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) 2012 by Delphix. All rights reserved.
26 #include <sys/dsl_pool.h>
27 #include <sys/dsl_dataset.h>
28 #include <sys/dsl_prop.h>
29 #include <sys/dsl_dir.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dsl_scan.h>
32 #include <sys/dnode.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/dmu_objset.h>
38 #include <sys/zfs_context.h>
39 #include <sys/fs/zfs.h>
40 #include <sys/zfs_znode.h>
41 #include <sys/spa_impl.h>
42 #include <sys/dsl_deadlist.h>
43 #include <sys/bptree.h>
44 #include <sys/zfeature.h>
46 int zfs_no_write_throttle = 0;
47 int zfs_write_limit_shift = 3; /* 1/8th of physical memory */
48 int zfs_txg_synctime_ms = 1000; /* target millisecs to sync a txg */
50 uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */
51 uint64_t zfs_write_limit_max = 0; /* max data payload per txg */
52 uint64_t zfs_write_limit_inflated = 0;
53 uint64_t zfs_write_limit_override = 0;
55 kmutex_t zfs_write_limit_lock;
57 static pgcnt_t old_physmem = 0;
59 SYSCTL_DECL(_vfs_zfs);
60 TUNABLE_INT("vfs.zfs.no_write_throttle", &zfs_no_write_throttle);
61 SYSCTL_INT(_vfs_zfs, OID_AUTO, no_write_throttle, CTLFLAG_RDTUN,
62 &zfs_no_write_throttle, 0, "");
63 TUNABLE_INT("vfs.zfs.write_limit_shift", &zfs_write_limit_shift);
64 SYSCTL_INT(_vfs_zfs, OID_AUTO, write_limit_shift, CTLFLAG_RDTUN,
65 &zfs_write_limit_shift, 0, "2^N of physical memory");
66 SYSCTL_DECL(_vfs_zfs_txg);
67 TUNABLE_INT("vfs.zfs.txg.synctime_ms", &zfs_txg_synctime_ms);
68 SYSCTL_INT(_vfs_zfs_txg, OID_AUTO, synctime_ms, CTLFLAG_RDTUN,
69 &zfs_txg_synctime_ms, 0, "Target milliseconds to sync a txg");
71 TUNABLE_QUAD("vfs.zfs.write_limit_min", &zfs_write_limit_min);
72 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_min, CTLFLAG_RDTUN,
73 &zfs_write_limit_min, 0, "Minimum write limit");
74 TUNABLE_QUAD("vfs.zfs.write_limit_max", &zfs_write_limit_max);
75 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_max, CTLFLAG_RDTUN,
76 &zfs_write_limit_max, 0, "Maximum data payload per txg");
77 TUNABLE_QUAD("vfs.zfs.write_limit_inflated", &zfs_write_limit_inflated);
78 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_inflated, CTLFLAG_RDTUN,
79 &zfs_write_limit_inflated, 0, "");
80 TUNABLE_QUAD("vfs.zfs.write_limit_override", &zfs_write_limit_override);
81 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_override, CTLFLAG_RDTUN,
82 &zfs_write_limit_override, 0, "");
85 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
90 err = zap_lookup(dp->dp_meta_objset,
91 dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
92 name, sizeof (obj), 1, &obj);
96 return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
100 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
103 blkptr_t *bp = spa_get_rootblkptr(spa);
105 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
107 dp->dp_meta_rootbp = *bp;
108 rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
109 dp->dp_write_limit = zfs_write_limit_min;
112 txg_list_create(&dp->dp_dirty_datasets,
113 offsetof(dsl_dataset_t, ds_dirty_link));
114 txg_list_create(&dp->dp_dirty_dirs,
115 offsetof(dsl_dir_t, dd_dirty_link));
116 txg_list_create(&dp->dp_sync_tasks,
117 offsetof(dsl_sync_task_group_t, dstg_node));
118 list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t),
119 offsetof(dsl_dataset_t, ds_synced_link));
121 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
123 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
130 dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
133 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
135 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp,
136 &dp->dp_meta_objset);
146 dsl_pool_open(dsl_pool_t *dp)
153 ASSERT(!dmu_objset_is_dirty_anywhere(dp->dp_meta_objset));
155 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
156 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
157 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
158 &dp->dp_root_dir_obj);
162 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
163 NULL, dp, &dp->dp_root_dir);
167 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
171 if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) {
172 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
175 err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
178 err = dsl_dataset_hold_obj(dp,
179 ds->ds_phys->ds_prev_snap_obj, dp,
180 &dp->dp_origin_snap);
181 dsl_dataset_rele(ds, FTAG);
183 dsl_dir_close(dd, dp);
188 if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
189 err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME,
194 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
195 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj);
198 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
199 dp->dp_meta_objset, obj));
202 if (spa_feature_is_active(dp->dp_spa,
203 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
204 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
205 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
211 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
212 DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1,
213 &dp->dp_tmp_userrefs_obj);
219 err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg);
222 rw_exit(&dp->dp_config_rwlock);
227 dsl_pool_close(dsl_pool_t *dp)
229 /* drop our references from dsl_pool_open() */
232 * Since we held the origin_snap from "syncing" context (which
233 * includes pool-opening context), it actually only got a "ref"
234 * and not a hold, so just drop that here.
236 if (dp->dp_origin_snap)
237 dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
239 dsl_dir_close(dp->dp_mos_dir, dp);
241 dsl_dir_close(dp->dp_free_dir, dp);
243 dsl_dir_close(dp->dp_root_dir, dp);
245 bpobj_close(&dp->dp_free_bpobj);
247 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
248 if (dp->dp_meta_objset)
249 dmu_objset_evict(dp->dp_meta_objset);
251 txg_list_destroy(&dp->dp_dirty_datasets);
252 txg_list_destroy(&dp->dp_sync_tasks);
253 txg_list_destroy(&dp->dp_dirty_dirs);
254 list_destroy(&dp->dp_synced_datasets);
256 arc_flush(dp->dp_spa);
259 rw_destroy(&dp->dp_config_rwlock);
260 mutex_destroy(&dp->dp_lock);
261 taskq_destroy(dp->dp_vnrele_taskq);
263 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
264 kmem_free(dp, sizeof (dsl_pool_t));
268 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
271 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
272 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
277 /* create and open the MOS (meta-objset) */
278 dp->dp_meta_objset = dmu_objset_create_impl(spa,
279 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
281 /* create the pool directory */
282 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
283 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
284 ASSERT3U(err, ==, 0);
286 /* Initialize scan structures */
287 VERIFY3U(0, ==, dsl_scan_init(dp, txg));
289 /* create and open the root dir */
290 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
291 VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
292 NULL, dp, &dp->dp_root_dir));
294 /* create and open the meta-objset dir */
295 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
296 VERIFY(0 == dsl_pool_open_special_dir(dp,
297 MOS_DIR_NAME, &dp->dp_mos_dir));
299 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
300 /* create and open the free dir */
301 (void) dsl_dir_create_sync(dp, dp->dp_root_dir,
303 VERIFY(0 == dsl_pool_open_special_dir(dp,
304 FREE_DIR_NAME, &dp->dp_free_dir));
306 /* create and open the free_bplist */
307 obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx);
308 VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
309 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
310 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
311 dp->dp_meta_objset, obj));
314 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
315 dsl_pool_create_origin(dp, tx);
317 /* create the root dataset */
318 obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
320 /* create the root objset */
321 VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
322 os = dmu_objset_create_impl(dp->dp_spa, ds,
323 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
325 zfs_create_fs(os, kcred, zplprops, tx);
327 dsl_dataset_rele(ds, FTAG);
335 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
337 dsl_deadlist_t *dl = arg;
338 dsl_pool_t *dp = dmu_objset_pool(dl->dl_os);
339 rw_enter(&dp->dp_config_rwlock, RW_READER);
340 dsl_deadlist_insert(dl, bp, tx);
341 rw_exit(&dp->dp_config_rwlock);
346 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
352 dsl_sync_task_group_t *dstg;
353 objset_t *mos = dp->dp_meta_objset;
354 hrtime_t start, write_time;
355 uint64_t data_written;
359 * We need to copy dp_space_towrite() before doing
360 * dsl_sync_task_group_sync(), because
361 * dsl_dataset_snapshot_reserve_space() will increase
362 * dp_space_towrite but not actually write anything.
364 data_written = dp->dp_space_towrite[txg & TXG_MASK];
366 tx = dmu_tx_create_assigned(dp, txg);
368 dp->dp_read_overhead = 0;
371 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
372 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
374 * We must not sync any non-MOS datasets twice, because
375 * we may have taken a snapshot of them. However, we
376 * may sync newly-created datasets on pass 2.
378 ASSERT(!list_link_active(&ds->ds_synced_link));
379 list_insert_tail(&dp->dp_synced_datasets, ds);
380 dsl_dataset_sync(ds, zio, tx);
382 DTRACE_PROBE(pool_sync__1setup);
385 write_time = gethrtime() - start;
387 DTRACE_PROBE(pool_sync__2rootzio);
389 for (ds = list_head(&dp->dp_synced_datasets); ds;
390 ds = list_next(&dp->dp_synced_datasets, ds))
391 dmu_objset_do_userquota_updates(ds->ds_objset, tx);
394 * Sync the datasets again to push out the changes due to
395 * userspace updates. This must be done before we process the
396 * sync tasks, because that could cause a snapshot of a dataset
397 * whose ds_bp will be rewritten when we do this 2nd sync.
399 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
400 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
401 ASSERT(list_link_active(&ds->ds_synced_link));
402 dmu_buf_rele(ds->ds_dbuf, ds);
403 dsl_dataset_sync(ds, zio, tx);
408 * Move dead blocks from the pending deadlist to the on-disk
411 for (ds = list_head(&dp->dp_synced_datasets); ds;
412 ds = list_next(&dp->dp_synced_datasets, ds)) {
413 bplist_iterate(&ds->ds_pending_deadlist,
414 deadlist_enqueue_cb, &ds->ds_deadlist, tx);
417 while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) {
419 * No more sync tasks should have been added while we
422 ASSERT(spa_sync_pass(dp->dp_spa) == 1);
423 dsl_sync_task_group_sync(dstg, tx);
425 DTRACE_PROBE(pool_sync__3task);
428 while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
429 dsl_dir_sync(dd, tx);
430 write_time += gethrtime() - start;
433 if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
434 list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) {
435 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
436 dmu_objset_sync(mos, zio, tx);
439 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
440 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
442 write_time += gethrtime() - start;
443 DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
444 hrtime_t, dp->dp_read_overhead);
445 write_time -= dp->dp_read_overhead;
449 dp->dp_space_towrite[txg & TXG_MASK] = 0;
450 ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
453 * If the write limit max has not been explicitly set, set it
454 * to a fraction of available physical memory (default 1/8th).
455 * Note that we must inflate the limit because the spa
456 * inflates write sizes to account for data replication.
457 * Check this each sync phase to catch changing memory size.
459 if (physmem != old_physmem && zfs_write_limit_shift) {
460 mutex_enter(&zfs_write_limit_lock);
461 old_physmem = physmem;
462 zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
463 zfs_write_limit_inflated = MAX(zfs_write_limit_min,
464 spa_get_asize(dp->dp_spa, zfs_write_limit_max));
465 mutex_exit(&zfs_write_limit_lock);
469 * Attempt to keep the sync time consistent by adjusting the
470 * amount of write traffic allowed into each transaction group.
471 * Weight the throughput calculation towards the current value:
472 * thru = 3/4 old_thru + 1/4 new_thru
474 * Note: write_time is in nanosecs, so write_time/MICROSEC
477 ASSERT(zfs_write_limit_min > 0);
478 if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) {
479 uint64_t throughput = data_written / (write_time / MICROSEC);
481 if (dp->dp_throughput)
482 dp->dp_throughput = throughput / 4 +
483 3 * dp->dp_throughput / 4;
485 dp->dp_throughput = throughput;
486 dp->dp_write_limit = MIN(zfs_write_limit_inflated,
487 MAX(zfs_write_limit_min,
488 dp->dp_throughput * zfs_txg_synctime_ms));
493 dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg)
498 while (ds = list_head(&dp->dp_synced_datasets)) {
499 list_remove(&dp->dp_synced_datasets, ds);
501 zil_clean(os->os_zil, txg);
502 ASSERT(!dmu_objset_is_dirty(os, txg));
503 dmu_buf_rele(ds->ds_dbuf, ds);
505 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg));
509 * TRUE if the current thread is the tx_sync_thread or if we
510 * are being called from SPA context during pool initialization.
513 dsl_pool_sync_context(dsl_pool_t *dp)
515 return (curthread == dp->dp_tx.tx_sync_thread ||
516 spa_is_initializing(dp->dp_spa));
520 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
522 uint64_t space, resv;
525 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
527 * XXX The intent log is not accounted for, so it must fit
530 * If we're trying to assess whether it's OK to do a free,
531 * cut the reservation in half to allow forward progress
532 * (e.g. make it possible to rm(1) files from a full pool).
534 space = spa_get_dspace(dp->dp_spa);
535 resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
539 return (space - resv);
543 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
545 uint64_t reserved = 0;
546 uint64_t write_limit = (zfs_write_limit_override ?
547 zfs_write_limit_override : dp->dp_write_limit);
549 if (zfs_no_write_throttle) {
550 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
556 * Check to see if we have exceeded the maximum allowed IO for
557 * this transaction group. We can do this without locks since
558 * a little slop here is ok. Note that we do the reserved check
559 * with only half the requested reserve: this is because the
560 * reserve requests are worst-case, and we really don't want to
561 * throttle based off of worst-case estimates.
563 if (write_limit > 0) {
564 reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
565 + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
567 if (reserved && reserved > write_limit)
571 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
574 * If this transaction group is over 7/8ths capacity, delay
575 * the caller 1 clock tick. This will slow down the "fill"
576 * rate until the sync process can catch up with us.
578 if (reserved && reserved > (write_limit - (write_limit >> 3)))
579 txg_delay(dp, tx->tx_txg, 1);
585 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
587 ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
588 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
592 dsl_pool_memory_pressure(dsl_pool_t *dp)
594 uint64_t space_inuse = 0;
597 if (dp->dp_write_limit == zfs_write_limit_min)
600 for (i = 0; i < TXG_SIZE; i++) {
601 space_inuse += dp->dp_space_towrite[i];
602 space_inuse += dp->dp_tempreserved[i];
604 dp->dp_write_limit = MAX(zfs_write_limit_min,
605 MIN(dp->dp_write_limit, space_inuse / 4));
609 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
612 mutex_enter(&dp->dp_lock);
613 dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
614 mutex_exit(&dp->dp_lock);
620 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
623 dsl_dataset_t *ds, *prev = NULL;
625 dsl_pool_t *dp = spa_get_dsl(spa);
627 err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
631 while (ds->ds_phys->ds_prev_snap_obj != 0) {
632 err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
635 dsl_dataset_rele(ds, FTAG);
639 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
641 dsl_dataset_rele(ds, FTAG);
647 prev = dp->dp_origin_snap;
650 * The $ORIGIN can't have any data, or the accounting
653 ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
655 /* The origin doesn't get attached to itself */
656 if (ds->ds_object == prev->ds_object) {
657 dsl_dataset_rele(ds, FTAG);
661 dmu_buf_will_dirty(ds->ds_dbuf, tx);
662 ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
663 ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
665 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
666 ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
668 dmu_buf_will_dirty(prev->ds_dbuf, tx);
669 prev->ds_phys->ds_num_children++;
671 if (ds->ds_phys->ds_next_snap_obj == 0) {
672 ASSERT(ds->ds_prev == NULL);
673 VERIFY(0 == dsl_dataset_hold_obj(dp,
674 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
678 ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
679 ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
681 if (prev->ds_phys->ds_next_clones_obj == 0) {
682 dmu_buf_will_dirty(prev->ds_dbuf, tx);
683 prev->ds_phys->ds_next_clones_obj =
684 zap_create(dp->dp_meta_objset,
685 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
687 VERIFY(0 == zap_add_int(dp->dp_meta_objset,
688 prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
690 dsl_dataset_rele(ds, FTAG);
691 if (prev != dp->dp_origin_snap)
692 dsl_dataset_rele(prev, FTAG);
697 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
699 ASSERT(dmu_tx_is_syncing(tx));
700 ASSERT(dp->dp_origin_snap != NULL);
702 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
703 tx, DS_FIND_CHILDREN));
708 upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
712 dsl_pool_t *dp = spa_get_dsl(spa);
713 objset_t *mos = dp->dp_meta_objset;
715 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
717 if (ds->ds_dir->dd_phys->dd_origin_obj) {
718 dsl_dataset_t *origin;
720 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
721 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin));
723 if (origin->ds_dir->dd_phys->dd_clones == 0) {
724 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
725 origin->ds_dir->dd_phys->dd_clones = zap_create(mos,
726 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
729 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
730 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
732 dsl_dataset_rele(origin, FTAG);
735 dsl_dataset_rele(ds, FTAG);
740 dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
742 ASSERT(dmu_tx_is_syncing(tx));
745 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx);
746 VERIFY(0 == dsl_pool_open_special_dir(dp,
747 FREE_DIR_NAME, &dp->dp_free_dir));
750 * We can't use bpobj_alloc(), because spa_version() still
751 * returns the old version, and we need a new-version bpobj with
752 * subobj support. So call dmu_object_alloc() directly.
754 obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ,
755 SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx);
756 VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
757 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
758 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
759 dp->dp_meta_objset, obj));
761 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL,
762 upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN));
766 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
771 ASSERT(dmu_tx_is_syncing(tx));
772 ASSERT(dp->dp_origin_snap == NULL);
774 /* create the origin dir, ds, & snap-ds */
775 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
776 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
778 VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
779 dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx);
780 VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
781 dp, &dp->dp_origin_snap));
782 dsl_dataset_rele(ds, FTAG);
783 rw_exit(&dp->dp_config_rwlock);
787 dsl_pool_vnrele_taskq(dsl_pool_t *dp)
789 return (dp->dp_vnrele_taskq);
793 * Walk through the pool-wide zap object of temporary snapshot user holds
797 dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp)
801 objset_t *mos = dp->dp_meta_objset;
802 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
806 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
808 for (zap_cursor_init(&zc, mos, zapobj);
809 zap_cursor_retrieve(&zc, &za) == 0;
810 zap_cursor_advance(&zc)) {
814 htag = strchr(za.za_name, '-');
817 dsobj = strtonum(za.za_name, NULL);
818 (void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE);
820 zap_cursor_fini(&zc);
824 * Create the pool-wide zap object for storing temporary snapshot holds.
827 dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx)
829 objset_t *mos = dp->dp_meta_objset;
831 ASSERT(dp->dp_tmp_userrefs_obj == 0);
832 ASSERT(dmu_tx_is_syncing(tx));
834 dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS,
835 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx);
839 dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
840 const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding)
842 objset_t *mos = dp->dp_meta_objset;
843 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
847 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
848 ASSERT(dmu_tx_is_syncing(tx));
851 * If the pool was created prior to SPA_VERSION_USERREFS, the
852 * zap object for temporary holds might not exist yet.
856 dsl_pool_user_hold_create_obj(dp, tx);
857 zapobj = dp->dp_tmp_userrefs_obj;
863 name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag);
865 error = zap_add(mos, zapobj, name, 8, 1, now, tx);
867 error = zap_remove(mos, zapobj, name, tx);
874 * Add a temporary hold for the given dataset object and tag.
877 dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
878 uint64_t *now, dmu_tx_t *tx)
880 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE));
884 * Release a temporary hold for the given dataset object and tag.
887 dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
890 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL,