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 atomic_t afs_server_debug_id;
  16
  17 static struct afs_server *afs_maybe_use_server(struct afs_server *,
  18                                                enum afs_server_trace);
  19 static void __afs_put_server(struct afs_net *, struct afs_server *);
  20
  21 /*
  22  * Find a server by one of its addresses.
  23  */
  24 struct afs_server *afs_find_server(struct afs_net *net,
  25                                    const struct sockaddr_rxrpc *srx)
  26 {
  27         const struct afs_addr_list *alist;
  28         struct afs_server *server = NULL;
  29         unsigned int i;
  30         int seq = 0, diff;
  31
  32         rcu_read_lock();
  33
  34         do {
  35                 if (server)
  36                         afs_unuse_server_notime(net, server, afs_server_trace_put_find_rsq);
  37                 server = NULL;
  38                 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
  39
  40                 if (srx->transport.family == AF_INET6) {
  41                         const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
  42                         hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
  43                                 alist = rcu_dereference(server->addresses);
  44                                 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
  45                                         b = &alist->addrs[i].transport.sin6;
  46                                         diff = ((u16 __force)a->sin6_port -
  47                                                 (u16 __force)b->sin6_port);
  48                                         if (diff == 0)
  49                                                 diff = memcmp(&a->sin6_addr,
  50                                                               &b->sin6_addr,
  51                                                               sizeof(struct in6_addr));
  52                                         if (diff == 0)
  53                                                 goto found;
  54                                 }
  55                         }
  56                 } else {
  57                         const struct sockaddr_in *a = &srx->transport.sin, *b;
  58                         hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
  59                                 alist = rcu_dereference(server->addresses);
  60                                 for (i = 0; i < alist->nr_ipv4; i++) {
  61                                         b = &alist->addrs[i].transport.sin;
  62                                         diff = ((u16 __force)a->sin_port -
  63                                                 (u16 __force)b->sin_port);
  64                                         if (diff == 0)
  65                                                 diff = ((u32 __force)a->sin_addr.s_addr -
  66                                                         (u32 __force)b->sin_addr.s_addr);
  67                                         if (diff == 0)
  68                                                 goto found;
  69                                 }
  70                         }
  71                 }
  72
  73                 server = NULL;
  74                 continue;
  75         found:
  76                 server = afs_maybe_use_server(server, afs_server_trace_get_by_addr);
  77
  78         } while (need_seqretry(&net->fs_addr_lock, seq));
  79
  80         done_seqretry(&net->fs_addr_lock, seq);
  81
  82         rcu_read_unlock();
  83         return server;
  84 }
  85
  86 /*
  87  * Look up a server by its UUID and mark it active.
  88  */
  89 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
  90 {
  91         struct afs_server *server = NULL;
  92         struct rb_node *p;
  93         int diff, seq = 0;
  94
  95         _enter("%pU", uuid);
  96
  97         do {
  98                 /* Unfortunately, rbtree walking doesn't give reliable results
  99                  * under just the RCU read lock, so we have to check for
 100                  * changes.
 101                  */
 102                 if (server)
 103                         afs_unuse_server(net, server, afs_server_trace_put_uuid_rsq);
 104                 server = NULL;
 105
 106                 read_seqbegin_or_lock(&net->fs_lock, &seq);
 107
 108                 p = net->fs_servers.rb_node;
 109                 while (p) {
 110                         server = rb_entry(p, struct afs_server, uuid_rb);
 111
 112                         diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
 113                         if (diff < 0) {
 114                                 p = p->rb_left;
 115                         } else if (diff > 0) {
 116                                 p = p->rb_right;
 117                         } else {
 118                                 afs_use_server(server, afs_server_trace_get_by_uuid);
 119                                 break;
 120                         }
 121
 122                         server = NULL;
 123                 }
 124         } while (need_seqretry(&net->fs_lock, seq));
 125
 126         done_seqretry(&net->fs_lock, seq);
 127
 128         _leave(" = %p", server);
 129         return server;
 130 }
 131
 132 /*
 133  * Install a server record in the namespace tree
 134  */
 135 static struct afs_server *afs_install_server(struct afs_net *net,
 136                                              struct afs_server *candidate)
 137 {
 138         const struct afs_addr_list *alist;
 139         struct afs_server *server;
 140         struct rb_node **pp, *p;
 141         int diff;
 142
 143         _enter("%p", candidate);
 144
 145         write_seqlock(&net->fs_lock);
 146
 147         /* Firstly install the server in the UUID lookup tree */
 148         pp = &net->fs_servers.rb_node;
 149         p = NULL;
 150         while (*pp) {
 151                 p = *pp;
 152                 _debug("- consider %p", p);
 153                 server = rb_entry(p, struct afs_server, uuid_rb);
 154                 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
 155                 if (diff < 0)
 156                         pp = &(*pp)->rb_left;
 157                 else if (diff > 0)
 158                         pp = &(*pp)->rb_right;
 159                 else
 160                         goto exists;
 161         }
 162
 163         server = candidate;
 164         rb_link_node(&server->uuid_rb, p, pp);
 165         rb_insert_color(&server->uuid_rb, &net->fs_servers);
 166         hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
 167
 168         write_seqlock(&net->fs_addr_lock);
 169         alist = rcu_dereference_protected(server->addresses,
 170                                           lockdep_is_held(&net->fs_addr_lock.lock));
 171
 172         /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
 173          * it in the IPv4 and/or IPv6 reverse-map lists.
 174          *
 175          * TODO: For speed we want to use something other than a flat list
 176          * here; even sorting the list in terms of lowest address would help a
 177          * bit, but anything we might want to do gets messy and memory
 178          * intensive.
 179          */
 180         if (alist->nr_ipv4 > 0)
 181                 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
 182         if (alist->nr_addrs > alist->nr_ipv4)
 183                 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
 184
 185         write_sequnlock(&net->fs_addr_lock);
 186
 187 exists:
 188         afs_get_server(server, afs_server_trace_get_install);
 189         write_sequnlock(&net->fs_lock);
 190         return server;
 191 }
 192
 193 /*
 194  * Allocate a new server record and mark it active.
 195  */
 196 static struct afs_server *afs_alloc_server(struct afs_net *net,
 197                                            const uuid_t *uuid,
 198                                            struct afs_addr_list *alist)
 199 {
 200         struct afs_server *server;
 201
 202         _enter("");
 203
 204         server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
 205         if (!server)
 206                 goto enomem;
 207
 208         atomic_set(&server->ref, 1);
 209         atomic_set(&server->active, 1);
 210         server->debug_id = atomic_inc_return(&afs_server_debug_id);
 211         RCU_INIT_POINTER(server->addresses, alist);
 212         server->addr_version = alist->version;
 213         server->uuid = *uuid;
 214         rwlock_init(&server->fs_lock);
 215         server->cb_volumes = RB_ROOT;
 216         seqlock_init(&server->cb_break_lock);
 217         init_waitqueue_head(&server->probe_wq);
 218         INIT_LIST_HEAD(&server->probe_link);
 219         spin_lock_init(&server->probe_lock);
 220
 221         afs_inc_servers_outstanding(net);
 222         trace_afs_server(server, 1, 1, afs_server_trace_alloc);
 223         _leave(" = %p", server);
 224         return server;
 225
 226 enomem:
 227         _leave(" = NULL [nomem]");
 228         return NULL;
 229 }
 230
 231 /*
 232  * Look up an address record for a server
 233  */
 234 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
 235                                                  struct key *key, const uuid_t *uuid)
 236 {
 237         struct afs_vl_cursor vc;
 238         struct afs_addr_list *alist = NULL;
 239         int ret;
 240
 241         ret = -ERESTARTSYS;
 242         if (afs_begin_vlserver_operation(&vc, cell, key)) {
 243                 while (afs_select_vlserver(&vc)) {
 244                         if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
 245                                 alist = afs_yfsvl_get_endpoints(&vc, uuid);
 246                         else
 247                                 alist = afs_vl_get_addrs_u(&vc, uuid);
 248                 }
 249
 250                 ret = afs_end_vlserver_operation(&vc);
 251         }
 252
 253         return ret < 0 ? ERR_PTR(ret) : alist;
 254 }
 255
 256 /*
 257  * Get or create a fileserver record.
 258  */
 259 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
 260                                      const uuid_t *uuid, u32 addr_version)
 261 {
 262         struct afs_addr_list *alist;
 263         struct afs_server *server, *candidate;
 264
 265         _enter("%p,%pU", cell->net, uuid);
 266
 267         server = afs_find_server_by_uuid(cell->net, uuid);
 268         if (server) {
 269                 if (server->addr_version != addr_version)
 270                         set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
 271                 return server;
 272         }
 273
 274         alist = afs_vl_lookup_addrs(cell, key, uuid);
 275         if (IS_ERR(alist))
 276                 return ERR_CAST(alist);
 277
 278         candidate = afs_alloc_server(cell->net, uuid, alist);
 279         if (!candidate) {
 280                 afs_put_addrlist(alist);
 281                 return ERR_PTR(-ENOMEM);
 282         }
 283
 284         server = afs_install_server(cell->net, candidate);
 285         if (server != candidate) {
 286                 afs_put_addrlist(alist);
 287                 kfree(candidate);
 288         } else {
 289                 /* Immediately dispatch an asynchronous probe to each interface
 290                  * on the fileserver.  This will make sure the repeat-probing
 291                  * service is started.
 292                  */
 293                 afs_fs_probe_fileserver(cell->net, server, key, true);
 294         }
 295
 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->ref);
 332
 333         trace_afs_server(server, u, atomic_read(&server->active), reason);
 334         return server;
 335 }
 336
 337 /*
 338  * Try to get a reference on a server object.
 339  */
 340 static struct afs_server *afs_maybe_use_server(struct afs_server *server,
 341                                                enum afs_server_trace reason)
 342 {
 343         unsigned int r = atomic_fetch_add_unless(&server->ref, 1, 0);
 344         unsigned int a;
 345
 346         if (r == 0)
 347                 return NULL;
 348
 349         a = atomic_inc_return(&server->active);
 350         trace_afs_server(server, r, a, reason);
 351         return server;
 352 }
 353
 354 /*
 355  * Get an active count on a server object.
 356  */
 357 struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
 358 {
 359         unsigned int r = atomic_inc_return(&server->ref);
 360         unsigned int a = atomic_inc_return(&server->active);
 361
 362         trace_afs_server(server, r, a, reason);
 363         return server;
 364 }
 365
 366 /*
 367  * Release a reference on a server record.
 368  */
 369 void afs_put_server(struct afs_net *net, struct afs_server *server,
 370                     enum afs_server_trace reason)
 371 {
 372         unsigned int usage;
 373
 374         if (!server)
 375                 return;
 376
 377         usage = atomic_dec_return(&server->ref);
 378         trace_afs_server(server, usage, atomic_read(&server->active), reason);
 379         if (unlikely(usage == 0))
 380                 __afs_put_server(net, server);
 381 }
 382
 383 /*
 384  * Drop an active count on a server object without updating the last-unused
 385  * time.
 386  */
 387 void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
 388                              enum afs_server_trace reason)
 389 {
 390         if (server) {
 391                 unsigned int active = atomic_dec_return(&server->active);
 392
 393                 if (active == 0)
 394                         afs_set_server_timer(net, afs_server_gc_delay);
 395                 afs_put_server(net, server, reason);
 396         }
 397 }
 398
 399 /*
 400  * Drop an active count on a server object.
 401  */
 402 void afs_unuse_server(struct afs_net *net, struct afs_server *server,
 403                       enum afs_server_trace reason)
 404 {
 405         if (server) {
 406                 server->unuse_time = ktime_get_real_seconds();
 407                 afs_unuse_server_notime(net, server, reason);
 408         }
 409 }
 410
 411 static void afs_server_rcu(struct rcu_head *rcu)
 412 {
 413         struct afs_server *server = container_of(rcu, struct afs_server, rcu);
 414
 415         trace_afs_server(server, atomic_read(&server->ref),
 416                          atomic_read(&server->active), afs_server_trace_free);
 417         afs_put_addrlist(rcu_access_pointer(server->addresses));
 418         kfree(server);
 419 }
 420
 421 static void __afs_put_server(struct afs_net *net, struct afs_server *server)
 422 {
 423         call_rcu(&server->rcu, afs_server_rcu);
 424         afs_dec_servers_outstanding(net);
 425 }
 426
 427 /*
 428  * destroy a dead server
 429  */
 430 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
 431 {
 432         struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
 433         struct afs_addr_cursor ac = {
 434                 .alist  = alist,
 435                 .index  = alist->preferred,
 436                 .error  = 0,
 437         };
 438
 439         if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
 440                 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
 441
 442         afs_put_server(net, server, afs_server_trace_destroy);
 443 }
 444
 445 /*
 446  * Garbage collect any expired servers.
 447  */
 448 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
 449 {
 450         struct afs_server *server;
 451         int active;
 452
 453         while ((server = gc_list)) {
 454                 gc_list = server->gc_next;
 455
 456                 write_seqlock(&net->fs_lock);
 457
 458                 active = atomic_read(&server->active);
 459                 if (active == 0) {
 460                         trace_afs_server(server, atomic_read(&server->ref),
 461                                          active, afs_server_trace_gc);
 462                         rb_erase(&server->uuid_rb, &net->fs_servers);
 463                         list_del(&server->probe_link);
 464                         hlist_del_rcu(&server->proc_link);
 465                         if (!hlist_unhashed(&server->addr4_link))
 466                                 hlist_del_rcu(&server->addr4_link);
 467                         if (!hlist_unhashed(&server->addr6_link))
 468                                 hlist_del_rcu(&server->addr6_link);
 469                 }
 470                 write_sequnlock(&net->fs_lock);
 471
 472                 if (active == 0)
 473                         afs_destroy_server(net, server);
 474         }
 475 }
 476
 477 /*
 478  * Manage the records of servers known to be within a network namespace.  This
 479  * includes garbage collecting unused servers.
 480  *
 481  * Note also that we were given an increment on net->servers_outstanding by
 482  * whoever queued us that we need to deal with before returning.
 483  */
 484 void afs_manage_servers(struct work_struct *work)
 485 {
 486         struct afs_net *net = container_of(work, struct afs_net, fs_manager);
 487         struct afs_server *gc_list = NULL;
 488         struct rb_node *cursor;
 489         time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
 490         bool purging = !net->live;
 491
 492         _enter("");
 493
 494         /* Trawl the server list looking for servers that have expired from
 495          * lack of use.
 496          */
 497         read_seqlock_excl(&net->fs_lock);
 498
 499         for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
 500                 struct afs_server *server =
 501                         rb_entry(cursor, struct afs_server, uuid_rb);
 502                 int active = atomic_read(&server->active);
 503
 504                 _debug("manage %pU %u", &server->uuid, active);
 505
 506                 ASSERTIFCMP(purging, active, ==, 0);
 507
 508                 if (active == 0) {
 509                         time64_t expire_at = server->unuse_time;
 510
 511                         if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
 512                             !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
 513                                 expire_at += afs_server_gc_delay;
 514                         if (purging || expire_at <= now) {
 515                                 server->gc_next = gc_list;
 516                                 gc_list = server;
 517                         } else if (expire_at < next_manage) {
 518                                 next_manage = expire_at;
 519                         }
 520                 }
 521         }
 522
 523         read_sequnlock_excl(&net->fs_lock);
 524
 525         /* Update the timer on the way out.  We have to pass an increment on
 526          * servers_outstanding in the namespace that we are in to the timer or
 527          * the work scheduler.
 528          */
 529         if (!purging && next_manage < TIME64_MAX) {
 530                 now = ktime_get_real_seconds();
 531
 532                 if (next_manage - now <= 0) {
 533                         if (queue_work(afs_wq, &net->fs_manager))
 534                                 afs_inc_servers_outstanding(net);
 535                 } else {
 536                         afs_set_server_timer(net, next_manage - now);
 537                 }
 538         }
 539
 540         afs_gc_servers(net, gc_list);
 541
 542         afs_dec_servers_outstanding(net);
 543         _leave(" [%d]", atomic_read(&net->servers_outstanding));
 544 }
 545
 546 static void afs_queue_server_manager(struct afs_net *net)
 547 {
 548         afs_inc_servers_outstanding(net);
 549         if (!queue_work(afs_wq, &net->fs_manager))
 550                 afs_dec_servers_outstanding(net);
 551 }
 552
 553 /*
 554  * Purge list of servers.
 555  */
 556 void afs_purge_servers(struct afs_net *net)
 557 {
 558         _enter("");
 559
 560         if (del_timer_sync(&net->fs_timer))
 561                 atomic_dec(&net->servers_outstanding);
 562
 563         afs_queue_server_manager(net);
 564
 565         _debug("wait");
 566         wait_var_event(&net->servers_outstanding,
 567                        !atomic_read(&net->servers_outstanding));
 568         _leave("");
 569 }
 570
 571 /*
 572  * Get an update for a server's address list.
 573  */
 574 static noinline bool afs_update_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 575 {
 576         struct afs_addr_list *alist, *discard;
 577
 578         _enter("");
 579
 580         trace_afs_server(server, atomic_read(&server->ref), atomic_read(&server->active),
 581                          afs_server_trace_update);
 582
 583         alist = afs_vl_lookup_addrs(fc->vnode->volume->cell, fc->key,
 584                                     &server->uuid);
 585         if (IS_ERR(alist)) {
 586                 if ((PTR_ERR(alist) == -ERESTARTSYS ||
 587                      PTR_ERR(alist) == -EINTR) &&
 588                     !(fc->flags & AFS_FS_CURSOR_INTR) &&
 589                     server->addresses) {
 590                         _leave(" = t [intr]");
 591                         return true;
 592                 }
 593                 fc->error = PTR_ERR(alist);
 594                 _leave(" = f [%d]", fc->error);
 595                 return false;
 596         }
 597
 598         discard = alist;
 599         if (server->addr_version != alist->version) {
 600                 write_lock(&server->fs_lock);
 601                 discard = rcu_dereference_protected(server->addresses,
 602                                                     lockdep_is_held(&server->fs_lock));
 603                 rcu_assign_pointer(server->addresses, alist);
 604                 server->addr_version = alist->version;
 605                 write_unlock(&server->fs_lock);
 606         }
 607
 608         afs_put_addrlist(discard);
 609         _leave(" = t");
 610         return true;
 611 }
 612
 613 /*
 614  * See if a server's address list needs updating.
 615  */
 616 bool afs_check_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 617 {
 618         bool success;
 619         int ret, retries = 0;
 620
 621         _enter("");
 622
 623         ASSERT(server);
 624
 625 retry:
 626         if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
 627                 goto wait;
 628         if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
 629                 goto update;
 630         _leave(" = t [good]");
 631         return true;
 632
 633 update:
 634         if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
 635                 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
 636                 success = afs_update_server_record(fc, server);
 637                 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
 638                 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
 639                 _leave(" = %d", success);
 640                 return success;
 641         }
 642
 643 wait:
 644         ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
 645                           (fc->flags & AFS_FS_CURSOR_INTR) ?
 646                           TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE);
 647         if (ret == -ERESTARTSYS) {
 648                 fc->error = ret;
 649                 _leave(" = f [intr]");
 650                 return false;
 651         }
 652
 653         retries++;
 654         if (retries == 4) {
 655                 _leave(" = f [stale]");
 656                 ret = -ESTALE;
 657                 return false;
 658         }
 659         goto retry;
 660 }