spa_misc.c revision bc9014e6a81272073b9854d9f65dd59e18d18c35
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.
232a104a5Alex Reece * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
24e9103aaGarrett D'Amore * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25bc9014eJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26fa9e406ahrens */
27fa9e406ahrens
28fa9e406ahrens#include <sys/zfs_context.h>
29fa9e406ahrens#include <sys/spa_impl.h>
30283b846George.Wilson#include <sys/spa_boot.h>
31fa9e406ahrens#include <sys/zio.h>
32fa9e406ahrens#include <sys/zio_checksum.h>
33fa9e406ahrens#include <sys/zio_compress.h>
34fa9e406ahrens#include <sys/dmu.h>
35fa9e406ahrens#include <sys/dmu_tx.h>
36fa9e406ahrens#include <sys/zap.h>
37fa9e406ahrens#include <sys/zil.h>
38fa9e406ahrens#include <sys/vdev_impl.h>
39fa9e406ahrens#include <sys/metaslab.h>
40fa9e406ahrens#include <sys/uberblock_impl.h>
41fa9e406ahrens#include <sys/txg.h>
42fa9e406ahrens#include <sys/avl.h>
43fa9e406ahrens#include <sys/unique.h>
44fa9e406ahrens#include <sys/dsl_pool.h>
45fa9e406ahrens#include <sys/dsl_dir.h>
46fa9e406ahrens#include <sys/dsl_prop.h>
473f9d6adLin Ling#include <sys/dsl_scan.h>
48fa9e406ahrens#include <sys/fs/zfs.h>
496ce0521perrin#include <sys/metaslab_impl.h>
50e14bb32Jeff Bonwick#include <sys/arc.h>
51485bbbfGeorge Wilson#include <sys/ddt.h>
5291ebeefahrens#include "zfs_prop.h"
53ad135b5Christopher Siden#include "zfeature_common.h"
54fa9e406ahrens
55fa9e406ahrens/*
56fa9e406ahrens * SPA locking
57fa9e406ahrens *
58fa9e406ahrens * There are four basic locks for managing spa_t structures:
59fa9e406ahrens *
60fa9e406ahrens * spa_namespace_lock (global mutex)
61fa9e406ahrens *
6244cd46cbillm *	This lock must be acquired to do any of the following:
63fa9e406ahrens *
6444cd46cbillm *		- Lookup a spa_t by name
6544cd46cbillm *		- Add or remove a spa_t from the namespace
6644cd46cbillm *		- Increase spa_refcount from non-zero
6744cd46cbillm *		- Check if spa_refcount is zero
6844cd46cbillm *		- Rename a spa_t
69ea8dc4beschrock *		- add/remove/attach/detach devices
7044cd46cbillm *		- Held for the duration of create/destroy/import/export
71fa9e406ahrens *
7244cd46cbillm *	It does not need to handle recursion.  A create or destroy may
7344cd46cbillm *	reference objects (files or zvols) in other pools, but by
7444cd46cbillm *	definition they must have an existing reference, and will never need
7544cd46cbillm *	to lookup a spa_t by name.
76fa9e406ahrens *
77fa9e406ahrens * spa_refcount (per-spa refcount_t protected by mutex)
78fa9e406ahrens *
7944cd46cbillm *	This reference count keep track of any active users of the spa_t.  The
8044cd46cbillm *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
8144cd46cbillm *	the refcount is never really 'zero' - opening a pool implicitly keeps
82088f389ahrens *	some references in the DMU.  Internally we check against spa_minref, but
8344cd46cbillm *	present the image of a zero/non-zero value to consumers.
84fa9e406ahrens *
85e14bb32Jeff Bonwick * spa_config_lock[] (per-spa array of rwlocks)
86fa9e406ahrens *
8791ebeefahrens *	This protects the spa_t from config changes, and must be held in
8891ebeefahrens *	the following circumstances:
89fa9e406ahrens *
9044cd46cbillm *		- RW_READER to perform I/O to the spa
9144cd46cbillm *		- RW_WRITER to change the vdev config
92fa9e406ahrens *
93fa9e406ahrens * The locking order is fairly straightforward:
94fa9e406ahrens *
9544cd46cbillm *		spa_namespace_lock	->	spa_refcount
96fa9e406ahrens *
9744cd46cbillm *	The namespace lock must be acquired to increase the refcount from 0
9844cd46cbillm *	or to check if it is zero.
99fa9e406ahrens *
100e14bb32Jeff Bonwick *		spa_refcount		->	spa_config_lock[]
101fa9e406ahrens *
10244cd46cbillm *	There must be at least one valid reference on the spa_t to acquire
10344cd46cbillm *	the config lock.
104fa9e406ahrens *
105e14bb32Jeff Bonwick *		spa_namespace_lock	->	spa_config_lock[]
106fa9e406ahrens *
10744cd46cbillm *	The namespace lock must always be taken before the config lock.
108fa9e406ahrens *
109fa9e406ahrens *
110e14bb32Jeff Bonwick * The spa_namespace_lock can be acquired directly and is globally visible.
111fa9e406ahrens *
112e14bb32Jeff Bonwick * The namespace is manipulated using the following functions, all of which
113e14bb32Jeff Bonwick * require the spa_namespace_lock to be held.
114fa9e406ahrens *
11544cd46cbillm *	spa_lookup()		Lookup a spa_t by name.
116fa9e406ahrens *
11744cd46cbillm *	spa_add()		Create a new spa_t in the namespace.
118fa9e406ahrens *
11944cd46cbillm *	spa_remove()		Remove a spa_t from the namespace.  This also
12044cd46cbillm *				frees up any memory associated with the spa_t.
121fa9e406ahrens *
12244cd46cbillm *	spa_next()		Returns the next spa_t in the system, or the
12344cd46cbillm *				first if NULL is passed.
124fa9e406ahrens *
12544cd46cbillm *	spa_evict_all()		Shutdown and remove all spa_t structures in
12644cd46cbillm *				the system.
127fa9e406ahrens *
128ea8dc4beschrock *	spa_guid_exists()	Determine whether a pool/device guid exists.
129fa9e406ahrens *
130fa9e406ahrens * The spa_refcount is manipulated using the following functions:
131fa9e406ahrens *
13244cd46cbillm *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
13344cd46cbillm *				called with spa_namespace_lock held if the
13444cd46cbillm *				refcount is currently zero.
135fa9e406ahrens *
13644cd46cbillm *	spa_close()		Remove a reference from the spa_t.  This will
13744cd46cbillm *				not free the spa_t or remove it from the
13844cd46cbillm *				namespace.  No locking is required.
139fa9e406ahrens *
14044cd46cbillm *	spa_refcount_zero()	Returns true if the refcount is currently
14144cd46cbillm *				zero.  Must be called with spa_namespace_lock
14244cd46cbillm *				held.
143fa9e406ahrens *
144e14bb32Jeff Bonwick * The spa_config_lock[] is an array of rwlocks, ordered as follows:
145e14bb32Jeff Bonwick * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
146e14bb32Jeff Bonwick * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
147e14bb32Jeff Bonwick *
148e14bb32Jeff Bonwick * To read the configuration, it suffices to hold one of these locks as reader.
149e14bb32Jeff Bonwick * To modify the configuration, you must hold all locks as writer.  To modify
150e14bb32Jeff Bonwick * vdev state without altering the vdev tree's topology (e.g. online/offline),
151e14bb32Jeff Bonwick * you must hold SCL_STATE and SCL_ZIO as writer.
152e14bb32Jeff Bonwick *
153e14bb32Jeff Bonwick * We use these distinct config locks to avoid recursive lock entry.
154e14bb32Jeff Bonwick * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
155e14bb32Jeff Bonwick * block allocations (SCL_ALLOC), which may require reading space maps
156e14bb32Jeff Bonwick * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
157e14bb32Jeff Bonwick *
158e14bb32Jeff Bonwick * The spa config locks cannot be normal rwlocks because we need the
159e14bb32Jeff Bonwick * ability to hand off ownership.  For example, SCL_ZIO is acquired
160e14bb32Jeff Bonwick * by the issuing thread and later released by an interrupt thread.
161e14bb32Jeff Bonwick * They do, however, obey the usual write-wanted semantics to prevent
162e14bb32Jeff Bonwick * writer (i.e. system administrator) starvation.
163e14bb32Jeff Bonwick *
164e14bb32Jeff Bonwick * The lock acquisition rules are as follows:
165e14bb32Jeff Bonwick *
166e14bb32Jeff Bonwick * SCL_CONFIG
167e14bb32Jeff Bonwick *	Protects changes to the vdev tree topology, such as vdev
168e14bb32Jeff Bonwick *	add/remove/attach/detach.  Protects the dirty config list
169e14bb32Jeff Bonwick *	(spa_config_dirty_list) and the set of spares and l2arc devices.
170e14bb32Jeff Bonwick *
171e14bb32Jeff Bonwick * SCL_STATE
172e14bb32Jeff Bonwick *	Protects changes to pool state and vdev state, such as vdev
173e14bb32Jeff Bonwick *	online/offline/fault/degrade/clear.  Protects the dirty state list
174e14bb32Jeff Bonwick *	(spa_state_dirty_list) and global pool state (spa_state).
175e14bb32Jeff Bonwick *
176e14bb32Jeff Bonwick * SCL_ALLOC
177e14bb32Jeff Bonwick *	Protects changes to metaslab groups and classes.
178e14bb32Jeff Bonwick *	Held as reader by metaslab_alloc() and metaslab_claim().
179e14bb32Jeff Bonwick *
180e14bb32Jeff Bonwick * SCL_ZIO
181e14bb32Jeff Bonwick *	Held by bp-level zios (those which have no io_vd upon entry)
182e14bb32Jeff Bonwick *	to prevent changes to the vdev tree.  The bp-level zio implicitly
183e14bb32Jeff Bonwick *	protects all of its vdev child zios, which do not hold SCL_ZIO.
184e14bb32Jeff Bonwick *
185e14bb32Jeff Bonwick * SCL_FREE
186e14bb32Jeff Bonwick *	Protects changes to metaslab groups and classes.
187e14bb32Jeff Bonwick *	Held as reader by metaslab_free().  SCL_FREE is distinct from
188e14bb32Jeff Bonwick *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
189e14bb32Jeff Bonwick *	blocks in zio_done() while another i/o that holds either
190e14bb32Jeff Bonwick *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
191e14bb32Jeff Bonwick *
192e14bb32Jeff Bonwick * SCL_VDEV
193e14bb32Jeff Bonwick *	Held as reader to prevent changes to the vdev tree during trivial
194b24ab67Jeff Bonwick *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
195e14bb32Jeff Bonwick *	other locks, and lower than all of them, to ensure that it's safe
196e14bb32Jeff Bonwick *	to acquire regardless of caller context.
197e14bb32Jeff Bonwick *
198e14bb32Jeff Bonwick * In addition, the following rules apply:
199e14bb32Jeff Bonwick *
200e14bb32Jeff Bonwick * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
201e14bb32Jeff Bonwick *	The lock ordering is SCL_CONFIG > spa_props_lock.
202e14bb32Jeff Bonwick *
203e14bb32Jeff Bonwick * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
204e14bb32Jeff Bonwick *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
205e14bb32Jeff Bonwick *	or zio_write_phys() -- the caller must ensure that the config cannot
206e14bb32Jeff Bonwick *	cannot change in the interim, and that the vdev cannot be reopened.
207e14bb32Jeff Bonwick *	SCL_STATE as reader suffices for both.
208fa9e406ahrens *
209ea8dc4beschrock * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
210fa9e406ahrens *
21144cd46cbillm *	spa_vdev_enter()	Acquire the namespace lock and the config lock
212ea8dc4beschrock *				for writing.
213fa9e406ahrens *
21444cd46cbillm *	spa_vdev_exit()		Release the config lock, wait for all I/O
21544cd46cbillm *				to complete, sync the updated configs to the
216ea8dc4beschrock *				cache, and release the namespace lock.
217fa9e406ahrens *
218e14bb32Jeff Bonwick * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
219e14bb32Jeff Bonwick * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
220e14bb32Jeff Bonwick * locking is, always, based on spa_namespace_lock and spa_config_lock[].
221e14bb32Jeff Bonwick *
222ad135b5Christopher Siden * spa_rename() is also implemented within this file since it requires
223e14bb32Jeff Bonwick * manipulation of the namespace.
224fa9e406ahrens */
225fa9e406ahrens
226fa9e406ahrensstatic avl_tree_t spa_namespace_avl;
227fa9e406ahrenskmutex_t spa_namespace_lock;
228fa9e406ahrensstatic kcondvar_t spa_namespace_cv;
2290373e76bonwickstatic int spa_active_count;
230416e0cdekint spa_max_replication_override = SPA_DVAS_PER_BP;
231fa9e406ahrens
23299653d4eschrockstatic kmutex_t spa_spare_lock;
23339c2341eschrockstatic avl_tree_t spa_spare_avl;
234fa94a07brendanstatic kmutex_t spa_l2cache_lock;
235fa94a07brendanstatic avl_tree_t spa_l2cache_avl;
23699653d4eschrock
237fa9e406ahrenskmem_cache_t *spa_buffer_pool;
2388ad4d6dJeff Bonwickint spa_mode_global;
239fa9e406ahrens
240fa9e406ahrens#ifdef ZFS_DEBUG
2413b2aab1Matthew Ahrens/* Everything except dprintf and spa is on by default in debug builds */
2423b2aab1Matthew Ahrensint zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
243fa9e406ahrens#else
244fa9e406ahrensint zfs_flags = 0;
245fa9e406ahrens#endif
246fa9e406ahrens
2470125049ahrens/*
2480125049ahrens * zfs_recover can be set to nonzero to attempt to recover from
2490125049ahrens * otherwise-fatal errors, typically caused by on-disk corruption.  When
2500125049ahrens * set, calls to zfs_panic_recover() will turn into warning messages.
2518b36997Matthew Ahrens * This should only be used as a last resort, as it typically results
2528b36997Matthew Ahrens * in leaked space, or worse.
2530125049ahrens */
2547fd05acMatthew Ahrensboolean_t zfs_recover = B_FALSE;
2557fd05acMatthew Ahrens
2567fd05acMatthew Ahrens/*
2577fd05acMatthew Ahrens * If destroy encounters an EIO while reading metadata (e.g. indirect
2587fd05acMatthew Ahrens * blocks), space referenced by the missing metadata can not be freed.
2597fd05acMatthew Ahrens * Normally this causes the background destroy to become "stalled", as
2607fd05acMatthew Ahrens * it is unable to make forward progress.  While in this stalled state,
2617fd05acMatthew Ahrens * all remaining space to free from the error-encountering filesystem is
2627fd05acMatthew Ahrens * "temporarily leaked".  Set this flag to cause it to ignore the EIO,
2637fd05acMatthew Ahrens * permanently leak the space from indirect blocks that can not be read,
2647fd05acMatthew Ahrens * and continue to free everything else that it can.
2657fd05acMatthew Ahrens *
2667fd05acMatthew Ahrens * The default, "stalling" behavior is useful if the storage partially
2677fd05acMatthew Ahrens * fails (i.e. some but not all i/os fail), and then later recovers.  In
2687fd05acMatthew Ahrens * this case, we will be able to continue pool operations while it is
2697fd05acMatthew Ahrens * partially failed, and when it recovers, we can continue to free the
2707fd05acMatthew Ahrens * space, with no leaks.  However, note that this case is actually
2717fd05acMatthew Ahrens * fairly rare.
2727fd05acMatthew Ahrens *
2737fd05acMatthew Ahrens * Typically pools either (a) fail completely (but perhaps temporarily,
2747fd05acMatthew Ahrens * e.g. a top-level vdev going offline), or (b) have localized,
2757fd05acMatthew Ahrens * permanent errors (e.g. disk returns the wrong data due to bit flip or
2767fd05acMatthew Ahrens * firmware bug).  In case (a), this setting does not matter because the
2777fd05acMatthew Ahrens * pool will be suspended and the sync thread will not be able to make
2787fd05acMatthew Ahrens * forward progress regardless.  In case (b), because the error is
2797fd05acMatthew Ahrens * permanent, the best we can do is leak the minimum amount of space,
2807fd05acMatthew Ahrens * which is what setting this flag will do.  Therefore, it is reasonable
2817fd05acMatthew Ahrens * for this flag to normally be set, but we chose the more conservative
2827fd05acMatthew Ahrens * approach of not setting it, so that there is no possibility of
2837fd05acMatthew Ahrens * leaking space in the "partial temporary" failure case.
2847fd05acMatthew Ahrens */
2857fd05acMatthew Ahrensboolean_t zfs_free_leak_on_eio = B_FALSE;
2860125049ahrens
28769962b5Matthew Ahrens/*
28869962b5Matthew Ahrens * Expiration time in milliseconds. This value has two meanings. First it is
28969962b5Matthew Ahrens * used to determine when the spa_deadman() logic should fire. By default the
29069962b5Matthew Ahrens * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
29169962b5Matthew Ahrens * Secondly, the value determines if an I/O is considered "hung". Any I/O that
29269962b5Matthew Ahrens * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
29369962b5Matthew Ahrens * in a system panic.
29469962b5Matthew Ahrens */
29569962b5Matthew Ahrensuint64_t zfs_deadman_synctime_ms = 1000000ULL;
296283b846George.Wilson
297283b846George.Wilson/*
29869962b5Matthew Ahrens * Check time in milliseconds. This defines the frequency at which we check
29969962b5Matthew Ahrens * for hung I/O.
300283b846George.Wilson */
30169962b5Matthew Ahrensuint64_t zfs_deadman_checktime_ms = 5000ULL;
302283b846George.Wilson
303283b846George.Wilson/*
304283b846George.Wilson * Override the zfs deadman behavior via /etc/system. By default the
305283b846George.Wilson * deadman is enabled except on VMware and sparc deployments.
306283b846George.Wilson */
307283b846George.Wilsonint zfs_deadman_enabled = -1;
308283b846George.Wilson
30969962b5Matthew Ahrens/*
31069962b5Matthew Ahrens * The worst case is single-sector max-parity RAID-Z blocks, in which
31169962b5Matthew Ahrens * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
31269962b5Matthew Ahrens * times the size; so just assume that.  Add to this the fact that
31369962b5Matthew Ahrens * we can have up to 3 DVAs per bp, and one more factor of 2 because
31469962b5Matthew Ahrens * the block may be dittoed with up to 3 DVAs by ddt_sync().  All together,
31569962b5Matthew Ahrens * the worst case is:
31669962b5Matthew Ahrens *     (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
31769962b5Matthew Ahrens */
31869962b5Matthew Ahrensint spa_asize_inflation = 24;
319fa9e406ahrens
320fa9e406ahrens/*
3217d46dc6Matthew Ahrens * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
3227d46dc6Matthew Ahrens * the pool to be consumed.  This ensures that we don't run the pool
3237d46dc6Matthew Ahrens * completely out of space, due to unaccounted changes (e.g. to the MOS).
3247d46dc6Matthew Ahrens * It also limits the worst-case time to allocate space.  If we have
3257d46dc6Matthew Ahrens * less than this amount of free space, most ZPL operations (e.g. write,
3267d46dc6Matthew Ahrens * create) will return ENOSPC.
3277d46dc6Matthew Ahrens *
3287d46dc6Matthew Ahrens * Certain operations (e.g. file removal, most administrative actions) can
3297d46dc6Matthew Ahrens * use half the slop space.  They will only return ENOSPC if less than half
3307d46dc6Matthew Ahrens * the slop space is free.  Typically, once the pool has less than the slop
3317d46dc6Matthew Ahrens * space free, the user will use these operations to free up space in the pool.
3327d46dc6Matthew Ahrens * These are the operations that call dsl_pool_adjustedsize() with the netfree
3337d46dc6Matthew Ahrens * argument set to TRUE.
3347d46dc6Matthew Ahrens *
3357d46dc6Matthew Ahrens * A very restricted set of operations are always permitted, regardless of
3367d46dc6Matthew Ahrens * the amount of free space.  These are the operations that call
3377d46dc6Matthew Ahrens * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy".  If these
3387d46dc6Matthew Ahrens * operations result in a net increase in the amount of space used,
3397d46dc6Matthew Ahrens * it is possible to run the pool completely out of space, causing it to
3407d46dc6Matthew Ahrens * be permanently read-only.
3417d46dc6Matthew Ahrens *
3427d46dc6Matthew Ahrens * See also the comments in zfs_space_check_t.
3437d46dc6Matthew Ahrens */
3447d46dc6Matthew Ahrensint spa_slop_shift = 5;
3457d46dc6Matthew Ahrens
3467d46dc6Matthew Ahrens/*
347fa9e406ahrens * ==========================================================================
348e05725bbonwick * SPA config locking
349e05725bbonwick * ==========================================================================
350e05725bbonwick */
351e05725bbonwickstatic void
352e14bb32Jeff Bonwickspa_config_lock_init(spa_t *spa)
353e14bb32Jeff Bonwick{
354e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
355e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
356e14bb32Jeff Bonwick		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
357e14bb32Jeff Bonwick		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
3583b2aab1Matthew Ahrens		refcount_create_untracked(&scl->scl_count);
359e14bb32Jeff Bonwick		scl->scl_writer = NULL;
360e14bb32Jeff Bonwick		scl->scl_write_wanted = 0;
361e14bb32Jeff Bonwick	}
362e05725bbonwick}
363e05725bbonwick
364e05725bbonwickstatic void
365e14bb32Jeff Bonwickspa_config_lock_destroy(spa_t *spa)
366e14bb32Jeff Bonwick{
367e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
368e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
369e14bb32Jeff Bonwick		mutex_destroy(&scl->scl_lock);
370e14bb32Jeff Bonwick		cv_destroy(&scl->scl_cv);
371e14bb32Jeff Bonwick		refcount_destroy(&scl->scl_count);
372e14bb32Jeff Bonwick		ASSERT(scl->scl_writer == NULL);
373e14bb32Jeff Bonwick		ASSERT(scl->scl_write_wanted == 0);
374e14bb32Jeff Bonwick	}
375e14bb32Jeff Bonwick}
376e14bb32Jeff Bonwick
377e14bb32Jeff Bonwickint
378e14bb32Jeff Bonwickspa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
379e05725bbonwick{
380e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
381e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
382e14bb32Jeff Bonwick		if (!(locks & (1 << i)))
383e14bb32Jeff Bonwick			continue;
384e14bb32Jeff Bonwick		mutex_enter(&scl->scl_lock);
385e14bb32Jeff Bonwick		if (rw == RW_READER) {
386e14bb32Jeff Bonwick			if (scl->scl_writer || scl->scl_write_wanted) {
387e14bb32Jeff Bonwick				mutex_exit(&scl->scl_lock);
388e14bb32Jeff Bonwick				spa_config_exit(spa, locks ^ (1 << i), tag);
389e14bb32Jeff Bonwick				return (0);
390e14bb32Jeff Bonwick			}
391e14bb32Jeff Bonwick		} else {
392e14bb32Jeff Bonwick			ASSERT(scl->scl_writer != curthread);
393e14bb32Jeff Bonwick			if (!refcount_is_zero(&scl->scl_count)) {
394e14bb32Jeff Bonwick				mutex_exit(&scl->scl_lock);
395e14bb32Jeff Bonwick				spa_config_exit(spa, locks ^ (1 << i), tag);
396e14bb32Jeff Bonwick				return (0);
397e14bb32Jeff Bonwick			}
398e14bb32Jeff Bonwick			scl->scl_writer = curthread;
399e14bb32Jeff Bonwick		}
400e14bb32Jeff Bonwick		(void) refcount_add(&scl->scl_count, tag);
401e14bb32Jeff Bonwick		mutex_exit(&scl->scl_lock);
402e14bb32Jeff Bonwick	}
403e14bb32Jeff Bonwick	return (1);
404e05725bbonwick}
405e05725bbonwick
406e05725bbonwickvoid
407e14bb32Jeff Bonwickspa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
408e05725bbonwick{
409f64c0e3Eric Taylor	int wlocks_held = 0;
410f64c0e3Eric Taylor
4113b2aab1Matthew Ahrens	ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
4123b2aab1Matthew Ahrens
413e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
414e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
415f64c0e3Eric Taylor		if (scl->scl_writer == curthread)
416f64c0e3Eric Taylor			wlocks_held |= (1 << i);
417e14bb32Jeff Bonwick		if (!(locks & (1 << i)))
418e14bb32Jeff Bonwick			continue;
419e14bb32Jeff Bonwick		mutex_enter(&scl->scl_lock);
420e14bb32Jeff Bonwick		if (rw == RW_READER) {
421e14bb32Jeff Bonwick			while (scl->scl_writer || scl->scl_write_wanted) {
422e14bb32Jeff Bonwick				cv_wait(&scl->scl_cv, &scl->scl_lock);
423e14bb32Jeff Bonwick			}
424e14bb32Jeff Bonwick		} else {
425e14bb32Jeff Bonwick			ASSERT(scl->scl_writer != curthread);
426e14bb32Jeff Bonwick			while (!refcount_is_zero(&scl->scl_count)) {
427e14bb32Jeff Bonwick				scl->scl_write_wanted++;
428e14bb32Jeff Bonwick				cv_wait(&scl->scl_cv, &scl->scl_lock);
429e14bb32Jeff Bonwick				scl->scl_write_wanted--;
430e14bb32Jeff Bonwick			}
431e14bb32Jeff Bonwick			scl->scl_writer = curthread;
432e14bb32Jeff Bonwick		}
433e14bb32Jeff Bonwick		(void) refcount_add(&scl->scl_count, tag);
434e14bb32Jeff Bonwick		mutex_exit(&scl->scl_lock);
435e05725bbonwick	}
436f64c0e3Eric Taylor	ASSERT(wlocks_held <= locks);
437e05725bbonwick}
438e05725bbonwick
439e05725bbonwickvoid
440e14bb32Jeff Bonwickspa_config_exit(spa_t *spa, int locks, void *tag)
441e05725bbonwick{
442e14bb32Jeff Bonwick	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
443e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
444e14bb32Jeff Bonwick		if (!(locks & (1 << i)))
445e14bb32Jeff Bonwick			continue;
446e14bb32Jeff Bonwick		mutex_enter(&scl->scl_lock);
447e14bb32Jeff Bonwick		ASSERT(!refcount_is_zero(&scl->scl_count));
448e14bb32Jeff Bonwick		if (refcount_remove(&scl->scl_count, tag) == 0) {
449e14bb32Jeff Bonwick			ASSERT(scl->scl_writer == NULL ||
450e14bb32Jeff Bonwick			    scl->scl_writer == curthread);
451e14bb32Jeff Bonwick			scl->scl_writer = NULL;	/* OK in either case */
452e14bb32Jeff Bonwick			cv_broadcast(&scl->scl_cv);
453e14bb32Jeff Bonwick		}
454e14bb32Jeff Bonwick		mutex_exit(&scl->scl_lock);
455e05725bbonwick	}
456e05725bbonwick}
457e05725bbonwick
458e14bb32Jeff Bonwickint
459e14bb32Jeff Bonwickspa_config_held(spa_t *spa, int locks, krw_t rw)
460e05725bbonwick{
461e14bb32Jeff Bonwick	int locks_held = 0;
462e05725bbonwick
463e14bb32Jeff Bonwick	for (int i = 0; i < SCL_LOCKS; i++) {
464e14bb32Jeff Bonwick		spa_config_lock_t *scl = &spa->spa_config_lock[i];
465e14bb32Jeff Bonwick		if (!(locks & (1 << i)))
466e14bb32Jeff Bonwick			continue;
467e14bb32Jeff Bonwick		if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
468e14bb32Jeff Bonwick		    (rw == RW_WRITER && scl->scl_writer == curthread))
469e14bb32Jeff Bonwick			locks_held |= 1 << i;
470e14bb32Jeff Bonwick	}
471e14bb32Jeff Bonwick
472e14bb32Jeff Bonwick	return (locks_held);
473e05725bbonwick}
474e05725bbonwick
475e05725bbonwick/*
476e05725bbonwick * ==========================================================================
477fa9e406ahrens * SPA namespace functions
478fa9e406ahrens * ==========================================================================
479fa9e406ahrens */
480fa9e406ahrens
481fa9e406ahrens/*
482fa9e406ahrens * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
483fa9e406ahrens * Returns NULL if no matching spa_t is found.
484fa9e406ahrens */
485fa9e406ahrensspa_t *
486fa9e406ahrensspa_lookup(const char *name)
487fa9e406ahrens{
488e14bb32Jeff Bonwick	static spa_t search;	/* spa_t is large; don't allocate on stack */
489e14bb32Jeff Bonwick	spa_t *spa;
490fa9e406ahrens	avl_index_t where;
49140feaa9ahrens	char *cp;
492fa9e406ahrens
493fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
494fa9e406ahrens
4953b2aab1Matthew Ahrens	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
4963b2aab1Matthew Ahrens
49740feaa9ahrens	/*
49840feaa9ahrens	 * If it's a full dataset name, figure out the pool name and
49940feaa9ahrens	 * just use that.
50040feaa9ahrens	 */
50178f1710Matthew Ahrens	cp = strpbrk(search.spa_name, "/@#");
5023b2aab1Matthew Ahrens	if (cp != NULL)
50340feaa9ahrens		*cp = '\0';
50440feaa9ahrens
505fa9e406ahrens	spa = avl_find(&spa_namespace_avl, &search, &where);
506fa9e406ahrens
507fa9e406ahrens	return (spa);
508fa9e406ahrens}
509fa9e406ahrens
510fa9e406ahrens/*
511283b846George.Wilson * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
512283b846George.Wilson * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
513283b846George.Wilson * looking for potentially hung I/Os.
514283b846George.Wilson */
515283b846George.Wilsonvoid
516283b846George.Wilsonspa_deadman(void *arg)
517283b846George.Wilson{
518283b846George.Wilson	spa_t *spa = arg;
519283b846George.Wilson
5200713e23George Wilson	/*
5210713e23George Wilson	 * Disable the deadman timer if the pool is suspended.
5220713e23George Wilson	 */
5230713e23George Wilson	if (spa_suspended(spa)) {
5240713e23George Wilson		VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
5250713e23George Wilson		return;
5260713e23George Wilson	}
5270713e23George Wilson
528283b846George.Wilson	zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
529283b846George.Wilson	    (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
530283b846George.Wilson	    ++spa->spa_deadman_calls);
531283b846George.Wilson	if (zfs_deadman_enabled)
532283b846George.Wilson		vdev_deadman(spa->spa_root_vdev);
533283b846George.Wilson}
534283b846George.Wilson
535283b846George.Wilson/*
536fa9e406ahrens * Create an uninitialized spa_t with the given name.  Requires
537fa9e406ahrens * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
538fa9e406ahrens * exist by calling spa_lookup() first.
539fa9e406ahrens */
540fa9e406ahrensspa_t *
541468c413Tim Haleyspa_add(const char *name, nvlist_t *config, const char *altroot)
542fa9e406ahrens{
543fa9e406ahrens	spa_t *spa;
544c5904d1eschrock	spa_config_dirent_t *dp;
545283b846George.Wilson	cyc_handler_t hdlr;
546283b846George.Wilson	cyc_time_t when;
547fa9e406ahrens
548fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
549fa9e406ahrens
550fa9e406ahrens	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
551fa9e406ahrens
552c25056dgw	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
553c25056dgw	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
55435a5a35Jonathan Adams	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
555bc9014eJustin Gibbs	mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
556c25056dgw	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
55735a5a35Jonathan Adams	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
558c25056dgw	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
55935a5a35Jonathan Adams	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
560a152156Jeff Bonwick	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
561a152156Jeff Bonwick	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
562c3a6601Matthew Ahrens	mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
563c25056dgw
564c25056dgw	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
565bc9014eJustin Gibbs	cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
56635a5a35Jonathan Adams	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
567c25056dgw	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
568e14bb32Jeff Bonwick	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
569c25056dgw
570b24ab67Jeff Bonwick	for (int t = 0; t < TXG_SIZE; t++)
571cde58dbMatthew Ahrens		bplist_create(&spa->spa_free_bplist[t]);
572b24ab67Jeff Bonwick
573e14bb32Jeff Bonwick	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
574fa9e406ahrens	spa->spa_state = POOL_STATE_UNINITIALIZED;
575fa9e406ahrens	spa->spa_freeze_txg = UINT64_MAX;
5760373e76bonwick	spa->spa_final_txg = UINT64_MAX;
577468c413Tim Haley	spa->spa_load_max_txg = UINT64_MAX;
57835a5a35Jonathan Adams	spa->spa_proc = &p0;
57935a5a35Jonathan Adams	spa->spa_proc_state = SPA_PROC_NONE;
580fa9e406ahrens
581283b846George.Wilson	hdlr.cyh_func = spa_deadman;
582283b846George.Wilson	hdlr.cyh_arg = spa;
583283b846George.Wilson	hdlr.cyh_level = CY_LOW_LEVEL;
584283b846George.Wilson
58569962b5Matthew Ahrens	spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
586283b846George.Wilson
587283b846George.Wilson	/*
588283b846George.Wilson	 * This determines how often we need to check for hung I/Os after
589283b846George.Wilson	 * the cyclic has already fired. Since checking for hung I/Os is
590283b846George.Wilson	 * an expensive operation we don't want to check too frequently.
59169962b5Matthew Ahrens	 * Instead wait for 5 seconds before checking again.
592283b846George.Wilson	 */
59369962b5Matthew Ahrens	when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
594283b846George.Wilson	when.cyt_when = CY_INFINITY;
595283b846George.Wilson	mutex_enter(&cpu_lock);
596283b846George.Wilson	spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
597283b846George.Wilson	mutex_exit(&cpu_lock);
598283b846George.Wilson
599fa9e406ahrens	refcount_create(&spa->spa_refcount);
600e14bb32Jeff Bonwick	spa_config_lock_init(spa);
601fa9e406ahrens
602fa9e406ahrens	avl_add(&spa_namespace_avl, spa);
603fa9e406ahrens
6040373e76bonwick	/*
6050373e76bonwick	 * Set the alternate root, if there is one.
6060373e76bonwick	 */
6070373e76bonwick	if (altroot) {
6080373e76bonwick		spa->spa_root = spa_strdup(altroot);
6090373e76bonwick		spa_active_count++;
6100373e76bonwick	}
6110373e76bonwick
612c5904d1eschrock	/*
613c5904d1eschrock	 * Every pool starts with the default cachefile
614c5904d1eschrock	 */
615c5904d1eschrock	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
616c5904d1eschrock	    offsetof(spa_config_dirent_t, scd_link));
617c5904d1eschrock
618c5904d1eschrock	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
619ef912c8Tim Haley	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
620c5904d1eschrock	list_insert_head(&spa->spa_config_list, dp);
621c5904d1eschrock
6224b964adGeorge Wilson	VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
6234b964adGeorge Wilson	    KM_SLEEP) == 0);
6244b964adGeorge Wilson
625ad135b5Christopher Siden	if (config != NULL) {
626ad135b5Christopher Siden		nvlist_t *features;
627ad135b5Christopher Siden
628ad135b5Christopher Siden		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
629ad135b5Christopher Siden		    &features) == 0) {
630ad135b5Christopher Siden			VERIFY(nvlist_dup(features, &spa->spa_label_features,
631ad135b5Christopher Siden			    0) == 0);
632ad135b5Christopher Siden		}
633ad135b5Christopher Siden
634468c413Tim Haley		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
635ad135b5Christopher Siden	}
636ad135b5Christopher Siden
637ad135b5Christopher Siden	if (spa->spa_label_features == NULL) {
638ad135b5Christopher Siden		VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
639ad135b5Christopher Siden		    KM_SLEEP) == 0);
640ad135b5Christopher Siden	}
641468c413Tim Haley
642c3a6601Matthew Ahrens	spa->spa_iokstat = kstat_create("zfs", 0, name,
643c3a6601Matthew Ahrens	    "disk", KSTAT_TYPE_IO, 1, 0);
644c3a6601Matthew Ahrens	if (spa->spa_iokstat) {
645c3a6601Matthew Ahrens		spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
646c3a6601Matthew Ahrens		kstat_install(spa->spa_iokstat);
647c3a6601Matthew Ahrens	}
648c3a6601Matthew Ahrens
6493b2aab1Matthew Ahrens	spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
6503b2aab1Matthew Ahrens
65143466aaMax Grossman	/*
65243466aaMax Grossman	 * As a pool is being created, treat all features as disabled by
65343466aaMax Grossman	 * setting SPA_FEATURE_DISABLED for all entries in the feature
65443466aaMax Grossman	 * refcount cache.
65543466aaMax Grossman	 */
65643466aaMax Grossman	for (int i = 0; i < SPA_FEATURES; i++) {
65743466aaMax Grossman		spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
65843466aaMax Grossman	}
65943466aaMax Grossman
660fa9e406ahrens	return (spa);
661fa9e406ahrens}
662fa9e406ahrens
663fa9e406ahrens/*
664fa9e406ahrens * Removes a spa_t from the namespace, freeing up any memory used.  Requires
665fa9e406ahrens * spa_namespace_lock.  This is called only after the spa_t has been closed and
666fa9e406ahrens * deactivated.
667fa9e406ahrens */
668fa9e406ahrensvoid
669fa9e406ahrensspa_remove(spa_t *spa)
670fa9e406ahrens{
671c5904d1eschrock	spa_config_dirent_t *dp;
672c5904d1eschrock
673fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
674fa9e406ahrens	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
675bc9014eJustin Gibbs	ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
676fa9e406ahrens
6771195e68Mark J Musante	nvlist_free(spa->spa_config_splitting);
6781195e68Mark J Musante
679fa9e406ahrens	avl_remove(&spa_namespace_avl, spa);
680fa9e406ahrens	cv_broadcast(&spa_namespace_cv);
681fa9e406ahrens
6820373e76bonwick	if (spa->spa_root) {
683fa9e406ahrens		spa_strfree(spa->spa_root);
6840373e76bonwick		spa_active_count--;
6850373e76bonwick	}
686fa9e406ahrens
687c5904d1eschrock	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
688c5904d1eschrock		list_remove(&spa->spa_config_list, dp);
689c5904d1eschrock		if (dp->scd_path != NULL)
690c5904d1eschrock			spa_strfree(dp->scd_path);
691c5904d1eschrock		kmem_free(dp, sizeof (spa_config_dirent_t));
692c5904d1eschrock	}
693c5904d1eschrock
694c5904d1eschrock	list_destroy(&spa->spa_config_list);
6952f8aaabeschrock
696ad135b5Christopher Siden	nvlist_free(spa->spa_label_features);
6974b964adGeorge Wilson	nvlist_free(spa->spa_load_info);
698fa9e406ahrens	spa_config_set(spa, NULL);
699fa9e406ahrens
700283b846George.Wilson	mutex_enter(&cpu_lock);
701283b846George.Wilson	if (spa->spa_deadman_cycid != CYCLIC_NONE)
702283b846George.Wilson		cyclic_remove(spa->spa_deadman_cycid);
703283b846George.Wilson	mutex_exit(&cpu_lock);
704283b846George.Wilson	spa->spa_deadman_cycid = CYCLIC_NONE;
705283b846George.Wilson
706fa9e406ahrens	refcount_destroy(&spa->spa_refcount);
70791ebeefahrens
708e14bb32Jeff Bonwick	spa_config_lock_destroy(spa);
709fa9e406ahrens
710c3a6601Matthew Ahrens	kstat_delete(spa->spa_iokstat);
711c3a6601Matthew Ahrens	spa->spa_iokstat = NULL;
712c3a6601Matthew Ahrens
713b24ab67Jeff Bonwick	for (int t = 0; t < TXG_SIZE; t++)
714cde58dbMatthew Ahrens		bplist_destroy(&spa->spa_free_bplist[t]);
715b24ab67Jeff Bonwick
716c25056dgw	cv_destroy(&spa->spa_async_cv);
717bc9014eJustin Gibbs	cv_destroy(&spa->spa_evicting_os_cv);
71835a5a35Jonathan Adams	cv_destroy(&spa->spa_proc_cv);
719c25056dgw	cv_destroy(&spa->spa_scrub_io_cv);
720e14bb32Jeff Bonwick	cv_destroy(&spa->spa_suspend_cv);
721c25056dgw
7225ad8204nd	mutex_destroy(&spa->spa_async_lock);
723c25056dgw	mutex_destroy(&spa->spa_errlist_lock);
72435a5a35Jonathan Adams	mutex_destroy(&spa->spa_errlog_lock);
725bc9014eJustin Gibbs	mutex_destroy(&spa->spa_evicting_os_lock);
72606eeb2aek	mutex_destroy(&spa->spa_history_lock);
72735a5a35Jonathan Adams	mutex_destroy(&spa->spa_proc_lock);
728b1b8ab3lling	mutex_destroy(&spa->spa_props_lock);
72935a5a35Jonathan Adams	mutex_destroy(&spa->spa_scrub_lock);
730e14bb32Jeff Bonwick	mutex_destroy(&spa->spa_suspend_lock);
731a152156Jeff Bonwick	mutex_destroy(&spa->spa_vdev_top_lock);
732c3a6601Matthew Ahrens	mutex_destroy(&spa->spa_iokstat_lock);
7335ad8204nd
734fa9e406ahrens	kmem_free(spa, sizeof (spa_t));
735fa9e406ahrens}
736fa9e406ahrens
737fa9e406ahrens/*
738fa9e406ahrens * Given a pool, return the next pool in the namespace, or NULL if there is
739fa9e406ahrens * none.  If 'prev' is NULL, return the first pool.
740fa9e406ahrens */
741fa9e406ahrensspa_t *
742fa9e406ahrensspa_next(spa_t *prev)
743fa9e406ahrens{
744fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
745fa9e406ahrens
746fa9e406ahrens	if (prev)
747fa9e406ahrens		return (AVL_NEXT(&spa_namespace_avl, prev));
748fa9e406ahrens	else
749fa9e406ahrens		return (avl_first(&spa_namespace_avl));
750fa9e406ahrens}
751fa9e406ahrens
752fa9e406ahrens/*
753fa9e406ahrens * ==========================================================================
754fa9e406ahrens * SPA refcount functions
755fa9e406ahrens * ==========================================================================
756fa9e406ahrens */
757fa9e406ahrens
758fa9e406ahrens/*
759fa9e406ahrens * Add a reference to the given spa_t.  Must have at least one reference, or
760fa9e406ahrens * have the namespace lock held.
761fa9e406ahrens */
762fa9e406ahrensvoid
763fa9e406ahrensspa_open_ref(spa_t *spa, void *tag)
764fa9e406ahrens{
765088f389ahrens	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
766fa9e406ahrens	    MUTEX_HELD(&spa_namespace_lock));
767fa9e406ahrens	(void) refcount_add(&spa->spa_refcount, tag);
768fa9e406ahrens}
769fa9e406ahrens
770fa9e406ahrens/*
771fa9e406ahrens * Remove a reference to the given spa_t.  Must have at least one reference, or
772fa9e406ahrens * have the namespace lock held.
773fa9e406ahrens */
774fa9e406ahrensvoid
775fa9e406ahrensspa_close(spa_t *spa, void *tag)
776fa9e406ahrens{
777088f389ahrens	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
778fa9e406ahrens	    MUTEX_HELD(&spa_namespace_lock));
779fa9e406ahrens	(void) refcount_remove(&spa->spa_refcount, tag);
780fa9e406ahrens}
781fa9e406ahrens
782fa9e406ahrens/*
783bc9014eJustin Gibbs * Remove a reference to the given spa_t held by a dsl dir that is
784bc9014eJustin Gibbs * being asynchronously released.  Async releases occur from a taskq
785bc9014eJustin Gibbs * performing eviction of dsl datasets and dirs.  The namespace lock
786bc9014eJustin Gibbs * isn't held and the hold by the object being evicted may contribute to
787bc9014eJustin Gibbs * spa_minref (e.g. dataset or directory released during pool export),
788bc9014eJustin Gibbs * so the asserts in spa_close() do not apply.
789bc9014eJustin Gibbs */
790bc9014eJustin Gibbsvoid
791bc9014eJustin Gibbsspa_async_close(spa_t *spa, void *tag)
792bc9014eJustin Gibbs{
793bc9014eJustin Gibbs	(void) refcount_remove(&spa->spa_refcount, tag);
794bc9014eJustin Gibbs}
795bc9014eJustin Gibbs
796bc9014eJustin Gibbs/*
797fa9e406ahrens * Check to see if the spa refcount is zero.  Must be called with
798088f389ahrens * spa_namespace_lock held.  We really compare against spa_minref, which is the
799fa9e406ahrens * number of references acquired when opening a pool
800fa9e406ahrens */
801fa9e406ahrensboolean_t
802fa9e406ahrensspa_refcount_zero(spa_t *spa)
803fa9e406ahrens{
804fa9e406ahrens	ASSERT(MUTEX_HELD(&spa_namespace_lock));
805fa9e406ahrens
806088f389ahrens	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
807fa9e406ahrens}
808fa9e406ahrens
809fa9e406ahrens/*
810fa9e406ahrens * ==========================================================================
811fa94a07brendan * SPA spare and l2cache tracking
81299653d4eschrock * ==========================================================================
81399653d4eschrock */
81499653d4eschrock
81599653d4eschrock/*
816fa94a07brendan * Hot spares and cache devices are tracked using the same code below,
817fa94a07brendan * for 'auxiliary' devices.
818fa94a07brendan */
819fa94a07brendan
820fa94a07brendantypedef struct spa_aux {
821fa94a07brendan	uint64_t	aux_guid;
822fa94a07brendan	uint64_t	aux_pool;
823fa94a07brendan	avl_node_t	aux_avl;
824fa94a07brendan	int		aux_count;
825fa94a07brendan} spa_aux_t;
826fa94a07brendan
827fa94a07brendanstatic int
828fa94a07brendanspa_aux_compare(const void *a, const void *b)
829fa94a07brendan{
830fa94a07brendan	const spa_aux_t *sa = a;
831fa94a07brendan	const spa_aux_t *sb = b;
832fa94a07brendan
833fa94a07brendan	if (sa->aux_guid < sb->aux_guid)
834fa94a07brendan		return (-1);
835fa94a07brendan	else if (sa->aux_guid > sb->aux_guid)
836fa94a07brendan		return (1);
837fa94a07brendan	else
838fa94a07brendan		return (0);
839fa94a07brendan}
840fa94a07brendan
841fa94a07brendanvoid
842fa94a07brendanspa_aux_add(vdev_t *vd, avl_tree_t *avl)
843fa94a07brendan{
844fa94a07brendan	avl_index_t where;
845fa94a07brendan	spa_aux_t search;
846fa94a07brendan	spa_aux_t *aux;
847fa94a07brendan
848fa94a07brendan	search.aux_guid = vd->vdev_guid;
849fa94a07brendan	if ((aux = avl_find(avl, &search, &where)) != NULL) {
850fa94a07brendan		aux->aux_count++;
851fa94a07brendan	} else {
852fa94a07brendan		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
853fa94a07brendan		aux->aux_guid = vd->vdev_guid;
854fa94a07brendan		aux->aux_count = 1;
855fa94a07brendan		avl_insert(avl, aux, where);
856fa94a07brendan	}
857fa94a07brendan}
858fa94a07brendan
859fa94a07brendanvoid
860fa94a07brendanspa_aux_remove(vdev_t *vd, avl_tree_t *avl)
861fa94a07brendan{
862fa94a07brendan	spa_aux_t search;
863fa94a07brendan	spa_aux_t *aux;
864fa94a07brendan	avl_index_t where;
865fa94a07brendan
866fa94a07brendan	search.aux_guid = vd->vdev_guid;
867fa94a07brendan	aux = avl_find(avl, &search, &where);
868fa94a07brendan
869fa94a07brendan	ASSERT(aux != NULL);
870fa94a07brendan
871fa94a07brendan	if (--aux->aux_count == 0) {
872fa94a07brendan		avl_remove(avl, aux);
873fa94a07brendan		kmem_free(aux, sizeof (spa_aux_t));
874fa94a07brendan	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
875fa94a07brendan		aux->aux_pool = 0ULL;
876fa94a07brendan	}
877fa94a07brendan}
878fa94a07brendan
879fa94a07brendanboolean_t
88089a89ebllingspa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
881fa94a07brendan{
882fa94a07brendan	spa_aux_t search, *found;
883fa94a07brendan
884fa94a07brendan	search.aux_guid = guid;
88589a89eblling	found = avl_find(avl, &search, NULL);
886fa94a07brendan
887fa94a07brendan	if (pool) {
888fa94a07brendan		if (found)
889fa94a07brendan			*pool = found->aux_pool;
890fa94a07brendan		else
891fa94a07brendan			*pool = 0ULL;
892fa94a07brendan	}
893fa94a07brendan
89489a89eblling	if (refcnt) {
89589a89eblling		if (found)
89689a89eblling			*refcnt = found->aux_count;
89789a89eblling		else
89889a89eblling			*refcnt = 0;
89989a89eblling	}
90089a89eblling
901fa94a07brendan	return (found != NULL);
902fa94a07brendan}
903fa94a07brendan
904fa94a07brendanvoid
905fa94a07brendanspa_aux_activate(vdev_t *vd, avl_tree_t *avl)
906fa94a07brendan{
907fa94a07brendan	spa_aux_t search, *found;
908fa94a07brendan	avl_index_t where;
909fa94a07brendan
910fa94a07brendan	search.aux_guid = vd->vdev_guid;
911fa94a07brendan	found = avl_find(avl, &search, &where);
912fa94a07brendan	ASSERT(found != NULL);
913fa94a07brendan	ASSERT(found->aux_pool == 0ULL);
914fa94a07brendan
915fa94a07brendan	found->aux_pool = spa_guid(vd->vdev_spa);
916fa94a07brendan}
917fa94a07brendan
918fa94a07brendan/*
91939c2341eschrock * Spares are tracked globally due to the following constraints:
92039c2341eschrock *
92139c2341eschrock *