fs/afs/volume.c

   1 /* AFS volume 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/kernel.h>
  13 #include <linux/slab.h>
  14 #include "internal.h"
  15
  16 unsigned __read_mostly afs_volume_gc_delay = 10;
  17 unsigned __read_mostly afs_volume_record_life = 60 * 60;
  18
  19 static const char *const afs_voltypes[] = { "R/W", "R/O", "BAK" };
  20
  21 /*
  22  * Allocate a volume record and load it up from a vldb record.
  23  */
  24 static struct afs_volume *afs_alloc_volume(struct afs_mount_params *params,
  25                                            struct afs_vldb_entry *vldb,
  26                                            unsigned long type_mask)
  27 {
  28         struct afs_server_list *slist;
  29         struct afs_volume *volume;
  30         int ret = -ENOMEM, nr_servers = 0, i;
  31
  32         for (i = 0; i < vldb->nr_servers; i++)
  33                 if (vldb->fs_mask[i] & type_mask)
  34                         nr_servers++;
  35
  36         volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
  37         if (!volume)
  38                 goto error_0;
  39
  40         volume->vid             = vldb->vid[params->type];
  41         volume->update_at       = ktime_get_real_seconds() + afs_volume_record_life;
  42         volume->cell            = afs_get_cell(params->cell);
  43         volume->type            = params->type;
  44         volume->type_force      = params->force;
  45         volume->name_len        = vldb->name_len;
  46
  47         atomic_set(&volume->usage, 1);
  48         INIT_LIST_HEAD(&volume->proc_link);
  49         rwlock_init(&volume->servers_lock);
  50         memcpy(volume->name, vldb->name, vldb->name_len + 1);
  51
  52         slist = afs_alloc_server_list(params->cell, params->key, vldb, type_mask);
  53         if (IS_ERR(slist)) {
  54                 ret = PTR_ERR(slist);
  55                 goto error_1;
  56         }
  57
  58         refcount_set(&slist->usage, 1);
  59         volume->servers = slist;
  60         return volume;
  61
  62 error_1:
  63         afs_put_cell(params->net, volume->cell);
  64         kfree(volume);
  65 error_0:
  66         return ERR_PTR(ret);
  67 }
  68
  69 /*
  70  * Look up a VLDB record for a volume.
  71  */
  72 static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
  73                                                  struct key *key,
  74                                                  const char *volname,
  75                                                  size_t volnamesz)
  76 {
  77         struct afs_addr_cursor ac;
  78         struct afs_vldb_entry *vldb;
  79         int ret;
  80
  81         ret = afs_set_vl_cursor(&ac, cell);
  82         if (ret < 0)
  83                 return ERR_PTR(ret);
  84
  85         while (afs_iterate_addresses(&ac)) {
  86                 if (!test_bit(ac.index, &ac.alist->probed)) {
  87                         ret = afs_vl_get_capabilities(cell->net, &ac, key);
  88                         switch (ret) {
  89                         case VL_SERVICE:
  90                                 clear_bit(ac.index, &ac.alist->yfs);
  91                                 set_bit(ac.index, &ac.alist->probed);
  92                                 ac.addr->srx_service = ret;
  93                                 break;
  94                         case YFS_VL_SERVICE:
  95                                 set_bit(ac.index, &ac.alist->yfs);
  96                                 set_bit(ac.index, &ac.alist->probed);
  97                                 ac.addr->srx_service = ret;
  98                                 break;
  99                         }
 100                 }
 101
 102                 vldb = afs_vl_get_entry_by_name_u(cell->net, &ac, key,
 103                                                   volname, volnamesz);
 104                 switch (ac.error) {
 105                 case 0:
 106                         afs_end_cursor(&ac);
 107                         return vldb;
 108                 case -ECONNABORTED:
 109                         ac.error = afs_abort_to_error(ac.abort_code);
 110                         goto error;
 111                 case -ENOMEM:
 112                 case -ENONET:
 113                         goto error;
 114                 case -ENETUNREACH:
 115                 case -EHOSTUNREACH:
 116                 case -ECONNREFUSED:
 117                         break;
 118                 default:
 119                         ac.error = -EIO;
 120                         goto error;
 121                 }
 122         }
 123
 124 error:
 125         return ERR_PTR(afs_end_cursor(&ac));
 126 }
 127
 128 /*
 129  * Look up a volume in the VL server and create a candidate volume record for
 130  * it.
 131  *
 132  * The volume name can be one of the following:
 133  *      "%[cell:]volume[.]"             R/W volume
 134  *      "#[cell:]volume[.]"             R/O or R/W volume (rwparent=0),
 135  *                                       or R/W (rwparent=1) volume
 136  *      "%[cell:]volume.readonly"       R/O volume
 137  *      "#[cell:]volume.readonly"       R/O volume
 138  *      "%[cell:]volume.backup"         Backup volume
 139  *      "#[cell:]volume.backup"         Backup volume
 140  *
 141  * The cell name is optional, and defaults to the current cell.
 142  *
 143  * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
 144  * Guide
 145  * - Rule 1: Explicit type suffix forces access of that type or nothing
 146  *           (no suffix, then use Rule 2 & 3)
 147  * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
 148  *           if not available
 149  * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
 150  *           explicitly told otherwise
 151  */
 152 struct afs_volume *afs_create_volume(struct afs_mount_params *params)
 153 {
 154         struct afs_vldb_entry *vldb;
 155         struct afs_volume *volume;
 156         unsigned long type_mask = 1UL << params->type;
 157
 158         vldb = afs_vl_lookup_vldb(params->cell, params->key,
 159                                   params->volname, params->volnamesz);
 160         if (IS_ERR(vldb))
 161                 return ERR_CAST(vldb);
 162
 163         if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
 164                 volume = ERR_PTR(vldb->error);
 165                 goto error;
 166         }
 167
 168         /* Make the final decision on the type we want */
 169         volume = ERR_PTR(-ENOMEDIUM);
 170         if (params->force) {
 171                 if (!(vldb->flags & type_mask))
 172                         goto error;
 173         } else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
 174                 params->type = AFSVL_ROVOL;
 175         } else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
 176                 params->type = AFSVL_RWVOL;
 177         } else {
 178                 goto error;
 179         }
 180
 181         type_mask = 1UL << params->type;
 182         volume = afs_alloc_volume(params, vldb, type_mask);
 183
 184 error:
 185         kfree(vldb);
 186         return volume;
 187 }
 188
 189 /*
 190  * Destroy a volume record
 191  */
 192 static void afs_destroy_volume(struct afs_net *net, struct afs_volume *volume)
 193 {
 194         _enter("%p", volume);
 195
 196 #ifdef CONFIG_AFS_FSCACHE
 197         ASSERTCMP(volume->cache, ==, NULL);
 198 #endif
 199
 200         afs_put_serverlist(net, volume->servers);
 201         afs_put_cell(net, volume->cell);
 202         kfree(volume);
 203
 204         _leave(" [destroyed]");
 205 }
 206
 207 /*
 208  * Drop a reference on a volume record.
 209  */
 210 void afs_put_volume(struct afs_cell *cell, struct afs_volume *volume)
 211 {
 212         if (volume) {
 213                 _enter("%s", volume->name);
 214
 215                 if (atomic_dec_and_test(&volume->usage))
 216                         afs_destroy_volume(cell->net, volume);
 217         }
 218 }
 219
 220 /*
 221  * Activate a volume.
 222  */
 223 void afs_activate_volume(struct afs_volume *volume)
 224 {
 225 #ifdef CONFIG_AFS_FSCACHE
 226         volume->cache = fscache_acquire_cookie(volume->cell->cache,
 227                                                &afs_volume_cache_index_def,
 228                                                volume, true);
 229 #endif
 230
 231         write_lock(&volume->cell->proc_lock);
 232         list_add_tail(&volume->proc_link, &volume->cell->proc_volumes);
 233         write_unlock(&volume->cell->proc_lock);
 234 }
 235
 236 /*
 237  * Deactivate a volume.
 238  */
 239 void afs_deactivate_volume(struct afs_volume *volume)
 240 {
 241         _enter("%s", volume->name);
 242
 243         write_lock(&volume->cell->proc_lock);
 244         list_del_init(&volume->proc_link);
 245         write_unlock(&volume->cell->proc_lock);
 246
 247 #ifdef CONFIG_AFS_FSCACHE
 248         fscache_relinquish_cookie(volume->cache,
 249                                   test_bit(AFS_VOLUME_DELETED, &volume->flags));
 250         volume->cache = NULL;
 251 #endif
 252
 253         _leave("");
 254 }
 255
 256 /*
 257  * Query the VL service to update the volume status.
 258  */
 259 static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
 260 {
 261         struct afs_server_list *new, *old, *discard;
 262         struct afs_vldb_entry *vldb;
 263         char idbuf[16];
 264         int ret, idsz;
 265
 266         _enter("");
 267
 268         /* We look up an ID by passing it as a decimal string in the
 269          * operation's name parameter.
 270          */
 271         idsz = sprintf(idbuf, "%u", volume->vid);
 272
 273         vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
 274         if (IS_ERR(vldb)) {
 275                 ret = PTR_ERR(vldb);
 276                 goto error;
 277         }
 278
 279         /* See if the volume got renamed. */
 280         if (vldb->name_len != volume->name_len ||
 281             memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
 282                 /* TODO: Use RCU'd string. */
 283                 memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
 284                 volume->name_len = vldb->name_len;
 285         }
 286
 287         /* See if the volume's server list got updated. */
 288         new = afs_alloc_server_list(volume->cell, key,
 289                                     vldb, (1 << volume->type));
 290         if (IS_ERR(new)) {
 291                 ret = PTR_ERR(new);
 292                 goto error_vldb;
 293         }
 294
 295         write_lock(&volume->servers_lock);
 296
 297         discard = new;
 298         old = volume->servers;
 299         if (afs_annotate_server_list(new, old)) {
 300                 new->seq = volume->servers_seq + 1;
 301                 volume->servers = new;
 302                 smp_wmb();
 303                 volume->servers_seq++;
 304                 discard = old;
 305         }
 306
 307         volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
 308         clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
 309         write_unlock(&volume->servers_lock);
 310         ret = 0;
 311
 312         afs_put_serverlist(volume->cell->net, discard);
 313 error_vldb:
 314         kfree(vldb);
 315 error:
 316         _leave(" = %d", ret);
 317         return ret;
 318 }
 319
 320 /*
 321  * Make sure the volume record is up to date.
 322  */
 323 int afs_check_volume_status(struct afs_volume *volume, struct key *key)
 324 {
 325         time64_t now = ktime_get_real_seconds();
 326         int ret, retries = 0;
 327
 328         _enter("");
 329
 330         if (volume->update_at <= now)
 331                 set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
 332
 333 retry:
 334         if (!test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags) &&
 335             !test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
 336                 _leave(" = 0");
 337                 return 0;
 338         }
 339
 340         if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
 341                 ret = afs_update_volume_status(volume, key);
 342                 clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
 343                 clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
 344                 wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
 345                 _leave(" = %d", ret);
 346                 return ret;
 347         }
 348
 349         if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
 350                 _leave(" = 0 [no wait]");
 351                 return 0;
 352         }
 353
 354         ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT, TASK_INTERRUPTIBLE);
 355         if (ret == -ERESTARTSYS) {
 356                 _leave(" = %d", ret);
 357                 return ret;
 358         }
 359
 360         retries++;
 361         if (retries == 4) {
 362                 _leave(" = -ESTALE");
 363                 return -ESTALE;
 364         }
 365         goto retry;
 366 }