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
  17 static void afs_inc_servers_outstanding(struct afs_net *net)
  18 {
  19         atomic_inc(&net->servers_outstanding);
  20 }
  21
  22 static void afs_dec_servers_outstanding(struct afs_net *net)
  23 {
  24         if (atomic_dec_and_test(&net->servers_outstanding))
  25                 wake_up_var(&net->servers_outstanding);
  26 }
  27
  28 /*
  29  * Find a server by one of its addresses.
  30  */
  31 struct afs_server *afs_find_server(struct afs_net *net,
  32                                    const struct sockaddr_rxrpc *srx)
  33 {
  34         const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
  35         const struct afs_addr_list *alist;
  36         struct afs_server *server = NULL;
  37         unsigned int i;
  38         bool ipv6 = true;
  39         int seq = 0, diff;
  40
  41         if (srx->transport.sin6.sin6_addr.s6_addr32[0] == 0 ||
  42             srx->transport.sin6.sin6_addr.s6_addr32[1] == 0 ||
  43             srx->transport.sin6.sin6_addr.s6_addr32[2] == htonl(0xffff))
  44                 ipv6 = false;
  45
  46         rcu_read_lock();
  47
  48         do {
  49                 if (server)
  50                         afs_put_server(net, server);
  51                 server = NULL;
  52                 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
  53
  54                 if (ipv6) {
  55                         hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
  56                                 alist = rcu_dereference(server->addresses);
  57                                 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
  58                                         b = &alist->addrs[i].transport.sin6;
  59                                         diff = ((u16 __force)a->sin6_port -
  60                                                 (u16 __force)b->sin6_port);
  61                                         if (diff == 0)
  62                                                 diff = memcmp(&a->sin6_addr,
  63                                                               &b->sin6_addr,
  64                                                               sizeof(struct in6_addr));
  65                                         if (diff == 0)
  66                                                 goto found;
  67                                 }
  68                         }
  69                 } else {
  70                         hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
  71                                 alist = rcu_dereference(server->addresses);
  72                                 for (i = 0; i < alist->nr_ipv4; i++) {
  73                                         b = &alist->addrs[i].transport.sin6;
  74                                         diff = ((u16 __force)a->sin6_port -
  75                                                 (u16 __force)b->sin6_port);
  76                                         if (diff == 0)
  77                                                 diff = ((u32 __force)a->sin6_addr.s6_addr32[3] -
  78                                                         (u32 __force)b->sin6_addr.s6_addr32[3]);
  79                                         if (diff == 0)
  80                                                 goto found;
  81                                 }
  82                         }
  83                 }
  84
  85                 server = NULL;
  86         found:
  87                 if (server && !atomic_inc_not_zero(&server->usage))
  88                         server = NULL;
  89
  90         } while (need_seqretry(&net->fs_addr_lock, seq));
  91
  92         done_seqretry(&net->fs_addr_lock, seq);
  93
  94         rcu_read_unlock();
  95         return server;
  96 }
  97
  98 /*
  99  * Look up a server by its UUID
 100  */
 101 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
 102 {
 103         struct afs_server *server = NULL;
 104         struct rb_node *p;
 105         int diff, seq = 0;
 106
 107         _enter("%pU", uuid);
 108
 109         do {
 110                 /* Unfortunately, rbtree walking doesn't give reliable results
 111                  * under just the RCU read lock, so we have to check for
 112                  * changes.
 113                  */
 114                 if (server)
 115                         afs_put_server(net, server);
 116                 server = NULL;
 117
 118                 read_seqbegin_or_lock(&net->fs_lock, &seq);
 119
 120                 p = net->fs_servers.rb_node;
 121                 while (p) {
 122                         server = rb_entry(p, struct afs_server, uuid_rb);
 123
 124                         diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
 125                         if (diff < 0) {
 126                                 p = p->rb_left;
 127                         } else if (diff > 0) {
 128                                 p = p->rb_right;
 129                         } else {
 130                                 afs_get_server(server);
 131                                 break;
 132                         }
 133
 134                         server = NULL;
 135                 }
 136         } while (need_seqretry(&net->fs_lock, seq));
 137
 138         done_seqretry(&net->fs_lock, seq);
 139
 140         _leave(" = %p", server);
 141         return server;
 142 }
 143
 144 /*
 145  * Install a server record in the namespace tree
 146  */
 147 static struct afs_server *afs_install_server(struct afs_net *net,
 148                                              struct afs_server *candidate)
 149 {
 150         const struct afs_addr_list *alist;
 151         struct afs_server *server;
 152         struct rb_node **pp, *p;
 153         int ret = -EEXIST, diff;
 154
 155         _enter("%p", candidate);
 156
 157         write_seqlock(&net->fs_lock);
 158
 159         /* Firstly install the server in the UUID lookup tree */
 160         pp = &net->fs_servers.rb_node;
 161         p = NULL;
 162         while (*pp) {
 163                 p = *pp;
 164                 _debug("- consider %p", p);
 165                 server = rb_entry(p, struct afs_server, uuid_rb);
 166                 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
 167                 if (diff < 0)
 168                         pp = &(*pp)->rb_left;
 169                 else if (diff > 0)
 170                         pp = &(*pp)->rb_right;
 171                 else
 172                         goto exists;
 173         }
 174
 175         server = candidate;
 176         rb_link_node(&server->uuid_rb, p, pp);
 177         rb_insert_color(&server->uuid_rb, &net->fs_servers);
 178         hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
 179
 180         write_seqlock(&net->fs_addr_lock);
 181         alist = rcu_dereference_protected(server->addresses,
 182                                           lockdep_is_held(&net->fs_addr_lock.lock));
 183
 184         /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
 185          * it in the IPv4 and/or IPv6 reverse-map lists.
 186          *
 187          * TODO: For speed we want to use something other than a flat list
 188          * here; even sorting the list in terms of lowest address would help a
 189          * bit, but anything we might want to do gets messy and memory
 190          * intensive.
 191          */
 192         if (alist->nr_ipv4 > 0)
 193                 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
 194         if (alist->nr_addrs > alist->nr_ipv4)
 195                 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
 196
 197         write_sequnlock(&net->fs_addr_lock);
 198         ret = 0;
 199
 200 exists:
 201         afs_get_server(server);
 202         write_sequnlock(&net->fs_lock);
 203         return server;
 204 }
 205
 206 /*
 207  * allocate a new server record
 208  */
 209 static struct afs_server *afs_alloc_server(struct afs_net *net,
 210                                            const uuid_t *uuid,
 211                                            struct afs_addr_list *alist)
 212 {
 213         struct afs_server *server;
 214
 215         _enter("");
 216
 217         server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
 218         if (!server)
 219                 goto enomem;
 220
 221         atomic_set(&server->usage, 1);
 222         RCU_INIT_POINTER(server->addresses, alist);
 223         server->addr_version = alist->version;
 224         server->uuid = *uuid;
 225         server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
 226         rwlock_init(&server->fs_lock);
 227         INIT_HLIST_HEAD(&server->cb_volumes);
 228         rwlock_init(&server->cb_break_lock);
 229         init_waitqueue_head(&server->probe_wq);
 230         spin_lock_init(&server->probe_lock);
 231
 232         afs_inc_servers_outstanding(net);
 233         _leave(" = %p", server);
 234         return server;
 235
 236 enomem:
 237         _leave(" = NULL [nomem]");
 238         return NULL;
 239 }
 240
 241 /*
 242  * Look up an address record for a server
 243  */
 244 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
 245                                                  struct key *key, const uuid_t *uuid)
 246 {
 247         struct afs_vl_cursor vc;
 248         struct afs_addr_list *alist = NULL;
 249         int ret;
 250
 251         ret = -ERESTARTSYS;
 252         if (afs_begin_vlserver_operation(&vc, cell, key)) {
 253                 while (afs_select_vlserver(&vc)) {
 254                         if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
 255                                 alist = afs_yfsvl_get_endpoints(&vc, uuid);
 256                         else
 257                                 alist = afs_vl_get_addrs_u(&vc, uuid);
 258                 }
 259
 260                 ret = afs_end_vlserver_operation(&vc);
 261         }
 262
 263         return ret < 0 ? ERR_PTR(ret) : alist;
 264 }
 265
 266 /*
 267  * Get or create a fileserver record.
 268  */
 269 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
 270                                      const uuid_t *uuid)
 271 {
 272         struct afs_addr_list *alist;
 273         struct afs_server *server, *candidate;
 274
 275         _enter("%p,%pU", cell->net, uuid);
 276
 277         server = afs_find_server_by_uuid(cell->net, uuid);
 278         if (server)
 279                 return server;
 280
 281         alist = afs_vl_lookup_addrs(cell, key, uuid);
 282         if (IS_ERR(alist))
 283                 return ERR_CAST(alist);
 284
 285         candidate = afs_alloc_server(cell->net, uuid, alist);
 286         if (!candidate) {
 287                 afs_put_addrlist(alist);
 288                 return ERR_PTR(-ENOMEM);
 289         }
 290
 291         server = afs_install_server(cell->net, candidate);
 292         if (server != candidate) {
 293                 afs_put_addrlist(alist);
 294                 kfree(candidate);
 295         }
 296
 297         _leave(" = %p{%d}", server, atomic_read(&server->usage));
 298         return server;
 299 }
 300
 301 /*
 302  * Set the server timer to fire after a given delay, assuming it's not already
 303  * set for an earlier time.
 304  */
 305 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
 306 {
 307         if (net->live) {
 308                 afs_inc_servers_outstanding(net);
 309                 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
 310                         afs_dec_servers_outstanding(net);
 311         }
 312 }
 313
 314 /*
 315  * Server management timer.  We have an increment on fs_outstanding that we
 316  * need to pass along to the work item.
 317  */
 318 void afs_servers_timer(struct timer_list *timer)
 319 {
 320         struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
 321
 322         _enter("");
 323         if (!queue_work(afs_wq, &net->fs_manager))
 324                 afs_dec_servers_outstanding(net);
 325 }
 326
 327 /*
 328  * Release a reference on a server record.
 329  */
 330 void afs_put_server(struct afs_net *net, struct afs_server *server)
 331 {
 332         unsigned int usage;
 333
 334         if (!server)
 335                 return;
 336
 337         server->put_time = ktime_get_real_seconds();
 338
 339         usage = atomic_dec_return(&server->usage);
 340
 341         _enter("{%u}", usage);
 342
 343         if (likely(usage > 0))
 344                 return;
 345
 346         afs_set_server_timer(net, afs_server_gc_delay);
 347 }
 348
 349 static void afs_server_rcu(struct rcu_head *rcu)
 350 {
 351         struct afs_server *server = container_of(rcu, struct afs_server, rcu);
 352
 353         afs_put_addrlist(rcu_access_pointer(server->addresses));
 354         kfree(server);
 355 }
 356
 357 /*
 358  * destroy a dead server
 359  */
 360 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
 361 {
 362         struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
 363         struct afs_addr_cursor ac = {
 364                 .alist  = alist,
 365                 .index  = alist->preferred,
 366                 .error  = 0,
 367         };
 368         _enter("%p", server);
 369
 370         if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
 371                 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
 372
 373         wait_var_event(&server->probe_outstanding,
 374                        atomic_read(&server->probe_outstanding) == 0);
 375
 376         call_rcu(&server->rcu, afs_server_rcu);
 377         afs_dec_servers_outstanding(net);
 378 }
 379
 380 /*
 381  * Garbage collect any expired servers.
 382  */
 383 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
 384 {
 385         struct afs_server *server;
 386         bool deleted;
 387         int usage;
 388
 389         while ((server = gc_list)) {
 390                 gc_list = server->gc_next;
 391
 392                 write_seqlock(&net->fs_lock);
 393                 usage = 1;
 394                 deleted = atomic_try_cmpxchg(&server->usage, &usage, 0);
 395                 if (deleted) {
 396                         rb_erase(&server->uuid_rb, &net->fs_servers);
 397                         hlist_del_rcu(&server->proc_link);
 398                 }
 399                 write_sequnlock(&net->fs_lock);
 400
 401                 if (deleted) {
 402                         write_seqlock(&net->fs_addr_lock);
 403                         if (!hlist_unhashed(&server->addr4_link))
 404                                 hlist_del_rcu(&server->addr4_link);
 405                         if (!hlist_unhashed(&server->addr6_link))
 406                                 hlist_del_rcu(&server->addr6_link);
 407                         write_sequnlock(&net->fs_addr_lock);
 408                         afs_destroy_server(net, server);
 409                 }
 410         }
 411 }
 412
 413 /*
 414  * Manage the records of servers known to be within a network namespace.  This
 415  * includes garbage collecting unused servers.
 416  *
 417  * Note also that we were given an increment on net->servers_outstanding by
 418  * whoever queued us that we need to deal with before returning.
 419  */
 420 void afs_manage_servers(struct work_struct *work)
 421 {
 422         struct afs_net *net = container_of(work, struct afs_net, fs_manager);
 423         struct afs_server *gc_list = NULL;
 424         struct rb_node *cursor;
 425         time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
 426         bool purging = !net->live;
 427
 428         _enter("");
 429
 430         /* Trawl the server list looking for servers that have expired from
 431          * lack of use.
 432          */
 433         read_seqlock_excl(&net->fs_lock);
 434
 435         for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
 436                 struct afs_server *server =
 437                         rb_entry(cursor, struct afs_server, uuid_rb);
 438                 int usage = atomic_read(&server->usage);
 439
 440                 _debug("manage %pU %u", &server->uuid, usage);
 441
 442                 ASSERTCMP(usage, >=, 1);
 443                 ASSERTIFCMP(purging, usage, ==, 1);
 444
 445                 if (usage == 1) {
 446                         time64_t expire_at = server->put_time;
 447
 448                         if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
 449                             !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
 450                                 expire_at += afs_server_gc_delay;
 451                         if (purging || expire_at <= now) {
 452                                 server->gc_next = gc_list;
 453                                 gc_list = server;
 454                         } else if (expire_at < next_manage) {
 455                                 next_manage = expire_at;
 456                         }
 457                 }
 458         }
 459
 460         read_sequnlock_excl(&net->fs_lock);
 461
 462         /* Update the timer on the way out.  We have to pass an increment on
 463          * servers_outstanding in the namespace that we are in to the timer or
 464          * the work scheduler.
 465          */
 466         if (!purging && next_manage < TIME64_MAX) {
 467                 now = ktime_get_real_seconds();
 468
 469                 if (next_manage - now <= 0) {
 470                         if (queue_work(afs_wq, &net->fs_manager))
 471                                 afs_inc_servers_outstanding(net);
 472                 } else {
 473                         afs_set_server_timer(net, next_manage - now);
 474                 }
 475         }
 476
 477         afs_gc_servers(net, gc_list);
 478
 479         afs_dec_servers_outstanding(net);
 480         _leave(" [%d]", atomic_read(&net->servers_outstanding));
 481 }
 482
 483 static void afs_queue_server_manager(struct afs_net *net)
 484 {
 485         afs_inc_servers_outstanding(net);
 486         if (!queue_work(afs_wq, &net->fs_manager))
 487                 afs_dec_servers_outstanding(net);
 488 }
 489
 490 /*
 491  * Purge list of servers.
 492  */
 493 void afs_purge_servers(struct afs_net *net)
 494 {
 495         _enter("");
 496
 497         if (del_timer_sync(&net->fs_timer))
 498                 atomic_dec(&net->servers_outstanding);
 499
 500         afs_queue_server_manager(net);
 501
 502         _debug("wait");
 503         wait_var_event(&net->servers_outstanding,
 504                        !atomic_read(&net->servers_outstanding));
 505         _leave("");
 506 }
 507
 508 /*
 509  * Get an update for a server's address list.
 510  */
 511 static noinline bool afs_update_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 512 {
 513         struct afs_addr_list *alist, *discard;
 514
 515         _enter("");
 516
 517         alist = afs_vl_lookup_addrs(fc->vnode->volume->cell, fc->key,
 518                                     &server->uuid);
 519         if (IS_ERR(alist)) {
 520                 if ((PTR_ERR(alist) == -ERESTARTSYS ||
 521                      PTR_ERR(alist) == -EINTR) &&
 522                     !(fc->flags & AFS_FS_CURSOR_INTR) &&
 523                     server->addresses) {
 524                         _leave(" = t [intr]");
 525                         return true;
 526                 }
 527                 fc->error = PTR_ERR(alist);
 528                 _leave(" = f [%d]", fc->error);
 529                 return false;
 530         }
 531
 532         discard = alist;
 533         if (server->addr_version != alist->version) {
 534                 write_lock(&server->fs_lock);
 535                 discard = rcu_dereference_protected(server->addresses,
 536                                                     lockdep_is_held(&server->fs_lock));
 537                 rcu_assign_pointer(server->addresses, alist);
 538                 server->addr_version = alist->version;
 539                 write_unlock(&server->fs_lock);
 540         }
 541
 542         server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
 543         afs_put_addrlist(discard);
 544         _leave(" = t");
 545         return true;
 546 }
 547
 548 /*
 549  * See if a server's address list needs updating.
 550  */
 551 bool afs_check_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 552 {
 553         time64_t now = ktime_get_real_seconds();
 554         long diff;
 555         bool success;
 556         int ret, retries = 0;
 557
 558         _enter("");
 559
 560         ASSERT(server);
 561
 562 retry:
 563         diff = READ_ONCE(server->update_at) - now;
 564         if (diff > 0) {
 565                 _leave(" = t [not now %ld]", diff);
 566                 return true;
 567         }
 568
 569         if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
 570                 success = afs_update_server_record(fc, server);
 571                 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
 572                 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
 573                 _leave(" = %d", success);
 574                 return success;
 575         }
 576
 577         ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
 578                           TASK_INTERRUPTIBLE);
 579         if (ret == -ERESTARTSYS) {
 580                 if (!(fc->flags & AFS_FS_CURSOR_INTR) && server->addresses) {
 581                         _leave(" = t [intr]");
 582                         return true;
 583                 }
 584                 fc->error = ret;
 585                 _leave(" = f [intr]");
 586                 return false;
 587         }
 588
 589         retries++;
 590         if (retries == 4) {
 591                 _leave(" = f [stale]");
 592                 ret = -ESTALE;
 593                 return false;
 594         }
 595         goto retry;
 596 }