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.
26 * Copyright 2017 Joyent, Inc.
29 #include <sys/zfs_context.h>
31 #include <sys/fm/fs/zfs.h>
32 #include <sys/spa_impl.h>
33 #include <sys/nvpair.h>
35 #include <sys/fs/zfs.h>
36 #include <sys/vdev_impl.h>
37 #include <sys/zfs_ioctl.h>
38 #include <sys/utsname.h>
39 #include <sys/sunddi.h>
40 #include <sys/zfeature.h>
47 * Pool configuration repository.
49 * Pool configuration is stored as a packed nvlist on the filesystem. By
50 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
51 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
52 * property set that allows them to be stored in an alternate location until
53 * the control of external software.
55 * For each cache file, we have a single nvlist which holds all the
56 * configuration information. When the module loads, we read this information
57 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
58 * maintained independently in spa.c. Whenever the namespace is modified, or
59 * the configuration of a pool is changed, we call spa_write_cachefile(), which
60 * walks through all the active pools and writes the configuration to disk.
63 static uint64_t spa_config_generation = 1;
66 * This can be overridden in userland to preserve an alternate namespace for
67 * userland pools when doing testing.
69 const char *spa_config_path = ZPOOL_CACHE;
72 * Called when the module is first loaded, this routine loads the configuration
73 * file into the SPA namespace. It does not actually open or load the pools; it
74 * only populates the namespace.
80 nvlist_t *nvlist, *child;
87 * Open the configuration file.
89 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
91 (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path);
93 file = kobj_open_file(pathname);
95 kmem_free(pathname, MAXPATHLEN);
97 if (file == (struct _buf *)-1)
100 if (kobj_get_filesize(file, &fsize) != 0)
103 buf = kmem_alloc(fsize, KM_SLEEP);
106 * Read the nvlist from the file.
108 if (kobj_read_file(file, buf, fsize, 0) < 0)
114 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
118 * Iterate over all elements in the nvlist, creating a new spa_t for
119 * each one with the specified configuration.
121 mutex_enter(&spa_namespace_lock);
123 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
124 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
127 child = fnvpair_value_nvlist(nvpair);
129 if (spa_lookup(nvpair_name(nvpair)) != NULL)
131 (void) spa_add(nvpair_name(nvpair), child, NULL);
133 mutex_exit(&spa_namespace_lock);
139 kmem_free(buf, fsize);
141 kobj_close_file(file);
145 spa_config_clean(nvlist_t *nvl)
148 nvlist_t *nvroot = NULL;
151 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
153 for (c = 0; c < children; c++)
154 spa_config_clean(child[c]);
157 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0)
158 spa_config_clean(nvroot);
160 nvlist_remove(nvl, ZPOOL_CONFIG_VDEV_STATS, DATA_TYPE_UINT64_ARRAY);
161 nvlist_remove(nvl, ZPOOL_CONFIG_SCAN_STATS, DATA_TYPE_UINT64_ARRAY);
165 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
170 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
175 * If the nvlist is empty (NULL), then remove the old cachefile.
178 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
183 * Pack the configuration into a buffer.
185 buf = fnvlist_pack(nvl, &buflen);
186 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
189 * Write the configuration to disk. We need to do the traditional
190 * 'write to temporary file, sync, move over original' to make sure we
191 * always have a consistent view of the data.
193 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
195 err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
197 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
198 0, RLIM64_INFINITY, kcred, NULL);
200 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
202 err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
203 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
206 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
208 fnvlist_pack_free(buf, buflen);
209 kmem_free(temp, MAXPATHLEN);
214 * Synchronize pool configuration to disk. This must be called with the
215 * namespace lock held. Synchronizing the pool cache is typically done after
216 * the configuration has been synced to the MOS. This exposes a window where
217 * the MOS config will have been updated but the cache file has not. If
218 * the system were to crash at that instant then the cached config may not
219 * contain the correct information to open the pool and an explicit import
223 spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent)
225 spa_config_dirent_t *dp, *tdp;
227 boolean_t ccw_failure;
230 ASSERT(MUTEX_HELD(&spa_namespace_lock));
232 if (rootdir == NULL || !(spa_mode_global & FWRITE))
236 * Iterate over all cachefiles for the pool, past or present. When the
237 * cachefile is changed, the new one is pushed onto this list, allowing
238 * us to update previous cachefiles that no longer contain this pool.
240 ccw_failure = B_FALSE;
241 for (dp = list_head(&target->spa_config_list); dp != NULL;
242 dp = list_next(&target->spa_config_list, dp)) {
244 if (dp->scd_path == NULL)
248 * Iterate over all pools, adding any matching pools to 'nvl'.
251 while ((spa = spa_next(spa)) != NULL) {
252 nvlist_t *nvroot = NULL;
254 * Skip over our own pool if we're about to remove
255 * ourselves from the spa namespace or any pool that
256 * is readonly. Since we cannot guarantee that a
257 * readonly pool would successfully import upon reboot,
258 * we don't allow them to be written to the cache file.
260 if ((spa == target && removing) ||
261 (spa_state(spa) == POOL_STATE_ACTIVE &&
262 !spa_writeable(spa)))
265 mutex_enter(&spa->spa_props_lock);
266 tdp = list_head(&spa->spa_config_list);
267 if (spa->spa_config == NULL ||
268 tdp->scd_path == NULL ||
269 strcmp(tdp->scd_path, dp->scd_path) != 0) {
270 mutex_exit(&spa->spa_props_lock);
275 nvl = fnvlist_alloc();
277 fnvlist_add_nvlist(nvl, spa->spa_name,
279 mutex_exit(&spa->spa_props_lock);
281 if (nvlist_lookup_nvlist(nvl, spa->spa_name, &nvroot) == 0)
282 spa_config_clean(nvroot);
285 error = spa_config_write(dp, nvl);
287 ccw_failure = B_TRUE;
293 * Keep trying so that configuration data is
294 * written if/when any temporary filesystem
295 * resource issues are resolved.
297 if (target->spa_ccw_fail_time == 0) {
298 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
299 target, NULL, NULL, 0, 0);
301 target->spa_ccw_fail_time = gethrtime();
302 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
305 * Do not rate limit future attempts to update
308 target->spa_ccw_fail_time = 0;
312 * Remove any config entries older than the current one.
314 dp = list_head(&target->spa_config_list);
315 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
316 list_remove(&target->spa_config_list, tdp);
317 if (tdp->scd_path != NULL)
318 spa_strfree(tdp->scd_path);
319 kmem_free(tdp, sizeof (spa_config_dirent_t));
322 spa_config_generation++;
325 spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
329 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
330 * and we don't want to allow the local zone to see all the pools anyway.
331 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
332 * information for all pool visible within the zone.
335 spa_all_configs(uint64_t *generation)
340 if (*generation == spa_config_generation)
343 pools = fnvlist_alloc();
345 mutex_enter(&spa_namespace_lock);
346 while ((spa = spa_next(spa)) != NULL) {
347 if (INGLOBALZONE(curthread) ||
348 zone_dataset_visible(spa_name(spa), NULL)) {
349 mutex_enter(&spa->spa_props_lock);
350 fnvlist_add_nvlist(pools, spa_name(spa),
352 mutex_exit(&spa->spa_props_lock);
355 *generation = spa_config_generation;
356 mutex_exit(&spa_namespace_lock);
362 spa_config_set(spa_t *spa, nvlist_t *config)
364 mutex_enter(&spa->spa_props_lock);
365 nvlist_free(spa->spa_config);
366 spa->spa_config = config;
367 mutex_exit(&spa->spa_props_lock);
371 * Generate the pool's configuration based on the current in-core state.
373 * We infer whether to generate a complete config or just one top-level config
374 * based on whether vd is the root vdev.
377 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
379 nvlist_t *config, *nvroot;
380 vdev_t *rvd = spa->spa_root_vdev;
381 unsigned long hostid = 0;
382 boolean_t locked = B_FALSE;
388 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
391 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
392 (SCL_CONFIG | SCL_STATE));
395 * If txg is -1, report the current value of spa->spa_config_txg.
398 txg = spa->spa_config_txg;
400 config = fnvlist_alloc();
402 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
403 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, spa_name(spa));
404 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
405 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
406 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
407 if (spa->spa_comment != NULL) {
408 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
413 hostid = zone_get_hostid(NULL);
416 * We're emulating the system's hostid in userland, so we can't use
419 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
422 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
424 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname.nodename);
426 int config_gen_flags = 0;
428 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
429 vd->vdev_top->vdev_guid);
430 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
432 if (vd->vdev_isspare) {
433 fnvlist_add_uint64(config,
434 ZPOOL_CONFIG_IS_SPARE, 1ULL);
436 if (vd->vdev_islog) {
437 fnvlist_add_uint64(config,
438 ZPOOL_CONFIG_IS_LOG, 1ULL);
440 vd = vd->vdev_top; /* label contains top config */
443 * Only add the (potentially large) split information
444 * in the mos config, and not in the vdev labels
446 if (spa->spa_config_splitting != NULL)
447 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
448 spa->spa_config_splitting);
449 fnvlist_add_boolean(config,
450 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
452 config_gen_flags |= VDEV_CONFIG_MOS;
456 * Add the top-level config. We even add this on pools which
457 * don't support holes in the namespace.
459 vdev_top_config_generate(spa, config);
462 * If we're splitting, record the original pool's guid.
464 if (spa->spa_config_splitting != NULL &&
465 nvlist_lookup_uint64(spa->spa_config_splitting,
466 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
467 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
471 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
472 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
476 * Store what's necessary for reading the MOS in the label.
478 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
479 spa->spa_label_features);
481 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
482 ddt_histogram_t *ddh;
486 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
487 ddt_get_dedup_histogram(spa, ddh);
488 fnvlist_add_uint64_array(config,
489 ZPOOL_CONFIG_DDT_HISTOGRAM,
490 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
491 kmem_free(ddh, sizeof (ddt_histogram_t));
493 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
494 ddt_get_dedup_object_stats(spa, ddo);
495 fnvlist_add_uint64_array(config,
496 ZPOOL_CONFIG_DDT_OBJ_STATS,
497 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
498 kmem_free(ddo, sizeof (ddt_object_t));
500 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
501 ddt_get_dedup_stats(spa, dds);
502 fnvlist_add_uint64_array(config,
503 ZPOOL_CONFIG_DDT_STATS,
504 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
505 kmem_free(dds, sizeof (ddt_stat_t));
509 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
515 * Update all disk labels, generate a fresh config based on the current
516 * in-core state, and sync the global config cache (do not sync the config
517 * cache if this is a booting rootpool).
520 spa_config_update(spa_t *spa, int what)
522 vdev_t *rvd = spa->spa_root_vdev;
526 ASSERT(MUTEX_HELD(&spa_namespace_lock));
528 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
529 txg = spa_last_synced_txg(spa) + 1;
530 if (what == SPA_CONFIG_UPDATE_POOL) {
531 vdev_config_dirty(rvd);
534 * If we have top-level vdevs that were added but have
535 * not yet been prepared for allocation, do that now.
536 * (It's safe now because the config cache is up to date,
537 * so it will be able to translate the new DVAs.)
538 * See comments in spa_vdev_add() for full details.
540 for (c = 0; c < rvd->vdev_children; c++) {
541 vdev_t *tvd = rvd->vdev_child[c];
542 if (tvd->vdev_ms_array == 0) {
543 vdev_ashift_optimize(tvd);
544 vdev_metaslab_set_size(tvd);
546 vdev_expand(tvd, txg);
549 spa_config_exit(spa, SCL_ALL, FTAG);
552 * Wait for the mosconfig to be regenerated and synced.
554 txg_wait_synced(spa->spa_dsl_pool, txg);
557 * Update the global config cache to reflect the new mosconfig.
559 spa_write_cachefile(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
561 if (what == SPA_CONFIG_UPDATE_POOL)
562 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);