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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
30 #include <sys/dsl_pool.h>
31 #include <sys/dsl_dataset.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_synctask.h>
35 #include <sys/dsl_scan.h>
36 #include <sys/dnode.h>
37 #include <sys/dmu_tx.h>
38 #include <sys/dmu_objset.h>
42 #include <sys/zfs_context.h>
43 #include <sys/fs/zfs.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/spa_impl.h>
46 #include <sys/dsl_deadlist.h>
47 #include <sys/vdev_impl.h>
48 #include <sys/metaslab_impl.h>
49 #include <sys/bptree.h>
50 #include <sys/zfeature.h>
51 #include <sys/zil_impl.h>
52 #include <sys/dsl_userhold.h>
58 * ZFS must limit the rate of incoming writes to the rate at which it is able
59 * to sync data modifications to the backend storage. Throttling by too much
60 * creates an artificial limit; throttling by too little can only be sustained
61 * for short periods and would lead to highly lumpy performance. On a per-pool
62 * basis, ZFS tracks the amount of modified (dirty) data. As operations change
63 * data, the amount of dirty data increases; as ZFS syncs out data, the amount
64 * of dirty data decreases. When the amount of dirty data exceeds a
65 * predetermined threshold further modifications are blocked until the amount
66 * of dirty data decreases (as data is synced out).
68 * The limit on dirty data is tunable, and should be adjusted according to
69 * both the IO capacity and available memory of the system. The larger the
70 * window, the more ZFS is able to aggregate and amortize metadata (and data)
71 * changes. However, memory is a limited resource, and allowing for more dirty
72 * data comes at the cost of keeping other useful data in memory (for example
73 * ZFS data cached by the ARC).
77 * As buffers are modified dsl_pool_willuse_space() increments both the per-
78 * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of
79 * dirty space used; dsl_pool_dirty_space() decrements those values as data
80 * is synced out from dsl_pool_sync(). While only the poolwide value is
81 * relevant, the per-txg value is useful for debugging. The tunable
82 * zfs_dirty_data_max determines the dirty space limit. Once that value is
83 * exceeded, new writes are halted until space frees up.
85 * The zfs_dirty_data_sync tunable dictates the threshold at which we
86 * ensure that there is a txg syncing (see the comment in txg.c for a full
87 * description of transaction group stages).
89 * The IO scheduler uses both the dirty space limit and current amount of
90 * dirty data as inputs. Those values affect the number of concurrent IOs ZFS
91 * issues. See the comment in vdev_queue.c for details of the IO scheduler.
93 * The delay is also calculated based on the amount of dirty data. See the
94 * comment above dmu_tx_delay() for details.
98 * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory,
99 * capped at zfs_dirty_data_max_max. It can also be overridden in /etc/system.
101 uint64_t zfs_dirty_data_max
;
102 uint64_t zfs_dirty_data_max_max
= 4ULL * 1024 * 1024 * 1024;
103 int zfs_dirty_data_max_percent
= 10;
106 * If there's at least this much dirty data (as a percentage of
107 * zfs_dirty_data_max), push out a txg. This should be less than
108 * zfs_vdev_async_write_active_min_dirty_percent.
110 uint64_t zfs_dirty_data_sync_pct
= 20;
113 * Once there is this amount of dirty data, the dmu_tx_delay() will kick in
114 * and delay each transaction.
115 * This value should be >= zfs_vdev_async_write_active_max_dirty_percent.
117 int zfs_delay_min_dirty_percent
= 60;
120 * This controls how quickly the delay approaches infinity.
121 * Larger values cause it to delay more for a given amount of dirty data.
122 * Therefore larger values will cause there to be less dirty data for a
125 * For the smoothest delay, this value should be about 1 billion divided
126 * by the maximum number of operations per second. This will smoothly
127 * handle between 10x and 1/10th this number.
129 * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the
130 * multiply in dmu_tx_delay().
132 uint64_t zfs_delay_scale
= 1000 * 1000 * 1000 / 2000;
135 * This determines the number of threads used by the dp_sync_taskq.
137 int zfs_sync_taskq_batch_pct
= 75;
140 * These tunables determine the behavior of how zil_itxg_clean() is
141 * called via zil_clean() in the context of spa_sync(). When an itxg
142 * list needs to be cleaned, TQ_NOSLEEP will be used when dispatching.
143 * If the dispatch fails, the call to zil_itxg_clean() will occur
144 * synchronously in the context of spa_sync(), which can negatively
145 * impact the performance of spa_sync() (e.g. in the case of the itxg
146 * list having a large number of itxs that needs to be cleaned).
148 * Thus, these tunables can be used to manipulate the behavior of the
149 * taskq used by zil_clean(); they determine the number of taskq entries
150 * that are pre-populated when the taskq is first created (via the
151 * "zfs_zil_clean_taskq_minalloc" tunable) and the maximum number of
152 * taskq entries that are cached after an on-demand allocation (via the
153 * "zfs_zil_clean_taskq_maxalloc").
155 * The idea being, we want to try reasonably hard to ensure there will
156 * already be a taskq entry pre-allocated by the time that it is needed
157 * by zil_clean(). This way, we can avoid the possibility of an
158 * on-demand allocation of a new taskq entry from failing, which would
159 * result in zil_itxg_clean() being called synchronously from zil_clean()
160 * (which can adversely affect performance of spa_sync()).
162 * Additionally, the number of threads used by the taskq can be
163 * configured via the "zfs_zil_clean_taskq_nthr_pct" tunable.
165 int zfs_zil_clean_taskq_nthr_pct
= 100;
166 int zfs_zil_clean_taskq_minalloc
= 1024;
167 int zfs_zil_clean_taskq_maxalloc
= 1024 * 1024;
170 dsl_pool_open_special_dir(dsl_pool_t
*dp
, const char *name
, dsl_dir_t
**ddp
)
175 err
= zap_lookup(dp
->dp_meta_objset
,
176 dsl_dir_phys(dp
->dp_root_dir
)->dd_child_dir_zapobj
,
177 name
, sizeof (obj
), 1, &obj
);
181 return (dsl_dir_hold_obj(dp
, obj
, name
, dp
, ddp
));
185 dsl_pool_open_impl(spa_t
*spa
, uint64_t txg
)
188 blkptr_t
*bp
= spa_get_rootblkptr(spa
);
190 dp
= kmem_zalloc(sizeof (dsl_pool_t
), KM_SLEEP
);
192 dp
->dp_meta_rootbp
= *bp
;
193 rrw_init(&dp
->dp_config_rwlock
, B_TRUE
);
196 txg_list_create(&dp
->dp_dirty_datasets
, spa
,
197 offsetof(dsl_dataset_t
, ds_dirty_link
));
198 txg_list_create(&dp
->dp_dirty_zilogs
, spa
,
199 offsetof(zilog_t
, zl_dirty_link
));
200 txg_list_create(&dp
->dp_dirty_dirs
, spa
,
201 offsetof(dsl_dir_t
, dd_dirty_link
));
202 txg_list_create(&dp
->dp_sync_tasks
, spa
,
203 offsetof(dsl_sync_task_t
, dst_node
));
204 txg_list_create(&dp
->dp_early_sync_tasks
, spa
,
205 offsetof(dsl_sync_task_t
, dst_node
));
207 dp
->dp_sync_taskq
= taskq_create("dp_sync_taskq",
208 zfs_sync_taskq_batch_pct
, minclsyspri
, 1, INT_MAX
,
209 TASKQ_THREADS_CPU_PCT
);
211 dp
->dp_zil_clean_taskq
= taskq_create("dp_zil_clean_taskq",
212 zfs_zil_clean_taskq_nthr_pct
, minclsyspri
,
213 zfs_zil_clean_taskq_minalloc
,
214 zfs_zil_clean_taskq_maxalloc
,
215 TASKQ_PREPOPULATE
| TASKQ_THREADS_CPU_PCT
);
217 mutex_init(&dp
->dp_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
218 cv_init(&dp
->dp_spaceavail_cv
, NULL
, CV_DEFAULT
, NULL
);
220 dp
->dp_vnrele_taskq
= taskq_create("zfs_vn_rele_taskq", 1, minclsyspri
,
227 dsl_pool_init(spa_t
*spa
, uint64_t txg
, dsl_pool_t
**dpp
)
230 dsl_pool_t
*dp
= dsl_pool_open_impl(spa
, txg
);
232 err
= dmu_objset_open_impl(spa
, NULL
, &dp
->dp_meta_rootbp
,
233 &dp
->dp_meta_objset
);
243 dsl_pool_open(dsl_pool_t
*dp
)
250 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
251 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
252 DMU_POOL_ROOT_DATASET
, sizeof (uint64_t), 1,
253 &dp
->dp_root_dir_obj
);
257 err
= dsl_dir_hold_obj(dp
, dp
->dp_root_dir_obj
,
258 NULL
, dp
, &dp
->dp_root_dir
);
262 err
= dsl_pool_open_special_dir(dp
, MOS_DIR_NAME
, &dp
->dp_mos_dir
);
266 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_ORIGIN
) {
267 err
= dsl_pool_open_special_dir(dp
, ORIGIN_DIR_NAME
, &dd
);
270 err
= dsl_dataset_hold_obj(dp
,
271 dsl_dir_phys(dd
)->dd_head_dataset_obj
, FTAG
, &ds
);
273 err
= dsl_dataset_hold_obj(dp
,
274 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, dp
,
275 &dp
->dp_origin_snap
);
276 dsl_dataset_rele(ds
, FTAG
);
278 dsl_dir_rele(dd
, dp
);
283 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DEADLISTS
) {
284 err
= dsl_pool_open_special_dir(dp
, FREE_DIR_NAME
,
289 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
290 DMU_POOL_FREE_BPOBJ
, sizeof (uint64_t), 1, &obj
);
293 VERIFY0(bpobj_open(&dp
->dp_free_bpobj
,
294 dp
->dp_meta_objset
, obj
));
297 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_OBSOLETE_COUNTS
)) {
298 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
299 DMU_POOL_OBSOLETE_BPOBJ
, sizeof (uint64_t), 1, &obj
);
301 VERIFY0(bpobj_open(&dp
->dp_obsolete_bpobj
,
302 dp
->dp_meta_objset
, obj
));
303 } else if (err
== ENOENT
) {
305 * We might not have created the remap bpobj yet.
314 * Note: errors ignored, because the these special dirs, used for
315 * space accounting, are only created on demand.
317 (void) dsl_pool_open_special_dir(dp
, LEAK_DIR_NAME
,
320 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_ASYNC_DESTROY
)) {
321 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
322 DMU_POOL_BPTREE_OBJ
, sizeof (uint64_t), 1,
328 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_EMPTY_BPOBJ
)) {
329 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
330 DMU_POOL_EMPTY_BPOBJ
, sizeof (uint64_t), 1,
331 &dp
->dp_empty_bpobj
);
336 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
337 DMU_POOL_TMP_USERREFS
, sizeof (uint64_t), 1,
338 &dp
->dp_tmp_userrefs_obj
);
344 err
= dsl_scan_init(dp
, dp
->dp_tx
.tx_open_txg
);
347 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
352 dsl_pool_close(dsl_pool_t
*dp
)
355 * Drop our references from dsl_pool_open().
357 * Since we held the origin_snap from "syncing" context (which
358 * includes pool-opening context), it actually only got a "ref"
359 * and not a hold, so just drop that here.
361 if (dp
->dp_origin_snap
!= NULL
)
362 dsl_dataset_rele(dp
->dp_origin_snap
, dp
);
363 if (dp
->dp_mos_dir
!= NULL
)
364 dsl_dir_rele(dp
->dp_mos_dir
, dp
);
365 if (dp
->dp_free_dir
!= NULL
)
366 dsl_dir_rele(dp
->dp_free_dir
, dp
);
367 if (dp
->dp_leak_dir
!= NULL
)
368 dsl_dir_rele(dp
->dp_leak_dir
, dp
);
369 if (dp
->dp_root_dir
!= NULL
)
370 dsl_dir_rele(dp
->dp_root_dir
, dp
);
372 bpobj_close(&dp
->dp_free_bpobj
);
373 bpobj_close(&dp
->dp_obsolete_bpobj
);
375 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
376 if (dp
->dp_meta_objset
!= NULL
)
377 dmu_objset_evict(dp
->dp_meta_objset
);
379 txg_list_destroy(&dp
->dp_dirty_datasets
);
380 txg_list_destroy(&dp
->dp_dirty_zilogs
);
381 txg_list_destroy(&dp
->dp_sync_tasks
);
382 txg_list_destroy(&dp
->dp_early_sync_tasks
);
383 txg_list_destroy(&dp
->dp_dirty_dirs
);
385 taskq_destroy(dp
->dp_zil_clean_taskq
);
386 taskq_destroy(dp
->dp_sync_taskq
);
389 * We can't set retry to TRUE since we're explicitly specifying
390 * a spa to flush. This is good enough; any missed buffers for
391 * this spa won't cause trouble, and they'll eventually fall
392 * out of the ARC just like any other unused buffer.
394 arc_flush(dp
->dp_spa
, FALSE
);
398 dmu_buf_user_evict_wait();
400 rrw_destroy(&dp
->dp_config_rwlock
);
401 mutex_destroy(&dp
->dp_lock
);
402 taskq_destroy(dp
->dp_vnrele_taskq
);
403 if (dp
->dp_blkstats
!= NULL
)
404 kmem_free(dp
->dp_blkstats
, sizeof (zfs_all_blkstats_t
));
405 kmem_free(dp
, sizeof (dsl_pool_t
));
409 dsl_pool_create_obsolete_bpobj(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
413 * Currently, we only create the obsolete_bpobj where there are
414 * indirect vdevs with referenced mappings.
416 ASSERT(spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_DEVICE_REMOVAL
));
417 /* create and open the obsolete_bpobj */
418 obj
= bpobj_alloc(dp
->dp_meta_objset
, SPA_OLD_MAXBLOCKSIZE
, tx
);
419 VERIFY0(bpobj_open(&dp
->dp_obsolete_bpobj
, dp
->dp_meta_objset
, obj
));
420 VERIFY0(zap_add(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
421 DMU_POOL_OBSOLETE_BPOBJ
, sizeof (uint64_t), 1, &obj
, tx
));
422 spa_feature_incr(dp
->dp_spa
, SPA_FEATURE_OBSOLETE_COUNTS
, tx
);
426 dsl_pool_destroy_obsolete_bpobj(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
428 spa_feature_decr(dp
->dp_spa
, SPA_FEATURE_OBSOLETE_COUNTS
, tx
);
429 VERIFY0(zap_remove(dp
->dp_meta_objset
,
430 DMU_POOL_DIRECTORY_OBJECT
,
431 DMU_POOL_OBSOLETE_BPOBJ
, tx
));
432 bpobj_free(dp
->dp_meta_objset
,
433 dp
->dp_obsolete_bpobj
.bpo_object
, tx
);
434 bpobj_close(&dp
->dp_obsolete_bpobj
);
438 dsl_pool_create(spa_t
*spa
, nvlist_t
*zplprops
, uint64_t txg
)
441 dsl_pool_t
*dp
= dsl_pool_open_impl(spa
, txg
);
442 dmu_tx_t
*tx
= dmu_tx_create_assigned(dp
, txg
);
446 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
448 /* create and open the MOS (meta-objset) */
449 dp
->dp_meta_objset
= dmu_objset_create_impl(spa
,
450 NULL
, &dp
->dp_meta_rootbp
, DMU_OST_META
, tx
);
452 /* create the pool directory */
453 err
= zap_create_claim(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
454 DMU_OT_OBJECT_DIRECTORY
, DMU_OT_NONE
, 0, tx
);
457 /* Initialize scan structures */
458 VERIFY0(dsl_scan_init(dp
, txg
));
460 /* create and open the root dir */
461 dp
->dp_root_dir_obj
= dsl_dir_create_sync(dp
, NULL
, NULL
, tx
);
462 VERIFY0(dsl_dir_hold_obj(dp
, dp
->dp_root_dir_obj
,
463 NULL
, dp
, &dp
->dp_root_dir
));
465 /* create and open the meta-objset dir */
466 (void) dsl_dir_create_sync(dp
, dp
->dp_root_dir
, MOS_DIR_NAME
, tx
);
467 VERIFY0(dsl_pool_open_special_dir(dp
,
468 MOS_DIR_NAME
, &dp
->dp_mos_dir
));
470 if (spa_version(spa
) >= SPA_VERSION_DEADLISTS
) {
471 /* create and open the free dir */
472 (void) dsl_dir_create_sync(dp
, dp
->dp_root_dir
,
474 VERIFY0(dsl_pool_open_special_dir(dp
,
475 FREE_DIR_NAME
, &dp
->dp_free_dir
));
477 /* create and open the free_bplist */
478 obj
= bpobj_alloc(dp
->dp_meta_objset
, SPA_OLD_MAXBLOCKSIZE
, tx
);
479 VERIFY(zap_add(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
480 DMU_POOL_FREE_BPOBJ
, sizeof (uint64_t), 1, &obj
, tx
) == 0);
481 VERIFY0(bpobj_open(&dp
->dp_free_bpobj
,
482 dp
->dp_meta_objset
, obj
));
485 if (spa_version(spa
) >= SPA_VERSION_DSL_SCRUB
)
486 dsl_pool_create_origin(dp
, tx
);
488 /* create the root dataset */
489 obj
= dsl_dataset_create_sync_dd(dp
->dp_root_dir
, NULL
, 0, tx
);
491 /* create the root objset */
492 VERIFY0(dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
));
496 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
497 os
= dmu_objset_create_impl(dp
->dp_spa
, ds
,
498 dsl_dataset_get_blkptr(ds
), DMU_OST_ZFS
, tx
);
499 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
500 zfs_create_fs(os
, kcred
, zplprops
, tx
);
503 dsl_dataset_rele(ds
, FTAG
);
507 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
513 * Account for the meta-objset space in its placeholder dsl_dir.
516 dsl_pool_mos_diduse_space(dsl_pool_t
*dp
,
517 int64_t used
, int64_t comp
, int64_t uncomp
)
519 ASSERT3U(comp
, ==, uncomp
); /* it's all metadata */
520 mutex_enter(&dp
->dp_lock
);
521 dp
->dp_mos_used_delta
+= used
;
522 dp
->dp_mos_compressed_delta
+= comp
;
523 dp
->dp_mos_uncompressed_delta
+= uncomp
;
524 mutex_exit(&dp
->dp_lock
);
528 dsl_pool_sync_mos(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
530 zio_t
*zio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
531 dmu_objset_sync(dp
->dp_meta_objset
, zio
, tx
);
532 VERIFY0(zio_wait(zio
));
533 dprintf_bp(&dp
->dp_meta_rootbp
, "meta objset rootbp is %s", "");
534 spa_set_rootblkptr(dp
->dp_spa
, &dp
->dp_meta_rootbp
);
538 dsl_pool_dirty_delta(dsl_pool_t
*dp
, int64_t delta
)
540 ASSERT(MUTEX_HELD(&dp
->dp_lock
));
543 ASSERT3U(-delta
, <=, dp
->dp_dirty_total
);
545 dp
->dp_dirty_total
+= delta
;
548 * Note: we signal even when increasing dp_dirty_total.
549 * This ensures forward progress -- each thread wakes the next waiter.
551 if (dp
->dp_dirty_total
< zfs_dirty_data_max
)
552 cv_signal(&dp
->dp_spaceavail_cv
);
556 dsl_early_sync_task_verify(dsl_pool_t
*dp
, uint64_t txg
)
558 spa_t
*spa
= dp
->dp_spa
;
559 vdev_t
*rvd
= spa
->spa_root_vdev
;
561 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
562 vdev_t
*vd
= rvd
->vdev_child
[c
];
563 txg_list_t
*tl
= &vd
->vdev_ms_list
;
566 for (ms
= txg_list_head(tl
, TXG_CLEAN(txg
)); ms
;
567 ms
= txg_list_next(tl
, ms
, TXG_CLEAN(txg
))) {
568 VERIFY(range_tree_is_empty(ms
->ms_freeing
));
569 VERIFY(range_tree_is_empty(ms
->ms_checkpointing
));
577 dsl_pool_sync(dsl_pool_t
*dp
, uint64_t txg
)
583 objset_t
*mos
= dp
->dp_meta_objset
;
584 list_t synced_datasets
;
586 list_create(&synced_datasets
, sizeof (dsl_dataset_t
),
587 offsetof(dsl_dataset_t
, ds_synced_link
));
589 tx
= dmu_tx_create_assigned(dp
, txg
);
592 * Run all early sync tasks before writing out any dirty blocks.
593 * For more info on early sync tasks see block comment in
594 * dsl_early_sync_task().
596 if (!txg_list_empty(&dp
->dp_early_sync_tasks
, txg
)) {
597 dsl_sync_task_t
*dst
;
599 ASSERT3U(spa_sync_pass(dp
->dp_spa
), ==, 1);
601 txg_list_remove(&dp
->dp_early_sync_tasks
, txg
)) != NULL
) {
602 ASSERT(dsl_early_sync_task_verify(dp
, txg
));
603 dsl_sync_task_sync(dst
, tx
);
605 ASSERT(dsl_early_sync_task_verify(dp
, txg
));
609 * Write out all dirty blocks of dirty datasets.
611 zio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
612 while ((ds
= txg_list_remove(&dp
->dp_dirty_datasets
, txg
)) != NULL
) {
614 * We must not sync any non-MOS datasets twice, because
615 * we may have taken a snapshot of them. However, we
616 * may sync newly-created datasets on pass 2.
618 ASSERT(!list_link_active(&ds
->ds_synced_link
));
619 list_insert_tail(&synced_datasets
, ds
);
620 dsl_dataset_sync(ds
, zio
, tx
);
622 VERIFY0(zio_wait(zio
));
625 * We have written all of the accounted dirty data, so our
626 * dp_space_towrite should now be zero. However, some seldom-used
627 * code paths do not adhere to this (e.g. dbuf_undirty(), also
628 * rounding error in dbuf_write_physdone).
629 * Shore up the accounting of any dirtied space now.
631 dsl_pool_undirty_space(dp
, dp
->dp_dirty_pertxg
[txg
& TXG_MASK
], txg
);
634 * Update the long range free counter after
635 * we're done syncing user data
637 mutex_enter(&dp
->dp_lock
);
638 ASSERT(spa_sync_pass(dp
->dp_spa
) == 1 ||
639 dp
->dp_long_free_dirty_pertxg
[txg
& TXG_MASK
] == 0);
640 dp
->dp_long_free_dirty_pertxg
[txg
& TXG_MASK
] = 0;
641 mutex_exit(&dp
->dp_lock
);
644 * After the data blocks have been written (ensured by the zio_wait()
645 * above), update the user/group space accounting. This happens
646 * in tasks dispatched to dp_sync_taskq, so wait for them before
649 for (ds
= list_head(&synced_datasets
); ds
!= NULL
;
650 ds
= list_next(&synced_datasets
, ds
)) {
651 dmu_objset_do_userquota_updates(ds
->ds_objset
, tx
);
653 taskq_wait(dp
->dp_sync_taskq
);
656 * Sync the datasets again to push out the changes due to
657 * userspace updates. This must be done before we process the
658 * sync tasks, so that any snapshots will have the correct
659 * user accounting information (and we won't get confused
660 * about which blocks are part of the snapshot).
662 zio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
663 while ((ds
= txg_list_remove(&dp
->dp_dirty_datasets
, txg
)) != NULL
) {
664 ASSERT(list_link_active(&ds
->ds_synced_link
));
665 dmu_buf_rele(ds
->ds_dbuf
, ds
);
666 dsl_dataset_sync(ds
, zio
, tx
);
668 VERIFY0(zio_wait(zio
));
671 * Now that the datasets have been completely synced, we can
672 * clean up our in-memory structures accumulated while syncing:
674 * - move dead blocks from the pending deadlist to the on-disk deadlist
675 * - release hold from dsl_dataset_dirty()
677 while ((ds
= list_remove_head(&synced_datasets
)) != NULL
) {
678 dsl_dataset_sync_done(ds
, tx
);
680 while ((dd
= txg_list_remove(&dp
->dp_dirty_dirs
, txg
)) != NULL
) {
681 dsl_dir_sync(dd
, tx
);
685 * The MOS's space is accounted for in the pool/$MOS
686 * (dp_mos_dir). We can't modify the mos while we're syncing
687 * it, so we remember the deltas and apply them here.
689 if (dp
->dp_mos_used_delta
!= 0 || dp
->dp_mos_compressed_delta
!= 0 ||
690 dp
->dp_mos_uncompressed_delta
!= 0) {
691 dsl_dir_diduse_space(dp
->dp_mos_dir
, DD_USED_HEAD
,
692 dp
->dp_mos_used_delta
,
693 dp
->dp_mos_compressed_delta
,
694 dp
->dp_mos_uncompressed_delta
, tx
);
695 dp
->dp_mos_used_delta
= 0;
696 dp
->dp_mos_compressed_delta
= 0;
697 dp
->dp_mos_uncompressed_delta
= 0;
700 if (!multilist_is_empty(mos
->os_dirty_dnodes
[txg
& TXG_MASK
])) {
701 dsl_pool_sync_mos(dp
, tx
);
705 * If we modify a dataset in the same txg that we want to destroy it,
706 * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it.
707 * dsl_dir_destroy_check() will fail if there are unexpected holds.
708 * Therefore, we want to sync the MOS (thus syncing the dd_dbuf
709 * and clearing the hold on it) before we process the sync_tasks.
710 * The MOS data dirtied by the sync_tasks will be synced on the next
713 if (!txg_list_empty(&dp
->dp_sync_tasks
, txg
)) {
714 dsl_sync_task_t
*dst
;
716 * No more sync tasks should have been added while we
719 ASSERT3U(spa_sync_pass(dp
->dp_spa
), ==, 1);
720 while ((dst
= txg_list_remove(&dp
->dp_sync_tasks
, txg
)) != NULL
)
721 dsl_sync_task_sync(dst
, tx
);
726 DTRACE_PROBE2(dsl_pool_sync__done
, dsl_pool_t
*dp
, dp
, uint64_t, txg
);
730 dsl_pool_sync_done(dsl_pool_t
*dp
, uint64_t txg
)
734 while (zilog
= txg_list_head(&dp
->dp_dirty_zilogs
, txg
)) {
735 dsl_dataset_t
*ds
= dmu_objset_ds(zilog
->zl_os
);
737 * We don't remove the zilog from the dp_dirty_zilogs
738 * list until after we've cleaned it. This ensures that
739 * callers of zilog_is_dirty() receive an accurate
740 * answer when they are racing with the spa sync thread.
742 zil_clean(zilog
, txg
);
743 (void) txg_list_remove_this(&dp
->dp_dirty_zilogs
, zilog
, txg
);
744 ASSERT(!dmu_objset_is_dirty(zilog
->zl_os
, txg
));
745 dmu_buf_rele(ds
->ds_dbuf
, zilog
);
747 ASSERT(!dmu_objset_is_dirty(dp
->dp_meta_objset
, txg
));
751 * TRUE if the current thread is the tx_sync_thread or if we
752 * are being called from SPA context during pool initialization.
755 dsl_pool_sync_context(dsl_pool_t
*dp
)
757 return (curthread
== dp
->dp_tx
.tx_sync_thread
||
758 spa_is_initializing(dp
->dp_spa
) ||
759 taskq_member(dp
->dp_sync_taskq
, curthread
));
763 * This function returns the amount of allocatable space in the pool
764 * minus whatever space is currently reserved by ZFS for specific
765 * purposes. Specifically:
767 * 1] Any reserved SLOP space
768 * 2] Any space used by the checkpoint
769 * 3] Any space used for deferred frees
771 * The latter 2 are especially important because they are needed to
772 * rectify the SPA's and DMU's different understanding of how much space
773 * is used. Now the DMU is aware of that extra space tracked by the SPA
774 * without having to maintain a separate special dir (e.g similar to
775 * $MOS, $FREEING, and $LEAKED).
777 * Note: By deferred frees here, we mean the frees that were deferred
778 * in spa_sync() after sync pass 1 (spa_deferred_bpobj), and not the
779 * segments placed in ms_defer trees during metaslab_sync_done().
782 dsl_pool_adjustedsize(dsl_pool_t
*dp
, zfs_space_check_t slop_policy
)
784 spa_t
*spa
= dp
->dp_spa
;
785 uint64_t space
, resv
, adjustedsize
;
786 uint64_t spa_deferred_frees
=
787 spa
->spa_deferred_bpobj
.bpo_phys
->bpo_bytes
;
789 space
= spa_get_dspace(spa
)
790 - spa_get_checkpoint_space(spa
) - spa_deferred_frees
;
791 resv
= spa_get_slop_space(spa
);
793 switch (slop_policy
) {
794 case ZFS_SPACE_CHECK_NORMAL
:
796 case ZFS_SPACE_CHECK_RESERVED
:
799 case ZFS_SPACE_CHECK_EXTRA_RESERVED
:
802 case ZFS_SPACE_CHECK_NONE
:
806 panic("invalid slop policy value: %d", slop_policy
);
809 adjustedsize
= (space
>= resv
) ? (space
- resv
) : 0;
811 return (adjustedsize
);
815 dsl_pool_unreserved_space(dsl_pool_t
*dp
, zfs_space_check_t slop_policy
)
817 uint64_t poolsize
= dsl_pool_adjustedsize(dp
, slop_policy
);
819 metaslab_class_get_deferred(spa_normal_class(dp
->dp_spa
));
820 uint64_t quota
= (poolsize
>= deferred
) ? (poolsize
- deferred
) : 0;
825 dsl_pool_need_dirty_delay(dsl_pool_t
*dp
)
827 uint64_t delay_min_bytes
=
828 zfs_dirty_data_max
* zfs_delay_min_dirty_percent
/ 100;
829 uint64_t dirty_min_bytes
=
830 zfs_dirty_data_max
* zfs_dirty_data_sync_pct
/ 100;
833 mutex_enter(&dp
->dp_lock
);
834 if (dp
->dp_dirty_total
> dirty_min_bytes
)
836 rv
= (dp
->dp_dirty_total
> delay_min_bytes
);
837 mutex_exit(&dp
->dp_lock
);
842 dsl_pool_dirty_space(dsl_pool_t
*dp
, int64_t space
, dmu_tx_t
*tx
)
845 mutex_enter(&dp
->dp_lock
);
846 dp
->dp_dirty_pertxg
[tx
->tx_txg
& TXG_MASK
] += space
;
847 dsl_pool_dirty_delta(dp
, space
);
848 mutex_exit(&dp
->dp_lock
);
853 dsl_pool_undirty_space(dsl_pool_t
*dp
, int64_t space
, uint64_t txg
)
855 ASSERT3S(space
, >=, 0);
858 mutex_enter(&dp
->dp_lock
);
859 if (dp
->dp_dirty_pertxg
[txg
& TXG_MASK
] < space
) {
860 /* XXX writing something we didn't dirty? */
861 space
= dp
->dp_dirty_pertxg
[txg
& TXG_MASK
];
863 ASSERT3U(dp
->dp_dirty_pertxg
[txg
& TXG_MASK
], >=, space
);
864 dp
->dp_dirty_pertxg
[txg
& TXG_MASK
] -= space
;
865 ASSERT3U(dp
->dp_dirty_total
, >=, space
);
866 dsl_pool_dirty_delta(dp
, -space
);
867 mutex_exit(&dp
->dp_lock
);
872 upgrade_clones_cb(dsl_pool_t
*dp
, dsl_dataset_t
*hds
, void *arg
)
875 dsl_dataset_t
*ds
, *prev
= NULL
;
878 err
= dsl_dataset_hold_obj(dp
, hds
->ds_object
, FTAG
, &ds
);
882 while (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
883 err
= dsl_dataset_hold_obj(dp
,
884 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
886 dsl_dataset_rele(ds
, FTAG
);
890 if (dsl_dataset_phys(prev
)->ds_next_snap_obj
!= ds
->ds_object
)
892 dsl_dataset_rele(ds
, FTAG
);
898 prev
= dp
->dp_origin_snap
;
901 * The $ORIGIN can't have any data, or the accounting
904 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
905 ASSERT0(dsl_dataset_phys(prev
)->ds_bp
.blk_birth
);
906 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
908 /* The origin doesn't get attached to itself */
909 if (ds
->ds_object
== prev
->ds_object
) {
910 dsl_dataset_rele(ds
, FTAG
);
914 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
915 dsl_dataset_phys(ds
)->ds_prev_snap_obj
= prev
->ds_object
;
916 dsl_dataset_phys(ds
)->ds_prev_snap_txg
=
917 dsl_dataset_phys(prev
)->ds_creation_txg
;
919 dmu_buf_will_dirty(ds
->ds_dir
->dd_dbuf
, tx
);
920 dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
= prev
->ds_object
;
922 dmu_buf_will_dirty(prev
->ds_dbuf
, tx
);
923 dsl_dataset_phys(prev
)->ds_num_children
++;
925 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0) {
926 ASSERT(ds
->ds_prev
== NULL
);
927 VERIFY0(dsl_dataset_hold_obj(dp
,
928 dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
933 ASSERT3U(dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
, ==, prev
->ds_object
);
934 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_obj
, ==, prev
->ds_object
);
936 if (dsl_dataset_phys(prev
)->ds_next_clones_obj
== 0) {
937 dmu_buf_will_dirty(prev
->ds_dbuf
, tx
);
938 dsl_dataset_phys(prev
)->ds_next_clones_obj
=
939 zap_create(dp
->dp_meta_objset
,
940 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
942 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
943 dsl_dataset_phys(prev
)->ds_next_clones_obj
, ds
->ds_object
, tx
));
945 dsl_dataset_rele(ds
, FTAG
);
946 if (prev
!= dp
->dp_origin_snap
)
947 dsl_dataset_rele(prev
, FTAG
);
952 dsl_pool_upgrade_clones(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
954 ASSERT(dmu_tx_is_syncing(tx
));
955 ASSERT(dp
->dp_origin_snap
!= NULL
);
957 VERIFY0(dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
, upgrade_clones_cb
,
958 tx
, DS_FIND_CHILDREN
| DS_FIND_SERIALIZE
));
963 upgrade_dir_clones_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
966 objset_t
*mos
= dp
->dp_meta_objset
;
968 if (dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
!= 0) {
969 dsl_dataset_t
*origin
;
971 VERIFY0(dsl_dataset_hold_obj(dp
,
972 dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
, FTAG
, &origin
));
974 if (dsl_dir_phys(origin
->ds_dir
)->dd_clones
== 0) {
975 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
976 dsl_dir_phys(origin
->ds_dir
)->dd_clones
=
977 zap_create(mos
, DMU_OT_DSL_CLONES
, DMU_OT_NONE
,
981 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
982 dsl_dir_phys(origin
->ds_dir
)->dd_clones
,
985 dsl_dataset_rele(origin
, FTAG
);
991 dsl_pool_upgrade_dir_clones(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
993 ASSERT(dmu_tx_is_syncing(tx
));
996 (void) dsl_dir_create_sync(dp
, dp
->dp_root_dir
, FREE_DIR_NAME
, tx
);
997 VERIFY0(dsl_pool_open_special_dir(dp
,
998 FREE_DIR_NAME
, &dp
->dp_free_dir
));
1001 * We can't use bpobj_alloc(), because spa_version() still
1002 * returns the old version, and we need a new-version bpobj with
1003 * subobj support. So call dmu_object_alloc() directly.
1005 obj
= dmu_object_alloc(dp
->dp_meta_objset
, DMU_OT_BPOBJ
,
1006 SPA_OLD_MAXBLOCKSIZE
, DMU_OT_BPOBJ_HDR
, sizeof (bpobj_phys_t
), tx
);
1007 VERIFY0(zap_add(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1008 DMU_POOL_FREE_BPOBJ
, sizeof (uint64_t), 1, &obj
, tx
));
1009 VERIFY0(bpobj_open(&dp
->dp_free_bpobj
, dp
->dp_meta_objset
, obj
));
1011 VERIFY0(dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1012 upgrade_dir_clones_cb
, tx
, DS_FIND_CHILDREN
| DS_FIND_SERIALIZE
));
1016 dsl_pool_create_origin(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
1021 ASSERT(dmu_tx_is_syncing(tx
));
1022 ASSERT(dp
->dp_origin_snap
== NULL
);
1023 ASSERT(rrw_held(&dp
->dp_config_rwlock
, RW_WRITER
));
1025 /* create the origin dir, ds, & snap-ds */
1026 dsobj
= dsl_dataset_create_sync(dp
->dp_root_dir
, ORIGIN_DIR_NAME
,
1027 NULL
, 0, kcred
, tx
);
1028 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
1029 dsl_dataset_snapshot_sync_impl(ds
, ORIGIN_DIR_NAME
, tx
);
1030 VERIFY0(dsl_dataset_hold_obj(dp
, dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
1031 dp
, &dp
->dp_origin_snap
));
1032 dsl_dataset_rele(ds
, FTAG
);
1036 dsl_pool_vnrele_taskq(dsl_pool_t
*dp
)
1038 return (dp
->dp_vnrele_taskq
);
1042 * Walk through the pool-wide zap object of temporary snapshot user holds
1046 dsl_pool_clean_tmp_userrefs(dsl_pool_t
*dp
)
1050 objset_t
*mos
= dp
->dp_meta_objset
;
1051 uint64_t zapobj
= dp
->dp_tmp_userrefs_obj
;
1056 ASSERT(spa_version(dp
->dp_spa
) >= SPA_VERSION_USERREFS
);
1058 holds
= fnvlist_alloc();
1060 for (zap_cursor_init(&zc
, mos
, zapobj
);
1061 zap_cursor_retrieve(&zc
, &za
) == 0;
1062 zap_cursor_advance(&zc
)) {
1066 htag
= strchr(za
.za_name
, '-');
1069 if (nvlist_lookup_nvlist(holds
, za
.za_name
, &tags
) != 0) {
1070 tags
= fnvlist_alloc();
1071 fnvlist_add_boolean(tags
, htag
);
1072 fnvlist_add_nvlist(holds
, za
.za_name
, tags
);
1075 fnvlist_add_boolean(tags
, htag
);
1078 dsl_dataset_user_release_tmp(dp
, holds
);
1079 fnvlist_free(holds
);
1080 zap_cursor_fini(&zc
);
1084 * Create the pool-wide zap object for storing temporary snapshot holds.
1087 dsl_pool_user_hold_create_obj(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
1089 objset_t
*mos
= dp
->dp_meta_objset
;
1091 ASSERT(dp
->dp_tmp_userrefs_obj
== 0);
1092 ASSERT(dmu_tx_is_syncing(tx
));
1094 dp
->dp_tmp_userrefs_obj
= zap_create_link(mos
, DMU_OT_USERREFS
,
1095 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_TMP_USERREFS
, tx
);
1099 dsl_pool_user_hold_rele_impl(dsl_pool_t
*dp
, uint64_t dsobj
,
1100 const char *tag
, uint64_t now
, dmu_tx_t
*tx
, boolean_t holding
)
1102 objset_t
*mos
= dp
->dp_meta_objset
;
1103 uint64_t zapobj
= dp
->dp_tmp_userrefs_obj
;
1107 ASSERT(spa_version(dp
->dp_spa
) >= SPA_VERSION_USERREFS
);
1108 ASSERT(dmu_tx_is_syncing(tx
));
1111 * If the pool was created prior to SPA_VERSION_USERREFS, the
1112 * zap object for temporary holds might not exist yet.
1116 dsl_pool_user_hold_create_obj(dp
, tx
);
1117 zapobj
= dp
->dp_tmp_userrefs_obj
;
1119 return (SET_ERROR(ENOENT
));
1123 name
= kmem_asprintf("%llx-%s", (u_longlong_t
)dsobj
, tag
);
1125 error
= zap_add(mos
, zapobj
, name
, 8, 1, &now
, tx
);
1127 error
= zap_remove(mos
, zapobj
, name
, tx
);
1134 * Add a temporary hold for the given dataset object and tag.
1137 dsl_pool_user_hold(dsl_pool_t
*dp
, uint64_t dsobj
, const char *tag
,
1138 uint64_t now
, dmu_tx_t
*tx
)
1140 return (dsl_pool_user_hold_rele_impl(dp
, dsobj
, tag
, now
, tx
, B_TRUE
));
1144 * Release a temporary hold for the given dataset object and tag.
1147 dsl_pool_user_release(dsl_pool_t
*dp
, uint64_t dsobj
, const char *tag
,
1150 return (dsl_pool_user_hold_rele_impl(dp
, dsobj
, tag
, (uintptr_t)NULL
,
1155 * DSL Pool Configuration Lock
1157 * The dp_config_rwlock protects against changes to DSL state (e.g. dataset
1158 * creation / destruction / rename / property setting). It must be held for
1159 * read to hold a dataset or dsl_dir. I.e. you must call
1160 * dsl_pool_config_enter() or dsl_pool_hold() before calling
1161 * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock
1162 * must be held continuously until all datasets and dsl_dirs are released.
1164 * The only exception to this rule is that if a "long hold" is placed on
1165 * a dataset, then the dp_config_rwlock may be dropped while the dataset
1166 * is still held. The long hold will prevent the dataset from being
1167 * destroyed -- the destroy will fail with EBUSY. A long hold can be
1168 * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset
1169 * (by calling dsl_{dataset,objset}_{try}own{_obj}).
1171 * Legitimate long-holders (including owners) should be long-running, cancelable
1172 * tasks that should cause "zfs destroy" to fail. This includes DMU
1173 * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open),
1174 * "zfs send", and "zfs diff". There are several other long-holders whose
1175 * uses are suboptimal (e.g. "zfs promote", and zil_suspend()).
1177 * The usual formula for long-holding would be:
1179 * dsl_dataset_hold()
1180 * ... perform checks ...
1181 * dsl_dataset_long_hold()
1183 * ... perform long-running task ...
1184 * dsl_dataset_long_rele()
1185 * dsl_dataset_rele()
1187 * Note that when the long hold is released, the dataset is still held but
1188 * the pool is not held. The dataset may change arbitrarily during this time
1189 * (e.g. it could be destroyed). Therefore you shouldn't do anything to the
1190 * dataset except release it.
1192 * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only
1193 * or modifying operations.
1195 * Modifying operations should generally use dsl_sync_task(). The synctask
1196 * infrastructure enforces proper locking strategy with respect to the
1197 * dp_config_rwlock. See the comment above dsl_sync_task() for details.
1199 * Read-only operations will manually hold the pool, then the dataset, obtain
1200 * information from the dataset, then release the pool and dataset.
1201 * dmu_objset_{hold,rele}() are convenience routines that also do the pool
1206 dsl_pool_hold(const char *name
, void *tag
, dsl_pool_t
**dp
)
1211 error
= spa_open(name
, &spa
, tag
);
1213 *dp
= spa_get_dsl(spa
);
1214 dsl_pool_config_enter(*dp
, tag
);
1220 dsl_pool_rele(dsl_pool_t
*dp
, void *tag
)
1222 dsl_pool_config_exit(dp
, tag
);
1223 spa_close(dp
->dp_spa
, tag
);
1227 dsl_pool_config_enter(dsl_pool_t
*dp
, void *tag
)
1230 * We use a "reentrant" reader-writer lock, but not reentrantly.
1232 * The rrwlock can (with the track_all flag) track all reading threads,
1233 * which is very useful for debugging which code path failed to release
1234 * the lock, and for verifying that the *current* thread does hold
1237 * (Unlike a rwlock, which knows that N threads hold it for
1238 * read, but not *which* threads, so rw_held(RW_READER) returns TRUE
1239 * if any thread holds it for read, even if this thread doesn't).
1241 ASSERT(!rrw_held(&dp
->dp_config_rwlock
, RW_READER
));
1242 rrw_enter(&dp
->dp_config_rwlock
, RW_READER
, tag
);
1246 dsl_pool_config_enter_prio(dsl_pool_t
*dp
, void *tag
)
1248 ASSERT(!rrw_held(&dp
->dp_config_rwlock
, RW_READER
));
1249 rrw_enter_read_prio(&dp
->dp_config_rwlock
, tag
);
1253 dsl_pool_config_exit(dsl_pool_t
*dp
, void *tag
)
1255 rrw_exit(&dp
->dp_config_rwlock
, tag
);
1259 dsl_pool_config_held(dsl_pool_t
*dp
)
1261 return (RRW_LOCK_HELD(&dp
->dp_config_rwlock
));
1265 dsl_pool_config_held_writer(dsl_pool_t
*dp
)
1267 return (RRW_WRITE_HELD(&dp
->dp_config_rwlock
));