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) 2012 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/utsname.h>
  36 #include <sys/systeminfo.h>
  37 #include <sys/sunddi.h>
  38 #include <sys/zfeature.h>
  39 #ifdef _KERNEL
  40 #include <sys/kobj.h>
  41 #include <sys/zone.h>
  42 #endif
  43
  44 /*
  45  * Pool configuration repository.
  46  *
  47  * Pool configuration is stored as a packed nvlist on the filesystem.  By
  48  * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
  49  * (when the ZFS module is loaded).  Pools can also have the 'cachefile'
  50  * property set that allows them to be stored in an alternate location until
  51  * the control of external software.
  52  *
  53  * For each cache file, we have a single nvlist which holds all the
  54  * configuration information.  When the module loads, we read this information
  55  * from /etc/zfs/zpool.cache and populate the SPA namespace.  This namespace is
  56  * maintained independently in spa.c.  Whenever the namespace is modified, or
  57  * the configuration of a pool is changed, we call spa_config_sync(), which
  58  * walks through all the active pools and writes the configuration to disk.
  59  */
  60
  61 static uint64_t spa_config_generation = 1;
  62
  63 /*
  64  * This can be overridden in userland to preserve an alternate namespace for
  65  * userland pools when doing testing.
  66  */
  67 char *spa_config_path = ZPOOL_CACHE;
  68 int zfs_autoimport_disable = 0;
  69
  70 /*
  71  * Called when the module is first loaded, this routine loads the configuration
  72  * file into the SPA namespace.  It does not actually open or load the pools; it
  73  * only populates the namespace.
  74  */
  75 void
  76 spa_config_load(void)
  77 {
  78         void *buf = NULL;
  79         nvlist_t *nvlist, *child;
  80         nvpair_t *nvpair;
  81         char *pathname;
  82         struct _buf *file;
  83         uint64_t fsize;
  84
  85         if (zfs_autoimport_disable)
  86                 return;
  87
  88         /*
  89          * Open the configuration file.
  90          */
  91         pathname = kmem_alloc(MAXPATHLEN, KM_PUSHPAGE);
  92
  93         (void) snprintf(pathname, MAXPATHLEN, "%s%s",
  94             (rootdir != NULL) ? "./" : "", spa_config_path);
  95
  96         file = kobj_open_file(pathname);
  97
  98         kmem_free(pathname, MAXPATHLEN);
  99
 100         if (file == (struct _buf *)-1)
 101                 return;
 102
 103         if (kobj_get_filesize(file, &fsize) != 0)
 104                 goto out;
 105
 106         buf = kmem_alloc(fsize, KM_PUSHPAGE | KM_NODEBUG);
 107
 108         /*
 109          * Read the nvlist from the file.
 110          */
 111         if (kobj_read_file(file, buf, fsize, 0) < 0)
 112                 goto out;
 113
 114         /*
 115          * Unpack the nvlist.
 116          */
 117         if (nvlist_unpack(buf, fsize, &nvlist, KM_PUSHPAGE) != 0)
 118                 goto out;
 119
 120         /*
 121          * Iterate over all elements in the nvlist, creating a new spa_t for
 122          * each one with the specified configuration.
 123          */
 124         mutex_enter(&spa_namespace_lock);
 125         nvpair = NULL;
 126         while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
 127                 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
 128                         continue;
 129
 130                 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
 131
 132                 if (spa_lookup(nvpair_name(nvpair)) != NULL)
 133                         continue;
 134                 (void) spa_add(nvpair_name(nvpair), child, NULL);
 135         }
 136         mutex_exit(&spa_namespace_lock);
 137
 138         nvlist_free(nvlist);
 139
 140 out:
 141         if (buf != NULL)
 142                 kmem_free(buf, fsize);
 143
 144         kobj_close_file(file);
 145 }
 146
 147 static void
 148 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 149 {
 150         size_t buflen;
 151         char *buf;
 152         vnode_t *vp;
 153         int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
 154         char *temp;
 155
 156         /*
 157          * If the nvlist is empty (NULL), then remove the old cachefile.
 158          */
 159         if (nvl == NULL) {
 160                 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
 161                 return;
 162         }
 163
 164         /*
 165          * Pack the configuration into a buffer.
 166          */
 167         VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
 168
 169         buf = kmem_alloc(buflen, KM_PUSHPAGE | KM_NODEBUG);
 170         temp = kmem_zalloc(MAXPATHLEN, KM_PUSHPAGE);
 171
 172         VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
 173             KM_PUSHPAGE) == 0);
 174
 175         /*
 176          * Write the configuration to disk.  We need to do the traditional
 177          * 'write to temporary file, sync, move over original' to make sure we
 178          * always have a consistent view of the data.
 179          */
 180         (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
 181
 182         if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
 183                 if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
 184                     0, RLIM64_INFINITY, kcred, NULL) == 0 &&
 185                     VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
 186                         (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
 187                 }
 188                 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
 189         }
 190
 191         (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
 192
 193         kmem_free(buf, buflen);
 194         kmem_free(temp, MAXPATHLEN);
 195 }
 196
 197 /*
 198  * Synchronize pool configuration to disk.  This must be called with the
 199  * namespace lock held.
 200  */
 201 void
 202 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
 203 {
 204         spa_config_dirent_t *dp, *tdp;
 205         nvlist_t *nvl;
 206
 207         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 208
 209         if (rootdir == NULL || !(spa_mode_global & FWRITE))
 210                 return;
 211
 212         /*
 213          * Iterate over all cachefiles for the pool, past or present.  When the
 214          * cachefile is changed, the new one is pushed onto this list, allowing
 215          * us to update previous cachefiles that no longer contain this pool.
 216          */
 217         for (dp = list_head(&target->spa_config_list); dp != NULL;
 218             dp = list_next(&target->spa_config_list, dp)) {
 219                 spa_t *spa = NULL;
 220                 if (dp->scd_path == NULL)
 221                         continue;
 222
 223                 /*
 224                  * Iterate over all pools, adding any matching pools to 'nvl'.
 225                  */
 226                 nvl = NULL;
 227                 while ((spa = spa_next(spa)) != NULL) {
 228                         /*
 229                          * Skip over our own pool if we're about to remove
 230                          * ourselves from the spa namespace or any pool that
 231                          * is readonly. Since we cannot guarantee that a
 232                          * readonly pool would successfully import upon reboot,
 233                          * we don't allow them to be written to the cache file.
 234                          */
 235                         if ((spa == target && removing) ||
 236                             !spa_writeable(spa))
 237                                 continue;
 238
 239                         mutex_enter(&spa->spa_props_lock);
 240                         tdp = list_head(&spa->spa_config_list);
 241                         if (spa->spa_config == NULL ||
 242                             tdp->scd_path == NULL ||
 243                             strcmp(tdp->scd_path, dp->scd_path) != 0) {
 244                                 mutex_exit(&spa->spa_props_lock);
 245                                 continue;
 246                         }
 247
 248                         if (nvl == NULL)
 249                                 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
 250                                     KM_PUSHPAGE) == 0);
 251
 252                         VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
 253                             spa->spa_config) == 0);
 254                         mutex_exit(&spa->spa_props_lock);
 255                 }
 256
 257                 spa_config_write(dp, nvl);
 258                 nvlist_free(nvl);
 259         }
 260
 261         /*
 262          * Remove any config entries older than the current one.
 263          */
 264         dp = list_head(&target->spa_config_list);
 265         while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
 266                 list_remove(&target->spa_config_list, tdp);
 267                 if (tdp->scd_path != NULL)
 268                         spa_strfree(tdp->scd_path);
 269                 kmem_free(tdp, sizeof (spa_config_dirent_t));
 270         }
 271
 272         spa_config_generation++;
 273
 274         if (postsysevent)
 275                 spa_event_notify(target, NULL, FM_EREPORT_ZFS_CONFIG_SYNC);
 276 }
 277
 278 /*
 279  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
 280  * and we don't want to allow the local zone to see all the pools anyway.
 281  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
 282  * information for all pool visible within the zone.
 283  */
 284 nvlist_t *
 285 spa_all_configs(uint64_t *generation)
 286 {
 287         nvlist_t *pools;
 288         spa_t *spa = NULL;
 289
 290         if (*generation == spa_config_generation)
 291                 return (NULL);
 292
 293         VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
 294
 295         mutex_enter(&spa_namespace_lock);
 296         while ((spa = spa_next(spa)) != NULL) {
 297                 if (INGLOBALZONE(curproc) ||
 298                     zone_dataset_visible(spa_name(spa), NULL)) {
 299                         mutex_enter(&spa->spa_props_lock);
 300                         VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
 301                             spa->spa_config) == 0);
 302                         mutex_exit(&spa->spa_props_lock);
 303                 }
 304         }
 305         *generation = spa_config_generation;
 306         mutex_exit(&spa_namespace_lock);
 307
 308         return (pools);
 309 }
 310
 311 void
 312 spa_config_set(spa_t *spa, nvlist_t *config)
 313 {
 314         mutex_enter(&spa->spa_props_lock);
 315         if (spa->spa_config != NULL)
 316                 nvlist_free(spa->spa_config);
 317         spa->spa_config = config;
 318         mutex_exit(&spa->spa_props_lock);
 319 }
 320
 321 /*
 322  * Generate the pool's configuration based on the current in-core state.
 323  * We infer whether to generate a complete config or just one top-level config
 324  * based on whether vd is the root vdev.
 325  */
 326 nvlist_t *
 327 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 328 {
 329         nvlist_t *config, *nvroot;
 330         vdev_t *rvd = spa->spa_root_vdev;
 331         unsigned long hostid = 0;
 332         boolean_t locked = B_FALSE;
 333         uint64_t split_guid;
 334
 335         if (vd == NULL) {
 336                 vd = rvd;
 337                 locked = B_TRUE;
 338                 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
 339         }
 340
 341         ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
 342             (SCL_CONFIG | SCL_STATE));
 343
 344         /*
 345          * If txg is -1, report the current value of spa->spa_config_txg.
 346          */
 347         if (txg == -1ULL)
 348                 txg = spa->spa_config_txg;
 349
 350         VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
 351
 352         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
 353             spa_version(spa)) == 0);
 354         VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
 355             spa_name(spa)) == 0);
 356         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 357             spa_state(spa)) == 0);
 358         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 359             txg) == 0);
 360         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 361             spa_guid(spa)) == 0);
 362         VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
 363             ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
 364
 365
 366 #ifdef  _KERNEL
 367         hostid = zone_get_hostid(NULL);
 368 #else   /* _KERNEL */
 369         /*
 370          * We're emulating the system's hostid in userland, so we can't use
 371          * zone_get_hostid().
 372          */
 373         (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
 374 #endif  /* _KERNEL */
 375         if (hostid != 0) {
 376                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 377                     hostid) == 0);
 378         }
 379         VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 380             utsname.nodename) == 0);
 381
 382         if (vd != rvd) {
 383                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 384                     vd->vdev_top->vdev_guid) == 0);
 385                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
 386                     vd->vdev_guid) == 0);
 387                 if (vd->vdev_isspare)
 388                         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
 389                             1ULL) == 0);
 390                 if (vd->vdev_islog)
 391                         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
 392                             1ULL) == 0);
 393                 vd = vd->vdev_top;              /* label contains top config */
 394         } else {
 395                 /*
 396                  * Only add the (potentially large) split information
 397                  * in the mos config, and not in the vdev labels
 398                  */
 399                 if (spa->spa_config_splitting != NULL)
 400                         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
 401                             spa->spa_config_splitting) == 0);
 402         }
 403
 404         /*
 405          * Add the top-level config.  We even add this on pools which
 406          * don't support holes in the namespace.
 407          */
 408         vdev_top_config_generate(spa, config);
 409
 410         /*
 411          * If we're splitting, record the original pool's guid.
 412          */
 413         if (spa->spa_config_splitting != NULL &&
 414             nvlist_lookup_uint64(spa->spa_config_splitting,
 415             ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
 416                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
 417                     split_guid) == 0);
 418         }
 419
 420         nvroot = vdev_config_generate(spa, vd, getstats, 0);
 421         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
 422         nvlist_free(nvroot);
 423
 424         /*
 425          * Store what's necessary for reading the MOS in the label.
 426          */
 427         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
 428             spa->spa_label_features) == 0);
 429
 430         if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
 431                 ddt_histogram_t *ddh;
 432                 ddt_stat_t *dds;
 433                 ddt_object_t *ddo;
 434
 435                 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_PUSHPAGE);
 436                 ddt_get_dedup_histogram(spa, ddh);
 437                 VERIFY(nvlist_add_uint64_array(config,
 438                     ZPOOL_CONFIG_DDT_HISTOGRAM,
 439                     (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
 440                 kmem_free(ddh, sizeof (ddt_histogram_t));
 441
 442                 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_PUSHPAGE);
 443                 ddt_get_dedup_object_stats(spa, ddo);
 444                 VERIFY(nvlist_add_uint64_array(config,
 445                     ZPOOL_CONFIG_DDT_OBJ_STATS,
 446                     (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
 447                 kmem_free(ddo, sizeof (ddt_object_t));
 448
 449                 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_PUSHPAGE);
 450                 ddt_get_dedup_stats(spa, dds);
 451                 VERIFY(nvlist_add_uint64_array(config,
 452                     ZPOOL_CONFIG_DDT_STATS,
 453                     (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
 454                 kmem_free(dds, sizeof (ddt_stat_t));
 455         }
 456
 457         if (locked)
 458                 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
 459
 460         return (config);
 461 }
 462
 463 /*
 464  * Update all disk labels, generate a fresh config based on the current
 465  * in-core state, and sync the global config cache (do not sync the config
 466  * cache if this is a booting rootpool).
 467  */
 468 void
 469 spa_config_update(spa_t *spa, int what)
 470 {
 471         vdev_t *rvd = spa->spa_root_vdev;
 472         uint64_t txg;
 473         int c;
 474
 475         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 476
 477         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 478         txg = spa_last_synced_txg(spa) + 1;
 479         if (what == SPA_CONFIG_UPDATE_POOL) {
 480                 vdev_config_dirty(rvd);
 481         } else {
 482                 /*
 483                  * If we have top-level vdevs that were added but have
 484                  * not yet been prepared for allocation, do that now.
 485                  * (It's safe now because the config cache is up to date,
 486                  * so it will be able to translate the new DVAs.)
 487                  * See comments in spa_vdev_add() for full details.
 488                  */
 489                 for (c = 0; c < rvd->vdev_children; c++) {
 490                         vdev_t *tvd = rvd->vdev_child[c];
 491                         if (tvd->vdev_ms_array == 0)
 492                                 vdev_metaslab_set_size(tvd);
 493                         vdev_expand(tvd, txg);
 494                 }
 495         }
 496         spa_config_exit(spa, SCL_ALL, FTAG);
 497
 498         /*
 499          * Wait for the mosconfig to be regenerated and synced.
 500          */
 501         txg_wait_synced(spa->spa_dsl_pool, txg);
 502
 503         /*
 504          * Update the global config cache to reflect the new mosconfig.
 505          */
 506         if (!spa->spa_is_root)
 507                 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
 508
 509         if (what == SPA_CONFIG_UPDATE_POOL)
 510                 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
 511 }
 512
 513 #if defined(_KERNEL) && defined(HAVE_SPL)
 514 EXPORT_SYMBOL(spa_config_sync);
 515 EXPORT_SYMBOL(spa_config_load);
 516 EXPORT_SYMBOL(spa_all_configs);
 517 EXPORT_SYMBOL(spa_config_set);
 518 EXPORT_SYMBOL(spa_config_generate);
 519 EXPORT_SYMBOL(spa_config_update);
 520
 521 module_param(spa_config_path, charp, 0444);
 522 MODULE_PARM_DESC(spa_config_path, "SPA config file (/etc/zfs/zpool.cache)");
 523
 524 module_param(zfs_autoimport_disable, int, 0644);
 525 MODULE_PARM_DESC(zfs_autoimport_disable, "Disable pool import at module load");
 526
 527 #endif