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.
29 #include <sys/fm/fs/zfs.h>
30 #include <sys/spa_impl.h>
31 #include <sys/nvpair.h>
33 #include <sys/fs/zfs.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/zfs_ioctl.h>
36 #include <sys/systeminfo.h>
37 #include <sys/sunddi.h>
38 #include <sys/zfeature.h>
45 * Pool configuration repository.
47 * Pool configuration is stored as a packed nvlist on the filesystem. By
48 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
49 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
50 * property set that allows them to be stored in an alternate location until
51 * the control of external software.
53 * For each cache file, we have a single nvlist which holds all the
54 * configuration information. When the module loads, we read this information
55 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
56 * maintained independently in spa.c. Whenever the namespace is modified, or
57 * the configuration of a pool is changed, we call spa_config_sync(), which
58 * walks through all the active pools and writes the configuration to disk.
61 static uint64_t spa_config_generation
= 1;
64 * This can be overridden in userland to preserve an alternate namespace for
65 * userland pools when doing testing.
67 char *spa_config_path
= ZPOOL_CACHE
;
68 int zfs_autoimport_disable
= 1;
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 if (zfs_autoimport_disable
)
91 * Open the configuration file.
93 pathname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
95 (void) snprintf(pathname
, MAXPATHLEN
, "%s%s",
96 (rootdir
!= NULL
) ? "./" : "", spa_config_path
);
98 file
= kobj_open_file(pathname
);
100 kmem_free(pathname
, MAXPATHLEN
);
102 if (file
== (struct _buf
*)-1)
105 if (kobj_get_filesize(file
, &fsize
) != 0)
108 buf
= kmem_alloc(fsize
, KM_SLEEP
);
111 * Read the nvlist from the file.
113 if (kobj_read_file(file
, buf
, fsize
, 0) < 0)
119 if (nvlist_unpack(buf
, fsize
, &nvlist
, KM_SLEEP
) != 0)
123 * Iterate over all elements in the nvlist, creating a new spa_t for
124 * each one with the specified configuration.
126 mutex_enter(&spa_namespace_lock
);
128 while ((nvpair
= nvlist_next_nvpair(nvlist
, nvpair
)) != NULL
) {
129 if (nvpair_type(nvpair
) != DATA_TYPE_NVLIST
)
132 child
= fnvpair_value_nvlist(nvpair
);
134 if (spa_lookup(nvpair_name(nvpair
)) != NULL
)
136 (void) spa_add(nvpair_name(nvpair
), child
, NULL
);
138 mutex_exit(&spa_namespace_lock
);
144 kmem_free(buf
, fsize
);
146 kobj_close_file(file
);
150 spa_config_write(spa_config_dirent_t
*dp
, nvlist_t
*nvl
)
155 int oflags
= FWRITE
| FTRUNC
| FCREAT
| FOFFMAX
;
160 * If the nvlist is empty (NULL), then remove the old cachefile.
163 err
= vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
168 * Pack the configuration into a buffer.
170 buf
= fnvlist_pack(nvl
, &buflen
);
171 temp
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
173 #if defined(__linux__) && defined(_KERNEL)
175 * Write the configuration to disk. Due to the complexity involved
176 * in performing a rename from within the kernel the file is truncated
177 * and overwritten in place. In the event of an error the file is
178 * unlinked to make sure we always have a consistent view of the data.
180 err
= vn_open(dp
->scd_path
, UIO_SYSSPACE
, oflags
, 0644, &vp
, 0, 0);
182 err
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0,
183 UIO_SYSSPACE
, 0, RLIM64_INFINITY
, kcred
, NULL
);
185 err
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
187 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
190 (void) vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
194 * Write the configuration to disk. We need to do the traditional
195 * 'write to temporary file, sync, move over original' to make sure we
196 * always have a consistent view of the data.
198 (void) snprintf(temp
, MAXPATHLEN
, "%s.tmp", dp
->scd_path
);
200 err
= vn_open(temp
, UIO_SYSSPACE
, oflags
, 0644, &vp
, CRCREAT
, 0);
202 err
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0, UIO_SYSSPACE
,
203 0, RLIM64_INFINITY
, kcred
, NULL
);
205 err
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
207 err
= vn_rename(temp
, dp
->scd_path
, UIO_SYSSPACE
);
208 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
211 (void) vn_remove(temp
, UIO_SYSSPACE
, RMFILE
);
214 fnvlist_pack_free(buf
, buflen
);
215 kmem_free(temp
, MAXPATHLEN
);
220 * Synchronize pool configuration to disk. This must be called with the
221 * namespace lock held. Synchronizing the pool cache is typically done after
222 * the configuration has been synced to the MOS. This exposes a window where
223 * the MOS config will have been updated but the cache file has not. If
224 * the system were to crash at that instant then the cached config may not
225 * contain the correct information to open the pool and an explicity import
229 spa_config_sync(spa_t
*target
, boolean_t removing
, boolean_t postsysevent
)
231 spa_config_dirent_t
*dp
, *tdp
;
234 boolean_t ccw_failure
;
237 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
239 if (rootdir
== NULL
|| !(spa_mode_global
& FWRITE
))
243 * Iterate over all cachefiles for the pool, past or present. When the
244 * cachefile is changed, the new one is pushed onto this list, allowing
245 * us to update previous cachefiles that no longer contain this pool.
247 ccw_failure
= B_FALSE
;
248 for (dp
= list_head(&target
->spa_config_list
); dp
!= NULL
;
249 dp
= list_next(&target
->spa_config_list
, dp
)) {
251 if (dp
->scd_path
== NULL
)
255 * Iterate over all pools, adding any matching pools to 'nvl'.
258 while ((spa
= spa_next(spa
)) != NULL
) {
260 * Skip over our own pool if we're about to remove
261 * ourselves from the spa namespace or any pool that
262 * is readonly. Since we cannot guarantee that a
263 * readonly pool would successfully import upon reboot,
264 * we don't allow them to be written to the cache file.
266 if ((spa
== target
&& removing
) ||
270 mutex_enter(&spa
->spa_props_lock
);
271 tdp
= list_head(&spa
->spa_config_list
);
272 if (spa
->spa_config
== NULL
||
273 tdp
->scd_path
== NULL
||
274 strcmp(tdp
->scd_path
, dp
->scd_path
) != 0) {
275 mutex_exit(&spa
->spa_props_lock
);
280 nvl
= fnvlist_alloc();
282 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
)
283 pool_name
= fnvlist_lookup_string(
284 spa
->spa_config
, ZPOOL_CONFIG_POOL_NAME
);
286 pool_name
= spa_name(spa
);
288 fnvlist_add_nvlist(nvl
, pool_name
, spa
->spa_config
);
289 mutex_exit(&spa
->spa_props_lock
);
292 error
= spa_config_write(dp
, nvl
);
294 ccw_failure
= B_TRUE
;
300 * Keep trying so that configuration data is
301 * written if/when any temporary filesystem
302 * resource issues are resolved.
304 if (target
->spa_ccw_fail_time
== 0) {
305 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE
,
306 target
, NULL
, NULL
, 0, 0);
308 target
->spa_ccw_fail_time
= gethrtime();
309 spa_async_request(target
, SPA_ASYNC_CONFIG_UPDATE
);
312 * Do not rate limit future attempts to update
315 target
->spa_ccw_fail_time
= 0;
319 * Remove any config entries older than the current one.
321 dp
= list_head(&target
->spa_config_list
);
322 while ((tdp
= list_next(&target
->spa_config_list
, dp
)) != NULL
) {
323 list_remove(&target
->spa_config_list
, tdp
);
324 if (tdp
->scd_path
!= NULL
)
325 spa_strfree(tdp
->scd_path
);
326 kmem_free(tdp
, sizeof (spa_config_dirent_t
));
329 spa_config_generation
++;
332 spa_event_notify(target
, NULL
, FM_EREPORT_ZFS_CONFIG_SYNC
);
336 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
337 * and we don't want to allow the local zone to see all the pools anyway.
338 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
339 * information for all pool visible within the zone.
342 spa_all_configs(uint64_t *generation
)
347 if (*generation
== spa_config_generation
)
350 pools
= fnvlist_alloc();
352 mutex_enter(&spa_namespace_lock
);
353 while ((spa
= spa_next(spa
)) != NULL
) {
354 if (INGLOBALZONE(curproc
) ||
355 zone_dataset_visible(spa_name(spa
), NULL
)) {
356 mutex_enter(&spa
->spa_props_lock
);
357 fnvlist_add_nvlist(pools
, spa_name(spa
),
359 mutex_exit(&spa
->spa_props_lock
);
362 *generation
= spa_config_generation
;
363 mutex_exit(&spa_namespace_lock
);
369 spa_config_set(spa_t
*spa
, nvlist_t
*config
)
371 mutex_enter(&spa
->spa_props_lock
);
372 nvlist_free(spa
->spa_config
);
373 spa
->spa_config
= config
;
374 mutex_exit(&spa
->spa_props_lock
);
378 * Generate the pool's configuration based on the current in-core state.
380 * We infer whether to generate a complete config or just one top-level config
381 * based on whether vd is the root vdev.
384 spa_config_generate(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int getstats
)
386 nvlist_t
*config
, *nvroot
;
387 vdev_t
*rvd
= spa
->spa_root_vdev
;
388 unsigned long hostid
= 0;
389 boolean_t locked
= B_FALSE
;
392 int config_gen_flags
= 0;
397 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
400 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
) ==
401 (SCL_CONFIG
| SCL_STATE
));
404 * If txg is -1, report the current value of spa->spa_config_txg.
407 txg
= spa
->spa_config_txg
;
410 * Originally, users had to handle spa namespace collisions by either
411 * exporting the already imported pool or by specifying a new name for
412 * the pool with a conflicting name. In the case of root pools from
413 * virtual guests, neither approach to collision resolution is
414 * reasonable. This is addressed by extending the new name syntax with
415 * an option to specify that the new name is temporary. When specified,
416 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
417 * to use the previous name, which we do below.
419 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
) {
420 VERIFY0(nvlist_lookup_string(spa
->spa_config
,
421 ZPOOL_CONFIG_POOL_NAME
, &pool_name
));
423 pool_name
= spa_name(spa
);
425 config
= fnvlist_alloc();
427 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
, spa_version(spa
));
428 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, spa_name(spa
));
429 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, spa_state(spa
));
430 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, txg
);
431 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, spa_guid(spa
));
432 fnvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
, spa
->spa_errata
);
433 if (spa
->spa_comment
!= NULL
)
434 fnvlist_add_string(config
, ZPOOL_CONFIG_COMMENT
,
438 hostid
= zone_get_hostid(NULL
);
441 * We're emulating the system's hostid in userland, so we can't use
444 (void) ddi_strtoul(hw_serial
, NULL
, 10, &hostid
);
447 fnvlist_add_uint64(config
, ZPOOL_CONFIG_HOSTID
, hostid
);
448 fnvlist_add_string(config
, ZPOOL_CONFIG_HOSTNAME
, utsname()->nodename
);
451 fnvlist_add_uint64(config
, ZPOOL_CONFIG_TOP_GUID
,
452 vd
->vdev_top
->vdev_guid
);
453 fnvlist_add_uint64(config
, ZPOOL_CONFIG_GUID
,
455 if (vd
->vdev_isspare
)
456 fnvlist_add_uint64(config
,
457 ZPOOL_CONFIG_IS_SPARE
, 1ULL);
459 fnvlist_add_uint64(config
,
460 ZPOOL_CONFIG_IS_LOG
, 1ULL);
461 vd
= vd
->vdev_top
; /* label contains top config */
464 * Only add the (potentially large) split information
465 * in the mos config, and not in the vdev labels
467 if (spa
->spa_config_splitting
!= NULL
)
468 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
469 spa
->spa_config_splitting
);
471 fnvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
);
473 config_gen_flags
|= VDEV_CONFIG_MOS
;
477 * Add the top-level config. We even add this on pools which
478 * don't support holes in the namespace.
480 vdev_top_config_generate(spa
, config
);
483 * If we're splitting, record the original pool's guid.
485 if (spa
->spa_config_splitting
!= NULL
&&
486 nvlist_lookup_uint64(spa
->spa_config_splitting
,
487 ZPOOL_CONFIG_SPLIT_GUID
, &split_guid
) == 0) {
488 fnvlist_add_uint64(config
, ZPOOL_CONFIG_SPLIT_GUID
, split_guid
);
491 nvroot
= vdev_config_generate(spa
, vd
, getstats
, config_gen_flags
);
492 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
);
496 * Store what's necessary for reading the MOS in the label.
498 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
499 spa
->spa_label_features
);
501 if (getstats
&& spa_load_state(spa
) == SPA_LOAD_NONE
) {
502 ddt_histogram_t
*ddh
;
506 ddh
= kmem_zalloc(sizeof (ddt_histogram_t
), KM_SLEEP
);
507 ddt_get_dedup_histogram(spa
, ddh
);
508 fnvlist_add_uint64_array(config
,
509 ZPOOL_CONFIG_DDT_HISTOGRAM
,
510 (uint64_t *)ddh
, sizeof (*ddh
) / sizeof (uint64_t));
511 kmem_free(ddh
, sizeof (ddt_histogram_t
));
513 ddo
= kmem_zalloc(sizeof (ddt_object_t
), KM_SLEEP
);
514 ddt_get_dedup_object_stats(spa
, ddo
);
515 fnvlist_add_uint64_array(config
,
516 ZPOOL_CONFIG_DDT_OBJ_STATS
,
517 (uint64_t *)ddo
, sizeof (*ddo
) / sizeof (uint64_t));
518 kmem_free(ddo
, sizeof (ddt_object_t
));
520 dds
= kmem_zalloc(sizeof (ddt_stat_t
), KM_SLEEP
);
521 ddt_get_dedup_stats(spa
, dds
);
522 fnvlist_add_uint64_array(config
,
523 ZPOOL_CONFIG_DDT_STATS
,
524 (uint64_t *)dds
, sizeof (*dds
) / sizeof (uint64_t));
525 kmem_free(dds
, sizeof (ddt_stat_t
));
529 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
535 * Update all disk labels, generate a fresh config based on the current
536 * in-core state, and sync the global config cache (do not sync the config
537 * cache if this is a booting rootpool).
540 spa_config_update(spa_t
*spa
, int what
)
542 vdev_t
*rvd
= spa
->spa_root_vdev
;
546 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
548 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
549 txg
= spa_last_synced_txg(spa
) + 1;
550 if (what
== SPA_CONFIG_UPDATE_POOL
) {
551 vdev_config_dirty(rvd
);
554 * If we have top-level vdevs that were added but have
555 * not yet been prepared for allocation, do that now.
556 * (It's safe now because the config cache is up to date,
557 * so it will be able to translate the new DVAs.)
558 * See comments in spa_vdev_add() for full details.
560 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
561 vdev_t
*tvd
= rvd
->vdev_child
[c
];
562 if (tvd
->vdev_ms_array
== 0)
563 vdev_metaslab_set_size(tvd
);
564 vdev_expand(tvd
, txg
);
567 spa_config_exit(spa
, SCL_ALL
, FTAG
);
570 * Wait for the mosconfig to be regenerated and synced.
572 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
575 * Update the global config cache to reflect the new mosconfig.
577 if (!spa
->spa_is_root
)
578 spa_config_sync(spa
, B_FALSE
, what
!= SPA_CONFIG_UPDATE_POOL
);
580 if (what
== SPA_CONFIG_UPDATE_POOL
)
581 spa_config_update(spa
, SPA_CONFIG_UPDATE_VDEVS
);
584 #if defined(_KERNEL) && defined(HAVE_SPL)
585 EXPORT_SYMBOL(spa_config_sync
);
586 EXPORT_SYMBOL(spa_config_load
);
587 EXPORT_SYMBOL(spa_all_configs
);
588 EXPORT_SYMBOL(spa_config_set
);
589 EXPORT_SYMBOL(spa_config_generate
);
590 EXPORT_SYMBOL(spa_config_update
);
592 module_param(spa_config_path
, charp
, 0444);
593 MODULE_PARM_DESC(spa_config_path
, "SPA config file (/etc/zfs/zpool.cache)");
595 module_param(zfs_autoimport_disable
, int, 0644);
596 MODULE_PARM_DESC(zfs_autoimport_disable
, "Disable pool import at module load");