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.
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/systeminfo.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%s",
92 (rootdir
!= NULL
) ? "./" : "", spa_config_path
);
94 file
= kobj_open_file(pathname
);
96 kmem_free(pathname
, MAXPATHLEN
);
98 if (file
== (struct _buf
*)-1)
101 if (kobj_get_filesize(file
, &fsize
) != 0)
104 buf
= kmem_alloc(fsize
, KM_SLEEP
);
107 * Read the nvlist from the file.
109 if (kobj_read_file(file
, buf
, fsize
, 0) < 0)
115 if (nvlist_unpack(buf
, fsize
, &nvlist
, KM_SLEEP
) != 0)
119 * Iterate over all elements in the nvlist, creating a new spa_t for
120 * each one with the specified configuration.
122 mutex_enter(&spa_namespace_lock
);
124 while ((nvpair
= nvlist_next_nvpair(nvlist
, nvpair
)) != NULL
) {
125 if (nvpair_type(nvpair
) != DATA_TYPE_NVLIST
)
128 child
= fnvpair_value_nvlist(nvpair
);
130 if (spa_lookup(nvpair_name(nvpair
)) != NULL
)
132 (void) spa_add(nvpair_name(nvpair
), child
, NULL
);
134 mutex_exit(&spa_namespace_lock
);
140 kmem_free(buf
, fsize
);
142 kobj_close_file(file
);
146 spa_config_write(spa_config_dirent_t
*dp
, nvlist_t
*nvl
)
151 int oflags
= FWRITE
| FTRUNC
| FCREAT
| FOFFMAX
;
156 * If the nvlist is empty (NULL), then remove the old cachefile.
159 err
= vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
164 * Pack the configuration into a buffer.
166 buf
= fnvlist_pack(nvl
, &buflen
);
167 temp
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
170 * Write the configuration to disk. We need to do the traditional
171 * 'write to temporary file, sync, move over original' to make sure we
172 * always have a consistent view of the data.
174 (void) snprintf(temp
, MAXPATHLEN
, "%s.tmp", dp
->scd_path
);
176 err
= vn_open(temp
, UIO_SYSSPACE
, oflags
, 0644, &vp
, CRCREAT
, 0);
178 err
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0, UIO_SYSSPACE
,
179 0, RLIM64_INFINITY
, kcred
, NULL
);
181 err
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
183 err
= vn_rename(temp
, dp
->scd_path
, UIO_SYSSPACE
);
184 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
188 (void) vn_remove(temp
, UIO_SYSSPACE
, RMFILE
);
190 fnvlist_pack_free(buf
, buflen
);
191 kmem_free(temp
, MAXPATHLEN
);
196 * Synchronize pool configuration to disk. This must be called with the
197 * namespace lock held. Synchronizing the pool cache is typically done after
198 * the configuration has been synced to the MOS. This exposes a window where
199 * the MOS config will have been updated but the cache file has not. If
200 * the system were to crash at that instant then the cached config may not
201 * contain the correct information to open the pool and an explicit import
205 spa_write_cachefile(spa_t
*target
, boolean_t removing
, boolean_t postsysevent
)
207 spa_config_dirent_t
*dp
, *tdp
;
209 boolean_t ccw_failure
;
212 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
214 if (rootdir
== NULL
|| !(spa_mode_global
& FWRITE
))
218 * Iterate over all cachefiles for the pool, past or present. When the
219 * cachefile is changed, the new one is pushed onto this list, allowing
220 * us to update previous cachefiles that no longer contain this pool.
222 ccw_failure
= B_FALSE
;
223 for (dp
= list_head(&target
->spa_config_list
); dp
!= NULL
;
224 dp
= list_next(&target
->spa_config_list
, dp
)) {
226 if (dp
->scd_path
== NULL
)
230 * Iterate over all pools, adding any matching pools to 'nvl'.
233 while ((spa
= spa_next(spa
)) != NULL
) {
235 * Skip over our own pool if we're about to remove
236 * ourselves from the spa namespace or any pool that
237 * is readonly. Since we cannot guarantee that a
238 * readonly pool would successfully import upon reboot,
239 * we don't allow them to be written to the cache file.
241 if ((spa
== target
&& removing
) ||
245 mutex_enter(&spa
->spa_props_lock
);
246 tdp
= list_head(&spa
->spa_config_list
);
247 if (spa
->spa_config
== NULL
||
248 tdp
->scd_path
== NULL
||
249 strcmp(tdp
->scd_path
, dp
->scd_path
) != 0) {
250 mutex_exit(&spa
->spa_props_lock
);
255 nvl
= fnvlist_alloc();
257 fnvlist_add_nvlist(nvl
, spa
->spa_name
,
259 mutex_exit(&spa
->spa_props_lock
);
262 error
= spa_config_write(dp
, nvl
);
264 ccw_failure
= B_TRUE
;
270 * Keep trying so that configuration data is
271 * written if/when any temporary filesystem
272 * resource issues are resolved.
274 if (target
->spa_ccw_fail_time
== 0) {
275 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE
,
276 target
, NULL
, NULL
, 0, 0);
278 target
->spa_ccw_fail_time
= gethrtime();
279 spa_async_request(target
, SPA_ASYNC_CONFIG_UPDATE
);
282 * Do not rate limit future attempts to update
285 target
->spa_ccw_fail_time
= 0;
289 * Remove any config entries older than the current one.
291 dp
= list_head(&target
->spa_config_list
);
292 while ((tdp
= list_next(&target
->spa_config_list
, dp
)) != NULL
) {
293 list_remove(&target
->spa_config_list
, tdp
);
294 if (tdp
->scd_path
!= NULL
)
295 spa_strfree(tdp
->scd_path
);
296 kmem_free(tdp
, sizeof (spa_config_dirent_t
));
299 spa_config_generation
++;
302 spa_event_notify(target
, NULL
, NULL
, ESC_ZFS_CONFIG_SYNC
);
306 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
307 * and we don't want to allow the local zone to see all the pools anyway.
308 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
309 * information for all pool visible within the zone.
312 spa_all_configs(uint64_t *generation
)
317 if (*generation
== spa_config_generation
)
320 pools
= fnvlist_alloc();
322 mutex_enter(&spa_namespace_lock
);
323 while ((spa
= spa_next(spa
)) != NULL
) {
324 if (INGLOBALZONE(curproc
) ||
325 zone_dataset_visible(spa_name(spa
), NULL
)) {
326 mutex_enter(&spa
->spa_props_lock
);
327 fnvlist_add_nvlist(pools
, spa_name(spa
),
329 mutex_exit(&spa
->spa_props_lock
);
332 *generation
= spa_config_generation
;
333 mutex_exit(&spa_namespace_lock
);
339 spa_config_set(spa_t
*spa
, nvlist_t
*config
)
341 mutex_enter(&spa
->spa_props_lock
);
342 if (spa
->spa_config
!= NULL
&& spa
->spa_config
!= config
)
343 nvlist_free(spa
->spa_config
);
344 spa
->spa_config
= config
;
345 mutex_exit(&spa
->spa_props_lock
);
349 * Generate the pool's configuration based on the current in-core state.
351 * We infer whether to generate a complete config or just one top-level config
352 * based on whether vd is the root vdev.
355 spa_config_generate(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int getstats
)
357 nvlist_t
*config
, *nvroot
;
358 vdev_t
*rvd
= spa
->spa_root_vdev
;
359 unsigned long hostid
= 0;
360 boolean_t locked
= B_FALSE
;
366 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
369 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
) ==
370 (SCL_CONFIG
| SCL_STATE
));
373 * If txg is -1, report the current value of spa->spa_config_txg.
376 txg
= spa
->spa_config_txg
;
378 config
= fnvlist_alloc();
380 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
, spa_version(spa
));
381 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, spa_name(spa
));
382 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, spa_state(spa
));
383 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, txg
);
384 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, spa_guid(spa
));
385 if (spa
->spa_comment
!= NULL
) {
386 fnvlist_add_string(config
, ZPOOL_CONFIG_COMMENT
,
390 hostid
= zone_get_hostid(NULL
);
393 fnvlist_add_uint64(config
, ZPOOL_CONFIG_HOSTID
, hostid
);
395 fnvlist_add_string(config
, ZPOOL_CONFIG_HOSTNAME
, utsname
.nodename
);
397 int config_gen_flags
= 0;
399 fnvlist_add_uint64(config
, ZPOOL_CONFIG_TOP_GUID
,
400 vd
->vdev_top
->vdev_guid
);
401 fnvlist_add_uint64(config
, ZPOOL_CONFIG_GUID
,
403 if (vd
->vdev_isspare
) {
404 fnvlist_add_uint64(config
,
405 ZPOOL_CONFIG_IS_SPARE
, 1ULL);
407 if (vd
->vdev_islog
) {
408 fnvlist_add_uint64(config
,
409 ZPOOL_CONFIG_IS_LOG
, 1ULL);
411 vd
= vd
->vdev_top
; /* label contains top config */
414 * Only add the (potentially large) split information
415 * in the mos config, and not in the vdev labels
417 if (spa
->spa_config_splitting
!= NULL
)
418 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
419 spa
->spa_config_splitting
);
420 fnvlist_add_boolean(config
,
421 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
);
423 config_gen_flags
|= VDEV_CONFIG_MOS
;
427 * Add the top-level config. We even add this on pools which
428 * don't support holes in the namespace.
430 vdev_top_config_generate(spa
, config
);
433 * If we're splitting, record the original pool's guid.
435 if (spa
->spa_config_splitting
!= NULL
&&
436 nvlist_lookup_uint64(spa
->spa_config_splitting
,
437 ZPOOL_CONFIG_SPLIT_GUID
, &split_guid
) == 0) {
438 fnvlist_add_uint64(config
, ZPOOL_CONFIG_SPLIT_GUID
,
442 nvroot
= vdev_config_generate(spa
, vd
, getstats
, config_gen_flags
);
443 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
);
447 * Store what's necessary for reading the MOS in the label.
449 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
450 spa
->spa_label_features
);
452 if (getstats
&& spa_load_state(spa
) == SPA_LOAD_NONE
) {
453 ddt_histogram_t
*ddh
;
457 ddh
= kmem_zalloc(sizeof (ddt_histogram_t
), KM_SLEEP
);
458 ddt_get_dedup_histogram(spa
, ddh
);
459 fnvlist_add_uint64_array(config
,
460 ZPOOL_CONFIG_DDT_HISTOGRAM
,
461 (uint64_t *)ddh
, sizeof (*ddh
) / sizeof (uint64_t));
462 kmem_free(ddh
, sizeof (ddt_histogram_t
));
464 ddo
= kmem_zalloc(sizeof (ddt_object_t
), KM_SLEEP
);
465 ddt_get_dedup_object_stats(spa
, ddo
);
466 fnvlist_add_uint64_array(config
,
467 ZPOOL_CONFIG_DDT_OBJ_STATS
,
468 (uint64_t *)ddo
, sizeof (*ddo
) / sizeof (uint64_t));
469 kmem_free(ddo
, sizeof (ddt_object_t
));
471 dds
= kmem_zalloc(sizeof (ddt_stat_t
), KM_SLEEP
);
472 ddt_get_dedup_stats(spa
, dds
);
473 fnvlist_add_uint64_array(config
,
474 ZPOOL_CONFIG_DDT_STATS
,
475 (uint64_t *)dds
, sizeof (*dds
) / sizeof (uint64_t));
476 kmem_free(dds
, sizeof (ddt_stat_t
));
480 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
486 * Update all disk labels, generate a fresh config based on the current
487 * in-core state, and sync the global config cache (do not sync the config
488 * cache if this is a booting rootpool).
491 spa_config_update(spa_t
*spa
, int what
)
493 vdev_t
*rvd
= spa
->spa_root_vdev
;
497 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
499 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
500 txg
= spa_last_synced_txg(spa
) + 1;
501 if (what
== SPA_CONFIG_UPDATE_POOL
) {
502 vdev_config_dirty(rvd
);
505 * If we have top-level vdevs that were added but have
506 * not yet been prepared for allocation, do that now.
507 * (It's safe now because the config cache is up to date,
508 * so it will be able to translate the new DVAs.)
509 * See comments in spa_vdev_add() for full details.
511 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
512 vdev_t
*tvd
= rvd
->vdev_child
[c
];
513 if (tvd
->vdev_ms_array
== 0)
514 vdev_metaslab_set_size(tvd
);
515 vdev_expand(tvd
, txg
);
518 spa_config_exit(spa
, SCL_ALL
, FTAG
);
521 * Wait for the mosconfig to be regenerated and synced.
523 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
526 * Update the global config cache to reflect the new mosconfig.
528 if (!spa
->spa_is_root
) {
529 spa_write_cachefile(spa
, B_FALSE
,
530 what
!= SPA_CONFIG_UPDATE_POOL
);
533 if (what
== SPA_CONFIG_UPDATE_POOL
)
534 spa_config_update(spa
, SPA_CONFIG_UPDATE_VDEVS
);