spa_misc.c revision 9842588ba779c98194a9c238d52784a2e45a850e
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/*
229842588George Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23fa9e406ahrens */
24fa9e406ahrens
25fa9e406ahrens#include <sys/zfs_context.h>
26fa9e406ahrens#include <sys/spa_impl.h>
27fa9e406ahrens#include <sys/zio.h>
28fa9e406ahrens#include <sys/zio_checksum.h>
29fa9e406ahrens#include <sys/zio_compress.h>
30fa9e406ahrens#include <sys/dmu.h>
31fa9e406ahrens#include <sys/dmu_tx.h>
32fa9e406ahrens#include <sys/zap.h>
33fa9e406ahrens#include <sys/zil.h>
34fa9e406ahrens#include <sys/vdev_impl.h>
35fa9e406ahrens#include <sys/metaslab.h>
36fa9e406ahrens#include <sys/uberblock_impl.h>
37fa9e406ahrens#include <sys/txg.h>
38fa9e406ahrens#include <sys/avl.h>
39fa9e406ahrens#include <sys/unique.h>
40fa9e406ahrens#include <sys/dsl_pool.h>
41fa9e406ahrens#include <sys/dsl_dir.h>
42fa9e406ahrens#include <sys/dsl_prop.h>
43fa9e406ahrens#include <sys/fs/zfs.h>
446ce0521perrin#include <sys/metaslab_impl.h>
45e14bb32Jeff Bonwick#include <sys/arc.h>
46485bbbfGeorge Wilson#include <sys/ddt.h>
4791ebeefahrens#include "zfs_prop.h"
48fa9e406ahrens
49fa9e406ahrens/*
50fa9e406ahrens * SPA locking
51fa9e406ahrens *
52fa9e406ahrens * There are four basic locks for managing spa_t structures:
53fa9e406ahrens *
54fa9e406ahrens * spa_namespace_lock (global mutex)
55fa9e406ahrens *
5644cd46cbillm *	This lock must be acquired to do any of the following:
57fa9e406ahrens *
5844cd46cbillm *		- Lookup a spa_t by name
5944cd46cbillm *		- Add or remove a spa_t from the namespace
6044cd46cbillm *		- Increase spa_refcount from non-zero
6144cd46cbillm *		- Check if spa_refcount is zero
6244cd46cbillm *		- Rename a spa_t
63ea8dc4beschrock *		- add/remove/attach/detach devices
6444cd46cbillm *		- Held for the duration of create/destroy/import/export
65fa9e406ahrens *
6644cd46cbillm *	It does not need to handle recursion.  A create or destroy may
6744cd46cbillm *	reference objects (files or zvols) in other pools, but by
6844cd46cbillm *	definition they must have an existing reference, and will never need
6944cd46cbillm *	to lookup a spa_t by name.
70fa9e406ahrens *
71fa9e406ahrens * spa_refcount (per-spa refcount_t protected by mutex)
72fa9e406ahrens *
7344cd46cbillm *	This reference count keep track of any active users of the spa_t.  The
7444cd46cbillm *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
7544cd46cbillm *	the refcount is never really 'zero' - opening a pool implicitly keeps
76088f389ahrens *	some references in the DMU.  Internally we check against spa_minref, but
7744cd46cbillm *	present the image of a zero/non-zero value to consumers.
78fa9e406ahrens *
79e14bb32Jeff Bonwick * spa_config_lock[] (per-spa array of rwlocks)
80fa9e406ahrens *
8191ebeefahrens *	This protects the spa_t from config changes, and must be held in
8291ebeefahrens *	the following circumstances:
83fa9e406ahrens *
8444cd46cbillm *		- RW_READER to perform I/O to the spa
8544cd46cbillm *		- RW_WRITER to change the vdev config
86fa9e406ahrens *
87fa9e406ahrens * The locking order is fairly straightforward:
88fa9e406ahrens *
8944cd46cbillm *		spa_namespace_lock	->	spa_refcount
90fa9e406ahrens *
9144cd46cbillm *	The namespace lock must be acquired to increase the refcount from 0
9244cd46cbillm *	or to check if it is zero.
93fa9e406ahrens *
94e14bb32Jeff Bonwick *		spa_refcount		->	spa_config_lock[]
95fa9e406ahrens *
9644cd46cbillm *	There must be at least one valid reference on the spa_t to acquire
9744cd46cbillm *	the config lock.
98fa9e406ahrens *
99e14bb32Jeff Bonwick *		spa_namespace_lock	->	spa_config_lock[]
100fa9e406ahrens *
10144cd46cbillm *	The namespace lock must always be taken before the config lock.
102fa9e406ahrens *
103fa9e406ahrens *
104e14bb32Jeff Bonwick * The spa_namespace_lock can be acquired directly and is globally visible.
105fa9e406ahrens *
106e14bb32Jeff Bonwick * The namespace is manipulated using the following functions, all of which
107e14bb32Jeff Bonwick * require the spa_namespace_lock to be held.
108fa9e406ahrens *
10944cd46cbillm *	spa_lookup()		Lookup a spa_t by name.
110fa9e406ahrens *
11144cd46cbillm *	spa_add()		Create a new spa_t in the namespace.
112fa9e406ahrens *
11344cd46cbillm *	spa_remove()		Remove a spa_t from the namespace.  This also
11444cd46cbillm *				frees up any memory associated with the spa_t.
115fa9e406ahrens *
11644cd46cbillm *	spa_next()		Returns the next spa_t in the system, or the
11744cd46cbillm *				first if NULL is passed.
118fa9e406ahrens *
11944cd46cbillm *	spa_evict_all()		Shutdown and remove all spa_t structures in
12044cd46cbillm *				the system.
121fa9e406ahrens *
122ea8dc4beschrock *	spa_guid_exists()	Determine whether a pool/device guid exists.
123fa9e406ahrens *
124fa9e406ahrens * The spa_refcount is manipulated using the following functions:
125fa9e406ahrens *
12644cd46cbillm *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
12744cd46cbillm *				called with spa_namespace_lock held if the
12844cd46cbillm *				refcount is currently zero.
129fa9e406ahrens *
13044cd46cbillm *	spa_close()		Remove a reference from the spa_t.  This will
13144cd46cbillm *				not free the spa_t or remove it from the
13244cd46cbillm *				namespace.  No locking is required.
133fa9e406ahrens *
13444cd46cbillm *	spa_refcount_zero()	Returns true if the refcount is currently
13544cd46cbillm *				zero.  Must be called with spa_namespace_lock
13644cd46cbillm *				held.
137fa9e406ahrens *
138e14bb32Jeff Bonwick * The spa_config_lock[] is an array of rwlocks, ordered as follows:
139e14bb32Jeff Bonwick * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
140e14bb32Jeff Bonwick * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
141e14bb32Jeff Bonwick *
142e14bb32Jeff Bonwick * To read the configuration, it suffices to hold one of these locks as reader.
143e14bb32Jeff Bonwick * To modify the configuration, you must hold all locks as writer.  To modify
144e14bb32Jeff Bonwick * vdev state without altering the vdev tree's topology (e.g. online/offline),
145e14bb32Jeff Bonwick * you must hold SCL_STATE and SCL_ZIO as writer.
146e14bb32Jeff Bonwick *
147e14bb32Jeff Bonwick * We use these distinct config locks to avoid recursive lock entry.
148e14bb32Jeff Bonwick * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
149e14bb32Jeff Bonwick * block allocations (SCL_ALLOC), which may require reading space maps
150e14bb32Jeff Bonwick * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
151e14bb32Jeff Bonwick *
152e14bb32Jeff Bonwick * The spa config locks cannot be normal rwlocks because we need the
153e14bb32Jeff Bonwick * ability to hand off ownership.  For example, SCL_ZIO is acquired
154e14bb32Jeff Bonwick * by the issuing thread and later released by an interrupt thread.
155e14bb32Jeff Bonwick * They do, however, obey the usual write-wanted semantics to prevent
156e14bb32Jeff Bonwick * writer (i.e. system administrator) starvation.
157e14bb32Jeff Bonwick *
158e14bb32Jeff Bonwick * The lock acquisition rules are as follows:
159e14bb32Jeff Bonwick *
160e14bb32Jeff Bonwick * SCL_CONFIG
161e14bb32Jeff Bonwick *	Protects changes to the vdev tree topology, such as vdev
162e14bb32Jeff Bonwick *	add/remove/attach/detach.  Protects the dirty config list
163e14bb32Jeff Bonwick *	(spa_config_dirty_list) and the set of spares and l2arc devices.
164e14bb32Jeff Bonwick *
165e14bb32Jeff Bonwick * SCL_STATE
166e14bb32Jeff Bonwick *	Protects changes to pool state and vdev state, such as vdev
167e14bb32Jeff Bonwick *	online/offline/fault/degrade/clear.  Protects the dirty state list
168e14bb32Jeff Bonwick *	(spa_state_dirty_list) and global pool state (spa_state).
169e14bb32Jeff Bonwick *
170e14bb32Jeff Bonwick * SCL_ALLOC
171e14bb32Jeff Bonwick *	Protects changes to metaslab groups and classes.
172e14bb32Jeff Bonwick *	Held as reader by metaslab_alloc() and metaslab_claim().
173e14bb32Jeff Bonwick *
174e14bb32Jeff Bonwick * SCL_ZIO
175e14bb32Jeff Bonwick *	Held by bp-level zios (those which have no io_vd upon entry)
176e14bb32Jeff Bonwick *	to prevent changes to the vdev tree.  The bp-level zio implicitly
177e14bb32Jeff Bonwick *	protects all of its vdev child zios, which do not hold SCL_ZIO.
178e14bb32Jeff Bonwick *
179e14bb32Jeff Bonwick * SCL_FREE
180e14bb32Jeff Bonwick *	Protects changes to metaslab groups and classes.
181e14bb32Jeff Bonwick *	Held as reader by metaslab_free().  SCL_FREE is distinct from
182e14bb32Jeff Bonwick *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
183e14bb32Jeff Bonwick *	blocks in zio_done() while another i/o that holds either
184e14bb32Jeff Bonwick *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
185e14bb32Jeff Bonwick *
186e14bb32Jeff Bonwick * SCL_VDEV
187e14bb32Jeff Bonwick *	Held as reader to prevent changes to the vdev tree during trivial
188b24ab67Jeff Bonwick *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
189e14bb32Jeff Bonwick *	other locks, and lower than all of them, to ensure that it's safe
190e14bb32Jeff Bonwick *	to acquire regardless of caller context.
191e14bb32Jeff Bonwick *
192e14bb32Jeff Bonwick * In addition, the following rules apply:
193e14bb32Jeff Bonwick *
194e14bb32Jeff Bonwick * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
195e14bb32Jeff Bonwick *	The lock ordering is SCL_CONFIG > spa_props_lock.
196e14bb32Jeff Bonwick *
197e14bb32Jeff Bonwick * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
198e14bb32Jeff Bonwick *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
199e14bb32Jeff Bonwick *	or zio_write_phys() -- the caller must ensure that the config cannot
200e14bb32Jeff Bonwick *	cannot change in the interim, and that the vdev cannot be reopened.
201e14bb32Jeff Bonwick *	SCL_STATE as reader suffices for both.
202fa9e406ahrens *
203ea8dc4beschrock * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
204fa9e406ahrens *
20544cd46cbillm *	spa_vdev_enter()	Acquire the namespace lock and the config lock
206ea8dc4beschrock *				for writing.
207fa9e406ahrens *
20844cd46cbillm *	spa_vdev_exit()		Release the config lock, wait for all I/O
20944cd46cbillm *				to complete, sync the updated configs to the
210ea8dc4beschrock *				cache, and release the namespace lock.
211fa9e406ahrens *
212e14bb32Jeff Bonwick * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
213e14bb32Jeff Bonwick * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
214e14bb32Jeff Bonwick * locking is, always, based on spa_namespace_lock and spa_config_lock[].
215e14bb32Jeff Bonwick *
216e14bb32Jeff Bonwick * spa_rename() is also implemented within this file since is requires
217e14bb32Jeff Bonwick * manipulation of the namespace.
218fa9e406ahrens */
219fa9e406ahrens
220fa9e406ahrensstatic avl_tree_t spa_namespace_avl;
221fa9e406ahrenskmutex_t spa_namespace_lock;
222fa9e406ahrensstatic kcondvar_t spa_namespace_cv;
2230373e76bonwickstatic int spa_active_count;
224416e0cdekint spa_max_replication_override = SPA_DVAS_PER_BP;
225fa9e406ahrens
22699653d4eschrockstatic kmutex_t spa_spare_lock;
22739c2341eschrockstatic avl_tree_t spa_spare_avl;
228fa94a07brendanstatic kmutex_t spa_l2cache_lock;
229fa94a07brendanstatic avl_tree_t spa_l2cache_avl;
23099653d4eschrock
231fa9e406ahrenskmem_cache_t *spa_buffer_pool;
2328ad4d6dJeff Bonwickint spa_mode_global;
233fa9e406ahrens
234fa9e406ahrens#ifdef ZFS_DEBUG
23540feaa9ahrens/* Everything except dprintf is on by default in debug builds */
23640feaa9ahrensint zfs_flags = ~ZFS_DEBUG_DPRINTF;
237fa9e406ahrens#else
238fa9e406ahrensint zfs_flags = 0;
239fa9e406ahrens#endif
240fa9e406ahrens
2410125049ahrens/*
2420125049ahrens * zfs_recover can be set to nonzero to attempt to recover from
2430125049ahrens * otherwise-fatal errors, typically caused by on-disk corruption.  When
2440125049ahrens * set, calls to zfs_panic_recover() will turn into warning messages.
2450125049ahrens */
2460125049ahrensint zfs_recover = 0;
2470125049ahrens
248fa9e406ahrens
249fa9e406ahrens/*
250fa9e406ahrens * ==========================================================================
251e05725bbonwick * SPA config locking
252e05725bbonwick * ==========================================================================
253e05725bbonwick */
254e05725bbonwickstatic void
255e14bb32Jeff Bonwickspa_config_lock_init(spa_t *spa)
256e14bb32Jeff Bonwick{
257e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
258e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
259e14bb32Jeff Bonwick		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
260e14bb32Jeff Bonwick		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
261e14bb32Jeff Bonwick		refcount_create(&scl->scl_count);
262e14bb32Jeff Bonwick		scl->scl_writer = NULL;
263e14bb32Jeff Bonwick		scl->scl_write_wanted = 0;
264e14bb32Jeff Bonwick	}
265e05725bbonwick}
266e05725bbonwick
267e05725bbonwickstatic void
268e14bb32Jeff Bonwickspa_config_lock_destroy(spa_t *spa)
269e14bb32Jeff Bonwick{
270e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
271e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
272e14bb32Jeff Bonwick		mutex_destroy(&scl->scl_lock);
273e14bb32Jeff Bonwick		cv_destroy(&scl->scl_cv);
274e14bb32Jeff Bonwick		refcount_destroy(&scl->scl_count);
275e14bb32Jeff Bonwick		ASSERT(scl->scl_writer == NULL);
276e14bb32Jeff Bonwick		ASSERT(scl->scl_write_wanted == 0);
277e14bb32Jeff Bonwick	}
278e14bb32Jeff Bonwick}
279e14bb32Jeff Bonwick
280e14bb32Jeff Bonwickint
281e14bb32Jeff Bonwickspa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
282e05725bbonwick{
283e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
284e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
285e14bb32Jeff Bonwick		if (!(locks & (1 << i)))
286e14bb32Jeff Bonwick			continue;
287e14bb32Jeff Bonwick		mutex_enter(&scl->scl_lock);
288e14bb32Jeff Bonwick		if (rw == RW_READER) {
289e14bb32Jeff Bonwick			if (scl->scl_writer || scl->scl_write_wanted) {
290e14bb32Jeff Bonwick				mutex_exit(&scl->scl_lock);
291e14bb32Jeff Bonwick				spa_config_exit(spa, locks ^ (1 << i), tag);
292e14bb32Jeff Bonwick				return (0);
293e14bb32Jeff Bonwick			}
294e14bb32Jeff Bonwick		} else {
295e14bb32Jeff Bonwick			ASSERT(scl->scl_writer != curthread);
296e14bb32Jeff Bonwick			if (!refcount_is_zero(&scl->scl_count)) {
297e14bb32Jeff Bonwick				mutex_exit(&scl->scl_lock);
298e14bb32Jeff Bonwick				spa_config_exit(spa, locks ^ (1 << i), tag);
299e14bb32Jeff Bonwick				return (0);
300e14bb32Jeff Bonwick			}
301e14bb32Jeff Bonwick			scl->scl_writer = curthread;
302e14bb32Jeff Bonwick		}
303e14bb32Jeff Bonwick		(void) refcount_add(&scl->scl_count, tag);
304e14bb32Jeff Bonwick		mutex_exit(&scl->scl_lock);
305e14bb32Jeff Bonwick	}
306e14bb32Jeff Bonwick	return (1);
307e05725bbonwick}
308e05725bbonwick
309e05725bbonwickvoid
310e14bb32Jeff Bonwickspa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
311e05725bbonwick{
312f64c0e3Eric Taylor	int wlocks_held = 0;
313f64c0e3Eric Taylor
314e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
315e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
316f64c0e3Eric Taylor		if (scl->scl_writer == curthread)
317f64c0e3Eric Taylor			wlocks_held |= (1 << i);
318e14bb32Jeff Bonwick		if (!(locks & (1 << i)))
319e14bb32Jeff Bonwick			continue;
320e14bb32Jeff Bonwick		mutex_enter(&scl->scl_lock);
321e14bb32Jeff Bonwick		if (rw == RW_READER) {
322e14bb32Jeff Bonwick			while (scl->scl_writer || scl->scl_write_wanted) {
323e14bb32Jeff Bonwick				cv_wait(&scl->scl_cv, &scl->scl_lock);
324e14bb32Jeff Bonwick			}
325e14bb32Jeff Bonwick		} else {
326e14bb32Jeff Bonwick			ASSERT(scl->scl_writer != curthread);
327e14bb32Jeff Bonwick			while (!refcount_is_zero(&scl->scl_count)) {
328e14bb32Jeff Bonwick				scl->scl_write_wanted++;
329e14bb32Jeff Bonwick				cv_wait(&scl->scl_cv, &scl->scl_lock);
330e14bb32Jeff Bonwick				scl->scl_write_wanted--;
331e14bb32Jeff Bonwick			}
332e14bb32Jeff Bonwick			scl->scl_writer = curthread;
333e14bb32Jeff Bonwick		}
334e14bb32Jeff Bonwick		(void) refcount_add(&scl->scl_count, tag);
335e14bb32Jeff Bonwick		mutex_exit(&scl->scl_lock);
336e05725bbonwick	}
337f64c0e3Eric Taylor	ASSERT(wlocks_held <= locks);
338e05725bbonwick}
339e05725bbonwick
340e05725bbonwickvoid
341e14bb32Jeff Bonwickspa_config_exit(spa_t *spa, int locks, void *tag)
342e05725bbonwick{
343e14bb32Jeff Bonwick	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
344e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
345e14bb32Jeff Bonwick		if (!(locks & (1 << i)))
346e14bb32Jeff Bonwick			continue;
347e14bb32Jeff Bonwick		mutex_enter(&scl->scl_lock);
348e14bb32Jeff Bonwick		ASSERT(!refcount_is_zero(&scl->scl_count));
349e14bb32Jeff Bonwick		if (refcount_remove(&scl->scl_count, tag) == 0) {
350e14bb32Jeff Bonwick			ASSERT(scl->scl_writer == NULL ||
351e14bb32Jeff Bonwick			    scl->scl_writer == curthread);
352e14bb32Jeff Bonwick			scl->scl_writer = NULL;	/* OK in either case */
353e14bb32Jeff Bonwick			cv_broadcast(&scl->scl_cv);
354e14bb32Jeff Bonwick		}
355e14bb32Jeff Bonwick		mutex_exit(&scl->scl_lock);
356e05725bbonwick	}
357e05725bbonwick}
358e05725bbonwick
359e14bb32Jeff Bonwickint
360e14bb32Jeff Bonwickspa_config_held(spa_t *spa, int locks, krw_t rw)
361e05725bbonwick{
362e14bb32Jeff Bonwick	int locks_held = 0;
363e05725bbonwick
364e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
365e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
366e14bb32Jeff Bonwick		if (!(locks & (1 << i)))
367e14bb32Jeff Bonwick			continue;
368e14bb32Jeff Bonwick		if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
369e14bb32Jeff Bonwick		    (rw == RW_WRITER && scl->scl_writer == curthread))
370e14bb32Jeff Bonwick			locks_held |= 1 << i;
371e14bb32Jeff Bonwick	}
372e14bb32Jeff Bonwick
373e14bb32Jeff Bonwick	return (locks_held);
374e05725bbonwick}
375e05725bbonwick
376e05725bbonwick/*
377e05725bbonwick * ==========================================================================
378fa9e406ahrens * SPA namespace functions
379fa9e406ahrens * ==========================================================================
380fa9e406ahrens */
381fa9e406ahrens
382fa9e406ahrens/*
383fa9e406ahrens * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
384fa9e406ahrens * Returns NULL if no matching spa_t is found.
385fa9e406ahrens */
386fa9e406ahrensspa_t *
387fa9e406ahrensspa_lookup(const char *name)
388fa9e406ahrens{
389e14bb32Jeff Bonwick	static spa_t search;	/* spa_t is large; don't allocate on stack */
390e14bb32Jeff Bonwick	spa_t *spa;
391fa9e406ahrens	avl_index_t where;
39240feaa9ahrens	char c;
39340feaa9ahrens	char *cp;
394fa9e406ahrens
395fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
396fa9e406ahrens
39740feaa9ahrens	/*
39840feaa9ahrens	 * If it's a full dataset name, figure out the pool name and
39940feaa9ahrens	 * just use that.
40040feaa9ahrens	 */
40140feaa9ahrens	cp = strpbrk(name, "/@");
40240feaa9ahrens	if (cp) {
40340feaa9ahrens		c = *cp;
40440feaa9ahrens		*cp = '\0';
40540feaa9ahrens	}
40640feaa9ahrens
407e14bb32Jeff Bonwick	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
408fa9e406ahrens	spa = avl_find(&spa_namespace_avl, &search, &where);
409fa9e406ahrens
41040feaa9ahrens	if (cp)
41140feaa9ahrens		*cp = c;
41240feaa9ahrens
413fa9e406ahrens	return (spa);
414fa9e406ahrens}
415fa9e406ahrens
416fa9e406ahrens/*
417fa9e406ahrens * Create an uninitialized spa_t with the given name.  Requires
418fa9e406ahrens * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
419fa9e406ahrens * exist by calling spa_lookup() first.
420fa9e406ahrens */
421fa9e406ahrensspa_t *
422468c413Tim Haleyspa_add(const char *name, nvlist_t *config, const char *altroot)
423fa9e406ahrens{
424fa9e406ahrens	spa_t *spa;
425c5904d1eschrock	spa_config_dirent_t *dp;
426fa9e406ahrens
427fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
428fa9e406ahrens
429fa9e406ahrens	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
430fa9e406ahrens
431c25056dgw	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
432c25056dgw	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
43335a5a35Jonathan Adams	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
434c25056dgw	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
43535a5a35Jonathan Adams	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
436c25056dgw	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
43735a5a35Jonathan Adams	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
438a152156Jeff Bonwick	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
439a152156Jeff Bonwick	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
440c25056dgw
441c25056dgw	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
44235a5a35Jonathan Adams	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
443c25056dgw	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
444e14bb32Jeff Bonwick	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
445c25056dgw
446b24ab67Jeff Bonwick	for (int t = 0; t < TXG_SIZE; t++)
447b24ab67Jeff Bonwick		bplist_init(&spa->spa_free_bplist[t]);
448b24ab67Jeff Bonwick	bplist_init(&spa->spa_deferred_bplist);
449b24ab67Jeff Bonwick
450e14bb32Jeff Bonwick	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
451fa9e406ahrens	spa->spa_state = POOL_STATE_UNINITIALIZED;
452fa9e406ahrens	spa->spa_freeze_txg = UINT64_MAX;
4530373e76bonwick	spa->spa_final_txg = UINT64_MAX;
454468c413Tim Haley	spa->spa_load_max_txg = UINT64_MAX;
45535a5a35Jonathan Adams	spa->spa_proc = &p0;
45635a5a35Jonathan Adams	spa->spa_proc_state = SPA_PROC_NONE;
457fa9e406ahrens
458fa9e406ahrens	refcount_create(&spa->spa_refcount);
459e14bb32Jeff Bonwick	spa_config_lock_init(spa);
460fa9e406ahrens
461fa9e406ahrens	avl_add(&spa_namespace_avl, spa);
462fa9e406ahrens
4630373e76bonwick	/*
4640373e76bonwick	 * Set the alternate root, if there is one.
4650373e76bonwick	 */
4660373e76bonwick	if (altroot) {
4670373e76bonwick		spa->spa_root = spa_strdup(altroot);
4680373e76bonwick		spa_active_count++;
4690373e76bonwick	}
4700373e76bonwick
471c5904d1eschrock	/*
472c5904d1eschrock	 * Every pool starts with the default cachefile
473c5904d1eschrock	 */
474c5904d1eschrock	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
475c5904d1eschrock	    offsetof(spa_config_dirent_t, scd_link));
476c5904d1eschrock
477c5904d1eschrock	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
478ef912c8Tim Haley	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
479c5904d1eschrock	list_insert_head(&spa->spa_config_list, dp);
480c5904d1eschrock
481468c413Tim Haley	if (config != NULL)
482468c413Tim Haley		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
483468c413Tim Haley
484fa9e406ahrens	return (spa);
485fa9e406ahrens}
486fa9e406ahrens
487fa9e406ahrens/*
488fa9e406ahrens * Removes a spa_t from the namespace, freeing up any memory used.  Requires
489fa9e406ahrens * spa_namespace_lock.  This is called only after the spa_t has been closed and
490fa9e406ahrens * deactivated.
491fa9e406ahrens */
492fa9e406ahrensvoid
493fa9e406ahrensspa_remove(spa_t *spa)
494fa9e406ahrens{
495c5904d1eschrock	spa_config_dirent_t *dp;
496c5904d1eschrock
497fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
498fa9e406ahrens	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
499fa9e406ahrens
5001195e68Mark J Musante	nvlist_free(spa->spa_config_splitting);
5011195e68Mark J Musante
502fa9e406ahrens	avl_remove(&spa_namespace_avl, spa);
503fa9e406ahrens	cv_broadcast(&spa_namespace_cv);
504fa9e406ahrens
5050373e76bonwick	if (spa->spa_root) {
506fa9e406ahrens		spa_strfree(spa->spa_root);
5070373e76bonwick		spa_active_count--;
5080373e76bonwick	}
509fa9e406ahrens
510c5904d1eschrock	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
511c5904d1eschrock		list_remove(&spa->spa_config_list, dp);
512c5904d1eschrock		if (dp->scd_path != NULL)
513c5904d1eschrock			spa_strfree(dp->scd_path);
514c5904d1eschrock		kmem_free(dp, sizeof (spa_config_dirent_t));
515c5904d1eschrock	}
516c5904d1eschrock
517c5904d1eschrock	list_destroy(&spa->spa_config_list);
5182f8aaabeschrock
519fa9e406ahrens	spa_config_set(spa, NULL);
520fa9e406ahrens
521fa9e406ahrens	refcount_destroy(&spa->spa_refcount);
52291ebeefahrens
523e14bb32Jeff Bonwick	spa_config_lock_destroy(spa);
524fa9e406ahrens
525b24ab67Jeff Bonwick	for (int t = 0; t < TXG_SIZE; t++)
526b24ab67Jeff Bonwick		bplist_fini(&spa->spa_free_bplist[t]);
527b24ab67Jeff Bonwick	bplist_fini(&spa->spa_deferred_bplist);
528b24ab67Jeff Bonwick
529c25056dgw	cv_destroy(&spa->spa_async_cv);
53035a5a35Jonathan Adams	cv_destroy(&spa->spa_proc_cv);
531c25056dgw	cv_destroy(&spa->spa_scrub_io_cv);
532e14bb32Jeff Bonwick	cv_destroy(&spa->spa_suspend_cv);
533c25056dgw
5345ad8204nd	mutex_destroy(&spa->spa_async_lock);
535c25056dgw	mutex_destroy(&spa->spa_errlist_lock);
53635a5a35Jonathan Adams	mutex_destroy(&spa->spa_errlog_lock);
53706eeb2aek	mutex_destroy(&spa->spa_history_lock);
53835a5a35Jonathan Adams	mutex_destroy(&spa->spa_proc_lock);
539b1b8ab3lling	mutex_destroy(&spa->spa_props_lock);
54035a5a35Jonathan Adams	mutex_destroy(&spa->spa_scrub_lock);
541e14bb32Jeff Bonwick	mutex_destroy(&spa->spa_suspend_lock);
542a152156Jeff Bonwick	mutex_destroy(&spa->spa_vdev_top_lock);
5435ad8204nd
544fa9e406ahrens	kmem_free(spa, sizeof (spa_t));
545fa9e406ahrens}
546fa9e406ahrens
547fa9e406ahrens/*
548fa9e406ahrens * Given a pool, return the next pool in the namespace, or NULL if there is
549fa9e406ahrens * none.  If 'prev' is NULL, return the first pool.
550fa9e406ahrens */
551fa9e406ahrensspa_t *
552fa9e406ahrensspa_next(spa_t *prev)
553fa9e406ahrens{
554fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
555fa9e406ahrens
556fa9e406ahrens	if (prev)
557fa9e406ahrens		return (AVL_NEXT(&spa_namespace_avl, prev));
558fa9e406ahrens	else
559fa9e406ahrens		return (avl_first(&spa_namespace_avl));
560fa9e406ahrens}
561fa9e406ahrens
562fa9e406ahrens/*
563fa9e406ahrens * ==========================================================================
564fa9e406ahrens * SPA refcount functions
565fa9e406ahrens * ==========================================================================
566fa9e406ahrens */
567fa9e406ahrens
568fa9e406ahrens/*
569fa9e406ahrens * Add a reference to the given spa_t.  Must have at least one reference, or
570fa9e406ahrens * have the namespace lock held.
571fa9e406ahrens */
572fa9e406ahrensvoid
573fa9e406ahrensspa_open_ref(spa_t *spa, void *tag)
574fa9e406ahrens{
575088f389ahrens	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
576fa9e406ahrens	    MUTEX_HELD(&spa_namespace_lock));
577fa9e406ahrens	(void) refcount_add(&spa->spa_refcount, tag);
578fa9e406ahrens}
579fa9e406ahrens
580fa9e406ahrens/*
581fa9e406ahrens * Remove a reference to the given spa_t.  Must have at least one reference, or
582fa9e406ahrens * have the namespace lock held.
583fa9e406ahrens */
584fa9e406ahrensvoid
585fa9e406ahrensspa_close(spa_t *spa, void *tag)
586fa9e406ahrens{
587088f389ahrens	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
588fa9e406ahrens	    MUTEX_HELD(&spa_namespace_lock));
589fa9e406ahrens	(void) refcount_remove(&spa->spa_refcount, tag);
590fa9e406ahrens}
591fa9e406ahrens
592fa9e406ahrens/*
593fa9e406ahrens * Check to see if the spa refcount is zero.  Must be called with
594088f389ahrens * spa_namespace_lock held.  We really compare against spa_minref, which is the
595fa9e406ahrens * number of references acquired when opening a pool
596fa9e406ahrens */
597fa9e406ahrensboolean_t
598fa9e406ahrensspa_refcount_zero(spa_t *spa)
599fa9e406ahrens{
600fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
601fa9e406ahrens
602088f389ahrens	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
603fa9e406ahrens}
604fa9e406ahrens
605fa9e406ahrens/*
606fa9e406ahrens * ==========================================================================
607fa94a07brendan * SPA spare and l2cache tracking
60899653d4eschrock * ==========================================================================
60999653d4eschrock */
61099653d4eschrock
61199653d4eschrock/*
612fa94a07brendan * Hot spares and cache devices are tracked using the same code below,
613fa94a07brendan * for 'auxiliary' devices.
614fa94a07brendan */
615fa94a07brendan
616fa94a07brendantypedef struct spa_aux {
617fa94a07brendan	uint64_t	aux_guid;
618fa94a07brendan	uint64_t	aux_pool;
619fa94a07brendan	avl_node_t	aux_avl;
620fa94a07brendan	int		aux_count;
621fa94a07brendan} spa_aux_t;
622fa94a07brendan
623fa94a07brendanstatic int
624fa94a07brendanspa_aux_compare(const void *a, const void *b)
625fa94a07brendan{
626fa94a07brendan	const spa_aux_t *sa = a;
627fa94a07brendan	const spa_aux_t *sb = b;
628fa94a07brendan
629fa94a07brendan	if (sa->aux_guid < sb->aux_guid)
630fa94a07brendan		return (-1);
631fa94a07brendan	else if (sa->aux_guid > sb->aux_guid)
632fa94a07brendan		return (1);
633fa94a07brendan	else
634fa94a07brendan		return (0);
635fa94a07brendan}
636fa94a07brendan
637fa94a07brendanvoid
638fa94a07brendanspa_aux_add(vdev_t *vd, avl_tree_t *avl)
639fa94a07brendan{
640fa94a07brendan	avl_index_t where;
641fa94a07brendan	spa_aux_t search;
642fa94a07brendan	spa_aux_t *aux;
643fa94a07brendan
644fa94a07brendan	search.aux_guid = vd->vdev_guid;
645fa94a07brendan	if ((aux = avl_find(avl, &search, &where)) != NULL) {
646fa94a07brendan		aux->aux_count++;
647fa94a07brendan	} else {
648fa94a07brendan		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
649fa94a07brendan		aux->aux_guid = vd->vdev_guid;
650fa94a07brendan		aux->aux_count = 1;
651fa94a07brendan		avl_insert(avl, aux, where);
652fa94a07brendan	}
653fa94a07brendan}
654fa94a07brendan
655fa94a07brendanvoid
656fa94a07brendanspa_aux_remove(vdev_t *vd, avl_tree_t *avl)
657fa94a07brendan{
658fa94a07brendan	spa_aux_t search;
659fa94a07brendan	spa_aux_t *aux;
660fa94a07brendan	avl_index_t where;
661fa94a07brendan
662fa94a07brendan	search.aux_guid = vd->vdev_guid;
663fa94a07brendan	aux = avl_find(avl, &search, &where);
664fa94a07brendan
665fa94a07brendan	ASSERT(aux != NULL);
666fa94a07brendan
667fa94a07brendan	if (--aux->aux_count == 0) {
668fa94a07brendan		avl_remove(avl, aux);
669fa94a07brendan		kmem_free(aux, sizeof (spa_aux_t));
670fa94a07brendan	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
671fa94a07brendan		aux->aux_pool = 0ULL;
672fa94a07brendan	}
673fa94a07brendan}
674fa94a07brendan
675fa94a07brendanboolean_t
67689a89ebllingspa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
677fa94a07brendan{
678fa94a07brendan	spa_aux_t search, *found;
679fa94a07brendan
680fa94a07brendan	search.aux_guid = guid;
68189a89eblling	found = avl_find(avl, &search, NULL);
682fa94a07brendan
683fa94a07brendan	if (pool) {
684fa94a07brendan		if (found)
685fa94a07brendan			*pool = found->aux_pool;
686fa94a07brendan		else
687fa94a07brendan			*pool = 0ULL;
688fa94a07brendan	}
689fa94a07brendan
69089a89eblling	if (refcnt) {
69189a89eblling		if (found)
69289a89eblling			*refcnt = found->aux_count;
69389a89eblling		else
69489a89eblling			*refcnt = 0;
69589a89eblling	}
69689a89eblling
697fa94a07brendan	return (found != NULL);
698fa94a07brendan}
699fa94a07brendan
700fa94a07brendanvoid
701fa94a07brendanspa_aux_activate(vdev_t *vd, avl_tree_t *avl)
702fa94a07brendan{
703fa94a07brendan	spa_aux_t search, *found;
704fa94a07brendan	avl_index_t where;
705fa94a07brendan
706fa94a07brendan	search.aux_guid = vd->vdev_guid;
707fa94a07brendan	found = avl_find(avl, &search, &where);
708fa94a07brendan	ASSERT(found != NULL);
709fa94a07brendan	ASSERT(found->aux_pool == 0ULL);
710fa94a07brendan
711fa94a07brendan	found->aux_pool = spa_guid(vd->vdev_spa);
712fa94a07brendan}
713fa94a07brendan
714fa94a07brendan/*
71539c2341eschrock * Spares are tracked globally due to the following constraints:
71639c2341eschrock *
71739c2341eschrock * 	- A spare may be part of multiple pools.
71839c2341eschrock * 	- A spare may be added to a pool even if it's actively in use within
71939c2341eschrock *	  another pool.
72039c2341eschrock * 	- A spare in use in any pool can only be the source of a replacement if
72139c2341eschrock *	  the target is a spare in the same pool.
72239c2341eschrock *
72339c2341eschrock * We keep track of all spares on the system through the use of a reference
72439c2341eschrock * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
72539c2341eschrock * spare, then we bump the reference count in the AVL tree.  In addition, we set
72639c2341eschrock * the 'vdev_isspare' member to indicate that the device is a spare (active or
72739c2341eschrock * inactive).  When a spare is made active (used to replace a device in the
72839c2341eschrock * pool), we also keep track of which pool its been made a part of.
72939c2341eschrock *
73039c2341eschrock * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
73139c2341eschrock * called under the spa_namespace lock as part of vdev reconfiguration.  The
73239c2341eschrock * separate spare lock exists for the status query path, which does not need to
73339c2341eschrock * be completely consistent with respect to other vdev configuration changes.
73499653d4eschrock */
73539c2341eschrock
73699653d4eschrockstatic int
73799653d4eschrockspa_spare_compare(const void *a, const void *b)
73899653d4eschrock{
739fa94a07brendan	return (spa_aux_compare(a, b));
74099653d4eschrock}
74199653d4eschrock
74299653d4eschrockvoid
74339c2341eschrockspa_spare_add(vdev_t *vd)
74499653d4eschrock{
74599653d4eschrock	mutex_enter(&spa_spare_lock);
74639c2341eschrock	ASSERT(!vd->vdev_isspare);
747fa94a07brendan	spa_aux_add(vd, &spa_spare_avl);
74839c2341eschrock	vd->vdev_isspare = B_TRUE;
74999653d4eschrock	mutex_exit(&spa_spare_lock);
75099653d4eschrock}
75199653d4eschrock
75299653d4eschrockvoid
75339c2341eschrockspa_spare_remove(vdev_t *vd)
75499653d4eschrock{
75599653d4eschrock	mutex_enter(&spa_spare_lock);
75639c2341eschrock	ASSERT(vd->vdev_isspare);
757fa94a07brendan	spa_aux_remove(vd, &spa_spare_avl);
75839c2341eschrock	vd->vdev_isspare = B_FALSE;
75999653d4eschrock	mutex_exit(&spa_spare_lock);
76099653d4eschrock}
76199653d4eschrock
76299653d4eschrockboolean_t
76389a89ebllingspa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
76499653d4eschrock{
765fa94a07brendan	boolean_t found;
76699653d4eschrock
76799653d4eschrock	mutex_enter(&spa_spare_lock);
76889a89eblling	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
76999653d4eschrock	mutex_exit(&spa_spare_lock);
77099653d4eschrock
771fa94a07brendan	return (found);
77239c2341eschrock}
77339c2341eschrock
77439c2341eschrockvoid
77539c2341eschrockspa_spare_activate(vdev_t *vd)
77639c2341eschrock{
77739c2341eschrock	mutex_enter(&spa_spare_lock);
77839c2341eschrock	ASSERT(vd->vdev_isspare);
779fa94a07brendan	spa_aux_activate(vd, &spa_spare_avl);
780fa94a07brendan	mutex_exit(&spa_spare_lock);
781fa94a07brendan}
78239c2341eschrock
783fa94a07brendan/*
784fa94a07brendan * Level 2 ARC devices are tracked globally for the same reasons as spares.
785fa94a07brendan * Cache devices currently only support one pool per cache device, and so
786fa94a07brendan * for these devices the aux reference count is currently unused beyond 1.
787fa94a07brendan */
78839c2341eschrock
789fa94a07brendanstatic int
790fa94a07brendanspa_l2cache_compare(const void *a, const void *b)
791fa94a07brendan{
792fa94a07brendan	return (spa_aux_compare(a, b));
793fa94a07brendan}
794fa94a07brendan
795fa94a07brendanvoid
796fa94a07brendanspa_l2cache_add(vdev_t *vd)
797fa94a07brendan{
798fa94a07brendan	mutex_enter(&spa_l2cache_lock);
799fa94a07brendan	ASSERT(!vd->vdev_isl2cache);
800fa94a07brendan	spa_aux_add(vd, &spa_l2cache_avl);
801fa94a07brendan	vd->vdev_isl2cache = B_TRUE;
802fa94a07brendan	mutex_exit(&spa_l2cache_lock);
803fa94a07brendan}
804fa94a07brendan
805fa94a07brendanvoid
806fa94a07brendanspa_l2cache_remove(vdev_t *vd)
807fa94a07brendan{
808fa94a07brendan	mutex_enter(&spa_l2cache_lock);
809fa94a07brendan	ASSERT(vd->vdev_isl2cache);
810fa94a07brendan	spa_aux_remove(vd, &spa_l2cache_avl);
811fa94a07brendan	vd->vdev_isl2cache = B_FALSE;
812fa94a07brendan	mutex_exit(&spa_l2cache_lock);
813fa94a07brendan}
814fa94a07brendan
815fa94a07brendanboolean_t
816fa94a07brendanspa_l2cache_exists(uint64_t guid, uint64_t *pool)
817fa94a07brendan{
818fa94a07brendan	boolean_t found;
819fa94a07brendan
820fa94a07brendan	mutex_enter(&spa_l2cache_lock);
82189a89eblling	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
822fa94a07brendan	mutex_exit(&spa_l2cache_lock);
823fa94a07brendan
824fa94a07brendan	return (found);
825fa94a07brendan}
826fa94a07brendan
827fa94a07brendanvoid
828fa94a07brendanspa_l2cache_activate(vdev_t *vd)
829fa94a07brendan{
830fa94a07brendan	mutex_enter(&spa_l2cache_lock);
831fa94a07brendan	ASSERT(vd->vdev_isl2cache);
832fa94a07brendan	spa_aux_activate(vd, &spa_l2cache_avl);
833fa94a07brendan	mutex_exit(&spa_l2cache_lock);
834fa94a07brendan}
835fa94a07brendan
83699653d4eschrock/*
83799653d4eschrock * ==========================================================================
838fa9e406ahrens * SPA vdev locking
839fa9e406ahrens * ==========================================================================
840fa9e406ahrens */
841fa9e406ahrens
842fa9e406ahrens/*
843ea8dc4beschrock * Lock the given spa_t for the purpose of adding or removing a vdev.
844ea8dc4beschrock * Grabs the global spa_namespace_lock plus the spa config lock for writing.
845fa9e406ahrens * It returns the next transaction group for the spa_t.
846fa9e406ahrens */
847fa9e406ahrensuint64_t
848fa9e406ahrensspa_vdev_enter(spa_t *spa)
849fa9e406ahrens{
850a152156Jeff Bonwick	mutex_enter(&spa->spa_vdev_top_lock);
851bbfd46cJeff Bonwick	mutex_enter(&spa_namespace_lock);
85288ecc94George Wilson	return (spa_vdev_config_enter(spa));
85388ecc94George Wilson}
85488ecc94George Wilson
85588ecc94George Wilson/*
85688ecc94George Wilson * Internal implementation for spa_vdev_enter().  Used when a vdev
85788ecc94George Wilson * operation requires multiple syncs (i.e. removing a device) while
85888ecc94George Wilson * keeping the spa_namespace_lock held.
85988ecc94George Wilson */
86088ecc94George Wilsonuint64_t
86188ecc94George Wilsonspa_vdev_config_enter(spa_t *spa)
86288ecc94George Wilson{
86388ecc94George Wilson	ASSERT(MUTEX_HELD(&spa_namespace_lock));
8643d7072feschrock
865e14bb32Jeff Bonwick	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
866fa9e406ahrens
867fa9e406ahrens	return (spa_last_synced_txg(spa) + 1);
868fa9e406ahrens}
869fa9e406ahrens
870fa9e406ahrens/*
87188ecc94George Wilson * Used in combination with spa_vdev_config_enter() to allow the syncing
87288ecc94George Wilson * of multiple transactions without releasing the spa_namespace_lock.
873fa9e406ahrens */
87488ecc94George Wilsonvoid
87588ecc94George Wilsonspa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
876fa9e406ahrens{
87788ecc94George Wilson	ASSERT(MUTEX_HELD(&spa_namespace_lock));
87888ecc94George Wilson
8790e34b6abonwick	int config_changed = B_FALSE;
880ea8dc4beschrock
8810373e76bonwick	ASSERT(txg > spa_last_synced_txg(spa));
8820e34b6abonwick
883e14bb32Jeff Bonwick	spa->spa_pending_vdev = NULL;
884e14bb32Jeff Bonwick
8850e34b6abonwick	/*
8860e34b6abonwick	 * Reassess the DTLs.
8870e34b6abonwick	 */
8880373e76bonwick	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
8890e34b6abonwick
8900e34b6abonwick	/*
8910373e76bonwick	 * If the config changed, notify the scrub thread that it must restart.
8920e34b6abonwick	 */
893e14bb32Jeff Bonwick	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
894088f389ahrens		dsl_pool_scrub_restart(spa->spa_dsl_pool);
8950e34b6abonwick		config_changed = B_TRUE;
8968f18d1fGeorge Wilson		spa->spa_config_generation++;
8970e34b6abonwick	}
898ea8dc4beschrock
89988ecc94George Wilson	/*
90088ecc94George Wilson	 * Verify the metaslab classes.
90188ecc94George Wilson	 */
902b24ab67Jeff Bonwick	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
903b24ab67Jeff Bonwick	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
90488ecc94George Wilson
905e14bb32Jeff Bonwick	spa_config_exit(spa, SCL_ALL, spa);
906fa9e406ahrens
907ea8dc4beschrock	/*
90888ecc94George Wilson	 * Panic the system if the specified tag requires it.  This
90988ecc94George Wilson	 * is useful for ensuring that configurations are updated
91088ecc94George Wilson	 * transactionally.
91188ecc94George Wilson	 */
91288ecc94George Wilson	if (zio_injection_enabled)
913