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/systeminfo.h>
38 #include <sys/sunddi.h>
39 #include <sys/zfeature.h>
46 * Pool configuration repository.
48 * Pool configuration is stored as a packed nvlist on the filesystem. By
49 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
50 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
51 * property set that allows them to be stored in an alternate location until
52 * the control of external software.
54 * For each cache file, we have a single nvlist which holds all the
55 * configuration information. When the module loads, we read this information
56 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
57 * maintained independently in spa.c. Whenever the namespace is modified, or
58 * the configuration of a pool is changed, we call spa_write_cachefile(), which
59 * walks through all the active pools and writes the configuration to disk.
62 static uint64_t spa_config_generation
= 1;
65 * This can be overridden in userland to preserve an alternate namespace for
66 * userland pools when doing testing.
68 char *spa_config_path
= ZPOOL_CACHE
;
69 int zfs_autoimport_disable
= 1;
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 if (zfs_autoimport_disable
)
92 * Open the configuration file.
94 pathname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
96 (void) snprintf(pathname
, MAXPATHLEN
, "%s%s",
97 (rootdir
!= NULL
) ? "./" : "", spa_config_path
);
99 file
= kobj_open_file(pathname
);
101 kmem_free(pathname
, MAXPATHLEN
);
103 if (file
== (struct _buf
*)-1)
106 if (kobj_get_filesize(file
, &fsize
) != 0)
109 buf
= kmem_alloc(fsize
, KM_SLEEP
);
112 * Read the nvlist from the file.
114 if (kobj_read_file(file
, buf
, fsize
, 0) < 0)
120 if (nvlist_unpack(buf
, fsize
, &nvlist
, KM_SLEEP
) != 0)
124 * Iterate over all elements in the nvlist, creating a new spa_t for
125 * each one with the specified configuration.
127 mutex_enter(&spa_namespace_lock
);
129 while ((nvpair
= nvlist_next_nvpair(nvlist
, nvpair
)) != NULL
) {
130 if (nvpair_type(nvpair
) != DATA_TYPE_NVLIST
)
133 child
= fnvpair_value_nvlist(nvpair
);
135 if (spa_lookup(nvpair_name(nvpair
)) != NULL
)
137 (void) spa_add(nvpair_name(nvpair
), child
, NULL
);
139 mutex_exit(&spa_namespace_lock
);
145 kmem_free(buf
, fsize
);
147 kobj_close_file(file
);
151 spa_config_remove(spa_config_dirent_t
*dp
)
153 #if defined(__linux__) && defined(_KERNEL)
154 int error
, flags
= FWRITE
| FTRUNC
;
155 uio_seg_t seg
= UIO_SYSSPACE
;
158 error
= vn_open(dp
->scd_path
, seg
, flags
, 0644, &vp
, 0, 0);
160 (void) VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
161 (void) VOP_CLOSE(vp
, 0, 1, 0, kcred
, NULL
);
166 return (vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
));
171 spa_config_write(spa_config_dirent_t
*dp
, nvlist_t
*nvl
)
176 int oflags
= FWRITE
| FTRUNC
| FCREAT
| FOFFMAX
;
181 * If the nvlist is empty (NULL), then remove the old cachefile.
184 err
= spa_config_remove(dp
);
192 * Pack the configuration into a buffer.
194 buf
= fnvlist_pack(nvl
, &buflen
);
195 temp
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
197 #if defined(__linux__) && defined(_KERNEL)
199 * Write the configuration to disk. Due to the complexity involved
200 * in performing a rename and remove from within the kernel the file
201 * is instead truncated and overwritten in place. This way we always
202 * have a consistent view of the data or a zero length file.
204 err
= vn_open(dp
->scd_path
, UIO_SYSSPACE
, oflags
, 0644, &vp
, 0, 0);
206 err
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0,
207 UIO_SYSSPACE
, 0, RLIM64_INFINITY
, kcred
, NULL
);
209 err
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
211 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
213 (void) spa_config_remove(dp
);
217 * Write the configuration to disk. We need to do the traditional
218 * 'write to temporary file, sync, move over original' to make sure we
219 * always have a consistent view of the data.
221 (void) snprintf(temp
, MAXPATHLEN
, "%s.tmp", dp
->scd_path
);
223 err
= vn_open(temp
, UIO_SYSSPACE
, oflags
, 0644, &vp
, CRCREAT
, 0);
225 err
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0, UIO_SYSSPACE
,
226 0, RLIM64_INFINITY
, kcred
, NULL
);
228 err
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
230 err
= vn_rename(temp
, dp
->scd_path
, UIO_SYSSPACE
);
231 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
234 (void) vn_remove(temp
, UIO_SYSSPACE
, RMFILE
);
237 fnvlist_pack_free(buf
, buflen
);
238 kmem_free(temp
, MAXPATHLEN
);
243 * Synchronize pool configuration to disk. This must be called with the
244 * namespace lock held. Synchronizing the pool cache is typically done after
245 * the configuration has been synced to the MOS. This exposes a window where
246 * the MOS config will have been updated but the cache file has not. If
247 * the system were to crash at that instant then the cached config may not
248 * contain the correct information to open the pool and an explicit import
252 spa_write_cachefile(spa_t
*target
, boolean_t removing
, boolean_t postsysevent
)
254 spa_config_dirent_t
*dp
, *tdp
;
257 boolean_t ccw_failure
;
260 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
262 if (rootdir
== NULL
|| !(spa_mode_global
& FWRITE
))
266 * Iterate over all cachefiles for the pool, past or present. When the
267 * cachefile is changed, the new one is pushed onto this list, allowing
268 * us to update previous cachefiles that no longer contain this pool.
270 ccw_failure
= B_FALSE
;
271 for (dp
= list_head(&target
->spa_config_list
); dp
!= NULL
;
272 dp
= list_next(&target
->spa_config_list
, dp
)) {
274 if (dp
->scd_path
== NULL
)
278 * Iterate over all pools, adding any matching pools to 'nvl'.
281 while ((spa
= spa_next(spa
)) != NULL
) {
283 * Skip over our own pool if we're about to remove
284 * ourselves from the spa namespace or any pool that
285 * is readonly. Since we cannot guarantee that a
286 * readonly pool would successfully import upon reboot,
287 * we don't allow them to be written to the cache file.
289 if ((spa
== target
&& removing
) ||
293 mutex_enter(&spa
->spa_props_lock
);
294 tdp
= list_head(&spa
->spa_config_list
);
295 if (spa
->spa_config
== NULL
||
297 tdp
->scd_path
== NULL
||
298 strcmp(tdp
->scd_path
, dp
->scd_path
) != 0) {
299 mutex_exit(&spa
->spa_props_lock
);
304 nvl
= fnvlist_alloc();
306 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
)
307 pool_name
= fnvlist_lookup_string(
308 spa
->spa_config
, ZPOOL_CONFIG_POOL_NAME
);
310 pool_name
= spa_name(spa
);
312 fnvlist_add_nvlist(nvl
, pool_name
, spa
->spa_config
);
313 mutex_exit(&spa
->spa_props_lock
);
316 error
= spa_config_write(dp
, nvl
);
318 ccw_failure
= B_TRUE
;
324 * Keep trying so that configuration data is
325 * written if/when any temporary filesystem
326 * resource issues are resolved.
328 if (target
->spa_ccw_fail_time
== 0) {
329 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE
,
330 target
, NULL
, NULL
, NULL
, 0, 0);
332 target
->spa_ccw_fail_time
= gethrtime();
333 spa_async_request(target
, SPA_ASYNC_CONFIG_UPDATE
);
336 * Do not rate limit future attempts to update
339 target
->spa_ccw_fail_time
= 0;
343 * Remove any config entries older than the current one.
345 dp
= list_head(&target
->spa_config_list
);
346 while ((tdp
= list_next(&target
->spa_config_list
, dp
)) != NULL
) {
347 list_remove(&target
->spa_config_list
, tdp
);
348 if (tdp
->scd_path
!= NULL
)
349 spa_strfree(tdp
->scd_path
);
350 kmem_free(tdp
, sizeof (spa_config_dirent_t
));
353 spa_config_generation
++;
356 spa_event_notify(target
, NULL
, NULL
, ESC_ZFS_CONFIG_SYNC
);
360 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
361 * and we don't want to allow the local zone to see all the pools anyway.
362 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
363 * information for all pool visible within the zone.
366 spa_all_configs(uint64_t *generation
)
371 if (*generation
== spa_config_generation
)
374 pools
= fnvlist_alloc();
376 mutex_enter(&spa_namespace_lock
);
377 while ((spa
= spa_next(spa
)) != NULL
) {
378 if (INGLOBALZONE(curproc
) ||
379 zone_dataset_visible(spa_name(spa
), NULL
)) {
380 mutex_enter(&spa
->spa_props_lock
);
381 fnvlist_add_nvlist(pools
, spa_name(spa
),
383 mutex_exit(&spa
->spa_props_lock
);
386 *generation
= spa_config_generation
;
387 mutex_exit(&spa_namespace_lock
);
393 spa_config_set(spa_t
*spa
, nvlist_t
*config
)
395 mutex_enter(&spa
->spa_props_lock
);
396 nvlist_free(spa
->spa_config
);
397 spa
->spa_config
= config
;
398 mutex_exit(&spa
->spa_props_lock
);
402 * Generate the pool's configuration based on the current in-core state.
404 * We infer whether to generate a complete config or just one top-level config
405 * based on whether vd is the root vdev.
408 spa_config_generate(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int getstats
)
410 nvlist_t
*config
, *nvroot
;
411 vdev_t
*rvd
= spa
->spa_root_vdev
;
412 unsigned long hostid
= 0;
413 boolean_t locked
= B_FALSE
;
420 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
423 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
) ==
424 (SCL_CONFIG
| SCL_STATE
));
427 * If txg is -1, report the current value of spa->spa_config_txg.
430 txg
= spa
->spa_config_txg
;
433 * Originally, users had to handle spa namespace collisions by either
434 * exporting the already imported pool or by specifying a new name for
435 * the pool with a conflicting name. In the case of root pools from
436 * virtual guests, neither approach to collision resolution is
437 * reasonable. This is addressed by extending the new name syntax with
438 * an option to specify that the new name is temporary. When specified,
439 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
440 * to use the previous name, which we do below.
442 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
) {
443 VERIFY0(nvlist_lookup_string(spa
->spa_config
,
444 ZPOOL_CONFIG_POOL_NAME
, &pool_name
));
446 pool_name
= spa_name(spa
);
448 config
= fnvlist_alloc();
450 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
, spa_version(spa
));
451 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, pool_name
);
452 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, spa_state(spa
));
453 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, txg
);
454 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, spa_guid(spa
));
455 fnvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
, spa
->spa_errata
);
456 if (spa
->spa_comment
!= NULL
)
457 fnvlist_add_string(config
, ZPOOL_CONFIG_COMMENT
,
460 hostid
= spa_get_hostid();
462 fnvlist_add_uint64(config
, ZPOOL_CONFIG_HOSTID
, hostid
);
463 fnvlist_add_string(config
, ZPOOL_CONFIG_HOSTNAME
, utsname()->nodename
);
465 int config_gen_flags
= 0;
467 fnvlist_add_uint64(config
, ZPOOL_CONFIG_TOP_GUID
,
468 vd
->vdev_top
->vdev_guid
);
469 fnvlist_add_uint64(config
, ZPOOL_CONFIG_GUID
,
471 if (vd
->vdev_isspare
)
472 fnvlist_add_uint64(config
,
473 ZPOOL_CONFIG_IS_SPARE
, 1ULL);
475 fnvlist_add_uint64(config
,
476 ZPOOL_CONFIG_IS_LOG
, 1ULL);
477 vd
= vd
->vdev_top
; /* label contains top config */
480 * Only add the (potentially large) split information
481 * in the mos config, and not in the vdev labels
483 if (spa
->spa_config_splitting
!= NULL
)
484 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
485 spa
->spa_config_splitting
);
487 fnvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
);
489 config_gen_flags
|= VDEV_CONFIG_MOS
;
493 * Add the top-level config. We even add this on pools which
494 * don't support holes in the namespace.
496 vdev_top_config_generate(spa
, config
);
499 * If we're splitting, record the original pool's guid.
501 if (spa
->spa_config_splitting
!= NULL
&&
502 nvlist_lookup_uint64(spa
->spa_config_splitting
,
503 ZPOOL_CONFIG_SPLIT_GUID
, &split_guid
) == 0) {
504 fnvlist_add_uint64(config
, ZPOOL_CONFIG_SPLIT_GUID
, split_guid
);
507 nvroot
= vdev_config_generate(spa
, vd
, getstats
, config_gen_flags
);
508 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
);
512 * Store what's necessary for reading the MOS in the label.
514 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
515 spa
->spa_label_features
);
517 if (getstats
&& spa_load_state(spa
) == SPA_LOAD_NONE
) {
518 ddt_histogram_t
*ddh
;
522 ddh
= kmem_zalloc(sizeof (ddt_histogram_t
), KM_SLEEP
);
523 ddt_get_dedup_histogram(spa
, ddh
);
524 fnvlist_add_uint64_array(config
,
525 ZPOOL_CONFIG_DDT_HISTOGRAM
,
526 (uint64_t *)ddh
, sizeof (*ddh
) / sizeof (uint64_t));
527 kmem_free(ddh
, sizeof (ddt_histogram_t
));
529 ddo
= kmem_zalloc(sizeof (ddt_object_t
), KM_SLEEP
);
530 ddt_get_dedup_object_stats(spa
, ddo
);
531 fnvlist_add_uint64_array(config
,
532 ZPOOL_CONFIG_DDT_OBJ_STATS
,
533 (uint64_t *)ddo
, sizeof (*ddo
) / sizeof (uint64_t));
534 kmem_free(ddo
, sizeof (ddt_object_t
));
536 dds
= kmem_zalloc(sizeof (ddt_stat_t
), KM_SLEEP
);
537 ddt_get_dedup_stats(spa
, dds
);
538 fnvlist_add_uint64_array(config
,
539 ZPOOL_CONFIG_DDT_STATS
,
540 (uint64_t *)dds
, sizeof (*dds
) / sizeof (uint64_t));
541 kmem_free(dds
, sizeof (ddt_stat_t
));
545 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
551 * Update all disk labels, generate a fresh config based on the current
552 * in-core state, and sync the global config cache (do not sync the config
553 * cache if this is a booting rootpool).
556 spa_config_update(spa_t
*spa
, int what
)
558 vdev_t
*rvd
= spa
->spa_root_vdev
;
562 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
564 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
565 txg
= spa_last_synced_txg(spa
) + 1;
566 if (what
== SPA_CONFIG_UPDATE_POOL
) {
567 vdev_config_dirty(rvd
);
570 * If we have top-level vdevs that were added but have
571 * not yet been prepared for allocation, do that now.
572 * (It's safe now because the config cache is up to date,
573 * so it will be able to translate the new DVAs.)
574 * See comments in spa_vdev_add() for full details.
576 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
577 vdev_t
*tvd
= rvd
->vdev_child
[c
];
578 if (tvd
->vdev_ms_array
== 0)
579 vdev_metaslab_set_size(tvd
);
580 vdev_expand(tvd
, txg
);
583 spa_config_exit(spa
, SCL_ALL
, FTAG
);
586 * Wait for the mosconfig to be regenerated and synced.
588 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
591 * Update the global config cache to reflect the new mosconfig.
593 if (!spa
->spa_is_root
) {
594 spa_write_cachefile(spa
, B_FALSE
,
595 what
!= SPA_CONFIG_UPDATE_POOL
);
598 if (what
== SPA_CONFIG_UPDATE_POOL
)
599 spa_config_update(spa
, SPA_CONFIG_UPDATE_VDEVS
);
602 #if defined(_KERNEL) && defined(HAVE_SPL)
603 EXPORT_SYMBOL(spa_config_load
);
604 EXPORT_SYMBOL(spa_all_configs
);
605 EXPORT_SYMBOL(spa_config_set
);
606 EXPORT_SYMBOL(spa_config_generate
);
607 EXPORT_SYMBOL(spa_config_update
);
609 module_param(spa_config_path
, charp
, 0444);
610 MODULE_PARM_DESC(spa_config_path
, "SPA config file (/etc/zfs/zpool.cache)");
612 module_param(zfs_autoimport_disable
, int, 0644);
613 MODULE_PARM_DESC(zfs_autoimport_disable
, "Disable pool import at module load");