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.
25 #include <sys/dsl_pool.h>
26 #include <sys/dsl_dataset.h>
27 #include <sys/dsl_prop.h>
28 #include <sys/dsl_dir.h>
29 #include <sys/dsl_synctask.h>
30 #include <sys/dsl_scan.h>
31 #include <sys/dnode.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dmu_objset.h>
37 #include <sys/zfs_context.h>
38 #include <sys/fs/zfs.h>
39 #include <sys/zfs_znode.h>
40 #include <sys/spa_impl.h>
41 #include <sys/dsl_deadlist.h>
43 int zfs_no_write_throttle = 0;
44 int zfs_write_limit_shift = 3; /* 1/8th of physical memory */
45 int zfs_txg_synctime_ms = 1000; /* target millisecs to sync a txg */
47 uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */
48 uint64_t zfs_write_limit_max = 0; /* max data payload per txg */
49 uint64_t zfs_write_limit_inflated = 0;
50 uint64_t zfs_write_limit_override = 0;
52 kmutex_t zfs_write_limit_lock;
54 static pgcnt_t old_physmem = 0;
56 SYSCTL_DECL(_vfs_zfs);
57 TUNABLE_INT("vfs.zfs.no_write_throttle", &zfs_no_write_throttle);
58 SYSCTL_INT(_vfs_zfs, OID_AUTO, no_write_throttle, CTLFLAG_RDTUN,
59 &zfs_no_write_throttle, 0, "");
60 TUNABLE_INT("vfs.zfs.write_limit_shift", &zfs_write_limit_shift);
61 SYSCTL_INT(_vfs_zfs, OID_AUTO, write_limit_shift, CTLFLAG_RDTUN,
62 &zfs_write_limit_shift, 0, "2^N of physical memory");
63 SYSCTL_DECL(_vfs_zfs_txg);
64 TUNABLE_INT("vfs.zfs.txg.synctime_ms", &zfs_txg_synctime_ms);
65 SYSCTL_INT(_vfs_zfs_txg, OID_AUTO, synctime_ms, CTLFLAG_RDTUN,
66 &zfs_txg_synctime_ms, 0, "Target milliseconds to sync a txg");
68 TUNABLE_QUAD("vfs.zfs.write_limit_min", &zfs_write_limit_min);
69 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_min, CTLFLAG_RDTUN,
70 &zfs_write_limit_min, 0, "Minimum write limit");
71 TUNABLE_QUAD("vfs.zfs.write_limit_max", &zfs_write_limit_max);
72 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_max, CTLFLAG_RDTUN,
73 &zfs_write_limit_max, 0, "Maximum data payload per txg");
74 TUNABLE_QUAD("vfs.zfs.write_limit_inflated", &zfs_write_limit_inflated);
75 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_inflated, CTLFLAG_RDTUN,
76 &zfs_write_limit_inflated, 0, "");
77 TUNABLE_QUAD("vfs.zfs.write_limit_override", &zfs_write_limit_override);
78 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_override, CTLFLAG_RDTUN,
79 &zfs_write_limit_override, 0, "");
82 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
87 err = zap_lookup(dp->dp_meta_objset,
88 dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
89 name, sizeof (obj), 1, &obj);
93 return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
97 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
100 blkptr_t *bp = spa_get_rootblkptr(spa);
102 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
104 dp->dp_meta_rootbp = *bp;
105 rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
106 dp->dp_write_limit = zfs_write_limit_min;
109 txg_list_create(&dp->dp_dirty_datasets,
110 offsetof(dsl_dataset_t, ds_dirty_link));
111 txg_list_create(&dp->dp_dirty_dirs,
112 offsetof(dsl_dir_t, dd_dirty_link));
113 txg_list_create(&dp->dp_sync_tasks,
114 offsetof(dsl_sync_task_group_t, dstg_node));
115 list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t),
116 offsetof(dsl_dataset_t, ds_synced_link));
118 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
120 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
127 dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
130 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
135 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
136 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp,
137 &dp->dp_meta_objset);
141 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
142 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
143 &dp->dp_root_dir_obj);
147 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
148 NULL, dp, &dp->dp_root_dir);
152 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
156 if (spa_version(spa) >= SPA_VERSION_ORIGIN) {
157 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
160 err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
163 err = dsl_dataset_hold_obj(dp,
164 ds->ds_phys->ds_prev_snap_obj, dp,
165 &dp->dp_origin_snap);
166 dsl_dataset_rele(ds, FTAG);
168 dsl_dir_close(dd, dp);
173 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
174 err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME,
179 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
180 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj);
183 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
184 dp->dp_meta_objset, obj));
187 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
188 DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1,
189 &dp->dp_tmp_userrefs_obj);
195 err = dsl_scan_init(dp, txg);
198 rw_exit(&dp->dp_config_rwlock);
208 dsl_pool_close(dsl_pool_t *dp)
210 /* drop our references from dsl_pool_open() */
213 * Since we held the origin_snap from "syncing" context (which
214 * includes pool-opening context), it actually only got a "ref"
215 * and not a hold, so just drop that here.
217 if (dp->dp_origin_snap)
218 dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
220 dsl_dir_close(dp->dp_mos_dir, dp);
222 dsl_dir_close(dp->dp_free_dir, dp);
224 dsl_dir_close(dp->dp_root_dir, dp);
226 bpobj_close(&dp->dp_free_bpobj);
228 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
229 if (dp->dp_meta_objset)
230 dmu_objset_evict(dp->dp_meta_objset);
232 txg_list_destroy(&dp->dp_dirty_datasets);
233 txg_list_destroy(&dp->dp_sync_tasks);
234 txg_list_destroy(&dp->dp_dirty_dirs);
235 list_destroy(&dp->dp_synced_datasets);
237 arc_flush(dp->dp_spa);
240 rw_destroy(&dp->dp_config_rwlock);
241 mutex_destroy(&dp->dp_lock);
242 taskq_destroy(dp->dp_vnrele_taskq);
244 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
245 kmem_free(dp, sizeof (dsl_pool_t));
249 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
252 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
253 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
258 /* create and open the MOS (meta-objset) */
259 dp->dp_meta_objset = dmu_objset_create_impl(spa,
260 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
262 /* create the pool directory */
263 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
264 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
265 ASSERT3U(err, ==, 0);
267 /* Initialize scan structures */
268 VERIFY3U(0, ==, dsl_scan_init(dp, txg));
270 /* create and open the root dir */
271 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
272 VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
273 NULL, dp, &dp->dp_root_dir));
275 /* create and open the meta-objset dir */
276 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
277 VERIFY(0 == dsl_pool_open_special_dir(dp,
278 MOS_DIR_NAME, &dp->dp_mos_dir));
280 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
281 /* create and open the free dir */
282 (void) dsl_dir_create_sync(dp, dp->dp_root_dir,
284 VERIFY(0 == dsl_pool_open_special_dir(dp,
285 FREE_DIR_NAME, &dp->dp_free_dir));
287 /* create and open the free_bplist */
288 obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx);
289 VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
290 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
291 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
292 dp->dp_meta_objset, obj));
295 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
296 dsl_pool_create_origin(dp, tx);
298 /* create the root dataset */
299 obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
301 /* create the root objset */
302 VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
303 os = dmu_objset_create_impl(dp->dp_spa, ds,
304 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
306 zfs_create_fs(os, kcred, zplprops, tx);
308 dsl_dataset_rele(ds, FTAG);
316 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
318 dsl_deadlist_t *dl = arg;
319 dsl_deadlist_insert(dl, bp, tx);
324 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
330 dsl_sync_task_group_t *dstg;
331 objset_t *mos = dp->dp_meta_objset;
332 hrtime_t start, write_time;
333 uint64_t data_written;
337 * We need to copy dp_space_towrite() before doing
338 * dsl_sync_task_group_sync(), because
339 * dsl_dataset_snapshot_reserve_space() will increase
340 * dp_space_towrite but not actually write anything.
342 data_written = dp->dp_space_towrite[txg & TXG_MASK];
344 tx = dmu_tx_create_assigned(dp, txg);
346 dp->dp_read_overhead = 0;
349 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
350 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
352 * We must not sync any non-MOS datasets twice, because
353 * we may have taken a snapshot of them. However, we
354 * may sync newly-created datasets on pass 2.
356 ASSERT(!list_link_active(&ds->ds_synced_link));
357 list_insert_tail(&dp->dp_synced_datasets, ds);
358 dsl_dataset_sync(ds, zio, tx);
360 DTRACE_PROBE(pool_sync__1setup);
363 write_time = gethrtime() - start;
365 DTRACE_PROBE(pool_sync__2rootzio);
367 for (ds = list_head(&dp->dp_synced_datasets); ds;
368 ds = list_next(&dp->dp_synced_datasets, ds))
369 dmu_objset_do_userquota_updates(ds->ds_objset, tx);
372 * Sync the datasets again to push out the changes due to
373 * userspace updates. This must be done before we process the
374 * sync tasks, because that could cause a snapshot of a dataset
375 * whose ds_bp will be rewritten when we do this 2nd sync.
377 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
378 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
379 ASSERT(list_link_active(&ds->ds_synced_link));
380 dmu_buf_rele(ds->ds_dbuf, ds);
381 dsl_dataset_sync(ds, zio, tx);
386 * Move dead blocks from the pending deadlist to the on-disk
389 for (ds = list_head(&dp->dp_synced_datasets); ds;
390 ds = list_next(&dp->dp_synced_datasets, ds)) {
391 bplist_iterate(&ds->ds_pending_deadlist,
392 deadlist_enqueue_cb, &ds->ds_deadlist, tx);
395 while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) {
397 * No more sync tasks should have been added while we
400 ASSERT(spa_sync_pass(dp->dp_spa) == 1);
401 dsl_sync_task_group_sync(dstg, tx);
403 DTRACE_PROBE(pool_sync__3task);
406 while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
407 dsl_dir_sync(dd, tx);
408 write_time += gethrtime() - start;
411 if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
412 list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) {
413 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
414 dmu_objset_sync(mos, zio, tx);
417 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
418 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
420 write_time += gethrtime() - start;
421 DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
422 hrtime_t, dp->dp_read_overhead);
423 write_time -= dp->dp_read_overhead;
427 dp->dp_space_towrite[txg & TXG_MASK] = 0;
428 ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
431 * If the write limit max has not been explicitly set, set it
432 * to a fraction of available physical memory (default 1/8th).
433 * Note that we must inflate the limit because the spa
434 * inflates write sizes to account for data replication.
435 * Check this each sync phase to catch changing memory size.
437 if (physmem != old_physmem && zfs_write_limit_shift) {
438 mutex_enter(&zfs_write_limit_lock);
439 old_physmem = physmem;
440 zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
441 zfs_write_limit_inflated = MAX(zfs_write_limit_min,
442 spa_get_asize(dp->dp_spa, zfs_write_limit_max));
443 mutex_exit(&zfs_write_limit_lock);
447 * Attempt to keep the sync time consistent by adjusting the
448 * amount of write traffic allowed into each transaction group.
449 * Weight the throughput calculation towards the current value:
450 * thru = 3/4 old_thru + 1/4 new_thru
452 * Note: write_time is in nanosecs, so write_time/MICROSEC
455 ASSERT(zfs_write_limit_min > 0);
456 if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) {
457 uint64_t throughput = data_written / (write_time / MICROSEC);
459 if (dp->dp_throughput)
460 dp->dp_throughput = throughput / 4 +
461 3 * dp->dp_throughput / 4;
463 dp->dp_throughput = throughput;
464 dp->dp_write_limit = MIN(zfs_write_limit_inflated,
465 MAX(zfs_write_limit_min,
466 dp->dp_throughput * zfs_txg_synctime_ms));
471 dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg)
476 while (ds = list_head(&dp->dp_synced_datasets)) {
477 list_remove(&dp->dp_synced_datasets, ds);
479 zil_clean(os->os_zil, txg);
480 ASSERT(!dmu_objset_is_dirty(os, txg));
481 dmu_buf_rele(ds->ds_dbuf, ds);
483 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg));
487 * TRUE if the current thread is the tx_sync_thread or if we
488 * are being called from SPA context during pool initialization.
491 dsl_pool_sync_context(dsl_pool_t *dp)
493 return (curthread == dp->dp_tx.tx_sync_thread ||
494 spa_get_dsl(dp->dp_spa) == NULL);
498 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
500 uint64_t space, resv;
503 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
505 * XXX The intent log is not accounted for, so it must fit
508 * If we're trying to assess whether it's OK to do a free,
509 * cut the reservation in half to allow forward progress
510 * (e.g. make it possible to rm(1) files from a full pool).
512 space = spa_get_dspace(dp->dp_spa);
513 resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
517 return (space - resv);
521 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
523 uint64_t reserved = 0;
524 uint64_t write_limit = (zfs_write_limit_override ?
525 zfs_write_limit_override : dp->dp_write_limit);
527 if (zfs_no_write_throttle) {
528 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
534 * Check to see if we have exceeded the maximum allowed IO for
535 * this transaction group. We can do this without locks since
536 * a little slop here is ok. Note that we do the reserved check
537 * with only half the requested reserve: this is because the
538 * reserve requests are worst-case, and we really don't want to
539 * throttle based off of worst-case estimates.
541 if (write_limit > 0) {
542 reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
543 + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
545 if (reserved && reserved > write_limit)
549 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
552 * If this transaction group is over 7/8ths capacity, delay
553 * the caller 1 clock tick. This will slow down the "fill"
554 * rate until the sync process can catch up with us.
556 if (reserved && reserved > (write_limit - (write_limit >> 3)))
557 txg_delay(dp, tx->tx_txg, 1);
563 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
565 ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
566 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
570 dsl_pool_memory_pressure(dsl_pool_t *dp)
572 uint64_t space_inuse = 0;
575 if (dp->dp_write_limit == zfs_write_limit_min)
578 for (i = 0; i < TXG_SIZE; i++) {
579 space_inuse += dp->dp_space_towrite[i];
580 space_inuse += dp->dp_tempreserved[i];
582 dp->dp_write_limit = MAX(zfs_write_limit_min,
583 MIN(dp->dp_write_limit, space_inuse / 4));
587 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
590 mutex_enter(&dp->dp_lock);
591 dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
592 mutex_exit(&dp->dp_lock);
598 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
601 dsl_dataset_t *ds, *prev = NULL;
603 dsl_pool_t *dp = spa_get_dsl(spa);
605 err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
609 while (ds->ds_phys->ds_prev_snap_obj != 0) {
610 err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
613 dsl_dataset_rele(ds, FTAG);
617 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
619 dsl_dataset_rele(ds, FTAG);
625 prev = dp->dp_origin_snap;
628 * The $ORIGIN can't have any data, or the accounting
631 ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
633 /* The origin doesn't get attached to itself */
634 if (ds->ds_object == prev->ds_object) {
635 dsl_dataset_rele(ds, FTAG);
639 dmu_buf_will_dirty(ds->ds_dbuf, tx);
640 ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
641 ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
643 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
644 ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
646 dmu_buf_will_dirty(prev->ds_dbuf, tx);
647 prev->ds_phys->ds_num_children++;
649 if (ds->ds_phys->ds_next_snap_obj == 0) {
650 ASSERT(ds->ds_prev == NULL);
651 VERIFY(0 == dsl_dataset_hold_obj(dp,
652 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
656 ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
657 ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
659 if (prev->ds_phys->ds_next_clones_obj == 0) {
660 dmu_buf_will_dirty(prev->ds_dbuf, tx);
661 prev->ds_phys->ds_next_clones_obj =
662 zap_create(dp->dp_meta_objset,
663 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
665 VERIFY(0 == zap_add_int(dp->dp_meta_objset,
666 prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
668 dsl_dataset_rele(ds, FTAG);
669 if (prev != dp->dp_origin_snap)
670 dsl_dataset_rele(prev, FTAG);
675 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
677 ASSERT(dmu_tx_is_syncing(tx));
678 ASSERT(dp->dp_origin_snap != NULL);
680 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
681 tx, DS_FIND_CHILDREN));
686 upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
690 dsl_pool_t *dp = spa_get_dsl(spa);
691 objset_t *mos = dp->dp_meta_objset;
693 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
695 if (ds->ds_dir->dd_phys->dd_origin_obj) {
696 dsl_dataset_t *origin;
698 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
699 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin));
701 if (origin->ds_dir->dd_phys->dd_clones == 0) {
702 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
703 origin->ds_dir->dd_phys->dd_clones = zap_create(mos,
704 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
707 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
708 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
710 dsl_dataset_rele(origin, FTAG);
713 dsl_dataset_rele(ds, FTAG);
718 dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
720 ASSERT(dmu_tx_is_syncing(tx));
723 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx);
724 VERIFY(0 == dsl_pool_open_special_dir(dp,
725 FREE_DIR_NAME, &dp->dp_free_dir));
728 * We can't use bpobj_alloc(), because spa_version() still
729 * returns the old version, and we need a new-version bpobj with
730 * subobj support. So call dmu_object_alloc() directly.
732 obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ,
733 SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx);
734 VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
735 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
736 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
737 dp->dp_meta_objset, obj));
739 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL,
740 upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN));
744 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
749 ASSERT(dmu_tx_is_syncing(tx));
750 ASSERT(dp->dp_origin_snap == NULL);
752 /* create the origin dir, ds, & snap-ds */
753 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
754 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
756 VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
757 dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx);
758 VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
759 dp, &dp->dp_origin_snap));
760 dsl_dataset_rele(ds, FTAG);
761 rw_exit(&dp->dp_config_rwlock);
765 dsl_pool_vnrele_taskq(dsl_pool_t *dp)
767 return (dp->dp_vnrele_taskq);
771 * Walk through the pool-wide zap object of temporary snapshot user holds
775 dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp)
779 objset_t *mos = dp->dp_meta_objset;
780 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
784 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
786 for (zap_cursor_init(&zc, mos, zapobj);
787 zap_cursor_retrieve(&zc, &za) == 0;
788 zap_cursor_advance(&zc)) {
792 htag = strchr(za.za_name, '-');
795 dsobj = strtonum(za.za_name, NULL);
796 (void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE);
798 zap_cursor_fini(&zc);
802 * Create the pool-wide zap object for storing temporary snapshot holds.
805 dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx)
807 objset_t *mos = dp->dp_meta_objset;
809 ASSERT(dp->dp_tmp_userrefs_obj == 0);
810 ASSERT(dmu_tx_is_syncing(tx));
812 dp->dp_tmp_userrefs_obj = zap_create(mos, DMU_OT_USERREFS,
815 VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS,
816 sizeof (uint64_t), 1, &dp->dp_tmp_userrefs_obj, tx) == 0);
820 dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
821 const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding)
823 objset_t *mos = dp->dp_meta_objset;
824 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
828 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
829 ASSERT(dmu_tx_is_syncing(tx));
832 * If the pool was created prior to SPA_VERSION_USERREFS, the
833 * zap object for temporary holds might not exist yet.
837 dsl_pool_user_hold_create_obj(dp, tx);
838 zapobj = dp->dp_tmp_userrefs_obj;
844 name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag);
846 error = zap_add(mos, zapobj, name, 8, 1, now, tx);
848 error = zap_remove(mos, zapobj, name, tx);
855 * Add a temporary hold for the given dataset object and tag.
858 dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
859 uint64_t *now, dmu_tx_t *tx)
861 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE));
865 * Release a temporary hold for the given dataset object and tag.
868 dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
871 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL,