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 https://opensource.org/licenses/CDDL-1.0.
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, 2020 by Delphix. All rights reserved.
26 * Copyright 2017 Joyent, Inc.
27 * Copyright (c) 2021, Colm Buckley <colm@tuatha.org>
32 #include <sys/fm/fs/zfs.h>
33 #include <sys/spa_impl.h>
34 #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/systeminfo.h>
39 #include <sys/sunddi.h>
40 #include <sys/zfeature.h>
41 #include <sys/zfs_file.h>
42 #include <sys/zfs_context.h>
48 * Pool configuration repository.
50 * Pool configuration is stored as a packed nvlist on the filesystem. By
51 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
52 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
53 * property set that allows them to be stored in an alternate location until
54 * the control of external software.
56 * For each cache file, we have a single nvlist which holds all the
57 * configuration information. When the module loads, we read this information
58 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
59 * maintained independently in spa.c. Whenever the namespace is modified, or
60 * the configuration of a pool is changed, we call spa_write_cachefile(), which
61 * walks through all the active pools and writes the configuration to disk.
64 static uint64_t spa_config_generation = 1;
67 * This can be overridden in userland to preserve an alternate namespace for
68 * userland pools when doing testing.
70 char *spa_config_path = (char *)ZPOOL_CACHE;
72 static int zfs_autoimport_disable = B_TRUE;
76 * Called when the module is first loaded, this routine loads the configuration
77 * file into the SPA namespace. It does not actually open or load the pools; it
78 * only populates the namespace.
84 nvlist_t *nvlist, *child;
93 if (zfs_autoimport_disable)
98 * Open the configuration file.
100 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
102 (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path);
104 err = zfs_file_open(pathname, O_RDONLY, 0, &fp);
108 err = zfs_file_open(ZPOOL_CACHE_BOOT, O_RDONLY, 0, &fp);
110 kmem_free(pathname, MAXPATHLEN);
115 if (zfs_file_getattr(fp, &zfa))
118 fsize = zfa.zfa_size;
119 buf = kmem_alloc(fsize, KM_SLEEP);
122 * Read the nvlist from the file.
124 if (zfs_file_read(fp, buf, fsize, NULL) < 0)
130 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
134 * Iterate over all elements in the nvlist, creating a new spa_t for
135 * each one with the specified configuration.
137 mutex_enter(&spa_namespace_lock);
139 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
140 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
143 child = fnvpair_value_nvlist(nvpair);
145 if (spa_lookup(nvpair_name(nvpair)) != NULL)
147 (void) spa_add(nvpair_name(nvpair), child, NULL);
149 mutex_exit(&spa_namespace_lock);
155 kmem_free(buf, fsize);
161 spa_config_remove(spa_config_dirent_t *dp)
166 * Remove the cache file. If zfs_file_unlink() in not supported by the
167 * platform fallback to truncating the file which is functionally
170 error = zfs_file_unlink(dp->scd_path);
171 if (error == EOPNOTSUPP) {
172 int flags = O_RDWR | O_TRUNC;
175 error = zfs_file_open(dp->scd_path, flags, 0644, &fp);
177 (void) zfs_file_fsync(fp, O_SYNC);
178 (void) zfs_file_close(fp);
186 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
190 int oflags = O_RDWR | O_TRUNC | O_CREAT | O_LARGEFILE;
196 * If the nvlist is empty (NULL), then remove the old cachefile.
199 err = spa_config_remove(dp);
207 * Pack the configuration into a buffer.
209 buf = fnvlist_pack(nvl, &buflen);
210 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
213 * Write the configuration to disk. Due to the complexity involved
214 * in performing a rename and remove from within the kernel the file
215 * is instead truncated and overwritten in place. This way we always
216 * have a consistent view of the data or a zero length file.
218 err = zfs_file_open(dp->scd_path, oflags, 0644, &fp);
220 err = zfs_file_write(fp, buf, buflen, NULL);
222 err = zfs_file_fsync(fp, O_SYNC);
226 (void) spa_config_remove(dp);
228 fnvlist_pack_free(buf, buflen);
229 kmem_free(temp, MAXPATHLEN);
234 * Synchronize pool configuration to disk. This must be called with the
235 * namespace lock held. Synchronizing the pool cache is typically done after
236 * the configuration has been synced to the MOS. This exposes a window where
237 * the MOS config will have been updated but the cache file has not. If
238 * the system were to crash at that instant then the cached config may not
239 * contain the correct information to open the pool and an explicit import
243 spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent,
244 boolean_t postblkidevent)
246 spa_config_dirent_t *dp, *tdp;
249 boolean_t ccw_failure;
252 ASSERT(MUTEX_HELD(&spa_namespace_lock));
254 if (!(spa_mode_global & SPA_MODE_WRITE))
258 * Iterate over all cachefiles for the pool, past or present. When the
259 * cachefile is changed, the new one is pushed onto this list, allowing
260 * us to update previous cachefiles that no longer contain this pool.
262 ccw_failure = B_FALSE;
263 for (dp = list_head(&target->spa_config_list); dp != NULL;
264 dp = list_next(&target->spa_config_list, dp)) {
266 if (dp->scd_path == NULL)
270 * Iterate over all pools, adding any matching pools to 'nvl'.
273 while ((spa = spa_next(spa)) != NULL) {
275 * Skip over our own pool if we're about to remove
276 * ourselves from the spa namespace or any pool that
277 * is readonly. Since we cannot guarantee that a
278 * readonly pool would successfully import upon reboot,
279 * we don't allow them to be written to the cache file.
281 if ((spa == target && removing) ||
285 mutex_enter(&spa->spa_props_lock);
286 tdp = list_head(&spa->spa_config_list);
287 if (spa->spa_config == NULL ||
289 tdp->scd_path == NULL ||
290 strcmp(tdp->scd_path, dp->scd_path) != 0) {
291 mutex_exit(&spa->spa_props_lock);
296 nvl = fnvlist_alloc();
298 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME)
299 pool_name = fnvlist_lookup_string(
300 spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
302 pool_name = spa_name(spa);
304 fnvlist_add_nvlist(nvl, pool_name, spa->spa_config);
305 mutex_exit(&spa->spa_props_lock);
308 error = spa_config_write(dp, nvl);
310 ccw_failure = B_TRUE;
316 * Keep trying so that configuration data is
317 * written if/when any temporary filesystem
318 * resource issues are resolved.
320 if (target->spa_ccw_fail_time == 0) {
321 (void) zfs_ereport_post(
322 FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
323 target, NULL, NULL, NULL, 0);
325 target->spa_ccw_fail_time = gethrtime();
326 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
329 * Do not rate limit future attempts to update
332 target->spa_ccw_fail_time = 0;
336 * Remove any config entries older than the current one.
338 dp = list_head(&target->spa_config_list);
339 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
340 list_remove(&target->spa_config_list, tdp);
341 if (tdp->scd_path != NULL)
342 spa_strfree(tdp->scd_path);
343 kmem_free(tdp, sizeof (spa_config_dirent_t));
346 spa_config_generation++;
349 spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
352 * Post udev event to sync blkid information if the pool is created
353 * or a new vdev is added to the pool.
355 if ((target->spa_root_vdev) && postblkidevent) {
356 vdev_post_kobj_evt(target->spa_root_vdev);
357 for (int i = 0; i < target->spa_l2cache.sav_count; i++)
358 vdev_post_kobj_evt(target->spa_l2cache.sav_vdevs[i]);
363 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
364 * and we don't want to allow the local zone to see all the pools anyway.
365 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
366 * information for all pool visible within the zone.
369 spa_all_configs(uint64_t *generation)
374 if (*generation == spa_config_generation)
377 pools = fnvlist_alloc();
379 mutex_enter(&spa_namespace_lock);
380 while ((spa = spa_next(spa)) != NULL) {
381 if (INGLOBALZONE(curproc) ||
382 zone_dataset_visible(spa_name(spa), NULL)) {
383 mutex_enter(&spa->spa_props_lock);
384 fnvlist_add_nvlist(pools, spa_name(spa),
386 mutex_exit(&spa->spa_props_lock);
389 *generation = spa_config_generation;
390 mutex_exit(&spa_namespace_lock);
396 spa_config_set(spa_t *spa, nvlist_t *config)
398 mutex_enter(&spa->spa_props_lock);
399 if (spa->spa_config != NULL && spa->spa_config != config)
400 nvlist_free(spa->spa_config);
401 spa->spa_config = config;
402 mutex_exit(&spa->spa_props_lock);
406 * Generate the pool's configuration based on the current in-core state.
408 * We infer whether to generate a complete config or just one top-level config
409 * based on whether vd is the root vdev.
412 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
414 nvlist_t *config, *nvroot;
415 vdev_t *rvd = spa->spa_root_vdev;
416 unsigned long hostid = 0;
417 boolean_t locked = B_FALSE;
424 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
427 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
428 (SCL_CONFIG | SCL_STATE));
431 * If txg is -1, report the current value of spa->spa_config_txg.
434 txg = spa->spa_config_txg;
437 * Originally, users had to handle spa namespace collisions by either
438 * exporting the already imported pool or by specifying a new name for
439 * the pool with a conflicting name. In the case of root pools from
440 * virtual guests, neither approach to collision resolution is
441 * reasonable. This is addressed by extending the new name syntax with
442 * an option to specify that the new name is temporary. When specified,
443 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
444 * to use the previous name, which we do below.
446 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
447 VERIFY0(nvlist_lookup_string(spa->spa_config,
448 ZPOOL_CONFIG_POOL_NAME, &pool_name));
450 pool_name = spa_name(spa);
452 config = fnvlist_alloc();
454 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
455 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name);
456 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
457 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
458 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
459 fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata);
460 if (spa->spa_comment != NULL)
461 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
463 if (spa->spa_compatibility != NULL)
464 fnvlist_add_string(config, ZPOOL_CONFIG_COMPATIBILITY,
465 spa->spa_compatibility);
467 hostid = spa_get_hostid(spa);
469 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
470 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname()->nodename);
472 int config_gen_flags = 0;
474 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
475 vd->vdev_top->vdev_guid);
476 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
478 if (vd->vdev_isspare)
479 fnvlist_add_uint64(config,
480 ZPOOL_CONFIG_IS_SPARE, 1ULL);
482 fnvlist_add_uint64(config,
483 ZPOOL_CONFIG_IS_LOG, 1ULL);
484 vd = vd->vdev_top; /* label contains top config */
487 * Only add the (potentially large) split information
488 * in the mos config, and not in the vdev labels
490 if (spa->spa_config_splitting != NULL)
491 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
492 spa->spa_config_splitting);
494 fnvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
496 config_gen_flags |= VDEV_CONFIG_MOS;
500 * Add the top-level config. We even add this on pools which
501 * don't support holes in the namespace.
503 vdev_top_config_generate(spa, config);
506 * If we're splitting, record the original pool's guid.
508 if (spa->spa_config_splitting != NULL &&
509 nvlist_lookup_uint64(spa->spa_config_splitting,
510 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
511 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, split_guid);
514 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
515 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
519 * Store what's necessary for reading the MOS in the label.
521 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
522 spa->spa_label_features);
524 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
525 ddt_histogram_t *ddh;
529 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
530 ddt_get_dedup_histogram(spa, ddh);
531 fnvlist_add_uint64_array(config,
532 ZPOOL_CONFIG_DDT_HISTOGRAM,
533 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
534 kmem_free(ddh, sizeof (ddt_histogram_t));
536 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
537 ddt_get_dedup_object_stats(spa, ddo);
538 fnvlist_add_uint64_array(config,
539 ZPOOL_CONFIG_DDT_OBJ_STATS,
540 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
541 kmem_free(ddo, sizeof (ddt_object_t));
543 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
544 ddt_get_dedup_stats(spa, dds);
545 fnvlist_add_uint64_array(config,
546 ZPOOL_CONFIG_DDT_STATS,
547 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
548 kmem_free(dds, sizeof (ddt_stat_t));
552 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
558 * Update all disk labels, generate a fresh config based on the current
559 * in-core state, and sync the global config cache (do not sync the config
560 * cache if this is a booting rootpool).
563 spa_config_update(spa_t *spa, int what)
565 vdev_t *rvd = spa->spa_root_vdev;
569 ASSERT(MUTEX_HELD(&spa_namespace_lock));
571 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
572 txg = spa_last_synced_txg(spa) + 1;
573 if (what == SPA_CONFIG_UPDATE_POOL) {
574 vdev_config_dirty(rvd);
577 * If we have top-level vdevs that were added but have
578 * not yet been prepared for allocation, do that now.
579 * (It's safe now because the config cache is up to date,
580 * so it will be able to translate the new DVAs.)
581 * See comments in spa_vdev_add() for full details.
583 for (c = 0; c < rvd->vdev_children; c++) {
584 vdev_t *tvd = rvd->vdev_child[c];
587 * Explicitly skip vdevs that are indirect or
588 * log vdevs that are being removed. The reason
589 * is that both of those can have vdev_ms_array
590 * set to 0 and we wouldn't want to change their
591 * metaslab size nor call vdev_expand() on them.
593 if (!vdev_is_concrete(tvd) ||
594 (tvd->vdev_islog && tvd->vdev_removing))
597 if (tvd->vdev_ms_array == 0)
598 vdev_metaslab_set_size(tvd);
599 vdev_expand(tvd, txg);
602 spa_config_exit(spa, SCL_ALL, FTAG);
605 * Wait for the mosconfig to be regenerated and synced.
607 txg_wait_synced(spa->spa_dsl_pool, txg);
610 * Update the global config cache to reflect the new mosconfig.
612 if (!spa->spa_is_root) {
613 spa_write_cachefile(spa, B_FALSE,
614 what != SPA_CONFIG_UPDATE_POOL,
615 what != SPA_CONFIG_UPDATE_POOL);
618 if (what == SPA_CONFIG_UPDATE_POOL)
619 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
622 EXPORT_SYMBOL(spa_config_load);
623 EXPORT_SYMBOL(spa_all_configs);
624 EXPORT_SYMBOL(spa_config_set);
625 EXPORT_SYMBOL(spa_config_generate);
626 EXPORT_SYMBOL(spa_config_update);
629 /* string sysctls require a char array on FreeBSD */
630 ZFS_MODULE_PARAM(zfs_spa, spa_, config_path, STRING, ZMOD_RD,
631 "SPA config file (/etc/zfs/zpool.cache)");
634 ZFS_MODULE_PARAM(zfs, zfs_, autoimport_disable, INT, ZMOD_RW,
635 "Disable pool import at module load");