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) 2012, 2014 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 ZFS_SPACE_CHECK_RESERVED);
647
648 if (error == EALREADY)
649 error = 0;
650
651 return (error);
652}
653
654/*
655 * Used to determine if the filesystem_limit or snapshot_limit should be
656 * enforced. We allow the limit to be exceeded if the user has permission to
657 * write the property value. We pass in the creds that we got in the open
658 * context since we will always be the GZ root in syncing context. We also have
659 * to handle the case where we are allowed to change the limit on the current
660 * dataset, but there may be another limit in the tree above.
661 *
662 * We can never modify these two properties within a non-global zone. In
663 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
664 * can't use that function since we are already holding the dp_config_rwlock.
665 * In addition, we already have the dd and dealing with snapshots is simplified
666 * in this code.
667 */
668
669typedef enum {
670 ENFORCE_ALWAYS,
671 ENFORCE_NEVER,
672 ENFORCE_ABOVE
673} enforce_res_t;
674
675static enforce_res_t
676dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
677{
678 enforce_res_t enforce = ENFORCE_ALWAYS;
679 uint64_t obj;
680 dsl_dataset_t *ds;
681 uint64_t zoned;
682
683 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
684 prop == ZFS_PROP_SNAPSHOT_LIMIT);
685
686#ifdef _KERNEL
687#ifdef __FreeBSD__
688 if (jailed(cr))
689#else
690 if (crgetzoneid(cr) != GLOBAL_ZONEID)
691#endif
692 return (ENFORCE_ALWAYS);
693
694 if (secpolicy_zfs(cr) == 0)
695 return (ENFORCE_NEVER);
696#endif
697
698 if ((obj = dd->dd_phys->dd_head_dataset_obj) == 0)
699 return (ENFORCE_ALWAYS);
700
701 ASSERT(dsl_pool_config_held(dd->dd_pool));
702
703 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
704 return (ENFORCE_ALWAYS);
705
706 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
707 /* Only root can access zoned fs's from the GZ */
708 enforce = ENFORCE_ALWAYS;
709 } else {
710 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
711 enforce = ENFORCE_ABOVE;
712 }
713
714 dsl_dataset_rele(ds, FTAG);
715 return (enforce);
716}
717
718/*
719 * Check if adding additional child filesystem(s) would exceed any filesystem
720 * limits or adding additional snapshot(s) would exceed any snapshot limits.
721 * The prop argument indicates which limit to check.
722 *
723 * Note that all filesystem limits up to the root (or the highest
724 * initialized) filesystem or the given ancestor must be satisfied.
725 */
726int
727dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
728 dsl_dir_t *ancestor, cred_t *cr)
729{
730 objset_t *os = dd->dd_pool->dp_meta_objset;
731 uint64_t limit, count;
732 char *count_prop;
733 enforce_res_t enforce;
734 int err = 0;
735
736 ASSERT(dsl_pool_config_held(dd->dd_pool));
737 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
738 prop == ZFS_PROP_SNAPSHOT_LIMIT);
739
740 /*
741 * If we're allowed to change the limit, don't enforce the limit
742 * e.g. this can happen if a snapshot is taken by an administrative
743 * user in the global zone (i.e. a recursive snapshot by root).
744 * However, we must handle the case of delegated permissions where we
745 * are allowed to change the limit on the current dataset, but there
746 * is another limit in the tree above.
747 */
748 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
749 if (enforce == ENFORCE_NEVER)
750 return (0);
751
752 /*
753 * e.g. if renaming a dataset with no snapshots, count adjustment
754 * is 0.
755 */
756 if (delta == 0)
757 return (0);
758
759 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
760 /*
761 * We don't enforce the limit for temporary snapshots. This is
762 * indicated by a NULL cred_t argument.
763 */
764 if (cr == NULL)
765 return (0);
766
767 count_prop = DD_FIELD_SNAPSHOT_COUNT;
768 } else {
769 count_prop = DD_FIELD_FILESYSTEM_COUNT;
770 }
771
772 /*
773 * If an ancestor has been provided, stop checking the limit once we
774 * hit that dir. We need this during rename so that we don't overcount
775 * the check once we recurse up to the common ancestor.
776 */
777 if (ancestor == dd)
778 return (0);
779
780 /*
781 * If we hit an uninitialized node while recursing up the tree, we can
782 * stop since we know there is no limit here (or above). The counts are
783 * not valid on this node and we know we won't touch this node's counts.
784 */
785 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
786 count_prop, sizeof (count), 1, &count) == ENOENT)
787 return (0);
788
789 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
790 B_FALSE);
791 if (err != 0)
792 return (err);
793
794 /* Is there a limit which we've hit? */
795 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
796 return (SET_ERROR(EDQUOT));
797
798 if (dd->dd_parent != NULL)
799 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
800 ancestor, cr);
801
802 return (err);
803}
804
805/*
806 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
807 * parents. When a new filesystem/snapshot is created, increment the count on
808 * all parents, and when a filesystem/snapshot is destroyed, decrement the
809 * count.
810 */
811void
812dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
813 dmu_tx_t *tx)
814{
815 int err;
816 objset_t *os = dd->dd_pool->dp_meta_objset;
817 uint64_t count;
818
819 ASSERT(dsl_pool_config_held(dd->dd_pool));
820 ASSERT(dmu_tx_is_syncing(tx));
821 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
822 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
823
824 /*
825 * When we receive an incremental stream into a filesystem that already
826 * exists, a temporary clone is created. We don't count this temporary
827 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
828 * $MOS & $ORIGIN) objsets.
829 */
830 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
831 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
832 return;
833
834 /*
835 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
836 */
837 if (delta == 0)
838 return;
839
840 /*
841 * If we hit an uninitialized node while recursing up the tree, we can
842 * stop since we know the counts are not valid on this node and we
843 * know we shouldn't touch this node's counts. An uninitialized count
844 * on the node indicates that either the feature has not yet been
845 * activated or there are no limits on this part of the tree.
846 */
847 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
848 prop, sizeof (count), 1, &count)) == ENOENT)
849 return;
850 VERIFY0(err);
851
852 count += delta;
853 /* Use a signed verify to make sure we're not neg. */
854 VERIFY3S(count, >=, 0);
855
856 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
857 tx));
858
859 /* Roll up this additional count into our ancestors */
860 if (dd->dd_parent != NULL)
861 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
862}
863
864uint64_t
865dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
866 dmu_tx_t *tx)
867{
868 objset_t *mos = dp->dp_meta_objset;
869 uint64_t ddobj;
870 dsl_dir_phys_t *ddphys;
871 dmu_buf_t *dbuf;
872
873 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
874 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
875 if (pds) {
876 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
877 name, sizeof (uint64_t), 1, &ddobj, tx));
878 } else {
879 /* it's the root dir */
880 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
881 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
882 }
883 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
884 dmu_buf_will_dirty(dbuf, tx);
885 ddphys = dbuf->db_data;
886
887 ddphys->dd_creation_time = gethrestime_sec();
888 if (pds) {
889 ddphys->dd_parent_obj = pds->dd_object;
890
891 /* update the filesystem counts */
892 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
893 }
894 ddphys->dd_props_zapobj = zap_create(mos,
895 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
896 ddphys->dd_child_dir_zapobj = zap_create(mos,
897 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
898 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
899 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
900 dmu_buf_rele(dbuf, FTAG);
901
902 return (ddobj);
903}
904
905boolean_t
906dsl_dir_is_clone(dsl_dir_t *dd)
907{
908 return (dd->dd_phys->dd_origin_obj &&
909 (dd->dd_pool->dp_origin_snap == NULL ||
910 dd->dd_phys->dd_origin_obj !=
911 dd->dd_pool->dp_origin_snap->ds_object));
912}
913
914void
915dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
916{
917 mutex_enter(&dd->dd_lock);
918 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
919 dd->dd_phys->dd_used_bytes);
920 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
921 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
922 dd->dd_phys->dd_reserved);
923 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
924 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
925 (dd->dd_phys->dd_uncompressed_bytes * 100 /
926 dd->dd_phys->dd_compressed_bytes));
927 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
928 dd->dd_phys->dd_uncompressed_bytes);
929 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
930 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
931 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
932 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
933 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
934 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
935 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
936 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
937 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
938 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
939 }
940 mutex_exit(&dd->dd_lock);
941
942 if (dsl_dir_is_zapified(dd)) {
943 uint64_t count;
944 objset_t *os = dd->dd_pool->dp_meta_objset;
945
946 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
947 sizeof (count), 1, &count) == 0) {
948 dsl_prop_nvlist_add_uint64(nv,
949 ZFS_PROP_FILESYSTEM_COUNT, count);
950 }
951 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
952 sizeof (count), 1, &count) == 0) {
953 dsl_prop_nvlist_add_uint64(nv,
954 ZFS_PROP_SNAPSHOT_COUNT, count);
955 }
956 }
957
958 if (dsl_dir_is_clone(dd)) {
959 dsl_dataset_t *ds;
960 char buf[MAXNAMELEN];
961
962 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
963 dd->dd_phys->dd_origin_obj, FTAG, &ds));
964 dsl_dataset_name(ds, buf);
965 dsl_dataset_rele(ds, FTAG);
966 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
967 }
968}
969
970void
971dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
972{
973 dsl_pool_t *dp = dd->dd_pool;
974
975 ASSERT(dd->dd_phys);
976
977 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
978 /* up the hold count until we can be written out */
979 dmu_buf_add_ref(dd->dd_dbuf, dd);
980 }
981}
982
983static int64_t
984parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
985{
986 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
987 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
988 return (new_accounted - old_accounted);
989}
990
991void
992dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
993{
994 ASSERT(dmu_tx_is_syncing(tx));
995
996 mutex_enter(&dd->dd_lock);
997 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
998 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
999 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1000 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1001 mutex_exit(&dd->dd_lock);
1002
1003 /* release the hold from dsl_dir_dirty */
1004 dmu_buf_rele(dd->dd_dbuf, dd);
1005}
1006
1007static uint64_t
1008dsl_dir_space_towrite(dsl_dir_t *dd)
1009{
1010 uint64_t space = 0;
1011 int i;
1012
1013 ASSERT(MUTEX_HELD(&dd->dd_lock));
1014
1015 for (i = 0; i < TXG_SIZE; i++) {
1016 space += dd->dd_space_towrite[i&TXG_MASK];
1017 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1018 }
1019 return (space);
1020}
1021
1022/*
1023 * How much space would dd have available if ancestor had delta applied
1024 * to it? If ondiskonly is set, we're only interested in what's
1025 * on-disk, not estimated pending changes.
1026 */
1027uint64_t
1028dsl_dir_space_available(dsl_dir_t *dd,
1029 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1030{
1031 uint64_t parentspace, myspace, quota, used;
1032
1033 /*
1034 * If there are no restrictions otherwise, assume we have
1035 * unlimited space available.
1036 */
1037 quota = UINT64_MAX;
1038 parentspace = UINT64_MAX;
1039
1040 if (dd->dd_parent != NULL) {
1041 parentspace = dsl_dir_space_available(dd->dd_parent,
1042 ancestor, delta, ondiskonly);
1043 }
1044
1045 mutex_enter(&dd->dd_lock);
1046 if (dd->dd_phys->dd_quota != 0)
1047 quota = dd->dd_phys->dd_quota;
1048 used = dd->dd_phys->dd_used_bytes;
1049 if (!ondiskonly)
1050 used += dsl_dir_space_towrite(dd);
1051
1052 if (dd->dd_parent == NULL) {
1053 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1054 quota = MIN(quota, poolsize);
1055 }
1056
1057 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
1058 /*
1059 * We have some space reserved, in addition to what our
1060 * parent gave us.
1061 */
1062 parentspace += dd->dd_phys->dd_reserved - used;
1063 }
1064
1065 if (dd == ancestor) {
1066 ASSERT(delta <= 0);
1067 ASSERT(used >= -delta);
1068 used += delta;
1069 if (parentspace != UINT64_MAX)
1070 parentspace -= delta;
1071 }
1072
1073 if (used > quota) {
1074 /* over quota */
1075 myspace = 0;
1076 } else {
1077 /*
1078 * the lesser of the space provided by our parent and
1079 * the space left in our quota
1080 */
1081 myspace = MIN(parentspace, quota - used);
1082 }
1083
1084 mutex_exit(&dd->dd_lock);
1085
1086 return (myspace);
1087}
1088
1089struct tempreserve {
1090 list_node_t tr_node;
1091 dsl_dir_t *tr_ds;
1092 uint64_t tr_size;
1093};
1094
1095static int
1096dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1097 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1098 dmu_tx_t *tx, boolean_t first)
1099{
1100 uint64_t txg = tx->tx_txg;
1101 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1102 uint64_t deferred = 0;
1103 struct tempreserve *tr;
1104 int retval = EDQUOT;
1105 int txgidx = txg & TXG_MASK;
1106 int i;
1107 uint64_t ref_rsrv = 0;
1108
1109 ASSERT3U(txg, !=, 0);
1110 ASSERT3S(asize, >, 0);
1111
1112 mutex_enter(&dd->dd_lock);
1113
1114 /*
1115 * Check against the dsl_dir's quota. We don't add in the delta
1116 * when checking for over-quota because they get one free hit.
1117 */
1118 est_inflight = dsl_dir_space_towrite(dd);
1119 for (i = 0; i < TXG_SIZE; i++)
1120 est_inflight += dd->dd_tempreserved[i];
1121 used_on_disk = dd->dd_phys->dd_used_bytes;
1122
1123 /*
1124 * On the first iteration, fetch the dataset's used-on-disk and
1125 * refreservation values. Also, if checkrefquota is set, test if
1126 * allocating this space would exceed the dataset's refquota.
1127 */
1128 if (first && tx->tx_objset) {
1129 int error;
1130 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1131
1132 error = dsl_dataset_check_quota(ds, checkrefquota,
1133 asize, est_inflight, &used_on_disk, &ref_rsrv);
1134 if (error) {
1135 mutex_exit(&dd->dd_lock);
1136 return (error);
1137 }
1138 }
1139
1140 /*
1141 * If this transaction will result in a net free of space,
1142 * we want to let it through.
1143 */
1144 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
1145 quota = UINT64_MAX;
1146 else
1147 quota = dd->dd_phys->dd_quota;
1148
1149 /*
1150 * Adjust the quota against the actual pool size at the root
1151 * minus any outstanding deferred frees.
1152 * To ensure that it's possible to remove files from a full
1153 * pool without inducing transient overcommits, we throttle
1154 * netfree transactions against a quota that is slightly larger,
1155 * but still within the pool's allocation slop. In cases where
1156 * we're very close to full, this will allow a steady trickle of
1157 * removes to get through.
1158 */
1159 if (dd->dd_parent == NULL) {
1160 spa_t *spa = dd->dd_pool->dp_spa;
1161 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1162 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1163 if (poolsize - deferred < quota) {
1164 quota = poolsize - deferred;
1165 retval = ENOSPC;
1166 }
1167 }
1168
1169 /*
1170 * If they are requesting more space, and our current estimate
1171 * is over quota, they get to try again unless the actual
1172 * on-disk is over quota and there are no pending changes (which
1173 * may free up space for us).
1174 */
1175 if (used_on_disk + est_inflight >= quota) {
1176 if (est_inflight > 0 || used_on_disk < quota ||
1177 (retval == ENOSPC && used_on_disk < quota + deferred))
1178 retval = ERESTART;
1179 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1180 "quota=%lluK tr=%lluK err=%d\n",
1181 used_on_disk>>10, est_inflight>>10,
1182 quota>>10, asize>>10, retval);
1183 mutex_exit(&dd->dd_lock);
1184 return (SET_ERROR(retval));
1185 }
1186
1187 /* We need to up our estimated delta before dropping dd_lock */
1188 dd->dd_tempreserved[txgidx] += asize;
1189
1190 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1191 asize - ref_rsrv);
1192 mutex_exit(&dd->dd_lock);
1193
1194 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1195 tr->tr_ds = dd;
1196 tr->tr_size = asize;
1197 list_insert_tail(tr_list, tr);
1198
1199 /* see if it's OK with our parent */
1200 if (dd->dd_parent && parent_rsrv) {
1201 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
1202
1203 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1204 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1205 } else {
1206 return (0);
1207 }
1208}
1209
1210/*
1211 * Reserve space in this dsl_dir, to be used in this tx's txg.
1212 * After the space has been dirtied (and dsl_dir_willuse_space()
1213 * has been called), the reservation should be canceled, using
1214 * dsl_dir_tempreserve_clear().
1215 */
1216int
1217dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1218 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1219{
1220 int err;
1221 list_t *tr_list;
1222
1223 if (asize == 0) {
1224 *tr_cookiep = NULL;
1225 return (0);
1226 }
1227
1228 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1229 list_create(tr_list, sizeof (struct tempreserve),
1230 offsetof(struct tempreserve, tr_node));
1231 ASSERT3S(asize, >, 0);
1232 ASSERT3S(fsize, >=, 0);
1233
1234 err = arc_tempreserve_space(lsize, tx->tx_txg);
1235 if (err == 0) {
1236 struct tempreserve *tr;
1237
1238 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1239 tr->tr_size = lsize;
1240 list_insert_tail(tr_list, tr);
1241 } else {
1242 if (err == EAGAIN) {
1243 /*
1244 * If arc_memory_throttle() detected that pageout
1245 * is running and we are low on memory, we delay new
1246 * non-pageout transactions to give pageout an
1247 * advantage.
1248 *
1249 * It is unfortunate to be delaying while the caller's
1250 * locks are held.
1251 */
1252 txg_delay(dd->dd_pool, tx->tx_txg,
1253 MSEC2NSEC(10), MSEC2NSEC(10));
1254 err = SET_ERROR(ERESTART);
1255 }
1256 }
1257
1258 if (err == 0) {
1259 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1260 FALSE, asize > usize, tr_list, tx, TRUE);
1261 }
1262
1263 if (err != 0)
1264 dsl_dir_tempreserve_clear(tr_list, tx);
1265 else
1266 *tr_cookiep = tr_list;
1267
1268 return (err);
1269}
1270
1271/*
1272 * Clear a temporary reservation that we previously made with
1273 * dsl_dir_tempreserve_space().
1274 */
1275void
1276dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1277{
1278 int txgidx = tx->tx_txg & TXG_MASK;
1279 list_t *tr_list = tr_cookie;
1280 struct tempreserve *tr;
1281
1282 ASSERT3U(tx->tx_txg, !=, 0);
1283
1284 if (tr_cookie == NULL)
1285 return;
1286
1287 while ((tr = list_head(tr_list)) != NULL) {
1288 if (tr->tr_ds) {
1289 mutex_enter(&tr->tr_ds->dd_lock);
1290 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1291 tr->tr_size);
1292 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1293 mutex_exit(&tr->tr_ds->dd_lock);
1294 } else {
1295 arc_tempreserve_clear(tr->tr_size);
1296 }
1297 list_remove(tr_list, tr);
1298 kmem_free(tr, sizeof (struct tempreserve));
1299 }
1300
1301 kmem_free(tr_list, sizeof (list_t));
1302}
1303
1304/*
1305 * This should be called from open context when we think we're going to write
1306 * or free space, for example when dirtying data. Be conservative; it's okay
1307 * to write less space or free more, but we don't want to write more or free
1308 * less than the amount specified.
1309 */
1310void
1311dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1312{
1313 int64_t parent_space;
1314 uint64_t est_used;
1315
1316 mutex_enter(&dd->dd_lock);
1317 if (space > 0)
1318 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1319
1320 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
1321 parent_space = parent_delta(dd, est_used, space);
1322 mutex_exit(&dd->dd_lock);
1323
1324 /* Make sure that we clean up dd_space_to* */
1325 dsl_dir_dirty(dd, tx);
1326
1327 /* XXX this is potentially expensive and unnecessary... */
1328 if (parent_space && dd->dd_parent)
1329 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1330}
1331
1332/* call from syncing context when we actually write/free space for this dd */
1333void
1334dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1335 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1336{
1337 int64_t accounted_delta;
1338
1339 /*
1340 * dsl_dataset_set_refreservation_sync_impl() calls this with
1341 * dd_lock held, so that it can atomically update
1342 * ds->ds_reserved and the dsl_dir accounting, so that
1343 * dsl_dataset_check_quota() can see dataset and dir accounting
1344 * consistently.
1345 */
1346 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1347
1348 ASSERT(dmu_tx_is_syncing(tx));
1349 ASSERT(type < DD_USED_NUM);
1350
1351 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1352
1353 if (needlock)
1354 mutex_enter(&dd->dd_lock);
1355 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
1356 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
1357 ASSERT(compressed >= 0 ||
1358 dd->dd_phys->dd_compressed_bytes >= -compressed);
1359 ASSERT(uncompressed >= 0 ||
1360 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
1361 dd->dd_phys->dd_used_bytes += used;
1362 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
1363 dd->dd_phys->dd_compressed_bytes += compressed;
1364
1365 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1366 ASSERT(used > 0 ||
1367 dd->dd_phys->dd_used_breakdown[type] >= -used);
1368 dd->dd_phys->dd_used_breakdown[type] += used;
1369#ifdef DEBUG
1370 dd_used_t t;
1371 uint64_t u = 0;
1372 for (t = 0; t < DD_USED_NUM; t++)
1373 u += dd->dd_phys->dd_used_breakdown[t];
1374 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
1375#endif
1376 }
1377 if (needlock)
1378 mutex_exit(&dd->dd_lock);
1379
1380 if (dd->dd_parent != NULL) {
1381 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1382 accounted_delta, compressed, uncompressed, tx);
1383 dsl_dir_transfer_space(dd->dd_parent,
1384 used - accounted_delta,
1385 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1386 }
1387}
1388
1389void
1390dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1391 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1392{
1393 ASSERT(dmu_tx_is_syncing(tx));
1394 ASSERT(oldtype < DD_USED_NUM);
1395 ASSERT(newtype < DD_USED_NUM);
1396
1397 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
1398 return;
1399
1400 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1401 mutex_enter(&dd->dd_lock);
1402 ASSERT(delta > 0 ?
1403 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1404 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1405 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1406 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1407 dd->dd_phys->dd_used_breakdown[newtype] += delta;
1408 mutex_exit(&dd->dd_lock);
1409}
1410
1411typedef struct dsl_dir_set_qr_arg {
1412 const char *ddsqra_name;
1413 zprop_source_t ddsqra_source;
1414 uint64_t ddsqra_value;
1415} dsl_dir_set_qr_arg_t;
1416
1417static int
1418dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1419{
1420 dsl_dir_set_qr_arg_t *ddsqra = arg;
1421 dsl_pool_t *dp = dmu_tx_pool(tx);
1422 dsl_dataset_t *ds;
1423 int error;
1424 uint64_t towrite, newval;
1425
1426 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1427 if (error != 0)
1428 return (error);
1429
1430 error = dsl_prop_predict(ds->ds_dir, "quota",
1431 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1432 if (error != 0) {
1433 dsl_dataset_rele(ds, FTAG);
1434 return (error);
1435 }
1436
1437 if (newval == 0) {
1438 dsl_dataset_rele(ds, FTAG);
1439 return (0);
1440 }
1441
1442 mutex_enter(&ds->ds_dir->dd_lock);
1443 /*
1444 * If we are doing the preliminary check in open context, and
1445 * there are pending changes, then don't fail it, since the
1446 * pending changes could under-estimate the amount of space to be
1447 * freed up.
1448 */
1449 towrite = dsl_dir_space_towrite(ds->ds_dir);
1450 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1451 (newval < ds->ds_dir->dd_phys->dd_reserved ||
1452 newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
1453 error = SET_ERROR(ENOSPC);
1454 }
1455 mutex_exit(&ds->ds_dir->dd_lock);
1456 dsl_dataset_rele(ds, FTAG);
1457 return (error);
1458}
1459
1460static void
1461dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1462{
1463 dsl_dir_set_qr_arg_t *ddsqra = arg;
1464 dsl_pool_t *dp = dmu_tx_pool(tx);
1465 dsl_dataset_t *ds;
1466 uint64_t newval;
1467
1468 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1469
1470 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1471 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1472 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1473 &ddsqra->ddsqra_value, tx);
1474
1475 VERIFY0(dsl_prop_get_int_ds(ds,
1476 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1477 } else {
1478 newval = ddsqra->ddsqra_value;
1479 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1480 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1481 }
1482
1483 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1484 mutex_enter(&ds->ds_dir->dd_lock);
1485 ds->ds_dir->dd_phys->dd_quota = newval;
1486 mutex_exit(&ds->ds_dir->dd_lock);
1487 dsl_dataset_rele(ds, FTAG);
1488}
1489
1490int
1491dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1492{
1493 dsl_dir_set_qr_arg_t ddsqra;
1494
1495 ddsqra.ddsqra_name = ddname;
1496 ddsqra.ddsqra_source = source;
1497 ddsqra.ddsqra_value = quota;
1498
1499 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1500 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1501}
1502
1503int
1504dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1505{
1506 dsl_dir_set_qr_arg_t *ddsqra = arg;
1507 dsl_pool_t *dp = dmu_tx_pool(tx);
1508 dsl_dataset_t *ds;
1509 dsl_dir_t *dd;
1510 uint64_t newval, used, avail;
1511 int error;
1512
1513 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1514 if (error != 0)
1515 return (error);
1516 dd = ds->ds_dir;
1517
1518 /*
1519 * If we are doing the preliminary check in open context, the
1520 * space estimates may be inaccurate.
1521 */
1522 if (!dmu_tx_is_syncing(tx)) {
1523 dsl_dataset_rele(ds, FTAG);
1524 return (0);
1525 }
1526
1527 error = dsl_prop_predict(ds->ds_dir,
1528 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1529 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1530 if (error != 0) {
1531 dsl_dataset_rele(ds, FTAG);
1532 return (error);
1533 }
1534
1535 mutex_enter(&dd->dd_lock);
1536 used = dd->dd_phys->dd_used_bytes;
1537 mutex_exit(&dd->dd_lock);
1538
1539 if (dd->dd_parent) {
1540 avail = dsl_dir_space_available(dd->dd_parent,
1541 NULL, 0, FALSE);
1542 } else {
1543 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1544 }
1545
1546 if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1547 uint64_t delta = MAX(used, newval) -
1548 MAX(used, dd->dd_phys->dd_reserved);
1549
1550 if (delta > avail ||
1551 (dd->dd_phys->dd_quota > 0 &&
1552 newval > dd->dd_phys->dd_quota))
1553 error = SET_ERROR(ENOSPC);
1554 }
1555
1556 dsl_dataset_rele(ds, FTAG);
1557 return (error);
1558}
1559
1560void
1561dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1562{
1563 uint64_t used;
1564 int64_t delta;
1565
1566 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1567
1568 mutex_enter(&dd->dd_lock);
1569 used = dd->dd_phys->dd_used_bytes;
1570 delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1571 dd->dd_phys->dd_reserved = value;
1572
1573 if (dd->dd_parent != NULL) {
1574 /* Roll up this additional usage into our ancestors */
1575 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1576 delta, 0, 0, tx);
1577 }
1578 mutex_exit(&dd->dd_lock);
1579}
1580
1581static void
1582dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1583{
1584 dsl_dir_set_qr_arg_t *ddsqra = arg;
1585 dsl_pool_t *dp = dmu_tx_pool(tx);
1586 dsl_dataset_t *ds;
1587 uint64_t newval;
1588
1589 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1590
1591 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1592 dsl_prop_set_sync_impl(ds,
1593 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1594 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1595 &ddsqra->ddsqra_value, tx);
1596
1597 VERIFY0(dsl_prop_get_int_ds(ds,
1598 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1599 } else {
1600 newval = ddsqra->ddsqra_value;
1601 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1602 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1603 (longlong_t)newval);
1604 }
1605
1606 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1607 dsl_dataset_rele(ds, FTAG);
1608}
1609
1610int
1611dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1612 uint64_t reservation)
1613{
1614 dsl_dir_set_qr_arg_t ddsqra;
1615
1616 ddsqra.ddsqra_name = ddname;
1617 ddsqra.ddsqra_source = source;
1618 ddsqra.ddsqra_value = reservation;
1619
1620 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1621 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1622}
1623
1624static dsl_dir_t *
1625closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1626{
1627 for (; ds1; ds1 = ds1->dd_parent) {
1628 dsl_dir_t *dd;
1629 for (dd = ds2; dd; dd = dd->dd_parent) {
1630 if (ds1 == dd)
1631 return (dd);
1632 }
1633 }
1634 return (NULL);
1635}
1636
1637/*
1638 * If delta is applied to dd, how much of that delta would be applied to
1639 * ancestor? Syncing context only.
1640 */
1641static int64_t
1642would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1643{
1644 if (dd == ancestor)
1645 return (delta);
1646
1647 mutex_enter(&dd->dd_lock);
1648 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1649 mutex_exit(&dd->dd_lock);
1650 return (would_change(dd->dd_parent, delta, ancestor));
1651}
1652
1653typedef struct dsl_dir_rename_arg {
1654 const char *ddra_oldname;
1655 const char *ddra_newname;
1656 cred_t *ddra_cred;
1657} dsl_dir_rename_arg_t;
1658
1659/* ARGSUSED */
1660static int
1661dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1662{
1663 int *deltap = arg;
1664 char namebuf[MAXNAMELEN];
1665
1666 dsl_dataset_name(ds, namebuf);
1667
1668 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1669 return (SET_ERROR(ENAMETOOLONG));
1670 return (0);
1671}
1672
1673static int
1674dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1675{
1676 dsl_dir_rename_arg_t *ddra = arg;
1677 dsl_pool_t *dp = dmu_tx_pool(tx);
1678 dsl_dir_t *dd, *newparent;
1679 const char *mynewname;
1680 int error;
1681 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1682
1683 /* target dir should exist */
1684 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1685 if (error != 0)
1686 return (error);
1687
1688 /* new parent should exist */
1689 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1690 &newparent, &mynewname);
1691 if (error != 0) {
1692 dsl_dir_rele(dd, FTAG);
1693 return (error);
1694 }
1695
1696 /* can't rename to different pool */
1697 if (dd->dd_pool != newparent->dd_pool) {
1698 dsl_dir_rele(newparent, FTAG);
1699 dsl_dir_rele(dd, FTAG);
1700 return (SET_ERROR(ENXIO));
1701 }
1702
1703 /* new name should not already exist */
1704 if (mynewname == NULL) {
1705 dsl_dir_rele(newparent, FTAG);
1706 dsl_dir_rele(dd, FTAG);
1707 return (SET_ERROR(EEXIST));
1708 }
1709
1710 /* if the name length is growing, validate child name lengths */
1711 if (delta > 0) {
1712 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1713 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1714 if (error != 0) {
1715 dsl_dir_rele(newparent, FTAG);
1716 dsl_dir_rele(dd, FTAG);
1717 return (error);
1718 }
1719 }
1720
1721 if (dmu_tx_is_syncing(tx)) {
1722 if (spa_feature_is_active(dp->dp_spa,
1723 SPA_FEATURE_FS_SS_LIMIT)) {
1724 /*
1725 * Although this is the check function and we don't
1726 * normally make on-disk changes in check functions,
1727 * we need to do that here.
1728 *
1729 * Ensure this portion of the tree's counts have been
1730 * initialized in case the new parent has limits set.
1731 */
1732 dsl_dir_init_fs_ss_count(dd, tx);
1733 }
1734 }
1735
1736 if (newparent != dd->dd_parent) {
1737 /* is there enough space? */
1738 uint64_t myspace =
1739 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1740 objset_t *os = dd->dd_pool->dp_meta_objset;
1741 uint64_t fs_cnt = 0;
1742 uint64_t ss_cnt = 0;
1743
1744 if (dsl_dir_is_zapified(dd)) {
1745 int err;
1746
1747 err = zap_lookup(os, dd->dd_object,
1748 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1749 &fs_cnt);
1750 if (err != ENOENT && err != 0) {
1751 dsl_dir_rele(newparent, FTAG);
1752 dsl_dir_rele(dd, FTAG);
1753 return (err);
1754 }
1755
1756 /*
1757 * have to add 1 for the filesystem itself that we're
1758 * moving
1759 */
1760 fs_cnt++;
1761
1762 err = zap_lookup(os, dd->dd_object,
1763 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1764 &ss_cnt);
1765 if (err != ENOENT && err != 0) {
1766 dsl_dir_rele(newparent, FTAG);
1767 dsl_dir_rele(dd, FTAG);
1768 return (err);
1769 }
1770 }
1771
1772 /* no rename into our descendant */
1773 if (closest_common_ancestor(dd, newparent) == dd) {
1774 dsl_dir_rele(newparent, FTAG);
1775 dsl_dir_rele(dd, FTAG);
1776 return (SET_ERROR(EINVAL));
1777 }
1778
1779 error = dsl_dir_transfer_possible(dd->dd_parent,
1780 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1781 if (error != 0) {
1782 dsl_dir_rele(newparent, FTAG);
1783 dsl_dir_rele(dd, FTAG);
1784 return (error);
1785 }
1786 }
1787
1788 dsl_dir_rele(newparent, FTAG);
1789 dsl_dir_rele(dd, FTAG);
1790 return (0);
1791}
1792
1793static void
1794dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1795{
1796 dsl_dir_rename_arg_t *ddra = arg;
1797 dsl_pool_t *dp = dmu_tx_pool(tx);
1798 dsl_dir_t *dd, *newparent;
1799 const char *mynewname;
1800 int error;
1801 objset_t *mos = dp->dp_meta_objset;
1802
1803 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1804 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1805 &mynewname));
1806
1807 /* Log this before we change the name. */
1808 spa_history_log_internal_dd(dd, "rename", tx,
1809 "-> %s", ddra->ddra_newname);
1810
1811 if (newparent != dd->dd_parent) {
1812 objset_t *os = dd->dd_pool->dp_meta_objset;
1813 uint64_t fs_cnt = 0;
1814 uint64_t ss_cnt = 0;
1815
1816 /*
1817 * We already made sure the dd counts were initialized in the
1818 * check function.
1819 */
1820 if (spa_feature_is_active(dp->dp_spa,
1821 SPA_FEATURE_FS_SS_LIMIT)) {
1822 VERIFY0(zap_lookup(os, dd->dd_object,
1823 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1824 &fs_cnt));
1825 /* add 1 for the filesystem itself that we're moving */
1826 fs_cnt++;
1827
1828 VERIFY0(zap_lookup(os, dd->dd_object,
1829 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1830 &ss_cnt));
1831 }
1832
1833 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1834 DD_FIELD_FILESYSTEM_COUNT, tx);
1835 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1836 DD_FIELD_FILESYSTEM_COUNT, tx);
1837
1838 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1839 DD_FIELD_SNAPSHOT_COUNT, tx);
1840 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1841 DD_FIELD_SNAPSHOT_COUNT, tx);
1842
1843 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1844 -dd->dd_phys->dd_used_bytes,
1845 -dd->dd_phys->dd_compressed_bytes,
1846 -dd->dd_phys->dd_uncompressed_bytes, tx);
1847 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1848 dd->dd_phys->dd_used_bytes,
1849 dd->dd_phys->dd_compressed_bytes,
1850 dd->dd_phys->dd_uncompressed_bytes, tx);
1851
1852 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1853 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1854 dd->dd_phys->dd_used_bytes;
1855
1856 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1857 -unused_rsrv, 0, 0, tx);
1858 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1859 unused_rsrv, 0, 0, tx);
1860 }
1861 }
1862
1863 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1864
1865 /* remove from old parent zapobj */
1866 error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1867 dd->dd_myname, tx);
1868 ASSERT0(error);
1869
1870 (void) strcpy(dd->dd_myname, mynewname);
1871 dsl_dir_rele(dd->dd_parent, dd);
1872 dd->dd_phys->dd_parent_obj = newparent->dd_object;
1873 VERIFY0(dsl_dir_hold_obj(dp,
1874 newparent->dd_object, NULL, dd, &dd->dd_parent));
1875
1876 /* add to new parent zapobj */
1877 VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1878 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1879
1880#ifdef __FreeBSD__
1881#ifdef _KERNEL
1882 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1883 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1884#endif
1885#endif
1886
1887 dsl_prop_notify_all(dd);
1888
1889 dsl_dir_rele(newparent, FTAG);
1890 dsl_dir_rele(dd, FTAG);
1891}
1892
1893int
1894dsl_dir_rename(const char *oldname, const char *newname)
1895{
1896 dsl_dir_rename_arg_t ddra;
1897
1898 ddra.ddra_oldname = oldname;
1899 ddra.ddra_newname = newname;
1900 ddra.ddra_cred = CRED();
1901
1902 return (dsl_sync_task(oldname,
1903 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
1904 3, ZFS_SPACE_CHECK_RESERVED));
1905}
1906
1907int
1908dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1909 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1910{
1911 dsl_dir_t *ancestor;
1912 int64_t adelta;
1913 uint64_t avail;
1914 int err;
1915
1916 ancestor = closest_common_ancestor(sdd, tdd);
1917 adelta = would_change(sdd, -space, ancestor);
1918 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1919 if (avail < space)
1920 return (SET_ERROR(ENOSPC));
1921
1922 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1923 ancestor, cr);
1924 if (err != 0)
1925 return (err);
1926 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1927 ancestor, cr);
1928 if (err != 0)
1929 return (err);
1930
1931 return (0);
1932}
1933
1934timestruc_t
1935dsl_dir_snap_cmtime(dsl_dir_t *dd)
1936{
1937 timestruc_t t;
1938
1939 mutex_enter(&dd->dd_lock);
1940 t = dd->dd_snap_cmtime;
1941 mutex_exit(&dd->dd_lock);
1942
1943 return (t);
1944}
1945
1946void
1947dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1948{
1949 timestruc_t t;
1950
1951 gethrestime(&t);
1952 mutex_enter(&dd->dd_lock);
1953 dd->dd_snap_cmtime = t;
1954 mutex_exit(&dd->dd_lock);
1955}
1956
1957void
1958dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1959{
1960 objset_t *mos = dd->dd_pool->dp_meta_objset;
1961 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1962}
1963
1964boolean_t
1965dsl_dir_is_zapified(dsl_dir_t *dd)
1966{
1967 dmu_object_info_t doi;
1968
1969 dmu_object_info_from_db(dd->dd_dbuf, &doi);
1970 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
1971}
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) 2012, 2014 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 ZFS_SPACE_CHECK_RESERVED);
647
648 if (error == EALREADY)
649 error = 0;
650
651 return (error);
652}
653
654/*
655 * Used to determine if the filesystem_limit or snapshot_limit should be
656 * enforced. We allow the limit to be exceeded if the user has permission to
657 * write the property value. We pass in the creds that we got in the open
658 * context since we will always be the GZ root in syncing context. We also have
659 * to handle the case where we are allowed to change the limit on the current
660 * dataset, but there may be another limit in the tree above.
661 *
662 * We can never modify these two properties within a non-global zone. In
663 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
664 * can't use that function since we are already holding the dp_config_rwlock.
665 * In addition, we already have the dd and dealing with snapshots is simplified
666 * in this code.
667 */
668
669typedef enum {
670 ENFORCE_ALWAYS,
671 ENFORCE_NEVER,
672 ENFORCE_ABOVE
673} enforce_res_t;
674
675static enforce_res_t
676dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
677{
678 enforce_res_t enforce = ENFORCE_ALWAYS;
679 uint64_t obj;
680 dsl_dataset_t *ds;
681 uint64_t zoned;
682
683 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
684 prop == ZFS_PROP_SNAPSHOT_LIMIT);
685
686#ifdef _KERNEL
687#ifdef __FreeBSD__
688 if (jailed(cr))
689#else
690 if (crgetzoneid(cr) != GLOBAL_ZONEID)
691#endif
692 return (ENFORCE_ALWAYS);
693
694 if (secpolicy_zfs(cr) == 0)
695 return (ENFORCE_NEVER);
696#endif
697
698 if ((obj = dd->dd_phys->dd_head_dataset_obj) == 0)
699 return (ENFORCE_ALWAYS);
700
701 ASSERT(dsl_pool_config_held(dd->dd_pool));
702
703 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
704 return (ENFORCE_ALWAYS);
705
706 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
707 /* Only root can access zoned fs's from the GZ */
708 enforce = ENFORCE_ALWAYS;
709 } else {
710 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
711 enforce = ENFORCE_ABOVE;
712 }
713
714 dsl_dataset_rele(ds, FTAG);
715 return (enforce);
716}
717
718/*
719 * Check if adding additional child filesystem(s) would exceed any filesystem
720 * limits or adding additional snapshot(s) would exceed any snapshot limits.
721 * The prop argument indicates which limit to check.
722 *
723 * Note that all filesystem limits up to the root (or the highest
724 * initialized) filesystem or the given ancestor must be satisfied.
725 */
726int
727dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
728 dsl_dir_t *ancestor, cred_t *cr)
729{
730 objset_t *os = dd->dd_pool->dp_meta_objset;
731 uint64_t limit, count;
732 char *count_prop;
733 enforce_res_t enforce;
734 int err = 0;
735
736 ASSERT(dsl_pool_config_held(dd->dd_pool));
737 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
738 prop == ZFS_PROP_SNAPSHOT_LIMIT);
739
740 /*
741 * If we're allowed to change the limit, don't enforce the limit
742 * e.g. this can happen if a snapshot is taken by an administrative
743 * user in the global zone (i.e. a recursive snapshot by root).
744 * However, we must handle the case of delegated permissions where we
745 * are allowed to change the limit on the current dataset, but there
746 * is another limit in the tree above.
747 */
748 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
749 if (enforce == ENFORCE_NEVER)
750 return (0);
751
752 /*
753 * e.g. if renaming a dataset with no snapshots, count adjustment
754 * is 0.
755 */
756 if (delta == 0)
757 return (0);
758
759 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
760 /*
761 * We don't enforce the limit for temporary snapshots. This is
762 * indicated by a NULL cred_t argument.
763 */
764 if (cr == NULL)
765 return (0);
766
767 count_prop = DD_FIELD_SNAPSHOT_COUNT;
768 } else {
769 count_prop = DD_FIELD_FILESYSTEM_COUNT;
770 }
771
772 /*
773 * If an ancestor has been provided, stop checking the limit once we
774 * hit that dir. We need this during rename so that we don't overcount
775 * the check once we recurse up to the common ancestor.
776 */
777 if (ancestor == dd)
778 return (0);
779
780 /*
781 * If we hit an uninitialized node while recursing up the tree, we can
782 * stop since we know there is no limit here (or above). The counts are
783 * not valid on this node and we know we won't touch this node's counts.
784 */
785 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
786 count_prop, sizeof (count), 1, &count) == ENOENT)
787 return (0);
788
789 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
790 B_FALSE);
791 if (err != 0)
792 return (err);
793
794 /* Is there a limit which we've hit? */
795 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
796 return (SET_ERROR(EDQUOT));
797
798 if (dd->dd_parent != NULL)
799 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
800 ancestor, cr);
801
802 return (err);
803}
804
805/*
806 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
807 * parents. When a new filesystem/snapshot is created, increment the count on
808 * all parents, and when a filesystem/snapshot is destroyed, decrement the
809 * count.
810 */
811void
812dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
813 dmu_tx_t *tx)
814{
815 int err;
816 objset_t *os = dd->dd_pool->dp_meta_objset;
817 uint64_t count;
818
819 ASSERT(dsl_pool_config_held(dd->dd_pool));
820 ASSERT(dmu_tx_is_syncing(tx));
821 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
822 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
823
824 /*
825 * When we receive an incremental stream into a filesystem that already
826 * exists, a temporary clone is created. We don't count this temporary
827 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
828 * $MOS & $ORIGIN) objsets.
829 */
830 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
831 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
832 return;
833
834 /*
835 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
836 */
837 if (delta == 0)
838 return;
839
840 /*
841 * If we hit an uninitialized node while recursing up the tree, we can
842 * stop since we know the counts are not valid on this node and we
843 * know we shouldn't touch this node's counts. An uninitialized count
844 * on the node indicates that either the feature has not yet been
845 * activated or there are no limits on this part of the tree.
846 */
847 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
848 prop, sizeof (count), 1, &count)) == ENOENT)
849 return;
850 VERIFY0(err);
851
852 count += delta;
853 /* Use a signed verify to make sure we're not neg. */
854 VERIFY3S(count, >=, 0);
855
856 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
857 tx));
858
859 /* Roll up this additional count into our ancestors */
860 if (dd->dd_parent != NULL)
861 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
862}
863
864uint64_t
865dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
866 dmu_tx_t *tx)
867{
868 objset_t *mos = dp->dp_meta_objset;
869 uint64_t ddobj;
870 dsl_dir_phys_t *ddphys;
871 dmu_buf_t *dbuf;
872
873 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
874 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
875 if (pds) {
876 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
877 name, sizeof (uint64_t), 1, &ddobj, tx));
878 } else {
879 /* it's the root dir */
880 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
881 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
882 }
883 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
884 dmu_buf_will_dirty(dbuf, tx);
885 ddphys = dbuf->db_data;
886
887 ddphys->dd_creation_time = gethrestime_sec();
888 if (pds) {
889 ddphys->dd_parent_obj = pds->dd_object;
890
891 /* update the filesystem counts */
892 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
893 }
894 ddphys->dd_props_zapobj = zap_create(mos,
895 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
896 ddphys->dd_child_dir_zapobj = zap_create(mos,
897 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
898 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
899 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
900 dmu_buf_rele(dbuf, FTAG);
901
902 return (ddobj);
903}
904
905boolean_t
906dsl_dir_is_clone(dsl_dir_t *dd)
907{
908 return (dd->dd_phys->dd_origin_obj &&
909 (dd->dd_pool->dp_origin_snap == NULL ||
910 dd->dd_phys->dd_origin_obj !=
911 dd->dd_pool->dp_origin_snap->ds_object));
912}
913
914void
915dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
916{
917 mutex_enter(&dd->dd_lock);
918 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
919 dd->dd_phys->dd_used_bytes);
920 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
921 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
922 dd->dd_phys->dd_reserved);
923 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
924 dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
925 (dd->dd_phys->dd_uncompressed_bytes * 100 /
926 dd->dd_phys->dd_compressed_bytes));
927 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
928 dd->dd_phys->dd_uncompressed_bytes);
929 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
930 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
931 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
932 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
933 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
934 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
935 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
936 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
937 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
938 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
939 }
940 mutex_exit(&dd->dd_lock);
941
942 if (dsl_dir_is_zapified(dd)) {
943 uint64_t count;
944 objset_t *os = dd->dd_pool->dp_meta_objset;
945
946 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
947 sizeof (count), 1, &count) == 0) {
948 dsl_prop_nvlist_add_uint64(nv,
949 ZFS_PROP_FILESYSTEM_COUNT, count);
950 }
951 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
952 sizeof (count), 1, &count) == 0) {
953 dsl_prop_nvlist_add_uint64(nv,
954 ZFS_PROP_SNAPSHOT_COUNT, count);
955 }
956 }
957
958 if (dsl_dir_is_clone(dd)) {
959 dsl_dataset_t *ds;
960 char buf[MAXNAMELEN];
961
962 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
963 dd->dd_phys->dd_origin_obj, FTAG, &ds));
964 dsl_dataset_name(ds, buf);
965 dsl_dataset_rele(ds, FTAG);
966 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
967 }
968}
969
970void
971dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
972{
973 dsl_pool_t *dp = dd->dd_pool;
974
975 ASSERT(dd->dd_phys);
976
977 if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
978 /* up the hold count until we can be written out */
979 dmu_buf_add_ref(dd->dd_dbuf, dd);
980 }
981}
982
983static int64_t
984parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
985{
986 uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
987 uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
988 return (new_accounted - old_accounted);
989}
990
991void
992dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
993{
994 ASSERT(dmu_tx_is_syncing(tx));
995
996 mutex_enter(&dd->dd_lock);
997 ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
998 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
999 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
1000 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
1001 mutex_exit(&dd->dd_lock);
1002
1003 /* release the hold from dsl_dir_dirty */
1004 dmu_buf_rele(dd->dd_dbuf, dd);
1005}
1006
1007static uint64_t
1008dsl_dir_space_towrite(dsl_dir_t *dd)
1009{
1010 uint64_t space = 0;
1011 int i;
1012
1013 ASSERT(MUTEX_HELD(&dd->dd_lock));
1014
1015 for (i = 0; i < TXG_SIZE; i++) {
1016 space += dd->dd_space_towrite[i&TXG_MASK];
1017 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
1018 }
1019 return (space);
1020}
1021
1022/*
1023 * How much space would dd have available if ancestor had delta applied
1024 * to it? If ondiskonly is set, we're only interested in what's
1025 * on-disk, not estimated pending changes.
1026 */
1027uint64_t
1028dsl_dir_space_available(dsl_dir_t *dd,
1029 dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
1030{
1031 uint64_t parentspace, myspace, quota, used;
1032
1033 /*
1034 * If there are no restrictions otherwise, assume we have
1035 * unlimited space available.
1036 */
1037 quota = UINT64_MAX;
1038 parentspace = UINT64_MAX;
1039
1040 if (dd->dd_parent != NULL) {
1041 parentspace = dsl_dir_space_available(dd->dd_parent,
1042 ancestor, delta, ondiskonly);
1043 }
1044
1045 mutex_enter(&dd->dd_lock);
1046 if (dd->dd_phys->dd_quota != 0)
1047 quota = dd->dd_phys->dd_quota;
1048 used = dd->dd_phys->dd_used_bytes;
1049 if (!ondiskonly)
1050 used += dsl_dir_space_towrite(dd);
1051
1052 if (dd->dd_parent == NULL) {
1053 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
1054 quota = MIN(quota, poolsize);
1055 }
1056
1057 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
1058 /*
1059 * We have some space reserved, in addition to what our
1060 * parent gave us.
1061 */
1062 parentspace += dd->dd_phys->dd_reserved - used;
1063 }
1064
1065 if (dd == ancestor) {
1066 ASSERT(delta <= 0);
1067 ASSERT(used >= -delta);
1068 used += delta;
1069 if (parentspace != UINT64_MAX)
1070 parentspace -= delta;
1071 }
1072
1073 if (used > quota) {
1074 /* over quota */
1075 myspace = 0;
1076 } else {
1077 /*
1078 * the lesser of the space provided by our parent and
1079 * the space left in our quota
1080 */
1081 myspace = MIN(parentspace, quota - used);
1082 }
1083
1084 mutex_exit(&dd->dd_lock);
1085
1086 return (myspace);
1087}
1088
1089struct tempreserve {
1090 list_node_t tr_node;
1091 dsl_dir_t *tr_ds;
1092 uint64_t tr_size;
1093};
1094
1095static int
1096dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
1097 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
1098 dmu_tx_t *tx, boolean_t first)
1099{
1100 uint64_t txg = tx->tx_txg;
1101 uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
1102 uint64_t deferred = 0;
1103 struct tempreserve *tr;
1104 int retval = EDQUOT;
1105 int txgidx = txg & TXG_MASK;
1106 int i;
1107 uint64_t ref_rsrv = 0;
1108
1109 ASSERT3U(txg, !=, 0);
1110 ASSERT3S(asize, >, 0);
1111
1112 mutex_enter(&dd->dd_lock);
1113
1114 /*
1115 * Check against the dsl_dir's quota. We don't add in the delta
1116 * when checking for over-quota because they get one free hit.
1117 */
1118 est_inflight = dsl_dir_space_towrite(dd);
1119 for (i = 0; i < TXG_SIZE; i++)
1120 est_inflight += dd->dd_tempreserved[i];
1121 used_on_disk = dd->dd_phys->dd_used_bytes;
1122
1123 /*
1124 * On the first iteration, fetch the dataset's used-on-disk and
1125 * refreservation values. Also, if checkrefquota is set, test if
1126 * allocating this space would exceed the dataset's refquota.
1127 */
1128 if (first && tx->tx_objset) {
1129 int error;
1130 dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
1131
1132 error = dsl_dataset_check_quota(ds, checkrefquota,
1133 asize, est_inflight, &used_on_disk, &ref_rsrv);
1134 if (error) {
1135 mutex_exit(&dd->dd_lock);
1136 return (error);
1137 }
1138 }
1139
1140 /*
1141 * If this transaction will result in a net free of space,
1142 * we want to let it through.
1143 */
1144 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
1145 quota = UINT64_MAX;
1146 else
1147 quota = dd->dd_phys->dd_quota;
1148
1149 /*
1150 * Adjust the quota against the actual pool size at the root
1151 * minus any outstanding deferred frees.
1152 * To ensure that it's possible to remove files from a full
1153 * pool without inducing transient overcommits, we throttle
1154 * netfree transactions against a quota that is slightly larger,
1155 * but still within the pool's allocation slop. In cases where
1156 * we're very close to full, this will allow a steady trickle of
1157 * removes to get through.
1158 */
1159 if (dd->dd_parent == NULL) {
1160 spa_t *spa = dd->dd_pool->dp_spa;
1161 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
1162 deferred = metaslab_class_get_deferred(spa_normal_class(spa));
1163 if (poolsize - deferred < quota) {
1164 quota = poolsize - deferred;
1165 retval = ENOSPC;
1166 }
1167 }
1168
1169 /*
1170 * If they are requesting more space, and our current estimate
1171 * is over quota, they get to try again unless the actual
1172 * on-disk is over quota and there are no pending changes (which
1173 * may free up space for us).
1174 */
1175 if (used_on_disk + est_inflight >= quota) {
1176 if (est_inflight > 0 || used_on_disk < quota ||
1177 (retval == ENOSPC && used_on_disk < quota + deferred))
1178 retval = ERESTART;
1179 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
1180 "quota=%lluK tr=%lluK err=%d\n",
1181 used_on_disk>>10, est_inflight>>10,
1182 quota>>10, asize>>10, retval);
1183 mutex_exit(&dd->dd_lock);
1184 return (SET_ERROR(retval));
1185 }
1186
1187 /* We need to up our estimated delta before dropping dd_lock */
1188 dd->dd_tempreserved[txgidx] += asize;
1189
1190 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
1191 asize - ref_rsrv);
1192 mutex_exit(&dd->dd_lock);
1193
1194 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1195 tr->tr_ds = dd;
1196 tr->tr_size = asize;
1197 list_insert_tail(tr_list, tr);
1198
1199 /* see if it's OK with our parent */
1200 if (dd->dd_parent && parent_rsrv) {
1201 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);
1202
1203 return (dsl_dir_tempreserve_impl(dd->dd_parent,
1204 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
1205 } else {
1206 return (0);
1207 }
1208}
1209
1210/*
1211 * Reserve space in this dsl_dir, to be used in this tx's txg.
1212 * After the space has been dirtied (and dsl_dir_willuse_space()
1213 * has been called), the reservation should be canceled, using
1214 * dsl_dir_tempreserve_clear().
1215 */
1216int
1217dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
1218 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
1219{
1220 int err;
1221 list_t *tr_list;
1222
1223 if (asize == 0) {
1224 *tr_cookiep = NULL;
1225 return (0);
1226 }
1227
1228 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
1229 list_create(tr_list, sizeof (struct tempreserve),
1230 offsetof(struct tempreserve, tr_node));
1231 ASSERT3S(asize, >, 0);
1232 ASSERT3S(fsize, >=, 0);
1233
1234 err = arc_tempreserve_space(lsize, tx->tx_txg);
1235 if (err == 0) {
1236 struct tempreserve *tr;
1237
1238 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
1239 tr->tr_size = lsize;
1240 list_insert_tail(tr_list, tr);
1241 } else {
1242 if (err == EAGAIN) {
1243 /*
1244 * If arc_memory_throttle() detected that pageout
1245 * is running and we are low on memory, we delay new
1246 * non-pageout transactions to give pageout an
1247 * advantage.
1248 *
1249 * It is unfortunate to be delaying while the caller's
1250 * locks are held.
1251 */
1252 txg_delay(dd->dd_pool, tx->tx_txg,
1253 MSEC2NSEC(10), MSEC2NSEC(10));
1254 err = SET_ERROR(ERESTART);
1255 }
1256 }
1257
1258 if (err == 0) {
1259 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
1260 FALSE, asize > usize, tr_list, tx, TRUE);
1261 }
1262
1263 if (err != 0)
1264 dsl_dir_tempreserve_clear(tr_list, tx);
1265 else
1266 *tr_cookiep = tr_list;
1267
1268 return (err);
1269}
1270
1271/*
1272 * Clear a temporary reservation that we previously made with
1273 * dsl_dir_tempreserve_space().
1274 */
1275void
1276dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
1277{
1278 int txgidx = tx->tx_txg & TXG_MASK;
1279 list_t *tr_list = tr_cookie;
1280 struct tempreserve *tr;
1281
1282 ASSERT3U(tx->tx_txg, !=, 0);
1283
1284 if (tr_cookie == NULL)
1285 return;
1286
1287 while ((tr = list_head(tr_list)) != NULL) {
1288 if (tr->tr_ds) {
1289 mutex_enter(&tr->tr_ds->dd_lock);
1290 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
1291 tr->tr_size);
1292 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
1293 mutex_exit(&tr->tr_ds->dd_lock);
1294 } else {
1295 arc_tempreserve_clear(tr->tr_size);
1296 }
1297 list_remove(tr_list, tr);
1298 kmem_free(tr, sizeof (struct tempreserve));
1299 }
1300
1301 kmem_free(tr_list, sizeof (list_t));
1302}
1303
1304/*
1305 * This should be called from open context when we think we're going to write
1306 * or free space, for example when dirtying data. Be conservative; it's okay
1307 * to write less space or free more, but we don't want to write more or free
1308 * less than the amount specified.
1309 */
1310void
1311dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
1312{
1313 int64_t parent_space;
1314 uint64_t est_used;
1315
1316 mutex_enter(&dd->dd_lock);
1317 if (space > 0)
1318 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
1319
1320 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
1321 parent_space = parent_delta(dd, est_used, space);
1322 mutex_exit(&dd->dd_lock);
1323
1324 /* Make sure that we clean up dd_space_to* */
1325 dsl_dir_dirty(dd, tx);
1326
1327 /* XXX this is potentially expensive and unnecessary... */
1328 if (parent_space && dd->dd_parent)
1329 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx);
1330}
1331
1332/* call from syncing context when we actually write/free space for this dd */
1333void
1334dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
1335 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
1336{
1337 int64_t accounted_delta;
1338
1339 /*
1340 * dsl_dataset_set_refreservation_sync_impl() calls this with
1341 * dd_lock held, so that it can atomically update
1342 * ds->ds_reserved and the dsl_dir accounting, so that
1343 * dsl_dataset_check_quota() can see dataset and dir accounting
1344 * consistently.
1345 */
1346 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
1347
1348 ASSERT(dmu_tx_is_syncing(tx));
1349 ASSERT(type < DD_USED_NUM);
1350
1351 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1352
1353 if (needlock)
1354 mutex_enter(&dd->dd_lock);
1355 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
1356 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
1357 ASSERT(compressed >= 0 ||
1358 dd->dd_phys->dd_compressed_bytes >= -compressed);
1359 ASSERT(uncompressed >= 0 ||
1360 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
1361 dd->dd_phys->dd_used_bytes += used;
1362 dd->dd_phys->dd_uncompressed_bytes += uncompressed;
1363 dd->dd_phys->dd_compressed_bytes += compressed;
1364
1365 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1366 ASSERT(used > 0 ||
1367 dd->dd_phys->dd_used_breakdown[type] >= -used);
1368 dd->dd_phys->dd_used_breakdown[type] += used;
1369#ifdef DEBUG
1370 dd_used_t t;
1371 uint64_t u = 0;
1372 for (t = 0; t < DD_USED_NUM; t++)
1373 u += dd->dd_phys->dd_used_breakdown[t];
1374 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
1375#endif
1376 }
1377 if (needlock)
1378 mutex_exit(&dd->dd_lock);
1379
1380 if (dd->dd_parent != NULL) {
1381 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1382 accounted_delta, compressed, uncompressed, tx);
1383 dsl_dir_transfer_space(dd->dd_parent,
1384 used - accounted_delta,
1385 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
1386 }
1387}
1388
1389void
1390dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
1391 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
1392{
1393 ASSERT(dmu_tx_is_syncing(tx));
1394 ASSERT(oldtype < DD_USED_NUM);
1395 ASSERT(newtype < DD_USED_NUM);
1396
1397 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
1398 return;
1399
1400 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1401 mutex_enter(&dd->dd_lock);
1402 ASSERT(delta > 0 ?
1403 dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
1404 dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
1405 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
1406 dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
1407 dd->dd_phys->dd_used_breakdown[newtype] += delta;
1408 mutex_exit(&dd->dd_lock);
1409}
1410
1411typedef struct dsl_dir_set_qr_arg {
1412 const char *ddsqra_name;
1413 zprop_source_t ddsqra_source;
1414 uint64_t ddsqra_value;
1415} dsl_dir_set_qr_arg_t;
1416
1417static int
1418dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
1419{
1420 dsl_dir_set_qr_arg_t *ddsqra = arg;
1421 dsl_pool_t *dp = dmu_tx_pool(tx);
1422 dsl_dataset_t *ds;
1423 int error;
1424 uint64_t towrite, newval;
1425
1426 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1427 if (error != 0)
1428 return (error);
1429
1430 error = dsl_prop_predict(ds->ds_dir, "quota",
1431 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1432 if (error != 0) {
1433 dsl_dataset_rele(ds, FTAG);
1434 return (error);
1435 }
1436
1437 if (newval == 0) {
1438 dsl_dataset_rele(ds, FTAG);
1439 return (0);
1440 }
1441
1442 mutex_enter(&ds->ds_dir->dd_lock);
1443 /*
1444 * If we are doing the preliminary check in open context, and
1445 * there are pending changes, then don't fail it, since the
1446 * pending changes could under-estimate the amount of space to be
1447 * freed up.
1448 */
1449 towrite = dsl_dir_space_towrite(ds->ds_dir);
1450 if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
1451 (newval < ds->ds_dir->dd_phys->dd_reserved ||
1452 newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
1453 error = SET_ERROR(ENOSPC);
1454 }
1455 mutex_exit(&ds->ds_dir->dd_lock);
1456 dsl_dataset_rele(ds, FTAG);
1457 return (error);
1458}
1459
1460static void
1461dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
1462{
1463 dsl_dir_set_qr_arg_t *ddsqra = arg;
1464 dsl_pool_t *dp = dmu_tx_pool(tx);
1465 dsl_dataset_t *ds;
1466 uint64_t newval;
1467
1468 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1469
1470 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1471 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
1472 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1473 &ddsqra->ddsqra_value, tx);
1474
1475 VERIFY0(dsl_prop_get_int_ds(ds,
1476 zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
1477 } else {
1478 newval = ddsqra->ddsqra_value;
1479 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1480 zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
1481 }
1482
1483 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1484 mutex_enter(&ds->ds_dir->dd_lock);
1485 ds->ds_dir->dd_phys->dd_quota = newval;
1486 mutex_exit(&ds->ds_dir->dd_lock);
1487 dsl_dataset_rele(ds, FTAG);
1488}
1489
1490int
1491dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
1492{
1493 dsl_dir_set_qr_arg_t ddsqra;
1494
1495 ddsqra.ddsqra_name = ddname;
1496 ddsqra.ddsqra_source = source;
1497 ddsqra.ddsqra_value = quota;
1498
1499 return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
1500 dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1501}
1502
1503int
1504dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
1505{
1506 dsl_dir_set_qr_arg_t *ddsqra = arg;
1507 dsl_pool_t *dp = dmu_tx_pool(tx);
1508 dsl_dataset_t *ds;
1509 dsl_dir_t *dd;
1510 uint64_t newval, used, avail;
1511 int error;
1512
1513 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
1514 if (error != 0)
1515 return (error);
1516 dd = ds->ds_dir;
1517
1518 /*
1519 * If we are doing the preliminary check in open context, the
1520 * space estimates may be inaccurate.
1521 */
1522 if (!dmu_tx_is_syncing(tx)) {
1523 dsl_dataset_rele(ds, FTAG);
1524 return (0);
1525 }
1526
1527 error = dsl_prop_predict(ds->ds_dir,
1528 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1529 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
1530 if (error != 0) {
1531 dsl_dataset_rele(ds, FTAG);
1532 return (error);
1533 }
1534
1535 mutex_enter(&dd->dd_lock);
1536 used = dd->dd_phys->dd_used_bytes;
1537 mutex_exit(&dd->dd_lock);
1538
1539 if (dd->dd_parent) {
1540 avail = dsl_dir_space_available(dd->dd_parent,
1541 NULL, 0, FALSE);
1542 } else {
1543 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
1544 }
1545
1546 if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
1547 uint64_t delta = MAX(used, newval) -
1548 MAX(used, dd->dd_phys->dd_reserved);
1549
1550 if (delta > avail ||
1551 (dd->dd_phys->dd_quota > 0 &&
1552 newval > dd->dd_phys->dd_quota))
1553 error = SET_ERROR(ENOSPC);
1554 }
1555
1556 dsl_dataset_rele(ds, FTAG);
1557 return (error);
1558}
1559
1560void
1561dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
1562{
1563 uint64_t used;
1564 int64_t delta;
1565
1566 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1567
1568 mutex_enter(&dd->dd_lock);
1569 used = dd->dd_phys->dd_used_bytes;
1570 delta = MAX(used, value) - MAX(used, dd->dd_phys->dd_reserved);
1571 dd->dd_phys->dd_reserved = value;
1572
1573 if (dd->dd_parent != NULL) {
1574 /* Roll up this additional usage into our ancestors */
1575 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1576 delta, 0, 0, tx);
1577 }
1578 mutex_exit(&dd->dd_lock);
1579}
1580
1581static void
1582dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
1583{
1584 dsl_dir_set_qr_arg_t *ddsqra = arg;
1585 dsl_pool_t *dp = dmu_tx_pool(tx);
1586 dsl_dataset_t *ds;
1587 uint64_t newval;
1588
1589 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
1590
1591 if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
1592 dsl_prop_set_sync_impl(ds,
1593 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1594 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
1595 &ddsqra->ddsqra_value, tx);
1596
1597 VERIFY0(dsl_prop_get_int_ds(ds,
1598 zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
1599 } else {
1600 newval = ddsqra->ddsqra_value;
1601 spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
1602 zfs_prop_to_name(ZFS_PROP_RESERVATION),
1603 (longlong_t)newval);
1604 }
1605
1606 dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
1607 dsl_dataset_rele(ds, FTAG);
1608}
1609
1610int
1611dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
1612 uint64_t reservation)
1613{
1614 dsl_dir_set_qr_arg_t ddsqra;
1615
1616 ddsqra.ddsqra_name = ddname;
1617 ddsqra.ddsqra_source = source;
1618 ddsqra.ddsqra_value = reservation;
1619
1620 return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
1621 dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
1622}
1623
1624static dsl_dir_t *
1625closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
1626{
1627 for (; ds1; ds1 = ds1->dd_parent) {
1628 dsl_dir_t *dd;
1629 for (dd = ds2; dd; dd = dd->dd_parent) {
1630 if (ds1 == dd)
1631 return (dd);
1632 }
1633 }
1634 return (NULL);
1635}
1636
1637/*
1638 * If delta is applied to dd, how much of that delta would be applied to
1639 * ancestor? Syncing context only.
1640 */
1641static int64_t
1642would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
1643{
1644 if (dd == ancestor)
1645 return (delta);
1646
1647 mutex_enter(&dd->dd_lock);
1648 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1649 mutex_exit(&dd->dd_lock);
1650 return (would_change(dd->dd_parent, delta, ancestor));
1651}
1652
1653typedef struct dsl_dir_rename_arg {
1654 const char *ddra_oldname;
1655 const char *ddra_newname;
1656 cred_t *ddra_cred;
1657} dsl_dir_rename_arg_t;
1658
1659/* ARGSUSED */
1660static int
1661dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1662{
1663 int *deltap = arg;
1664 char namebuf[MAXNAMELEN];
1665
1666 dsl_dataset_name(ds, namebuf);
1667
1668 if (strlen(namebuf) + *deltap >= MAXNAMELEN)
1669 return (SET_ERROR(ENAMETOOLONG));
1670 return (0);
1671}
1672
1673static int
1674dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
1675{
1676 dsl_dir_rename_arg_t *ddra = arg;
1677 dsl_pool_t *dp = dmu_tx_pool(tx);
1678 dsl_dir_t *dd, *newparent;
1679 const char *mynewname;
1680 int error;
1681 int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
1682
1683 /* target dir should exist */
1684 error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
1685 if (error != 0)
1686 return (error);
1687
1688 /* new parent should exist */
1689 error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
1690 &newparent, &mynewname);
1691 if (error != 0) {
1692 dsl_dir_rele(dd, FTAG);
1693 return (error);
1694 }
1695
1696 /* can't rename to different pool */
1697 if (dd->dd_pool != newparent->dd_pool) {
1698 dsl_dir_rele(newparent, FTAG);
1699 dsl_dir_rele(dd, FTAG);
1700 return (SET_ERROR(ENXIO));
1701 }
1702
1703 /* new name should not already exist */
1704 if (mynewname == NULL) {
1705 dsl_dir_rele(newparent, FTAG);
1706 dsl_dir_rele(dd, FTAG);
1707 return (SET_ERROR(EEXIST));
1708 }
1709
1710 /* if the name length is growing, validate child name lengths */
1711 if (delta > 0) {
1712 error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
1713 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1714 if (error != 0) {
1715 dsl_dir_rele(newparent, FTAG);
1716 dsl_dir_rele(dd, FTAG);
1717 return (error);
1718 }
1719 }
1720
1721 if (dmu_tx_is_syncing(tx)) {
1722 if (spa_feature_is_active(dp->dp_spa,
1723 SPA_FEATURE_FS_SS_LIMIT)) {
1724 /*
1725 * Although this is the check function and we don't
1726 * normally make on-disk changes in check functions,
1727 * we need to do that here.
1728 *
1729 * Ensure this portion of the tree's counts have been
1730 * initialized in case the new parent has limits set.
1731 */
1732 dsl_dir_init_fs_ss_count(dd, tx);
1733 }
1734 }
1735
1736 if (newparent != dd->dd_parent) {
1737 /* is there enough space? */
1738 uint64_t myspace =
1739 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1740 objset_t *os = dd->dd_pool->dp_meta_objset;
1741 uint64_t fs_cnt = 0;
1742 uint64_t ss_cnt = 0;
1743
1744 if (dsl_dir_is_zapified(dd)) {
1745 int err;
1746
1747 err = zap_lookup(os, dd->dd_object,
1748 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1749 &fs_cnt);
1750 if (err != ENOENT && err != 0) {
1751 dsl_dir_rele(newparent, FTAG);
1752 dsl_dir_rele(dd, FTAG);
1753 return (err);
1754 }
1755
1756 /*
1757 * have to add 1 for the filesystem itself that we're
1758 * moving
1759 */
1760 fs_cnt++;
1761
1762 err = zap_lookup(os, dd->dd_object,
1763 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1764 &ss_cnt);
1765 if (err != ENOENT && err != 0) {
1766 dsl_dir_rele(newparent, FTAG);
1767 dsl_dir_rele(dd, FTAG);
1768 return (err);
1769 }
1770 }
1771
1772 /* no rename into our descendant */
1773 if (closest_common_ancestor(dd, newparent) == dd) {
1774 dsl_dir_rele(newparent, FTAG);
1775 dsl_dir_rele(dd, FTAG);
1776 return (SET_ERROR(EINVAL));
1777 }
1778
1779 error = dsl_dir_transfer_possible(dd->dd_parent,
1780 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1781 if (error != 0) {
1782 dsl_dir_rele(newparent, FTAG);
1783 dsl_dir_rele(dd, FTAG);
1784 return (error);
1785 }
1786 }
1787
1788 dsl_dir_rele(newparent, FTAG);
1789 dsl_dir_rele(dd, FTAG);
1790 return (0);
1791}
1792
1793static void
1794dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
1795{
1796 dsl_dir_rename_arg_t *ddra = arg;
1797 dsl_pool_t *dp = dmu_tx_pool(tx);
1798 dsl_dir_t *dd, *newparent;
1799 const char *mynewname;
1800 int error;
1801 objset_t *mos = dp->dp_meta_objset;
1802
1803 VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
1804 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1805 &mynewname));
1806
1807 /* Log this before we change the name. */
1808 spa_history_log_internal_dd(dd, "rename", tx,
1809 "-> %s", ddra->ddra_newname);
1810
1811 if (newparent != dd->dd_parent) {
1812 objset_t *os = dd->dd_pool->dp_meta_objset;
1813 uint64_t fs_cnt = 0;
1814 uint64_t ss_cnt = 0;
1815
1816 /*
1817 * We already made sure the dd counts were initialized in the
1818 * check function.
1819 */
1820 if (spa_feature_is_active(dp->dp_spa,
1821 SPA_FEATURE_FS_SS_LIMIT)) {
1822 VERIFY0(zap_lookup(os, dd->dd_object,
1823 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1824 &fs_cnt));
1825 /* add 1 for the filesystem itself that we're moving */
1826 fs_cnt++;
1827
1828 VERIFY0(zap_lookup(os, dd->dd_object,
1829 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1830 &ss_cnt));
1831 }
1832
1833 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1834 DD_FIELD_FILESYSTEM_COUNT, tx);
1835 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1836 DD_FIELD_FILESYSTEM_COUNT, tx);
1837
1838 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1839 DD_FIELD_SNAPSHOT_COUNT, tx);
1840 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1841 DD_FIELD_SNAPSHOT_COUNT, tx);
1842
1843 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1844 -dd->dd_phys->dd_used_bytes,
1845 -dd->dd_phys->dd_compressed_bytes,
1846 -dd->dd_phys->dd_uncompressed_bytes, tx);
1847 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1848 dd->dd_phys->dd_used_bytes,
1849 dd->dd_phys->dd_compressed_bytes,
1850 dd->dd_phys->dd_uncompressed_bytes, tx);
1851
1852 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
1853 uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
1854 dd->dd_phys->dd_used_bytes;
1855
1856 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
1857 -unused_rsrv, 0, 0, tx);
1858 dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
1859 unused_rsrv, 0, 0, tx);
1860 }
1861 }
1862
1863 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1864
1865 /* remove from old parent zapobj */
1866 error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
1867 dd->dd_myname, tx);
1868 ASSERT0(error);
1869
1870 (void) strcpy(dd->dd_myname, mynewname);
1871 dsl_dir_rele(dd->dd_parent, dd);
1872 dd->dd_phys->dd_parent_obj = newparent->dd_object;
1873 VERIFY0(dsl_dir_hold_obj(dp,
1874 newparent->dd_object, NULL, dd, &dd->dd_parent));
1875
1876 /* add to new parent zapobj */
1877 VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
1878 dd->dd_myname, 8, 1, &dd->dd_object, tx));
1879
1880#ifdef __FreeBSD__
1881#ifdef _KERNEL
1882 zfsvfs_update_fromname(ddra->ddra_oldname, ddra->ddra_newname);
1883 zvol_rename_minors(ddra->ddra_oldname, ddra->ddra_newname);
1884#endif
1885#endif
1886
1887 dsl_prop_notify_all(dd);
1888
1889 dsl_dir_rele(newparent, FTAG);
1890 dsl_dir_rele(dd, FTAG);
1891}
1892
1893int
1894dsl_dir_rename(const char *oldname, const char *newname)
1895{
1896 dsl_dir_rename_arg_t ddra;
1897
1898 ddra.ddra_oldname = oldname;
1899 ddra.ddra_newname = newname;
1900 ddra.ddra_cred = CRED();
1901
1902 return (dsl_sync_task(oldname,
1903 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
1904 3, ZFS_SPACE_CHECK_RESERVED));
1905}
1906
1907int
1908dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1909 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1910{
1911 dsl_dir_t *ancestor;
1912 int64_t adelta;
1913 uint64_t avail;
1914 int err;
1915
1916 ancestor = closest_common_ancestor(sdd, tdd);
1917 adelta = would_change(sdd, -space, ancestor);
1918 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1919 if (avail < space)
1920 return (SET_ERROR(ENOSPC));
1921
1922 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1923 ancestor, cr);
1924 if (err != 0)
1925 return (err);
1926 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1927 ancestor, cr);
1928 if (err != 0)
1929 return (err);
1930
1931 return (0);
1932}
1933
1934timestruc_t
1935dsl_dir_snap_cmtime(dsl_dir_t *dd)
1936{
1937 timestruc_t t;
1938
1939 mutex_enter(&dd->dd_lock);
1940 t = dd->dd_snap_cmtime;
1941 mutex_exit(&dd->dd_lock);
1942
1943 return (t);
1944}
1945
1946void
1947dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
1948{
1949 timestruc_t t;
1950
1951 gethrestime(&t);
1952 mutex_enter(&dd->dd_lock);
1953 dd->dd_snap_cmtime = t;
1954 mutex_exit(&dd->dd_lock);
1955}
1956
1957void
1958dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1959{
1960 objset_t *mos = dd->dd_pool->dp_meta_objset;
1961 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1962}
1963
1964boolean_t
1965dsl_dir_is_zapified(dsl_dir_t *dd)
1966{
1967 dmu_object_info_t doi;
1968
1969 dmu_object_info_from_db(dd->dd_dbuf, &doi);
1970 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
1971}