2 * Copyright (C) 2011 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/rbtree.h>
24 #include <linux/slab.h>
25 #include <linux/workqueue.h>
26 #include <linux/btrfs.h>
29 #include "transaction.h"
34 #include "extent_io.h"
39 * - subvol delete -> delete when ref goes to 0? delete limits also?
43 * - copy also limits on subvol creation
45 * - caches fuer ulists
46 * - performance benchmarks
47 * - check all ioctl parameters
50 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup
*qg
, u64 seq
,
53 if (qg
->old_refcnt
< seq
)
55 qg
->old_refcnt
+= mod
;
58 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup
*qg
, u64 seq
,
61 if (qg
->new_refcnt
< seq
)
63 qg
->new_refcnt
+= mod
;
66 static inline u64
btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup
*qg
, u64 seq
)
68 if (qg
->old_refcnt
< seq
)
70 return qg
->old_refcnt
- seq
;
73 static inline u64
btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup
*qg
, u64 seq
)
75 if (qg
->new_refcnt
< seq
)
77 return qg
->new_refcnt
- seq
;
81 * glue structure to represent the relations between qgroups.
83 struct btrfs_qgroup_list
{
84 struct list_head next_group
;
85 struct list_head next_member
;
86 struct btrfs_qgroup
*group
;
87 struct btrfs_qgroup
*member
;
90 static inline u64
qgroup_to_aux(struct btrfs_qgroup
*qg
)
92 return (u64
)(uintptr_t)qg
;
95 static inline struct btrfs_qgroup
* unode_aux_to_qgroup(struct ulist_node
*n
)
97 return (struct btrfs_qgroup
*)(uintptr_t)n
->aux
;
101 qgroup_rescan_init(struct btrfs_fs_info
*fs_info
, u64 progress_objectid
,
103 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info
*fs_info
);
105 /* must be called with qgroup_ioctl_lock held */
106 static struct btrfs_qgroup
*find_qgroup_rb(struct btrfs_fs_info
*fs_info
,
109 struct rb_node
*n
= fs_info
->qgroup_tree
.rb_node
;
110 struct btrfs_qgroup
*qgroup
;
113 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
114 if (qgroup
->qgroupid
< qgroupid
)
116 else if (qgroup
->qgroupid
> qgroupid
)
124 /* must be called with qgroup_lock held */
125 static struct btrfs_qgroup
*add_qgroup_rb(struct btrfs_fs_info
*fs_info
,
128 struct rb_node
**p
= &fs_info
->qgroup_tree
.rb_node
;
129 struct rb_node
*parent
= NULL
;
130 struct btrfs_qgroup
*qgroup
;
134 qgroup
= rb_entry(parent
, struct btrfs_qgroup
, node
);
136 if (qgroup
->qgroupid
< qgroupid
)
138 else if (qgroup
->qgroupid
> qgroupid
)
144 qgroup
= kzalloc(sizeof(*qgroup
), GFP_ATOMIC
);
146 return ERR_PTR(-ENOMEM
);
148 qgroup
->qgroupid
= qgroupid
;
149 INIT_LIST_HEAD(&qgroup
->groups
);
150 INIT_LIST_HEAD(&qgroup
->members
);
151 INIT_LIST_HEAD(&qgroup
->dirty
);
153 rb_link_node(&qgroup
->node
, parent
, p
);
154 rb_insert_color(&qgroup
->node
, &fs_info
->qgroup_tree
);
159 static void __del_qgroup_rb(struct btrfs_qgroup
*qgroup
)
161 struct btrfs_qgroup_list
*list
;
163 list_del(&qgroup
->dirty
);
164 while (!list_empty(&qgroup
->groups
)) {
165 list
= list_first_entry(&qgroup
->groups
,
166 struct btrfs_qgroup_list
, next_group
);
167 list_del(&list
->next_group
);
168 list_del(&list
->next_member
);
172 while (!list_empty(&qgroup
->members
)) {
173 list
= list_first_entry(&qgroup
->members
,
174 struct btrfs_qgroup_list
, next_member
);
175 list_del(&list
->next_group
);
176 list_del(&list
->next_member
);
182 /* must be called with qgroup_lock held */
183 static int del_qgroup_rb(struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
185 struct btrfs_qgroup
*qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
190 rb_erase(&qgroup
->node
, &fs_info
->qgroup_tree
);
191 __del_qgroup_rb(qgroup
);
195 /* must be called with qgroup_lock held */
196 static int add_relation_rb(struct btrfs_fs_info
*fs_info
,
197 u64 memberid
, u64 parentid
)
199 struct btrfs_qgroup
*member
;
200 struct btrfs_qgroup
*parent
;
201 struct btrfs_qgroup_list
*list
;
203 member
= find_qgroup_rb(fs_info
, memberid
);
204 parent
= find_qgroup_rb(fs_info
, parentid
);
205 if (!member
|| !parent
)
208 list
= kzalloc(sizeof(*list
), GFP_ATOMIC
);
212 list
->group
= parent
;
213 list
->member
= member
;
214 list_add_tail(&list
->next_group
, &member
->groups
);
215 list_add_tail(&list
->next_member
, &parent
->members
);
220 /* must be called with qgroup_lock held */
221 static int del_relation_rb(struct btrfs_fs_info
*fs_info
,
222 u64 memberid
, u64 parentid
)
224 struct btrfs_qgroup
*member
;
225 struct btrfs_qgroup
*parent
;
226 struct btrfs_qgroup_list
*list
;
228 member
= find_qgroup_rb(fs_info
, memberid
);
229 parent
= find_qgroup_rb(fs_info
, parentid
);
230 if (!member
|| !parent
)
233 list_for_each_entry(list
, &member
->groups
, next_group
) {
234 if (list
->group
== parent
) {
235 list_del(&list
->next_group
);
236 list_del(&list
->next_member
);
244 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
245 int btrfs_verify_qgroup_counts(struct btrfs_fs_info
*fs_info
, u64 qgroupid
,
248 struct btrfs_qgroup
*qgroup
;
250 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
253 if (qgroup
->rfer
!= rfer
|| qgroup
->excl
!= excl
)
260 * The full config is read in one go, only called from open_ctree()
261 * It doesn't use any locking, as at this point we're still single-threaded
263 int btrfs_read_qgroup_config(struct btrfs_fs_info
*fs_info
)
265 struct btrfs_key key
;
266 struct btrfs_key found_key
;
267 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
268 struct btrfs_path
*path
= NULL
;
269 struct extent_buffer
*l
;
273 u64 rescan_progress
= 0;
275 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
))
278 fs_info
->qgroup_ulist
= ulist_alloc(GFP_KERNEL
);
279 if (!fs_info
->qgroup_ulist
) {
284 path
= btrfs_alloc_path();
290 /* default this to quota off, in case no status key is found */
291 fs_info
->qgroup_flags
= 0;
294 * pass 1: read status, all qgroup infos and limits
299 ret
= btrfs_search_slot_for_read(quota_root
, &key
, path
, 1, 1);
304 struct btrfs_qgroup
*qgroup
;
306 slot
= path
->slots
[0];
308 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
310 if (found_key
.type
== BTRFS_QGROUP_STATUS_KEY
) {
311 struct btrfs_qgroup_status_item
*ptr
;
313 ptr
= btrfs_item_ptr(l
, slot
,
314 struct btrfs_qgroup_status_item
);
316 if (btrfs_qgroup_status_version(l
, ptr
) !=
317 BTRFS_QGROUP_STATUS_VERSION
) {
319 "old qgroup version, quota disabled");
322 if (btrfs_qgroup_status_generation(l
, ptr
) !=
323 fs_info
->generation
) {
324 flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
326 "qgroup generation mismatch, marked as inconsistent");
328 fs_info
->qgroup_flags
= btrfs_qgroup_status_flags(l
,
330 rescan_progress
= btrfs_qgroup_status_rescan(l
, ptr
);
334 if (found_key
.type
!= BTRFS_QGROUP_INFO_KEY
&&
335 found_key
.type
!= BTRFS_QGROUP_LIMIT_KEY
)
338 qgroup
= find_qgroup_rb(fs_info
, found_key
.offset
);
339 if ((qgroup
&& found_key
.type
== BTRFS_QGROUP_INFO_KEY
) ||
340 (!qgroup
&& found_key
.type
== BTRFS_QGROUP_LIMIT_KEY
)) {
341 btrfs_err(fs_info
, "inconsistent qgroup config");
342 flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
345 qgroup
= add_qgroup_rb(fs_info
, found_key
.offset
);
346 if (IS_ERR(qgroup
)) {
347 ret
= PTR_ERR(qgroup
);
351 switch (found_key
.type
) {
352 case BTRFS_QGROUP_INFO_KEY
: {
353 struct btrfs_qgroup_info_item
*ptr
;
355 ptr
= btrfs_item_ptr(l
, slot
,
356 struct btrfs_qgroup_info_item
);
357 qgroup
->rfer
= btrfs_qgroup_info_rfer(l
, ptr
);
358 qgroup
->rfer_cmpr
= btrfs_qgroup_info_rfer_cmpr(l
, ptr
);
359 qgroup
->excl
= btrfs_qgroup_info_excl(l
, ptr
);
360 qgroup
->excl_cmpr
= btrfs_qgroup_info_excl_cmpr(l
, ptr
);
361 /* generation currently unused */
364 case BTRFS_QGROUP_LIMIT_KEY
: {
365 struct btrfs_qgroup_limit_item
*ptr
;
367 ptr
= btrfs_item_ptr(l
, slot
,
368 struct btrfs_qgroup_limit_item
);
369 qgroup
->lim_flags
= btrfs_qgroup_limit_flags(l
, ptr
);
370 qgroup
->max_rfer
= btrfs_qgroup_limit_max_rfer(l
, ptr
);
371 qgroup
->max_excl
= btrfs_qgroup_limit_max_excl(l
, ptr
);
372 qgroup
->rsv_rfer
= btrfs_qgroup_limit_rsv_rfer(l
, ptr
);
373 qgroup
->rsv_excl
= btrfs_qgroup_limit_rsv_excl(l
, ptr
);
378 ret
= btrfs_next_item(quota_root
, path
);
384 btrfs_release_path(path
);
387 * pass 2: read all qgroup relations
390 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
392 ret
= btrfs_search_slot_for_read(quota_root
, &key
, path
, 1, 0);
396 slot
= path
->slots
[0];
398 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
400 if (found_key
.type
!= BTRFS_QGROUP_RELATION_KEY
)
403 if (found_key
.objectid
> found_key
.offset
) {
404 /* parent <- member, not needed to build config */
405 /* FIXME should we omit the key completely? */
409 ret
= add_relation_rb(fs_info
, found_key
.objectid
,
411 if (ret
== -ENOENT
) {
413 "orphan qgroup relation 0x%llx->0x%llx",
414 found_key
.objectid
, found_key
.offset
);
415 ret
= 0; /* ignore the error */
420 ret
= btrfs_next_item(quota_root
, path
);
427 fs_info
->qgroup_flags
|= flags
;
428 if (!(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
))
429 clear_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
430 else if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
&&
432 ret
= qgroup_rescan_init(fs_info
, rescan_progress
, 0);
433 btrfs_free_path(path
);
436 ulist_free(fs_info
->qgroup_ulist
);
437 fs_info
->qgroup_ulist
= NULL
;
438 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
441 return ret
< 0 ? ret
: 0;
445 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
446 * first two are in single-threaded paths.And for the third one, we have set
447 * quota_root to be null with qgroup_lock held before, so it is safe to clean
448 * up the in-memory structures without qgroup_lock held.
450 void btrfs_free_qgroup_config(struct btrfs_fs_info
*fs_info
)
453 struct btrfs_qgroup
*qgroup
;
455 while ((n
= rb_first(&fs_info
->qgroup_tree
))) {
456 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
457 rb_erase(n
, &fs_info
->qgroup_tree
);
458 __del_qgroup_rb(qgroup
);
461 * we call btrfs_free_qgroup_config() when umounting
462 * filesystem and disabling quota, so we set qgroup_ulist
463 * to be null here to avoid double free.
465 ulist_free(fs_info
->qgroup_ulist
);
466 fs_info
->qgroup_ulist
= NULL
;
469 static int add_qgroup_relation_item(struct btrfs_trans_handle
*trans
,
470 struct btrfs_root
*quota_root
,
474 struct btrfs_path
*path
;
475 struct btrfs_key key
;
477 path
= btrfs_alloc_path();
482 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
485 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
, 0);
487 btrfs_mark_buffer_dirty(path
->nodes
[0]);
489 btrfs_free_path(path
);
493 static int del_qgroup_relation_item(struct btrfs_trans_handle
*trans
,
494 struct btrfs_root
*quota_root
,
498 struct btrfs_path
*path
;
499 struct btrfs_key key
;
501 path
= btrfs_alloc_path();
506 key
.type
= BTRFS_QGROUP_RELATION_KEY
;
509 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
518 ret
= btrfs_del_item(trans
, quota_root
, path
);
520 btrfs_free_path(path
);
524 static int add_qgroup_item(struct btrfs_trans_handle
*trans
,
525 struct btrfs_root
*quota_root
, u64 qgroupid
)
528 struct btrfs_path
*path
;
529 struct btrfs_qgroup_info_item
*qgroup_info
;
530 struct btrfs_qgroup_limit_item
*qgroup_limit
;
531 struct extent_buffer
*leaf
;
532 struct btrfs_key key
;
534 if (btrfs_is_testing(quota_root
->fs_info
))
537 path
= btrfs_alloc_path();
542 key
.type
= BTRFS_QGROUP_INFO_KEY
;
543 key
.offset
= qgroupid
;
546 * Avoid a transaction abort by catching -EEXIST here. In that
547 * case, we proceed by re-initializing the existing structure
551 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
552 sizeof(*qgroup_info
));
553 if (ret
&& ret
!= -EEXIST
)
556 leaf
= path
->nodes
[0];
557 qgroup_info
= btrfs_item_ptr(leaf
, path
->slots
[0],
558 struct btrfs_qgroup_info_item
);
559 btrfs_set_qgroup_info_generation(leaf
, qgroup_info
, trans
->transid
);
560 btrfs_set_qgroup_info_rfer(leaf
, qgroup_info
, 0);
561 btrfs_set_qgroup_info_rfer_cmpr(leaf
, qgroup_info
, 0);
562 btrfs_set_qgroup_info_excl(leaf
, qgroup_info
, 0);
563 btrfs_set_qgroup_info_excl_cmpr(leaf
, qgroup_info
, 0);
565 btrfs_mark_buffer_dirty(leaf
);
567 btrfs_release_path(path
);
569 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
570 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
571 sizeof(*qgroup_limit
));
572 if (ret
&& ret
!= -EEXIST
)
575 leaf
= path
->nodes
[0];
576 qgroup_limit
= btrfs_item_ptr(leaf
, path
->slots
[0],
577 struct btrfs_qgroup_limit_item
);
578 btrfs_set_qgroup_limit_flags(leaf
, qgroup_limit
, 0);
579 btrfs_set_qgroup_limit_max_rfer(leaf
, qgroup_limit
, 0);
580 btrfs_set_qgroup_limit_max_excl(leaf
, qgroup_limit
, 0);
581 btrfs_set_qgroup_limit_rsv_rfer(leaf
, qgroup_limit
, 0);
582 btrfs_set_qgroup_limit_rsv_excl(leaf
, qgroup_limit
, 0);
584 btrfs_mark_buffer_dirty(leaf
);
588 btrfs_free_path(path
);
592 static int del_qgroup_item(struct btrfs_trans_handle
*trans
,
593 struct btrfs_root
*quota_root
, u64 qgroupid
)
596 struct btrfs_path
*path
;
597 struct btrfs_key key
;
599 path
= btrfs_alloc_path();
604 key
.type
= BTRFS_QGROUP_INFO_KEY
;
605 key
.offset
= qgroupid
;
606 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
615 ret
= btrfs_del_item(trans
, quota_root
, path
);
619 btrfs_release_path(path
);
621 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
622 ret
= btrfs_search_slot(trans
, quota_root
, &key
, path
, -1, 1);
631 ret
= btrfs_del_item(trans
, quota_root
, path
);
634 btrfs_free_path(path
);
638 static int update_qgroup_limit_item(struct btrfs_trans_handle
*trans
,
639 struct btrfs_root
*root
,
640 struct btrfs_qgroup
*qgroup
)
642 struct btrfs_path
*path
;
643 struct btrfs_key key
;
644 struct extent_buffer
*l
;
645 struct btrfs_qgroup_limit_item
*qgroup_limit
;
650 key
.type
= BTRFS_QGROUP_LIMIT_KEY
;
651 key
.offset
= qgroup
->qgroupid
;
653 path
= btrfs_alloc_path();
657 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
665 slot
= path
->slots
[0];
666 qgroup_limit
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_limit_item
);
667 btrfs_set_qgroup_limit_flags(l
, qgroup_limit
, qgroup
->lim_flags
);
668 btrfs_set_qgroup_limit_max_rfer(l
, qgroup_limit
, qgroup
->max_rfer
);
669 btrfs_set_qgroup_limit_max_excl(l
, qgroup_limit
, qgroup
->max_excl
);
670 btrfs_set_qgroup_limit_rsv_rfer(l
, qgroup_limit
, qgroup
->rsv_rfer
);
671 btrfs_set_qgroup_limit_rsv_excl(l
, qgroup_limit
, qgroup
->rsv_excl
);
673 btrfs_mark_buffer_dirty(l
);
676 btrfs_free_path(path
);
680 static int update_qgroup_info_item(struct btrfs_trans_handle
*trans
,
681 struct btrfs_root
*root
,
682 struct btrfs_qgroup
*qgroup
)
684 struct btrfs_path
*path
;
685 struct btrfs_key key
;
686 struct extent_buffer
*l
;
687 struct btrfs_qgroup_info_item
*qgroup_info
;
691 if (btrfs_is_testing(root
->fs_info
))
695 key
.type
= BTRFS_QGROUP_INFO_KEY
;
696 key
.offset
= qgroup
->qgroupid
;
698 path
= btrfs_alloc_path();
702 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
710 slot
= path
->slots
[0];
711 qgroup_info
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_info_item
);
712 btrfs_set_qgroup_info_generation(l
, qgroup_info
, trans
->transid
);
713 btrfs_set_qgroup_info_rfer(l
, qgroup_info
, qgroup
->rfer
);
714 btrfs_set_qgroup_info_rfer_cmpr(l
, qgroup_info
, qgroup
->rfer_cmpr
);
715 btrfs_set_qgroup_info_excl(l
, qgroup_info
, qgroup
->excl
);
716 btrfs_set_qgroup_info_excl_cmpr(l
, qgroup_info
, qgroup
->excl_cmpr
);
718 btrfs_mark_buffer_dirty(l
);
721 btrfs_free_path(path
);
725 static int update_qgroup_status_item(struct btrfs_trans_handle
*trans
,
726 struct btrfs_fs_info
*fs_info
,
727 struct btrfs_root
*root
)
729 struct btrfs_path
*path
;
730 struct btrfs_key key
;
731 struct extent_buffer
*l
;
732 struct btrfs_qgroup_status_item
*ptr
;
737 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
740 path
= btrfs_alloc_path();
744 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
752 slot
= path
->slots
[0];
753 ptr
= btrfs_item_ptr(l
, slot
, struct btrfs_qgroup_status_item
);
754 btrfs_set_qgroup_status_flags(l
, ptr
, fs_info
->qgroup_flags
);
755 btrfs_set_qgroup_status_generation(l
, ptr
, trans
->transid
);
756 btrfs_set_qgroup_status_rescan(l
, ptr
,
757 fs_info
->qgroup_rescan_progress
.objectid
);
759 btrfs_mark_buffer_dirty(l
);
762 btrfs_free_path(path
);
767 * called with qgroup_lock held
769 static int btrfs_clean_quota_tree(struct btrfs_trans_handle
*trans
,
770 struct btrfs_root
*root
)
772 struct btrfs_path
*path
;
773 struct btrfs_key key
;
774 struct extent_buffer
*leaf
= NULL
;
778 path
= btrfs_alloc_path();
782 path
->leave_spinning
= 1;
789 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
792 leaf
= path
->nodes
[0];
793 nr
= btrfs_header_nritems(leaf
);
797 * delete the leaf one by one
798 * since the whole tree is going
802 ret
= btrfs_del_items(trans
, root
, path
, 0, nr
);
806 btrfs_release_path(path
);
810 set_bit(BTRFS_FS_QUOTA_DISABLING
, &root
->fs_info
->flags
);
811 btrfs_free_path(path
);
815 int btrfs_quota_enable(struct btrfs_trans_handle
*trans
,
816 struct btrfs_fs_info
*fs_info
)
818 struct btrfs_root
*quota_root
;
819 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
820 struct btrfs_path
*path
= NULL
;
821 struct btrfs_qgroup_status_item
*ptr
;
822 struct extent_buffer
*leaf
;
823 struct btrfs_key key
;
824 struct btrfs_key found_key
;
825 struct btrfs_qgroup
*qgroup
= NULL
;
829 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
830 if (fs_info
->quota_root
) {
831 set_bit(BTRFS_FS_QUOTA_ENABLING
, &fs_info
->flags
);
835 fs_info
->qgroup_ulist
= ulist_alloc(GFP_KERNEL
);
836 if (!fs_info
->qgroup_ulist
) {
842 * initially create the quota tree
844 quota_root
= btrfs_create_tree(trans
, fs_info
,
845 BTRFS_QUOTA_TREE_OBJECTID
);
846 if (IS_ERR(quota_root
)) {
847 ret
= PTR_ERR(quota_root
);
851 path
= btrfs_alloc_path();
858 key
.type
= BTRFS_QGROUP_STATUS_KEY
;
861 ret
= btrfs_insert_empty_item(trans
, quota_root
, path
, &key
,
866 leaf
= path
->nodes
[0];
867 ptr
= btrfs_item_ptr(leaf
, path
->slots
[0],
868 struct btrfs_qgroup_status_item
);
869 btrfs_set_qgroup_status_generation(leaf
, ptr
, trans
->transid
);
870 btrfs_set_qgroup_status_version(leaf
, ptr
, BTRFS_QGROUP_STATUS_VERSION
);
871 fs_info
->qgroup_flags
= BTRFS_QGROUP_STATUS_FLAG_ON
|
872 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
873 btrfs_set_qgroup_status_flags(leaf
, ptr
, fs_info
->qgroup_flags
);
874 btrfs_set_qgroup_status_rescan(leaf
, ptr
, 0);
876 btrfs_mark_buffer_dirty(leaf
);
879 key
.type
= BTRFS_ROOT_REF_KEY
;
882 btrfs_release_path(path
);
883 ret
= btrfs_search_slot_for_read(tree_root
, &key
, path
, 1, 0);
891 slot
= path
->slots
[0];
892 leaf
= path
->nodes
[0];
893 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
895 if (found_key
.type
== BTRFS_ROOT_REF_KEY
) {
896 ret
= add_qgroup_item(trans
, quota_root
,
901 qgroup
= add_qgroup_rb(fs_info
, found_key
.offset
);
902 if (IS_ERR(qgroup
)) {
903 ret
= PTR_ERR(qgroup
);
907 ret
= btrfs_next_item(tree_root
, path
);
915 btrfs_release_path(path
);
916 ret
= add_qgroup_item(trans
, quota_root
, BTRFS_FS_TREE_OBJECTID
);
920 qgroup
= add_qgroup_rb(fs_info
, BTRFS_FS_TREE_OBJECTID
);
921 if (IS_ERR(qgroup
)) {
922 ret
= PTR_ERR(qgroup
);
925 spin_lock(&fs_info
->qgroup_lock
);
926 fs_info
->quota_root
= quota_root
;
927 set_bit(BTRFS_FS_QUOTA_ENABLING
, &fs_info
->flags
);
928 spin_unlock(&fs_info
->qgroup_lock
);
930 btrfs_free_path(path
);
933 free_extent_buffer(quota_root
->node
);
934 free_extent_buffer(quota_root
->commit_root
);
939 ulist_free(fs_info
->qgroup_ulist
);
940 fs_info
->qgroup_ulist
= NULL
;
942 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
946 int btrfs_quota_disable(struct btrfs_trans_handle
*trans
,
947 struct btrfs_fs_info
*fs_info
)
949 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
950 struct btrfs_root
*quota_root
;
953 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
954 if (!fs_info
->quota_root
)
956 clear_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
957 set_bit(BTRFS_FS_QUOTA_DISABLING
, &fs_info
->flags
);
958 btrfs_qgroup_wait_for_completion(fs_info
, false);
959 spin_lock(&fs_info
->qgroup_lock
);
960 quota_root
= fs_info
->quota_root
;
961 fs_info
->quota_root
= NULL
;
962 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_ON
;
963 spin_unlock(&fs_info
->qgroup_lock
);
965 btrfs_free_qgroup_config(fs_info
);
967 ret
= btrfs_clean_quota_tree(trans
, quota_root
);
971 ret
= btrfs_del_root(trans
, tree_root
, "a_root
->root_key
);
975 list_del("a_root
->dirty_list
);
977 btrfs_tree_lock(quota_root
->node
);
978 clean_tree_block(fs_info
, quota_root
->node
);
979 btrfs_tree_unlock(quota_root
->node
);
980 btrfs_free_tree_block(trans
, quota_root
, quota_root
->node
, 0, 1);
982 free_extent_buffer(quota_root
->node
);
983 free_extent_buffer(quota_root
->commit_root
);
986 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
990 static void qgroup_dirty(struct btrfs_fs_info
*fs_info
,
991 struct btrfs_qgroup
*qgroup
)
993 if (list_empty(&qgroup
->dirty
))
994 list_add(&qgroup
->dirty
, &fs_info
->dirty_qgroups
);
997 static void report_reserved_underflow(struct btrfs_fs_info
*fs_info
,
998 struct btrfs_qgroup
*qgroup
,
1001 #ifdef CONFIG_BTRFS_DEBUG
1002 WARN_ON(qgroup
->reserved
< num_bytes
);
1003 btrfs_debug(fs_info
,
1004 "qgroup %llu reserved space underflow, have: %llu, to free: %llu",
1005 qgroup
->qgroupid
, qgroup
->reserved
, num_bytes
);
1007 qgroup
->reserved
= 0;
1010 * The easy accounting, if we are adding/removing the only ref for an extent
1011 * then this qgroup and all of the parent qgroups get their reference and
1012 * exclusive counts adjusted.
1014 * Caller should hold fs_info->qgroup_lock.
1016 static int __qgroup_excl_accounting(struct btrfs_fs_info
*fs_info
,
1017 struct ulist
*tmp
, u64 ref_root
,
1018 u64 num_bytes
, int sign
)
1020 struct btrfs_qgroup
*qgroup
;
1021 struct btrfs_qgroup_list
*glist
;
1022 struct ulist_node
*unode
;
1023 struct ulist_iterator uiter
;
1026 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
1030 qgroup
->rfer
+= sign
* num_bytes
;
1031 qgroup
->rfer_cmpr
+= sign
* num_bytes
;
1033 WARN_ON(sign
< 0 && qgroup
->excl
< num_bytes
);
1034 qgroup
->excl
+= sign
* num_bytes
;
1035 qgroup
->excl_cmpr
+= sign
* num_bytes
;
1037 trace_qgroup_update_reserve(fs_info
, qgroup
, -(s64
)num_bytes
);
1038 if (qgroup
->reserved
< num_bytes
)
1039 report_reserved_underflow(fs_info
, qgroup
, num_bytes
);
1041 qgroup
->reserved
-= num_bytes
;
1044 qgroup_dirty(fs_info
, qgroup
);
1046 /* Get all of the parent groups that contain this qgroup */
1047 list_for_each_entry(glist
, &qgroup
->groups
, next_group
) {
1048 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1049 qgroup_to_aux(glist
->group
), GFP_ATOMIC
);
1054 /* Iterate all of the parents and adjust their reference counts */
1055 ULIST_ITER_INIT(&uiter
);
1056 while ((unode
= ulist_next(tmp
, &uiter
))) {
1057 qgroup
= unode_aux_to_qgroup(unode
);
1058 qgroup
->rfer
+= sign
* num_bytes
;
1059 qgroup
->rfer_cmpr
+= sign
* num_bytes
;
1060 WARN_ON(sign
< 0 && qgroup
->excl
< num_bytes
);
1061 qgroup
->excl
+= sign
* num_bytes
;
1063 trace_qgroup_update_reserve(fs_info
, qgroup
,
1065 if (qgroup
->reserved
< num_bytes
)
1066 report_reserved_underflow(fs_info
, qgroup
,
1069 qgroup
->reserved
-= num_bytes
;
1071 qgroup
->excl_cmpr
+= sign
* num_bytes
;
1072 qgroup_dirty(fs_info
, qgroup
);
1074 /* Add any parents of the parents */
1075 list_for_each_entry(glist
, &qgroup
->groups
, next_group
) {
1076 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1077 qgroup_to_aux(glist
->group
), GFP_ATOMIC
);
1089 * Quick path for updating qgroup with only excl refs.
1091 * In that case, just update all parent will be enough.
1092 * Or we needs to do a full rescan.
1093 * Caller should also hold fs_info->qgroup_lock.
1095 * Return 0 for quick update, return >0 for need to full rescan
1096 * and mark INCONSISTENT flag.
1097 * Return < 0 for other error.
1099 static int quick_update_accounting(struct btrfs_fs_info
*fs_info
,
1100 struct ulist
*tmp
, u64 src
, u64 dst
,
1103 struct btrfs_qgroup
*qgroup
;
1107 qgroup
= find_qgroup_rb(fs_info
, src
);
1110 if (qgroup
->excl
== qgroup
->rfer
) {
1112 err
= __qgroup_excl_accounting(fs_info
, tmp
, dst
,
1113 qgroup
->excl
, sign
);
1121 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1125 int btrfs_add_qgroup_relation(struct btrfs_trans_handle
*trans
,
1126 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
1128 struct btrfs_root
*quota_root
;
1129 struct btrfs_qgroup
*parent
;
1130 struct btrfs_qgroup
*member
;
1131 struct btrfs_qgroup_list
*list
;
1135 /* Check the level of src and dst first */
1136 if (btrfs_qgroup_level(src
) >= btrfs_qgroup_level(dst
))
1139 tmp
= ulist_alloc(GFP_KERNEL
);
1143 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1144 quota_root
= fs_info
->quota_root
;
1149 member
= find_qgroup_rb(fs_info
, src
);
1150 parent
= find_qgroup_rb(fs_info
, dst
);
1151 if (!member
|| !parent
) {
1156 /* check if such qgroup relation exist firstly */
1157 list_for_each_entry(list
, &member
->groups
, next_group
) {
1158 if (list
->group
== parent
) {
1164 ret
= add_qgroup_relation_item(trans
, quota_root
, src
, dst
);
1168 ret
= add_qgroup_relation_item(trans
, quota_root
, dst
, src
);
1170 del_qgroup_relation_item(trans
, quota_root
, src
, dst
);
1174 spin_lock(&fs_info
->qgroup_lock
);
1175 ret
= add_relation_rb(fs_info
, src
, dst
);
1177 spin_unlock(&fs_info
->qgroup_lock
);
1180 ret
= quick_update_accounting(fs_info
, tmp
, src
, dst
, 1);
1181 spin_unlock(&fs_info
->qgroup_lock
);
1183 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1188 static int __del_qgroup_relation(struct btrfs_trans_handle
*trans
,
1189 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
1191 struct btrfs_root
*quota_root
;
1192 struct btrfs_qgroup
*parent
;
1193 struct btrfs_qgroup
*member
;
1194 struct btrfs_qgroup_list
*list
;
1199 tmp
= ulist_alloc(GFP_KERNEL
);
1203 quota_root
= fs_info
->quota_root
;
1209 member
= find_qgroup_rb(fs_info
, src
);
1210 parent
= find_qgroup_rb(fs_info
, dst
);
1211 if (!member
|| !parent
) {
1216 /* check if such qgroup relation exist firstly */
1217 list_for_each_entry(list
, &member
->groups
, next_group
) {
1218 if (list
->group
== parent
)
1224 ret
= del_qgroup_relation_item(trans
, quota_root
, src
, dst
);
1225 err
= del_qgroup_relation_item(trans
, quota_root
, dst
, src
);
1229 spin_lock(&fs_info
->qgroup_lock
);
1230 del_relation_rb(fs_info
, src
, dst
);
1231 ret
= quick_update_accounting(fs_info
, tmp
, src
, dst
, -1);
1232 spin_unlock(&fs_info
->qgroup_lock
);
1238 int btrfs_del_qgroup_relation(struct btrfs_trans_handle
*trans
,
1239 struct btrfs_fs_info
*fs_info
, u64 src
, u64 dst
)
1243 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1244 ret
= __del_qgroup_relation(trans
, fs_info
, src
, dst
);
1245 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1250 int btrfs_create_qgroup(struct btrfs_trans_handle
*trans
,
1251 struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
1253 struct btrfs_root
*quota_root
;
1254 struct btrfs_qgroup
*qgroup
;
1257 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1258 quota_root
= fs_info
->quota_root
;
1263 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1269 ret
= add_qgroup_item(trans
, quota_root
, qgroupid
);
1273 spin_lock(&fs_info
->qgroup_lock
);
1274 qgroup
= add_qgroup_rb(fs_info
, qgroupid
);
1275 spin_unlock(&fs_info
->qgroup_lock
);
1278 ret
= PTR_ERR(qgroup
);
1280 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1284 int btrfs_remove_qgroup(struct btrfs_trans_handle
*trans
,
1285 struct btrfs_fs_info
*fs_info
, u64 qgroupid
)
1287 struct btrfs_root
*quota_root
;
1288 struct btrfs_qgroup
*qgroup
;
1289 struct btrfs_qgroup_list
*list
;
1292 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1293 quota_root
= fs_info
->quota_root
;
1299 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1304 /* check if there are no children of this qgroup */
1305 if (!list_empty(&qgroup
->members
)) {
1310 ret
= del_qgroup_item(trans
, quota_root
, qgroupid
);
1312 while (!list_empty(&qgroup
->groups
)) {
1313 list
= list_first_entry(&qgroup
->groups
,
1314 struct btrfs_qgroup_list
, next_group
);
1315 ret
= __del_qgroup_relation(trans
, fs_info
,
1317 list
->group
->qgroupid
);
1322 spin_lock(&fs_info
->qgroup_lock
);
1323 del_qgroup_rb(fs_info
, qgroupid
);
1324 spin_unlock(&fs_info
->qgroup_lock
);
1326 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1330 int btrfs_limit_qgroup(struct btrfs_trans_handle
*trans
,
1331 struct btrfs_fs_info
*fs_info
, u64 qgroupid
,
1332 struct btrfs_qgroup_limit
*limit
)
1334 struct btrfs_root
*quota_root
;
1335 struct btrfs_qgroup
*qgroup
;
1337 /* Sometimes we would want to clear the limit on this qgroup.
1338 * To meet this requirement, we treat the -1 as a special value
1339 * which tell kernel to clear the limit on this qgroup.
1341 const u64 CLEAR_VALUE
= -1;
1343 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
1344 quota_root
= fs_info
->quota_root
;
1350 qgroup
= find_qgroup_rb(fs_info
, qgroupid
);
1356 spin_lock(&fs_info
->qgroup_lock
);
1357 if (limit
->flags
& BTRFS_QGROUP_LIMIT_MAX_RFER
) {
1358 if (limit
->max_rfer
== CLEAR_VALUE
) {
1359 qgroup
->lim_flags
&= ~BTRFS_QGROUP_LIMIT_MAX_RFER
;
1360 limit
->flags
&= ~BTRFS_QGROUP_LIMIT_MAX_RFER
;
1361 qgroup
->max_rfer
= 0;
1363 qgroup
->max_rfer
= limit
->max_rfer
;
1366 if (limit
->flags
& BTRFS_QGROUP_LIMIT_MAX_EXCL
) {
1367 if (limit
->max_excl
== CLEAR_VALUE
) {
1368 qgroup
->lim_flags
&= ~BTRFS_QGROUP_LIMIT_MAX_EXCL
;
1369 limit
->flags
&= ~BTRFS_QGROUP_LIMIT_MAX_EXCL
;
1370 qgroup
->max_excl
= 0;
1372 qgroup
->max_excl
= limit
->max_excl
;
1375 if (limit
->flags
& BTRFS_QGROUP_LIMIT_RSV_RFER
) {
1376 if (limit
->rsv_rfer
== CLEAR_VALUE
) {
1377 qgroup
->lim_flags
&= ~BTRFS_QGROUP_LIMIT_RSV_RFER
;
1378 limit
->flags
&= ~BTRFS_QGROUP_LIMIT_RSV_RFER
;
1379 qgroup
->rsv_rfer
= 0;
1381 qgroup
->rsv_rfer
= limit
->rsv_rfer
;
1384 if (limit
->flags
& BTRFS_QGROUP_LIMIT_RSV_EXCL
) {
1385 if (limit
->rsv_excl
== CLEAR_VALUE
) {
1386 qgroup
->lim_flags
&= ~BTRFS_QGROUP_LIMIT_RSV_EXCL
;
1387 limit
->flags
&= ~BTRFS_QGROUP_LIMIT_RSV_EXCL
;
1388 qgroup
->rsv_excl
= 0;
1390 qgroup
->rsv_excl
= limit
->rsv_excl
;
1393 qgroup
->lim_flags
|= limit
->flags
;
1395 spin_unlock(&fs_info
->qgroup_lock
);
1397 ret
= update_qgroup_limit_item(trans
, quota_root
, qgroup
);
1399 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
1400 btrfs_info(fs_info
, "unable to update quota limit for %llu",
1405 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
1409 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle
*trans
,
1410 struct btrfs_fs_info
*fs_info
)
1412 struct btrfs_qgroup_extent_record
*record
;
1413 struct btrfs_delayed_ref_root
*delayed_refs
;
1414 struct rb_node
*node
;
1418 delayed_refs
= &trans
->transaction
->delayed_refs
;
1419 qgroup_to_skip
= delayed_refs
->qgroup_to_skip
;
1422 * No need to do lock, since this function will only be called in
1423 * btrfs_commit_transaction().
1425 node
= rb_first(&delayed_refs
->dirty_extent_root
);
1427 record
= rb_entry(node
, struct btrfs_qgroup_extent_record
,
1429 if (WARN_ON(!record
->old_roots
))
1430 ret
= btrfs_find_all_roots(NULL
, fs_info
,
1431 record
->bytenr
, 0, &record
->old_roots
);
1435 ulist_del(record
->old_roots
, qgroup_to_skip
, 0);
1436 node
= rb_next(node
);
1441 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info
*fs_info
,
1442 struct btrfs_delayed_ref_root
*delayed_refs
,
1443 struct btrfs_qgroup_extent_record
*record
)
1445 struct rb_node
**p
= &delayed_refs
->dirty_extent_root
.rb_node
;
1446 struct rb_node
*parent_node
= NULL
;
1447 struct btrfs_qgroup_extent_record
*entry
;
1448 u64 bytenr
= record
->bytenr
;
1450 assert_spin_locked(&delayed_refs
->lock
);
1451 trace_btrfs_qgroup_trace_extent(fs_info
, record
);
1455 entry
= rb_entry(parent_node
, struct btrfs_qgroup_extent_record
,
1457 if (bytenr
< entry
->bytenr
)
1459 else if (bytenr
> entry
->bytenr
)
1460 p
= &(*p
)->rb_right
;
1465 rb_link_node(&record
->node
, parent_node
, p
);
1466 rb_insert_color(&record
->node
, &delayed_refs
->dirty_extent_root
);
1470 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info
*fs_info
,
1471 struct btrfs_qgroup_extent_record
*qrecord
)
1473 struct ulist
*old_root
;
1474 u64 bytenr
= qrecord
->bytenr
;
1477 ret
= btrfs_find_all_roots(NULL
, fs_info
, bytenr
, 0, &old_root
);
1482 * Here we don't need to get the lock of
1483 * trans->transaction->delayed_refs, since inserted qrecord won't
1484 * be deleted, only qrecord->node may be modified (new qrecord insert)
1486 * So modifying qrecord->old_roots is safe here
1488 qrecord
->old_roots
= old_root
;
1492 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle
*trans
,
1493 struct btrfs_fs_info
*fs_info
, u64 bytenr
, u64 num_bytes
,
1496 struct btrfs_qgroup_extent_record
*record
;
1497 struct btrfs_delayed_ref_root
*delayed_refs
;
1500 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)
1501 || bytenr
== 0 || num_bytes
== 0)
1503 if (WARN_ON(trans
== NULL
))
1505 record
= kmalloc(sizeof(*record
), gfp_flag
);
1509 delayed_refs
= &trans
->transaction
->delayed_refs
;
1510 record
->bytenr
= bytenr
;
1511 record
->num_bytes
= num_bytes
;
1512 record
->old_roots
= NULL
;
1514 spin_lock(&delayed_refs
->lock
);
1515 ret
= btrfs_qgroup_trace_extent_nolock(fs_info
, delayed_refs
, record
);
1516 spin_unlock(&delayed_refs
->lock
);
1521 return btrfs_qgroup_trace_extent_post(fs_info
, record
);
1524 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle
*trans
,
1525 struct btrfs_fs_info
*fs_info
,
1526 struct extent_buffer
*eb
)
1528 int nr
= btrfs_header_nritems(eb
);
1529 int i
, extent_type
, ret
;
1530 struct btrfs_key key
;
1531 struct btrfs_file_extent_item
*fi
;
1532 u64 bytenr
, num_bytes
;
1534 /* We can be called directly from walk_up_proc() */
1535 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
))
1538 for (i
= 0; i
< nr
; i
++) {
1539 btrfs_item_key_to_cpu(eb
, &key
, i
);
1541 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
1544 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
1545 /* filter out non qgroup-accountable extents */
1546 extent_type
= btrfs_file_extent_type(eb
, fi
);
1548 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
1551 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
1555 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
1557 ret
= btrfs_qgroup_trace_extent(trans
, fs_info
, bytenr
,
1558 num_bytes
, GFP_NOFS
);
1566 * Walk up the tree from the bottom, freeing leaves and any interior
1567 * nodes which have had all slots visited. If a node (leaf or
1568 * interior) is freed, the node above it will have it's slot
1569 * incremented. The root node will never be freed.
1571 * At the end of this function, we should have a path which has all
1572 * slots incremented to the next position for a search. If we need to
1573 * read a new node it will be NULL and the node above it will have the
1574 * correct slot selected for a later read.
1576 * If we increment the root nodes slot counter past the number of
1577 * elements, 1 is returned to signal completion of the search.
1579 static int adjust_slots_upwards(struct btrfs_path
*path
, int root_level
)
1583 struct extent_buffer
*eb
;
1585 if (root_level
== 0)
1588 while (level
<= root_level
) {
1589 eb
= path
->nodes
[level
];
1590 nr
= btrfs_header_nritems(eb
);
1591 path
->slots
[level
]++;
1592 slot
= path
->slots
[level
];
1593 if (slot
>= nr
|| level
== 0) {
1595 * Don't free the root - we will detect this
1596 * condition after our loop and return a
1597 * positive value for caller to stop walking the tree.
1599 if (level
!= root_level
) {
1600 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
1601 path
->locks
[level
] = 0;
1603 free_extent_buffer(eb
);
1604 path
->nodes
[level
] = NULL
;
1605 path
->slots
[level
] = 0;
1609 * We have a valid slot to walk back down
1610 * from. Stop here so caller can process these
1619 eb
= path
->nodes
[root_level
];
1620 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
1626 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle
*trans
,
1627 struct btrfs_root
*root
,
1628 struct extent_buffer
*root_eb
,
1629 u64 root_gen
, int root_level
)
1631 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1634 struct extent_buffer
*eb
= root_eb
;
1635 struct btrfs_path
*path
= NULL
;
1637 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
1638 BUG_ON(root_eb
== NULL
);
1640 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
))
1643 if (!extent_buffer_uptodate(root_eb
)) {
1644 ret
= btrfs_read_buffer(root_eb
, root_gen
);
1649 if (root_level
== 0) {
1650 ret
= btrfs_qgroup_trace_leaf_items(trans
, fs_info
, root_eb
);
1654 path
= btrfs_alloc_path();
1659 * Walk down the tree. Missing extent blocks are filled in as
1660 * we go. Metadata is accounted every time we read a new
1663 * When we reach a leaf, we account for file extent items in it,
1664 * walk back up the tree (adjusting slot pointers as we go)
1665 * and restart the search process.
1667 extent_buffer_get(root_eb
); /* For path */
1668 path
->nodes
[root_level
] = root_eb
;
1669 path
->slots
[root_level
] = 0;
1670 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
1673 while (level
>= 0) {
1674 if (path
->nodes
[level
] == NULL
) {
1680 * We need to get child blockptr/gen from parent before
1683 eb
= path
->nodes
[level
+ 1];
1684 parent_slot
= path
->slots
[level
+ 1];
1685 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
1686 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
1688 eb
= read_tree_block(fs_info
, child_bytenr
, child_gen
);
1692 } else if (!extent_buffer_uptodate(eb
)) {
1693 free_extent_buffer(eb
);
1698 path
->nodes
[level
] = eb
;
1699 path
->slots
[level
] = 0;
1701 btrfs_tree_read_lock(eb
);
1702 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
1703 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
1705 ret
= btrfs_qgroup_trace_extent(trans
, fs_info
,
1714 ret
= btrfs_qgroup_trace_leaf_items(trans
,fs_info
,
1715 path
->nodes
[level
]);
1719 /* Nonzero return here means we completed our search */
1720 ret
= adjust_slots_upwards(path
, root_level
);
1724 /* Restart search with new slots */
1733 btrfs_free_path(path
);
1738 #define UPDATE_NEW 0
1739 #define UPDATE_OLD 1
1741 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1743 static int qgroup_update_refcnt(struct btrfs_fs_info
*fs_info
,
1744 struct ulist
*roots
, struct ulist
*tmp
,
1745 struct ulist
*qgroups
, u64 seq
, int update_old
)
1747 struct ulist_node
*unode
;
1748 struct ulist_iterator uiter
;
1749 struct ulist_node
*tmp_unode
;
1750 struct ulist_iterator tmp_uiter
;
1751 struct btrfs_qgroup
*qg
;
1756 ULIST_ITER_INIT(&uiter
);
1757 while ((unode
= ulist_next(roots
, &uiter
))) {
1758 qg
= find_qgroup_rb(fs_info
, unode
->val
);
1763 ret
= ulist_add(qgroups
, qg
->qgroupid
, qgroup_to_aux(qg
),
1767 ret
= ulist_add(tmp
, qg
->qgroupid
, qgroup_to_aux(qg
), GFP_ATOMIC
);
1770 ULIST_ITER_INIT(&tmp_uiter
);
1771 while ((tmp_unode
= ulist_next(tmp
, &tmp_uiter
))) {
1772 struct btrfs_qgroup_list
*glist
;
1774 qg
= unode_aux_to_qgroup(tmp_unode
);
1776 btrfs_qgroup_update_old_refcnt(qg
, seq
, 1);
1778 btrfs_qgroup_update_new_refcnt(qg
, seq
, 1);
1779 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
1780 ret
= ulist_add(qgroups
, glist
->group
->qgroupid
,
1781 qgroup_to_aux(glist
->group
),
1785 ret
= ulist_add(tmp
, glist
->group
->qgroupid
,
1786 qgroup_to_aux(glist
->group
),
1797 * Update qgroup rfer/excl counters.
1798 * Rfer update is easy, codes can explain themselves.
1800 * Excl update is tricky, the update is split into 2 part.
1801 * Part 1: Possible exclusive <-> sharing detect:
1803 * -------------------------------------
1805 * -------------------------------------
1807 * -------------------------------------
1810 * A: cur_old_roots < nr_old_roots (not exclusive before)
1811 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1812 * B: cur_new_roots < nr_new_roots (not exclusive now)
1813 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
1816 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1817 * *: Definitely not changed. **: Possible unchanged.
1819 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1821 * To make the logic clear, we first use condition A and B to split
1822 * combination into 4 results.
1824 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1825 * only on variant maybe 0.
1827 * Lastly, check result **, since there are 2 variants maybe 0, split them
1829 * But this time we don't need to consider other things, the codes and logic
1830 * is easy to understand now.
1832 static int qgroup_update_counters(struct btrfs_fs_info
*fs_info
,
1833 struct ulist
*qgroups
,
1836 u64 num_bytes
, u64 seq
)
1838 struct ulist_node
*unode
;
1839 struct ulist_iterator uiter
;
1840 struct btrfs_qgroup
*qg
;
1841 u64 cur_new_count
, cur_old_count
;
1843 ULIST_ITER_INIT(&uiter
);
1844 while ((unode
= ulist_next(qgroups
, &uiter
))) {
1847 qg
= unode_aux_to_qgroup(unode
);
1848 cur_old_count
= btrfs_qgroup_get_old_refcnt(qg
, seq
);
1849 cur_new_count
= btrfs_qgroup_get_new_refcnt(qg
, seq
);
1851 trace_qgroup_update_counters(fs_info
, qg
->qgroupid
,
1852 cur_old_count
, cur_new_count
);
1854 /* Rfer update part */
1855 if (cur_old_count
== 0 && cur_new_count
> 0) {
1856 qg
->rfer
+= num_bytes
;
1857 qg
->rfer_cmpr
+= num_bytes
;
1860 if (cur_old_count
> 0 && cur_new_count
== 0) {
1861 qg
->rfer
-= num_bytes
;
1862 qg
->rfer_cmpr
-= num_bytes
;
1866 /* Excl update part */
1867 /* Exclusive/none -> shared case */
1868 if (cur_old_count
== nr_old_roots
&&
1869 cur_new_count
< nr_new_roots
) {
1870 /* Exclusive -> shared */
1871 if (cur_old_count
!= 0) {
1872 qg
->excl
-= num_bytes
;
1873 qg
->excl_cmpr
-= num_bytes
;
1878 /* Shared -> exclusive/none case */
1879 if (cur_old_count
< nr_old_roots
&&
1880 cur_new_count
== nr_new_roots
) {
1881 /* Shared->exclusive */
1882 if (cur_new_count
!= 0) {
1883 qg
->excl
+= num_bytes
;
1884 qg
->excl_cmpr
+= num_bytes
;
1889 /* Exclusive/none -> exclusive/none case */
1890 if (cur_old_count
== nr_old_roots
&&
1891 cur_new_count
== nr_new_roots
) {
1892 if (cur_old_count
== 0) {
1893 /* None -> exclusive/none */
1895 if (cur_new_count
!= 0) {
1896 /* None -> exclusive */
1897 qg
->excl
+= num_bytes
;
1898 qg
->excl_cmpr
+= num_bytes
;
1901 /* None -> none, nothing changed */
1903 /* Exclusive -> exclusive/none */
1905 if (cur_new_count
== 0) {
1906 /* Exclusive -> none */
1907 qg
->excl
-= num_bytes
;
1908 qg
->excl_cmpr
-= num_bytes
;
1911 /* Exclusive -> exclusive, nothing changed */
1916 qgroup_dirty(fs_info
, qg
);
1922 btrfs_qgroup_account_extent(struct btrfs_trans_handle
*trans
,
1923 struct btrfs_fs_info
*fs_info
,
1924 u64 bytenr
, u64 num_bytes
,
1925 struct ulist
*old_roots
, struct ulist
*new_roots
)
1927 struct ulist
*qgroups
= NULL
;
1928 struct ulist
*tmp
= NULL
;
1930 u64 nr_new_roots
= 0;
1931 u64 nr_old_roots
= 0;
1934 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
))
1938 nr_new_roots
= new_roots
->nnodes
;
1940 nr_old_roots
= old_roots
->nnodes
;
1942 BUG_ON(!fs_info
->quota_root
);
1944 trace_btrfs_qgroup_account_extent(fs_info
, bytenr
, num_bytes
,
1945 nr_old_roots
, nr_new_roots
);
1947 qgroups
= ulist_alloc(GFP_NOFS
);
1952 tmp
= ulist_alloc(GFP_NOFS
);
1958 mutex_lock(&fs_info
->qgroup_rescan_lock
);
1959 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
) {
1960 if (fs_info
->qgroup_rescan_progress
.objectid
<= bytenr
) {
1961 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
1966 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
1968 spin_lock(&fs_info
->qgroup_lock
);
1969 seq
= fs_info
->qgroup_seq
;
1971 /* Update old refcnts using old_roots */
1972 ret
= qgroup_update_refcnt(fs_info
, old_roots
, tmp
, qgroups
, seq
,
1977 /* Update new refcnts using new_roots */
1978 ret
= qgroup_update_refcnt(fs_info
, new_roots
, tmp
, qgroups
, seq
,
1983 qgroup_update_counters(fs_info
, qgroups
, nr_old_roots
, nr_new_roots
,
1987 * Bump qgroup_seq to avoid seq overlap
1989 fs_info
->qgroup_seq
+= max(nr_old_roots
, nr_new_roots
) + 1;
1991 spin_unlock(&fs_info
->qgroup_lock
);
1994 ulist_free(qgroups
);
1995 ulist_free(old_roots
);
1996 ulist_free(new_roots
);
2000 int btrfs_qgroup_account_extents(struct btrfs_trans_handle
*trans
,
2001 struct btrfs_fs_info
*fs_info
)
2003 struct btrfs_qgroup_extent_record
*record
;
2004 struct btrfs_delayed_ref_root
*delayed_refs
;
2005 struct ulist
*new_roots
= NULL
;
2006 struct rb_node
*node
;
2010 delayed_refs
= &trans
->transaction
->delayed_refs
;
2011 qgroup_to_skip
= delayed_refs
->qgroup_to_skip
;
2012 while ((node
= rb_first(&delayed_refs
->dirty_extent_root
))) {
2013 record
= rb_entry(node
, struct btrfs_qgroup_extent_record
,
2016 trace_btrfs_qgroup_account_extents(fs_info
, record
);
2020 * Use SEQ_LAST as time_seq to do special search, which
2021 * doesn't lock tree or delayed_refs and search current
2022 * root. It's safe inside commit_transaction().
2024 ret
= btrfs_find_all_roots(trans
, fs_info
,
2025 record
->bytenr
, SEQ_LAST
, &new_roots
);
2029 ulist_del(new_roots
, qgroup_to_skip
, 0);
2030 ret
= btrfs_qgroup_account_extent(trans
, fs_info
,
2031 record
->bytenr
, record
->num_bytes
,
2032 record
->old_roots
, new_roots
);
2033 record
->old_roots
= NULL
;
2037 ulist_free(record
->old_roots
);
2038 ulist_free(new_roots
);
2040 rb_erase(node
, &delayed_refs
->dirty_extent_root
);
2048 * called from commit_transaction. Writes all changed qgroups to disk.
2050 int btrfs_run_qgroups(struct btrfs_trans_handle
*trans
,
2051 struct btrfs_fs_info
*fs_info
)
2053 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
2055 int start_rescan_worker
= 0;
2060 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
) &&
2061 test_bit(BTRFS_FS_QUOTA_ENABLING
, &fs_info
->flags
))
2062 start_rescan_worker
= 1;
2064 if (test_and_clear_bit(BTRFS_FS_QUOTA_ENABLING
, &fs_info
->flags
))
2065 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
2066 if (test_and_clear_bit(BTRFS_FS_QUOTA_DISABLING
, &fs_info
->flags
))
2067 clear_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
2069 spin_lock(&fs_info
->qgroup_lock
);
2070 while (!list_empty(&fs_info
->dirty_qgroups
)) {
2071 struct btrfs_qgroup
*qgroup
;
2072 qgroup
= list_first_entry(&fs_info
->dirty_qgroups
,
2073 struct btrfs_qgroup
, dirty
);
2074 list_del_init(&qgroup
->dirty
);
2075 spin_unlock(&fs_info
->qgroup_lock
);
2076 ret
= update_qgroup_info_item(trans
, quota_root
, qgroup
);
2078 fs_info
->qgroup_flags
|=
2079 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2080 ret
= update_qgroup_limit_item(trans
, quota_root
, qgroup
);
2082 fs_info
->qgroup_flags
|=
2083 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2084 spin_lock(&fs_info
->qgroup_lock
);
2086 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
))
2087 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_ON
;
2089 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_ON
;
2090 spin_unlock(&fs_info
->qgroup_lock
);
2092 ret
= update_qgroup_status_item(trans
, fs_info
, quota_root
);
2094 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2096 if (!ret
&& start_rescan_worker
) {
2097 ret
= qgroup_rescan_init(fs_info
, 0, 1);
2099 qgroup_rescan_zero_tracking(fs_info
);
2100 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
2101 &fs_info
->qgroup_rescan_work
);
2112 * Copy the accounting information between qgroups. This is necessary
2113 * when a snapshot or a subvolume is created. Throwing an error will
2114 * cause a transaction abort so we take extra care here to only error
2115 * when a readonly fs is a reasonable outcome.
2117 int btrfs_qgroup_inherit(struct btrfs_trans_handle
*trans
,
2118 struct btrfs_fs_info
*fs_info
, u64 srcid
, u64 objectid
,
2119 struct btrfs_qgroup_inherit
*inherit
)
2124 struct btrfs_root
*quota_root
= fs_info
->quota_root
;
2125 struct btrfs_qgroup
*srcgroup
;
2126 struct btrfs_qgroup
*dstgroup
;
2130 mutex_lock(&fs_info
->qgroup_ioctl_lock
);
2131 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
))
2140 i_qgroups
= (u64
*)(inherit
+ 1);
2141 nums
= inherit
->num_qgroups
+ 2 * inherit
->num_ref_copies
+
2142 2 * inherit
->num_excl_copies
;
2143 for (i
= 0; i
< nums
; ++i
) {
2144 srcgroup
= find_qgroup_rb(fs_info
, *i_qgroups
);
2147 * Zero out invalid groups so we can ignore
2151 ((srcgroup
->qgroupid
>> 48) <= (objectid
>> 48)))
2159 * create a tracking group for the subvol itself
2161 ret
= add_qgroup_item(trans
, quota_root
, objectid
);
2166 struct btrfs_root
*srcroot
;
2167 struct btrfs_key srckey
;
2169 srckey
.objectid
= srcid
;
2170 srckey
.type
= BTRFS_ROOT_ITEM_KEY
;
2171 srckey
.offset
= (u64
)-1;
2172 srcroot
= btrfs_read_fs_root_no_name(fs_info
, &srckey
);
2173 if (IS_ERR(srcroot
)) {
2174 ret
= PTR_ERR(srcroot
);
2178 level_size
= fs_info
->nodesize
;
2182 * add qgroup to all inherited groups
2185 i_qgroups
= (u64
*)(inherit
+ 1);
2186 for (i
= 0; i
< inherit
->num_qgroups
; ++i
, ++i_qgroups
) {
2187 if (*i_qgroups
== 0)
2189 ret
= add_qgroup_relation_item(trans
, quota_root
,
2190 objectid
, *i_qgroups
);
2191 if (ret
&& ret
!= -EEXIST
)
2193 ret
= add_qgroup_relation_item(trans
, quota_root
,
2194 *i_qgroups
, objectid
);
2195 if (ret
&& ret
!= -EEXIST
)
2202 spin_lock(&fs_info
->qgroup_lock
);
2204 dstgroup
= add_qgroup_rb(fs_info
, objectid
);
2205 if (IS_ERR(dstgroup
)) {
2206 ret
= PTR_ERR(dstgroup
);
2210 if (inherit
&& inherit
->flags
& BTRFS_QGROUP_INHERIT_SET_LIMITS
) {
2211 dstgroup
->lim_flags
= inherit
->lim
.flags
;
2212 dstgroup
->max_rfer
= inherit
->lim
.max_rfer
;
2213 dstgroup
->max_excl
= inherit
->lim
.max_excl
;
2214 dstgroup
->rsv_rfer
= inherit
->lim
.rsv_rfer
;
2215 dstgroup
->rsv_excl
= inherit
->lim
.rsv_excl
;
2217 ret
= update_qgroup_limit_item(trans
, quota_root
, dstgroup
);
2219 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2221 "unable to update quota limit for %llu",
2222 dstgroup
->qgroupid
);
2228 srcgroup
= find_qgroup_rb(fs_info
, srcid
);
2233 * We call inherit after we clone the root in order to make sure
2234 * our counts don't go crazy, so at this point the only
2235 * difference between the two roots should be the root node.
2237 dstgroup
->rfer
= srcgroup
->rfer
;
2238 dstgroup
->rfer_cmpr
= srcgroup
->rfer_cmpr
;
2239 dstgroup
->excl
= level_size
;
2240 dstgroup
->excl_cmpr
= level_size
;
2241 srcgroup
->excl
= level_size
;
2242 srcgroup
->excl_cmpr
= level_size
;
2244 /* inherit the limit info */
2245 dstgroup
->lim_flags
= srcgroup
->lim_flags
;
2246 dstgroup
->max_rfer
= srcgroup
->max_rfer
;
2247 dstgroup
->max_excl
= srcgroup
->max_excl
;
2248 dstgroup
->rsv_rfer
= srcgroup
->rsv_rfer
;
2249 dstgroup
->rsv_excl
= srcgroup
->rsv_excl
;
2251 qgroup_dirty(fs_info
, dstgroup
);
2252 qgroup_dirty(fs_info
, srcgroup
);
2258 i_qgroups
= (u64
*)(inherit
+ 1);
2259 for (i
= 0; i
< inherit
->num_qgroups
; ++i
) {
2261 ret
= add_relation_rb(fs_info
, objectid
, *i_qgroups
);
2268 for (i
= 0; i
< inherit
->num_ref_copies
; ++i
, i_qgroups
+= 2) {
2269 struct btrfs_qgroup
*src
;
2270 struct btrfs_qgroup
*dst
;
2272 if (!i_qgroups
[0] || !i_qgroups
[1])
2275 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
2276 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
2283 dst
->rfer
= src
->rfer
- level_size
;
2284 dst
->rfer_cmpr
= src
->rfer_cmpr
- level_size
;
2286 for (i
= 0; i
< inherit
->num_excl_copies
; ++i
, i_qgroups
+= 2) {
2287 struct btrfs_qgroup
*src
;
2288 struct btrfs_qgroup
*dst
;
2290 if (!i_qgroups
[0] || !i_qgroups
[1])
2293 src
= find_qgroup_rb(fs_info
, i_qgroups
[0]);
2294 dst
= find_qgroup_rb(fs_info
, i_qgroups
[1]);
2301 dst
->excl
= src
->excl
+ level_size
;
2302 dst
->excl_cmpr
= src
->excl_cmpr
+ level_size
;
2306 spin_unlock(&fs_info
->qgroup_lock
);
2308 mutex_unlock(&fs_info
->qgroup_ioctl_lock
);
2312 static bool qgroup_check_limits(const struct btrfs_qgroup
*qg
, u64 num_bytes
)
2314 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_RFER
) &&
2315 qg
->reserved
+ (s64
)qg
->rfer
+ num_bytes
> qg
->max_rfer
)
2318 if ((qg
->lim_flags
& BTRFS_QGROUP_LIMIT_MAX_EXCL
) &&
2319 qg
->reserved
+ (s64
)qg
->excl
+ num_bytes
> qg
->max_excl
)
2325 static int qgroup_reserve(struct btrfs_root
*root
, u64 num_bytes
, bool enforce
)
2327 struct btrfs_root
*quota_root
;
2328 struct btrfs_qgroup
*qgroup
;
2329 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2330 u64 ref_root
= root
->root_key
.objectid
;
2333 struct ulist_node
*unode
;
2334 struct ulist_iterator uiter
;
2336 if (!is_fstree(ref_root
))
2342 spin_lock(&fs_info
->qgroup_lock
);
2343 quota_root
= fs_info
->quota_root
;
2347 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
2352 * in a first step, we check all affected qgroups if any limits would
2355 ulist_reinit(fs_info
->qgroup_ulist
);
2356 ret
= ulist_add(fs_info
->qgroup_ulist
, qgroup
->qgroupid
,
2357 (uintptr_t)qgroup
, GFP_ATOMIC
);
2360 ULIST_ITER_INIT(&uiter
);
2361 while ((unode
= ulist_next(fs_info
->qgroup_ulist
, &uiter
))) {
2362 struct btrfs_qgroup
*qg
;
2363 struct btrfs_qgroup_list
*glist
;
2365 qg
= unode_aux_to_qgroup(unode
);
2367 if (enforce
&& !qgroup_check_limits(qg
, num_bytes
)) {
2369 * Commit the tree and retry, since we may have
2370 * deletions which would free up space.
2372 if (!retried
&& qg
->reserved
> 0) {
2373 struct btrfs_trans_handle
*trans
;
2375 spin_unlock(&fs_info
->qgroup_lock
);
2376 ret
= btrfs_start_delalloc_inodes(root
, 0);
2379 btrfs_wait_ordered_extents(root
, -1, 0, (u64
)-1);
2380 trans
= btrfs_join_transaction(root
);
2382 return PTR_ERR(trans
);
2383 ret
= btrfs_commit_transaction(trans
);
2393 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
2394 ret
= ulist_add(fs_info
->qgroup_ulist
,
2395 glist
->group
->qgroupid
,
2396 (uintptr_t)glist
->group
, GFP_ATOMIC
);
2403 * no limits exceeded, now record the reservation into all qgroups
2405 ULIST_ITER_INIT(&uiter
);
2406 while ((unode
= ulist_next(fs_info
->qgroup_ulist
, &uiter
))) {
2407 struct btrfs_qgroup
*qg
;
2409 qg
= unode_aux_to_qgroup(unode
);
2411 trace_qgroup_update_reserve(fs_info
, qg
, num_bytes
);
2412 qg
->reserved
+= num_bytes
;
2416 spin_unlock(&fs_info
->qgroup_lock
);
2420 void btrfs_qgroup_free_refroot(struct btrfs_fs_info
*fs_info
,
2421 u64 ref_root
, u64 num_bytes
)
2423 struct btrfs_root
*quota_root
;
2424 struct btrfs_qgroup
*qgroup
;
2425 struct ulist_node
*unode
;
2426 struct ulist_iterator uiter
;
2429 if (!is_fstree(ref_root
))
2435 spin_lock(&fs_info
->qgroup_lock
);
2437 quota_root
= fs_info
->quota_root
;
2441 qgroup
= find_qgroup_rb(fs_info
, ref_root
);
2445 ulist_reinit(fs_info
->qgroup_ulist
);
2446 ret
= ulist_add(fs_info
->qgroup_ulist
, qgroup
->qgroupid
,
2447 (uintptr_t)qgroup
, GFP_ATOMIC
);
2450 ULIST_ITER_INIT(&uiter
);
2451 while ((unode
= ulist_next(fs_info
->qgroup_ulist
, &uiter
))) {
2452 struct btrfs_qgroup
*qg
;
2453 struct btrfs_qgroup_list
*glist
;
2455 qg
= unode_aux_to_qgroup(unode
);
2457 trace_qgroup_update_reserve(fs_info
, qg
, -(s64
)num_bytes
);
2458 if (qg
->reserved
< num_bytes
)
2459 report_reserved_underflow(fs_info
, qg
, num_bytes
);
2461 qg
->reserved
-= num_bytes
;
2463 list_for_each_entry(glist
, &qg
->groups
, next_group
) {
2464 ret
= ulist_add(fs_info
->qgroup_ulist
,
2465 glist
->group
->qgroupid
,
2466 (uintptr_t)glist
->group
, GFP_ATOMIC
);
2473 spin_unlock(&fs_info
->qgroup_lock
);
2477 * returns < 0 on error, 0 when more leafs are to be scanned.
2478 * returns 1 when done.
2481 qgroup_rescan_leaf(struct btrfs_fs_info
*fs_info
, struct btrfs_path
*path
,
2482 struct btrfs_trans_handle
*trans
)
2484 struct btrfs_key found
;
2485 struct extent_buffer
*scratch_leaf
= NULL
;
2486 struct ulist
*roots
= NULL
;
2487 struct seq_list tree_mod_seq_elem
= SEQ_LIST_INIT(tree_mod_seq_elem
);
2492 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2493 ret
= btrfs_search_slot_for_read(fs_info
->extent_root
,
2494 &fs_info
->qgroup_rescan_progress
,
2497 btrfs_debug(fs_info
,
2498 "current progress key (%llu %u %llu), search_slot ret %d",
2499 fs_info
->qgroup_rescan_progress
.objectid
,
2500 fs_info
->qgroup_rescan_progress
.type
,
2501 fs_info
->qgroup_rescan_progress
.offset
, ret
);
2505 * The rescan is about to end, we will not be scanning any
2506 * further blocks. We cannot unset the RESCAN flag here, because
2507 * we want to commit the transaction if everything went well.
2508 * To make the live accounting work in this phase, we set our
2509 * scan progress pointer such that every real extent objectid
2512 fs_info
->qgroup_rescan_progress
.objectid
= (u64
)-1;
2513 btrfs_release_path(path
);
2514 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2518 btrfs_item_key_to_cpu(path
->nodes
[0], &found
,
2519 btrfs_header_nritems(path
->nodes
[0]) - 1);
2520 fs_info
->qgroup_rescan_progress
.objectid
= found
.objectid
+ 1;
2522 btrfs_get_tree_mod_seq(fs_info
, &tree_mod_seq_elem
);
2523 scratch_leaf
= btrfs_clone_extent_buffer(path
->nodes
[0]);
2524 if (!scratch_leaf
) {
2526 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2529 extent_buffer_get(scratch_leaf
);
2530 btrfs_tree_read_lock(scratch_leaf
);
2531 btrfs_set_lock_blocking_rw(scratch_leaf
, BTRFS_READ_LOCK
);
2532 slot
= path
->slots
[0];
2533 btrfs_release_path(path
);
2534 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2536 for (; slot
< btrfs_header_nritems(scratch_leaf
); ++slot
) {
2537 btrfs_item_key_to_cpu(scratch_leaf
, &found
, slot
);
2538 if (found
.type
!= BTRFS_EXTENT_ITEM_KEY
&&
2539 found
.type
!= BTRFS_METADATA_ITEM_KEY
)
2541 if (found
.type
== BTRFS_METADATA_ITEM_KEY
)
2542 num_bytes
= fs_info
->nodesize
;
2544 num_bytes
= found
.offset
;
2546 ret
= btrfs_find_all_roots(NULL
, fs_info
, found
.objectid
, 0,
2550 /* For rescan, just pass old_roots as NULL */
2551 ret
= btrfs_qgroup_account_extent(trans
, fs_info
,
2552 found
.objectid
, num_bytes
, NULL
, roots
);
2558 btrfs_tree_read_unlock_blocking(scratch_leaf
);
2559 free_extent_buffer(scratch_leaf
);
2561 btrfs_put_tree_mod_seq(fs_info
, &tree_mod_seq_elem
);
2566 static void btrfs_qgroup_rescan_worker(struct btrfs_work
*work
)
2568 struct btrfs_fs_info
*fs_info
= container_of(work
, struct btrfs_fs_info
,
2569 qgroup_rescan_work
);
2570 struct btrfs_path
*path
;
2571 struct btrfs_trans_handle
*trans
= NULL
;
2575 path
= btrfs_alloc_path();
2580 while (!err
&& !btrfs_fs_closing(fs_info
)) {
2581 trans
= btrfs_start_transaction(fs_info
->fs_root
, 0);
2582 if (IS_ERR(trans
)) {
2583 err
= PTR_ERR(trans
);
2586 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
2589 err
= qgroup_rescan_leaf(fs_info
, path
, trans
);
2592 btrfs_commit_transaction(trans
);
2594 btrfs_end_transaction(trans
);
2598 btrfs_free_path(path
);
2600 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2601 if (!btrfs_fs_closing(fs_info
))
2602 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2605 fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
) {
2606 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2607 } else if (err
< 0) {
2608 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT
;
2610 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2613 * only update status, since the previous part has already updated the
2616 trans
= btrfs_start_transaction(fs_info
->quota_root
, 1);
2617 if (IS_ERR(trans
)) {
2618 err
= PTR_ERR(trans
);
2620 "fail to start transaction for status update: %d\n",
2624 ret
= update_qgroup_status_item(trans
, fs_info
, fs_info
->quota_root
);
2627 btrfs_err(fs_info
, "fail to update qgroup status: %d", err
);
2629 btrfs_end_transaction(trans
);
2631 if (btrfs_fs_closing(fs_info
)) {
2632 btrfs_info(fs_info
, "qgroup scan paused");
2633 } else if (err
>= 0) {
2634 btrfs_info(fs_info
, "qgroup scan completed%s",
2635 err
> 0 ? " (inconsistency flag cleared)" : "");
2637 btrfs_err(fs_info
, "qgroup scan failed with %d", err
);
2641 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2642 fs_info
->qgroup_rescan_running
= false;
2643 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2644 complete_all(&fs_info
->qgroup_rescan_completion
);
2648 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2649 * memory required for the rescan context.
2652 qgroup_rescan_init(struct btrfs_fs_info
*fs_info
, u64 progress_objectid
,
2658 (!(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
) ||
2659 !(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
))) {
2664 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2665 spin_lock(&fs_info
->qgroup_lock
);
2668 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
)
2670 else if (!(fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_ON
))
2674 spin_unlock(&fs_info
->qgroup_lock
);
2675 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2678 fs_info
->qgroup_flags
|= BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2681 memset(&fs_info
->qgroup_rescan_progress
, 0,
2682 sizeof(fs_info
->qgroup_rescan_progress
));
2683 fs_info
->qgroup_rescan_progress
.objectid
= progress_objectid
;
2684 init_completion(&fs_info
->qgroup_rescan_completion
);
2685 fs_info
->qgroup_rescan_running
= true;
2687 spin_unlock(&fs_info
->qgroup_lock
);
2688 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2690 memset(&fs_info
->qgroup_rescan_work
, 0,
2691 sizeof(fs_info
->qgroup_rescan_work
));
2692 btrfs_init_work(&fs_info
->qgroup_rescan_work
,
2693 btrfs_qgroup_rescan_helper
,
2694 btrfs_qgroup_rescan_worker
, NULL
, NULL
);
2698 btrfs_info(fs_info
, "qgroup_rescan_init failed with %d", ret
);
2706 qgroup_rescan_zero_tracking(struct btrfs_fs_info
*fs_info
)
2709 struct btrfs_qgroup
*qgroup
;
2711 spin_lock(&fs_info
->qgroup_lock
);
2712 /* clear all current qgroup tracking information */
2713 for (n
= rb_first(&fs_info
->qgroup_tree
); n
; n
= rb_next(n
)) {
2714 qgroup
= rb_entry(n
, struct btrfs_qgroup
, node
);
2716 qgroup
->rfer_cmpr
= 0;
2718 qgroup
->excl_cmpr
= 0;
2720 spin_unlock(&fs_info
->qgroup_lock
);
2724 btrfs_qgroup_rescan(struct btrfs_fs_info
*fs_info
)
2727 struct btrfs_trans_handle
*trans
;
2729 ret
= qgroup_rescan_init(fs_info
, 0, 1);
2734 * We have set the rescan_progress to 0, which means no more
2735 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2736 * However, btrfs_qgroup_account_ref may be right after its call
2737 * to btrfs_find_all_roots, in which case it would still do the
2739 * To solve this, we're committing the transaction, which will
2740 * ensure we run all delayed refs and only after that, we are
2741 * going to clear all tracking information for a clean start.
2744 trans
= btrfs_join_transaction(fs_info
->fs_root
);
2745 if (IS_ERR(trans
)) {
2746 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2747 return PTR_ERR(trans
);
2749 ret
= btrfs_commit_transaction(trans
);
2751 fs_info
->qgroup_flags
&= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN
;
2755 qgroup_rescan_zero_tracking(fs_info
);
2757 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
2758 &fs_info
->qgroup_rescan_work
);
2763 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info
*fs_info
,
2769 mutex_lock(&fs_info
->qgroup_rescan_lock
);
2770 spin_lock(&fs_info
->qgroup_lock
);
2771 running
= fs_info
->qgroup_rescan_running
;
2772 spin_unlock(&fs_info
->qgroup_lock
);
2773 mutex_unlock(&fs_info
->qgroup_rescan_lock
);
2779 ret
= wait_for_completion_interruptible(
2780 &fs_info
->qgroup_rescan_completion
);
2782 wait_for_completion(&fs_info
->qgroup_rescan_completion
);
2788 * this is only called from open_ctree where we're still single threaded, thus
2789 * locking is omitted here.
2792 btrfs_qgroup_rescan_resume(struct btrfs_fs_info
*fs_info
)
2794 if (fs_info
->qgroup_flags
& BTRFS_QGROUP_STATUS_FLAG_RESCAN
)
2795 btrfs_queue_work(fs_info
->qgroup_rescan_workers
,
2796 &fs_info
->qgroup_rescan_work
);
2800 * Reserve qgroup space for range [start, start + len).
2802 * This function will either reserve space from related qgroups or doing
2803 * nothing if the range is already reserved.
2805 * Return 0 for successful reserve
2806 * Return <0 for error (including -EQUOT)
2808 * NOTE: this function may sleep for memory allocation.
2810 int btrfs_qgroup_reserve_data(struct inode
*inode
, u64 start
, u64 len
)
2812 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2813 struct extent_changeset changeset
;
2814 struct ulist_node
*unode
;
2815 struct ulist_iterator uiter
;
2818 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &root
->fs_info
->flags
) ||
2819 !is_fstree(root
->objectid
) || len
== 0)
2822 changeset
.bytes_changed
= 0;
2823 ulist_init(&changeset
.range_changed
);
2824 ret
= set_record_extent_bits(&BTRFS_I(inode
)->io_tree
, start
,
2825 start
+ len
-1, EXTENT_QGROUP_RESERVED
, &changeset
);
2826 trace_btrfs_qgroup_reserve_data(inode
, start
, len
,
2827 changeset
.bytes_changed
,
2831 ret
= qgroup_reserve(root
, changeset
.bytes_changed
, true);
2835 ulist_release(&changeset
.range_changed
);
2839 /* cleanup already reserved ranges */
2840 ULIST_ITER_INIT(&uiter
);
2841 while ((unode
= ulist_next(&changeset
.range_changed
, &uiter
)))
2842 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, unode
->val
,
2843 unode
->aux
, EXTENT_QGROUP_RESERVED
, 0, 0, NULL
,
2845 ulist_release(&changeset
.range_changed
);
2849 static int __btrfs_qgroup_release_data(struct inode
*inode
, u64 start
, u64 len
,
2852 struct extent_changeset changeset
;
2853 int trace_op
= QGROUP_RELEASE
;
2856 changeset
.bytes_changed
= 0;
2857 ulist_init(&changeset
.range_changed
);
2858 ret
= clear_record_extent_bits(&BTRFS_I(inode
)->io_tree
, start
,
2859 start
+ len
-1, EXTENT_QGROUP_RESERVED
, &changeset
);
2864 trace_op
= QGROUP_FREE
;
2865 trace_btrfs_qgroup_release_data(inode
, start
, len
,
2866 changeset
.bytes_changed
, trace_op
);
2868 btrfs_qgroup_free_refroot(BTRFS_I(inode
)->root
->fs_info
,
2869 BTRFS_I(inode
)->root
->objectid
,
2870 changeset
.bytes_changed
);
2872 ulist_release(&changeset
.range_changed
);
2877 * Free a reserved space range from io_tree and related qgroups
2879 * Should be called when a range of pages get invalidated before reaching disk.
2880 * Or for error cleanup case.
2882 * For data written to disk, use btrfs_qgroup_release_data().
2884 * NOTE: This function may sleep for memory allocation.
2886 int btrfs_qgroup_free_data(struct inode
*inode
, u64 start
, u64 len
)
2888 return __btrfs_qgroup_release_data(inode
, start
, len
, 1);
2892 * Release a reserved space range from io_tree only.
2894 * Should be called when a range of pages get written to disk and corresponding
2895 * FILE_EXTENT is inserted into corresponding root.
2897 * Since new qgroup accounting framework will only update qgroup numbers at
2898 * commit_transaction() time, its reserved space shouldn't be freed from
2901 * But we should release the range from io_tree, to allow further write to be
2904 * NOTE: This function may sleep for memory allocation.
2906 int btrfs_qgroup_release_data(struct inode
*inode
, u64 start
, u64 len
)
2908 return __btrfs_qgroup_release_data(inode
, start
, len
, 0);
2911 int btrfs_qgroup_reserve_meta(struct btrfs_root
*root
, int num_bytes
,
2914 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2917 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
) ||
2918 !is_fstree(root
->objectid
) || num_bytes
== 0)
2921 BUG_ON(num_bytes
!= round_down(num_bytes
, fs_info
->nodesize
));
2922 trace_qgroup_meta_reserve(root
, (s64
)num_bytes
);
2923 ret
= qgroup_reserve(root
, num_bytes
, enforce
);
2926 atomic64_add(num_bytes
, &root
->qgroup_meta_rsv
);
2930 void btrfs_qgroup_free_meta_all(struct btrfs_root
*root
)
2932 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2935 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
) ||
2936 !is_fstree(root
->objectid
))
2939 reserved
= atomic64_xchg(&root
->qgroup_meta_rsv
, 0);
2942 trace_qgroup_meta_reserve(root
, -(s64
)reserved
);
2943 btrfs_qgroup_free_refroot(fs_info
, root
->objectid
, reserved
);
2946 void btrfs_qgroup_free_meta(struct btrfs_root
*root
, int num_bytes
)
2948 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2950 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
) ||
2951 !is_fstree(root
->objectid
))
2954 BUG_ON(num_bytes
!= round_down(num_bytes
, fs_info
->nodesize
));
2955 WARN_ON(atomic64_read(&root
->qgroup_meta_rsv
) < num_bytes
);
2956 atomic64_sub(num_bytes
, &root
->qgroup_meta_rsv
);
2957 trace_qgroup_meta_reserve(root
, -(s64
)num_bytes
);
2958 btrfs_qgroup_free_refroot(fs_info
, root
->objectid
, num_bytes
);
2962 * Check qgroup reserved space leaking, normally at destroy inode
2965 void btrfs_qgroup_check_reserved_leak(struct inode
*inode
)
2967 struct extent_changeset changeset
;
2968 struct ulist_node
*unode
;
2969 struct ulist_iterator iter
;
2972 changeset
.bytes_changed
= 0;
2973 ulist_init(&changeset
.range_changed
);
2974 ret
= clear_record_extent_bits(&BTRFS_I(inode
)->io_tree
, 0, (u64
)-1,
2975 EXTENT_QGROUP_RESERVED
, &changeset
);
2978 if (WARN_ON(changeset
.bytes_changed
)) {
2979 ULIST_ITER_INIT(&iter
);
2980 while ((unode
= ulist_next(&changeset
.range_changed
, &iter
))) {
2981 btrfs_warn(BTRFS_I(inode
)->root
->fs_info
,
2982 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
2983 inode
->i_ino
, unode
->val
, unode
->aux
);
2985 btrfs_qgroup_free_refroot(BTRFS_I(inode
)->root
->fs_info
,
2986 BTRFS_I(inode
)->root
->objectid
,
2987 changeset
.bytes_changed
);
2990 ulist_release(&changeset
.range_changed
);