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_QUAD(_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_QUAD(_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_QUAD(_vfs_zfs, OID_AUTO, write_limit_inflated, CTLFLAG_RDTUN,
80 &zfs_write_limit_inflated, 0, "Maximum size of the dynamic write limit");
81 TUNABLE_QUAD("vfs.zfs.write_limit_override", &zfs_write_limit_override);
82 SYSCTL_QUAD(_vfs_zfs, OID_AUTO, write_limit_override, CTLFLAG_RDTUN,
83 &zfs_write_limit_override, 0,
84 "Force a txg if dirty buffers exceed this value (bytes)");
87 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
92 err = zap_lookup(dp->dp_meta_objset,
93 dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
94 name, sizeof (obj), 1, &obj);
98 return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
102 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
105 blkptr_t *bp = spa_get_rootblkptr(spa);
107 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
109 dp->dp_meta_rootbp = *bp;
110 rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
111 dp->dp_write_limit = zfs_write_limit_min;
114 txg_list_create(&dp->dp_dirty_datasets,
115 offsetof(dsl_dataset_t, ds_dirty_link));
116 txg_list_create(&dp->dp_dirty_zilogs,
117 offsetof(zilog_t, zl_dirty_link));
118 txg_list_create(&dp->dp_dirty_dirs,
119 offsetof(dsl_dir_t, dd_dirty_link));
120 txg_list_create(&dp->dp_sync_tasks,
121 offsetof(dsl_sync_task_group_t, dstg_node));
123 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
125 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
132 dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
135 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
137 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp,
138 &dp->dp_meta_objset);
148 dsl_pool_open(dsl_pool_t *dp)
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 if (spa_feature_is_active(dp->dp_spa,
212 &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ])) {
213 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
214 DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
215 &dp->dp_empty_bpobj);
220 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
221 DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1,
222 &dp->dp_tmp_userrefs_obj);
228 err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg);
231 rw_exit(&dp->dp_config_rwlock);
236 dsl_pool_close(dsl_pool_t *dp)
238 /* drop our references from dsl_pool_open() */
241 * Since we held the origin_snap from "syncing" context (which
242 * includes pool-opening context), it actually only got a "ref"
243 * and not a hold, so just drop that here.
245 if (dp->dp_origin_snap)
246 dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
248 dsl_dir_close(dp->dp_mos_dir, dp);
250 dsl_dir_close(dp->dp_free_dir, dp);
252 dsl_dir_close(dp->dp_root_dir, dp);
254 bpobj_close(&dp->dp_free_bpobj);
256 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
257 if (dp->dp_meta_objset)
258 dmu_objset_evict(dp->dp_meta_objset);
260 txg_list_destroy(&dp->dp_dirty_datasets);
261 txg_list_destroy(&dp->dp_dirty_zilogs);
262 txg_list_destroy(&dp->dp_sync_tasks);
263 txg_list_destroy(&dp->dp_dirty_dirs);
265 arc_flush(dp->dp_spa);
268 rw_destroy(&dp->dp_config_rwlock);
269 mutex_destroy(&dp->dp_lock);
270 taskq_destroy(dp->dp_vnrele_taskq);
272 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
273 kmem_free(dp, sizeof (dsl_pool_t));
277 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
280 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
281 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
286 /* create and open the MOS (meta-objset) */
287 dp->dp_meta_objset = dmu_objset_create_impl(spa,
288 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
290 /* create the pool directory */
291 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
292 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
295 /* Initialize scan structures */
296 VERIFY3U(0, ==, dsl_scan_init(dp, txg));
298 /* create and open the root dir */
299 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
300 VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
301 NULL, dp, &dp->dp_root_dir));
303 /* create and open the meta-objset dir */
304 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
305 VERIFY(0 == dsl_pool_open_special_dir(dp,
306 MOS_DIR_NAME, &dp->dp_mos_dir));
308 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
309 /* create and open the free dir */
310 (void) dsl_dir_create_sync(dp, dp->dp_root_dir,
312 VERIFY(0 == dsl_pool_open_special_dir(dp,
313 FREE_DIR_NAME, &dp->dp_free_dir));
315 /* create and open the free_bplist */
316 obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx);
317 VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
318 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
319 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
320 dp->dp_meta_objset, obj));
323 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
324 dsl_pool_create_origin(dp, tx);
326 /* create the root dataset */
327 obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
329 /* create the root objset */
330 VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
331 os = dmu_objset_create_impl(dp->dp_spa, ds,
332 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
334 zfs_create_fs(os, kcred, zplprops, tx);
336 dsl_dataset_rele(ds, FTAG);
344 * Account for the meta-objset space in its placeholder dsl_dir.
347 dsl_pool_mos_diduse_space(dsl_pool_t *dp,
348 int64_t used, int64_t comp, int64_t uncomp)
350 ASSERT3U(comp, ==, uncomp); /* it's all metadata */
351 mutex_enter(&dp->dp_lock);
352 dp->dp_mos_used_delta += used;
353 dp->dp_mos_compressed_delta += comp;
354 dp->dp_mos_uncompressed_delta += uncomp;
355 mutex_exit(&dp->dp_lock);
359 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
361 dsl_deadlist_t *dl = arg;
362 dsl_pool_t *dp = dmu_objset_pool(dl->dl_os);
363 rw_enter(&dp->dp_config_rwlock, RW_READER);
364 dsl_deadlist_insert(dl, bp, tx);
365 rw_exit(&dp->dp_config_rwlock);
370 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
376 objset_t *mos = dp->dp_meta_objset;
377 hrtime_t start, write_time;
378 uint64_t data_written;
380 list_t synced_datasets;
382 list_create(&synced_datasets, sizeof (dsl_dataset_t),
383 offsetof(dsl_dataset_t, ds_synced_link));
386 * We need to copy dp_space_towrite() before doing
387 * dsl_sync_task_group_sync(), because
388 * dsl_dataset_snapshot_reserve_space() will increase
389 * dp_space_towrite but not actually write anything.
391 data_written = dp->dp_space_towrite[txg & TXG_MASK];
393 tx = dmu_tx_create_assigned(dp, txg);
395 dp->dp_read_overhead = 0;
398 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
399 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
401 * We must not sync any non-MOS datasets twice, because
402 * we may have taken a snapshot of them. However, we
403 * may sync newly-created datasets on pass 2.
405 ASSERT(!list_link_active(&ds->ds_synced_link));
406 list_insert_tail(&synced_datasets, ds);
407 dsl_dataset_sync(ds, zio, tx);
409 DTRACE_PROBE(pool_sync__1setup);
412 write_time = gethrtime() - start;
414 DTRACE_PROBE(pool_sync__2rootzio);
417 * After the data blocks have been written (ensured by the zio_wait()
418 * above), update the user/group space accounting.
420 for (ds = list_head(&synced_datasets); ds;
421 ds = list_next(&synced_datasets, ds))
422 dmu_objset_do_userquota_updates(ds->ds_objset, tx);
425 * Sync the datasets again to push out the changes due to
426 * userspace updates. This must be done before we process the
427 * sync tasks, so that any snapshots will have the correct
428 * user accounting information (and we won't get confused
429 * about which blocks are part of the snapshot).
431 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
432 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
433 ASSERT(list_link_active(&ds->ds_synced_link));
434 dmu_buf_rele(ds->ds_dbuf, ds);
435 dsl_dataset_sync(ds, zio, tx);
440 * Now that the datasets have been completely synced, we can
441 * clean up our in-memory structures accumulated while syncing:
443 * - move dead blocks from the pending deadlist to the on-disk deadlist
444 * - release hold from dsl_dataset_dirty()
446 while (ds = list_remove_head(&synced_datasets)) {
447 objset_t *os = ds->ds_objset;
448 bplist_iterate(&ds->ds_pending_deadlist,
449 deadlist_enqueue_cb, &ds->ds_deadlist, tx);
450 ASSERT(!dmu_objset_is_dirty(os, txg));
451 dmu_buf_rele(ds->ds_dbuf, ds);
455 while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
456 dsl_dir_sync(dd, tx);
457 write_time += gethrtime() - start;
460 * The MOS's space is accounted for in the pool/$MOS
461 * (dp_mos_dir). We can't modify the mos while we're syncing
462 * it, so we remember the deltas and apply them here.
464 if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 ||
465 dp->dp_mos_uncompressed_delta != 0) {
466 dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD,
467 dp->dp_mos_used_delta,
468 dp->dp_mos_compressed_delta,
469 dp->dp_mos_uncompressed_delta, tx);
470 dp->dp_mos_used_delta = 0;
471 dp->dp_mos_compressed_delta = 0;
472 dp->dp_mos_uncompressed_delta = 0;
476 if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
477 list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) {
478 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
479 dmu_objset_sync(mos, zio, tx);
482 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
483 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
485 write_time += gethrtime() - start;
486 DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
487 hrtime_t, dp->dp_read_overhead);
488 write_time -= dp->dp_read_overhead;
491 * If we modify a dataset in the same txg that we want to destroy it,
492 * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it.
493 * dsl_dir_destroy_check() will fail if there are unexpected holds.
494 * Therefore, we want to sync the MOS (thus syncing the dd_dbuf
495 * and clearing the hold on it) before we process the sync_tasks.
496 * The MOS data dirtied by the sync_tasks will be synced on the next
499 DTRACE_PROBE(pool_sync__3task);
500 if (!txg_list_empty(&dp->dp_sync_tasks, txg)) {
501 dsl_sync_task_group_t *dstg;
503 * No more sync tasks should have been added while we
506 ASSERT(spa_sync_pass(dp->dp_spa) == 1);
507 while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
508 dsl_sync_task_group_sync(dstg, tx);
513 dp->dp_space_towrite[txg & TXG_MASK] = 0;
514 ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
517 * If the write limit max has not been explicitly set, set it
518 * to a fraction of available physical memory (default 1/8th).
519 * Note that we must inflate the limit because the spa
520 * inflates write sizes to account for data replication.
521 * Check this each sync phase to catch changing memory size.
523 if (physmem != old_physmem && zfs_write_limit_shift) {
524 mutex_enter(&zfs_write_limit_lock);
525 old_physmem = physmem;
526 zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
527 zfs_write_limit_inflated = MAX(zfs_write_limit_min,
528 spa_get_asize(dp->dp_spa, zfs_write_limit_max));
529 mutex_exit(&zfs_write_limit_lock);
533 * Attempt to keep the sync time consistent by adjusting the
534 * amount of write traffic allowed into each transaction group.
535 * Weight the throughput calculation towards the current value:
536 * thru = 3/4 old_thru + 1/4 new_thru
538 * Note: write_time is in nanosecs, so write_time/MICROSEC
541 ASSERT(zfs_write_limit_min > 0);
542 if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) {
543 uint64_t throughput = data_written / (write_time / MICROSEC);
545 if (dp->dp_throughput)
546 dp->dp_throughput = throughput / 4 +
547 3 * dp->dp_throughput / 4;
549 dp->dp_throughput = throughput;
550 dp->dp_write_limit = MIN(zfs_write_limit_inflated,
551 MAX(zfs_write_limit_min,
552 dp->dp_throughput * zfs_txg_synctime_ms));
557 dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg)
562 while (zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg)) {
563 ds = dmu_objset_ds(zilog->zl_os);
564 zil_clean(zilog, txg);
565 ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg));
566 dmu_buf_rele(ds->ds_dbuf, zilog);
568 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg));
572 * TRUE if the current thread is the tx_sync_thread or if we
573 * are being called from SPA context during pool initialization.
576 dsl_pool_sync_context(dsl_pool_t *dp)
578 return (curthread == dp->dp_tx.tx_sync_thread ||
579 spa_is_initializing(dp->dp_spa));
583 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
585 uint64_t space, resv;
588 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
590 * XXX The intent log is not accounted for, so it must fit
593 * If we're trying to assess whether it's OK to do a free,
594 * cut the reservation in half to allow forward progress
595 * (e.g. make it possible to rm(1) files from a full pool).
597 space = spa_get_dspace(dp->dp_spa);
598 resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
602 return (space - resv);
606 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
608 uint64_t reserved = 0;
609 uint64_t write_limit = (zfs_write_limit_override ?
610 zfs_write_limit_override : dp->dp_write_limit);
612 if (zfs_no_write_throttle) {
613 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
619 * Check to see if we have exceeded the maximum allowed IO for
620 * this transaction group. We can do this without locks since
621 * a little slop here is ok. Note that we do the reserved check
622 * with only half the requested reserve: this is because the
623 * reserve requests are worst-case, and we really don't want to
624 * throttle based off of worst-case estimates.
626 if (write_limit > 0) {
627 reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
628 + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
630 if (reserved && reserved > write_limit)
634 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
637 * If this transaction group is over 7/8ths capacity, delay
638 * the caller 1 clock tick. This will slow down the "fill"
639 * rate until the sync process can catch up with us.
641 if (reserved && reserved > (write_limit - (write_limit >> 3)))
642 txg_delay(dp, tx->tx_txg, 1);
648 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
650 ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
651 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
655 dsl_pool_memory_pressure(dsl_pool_t *dp)
657 uint64_t space_inuse = 0;
660 if (dp->dp_write_limit == zfs_write_limit_min)
663 for (i = 0; i < TXG_SIZE; i++) {
664 space_inuse += dp->dp_space_towrite[i];
665 space_inuse += dp->dp_tempreserved[i];
667 dp->dp_write_limit = MAX(zfs_write_limit_min,
668 MIN(dp->dp_write_limit, space_inuse / 4));
672 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
675 mutex_enter(&dp->dp_lock);
676 dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
677 mutex_exit(&dp->dp_lock);
683 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
686 dsl_dataset_t *ds, *prev = NULL;
688 dsl_pool_t *dp = spa_get_dsl(spa);
690 err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
694 while (ds->ds_phys->ds_prev_snap_obj != 0) {
695 err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
698 dsl_dataset_rele(ds, FTAG);
702 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
704 dsl_dataset_rele(ds, FTAG);
710 prev = dp->dp_origin_snap;
713 * The $ORIGIN can't have any data, or the accounting
716 ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
718 /* The origin doesn't get attached to itself */
719 if (ds->ds_object == prev->ds_object) {
720 dsl_dataset_rele(ds, FTAG);
724 dmu_buf_will_dirty(ds->ds_dbuf, tx);
725 ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
726 ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
728 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
729 ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
731 dmu_buf_will_dirty(prev->ds_dbuf, tx);
732 prev->ds_phys->ds_num_children++;
734 if (ds->ds_phys->ds_next_snap_obj == 0) {
735 ASSERT(ds->ds_prev == NULL);
736 VERIFY(0 == dsl_dataset_hold_obj(dp,
737 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
741 ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
742 ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
744 if (prev->ds_phys->ds_next_clones_obj == 0) {
745 dmu_buf_will_dirty(prev->ds_dbuf, tx);
746 prev->ds_phys->ds_next_clones_obj =
747 zap_create(dp->dp_meta_objset,
748 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
750 VERIFY(0 == zap_add_int(dp->dp_meta_objset,
751 prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
753 dsl_dataset_rele(ds, FTAG);
754 if (prev != dp->dp_origin_snap)
755 dsl_dataset_rele(prev, FTAG);
760 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
762 ASSERT(dmu_tx_is_syncing(tx));
763 ASSERT(dp->dp_origin_snap != NULL);
765 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
766 tx, DS_FIND_CHILDREN));
771 upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
775 dsl_pool_t *dp = spa_get_dsl(spa);
776 objset_t *mos = dp->dp_meta_objset;
778 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
780 if (ds->ds_dir->dd_phys->dd_origin_obj) {
781 dsl_dataset_t *origin;
783 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
784 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin));
786 if (origin->ds_dir->dd_phys->dd_clones == 0) {
787 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
788 origin->ds_dir->dd_phys->dd_clones = zap_create(mos,
789 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
792 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
793 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
795 dsl_dataset_rele(origin, FTAG);
798 dsl_dataset_rele(ds, FTAG);
803 dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
805 ASSERT(dmu_tx_is_syncing(tx));
808 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx);
809 VERIFY(0 == dsl_pool_open_special_dir(dp,
810 FREE_DIR_NAME, &dp->dp_free_dir));
813 * We can't use bpobj_alloc(), because spa_version() still
814 * returns the old version, and we need a new-version bpobj with
815 * subobj support. So call dmu_object_alloc() directly.
817 obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ,
818 SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx);
819 VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
820 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
821 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
822 dp->dp_meta_objset, obj));
824 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL,
825 upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN));
829 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
834 ASSERT(dmu_tx_is_syncing(tx));
835 ASSERT(dp->dp_origin_snap == NULL);
837 /* create the origin dir, ds, & snap-ds */
838 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
839 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
841 VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
842 dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx);
843 VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
844 dp, &dp->dp_origin_snap));
845 dsl_dataset_rele(ds, FTAG);
846 rw_exit(&dp->dp_config_rwlock);
850 dsl_pool_vnrele_taskq(dsl_pool_t *dp)
852 return (dp->dp_vnrele_taskq);
856 * Walk through the pool-wide zap object of temporary snapshot user holds
860 dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp)
864 objset_t *mos = dp->dp_meta_objset;
865 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
869 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
871 for (zap_cursor_init(&zc, mos, zapobj);
872 zap_cursor_retrieve(&zc, &za) == 0;
873 zap_cursor_advance(&zc)) {
877 htag = strchr(za.za_name, '-');
880 dsobj = strtonum(za.za_name, NULL);
881 (void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE);
883 zap_cursor_fini(&zc);
887 * Create the pool-wide zap object for storing temporary snapshot holds.
890 dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx)
892 objset_t *mos = dp->dp_meta_objset;
894 ASSERT(dp->dp_tmp_userrefs_obj == 0);
895 ASSERT(dmu_tx_is_syncing(tx));
897 dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS,
898 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx);
902 dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
903 const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding)
905 objset_t *mos = dp->dp_meta_objset;
906 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
910 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
911 ASSERT(dmu_tx_is_syncing(tx));
914 * If the pool was created prior to SPA_VERSION_USERREFS, the
915 * zap object for temporary holds might not exist yet.
919 dsl_pool_user_hold_create_obj(dp, tx);
920 zapobj = dp->dp_tmp_userrefs_obj;
926 name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag);
928 error = zap_add(mos, zapobj, name, 8, 1, now, tx);
930 error = zap_remove(mos, zapobj, name, tx);
937 * Add a temporary hold for the given dataset object and tag.
940 dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
941 uint64_t *now, dmu_tx_t *tx)
943 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE));
947 * Release a temporary hold for the given dataset object and tag.
950 dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
953 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL,