xref: /titanic_51/usr/src/uts/common/fs/zfs/dsl_pool.c (revision d29f5a711240f866521445b1656d114da090335e)
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 2008 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 	return (dp);
94 }
95 
96 int
97 dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
98 {
99 	int err;
100 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
101 	dsl_dir_t *dd;
102 	dsl_dataset_t *ds;
103 	objset_impl_t *osi;
104 
105 	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
106 	err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &osi);
107 	if (err)
108 		goto out;
109 	dp->dp_meta_objset = &osi->os;
110 
111 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
112 	    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
113 	    &dp->dp_root_dir_obj);
114 	if (err)
115 		goto out;
116 
117 	err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
118 	    NULL, dp, &dp->dp_root_dir);
119 	if (err)
120 		goto out;
121 
122 	err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
123 	if (err)
124 		goto out;
125 
126 	if (spa_version(spa) >= SPA_VERSION_ORIGIN) {
127 		err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
128 		if (err)
129 			goto out;
130 		err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
131 		    FTAG, &ds);
132 		if (err)
133 			goto out;
134 		err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
135 		    dp, &dp->dp_origin_snap);
136 		if (err)
137 			goto out;
138 		dsl_dataset_rele(ds, FTAG);
139 		dsl_dir_close(dd, dp);
140 	}
141 
142 	/* get scrub status */
143 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
144 	    DMU_POOL_SCRUB_FUNC, sizeof (uint32_t), 1,
145 	    &dp->dp_scrub_func);
146 	if (err == 0) {
147 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
148 		    DMU_POOL_SCRUB_QUEUE, sizeof (uint64_t), 1,
149 		    &dp->dp_scrub_queue_obj);
150 		if (err)
151 			goto out;
152 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
153 		    DMU_POOL_SCRUB_MIN_TXG, sizeof (uint64_t), 1,
154 		    &dp->dp_scrub_min_txg);
155 		if (err)
156 			goto out;
157 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
158 		    DMU_POOL_SCRUB_MAX_TXG, sizeof (uint64_t), 1,
159 		    &dp->dp_scrub_max_txg);
160 		if (err)
161 			goto out;
162 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
163 		    DMU_POOL_SCRUB_BOOKMARK, sizeof (uint64_t), 4,
164 		    &dp->dp_scrub_bookmark);
165 		if (err)
166 			goto out;
167 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
168 		    DMU_POOL_SCRUB_ERRORS, sizeof (uint64_t), 1,
169 		    &spa->spa_scrub_errors);
170 		if (err)
171 			goto out;
172 		if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) {
173 			/*
174 			 * A new-type scrub was in progress on an old
175 			 * pool.  Restart from the beginning, since the
176 			 * old software may have changed the pool in the
177 			 * meantime.
178 			 */
179 			dsl_pool_scrub_restart(dp);
180 		}
181 	} else {
182 		/*
183 		 * It's OK if there is no scrub in progress (and if
184 		 * there was an I/O error, ignore it).
185 		 */
186 		err = 0;
187 	}
188 
189 out:
190 	rw_exit(&dp->dp_config_rwlock);
191 	if (err)
192 		dsl_pool_close(dp);
193 	else
194 		*dpp = dp;
195 
196 	return (err);
197 }
198 
199 void
200 dsl_pool_close(dsl_pool_t *dp)
201 {
202 	/* drop our references from dsl_pool_open() */
203 
204 	/*
205 	 * Since we held the origin_snap from "syncing" context (which
206 	 * includes pool-opening context), it actually only got a "ref"
207 	 * and not a hold, so just drop that here.
208 	 */
209 	if (dp->dp_origin_snap)
210 		dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
211 	if (dp->dp_mos_dir)
212 		dsl_dir_close(dp->dp_mos_dir, dp);
213 	if (dp->dp_root_dir)
214 		dsl_dir_close(dp->dp_root_dir, dp);
215 
216 	/* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
217 	if (dp->dp_meta_objset)
218 		dmu_objset_evict(NULL, dp->dp_meta_objset->os);
219 
220 	txg_list_destroy(&dp->dp_dirty_datasets);
221 	txg_list_destroy(&dp->dp_dirty_dirs);
222 	list_destroy(&dp->dp_synced_datasets);
223 
224 	arc_flush(dp->dp_spa);
225 	txg_fini(dp);
226 	rw_destroy(&dp->dp_config_rwlock);
227 	mutex_destroy(&dp->dp_lock);
228 	mutex_destroy(&dp->dp_scrub_cancel_lock);
229 	kmem_free(dp, sizeof (dsl_pool_t));
230 }
231 
232 dsl_pool_t *
233 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
234 {
235 	int err;
236 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
237 	dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
238 	objset_impl_t *osip;
239 	dsl_dataset_t *ds;
240 	uint64_t dsobj;
241 
242 	/* create and open the MOS (meta-objset) */
243 	dp->dp_meta_objset = &dmu_objset_create_impl(spa,
244 	    NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx)->os;
245 
246 	/* create the pool directory */
247 	err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
248 	    DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
249 	ASSERT3U(err, ==, 0);
250 
251 	/* create and open the root dir */
252 	dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
253 	VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
254 	    NULL, dp, &dp->dp_root_dir));
255 
256 	/* create and open the meta-objset dir */
257 	(void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
258 	VERIFY(0 == dsl_pool_open_special_dir(dp,
259 	    MOS_DIR_NAME, &dp->dp_mos_dir));
260 
261 	if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
262 		dsl_pool_create_origin(dp, tx);
263 
264 	/* create the root dataset */
265 	dsobj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
266 
267 	/* create the root objset */
268 	VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
269 	osip = dmu_objset_create_impl(dp->dp_spa, ds,
270 	    dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
271 #ifdef _KERNEL
272 	zfs_create_fs(&osip->os, kcred, zplprops, tx);
273 #endif
274 	dsl_dataset_rele(ds, FTAG);
275 
276 	dmu_tx_commit(tx);
277 
278 	return (dp);
279 }
280 
281 void
282 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
283 {
284 	zio_t *zio;
285 	dmu_tx_t *tx;
286 	dsl_dir_t *dd;
287 	dsl_dataset_t *ds;
288 	dsl_sync_task_group_t *dstg;
289 	objset_impl_t *mosi = dp->dp_meta_objset->os;
290 	hrtime_t start, write_time;
291 	uint64_t data_written;
292 	int err;
293 
294 	tx = dmu_tx_create_assigned(dp, txg);
295 
296 	dp->dp_read_overhead = 0;
297 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
298 	while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
299 		if (!list_link_active(&ds->ds_synced_link))
300 			list_insert_tail(&dp->dp_synced_datasets, ds);
301 		else
302 			dmu_buf_rele(ds->ds_dbuf, ds);
303 		dsl_dataset_sync(ds, zio, tx);
304 	}
305 	DTRACE_PROBE(pool_sync__1setup);
306 
307 	start = gethrtime();
308 	err = zio_wait(zio);
309 	write_time = gethrtime() - start;
310 	ASSERT(err == 0);
311 	DTRACE_PROBE(pool_sync__2rootzio);
312 
313 	while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
314 		dsl_sync_task_group_sync(dstg, tx);
315 	DTRACE_PROBE(pool_sync__3task);
316 
317 	start = gethrtime();
318 	while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
319 		dsl_dir_sync(dd, tx);
320 	write_time += gethrtime() - start;
321 
322 	if (spa_sync_pass(dp->dp_spa) == 1)
323 		dsl_pool_scrub_sync(dp, tx);
324 
325 	start = gethrtime();
326 	if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
327 	    list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
328 		zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
329 		dmu_objset_sync(mosi, zio, tx);
330 		err = zio_wait(zio);
331 		ASSERT(err == 0);
332 		dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
333 		spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
334 	}
335 	write_time += gethrtime() - start;
336 	DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
337 	    hrtime_t, dp->dp_read_overhead);
338 	write_time -= dp->dp_read_overhead;
339 
340 	dmu_tx_commit(tx);
341 
342 	data_written = dp->dp_space_towrite[txg & TXG_MASK];
343 	dp->dp_space_towrite[txg & TXG_MASK] = 0;
344 	ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
345 
346 	/*
347 	 * If the write limit max has not been explicitly set, set it
348 	 * to a fraction of available physical memory (default 1/8th).
349 	 * Note that we must inflate the limit because the spa
350 	 * inflates write sizes to account for data replication.
351 	 * Check this each sync phase to catch changing memory size.
352 	 */
353 	if (physmem != old_physmem && zfs_write_limit_shift) {
354 		mutex_enter(&zfs_write_limit_lock);
355 		old_physmem = physmem;
356 		zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
357 		zfs_write_limit_inflated = MAX(zfs_write_limit_min,
358 		    spa_get_asize(dp->dp_spa, zfs_write_limit_max));
359 		mutex_exit(&zfs_write_limit_lock);
360 	}
361 
362 	/*
363 	 * Attempt to keep the sync time consistent by adjusting the
364 	 * amount of write traffic allowed into each transaction group.
365 	 * Weight the throughput calculation towards the current value:
366 	 * 	thru = 3/4 old_thru + 1/4 new_thru
367 	 */
368 	ASSERT(zfs_write_limit_min > 0);
369 	if (data_written > zfs_write_limit_min / 8 && write_time > 0) {
370 		uint64_t throughput = (data_written * NANOSEC) / write_time;
371 		if (dp->dp_throughput)
372 			dp->dp_throughput = throughput / 4 +
373 			    3 * dp->dp_throughput / 4;
374 		else
375 			dp->dp_throughput = throughput;
376 		dp->dp_write_limit = MIN(zfs_write_limit_inflated,
377 		    MAX(zfs_write_limit_min,
378 		    dp->dp_throughput * zfs_txg_synctime));
379 	}
380 }
381 
382 void
383 dsl_pool_zil_clean(dsl_pool_t *dp)
384 {
385 	dsl_dataset_t *ds;
386 
387 	while (ds = list_head(&dp->dp_synced_datasets)) {
388 		list_remove(&dp->dp_synced_datasets, ds);
389 		ASSERT(ds->ds_user_ptr != NULL);
390 		zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
391 		dmu_buf_rele(ds->ds_dbuf, ds);
392 	}
393 }
394 
395 /*
396  * TRUE if the current thread is the tx_sync_thread or if we
397  * are being called from SPA context during pool initialization.
398  */
399 int
400 dsl_pool_sync_context(dsl_pool_t *dp)
401 {
402 	return (curthread == dp->dp_tx.tx_sync_thread ||
403 	    spa_get_dsl(dp->dp_spa) == NULL);
404 }
405 
406 uint64_t
407 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
408 {
409 	uint64_t space, resv;
410 
411 	/*
412 	 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
413 	 * efficiency.
414 	 * XXX The intent log is not accounted for, so it must fit
415 	 * within this slop.
416 	 *
417 	 * If we're trying to assess whether it's OK to do a free,
418 	 * cut the reservation in half to allow forward progress
419 	 * (e.g. make it possible to rm(1) files from a full pool).
420 	 */
421 	space = spa_get_dspace(dp->dp_spa);
422 	resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
423 	if (netfree)
424 		resv >>= 1;
425 
426 	return (space - resv);
427 }
428 
429 int
430 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
431 {
432 	uint64_t reserved = 0;
433 	uint64_t write_limit = (zfs_write_limit_override ?
434 	    zfs_write_limit_override : dp->dp_write_limit);
435 
436 	if (zfs_no_write_throttle) {
437 		atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
438 		    space);
439 		return (0);
440 	}
441 
442 	/*
443 	 * Check to see if we have exceeded the maximum allowed IO for
444 	 * this transaction group.  We can do this without locks since
445 	 * a little slop here is ok.  Note that we do the reserved check
446 	 * with only half the requested reserve: this is because the
447 	 * reserve requests are worst-case, and we really don't want to
448 	 * throttle based off of worst-case estimates.
449 	 */
450 	if (write_limit > 0) {
451 		reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
452 		    + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
453 
454 		if (reserved && reserved > write_limit)
455 			return (ERESTART);
456 	}
457 
458 	atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
459 
460 	/*
461 	 * If this transaction group is over 7/8ths capacity, delay
462 	 * the caller 1 clock tick.  This will slow down the "fill"
463 	 * rate until the sync process can catch up with us.
464 	 */
465 	if (reserved && reserved > (write_limit - (write_limit >> 3)))
466 		txg_delay(dp, tx->tx_txg, 1);
467 
468 	return (0);
469 }
470 
471 void
472 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
473 {
474 	ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
475 	atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
476 }
477 
478 void
479 dsl_pool_memory_pressure(dsl_pool_t *dp)
480 {
481 	uint64_t space_inuse = 0;
482 	int i;
483 
484 	if (dp->dp_write_limit == zfs_write_limit_min)
485 		return;
486 
487 	for (i = 0; i < TXG_SIZE; i++) {
488 		space_inuse += dp->dp_space_towrite[i];
489 		space_inuse += dp->dp_tempreserved[i];
490 	}
491 	dp->dp_write_limit = MAX(zfs_write_limit_min,
492 	    MIN(dp->dp_write_limit, space_inuse / 4));
493 }
494 
495 void
496 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
497 {
498 	if (space > 0) {
499 		mutex_enter(&dp->dp_lock);
500 		dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
501 		mutex_exit(&dp->dp_lock);
502 	}
503 }
504 
505 /* ARGSUSED */
506 static int
507 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
508 {
509 	dmu_tx_t *tx = arg;
510 	dsl_dataset_t *ds, *prev = NULL;
511 	int err;
512 	dsl_pool_t *dp = spa_get_dsl(spa);
513 
514 	err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
515 	if (err)
516 		return (err);
517 
518 	while (ds->ds_phys->ds_prev_snap_obj != 0) {
519 		err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
520 		    FTAG, &prev);
521 		if (err) {
522 			dsl_dataset_rele(ds, FTAG);
523 			return (err);
524 		}
525 
526 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
527 			break;
528 		dsl_dataset_rele(ds, FTAG);
529 		ds = prev;
530 		prev = NULL;
531 	}
532 
533 	if (prev == NULL) {
534 		prev = dp->dp_origin_snap;
535 
536 		/*
537 		 * The $ORIGIN can't have any data, or the accounting
538 		 * will be wrong.
539 		 */
540 		ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
541 
542 		/* The origin doesn't get attached to itself */
543 		if (ds->ds_object == prev->ds_object) {
544 			dsl_dataset_rele(ds, FTAG);
545 			return (0);
546 		}
547 
548 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
549 		ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
550 		ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
551 
552 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
553 		ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
554 
555 		dmu_buf_will_dirty(prev->ds_dbuf, tx);
556 		prev->ds_phys->ds_num_children++;
557 
558 		if (ds->ds_phys->ds_next_snap_obj == 0) {
559 			ASSERT(ds->ds_prev == NULL);
560 			VERIFY(0 == dsl_dataset_hold_obj(dp,
561 			    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
562 		}
563 	}
564 
565 	ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
566 	ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
567 
568 	if (prev->ds_phys->ds_next_clones_obj == 0) {
569 		prev->ds_phys->ds_next_clones_obj =
570 		    zap_create(dp->dp_meta_objset,
571 		    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
572 	}
573 	VERIFY(0 == zap_add_int(dp->dp_meta_objset,
574 	    prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
575 
576 	dsl_dataset_rele(ds, FTAG);
577 	if (prev != dp->dp_origin_snap)
578 		dsl_dataset_rele(prev, FTAG);
579 	return (0);
580 }
581 
582 void
583 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
584 {
585 	ASSERT(dmu_tx_is_syncing(tx));
586 	ASSERT(dp->dp_origin_snap != NULL);
587 
588 	(void) dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
589 	    tx, DS_FIND_CHILDREN);
590 }
591 
592 void
593 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
594 {
595 	uint64_t dsobj;
596 	dsl_dataset_t *ds;
597 
598 	ASSERT(dmu_tx_is_syncing(tx));
599 	ASSERT(dp->dp_origin_snap == NULL);
600 
601 	/* create the origin dir, ds, & snap-ds */
602 	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
603 	dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
604 	    NULL, 0, kcred, tx);
605 	VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
606 	dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, kcred, tx);
607 	VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
608 	    dp, &dp->dp_origin_snap));
609 	dsl_dataset_rele(ds, FTAG);
610 	rw_exit(&dp->dp_config_rwlock);
611 }
612