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.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
28 #include <sys/zfs_context.h>
30 #include <sys/fm/fs/zfs.h>
31 #include <sys/spa_impl.h>
32 #include <sys/nvpair.h>
34 #include <sys/fs/zfs.h>
35 #include <sys/vdev_impl.h>
36 #include <sys/zfs_ioctl.h>
37 #include <sys/utsname.h>
38 #include <sys/sunddi.h>
39 #include <sys/zfeature.h>
46 * Pool configuration repository.
48 * Pool configuration is stored as a packed nvlist on the filesystem. By
49 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
50 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
51 * property set that allows them to be stored in an alternate location until
52 * the control of external software.
54 * For each cache file, we have a single nvlist which holds all the
55 * configuration information. When the module loads, we read this information
56 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
57 * maintained independently in spa.c. Whenever the namespace is modified, or
58 * the configuration of a pool is changed, we call spa_config_sync(), which
59 * walks through all the active pools and writes the configuration to disk.
62 static uint64_t spa_config_generation = 1;
65 * This can be overridden in userland to preserve an alternate namespace for
66 * userland pools when doing testing.
68 const char *spa_config_path = ZPOOL_CACHE;
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.
79 nvlist_t *nvlist, *child;
86 * Open the configuration file.
88 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
90 (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path);
92 file = kobj_open_file(pathname);
94 kmem_free(pathname, MAXPATHLEN);
96 if (file == (struct _buf *)-1)
99 if (kobj_get_filesize(file, &fsize) != 0)
102 buf = kmem_alloc(fsize, KM_SLEEP);
105 * Read the nvlist from the file.
107 if (kobj_read_file(file, buf, fsize, 0) < 0)
113 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
117 * Iterate over all elements in the nvlist, creating a new spa_t for
118 * each one with the specified configuration.
120 mutex_enter(&spa_namespace_lock);
122 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
123 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
126 child = fnvpair_value_nvlist(nvpair);
128 if (spa_lookup(nvpair_name(nvpair)) != NULL)
130 (void) spa_add(nvpair_name(nvpair), child, NULL);
132 mutex_exit(&spa_namespace_lock);
138 kmem_free(buf, fsize);
140 kobj_close_file(file);
144 spa_config_clean(nvlist_t *nvl)
147 nvlist_t *nvroot = NULL;
150 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
152 for (c = 0; c < children; c++)
153 spa_config_clean(child[c]);
156 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0)
157 spa_config_clean(nvroot);
159 nvlist_remove(nvl, ZPOOL_CONFIG_VDEV_STATS, DATA_TYPE_UINT64_ARRAY);
160 nvlist_remove(nvl, ZPOOL_CONFIG_SCAN_STATS, DATA_TYPE_UINT64_ARRAY);
164 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
169 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
174 * If the nvlist is empty (NULL), then remove the old cachefile.
177 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
182 * Pack the configuration into a buffer.
184 buf = fnvlist_pack(nvl, &buflen);
185 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
188 * Write the configuration to disk. We need to do the traditional
189 * 'write to temporary file, sync, move over original' to make sure we
190 * always have a consistent view of the data.
192 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
194 err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
196 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
197 0, RLIM64_INFINITY, kcred, NULL);
199 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
201 err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
202 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
205 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
207 fnvlist_pack_free(buf, buflen);
208 kmem_free(temp, MAXPATHLEN);
213 * Synchronize pool configuration to disk. This must be called with the
214 * namespace lock held. Synchronizing the pool cache is typically done after
215 * the configuration has been synced to the MOS. This exposes a window where
216 * the MOS config will have been updated but the cache file has not. If
217 * the system were to crash at that instant then the cached config may not
218 * contain the correct information to open the pool and an explicity import
222 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
224 spa_config_dirent_t *dp, *tdp;
226 boolean_t ccw_failure;
229 ASSERT(MUTEX_HELD(&spa_namespace_lock));
231 if (rootdir == NULL || !(spa_mode_global & FWRITE))
235 * Iterate over all cachefiles for the pool, past or present. When the
236 * cachefile is changed, the new one is pushed onto this list, allowing
237 * us to update previous cachefiles that no longer contain this pool.
239 ccw_failure = B_FALSE;
240 for (dp = list_head(&target->spa_config_list); dp != NULL;
241 dp = list_next(&target->spa_config_list, dp)) {
243 if (dp->scd_path == NULL)
247 * Iterate over all pools, adding any matching pools to 'nvl'.
250 while ((spa = spa_next(spa)) != NULL) {
251 nvlist_t *nvroot = NULL;
253 * Skip over our own pool if we're about to remove
254 * ourselves from the spa namespace or any pool that
255 * is readonly. Since we cannot guarantee that a
256 * readonly pool would successfully import upon reboot,
257 * we don't allow them to be written to the cache file.
259 if ((spa == target && removing) ||
260 (spa_state(spa) == POOL_STATE_ACTIVE &&
261 !spa_writeable(spa)))
264 mutex_enter(&spa->spa_props_lock);
265 tdp = list_head(&spa->spa_config_list);
266 if (spa->spa_config == NULL ||
267 tdp->scd_path == NULL ||
268 strcmp(tdp->scd_path, dp->scd_path) != 0) {
269 mutex_exit(&spa->spa_props_lock);
274 nvl = fnvlist_alloc();
276 fnvlist_add_nvlist(nvl, spa->spa_name,
278 mutex_exit(&spa->spa_props_lock);
280 if (nvlist_lookup_nvlist(nvl, spa->spa_name, &nvroot) == 0)
281 spa_config_clean(nvroot);
284 error = spa_config_write(dp, nvl);
286 ccw_failure = B_TRUE;
292 * Keep trying so that configuration data is
293 * written if/when any temporary filesystem
294 * resource issues are resolved.
296 if (target->spa_ccw_fail_time == 0) {
297 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
298 target, NULL, NULL, 0, 0);
300 target->spa_ccw_fail_time = gethrtime();
301 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
304 * Do not rate limit future attempts to update
307 target->spa_ccw_fail_time = 0;
311 * Remove any config entries older than the current one.
313 dp = list_head(&target->spa_config_list);
314 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
315 list_remove(&target->spa_config_list, tdp);
316 if (tdp->scd_path != NULL)
317 spa_strfree(tdp->scd_path);
318 kmem_free(tdp, sizeof (spa_config_dirent_t));
321 spa_config_generation++;
324 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
328 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
329 * and we don't want to allow the local zone to see all the pools anyway.
330 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
331 * information for all pool visible within the zone.
334 spa_all_configs(uint64_t *generation)
339 if (*generation == spa_config_generation)
342 pools = fnvlist_alloc();
344 mutex_enter(&spa_namespace_lock);
345 while ((spa = spa_next(spa)) != NULL) {
346 if (INGLOBALZONE(curthread) ||
347 zone_dataset_visible(spa_name(spa), NULL)) {
348 mutex_enter(&spa->spa_props_lock);
349 fnvlist_add_nvlist(pools, spa_name(spa),
351 mutex_exit(&spa->spa_props_lock);
354 *generation = spa_config_generation;
355 mutex_exit(&spa_namespace_lock);
361 spa_config_set(spa_t *spa, nvlist_t *config)
363 mutex_enter(&spa->spa_props_lock);
364 nvlist_free(spa->spa_config);
365 spa->spa_config = config;
366 mutex_exit(&spa->spa_props_lock);
370 * Generate the pool's configuration based on the current in-core state.
372 * We infer whether to generate a complete config or just one top-level config
373 * based on whether vd is the root vdev.
376 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
378 nvlist_t *config, *nvroot;
379 vdev_t *rvd = spa->spa_root_vdev;
380 unsigned long hostid = 0;
381 boolean_t locked = B_FALSE;
387 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
390 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
391 (SCL_CONFIG | SCL_STATE));
394 * If txg is -1, report the current value of spa->spa_config_txg.
397 txg = spa->spa_config_txg;
399 config = fnvlist_alloc();
401 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
402 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, spa_name(spa));
403 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
404 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
405 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
406 if (spa->spa_comment != NULL) {
407 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
412 hostid = zone_get_hostid(NULL);
415 * We're emulating the system's hostid in userland, so we can't use
418 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
421 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
423 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname.nodename);
425 int config_gen_flags = 0;
427 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
428 vd->vdev_top->vdev_guid);
429 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
431 if (vd->vdev_isspare) {
432 fnvlist_add_uint64(config,
433 ZPOOL_CONFIG_IS_SPARE, 1ULL);
435 if (vd->vdev_islog) {
436 fnvlist_add_uint64(config,
437 ZPOOL_CONFIG_IS_LOG, 1ULL);
439 vd = vd->vdev_top; /* label contains top config */
442 * Only add the (potentially large) split information
443 * in the mos config, and not in the vdev labels
445 if (spa->spa_config_splitting != NULL)
446 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
447 spa->spa_config_splitting);
448 fnvlist_add_boolean(config,
449 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
451 config_gen_flags |= VDEV_CONFIG_MOS;
455 * Add the top-level config. We even add this on pools which
456 * don't support holes in the namespace.
458 vdev_top_config_generate(spa, config);
461 * If we're splitting, record the original pool's guid.
463 if (spa->spa_config_splitting != NULL &&
464 nvlist_lookup_uint64(spa->spa_config_splitting,
465 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
466 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
470 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
471 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
475 * Store what's necessary for reading the MOS in the label.
477 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
478 spa->spa_label_features);
480 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
481 ddt_histogram_t *ddh;
485 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
486 ddt_get_dedup_histogram(spa, ddh);
487 fnvlist_add_uint64_array(config,
488 ZPOOL_CONFIG_DDT_HISTOGRAM,
489 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
490 kmem_free(ddh, sizeof (ddt_histogram_t));
492 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
493 ddt_get_dedup_object_stats(spa, ddo);
494 fnvlist_add_uint64_array(config,
495 ZPOOL_CONFIG_DDT_OBJ_STATS,
496 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
497 kmem_free(ddo, sizeof (ddt_object_t));
499 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
500 ddt_get_dedup_stats(spa, dds);
501 fnvlist_add_uint64_array(config,
502 ZPOOL_CONFIG_DDT_STATS,
503 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
504 kmem_free(dds, sizeof (ddt_stat_t));
508 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
514 * Update all disk labels, generate a fresh config based on the current
515 * in-core state, and sync the global config cache (do not sync the config
516 * cache if this is a booting rootpool).
519 spa_config_update(spa_t *spa, int what)
521 vdev_t *rvd = spa->spa_root_vdev;
525 ASSERT(MUTEX_HELD(&spa_namespace_lock));
527 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
528 txg = spa_last_synced_txg(spa) + 1;
529 if (what == SPA_CONFIG_UPDATE_POOL) {
530 vdev_config_dirty(rvd);
533 * If we have top-level vdevs that were added but have
534 * not yet been prepared for allocation, do that now.
535 * (It's safe now because the config cache is up to date,
536 * so it will be able to translate the new DVAs.)
537 * See comments in spa_vdev_add() for full details.
539 for (c = 0; c < rvd->vdev_children; c++) {
540 vdev_t *tvd = rvd->vdev_child[c];
541 if (tvd->vdev_ms_array == 0) {
542 vdev_ashift_optimize(tvd);
543 vdev_metaslab_set_size(tvd);
545 vdev_expand(tvd, txg);
548 spa_config_exit(spa, SCL_ALL, FTAG);
551 * Wait for the mosconfig to be regenerated and synced.
553 txg_wait_synced(spa->spa_dsl_pool, txg);
556 * Update the global config cache to reflect the new mosconfig.
558 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
560 if (what == SPA_CONFIG_UPDATE_POOL)
561 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);