
 143         _debug("PUT B=%x %u", d, u);
 144         if (u == 0) {
 145                 rxrpc_put_peer(bundle->params.peer);
 146                 kfree(bundle);
 147         }
 148 }
 149
 150 /*
 151  * Allocate a client connection.
 152  */
 153 static struct rxrpc_connection *
 154 rxrpc_alloc_client_connection(struct rxrpc_bundle *bundle, gfp_t gfp)
 155 {
 156         struct rxrpc_connection *conn;
 157         struct rxrpc_net *rxnet = bundle->params.local->rxnet;
 158         int ret;
 159
 160         _enter("");
 161
 162         conn = rxrpc_alloc_connection(gfp);
 163         if (!conn) {
 164                 _leave(" = -ENOMEM");
 165                 return ERR_PTR(-ENOMEM);
 166         }
 167
 168         atomic_set(&conn->usage, 1);
 169         conn->bundle            = bundle;
 170         conn->params            = bundle->params;
 171         conn->out_clientflag    = RXRPC_CLIENT_INITIATED;
 172         conn->state             = RXRPC_CONN_CLIENT;
 173         conn->service_id        = conn->params.service_id;
 174
 175         ret = rxrpc_get_client_connection_id(conn, gfp);
 176         if (ret < 0)
 177                 goto error_0;
 178
 179         ret = rxrpc_init_client_conn_security(conn);
 180         if (ret < 0)
 181                 goto error_1;
 182
 183         atomic_inc(&rxnet->nr_conns);
 184         write_lock(&rxnet->conn_lock);
 185         list_add_tail(&conn->proc_link, &rxnet->conn_proc_list);
 186         write_unlock(&rxnet->conn_lock);
 187
 188         rxrpc_get_bundle(bundle);
 189         rxrpc_get_peer(conn->params.peer);
 190         rxrpc_get_local(conn->params.local);
 191         key_get(conn->params.key);
 192
 193         trace_rxrpc_conn(conn->debug_id, rxrpc_conn_new_client,
 194                          atomic_read(&conn->usage),
 195                          __builtin_return_address(0));
 196
 197         atomic_inc(&rxnet->nr_client_conns);
 198         trace_rxrpc_client(conn, -1, rxrpc_client_alloc);
 199         _leave(" = %p", conn);
 200         return conn;
 201
 202 error_1:
 203         rxrpc_put_client_connection_id(conn);
 204 error_0:
 205         kfree(conn);
 206         _leave(" = %d", ret);
 207         return ERR_PTR(ret);
 208 }
 209
 210 /*
 211  * Determine if a connection may be reused.
 212  */
 213 static bool rxrpc_may_reuse_conn(struct rxrpc_connection *conn)
 214 {
 215         struct rxrpc_net *rxnet;
 216         int id_cursor, id, distance, limit;
 217
 218         if (!conn)
 219                 goto dont_reuse;
 220
 221         rxnet = conn->params.local->rxnet;
 222         if (test_bit(RXRPC_CONN_DONT_REUSE, &conn->flags))
 223                 goto dont_reuse;
 224
 225         if (conn->state != RXRPC_CONN_CLIENT ||
 226             conn->proto.epoch != rxnet->epoch)
 227                 goto mark_dont_reuse;
 228
 229         /* The IDR tree gets very expensive on memory if the connection IDs are
 230          * widely scattered throughout the number space, so we shall want to
 231          * kill off connections that, say, have an ID more than about four
 232          * times the maximum number of client conns away from the current
 233          * allocation point to try and keep the IDs concentrated.
 234          */
 235         id_cursor = idr_get_cursor(&rxrpc_client_conn_ids);
 236         id = conn->proto.cid >> RXRPC_CIDSHIFT;
 237         distance = id - id_cursor;
 238         if (distance < 0)
 239                 distance = -distance;
 240         limit = max_t(unsigned long, atomic_read(&rxnet->nr_conns) * 4, 1024);
 241         if (distance > limit)
 242                 goto mark_dont_reuse;
 243
 244         return true;
 245
 246 mark_dont_reuse:
 247         set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
 248 dont_reuse:
 249         return false;
 250 }
 251
 252 /*
 253  * Look up the conn bundle that matches the connection parameters, adding it if
 254  * it doesn't yet exist.
 255  */
 256 static struct rxrpc_bundle *rxrpc_look_up_bundle(struct rxrpc_conn_parameters *cp,
 257                                                  gfp_t gfp)
 258 {
 259         static atomic_t rxrpc_bundle_id;
 260         struct rxrpc_bundle *bundle, *candidate;
 261         struct rxrpc_local *local = cp->local;
 262         struct rb_node *p, **pp, *parent;
 263         long diff;
 264
 265         _enter("{%px,%x,%u,%u}",
 266                cp->peer, key_serial(cp->key), cp->security_level, cp->upgrade);
 267
 268         if (cp->exclusive)
 269                 return rxrpc_alloc_bundle(cp, gfp);
 270
 271         /* First, see if the bundle is already there. */
 272         _debug("search 1");
 273         spin_lock(&local->client_bundles_lock);
 274         p = local->client_bundles.rb_node;
 275         while (p) {
 276                 bundle = rb_entry(p, struct rxrpc_bundle, local_node);
 277
 278 #define cmp(X) ((long)bundle->params.X - (long)cp->X)
 279                 diff = (cmp(peer) ?:
 280                         cmp(key) ?:
 281                         cmp(security_level) ?:
 282                         cmp(upgrade));
 283 #undef cmp
 284                 if (diff < 0)
 285                         p = p->rb_left;
 286                 else if (diff > 0)
 287                         p = p->rb_right;
 288                 else
 289                         goto found_bundle;
 290         }
 291         spin_unlock(&local->client_bundles_lock);
 292         _debug("not found");
 293
 294         /* It wasn't.  We need to add one. */
 295         candidate = rxrpc_alloc_bundle(cp, gfp);
 296         if (!candidate)
 297                 return NULL;
 298
 299         _debug("search 2");
 300         spin_lock(&local->client_bundles_lock);
 301         pp = &local->client_bundles.rb_node;
 302         parent = NULL;
 303         while (*pp) {
 304                 parent = *pp;
 305                 bundle = rb_entry(parent, struct rxrpc_bundle, local_node);
 306
 307 #define cmp(X) ((long)bundle->params.X - (long)cp->X)
 308                 diff = (cmp(peer) ?:
 309                         cmp(key) ?:
 310                         cmp(security_level) ?:
 311                         cmp(upgrade));
 312 #undef cmp
 313                 if (diff < 0)
 314                         pp = &(*pp)->rb_left;
 315                 else if (diff > 0)
 316                         pp = &(*pp)->rb_right;
 317                 else
 318                         goto found_bundle_free;
 319         }
 320
 321         _debug("new bundle");
 322         candidate->debug_id = atomic_inc_return(&rxrpc_bundle_id);
 323         rb_link_node(&candidate->local_node, parent, pp);
 324         rb_insert_color(&candidate->local_node, &local->client_bundles);
 325         rxrpc_get_bundle(candidate);
 326         spin_unlock(&local->client_bundles_lock);
 327         _leave(" = %u [new]", candidate->debug_id);
 328         return candidate;
 329
 330 found_bundle_free:
 331         kfree(candidate);
 332 found_bundle:
 333         rxrpc_get_bundle(bundle);
 334         spin_unlock(&local->client_bundles_lock);
 335         _leave(" = %u [found]", bundle->debug_id);
 336         return bundle;
 337 }
 338
 339 /*
 340  * Create or find a client bundle to use for a call.
 341  *
 342  * If we return with a connection, the call will be on its waiting list.  It's
 343  * left to the caller to assign a channel and wake up the call.
 344  */
 345 static struct rxrpc_bundle *rxrpc_prep_call(struct rxrpc_sock *rx,
 346                                             struct rxrpc_call *call,
 347                                             struct rxrpc_conn_parameters *cp,
 348                                             struct sockaddr_rxrpc *srx,
 349                                             gfp_t gfp)
 350 {
 351         struct rxrpc_bundle *bundle;
 352
 353         _enter("{%d,%lx},", call->debug_id, call->user_call_ID);
 354
 355         cp->peer = rxrpc_lookup_peer(rx, cp->local, srx, gfp);
 356         if (!cp->peer)
 357                 goto error;
 358
 359         call->cong_cwnd = cp->peer->cong_cwnd;
 360         if (call->cong_cwnd >= call->cong_ssthresh)
 361                 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
 362         else
 363                 call->cong_mode = RXRPC_CALL_SLOW_START;
 364         if (cp->upgrade)
 365                 __set_bit(RXRPC_CALL_UPGRADE, &call->flags);
 366
 367         /* Find the client connection bundle. */
 368         bundle = rxrpc_look_up_bundle(cp, gfp);
 369         if (!bundle)
 370                 goto error;
 371
 372         /* Get this call queued.  Someone else may activate it whilst we're
 373          * lining up a new connection, but that's fine.
 374          */
 375         spin_lock(&bundle->channel_lock);
 376         list_add_tail(&call->chan_wait_link, &bundle->waiting_calls);
 377         spin_unlock(&bundle->channel_lock);
 378
 379         _leave(" = [B=%x]", bundle->debug_id);
 380         return bundle;
 381
 382 error:
 383         _leave(" = -ENOMEM");
 384         return ERR_PTR(-ENOMEM);
 385 }
 386
 387 /*
 388  * Allocate a new connection and add it into a bundle.
 389  */
 390 static void rxrpc_add_conn_to_bundle(struct rxrpc_bundle *bundle, gfp_t gfp)
 391         __releases(bundle->channel_lock)
 392 {
 393         struct rxrpc_connection *candidate = NULL, *old = NULL;
 394         bool conflict;
 395         int i;
 396
 397         _enter("");
 398
 399         conflict = bundle->alloc_conn;
 400         if (!conflict)
 401                 bundle->alloc_conn = true;
 402         spin_unlock(&bundle->channel_lock);
 403         if (conflict) {
 404                 _leave(" [conf]");
 405                 return;
 406         }
 407
 408         candidate = rxrpc_alloc_client_connection(bundle, gfp);
 409
 410         spin_lock(&bundle->channel_lock);
 411         bundle->alloc_conn = false;
 412
 413         if (IS_ERR(candidate)) {
 414                 bundle->alloc_error = PTR_ERR(candidate);
 415                 spin_unlock(&bundle->channel_lock);
 416                 _leave(" [err %ld]", PTR_ERR(candidate));
 417                 return;
 418         }
 419
 420         bundle->alloc_error = 0;
 421
 422         for (i = 0; i < ARRAY_SIZE(bundle->conns); i++) {
 423                 unsigned int shift = i * RXRPC_MAXCALLS;
 424                 int j;
 425
 426                 old = bundle->conns[i];
 427                 if (!rxrpc_may_reuse_conn(old)) {
 428                         if (old)
 429                                 trace_rxrpc_client(old, -1, rxrpc_client_replace);
 430                         candidate->bundle_shift = shift;
 431                         bundle->conns[i] = candidate;
 432                         for (j = 0; j < RXRPC_MAXCALLS; j++)
 433                                 set_bit(shift + j, &bundle->avail_chans);
 434                         candidate = NULL;
 435                         break;
 436                 }
 437
 438                 old = NULL;
 439         }
 440
 441         spin_unlock(&bundle->channel_lock);
 442
 443         if (candidate) {
 444                 _debug("discard C=%x", candidate->debug_id);
 445                 trace_rxrpc_client(candidate, -1, rxrpc_client_duplicate);
 446                 rxrpc_put_connection(candidate);
 447         }
 448
 449         rxrpc_put_connection(old);
 450         _leave("");
 451 }
 452
 453 /*
 454  * Add a connection to a bundle if there are no usable connections or we have
 455  * connections waiting for extra capacity.
 456  */
 457 static void rxrpc_maybe_add_conn(struct rxrpc_bundle *bundle, gfp_t gfp)
 458 {
 459         struct rxrpc_call *call;
 460         int i, usable;
 461
 462         _enter("");
 463
 464         spin_lock(&bundle->channel_lock);
 465
 466         /* See if there are any usable connections. */
 467         usable = 0;
 468         for (i = 0; i < ARRAY_SIZE(bundle->conns); i++)
 469                 if (rxrpc_may_reuse_conn(bundle->conns[i]))
 470                         usable++;
 471
 472         if (!usable && !list_empty(&bundle->waiting_calls)) {
 473                 call = list_first_entry(&bundle->waiting_calls,
 474                                         struct rxrpc_call, chan_wait_link);
 475                 if (test_bit(RXRPC_CALL_UPGRADE, &call->flags))
 476                         bundle->try_upgrade = true;
 477         }
 478
 479         if (!usable)
 480                 goto alloc_conn;
 481
 482         if (!bundle->avail_chans &&
 483             !bundle->try_upgrade &&
 484             !list_empty(&bundle->waiting_calls) &&
 485             usable < ARRAY_SIZE(bundle->conns))
 486                 goto alloc_conn;
 487
 488         spin_unlock(&bundle->channel_lock);
 489         _leave("");
 490         return;
 491
 492 alloc_conn:
 493         return rxrpc_add_conn_to_bundle(bundle, gfp);
 494 }
 495
 496 /*
 497  * Assign a channel to the call at the front of the queue and wake the call up.
 498  * We don't increment the callNumber counter until this number has been exposed
 499  * to the world.
 500  */
 501 static void rxrpc_activate_one_channel(struct rxrpc_connection *conn,
 502                                        unsigned int channel)
 503 {
 504         struct rxrpc_channel *chan = &conn->channels[channel];
 505         struct rxrpc_bundle *bundle = conn->bundle;
 506         struct rxrpc_call *call = list_entry(bundle->waiting_calls.next,
 507                                              struct rxrpc_call, chan_wait_link);
 508         u32 call_id = chan->call_counter + 1;
 509
 510         _enter("C=%x,%u", conn->debug_id, channel);
 511
 512         trace_rxrpc_client(conn, channel, rxrpc_client_chan_activate);
 513
 514         /* Cancel the final ACK on the previous call if it hasn't been sent yet
 515          * as the DATA packet will implicitly ACK it.
 516          */
 517         clear_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
 518         clear_bit(conn->bundle_shift + channel, &bundle->avail_chans);
 519
 520         rxrpc_see_call(call);
 521         list_del_init(&call->chan_wait_link);
 522         call->peer      = rxrpc_get_peer(conn->params.peer);
 523         call->conn      = rxrpc_get_connection(conn);
 524         call->cid       = conn->proto.cid | channel;
 525         call->call_id   = call_id;
 526         call->security  = conn->security;
 527         call->security_ix = conn->security_ix;
 528         call->service_id = conn->service_id;
 529
 530         trace_rxrpc_connect_call(call);
 531         _net("CONNECT call %08x:%08x as call %d on conn %d",
 532              call->cid, call->call_id, call->debug_id, conn->debug_id);
 533
 534         write_lock_bh(&call->state_lock);
 535         call->state = RXRPC_CALL_CLIENT_SEND_REQUEST;
 536         write_unlock_bh(&call->state_lock);
 537
 538         /* Paired with the read barrier in rxrpc_connect_call().  This orders
 539          * cid and epoch in the connection wrt to call_id without the need to
 540          * take the channel_lock.
 541          *
 542          * We provisionally assign a callNumber at this point, but we don't
 543          * confirm it until the call is about to be exposed.
 544          *
 545          * TODO: Pair with a barrier in the data_ready handler when that looks
 546          * at the call ID through a connection channel.
 547          */
 548         smp_wmb();
 549
 550         chan->call_id           = call_id;
 551         chan->call_debug_id     = call->debug_id;
 552         rcu_assign_pointer(chan->call, call);
 553         wake_up(&call->waitq);
 554 }
 555
 556 /*
 557  * Remove a connection from the idle list if it's on it.
 558  */
 559 static void rxrpc_unidle_conn(struct rxrpc_bundle *bundle, struct rxrpc_connection *conn)
 560 {
 561         struct rxrpc_net *rxnet = bundle->params.local->rxnet;
 562         bool drop_ref;
 563
 564         if (!list_empty(&conn->cache_link)) {
 565                 drop_ref = false;
 566                 spin_lock(&rxnet->client_conn_cache_lock);
 567                 if (!list_empty(&conn->cache_link)) {
 568                         list_del_init(&conn->cache_link);
 569                         drop_ref = true;
 570                 }
 571                 spin_unlock(&rxnet->client_conn_cache_lock);
 572                 if (drop_ref)
 573                         rxrpc_put_connection(conn);
 574         }
 575 }
 576
 577 /*
 578  * Assign channels and callNumbers to waiting calls with channel_lock
 579  * held by caller.
 580  */
 581 static void rxrpc_activate_channels_locked(struct rxrpc_bundle *bundle)
 582 {
 583         struct rxrpc_connection *conn;
 584         unsigned long avail, mask;
 585         unsigned int channel, slot;
 586
 587         if (bundle->try_upgrade)
 588                 mask = 1;
 589         else
 590                 mask = ULONG_MAX;
 591
 592         while (!list_empty(&bundle->waiting_calls)) {
 593                 avail = bundle->avail_chans & mask;
 594                 if (!avail)
 595                         break;
 596                 channel = __ffs(avail);
 597                 clear_bit(channel, &bundle->avail_chans);
 598
 599                 slot = channel / RXRPC_MAXCALLS;
 600                 conn = bundle->conns[slot];
 601                 if (!conn)
 602                         break;
 603
 604                 if (bundle->try_upgrade)
 605                         set_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags);
 606                 rxrpc_unidle_conn(bundle, conn);
 607
 608                 channel &= (RXRPC_MAXCALLS - 1);
 609                 conn->act_chans |= 1 << channel;
 610                 rxrpc_activate_one_channel(conn, channel);
 611         }
 612 }
 613
 614 /*
 615  * Assign channels and callNumbers to waiting calls.
 616  */
 617 static void rxrpc_activate_channels(struct rxrpc_bundle *bundle)
 618 {
 619         _enter("B=%x", bundle->debug_id);
 620
 621         trace_rxrpc_client(NULL, -1, rxrpc_client_activate_chans);
 622
 623         if (!bundle->avail_chans)
 624                 return;
 625
 626         spin_lock(&bundle->channel_lock);
 627         rxrpc_activate_channels_locked(bundle);
 628         spin_unlock(&bundle->channel_lock);
 629         _leave("");
 630 }
 631
 632 /*
 633  * Wait for a callNumber and a channel to be granted to a call.
 634  */
 635 static int rxrpc_wait_for_channel(struct rxrpc_bundle *bundle,
 636                                   struct rxrpc_call *call, gfp_t gfp)
 637 {
 638         DECLARE_WAITQUEUE(myself, current);
 639         int ret = 0;
 640
 641         _enter("%d", call->debug_id);
 642
 643         if (!gfpflags_allow_blocking(gfp)) {
 644                 rxrpc_maybe_add_conn(bundle, gfp);
 645                 rxrpc_activate_channels(bundle);
 646                 ret = bundle->alloc_error ?: -EAGAIN;
 647                 goto out;
 648         }
 649
 650         add_wait_queue_exclusive(&call->waitq, &myself);
 651         for (;;) {
 652                 rxrpc_maybe_add_conn(bundle, gfp);
 653                 rxrpc_activate_channels(bundle);
 654                 ret = bundle->alloc_error;
 655                 if (ret < 0)
 656                         break;
 657
 658                 switch (call->interruptibility) {
 659                 case RXRPC_INTERRUPTIBLE:
 660                 case RXRPC_PREINTERRUPTIBLE:
 661                         set_current_state(TASK_INTERRUPTIBLE);
 662                         break;
 663                 case RXRPC_UNINTERRUPTIBLE:
 664                 default:
 665                         set_current_state(TASK_UNINTERRUPTIBLE);
 666                         break;
 667                 }
 668                 if (READ_ONCE(call->state) != RXRPC_CALL_CLIENT_AWAIT_CONN)
 669                         break;
 670                 if ((call->interruptibility == RXRPC_INTERRUPTIBLE ||
 671                      call->interruptibility == RXRPC_PREINTERRUPTIBLE) &&
 672                     signal_pending(current)) {
 673                         ret = -ERESTARTSYS;
 674                         break;
 675                 }
 676                 schedule();
 677         }
 678         remove_wait_queue(&call->waitq, &myself);
 679         __set_current_state(TASK_RUNNING);
 680
 681 out:
 682         _leave(" = %d", ret);
 683         return ret;
 684 }
 685
 686 /*
 687  * find a connection for a call
 688  * - called in process context with IRQs enabled
 689  */
 690 int rxrpc_connect_call(struct rxrpc_sock *rx,
 691                        struct rxrpc_call *call,
 692                        struct rxrpc_conn_parameters *cp,
 693                        struct sockaddr_rxrpc *srx,
 694                        gfp_t gfp)
 695 {
 696         struct rxrpc_bundle *bundle;
 697         struct rxrpc_net *rxnet = cp->local->rxnet;
 698         int ret = 0;
 699
 700         _enter("{%d,%lx},", call->debug_id, call->user_call_ID);
 701
 702         rxrpc_discard_expired_client_conns(&rxnet->client_conn_reaper);
 703
 704         bundle = rxrpc_prep_call(rx, call, cp, srx, gfp);
 705         if (IS_ERR(bundle)) {
 706                 ret = PTR_ERR(bundle);
 707                 goto out;
 708         }
 709
 710         if (call->state == RXRPC_CALL_CLIENT_AWAIT_CONN) {
 711                 ret = rxrpc_wait_for_channel(bundle, call, gfp);
 712                 if (ret < 0)
 713                         goto wait_failed;
 714         }
 715
 716 granted_channel:
 717         /* Paired with the write barrier in rxrpc_activate_one_channel(). */
 718         smp_rmb();
 719
 720 out_put_bundle:
 721         rxrpc_put_bundle(bundle);
 722 out:
 723         _leave(" = %d", ret);
 724         return ret;
 725
 726 wait_failed:
 727         spin_lock(&bundle->channel_lock);
 728         list_del_init(&call->chan_wait_link);
 729         spin_unlock(&bundle->channel_lock);
 730
 731         if (call->state != RXRPC_CALL_CLIENT_AWAIT_CONN) {
 732                 ret = 0;
 733                 goto granted_channel;
 734         }
 735
 736         trace_rxrpc_client(call->conn, ret, rxrpc_client_chan_wait_failed);
 737         rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR, 0, ret);
 738         rxrpc_disconnect_client_call(bundle, call);
 739         goto out_put_bundle;
 740 }
 741
 742 /*
 743  * Note that a call, and thus a connection, is about to be exposed to the
 744  * world.
 745  */
 746 void rxrpc_expose_client_call(struct rxrpc_call *call)
 747 {
 748         unsigned int channel = call->cid & RXRPC_CHANNELMASK;
 749         struct rxrpc_connection *conn = call->conn;
 750         struct rxrpc_channel *chan = &conn->channels[channel];
 751
 752         if (!test_and_set_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
 753                 /* Mark the call ID as being used.  If the callNumber counter
 754                  * exceeds ~2 billion, we kill the connection after its
 755                  * outstanding calls have finished so that the counter doesn't
 756                  * wrap.
 757                  */
 758                 chan->call_counter++;
 759                 if (chan->call_counter >= INT_MAX)
 760                         set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
 761                 trace_rxrpc_client(conn, channel, rxrpc_client_exposed);
 762         }
 763 }
 764
 765 /*
 766  * Set the reap timer.
 767  */
 768 static void rxrpc_set_client_reap_timer(struct rxrpc_net *rxnet)
 769 {
 770         if (!rxnet->kill_all_client_conns) {
 771                 unsigned long now = jiffies;
 772                 unsigned long reap_at = now + rxrpc_conn_idle_client_expiry;
 773
 774                 if (rxnet->live)
 775                         timer_reduce(&rxnet->client_conn_reap_timer, reap_at);
 776         }
 777 }
 778
 779 /*
 780  * Disconnect a client call.
 781  */
 782 void rxrpc_disconnect_client_call(struct rxrpc_bundle *bundle, struct rxrpc_call *call)
 783 {
 784         struct rxrpc_connection *conn;
 785         struct rxrpc_channel *chan = NULL;
 786         struct rxrpc_net *rxnet = bundle->params.local->rxnet;
 787         unsigned int channel;
 788         bool may_reuse;
 789         u32 cid;
 790
 791         _enter("c=%x", call->debug_id);
 792
 793         spin_lock(&bundle->channel_lock);
 794         set_bit(RXRPC_CALL_DISCONNECTED, &call->flags);
 795
 796         /* Calls that have never actually been assigned a channel can simply be
 797          * discarded.
 798          */
 799         conn = call->conn;
 800         if (!conn) {
 801                 _debug("call is waiting");
 802                 ASSERTCMP(call->call_id, ==, 0);
 803                 ASSERT(!test_bit(RXRPC_CALL_EXPOSED, &call->flags));
 804                 list_del_init(&call->chan_wait_link);
 805                 goto out;
 806         }
 807
 808         cid = call->cid;
 809         channel = cid & RXRPC_CHANNELMASK;
 810         chan = &conn->channels[channel];
 811         trace_rxrpc_client(conn, channel, rxrpc_client_chan_disconnect);
 812
 813         if (rcu_access_pointer(chan->call) != call) {
 814                 spin_unlock(&bundle->channel_lock);
 815                 BUG();
 816         }
 817
 818         may_reuse = rxrpc_may_reuse_conn(conn);
 819
 820         /* If a client call was exposed to the world, we save the result for
 821          * retransmission.
 822          *
 823          * We use a barrier here so that the call number and abort code can be
 824          * read without needing to take a lock.
 825          *
 826          * TODO: Make the incoming packet handler check this and handle
 827          * terminal retransmission without requiring access to the call.
 828          */
 829         if (test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
 830                 _debug("exposed %u,%u", call->call_id, call->abort_code);
 831                 __rxrpc_disconnect_call(conn, call);
 832
 833                 if (test_and_clear_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags)) {
 834                         trace_rxrpc_client(conn, channel, rxrpc_client_to_active);
 835                         bundle->try_upgrade = false;
 836                         if (may_reuse)
 837                                 rxrpc_activate_channels_locked(bundle);
 838                 }
 839
 840         }
 841
 842         /* See if we can pass the channel directly to another call. */
 843         if (may_reuse && !list_empty(&bundle->waiting_calls)) {
 844                 trace_rxrpc_client(conn, channel, rxrpc_client_chan_pass);
 845                 rxrpc_activate_one_channel(conn, channel);
 846                 goto out;
 847         }
 848
 849         /* Schedule the final ACK to be transmitted in a short while so that it
 850          * can be skipped if we find a follow-on call.  The first DATA packet
 851          * of the follow on call will implicitly ACK this call.
 852          */
 853         if (call->completion == RXRPC_CALL_SUCCEEDED &&
 854             test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
 855                 unsigned long final_ack_at = jiffies + 2;
 856
 857                 WRITE_ONCE(chan->final_ack_at, final_ack_at);
 858                 smp_wmb(); /* vs rxrpc_process_delayed_final_acks() */
 859                 set_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
 860                 rxrpc_reduce_conn_timer(conn, final_ack_at);
 861         }
 862
 863         /* Deactivate the channel. */
 864         rcu_assign_pointer(chan->call, NULL);
 865         set_bit(conn->bundle_shift + channel, &conn->bundle->avail_chans);
 866         conn->act_chans &= ~(1 << channel);
 867
 868         /* If no channels remain active, then put the connection on the idle
 869          * list for a short while.  Give it a ref to stop it going away if it
 870          * becomes unbundled.
 871          */
 872         if (!conn->act_chans) {
 873                 trace_rxrpc_client(conn, channel, rxrpc_client_to_idle);
 874                 conn->idle_timestamp = jiffies;
 875
 876                 rxrpc_get_connection(conn);
 877                 spin_lock(&rxnet->client_conn_cache_lock);
 878                 list_move_tail(&conn->cache_link, &rxnet->idle_client_conns);
 879                 spin_unlock(&rxnet->client_conn_cache_lock);
 880
 881                 rxrpc_set_client_reap_timer(rxnet);
 882         }
 883
 884 out:
 885         spin_unlock(&bundle->channel_lock);
 886         _leave("");
 887         return;
 888 }
 889
 890 /*
 891  * Remove a connection from a bundle.
 892  */
 893 static void rxrpc_unbundle_conn(struct rxrpc_connection *conn)
 894 {
 895         struct rxrpc_bundle *bundle = conn->bundle;
 896         struct rxrpc_local *local = bundle->params.local;
 897         unsigned int bindex;
 898         bool need_drop = false, need_put = false;
 899         int i;
 900
 901         _enter("C=%x", conn->debug_id);
 902
 903         if (conn->flags & RXRPC_CONN_FINAL_ACK_MASK)
 904                 rxrpc_process_delayed_final_acks(conn, true);
 905
 906         spin_lock(&bundle->channel_lock);
 907         bindex = conn->bundle_shift / RXRPC_MAXCALLS;
 908         if (bundle->conns[bindex] == conn) {
 909                 _debug("clear slot %u", bindex);
 910                 bundle->conns[bindex] = NULL;
 911                 for (i = 0; i < RXRPC_MAXCALLS; i++)
 912                         clear_bit(conn->bundle_shift + i, &bundle->avail_chans);
 913                 need_drop = true;
 914         }
 915         spin_unlock(&bundle->channel_lock);
 916
 917         /* If there are no more connections, remove the bundle */
 918         if (!bundle->avail_chans) {
 919                 _debug("maybe unbundle");
 920                 spin_lock(&local->client_bundles_lock);
 921
 922                 for (i = 0; i < ARRAY_SIZE(bundle->conns); i++)
 923                         if (bundle->conns[i])
 924                                 break;
 925                 if (i == ARRAY_SIZE(bundle->conns) && !bundle->params.exclusive) {
 926                         _debug("erase bundle");
 927                         rb_erase(&bundle->local_node, &local->client_bundles);
 928                         need_put = true;
 929                 }
 930
 931                 spin_unlock(&local->client_bundles_lock);
 932                 if (need_put)
 933                         rxrpc_put_bundle(bundle);
 934         }
 935
 936         if (need_drop)
 937                 rxrpc_put_connection(conn);
 938         _leave("");
 939 }
 940
 941 /*
 942  * Clean up a dead client connection.
 943  */
 944 static void rxrpc_kill_client_conn(struct rxrpc_connection *conn)
 945 {
 946         struct rxrpc_local *local = conn->params.local;
 947         struct rxrpc_net *rxnet = local->rxnet;
 948
 949         _enter("C=%x", conn->debug_id);
 950
 951         trace_rxrpc_client(conn, -1, rxrpc_client_cleanup);
 952         atomic_dec(&rxnet->nr_client_conns);
 953
 954         rxrpc_put_client_connection_id(conn);
 955         rxrpc_kill_connection(conn);
 956 }
 957
 958 /*
 959  * Clean up a dead client connections.
 960  */
 961 void rxrpc_put_client_conn(struct rxrpc_connection *conn)
 962 {
 963         const void *here = __builtin_return_address(0);
 964         unsigned int debug_id = conn->debug_id;
 965         int n;
 966
 967         n = atomic_dec_return(&conn->usage);
 968         trace_rxrpc_conn(debug_id, rxrpc_conn_put_client, n, here);
 969         if (n <= 0) {
 970                 ASSERTCMP(n, >=, 0);
 971                 rxrpc_kill_client_conn(conn);
 972         }
 973 }
 974
 975 /*
 976  * Discard expired client connections from the idle list.  Each conn in the
 977  * idle list has been exposed and holds an extra ref because of that.
 978  *
 979  * This may be called from conn setup or from a work item so cannot be
 980  * considered non-reentrant.
 981  */
 982 void rxrpc_discard_expired_client_conns(struct work_struct *work)
 983 {
 984         struct rxrpc_connection *conn;
 985         struct rxrpc_net *rxnet =
 986                 container_of(work, struct rxrpc_net, client_conn_reaper);
 987         unsigned long expiry, conn_expires_at, now;
 988         unsigned int nr_conns;
 989
 990         _enter("");
 991
 992         if (list_empty(&rxnet->idle_client_conns)) {
 993                 _leave(" [empty]");
 994                 return;
 995         }
 996
 997         /* Don't double up on the discarding */
 998         if (!spin_trylock(&rxnet->client_conn_discard_lock)) {
 999                 _leave(" [already]");
1000                 return;
1001         }
1002
1003         /* We keep an estimate of what the number of conns ought to be after
1004          * we've discarded some so that we don't overdo the discarding.
1005          */
1006         nr_conns = atomic_read(&rxnet->nr_client_conns);
1007
1008 next:
1009         spin_lock(&rxnet->client_conn_cache_lock);
1010
1011         if (list_empty(&rxnet->idle_client_conns))
1012                 goto out;
1013
1014         conn = list_entry(rxnet->idle_client_conns.next,
1015                           struct rxrpc_connection, cache_link);
1016
1017         if (!rxnet->kill_all_client_conns) {
1018                 /* If the number of connections is over the reap limit, we
1019                  * expedite discard by reducing the expiry timeout.  We must,
1020                  * however, have at least a short grace period to be able to do
1021                  * final-ACK or ABORT retransmission.
1022                  */
1023                 expiry = rxrpc_conn_idle_client_expiry;
1024                 if (nr_conns > rxrpc_reap_client_connections)
1025                         expiry = rxrpc_conn_idle_client_fast_expiry;
1026                 if (conn->params.local->service_closed)
1027                         expiry = rxrpc_closed_conn_expiry * HZ;
1028
1029                 conn_expires_at = conn->idle_timestamp + expiry;
1030
1031                 now = READ_ONCE(jiffies);
1032                 if (time_after(conn_expires_at, now))
1033                         goto not_yet_expired;
1034         }
1035
1036         trace_rxrpc_client(conn, -1, rxrpc_client_discard);
1037         list_del_init(&conn->cache_link);
1038
1039         spin_unlock(&rxnet->client_conn_cache_lock);
1040
1041         rxrpc_unbundle_conn(conn);
1042         rxrpc_put_connection(conn); /* Drop the ->cache_link ref */
1043
1044         nr_conns--;
1045         goto next;
1046
1047 not_yet_expired:
1048         /* The connection at the front of the queue hasn't yet expired, so
1049          * schedule the work item for that point if we discarded something.
1050          *
1051          * We don't worry if the work item is already scheduled - it can look
1052          * after rescheduling itself at a later time.  We could cancel it, but
1053          * then things get messier.
1054          */
1055         _debug("not yet");
1056         if (!rxnet->kill_all_client_conns)
1057                 timer_reduce(&rxnet->client_conn_reap_timer, conn_expires_at);
1058
1059 out:
1060         spin_unlock(&rxnet->client_conn_cache_lock);
1061         spin_unlock(&rxnet->client_conn_discard_lock);
1062         _leave("");
1063 }
1064
1065 /*
1066  * Preemptively destroy all the client connection records rather than waiting
1067  * for them to time out
1068  */
1069 void rxrpc_destroy_all_client_connections(struct rxrpc_net *rxnet)
1070 {
1071         _enter("");
1072
1073         spin_lock(&rxnet->client_conn_cache_lock);
1074         rxnet->kill_all_client_conns = true;
1075         spin_unlock(&rxnet->client_conn_cache_lock);
1076
1077         del_timer_sync(&rxnet->client_conn_reap_timer);
1078
1079         if (!rxrpc_queue_work(&rxnet->client_conn_reaper))
1080                 _debug("destroy: queue failed");
1081
1082         _leave("");
1083 }
1084
1085 /*
1086  * Clean up the client connections on a local endpoint.
1087  */
1088 void rxrpc_clean_up_local_conns(struct rxrpc_local *local)
1089 {
1090         struct rxrpc_connection *conn, *tmp;
1091         struct rxrpc_net *rxnet = local->rxnet;
1092         LIST_HEAD(graveyard);
1093
1094         _enter("");
1095
1096         spin_lock(&rxnet->client_conn_cache_lock);
1097
1098         list_for_each_entry_safe(conn, tmp, &rxnet->idle_client_conns,
1099                                  cache_link) {
1100                 if (conn->params.local == local) {
1101                         trace_rxrpc_client(conn, -1, rxrpc_client_discard);
1102                         list_move(&conn->cache_link, &graveyard);
1103                 }
1104         }
1105
1106         spin_unlock(&rxnet->client_conn_cache_lock);
1107
1108         while (!list_empty(&graveyard)) {
1109                 conn = list_entry(graveyard.next,
1110                                   struct rxrpc_connection, cache_link);
1111                 list_del_init(&conn->cache_link);
1112                 rxrpc_unbundle_conn(conn);
1113                 rxrpc_put_connection(conn);
1114         }
1115
1116         _leave(" [culled]");
1117 }