xref: /titanic_41/usr/src/uts/common/fs/zfs/spa_misc.c (revision 18c2aff776a775d34a4c9893a4c72e0434d68e36)
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 
172 static avl_tree_t spa_namespace_avl;
173 kmutex_t spa_namespace_lock;
174 static kcondvar_t spa_namespace_cv;
175 static int spa_active_count;
176 static int spa_max_replication_override = SPA_DVAS_PER_BP;
177 
178 static avl_tree_t spa_spare_avl;
179 static kmutex_t spa_spare_lock;
180 
181 kmem_cache_t *spa_buffer_pool;
182 int spa_mode;
183 
184 #ifdef ZFS_DEBUG
185 int zfs_flags = ~0;
186 #else
187 int zfs_flags = 0;
188 #endif
189 
190 #define	SPA_MINREF	5	/* spa_refcnt for an open-but-idle pool */
191 
192 /*
193  * ==========================================================================
194  * SPA namespace functions
195  * ==========================================================================
196  */
197 
198 /*
199  * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
200  * Returns NULL if no matching spa_t is found.
201  */
202 spa_t *
203 spa_lookup(const char *name)
204 {
205 	spa_t search, *spa;
206 	avl_index_t where;
207 
208 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
209 
210 	search.spa_name = (char *)name;
211 	spa = avl_find(&spa_namespace_avl, &search, &where);
212 
213 	return (spa);
214 }
215 
216 /*
217  * Create an uninitialized spa_t with the given name.  Requires
218  * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
219  * exist by calling spa_lookup() first.
220  */
221 spa_t *
222 spa_add(const char *name, const char *altroot)
223 {
224 	spa_t *spa;
225 
226 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
227 
228 	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
229 
230 	spa->spa_name = spa_strdup(name);
231 	spa->spa_state = POOL_STATE_UNINITIALIZED;
232 	spa->spa_freeze_txg = UINT64_MAX;
233 	spa->spa_final_txg = UINT64_MAX;
234 
235 	refcount_create(&spa->spa_refcount);
236 	refcount_create(&spa->spa_config_lock.scl_count);
237 
238 	avl_add(&spa_namespace_avl, spa);
239 
240 	/*
241 	 * Set the alternate root, if there is one.
242 	 */
243 	if (altroot) {
244 		spa->spa_root = spa_strdup(altroot);
245 		spa_active_count++;
246 	}
247 
248 	return (spa);
249 }
250 
251 /*
252  * Removes a spa_t from the namespace, freeing up any memory used.  Requires
253  * spa_namespace_lock.  This is called only after the spa_t has been closed and
254  * deactivated.
255  */
256 void
257 spa_remove(spa_t *spa)
258 {
259 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
260 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
261 	ASSERT(spa->spa_scrub_thread == NULL);
262 
263 	avl_remove(&spa_namespace_avl, spa);
264 	cv_broadcast(&spa_namespace_cv);
265 
266 	if (spa->spa_root) {
267 		spa_strfree(spa->spa_root);
268 		spa_active_count--;
269 	}
270 
271 	if (spa->spa_name)
272 		spa_strfree(spa->spa_name);
273 
274 	spa_config_set(spa, NULL);
275 
276 	refcount_destroy(&spa->spa_refcount);
277 	refcount_destroy(&spa->spa_config_lock.scl_count);
278 
279 	mutex_destroy(&spa->spa_sync_bplist.bpl_lock);
280 	mutex_destroy(&spa->spa_config_lock.scl_lock);
281 	mutex_destroy(&spa->spa_errlist_lock);
282 	mutex_destroy(&spa->spa_errlog_lock);
283 	mutex_destroy(&spa->spa_scrub_lock);
284 	mutex_destroy(&spa->spa_config_cache_lock);
285 	mutex_destroy(&spa->spa_async_lock);
286 
287 	kmem_free(spa, sizeof (spa_t));
288 }
289 
290 /*
291  * Given a pool, return the next pool in the namespace, or NULL if there is
292  * none.  If 'prev' is NULL, return the first pool.
293  */
294 spa_t *
295 spa_next(spa_t *prev)
296 {
297 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
298 
299 	if (prev)
300 		return (AVL_NEXT(&spa_namespace_avl, prev));
301 	else
302 		return (avl_first(&spa_namespace_avl));
303 }
304 
305 /*
306  * ==========================================================================
307  * SPA refcount functions
308  * ==========================================================================
309  */
310 
311 /*
312  * Add a reference to the given spa_t.  Must have at least one reference, or
313  * have the namespace lock held.
314  */
315 void
316 spa_open_ref(spa_t *spa, void *tag)
317 {
318 	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
319 	    MUTEX_HELD(&spa_namespace_lock));
320 
321 	(void) refcount_add(&spa->spa_refcount, tag);
322 }
323 
324 /*
325  * Remove a reference to the given spa_t.  Must have at least one reference, or
326  * have the namespace lock held.
327  */
328 void
329 spa_close(spa_t *spa, void *tag)
330 {
331 	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
332 	    MUTEX_HELD(&spa_namespace_lock));
333 
334 	(void) refcount_remove(&spa->spa_refcount, tag);
335 }
336 
337 /*
338  * Check to see if the spa refcount is zero.  Must be called with
339  * spa_namespace_lock held.  We really compare against SPA_MINREF, which is the
340  * number of references acquired when opening a pool
341  */
342 boolean_t
343 spa_refcount_zero(spa_t *spa)
344 {
345 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
346 
347 	return (refcount_count(&spa->spa_refcount) == SPA_MINREF);
348 }
349 
350 /*
351  * ==========================================================================
352  * SPA spare tracking
353  * ==========================================================================
354  */
355 
356 /*
357  * We track spare information on a global basis.  This allows us to do two
358  * things: determine when a spare is no longer referenced by any active pool,
359  * and (quickly) determine if a spare is currently in use in another pool on the
360  * system.
361  */
362 typedef struct spa_spare {
363 	uint64_t	spare_guid;
364 	avl_node_t	spare_avl;
365 	int		spare_count;
366 } spa_spare_t;
367 
368 static int
369 spa_spare_compare(const void *a, const void *b)
370 {
371 	const spa_spare_t *sa = a;
372 	const spa_spare_t *sb = b;
373 
374 	if (sa->spare_guid < sb->spare_guid)
375 		return (-1);
376 	else if (sa->spare_guid > sb->spare_guid)
377 		return (1);
378 	else
379 		return (0);
380 }
381 
382 void
383 spa_spare_add(uint64_t guid)
384 {
385 	avl_index_t where;
386 	spa_spare_t search;
387 	spa_spare_t *spare;
388 
389 	mutex_enter(&spa_spare_lock);
390 
391 	search.spare_guid = guid;
392 	if ((spare = avl_find(&spa_spare_avl, &search, &where)) != NULL) {
393 		spare->spare_count++;
394 	} else {
395 		spare = kmem_alloc(sizeof (spa_spare_t), KM_SLEEP);
396 		spare->spare_guid = guid;
397 		spare->spare_count = 1;
398 		avl_insert(&spa_spare_avl, spare, where);
399 	}
400 
401 	mutex_exit(&spa_spare_lock);
402 }
403 
404 void
405 spa_spare_remove(uint64_t guid)
406 {
407 	spa_spare_t search;
408 	spa_spare_t *spare;
409 	avl_index_t where;
410 
411 	mutex_enter(&spa_spare_lock);
412 
413 	search.spare_guid = guid;
414 	spare = avl_find(&spa_spare_avl, &search, &where);
415 
416 	ASSERT(spare != NULL);
417 
418 	if (--spare->spare_count == 0) {
419 		avl_remove(&spa_spare_avl, spare);
420 		kmem_free(spare, sizeof (spa_spare_t));
421 	}
422 
423 	mutex_exit(&spa_spare_lock);
424 }
425 
426 boolean_t
427 spa_spare_inuse(uint64_t guid)
428 {
429 	spa_spare_t search;
430 	avl_index_t where;
431 	boolean_t ret;
432 
433 	mutex_enter(&spa_spare_lock);
434 
435 	search.spare_guid = guid;
436 	ret = (avl_find(&spa_spare_avl, &search, &where) != NULL);
437 
438 	mutex_exit(&spa_spare_lock);
439 
440 	return (ret);
441 }
442 
443 /*
444  * ==========================================================================
445  * SPA config locking
446  * ==========================================================================
447  */
448 
449 /*
450  * Acquire the config lock.  The config lock is a special rwlock that allows for
451  * recursive enters.  Because these enters come from the same thread as well as
452  * asynchronous threads working on behalf of the owner, we must unilaterally
453  * allow all reads access as long at least one reader is held (even if a write
454  * is requested).  This has the side effect of write starvation, but write locks
455  * are extremely rare, and a solution to this problem would be significantly
456  * more complex (if even possible).
457  *
458  * We would like to assert that the namespace lock isn't held, but this is a
459  * valid use during create.
460  */
461 void
462 spa_config_enter(spa_t *spa, krw_t rw, void *tag)
463 {
464 	spa_config_lock_t *scl = &spa->spa_config_lock;
465 
466 	mutex_enter(&scl->scl_lock);
467 
468 	if (scl->scl_writer != curthread) {
469 		if (rw == RW_READER) {
470 			while (scl->scl_writer != NULL)
471 				cv_wait(&scl->scl_cv, &scl->scl_lock);
472 		} else {
473 			while (scl->scl_writer != NULL ||
474 			    !refcount_is_zero(&scl->scl_count))
475 				cv_wait(&scl->scl_cv, &scl->scl_lock);
476 			scl->scl_writer = curthread;
477 		}
478 	}
479 
480 	(void) refcount_add(&scl->scl_count, tag);
481 
482 	mutex_exit(&scl->scl_lock);
483 }
484 
485 /*
486  * Release the spa config lock, notifying any waiters in the process.
487  */
488 void
489 spa_config_exit(spa_t *spa, void *tag)
490 {
491 	spa_config_lock_t *scl = &spa->spa_config_lock;
492 
493 	mutex_enter(&scl->scl_lock);
494 
495 	ASSERT(!refcount_is_zero(&scl->scl_count));
496 	if (refcount_remove(&scl->scl_count, tag) == 0) {
497 		cv_broadcast(&scl->scl_cv);
498 		scl->scl_writer = NULL;  /* OK in either case */
499 	}
500 
501 	mutex_exit(&scl->scl_lock);
502 }
503 
504 /*
505  * Returns true if the config lock is held in the given manner.
506  */
507 boolean_t
508 spa_config_held(spa_t *spa, krw_t rw)
509 {
510 	spa_config_lock_t *scl = &spa->spa_config_lock;
511 	boolean_t held;
512 
513 	mutex_enter(&scl->scl_lock);
514 	if (rw == RW_WRITER)
515 		held = (scl->scl_writer == curthread);
516 	else
517 		held = !refcount_is_zero(&scl->scl_count);
518 	mutex_exit(&scl->scl_lock);
519 
520 	return (held);
521 }
522 
523 /*
524  * ==========================================================================
525  * SPA vdev locking
526  * ==========================================================================
527  */
528 
529 /*
530  * Lock the given spa_t for the purpose of adding or removing a vdev.
531  * Grabs the global spa_namespace_lock plus the spa config lock for writing.
532  * It returns the next transaction group for the spa_t.
533  */
534 uint64_t
535 spa_vdev_enter(spa_t *spa)
536 {
537 	/*
538 	 * Suspend scrub activity while we mess with the config.
539 	 */
540 	spa_scrub_suspend(spa);
541 
542 	mutex_enter(&spa_namespace_lock);
543 
544 	spa_config_enter(spa, RW_WRITER, spa);
545 
546 	return (spa_last_synced_txg(spa) + 1);
547 }
548 
549 /*
550  * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
551  * locking of spa_vdev_enter(), we also want make sure the transactions have
552  * synced to disk, and then update the global configuration cache with the new
553  * information.
554  */
555 int
556 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
557 {
558 	int config_changed = B_FALSE;
559 
560 	ASSERT(txg > spa_last_synced_txg(spa));
561 
562 	/*
563 	 * Reassess the DTLs.
564 	 */
565 	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
566 
567 	/*
568 	 * If the config changed, notify the scrub thread that it must restart.
569 	 */
570 	if (error == 0 && !list_is_empty(&spa->spa_dirty_list)) {
571 		config_changed = B_TRUE;
572 		spa_scrub_restart(spa, txg);
573 	}
574 
575 	spa_config_exit(spa, spa);
576 
577 	/*
578 	 * Allow scrubbing to resume.
579 	 */
580 	spa_scrub_resume(spa);
581 
582 	/*
583 	 * Note: this txg_wait_synced() is important because it ensures
584 	 * that there won't be more than one config change per txg.
585 	 * This allows us to use the txg as the generation number.
586 	 */
587 	if (error == 0)
588 		txg_wait_synced(spa->spa_dsl_pool, txg);
589 
590 	if (vd != NULL) {
591 		ASSERT(!vd->vdev_detached || vd->vdev_dtl.smo_object == 0);
592 		vdev_free(vd);
593 	}
594 
595 	/*
596 	 * If the config changed, update the config cache.
597 	 */
598 	if (config_changed)
599 		spa_config_sync();
600 
601 	mutex_exit(&spa_namespace_lock);
602 
603 	return (error);
604 }
605 
606 /*
607  * ==========================================================================
608  * Miscellaneous functions
609  * ==========================================================================
610  */
611 
612 /*
613  * Rename a spa_t.
614  */
615 int
616 spa_rename(const char *name, const char *newname)
617 {
618 	spa_t *spa;
619 	int err;
620 
621 	/*
622 	 * Lookup the spa_t and grab the config lock for writing.  We need to
623 	 * actually open the pool so that we can sync out the necessary labels.
624 	 * It's OK to call spa_open() with the namespace lock held because we
625 	 * allow recursive calls for other reasons.
626 	 */
627 	mutex_enter(&spa_namespace_lock);
628 	if ((err = spa_open(name, &spa, FTAG)) != 0) {
629 		mutex_exit(&spa_namespace_lock);
630 		return (err);
631 	}
632 
633 	spa_config_enter(spa, RW_WRITER, FTAG);
634 
635 	avl_remove(&spa_namespace_avl, spa);
636 	spa_strfree(spa->spa_name);
637 	spa->spa_name = spa_strdup(newname);
638 	avl_add(&spa_namespace_avl, spa);
639 
640 	/*
641 	 * Sync all labels to disk with the new names by marking the root vdev
642 	 * dirty and waiting for it to sync.  It will pick up the new pool name
643 	 * during the sync.
644 	 */
645 	vdev_config_dirty(spa->spa_root_vdev);
646 
647 	spa_config_exit(spa, FTAG);
648 
649 	txg_wait_synced(spa->spa_dsl_pool, 0);
650 
651 	/*
652 	 * Sync the updated config cache.
653 	 */
654 	spa_config_sync();
655 
656 	spa_close(spa, FTAG);
657 
658 	mutex_exit(&spa_namespace_lock);
659 
660 	return (0);
661 }
662 
663 
664 /*
665  * Determine whether a pool with given pool_guid exists.  If device_guid is
666  * non-zero, determine whether the pool exists *and* contains a device with the
667  * specified device_guid.
668  */
669 boolean_t
670 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
671 {
672 	spa_t *spa;
673 	avl_tree_t *t = &spa_namespace_avl;
674 
675 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
676 
677 	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
678 		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
679 			continue;
680 		if (spa->spa_root_vdev == NULL)
681 			continue;
682 		if (spa_guid(spa) == pool_guid && (device_guid == 0 ||
683 		    vdev_lookup_by_guid(spa->spa_root_vdev, device_guid)))
684 			break;
685 	}
686 
687 	return (spa != NULL);
688 }
689 
690 char *
691 spa_strdup(const char *s)
692 {
693 	size_t len;
694 	char *new;
695 
696 	len = strlen(s);
697 	new = kmem_alloc(len + 1, KM_SLEEP);
698 	bcopy(s, new, len);
699 	new[len] = '\0';
700 
701 	return (new);
702 }
703 
704 void
705 spa_strfree(char *s)
706 {
707 	kmem_free(s, strlen(s) + 1);
708 }
709 
710 uint64_t
711 spa_get_random(uint64_t range)
712 {
713 	uint64_t r;
714 
715 	ASSERT(range != 0);
716 
717 	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
718 
719 	return (r % range);
720 }
721 
722 void
723 sprintf_blkptr(char *buf, int len, const blkptr_t *bp)
724 {
725 	int d;
726 
727 	if (bp == NULL) {
728 		(void) snprintf(buf, len, "<NULL>");
729 		return;
730 	}
731 
732 	if (BP_IS_HOLE(bp)) {
733 		(void) snprintf(buf, len, "<hole>");
734 		return;
735 	}
736 
737 	(void) snprintf(buf, len, "[L%llu %s] %llxL/%llxP ",
738 	    (u_longlong_t)BP_GET_LEVEL(bp),
739 	    dmu_ot[BP_GET_TYPE(bp)].ot_name,
740 	    (u_longlong_t)BP_GET_LSIZE(bp),
741 	    (u_longlong_t)BP_GET_PSIZE(bp));
742 
743 	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
744 		const dva_t *dva = &bp->blk_dva[d];
745 		(void) snprintf(buf + strlen(buf), len - strlen(buf),
746 		    "DVA[%d]=<%llu:%llx:%llx> ", d,
747 		    (u_longlong_t)DVA_GET_VDEV(dva),
748 		    (u_longlong_t)DVA_GET_OFFSET(dva),
749 		    (u_longlong_t)DVA_GET_ASIZE(dva));
750 	}
751 
752 	(void) snprintf(buf + strlen(buf), len - strlen(buf),
753 	    "%s %s %s %s birth=%llu fill=%llu cksum=%llx:%llx:%llx:%llx",
754 	    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name,
755 	    zio_compress_table[BP_GET_COMPRESS(bp)].ci_name,
756 	    BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE",
757 	    BP_IS_GANG(bp) ? "gang" : "contiguous",
758 	    (u_longlong_t)bp->blk_birth,
759 	    (u_longlong_t)bp->blk_fill,
760 	    (u_longlong_t)bp->blk_cksum.zc_word[0],
761 	    (u_longlong_t)bp->blk_cksum.zc_word[1],
762 	    (u_longlong_t)bp->blk_cksum.zc_word[2],
763 	    (u_longlong_t)bp->blk_cksum.zc_word[3]);
764 }
765 
766 void
767 spa_freeze(spa_t *spa)
768 {
769 	uint64_t freeze_txg = 0;
770 
771 	spa_config_enter(spa, RW_WRITER, FTAG);
772 	if (spa->spa_freeze_txg == UINT64_MAX) {
773 		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
774 		spa->spa_freeze_txg = freeze_txg;
775 	}
776 	spa_config_exit(spa, FTAG);
777 	if (freeze_txg != 0)
778 		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
779 }
780 
781 /*
782  * ==========================================================================
783  * Accessor functions
784  * ==========================================================================
785  */
786 
787 krwlock_t *
788 spa_traverse_rwlock(spa_t *spa)
789 {
790 	return (&spa->spa_traverse_lock);
791 }
792 
793 int
794 spa_traverse_wanted(spa_t *spa)
795 {
796 	return (spa->spa_traverse_wanted);
797 }
798 
799 dsl_pool_t *
800 spa_get_dsl(spa_t *spa)
801 {
802 	return (spa->spa_dsl_pool);
803 }
804 
805 blkptr_t *
806 spa_get_rootblkptr(spa_t *spa)
807 {
808 	return (&spa->spa_ubsync.ub_rootbp);
809 }
810 
811 void
812 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
813 {
814 	spa->spa_uberblock.ub_rootbp = *bp;
815 }
816 
817 void
818 spa_altroot(spa_t *spa, char *buf, size_t buflen)
819 {
820 	if (spa->spa_root == NULL)
821 		buf[0] = '\0';
822 	else
823 		(void) strncpy(buf, spa->spa_root, buflen);
824 }
825 
826 int
827 spa_sync_pass(spa_t *spa)
828 {
829 	return (spa->spa_sync_pass);
830 }
831 
832 char *
833 spa_name(spa_t *spa)
834 {
835 	/*
836 	 * Accessing the name requires holding either the namespace lock or the
837 	 * config lock, both of which are required to do a rename.
838 	 */
839 	ASSERT(MUTEX_HELD(&spa_namespace_lock) ||
840 	    spa_config_held(spa, RW_READER) || spa_config_held(spa, RW_WRITER));
841 
842 	return (spa->spa_name);
843 }
844 
845 uint64_t
846 spa_guid(spa_t *spa)
847 {
848 	/*
849 	 * If we fail to parse the config during spa_load(), we can go through
850 	 * the error path (which posts an ereport) and end up here with no root
851 	 * vdev.  We stash the original pool guid in 'spa_load_guid' to handle
852 	 * this case.
853 	 */
854 	if (spa->spa_root_vdev != NULL)
855 		return (spa->spa_root_vdev->vdev_guid);
856 	else
857 		return (spa->spa_load_guid);
858 }
859 
860 uint64_t
861 spa_last_synced_txg(spa_t *spa)
862 {
863 	return (spa->spa_ubsync.ub_txg);
864 }
865 
866 uint64_t
867 spa_first_txg(spa_t *spa)
868 {
869 	return (spa->spa_first_txg);
870 }
871 
872 int
873 spa_state(spa_t *spa)
874 {
875 	return (spa->spa_state);
876 }
877 
878 uint64_t
879 spa_freeze_txg(spa_t *spa)
880 {
881 	return (spa->spa_freeze_txg);
882 }
883 
884 /*
885  * In the future, this may select among different metaslab classes
886  * depending on the zdp.  For now, there's no such distinction.
887  */
888 metaslab_class_t *
889 spa_metaslab_class_select(spa_t *spa)
890 {
891 	return (spa->spa_normal_class);
892 }
893 
894 /*
895  * Return how much space is allocated in the pool (ie. sum of all asize)
896  */
897 uint64_t
898 spa_get_alloc(spa_t *spa)
899 {
900 	return (spa->spa_root_vdev->vdev_stat.vs_alloc);
901 }
902 
903 /*
904  * Return how much (raid-z inflated) space there is in the pool.
905  */
906 uint64_t
907 spa_get_space(spa_t *spa)
908 {
909 	return (spa->spa_root_vdev->vdev_stat.vs_space);
910 }
911 
912 /*
913  * Return the amount of raid-z-deflated space in the pool.
914  */
915 uint64_t
916 spa_get_dspace(spa_t *spa)
917 {
918 	if (spa->spa_deflate)
919 		return (spa->spa_root_vdev->vdev_stat.vs_dspace);
920 	else
921 		return (spa->spa_root_vdev->vdev_stat.vs_space);
922 }
923 
924 /* ARGSUSED */
925 uint64_t
926 spa_get_asize(spa_t *spa, uint64_t lsize)
927 {
928 	/*
929 	 * For now, the worst case is 512-byte RAID-Z blocks, in which
930 	 * case the space requirement is exactly 2x; so just assume that.
931 	 * Add to this the fact that we can have up to 3 DVAs per bp, and
932 	 * we have to multiply by a total of 6x.
933 	 */
934 	return (lsize * 6);
935 }
936 
937 uint64_t
938 spa_version(spa_t *spa)
939 {
940 	return (spa->spa_ubsync.ub_version);
941 }
942 
943 int
944 spa_max_replication(spa_t *spa)
945 {
946 	/*
947 	 * As of ZFS_VERSION == ZFS_VERSION_DITTO_BLOCKS, we are able to
948 	 * handle BPs with more than one DVA allocated.  Set our max
949 	 * replication level accordingly.
950 	 */
951 	if (spa_version(spa) < ZFS_VERSION_DITTO_BLOCKS)
952 		return (1);
953 	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
954 }
955 
956 uint64_t
957 bp_get_dasize(spa_t *spa, const blkptr_t *bp)
958 {
959 	int sz = 0, i;
960 
961 	if (!spa->spa_deflate)
962 		return (BP_GET_ASIZE(bp));
963 
964 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
965 		vdev_t *vd =
966 		    vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[i]));
967 		sz += (DVA_GET_ASIZE(&bp->blk_dva[i]) >> SPA_MINBLOCKSHIFT) *
968 		    vd->vdev_deflate_ratio;
969 	}
970 	return (sz);
971 }
972 
973 /*
974  * ==========================================================================
975  * Initialization and Termination
976  * ==========================================================================
977  */
978 
979 static int
980 spa_name_compare(const void *a1, const void *a2)
981 {
982 	const spa_t *s1 = a1;
983 	const spa_t *s2 = a2;
984 	int s;
985 
986 	s = strcmp(s1->spa_name, s2->spa_name);
987 	if (s > 0)
988 		return (1);
989 	if (s < 0)
990 		return (-1);
991 	return (0);
992 }
993 
994 int
995 spa_busy(void)
996 {
997 	return (spa_active_count);
998 }
999 
1000 void
1001 spa_init(int mode)
1002 {
1003 	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1004 	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1005 
1006 	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1007 	    offsetof(spa_t, spa_avl));
1008 
1009 	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_spare_t),
1010 	    offsetof(spa_spare_t, spare_avl));
1011 
1012 	spa_mode = mode;
1013 
1014 	refcount_init();
1015 	unique_init();
1016 	zio_init();
1017 	dmu_init();
1018 	zil_init();
1019 	spa_config_load();
1020 }
1021 
1022 void
1023 spa_fini(void)
1024 {
1025 	spa_evict_all();
1026 
1027 	zil_fini();
1028 	dmu_fini();
1029 	zio_fini();
1030 	refcount_fini();
1031 
1032 	avl_destroy(&spa_namespace_avl);
1033 	avl_destroy(&spa_spare_avl);
1034 
1035 	cv_destroy(&spa_namespace_cv);
1036 	mutex_destroy(&spa_namespace_lock);
1037 }
1038