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 sockaddr_in6 *a = &srx->transport.sin6, *b;
  36         const struct afs_addr_list *alist;
  37         struct afs_server *server = NULL;
  38         unsigned int i;
  39         bool ipv6 = true;
  40         int seq = 0, diff;
  41
  42         if (srx->transport.sin6.sin6_addr.s6_addr32[0] == 0 ||
  43             srx->transport.sin6.sin6_addr.s6_addr32[1] == 0 ||
  44             srx->transport.sin6.sin6_addr.s6_addr32[2] == htonl(0xffff))
  45                 ipv6 = false;
  46
  47         rcu_read_lock();
  48
  49         do {
  50                 if (server)
  51                         afs_put_server(net, server, afs_server_trace_put_find_rsq);
  52                 server = NULL;
  53                 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
  54
  55                 if (ipv6) {
  56                         hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
  57                                 alist = rcu_dereference(server->addresses);
  58                                 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
  59                                         b = &alist->addrs[i].transport.sin6;
  60                                         diff = ((u16 __force)a->sin6_port -
  61                                                 (u16 __force)b->sin6_port);
  62                                         if (diff == 0)
  63                                                 diff = memcmp(&a->sin6_addr,
  64                                                               &b->sin6_addr,
  65                                                               sizeof(struct in6_addr));
  66                                         if (diff == 0)
  67                                                 goto found;
  68                                 }
  69                         }
  70                 } else {
  71                         hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
  72                                 alist = rcu_dereference(server->addresses);
  73                                 for (i = 0; i < alist->nr_ipv4; i++) {
  74                                         b = &alist->addrs[i].transport.sin6;
  75                                         diff = ((u16 __force)a->sin6_port -
  76                                                 (u16 __force)b->sin6_port);
  77                                         if (diff == 0)
  78                                                 diff = ((u32 __force)a->sin6_addr.s6_addr32[3] -
  79                                                         (u32 __force)b->sin6_addr.s6_addr32[3]);
  80                                         if (diff == 0)
  81                                                 goto found;
  82                                 }
  83                         }
  84                 }
  85
  86                 server = NULL;
  87         found:
  88                 if (server && !atomic_inc_not_zero(&server->usage))
  89                         server = NULL;
  90
  91         } while (need_seqretry(&net->fs_addr_lock, seq));
  92
  93         done_seqretry(&net->fs_addr_lock, seq);
  94
  95         rcu_read_unlock();
  96         return server;
  97 }
  98
  99 /*
 100  * Look up a server by its UUID
 101  */
 102 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
 103 {
 104         struct afs_server *server = NULL;
 105         struct rb_node *p;
 106         int diff, seq = 0;
 107
 108         _enter("%pU", uuid);
 109
 110         do {
 111                 /* Unfortunately, rbtree walking doesn't give reliable results
 112                  * under just the RCU read lock, so we have to check for
 113                  * changes.
 114                  */
 115                 if (server)
 116                         afs_put_server(net, server, afs_server_trace_put_uuid_rsq);
 117                 server = NULL;
 118
 119                 read_seqbegin_or_lock(&net->fs_lock, &seq);
 120
 121                 p = net->fs_servers.rb_node;
 122                 while (p) {
 123                         server = rb_entry(p, struct afs_server, uuid_rb);
 124
 125                         diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
 126                         if (diff < 0) {
 127                                 p = p->rb_left;
 128                         } else if (diff > 0) {
 129                                 p = p->rb_right;
 130                         } else {
 131                                 afs_get_server(server, afs_server_trace_get_by_uuid);
 132                                 break;
 133                         }
 134
 135                         server = NULL;
 136                 }
 137         } while (need_seqretry(&net->fs_lock, seq));
 138
 139         done_seqretry(&net->fs_lock, seq);
 140
 141         _leave(" = %p", server);
 142         return server;
 143 }
 144
 145 /*
 146  * Install a server record in the namespace tree
 147  */
 148 static struct afs_server *afs_install_server(struct afs_net *net,
 149                                              struct afs_server *candidate)
 150 {
 151         const struct afs_addr_list *alist;
 152         struct afs_server *server;
 153         struct rb_node **pp, *p;
 154         int diff;
 155
 156         _enter("%p", candidate);
 157
 158         write_seqlock(&net->fs_lock);
 159
 160         /* Firstly install the server in the UUID lookup tree */
 161         pp = &net->fs_servers.rb_node;
 162         p = NULL;
 163         while (*pp) {
 164                 p = *pp;
 165                 _debug("- consider %p", p);
 166                 server = rb_entry(p, struct afs_server, uuid_rb);
 167                 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
 168                 if (diff < 0)
 169                         pp = &(*pp)->rb_left;
 170                 else if (diff > 0)
 171                         pp = &(*pp)->rb_right;
 172                 else
 173                         goto exists;
 174         }
 175
 176         server = candidate;
 177         rb_link_node(&server->uuid_rb, p, pp);
 178         rb_insert_color(&server->uuid_rb, &net->fs_servers);
 179         hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
 180
 181         write_seqlock(&net->fs_addr_lock);
 182         alist = rcu_dereference_protected(server->addresses,
 183                                           lockdep_is_held(&net->fs_addr_lock.lock));
 184
 185         /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
 186          * it in the IPv4 and/or IPv6 reverse-map lists.
 187          *
 188          * TODO: For speed we want to use something other than a flat list
 189          * here; even sorting the list in terms of lowest address would help a
 190          * bit, but anything we might want to do gets messy and memory
 191          * intensive.
 192          */
 193         if (alist->nr_ipv4 > 0)
 194                 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
 195         if (alist->nr_addrs > alist->nr_ipv4)
 196                 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
 197
 198         write_sequnlock(&net->fs_addr_lock);
 199
 200 exists:
 201         afs_get_server(server, afs_server_trace_get_install);
 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         server->debug_id = atomic_inc_return(&afs_server_debug_id);
 223         RCU_INIT_POINTER(server->addresses, alist);
 224         server->addr_version = alist->version;
 225         server->uuid = *uuid;
 226         server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
 227         rwlock_init(&server->fs_lock);
 228         INIT_HLIST_HEAD(&server->cb_volumes);
 229         rwlock_init(&server->cb_break_lock);
 230         init_waitqueue_head(&server->probe_wq);
 231         spin_lock_init(&server->probe_lock);
 232
 233         afs_inc_servers_outstanding(net);
 234         trace_afs_server(server, 1, afs_server_trace_alloc);
 235         _leave(" = %p", server);
 236         return server;
 237
 238 enomem:
 239         _leave(" = NULL [nomem]");
 240         return NULL;
 241 }
 242
 243 /*
 244  * Look up an address record for a server
 245  */
 246 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
 247                                                  struct key *key, const uuid_t *uuid)
 248 {
 249         struct afs_vl_cursor vc;
 250         struct afs_addr_list *alist = NULL;
 251         int ret;
 252
 253         ret = -ERESTARTSYS;
 254         if (afs_begin_vlserver_operation(&vc, cell, key)) {
 255                 while (afs_select_vlserver(&vc)) {
 256                         if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
 257                                 alist = afs_yfsvl_get_endpoints(&vc, uuid);
 258                         else
 259                                 alist = afs_vl_get_addrs_u(&vc, uuid);
 260                 }
 261
 262                 ret = afs_end_vlserver_operation(&vc);
 263         }
 264
 265         return ret < 0 ? ERR_PTR(ret) : alist;
 266 }
 267
 268 /*
 269  * Get or create a fileserver record.
 270  */
 271 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
 272                                      const uuid_t *uuid)
 273 {
 274         struct afs_addr_list *alist;
 275         struct afs_server *server, *candidate;
 276
 277         _enter("%p,%pU", cell->net, uuid);
 278
 279         server = afs_find_server_by_uuid(cell->net, uuid);
 280         if (server)
 281                 return server;
 282
 283         alist = afs_vl_lookup_addrs(cell, key, uuid);
 284         if (IS_ERR(alist))
 285                 return ERR_CAST(alist);
 286
 287         candidate = afs_alloc_server(cell->net, uuid, alist);
 288         if (!candidate) {
 289                 afs_put_addrlist(alist);
 290                 return ERR_PTR(-ENOMEM);
 291         }
 292
 293         server = afs_install_server(cell->net, candidate);
 294         if (server != candidate) {
 295                 afs_put_addrlist(alist);
 296                 kfree(candidate);
 297         }
 298
 299         _leave(" = %p{%d}", server, atomic_read(&server->usage));
 300         return server;
 301 }
 302
 303 /*
 304  * Set the server timer to fire after a given delay, assuming it's not already
 305  * set for an earlier time.
 306  */
 307 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
 308 {
 309         if (net->live) {
 310                 afs_inc_servers_outstanding(net);
 311                 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
 312                         afs_dec_servers_outstanding(net);
 313         }
 314 }
 315
 316 /*
 317  * Server management timer.  We have an increment on fs_outstanding that we
 318  * need to pass along to the work item.
 319  */
 320 void afs_servers_timer(struct timer_list *timer)
 321 {
 322         struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
 323
 324         _enter("");
 325         if (!queue_work(afs_wq, &net->fs_manager))
 326                 afs_dec_servers_outstanding(net);
 327 }
 328
 329 /*
 330  * Get a reference on a server object.
 331  */
 332 struct afs_server *afs_get_server(struct afs_server *server,
 333                                   enum afs_server_trace reason)
 334 {
 335         unsigned int u = atomic_inc_return(&server->usage);
 336
 337         trace_afs_server(server, u, reason);
 338         return server;
 339 }
 340
 341 /*
 342  * Release a reference on a server record.
 343  */
 344 void afs_put_server(struct afs_net *net, struct afs_server *server,
 345                     enum afs_server_trace reason)
 346 {
 347         unsigned int usage;
 348
 349         if (!server)
 350                 return;
 351
 352         server->put_time = ktime_get_real_seconds();
 353
 354         usage = atomic_dec_return(&server->usage);
 355
 356         trace_afs_server(server, usage, reason);
 357
 358         if (likely(usage > 0))
 359                 return;
 360
 361         afs_set_server_timer(net, afs_server_gc_delay);
 362 }
 363
 364 static void afs_server_rcu(struct rcu_head *rcu)
 365 {
 366         struct afs_server *server = container_of(rcu, struct afs_server, rcu);
 367
 368         trace_afs_server(server, atomic_read(&server->usage),
 369                          afs_server_trace_free);
 370         afs_put_addrlist(rcu_access_pointer(server->addresses));
 371         kfree(server);
 372 }
 373
 374 /*
 375  * destroy a dead server
 376  */
 377 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
 378 {
 379         struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
 380         struct afs_addr_cursor ac = {
 381                 .alist  = alist,
 382                 .index  = alist->preferred,
 383                 .error  = 0,
 384         };
 385
 386         trace_afs_server(server, atomic_read(&server->usage),
 387                          afs_server_trace_give_up_cb);
 388
 389         if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
 390                 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
 391
 392         wait_var_event(&server->probe_outstanding,
 393                        atomic_read(&server->probe_outstanding) == 0);
 394
 395         trace_afs_server(server, atomic_read(&server->usage),
 396                          afs_server_trace_destroy);
 397         call_rcu(&server->rcu, afs_server_rcu);
 398         afs_dec_servers_outstanding(net);
 399 }
 400
 401 /*
 402  * Garbage collect any expired servers.
 403  */
 404 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
 405 {
 406         struct afs_server *server;
 407         bool deleted;
 408         int usage;
 409
 410         while ((server = gc_list)) {
 411                 gc_list = server->gc_next;
 412
 413                 write_seqlock(&net->fs_lock);
 414                 usage = 1;
 415                 deleted = atomic_try_cmpxchg(&server->usage, &usage, 0);
 416                 trace_afs_server(server, usage, afs_server_trace_gc);
 417                 if (deleted) {
 418                         rb_erase(&server->uuid_rb, &net->fs_servers);
 419                         hlist_del_rcu(&server->proc_link);
 420                 }
 421                 write_sequnlock(&net->fs_lock);
 422
 423                 if (deleted) {
 424                         write_seqlock(&net->fs_addr_lock);
 425                         if (!hlist_unhashed(&server->addr4_link))
 426                                 hlist_del_rcu(&server->addr4_link);
 427                         if (!hlist_unhashed(&server->addr6_link))
 428                                 hlist_del_rcu(&server->addr6_link);
 429                         write_sequnlock(&net->fs_addr_lock);
 430                         afs_destroy_server(net, server);
 431                 }
 432         }
 433 }
 434
 435 /*
 436  * Manage the records of servers known to be within a network namespace.  This
 437  * includes garbage collecting unused servers.
 438  *
 439  * Note also that we were given an increment on net->servers_outstanding by
 440  * whoever queued us that we need to deal with before returning.
 441  */
 442 void afs_manage_servers(struct work_struct *work)
 443 {
 444         struct afs_net *net = container_of(work, struct afs_net, fs_manager);
 445         struct afs_server *gc_list = NULL;
 446         struct rb_node *cursor;
 447         time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
 448         bool purging = !net->live;
 449
 450         _enter("");
 451
 452         /* Trawl the server list looking for servers that have expired from
 453          * lack of use.
 454          */
 455         read_seqlock_excl(&net->fs_lock);
 456
 457         for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
 458                 struct afs_server *server =
 459                         rb_entry(cursor, struct afs_server, uuid_rb);
 460                 int usage = atomic_read(&server->usage);
 461
 462                 _debug("manage %pU %u", &server->uuid, usage);
 463
 464                 ASSERTCMP(usage, >=, 1);
 465                 ASSERTIFCMP(purging, usage, ==, 1);
 466
 467                 if (usage == 1) {
 468                         time64_t expire_at = server->put_time;
 469
 470                         if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
 471                             !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
 472                                 expire_at += afs_server_gc_delay;
 473                         if (purging || expire_at <= now) {
 474                                 server->gc_next = gc_list;
 475                                 gc_list = server;
 476                         } else if (expire_at < next_manage) {
 477                                 next_manage = expire_at;
 478                         }
 479                 }
 480         }
 481
 482         read_sequnlock_excl(&net->fs_lock);
 483
 484         /* Update the timer on the way out.  We have to pass an increment on
 485          * servers_outstanding in the namespace that we are in to the timer or
 486          * the work scheduler.
 487          */
 488         if (!purging && next_manage < TIME64_MAX) {
 489                 now = ktime_get_real_seconds();
 490
 491                 if (next_manage - now <= 0) {
 492                         if (queue_work(afs_wq, &net->fs_manager))
 493                                 afs_inc_servers_outstanding(net);
 494                 } else {
 495                         afs_set_server_timer(net, next_manage - now);
 496                 }
 497         }
 498
 499         afs_gc_servers(net, gc_list);
 500
 501         afs_dec_servers_outstanding(net);
 502         _leave(" [%d]", atomic_read(&net->servers_outstanding));
 503 }
 504
 505 static void afs_queue_server_manager(struct afs_net *net)
 506 {
 507         afs_inc_servers_outstanding(net);
 508         if (!queue_work(afs_wq, &net->fs_manager))
 509                 afs_dec_servers_outstanding(net);
 510 }
 511
 512 /*
 513  * Purge list of servers.
 514  */
 515 void afs_purge_servers(struct afs_net *net)
 516 {
 517         _enter("");
 518
 519         if (del_timer_sync(&net->fs_timer))
 520                 atomic_dec(&net->servers_outstanding);
 521
 522         afs_queue_server_manager(net);
 523
 524         _debug("wait");
 525         wait_var_event(&net->servers_outstanding,
 526                        !atomic_read(&net->servers_outstanding));
 527         _leave("");
 528 }
 529
 530 /*
 531  * Get an update for a server's address list.
 532  */
 533 static noinline bool afs_update_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 534 {
 535         struct afs_addr_list *alist, *discard;
 536
 537         _enter("");
 538
 539         trace_afs_server(server, atomic_read(&server->usage), afs_server_trace_update);
 540
 541         alist = afs_vl_lookup_addrs(fc->vnode->volume->cell, fc->key,
 542                                     &server->uuid);
 543         if (IS_ERR(alist)) {
 544                 if ((PTR_ERR(alist) == -ERESTARTSYS ||
 545                      PTR_ERR(alist) == -EINTR) &&
 546                     !(fc->flags & AFS_FS_CURSOR_INTR) &&
 547                     server->addresses) {
 548                         _leave(" = t [intr]");
 549                         return true;
 550                 }
 551                 fc->error = PTR_ERR(alist);
 552                 _leave(" = f [%d]", fc->error);
 553                 return false;
 554         }
 555
 556         discard = alist;
 557         if (server->addr_version != alist->version) {
 558                 write_lock(&server->fs_lock);
 559                 discard = rcu_dereference_protected(server->addresses,
 560                                                     lockdep_is_held(&server->fs_lock));
 561                 rcu_assign_pointer(server->addresses, alist);
 562                 server->addr_version = alist->version;
 563                 write_unlock(&server->fs_lock);
 564         }
 565
 566         server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
 567         afs_put_addrlist(discard);
 568         _leave(" = t");
 569         return true;
 570 }
 571
 572 /*
 573  * See if a server's address list needs updating.
 574  */
 575 bool afs_check_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 576 {
 577         time64_t now = ktime_get_real_seconds();
 578         long diff;
 579         bool success;
 580         int ret, retries = 0;
 581
 582         _enter("");
 583
 584         ASSERT(server);
 585
 586 retry:
 587         diff = READ_ONCE(server->update_at) - now;
 588         if (diff > 0) {
 589                 _leave(" = t [not now %ld]", diff);
 590                 return true;
 591         }
 592
 593         if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
 594                 success = afs_update_server_record(fc, server);
 595                 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
 596                 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
 597                 _leave(" = %d", success);
 598                 return success;
 599         }
 600
 601         ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
 602                           TASK_INTERRUPTIBLE);
 603         if (ret == -ERESTARTSYS) {
 604                 if (!(fc->flags & AFS_FS_CURSOR_INTR) && server->addresses) {
 605                         _leave(" = t [intr]");
 606                         return true;
 607                 }
 608                 fc->error = ret;
 609                 _leave(" = f [intr]");
 610                 return false;
 611         }
 612
 613         retries++;
 614         if (retries == 4) {
 615                 _leave(" = f [stale]");
 616                 ret = -ESTALE;
 617                 return false;
 618         }
 619         goto retry;
 620 }