xref: /titanic_51/usr/src/uts/common/fs/zfs/dsl_dir.c (revision cf988fac1debd92859f8068ee3d3e53782043469)
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) 2013 by Delphix. All rights reserved.
24  * Copyright (c) 2013 Martin Matuska. All rights reserved.
25  */
26 
27 #include <sys/dmu.h>
28 #include <sys/dmu_objset.h>
29 #include <sys/dmu_tx.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dsl_deleg.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/spa.h>
37 #include <sys/metaslab.h>
38 #include <sys/zap.h>
39 #include <sys/zio.h>
40 #include <sys/arc.h>
41 #include <sys/sunddi.h>
42 #include "zfs_namecheck.h"
43 
44 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
45 
46 /* ARGSUSED */
47 static void
48 dsl_dir_evict(dmu_buf_t *db, void *arg)
49 {
50 	dsl_dir_t *dd = arg;
51 	dsl_pool_t *dp = dd->dd_pool;
52 	int t;
53 
54 	for (t = 0; t < TXG_SIZE; t++) {
55 		ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
56 		ASSERT(dd->dd_tempreserved[t] == 0);
57 		ASSERT(dd->dd_space_towrite[t] == 0);
58 	}
59 
60 	if (dd->dd_parent)
61 		dsl_dir_rele(dd->dd_parent, dd);
62 
63 	spa_close(dd->dd_pool->dp_spa, dd);
64 
65 	/*
66 	 * The props callback list should have been cleaned up by
67 	 * objset_evict().
68 	 */
69 	list_destroy(&dd->dd_prop_cbs);
70 	mutex_destroy(&dd->dd_lock);
71 	kmem_free(dd, sizeof (dsl_dir_t));
72 }
73 
74 int
75 dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
76     const char *tail, void *tag, dsl_dir_t **ddp)
77 {
78 	dmu_buf_t *dbuf;
79 	dsl_dir_t *dd;
80 	int err;
81 
82 	ASSERT(dsl_pool_config_held(dp));
83 
84 	err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
85 	if (err != 0)
86 		return (err);
87 	dd = dmu_buf_get_user(dbuf);
88 #ifdef ZFS_DEBUG
89 	{
90 		dmu_object_info_t doi;
91 		dmu_object_info_from_db(dbuf, &doi);
92 		ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
93 		ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
94 	}
95 #endif
96 	if (dd == NULL) {
97 		dsl_dir_t *winner;
98 
99 		dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
100 		dd->dd_object = ddobj;
101 		dd->dd_dbuf = dbuf;
102 		dd->dd_pool = dp;
103 		dd->dd_phys = dbuf->db_data;
104 		mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
105 
106 		list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
107 		    offsetof(dsl_prop_cb_record_t, cbr_node));
108 
109 		dsl_dir_snap_cmtime_update(dd);
110 
111 		if (dd->dd_phys->dd_parent_obj) {
112 			err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
113 			    NULL, dd, &dd->dd_parent);
114 			if (err != 0)
115 				goto errout;
116 			if (tail) {
117 #ifdef ZFS_DEBUG
118 				uint64_t foundobj;
119 
120 				err = zap_lookup(dp->dp_meta_objset,
121 				    dd->dd_parent->dd_phys->dd_child_dir_zapobj,
122 				    tail, sizeof (foundobj), 1, &foundobj);
123 				ASSERT(err || foundobj == ddobj);
124 #endif
125 				(void) strcpy(dd->dd_myname, tail);
126 			} else {
127 				err = zap_value_search(dp->dp_meta_objset,
128 				    dd->dd_parent->dd_phys->dd_child_dir_zapobj,
129 				    ddobj, 0, dd->dd_myname);
130 			}
131 			if (err != 0)
132 				goto errout;
133 		} else {
134 			(void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
135 		}
136 
137 		if (dsl_dir_is_clone(dd)) {
138 			dmu_buf_t *origin_bonus;
139 			dsl_dataset_phys_t *origin_phys;
140 
141 			/*
142 			 * We can't open the origin dataset, because
143 			 * that would require opening this dsl_dir.
144 			 * Just look at its phys directly instead.
145 			 */
146 			err = dmu_bonus_hold(dp->dp_meta_objset,
147 			    dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
148 			if (err != 0)
149 				goto errout;
150 			origin_phys = origin_bonus->db_data;
151 			dd->dd_origin_txg =
152 			    origin_phys->ds_creation_txg;
153 			dmu_buf_rele(origin_bonus, FTAG);
154 		}
155 
156 		winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
157 		    dsl_dir_evict);
158 		if (winner) {
159 			if (dd->dd_parent)
160 				dsl_dir_rele(dd->dd_parent, dd);
161 			mutex_destroy(&dd->dd_lock);
162 			kmem_free(dd, sizeof (dsl_dir_t));
163 			dd = winner;
164 		} else {
165 			spa_open_ref(dp->dp_spa, dd);
166 		}
167 	}
168 
169 	/*
170 	 * The dsl_dir_t has both open-to-close and instantiate-to-evict
171 	 * holds on the spa.  We need the open-to-close holds because
172 	 * otherwise the spa_refcnt wouldn't change when we open a
173 	 * dir which the spa also has open, so we could incorrectly
174 	 * think it was OK to unload/export/destroy the pool.  We need
175 	 * the instantiate-to-evict hold because the dsl_dir_t has a
176 	 * pointer to the dd_pool, which has a pointer to the spa_t.
177 	 */
178 	spa_open_ref(dp->dp_spa, tag);
179 	ASSERT3P(dd->dd_pool, ==, dp);
180 	ASSERT3U(dd->dd_object, ==, ddobj);
181 	ASSERT3P(dd->dd_dbuf, ==, dbuf);
182 	*ddp = dd;
183 	return (0);
184 
185 errout:
186 	if (dd->dd_parent)
187 		dsl_dir_rele(dd->dd_parent, dd);
188 	mutex_destroy(&dd->dd_lock);
189 	kmem_free(dd, sizeof (dsl_dir_t));
190 	dmu_buf_rele(dbuf, tag);
191 	return (err);
192 }
193 
194 void
195 dsl_dir_rele(dsl_dir_t *dd, void *tag)
196 {
197 	dprintf_dd(dd, "%s\n", "");
198 	spa_close(dd->dd_pool->dp_spa, tag);
199 	dmu_buf_rele(dd->dd_dbuf, tag);
200 }
201 
202 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
203 void
204 dsl_dir_name(dsl_dir_t *dd, char *buf)
205 {
206 	if (dd->dd_parent) {
207 		dsl_dir_name(dd->dd_parent, buf);
208 		(void) strcat(buf, "/");
209 	} else {
210 		buf[0] = '\0';
211 	}
212 	if (!MUTEX_HELD(&dd->dd_lock)) {
213 		/*
214 		 * recursive mutex so that we can use
215 		 * dprintf_dd() with dd_lock held
216 		 */
217 		mutex_enter(&dd->dd_lock);
218 		(void) strcat(buf, dd->dd_myname);
219 		mutex_exit(&dd->dd_lock);
220 	} else {
221 		(void) strcat(buf, dd->dd_myname);
222 	}
223 }
224 
225 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
226 int
227 dsl_dir_namelen(dsl_dir_t *dd)
228 {
229 	int result = 0;
230 
231 	if (dd->dd_parent) {
232 		/* parent's name + 1 for the "/" */
233 		result = dsl_dir_namelen(dd->dd_parent) + 1;
234 	}
235 
236 	if (!MUTEX_HELD(&dd->dd_lock)) {
237 		/* see dsl_dir_name */
238 		mutex_enter(&dd->dd_lock);
239 		result += strlen(dd->dd_myname);
240 		mutex_exit(&dd->dd_lock);
241 	} else {
242 		result += strlen(dd->dd_myname);
243 	}
244 
245 	return (result);
246 }
247 
248 static int
249 getcomponent(const char *path, char *component, const char **nextp)
250 {
251 	char *p;
252 
253 	if ((path == NULL) || (path[0] == '\0'))
254 		return (SET_ERROR(ENOENT));
255 	/* This would be a good place to reserve some namespace... */
256 	p = strpbrk(path, "/@");
257 	if (p && (p[1] == '/' || p[1] == '@')) {
258 		/* two separators in a row */
259 		return (SET_ERROR(EINVAL));
260 	}
261 	if (p == NULL || p == path) {
262 		/*
263 		 * if the first thing is an @ or /, it had better be an
264 		 * @ and it had better not have any more ats or slashes,
265 		 * and it had better have something after the @.
266 		 */
267 		if (p != NULL &&
268 		    (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
269 			return (SET_ERROR(EINVAL));
270 		if (strlen(path) >= MAXNAMELEN)
271 			return (SET_ERROR(ENAMETOOLONG));
272 		(void) strcpy(component, path);
273 		p = NULL;
274 	} else if (p[0] == '/') {
275 		if (p - path >= MAXNAMELEN)
276 			return (SET_ERROR(ENAMETOOLONG));
277 		(void) strncpy(component, path, p - path);
278 		component[p - path] = '\0';
279 		p++;
280 	} else if (p[0] == '@') {
281 		/*
282 		 * if the next separator is an @, there better not be
283 		 * any more slashes.
284 		 */
285 		if (strchr(path, '/'))
286 			return (SET_ERROR(EINVAL));
287 		if (p - path >= MAXNAMELEN)
288 			return (SET_ERROR(ENAMETOOLONG));
289 		(void) strncpy(component, path, p - path);
290 		component[p - path] = '\0';
291 	} else {
292 		panic("invalid p=%p", (void *)p);
293 	}
294 	*nextp = p;
295 	return (0);
296 }
297 
298 /*
299  * Return the dsl_dir_t, and possibly the last component which couldn't
300  * be found in *tail.  The name must be in the specified dsl_pool_t.  This
301  * thread must hold the dp_config_rwlock for the pool.  Returns NULL if the
302  * path is bogus, or if tail==NULL and we couldn't parse the whole name.
303  * (*tail)[0] == '@' means that the last component is a snapshot.
304  */
305 int
306 dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
307     dsl_dir_t **ddp, const char **tailp)
308 {
309 	char buf[MAXNAMELEN];
310 	const char *spaname, *next, *nextnext = NULL;
311 	int err;
312 	dsl_dir_t *dd;
313 	uint64_t ddobj;
314 
315 	err = getcomponent(name, buf, &next);
316 	if (err != 0)
317 		return (err);
318 
319 	/* Make sure the name is in the specified pool. */
320 	spaname = spa_name(dp->dp_spa);
321 	if (strcmp(buf, spaname) != 0)
322 		return (SET_ERROR(EINVAL));
323 
324 	ASSERT(dsl_pool_config_held(dp));
325 
326 	err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
327 	if (err != 0) {
328 		return (err);
329 	}
330 
331 	while (next != NULL) {
332 		dsl_dir_t *child_ds;
333 		err = getcomponent(next, buf, &nextnext);
334 		if (err != 0)
335 			break;
336 		ASSERT(next[0] != '\0');
337 		if (next[0] == '@')
338 			break;
339 		dprintf("looking up %s in obj%lld\n",
340 		    buf, dd->dd_phys->dd_child_dir_zapobj);
341 
342 		err = zap_lookup(dp->dp_meta_objset,
343 		    dd->dd_phys->dd_child_dir_zapobj,
344 		    buf, sizeof (ddobj), 1, &ddobj);
345 		if (err != 0) {
346 			if (err == ENOENT)
347 				err = 0;
348 			break;
349 		}
350 
351 		err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
352 		if (err != 0)
353 			break;
354 		dsl_dir_rele(dd, tag);
355 		dd = child_ds;
356 		next = nextnext;
357 	}
358 
359 	if (err != 0) {
360 		dsl_dir_rele(dd, tag);
361 		return (err);
362 	}
363 
364 	/*
365 	 * It's an error if there's more than one component left, or
366 	 * tailp==NULL and there's any component left.
367 	 */
368 	if (next != NULL &&
369 	    (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
370 		/* bad path name */
371 		dsl_dir_rele(dd, tag);
372 		dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
373 		err = SET_ERROR(ENOENT);
374 	}
375 	if (tailp != NULL)
376 		*tailp = next;
377 	*ddp = dd;
378 	return (err);
379 }
380 
381 uint64_t
382 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
383     dmu_tx_t *tx)
384 {
385 	objset_t *mos = dp->dp_meta_objset;
386 	uint64_t ddobj;
387 	dsl_dir_phys_t *ddphys;
388 	dmu_buf_t *dbuf;
389 
390 	ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
391 	    DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
392 	if (pds) {
393 		VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
394 		    name, sizeof (uint64_t), 1, &ddobj, tx));
395 	} else {
396 		/* it's the root dir */
397 		VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
398 		    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
399 	}
400 	VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
401 	dmu_buf_will_dirty(dbuf, tx);
402 	ddphys = dbuf->db_data;
403 
404 	ddphys->dd_creation_time = gethrestime_sec();
405 	if (pds)
406 		ddphys->dd_parent_obj = pds->dd_object;
407 	ddphys->dd_props_zapobj = zap_create(mos,
408 	    DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
409 	ddphys->dd_child_dir_zapobj = zap_create(mos,
410 	    DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
411 	if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
412 		ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
413 	dmu_buf_rele(dbuf, FTAG);
414 
415 	return (ddobj);
416 }
417 
418 boolean_t
419 dsl_dir_is_clone(dsl_dir_t *dd)
420 {
421 	return (dd->dd_phys->dd_origin_obj &&
422 	    (dd->dd_pool->dp_origin_snap == NULL ||
423 	    dd->dd_phys->dd_origin_obj !=
424 	    dd->dd_pool->dp_origin_snap->ds_object));
425 }
426 
427 void
428 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
429 {
430 	mutex_enter(&dd->dd_lock);
431 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
432 	    dd->dd_phys->dd_used_bytes);
433 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
434 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
435 	    dd->dd_phys->dd_reserved);
436 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
437 	    dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
438 	    (dd->dd_phys->dd_uncompressed_bytes * 100 /
439 	    dd->dd_phys->dd_compressed_bytes));
440 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
441 	    dd->dd_phys->dd_uncompressed_bytes);
442 	if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
443 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
444 		    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
445 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
446 		    dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
447 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
448 		    dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
449 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
450 		    dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
451 		    dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
452 	}
453 	mutex_exit(&dd->dd_lock);
454 
455 	if (dsl_dir_is_clone(dd)) {
456 		dsl_dataset_t *ds;
457 		char buf[MAXNAMELEN];
458 
459 		VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
460 		    dd->dd_phys->dd_origin_obj, FTAG, &ds));
461 		dsl_dataset_name(ds, buf);
462 		dsl_dataset_rele(ds, FTAG);
463 		dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
464 	}
465 }
466 
467 void
468 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
469 {
470 	dsl_pool_t *dp = dd->dd_pool;
471 
472 	ASSERT(dd->dd_phys);
473 
474 	if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
475 		/* up the hold count until we can be written out */
476 		dmu_buf_add_ref(dd->dd_dbuf, dd);
477 	}
478 }
479 
480 static int64_t
481 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
482 {
483 	uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
484 	uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
485 	return (new_accounted - old_accounted);
486 }
487 
488 void
489 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
490 {
491 	ASSERT(dmu_tx_is_syncing(tx));
492 
493 	mutex_enter(&dd->dd_lock);
494 	ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
495 	dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
496 	    dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
497 	dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
498 	mutex_exit(&dd->dd_lock);
499 
500 	/* release the hold from dsl_dir_dirty */
501 	dmu_buf_rele(dd->dd_dbuf, dd);
502 }
503 
504 static uint64_t
505 dsl_dir_space_towrite(dsl_dir_t *dd)
506 {
507 	uint64_t space = 0;
508 	int i;
509 
510 	ASSERT(MUTEX_HELD(&dd->dd_lock));
511 
512 	for (i = 0; i < TXG_SIZE; i++) {
513 		space += dd->dd_space_towrite[i&TXG_MASK];
514 		ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
515 	}
516 	return (space);
517 }
518 
519 /*
520  * How much space would dd have available if ancestor had delta applied
521  * to it?  If ondiskonly is set, we're only interested in what's
522  * on-disk, not estimated pending changes.
523  */
524 uint64_t
525 dsl_dir_space_available(dsl_dir_t *dd,
526     dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
527 {
528 	uint64_t parentspace, myspace, quota, used;
529 
530 	/*
531 	 * If there are no restrictions otherwise, assume we have
532 	 * unlimited space available.
533 	 */
534 	quota = UINT64_MAX;
535 	parentspace = UINT64_MAX;
536 
537 	if (dd->dd_parent != NULL) {
538 		parentspace = dsl_dir_space_available(dd->dd_parent,
539 		    ancestor, delta, ondiskonly);
540 	}
541 
542 	mutex_enter(&dd->dd_lock);
543 	if (dd->dd_phys->dd_quota != 0)
544 		quota = dd->dd_phys->dd_quota;
545 	used = dd->dd_phys->dd_used_bytes;
546 	if (!ondiskonly)
547 		used += dsl_dir_space_towrite(dd);
548 
549 	if (dd->dd_parent == NULL) {
550 		uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
551 		quota = MIN(quota, poolsize);
552 	}
553 
554 	if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
555 		/*
556 		 * We have some space reserved, in addition to what our
557 		 * parent gave us.
558 		 */
559 		parentspace += dd->dd_phys->dd_reserved - used;
560 	}
561 
562 	if (dd == ancestor) {
563 		ASSERT(delta <= 0);
564 		ASSERT(used >= -delta);
565 		used += delta;
566 		if (parentspace != UINT64_MAX)
567 			parentspace -= delta;
568 	}
569 
570 	if (used > quota) {
571 		/* over quota */
572 		myspace = 0;
573 	} else {
574 		/*
575 		 * the lesser of the space provided by our parent and
576 		 * the space left in our quota
577 		 */
578 		myspace = MIN(parentspace, quota - used);
579 	}
580 
581 	mutex_exit(&dd->dd_lock);
582 
583 	return (myspace);
584 }
585 
586 struct tempreserve {
587 	list_node_t tr_node;
588 	dsl_dir_t *tr_ds;
589 	uint64_t tr_size;
590 };
591 
592 static int
593 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
594     boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
595     dmu_tx_t *tx, boolean_t first)
596 {
597 	uint64_t txg = tx->tx_txg;
598 	uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
599 	uint64_t deferred = 0;
600 	struct tempreserve *tr;
601 	int retval = EDQUOT;
602 	int txgidx = txg & TXG_MASK;
603 	int i;
604 	uint64_t ref_rsrv = 0;
605 
606 	ASSERT3U(txg, !=, 0);
607 	ASSERT3S(asize, >, 0);
608 
609 	mutex_enter(&dd->dd_lock);
610 
611 	/*
612 	 * Check against the dsl_dir's quota.  We don't add in the delta
613 	 * when checking for over-quota because they get one free hit.
614 	 */
615 	est_inflight = dsl_dir_space_towrite(dd);
616 	for (i = 0; i < TXG_SIZE; i++)
617 		est_inflight += dd->dd_tempreserved[i];
618 	used_on_disk = dd->dd_phys->dd_used_bytes;
619 
620 	/*
621 	 * On the first iteration, fetch the dataset's used-on-disk and
622 	 * refreservation values. Also, if checkrefquota is set, test if
623 	 * allocating this space would exceed the dataset's refquota.
624 	 */
625 	if (first && tx->tx_objset) {
626 		int error;
627 		dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
628 
629 		error = dsl_dataset_check_quota(ds, checkrefquota,
630 		    asize, est_inflight, &used_on_disk, &ref_rsrv);
631 		if (error) {
632 			mutex_exit(&dd->dd_lock);
633 			return (error);
634 		}
635 	}
636 
637 	/*
638 	 * If this transaction will result in a net free of space,
639 	 * we want to let it through.
640 	 */
641 	if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
642 		quota = UINT64_MAX;
643 	else
644 		quota = dd->dd_phys->dd_quota;
645 
646 	/*
647 	 * Adjust the quota against the actual pool size at the root
648 	 * minus any outstanding deferred frees.
649 	 * To ensure that it's possible to remove files from a full
650 	 * pool without inducing transient overcommits, we throttle
651 	 * netfree transactions against a quota that is slightly larger,
652 	 * but still within the pool's allocation slop.  In cases where
653 	 * we're very close to full, this will allow a steady trickle of
654 	 * removes to get through.
655 	 */
656 	if (dd->dd_parent == NULL) {
657 		spa_t *spa = dd->dd_pool->dp_spa;
658 		uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
659 		deferred = metaslab_class_get_deferred(spa_normal_class(spa));
660 		if (poolsize - deferred < quota) {
661 			quota = poolsize - deferred;
662 			retval = ENOSPC;
663 		}
664 	}
665 
666 	/*
667 	 * If they are requesting more space, and our current estimate
668 	 * is over quota, they get to try again unless the actual
669 	 * on-disk is over quota and there are no pending changes (which
670 	 * may free up space for us).
671 	 */
672 	if (used_on_disk + est_inflight >= quota) {
673 		if (est_inflight > 0 || used_on_disk < quota ||
674 		    (retval == ENOSPC && used_on_disk < quota + deferred))
675 			retval = ERESTART;
676 		dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
677 		    "quota=%lluK tr=%lluK err=%d\n",
678 		    used_on_disk>>10, est_inflight>>10,
679 		    quota>>10, asize>>10, retval);
680 		mutex_exit(&dd->dd_lock);
681 		return (SET_ERROR(retval));
682 	}
683 
684 	/* We need to up our estimated delta before dropping dd_lock */
685 	dd->dd_tempreserved[txgidx] += asize;
686 
687 	parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
688 	    asize - ref_rsrv);
689 	mutex_exit(&dd->dd_lock);
690 
691 	tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
692 	tr->tr_ds = dd;
693 	tr->tr_size = asize;
694 	list_insert_tail(tr_list, tr);
695 
696 	/* see if it's OK with our parent */
697 	if (dd->dd_parent && parent_rsrv) {
698 		boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
699 
700 		return (dsl_dir_tempreserve_impl(dd->dd_parent,
701 		    parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
702 	} else {
703 		return (0);
704 	}
705 }
706 
707 /*
708  * Reserve space in this dsl_dir, to be used in this tx's txg.
709  * After the space has been dirtied (and dsl_dir_willuse_space()
710  * has been called), the reservation should be canceled, using
711  * dsl_dir_tempreserve_clear().
712  */
713 int
714 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
715     uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
716 {
717 	int err;
718 	list_t *tr_list;
719 
720 	if (asize == 0) {
721 		*tr_cookiep = NULL;
722 		return (0);
723 	}
724 
725 	tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
726 	list_create(tr_list, sizeof (struct tempreserve),
727 	    offsetof(struct tempreserve, tr_node));
728 	ASSERT3S(asize, >, 0);
729 	ASSERT3S(fsize, >=, 0);
730 
731 	err = arc_tempreserve_space(lsize, tx->tx_txg);
732 	if (err == 0) {
733 		struct tempreserve *tr;
734 
735 		tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
736 		tr->tr_size = lsize;
737 		list_insert_tail(tr_list, tr);
738 	} else {
739 		if (err == EAGAIN) {
740 			/*
741 			 * If arc_memory_throttle() detected that pageout
742 			 * is running and we are low on memory, we delay new
743 			 * non-pageout transactions to give pageout an
744 			 * advantage.
745 			 *
746 			 * It is unfortunate to be delaying while the caller's
747 			 * locks are held.
748 			 */
749 			txg_delay(dd->dd_pool, tx->tx_txg,
750 			    MSEC2NSEC(10), MSEC2NSEC(10));
751 			err = SET_ERROR(ERESTART);
752 		}
753 	}
754 
755 	if (err == 0) {
756 		err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
757 		    FALSE, asize > usize, tr_list, tx, TRUE);
758 	}
759 
760 	if (err != 0)
761 		dsl_dir_tempreserve_clear(tr_list, tx);
762 	else
763 		*tr_cookiep = tr_list;
764 
765 	return (err);
766 }
767 
768 /*
769  * Clear a temporary reservation that we previously made with
770  * dsl_dir_tempreserve_space().
771  */
772 void
773 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
774 {
775 	int txgidx = tx->tx_txg & TXG_MASK;
776 	list_t *tr_list = tr_cookie;
777 	struct tempreserve *tr;
778 
779 	ASSERT3U(tx->tx_txg, !=, 0);
780 
781 	if (tr_cookie == NULL)
782 		return;
783 
784 	while ((tr = list_head(tr_list)) != NULL) {
785 		if (tr->tr_ds) {
786 			mutex_enter(&tr->tr_ds->dd_lock);
787 			ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
788 			    tr->tr_size);
789 			tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
790 			mutex_exit(&tr->tr_ds->dd_lock);
791 		} else {
792 			arc_tempreserve_clear(tr->tr_size);
793 		}
794 		list_remove(tr_list, tr);
795 		kmem_free(tr, sizeof (struct tempreserve));
796 	}
797 
798 	kmem_free(tr_list, sizeof (list_t));
799 }
800 
801 /*
802  * This should be called from open context when we think we're going to write
803  * or free space, for example when dirtying data. Be conservative; it's okay
804  * to write less space or free more, but we don't want to write more or free
805  * less than the amount specified.
806  */
807 void
808 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
809 {
810 	int64_t parent_space;
811 	uint64_t est_used;
812 
813 	mutex_enter(&dd->dd_lock);
814 	if (space > 0)
815 		dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
816 
817 	est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
818 	parent_space = parent_delta(dd, est_used, space);
819 	mutex_exit(&dd->dd_lock);
820 
821 	/* Make sure that we clean up dd_space_to* */
822 	dsl_dir_dirty(dd, tx);
823 
824 	/* XXX this is potentially expensive and unnecessary... */
825 	if (parent_space && dd->dd_parent)
826 		dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
827 }
828 
829 /* call from syncing context when we actually write/free space for this dd */
830 void
831 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
832     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
833 {
834 	int64_t accounted_delta;
835 
836 	/*
837 	 * dsl_dataset_set_refreservation_sync_impl() calls this with
838 	 * dd_lock held, so that it can atomically update
839 	 * ds->ds_reserved and the dsl_dir accounting, so that
840 	 * dsl_dataset_check_quota() can see dataset and dir accounting
841 	 * consistently.
842 	 */
843 	boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
844 
845 	ASSERT(dmu_tx_is_syncing(tx));
846 	ASSERT(type < DD_USED_NUM);
847 
848 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
849 
850 	if (needlock)
851 		mutex_enter(&dd->dd_lock);
852 	accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
853 	ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
854 	ASSERT(compressed >= 0 ||
855 	    dd->dd_phys->dd_compressed_bytes >= -compressed);
856 	ASSERT(uncompressed >= 0 ||
857 	    dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
858 	dd->dd_phys->dd_used_bytes += used;
859 	dd->dd_phys->dd_uncompressed_bytes += uncompressed;
860 	dd->dd_phys->dd_compressed_bytes += compressed;
861 
862 	if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
863 		ASSERT(used > 0 ||
864 		    dd->dd_phys->dd_used_breakdown[type] >= -used);
865 		dd->dd_phys->dd_used_breakdown[type] += used;
866 #ifdef DEBUG
867 		dd_used_t t;
868 		uint64_t u = 0;
869 		for (t = 0; t < DD_USED_NUM; t++)
870 			u += dd->dd_phys->dd_used_breakdown[t];
871 		ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
872 #endif
873 	}
874 	if (needlock)
875 		mutex_exit(&dd->dd_lock);
876 
877 	if (dd->dd_parent != NULL) {
878 		dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
879 		    accounted_delta, compressed, uncompressed, tx);
880 		dsl_dir_transfer_space(dd->dd_parent,
881 		    used - accounted_delta,
882 		    DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
883 	}
884 }
885 
886 void
887 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
888     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
889 {
890 	ASSERT(dmu_tx_is_syncing(tx));
891 	ASSERT(oldtype < DD_USED_NUM);
892 	ASSERT(newtype < DD_USED_NUM);
893 
894 	if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
895 		return;
896 
897 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
898 	mutex_enter(&dd->dd_lock);
899 	ASSERT(delta > 0 ?
900 	    dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
901 	    dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
902 	ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
903 	dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
904 	dd->dd_phys->dd_used_breakdown[newtype] += delta;
905 	mutex_exit(&dd->dd_lock);
906 }
907 
908 typedef struct dsl_dir_set_qr_arg {
909 	const char *ddsqra_name;
910 	zprop_source_t ddsqra_source;
911 	uint64_t ddsqra_value;
912 } dsl_dir_set_qr_arg_t;
913 
914 static int
915 dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
916 {
917 	dsl_dir_set_qr_arg_t *ddsqra = arg;
918 	dsl_pool_t *dp = dmu_tx_pool(tx);
919 	dsl_dataset_t *ds;
920 	int error;
921 	uint64_t towrite, newval;
922 
923 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
924 	if (error != 0)
925 		return (error);
926 
927 	error = dsl_prop_predict(ds->ds_dir, "quota",
928 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
929 	if (error != 0) {
930 		dsl_dataset_rele(ds, FTAG);
931 		return (error);
932 	}
933 
934 	if (newval == 0) {
935 		dsl_dataset_rele(ds, FTAG);
936 		return (0);
937 	}
938 
939 	mutex_enter(&ds->ds_dir->dd_lock);
940 	/*
941 	 * If we are doing the preliminary check in open context, and
942 	 * there are pending changes, then don't fail it, since the
943 	 * pending changes could under-estimate the amount of space to be
944 	 * freed up.
945 	 */
946 	towrite = dsl_dir_space_towrite(ds->ds_dir);
947 	if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
948 	    (newval < ds->ds_dir->dd_phys->dd_reserved ||
949 	    newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
950 		error = SET_ERROR(ENOSPC);
951 	}
952 	mutex_exit(&ds->ds_dir->dd_lock);
953 	dsl_dataset_rele(ds, FTAG);
954 	return (error);
955 }
956 
957 static void
958 dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
959 {
960 	dsl_dir_set_qr_arg_t *ddsqra = arg;
961 	dsl_pool_t *dp = dmu_tx_pool(tx);
962 	dsl_dataset_t *ds;
963 	uint64_t newval;
964 
965 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
966 
967 	if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
968 		dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
969 		    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
970 		    &ddsqra->ddsqra_value, tx);
971 
972 		VERIFY0(dsl_prop_get_int_ds(ds,
973 		    zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
974 	} else {
975 		newval = ddsqra->ddsqra_value;
976 		spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
977 		    zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
978 	}
979 
980 	dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
981 	mutex_enter(&ds->ds_dir->dd_lock);
982 	ds->ds_dir->dd_phys->dd_quota = newval;
983 	mutex_exit(&ds->ds_dir->dd_lock);
984 	dsl_dataset_rele(ds, FTAG);
985 }
986 
987 int
988 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
989 {
990 	dsl_dir_set_qr_arg_t ddsqra;
991 
992 	ddsqra.ddsqra_name = ddname;
993 	ddsqra.ddsqra_source = source;
994 	ddsqra.ddsqra_value = quota;
995 
996 	return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
997 	    dsl_dir_set_quota_sync, &ddsqra, 0));
998 }
999 
1000 int
1001 dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1002 {
1003 	dsl_dir_set_qr_arg_t *ddsqra = arg;
1004 	dsl_pool_t *dp = dmu_tx_pool(tx);
1005 	dsl_dataset_t *ds;
1006 	dsl_dir_t *dd;
1007 	uint64_t newval, used, avail;
1008 	int error;
1009 
1010 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1011 	if (error != 0)
1012 		return (error);
1013 	dd = ds->ds_dir;
1014 
1015 	/*
1016 	 * If we are doing the preliminary check in open context, the
1017 	 * space estimates may be inaccurate.
1018 	 */
1019 	if (!dmu_tx_is_syncing(tx)) {
1020 		dsl_dataset_rele(ds, FTAG);
1021 		return (0);
1022 	}
1023 
1024 	error = dsl_prop_predict(ds->ds_dir,
1025 	    zfs_prop_to_name(ZFS_PROP_RESERVATION),
1026 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1027 	if (error != 0) {
1028 		dsl_dataset_rele(ds, FTAG);
1029 		return (error);
1030 	}
1031 
1032 	mutex_enter(&dd->dd_lock);
1033 	used = dd->dd_phys->dd_used_bytes;
1034 	mutex_exit(&dd->dd_lock);
1035 
1036 	if (dd->dd_parent) {
1037 		avail = dsl_dir_space_available(dd->dd_parent,
1038 		    NULL, 0, FALSE);
1039 	} else {
1040 		avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1041 	}
1042 
1043 	if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1044 		uint64_t delta = MAX(used, newval) -
1045 		    MAX(used, dd->dd_phys->dd_reserved);
1046 
1047 		if (delta > avail ||
1048 		    (dd->dd_phys->dd_quota > 0 &&
1049 		    newval > dd->dd_phys->dd_quota))
1050 			error = SET_ERROR(ENOSPC);
1051 	}
1052 
1053 	dsl_dataset_rele(ds, FTAG);
1054 	return (error);
1055 }
1056 
1057 void
1058 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1059 {
1060 	uint64_t used;
1061 	int64_t delta;
1062 
1063 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1064 
1065 	mutex_enter(&dd->dd_lock);
1066 	used = dd->dd_phys->dd_used_bytes;
1067 	delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1068 	dd->dd_phys->dd_reserved = value;
1069 
1070 	if (dd->dd_parent != NULL) {
1071 		/* Roll up this additional usage into our ancestors */
1072 		dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1073 		    delta, 0, 0, tx);
1074 	}
1075 	mutex_exit(&dd->dd_lock);
1076 }
1077 
1078 
1079 static void
1080 dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1081 {
1082 	dsl_dir_set_qr_arg_t *ddsqra = arg;
1083 	dsl_pool_t *dp = dmu_tx_pool(tx);
1084 	dsl_dataset_t *ds;
1085 	uint64_t newval;
1086 
1087 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1088 
1089 	if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1090 		dsl_prop_set_sync_impl(ds,
1091 		    zfs_prop_to_name(ZFS_PROP_RESERVATION),
1092 		    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1093 		    &ddsqra->ddsqra_value, tx);
1094 
1095 		VERIFY0(dsl_prop_get_int_ds(ds,
1096 		    zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1097 	} else {
1098 		newval = ddsqra->ddsqra_value;
1099 		spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1100 		    zfs_prop_to_name(ZFS_PROP_RESERVATION),
1101 		    (longlong_t)newval);
1102 	}
1103 
1104 	dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1105 	dsl_dataset_rele(ds, FTAG);
1106 }
1107 
1108 int
1109 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1110     uint64_t reservation)
1111 {
1112 	dsl_dir_set_qr_arg_t ddsqra;
1113 
1114 	ddsqra.ddsqra_name = ddname;
1115 	ddsqra.ddsqra_source = source;
1116 	ddsqra.ddsqra_value = reservation;
1117 
1118 	return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1119 	    dsl_dir_set_reservation_sync, &ddsqra, 0));
1120 }
1121 
1122 static dsl_dir_t *
1123 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1124 {
1125 	for (; ds1; ds1 = ds1->dd_parent) {
1126 		dsl_dir_t *dd;
1127 		for (dd = ds2; dd; dd = dd->dd_parent) {
1128 			if (ds1 == dd)
1129 				return (dd);
1130 		}
1131 	}
1132 	return (NULL);
1133 }
1134 
1135 /*
1136  * If delta is applied to dd, how much of that delta would be applied to
1137  * ancestor?  Syncing context only.
1138  */
1139 static int64_t
1140 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1141 {
1142 	if (dd == ancestor)
1143 		return (delta);
1144 
1145 	mutex_enter(&dd->dd_lock);
1146 	delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1147 	mutex_exit(&dd->dd_lock);
1148 	return (would_change(dd->dd_parent, delta, ancestor));
1149 }
1150 
1151 typedef struct dsl_dir_rename_arg {
1152 	const char *ddra_oldname;
1153 	const char *ddra_newname;
1154 } dsl_dir_rename_arg_t;
1155 
1156 /* ARGSUSED */
1157 static int
1158 dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1159 {
1160 	int *deltap = arg;
1161 	char namebuf[MAXNAMELEN];
1162 
1163 	dsl_dataset_name(ds, namebuf);
1164 
1165 	if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1166 		return (SET_ERROR(ENAMETOOLONG));
1167 	return (0);
1168 }
1169 
1170 static int
1171 dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1172 {
1173 	dsl_dir_rename_arg_t *ddra = arg;
1174 	dsl_pool_t *dp = dmu_tx_pool(tx);
1175 	dsl_dir_t *dd, *newparent;
1176 	const char *mynewname;
1177 	int error;
1178 	int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1179 
1180 	/* target dir should exist */
1181 	error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1182 	if (error != 0)
1183 		return (error);
1184 
1185 	/* new parent should exist */
1186 	error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1187 	    &newparent, &mynewname);
1188 	if (error != 0) {
1189 		dsl_dir_rele(dd, FTAG);
1190 		return (error);
1191 	}
1192 
1193 	/* can't rename to different pool */
1194 	if (dd->dd_pool != newparent->dd_pool) {
1195 		dsl_dir_rele(newparent, FTAG);
1196 		dsl_dir_rele(dd, FTAG);
1197 		return (SET_ERROR(ENXIO));
1198 	}
1199 
1200 	/* new name should not already exist */
1201 	if (mynewname == NULL) {
1202 		dsl_dir_rele(newparent, FTAG);
1203 		dsl_dir_rele(dd, FTAG);
1204 		return (SET_ERROR(EEXIST));
1205 	}
1206 
1207 	/* if the name length is growing, validate child name lengths */
1208 	if (delta > 0) {
1209 		error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1210 		    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1211 		if (error != 0) {
1212 			dsl_dir_rele(newparent, FTAG);
1213 			dsl_dir_rele(dd, FTAG);
1214 			return (error);
1215 		}
1216 	}
1217 
1218 	if (newparent != dd->dd_parent) {
1219 		/* is there enough space? */
1220 		uint64_t myspace =
1221 		    MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1222 
1223 		/* no rename into our descendant */
1224 		if (closest_common_ancestor(dd, newparent) == dd) {
1225 			dsl_dir_rele(newparent, FTAG);
1226 			dsl_dir_rele(dd, FTAG);
1227 			return (SET_ERROR(EINVAL));
1228 		}
1229 
1230 		error = dsl_dir_transfer_possible(dd->dd_parent,
1231 		    newparent, myspace);
1232 		if (error != 0) {
1233 			dsl_dir_rele(newparent, FTAG);
1234 			dsl_dir_rele(dd, FTAG);
1235 			return (error);
1236 		}
1237 	}
1238 
1239 	dsl_dir_rele(newparent, FTAG);
1240 	dsl_dir_rele(dd, FTAG);
1241 	return (0);
1242 }
1243 
1244 static void
1245 dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1246 {
1247 	dsl_dir_rename_arg_t *ddra = arg;
1248 	dsl_pool_t *dp = dmu_tx_pool(tx);
1249 	dsl_dir_t *dd, *newparent;
1250 	const char *mynewname;
1251 	int error;
1252 	objset_t *mos = dp->dp_meta_objset;
1253 
1254 	VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1255 	VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1256 	    &mynewname));
1257 
1258 	/* Log this before we change the name. */
1259 	spa_history_log_internal_dd(dd, "rename", tx,
1260 	    "-> %s", ddra->ddra_newname);
1261 
1262 	if (newparent != dd->dd_parent) {
1263 		dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1264 		    -dd->dd_phys->dd_used_bytes,
1265 		    -dd->dd_phys->dd_compressed_bytes,
1266 		    -dd->dd_phys->dd_uncompressed_bytes, tx);
1267 		dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1268 		    dd->dd_phys->dd_used_bytes,
1269 		    dd->dd_phys->dd_compressed_bytes,
1270 		    dd->dd_phys->dd_uncompressed_bytes, tx);
1271 
1272 		if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1273 			uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1274 			    dd->dd_phys->dd_used_bytes;
1275 
1276 			dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1277 			    -unused_rsrv, 0, 0, tx);
1278 			dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1279 			    unused_rsrv, 0, 0, tx);
1280 		}
1281 	}
1282 
1283 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1284 
1285 	/* remove from old parent zapobj */
1286 	error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1287 	    dd->dd_myname, tx);
1288 	ASSERT0(error);
1289 
1290 	(void) strcpy(dd->dd_myname, mynewname);
1291 	dsl_dir_rele(dd->dd_parent, dd);
1292 	dd->dd_phys->dd_parent_obj = newparent->dd_object;
1293 	VERIFY0(dsl_dir_hold_obj(dp,
1294 	    newparent->dd_object, NULL, dd, &dd->dd_parent));
1295 
1296 	/* add to new parent zapobj */
1297 	VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1298 	    dd->dd_myname, 8, 1, &dd->dd_object, tx));
1299 
1300 	dsl_prop_notify_all(dd);
1301 
1302 	dsl_dir_rele(newparent, FTAG);
1303 	dsl_dir_rele(dd, FTAG);
1304 }
1305 
1306 int
1307 dsl_dir_rename(const char *oldname, const char *newname)
1308 {
1309 	dsl_dir_rename_arg_t ddra;
1310 
1311 	ddra.ddra_oldname = oldname;
1312 	ddra.ddra_newname = newname;
1313 
1314 	return (dsl_sync_task(oldname,
1315 	    dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1316 }
1317 
1318 int
1319 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1320 {
1321 	dsl_dir_t *ancestor;
1322 	int64_t adelta;
1323 	uint64_t avail;
1324 
1325 	ancestor = closest_common_ancestor(sdd, tdd);
1326 	adelta = would_change(sdd, -space, ancestor);
1327 	avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1328 	if (avail < space)
1329 		return (SET_ERROR(ENOSPC));
1330 
1331 	return (0);
1332 }
1333 
1334 timestruc_t
1335 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1336 {
1337 	timestruc_t t;
1338 
1339 	mutex_enter(&dd->dd_lock);
1340 	t = dd->dd_snap_cmtime;
1341 	mutex_exit(&dd->dd_lock);
1342 
1343 	return (t);
1344 }
1345 
1346 void
1347 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1348 {
1349 	timestruc_t t;
1350 
1351 	gethrestime(&t);
1352 	mutex_enter(&dd->dd_lock);
1353 	dd->dd_snap_cmtime = t;
1354 	mutex_exit(&dd->dd_lock);
1355 }
1356 
1357 void
1358 dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1359 {
1360 	objset_t *mos = dd->dd_pool->dp_meta_objset;
1361 	dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1362 }
1363