xref: /illumos-gate/usr/src/uts/common/fs/zfs/dsl_pool.c (revision e5351341b58845eee9d722bd71543d5a7c26b6cc)
1 /*
2  * CDDL HEADER START
3  *
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.
7  *
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.
12  *
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]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011 by Delphix. All rights reserved.
24  */
25 
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>
35 #include <sys/arc.h>
36 #include <sys/zap.h>
37 #include <sys/zio.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 
44 int zfs_no_write_throttle = 0;
45 int zfs_write_limit_shift = 3;			/* 1/8th of physical memory */
46 int zfs_txg_synctime_ms = 1000;		/* target millisecs to sync a txg */
47 
48 uint64_t zfs_write_limit_min = 32 << 20;	/* min write limit is 32MB */
49 uint64_t zfs_write_limit_max = 0;		/* max data payload per txg */
50 uint64_t zfs_write_limit_inflated = 0;
51 uint64_t zfs_write_limit_override = 0;
52 
53 kmutex_t zfs_write_limit_lock;
54 
55 static pgcnt_t old_physmem = 0;
56 
57 int
58 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
59 {
60 	uint64_t obj;
61 	int err;
62 
63 	err = zap_lookup(dp->dp_meta_objset,
64 	    dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
65 	    name, sizeof (obj), 1, &obj);
66 	if (err)
67 		return (err);
68 
69 	return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
70 }
71 
72 static dsl_pool_t *
73 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
74 {
75 	dsl_pool_t *dp;
76 	blkptr_t *bp = spa_get_rootblkptr(spa);
77 
78 	dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
79 	dp->dp_spa = spa;
80 	dp->dp_meta_rootbp = *bp;
81 	rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
82 	dp->dp_write_limit = zfs_write_limit_min;
83 	txg_init(dp, txg);
84 
85 	txg_list_create(&dp->dp_dirty_datasets,
86 	    offsetof(dsl_dataset_t, ds_dirty_link));
87 	txg_list_create(&dp->dp_dirty_dirs,
88 	    offsetof(dsl_dir_t, dd_dirty_link));
89 	txg_list_create(&dp->dp_sync_tasks,
90 	    offsetof(dsl_sync_task_group_t, dstg_node));
91 	list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t),
92 	    offsetof(dsl_dataset_t, ds_synced_link));
93 
94 	mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
95 
96 	dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
97 	    1, 4, 0);
98 
99 	return (dp);
100 }
101 
102 int
103 dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
104 {
105 	int err;
106 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
107 	dsl_dir_t *dd;
108 	dsl_dataset_t *ds;
109 	uint64_t obj;
110 
111 	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
112 	err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp,
113 	    &dp->dp_meta_objset);
114 	if (err)
115 		goto out;
116 
117 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
118 	    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
119 	    &dp->dp_root_dir_obj);
120 	if (err)
121 		goto out;
122 
123 	err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
124 	    NULL, dp, &dp->dp_root_dir);
125 	if (err)
126 		goto out;
127 
128 	err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
129 	if (err)
130 		goto out;
131 
132 	if (spa_version(spa) >= SPA_VERSION_ORIGIN) {
133 		err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
134 		if (err)
135 			goto out;
136 		err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
137 		    FTAG, &ds);
138 		if (err == 0) {
139 			err = dsl_dataset_hold_obj(dp,
140 			    ds->ds_phys->ds_prev_snap_obj, dp,
141 			    &dp->dp_origin_snap);
142 			dsl_dataset_rele(ds, FTAG);
143 		}
144 		dsl_dir_close(dd, dp);
145 		if (err)
146 			goto out;
147 	}
148 
149 	if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
150 		err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME,
151 		    &dp->dp_free_dir);
152 		if (err)
153 			goto out;
154 
155 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
156 		    DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj);
157 		if (err)
158 			goto out;
159 		VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
160 		    dp->dp_meta_objset, obj));
161 	}
162 
163 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
164 	    DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1,
165 	    &dp->dp_tmp_userrefs_obj);
166 	if (err == ENOENT)
167 		err = 0;
168 	if (err)
169 		goto out;
170 
171 	err = dsl_scan_init(dp, txg);
172 
173 out:
174 	rw_exit(&dp->dp_config_rwlock);
175 	if (err)
176 		dsl_pool_close(dp);
177 	else
178 		*dpp = dp;
179 
180 	return (err);
181 }
182 
183 void
184 dsl_pool_close(dsl_pool_t *dp)
185 {
186 	/* drop our references from dsl_pool_open() */
187 
188 	/*
189 	 * Since we held the origin_snap from "syncing" context (which
190 	 * includes pool-opening context), it actually only got a "ref"
191 	 * and not a hold, so just drop that here.
192 	 */
193 	if (dp->dp_origin_snap)
194 		dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
195 	if (dp->dp_mos_dir)
196 		dsl_dir_close(dp->dp_mos_dir, dp);
197 	if (dp->dp_free_dir)
198 		dsl_dir_close(dp->dp_free_dir, dp);
199 	if (dp->dp_root_dir)
200 		dsl_dir_close(dp->dp_root_dir, dp);
201 
202 	bpobj_close(&dp->dp_free_bpobj);
203 
204 	/* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
205 	if (dp->dp_meta_objset)
206 		dmu_objset_evict(dp->dp_meta_objset);
207 
208 	txg_list_destroy(&dp->dp_dirty_datasets);
209 	txg_list_destroy(&dp->dp_sync_tasks);
210 	txg_list_destroy(&dp->dp_dirty_dirs);
211 	list_destroy(&dp->dp_synced_datasets);
212 
213 	arc_flush(dp->dp_spa);
214 	txg_fini(dp);
215 	dsl_scan_fini(dp);
216 	rw_destroy(&dp->dp_config_rwlock);
217 	mutex_destroy(&dp->dp_lock);
218 	taskq_destroy(dp->dp_vnrele_taskq);
219 	if (dp->dp_blkstats)
220 		kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
221 	kmem_free(dp, sizeof (dsl_pool_t));
222 }
223 
224 dsl_pool_t *
225 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
226 {
227 	int err;
228 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
229 	dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
230 	objset_t *os;
231 	dsl_dataset_t *ds;
232 	uint64_t obj;
233 
234 	/* create and open the MOS (meta-objset) */
235 	dp->dp_meta_objset = dmu_objset_create_impl(spa,
236 	    NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
237 
238 	/* create the pool directory */
239 	err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
240 	    DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
241 	ASSERT3U(err, ==, 0);
242 
243 	/* Initialize scan structures */
244 	VERIFY3U(0, ==, dsl_scan_init(dp, txg));
245 
246 	/* create and open the root dir */
247 	dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
248 	VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
249 	    NULL, dp, &dp->dp_root_dir));
250 
251 	/* create and open the meta-objset dir */
252 	(void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
253 	VERIFY(0 == dsl_pool_open_special_dir(dp,
254 	    MOS_DIR_NAME, &dp->dp_mos_dir));
255 
256 	if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
257 		/* create and open the free dir */
258 		(void) dsl_dir_create_sync(dp, dp->dp_root_dir,
259 		    FREE_DIR_NAME, tx);
260 		VERIFY(0 == dsl_pool_open_special_dir(dp,
261 		    FREE_DIR_NAME, &dp->dp_free_dir));
262 
263 		/* create and open the free_bplist */
264 		obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx);
265 		VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
266 		    DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
267 		VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
268 		    dp->dp_meta_objset, obj));
269 	}
270 
271 	if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
272 		dsl_pool_create_origin(dp, tx);
273 
274 	/* create the root dataset */
275 	obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
276 
277 	/* create the root objset */
278 	VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
279 	os = dmu_objset_create_impl(dp->dp_spa, ds,
280 	    dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
281 #ifdef _KERNEL
282 	zfs_create_fs(os, kcred, zplprops, tx);
283 #endif
284 	dsl_dataset_rele(ds, FTAG);
285 
286 	dmu_tx_commit(tx);
287 
288 	return (dp);
289 }
290 
291 static int
292 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
293 {
294 	dsl_deadlist_t *dl = arg;
295 	dsl_pool_t *dp = dmu_objset_pool(dl->dl_os);
296 	rw_enter(&dp->dp_config_rwlock, RW_READER);
297 	dsl_deadlist_insert(dl, bp, tx);
298 	rw_exit(&dp->dp_config_rwlock);
299 	return (0);
300 }
301 
302 void
303 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
304 {
305 	zio_t *zio;
306 	dmu_tx_t *tx;
307 	dsl_dir_t *dd;
308 	dsl_dataset_t *ds;
309 	dsl_sync_task_group_t *dstg;
310 	objset_t *mos = dp->dp_meta_objset;
311 	hrtime_t start, write_time;
312 	uint64_t data_written;
313 	int err;
314 
315 	/*
316 	 * We need to copy dp_space_towrite() before doing
317 	 * dsl_sync_task_group_sync(), because
318 	 * dsl_dataset_snapshot_reserve_space() will increase
319 	 * dp_space_towrite but not actually write anything.
320 	 */
321 	data_written = dp->dp_space_towrite[txg & TXG_MASK];
322 
323 	tx = dmu_tx_create_assigned(dp, txg);
324 
325 	dp->dp_read_overhead = 0;
326 	start = gethrtime();
327 
328 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
329 	while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
330 		/*
331 		 * We must not sync any non-MOS datasets twice, because
332 		 * we may have taken a snapshot of them.  However, we
333 		 * may sync newly-created datasets on pass 2.
334 		 */
335 		ASSERT(!list_link_active(&ds->ds_synced_link));
336 		list_insert_tail(&dp->dp_synced_datasets, ds);
337 		dsl_dataset_sync(ds, zio, tx);
338 	}
339 	DTRACE_PROBE(pool_sync__1setup);
340 	err = zio_wait(zio);
341 
342 	write_time = gethrtime() - start;
343 	ASSERT(err == 0);
344 	DTRACE_PROBE(pool_sync__2rootzio);
345 
346 	for (ds = list_head(&dp->dp_synced_datasets); ds;
347 	    ds = list_next(&dp->dp_synced_datasets, ds))
348 		dmu_objset_do_userquota_updates(ds->ds_objset, tx);
349 
350 	/*
351 	 * Sync the datasets again to push out the changes due to
352 	 * userspace updates.  This must be done before we process the
353 	 * sync tasks, because that could cause a snapshot of a dataset
354 	 * whose ds_bp will be rewritten when we do this 2nd sync.
355 	 */
356 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
357 	while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
358 		ASSERT(list_link_active(&ds->ds_synced_link));
359 		dmu_buf_rele(ds->ds_dbuf, ds);
360 		dsl_dataset_sync(ds, zio, tx);
361 	}
362 	err = zio_wait(zio);
363 
364 	/*
365 	 * Move dead blocks from the pending deadlist to the on-disk
366 	 * deadlist.
367 	 */
368 	for (ds = list_head(&dp->dp_synced_datasets); ds;
369 	    ds = list_next(&dp->dp_synced_datasets, ds)) {
370 		bplist_iterate(&ds->ds_pending_deadlist,
371 		    deadlist_enqueue_cb, &ds->ds_deadlist, tx);
372 	}
373 
374 	while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) {
375 		/*
376 		 * No more sync tasks should have been added while we
377 		 * were syncing.
378 		 */
379 		ASSERT(spa_sync_pass(dp->dp_spa) == 1);
380 		dsl_sync_task_group_sync(dstg, tx);
381 	}
382 	DTRACE_PROBE(pool_sync__3task);
383 
384 	start = gethrtime();
385 	while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
386 		dsl_dir_sync(dd, tx);
387 	write_time += gethrtime() - start;
388 
389 	start = gethrtime();
390 	if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
391 	    list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) {
392 		zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
393 		dmu_objset_sync(mos, zio, tx);
394 		err = zio_wait(zio);
395 		ASSERT(err == 0);
396 		dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
397 		spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
398 	}
399 	write_time += gethrtime() - start;
400 	DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
401 	    hrtime_t, dp->dp_read_overhead);
402 	write_time -= dp->dp_read_overhead;
403 
404 	dmu_tx_commit(tx);
405 
406 	dp->dp_space_towrite[txg & TXG_MASK] = 0;
407 	ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
408 
409 	/*
410 	 * If the write limit max has not been explicitly set, set it
411 	 * to a fraction of available physical memory (default 1/8th).
412 	 * Note that we must inflate the limit because the spa
413 	 * inflates write sizes to account for data replication.
414 	 * Check this each sync phase to catch changing memory size.
415 	 */
416 	if (physmem != old_physmem && zfs_write_limit_shift) {
417 		mutex_enter(&zfs_write_limit_lock);
418 		old_physmem = physmem;
419 		zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
420 		zfs_write_limit_inflated = MAX(zfs_write_limit_min,
421 		    spa_get_asize(dp->dp_spa, zfs_write_limit_max));
422 		mutex_exit(&zfs_write_limit_lock);
423 	}
424 
425 	/*
426 	 * Attempt to keep the sync time consistent by adjusting the
427 	 * amount of write traffic allowed into each transaction group.
428 	 * Weight the throughput calculation towards the current value:
429 	 * 	thru = 3/4 old_thru + 1/4 new_thru
430 	 *
431 	 * Note: write_time is in nanosecs, so write_time/MICROSEC
432 	 * yields millisecs
433 	 */
434 	ASSERT(zfs_write_limit_min > 0);
435 	if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) {
436 		uint64_t throughput = data_written / (write_time / MICROSEC);
437 
438 		if (dp->dp_throughput)
439 			dp->dp_throughput = throughput / 4 +
440 			    3 * dp->dp_throughput / 4;
441 		else
442 			dp->dp_throughput = throughput;
443 		dp->dp_write_limit = MIN(zfs_write_limit_inflated,
444 		    MAX(zfs_write_limit_min,
445 		    dp->dp_throughput * zfs_txg_synctime_ms));
446 	}
447 }
448 
449 void
450 dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg)
451 {
452 	dsl_dataset_t *ds;
453 	objset_t *os;
454 
455 	while (ds = list_head(&dp->dp_synced_datasets)) {
456 		list_remove(&dp->dp_synced_datasets, ds);
457 		os = ds->ds_objset;
458 		zil_clean(os->os_zil, txg);
459 		ASSERT(!dmu_objset_is_dirty(os, txg));
460 		dmu_buf_rele(ds->ds_dbuf, ds);
461 	}
462 	ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg));
463 }
464 
465 /*
466  * TRUE if the current thread is the tx_sync_thread or if we
467  * are being called from SPA context during pool initialization.
468  */
469 int
470 dsl_pool_sync_context(dsl_pool_t *dp)
471 {
472 	return (curthread == dp->dp_tx.tx_sync_thread ||
473 	    spa_get_dsl(dp->dp_spa) == NULL);
474 }
475 
476 uint64_t
477 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
478 {
479 	uint64_t space, resv;
480 
481 	/*
482 	 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
483 	 * efficiency.
484 	 * XXX The intent log is not accounted for, so it must fit
485 	 * within this slop.
486 	 *
487 	 * If we're trying to assess whether it's OK to do a free,
488 	 * cut the reservation in half to allow forward progress
489 	 * (e.g. make it possible to rm(1) files from a full pool).
490 	 */
491 	space = spa_get_dspace(dp->dp_spa);
492 	resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
493 	if (netfree)
494 		resv >>= 1;
495 
496 	return (space - resv);
497 }
498 
499 int
500 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
501 {
502 	uint64_t reserved = 0;
503 	uint64_t write_limit = (zfs_write_limit_override ?
504 	    zfs_write_limit_override : dp->dp_write_limit);
505 
506 	if (zfs_no_write_throttle) {
507 		atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
508 		    space);
509 		return (0);
510 	}
511 
512 	/*
513 	 * Check to see if we have exceeded the maximum allowed IO for
514 	 * this transaction group.  We can do this without locks since
515 	 * a little slop here is ok.  Note that we do the reserved check
516 	 * with only half the requested reserve: this is because the
517 	 * reserve requests are worst-case, and we really don't want to
518 	 * throttle based off of worst-case estimates.
519 	 */
520 	if (write_limit > 0) {
521 		reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
522 		    + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
523 
524 		if (reserved && reserved > write_limit)
525 			return (ERESTART);
526 	}
527 
528 	atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
529 
530 	/*
531 	 * If this transaction group is over 7/8ths capacity, delay
532 	 * the caller 1 clock tick.  This will slow down the "fill"
533 	 * rate until the sync process can catch up with us.
534 	 */
535 	if (reserved && reserved > (write_limit - (write_limit >> 3)))
536 		txg_delay(dp, tx->tx_txg, 1);
537 
538 	return (0);
539 }
540 
541 void
542 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
543 {
544 	ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
545 	atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
546 }
547 
548 void
549 dsl_pool_memory_pressure(dsl_pool_t *dp)
550 {
551 	uint64_t space_inuse = 0;
552 	int i;
553 
554 	if (dp->dp_write_limit == zfs_write_limit_min)
555 		return;
556 
557 	for (i = 0; i < TXG_SIZE; i++) {
558 		space_inuse += dp->dp_space_towrite[i];
559 		space_inuse += dp->dp_tempreserved[i];
560 	}
561 	dp->dp_write_limit = MAX(zfs_write_limit_min,
562 	    MIN(dp->dp_write_limit, space_inuse / 4));
563 }
564 
565 void
566 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
567 {
568 	if (space > 0) {
569 		mutex_enter(&dp->dp_lock);
570 		dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
571 		mutex_exit(&dp->dp_lock);
572 	}
573 }
574 
575 /* ARGSUSED */
576 static int
577 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
578 {
579 	dmu_tx_t *tx = arg;
580 	dsl_dataset_t *ds, *prev = NULL;
581 	int err;
582 	dsl_pool_t *dp = spa_get_dsl(spa);
583 
584 	err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
585 	if (err)
586 		return (err);
587 
588 	while (ds->ds_phys->ds_prev_snap_obj != 0) {
589 		err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
590 		    FTAG, &prev);
591 		if (err) {
592 			dsl_dataset_rele(ds, FTAG);
593 			return (err);
594 		}
595 
596 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
597 			break;
598 		dsl_dataset_rele(ds, FTAG);
599 		ds = prev;
600 		prev = NULL;
601 	}
602 
603 	if (prev == NULL) {
604 		prev = dp->dp_origin_snap;
605 
606 		/*
607 		 * The $ORIGIN can't have any data, or the accounting
608 		 * will be wrong.
609 		 */
610 		ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
611 
612 		/* The origin doesn't get attached to itself */
613 		if (ds->ds_object == prev->ds_object) {
614 			dsl_dataset_rele(ds, FTAG);
615 			return (0);
616 		}
617 
618 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
619 		ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
620 		ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
621 
622 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
623 		ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
624 
625 		dmu_buf_will_dirty(prev->ds_dbuf, tx);
626 		prev->ds_phys->ds_num_children++;
627 
628 		if (ds->ds_phys->ds_next_snap_obj == 0) {
629 			ASSERT(ds->ds_prev == NULL);
630 			VERIFY(0 == dsl_dataset_hold_obj(dp,
631 			    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
632 		}
633 	}
634 
635 	ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
636 	ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
637 
638 	if (prev->ds_phys->ds_next_clones_obj == 0) {
639 		dmu_buf_will_dirty(prev->ds_dbuf, tx);
640 		prev->ds_phys->ds_next_clones_obj =
641 		    zap_create(dp->dp_meta_objset,
642 		    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
643 	}
644 	VERIFY(0 == zap_add_int(dp->dp_meta_objset,
645 	    prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
646 
647 	dsl_dataset_rele(ds, FTAG);
648 	if (prev != dp->dp_origin_snap)
649 		dsl_dataset_rele(prev, FTAG);
650 	return (0);
651 }
652 
653 void
654 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
655 {
656 	ASSERT(dmu_tx_is_syncing(tx));
657 	ASSERT(dp->dp_origin_snap != NULL);
658 
659 	VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
660 	    tx, DS_FIND_CHILDREN));
661 }
662 
663 /* ARGSUSED */
664 static int
665 upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
666 {
667 	dmu_tx_t *tx = arg;
668 	dsl_dataset_t *ds;
669 	dsl_pool_t *dp = spa_get_dsl(spa);
670 	objset_t *mos = dp->dp_meta_objset;
671 
672 	VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
673 
674 	if (ds->ds_dir->dd_phys->dd_origin_obj) {
675 		dsl_dataset_t *origin;
676 
677 		VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
678 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin));
679 
680 		if (origin->ds_dir->dd_phys->dd_clones == 0) {
681 			dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
682 			origin->ds_dir->dd_phys->dd_clones = zap_create(mos,
683 			    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
684 		}
685 
686 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
687 		    origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
688 
689 		dsl_dataset_rele(origin, FTAG);
690 	}
691 
692 	dsl_dataset_rele(ds, FTAG);
693 	return (0);
694 }
695 
696 void
697 dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
698 {
699 	ASSERT(dmu_tx_is_syncing(tx));
700 	uint64_t obj;
701 
702 	(void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx);
703 	VERIFY(0 == dsl_pool_open_special_dir(dp,
704 	    FREE_DIR_NAME, &dp->dp_free_dir));
705 
706 	/*
707 	 * We can't use bpobj_alloc(), because spa_version() still
708 	 * returns the old version, and we need a new-version bpobj with
709 	 * subobj support.  So call dmu_object_alloc() directly.
710 	 */
711 	obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ,
712 	    SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx);
713 	VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
714 	    DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
715 	VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
716 	    dp->dp_meta_objset, obj));
717 
718 	VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL,
719 	    upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN));
720 }
721 
722 void
723 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
724 {
725 	uint64_t dsobj;
726 	dsl_dataset_t *ds;
727 
728 	ASSERT(dmu_tx_is_syncing(tx));
729 	ASSERT(dp->dp_origin_snap == NULL);
730 
731 	/* create the origin dir, ds, & snap-ds */
732 	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
733 	dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
734 	    NULL, 0, kcred, tx);
735 	VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
736 	dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx);
737 	VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
738 	    dp, &dp->dp_origin_snap));
739 	dsl_dataset_rele(ds, FTAG);
740 	rw_exit(&dp->dp_config_rwlock);
741 }
742 
743 taskq_t *
744 dsl_pool_vnrele_taskq(dsl_pool_t *dp)
745 {
746 	return (dp->dp_vnrele_taskq);
747 }
748 
749 /*
750  * Walk through the pool-wide zap object of temporary snapshot user holds
751  * and release them.
752  */
753 void
754 dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp)
755 {
756 	zap_attribute_t za;
757 	zap_cursor_t zc;
758 	objset_t *mos = dp->dp_meta_objset;
759 	uint64_t zapobj = dp->dp_tmp_userrefs_obj;
760 
761 	if (zapobj == 0)
762 		return;
763 	ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
764 
765 	for (zap_cursor_init(&zc, mos, zapobj);
766 	    zap_cursor_retrieve(&zc, &za) == 0;
767 	    zap_cursor_advance(&zc)) {
768 		char *htag;
769 		uint64_t dsobj;
770 
771 		htag = strchr(za.za_name, '-');
772 		*htag = '\0';
773 		++htag;
774 		dsobj = strtonum(za.za_name, NULL);
775 		(void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE);
776 	}
777 	zap_cursor_fini(&zc);
778 }
779 
780 /*
781  * Create the pool-wide zap object for storing temporary snapshot holds.
782  */
783 void
784 dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx)
785 {
786 	objset_t *mos = dp->dp_meta_objset;
787 
788 	ASSERT(dp->dp_tmp_userrefs_obj == 0);
789 	ASSERT(dmu_tx_is_syncing(tx));
790 
791 	dp->dp_tmp_userrefs_obj = zap_create(mos, DMU_OT_USERREFS,
792 	    DMU_OT_NONE, 0, tx);
793 
794 	VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS,
795 	    sizeof (uint64_t), 1, &dp->dp_tmp_userrefs_obj, tx) == 0);
796 }
797 
798 static int
799 dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
800     const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding)
801 {
802 	objset_t *mos = dp->dp_meta_objset;
803 	uint64_t zapobj = dp->dp_tmp_userrefs_obj;
804 	char *name;
805 	int error;
806 
807 	ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
808 	ASSERT(dmu_tx_is_syncing(tx));
809 
810 	/*
811 	 * If the pool was created prior to SPA_VERSION_USERREFS, the
812 	 * zap object for temporary holds might not exist yet.
813 	 */
814 	if (zapobj == 0) {
815 		if (holding) {
816 			dsl_pool_user_hold_create_obj(dp, tx);
817 			zapobj = dp->dp_tmp_userrefs_obj;
818 		} else {
819 			return (ENOENT);
820 		}
821 	}
822 
823 	name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag);
824 	if (holding)
825 		error = zap_add(mos, zapobj, name, 8, 1, now, tx);
826 	else
827 		error = zap_remove(mos, zapobj, name, tx);
828 	strfree(name);
829 
830 	return (error);
831 }
832 
833 /*
834  * Add a temporary hold for the given dataset object and tag.
835  */
836 int
837 dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
838     uint64_t *now, dmu_tx_t *tx)
839 {
840 	return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE));
841 }
842 
843 /*
844  * Release a temporary hold for the given dataset object and tag.
845  */
846 int
847 dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
848     dmu_tx_t *tx)
849 {
850 	return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL,
851 	    tx, B_FALSE));
852 }
853