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) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Martin Matuska. All rights reserved.
25 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
28 */
29
30#include <sys/dmu.h>
31#include <sys/dmu_objset.h>
32#include <sys/dmu_tx.h>
33#include <sys/dsl_dataset.h>
34#include <sys/dsl_dir.h>
35#include <sys/dsl_prop.h>
36#include <sys/dsl_synctask.h>
37#include <sys/dsl_deleg.h>
38#include <sys/dmu_impl.h>
39#include <sys/spa.h>
40#include <sys/metaslab.h>
41#include <sys/zap.h>
42#include <sys/zio.h>
43#include <sys/arc.h>
44#include <sys/sunddi.h>
45#include <sys/zfeature.h>
46#include <sys/policy.h>
47#include <sys/zfs_znode.h>
48#include "zfs_namecheck.h"
49#include "zfs_prop.h"
50
51/*
52 * Filesystem and Snapshot Limits
53 * ------------------------------
54 *
55 * These limits are used to restrict the number of filesystems and/or snapshots
56 * that can be created at a given level in the tree or below. A typical
57 * use-case is with a delegated dataset where the administrator wants to ensure
58 * that a user within the zone is not creating too many additional filesystems
59 * or snapshots, even though they're not exceeding their space quota.
60 *
61 * The filesystem and snapshot counts are stored as extensible properties. This
62 * capability is controlled by a feature flag and must be enabled to be used.
63 * Once enabled, the feature is not active until the first limit is set. At
64 * that point, future operations to create/destroy filesystems or snapshots
65 * will validate and update the counts.
66 *
67 * Because the count properties will not exist before the feature is active,
68 * the counts are updated when a limit is first set on an uninitialized
69 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
70 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
71 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
72 * snapshot count properties on a node indicate uninitialized counts on that
73 * node.) When first setting a limit on an uninitialized node, the code starts
74 * at the filesystem with the new limit and descends into all sub-filesystems
75 * to add the count properties.
76 *
77 * In practice this is lightweight since a limit is typically set when the
78 * filesystem is created and thus has no children. Once valid, changing the
79 * limit value won't require a re-traversal since the counts are already valid.
80 * When recursively fixing the counts, if a node with a limit is encountered
81 * during the descent, the counts are known to be valid and there is no need to
82 * descend into that filesystem's children. The counts on filesystems above the
83 * one with the new limit will still be uninitialized, unless a limit is
84 * eventually set on one of those filesystems. The counts are always recursively
85 * updated when a limit is set on a dataset, unless there is already a limit.
86 * When a new limit value is set on a filesystem with an existing limit, it is
87 * possible for the new limit to be less than the current count at that level
88 * since a user who can change the limit is also allowed to exceed the limit.
89 *
90 * Once the feature is active, then whenever a filesystem or snapshot is
91 * created, the code recurses up the tree, validating the new count against the
92 * limit at each initialized level. In practice, most levels will not have a
93 * limit set. If there is a limit at any initialized level up the tree, the
94 * check must pass or the creation will fail. Likewise, when a filesystem or
95 * snapshot is destroyed, the counts are recursively adjusted all the way up
96 * the initizized nodes in the tree. Renaming a filesystem into different point
97 * in the tree will first validate, then update the counts on each branch up to
98 * the common ancestor. A receive will also validate the counts and then update
99 * them.
100 *
101 * An exception to the above behavior is that the limit is not enforced if the
102 * user has permission to modify the limit. This is primarily so that
103 * recursive snapshots in the global zone always work. We want to prevent a
104 * denial-of-service in which a lower level delegated dataset could max out its
105 * limit and thus block recursive snapshots from being taken in the global zone.
106 * Because of this, it is possible for the snapshot count to be over the limit
107 * and snapshots taken in the global zone could cause a lower level dataset to
108 * hit or exceed its limit. The administrator taking the global zone recursive
109 * snapshot should be aware of this side-effect and behave accordingly.
110 * For consistency, the filesystem limit is also not enforced if the user can
111 * modify the limit.
112 *
113 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
114 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
115 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
116 * dsl_dir_init_fs_ss_count().
117 *
118 * There is a special case when we receive a filesystem that already exists. In
119 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
120 * never update the filesystem counts for temporary clones.
121 *
122 * Likewise, we do not update the snapshot counts for temporary snapshots,
123 * such as those created by zfs diff.
124 */
125
126extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
127
128static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
129
130static void
131dsl_dir_evict(void *dbu)
132{
133	dsl_dir_t *dd = dbu;
134	dsl_pool_t *dp = dd->dd_pool;
135	int t;
136
137	dd->dd_dbuf = NULL;
138
139	for (t = 0; t < TXG_SIZE; t++) {
140		ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
141		ASSERT(dd->dd_tempreserved[t] == 0);
142		ASSERT(dd->dd_space_towrite[t] == 0);
143	}
144
145	if (dd->dd_parent)
146		dsl_dir_async_rele(dd->dd_parent, dd);
147
148	spa_async_close(dd->dd_pool->dp_spa, dd);
149
150	/*
151	 * The props callback list should have been cleaned up by
152	 * objset_evict().
153	 */
154	list_destroy(&dd->dd_prop_cbs);
155	mutex_destroy(&dd->dd_lock);
156	kmem_free(dd, sizeof (dsl_dir_t));
157}
158
159int
160dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
161    const char *tail, void *tag, dsl_dir_t **ddp)
162{
163	dmu_buf_t *dbuf;
164	dsl_dir_t *dd;
165	int err;
166
167	ASSERT(dsl_pool_config_held(dp));
168
169	err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
170	if (err != 0)
171		return (err);
172	dd = dmu_buf_get_user(dbuf);
173#ifdef ZFS_DEBUG
174	{
175		dmu_object_info_t doi;
176		dmu_object_info_from_db(dbuf, &doi);
177		ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
178		ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
179	}
180#endif
181	if (dd == NULL) {
182		dsl_dir_t *winner;
183
184		dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
185		dd->dd_object = ddobj;
186		dd->dd_dbuf = dbuf;
187		dd->dd_pool = dp;
188		mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
189
190		list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
191		    offsetof(dsl_prop_cb_record_t, cbr_node));
192
193		dsl_dir_snap_cmtime_update(dd);
194
195		if (dsl_dir_phys(dd)->dd_parent_obj) {
196			err = dsl_dir_hold_obj(dp,
197			    dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
198			    &dd->dd_parent);
199			if (err != 0)
200				goto errout;
201			if (tail) {
202#ifdef ZFS_DEBUG
203				uint64_t foundobj;
204
205				err = zap_lookup(dp->dp_meta_objset,
206				    dsl_dir_phys(dd->dd_parent)->
207				    dd_child_dir_zapobj, tail,
208				    sizeof (foundobj), 1, &foundobj);
209				ASSERT(err || foundobj == ddobj);
210#endif
211				(void) strcpy(dd->dd_myname, tail);
212			} else {
213				err = zap_value_search(dp->dp_meta_objset,
214				    dsl_dir_phys(dd->dd_parent)->
215				    dd_child_dir_zapobj,
216				    ddobj, 0, dd->dd_myname);
217			}
218			if (err != 0)
219				goto errout;
220		} else {
221			(void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
222		}
223
224		if (dsl_dir_is_clone(dd)) {
225			dmu_buf_t *origin_bonus;
226			dsl_dataset_phys_t *origin_phys;
227
228			/*
229			 * We can't open the origin dataset, because
230			 * that would require opening this dsl_dir.
231			 * Just look at its phys directly instead.
232			 */
233			err = dmu_bonus_hold(dp->dp_meta_objset,
234			    dsl_dir_phys(dd)->dd_origin_obj, FTAG,
235			    &origin_bonus);
236			if (err != 0)
237				goto errout;
238			origin_phys = origin_bonus->db_data;
239			dd->dd_origin_txg =
240			    origin_phys->ds_creation_txg;
241			dmu_buf_rele(origin_bonus, FTAG);
242		}
243
244		dmu_buf_init_user(&dd->dd_dbu, dsl_dir_evict, &dd->dd_dbuf);
245		winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
246		if (winner != NULL) {
247			if (dd->dd_parent)
248				dsl_dir_rele(dd->dd_parent, dd);
249			mutex_destroy(&dd->dd_lock);
250			kmem_free(dd, sizeof (dsl_dir_t));
251			dd = winner;
252		} else {
253			spa_open_ref(dp->dp_spa, dd);
254		}
255	}
256
257	/*
258	 * The dsl_dir_t has both open-to-close and instantiate-to-evict
259	 * holds on the spa.  We need the open-to-close holds because
260	 * otherwise the spa_refcnt wouldn't change when we open a
261	 * dir which the spa also has open, so we could incorrectly
262	 * think it was OK to unload/export/destroy the pool.  We need
263	 * the instantiate-to-evict hold because the dsl_dir_t has a
264	 * pointer to the dd_pool, which has a pointer to the spa_t.
265	 */
266	spa_open_ref(dp->dp_spa, tag);
267	ASSERT3P(dd->dd_pool, ==, dp);
268	ASSERT3U(dd->dd_object, ==, ddobj);
269	ASSERT3P(dd->dd_dbuf, ==, dbuf);
270	*ddp = dd;
271	return (0);
272
273errout:
274	if (dd->dd_parent)
275		dsl_dir_rele(dd->dd_parent, dd);
276	mutex_destroy(&dd->dd_lock);
277	kmem_free(dd, sizeof (dsl_dir_t));
278	dmu_buf_rele(dbuf, tag);
279	return (err);
280}
281
282void
283dsl_dir_rele(dsl_dir_t *dd, void *tag)
284{
285	dprintf_dd(dd, "%s\n", "");
286	spa_close(dd->dd_pool->dp_spa, tag);
287	dmu_buf_rele(dd->dd_dbuf, tag);
288}
289
290/*
291 * Remove a reference to the given dsl dir that is being asynchronously
292 * released.  Async releases occur from a taskq performing eviction of
293 * dsl datasets and dirs.  This process is identical to a normal release
294 * with the exception of using the async API for releasing the reference on
295 * the spa.
296 */
297void
298dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
299{
300	dprintf_dd(dd, "%s\n", "");
301	spa_async_close(dd->dd_pool->dp_spa, tag);
302	dmu_buf_rele(dd->dd_dbuf, tag);
303}
304
305/* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
306void
307dsl_dir_name(dsl_dir_t *dd, char *buf)
308{
309	if (dd->dd_parent) {
310		dsl_dir_name(dd->dd_parent, buf);
311		(void) strcat(buf, "/");
312	} else {
313		buf[0] = '\0';
314	}
315	if (!MUTEX_HELD(&dd->dd_lock)) {
316		/*
317		 * recursive mutex so that we can use
318		 * dprintf_dd() with dd_lock held
319		 */
320		mutex_enter(&dd->dd_lock);
321		(void) strcat(buf, dd->dd_myname);
322		mutex_exit(&dd->dd_lock);
323	} else {
324		(void) strcat(buf, dd->dd_myname);
325	}
326}
327
328/* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
329int
330dsl_dir_namelen(dsl_dir_t *dd)
331{
332	int result = 0;
333
334	if (dd->dd_parent) {
335		/* parent's name + 1 for the "/" */
336		result = dsl_dir_namelen(dd->dd_parent) + 1;
337	}
338
339	if (!MUTEX_HELD(&dd->dd_lock)) {
340		/* see dsl_dir_name */
341		mutex_enter(&dd->dd_lock);
342		result += strlen(dd->dd_myname);
343		mutex_exit(&dd->dd_lock);
344	} else {
345		result += strlen(dd->dd_myname);
346	}
347
348	return (result);
349}
350
351static int
352getcomponent(const char *path, char *component, const char **nextp)
353{
354	char *p;
355
356	if ((path == NULL) || (path[0] == '\0'))
357		return (SET_ERROR(ENOENT));
358	/* This would be a good place to reserve some namespace... */
359	p = strpbrk(path, "/@");
360	if (p && (p[1] == '/' || p[1] == '@')) {
361		/* two separators in a row */
362		return (SET_ERROR(EINVAL));
363	}
364	if (p == NULL || p == path) {
365		/*
366		 * if the first thing is an @ or /, it had better be an
367		 * @ and it had better not have any more ats or slashes,
368		 * and it had better have something after the @.
369		 */
370		if (p != NULL &&
371		    (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
372			return (SET_ERROR(EINVAL));
373		if (strlen(path) >= MAXNAMELEN)
374			return (SET_ERROR(ENAMETOOLONG));
375		(void) strcpy(component, path);
376		p = NULL;
377	} else if (p[0] == '/') {
378		if (p - path >= MAXNAMELEN)
379			return (SET_ERROR(ENAMETOOLONG));
380		(void) strncpy(component, path, p - path);
381		component[p - path] = '\0';
382		p++;
383	} else if (p[0] == '@') {
384		/*
385		 * if the next separator is an @, there better not be
386		 * any more slashes.
387		 */
388		if (strchr(path, '/'))
389			return (SET_ERROR(EINVAL));
390		if (p - path >= MAXNAMELEN)
391			return (SET_ERROR(ENAMETOOLONG));
392		(void) strncpy(component, path, p - path);
393		component[p - path] = '\0';
394	} else {
395		panic("invalid p=%p", (void *)p);
396	}
397	*nextp = p;
398	return (0);
399}
400
401/*
402 * Return the dsl_dir_t, and possibly the last component which couldn't
403 * be found in *tail.  The name must be in the specified dsl_pool_t.  This
404 * thread must hold the dp_config_rwlock for the pool.  Returns NULL if the
405 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
406 * (*tail)[0] == '@' means that the last component is a snapshot.
407 */
408int
409dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
410    dsl_dir_t **ddp, const char **tailp)
411{
412	char buf[MAXNAMELEN];
413	const char *spaname, *next, *nextnext = NULL;
414	int err;
415	dsl_dir_t *dd;
416	uint64_t ddobj;
417
418	err = getcomponent(name, buf, &next);
419	if (err != 0)
420		return (err);
421
422	/* Make sure the name is in the specified pool. */
423	spaname = spa_name(dp->dp_spa);
424	if (strcmp(buf, spaname) != 0)
425		return (SET_ERROR(EXDEV));
426
427	ASSERT(dsl_pool_config_held(dp));
428
429	err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
430	if (err != 0) {
431		return (err);
432	}
433
434	while (next != NULL) {
435		dsl_dir_t *child_dd;
436		err = getcomponent(next, buf, &nextnext);
437		if (err != 0)
438			break;
439		ASSERT(next[0] != '\0');
440		if (next[0] == '@')
441			break;
442		dprintf("looking up %s in obj%lld\n",
443		    buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
444
445		err = zap_lookup(dp->dp_meta_objset,
446		    dsl_dir_phys(dd)->dd_child_dir_zapobj,
447		    buf, sizeof (ddobj), 1, &ddobj);
448		if (err != 0) {
449			if (err == ENOENT)
450				err = 0;
451			break;
452		}
453
454		err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
455		if (err != 0)
456			break;
457		dsl_dir_rele(dd, tag);
458		dd = child_dd;
459		next = nextnext;
460	}
461
462	if (err != 0) {
463		dsl_dir_rele(dd, tag);
464		return (err);
465	}
466
467	/*
468	 * It's an error if there's more than one component left, or
469	 * tailp==NULL and there's any component left.
470	 */
471	if (next != NULL &&
472	    (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
473		/* bad path name */
474		dsl_dir_rele(dd, tag);
475		dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
476		err = SET_ERROR(ENOENT);
477	}
478	if (tailp != NULL)
479		*tailp = next;
480	*ddp = dd;
481	return (err);
482}
483
484/*
485 * If the counts are already initialized for this filesystem and its
486 * descendants then do nothing, otherwise initialize the counts.
487 *
488 * The counts on this filesystem, and those below, may be uninitialized due to
489 * either the use of a pre-existing pool which did not support the
490 * filesystem/snapshot limit feature, or one in which the feature had not yet
491 * been enabled.
492 *
493 * Recursively descend the filesystem tree and update the filesystem/snapshot
494 * counts on each filesystem below, then update the cumulative count on the
495 * current filesystem. If the filesystem already has a count set on it,
496 * then we know that its counts, and the counts on the filesystems below it,
497 * are already correct, so we don't have to update this filesystem.
498 */
499static void
500dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
501{
502	uint64_t my_fs_cnt = 0;
503	uint64_t my_ss_cnt = 0;
504	dsl_pool_t *dp = dd->dd_pool;
505	objset_t *os = dp->dp_meta_objset;
506	zap_cursor_t *zc;
507	zap_attribute_t *za;
508	dsl_dataset_t *ds;
509
510	ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
511	ASSERT(dsl_pool_config_held(dp));
512	ASSERT(dmu_tx_is_syncing(tx));
513
514	dsl_dir_zapify(dd, tx);
515
516	/*
517	 * If the filesystem count has already been initialized then we
518	 * don't need to recurse down any further.
519	 */
520	if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
521		return;
522
523	zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
524	za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
525
526	/* Iterate my child dirs */
527	for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
528	    zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
529		dsl_dir_t *chld_dd;
530		uint64_t count;
531
532		VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
533		    &chld_dd));
534
535		/*
536		 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
537		 * temporary datasets.
538		 */
539		if (chld_dd->dd_myname[0] == '$' ||
540		    chld_dd->dd_myname[0] == '%') {
541			dsl_dir_rele(chld_dd, FTAG);
542			continue;
543		}
544
545		my_fs_cnt++;	/* count this child */
546
547		dsl_dir_init_fs_ss_count(chld_dd, tx);
548
549		VERIFY0(zap_lookup(os, chld_dd->dd_object,
550		    DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
551		my_fs_cnt += count;
552		VERIFY0(zap_lookup(os, chld_dd->dd_object,
553		    DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
554		my_ss_cnt += count;
555
556		dsl_dir_rele(chld_dd, FTAG);
557	}
558	zap_cursor_fini(zc);
559	/* Count my snapshots (we counted children's snapshots above) */
560	VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
561	    dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
562
563	for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
564	    zap_cursor_retrieve(zc, za) == 0;
565	    zap_cursor_advance(zc)) {
566		/* Don't count temporary snapshots */
567		if (za->za_name[0] != '%')
568			my_ss_cnt++;
569	}
570	zap_cursor_fini(zc);
571
572	dsl_dataset_rele(ds, FTAG);
573
574	kmem_free(zc, sizeof (zap_cursor_t));
575	kmem_free(za, sizeof (zap_attribute_t));
576
577	/* we're in a sync task, update counts */
578	dmu_buf_will_dirty(dd->dd_dbuf, tx);
579	VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
580	    sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
581	VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
582	    sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
583}
584
585static int
586dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
587{
588	char *ddname = (char *)arg;
589	dsl_pool_t *dp = dmu_tx_pool(tx);
590	dsl_dataset_t *ds;
591	dsl_dir_t *dd;
592	int error;
593
594	error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
595	if (error != 0)
596		return (error);
597
598	if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
599		dsl_dataset_rele(ds, FTAG);
600		return (SET_ERROR(ENOTSUP));
601	}
602
603	dd = ds->ds_dir;
604	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
605	    dsl_dir_is_zapified(dd) &&
606	    zap_contains(dp->dp_meta_objset, dd->dd_object,
607	    DD_FIELD_FILESYSTEM_COUNT) == 0) {
608		dsl_dataset_rele(ds, FTAG);
609		return (SET_ERROR(EALREADY));
610	}
611
612	dsl_dataset_rele(ds, FTAG);
613	return (0);
614}
615
616static void
617dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
618{
619	char *ddname = (char *)arg;
620	dsl_pool_t *dp = dmu_tx_pool(tx);
621	dsl_dataset_t *ds;
622	spa_t *spa;
623
624	VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
625
626	spa = dsl_dataset_get_spa(ds);
627
628	if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
629		/*
630		 * Since the feature was not active and we're now setting a
631		 * limit, increment the feature-active counter so that the
632		 * feature becomes active for the first time.
633		 *
634		 * We are already in a sync task so we can update the MOS.
635		 */
636		spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
637	}
638
639	/*
640	 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
641	 * we need to ensure the counts are correct. Descend down the tree from
642	 * this point and update all of the counts to be accurate.
643	 */
644	dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
645
646	dsl_dataset_rele(ds, FTAG);
647}
648
649/*
650 * Make sure the feature is enabled and activate it if necessary.
651 * Since we're setting a limit, ensure the on-disk counts are valid.
652 * This is only called by the ioctl path when setting a limit value.
653 *
654 * We do not need to validate the new limit, since users who can change the
655 * limit are also allowed to exceed the limit.
656 */
657int
658dsl_dir_activate_fs_ss_limit(const char *ddname)
659{
660	int error;
661
662	error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
663	    dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
664	    ZFS_SPACE_CHECK_RESERVED);
665
666	if (error == EALREADY)
667		error = 0;
668
669	return (error);
670}
671
672/*
673 * Used to determine if the filesystem_limit or snapshot_limit should be
674 * enforced. We allow the limit to be exceeded if the user has permission to
675 * write the property value. We pass in the creds that we got in the open
676 * context since we will always be the GZ root in syncing context. We also have
677 * to handle the case where we are allowed to change the limit on the current
678 * dataset, but there may be another limit in the tree above.
679 *
680 * We can never modify these two properties within a non-global zone. In
681 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
682 * can't use that function since we are already holding the dp_config_rwlock.
683 * In addition, we already have the dd and dealing with snapshots is simplified
684 * in this code.
685 */
686
687typedef enum {
688	ENFORCE_ALWAYS,
689	ENFORCE_NEVER,
690	ENFORCE_ABOVE
691} enforce_res_t;
692
693static enforce_res_t
694dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
695{
696	enforce_res_t enforce = ENFORCE_ALWAYS;
697	uint64_t obj;
698	dsl_dataset_t *ds;
699	uint64_t zoned;
700
701	ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
702	    prop == ZFS_PROP_SNAPSHOT_LIMIT);
703
704#ifdef _KERNEL
705	if (crgetzoneid(cr) != GLOBAL_ZONEID)
706		return (ENFORCE_ALWAYS);
707
708	if (secpolicy_zfs(cr) == 0)
709		return (ENFORCE_NEVER);
710#endif
711
712	if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
713		return (ENFORCE_ALWAYS);
714
715	ASSERT(dsl_pool_config_held(dd->dd_pool));
716
717	if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
718		return (ENFORCE_ALWAYS);
719
720	if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
721		/* Only root can access zoned fs's from the GZ */
722		enforce = ENFORCE_ALWAYS;
723	} else {
724		if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
725			enforce = ENFORCE_ABOVE;
726	}
727
728	dsl_dataset_rele(ds, FTAG);
729	return (enforce);
730}
731
732/*
733 * Check if adding additional child filesystem(s) would exceed any filesystem
734 * limits or adding additional snapshot(s) would exceed any snapshot limits.
735 * The prop argument indicates which limit to check.
736 *
737 * Note that all filesystem limits up to the root (or the highest
738 * initialized) filesystem or the given ancestor must be satisfied.
739 */
740int
741dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
742    dsl_dir_t *ancestor, cred_t *cr)
743{
744	objset_t *os = dd->dd_pool->dp_meta_objset;
745	uint64_t limit, count;
746	char *count_prop;
747	enforce_res_t enforce;
748	int err = 0;
749
750	ASSERT(dsl_pool_config_held(dd->dd_pool));
751	ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
752	    prop == ZFS_PROP_SNAPSHOT_LIMIT);
753
754	/*
755	 * If we're allowed to change the limit, don't enforce the limit
756	 * e.g. this can happen if a snapshot is taken by an administrative
757	 * user in the global zone (i.e. a recursive snapshot by root).
758	 * However, we must handle the case of delegated permissions where we
759	 * are allowed to change the limit on the current dataset, but there
760	 * is another limit in the tree above.
761	 */
762	enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
763	if (enforce == ENFORCE_NEVER)
764		return (0);
765
766	/*
767	 * e.g. if renaming a dataset with no snapshots, count adjustment
768	 * is 0.
769	 */
770	if (delta == 0)
771		return (0);
772
773	if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
774		/*
775		 * We don't enforce the limit for temporary snapshots. This is
776		 * indicated by a NULL cred_t argument.
777		 */
778		if (cr == NULL)
779			return (0);
780
781		count_prop = DD_FIELD_SNAPSHOT_COUNT;
782	} else {
783		count_prop = DD_FIELD_FILESYSTEM_COUNT;
784	}
785
786	/*
787	 * If an ancestor has been provided, stop checking the limit once we
788	 * hit that dir. We need this during rename so that we don't overcount
789	 * the check once we recurse up to the common ancestor.
790	 */
791	if (ancestor == dd)
792		return (0);
793
794	/*
795	 * If we hit an uninitialized node while recursing up the tree, we can
796	 * stop since we know there is no limit here (or above). The counts are
797	 * not valid on this node and we know we won't touch this node's counts.
798	 */
799	if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
800	    count_prop, sizeof (count), 1, &count) == ENOENT)
801		return (0);
802
803	err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
804	    B_FALSE);
805	if (err != 0)
806		return (err);
807
808	/* Is there a limit which we've hit? */
809	if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
810		return (SET_ERROR(EDQUOT));
811
812	if (dd->dd_parent != NULL)
813		err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
814		    ancestor, cr);
815
816	return (err);
817}
818
819/*
820 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
821 * parents. When a new filesystem/snapshot is created, increment the count on
822 * all parents, and when a filesystem/snapshot is destroyed, decrement the
823 * count.
824 */
825void
826dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
827    dmu_tx_t *tx)
828{
829	int err;
830	objset_t *os = dd->dd_pool->dp_meta_objset;
831	uint64_t count;
832
833	ASSERT(dsl_pool_config_held(dd->dd_pool));
834	ASSERT(dmu_tx_is_syncing(tx));
835	ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
836	    strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
837
838	/*
839	 * When we receive an incremental stream into a filesystem that already
840	 * exists, a temporary clone is created.  We don't count this temporary
841	 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
842	 * $MOS & $ORIGIN) objsets.
843	 */
844	if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
845	    strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
846		return;
847
848	/*
849	 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
850	 */
851	if (delta == 0)
852		return;
853
854	/*
855	 * If we hit an uninitialized node while recursing up the tree, we can
856	 * stop since we know the counts are not valid on this node and we
857	 * know we shouldn't touch this node's counts. An uninitialized count
858	 * on the node indicates that either the feature has not yet been
859	 * activated or there are no limits on this part of the tree.
860	 */
861	if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
862	    prop, sizeof (count), 1, &count)) == ENOENT)
863		return;
864	VERIFY0(err);
865
866	count += delta;
867	/* Use a signed verify to make sure we're not neg. */
868	VERIFY3S(count, >=, 0);
869
870	VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
871	    tx));
872
873	/* Roll up this additional count into our ancestors */
874	if (dd->dd_parent != NULL)
875		dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
876}
877
878uint64_t
879dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
880    dmu_tx_t *tx)
881{
882	objset_t *mos = dp->dp_meta_objset;
883	uint64_t ddobj;
884	dsl_dir_phys_t *ddphys;
885	dmu_buf_t *dbuf;
886
887	ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
888	    DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
889	if (pds) {
890		VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
891		    name, sizeof (uint64_t), 1, &ddobj, tx));
892	} else {
893		/* it's the root dir */
894		VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
895		    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
896	}
897	VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
898	dmu_buf_will_dirty(dbuf, tx);
899	ddphys = dbuf->db_data;
900
901	ddphys->dd_creation_time = gethrestime_sec();
902	if (pds) {
903		ddphys->dd_parent_obj = pds->dd_object;
904
905		/* update the filesystem counts */
906		dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
907	}
908	ddphys->dd_props_zapobj = zap_create(mos,
909	    DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
910	ddphys->dd_child_dir_zapobj = zap_create(mos,
911	    DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
912	if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
913		ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
914	dmu_buf_rele(dbuf, FTAG);
915
916	return (ddobj);
917}
918
919boolean_t
920dsl_dir_is_clone(dsl_dir_t *dd)
921{
922	return (dsl_dir_phys(dd)->dd_origin_obj &&
923	    (dd->dd_pool->dp_origin_snap == NULL ||
924	    dsl_dir_phys(dd)->dd_origin_obj !=
925	    dd->dd_pool->dp_origin_snap->ds_object));
926}
927
928void
929dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
930{
931	mutex_enter(&dd->dd_lock);
932	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
933	    dsl_dir_phys(dd)->dd_used_bytes);
934	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
935	    dsl_dir_phys(dd)->dd_quota);
936	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
937	    dsl_dir_phys(dd)->dd_reserved);
938	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
939	    dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
940	    (dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
941	    dsl_dir_phys(dd)->dd_compressed_bytes));
942	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
943	    dsl_dir_phys(dd)->dd_uncompressed_bytes);
944	if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
945		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
946		    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
947		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
948		    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
949		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
950		    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
951		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
952		    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
953		    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
954	}
955	mutex_exit(&dd->dd_lock);
956
957	if (dsl_dir_is_zapified(dd)) {
958		uint64_t count;
959		objset_t *os = dd->dd_pool->dp_meta_objset;
960
961		if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
962		    sizeof (count), 1, &count) == 0) {
963			dsl_prop_nvlist_add_uint64(nv,
964			    ZFS_PROP_FILESYSTEM_COUNT, count);
965		}
966		if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
967		    sizeof (count), 1, &count) == 0) {
968			dsl_prop_nvlist_add_uint64(nv,
969			    ZFS_PROP_SNAPSHOT_COUNT, count);
970		}
971	}
972
973	if (dsl_dir_is_clone(dd)) {
974		dsl_dataset_t *ds;
975		char buf[MAXNAMELEN];
976
977		VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
978		    dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
979		dsl_dataset_name(ds, buf);
980		dsl_dataset_rele(ds, FTAG);
981		dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
982	}
983}
984
985void
986dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
987{
988	dsl_pool_t *dp = dd->dd_pool;
989
990	ASSERT(dsl_dir_phys(dd));
991
992	if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
993		/* up the hold count until we can be written out */
994		dmu_buf_add_ref(dd->dd_dbuf, dd);
995	}
996}
997
998static int64_t
999parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1000{
1001	uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
1002	uint64_t new_accounted =
1003	    MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
1004	return (new_accounted - old_accounted);
1005}
1006
1007void
1008dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1009{
1010	ASSERT(dmu_tx_is_syncing(tx));
1011
1012	mutex_enter(&dd->dd_lock);
1013	ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1014	dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1015	    dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1016	dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1017	mutex_exit(&dd->dd_lock);
1018
1019	/* release the hold from dsl_dir_dirty */
1020	dmu_buf_rele(dd->dd_dbuf, dd);
1021}
1022
1023static uint64_t
1024dsl_dir_space_towrite(dsl_dir_t *dd)
1025{
1026	uint64_t space = 0;
1027	int i;
1028
1029	ASSERT(MUTEX_HELD(&dd->dd_lock));
1030
1031	for (i = 0; i < TXG_SIZE; i++) {
1032		space += dd->dd_space_towrite[i&TXG_MASK];
1033		ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1034	}
1035	return (space);
1036}
1037
1038/*
1039 * How much space would dd have available if ancestor had delta applied
1040 * to it?  If ondiskonly is set, we're only interested in what's
1041 * on-disk, not estimated pending changes.
1042 */
1043uint64_t
1044dsl_dir_space_available(dsl_dir_t *dd,
1045    dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1046{
1047	uint64_t parentspace, myspace, quota, used;
1048
1049	/*
1050	 * If there are no restrictions otherwise, assume we have
1051	 * unlimited space available.
1052	 */
1053	quota = UINT64_MAX;
1054	parentspace = UINT64_MAX;
1055
1056	if (dd->dd_parent != NULL) {
1057		parentspace = dsl_dir_space_available(dd->dd_parent,
1058		    ancestor, delta, ondiskonly);
1059	}
1060
1061	mutex_enter(&dd->dd_lock);
1062	if (dsl_dir_phys(dd)->dd_quota != 0)
1063		quota = dsl_dir_phys(dd)->dd_quota;
1064	used = dsl_dir_phys(dd)->dd_used_bytes;
1065	if (!ondiskonly)
1066		used += dsl_dir_space_towrite(dd);
1067
1068	if (dd->dd_parent == NULL) {
1069		uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1070		quota = MIN(quota, poolsize);
1071	}
1072
1073	if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
1074		/*
1075		 * We have some space reserved, in addition to what our
1076		 * parent gave us.
1077		 */
1078		parentspace += dsl_dir_phys(dd)->dd_reserved - used;
1079	}
1080
1081	if (dd == ancestor) {
1082		ASSERT(delta <= 0);
1083		ASSERT(used >= -delta);
1084		used += delta;
1085		if (parentspace != UINT64_MAX)
1086			parentspace -= delta;
1087	}
1088
1089	if (used > quota) {
1090		/* over quota */
1091		myspace = 0;
1092	} else {
1093		/*
1094		 * the lesser of the space provided by our parent and
1095		 * the space left in our quota
1096		 */
1097		myspace = MIN(parentspace, quota - used);
1098	}
1099
1100	mutex_exit(&dd->dd_lock);
1101
1102	return (myspace);
1103}
1104
1105struct tempreserve {
1106	list_node_t tr_node;
1107	dsl_dir_t *tr_ds;
1108	uint64_t tr_size;
1109};
1110
1111static int
1112dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1113    boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1114    dmu_tx_t *tx, boolean_t first)
1115{
1116	uint64_t txg = tx->tx_txg;
1117	uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1118	uint64_t deferred = 0;
1119	struct tempreserve *tr;
1120	int retval = EDQUOT;
1121	int txgidx = txg & TXG_MASK;
1122	int i;
1123	uint64_t ref_rsrv = 0;
1124
1125	ASSERT3U(txg, !=, 0);
1126	ASSERT3S(asize, >, 0);
1127
1128	mutex_enter(&dd->dd_lock);
1129
1130	/*
1131	 * Check against the dsl_dir's quota.  We don't add in the delta
1132	 * when checking for over-quota because they get one free hit.
1133	 */
1134	est_inflight = dsl_dir_space_towrite(dd);
1135	for (i = 0; i < TXG_SIZE; i++)
1136		est_inflight += dd->dd_tempreserved[i];
1137	used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
1138
1139	/*
1140	 * On the first iteration, fetch the dataset's used-on-disk and
1141	 * refreservation values. Also, if checkrefquota is set, test if
1142	 * allocating this space would exceed the dataset's refquota.
1143	 */
1144	if (first && tx->tx_objset) {
1145		int error;
1146		dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1147
1148		error = dsl_dataset_check_quota(ds, checkrefquota,
1149		    asize, est_inflight, &used_on_disk, &ref_rsrv);
1150		if (error) {
1151			mutex_exit(&dd->dd_lock);
1152			return (error);
1153		}
1154	}
1155
1156	/*
1157	 * If this transaction will result in a net free of space,
1158	 * we want to let it through.
1159	 */
1160	if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
1161		quota = UINT64_MAX;
1162	else
1163		quota = dsl_dir_phys(dd)->dd_quota;
1164
1165	/*
1166	 * Adjust the quota against the actual pool size at the root
1167	 * minus any outstanding deferred frees.
1168	 * To ensure that it's possible to remove files from a full
1169	 * pool without inducing transient overcommits, we throttle
1170	 * netfree transactions against a quota that is slightly larger,
1171	 * but still within the pool's allocation slop.  In cases where
1172	 * we're very close to full, this will allow a steady trickle of
1173	 * removes to get through.
1174	 */
1175	if (dd->dd_parent == NULL) {
1176		spa_t *spa = dd->dd_pool->dp_spa;
1177		uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1178		deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1179		if (poolsize - deferred < quota) {
1180			quota = poolsize - deferred;
1181			retval = ENOSPC;
1182		}
1183	}
1184
1185	/*
1186	 * If they are requesting more space, and our current estimate
1187	 * is over quota, they get to try again unless the actual
1188	 * on-disk is over quota and there are no pending changes (which
1189	 * may free up space for us).
1190	 */
1191	if (used_on_disk + est_inflight >= quota) {
1192		if (est_inflight > 0 || used_on_disk < quota ||
1193		    (retval == ENOSPC && used_on_disk < quota + deferred))
1194			retval = ERESTART;
1195		dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1196		    "quota=%lluK tr=%lluK err=%d\n",
1197		    used_on_disk>>10, est_inflight>>10,
1198		    quota>>10, asize>>10, retval);
1199		mutex_exit(&dd->dd_lock);
1200		return (SET_ERROR(retval));
1201	}
1202
1203	/* We need to up our estimated delta before dropping dd_lock */
1204	dd->dd_tempreserved[txgidx] += asize;
1205
1206	parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1207	    asize - ref_rsrv);
1208	mutex_exit(&dd->dd_lock);
1209
1210	tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1211	tr->tr_ds = dd;
1212	tr->tr_size = asize;
1213	list_insert_tail(tr_list, tr);
1214
1215	/* see if it's OK with our parent */
1216	if (dd->dd_parent && parent_rsrv) {
1217		boolean_t ismos = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1218
1219		return (dsl_dir_tempreserve_impl(dd->dd_parent,
1220		    parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1221	} else {
1222		return (0);
1223	}
1224}
1225
1226/*
1227 * Reserve space in this dsl_dir, to be used in this tx's txg.
1228 * After the space has been dirtied (and dsl_dir_willuse_space()
1229 * has been called), the reservation should be canceled, using
1230 * dsl_dir_tempreserve_clear().
1231 */
1232int
1233dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1234    uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1235{
1236	int err;
1237	list_t *tr_list;
1238
1239	if (asize == 0) {
1240		*tr_cookiep = NULL;
1241		return (0);
1242	}
1243
1244	tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1245	list_create(tr_list, sizeof (struct tempreserve),
1246	    offsetof(struct tempreserve, tr_node));
1247	ASSERT3S(asize, >, 0);
1248	ASSERT3S(fsize, >=, 0);
1249
1250	err = arc_tempreserve_space(lsize, tx->tx_txg);
1251	if (err == 0) {
1252		struct tempreserve *tr;
1253
1254		tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1255		tr->tr_size = lsize;
1256		list_insert_tail(tr_list, tr);
1257	} else {
1258		if (err == EAGAIN) {
1259			/*
1260			 * If arc_memory_throttle() detected that pageout
1261			 * is running and we are low on memory, we delay new
1262			 * non-pageout transactions to give pageout an
1263			 * advantage.
1264			 *
1265			 * It is unfortunate to be delaying while the caller's
1266			 * locks are held.
1267			 */
1268			txg_delay(dd->dd_pool, tx->tx_txg,
1269			    MSEC2NSEC(10), MSEC2NSEC(10));
1270			err = SET_ERROR(ERESTART);
1271		}
1272	}
1273
1274	if (err == 0) {
1275		err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1276		    FALSE, asize > usize, tr_list, tx, TRUE);
1277	}
1278
1279	if (err != 0)
1280		dsl_dir_tempreserve_clear(tr_list, tx);
1281	else
1282		*tr_cookiep = tr_list;
1283
1284	return (err);
1285}
1286
1287/*
1288 * Clear a temporary reservation that we previously made with
1289 * dsl_dir_tempreserve_space().
1290 */
1291void
1292dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1293{
1294	int txgidx = tx->tx_txg & TXG_MASK;
1295	list_t *tr_list = tr_cookie;
1296	struct tempreserve *tr;
1297
1298	ASSERT3U(tx->tx_txg, !=, 0);
1299
1300	if (tr_cookie == NULL)
1301		return;
1302
1303	while ((tr = list_head(tr_list)) != NULL) {
1304		if (tr->tr_ds) {
1305			mutex_enter(&tr->tr_ds->dd_lock);
1306			ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1307			    tr->tr_size);
1308			tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1309			mutex_exit(&tr->tr_ds->dd_lock);
1310		} else {
1311			arc_tempreserve_clear(tr->tr_size);
1312		}
1313		list_remove(tr_list, tr);
1314		kmem_free(tr, sizeof (struct tempreserve));
1315	}
1316
1317	kmem_free(tr_list, sizeof (list_t));
1318}
1319
1320/*
1321 * This should be called from open context when we think we're going to write
1322 * or free space, for example when dirtying data. Be conservative; it's okay
1323 * to write less space or free more, but we don't want to write more or free
1324 * less than the amount specified.
1325 */
1326void
1327dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1328{
1329	int64_t parent_space;
1330	uint64_t est_used;
1331
1332	mutex_enter(&dd->dd_lock);
1333	if (space > 0)
1334		dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1335
1336	est_used = dsl_dir_space_towrite(dd) + dsl_dir_phys(dd)->dd_used_bytes;
1337	parent_space = parent_delta(dd, est_used, space);
1338	mutex_exit(&dd->dd_lock);
1339
1340	/* Make sure that we clean up dd_space_to* */
1341	dsl_dir_dirty(dd, tx);
1342
1343	/* XXX this is potentially expensive and unnecessary... */
1344	if (parent_space && dd->dd_parent)
1345		dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1346}
1347
1348/* call from syncing context when we actually write/free space for this dd */
1349void
1350dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1351    int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1352{
1353	int64_t accounted_delta;
1354
1355	/*
1356	 * dsl_dataset_set_refreservation_sync_impl() calls this with
1357	 * dd_lock held, so that it can atomically update
1358	 * ds->ds_reserved and the dsl_dir accounting, so that
1359	 * dsl_dataset_check_quota() can see dataset and dir accounting
1360	 * consistently.
1361	 */
1362	boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1363
1364	ASSERT(dmu_tx_is_syncing(tx));
1365	ASSERT(type < DD_USED_NUM);
1366
1367	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1368
1369	if (needlock)
1370		mutex_enter(&dd->dd_lock);
1371	accounted_delta =
1372	    parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
1373	ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
1374	ASSERT(compressed >= 0 ||
1375	    dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
1376	ASSERT(uncompressed >= 0 ||
1377	    dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
1378	dsl_dir_phys(dd)->dd_used_bytes += used;
1379	dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
1380	dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
1381
1382	if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1383		ASSERT(used > 0 ||
1384		    dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
1385		dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
1386#ifdef DEBUG
1387		dd_used_t t;
1388		uint64_t u = 0;
1389		for (t = 0; t < DD_USED_NUM; t++)
1390			u += dsl_dir_phys(dd)->dd_used_breakdown[t];
1391		ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
1392#endif
1393	}
1394	if (needlock)
1395		mutex_exit(&dd->dd_lock);
1396
1397	if (dd->dd_parent != NULL) {
1398		dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1399		    accounted_delta, compressed, uncompressed, tx);
1400		dsl_dir_transfer_space(dd->dd_parent,
1401		    used - accounted_delta,
1402		    DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1403	}
1404}
1405
1406void
1407dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1408    dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1409{
1410	ASSERT(dmu_tx_is_syncing(tx));
1411	ASSERT(oldtype < DD_USED_NUM);
1412	ASSERT(newtype < DD_USED_NUM);
1413
1414	if (delta == 0 ||
1415	    !(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
1416		return;
1417
1418	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1419	mutex_enter(&dd->dd_lock);
1420	ASSERT(delta > 0 ?
1421	    dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
1422	    dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
1423	ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
1424	dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
1425	dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
1426	mutex_exit(&dd->dd_lock);
1427}
1428
1429typedef struct dsl_dir_set_qr_arg {
1430	const char *ddsqra_name;
1431	zprop_source_t ddsqra_source;
1432	uint64_t ddsqra_value;
1433} dsl_dir_set_qr_arg_t;
1434
1435static int
1436dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1437{
1438	dsl_dir_set_qr_arg_t *ddsqra = arg;
1439	dsl_pool_t *dp = dmu_tx_pool(tx);
1440	dsl_dataset_t *ds;
1441	int error;
1442	uint64_t towrite, newval;
1443
1444	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1445	if (error != 0)
1446		return (error);
1447
1448	error = dsl_prop_predict(ds->ds_dir, "quota",
1449	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1450	if (error != 0) {
1451		dsl_dataset_rele(ds, FTAG);
1452		return (error);
1453	}
1454
1455	if (newval == 0) {
1456		dsl_dataset_rele(ds, FTAG);
1457		return (0);
1458	}
1459
1460	mutex_enter(&ds->ds_dir->dd_lock);
1461	/*
1462	 * If we are doing the preliminary check in open context, and
1463	 * there are pending changes, then don't fail it, since the
1464	 * pending changes could under-estimate the amount of space to be
1465	 * freed up.
1466	 */
1467	towrite = dsl_dir_space_towrite(ds->ds_dir);
1468	if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1469	    (newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
1470	    newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
1471		error = SET_ERROR(ENOSPC);
1472	}
1473	mutex_exit(&ds->ds_dir->dd_lock);
1474	dsl_dataset_rele(ds, FTAG);
1475	return (error);
1476}
1477
1478static void
1479dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1480{
1481	dsl_dir_set_qr_arg_t *ddsqra = arg;
1482	dsl_pool_t *dp = dmu_tx_pool(tx);
1483	dsl_dataset_t *ds;
1484	uint64_t newval;
1485
1486	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1487
1488	if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1489		dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1490		    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1491		    &ddsqra->ddsqra_value, tx);
1492
1493		VERIFY0(dsl_prop_get_int_ds(ds,
1494		    zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1495	} else {
1496		newval = ddsqra->ddsqra_value;
1497		spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1498		    zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1499	}
1500
1501	dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1502	mutex_enter(&ds->ds_dir->dd_lock);
1503	dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
1504	mutex_exit(&ds->ds_dir->dd_lock);
1505	dsl_dataset_rele(ds, FTAG);
1506}
1507
1508int
1509dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1510{
1511	dsl_dir_set_qr_arg_t ddsqra;
1512
1513	ddsqra.ddsqra_name = ddname;
1514	ddsqra.ddsqra_source = source;
1515	ddsqra.ddsqra_value = quota;
1516
1517	return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1518	    dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1519}
1520
1521int
1522dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1523{
1524	dsl_dir_set_qr_arg_t *ddsqra = arg;
1525	dsl_pool_t *dp = dmu_tx_pool(tx);
1526	dsl_dataset_t *ds;
1527	dsl_dir_t *dd;
1528	uint64_t newval, used, avail;
1529	int error;
1530
1531	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1532	if (error != 0)
1533		return (error);
1534	dd = ds->ds_dir;
1535
1536	/*
1537	 * If we are doing the preliminary check in open context, the
1538	 * space estimates may be inaccurate.
1539	 */
1540	if (!dmu_tx_is_syncing(tx)) {
1541		dsl_dataset_rele(ds, FTAG);
1542		return (0);
1543	}
1544
1545	error = dsl_prop_predict(ds->ds_dir,
1546	    zfs_prop_to_name(ZFS_PROP_RESERVATION),
1547	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1548	if (error != 0) {
1549		dsl_dataset_rele(ds, FTAG);
1550		return (error);
1551	}
1552
1553	mutex_enter(&dd->dd_lock);
1554	used = dsl_dir_phys(dd)->dd_used_bytes;
1555	mutex_exit(&dd->dd_lock);
1556
1557	if (dd->dd_parent) {
1558		avail = dsl_dir_space_available(dd->dd_parent,
1559		    NULL, 0, FALSE);
1560	} else {
1561		avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1562	}
1563
1564	if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
1565		uint64_t delta = MAX(used, newval) -
1566		    MAX(used, dsl_dir_phys(dd)->dd_reserved);
1567
1568		if (delta > avail ||
1569		    (dsl_dir_phys(dd)->dd_quota > 0 &&
1570		    newval > dsl_dir_phys(dd)->dd_quota))
1571			error = SET_ERROR(ENOSPC);
1572	}
1573
1574	dsl_dataset_rele(ds, FTAG);
1575	return (error);
1576}
1577
1578void
1579dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1580{
1581	uint64_t used;
1582	int64_t delta;
1583
1584	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1585
1586	mutex_enter(&dd->dd_lock);
1587	used = dsl_dir_phys(dd)->dd_used_bytes;
1588	delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
1589	dsl_dir_phys(dd)->dd_reserved = value;
1590
1591	if (dd->dd_parent != NULL) {
1592		/* Roll up this additional usage into our ancestors */
1593		dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1594		    delta, 0, 0, tx);
1595	}
1596	mutex_exit(&dd->dd_lock);
1597}
1598
1599
1600static void
1601dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1602{
1603	dsl_dir_set_qr_arg_t *ddsqra = arg;
1604	dsl_pool_t *dp = dmu_tx_pool(tx);
1605	dsl_dataset_t *ds;
1606	uint64_t newval;
1607
1608	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1609
1610	if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1611		dsl_prop_set_sync_impl(ds,
1612		    zfs_prop_to_name(ZFS_PROP_RESERVATION),
1613		    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1614		    &ddsqra->ddsqra_value, tx);
1615
1616		VERIFY0(dsl_prop_get_int_ds(ds,
1617		    zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1618	} else {
1619		newval = ddsqra->ddsqra_value;
1620		spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1621		    zfs_prop_to_name(ZFS_PROP_RESERVATION),
1622		    (longlong_t)newval);
1623	}
1624
1625	dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1626	dsl_dataset_rele(ds, FTAG);
1627}
1628
1629int
1630dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1631    uint64_t reservation)
1632{
1633	dsl_dir_set_qr_arg_t ddsqra;
1634
1635	ddsqra.ddsqra_name = ddname;
1636	ddsqra.ddsqra_source = source;
1637	ddsqra.ddsqra_value = reservation;
1638
1639	return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1640	    dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1641}
1642
1643static dsl_dir_t *
1644closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1645{
1646	for (; ds1; ds1 = ds1->dd_parent) {
1647		dsl_dir_t *dd;
1648		for (dd = ds2; dd; dd = dd->dd_parent) {
1649			if (ds1 == dd)
1650				return (dd);
1651		}
1652	}
1653	return (NULL);
1654}
1655
1656/*
1657 * If delta is applied to dd, how much of that delta would be applied to
1658 * ancestor?  Syncing context only.
1659 */
1660static int64_t
1661would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1662{
1663	if (dd == ancestor)
1664		return (delta);
1665
1666	mutex_enter(&dd->dd_lock);
1667	delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
1668	mutex_exit(&dd->dd_lock);
1669	return (would_change(dd->dd_parent, delta, ancestor));
1670}
1671
1672typedef struct dsl_dir_rename_arg {
1673	const char *ddra_oldname;
1674	const char *ddra_newname;
1675	cred_t *ddra_cred;
1676} dsl_dir_rename_arg_t;
1677
1678/* ARGSUSED */
1679static int
1680dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1681{
1682	int *deltap = arg;
1683	char namebuf[MAXNAMELEN];
1684
1685	dsl_dataset_name(ds, namebuf);
1686
1687	if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1688		return (SET_ERROR(ENAMETOOLONG));
1689	return (0);
1690}
1691
1692static int
1693dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1694{
1695	dsl_dir_rename_arg_t *ddra = arg;
1696	dsl_pool_t *dp = dmu_tx_pool(tx);
1697	dsl_dir_t *dd, *newparent;
1698	const char *mynewname;
1699	int error;
1700	int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1701
1702	/* target dir should exist */
1703	error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1704	if (error != 0)
1705		return (error);
1706
1707	/* new parent should exist */
1708	error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1709	    &newparent, &mynewname);
1710	if (error != 0) {
1711		dsl_dir_rele(dd, FTAG);
1712		return (error);
1713	}
1714
1715	/* can't rename to different pool */
1716	if (dd->dd_pool != newparent->dd_pool) {
1717		dsl_dir_rele(newparent, FTAG);
1718		dsl_dir_rele(dd, FTAG);
1719		return (SET_ERROR(ENXIO));
1720	}
1721
1722	/* new name should not already exist */
1723	if (mynewname == NULL) {
1724		dsl_dir_rele(newparent, FTAG);
1725		dsl_dir_rele(dd, FTAG);
1726		return (SET_ERROR(EEXIST));
1727	}
1728
1729	/* if the name length is growing, validate child name lengths */
1730	if (delta > 0) {
1731		error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1732		    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1733		if (error != 0) {
1734			dsl_dir_rele(newparent, FTAG);
1735			dsl_dir_rele(dd, FTAG);
1736			return (error);
1737		}
1738	}
1739
1740	if (dmu_tx_is_syncing(tx)) {
1741		if (spa_feature_is_active(dp->dp_spa,
1742		    SPA_FEATURE_FS_SS_LIMIT)) {
1743			/*
1744			 * Although this is the check function and we don't
1745			 * normally make on-disk changes in check functions,
1746			 * we need to do that here.
1747			 *
1748			 * Ensure this portion of the tree's counts have been
1749			 * initialized in case the new parent has limits set.
1750			 */
1751			dsl_dir_init_fs_ss_count(dd, tx);
1752		}
1753	}
1754
1755	if (newparent != dd->dd_parent) {
1756		/* is there enough space? */
1757		uint64_t myspace =
1758		    MAX(dsl_dir_phys(dd)->dd_used_bytes,
1759		    dsl_dir_phys(dd)->dd_reserved);
1760		objset_t *os = dd->dd_pool->dp_meta_objset;
1761		uint64_t fs_cnt = 0;
1762		uint64_t ss_cnt = 0;
1763
1764		if (dsl_dir_is_zapified(dd)) {
1765			int err;
1766
1767			err = zap_lookup(os, dd->dd_object,
1768			    DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1769			    &fs_cnt);
1770			if (err != ENOENT && err != 0) {
1771				dsl_dir_rele(newparent, FTAG);
1772				dsl_dir_rele(dd, FTAG);
1773				return (err);
1774			}
1775
1776			/*
1777			 * have to add 1 for the filesystem itself that we're
1778			 * moving
1779			 */
1780			fs_cnt++;
1781
1782			err = zap_lookup(os, dd->dd_object,
1783			    DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1784			    &ss_cnt);
1785			if (err != ENOENT && err != 0) {
1786				dsl_dir_rele(newparent, FTAG);
1787				dsl_dir_rele(dd, FTAG);
1788				return (err);
1789			}
1790		}
1791
1792		/* no rename into our descendant */
1793		if (closest_common_ancestor(dd, newparent) == dd) {
1794			dsl_dir_rele(newparent, FTAG);
1795			dsl_dir_rele(dd, FTAG);
1796			return (SET_ERROR(EINVAL));
1797		}
1798
1799		error = dsl_dir_transfer_possible(dd->dd_parent,
1800		    newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1801		if (error != 0) {
1802			dsl_dir_rele(newparent, FTAG);
1803			dsl_dir_rele(dd, FTAG);
1804			return (error);
1805		}
1806	}
1807
1808	dsl_dir_rele(newparent, FTAG);
1809	dsl_dir_rele(dd, FTAG);
1810	return (0);
1811}
1812
1813static void
1814dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1815{
1816	dsl_dir_rename_arg_t *ddra = arg;
1817	dsl_pool_t *dp = dmu_tx_pool(tx);
1818	dsl_dir_t *dd, *newparent;
1819	const char *mynewname;
1820	int error;
1821	objset_t *mos = dp->dp_meta_objset;
1822
1823	VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1824	VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1825	    &mynewname));
1826
1827	/* Log this before we change the name. */
1828	spa_history_log_internal_dd(dd, "rename", tx,
1829	    "-> %s", ddra->ddra_newname);
1830
1831	if (newparent != dd->dd_parent) {
1832		objset_t *os = dd->dd_pool->dp_meta_objset;
1833		uint64_t fs_cnt = 0;
1834		uint64_t ss_cnt = 0;
1835
1836		/*
1837		 * We already made sure the dd counts were initialized in the
1838		 * check function.
1839		 */
1840		if (spa_feature_is_active(dp->dp_spa,
1841		    SPA_FEATURE_FS_SS_LIMIT)) {
1842			VERIFY0(zap_lookup(os, dd->dd_object,
1843			    DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1844			    &fs_cnt));
1845			/* add 1 for the filesystem itself that we're moving */
1846			fs_cnt++;
1847
1848			VERIFY0(zap_lookup(os, dd->dd_object,
1849			    DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1850			    &ss_cnt));
1851		}
1852
1853		dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1854		    DD_FIELD_FILESYSTEM_COUNT, tx);
1855		dsl_fs_ss_count_adjust(newparent, fs_cnt,
1856		    DD_FIELD_FILESYSTEM_COUNT, tx);
1857
1858		dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1859		    DD_FIELD_SNAPSHOT_COUNT, tx);
1860		dsl_fs_ss_count_adjust(newparent, ss_cnt,
1861		    DD_FIELD_SNAPSHOT_COUNT, tx);
1862
1863		dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1864		    -dsl_dir_phys(dd)->dd_used_bytes,
1865		    -dsl_dir_phys(dd)->dd_compressed_bytes,
1866		    -dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1867		dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1868		    dsl_dir_phys(dd)->dd_used_bytes,
1869		    dsl_dir_phys(dd)->dd_compressed_bytes,
1870		    dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
1871
1872		if (dsl_dir_phys(dd)->dd_reserved >
1873		    dsl_dir_phys(dd)->dd_used_bytes) {
1874			uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
1875			    dsl_dir_phys(dd)->dd_used_bytes;
1876
1877			dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1878			    -unused_rsrv, 0, 0, tx);
1879			dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1880			    unused_rsrv, 0, 0, tx);
1881		}
1882	}
1883
1884	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1885
1886	/* remove from old parent zapobj */
1887	error = zap_remove(mos,
1888	    dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
1889	    dd->dd_myname, tx);
1890	ASSERT0(error);
1891
1892	(void) strcpy(dd->dd_myname, mynewname);
1893	dsl_dir_rele(dd->dd_parent, dd);
1894	dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
1895	VERIFY0(dsl_dir_hold_obj(dp,
1896	    newparent->dd_object, NULL, dd, &dd->dd_parent));
1897
1898	/* add to new parent zapobj */
1899	VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
1900	    dd->dd_myname, 8, 1, &dd->dd_object, tx));
1901
1902	dsl_prop_notify_all(dd);
1903
1904	dsl_dir_rele(newparent, FTAG);
1905	dsl_dir_rele(dd, FTAG);
1906}
1907
1908int
1909dsl_dir_rename(const char *oldname, const char *newname)
1910{
1911	dsl_dir_rename_arg_t ddra;
1912
1913	ddra.ddra_oldname = oldname;
1914	ddra.ddra_newname = newname;
1915	ddra.ddra_cred = CRED();
1916
1917	return (dsl_sync_task(oldname,
1918	    dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
1919	    3, ZFS_SPACE_CHECK_RESERVED));
1920}
1921
1922int
1923dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1924    uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1925{
1926	dsl_dir_t *ancestor;
1927	int64_t adelta;
1928	uint64_t avail;
1929	int err;
1930
1931	ancestor = closest_common_ancestor(sdd, tdd);
1932	adelta = would_change(sdd, -space, ancestor);
1933	avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1934	if (avail < space)
1935		return (SET_ERROR(ENOSPC));
1936
1937	err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1938	    ancestor, cr);
1939	if (err != 0)
1940		return (err);
1941	err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1942	    ancestor, cr);
1943	if (err != 0)
1944		return (err);
1945
1946	return (0);
1947}
1948
1949timestruc_t
1950dsl_dir_snap_cmtime(dsl_dir_t *dd)
1951{
1952	timestruc_t t;
1953
1954	mutex_enter(&dd->dd_lock);
1955	t = dd->dd_snap_cmtime;
1956	mutex_exit(&dd->dd_lock);
1957
1958	return (t);
1959}
1960
1961void
1962dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1963{
1964	timestruc_t t;
1965
1966	gethrestime(&t);
1967	mutex_enter(&dd->dd_lock);
1968	dd->dd_snap_cmtime = t;
1969	mutex_exit(&dd->dd_lock);
1970}
1971
1972void
1973dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1974{
1975	objset_t *mos = dd->dd_pool->dp_meta_objset;
1976	dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1977}
1978
1979boolean_t
1980dsl_dir_is_zapified(dsl_dir_t *dd)
1981{
1982	dmu_object_info_t doi;
1983
1984	dmu_object_info_from_db(dd->dd_dbuf, &doi);
1985	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
1986}
1987