fs/afs/server.c

   1 /* AFS server record management
   2  *
   3  * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
   4  * Written by David Howells (dhowells@redhat.com)
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the License, or (at your option) any later version.
  10  */
  11
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include "afs_fs.h"
  15 #include "internal.h"
  16
  17 static unsigned afs_server_gc_delay = 10;       /* Server record timeout in seconds */
  18 static unsigned afs_server_update_delay = 30;   /* Time till VLDB recheck in secs */
  19
  20 static void afs_inc_servers_outstanding(struct afs_net *net)
  21 {
  22         atomic_inc(&net->servers_outstanding);
  23 }
  24
  25 static void afs_dec_servers_outstanding(struct afs_net *net)
  26 {
  27         if (atomic_dec_and_test(&net->servers_outstanding))
  28                 wake_up_var(&net->servers_outstanding);
  29 }
  30
  31 /*
  32  * Find a server by one of its addresses.
  33  */
  34 struct afs_server *afs_find_server(struct afs_net *net,
  35                                    const struct sockaddr_rxrpc *srx)
  36 {
  37         const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
  38         const struct afs_addr_list *alist;
  39         struct afs_server *server = NULL;
  40         unsigned int i;
  41         bool ipv6 = true;
  42         int seq = 0, diff;
  43
  44         if (srx->transport.sin6.sin6_addr.s6_addr32[0] == 0 ||
  45             srx->transport.sin6.sin6_addr.s6_addr32[1] == 0 ||
  46             srx->transport.sin6.sin6_addr.s6_addr32[2] == htonl(0xffff))
  47                 ipv6 = false;
  48
  49         rcu_read_lock();
  50
  51         do {
  52                 if (server)
  53                         afs_put_server(net, server);
  54                 server = NULL;
  55                 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
  56
  57                 if (ipv6) {
  58                         hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
  59                                 alist = rcu_dereference(server->addresses);
  60                                 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
  61                                         b = &alist->addrs[i].transport.sin6;
  62                                         diff = ((u16 __force)a->sin6_port -
  63                                                 (u16 __force)b->sin6_port);
  64                                         if (diff == 0)
  65                                                 diff = memcmp(&a->sin6_addr,
  66                                                               &b->sin6_addr,
  67                                                               sizeof(struct in6_addr));
  68                                         if (diff == 0)
  69                                                 goto found;
  70                                         if (diff < 0) {
  71                                                 // TODO: Sort the list
  72                                                 //if (i == alist->nr_ipv4)
  73                                                 //      goto not_found;
  74                                                 break;
  75                                         }
  76                                 }
  77                         }
  78                 } else {
  79                         hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
  80                                 alist = rcu_dereference(server->addresses);
  81                                 for (i = 0; i < alist->nr_ipv4; i++) {
  82                                         b = &alist->addrs[i].transport.sin6;
  83                                         diff = ((u16 __force)a->sin6_port -
  84                                                 (u16 __force)b->sin6_port);
  85                                         if (diff == 0)
  86                                                 diff = ((u32 __force)a->sin6_addr.s6_addr32[3] -
  87                                                         (u32 __force)b->sin6_addr.s6_addr32[3]);
  88                                         if (diff == 0)
  89                                                 goto found;
  90                                         if (diff < 0) {
  91                                                 // TODO: Sort the list
  92                                                 //if (i == 0)
  93                                                 //      goto not_found;
  94                                                 break;
  95                                         }
  96                                 }
  97                         }
  98                 }
  99
 100         //not_found:
 101                 server = NULL;
 102         found:
 103                 if (server && !atomic_inc_not_zero(&server->usage))
 104                         server = NULL;
 105
 106         } while (need_seqretry(&net->fs_addr_lock, seq));
 107
 108         done_seqretry(&net->fs_addr_lock, seq);
 109
 110         rcu_read_unlock();
 111         return server;
 112 }
 113
 114 /*
 115  * Look up a server by its UUID
 116  */
 117 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
 118 {
 119         struct afs_server *server = NULL;
 120         struct rb_node *p;
 121         int diff, seq = 0;
 122
 123         _enter("%pU", uuid);
 124
 125         do {
 126                 /* Unfortunately, rbtree walking doesn't give reliable results
 127                  * under just the RCU read lock, so we have to check for
 128                  * changes.
 129                  */
 130                 if (server)
 131                         afs_put_server(net, server);
 132                 server = NULL;
 133
 134                 read_seqbegin_or_lock(&net->fs_lock, &seq);
 135
 136                 p = net->fs_servers.rb_node;
 137                 while (p) {
 138                         server = rb_entry(p, struct afs_server, uuid_rb);
 139
 140                         diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
 141                         if (diff < 0) {
 142                                 p = p->rb_left;
 143                         } else if (diff > 0) {
 144                                 p = p->rb_right;
 145                         } else {
 146                                 afs_get_server(server);
 147                                 break;
 148                         }
 149
 150                         server = NULL;
 151                 }
 152         } while (need_seqretry(&net->fs_lock, seq));
 153
 154         done_seqretry(&net->fs_lock, seq);
 155
 156         _leave(" = %p", server);
 157         return server;
 158 }
 159
 160 /*
 161  * Install a server record in the namespace tree
 162  */
 163 static struct afs_server *afs_install_server(struct afs_net *net,
 164                                              struct afs_server *candidate)
 165 {
 166         const struct afs_addr_list *alist;
 167         struct afs_server *server;
 168         struct rb_node **pp, *p;
 169         int ret = -EEXIST, diff;
 170
 171         _enter("%p", candidate);
 172
 173         write_seqlock(&net->fs_lock);
 174
 175         /* Firstly install the server in the UUID lookup tree */
 176         pp = &net->fs_servers.rb_node;
 177         p = NULL;
 178         while (*pp) {
 179                 p = *pp;
 180                 _debug("- consider %p", p);
 181                 server = rb_entry(p, struct afs_server, uuid_rb);
 182                 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
 183                 if (diff < 0)
 184                         pp = &(*pp)->rb_left;
 185                 else if (diff > 0)
 186                         pp = &(*pp)->rb_right;
 187                 else
 188                         goto exists;
 189         }
 190
 191         server = candidate;
 192         rb_link_node(&server->uuid_rb, p, pp);
 193         rb_insert_color(&server->uuid_rb, &net->fs_servers);
 194         hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
 195
 196         write_seqlock(&net->fs_addr_lock);
 197         alist = rcu_dereference_protected(server->addresses,
 198                                           lockdep_is_held(&net->fs_addr_lock.lock));
 199
 200         /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
 201          * it in the IPv4 and/or IPv6 reverse-map lists.
 202          *
 203          * TODO: For speed we want to use something other than a flat list
 204          * here; even sorting the list in terms of lowest address would help a
 205          * bit, but anything we might want to do gets messy and memory
 206          * intensive.
 207          */
 208         if (alist->nr_ipv4 > 0)
 209                 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
 210         if (alist->nr_addrs > alist->nr_ipv4)
 211                 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
 212
 213         write_sequnlock(&net->fs_addr_lock);
 214         ret = 0;
 215
 216 exists:
 217         afs_get_server(server);
 218         write_sequnlock(&net->fs_lock);
 219         return server;
 220 }
 221
 222 /*
 223  * allocate a new server record
 224  */
 225 static struct afs_server *afs_alloc_server(struct afs_net *net,
 226                                            const uuid_t *uuid,
 227                                            struct afs_addr_list *alist)
 228 {
 229         struct afs_server *server;
 230
 231         _enter("");
 232
 233         server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
 234         if (!server)
 235                 goto enomem;
 236
 237         atomic_set(&server->usage, 1);
 238         RCU_INIT_POINTER(server->addresses, alist);
 239         server->addr_version = alist->version;
 240         server->uuid = *uuid;
 241         server->flags = (1UL << AFS_SERVER_FL_NEW);
 242         server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
 243         rwlock_init(&server->fs_lock);
 244         INIT_LIST_HEAD(&server->cb_interests);
 245         rwlock_init(&server->cb_break_lock);
 246
 247         afs_inc_servers_outstanding(net);
 248         _leave(" = %p", server);
 249         return server;
 250
 251 enomem:
 252         _leave(" = NULL [nomem]");
 253         return NULL;
 254 }
 255
 256 /*
 257  * Look up an address record for a server
 258  */
 259 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
 260                                                  struct key *key, const uuid_t *uuid)
 261 {
 262         struct afs_addr_cursor ac;
 263         struct afs_addr_list *alist;
 264         int ret;
 265
 266         ret = afs_set_vl_cursor(&ac, cell);
 267         if (ret < 0)
 268                 return ERR_PTR(ret);
 269
 270         while (afs_iterate_addresses(&ac)) {
 271                 if (test_bit(ac.index, &ac.alist->yfs))
 272                         alist = afs_yfsvl_get_endpoints(cell->net, &ac, key, uuid);
 273                 else
 274                         alist = afs_vl_get_addrs_u(cell->net, &ac, key, uuid);
 275                 switch (ac.error) {
 276                 case 0:
 277                         afs_end_cursor(&ac);
 278                         return alist;
 279                 case -ECONNABORTED:
 280                         ac.error = afs_abort_to_error(ac.abort_code);
 281                         goto error;
 282                 case -ENOMEM:
 283                 case -ENONET:
 284                         goto error;
 285                 case -ENETUNREACH:
 286                 case -EHOSTUNREACH:
 287                 case -ECONNREFUSED:
 288                         break;
 289                 default:
 290                         ac.error = -EIO;
 291                         goto error;
 292                 }
 293         }
 294
 295 error:
 296         return ERR_PTR(afs_end_cursor(&ac));
 297 }
 298
 299 /*
 300  * Get or create a fileserver record.
 301  */
 302 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
 303                                      const uuid_t *uuid)
 304 {
 305         struct afs_addr_list *alist;
 306         struct afs_server *server, *candidate;
 307
 308         _enter("%p,%pU", cell->net, uuid);
 309
 310         server = afs_find_server_by_uuid(cell->net, uuid);
 311         if (server)
 312                 return server;
 313
 314         alist = afs_vl_lookup_addrs(cell, key, uuid);
 315         if (IS_ERR(alist))
 316                 return ERR_CAST(alist);
 317
 318         candidate = afs_alloc_server(cell->net, uuid, alist);
 319         if (!candidate) {
 320                 afs_put_addrlist(alist);
 321                 return ERR_PTR(-ENOMEM);
 322         }
 323
 324         server = afs_install_server(cell->net, candidate);
 325         if (server != candidate) {
 326                 afs_put_addrlist(alist);
 327                 kfree(candidate);
 328         }
 329
 330         _leave(" = %p{%d}", server, atomic_read(&server->usage));
 331         return server;
 332 }
 333
 334 /*
 335  * Set the server timer to fire after a given delay, assuming it's not already
 336  * set for an earlier time.
 337  */
 338 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
 339 {
 340         if (net->live) {
 341                 afs_inc_servers_outstanding(net);
 342                 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
 343                         afs_dec_servers_outstanding(net);
 344         }
 345 }
 346
 347 /*
 348  * Server management timer.  We have an increment on fs_outstanding that we
 349  * need to pass along to the work item.
 350  */
 351 void afs_servers_timer(struct timer_list *timer)
 352 {
 353         struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
 354
 355         _enter("");
 356         if (!queue_work(afs_wq, &net->fs_manager))
 357                 afs_dec_servers_outstanding(net);
 358 }
 359
 360 /*
 361  * Release a reference on a server record.
 362  */
 363 void afs_put_server(struct afs_net *net, struct afs_server *server)
 364 {
 365         unsigned int usage;
 366
 367         if (!server)
 368                 return;
 369
 370         server->put_time = ktime_get_real_seconds();
 371
 372         usage = atomic_dec_return(&server->usage);
 373
 374         _enter("{%u}", usage);
 375
 376         if (likely(usage > 0))
 377                 return;
 378
 379         afs_set_server_timer(net, afs_server_gc_delay);
 380 }
 381
 382 static void afs_server_rcu(struct rcu_head *rcu)
 383 {
 384         struct afs_server *server = container_of(rcu, struct afs_server, rcu);
 385
 386         afs_put_addrlist(rcu_access_pointer(server->addresses));
 387         kfree(server);
 388 }
 389
 390 /*
 391  * destroy a dead server
 392  */
 393 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
 394 {
 395         struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
 396         struct afs_addr_cursor ac = {
 397                 .alist  = alist,
 398                 .addr   = &alist->addrs[0],
 399                 .start  = alist->index,
 400                 .index  = alist->index,
 401                 .error  = 0,
 402         };
 403         _enter("%p", server);
 404
 405         afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
 406         call_rcu(&server->rcu, afs_server_rcu);
 407         afs_dec_servers_outstanding(net);
 408 }
 409
 410 /*
 411  * Garbage collect any expired servers.
 412  */
 413 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
 414 {
 415         struct afs_server *server;
 416         bool deleted;
 417         int usage;
 418
 419         while ((server = gc_list)) {
 420                 gc_list = server->gc_next;
 421
 422                 write_seqlock(&net->fs_lock);
 423                 usage = 1;
 424                 deleted = atomic_try_cmpxchg(&server->usage, &usage, 0);
 425                 if (deleted) {
 426                         rb_erase(&server->uuid_rb, &net->fs_servers);
 427                         hlist_del_rcu(&server->proc_link);
 428                 }
 429                 write_sequnlock(&net->fs_lock);
 430
 431                 if (deleted)
 432                         afs_destroy_server(net, server);
 433         }
 434 }
 435
 436 /*
 437  * Manage the records of servers known to be within a network namespace.  This
 438  * includes garbage collecting unused servers.
 439  *
 440  * Note also that we were given an increment on net->servers_outstanding by
 441  * whoever queued us that we need to deal with before returning.
 442  */
 443 void afs_manage_servers(struct work_struct *work)
 444 {
 445         struct afs_net *net = container_of(work, struct afs_net, fs_manager);
 446         struct afs_server *gc_list = NULL;
 447         struct rb_node *cursor;
 448         time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
 449         bool purging = !net->live;
 450
 451         _enter("");
 452
 453         /* Trawl the server list looking for servers that have expired from
 454          * lack of use.
 455          */
 456         read_seqlock_excl(&net->fs_lock);
 457
 458         for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
 459                 struct afs_server *server =
 460                         rb_entry(cursor, struct afs_server, uuid_rb);
 461                 int usage = atomic_read(&server->usage);
 462
 463                 _debug("manage %pU %u", &server->uuid, usage);
 464
 465                 ASSERTCMP(usage, >=, 1);
 466                 ASSERTIFCMP(purging, usage, ==, 1);
 467
 468                 if (usage == 1) {
 469                         time64_t expire_at = server->put_time;
 470
 471                         if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
 472                             !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
 473                                 expire_at += afs_server_gc_delay;
 474                         if (purging || expire_at <= now) {
 475                                 server->gc_next = gc_list;
 476                                 gc_list = server;
 477                         } else if (expire_at < next_manage) {
 478                                 next_manage = expire_at;
 479                         }
 480                 }
 481         }
 482
 483         read_sequnlock_excl(&net->fs_lock);
 484
 485         /* Update the timer on the way out.  We have to pass an increment on
 486          * servers_outstanding in the namespace that we are in to the timer or
 487          * the work scheduler.
 488          */
 489         if (!purging && next_manage < TIME64_MAX) {
 490                 now = ktime_get_real_seconds();
 491
 492                 if (next_manage - now <= 0) {
 493                         if (queue_work(afs_wq, &net->fs_manager))
 494                                 afs_inc_servers_outstanding(net);
 495                 } else {
 496                         afs_set_server_timer(net, next_manage - now);
 497                 }
 498         }
 499
 500         afs_gc_servers(net, gc_list);
 501
 502         afs_dec_servers_outstanding(net);
 503         _leave(" [%d]", atomic_read(&net->servers_outstanding));
 504 }
 505
 506 static void afs_queue_server_manager(struct afs_net *net)
 507 {
 508         afs_inc_servers_outstanding(net);
 509         if (!queue_work(afs_wq, &net->fs_manager))
 510                 afs_dec_servers_outstanding(net);
 511 }
 512
 513 /*
 514  * Purge list of servers.
 515  */
 516 void afs_purge_servers(struct afs_net *net)
 517 {
 518         _enter("");
 519
 520         if (del_timer_sync(&net->fs_timer))
 521                 atomic_dec(&net->servers_outstanding);
 522
 523         afs_queue_server_manager(net);
 524
 525         _debug("wait");
 526         wait_var_event(&net->servers_outstanding,
 527                        !atomic_read(&net->servers_outstanding));
 528         _leave("");
 529 }
 530
 531 /*
 532  * Probe a fileserver to find its capabilities.
 533  *
 534  * TODO: Try service upgrade.
 535  */
 536 static bool afs_do_probe_fileserver(struct afs_fs_cursor *fc)
 537 {
 538         _enter("");
 539
 540         fc->ac.addr = NULL;
 541         fc->ac.start = READ_ONCE(fc->ac.alist->index);
 542         fc->ac.index = fc->ac.start;
 543         fc->ac.error = 0;
 544         fc->ac.begun = false;
 545
 546         while (afs_iterate_addresses(&fc->ac)) {
 547                 afs_fs_get_capabilities(afs_v2net(fc->vnode), fc->cbi->server,
 548                                         &fc->ac, fc->key);
 549                 switch (fc->ac.error) {
 550                 case 0:
 551                         afs_end_cursor(&fc->ac);
 552                         set_bit(AFS_SERVER_FL_PROBED, &fc->cbi->server->flags);
 553                         return true;
 554                 case -ECONNABORTED:
 555                         fc->ac.error = afs_abort_to_error(fc->ac.abort_code);
 556                         goto error;
 557                 case -ENOMEM:
 558                 case -ENONET:
 559                         goto error;
 560                 case -ENETUNREACH:
 561                 case -EHOSTUNREACH:
 562                 case -ECONNREFUSED:
 563                 case -ETIMEDOUT:
 564                 case -ETIME:
 565                         break;
 566                 default:
 567                         fc->ac.error = -EIO;
 568                         goto error;
 569                 }
 570         }
 571
 572 error:
 573         afs_end_cursor(&fc->ac);
 574         return false;
 575 }
 576
 577 /*
 578  * If we haven't already, try probing the fileserver to get its capabilities.
 579  * We try not to instigate parallel probes, but it's possible that the parallel
 580  * probes will fail due to authentication failure when ours would succeed.
 581  *
 582  * TODO: Try sending an anonymous probe if an authenticated probe fails.
 583  */
 584 bool afs_probe_fileserver(struct afs_fs_cursor *fc)
 585 {
 586         bool success;
 587         int ret, retries = 0;
 588
 589         _enter("");
 590
 591 retry:
 592         if (test_bit(AFS_SERVER_FL_PROBED, &fc->cbi->server->flags)) {
 593                 _leave(" = t");
 594                 return true;
 595         }
 596
 597         if (!test_and_set_bit_lock(AFS_SERVER_FL_PROBING, &fc->cbi->server->flags)) {
 598                 success = afs_do_probe_fileserver(fc);
 599                 clear_bit_unlock(AFS_SERVER_FL_PROBING, &fc->cbi->server->flags);
 600                 wake_up_bit(&fc->cbi->server->flags, AFS_SERVER_FL_PROBING);
 601                 _leave(" = t");
 602                 return success;
 603         }
 604
 605         _debug("wait");
 606         ret = wait_on_bit(&fc->cbi->server->flags, AFS_SERVER_FL_PROBING,
 607                           TASK_INTERRUPTIBLE);
 608         if (ret == -ERESTARTSYS) {
 609                 fc->ac.error = ret;
 610                 _leave(" = f [%d]", ret);
 611                 return false;
 612         }
 613
 614         retries++;
 615         if (retries == 4) {
 616                 fc->ac.error = -ESTALE;
 617                 _leave(" = f [stale]");
 618                 return false;
 619         }
 620         _debug("retry");
 621         goto retry;
 622 }
 623
 624 /*
 625  * Get an update for a server's address list.
 626  */
 627 static noinline bool afs_update_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 628 {
 629         struct afs_addr_list *alist, *discard;
 630
 631         _enter("");
 632
 633         alist = afs_vl_lookup_addrs(fc->vnode->volume->cell, fc->key,
 634                                     &server->uuid);
 635         if (IS_ERR(alist)) {
 636                 fc->ac.error = PTR_ERR(alist);
 637                 _leave(" = f [%d]", fc->ac.error);
 638                 return false;
 639         }
 640
 641         discard = alist;
 642         if (server->addr_version != alist->version) {
 643                 write_lock(&server->fs_lock);
 644                 discard = rcu_dereference_protected(server->addresses,
 645                                                     lockdep_is_held(&server->fs_lock));
 646                 rcu_assign_pointer(server->addresses, alist);
 647                 server->addr_version = alist->version;
 648                 write_unlock(&server->fs_lock);
 649         }
 650
 651         server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
 652         afs_put_addrlist(discard);
 653         _leave(" = t");
 654         return true;
 655 }
 656
 657 /*
 658  * See if a server's address list needs updating.
 659  */
 660 bool afs_check_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
 661 {
 662         time64_t now = ktime_get_real_seconds();
 663         long diff;
 664         bool success;
 665         int ret, retries = 0;
 666
 667         _enter("");
 668
 669         ASSERT(server);
 670
 671 retry:
 672         diff = READ_ONCE(server->update_at) - now;
 673         if (diff > 0) {
 674                 _leave(" = t [not now %ld]", diff);
 675                 return true;
 676         }
 677
 678         if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
 679                 success = afs_update_server_record(fc, server);
 680                 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
 681                 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
 682                 _leave(" = %d", success);
 683                 return success;
 684         }
 685
 686         ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
 687                           TASK_INTERRUPTIBLE);
 688         if (ret == -ERESTARTSYS) {
 689                 fc->ac.error = ret;
 690                 _leave(" = f [intr]");
 691                 return false;
 692         }
 693
 694         retries++;
 695         if (retries == 4) {
 696                 _leave(" = f [stale]");
 697                 ret = -ESTALE;
 698                 return false;
 699         }
 700         goto retry;
 701 }