fs/afs/server.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* AFS server record management
   3  *
   4  * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
   5  * Written by David Howells (dhowells@redhat.com)
   6  */
   7
   8 #include <linux/sched.h>
   9 #include <linux/slab.h>
  10 #include "afs_fs.h"
  11 #include "internal.h"
  12 #include "protocol_yfs.h"
  13
  14 static unsigned afs_server_gc_delay = 10;       /* Server record timeout in seconds */
  15 static unsigned afs_server_update_delay = 30;   /* Time till VLDB recheck in secs */
  16 static atomic_t afs_server_debug_id;
  17
  18 static void afs_inc_servers_outstanding(struct afs_net *net)
  19 {
  20         atomic_inc(&net->servers_outstanding);
  21 }
  22
  23 static void afs_dec_servers_outstanding(struct afs_net *net)
  24 {
  25         if (atomic_dec_and_test(&net->servers_outstanding))
  26                 wake_up_var(&net->servers_outstanding);
  27 }
  28
  29 /*
  30  * Find a server by one of its addresses.
  31  */
  32 struct afs_server *afs_find_server(struct afs_net *net,
  33                                    const struct sockaddr_rxrpc *srx)
  34 {
  35         const struct afs_addr_list *alist;
  36         struct afs_server *server = NULL;
  37         unsigned int i;
  38         int seq = 0, diff;
  39
  40         rcu_read_lock();
  41
  42         do {
  43                 if (server)
  44                         afs_put_server(net, server, afs_server_trace_put_find_rsq);
  45                 server = NULL;
  46                 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
  47
  48                 if (srx->transport.family == AF_INET6) {
  49                         const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
  50                         hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
  51                                 alist = rcu_dereference(server->addresses);
  52                                 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
  53                                         b = &alist->addrs[i].transport.sin6;
  54                                         diff = ((u16 __force)a->sin6_port -
  55                                                 (u16 __force)b->sin6_port);
  56                                         if (diff == 0)
  57                                                 diff = memcmp(&a->sin6_addr,
  58                                                               &b->sin6_addr,
  59                                                               sizeof(struct in6_addr));
  60                                         if (diff == 0)
  61                                                 goto found;
  62                                 }
  63                         }
  64                 } else {
  65                         const struct sockaddr_in *a = &srx->transport.sin, *b;
  66                         hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
  67                                 alist = rcu_dereference(server->addresses);
  68                                 for (i = 0; i < alist->nr_ipv4; i++) {
  69                                         b = &alist->addrs[i].transport.sin;
  70                                         diff = ((u16 __force)a->sin_port -
  71                                                 (u16 __force)b->sin_port);
  72                                         if (diff == 0)
  73                                                 diff = ((u32 __force)a->sin_addr.s_addr -
  74                                                         (u32 __force)b->sin_addr.s_addr);
  75                                         if (diff == 0)
  76                                                 goto found;
  77                                 }
  78                         }
  79                 }
  80
  81                 server = NULL;
  82         found:
  83                 if (server && !atomic_inc_not_zero(&server->usage))
  84                         server = NULL;
  85
  86         } while (need_seqretry(&net->fs_addr_lock, seq));
  87
  88         done_seqretry(&net->fs_addr_lock, seq);
  89
  90         rcu_read_unlock();
  91         return server;
  92 }
  93
  94 /*
  95  * Look up a server by its UUID
  96  */
  97 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
  98 {
  99         struct afs_server *server = NULL;
 100         struct rb_node *p;
 101         int diff, seq = 0;
 102
 103         _enter("%pU", uuid);
 104
 105         do {
 106                 /* Unfortunately, rbtree walking doesn't give reliable results
 107                  * under just the RCU read lock, so we have to check for
 108                  * changes.
 109                  */
 110                 if (server)
 111                         afs_put_server(net, server, afs_server_trace_put_uuid_rsq);
 112                 server = NULL;
 113
 114                 read_seqbegin_or_lock(&net->fs_lock, &seq);
 115
 116                 p = net->fs_servers.rb_node;
 117                 while (p) {
 118                         server = rb_entry(p, struct afs_server, uuid_rb);
 119
 120                         diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
 121                         if (diff < 0) {
 122                                 p = p->rb_left;
 123                         } else if (diff > 0) {
 124                                 p = p->rb_right;
 125                         } else {
 126                                 afs_get_server(server, afs_server_trace_get_by_uuid);
 127                                 break;
 128                         }
 129
 130                         server = NULL;
 131                 }
 132         } while (need_seqretry(&net->fs_lock, seq));
 133
 134         done_seqretry(&net->fs_lock, seq);
 135
 136         _leave(" = %p", server);
 137         return server;
 138 }
 139
 140 /*
 141  * Install a server record in the namespace tree
 142  */
 143 static struct afs_server *afs_install_server(struct afs_net *net,
 144                                              struct afs_server *candidate)
 145 {
 146         const struct afs_addr_list *alist;
 147         struct afs_server *server;
 148         struct rb_node **pp, *p;
 149         int ret = -EEXIST, diff;
 150
 151         _enter("%p", candidate);
 152
 153         write_seqlock(&net->fs_lock);
 154
 155         /* Firstly install the server in the UUID lookup tree */
 156         pp = &net->fs_servers.rb_node;
 157         p = NULL;
 158         while (*pp) {
 159                 p = *pp;
 160                 _debug("- consider %p", p);
 161                 server = rb_entry(p, struct afs_server, uuid_rb);
 162                 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
 163                 if (diff < 0)
 164                         pp = &(*pp)->rb_left;
 165                 else if (diff > 0)
 166                         pp = &(*pp)->rb_right;
 167                 else
 168                         goto exists;
 169         }
 170
 171         server = candidate;
 172         rb_link_node(&server->uuid_rb, p, pp);
 173         rb_insert_color(&server->uuid_rb, &net->fs_servers);
 174         hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
 175
 176         write_seqlock(&net->fs_addr_lock);
 177         alist = rcu_dereference_protected(server->addresses,
 178                                           lockdep_is_held(&net->fs_addr_lock.lock));
 179
 180         /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
 181          * it in the IPv4 and/or IPv6 reverse-map lists.
 182          *
 183          * TODO: For speed we want to use something other than a flat list
 184          * here; even sorting the list in terms of lowest address would help a
 185          * bit, but anything we might want to do gets messy and memory
 186          * intensive.
 187          */
 188         if (alist->nr_ipv4 > 0)
 189                 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
 190         if (alist->nr_addrs > alist->nr_ipv4)
 191                 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
 192
 193         write_sequnlock(&net->fs_addr_lock);
 194         ret = 0;
 195
 196 exists:
 197         afs_get_server(server, afs_server_trace_get_install);
 198         write_sequnlock(&net->fs_lock);
 199         return server;
 200 }
 201
 202 /*
 203  * allocate a new server record
 204  */
 205 static struct afs_server *afs_alloc_server(struct afs_net *net,
 206                                            const uuid_t *uuid,
 207                                            struct afs_addr_list *alist)
 208 {
 209         struct afs_server *server;
 210
 211         _enter("");
 212
 213         server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
 214         if (!server)
 215                 goto enomem;
 216
 217         atomic_set(&server->usage, 1);
 218         server->debug_id = atomic_inc_return(&afs_server_debug_id);
 219         RCU_INIT_POINTER(server->addresses, alist);
 220         server->addr_version = alist->version;
 221         server->uuid = *uuid;
 222         server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
 223         rwlock_init(&server->fs_lock);
 224         INIT_HLIST_HEAD(&server->cb_volumes);
 225         rwlock_init(&server->cb_break_lock);
 226         init_waitqueue_head(&server->probe_wq);
 227         spin_lock_init(&server->probe_lock);
 228
 229         afs_inc_servers_outstanding(net);
 230         trace_afs_server(server, 1, afs_server_trace_alloc);
 231         _leave(" = %p", server);
 232         return server;
 233
 234 enomem:
 235         _leave(" = NULL [nomem]");
 236         return NULL;
 237 }
 238
 239 /*
 240  * Look up an address record for a server
 241  */
 242 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
 243                                                  struct key *key, const uuid_t *uuid)
 244 {
 245         struct afs_vl_cursor vc;
 246         struct afs_addr_list *alist = NULL;
 247         int ret;
 248
 249         ret = -ERESTARTSYS;
 250         if (afs_begin_vlserver_operation(&vc, cell, key)) {
 251                 while (afs_select_vlserver(&vc)) {
 252                         if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
 253                                 alist = afs_yfsvl_get_endpoints(&vc, uuid);
 254                         else
 255                                 alist = afs_vl_get_addrs_u(&vc, uuid);
 256                 }
 257
 258                 ret = afs_end_vlserver_operation(&vc);
 259         }
 260
 261         return ret < 0 ? ERR_PTR(ret) : alist;
 262 }
 263
 264 /*
 265  * Get or create a fileserver record.
 266  */
 267 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
 268                                      const uuid_t *uuid)
 269 {
 270         struct afs_addr_list *alist;
 271         struct afs_server *server, *candidate;
 272
 273         _enter("%p,%pU", cell->net, uuid);
 274
 275         server = afs_find_server_by_uuid(cell->net, uuid);
 276         if (server)
 277                 return server;
 278
 279         alist = afs_vl_lookup_addrs(cell, key, uuid);
 280         if (IS_ERR(alist))
 281                 return ERR_CAST(alist);
 282
 283         candidate = afs_alloc_server(cell->net, uuid, alist);
 284         if (!candidate) {
 285                 afs_put_addrlist(alist);
 286                 return ERR_PTR(-ENOMEM);
 287         }
 288
 289         server = afs_install_server(cell->net, candidate);
 290         if (server != candidate) {
 291                 afs_put_addrlist(alist);
 292                 kfree(candidate);
 293         }
 294
 295         _leave(" = %p{%d}", server, atomic_read(&server->usage));
 296         return server;
 297 }
 298
 299 /*
 300  * Set the server timer to fire after a given delay, assuming it's not already
 301  * set for an earlier time.
 302  */
 303 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
 304 {
 305         if (net->live) {
 306                 afs_inc_servers_outstanding(net);
 307                 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
 308                         afs_dec_servers_outstanding(net);
 309         }
 310 }
 311
 312 /*
 313  * Server management timer.  We have an increment on fs_outstanding that we
 314  * need to pass along to the work item.
 315  */
 316 void afs_servers_timer(struct timer_list *timer)
 317 {
 318         struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
 319
 320         _enter("");
 321         if (!queue_work(afs_wq, &net->fs_manager))
 322                 afs_dec_servers_outstanding(net);
 323 }
 324
 325 /*
 326  * Get a reference on a server object.
 327  */
 328 struct afs_server *afs_get_server(struct afs_server *server,
 329                                   enum afs_server_trace reason)
 330 {
 331         unsigned int u = atomic_inc_return(&server->usage);
 332
 333         trace_afs_server(server, u, reason);
 334         return server;
 335 }
 336
 337 /*
 338  * Release a reference on a server record.
 339  */
 340 void afs_put_server(struct afs_net *net, struct afs_server *server,
 341                     enum afs_server_trace reason)
 342 {
 343         unsigned int usage;
 344
 345         if (!server)
 346                 return;
 347
 348         server->put_time = ktime_get_real_seconds();
 349
 350         usage = atomic_dec_return(&server->usage);
 351
 352         trace_afs_server(server, usage, reason);
 353
 354         if (likely(usage > 0))
 355                 return;
 356
 357         afs_set_server_timer(net, afs_server_gc_delay);
 358 }
 359
 360 static void afs_server_rcu(struct rcu_head *rcu)
 361 {
 362         struct afs_server *server = container_of(rcu, struct afs_server, rcu);
 363
 364         trace_afs_server(server, atomic_read(&server->usage),
 365                          afs_server_trace_free);
 366         afs_put_addrlist(rcu_access_pointer(server->addresses));
 367         kfree(server);
 368 }
 369
 370 /*
 371  * destroy a dead server
 372  */
 373 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
 374 {
 375         struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
 376         struct afs_addr_cursor ac = {
 377                 .alist  = alist,
 378                 .index  = alist->preferred,
 379                 .error  = 0,
 380         };
 381
 382         trace_afs_server(server, atomic_read(&server->usage),
 383                          afs_server_trace_give_up_cb);
 384
 385         if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
 386                 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
 387
 388         wait_var_event(&server->probe_outstanding,
 389                        atomic_read(&server->probe_outstanding) == 0);
 390
 391         trace_afs_server(server, atomic_read(&server->usage),
 392                          afs_server_trace_destroy);
 393         call_rcu(&server->rcu, afs_server_rcu);
 394         afs_dec_servers_outstanding(net);
 395 }
 396
 397 /*
 398  * Garbage collect any expired servers.
 399  */
 400 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
 401 {
 402         struct afs_server *server;
 403         bool deleted;
 404         int usage;
 405
 406         while ((server = gc_list)) {
 407                 gc_list = server->gc_next;
 408
 409                 write_seqlock(&net->fs_lock);
 410                 usage = 1;
 411                 deleted = atomic_try_cmpxchg(&server->usage, &usage, 0);
 412                 trace_afs_server(server, usage, afs_server_trace_gc);
 413                 if (deleted) {
 414                         rb_erase(&server->uuid_rb, &net->fs_servers);
 415                         hlist_del_rcu(&server->proc_link);
 416                 }
 417                 write_sequnlock(&net->fs_lock);
 418
 419                 if (deleted) {
 420                         write_seqlock(&net->fs_addr_lock);
 421                         if (!hlist_unhashed(&server->addr4_link))
 422                                 hlist_del_rcu(&server->addr4_link);
 423                         if (!hlist_unhashed(&server->addr6_link))
 424                                 hlist_del_rcu(&server->addr6_link);
 425                         write_sequnlock(&net->fs_addr_lock);
 426                         afs_destroy_server(net, server);
 427                 }
 428         }
 429 }
 430
 431 /*
 432  * Manage the records of servers known to be within a network namespace.  This
 433  * includes garbage collecting unused servers.
 434  *
 435  * Note also that we were given an increment on net->servers_outstanding by
 436  * whoever queued us that we need to deal with before returning.
 437  */
 438 void afs_manage_servers(struct work_struct *work)
 439 {
 440         struct afs_net *net = container_of(work, struct afs_net, fs_manager);
 441         struct afs_server *gc_list = NULL;
 442         struct rb_node *cursor;
 443         time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
 444         bool purging = !net->live;
 445
 446         _enter("");
 447
 448         /* Trawl the server list looking for servers that have expired from
 449          * lack of use.
 450          */
 451         read_seqlock_excl(&net->fs_lock);
 452
 453         for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
 454                 struct afs_server *server =
 455                         rb_entry(cursor, struct afs_server, uuid_rb);
 456                 int usage = atomic_read(&server->usage);
 457
 458                 _debug("manage %pU %u", &server->uuid, usage);
 459
 460                 ASSERTCMP(usage, >=, 1);
 461                 ASSERTIFCMP(purging, usage, ==, 1);
 462
 463                 if (usage == 1) {
 464                         time64_t expire_at = server->put_time;
 465
 466                         if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
 467                             !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
 468                                 expire_at += afs_server_gc_delay;
 469                         if (purging || expire_at <= now) {
 470                                 server->gc_next = gc_list;
 471                                 gc_list = server;
 472                         } else if (expire_at < next_manage) {
 473                                 next_manage = expire_at;
 474                         }
 475                 }
 476         }
 477
 478         read_sequnlock_excl(&net->fs_lock);
 479
 480         /* Update the timer on the way out.  We have to pass an increment on
 481          * servers_outstanding in the namespace that we are in to the timer or
 482          * the work scheduler.
 483          */
 484         if (!purging && next_manage < TIME64_MAX) {
 485                 now = ktime_get_real_seconds();
 486
 487                 if (next_manage - now <= 0) {
 488                         if (queue_work(afs_wq, &net->fs_manager))
 489                                 afs_inc_servers_outstanding(net);
 490                 } else {
 491                         afs_set_server_timer(net, next_manage - now);
 492                 }
 493         }
 494
 495         afs_gc_servers(net, gc_list);
 496
 497         afs_dec_servers_outstanding(net);
 498         _leave(" [%d]", atomic_read(&net->servers_outstanding));
 499 }
 500
 501 static void afs_queue_server_manager(struct afs_net *net)
 502 {
 503         afs_inc_servers_outstanding(net);
 504         if (!queue_work(afs_wq, &net->fs_manager))
 505                 afs_dec_servers_outstanding(net);
 506 }
 507
 508 /*
 509  * Purge list of servers.
 510  */
 511 void afs_purge_servers(struct afs_net *net)
 512 {
 513         _enter("");
 514
 515         if (del_timer_sync(&net->fs_timer))
 516                 atomic_dec(&net->servers_outstanding);
 517
 518         afs_queue_server_manager(net);
 519
 520         _debug("wait");
 521         wait_var_event(&net->servers_outstanding,
 522                        !atomic_read(&net->servers_outstanding));
 523         _leave("");
 524 }
 525
 526 /*
 527  * Get an update for a server's address list.
 528  */
 529 static noinline bool afs_update_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 530 {
 531         struct afs_addr_list *alist, *discard;
 532
 533         _enter("");
 534
 535         trace_afs_server(server, atomic_read(&server->usage), afs_server_trace_update);
 536
 537         alist = afs_vl_lookup_addrs(fc->vnode->volume->cell, fc->key,
 538                                     &server->uuid);
 539         if (IS_ERR(alist)) {
 540                 if ((PTR_ERR(alist) == -ERESTARTSYS ||
 541                      PTR_ERR(alist) == -EINTR) &&
 542                     !(fc->flags & AFS_FS_CURSOR_INTR) &&
 543                     server->addresses) {
 544                         _leave(" = t [intr]");
 545                         return true;
 546                 }
 547                 fc->error = PTR_ERR(alist);
 548                 _leave(" = f [%d]", fc->error);
 549                 return false;
 550         }
 551
 552         discard = alist;
 553         if (server->addr_version != alist->version) {
 554                 write_lock(&server->fs_lock);
 555                 discard = rcu_dereference_protected(server->addresses,
 556                                                     lockdep_is_held(&server->fs_lock));
 557                 rcu_assign_pointer(server->addresses, alist);
 558                 server->addr_version = alist->version;
 559                 write_unlock(&server->fs_lock);
 560         }
 561
 562         server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
 563         afs_put_addrlist(discard);
 564         _leave(" = t");
 565         return true;
 566 }
 567
 568 /*
 569  * See if a server's address list needs updating.
 570  */
 571 bool afs_check_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 572 {
 573         time64_t now = ktime_get_real_seconds();
 574         long diff;
 575         bool success;
 576         int ret, retries = 0;
 577
 578         _enter("");
 579
 580         ASSERT(server);
 581
 582 retry:
 583         diff = READ_ONCE(server->update_at) - now;
 584         if (diff > 0) {
 585                 _leave(" = t [not now %ld]", diff);
 586                 return true;
 587         }
 588
 589         if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
 590                 success = afs_update_server_record(fc, server);
 591                 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
 592                 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
 593                 _leave(" = %d", success);
 594                 return success;
 595         }
 596
 597         ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
 598                           (fc->flags & AFS_FS_CURSOR_INTR) ?
 599                           TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE);
 600         if (ret == -ERESTARTSYS) {
 601                 fc->error = ret;
 602                 _leave(" = f [intr]");
 603                 return false;
 604         }
 605
 606         retries++;
 607         if (retries == 4) {
 608                 _leave(" = f [stale]");
 609                 ret = -ESTALE;
 610                 return false;
 611         }
 612         goto retry;
 613 }