xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa_misc.c (revision 990b4856d0eaada6f8140335733a1b1771ed2746)
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 2007 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 #include <sys/metaslab_impl.h>
48 #include "zfs_prop.h"
49 
50 /*
51  * SPA locking
52  *
53  * There are four basic locks for managing spa_t structures:
54  *
55  * spa_namespace_lock (global mutex)
56  *
57  *	This lock must be acquired to do any of the following:
58  *
59  *		- Lookup a spa_t by name
60  *		- Add or remove a spa_t from the namespace
61  *		- Increase spa_refcount from non-zero
62  *		- Check if spa_refcount is zero
63  *		- Rename a spa_t
64  *		- add/remove/attach/detach devices
65  *		- Held for the duration of create/destroy/import/export
66  *
67  *	It does not need to handle recursion.  A create or destroy may
68  *	reference objects (files or zvols) in other pools, but by
69  *	definition they must have an existing reference, and will never need
70  *	to lookup a spa_t by name.
71  *
72  * spa_refcount (per-spa refcount_t protected by mutex)
73  *
74  *	This reference count keep track of any active users of the spa_t.  The
75  *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
76  *	the refcount is never really 'zero' - opening a pool implicitly keeps
77  *	some references in the DMU.  Internally we check against SPA_MINREF, but
78  *	present the image of a zero/non-zero value to consumers.
79  *
80  * spa_config_lock (per-spa read-priority rwlock)
81  *
82  *	This protects the spa_t from config changes, and must be held in
83  *	the following circumstances:
84  *
85  *		- RW_READER to perform I/O to the spa
86  *		- RW_WRITER to change the vdev config
87  *
88  * spa_config_cache_lock (per-spa mutex)
89  *
90  *	This mutex prevents the spa_config nvlist from being updated.  No
91  *      other locks are required to obtain this lock, although implicitly you
92  *      must have the namespace lock or non-zero refcount to have any kind
93  *      of spa_t pointer at all.
94  *
95  * The locking order is fairly straightforward:
96  *
97  *		spa_namespace_lock	->	spa_refcount
98  *
99  *	The namespace lock must be acquired to increase the refcount from 0
100  *	or to check if it is zero.
101  *
102  *		spa_refcount		->	spa_config_lock
103  *
104  *	There must be at least one valid reference on the spa_t to acquire
105  *	the config lock.
106  *
107  *		spa_namespace_lock	->	spa_config_lock
108  *
109  *	The namespace lock must always be taken before the config lock.
110  *
111  *
112  * The spa_namespace_lock and spa_config_cache_lock can be acquired directly and
113  * are globally visible.
114  *
115  * The namespace is manipulated using the following functions, all which require
116  * the spa_namespace_lock to be held.
117  *
118  *	spa_lookup()		Lookup a spa_t by name.
119  *
120  *	spa_add()		Create a new spa_t in the namespace.
121  *
122  *	spa_remove()		Remove a spa_t from the namespace.  This also
123  *				frees up any memory associated with the spa_t.
124  *
125  *	spa_next()		Returns the next spa_t in the system, or the
126  *				first if NULL is passed.
127  *
128  *	spa_evict_all()		Shutdown and remove all spa_t structures in
129  *				the system.
130  *
131  *	spa_guid_exists()	Determine whether a pool/device guid exists.
132  *
133  * The spa_refcount is manipulated using the following functions:
134  *
135  *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
136  *				called with spa_namespace_lock held if the
137  *				refcount is currently zero.
138  *
139  *	spa_close()		Remove a reference from the spa_t.  This will
140  *				not free the spa_t or remove it from the
141  *				namespace.  No locking is required.
142  *
143  *	spa_refcount_zero()	Returns true if the refcount is currently
144  *				zero.  Must be called with spa_namespace_lock
145  *				held.
146  *
147  * The spa_config_lock is manipulated using the following functions:
148  *
149  *	spa_config_enter()	Acquire the config lock as RW_READER or
150  *				RW_WRITER.  At least one reference on the spa_t
151  *				must exist.
152  *
153  *	spa_config_exit()	Release the config lock.
154  *
155  *	spa_config_held()	Returns true if the config lock is currently
156  *				held in the given state.
157  *
158  * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
159  *
160  *	spa_vdev_enter()	Acquire the namespace lock and the config lock
161  *				for writing.
162  *
163  *	spa_vdev_exit()		Release the config lock, wait for all I/O
164  *				to complete, sync the updated configs to the
165  *				cache, and release the namespace lock.
166  *
167  * The spa_name() function also requires either the spa_namespace_lock
168  * or the spa_config_lock, as both are needed to do a rename.  spa_rename() is
169  * also implemented within this file since is requires manipulation of the
170  * namespace.
171  */
172 
173 static avl_tree_t spa_namespace_avl;
174 kmutex_t spa_namespace_lock;
175 static kcondvar_t spa_namespace_cv;
176 static int spa_active_count;
177 int spa_max_replication_override = SPA_DVAS_PER_BP;
178 
179 static kmutex_t spa_spare_lock;
180 static avl_tree_t spa_spare_avl;
181 
182 kmem_cache_t *spa_buffer_pool;
183 int spa_mode;
184 
185 #ifdef ZFS_DEBUG
186 /* Everything except dprintf is on by default in debug builds */
187 int zfs_flags = ~ZFS_DEBUG_DPRINTF;
188 #else
189 int zfs_flags = 0;
190 #endif
191 
192 /*
193  * zfs_recover can be set to nonzero to attempt to recover from
194  * otherwise-fatal errors, typically caused by on-disk corruption.  When
195  * set, calls to zfs_panic_recover() will turn into warning messages.
196  */
197 int zfs_recover = 0;
198 
199 #define	SPA_MINREF	5	/* spa_refcnt for an open-but-idle pool */
200 
201 /*
202  * ==========================================================================
203  * SPA namespace functions
204  * ==========================================================================
205  */
206 
207 /*
208  * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
209  * Returns NULL if no matching spa_t is found.
210  */
211 spa_t *
212 spa_lookup(const char *name)
213 {
214 	spa_t search, *spa;
215 	avl_index_t where;
216 	char c;
217 	char *cp;
218 
219 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
220 
221 	/*
222 	 * If it's a full dataset name, figure out the pool name and
223 	 * just use that.
224 	 */
225 	cp = strpbrk(name, "/@");
226 	if (cp) {
227 		c = *cp;
228 		*cp = '\0';
229 	}
230 
231 	search.spa_name = (char *)name;
232 	spa = avl_find(&spa_namespace_avl, &search, &where);
233 
234 	if (cp)
235 		*cp = c;
236 
237 	return (spa);
238 }
239 
240 /*
241  * Create an uninitialized spa_t with the given name.  Requires
242  * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
243  * exist by calling spa_lookup() first.
244  */
245 spa_t *
246 spa_add(const char *name, const char *altroot)
247 {
248 	spa_t *spa;
249 
250 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
251 
252 	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
253 
254 	rw_init(&spa->spa_traverse_lock, NULL, RW_DEFAULT, NULL);
255 
256 	mutex_init(&spa->spa_uberblock_lock, NULL, MUTEX_DEFAULT, NULL);
257 	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
258 	mutex_init(&spa->spa_config_cache_lock, NULL, MUTEX_DEFAULT, NULL);
259 	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
260 	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
261 	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
262 	mutex_init(&spa->spa_sync_bplist.bpl_lock, NULL, MUTEX_DEFAULT, NULL);
263 	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
264 	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
265 
266 	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
267 	cv_init(&spa->spa_scrub_cv, NULL, CV_DEFAULT, NULL);
268 	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
269 
270 	spa->spa_name = spa_strdup(name);
271 	spa->spa_state = POOL_STATE_UNINITIALIZED;
272 	spa->spa_freeze_txg = UINT64_MAX;
273 	spa->spa_final_txg = UINT64_MAX;
274 
275 	refcount_create(&spa->spa_refcount);
276 	rprw_init(&spa->spa_config_lock);
277 
278 	avl_add(&spa_namespace_avl, spa);
279 
280 	/*
281 	 * Set the alternate root, if there is one.
282 	 */
283 	if (altroot) {
284 		spa->spa_root = spa_strdup(altroot);
285 		spa_active_count++;
286 	}
287 
288 	return (spa);
289 }
290 
291 /*
292  * Removes a spa_t from the namespace, freeing up any memory used.  Requires
293  * spa_namespace_lock.  This is called only after the spa_t has been closed and
294  * deactivated.
295  */
296 void
297 spa_remove(spa_t *spa)
298 {
299 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
300 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
301 	ASSERT(spa->spa_scrub_thread == NULL);
302 
303 	avl_remove(&spa_namespace_avl, spa);
304 	cv_broadcast(&spa_namespace_cv);
305 
306 	if (spa->spa_root) {
307 		spa_strfree(spa->spa_root);
308 		spa_active_count--;
309 	}
310 
311 	if (spa->spa_name)
312 		spa_strfree(spa->spa_name);
313 
314 	spa_config_set(spa, NULL);
315 
316 	refcount_destroy(&spa->spa_refcount);
317 
318 	rprw_destroy(&spa->spa_config_lock);
319 
320 	rw_destroy(&spa->spa_traverse_lock);
321 
322 	cv_destroy(&spa->spa_async_cv);
323 	cv_destroy(&spa->spa_scrub_cv);
324 	cv_destroy(&spa->spa_scrub_io_cv);
325 
326 	mutex_destroy(&spa->spa_uberblock_lock);
327 	mutex_destroy(&spa->spa_async_lock);
328 	mutex_destroy(&spa->spa_config_cache_lock);
329 	mutex_destroy(&spa->spa_scrub_lock);
330 	mutex_destroy(&spa->spa_errlog_lock);
331 	mutex_destroy(&spa->spa_errlist_lock);
332 	mutex_destroy(&spa->spa_sync_bplist.bpl_lock);
333 	mutex_destroy(&spa->spa_history_lock);
334 	mutex_destroy(&spa->spa_props_lock);
335 
336 	kmem_free(spa, sizeof (spa_t));
337 }
338 
339 /*
340  * Given a pool, return the next pool in the namespace, or NULL if there is
341  * none.  If 'prev' is NULL, return the first pool.
342  */
343 spa_t *
344 spa_next(spa_t *prev)
345 {
346 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
347 
348 	if (prev)
349 		return (AVL_NEXT(&spa_namespace_avl, prev));
350 	else
351 		return (avl_first(&spa_namespace_avl));
352 }
353 
354 /*
355  * ==========================================================================
356  * SPA refcount functions
357  * ==========================================================================
358  */
359 
360 /*
361  * Add a reference to the given spa_t.  Must have at least one reference, or
362  * have the namespace lock held.
363  */
364 void
365 spa_open_ref(spa_t *spa, void *tag)
366 {
367 	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
368 	    MUTEX_HELD(&spa_namespace_lock));
369 
370 	(void) refcount_add(&spa->spa_refcount, tag);
371 }
372 
373 /*
374  * Remove a reference to the given spa_t.  Must have at least one reference, or
375  * have the namespace lock held.
376  */
377 void
378 spa_close(spa_t *spa, void *tag)
379 {
380 	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
381 	    MUTEX_HELD(&spa_namespace_lock));
382 
383 	(void) refcount_remove(&spa->spa_refcount, tag);
384 }
385 
386 /*
387  * Check to see if the spa refcount is zero.  Must be called with
388  * spa_namespace_lock held.  We really compare against SPA_MINREF, which is the
389  * number of references acquired when opening a pool
390  */
391 boolean_t
392 spa_refcount_zero(spa_t *spa)
393 {
394 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
395 
396 	return (refcount_count(&spa->spa_refcount) == SPA_MINREF);
397 }
398 
399 /*
400  * ==========================================================================
401  * SPA spare tracking
402  * ==========================================================================
403  */
404 
405 /*
406  * Spares are tracked globally due to the following constraints:
407  *
408  * 	- A spare may be part of multiple pools.
409  * 	- A spare may be added to a pool even if it's actively in use within
410  *	  another pool.
411  * 	- A spare in use in any pool can only be the source of a replacement if
412  *	  the target is a spare in the same pool.
413  *
414  * We keep track of all spares on the system through the use of a reference
415  * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
416  * spare, then we bump the reference count in the AVL tree.  In addition, we set
417  * the 'vdev_isspare' member to indicate that the device is a spare (active or
418  * inactive).  When a spare is made active (used to replace a device in the
419  * pool), we also keep track of which pool its been made a part of.
420  *
421  * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
422  * called under the spa_namespace lock as part of vdev reconfiguration.  The
423  * separate spare lock exists for the status query path, which does not need to
424  * be completely consistent with respect to other vdev configuration changes.
425  */
426 
427 typedef struct spa_spare {
428 	uint64_t	spare_guid;
429 	uint64_t	spare_pool;
430 	avl_node_t	spare_avl;
431 	int		spare_count;
432 } spa_spare_t;
433 
434 static int
435 spa_spare_compare(const void *a, const void *b)
436 {
437 	const spa_spare_t *sa = a;
438 	const spa_spare_t *sb = b;
439 
440 	if (sa->spare_guid < sb->spare_guid)
441 		return (-1);
442 	else if (sa->spare_guid > sb->spare_guid)
443 		return (1);
444 	else
445 		return (0);
446 }
447 
448 void
449 spa_spare_add(vdev_t *vd)
450 {
451 	avl_index_t where;
452 	spa_spare_t search;
453 	spa_spare_t *spare;
454 
455 	mutex_enter(&spa_spare_lock);
456 	ASSERT(!vd->vdev_isspare);
457 
458 	search.spare_guid = vd->vdev_guid;
459 	if ((spare = avl_find(&spa_spare_avl, &search, &where)) != NULL) {
460 		spare->spare_count++;
461 	} else {
462 		spare = kmem_zalloc(sizeof (spa_spare_t), KM_SLEEP);
463 		spare->spare_guid = vd->vdev_guid;
464 		spare->spare_count = 1;
465 		avl_insert(&spa_spare_avl, spare, where);
466 	}
467 	vd->vdev_isspare = B_TRUE;
468 
469 	mutex_exit(&spa_spare_lock);
470 }
471 
472 void
473 spa_spare_remove(vdev_t *vd)
474 {
475 	spa_spare_t search;
476 	spa_spare_t *spare;
477 	avl_index_t where;
478 
479 	mutex_enter(&spa_spare_lock);
480 
481 	search.spare_guid = vd->vdev_guid;
482 	spare = avl_find(&spa_spare_avl, &search, &where);
483 
484 	ASSERT(vd->vdev_isspare);
485 	ASSERT(spare != NULL);
486 
487 	if (--spare->spare_count == 0) {
488 		avl_remove(&spa_spare_avl, spare);
489 		kmem_free(spare, sizeof (spa_spare_t));
490 	} else if (spare->spare_pool == spa_guid(vd->vdev_spa)) {
491 		spare->spare_pool = 0ULL;
492 	}
493 
494 	vd->vdev_isspare = B_FALSE;
495 	mutex_exit(&spa_spare_lock);
496 }
497 
498 boolean_t
499 spa_spare_exists(uint64_t guid, uint64_t *pool)
500 {
501 	spa_spare_t search, *found;
502 	avl_index_t where;
503 
504 	mutex_enter(&spa_spare_lock);
505 
506 	search.spare_guid = guid;
507 	found = avl_find(&spa_spare_avl, &search, &where);
508 
509 	if (pool) {
510 		if (found)
511 			*pool = found->spare_pool;
512 		else
513 			*pool = 0ULL;
514 	}
515 
516 	mutex_exit(&spa_spare_lock);
517 
518 	return (found != NULL);
519 }
520 
521 void
522 spa_spare_activate(vdev_t *vd)
523 {
524 	spa_spare_t search, *found;
525 	avl_index_t where;
526 
527 	mutex_enter(&spa_spare_lock);
528 	ASSERT(vd->vdev_isspare);
529 
530 	search.spare_guid = vd->vdev_guid;
531 	found = avl_find(&spa_spare_avl, &search, &where);
532 	ASSERT(found != NULL);
533 	ASSERT(found->spare_pool == 0ULL);
534 
535 	found->spare_pool = spa_guid(vd->vdev_spa);
536 	mutex_exit(&spa_spare_lock);
537 }
538 
539 /*
540  * ==========================================================================
541  * SPA config locking
542  * ==========================================================================
543  */
544 void
545 spa_config_enter(spa_t *spa, krw_t rw, void *tag)
546 {
547 	rprw_enter(&spa->spa_config_lock, rw, tag);
548 }
549 
550 void
551 spa_config_exit(spa_t *spa, void *tag)
552 {
553 	rprw_exit(&spa->spa_config_lock, tag);
554 }
555 
556 boolean_t
557 spa_config_held(spa_t *spa, krw_t rw)
558 {
559 	return (rprw_held(&spa->spa_config_lock, rw));
560 }
561 
562 /*
563  * ==========================================================================
564  * SPA vdev locking
565  * ==========================================================================
566  */
567 
568 /*
569  * Lock the given spa_t for the purpose of adding or removing a vdev.
570  * Grabs the global spa_namespace_lock plus the spa config lock for writing.
571  * It returns the next transaction group for the spa_t.
572  */
573 uint64_t
574 spa_vdev_enter(spa_t *spa)
575 {
576 	mutex_enter(&spa_namespace_lock);
577 
578 	/*
579 	 * Suspend scrub activity while we mess with the config.  We must do
580 	 * this after acquiring the namespace lock to avoid a 3-way deadlock
581 	 * with spa_scrub_stop() and the scrub thread.
582 	 */
583 	spa_scrub_suspend(spa);
584 
585 	spa_config_enter(spa, RW_WRITER, spa);
586 
587 	return (spa_last_synced_txg(spa) + 1);
588 }
589 
590 /*
591  * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
592  * locking of spa_vdev_enter(), we also want make sure the transactions have
593  * synced to disk, and then update the global configuration cache with the new
594  * information.
595  */
596 int
597 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
598 {
599 	int config_changed = B_FALSE;
600 
601 	ASSERT(txg > spa_last_synced_txg(spa));
602 
603 	/*
604 	 * Reassess the DTLs.
605 	 */
606 	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
607 
608 	/*
609 	 * If the config changed, notify the scrub thread that it must restart.
610 	 */
611 	if (error == 0 && !list_is_empty(&spa->spa_dirty_list)) {
612 		config_changed = B_TRUE;
613 		spa_scrub_restart(spa, txg);
614 	}
615 
616 	spa_config_exit(spa, spa);
617 
618 	/*
619 	 * Allow scrubbing to resume.
620 	 */
621 	spa_scrub_resume(spa);
622 
623 	/*
624 	 * Note: this txg_wait_synced() is important because it ensures
625 	 * that there won't be more than one config change per txg.
626 	 * This allows us to use the txg as the generation number.
627 	 */
628 	if (error == 0)
629 		txg_wait_synced(spa->spa_dsl_pool, txg);
630 
631 	if (vd != NULL) {
632 		ASSERT(!vd->vdev_detached || vd->vdev_dtl.smo_object == 0);
633 		vdev_free(vd);
634 	}
635 
636 	/*
637 	 * If the config changed, update the config cache.
638 	 */
639 	if (config_changed)
640 		spa_config_sync();
641 
642 	mutex_exit(&spa_namespace_lock);
643 
644 	return (error);
645 }
646 
647 /*
648  * ==========================================================================
649  * Miscellaneous functions
650  * ==========================================================================
651  */
652 
653 /*
654  * Rename a spa_t.
655  */
656 int
657 spa_rename(const char *name, const char *newname)
658 {
659 	spa_t *spa;
660 	int err;
661 
662 	/*
663 	 * Lookup the spa_t and grab the config lock for writing.  We need to
664 	 * actually open the pool so that we can sync out the necessary labels.
665 	 * It's OK to call spa_open() with the namespace lock held because we
666 	 * allow recursive calls for other reasons.
667 	 */
668 	mutex_enter(&spa_namespace_lock);
669 	if ((err = spa_open(name, &spa, FTAG)) != 0) {
670 		mutex_exit(&spa_namespace_lock);
671 		return (err);
672 	}
673 
674 	spa_config_enter(spa, RW_WRITER, FTAG);
675 
676 	avl_remove(&spa_namespace_avl, spa);
677 	spa_strfree(spa->spa_name);
678 	spa->spa_name = spa_strdup(newname);
679 	avl_add(&spa_namespace_avl, spa);
680 
681 	/*
682 	 * Sync all labels to disk with the new names by marking the root vdev
683 	 * dirty and waiting for it to sync.  It will pick up the new pool name
684 	 * during the sync.
685 	 */
686 	vdev_config_dirty(spa->spa_root_vdev);
687 
688 	spa_config_exit(spa, FTAG);
689 
690 	txg_wait_synced(spa->spa_dsl_pool, 0);
691 
692 	/*
693 	 * Sync the updated config cache.
694 	 */
695 	spa_config_sync();
696 
697 	spa_close(spa, FTAG);
698 
699 	mutex_exit(&spa_namespace_lock);
700 
701 	return (0);
702 }
703 
704 
705 /*
706  * Determine whether a pool with given pool_guid exists.  If device_guid is
707  * non-zero, determine whether the pool exists *and* contains a device with the
708  * specified device_guid.
709  */
710 boolean_t
711 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
712 {
713 	spa_t *spa;
714 	avl_tree_t *t = &spa_namespace_avl;
715 
716 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
717 
718 	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
719 		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
720 			continue;
721 		if (spa->spa_root_vdev == NULL)
722 			continue;
723 		if (spa_guid(spa) == pool_guid) {
724 			if (device_guid == 0)
725 				break;
726 
727 			if (vdev_lookup_by_guid(spa->spa_root_vdev,
728 			    device_guid) != NULL)
729 				break;
730 
731 			/*
732 			 * Check any devices we may be in the process of adding.
733 			 */
734 			if (spa->spa_pending_vdev) {
735 				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
736 				    device_guid) != NULL)
737 					break;
738 			}
739 		}
740 	}
741 
742 	return (spa != NULL);
743 }
744 
745 char *
746 spa_strdup(const char *s)
747 {
748 	size_t len;
749 	char *new;
750 
751 	len = strlen(s);
752 	new = kmem_alloc(len + 1, KM_SLEEP);
753 	bcopy(s, new, len);
754 	new[len] = '\0';
755 
756 	return (new);
757 }
758 
759 void
760 spa_strfree(char *s)
761 {
762 	kmem_free(s, strlen(s) + 1);
763 }
764 
765 uint64_t
766 spa_get_random(uint64_t range)
767 {
768 	uint64_t r;
769 
770 	ASSERT(range != 0);
771 
772 	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
773 
774 	return (r % range);
775 }
776 
777 void
778 sprintf_blkptr(char *buf, int len, const blkptr_t *bp)
779 {
780 	int d;
781 
782 	if (bp == NULL) {
783 		(void) snprintf(buf, len, "<NULL>");
784 		return;
785 	}
786 
787 	if (BP_IS_HOLE(bp)) {
788 		(void) snprintf(buf, len, "<hole>");
789 		return;
790 	}
791 
792 	(void) snprintf(buf, len, "[L%llu %s] %llxL/%llxP ",
793 	    (u_longlong_t)BP_GET_LEVEL(bp),
794 	    dmu_ot[BP_GET_TYPE(bp)].ot_name,
795 	    (u_longlong_t)BP_GET_LSIZE(bp),
796 	    (u_longlong_t)BP_GET_PSIZE(bp));
797 
798 	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
799 		const dva_t *dva = &bp->blk_dva[d];
800 		(void) snprintf(buf + strlen(buf), len - strlen(buf),
801 		    "DVA[%d]=<%llu:%llx:%llx> ", d,
802 		    (u_longlong_t)DVA_GET_VDEV(dva),
803 		    (u_longlong_t)DVA_GET_OFFSET(dva),
804 		    (u_longlong_t)DVA_GET_ASIZE(dva));
805 	}
806 
807 	(void) snprintf(buf + strlen(buf), len - strlen(buf),
808 	    "%s %s %s %s birth=%llu fill=%llu cksum=%llx:%llx:%llx:%llx",
809 	    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name,
810 	    zio_compress_table[BP_GET_COMPRESS(bp)].ci_name,
811 	    BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE",
812 	    BP_IS_GANG(bp) ? "gang" : "contiguous",
813 	    (u_longlong_t)bp->blk_birth,
814 	    (u_longlong_t)bp->blk_fill,
815 	    (u_longlong_t)bp->blk_cksum.zc_word[0],
816 	    (u_longlong_t)bp->blk_cksum.zc_word[1],
817 	    (u_longlong_t)bp->blk_cksum.zc_word[2],
818 	    (u_longlong_t)bp->blk_cksum.zc_word[3]);
819 }
820 
821 void
822 spa_freeze(spa_t *spa)
823 {
824 	uint64_t freeze_txg = 0;
825 
826 	spa_config_enter(spa, RW_WRITER, FTAG);
827 	if (spa->spa_freeze_txg == UINT64_MAX) {
828 		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
829 		spa->spa_freeze_txg = freeze_txg;
830 	}
831 	spa_config_exit(spa, FTAG);
832 	if (freeze_txg != 0)
833 		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
834 }
835 
836 void
837 zfs_panic_recover(const char *fmt, ...)
838 {
839 	va_list adx;
840 
841 	va_start(adx, fmt);
842 	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
843 	va_end(adx);
844 }
845 
846 /*
847  * ==========================================================================
848  * Accessor functions
849  * ==========================================================================
850  */
851 
852 krwlock_t *
853 spa_traverse_rwlock(spa_t *spa)
854 {
855 	return (&spa->spa_traverse_lock);
856 }
857 
858 int
859 spa_traverse_wanted(spa_t *spa)
860 {
861 	return (spa->spa_traverse_wanted);
862 }
863 
864 dsl_pool_t *
865 spa_get_dsl(spa_t *spa)
866 {
867 	return (spa->spa_dsl_pool);
868 }
869 
870 blkptr_t *
871 spa_get_rootblkptr(spa_t *spa)
872 {
873 	return (&spa->spa_ubsync.ub_rootbp);
874 }
875 
876 void
877 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
878 {
879 	spa->spa_uberblock.ub_rootbp = *bp;
880 }
881 
882 void
883 spa_altroot(spa_t *spa, char *buf, size_t buflen)
884 {
885 	if (spa->spa_root == NULL)
886 		buf[0] = '\0';
887 	else
888 		(void) strncpy(buf, spa->spa_root, buflen);
889 }
890 
891 int
892 spa_sync_pass(spa_t *spa)
893 {
894 	return (spa->spa_sync_pass);
895 }
896 
897 char *
898 spa_name(spa_t *spa)
899 {
900 	/*
901 	 * Accessing the name requires holding either the namespace lock or the
902 	 * config lock, both of which are required to do a rename.
903 	 */
904 	ASSERT(MUTEX_HELD(&spa_namespace_lock) ||
905 	    spa_config_held(spa, RW_READER) || spa_config_held(spa, RW_WRITER));
906 
907 	return (spa->spa_name);
908 }
909 
910 uint64_t
911 spa_guid(spa_t *spa)
912 {
913 	/*
914 	 * If we fail to parse the config during spa_load(), we can go through
915 	 * the error path (which posts an ereport) and end up here with no root
916 	 * vdev.  We stash the original pool guid in 'spa_load_guid' to handle
917 	 * this case.
918 	 */
919 	if (spa->spa_root_vdev != NULL)
920 		return (spa->spa_root_vdev->vdev_guid);
921 	else
922 		return (spa->spa_load_guid);
923 }
924 
925 uint64_t
926 spa_last_synced_txg(spa_t *spa)
927 {
928 	return (spa->spa_ubsync.ub_txg);
929 }
930 
931 uint64_t
932 spa_first_txg(spa_t *spa)
933 {
934 	return (spa->spa_first_txg);
935 }
936 
937 int
938 spa_state(spa_t *spa)
939 {
940 	return (spa->spa_state);
941 }
942 
943 uint64_t
944 spa_freeze_txg(spa_t *spa)
945 {
946 	return (spa->spa_freeze_txg);
947 }
948 
949 /*
950  * Return how much space is allocated in the pool (ie. sum of all asize)
951  */
952 uint64_t
953 spa_get_alloc(spa_t *spa)
954 {
955 	return (spa->spa_root_vdev->vdev_stat.vs_alloc);
956 }
957 
958 /*
959  * Return how much (raid-z inflated) space there is in the pool.
960  */
961 uint64_t
962 spa_get_space(spa_t *spa)
963 {
964 	return (spa->spa_root_vdev->vdev_stat.vs_space);
965 }
966 
967 /*
968  * Return the amount of raid-z-deflated space in the pool.
969  */
970 uint64_t
971 spa_get_dspace(spa_t *spa)
972 {
973 	if (spa->spa_deflate)
974 		return (spa->spa_root_vdev->vdev_stat.vs_dspace);
975 	else
976 		return (spa->spa_root_vdev->vdev_stat.vs_space);
977 }
978 
979 /* ARGSUSED */
980 uint64_t
981 spa_get_asize(spa_t *spa, uint64_t lsize)
982 {
983 	/*
984 	 * For now, the worst case is 512-byte RAID-Z blocks, in which
985 	 * case the space requirement is exactly 2x; so just assume that.
986 	 * Add to this the fact that we can have up to 3 DVAs per bp, and
987 	 * we have to multiply by a total of 6x.
988 	 */
989 	return (lsize * 6);
990 }
991 
992 uint64_t
993 spa_version(spa_t *spa)
994 {
995 	return (spa->spa_ubsync.ub_version);
996 }
997 
998 int
999 spa_max_replication(spa_t *spa)
1000 {
1001 	/*
1002 	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1003 	 * handle BPs with more than one DVA allocated.  Set our max
1004 	 * replication level accordingly.
1005 	 */
1006 	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1007 		return (1);
1008 	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1009 }
1010 
1011 uint64_t
1012 bp_get_dasize(spa_t *spa, const blkptr_t *bp)
1013 {
1014 	int sz = 0, i;
1015 
1016 	if (!spa->spa_deflate)
1017 		return (BP_GET_ASIZE(bp));
1018 
1019 	spa_config_enter(spa, RW_READER, FTAG);
1020 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1021 		vdev_t *vd =
1022 		    vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[i]));
1023 		if (vd)
1024 			sz += (DVA_GET_ASIZE(&bp->blk_dva[i]) >>
1025 			    SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1026 	}
1027 	spa_config_exit(spa, FTAG);
1028 	return (sz);
1029 }
1030 
1031 /*
1032  * ==========================================================================
1033  * Initialization and Termination
1034  * ==========================================================================
1035  */
1036 
1037 static int
1038 spa_name_compare(const void *a1, const void *a2)
1039 {
1040 	const spa_t *s1 = a1;
1041 	const spa_t *s2 = a2;
1042 	int s;
1043 
1044 	s = strcmp(s1->spa_name, s2->spa_name);
1045 	if (s > 0)
1046 		return (1);
1047 	if (s < 0)
1048 		return (-1);
1049 	return (0);
1050 }
1051 
1052 int
1053 spa_busy(void)
1054 {
1055 	return (spa_active_count);
1056 }
1057 
1058 void
1059 spa_init(int mode)
1060 {
1061 	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1062 	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1063 	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1064 
1065 	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1066 	    offsetof(spa_t, spa_avl));
1067 
1068 	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_spare_t),
1069 	    offsetof(spa_spare_t, spare_avl));
1070 
1071 	spa_mode = mode;
1072 
1073 	refcount_init();
1074 	unique_init();
1075 	zio_init();
1076 	dmu_init();
1077 	zil_init();
1078 	zfs_prop_init();
1079 	zpool_prop_init();
1080 	spa_config_load();
1081 }
1082 
1083 void
1084 spa_fini(void)
1085 {
1086 	spa_evict_all();
1087 
1088 	zil_fini();
1089 	dmu_fini();
1090 	zio_fini();
1091 	unique_fini();
1092 	refcount_fini();
1093 
1094 	avl_destroy(&spa_namespace_avl);
1095 	avl_destroy(&spa_spare_avl);
1096 
1097 	cv_destroy(&spa_namespace_cv);
1098 	mutex_destroy(&spa_namespace_lock);
1099 	mutex_destroy(&spa_spare_lock);
1100 }
1101 
1102 /*
1103  * Return whether this pool has slogs. No locking needed.
1104  * It's not a problem if the wrong answer is returned as it's only for
1105  * performance and not correctness
1106  */
1107 boolean_t
1108 spa_has_slogs(spa_t *spa)
1109 {
1110 	return (spa->spa_log_class->mc_rotor != NULL);
1111 }
1112