1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2011 STRATO. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sched/mm.h>
17 #include "transaction.h"
22 #include "extent_io.h"
24 #include "block-group.h"
26 #include "tree-mod-log.h"
28 #include "accessors.h"
29 #include "extent-tree.h"
30 #include "root-tree.h"
31 #include "tree-checker.h"
33 enum btrfs_qgroup_mode
btrfs_qgroup_mode(const struct btrfs_fs_info
*fs_info
)
35 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
))
36 return BTRFS_QGROUP_MODE_DISABLED
;
37 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_SIMPLE_MODE
)
38 return BTRFS_QGROUP_MODE_SIMPLE
;
39 return BTRFS_QGROUP_MODE_FULL
;
42 bool btrfs_qgroup_enabled(const struct btrfs_fs_info
*fs_info
)
44 return btrfs_qgroup_mode(fs_info
) != BTRFS_QGROUP_MODE_DISABLED
;
47 bool btrfs_qgroup_full_accounting(const struct btrfs_fs_info
*fs_info
)
49 return btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_FULL
;
53 * Helpers to access qgroup reservation
55 * Callers should ensure the lock context and type are valid
58 static u64
qgroup_rsv_total(const struct btrfs_qgroup
*qgroup
)
63 for (i
= 0; i
< BTRFS_QGROUP_RSV_LAST
; i
++)
64 ret
+= qgroup
->rsv
.values
[i
];
69 #ifdef CONFIG_BTRFS_DEBUG
70 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type
)
72 if (type
== BTRFS_QGROUP_RSV_DATA
)
74 if (type
== BTRFS_QGROUP_RSV_META_PERTRANS
)
75 return "meta_pertrans";
76 if (type
== BTRFS_QGROUP_RSV_META_PREALLOC
)
77 return "meta_prealloc";
82 static void qgroup_rsv_add(struct btrfs_fs_info
*fs_info
,
83 struct btrfs_qgroup
*qgroup
, u64 num_bytes
,
84 enum btrfs_qgroup_rsv_type type
)
86 trace_qgroup_update_reserve(fs_info
, qgroup
, num_bytes
, type
);
87 qgroup
->rsv
.values
[type
] += num_bytes
;
90 static void qgroup_rsv_release(struct btrfs_fs_info
*fs_info
,
91 struct btrfs_qgroup
*qgroup
, u64 num_bytes
,
92 enum btrfs_qgroup_rsv_type type
)
94 trace_qgroup_update_reserve(fs_info
, qgroup
, -(s64
)num_bytes
, type
);
95 if (qgroup
->rsv
.values
[type
] >= num_bytes
) {
96 qgroup
->rsv
.values
[type
] -= num_bytes
;
99 #ifdef CONFIG_BTRFS_DEBUG
101 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
102 qgroup
->qgroupid
, qgroup_rsv_type_str(type
),
103 qgroup
->rsv
.values
[type
], num_bytes
);
105 qgroup
->rsv
.values
[type
] = 0;
108 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info
*fs_info
,
109 struct btrfs_qgroup
*dest
,
110 const struct btrfs_qgroup
*src
)
114 for (i
= 0; i
< BTRFS_QGROUP_RSV_LAST
; i
++)
115 qgroup_rsv_add(fs_info
, dest
, src
->rsv
.values
[i
], i
);
118 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info
*fs_info
,
119 struct btrfs_qgroup
*dest
,
120 const struct btrfs_qgroup
*src
)
124 for (i
= 0; i
< BTRFS_QGROUP_RSV_LAST
; i
++)
125 qgroup_rsv_release(fs_info
, dest
, src
->rsv
.values
[i
], i
);
128 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup
*qg
, u64 seq
,
131 if (qg
->old_refcnt
< seq
)
132 qg
->old_refcnt
= seq
;
133 qg
->old_refcnt
+= mod
;
136 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup
*qg
, u64 seq
,
139 if (qg
->new_refcnt
< seq
)
140 qg
->new_refcnt
= seq
;
141 qg
->new_refcnt
+= mod
;
144 static inline u64
btrfs_qgroup_get_old_refcnt(const struct btrfs_qgroup
*qg
, u64 seq
)
146 if (qg
->old_refcnt
< seq
)
148 return qg
->old_refcnt
- seq
;
151 static inline u64
btrfs_qgroup_get_new_refcnt(const struct btrfs_qgroup
*qg
, u64 seq
)
153 if (qg
->new_refcnt
< seq
)
155 return qg
->new_refcnt
- seq
;
159 qgroup_rescan_init(struct btrfs_fs_info
*fs_info
, u64 progress_objectid
,
161 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info
*fs_info
);
163 /* must be called with qgroup_ioctl_lock held */
164 static struct btrfs_qgroup
*find_qgroup_rb(const struct btrfs_fs_info
*fs_info
,
167 struct rb_node
*n
= fs_info
->qgroup_tree
.rb_node
;
168 struct btrfs_qgroup
*qgroup
;
171 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
172 if (qgroup
->qgroupid
< qgroupid
)
174 else if (qgroup
->qgroupid
> qgroupid
)
183 * Add qgroup to the filesystem's qgroup tree.
185 * Must be called with qgroup_lock held and @prealloc preallocated.
187 * The control on the lifespan of @prealloc would be transferred to this
188 * function, thus caller should no longer touch @prealloc.
190 static struct btrfs_qgroup
*add_qgroup_rb(struct btrfs_fs_info
*fs_info
,
191 struct btrfs_qgroup
*prealloc
,
194 struct rb_node
**p
= &fs_info
->qgroup_tree
.rb_node
;
195 struct rb_node
*parent
= NULL
;
196 struct btrfs_qgroup
*qgroup
;
198 /* Caller must have pre-allocated @prealloc. */
203 qgroup
= rb_entry(parent
, struct btrfs_qgroup
, node
);
205 if (qgroup
->qgroupid
< qgroupid
) {
207 } else if (qgroup
->qgroupid
> qgroupid
) {
216 qgroup
->qgroupid
= qgroupid
;
217 INIT_LIST_HEAD(&qgroup
->groups
);
218 INIT_LIST_HEAD(&qgroup
->members
);
219 INIT_LIST_HEAD(&qgroup
->dirty
);
220 INIT_LIST_HEAD(&qgroup
->iterator
);
221 INIT_LIST_HEAD(&qgroup
->nested_iterator
);
223 rb_link_node(&qgroup
->node
, parent
, p
);
224 rb_insert_color(&qgroup
->node
, &fs_info
->qgroup_tree
);
229 static void __del_qgroup_rb(struct btrfs_fs_info
*fs_info
,
230 struct btrfs_qgroup
*qgroup
)
232 struct btrfs_qgroup_list
*list
;
234 list_del(&qgroup
->dirty
);
235 while (!list_empty(&qgroup
->groups
)) {
236 list
= list_first_entry(&qgroup
->groups
,
237 struct btrfs_qgroup_list
, next_group
);
238 list_del(&list
->next_group
);
239 list_del(&list
->next_member
);
243 while (!list_empty(&qgroup
->members
)) {
244 list
= list_first_entry(&qgroup
->members
,
245 struct btrfs_qgroup_list
, next_member
);
246 list_del(&list
->next_group
);
247 list_del(&list
->next_member
);
252 /* must be called with qgroup_lock held */
253 static int del_qgroup_rb(struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
255 struct btrfs_qgroup
*qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
260 rb_erase(&qgroup
->node
, &fs_info
->qgroup_tree
);
261 __del_qgroup_rb(fs_info
, qgroup
);
266 * Add relation specified by two qgroups.
268 * Must be called with qgroup_lock held, the ownership of @prealloc is
269 * transferred to this function and caller should not touch it anymore.
271 * Return: 0 on success
272 * -ENOENT if one of the qgroups is NULL
275 static int __add_relation_rb(struct btrfs_qgroup_list
*prealloc
,
276 struct btrfs_qgroup
*member
,
277 struct btrfs_qgroup
*parent
)
279 if (!member
|| !parent
) {
284 prealloc
->group
= parent
;
285 prealloc
->member
= member
;
286 list_add_tail(&prealloc
->next_group
, &member
->groups
);
287 list_add_tail(&prealloc
->next_member
, &parent
->members
);
293 * Add relation specified by two qgroup ids.
295 * Must be called with qgroup_lock held.
297 * Return: 0 on success
298 * -ENOENT if one of the ids does not exist
301 static int add_relation_rb(struct btrfs_fs_info
*fs_info
,
302 struct btrfs_qgroup_list
*prealloc
,
303 u64 memberid
, u64 parentid
)
305 struct btrfs_qgroup
*member
;
306 struct btrfs_qgroup
*parent
;
308 member
= find_qgroup_rb(fs_info
, memberid
);
309 parent
= find_qgroup_rb(fs_info
, parentid
);
311 return __add_relation_rb(prealloc
, member
, parent
);
314 /* Must be called with qgroup_lock held */
315 static int del_relation_rb(struct btrfs_fs_info
*fs_info
,
316 u64 memberid
, u64 parentid
)
318 struct btrfs_qgroup
*member
;
319 struct btrfs_qgroup
*parent
;
320 struct btrfs_qgroup_list
*list
;
322 member
= find_qgroup_rb(fs_info
, memberid
);
323 parent
= find_qgroup_rb(fs_info
, parentid
);
324 if (!member
|| !parent
)
327 list_for_each_entry(list
, &member
->groups
, next_group
) {
328 if (list
->group
== parent
) {
329 list_del(&list
->next_group
);
330 list_del(&list
->next_member
);
338 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
339 int btrfs_verify_qgroup_counts(const struct btrfs_fs_info
*fs_info
, u64 qgroupid
,
342 struct btrfs_qgroup
*qgroup
;
344 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
347 if (qgroup
->rfer
!= rfer
|| qgroup
->excl
!= excl
)
353 static void qgroup_mark_inconsistent(struct btrfs_fs_info
*fs_info
)
355 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_SIMPLE
)
357 fs_info
->qgroup_flags
|= (BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
|
358 BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN
|
359 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING
);
362 static void qgroup_read_enable_gen(struct btrfs_fs_info
*fs_info
,
363 struct extent_buffer
*leaf
, int slot
,
364 struct btrfs_qgroup_status_item
*ptr
)
366 ASSERT(btrfs_fs_incompat(fs_info
, SIMPLE_QUOTA
));
367 ASSERT(btrfs_item_size(leaf
, slot
) >= sizeof(*ptr
));
368 fs_info
->qgroup_enable_gen
= btrfs_qgroup_status_enable_gen(leaf
, ptr
);
372 * The full config is read in one go, only called from open_ctree()
373 * It doesn't use any locking, as at this point we're still single-threaded
375 int btrfs_read_qgroup_config(struct btrfs_fs_info
*fs_info
)
377 struct btrfs_key key
;
378 struct btrfs_key found_key
;
379 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
380 struct btrfs_path
*path
= NULL
;
381 struct extent_buffer
*l
;
385 u64 rescan_progress
= 0;
387 if (!fs_info
->quota_root
)
390 fs_info
->qgroup_ulist
= ulist_alloc(GFP_KERNEL
);
391 if (!fs_info
->qgroup_ulist
) {
396 path
= btrfs_alloc_path();
402 ret
= btrfs_sysfs_add_qgroups(fs_info
);
405 /* default this to quota off, in case no status key is found */
406 fs_info
->qgroup_flags
= 0;
409 * pass 1: read status, all qgroup infos and limits
414 ret
= btrfs_search_slot_for_read(quota_root
, &key
, path
, 1, 1);
419 struct btrfs_qgroup
*qgroup
;
421 slot
= path
->slots
[0];
423 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
425 if (found_key
.type
== BTRFS_QGROUP_STATUS_KEY
) {
426 struct btrfs_qgroup_status_item
*ptr
;
428 ptr
= btrfs_item_ptr(l
, slot
,
429 struct btrfs_qgroup_status_item
);
431 if (btrfs_qgroup_status_version(l
, ptr
) !=
432 BTRFS_QGROUP_STATUS_VERSION
) {
434 "old qgroup version, quota disabled");
437 fs_info
->qgroup_flags
= btrfs_qgroup_status_flags(l
, ptr
);
438 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_SIMPLE_MODE
) {
439 qgroup_read_enable_gen(fs_info
, l
, slot
, ptr
);
440 } else if (btrfs_qgroup_status_generation(l
, ptr
) != fs_info
->generation
) {
441 qgroup_mark_inconsistent(fs_info
);
443 "qgroup generation mismatch, marked as inconsistent");
445 rescan_progress
= btrfs_qgroup_status_rescan(l
, ptr
);
449 if (found_key
.type
!= BTRFS_QGROUP_INFO_KEY
&&
450 found_key
.type
!= BTRFS_QGROUP_LIMIT_KEY
)
453 qgroup
= find_qgroup_rb(fs_info
, found_key
.offset
);
454 if ((qgroup
&& found_key
.type
== BTRFS_QGROUP_INFO_KEY
) ||
455 (!qgroup
&& found_key
.type
== BTRFS_QGROUP_LIMIT_KEY
)) {
456 btrfs_err(fs_info
, "inconsistent qgroup config");
457 qgroup_mark_inconsistent(fs_info
);
460 struct btrfs_qgroup
*prealloc
;
461 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
463 prealloc
= kzalloc(sizeof(*prealloc
), GFP_KERNEL
);
468 qgroup
= add_qgroup_rb(fs_info
, prealloc
, found_key
.offset
);
470 * If a qgroup exists for a subvolume ID, it is possible
471 * that subvolume has been deleted, in which case
472 * re-using that ID would lead to incorrect accounting.
474 * Ensure that we skip any such subvol ids.
476 * We don't need to lock because this is only called
477 * during mount before we start doing things like creating
480 if (is_fstree(qgroup
->qgroupid
) &&
481 qgroup
->qgroupid
> tree_root
->free_objectid
)
483 * Don't need to check against BTRFS_LAST_FREE_OBJECTID,
484 * as it will get checked on the next call to
485 * btrfs_get_free_objectid.
487 tree_root
->free_objectid
= qgroup
->qgroupid
+ 1;
489 ret
= btrfs_sysfs_add_one_qgroup(fs_info
, qgroup
);
493 switch (found_key
.type
) {
494 case BTRFS_QGROUP_INFO_KEY
: {
495 struct btrfs_qgroup_info_item
*ptr
;
497 ptr
= btrfs_item_ptr(l
, slot
,
498 struct btrfs_qgroup_info_item
);
499 qgroup
->rfer
= btrfs_qgroup_info_rfer(l
, ptr
);
500 qgroup
->rfer_cmpr
= btrfs_qgroup_info_rfer_cmpr(l
, ptr
);
501 qgroup
->excl
= btrfs_qgroup_info_excl(l
, ptr
);
502 qgroup
->excl_cmpr
= btrfs_qgroup_info_excl_cmpr(l
, ptr
);
503 /* generation currently unused */
506 case BTRFS_QGROUP_LIMIT_KEY
: {
507 struct btrfs_qgroup_limit_item
*ptr
;
509 ptr
= btrfs_item_ptr(l
, slot
,
510 struct btrfs_qgroup_limit_item
);
511 qgroup
->lim_flags
= btrfs_qgroup_limit_flags(l
, ptr
);
512 qgroup
->max_rfer
= btrfs_qgroup_limit_max_rfer(l
, ptr
);
513 qgroup
->max_excl
= btrfs_qgroup_limit_max_excl(l
, ptr
);
514 qgroup
->rsv_rfer
= btrfs_qgroup_limit_rsv_rfer(l
, ptr
);
515 qgroup
->rsv_excl
= btrfs_qgroup_limit_rsv_excl(l
, ptr
);
520 ret
= btrfs_next_item(quota_root
, path
);
526 btrfs_release_path(path
);
529 * pass 2: read all qgroup relations
532 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
534 ret
= btrfs_search_slot_for_read(quota_root
, &key
, path
, 1, 0);
538 struct btrfs_qgroup_list
*list
= NULL
;
540 slot
= path
->slots
[0];
542 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
544 if (found_key
.type
!= BTRFS_QGROUP_RELATION_KEY
)
547 if (found_key
.objectid
> found_key
.offset
) {
548 /* parent <- member, not needed to build config */
549 /* FIXME should we omit the key completely? */
553 list
= kzalloc(sizeof(*list
), GFP_KERNEL
);
558 ret
= add_relation_rb(fs_info
, list
, found_key
.objectid
,
561 if (ret
== -ENOENT
) {
563 "orphan qgroup relation 0x%llx->0x%llx",
564 found_key
.objectid
, found_key
.offset
);
565 ret
= 0; /* ignore the error */
570 ret
= btrfs_next_item(quota_root
, path
);
577 btrfs_free_path(path
);
578 fs_info
->qgroup_flags
|= flags
;
580 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
)
581 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
582 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
)
583 ret
= qgroup_rescan_init(fs_info
, rescan_progress
, 0);
585 ulist_free(fs_info
->qgroup_ulist
);
586 fs_info
->qgroup_ulist
= NULL
;
587 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
588 btrfs_sysfs_del_qgroups(fs_info
);
591 return ret
< 0 ? ret
: 0;
595 * Called in close_ctree() when quota is still enabled. This verifies we don't
596 * leak some reserved space.
598 * Return false if no reserved space is left.
599 * Return true if some reserved space is leaked.
601 bool btrfs_check_quota_leak(const struct btrfs_fs_info
*fs_info
)
603 struct rb_node
*node
;
606 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_DISABLED
)
609 * Since we're unmounting, there is no race and no need to grab qgroup
610 * lock. And here we don't go post-order to provide a more user
611 * friendly sorted result.
613 for (node
= rb_first(&fs_info
->qgroup_tree
); node
; node
= rb_next(node
)) {
614 struct btrfs_qgroup
*qgroup
;
617 qgroup
= rb_entry(node
, struct btrfs_qgroup
, node
);
618 for (i
= 0; i
< BTRFS_QGROUP_RSV_LAST
; i
++) {
619 if (qgroup
->rsv
.values
[i
]) {
622 "qgroup %hu/%llu has unreleased space, type %d rsv %llu",
623 btrfs_qgroup_level(qgroup
->qgroupid
),
624 btrfs_qgroup_subvolid(qgroup
->qgroupid
),
625 i
, qgroup
->rsv
.values
[i
]);
633 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
634 * first two are in single-threaded paths.And for the third one, we have set
635 * quota_root to be null with qgroup_lock held before, so it is safe to clean
636 * up the in-memory structures without qgroup_lock held.
638 void btrfs_free_qgroup_config(struct btrfs_fs_info
*fs_info
)
641 struct btrfs_qgroup
*qgroup
;
643 while ((n
= rb_first(&fs_info
->qgroup_tree
))) {
644 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
645 rb_erase(n
, &fs_info
->qgroup_tree
);
646 __del_qgroup_rb(fs_info
, qgroup
);
647 btrfs_sysfs_del_one_qgroup(fs_info
, qgroup
);
651 * We call btrfs_free_qgroup_config() when unmounting
652 * filesystem and disabling quota, so we set qgroup_ulist
653 * to be null here to avoid double free.
655 ulist_free(fs_info
->qgroup_ulist
);
656 fs_info
->qgroup_ulist
= NULL
;
657 btrfs_sysfs_del_qgroups(fs_info
);
660 static int add_qgroup_relation_item(struct btrfs_trans_handle
*trans
, u64 src
,
664 struct btrfs_root
*quota_root
= trans
->fs_info
->quota_root
;
665 struct btrfs_path
*path
;
666 struct btrfs_key key
;
668 path
= btrfs_alloc_path();
673 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
676 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
, 0);
678 btrfs_mark_buffer_dirty(trans
, path
->nodes
[0]);
680 btrfs_free_path(path
);
684 static int del_qgroup_relation_item(struct btrfs_trans_handle
*trans
, u64 src
,
688 struct btrfs_root
*quota_root
= trans
->fs_info
->quota_root
;
689 struct btrfs_path
*path
;
690 struct btrfs_key key
;
692 path
= btrfs_alloc_path();
697 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
700 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
709 ret
= btrfs_del_item(trans
, quota_root
, path
);
711 btrfs_free_path(path
);
715 static int add_qgroup_item(struct btrfs_trans_handle
*trans
,
716 struct btrfs_root
*quota_root
, u64 qgroupid
)
719 struct btrfs_path
*path
;
720 struct btrfs_qgroup_info_item
*qgroup_info
;
721 struct btrfs_qgroup_limit_item
*qgroup_limit
;
722 struct extent_buffer
*leaf
;
723 struct btrfs_key key
;
725 if (btrfs_is_testing(quota_root
->fs_info
))
728 path
= btrfs_alloc_path();
733 key
.type
= BTRFS_QGROUP_INFO_KEY
;
734 key
.offset
= qgroupid
;
737 * Avoid a transaction abort by catching -EEXIST here. In that
738 * case, we proceed by re-initializing the existing structure
742 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
743 sizeof(*qgroup_info
));
744 if (ret
&& ret
!= -EEXIST
)
747 leaf
= path
->nodes
[0];
748 qgroup_info
= btrfs_item_ptr(leaf
, path
->slots
[0],
749 struct btrfs_qgroup_info_item
);
750 btrfs_set_qgroup_info_generation(leaf
, qgroup_info
, trans
->transid
);
751 btrfs_set_qgroup_info_rfer(leaf
, qgroup_info
, 0);
752 btrfs_set_qgroup_info_rfer_cmpr(leaf
, qgroup_info
, 0);
753 btrfs_set_qgroup_info_excl(leaf
, qgroup_info
, 0);
754 btrfs_set_qgroup_info_excl_cmpr(leaf
, qgroup_info
, 0);
756 btrfs_mark_buffer_dirty(trans
, leaf
);
758 btrfs_release_path(path
);
760 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
761 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
762 sizeof(*qgroup_limit
));
763 if (ret
&& ret
!= -EEXIST
)
766 leaf
= path
->nodes
[0];
767 qgroup_limit
= btrfs_item_ptr(leaf
, path
->slots
[0],
768 struct btrfs_qgroup_limit_item
);
769 btrfs_set_qgroup_limit_flags(leaf
, qgroup_limit
, 0);
770 btrfs_set_qgroup_limit_max_rfer(leaf
, qgroup_limit
, 0);
771 btrfs_set_qgroup_limit_max_excl(leaf
, qgroup_limit
, 0);
772 btrfs_set_qgroup_limit_rsv_rfer(leaf
, qgroup_limit
, 0);
773 btrfs_set_qgroup_limit_rsv_excl(leaf
, qgroup_limit
, 0);
775 btrfs_mark_buffer_dirty(trans
, leaf
);
779 btrfs_free_path(path
);
783 static int del_qgroup_item(struct btrfs_trans_handle
*trans
, u64 qgroupid
)
786 struct btrfs_root
*quota_root
= trans
->fs_info
->quota_root
;
787 struct btrfs_path
*path
;
788 struct btrfs_key key
;
790 path
= btrfs_alloc_path();
795 key
.type
= BTRFS_QGROUP_INFO_KEY
;
796 key
.offset
= qgroupid
;
797 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
806 ret
= btrfs_del_item(trans
, quota_root
, path
);
810 btrfs_release_path(path
);
812 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
813 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
822 ret
= btrfs_del_item(trans
, quota_root
, path
);
825 btrfs_free_path(path
);
829 static int update_qgroup_limit_item(struct btrfs_trans_handle
*trans
,
830 struct btrfs_qgroup
*qgroup
)
832 struct btrfs_root
*quota_root
= trans
->fs_info
->quota_root
;
833 struct btrfs_path
*path
;
834 struct btrfs_key key
;
835 struct extent_buffer
*l
;
836 struct btrfs_qgroup_limit_item
*qgroup_limit
;
841 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
842 key
.offset
= qgroup
->qgroupid
;
844 path
= btrfs_alloc_path();
848 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, 0, 1);
856 slot
= path
->slots
[0];
857 qgroup_limit
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_limit_item
);
858 btrfs_set_qgroup_limit_flags(l
, qgroup_limit
, qgroup
->lim_flags
);
859 btrfs_set_qgroup_limit_max_rfer(l
, qgroup_limit
, qgroup
->max_rfer
);
860 btrfs_set_qgroup_limit_max_excl(l
, qgroup_limit
, qgroup
->max_excl
);
861 btrfs_set_qgroup_limit_rsv_rfer(l
, qgroup_limit
, qgroup
->rsv_rfer
);
862 btrfs_set_qgroup_limit_rsv_excl(l
, qgroup_limit
, qgroup
->rsv_excl
);
864 btrfs_mark_buffer_dirty(trans
, l
);
867 btrfs_free_path(path
);
871 static int update_qgroup_info_item(struct btrfs_trans_handle
*trans
,
872 struct btrfs_qgroup
*qgroup
)
874 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
875 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
876 struct btrfs_path
*path
;
877 struct btrfs_key key
;
878 struct extent_buffer
*l
;
879 struct btrfs_qgroup_info_item
*qgroup_info
;
883 if (btrfs_is_testing(fs_info
))
887 key
.type
= BTRFS_QGROUP_INFO_KEY
;
888 key
.offset
= qgroup
->qgroupid
;
890 path
= btrfs_alloc_path();
894 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, 0, 1);
902 slot
= path
->slots
[0];
903 qgroup_info
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_info_item
);
904 btrfs_set_qgroup_info_generation(l
, qgroup_info
, trans
->transid
);
905 btrfs_set_qgroup_info_rfer(l
, qgroup_info
, qgroup
->rfer
);
906 btrfs_set_qgroup_info_rfer_cmpr(l
, qgroup_info
, qgroup
->rfer_cmpr
);
907 btrfs_set_qgroup_info_excl(l
, qgroup_info
, qgroup
->excl
);
908 btrfs_set_qgroup_info_excl_cmpr(l
, qgroup_info
, qgroup
->excl_cmpr
);
910 btrfs_mark_buffer_dirty(trans
, l
);
913 btrfs_free_path(path
);
917 static int update_qgroup_status_item(struct btrfs_trans_handle
*trans
)
919 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
920 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
921 struct btrfs_path
*path
;
922 struct btrfs_key key
;
923 struct extent_buffer
*l
;
924 struct btrfs_qgroup_status_item
*ptr
;
929 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
932 path
= btrfs_alloc_path();
936 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, 0, 1);
944 slot
= path
->slots
[0];
945 ptr
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_status_item
);
946 btrfs_set_qgroup_status_flags(l
, ptr
, fs_info
->qgroup_flags
&
947 BTRFS_QGROUP_STATUS_FLAGS_MASK
);
948 btrfs_set_qgroup_status_generation(l
, ptr
, trans
->transid
);
949 btrfs_set_qgroup_status_rescan(l
, ptr
,
950 fs_info
->qgroup_rescan_progress
.objectid
);
952 btrfs_mark_buffer_dirty(trans
, l
);
955 btrfs_free_path(path
);
960 * called with qgroup_lock held
962 static int btrfs_clean_quota_tree(struct btrfs_trans_handle
*trans
,
963 struct btrfs_root
*root
)
965 struct btrfs_path
*path
;
966 struct btrfs_key key
;
967 struct extent_buffer
*leaf
= NULL
;
971 path
= btrfs_alloc_path();
980 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
983 leaf
= path
->nodes
[0];
984 nr
= btrfs_header_nritems(leaf
);
988 * delete the leaf one by one
989 * since the whole tree is going
993 ret
= btrfs_del_items(trans
, root
, path
, 0, nr
);
997 btrfs_release_path(path
);
1001 btrfs_free_path(path
);
1005 int btrfs_quota_enable(struct btrfs_fs_info
*fs_info
,
1006 struct btrfs_ioctl_quota_ctl_args
*quota_ctl_args
)
1008 struct btrfs_root
*quota_root
;
1009 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1010 struct btrfs_path
*path
= NULL
;
1011 struct btrfs_qgroup_status_item
*ptr
;
1012 struct extent_buffer
*leaf
;
1013 struct btrfs_key key
;
1014 struct btrfs_key found_key
;
1015 struct btrfs_qgroup
*qgroup
= NULL
;
1016 struct btrfs_qgroup
*prealloc
= NULL
;
1017 struct btrfs_trans_handle
*trans
= NULL
;
1018 struct ulist
*ulist
= NULL
;
1019 const bool simple
= (quota_ctl_args
->cmd
== BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA
);
1024 * We need to have subvol_sem write locked, to prevent races between
1025 * concurrent tasks trying to enable quotas, because we will unlock
1026 * and relock qgroup_ioctl_lock before setting fs_info->quota_root
1027 * and before setting BTRFS_FS_QUOTA_ENABLED.
1029 lockdep_assert_held_write(&fs_info
->subvol_sem
);
1031 if (btrfs_fs_incompat(fs_info
, EXTENT_TREE_V2
)) {
1033 "qgroups are currently unsupported in extent tree v2");
1037 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1038 if (fs_info
->quota_root
)
1041 ulist
= ulist_alloc(GFP_KERNEL
);
1047 ret
= btrfs_sysfs_add_qgroups(fs_info
);
1052 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
1053 * avoid lock acquisition inversion problems (reported by lockdep) between
1054 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
1055 * start a transaction.
1056 * After we started the transaction lock qgroup_ioctl_lock again and
1057 * check if someone else created the quota root in the meanwhile. If so,
1058 * just return success and release the transaction handle.
1060 * Also we don't need to worry about someone else calling
1061 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
1062 * that function returns 0 (success) when the sysfs entries already exist.
1064 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1067 * 1 for quota root item
1068 * 1 for BTRFS_QGROUP_STATUS item
1070 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
1071 * per subvolume. However those are not currently reserved since it
1072 * would be a lot of overkill.
1074 trans
= btrfs_start_transaction(tree_root
, 2);
1076 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1077 if (IS_ERR(trans
)) {
1078 ret
= PTR_ERR(trans
);
1083 if (fs_info
->quota_root
)
1086 fs_info
->qgroup_ulist
= ulist
;
1090 * initially create the quota tree
1092 quota_root
= btrfs_create_tree(trans
, BTRFS_QUOTA_TREE_OBJECTID
);
1093 if (IS_ERR(quota_root
)) {
1094 ret
= PTR_ERR(quota_root
);
1095 btrfs_abort_transaction(trans
, ret
);
1099 path
= btrfs_alloc_path();
1102 btrfs_abort_transaction(trans
, ret
);
1107 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
1110 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
1113 btrfs_abort_transaction(trans
, ret
);
1117 leaf
= path
->nodes
[0];
1118 ptr
= btrfs_item_ptr(leaf
, path
->slots
[0],
1119 struct btrfs_qgroup_status_item
);
1120 btrfs_set_qgroup_status_generation(leaf
, ptr
, trans
->transid
);
1121 btrfs_set_qgroup_status_version(leaf
, ptr
, BTRFS_QGROUP_STATUS_VERSION
);
1122 fs_info
->qgroup_flags
= BTRFS_QGROUP_STATUS_FLAG_ON
;
1124 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_SIMPLE_MODE
;
1125 btrfs_set_qgroup_status_enable_gen(leaf
, ptr
, trans
->transid
);
1127 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1129 btrfs_set_qgroup_status_flags(leaf
, ptr
, fs_info
->qgroup_flags
&
1130 BTRFS_QGROUP_STATUS_FLAGS_MASK
);
1131 btrfs_set_qgroup_status_rescan(leaf
, ptr
, 0);
1133 btrfs_mark_buffer_dirty(trans
, leaf
);
1136 key
.type
= BTRFS_ROOT_REF_KEY
;
1139 btrfs_release_path(path
);
1140 ret
= btrfs_search_slot_for_read(tree_root
, &key
, path
, 1, 0);
1144 btrfs_abort_transaction(trans
, ret
);
1149 slot
= path
->slots
[0];
1150 leaf
= path
->nodes
[0];
1151 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
1153 if (found_key
.type
== BTRFS_ROOT_REF_KEY
) {
1155 /* Release locks on tree_root before we access quota_root */
1156 btrfs_release_path(path
);
1158 /* We should not have a stray @prealloc pointer. */
1159 ASSERT(prealloc
== NULL
);
1160 prealloc
= kzalloc(sizeof(*prealloc
), GFP_NOFS
);
1163 btrfs_abort_transaction(trans
, ret
);
1167 ret
= add_qgroup_item(trans
, quota_root
,
1170 btrfs_abort_transaction(trans
, ret
);
1174 qgroup
= add_qgroup_rb(fs_info
, prealloc
, found_key
.offset
);
1176 if (IS_ERR(qgroup
)) {
1177 ret
= PTR_ERR(qgroup
);
1178 btrfs_abort_transaction(trans
, ret
);
1181 ret
= btrfs_sysfs_add_one_qgroup(fs_info
, qgroup
);
1183 btrfs_abort_transaction(trans
, ret
);
1186 ret
= btrfs_search_slot_for_read(tree_root
, &found_key
,
1189 btrfs_abort_transaction(trans
, ret
);
1194 * Shouldn't happen, but in case it does we
1195 * don't need to do the btrfs_next_item, just
1201 ret
= btrfs_next_item(tree_root
, path
);
1203 btrfs_abort_transaction(trans
, ret
);
1211 btrfs_release_path(path
);
1212 ret
= add_qgroup_item(trans
, quota_root
, BTRFS_FS_TREE_OBJECTID
);
1214 btrfs_abort_transaction(trans
, ret
);
1218 ASSERT(prealloc
== NULL
);
1219 prealloc
= kzalloc(sizeof(*prealloc
), GFP_NOFS
);
1224 qgroup
= add_qgroup_rb(fs_info
, prealloc
, BTRFS_FS_TREE_OBJECTID
);
1226 ret
= btrfs_sysfs_add_one_qgroup(fs_info
, qgroup
);
1228 btrfs_abort_transaction(trans
, ret
);
1232 fs_info
->qgroup_enable_gen
= trans
->transid
;
1234 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1236 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
1237 * a deadlock with tasks concurrently doing other qgroup operations, such
1238 * adding/removing qgroups or adding/deleting qgroup relations for example,
1239 * because all qgroup operations first start or join a transaction and then
1240 * lock the qgroup_ioctl_lock mutex.
1241 * We are safe from a concurrent task trying to enable quotas, by calling
1242 * this function, since we are serialized by fs_info->subvol_sem.
1244 ret
= btrfs_commit_transaction(trans
);
1246 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1251 * Set quota enabled flag after committing the transaction, to avoid
1252 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1255 spin_lock(&fs_info
->qgroup_lock
);
1256 fs_info
->quota_root
= quota_root
;
1257 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
1259 btrfs_set_fs_incompat(fs_info
, SIMPLE_QUOTA
);
1260 spin_unlock(&fs_info
->qgroup_lock
);
1262 /* Skip rescan for simple qgroups. */
1263 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_SIMPLE
)
1266 ret
= qgroup_rescan_init(fs_info
, 0, 1);
1268 qgroup_rescan_zero_tracking(fs_info
);
1269 fs_info
->qgroup_rescan_running
= true;
1270 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
1271 &fs_info
->qgroup_rescan_work
);
1274 * We have set both BTRFS_FS_QUOTA_ENABLED and
1275 * BTRFS_QGROUP_STATUS_FLAG_ON, so we can only fail with
1276 * -EINPROGRESS. That can happen because someone started the
1277 * rescan worker by calling quota rescan ioctl before we
1278 * attempted to initialize the rescan worker. Failure due to
1279 * quotas disabled in the meanwhile is not possible, because
1280 * we are holding a write lock on fs_info->subvol_sem, which
1281 * is also acquired when disabling quotas.
1282 * Ignore such error, and any other error would need to undo
1283 * everything we did in the transaction we just committed.
1285 ASSERT(ret
== -EINPROGRESS
);
1290 btrfs_free_path(path
);
1293 btrfs_put_root(quota_root
);
1296 ulist_free(fs_info
->qgroup_ulist
);
1297 fs_info
->qgroup_ulist
= NULL
;
1298 btrfs_sysfs_del_qgroups(fs_info
);
1300 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1302 btrfs_end_transaction(trans
);
1304 ret
= btrfs_end_transaction(trans
);
1311 * It is possible to have outstanding ordered extents which reserved bytes
1312 * before we disabled. We need to fully flush delalloc, ordered extents, and a
1313 * commit to ensure that we don't leak such reservations, only to have them
1314 * come back if we re-enable.
1316 * - enable simple quotas
1318 * - release it, store rsv_bytes in OE
1320 * - enable simple quotas (qgroup rsv are all 0)
1322 * - run delayed refs
1323 * - free rsv_bytes, resulting in miscounting or even underflow
1325 static int flush_reservations(struct btrfs_fs_info
*fs_info
)
1329 ret
= btrfs_start_delalloc_roots(fs_info
, LONG_MAX
, false);
1332 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, NULL
);
1334 return btrfs_commit_current_transaction(fs_info
->tree_root
);
1337 int btrfs_quota_disable(struct btrfs_fs_info
*fs_info
)
1339 struct btrfs_root
*quota_root
= NULL
;
1340 struct btrfs_trans_handle
*trans
= NULL
;
1344 * We need to have subvol_sem write locked to prevent races with
1345 * snapshot creation.
1347 lockdep_assert_held_write(&fs_info
->subvol_sem
);
1350 * Relocation will mess with backrefs, so make sure we have the
1351 * cleaner_mutex held to protect us from relocate.
1353 lockdep_assert_held(&fs_info
->cleaner_mutex
);
1355 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1356 if (!fs_info
->quota_root
)
1360 * Unlock the qgroup_ioctl_lock mutex before waiting for the rescan worker to
1361 * complete. Otherwise we can deadlock because btrfs_remove_qgroup() needs
1362 * to lock that mutex while holding a transaction handle and the rescan
1363 * worker needs to commit a transaction.
1365 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1368 * Request qgroup rescan worker to complete and wait for it. This wait
1369 * must be done before transaction start for quota disable since it may
1370 * deadlock with transaction by the qgroup rescan worker.
1372 clear_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
1373 btrfs_qgroup_wait_for_completion(fs_info
, false);
1376 * We have nothing held here and no trans handle, just return the error
1379 ret
= flush_reservations(fs_info
);
1384 * 1 For the root item
1386 * We should also reserve enough items for the quota tree deletion in
1387 * btrfs_clean_quota_tree but this is not done.
1389 * Also, we must always start a transaction without holding the mutex
1390 * qgroup_ioctl_lock, see btrfs_quota_enable().
1392 trans
= btrfs_start_transaction(fs_info
->tree_root
, 1);
1394 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1395 if (IS_ERR(trans
)) {
1396 ret
= PTR_ERR(trans
);
1398 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
1402 if (!fs_info
->quota_root
)
1405 spin_lock(&fs_info
->qgroup_lock
);
1406 quota_root
= fs_info
->quota_root
;
1407 fs_info
->quota_root
= NULL
;
1408 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_ON
;
1409 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_SIMPLE_MODE
;
1410 fs_info
->qgroup_drop_subtree_thres
= BTRFS_MAX_LEVEL
;
1411 spin_unlock(&fs_info
->qgroup_lock
);
1413 btrfs_free_qgroup_config(fs_info
);
1415 ret
= btrfs_clean_quota_tree(trans
, quota_root
);
1417 btrfs_abort_transaction(trans
, ret
);
1421 ret
= btrfs_del_root(trans
, "a_root
->root_key
);
1423 btrfs_abort_transaction(trans
, ret
);
1427 spin_lock(&fs_info
->trans_lock
);
1428 list_del("a_root
->dirty_list
);
1429 spin_unlock(&fs_info
->trans_lock
);
1431 btrfs_tree_lock(quota_root
->node
);
1432 btrfs_clear_buffer_dirty(trans
, quota_root
->node
);
1433 btrfs_tree_unlock(quota_root
->node
);
1434 ret
= btrfs_free_tree_block(trans
, btrfs_root_id(quota_root
),
1435 quota_root
->node
, 0, 1);
1438 btrfs_abort_transaction(trans
, ret
);
1441 btrfs_put_root(quota_root
);
1442 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1444 btrfs_end_transaction(trans
);
1446 ret
= btrfs_commit_transaction(trans
);
1450 static void qgroup_dirty(struct btrfs_fs_info
*fs_info
,
1451 struct btrfs_qgroup
*qgroup
)
1453 if (list_empty(&qgroup
->dirty
))
1454 list_add(&qgroup
->dirty
, &fs_info
->dirty_qgroups
);
1457 static void qgroup_iterator_add(struct list_head
*head
, struct btrfs_qgroup
*qgroup
)
1459 if (!list_empty(&qgroup
->iterator
))
1462 list_add_tail(&qgroup
->iterator
, head
);
1465 static void qgroup_iterator_clean(struct list_head
*head
)
1467 while (!list_empty(head
)) {
1468 struct btrfs_qgroup
*qgroup
;
1470 qgroup
= list_first_entry(head
, struct btrfs_qgroup
, iterator
);
1471 list_del_init(&qgroup
->iterator
);
1476 * The easy accounting, we're updating qgroup relationship whose child qgroup
1477 * only has exclusive extents.
1479 * In this case, all exclusive extents will also be exclusive for parent, so
1480 * excl/rfer just get added/removed.
1482 * So is qgroup reservation space, which should also be added/removed to
1484 * Or when child tries to release reservation space, parent will underflow its
1485 * reservation (for relationship adding case).
1487 * Caller should hold fs_info->qgroup_lock.
1489 static int __qgroup_excl_accounting(struct btrfs_fs_info
*fs_info
, u64 ref_root
,
1490 struct btrfs_qgroup
*src
, int sign
)
1492 struct btrfs_qgroup
*qgroup
;
1493 struct btrfs_qgroup
*cur
;
1494 LIST_HEAD(qgroup_list
);
1495 u64 num_bytes
= src
->excl
;
1498 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
1502 qgroup_iterator_add(&qgroup_list
, qgroup
);
1503 list_for_each_entry(cur
, &qgroup_list
, iterator
) {
1504 struct btrfs_qgroup_list
*glist
;
1506 qgroup
->rfer
+= sign
* num_bytes
;
1507 qgroup
->rfer_cmpr
+= sign
* num_bytes
;
1509 WARN_ON(sign
< 0 && qgroup
->excl
< num_bytes
);
1510 qgroup
->excl
+= sign
* num_bytes
;
1511 qgroup
->excl_cmpr
+= sign
* num_bytes
;
1514 qgroup_rsv_add_by_qgroup(fs_info
, qgroup
, src
);
1516 qgroup_rsv_release_by_qgroup(fs_info
, qgroup
, src
);
1517 qgroup_dirty(fs_info
, qgroup
);
1519 /* Append parent qgroups to @qgroup_list. */
1520 list_for_each_entry(glist
, &qgroup
->groups
, next_group
)
1521 qgroup_iterator_add(&qgroup_list
, glist
->group
);
1525 qgroup_iterator_clean(&qgroup_list
);
1531 * Quick path for updating qgroup with only excl refs.
1533 * In that case, just update all parent will be enough.
1534 * Or we needs to do a full rescan.
1535 * Caller should also hold fs_info->qgroup_lock.
1537 * Return 0 for quick update, return >0 for need to full rescan
1538 * and mark INCONSISTENT flag.
1539 * Return < 0 for other error.
1541 static int quick_update_accounting(struct btrfs_fs_info
*fs_info
,
1542 u64 src
, u64 dst
, int sign
)
1544 struct btrfs_qgroup
*qgroup
;
1547 qgroup
= find_qgroup_rb(fs_info
, src
);
1550 if (qgroup
->excl
== qgroup
->rfer
) {
1551 ret
= __qgroup_excl_accounting(fs_info
, dst
, qgroup
, sign
);
1558 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1563 * Add relation between @src and @dst qgroup. The @prealloc is allocated by the
1564 * callers and transferred here (either used or freed on error).
1566 int btrfs_add_qgroup_relation(struct btrfs_trans_handle
*trans
, u64 src
, u64 dst
,
1567 struct btrfs_qgroup_list
*prealloc
)
1569 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1570 struct btrfs_qgroup
*parent
;
1571 struct btrfs_qgroup
*member
;
1572 struct btrfs_qgroup_list
*list
;
1577 /* Check the level of src and dst first */
1578 if (btrfs_qgroup_level(src
) >= btrfs_qgroup_level(dst
))
1581 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1582 if (!fs_info
->quota_root
) {
1586 member
= find_qgroup_rb(fs_info
, src
);
1587 parent
= find_qgroup_rb(fs_info
, dst
);
1588 if (!member
|| !parent
) {
1593 /* check if such qgroup relation exist firstly */
1594 list_for_each_entry(list
, &member
->groups
, next_group
) {
1595 if (list
->group
== parent
) {
1601 ret
= add_qgroup_relation_item(trans
, src
, dst
);
1605 ret
= add_qgroup_relation_item(trans
, dst
, src
);
1607 del_qgroup_relation_item(trans
, src
, dst
);
1611 spin_lock(&fs_info
->qgroup_lock
);
1612 ret
= __add_relation_rb(prealloc
, member
, parent
);
1615 spin_unlock(&fs_info
->qgroup_lock
);
1618 ret
= quick_update_accounting(fs_info
, src
, dst
, 1);
1619 spin_unlock(&fs_info
->qgroup_lock
);
1622 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1626 static int __del_qgroup_relation(struct btrfs_trans_handle
*trans
, u64 src
,
1629 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1630 struct btrfs_qgroup
*parent
;
1631 struct btrfs_qgroup
*member
;
1632 struct btrfs_qgroup_list
*list
;
1637 if (!fs_info
->quota_root
) {
1642 member
= find_qgroup_rb(fs_info
, src
);
1643 parent
= find_qgroup_rb(fs_info
, dst
);
1645 * The parent/member pair doesn't exist, then try to delete the dead
1646 * relation items only.
1648 if (!member
|| !parent
)
1651 /* check if such qgroup relation exist firstly */
1652 list_for_each_entry(list
, &member
->groups
, next_group
) {
1653 if (list
->group
== parent
) {
1660 ret
= del_qgroup_relation_item(trans
, src
, dst
);
1661 if (ret
< 0 && ret
!= -ENOENT
)
1663 ret2
= del_qgroup_relation_item(trans
, dst
, src
);
1664 if (ret2
< 0 && ret2
!= -ENOENT
)
1667 /* At least one deletion succeeded, return 0 */
1672 spin_lock(&fs_info
->qgroup_lock
);
1673 del_relation_rb(fs_info
, src
, dst
);
1674 ret
= quick_update_accounting(fs_info
, src
, dst
, -1);
1675 spin_unlock(&fs_info
->qgroup_lock
);
1681 int btrfs_del_qgroup_relation(struct btrfs_trans_handle
*trans
, u64 src
,
1684 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1687 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1688 ret
= __del_qgroup_relation(trans
, src
, dst
);
1689 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1694 int btrfs_create_qgroup(struct btrfs_trans_handle
*trans
, u64 qgroupid
)
1696 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1697 struct btrfs_root
*quota_root
;
1698 struct btrfs_qgroup
*qgroup
;
1699 struct btrfs_qgroup
*prealloc
= NULL
;
1702 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_DISABLED
)
1705 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1706 if (!fs_info
->quota_root
) {
1710 quota_root
= fs_info
->quota_root
;
1711 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1717 prealloc
= kzalloc(sizeof(*prealloc
), GFP_NOFS
);
1723 ret
= add_qgroup_item(trans
, quota_root
, qgroupid
);
1727 spin_lock(&fs_info
->qgroup_lock
);
1728 qgroup
= add_qgroup_rb(fs_info
, prealloc
, qgroupid
);
1729 spin_unlock(&fs_info
->qgroup_lock
);
1732 ret
= btrfs_sysfs_add_one_qgroup(fs_info
, qgroup
);
1734 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1740 * Return 0 if we can not delete the qgroup (not empty or has children etc).
1741 * Return >0 if we can delete the qgroup.
1742 * Return <0 for other errors during tree search.
1744 static int can_delete_qgroup(struct btrfs_fs_info
*fs_info
, struct btrfs_qgroup
*qgroup
)
1746 struct btrfs_key key
;
1747 struct btrfs_path
*path
;
1751 * Squota would never be inconsistent, but there can still be case
1752 * where a dropped subvolume still has qgroup numbers, and squota
1753 * relies on such qgroup for future accounting.
1755 * So for squota, do not allow dropping any non-zero qgroup.
1757 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_SIMPLE
&&
1758 (qgroup
->rfer
|| qgroup
->excl
|| qgroup
->excl_cmpr
|| qgroup
->rfer_cmpr
))
1761 /* For higher level qgroup, we can only delete it if it has no child. */
1762 if (btrfs_qgroup_level(qgroup
->qgroupid
)) {
1763 if (!list_empty(&qgroup
->members
))
1769 * For level-0 qgroups, we can only delete it if it has no subvolume
1771 * This means even a subvolume is unlinked but not yet fully dropped,
1772 * we can not delete the qgroup.
1774 key
.objectid
= qgroup
->qgroupid
;
1775 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1777 path
= btrfs_alloc_path();
1781 ret
= btrfs_find_root(fs_info
->tree_root
, &key
, path
, NULL
, NULL
);
1782 btrfs_free_path(path
);
1784 * The @ret from btrfs_find_root() exactly matches our definition for
1785 * the return value, thus can be returned directly.
1790 int btrfs_remove_qgroup(struct btrfs_trans_handle
*trans
, u64 qgroupid
)
1792 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1793 struct btrfs_qgroup
*qgroup
;
1794 struct btrfs_qgroup_list
*list
;
1797 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1798 if (!fs_info
->quota_root
) {
1803 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1809 ret
= can_delete_qgroup(fs_info
, qgroup
);
1817 /* Check if there are no children of this qgroup */
1818 if (!list_empty(&qgroup
->members
)) {
1823 ret
= del_qgroup_item(trans
, qgroupid
);
1824 if (ret
&& ret
!= -ENOENT
)
1827 while (!list_empty(&qgroup
->groups
)) {
1828 list
= list_first_entry(&qgroup
->groups
,
1829 struct btrfs_qgroup_list
, next_group
);
1830 ret
= __del_qgroup_relation(trans
, qgroupid
,
1831 list
->group
->qgroupid
);
1836 spin_lock(&fs_info
->qgroup_lock
);
1838 * Warn on reserved space. The subvolume should has no child nor
1839 * corresponding subvolume.
1840 * Thus its reserved space should all be zero, no matter if qgroup
1841 * is consistent or the mode.
1843 WARN_ON(qgroup
->rsv
.values
[BTRFS_QGROUP_RSV_DATA
] ||
1844 qgroup
->rsv
.values
[BTRFS_QGROUP_RSV_META_PREALLOC
] ||
1845 qgroup
->rsv
.values
[BTRFS_QGROUP_RSV_META_PERTRANS
]);
1847 * The same for rfer/excl numbers, but that's only if our qgroup is
1848 * consistent and if it's in regular qgroup mode.
1849 * For simple mode it's not as accurate thus we can hit non-zero values
1852 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_FULL
&&
1853 !(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
)) {
1854 if (WARN_ON(qgroup
->rfer
|| qgroup
->excl
||
1855 qgroup
->rfer_cmpr
|| qgroup
->excl_cmpr
)) {
1856 btrfs_warn_rl(fs_info
,
1857 "to be deleted qgroup %u/%llu has non-zero numbers, rfer %llu rfer_cmpr %llu excl %llu excl_cmpr %llu",
1858 btrfs_qgroup_level(qgroup
->qgroupid
),
1859 btrfs_qgroup_subvolid(qgroup
->qgroupid
),
1860 qgroup
->rfer
, qgroup
->rfer_cmpr
,
1861 qgroup
->excl
, qgroup
->excl_cmpr
);
1862 qgroup_mark_inconsistent(fs_info
);
1865 del_qgroup_rb(fs_info
, qgroupid
);
1866 spin_unlock(&fs_info
->qgroup_lock
);
1869 * Remove the qgroup from sysfs now without holding the qgroup_lock
1870 * spinlock, since the sysfs_remove_group() function needs to take
1871 * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1873 btrfs_sysfs_del_one_qgroup(fs_info
, qgroup
);
1876 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1880 int btrfs_qgroup_cleanup_dropped_subvolume(struct btrfs_fs_info
*fs_info
, u64 subvolid
)
1882 struct btrfs_trans_handle
*trans
;
1885 if (!is_fstree(subvolid
) || !btrfs_qgroup_enabled(fs_info
) || !fs_info
->quota_root
)
1889 * Commit current transaction to make sure all the rfer/excl numbers
1892 trans
= btrfs_start_transaction(fs_info
->quota_root
, 0);
1894 return PTR_ERR(trans
);
1896 ret
= btrfs_commit_transaction(trans
);
1900 /* Start new trans to delete the qgroup info and limit items. */
1901 trans
= btrfs_start_transaction(fs_info
->quota_root
, 2);
1903 return PTR_ERR(trans
);
1904 ret
= btrfs_remove_qgroup(trans
, subvolid
);
1905 btrfs_end_transaction(trans
);
1907 * It's squota and the subvolume still has numbers needed for future
1908 * accounting, in this case we can not delete it. Just skip it.
1915 int btrfs_limit_qgroup(struct btrfs_trans_handle
*trans
, u64 qgroupid
,
1916 struct btrfs_qgroup_limit
*limit
)
1918 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
1919 struct btrfs_qgroup
*qgroup
;
1921 /* Sometimes we would want to clear the limit on this qgroup.
1922 * To meet this requirement, we treat the -1 as a special value
1923 * which tell kernel to clear the limit on this qgroup.
1925 const u64 CLEAR_VALUE
= -1;
1927 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1928 if (!fs_info
->quota_root
) {
1933 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1939 spin_lock(&fs_info
->qgroup_lock
);
1940 if (limit
->flags
& BTRFS_QGROUP_LIMIT_MAX_RFER
) {
1941 if (limit
->max_rfer
== CLEAR_VALUE
) {
1942 qgroup
->lim_flags
&= ~BTRFS_QGROUP_LIMIT_MAX_RFER
;
1943 limit
->flags
&= ~BTRFS_QGROUP_LIMIT_MAX_RFER
;
1944 qgroup
->max_rfer
= 0;
1946 qgroup
->max_rfer
= limit
->max_rfer
;
1949 if (limit
->flags
& BTRFS_QGROUP_LIMIT_MAX_EXCL
) {
1950 if (limit
->max_excl
== CLEAR_VALUE
) {
1951 qgroup
->lim_flags
&= ~BTRFS_QGROUP_LIMIT_MAX_EXCL
;
1952 limit
->flags
&= ~BTRFS_QGROUP_LIMIT_MAX_EXCL
;
1953 qgroup
->max_excl
= 0;
1955 qgroup
->max_excl
= limit
->max_excl
;
1958 if (limit
->flags
& BTRFS_QGROUP_LIMIT_RSV_RFER
) {
1959 if (limit
->rsv_rfer
== CLEAR_VALUE
) {
1960 qgroup
->lim_flags
&= ~BTRFS_QGROUP_LIMIT_RSV_RFER
;
1961 limit
->flags
&= ~BTRFS_QGROUP_LIMIT_RSV_RFER
;
1962 qgroup
->rsv_rfer
= 0;
1964 qgroup
->rsv_rfer
= limit
->rsv_rfer
;
1967 if (limit
->flags
& BTRFS_QGROUP_LIMIT_RSV_EXCL
) {
1968 if (limit
->rsv_excl
== CLEAR_VALUE
) {
1969 qgroup
->lim_flags
&= ~BTRFS_QGROUP_LIMIT_RSV_EXCL
;
1970 limit
->flags
&= ~BTRFS_QGROUP_LIMIT_RSV_EXCL
;
1971 qgroup
->rsv_excl
= 0;
1973 qgroup
->rsv_excl
= limit
->rsv_excl
;
1976 qgroup
->lim_flags
|= limit
->flags
;
1978 spin_unlock(&fs_info
->qgroup_lock
);
1980 ret
= update_qgroup_limit_item(trans
, qgroup
);
1982 qgroup_mark_inconsistent(fs_info
);
1983 btrfs_info(fs_info
, "unable to update quota limit for %llu",
1988 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1993 * Inform qgroup to trace one dirty extent, its info is recorded in @record.
1994 * So qgroup can account it at transaction committing time.
1996 * No lock version, caller must acquire delayed ref lock and allocated memory,
1997 * then call btrfs_qgroup_trace_extent_post() after exiting lock context.
1999 * Return 0 for success insert
2000 * Return >0 for existing record, caller can free @record safely.
2001 * Return <0 for insertion failure, caller can free @record safely.
2003 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info
*fs_info
,
2004 struct btrfs_delayed_ref_root
*delayed_refs
,
2005 struct btrfs_qgroup_extent_record
*record
)
2007 struct btrfs_qgroup_extent_record
*existing
, *ret
;
2008 unsigned long bytenr
= record
->bytenr
;
2010 if (!btrfs_qgroup_full_accounting(fs_info
))
2013 lockdep_assert_held(&delayed_refs
->lock
);
2014 trace_btrfs_qgroup_trace_extent(fs_info
, record
);
2016 xa_lock(&delayed_refs
->dirty_extents
);
2017 existing
= xa_load(&delayed_refs
->dirty_extents
, bytenr
);
2019 if (record
->data_rsv
&& !existing
->data_rsv
) {
2020 existing
->data_rsv
= record
->data_rsv
;
2021 existing
->data_rsv_refroot
= record
->data_rsv_refroot
;
2023 xa_unlock(&delayed_refs
->dirty_extents
);
2027 ret
= __xa_store(&delayed_refs
->dirty_extents
, record
->bytenr
, record
, GFP_ATOMIC
);
2028 xa_unlock(&delayed_refs
->dirty_extents
);
2029 if (xa_is_err(ret
)) {
2030 qgroup_mark_inconsistent(fs_info
);
2038 * Post handler after qgroup_trace_extent_nolock().
2040 * NOTE: Current qgroup does the expensive backref walk at transaction
2041 * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
2043 * This is designed to allow btrfs_find_all_roots() to get correct new_roots
2046 * However for old_roots there is no need to do backref walk at that time,
2047 * since we search commit roots to walk backref and result will always be
2050 * Due to the nature of no lock version, we can't do backref there.
2051 * So we must call btrfs_qgroup_trace_extent_post() after exiting
2054 * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result
2055 * using current root, then we can move all expensive backref walk out of
2056 * transaction committing, but not now as qgroup accounting will be wrong again.
2058 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle
*trans
,
2059 struct btrfs_qgroup_extent_record
*qrecord
)
2061 struct btrfs_backref_walk_ctx ctx
= { 0 };
2064 if (!btrfs_qgroup_full_accounting(trans
->fs_info
))
2067 * We are always called in a context where we are already holding a
2068 * transaction handle. Often we are called when adding a data delayed
2069 * reference from btrfs_truncate_inode_items() (truncating or unlinking),
2070 * in which case we will be holding a write lock on extent buffer from a
2071 * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
2072 * acquire fs_info->commit_root_sem, because that is a higher level lock
2073 * that must be acquired before locking any extent buffers.
2075 * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
2076 * but we can't pass it a non-NULL transaction handle, because otherwise
2077 * it would not use commit roots and would lock extent buffers, causing
2078 * a deadlock if it ends up trying to read lock the same extent buffer
2079 * that was previously write locked at btrfs_truncate_inode_items().
2081 * So pass a NULL transaction handle to btrfs_find_all_roots() and
2082 * explicitly tell it to not acquire the commit_root_sem - if we are
2083 * holding a transaction handle we don't need its protection.
2085 ASSERT(trans
!= NULL
);
2087 if (trans
->fs_info
->qgroup_flags
& BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING
)
2090 ctx
.bytenr
= qrecord
->bytenr
;
2091 ctx
.fs_info
= trans
->fs_info
;
2093 ret
= btrfs_find_all_roots(&ctx
, true);
2095 qgroup_mark_inconsistent(trans
->fs_info
);
2096 btrfs_warn(trans
->fs_info
,
2097 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
2103 * Here we don't need to get the lock of
2104 * trans->transaction->delayed_refs, since inserted qrecord won't
2105 * be deleted, only qrecord->node may be modified (new qrecord insert)
2107 * So modifying qrecord->old_roots is safe here
2109 qrecord
->old_roots
= ctx
.roots
;
2114 * Inform qgroup to trace one dirty extent, specified by @bytenr and
2116 * So qgroup can account it at commit trans time.
2118 * Better encapsulated version, with memory allocation and backref walk for
2120 * So this can sleep.
2122 * Return 0 if the operation is done.
2123 * Return <0 for error, like memory allocation failure or invalid parameter
2126 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle
*trans
, u64 bytenr
,
2129 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2130 struct btrfs_qgroup_extent_record
*record
;
2131 struct btrfs_delayed_ref_root
*delayed_refs
;
2134 if (!btrfs_qgroup_full_accounting(fs_info
) || bytenr
== 0 || num_bytes
== 0)
2136 record
= kzalloc(sizeof(*record
), GFP_NOFS
);
2140 if (xa_reserve(&trans
->transaction
->delayed_refs
.dirty_extents
, bytenr
, GFP_NOFS
)) {
2145 delayed_refs
= &trans
->transaction
->delayed_refs
;
2146 record
->bytenr
= bytenr
;
2147 record
->num_bytes
= num_bytes
;
2148 record
->old_roots
= NULL
;
2150 spin_lock(&delayed_refs
->lock
);
2151 ret
= btrfs_qgroup_trace_extent_nolock(fs_info
, delayed_refs
, record
);
2152 spin_unlock(&delayed_refs
->lock
);
2154 /* Clean up if insertion fails or item exists. */
2155 xa_release(&delayed_refs
->dirty_extents
, record
->bytenr
);
2159 return btrfs_qgroup_trace_extent_post(trans
, record
);
2163 * Inform qgroup to trace all leaf items of data
2165 * Return 0 for success
2166 * Return <0 for error(ENOMEM)
2168 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle
*trans
,
2169 struct extent_buffer
*eb
)
2171 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2172 int nr
= btrfs_header_nritems(eb
);
2173 int i
, extent_type
, ret
;
2174 struct btrfs_key key
;
2175 struct btrfs_file_extent_item
*fi
;
2176 u64 bytenr
, num_bytes
;
2178 /* We can be called directly from walk_up_proc() */
2179 if (!btrfs_qgroup_full_accounting(fs_info
))
2182 for (i
= 0; i
< nr
; i
++) {
2183 btrfs_item_key_to_cpu(eb
, &key
, i
);
2185 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2188 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
2189 /* filter out non qgroup-accountable extents */
2190 extent_type
= btrfs_file_extent_type(eb
, fi
);
2192 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
2195 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
2199 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
2201 ret
= btrfs_qgroup_trace_extent(trans
, bytenr
, num_bytes
);
2210 * Walk up the tree from the bottom, freeing leaves and any interior
2211 * nodes which have had all slots visited. If a node (leaf or
2212 * interior) is freed, the node above it will have it's slot
2213 * incremented. The root node will never be freed.
2215 * At the end of this function, we should have a path which has all
2216 * slots incremented to the next position for a search. If we need to
2217 * read a new node it will be NULL and the node above it will have the
2218 * correct slot selected for a later read.
2220 * If we increment the root nodes slot counter past the number of
2221 * elements, 1 is returned to signal completion of the search.
2223 static int adjust_slots_upwards(struct btrfs_path
*path
, int root_level
)
2227 struct extent_buffer
*eb
;
2229 if (root_level
== 0)
2232 while (level
<= root_level
) {
2233 eb
= path
->nodes
[level
];
2234 nr
= btrfs_header_nritems(eb
);
2235 path
->slots
[level
]++;
2236 slot
= path
->slots
[level
];
2237 if (slot
>= nr
|| level
== 0) {
2239 * Don't free the root - we will detect this
2240 * condition after our loop and return a
2241 * positive value for caller to stop walking the tree.
2243 if (level
!= root_level
) {
2244 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
2245 path
->locks
[level
] = 0;
2247 free_extent_buffer(eb
);
2248 path
->nodes
[level
] = NULL
;
2249 path
->slots
[level
] = 0;
2253 * We have a valid slot to walk back down
2254 * from. Stop here so caller can process these
2263 eb
= path
->nodes
[root_level
];
2264 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
2271 * Helper function to trace a subtree tree block swap.
2273 * The swap will happen in highest tree block, but there may be a lot of
2274 * tree blocks involved.
2277 * OO = Old tree blocks
2278 * NN = New tree blocks allocated during balance
2280 * File tree (257) Reloc tree for 257
2283 * L1 OO OO (a) OO NN (a)
2285 * L0 OO OO OO OO OO OO NN NN
2288 * When calling qgroup_trace_extent_swap(), we will pass:
2290 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
2294 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
2295 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
2297 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
2299 * 1) Tree search from @src_eb
2300 * It should acts as a simplified btrfs_search_slot().
2301 * The key for search can be extracted from @dst_path->nodes[dst_level]
2304 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
2305 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
2306 * They should be marked during previous (@dst_level = 1) iteration.
2308 * 3) Mark file extents in leaves dirty
2309 * We don't have good way to pick out new file extents only.
2310 * So we still follow the old method by scanning all file extents in
2313 * This function can free us from keeping two paths, thus later we only need
2314 * to care about how to iterate all new tree blocks in reloc tree.
2316 static int qgroup_trace_extent_swap(struct btrfs_trans_handle
* trans
,
2317 struct extent_buffer
*src_eb
,
2318 struct btrfs_path
*dst_path
,
2319 int dst_level
, int root_level
,
2322 struct btrfs_key key
;
2323 struct btrfs_path
*src_path
;
2324 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2325 u32 nodesize
= fs_info
->nodesize
;
2326 int cur_level
= root_level
;
2329 BUG_ON(dst_level
> root_level
);
2330 /* Level mismatch */
2331 if (btrfs_header_level(src_eb
) != root_level
)
2334 src_path
= btrfs_alloc_path();
2341 btrfs_node_key_to_cpu(dst_path
->nodes
[dst_level
], &key
, 0);
2343 btrfs_item_key_to_cpu(dst_path
->nodes
[dst_level
], &key
, 0);
2346 atomic_inc(&src_eb
->refs
);
2347 src_path
->nodes
[root_level
] = src_eb
;
2348 src_path
->slots
[root_level
] = dst_path
->slots
[root_level
];
2349 src_path
->locks
[root_level
] = 0;
2351 /* A simplified version of btrfs_search_slot() */
2352 while (cur_level
>= dst_level
) {
2353 struct btrfs_key src_key
;
2354 struct btrfs_key dst_key
;
2356 if (src_path
->nodes
[cur_level
] == NULL
) {
2357 struct extent_buffer
*eb
;
2360 eb
= src_path
->nodes
[cur_level
+ 1];
2361 parent_slot
= src_path
->slots
[cur_level
+ 1];
2363 eb
= btrfs_read_node_slot(eb
, parent_slot
);
2369 src_path
->nodes
[cur_level
] = eb
;
2371 btrfs_tree_read_lock(eb
);
2372 src_path
->locks
[cur_level
] = BTRFS_READ_LOCK
;
2375 src_path
->slots
[cur_level
] = dst_path
->slots
[cur_level
];
2377 btrfs_node_key_to_cpu(dst_path
->nodes
[cur_level
],
2378 &dst_key
, dst_path
->slots
[cur_level
]);
2379 btrfs_node_key_to_cpu(src_path
->nodes
[cur_level
],
2380 &src_key
, src_path
->slots
[cur_level
]);
2382 btrfs_item_key_to_cpu(dst_path
->nodes
[cur_level
],
2383 &dst_key
, dst_path
->slots
[cur_level
]);
2384 btrfs_item_key_to_cpu(src_path
->nodes
[cur_level
],
2385 &src_key
, src_path
->slots
[cur_level
]);
2387 /* Content mismatch, something went wrong */
2388 if (btrfs_comp_cpu_keys(&dst_key
, &src_key
)) {
2396 * Now both @dst_path and @src_path have been populated, record the tree
2397 * blocks for qgroup accounting.
2399 ret
= btrfs_qgroup_trace_extent(trans
, src_path
->nodes
[dst_level
]->start
,
2403 ret
= btrfs_qgroup_trace_extent(trans
, dst_path
->nodes
[dst_level
]->start
,
2408 /* Record leaf file extents */
2409 if (dst_level
== 0 && trace_leaf
) {
2410 ret
= btrfs_qgroup_trace_leaf_items(trans
, src_path
->nodes
[0]);
2413 ret
= btrfs_qgroup_trace_leaf_items(trans
, dst_path
->nodes
[0]);
2416 btrfs_free_path(src_path
);
2421 * Helper function to do recursive generation-aware depth-first search, to
2422 * locate all new tree blocks in a subtree of reloc tree.
2424 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2433 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2437 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2438 * above tree blocks along with their counter parts in file tree.
2439 * While during search, old tree blocks OO(c) will be skipped as tree block swap
2440 * won't affect OO(c).
2442 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle
* trans
,
2443 struct extent_buffer
*src_eb
,
2444 struct btrfs_path
*dst_path
,
2445 int cur_level
, int root_level
,
2446 u64 last_snapshot
, bool trace_leaf
)
2448 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2449 struct extent_buffer
*eb
;
2450 bool need_cleanup
= false;
2454 /* Level sanity check */
2455 if (cur_level
< 0 || cur_level
>= BTRFS_MAX_LEVEL
- 1 ||
2456 root_level
< 0 || root_level
>= BTRFS_MAX_LEVEL
- 1 ||
2457 root_level
< cur_level
) {
2458 btrfs_err_rl(fs_info
,
2459 "%s: bad levels, cur_level=%d root_level=%d",
2460 __func__
, cur_level
, root_level
);
2464 /* Read the tree block if needed */
2465 if (dst_path
->nodes
[cur_level
] == NULL
) {
2470 * dst_path->nodes[root_level] must be initialized before
2471 * calling this function.
2473 if (cur_level
== root_level
) {
2474 btrfs_err_rl(fs_info
,
2475 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2476 __func__
, root_level
, root_level
, cur_level
);
2481 * We need to get child blockptr/gen from parent before we can
2484 eb
= dst_path
->nodes
[cur_level
+ 1];
2485 parent_slot
= dst_path
->slots
[cur_level
+ 1];
2486 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
2488 /* This node is old, no need to trace */
2489 if (child_gen
< last_snapshot
)
2492 eb
= btrfs_read_node_slot(eb
, parent_slot
);
2498 dst_path
->nodes
[cur_level
] = eb
;
2499 dst_path
->slots
[cur_level
] = 0;
2501 btrfs_tree_read_lock(eb
);
2502 dst_path
->locks
[cur_level
] = BTRFS_READ_LOCK
;
2503 need_cleanup
= true;
2506 /* Now record this tree block and its counter part for qgroups */
2507 ret
= qgroup_trace_extent_swap(trans
, src_eb
, dst_path
, cur_level
,
2508 root_level
, trace_leaf
);
2512 eb
= dst_path
->nodes
[cur_level
];
2514 if (cur_level
> 0) {
2515 /* Iterate all child tree blocks */
2516 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
2517 /* Skip old tree blocks as they won't be swapped */
2518 if (btrfs_node_ptr_generation(eb
, i
) < last_snapshot
)
2520 dst_path
->slots
[cur_level
] = i
;
2522 /* Recursive call (at most 7 times) */
2523 ret
= qgroup_trace_new_subtree_blocks(trans
, src_eb
,
2524 dst_path
, cur_level
- 1, root_level
,
2525 last_snapshot
, trace_leaf
);
2534 btrfs_tree_unlock_rw(dst_path
->nodes
[cur_level
],
2535 dst_path
->locks
[cur_level
]);
2536 free_extent_buffer(dst_path
->nodes
[cur_level
]);
2537 dst_path
->nodes
[cur_level
] = NULL
;
2538 dst_path
->slots
[cur_level
] = 0;
2539 dst_path
->locks
[cur_level
] = 0;
2545 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle
*trans
,
2546 struct extent_buffer
*src_eb
,
2547 struct extent_buffer
*dst_eb
,
2548 u64 last_snapshot
, bool trace_leaf
)
2550 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2551 struct btrfs_path
*dst_path
= NULL
;
2555 if (!btrfs_qgroup_full_accounting(fs_info
))
2558 /* Wrong parameter order */
2559 if (btrfs_header_generation(src_eb
) > btrfs_header_generation(dst_eb
)) {
2560 btrfs_err_rl(fs_info
,
2561 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__
,
2562 btrfs_header_generation(src_eb
),
2563 btrfs_header_generation(dst_eb
));
2567 if (!extent_buffer_uptodate(src_eb
) || !extent_buffer_uptodate(dst_eb
)) {
2572 level
= btrfs_header_level(dst_eb
);
2573 dst_path
= btrfs_alloc_path();
2579 atomic_inc(&dst_eb
->refs
);
2580 dst_path
->nodes
[level
] = dst_eb
;
2581 dst_path
->slots
[level
] = 0;
2582 dst_path
->locks
[level
] = 0;
2584 /* Do the generation aware breadth-first search */
2585 ret
= qgroup_trace_new_subtree_blocks(trans
, src_eb
, dst_path
, level
,
2586 level
, last_snapshot
, trace_leaf
);
2592 btrfs_free_path(dst_path
);
2594 qgroup_mark_inconsistent(fs_info
);
2599 * Inform qgroup to trace a whole subtree, including all its child tree
2601 * The root tree block is specified by @root_eb.
2603 * Normally used by relocation(tree block swap) and subvolume deletion.
2605 * Return 0 for success
2606 * Return <0 for error(ENOMEM or tree search error)
2608 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle
*trans
,
2609 struct extent_buffer
*root_eb
,
2610 u64 root_gen
, int root_level
)
2612 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2615 u8 drop_subptree_thres
;
2616 struct extent_buffer
*eb
= root_eb
;
2617 struct btrfs_path
*path
= NULL
;
2619 ASSERT(0 <= root_level
&& root_level
< BTRFS_MAX_LEVEL
);
2620 ASSERT(root_eb
!= NULL
);
2622 if (!btrfs_qgroup_full_accounting(fs_info
))
2625 spin_lock(&fs_info
->qgroup_lock
);
2626 drop_subptree_thres
= fs_info
->qgroup_drop_subtree_thres
;
2627 spin_unlock(&fs_info
->qgroup_lock
);
2630 * This function only gets called for snapshot drop, if we hit a high
2631 * node here, it means we are going to change ownership for quite a lot
2632 * of extents, which will greatly slow down btrfs_commit_transaction().
2634 * So here if we find a high tree here, we just skip the accounting and
2635 * mark qgroup inconsistent.
2637 if (root_level
>= drop_subptree_thres
) {
2638 qgroup_mark_inconsistent(fs_info
);
2642 if (!extent_buffer_uptodate(root_eb
)) {
2643 struct btrfs_tree_parent_check check
= {
2644 .has_first_key
= false,
2645 .transid
= root_gen
,
2649 ret
= btrfs_read_extent_buffer(root_eb
, &check
);
2654 if (root_level
== 0) {
2655 ret
= btrfs_qgroup_trace_leaf_items(trans
, root_eb
);
2659 path
= btrfs_alloc_path();
2664 * Walk down the tree. Missing extent blocks are filled in as
2665 * we go. Metadata is accounted every time we read a new
2668 * When we reach a leaf, we account for file extent items in it,
2669 * walk back up the tree (adjusting slot pointers as we go)
2670 * and restart the search process.
2672 atomic_inc(&root_eb
->refs
); /* For path */
2673 path
->nodes
[root_level
] = root_eb
;
2674 path
->slots
[root_level
] = 0;
2675 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
2678 while (level
>= 0) {
2679 if (path
->nodes
[level
] == NULL
) {
2684 * We need to get child blockptr from parent before we
2687 eb
= path
->nodes
[level
+ 1];
2688 parent_slot
= path
->slots
[level
+ 1];
2689 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
2691 eb
= btrfs_read_node_slot(eb
, parent_slot
);
2697 path
->nodes
[level
] = eb
;
2698 path
->slots
[level
] = 0;
2700 btrfs_tree_read_lock(eb
);
2701 path
->locks
[level
] = BTRFS_READ_LOCK
;
2703 ret
= btrfs_qgroup_trace_extent(trans
, child_bytenr
,
2710 ret
= btrfs_qgroup_trace_leaf_items(trans
,
2711 path
->nodes
[level
]);
2715 /* Nonzero return here means we completed our search */
2716 ret
= adjust_slots_upwards(path
, root_level
);
2720 /* Restart search with new slots */
2729 btrfs_free_path(path
);
2734 static void qgroup_iterator_nested_add(struct list_head
*head
, struct btrfs_qgroup
*qgroup
)
2736 if (!list_empty(&qgroup
->nested_iterator
))
2739 list_add_tail(&qgroup
->nested_iterator
, head
);
2742 static void qgroup_iterator_nested_clean(struct list_head
*head
)
2744 while (!list_empty(head
)) {
2745 struct btrfs_qgroup
*qgroup
;
2747 qgroup
= list_first_entry(head
, struct btrfs_qgroup
, nested_iterator
);
2748 list_del_init(&qgroup
->nested_iterator
);
2752 #define UPDATE_NEW 0
2753 #define UPDATE_OLD 1
2755 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2757 static void qgroup_update_refcnt(struct btrfs_fs_info
*fs_info
,
2758 struct ulist
*roots
, struct list_head
*qgroups
,
2759 u64 seq
, int update_old
)
2761 struct ulist_node
*unode
;
2762 struct ulist_iterator uiter
;
2763 struct btrfs_qgroup
*qg
;
2767 ULIST_ITER_INIT(&uiter
);
2768 while ((unode
= ulist_next(roots
, &uiter
))) {
2771 qg
= find_qgroup_rb(fs_info
, unode
->val
);
2775 qgroup_iterator_nested_add(qgroups
, qg
);
2776 qgroup_iterator_add(&tmp
, qg
);
2777 list_for_each_entry(qg
, &tmp
, iterator
) {
2778 struct btrfs_qgroup_list
*glist
;
2781 btrfs_qgroup_update_old_refcnt(qg
, seq
, 1);
2783 btrfs_qgroup_update_new_refcnt(qg
, seq
, 1);
2785 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
2786 qgroup_iterator_nested_add(qgroups
, glist
->group
);
2787 qgroup_iterator_add(&tmp
, glist
->group
);
2790 qgroup_iterator_clean(&tmp
);
2795 * Update qgroup rfer/excl counters.
2796 * Rfer update is easy, codes can explain themselves.
2798 * Excl update is tricky, the update is split into 2 parts.
2799 * Part 1: Possible exclusive <-> sharing detect:
2801 * -------------------------------------
2803 * -------------------------------------
2805 * -------------------------------------
2808 * A: cur_old_roots < nr_old_roots (not exclusive before)
2809 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2810 * B: cur_new_roots < nr_new_roots (not exclusive now)
2811 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2814 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2815 * *: Definitely not changed. **: Possible unchanged.
2817 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2819 * To make the logic clear, we first use condition A and B to split
2820 * combination into 4 results.
2822 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2823 * only on variant maybe 0.
2825 * Lastly, check result **, since there are 2 variants maybe 0, split them
2827 * But this time we don't need to consider other things, the codes and logic
2828 * is easy to understand now.
2830 static void qgroup_update_counters(struct btrfs_fs_info
*fs_info
,
2831 struct list_head
*qgroups
, u64 nr_old_roots
,
2832 u64 nr_new_roots
, u64 num_bytes
, u64 seq
)
2834 struct btrfs_qgroup
*qg
;
2836 list_for_each_entry(qg
, qgroups
, nested_iterator
) {
2837 u64 cur_new_count
, cur_old_count
;
2840 cur_old_count
= btrfs_qgroup_get_old_refcnt(qg
, seq
);
2841 cur_new_count
= btrfs_qgroup_get_new_refcnt(qg
, seq
);
2843 trace_qgroup_update_counters(fs_info
, qg
, cur_old_count
,
2846 /* Rfer update part */
2847 if (cur_old_count
== 0 && cur_new_count
> 0) {
2848 qg
->rfer
+= num_bytes
;
2849 qg
->rfer_cmpr
+= num_bytes
;
2852 if (cur_old_count
> 0 && cur_new_count
== 0) {
2853 qg
->rfer
-= num_bytes
;
2854 qg
->rfer_cmpr
-= num_bytes
;
2858 /* Excl update part */
2859 /* Exclusive/none -> shared case */
2860 if (cur_old_count
== nr_old_roots
&&
2861 cur_new_count
< nr_new_roots
) {
2862 /* Exclusive -> shared */
2863 if (cur_old_count
!= 0) {
2864 qg
->excl
-= num_bytes
;
2865 qg
->excl_cmpr
-= num_bytes
;
2870 /* Shared -> exclusive/none case */
2871 if (cur_old_count
< nr_old_roots
&&
2872 cur_new_count
== nr_new_roots
) {
2873 /* Shared->exclusive */
2874 if (cur_new_count
!= 0) {
2875 qg
->excl
+= num_bytes
;
2876 qg
->excl_cmpr
+= num_bytes
;
2881 /* Exclusive/none -> exclusive/none case */
2882 if (cur_old_count
== nr_old_roots
&&
2883 cur_new_count
== nr_new_roots
) {
2884 if (cur_old_count
== 0) {
2885 /* None -> exclusive/none */
2887 if (cur_new_count
!= 0) {
2888 /* None -> exclusive */
2889 qg
->excl
+= num_bytes
;
2890 qg
->excl_cmpr
+= num_bytes
;
2893 /* None -> none, nothing changed */
2895 /* Exclusive -> exclusive/none */
2897 if (cur_new_count
== 0) {
2898 /* Exclusive -> none */
2899 qg
->excl
-= num_bytes
;
2900 qg
->excl_cmpr
-= num_bytes
;
2903 /* Exclusive -> exclusive, nothing changed */
2908 qgroup_dirty(fs_info
, qg
);
2913 * Check if the @roots potentially is a list of fs tree roots
2915 * Return 0 for definitely not a fs/subvol tree roots ulist
2916 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2919 static int maybe_fs_roots(struct ulist
*roots
)
2921 struct ulist_node
*unode
;
2922 struct ulist_iterator uiter
;
2924 /* Empty one, still possible for fs roots */
2925 if (!roots
|| roots
->nnodes
== 0)
2928 ULIST_ITER_INIT(&uiter
);
2929 unode
= ulist_next(roots
, &uiter
);
2934 * If it contains fs tree roots, then it must belong to fs/subvol
2936 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2938 return is_fstree(unode
->val
);
2941 int btrfs_qgroup_account_extent(struct btrfs_trans_handle
*trans
, u64 bytenr
,
2942 u64 num_bytes
, struct ulist
*old_roots
,
2943 struct ulist
*new_roots
)
2945 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2948 u64 nr_new_roots
= 0;
2949 u64 nr_old_roots
= 0;
2953 * If quotas get disabled meanwhile, the resources need to be freed and
2954 * we can't just exit here.
2956 if (!btrfs_qgroup_full_accounting(fs_info
) ||
2957 fs_info
->qgroup_flags
& BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING
)
2961 if (!maybe_fs_roots(new_roots
))
2963 nr_new_roots
= new_roots
->nnodes
;
2966 if (!maybe_fs_roots(old_roots
))
2968 nr_old_roots
= old_roots
->nnodes
;
2971 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2972 if (nr_old_roots
== 0 && nr_new_roots
== 0)
2975 trace_btrfs_qgroup_account_extent(fs_info
, trans
->transid
, bytenr
,
2976 num_bytes
, nr_old_roots
, nr_new_roots
);
2978 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2979 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
) {
2980 if (fs_info
->qgroup_rescan_progress
.objectid
<= bytenr
) {
2981 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2986 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2988 spin_lock(&fs_info
->qgroup_lock
);
2989 seq
= fs_info
->qgroup_seq
;
2991 /* Update old refcnts using old_roots */
2992 qgroup_update_refcnt(fs_info
, old_roots
, &qgroups
, seq
, UPDATE_OLD
);
2994 /* Update new refcnts using new_roots */
2995 qgroup_update_refcnt(fs_info
, new_roots
, &qgroups
, seq
, UPDATE_NEW
);
2997 qgroup_update_counters(fs_info
, &qgroups
, nr_old_roots
, nr_new_roots
,
3001 * We're done using the iterator, release all its qgroups while holding
3002 * fs_info->qgroup_lock so that we don't race with btrfs_remove_qgroup()
3003 * and trigger use-after-free accesses to qgroups.
3005 qgroup_iterator_nested_clean(&qgroups
);
3008 * Bump qgroup_seq to avoid seq overlap
3010 fs_info
->qgroup_seq
+= max(nr_old_roots
, nr_new_roots
) + 1;
3011 spin_unlock(&fs_info
->qgroup_lock
);
3013 ulist_free(old_roots
);
3014 ulist_free(new_roots
);
3018 int btrfs_qgroup_account_extents(struct btrfs_trans_handle
*trans
)
3020 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3021 struct btrfs_qgroup_extent_record
*record
;
3022 struct btrfs_delayed_ref_root
*delayed_refs
;
3023 struct ulist
*new_roots
= NULL
;
3024 unsigned long index
;
3025 u64 num_dirty_extents
= 0;
3029 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_SIMPLE
)
3032 delayed_refs
= &trans
->transaction
->delayed_refs
;
3033 qgroup_to_skip
= delayed_refs
->qgroup_to_skip
;
3034 xa_for_each(&delayed_refs
->dirty_extents
, index
, record
) {
3035 num_dirty_extents
++;
3036 trace_btrfs_qgroup_account_extents(fs_info
, record
);
3038 if (!ret
&& !(fs_info
->qgroup_flags
&
3039 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING
)) {
3040 struct btrfs_backref_walk_ctx ctx
= { 0 };
3042 ctx
.bytenr
= record
->bytenr
;
3043 ctx
.fs_info
= fs_info
;
3046 * Old roots should be searched when inserting qgroup
3049 * But for INCONSISTENT (NO_ACCOUNTING) -> rescan case,
3050 * we may have some record inserted during
3051 * NO_ACCOUNTING (thus no old_roots populated), but
3052 * later we start rescan, which clears NO_ACCOUNTING,
3053 * leaving some inserted records without old_roots
3056 * Those cases are rare and should not cause too much
3057 * time spent during commit_transaction().
3059 if (!record
->old_roots
) {
3060 /* Search commit root to find old_roots */
3061 ret
= btrfs_find_all_roots(&ctx
, false);
3064 record
->old_roots
= ctx
.roots
;
3069 * Use BTRFS_SEQ_LAST as time_seq to do special search,
3070 * which doesn't lock tree or delayed_refs and search
3071 * current root. It's safe inside commit_transaction().
3074 ctx
.time_seq
= BTRFS_SEQ_LAST
;
3075 ret
= btrfs_find_all_roots(&ctx
, false);
3078 new_roots
= ctx
.roots
;
3079 if (qgroup_to_skip
) {
3080 ulist_del(new_roots
, qgroup_to_skip
, 0);
3081 ulist_del(record
->old_roots
, qgroup_to_skip
,
3084 ret
= btrfs_qgroup_account_extent(trans
, record
->bytenr
,
3088 record
->old_roots
= NULL
;
3091 /* Free the reserved data space */
3092 btrfs_qgroup_free_refroot(fs_info
,
3093 record
->data_rsv_refroot
,
3095 BTRFS_QGROUP_RSV_DATA
);
3097 ulist_free(record
->old_roots
);
3098 ulist_free(new_roots
);
3100 xa_erase(&delayed_refs
->dirty_extents
, index
);
3104 trace_qgroup_num_dirty_extents(fs_info
, trans
->transid
,
3110 * Writes all changed qgroups to disk.
3111 * Called by the transaction commit path and the qgroup assign ioctl.
3113 int btrfs_run_qgroups(struct btrfs_trans_handle
*trans
)
3115 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3119 * In case we are called from the qgroup assign ioctl, assert that we
3120 * are holding the qgroup_ioctl_lock, otherwise we can race with a quota
3121 * disable operation (ioctl) and access a freed quota root.
3123 if (trans
->transaction
->state
!= TRANS_STATE_COMMIT_DOING
)
3124 lockdep_assert_held(&fs_info
->qgroup_ioctl_lock
);
3126 if (!fs_info
->quota_root
)
3129 spin_lock(&fs_info
->qgroup_lock
);
3130 while (!list_empty(&fs_info
->dirty_qgroups
)) {
3131 struct btrfs_qgroup
*qgroup
;
3132 qgroup
= list_first_entry(&fs_info
->dirty_qgroups
,
3133 struct btrfs_qgroup
, dirty
);
3134 list_del_init(&qgroup
->dirty
);
3135 spin_unlock(&fs_info
->qgroup_lock
);
3136 ret
= update_qgroup_info_item(trans
, qgroup
);
3138 qgroup_mark_inconsistent(fs_info
);
3139 ret
= update_qgroup_limit_item(trans
, qgroup
);
3141 qgroup_mark_inconsistent(fs_info
);
3142 spin_lock(&fs_info
->qgroup_lock
);
3144 if (btrfs_qgroup_enabled(fs_info
))
3145 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_ON
;
3147 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_ON
;
3148 spin_unlock(&fs_info
->qgroup_lock
);
3150 ret
= update_qgroup_status_item(trans
);
3152 qgroup_mark_inconsistent(fs_info
);
3157 int btrfs_qgroup_check_inherit(struct btrfs_fs_info
*fs_info
,
3158 struct btrfs_qgroup_inherit
*inherit
,
3161 if (inherit
->flags
& ~BTRFS_QGROUP_INHERIT_FLAGS_SUPP
)
3163 if (size
< sizeof(*inherit
) || size
> PAGE_SIZE
)
3167 * In the past we allowed btrfs_qgroup_inherit to specify to copy
3168 * rfer/excl numbers directly from other qgroups. This behavior has
3169 * been disabled in userspace for a very long time, but here we should
3170 * also disable it in kernel, as this behavior is known to mark qgroup
3171 * inconsistent, and a rescan would wipe out the changes anyway.
3173 * Reject any btrfs_qgroup_inherit with num_ref_copies or num_excl_copies.
3175 if (inherit
->num_ref_copies
> 0 || inherit
->num_excl_copies
> 0)
3178 if (size
!= struct_size(inherit
, qgroups
, inherit
->num_qgroups
))
3182 * Skip the inherit source qgroups check if qgroup is not enabled.
3183 * Qgroup can still be later enabled causing problems, but in that case
3184 * btrfs_qgroup_inherit() would just ignore those invalid ones.
3186 if (!btrfs_qgroup_enabled(fs_info
))
3190 * Now check all the remaining qgroups, they should all:
3193 * - Be higher level qgroups.
3195 for (int i
= 0; i
< inherit
->num_qgroups
; i
++) {
3196 struct btrfs_qgroup
*qgroup
;
3197 u64 qgroupid
= inherit
->qgroups
[i
];
3199 if (btrfs_qgroup_level(qgroupid
) == 0)
3202 spin_lock(&fs_info
->qgroup_lock
);
3203 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
3205 spin_unlock(&fs_info
->qgroup_lock
);
3208 spin_unlock(&fs_info
->qgroup_lock
);
3213 static int qgroup_auto_inherit(struct btrfs_fs_info
*fs_info
,
3215 struct btrfs_qgroup_inherit
**inherit
)
3218 u64 num_qgroups
= 0;
3219 struct btrfs_qgroup
*inode_qg
;
3220 struct btrfs_qgroup_list
*qg_list
;
3221 struct btrfs_qgroup_inherit
*res
;
3228 inode_qg
= find_qgroup_rb(fs_info
, inode_rootid
);
3232 num_qgroups
= list_count_nodes(&inode_qg
->groups
);
3237 struct_sz
= struct_size(res
, qgroups
, num_qgroups
);
3238 if (struct_sz
== SIZE_MAX
)
3241 res
= kzalloc(struct_sz
, GFP_NOFS
);
3244 res
->num_qgroups
= num_qgroups
;
3245 qgids
= res
->qgroups
;
3247 list_for_each_entry(qg_list
, &inode_qg
->groups
, next_group
)
3248 qgids
[i
++] = qg_list
->group
->qgroupid
;
3255 * Check if we can skip rescan when inheriting qgroups. If @src has a single
3256 * @parent, and that @parent is owning all its bytes exclusively, we can skip
3257 * the full rescan, by just adding nodesize to the @parent's excl/rfer.
3259 * Return <0 for fatal errors (like srcid/parentid has no qgroup).
3260 * Return 0 if a quick inherit is done.
3261 * Return >0 if a quick inherit is not possible, and a full rescan is needed.
3263 static int qgroup_snapshot_quick_inherit(struct btrfs_fs_info
*fs_info
,
3264 u64 srcid
, u64 parentid
)
3266 struct btrfs_qgroup
*src
;
3267 struct btrfs_qgroup
*parent
;
3268 struct btrfs_qgroup_list
*list
;
3271 src
= find_qgroup_rb(fs_info
, srcid
);
3274 parent
= find_qgroup_rb(fs_info
, parentid
);
3279 * Source has no parent qgroup, but our new qgroup would have one.
3280 * Qgroup numbers would become inconsistent.
3282 if (list_empty(&src
->groups
))
3285 list_for_each_entry(list
, &src
->groups
, next_group
) {
3286 /* The parent is not the same, quick update is not possible. */
3287 if (list
->group
->qgroupid
!= parentid
)
3291 * More than one parent qgroup, we can't be sure about accounting
3299 * The parent is not exclusively owning all its bytes. We're not sure
3300 * if the source has any bytes not fully owned by the parent.
3302 if (parent
->excl
!= parent
->rfer
)
3305 parent
->excl
+= fs_info
->nodesize
;
3306 parent
->rfer
+= fs_info
->nodesize
;
3311 * Copy the accounting information between qgroups. This is necessary
3312 * when a snapshot or a subvolume is created. Throwing an error will
3313 * cause a transaction abort so we take extra care here to only error
3314 * when a readonly fs is a reasonable outcome.
3316 int btrfs_qgroup_inherit(struct btrfs_trans_handle
*trans
, u64 srcid
,
3317 u64 objectid
, u64 inode_rootid
,
3318 struct btrfs_qgroup_inherit
*inherit
)
3322 bool committing
= false;
3323 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3324 struct btrfs_root
*quota_root
;
3325 struct btrfs_qgroup
*srcgroup
;
3326 struct btrfs_qgroup
*dstgroup
;
3327 struct btrfs_qgroup
*prealloc
;
3328 struct btrfs_qgroup_list
**qlist_prealloc
= NULL
;
3329 bool free_inherit
= false;
3330 bool need_rescan
= false;
3334 prealloc
= kzalloc(sizeof(*prealloc
), GFP_NOFS
);
3339 * There are only two callers of this function.
3341 * One in create_subvol() in the ioctl context, which needs to hold
3342 * the qgroup_ioctl_lock.
3344 * The other one in create_pending_snapshot() where no other qgroup
3345 * code can modify the fs as they all need to either start a new trans
3346 * or hold a trans handler, thus we don't need to hold
3347 * qgroup_ioctl_lock.
3348 * This would avoid long and complex lock chain and make lockdep happy.
3350 spin_lock(&fs_info
->trans_lock
);
3351 if (trans
->transaction
->state
== TRANS_STATE_COMMIT_DOING
)
3353 spin_unlock(&fs_info
->trans_lock
);
3356 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
3357 if (!btrfs_qgroup_enabled(fs_info
))
3360 quota_root
= fs_info
->quota_root
;
3366 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_SIMPLE
&& !inherit
) {
3367 ret
= qgroup_auto_inherit(fs_info
, inode_rootid
, &inherit
);
3370 free_inherit
= true;
3374 i_qgroups
= (u64
*)(inherit
+ 1);
3375 nums
= inherit
->num_qgroups
+ 2 * inherit
->num_ref_copies
+
3376 2 * inherit
->num_excl_copies
;
3377 for (int i
= 0; i
< nums
; i
++) {
3378 srcgroup
= find_qgroup_rb(fs_info
, *i_qgroups
);
3381 * Zero out invalid groups so we can ignore
3385 ((srcgroup
->qgroupid
>> 48) <= (objectid
>> 48)))
3393 * create a tracking group for the subvol itself
3395 ret
= add_qgroup_item(trans
, quota_root
, objectid
);
3400 * add qgroup to all inherited groups
3403 i_qgroups
= (u64
*)(inherit
+ 1);
3404 for (int i
= 0; i
< inherit
->num_qgroups
; i
++, i_qgroups
++) {
3405 if (*i_qgroups
== 0)
3407 ret
= add_qgroup_relation_item(trans
, objectid
,
3409 if (ret
&& ret
!= -EEXIST
)
3411 ret
= add_qgroup_relation_item(trans
, *i_qgroups
,
3413 if (ret
&& ret
!= -EEXIST
)
3418 qlist_prealloc
= kcalloc(inherit
->num_qgroups
,
3419 sizeof(struct btrfs_qgroup_list
*),
3421 if (!qlist_prealloc
) {
3425 for (int i
= 0; i
< inherit
->num_qgroups
; i
++) {
3426 qlist_prealloc
[i
] = kzalloc(sizeof(struct btrfs_qgroup_list
),
3428 if (!qlist_prealloc
[i
]) {
3435 spin_lock(&fs_info
->qgroup_lock
);
3437 dstgroup
= add_qgroup_rb(fs_info
, prealloc
, objectid
);
3440 if (inherit
&& inherit
->flags
& BTRFS_QGROUP_INHERIT_SET_LIMITS
) {
3441 dstgroup
->lim_flags
= inherit
->lim
.flags
;
3442 dstgroup
->max_rfer
= inherit
->lim
.max_rfer
;
3443 dstgroup
->max_excl
= inherit
->lim
.max_excl
;
3444 dstgroup
->rsv_rfer
= inherit
->lim
.rsv_rfer
;
3445 dstgroup
->rsv_excl
= inherit
->lim
.rsv_excl
;
3447 qgroup_dirty(fs_info
, dstgroup
);
3450 if (srcid
&& btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_FULL
) {
3451 srcgroup
= find_qgroup_rb(fs_info
, srcid
);
3456 * We call inherit after we clone the root in order to make sure
3457 * our counts don't go crazy, so at this point the only
3458 * difference between the two roots should be the root node.
3460 level_size
= fs_info
->nodesize
;
3461 dstgroup
->rfer
= srcgroup
->rfer
;
3462 dstgroup
->rfer_cmpr
= srcgroup
->rfer_cmpr
;
3463 dstgroup
->excl
= level_size
;
3464 dstgroup
->excl_cmpr
= level_size
;
3465 srcgroup
->excl
= level_size
;
3466 srcgroup
->excl_cmpr
= level_size
;
3468 /* inherit the limit info */
3469 dstgroup
->lim_flags
= srcgroup
->lim_flags
;
3470 dstgroup
->max_rfer
= srcgroup
->max_rfer
;
3471 dstgroup
->max_excl
= srcgroup
->max_excl
;
3472 dstgroup
->rsv_rfer
= srcgroup
->rsv_rfer
;
3473 dstgroup
->rsv_excl
= srcgroup
->rsv_excl
;
3475 qgroup_dirty(fs_info
, dstgroup
);
3476 qgroup_dirty(fs_info
, srcgroup
);
3479 * If the source qgroup has parent but the new one doesn't,
3480 * we need a full rescan.
3482 if (!inherit
&& !list_empty(&srcgroup
->groups
))
3489 i_qgroups
= (u64
*)(inherit
+ 1);
3490 for (int i
= 0; i
< inherit
->num_qgroups
; i
++) {
3492 ret
= add_relation_rb(fs_info
, qlist_prealloc
[i
], objectid
,
3494 qlist_prealloc
[i
] = NULL
;
3499 /* Check if we can do a quick inherit. */
3500 ret
= qgroup_snapshot_quick_inherit(fs_info
, srcid
, *i_qgroups
);
3510 for (int i
= 0; i
< inherit
->num_ref_copies
; i
++, i_qgroups
+= 2) {
3511 struct btrfs_qgroup
*src
;
3512 struct btrfs_qgroup
*dst
;
3514 if (!i_qgroups
[0] || !i_qgroups
[1])
3517 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
3518 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
3525 dst
->rfer
= src
->rfer
- level_size
;
3526 dst
->rfer_cmpr
= src
->rfer_cmpr
- level_size
;
3528 /* Manually tweaking numbers certainly needs a rescan */
3531 for (int i
= 0; i
< inherit
->num_excl_copies
; i
++, i_qgroups
+= 2) {
3532 struct btrfs_qgroup
*src
;
3533 struct btrfs_qgroup
*dst
;
3535 if (!i_qgroups
[0] || !i_qgroups
[1])
3538 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
3539 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
3546 dst
->excl
= src
->excl
+ level_size
;
3547 dst
->excl_cmpr
= src
->excl_cmpr
+ level_size
;
3552 spin_unlock(&fs_info
->qgroup_lock
);
3554 ret
= btrfs_sysfs_add_one_qgroup(fs_info
, dstgroup
);
3557 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
3559 qgroup_mark_inconsistent(fs_info
);
3560 if (qlist_prealloc
) {
3561 for (int i
= 0; i
< inherit
->num_qgroups
; i
++)
3562 kfree(qlist_prealloc
[i
]);
3563 kfree(qlist_prealloc
);
3571 static bool qgroup_check_limits(const struct btrfs_qgroup
*qg
, u64 num_bytes
)
3573 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_RFER
) &&
3574 qgroup_rsv_total(qg
) + (s64
)qg
->rfer
+ num_bytes
> qg
->max_rfer
)
3577 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_EXCL
) &&
3578 qgroup_rsv_total(qg
) + (s64
)qg
->excl
+ num_bytes
> qg
->max_excl
)
3584 static int qgroup_reserve(struct btrfs_root
*root
, u64 num_bytes
, bool enforce
,
3585 enum btrfs_qgroup_rsv_type type
)
3587 struct btrfs_qgroup
*qgroup
;
3588 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3589 u64 ref_root
= btrfs_root_id(root
);
3591 LIST_HEAD(qgroup_list
);
3593 if (!is_fstree(ref_root
))
3599 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE
, &fs_info
->flags
) &&
3600 capable(CAP_SYS_RESOURCE
))
3603 spin_lock(&fs_info
->qgroup_lock
);
3604 if (!fs_info
->quota_root
)
3607 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
3611 qgroup_iterator_add(&qgroup_list
, qgroup
);
3612 list_for_each_entry(qgroup
, &qgroup_list
, iterator
) {
3613 struct btrfs_qgroup_list
*glist
;
3615 if (enforce
&& !qgroup_check_limits(qgroup
, num_bytes
)) {
3620 list_for_each_entry(glist
, &qgroup
->groups
, next_group
)
3621 qgroup_iterator_add(&qgroup_list
, glist
->group
);
3626 * no limits exceeded, now record the reservation into all qgroups
3628 list_for_each_entry(qgroup
, &qgroup_list
, iterator
)
3629 qgroup_rsv_add(fs_info
, qgroup
, num_bytes
, type
);
3632 qgroup_iterator_clean(&qgroup_list
);
3633 spin_unlock(&fs_info
->qgroup_lock
);
3638 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
3641 * Will handle all higher level qgroup too.
3643 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3644 * This special case is only used for META_PERTRANS type.
3646 void btrfs_qgroup_free_refroot(struct btrfs_fs_info
*fs_info
,
3647 u64 ref_root
, u64 num_bytes
,
3648 enum btrfs_qgroup_rsv_type type
)
3650 struct btrfs_qgroup
*qgroup
;
3651 LIST_HEAD(qgroup_list
);
3653 if (!is_fstree(ref_root
))
3659 if (num_bytes
== (u64
)-1 && type
!= BTRFS_QGROUP_RSV_META_PERTRANS
) {
3660 WARN(1, "%s: Invalid type to free", __func__
);
3663 spin_lock(&fs_info
->qgroup_lock
);
3665 if (!fs_info
->quota_root
)
3668 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
3672 if (num_bytes
== (u64
)-1)
3674 * We're freeing all pertrans rsv, get reserved value from
3675 * level 0 qgroup as real num_bytes to free.
3677 num_bytes
= qgroup
->rsv
.values
[type
];
3679 qgroup_iterator_add(&qgroup_list
, qgroup
);
3680 list_for_each_entry(qgroup
, &qgroup_list
, iterator
) {
3681 struct btrfs_qgroup_list
*glist
;
3683 qgroup_rsv_release(fs_info
, qgroup
, num_bytes
, type
);
3684 list_for_each_entry(glist
, &qgroup
->groups
, next_group
) {
3685 qgroup_iterator_add(&qgroup_list
, glist
->group
);
3689 qgroup_iterator_clean(&qgroup_list
);
3690 spin_unlock(&fs_info
->qgroup_lock
);
3694 * Check if the leaf is the last leaf. Which means all node pointers
3695 * are at their last position.
3697 static bool is_last_leaf(struct btrfs_path
*path
)
3701 for (i
= 1; i
< BTRFS_MAX_LEVEL
&& path
->nodes
[i
]; i
++) {
3702 if (path
->slots
[i
] != btrfs_header_nritems(path
->nodes
[i
]) - 1)
3709 * returns < 0 on error, 0 when more leafs are to be scanned.
3710 * returns 1 when done.
3712 static int qgroup_rescan_leaf(struct btrfs_trans_handle
*trans
,
3713 struct btrfs_path
*path
)
3715 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3716 struct btrfs_root
*extent_root
;
3717 struct btrfs_key found
;
3718 struct extent_buffer
*scratch_leaf
= NULL
;
3724 if (!btrfs_qgroup_full_accounting(fs_info
))
3727 mutex_lock(&fs_info
->qgroup_rescan_lock
);
3728 extent_root
= btrfs_extent_root(fs_info
,
3729 fs_info
->qgroup_rescan_progress
.objectid
);
3730 ret
= btrfs_search_slot_for_read(extent_root
,
3731 &fs_info
->qgroup_rescan_progress
,
3734 btrfs_debug(fs_info
,
3735 "current progress key (%llu %u %llu), search_slot ret %d",
3736 fs_info
->qgroup_rescan_progress
.objectid
,
3737 fs_info
->qgroup_rescan_progress
.type
,
3738 fs_info
->qgroup_rescan_progress
.offset
, ret
);
3742 * The rescan is about to end, we will not be scanning any
3743 * further blocks. We cannot unset the RESCAN flag here, because
3744 * we want to commit the transaction if everything went well.
3745 * To make the live accounting work in this phase, we set our
3746 * scan progress pointer such that every real extent objectid
3749 fs_info
->qgroup_rescan_progress
.objectid
= (u64
)-1;
3750 btrfs_release_path(path
);
3751 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
3754 done
= is_last_leaf(path
);
3756 btrfs_item_key_to_cpu(path
->nodes
[0], &found
,
3757 btrfs_header_nritems(path
->nodes
[0]) - 1);
3758 fs_info
->qgroup_rescan_progress
.objectid
= found
.objectid
+ 1;
3760 scratch_leaf
= btrfs_clone_extent_buffer(path
->nodes
[0]);
3761 if (!scratch_leaf
) {
3763 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
3766 slot
= path
->slots
[0];
3767 btrfs_release_path(path
);
3768 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
3770 for (; slot
< btrfs_header_nritems(scratch_leaf
); ++slot
) {
3771 struct btrfs_backref_walk_ctx ctx
= { 0 };
3773 btrfs_item_key_to_cpu(scratch_leaf
, &found
, slot
);
3774 if (found
.type
!= BTRFS_EXTENT_ITEM_KEY
&&
3775 found
.type
!= BTRFS_METADATA_ITEM_KEY
)
3777 if (found
.type
== BTRFS_METADATA_ITEM_KEY
)
3778 num_bytes
= fs_info
->nodesize
;
3780 num_bytes
= found
.offset
;
3782 ctx
.bytenr
= found
.objectid
;
3783 ctx
.fs_info
= fs_info
;
3785 ret
= btrfs_find_all_roots(&ctx
, false);
3788 /* For rescan, just pass old_roots as NULL */
3789 ret
= btrfs_qgroup_account_extent(trans
, found
.objectid
,
3790 num_bytes
, NULL
, ctx
.roots
);
3796 free_extent_buffer(scratch_leaf
);
3800 fs_info
->qgroup_rescan_progress
.objectid
= (u64
)-1;
3805 static bool rescan_should_stop(struct btrfs_fs_info
*fs_info
)
3807 if (btrfs_fs_closing(fs_info
))
3809 if (test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
))
3811 if (!btrfs_qgroup_enabled(fs_info
))
3813 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN
)
3818 static void btrfs_qgroup_rescan_worker(struct btrfs_work
*work
)
3820 struct btrfs_fs_info
*fs_info
= container_of(work
, struct btrfs_fs_info
,
3821 qgroup_rescan_work
);
3822 struct btrfs_path
*path
;
3823 struct btrfs_trans_handle
*trans
= NULL
;
3825 bool stopped
= false;
3826 bool did_leaf_rescans
= false;
3828 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_SIMPLE
)
3831 path
= btrfs_alloc_path();
3837 * Rescan should only search for commit root, and any later difference
3838 * should be recorded by qgroup
3840 path
->search_commit_root
= 1;
3841 path
->skip_locking
= 1;
3843 while (!ret
&& !(stopped
= rescan_should_stop(fs_info
))) {
3844 trans
= btrfs_start_transaction(fs_info
->fs_root
, 0);
3845 if (IS_ERR(trans
)) {
3846 ret
= PTR_ERR(trans
);
3850 ret
= qgroup_rescan_leaf(trans
, path
);
3851 did_leaf_rescans
= true;
3854 btrfs_commit_transaction(trans
);
3856 btrfs_end_transaction(trans
);
3860 btrfs_free_path(path
);
3862 mutex_lock(&fs_info
->qgroup_rescan_lock
);
3864 fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
) {
3865 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
3866 } else if (ret
< 0 || stopped
) {
3867 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
3869 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
3872 * Only update status, since the previous part has already updated the
3873 * qgroup info, and only if we did any actual work. This also prevents
3874 * race with a concurrent quota disable, which has already set
3875 * fs_info->quota_root to NULL and cleared BTRFS_FS_QUOTA_ENABLED at
3876 * btrfs_quota_disable().
3878 if (did_leaf_rescans
) {
3879 trans
= btrfs_start_transaction(fs_info
->quota_root
, 1);
3880 if (IS_ERR(trans
)) {
3881 ret
= PTR_ERR(trans
);
3884 "fail to start transaction for status update: %d",
3891 mutex_lock(&fs_info
->qgroup_rescan_lock
);
3893 fs_info
->qgroup_flags
& BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN
)
3894 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
3896 int ret2
= update_qgroup_status_item(trans
);
3900 btrfs_err(fs_info
, "fail to update qgroup status: %d", ret
);
3903 fs_info
->qgroup_rescan_running
= false;
3904 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN
;
3905 complete_all(&fs_info
->qgroup_rescan_completion
);
3906 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
3911 btrfs_end_transaction(trans
);
3914 btrfs_info(fs_info
, "qgroup scan paused");
3915 } else if (fs_info
->qgroup_flags
& BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN
) {
3916 btrfs_info(fs_info
, "qgroup scan cancelled");
3917 } else if (ret
>= 0) {
3918 btrfs_info(fs_info
, "qgroup scan completed%s",
3919 ret
> 0 ? " (inconsistency flag cleared)" : "");
3921 btrfs_err(fs_info
, "qgroup scan failed with %d", ret
);
3926 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3927 * memory required for the rescan context.
3930 qgroup_rescan_init(struct btrfs_fs_info
*fs_info
, u64 progress_objectid
,
3935 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_SIMPLE
) {
3936 btrfs_warn(fs_info
, "qgroup rescan init failed, running in simple mode");
3941 /* we're resuming qgroup rescan at mount time */
3942 if (!(fs_info
->qgroup_flags
&
3943 BTRFS_QGROUP_STATUS_FLAG_RESCAN
)) {
3944 btrfs_debug(fs_info
,
3945 "qgroup rescan init failed, qgroup rescan is not queued");
3947 } else if (!(fs_info
->qgroup_flags
&
3948 BTRFS_QGROUP_STATUS_FLAG_ON
)) {
3949 btrfs_debug(fs_info
,
3950 "qgroup rescan init failed, qgroup is not enabled");
3958 mutex_lock(&fs_info
->qgroup_rescan_lock
);
3961 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
) {
3963 } else if (!(fs_info
->qgroup_flags
&
3964 BTRFS_QGROUP_STATUS_FLAG_ON
)) {
3965 btrfs_debug(fs_info
,
3966 "qgroup rescan init failed, qgroup is not enabled");
3968 } else if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_DISABLED
) {
3969 /* Quota disable is in progress */
3974 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
3977 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
3980 memset(&fs_info
->qgroup_rescan_progress
, 0,
3981 sizeof(fs_info
->qgroup_rescan_progress
));
3982 fs_info
->qgroup_flags
&= ~(BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN
|
3983 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING
);
3984 fs_info
->qgroup_rescan_progress
.objectid
= progress_objectid
;
3985 init_completion(&fs_info
->qgroup_rescan_completion
);
3986 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
3988 btrfs_init_work(&fs_info
->qgroup_rescan_work
,
3989 btrfs_qgroup_rescan_worker
, NULL
);
3994 qgroup_rescan_zero_tracking(struct btrfs_fs_info
*fs_info
)
3997 struct btrfs_qgroup
*qgroup
;
3999 spin_lock(&fs_info
->qgroup_lock
);
4000 /* clear all current qgroup tracking information */
4001 for (n
= rb_first(&fs_info
->qgroup_tree
); n
; n
= rb_next(n
)) {
4002 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
4004 qgroup
->rfer_cmpr
= 0;
4006 qgroup
->excl_cmpr
= 0;
4007 qgroup_dirty(fs_info
, qgroup
);
4009 spin_unlock(&fs_info
->qgroup_lock
);
4013 btrfs_qgroup_rescan(struct btrfs_fs_info
*fs_info
)
4017 ret
= qgroup_rescan_init(fs_info
, 0, 1);
4022 * We have set the rescan_progress to 0, which means no more
4023 * delayed refs will be accounted by btrfs_qgroup_account_ref.
4024 * However, btrfs_qgroup_account_ref may be right after its call
4025 * to btrfs_find_all_roots, in which case it would still do the
4027 * To solve this, we're committing the transaction, which will
4028 * ensure we run all delayed refs and only after that, we are
4029 * going to clear all tracking information for a clean start.
4032 ret
= btrfs_commit_current_transaction(fs_info
->fs_root
);
4034 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
4038 qgroup_rescan_zero_tracking(fs_info
);
4040 mutex_lock(&fs_info
->qgroup_rescan_lock
);
4041 fs_info
->qgroup_rescan_running
= true;
4042 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
4043 &fs_info
->qgroup_rescan_work
);
4044 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
4049 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info
*fs_info
,
4055 mutex_lock(&fs_info
->qgroup_rescan_lock
);
4056 running
= fs_info
->qgroup_rescan_running
;
4057 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
4063 ret
= wait_for_completion_interruptible(
4064 &fs_info
->qgroup_rescan_completion
);
4066 wait_for_completion(&fs_info
->qgroup_rescan_completion
);
4072 * this is only called from open_ctree where we're still single threaded, thus
4073 * locking is omitted here.
4076 btrfs_qgroup_rescan_resume(struct btrfs_fs_info
*fs_info
)
4078 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
) {
4079 mutex_lock(&fs_info
->qgroup_rescan_lock
);
4080 fs_info
->qgroup_rescan_running
= true;
4081 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
4082 &fs_info
->qgroup_rescan_work
);
4083 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
4087 #define rbtree_iterate_from_safe(node, next, start) \
4088 for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
4090 static int qgroup_unreserve_range(struct btrfs_inode
*inode
,
4091 struct extent_changeset
*reserved
, u64 start
,
4094 struct rb_node
*node
;
4095 struct rb_node
*next
;
4096 struct ulist_node
*entry
;
4099 node
= reserved
->range_changed
.root
.rb_node
;
4103 entry
= rb_entry(node
, struct ulist_node
, rb_node
);
4104 if (entry
->val
< start
)
4105 node
= node
->rb_right
;
4107 node
= node
->rb_left
;
4110 if (entry
->val
> start
&& rb_prev(&entry
->rb_node
))
4111 entry
= rb_entry(rb_prev(&entry
->rb_node
), struct ulist_node
,
4114 rbtree_iterate_from_safe(node
, next
, &entry
->rb_node
) {
4120 entry
= rb_entry(node
, struct ulist_node
, rb_node
);
4121 entry_start
= entry
->val
;
4122 entry_end
= entry
->aux
;
4123 entry_len
= entry_end
- entry_start
+ 1;
4125 if (entry_start
>= start
+ len
)
4127 if (entry_start
+ entry_len
<= start
)
4130 * Now the entry is in [start, start + len), revert the
4131 * EXTENT_QGROUP_RESERVED bit.
4133 clear_ret
= clear_extent_bits(&inode
->io_tree
, entry_start
,
4134 entry_end
, EXTENT_QGROUP_RESERVED
);
4135 if (!ret
&& clear_ret
< 0)
4138 ulist_del(&reserved
->range_changed
, entry
->val
, entry
->aux
);
4139 if (likely(reserved
->bytes_changed
>= entry_len
)) {
4140 reserved
->bytes_changed
-= entry_len
;
4143 reserved
->bytes_changed
= 0;
4151 * Try to free some space for qgroup.
4153 * For qgroup, there are only 3 ways to free qgroup space:
4154 * - Flush nodatacow write
4155 * Any nodatacow write will free its reserved data space at run_delalloc_range().
4156 * In theory, we should only flush nodatacow inodes, but it's not yet
4157 * possible, so we need to flush the whole root.
4159 * - Wait for ordered extents
4160 * When ordered extents are finished, their reserved metadata is finally
4161 * converted to per_trans status, which can be freed by later commit
4164 * - Commit transaction
4165 * This would free the meta_per_trans space.
4166 * In theory this shouldn't provide much space, but any more qgroup space
4169 static int try_flush_qgroup(struct btrfs_root
*root
)
4173 /* Can't hold an open transaction or we run the risk of deadlocking. */
4174 ASSERT(current
->journal_info
== NULL
);
4175 if (WARN_ON(current
->journal_info
))
4179 * We don't want to run flush again and again, so if there is a running
4180 * one, we won't try to start a new flush, but exit directly.
4182 if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING
, &root
->state
)) {
4183 wait_event(root
->qgroup_flush_wait
,
4184 !test_bit(BTRFS_ROOT_QGROUP_FLUSHING
, &root
->state
));
4188 btrfs_run_delayed_iputs(root
->fs_info
);
4189 btrfs_wait_on_delayed_iputs(root
->fs_info
);
4190 ret
= btrfs_start_delalloc_snapshot(root
, true);
4193 btrfs_wait_ordered_extents(root
, U64_MAX
, NULL
);
4195 ret
= btrfs_commit_current_transaction(root
);
4197 clear_bit(BTRFS_ROOT_QGROUP_FLUSHING
, &root
->state
);
4198 wake_up(&root
->qgroup_flush_wait
);
4202 static int qgroup_reserve_data(struct btrfs_inode
*inode
,
4203 struct extent_changeset
**reserved_ret
, u64 start
,
4206 struct btrfs_root
*root
= inode
->root
;
4207 struct extent_changeset
*reserved
;
4208 bool new_reserved
= false;
4213 if (btrfs_qgroup_mode(root
->fs_info
) == BTRFS_QGROUP_MODE_DISABLED
||
4214 !is_fstree(btrfs_root_id(root
)) || len
== 0)
4217 /* @reserved parameter is mandatory for qgroup */
4218 if (WARN_ON(!reserved_ret
))
4220 if (!*reserved_ret
) {
4221 new_reserved
= true;
4222 *reserved_ret
= extent_changeset_alloc();
4226 reserved
= *reserved_ret
;
4227 /* Record already reserved space */
4228 orig_reserved
= reserved
->bytes_changed
;
4229 ret
= set_record_extent_bits(&inode
->io_tree
, start
,
4230 start
+ len
-1, EXTENT_QGROUP_RESERVED
, reserved
);
4232 /* Newly reserved space */
4233 to_reserve
= reserved
->bytes_changed
- orig_reserved
;
4234 trace_btrfs_qgroup_reserve_data(&inode
->vfs_inode
, start
, len
,
4235 to_reserve
, QGROUP_RESERVE
);
4238 ret
= qgroup_reserve(root
, to_reserve
, true, BTRFS_QGROUP_RSV_DATA
);
4245 qgroup_unreserve_range(inode
, reserved
, start
, len
);
4248 extent_changeset_free(reserved
);
4249 *reserved_ret
= NULL
;
4255 * Reserve qgroup space for range [start, start + len).
4257 * This function will either reserve space from related qgroups or do nothing
4258 * if the range is already reserved.
4260 * Return 0 for successful reservation
4261 * Return <0 for error (including -EQUOT)
4263 * NOTE: This function may sleep for memory allocation, dirty page flushing and
4264 * commit transaction. So caller should not hold any dirty page locked.
4266 int btrfs_qgroup_reserve_data(struct btrfs_inode
*inode
,
4267 struct extent_changeset
**reserved_ret
, u64 start
,
4272 ret
= qgroup_reserve_data(inode
, reserved_ret
, start
, len
);
4273 if (ret
<= 0 && ret
!= -EDQUOT
)
4276 ret
= try_flush_qgroup(inode
->root
);
4279 return qgroup_reserve_data(inode
, reserved_ret
, start
, len
);
4282 /* Free ranges specified by @reserved, normally in error path */
4283 static int qgroup_free_reserved_data(struct btrfs_inode
*inode
,
4284 struct extent_changeset
*reserved
,
4285 u64 start
, u64 len
, u64
*freed_ret
)
4287 struct btrfs_root
*root
= inode
->root
;
4288 struct ulist_node
*unode
;
4289 struct ulist_iterator uiter
;
4290 struct extent_changeset changeset
;
4294 extent_changeset_init(&changeset
);
4295 len
= round_up(start
+ len
, root
->fs_info
->sectorsize
);
4296 start
= round_down(start
, root
->fs_info
->sectorsize
);
4298 ULIST_ITER_INIT(&uiter
);
4299 while ((unode
= ulist_next(&reserved
->range_changed
, &uiter
))) {
4300 u64 range_start
= unode
->val
;
4301 /* unode->aux is the inclusive end */
4302 u64 range_len
= unode
->aux
- range_start
+ 1;
4306 extent_changeset_release(&changeset
);
4308 /* Only free range in range [start, start + len) */
4309 if (range_start
>= start
+ len
||
4310 range_start
+ range_len
<= start
)
4312 free_start
= max(range_start
, start
);
4313 free_len
= min(start
+ len
, range_start
+ range_len
) -
4316 * TODO: To also modify reserved->ranges_reserved to reflect
4319 * However as long as we free qgroup reserved according to
4320 * EXTENT_QGROUP_RESERVED, we won't double free.
4321 * So not need to rush.
4323 ret
= clear_record_extent_bits(&inode
->io_tree
, free_start
,
4324 free_start
+ free_len
- 1,
4325 EXTENT_QGROUP_RESERVED
, &changeset
);
4328 freed
+= changeset
.bytes_changed
;
4330 btrfs_qgroup_free_refroot(root
->fs_info
, btrfs_root_id(root
), freed
,
4331 BTRFS_QGROUP_RSV_DATA
);
4336 extent_changeset_release(&changeset
);
4340 static int __btrfs_qgroup_release_data(struct btrfs_inode
*inode
,
4341 struct extent_changeset
*reserved
, u64 start
, u64 len
,
4342 u64
*released
, int free
)
4344 struct extent_changeset changeset
;
4345 int trace_op
= QGROUP_RELEASE
;
4348 if (btrfs_qgroup_mode(inode
->root
->fs_info
) == BTRFS_QGROUP_MODE_DISABLED
) {
4349 return clear_record_extent_bits(&inode
->io_tree
, start
,
4351 EXTENT_QGROUP_RESERVED
, NULL
);
4354 /* In release case, we shouldn't have @reserved */
4355 WARN_ON(!free
&& reserved
);
4356 if (free
&& reserved
)
4357 return qgroup_free_reserved_data(inode
, reserved
, start
, len
, released
);
4358 extent_changeset_init(&changeset
);
4359 ret
= clear_record_extent_bits(&inode
->io_tree
, start
, start
+ len
-1,
4360 EXTENT_QGROUP_RESERVED
, &changeset
);
4365 trace_op
= QGROUP_FREE
;
4366 trace_btrfs_qgroup_release_data(&inode
->vfs_inode
, start
, len
,
4367 changeset
.bytes_changed
, trace_op
);
4369 btrfs_qgroup_free_refroot(inode
->root
->fs_info
,
4370 btrfs_root_id(inode
->root
),
4371 changeset
.bytes_changed
, BTRFS_QGROUP_RSV_DATA
);
4373 *released
= changeset
.bytes_changed
;
4375 extent_changeset_release(&changeset
);
4380 * Free a reserved space range from io_tree and related qgroups
4382 * Should be called when a range of pages get invalidated before reaching disk.
4383 * Or for error cleanup case.
4384 * if @reserved is given, only reserved range in [@start, @start + @len) will
4387 * For data written to disk, use btrfs_qgroup_release_data().
4389 * NOTE: This function may sleep for memory allocation.
4391 int btrfs_qgroup_free_data(struct btrfs_inode
*inode
,
4392 struct extent_changeset
*reserved
,
4393 u64 start
, u64 len
, u64
*freed
)
4395 return __btrfs_qgroup_release_data(inode
, reserved
, start
, len
, freed
, 1);
4399 * Release a reserved space range from io_tree only.
4401 * Should be called when a range of pages get written to disk and corresponding
4402 * FILE_EXTENT is inserted into corresponding root.
4404 * Since new qgroup accounting framework will only update qgroup numbers at
4405 * commit_transaction() time, its reserved space shouldn't be freed from
4408 * But we should release the range from io_tree, to allow further write to be
4411 * NOTE: This function may sleep for memory allocation.
4413 int btrfs_qgroup_release_data(struct btrfs_inode
*inode
, u64 start
, u64 len
, u64
*released
)
4415 return __btrfs_qgroup_release_data(inode
, NULL
, start
, len
, released
, 0);
4418 static void add_root_meta_rsv(struct btrfs_root
*root
, int num_bytes
,
4419 enum btrfs_qgroup_rsv_type type
)
4421 if (type
!= BTRFS_QGROUP_RSV_META_PREALLOC
&&
4422 type
!= BTRFS_QGROUP_RSV_META_PERTRANS
)
4427 spin_lock(&root
->qgroup_meta_rsv_lock
);
4428 if (type
== BTRFS_QGROUP_RSV_META_PREALLOC
)
4429 root
->qgroup_meta_rsv_prealloc
+= num_bytes
;
4431 root
->qgroup_meta_rsv_pertrans
+= num_bytes
;
4432 spin_unlock(&root
->qgroup_meta_rsv_lock
);
4435 static int sub_root_meta_rsv(struct btrfs_root
*root
, int num_bytes
,
4436 enum btrfs_qgroup_rsv_type type
)
4438 if (type
!= BTRFS_QGROUP_RSV_META_PREALLOC
&&
4439 type
!= BTRFS_QGROUP_RSV_META_PERTRANS
)
4444 spin_lock(&root
->qgroup_meta_rsv_lock
);
4445 if (type
== BTRFS_QGROUP_RSV_META_PREALLOC
) {
4446 num_bytes
= min_t(u64
, root
->qgroup_meta_rsv_prealloc
,
4448 root
->qgroup_meta_rsv_prealloc
-= num_bytes
;
4450 num_bytes
= min_t(u64
, root
->qgroup_meta_rsv_pertrans
,
4452 root
->qgroup_meta_rsv_pertrans
-= num_bytes
;
4454 spin_unlock(&root
->qgroup_meta_rsv_lock
);
4458 int btrfs_qgroup_reserve_meta(struct btrfs_root
*root
, int num_bytes
,
4459 enum btrfs_qgroup_rsv_type type
, bool enforce
)
4461 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4464 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_DISABLED
||
4465 !is_fstree(btrfs_root_id(root
)) || num_bytes
== 0)
4468 BUG_ON(num_bytes
!= round_down(num_bytes
, fs_info
->nodesize
));
4469 trace_qgroup_meta_reserve(root
, (s64
)num_bytes
, type
);
4470 ret
= qgroup_reserve(root
, num_bytes
, enforce
, type
);
4474 * Record what we have reserved into root.
4476 * To avoid quota disabled->enabled underflow.
4477 * In that case, we may try to free space we haven't reserved
4478 * (since quota was disabled), so record what we reserved into root.
4479 * And ensure later release won't underflow this number.
4481 add_root_meta_rsv(root
, num_bytes
, type
);
4485 int __btrfs_qgroup_reserve_meta(struct btrfs_root
*root
, int num_bytes
,
4486 enum btrfs_qgroup_rsv_type type
, bool enforce
,
4491 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
, type
, enforce
);
4492 if ((ret
<= 0 && ret
!= -EDQUOT
) || noflush
)
4495 ret
= try_flush_qgroup(root
);
4498 return btrfs_qgroup_reserve_meta(root
, num_bytes
, type
, enforce
);
4502 * Per-transaction meta reservation should be all freed at transaction commit
4505 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root
*root
)
4507 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4509 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_DISABLED
||
4510 !is_fstree(btrfs_root_id(root
)))
4513 /* TODO: Update trace point to handle such free */
4514 trace_qgroup_meta_free_all_pertrans(root
);
4515 /* Special value -1 means to free all reserved space */
4516 btrfs_qgroup_free_refroot(fs_info
, btrfs_root_id(root
), (u64
)-1,
4517 BTRFS_QGROUP_RSV_META_PERTRANS
);
4520 void __btrfs_qgroup_free_meta(struct btrfs_root
*root
, int num_bytes
,
4521 enum btrfs_qgroup_rsv_type type
)
4523 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4525 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_DISABLED
||
4526 !is_fstree(btrfs_root_id(root
)))
4530 * reservation for META_PREALLOC can happen before quota is enabled,
4531 * which can lead to underflow.
4532 * Here ensure we will only free what we really have reserved.
4534 num_bytes
= sub_root_meta_rsv(root
, num_bytes
, type
);
4535 BUG_ON(num_bytes
!= round_down(num_bytes
, fs_info
->nodesize
));
4536 trace_qgroup_meta_reserve(root
, -(s64
)num_bytes
, type
);
4537 btrfs_qgroup_free_refroot(fs_info
, btrfs_root_id(root
), num_bytes
, type
);
4540 static void qgroup_convert_meta(struct btrfs_fs_info
*fs_info
, u64 ref_root
,
4543 struct btrfs_qgroup
*qgroup
;
4544 LIST_HEAD(qgroup_list
);
4548 if (!fs_info
->quota_root
)
4551 spin_lock(&fs_info
->qgroup_lock
);
4552 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
4556 qgroup_iterator_add(&qgroup_list
, qgroup
);
4557 list_for_each_entry(qgroup
, &qgroup_list
, iterator
) {
4558 struct btrfs_qgroup_list
*glist
;
4560 qgroup_rsv_release(fs_info
, qgroup
, num_bytes
,
4561 BTRFS_QGROUP_RSV_META_PREALLOC
);
4562 if (!sb_rdonly(fs_info
->sb
))
4563 qgroup_rsv_add(fs_info
, qgroup
, num_bytes
,
4564 BTRFS_QGROUP_RSV_META_PERTRANS
);
4566 list_for_each_entry(glist
, &qgroup
->groups
, next_group
)
4567 qgroup_iterator_add(&qgroup_list
, glist
->group
);
4570 qgroup_iterator_clean(&qgroup_list
);
4571 spin_unlock(&fs_info
->qgroup_lock
);
4575 * Convert @num_bytes of META_PREALLOCATED reservation to META_PERTRANS.
4577 * This is called when preallocated meta reservation needs to be used.
4578 * Normally after btrfs_join_transaction() call.
4580 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root
*root
, int num_bytes
)
4582 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4584 if (btrfs_qgroup_mode(fs_info
) == BTRFS_QGROUP_MODE_DISABLED
||
4585 !is_fstree(btrfs_root_id(root
)))
4587 /* Same as btrfs_qgroup_free_meta_prealloc() */
4588 num_bytes
= sub_root_meta_rsv(root
, num_bytes
,
4589 BTRFS_QGROUP_RSV_META_PREALLOC
);
4590 trace_qgroup_meta_convert(root
, num_bytes
);
4591 qgroup_convert_meta(fs_info
, btrfs_root_id(root
), num_bytes
);
4592 if (!sb_rdonly(fs_info
->sb
))
4593 add_root_meta_rsv(root
, num_bytes
, BTRFS_QGROUP_RSV_META_PERTRANS
);
4597 * Check qgroup reserved space leaking, normally at destroy inode
4600 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode
*inode
)
4602 struct extent_changeset changeset
;
4603 struct ulist_node
*unode
;
4604 struct ulist_iterator iter
;
4607 extent_changeset_init(&changeset
);
4608 ret
= clear_record_extent_bits(&inode
->io_tree
, 0, (u64
)-1,
4609 EXTENT_QGROUP_RESERVED
, &changeset
);
4612 if (WARN_ON(changeset
.bytes_changed
)) {
4613 ULIST_ITER_INIT(&iter
);
4614 while ((unode
= ulist_next(&changeset
.range_changed
, &iter
))) {
4615 btrfs_warn(inode
->root
->fs_info
,
4616 "leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4617 btrfs_ino(inode
), unode
->val
, unode
->aux
);
4619 btrfs_qgroup_free_refroot(inode
->root
->fs_info
,
4620 btrfs_root_id(inode
->root
),
4621 changeset
.bytes_changed
, BTRFS_QGROUP_RSV_DATA
);
4624 extent_changeset_release(&changeset
);
4627 void btrfs_qgroup_init_swapped_blocks(
4628 struct btrfs_qgroup_swapped_blocks
*swapped_blocks
)
4632 spin_lock_init(&swapped_blocks
->lock
);
4633 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++)
4634 swapped_blocks
->blocks
[i
] = RB_ROOT
;
4635 swapped_blocks
->swapped
= false;
4639 * Delete all swapped blocks record of @root.
4640 * Every record here means we skipped a full subtree scan for qgroup.
4642 * Gets called when committing one transaction.
4644 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root
*root
)
4646 struct btrfs_qgroup_swapped_blocks
*swapped_blocks
;
4649 swapped_blocks
= &root
->swapped_blocks
;
4651 spin_lock(&swapped_blocks
->lock
);
4652 if (!swapped_blocks
->swapped
)
4654 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
4655 struct rb_root
*cur_root
= &swapped_blocks
->blocks
[i
];
4656 struct btrfs_qgroup_swapped_block
*entry
;
4657 struct btrfs_qgroup_swapped_block
*next
;
4659 rbtree_postorder_for_each_entry_safe(entry
, next
, cur_root
,
4662 swapped_blocks
->blocks
[i
] = RB_ROOT
;
4664 swapped_blocks
->swapped
= false;
4666 spin_unlock(&swapped_blocks
->lock
);
4670 * Add subtree roots record into @subvol_root.
4672 * @subvol_root: tree root of the subvolume tree get swapped
4673 * @bg: block group under balance
4674 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
4675 * @reloc_parent/slot: pointer to the subtree root in reloc tree
4676 * BOTH POINTERS ARE BEFORE TREE SWAP
4677 * @last_snapshot: last snapshot generation of the subvolume tree
4679 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle
*trans
,
4680 struct btrfs_root
*subvol_root
,
4681 struct btrfs_block_group
*bg
,
4682 struct extent_buffer
*subvol_parent
, int subvol_slot
,
4683 struct extent_buffer
*reloc_parent
, int reloc_slot
,
4686 struct btrfs_fs_info
*fs_info
= subvol_root
->fs_info
;
4687 struct btrfs_qgroup_swapped_blocks
*blocks
= &subvol_root
->swapped_blocks
;
4688 struct btrfs_qgroup_swapped_block
*block
;
4689 struct rb_node
**cur
;
4690 struct rb_node
*parent
= NULL
;
4691 int level
= btrfs_header_level(subvol_parent
) - 1;
4694 if (!btrfs_qgroup_full_accounting(fs_info
))
4697 if (btrfs_node_ptr_generation(subvol_parent
, subvol_slot
) >
4698 btrfs_node_ptr_generation(reloc_parent
, reloc_slot
)) {
4699 btrfs_err_rl(fs_info
,
4700 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4702 btrfs_node_ptr_generation(subvol_parent
, subvol_slot
),
4703 btrfs_node_ptr_generation(reloc_parent
, reloc_slot
));
4707 block
= kmalloc(sizeof(*block
), GFP_NOFS
);
4714 * @reloc_parent/slot is still before swap, while @block is going to
4715 * record the bytenr after swap, so we do the swap here.
4717 block
->subvol_bytenr
= btrfs_node_blockptr(reloc_parent
, reloc_slot
);
4718 block
->subvol_generation
= btrfs_node_ptr_generation(reloc_parent
,
4720 block
->reloc_bytenr
= btrfs_node_blockptr(subvol_parent
, subvol_slot
);
4721 block
->reloc_generation
= btrfs_node_ptr_generation(subvol_parent
,
4723 block
->last_snapshot
= last_snapshot
;
4724 block
->level
= level
;
4727 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4728 * no one else can modify tree blocks thus we qgroup will not change
4729 * no matter the value of trace_leaf.
4731 if (bg
&& bg
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4732 block
->trace_leaf
= true;
4734 block
->trace_leaf
= false;
4735 btrfs_node_key_to_cpu(reloc_parent
, &block
->first_key
, reloc_slot
);
4737 /* Insert @block into @blocks */
4738 spin_lock(&blocks
->lock
);
4739 cur
= &blocks
->blocks
[level
].rb_node
;
4741 struct btrfs_qgroup_swapped_block
*entry
;
4744 entry
= rb_entry(parent
, struct btrfs_qgroup_swapped_block
,
4747 if (entry
->subvol_bytenr
< block
->subvol_bytenr
) {
4748 cur
= &(*cur
)->rb_left
;
4749 } else if (entry
->subvol_bytenr
> block
->subvol_bytenr
) {
4750 cur
= &(*cur
)->rb_right
;
4752 if (entry
->subvol_generation
!=
4753 block
->subvol_generation
||
4754 entry
->reloc_bytenr
!= block
->reloc_bytenr
||
4755 entry
->reloc_generation
!=
4756 block
->reloc_generation
) {
4758 * Duplicated but mismatch entry found.
4761 * Marking qgroup inconsistent should be enough
4764 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG
));
4771 rb_link_node(&block
->node
, parent
, cur
);
4772 rb_insert_color(&block
->node
, &blocks
->blocks
[level
]);
4773 blocks
->swapped
= true;
4775 spin_unlock(&blocks
->lock
);
4778 qgroup_mark_inconsistent(fs_info
);
4783 * Check if the tree block is a subtree root, and if so do the needed
4784 * delayed subtree trace for qgroup.
4786 * This is called during btrfs_cow_block().
4788 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle
*trans
,
4789 struct btrfs_root
*root
,
4790 struct extent_buffer
*subvol_eb
)
4792 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4793 struct btrfs_tree_parent_check check
= { 0 };
4794 struct btrfs_qgroup_swapped_blocks
*blocks
= &root
->swapped_blocks
;
4795 struct btrfs_qgroup_swapped_block
*block
;
4796 struct extent_buffer
*reloc_eb
= NULL
;
4797 struct rb_node
*node
;
4799 bool swapped
= false;
4800 int level
= btrfs_header_level(subvol_eb
);
4804 if (!btrfs_qgroup_full_accounting(fs_info
))
4806 if (!is_fstree(btrfs_root_id(root
)) || !root
->reloc_root
)
4809 spin_lock(&blocks
->lock
);
4810 if (!blocks
->swapped
) {
4811 spin_unlock(&blocks
->lock
);
4814 node
= blocks
->blocks
[level
].rb_node
;
4817 block
= rb_entry(node
, struct btrfs_qgroup_swapped_block
, node
);
4818 if (block
->subvol_bytenr
< subvol_eb
->start
) {
4819 node
= node
->rb_left
;
4820 } else if (block
->subvol_bytenr
> subvol_eb
->start
) {
4821 node
= node
->rb_right
;
4828 spin_unlock(&blocks
->lock
);
4831 /* Found one, remove it from @blocks first and update blocks->swapped */
4832 rb_erase(&block
->node
, &blocks
->blocks
[level
]);
4833 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
4834 if (RB_EMPTY_ROOT(&blocks
->blocks
[i
])) {
4839 blocks
->swapped
= swapped
;
4840 spin_unlock(&blocks
->lock
);
4842 check
.level
= block
->level
;
4843 check
.transid
= block
->reloc_generation
;
4844 check
.has_first_key
= true;
4845 memcpy(&check
.first_key
, &block
->first_key
, sizeof(check
.first_key
));
4847 /* Read out reloc subtree root */
4848 reloc_eb
= read_tree_block(fs_info
, block
->reloc_bytenr
, &check
);
4849 if (IS_ERR(reloc_eb
)) {
4850 ret
= PTR_ERR(reloc_eb
);
4854 if (!extent_buffer_uptodate(reloc_eb
)) {
4859 ret
= qgroup_trace_subtree_swap(trans
, reloc_eb
, subvol_eb
,
4860 block
->last_snapshot
, block
->trace_leaf
);
4863 free_extent_buffer(reloc_eb
);
4866 btrfs_err_rl(fs_info
,
4867 "failed to account subtree at bytenr %llu: %d",
4868 subvol_eb
->start
, ret
);
4869 qgroup_mark_inconsistent(fs_info
);
4874 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction
*trans
)
4876 struct btrfs_qgroup_extent_record
*entry
;
4877 unsigned long index
;
4879 xa_for_each(&trans
->delayed_refs
.dirty_extents
, index
, entry
) {
4880 ulist_free(entry
->old_roots
);
4883 xa_destroy(&trans
->delayed_refs
.dirty_extents
);
4886 void btrfs_free_squota_rsv(struct btrfs_fs_info
*fs_info
, u64 root
, u64 rsv_bytes
)
4888 if (btrfs_qgroup_mode(fs_info
) != BTRFS_QGROUP_MODE_SIMPLE
)
4891 if (!is_fstree(root
))
4894 btrfs_qgroup_free_refroot(fs_info
, root
, rsv_bytes
, BTRFS_QGROUP_RSV_DATA
);
4897 int btrfs_record_squota_delta(struct btrfs_fs_info
*fs_info
,
4898 const struct btrfs_squota_delta
*delta
)
4901 struct btrfs_qgroup
*qgroup
;
4902 struct btrfs_qgroup
*qg
;
4903 LIST_HEAD(qgroup_list
);
4904 u64 root
= delta
->root
;
4905 u64 num_bytes
= delta
->num_bytes
;
4906 const int sign
= (delta
->is_inc
? 1 : -1);
4908 if (btrfs_qgroup_mode(fs_info
) != BTRFS_QGROUP_MODE_SIMPLE
)
4911 if (!is_fstree(root
))
4914 /* If the extent predates enabling quotas, don't count it. */
4915 if (delta
->generation
< fs_info
->qgroup_enable_gen
)
4918 spin_lock(&fs_info
->qgroup_lock
);
4919 qgroup
= find_qgroup_rb(fs_info
, root
);
4926 qgroup_iterator_add(&qgroup_list
, qgroup
);
4927 list_for_each_entry(qg
, &qgroup_list
, iterator
) {
4928 struct btrfs_qgroup_list
*glist
;
4930 qg
->excl
+= num_bytes
* sign
;
4931 qg
->rfer
+= num_bytes
* sign
;
4932 qgroup_dirty(fs_info
, qg
);
4934 list_for_each_entry(glist
, &qg
->groups
, next_group
)
4935 qgroup_iterator_add(&qgroup_list
, glist
->group
);
4937 qgroup_iterator_clean(&qgroup_list
);
4940 spin_unlock(&fs_info
->qgroup_lock
);