spa_misc.c revision d80c45e0f58fa434ba37259ea2e2b12e0380c19a
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;
175static int spa_active_count;
176static int spa_max_replication_override = SPA_DVAS_PER_BP;
177
178kmem_cache_t *spa_buffer_pool;
179int spa_mode;
180
181#ifdef ZFS_DEBUG
182int zfs_flags = ~0;
183#else
184int zfs_flags = 0;
185#endif
186
187#define	SPA_MINREF	5	/* spa_refcnt for an open-but-idle pool */
188
189/*
190 * ==========================================================================
191 * SPA namespace functions
192 * ==========================================================================
193 */
194
195/*
196 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
197 * Returns NULL if no matching spa_t is found.
198 */
199spa_t *
200spa_lookup(const char *name)
201{
202	spa_t search, *spa;
203	avl_index_t where;
204
205	ASSERT(MUTEX_HELD(&spa_namespace_lock));
206
207	search.spa_name = (char *)name;
208	spa = avl_find(&spa_namespace_avl, &search, &where);
209
210	return (spa);
211}
212
213/*
214 * Create an uninitialized spa_t with the given name.  Requires
215 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
216 * exist by calling spa_lookup() first.
217 */
218spa_t *
219spa_add(const char *name, const char *altroot)
220{
221	spa_t *spa;
222
223	ASSERT(MUTEX_HELD(&spa_namespace_lock));
224
225	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
226
227	spa->spa_name = spa_strdup(name);
228	spa->spa_state = POOL_STATE_UNINITIALIZED;
229	spa->spa_freeze_txg = UINT64_MAX;
230	spa->spa_final_txg = UINT64_MAX;
231
232	refcount_create(&spa->spa_refcount);
233	refcount_create(&spa->spa_config_lock.scl_count);
234
235	avl_add(&spa_namespace_avl, spa);
236
237	/*
238	 * Set the alternate root, if there is one.
239	 */
240	if (altroot) {
241		spa->spa_root = spa_strdup(altroot);
242		spa_active_count++;
243	}
244
245	return (spa);
246}
247
248/*
249 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
250 * spa_namespace_lock.  This is called only after the spa_t has been closed and
251 * deactivated.
252 */
253void
254spa_remove(spa_t *spa)
255{
256	ASSERT(MUTEX_HELD(&spa_namespace_lock));
257	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
258	ASSERT(spa->spa_scrub_thread == NULL);
259
260	avl_remove(&spa_namespace_avl, spa);
261	cv_broadcast(&spa_namespace_cv);
262
263	if (spa->spa_root) {
264		spa_strfree(spa->spa_root);
265		spa_active_count--;
266	}
267
268	if (spa->spa_name)
269		spa_strfree(spa->spa_name);
270
271	spa_config_set(spa, NULL);
272
273	refcount_destroy(&spa->spa_refcount);
274	refcount_destroy(&spa->spa_config_lock.scl_count);
275
276	kmem_free(spa, sizeof (spa_t));
277}
278
279/*
280 * Given a pool, return the next pool in the namespace, or NULL if there is
281 * none.  If 'prev' is NULL, return the first pool.
282 */
283spa_t *
284spa_next(spa_t *prev)
285{
286	ASSERT(MUTEX_HELD(&spa_namespace_lock));
287
288	if (prev)
289		return (AVL_NEXT(&spa_namespace_avl, prev));
290	else
291		return (avl_first(&spa_namespace_avl));
292}
293
294/*
295 * ==========================================================================
296 * SPA refcount functions
297 * ==========================================================================
298 */
299
300/*
301 * Add a reference to the given spa_t.  Must have at least one reference, or
302 * have the namespace lock held.
303 */
304void
305spa_open_ref(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_add(&spa->spa_refcount, tag);
311}
312
313/*
314 * Remove a reference to the given spa_t.  Must have at least one reference, or
315 * have the namespace lock held.
316 */
317void
318spa_close(spa_t *spa, void *tag)
319{
320	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
321	    MUTEX_HELD(&spa_namespace_lock));
322
323	(void) refcount_remove(&spa->spa_refcount, tag);
324}
325
326/*
327 * Check to see if the spa refcount is zero.  Must be called with
328 * spa_namespace_lock held.  We really compare against SPA_MINREF, which is the
329 * number of references acquired when opening a pool
330 */
331boolean_t
332spa_refcount_zero(spa_t *spa)
333{
334	ASSERT(MUTEX_HELD(&spa_namespace_lock));
335
336	return (refcount_count(&spa->spa_refcount) == SPA_MINREF);
337}
338
339/*
340 * ==========================================================================
341 * SPA config locking
342 * ==========================================================================
343 */
344
345/*
346 * Acquire the config lock.  The config lock is a special rwlock that allows for
347 * recursive enters.  Because these enters come from the same thread as well as
348 * asynchronous threads working on behalf of the owner, we must unilaterally
349 * allow all reads access as long at least one reader is held (even if a write
350 * is requested).  This has the side effect of write starvation, but write locks
351 * are extremely rare, and a solution to this problem would be significantly
352 * more complex (if even possible).
353 *
354 * We would like to assert that the namespace lock isn't held, but this is a
355 * valid use during create.
356 */
357void
358spa_config_enter(spa_t *spa, krw_t rw, void *tag)
359{
360	spa_config_lock_t *scl = &spa->spa_config_lock;
361
362	mutex_enter(&scl->scl_lock);
363
364	if (scl->scl_writer != curthread) {
365		if (rw == RW_READER) {
366			while (scl->scl_writer != NULL)
367				cv_wait(&scl->scl_cv, &scl->scl_lock);
368		} else {
369			while (scl->scl_writer != NULL ||
370			    !refcount_is_zero(&scl->scl_count))
371				cv_wait(&scl->scl_cv, &scl->scl_lock);
372			scl->scl_writer = curthread;
373		}
374	}
375
376	(void) refcount_add(&scl->scl_count, tag);
377
378	mutex_exit(&scl->scl_lock);
379}
380
381/*
382 * Release the spa config lock, notifying any waiters in the process.
383 */
384void
385spa_config_exit(spa_t *spa, void *tag)
386{
387	spa_config_lock_t *scl = &spa->spa_config_lock;
388
389	mutex_enter(&scl->scl_lock);
390
391	ASSERT(!refcount_is_zero(&scl->scl_count));
392	if (refcount_remove(&scl->scl_count, tag) == 0) {
393		cv_broadcast(&scl->scl_cv);
394		scl->scl_writer = NULL;  /* OK in either case */
395	}
396
397	mutex_exit(&scl->scl_lock);
398}
399
400/*
401 * Returns true if the config lock is held in the given manner.
402 */
403boolean_t
404spa_config_held(spa_t *spa, krw_t rw)
405{
406	spa_config_lock_t *scl = &spa->spa_config_lock;
407	boolean_t held;
408
409	mutex_enter(&scl->scl_lock);
410	if (rw == RW_WRITER)
411		held = (scl->scl_writer == curthread);
412	else
413		held = !refcount_is_zero(&scl->scl_count);
414	mutex_exit(&scl->scl_lock);
415
416	return (held);
417}
418
419/*
420 * ==========================================================================
421 * SPA vdev locking
422 * ==========================================================================
423 */
424
425/*
426 * Lock the given spa_t for the purpose of adding or removing a vdev.
427 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
428 * It returns the next transaction group for the spa_t.
429 */
430uint64_t
431spa_vdev_enter(spa_t *spa)
432{
433	/*
434	 * Suspend scrub activity while we mess with the config.
435	 */
436	spa_scrub_suspend(spa);
437
438	mutex_enter(&spa_namespace_lock);
439
440	spa_config_enter(spa, RW_WRITER, spa);
441
442	return (spa_last_synced_txg(spa) + 1);
443}
444
445/*
446 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
447 * locking of spa_vdev_enter(), we also want make sure the transactions have
448 * synced to disk, and then update the global configuration cache with the new
449 * information.
450 */
451int
452spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
453{
454	int config_changed = B_FALSE;
455
456	ASSERT(txg > spa_last_synced_txg(spa));
457
458	/*
459	 * Reassess the DTLs.
460	 */
461	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
462
463	/*
464	 * If the config changed, notify the scrub thread that it must restart.
465	 */
466	if (error == 0 && !list_is_empty(&spa->spa_dirty_list)) {
467		config_changed = B_TRUE;
468		spa_scrub_restart(spa, txg);
469	}
470
471	spa_config_exit(spa, spa);
472
473	/*
474	 * Allow scrubbing to resume.
475	 */
476	spa_scrub_resume(spa);
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	int err;
516
517	/*
518	 * Lookup the spa_t and grab the config lock for writing.  We need to
519	 * actually open the pool so that we can sync out the necessary labels.
520	 * It's OK to call spa_open() with the namespace lock held because we
521	 * allow recursive calls for other reasons.
522	 */
523	mutex_enter(&spa_namespace_lock);
524	if ((err = spa_open(name, &spa, FTAG)) != 0) {
525		mutex_exit(&spa_namespace_lock);
526		return (err);
527	}
528
529	spa_config_enter(spa, RW_WRITER, FTAG);
530
531	avl_remove(&spa_namespace_avl, spa);
532	spa_strfree(spa->spa_name);
533	spa->spa_name = spa_strdup(newname);
534	avl_add(&spa_namespace_avl, spa);
535
536	/*
537	 * Sync all labels to disk with the new names by marking the root vdev
538	 * dirty and waiting for it to sync.  It will pick up the new pool name
539	 * during the sync.
540	 */
541	vdev_config_dirty(spa->spa_root_vdev);
542
543	spa_config_exit(spa, FTAG);
544
545	txg_wait_synced(spa->spa_dsl_pool, 0);
546
547	/*
548	 * Sync the updated config cache.
549	 */
550	spa_config_sync();
551
552	spa_close(spa, FTAG);
553
554	mutex_exit(&spa_namespace_lock);
555
556	return (0);
557}
558
559
560/*
561 * Determine whether a pool with given pool_guid exists.  If device_guid is
562 * non-zero, determine whether the pool exists *and* contains a device with the
563 * specified device_guid.
564 */
565boolean_t
566spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
567{
568	spa_t *spa;
569	avl_tree_t *t = &spa_namespace_avl;
570
571	ASSERT(MUTEX_HELD(&spa_namespace_lock));
572
573	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
574		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
575			continue;
576		if (spa->spa_root_vdev == NULL)
577			continue;
578		if (spa_guid(spa) == pool_guid && (device_guid == 0 ||
579		    vdev_lookup_by_guid(spa->spa_root_vdev, device_guid)))
580			break;
581	}
582
583	return (spa != NULL);
584}
585
586char *
587spa_strdup(const char *s)
588{
589	size_t len;
590	char *new;
591
592	len = strlen(s);
593	new = kmem_alloc(len + 1, KM_SLEEP);
594	bcopy(s, new, len);
595	new[len] = '\0';
596
597	return (new);
598}
599
600void
601spa_strfree(char *s)
602{
603	kmem_free(s, strlen(s) + 1);
604}
605
606uint64_t
607spa_get_random(uint64_t range)
608{
609	uint64_t r;
610
611	ASSERT(range != 0);
612
613	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
614
615	return (r % range);
616}
617
618void
619sprintf_blkptr(char *buf, int len, const blkptr_t *bp)
620{
621	int d;
622
623	if (bp == NULL) {
624		(void) snprintf(buf, len, "<NULL>");
625		return;
626	}
627
628	if (BP_IS_HOLE(bp)) {
629		(void) snprintf(buf, len, "<hole>");
630		return;
631	}
632
633	(void) snprintf(buf, len, "[L%llu %s] %llxL/%llxP ",
634	    (u_longlong_t)BP_GET_LEVEL(bp),
635	    dmu_ot[BP_GET_TYPE(bp)].ot_name,
636	    (u_longlong_t)BP_GET_LSIZE(bp),
637	    (u_longlong_t)BP_GET_PSIZE(bp));
638
639	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
640		const dva_t *dva = &bp->blk_dva[d];
641		(void) snprintf(buf + strlen(buf), len - strlen(buf),
642		    "DVA[%d]=<%llu:%llx:%llx> ", d,
643		    (u_longlong_t)DVA_GET_VDEV(dva),
644		    (u_longlong_t)DVA_GET_OFFSET(dva),
645		    (u_longlong_t)DVA_GET_ASIZE(dva));
646	}
647
648	(void) snprintf(buf + strlen(buf), len - strlen(buf),
649	    "%s %s %s %s birth=%llu fill=%llu cksum=%llx:%llx:%llx:%llx",
650	    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name,
651	    zio_compress_table[BP_GET_COMPRESS(bp)].ci_name,
652	    BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE",
653	    BP_IS_GANG(bp) ? "gang" : "contiguous",
654	    (u_longlong_t)bp->blk_birth,
655	    (u_longlong_t)bp->blk_fill,
656	    (u_longlong_t)bp->blk_cksum.zc_word[0],
657	    (u_longlong_t)bp->blk_cksum.zc_word[1],
658	    (u_longlong_t)bp->blk_cksum.zc_word[2],
659	    (u_longlong_t)bp->blk_cksum.zc_word[3]);
660}
661
662void
663spa_freeze(spa_t *spa)
664{
665	uint64_t freeze_txg = 0;
666
667	spa_config_enter(spa, RW_WRITER, FTAG);
668	if (spa->spa_freeze_txg == UINT64_MAX) {
669		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
670		spa->spa_freeze_txg = freeze_txg;
671	}
672	spa_config_exit(spa, FTAG);
673	if (freeze_txg != 0)
674		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
675}
676
677/*
678 * ==========================================================================
679 * Accessor functions
680 * ==========================================================================
681 */
682
683krwlock_t *
684spa_traverse_rwlock(spa_t *spa)
685{
686	return (&spa->spa_traverse_lock);
687}
688
689int
690spa_traverse_wanted(spa_t *spa)
691{
692	return (spa->spa_traverse_wanted);
693}
694
695dsl_pool_t *
696spa_get_dsl(spa_t *spa)
697{
698	return (spa->spa_dsl_pool);
699}
700
701blkptr_t *
702spa_get_rootblkptr(spa_t *spa)
703{
704	return (&spa->spa_ubsync.ub_rootbp);
705}
706
707void
708spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
709{
710	spa->spa_uberblock.ub_rootbp = *bp;
711}
712
713void
714spa_altroot(spa_t *spa, char *buf, size_t buflen)
715{
716	if (spa->spa_root == NULL)
717		buf[0] = '\0';
718	else
719		(void) strncpy(buf, spa->spa_root, buflen);
720}
721
722int
723spa_sync_pass(spa_t *spa)
724{
725	return (spa->spa_sync_pass);
726}
727
728char *
729spa_name(spa_t *spa)
730{
731	/*
732	 * Accessing the name requires holding either the namespace lock or the
733	 * config lock, both of which are required to do a rename.
734	 */
735	ASSERT(MUTEX_HELD(&spa_namespace_lock) ||
736	    spa_config_held(spa, RW_READER) || spa_config_held(spa, RW_WRITER));
737
738	return (spa->spa_name);
739}
740
741uint64_t
742spa_guid(spa_t *spa)
743{
744	return (spa->spa_root_vdev->vdev_guid);
745}
746
747uint64_t
748spa_last_synced_txg(spa_t *spa)
749{
750	return (spa->spa_ubsync.ub_txg);
751}
752
753uint64_t
754spa_first_txg(spa_t *spa)
755{
756	return (spa->spa_first_txg);
757}
758
759int
760spa_state(spa_t *spa)
761{
762	return (spa->spa_state);
763}
764
765uint64_t
766spa_freeze_txg(spa_t *spa)
767{
768	return (spa->spa_freeze_txg);
769}
770
771/*
772 * In the future, this may select among different metaslab classes
773 * depending on the zdp.  For now, there's no such distinction.
774 */
775metaslab_class_t *
776spa_metaslab_class_select(spa_t *spa)
777{
778	return (spa->spa_normal_class);
779}
780
781/*
782 * Return pool-wide allocated space.
783 */
784uint64_t
785spa_get_alloc(spa_t *spa)
786{
787	return (spa->spa_root_vdev->vdev_stat.vs_alloc);
788}
789
790/*
791 * Return pool-wide allocated space.
792 */
793uint64_t
794spa_get_space(spa_t *spa)
795{
796	return (spa->spa_root_vdev->vdev_stat.vs_space);
797}
798
799/* ARGSUSED */
800uint64_t
801spa_get_asize(spa_t *spa, uint64_t lsize)
802{
803	/*
804	 * For now, the worst case is 512-byte RAID-Z blocks, in which
805	 * case the space requirement is exactly 2x; so just assume that.
806	 * Add to this the fact that we can have up to 3 DVAs per bp, and
807	 * we have to multiply by a total of 6x.
808	 */
809	return (lsize * 6);
810}
811
812uint64_t
813spa_version(spa_t *spa)
814{
815	return (spa->spa_ubsync.ub_version);
816}
817
818int
819spa_max_replication(spa_t *spa)
820{
821	/*
822	 * As of ZFS_VERSION == ZFS_VERSION_DITTO_BLOCKS, we are able to
823	 * handle BPs with more than one DVA allocated.  Set our max
824	 * replication level accordingly.
825	 */
826	if (spa_version(spa) < ZFS_VERSION_DITTO_BLOCKS)
827		return (1);
828	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
829}
830
831/*
832 * ==========================================================================
833 * Initialization and Termination
834 * ==========================================================================
835 */
836
837static int
838spa_name_compare(const void *a1, const void *a2)
839{
840	const spa_t *s1 = a1;
841	const spa_t *s2 = a2;
842	int s;
843
844	s = strcmp(s1->spa_name, s2->spa_name);
845	if (s > 0)
846		return (1);
847	if (s < 0)
848		return (-1);
849	return (0);
850}
851
852int
853spa_busy(void)
854{
855	return (spa_active_count);
856}
857
858void
859spa_init(int mode)
860{
861	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
862	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
863
864	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
865	    offsetof(spa_t, spa_avl));
866
867	spa_mode = mode;
868
869	refcount_init();
870	unique_init();
871	zio_init();
872	dmu_init();
873	zil_init();
874	spa_config_load();
875}
876
877void
878spa_fini(void)
879{
880	spa_evict_all();
881
882	zil_fini();
883	dmu_fini();
884	zio_fini();
885	refcount_fini();
886
887	avl_destroy(&spa_namespace_avl);
888
889	cv_destroy(&spa_namespace_cv);
890	mutex_destroy(&spa_namespace_lock);
891}
892