spa_misc.c revision 99653d4ee642c6528e88224f12409a5f23060994
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/*
22ea8dc4beschrock * Copyright 2006 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;
17644cd46cbillmstatic int spa_max_replication_override = SPA_DVAS_PER_BP;
177fa9e406ahrens
17899653d4eschrockstatic avl_tree_t spa_spare_avl;
17999653d4eschrockstatic kmutex_t spa_spare_lock;
18099653d4eschrock
181fa9e406ahrenskmem_cache_t *spa_buffer_pool;
182fa9e406ahrensint spa_mode;
183fa9e406ahrens
184fa9e406ahrens#ifdef ZFS_DEBUG
185fa9e406ahrensint zfs_flags = ~0;
186fa9e406ahrens#else
187fa9e406ahrensint zfs_flags = 0;
188fa9e406ahrens#endif
189fa9e406ahrens
190fa9e406ahrens#define	SPA_MINREF	5	/* spa_refcnt for an open-but-idle pool */
191fa9e406ahrens
192fa9e406ahrens/*
193fa9e406ahrens * ==========================================================================
194fa9e406ahrens * SPA namespace functions
195fa9e406ahrens * ==========================================================================
196fa9e406ahrens */
197fa9e406ahrens
198fa9e406ahrens/*
199fa9e406ahrens * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
200fa9e406ahrens * Returns NULL if no matching spa_t is found.
201fa9e406ahrens */
202fa9e406ahrensspa_t *
203fa9e406ahrensspa_lookup(const char *name)
204fa9e406ahrens{
205fa9e406ahrens	spa_t search, *spa;
206fa9e406ahrens	avl_index_t where;
207fa9e406ahrens
208fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
209fa9e406ahrens
210fa9e406ahrens	search.spa_name = (char *)name;
211fa9e406ahrens	spa = avl_find(&spa_namespace_avl, &search, &where);
212fa9e406ahrens
213fa9e406ahrens	return (spa);
214fa9e406ahrens}
215fa9e406ahrens
216fa9e406ahrens/*
217fa9e406ahrens * Create an uninitialized spa_t with the given name.  Requires
218fa9e406ahrens * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
219fa9e406ahrens * exist by calling spa_lookup() first.
220fa9e406ahrens */
221fa9e406ahrensspa_t *
2220373e76bonwickspa_add(const char *name, const char *altroot)
223fa9e406ahrens{
224fa9e406ahrens	spa_t *spa;
225fa9e406ahrens
226fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
227fa9e406ahrens
228fa9e406ahrens	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
229fa9e406ahrens
230fa9e406ahrens	spa->spa_name = spa_strdup(name);
231fa9e406ahrens	spa->spa_state = POOL_STATE_UNINITIALIZED;
232fa9e406ahrens	spa->spa_freeze_txg = UINT64_MAX;
2330373e76bonwick	spa->spa_final_txg = UINT64_MAX;
234fa9e406ahrens
235fa9e406ahrens	refcount_create(&spa->spa_refcount);
236ea8dc4beschrock	refcount_create(&spa->spa_config_lock.scl_count);
237fa9e406ahrens
238fa9e406ahrens	avl_add(&spa_namespace_avl, spa);
239fa9e406ahrens
2400373e76bonwick	/*
2410373e76bonwick	 * Set the alternate root, if there is one.
2420373e76bonwick	 */
2430373e76bonwick	if (altroot) {
2440373e76bonwick		spa->spa_root = spa_strdup(altroot);
2450373e76bonwick		spa_active_count++;
2460373e76bonwick	}
2470373e76bonwick
248fa9e406ahrens	return (spa);
249fa9e406ahrens}
250fa9e406ahrens
251fa9e406ahrens/*
252fa9e406ahrens * Removes a spa_t from the namespace, freeing up any memory used.  Requires
253fa9e406ahrens * spa_namespace_lock.  This is called only after the spa_t has been closed and
254fa9e406ahrens * deactivated.
255fa9e406ahrens */
256fa9e406ahrensvoid
257fa9e406ahrensspa_remove(spa_t *spa)
258fa9e406ahrens{
259fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
260fa9e406ahrens	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
261fa9e406ahrens	ASSERT(spa->spa_scrub_thread == NULL);
262fa9e406ahrens
263fa9e406ahrens	avl_remove(&spa_namespace_avl, spa);
264fa9e406ahrens	cv_broadcast(&spa_namespace_cv);
265fa9e406ahrens
2660373e76bonwick	if (spa->spa_root) {
267fa9e406ahrens		spa_strfree(spa->spa_root);
2680373e76bonwick		spa_active_count--;
2690373e76bonwick	}
270fa9e406ahrens
271fa9e406ahrens	if (spa->spa_name)
272fa9e406ahrens		spa_strfree(spa->spa_name);
273fa9e406ahrens
274fa9e406ahrens	spa_config_set(spa, NULL);
275fa9e406ahrens
276fa9e406ahrens	refcount_destroy(&spa->spa_refcount);
277ea8dc4beschrock	refcount_destroy(&spa->spa_config_lock.scl_count);
278fa9e406ahrens
279fa9e406ahrens	kmem_free(spa, sizeof (spa_t));
280fa9e406ahrens}
281fa9e406ahrens
282fa9e406ahrens/*
283fa9e406ahrens * Given a pool, return the next pool in the namespace, or NULL if there is
284fa9e406ahrens * none.  If 'prev' is NULL, return the first pool.
285fa9e406ahrens */
286fa9e406ahrensspa_t *
287fa9e406ahrensspa_next(spa_t *prev)
288fa9e406ahrens{
289fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
290fa9e406ahrens
291fa9e406ahrens	if (prev)
292fa9e406ahrens		return (AVL_NEXT(&spa_namespace_avl, prev));
293fa9e406ahrens	else
294fa9e406ahrens		return (avl_first(&spa_namespace_avl));
295fa9e406ahrens}
296fa9e406ahrens
297fa9e406ahrens/*
298fa9e406ahrens * ==========================================================================
299fa9e406ahrens * SPA refcount functions
300fa9e406ahrens * ==========================================================================
301fa9e406ahrens */
302fa9e406ahrens
303fa9e406ahrens/*
304fa9e406ahrens * Add a reference to the given spa_t.  Must have at least one reference, or
305fa9e406ahrens * have the namespace lock held.
306fa9e406ahrens */
307fa9e406ahrensvoid
308fa9e406ahrensspa_open_ref(spa_t *spa, void *tag)
309fa9e406ahrens{
310fa9e406ahrens	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
311fa9e406ahrens	    MUTEX_HELD(&spa_namespace_lock));
312fa9e406ahrens
313fa9e406ahrens	(void) refcount_add(&spa->spa_refcount, tag);
314fa9e406ahrens}
315fa9e406ahrens
316fa9e406ahrens/*
317fa9e406ahrens * Remove a reference to the given spa_t.  Must have at least one reference, or
318fa9e406ahrens * have the namespace lock held.
319fa9e406ahrens */
320fa9e406ahrensvoid
321fa9e406ahrensspa_close(spa_t *spa, void *tag)
322fa9e406ahrens{
323fa9e406ahrens	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
324fa9e406ahrens	    MUTEX_HELD(&spa_namespace_lock));
325fa9e406ahrens
326fa9e406ahrens	(void) refcount_remove(&spa->spa_refcount, tag);
327fa9e406ahrens}
328fa9e406ahrens
329fa9e406ahrens/*
330fa9e406ahrens * Check to see if the spa refcount is zero.  Must be called with
331fa9e406ahrens * spa_namespace_lock held.  We really compare against SPA_MINREF, which is the
332fa9e406ahrens * number of references acquired when opening a pool
333fa9e406ahrens */
334fa9e406ahrensboolean_t
335fa9e406ahrensspa_refcount_zero(spa_t *spa)
336fa9e406ahrens{
337fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
338fa9e406ahrens
339fa9e406ahrens	return (refcount_count(&spa->spa_refcount) == SPA_MINREF);
340fa9e406ahrens}
341fa9e406ahrens
342fa9e406ahrens/*
343fa9e406ahrens * ==========================================================================
34499653d4eschrock * SPA spare tracking
34599653d4eschrock * ==========================================================================
34699653d4eschrock */
34799653d4eschrock
34899653d4eschrock/*
34999653d4eschrock * We track spare information on a global basis.  This allows us to do two
35099653d4eschrock * things: determine when a spare is no longer referenced by any active pool,
35199653d4eschrock * and (quickly) determine if a spare is currently in use in another pool on the
35299653d4eschrock * system.
35399653d4eschrock */
35499653d4eschrocktypedef struct spa_spare {
35599653d4eschrock	uint64_t	spare_guid;
35699653d4eschrock	avl_node_t	spare_avl;
35799653d4eschrock	int		spare_count;
35899653d4eschrock} spa_spare_t;
35999653d4eschrock
36099653d4eschrockstatic int
36199653d4eschrockspa_spare_compare(const void *a, const void *b)
36299653d4eschrock{
36399653d4eschrock	const spa_spare_t *sa = a;
36499653d4eschrock	const spa_spare_t *sb = b;
36599653d4eschrock
36699653d4eschrock	if (sa->spare_guid < sb->spare_guid)
36799653d4eschrock		return (-1);
36899653d4eschrock	else if (sa->spare_guid > sb->spare_guid)
36999653d4eschrock		return (1);
37099653d4eschrock	else
37199653d4eschrock		return (0);
37299653d4eschrock}
37399653d4eschrock
37499653d4eschrockvoid
37599653d4eschrockspa_spare_add(uint64_t guid)
37699653d4eschrock{
37799653d4eschrock	avl_index_t where;
37899653d4eschrock	spa_spare_t search;
37999653d4eschrock	spa_spare_t *spare;
38099653d4eschrock
38199653d4eschrock	mutex_enter(&spa_spare_lock);
38299653d4eschrock
38399653d4eschrock	search.spare_guid = guid;
38499653d4eschrock	if ((spare = avl_find(&spa_spare_avl, &search, &where)) != NULL) {
38599653d4eschrock		spare->spare_count++;
38699653d4eschrock	} else {
38799653d4eschrock		spare = kmem_alloc(sizeof (spa_spare_t), KM_SLEEP);
38899653d4eschrock		spare->spare_guid = guid;
38999653d4eschrock		spare->spare_count = 1;
39099653d4eschrock		avl_insert(&spa_spare_avl, spare, where);
39199653d4eschrock	}
39299653d4eschrock
39399653d4eschrock	mutex_exit(&spa_spare_lock);
39499653d4eschrock}
39599653d4eschrock
39699653d4eschrockvoid
39799653d4eschrockspa_spare_remove(uint64_t guid)
39899653d4eschrock{
39999653d4eschrock	spa_spare_t search;
40099653d4eschrock	spa_spare_t *spare;
40199653d4eschrock	avl_index_t where;
40299653d4eschrock
40399653d4eschrock	mutex_enter(&spa_spare_lock);
40499653d4eschrock
40599653d4eschrock	search.spare_guid = guid;
40699653d4eschrock	spare = avl_find(&spa_spare_avl, &search, &where);
40799653d4eschrock
40899653d4eschrock	ASSERT(spare != NULL);
40999653d4eschrock
41099653d4eschrock	if (--spare->spare_count == 0) {
41199653d4eschrock		avl_remove(&spa_spare_avl, spare);
41299653d4eschrock		kmem_free(spare, sizeof (spa_spare_t));
41399653d4eschrock	}
41499653d4eschrock
41599653d4eschrock	mutex_exit(&spa_spare_lock);
41699653d4eschrock}
41799653d4eschrock
41899653d4eschrockboolean_t
41999653d4eschrockspa_spare_inuse(uint64_t guid)
42099653d4eschrock{
42199653d4eschrock	spa_spare_t search;
42299653d4eschrock	avl_index_t where;
42399653d4eschrock	boolean_t ret;
42499653d4eschrock
42599653d4eschrock	mutex_enter(&spa_spare_lock);
42699653d4eschrock
42799653d4eschrock	search.spare_guid = guid;
42899653d4eschrock	ret = (avl_find(&spa_spare_avl, &search, &where) != NULL);
42999653d4eschrock
43099653d4eschrock	mutex_exit(&spa_spare_lock);
43199653d4eschrock
43299653d4eschrock	return (ret);
43399653d4eschrock}
43499653d4eschrock
43599653d4eschrock/*
43699653d4eschrock * ==========================================================================
437fa9e406ahrens * SPA config locking
438fa9e406ahrens * ==========================================================================
439fa9e406ahrens */
440fa9e406ahrens
441fa9e406ahrens/*
442fa9e406ahrens * Acquire the config lock.  The config lock is a special rwlock that allows for
443fa9e406ahrens * recursive enters.  Because these enters come from the same thread as well as
444fa9e406ahrens * asynchronous threads working on behalf of the owner, we must unilaterally
445fa9e406ahrens * allow all reads access as long at least one reader is held (even if a write
446fa9e406ahrens * is requested).  This has the side effect of write starvation, but write locks
447fa9e406ahrens * are extremely rare, and a solution to this problem would be significantly
448fa9e406ahrens * more complex (if even possible).
449fa9e406ahrens *
450fa9e406ahrens * We would like to assert that the namespace lock isn't held, but this is a
451fa9e406ahrens * valid use during create.
452fa9e406ahrens */
453fa9e406ahrensvoid
454ea8dc4beschrockspa_config_enter(spa_t *spa, krw_t rw, void *tag)
455fa9e406ahrens{
456fa9e406ahrens	spa_config_lock_t *scl = &spa->spa_config_lock;
457fa9e406ahrens
458fa9e406ahrens	mutex_enter(&scl->scl_lock);
459fa9e406ahrens
460fa9e406ahrens	if (scl->scl_writer != curthread) {
461fa9e406ahrens		if (rw == RW_READER) {
462fa9e406ahrens			while (scl->scl_writer != NULL)
463fa9e406ahrens				cv_wait(&scl->scl_cv, &scl->scl_lock);
464fa9e406ahrens		} else {
465ea8dc4beschrock			while (scl->scl_writer != NULL ||
466ea8dc4beschrock			    !refcount_is_zero(&scl->scl_count))
467fa9e406ahrens				cv_wait(&scl->scl_cv, &scl->scl_lock);
468fa9e406ahrens			scl->scl_writer = curthread;
469fa9e406ahrens		}
470fa9e406ahrens	}
471fa9e406ahrens
472ea8dc4beschrock	(void) refcount_add(&scl->scl_count, tag);
473fa9e406ahrens
474fa9e406ahrens	mutex_exit(&scl->scl_lock);
475fa9e406ahrens}
476fa9e406ahrens
477fa9e406ahrens/*
478fa9e406ahrens * Release the spa config lock, notifying any waiters in the process.
479fa9e406ahrens */
480fa9e406ahrensvoid
481ea8dc4beschrockspa_config_exit(spa_t *spa, void *tag)
482fa9e406ahrens{
483fa9e406ahrens	spa_config_lock_t *scl = &spa->spa_config_lock;
484fa9e406ahrens
485fa9e406ahrens	mutex_enter(&scl->scl_lock);
486fa9e406ahrens
487ea8dc4beschrock	ASSERT(!refcount_is_zero(&scl->scl_count));
488ea8dc4beschrock	if (refcount_remove(&scl->scl_count, tag) == 0) {
489fa9e406ahrens		cv_broadcast(&scl->scl_cv);
490fa9e406ahrens		scl->scl_writer = NULL;  /* OK in either case */
491fa9e406ahrens	}
492fa9e406ahrens
493fa9e406ahrens	mutex_exit(&scl->scl_lock);
494fa9e406ahrens}
495fa9e406ahrens
496fa9e406ahrens/*
497fa9e406ahrens * Returns true if the config lock is held in the given manner.
498fa9e406ahrens */
499fa9e406ahrensboolean_t
500fa9e406ahrensspa_config_held(spa_t *spa, krw_t rw)
501fa9e406ahrens{
502fa9e406ahrens	spa_config_lock_t *scl = &spa->spa_config_lock;
503fa9e406ahrens	boolean_t held;
504fa9e406ahrens
505fa9e406ahrens	mutex_enter(&scl->scl_lock);
506fa9e406ahrens	if (rw == RW_WRITER)
507fa9e406ahrens		held = (scl->scl_writer == curthread);
508fa9e406ahrens	else
509ea8dc4beschrock		held = !refcount_is_zero(&scl->scl_count);
510fa9e406ahrens	mutex_exit(&scl->scl_lock);
511fa9e406ahrens
512fa9e406ahrens	return (held);
513fa9e406ahrens}
514fa9e406ahrens
515fa9e406ahrens/*
516fa9e406ahrens * ==========================================================================
517fa9e406ahrens * SPA vdev locking
518fa9e406ahrens * ==========================================================================
519fa9e406ahrens */
520fa9e406ahrens
521fa9e406ahrens/*
522ea8dc4beschrock * Lock the given spa_t for the purpose of adding or removing a vdev.
523ea8dc4beschrock * Grabs the global spa_namespace_lock plus the spa config lock for writing.
524fa9e406ahrens * It returns the next transaction group for the spa_t.
525fa9e406ahrens */
526fa9e406ahrensuint64_t
527fa9e406ahrensspa_vdev_enter(spa_t *spa)
528fa9e406ahrens{
529ea8dc4beschrock	/*
530ea8dc4beschrock	 * Suspend scrub activity while we mess with the config.
531ea8dc4beschrock	 */
532ea8dc4beschrock	spa_scrub_suspend(spa);
533fa9e406ahrens
5340373e76bonwick	mutex_enter(&spa_namespace_lock);
535ea8dc4beschrock
536ea8dc4beschrock	spa_config_enter(spa, RW_WRITER, spa);
537fa9e406ahrens
538fa9e406ahrens	return (spa_last_synced_txg(spa) + 1);
539fa9e406ahrens}
540fa9e406ahrens
541fa9e406ahrens/*
542fa9e406ahrens * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
543fa9e406ahrens * locking of spa_vdev_enter(), we also want make sure the transactions have
544fa9e406ahrens * synced to disk, and then update the global configuration cache with the new
545fa9e406ahrens * information.
546fa9e406ahrens */
547fa9e406ahrensint
548fa9e406ahrensspa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
549fa9e406ahrens{
5500e34b6abonwick	int config_changed = B_FALSE;
551ea8dc4beschrock
5520373e76bonwick	ASSERT(txg > spa_last_synced_txg(spa));
5530e34b6abonwick
5540e34b6abonwick	/*
5550e34b6abonwick	 * Reassess the DTLs.
5560e34b6abonwick	 */
5570373e76bonwick	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
5580e34b6abonwick
5590e34b6abonwick	/*
5600373e76bonwick	 * If the config changed, notify the scrub thread that it must restart.
5610e34b6abonwick	 */
5620e34b6abonwick	if (error == 0 && !list_is_empty(&spa->spa_dirty_list)) {
5630e34b6abonwick		config_changed = B_TRUE;
5640373e76bonwick		spa_scrub_restart(spa, txg);
5650e34b6abonwick	}
566ea8dc4beschrock
567ea8dc4beschrock	spa_config_exit(spa, spa);
568fa9e406ahrens
569ea8dc4beschrock	/*
5705dabedebonwick	 * Allow scrubbing to resume.
571ea8dc4beschrock	 */
572ea8dc4beschrock	spa_scrub_resume(spa);
573fa9e406ahrens
574fa9e406ahrens	/*
575fa9e406ahrens	 * Note: this txg_wait_synced() is important because it ensures
576fa9e406ahrens	 * that there won't be more than one config change per txg.
577fa9e406ahrens	 * This allows us to use the txg as the generation number.
578fa9e406ahrens	 */
579fa9e406ahrens	if (error == 0)
580fa9e406ahrens		txg_wait_synced(spa->spa_dsl_pool, txg);
581fa9e406ahrens
582fa9e406ahrens	if (vd != NULL) {
583fa9e406ahrens		ASSERT(!vd->vdev_detached || vd->vdev_dtl.smo_object == 0);
584fa9e406ahrens		vdev_free(vd);
585fa9e406ahrens	}
586fa9e406ahrens
587fa9e406ahrens	/*
5880e34b6abonwick	 * If the config changed, update the config cache.
589fa9e406ahrens	 */
5900e34b6abonwick	if (config_changed)
591fa9e406ahrens		spa_config_sync();
592ea8dc4beschrock
593ea8dc4beschrock	mutex_exit(&spa_namespace_lock);
594fa9e406ahrens
595fa9e406ahrens	return (error);
596fa9e406ahrens}
597fa9e406ahrens
598fa9e406ahrens/*
599fa9e406ahrens * ==========================================================================
600fa9e406ahrens * Miscellaneous functions
601fa9e406ahrens * ==========================================================================
602fa9e406ahrens */
603fa9e406ahrens
604fa9e406ahrens/*
605fa9e406ahrens * Rename a spa_t.
606fa9e406ahrens */
607fa9e406ahrensint
608fa9e406ahrensspa_rename(const char *name, const char *newname)
609fa9e406ahrens{
610fa9e406ahrens	spa_t *spa;
611fa9e406ahrens	int err;
612fa9e406ahrens
613fa9e406ahrens	/*
614fa9e406ahrens	 * Lookup the spa_t and grab the config lock for writing.  We need to
615fa9e406ahrens	 * actually open the pool so that we can sync out the necessary labels.
616fa9e406ahrens	 * It's OK to call spa_open() with the namespace lock held because we
617ea8dc4beschrock	 * allow recursive calls for other reasons.
618fa9e406ahrens	 */
619fa9e406ahrens	mutex_enter(&spa_namespace_lock);
620fa9e406ahrens	if ((err = spa_open(name, &spa, FTAG)) != 0) {
621fa9e406ahrens		mutex_exit(&spa_namespace_lock);
622fa9e406ahrens		return (err);
623fa9e406ahrens	}
624fa9e406ahrens
625ea8dc4beschrock	spa_config_enter(spa, RW_WRITER, FTAG);
626fa9e406ahrens
627fa9e406ahrens	avl_remove(&spa_namespace_avl, spa);
628fa9e406ahrens	spa_strfree(spa->spa_name);
629fa9e406ahrens	spa->spa_name = spa_strdup(newname);
630fa9e406ahrens	avl_add(&spa_namespace_avl, spa);
631fa9e406ahrens
632fa9e406ahrens	/*
633fa9e406ahrens	 * Sync all labels to disk with the new names by marking the root vdev
634fa9e406ahrens	 * dirty and waiting for it to sync.  It will pick up the new pool name
635fa9e406ahrens	 * during the sync.
636fa9e406ahrens	 */
637fa9e406ahrens	vdev_config_dirty(spa->spa_root_vdev);
638fa9e406ahrens
639ea8dc4beschrock	spa_config_exit(spa, FTAG);
640fa9e406ahrens
6410373e76bonwick	txg_wait_synced(spa->spa_dsl_pool, 0);
642fa9e406ahrens
643fa9e406ahrens	/*
644fa9e406ahrens	 * Sync the updated config cache.
645fa9e406ahrens	 */
646fa9e406ahrens	spa_config_sync();
647fa9e406ahrens
648fa9e406ahrens	spa_close(spa, FTAG);
649fa9e406ahrens
650fa9e406ahrens	mutex_exit(&spa_namespace_lock);
651fa9e406ahrens
652fa9e406ahrens	return (0);
653fa9e406ahrens}
654fa9e406ahrens
655fa9e406ahrens
656fa9e406ahrens/*
657fa9e406ahrens * Determine whether a pool with given pool_guid exists.  If device_guid is
658fa9e406ahrens * non-zero, determine whether the pool exists *and* contains a device with the
659fa9e406ahrens * specified device_guid.
660fa9e406ahrens */
661fa9e406ahrensboolean_t
662fa9e406ahrensspa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
663fa9e406ahrens{
664fa9e406ahrens	spa_t *spa;
665fa9e406ahrens	avl_tree_t *t = &spa_namespace_avl;
666fa9e406ahrens
667ea8dc4beschrock	ASSERT(MUTEX_HELD(&spa_namespace_lock));
668fa9e406ahrens
669fa9e406ahrens	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
670fa9e406ahrens		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
671fa9e406ahrens			continue;
672fa9e406ahrens		if (spa->spa_root_vdev == NULL)
673fa9e406ahrens			continue;
674fa9e406ahrens		if (spa_guid(spa) == pool_guid && (device_guid == 0 ||
675fa9e406ahrens		    vdev_lookup_by_guid(spa->spa_root_vdev, device_guid)))
676fa9e406ahrens			break;
677fa9e406ahrens	}
678fa9e406ahrens
679fa9e406ahrens	return (spa != NULL);
680fa9e406ahrens}
681fa9e406ahrens
682fa9e406ahrenschar *
683fa9e406ahrensspa_strdup(const char *s)
684fa9e406ahrens{
685fa9e406ahrens	size_t len;
686fa9e406ahrens	char *new;
687fa9e406ahrens
688fa9e406ahrens	len = strlen(s);
689fa9e406ahrens	new = kmem_alloc(len + 1, KM_SLEEP);
690fa9e406ahrens	bcopy(s, new, len);
691fa9e406ahrens	new[len] = '\0';
692fa9e406ahrens
693fa9e406ahrens	return (new);
694fa9e406ahrens}
695fa9e406ahrens
696fa9e406ahrensvoid
697fa9e406ahrensspa_strfree(char *s)
698fa9e406ahrens{
699fa9e406ahrens	kmem_free(s, strlen(s) + 1);
700fa9e406ahrens}
701fa9e406ahrens
702fa9e406ahrensuint64_t
703fa9e406ahrensspa_get_random(uint64_t range)
704fa9e406ahrens{
705fa9e406ahrens	uint64_t r;
706fa9e406ahrens
707fa9e406ahrens	ASSERT(range != 0);
708fa9e406ahrens
709fa9e406ahrens	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
710fa9e406ahrens
711fa9e406ahrens	return (r % range);
712fa9e406ahrens}
713fa9e406ahrens
714fa9e406ahrensvoid
715d80c45ebonwicksprintf_blkptr(char *buf, int len, const blkptr_t *bp)
716fa9e406ahrens{
71744cd46cbillm	int d;
718fa9e406ahrens
719fa9e406ahrens	if (bp == NULL) {
720fbabab8maybee		(void) snprintf(buf, len, "<NULL>");
721fa9e406ahrens		return;
722fa9e406ahrens	}
723fa9e406ahrens
724fa9e406ahrens	if (BP_IS_HOLE(bp)) {
725fbabab8maybee		(void) snprintf(buf, len, "<hole>");
726fa9e406ahrens		return;
727fa9e406ahrens	}
728fa9e406ahrens
72944cd46cbillm	(void) snprintf(buf, len, "[L%llu %s] %llxL/%llxP ",
730fa9e406ahrens	    (u_longlong_t)BP_GET_LEVEL(bp),
731fa9e406ahrens	    dmu_ot[BP_GET_TYPE(bp)].ot_name,
732fa9e406ahrens	    (u_longlong_t)BP_GET_LSIZE(bp),
73344cd46cbillm	    (u_longlong_t)BP_GET_PSIZE(bp));
73444cd46cbillm
73544cd46cbillm	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
736d80c45ebonwick		const dva_t *dva = &bp->blk_dva[d];
73744cd46cbillm		(void) snprintf(buf + strlen(buf), len - strlen(buf),
73844cd46cbillm		    "DVA[%d]=<%llu:%llx:%llx> ", d,
73944cd46cbillm		    (u_longlong_t)DVA_GET_VDEV(dva),
74044cd46cbillm		    (u_longlong_t)DVA_GET_OFFSET(dva),
74144cd46cbillm		    (u_longlong_t)DVA_GET_ASIZE(dva));
74244cd46cbillm	}
74344cd46cbillm
74444cd46cbillm	(void) snprintf(buf + strlen(buf), len - strlen(buf),
74544cd46cbillm	    "%s %s %s %s birth=%llu fill=%llu cksum=%llx:%llx:%llx:%llx",
746fa9e406ahrens	    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name,
747fa9e406ahrens	    zio_compress_table[BP_GET_COMPRESS(bp)].ci_name,
748fa9e406ahrens	    BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE",
74944cd46cbillm	    BP_IS_GANG(bp) ? "gang" : "contiguous",
750fa9e406ahrens	    (u_longlong_t)bp->blk_birth,
751fa9e406ahrens	    (u_longlong_t)bp->blk_fill,
752fa9e406ahrens	    (u_longlong_t)bp->blk_cksum.zc_word[0],
753fa9e406ahrens	    (u_longlong_t)bp->blk_cksum.zc_word[1],
754fa9e406ahrens	    (u_longlong_t)bp->blk_cksum.zc_word[2],
755fa9e406ahrens	    (u_longlong_t)bp->blk_cksum.zc_word[3]);
756fa9e406ahrens}
757fa9e406ahrens
758fa9e406ahrensvoid
759fa9e406ahrensspa_freeze(spa_t *spa)
760fa9e406ahrens{
761fa9e406ahrens	uint64_t freeze_txg = 0;
762fa9e406ahrens
763ea8dc4beschrock	spa_config_enter(spa, RW_WRITER, FTAG);
764fa9e406ahrens	if (spa->spa_freeze_txg == UINT64_MAX) {
765fa9e406ahrens		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
766fa9e406ahrens		spa->spa_freeze_txg = freeze_txg;
767fa9e406ahrens	}
768ea8dc4beschrock	spa_config_exit(spa, FTAG);
769fa9e406ahrens	if (freeze_txg != 0)
770fa9e406ahrens		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
771fa9e406ahrens}
772fa9e406ahrens
773fa9e406ahrens/*
774fa9e406ahrens * ==========================================================================
775fa9e406ahrens * Accessor functions
776fa9e406ahrens * ==========================================================================
777fa9e406ahrens */
778fa9e406ahrens
779fa9e406ahrenskrwlock_t *
780fa9e406ahrensspa_traverse_rwlock(spa_t *spa)
781fa9e406ahrens{
782fa9e406ahrens	return (&spa->spa_traverse_lock);
783fa9e406ahrens}
784fa9e406ahrens
785fa9e406ahrensint
786fa9e406ahrensspa_traverse_wanted(spa_t *spa)
787fa9e406ahrens{
788fa9e406ahrens	return (spa->spa_traverse_wanted);
789fa9e406ahrens}
790fa9e406ahrens
791fa9e406ahrensdsl_pool_t *
792fa9e406ahrensspa_get_dsl(spa_t *spa)
793fa9e406ahrens{
794fa9e406ahrens	return (spa->spa_dsl_pool);
795fa9e406ahrens}
796fa9e406ahrens
797fa9e406ahrensblkptr_t *
798fa9e406ahrensspa_get_rootblkptr(spa_t *spa)
799fa9e406ahrens{
800fa9e406ahrens	return (&spa->spa_ubsync.ub_rootbp);
801fa9e406ahrens}
802fa9e406ahrens
803fa9e406ahrensvoid
804fa9e406ahrensspa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
805fa9e406ahrens{
806fa9e406ahrens	spa->spa_uberblock.ub_rootbp = *bp;
807fa9e406ahrens}
808fa9e406ahrens
809fa9e406ahrensvoid
810fa9e406ahrensspa_altroot(spa_t *spa, char *buf, size_t buflen)
811fa9e406ahrens{
812fa9e406ahrens	if (spa->spa_root == NULL)
813fa9e406ahrens		buf[0] = '\0';
814fa9e406ahrens	else
815fa9e406ahrens		(void) strncpy(buf, spa->spa_root, buflen);
816fa9e406ahrens}
817