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) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2013 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
27 */
28
29#include <sys/dmu.h>
30#include <sys/dmu_objset.h>
31#include <sys/dmu_tx.h>
32#include <sys/dsl_dataset.h>
33#include <sys/dsl_dir.h>
34#include <sys/dsl_prop.h>
35#include <sys/dsl_synctask.h>
36#include <sys/dsl_deleg.h>
37#include <sys/dmu_impl.h>
38#include <sys/spa.h>
39#include <sys/metaslab.h>
40#include <sys/zap.h>
41#include <sys/zio.h>
42#include <sys/arc.h>
43#include <sys/sunddi.h>
44#include <sys/zvol.h>
45#ifdef _KERNEL
46#include <sys/zfs_vfsops.h>
47#endif
48#include <sys/zfeature.h>
49#include <sys/policy.h>
50#include <sys/zfs_znode.h>
51#include "zfs_namecheck.h"
52#include "zfs_prop.h"
53
54/*
55 * Filesystem and Snapshot Limits
56 * ------------------------------
57 *
58 * These limits are used to restrict the number of filesystems and/or snapshots
59 * that can be created at a given level in the tree or below. A typical
60 * use-case is with a delegated dataset where the administrator wants to ensure
61 * that a user within the zone is not creating too many additional filesystems
62 * or snapshots, even though they're not exceeding their space quota.
63 *
64 * The filesystem and snapshot counts are stored as extensible properties. This
65 * capability is controlled by a feature flag and must be enabled to be used.
66 * Once enabled, the feature is not active until the first limit is set. At
67 * that point, future operations to create/destroy filesystems or snapshots
68 * will validate and update the counts.
69 *
70 * Because the count properties will not exist before the feature is active,
71 * the counts are updated when a limit is first set on an uninitialized
72 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
73 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
74 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
75 * snapshot count properties on a node indicate uninitialized counts on that
76 * node.) When first setting a limit on an uninitialized node, the code starts
77 * at the filesystem with the new limit and descends into all sub-filesystems
78 * to add the count properties.
79 *
80 * In practice this is lightweight since a limit is typically set when the
81 * filesystem is created and thus has no children. Once valid, changing the
82 * limit value won't require a re-traversal since the counts are already valid.
83 * When recursively fixing the counts, if a node with a limit is encountered
84 * during the descent, the counts are known to be valid and there is no need to
85 * descend into that filesystem's children. The counts on filesystems above the
86 * one with the new limit will still be uninitialized, unless a limit is
87 * eventually set on one of those filesystems. The counts are always recursively
88 * updated when a limit is set on a dataset, unless there is already a limit.
89 * When a new limit value is set on a filesystem with an existing limit, it is
90 * possible for the new limit to be less than the current count at that level
91 * since a user who can change the limit is also allowed to exceed the limit.
92 *
93 * Once the feature is active, then whenever a filesystem or snapshot is
94 * created, the code recurses up the tree, validating the new count against the
95 * limit at each initialized level. In practice, most levels will not have a
96 * limit set. If there is a limit at any initialized level up the tree, the
97 * check must pass or the creation will fail. Likewise, when a filesystem or
98 * snapshot is destroyed, the counts are recursively adjusted all the way up
99 * the initizized nodes in the tree. Renaming a filesystem into different point
100 * in the tree will first validate, then update the counts on each branch up to
101 * the common ancestor. A receive will also validate the counts and then update
102 * them.
103 *
104 * An exception to the above behavior is that the limit is not enforced if the
105 * user has permission to modify the limit. This is primarily so that
106 * recursive snapshots in the global zone always work. We want to prevent a
107 * denial-of-service in which a lower level delegated dataset could max out its
108 * limit and thus block recursive snapshots from being taken in the global zone.
109 * Because of this, it is possible for the snapshot count to be over the limit
110 * and snapshots taken in the global zone could cause a lower level dataset to
111 * hit or exceed its limit. The administrator taking the global zone recursive
112 * snapshot should be aware of this side-effect and behave accordingly.
113 * For consistency, the filesystem limit is also not enforced if the user can
114 * modify the limit.
115 *
116 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
117 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
118 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
119 * dsl_dir_init_fs_ss_count().
120 *
121 * There is a special case when we receive a filesystem that already exists. In
122 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
123 * never update the filesystem counts for temporary clones.
124 *
125 * Likewise, we do not update the snapshot counts for temporary snapshots,
126 * such as those created by zfs diff.
127 */
128
129static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
130
131/* ARGSUSED */
132static void
133dsl_dir_evict(dmu_buf_t *db, void *arg)
134{
135 dsl_dir_t *dd = arg;
136 dsl_pool_t *dp = dd->dd_pool;
137 int t;
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_rele(dd->dd_parent, dd);
147
148 spa_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 dd->dd_phys = dbuf->db_data;
189 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
190
191 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
192 offsetof(dsl_prop_cb_record_t, cbr_node));
193
194 dsl_dir_snap_cmtime_update(dd);
195
196 if (dd->dd_phys->dd_parent_obj) {
197 err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
198 NULL, dd, &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 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
207 tail, sizeof (foundobj), 1, &foundobj);
208 ASSERT(err || foundobj == ddobj);
209#endif
210 (void) strcpy(dd->dd_myname, tail);
211 } else {
212 err = zap_value_search(dp->dp_meta_objset,
213 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
214 ddobj, 0, dd->dd_myname);
215 }
216 if (err != 0)
217 goto errout;
218 } else {
219 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
220 }
221
222 if (dsl_dir_is_clone(dd)) {
223 dmu_buf_t *origin_bonus;
224 dsl_dataset_phys_t *origin_phys;
225
226 /*
227 * We can't open the origin dataset, because
228 * that would require opening this dsl_dir.
229 * Just look at its phys directly instead.
230 */
231 err = dmu_bonus_hold(dp->dp_meta_objset,
232 dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
233 if (err != 0)
234 goto errout;
235 origin_phys = origin_bonus->db_data;
236 dd->dd_origin_txg =
237 origin_phys->ds_creation_txg;
238 dmu_buf_rele(origin_bonus, FTAG);
239 }
240
241 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
242 dsl_dir_evict);
243 if (winner) {
244 if (dd->dd_parent)
245 dsl_dir_rele(dd->dd_parent, dd);
246 mutex_destroy(&dd->dd_lock);
247 kmem_free(dd, sizeof (dsl_dir_t));
248 dd = winner;
249 } else {
250 spa_open_ref(dp->dp_spa, dd);
251 }
252 }
253
254 /*
255 * The dsl_dir_t has both open-to-close and instantiate-to-evict
256 * holds on the spa. We need the open-to-close holds because
257 * otherwise the spa_refcnt wouldn't change when we open a
258 * dir which the spa also has open, so we could incorrectly
259 * think it was OK to unload/export/destroy the pool. We need
260 * the instantiate-to-evict hold because the dsl_dir_t has a
261 * pointer to the dd_pool, which has a pointer to the spa_t.
262 */
263 spa_open_ref(dp->dp_spa, tag);
264 ASSERT3P(dd->dd_pool, ==, dp);
265 ASSERT3U(dd->dd_object, ==, ddobj);
266 ASSERT3P(dd->dd_dbuf, ==, dbuf);
267 *ddp = dd;
268 return (0);
269
270errout:
271 if (dd->dd_parent)
272 dsl_dir_rele(dd->dd_parent, dd);
273 mutex_destroy(&dd->dd_lock);
274 kmem_free(dd, sizeof (dsl_dir_t));
275 dmu_buf_rele(dbuf, tag);
276 return (err);
277}
278
279void
280dsl_dir_rele(dsl_dir_t *dd, void *tag)
281{
282 dprintf_dd(dd, "%s\n", "");
283 spa_close(dd->dd_pool->dp_spa, tag);
284 dmu_buf_rele(dd->dd_dbuf, tag);
285}
286
287/* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
288void
289dsl_dir_name(dsl_dir_t *dd, char *buf)
290{
291 if (dd->dd_parent) {
292 dsl_dir_name(dd->dd_parent, buf);
293 (void) strcat(buf, "/");
294 } else {
295 buf[0] = '\0';
296 }
297 if (!MUTEX_HELD(&dd->dd_lock)) {
298 /*
299 * recursive mutex so that we can use
300 * dprintf_dd() with dd_lock held
301 */
302 mutex_enter(&dd->dd_lock);
303 (void) strcat(buf, dd->dd_myname);
304 mutex_exit(&dd->dd_lock);
305 } else {
306 (void) strcat(buf, dd->dd_myname);
307 }
308}
309
310/* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
311int
312dsl_dir_namelen(dsl_dir_t *dd)
313{
314 int result = 0;
315
316 if (dd->dd_parent) {
317 /* parent's name + 1 for the "/" */
318 result = dsl_dir_namelen(dd->dd_parent) + 1;
319 }
320
321 if (!MUTEX_HELD(&dd->dd_lock)) {
322 /* see dsl_dir_name */
323 mutex_enter(&dd->dd_lock);
324 result += strlen(dd->dd_myname);
325 mutex_exit(&dd->dd_lock);
326 } else {
327 result += strlen(dd->dd_myname);
328 }
329
330 return (result);
331}
332
333static int
334getcomponent(const char *path, char *component, const char **nextp)
335{
336 char *p;
337
338 if ((path == NULL) || (path[0] == '\0'))
339 return (SET_ERROR(ENOENT));
340 /* This would be a good place to reserve some namespace... */
341 p = strpbrk(path, "/@");
342 if (p && (p[1] == '/' || p[1] == '@')) {
343 /* two separators in a row */
344 return (SET_ERROR(EINVAL));
345 }
346 if (p == NULL || p == path) {
347 /*
348 * if the first thing is an @ or /, it had better be an
349 * @ and it had better not have any more ats or slashes,
350 * and it had better have something after the @.
351 */
352 if (p != NULL &&
353 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
354 return (SET_ERROR(EINVAL));
355 if (strlen(path) >= MAXNAMELEN)
356 return (SET_ERROR(ENAMETOOLONG));
357 (void) strcpy(component, path);
358 p = NULL;
359 } else if (p[0] == '/') {
360 if (p - path >= MAXNAMELEN)
361 return (SET_ERROR(ENAMETOOLONG));
362 (void) strncpy(component, path, p - path);
363 component[p - path] = '\0';
364 p++;
365 } else if (p[0] == '@') {
366 /*
367 * if the next separator is an @, there better not be
368 * any more slashes.
369 */
370 if (strchr(path, '/'))
371 return (SET_ERROR(EINVAL));
372 if (p - path >= MAXNAMELEN)
373 return (SET_ERROR(ENAMETOOLONG));
374 (void) strncpy(component, path, p - path);
375 component[p - path] = '\0';
376 } else {
377 panic("invalid p=%p", (void *)p);
378 }
379 *nextp = p;
380 return (0);
381}
382
383/*
384 * Return the dsl_dir_t, and possibly the last component which couldn't
385 * be found in *tail. The name must be in the specified dsl_pool_t. This
386 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
387 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
388 * (*tail)[0] == '@' means that the last component is a snapshot.
389 */
390int
391dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
392 dsl_dir_t **ddp, const char **tailp)
393{
394 char buf[MAXNAMELEN];
395 const char *spaname, *next, *nextnext = NULL;
396 int err;
397 dsl_dir_t *dd;
398 uint64_t ddobj;
399
400 err = getcomponent(name, buf, &next);
401 if (err != 0)
402 return (err);
403
404 /* Make sure the name is in the specified pool. */
405 spaname = spa_name(dp->dp_spa);
406 if (strcmp(buf, spaname) != 0)
407 return (SET_ERROR(EINVAL));
408
409 ASSERT(dsl_pool_config_held(dp));
410
411 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
412 if (err != 0) {
413 return (err);
414 }
415
416 while (next != NULL) {
417 dsl_dir_t *child_ds;
418 err = getcomponent(next, buf, &nextnext);
419 if (err != 0)
420 break;
421 ASSERT(next[0] != '\0');
422 if (next[0] == '@')
423 break;
424 dprintf("looking up %s in obj%lld\n",
425 buf, dd->dd_phys->dd_child_dir_zapobj);
426
427 err = zap_lookup(dp->dp_meta_objset,
428 dd->dd_phys->dd_child_dir_zapobj,
429 buf, sizeof (ddobj), 1, &ddobj);
430 if (err != 0) {
431 if (err == ENOENT)
432 err = 0;
433 break;
434 }
435
436 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
437 if (err != 0)
438 break;
439 dsl_dir_rele(dd, tag);
440 dd = child_ds;
441 next = nextnext;
442 }
443
444 if (err != 0) {
445 dsl_dir_rele(dd, tag);
446 return (err);
447 }
448
449 /*
450 * It's an error if there's more than one component left, or
451 * tailp==NULL and there's any component left.
452 */
453 if (next != NULL &&
454 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
455 /* bad path name */
456 dsl_dir_rele(dd, tag);
457 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
458 err = SET_ERROR(ENOENT);
459 }
460 if (tailp != NULL)
461 *tailp = next;
462 *ddp = dd;
463 return (err);
464}
465
466/*
467 * If the counts are already initialized for this filesystem and its
468 * descendants then do nothing, otherwise initialize the counts.
469 *
470 * The counts on this filesystem, and those below, may be uninitialized due to
471 * either the use of a pre-existing pool which did not support the
472 * filesystem/snapshot limit feature, or one in which the feature had not yet
473 * been enabled.
474 *
475 * Recursively descend the filesystem tree and update the filesystem/snapshot
476 * counts on each filesystem below, then update the cumulative count on the
477 * current filesystem. If the filesystem already has a count set on it,
478 * then we know that its counts, and the counts on the filesystems below it,
479 * are already correct, so we don't have to update this filesystem.
480 */
481static void
482dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
483{
484 uint64_t my_fs_cnt = 0;
485 uint64_t my_ss_cnt = 0;
486 dsl_pool_t *dp = dd->dd_pool;
487 objset_t *os = dp->dp_meta_objset;
488 zap_cursor_t *zc;
489 zap_attribute_t *za;
490 dsl_dataset_t *ds;
491
492 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
493 ASSERT(dsl_pool_config_held(dp));
494 ASSERT(dmu_tx_is_syncing(tx));
495
496 dsl_dir_zapify(dd, tx);
497
498 /*
499 * If the filesystem count has already been initialized then we
500 * don't need to recurse down any further.
501 */
502 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
503 return;
504
505 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
506 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
507
508 /* Iterate my child dirs */
509 for (zap_cursor_init(zc, os, dd->dd_phys->dd_child_dir_zapobj);
510 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
511 dsl_dir_t *chld_dd;
512 uint64_t count;
513
514 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
515 &chld_dd));
516
517 /*
518 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
519 * temporary datasets.
520 */
521 if (chld_dd->dd_myname[0] == '$' ||
522 chld_dd->dd_myname[0] == '%') {
523 dsl_dir_rele(chld_dd, FTAG);
524 continue;
525 }
526
527 my_fs_cnt++; /* count this child */
528
529 dsl_dir_init_fs_ss_count(chld_dd, tx);
530
531 VERIFY0(zap_lookup(os, chld_dd->dd_object,
532 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
533 my_fs_cnt += count;
534 VERIFY0(zap_lookup(os, chld_dd->dd_object,
535 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
536 my_ss_cnt += count;
537
538 dsl_dir_rele(chld_dd, FTAG);
539 }
540 zap_cursor_fini(zc);
541 /* Count my snapshots (we counted children's snapshots above) */
542 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
543 dd->dd_phys->dd_head_dataset_obj, FTAG, &ds));
544
545 for (zap_cursor_init(zc, os, ds->ds_phys->ds_snapnames_zapobj);
546 zap_cursor_retrieve(zc, za) == 0;
547 zap_cursor_advance(zc)) {
548 /* Don't count temporary snapshots */
549 if (za->za_name[0] != '%')
550 my_ss_cnt++;
551 }
552 zap_cursor_fini(zc);
553
554 dsl_dataset_rele(ds, FTAG);
555
556 kmem_free(zc, sizeof (zap_cursor_t));
557 kmem_free(za, sizeof (zap_attribute_t));
558
559 /* we're in a sync task, update counts */
560 dmu_buf_will_dirty(dd->dd_dbuf, tx);
561 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
562 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
563 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
564 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
565}
566
567static int
568dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
569{
570 char *ddname = (char *)arg;
571 dsl_pool_t *dp = dmu_tx_pool(tx);
572 dsl_dataset_t *ds;
573 dsl_dir_t *dd;
574 int error;
575
576 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
577 if (error != 0)
578 return (error);
579
580 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
581 dsl_dataset_rele(ds, FTAG);
582 return (SET_ERROR(ENOTSUP));
583 }
584
585 dd = ds->ds_dir;
586 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
587 dsl_dir_is_zapified(dd) &&
588 zap_contains(dp->dp_meta_objset, dd->dd_object,
589 DD_FIELD_FILESYSTEM_COUNT) == 0) {
590 dsl_dataset_rele(ds, FTAG);
591 return (SET_ERROR(EALREADY));
592 }
593
594 dsl_dataset_rele(ds, FTAG);
595 return (0);
596}
597
598static void
599dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
600{
601 char *ddname = (char *)arg;
602 dsl_pool_t *dp = dmu_tx_pool(tx);
603 dsl_dataset_t *ds;
604 spa_t *spa;
605
606 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
607
608 spa = dsl_dataset_get_spa(ds);
609
610 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
611 /*
612 * Since the feature was not active and we're now setting a
613 * limit, increment the feature-active counter so that the
614 * feature becomes active for the first time.
615 *
616 * We are already in a sync task so we can update the MOS.
617 */
618 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
619 }
620
621 /*
622 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
623 * we need to ensure the counts are correct. Descend down the tree from
624 * this point and update all of the counts to be accurate.
625 */
626 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
627
628 dsl_dataset_rele(ds, FTAG);
629}
630
631/*
632 * Make sure the feature is enabled and activate it if necessary.
633 * Since we're setting a limit, ensure the on-disk counts are valid.
634 * This is only called by the ioctl path when setting a limit value.
635 *
636 * We do not need to validate the new limit, since users who can change the
637 * limit are also allowed to exceed the limit.
638 */
639int
640dsl_dir_activate_fs_ss_limit(const char *ddname)
641{
642 int error;
643
644 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
645 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0);
646
647 if (error == EALREADY)
648 error = 0;
649
650 return (error);
651}
652
653/*
654 * Used to determine if the filesystem_limit or snapshot_limit should be
655 * enforced. We allow the limit to be exceeded if the user has permission to
656 * write the property value. We pass in the creds that we got in the open
657 * context since we will always be the GZ root in syncing context. We also have
658 * to handle the case where we are allowed to change the limit on the current
659 * dataset, but there may be another limit in the tree above.
660 *
661 * We can never modify these two properties within a non-global zone. In
662 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
663 * can't use that function since we are already holding the dp_config_rwlock.
664 * In addition, we already have the dd and dealing with snapshots is simplified
665 * in this code.
666 */
667
668typedef enum {
669 ENFORCE_ALWAYS,
670 ENFORCE_NEVER,
671 ENFORCE_ABOVE
672} enforce_res_t;
673
674static enforce_res_t
675dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
676{
677 enforce_res_t enforce = ENFORCE_ALWAYS;
678 uint64_t obj;
679 dsl_dataset_t *ds;
680 uint64_t zoned;
681
682 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
683 prop == ZFS_PROP_SNAPSHOT_LIMIT);
684
685#ifdef _KERNEL
686#ifdef __FreeBSD__
687 if (jailed(cr))
688#else
689 if (crgetzoneid(cr) != GLOBAL_ZONEID)
690#endif
691 return (ENFORCE_ALWAYS);
692
693 if (secpolicy_zfs(cr) == 0)
694 return (ENFORCE_NEVER);
695#endif
696
697 if ((obj = dd->dd_phys->dd_head_dataset_obj) == 0)
698 return (ENFORCE_ALWAYS);
699
700 ASSERT(dsl_pool_config_held(dd->dd_pool));
701
702 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
703 return (ENFORCE_ALWAYS);
704
705 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
706 /* Only root can access zoned fs's from the GZ */
707 enforce = ENFORCE_ALWAYS;
708 } else {
709 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
710 enforce = ENFORCE_ABOVE;
711 }
712
713 dsl_dataset_rele(ds, FTAG);
714 return (enforce);
715}
716
717/*
718 * Check if adding additional child filesystem(s) would exceed any filesystem
719 * limits or adding additional snapshot(s) would exceed any snapshot limits.
720 * The prop argument indicates which limit to check.
721 *
722 * Note that all filesystem limits up to the root (or the highest
723 * initialized) filesystem or the given ancestor must be satisfied.
724 */
725int
726dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
727 dsl_dir_t *ancestor, cred_t *cr)
728{
729 objset_t *os = dd->dd_pool->dp_meta_objset;
730 uint64_t limit, count;
731 char *count_prop;
732 enforce_res_t enforce;
733 int err = 0;
734
735 ASSERT(dsl_pool_config_held(dd->dd_pool));
736 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
737 prop == ZFS_PROP_SNAPSHOT_LIMIT);
738
739 /*
740 * If we're allowed to change the limit, don't enforce the limit
741 * e.g. this can happen if a snapshot is taken by an administrative
742 * user in the global zone (i.e. a recursive snapshot by root).
743 * However, we must handle the case of delegated permissions where we
744 * are allowed to change the limit on the current dataset, but there
745 * is another limit in the tree above.
746 */
747 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
748 if (enforce == ENFORCE_NEVER)
749 return (0);
750
751 /*
752 * e.g. if renaming a dataset with no snapshots, count adjustment
753 * is 0.
754 */
755 if (delta == 0)
756 return (0);
757
758 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
759 /*
760 * We don't enforce the limit for temporary snapshots. This is
761 * indicated by a NULL cred_t argument.
762 */
763 if (cr == NULL)
764 return (0);
765
766 count_prop = DD_FIELD_SNAPSHOT_COUNT;
767 } else {
768 count_prop = DD_FIELD_FILESYSTEM_COUNT;
769 }
770
771 /*
772 * If an ancestor has been provided, stop checking the limit once we
773 * hit that dir. We need this during rename so that we don't overcount
774 * the check once we recurse up to the common ancestor.
775 */
776 if (ancestor == dd)
777 return (0);
778
779 /*
780 * If we hit an uninitialized node while recursing up the tree, we can
781 * stop since we know there is no limit here (or above). The counts are
782 * not valid on this node and we know we won't touch this node's counts.
783 */
784 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
785 count_prop, sizeof (count), 1, &count) == ENOENT)
786 return (0);
787
788 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
789 B_FALSE);
790 if (err != 0)
791 return (err);
792
793 /* Is there a limit which we've hit? */
794 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
795 return (SET_ERROR(EDQUOT));
796
797 if (dd->dd_parent != NULL)
798 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
799 ancestor, cr);
800
801 return (err);
802}
803
804/*
805 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
806 * parents. When a new filesystem/snapshot is created, increment the count on
807 * all parents, and when a filesystem/snapshot is destroyed, decrement the
808 * count.
809 */
810void
811dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
812 dmu_tx_t *tx)
813{
814 int err;
815 objset_t *os = dd->dd_pool->dp_meta_objset;
816 uint64_t count;
817
818 ASSERT(dsl_pool_config_held(dd->dd_pool));
819 ASSERT(dmu_tx_is_syncing(tx));
820 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
821 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
822
823 /*
824 * When we receive an incremental stream into a filesystem that already
825 * exists, a temporary clone is created. We don't count this temporary
826 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
827 * $MOS & $ORIGIN) objsets.
828 */
829 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
830 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
831 return;
832
833 /*
834 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
835 */
836 if (delta == 0)
837 return;
838
839 /*
840 * If we hit an uninitialized node while recursing up the tree, we can
841 * stop since we know the counts are not valid on this node and we
842 * know we shouldn't touch this node's counts. An uninitialized count
843 * on the node indicates that either the feature has not yet been
844 * activated or there are no limits on this part of the tree.
845 */
846 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
847 prop, sizeof (count), 1, &count)) == ENOENT)
848 return;
849 VERIFY0(err);
850
851 count += delta;
852 /* Use a signed verify to make sure we're not neg. */
853 VERIFY3S(count, >=, 0);
854
855 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
856 tx));
857
858 /* Roll up this additional count into our ancestors */
859 if (dd->dd_parent != NULL)
860 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
861}
862
863uint64_t
864dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
865 dmu_tx_t *tx)
866{
867 objset_t *mos = dp->dp_meta_objset;
868 uint64_t ddobj;
869 dsl_dir_phys_t *ddphys;
870 dmu_buf_t *dbuf;
871
872 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
873 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
874 if (pds) {
875 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
876 name, sizeof (uint64_t), 1, &ddobj, tx));
877 } else {
878 /* it's the root dir */
879 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
880 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
881 }
882 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
883 dmu_buf_will_dirty(dbuf, tx);
884 ddphys = dbuf->db_data;
885
886 ddphys->dd_creation_time = gethrestime_sec();
887 if (pds) {
888 ddphys->dd_parent_obj = pds->dd_object;
889
890 /* update the filesystem counts */
891 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
892 }
893 ddphys->dd_props_zapobj = zap_create(mos,
894 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
895 ddphys->dd_child_dir_zapobj = zap_create(mos,
896 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
897 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
898 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
899 dmu_buf_rele(dbuf, FTAG);
900
901 return (ddobj);
902}
903
904boolean_t
905dsl_dir_is_clone(dsl_dir_t *dd)
906{
907 return (dd->dd_phys->dd_origin_obj &&
908 (dd->dd_pool->dp_origin_snap == NULL ||
909 dd->dd_phys->dd_origin_obj !=
910 dd->dd_pool->dp_origin_snap->ds_object));
911}
912
913void
914dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
915{
916 mutex_enter(&dd->dd_lock);
917 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
918 dd->dd_phys->dd_used_bytes);
919 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
920 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
921 dd->dd_phys->dd_reserved);
922 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
923 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
924 (dd->dd_phys->dd_uncompressed_bytes * 100 /
925 dd->dd_phys->dd_compressed_bytes));
926 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
927 dd->dd_phys->dd_uncompressed_bytes);
928 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
929 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
930 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
931 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
932 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
933 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
934 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
935 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
936 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
937 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
938 }
939 mutex_exit(&dd->dd_lock);
940
941 if (dsl_dir_is_zapified(dd)) {
942 uint64_t count;
943 objset_t *os = dd->dd_pool->dp_meta_objset;
944
945 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
946 sizeof (count), 1, &count) == 0) {
947 dsl_prop_nvlist_add_uint64(nv,
948 ZFS_PROP_FILESYSTEM_COUNT, count);
949 }
950 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
951 sizeof (count), 1, &count) == 0) {
952 dsl_prop_nvlist_add_uint64(nv,
953 ZFS_PROP_SNAPSHOT_COUNT, count);
954 }
955 }
956
957 if (dsl_dir_is_clone(dd)) {
958 dsl_dataset_t *ds;
959 char buf[MAXNAMELEN];
960
961 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
962 dd->dd_phys->dd_origin_obj, FTAG, &ds));
963 dsl_dataset_name(ds, buf);
964 dsl_dataset_rele(ds, FTAG);
965 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
966 }
967}
968
969void
970dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
971{
972 dsl_pool_t *dp = dd->dd_pool;
973
974 ASSERT(dd->dd_phys);
975
976 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
977 /* up the hold count until we can be written out */
978 dmu_buf_add_ref(dd->dd_dbuf, dd);
979 }
980}
981
982static int64_t
983parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
984{
985 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
986 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
987 return (new_accounted - old_accounted);
988}
989
990void
991dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
992{
993 ASSERT(dmu_tx_is_syncing(tx));
994
995 mutex_enter(&dd->dd_lock);
996 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
997 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
998 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
999 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1000 mutex_exit(&dd->dd_lock);
1001
1002 /* release the hold from dsl_dir_dirty */
1003 dmu_buf_rele(dd->dd_dbuf, dd);
1004}
1005
1006static uint64_t
1007dsl_dir_space_towrite(dsl_dir_t *dd)
1008{
1009 uint64_t space = 0;
1010 int i;
1011
1012 ASSERT(MUTEX_HELD(&dd->dd_lock));
1013
1014 for (i = 0; i < TXG_SIZE; i++) {
1015 space += dd->dd_space_towrite[i&TXG_MASK];
1016 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1017 }
1018 return (space);
1019}
1020
1021/*
1022 * How much space would dd have available if ancestor had delta applied
1023 * to it? If ondiskonly is set, we're only interested in what's
1024 * on-disk, not estimated pending changes.
1025 */
1026uint64_t
1027dsl_dir_space_available(dsl_dir_t *dd,
1028 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1029{
1030 uint64_t parentspace, myspace, quota, used;
1031
1032 /*
1033 * If there are no restrictions otherwise, assume we have
1034 * unlimited space available.
1035 */
1036 quota = UINT64_MAX;
1037 parentspace = UINT64_MAX;
1038
1039 if (dd->dd_parent != NULL) {
1040 parentspace = dsl_dir_space_available(dd->dd_parent,
1041 ancestor, delta, ondiskonly);
1042 }
1043
1044 mutex_enter(&dd->dd_lock);
1045 if (dd->dd_phys->dd_quota != 0)
1046 quota = dd->dd_phys->dd_quota;
1047 used = dd->dd_phys->dd_used_bytes;
1048 if (!ondiskonly)
1049 used += dsl_dir_space_towrite(dd);
1050
1051 if (dd->dd_parent == NULL) {
1052 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1053 quota = MIN(quota, poolsize);
1054 }
1055
1056 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
1057 /*
1058 * We have some space reserved, in addition to what our
1059 * parent gave us.
1060 */
1061 parentspace += dd->dd_phys->dd_reserved - used;
1062 }
1063
1064 if (dd == ancestor) {
1065 ASSERT(delta <= 0);
1066 ASSERT(used >= -delta);
1067 used += delta;
1068 if (parentspace != UINT64_MAX)
1069 parentspace -= delta;
1070 }
1071
1072 if (used > quota) {
1073 /* over quota */
1074 myspace = 0;
1075 } else {
1076 /*
1077 * the lesser of the space provided by our parent and
1078 * the space left in our quota
1079 */
1080 myspace = MIN(parentspace, quota - used);
1081 }
1082
1083 mutex_exit(&dd->dd_lock);
1084
1085 return (myspace);
1086}
1087
1088struct tempreserve {
1089 list_node_t tr_node;
1090 dsl_dir_t *tr_ds;
1091 uint64_t tr_size;
1092};
1093
1094static int
1095dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1096 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1097 dmu_tx_t *tx, boolean_t first)
1098{
1099 uint64_t txg = tx->tx_txg;
1100 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1101 uint64_t deferred = 0;
1102 struct tempreserve *tr;
1103 int retval = EDQUOT;
1104 int txgidx = txg & TXG_MASK;
1105 int i;
1106 uint64_t ref_rsrv = 0;
1107
1108 ASSERT3U(txg, !=, 0);
1109 ASSERT3S(asize, >, 0);
1110
1111 mutex_enter(&dd->dd_lock);
1112
1113 /*
1114 * Check against the dsl_dir's quota. We don't add in the delta
1115 * when checking for over-quota because they get one free hit.
1116 */
1117 est_inflight = dsl_dir_space_towrite(dd);
1118 for (i = 0; i < TXG_SIZE; i++)
1119 est_inflight += dd->dd_tempreserved[i];
1120 used_on_disk = dd->dd_phys->dd_used_bytes;
1121
1122 /*
1123 * On the first iteration, fetch the dataset's used-on-disk and
1124 * refreservation values. Also, if checkrefquota is set, test if
1125 * allocating this space would exceed the dataset's refquota.
1126 */
1127 if (first && tx->tx_objset) {
1128 int error;
1129 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1130
1131 error = dsl_dataset_check_quota(ds, checkrefquota,
1132 asize, est_inflight, &used_on_disk, &ref_rsrv);
1133 if (error) {
1134 mutex_exit(&dd->dd_lock);
1135 return (error);
1136 }
1137 }
1138
1139 /*
1140 * If this transaction will result in a net free of space,
1141 * we want to let it through.
1142 */
1143 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
1144 quota = UINT64_MAX;
1145 else
1146 quota = dd->dd_phys->dd_quota;
1147
1148 /*
1149 * Adjust the quota against the actual pool size at the root
1150 * minus any outstanding deferred frees.
1151 * To ensure that it's possible to remove files from a full
1152 * pool without inducing transient overcommits, we throttle
1153 * netfree transactions against a quota that is slightly larger,
1154 * but still within the pool's allocation slop. In cases where
1155 * we're very close to full, this will allow a steady trickle of
1156 * removes to get through.
1157 */
1158 if (dd->dd_parent == NULL) {
1159 spa_t *spa = dd->dd_pool->dp_spa;
1160 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1161 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1162 if (poolsize - deferred < quota) {
1163 quota = poolsize - deferred;
1164 retval = ENOSPC;
1165 }
1166 }
1167
1168 /*
1169 * If they are requesting more space, and our current estimate
1170 * is over quota, they get to try again unless the actual
1171 * on-disk is over quota and there are no pending changes (which
1172 * may free up space for us).
1173 */
1174 if (used_on_disk + est_inflight >= quota) {
1175 if (est_inflight > 0 || used_on_disk < quota ||
1176 (retval == ENOSPC && used_on_disk < quota + deferred))
1177 retval = ERESTART;
1178 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1179 "quota=%lluK tr=%lluK err=%d\n",
1180 used_on_disk>>10, est_inflight>>10,
1181 quota>>10, asize>>10, retval);
1182 mutex_exit(&dd->dd_lock);
1183 return (SET_ERROR(retval));
1184 }
1185
1186 /* We need to up our estimated delta before dropping dd_lock */
1187 dd->dd_tempreserved[txgidx] += asize;
1188
1189 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1190 asize - ref_rsrv);
1191 mutex_exit(&dd->dd_lock);
1192
1193 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1194 tr->tr_ds = dd;
1195 tr->tr_size = asize;
1196 list_insert_tail(tr_list, tr);
1197
1198 /* see if it's OK with our parent */
1199 if (dd->dd_parent && parent_rsrv) {
1200 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
1201
1202 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1203 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1204 } else {
1205 return (0);
1206 }
1207}
1208
1209/*
1210 * Reserve space in this dsl_dir, to be used in this tx's txg.
1211 * After the space has been dirtied (and dsl_dir_willuse_space()
1212 * has been called), the reservation should be canceled, using
1213 * dsl_dir_tempreserve_clear().
1214 */
1215int
1216dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1217 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1218{
1219 int err;
1220 list_t *tr_list;
1221
1222 if (asize == 0) {
1223 *tr_cookiep = NULL;
1224 return (0);
1225 }
1226
1227 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1228 list_create(tr_list, sizeof (struct tempreserve),
1229 offsetof(struct tempreserve, tr_node));
1230 ASSERT3S(asize, >, 0);
1231 ASSERT3S(fsize, >=, 0);
1232
1233 err = arc_tempreserve_space(lsize, tx->tx_txg);
1234 if (err == 0) {
1235 struct tempreserve *tr;
1236
1237 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1238 tr->tr_size = lsize;
1239 list_insert_tail(tr_list, tr);
1240 } else {
1241 if (err == EAGAIN) {
1242 /*
1243 * If arc_memory_throttle() detected that pageout
1244 * is running and we are low on memory, we delay new
1245 * non-pageout transactions to give pageout an
1246 * advantage.
1247 *
1248 * It is unfortunate to be delaying while the caller's
1249 * locks are held.
1250 */
1251 txg_delay(dd->dd_pool, tx->tx_txg,
1252 MSEC2NSEC(10), MSEC2NSEC(10));
1253 err = SET_ERROR(ERESTART);
1254 }
1255 }
1256
1257 if (err == 0) {
1258 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1259 FALSE, asize > usize, tr_list, tx, TRUE);
1260 }
1261
1262 if (err != 0)
1263 dsl_dir_tempreserve_clear(tr_list, tx);
1264 else
1265 *tr_cookiep = tr_list;
1266
1267 return (err);
1268}
1269
1270/*
1271 * Clear a temporary reservation that we previously made with
1272 * dsl_dir_tempreserve_space().
1273 */
1274void
1275dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1276{
1277 int txgidx = tx->tx_txg & TXG_MASK;
1278 list_t *tr_list = tr_cookie;
1279 struct tempreserve *tr;
1280
1281 ASSERT3U(tx->tx_txg, !=, 0);
1282
1283 if (tr_cookie == NULL)
1284 return;
1285
1286 while ((tr = list_head(tr_list)) != NULL) {
1287 if (tr->tr_ds) {
1288 mutex_enter(&tr->tr_ds->dd_lock);
1289 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1290 tr->tr_size);
1291 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1292 mutex_exit(&tr->tr_ds->dd_lock);
1293 } else {
1294 arc_tempreserve_clear(tr->tr_size);
1295 }
1296 list_remove(tr_list, tr);
1297 kmem_free(tr, sizeof (struct tempreserve));
1298 }
1299
1300 kmem_free(tr_list, sizeof (list_t));
1301}
1302
1303/*
1304 * This should be called from open context when we think we're going to write
1305 * or free space, for example when dirtying data. Be conservative; it's okay
1306 * to write less space or free more, but we don't want to write more or free
1307 * less than the amount specified.
1308 */
1309void
1310dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1311{
1312 int64_t parent_space;
1313 uint64_t est_used;
1314
1315 mutex_enter(&dd->dd_lock);
1316 if (space > 0)
1317 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1318
1319 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
1320 parent_space = parent_delta(dd, est_used, space);
1321 mutex_exit(&dd->dd_lock);
1322
1323 /* Make sure that we clean up dd_space_to* */
1324 dsl_dir_dirty(dd, tx);
1325
1326 /* XXX this is potentially expensive and unnecessary... */
1327 if (parent_space && dd->dd_parent)
1328 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1329}
1330
1331/* call from syncing context when we actually write/free space for this dd */
1332void
1333dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1334 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1335{
1336 int64_t accounted_delta;
1337
1338 /*
1339 * dsl_dataset_set_refreservation_sync_impl() calls this with
1340 * dd_lock held, so that it can atomically update
1341 * ds->ds_reserved and the dsl_dir accounting, so that
1342 * dsl_dataset_check_quota() can see dataset and dir accounting
1343 * consistently.
1344 */
1345 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1346
1347 ASSERT(dmu_tx_is_syncing(tx));
1348 ASSERT(type < DD_USED_NUM);
1349
1350 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1351
1352 if (needlock)
1353 mutex_enter(&dd->dd_lock);
1354 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
1355 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
1356 ASSERT(compressed >= 0 ||
1357 dd->dd_phys->dd_compressed_bytes >= -compressed);
1358 ASSERT(uncompressed >= 0 ||
1359 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
1360 dd->dd_phys->dd_used_bytes += used;
1361 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
1362 dd->dd_phys->dd_compressed_bytes += compressed;
1363
1364 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1365 ASSERT(used > 0 ||
1366 dd->dd_phys->dd_used_breakdown[type] >= -used);
1367 dd->dd_phys->dd_used_breakdown[type] += used;
1368#ifdef DEBUG
1369 dd_used_t t;
1370 uint64_t u = 0;
1371 for (t = 0; t < DD_USED_NUM; t++)
1372 u += dd->dd_phys->dd_used_breakdown[t];
1373 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
1374#endif
1375 }
1376 if (needlock)
1377 mutex_exit(&dd->dd_lock);
1378
1379 if (dd->dd_parent != NULL) {
1380 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1381 accounted_delta, compressed, uncompressed, tx);
1382 dsl_dir_transfer_space(dd->dd_parent,
1383 used - accounted_delta,
1384 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1385 }
1386}
1387
1388void
1389dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1390 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1391{
1392 ASSERT(dmu_tx_is_syncing(tx));
1393 ASSERT(oldtype < DD_USED_NUM);
1394 ASSERT(newtype < DD_USED_NUM);
1395
1396 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
1397 return;
1398
1399 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1400 mutex_enter(&dd->dd_lock);
1401 ASSERT(delta > 0 ?
1402 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1403 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1404 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1405 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1406 dd->dd_phys->dd_used_breakdown[newtype] += delta;
1407 mutex_exit(&dd->dd_lock);
1408}
1409
1410typedef struct dsl_dir_set_qr_arg {
1411 const char *ddsqra_name;
1412 zprop_source_t ddsqra_source;
1413 uint64_t ddsqra_value;
1414} dsl_dir_set_qr_arg_t;
1415
1416static int
1417dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1418{
1419 dsl_dir_set_qr_arg_t *ddsqra = arg;
1420 dsl_pool_t *dp = dmu_tx_pool(tx);
1421 dsl_dataset_t *ds;
1422 int error;
1423 uint64_t towrite, newval;
1424
1425 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1426 if (error != 0)
1427 return (error);
1428
1429 error = dsl_prop_predict(ds->ds_dir, "quota",
1430 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1431 if (error != 0) {
1432 dsl_dataset_rele(ds, FTAG);
1433 return (error);
1434 }
1435
1436 if (newval == 0) {
1437 dsl_dataset_rele(ds, FTAG);
1438 return (0);
1439 }
1440
1441 mutex_enter(&ds->ds_dir->dd_lock);
1442 /*
1443 * If we are doing the preliminary check in open context, and
1444 * there are pending changes, then don't fail it, since the
1445 * pending changes could under-estimate the amount of space to be
1446 * freed up.
1447 */
1448 towrite = dsl_dir_space_towrite(ds->ds_dir);
1449 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1450 (newval < ds->ds_dir->dd_phys->dd_reserved ||
1451 newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
1452 error = SET_ERROR(ENOSPC);
1453 }
1454 mutex_exit(&ds->ds_dir->dd_lock);
1455 dsl_dataset_rele(ds, FTAG);
1456 return (error);
1457}
1458
1459static void
1460dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1461{
1462 dsl_dir_set_qr_arg_t *ddsqra = arg;
1463 dsl_pool_t *dp = dmu_tx_pool(tx);
1464 dsl_dataset_t *ds;
1465 uint64_t newval;
1466
1467 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1468
1469 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1470 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1471 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1472 &ddsqra->ddsqra_value, tx);
1473
1474 VERIFY0(dsl_prop_get_int_ds(ds,
1475 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1476 } else {
1477 newval = ddsqra->ddsqra_value;
1478 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1479 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1480 }
1481
1482 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1483 mutex_enter(&ds->ds_dir->dd_lock);
1484 ds->ds_dir->dd_phys->dd_quota = newval;
1485 mutex_exit(&ds->ds_dir->dd_lock);
1486 dsl_dataset_rele(ds, FTAG);
1487}
1488
1489int
1490dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1491{
1492 dsl_dir_set_qr_arg_t ddsqra;
1493
1494 ddsqra.ddsqra_name = ddname;
1495 ddsqra.ddsqra_source = source;
1496 ddsqra.ddsqra_value = quota;
1497
1498 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1499 dsl_dir_set_quota_sync, &ddsqra, 0));
1500}
1501
1502int
1503dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1504{
1505 dsl_dir_set_qr_arg_t *ddsqra = arg;
1506 dsl_pool_t *dp = dmu_tx_pool(tx);
1507 dsl_dataset_t *ds;
1508 dsl_dir_t *dd;
1509 uint64_t newval, used, avail;
1510 int error;
1511
1512 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1513 if (error != 0)
1514 return (error);
1515 dd = ds->ds_dir;
1516
1517 /*
1518 * If we are doing the preliminary check in open context, the
1519 * space estimates may be inaccurate.
1520 */
1521 if (!dmu_tx_is_syncing(tx)) {
1522 dsl_dataset_rele(ds, FTAG);
1523 return (0);
1524 }
1525
1526 error = dsl_prop_predict(ds->ds_dir,
1527 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1528 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1529 if (error != 0) {
1530 dsl_dataset_rele(ds, FTAG);
1531 return (error);
1532 }
1533
1534 mutex_enter(&dd->dd_lock);
1535 used = dd->dd_phys->dd_used_bytes;
1536 mutex_exit(&dd->dd_lock);
1537
1538 if (dd->dd_parent) {
1539 avail = dsl_dir_space_available(dd->dd_parent,
1540 NULL, 0, FALSE);
1541 } else {
1542 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1543 }
1544
1545 if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1546 uint64_t delta = MAX(used, newval) -
1547 MAX(used, dd->dd_phys->dd_reserved);
1548
1549 if (delta > avail ||
1550 (dd->dd_phys->dd_quota > 0 &&
1551 newval > dd->dd_phys->dd_quota))
1552 error = SET_ERROR(ENOSPC);
1553 }
1554
1555 dsl_dataset_rele(ds, FTAG);
1556 return (error);
1557}
1558
1559void
1560dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1561{
1562 uint64_t used;
1563 int64_t delta;
1564
1565 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1566
1567 mutex_enter(&dd->dd_lock);
1568 used = dd->dd_phys->dd_used_bytes;
1569 delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1570 dd->dd_phys->dd_reserved = value;
1571
1572 if (dd->dd_parent != NULL) {
1573 /* Roll up this additional usage into our ancestors */
1574 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1575 delta, 0, 0, tx);
1576 }
1577 mutex_exit(&dd->dd_lock);
1578}
1579
1580static void
1581dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1582{
1583 dsl_dir_set_qr_arg_t *ddsqra = arg;
1584 dsl_pool_t *dp = dmu_tx_pool(tx);
1585 dsl_dataset_t *ds;
1586 uint64_t newval;
1587
1588 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1589
1590 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1591 dsl_prop_set_sync_impl(ds,
1592 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1593 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1594 &ddsqra->ddsqra_value, tx);
1595
1596 VERIFY0(dsl_prop_get_int_ds(ds,
1597 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1598 } else {
1599 newval = ddsqra->ddsqra_value;
1600 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1601 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1602 (longlong_t)newval);
1603 }
1604
1605 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1606 dsl_dataset_rele(ds, FTAG);
1607}
1608
1609int
1610dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1611 uint64_t reservation)
1612{
1613 dsl_dir_set_qr_arg_t ddsqra;
1614
1615 ddsqra.ddsqra_name = ddname;
1616 ddsqra.ddsqra_source = source;
1617 ddsqra.ddsqra_value = reservation;
1618
1619 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1620 dsl_dir_set_reservation_sync, &ddsqra, 0));
1621}
1622
1623static dsl_dir_t *
1624closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1625{
1626 for (; ds1; ds1 = ds1->dd_parent) {
1627 dsl_dir_t *dd;
1628 for (dd = ds2; dd; dd = dd->dd_parent) {
1629 if (ds1 == dd)
1630 return (dd);
1631 }
1632 }
1633 return (NULL);
1634}
1635
1636/*
1637 * If delta is applied to dd, how much of that delta would be applied to
1638 * ancestor? Syncing context only.
1639 */
1640static int64_t
1641would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1642{
1643 if (dd == ancestor)
1644 return (delta);
1645
1646 mutex_enter(&dd->dd_lock);
1647 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1648 mutex_exit(&dd->dd_lock);
1649 return (would_change(dd->dd_parent, delta, ancestor));
1650}
1651
1652typedef struct dsl_dir_rename_arg {
1653 const char *ddra_oldname;
1654 const char *ddra_newname;
1655 cred_t *ddra_cred;
1656} dsl_dir_rename_arg_t;
1657
1658/* ARGSUSED */
1659static int
1660dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1661{
1662 int *deltap = arg;
1663 char namebuf[MAXNAMELEN];
1664
1665 dsl_dataset_name(ds, namebuf);
1666
1667 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1668 return (SET_ERROR(ENAMETOOLONG));
1669 return (0);
1670}
1671
1672static int
1673dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1674{
1675 dsl_dir_rename_arg_t *ddra = arg;
1676 dsl_pool_t *dp = dmu_tx_pool(tx);
1677 dsl_dir_t *dd, *newparent;
1678 const char *mynewname;
1679 int error;
1680 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1681
1682 /* target dir should exist */
1683 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1684 if (error != 0)
1685 return (error);
1686
1687 /* new parent should exist */
1688 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1689 &newparent, &mynewname);
1690 if (error != 0) {
1691 dsl_dir_rele(dd, FTAG);
1692 return (error);
1693 }
1694
1695 /* can't rename to different pool */
1696 if (dd->dd_pool != newparent->dd_pool) {
1697 dsl_dir_rele(newparent, FTAG);
1698 dsl_dir_rele(dd, FTAG);
1699 return (SET_ERROR(ENXIO));
1700 }
1701
1702 /* new name should not already exist */
1703 if (mynewname == NULL) {
1704 dsl_dir_rele(newparent, FTAG);
1705 dsl_dir_rele(dd, FTAG);
1706 return (SET_ERROR(EEXIST));
1707 }
1708
1709 /* if the name length is growing, validate child name lengths */
1710 if (delta > 0) {
1711 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1712 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1713 if (error != 0) {
1714 dsl_dir_rele(newparent, FTAG);
1715 dsl_dir_rele(dd, FTAG);
1716 return (error);
1717 }
1718 }
1719
1720 if (dmu_tx_is_syncing(tx)) {
1721 if (spa_feature_is_active(dp->dp_spa,
1722 SPA_FEATURE_FS_SS_LIMIT)) {
1723 /*
1724 * Although this is the check function and we don't
1725 * normally make on-disk changes in check functions,
1726 * we need to do that here.
1727 *
1728 * Ensure this portion of the tree's counts have been
1729 * initialized in case the new parent has limits set.
1730 */
1731 dsl_dir_init_fs_ss_count(dd, tx);
1732 }
1733 }
1734
1735 if (newparent != dd->dd_parent) {
1736 /* is there enough space? */
1737 uint64_t myspace =
1738 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1739 objset_t *os = dd->dd_pool->dp_meta_objset;
1740 uint64_t fs_cnt = 0;
1741 uint64_t ss_cnt = 0;
1742
1743 if (dsl_dir_is_zapified(dd)) {
1744 int err;
1745
1746 err = zap_lookup(os, dd->dd_object,
1747 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1748 &fs_cnt);
1749 if (err != ENOENT && err != 0) {
1750 dsl_dir_rele(newparent, FTAG);
1751 dsl_dir_rele(dd, FTAG);
1752 return (err);
1753 }
1754
1755 /*
1756 * have to add 1 for the filesystem itself that we're
1757 * moving
1758 */
1759 fs_cnt++;
1760
1761 err = zap_lookup(os, dd->dd_object,
1762 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1763 &ss_cnt);
1764 if (err != ENOENT && err != 0) {
1765 dsl_dir_rele(newparent, FTAG);
1766 dsl_dir_rele(dd, FTAG);
1767 return (err);
1768 }
1769 }
1770
1771 /* no rename into our descendant */
1772 if (closest_common_ancestor(dd, newparent) == dd) {
1773 dsl_dir_rele(newparent, FTAG);
1774 dsl_dir_rele(dd, FTAG);
1775 return (SET_ERROR(EINVAL));
1776 }
1777
1778 error = dsl_dir_transfer_possible(dd->dd_parent,
1779 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1780 if (error != 0) {
1781 dsl_dir_rele(newparent, FTAG);
1782 dsl_dir_rele(dd, FTAG);
1783 return (error);
1784 }
1785 }
1786
1787 dsl_dir_rele(newparent, FTAG);
1788 dsl_dir_rele(dd, FTAG);
1789 return (0);
1790}
1791
1792static void
1793dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1794{
1795 dsl_dir_rename_arg_t *ddra = arg;
1796 dsl_pool_t *dp = dmu_tx_pool(tx);
1797 dsl_dir_t *dd, *newparent;
1798 const char *mynewname;
1799 int error;
1800 objset_t *mos = dp->dp_meta_objset;
1801
1802 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1803 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1804 &mynewname));
1805
1806 /* Log this before we change the name. */
1807 spa_history_log_internal_dd(dd, "rename", tx,
1808 "-> %s", ddra->ddra_newname);
1809
1810 if (newparent != dd->dd_parent) {
1811 objset_t *os = dd->dd_pool->dp_meta_objset;
1812 uint64_t fs_cnt = 0;
1813 uint64_t ss_cnt = 0;
1814
1815 /*
1816 * We already made sure the dd counts were initialized in the
1817 * check function.
1818 */
1819 if (spa_feature_is_active(dp->dp_spa,
1820 SPA_FEATURE_FS_SS_LIMIT)) {
1821 VERIFY0(zap_lookup(os, dd->dd_object,
1822 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1823 &fs_cnt));
1824 /* add 1 for the filesystem itself that we're moving */
1825 fs_cnt++;
1826
1827 VERIFY0(zap_lookup(os, dd->dd_object,
1828 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1829 &ss_cnt));
1830 }
1831
1832 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1833 DD_FIELD_FILESYSTEM_COUNT, tx);
1834 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1835 DD_FIELD_FILESYSTEM_COUNT, tx);
1836
1837 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1838 DD_FIELD_SNAPSHOT_COUNT, tx);
1839 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1840 DD_FIELD_SNAPSHOT_COUNT, tx);
1841
1842 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1843 -dd->dd_phys->dd_used_bytes,
1844 -dd->dd_phys->dd_compressed_bytes,
1845 -dd->dd_phys->dd_uncompressed_bytes, tx);
1846 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1847 dd->dd_phys->dd_used_bytes,
1848 dd->dd_phys->dd_compressed_bytes,
1849 dd->dd_phys->dd_uncompressed_bytes, tx);
1850
1851 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1852 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1853 dd->dd_phys->dd_used_bytes;
1854
1855 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1856 -unused_rsrv, 0, 0, tx);
1857 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1858 unused_rsrv, 0, 0, tx);
1859 }
1860 }
1861
1862 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1863
1864 /* remove from old parent zapobj */
1865 error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1866 dd->dd_myname, tx);
1867 ASSERT0(error);
1868
1869 (void) strcpy(dd->dd_myname, mynewname);
1870 dsl_dir_rele(dd->dd_parent, dd);
1871 dd->dd_phys->dd_parent_obj = newparent->dd_object;
1872 VERIFY0(dsl_dir_hold_obj(dp,
1873 newparent->dd_object, NULL, dd, &dd->dd_parent));
1874
1875 /* add to new parent zapobj */
1876 VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1877 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1878
1879#ifdef __FreeBSD__
1880#ifdef _KERNEL
1881 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1882 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1883#endif
1884#endif
1885
1886 dsl_prop_notify_all(dd);
1887
1888 dsl_dir_rele(newparent, FTAG);
1889 dsl_dir_rele(dd, FTAG);
1890}
1891
1892int
1893dsl_dir_rename(const char *oldname, const char *newname)
1894{
1895 dsl_dir_rename_arg_t ddra;
1896
1897 ddra.ddra_oldname = oldname;
1898 ddra.ddra_newname = newname;
1899 ddra.ddra_cred = CRED();
1900
1901 return (dsl_sync_task(oldname,
1902 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1903}
1904
1905int
1906dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1907 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1908{
1909 dsl_dir_t *ancestor;
1910 int64_t adelta;
1911 uint64_t avail;
1912 int err;
1913
1914 ancestor = closest_common_ancestor(sdd, tdd);
1915 adelta = would_change(sdd, -space, ancestor);
1916 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1917 if (avail < space)
1918 return (SET_ERROR(ENOSPC));
1919
1920 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1921 ancestor, cr);
1922 if (err != 0)
1923 return (err);
1924 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1925 ancestor, cr);
1926 if (err != 0)
1927 return (err);
1928
1929 return (0);
1930}
1931
1932timestruc_t
1933dsl_dir_snap_cmtime(dsl_dir_t *dd)
1934{
1935 timestruc_t t;
1936
1937 mutex_enter(&dd->dd_lock);
1938 t = dd->dd_snap_cmtime;
1939 mutex_exit(&dd->dd_lock);
1940
1941 return (t);
1942}
1943
1944void
1945dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1946{
1947 timestruc_t t;
1948
1949 gethrestime(&t);
1950 mutex_enter(&dd->dd_lock);
1951 dd->dd_snap_cmtime = t;
1952 mutex_exit(&dd->dd_lock);
1953}
1954
1955void
1956dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1957{
1958 objset_t *mos = dd->dd_pool->dp_meta_objset;
1959 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1960}
1961
1962boolean_t
1963dsl_dir_is_zapified(dsl_dir_t *dd)
1964{
1965 dmu_object_info_t doi;
1966
1967 dmu_object_info_from_db(dd->dd_dbuf, &doi);
1968 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
1969}
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) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2013 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
27 */
28
29#include <sys/dmu.h>
30#include <sys/dmu_objset.h>
31#include <sys/dmu_tx.h>
32#include <sys/dsl_dataset.h>
33#include <sys/dsl_dir.h>
34#include <sys/dsl_prop.h>
35#include <sys/dsl_synctask.h>
36#include <sys/dsl_deleg.h>
37#include <sys/dmu_impl.h>
38#include <sys/spa.h>
39#include <sys/metaslab.h>
40#include <sys/zap.h>
41#include <sys/zio.h>
42#include <sys/arc.h>
43#include <sys/sunddi.h>
44#include <sys/zvol.h>
45#ifdef _KERNEL
46#include <sys/zfs_vfsops.h>
47#endif
48#include <sys/zfeature.h>
49#include <sys/policy.h>
50#include <sys/zfs_znode.h>
51#include "zfs_namecheck.h"
52#include "zfs_prop.h"
53
54/*
55 * Filesystem and Snapshot Limits
56 * ------------------------------
57 *
58 * These limits are used to restrict the number of filesystems and/or snapshots
59 * that can be created at a given level in the tree or below. A typical
60 * use-case is with a delegated dataset where the administrator wants to ensure
61 * that a user within the zone is not creating too many additional filesystems
62 * or snapshots, even though they're not exceeding their space quota.
63 *
64 * The filesystem and snapshot counts are stored as extensible properties. This
65 * capability is controlled by a feature flag and must be enabled to be used.
66 * Once enabled, the feature is not active until the first limit is set. At
67 * that point, future operations to create/destroy filesystems or snapshots
68 * will validate and update the counts.
69 *
70 * Because the count properties will not exist before the feature is active,
71 * the counts are updated when a limit is first set on an uninitialized
72 * dsl_dir node in the tree (The filesystem/snapshot count on a node includes
73 * all of the nested filesystems/snapshots. Thus, a new leaf node has a
74 * filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
75 * snapshot count properties on a node indicate uninitialized counts on that
76 * node.) When first setting a limit on an uninitialized node, the code starts
77 * at the filesystem with the new limit and descends into all sub-filesystems
78 * to add the count properties.
79 *
80 * In practice this is lightweight since a limit is typically set when the
81 * filesystem is created and thus has no children. Once valid, changing the
82 * limit value won't require a re-traversal since the counts are already valid.
83 * When recursively fixing the counts, if a node with a limit is encountered
84 * during the descent, the counts are known to be valid and there is no need to
85 * descend into that filesystem's children. The counts on filesystems above the
86 * one with the new limit will still be uninitialized, unless a limit is
87 * eventually set on one of those filesystems. The counts are always recursively
88 * updated when a limit is set on a dataset, unless there is already a limit.
89 * When a new limit value is set on a filesystem with an existing limit, it is
90 * possible for the new limit to be less than the current count at that level
91 * since a user who can change the limit is also allowed to exceed the limit.
92 *
93 * Once the feature is active, then whenever a filesystem or snapshot is
94 * created, the code recurses up the tree, validating the new count against the
95 * limit at each initialized level. In practice, most levels will not have a
96 * limit set. If there is a limit at any initialized level up the tree, the
97 * check must pass or the creation will fail. Likewise, when a filesystem or
98 * snapshot is destroyed, the counts are recursively adjusted all the way up
99 * the initizized nodes in the tree. Renaming a filesystem into different point
100 * in the tree will first validate, then update the counts on each branch up to
101 * the common ancestor. A receive will also validate the counts and then update
102 * them.
103 *
104 * An exception to the above behavior is that the limit is not enforced if the
105 * user has permission to modify the limit. This is primarily so that
106 * recursive snapshots in the global zone always work. We want to prevent a
107 * denial-of-service in which a lower level delegated dataset could max out its
108 * limit and thus block recursive snapshots from being taken in the global zone.
109 * Because of this, it is possible for the snapshot count to be over the limit
110 * and snapshots taken in the global zone could cause a lower level dataset to
111 * hit or exceed its limit. The administrator taking the global zone recursive
112 * snapshot should be aware of this side-effect and behave accordingly.
113 * For consistency, the filesystem limit is also not enforced if the user can
114 * modify the limit.
115 *
116 * The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
117 * and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
118 * dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
119 * dsl_dir_init_fs_ss_count().
120 *
121 * There is a special case when we receive a filesystem that already exists. In
122 * this case a temporary clone name of %X is created (see dmu_recv_begin). We
123 * never update the filesystem counts for temporary clones.
124 *
125 * Likewise, we do not update the snapshot counts for temporary snapshots,
126 * such as those created by zfs diff.
127 */
128
129static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
130
131/* ARGSUSED */
132static void
133dsl_dir_evict(dmu_buf_t *db, void *arg)
134{
135 dsl_dir_t *dd = arg;
136 dsl_pool_t *dp = dd->dd_pool;
137 int t;
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_rele(dd->dd_parent, dd);
147
148 spa_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 dd->dd_phys = dbuf->db_data;
189 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
190
191 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
192 offsetof(dsl_prop_cb_record_t, cbr_node));
193
194 dsl_dir_snap_cmtime_update(dd);
195
196 if (dd->dd_phys->dd_parent_obj) {
197 err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
198 NULL, dd, &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 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
207 tail, sizeof (foundobj), 1, &foundobj);
208 ASSERT(err || foundobj == ddobj);
209#endif
210 (void) strcpy(dd->dd_myname, tail);
211 } else {
212 err = zap_value_search(dp->dp_meta_objset,
213 dd->dd_parent->dd_phys->dd_child_dir_zapobj,
214 ddobj, 0, dd->dd_myname);
215 }
216 if (err != 0)
217 goto errout;
218 } else {
219 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
220 }
221
222 if (dsl_dir_is_clone(dd)) {
223 dmu_buf_t *origin_bonus;
224 dsl_dataset_phys_t *origin_phys;
225
226 /*
227 * We can't open the origin dataset, because
228 * that would require opening this dsl_dir.
229 * Just look at its phys directly instead.
230 */
231 err = dmu_bonus_hold(dp->dp_meta_objset,
232 dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
233 if (err != 0)
234 goto errout;
235 origin_phys = origin_bonus->db_data;
236 dd->dd_origin_txg =
237 origin_phys->ds_creation_txg;
238 dmu_buf_rele(origin_bonus, FTAG);
239 }
240
241 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
242 dsl_dir_evict);
243 if (winner) {
244 if (dd->dd_parent)
245 dsl_dir_rele(dd->dd_parent, dd);
246 mutex_destroy(&dd->dd_lock);
247 kmem_free(dd, sizeof (dsl_dir_t));
248 dd = winner;
249 } else {
250 spa_open_ref(dp->dp_spa, dd);
251 }
252 }
253
254 /*
255 * The dsl_dir_t has both open-to-close and instantiate-to-evict
256 * holds on the spa. We need the open-to-close holds because
257 * otherwise the spa_refcnt wouldn't change when we open a
258 * dir which the spa also has open, so we could incorrectly
259 * think it was OK to unload/export/destroy the pool. We need
260 * the instantiate-to-evict hold because the dsl_dir_t has a
261 * pointer to the dd_pool, which has a pointer to the spa_t.
262 */
263 spa_open_ref(dp->dp_spa, tag);
264 ASSERT3P(dd->dd_pool, ==, dp);
265 ASSERT3U(dd->dd_object, ==, ddobj);
266 ASSERT3P(dd->dd_dbuf, ==, dbuf);
267 *ddp = dd;
268 return (0);
269
270errout:
271 if (dd->dd_parent)
272 dsl_dir_rele(dd->dd_parent, dd);
273 mutex_destroy(&dd->dd_lock);
274 kmem_free(dd, sizeof (dsl_dir_t));
275 dmu_buf_rele(dbuf, tag);
276 return (err);
277}
278
279void
280dsl_dir_rele(dsl_dir_t *dd, void *tag)
281{
282 dprintf_dd(dd, "%s\n", "");
283 spa_close(dd->dd_pool->dp_spa, tag);
284 dmu_buf_rele(dd->dd_dbuf, tag);
285}
286
287/* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
288void
289dsl_dir_name(dsl_dir_t *dd, char *buf)
290{
291 if (dd->dd_parent) {
292 dsl_dir_name(dd->dd_parent, buf);
293 (void) strcat(buf, "/");
294 } else {
295 buf[0] = '\0';
296 }
297 if (!MUTEX_HELD(&dd->dd_lock)) {
298 /*
299 * recursive mutex so that we can use
300 * dprintf_dd() with dd_lock held
301 */
302 mutex_enter(&dd->dd_lock);
303 (void) strcat(buf, dd->dd_myname);
304 mutex_exit(&dd->dd_lock);
305 } else {
306 (void) strcat(buf, dd->dd_myname);
307 }
308}
309
310/* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
311int
312dsl_dir_namelen(dsl_dir_t *dd)
313{
314 int result = 0;
315
316 if (dd->dd_parent) {
317 /* parent's name + 1 for the "/" */
318 result = dsl_dir_namelen(dd->dd_parent) + 1;
319 }
320
321 if (!MUTEX_HELD(&dd->dd_lock)) {
322 /* see dsl_dir_name */
323 mutex_enter(&dd->dd_lock);
324 result += strlen(dd->dd_myname);
325 mutex_exit(&dd->dd_lock);
326 } else {
327 result += strlen(dd->dd_myname);
328 }
329
330 return (result);
331}
332
333static int
334getcomponent(const char *path, char *component, const char **nextp)
335{
336 char *p;
337
338 if ((path == NULL) || (path[0] == '\0'))
339 return (SET_ERROR(ENOENT));
340 /* This would be a good place to reserve some namespace... */
341 p = strpbrk(path, "/@");
342 if (p && (p[1] == '/' || p[1] == '@')) {
343 /* two separators in a row */
344 return (SET_ERROR(EINVAL));
345 }
346 if (p == NULL || p == path) {
347 /*
348 * if the first thing is an @ or /, it had better be an
349 * @ and it had better not have any more ats or slashes,
350 * and it had better have something after the @.
351 */
352 if (p != NULL &&
353 (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
354 return (SET_ERROR(EINVAL));
355 if (strlen(path) >= MAXNAMELEN)
356 return (SET_ERROR(ENAMETOOLONG));
357 (void) strcpy(component, path);
358 p = NULL;
359 } else if (p[0] == '/') {
360 if (p - path >= MAXNAMELEN)
361 return (SET_ERROR(ENAMETOOLONG));
362 (void) strncpy(component, path, p - path);
363 component[p - path] = '\0';
364 p++;
365 } else if (p[0] == '@') {
366 /*
367 * if the next separator is an @, there better not be
368 * any more slashes.
369 */
370 if (strchr(path, '/'))
371 return (SET_ERROR(EINVAL));
372 if (p - path >= MAXNAMELEN)
373 return (SET_ERROR(ENAMETOOLONG));
374 (void) strncpy(component, path, p - path);
375 component[p - path] = '\0';
376 } else {
377 panic("invalid p=%p", (void *)p);
378 }
379 *nextp = p;
380 return (0);
381}
382
383/*
384 * Return the dsl_dir_t, and possibly the last component which couldn't
385 * be found in *tail. The name must be in the specified dsl_pool_t. This
386 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
387 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
388 * (*tail)[0] == '@' means that the last component is a snapshot.
389 */
390int
391dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
392 dsl_dir_t **ddp, const char **tailp)
393{
394 char buf[MAXNAMELEN];
395 const char *spaname, *next, *nextnext = NULL;
396 int err;
397 dsl_dir_t *dd;
398 uint64_t ddobj;
399
400 err = getcomponent(name, buf, &next);
401 if (err != 0)
402 return (err);
403
404 /* Make sure the name is in the specified pool. */
405 spaname = spa_name(dp->dp_spa);
406 if (strcmp(buf, spaname) != 0)
407 return (SET_ERROR(EINVAL));
408
409 ASSERT(dsl_pool_config_held(dp));
410
411 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
412 if (err != 0) {
413 return (err);
414 }
415
416 while (next != NULL) {
417 dsl_dir_t *child_ds;
418 err = getcomponent(next, buf, &nextnext);
419 if (err != 0)
420 break;
421 ASSERT(next[0] != '\0');
422 if (next[0] == '@')
423 break;
424 dprintf("looking up %s in obj%lld\n",
425 buf, dd->dd_phys->dd_child_dir_zapobj);
426
427 err = zap_lookup(dp->dp_meta_objset,
428 dd->dd_phys->dd_child_dir_zapobj,
429 buf, sizeof (ddobj), 1, &ddobj);
430 if (err != 0) {
431 if (err == ENOENT)
432 err = 0;
433 break;
434 }
435
436 err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
437 if (err != 0)
438 break;
439 dsl_dir_rele(dd, tag);
440 dd = child_ds;
441 next = nextnext;
442 }
443
444 if (err != 0) {
445 dsl_dir_rele(dd, tag);
446 return (err);
447 }
448
449 /*
450 * It's an error if there's more than one component left, or
451 * tailp==NULL and there's any component left.
452 */
453 if (next != NULL &&
454 (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
455 /* bad path name */
456 dsl_dir_rele(dd, tag);
457 dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
458 err = SET_ERROR(ENOENT);
459 }
460 if (tailp != NULL)
461 *tailp = next;
462 *ddp = dd;
463 return (err);
464}
465
466/*
467 * If the counts are already initialized for this filesystem and its
468 * descendants then do nothing, otherwise initialize the counts.
469 *
470 * The counts on this filesystem, and those below, may be uninitialized due to
471 * either the use of a pre-existing pool which did not support the
472 * filesystem/snapshot limit feature, or one in which the feature had not yet
473 * been enabled.
474 *
475 * Recursively descend the filesystem tree and update the filesystem/snapshot
476 * counts on each filesystem below, then update the cumulative count on the
477 * current filesystem. If the filesystem already has a count set on it,
478 * then we know that its counts, and the counts on the filesystems below it,
479 * are already correct, so we don't have to update this filesystem.
480 */
481static void
482dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
483{
484 uint64_t my_fs_cnt = 0;
485 uint64_t my_ss_cnt = 0;
486 dsl_pool_t *dp = dd->dd_pool;
487 objset_t *os = dp->dp_meta_objset;
488 zap_cursor_t *zc;
489 zap_attribute_t *za;
490 dsl_dataset_t *ds;
491
492 ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
493 ASSERT(dsl_pool_config_held(dp));
494 ASSERT(dmu_tx_is_syncing(tx));
495
496 dsl_dir_zapify(dd, tx);
497
498 /*
499 * If the filesystem count has already been initialized then we
500 * don't need to recurse down any further.
501 */
502 if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
503 return;
504
505 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
506 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
507
508 /* Iterate my child dirs */
509 for (zap_cursor_init(zc, os, dd->dd_phys->dd_child_dir_zapobj);
510 zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
511 dsl_dir_t *chld_dd;
512 uint64_t count;
513
514 VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
515 &chld_dd));
516
517 /*
518 * Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
519 * temporary datasets.
520 */
521 if (chld_dd->dd_myname[0] == '$' ||
522 chld_dd->dd_myname[0] == '%') {
523 dsl_dir_rele(chld_dd, FTAG);
524 continue;
525 }
526
527 my_fs_cnt++; /* count this child */
528
529 dsl_dir_init_fs_ss_count(chld_dd, tx);
530
531 VERIFY0(zap_lookup(os, chld_dd->dd_object,
532 DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
533 my_fs_cnt += count;
534 VERIFY0(zap_lookup(os, chld_dd->dd_object,
535 DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
536 my_ss_cnt += count;
537
538 dsl_dir_rele(chld_dd, FTAG);
539 }
540 zap_cursor_fini(zc);
541 /* Count my snapshots (we counted children's snapshots above) */
542 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
543 dd->dd_phys->dd_head_dataset_obj, FTAG, &ds));
544
545 for (zap_cursor_init(zc, os, ds->ds_phys->ds_snapnames_zapobj);
546 zap_cursor_retrieve(zc, za) == 0;
547 zap_cursor_advance(zc)) {
548 /* Don't count temporary snapshots */
549 if (za->za_name[0] != '%')
550 my_ss_cnt++;
551 }
552 zap_cursor_fini(zc);
553
554 dsl_dataset_rele(ds, FTAG);
555
556 kmem_free(zc, sizeof (zap_cursor_t));
557 kmem_free(za, sizeof (zap_attribute_t));
558
559 /* we're in a sync task, update counts */
560 dmu_buf_will_dirty(dd->dd_dbuf, tx);
561 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
562 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
563 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
564 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
565}
566
567static int
568dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
569{
570 char *ddname = (char *)arg;
571 dsl_pool_t *dp = dmu_tx_pool(tx);
572 dsl_dataset_t *ds;
573 dsl_dir_t *dd;
574 int error;
575
576 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
577 if (error != 0)
578 return (error);
579
580 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
581 dsl_dataset_rele(ds, FTAG);
582 return (SET_ERROR(ENOTSUP));
583 }
584
585 dd = ds->ds_dir;
586 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
587 dsl_dir_is_zapified(dd) &&
588 zap_contains(dp->dp_meta_objset, dd->dd_object,
589 DD_FIELD_FILESYSTEM_COUNT) == 0) {
590 dsl_dataset_rele(ds, FTAG);
591 return (SET_ERROR(EALREADY));
592 }
593
594 dsl_dataset_rele(ds, FTAG);
595 return (0);
596}
597
598static void
599dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
600{
601 char *ddname = (char *)arg;
602 dsl_pool_t *dp = dmu_tx_pool(tx);
603 dsl_dataset_t *ds;
604 spa_t *spa;
605
606 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
607
608 spa = dsl_dataset_get_spa(ds);
609
610 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
611 /*
612 * Since the feature was not active and we're now setting a
613 * limit, increment the feature-active counter so that the
614 * feature becomes active for the first time.
615 *
616 * We are already in a sync task so we can update the MOS.
617 */
618 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
619 }
620
621 /*
622 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
623 * we need to ensure the counts are correct. Descend down the tree from
624 * this point and update all of the counts to be accurate.
625 */
626 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
627
628 dsl_dataset_rele(ds, FTAG);
629}
630
631/*
632 * Make sure the feature is enabled and activate it if necessary.
633 * Since we're setting a limit, ensure the on-disk counts are valid.
634 * This is only called by the ioctl path when setting a limit value.
635 *
636 * We do not need to validate the new limit, since users who can change the
637 * limit are also allowed to exceed the limit.
638 */
639int
640dsl_dir_activate_fs_ss_limit(const char *ddname)
641{
642 int error;
643
644 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
645 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0);
646
647 if (error == EALREADY)
648 error = 0;
649
650 return (error);
651}
652
653/*
654 * Used to determine if the filesystem_limit or snapshot_limit should be
655 * enforced. We allow the limit to be exceeded if the user has permission to
656 * write the property value. We pass in the creds that we got in the open
657 * context since we will always be the GZ root in syncing context. We also have
658 * to handle the case where we are allowed to change the limit on the current
659 * dataset, but there may be another limit in the tree above.
660 *
661 * We can never modify these two properties within a non-global zone. In
662 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
663 * can't use that function since we are already holding the dp_config_rwlock.
664 * In addition, we already have the dd and dealing with snapshots is simplified
665 * in this code.
666 */
667
668typedef enum {
669 ENFORCE_ALWAYS,
670 ENFORCE_NEVER,
671 ENFORCE_ABOVE
672} enforce_res_t;
673
674static enforce_res_t
675dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
676{
677 enforce_res_t enforce = ENFORCE_ALWAYS;
678 uint64_t obj;
679 dsl_dataset_t *ds;
680 uint64_t zoned;
681
682 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
683 prop == ZFS_PROP_SNAPSHOT_LIMIT);
684
685#ifdef _KERNEL
686#ifdef __FreeBSD__
687 if (jailed(cr))
688#else
689 if (crgetzoneid(cr) != GLOBAL_ZONEID)
690#endif
691 return (ENFORCE_ALWAYS);
692
693 if (secpolicy_zfs(cr) == 0)
694 return (ENFORCE_NEVER);
695#endif
696
697 if ((obj = dd->dd_phys->dd_head_dataset_obj) == 0)
698 return (ENFORCE_ALWAYS);
699
700 ASSERT(dsl_pool_config_held(dd->dd_pool));
701
702 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
703 return (ENFORCE_ALWAYS);
704
705 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
706 /* Only root can access zoned fs's from the GZ */
707 enforce = ENFORCE_ALWAYS;
708 } else {
709 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
710 enforce = ENFORCE_ABOVE;
711 }
712
713 dsl_dataset_rele(ds, FTAG);
714 return (enforce);
715}
716
717/*
718 * Check if adding additional child filesystem(s) would exceed any filesystem
719 * limits or adding additional snapshot(s) would exceed any snapshot limits.
720 * The prop argument indicates which limit to check.
721 *
722 * Note that all filesystem limits up to the root (or the highest
723 * initialized) filesystem or the given ancestor must be satisfied.
724 */
725int
726dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
727 dsl_dir_t *ancestor, cred_t *cr)
728{
729 objset_t *os = dd->dd_pool->dp_meta_objset;
730 uint64_t limit, count;
731 char *count_prop;
732 enforce_res_t enforce;
733 int err = 0;
734
735 ASSERT(dsl_pool_config_held(dd->dd_pool));
736 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
737 prop == ZFS_PROP_SNAPSHOT_LIMIT);
738
739 /*
740 * If we're allowed to change the limit, don't enforce the limit
741 * e.g. this can happen if a snapshot is taken by an administrative
742 * user in the global zone (i.e. a recursive snapshot by root).
743 * However, we must handle the case of delegated permissions where we
744 * are allowed to change the limit on the current dataset, but there
745 * is another limit in the tree above.
746 */
747 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
748 if (enforce == ENFORCE_NEVER)
749 return (0);
750
751 /*
752 * e.g. if renaming a dataset with no snapshots, count adjustment
753 * is 0.
754 */
755 if (delta == 0)
756 return (0);
757
758 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
759 /*
760 * We don't enforce the limit for temporary snapshots. This is
761 * indicated by a NULL cred_t argument.
762 */
763 if (cr == NULL)
764 return (0);
765
766 count_prop = DD_FIELD_SNAPSHOT_COUNT;
767 } else {
768 count_prop = DD_FIELD_FILESYSTEM_COUNT;
769 }
770
771 /*
772 * If an ancestor has been provided, stop checking the limit once we
773 * hit that dir. We need this during rename so that we don't overcount
774 * the check once we recurse up to the common ancestor.
775 */
776 if (ancestor == dd)
777 return (0);
778
779 /*
780 * If we hit an uninitialized node while recursing up the tree, we can
781 * stop since we know there is no limit here (or above). The counts are
782 * not valid on this node and we know we won't touch this node's counts.
783 */
784 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
785 count_prop, sizeof (count), 1, &count) == ENOENT)
786 return (0);
787
788 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
789 B_FALSE);
790 if (err != 0)
791 return (err);
792
793 /* Is there a limit which we've hit? */
794 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
795 return (SET_ERROR(EDQUOT));
796
797 if (dd->dd_parent != NULL)
798 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
799 ancestor, cr);
800
801 return (err);
802}
803
804/*
805 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
806 * parents. When a new filesystem/snapshot is created, increment the count on
807 * all parents, and when a filesystem/snapshot is destroyed, decrement the
808 * count.
809 */
810void
811dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
812 dmu_tx_t *tx)
813{
814 int err;
815 objset_t *os = dd->dd_pool->dp_meta_objset;
816 uint64_t count;
817
818 ASSERT(dsl_pool_config_held(dd->dd_pool));
819 ASSERT(dmu_tx_is_syncing(tx));
820 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
821 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
822
823 /*
824 * When we receive an incremental stream into a filesystem that already
825 * exists, a temporary clone is created. We don't count this temporary
826 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
827 * $MOS & $ORIGIN) objsets.
828 */
829 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
830 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
831 return;
832
833 /*
834 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
835 */
836 if (delta == 0)
837 return;
838
839 /*
840 * If we hit an uninitialized node while recursing up the tree, we can
841 * stop since we know the counts are not valid on this node and we
842 * know we shouldn't touch this node's counts. An uninitialized count
843 * on the node indicates that either the feature has not yet been
844 * activated or there are no limits on this part of the tree.
845 */
846 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
847 prop, sizeof (count), 1, &count)) == ENOENT)
848 return;
849 VERIFY0(err);
850
851 count += delta;
852 /* Use a signed verify to make sure we're not neg. */
853 VERIFY3S(count, >=, 0);
854
855 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
856 tx));
857
858 /* Roll up this additional count into our ancestors */
859 if (dd->dd_parent != NULL)
860 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
861}
862
863uint64_t
864dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
865 dmu_tx_t *tx)
866{
867 objset_t *mos = dp->dp_meta_objset;
868 uint64_t ddobj;
869 dsl_dir_phys_t *ddphys;
870 dmu_buf_t *dbuf;
871
872 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
873 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
874 if (pds) {
875 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
876 name, sizeof (uint64_t), 1, &ddobj, tx));
877 } else {
878 /* it's the root dir */
879 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
880 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
881 }
882 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
883 dmu_buf_will_dirty(dbuf, tx);
884 ddphys = dbuf->db_data;
885
886 ddphys->dd_creation_time = gethrestime_sec();
887 if (pds) {
888 ddphys->dd_parent_obj = pds->dd_object;
889
890 /* update the filesystem counts */
891 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
892 }
893 ddphys->dd_props_zapobj = zap_create(mos,
894 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
895 ddphys->dd_child_dir_zapobj = zap_create(mos,
896 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
897 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
898 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
899 dmu_buf_rele(dbuf, FTAG);
900
901 return (ddobj);
902}
903
904boolean_t
905dsl_dir_is_clone(dsl_dir_t *dd)
906{
907 return (dd->dd_phys->dd_origin_obj &&
908 (dd->dd_pool->dp_origin_snap == NULL ||
909 dd->dd_phys->dd_origin_obj !=
910 dd->dd_pool->dp_origin_snap->ds_object));
911}
912
913void
914dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
915{
916 mutex_enter(&dd->dd_lock);
917 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
918 dd->dd_phys->dd_used_bytes);
919 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
920 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
921 dd->dd_phys->dd_reserved);
922 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
923 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
924 (dd->dd_phys->dd_uncompressed_bytes * 100 /
925 dd->dd_phys->dd_compressed_bytes));
926 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
927 dd->dd_phys->dd_uncompressed_bytes);
928 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
929 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
930 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
931 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
932 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
933 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
934 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
935 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
936 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
937 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
938 }
939 mutex_exit(&dd->dd_lock);
940
941 if (dsl_dir_is_zapified(dd)) {
942 uint64_t count;
943 objset_t *os = dd->dd_pool->dp_meta_objset;
944
945 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
946 sizeof (count), 1, &count) == 0) {
947 dsl_prop_nvlist_add_uint64(nv,
948 ZFS_PROP_FILESYSTEM_COUNT, count);
949 }
950 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
951 sizeof (count), 1, &count) == 0) {
952 dsl_prop_nvlist_add_uint64(nv,
953 ZFS_PROP_SNAPSHOT_COUNT, count);
954 }
955 }
956
957 if (dsl_dir_is_clone(dd)) {
958 dsl_dataset_t *ds;
959 char buf[MAXNAMELEN];
960
961 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
962 dd->dd_phys->dd_origin_obj, FTAG, &ds));
963 dsl_dataset_name(ds, buf);
964 dsl_dataset_rele(ds, FTAG);
965 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
966 }
967}
968
969void
970dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
971{
972 dsl_pool_t *dp = dd->dd_pool;
973
974 ASSERT(dd->dd_phys);
975
976 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
977 /* up the hold count until we can be written out */
978 dmu_buf_add_ref(dd->dd_dbuf, dd);
979 }
980}
981
982static int64_t
983parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
984{
985 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
986 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
987 return (new_accounted - old_accounted);
988}
989
990void
991dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
992{
993 ASSERT(dmu_tx_is_syncing(tx));
994
995 mutex_enter(&dd->dd_lock);
996 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
997 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
998 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
999 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1000 mutex_exit(&dd->dd_lock);
1001
1002 /* release the hold from dsl_dir_dirty */
1003 dmu_buf_rele(dd->dd_dbuf, dd);
1004}
1005
1006static uint64_t
1007dsl_dir_space_towrite(dsl_dir_t *dd)
1008{
1009 uint64_t space = 0;
1010 int i;
1011
1012 ASSERT(MUTEX_HELD(&dd->dd_lock));
1013
1014 for (i = 0; i < TXG_SIZE; i++) {
1015 space += dd->dd_space_towrite[i&TXG_MASK];
1016 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1017 }
1018 return (space);
1019}
1020
1021/*
1022 * How much space would dd have available if ancestor had delta applied
1023 * to it? If ondiskonly is set, we're only interested in what's
1024 * on-disk, not estimated pending changes.
1025 */
1026uint64_t
1027dsl_dir_space_available(dsl_dir_t *dd,
1028 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1029{
1030 uint64_t parentspace, myspace, quota, used;
1031
1032 /*
1033 * If there are no restrictions otherwise, assume we have
1034 * unlimited space available.
1035 */
1036 quota = UINT64_MAX;
1037 parentspace = UINT64_MAX;
1038
1039 if (dd->dd_parent != NULL) {
1040 parentspace = dsl_dir_space_available(dd->dd_parent,
1041 ancestor, delta, ondiskonly);
1042 }
1043
1044 mutex_enter(&dd->dd_lock);
1045 if (dd->dd_phys->dd_quota != 0)
1046 quota = dd->dd_phys->dd_quota;
1047 used = dd->dd_phys->dd_used_bytes;
1048 if (!ondiskonly)
1049 used += dsl_dir_space_towrite(dd);
1050
1051 if (dd->dd_parent == NULL) {
1052 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1053 quota = MIN(quota, poolsize);
1054 }
1055
1056 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
1057 /*
1058 * We have some space reserved, in addition to what our
1059 * parent gave us.
1060 */
1061 parentspace += dd->dd_phys->dd_reserved - used;
1062 }
1063
1064 if (dd == ancestor) {
1065 ASSERT(delta <= 0);
1066 ASSERT(used >= -delta);
1067 used += delta;
1068 if (parentspace != UINT64_MAX)
1069 parentspace -= delta;
1070 }
1071
1072 if (used > quota) {
1073 /* over quota */
1074 myspace = 0;
1075 } else {
1076 /*
1077 * the lesser of the space provided by our parent and
1078 * the space left in our quota
1079 */
1080 myspace = MIN(parentspace, quota - used);
1081 }
1082
1083 mutex_exit(&dd->dd_lock);
1084
1085 return (myspace);
1086}
1087
1088struct tempreserve {
1089 list_node_t tr_node;
1090 dsl_dir_t *tr_ds;
1091 uint64_t tr_size;
1092};
1093
1094static int
1095dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1096 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1097 dmu_tx_t *tx, boolean_t first)
1098{
1099 uint64_t txg = tx->tx_txg;
1100 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1101 uint64_t deferred = 0;
1102 struct tempreserve *tr;
1103 int retval = EDQUOT;
1104 int txgidx = txg & TXG_MASK;
1105 int i;
1106 uint64_t ref_rsrv = 0;
1107
1108 ASSERT3U(txg, !=, 0);
1109 ASSERT3S(asize, >, 0);
1110
1111 mutex_enter(&dd->dd_lock);
1112
1113 /*
1114 * Check against the dsl_dir's quota. We don't add in the delta
1115 * when checking for over-quota because they get one free hit.
1116 */
1117 est_inflight = dsl_dir_space_towrite(dd);
1118 for (i = 0; i < TXG_SIZE; i++)
1119 est_inflight += dd->dd_tempreserved[i];
1120 used_on_disk = dd->dd_phys->dd_used_bytes;
1121
1122 /*
1123 * On the first iteration, fetch the dataset's used-on-disk and
1124 * refreservation values. Also, if checkrefquota is set, test if
1125 * allocating this space would exceed the dataset's refquota.
1126 */
1127 if (first && tx->tx_objset) {
1128 int error;
1129 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1130
1131 error = dsl_dataset_check_quota(ds, checkrefquota,
1132 asize, est_inflight, &used_on_disk, &ref_rsrv);
1133 if (error) {
1134 mutex_exit(&dd->dd_lock);
1135 return (error);
1136 }
1137 }
1138
1139 /*
1140 * If this transaction will result in a net free of space,
1141 * we want to let it through.
1142 */
1143 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
1144 quota = UINT64_MAX;
1145 else
1146 quota = dd->dd_phys->dd_quota;
1147
1148 /*
1149 * Adjust the quota against the actual pool size at the root
1150 * minus any outstanding deferred frees.
1151 * To ensure that it's possible to remove files from a full
1152 * pool without inducing transient overcommits, we throttle
1153 * netfree transactions against a quota that is slightly larger,
1154 * but still within the pool's allocation slop. In cases where
1155 * we're very close to full, this will allow a steady trickle of
1156 * removes to get through.
1157 */
1158 if (dd->dd_parent == NULL) {
1159 spa_t *spa = dd->dd_pool->dp_spa;
1160 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1161 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1162 if (poolsize - deferred < quota) {
1163 quota = poolsize - deferred;
1164 retval = ENOSPC;
1165 }
1166 }
1167
1168 /*
1169 * If they are requesting more space, and our current estimate
1170 * is over quota, they get to try again unless the actual
1171 * on-disk is over quota and there are no pending changes (which
1172 * may free up space for us).
1173 */
1174 if (used_on_disk + est_inflight >= quota) {
1175 if (est_inflight > 0 || used_on_disk < quota ||
1176 (retval == ENOSPC && used_on_disk < quota + deferred))
1177 retval = ERESTART;
1178 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1179 "quota=%lluK tr=%lluK err=%d\n",
1180 used_on_disk>>10, est_inflight>>10,
1181 quota>>10, asize>>10, retval);
1182 mutex_exit(&dd->dd_lock);
1183 return (SET_ERROR(retval));
1184 }
1185
1186 /* We need to up our estimated delta before dropping dd_lock */
1187 dd->dd_tempreserved[txgidx] += asize;
1188
1189 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1190 asize - ref_rsrv);
1191 mutex_exit(&dd->dd_lock);
1192
1193 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1194 tr->tr_ds = dd;
1195 tr->tr_size = asize;
1196 list_insert_tail(tr_list, tr);
1197
1198 /* see if it's OK with our parent */
1199 if (dd->dd_parent && parent_rsrv) {
1200 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
1201
1202 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1203 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1204 } else {
1205 return (0);
1206 }
1207}
1208
1209/*
1210 * Reserve space in this dsl_dir, to be used in this tx's txg.
1211 * After the space has been dirtied (and dsl_dir_willuse_space()
1212 * has been called), the reservation should be canceled, using
1213 * dsl_dir_tempreserve_clear().
1214 */
1215int
1216dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1217 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1218{
1219 int err;
1220 list_t *tr_list;
1221
1222 if (asize == 0) {
1223 *tr_cookiep = NULL;
1224 return (0);
1225 }
1226
1227 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1228 list_create(tr_list, sizeof (struct tempreserve),
1229 offsetof(struct tempreserve, tr_node));
1230 ASSERT3S(asize, >, 0);
1231 ASSERT3S(fsize, >=, 0);
1232
1233 err = arc_tempreserve_space(lsize, tx->tx_txg);
1234 if (err == 0) {
1235 struct tempreserve *tr;
1236
1237 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1238 tr->tr_size = lsize;
1239 list_insert_tail(tr_list, tr);
1240 } else {
1241 if (err == EAGAIN) {
1242 /*
1243 * If arc_memory_throttle() detected that pageout
1244 * is running and we are low on memory, we delay new
1245 * non-pageout transactions to give pageout an
1246 * advantage.
1247 *
1248 * It is unfortunate to be delaying while the caller's
1249 * locks are held.
1250 */
1251 txg_delay(dd->dd_pool, tx->tx_txg,
1252 MSEC2NSEC(10), MSEC2NSEC(10));
1253 err = SET_ERROR(ERESTART);
1254 }
1255 }
1256
1257 if (err == 0) {
1258 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1259 FALSE, asize > usize, tr_list, tx, TRUE);
1260 }
1261
1262 if (err != 0)
1263 dsl_dir_tempreserve_clear(tr_list, tx);
1264 else
1265 *tr_cookiep = tr_list;
1266
1267 return (err);
1268}
1269
1270/*
1271 * Clear a temporary reservation that we previously made with
1272 * dsl_dir_tempreserve_space().
1273 */
1274void
1275dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1276{
1277 int txgidx = tx->tx_txg & TXG_MASK;
1278 list_t *tr_list = tr_cookie;
1279 struct tempreserve *tr;
1280
1281 ASSERT3U(tx->tx_txg, !=, 0);
1282
1283 if (tr_cookie == NULL)
1284 return;
1285
1286 while ((tr = list_head(tr_list)) != NULL) {
1287 if (tr->tr_ds) {
1288 mutex_enter(&tr->tr_ds->dd_lock);
1289 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1290 tr->tr_size);
1291 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1292 mutex_exit(&tr->tr_ds->dd_lock);
1293 } else {
1294 arc_tempreserve_clear(tr->tr_size);
1295 }
1296 list_remove(tr_list, tr);
1297 kmem_free(tr, sizeof (struct tempreserve));
1298 }
1299
1300 kmem_free(tr_list, sizeof (list_t));
1301}
1302
1303/*
1304 * This should be called from open context when we think we're going to write
1305 * or free space, for example when dirtying data. Be conservative; it's okay
1306 * to write less space or free more, but we don't want to write more or free
1307 * less than the amount specified.
1308 */
1309void
1310dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1311{
1312 int64_t parent_space;
1313 uint64_t est_used;
1314
1315 mutex_enter(&dd->dd_lock);
1316 if (space > 0)
1317 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1318
1319 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
1320 parent_space = parent_delta(dd, est_used, space);
1321 mutex_exit(&dd->dd_lock);
1322
1323 /* Make sure that we clean up dd_space_to* */
1324 dsl_dir_dirty(dd, tx);
1325
1326 /* XXX this is potentially expensive and unnecessary... */
1327 if (parent_space && dd->dd_parent)
1328 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1329}
1330
1331/* call from syncing context when we actually write/free space for this dd */
1332void
1333dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1334 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1335{
1336 int64_t accounted_delta;
1337
1338 /*
1339 * dsl_dataset_set_refreservation_sync_impl() calls this with
1340 * dd_lock held, so that it can atomically update
1341 * ds->ds_reserved and the dsl_dir accounting, so that
1342 * dsl_dataset_check_quota() can see dataset and dir accounting
1343 * consistently.
1344 */
1345 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1346
1347 ASSERT(dmu_tx_is_syncing(tx));
1348 ASSERT(type < DD_USED_NUM);
1349
1350 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1351
1352 if (needlock)
1353 mutex_enter(&dd->dd_lock);
1354 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
1355 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
1356 ASSERT(compressed >= 0 ||
1357 dd->dd_phys->dd_compressed_bytes >= -compressed);
1358 ASSERT(uncompressed >= 0 ||
1359 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
1360 dd->dd_phys->dd_used_bytes += used;
1361 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
1362 dd->dd_phys->dd_compressed_bytes += compressed;
1363
1364 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1365 ASSERT(used > 0 ||
1366 dd->dd_phys->dd_used_breakdown[type] >= -used);
1367 dd->dd_phys->dd_used_breakdown[type] += used;
1368#ifdef DEBUG
1369 dd_used_t t;
1370 uint64_t u = 0;
1371 for (t = 0; t < DD_USED_NUM; t++)
1372 u += dd->dd_phys->dd_used_breakdown[t];
1373 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
1374#endif
1375 }
1376 if (needlock)
1377 mutex_exit(&dd->dd_lock);
1378
1379 if (dd->dd_parent != NULL) {
1380 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1381 accounted_delta, compressed, uncompressed, tx);
1382 dsl_dir_transfer_space(dd->dd_parent,
1383 used - accounted_delta,
1384 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1385 }
1386}
1387
1388void
1389dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1390 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1391{
1392 ASSERT(dmu_tx_is_syncing(tx));
1393 ASSERT(oldtype < DD_USED_NUM);
1394 ASSERT(newtype < DD_USED_NUM);
1395
1396 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
1397 return;
1398
1399 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1400 mutex_enter(&dd->dd_lock);
1401 ASSERT(delta > 0 ?
1402 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1403 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1404 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1405 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1406 dd->dd_phys->dd_used_breakdown[newtype] += delta;
1407 mutex_exit(&dd->dd_lock);
1408}
1409
1410typedef struct dsl_dir_set_qr_arg {
1411 const char *ddsqra_name;
1412 zprop_source_t ddsqra_source;
1413 uint64_t ddsqra_value;
1414} dsl_dir_set_qr_arg_t;
1415
1416static int
1417dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1418{
1419 dsl_dir_set_qr_arg_t *ddsqra = arg;
1420 dsl_pool_t *dp = dmu_tx_pool(tx);
1421 dsl_dataset_t *ds;
1422 int error;
1423 uint64_t towrite, newval;
1424
1425 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1426 if (error != 0)
1427 return (error);
1428
1429 error = dsl_prop_predict(ds->ds_dir, "quota",
1430 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1431 if (error != 0) {
1432 dsl_dataset_rele(ds, FTAG);
1433 return (error);
1434 }
1435
1436 if (newval == 0) {
1437 dsl_dataset_rele(ds, FTAG);
1438 return (0);
1439 }
1440
1441 mutex_enter(&ds->ds_dir->dd_lock);
1442 /*
1443 * If we are doing the preliminary check in open context, and
1444 * there are pending changes, then don't fail it, since the
1445 * pending changes could under-estimate the amount of space to be
1446 * freed up.
1447 */
1448 towrite = dsl_dir_space_towrite(ds->ds_dir);
1449 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1450 (newval < ds->ds_dir->dd_phys->dd_reserved ||
1451 newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
1452 error = SET_ERROR(ENOSPC);
1453 }
1454 mutex_exit(&ds->ds_dir->dd_lock);
1455 dsl_dataset_rele(ds, FTAG);
1456 return (error);
1457}
1458
1459static void
1460dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1461{
1462 dsl_dir_set_qr_arg_t *ddsqra = arg;
1463 dsl_pool_t *dp = dmu_tx_pool(tx);
1464 dsl_dataset_t *ds;
1465 uint64_t newval;
1466
1467 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1468
1469 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1470 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1471 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1472 &ddsqra->ddsqra_value, tx);
1473
1474 VERIFY0(dsl_prop_get_int_ds(ds,
1475 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1476 } else {
1477 newval = ddsqra->ddsqra_value;
1478 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1479 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1480 }
1481
1482 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1483 mutex_enter(&ds->ds_dir->dd_lock);
1484 ds->ds_dir->dd_phys->dd_quota = newval;
1485 mutex_exit(&ds->ds_dir->dd_lock);
1486 dsl_dataset_rele(ds, FTAG);
1487}
1488
1489int
1490dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1491{
1492 dsl_dir_set_qr_arg_t ddsqra;
1493
1494 ddsqra.ddsqra_name = ddname;
1495 ddsqra.ddsqra_source = source;
1496 ddsqra.ddsqra_value = quota;
1497
1498 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1499 dsl_dir_set_quota_sync, &ddsqra, 0));
1500}
1501
1502int
1503dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1504{
1505 dsl_dir_set_qr_arg_t *ddsqra = arg;
1506 dsl_pool_t *dp = dmu_tx_pool(tx);
1507 dsl_dataset_t *ds;
1508 dsl_dir_t *dd;
1509 uint64_t newval, used, avail;
1510 int error;
1511
1512 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1513 if (error != 0)
1514 return (error);
1515 dd = ds->ds_dir;
1516
1517 /*
1518 * If we are doing the preliminary check in open context, the
1519 * space estimates may be inaccurate.
1520 */
1521 if (!dmu_tx_is_syncing(tx)) {
1522 dsl_dataset_rele(ds, FTAG);
1523 return (0);
1524 }
1525
1526 error = dsl_prop_predict(ds->ds_dir,
1527 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1528 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1529 if (error != 0) {
1530 dsl_dataset_rele(ds, FTAG);
1531 return (error);
1532 }
1533
1534 mutex_enter(&dd->dd_lock);
1535 used = dd->dd_phys->dd_used_bytes;
1536 mutex_exit(&dd->dd_lock);
1537
1538 if (dd->dd_parent) {
1539 avail = dsl_dir_space_available(dd->dd_parent,
1540 NULL, 0, FALSE);
1541 } else {
1542 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1543 }
1544
1545 if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1546 uint64_t delta = MAX(used, newval) -
1547 MAX(used, dd->dd_phys->dd_reserved);
1548
1549 if (delta > avail ||
1550 (dd->dd_phys->dd_quota > 0 &&
1551 newval > dd->dd_phys->dd_quota))
1552 error = SET_ERROR(ENOSPC);
1553 }
1554
1555 dsl_dataset_rele(ds, FTAG);
1556 return (error);
1557}
1558
1559void
1560dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1561{
1562 uint64_t used;
1563 int64_t delta;
1564
1565 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1566
1567 mutex_enter(&dd->dd_lock);
1568 used = dd->dd_phys->dd_used_bytes;
1569 delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1570 dd->dd_phys->dd_reserved = value;
1571
1572 if (dd->dd_parent != NULL) {
1573 /* Roll up this additional usage into our ancestors */
1574 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1575 delta, 0, 0, tx);
1576 }
1577 mutex_exit(&dd->dd_lock);
1578}
1579
1580static void
1581dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1582{
1583 dsl_dir_set_qr_arg_t *ddsqra = arg;
1584 dsl_pool_t *dp = dmu_tx_pool(tx);
1585 dsl_dataset_t *ds;
1586 uint64_t newval;
1587
1588 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1589
1590 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1591 dsl_prop_set_sync_impl(ds,
1592 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1593 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1594 &ddsqra->ddsqra_value, tx);
1595
1596 VERIFY0(dsl_prop_get_int_ds(ds,
1597 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1598 } else {
1599 newval = ddsqra->ddsqra_value;
1600 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1601 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1602 (longlong_t)newval);
1603 }
1604
1605 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1606 dsl_dataset_rele(ds, FTAG);
1607}
1608
1609int
1610dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1611 uint64_t reservation)
1612{
1613 dsl_dir_set_qr_arg_t ddsqra;
1614
1615 ddsqra.ddsqra_name = ddname;
1616 ddsqra.ddsqra_source = source;
1617 ddsqra.ddsqra_value = reservation;
1618
1619 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1620 dsl_dir_set_reservation_sync, &ddsqra, 0));
1621}
1622
1623static dsl_dir_t *
1624closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1625{
1626 for (; ds1; ds1 = ds1->dd_parent) {
1627 dsl_dir_t *dd;
1628 for (dd = ds2; dd; dd = dd->dd_parent) {
1629 if (ds1 == dd)
1630 return (dd);
1631 }
1632 }
1633 return (NULL);
1634}
1635
1636/*
1637 * If delta is applied to dd, how much of that delta would be applied to
1638 * ancestor? Syncing context only.
1639 */
1640static int64_t
1641would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1642{
1643 if (dd == ancestor)
1644 return (delta);
1645
1646 mutex_enter(&dd->dd_lock);
1647 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1648 mutex_exit(&dd->dd_lock);
1649 return (would_change(dd->dd_parent, delta, ancestor));
1650}
1651
1652typedef struct dsl_dir_rename_arg {
1653 const char *ddra_oldname;
1654 const char *ddra_newname;
1655 cred_t *ddra_cred;
1656} dsl_dir_rename_arg_t;
1657
1658/* ARGSUSED */
1659static int
1660dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1661{
1662 int *deltap = arg;
1663 char namebuf[MAXNAMELEN];
1664
1665 dsl_dataset_name(ds, namebuf);
1666
1667 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1668 return (SET_ERROR(ENAMETOOLONG));
1669 return (0);
1670}
1671
1672static int
1673dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1674{
1675 dsl_dir_rename_arg_t *ddra = arg;
1676 dsl_pool_t *dp = dmu_tx_pool(tx);
1677 dsl_dir_t *dd, *newparent;
1678 const char *mynewname;
1679 int error;
1680 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1681
1682 /* target dir should exist */
1683 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1684 if (error != 0)
1685 return (error);
1686
1687 /* new parent should exist */
1688 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1689 &newparent, &mynewname);
1690 if (error != 0) {
1691 dsl_dir_rele(dd, FTAG);
1692 return (error);
1693 }
1694
1695 /* can't rename to different pool */
1696 if (dd->dd_pool != newparent->dd_pool) {
1697 dsl_dir_rele(newparent, FTAG);
1698 dsl_dir_rele(dd, FTAG);
1699 return (SET_ERROR(ENXIO));
1700 }
1701
1702 /* new name should not already exist */
1703 if (mynewname == NULL) {
1704 dsl_dir_rele(newparent, FTAG);
1705 dsl_dir_rele(dd, FTAG);
1706 return (SET_ERROR(EEXIST));
1707 }
1708
1709 /* if the name length is growing, validate child name lengths */
1710 if (delta > 0) {
1711 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1712 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1713 if (error != 0) {
1714 dsl_dir_rele(newparent, FTAG);
1715 dsl_dir_rele(dd, FTAG);
1716 return (error);
1717 }
1718 }
1719
1720 if (dmu_tx_is_syncing(tx)) {
1721 if (spa_feature_is_active(dp->dp_spa,
1722 SPA_FEATURE_FS_SS_LIMIT)) {
1723 /*
1724 * Although this is the check function and we don't
1725 * normally make on-disk changes in check functions,
1726 * we need to do that here.
1727 *
1728 * Ensure this portion of the tree's counts have been
1729 * initialized in case the new parent has limits set.
1730 */
1731 dsl_dir_init_fs_ss_count(dd, tx);
1732 }
1733 }
1734
1735 if (newparent != dd->dd_parent) {
1736 /* is there enough space? */
1737 uint64_t myspace =
1738 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1739 objset_t *os = dd->dd_pool->dp_meta_objset;
1740 uint64_t fs_cnt = 0;
1741 uint64_t ss_cnt = 0;
1742
1743 if (dsl_dir_is_zapified(dd)) {
1744 int err;
1745
1746 err = zap_lookup(os, dd->dd_object,
1747 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1748 &fs_cnt);
1749 if (err != ENOENT && err != 0) {
1750 dsl_dir_rele(newparent, FTAG);
1751 dsl_dir_rele(dd, FTAG);
1752 return (err);
1753 }
1754
1755 /*
1756 * have to add 1 for the filesystem itself that we're
1757 * moving
1758 */
1759 fs_cnt++;
1760
1761 err = zap_lookup(os, dd->dd_object,
1762 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1763 &ss_cnt);
1764 if (err != ENOENT && err != 0) {
1765 dsl_dir_rele(newparent, FTAG);
1766 dsl_dir_rele(dd, FTAG);
1767 return (err);
1768 }
1769 }
1770
1771 /* no rename into our descendant */
1772 if (closest_common_ancestor(dd, newparent) == dd) {
1773 dsl_dir_rele(newparent, FTAG);
1774 dsl_dir_rele(dd, FTAG);
1775 return (SET_ERROR(EINVAL));
1776 }
1777
1778 error = dsl_dir_transfer_possible(dd->dd_parent,
1779 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1780 if (error != 0) {
1781 dsl_dir_rele(newparent, FTAG);
1782 dsl_dir_rele(dd, FTAG);
1783 return (error);
1784 }
1785 }
1786
1787 dsl_dir_rele(newparent, FTAG);
1788 dsl_dir_rele(dd, FTAG);
1789 return (0);
1790}
1791
1792static void
1793dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1794{
1795 dsl_dir_rename_arg_t *ddra = arg;
1796 dsl_pool_t *dp = dmu_tx_pool(tx);
1797 dsl_dir_t *dd, *newparent;
1798 const char *mynewname;
1799 int error;
1800 objset_t *mos = dp->dp_meta_objset;
1801
1802 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1803 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1804 &mynewname));
1805
1806 /* Log this before we change the name. */
1807 spa_history_log_internal_dd(dd, "rename", tx,
1808 "-> %s", ddra->ddra_newname);
1809
1810 if (newparent != dd->dd_parent) {
1811 objset_t *os = dd->dd_pool->dp_meta_objset;
1812 uint64_t fs_cnt = 0;
1813 uint64_t ss_cnt = 0;
1814
1815 /*
1816 * We already made sure the dd counts were initialized in the
1817 * check function.
1818 */
1819 if (spa_feature_is_active(dp->dp_spa,
1820 SPA_FEATURE_FS_SS_LIMIT)) {
1821 VERIFY0(zap_lookup(os, dd->dd_object,
1822 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1823 &fs_cnt));
1824 /* add 1 for the filesystem itself that we're moving */
1825 fs_cnt++;
1826
1827 VERIFY0(zap_lookup(os, dd->dd_object,
1828 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1829 &ss_cnt));
1830 }
1831
1832 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1833 DD_FIELD_FILESYSTEM_COUNT, tx);
1834 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1835 DD_FIELD_FILESYSTEM_COUNT, tx);
1836
1837 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1838 DD_FIELD_SNAPSHOT_COUNT, tx);
1839 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1840 DD_FIELD_SNAPSHOT_COUNT, tx);
1841
1842 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1843 -dd->dd_phys->dd_used_bytes,
1844 -dd->dd_phys->dd_compressed_bytes,
1845 -dd->dd_phys->dd_uncompressed_bytes, tx);
1846 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1847 dd->dd_phys->dd_used_bytes,
1848 dd->dd_phys->dd_compressed_bytes,
1849 dd->dd_phys->dd_uncompressed_bytes, tx);
1850
1851 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1852 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1853 dd->dd_phys->dd_used_bytes;
1854
1855 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1856 -unused_rsrv, 0, 0, tx);
1857 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1858 unused_rsrv, 0, 0, tx);
1859 }
1860 }
1861
1862 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1863
1864 /* remove from old parent zapobj */
1865 error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1866 dd->dd_myname, tx);
1867 ASSERT0(error);
1868
1869 (void) strcpy(dd->dd_myname, mynewname);
1870 dsl_dir_rele(dd->dd_parent, dd);
1871 dd->dd_phys->dd_parent_obj = newparent->dd_object;
1872 VERIFY0(dsl_dir_hold_obj(dp,
1873 newparent->dd_object, NULL, dd, &dd->dd_parent));
1874
1875 /* add to new parent zapobj */
1876 VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1877 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1878
1879#ifdef __FreeBSD__
1880#ifdef _KERNEL
1881 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1882 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1883#endif
1884#endif
1885
1886 dsl_prop_notify_all(dd);
1887
1888 dsl_dir_rele(newparent, FTAG);
1889 dsl_dir_rele(dd, FTAG);
1890}
1891
1892int
1893dsl_dir_rename(const char *oldname, const char *newname)
1894{
1895 dsl_dir_rename_arg_t ddra;
1896
1897 ddra.ddra_oldname = oldname;
1898 ddra.ddra_newname = newname;
1899 ddra.ddra_cred = CRED();
1900
1901 return (dsl_sync_task(oldname,
1902 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1903}
1904
1905int
1906dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1907 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1908{
1909 dsl_dir_t *ancestor;
1910 int64_t adelta;
1911 uint64_t avail;
1912 int err;
1913
1914 ancestor = closest_common_ancestor(sdd, tdd);
1915 adelta = would_change(sdd, -space, ancestor);
1916 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1917 if (avail < space)
1918 return (SET_ERROR(ENOSPC));
1919
1920 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1921 ancestor, cr);
1922 if (err != 0)
1923 return (err);
1924 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1925 ancestor, cr);
1926 if (err != 0)
1927 return (err);
1928
1929 return (0);
1930}
1931
1932timestruc_t
1933dsl_dir_snap_cmtime(dsl_dir_t *dd)
1934{
1935 timestruc_t t;
1936
1937 mutex_enter(&dd->dd_lock);
1938 t = dd->dd_snap_cmtime;
1939 mutex_exit(&dd->dd_lock);
1940
1941 return (t);
1942}
1943
1944void
1945dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1946{
1947 timestruc_t t;
1948
1949 gethrestime(&t);
1950 mutex_enter(&dd->dd_lock);
1951 dd->dd_snap_cmtime = t;
1952 mutex_exit(&dd->dd_lock);
1953}
1954
1955void
1956dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1957{
1958 objset_t *mos = dd->dd_pool->dp_meta_objset;
1959 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1960}
1961
1962boolean_t
1963dsl_dir_is_zapified(dsl_dir_t *dd)
1964{
1965 dmu_object_info_t doi;
1966
1967 dmu_object_info_from_db(dd->dd_dbuf, &doi);
1968 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
1969}