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) 2013 by Delphix. All rights reserved.
29 #include <sys/spa_impl.h>
30 #include <sys/nvpair.h>
32 #include <sys/fs/zfs.h>
33 #include <sys/vdev_impl.h>
34 #include <sys/zfs_ioctl.h>
35 #include <sys/systeminfo.h>
36 #include <sys/sunddi.h>
37 #include <sys/zfeature.h>
44 * Pool configuration repository.
46 * Pool configuration is stored as a packed nvlist on the filesystem. By
47 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
48 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
49 * property set that allows them to be stored in an alternate location until
50 * the control of external software.
52 * For each cache file, we have a single nvlist which holds all the
53 * configuration information. When the module loads, we read this information
54 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
55 * maintained independently in spa.c. Whenever the namespace is modified, or
56 * the configuration of a pool is changed, we call spa_config_sync(), which
57 * walks through all the active pools and writes the configuration to disk.
60 static uint64_t spa_config_generation
= 1;
63 * This can be overridden in userland to preserve an alternate namespace for
64 * userland pools when doing testing.
66 char *spa_config_path
= ZPOOL_CACHE
;
67 int zfs_autoimport_disable
= 1;
70 * Called when the module is first loaded, this routine loads the configuration
71 * file into the SPA namespace. It does not actually open or load the pools; it
72 * only populates the namespace.
78 nvlist_t
*nvlist
, *child
;
85 if (zfs_autoimport_disable
)
90 * Open the configuration file.
92 pathname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
94 (void) snprintf(pathname
, MAXPATHLEN
, "%s%s",
95 (rootdir
!= NULL
) ? "./" : "", spa_config_path
);
97 file
= kobj_open_file(pathname
);
99 kmem_free(pathname
, MAXPATHLEN
);
101 if (file
== (struct _buf
*)-1)
104 if (kobj_get_filesize(file
, &fsize
) != 0)
107 buf
= kmem_alloc(fsize
, KM_SLEEP
);
110 * Read the nvlist from the file.
112 if (kobj_read_file(file
, buf
, fsize
, 0) < 0)
118 if (nvlist_unpack(buf
, fsize
, &nvlist
, KM_SLEEP
) != 0)
122 * Iterate over all elements in the nvlist, creating a new spa_t for
123 * each one with the specified configuration.
125 mutex_enter(&spa_namespace_lock
);
127 while ((nvpair
= nvlist_next_nvpair(nvlist
, nvpair
)) != NULL
) {
128 if (nvpair_type(nvpair
) != DATA_TYPE_NVLIST
)
131 VERIFY(nvpair_value_nvlist(nvpair
, &child
) == 0);
133 if (spa_lookup(nvpair_name(nvpair
)) != NULL
)
135 (void) spa_add(nvpair_name(nvpair
), child
, NULL
);
137 mutex_exit(&spa_namespace_lock
);
143 kmem_free(buf
, fsize
);
145 kobj_close_file(file
);
149 spa_config_write(spa_config_dirent_t
*dp
, nvlist_t
*nvl
)
154 int oflags
= FWRITE
| FTRUNC
| FCREAT
| FOFFMAX
;
159 * If the nvlist is empty (NULL), then remove the old cachefile.
162 (void) vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
167 * Pack the configuration into a buffer.
169 VERIFY(nvlist_size(nvl
, &buflen
, NV_ENCODE_XDR
) == 0);
171 buf
= vmem_alloc(buflen
, KM_SLEEP
);
172 temp
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
174 VERIFY(nvlist_pack(nvl
, &buf
, &buflen
, NV_ENCODE_XDR
,
177 #if defined(__linux__) && defined(_KERNEL)
179 * Write the configuration to disk. Due to the complexity involved
180 * in performing a rename from within the kernel the file is truncated
181 * and overwritten in place. In the event of an error the file is
182 * unlinked to make sure we always have a consistent view of the data.
184 error
= vn_open(dp
->scd_path
, UIO_SYSSPACE
, oflags
, 0644, &vp
, 0, 0);
186 error
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0,
187 UIO_SYSSPACE
, 0, RLIM64_INFINITY
, kcred
, NULL
);
189 error
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
191 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
194 (void) vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
198 * Write the configuration to disk. We need to do the traditional
199 * 'write to temporary file, sync, move over original' to make sure we
200 * always have a consistent view of the data.
202 (void) snprintf(temp
, MAXPATHLEN
, "%s.tmp", dp
->scd_path
);
204 error
= vn_open(temp
, UIO_SYSSPACE
, oflags
, 0644, &vp
, CRCREAT
, 0);
206 if (vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0, UIO_SYSSPACE
,
207 0, RLIM64_INFINITY
, kcred
, NULL
) == 0 &&
208 VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
) == 0) {
209 (void) vn_rename(temp
, dp
->scd_path
, UIO_SYSSPACE
);
211 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
214 (void) vn_remove(temp
, UIO_SYSSPACE
, RMFILE
);
217 vmem_free(buf
, buflen
);
218 kmem_free(temp
, MAXPATHLEN
);
222 * Synchronize pool configuration to disk. This must be called with the
223 * namespace lock held. Synchronizing the pool cache is typically done after
224 * the configuration has been synced to the MOS. This exposes a window where
225 * the MOS config will have been updated but the cache file has not. If
226 * the system were to crash at that instant then the cached config may not
227 * contain the correct information to open the pool and an explicity import
231 spa_config_sync(spa_t
*target
, boolean_t removing
, boolean_t postsysevent
)
233 spa_config_dirent_t
*dp
, *tdp
;
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 for (dp
= list_head(&target
->spa_config_list
); dp
!= NULL
;
248 dp
= list_next(&target
->spa_config_list
, dp
)) {
250 if (dp
->scd_path
== NULL
)
254 * Iterate over all pools, adding any matching pools to 'nvl'.
257 while ((spa
= spa_next(spa
)) != NULL
) {
259 * Skip over our own pool if we're about to remove
260 * ourselves from the spa namespace or any pool that
261 * is readonly. Since we cannot guarantee that a
262 * readonly pool would successfully import upon reboot,
263 * we don't allow them to be written to the cache file.
265 if ((spa
== target
&& removing
) ||
269 mutex_enter(&spa
->spa_props_lock
);
270 tdp
= list_head(&spa
->spa_config_list
);
271 if (spa
->spa_config
== NULL
||
272 tdp
->scd_path
== NULL
||
273 strcmp(tdp
->scd_path
, dp
->scd_path
) != 0) {
274 mutex_exit(&spa
->spa_props_lock
);
279 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
,
282 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
) {
283 VERIFY0(nvlist_lookup_string(spa
->spa_config
,
284 ZPOOL_CONFIG_POOL_NAME
, &pool_name
));
286 pool_name
= spa_name(spa
);
288 VERIFY(nvlist_add_nvlist(nvl
, pool_name
,
289 spa
->spa_config
) == 0);
290 mutex_exit(&spa
->spa_props_lock
);
293 spa_config_write(dp
, nvl
);
298 * Remove any config entries older than the current one.
300 dp
= list_head(&target
->spa_config_list
);
301 while ((tdp
= list_next(&target
->spa_config_list
, dp
)) != NULL
) {
302 list_remove(&target
->spa_config_list
, tdp
);
303 if (tdp
->scd_path
!= NULL
)
304 spa_strfree(tdp
->scd_path
);
305 kmem_free(tdp
, sizeof (spa_config_dirent_t
));
308 spa_config_generation
++;
311 spa_event_notify(target
, NULL
, FM_EREPORT_ZFS_CONFIG_SYNC
);
315 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
316 * and we don't want to allow the local zone to see all the pools anyway.
317 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
318 * information for all pool visible within the zone.
321 spa_all_configs(uint64_t *generation
)
326 if (*generation
== spa_config_generation
)
329 VERIFY(nvlist_alloc(&pools
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
331 mutex_enter(&spa_namespace_lock
);
332 while ((spa
= spa_next(spa
)) != NULL
) {
333 if (INGLOBALZONE(curproc
) ||
334 zone_dataset_visible(spa_name(spa
), NULL
)) {
335 mutex_enter(&spa
->spa_props_lock
);
336 VERIFY(nvlist_add_nvlist(pools
, spa_name(spa
),
337 spa
->spa_config
) == 0);
338 mutex_exit(&spa
->spa_props_lock
);
341 *generation
= spa_config_generation
;
342 mutex_exit(&spa_namespace_lock
);
348 spa_config_set(spa_t
*spa
, nvlist_t
*config
)
350 mutex_enter(&spa
->spa_props_lock
);
351 if (spa
->spa_config
!= NULL
)
352 nvlist_free(spa
->spa_config
);
353 spa
->spa_config
= config
;
354 mutex_exit(&spa
->spa_props_lock
);
358 * Generate the pool's configuration based on the current in-core state.
360 * We infer whether to generate a complete config or just one top-level config
361 * based on whether vd is the root vdev.
364 spa_config_generate(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int getstats
)
366 nvlist_t
*config
, *nvroot
;
367 vdev_t
*rvd
= spa
->spa_root_vdev
;
368 unsigned long hostid
= 0;
369 boolean_t locked
= B_FALSE
;
376 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
379 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
) ==
380 (SCL_CONFIG
| SCL_STATE
));
383 * If txg is -1, report the current value of spa->spa_config_txg.
386 txg
= spa
->spa_config_txg
;
389 * Originally, users had to handle spa namespace collisions by either
390 * exporting the already imported pool or by specifying a new name for
391 * the pool with a conflicting name. In the case of root pools from
392 * virtual guests, neither approach to collision resolution is
393 * reasonable. This is addressed by extending the new name syntax with
394 * an option to specify that the new name is temporary. When specified,
395 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
396 * to use the previous name, which we do below.
398 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
) {
399 VERIFY0(nvlist_lookup_string(spa
->spa_config
,
400 ZPOOL_CONFIG_POOL_NAME
, &pool_name
));
402 pool_name
= spa_name(spa
);
404 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
406 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
407 spa_version(spa
)) == 0);
408 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
410 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
411 spa_state(spa
)) == 0);
412 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
414 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
415 spa_guid(spa
)) == 0);
416 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
417 spa
->spa_errata
) == 0);
418 VERIFY(spa
->spa_comment
== NULL
|| nvlist_add_string(config
,
419 ZPOOL_CONFIG_COMMENT
, spa
->spa_comment
) == 0);
423 hostid
= zone_get_hostid(NULL
);
426 * We're emulating the system's hostid in userland, so we can't use
429 (void) ddi_strtoul(hw_serial
, NULL
, 10, &hostid
);
432 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_HOSTID
,
435 VERIFY0(nvlist_add_string(config
, ZPOOL_CONFIG_HOSTNAME
,
436 utsname()->nodename
));
439 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TOP_GUID
,
440 vd
->vdev_top
->vdev_guid
) == 0);
441 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_GUID
,
442 vd
->vdev_guid
) == 0);
443 if (vd
->vdev_isspare
)
444 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_IS_SPARE
,
447 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_IS_LOG
,
449 vd
= vd
->vdev_top
; /* label contains top config */
452 * Only add the (potentially large) split information
453 * in the mos config, and not in the vdev labels
455 if (spa
->spa_config_splitting
!= NULL
)
456 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
457 spa
->spa_config_splitting
) == 0);
461 * Add the top-level config. We even add this on pools which
462 * don't support holes in the namespace.
464 vdev_top_config_generate(spa
, config
);
467 * If we're splitting, record the original pool's guid.
469 if (spa
->spa_config_splitting
!= NULL
&&
470 nvlist_lookup_uint64(spa
->spa_config_splitting
,
471 ZPOOL_CONFIG_SPLIT_GUID
, &split_guid
) == 0) {
472 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_SPLIT_GUID
,
476 nvroot
= vdev_config_generate(spa
, vd
, getstats
, 0);
477 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
481 * Store what's necessary for reading the MOS in the label.
483 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
484 spa
->spa_label_features
) == 0);
486 if (getstats
&& spa_load_state(spa
) == SPA_LOAD_NONE
) {
487 ddt_histogram_t
*ddh
;
491 ddh
= kmem_zalloc(sizeof (ddt_histogram_t
), KM_SLEEP
);
492 ddt_get_dedup_histogram(spa
, ddh
);
493 VERIFY(nvlist_add_uint64_array(config
,
494 ZPOOL_CONFIG_DDT_HISTOGRAM
,
495 (uint64_t *)ddh
, sizeof (*ddh
) / sizeof (uint64_t)) == 0);
496 kmem_free(ddh
, sizeof (ddt_histogram_t
));
498 ddo
= kmem_zalloc(sizeof (ddt_object_t
), KM_SLEEP
);
499 ddt_get_dedup_object_stats(spa
, ddo
);
500 VERIFY(nvlist_add_uint64_array(config
,
501 ZPOOL_CONFIG_DDT_OBJ_STATS
,
502 (uint64_t *)ddo
, sizeof (*ddo
) / sizeof (uint64_t)) == 0);
503 kmem_free(ddo
, sizeof (ddt_object_t
));
505 dds
= kmem_zalloc(sizeof (ddt_stat_t
), KM_SLEEP
);
506 ddt_get_dedup_stats(spa
, dds
);
507 VERIFY(nvlist_add_uint64_array(config
,
508 ZPOOL_CONFIG_DDT_STATS
,
509 (uint64_t *)dds
, sizeof (*dds
) / sizeof (uint64_t)) == 0);
510 kmem_free(dds
, sizeof (ddt_stat_t
));
514 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
520 * Update all disk labels, generate a fresh config based on the current
521 * in-core state, and sync the global config cache (do not sync the config
522 * cache if this is a booting rootpool).
525 spa_config_update(spa_t
*spa
, int what
)
527 vdev_t
*rvd
= spa
->spa_root_vdev
;
531 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
533 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
534 txg
= spa_last_synced_txg(spa
) + 1;
535 if (what
== SPA_CONFIG_UPDATE_POOL
) {
536 vdev_config_dirty(rvd
);
539 * If we have top-level vdevs that were added but have
540 * not yet been prepared for allocation, do that now.
541 * (It's safe now because the config cache is up to date,
542 * so it will be able to translate the new DVAs.)
543 * See comments in spa_vdev_add() for full details.
545 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
546 vdev_t
*tvd
= rvd
->vdev_child
[c
];
547 if (tvd
->vdev_ms_array
== 0)
548 vdev_metaslab_set_size(tvd
);
549 vdev_expand(tvd
, txg
);
552 spa_config_exit(spa
, SCL_ALL
, FTAG
);
555 * Wait for the mosconfig to be regenerated and synced.
557 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
560 * Update the global config cache to reflect the new mosconfig.
562 if (!spa
->spa_is_root
)
563 spa_config_sync(spa
, B_FALSE
, what
!= SPA_CONFIG_UPDATE_POOL
);
565 if (what
== SPA_CONFIG_UPDATE_POOL
)
566 spa_config_update(spa
, SPA_CONFIG_UPDATE_VDEVS
);
569 #if defined(_KERNEL) && defined(HAVE_SPL)
570 EXPORT_SYMBOL(spa_config_sync
);
571 EXPORT_SYMBOL(spa_config_load
);
572 EXPORT_SYMBOL(spa_all_configs
);
573 EXPORT_SYMBOL(spa_config_set
);
574 EXPORT_SYMBOL(spa_config_generate
);
575 EXPORT_SYMBOL(spa_config_update
);
577 module_param(spa_config_path
, charp
, 0444);
578 MODULE_PARM_DESC(spa_config_path
, "SPA config file (/etc/zfs/zpool.cache)");
580 module_param(zfs_autoimport_disable
, int, 0644);
581 MODULE_PARM_DESC(zfs_autoimport_disable
, "Disable pool import at module load");