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 by Delphix. All rights reserved.
24 * Copyright (c) 2012 Joyent, Inc. 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/spa.h>
36 #include <sys/metaslab.h>
37 #include <sys/zap.h>
38 #include <sys/zio.h>
39 #include <sys/arc.h>
40 #include <sys/sunddi.h>
41 #include <sys/zfs_zone.h>
42 #include <sys/zfeature.h>
43 #include "zfs_namecheck.h"
44 #include "zfs_prop.h"
45
46 /*
47 * Filesystem and Snapshot Limits
48 * ------------------------------
49 *
50 * These limits are used to restrict the number of filesystems and/or snapshots
51 * that can be created at a given level in the tree or below. The standard
52 * use-case is with a delegated dataset where the administrator wants to ensure
53 * that a user within the zone is not creating too many additional filesystems
54 * or snapshots, even though they're not exceeding their space quota.
55 *
56 * The count of filesystems and snapshots is stored in the dsl_dir_phys_t which
57 * impacts the on-disk format. As such, this capability is controlled by a
58 * feature flag and must be enabled to be used. Once enabled, the feature is
59 * not active until the first limit is set. At that point, future operations to
60 * create/destroy filesystems or snapshots will validate and update the counts.
61 *
62 * Because the on-disk counts will be uninitialized (0) before the feature is
63 * active, the counts are updated when a limit is first set on an uninitialized
64 * node (The filesystem/snapshot counts on a node includes all of the nested
65 * filesystems/snapshots, plus the node itself. Thus, a new leaf node has a
66 * filesystem count of 1 and a snapshot count of 0. A filesystem count of 0 on
67 * a node indicates uninitialized counts on that node.) When setting a limit on
68 * an uninitialized node, the code starts at the filesystem with the new limit
69 * and descends into all sub-filesystems and updates the counts to be accurate.
70 * In practice this is lightweight since a limit is typically set when the
71 * filesystem is created and thus has no children. Once valid, changing the
72 * limit value won't require a re-traversal since the counts are already valid.
73 * When recursively fixing the counts, if a node with a limit is encountered
74 * during the descent, the counts are known to be valid and there is no need to
75 * descend into that filesystem's children. The counts on filesystems above the
76 * one with the new limit will still be uninitialized (0), unless a limit is
77 * eventually set on one of those filesystems. It is possible for the counts
78 * to appear initialized, but be invalid, if the feature was previously active
79 * but then deactivated. For this reason, the counts are always recursively
80 * updated when a limit is set on a dataset, unless there is already a limit.
81 * When a new limit value is set on a filesystem with an existing limit, the
82 * new limit must be greater than the current count at that level or an error
83 * is returned and the limit is not changed.
84 *
85 * Once the feature is active, then whenever a filesystem or snapshot is
86 * created, the code recurses up the tree, validating the new count against the
87 * limit at each initialized level. In practice, most levels will not have a
88 * limit set. If there is a limit at any initialized level up the tree, the
89 * check must pass or the creation will fail. Likewise, when a filesystem or
90 * snapshot is destroyed, the counts are recursively adjusted all the way up
91 * the initizized nodes in the tree. Renaming a filesystem into different point
92 * in the tree will first validate, then update the counts on each branch up to
93 * the common ancestor. A receive will also validate the counts and then update
94 * them.
95 *
96 * An exception to the above behavior is that the limits are never enforced
97 * for the administrative user in the global zone. This is primarily so that
98 * recursive snapshots in the global zone always work. We want to prevent a
99 * denial-of-service in which a lower level delegated dataset could max out its
100 * limit and thus block recursive snapshots from being taken in the global zone.
101 * Because of this, it is possible for the snapshot count to be over the limit
102 * and snapshots taken in the global zone could cause a lower level dataset to
103 * hit or exceed its limit. The administrator taking the global zone recursive
104 * snapshot should be aware of this side-effect and behave accordingly.
105 * For consistency, the filesystem limit is also not enforced for the admin
106 * user in the global zone.
107 *
108 * The filesystem limit is validated by dsl_dir_fscount_check() and updated by
109 * dsl_dir_fscount_adjust(). The snapshot limit is validated by
110 * dsl_snapcount_check() and updated by dsl_snapcount_adjust().
111 * A new limit value is validated in dsl_dir_validate_fs_ss_limit() and the
112 * filesystem counts are adjusted, if necessary, by dsl_dir_set_fs_ss_count().
113 */
114
115 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
116 static void dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd,
117 uint64_t value, dmu_tx_t *tx);
118
119 extern dsl_syncfunc_t dsl_prop_set_sync;
120
121 /* ARGSUSED */
122 static void
123 dsl_dir_evict(dmu_buf_t *db, void *arg)
124 {
125 dsl_dir_t *dd = arg;
126 dsl_pool_t *dp = dd->dd_pool;
127 int t;
128
129 for (t = 0; t < TXG_SIZE; t++) {
130 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
131 ASSERT(dd->dd_tempreserved[t] == 0);
132 ASSERT(dd->dd_space_towrite[t] == 0);
133 }
134
135 if (dd->dd_parent)
136 dsl_dir_close(dd->dd_parent, dd);
137
138 spa_close(dd->dd_pool->dp_spa, dd);
139
140 /*
141 * The props callback list should have been cleaned up by
142 * objset_evict().
143 */
144 list_destroy(&dd->dd_prop_cbs);
145 mutex_destroy(&dd->dd_lock);
146 kmem_free(dd, sizeof (dsl_dir_t));
147 }
148
149 int
150 dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
151 const char *tail, void *tag, dsl_dir_t **ddp)
152 {
153 dmu_buf_t *dbuf;
154 dsl_dir_t *dd;
155 int err;
156
157 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
158 dsl_pool_sync_context(dp));
159
160 err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
161 if (err)
162 return (err);
163 dd = dmu_buf_get_user(dbuf);
164 #ifdef ZFS_DEBUG
165 {
166 dmu_object_info_t doi;
167 dmu_object_info_from_db(dbuf, &doi);
168 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
169 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
170 }
171 #endif
172 if (dd == NULL) {
173 dsl_dir_t *winner;
174
175 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
176 dd->dd_object = ddobj;
177 dd->dd_dbuf = dbuf;
178 dd->dd_pool = dp;
179 dd->dd_phys = dbuf->db_data;
180 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
181
182 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
183 offsetof(dsl_prop_cb_record_t, cbr_node));
184
185 dsl_dir_snap_cmtime_update(dd);
186
187 if (dd->dd_phys->dd_parent_obj) {
188 err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
189 NULL, dd, &dd->dd_parent);
190 if (err)
191 goto errout;
192 if (tail) {
193 #ifdef ZFS_DEBUG
194 uint64_t foundobj;
195
196 err = zap_lookup(dp->dp_meta_objset,
197 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
198 tail, sizeof (foundobj), 1, &foundobj);
199 ASSERT(err || foundobj == ddobj);
200 #endif
201 (void) strcpy(dd->dd_myname, tail);
202 } else {
203 err = zap_value_search(dp->dp_meta_objset,
204 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
205 ddobj, 0, dd->dd_myname);
206 }
207 if (err)
208 goto errout;
209 } else {
210 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
211 }
212
213 if (dsl_dir_is_clone(dd)) {
214 dmu_buf_t *origin_bonus;
215 dsl_dataset_phys_t *origin_phys;
216
217 /*
218 * We can't open the origin dataset, because
219 * that would require opening this dsl_dir.
220 * Just look at its phys directly instead.
221 */
222 err = dmu_bonus_hold(dp->dp_meta_objset,
223 dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
224 if (err)
225 goto errout;
226 origin_phys = origin_bonus->db_data;
227 dd->dd_origin_txg =
228 origin_phys->ds_creation_txg;
229 dmu_buf_rele(origin_bonus, FTAG);
230 }
231
232 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
233 dsl_dir_evict);
234 if (winner) {
235 if (dd->dd_parent)
236 dsl_dir_close(dd->dd_parent, dd);
237 mutex_destroy(&dd->dd_lock);
238 kmem_free(dd, sizeof (dsl_dir_t));
239 dd = winner;
240 } else {
241 spa_open_ref(dp->dp_spa, dd);
242 }
243 }
244
245 /*
246 * The dsl_dir_t has both open-to-close and instantiate-to-evict
247 * holds on the spa. We need the open-to-close holds because
248 * otherwise the spa_refcnt wouldn't change when we open a
249 * dir which the spa also has open, so we could incorrectly
250 * think it was OK to unload/export/destroy the pool. We need
251 * the instantiate-to-evict hold because the dsl_dir_t has a
252 * pointer to the dd_pool, which has a pointer to the spa_t.
253 */
254 spa_open_ref(dp->dp_spa, tag);
255 ASSERT3P(dd->dd_pool, ==, dp);
256 ASSERT3U(dd->dd_object, ==, ddobj);
257 ASSERT3P(dd->dd_dbuf, ==, dbuf);
258 *ddp = dd;
259 return (0);
260
261 errout:
262 if (dd->dd_parent)
263 dsl_dir_close(dd->dd_parent, dd);
264 mutex_destroy(&dd->dd_lock);
265 kmem_free(dd, sizeof (dsl_dir_t));
266 dmu_buf_rele(dbuf, tag);
267 return (err);
268 }
269
270 void
271 dsl_dir_close(dsl_dir_t *dd, void *tag)
272 {
273 dprintf_dd(dd, "%s\n", "");
274 spa_close(dd->dd_pool->dp_spa, tag);
275 dmu_buf_rele(dd->dd_dbuf, tag);
276 }
277
278 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
279 void
280 dsl_dir_name(dsl_dir_t *dd, char *buf)
281 {
282 if (dd->dd_parent) {
283 dsl_dir_name(dd->dd_parent, buf);
284 (void) strcat(buf, "/");
285 } else {
286 buf[0] = '\0';
287 }
288 if (!MUTEX_HELD(&dd->dd_lock)) {
289 /*
290 * recursive mutex so that we can use
291 * dprintf_dd() with dd_lock held
292 */
293 mutex_enter(&dd->dd_lock);
294 (void) strcat(buf, dd->dd_myname);
295 mutex_exit(&dd->dd_lock);
296 } else {
297 (void) strcat(buf, dd->dd_myname);
298 }
299 }
300
301 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
302 int
303 dsl_dir_namelen(dsl_dir_t *dd)
304 {
305 int result = 0;
306
307 if (dd->dd_parent) {
308 /* parent's name + 1 for the "/" */
309 result = dsl_dir_namelen(dd->dd_parent) + 1;
310 }
311
312 if (!MUTEX_HELD(&dd->dd_lock)) {
313 /* see dsl_dir_name */
314 mutex_enter(&dd->dd_lock);
315 result += strlen(dd->dd_myname);
316 mutex_exit(&dd->dd_lock);
317 } else {
318 result += strlen(dd->dd_myname);
319 }
320
321 return (result);
322 }
323
324 static int
325 getcomponent(const char *path, char *component, const char **nextp)
326 {
327 char *p;
328 if ((path == NULL) || (path[0] == '\0'))
329 return (ENOENT);
330 /* This would be a good place to reserve some namespace... */
331 p = strpbrk(path, "/@");
332 if (p && (p[1] == '/' || p[1] == '@')) {
333 /* two separators in a row */
334 return (EINVAL);
335 }
336 if (p == NULL || p == path) {
337 /*
338 * if the first thing is an @ or /, it had better be an
339 * @ and it had better not have any more ats or slashes,
340 * and it had better have something after the @.
341 */
342 if (p != NULL &&
343 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
344 return (EINVAL);
345 if (strlen(path) >= MAXNAMELEN)
346 return (ENAMETOOLONG);
347 (void) strcpy(component, path);
348 p = NULL;
349 } else if (p[0] == '/') {
350 if (p-path >= MAXNAMELEN)
351 return (ENAMETOOLONG);
352 (void) strncpy(component, path, p - path);
353 component[p-path] = '\0';
354 p++;
355 } else if (p[0] == '@') {
356 /*
357 * if the next separator is an @, there better not be
358 * any more slashes.
359 */
360 if (strchr(path, '/'))
361 return (EINVAL);
362 if (p-path >= MAXNAMELEN)
363 return (ENAMETOOLONG);
364 (void) strncpy(component, path, p - path);
365 component[p-path] = '\0';
366 } else {
367 ASSERT(!"invalid p");
368 }
369 *nextp = p;
370 return (0);
371 }
372
373 /*
374 * same as dsl_open_dir, ignore the first component of name and use the
375 * spa instead
376 */
377 int
378 dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
379 dsl_dir_t **ddp, const char **tailp)
380 {
381 char buf[MAXNAMELEN];
382 const char *next, *nextnext = NULL;
383 int err;
384 dsl_dir_t *dd;
385 dsl_pool_t *dp;
386 uint64_t ddobj;
387 int openedspa = FALSE;
388
389 dprintf("%s\n", name);
390
391 err = getcomponent(name, buf, &next);
392 if (err)
393 return (err);
394 if (spa == NULL) {
395 err = spa_open(buf, &spa, FTAG);
396 if (err) {
397 dprintf("spa_open(%s) failed\n", buf);
398 return (err);
399 }
400 openedspa = TRUE;
401
402 /* XXX this assertion belongs in spa_open */
403 ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
404 }
405
406 dp = spa_get_dsl(spa);
407
408 rw_enter(&dp->dp_config_rwlock, RW_READER);
409 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
410 if (err) {
411 rw_exit(&dp->dp_config_rwlock);
412 if (openedspa)
413 spa_close(spa, FTAG);
414 return (err);
415 }
416
417 while (next != NULL) {
418 dsl_dir_t *child_ds;
419 err = getcomponent(next, buf, &nextnext);
420 if (err)
421 break;
422 ASSERT(next[0] != '\0');
423 if (next[0] == '@')
424 break;
425 dprintf("looking up %s in obj%lld\n",
426 buf, dd->dd_phys->dd_child_dir_zapobj);
427
428 err = zap_lookup(dp->dp_meta_objset,
429 dd->dd_phys->dd_child_dir_zapobj,
430 buf, sizeof (ddobj), 1, &ddobj);
431 if (err) {
432 if (err == ENOENT)
433 err = 0;
434 break;
435 }
436
437 err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
438 if (err)
439 break;
440 dsl_dir_close(dd, tag);
441 dd = child_ds;
442 next = nextnext;
443 }
444 rw_exit(&dp->dp_config_rwlock);
445
446 if (err) {
447 dsl_dir_close(dd, tag);
448 if (openedspa)
449 spa_close(spa, FTAG);
450 return (err);
451 }
452
453 /*
454 * It's an error if there's more than one component left, or
455 * tailp==NULL and there's any component left.
456 */
457 if (next != NULL &&
458 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
459 /* bad path name */
460 dsl_dir_close(dd, tag);
461 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
462 err = ENOENT;
463 }
464 if (tailp)
465 *tailp = next;
466 if (openedspa)
467 spa_close(spa, FTAG);
468 *ddp = dd;
469 return (err);
470 }
471
472 /*
473 * Return the dsl_dir_t, and possibly the last component which couldn't
474 * be found in *tail. Return NULL if the path is bogus, or if
475 * tail==NULL and we couldn't parse the whole name. (*tail)[0] == '@'
476 * means that the last component is a snapshot.
477 */
478 int
479 dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
480 {
481 return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
482 }
483
484 /*
485 * Check if the counts are already valid for this filesystem and its
486 * descendants. The counts on this filesystem, and those below, may be
487 * uninitialized due to either the use of a pre-existing pool which did not
488 * support the filesystem/snapshot limit feature, or one in which the feature
489 * had not yet been enabled. The counts can also be invalid if the feature was
490 * previously active but then deactivated.
491 *
492 * Recursively descend the filesystem tree and update the filesystem/snapshot
493 * counts on each filesystem below, then update the cumulative count on the
494 * current filesystem. If the filesystem already has a limit set on it,
495 * then we know that its counts, and the counts on the filesystems below it,
496 * have been updated to be correct, so we can skip this filesystem.
497 */
498 static void
499 dsl_dir_set_fs_ss_count(const char *nm, dsl_dir_t *dd, dmu_tx_t *tx,
500 uint64_t *fscnt, uint64_t *sscnt)
501 {
502 uint64_t my_fs_cnt = 0;
503 uint64_t my_ss_cnt = 0;
504 objset_t *os = dd->dd_pool->dp_meta_objset;
505 zap_cursor_t *zc;
506 zap_attribute_t *za;
507 char *namebuf;
508 int err;
509 boolean_t limit_set = B_FALSE;
510 uint64_t fslimit, sslimit;
511 dsl_dataset_t *ds;
512
513 err = dsl_prop_get_dd(dd, zfs_prop_to_name(ZFS_PROP_FILESYSTEM_LIMIT),
514 8, 1, &fslimit, NULL, B_FALSE);
515 if (err == 0 && fslimit != MAXLIMIT)
516 limit_set = B_TRUE;
517
518 if (!limit_set) {
519 err = dsl_prop_get_dd(dd,
520 zfs_prop_to_name(ZFS_PROP_SNAPSHOT_LIMIT), 8, 1, &sslimit,
521 NULL, B_FALSE);
522 if (err == 0 && sslimit != MAXLIMIT)
523 limit_set = B_TRUE;
524 }
525
526 /*
527 * If the dd has a limit, we know its count is already good and we
528 * don't need to recurse down any further.
529 *
530 * We can't check for an initialized (non-0) count since the feature
531 * might have been previously active, then deactivated and is now
532 * being activated again.
533 */
534 if (limit_set) {
535 *fscnt = dd->dd_phys->dd_filesystem_count;
536 *sscnt = dd->dd_phys->dd_snapshot_count;
537 return;
538 }
539
540 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
541 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
542 namebuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
543
544 mutex_enter(&dd->dd_lock);
545
546 /* Iterate datasets */
547 for (zap_cursor_init(zc, os, dd->dd_phys->dd_child_dir_zapobj);
548 zap_cursor_retrieve(zc, za) == 0;
549 zap_cursor_advance(zc)) {
550 dsl_dir_t *chld_dd;
551 uint64_t chld_fs_cnt = 0;
552 uint64_t chld_ss_cnt = 0;
553
554 (void) snprintf(namebuf, MAXPATHLEN, "%s/%s", nm, za->za_name);
555
556 if (dsl_dir_open(namebuf, FTAG, &chld_dd, NULL))
557 continue;
558
559 dsl_dir_set_fs_ss_count(namebuf, chld_dd, tx, &chld_fs_cnt,
560 &chld_ss_cnt);
561
562 dsl_dir_close(chld_dd, FTAG);
563
564 my_fs_cnt += chld_fs_cnt;
565 my_ss_cnt += chld_ss_cnt;
566 }
567 zap_cursor_fini(zc);
568
569 kmem_free(namebuf, MAXPATHLEN);
570
571 /* Iterate snapshots */
572 if (dsl_dataset_hold(nm, FTAG, &ds) == 0) {
573 for (zap_cursor_init(zc, os, ds->ds_phys->ds_snapnames_zapobj);
574 zap_cursor_retrieve(zc, za) == 0;
575 zap_cursor_advance(zc)) {
576 my_ss_cnt++;
577 }
578 zap_cursor_fini(zc);
579 dsl_dataset_rele(ds, FTAG);
580 }
581
582 kmem_free(zc, sizeof (zap_cursor_t));
583 kmem_free(za, sizeof (zap_attribute_t));
584
585 /* Add 1 for self */
586 my_fs_cnt++;
587
588 #ifdef _KERNEL
589 extern void __dtrace_probe_zfs__fs__fix__count(char *, uint64_t,
590 uint64_t);
591 __dtrace_probe_zfs__fs__fix__count((char *)nm, my_fs_cnt, my_ss_cnt);
592 #endif
593
594 /* save updated counts */
595 dmu_buf_will_dirty(dd->dd_dbuf, tx);
596 dd->dd_phys->dd_filesystem_count = my_fs_cnt;
597 dd->dd_phys->dd_snapshot_count = my_ss_cnt;
598
599 mutex_exit(&dd->dd_lock);
600
601 /* Return child dataset count plus self */
602 *fscnt = my_fs_cnt;
603 *sscnt = my_ss_cnt;
604 }
605
606 /*
607 * Return ENOSPC if new limit is less than the existing count, otherwise return
608 * -1 to force the zfs_set_prop_nvlist code down the default path to set the
609 * value in the nvlist.
610 */
611 int
612 dsl_dir_validate_fs_ss_limit(const char *ddname, uint64_t limit,
613 zfs_prop_t ptype)
614 {
615 dsl_dir_t *dd;
616 dsl_dataset_t *ds;
617 int err = -1;
618 uint64_t count;
619 dmu_tx_t *tx;
620 uint64_t my_fs_cnt = 0;
621 uint64_t my_ss_cnt = 0;
622 uint64_t curr_limit;
623 spa_t *spa;
624 zfeature_info_t *limit_feat =
625 &spa_feature_table[SPA_FEATURE_FS_SS_LIMIT];
626
627 if (dsl_dataset_hold(ddname, FTAG, &ds))
628 return (EACCES);
629
630 spa = dsl_dataset_get_spa(ds);
631 if (!spa_feature_is_enabled(spa,
632 &spa_feature_table[SPA_FEATURE_FS_SS_LIMIT])) {
633 dsl_dataset_rele(ds, FTAG);
634 return (ENOTSUP);
635 }
636
637 if (dsl_dir_open(ddname, FTAG, &dd, NULL)) {
638 dsl_dataset_rele(ds, FTAG);
639 return (EACCES);
640 }
641
642 ASSERT(ds->ds_dir == dd);
643
644 if (dsl_prop_get_dd(dd, zfs_prop_to_name(ptype), 8, 1, &curr_limit,
645 NULL, B_FALSE) != 0)
646 curr_limit = MAXLIMIT;
647
648 tx = dmu_tx_create_dd(dd);
649 if (dmu_tx_assign(tx, TXG_WAIT)) {
650 dmu_tx_abort(tx);
651 dsl_dir_close(dd, FTAG);
652 dsl_dataset_rele(ds, FTAG);
653 return (ENOSPC);
654 }
655
656 if (limit == MAXLIMIT) {
657 /*
658 * If we had a limit, since we're now removing that limit,
659 * decrement the feature-active counter so that the feature
660 * becomes inactive (only enabled) if we remove the last limit.
661 */
662 if (curr_limit != MAXLIMIT)
663 spa_feature_decr(spa, limit_feat, tx);
664
665 dmu_tx_commit(tx);
666 dsl_dir_close(dd, FTAG);
667 dsl_dataset_rele(ds, FTAG);
668 return (-1);
669 }
670
671 /*
672 * Since we are now setting a non-MAXLIMIT on the filesystem, we need
673 * to ensure the counts are correct. Descend down the tree from this
674 * point and update all of the counts to be accurate.
675 */
676 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
677 dsl_dir_set_fs_ss_count(ddname, dd, tx, &my_fs_cnt, &my_ss_cnt);
678 rw_exit(&dd->dd_pool->dp_config_rwlock);
679
680 if (ptype == ZFS_PROP_FILESYSTEM_LIMIT)
681 count = dd->dd_phys->dd_filesystem_count;
682 else
683 count = dd->dd_phys->dd_snapshot_count;
684
685 if (limit < count) {
686 err = ENOSPC;
687 } else {
688 /*
689 * If we had no limit, since we're now setting a limit
690 * increment the feature-active counter so that the feature
691 * either becomes active for the first time, or the count
692 * simply increases so that we can decrement it when we remove
693 * the limit.
694 */
695 if (curr_limit == MAXLIMIT)
696 spa_feature_incr(spa, limit_feat, tx);
697 }
698
699 dmu_tx_commit(tx);
700
701 dsl_dir_close(dd, FTAG);
702 dsl_dataset_rele(ds, FTAG);
703
704 return (err);
705 }
706
707 /*
708 * Check if adding additional child filesystem(s) would exceed any filesystem
709 * limits. Note that all filesystem limits up to the root (or the highest
710 * initialized) filesystem or the given ancestor must be satisfied.
711 */
712 int
713 dsl_dir_fscount_check(dsl_dir_t *dd, uint64_t cnt, dsl_dir_t *ancestor)
714 {
715 uint64_t limit;
716 int err = 0;
717
718 VERIFY(RW_LOCK_HELD(&dd->dd_pool->dp_config_rwlock));
719
720 /*
721 * The limit is never enforced for the admin user in global zone.
722 * If we're not in the global zone then we need to run this check in
723 * open context, since thats when we know what zone we're in and
724 * syncing is only performed in the global zone.
725 */
726 if (INGLOBALZONE(curproc))
727 return (0);
728
729 /*
730 * If an ancestor has been provided, stop checking the limit once we
731 * hit that dir. We need this during rename so that we don't overcount
732 * the check once we recurse up to the common ancestor.
733 */
734 if (ancestor == dd)
735 return (0);
736
737 /*
738 * If we hit an uninitialized node while recursing up the tree, we can
739 * stop since we know the counts are not valid on this node and we
740 * know we won't touch this node's counts.
741 */
742 if (dd->dd_phys->dd_filesystem_count == 0)
743 return (0);
744
745 /*
746 * If there's no value for this property, there's no need to enforce a
747 * filesystem limit.
748 */
749 err = dsl_prop_get_dd(dd, zfs_prop_to_name(ZFS_PROP_FILESYSTEM_LIMIT),
750 8, 1, &limit, NULL, B_FALSE);
751 if (err == ENOENT)
752 return (0);
753 else if (err != 0)
754 return (err);
755
756 #ifdef _KERNEL
757 extern void __dtrace_probe_zfs__fs__limit(uint64_t, uint64_t, char *);
758 __dtrace_probe_zfs__fs__limit(
759 (uint64_t)dd->dd_phys->dd_filesystem_count, (uint64_t)limit,
760 dd->dd_myname);
761 #endif
762
763 if (limit != MAXLIMIT &&
764 (dd->dd_phys->dd_filesystem_count + cnt) > limit)
765 return (EDQUOT);
766
767 if (dd->dd_parent != NULL)
768 err = dsl_dir_fscount_check(dd->dd_parent, cnt, ancestor);
769
770 return (err);
771 }
772
773 /*
774 * Adjust the filesystem count for the specified dsl_dir_t and all parent
775 * filesystems. When a new filesystem is created, increment the count on all
776 * parents, and when a filesystem is destroyed, decrement the count.
777 */
778 void
779 dsl_dir_fscount_adjust(dsl_dir_t *dd, dmu_tx_t *tx, int64_t delta,
780 boolean_t syncing, boolean_t first)
781 {
782 VERIFY(RW_LOCK_HELD(&dd->dd_pool->dp_config_rwlock));
783 if (syncing)
784 VERIFY(dmu_tx_is_syncing(tx));
785
786 /*
787 * There is a special case where we are receiving a filesystem that
788 * already exists. In this case a temporary clone name of %X is created
789 * (see dmu_recv_begin). In dmu_recv_existing_end we destroy this
790 * temporary clone. We never update the filesystem counts for temporary
791 * clones. To detect this case we check the filesystem name to see if
792 * its a hidden filesystem (%X).
793 */
794 if (dd->dd_myname[0] == '%')
795 return;
796
797 /*
798 * If we hit an uninitialized node while recursing up the tree, we can
799 * stop since we know the counts are not valid on this node and we
800 * know we shouldn't touch this node's counts. An uninitialized count
801 * on the node indicates that either the feature has not yet been
802 * activated or there are no limits on this part of the tree.
803 */
804 if (dd->dd_phys->dd_filesystem_count == 0)
805 return;
806
807 /*
808 * The feature might have previously been active, so there could be
809 * non-0 counts on the nodes, but it might now be inactive.
810 *
811 * On initial entry we need to check if this feature is active, but
812 * we don't want to re-check this on each recursive call. Note: the
813 * feature cannot be active if its not enabled. If the feature is not
814 * active, don't touch the on-disk count fields.
815 */
816 if (first) {
817 dsl_dataset_t *ds = NULL;
818 spa_t *spa;
819 zfeature_info_t *quota_feat =
820 &spa_feature_table[SPA_FEATURE_FS_SS_LIMIT];
821
822 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
823 dd->dd_phys->dd_head_dataset_obj, FTAG, &ds));
824 spa = dsl_dataset_get_spa(ds);
825 dsl_dataset_rele(ds, FTAG);
826 if (!spa_feature_is_active(spa, quota_feat))
827 return;
828 }
829
830 dmu_buf_will_dirty(dd->dd_dbuf, tx);
831
832 mutex_enter(&dd->dd_lock);
833
834 dd->dd_phys->dd_filesystem_count += delta;
835
836 if (dd->dd_parent != NULL)
837 dsl_dir_fscount_adjust(dd->dd_parent, tx, delta, syncing,
838 B_FALSE);
839
840 mutex_exit(&dd->dd_lock);
841 }
842
843 uint64_t
844 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
845 dmu_tx_t *tx)
846 {
847 objset_t *mos = dp->dp_meta_objset;
848 uint64_t ddobj;
849 dsl_dir_phys_t *ddphys;
850 dmu_buf_t *dbuf;
851 zfeature_info_t *limit_feat =
852 &spa_feature_table[SPA_FEATURE_FS_SS_LIMIT];
853
854
855 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
856 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
857 if (pds) {
858 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
859 name, sizeof (uint64_t), 1, &ddobj, tx));
860 } else {
861 /* it's the root dir */
862 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
863 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
864 }
865 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
866 dmu_buf_will_dirty(dbuf, tx);
867 ddphys = dbuf->db_data;
868
869 ddphys->dd_creation_time = gethrestime_sec();
870 /* Only initialize the count if the limit feature is active */
871 if (spa_feature_is_active(dp->dp_spa, limit_feat))
872 ddphys->dd_filesystem_count = 1;
873 if (pds)
874 ddphys->dd_parent_obj = pds->dd_object;
875 ddphys->dd_props_zapobj = zap_create(mos,
876 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
877 ddphys->dd_child_dir_zapobj = zap_create(mos,
878 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
879 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
880 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
881 dmu_buf_rele(dbuf, FTAG);
882
883 return (ddobj);
884 }
885
886 /* ARGSUSED */
887 int
888 dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
889 {
890 dsl_dir_t *dd = arg1;
891 dsl_pool_t *dp = dd->dd_pool;
892 objset_t *mos = dp->dp_meta_objset;
893 int err;
894 uint64_t count;
895
896 /*
897 * There should be exactly two holds, both from
898 * dsl_dataset_destroy: one on the dd directory, and one on its
899 * head ds. If there are more holds, then a concurrent thread is
900 * performing a lookup inside this dir while we're trying to destroy
901 * it. To minimize this possibility, we perform this check only
902 * in syncing context and fail the operation if we encounter
903 * additional holds. The dp_config_rwlock ensures that nobody else
904 * opens it after we check.
905 */
906 if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 2)
907 return (EBUSY);
908
909 err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
910 if (err)
911 return (err);
912 if (count != 0)
913 return (EEXIST);
914
915 return (0);
916 }
917
918 void
919 dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
920 {
921 dsl_dir_t *dd = arg1;
922 objset_t *mos = dd->dd_pool->dp_meta_objset;
923 uint64_t obj;
924 dd_used_t t;
925
926 ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
927 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
928
929 /* Decrement the filesystem count for all parent filesystems. */
930 if (dd->dd_parent != NULL)
931 dsl_dir_fscount_adjust(dd->dd_parent, tx, -1, B_TRUE, B_TRUE);
932
933 /*
934 * Remove our reservation. The impl() routine avoids setting the
935 * actual property, which would require the (already destroyed) ds.
936 */
937 dsl_dir_set_reservation_sync_impl(dd, 0, tx);
938
939 ASSERT0(dd->dd_phys->dd_used_bytes);
940 ASSERT0(dd->dd_phys->dd_reserved);
941 for (t = 0; t < DD_USED_NUM; t++)
942 ASSERT0(dd->dd_phys->dd_used_breakdown[t]);
943
944 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
945 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
946 VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
947 VERIFY(0 == zap_remove(mos,
948 dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
949
950 obj = dd->dd_object;
951 dsl_dir_close(dd, tag);
952 VERIFY(0 == dmu_object_free(mos, obj, tx));
953 }
954
955 boolean_t
956 dsl_dir_is_clone(dsl_dir_t *dd)
957 {
958 return (dd->dd_phys->dd_origin_obj &&
959 (dd->dd_pool->dp_origin_snap == NULL ||
960 dd->dd_phys->dd_origin_obj !=
961 dd->dd_pool->dp_origin_snap->ds_object));
962 }
963
964 void
965 dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
966 {
967 mutex_enter(&dd->dd_lock);
968 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
969 dd->dd_phys->dd_used_bytes);
970 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
971 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
972 dd->dd_phys->dd_reserved);
973 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
974 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
975 (dd->dd_phys->dd_uncompressed_bytes * 100 /
976 dd->dd_phys->dd_compressed_bytes));
977 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
978 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
979 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
980 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
981 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
982 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
983 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
984 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
985 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
986 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
987 }
988 mutex_exit(&dd->dd_lock);
989
990 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
991 if (dsl_dir_is_clone(dd)) {
992 dsl_dataset_t *ds;
993 char buf[MAXNAMELEN];
994
995 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
996 dd->dd_phys->dd_origin_obj, FTAG, &ds));
997 dsl_dataset_name(ds, buf);
998 dsl_dataset_rele(ds, FTAG);
999 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
1000 }
1001 rw_exit(&dd->dd_pool->dp_config_rwlock);
1002 }
1003
1004 void
1005 dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
1006 {
1007 dsl_pool_t *dp = dd->dd_pool;
1008
1009 ASSERT(dd->dd_phys);
1010
1011 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
1012 /* up the hold count until we can be written out */
1013 dmu_buf_add_ref(dd->dd_dbuf, dd);
1014 }
1015 }
1016
1017 static int64_t
1018 parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
1019 {
1020 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
1021 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
1022 return (new_accounted - old_accounted);
1023 }
1024
1025 void
1026 dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
1027 {
1028 ASSERT(dmu_tx_is_syncing(tx));
1029
1030 mutex_enter(&dd->dd_lock);
1031 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
1032 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
1033 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1034 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1035 mutex_exit(&dd->dd_lock);
1036
1037 /* release the hold from dsl_dir_dirty */
1038 dmu_buf_rele(dd->dd_dbuf, dd);
1039 }
1040
1041 static uint64_t
1042 dsl_dir_space_towrite(dsl_dir_t *dd)
1043 {
1044 uint64_t space = 0;
1045 int i;
1046
1047 ASSERT(MUTEX_HELD(&dd->dd_lock));
1048
1049 for (i = 0; i < TXG_SIZE; i++) {
1050 space += dd->dd_space_towrite[i&TXG_MASK];
1051 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1052 }
1053 return (space);
1054 }
1055
1056 /*
1057 * How much space would dd have available if ancestor had delta applied
1058 * to it? If ondiskonly is set, we're only interested in what's
1059 * on-disk, not estimated pending changes.
1060 */
1061 uint64_t
1062 dsl_dir_space_available(dsl_dir_t *dd,
1063 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1064 {
1065 uint64_t parentspace, myspace, quota, used;
1066
1067 /*
1068 * If there are no restrictions otherwise, assume we have
1069 * unlimited space available.
1070 */
1071 quota = UINT64_MAX;
1072 parentspace = UINT64_MAX;
1073
1074 if (dd->dd_parent != NULL) {
1075 parentspace = dsl_dir_space_available(dd->dd_parent,
1076 ancestor, delta, ondiskonly);
1077 }
1078
1079 mutex_enter(&dd->dd_lock);
1080 if (dd->dd_phys->dd_quota != 0)
1081 quota = dd->dd_phys->dd_quota;
1082 used = dd->dd_phys->dd_used_bytes;
1083 if (!ondiskonly)
1084 used += dsl_dir_space_towrite(dd);
1085
1086 if (dd->dd_parent == NULL) {
1087 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1088 quota = MIN(quota, poolsize);
1089 }
1090
1091 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
1092 /*
1093 * We have some space reserved, in addition to what our
1094 * parent gave us.
1095 */
1096 parentspace += dd->dd_phys->dd_reserved - used;
1097 }
1098
1099 if (dd == ancestor) {
1100 ASSERT(delta <= 0);
1101 ASSERT(used >= -delta);
1102 used += delta;
1103 if (parentspace != UINT64_MAX)
1104 parentspace -= delta;
1105 }
1106
1107 if (used > quota) {
1108 /* over quota */
1109 myspace = 0;
1110 } else {
1111 /*
1112 * the lesser of the space provided by our parent and
1113 * the space left in our quota
1114 */
1115 myspace = MIN(parentspace, quota - used);
1116 }
1117
1118 mutex_exit(&dd->dd_lock);
1119
1120 return (myspace);
1121 }
1122
1123 struct tempreserve {
1124 list_node_t tr_node;
1125 dsl_pool_t *tr_dp;
1126 dsl_dir_t *tr_ds;
1127 uint64_t tr_size;
1128 };
1129
1130 static int
1131 dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1132 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1133 dmu_tx_t *tx, boolean_t first)
1134 {
1135 uint64_t txg = tx->tx_txg;
1136 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1137 uint64_t deferred = 0;
1138 struct tempreserve *tr;
1139 int retval = EDQUOT;
1140 int txgidx = txg & TXG_MASK;
1141 int i;
1142 uint64_t ref_rsrv = 0;
1143
1144 ASSERT3U(txg, !=, 0);
1145 ASSERT3S(asize, >, 0);
1146
1147 mutex_enter(&dd->dd_lock);
1148
1149 /*
1150 * Check against the dsl_dir's quota. We don't add in the delta
1151 * when checking for over-quota because they get one free hit.
1152 */
1153 est_inflight = dsl_dir_space_towrite(dd);
1154 for (i = 0; i < TXG_SIZE; i++)
1155 est_inflight += dd->dd_tempreserved[i];
1156 used_on_disk = dd->dd_phys->dd_used_bytes;
1157
1158 /*
1159 * On the first iteration, fetch the dataset's used-on-disk and
1160 * refreservation values. Also, if checkrefquota is set, test if
1161 * allocating this space would exceed the dataset's refquota.
1162 */
1163 if (first && tx->tx_objset) {
1164 int error;
1165 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1166
1167 error = dsl_dataset_check_quota(ds, checkrefquota,
1168 asize, est_inflight, &used_on_disk, &ref_rsrv);
1169 if (error) {
1170 mutex_exit(&dd->dd_lock);
1171 return (error);
1172 }
1173 }
1174
1175 /*
1176 * If this transaction will result in a net free of space,
1177 * we want to let it through.
1178 */
1179 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
1180 quota = UINT64_MAX;
1181 else
1182 quota = dd->dd_phys->dd_quota;
1183
1184 /*
1185 * Adjust the quota against the actual pool size at the root
1186 * minus any outstanding deferred frees.
1187 * To ensure that it's possible to remove files from a full
1188 * pool without inducing transient overcommits, we throttle
1189 * netfree transactions against a quota that is slightly larger,
1190 * but still within the pool's allocation slop. In cases where
1191 * we're very close to full, this will allow a steady trickle of
1192 * removes to get through.
1193 */
1194 if (dd->dd_parent == NULL) {
1195 spa_t *spa = dd->dd_pool->dp_spa;
1196 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1197 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1198 if (poolsize - deferred < quota) {
1199 quota = poolsize - deferred;
1200 retval = ENOSPC;
1201 }
1202 }
1203
1204 /*
1205 * If they are requesting more space, and our current estimate
1206 * is over quota, they get to try again unless the actual
1207 * on-disk is over quota and there are no pending changes (which
1208 * may free up space for us).
1209 */
1210 if (used_on_disk + est_inflight >= quota) {
1211 if (est_inflight > 0 || used_on_disk < quota ||
1212 (retval == ENOSPC && used_on_disk < quota + deferred))
1213 retval = ERESTART;
1214 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1215 "quota=%lluK tr=%lluK err=%d\n",
1216 used_on_disk>>10, est_inflight>>10,
1217 quota>>10, asize>>10, retval);
1218 mutex_exit(&dd->dd_lock);
1219 return (retval);
1220 }
1221
1222 /* We need to up our estimated delta before dropping dd_lock */
1223 dd->dd_tempreserved[txgidx] += asize;
1224
1225 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1226 asize - ref_rsrv);
1227 mutex_exit(&dd->dd_lock);
1228
1229 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1230 tr->tr_ds = dd;
1231 tr->tr_size = asize;
1232 list_insert_tail(tr_list, tr);
1233
1234 /* see if it's OK with our parent */
1235 if (dd->dd_parent && parent_rsrv) {
1236 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
1237
1238 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1239 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1240 } else {
1241 return (0);
1242 }
1243 }
1244
1245 /*
1246 * Reserve space in this dsl_dir, to be used in this tx's txg.
1247 * After the space has been dirtied (and dsl_dir_willuse_space()
1248 * has been called), the reservation should be canceled, using
1249 * dsl_dir_tempreserve_clear().
1250 */
1251 int
1252 dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1253 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1254 {
1255 int err;
1256 list_t *tr_list;
1257
1258 if (asize == 0) {
1259 *tr_cookiep = NULL;
1260 return (0);
1261 }
1262
1263 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1264 list_create(tr_list, sizeof (struct tempreserve),
1265 offsetof(struct tempreserve, tr_node));
1266 ASSERT3S(asize, >, 0);
1267 ASSERT3S(fsize, >=, 0);
1268
1269 err = arc_tempreserve_space(lsize, tx->tx_txg);
1270 if (err == 0) {
1271 struct tempreserve *tr;
1272
1273 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1274 tr->tr_size = lsize;
1275 list_insert_tail(tr_list, tr);
1276
1277 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
1278 } else {
1279 if (err == EAGAIN) {
1280 txg_delay(dd->dd_pool, tx->tx_txg,
1281 zfs_zone_txg_delay());
1282 err = ERESTART;
1283 }
1284 dsl_pool_memory_pressure(dd->dd_pool);
1285 }
1286
1287 if (err == 0) {
1288 struct tempreserve *tr;
1289
1290 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1291 tr->tr_dp = dd->dd_pool;
1292 tr->tr_size = asize;
1293 list_insert_tail(tr_list, tr);
1294
1295 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1296 FALSE, asize > usize, tr_list, tx, TRUE);
1297 }
1298
1299 if (err)
1300 dsl_dir_tempreserve_clear(tr_list, tx);
1301 else
1302 *tr_cookiep = tr_list;
1303
1304 return (err);
1305 }
1306
1307 /*
1308 * Clear a temporary reservation that we previously made with
1309 * dsl_dir_tempreserve_space().
1310 */
1311 void
1312 dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1313 {
1314 int txgidx = tx->tx_txg & TXG_MASK;
1315 list_t *tr_list = tr_cookie;
1316 struct tempreserve *tr;
1317
1318 ASSERT3U(tx->tx_txg, !=, 0);
1319
1320 if (tr_cookie == NULL)
1321 return;
1322
1323 while (tr = list_head(tr_list)) {
1324 if (tr->tr_dp) {
1325 dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
1326 } else if (tr->tr_ds) {
1327 mutex_enter(&tr->tr_ds->dd_lock);
1328 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1329 tr->tr_size);
1330 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1331 mutex_exit(&tr->tr_ds->dd_lock);
1332 } else {
1333 arc_tempreserve_clear(tr->tr_size);
1334 }
1335 list_remove(tr_list, tr);
1336 kmem_free(tr, sizeof (struct tempreserve));
1337 }
1338
1339 kmem_free(tr_list, sizeof (list_t));
1340 }
1341
1342 static void
1343 dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1344 {
1345 int64_t parent_space;
1346 uint64_t est_used;
1347
1348 mutex_enter(&dd->dd_lock);
1349 if (space > 0)
1350 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1351
1352 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
1353 parent_space = parent_delta(dd, est_used, space);
1354 mutex_exit(&dd->dd_lock);
1355
1356 /* Make sure that we clean up dd_space_to* */
1357 dsl_dir_dirty(dd, tx);
1358
1359 /* XXX this is potentially expensive and unnecessary... */
1360 if (parent_space && dd->dd_parent)
1361 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
1362 }
1363
1364 /*
1365 * Call in open context when we think we're going to write/free space,
1366 * eg. when dirtying data. Be conservative (ie. OK to write less than
1367 * this or free more than this, but don't write more or free less).
1368 */
1369 void
1370 dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1371 {
1372 dsl_pool_willuse_space(dd->dd_pool, space, tx);
1373 dsl_dir_willuse_space_impl(dd, space, tx);
1374 }
1375
1376 /* call from syncing context when we actually write/free space for this dd */
1377 void
1378 dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1379 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1380 {
1381 int64_t accounted_delta;
1382 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1383
1384 ASSERT(dmu_tx_is_syncing(tx));
1385 ASSERT(type < DD_USED_NUM);
1386
1387 if (needlock)
1388 mutex_enter(&dd->dd_lock);
1389 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
1390 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
1391 ASSERT(compressed >= 0 ||
1392 dd->dd_phys->dd_compressed_bytes >= -compressed);
1393 ASSERT(uncompressed >= 0 ||
1394 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
1395 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1396 dd->dd_phys->dd_used_bytes += used;
1397 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
1398 dd->dd_phys->dd_compressed_bytes += compressed;
1399
1400 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1401 ASSERT(used > 0 ||
1402 dd->dd_phys->dd_used_breakdown[type] >= -used);
1403 dd->dd_phys->dd_used_breakdown[type] += used;
1404 #ifdef DEBUG
1405 dd_used_t t;
1406 uint64_t u = 0;
1407 for (t = 0; t < DD_USED_NUM; t++)
1408 u += dd->dd_phys->dd_used_breakdown[t];
1409 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
1410 #endif
1411 }
1412 if (needlock)
1413 mutex_exit(&dd->dd_lock);
1414
1415 if (dd->dd_parent != NULL) {
1416 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1417 accounted_delta, compressed, uncompressed, tx);
1418 dsl_dir_transfer_space(dd->dd_parent,
1419 used - accounted_delta,
1420 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1421 }
1422 }
1423
1424 void
1425 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1426 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1427 {
1428 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1429
1430 ASSERT(dmu_tx_is_syncing(tx));
1431 ASSERT(oldtype < DD_USED_NUM);
1432 ASSERT(newtype < DD_USED_NUM);
1433
1434 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
1435 return;
1436
1437 if (needlock)
1438 mutex_enter(&dd->dd_lock);
1439 ASSERT(delta > 0 ?
1440 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1441 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1442 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1443 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1444 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1445 dd->dd_phys->dd_used_breakdown[newtype] += delta;
1446 if (needlock)
1447 mutex_exit(&dd->dd_lock);
1448 }
1449
1450 static int
1451 dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
1452 {
1453 dsl_dataset_t *ds = arg1;
1454 dsl_dir_t *dd = ds->ds_dir;
1455 dsl_prop_setarg_t *psa = arg2;
1456 int err;
1457 uint64_t towrite;
1458
1459 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1460 return (err);
1461
1462 if (psa->psa_effective_value == 0)
1463 return (0);
1464
1465 mutex_enter(&dd->dd_lock);
1466 /*
1467 * If we are doing the preliminary check in open context, and
1468 * there are pending changes, then don't fail it, since the
1469 * pending changes could under-estimate the amount of space to be
1470 * freed up.
1471 */
1472 towrite = dsl_dir_space_towrite(dd);
1473 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1474 (psa->psa_effective_value < dd->dd_phys->dd_reserved ||
1475 psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) {
1476 err = ENOSPC;
1477 }
1478 mutex_exit(&dd->dd_lock);
1479 return (err);
1480 }
1481
1482 static void
1483 dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1484 {
1485 dsl_dataset_t *ds = arg1;
1486 dsl_dir_t *dd = ds->ds_dir;
1487 dsl_prop_setarg_t *psa = arg2;
1488 uint64_t effective_value = psa->psa_effective_value;
1489
1490 dsl_prop_set_sync(ds, psa, tx);
1491 DSL_PROP_CHECK_PREDICTION(dd, psa);
1492
1493 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1494
1495 mutex_enter(&dd->dd_lock);
1496 dd->dd_phys->dd_quota = effective_value;
1497 mutex_exit(&dd->dd_lock);
1498 }
1499
1500 int
1501 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1502 {
1503 dsl_dir_t *dd;
1504 dsl_dataset_t *ds;
1505 dsl_prop_setarg_t psa;
1506 int err;
1507
1508 dsl_prop_setarg_init_uint64(&psa, "quota", source, "a);
1509
1510 err = dsl_dataset_hold(ddname, FTAG, &ds);
1511 if (err)
1512 return (err);
1513
1514 err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1515 if (err) {
1516 dsl_dataset_rele(ds, FTAG);
1517 return (err);
1518 }
1519
1520 ASSERT(ds->ds_dir == dd);
1521
1522 /*
1523 * If someone removes a file, then tries to set the quota, we want to
1524 * make sure the file freeing takes effect.
1525 */
1526 txg_wait_open(dd->dd_pool, 0);
1527
1528 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
1529 dsl_dir_set_quota_sync, ds, &psa, 0);
1530
1531 dsl_dir_close(dd, FTAG);
1532 dsl_dataset_rele(ds, FTAG);
1533 return (err);
1534 }
1535
1536 int
1537 dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
1538 {
1539 dsl_dataset_t *ds = arg1;
1540 dsl_dir_t *dd = ds->ds_dir;
1541 dsl_prop_setarg_t *psa = arg2;
1542 uint64_t effective_value;
1543 uint64_t used, avail;
1544 int err;
1545
1546 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
1547 return (err);
1548
1549 effective_value = psa->psa_effective_value;
1550
1551 /*
1552 * If we are doing the preliminary check in open context, the
1553 * space estimates may be inaccurate.
1554 */
1555 if (!dmu_tx_is_syncing(tx))
1556 return (0);
1557
1558 mutex_enter(&dd->dd_lock);
1559 used = dd->dd_phys->dd_used_bytes;
1560 mutex_exit(&dd->dd_lock);
1561
1562 if (dd->dd_parent) {
1563 avail = dsl_dir_space_available(dd->dd_parent,
1564 NULL, 0, FALSE);
1565 } else {
1566 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1567 }
1568
1569 if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) {
1570 uint64_t delta = MAX(used, effective_value) -
1571 MAX(used, dd->dd_phys->dd_reserved);
1572
1573 if (delta > avail)
1574 return (ENOSPC);
1575 if (dd->dd_phys->dd_quota > 0 &&
1576 effective_value > dd->dd_phys->dd_quota)
1577 return (ENOSPC);
1578 }
1579
1580 return (0);
1581 }
1582
1583 static void
1584 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1585 {
1586 uint64_t used;
1587 int64_t delta;
1588
1589 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1590
1591 mutex_enter(&dd->dd_lock);
1592 used = dd->dd_phys->dd_used_bytes;
1593 delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1594 dd->dd_phys->dd_reserved = value;
1595
1596 if (dd->dd_parent != NULL) {
1597 /* Roll up this additional usage into our ancestors */
1598 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1599 delta, 0, 0, tx);
1600 }
1601 mutex_exit(&dd->dd_lock);
1602 }
1603
1604
1605 static void
1606 dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1607 {
1608 dsl_dataset_t *ds = arg1;
1609 dsl_dir_t *dd = ds->ds_dir;
1610 dsl_prop_setarg_t *psa = arg2;
1611 uint64_t value = psa->psa_effective_value;
1612
1613 dsl_prop_set_sync(ds, psa, tx);
1614 DSL_PROP_CHECK_PREDICTION(dd, psa);
1615
1616 dsl_dir_set_reservation_sync_impl(dd, value, tx);
1617 }
1618
1619 int
1620 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1621 uint64_t reservation)
1622 {
1623 dsl_dir_t *dd;
1624 dsl_dataset_t *ds;
1625 dsl_prop_setarg_t psa;
1626 int err;
1627
1628 dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation);
1629
1630 err = dsl_dataset_hold(ddname, FTAG, &ds);
1631 if (err)
1632 return (err);
1633
1634 err = dsl_dir_open(ddname, FTAG, &dd, NULL);
1635 if (err) {
1636 dsl_dataset_rele(ds, FTAG);
1637 return (err);
1638 }
1639
1640 ASSERT(ds->ds_dir == dd);
1641
1642 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
1643 dsl_dir_set_reservation_sync, ds, &psa, 0);
1644
1645 dsl_dir_close(dd, FTAG);
1646 dsl_dataset_rele(ds, FTAG);
1647 return (err);
1648 }
1649
1650 static dsl_dir_t *
1651 closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1652 {
1653 for (; ds1; ds1 = ds1->dd_parent) {
1654 dsl_dir_t *dd;
1655 for (dd = ds2; dd; dd = dd->dd_parent) {
1656 if (ds1 == dd)
1657 return (dd);
1658 }
1659 }
1660 return (NULL);
1661 }
1662
1663 /*
1664 * If delta is applied to dd, how much of that delta would be applied to
1665 * ancestor? Syncing context only.
1666 */
1667 static int64_t
1668 would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1669 {
1670 if (dd == ancestor)
1671 return (delta);
1672
1673 mutex_enter(&dd->dd_lock);
1674 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1675 mutex_exit(&dd->dd_lock);
1676 return (would_change(dd->dd_parent, delta, ancestor));
1677 }
1678
1679 struct renamearg {
1680 dsl_dir_t *newparent;
1681 const char *mynewname;
1682 };
1683
1684 static int
1685 dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1686 {
1687 dsl_dir_t *dd = arg1;
1688 struct renamearg *ra = arg2;
1689 dsl_pool_t *dp = dd->dd_pool;
1690 objset_t *mos = dp->dp_meta_objset;
1691 int err;
1692 uint64_t val;
1693
1694 /*
1695 * There should only be one reference, from dmu_objset_rename().
1696 * Fleeting holds are also possible (eg, from "zfs list" getting
1697 * stats), but any that are present in open context will likely
1698 * be gone by syncing context, so only fail from syncing
1699 * context.
1700 */
1701 if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 1)
1702 return (EBUSY);
1703
1704 /* check for existing name */
1705 err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1706 ra->mynewname, 8, 1, &val);
1707 if (err == 0)
1708 return (EEXIST);
1709 if (err != ENOENT)
1710 return (err);
1711
1712 if (ra->newparent != dd->dd_parent) {
1713 /* is there enough space? */
1714 uint64_t myspace =
1715 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1716
1717 /* no rename into our descendant */
1718 if (closest_common_ancestor(dd, ra->newparent) == dd)
1719 return (EINVAL);
1720
1721 if (err = dsl_dir_transfer_possible(dd->dd_parent,
1722 ra->newparent, dd, myspace, tx))
1723 return (err);
1724 }
1725
1726 return (0);
1727 }
1728
1729 static void
1730 dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1731 {
1732 dsl_dir_t *dd = arg1;
1733 struct renamearg *ra = arg2;
1734 dsl_pool_t *dp = dd->dd_pool;
1735 objset_t *mos = dp->dp_meta_objset;
1736 int err;
1737 char namebuf[MAXNAMELEN];
1738
1739 ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2);
1740
1741 /* Log this before we change the name. */
1742 dsl_dir_name(ra->newparent, namebuf);
1743 spa_history_log_internal_dd(dd, "rename", tx,
1744 "-> %s/%s", namebuf, ra->mynewname);
1745
1746 if (ra->newparent != dd->dd_parent) {
1747 int cnt;
1748
1749 mutex_enter(&dd->dd_lock);
1750
1751 cnt = dd->dd_phys->dd_filesystem_count;
1752 dsl_dir_fscount_adjust(dd->dd_parent, tx, -cnt, B_TRUE, B_TRUE);
1753 dsl_dir_fscount_adjust(ra->newparent, tx, cnt, B_TRUE, B_TRUE);
1754
1755 cnt = dd->dd_phys->dd_snapshot_count;
1756 dsl_snapcount_adjust(dd->dd_parent, tx, -cnt, B_TRUE);
1757 dsl_snapcount_adjust(ra->newparent, tx, cnt, B_TRUE);
1758
1759 mutex_exit(&dd->dd_lock);
1760
1761 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1762 -dd->dd_phys->dd_used_bytes,
1763 -dd->dd_phys->dd_compressed_bytes,
1764 -dd->dd_phys->dd_uncompressed_bytes, tx);
1765 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
1766 dd->dd_phys->dd_used_bytes,
1767 dd->dd_phys->dd_compressed_bytes,
1768 dd->dd_phys->dd_uncompressed_bytes, tx);
1769
1770 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1771 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1772 dd->dd_phys->dd_used_bytes;
1773
1774 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1775 -unused_rsrv, 0, 0, tx);
1776 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
1777 unused_rsrv, 0, 0, tx);
1778 }
1779 }
1780
1781 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1782
1783 /* remove from old parent zapobj */
1784 err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1785 dd->dd_myname, tx);
1786 ASSERT0(err);
1787
1788 (void) strcpy(dd->dd_myname, ra->mynewname);
1789 dsl_dir_close(dd->dd_parent, dd);
1790 dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
1791 VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
1792 ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
1793
1794 /* add to new parent zapobj */
1795 err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
1796 dd->dd_myname, 8, 1, &dd->dd_object, tx);
1797 ASSERT0(err);
1798
1799 }
1800
1801 int
1802 dsl_dir_rename(dsl_dir_t *dd, const char *newname)
1803 {
1804 struct renamearg ra;
1805 int err;
1806
1807 /* new parent should exist */
1808 err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
1809 if (err)
1810 return (err);
1811
1812 /* can't rename to different pool */
1813 if (dd->dd_pool != ra.newparent->dd_pool) {
1814 err = ENXIO;
1815 goto out;
1816 }
1817
1818 /* new name should not already exist */
1819 if (ra.mynewname == NULL) {
1820 err = EEXIST;
1821 goto out;
1822 }
1823
1824 err = dsl_sync_task_do(dd->dd_pool,
1825 dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
1826
1827 out:
1828 dsl_dir_close(ra.newparent, FTAG);
1829 return (err);
1830 }
1831
1832 int
1833 dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, dsl_dir_t *moving_dd,
1834 uint64_t space, dmu_tx_t *tx)
1835 {
1836 dsl_dir_t *ancestor;
1837 int64_t adelta;
1838 uint64_t avail;
1839 int err;
1840
1841 ancestor = closest_common_ancestor(sdd, tdd);
1842 adelta = would_change(sdd, -space, ancestor);
1843 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1844 if (avail < space)
1845 return (ENOSPC);
1846
1847 if (sdd != moving_dd) {
1848 err = dsl_dir_fscount_check(tdd,
1849 moving_dd->dd_phys->dd_filesystem_count, ancestor);
1850 if (err != 0)
1851 return (err);
1852 }
1853 err = dsl_snapcount_check(tdd, moving_dd->dd_phys->dd_snapshot_count,
1854 ancestor);
1855 if (err != 0)
1856 return (err);
1857
1858 return (0);
1859 }
1860
1861 timestruc_t
1862 dsl_dir_snap_cmtime(dsl_dir_t *dd)
1863 {
1864 timestruc_t t;
1865
1866 mutex_enter(&dd->dd_lock);
1867 t = dd->dd_snap_cmtime;
1868 mutex_exit(&dd->dd_lock);
1869
1870 return (t);
1871 }
1872
1873 void
1874 dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1875 {
1876 timestruc_t t;
1877
1878 gethrestime(&t);
1879 mutex_enter(&dd->dd_lock);
1880 dd->dd_snap_cmtime = t;
1881 mutex_exit(&dd->dd_lock);
1882 }