spa_misc.c revision 40feaa914945406e86e193599d115ea71a171d18
1fa9e406ahrens/*
2fa9e406ahrens * CDDL HEADER START
3fa9e406ahrens *
4fa9e406ahrens * The contents of this file are subject to the terms of the
5ea8dc4beschrock * Common Development and Distribution License (the "License").
6ea8dc4beschrock * You may not use this file except in compliance with the License.
7fa9e406ahrens *
8fa9e406ahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9fa9e406ahrens * or http://www.opensolaris.org/os/licensing.
10fa9e406ahrens * See the License for the specific language governing permissions
11fa9e406ahrens * and limitations under the License.
12fa9e406ahrens *
13fa9e406ahrens * When distributing Covered Code, include this CDDL HEADER in each
14fa9e406ahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15fa9e406ahrens * If applicable, add the following below this CDDL HEADER, with the
16fa9e406ahrens * fields enclosed by brackets "[]" replaced with your own identifying
17fa9e406ahrens * information: Portions Copyright [yyyy] [name of copyright owner]
18fa9e406ahrens *
19fa9e406ahrens * CDDL HEADER END
20fa9e406ahrens */
21fa9e406ahrens/*
2239c2341eschrock * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23fa9e406ahrens * Use is subject to license terms.
24fa9e406ahrens */
25fa9e406ahrens
26fa9e406ahrens#pragma ident	"%Z%%M%	%I%	%E% SMI"
27fa9e406ahrens
28fa9e406ahrens#include <sys/zfs_context.h>
29fa9e406ahrens#include <sys/spa_impl.h>
30fa9e406ahrens#include <sys/zio.h>
31fa9e406ahrens#include <sys/zio_checksum.h>
32fa9e406ahrens#include <sys/zio_compress.h>
33fa9e406ahrens#include <sys/dmu.h>
34fa9e406ahrens#include <sys/dmu_tx.h>
35fa9e406ahrens#include <sys/zap.h>
36fa9e406ahrens#include <sys/zil.h>
37fa9e406ahrens#include <sys/vdev_impl.h>
38fa9e406ahrens#include <sys/metaslab.h>
39fa9e406ahrens#include <sys/uberblock_impl.h>
40fa9e406ahrens#include <sys/txg.h>
41fa9e406ahrens#include <sys/avl.h>
42fa9e406ahrens#include <sys/unique.h>
43fa9e406ahrens#include <sys/dsl_pool.h>
44fa9e406ahrens#include <sys/dsl_dir.h>
45fa9e406ahrens#include <sys/dsl_prop.h>
46fa9e406ahrens#include <sys/fs/zfs.h>
47fa9e406ahrens
48fa9e406ahrens/*
49fa9e406ahrens * SPA locking
50fa9e406ahrens *
51fa9e406ahrens * There are four basic locks for managing spa_t structures:
52fa9e406ahrens *
53fa9e406ahrens * spa_namespace_lock (global mutex)
54fa9e406ahrens *
5544cd46cbillm *	This lock must be acquired to do any of the following:
56fa9e406ahrens *
5744cd46cbillm *		- Lookup a spa_t by name
5844cd46cbillm *		- Add or remove a spa_t from the namespace
5944cd46cbillm *		- Increase spa_refcount from non-zero
6044cd46cbillm *		- Check if spa_refcount is zero
6144cd46cbillm *		- Rename a spa_t
62ea8dc4beschrock *		- add/remove/attach/detach devices
6344cd46cbillm *		- Held for the duration of create/destroy/import/export
64fa9e406ahrens *
6544cd46cbillm *	It does not need to handle recursion.  A create or destroy may
6644cd46cbillm *	reference objects (files or zvols) in other pools, but by
6744cd46cbillm *	definition they must have an existing reference, and will never need
6844cd46cbillm *	to lookup a spa_t by name.
69fa9e406ahrens *
70fa9e406ahrens * spa_refcount (per-spa refcount_t protected by mutex)
71fa9e406ahrens *
7244cd46cbillm *	This reference count keep track of any active users of the spa_t.  The
7344cd46cbillm *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
7444cd46cbillm *	the refcount is never really 'zero' - opening a pool implicitly keeps
7544cd46cbillm *	some references in the DMU.  Internally we check against SPA_MINREF, but
7644cd46cbillm *	present the image of a zero/non-zero value to consumers.
77fa9e406ahrens *
78fa9e406ahrens * spa_config_lock (per-spa crazy rwlock)
79fa9e406ahrens *
8044cd46cbillm *	This SPA special is a recursive rwlock, capable of being acquired from
8144cd46cbillm *	asynchronous threads.  It has protects the spa_t from config changes,
8244cd46cbillm *	and must be held in the following circumstances:
83fa9e406ahrens *
8444cd46cbillm *		- RW_READER to perform I/O to the spa
8544cd46cbillm *		- RW_WRITER to change the vdev config
86fa9e406ahrens *
87fa9e406ahrens * spa_config_cache_lock (per-spa mutex)
88fa9e406ahrens *
8944cd46cbillm *	This mutex prevents the spa_config nvlist from being updated.  No
90fa9e406ahrens *      other locks are required to obtain this lock, although implicitly you
91fa9e406ahrens *      must have the namespace lock or non-zero refcount to have any kind
92fa9e406ahrens *      of spa_t pointer at all.
93fa9e406ahrens *
94fa9e406ahrens * The locking order is fairly straightforward:
95fa9e406ahrens *
9644cd46cbillm *		spa_namespace_lock	->	spa_refcount
97fa9e406ahrens *
9844cd46cbillm *	The namespace lock must be acquired to increase the refcount from 0
9944cd46cbillm *	or to check if it is zero.
100fa9e406ahrens *
10144cd46cbillm *		spa_refcount		->	spa_config_lock
102fa9e406ahrens *
10344cd46cbillm *	There must be at least one valid reference on the spa_t to acquire
10444cd46cbillm *	the config lock.
105fa9e406ahrens *
10644cd46cbillm *		spa_namespace_lock	->	spa_config_lock
107fa9e406ahrens *
10844cd46cbillm *	The namespace lock must always be taken before the config lock.
109fa9e406ahrens *
110fa9e406ahrens *
111fa9e406ahrens * The spa_namespace_lock and spa_config_cache_lock can be acquired directly and
112fa9e406ahrens * are globally visible.
113fa9e406ahrens *
114fa9e406ahrens * The namespace is manipulated using the following functions, all which require
115fa9e406ahrens * the spa_namespace_lock to be held.
116fa9e406ahrens *
11744cd46cbillm *	spa_lookup()		Lookup a spa_t by name.
118fa9e406ahrens *
11944cd46cbillm *	spa_add()		Create a new spa_t in the namespace.
120fa9e406ahrens *
12144cd46cbillm *	spa_remove()		Remove a spa_t from the namespace.  This also
12244cd46cbillm *				frees up any memory associated with the spa_t.
123fa9e406ahrens *
12444cd46cbillm *	spa_next()		Returns the next spa_t in the system, or the
12544cd46cbillm *				first if NULL is passed.
126fa9e406ahrens *
12744cd46cbillm *	spa_evict_all()		Shutdown and remove all spa_t structures in
12844cd46cbillm *				the system.
129fa9e406ahrens *
130ea8dc4beschrock *	spa_guid_exists()	Determine whether a pool/device guid exists.
131fa9e406ahrens *
132fa9e406ahrens * The spa_refcount is manipulated using the following functions:
133fa9e406ahrens *
13444cd46cbillm *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
13544cd46cbillm *				called with spa_namespace_lock held if the
13644cd46cbillm *				refcount is currently zero.
137fa9e406ahrens *
13844cd46cbillm *	spa_close()		Remove a reference from the spa_t.  This will
13944cd46cbillm *				not free the spa_t or remove it from the
14044cd46cbillm *				namespace.  No locking is required.
141fa9e406ahrens *
14244cd46cbillm *	spa_refcount_zero()	Returns true if the refcount is currently
14344cd46cbillm *				zero.  Must be called with spa_namespace_lock
14444cd46cbillm *				held.
145fa9e406ahrens *
146fa9e406ahrens * The spa_config_lock is manipulated using the following functions:
147fa9e406ahrens *
14844cd46cbillm *	spa_config_enter()	Acquire the config lock as RW_READER or
14944cd46cbillm *				RW_WRITER.  At least one reference on the spa_t
15044cd46cbillm *				must exist.
151fa9e406ahrens *
15244cd46cbillm *	spa_config_exit()	Release the config lock.
153fa9e406ahrens *
15444cd46cbillm *	spa_config_held()	Returns true if the config lock is currently
15544cd46cbillm *				held in the given state.
156fa9e406ahrens *
157ea8dc4beschrock * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
158fa9e406ahrens *
15944cd46cbillm *	spa_vdev_enter()	Acquire the namespace lock and the config lock
160ea8dc4beschrock *				for writing.
161fa9e406ahrens *
16244cd46cbillm *	spa_vdev_exit()		Release the config lock, wait for all I/O
16344cd46cbillm *				to complete, sync the updated configs to the
164ea8dc4beschrock *				cache, and release the namespace lock.
165fa9e406ahrens *
166fa9e406ahrens * The spa_name() function also requires either the spa_namespace_lock
167fa9e406ahrens * or the spa_config_lock, as both are needed to do a rename.  spa_rename() is
168fa9e406ahrens * also implemented within this file since is requires manipulation of the
169fa9e406ahrens * namespace.
170fa9e406ahrens */
171fa9e406ahrens
172fa9e406ahrensstatic avl_tree_t spa_namespace_avl;
173fa9e406ahrenskmutex_t spa_namespace_lock;
174fa9e406ahrensstatic kcondvar_t spa_namespace_cv;
1750373e76bonwickstatic int spa_active_count;
176416e0cdekint spa_max_replication_override = SPA_DVAS_PER_BP;
177fa9e406ahrens
17899653d4eschrockstatic kmutex_t spa_spare_lock;
17939c2341eschrockstatic avl_tree_t spa_spare_avl;
18099653d4eschrock
181fa9e406ahrenskmem_cache_t *spa_buffer_pool;
182fa9e406ahrensint spa_mode;
183fa9e406ahrens
184fa9e406ahrens#ifdef ZFS_DEBUG
18540feaa9ahrens/* Everything except dprintf is on by default in debug builds */
18640feaa9ahrensint zfs_flags = ~ZFS_DEBUG_DPRINTF;
187fa9e406ahrens#else
188fa9e406ahrensint zfs_flags = 0;
189fa9e406ahrens#endif
190fa9e406ahrens
1910125049ahrens/*
1920125049ahrens * zfs_recover can be set to nonzero to attempt to recover from
1930125049ahrens * otherwise-fatal errors, typically caused by on-disk corruption.  When
1940125049ahrens * set, calls to zfs_panic_recover() will turn into warning messages.
1950125049ahrens */
1960125049ahrensint zfs_recover = 0;
1970125049ahrens
198fa9e406ahrens#define	SPA_MINREF	5	/* spa_refcnt for an open-but-idle pool */
199fa9e406ahrens
200fa9e406ahrens/*
201fa9e406ahrens * ==========================================================================
202fa9e406ahrens * SPA namespace functions
203fa9e406ahrens * ==========================================================================
204fa9e406ahrens */
205fa9e406ahrens
206fa9e406ahrens/*
207fa9e406ahrens * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
208fa9e406ahrens * Returns NULL if no matching spa_t is found.
209fa9e406ahrens */
210fa9e406ahrensspa_t *
211fa9e406ahrensspa_lookup(const char *name)
212fa9e406ahrens{
213fa9e406ahrens	spa_t search, *spa;
214fa9e406ahrens	avl_index_t where;
21540feaa9ahrens	char c;
21640feaa9ahrens	char *cp;
217fa9e406ahrens
218fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
219fa9e406ahrens
22040feaa9ahrens	/*
22140feaa9ahrens	 * If it's a full dataset name, figure out the pool name and
22240feaa9ahrens	 * just use that.
22340feaa9ahrens	 */
22440feaa9ahrens	cp = strpbrk(name, "/@");
22540feaa9ahrens	if (cp) {
22640feaa9ahrens		c = *cp;
22740feaa9ahrens		*cp = '\0';
22840feaa9ahrens	}
22940feaa9ahrens
230fa9e406ahrens	search.spa_name = (char *)name;
231fa9e406ahrens	spa = avl_find(&spa_namespace_avl, &search, &where);
232fa9e406ahrens
23340feaa9ahrens	if (cp)
23440feaa9ahrens		*cp = c;
23540feaa9ahrens
236fa9e406ahrens	return (spa);
237fa9e406ahrens}
238fa9e406ahrens
239fa9e406ahrens/*
240fa9e406ahrens * Create an uninitialized spa_t with the given name.  Requires
241fa9e406ahrens * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
242fa9e406ahrens * exist by calling spa_lookup() first.
243fa9e406ahrens */
244fa9e406ahrensspa_t *
2450373e76bonwickspa_add(const char *name, const char *altroot)
246fa9e406ahrens{
247fa9e406ahrens	spa_t *spa;
248fa9e406ahrens
249fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
250fa9e406ahrens
251fa9e406ahrens	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
252fa9e406ahrens
253fa9e406ahrens	spa->spa_name = spa_strdup(name);
254fa9e406ahrens	spa->spa_state = POOL_STATE_UNINITIALIZED;
255fa9e406ahrens	spa->spa_freeze_txg = UINT64_MAX;
2560373e76bonwick	spa->spa_final_txg = UINT64_MAX;
257fa9e406ahrens
258fa9e406ahrens	refcount_create(&spa->spa_refcount);
259ea8dc4beschrock	refcount_create(&spa->spa_config_lock.scl_count);
260fa9e406ahrens
261fa9e406ahrens	avl_add(&spa_namespace_avl, spa);
262fa9e406ahrens
2630373e76bonwick	/*
2640373e76bonwick	 * Set the alternate root, if there is one.
2650373e76bonwick	 */
2660373e76bonwick	if (altroot) {
2670373e76bonwick		spa->spa_root = spa_strdup(altroot);
2680373e76bonwick		spa_active_count++;
2690373e76bonwick	}
2700373e76bonwick
271fa9e406ahrens	return (spa);
272fa9e406ahrens}
273fa9e406ahrens
274fa9e406ahrens/*
275fa9e406ahrens * Removes a spa_t from the namespace, freeing up any memory used.  Requires
276fa9e406ahrens * spa_namespace_lock.  This is called only after the spa_t has been closed and
277fa9e406ahrens * deactivated.
278fa9e406ahrens */
279fa9e406ahrensvoid
280fa9e406ahrensspa_remove(spa_t *spa)
281fa9e406ahrens{
282fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
283fa9e406ahrens	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
284fa9e406ahrens	ASSERT(spa->spa_scrub_thread == NULL);
285fa9e406ahrens
286fa9e406ahrens	avl_remove(&spa_namespace_avl, spa);
287fa9e406ahrens	cv_broadcast(&spa_namespace_cv);
288fa9e406ahrens
2890373e76bonwick	if (spa->spa_root) {
290fa9e406ahrens		spa_strfree(spa->spa_root);
2910373e76bonwick		spa_active_count--;
2920373e76bonwick	}
293fa9e406ahrens
294fa9e406ahrens	if (spa->spa_name)
295fa9e406ahrens		spa_strfree(spa->spa_name);
296fa9e406ahrens
297fa9e406ahrens	spa_config_set(spa, NULL);
298fa9e406ahrens
299fa9e406ahrens	refcount_destroy(&spa->spa_refcount);
300ea8dc4beschrock	refcount_destroy(&spa->spa_config_lock.scl_count);
301fa9e406ahrens
3025ad8204nd	mutex_destroy(&spa->spa_sync_bplist.bpl_lock);
3035ad8204nd	mutex_destroy(&spa->spa_config_lock.scl_lock);
3045ad8204nd	mutex_destroy(&spa->spa_errlist_lock);
3055ad8204nd	mutex_destroy(&spa->spa_errlog_lock);
3065ad8204nd	mutex_destroy(&spa->spa_scrub_lock);
3075ad8204nd	mutex_destroy(&spa->spa_config_cache_lock);
3085ad8204nd	mutex_destroy(&spa->spa_async_lock);
30906eeb2aek	mutex_destroy(&spa->spa_history_lock);
310b1b8ab3lling	mutex_destroy(&spa->spa_props_lock);
3115ad8204nd
312fa9e406ahrens	kmem_free(spa, sizeof (spa_t));
313fa9e406ahrens}
314fa9e406ahrens
315fa9e406ahrens/*
316fa9e406ahrens * Given a pool, return the next pool in the namespace, or NULL if there is
317fa9e406ahrens * none.  If 'prev' is NULL, return the first pool.
318fa9e406ahrens */
319fa9e406ahrensspa_t *
320fa9e406ahrensspa_next(spa_t *prev)
321fa9e406ahrens{
322fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
323fa9e406ahrens
324fa9e406ahrens	if (prev)
325fa9e406ahrens		return (AVL_NEXT(&spa_namespace_avl, prev));
326fa9e406ahrens	else
327fa9e406ahrens		return (avl_first(&spa_namespace_avl));
328fa9e406ahrens}
329fa9e406ahrens
330fa9e406ahrens/*
331fa9e406ahrens * ==========================================================================
332fa9e406ahrens * SPA refcount functions
333fa9e406ahrens * ==========================================================================
334fa9e406ahrens */
335fa9e406ahrens
336fa9e406ahrens/*
337fa9e406ahrens * Add a reference to the given spa_t.  Must have at least one reference, or
338fa9e406ahrens * have the namespace lock held.
339fa9e406ahrens */
340fa9e406ahrensvoid
341fa9e406ahrensspa_open_ref(spa_t *spa, void *tag)
342fa9e406ahrens{
343fa9e406ahrens	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
344fa9e406ahrens	    MUTEX_HELD(&spa_namespace_lock));
345fa9e406ahrens
346fa9e406ahrens	(void) refcount_add(&spa->spa_refcount, tag);
347fa9e406ahrens}
348fa9e406ahrens
349fa9e406ahrens/*
350fa9e406ahrens * Remove a reference to the given spa_t.  Must have at least one reference, or
351fa9e406ahrens * have the namespace lock held.
352fa9e406ahrens */
353fa9e406ahrensvoid
354fa9e406ahrensspa_close(spa_t *spa, void *tag)
355fa9e406ahrens{
356fa9e406ahrens	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
357fa9e406ahrens	    MUTEX_HELD(&spa_namespace_lock));
358fa9e406ahrens
359fa9e406ahrens	(void) refcount_remove(&spa->spa_refcount, tag);
360fa9e406ahrens}
361fa9e406ahrens
362fa9e406ahrens/*
363fa9e406ahrens * Check to see if the spa refcount is zero.  Must be called with
364fa9e406ahrens * spa_namespace_lock held.  We really compare against SPA_MINREF, which is the
365fa9e406ahrens * number of references acquired when opening a pool
366fa9e406ahrens */
367fa9e406ahrensboolean_t
368fa9e406ahrensspa_refcount_zero(spa_t *spa)
369fa9e406ahrens{
370fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
371fa9e406ahrens
372fa9e406ahrens	return (refcount_count(&spa->spa_refcount) == SPA_MINREF);
373fa9e406ahrens}
374fa9e406ahrens
375fa9e406ahrens/*
376fa9e406ahrens * ==========================================================================
37799653d4eschrock * SPA spare tracking
37899653d4eschrock * ==========================================================================
37999653d4eschrock */
38099653d4eschrock
38199653d4eschrock/*
38239c2341eschrock * Spares are tracked globally due to the following constraints:
38339c2341eschrock *
38439c2341eschrock * 	- A spare may be part of multiple pools.
38539c2341eschrock * 	- A spare may be added to a pool even if it's actively in use within
38639c2341eschrock *	  another pool.
38739c2341eschrock * 	- A spare in use in any pool can only be the source of a replacement if
38839c2341eschrock *	  the target is a spare in the same pool.
38939c2341eschrock *
39039c2341eschrock * We keep track of all spares on the system through the use of a reference
39139c2341eschrock * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
39239c2341eschrock * spare, then we bump the reference count in the AVL tree.  In addition, we set
39339c2341eschrock * the 'vdev_isspare' member to indicate that the device is a spare (active or
39439c2341eschrock * inactive).  When a spare is made active (used to replace a device in the
39539c2341eschrock * pool), we also keep track of which pool its been made a part of.
39639c2341eschrock *
39739c2341eschrock * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
39839c2341eschrock * called under the spa_namespace lock as part of vdev reconfiguration.  The
39939c2341eschrock * separate spare lock exists for the status query path, which does not need to
40039c2341eschrock * be completely consistent with respect to other vdev configuration changes.
40199653d4eschrock */
40239c2341eschrock
40399653d4eschrocktypedef struct spa_spare {
40499653d4eschrock	uint64_t	spare_guid;
40539c2341eschrock	uint64_t	spare_pool;
40699653d4eschrock	avl_node_t	spare_avl;
40799653d4eschrock	int		spare_count;
40899653d4eschrock} spa_spare_t;
40999653d4eschrock
41099653d4eschrockstatic int
41199653d4eschrockspa_spare_compare(const void *a, const void *b)
41299653d4eschrock{
41399653d4eschrock	const spa_spare_t *sa = a;
41499653d4eschrock	const spa_spare_t *sb = b;
41599653d4eschrock
41699653d4eschrock	if (sa->spare_guid < sb->spare_guid)
41799653d4eschrock		return (-1);
41899653d4eschrock	else if (sa->spare_guid > sb->spare_guid)
41999653d4eschrock		return (1);
42099653d4eschrock	else
42199653d4eschrock		return (0);
42299653d4eschrock}
42399653d4eschrock
42499653d4eschrockvoid
42539c2341eschrockspa_spare_add(vdev_t *vd)
42699653d4eschrock{
42799653d4eschrock	avl_index_t where;
42899653d4eschrock	spa_spare_t search;
42999653d4eschrock	spa_spare_t *spare;
43099653d4eschrock
43199653d4eschrock	mutex_enter(&spa_spare_lock);
43239c2341eschrock	ASSERT(!vd->vdev_isspare);
43399653d4eschrock
43439c2341eschrock	search.spare_guid = vd->vdev_guid;
43599653d4eschrock	if ((spare = avl_find(&spa_spare_avl, &search, &where)) != NULL) {
43699653d4eschrock		spare->spare_count++;
43799653d4eschrock	} else {
43839c2341eschrock		spare = kmem_zalloc(sizeof (spa_spare_t), KM_SLEEP);
43939c2341eschrock		spare->spare_guid = vd->vdev_guid;
44099653d4eschrock		spare->spare_count = 1;
44199653d4eschrock		avl_insert(&spa_spare_avl, spare, where);
44299653d4eschrock	}
44339c2341eschrock	vd->vdev_isspare = B_TRUE;
44499653d4eschrock
44599653d4eschrock	mutex_exit(&spa_spare_lock);
44699653d4eschrock}
44799653d4eschrock
44899653d4eschrockvoid
44939c2341eschrockspa_spare_remove(vdev_t *vd)
45099653d4eschrock{
45199653d4eschrock	spa_spare_t search;
45299653d4eschrock	spa_spare_t *spare;
45399653d4eschrock	avl_index_t where;
45499653d4eschrock
45599653d4eschrock	mutex_enter(&spa_spare_lock);
45699653d4eschrock
45739c2341eschrock	search.spare_guid = vd->vdev_guid;
45899653d4eschrock	spare = avl_find(&spa_spare_avl, &search, &where);
45999653d4eschrock
46039c2341eschrock	ASSERT(vd->vdev_isspare);
46199653d4eschrock	ASSERT(spare != NULL);
46299653d4eschrock
46399653d4eschrock	if (--spare->spare_count == 0) {
46499653d4eschrock		avl_remove(&spa_spare_avl, spare);
46599653d4eschrock		kmem_free(spare, sizeof (spa_spare_t));
46639c2341eschrock	} else if (spare->spare_pool == spa_guid(vd->vdev_spa)) {
46739c2341eschrock		spare->spare_pool = 0ULL;
46899653d4eschrock	}
46999653d4eschrock
47039c2341eschrock	vd->vdev_isspare = B_FALSE;
47199653d4eschrock	mutex_exit(&spa_spare_lock);
47299653d4eschrock}
47399653d4eschrock
47499653d4eschrockboolean_t
47539c2341eschrockspa_spare_exists(uint64_t guid, uint64_t *pool)
47699653d4eschrock{
47739c2341eschrock	spa_spare_t search, *found;
47899653d4eschrock	avl_index_t where;
47999653d4eschrock
48099653d4eschrock	mutex_enter(&spa_spare_lock);
48199653d4eschrock
48299653d4eschrock	search.spare_guid = guid;
48339c2341eschrock	found = avl_find(&spa_spare_avl, &search, &where);
48439c2341eschrock
48539c2341eschrock	if (pool) {
48639c2341eschrock		if (found)
48739c2341eschrock			*pool = found->spare_pool;
48839c2341eschrock		else
48939c2341eschrock			*pool = 0ULL;
49039c2341eschrock	}
49199653d4eschrock
49299653d4eschrock	mutex_exit(&spa_spare_lock);
49399653d4eschrock
49439c2341eschrock	return (found != NULL);
49539c2341eschrock}
49639c2341eschrock
49739c2341eschrockvoid
49839c2341eschrockspa_spare_activate(vdev_t *vd)
49939c2341eschrock{
50039c2341eschrock	spa_spare_t search, *found;
50139c2341eschrock	avl_index_t where;
50239c2341eschrock
50339c2341eschrock	mutex_enter(&spa_spare_lock);
50439c2341eschrock	ASSERT(vd->vdev_isspare);
50539c2341eschrock
50639c2341eschrock	search.spare_guid = vd->vdev_guid;
50739c2341eschrock	found = avl_find(&spa_spare_avl, &search, &where);
50839c2341eschrock	ASSERT(found != NULL);
50939c2341eschrock	ASSERT(found->spare_pool == 0ULL);
51039c2341eschrock
51139c2341eschrock	found->spare_pool = spa_guid(vd->vdev_spa);
51239c2341eschrock	mutex_exit(&spa_spare_lock);
51399653d4eschrock}
51499653d4eschrock
51599653d4eschrock/*
51699653d4eschrock * ==========================================================================
517fa9e406ahrens * SPA config locking
518fa9e406ahrens * ==========================================================================
519fa9e406ahrens */
520fa9e406ahrens
521fa9e406ahrens/*
522fa9e406ahrens * Acquire the config lock.  The config lock is a special rwlock that allows for
523fa9e406ahrens * recursive enters.  Because these enters come from the same thread as well as
524fa9e406ahrens * asynchronous threads working on behalf of the owner, we must unilaterally
525fa9e406ahrens * allow all reads access as long at least one reader is held (even if a write
526fa9e406ahrens * is requested).  This has the side effect of write starvation, but write locks
527fa9e406ahrens * are extremely rare, and a solution to this problem would be significantly
528fa9e406ahrens * more complex (if even possible).
529fa9e406ahrens *
530fa9e406ahrens * We would like to assert that the namespace lock isn't held, but this is a
531fa9e406ahrens * valid use during create.
532fa9e406ahrens */
533fa9e406ahrensvoid
534ea8dc4beschrockspa_config_enter(spa_t *spa, krw_t rw, void *tag)
535fa9e406ahrens{
536fa9e406ahrens	spa_config_lock_t *scl = &spa->spa_config_lock;
537fa9e406ahrens
538fa9e406ahrens	mutex_enter(&scl->scl_lock);
539fa9e406ahrens
540fa9e406ahrens	if (scl->scl_writer != curthread) {
541fa9e406ahrens		if (rw == RW_READER) {
542fa9e406ahrens			while (scl->scl_writer != NULL)
543fa9e406ahrens				cv_wait(&scl->scl_cv, &scl->scl_lock);
544fa9e406ahrens		} else {
545ea8dc4beschrock			while (scl->scl_writer != NULL ||
546ea8dc4beschrock			    !refcount_is_zero(&scl->scl_count))
547fa9e406ahrens				cv_wait(&scl->scl_cv, &scl->scl_lock);
548fa9e406ahrens			scl->scl_writer = curthread;
549fa9e406ahrens		}
550fa9e406ahrens	}
551fa9e406ahrens
552ea8dc4beschrock	(void) refcount_add(&scl->scl_count, tag);
553fa9e406ahrens
554fa9e406ahrens	mutex_exit(&scl->scl_lock);
555fa9e406ahrens}
556fa9e406ahrens
557fa9e406ahrens/*
558fa9e406ahrens * Release the spa config lock, notifying any waiters in the process.
559fa9e406ahrens */
560fa9e406ahrensvoid
561ea8dc4beschrockspa_config_exit(spa_t *spa, void *tag)
562fa9e406ahrens{
563fa9e406ahrens	spa_config_lock_t *scl = &spa->spa_config_lock;
564fa9e406ahrens
565fa9e406ahrens	mutex_enter(&scl->scl_lock);
566fa9e406ahrens
567ea8dc4beschrock	ASSERT(!refcount_is_zero(&scl->scl_count));
568ea8dc4beschrock	if (refcount_remove(&scl->scl_count, tag) == 0) {
569fa9e406ahrens		cv_broadcast(&scl->scl_cv);
570fa9e406ahrens		scl->scl_writer = NULL;  /* OK in either case */
571fa9e406ahrens	}
572fa9e406ahrens
573fa9e406ahrens	mutex_exit(&scl->scl_lock);
574fa9e406ahrens}
575fa9e406ahrens
576fa9e406ahrens/*
577fa9e406ahrens * Returns true if the config lock is held in the given manner.
578fa9e406ahrens */
579fa9e406ahrensboolean_t
580fa9e406ahrensspa_config_held(spa_t *spa, krw_t rw)
581fa9e406ahrens{
582fa9e406ahrens	spa_config_lock_t *scl = &spa->spa_config_lock;
583fa9e406ahrens	boolean_t held;
584fa9e406ahrens
585fa9e406ahrens	mutex_enter(&scl->scl_lock);
586fa9e406ahrens	if (rw == RW_WRITER)
587fa9e406ahrens		held = (scl->scl_writer == curthread);
588fa9e406ahrens	else
589ea8dc4beschrock		held = !refcount_is_zero(&scl->scl_count);
590fa9e406ahrens	mutex_exit(&scl->scl_lock);
591fa9e406ahrens
592fa9e406ahrens	return (held);
593fa9e406ahrens}
594fa9e406ahrens
595fa9e406ahrens/*
596fa9e406ahrens * ==========================================================================
597fa9e406ahrens * SPA vdev locking
598fa9e406ahrens * ==========================================================================
599fa9e406ahrens */
600fa9e406ahrens
601fa9e406ahrens/*
602ea8dc4beschrock * Lock the given spa_t for the purpose of adding or removing a vdev.
603ea8dc4beschrock * Grabs the global spa_namespace_lock plus the spa config lock for writing.
604fa9e406ahrens * It returns the next transaction group for the spa_t.
605fa9e406ahrens */
606fa9e406ahrensuint64_t
607fa9e406ahrensspa_vdev_enter(spa_t *spa)
608fa9e406ahrens{
6093d7072feschrock	mutex_enter(&spa_namespace_lock);
6103d7072feschrock
611ea8dc4beschrock	/*
6123d7072feschrock	 * Suspend scrub activity while we mess with the config.  We must do
6133d7072feschrock	 * this after acquiring the namespace lock to avoid a 3-way deadlock
6143d7072feschrock	 * with spa_scrub_stop() and the scrub thread.
615ea8dc4beschrock	 */
616ea8dc4beschrock	spa_scrub_suspend(spa);
617fa9e406ahrens
618ea8dc4beschrock	spa_config_enter(spa, RW_WRITER, spa);
619fa9e406ahrens
620fa9e406ahrens	return (spa_last_synced_txg(spa) + 1);
621fa9e406ahrens}
622fa9e406ahrens
623fa9e406ahrens/*
624fa9e406ahrens * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
625fa9e406ahrens * locking of spa_vdev_enter(), we also want make sure the transactions have
626fa9e406ahrens * synced to disk, and then update the global configuration cache with the new
627fa9e406ahrens * information.
628fa9e406ahrens */
629fa9e406ahrensint
630fa9e406ahrensspa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
631fa9e406ahrens{
6320e34b6abonwick	int config_changed = B_FALSE;
633ea8dc4beschrock
6340373e76bonwick	ASSERT(txg > spa_last_synced_txg(spa));
6350e34b6abonwick
6360e34b6abonwick	/*
6370e34b6abonwick	 * Reassess the DTLs.
6380e34b6abonwick	 */
6390373e76bonwick	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
6400e34b6abonwick
6410e34b6abonwick	/*
6420373e76bonwick	 * If the config changed, notify the scrub thread that it must restart.
6430e34b6abonwick	 */
6440e34b6abonwick	if (error == 0 && !list_is_empty(&spa->spa_dirty_list)) {
6450e34b6abonwick		config_changed = B_TRUE;
6460373e76bonwick		spa_scrub_restart(spa, txg);
6470e34b6abonwick	}
648ea8dc4beschrock
649ea8dc4beschrock	spa_config_exit(spa, spa);
650fa9e406ahrens
651ea8dc4beschrock	/*
6525dabedebonwick	 * Allow scrubbing to resume.
653ea8dc4beschrock	 */
654ea8dc4beschrock	spa_scrub_resume(spa);
655fa9e406ahrens
656fa9e406ahrens	/*
657fa9e406ahrens	 * Note: this txg_wait_synced() is important because it ensures
658fa9e406ahrens	 * that there won't be more than one config change per txg.
659fa9e406ahrens	 * This allows us to use the txg as the generation number.
660fa9e406ahrens	 */
661fa9e406ahrens	if (error == 0)
662fa9e406ahrens		txg_wait_synced(spa->spa_dsl_pool, txg);
663fa9e406ahrens
664fa9e406ahrens	if (vd != NULL) {
665fa9e406ahrens		ASSERT(!vd->vdev_detached || vd->vdev_dtl.smo_object == 0);
666fa9e406ahrens		vdev_free(vd);
667fa9e406ahrens	}
668fa9e406ahrens
669fa9e406ahrens	/*
6700e34b6abonwick	 * If the config changed, update the config cache.
671fa9e406ahrens	 */
6720e34b6abonwick	if (config_changed)
673fa9e406ahrens		spa_config_sync();
674ea8dc4beschrock
675ea8dc4beschrock	mutex_exit(&spa_namespace_lock);
676fa9e406ahrens
677fa9e406ahrens	return (error);
678fa9e406ahrens}
679fa9e406ahrens
680fa9e406ahrens/*
681fa9e406ahrens * ==========================================================================
682fa9e406ahrens * Miscellaneous functions
683fa9e406ahrens * ==========================================================================
684fa9e406ahrens */
685fa9e406ahrens
686fa9e406ahrens/*
687fa9e406ahrens * Rename a spa_t.
688fa9e406ahrens */
689fa9e406ahrensint
690fa9e406ahrensspa_rename(const char *name, const char *newname)
691fa9e406ahrens{
692fa9e406ahrens	spa_t *spa;
693fa9e406ahrens	int err;
694fa9e406ahrens
695fa9e406ahrens	/*
696fa9e406ahrens	 * Lookup the spa_t and grab the config lock for writing.  We need to
697fa9e406ahrens	 * actually open the pool so that we can sync out the necessary labels.
698fa9e406ahrens	 * It's OK to call spa_open() with the namespace lock held because we
699ea8dc4beschrock	 * allow recursive calls for other reasons.
700fa9e406ahrens	 */
701fa9e406ahrens	mutex_enter(&spa_namespace_lock);
702fa9e406ahrens	if ((err = spa_open(name, &spa, FTAG)) != 0) {
703fa9e406ahrens		mutex_exit(&spa_namespace_lock);
704fa9e406ahrens		return (err);
705fa9e406ahrens	}
706fa9e406ahrens
707ea8dc4beschrock	spa_config_enter(spa, RW_WRITER, FTAG);
708fa9e406ahrens
709fa9e406ahrens	avl_remove(&spa_namespace_avl, spa);
710fa9e406ahrens	spa_strfree(spa->spa_name);
711fa9e406ahrens	spa->spa_name = spa_strdup(newname);
712fa9e406ahrens	avl_add(&spa_namespace_avl, spa);
713fa9e406ahrens
714fa9e406ahrens	/*
715fa9e406ahrens	 * Sync all labels to disk with the new names by marking the root vdev
716fa9e406ahrens	 * dirty and waiting for it to sync.  It will pick up the new pool name
717fa9e406ahrens	 * during the sync.
718fa9e406ahrens	 */
719fa9e406ahrens	vdev_config_dirty(spa->spa_root_vdev);
720fa9e406ahrens
721ea8dc4beschrock	spa_config_exit(spa, FTAG);
722fa9e406ahrens
7230373e76bonwick	txg_wait_synced(spa->spa_dsl_pool, 0);
724fa9e406ahrens
725fa9e406ahrens	/*
726fa9e406ahrens	 * Sync the updated config cache.
727fa9e406ahrens	 */
728fa9e406ahrens	spa_config_sync();
729fa9e406ahrens
730fa9e406ahrens	spa_close(spa, FTAG);
731fa9e406ahrens
732fa9e406ahrens	mutex_exit(&spa_namespace_lock);
733fa9e406ahrens
734fa9e406ahrens	return (0);
735fa9e406ahrens}
736fa9e406ahrens
737fa9e406ahrens
738fa9e406ahrens/*
739fa9e406ahrens * Determine whether a pool with given pool_guid exists.  If device_guid is
740fa9e406ahrens * non-zero, determine whether the pool exists *and* contains a device with the
741fa9e406ahrens * specified device_guid.
742fa9e406ahrens */
743fa9e406ahrensboolean_t
744fa9e406ahrensspa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
745fa9e406ahrens{
746fa9e406ahrens	spa_t *spa;
747fa9e406ahrens	avl_tree_t *t = &spa_namespace_avl;
748fa9e406ahrens
749ea8dc4beschrock	ASSERT(MUTEX_HELD(&spa_namespace_lock));
750fa9e406ahrens
751fa9e406ahrens	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
752fa9e406ahrens		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
753fa9e406ahrens			continue;
754fa9e406ahrens		if (spa->spa_root_vdev == NULL)
755fa9e406ahrens			continue;
75639c2341eschrock		if (spa_guid(spa) == pool_guid) {
75739c2341eschrock			if (device_guid == 0)
75839c2341eschrock				break;
759