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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 */
26
27 #include <sys/zfs_context.h>
28 #include <sys/spa_impl.h>
29 #include <sys/spa_boot.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/dsl_scan.h>
47 #include <sys/fs/zfs.h>
48 #include <sys/metaslab_impl.h>
49 #include <sys/arc.h>
50 #include <sys/ddt.h>
51 #include "zfs_prop.h"
52 #include "zfeature_common.h"
53
54 /*
55 * SPA locking
56 *
57 * There are four basic locks for managing spa_t structures:
58 *
59 * spa_namespace_lock (global mutex)
60 *
61 * This lock must be acquired to do any of the following:
62 *
63 * - Lookup a spa_t by name
64 * - Add or remove a spa_t from the namespace
65 * - Increase spa_refcount from non-zero
66 * - Check if spa_refcount is zero
67 * - Rename a spa_t
68 * - add/remove/attach/detach devices
69 * - Held for the duration of create/destroy/import/export
70 *
71 * It does not need to handle recursion. A create or destroy may
72 * reference objects (files or zvols) in other pools, but by
73 * definition they must have an existing reference, and will never need
74 * to lookup a spa_t by name.
75 *
76 * spa_refcount (per-spa refcount_t protected by mutex)
77 *
78 * This reference count keep track of any active users of the spa_t. The
79 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
80 * the refcount is never really 'zero' - opening a pool implicitly keeps
81 * some references in the DMU. Internally we check against spa_minref, but
82 * present the image of a zero/non-zero value to consumers.
83 *
84 * spa_config_lock[] (per-spa array of rwlocks)
85 *
86 * This protects the spa_t from config changes, and must be held in
87 * the following circumstances:
88 *
89 * - RW_READER to perform I/O to the spa
90 * - RW_WRITER to change the vdev config
91 *
92 * The locking order is fairly straightforward:
93 *
94 * spa_namespace_lock -> spa_refcount
95 *
96 * The namespace lock must be acquired to increase the refcount from 0
97 * or to check if it is zero.
98 *
99 * spa_refcount -> spa_config_lock[]
100 *
101 * There must be at least one valid reference on the spa_t to acquire
102 * the config lock.
103 *
104 * spa_namespace_lock -> spa_config_lock[]
105 *
106 * The namespace lock must always be taken before the config lock.
107 *
108 *
109 * The spa_namespace_lock can be acquired directly and is globally visible.
110 *
111 * The namespace is manipulated using the following functions, all of which
112 * require the spa_namespace_lock to be held.
113 *
114 * spa_lookup() Lookup a spa_t by name.
115 *
116 * spa_add() Create a new spa_t in the namespace.
117 *
118 * spa_remove() Remove a spa_t from the namespace. This also
119 * frees up any memory associated with the spa_t.
120 *
121 * spa_next() Returns the next spa_t in the system, or the
122 * first if NULL is passed.
123 *
124 * spa_evict_all() Shutdown and remove all spa_t structures in
125 * the system.
126 *
127 * spa_guid_exists() Determine whether a pool/device guid exists.
128 *
129 * The spa_refcount is manipulated using the following functions:
130 *
131 * spa_open_ref() Adds a reference to the given spa_t. Must be
132 * called with spa_namespace_lock held if the
133 * refcount is currently zero.
134 *
135 * spa_close() Remove a reference from the spa_t. This will
136 * not free the spa_t or remove it from the
137 * namespace. No locking is required.
138 *
139 * spa_refcount_zero() Returns true if the refcount is currently
140 * zero. Must be called with spa_namespace_lock
141 * held.
142 *
143 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
144 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
145 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
146 *
147 * To read the configuration, it suffices to hold one of these locks as reader.
148 * To modify the configuration, you must hold all locks as writer. To modify
149 * vdev state without altering the vdev tree's topology (e.g. online/offline),
150 * you must hold SCL_STATE and SCL_ZIO as writer.
151 *
152 * We use these distinct config locks to avoid recursive lock entry.
153 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
154 * block allocations (SCL_ALLOC), which may require reading space maps
155 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
156 *
157 * The spa config locks cannot be normal rwlocks because we need the
158 * ability to hand off ownership. For example, SCL_ZIO is acquired
159 * by the issuing thread and later released by an interrupt thread.
160 * They do, however, obey the usual write-wanted semantics to prevent
161 * writer (i.e. system administrator) starvation.
162 *
163 * The lock acquisition rules are as follows:
164 *
165 * SCL_CONFIG
166 * Protects changes to the vdev tree topology, such as vdev
167 * add/remove/attach/detach. Protects the dirty config list
168 * (spa_config_dirty_list) and the set of spares and l2arc devices.
169 *
170 * SCL_STATE
171 * Protects changes to pool state and vdev state, such as vdev
172 * online/offline/fault/degrade/clear. Protects the dirty state list
173 * (spa_state_dirty_list) and global pool state (spa_state).
174 *
175 * SCL_ALLOC
176 * Protects changes to metaslab groups and classes.
177 * Held as reader by metaslab_alloc() and metaslab_claim().
178 *
179 * SCL_ZIO
180 * Held by bp-level zios (those which have no io_vd upon entry)
181 * to prevent changes to the vdev tree. The bp-level zio implicitly
182 * protects all of its vdev child zios, which do not hold SCL_ZIO.
183 *
184 * SCL_FREE
185 * Protects changes to metaslab groups and classes.
186 * Held as reader by metaslab_free(). SCL_FREE is distinct from
187 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
188 * blocks in zio_done() while another i/o that holds either
189 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
190 *
191 * SCL_VDEV
192 * Held as reader to prevent changes to the vdev tree during trivial
193 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
194 * other locks, and lower than all of them, to ensure that it's safe
195 * to acquire regardless of caller context.
196 *
197 * In addition, the following rules apply:
198 *
199 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
200 * The lock ordering is SCL_CONFIG > spa_props_lock.
201 *
202 * (b) I/O operations on leaf vdevs. For any zio operation that takes
203 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
204 * or zio_write_phys() -- the caller must ensure that the config cannot
205 * cannot change in the interim, and that the vdev cannot be reopened.
206 * SCL_STATE as reader suffices for both.
207 *
208 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
209 *
210 * spa_vdev_enter() Acquire the namespace lock and the config lock
211 * for writing.
212 *
213 * spa_vdev_exit() Release the config lock, wait for all I/O
214 * to complete, sync the updated configs to the
215 * cache, and release the namespace lock.
216 *
217 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
218 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
219 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
220 *
221 * spa_rename() is also implemented within this file since it requires
222 * manipulation of the namespace.
223 */
224
225 static avl_tree_t spa_namespace_avl;
226 kmutex_t spa_namespace_lock;
227 static kcondvar_t spa_namespace_cv;
228 static int spa_active_count;
229 int spa_max_replication_override = SPA_DVAS_PER_BP;
230
231 static kmutex_t spa_spare_lock;
232 static avl_tree_t spa_spare_avl;
233 static kmutex_t spa_l2cache_lock;
234 static avl_tree_t spa_l2cache_avl;
235
236 kmem_cache_t *spa_buffer_pool;
237 int spa_mode_global;
238
239 #ifdef ZFS_DEBUG
240 /* Everything except dprintf and spa is on by default in debug builds */
241 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
242 #else
243 int zfs_flags = 0;
244 #endif
245
246 /*
247 * zfs_recover can be set to nonzero to attempt to recover from
248 * otherwise-fatal errors, typically caused by on-disk corruption. When
249 * set, calls to zfs_panic_recover() will turn into warning messages.
250 */
251 int zfs_recover = 0;
252
253 extern int zfs_txg_synctime_ms;
254
255 /*
256 * Expiration time in units of zfs_txg_synctime_ms. This value has two
257 * meanings. First it is used to determine when the spa_deadman logic
258 * should fire. By default the spa_deadman will fire if spa_sync has
259 * not completed in 1000 * zfs_txg_synctime_ms (i.e. 1000 seconds).
260 * Secondly, the value determines if an I/O is considered "hung".
261 * Any I/O that has not completed in zfs_deadman_synctime is considered
262 * "hung" resulting in a system panic.
263 */
264 uint64_t zfs_deadman_synctime = 1000ULL;
265
266 /*
267 * Override the zfs deadman behavior via /etc/system. By default the
268 * deadman is enabled except on VMware and sparc deployments.
269 */
270 int zfs_deadman_enabled = -1;
271
272
273 /*
274 * ==========================================================================
275 * SPA config locking
276 * ==========================================================================
277 */
278 static void
279 spa_config_lock_init(spa_t *spa)
280 {
281 for (int i = 0; i < SCL_LOCKS; i++) {
282 spa_config_lock_t *scl = &spa->spa_config_lock[i];
283 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
284 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
285 refcount_create_untracked(&scl->scl_count);
286 scl->scl_writer = NULL;
287 scl->scl_write_wanted = 0;
288 }
289 }
290
291 static void
292 spa_config_lock_destroy(spa_t *spa)
293 {
294 for (int i = 0; i < SCL_LOCKS; i++) {
295 spa_config_lock_t *scl = &spa->spa_config_lock[i];
296 mutex_destroy(&scl->scl_lock);
297 cv_destroy(&scl->scl_cv);
298 refcount_destroy(&scl->scl_count);
299 ASSERT(scl->scl_writer == NULL);
300 ASSERT(scl->scl_write_wanted == 0);
301 }
302 }
303
304 int
305 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
306 {
307 for (int i = 0; i < SCL_LOCKS; i++) {
308 spa_config_lock_t *scl = &spa->spa_config_lock[i];
309 if (!(locks & (1 << i)))
310 continue;
311 mutex_enter(&scl->scl_lock);
312 if (rw == RW_READER) {
313 if (scl->scl_writer || scl->scl_write_wanted) {
314 mutex_exit(&scl->scl_lock);
315 spa_config_exit(spa, locks ^ (1 << i), tag);
316 return (0);
317 }
318 } else {
319 ASSERT(scl->scl_writer != curthread);
320 if (!refcount_is_zero(&scl->scl_count)) {
321 mutex_exit(&scl->scl_lock);
322 spa_config_exit(spa, locks ^ (1 << i), tag);
323 return (0);
324 }
325 scl->scl_writer = curthread;
326 }
327 (void) refcount_add(&scl->scl_count, tag);
328 mutex_exit(&scl->scl_lock);
329 }
330 return (1);
331 }
332
333 void
334 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
335 {
336 int wlocks_held = 0;
337
338 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
339
340 for (int i = 0; i < SCL_LOCKS; i++) {
341 spa_config_lock_t *scl = &spa->spa_config_lock[i];
342 if (scl->scl_writer == curthread)
343 wlocks_held |= (1 << i);
344 if (!(locks & (1 << i)))
345 continue;
346 mutex_enter(&scl->scl_lock);
347 if (rw == RW_READER) {
348 while (scl->scl_writer || scl->scl_write_wanted) {
349 cv_wait(&scl->scl_cv, &scl->scl_lock);
350 }
351 } else {
352 ASSERT(scl->scl_writer != curthread);
353 while (!refcount_is_zero(&scl->scl_count)) {
354 scl->scl_write_wanted++;
355 cv_wait(&scl->scl_cv, &scl->scl_lock);
356 scl->scl_write_wanted--;
357 }
358 scl->scl_writer = curthread;
359 }
360 (void) refcount_add(&scl->scl_count, tag);
361 mutex_exit(&scl->scl_lock);
362 }
363 ASSERT(wlocks_held <= locks);
364 }
365
366 void
367 spa_config_exit(spa_t *spa, int locks, void *tag)
368 {
369 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
370 spa_config_lock_t *scl = &spa->spa_config_lock[i];
371 if (!(locks & (1 << i)))
372 continue;
373 mutex_enter(&scl->scl_lock);
374 ASSERT(!refcount_is_zero(&scl->scl_count));
375 if (refcount_remove(&scl->scl_count, tag) == 0) {
376 ASSERT(scl->scl_writer == NULL ||
377 scl->scl_writer == curthread);
378 scl->scl_writer = NULL; /* OK in either case */
379 cv_broadcast(&scl->scl_cv);
380 }
381 mutex_exit(&scl->scl_lock);
382 }
383 }
384
385 int
386 spa_config_held(spa_t *spa, int locks, krw_t rw)
387 {
388 int locks_held = 0;
389
390 for (int i = 0; i < SCL_LOCKS; i++) {
391 spa_config_lock_t *scl = &spa->spa_config_lock[i];
392 if (!(locks & (1 << i)))
393 continue;
394 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
395 (rw == RW_WRITER && scl->scl_writer == curthread))
396 locks_held |= 1 << i;
397 }
398
399 return (locks_held);
400 }
401
402 /*
403 * ==========================================================================
404 * SPA namespace functions
405 * ==========================================================================
406 */
407
408 /*
409 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
410 * Returns NULL if no matching spa_t is found.
411 */
412 spa_t *
413 spa_lookup(const char *name)
414 {
415 static spa_t search; /* spa_t is large; don't allocate on stack */
416 spa_t *spa;
417 avl_index_t where;
418 char *cp;
419
420 ASSERT(MUTEX_HELD(&spa_namespace_lock));
421
422 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
423
424 /*
425 * If it's a full dataset name, figure out the pool name and
426 * just use that.
427 */
428 cp = strpbrk(search.spa_name, "/@");
429 if (cp != NULL)
430 *cp = '\0';
431
432 spa = avl_find(&spa_namespace_avl, &search, &where);
433
434 return (spa);
435 }
436
437 /*
438 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
439 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
440 * looking for potentially hung I/Os.
441 */
442 void
443 spa_deadman(void *arg)
444 {
445 spa_t *spa = arg;
446
447 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
448 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
449 ++spa->spa_deadman_calls);
450 if (zfs_deadman_enabled)
451 vdev_deadman(spa->spa_root_vdev);
452 }
453
454 /*
455 * Create an uninitialized spa_t with the given name. Requires
456 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
457 * exist by calling spa_lookup() first.
458 */
459 spa_t *
460 spa_add(const char *name, nvlist_t *config, const char *altroot)
461 {
462 spa_t *spa;
463 spa_config_dirent_t *dp;
464 cyc_handler_t hdlr;
465 cyc_time_t when;
466
467 ASSERT(MUTEX_HELD(&spa_namespace_lock));
468
469 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
470
471 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
472 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
473 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
474 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
475 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
476 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
477 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
478 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
479 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
480 mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
481
482 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
483 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
484 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
485 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
486
487 for (int t = 0; t < TXG_SIZE; t++)
488 bplist_create(&spa->spa_free_bplist[t]);
489
490 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
491 spa->spa_state = POOL_STATE_UNINITIALIZED;
492 spa->spa_freeze_txg = UINT64_MAX;
493 spa->spa_final_txg = UINT64_MAX;
494 spa->spa_load_max_txg = UINT64_MAX;
495 spa->spa_proc = &p0;
496 spa->spa_proc_state = SPA_PROC_NONE;
497
498 hdlr.cyh_func = spa_deadman;
499 hdlr.cyh_arg = spa;
500 hdlr.cyh_level = CY_LOW_LEVEL;
501
502 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime *
503 zfs_txg_synctime_ms);
504
505 /*
506 * This determines how often we need to check for hung I/Os after
507 * the cyclic has already fired. Since checking for hung I/Os is
508 * an expensive operation we don't want to check too frequently.
509 * Instead wait for 5 synctimes before checking again.
510 */
511 when.cyt_interval = MSEC2NSEC(5 * zfs_txg_synctime_ms);
512 when.cyt_when = CY_INFINITY;
513 mutex_enter(&cpu_lock);
514 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
515 mutex_exit(&cpu_lock);
516
517 refcount_create(&spa->spa_refcount);
518 spa_config_lock_init(spa);
519
520 avl_add(&spa_namespace_avl, spa);
521
522 /*
523 * Set the alternate root, if there is one.
524 */
525 if (altroot) {
526 spa->spa_root = spa_strdup(altroot);
527 spa_active_count++;
528 }
529
530 /*
531 * Every pool starts with the default cachefile
532 */
533 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
534 offsetof(spa_config_dirent_t, scd_link));
535
536 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
537 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
538 list_insert_head(&spa->spa_config_list, dp);
539
540 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
541 KM_SLEEP) == 0);
542
543 if (config != NULL) {
544 nvlist_t *features;
545
546 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
547 &features) == 0) {
548 VERIFY(nvlist_dup(features, &spa->spa_label_features,
549 0) == 0);
550 }
551
552 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
553 }
554
555 if (spa->spa_label_features == NULL) {
556 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
557 KM_SLEEP) == 0);
558 }
559
560 spa->spa_iokstat = kstat_create("zfs", 0, name,
561 "disk", KSTAT_TYPE_IO, 1, 0);
562 if (spa->spa_iokstat) {
563 spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
564 kstat_install(spa->spa_iokstat);
565 }
566
567 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
568
569 return (spa);
570 }
571
572 /*
573 * Removes a spa_t from the namespace, freeing up any memory used. Requires
574 * spa_namespace_lock. This is called only after the spa_t has been closed and
575 * deactivated.
576 */
577 void
578 spa_remove(spa_t *spa)
579 {
580 spa_config_dirent_t *dp;
581
582 ASSERT(MUTEX_HELD(&spa_namespace_lock));
583 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
584
585 nvlist_free(spa->spa_config_splitting);
586
587 avl_remove(&spa_namespace_avl, spa);
588 cv_broadcast(&spa_namespace_cv);
589
590 if (spa->spa_root) {
591 spa_strfree(spa->spa_root);
592 spa_active_count--;
593 }
594
595 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
596 list_remove(&spa->spa_config_list, dp);
597 if (dp->scd_path != NULL)
598 spa_strfree(dp->scd_path);
599 kmem_free(dp, sizeof (spa_config_dirent_t));
600 }
601
602 list_destroy(&spa->spa_config_list);
603
604 nvlist_free(spa->spa_label_features);
605 nvlist_free(spa->spa_load_info);
606 spa_config_set(spa, NULL);
607
608 mutex_enter(&cpu_lock);
609 if (spa->spa_deadman_cycid != CYCLIC_NONE)
610 cyclic_remove(spa->spa_deadman_cycid);
611 mutex_exit(&cpu_lock);
612 spa->spa_deadman_cycid = CYCLIC_NONE;
613
614 refcount_destroy(&spa->spa_refcount);
615
616 spa_config_lock_destroy(spa);
617
618 kstat_delete(spa->spa_iokstat);
619 spa->spa_iokstat = NULL;
620
621 for (int t = 0; t < TXG_SIZE; t++)
622 bplist_destroy(&spa->spa_free_bplist[t]);
623
624 cv_destroy(&spa->spa_async_cv);
625 cv_destroy(&spa->spa_proc_cv);
626 cv_destroy(&spa->spa_scrub_io_cv);
627 cv_destroy(&spa->spa_suspend_cv);
628
629 mutex_destroy(&spa->spa_async_lock);
630 mutex_destroy(&spa->spa_errlist_lock);
631 mutex_destroy(&spa->spa_errlog_lock);
632 mutex_destroy(&spa->spa_history_lock);
633 mutex_destroy(&spa->spa_proc_lock);
634 mutex_destroy(&spa->spa_props_lock);
635 mutex_destroy(&spa->spa_scrub_lock);
636 mutex_destroy(&spa->spa_suspend_lock);
637 mutex_destroy(&spa->spa_vdev_top_lock);
638 mutex_destroy(&spa->spa_iokstat_lock);
639
640 kmem_free(spa, sizeof (spa_t));
641 }
642
643 /*
644 * Given a pool, return the next pool in the namespace, or NULL if there is
645 * none. If 'prev' is NULL, return the first pool.
646 */
647 spa_t *
648 spa_next(spa_t *prev)
649 {
650 ASSERT(MUTEX_HELD(&spa_namespace_lock));
651
652 if (prev)
653 return (AVL_NEXT(&spa_namespace_avl, prev));
654 else
655 return (avl_first(&spa_namespace_avl));
656 }
657
658 /*
659 * ==========================================================================
660 * SPA refcount functions
661 * ==========================================================================
662 */
663
664 /*
665 * Add a reference to the given spa_t. Must have at least one reference, or
666 * have the namespace lock held.
667 */
668 void
669 spa_open_ref(spa_t *spa, void *tag)
670 {
671 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
672 MUTEX_HELD(&spa_namespace_lock));
673 (void) refcount_add(&spa->spa_refcount, tag);
674 }
675
676 /*
677 * Remove a reference to the given spa_t. Must have at least one reference, or
678 * have the namespace lock held.
679 */
680 void
681 spa_close(spa_t *spa, void *tag)
682 {
683 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
684 MUTEX_HELD(&spa_namespace_lock));
685 (void) refcount_remove(&spa->spa_refcount, tag);
686 }
687
688 /*
689 * Check to see if the spa refcount is zero. Must be called with
690 * spa_namespace_lock held. We really compare against spa_minref, which is the
691 * number of references acquired when opening a pool
692 */
693 boolean_t
694 spa_refcount_zero(spa_t *spa)
695 {
696 ASSERT(MUTEX_HELD(&spa_namespace_lock));
697
698 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
699 }
700
701 /*
702 * ==========================================================================
703 * SPA spare and l2cache tracking
704 * ==========================================================================
705 */
706
707 /*
708 * Hot spares and cache devices are tracked using the same code below,
709 * for 'auxiliary' devices.
710 */
711
712 typedef struct spa_aux {
713 uint64_t aux_guid;
714 uint64_t aux_pool;
715 avl_node_t aux_avl;
716 int aux_count;
717 } spa_aux_t;
718
719 static int
720 spa_aux_compare(const void *a, const void *b)
721 {
722 const spa_aux_t *sa = a;
723 const spa_aux_t *sb = b;
724
725 if (sa->aux_guid < sb->aux_guid)
726 return (-1);
727 else if (sa->aux_guid > sb->aux_guid)
728 return (1);
729 else
730 return (0);
731 }
732
733 void
734 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
735 {
736 avl_index_t where;
737 spa_aux_t search;
738 spa_aux_t *aux;
739
740 search.aux_guid = vd->vdev_guid;
741 if ((aux = avl_find(avl, &search, &where)) != NULL) {
742 aux->aux_count++;
743 } else {
744 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
745 aux->aux_guid = vd->vdev_guid;
746 aux->aux_count = 1;
747 avl_insert(avl, aux, where);
748 }
749 }
750
751 void
752 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
753 {
754 spa_aux_t search;
755 spa_aux_t *aux;
756 avl_index_t where;
757
758 search.aux_guid = vd->vdev_guid;
759 aux = avl_find(avl, &search, &where);
760
761 ASSERT(aux != NULL);
762
763 if (--aux->aux_count == 0) {
764 avl_remove(avl, aux);
765 kmem_free(aux, sizeof (spa_aux_t));
766 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
767 aux->aux_pool = 0ULL;
768 }
769 }
770
771 boolean_t
772 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
773 {
774 spa_aux_t search, *found;
775
776 search.aux_guid = guid;
777 found = avl_find(avl, &search, NULL);
778
779 if (pool) {
780 if (found)
781 *pool = found->aux_pool;
782 else
783 *pool = 0ULL;
784 }
785
786 if (refcnt) {
787 if (found)
788 *refcnt = found->aux_count;
789 else
790 *refcnt = 0;
791 }
792
793 return (found != NULL);
794 }
795
796 void
797 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
798 {
799 spa_aux_t search, *found;
800 avl_index_t where;
801
802 search.aux_guid = vd->vdev_guid;
803 found = avl_find(avl, &search, &where);
804 ASSERT(found != NULL);
805 ASSERT(found->aux_pool == 0ULL);
806
807 found->aux_pool = spa_guid(vd->vdev_spa);
808 }
809
810 /*
811 * Spares are tracked globally due to the following constraints:
812 *
813 * - A spare may be part of multiple pools.
814 * - A spare may be added to a pool even if it's actively in use within
815 * another pool.
816 * - A spare in use in any pool can only be the source of a replacement if
817 * the target is a spare in the same pool.
818 *
819 * We keep track of all spares on the system through the use of a reference
820 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
821 * spare, then we bump the reference count in the AVL tree. In addition, we set
822 * the 'vdev_isspare' member to indicate that the device is a spare (active or
823 * inactive). When a spare is made active (used to replace a device in the
824 * pool), we also keep track of which pool its been made a part of.
825 *
826 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
827 * called under the spa_namespace lock as part of vdev reconfiguration. The
828 * separate spare lock exists for the status query path, which does not need to
829 * be completely consistent with respect to other vdev configuration changes.
830 */
831
832 static int
833 spa_spare_compare(const void *a, const void *b)
834 {
835 return (spa_aux_compare(a, b));
836 }
837
838 void
839 spa_spare_add(vdev_t *vd)
840 {
841 mutex_enter(&spa_spare_lock);
842 ASSERT(!vd->vdev_isspare);
843 spa_aux_add(vd, &spa_spare_avl);
844 vd->vdev_isspare = B_TRUE;
845 mutex_exit(&spa_spare_lock);
846 }
847
848 void
849 spa_spare_remove(vdev_t *vd)
850 {
851 mutex_enter(&spa_spare_lock);
852 ASSERT(vd->vdev_isspare);
853 spa_aux_remove(vd, &spa_spare_avl);
854 vd->vdev_isspare = B_FALSE;
855 mutex_exit(&spa_spare_lock);
856 }
857
858 boolean_t
859 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
860 {
861 boolean_t found;
862
863 mutex_enter(&spa_spare_lock);
864 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
865 mutex_exit(&spa_spare_lock);
866
867 return (found);
868 }
869
870 void
871 spa_spare_activate(vdev_t *vd)
872 {
873 mutex_enter(&spa_spare_lock);
874 ASSERT(vd->vdev_isspare);
875 spa_aux_activate(vd, &spa_spare_avl);
876 mutex_exit(&spa_spare_lock);
877 }
878
879 /*
880 * Level 2 ARC devices are tracked globally for the same reasons as spares.
881 * Cache devices currently only support one pool per cache device, and so
882 * for these devices the aux reference count is currently unused beyond 1.
883 */
884
885 static int
886 spa_l2cache_compare(const void *a, const void *b)
887 {
888 return (spa_aux_compare(a, b));
889 }
890
891 void
892 spa_l2cache_add(vdev_t *vd)
893 {
894 mutex_enter(&spa_l2cache_lock);
895 ASSERT(!vd->vdev_isl2cache);
896 spa_aux_add(vd, &spa_l2cache_avl);
897 vd->vdev_isl2cache = B_TRUE;
898 mutex_exit(&spa_l2cache_lock);
899 }
900
901 void
902 spa_l2cache_remove(vdev_t *vd)
903 {
904 mutex_enter(&spa_l2cache_lock);
905 ASSERT(vd->vdev_isl2cache);
906 spa_aux_remove(vd, &spa_l2cache_avl);
907 vd->vdev_isl2cache = B_FALSE;
908 mutex_exit(&spa_l2cache_lock);
909 }
910
911 boolean_t
912 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
913 {
914 boolean_t found;
915
916 mutex_enter(&spa_l2cache_lock);
917 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
918 mutex_exit(&spa_l2cache_lock);
919
920 return (found);
921 }
922
923 void
924 spa_l2cache_activate(vdev_t *vd)
925 {
926 mutex_enter(&spa_l2cache_lock);
927 ASSERT(vd->vdev_isl2cache);
928 spa_aux_activate(vd, &spa_l2cache_avl);
929 mutex_exit(&spa_l2cache_lock);
930 }
931
932 /*
933 * ==========================================================================
934 * SPA vdev locking
935 * ==========================================================================
936 */
937
938 /*
939 * Lock the given spa_t for the purpose of adding or removing a vdev.
940 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
941 * It returns the next transaction group for the spa_t.
942 */
943 uint64_t
944 spa_vdev_enter(spa_t *spa)
945 {
946 mutex_enter(&spa->spa_vdev_top_lock);
947 mutex_enter(&spa_namespace_lock);
948 return (spa_vdev_config_enter(spa));
949 }
950
951 /*
952 * Internal implementation for spa_vdev_enter(). Used when a vdev
953 * operation requires multiple syncs (i.e. removing a device) while
954 * keeping the spa_namespace_lock held.
955 */
956 uint64_t
957 spa_vdev_config_enter(spa_t *spa)
958 {
959 ASSERT(MUTEX_HELD(&spa_namespace_lock));
960
961 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
962
963 return (spa_last_synced_txg(spa) + 1);
964 }
965
966 /*
967 * Used in combination with spa_vdev_config_enter() to allow the syncing
968 * of multiple transactions without releasing the spa_namespace_lock.
969 */
970 void
971 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
972 {
973 ASSERT(MUTEX_HELD(&spa_namespace_lock));
974
975 int config_changed = B_FALSE;
976
977 ASSERT(txg > spa_last_synced_txg(spa));
978
979 spa->spa_pending_vdev = NULL;
980
981 /*
982 * Reassess the DTLs.
983 */
984 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
985
986 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
987 config_changed = B_TRUE;
988 spa->spa_config_generation++;
989 }
990
991 /*
992 * Verify the metaslab classes.
993 */
994 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
995 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
996
997 spa_config_exit(spa, SCL_ALL, spa);
998
999 /*
1000 * Panic the system if the specified tag requires it. This
1001 * is useful for ensuring that configurations are updated
1002 * transactionally.
1003 */
1004 if (zio_injection_enabled)
1005 zio_handle_panic_injection(spa, tag, 0);
1006
1007 /*
1008 * Note: this txg_wait_synced() is important because it ensures
1009 * that there won't be more than one config change per txg.
1010 * This allows us to use the txg as the generation number.
1011 */
1012 if (error == 0)
1013 txg_wait_synced(spa->spa_dsl_pool, txg);
1014
1015 if (vd != NULL) {
1016 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
1017 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1018 vdev_free(vd);
1019 spa_config_exit(spa, SCL_ALL, spa);
1020 }
1021
1022 /*
1023 * If the config changed, update the config cache.
1024 */
1025 if (config_changed)
1026 spa_config_sync(spa, B_FALSE, B_TRUE);
1027 }
1028
1029 /*
1030 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1031 * locking of spa_vdev_enter(), we also want make sure the transactions have
1032 * synced to disk, and then update the global configuration cache with the new
1033 * information.
1034 */
1035 int
1036 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1037 {
1038 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1039 mutex_exit(&spa_namespace_lock);
1040 mutex_exit(&spa->spa_vdev_top_lock);
1041
1042 return (error);
1043 }
1044
1045 /*
1046 * Lock the given spa_t for the purpose of changing vdev state.
1047 */
1048 void
1049 spa_vdev_state_enter(spa_t *spa, int oplocks)
1050 {
1051 int locks = SCL_STATE_ALL | oplocks;
1052
1053 /*
1054 * Root pools may need to read of the underlying devfs filesystem
1055 * when opening up a vdev. Unfortunately if we're holding the
1056 * SCL_ZIO lock it will result in a deadlock when we try to issue
1057 * the read from the root filesystem. Instead we "prefetch"
1058 * the associated vnodes that we need prior to opening the
1059 * underlying devices and cache them so that we can prevent
1060 * any I/O when we are doing the actual open.
1061 */
1062 if (spa_is_root(spa)) {
1063 int low = locks & ~(SCL_ZIO - 1);
1064 int high = locks & ~low;
1065
1066 spa_config_enter(spa, high, spa, RW_WRITER);
1067 vdev_hold(spa->spa_root_vdev);
1068 spa_config_enter(spa, low, spa, RW_WRITER);
1069 } else {
1070 spa_config_enter(spa, locks, spa, RW_WRITER);
1071 }
1072 spa->spa_vdev_locks = locks;
1073 }
1074
1075 int
1076 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1077 {
1078 boolean_t config_changed = B_FALSE;
1079
1080 if (vd != NULL || error == 0)
1081 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1082 0, 0, B_FALSE);
1083
1084 if (vd != NULL) {
1085 vdev_state_dirty(vd->vdev_top);
1086 config_changed = B_TRUE;
1087 spa->spa_config_generation++;
1088 }
1089
1090 if (spa_is_root(spa))
1091 vdev_rele(spa->spa_root_vdev);
1092
1093 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1094 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1095
1096 /*
1097 * If anything changed, wait for it to sync. This ensures that,
1098 * from the system administrator's perspective, zpool(1M) commands
1099 * are synchronous. This is important for things like zpool offline:
1100 * when the command completes, you expect no further I/O from ZFS.
1101 */
1102 if (vd != NULL)
1103 txg_wait_synced(spa->spa_dsl_pool, 0);
1104
1105 /*
1106 * If the config changed, update the config cache.
1107 */
1108 if (config_changed) {
1109 mutex_enter(&spa_namespace_lock);
1110 spa_config_sync(spa, B_FALSE, B_TRUE);
1111 mutex_exit(&spa_namespace_lock);
1112 }
1113
1114 return (error);
1115 }
1116
1117 /*
1118 * ==========================================================================
1119 * Miscellaneous functions
1120 * ==========================================================================
1121 */
1122
1123 void
1124 spa_activate_mos_feature(spa_t *spa, const char *feature)
1125 {
1126 (void) nvlist_add_boolean(spa->spa_label_features, feature);
1127 vdev_config_dirty(spa->spa_root_vdev);
1128 }
1129
1130 void
1131 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1132 {
1133 (void) nvlist_remove_all(spa->spa_label_features, feature);
1134 vdev_config_dirty(spa->spa_root_vdev);
1135 }
1136
1137 /*
1138 * Rename a spa_t.
1139 */
1140 int
1141 spa_rename(const char *name, const char *newname)
1142 {
1143 spa_t *spa;
1144 int err;
1145
1146 /*
1147 * Lookup the spa_t and grab the config lock for writing. We need to
1148 * actually open the pool so that we can sync out the necessary labels.
1149 * It's OK to call spa_open() with the namespace lock held because we
1150 * allow recursive calls for other reasons.
1151 */
1152 mutex_enter(&spa_namespace_lock);
1153 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1154 mutex_exit(&spa_namespace_lock);
1155 return (err);
1156 }
1157
1158 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1159
1160 avl_remove(&spa_namespace_avl, spa);
1161 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1162 avl_add(&spa_namespace_avl, spa);
1163
1164 /*
1165 * Sync all labels to disk with the new names by marking the root vdev
1166 * dirty and waiting for it to sync. It will pick up the new pool name
1167 * during the sync.
1168 */
1169 vdev_config_dirty(spa->spa_root_vdev);
1170
1171 spa_config_exit(spa, SCL_ALL, FTAG);
1172
1173 txg_wait_synced(spa->spa_dsl_pool, 0);
1174
1175 /*
1176 * Sync the updated config cache.
1177 */
1178 spa_config_sync(spa, B_FALSE, B_TRUE);
1179
1180 spa_close(spa, FTAG);
1181
1182 mutex_exit(&spa_namespace_lock);
1183
1184 return (0);
1185 }
1186
1187 /*
1188 * Return the spa_t associated with given pool_guid, if it exists. If
1189 * device_guid is non-zero, determine whether the pool exists *and* contains
1190 * a device with the specified device_guid.
1191 */
1192 spa_t *
1193 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1194 {
1195 spa_t *spa;
1196 avl_tree_t *t = &spa_namespace_avl;
1197
1198 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1199
1200 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1201 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1202 continue;
1203 if (spa->spa_root_vdev == NULL)
1204 continue;
1205 if (spa_guid(spa) == pool_guid) {
1206 if (device_guid == 0)
1207 break;
1208
1209 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1210 device_guid) != NULL)
1211 break;
1212
1213 /*
1214 * Check any devices we may be in the process of adding.
1215 */
1216 if (spa->spa_pending_vdev) {
1217 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1218 device_guid) != NULL)
1219 break;
1220 }
1221 }
1222 }
1223
1224 return (spa);
1225 }
1226
1227 /*
1228 * Determine whether a pool with the given pool_guid exists.
1229 */
1230 boolean_t
1231 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1232 {
1233 return (spa_by_guid(pool_guid, device_guid) != NULL);
1234 }
1235
1236 char *
1237 spa_strdup(const char *s)
1238 {
1239 size_t len;
1240 char *new;
1241
1242 len = strlen(s);
1243 new = kmem_alloc(len + 1, KM_SLEEP);
1244 bcopy(s, new, len);
1245 new[len] = '\0';
1246
1247 return (new);
1248 }
1249
1250 void
1251 spa_strfree(char *s)
1252 {
1253 kmem_free(s, strlen(s) + 1);
1254 }
1255
1256 uint64_t
1257 spa_get_random(uint64_t range)
1258 {
1259 uint64_t r;
1260
1261 ASSERT(range != 0);
1262
1263 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1264
1265 return (r % range);
1266 }
1267
1268 uint64_t
1269 spa_generate_guid(spa_t *spa)
1270 {
1271 uint64_t guid = spa_get_random(-1ULL);
1272
1273 if (spa != NULL) {
1274 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1275 guid = spa_get_random(-1ULL);
1276 } else {
1277 while (guid == 0 || spa_guid_exists(guid, 0))
1278 guid = spa_get_random(-1ULL);
1279 }
1280
1281 return (guid);
1282 }
1283
1284 void
1285 sprintf_blkptr(char *buf, const blkptr_t *bp)
1286 {
1287 char type[256];
1288 char *checksum = NULL;
1289 char *compress = NULL;
1290
1291 if (bp != NULL) {
1292 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1293 dmu_object_byteswap_t bswap =
1294 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1295 (void) snprintf(type, sizeof (type), "bswap %s %s",
1296 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1297 "metadata" : "data",
1298 dmu_ot_byteswap[bswap].ob_name);
1299 } else {
1300 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1301 sizeof (type));
1302 }
1303 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1304 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1305 }
1306
1307 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1308 }
1309
1310 void
1311 spa_freeze(spa_t *spa)
1312 {
1313 uint64_t freeze_txg = 0;
1314
1315 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1316 if (spa->spa_freeze_txg == UINT64_MAX) {
1317 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1318 spa->spa_freeze_txg = freeze_txg;
1319 }
1320 spa_config_exit(spa, SCL_ALL, FTAG);
1321 if (freeze_txg != 0)
1322 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1323 }
1324
1325 void
1326 zfs_panic_recover(const char *fmt, ...)
1327 {
1328 va_list adx;
1329
1330 va_start(adx, fmt);
1331 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1332 va_end(adx);
1333 }
1334
1335 /*
1336 * This is a stripped-down version of strtoull, suitable only for converting
1337 * lowercase hexadecimal numbers that don't overflow.
1338 */
1339 uint64_t
1340 strtonum(const char *str, char **nptr)
1341 {
1342 uint64_t val = 0;
1343 char c;
1344 int digit;
1345
1346 while ((c = *str) != '\0') {
1347 if (c >= '0' && c <= '9')
1348 digit = c - '0';
1349 else if (c >= 'a' && c <= 'f')
1350 digit = 10 + c - 'a';
1351 else
1352 break;
1353
1354 val *= 16;
1355 val += digit;
1356
1357 str++;
1358 }
1359
1360 if (nptr)
1361 *nptr = (char *)str;
1362
1363 return (val);
1364 }
1365
1366 /*
1367 * ==========================================================================
1368 * Accessor functions
1369 * ==========================================================================
1370 */
1371
1372 boolean_t
1373 spa_shutting_down(spa_t *spa)
1374 {
1375 return (spa->spa_async_suspended);
1376 }
1377
1378 dsl_pool_t *
1379 spa_get_dsl(spa_t *spa)
1380 {
1381 return (spa->spa_dsl_pool);
1382 }
1383
1384 boolean_t
1385 spa_is_initializing(spa_t *spa)
1386 {
1387 return (spa->spa_is_initializing);
1388 }
1389
1390 blkptr_t *
1391 spa_get_rootblkptr(spa_t *spa)
1392 {
1393 return (&spa->spa_ubsync.ub_rootbp);
1394 }
1395
1396 void
1397 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1398 {
1399 spa->spa_uberblock.ub_rootbp = *bp;
1400 }
1401
1402 void
1403 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1404 {
1405 if (spa->spa_root == NULL)
1406 buf[0] = '\0';
1407 else
1408 (void) strncpy(buf, spa->spa_root, buflen);
1409 }
1410
1411 int
1412 spa_sync_pass(spa_t *spa)
1413 {
1414 return (spa->spa_sync_pass);
1415 }
1416
1417 char *
1418 spa_name(spa_t *spa)
1419 {
1420 return (spa->spa_name);
1421 }
1422
1423 uint64_t
1424 spa_guid(spa_t *spa)
1425 {
1426 dsl_pool_t *dp = spa_get_dsl(spa);
1427 uint64_t guid;
1428
1429 /*
1430 * If we fail to parse the config during spa_load(), we can go through
1431 * the error path (which posts an ereport) and end up here with no root
1432 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1433 * this case.
1434 */
1435 if (spa->spa_root_vdev == NULL)
1436 return (spa->spa_config_guid);
1437
1438 guid = spa->spa_last_synced_guid != 0 ?
1439 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1440
1441 /*
1442 * Return the most recently synced out guid unless we're
1443 * in syncing context.
1444 */
1445 if (dp && dsl_pool_sync_context(dp))
1446 return (spa->spa_root_vdev->vdev_guid);
1447 else
1448 return (guid);
1449 }
1450
1451 uint64_t
1452 spa_load_guid(spa_t *spa)
1453 {
1454 /*
1455 * This is a GUID that exists solely as a reference for the
1456 * purposes of the arc. It is generated at load time, and
1457 * is never written to persistent storage.
1458 */
1459 return (spa->spa_load_guid);
1460 }
1461
1462 uint64_t
1463 spa_last_synced_txg(spa_t *spa)
1464 {
1465 return (spa->spa_ubsync.ub_txg);
1466 }
1467
1468 uint64_t
1469 spa_first_txg(spa_t *spa)
1470 {
1471 return (spa->spa_first_txg);
1472 }
1473
1474 uint64_t
1475 spa_syncing_txg(spa_t *spa)
1476 {
1477 return (spa->spa_syncing_txg);
1478 }
1479
1480 pool_state_t
1481 spa_state(spa_t *spa)
1482 {
1483 return (spa->spa_state);
1484 }
1485
1486 spa_load_state_t
1487 spa_load_state(spa_t *spa)
1488 {
1489 return (spa->spa_load_state);
1490 }
1491
1492 uint64_t
1493 spa_freeze_txg(spa_t *spa)
1494 {
1495 return (spa->spa_freeze_txg);
1496 }
1497
1498 /* ARGSUSED */
1499 uint64_t
1500 spa_get_asize(spa_t *spa, uint64_t lsize)
1501 {
1502 /*
1503 * The worst case is single-sector max-parity RAID-Z blocks, in which
1504 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1505 * times the size; so just assume that. Add to this the fact that
1506 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1507 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1508 */
1509 return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1510 }
1511
1512 uint64_t
1513 spa_get_dspace(spa_t *spa)
1514 {
1515 return (spa->spa_dspace);
1516 }
1517
1518 void
1519 spa_update_dspace(spa_t *spa)
1520 {
1521 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1522 ddt_get_dedup_dspace(spa);
1523 }
1524
1525 /*
1526 * Return the failure mode that has been set to this pool. The default
1527 * behavior will be to block all I/Os when a complete failure occurs.
1528 */
1529 uint8_t
1530 spa_get_failmode(spa_t *spa)
1531 {
1532 return (spa->spa_failmode);
1533 }
1534
1535 boolean_t
1536 spa_suspended(spa_t *spa)
1537 {
1538 return (spa->spa_suspended);
1539 }
1540
1541 uint64_t
1542 spa_version(spa_t *spa)
1543 {
1544 return (spa->spa_ubsync.ub_version);
1545 }
1546
1547 boolean_t
1548 spa_deflate(spa_t *spa)
1549 {
1550 return (spa->spa_deflate);
1551 }
1552
1553 metaslab_class_t *
1554 spa_normal_class(spa_t *spa)
1555 {
1556 return (spa->spa_normal_class);
1557 }
1558
1559 metaslab_class_t *
1560 spa_log_class(spa_t *spa)
1561 {
1562 return (spa->spa_log_class);
1563 }
1564
1565 int
1566 spa_max_replication(spa_t *spa)
1567 {
1568 /*
1569 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1570 * handle BPs with more than one DVA allocated. Set our max
1571 * replication level accordingly.
1572 */
1573 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1574 return (1);
1575 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1576 }
1577
1578 int
1579 spa_prev_software_version(spa_t *spa)
1580 {
1581 return (spa->spa_prev_software_version);
1582 }
1583
1584 uint64_t
1585 spa_deadman_synctime(spa_t *spa)
1586 {
1587 return (spa->spa_deadman_synctime);
1588 }
1589
1590 uint64_t
1591 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1592 {
1593 uint64_t asize = DVA_GET_ASIZE(dva);
1594 uint64_t dsize = asize;
1595
1596 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1597
1598 if (asize != 0 && spa->spa_deflate) {
1599 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1600 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1601 }
1602
1603 return (dsize);
1604 }
1605
1606 uint64_t
1607 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1608 {
1609 uint64_t dsize = 0;
1610
1611 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1612 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1613
1614 return (dsize);
1615 }
1616
1617 uint64_t
1618 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1619 {
1620 uint64_t dsize = 0;
1621
1622 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1623
1624 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1625 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1626
1627 spa_config_exit(spa, SCL_VDEV, FTAG);
1628
1629 return (dsize);
1630 }
1631
1632 /*
1633 * ==========================================================================
1634 * Initialization and Termination
1635 * ==========================================================================
1636 */
1637
1638 static int
1639 spa_name_compare(const void *a1, const void *a2)
1640 {
1641 const spa_t *s1 = a1;
1642 const spa_t *s2 = a2;
1643 int s;
1644
1645 s = strcmp(s1->spa_name, s2->spa_name);
1646 if (s > 0)
1647 return (1);
1648 if (s < 0)
1649 return (-1);
1650 return (0);
1651 }
1652
1653 int
1654 spa_busy(void)
1655 {
1656 return (spa_active_count);
1657 }
1658
1659 void
1660 spa_boot_init()
1661 {
1662 spa_config_load();
1663 }
1664
1665 void
1666 spa_init(int mode)
1667 {
1668 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1669 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1670 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1671 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1672
1673 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1674 offsetof(spa_t, spa_avl));
1675
1676 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1677 offsetof(spa_aux_t, aux_avl));
1678
1679 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1680 offsetof(spa_aux_t, aux_avl));
1681
1682 spa_mode_global = mode;
1683
1684 #ifdef _KERNEL
1685 spa_arch_init();
1686 #else
1687 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1688 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1689 if (arc_procfd == -1) {
1690 perror("could not enable watchpoints: "
1691 "opening /proc/self/ctl failed: ");
1692 } else {
1693 arc_watch = B_TRUE;
1694 }
1695 }
1696 #endif
1697
1698 refcount_init();
1699 unique_init();
1700 space_map_init();
1701 zio_init();
1702 dmu_init();
1703 zil_init();
1704 vdev_cache_stat_init();
1705 zfs_prop_init();
1706 zpool_prop_init();
1707 zpool_feature_init();
1708 spa_config_load();
1709 l2arc_start();
1710 }
1711
1712 void
1713 spa_fini(void)
1714 {
1715 l2arc_stop();
1716
1717 spa_evict_all();
1718
1719 vdev_cache_stat_fini();
1720 zil_fini();
1721 dmu_fini();
1722 zio_fini();
1723 space_map_fini();
1724 unique_fini();
1725 refcount_fini();
1726
1727 avl_destroy(&spa_namespace_avl);
1728 avl_destroy(&spa_spare_avl);
1729 avl_destroy(&spa_l2cache_avl);
1730
1731 cv_destroy(&spa_namespace_cv);
1732 mutex_destroy(&spa_namespace_lock);
1733 mutex_destroy(&spa_spare_lock);
1734 mutex_destroy(&spa_l2cache_lock);
1735 }
1736
1737 /*
1738 * Return whether this pool has slogs. No locking needed.
1739 * It's not a problem if the wrong answer is returned as it's only for
1740 * performance and not correctness
1741 */
1742 boolean_t
1743 spa_has_slogs(spa_t *spa)
1744 {
1745 return (spa->spa_log_class->mc_rotor != NULL);
1746 }
1747
1748 spa_log_state_t
1749 spa_get_log_state(spa_t *spa)
1750 {
1751 return (spa->spa_log_state);
1752 }
1753
1754 void
1755 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1756 {
1757 spa->spa_log_state = state;
1758 }
1759
1760 boolean_t
1761 spa_is_root(spa_t *spa)
1762 {
1763 return (spa->spa_is_root);
1764 }
1765
1766 boolean_t
1767 spa_writeable(spa_t *spa)
1768 {
1769 return (!!(spa->spa_mode & FWRITE));
1770 }
1771
1772 int
1773 spa_mode(spa_t *spa)
1774 {
1775 return (spa->spa_mode);
1776 }
1777
1778 uint64_t
1779 spa_bootfs(spa_t *spa)
1780 {
1781 return (spa->spa_bootfs);
1782 }
1783
1784 uint64_t
1785 spa_delegation(spa_t *spa)
1786 {
1787 return (spa->spa_delegation);
1788 }
1789
1790 objset_t *
1791 spa_meta_objset(spa_t *spa)
1792 {
1793 return (spa->spa_meta_objset);
1794 }
1795
1796 enum zio_checksum
1797 spa_dedup_checksum(spa_t *spa)
1798 {
1799 return (spa->spa_dedup_checksum);
1800 }
1801
1802 /*
1803 * Reset pool scan stat per scan pass (or reboot).
1804 */
1805 void
1806 spa_scan_stat_init(spa_t *spa)
1807 {
1808 /* data not stored on disk */
1809 spa->spa_scan_pass_start = gethrestime_sec();
1810 spa->spa_scan_pass_exam = 0;
1811 vdev_scan_stat_init(spa->spa_root_vdev);
1812 }
1813
1814 /*
1815 * Get scan stats for zpool status reports
1816 */
1817 int
1818 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1819 {
1820 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1821
1822 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1823 return (SET_ERROR(ENOENT));
1824 bzero(ps, sizeof (pool_scan_stat_t));
1825
1826 /* data stored on disk */
1827 ps->pss_func = scn->scn_phys.scn_func;
1828 ps->pss_start_time = scn->scn_phys.scn_start_time;
1829 ps->pss_end_time = scn->scn_phys.scn_end_time;
1830 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1831 ps->pss_examined = scn->scn_phys.scn_examined;
1832 ps->pss_to_process = scn->scn_phys.scn_to_process;
1833 ps->pss_processed = scn->scn_phys.scn_processed;
1834 ps->pss_errors = scn->scn_phys.scn_errors;
1835 ps->pss_state = scn->scn_phys.scn_state;
1836
1837 /* data not stored on disk */
1838 ps->pss_pass_start = spa->spa_scan_pass_start;
1839 ps->pss_pass_exam = spa->spa_scan_pass_exam;
1840
1841 return (0);
1842 }
1843
1844 boolean_t
1845 spa_debug_enabled(spa_t *spa)
1846 {
1847 return (spa->spa_debug);
1848 }