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>
45 #include <sys/zil_impl.h>
47 int zfs_no_write_throttle = 0;
48 int zfs_write_limit_shift = 3; /* 1/8th of physical memory */
49 int zfs_txg_synctime_ms = 1000; /* target millisecs to sync a txg */
51 uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */
52 uint64_t zfs_write_limit_max = 0; /* max data payload per txg */
53 uint64_t zfs_write_limit_inflated = 0;
54 uint64_t zfs_write_limit_override = 0;
56 kmutex_t zfs_write_limit_lock;
58 static pgcnt_t old_physmem = 0;
60 SYSCTL_DECL(_vfs_zfs);
61 TUNABLE_INT("vfs.zfs.no_write_throttle", &zfs_no_write_throttle);
62 SYSCTL_INT(_vfs_zfs, OID_AUTO, no_write_throttle, CTLFLAG_RDTUN,
63 &zfs_no_write_throttle, 0, "");
64 TUNABLE_INT("vfs.zfs.write_limit_shift", &zfs_write_limit_shift);
65 SYSCTL_INT(_vfs_zfs, OID_AUTO, write_limit_shift, CTLFLAG_RDTUN,
66 &zfs_write_limit_shift, 0, "2^N of physical memory");
67 SYSCTL_DECL(_vfs_zfs_txg);
68 TUNABLE_INT("vfs.zfs.txg.synctime_ms", &zfs_txg_synctime_ms);
69 SYSCTL_INT(_vfs_zfs_txg, OID_AUTO, synctime_ms, CTLFLAG_RDTUN,
70 &zfs_txg_synctime_ms, 0, "Target milliseconds to sync a txg");
72 TUNABLE_QUAD("vfs.zfs.write_limit_min", &zfs_write_limit_min);
73 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_min, CTLFLAG_RDTUN,
74 &zfs_write_limit_min, 0, "Minimum write limit");
75 TUNABLE_QUAD("vfs.zfs.write_limit_max", &zfs_write_limit_max);
76 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_max, CTLFLAG_RDTUN,
77 &zfs_write_limit_max, 0, "Maximum data payload per txg");
78 TUNABLE_QUAD("vfs.zfs.write_limit_inflated", &zfs_write_limit_inflated);
79 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_inflated, CTLFLAG_RDTUN,
80 &zfs_write_limit_inflated, 0, "");
81 TUNABLE_QUAD("vfs.zfs.write_limit_override", &zfs_write_limit_override);
82 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, write_limit_override, CTLFLAG_RDTUN,
83 &zfs_write_limit_override, 0, "");
86 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
91 err = zap_lookup(dp->dp_meta_objset,
92 dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
93 name, sizeof (obj), 1, &obj);
97 return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
101 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
104 blkptr_t *bp = spa_get_rootblkptr(spa);
106 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
108 dp->dp_meta_rootbp = *bp;
109 rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
110 dp->dp_write_limit = zfs_write_limit_min;
113 txg_list_create(&dp->dp_dirty_datasets,
114 offsetof(dsl_dataset_t, ds_dirty_link));
115 txg_list_create(&dp->dp_dirty_zilogs,
116 offsetof(zilog_t, zl_dirty_link));
117 txg_list_create(&dp->dp_dirty_dirs,
118 offsetof(dsl_dir_t, dd_dirty_link));
119 txg_list_create(&dp->dp_sync_tasks,
120 offsetof(dsl_sync_task_group_t, dstg_node));
122 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
124 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
131 dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
134 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
136 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp,
137 &dp->dp_meta_objset);
147 dsl_pool_open(dsl_pool_t *dp)
154 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
155 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
156 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
157 &dp->dp_root_dir_obj);
161 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
162 NULL, dp, &dp->dp_root_dir);
166 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
170 if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) {
171 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
174 err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
177 err = dsl_dataset_hold_obj(dp,
178 ds->ds_phys->ds_prev_snap_obj, dp,
179 &dp->dp_origin_snap);
180 dsl_dataset_rele(ds, FTAG);
182 dsl_dir_close(dd, dp);
187 if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
188 err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME,
193 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
194 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj);
197 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
198 dp->dp_meta_objset, obj));
201 if (spa_feature_is_active(dp->dp_spa,
202 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
203 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
204 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
210 if (spa_feature_is_active(dp->dp_spa,
211 &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ])) {
212 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
213 DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
214 &dp->dp_empty_bpobj);
219 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
220 DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1,
221 &dp->dp_tmp_userrefs_obj);
227 err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg);
230 rw_exit(&dp->dp_config_rwlock);
235 dsl_pool_close(dsl_pool_t *dp)
237 /* drop our references from dsl_pool_open() */
240 * Since we held the origin_snap from "syncing" context (which
241 * includes pool-opening context), it actually only got a "ref"
242 * and not a hold, so just drop that here.
244 if (dp->dp_origin_snap)
245 dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
247 dsl_dir_close(dp->dp_mos_dir, dp);
249 dsl_dir_close(dp->dp_free_dir, dp);
251 dsl_dir_close(dp->dp_root_dir, dp);
253 bpobj_close(&dp->dp_free_bpobj);
255 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
256 if (dp->dp_meta_objset)
257 dmu_objset_evict(dp->dp_meta_objset);
259 txg_list_destroy(&dp->dp_dirty_datasets);
260 txg_list_destroy(&dp->dp_dirty_zilogs);
261 txg_list_destroy(&dp->dp_sync_tasks);
262 txg_list_destroy(&dp->dp_dirty_dirs);
264 arc_flush(dp->dp_spa);
267 rw_destroy(&dp->dp_config_rwlock);
268 mutex_destroy(&dp->dp_lock);
269 taskq_destroy(dp->dp_vnrele_taskq);
271 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
272 kmem_free(dp, sizeof (dsl_pool_t));
276 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
279 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
280 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
285 /* create and open the MOS (meta-objset) */
286 dp->dp_meta_objset = dmu_objset_create_impl(spa,
287 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
289 /* create the pool directory */
290 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
291 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
294 /* Initialize scan structures */
295 VERIFY3U(0, ==, dsl_scan_init(dp, txg));
297 /* create and open the root dir */
298 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
299 VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
300 NULL, dp, &dp->dp_root_dir));
302 /* create and open the meta-objset dir */
303 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
304 VERIFY(0 == dsl_pool_open_special_dir(dp,
305 MOS_DIR_NAME, &dp->dp_mos_dir));
307 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
308 /* create and open the free dir */
309 (void) dsl_dir_create_sync(dp, dp->dp_root_dir,
311 VERIFY(0 == dsl_pool_open_special_dir(dp,
312 FREE_DIR_NAME, &dp->dp_free_dir));
314 /* create and open the free_bplist */
315 obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx);
316 VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
317 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
318 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
319 dp->dp_meta_objset, obj));
322 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
323 dsl_pool_create_origin(dp, tx);
325 /* create the root dataset */
326 obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
328 /* create the root objset */
329 VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
330 os = dmu_objset_create_impl(dp->dp_spa, ds,
331 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
333 zfs_create_fs(os, kcred, zplprops, tx);
335 dsl_dataset_rele(ds, FTAG);
343 * Account for the meta-objset space in its placeholder dsl_dir.
346 dsl_pool_mos_diduse_space(dsl_pool_t *dp,
347 int64_t used, int64_t comp, int64_t uncomp)
349 ASSERT3U(comp, ==, uncomp); /* it's all metadata */
350 mutex_enter(&dp->dp_lock);
351 dp->dp_mos_used_delta += used;
352 dp->dp_mos_compressed_delta += comp;
353 dp->dp_mos_uncompressed_delta += uncomp;
354 mutex_exit(&dp->dp_lock);
358 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
360 dsl_deadlist_t *dl = arg;
361 dsl_pool_t *dp = dmu_objset_pool(dl->dl_os);
362 rw_enter(&dp->dp_config_rwlock, RW_READER);
363 dsl_deadlist_insert(dl, bp, tx);
364 rw_exit(&dp->dp_config_rwlock);
369 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
375 objset_t *mos = dp->dp_meta_objset;
376 hrtime_t start, write_time;
377 uint64_t data_written;
379 list_t synced_datasets;
381 list_create(&synced_datasets, sizeof (dsl_dataset_t),
382 offsetof(dsl_dataset_t, ds_synced_link));
385 * We need to copy dp_space_towrite() before doing
386 * dsl_sync_task_group_sync(), because
387 * dsl_dataset_snapshot_reserve_space() will increase
388 * dp_space_towrite but not actually write anything.
390 data_written = dp->dp_space_towrite[txg & TXG_MASK];
392 tx = dmu_tx_create_assigned(dp, txg);
394 dp->dp_read_overhead = 0;
397 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
398 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
400 * We must not sync any non-MOS datasets twice, because
401 * we may have taken a snapshot of them. However, we
402 * may sync newly-created datasets on pass 2.
404 ASSERT(!list_link_active(&ds->ds_synced_link));
405 list_insert_tail(&synced_datasets, ds);
406 dsl_dataset_sync(ds, zio, tx);
408 DTRACE_PROBE(pool_sync__1setup);
411 write_time = gethrtime() - start;
413 DTRACE_PROBE(pool_sync__2rootzio);
416 * After the data blocks have been written (ensured by the zio_wait()
417 * above), update the user/group space accounting.
419 for (ds = list_head(&synced_datasets); ds;
420 ds = list_next(&synced_datasets, ds))
421 dmu_objset_do_userquota_updates(ds->ds_objset, tx);
424 * Sync the datasets again to push out the changes due to
425 * userspace updates. This must be done before we process the
426 * sync tasks, so that any snapshots will have the correct
427 * user accounting information (and we won't get confused
428 * about which blocks are part of the snapshot).
430 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
431 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
432 ASSERT(list_link_active(&ds->ds_synced_link));
433 dmu_buf_rele(ds->ds_dbuf, ds);
434 dsl_dataset_sync(ds, zio, tx);
439 * Now that the datasets have been completely synced, we can
440 * clean up our in-memory structures accumulated while syncing:
442 * - move dead blocks from the pending deadlist to the on-disk deadlist
443 * - release hold from dsl_dataset_dirty()
445 while (ds = list_remove_head(&synced_datasets)) {
446 objset_t *os = ds->ds_objset;
447 bplist_iterate(&ds->ds_pending_deadlist,
448 deadlist_enqueue_cb, &ds->ds_deadlist, tx);
449 ASSERT(!dmu_objset_is_dirty(os, txg));
450 dmu_buf_rele(ds->ds_dbuf, ds);
454 while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
455 dsl_dir_sync(dd, tx);
456 write_time += gethrtime() - start;
459 * The MOS's space is accounted for in the pool/$MOS
460 * (dp_mos_dir). We can't modify the mos while we're syncing
461 * it, so we remember the deltas and apply them here.
463 if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 ||
464 dp->dp_mos_uncompressed_delta != 0) {
465 dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD,
466 dp->dp_mos_used_delta,
467 dp->dp_mos_compressed_delta,
468 dp->dp_mos_uncompressed_delta, tx);
469 dp->dp_mos_used_delta = 0;
470 dp->dp_mos_compressed_delta = 0;
471 dp->dp_mos_uncompressed_delta = 0;
475 if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
476 list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) {
477 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
478 dmu_objset_sync(mos, zio, tx);
481 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
482 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
484 write_time += gethrtime() - start;
485 DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
486 hrtime_t, dp->dp_read_overhead);
487 write_time -= dp->dp_read_overhead;
490 * If we modify a dataset in the same txg that we want to destroy it,
491 * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it.
492 * dsl_dir_destroy_check() will fail if there are unexpected holds.
493 * Therefore, we want to sync the MOS (thus syncing the dd_dbuf
494 * and clearing the hold on it) before we process the sync_tasks.
495 * The MOS data dirtied by the sync_tasks will be synced on the next
498 DTRACE_PROBE(pool_sync__3task);
499 if (!txg_list_empty(&dp->dp_sync_tasks, txg)) {
500 dsl_sync_task_group_t *dstg;
502 * No more sync tasks should have been added while we
505 ASSERT(spa_sync_pass(dp->dp_spa) == 1);
506 while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
507 dsl_sync_task_group_sync(dstg, tx);
512 dp->dp_space_towrite[txg & TXG_MASK] = 0;
513 ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
516 * If the write limit max has not been explicitly set, set it
517 * to a fraction of available physical memory (default 1/8th).
518 * Note that we must inflate the limit because the spa
519 * inflates write sizes to account for data replication.
520 * Check this each sync phase to catch changing memory size.
522 if (physmem != old_physmem && zfs_write_limit_shift) {
523 mutex_enter(&zfs_write_limit_lock);
524 old_physmem = physmem;
525 zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
526 zfs_write_limit_inflated = MAX(zfs_write_limit_min,
527 spa_get_asize(dp->dp_spa, zfs_write_limit_max));
528 mutex_exit(&zfs_write_limit_lock);
532 * Attempt to keep the sync time consistent by adjusting the
533 * amount of write traffic allowed into each transaction group.
534 * Weight the throughput calculation towards the current value:
535 * thru = 3/4 old_thru + 1/4 new_thru
537 * Note: write_time is in nanosecs, so write_time/MICROSEC
540 ASSERT(zfs_write_limit_min > 0);
541 if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) {
542 uint64_t throughput = data_written / (write_time / MICROSEC);
544 if (dp->dp_throughput)
545 dp->dp_throughput = throughput / 4 +
546 3 * dp->dp_throughput / 4;
548 dp->dp_throughput = throughput;
549 dp->dp_write_limit = MIN(zfs_write_limit_inflated,
550 MAX(zfs_write_limit_min,
551 dp->dp_throughput * zfs_txg_synctime_ms));
556 dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg)
561 while (zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg)) {
562 ds = dmu_objset_ds(zilog->zl_os);
563 zil_clean(zilog, txg);
564 ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg));
565 dmu_buf_rele(ds->ds_dbuf, zilog);
567 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg));
571 * TRUE if the current thread is the tx_sync_thread or if we
572 * are being called from SPA context during pool initialization.
575 dsl_pool_sync_context(dsl_pool_t *dp)
577 return (curthread == dp->dp_tx.tx_sync_thread ||
578 spa_is_initializing(dp->dp_spa));
582 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
584 uint64_t space, resv;
587 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
589 * XXX The intent log is not accounted for, so it must fit
592 * If we're trying to assess whether it's OK to do a free,
593 * cut the reservation in half to allow forward progress
594 * (e.g. make it possible to rm(1) files from a full pool).
596 space = spa_get_dspace(dp->dp_spa);
597 resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
601 return (space - resv);
605 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
607 uint64_t reserved = 0;
608 uint64_t write_limit = (zfs_write_limit_override ?
609 zfs_write_limit_override : dp->dp_write_limit);
611 if (zfs_no_write_throttle) {
612 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
618 * Check to see if we have exceeded the maximum allowed IO for
619 * this transaction group. We can do this without locks since
620 * a little slop here is ok. Note that we do the reserved check
621 * with only half the requested reserve: this is because the
622 * reserve requests are worst-case, and we really don't want to
623 * throttle based off of worst-case estimates.
625 if (write_limit > 0) {
626 reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
627 + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
629 if (reserved && reserved > write_limit)
633 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
636 * If this transaction group is over 7/8ths capacity, delay
637 * the caller 1 clock tick. This will slow down the "fill"
638 * rate until the sync process can catch up with us.
640 if (reserved && reserved > (write_limit - (write_limit >> 3)))
641 txg_delay(dp, tx->tx_txg, 1);
647 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
649 ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
650 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
654 dsl_pool_memory_pressure(dsl_pool_t *dp)
656 uint64_t space_inuse = 0;
659 if (dp->dp_write_limit == zfs_write_limit_min)
662 for (i = 0; i < TXG_SIZE; i++) {
663 space_inuse += dp->dp_space_towrite[i];
664 space_inuse += dp->dp_tempreserved[i];
666 dp->dp_write_limit = MAX(zfs_write_limit_min,
667 MIN(dp->dp_write_limit, space_inuse / 4));
671 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
674 mutex_enter(&dp->dp_lock);
675 dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
676 mutex_exit(&dp->dp_lock);
682 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
685 dsl_dataset_t *ds, *prev = NULL;
687 dsl_pool_t *dp = spa_get_dsl(spa);
689 err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
693 while (ds->ds_phys->ds_prev_snap_obj != 0) {
694 err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
697 dsl_dataset_rele(ds, FTAG);
701 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
703 dsl_dataset_rele(ds, FTAG);
709 prev = dp->dp_origin_snap;
712 * The $ORIGIN can't have any data, or the accounting
715 ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
717 /* The origin doesn't get attached to itself */
718 if (ds->ds_object == prev->ds_object) {
719 dsl_dataset_rele(ds, FTAG);
723 dmu_buf_will_dirty(ds->ds_dbuf, tx);
724 ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
725 ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
727 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
728 ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
730 dmu_buf_will_dirty(prev->ds_dbuf, tx);
731 prev->ds_phys->ds_num_children++;
733 if (ds->ds_phys->ds_next_snap_obj == 0) {
734 ASSERT(ds->ds_prev == NULL);
735 VERIFY(0 == dsl_dataset_hold_obj(dp,
736 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
740 ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
741 ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
743 if (prev->ds_phys->ds_next_clones_obj == 0) {
744 dmu_buf_will_dirty(prev->ds_dbuf, tx);
745 prev->ds_phys->ds_next_clones_obj =
746 zap_create(dp->dp_meta_objset,
747 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
749 VERIFY(0 == zap_add_int(dp->dp_meta_objset,
750 prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
752 dsl_dataset_rele(ds, FTAG);
753 if (prev != dp->dp_origin_snap)
754 dsl_dataset_rele(prev, FTAG);
759 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
761 ASSERT(dmu_tx_is_syncing(tx));
762 ASSERT(dp->dp_origin_snap != NULL);
764 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
765 tx, DS_FIND_CHILDREN));
770 upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
774 dsl_pool_t *dp = spa_get_dsl(spa);
775 objset_t *mos = dp->dp_meta_objset;
777 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
779 if (ds->ds_dir->dd_phys->dd_origin_obj) {
780 dsl_dataset_t *origin;
782 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
783 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin));
785 if (origin->ds_dir->dd_phys->dd_clones == 0) {
786 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
787 origin->ds_dir->dd_phys->dd_clones = zap_create(mos,
788 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
791 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
792 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
794 dsl_dataset_rele(origin, FTAG);
797 dsl_dataset_rele(ds, FTAG);
802 dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
804 ASSERT(dmu_tx_is_syncing(tx));
807 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx);
808 VERIFY(0 == dsl_pool_open_special_dir(dp,
809 FREE_DIR_NAME, &dp->dp_free_dir));
812 * We can't use bpobj_alloc(), because spa_version() still
813 * returns the old version, and we need a new-version bpobj with
814 * subobj support. So call dmu_object_alloc() directly.
816 obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ,
817 SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx);
818 VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
819 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
820 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
821 dp->dp_meta_objset, obj));
823 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL,
824 upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN));
828 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
833 ASSERT(dmu_tx_is_syncing(tx));
834 ASSERT(dp->dp_origin_snap == NULL);
836 /* create the origin dir, ds, & snap-ds */
837 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
838 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
840 VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
841 dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx);
842 VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
843 dp, &dp->dp_origin_snap));
844 dsl_dataset_rele(ds, FTAG);
845 rw_exit(&dp->dp_config_rwlock);
849 dsl_pool_vnrele_taskq(dsl_pool_t *dp)
851 return (dp->dp_vnrele_taskq);
855 * Walk through the pool-wide zap object of temporary snapshot user holds
859 dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp)
863 objset_t *mos = dp->dp_meta_objset;
864 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
868 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
870 for (zap_cursor_init(&zc, mos, zapobj);
871 zap_cursor_retrieve(&zc, &za) == 0;
872 zap_cursor_advance(&zc)) {
876 htag = strchr(za.za_name, '-');
879 dsobj = strtonum(za.za_name, NULL);
880 (void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE);
882 zap_cursor_fini(&zc);
886 * Create the pool-wide zap object for storing temporary snapshot holds.
889 dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx)
891 objset_t *mos = dp->dp_meta_objset;
893 ASSERT(dp->dp_tmp_userrefs_obj == 0);
894 ASSERT(dmu_tx_is_syncing(tx));
896 dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS,
897 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx);
901 dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
902 const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding)
904 objset_t *mos = dp->dp_meta_objset;
905 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
909 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
910 ASSERT(dmu_tx_is_syncing(tx));
913 * If the pool was created prior to SPA_VERSION_USERREFS, the
914 * zap object for temporary holds might not exist yet.
918 dsl_pool_user_hold_create_obj(dp, tx);
919 zapobj = dp->dp_tmp_userrefs_obj;
925 name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag);
927 error = zap_add(mos, zapobj, name, 8, 1, now, tx);
929 error = zap_remove(mos, zapobj, name, tx);
936 * Add a temporary hold for the given dataset object and tag.
939 dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
940 uint64_t *now, dmu_tx_t *tx)
942 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE));
946 * Release a temporary hold for the given dataset object and tag.
949 dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
952 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL,