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_config_sync(), 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_write(spa_config_dirent_t
*dp
, nvlist_t
*nvl
)
156 int oflags
= FWRITE
| FTRUNC
| FCREAT
| FOFFMAX
;
161 * If the nvlist is empty (NULL), then remove the old cachefile.
164 err
= vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
169 * Pack the configuration into a buffer.
171 buf
= fnvlist_pack(nvl
, &buflen
);
172 temp
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
174 #if defined(__linux__) && defined(_KERNEL)
176 * Write the configuration to disk. Due to the complexity involved
177 * in performing a rename from within the kernel the file is truncated
178 * and overwritten in place. In the event of an error the file is
179 * unlinked to make sure we always have a consistent view of the data.
181 err
= vn_open(dp
->scd_path
, UIO_SYSSPACE
, oflags
, 0644, &vp
, 0, 0);
183 err
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0,
184 UIO_SYSSPACE
, 0, RLIM64_INFINITY
, kcred
, NULL
);
186 err
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
188 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
191 (void) vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
195 * Write the configuration to disk. We need to do the traditional
196 * 'write to temporary file, sync, move over original' to make sure we
197 * always have a consistent view of the data.
199 (void) snprintf(temp
, MAXPATHLEN
, "%s.tmp", dp
->scd_path
);
201 err
= vn_open(temp
, UIO_SYSSPACE
, oflags
, 0644, &vp
, CRCREAT
, 0);
203 err
= vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0, UIO_SYSSPACE
,
204 0, RLIM64_INFINITY
, kcred
, NULL
);
206 err
= VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
);
208 err
= vn_rename(temp
, dp
->scd_path
, UIO_SYSSPACE
);
209 (void) VOP_CLOSE(vp
, oflags
, 1, 0, kcred
, NULL
);
212 (void) vn_remove(temp
, UIO_SYSSPACE
, RMFILE
);
215 fnvlist_pack_free(buf
, buflen
);
216 kmem_free(temp
, MAXPATHLEN
);
221 * Synchronize pool configuration to disk. This must be called with the
222 * namespace lock held. Synchronizing the pool cache is typically done after
223 * the configuration has been synced to the MOS. This exposes a window where
224 * the MOS config will have been updated but the cache file has not. If
225 * the system were to crash at that instant then the cached config may not
226 * contain the correct information to open the pool and an explicit import
230 spa_config_sync(spa_t
*target
, boolean_t removing
, boolean_t postsysevent
)
232 spa_config_dirent_t
*dp
, *tdp
;
235 boolean_t ccw_failure
;
238 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
240 if (rootdir
== NULL
|| !(spa_mode_global
& FWRITE
))
244 * Iterate over all cachefiles for the pool, past or present. When the
245 * cachefile is changed, the new one is pushed onto this list, allowing
246 * us to update previous cachefiles that no longer contain this pool.
248 ccw_failure
= B_FALSE
;
249 for (dp
= list_head(&target
->spa_config_list
); dp
!= NULL
;
250 dp
= list_next(&target
->spa_config_list
, dp
)) {
252 if (dp
->scd_path
== NULL
)
256 * Iterate over all pools, adding any matching pools to 'nvl'.
259 while ((spa
= spa_next(spa
)) != NULL
) {
261 * Skip over our own pool if we're about to remove
262 * ourselves from the spa namespace or any pool that
263 * is readonly. Since we cannot guarantee that a
264 * readonly pool would successfully import upon reboot,
265 * we don't allow them to be written to the cache file.
267 if ((spa
== target
&& removing
) ||
271 mutex_enter(&spa
->spa_props_lock
);
272 tdp
= list_head(&spa
->spa_config_list
);
273 if (spa
->spa_config
== NULL
||
275 tdp
->scd_path
== NULL
||
276 strcmp(tdp
->scd_path
, dp
->scd_path
) != 0) {
277 mutex_exit(&spa
->spa_props_lock
);
282 nvl
= fnvlist_alloc();
284 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
)
285 pool_name
= fnvlist_lookup_string(
286 spa
->spa_config
, ZPOOL_CONFIG_POOL_NAME
);
288 pool_name
= spa_name(spa
);
290 fnvlist_add_nvlist(nvl
, pool_name
, spa
->spa_config
);
291 mutex_exit(&spa
->spa_props_lock
);
294 error
= spa_config_write(dp
, nvl
);
296 ccw_failure
= B_TRUE
;
302 * Keep trying so that configuration data is
303 * written if/when any temporary filesystem
304 * resource issues are resolved.
306 if (target
->spa_ccw_fail_time
== 0) {
307 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE
,
308 target
, NULL
, NULL
, 0, 0);
310 target
->spa_ccw_fail_time
= gethrtime();
311 spa_async_request(target
, SPA_ASYNC_CONFIG_UPDATE
);
314 * Do not rate limit future attempts to update
317 target
->spa_ccw_fail_time
= 0;
321 * Remove any config entries older than the current one.
323 dp
= list_head(&target
->spa_config_list
);
324 while ((tdp
= list_next(&target
->spa_config_list
, dp
)) != NULL
) {
325 list_remove(&target
->spa_config_list
, tdp
);
326 if (tdp
->scd_path
!= NULL
)
327 spa_strfree(tdp
->scd_path
);
328 kmem_free(tdp
, sizeof (spa_config_dirent_t
));
331 spa_config_generation
++;
334 spa_event_notify(target
, NULL
, NULL
, ESC_ZFS_CONFIG_SYNC
);
338 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
339 * and we don't want to allow the local zone to see all the pools anyway.
340 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
341 * information for all pool visible within the zone.
344 spa_all_configs(uint64_t *generation
)
349 if (*generation
== spa_config_generation
)
352 pools
= fnvlist_alloc();
354 mutex_enter(&spa_namespace_lock
);
355 while ((spa
= spa_next(spa
)) != NULL
) {
356 if (INGLOBALZONE(curproc
) ||
357 zone_dataset_visible(spa_name(spa
), NULL
)) {
358 mutex_enter(&spa
->spa_props_lock
);
359 fnvlist_add_nvlist(pools
, spa_name(spa
),
361 mutex_exit(&spa
->spa_props_lock
);
364 *generation
= spa_config_generation
;
365 mutex_exit(&spa_namespace_lock
);
371 spa_config_set(spa_t
*spa
, nvlist_t
*config
)
373 mutex_enter(&spa
->spa_props_lock
);
374 nvlist_free(spa
->spa_config
);
375 spa
->spa_config
= config
;
376 mutex_exit(&spa
->spa_props_lock
);
380 * Generate the pool's configuration based on the current in-core state.
382 * We infer whether to generate a complete config or just one top-level config
383 * based on whether vd is the root vdev.
386 spa_config_generate(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int getstats
)
388 nvlist_t
*config
, *nvroot
;
389 vdev_t
*rvd
= spa
->spa_root_vdev
;
390 unsigned long hostid
= 0;
391 boolean_t locked
= B_FALSE
;
394 int config_gen_flags
= 0;
399 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
402 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
) ==
403 (SCL_CONFIG
| SCL_STATE
));
406 * If txg is -1, report the current value of spa->spa_config_txg.
409 txg
= spa
->spa_config_txg
;
412 * Originally, users had to handle spa namespace collisions by either
413 * exporting the already imported pool or by specifying a new name for
414 * the pool with a conflicting name. In the case of root pools from
415 * virtual guests, neither approach to collision resolution is
416 * reasonable. This is addressed by extending the new name syntax with
417 * an option to specify that the new name is temporary. When specified,
418 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
419 * to use the previous name, which we do below.
421 if (spa
->spa_import_flags
& ZFS_IMPORT_TEMP_NAME
) {
422 VERIFY0(nvlist_lookup_string(spa
->spa_config
,
423 ZPOOL_CONFIG_POOL_NAME
, &pool_name
));
425 pool_name
= spa_name(spa
);
427 config
= fnvlist_alloc();
429 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
, spa_version(spa
));
430 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, pool_name
);
431 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, spa_state(spa
));
432 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, txg
);
433 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, spa_guid(spa
));
434 fnvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
, spa
->spa_errata
);
435 if (spa
->spa_comment
!= NULL
)
436 fnvlist_add_string(config
, ZPOOL_CONFIG_COMMENT
,
439 hostid
= spa_get_hostid();
441 fnvlist_add_uint64(config
, ZPOOL_CONFIG_HOSTID
, hostid
);
442 fnvlist_add_string(config
, ZPOOL_CONFIG_HOSTNAME
, utsname()->nodename
);
445 fnvlist_add_uint64(config
, ZPOOL_CONFIG_TOP_GUID
,
446 vd
->vdev_top
->vdev_guid
);
447 fnvlist_add_uint64(config
, ZPOOL_CONFIG_GUID
,
449 if (vd
->vdev_isspare
)
450 fnvlist_add_uint64(config
,
451 ZPOOL_CONFIG_IS_SPARE
, 1ULL);
453 fnvlist_add_uint64(config
,
454 ZPOOL_CONFIG_IS_LOG
, 1ULL);
455 vd
= vd
->vdev_top
; /* label contains top config */
458 * Only add the (potentially large) split information
459 * in the mos config, and not in the vdev labels
461 if (spa
->spa_config_splitting
!= NULL
)
462 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
463 spa
->spa_config_splitting
);
465 fnvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
);
467 config_gen_flags
|= VDEV_CONFIG_MOS
;
471 * Add the top-level config. We even add this on pools which
472 * don't support holes in the namespace.
474 vdev_top_config_generate(spa
, config
);
477 * If we're splitting, record the original pool's guid.
479 if (spa
->spa_config_splitting
!= NULL
&&
480 nvlist_lookup_uint64(spa
->spa_config_splitting
,
481 ZPOOL_CONFIG_SPLIT_GUID
, &split_guid
) == 0) {
482 fnvlist_add_uint64(config
, ZPOOL_CONFIG_SPLIT_GUID
, split_guid
);
485 nvroot
= vdev_config_generate(spa
, vd
, getstats
, config_gen_flags
);
486 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
);
490 * Store what's necessary for reading the MOS in the label.
492 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
493 spa
->spa_label_features
);
495 if (getstats
&& spa_load_state(spa
) == SPA_LOAD_NONE
) {
496 ddt_histogram_t
*ddh
;
500 ddh
= kmem_zalloc(sizeof (ddt_histogram_t
), KM_SLEEP
);
501 ddt_get_dedup_histogram(spa
, ddh
);
502 fnvlist_add_uint64_array(config
,
503 ZPOOL_CONFIG_DDT_HISTOGRAM
,
504 (uint64_t *)ddh
, sizeof (*ddh
) / sizeof (uint64_t));
505 kmem_free(ddh
, sizeof (ddt_histogram_t
));
507 ddo
= kmem_zalloc(sizeof (ddt_object_t
), KM_SLEEP
);
508 ddt_get_dedup_object_stats(spa
, ddo
);
509 fnvlist_add_uint64_array(config
,
510 ZPOOL_CONFIG_DDT_OBJ_STATS
,
511 (uint64_t *)ddo
, sizeof (*ddo
) / sizeof (uint64_t));
512 kmem_free(ddo
, sizeof (ddt_object_t
));
514 dds
= kmem_zalloc(sizeof (ddt_stat_t
), KM_SLEEP
);
515 ddt_get_dedup_stats(spa
, dds
);
516 fnvlist_add_uint64_array(config
,
517 ZPOOL_CONFIG_DDT_STATS
,
518 (uint64_t *)dds
, sizeof (*dds
) / sizeof (uint64_t));
519 kmem_free(dds
, sizeof (ddt_stat_t
));
523 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
529 * Update all disk labels, generate a fresh config based on the current
530 * in-core state, and sync the global config cache (do not sync the config
531 * cache if this is a booting rootpool).
534 spa_config_update(spa_t
*spa
, int what
)
536 vdev_t
*rvd
= spa
->spa_root_vdev
;
540 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
542 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
543 txg
= spa_last_synced_txg(spa
) + 1;
544 if (what
== SPA_CONFIG_UPDATE_POOL
) {
545 vdev_config_dirty(rvd
);
548 * If we have top-level vdevs that were added but have
549 * not yet been prepared for allocation, do that now.
550 * (It's safe now because the config cache is up to date,
551 * so it will be able to translate the new DVAs.)
552 * See comments in spa_vdev_add() for full details.
554 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
555 vdev_t
*tvd
= rvd
->vdev_child
[c
];
556 if (tvd
->vdev_ms_array
== 0)
557 vdev_metaslab_set_size(tvd
);
558 vdev_expand(tvd
, txg
);
561 spa_config_exit(spa
, SCL_ALL
, FTAG
);
564 * Wait for the mosconfig to be regenerated and synced.
566 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
569 * Update the global config cache to reflect the new mosconfig.
571 if (!spa
->spa_is_root
)
572 spa_config_sync(spa
, B_FALSE
, what
!= SPA_CONFIG_UPDATE_POOL
);
574 if (what
== SPA_CONFIG_UPDATE_POOL
)
575 spa_config_update(spa
, SPA_CONFIG_UPDATE_VDEVS
);
578 #if defined(_KERNEL) && defined(HAVE_SPL)
579 EXPORT_SYMBOL(spa_config_sync
);
580 EXPORT_SYMBOL(spa_config_load
);
581 EXPORT_SYMBOL(spa_all_configs
);
582 EXPORT_SYMBOL(spa_config_set
);
583 EXPORT_SYMBOL(spa_config_generate
);
584 EXPORT_SYMBOL(spa_config_update
);
586 module_param(spa_config_path
, charp
, 0444);
587 MODULE_PARM_DESC(spa_config_path
, "SPA config file (/etc/zfs/zpool.cache)");
589 module_param(zfs_autoimport_disable
, int, 0644);
590 MODULE_PARM_DESC(zfs_autoimport_disable
, "Disable pool import at module load");