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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #include <sys/dsl_pool.h>
27 #include <sys/dsl_dataset.h>
28 #include <sys/dsl_dir.h>
29 #include <sys/dsl_synctask.h>
30 #include <sys/dmu_tx.h>
31 #include <sys/dmu_objset.h>
35 #include <sys/zfs_context.h>
36 #include <sys/fs/zfs.h>
37 #include <sys/zfs_znode.h>
38 #include <sys/spa_impl.h>
40 int zfs_no_write_throttle = 0;
41 int zfs_write_limit_shift = 3; /* 1/8th of physical memory */
42 int zfs_txg_synctime = 5; /* target secs to sync a txg */
44 uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */
45 uint64_t zfs_write_limit_max = 0; /* max data payload per txg */
46 uint64_t zfs_write_limit_inflated = 0;
47 uint64_t zfs_write_limit_override = 0;
48 extern uint64_t zfs_write_limit_min;
50 kmutex_t zfs_write_limit_lock;
52 static pgcnt_t old_physmem = 0;
55 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
60 err = zap_lookup(dp->dp_meta_objset,
61 dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
62 name, sizeof (obj), 1, &obj);
66 return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
70 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
73 blkptr_t *bp = spa_get_rootblkptr(spa);
75 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
77 dp->dp_meta_rootbp = *bp;
78 rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
79 dp->dp_write_limit = zfs_write_limit_min;
82 txg_list_create(&dp->dp_dirty_datasets,
83 offsetof(dsl_dataset_t, ds_dirty_link));
84 txg_list_create(&dp->dp_dirty_dirs,
85 offsetof(dsl_dir_t, dd_dirty_link));
86 txg_list_create(&dp->dp_sync_tasks,
87 offsetof(dsl_sync_task_group_t, dstg_node));
88 list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t),
89 offsetof(dsl_dataset_t, ds_synced_link));
91 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
92 mutex_init(&dp->dp_scrub_cancel_lock, NULL, MUTEX_DEFAULT, NULL);
94 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
101 dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
104 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
109 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
110 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &osi);
113 dp->dp_meta_objset = &osi->os;
115 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
116 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
117 &dp->dp_root_dir_obj);
121 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
122 NULL, dp, &dp->dp_root_dir);
126 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
130 if (spa_version(spa) >= SPA_VERSION_ORIGIN) {
131 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
134 err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
137 err = dsl_dataset_hold_obj(dp,
138 ds->ds_phys->ds_prev_snap_obj, dp,
139 &dp->dp_origin_snap);
140 dsl_dataset_rele(ds, FTAG);
142 dsl_dir_close(dd, dp);
147 /* get scrub status */
148 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
149 DMU_POOL_SCRUB_FUNC, sizeof (uint32_t), 1,
152 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
153 DMU_POOL_SCRUB_QUEUE, sizeof (uint64_t), 1,
154 &dp->dp_scrub_queue_obj);
157 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
158 DMU_POOL_SCRUB_MIN_TXG, sizeof (uint64_t), 1,
159 &dp->dp_scrub_min_txg);
162 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
163 DMU_POOL_SCRUB_MAX_TXG, sizeof (uint64_t), 1,
164 &dp->dp_scrub_max_txg);
167 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
168 DMU_POOL_SCRUB_BOOKMARK, sizeof (uint64_t), 4,
169 &dp->dp_scrub_bookmark);
172 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
173 DMU_POOL_SCRUB_ERRORS, sizeof (uint64_t), 1,
174 &spa->spa_scrub_errors);
177 if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) {
179 * A new-type scrub was in progress on an old
180 * pool. Restart from the beginning, since the
181 * old software may have changed the pool in the
184 dsl_pool_scrub_restart(dp);
188 * It's OK if there is no scrub in progress (and if
189 * there was an I/O error, ignore it).
195 rw_exit(&dp->dp_config_rwlock);
205 dsl_pool_close(dsl_pool_t *dp)
207 /* drop our references from dsl_pool_open() */
210 * Since we held the origin_snap from "syncing" context (which
211 * includes pool-opening context), it actually only got a "ref"
212 * and not a hold, so just drop that here.
214 if (dp->dp_origin_snap)
215 dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
217 dsl_dir_close(dp->dp_mos_dir, dp);
219 dsl_dir_close(dp->dp_root_dir, dp);
221 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
222 if (dp->dp_meta_objset)
223 dmu_objset_evict(NULL, dp->dp_meta_objset->os);
225 txg_list_destroy(&dp->dp_dirty_datasets);
226 txg_list_destroy(&dp->dp_dirty_dirs);
227 txg_list_destroy(&dp->dp_sync_tasks);
228 list_destroy(&dp->dp_synced_datasets);
230 arc_flush(dp->dp_spa);
232 rw_destroy(&dp->dp_config_rwlock);
233 mutex_destroy(&dp->dp_lock);
234 mutex_destroy(&dp->dp_scrub_cancel_lock);
235 taskq_destroy(dp->dp_vnrele_taskq);
237 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
238 kmem_free(dp, sizeof (dsl_pool_t));
242 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
245 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
246 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
251 /* create and open the MOS (meta-objset) */
252 dp->dp_meta_objset = &dmu_objset_create_impl(spa,
253 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx)->os;
255 /* create the pool directory */
256 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
257 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
258 ASSERT3U(err, ==, 0);
260 /* create and open the root dir */
261 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
262 VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
263 NULL, dp, &dp->dp_root_dir));
265 /* create and open the meta-objset dir */
266 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
267 VERIFY(0 == dsl_pool_open_special_dir(dp,
268 MOS_DIR_NAME, &dp->dp_mos_dir));
270 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
271 dsl_pool_create_origin(dp, tx);
273 /* create the root dataset */
274 dsobj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
276 /* create the root objset */
277 VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
278 osip = dmu_objset_create_impl(dp->dp_spa, ds,
279 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
281 zfs_create_fs(&osip->os, kcred, zplprops, tx);
283 dsl_dataset_rele(ds, FTAG);
291 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
297 dsl_sync_task_group_t *dstg;
298 objset_impl_t *mosi = dp->dp_meta_objset->os;
299 hrtime_t start, write_time;
300 uint64_t data_written;
303 tx = dmu_tx_create_assigned(dp, txg);
305 dp->dp_read_overhead = 0;
308 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
309 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
311 * We must not sync any non-MOS datasets twice, because
312 * we may have taken a snapshot of them. However, we
313 * may sync newly-created datasets on pass 2.
315 ASSERT(!list_link_active(&ds->ds_synced_link));
316 list_insert_tail(&dp->dp_synced_datasets, ds);
317 dsl_dataset_sync(ds, zio, tx);
319 DTRACE_PROBE(pool_sync__1setup);
322 write_time = gethrtime() - start;
324 DTRACE_PROBE(pool_sync__2rootzio);
326 for (ds = list_head(&dp->dp_synced_datasets); ds;
327 ds = list_next(&dp->dp_synced_datasets, ds))
328 dmu_objset_do_userquota_callbacks(ds->ds_user_ptr, tx);
331 * Sync the datasets again to push out the changes due to
332 * userquota updates. This must be done before we process the
333 * sync tasks, because that could cause a snapshot of a dataset
334 * whose ds_bp will be rewritten when we do this 2nd sync.
336 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
337 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
338 ASSERT(list_link_active(&ds->ds_synced_link));
339 dmu_buf_rele(ds->ds_dbuf, ds);
340 dsl_dataset_sync(ds, zio, tx);
344 while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) {
346 * No more sync tasks should have been added while we
349 ASSERT(spa_sync_pass(dp->dp_spa) == 1);
350 dsl_sync_task_group_sync(dstg, tx);
352 DTRACE_PROBE(pool_sync__3task);
355 while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
356 dsl_dir_sync(dd, tx);
357 write_time += gethrtime() - start;
359 if (spa_sync_pass(dp->dp_spa) == 1)
360 dsl_pool_scrub_sync(dp, tx);
363 if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
364 list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
365 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
366 dmu_objset_sync(mosi, zio, tx);
369 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
370 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
372 write_time += gethrtime() - start;
373 DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
374 hrtime_t, dp->dp_read_overhead);
375 write_time -= dp->dp_read_overhead;
379 data_written = dp->dp_space_towrite[txg & TXG_MASK];
380 dp->dp_space_towrite[txg & TXG_MASK] = 0;
381 ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
384 * If the write limit max has not been explicitly set, set it
385 * to a fraction of available physical memory (default 1/8th).
386 * Note that we must inflate the limit because the spa
387 * inflates write sizes to account for data replication.
388 * Check this each sync phase to catch changing memory size.
390 if (physmem != old_physmem && zfs_write_limit_shift) {
391 mutex_enter(&zfs_write_limit_lock);
392 old_physmem = physmem;
393 zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
394 zfs_write_limit_inflated = MAX(zfs_write_limit_min,
395 spa_get_asize(dp->dp_spa, zfs_write_limit_max));
396 mutex_exit(&zfs_write_limit_lock);
400 * Attempt to keep the sync time consistent by adjusting the
401 * amount of write traffic allowed into each transaction group.
402 * Weight the throughput calculation towards the current value:
403 * thru = 3/4 old_thru + 1/4 new_thru
405 ASSERT(zfs_write_limit_min > 0);
406 if (data_written > zfs_write_limit_min / 8 && write_time > 0) {
407 uint64_t throughput = (data_written * NANOSEC) / write_time;
408 if (dp->dp_throughput)
409 dp->dp_throughput = throughput / 4 +
410 3 * dp->dp_throughput / 4;
412 dp->dp_throughput = throughput;
413 dp->dp_write_limit = MIN(zfs_write_limit_inflated,
414 MAX(zfs_write_limit_min,
415 dp->dp_throughput * zfs_txg_synctime));
420 dsl_pool_zil_clean(dsl_pool_t *dp)
424 while (ds = list_head(&dp->dp_synced_datasets)) {
425 list_remove(&dp->dp_synced_datasets, ds);
426 ASSERT(ds->ds_user_ptr != NULL);
427 zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
428 dmu_buf_rele(ds->ds_dbuf, ds);
433 * TRUE if the current thread is the tx_sync_thread or if we
434 * are being called from SPA context during pool initialization.
437 dsl_pool_sync_context(dsl_pool_t *dp)
439 return (curthread == dp->dp_tx.tx_sync_thread ||
440 spa_get_dsl(dp->dp_spa) == NULL);
444 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
446 uint64_t space, resv;
449 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
451 * XXX The intent log is not accounted for, so it must fit
454 * If we're trying to assess whether it's OK to do a free,
455 * cut the reservation in half to allow forward progress
456 * (e.g. make it possible to rm(1) files from a full pool).
458 space = spa_get_dspace(dp->dp_spa);
459 resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
463 return (space - resv);
467 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
469 uint64_t reserved = 0;
470 uint64_t write_limit = (zfs_write_limit_override ?
471 zfs_write_limit_override : dp->dp_write_limit);
473 if (zfs_no_write_throttle) {
474 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
480 * Check to see if we have exceeded the maximum allowed IO for
481 * this transaction group. We can do this without locks since
482 * a little slop here is ok. Note that we do the reserved check
483 * with only half the requested reserve: this is because the
484 * reserve requests are worst-case, and we really don't want to
485 * throttle based off of worst-case estimates.
487 if (write_limit > 0) {
488 reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
489 + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
491 if (reserved && reserved > write_limit)
495 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
498 * If this transaction group is over 7/8ths capacity, delay
499 * the caller 1 clock tick. This will slow down the "fill"
500 * rate until the sync process can catch up with us.
502 if (reserved && reserved > (write_limit - (write_limit >> 3)))
503 txg_delay(dp, tx->tx_txg, 1);
509 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
511 ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
512 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
516 dsl_pool_memory_pressure(dsl_pool_t *dp)
518 uint64_t space_inuse = 0;
521 if (dp->dp_write_limit == zfs_write_limit_min)
524 for (i = 0; i < TXG_SIZE; i++) {
525 space_inuse += dp->dp_space_towrite[i];
526 space_inuse += dp->dp_tempreserved[i];
528 dp->dp_write_limit = MAX(zfs_write_limit_min,
529 MIN(dp->dp_write_limit, space_inuse / 4));
533 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
536 mutex_enter(&dp->dp_lock);
537 dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
538 mutex_exit(&dp->dp_lock);
544 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
547 dsl_dataset_t *ds, *prev = NULL;
549 dsl_pool_t *dp = spa_get_dsl(spa);
551 err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
555 while (ds->ds_phys->ds_prev_snap_obj != 0) {
556 err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
559 dsl_dataset_rele(ds, FTAG);
563 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
565 dsl_dataset_rele(ds, FTAG);
571 prev = dp->dp_origin_snap;
574 * The $ORIGIN can't have any data, or the accounting
577 ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
579 /* The origin doesn't get attached to itself */
580 if (ds->ds_object == prev->ds_object) {
581 dsl_dataset_rele(ds, FTAG);
585 dmu_buf_will_dirty(ds->ds_dbuf, tx);
586 ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
587 ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
589 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
590 ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
592 dmu_buf_will_dirty(prev->ds_dbuf, tx);
593 prev->ds_phys->ds_num_children++;
595 if (ds->ds_phys->ds_next_snap_obj == 0) {
596 ASSERT(ds->ds_prev == NULL);
597 VERIFY(0 == dsl_dataset_hold_obj(dp,
598 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
602 ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
603 ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
605 if (prev->ds_phys->ds_next_clones_obj == 0) {
606 dmu_buf_will_dirty(prev->ds_dbuf, tx);
607 prev->ds_phys->ds_next_clones_obj =
608 zap_create(dp->dp_meta_objset,
609 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
611 VERIFY(0 == zap_add_int(dp->dp_meta_objset,
612 prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
614 dsl_dataset_rele(ds, FTAG);
615 if (prev != dp->dp_origin_snap)
616 dsl_dataset_rele(prev, FTAG);
621 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
623 ASSERT(dmu_tx_is_syncing(tx));
624 ASSERT(dp->dp_origin_snap != NULL);
626 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
627 tx, DS_FIND_CHILDREN));
631 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
636 ASSERT(dmu_tx_is_syncing(tx));
637 ASSERT(dp->dp_origin_snap == NULL);
639 /* create the origin dir, ds, & snap-ds */
640 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
641 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
643 VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
644 dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, kcred, tx);
645 VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
646 dp, &dp->dp_origin_snap));
647 dsl_dataset_rele(ds, FTAG);
648 rw_exit(&dp->dp_config_rwlock);
652 dsl_pool_vnrele_taskq(dsl_pool_t *dp)
654 return (dp->dp_vnrele_taskq);