Deleted Added
sdiff udiff text old ( 263391 ) new ( 265754 )
full compact
dsl_dir.c (263391) dsl_dir.c (265754)
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 unchanged lines hidden (view full) ---

18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2013 by Delphix. All rights reserved.
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 unchanged lines hidden (view full) ---

18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2013 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Joyent, Inc. All rights reserved.
26 */
27
28#include <sys/dmu.h>
29#include <sys/dmu_objset.h>
30#include <sys/dmu_tx.h>
31#include <sys/dsl_dataset.h>
32#include <sys/dsl_dir.h>
33#include <sys/dsl_prop.h>

--- 5 unchanged lines hidden (view full) ---

39#include <sys/zap.h>
40#include <sys/zio.h>
41#include <sys/arc.h>
42#include <sys/sunddi.h>
43#include <sys/zvol.h>
44#ifdef _KERNEL
45#include <sys/zfs_vfsops.h>
46#endif
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>

--- 5 unchanged lines hidden (view full) ---

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>
47#include "zfs_namecheck.h"
51#include "zfs_namecheck.h"
52#include "zfs_prop.h"
48
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
49static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
50
51/* ARGSUSED */
52static void
53dsl_dir_evict(dmu_buf_t *db, void *arg)
54{
55 dsl_dir_t *dd = arg;
56 dsl_pool_t *dp = dd->dd_pool;

--- 321 unchanged lines hidden (view full) ---

378 err = SET_ERROR(ENOENT);
379 }
380 if (tailp != NULL)
381 *tailp = next;
382 *ddp = dd;
383 return (err);
384}
385
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;

--- 321 unchanged lines hidden (view full) ---

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_enabled(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
553 dsl_dataset_rele(ds, FTAG);
554
555 kmem_free(zc, sizeof (zap_cursor_t));
556 kmem_free(za, sizeof (zap_attribute_t));
557
558 /* we're in a sync task, update counts */
559 dmu_buf_will_dirty(dd->dd_dbuf, tx);
560 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
561 sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
562 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
563 sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
564}
565
566static int
567dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
568{
569 char *ddname = (char *)arg;
570 dsl_pool_t *dp = dmu_tx_pool(tx);
571 dsl_dataset_t *ds;
572 dsl_dir_t *dd;
573 int error;
574
575 error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
576 if (error != 0)
577 return (error);
578
579 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
580 dsl_dataset_rele(ds, FTAG);
581 return (SET_ERROR(ENOTSUP));
582 }
583
584 dd = ds->ds_dir;
585 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
586 dsl_dir_is_zapified(dd) &&
587 zap_contains(dp->dp_meta_objset, dd->dd_object,
588 DD_FIELD_FILESYSTEM_COUNT) == 0) {
589 dsl_dataset_rele(ds, FTAG);
590 return (SET_ERROR(EALREADY));
591 }
592
593 dsl_dataset_rele(ds, FTAG);
594 return (0);
595}
596
597static void
598dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
599{
600 char *ddname = (char *)arg;
601 dsl_pool_t *dp = dmu_tx_pool(tx);
602 dsl_dataset_t *ds;
603 spa_t *spa;
604
605 VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
606
607 spa = dsl_dataset_get_spa(ds);
608
609 if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
610 /*
611 * Since the feature was not active and we're now setting a
612 * limit, increment the feature-active counter so that the
613 * feature becomes active for the first time.
614 *
615 * We are already in a sync task so we can update the MOS.
616 */
617 spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
618 }
619
620 /*
621 * Since we are now setting a non-UINT64_MAX limit on the filesystem,
622 * we need to ensure the counts are correct. Descend down the tree from
623 * this point and update all of the counts to be accurate.
624 */
625 dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
626
627 dsl_dataset_rele(ds, FTAG);
628}
629
630/*
631 * Make sure the feature is enabled and activate it if necessary.
632 * Since we're setting a limit, ensure the on-disk counts are valid.
633 * This is only called by the ioctl path when setting a limit value.
634 *
635 * We do not need to validate the new limit, since users who can change the
636 * limit are also allowed to exceed the limit.
637 */
638int
639dsl_dir_activate_fs_ss_limit(const char *ddname)
640{
641 int error;
642
643 error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
644 dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0);
645
646 if (error == EALREADY)
647 error = 0;
648
649 return (error);
650}
651
652/*
653 * Used to determine if the filesystem_limit or snapshot_limit should be
654 * enforced. We allow the limit to be exceeded if the user has permission to
655 * write the property value. We pass in the creds that we got in the open
656 * context since we will always be the GZ root in syncing context. We also have
657 * to handle the case where we are allowed to change the limit on the current
658 * dataset, but there may be another limit in the tree above.
659 *
660 * We can never modify these two properties within a non-global zone. In
661 * addition, the other checks are modeled on zfs_secpolicy_write_perms. We
662 * can't use that function since we are already holding the dp_config_rwlock.
663 * In addition, we already have the dd and dealing with snapshots is simplified
664 * in this code.
665 */
666
667typedef enum {
668 ENFORCE_ALWAYS,
669 ENFORCE_NEVER,
670 ENFORCE_ABOVE
671} enforce_res_t;
672
673static enforce_res_t
674dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
675{
676 enforce_res_t enforce = ENFORCE_ALWAYS;
677 uint64_t obj;
678 dsl_dataset_t *ds;
679 uint64_t zoned;
680
681 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
682 prop == ZFS_PROP_SNAPSHOT_LIMIT);
683
684#ifdef _KERNEL
685#ifdef __FreeBSD__
686 if (jailed(cr))
687#else
688 if (crgetzoneid(cr) != GLOBAL_ZONEID)
689#endif
690 return (ENFORCE_ALWAYS);
691
692 if (secpolicy_zfs(cr) == 0)
693 return (ENFORCE_NEVER);
694#endif
695
696 if ((obj = dd->dd_phys->dd_head_dataset_obj) == 0)
697 return (ENFORCE_ALWAYS);
698
699 ASSERT(dsl_pool_config_held(dd->dd_pool));
700
701 if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
702 return (ENFORCE_ALWAYS);
703
704 if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
705 /* Only root can access zoned fs's from the GZ */
706 enforce = ENFORCE_ALWAYS;
707 } else {
708 if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
709 enforce = ENFORCE_ABOVE;
710 }
711
712 dsl_dataset_rele(ds, FTAG);
713 return (enforce);
714}
715
716/*
717 * Check if adding additional child filesystem(s) would exceed any filesystem
718 * limits or adding additional snapshot(s) would exceed any snapshot limits.
719 * The prop argument indicates which limit to check.
720 *
721 * Note that all filesystem limits up to the root (or the highest
722 * initialized) filesystem or the given ancestor must be satisfied.
723 */
724int
725dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
726 dsl_dir_t *ancestor, cred_t *cr)
727{
728 objset_t *os = dd->dd_pool->dp_meta_objset;
729 uint64_t limit, count;
730 char *count_prop;
731 enforce_res_t enforce;
732 int err = 0;
733
734 ASSERT(dsl_pool_config_held(dd->dd_pool));
735 ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
736 prop == ZFS_PROP_SNAPSHOT_LIMIT);
737
738 /*
739 * If we're allowed to change the limit, don't enforce the limit
740 * e.g. this can happen if a snapshot is taken by an administrative
741 * user in the global zone (i.e. a recursive snapshot by root).
742 * However, we must handle the case of delegated permissions where we
743 * are allowed to change the limit on the current dataset, but there
744 * is another limit in the tree above.
745 */
746 enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
747 if (enforce == ENFORCE_NEVER)
748 return (0);
749
750 /*
751 * e.g. if renaming a dataset with no snapshots, count adjustment
752 * is 0.
753 */
754 if (delta == 0)
755 return (0);
756
757 if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
758 /*
759 * We don't enforce the limit for temporary snapshots. This is
760 * indicated by a NULL cred_t argument.
761 */
762 if (cr == NULL)
763 return (0);
764
765 count_prop = DD_FIELD_SNAPSHOT_COUNT;
766 } else {
767 count_prop = DD_FIELD_FILESYSTEM_COUNT;
768 }
769
770 /*
771 * If an ancestor has been provided, stop checking the limit once we
772 * hit that dir. We need this during rename so that we don't overcount
773 * the check once we recurse up to the common ancestor.
774 */
775 if (ancestor == dd)
776 return (0);
777
778 /*
779 * If we hit an uninitialized node while recursing up the tree, we can
780 * stop since we know there is no limit here (or above). The counts are
781 * not valid on this node and we know we won't touch this node's counts.
782 */
783 if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
784 count_prop, sizeof (count), 1, &count) == ENOENT)
785 return (0);
786
787 err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
788 B_FALSE);
789 if (err != 0)
790 return (err);
791
792 /* Is there a limit which we've hit? */
793 if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
794 return (SET_ERROR(EDQUOT));
795
796 if (dd->dd_parent != NULL)
797 err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
798 ancestor, cr);
799
800 return (err);
801}
802
803/*
804 * Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
805 * parents. When a new filesystem/snapshot is created, increment the count on
806 * all parents, and when a filesystem/snapshot is destroyed, decrement the
807 * count.
808 */
809void
810dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
811 dmu_tx_t *tx)
812{
813 int err;
814 objset_t *os = dd->dd_pool->dp_meta_objset;
815 uint64_t count;
816
817 ASSERT(dsl_pool_config_held(dd->dd_pool));
818 ASSERT(dmu_tx_is_syncing(tx));
819 ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
820 strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
821
822 /*
823 * When we receive an incremental stream into a filesystem that already
824 * exists, a temporary clone is created. We don't count this temporary
825 * clone, whose name begins with a '%'. We also ignore hidden ($FREE,
826 * $MOS & $ORIGIN) objsets.
827 */
828 if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
829 strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
830 return;
831
832 /*
833 * e.g. if renaming a dataset with no snapshots, count adjustment is 0
834 */
835 if (delta == 0)
836 return;
837
838 /*
839 * If we hit an uninitialized node while recursing up the tree, we can
840 * stop since we know the counts are not valid on this node and we
841 * know we shouldn't touch this node's counts. An uninitialized count
842 * on the node indicates that either the feature has not yet been
843 * activated or there are no limits on this part of the tree.
844 */
845 if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
846 prop, sizeof (count), 1, &count)) == ENOENT)
847 return;
848 VERIFY0(err);
849
850 count += delta;
851 /* Use a signed verify to make sure we're not neg. */
852 VERIFY3S(count, >=, 0);
853
854 VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
855 tx));
856
857 /* Roll up this additional count into our ancestors */
858 if (dd->dd_parent != NULL)
859 dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
860}
861
386uint64_t
387dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
388 dmu_tx_t *tx)
389{
390 objset_t *mos = dp->dp_meta_objset;
391 uint64_t ddobj;
392 dsl_dir_phys_t *ddphys;
393 dmu_buf_t *dbuf;

--- 8 unchanged lines hidden (view full) ---

402 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
403 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
404 }
405 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
406 dmu_buf_will_dirty(dbuf, tx);
407 ddphys = dbuf->db_data;
408
409 ddphys->dd_creation_time = gethrestime_sec();
862uint64_t
863dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
864 dmu_tx_t *tx)
865{
866 objset_t *mos = dp->dp_meta_objset;
867 uint64_t ddobj;
868 dsl_dir_phys_t *ddphys;
869 dmu_buf_t *dbuf;

--- 8 unchanged lines hidden (view full) ---

878 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
879 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
880 }
881 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
882 dmu_buf_will_dirty(dbuf, tx);
883 ddphys = dbuf->db_data;
884
885 ddphys->dd_creation_time = gethrestime_sec();
410 if (pds)
886 if (pds) {
411 ddphys->dd_parent_obj = pds->dd_object;
887 ddphys->dd_parent_obj = pds->dd_object;
888
889 /* update the filesystem counts */
890 dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
891 }
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);
419

--- 32 unchanged lines hidden (view full) ---

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]);
457 }
458 mutex_exit(&dd->dd_lock);
459
892 ddphys->dd_props_zapobj = zap_create(mos,
893 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
894 ddphys->dd_child_dir_zapobj = zap_create(mos,
895 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
896 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
897 ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
898 dmu_buf_rele(dbuf, FTAG);
899

--- 32 unchanged lines hidden (view full) ---

932 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
933 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
934 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
935 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
936 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
937 }
938 mutex_exit(&dd->dd_lock);
939
940 if (dsl_dir_is_zapified(dd)) {
941 uint64_t count;
942 objset_t *os = dd->dd_pool->dp_meta_objset;
943
944 if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
945 sizeof (count), 1, &count) == 0) {
946 dsl_prop_nvlist_add_uint64(nv,
947 ZFS_PROP_FILESYSTEM_COUNT, count);
948 }
949 if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
950 sizeof (count), 1, &count) == 0) {
951 dsl_prop_nvlist_add_uint64(nv,
952 ZFS_PROP_SNAPSHOT_COUNT, count);
953 }
954 }
955
460 if (dsl_dir_is_clone(dd)) {
461 dsl_dataset_t *ds;
462 char buf[MAXNAMELEN];
463
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);

--- 682 unchanged lines hidden (view full) ---

1150 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1151 mutex_exit(&dd->dd_lock);
1152 return (would_change(dd->dd_parent, delta, ancestor));
1153}
1154
1155typedef struct dsl_dir_rename_arg {
1156 const char *ddra_oldname;
1157 const char *ddra_newname;
956 if (dsl_dir_is_clone(dd)) {
957 dsl_dataset_t *ds;
958 char buf[MAXNAMELEN];
959
960 VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
961 dd->dd_phys->dd_origin_obj, FTAG, &ds));
962 dsl_dataset_name(ds, buf);
963 dsl_dataset_rele(ds, FTAG);

--- 682 unchanged lines hidden (view full) ---

1646 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
1647 mutex_exit(&dd->dd_lock);
1648 return (would_change(dd->dd_parent, delta, ancestor));
1649}
1650
1651typedef struct dsl_dir_rename_arg {
1652 const char *ddra_oldname;
1653 const char *ddra_newname;
1654 cred_t *ddra_cred;
1158} dsl_dir_rename_arg_t;
1159
1160/* ARGSUSED */
1161static int
1162dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1163{
1164 int *deltap = arg;
1165 char namebuf[MAXNAMELEN];

--- 48 unchanged lines hidden (view full) ---

1214 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1215 if (error != 0) {
1216 dsl_dir_rele(newparent, FTAG);
1217 dsl_dir_rele(dd, FTAG);
1218 return (error);
1219 }
1220 }
1221
1655} dsl_dir_rename_arg_t;
1656
1657/* ARGSUSED */
1658static int
1659dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1660{
1661 int *deltap = arg;
1662 char namebuf[MAXNAMELEN];

--- 48 unchanged lines hidden (view full) ---

1711 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1712 if (error != 0) {
1713 dsl_dir_rele(newparent, FTAG);
1714 dsl_dir_rele(dd, FTAG);
1715 return (error);
1716 }
1717 }
1718
1719 if (dmu_tx_is_syncing(tx)) {
1720 if (spa_feature_is_enabled(dp->dp_spa,
1721 SPA_FEATURE_FS_SS_LIMIT)) {
1722 /*
1723 * Although this is the check function and we don't
1724 * normally make on-disk changes in check functions,
1725 * we need to do that here.
1726 *
1727 * Ensure this portion of the tree's counts have been
1728 * initialized in case the new parent has limits set.
1729 */
1730 dsl_dir_init_fs_ss_count(dd, tx);
1731 }
1732 }
1733
1222 if (newparent != dd->dd_parent) {
1223 /* is there enough space? */
1224 uint64_t myspace =
1225 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1734 if (newparent != dd->dd_parent) {
1735 /* is there enough space? */
1736 uint64_t myspace =
1737 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
1738 objset_t *os = dd->dd_pool->dp_meta_objset;
1739 uint64_t fs_cnt = 0;
1740 uint64_t ss_cnt = 0;
1226
1741
1742 if (dsl_dir_is_zapified(dd)) {
1743 int err;
1744
1745 err = zap_lookup(os, dd->dd_object,
1746 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1747 &fs_cnt);
1748 if (err != ENOENT && err != 0)
1749 return (err);
1750
1751 /*
1752 * have to add 1 for the filesystem itself that we're
1753 * moving
1754 */
1755 fs_cnt++;
1756
1757 err = zap_lookup(os, dd->dd_object,
1758 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1759 &ss_cnt);
1760 if (err != ENOENT && err != 0)
1761 return (err);
1762 }
1763
1227 /* no rename into our descendant */
1228 if (closest_common_ancestor(dd, newparent) == dd) {
1229 dsl_dir_rele(newparent, FTAG);
1230 dsl_dir_rele(dd, FTAG);
1231 return (SET_ERROR(EINVAL));
1232 }
1233
1234 error = dsl_dir_transfer_possible(dd->dd_parent,
1764 /* no rename into our descendant */
1765 if (closest_common_ancestor(dd, newparent) == dd) {
1766 dsl_dir_rele(newparent, FTAG);
1767 dsl_dir_rele(dd, FTAG);
1768 return (SET_ERROR(EINVAL));
1769 }
1770
1771 error = dsl_dir_transfer_possible(dd->dd_parent,
1235 newparent, myspace);
1772 newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
1236 if (error != 0) {
1237 dsl_dir_rele(newparent, FTAG);
1238 dsl_dir_rele(dd, FTAG);
1239 return (error);
1240 }
1241 }
1242
1243 dsl_dir_rele(newparent, FTAG);

--- 15 unchanged lines hidden (view full) ---

1259 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1260 &mynewname));
1261
1262 /* Log this before we change the name. */
1263 spa_history_log_internal_dd(dd, "rename", tx,
1264 "-> %s", ddra->ddra_newname);
1265
1266 if (newparent != dd->dd_parent) {
1773 if (error != 0) {
1774 dsl_dir_rele(newparent, FTAG);
1775 dsl_dir_rele(dd, FTAG);
1776 return (error);
1777 }
1778 }
1779
1780 dsl_dir_rele(newparent, FTAG);

--- 15 unchanged lines hidden (view full) ---

1796 VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
1797 &mynewname));
1798
1799 /* Log this before we change the name. */
1800 spa_history_log_internal_dd(dd, "rename", tx,
1801 "-> %s", ddra->ddra_newname);
1802
1803 if (newparent != dd->dd_parent) {
1804 objset_t *os = dd->dd_pool->dp_meta_objset;
1805 uint64_t fs_cnt = 0;
1806 uint64_t ss_cnt = 0;
1807
1808 /*
1809 * We already made sure the dd counts were initialized in the
1810 * check function.
1811 */
1812 if (spa_feature_is_enabled(dp->dp_spa,
1813 SPA_FEATURE_FS_SS_LIMIT)) {
1814 VERIFY0(zap_lookup(os, dd->dd_object,
1815 DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
1816 &fs_cnt));
1817 /* add 1 for the filesystem itself that we're moving */
1818 fs_cnt++;
1819
1820 VERIFY0(zap_lookup(os, dd->dd_object,
1821 DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
1822 &ss_cnt));
1823 }
1824
1825 dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
1826 DD_FIELD_FILESYSTEM_COUNT, tx);
1827 dsl_fs_ss_count_adjust(newparent, fs_cnt,
1828 DD_FIELD_FILESYSTEM_COUNT, tx);
1829
1830 dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
1831 DD_FIELD_SNAPSHOT_COUNT, tx);
1832 dsl_fs_ss_count_adjust(newparent, ss_cnt,
1833 DD_FIELD_SNAPSHOT_COUNT, tx);
1834
1267 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1268 -dd->dd_phys->dd_used_bytes,
1269 -dd->dd_phys->dd_compressed_bytes,
1270 -dd->dd_phys->dd_uncompressed_bytes, tx);
1271 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1272 dd->dd_phys->dd_used_bytes,
1273 dd->dd_phys->dd_compressed_bytes,
1274 dd->dd_phys->dd_uncompressed_bytes, tx);

--- 41 unchanged lines hidden (view full) ---

1316
1317int
1318dsl_dir_rename(const char *oldname, const char *newname)
1319{
1320 dsl_dir_rename_arg_t ddra;
1321
1322 ddra.ddra_oldname = oldname;
1323 ddra.ddra_newname = newname;
1835 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
1836 -dd->dd_phys->dd_used_bytes,
1837 -dd->dd_phys->dd_compressed_bytes,
1838 -dd->dd_phys->dd_uncompressed_bytes, tx);
1839 dsl_dir_diduse_space(newparent, DD_USED_CHILD,
1840 dd->dd_phys->dd_used_bytes,
1841 dd->dd_phys->dd_compressed_bytes,
1842 dd->dd_phys->dd_uncompressed_bytes, tx);

--- 41 unchanged lines hidden (view full) ---

1884
1885int
1886dsl_dir_rename(const char *oldname, const char *newname)
1887{
1888 dsl_dir_rename_arg_t ddra;
1889
1890 ddra.ddra_oldname = oldname;
1891 ddra.ddra_newname = newname;
1892 ddra.ddra_cred = CRED();
1324
1325 return (dsl_sync_task(oldname,
1326 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1327}
1328
1329int
1893
1894 return (dsl_sync_task(oldname,
1895 dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
1896}
1897
1898int
1330dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
1899dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
1900 uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
1331{
1332 dsl_dir_t *ancestor;
1333 int64_t adelta;
1334 uint64_t avail;
1901{
1902 dsl_dir_t *ancestor;
1903 int64_t adelta;
1904 uint64_t avail;
1905 int err;
1335
1336 ancestor = closest_common_ancestor(sdd, tdd);
1337 adelta = would_change(sdd, -space, ancestor);
1338 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1339 if (avail < space)
1340 return (SET_ERROR(ENOSPC));
1341
1906
1907 ancestor = closest_common_ancestor(sdd, tdd);
1908 adelta = would_change(sdd, -space, ancestor);
1909 avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
1910 if (avail < space)
1911 return (SET_ERROR(ENOSPC));
1912
1913 err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
1914 ancestor, cr);
1915 if (err != 0)
1916 return (err);
1917 err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
1918 ancestor, cr);
1919 if (err != 0)
1920 return (err);
1921
1342 return (0);
1343}
1344
1345timestruc_t
1346dsl_dir_snap_cmtime(dsl_dir_t *dd)
1347{
1348 timestruc_t t;
1349

--- 16 unchanged lines hidden (view full) ---

1366}
1367
1368void
1369dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1370{
1371 objset_t *mos = dd->dd_pool->dp_meta_objset;
1372 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1373}
1922 return (0);
1923}
1924
1925timestruc_t
1926dsl_dir_snap_cmtime(dsl_dir_t *dd)
1927{
1928 timestruc_t t;
1929

--- 16 unchanged lines hidden (view full) ---

1946}
1947
1948void
1949dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
1950{
1951 objset_t *mos = dd->dd_pool->dp_meta_objset;
1952 dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
1953}
1954
1955boolean_t
1956dsl_dir_is_zapified(dsl_dir_t *dd)
1957{
1958 dmu_object_info_t doi;
1959
1960 dmu_object_info_from_db(dd->dd_dbuf, &doi);
1961 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
1962}