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) 2012 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/utsname.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 const char *spa_config_path
= ZPOOL_CACHE
;
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 * Open the configuration file.
87 pathname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
89 (void) snprintf(pathname
, MAXPATHLEN
, "%s%s",
90 (rootdir
!= NULL
) ? "./" : "", spa_config_path
);
92 file
= kobj_open_file(pathname
);
94 kmem_free(pathname
, MAXPATHLEN
);
96 if (file
== (struct _buf
*)-1)
99 if (kobj_get_filesize(file
, &fsize
) != 0)
102 buf
= kmem_alloc(fsize
, KM_SLEEP
);
105 * Read the nvlist from the file.
107 if (kobj_read_file(file
, buf
, fsize
, 0) < 0)
113 if (nvlist_unpack(buf
, fsize
, &nvlist
, KM_SLEEP
) != 0)
117 * Iterate over all elements in the nvlist, creating a new spa_t for
118 * each one with the specified configuration.
120 mutex_enter(&spa_namespace_lock
);
122 while ((nvpair
= nvlist_next_nvpair(nvlist
, nvpair
)) != NULL
) {
123 if (nvpair_type(nvpair
) != DATA_TYPE_NVLIST
)
126 VERIFY(nvpair_value_nvlist(nvpair
, &child
) == 0);
128 if (spa_lookup(nvpair_name(nvpair
)) != NULL
)
130 (void) spa_add(nvpair_name(nvpair
), child
, NULL
);
132 mutex_exit(&spa_namespace_lock
);
138 kmem_free(buf
, fsize
);
140 kobj_close_file(file
);
144 spa_config_write(spa_config_dirent_t
*dp
, nvlist_t
*nvl
)
149 int oflags
= FWRITE
| FTRUNC
| FCREAT
| FOFFMAX
;
153 * If the nvlist is empty (NULL), then remove the old cachefile.
156 (void) vn_remove(dp
->scd_path
, UIO_SYSSPACE
, RMFILE
);
161 * Pack the configuration into a buffer.
163 VERIFY(nvlist_size(nvl
, &buflen
, NV_ENCODE_XDR
) == 0);
165 buf
= kmem_alloc(buflen
, KM_SLEEP
);
166 temp
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
168 VERIFY(nvlist_pack(nvl
, &buf
, &buflen
, NV_ENCODE_XDR
,
172 * Write the configuration to disk. We need to do the traditional
173 * 'write to temporary file, sync, move over original' to make sure we
174 * always have a consistent view of the data.
176 (void) snprintf(temp
, MAXPATHLEN
, "%s.tmp", dp
->scd_path
);
178 if (vn_open(temp
, UIO_SYSSPACE
, oflags
, 0644, &vp
, CRCREAT
, 0) == 0) {
179 if (vn_rdwr(UIO_WRITE
, vp
, buf
, buflen
, 0, UIO_SYSSPACE
,
180 0, RLIM64_INFINITY
, kcred
, NULL
) == 0 &&
181 VOP_FSYNC(vp
, FSYNC
, kcred
, NULL
) == 0) {
182 (void) 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 kmem_free(buf
, buflen
);
191 kmem_free(temp
, MAXPATHLEN
);
195 * Synchronize pool configuration to disk. This must be called with the
196 * namespace lock held.
199 spa_config_sync(spa_t
*target
, boolean_t removing
, boolean_t postsysevent
)
201 spa_config_dirent_t
*dp
, *tdp
;
204 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
206 if (rootdir
== NULL
|| !(spa_mode_global
& FWRITE
))
210 * Iterate over all cachefiles for the pool, past or present. When the
211 * cachefile is changed, the new one is pushed onto this list, allowing
212 * us to update previous cachefiles that no longer contain this pool.
214 for (dp
= list_head(&target
->spa_config_list
); dp
!= NULL
;
215 dp
= list_next(&target
->spa_config_list
, dp
)) {
217 if (dp
->scd_path
== NULL
)
221 * Iterate over all pools, adding any matching pools to 'nvl'.
224 while ((spa
= spa_next(spa
)) != NULL
) {
226 * Skip over our own pool if we're about to remove
227 * ourselves from the spa namespace or any pool that
228 * is readonly. Since we cannot guarantee that a
229 * readonly pool would successfully import upon reboot,
230 * we don't allow them to be written to the cache file.
232 if ((spa
== target
&& removing
) ||
236 mutex_enter(&spa
->spa_props_lock
);
237 tdp
= list_head(&spa
->spa_config_list
);
238 if (spa
->spa_config
== NULL
||
239 tdp
->scd_path
== NULL
||
240 strcmp(tdp
->scd_path
, dp
->scd_path
) != 0) {
241 mutex_exit(&spa
->spa_props_lock
);
246 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
,
249 VERIFY(nvlist_add_nvlist(nvl
, spa
->spa_name
,
250 spa
->spa_config
) == 0);
251 mutex_exit(&spa
->spa_props_lock
);
254 spa_config_write(dp
, nvl
);
259 * Remove any config entries older than the current one.
261 dp
= list_head(&target
->spa_config_list
);
262 while ((tdp
= list_next(&target
->spa_config_list
, dp
)) != NULL
) {
263 list_remove(&target
->spa_config_list
, tdp
);
264 if (tdp
->scd_path
!= NULL
)
265 spa_strfree(tdp
->scd_path
);
266 kmem_free(tdp
, sizeof (spa_config_dirent_t
));
269 spa_config_generation
++;
272 spa_event_notify(target
, NULL
, ESC_ZFS_CONFIG_SYNC
);
276 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
277 * and we don't want to allow the local zone to see all the pools anyway.
278 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
279 * information for all pool visible within the zone.
282 spa_all_configs(uint64_t *generation
)
287 if (*generation
== spa_config_generation
)
290 VERIFY(nvlist_alloc(&pools
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
292 mutex_enter(&spa_namespace_lock
);
293 while ((spa
= spa_next(spa
)) != NULL
) {
294 if (INGLOBALZONE(curproc
) ||
295 zone_dataset_visible(spa_name(spa
), NULL
)) {
296 mutex_enter(&spa
->spa_props_lock
);
297 VERIFY(nvlist_add_nvlist(pools
, spa_name(spa
),
298 spa
->spa_config
) == 0);
299 mutex_exit(&spa
->spa_props_lock
);
302 *generation
= spa_config_generation
;
303 mutex_exit(&spa_namespace_lock
);
309 spa_config_set(spa_t
*spa
, nvlist_t
*config
)
311 mutex_enter(&spa
->spa_props_lock
);
312 if (spa
->spa_config
!= NULL
)
313 nvlist_free(spa
->spa_config
);
314 spa
->spa_config
= config
;
315 mutex_exit(&spa
->spa_props_lock
);
319 * Generate the pool's configuration based on the current in-core state.
320 * We infer whether to generate a complete config or just one top-level config
321 * based on whether vd is the root vdev.
324 spa_config_generate(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int getstats
)
326 nvlist_t
*config
, *nvroot
;
327 vdev_t
*rvd
= spa
->spa_root_vdev
;
328 unsigned long hostid
= 0;
329 boolean_t locked
= B_FALSE
;
335 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
338 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
) ==
339 (SCL_CONFIG
| SCL_STATE
));
342 * If txg is -1, report the current value of spa->spa_config_txg.
345 txg
= spa
->spa_config_txg
;
347 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
349 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
350 spa_version(spa
)) == 0);
351 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
352 spa_name(spa
)) == 0);
353 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
354 spa_state(spa
)) == 0);
355 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
357 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
358 spa_guid(spa
)) == 0);
359 VERIFY(spa
->spa_comment
== NULL
|| nvlist_add_string(config
,
360 ZPOOL_CONFIG_COMMENT
, spa
->spa_comment
) == 0);
364 hostid
= zone_get_hostid(NULL
);
367 * We're emulating the system's hostid in userland, so we can't use
370 (void) ddi_strtoul(hw_serial
, NULL
, 10, &hostid
);
373 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_HOSTID
,
376 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_HOSTNAME
,
377 utsname
.nodename
) == 0);
380 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TOP_GUID
,
381 vd
->vdev_top
->vdev_guid
) == 0);
382 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_GUID
,
383 vd
->vdev_guid
) == 0);
384 if (vd
->vdev_isspare
)
385 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_IS_SPARE
,
388 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_IS_LOG
,
390 vd
= vd
->vdev_top
; /* label contains top config */
393 * Only add the (potentially large) split information
394 * in the mos config, and not in the vdev labels
396 if (spa
->spa_config_splitting
!= NULL
)
397 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
398 spa
->spa_config_splitting
) == 0);
402 * Add the top-level config. We even add this on pools which
403 * don't support holes in the namespace.
405 vdev_top_config_generate(spa
, config
);
408 * If we're splitting, record the original pool's guid.
410 if (spa
->spa_config_splitting
!= NULL
&&
411 nvlist_lookup_uint64(spa
->spa_config_splitting
,
412 ZPOOL_CONFIG_SPLIT_GUID
, &split_guid
) == 0) {
413 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_SPLIT_GUID
,
417 nvroot
= vdev_config_generate(spa
, vd
, getstats
, 0);
418 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
422 * Store what's necessary for reading the MOS in the label.
424 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
425 spa
->spa_label_features
) == 0);
427 if (getstats
&& spa_load_state(spa
) == SPA_LOAD_NONE
) {
428 ddt_histogram_t
*ddh
;
432 ddh
= kmem_zalloc(sizeof (ddt_histogram_t
), KM_SLEEP
);
433 ddt_get_dedup_histogram(spa
, ddh
);
434 VERIFY(nvlist_add_uint64_array(config
,
435 ZPOOL_CONFIG_DDT_HISTOGRAM
,
436 (uint64_t *)ddh
, sizeof (*ddh
) / sizeof (uint64_t)) == 0);
437 kmem_free(ddh
, sizeof (ddt_histogram_t
));
439 ddo
= kmem_zalloc(sizeof (ddt_object_t
), KM_SLEEP
);
440 ddt_get_dedup_object_stats(spa
, ddo
);
441 VERIFY(nvlist_add_uint64_array(config
,
442 ZPOOL_CONFIG_DDT_OBJ_STATS
,
443 (uint64_t *)ddo
, sizeof (*ddo
) / sizeof (uint64_t)) == 0);
444 kmem_free(ddo
, sizeof (ddt_object_t
));
446 dds
= kmem_zalloc(sizeof (ddt_stat_t
), KM_SLEEP
);
447 ddt_get_dedup_stats(spa
, dds
);
448 VERIFY(nvlist_add_uint64_array(config
,
449 ZPOOL_CONFIG_DDT_STATS
,
450 (uint64_t *)dds
, sizeof (*dds
) / sizeof (uint64_t)) == 0);
451 kmem_free(dds
, sizeof (ddt_stat_t
));
455 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
461 * Update all disk labels, generate a fresh config based on the current
462 * in-core state, and sync the global config cache (do not sync the config
463 * cache if this is a booting rootpool).
466 spa_config_update(spa_t
*spa
, int what
)
468 vdev_t
*rvd
= spa
->spa_root_vdev
;
472 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
474 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
475 txg
= spa_last_synced_txg(spa
) + 1;
476 if (what
== SPA_CONFIG_UPDATE_POOL
) {
477 vdev_config_dirty(rvd
);
480 * If we have top-level vdevs that were added but have
481 * not yet been prepared for allocation, do that now.
482 * (It's safe now because the config cache is up to date,
483 * so it will be able to translate the new DVAs.)
484 * See comments in spa_vdev_add() for full details.
486 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
487 vdev_t
*tvd
= rvd
->vdev_child
[c
];
488 if (tvd
->vdev_ms_array
== 0)
489 vdev_metaslab_set_size(tvd
);
490 vdev_expand(tvd
, txg
);
493 spa_config_exit(spa
, SCL_ALL
, FTAG
);
496 * Wait for the mosconfig to be regenerated and synced.
498 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
501 * Update the global config cache to reflect the new mosconfig.
503 if (!spa
->spa_is_root
)
504 spa_config_sync(spa
, B_FALSE
, what
!= SPA_CONFIG_UPDATE_POOL
);
506 if (what
== SPA_CONFIG_UPDATE_POOL
)
507 spa_config_update(spa
, SPA_CONFIG_UPDATE_VDEVS
);