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.
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.
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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
26 #include <sys/zfs_context.h>
28 #include <sys/spa_impl.h>
29 #include <sys/nvpair.h>
31 #include <sys/fs/zfs.h>
32 #include <sys/vdev_impl.h>
33 #include <sys/zfs_ioctl.h>
34 #include <sys/utsname.h>
35 #include <sys/sunddi.h>
42 * Pool configuration repository.
44 * Pool configuration is stored as a packed nvlist on the filesystem. By
45 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
46 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
47 * property set that allows them to be stored in an alternate location until
48 * the control of external software.
50 * For each cache file, we have a single nvlist which holds all the
51 * configuration information. When the module loads, we read this information
52 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
53 * maintained independently in spa.c. Whenever the namespace is modified, or
54 * the configuration of a pool is changed, we call spa_config_sync(), which
55 * walks through all the active pools and writes the configuration to disk.
58 static uint64_t spa_config_generation = 1;
61 * This can be overridden in userland to preserve an alternate namespace for
62 * userland pools when doing testing.
64 const char *spa_config_path = ZPOOL_CACHE;
67 * Called when the module is first loaded, this routine loads the configuration
68 * file into the SPA namespace. It does not actually open or load the pools; it
69 * only populates the namespace.
75 nvlist_t *nvlist, *child;
82 * Open the configuration file.
84 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
86 (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path);
88 file = kobj_open_file(pathname);
90 kmem_free(pathname, MAXPATHLEN);
92 if (file == (struct _buf *)-1)
95 if (kobj_get_filesize(file, &fsize) != 0)
98 buf = kmem_alloc(fsize, KM_SLEEP);
101 * Read the nvlist from the file.
103 if (kobj_read_file(file, buf, fsize, 0) < 0)
109 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
113 * Iterate over all elements in the nvlist, creating a new spa_t for
114 * each one with the specified configuration.
116 mutex_enter(&spa_namespace_lock);
118 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
119 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
122 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
124 if (spa_lookup(nvpair_name(nvpair)) != NULL)
126 (void) spa_add(nvpair_name(nvpair), child, NULL);
128 mutex_exit(&spa_namespace_lock);
134 kmem_free(buf, fsize);
136 kobj_close_file(file);
140 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
145 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
149 * If the nvlist is empty (NULL), then remove the old cachefile.
152 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
157 * Pack the configuration into a buffer.
159 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
161 buf = kmem_alloc(buflen, KM_SLEEP);
162 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
164 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
168 * Write the configuration to disk. We need to do the traditional
169 * 'write to temporary file, sync, move over original' to make sure we
170 * always have a consistent view of the data.
172 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
174 if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
175 if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
176 0, RLIM64_INFINITY, kcred, NULL) == 0 &&
177 VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
178 (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
180 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
183 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
185 kmem_free(buf, buflen);
186 kmem_free(temp, MAXPATHLEN);
190 * Synchronize pool configuration to disk. This must be called with the
191 * namespace lock held.
194 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
196 spa_config_dirent_t *dp, *tdp;
199 ASSERT(MUTEX_HELD(&spa_namespace_lock));
201 if (rootdir == NULL || !(spa_mode_global & FWRITE))
205 * Iterate over all cachefiles for the pool, past or present. When the
206 * cachefile is changed, the new one is pushed onto this list, allowing
207 * us to update previous cachefiles that no longer contain this pool.
209 for (dp = list_head(&target->spa_config_list); dp != NULL;
210 dp = list_next(&target->spa_config_list, dp)) {
212 if (dp->scd_path == NULL)
216 * Iterate over all pools, adding any matching pools to 'nvl'.
219 while ((spa = spa_next(spa)) != NULL) {
220 if (spa == target && removing)
223 mutex_enter(&spa->spa_props_lock);
224 tdp = list_head(&spa->spa_config_list);
225 if (spa->spa_config == NULL ||
226 tdp->scd_path == NULL ||
227 strcmp(tdp->scd_path, dp->scd_path) != 0) {
228 mutex_exit(&spa->spa_props_lock);
233 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
236 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
237 spa->spa_config) == 0);
238 mutex_exit(&spa->spa_props_lock);
241 spa_config_write(dp, nvl);
246 * Remove any config entries older than the current one.
248 dp = list_head(&target->spa_config_list);
249 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
250 list_remove(&target->spa_config_list, tdp);
251 if (tdp->scd_path != NULL)
252 spa_strfree(tdp->scd_path);
253 kmem_free(tdp, sizeof (spa_config_dirent_t));
256 spa_config_generation++;
259 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
263 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
264 * and we don't want to allow the local zone to see all the pools anyway.
265 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
266 * information for all pool visible within the zone.
269 spa_all_configs(uint64_t *generation)
274 if (*generation == spa_config_generation)
277 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
279 mutex_enter(&spa_namespace_lock);
280 while ((spa = spa_next(spa)) != NULL) {
281 if (INGLOBALZONE(curthread) ||
282 zone_dataset_visible(spa_name(spa), NULL)) {
283 mutex_enter(&spa->spa_props_lock);
284 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
285 spa->spa_config) == 0);
286 mutex_exit(&spa->spa_props_lock);
289 *generation = spa_config_generation;
290 mutex_exit(&spa_namespace_lock);
296 spa_config_set(spa_t *spa, nvlist_t *config)
298 mutex_enter(&spa->spa_props_lock);
299 if (spa->spa_config != NULL)
300 nvlist_free(spa->spa_config);
301 spa->spa_config = config;
302 mutex_exit(&spa->spa_props_lock);
306 * Generate the pool's configuration based on the current in-core state.
307 * We infer whether to generate a complete config or just one top-level config
308 * based on whether vd is the root vdev.
311 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
313 nvlist_t *config, *nvroot;
314 vdev_t *rvd = spa->spa_root_vdev;
315 unsigned long hostid = 0;
316 boolean_t locked = B_FALSE;
322 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
325 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
326 (SCL_CONFIG | SCL_STATE));
329 * If txg is -1, report the current value of spa->spa_config_txg.
332 txg = spa->spa_config_txg;
334 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
336 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
337 spa_version(spa)) == 0);
338 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
339 spa_name(spa)) == 0);
340 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
341 spa_state(spa)) == 0);
342 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
344 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
345 spa_guid(spa)) == 0);
347 hostid = zone_get_hostid(NULL);
350 * We're emulating the system's hostid in userland, so we can't use
353 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
356 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
359 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
360 utsname.nodename) == 0);
363 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
364 vd->vdev_top->vdev_guid) == 0);
365 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
366 vd->vdev_guid) == 0);
367 if (vd->vdev_isspare)
368 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
371 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
373 vd = vd->vdev_top; /* label contains top config */
376 * Only add the (potentially large) split information
377 * in the mos config, and not in the vdev labels
379 if (spa->spa_config_splitting != NULL)
380 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
381 spa->spa_config_splitting) == 0);
385 * Add the top-level config. We even add this on pools which
386 * don't support holes in the namespace.
388 vdev_top_config_generate(spa, config);
391 * If we're splitting, record the original pool's guid.
393 if (spa->spa_config_splitting != NULL &&
394 nvlist_lookup_uint64(spa->spa_config_splitting,
395 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
396 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
400 nvroot = vdev_config_generate(spa, vd, getstats, 0);
401 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
404 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
405 ddt_histogram_t *ddh;
409 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
410 ddt_get_dedup_histogram(spa, ddh);
411 VERIFY(nvlist_add_uint64_array(config,
412 ZPOOL_CONFIG_DDT_HISTOGRAM,
413 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
414 kmem_free(ddh, sizeof (ddt_histogram_t));
416 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
417 ddt_get_dedup_object_stats(spa, ddo);
418 VERIFY(nvlist_add_uint64_array(config,
419 ZPOOL_CONFIG_DDT_OBJ_STATS,
420 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
421 kmem_free(ddo, sizeof (ddt_object_t));
423 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
424 ddt_get_dedup_stats(spa, dds);
425 VERIFY(nvlist_add_uint64_array(config,
426 ZPOOL_CONFIG_DDT_STATS,
427 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
428 kmem_free(dds, sizeof (ddt_stat_t));
432 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
438 * Update all disk labels, generate a fresh config based on the current
439 * in-core state, and sync the global config cache (do not sync the config
440 * cache if this is a booting rootpool).
443 spa_config_update(spa_t *spa, int what)
445 vdev_t *rvd = spa->spa_root_vdev;
449 ASSERT(MUTEX_HELD(&spa_namespace_lock));
451 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
452 txg = spa_last_synced_txg(spa) + 1;
453 if (what == SPA_CONFIG_UPDATE_POOL) {
454 vdev_config_dirty(rvd);
457 * If we have top-level vdevs that were added but have
458 * not yet been prepared for allocation, do that now.
459 * (It's safe now because the config cache is up to date,
460 * so it will be able to translate the new DVAs.)
461 * See comments in spa_vdev_add() for full details.
463 for (c = 0; c < rvd->vdev_children; c++) {
464 vdev_t *tvd = rvd->vdev_child[c];
465 if (tvd->vdev_ms_array == 0)
466 vdev_metaslab_set_size(tvd);
467 vdev_expand(tvd, txg);
470 spa_config_exit(spa, SCL_ALL, FTAG);
473 * Wait for the mosconfig to be regenerated and synced.
475 txg_wait_synced(spa->spa_dsl_pool, txg);
478 * Update the global config cache to reflect the new mosconfig.
480 if (!spa->spa_is_root)
481 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
483 if (what == SPA_CONFIG_UPDATE_POOL)
484 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);