module/zfs/spa_config.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2013 by Delphix. All rights reserved.
  26  */
  27
  28 #include <sys/spa.h>
  29 #include <sys/spa_impl.h>
  30 #include <sys/nvpair.h>
  31 #include <sys/uio.h>
  32 #include <sys/fs/zfs.h>
  33 #include <sys/vdev_impl.h>
  34 #include <sys/zfs_ioctl.h>
  35 #include <sys/systeminfo.h>
  36 #include <sys/sunddi.h>
  37 #include <sys/zfeature.h>
  38 #ifdef _KERNEL
  39 #include <sys/kobj.h>
  40 #include <sys/zone.h>
  41 #endif
  42
  43 /*
  44  * Pool configuration repository.
  45  *
  46  * Pool configuration is stored as a packed nvlist on the filesystem.  By
  47  * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
  48  * (when the ZFS module is loaded).  Pools can also have the 'cachefile'
  49  * property set that allows them to be stored in an alternate location until
  50  * the control of external software.
  51  *
  52  * For each cache file, we have a single nvlist which holds all the
  53  * configuration information.  When the module loads, we read this information
  54  * from /etc/zfs/zpool.cache and populate the SPA namespace.  This namespace is
  55  * maintained independently in spa.c.  Whenever the namespace is modified, or
  56  * the configuration of a pool is changed, we call spa_config_sync(), which
  57  * walks through all the active pools and writes the configuration to disk.
  58  */
  59
  60 static uint64_t spa_config_generation = 1;
  61
  62 /*
  63  * This can be overridden in userland to preserve an alternate namespace for
  64  * userland pools when doing testing.
  65  */
  66 char *spa_config_path = ZPOOL_CACHE;
  67 int zfs_autoimport_disable = 1;
  68
  69 /*
  70  * Called when the module is first loaded, this routine loads the configuration
  71  * file into the SPA namespace.  It does not actually open or load the pools; it
  72  * only populates the namespace.
  73  */
  74 void
  75 spa_config_load(void)
  76 {
  77         void *buf = NULL;
  78         nvlist_t *nvlist, *child;
  79         nvpair_t *nvpair;
  80         char *pathname;
  81         struct _buf *file;
  82         uint64_t fsize;
  83
  84 #ifdef _KERNEL
  85         if (zfs_autoimport_disable)
  86                 return;
  87 #endif
  88
  89         /*
  90          * Open the configuration file.
  91          */
  92         pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
  93
  94         (void) snprintf(pathname, MAXPATHLEN, "%s%s",
  95             (rootdir != NULL) ? "./" : "", spa_config_path);
  96
  97         file = kobj_open_file(pathname);
  98
  99         kmem_free(pathname, MAXPATHLEN);
 100
 101         if (file == (struct _buf *)-1)
 102                 return;
 103
 104         if (kobj_get_filesize(file, &fsize) != 0)
 105                 goto out;
 106
 107         buf = kmem_alloc(fsize, KM_SLEEP);
 108
 109         /*
 110          * Read the nvlist from the file.
 111          */
 112         if (kobj_read_file(file, buf, fsize, 0) < 0)
 113                 goto out;
 114
 115         /*
 116          * Unpack the nvlist.
 117          */
 118         if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
 119                 goto out;
 120
 121         /*
 122          * Iterate over all elements in the nvlist, creating a new spa_t for
 123          * each one with the specified configuration.
 124          */
 125         mutex_enter(&spa_namespace_lock);
 126         nvpair = NULL;
 127         while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
 128                 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
 129                         continue;
 130
 131                 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
 132
 133                 if (spa_lookup(nvpair_name(nvpair)) != NULL)
 134                         continue;
 135                 (void) spa_add(nvpair_name(nvpair), child, NULL);
 136         }
 137         mutex_exit(&spa_namespace_lock);
 138
 139         nvlist_free(nvlist);
 140
 141 out:
 142         if (buf != NULL)
 143                 kmem_free(buf, fsize);
 144
 145         kobj_close_file(file);
 146 }
 147
 148 static void
 149 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 150 {
 151         size_t buflen;
 152         char *buf;
 153         vnode_t *vp;
 154         int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
 155         char *temp;
 156
 157         /*
 158          * If the nvlist is empty (NULL), then remove the old cachefile.
 159          */
 160         if (nvl == NULL) {
 161                 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
 162                 return;
 163         }
 164
 165         /*
 166          * Pack the configuration into a buffer.
 167          */
 168         VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
 169
 170         buf = vmem_alloc(buflen, KM_SLEEP);
 171         temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
 172
 173         VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
 174             KM_SLEEP) == 0);
 175
 176         /*
 177          * Write the configuration to disk.  We need to do the traditional
 178          * 'write to temporary file, sync, move over original' to make sure we
 179          * always have a consistent view of the data.
 180          */
 181         (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
 182
 183         if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
 184                 if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
 185                     0, RLIM64_INFINITY, kcred, NULL) == 0 &&
 186                     VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
 187                         (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
 188                 }
 189                 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
 190         }
 191
 192         (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
 193
 194         vmem_free(buf, buflen);
 195         kmem_free(temp, MAXPATHLEN);
 196 }
 197
 198 /*
 199  * Synchronize pool configuration to disk.  This must be called with the
 200  * namespace lock held. Synchronizing the pool cache is typically done after
 201  * the configuration has been synced to the MOS. This exposes a window where
 202  * the MOS config will have been updated but the cache file has not. If
 203  * the system were to crash at that instant then the cached config may not
 204  * contain the correct information to open the pool and an explicity import
 205  * would be required.
 206  */
 207 void
 208 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
 209 {
 210         spa_config_dirent_t *dp, *tdp;
 211         nvlist_t *nvl;
 212         char *pool_name;
 213
 214         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 215
 216         if (rootdir == NULL || !(spa_mode_global & FWRITE))
 217                 return;
 218
 219         /*
 220          * Iterate over all cachefiles for the pool, past or present.  When the
 221          * cachefile is changed, the new one is pushed onto this list, allowing
 222          * us to update previous cachefiles that no longer contain this pool.
 223          */
 224         for (dp = list_head(&target->spa_config_list); dp != NULL;
 225             dp = list_next(&target->spa_config_list, dp)) {
 226                 spa_t *spa = NULL;
 227                 if (dp->scd_path == NULL)
 228                         continue;
 229
 230                 /*
 231                  * Iterate over all pools, adding any matching pools to 'nvl'.
 232                  */
 233                 nvl = NULL;
 234                 while ((spa = spa_next(spa)) != NULL) {
 235                         /*
 236                          * Skip over our own pool if we're about to remove
 237                          * ourselves from the spa namespace or any pool that
 238                          * is readonly. Since we cannot guarantee that a
 239                          * readonly pool would successfully import upon reboot,
 240                          * we don't allow them to be written to the cache file.
 241                          */
 242                         if ((spa == target && removing) ||
 243                             !spa_writeable(spa))
 244                                 continue;
 245
 246                         mutex_enter(&spa->spa_props_lock);
 247                         tdp = list_head(&spa->spa_config_list);
 248                         if (spa->spa_config == NULL ||
 249                             tdp->scd_path == NULL ||
 250                             strcmp(tdp->scd_path, dp->scd_path) != 0) {
 251                                 mutex_exit(&spa->spa_props_lock);
 252                                 continue;
 253                         }
 254
 255                         if (nvl == NULL)
 256                                 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
 257                                     KM_SLEEP) == 0);
 258
 259                         if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
 260                                 VERIFY0(nvlist_lookup_string(spa->spa_config,
 261                                         ZPOOL_CONFIG_POOL_NAME, &pool_name));
 262                         } else
 263                                 pool_name = spa_name(spa);
 264
 265                         VERIFY(nvlist_add_nvlist(nvl, pool_name,
 266                             spa->spa_config) == 0);
 267                         mutex_exit(&spa->spa_props_lock);
 268                 }
 269
 270                 spa_config_write(dp, nvl);
 271                 nvlist_free(nvl);
 272         }
 273
 274         /*
 275          * Remove any config entries older than the current one.
 276          */
 277         dp = list_head(&target->spa_config_list);
 278         while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
 279                 list_remove(&target->spa_config_list, tdp);
 280                 if (tdp->scd_path != NULL)
 281                         spa_strfree(tdp->scd_path);
 282                 kmem_free(tdp, sizeof (spa_config_dirent_t));
 283         }
 284
 285         spa_config_generation++;
 286
 287         if (postsysevent)
 288                 spa_event_notify(target, NULL, FM_EREPORT_ZFS_CONFIG_SYNC);
 289 }
 290
 291 /*
 292  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
 293  * and we don't want to allow the local zone to see all the pools anyway.
 294  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
 295  * information for all pool visible within the zone.
 296  */
 297 nvlist_t *
 298 spa_all_configs(uint64_t *generation)
 299 {
 300         nvlist_t *pools;
 301         spa_t *spa = NULL;
 302
 303         if (*generation == spa_config_generation)
 304                 return (NULL);
 305
 306         VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 307
 308         mutex_enter(&spa_namespace_lock);
 309         while ((spa = spa_next(spa)) != NULL) {
 310                 if (INGLOBALZONE(curproc) ||
 311                     zone_dataset_visible(spa_name(spa), NULL)) {
 312                         mutex_enter(&spa->spa_props_lock);
 313                         VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
 314                             spa->spa_config) == 0);
 315                         mutex_exit(&spa->spa_props_lock);
 316                 }
 317         }
 318         *generation = spa_config_generation;
 319         mutex_exit(&spa_namespace_lock);
 320
 321         return (pools);
 322 }
 323
 324 void
 325 spa_config_set(spa_t *spa, nvlist_t *config)
 326 {
 327         mutex_enter(&spa->spa_props_lock);
 328         if (spa->spa_config != NULL)
 329                 nvlist_free(spa->spa_config);
 330         spa->spa_config = config;
 331         mutex_exit(&spa->spa_props_lock);
 332 }
 333
 334 /*
 335  * Generate the pool's configuration based on the current in-core state.
 336  *
 337  * We infer whether to generate a complete config or just one top-level config
 338  * based on whether vd is the root vdev.
 339  */
 340 nvlist_t *
 341 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 342 {
 343         nvlist_t *config, *nvroot;
 344         vdev_t *rvd = spa->spa_root_vdev;
 345         unsigned long hostid = 0;
 346         boolean_t locked = B_FALSE;
 347         uint64_t split_guid;
 348         char *pool_name;
 349
 350         if (vd == NULL) {
 351                 vd = rvd;
 352                 locked = B_TRUE;
 353                 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
 354         }
 355
 356         ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
 357             (SCL_CONFIG | SCL_STATE));
 358
 359         /*
 360          * If txg is -1, report the current value of spa->spa_config_txg.
 361          */
 362         if (txg == -1ULL)
 363                 txg = spa->spa_config_txg;
 364
 365         /*
 366          * Originally, users had to handle spa namespace collisions by either
 367          * exporting the already imported pool or by specifying a new name for
 368          * the pool with a conflicting name. In the case of root pools from
 369          * virtual guests, neither approach to collision resolution is
 370          * reasonable. This is addressed by extending the new name syntax with
 371          * an option to specify that the new name is temporary. When specified,
 372          * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
 373          * to use the previous name, which we do below.
 374          */
 375         if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
 376                 VERIFY0(nvlist_lookup_string(spa->spa_config,
 377                         ZPOOL_CONFIG_POOL_NAME, &pool_name));
 378         } else
 379                 pool_name = spa_name(spa);
 380
 381         VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 382
 383         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
 384             spa_version(spa)) == 0);
 385         VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
 386             pool_name) == 0);
 387         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 388             spa_state(spa)) == 0);
 389         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 390             txg) == 0);
 391         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 392             spa_guid(spa)) == 0);
 393         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
 394             spa->spa_errata) == 0);
 395         VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
 396             ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
 397
 398
 399 #ifdef  _KERNEL
 400         hostid = zone_get_hostid(NULL);
 401 #else   /* _KERNEL */
 402         /*
 403          * We're emulating the system's hostid in userland, so we can't use
 404          * zone_get_hostid().
 405          */
 406         (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
 407 #endif  /* _KERNEL */
 408         if (hostid != 0) {
 409                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 410                     hostid) == 0);
 411         }
 412         VERIFY0(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 413             utsname()->nodename));
 414
 415         if (vd != rvd) {
 416                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 417                     vd->vdev_top->vdev_guid) == 0);
 418                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
 419                     vd->vdev_guid) == 0);
 420                 if (vd->vdev_isspare)
 421                         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
 422                             1ULL) == 0);
 423                 if (vd->vdev_islog)
 424                         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
 425                             1ULL) == 0);
 426                 vd = vd->vdev_top;              /* label contains top config */
 427         } else {
 428                 /*
 429                  * Only add the (potentially large) split information
 430                  * in the mos config, and not in the vdev labels
 431                  */
 432                 if (spa->spa_config_splitting != NULL)
 433                         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
 434                             spa->spa_config_splitting) == 0);
 435         }
 436
 437         /*
 438          * Add the top-level config.  We even add this on pools which
 439          * don't support holes in the namespace.
 440          */
 441         vdev_top_config_generate(spa, config);
 442
 443         /*
 444          * If we're splitting, record the original pool's guid.
 445          */
 446         if (spa->spa_config_splitting != NULL &&
 447             nvlist_lookup_uint64(spa->spa_config_splitting,
 448             ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
 449                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
 450                     split_guid) == 0);
 451         }
 452
 453         nvroot = vdev_config_generate(spa, vd, getstats, 0);
 454         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
 455         nvlist_free(nvroot);
 456
 457         /*
 458          * Store what's necessary for reading the MOS in the label.
 459          */
 460         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
 461             spa->spa_label_features) == 0);
 462
 463         if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
 464                 ddt_histogram_t *ddh;
 465                 ddt_stat_t *dds;
 466                 ddt_object_t *ddo;
 467
 468                 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
 469                 ddt_get_dedup_histogram(spa, ddh);
 470                 VERIFY(nvlist_add_uint64_array(config,
 471                     ZPOOL_CONFIG_DDT_HISTOGRAM,
 472                     (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
 473                 kmem_free(ddh, sizeof (ddt_histogram_t));
 474
 475                 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
 476                 ddt_get_dedup_object_stats(spa, ddo);
 477                 VERIFY(nvlist_add_uint64_array(config,
 478                     ZPOOL_CONFIG_DDT_OBJ_STATS,
 479                     (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
 480                 kmem_free(ddo, sizeof (ddt_object_t));
 481
 482                 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
 483                 ddt_get_dedup_stats(spa, dds);
 484                 VERIFY(nvlist_add_uint64_array(config,
 485                     ZPOOL_CONFIG_DDT_STATS,
 486                     (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
 487                 kmem_free(dds, sizeof (ddt_stat_t));
 488         }
 489
 490         if (locked)
 491                 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
 492
 493         return (config);
 494 }
 495
 496 /*
 497  * Update all disk labels, generate a fresh config based on the current
 498  * in-core state, and sync the global config cache (do not sync the config
 499  * cache if this is a booting rootpool).
 500  */
 501 void
 502 spa_config_update(spa_t *spa, int what)
 503 {
 504         vdev_t *rvd = spa->spa_root_vdev;
 505         uint64_t txg;
 506         int c;
 507
 508         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 509
 510         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 511         txg = spa_last_synced_txg(spa) + 1;
 512         if (what == SPA_CONFIG_UPDATE_POOL) {
 513                 vdev_config_dirty(rvd);
 514         } else {
 515                 /*
 516                  * If we have top-level vdevs that were added but have
 517                  * not yet been prepared for allocation, do that now.
 518                  * (It's safe now because the config cache is up to date,
 519                  * so it will be able to translate the new DVAs.)
 520                  * See comments in spa_vdev_add() for full details.
 521                  */
 522                 for (c = 0; c < rvd->vdev_children; c++) {
 523                         vdev_t *tvd = rvd->vdev_child[c];
 524                         if (tvd->vdev_ms_array == 0)
 525                                 vdev_metaslab_set_size(tvd);
 526                         vdev_expand(tvd, txg);
 527                 }
 528         }
 529         spa_config_exit(spa, SCL_ALL, FTAG);
 530
 531         /*
 532          * Wait for the mosconfig to be regenerated and synced.
 533          */
 534         txg_wait_synced(spa->spa_dsl_pool, txg);
 535
 536         /*
 537          * Update the global config cache to reflect the new mosconfig.
 538          */
 539         if (!spa->spa_is_root)
 540                 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
 541
 542         if (what == SPA_CONFIG_UPDATE_POOL)
 543                 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
 544 }
 545
 546 #if defined(_KERNEL) && defined(HAVE_SPL)
 547 EXPORT_SYMBOL(spa_config_sync);
 548 EXPORT_SYMBOL(spa_config_load);
 549 EXPORT_SYMBOL(spa_all_configs);
 550 EXPORT_SYMBOL(spa_config_set);
 551 EXPORT_SYMBOL(spa_config_generate);
 552 EXPORT_SYMBOL(spa_config_update);
 553
 554 module_param(spa_config_path, charp, 0444);
 555 MODULE_PARM_DESC(spa_config_path, "SPA config file (/etc/zfs/zpool.cache)");
 556
 557 module_param(zfs_autoimport_disable, int, 0644);
 558 MODULE_PARM_DESC(zfs_autoimport_disable, "Disable pool import at module load");
 559
 560 #endif