xref: /illumos-gate/usr/src/uts/common/fs/zfs/dsl_pool.c (revision 0fd90d51ca56bcfe45114c5f0b74d50ad612537f)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
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>
32 #include <sys/arc.h>
33 #include <sys/zap.h>
34 #include <sys/zio.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>
39 
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 */
43 
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 
49 kmutex_t zfs_write_limit_lock;
50 
51 static pgcnt_t old_physmem = 0;
52 
53 static int
54 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
55 {
56 	uint64_t obj;
57 	int err;
58 
59 	err = zap_lookup(dp->dp_meta_objset,
60 	    dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
61 	    name, sizeof (obj), 1, &obj);
62 	if (err)
63 		return (err);
64 
65 	return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
66 }
67 
68 static dsl_pool_t *
69 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
70 {
71 	dsl_pool_t *dp;
72 	blkptr_t *bp = spa_get_rootblkptr(spa);
73 
74 	dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
75 	dp->dp_spa = spa;
76 	dp->dp_meta_rootbp = *bp;
77 	rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
78 	dp->dp_write_limit = zfs_write_limit_min;
79 	txg_init(dp, txg);
80 
81 	txg_list_create(&dp->dp_dirty_datasets,
82 	    offsetof(dsl_dataset_t, ds_dirty_link));
83 	txg_list_create(&dp->dp_dirty_dirs,
84 	    offsetof(dsl_dir_t, dd_dirty_link));
85 	txg_list_create(&dp->dp_sync_tasks,
86 	    offsetof(dsl_sync_task_group_t, dstg_node));
87 	list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t),
88 	    offsetof(dsl_dataset_t, ds_synced_link));
89 
90 	mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
91 	mutex_init(&dp->dp_scrub_cancel_lock, NULL, MUTEX_DEFAULT, NULL);
92 
93 	dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
94 	    1, 4, 0);
95 
96 	return (dp);
97 }
98 
99 int
100 dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
101 {
102 	int err;
103 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
104 	dsl_dir_t *dd;
105 	dsl_dataset_t *ds;
106 	objset_impl_t *osi;
107 
108 	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
109 	err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &osi);
110 	if (err)
111 		goto out;
112 	dp->dp_meta_objset = &osi->os;
113 
114 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
115 	    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
116 	    &dp->dp_root_dir_obj);
117 	if (err)
118 		goto out;
119 
120 	err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
121 	    NULL, dp, &dp->dp_root_dir);
122 	if (err)
123 		goto out;
124 
125 	err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
126 	if (err)
127 		goto out;
128 
129 	if (spa_version(spa) >= SPA_VERSION_ORIGIN) {
130 		err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
131 		if (err)
132 			goto out;
133 		err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
134 		    FTAG, &ds);
135 		if (err == 0) {
136 			err = dsl_dataset_hold_obj(dp,
137 			    ds->ds_phys->ds_prev_snap_obj, dp,
138 			    &dp->dp_origin_snap);
139 			dsl_dataset_rele(ds, FTAG);
140 		}
141 		dsl_dir_close(dd, dp);
142 		if (err)
143 			goto out;
144 	}
145 
146 	/* get scrub status */
147 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
148 	    DMU_POOL_SCRUB_FUNC, sizeof (uint32_t), 1,
149 	    &dp->dp_scrub_func);
150 	if (err == 0) {
151 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
152 		    DMU_POOL_SCRUB_QUEUE, sizeof (uint64_t), 1,
153 		    &dp->dp_scrub_queue_obj);
154 		if (err)
155 			goto out;
156 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
157 		    DMU_POOL_SCRUB_MIN_TXG, sizeof (uint64_t), 1,
158 		    &dp->dp_scrub_min_txg);
159 		if (err)
160 			goto out;
161 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
162 		    DMU_POOL_SCRUB_MAX_TXG, sizeof (uint64_t), 1,
163 		    &dp->dp_scrub_max_txg);
164 		if (err)
165 			goto out;
166 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
167 		    DMU_POOL_SCRUB_BOOKMARK, sizeof (uint64_t), 4,
168 		    &dp->dp_scrub_bookmark);
169 		if (err)
170 			goto out;
171 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
172 		    DMU_POOL_SCRUB_ERRORS, sizeof (uint64_t), 1,
173 		    &spa->spa_scrub_errors);
174 		if (err)
175 			goto out;
176 		if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) {
177 			/*
178 			 * A new-type scrub was in progress on an old
179 			 * pool.  Restart from the beginning, since the
180 			 * old software may have changed the pool in the
181 			 * meantime.
182 			 */
183 			dsl_pool_scrub_restart(dp);
184 		}
185 	} else {
186 		/*
187 		 * It's OK if there is no scrub in progress (and if
188 		 * there was an I/O error, ignore it).
189 		 */
190 		err = 0;
191 	}
192 
193 out:
194 	rw_exit(&dp->dp_config_rwlock);
195 	if (err)
196 		dsl_pool_close(dp);
197 	else
198 		*dpp = dp;
199 
200 	return (err);
201 }
202 
203 void
204 dsl_pool_close(dsl_pool_t *dp)
205 {
206 	/* drop our references from dsl_pool_open() */
207 
208 	/*
209 	 * Since we held the origin_snap from "syncing" context (which
210 	 * includes pool-opening context), it actually only got a "ref"
211 	 * and not a hold, so just drop that here.
212 	 */
213 	if (dp->dp_origin_snap)
214 		dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
215 	if (dp->dp_mos_dir)
216 		dsl_dir_close(dp->dp_mos_dir, dp);
217 	if (dp->dp_root_dir)
218 		dsl_dir_close(dp->dp_root_dir, dp);
219 
220 	/* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
221 	if (dp->dp_meta_objset)
222 		dmu_objset_evict(NULL, dp->dp_meta_objset->os);
223 
224 	txg_list_destroy(&dp->dp_dirty_datasets);
225 	txg_list_destroy(&dp->dp_dirty_dirs);
226 	list_destroy(&dp->dp_synced_datasets);
227 
228 	arc_flush(dp->dp_spa);
229 	txg_fini(dp);
230 	rw_destroy(&dp->dp_config_rwlock);
231 	mutex_destroy(&dp->dp_lock);
232 	mutex_destroy(&dp->dp_scrub_cancel_lock);
233 	taskq_destroy(dp->dp_vnrele_taskq);
234 	if (dp->dp_blkstats)
235 		kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
236 	kmem_free(dp, sizeof (dsl_pool_t));
237 }
238 
239 dsl_pool_t *
240 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
241 {
242 	int err;
243 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
244 	dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
245 	objset_impl_t *osip;
246 	dsl_dataset_t *ds;
247 	uint64_t dsobj;
248 
249 	/* create and open the MOS (meta-objset) */
250 	dp->dp_meta_objset = &dmu_objset_create_impl(spa,
251 	    NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx)->os;
252 
253 	/* create the pool directory */
254 	err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
255 	    DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
256 	ASSERT3U(err, ==, 0);
257 
258 	/* create and open the root dir */
259 	dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
260 	VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
261 	    NULL, dp, &dp->dp_root_dir));
262 
263 	/* create and open the meta-objset dir */
264 	(void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
265 	VERIFY(0 == dsl_pool_open_special_dir(dp,
266 	    MOS_DIR_NAME, &dp->dp_mos_dir));
267 
268 	if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
269 		dsl_pool_create_origin(dp, tx);
270 
271 	/* create the root dataset */
272 	dsobj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
273 
274 	/* create the root objset */
275 	VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
276 	osip = dmu_objset_create_impl(dp->dp_spa, ds,
277 	    dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
278 #ifdef _KERNEL
279 	zfs_create_fs(&osip->os, kcred, zplprops, tx);
280 #endif
281 	dsl_dataset_rele(ds, FTAG);
282 
283 	dmu_tx_commit(tx);
284 
285 	return (dp);
286 }
287 
288 void
289 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
290 {
291 	zio_t *zio;
292 	dmu_tx_t *tx;
293 	dsl_dir_t *dd;
294 	dsl_dataset_t *ds;
295 	dsl_sync_task_group_t *dstg;
296 	objset_impl_t *mosi = dp->dp_meta_objset->os;
297 	hrtime_t start, write_time;
298 	uint64_t data_written;
299 	int err;
300 
301 	tx = dmu_tx_create_assigned(dp, txg);
302 
303 	dp->dp_read_overhead = 0;
304 	start = gethrtime();
305 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
306 	while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
307 		if (!list_link_active(&ds->ds_synced_link))
308 			list_insert_tail(&dp->dp_synced_datasets, ds);
309 		else
310 			dmu_buf_rele(ds->ds_dbuf, ds);
311 		dsl_dataset_sync(ds, zio, tx);
312 	}
313 	DTRACE_PROBE(pool_sync__1setup);
314 
315 	err = zio_wait(zio);
316 	write_time = gethrtime() - start;
317 	ASSERT(err == 0);
318 	DTRACE_PROBE(pool_sync__2rootzio);
319 
320 	while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
321 		dsl_sync_task_group_sync(dstg, tx);
322 	DTRACE_PROBE(pool_sync__3task);
323 
324 	start = gethrtime();
325 	while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
326 		dsl_dir_sync(dd, tx);
327 	write_time += gethrtime() - start;
328 
329 	if (spa_sync_pass(dp->dp_spa) == 1)
330 		dsl_pool_scrub_sync(dp, tx);
331 
332 	start = gethrtime();
333 	if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
334 	    list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
335 		zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
336 		dmu_objset_sync(mosi, zio, tx);
337 		err = zio_wait(zio);
338 		ASSERT(err == 0);
339 		dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
340 		spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
341 	}
342 	write_time += gethrtime() - start;
343 	DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
344 	    hrtime_t, dp->dp_read_overhead);
345 	write_time -= dp->dp_read_overhead;
346 
347 	dmu_tx_commit(tx);
348 
349 	data_written = dp->dp_space_towrite[txg & TXG_MASK];
350 	dp->dp_space_towrite[txg & TXG_MASK] = 0;
351 	ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
352 
353 	/*
354 	 * If the write limit max has not been explicitly set, set it
355 	 * to a fraction of available physical memory (default 1/8th).
356 	 * Note that we must inflate the limit because the spa
357 	 * inflates write sizes to account for data replication.
358 	 * Check this each sync phase to catch changing memory size.
359 	 */
360 	if (physmem != old_physmem && zfs_write_limit_shift) {
361 		mutex_enter(&zfs_write_limit_lock);
362 		old_physmem = physmem;
363 		zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
364 		zfs_write_limit_inflated = MAX(zfs_write_limit_min,
365 		    spa_get_asize(dp->dp_spa, zfs_write_limit_max));
366 		mutex_exit(&zfs_write_limit_lock);
367 	}
368 
369 	/*
370 	 * Attempt to keep the sync time consistent by adjusting the
371 	 * amount of write traffic allowed into each transaction group.
372 	 * Weight the throughput calculation towards the current value:
373 	 * 	thru = 3/4 old_thru + 1/4 new_thru
374 	 */
375 	ASSERT(zfs_write_limit_min > 0);
376 	if (data_written > zfs_write_limit_min / 8 && write_time > 0) {
377 		uint64_t throughput = (data_written * NANOSEC) / write_time;
378 		if (dp->dp_throughput)
379 			dp->dp_throughput = throughput / 4 +
380 			    3 * dp->dp_throughput / 4;
381 		else
382 			dp->dp_throughput = throughput;
383 		dp->dp_write_limit = MIN(zfs_write_limit_inflated,
384 		    MAX(zfs_write_limit_min,
385 		    dp->dp_throughput * zfs_txg_synctime));
386 	}
387 }
388 
389 void
390 dsl_pool_zil_clean(dsl_pool_t *dp)
391 {
392 	dsl_dataset_t *ds;
393 
394 	while (ds = list_head(&dp->dp_synced_datasets)) {
395 		list_remove(&dp->dp_synced_datasets, ds);
396 		ASSERT(ds->ds_user_ptr != NULL);
397 		zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
398 		dmu_buf_rele(ds->ds_dbuf, ds);
399 	}
400 }
401 
402 /*
403  * TRUE if the current thread is the tx_sync_thread or if we
404  * are being called from SPA context during pool initialization.
405  */
406 int
407 dsl_pool_sync_context(dsl_pool_t *dp)
408 {
409 	return (curthread == dp->dp_tx.tx_sync_thread ||
410 	    spa_get_dsl(dp->dp_spa) == NULL);
411 }
412 
413 uint64_t
414 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
415 {
416 	uint64_t space, resv;
417 
418 	/*
419 	 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
420 	 * efficiency.
421 	 * XXX The intent log is not accounted for, so it must fit
422 	 * within this slop.
423 	 *
424 	 * If we're trying to assess whether it's OK to do a free,
425 	 * cut the reservation in half to allow forward progress
426 	 * (e.g. make it possible to rm(1) files from a full pool).
427 	 */
428 	space = spa_get_dspace(dp->dp_spa);
429 	resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
430 	if (netfree)
431 		resv >>= 1;
432 
433 	return (space - resv);
434 }
435 
436 int
437 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
438 {
439 	uint64_t reserved = 0;
440 	uint64_t write_limit = (zfs_write_limit_override ?
441 	    zfs_write_limit_override : dp->dp_write_limit);
442 
443 	if (zfs_no_write_throttle) {
444 		atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
445 		    space);
446 		return (0);
447 	}
448 
449 	/*
450 	 * Check to see if we have exceeded the maximum allowed IO for
451 	 * this transaction group.  We can do this without locks since
452 	 * a little slop here is ok.  Note that we do the reserved check
453 	 * with only half the requested reserve: this is because the
454 	 * reserve requests are worst-case, and we really don't want to
455 	 * throttle based off of worst-case estimates.
456 	 */
457 	if (write_limit > 0) {
458 		reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
459 		    + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
460 
461 		if (reserved && reserved > write_limit)
462 			return (ERESTART);
463 	}
464 
465 	atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
466 
467 	/*
468 	 * If this transaction group is over 7/8ths capacity, delay
469 	 * the caller 1 clock tick.  This will slow down the "fill"
470 	 * rate until the sync process can catch up with us.
471 	 */
472 	if (reserved && reserved > (write_limit - (write_limit >> 3)))
473 		txg_delay(dp, tx->tx_txg, 1);
474 
475 	return (0);
476 }
477 
478 void
479 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
480 {
481 	ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
482 	atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
483 }
484 
485 void
486 dsl_pool_memory_pressure(dsl_pool_t *dp)
487 {
488 	uint64_t space_inuse = 0;
489 	int i;
490 
491 	if (dp->dp_write_limit == zfs_write_limit_min)
492 		return;
493 
494 	for (i = 0; i < TXG_SIZE; i++) {
495 		space_inuse += dp->dp_space_towrite[i];
496 		space_inuse += dp->dp_tempreserved[i];
497 	}
498 	dp->dp_write_limit = MAX(zfs_write_limit_min,
499 	    MIN(dp->dp_write_limit, space_inuse / 4));
500 }
501 
502 void
503 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
504 {
505 	if (space > 0) {
506 		mutex_enter(&dp->dp_lock);
507 		dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
508 		mutex_exit(&dp->dp_lock);
509 	}
510 }
511 
512 /* ARGSUSED */
513 static int
514 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
515 {
516 	dmu_tx_t *tx = arg;
517 	dsl_dataset_t *ds, *prev = NULL;
518 	int err;
519 	dsl_pool_t *dp = spa_get_dsl(spa);
520 
521 	err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
522 	if (err)
523 		return (err);
524 
525 	while (ds->ds_phys->ds_prev_snap_obj != 0) {
526 		err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
527 		    FTAG, &prev);
528 		if (err) {
529 			dsl_dataset_rele(ds, FTAG);
530 			return (err);
531 		}
532 
533 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
534 			break;
535 		dsl_dataset_rele(ds, FTAG);
536 		ds = prev;
537 		prev = NULL;
538 	}
539 
540 	if (prev == NULL) {
541 		prev = dp->dp_origin_snap;
542 
543 		/*
544 		 * The $ORIGIN can't have any data, or the accounting
545 		 * will be wrong.
546 		 */
547 		ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
548 
549 		/* The origin doesn't get attached to itself */
550 		if (ds->ds_object == prev->ds_object) {
551 			dsl_dataset_rele(ds, FTAG);
552 			return (0);
553 		}
554 
555 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
556 		ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
557 		ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
558 
559 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
560 		ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
561 
562 		dmu_buf_will_dirty(prev->ds_dbuf, tx);
563 		prev->ds_phys->ds_num_children++;
564 
565 		if (ds->ds_phys->ds_next_snap_obj == 0) {
566 			ASSERT(ds->ds_prev == NULL);
567 			VERIFY(0 == dsl_dataset_hold_obj(dp,
568 			    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
569 		}
570 	}
571 
572 	ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
573 	ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
574 
575 	if (prev->ds_phys->ds_next_clones_obj == 0) {
576 		prev->ds_phys->ds_next_clones_obj =
577 		    zap_create(dp->dp_meta_objset,
578 		    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
579 	}
580 	VERIFY(0 == zap_add_int(dp->dp_meta_objset,
581 	    prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
582 
583 	dsl_dataset_rele(ds, FTAG);
584 	if (prev != dp->dp_origin_snap)
585 		dsl_dataset_rele(prev, FTAG);
586 	return (0);
587 }
588 
589 void
590 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
591 {
592 	ASSERT(dmu_tx_is_syncing(tx));
593 	ASSERT(dp->dp_origin_snap != NULL);
594 
595 	(void) dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
596 	    tx, DS_FIND_CHILDREN);
597 }
598 
599 void
600 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
601 {
602 	uint64_t dsobj;
603 	dsl_dataset_t *ds;
604 
605 	ASSERT(dmu_tx_is_syncing(tx));
606 	ASSERT(dp->dp_origin_snap == NULL);
607 
608 	/* create the origin dir, ds, & snap-ds */
609 	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
610 	dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
611 	    NULL, 0, kcred, tx);
612 	VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
613 	dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, kcred, tx);
614 	VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
615 	    dp, &dp->dp_origin_snap));
616 	dsl_dataset_rele(ds, FTAG);
617 	rw_exit(&dp->dp_config_rwlock);
618 }
619 
620 taskq_t *
621 dsl_pool_vnrele_taskq(dsl_pool_t *dp)
622 {
623 	return (dp->dp_vnrele_taskq);
624 }
625