spa_misc.c revision 5dabedee088a5b1080fff5e8b615fff203828fad
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#pragma ident	"%Z%%M%	%I%	%E% SMI"
27
28#include <sys/zfs_context.h>
29#include <sys/spa_impl.h>
30#include <sys/zio.h>
31#include <sys/zio_checksum.h>
32#include <sys/zio_compress.h>
33#include <sys/dmu.h>
34#include <sys/dmu_tx.h>
35#include <sys/zap.h>
36#include <sys/zil.h>
37#include <sys/vdev_impl.h>
38#include <sys/metaslab.h>
39#include <sys/uberblock_impl.h>
40#include <sys/txg.h>
41#include <sys/avl.h>
42#include <sys/unique.h>
43#include <sys/dsl_pool.h>
44#include <sys/dsl_dir.h>
45#include <sys/dsl_prop.h>
46#include <sys/fs/zfs.h>
47
48/*
49 * SPA locking
50 *
51 * There are four basic locks for managing spa_t structures:
52 *
53 * spa_namespace_lock (global mutex)
54 *
55 * 	This lock must be acquired to do any of the following:
56 *
57 * 		- Lookup a spa_t by name
58 * 		- Add or remove a spa_t from the namespace
59 * 		- Increase spa_refcount from non-zero
60 * 		- Check if spa_refcount is zero
61 * 		- Rename a spa_t
62 *		- add/remove/attach/detach devices
63 * 		- Held for the duration of create/destroy/import/export
64 *
65 * 	It does not need to handle recursion.  A create or destroy may
66 * 	reference objects (files or zvols) in other pools, but by
67 * 	definition they must have an existing reference, and will never need
68 * 	to lookup a spa_t by name.
69 *
70 * spa_refcount (per-spa refcount_t protected by mutex)
71 *
72 * 	This reference count keep track of any active users of the spa_t.  The
73 * 	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
74 * 	the refcount is never really 'zero' - opening a pool implicitly keeps
75 * 	some references in the DMU.  Internally we check against SPA_MINREF, but
76 * 	present the image of a zero/non-zero value to consumers.
77 *
78 * spa_config_lock (per-spa crazy rwlock)
79 *
80 * 	This SPA special is a recursive rwlock, capable of being acquired from
81 * 	asynchronous threads.  It has protects the spa_t from config changes,
82 * 	and must be held in the following circumstances:
83 *
84 * 		- RW_READER to perform I/O to the spa
85 * 		- RW_WRITER to change the vdev config
86 *
87 * spa_config_cache_lock (per-spa mutex)
88 *
89 * 	This mutex prevents the spa_config nvlist from being updated.  No
90 *      other locks are required to obtain this lock, although implicitly you
91 *      must have the namespace lock or non-zero refcount to have any kind
92 *      of spa_t pointer at all.
93 *
94 * The locking order is fairly straightforward:
95 *
96 * 		spa_namespace_lock	->	spa_refcount
97 *
98 * 	The namespace lock must be acquired to increase the refcount from 0
99 * 	or to check if it is zero.
100 *
101 * 		spa_refcount 		->	spa_config_lock
102 *
103 * 	There must be at least one valid reference on the spa_t to acquire
104 * 	the config lock.
105 *
106 * 		spa_namespace_lock	->	spa_config_lock
107 *
108 * 	The namespace lock must always be taken before the config lock.
109 *
110 *
111 * The spa_namespace_lock and spa_config_cache_lock can be acquired directly and
112 * are globally visible.
113 *
114 * The namespace is manipulated using the following functions, all which require
115 * the spa_namespace_lock to be held.
116 *
117 * 	spa_lookup()		Lookup a spa_t by name.
118 *
119 * 	spa_add()		Create a new spa_t in the namespace.
120 *
121 * 	spa_remove()		Remove a spa_t from the namespace.  This also
122 * 				frees up any memory associated with the spa_t.
123 *
124 * 	spa_next()		Returns the next spa_t in the system, or the
125 * 				first if NULL is passed.
126 *
127 * 	spa_evict_all()		Shutdown and remove all spa_t structures in
128 * 				the system.
129 *
130 *	spa_guid_exists()	Determine whether a pool/device guid exists.
131 *
132 * The spa_refcount is manipulated using the following functions:
133 *
134 * 	spa_open_ref()		Adds a reference to the given spa_t.  Must be
135 * 				called with spa_namespace_lock held if the
136 * 				refcount is currently zero.
137 *
138 * 	spa_close()		Remove a reference from the spa_t.  This will
139 * 				not free the spa_t or remove it from the
140 * 				namespace.  No locking is required.
141 *
142 * 	spa_refcount_zero()	Returns true if the refcount is currently
143 * 				zero.  Must be called with spa_namespace_lock
144 * 				held.
145 *
146 * The spa_config_lock is manipulated using the following functions:
147 *
148 * 	spa_config_enter()	Acquire the config lock as RW_READER or
149 * 				RW_WRITER.  At least one reference on the spa_t
150 * 				must exist.
151 *
152 * 	spa_config_exit()	Release the config lock.
153 *
154 * 	spa_config_held()	Returns true if the config lock is currently
155 * 				held in the given state.
156 *
157 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
158 *
159 * 	spa_vdev_enter()	Acquire the namespace lock and the config lock
160 *				for writing.
161 *
162 * 	spa_vdev_exit()		Release the config lock, wait for all I/O
163 * 				to complete, sync the updated configs to the
164 *				cache, and release the namespace lock.
165 *
166 * The spa_name() function also requires either the spa_namespace_lock
167 * or the spa_config_lock, as both are needed to do a rename.  spa_rename() is
168 * also implemented within this file since is requires manipulation of the
169 * namespace.
170 */
171
172static avl_tree_t spa_namespace_avl;
173kmutex_t spa_namespace_lock;
174static kcondvar_t spa_namespace_cv;
175
176kmem_cache_t *spa_buffer_pool;
177int spa_mode;
178
179#ifdef ZFS_DEBUG
180int zfs_flags = ~0;
181#else
182int zfs_flags = 0;
183#endif
184
185#define	SPA_MINREF	5	/* spa_refcnt for an open-but-idle pool */
186
187/*
188 * ==========================================================================
189 * SPA namespace functions
190 * ==========================================================================
191 */
192
193/*
194 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
195 * Returns NULL if no matching spa_t is found.
196 */
197spa_t *
198spa_lookup(const char *name)
199{
200	spa_t search, *spa;
201	avl_index_t where;
202
203	ASSERT(MUTEX_HELD(&spa_namespace_lock));
204
205	search.spa_name = (char *)name;
206	spa = avl_find(&spa_namespace_avl, &search, &where);
207
208	return (spa);
209}
210
211/*
212 * Create an uninitialized spa_t with the given name.  Requires
213 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
214 * exist by calling spa_lookup() first.
215 */
216spa_t *
217spa_add(const char *name)
218{
219	spa_t *spa;
220
221	ASSERT(MUTEX_HELD(&spa_namespace_lock));
222
223	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
224
225	spa->spa_name = spa_strdup(name);
226	spa->spa_state = POOL_STATE_UNINITIALIZED;
227	spa->spa_freeze_txg = UINT64_MAX;
228
229	refcount_create(&spa->spa_refcount);
230	refcount_create(&spa->spa_config_lock.scl_count);
231
232	avl_add(&spa_namespace_avl, spa);
233
234	return (spa);
235}
236
237/*
238 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
239 * spa_namespace_lock.  This is called only after the spa_t has been closed and
240 * deactivated.
241 */
242void
243spa_remove(spa_t *spa)
244{
245	ASSERT(MUTEX_HELD(&spa_namespace_lock));
246	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
247	ASSERT(spa->spa_scrub_thread == NULL);
248
249	avl_remove(&spa_namespace_avl, spa);
250	cv_broadcast(&spa_namespace_cv);
251
252	if (spa->spa_root)
253		spa_strfree(spa->spa_root);
254
255	if (spa->spa_name)
256		spa_strfree(spa->spa_name);
257
258	spa_config_set(spa, NULL);
259
260	refcount_destroy(&spa->spa_refcount);
261	refcount_destroy(&spa->spa_config_lock.scl_count);
262
263	kmem_free(spa, sizeof (spa_t));
264}
265
266/*
267 * Given a pool, return the next pool in the namespace, or NULL if there is
268 * none.  If 'prev' is NULL, return the first pool.
269 */
270spa_t *
271spa_next(spa_t *prev)
272{
273	ASSERT(MUTEX_HELD(&spa_namespace_lock));
274
275	if (prev)
276		return (AVL_NEXT(&spa_namespace_avl, prev));
277	else
278		return (avl_first(&spa_namespace_avl));
279}
280
281/*
282 * ==========================================================================
283 * SPA refcount functions
284 * ==========================================================================
285 */
286
287/*
288 * Add a reference to the given spa_t.  Must have at least one reference, or
289 * have the namespace lock held.
290 */
291void
292spa_open_ref(spa_t *spa, void *tag)
293{
294	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
295	    MUTEX_HELD(&spa_namespace_lock));
296
297	(void) refcount_add(&spa->spa_refcount, tag);
298}
299
300/*
301 * Remove a reference to the given spa_t.  Must have at least one reference, or
302 * have the namespace lock held.
303 */
304void
305spa_close(spa_t *spa, void *tag)
306{
307	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
308	    MUTEX_HELD(&spa_namespace_lock));
309
310	(void) refcount_remove(&spa->spa_refcount, tag);
311}
312
313/*
314 * Check to see if the spa refcount is zero.  Must be called with
315 * spa_namespace_lock held.  We really compare against SPA_MINREF, which is the
316 * number of references acquired when opening a pool
317 */
318boolean_t
319spa_refcount_zero(spa_t *spa)
320{
321	ASSERT(MUTEX_HELD(&spa_namespace_lock));
322
323	return (refcount_count(&spa->spa_refcount) == SPA_MINREF);
324}
325
326/*
327 * ==========================================================================
328 * SPA config locking
329 * ==========================================================================
330 */
331
332/*
333 * Acquire the config lock.  The config lock is a special rwlock that allows for
334 * recursive enters.  Because these enters come from the same thread as well as
335 * asynchronous threads working on behalf of the owner, we must unilaterally
336 * allow all reads access as long at least one reader is held (even if a write
337 * is requested).  This has the side effect of write starvation, but write locks
338 * are extremely rare, and a solution to this problem would be significantly
339 * more complex (if even possible).
340 *
341 * We would like to assert that the namespace lock isn't held, but this is a
342 * valid use during create.
343 */
344void
345spa_config_enter(spa_t *spa, krw_t rw, void *tag)
346{
347	spa_config_lock_t *scl = &spa->spa_config_lock;
348
349	mutex_enter(&scl->scl_lock);
350
351	if (scl->scl_writer != curthread) {
352		if (rw == RW_READER) {
353			while (scl->scl_writer != NULL)
354				cv_wait(&scl->scl_cv, &scl->scl_lock);
355		} else {
356			while (scl->scl_writer != NULL ||
357			    !refcount_is_zero(&scl->scl_count))
358				cv_wait(&scl->scl_cv, &scl->scl_lock);
359			scl->scl_writer = curthread;
360		}
361	}
362
363	(void) refcount_add(&scl->scl_count, tag);
364
365	mutex_exit(&scl->scl_lock);
366}
367
368/*
369 * Release the spa config lock, notifying any waiters in the process.
370 */
371void
372spa_config_exit(spa_t *spa, void *tag)
373{
374	spa_config_lock_t *scl = &spa->spa_config_lock;
375
376	mutex_enter(&scl->scl_lock);
377
378	ASSERT(!refcount_is_zero(&scl->scl_count));
379	if (refcount_remove(&scl->scl_count, tag) == 0) {
380		cv_broadcast(&scl->scl_cv);
381		scl->scl_writer = NULL;  /* OK in either case */
382	}
383
384	mutex_exit(&scl->scl_lock);
385}
386
387/*
388 * Returns true if the config lock is held in the given manner.
389 */
390boolean_t
391spa_config_held(spa_t *spa, krw_t rw)
392{
393	spa_config_lock_t *scl = &spa->spa_config_lock;
394	boolean_t held;
395
396	mutex_enter(&scl->scl_lock);
397	if (rw == RW_WRITER)
398		held = (scl->scl_writer == curthread);
399	else
400		held = !refcount_is_zero(&scl->scl_count);
401	mutex_exit(&scl->scl_lock);
402
403	return (held);
404}
405
406/*
407 * ==========================================================================
408 * SPA vdev locking
409 * ==========================================================================
410 */
411
412/*
413 * Lock the given spa_t for the purpose of adding or removing a vdev.
414 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
415 * It returns the next transaction group for the spa_t.
416 */
417uint64_t
418spa_vdev_enter(spa_t *spa)
419{
420	/*
421	 * Suspend scrub activity while we mess with the config.
422	 */
423	spa_scrub_suspend(spa);
424
425	if (spa->spa_root_vdev != NULL)		/* not spa_create() */
426		mutex_enter(&spa_namespace_lock);
427
428	spa_config_enter(spa, RW_WRITER, spa);
429
430	return (spa_last_synced_txg(spa) + 1);
431}
432
433/*
434 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
435 * locking of spa_vdev_enter(), we also want make sure the transactions have
436 * synced to disk, and then update the global configuration cache with the new
437 * information.
438 */
439int
440spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
441{
442	vdev_t *rvd = spa->spa_root_vdev;
443	uint64_t next_txg = spa_last_synced_txg(spa) + 1;
444	int config_changed = B_FALSE;
445
446	/*
447	 * Usually txg == next_txg, but spa_vdev_attach()
448	 * actually needs to wait for the open txg to sync.
449	 */
450	ASSERT(txg >= next_txg);
451
452	/*
453	 * Reassess the DTLs.
454	 */
455	if (rvd != NULL)
456		vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
457
458	/*
459	 * If the config changed, update the in-core cached copy
460	 * and notify the scrub thread that it must restart.
461	 */
462	if (error == 0 && !list_is_empty(&spa->spa_dirty_list)) {
463		config_changed = B_TRUE;
464		spa_config_set(spa, spa_config_generate(spa, rvd, next_txg, 0));
465		spa_scrub_restart(spa, next_txg);
466	}
467
468	spa_config_exit(spa, spa);
469
470	/*
471	 * Allow scrubbing to resume.
472	 */
473	spa_scrub_resume(spa);
474
475	if (vd == rvd)				/* spa_create() */
476		return (error);
477
478	/*
479	 * Note: this txg_wait_synced() is important because it ensures
480	 * that there won't be more than one config change per txg.
481	 * This allows us to use the txg as the generation number.
482	 */
483	if (error == 0)
484		txg_wait_synced(spa->spa_dsl_pool, txg);
485
486	if (vd != NULL) {
487		ASSERT(!vd->vdev_detached || vd->vdev_dtl.smo_object == 0);
488		vdev_free(vd);
489	}
490
491	/*
492	 * If the config changed, update the config cache.
493	 */
494	if (config_changed)
495		spa_config_sync();
496
497	mutex_exit(&spa_namespace_lock);
498
499	return (error);
500}
501
502/*
503 * ==========================================================================
504 * Miscellaneous functions
505 * ==========================================================================
506 */
507
508/*
509 * Rename a spa_t.
510 */
511int
512spa_rename(const char *name, const char *newname)
513{
514	spa_t *spa;
515	uint64_t txg;
516	int err;
517
518	/*
519	 * Lookup the spa_t and grab the config lock for writing.  We need to
520	 * actually open the pool so that we can sync out the necessary labels.
521	 * It's OK to call spa_open() with the namespace lock held because we
522	 * allow recursive calls for other reasons.
523	 */
524	mutex_enter(&spa_namespace_lock);
525	if ((err = spa_open(name, &spa, FTAG)) != 0) {
526		mutex_exit(&spa_namespace_lock);
527		return (err);
528	}
529
530	spa_config_enter(spa, RW_WRITER, FTAG);
531
532	avl_remove(&spa_namespace_avl, spa);
533	spa_strfree(spa->spa_name);
534	spa->spa_name = spa_strdup(newname);
535	avl_add(&spa_namespace_avl, spa);
536
537	/*
538	 * Sync all labels to disk with the new names by marking the root vdev
539	 * dirty and waiting for it to sync.  It will pick up the new pool name
540	 * during the sync.
541	 */
542	vdev_config_dirty(spa->spa_root_vdev);
543	txg = spa_last_synced_txg(spa) + 1;
544
545	spa_config_set(spa, spa_config_generate(spa, NULL, txg, 0));
546
547	spa_config_exit(spa, FTAG);
548
549	txg_wait_synced(spa->spa_dsl_pool, txg);
550
551	/*
552	 * Sync the updated config cache.
553	 */
554	spa_config_sync();
555
556	spa_close(spa, FTAG);
557
558	mutex_exit(&spa_namespace_lock);
559
560	return (0);
561}
562
563
564/*
565 * Determine whether a pool with given pool_guid exists.  If device_guid is
566 * non-zero, determine whether the pool exists *and* contains a device with the
567 * specified device_guid.
568 */
569boolean_t
570spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
571{
572	spa_t *spa;
573	avl_tree_t *t = &spa_namespace_avl;
574
575	ASSERT(MUTEX_HELD(&spa_namespace_lock));
576
577	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
578		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
579			continue;
580		if (spa->spa_root_vdev == NULL)
581			continue;
582		if (spa_guid(spa) == pool_guid && (device_guid == 0 ||
583		    vdev_lookup_by_guid(spa->spa_root_vdev, device_guid)))
584			break;
585	}
586
587	return (spa != NULL);
588}
589
590char *
591spa_strdup(const char *s)
592{
593	size_t len;
594	char *new;
595
596	len = strlen(s);
597	new = kmem_alloc(len + 1, KM_SLEEP);
598	bcopy(s, new, len);
599	new[len] = '\0';
600
601	return (new);
602}
603
604void
605spa_strfree(char *s)
606{
607	kmem_free(s, strlen(s) + 1);
608}
609
610uint64_t
611spa_get_random(uint64_t range)
612{
613	uint64_t r;
614
615	ASSERT(range != 0);
616
617	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
618
619	return (r % range);
620}
621
622void
623sprintf_blkptr(char *buf, int len, blkptr_t *bp)
624{
625	/* XXBP - Need to see if we want all DVAs or not */
626	dva_t *dva = BP_IDENTITY(bp);
627
628	if (bp == NULL) {
629		(void) snprintf(buf, len, "<NULL>");
630		return;
631	}
632
633	if (BP_IS_HOLE(bp)) {
634		(void) snprintf(buf, len, "<hole>");
635		return;
636	}
637
638	(void) snprintf(buf, len, "[L%llu %s] vdev=%llu offset=%llx "
639	    "size=%llxL/%llxP/%llxA %s %s %s %s "
640	    "birth=%llu fill=%llu cksum=%llx:%llx:%llx:%llx",
641	    (u_longlong_t)BP_GET_LEVEL(bp),
642	    dmu_ot[BP_GET_TYPE(bp)].ot_name,
643	    (u_longlong_t)DVA_GET_VDEV(dva),
644	    (u_longlong_t)DVA_GET_OFFSET(dva),
645	    (u_longlong_t)BP_GET_LSIZE(bp),
646	    (u_longlong_t)BP_GET_PSIZE(bp),
647	    (u_longlong_t)DVA_GET_ASIZE(dva),
648	    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name,
649	    zio_compress_table[BP_GET_COMPRESS(bp)].ci_name,
650	    BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE",
651	    DVA_GET_GANG(dva) == 0 ? "contiguous" : "gang",
652	    (u_longlong_t)bp->blk_birth,
653	    (u_longlong_t)bp->blk_fill,
654	    (u_longlong_t)bp->blk_cksum.zc_word[0],
655	    (u_longlong_t)bp->blk_cksum.zc_word[1],
656	    (u_longlong_t)bp->blk_cksum.zc_word[2],
657	    (u_longlong_t)bp->blk_cksum.zc_word[3]);
658}
659
660void
661spa_freeze(spa_t *spa)
662{
663	uint64_t freeze_txg = 0;
664
665	spa_config_enter(spa, RW_WRITER, FTAG);
666	if (spa->spa_freeze_txg == UINT64_MAX) {
667		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
668		spa->spa_freeze_txg = freeze_txg;
669	}
670	spa_config_exit(spa, FTAG);
671	if (freeze_txg != 0)
672		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
673}
674
675/*
676 * ==========================================================================
677 * Accessor functions
678 * ==========================================================================
679 */
680
681krwlock_t *
682spa_traverse_rwlock(spa_t *spa)
683{
684	return (&spa->spa_traverse_lock);
685}
686
687int
688spa_traverse_wanted(spa_t *spa)
689{
690	return (spa->spa_traverse_wanted);
691}
692
693dsl_pool_t *
694spa_get_dsl(spa_t *spa)
695{
696	return (spa->spa_dsl_pool);
697}
698
699blkptr_t *
700spa_get_rootblkptr(spa_t *spa)
701{
702	return (&spa->spa_ubsync.ub_rootbp);
703}
704
705void
706spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
707{
708	spa->spa_uberblock.ub_rootbp = *bp;
709}
710
711void
712spa_altroot(spa_t *spa, char *buf, size_t buflen)
713{
714	if (spa->spa_root == NULL)
715		buf[0] = '\0';
716	else
717		(void) strncpy(buf, spa->spa_root, buflen);
718}
719
720int
721spa_sync_pass(spa_t *spa)
722{
723	return (spa->spa_sync_pass);
724}
725
726char *
727spa_name(spa_t *spa)
728{
729	/*
730	 * Accessing the name requires holding either the namespace lock or the
731	 * config lock, both of which are required to do a rename.
732	 */
733	ASSERT(MUTEX_HELD(&spa_namespace_lock) ||
734	    spa_config_held(spa, RW_READER) || spa_config_held(spa, RW_WRITER));
735
736	return (spa->spa_name);
737}
738
739uint64_t
740spa_guid(spa_t *spa)
741{
742	return (spa->spa_root_vdev->vdev_guid);
743}
744
745uint64_t
746spa_last_synced_txg(spa_t *spa)
747{
748	return (spa->spa_ubsync.ub_txg);
749}
750
751uint64_t
752spa_first_txg(spa_t *spa)
753{
754	return (spa->spa_first_txg);
755}
756
757int
758spa_state(spa_t *spa)
759{
760	return (spa->spa_state);
761}
762
763uint64_t
764spa_freeze_txg(spa_t *spa)
765{
766	return (spa->spa_freeze_txg);
767}
768
769/*
770 * In the future, this may select among different metaslab classes
771 * depending on the zdp.  For now, there's no such distinction.
772 */
773metaslab_class_t *
774spa_metaslab_class_select(spa_t *spa)
775{
776	return (spa->spa_normal_class);
777}
778
779/*
780 * Return pool-wide allocated space.
781 */
782uint64_t
783spa_get_alloc(spa_t *spa)
784{
785	return (spa->spa_root_vdev->vdev_stat.vs_alloc);
786}
787
788/*
789 * Return pool-wide allocated space.
790 */
791uint64_t
792spa_get_space(spa_t *spa)
793{
794	return (spa->spa_root_vdev->vdev_stat.vs_space);
795}
796
797/* ARGSUSED */
798uint64_t
799spa_get_asize(spa_t *spa, uint64_t lsize)
800{
801	/*
802	 * For now, the worst case is 512-byte RAID-Z blocks, in which
803	 * case the space requirement is exactly 2x; so just assume that.
804	 */
805	return (lsize << 1);
806}
807
808/*
809 * ==========================================================================
810 * Initialization and Termination
811 * ==========================================================================
812 */
813
814static int
815spa_name_compare(const void *a1, const void *a2)
816{
817	const spa_t *s1 = a1;
818	const spa_t *s2 = a2;
819	int s;
820
821	s = strcmp(s1->spa_name, s2->spa_name);
822	if (s > 0)
823		return (1);
824	if (s < 0)
825		return (-1);
826	return (0);
827}
828
829void
830spa_init(int mode)
831{
832	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
833	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
834
835	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
836	    offsetof(spa_t, spa_avl));
837
838	spa_mode = mode;
839
840	refcount_init();
841	unique_init();
842	zio_init();
843	dmu_init();
844	zil_init();
845	spa_config_load();
846}
847
848void
849spa_fini(void)
850{
851	spa_evict_all();
852
853	zil_fini();
854	dmu_fini();
855	zio_fini();
856	refcount_fini();
857
858	avl_destroy(&spa_namespace_avl);
859
860	cv_destroy(&spa_namespace_cv);
861	mutex_destroy(&spa_namespace_lock);
862}
863