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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Martin Matuska. All rights reserved.
28 #include <sys/dmu_objset.h>
29 #include <sys/dmu_tx.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dsl_deleg.h>
36 #include <sys/metaslab.h>
40 #include <sys/sunddi.h>
42 #include "zfs_namecheck.h"
44 static uint64_t dsl_dir_space_towrite(dsl_dir_t
*dd
);
48 dsl_dir_evict(dmu_buf_t
*db
, void *arg
)
52 ASSERTV(dsl_pool_t
*dp
= dd
->dd_pool
);
54 for (t
= 0; t
< TXG_SIZE
; t
++) {
55 ASSERT(!txg_list_member(&dp
->dp_dirty_dirs
, dd
, t
));
56 ASSERT(dd
->dd_tempreserved
[t
] == 0);
57 ASSERT(dd
->dd_space_towrite
[t
] == 0);
61 dsl_dir_rele(dd
->dd_parent
, dd
);
63 spa_close(dd
->dd_pool
->dp_spa
, dd
);
66 * The props callback list should have been cleaned up by
69 list_destroy(&dd
->dd_prop_cbs
);
70 mutex_destroy(&dd
->dd_lock
);
71 kmem_free(dd
, sizeof (dsl_dir_t
));
75 dsl_dir_hold_obj(dsl_pool_t
*dp
, uint64_t ddobj
,
76 const char *tail
, void *tag
, dsl_dir_t
**ddp
)
82 ASSERT(dsl_pool_config_held(dp
));
84 err
= dmu_bonus_hold(dp
->dp_meta_objset
, ddobj
, tag
, &dbuf
);
87 dd
= dmu_buf_get_user(dbuf
);
90 dmu_object_info_t doi
;
91 dmu_object_info_from_db(dbuf
, &doi
);
92 ASSERT3U(doi
.doi_type
, ==, DMU_OT_DSL_DIR
);
93 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (dsl_dir_phys_t
));
99 dd
= kmem_zalloc(sizeof (dsl_dir_t
), KM_PUSHPAGE
);
100 dd
->dd_object
= ddobj
;
103 dd
->dd_phys
= dbuf
->db_data
;
104 mutex_init(&dd
->dd_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
106 list_create(&dd
->dd_prop_cbs
, sizeof (dsl_prop_cb_record_t
),
107 offsetof(dsl_prop_cb_record_t
, cbr_node
));
109 dsl_dir_snap_cmtime_update(dd
);
111 if (dd
->dd_phys
->dd_parent_obj
) {
112 err
= dsl_dir_hold_obj(dp
, dd
->dd_phys
->dd_parent_obj
,
113 NULL
, dd
, &dd
->dd_parent
);
120 err
= zap_lookup(dp
->dp_meta_objset
,
121 dd
->dd_parent
->dd_phys
->dd_child_dir_zapobj
,
122 tail
, sizeof (foundobj
), 1, &foundobj
);
123 ASSERT(err
|| foundobj
== ddobj
);
125 (void) strcpy(dd
->dd_myname
, tail
);
127 err
= zap_value_search(dp
->dp_meta_objset
,
128 dd
->dd_parent
->dd_phys
->dd_child_dir_zapobj
,
129 ddobj
, 0, dd
->dd_myname
);
134 (void) strcpy(dd
->dd_myname
, spa_name(dp
->dp_spa
));
137 if (dsl_dir_is_clone(dd
)) {
138 dmu_buf_t
*origin_bonus
;
139 dsl_dataset_phys_t
*origin_phys
;
142 * We can't open the origin dataset, because
143 * that would require opening this dsl_dir.
144 * Just look at its phys directly instead.
146 err
= dmu_bonus_hold(dp
->dp_meta_objset
,
147 dd
->dd_phys
->dd_origin_obj
, FTAG
, &origin_bonus
);
150 origin_phys
= origin_bonus
->db_data
;
152 origin_phys
->ds_creation_txg
;
153 dmu_buf_rele(origin_bonus
, FTAG
);
156 winner
= dmu_buf_set_user_ie(dbuf
, dd
, &dd
->dd_phys
,
160 dsl_dir_rele(dd
->dd_parent
, dd
);
161 mutex_destroy(&dd
->dd_lock
);
162 kmem_free(dd
, sizeof (dsl_dir_t
));
165 spa_open_ref(dp
->dp_spa
, dd
);
170 * The dsl_dir_t has both open-to-close and instantiate-to-evict
171 * holds on the spa. We need the open-to-close holds because
172 * otherwise the spa_refcnt wouldn't change when we open a
173 * dir which the spa also has open, so we could incorrectly
174 * think it was OK to unload/export/destroy the pool. We need
175 * the instantiate-to-evict hold because the dsl_dir_t has a
176 * pointer to the dd_pool, which has a pointer to the spa_t.
178 spa_open_ref(dp
->dp_spa
, tag
);
179 ASSERT3P(dd
->dd_pool
, ==, dp
);
180 ASSERT3U(dd
->dd_object
, ==, ddobj
);
181 ASSERT3P(dd
->dd_dbuf
, ==, dbuf
);
187 dsl_dir_rele(dd
->dd_parent
, dd
);
188 mutex_destroy(&dd
->dd_lock
);
189 kmem_free(dd
, sizeof (dsl_dir_t
));
190 dmu_buf_rele(dbuf
, tag
);
195 dsl_dir_rele(dsl_dir_t
*dd
, void *tag
)
197 dprintf_dd(dd
, "%s\n", "");
198 spa_close(dd
->dd_pool
->dp_spa
, tag
);
199 dmu_buf_rele(dd
->dd_dbuf
, tag
);
202 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
204 dsl_dir_name(dsl_dir_t
*dd
, char *buf
)
207 dsl_dir_name(dd
->dd_parent
, buf
);
208 (void) strcat(buf
, "/");
212 if (!MUTEX_HELD(&dd
->dd_lock
)) {
214 * recursive mutex so that we can use
215 * dprintf_dd() with dd_lock held
217 mutex_enter(&dd
->dd_lock
);
218 (void) strcat(buf
, dd
->dd_myname
);
219 mutex_exit(&dd
->dd_lock
);
221 (void) strcat(buf
, dd
->dd_myname
);
225 /* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
227 dsl_dir_namelen(dsl_dir_t
*dd
)
232 /* parent's name + 1 for the "/" */
233 result
= dsl_dir_namelen(dd
->dd_parent
) + 1;
236 if (!MUTEX_HELD(&dd
->dd_lock
)) {
237 /* see dsl_dir_name */
238 mutex_enter(&dd
->dd_lock
);
239 result
+= strlen(dd
->dd_myname
);
240 mutex_exit(&dd
->dd_lock
);
242 result
+= strlen(dd
->dd_myname
);
249 getcomponent(const char *path
, char *component
, const char **nextp
)
253 if ((path
== NULL
) || (path
[0] == '\0'))
254 return (SET_ERROR(ENOENT
));
255 /* This would be a good place to reserve some namespace... */
256 p
= strpbrk(path
, "/@");
257 if (p
&& (p
[1] == '/' || p
[1] == '@')) {
258 /* two separators in a row */
259 return (SET_ERROR(EINVAL
));
261 if (p
== NULL
|| p
== path
) {
263 * if the first thing is an @ or /, it had better be an
264 * @ and it had better not have any more ats or slashes,
265 * and it had better have something after the @.
268 (p
[0] != '@' || strpbrk(path
+1, "/@") || p
[1] == '\0'))
269 return (SET_ERROR(EINVAL
));
270 if (strlen(path
) >= MAXNAMELEN
)
271 return (SET_ERROR(ENAMETOOLONG
));
272 (void) strcpy(component
, path
);
274 } else if (p
[0] == '/') {
275 if (p
- path
>= MAXNAMELEN
)
276 return (SET_ERROR(ENAMETOOLONG
));
277 (void) strncpy(component
, path
, p
- path
);
278 component
[p
- path
] = '\0';
280 } else if (p
[0] == '@') {
282 * if the next separator is an @, there better not be
285 if (strchr(path
, '/'))
286 return (SET_ERROR(EINVAL
));
287 if (p
- path
>= MAXNAMELEN
)
288 return (SET_ERROR(ENAMETOOLONG
));
289 (void) strncpy(component
, path
, p
- path
);
290 component
[p
- path
] = '\0';
292 panic("invalid p=%p", (void *)p
);
299 * Return the dsl_dir_t, and possibly the last component which couldn't
300 * be found in *tail. The name must be in the specified dsl_pool_t. This
301 * thread must hold the dp_config_rwlock for the pool. Returns NULL if the
302 * path is bogus, or if tail==NULL and we couldn't parse the whole name.
303 * (*tail)[0] == '@' means that the last component is a snapshot.
306 dsl_dir_hold(dsl_pool_t
*dp
, const char *name
, void *tag
,
307 dsl_dir_t
**ddp
, const char **tailp
)
310 const char *spaname
, *next
, *nextnext
= NULL
;
315 buf
= kmem_alloc(MAXNAMELEN
, KM_PUSHPAGE
);
316 err
= getcomponent(name
, buf
, &next
);
320 /* Make sure the name is in the specified pool. */
321 spaname
= spa_name(dp
->dp_spa
);
322 if (strcmp(buf
, spaname
) != 0) {
323 err
= SET_ERROR(EINVAL
);
327 ASSERT(dsl_pool_config_held(dp
));
329 err
= dsl_dir_hold_obj(dp
, dp
->dp_root_dir_obj
, NULL
, tag
, &dd
);
334 while (next
!= NULL
) {
336 err
= getcomponent(next
, buf
, &nextnext
);
339 ASSERT(next
[0] != '\0');
342 dprintf("looking up %s in obj%lld\n",
343 buf
, dd
->dd_phys
->dd_child_dir_zapobj
);
345 err
= zap_lookup(dp
->dp_meta_objset
,
346 dd
->dd_phys
->dd_child_dir_zapobj
,
347 buf
, sizeof (ddobj
), 1, &ddobj
);
354 err
= dsl_dir_hold_obj(dp
, ddobj
, buf
, tag
, &child_ds
);
357 dsl_dir_rele(dd
, tag
);
363 dsl_dir_rele(dd
, tag
);
368 * It's an error if there's more than one component left, or
369 * tailp==NULL and there's any component left.
372 (tailp
== NULL
|| (nextnext
&& nextnext
[0] != '\0'))) {
374 dsl_dir_rele(dd
, tag
);
375 dprintf("next=%p (%s) tail=%p\n", next
, next
?next
:"", tailp
);
376 err
= SET_ERROR(ENOENT
);
382 kmem_free(buf
, MAXNAMELEN
);
387 dsl_dir_create_sync(dsl_pool_t
*dp
, dsl_dir_t
*pds
, const char *name
,
390 objset_t
*mos
= dp
->dp_meta_objset
;
392 dsl_dir_phys_t
*ddphys
;
395 ddobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DIR
, 0,
396 DMU_OT_DSL_DIR
, sizeof (dsl_dir_phys_t
), tx
);
398 VERIFY(0 == zap_add(mos
, pds
->dd_phys
->dd_child_dir_zapobj
,
399 name
, sizeof (uint64_t), 1, &ddobj
, tx
));
401 /* it's the root dir */
402 VERIFY(0 == zap_add(mos
, DMU_POOL_DIRECTORY_OBJECT
,
403 DMU_POOL_ROOT_DATASET
, sizeof (uint64_t), 1, &ddobj
, tx
));
405 VERIFY(0 == dmu_bonus_hold(mos
, ddobj
, FTAG
, &dbuf
));
406 dmu_buf_will_dirty(dbuf
, tx
);
407 ddphys
= dbuf
->db_data
;
409 ddphys
->dd_creation_time
= gethrestime_sec();
411 ddphys
->dd_parent_obj
= pds
->dd_object
;
412 ddphys
->dd_props_zapobj
= zap_create(mos
,
413 DMU_OT_DSL_PROPS
, DMU_OT_NONE
, 0, tx
);
414 ddphys
->dd_child_dir_zapobj
= zap_create(mos
,
415 DMU_OT_DSL_DIR_CHILD_MAP
, DMU_OT_NONE
, 0, tx
);
416 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_USED_BREAKDOWN
)
417 ddphys
->dd_flags
|= DD_FLAG_USED_BREAKDOWN
;
418 dmu_buf_rele(dbuf
, FTAG
);
424 dsl_dir_is_clone(dsl_dir_t
*dd
)
426 return (dd
->dd_phys
->dd_origin_obj
&&
427 (dd
->dd_pool
->dp_origin_snap
== NULL
||
428 dd
->dd_phys
->dd_origin_obj
!=
429 dd
->dd_pool
->dp_origin_snap
->ds_object
));
433 dsl_dir_stats(dsl_dir_t
*dd
, nvlist_t
*nv
)
435 mutex_enter(&dd
->dd_lock
);
436 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
437 dd
->dd_phys
->dd_used_bytes
);
438 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_QUOTA
, dd
->dd_phys
->dd_quota
);
439 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_RESERVATION
,
440 dd
->dd_phys
->dd_reserved
);
441 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
,
442 dd
->dd_phys
->dd_compressed_bytes
== 0 ? 100 :
443 (dd
->dd_phys
->dd_uncompressed_bytes
* 100 /
444 dd
->dd_phys
->dd_compressed_bytes
));
445 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALUSED
,
446 dd
->dd_phys
->dd_uncompressed_bytes
);
447 if (dd
->dd_phys
->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
448 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDSNAP
,
449 dd
->dd_phys
->dd_used_breakdown
[DD_USED_SNAP
]);
450 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDDS
,
451 dd
->dd_phys
->dd_used_breakdown
[DD_USED_HEAD
]);
452 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDREFRESERV
,
453 dd
->dd_phys
->dd_used_breakdown
[DD_USED_REFRSRV
]);
454 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USEDCHILD
,
455 dd
->dd_phys
->dd_used_breakdown
[DD_USED_CHILD
] +
456 dd
->dd_phys
->dd_used_breakdown
[DD_USED_CHILD_RSRV
]);
458 mutex_exit(&dd
->dd_lock
);
460 if (dsl_dir_is_clone(dd
)) {
462 char buf
[MAXNAMELEN
];
464 VERIFY0(dsl_dataset_hold_obj(dd
->dd_pool
,
465 dd
->dd_phys
->dd_origin_obj
, FTAG
, &ds
));
466 dsl_dataset_name(ds
, buf
);
467 dsl_dataset_rele(ds
, FTAG
);
468 dsl_prop_nvlist_add_string(nv
, ZFS_PROP_ORIGIN
, buf
);
473 dsl_dir_dirty(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
475 dsl_pool_t
*dp
= dd
->dd_pool
;
479 if (txg_list_add(&dp
->dp_dirty_dirs
, dd
, tx
->tx_txg
)) {
480 /* up the hold count until we can be written out */
481 dmu_buf_add_ref(dd
->dd_dbuf
, dd
);
486 parent_delta(dsl_dir_t
*dd
, uint64_t used
, int64_t delta
)
488 uint64_t old_accounted
= MAX(used
, dd
->dd_phys
->dd_reserved
);
489 uint64_t new_accounted
= MAX(used
+ delta
, dd
->dd_phys
->dd_reserved
);
490 return (new_accounted
- old_accounted
);
494 dsl_dir_sync(dsl_dir_t
*dd
, dmu_tx_t
*tx
)
496 ASSERT(dmu_tx_is_syncing(tx
));
498 mutex_enter(&dd
->dd_lock
);
499 ASSERT0(dd
->dd_tempreserved
[tx
->tx_txg
&TXG_MASK
]);
500 dprintf_dd(dd
, "txg=%llu towrite=%lluK\n", tx
->tx_txg
,
501 dd
->dd_space_towrite
[tx
->tx_txg
&TXG_MASK
] / 1024);
502 dd
->dd_space_towrite
[tx
->tx_txg
&TXG_MASK
] = 0;
503 mutex_exit(&dd
->dd_lock
);
505 /* release the hold from dsl_dir_dirty */
506 dmu_buf_rele(dd
->dd_dbuf
, dd
);
510 dsl_dir_space_towrite(dsl_dir_t
*dd
)
515 ASSERT(MUTEX_HELD(&dd
->dd_lock
));
517 for (i
= 0; i
< TXG_SIZE
; i
++) {
518 space
+= dd
->dd_space_towrite
[i
&TXG_MASK
];
519 ASSERT3U(dd
->dd_space_towrite
[i
&TXG_MASK
], >=, 0);
525 * How much space would dd have available if ancestor had delta applied
526 * to it? If ondiskonly is set, we're only interested in what's
527 * on-disk, not estimated pending changes.
530 dsl_dir_space_available(dsl_dir_t
*dd
,
531 dsl_dir_t
*ancestor
, int64_t delta
, int ondiskonly
)
533 uint64_t parentspace
, myspace
, quota
, used
;
536 * If there are no restrictions otherwise, assume we have
537 * unlimited space available.
540 parentspace
= UINT64_MAX
;
542 if (dd
->dd_parent
!= NULL
) {
543 parentspace
= dsl_dir_space_available(dd
->dd_parent
,
544 ancestor
, delta
, ondiskonly
);
547 mutex_enter(&dd
->dd_lock
);
548 if (dd
->dd_phys
->dd_quota
!= 0)
549 quota
= dd
->dd_phys
->dd_quota
;
550 used
= dd
->dd_phys
->dd_used_bytes
;
552 used
+= dsl_dir_space_towrite(dd
);
554 if (dd
->dd_parent
== NULL
) {
555 uint64_t poolsize
= dsl_pool_adjustedsize(dd
->dd_pool
, FALSE
);
556 quota
= MIN(quota
, poolsize
);
559 if (dd
->dd_phys
->dd_reserved
> used
&& parentspace
!= UINT64_MAX
) {
561 * We have some space reserved, in addition to what our
564 parentspace
+= dd
->dd_phys
->dd_reserved
- used
;
567 if (dd
== ancestor
) {
569 ASSERT(used
>= -delta
);
571 if (parentspace
!= UINT64_MAX
)
572 parentspace
-= delta
;
580 * the lesser of the space provided by our parent and
581 * the space left in our quota
583 myspace
= MIN(parentspace
, quota
- used
);
586 mutex_exit(&dd
->dd_lock
);
598 dsl_dir_tempreserve_impl(dsl_dir_t
*dd
, uint64_t asize
, boolean_t netfree
,
599 boolean_t ignorequota
, boolean_t checkrefquota
, list_t
*tr_list
,
600 dmu_tx_t
*tx
, boolean_t first
)
602 uint64_t txg
= tx
->tx_txg
;
603 uint64_t est_inflight
, used_on_disk
, quota
, parent_rsrv
;
604 uint64_t deferred
= 0;
605 struct tempreserve
*tr
;
607 int txgidx
= txg
& TXG_MASK
;
609 uint64_t ref_rsrv
= 0;
611 ASSERT3U(txg
, !=, 0);
612 ASSERT3S(asize
, >, 0);
614 mutex_enter(&dd
->dd_lock
);
617 * Check against the dsl_dir's quota. We don't add in the delta
618 * when checking for over-quota because they get one free hit.
620 est_inflight
= dsl_dir_space_towrite(dd
);
621 for (i
= 0; i
< TXG_SIZE
; i
++)
622 est_inflight
+= dd
->dd_tempreserved
[i
];
623 used_on_disk
= dd
->dd_phys
->dd_used_bytes
;
626 * On the first iteration, fetch the dataset's used-on-disk and
627 * refreservation values. Also, if checkrefquota is set, test if
628 * allocating this space would exceed the dataset's refquota.
630 if (first
&& tx
->tx_objset
) {
632 dsl_dataset_t
*ds
= tx
->tx_objset
->os_dsl_dataset
;
634 error
= dsl_dataset_check_quota(ds
, checkrefquota
,
635 asize
, est_inflight
, &used_on_disk
, &ref_rsrv
);
637 mutex_exit(&dd
->dd_lock
);
643 * If this transaction will result in a net free of space,
644 * we want to let it through.
646 if (ignorequota
|| netfree
|| dd
->dd_phys
->dd_quota
== 0)
649 quota
= dd
->dd_phys
->dd_quota
;
652 * Adjust the quota against the actual pool size at the root
653 * minus any outstanding deferred frees.
654 * To ensure that it's possible to remove files from a full
655 * pool without inducing transient overcommits, we throttle
656 * netfree transactions against a quota that is slightly larger,
657 * but still within the pool's allocation slop. In cases where
658 * we're very close to full, this will allow a steady trickle of
659 * removes to get through.
661 if (dd
->dd_parent
== NULL
) {
662 spa_t
*spa
= dd
->dd_pool
->dp_spa
;
663 uint64_t poolsize
= dsl_pool_adjustedsize(dd
->dd_pool
, netfree
);
664 deferred
= metaslab_class_get_deferred(spa_normal_class(spa
));
665 if (poolsize
- deferred
< quota
) {
666 quota
= poolsize
- deferred
;
672 * If they are requesting more space, and our current estimate
673 * is over quota, they get to try again unless the actual
674 * on-disk is over quota and there are no pending changes (which
675 * may free up space for us).
677 if (used_on_disk
+ est_inflight
>= quota
) {
678 if (est_inflight
> 0 || used_on_disk
< quota
||
679 (retval
== ENOSPC
&& used_on_disk
< quota
+ deferred
))
681 dprintf_dd(dd
, "failing: used=%lluK inflight = %lluK "
682 "quota=%lluK tr=%lluK err=%d\n",
683 used_on_disk
>>10, est_inflight
>>10,
684 quota
>>10, asize
>>10, retval
);
685 mutex_exit(&dd
->dd_lock
);
686 return (SET_ERROR(retval
));
689 /* We need to up our estimated delta before dropping dd_lock */
690 dd
->dd_tempreserved
[txgidx
] += asize
;
692 parent_rsrv
= parent_delta(dd
, used_on_disk
+ est_inflight
,
694 mutex_exit(&dd
->dd_lock
);
696 tr
= kmem_zalloc(sizeof (struct tempreserve
), KM_PUSHPAGE
);
699 list_insert_tail(tr_list
, tr
);
701 /* see if it's OK with our parent */
702 if (dd
->dd_parent
&& parent_rsrv
) {
703 boolean_t ismos
= (dd
->dd_phys
->dd_head_dataset_obj
== 0);
705 return (dsl_dir_tempreserve_impl(dd
->dd_parent
,
706 parent_rsrv
, netfree
, ismos
, TRUE
, tr_list
, tx
, FALSE
));
713 * Reserve space in this dsl_dir, to be used in this tx's txg.
714 * After the space has been dirtied (and dsl_dir_willuse_space()
715 * has been called), the reservation should be canceled, using
716 * dsl_dir_tempreserve_clear().
719 dsl_dir_tempreserve_space(dsl_dir_t
*dd
, uint64_t lsize
, uint64_t asize
,
720 uint64_t fsize
, uint64_t usize
, void **tr_cookiep
, dmu_tx_t
*tx
)
730 tr_list
= kmem_alloc(sizeof (list_t
), KM_PUSHPAGE
);
731 list_create(tr_list
, sizeof (struct tempreserve
),
732 offsetof(struct tempreserve
, tr_node
));
733 ASSERT3S(asize
, >, 0);
734 ASSERT3S(fsize
, >=, 0);
736 err
= arc_tempreserve_space(lsize
, tx
->tx_txg
);
738 struct tempreserve
*tr
;
740 tr
= kmem_zalloc(sizeof (struct tempreserve
), KM_PUSHPAGE
);
742 list_insert_tail(tr_list
, tr
);
746 * If arc_memory_throttle() detected that pageout
747 * is running and we are low on memory, we delay new
748 * non-pageout transactions to give pageout an
751 * It is unfortunate to be delaying while the caller's
754 txg_delay(dd
->dd_pool
, tx
->tx_txg
,
755 MSEC2NSEC(10), MSEC2NSEC(10));
756 err
= SET_ERROR(ERESTART
);
761 err
= dsl_dir_tempreserve_impl(dd
, asize
, fsize
>= asize
,
762 FALSE
, asize
> usize
, tr_list
, tx
, TRUE
);
766 dsl_dir_tempreserve_clear(tr_list
, tx
);
768 *tr_cookiep
= tr_list
;
774 * Clear a temporary reservation that we previously made with
775 * dsl_dir_tempreserve_space().
778 dsl_dir_tempreserve_clear(void *tr_cookie
, dmu_tx_t
*tx
)
780 int txgidx
= tx
->tx_txg
& TXG_MASK
;
781 list_t
*tr_list
= tr_cookie
;
782 struct tempreserve
*tr
;
784 ASSERT3U(tx
->tx_txg
, !=, 0);
786 if (tr_cookie
== NULL
)
789 while ((tr
= list_head(tr_list
)) != NULL
) {
791 mutex_enter(&tr
->tr_ds
->dd_lock
);
792 ASSERT3U(tr
->tr_ds
->dd_tempreserved
[txgidx
], >=,
794 tr
->tr_ds
->dd_tempreserved
[txgidx
] -= tr
->tr_size
;
795 mutex_exit(&tr
->tr_ds
->dd_lock
);
797 arc_tempreserve_clear(tr
->tr_size
);
799 list_remove(tr_list
, tr
);
800 kmem_free(tr
, sizeof (struct tempreserve
));
803 kmem_free(tr_list
, sizeof (list_t
));
807 * This should be called from open context when we think we're going to write
808 * or free space, for example when dirtying data. Be conservative; it's okay
809 * to write less space or free more, but we don't want to write more or free
810 * less than the amount specified.
813 dsl_dir_willuse_space(dsl_dir_t
*dd
, int64_t space
, dmu_tx_t
*tx
)
815 int64_t parent_space
;
818 mutex_enter(&dd
->dd_lock
);
820 dd
->dd_space_towrite
[tx
->tx_txg
& TXG_MASK
] += space
;
822 est_used
= dsl_dir_space_towrite(dd
) + dd
->dd_phys
->dd_used_bytes
;
823 parent_space
= parent_delta(dd
, est_used
, space
);
824 mutex_exit(&dd
->dd_lock
);
826 /* Make sure that we clean up dd_space_to* */
827 dsl_dir_dirty(dd
, tx
);
829 /* XXX this is potentially expensive and unnecessary... */
830 if (parent_space
&& dd
->dd_parent
)
831 dsl_dir_willuse_space(dd
->dd_parent
, parent_space
, tx
);
834 /* call from syncing context when we actually write/free space for this dd */
836 dsl_dir_diduse_space(dsl_dir_t
*dd
, dd_used_t type
,
837 int64_t used
, int64_t compressed
, int64_t uncompressed
, dmu_tx_t
*tx
)
839 int64_t accounted_delta
;
842 * dsl_dataset_set_refreservation_sync_impl() calls this with
843 * dd_lock held, so that it can atomically update
844 * ds->ds_reserved and the dsl_dir accounting, so that
845 * dsl_dataset_check_quota() can see dataset and dir accounting
848 boolean_t needlock
= !MUTEX_HELD(&dd
->dd_lock
);
850 ASSERT(dmu_tx_is_syncing(tx
));
851 ASSERT(type
< DD_USED_NUM
);
853 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
856 mutex_enter(&dd
->dd_lock
);
857 accounted_delta
= parent_delta(dd
, dd
->dd_phys
->dd_used_bytes
, used
);
858 ASSERT(used
>= 0 || dd
->dd_phys
->dd_used_bytes
>= -used
);
859 ASSERT(compressed
>= 0 ||
860 dd
->dd_phys
->dd_compressed_bytes
>= -compressed
);
861 ASSERT(uncompressed
>= 0 ||
862 dd
->dd_phys
->dd_uncompressed_bytes
>= -uncompressed
);
863 dd
->dd_phys
->dd_used_bytes
+= used
;
864 dd
->dd_phys
->dd_uncompressed_bytes
+= uncompressed
;
865 dd
->dd_phys
->dd_compressed_bytes
+= compressed
;
867 if (dd
->dd_phys
->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
869 dd
->dd_phys
->dd_used_breakdown
[type
] >= -used
);
870 dd
->dd_phys
->dd_used_breakdown
[type
] += used
;
875 for (t
= 0; t
< DD_USED_NUM
; t
++)
876 u
+= dd
->dd_phys
->dd_used_breakdown
[t
];
877 ASSERT3U(u
, ==, dd
->dd_phys
->dd_used_bytes
);
882 mutex_exit(&dd
->dd_lock
);
884 if (dd
->dd_parent
!= NULL
) {
885 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD
,
886 accounted_delta
, compressed
, uncompressed
, tx
);
887 dsl_dir_transfer_space(dd
->dd_parent
,
888 used
- accounted_delta
,
889 DD_USED_CHILD_RSRV
, DD_USED_CHILD
, tx
);
894 dsl_dir_transfer_space(dsl_dir_t
*dd
, int64_t delta
,
895 dd_used_t oldtype
, dd_used_t newtype
, dmu_tx_t
*tx
)
897 ASSERT(dmu_tx_is_syncing(tx
));
898 ASSERT(oldtype
< DD_USED_NUM
);
899 ASSERT(newtype
< DD_USED_NUM
);
901 if (delta
== 0 || !(dd
->dd_phys
->dd_flags
& DD_FLAG_USED_BREAKDOWN
))
904 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
905 mutex_enter(&dd
->dd_lock
);
907 dd
->dd_phys
->dd_used_breakdown
[oldtype
] >= delta
:
908 dd
->dd_phys
->dd_used_breakdown
[newtype
] >= -delta
);
909 ASSERT(dd
->dd_phys
->dd_used_bytes
>= ABS(delta
));
910 dd
->dd_phys
->dd_used_breakdown
[oldtype
] -= delta
;
911 dd
->dd_phys
->dd_used_breakdown
[newtype
] += delta
;
912 mutex_exit(&dd
->dd_lock
);
915 typedef struct dsl_dir_set_qr_arg
{
916 const char *ddsqra_name
;
917 zprop_source_t ddsqra_source
;
918 uint64_t ddsqra_value
;
919 } dsl_dir_set_qr_arg_t
;
922 dsl_dir_set_quota_check(void *arg
, dmu_tx_t
*tx
)
924 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
925 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
928 uint64_t towrite
, newval
;
930 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
934 error
= dsl_prop_predict(ds
->ds_dir
, "quota",
935 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
937 dsl_dataset_rele(ds
, FTAG
);
942 dsl_dataset_rele(ds
, FTAG
);
946 mutex_enter(&ds
->ds_dir
->dd_lock
);
948 * If we are doing the preliminary check in open context, and
949 * there are pending changes, then don't fail it, since the
950 * pending changes could under-estimate the amount of space to be
953 towrite
= dsl_dir_space_towrite(ds
->ds_dir
);
954 if ((dmu_tx_is_syncing(tx
) || towrite
== 0) &&
955 (newval
< ds
->ds_dir
->dd_phys
->dd_reserved
||
956 newval
< ds
->ds_dir
->dd_phys
->dd_used_bytes
+ towrite
)) {
957 error
= SET_ERROR(ENOSPC
);
959 mutex_exit(&ds
->ds_dir
->dd_lock
);
960 dsl_dataset_rele(ds
, FTAG
);
965 dsl_dir_set_quota_sync(void *arg
, dmu_tx_t
*tx
)
967 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
968 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
972 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
974 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_RECVD_PROPS
) {
975 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_QUOTA
),
976 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
977 &ddsqra
->ddsqra_value
, tx
);
979 VERIFY0(dsl_prop_get_int_ds(ds
,
980 zfs_prop_to_name(ZFS_PROP_QUOTA
), &newval
));
982 newval
= ddsqra
->ddsqra_value
;
983 spa_history_log_internal_ds(ds
, "set", tx
, "%s=%lld",
984 zfs_prop_to_name(ZFS_PROP_QUOTA
), (longlong_t
)newval
);
987 dmu_buf_will_dirty(ds
->ds_dir
->dd_dbuf
, tx
);
988 mutex_enter(&ds
->ds_dir
->dd_lock
);
989 ds
->ds_dir
->dd_phys
->dd_quota
= newval
;
990 mutex_exit(&ds
->ds_dir
->dd_lock
);
991 dsl_dataset_rele(ds
, FTAG
);
995 dsl_dir_set_quota(const char *ddname
, zprop_source_t source
, uint64_t quota
)
997 dsl_dir_set_qr_arg_t ddsqra
;
999 ddsqra
.ddsqra_name
= ddname
;
1000 ddsqra
.ddsqra_source
= source
;
1001 ddsqra
.ddsqra_value
= quota
;
1003 return (dsl_sync_task(ddname
, dsl_dir_set_quota_check
,
1004 dsl_dir_set_quota_sync
, &ddsqra
, 0));
1008 dsl_dir_set_reservation_check(void *arg
, dmu_tx_t
*tx
)
1010 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1011 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1014 uint64_t newval
, used
, avail
;
1017 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
1023 * If we are doing the preliminary check in open context, the
1024 * space estimates may be inaccurate.
1026 if (!dmu_tx_is_syncing(tx
)) {
1027 dsl_dataset_rele(ds
, FTAG
);
1031 error
= dsl_prop_predict(ds
->ds_dir
,
1032 zfs_prop_to_name(ZFS_PROP_RESERVATION
),
1033 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
1035 dsl_dataset_rele(ds
, FTAG
);
1039 mutex_enter(&dd
->dd_lock
);
1040 used
= dd
->dd_phys
->dd_used_bytes
;
1041 mutex_exit(&dd
->dd_lock
);
1043 if (dd
->dd_parent
) {
1044 avail
= dsl_dir_space_available(dd
->dd_parent
,
1047 avail
= dsl_pool_adjustedsize(dd
->dd_pool
, B_FALSE
) - used
;
1050 if (MAX(used
, newval
) > MAX(used
, dd
->dd_phys
->dd_reserved
)) {
1051 uint64_t delta
= MAX(used
, newval
) -
1052 MAX(used
, dd
->dd_phys
->dd_reserved
);
1054 if (delta
> avail
||
1055 (dd
->dd_phys
->dd_quota
> 0 &&
1056 newval
> dd
->dd_phys
->dd_quota
))
1057 error
= SET_ERROR(ENOSPC
);
1060 dsl_dataset_rele(ds
, FTAG
);
1065 dsl_dir_set_reservation_sync_impl(dsl_dir_t
*dd
, uint64_t value
, dmu_tx_t
*tx
)
1070 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
1072 mutex_enter(&dd
->dd_lock
);
1073 used
= dd
->dd_phys
->dd_used_bytes
;
1074 delta
= MAX(used
, value
) - MAX(used
, dd
->dd_phys
->dd_reserved
);
1075 dd
->dd_phys
->dd_reserved
= value
;
1077 if (dd
->dd_parent
!= NULL
) {
1078 /* Roll up this additional usage into our ancestors */
1079 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD_RSRV
,
1082 mutex_exit(&dd
->dd_lock
);
1086 dsl_dir_set_reservation_sync(void *arg
, dmu_tx_t
*tx
)
1088 dsl_dir_set_qr_arg_t
*ddsqra
= arg
;
1089 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1093 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
1095 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_RECVD_PROPS
) {
1096 dsl_prop_set_sync_impl(ds
,
1097 zfs_prop_to_name(ZFS_PROP_RESERVATION
),
1098 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
1099 &ddsqra
->ddsqra_value
, tx
);
1101 VERIFY0(dsl_prop_get_int_ds(ds
,
1102 zfs_prop_to_name(ZFS_PROP_RESERVATION
), &newval
));
1104 newval
= ddsqra
->ddsqra_value
;
1105 spa_history_log_internal_ds(ds
, "set", tx
, "%s=%lld",
1106 zfs_prop_to_name(ZFS_PROP_RESERVATION
),
1107 (longlong_t
)newval
);
1110 dsl_dir_set_reservation_sync_impl(ds
->ds_dir
, newval
, tx
);
1111 dsl_dataset_rele(ds
, FTAG
);
1115 dsl_dir_set_reservation(const char *ddname
, zprop_source_t source
,
1116 uint64_t reservation
)
1118 dsl_dir_set_qr_arg_t ddsqra
;
1120 ddsqra
.ddsqra_name
= ddname
;
1121 ddsqra
.ddsqra_source
= source
;
1122 ddsqra
.ddsqra_value
= reservation
;
1124 return (dsl_sync_task(ddname
, dsl_dir_set_reservation_check
,
1125 dsl_dir_set_reservation_sync
, &ddsqra
, 0));
1129 closest_common_ancestor(dsl_dir_t
*ds1
, dsl_dir_t
*ds2
)
1131 for (; ds1
; ds1
= ds1
->dd_parent
) {
1133 for (dd
= ds2
; dd
; dd
= dd
->dd_parent
) {
1142 * If delta is applied to dd, how much of that delta would be applied to
1143 * ancestor? Syncing context only.
1146 would_change(dsl_dir_t
*dd
, int64_t delta
, dsl_dir_t
*ancestor
)
1151 mutex_enter(&dd
->dd_lock
);
1152 delta
= parent_delta(dd
, dd
->dd_phys
->dd_used_bytes
, delta
);
1153 mutex_exit(&dd
->dd_lock
);
1154 return (would_change(dd
->dd_parent
, delta
, ancestor
));
1157 typedef struct dsl_dir_rename_arg
{
1158 const char *ddra_oldname
;
1159 const char *ddra_newname
;
1160 } dsl_dir_rename_arg_t
;
1164 dsl_valid_rename(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
1167 char namebuf
[MAXNAMELEN
];
1169 dsl_dataset_name(ds
, namebuf
);
1171 if (strlen(namebuf
) + *deltap
>= MAXNAMELEN
)
1172 return (SET_ERROR(ENAMETOOLONG
));
1177 dsl_dir_rename_check(void *arg
, dmu_tx_t
*tx
)
1179 dsl_dir_rename_arg_t
*ddra
= arg
;
1180 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1181 dsl_dir_t
*dd
, *newparent
;
1182 const char *mynewname
;
1184 int delta
= strlen(ddra
->ddra_newname
) - strlen(ddra
->ddra_oldname
);
1186 /* target dir should exist */
1187 error
= dsl_dir_hold(dp
, ddra
->ddra_oldname
, FTAG
, &dd
, NULL
);
1191 /* new parent should exist */
1192 error
= dsl_dir_hold(dp
, ddra
->ddra_newname
, FTAG
,
1193 &newparent
, &mynewname
);
1195 dsl_dir_rele(dd
, FTAG
);
1199 /* can't rename to different pool */
1200 if (dd
->dd_pool
!= newparent
->dd_pool
) {
1201 dsl_dir_rele(newparent
, FTAG
);
1202 dsl_dir_rele(dd
, FTAG
);
1203 return (SET_ERROR(ENXIO
));
1206 /* new name should not already exist */
1207 if (mynewname
== NULL
) {
1208 dsl_dir_rele(newparent
, FTAG
);
1209 dsl_dir_rele(dd
, FTAG
);
1210 return (SET_ERROR(EEXIST
));
1213 /* if the name length is growing, validate child name lengths */
1215 error
= dmu_objset_find_dp(dp
, dd
->dd_object
, dsl_valid_rename
,
1216 &delta
, DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
1218 dsl_dir_rele(newparent
, FTAG
);
1219 dsl_dir_rele(dd
, FTAG
);
1224 if (newparent
!= dd
->dd_parent
) {
1225 /* is there enough space? */
1227 MAX(dd
->dd_phys
->dd_used_bytes
, dd
->dd_phys
->dd_reserved
);
1229 /* no rename into our descendant */
1230 if (closest_common_ancestor(dd
, newparent
) == dd
) {
1231 dsl_dir_rele(newparent
, FTAG
);
1232 dsl_dir_rele(dd
, FTAG
);
1233 return (SET_ERROR(EINVAL
));
1236 error
= dsl_dir_transfer_possible(dd
->dd_parent
,
1237 newparent
, myspace
);
1239 dsl_dir_rele(newparent
, FTAG
);
1240 dsl_dir_rele(dd
, FTAG
);
1245 dsl_dir_rele(newparent
, FTAG
);
1246 dsl_dir_rele(dd
, FTAG
);
1251 dsl_dir_rename_sync(void *arg
, dmu_tx_t
*tx
)
1253 dsl_dir_rename_arg_t
*ddra
= arg
;
1254 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1255 dsl_dir_t
*dd
, *newparent
;
1256 const char *mynewname
;
1258 objset_t
*mos
= dp
->dp_meta_objset
;
1260 VERIFY0(dsl_dir_hold(dp
, ddra
->ddra_oldname
, FTAG
, &dd
, NULL
));
1261 VERIFY0(dsl_dir_hold(dp
, ddra
->ddra_newname
, FTAG
, &newparent
,
1264 /* Log this before we change the name. */
1265 spa_history_log_internal_dd(dd
, "rename", tx
,
1266 "-> %s", ddra
->ddra_newname
);
1268 if (newparent
!= dd
->dd_parent
) {
1269 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD
,
1270 -dd
->dd_phys
->dd_used_bytes
,
1271 -dd
->dd_phys
->dd_compressed_bytes
,
1272 -dd
->dd_phys
->dd_uncompressed_bytes
, tx
);
1273 dsl_dir_diduse_space(newparent
, DD_USED_CHILD
,
1274 dd
->dd_phys
->dd_used_bytes
,
1275 dd
->dd_phys
->dd_compressed_bytes
,
1276 dd
->dd_phys
->dd_uncompressed_bytes
, tx
);
1278 if (dd
->dd_phys
->dd_reserved
> dd
->dd_phys
->dd_used_bytes
) {
1279 uint64_t unused_rsrv
= dd
->dd_phys
->dd_reserved
-
1280 dd
->dd_phys
->dd_used_bytes
;
1282 dsl_dir_diduse_space(dd
->dd_parent
, DD_USED_CHILD_RSRV
,
1283 -unused_rsrv
, 0, 0, tx
);
1284 dsl_dir_diduse_space(newparent
, DD_USED_CHILD_RSRV
,
1285 unused_rsrv
, 0, 0, tx
);
1289 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
1291 /* remove from old parent zapobj */
1292 error
= zap_remove(mos
, dd
->dd_parent
->dd_phys
->dd_child_dir_zapobj
,
1296 (void) strcpy(dd
->dd_myname
, mynewname
);
1297 dsl_dir_rele(dd
->dd_parent
, dd
);
1298 dd
->dd_phys
->dd_parent_obj
= newparent
->dd_object
;
1299 VERIFY0(dsl_dir_hold_obj(dp
,
1300 newparent
->dd_object
, NULL
, dd
, &dd
->dd_parent
));
1302 /* add to new parent zapobj */
1303 VERIFY0(zap_add(mos
, newparent
->dd_phys
->dd_child_dir_zapobj
,
1304 dd
->dd_myname
, 8, 1, &dd
->dd_object
, tx
));
1307 zvol_rename_minors(ddra
->ddra_oldname
, ddra
->ddra_newname
);
1310 dsl_prop_notify_all(dd
);
1312 dsl_dir_rele(newparent
, FTAG
);
1313 dsl_dir_rele(dd
, FTAG
);
1317 dsl_dir_rename(const char *oldname
, const char *newname
)
1319 dsl_dir_rename_arg_t ddra
;
1321 ddra
.ddra_oldname
= oldname
;
1322 ddra
.ddra_newname
= newname
;
1324 return (dsl_sync_task(oldname
,
1325 dsl_dir_rename_check
, dsl_dir_rename_sync
, &ddra
, 3));
1329 dsl_dir_transfer_possible(dsl_dir_t
*sdd
, dsl_dir_t
*tdd
, uint64_t space
)
1331 dsl_dir_t
*ancestor
;
1335 ancestor
= closest_common_ancestor(sdd
, tdd
);
1336 adelta
= would_change(sdd
, -space
, ancestor
);
1337 avail
= dsl_dir_space_available(tdd
, ancestor
, adelta
, FALSE
);
1339 return (SET_ERROR(ENOSPC
));
1345 dsl_dir_snap_cmtime(dsl_dir_t
*dd
)
1349 mutex_enter(&dd
->dd_lock
);
1350 t
= dd
->dd_snap_cmtime
;
1351 mutex_exit(&dd
->dd_lock
);
1357 dsl_dir_snap_cmtime_update(dsl_dir_t
*dd
)
1362 mutex_enter(&dd
->dd_lock
);
1363 dd
->dd_snap_cmtime
= t
;
1364 mutex_exit(&dd
->dd_lock
);
1367 #if defined(_KERNEL) && defined(HAVE_SPL)
1368 EXPORT_SYMBOL(dsl_dir_set_quota
);
1369 EXPORT_SYMBOL(dsl_dir_set_reservation
);